Loading...
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_PENDING 1
51#define STATE_READY 2
52
53struct posix_msg_tree_node {
54 struct rb_node rb_node;
55 struct list_head msg_list;
56 int priority;
57};
58
59struct ext_wait_queue { /* queue of sleeping tasks */
60 struct task_struct *task;
61 struct list_head list;
62 struct msg_msg *msg; /* ptr of loaded message */
63 int state; /* one of STATE_* values */
64};
65
66struct mqueue_inode_info {
67 spinlock_t lock;
68 struct inode vfs_inode;
69 wait_queue_head_t wait_q;
70
71 struct rb_root msg_tree;
72 struct posix_msg_tree_node *node_cache;
73 struct mq_attr attr;
74
75 struct sigevent notify;
76 struct pid *notify_owner;
77 struct user_namespace *notify_user_ns;
78 struct user_struct *user; /* user who created, for accounting */
79 struct sock *notify_sock;
80 struct sk_buff *notify_cookie;
81
82 /* for tasks waiting for free space and messages, respectively */
83 struct ext_wait_queue e_wait_q[2];
84
85 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
86};
87
88static const struct inode_operations mqueue_dir_inode_operations;
89static const struct file_operations mqueue_file_operations;
90static const struct super_operations mqueue_super_ops;
91static void remove_notification(struct mqueue_inode_info *info);
92
93static struct kmem_cache *mqueue_inode_cachep;
94
95static struct ctl_table_header *mq_sysctl_table;
96
97static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98{
99 return container_of(inode, struct mqueue_inode_info, vfs_inode);
100}
101
102/*
103 * This routine should be called with the mq_lock held.
104 */
105static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
106{
107 return get_ipc_ns(inode->i_sb->s_fs_info);
108}
109
110static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
111{
112 struct ipc_namespace *ns;
113
114 spin_lock(&mq_lock);
115 ns = __get_ns_from_inode(inode);
116 spin_unlock(&mq_lock);
117 return ns;
118}
119
120/* Auxiliary functions to manipulate messages' list */
121static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122{
123 struct rb_node **p, *parent = NULL;
124 struct posix_msg_tree_node *leaf;
125
126 p = &info->msg_tree.rb_node;
127 while (*p) {
128 parent = *p;
129 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130
131 if (likely(leaf->priority == msg->m_type))
132 goto insert_msg;
133 else if (msg->m_type < leaf->priority)
134 p = &(*p)->rb_left;
135 else
136 p = &(*p)->rb_right;
137 }
138 if (info->node_cache) {
139 leaf = info->node_cache;
140 info->node_cache = NULL;
141 } else {
142 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
143 if (!leaf)
144 return -ENOMEM;
145 INIT_LIST_HEAD(&leaf->msg_list);
146 info->qsize += sizeof(*leaf);
147 }
148 leaf->priority = msg->m_type;
149 rb_link_node(&leaf->rb_node, parent, p);
150 rb_insert_color(&leaf->rb_node, &info->msg_tree);
151insert_msg:
152 info->attr.mq_curmsgs++;
153 info->qsize += msg->m_ts;
154 list_add_tail(&msg->m_list, &leaf->msg_list);
155 return 0;
156}
157
158static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
159{
160 struct rb_node **p, *parent = NULL;
161 struct posix_msg_tree_node *leaf;
162 struct msg_msg *msg;
163
164try_again:
165 p = &info->msg_tree.rb_node;
166 while (*p) {
167 parent = *p;
168 /*
169 * During insert, low priorities go to the left and high to the
170 * right. On receive, we want the highest priorities first, so
171 * walk all the way to the right.
172 */
173 p = &(*p)->rb_right;
174 }
175 if (!parent) {
176 if (info->attr.mq_curmsgs) {
177 pr_warn_once("Inconsistency in POSIX message queue, "
178 "no tree element, but supposedly messages "
179 "should exist!\n");
180 info->attr.mq_curmsgs = 0;
181 }
182 return NULL;
183 }
184 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
185 if (unlikely(list_empty(&leaf->msg_list))) {
186 pr_warn_once("Inconsistency in POSIX message queue, "
187 "empty leaf node but we haven't implemented "
188 "lazy leaf delete!\n");
189 rb_erase(&leaf->rb_node, &info->msg_tree);
190 if (info->node_cache) {
191 info->qsize -= sizeof(*leaf);
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 info->qsize -= sizeof(*leaf);
205 kfree(leaf);
206 } else {
207 info->node_cache = leaf;
208 }
209 }
210 }
211 info->attr.mq_curmsgs--;
212 info->qsize -= msg->m_ts;
213 return msg;
214}
215
216static struct inode *mqueue_get_inode(struct super_block *sb,
217 struct ipc_namespace *ipc_ns, umode_t mode,
218 struct mq_attr *attr)
219{
220 struct user_struct *u = current_user();
221 struct inode *inode;
222 int ret = -ENOMEM;
223
224 inode = new_inode(sb);
225 if (!inode)
226 goto err;
227
228 inode->i_ino = get_next_ino();
229 inode->i_mode = mode;
230 inode->i_uid = current_fsuid();
231 inode->i_gid = current_fsgid();
232 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
233
234 if (S_ISREG(mode)) {
235 struct mqueue_inode_info *info;
236 unsigned long mq_bytes, mq_treesize;
237
238 inode->i_fop = &mqueue_file_operations;
239 inode->i_size = FILENT_SIZE;
240 /* mqueue specific info */
241 info = MQUEUE_I(inode);
242 spin_lock_init(&info->lock);
243 init_waitqueue_head(&info->wait_q);
244 INIT_LIST_HEAD(&info->e_wait_q[0].list);
245 INIT_LIST_HEAD(&info->e_wait_q[1].list);
246 info->notify_owner = NULL;
247 info->notify_user_ns = NULL;
248 info->qsize = 0;
249 info->user = NULL; /* set when all is ok */
250 info->msg_tree = RB_ROOT;
251 info->node_cache = NULL;
252 memset(&info->attr, 0, sizeof(info->attr));
253 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
254 ipc_ns->mq_msg_default);
255 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
256 ipc_ns->mq_msgsize_default);
257 if (attr) {
258 info->attr.mq_maxmsg = attr->mq_maxmsg;
259 info->attr.mq_msgsize = attr->mq_msgsize;
260 }
261 /*
262 * We used to allocate a static array of pointers and account
263 * the size of that array as well as one msg_msg struct per
264 * possible message into the queue size. That's no longer
265 * accurate as the queue is now an rbtree and will grow and
266 * shrink depending on usage patterns. We can, however, still
267 * account one msg_msg struct per message, but the nodes are
268 * allocated depending on priority usage, and most programs
269 * only use one, or a handful, of priorities. However, since
270 * this is pinned memory, we need to assume worst case, so
271 * that means the min(mq_maxmsg, max_priorities) * struct
272 * posix_msg_tree_node.
273 */
274 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
275 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
276 sizeof(struct posix_msg_tree_node);
277
278 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
279 info->attr.mq_msgsize);
280
281 spin_lock(&mq_lock);
282 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
283 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
284 spin_unlock(&mq_lock);
285 /* mqueue_evict_inode() releases info->messages */
286 ret = -EMFILE;
287 goto out_inode;
288 }
289 u->mq_bytes += mq_bytes;
290 spin_unlock(&mq_lock);
291
292 /* all is ok */
293 info->user = get_uid(u);
294 } else if (S_ISDIR(mode)) {
295 inc_nlink(inode);
296 /* Some things misbehave if size == 0 on a directory */
297 inode->i_size = 2 * DIRENT_SIZE;
298 inode->i_op = &mqueue_dir_inode_operations;
299 inode->i_fop = &simple_dir_operations;
300 }
301
302 return inode;
303out_inode:
304 iput(inode);
305err:
306 return ERR_PTR(ret);
307}
308
309static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
310{
311 struct inode *inode;
312 struct ipc_namespace *ns = data;
313
314 sb->s_blocksize = PAGE_CACHE_SIZE;
315 sb->s_blocksize_bits = PAGE_CACHE_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 if (!(flags & MS_KERNMOUNT)) {
334 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
335 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
336 * over the ipc namespace.
337 */
338 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
339 return ERR_PTR(-EPERM);
340
341 data = ns;
342 }
343 return mount_ns(fs_type, flags, data, mqueue_fill_super);
344}
345
346static void init_once(void *foo)
347{
348 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
349
350 inode_init_once(&p->vfs_inode);
351}
352
353static struct inode *mqueue_alloc_inode(struct super_block *sb)
354{
355 struct mqueue_inode_info *ei;
356
357 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
358 if (!ei)
359 return NULL;
360 return &ei->vfs_inode;
361}
362
363static void mqueue_i_callback(struct rcu_head *head)
364{
365 struct inode *inode = container_of(head, struct inode, i_rcu);
366 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
367}
368
369static void mqueue_destroy_inode(struct inode *inode)
370{
371 call_rcu(&inode->i_rcu, mqueue_i_callback);
372}
373
374static void mqueue_evict_inode(struct inode *inode)
375{
376 struct mqueue_inode_info *info;
377 struct user_struct *user;
378 unsigned long mq_bytes, mq_treesize;
379 struct ipc_namespace *ipc_ns;
380 struct msg_msg *msg;
381
382 clear_inode(inode);
383
384 if (S_ISDIR(inode->i_mode))
385 return;
386
387 ipc_ns = get_ns_from_inode(inode);
388 info = MQUEUE_I(inode);
389 spin_lock(&info->lock);
390 while ((msg = msg_get(info)) != NULL)
391 free_msg(msg);
392 kfree(info->node_cache);
393 spin_unlock(&info->lock);
394
395 /* Total amount of bytes accounted for the mqueue */
396 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
397 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
398 sizeof(struct posix_msg_tree_node);
399
400 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
401 info->attr.mq_msgsize);
402
403 user = info->user;
404 if (user) {
405 spin_lock(&mq_lock);
406 user->mq_bytes -= mq_bytes;
407 /*
408 * get_ns_from_inode() ensures that the
409 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
410 * to which we now hold a reference, or it is NULL.
411 * We can't put it here under mq_lock, though.
412 */
413 if (ipc_ns)
414 ipc_ns->mq_queues_count--;
415 spin_unlock(&mq_lock);
416 free_uid(user);
417 }
418 if (ipc_ns)
419 put_ipc_ns(ipc_ns);
420}
421
422static int mqueue_create(struct inode *dir, struct dentry *dentry,
423 umode_t mode, bool excl)
424{
425 struct inode *inode;
426 struct mq_attr *attr = dentry->d_fsdata;
427 int error;
428 struct ipc_namespace *ipc_ns;
429
430 spin_lock(&mq_lock);
431 ipc_ns = __get_ns_from_inode(dir);
432 if (!ipc_ns) {
433 error = -EACCES;
434 goto out_unlock;
435 }
436
437 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
438 !capable(CAP_SYS_RESOURCE)) {
439 error = -ENOSPC;
440 goto out_unlock;
441 }
442 ipc_ns->mq_queues_count++;
443 spin_unlock(&mq_lock);
444
445 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
446 if (IS_ERR(inode)) {
447 error = PTR_ERR(inode);
448 spin_lock(&mq_lock);
449 ipc_ns->mq_queues_count--;
450 goto out_unlock;
451 }
452
453 put_ipc_ns(ipc_ns);
454 dir->i_size += DIRENT_SIZE;
455 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
456
457 d_instantiate(dentry, inode);
458 dget(dentry);
459 return 0;
460out_unlock:
461 spin_unlock(&mq_lock);
462 if (ipc_ns)
463 put_ipc_ns(ipc_ns);
464 return error;
465}
466
467static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
468{
469 struct inode *inode = dentry->d_inode;
470
471 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
472 dir->i_size -= DIRENT_SIZE;
473 drop_nlink(inode);
474 dput(dentry);
475 return 0;
476}
477
478/*
479* This is routine for system read from queue file.
480* To avoid mess with doing here some sort of mq_receive we allow
481* to read only queue size & notification info (the only values
482* that are interesting from user point of view and aren't accessible
483* through std routines)
484*/
485static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
486 size_t count, loff_t *off)
487{
488 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
489 char buffer[FILENT_SIZE];
490 ssize_t ret;
491
492 spin_lock(&info->lock);
493 snprintf(buffer, sizeof(buffer),
494 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
495 info->qsize,
496 info->notify_owner ? info->notify.sigev_notify : 0,
497 (info->notify_owner &&
498 info->notify.sigev_notify == SIGEV_SIGNAL) ?
499 info->notify.sigev_signo : 0,
500 pid_vnr(info->notify_owner));
501 spin_unlock(&info->lock);
502 buffer[sizeof(buffer)-1] = '\0';
503
504 ret = simple_read_from_buffer(u_data, count, off, buffer,
505 strlen(buffer));
506 if (ret <= 0)
507 return ret;
508
509 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
510 return ret;
511}
512
513static int mqueue_flush_file(struct file *filp, fl_owner_t id)
514{
515 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
516
517 spin_lock(&info->lock);
518 if (task_tgid(current) == info->notify_owner)
519 remove_notification(info);
520
521 spin_unlock(&info->lock);
522 return 0;
523}
524
525static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
526{
527 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
528 int retval = 0;
529
530 poll_wait(filp, &info->wait_q, poll_tab);
531
532 spin_lock(&info->lock);
533 if (info->attr.mq_curmsgs)
534 retval = POLLIN | POLLRDNORM;
535
536 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
537 retval |= POLLOUT | POLLWRNORM;
538 spin_unlock(&info->lock);
539
540 return retval;
541}
542
543/* Adds current to info->e_wait_q[sr] before element with smaller prio */
544static void wq_add(struct mqueue_inode_info *info, int sr,
545 struct ext_wait_queue *ewp)
546{
547 struct ext_wait_queue *walk;
548
549 ewp->task = current;
550
551 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
552 if (walk->task->static_prio <= current->static_prio) {
553 list_add_tail(&ewp->list, &walk->list);
554 return;
555 }
556 }
557 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
558}
559
560/*
561 * Puts current task to sleep. Caller must hold queue lock. After return
562 * lock isn't held.
563 * sr: SEND or RECV
564 */
565static int wq_sleep(struct mqueue_inode_info *info, int sr,
566 ktime_t *timeout, struct ext_wait_queue *ewp)
567{
568 int retval;
569 signed long time;
570
571 wq_add(info, sr, ewp);
572
573 for (;;) {
574 set_current_state(TASK_INTERRUPTIBLE);
575
576 spin_unlock(&info->lock);
577 time = schedule_hrtimeout_range_clock(timeout, 0,
578 HRTIMER_MODE_ABS, CLOCK_REALTIME);
579
580 while (ewp->state == STATE_PENDING)
581 cpu_relax();
582
583 if (ewp->state == STATE_READY) {
584 retval = 0;
585 goto out;
586 }
587 spin_lock(&info->lock);
588 if (ewp->state == STATE_READY) {
589 retval = 0;
590 goto out_unlock;
591 }
592 if (signal_pending(current)) {
593 retval = -ERESTARTSYS;
594 break;
595 }
596 if (time == 0) {
597 retval = -ETIMEDOUT;
598 break;
599 }
600 }
601 list_del(&ewp->list);
602out_unlock:
603 spin_unlock(&info->lock);
604out:
605 return retval;
606}
607
608/*
609 * Returns waiting task that should be serviced first or NULL if none exists
610 */
611static struct ext_wait_queue *wq_get_first_waiter(
612 struct mqueue_inode_info *info, int sr)
613{
614 struct list_head *ptr;
615
616 ptr = info->e_wait_q[sr].list.prev;
617 if (ptr == &info->e_wait_q[sr].list)
618 return NULL;
619 return list_entry(ptr, struct ext_wait_queue, list);
620}
621
622
623static inline void set_cookie(struct sk_buff *skb, char code)
624{
625 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
626}
627
628/*
629 * The next function is only to split too long sys_mq_timedsend
630 */
631static void __do_notify(struct mqueue_inode_info *info)
632{
633 /* notification
634 * invoked when there is registered process and there isn't process
635 * waiting synchronously for message AND state of queue changed from
636 * empty to not empty. Here we are sure that no one is waiting
637 * synchronously. */
638 if (info->notify_owner &&
639 info->attr.mq_curmsgs == 1) {
640 struct siginfo sig_i;
641 switch (info->notify.sigev_notify) {
642 case SIGEV_NONE:
643 break;
644 case SIGEV_SIGNAL:
645 /* sends signal */
646
647 sig_i.si_signo = info->notify.sigev_signo;
648 sig_i.si_errno = 0;
649 sig_i.si_code = SI_MESGQ;
650 sig_i.si_value = info->notify.sigev_value;
651 /* map current pid/uid into info->owner's namespaces */
652 rcu_read_lock();
653 sig_i.si_pid = task_tgid_nr_ns(current,
654 ns_of_pid(info->notify_owner));
655 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
656 rcu_read_unlock();
657
658 kill_pid_info(info->notify.sigev_signo,
659 &sig_i, info->notify_owner);
660 break;
661 case SIGEV_THREAD:
662 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
663 netlink_sendskb(info->notify_sock, info->notify_cookie);
664 break;
665 }
666 /* after notification unregisters process */
667 put_pid(info->notify_owner);
668 put_user_ns(info->notify_user_ns);
669 info->notify_owner = NULL;
670 info->notify_user_ns = NULL;
671 }
672 wake_up(&info->wait_q);
673}
674
675static int prepare_timeout(const struct timespec __user *u_abs_timeout,
676 ktime_t *expires, struct timespec *ts)
677{
678 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
679 return -EFAULT;
680 if (!timespec_valid(ts))
681 return -EINVAL;
682
683 *expires = timespec_to_ktime(*ts);
684 return 0;
685}
686
687static void remove_notification(struct mqueue_inode_info *info)
688{
689 if (info->notify_owner != NULL &&
690 info->notify.sigev_notify == SIGEV_THREAD) {
691 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
692 netlink_sendskb(info->notify_sock, info->notify_cookie);
693 }
694 put_pid(info->notify_owner);
695 put_user_ns(info->notify_user_ns);
696 info->notify_owner = NULL;
697 info->notify_user_ns = NULL;
698}
699
700static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
701{
702 int mq_treesize;
703 unsigned long total_size;
704
705 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
706 return -EINVAL;
707 if (capable(CAP_SYS_RESOURCE)) {
708 if (attr->mq_maxmsg > HARD_MSGMAX ||
709 attr->mq_msgsize > HARD_MSGSIZEMAX)
710 return -EINVAL;
711 } else {
712 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
713 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
714 return -EINVAL;
715 }
716 /* check for overflow */
717 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
718 return -EOVERFLOW;
719 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
720 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
721 sizeof(struct posix_msg_tree_node);
722 total_size = attr->mq_maxmsg * attr->mq_msgsize;
723 if (total_size + mq_treesize < total_size)
724 return -EOVERFLOW;
725 return 0;
726}
727
728/*
729 * Invoked when creating a new queue via sys_mq_open
730 */
731static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
732 struct path *path, int oflag, umode_t mode,
733 struct mq_attr *attr)
734{
735 const struct cred *cred = current_cred();
736 int ret;
737
738 if (attr) {
739 ret = mq_attr_ok(ipc_ns, attr);
740 if (ret)
741 return ERR_PTR(ret);
742 /* store for use during create */
743 path->dentry->d_fsdata = attr;
744 } else {
745 struct mq_attr def_attr;
746
747 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
748 ipc_ns->mq_msg_default);
749 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
750 ipc_ns->mq_msgsize_default);
751 ret = mq_attr_ok(ipc_ns, &def_attr);
752 if (ret)
753 return ERR_PTR(ret);
754 }
755
756 mode &= ~current_umask();
757 ret = vfs_create(dir, path->dentry, mode, true);
758 path->dentry->d_fsdata = NULL;
759 if (ret)
760 return ERR_PTR(ret);
761 return dentry_open(path, oflag, cred);
762}
763
764/* Opens existing queue */
765static struct file *do_open(struct path *path, int oflag)
766{
767 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
768 MAY_READ | MAY_WRITE };
769 int acc;
770 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
771 return ERR_PTR(-EINVAL);
772 acc = oflag2acc[oflag & O_ACCMODE];
773 if (inode_permission(path->dentry->d_inode, acc))
774 return ERR_PTR(-EACCES);
775 return dentry_open(path, oflag, current_cred());
776}
777
778SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
779 struct mq_attr __user *, u_attr)
780{
781 struct path path;
782 struct file *filp;
783 struct filename *name;
784 struct mq_attr attr;
785 int fd, error;
786 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
787 struct vfsmount *mnt = ipc_ns->mq_mnt;
788 struct dentry *root = mnt->mnt_root;
789 int ro;
790
791 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
792 return -EFAULT;
793
794 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
795
796 if (IS_ERR(name = getname(u_name)))
797 return PTR_ERR(name);
798
799 fd = get_unused_fd_flags(O_CLOEXEC);
800 if (fd < 0)
801 goto out_putname;
802
803 ro = mnt_want_write(mnt); /* we'll drop it in any case */
804 error = 0;
805 mutex_lock(&root->d_inode->i_mutex);
806 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
807 if (IS_ERR(path.dentry)) {
808 error = PTR_ERR(path.dentry);
809 goto out_putfd;
810 }
811 path.mnt = mntget(mnt);
812
813 if (oflag & O_CREAT) {
814 if (path.dentry->d_inode) { /* entry already exists */
815 audit_inode(name, path.dentry, 0);
816 if (oflag & O_EXCL) {
817 error = -EEXIST;
818 goto out;
819 }
820 filp = do_open(&path, oflag);
821 } else {
822 if (ro) {
823 error = ro;
824 goto out;
825 }
826 audit_inode_parent_hidden(name, root);
827 filp = do_create(ipc_ns, root->d_inode,
828 &path, oflag, mode,
829 u_attr ? &attr : NULL);
830 }
831 } else {
832 if (!path.dentry->d_inode) {
833 error = -ENOENT;
834 goto out;
835 }
836 audit_inode(name, path.dentry, 0);
837 filp = do_open(&path, oflag);
838 }
839
840 if (!IS_ERR(filp))
841 fd_install(fd, filp);
842 else
843 error = PTR_ERR(filp);
844out:
845 path_put(&path);
846out_putfd:
847 if (error) {
848 put_unused_fd(fd);
849 fd = error;
850 }
851 mutex_unlock(&root->d_inode->i_mutex);
852 if (!ro)
853 mnt_drop_write(mnt);
854out_putname:
855 putname(name);
856 return fd;
857}
858
859SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
860{
861 int err;
862 struct filename *name;
863 struct dentry *dentry;
864 struct inode *inode = NULL;
865 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
866 struct vfsmount *mnt = ipc_ns->mq_mnt;
867
868 name = getname(u_name);
869 if (IS_ERR(name))
870 return PTR_ERR(name);
871
872 audit_inode_parent_hidden(name, mnt->mnt_root);
873 err = mnt_want_write(mnt);
874 if (err)
875 goto out_name;
876 mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
877 dentry = lookup_one_len(name->name, mnt->mnt_root,
878 strlen(name->name));
879 if (IS_ERR(dentry)) {
880 err = PTR_ERR(dentry);
881 goto out_unlock;
882 }
883
884 inode = dentry->d_inode;
885 if (!inode) {
886 err = -ENOENT;
887 } else {
888 ihold(inode);
889 err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
890 }
891 dput(dentry);
892
893out_unlock:
894 mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
895 if (inode)
896 iput(inode);
897 mnt_drop_write(mnt);
898out_name:
899 putname(name);
900
901 return err;
902}
903
904/* Pipelined send and receive functions.
905 *
906 * If a receiver finds no waiting message, then it registers itself in the
907 * list of waiting receivers. A sender checks that list before adding the new
908 * message into the message array. If there is a waiting receiver, then it
909 * bypasses the message array and directly hands the message over to the
910 * receiver.
911 * The receiver accepts the message and returns without grabbing the queue
912 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
913 * are necessary. The same algorithm is used for sysv semaphores, see
914 * ipc/sem.c for more details.
915 *
916 * The same algorithm is used for senders.
917 */
918
919/* pipelined_send() - send a message directly to the task waiting in
920 * sys_mq_timedreceive() (without inserting message into a queue).
921 */
922static inline void pipelined_send(struct mqueue_inode_info *info,
923 struct msg_msg *message,
924 struct ext_wait_queue *receiver)
925{
926 receiver->msg = message;
927 list_del(&receiver->list);
928 receiver->state = STATE_PENDING;
929 wake_up_process(receiver->task);
930 smp_wmb();
931 receiver->state = STATE_READY;
932}
933
934/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
935 * gets its message and put to the queue (we have one free place for sure). */
936static inline void pipelined_receive(struct mqueue_inode_info *info)
937{
938 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
939
940 if (!sender) {
941 /* for poll */
942 wake_up_interruptible(&info->wait_q);
943 return;
944 }
945 if (msg_insert(sender->msg, info))
946 return;
947 list_del(&sender->list);
948 sender->state = STATE_PENDING;
949 wake_up_process(sender->task);
950 smp_wmb();
951 sender->state = STATE_READY;
952}
953
954SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
955 size_t, msg_len, unsigned int, msg_prio,
956 const struct timespec __user *, u_abs_timeout)
957{
958 struct fd f;
959 struct inode *inode;
960 struct ext_wait_queue wait;
961 struct ext_wait_queue *receiver;
962 struct msg_msg *msg_ptr;
963 struct mqueue_inode_info *info;
964 ktime_t expires, *timeout = NULL;
965 struct timespec ts;
966 struct posix_msg_tree_node *new_leaf = NULL;
967 int ret = 0;
968
969 if (u_abs_timeout) {
970 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
971 if (res)
972 return res;
973 timeout = &expires;
974 }
975
976 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
977 return -EINVAL;
978
979 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
980
981 f = fdget(mqdes);
982 if (unlikely(!f.file)) {
983 ret = -EBADF;
984 goto out;
985 }
986
987 inode = file_inode(f.file);
988 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
989 ret = -EBADF;
990 goto out_fput;
991 }
992 info = MQUEUE_I(inode);
993 audit_inode(NULL, f.file->f_path.dentry, 0);
994
995 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
996 ret = -EBADF;
997 goto out_fput;
998 }
999
1000 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1001 ret = -EMSGSIZE;
1002 goto out_fput;
1003 }
1004
1005 /* First try to allocate memory, before doing anything with
1006 * existing queues. */
1007 msg_ptr = load_msg(u_msg_ptr, msg_len);
1008 if (IS_ERR(msg_ptr)) {
1009 ret = PTR_ERR(msg_ptr);
1010 goto out_fput;
1011 }
1012 msg_ptr->m_ts = msg_len;
1013 msg_ptr->m_type = msg_prio;
1014
1015 /*
1016 * msg_insert really wants us to have a valid, spare node struct so
1017 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1018 * fall back to that if necessary.
1019 */
1020 if (!info->node_cache)
1021 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1022
1023 spin_lock(&info->lock);
1024
1025 if (!info->node_cache && new_leaf) {
1026 /* Save our speculative allocation into the cache */
1027 INIT_LIST_HEAD(&new_leaf->msg_list);
1028 info->node_cache = new_leaf;
1029 info->qsize += sizeof(*new_leaf);
1030 new_leaf = NULL;
1031 } else {
1032 kfree(new_leaf);
1033 }
1034
1035 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1036 if (f.file->f_flags & O_NONBLOCK) {
1037 ret = -EAGAIN;
1038 } else {
1039 wait.task = current;
1040 wait.msg = (void *) msg_ptr;
1041 wait.state = STATE_NONE;
1042 ret = wq_sleep(info, SEND, timeout, &wait);
1043 /*
1044 * wq_sleep must be called with info->lock held, and
1045 * returns with the lock released
1046 */
1047 goto out_free;
1048 }
1049 } else {
1050 receiver = wq_get_first_waiter(info, RECV);
1051 if (receiver) {
1052 pipelined_send(info, msg_ptr, receiver);
1053 } else {
1054 /* adds message to the queue */
1055 ret = msg_insert(msg_ptr, info);
1056 if (ret)
1057 goto out_unlock;
1058 __do_notify(info);
1059 }
1060 inode->i_atime = inode->i_mtime = inode->i_ctime =
1061 CURRENT_TIME;
1062 }
1063out_unlock:
1064 spin_unlock(&info->lock);
1065out_free:
1066 if (ret)
1067 free_msg(msg_ptr);
1068out_fput:
1069 fdput(f);
1070out:
1071 return ret;
1072}
1073
1074SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1075 size_t, msg_len, unsigned int __user *, u_msg_prio,
1076 const struct timespec __user *, u_abs_timeout)
1077{
1078 ssize_t ret;
1079 struct msg_msg *msg_ptr;
1080 struct fd f;
1081 struct inode *inode;
1082 struct mqueue_inode_info *info;
1083 struct ext_wait_queue wait;
1084 ktime_t expires, *timeout = NULL;
1085 struct timespec ts;
1086 struct posix_msg_tree_node *new_leaf = NULL;
1087
1088 if (u_abs_timeout) {
1089 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1090 if (res)
1091 return res;
1092 timeout = &expires;
1093 }
1094
1095 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1096
1097 f = fdget(mqdes);
1098 if (unlikely(!f.file)) {
1099 ret = -EBADF;
1100 goto out;
1101 }
1102
1103 inode = file_inode(f.file);
1104 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1105 ret = -EBADF;
1106 goto out_fput;
1107 }
1108 info = MQUEUE_I(inode);
1109 audit_inode(NULL, f.file->f_path.dentry, 0);
1110
1111 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1112 ret = -EBADF;
1113 goto out_fput;
1114 }
1115
1116 /* checks if buffer is big enough */
1117 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1118 ret = -EMSGSIZE;
1119 goto out_fput;
1120 }
1121
1122 /*
1123 * msg_insert really wants us to have a valid, spare node struct so
1124 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1125 * fall back to that if necessary.
1126 */
1127 if (!info->node_cache)
1128 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1129
1130 spin_lock(&info->lock);
1131
1132 if (!info->node_cache && new_leaf) {
1133 /* Save our speculative allocation into the cache */
1134 INIT_LIST_HEAD(&new_leaf->msg_list);
1135 info->node_cache = new_leaf;
1136 info->qsize += sizeof(*new_leaf);
1137 } else {
1138 kfree(new_leaf);
1139 }
1140
1141 if (info->attr.mq_curmsgs == 0) {
1142 if (f.file->f_flags & O_NONBLOCK) {
1143 spin_unlock(&info->lock);
1144 ret = -EAGAIN;
1145 } else {
1146 wait.task = current;
1147 wait.state = STATE_NONE;
1148 ret = wq_sleep(info, RECV, timeout, &wait);
1149 msg_ptr = wait.msg;
1150 }
1151 } else {
1152 msg_ptr = msg_get(info);
1153
1154 inode->i_atime = inode->i_mtime = inode->i_ctime =
1155 CURRENT_TIME;
1156
1157 /* There is now free space in queue. */
1158 pipelined_receive(info);
1159 spin_unlock(&info->lock);
1160 ret = 0;
1161 }
1162 if (ret == 0) {
1163 ret = msg_ptr->m_ts;
1164
1165 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1166 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1167 ret = -EFAULT;
1168 }
1169 free_msg(msg_ptr);
1170 }
1171out_fput:
1172 fdput(f);
1173out:
1174 return ret;
1175}
1176
1177/*
1178 * Notes: the case when user wants us to deregister (with NULL as pointer)
1179 * and he isn't currently owner of notification, will be silently discarded.
1180 * It isn't explicitly defined in the POSIX.
1181 */
1182SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1183 const struct sigevent __user *, u_notification)
1184{
1185 int ret;
1186 struct fd f;
1187 struct sock *sock;
1188 struct inode *inode;
1189 struct sigevent notification;
1190 struct mqueue_inode_info *info;
1191 struct sk_buff *nc;
1192
1193 if (u_notification) {
1194 if (copy_from_user(¬ification, u_notification,
1195 sizeof(struct sigevent)))
1196 return -EFAULT;
1197 }
1198
1199 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1200
1201 nc = NULL;
1202 sock = NULL;
1203 if (u_notification != NULL) {
1204 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1205 notification.sigev_notify != SIGEV_SIGNAL &&
1206 notification.sigev_notify != SIGEV_THREAD))
1207 return -EINVAL;
1208 if (notification.sigev_notify == SIGEV_SIGNAL &&
1209 !valid_signal(notification.sigev_signo)) {
1210 return -EINVAL;
1211 }
1212 if (notification.sigev_notify == SIGEV_THREAD) {
1213 long timeo;
1214
1215 /* create the notify skb */
1216 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1217 if (!nc) {
1218 ret = -ENOMEM;
1219 goto out;
1220 }
1221 if (copy_from_user(nc->data,
1222 notification.sigev_value.sival_ptr,
1223 NOTIFY_COOKIE_LEN)) {
1224 ret = -EFAULT;
1225 goto out;
1226 }
1227
1228 /* TODO: add a header? */
1229 skb_put(nc, NOTIFY_COOKIE_LEN);
1230 /* and attach it to the socket */
1231retry:
1232 f = fdget(notification.sigev_signo);
1233 if (!f.file) {
1234 ret = -EBADF;
1235 goto out;
1236 }
1237 sock = netlink_getsockbyfilp(f.file);
1238 fdput(f);
1239 if (IS_ERR(sock)) {
1240 ret = PTR_ERR(sock);
1241 sock = NULL;
1242 goto out;
1243 }
1244
1245 timeo = MAX_SCHEDULE_TIMEOUT;
1246 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1247 if (ret == 1)
1248 goto retry;
1249 if (ret) {
1250 sock = NULL;
1251 nc = NULL;
1252 goto out;
1253 }
1254 }
1255 }
1256
1257 f = fdget(mqdes);
1258 if (!f.file) {
1259 ret = -EBADF;
1260 goto out;
1261 }
1262
1263 inode = file_inode(f.file);
1264 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1265 ret = -EBADF;
1266 goto out_fput;
1267 }
1268 info = MQUEUE_I(inode);
1269
1270 ret = 0;
1271 spin_lock(&info->lock);
1272 if (u_notification == NULL) {
1273 if (info->notify_owner == task_tgid(current)) {
1274 remove_notification(info);
1275 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1276 }
1277 } else if (info->notify_owner != NULL) {
1278 ret = -EBUSY;
1279 } else {
1280 switch (notification.sigev_notify) {
1281 case SIGEV_NONE:
1282 info->notify.sigev_notify = SIGEV_NONE;
1283 break;
1284 case SIGEV_THREAD:
1285 info->notify_sock = sock;
1286 info->notify_cookie = nc;
1287 sock = NULL;
1288 nc = NULL;
1289 info->notify.sigev_notify = SIGEV_THREAD;
1290 break;
1291 case SIGEV_SIGNAL:
1292 info->notify.sigev_signo = notification.sigev_signo;
1293 info->notify.sigev_value = notification.sigev_value;
1294 info->notify.sigev_notify = SIGEV_SIGNAL;
1295 break;
1296 }
1297
1298 info->notify_owner = get_pid(task_tgid(current));
1299 info->notify_user_ns = get_user_ns(current_user_ns());
1300 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1301 }
1302 spin_unlock(&info->lock);
1303out_fput:
1304 fdput(f);
1305out:
1306 if (sock)
1307 netlink_detachskb(sock, nc);
1308 else if (nc)
1309 dev_kfree_skb(nc);
1310
1311 return ret;
1312}
1313
1314SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1315 const struct mq_attr __user *, u_mqstat,
1316 struct mq_attr __user *, u_omqstat)
1317{
1318 int ret;
1319 struct mq_attr mqstat, omqstat;
1320 struct fd f;
1321 struct inode *inode;
1322 struct mqueue_inode_info *info;
1323
1324 if (u_mqstat != NULL) {
1325 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1326 return -EFAULT;
1327 if (mqstat.mq_flags & (~O_NONBLOCK))
1328 return -EINVAL;
1329 }
1330
1331 f = fdget(mqdes);
1332 if (!f.file) {
1333 ret = -EBADF;
1334 goto out;
1335 }
1336
1337 inode = file_inode(f.file);
1338 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1339 ret = -EBADF;
1340 goto out_fput;
1341 }
1342 info = MQUEUE_I(inode);
1343
1344 spin_lock(&info->lock);
1345
1346 omqstat = info->attr;
1347 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1348 if (u_mqstat) {
1349 audit_mq_getsetattr(mqdes, &mqstat);
1350 spin_lock(&f.file->f_lock);
1351 if (mqstat.mq_flags & O_NONBLOCK)
1352 f.file->f_flags |= O_NONBLOCK;
1353 else
1354 f.file->f_flags &= ~O_NONBLOCK;
1355 spin_unlock(&f.file->f_lock);
1356
1357 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1358 }
1359
1360 spin_unlock(&info->lock);
1361
1362 ret = 0;
1363 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1364 sizeof(struct mq_attr)))
1365 ret = -EFAULT;
1366
1367out_fput:
1368 fdput(f);
1369out:
1370 return ret;
1371}
1372
1373static const struct inode_operations mqueue_dir_inode_operations = {
1374 .lookup = simple_lookup,
1375 .create = mqueue_create,
1376 .unlink = mqueue_unlink,
1377};
1378
1379static const struct file_operations mqueue_file_operations = {
1380 .flush = mqueue_flush_file,
1381 .poll = mqueue_poll_file,
1382 .read = mqueue_read_file,
1383 .llseek = default_llseek,
1384};
1385
1386static const struct super_operations mqueue_super_ops = {
1387 .alloc_inode = mqueue_alloc_inode,
1388 .destroy_inode = mqueue_destroy_inode,
1389 .evict_inode = mqueue_evict_inode,
1390 .statfs = simple_statfs,
1391};
1392
1393static struct file_system_type mqueue_fs_type = {
1394 .name = "mqueue",
1395 .mount = mqueue_mount,
1396 .kill_sb = kill_litter_super,
1397 .fs_flags = FS_USERNS_MOUNT,
1398};
1399
1400int mq_init_ns(struct ipc_namespace *ns)
1401{
1402 ns->mq_queues_count = 0;
1403 ns->mq_queues_max = DFLT_QUEUESMAX;
1404 ns->mq_msg_max = DFLT_MSGMAX;
1405 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1406 ns->mq_msg_default = DFLT_MSG;
1407 ns->mq_msgsize_default = DFLT_MSGSIZE;
1408
1409 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1410 if (IS_ERR(ns->mq_mnt)) {
1411 int err = PTR_ERR(ns->mq_mnt);
1412 ns->mq_mnt = NULL;
1413 return err;
1414 }
1415 return 0;
1416}
1417
1418void mq_clear_sbinfo(struct ipc_namespace *ns)
1419{
1420 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1421}
1422
1423void mq_put_mnt(struct ipc_namespace *ns)
1424{
1425 kern_unmount(ns->mq_mnt);
1426}
1427
1428static int __init init_mqueue_fs(void)
1429{
1430 int error;
1431
1432 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1433 sizeof(struct mqueue_inode_info), 0,
1434 SLAB_HWCACHE_ALIGN, init_once);
1435 if (mqueue_inode_cachep == NULL)
1436 return -ENOMEM;
1437
1438 /* ignore failures - they are not fatal */
1439 mq_sysctl_table = mq_register_sysctl_table();
1440
1441 error = register_filesystem(&mqueue_fs_type);
1442 if (error)
1443 goto out_sysctl;
1444
1445 spin_lock_init(&mq_lock);
1446
1447 error = mq_init_ns(&init_ipc_ns);
1448 if (error)
1449 goto out_filesystem;
1450
1451 return 0;
1452
1453out_filesystem:
1454 unregister_filesystem(&mqueue_fs_type);
1455out_sysctl:
1456 if (mq_sysctl_table)
1457 unregister_sysctl_table(mq_sysctl_table);
1458 kmem_cache_destroy(mqueue_inode_cachep);
1459 return error;
1460}
1461
1462device_initcall(init_mqueue_fs);
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/fs_context.h>
22#include <linux/namei.h>
23#include <linux/sysctl.h>
24#include <linux/poll.h>
25#include <linux/mqueue.h>
26#include <linux/msg.h>
27#include <linux/skbuff.h>
28#include <linux/vmalloc.h>
29#include <linux/netlink.h>
30#include <linux/syscalls.h>
31#include <linux/audit.h>
32#include <linux/signal.h>
33#include <linux/mutex.h>
34#include <linux/nsproxy.h>
35#include <linux/pid.h>
36#include <linux/ipc_namespace.h>
37#include <linux/user_namespace.h>
38#include <linux/slab.h>
39#include <linux/sched/wake_q.h>
40#include <linux/sched/signal.h>
41#include <linux/sched/user.h>
42
43#include <net/sock.h>
44#include "util.h"
45
46struct mqueue_fs_context {
47 struct ipc_namespace *ipc_ns;
48};
49
50#define MQUEUE_MAGIC 0x19800202
51#define DIRENT_SIZE 20
52#define FILENT_SIZE 80
53
54#define SEND 0
55#define RECV 1
56
57#define STATE_NONE 0
58#define STATE_READY 1
59
60struct posix_msg_tree_node {
61 struct rb_node rb_node;
62 struct list_head msg_list;
63 int priority;
64};
65
66/*
67 * Locking:
68 *
69 * Accesses to a message queue are synchronized by acquiring info->lock.
70 *
71 * There are two notable exceptions:
72 * - The actual wakeup of a sleeping task is performed using the wake_q
73 * framework. info->lock is already released when wake_up_q is called.
74 * - The exit codepaths after sleeping check ext_wait_queue->state without
75 * any locks. If it is STATE_READY, then the syscall is completed without
76 * acquiring info->lock.
77 *
78 * MQ_BARRIER:
79 * To achieve proper release/acquire memory barrier pairing, the state is set to
80 * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
81 * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
82 *
83 * This prevents the following races:
84 *
85 * 1) With the simple wake_q_add(), the task could be gone already before
86 * the increase of the reference happens
87 * Thread A
88 * Thread B
89 * WRITE_ONCE(wait.state, STATE_NONE);
90 * schedule_hrtimeout()
91 * wake_q_add(A)
92 * if (cmpxchg()) // success
93 * ->state = STATE_READY (reordered)
94 * <timeout returns>
95 * if (wait.state == STATE_READY) return;
96 * sysret to user space
97 * sys_exit()
98 * get_task_struct() // UaF
99 *
100 * Solution: Use wake_q_add_safe() and perform the get_task_struct() before
101 * the smp_store_release() that does ->state = STATE_READY.
102 *
103 * 2) Without proper _release/_acquire barriers, the woken up task
104 * could read stale data
105 *
106 * Thread A
107 * Thread B
108 * do_mq_timedreceive
109 * WRITE_ONCE(wait.state, STATE_NONE);
110 * schedule_hrtimeout()
111 * state = STATE_READY;
112 * <timeout returns>
113 * if (wait.state == STATE_READY) return;
114 * msg_ptr = wait.msg; // Access to stale data!
115 * receiver->msg = message; (reordered)
116 *
117 * Solution: use _release and _acquire barriers.
118 *
119 * 3) There is intentionally no barrier when setting current->state
120 * to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
121 * release memory barrier, and the wakeup is triggered when holding
122 * info->lock, i.e. spin_lock(&info->lock) provided a pairing
123 * acquire memory barrier.
124 */
125
126struct ext_wait_queue { /* queue of sleeping tasks */
127 struct task_struct *task;
128 struct list_head list;
129 struct msg_msg *msg; /* ptr of loaded message */
130 int state; /* one of STATE_* values */
131};
132
133struct mqueue_inode_info {
134 spinlock_t lock;
135 struct inode vfs_inode;
136 wait_queue_head_t wait_q;
137
138 struct rb_root msg_tree;
139 struct rb_node *msg_tree_rightmost;
140 struct posix_msg_tree_node *node_cache;
141 struct mq_attr attr;
142
143 struct sigevent notify;
144 struct pid *notify_owner;
145 u32 notify_self_exec_id;
146 struct user_namespace *notify_user_ns;
147 struct user_struct *user; /* user who created, for accounting */
148 struct sock *notify_sock;
149 struct sk_buff *notify_cookie;
150
151 /* for tasks waiting for free space and messages, respectively */
152 struct ext_wait_queue e_wait_q[2];
153
154 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
155};
156
157static struct file_system_type mqueue_fs_type;
158static const struct inode_operations mqueue_dir_inode_operations;
159static const struct file_operations mqueue_file_operations;
160static const struct super_operations mqueue_super_ops;
161static const struct fs_context_operations mqueue_fs_context_ops;
162static void remove_notification(struct mqueue_inode_info *info);
163
164static struct kmem_cache *mqueue_inode_cachep;
165
166static struct ctl_table_header *mq_sysctl_table;
167
168static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
169{
170 return container_of(inode, struct mqueue_inode_info, vfs_inode);
171}
172
173/*
174 * This routine should be called with the mq_lock held.
175 */
176static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
177{
178 return get_ipc_ns(inode->i_sb->s_fs_info);
179}
180
181static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
182{
183 struct ipc_namespace *ns;
184
185 spin_lock(&mq_lock);
186 ns = __get_ns_from_inode(inode);
187 spin_unlock(&mq_lock);
188 return ns;
189}
190
191/* Auxiliary functions to manipulate messages' list */
192static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
193{
194 struct rb_node **p, *parent = NULL;
195 struct posix_msg_tree_node *leaf;
196 bool rightmost = true;
197
198 p = &info->msg_tree.rb_node;
199 while (*p) {
200 parent = *p;
201 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
202
203 if (likely(leaf->priority == msg->m_type))
204 goto insert_msg;
205 else if (msg->m_type < leaf->priority) {
206 p = &(*p)->rb_left;
207 rightmost = false;
208 } else
209 p = &(*p)->rb_right;
210 }
211 if (info->node_cache) {
212 leaf = info->node_cache;
213 info->node_cache = NULL;
214 } else {
215 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
216 if (!leaf)
217 return -ENOMEM;
218 INIT_LIST_HEAD(&leaf->msg_list);
219 }
220 leaf->priority = msg->m_type;
221
222 if (rightmost)
223 info->msg_tree_rightmost = &leaf->rb_node;
224
225 rb_link_node(&leaf->rb_node, parent, p);
226 rb_insert_color(&leaf->rb_node, &info->msg_tree);
227insert_msg:
228 info->attr.mq_curmsgs++;
229 info->qsize += msg->m_ts;
230 list_add_tail(&msg->m_list, &leaf->msg_list);
231 return 0;
232}
233
234static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
235 struct mqueue_inode_info *info)
236{
237 struct rb_node *node = &leaf->rb_node;
238
239 if (info->msg_tree_rightmost == node)
240 info->msg_tree_rightmost = rb_prev(node);
241
242 rb_erase(node, &info->msg_tree);
243 if (info->node_cache)
244 kfree(leaf);
245 else
246 info->node_cache = leaf;
247}
248
249static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
250{
251 struct rb_node *parent = NULL;
252 struct posix_msg_tree_node *leaf;
253 struct msg_msg *msg;
254
255try_again:
256 /*
257 * During insert, low priorities go to the left and high to the
258 * right. On receive, we want the highest priorities first, so
259 * walk all the way to the right.
260 */
261 parent = info->msg_tree_rightmost;
262 if (!parent) {
263 if (info->attr.mq_curmsgs) {
264 pr_warn_once("Inconsistency in POSIX message queue, "
265 "no tree element, but supposedly messages "
266 "should exist!\n");
267 info->attr.mq_curmsgs = 0;
268 }
269 return NULL;
270 }
271 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
272 if (unlikely(list_empty(&leaf->msg_list))) {
273 pr_warn_once("Inconsistency in POSIX message queue, "
274 "empty leaf node but we haven't implemented "
275 "lazy leaf delete!\n");
276 msg_tree_erase(leaf, info);
277 goto try_again;
278 } else {
279 msg = list_first_entry(&leaf->msg_list,
280 struct msg_msg, m_list);
281 list_del(&msg->m_list);
282 if (list_empty(&leaf->msg_list)) {
283 msg_tree_erase(leaf, info);
284 }
285 }
286 info->attr.mq_curmsgs--;
287 info->qsize -= msg->m_ts;
288 return msg;
289}
290
291static struct inode *mqueue_get_inode(struct super_block *sb,
292 struct ipc_namespace *ipc_ns, umode_t mode,
293 struct mq_attr *attr)
294{
295 struct user_struct *u = current_user();
296 struct inode *inode;
297 int ret = -ENOMEM;
298
299 inode = new_inode(sb);
300 if (!inode)
301 goto err;
302
303 inode->i_ino = get_next_ino();
304 inode->i_mode = mode;
305 inode->i_uid = current_fsuid();
306 inode->i_gid = current_fsgid();
307 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
308
309 if (S_ISREG(mode)) {
310 struct mqueue_inode_info *info;
311 unsigned long mq_bytes, mq_treesize;
312
313 inode->i_fop = &mqueue_file_operations;
314 inode->i_size = FILENT_SIZE;
315 /* mqueue specific info */
316 info = MQUEUE_I(inode);
317 spin_lock_init(&info->lock);
318 init_waitqueue_head(&info->wait_q);
319 INIT_LIST_HEAD(&info->e_wait_q[0].list);
320 INIT_LIST_HEAD(&info->e_wait_q[1].list);
321 info->notify_owner = NULL;
322 info->notify_user_ns = NULL;
323 info->qsize = 0;
324 info->user = NULL; /* set when all is ok */
325 info->msg_tree = RB_ROOT;
326 info->msg_tree_rightmost = NULL;
327 info->node_cache = NULL;
328 memset(&info->attr, 0, sizeof(info->attr));
329 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
330 ipc_ns->mq_msg_default);
331 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
332 ipc_ns->mq_msgsize_default);
333 if (attr) {
334 info->attr.mq_maxmsg = attr->mq_maxmsg;
335 info->attr.mq_msgsize = attr->mq_msgsize;
336 }
337 /*
338 * We used to allocate a static array of pointers and account
339 * the size of that array as well as one msg_msg struct per
340 * possible message into the queue size. That's no longer
341 * accurate as the queue is now an rbtree and will grow and
342 * shrink depending on usage patterns. We can, however, still
343 * account one msg_msg struct per message, but the nodes are
344 * allocated depending on priority usage, and most programs
345 * only use one, or a handful, of priorities. However, since
346 * this is pinned memory, we need to assume worst case, so
347 * that means the min(mq_maxmsg, max_priorities) * struct
348 * posix_msg_tree_node.
349 */
350
351 ret = -EINVAL;
352 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
353 goto out_inode;
354 if (capable(CAP_SYS_RESOURCE)) {
355 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
356 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
357 goto out_inode;
358 } else {
359 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
360 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
361 goto out_inode;
362 }
363 ret = -EOVERFLOW;
364 /* check for overflow */
365 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
366 goto out_inode;
367 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
368 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
369 sizeof(struct posix_msg_tree_node);
370 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
371 if (mq_bytes + mq_treesize < mq_bytes)
372 goto out_inode;
373 mq_bytes += mq_treesize;
374 spin_lock(&mq_lock);
375 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
376 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
377 spin_unlock(&mq_lock);
378 /* mqueue_evict_inode() releases info->messages */
379 ret = -EMFILE;
380 goto out_inode;
381 }
382 u->mq_bytes += mq_bytes;
383 spin_unlock(&mq_lock);
384
385 /* all is ok */
386 info->user = get_uid(u);
387 } else if (S_ISDIR(mode)) {
388 inc_nlink(inode);
389 /* Some things misbehave if size == 0 on a directory */
390 inode->i_size = 2 * DIRENT_SIZE;
391 inode->i_op = &mqueue_dir_inode_operations;
392 inode->i_fop = &simple_dir_operations;
393 }
394
395 return inode;
396out_inode:
397 iput(inode);
398err:
399 return ERR_PTR(ret);
400}
401
402static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
403{
404 struct inode *inode;
405 struct ipc_namespace *ns = sb->s_fs_info;
406
407 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
408 sb->s_blocksize = PAGE_SIZE;
409 sb->s_blocksize_bits = PAGE_SHIFT;
410 sb->s_magic = MQUEUE_MAGIC;
411 sb->s_op = &mqueue_super_ops;
412
413 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
414 if (IS_ERR(inode))
415 return PTR_ERR(inode);
416
417 sb->s_root = d_make_root(inode);
418 if (!sb->s_root)
419 return -ENOMEM;
420 return 0;
421}
422
423static int mqueue_get_tree(struct fs_context *fc)
424{
425 struct mqueue_fs_context *ctx = fc->fs_private;
426
427 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
428}
429
430static void mqueue_fs_context_free(struct fs_context *fc)
431{
432 struct mqueue_fs_context *ctx = fc->fs_private;
433
434 put_ipc_ns(ctx->ipc_ns);
435 kfree(ctx);
436}
437
438static int mqueue_init_fs_context(struct fs_context *fc)
439{
440 struct mqueue_fs_context *ctx;
441
442 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
443 if (!ctx)
444 return -ENOMEM;
445
446 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
447 put_user_ns(fc->user_ns);
448 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
449 fc->fs_private = ctx;
450 fc->ops = &mqueue_fs_context_ops;
451 return 0;
452}
453
454static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
455{
456 struct mqueue_fs_context *ctx;
457 struct fs_context *fc;
458 struct vfsmount *mnt;
459
460 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
461 if (IS_ERR(fc))
462 return ERR_CAST(fc);
463
464 ctx = fc->fs_private;
465 put_ipc_ns(ctx->ipc_ns);
466 ctx->ipc_ns = get_ipc_ns(ns);
467 put_user_ns(fc->user_ns);
468 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
469
470 mnt = fc_mount(fc);
471 put_fs_context(fc);
472 return mnt;
473}
474
475static void init_once(void *foo)
476{
477 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
478
479 inode_init_once(&p->vfs_inode);
480}
481
482static struct inode *mqueue_alloc_inode(struct super_block *sb)
483{
484 struct mqueue_inode_info *ei;
485
486 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
487 if (!ei)
488 return NULL;
489 return &ei->vfs_inode;
490}
491
492static void mqueue_free_inode(struct inode *inode)
493{
494 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
495}
496
497static void mqueue_evict_inode(struct inode *inode)
498{
499 struct mqueue_inode_info *info;
500 struct user_struct *user;
501 struct ipc_namespace *ipc_ns;
502 struct msg_msg *msg, *nmsg;
503 LIST_HEAD(tmp_msg);
504
505 clear_inode(inode);
506
507 if (S_ISDIR(inode->i_mode))
508 return;
509
510 ipc_ns = get_ns_from_inode(inode);
511 info = MQUEUE_I(inode);
512 spin_lock(&info->lock);
513 while ((msg = msg_get(info)) != NULL)
514 list_add_tail(&msg->m_list, &tmp_msg);
515 kfree(info->node_cache);
516 spin_unlock(&info->lock);
517
518 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
519 list_del(&msg->m_list);
520 free_msg(msg);
521 }
522
523 user = info->user;
524 if (user) {
525 unsigned long mq_bytes, mq_treesize;
526
527 /* Total amount of bytes accounted for the mqueue */
528 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
529 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
530 sizeof(struct posix_msg_tree_node);
531
532 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
533 info->attr.mq_msgsize);
534
535 spin_lock(&mq_lock);
536 user->mq_bytes -= mq_bytes;
537 /*
538 * get_ns_from_inode() ensures that the
539 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
540 * to which we now hold a reference, or it is NULL.
541 * We can't put it here under mq_lock, though.
542 */
543 if (ipc_ns)
544 ipc_ns->mq_queues_count--;
545 spin_unlock(&mq_lock);
546 free_uid(user);
547 }
548 if (ipc_ns)
549 put_ipc_ns(ipc_ns);
550}
551
552static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
553{
554 struct inode *dir = dentry->d_parent->d_inode;
555 struct inode *inode;
556 struct mq_attr *attr = arg;
557 int error;
558 struct ipc_namespace *ipc_ns;
559
560 spin_lock(&mq_lock);
561 ipc_ns = __get_ns_from_inode(dir);
562 if (!ipc_ns) {
563 error = -EACCES;
564 goto out_unlock;
565 }
566
567 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
568 !capable(CAP_SYS_RESOURCE)) {
569 error = -ENOSPC;
570 goto out_unlock;
571 }
572 ipc_ns->mq_queues_count++;
573 spin_unlock(&mq_lock);
574
575 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
576 if (IS_ERR(inode)) {
577 error = PTR_ERR(inode);
578 spin_lock(&mq_lock);
579 ipc_ns->mq_queues_count--;
580 goto out_unlock;
581 }
582
583 put_ipc_ns(ipc_ns);
584 dir->i_size += DIRENT_SIZE;
585 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
586
587 d_instantiate(dentry, inode);
588 dget(dentry);
589 return 0;
590out_unlock:
591 spin_unlock(&mq_lock);
592 if (ipc_ns)
593 put_ipc_ns(ipc_ns);
594 return error;
595}
596
597static int mqueue_create(struct inode *dir, struct dentry *dentry,
598 umode_t mode, bool excl)
599{
600 return mqueue_create_attr(dentry, mode, NULL);
601}
602
603static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
604{
605 struct inode *inode = d_inode(dentry);
606
607 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
608 dir->i_size -= DIRENT_SIZE;
609 drop_nlink(inode);
610 dput(dentry);
611 return 0;
612}
613
614/*
615* This is routine for system read from queue file.
616* To avoid mess with doing here some sort of mq_receive we allow
617* to read only queue size & notification info (the only values
618* that are interesting from user point of view and aren't accessible
619* through std routines)
620*/
621static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
622 size_t count, loff_t *off)
623{
624 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
625 char buffer[FILENT_SIZE];
626 ssize_t ret;
627
628 spin_lock(&info->lock);
629 snprintf(buffer, sizeof(buffer),
630 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
631 info->qsize,
632 info->notify_owner ? info->notify.sigev_notify : 0,
633 (info->notify_owner &&
634 info->notify.sigev_notify == SIGEV_SIGNAL) ?
635 info->notify.sigev_signo : 0,
636 pid_vnr(info->notify_owner));
637 spin_unlock(&info->lock);
638 buffer[sizeof(buffer)-1] = '\0';
639
640 ret = simple_read_from_buffer(u_data, count, off, buffer,
641 strlen(buffer));
642 if (ret <= 0)
643 return ret;
644
645 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
646 return ret;
647}
648
649static int mqueue_flush_file(struct file *filp, fl_owner_t id)
650{
651 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
652
653 spin_lock(&info->lock);
654 if (task_tgid(current) == info->notify_owner)
655 remove_notification(info);
656
657 spin_unlock(&info->lock);
658 return 0;
659}
660
661static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
662{
663 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
664 __poll_t retval = 0;
665
666 poll_wait(filp, &info->wait_q, poll_tab);
667
668 spin_lock(&info->lock);
669 if (info->attr.mq_curmsgs)
670 retval = EPOLLIN | EPOLLRDNORM;
671
672 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
673 retval |= EPOLLOUT | EPOLLWRNORM;
674 spin_unlock(&info->lock);
675
676 return retval;
677}
678
679/* Adds current to info->e_wait_q[sr] before element with smaller prio */
680static void wq_add(struct mqueue_inode_info *info, int sr,
681 struct ext_wait_queue *ewp)
682{
683 struct ext_wait_queue *walk;
684
685 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
686 if (walk->task->prio <= current->prio) {
687 list_add_tail(&ewp->list, &walk->list);
688 return;
689 }
690 }
691 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
692}
693
694/*
695 * Puts current task to sleep. Caller must hold queue lock. After return
696 * lock isn't held.
697 * sr: SEND or RECV
698 */
699static int wq_sleep(struct mqueue_inode_info *info, int sr,
700 ktime_t *timeout, struct ext_wait_queue *ewp)
701 __releases(&info->lock)
702{
703 int retval;
704 signed long time;
705
706 wq_add(info, sr, ewp);
707
708 for (;;) {
709 /* memory barrier not required, we hold info->lock */
710 __set_current_state(TASK_INTERRUPTIBLE);
711
712 spin_unlock(&info->lock);
713 time = schedule_hrtimeout_range_clock(timeout, 0,
714 HRTIMER_MODE_ABS, CLOCK_REALTIME);
715
716 if (READ_ONCE(ewp->state) == STATE_READY) {
717 /* see MQ_BARRIER for purpose/pairing */
718 smp_acquire__after_ctrl_dep();
719 retval = 0;
720 goto out;
721 }
722 spin_lock(&info->lock);
723
724 /* we hold info->lock, so no memory barrier required */
725 if (READ_ONCE(ewp->state) == STATE_READY) {
726 retval = 0;
727 goto out_unlock;
728 }
729 if (signal_pending(current)) {
730 retval = -ERESTARTSYS;
731 break;
732 }
733 if (time == 0) {
734 retval = -ETIMEDOUT;
735 break;
736 }
737 }
738 list_del(&ewp->list);
739out_unlock:
740 spin_unlock(&info->lock);
741out:
742 return retval;
743}
744
745/*
746 * Returns waiting task that should be serviced first or NULL if none exists
747 */
748static struct ext_wait_queue *wq_get_first_waiter(
749 struct mqueue_inode_info *info, int sr)
750{
751 struct list_head *ptr;
752
753 ptr = info->e_wait_q[sr].list.prev;
754 if (ptr == &info->e_wait_q[sr].list)
755 return NULL;
756 return list_entry(ptr, struct ext_wait_queue, list);
757}
758
759
760static inline void set_cookie(struct sk_buff *skb, char code)
761{
762 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
763}
764
765/*
766 * The next function is only to split too long sys_mq_timedsend
767 */
768static void __do_notify(struct mqueue_inode_info *info)
769{
770 /* notification
771 * invoked when there is registered process and there isn't process
772 * waiting synchronously for message AND state of queue changed from
773 * empty to not empty. Here we are sure that no one is waiting
774 * synchronously. */
775 if (info->notify_owner &&
776 info->attr.mq_curmsgs == 1) {
777 switch (info->notify.sigev_notify) {
778 case SIGEV_NONE:
779 break;
780 case SIGEV_SIGNAL: {
781 struct kernel_siginfo sig_i;
782 struct task_struct *task;
783
784 /* do_mq_notify() accepts sigev_signo == 0, why?? */
785 if (!info->notify.sigev_signo)
786 break;
787
788 clear_siginfo(&sig_i);
789 sig_i.si_signo = info->notify.sigev_signo;
790 sig_i.si_errno = 0;
791 sig_i.si_code = SI_MESGQ;
792 sig_i.si_value = info->notify.sigev_value;
793 rcu_read_lock();
794 /* map current pid/uid into info->owner's namespaces */
795 sig_i.si_pid = task_tgid_nr_ns(current,
796 ns_of_pid(info->notify_owner));
797 sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
798 current_uid());
799 /*
800 * We can't use kill_pid_info(), this signal should
801 * bypass check_kill_permission(). It is from kernel
802 * but si_fromuser() can't know this.
803 * We do check the self_exec_id, to avoid sending
804 * signals to programs that don't expect them.
805 */
806 task = pid_task(info->notify_owner, PIDTYPE_TGID);
807 if (task && task->self_exec_id ==
808 info->notify_self_exec_id) {
809 do_send_sig_info(info->notify.sigev_signo,
810 &sig_i, task, PIDTYPE_TGID);
811 }
812 rcu_read_unlock();
813 break;
814 }
815 case SIGEV_THREAD:
816 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
817 netlink_sendskb(info->notify_sock, info->notify_cookie);
818 break;
819 }
820 /* after notification unregisters process */
821 put_pid(info->notify_owner);
822 put_user_ns(info->notify_user_ns);
823 info->notify_owner = NULL;
824 info->notify_user_ns = NULL;
825 }
826 wake_up(&info->wait_q);
827}
828
829static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
830 struct timespec64 *ts)
831{
832 if (get_timespec64(ts, u_abs_timeout))
833 return -EFAULT;
834 if (!timespec64_valid(ts))
835 return -EINVAL;
836 return 0;
837}
838
839static void remove_notification(struct mqueue_inode_info *info)
840{
841 if (info->notify_owner != NULL &&
842 info->notify.sigev_notify == SIGEV_THREAD) {
843 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
844 netlink_sendskb(info->notify_sock, info->notify_cookie);
845 }
846 put_pid(info->notify_owner);
847 put_user_ns(info->notify_user_ns);
848 info->notify_owner = NULL;
849 info->notify_user_ns = NULL;
850}
851
852static int prepare_open(struct dentry *dentry, int oflag, int ro,
853 umode_t mode, struct filename *name,
854 struct mq_attr *attr)
855{
856 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
857 MAY_READ | MAY_WRITE };
858 int acc;
859
860 if (d_really_is_negative(dentry)) {
861 if (!(oflag & O_CREAT))
862 return -ENOENT;
863 if (ro)
864 return ro;
865 audit_inode_parent_hidden(name, dentry->d_parent);
866 return vfs_mkobj(dentry, mode & ~current_umask(),
867 mqueue_create_attr, attr);
868 }
869 /* it already existed */
870 audit_inode(name, dentry, 0);
871 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
872 return -EEXIST;
873 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
874 return -EINVAL;
875 acc = oflag2acc[oflag & O_ACCMODE];
876 return inode_permission(d_inode(dentry), acc);
877}
878
879static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
880 struct mq_attr *attr)
881{
882 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
883 struct dentry *root = mnt->mnt_root;
884 struct filename *name;
885 struct path path;
886 int fd, error;
887 int ro;
888
889 audit_mq_open(oflag, mode, attr);
890
891 if (IS_ERR(name = getname(u_name)))
892 return PTR_ERR(name);
893
894 fd = get_unused_fd_flags(O_CLOEXEC);
895 if (fd < 0)
896 goto out_putname;
897
898 ro = mnt_want_write(mnt); /* we'll drop it in any case */
899 inode_lock(d_inode(root));
900 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
901 if (IS_ERR(path.dentry)) {
902 error = PTR_ERR(path.dentry);
903 goto out_putfd;
904 }
905 path.mnt = mntget(mnt);
906 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
907 if (!error) {
908 struct file *file = dentry_open(&path, oflag, current_cred());
909 if (!IS_ERR(file))
910 fd_install(fd, file);
911 else
912 error = PTR_ERR(file);
913 }
914 path_put(&path);
915out_putfd:
916 if (error) {
917 put_unused_fd(fd);
918 fd = error;
919 }
920 inode_unlock(d_inode(root));
921 if (!ro)
922 mnt_drop_write(mnt);
923out_putname:
924 putname(name);
925 return fd;
926}
927
928SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
929 struct mq_attr __user *, u_attr)
930{
931 struct mq_attr attr;
932 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
933 return -EFAULT;
934
935 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
936}
937
938SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
939{
940 int err;
941 struct filename *name;
942 struct dentry *dentry;
943 struct inode *inode = NULL;
944 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
945 struct vfsmount *mnt = ipc_ns->mq_mnt;
946
947 name = getname(u_name);
948 if (IS_ERR(name))
949 return PTR_ERR(name);
950
951 audit_inode_parent_hidden(name, mnt->mnt_root);
952 err = mnt_want_write(mnt);
953 if (err)
954 goto out_name;
955 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
956 dentry = lookup_one_len(name->name, mnt->mnt_root,
957 strlen(name->name));
958 if (IS_ERR(dentry)) {
959 err = PTR_ERR(dentry);
960 goto out_unlock;
961 }
962
963 inode = d_inode(dentry);
964 if (!inode) {
965 err = -ENOENT;
966 } else {
967 ihold(inode);
968 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
969 }
970 dput(dentry);
971
972out_unlock:
973 inode_unlock(d_inode(mnt->mnt_root));
974 if (inode)
975 iput(inode);
976 mnt_drop_write(mnt);
977out_name:
978 putname(name);
979
980 return err;
981}
982
983/* Pipelined send and receive functions.
984 *
985 * If a receiver finds no waiting message, then it registers itself in the
986 * list of waiting receivers. A sender checks that list before adding the new
987 * message into the message array. If there is a waiting receiver, then it
988 * bypasses the message array and directly hands the message over to the
989 * receiver. The receiver accepts the message and returns without grabbing the
990 * queue spinlock:
991 *
992 * - Set pointer to message.
993 * - Queue the receiver task for later wakeup (without the info->lock).
994 * - Update its state to STATE_READY. Now the receiver can continue.
995 * - Wake up the process after the lock is dropped. Should the process wake up
996 * before this wakeup (due to a timeout or a signal) it will either see
997 * STATE_READY and continue or acquire the lock to check the state again.
998 *
999 * The same algorithm is used for senders.
1000 */
1001
1002static inline void __pipelined_op(struct wake_q_head *wake_q,
1003 struct mqueue_inode_info *info,
1004 struct ext_wait_queue *this)
1005{
1006 list_del(&this->list);
1007 get_task_struct(this->task);
1008
1009 /* see MQ_BARRIER for purpose/pairing */
1010 smp_store_release(&this->state, STATE_READY);
1011 wake_q_add_safe(wake_q, this->task);
1012}
1013
1014/* pipelined_send() - send a message directly to the task waiting in
1015 * sys_mq_timedreceive() (without inserting message into a queue).
1016 */
1017static inline void pipelined_send(struct wake_q_head *wake_q,
1018 struct mqueue_inode_info *info,
1019 struct msg_msg *message,
1020 struct ext_wait_queue *receiver)
1021{
1022 receiver->msg = message;
1023 __pipelined_op(wake_q, info, receiver);
1024}
1025
1026/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1027 * gets its message and put to the queue (we have one free place for sure). */
1028static inline void pipelined_receive(struct wake_q_head *wake_q,
1029 struct mqueue_inode_info *info)
1030{
1031 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1032
1033 if (!sender) {
1034 /* for poll */
1035 wake_up_interruptible(&info->wait_q);
1036 return;
1037 }
1038 if (msg_insert(sender->msg, info))
1039 return;
1040
1041 __pipelined_op(wake_q, info, sender);
1042}
1043
1044static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1045 size_t msg_len, unsigned int msg_prio,
1046 struct timespec64 *ts)
1047{
1048 struct fd f;
1049 struct inode *inode;
1050 struct ext_wait_queue wait;
1051 struct ext_wait_queue *receiver;
1052 struct msg_msg *msg_ptr;
1053 struct mqueue_inode_info *info;
1054 ktime_t expires, *timeout = NULL;
1055 struct posix_msg_tree_node *new_leaf = NULL;
1056 int ret = 0;
1057 DEFINE_WAKE_Q(wake_q);
1058
1059 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1060 return -EINVAL;
1061
1062 if (ts) {
1063 expires = timespec64_to_ktime(*ts);
1064 timeout = &expires;
1065 }
1066
1067 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1068
1069 f = fdget(mqdes);
1070 if (unlikely(!f.file)) {
1071 ret = -EBADF;
1072 goto out;
1073 }
1074
1075 inode = file_inode(f.file);
1076 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1077 ret = -EBADF;
1078 goto out_fput;
1079 }
1080 info = MQUEUE_I(inode);
1081 audit_file(f.file);
1082
1083 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1084 ret = -EBADF;
1085 goto out_fput;
1086 }
1087
1088 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1089 ret = -EMSGSIZE;
1090 goto out_fput;
1091 }
1092
1093 /* First try to allocate memory, before doing anything with
1094 * existing queues. */
1095 msg_ptr = load_msg(u_msg_ptr, msg_len);
1096 if (IS_ERR(msg_ptr)) {
1097 ret = PTR_ERR(msg_ptr);
1098 goto out_fput;
1099 }
1100 msg_ptr->m_ts = msg_len;
1101 msg_ptr->m_type = msg_prio;
1102
1103 /*
1104 * msg_insert really wants us to have a valid, spare node struct so
1105 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1106 * fall back to that if necessary.
1107 */
1108 if (!info->node_cache)
1109 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1110
1111 spin_lock(&info->lock);
1112
1113 if (!info->node_cache && new_leaf) {
1114 /* Save our speculative allocation into the cache */
1115 INIT_LIST_HEAD(&new_leaf->msg_list);
1116 info->node_cache = new_leaf;
1117 new_leaf = NULL;
1118 } else {
1119 kfree(new_leaf);
1120 }
1121
1122 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1123 if (f.file->f_flags & O_NONBLOCK) {
1124 ret = -EAGAIN;
1125 } else {
1126 wait.task = current;
1127 wait.msg = (void *) msg_ptr;
1128
1129 /* memory barrier not required, we hold info->lock */
1130 WRITE_ONCE(wait.state, STATE_NONE);
1131 ret = wq_sleep(info, SEND, timeout, &wait);
1132 /*
1133 * wq_sleep must be called with info->lock held, and
1134 * returns with the lock released
1135 */
1136 goto out_free;
1137 }
1138 } else {
1139 receiver = wq_get_first_waiter(info, RECV);
1140 if (receiver) {
1141 pipelined_send(&wake_q, info, msg_ptr, receiver);
1142 } else {
1143 /* adds message to the queue */
1144 ret = msg_insert(msg_ptr, info);
1145 if (ret)
1146 goto out_unlock;
1147 __do_notify(info);
1148 }
1149 inode->i_atime = inode->i_mtime = inode->i_ctime =
1150 current_time(inode);
1151 }
1152out_unlock:
1153 spin_unlock(&info->lock);
1154 wake_up_q(&wake_q);
1155out_free:
1156 if (ret)
1157 free_msg(msg_ptr);
1158out_fput:
1159 fdput(f);
1160out:
1161 return ret;
1162}
1163
1164static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1165 size_t msg_len, unsigned int __user *u_msg_prio,
1166 struct timespec64 *ts)
1167{
1168 ssize_t ret;
1169 struct msg_msg *msg_ptr;
1170 struct fd f;
1171 struct inode *inode;
1172 struct mqueue_inode_info *info;
1173 struct ext_wait_queue wait;
1174 ktime_t expires, *timeout = NULL;
1175 struct posix_msg_tree_node *new_leaf = NULL;
1176
1177 if (ts) {
1178 expires = timespec64_to_ktime(*ts);
1179 timeout = &expires;
1180 }
1181
1182 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1183
1184 f = fdget(mqdes);
1185 if (unlikely(!f.file)) {
1186 ret = -EBADF;
1187 goto out;
1188 }
1189
1190 inode = file_inode(f.file);
1191 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1192 ret = -EBADF;
1193 goto out_fput;
1194 }
1195 info = MQUEUE_I(inode);
1196 audit_file(f.file);
1197
1198 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1199 ret = -EBADF;
1200 goto out_fput;
1201 }
1202
1203 /* checks if buffer is big enough */
1204 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1205 ret = -EMSGSIZE;
1206 goto out_fput;
1207 }
1208
1209 /*
1210 * msg_insert really wants us to have a valid, spare node struct so
1211 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1212 * fall back to that if necessary.
1213 */
1214 if (!info->node_cache)
1215 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1216
1217 spin_lock(&info->lock);
1218
1219 if (!info->node_cache && new_leaf) {
1220 /* Save our speculative allocation into the cache */
1221 INIT_LIST_HEAD(&new_leaf->msg_list);
1222 info->node_cache = new_leaf;
1223 } else {
1224 kfree(new_leaf);
1225 }
1226
1227 if (info->attr.mq_curmsgs == 0) {
1228 if (f.file->f_flags & O_NONBLOCK) {
1229 spin_unlock(&info->lock);
1230 ret = -EAGAIN;
1231 } else {
1232 wait.task = current;
1233
1234 /* memory barrier not required, we hold info->lock */
1235 WRITE_ONCE(wait.state, STATE_NONE);
1236 ret = wq_sleep(info, RECV, timeout, &wait);
1237 msg_ptr = wait.msg;
1238 }
1239 } else {
1240 DEFINE_WAKE_Q(wake_q);
1241
1242 msg_ptr = msg_get(info);
1243
1244 inode->i_atime = inode->i_mtime = inode->i_ctime =
1245 current_time(inode);
1246
1247 /* There is now free space in queue. */
1248 pipelined_receive(&wake_q, info);
1249 spin_unlock(&info->lock);
1250 wake_up_q(&wake_q);
1251 ret = 0;
1252 }
1253 if (ret == 0) {
1254 ret = msg_ptr->m_ts;
1255
1256 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1257 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1258 ret = -EFAULT;
1259 }
1260 free_msg(msg_ptr);
1261 }
1262out_fput:
1263 fdput(f);
1264out:
1265 return ret;
1266}
1267
1268SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1269 size_t, msg_len, unsigned int, msg_prio,
1270 const struct __kernel_timespec __user *, u_abs_timeout)
1271{
1272 struct timespec64 ts, *p = NULL;
1273 if (u_abs_timeout) {
1274 int res = prepare_timeout(u_abs_timeout, &ts);
1275 if (res)
1276 return res;
1277 p = &ts;
1278 }
1279 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1280}
1281
1282SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1283 size_t, msg_len, unsigned int __user *, u_msg_prio,
1284 const struct __kernel_timespec __user *, u_abs_timeout)
1285{
1286 struct timespec64 ts, *p = NULL;
1287 if (u_abs_timeout) {
1288 int res = prepare_timeout(u_abs_timeout, &ts);
1289 if (res)
1290 return res;
1291 p = &ts;
1292 }
1293 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1294}
1295
1296/*
1297 * Notes: the case when user wants us to deregister (with NULL as pointer)
1298 * and he isn't currently owner of notification, will be silently discarded.
1299 * It isn't explicitly defined in the POSIX.
1300 */
1301static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1302{
1303 int ret;
1304 struct fd f;
1305 struct sock *sock;
1306 struct inode *inode;
1307 struct mqueue_inode_info *info;
1308 struct sk_buff *nc;
1309
1310 audit_mq_notify(mqdes, notification);
1311
1312 nc = NULL;
1313 sock = NULL;
1314 if (notification != NULL) {
1315 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1316 notification->sigev_notify != SIGEV_SIGNAL &&
1317 notification->sigev_notify != SIGEV_THREAD))
1318 return -EINVAL;
1319 if (notification->sigev_notify == SIGEV_SIGNAL &&
1320 !valid_signal(notification->sigev_signo)) {
1321 return -EINVAL;
1322 }
1323 if (notification->sigev_notify == SIGEV_THREAD) {
1324 long timeo;
1325
1326 /* create the notify skb */
1327 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1328 if (!nc)
1329 return -ENOMEM;
1330
1331 if (copy_from_user(nc->data,
1332 notification->sigev_value.sival_ptr,
1333 NOTIFY_COOKIE_LEN)) {
1334 ret = -EFAULT;
1335 goto free_skb;
1336 }
1337
1338 /* TODO: add a header? */
1339 skb_put(nc, NOTIFY_COOKIE_LEN);
1340 /* and attach it to the socket */
1341retry:
1342 f = fdget(notification->sigev_signo);
1343 if (!f.file) {
1344 ret = -EBADF;
1345 goto out;
1346 }
1347 sock = netlink_getsockbyfilp(f.file);
1348 fdput(f);
1349 if (IS_ERR(sock)) {
1350 ret = PTR_ERR(sock);
1351 goto free_skb;
1352 }
1353
1354 timeo = MAX_SCHEDULE_TIMEOUT;
1355 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1356 if (ret == 1) {
1357 sock = NULL;
1358 goto retry;
1359 }
1360 if (ret)
1361 return ret;
1362 }
1363 }
1364
1365 f = fdget(mqdes);
1366 if (!f.file) {
1367 ret = -EBADF;
1368 goto out;
1369 }
1370
1371 inode = file_inode(f.file);
1372 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1373 ret = -EBADF;
1374 goto out_fput;
1375 }
1376 info = MQUEUE_I(inode);
1377
1378 ret = 0;
1379 spin_lock(&info->lock);
1380 if (notification == NULL) {
1381 if (info->notify_owner == task_tgid(current)) {
1382 remove_notification(info);
1383 inode->i_atime = inode->i_ctime = current_time(inode);
1384 }
1385 } else if (info->notify_owner != NULL) {
1386 ret = -EBUSY;
1387 } else {
1388 switch (notification->sigev_notify) {
1389 case SIGEV_NONE:
1390 info->notify.sigev_notify = SIGEV_NONE;
1391 break;
1392 case SIGEV_THREAD:
1393 info->notify_sock = sock;
1394 info->notify_cookie = nc;
1395 sock = NULL;
1396 nc = NULL;
1397 info->notify.sigev_notify = SIGEV_THREAD;
1398 break;
1399 case SIGEV_SIGNAL:
1400 info->notify.sigev_signo = notification->sigev_signo;
1401 info->notify.sigev_value = notification->sigev_value;
1402 info->notify.sigev_notify = SIGEV_SIGNAL;
1403 info->notify_self_exec_id = current->self_exec_id;
1404 break;
1405 }
1406
1407 info->notify_owner = get_pid(task_tgid(current));
1408 info->notify_user_ns = get_user_ns(current_user_ns());
1409 inode->i_atime = inode->i_ctime = current_time(inode);
1410 }
1411 spin_unlock(&info->lock);
1412out_fput:
1413 fdput(f);
1414out:
1415 if (sock)
1416 netlink_detachskb(sock, nc);
1417 else
1418free_skb:
1419 dev_kfree_skb(nc);
1420
1421 return ret;
1422}
1423
1424SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1425 const struct sigevent __user *, u_notification)
1426{
1427 struct sigevent n, *p = NULL;
1428 if (u_notification) {
1429 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1430 return -EFAULT;
1431 p = &n;
1432 }
1433 return do_mq_notify(mqdes, p);
1434}
1435
1436static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1437{
1438 struct fd f;
1439 struct inode *inode;
1440 struct mqueue_inode_info *info;
1441
1442 if (new && (new->mq_flags & (~O_NONBLOCK)))
1443 return -EINVAL;
1444
1445 f = fdget(mqdes);
1446 if (!f.file)
1447 return -EBADF;
1448
1449 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1450 fdput(f);
1451 return -EBADF;
1452 }
1453
1454 inode = file_inode(f.file);
1455 info = MQUEUE_I(inode);
1456
1457 spin_lock(&info->lock);
1458
1459 if (old) {
1460 *old = info->attr;
1461 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1462 }
1463 if (new) {
1464 audit_mq_getsetattr(mqdes, new);
1465 spin_lock(&f.file->f_lock);
1466 if (new->mq_flags & O_NONBLOCK)
1467 f.file->f_flags |= O_NONBLOCK;
1468 else
1469 f.file->f_flags &= ~O_NONBLOCK;
1470 spin_unlock(&f.file->f_lock);
1471
1472 inode->i_atime = inode->i_ctime = current_time(inode);
1473 }
1474
1475 spin_unlock(&info->lock);
1476 fdput(f);
1477 return 0;
1478}
1479
1480SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1481 const struct mq_attr __user *, u_mqstat,
1482 struct mq_attr __user *, u_omqstat)
1483{
1484 int ret;
1485 struct mq_attr mqstat, omqstat;
1486 struct mq_attr *new = NULL, *old = NULL;
1487
1488 if (u_mqstat) {
1489 new = &mqstat;
1490 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1491 return -EFAULT;
1492 }
1493 if (u_omqstat)
1494 old = &omqstat;
1495
1496 ret = do_mq_getsetattr(mqdes, new, old);
1497 if (ret || !old)
1498 return ret;
1499
1500 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1501 return -EFAULT;
1502 return 0;
1503}
1504
1505#ifdef CONFIG_COMPAT
1506
1507struct compat_mq_attr {
1508 compat_long_t mq_flags; /* message queue flags */
1509 compat_long_t mq_maxmsg; /* maximum number of messages */
1510 compat_long_t mq_msgsize; /* maximum message size */
1511 compat_long_t mq_curmsgs; /* number of messages currently queued */
1512 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1513};
1514
1515static inline int get_compat_mq_attr(struct mq_attr *attr,
1516 const struct compat_mq_attr __user *uattr)
1517{
1518 struct compat_mq_attr v;
1519
1520 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1521 return -EFAULT;
1522
1523 memset(attr, 0, sizeof(*attr));
1524 attr->mq_flags = v.mq_flags;
1525 attr->mq_maxmsg = v.mq_maxmsg;
1526 attr->mq_msgsize = v.mq_msgsize;
1527 attr->mq_curmsgs = v.mq_curmsgs;
1528 return 0;
1529}
1530
1531static inline int put_compat_mq_attr(const struct mq_attr *attr,
1532 struct compat_mq_attr __user *uattr)
1533{
1534 struct compat_mq_attr v;
1535
1536 memset(&v, 0, sizeof(v));
1537 v.mq_flags = attr->mq_flags;
1538 v.mq_maxmsg = attr->mq_maxmsg;
1539 v.mq_msgsize = attr->mq_msgsize;
1540 v.mq_curmsgs = attr->mq_curmsgs;
1541 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1542 return -EFAULT;
1543 return 0;
1544}
1545
1546COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1547 int, oflag, compat_mode_t, mode,
1548 struct compat_mq_attr __user *, u_attr)
1549{
1550 struct mq_attr attr, *p = NULL;
1551 if (u_attr && oflag & O_CREAT) {
1552 p = &attr;
1553 if (get_compat_mq_attr(&attr, u_attr))
1554 return -EFAULT;
1555 }
1556 return do_mq_open(u_name, oflag, mode, p);
1557}
1558
1559COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1560 const struct compat_sigevent __user *, u_notification)
1561{
1562 struct sigevent n, *p = NULL;
1563 if (u_notification) {
1564 if (get_compat_sigevent(&n, u_notification))
1565 return -EFAULT;
1566 if (n.sigev_notify == SIGEV_THREAD)
1567 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1568 p = &n;
1569 }
1570 return do_mq_notify(mqdes, p);
1571}
1572
1573COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1574 const struct compat_mq_attr __user *, u_mqstat,
1575 struct compat_mq_attr __user *, u_omqstat)
1576{
1577 int ret;
1578 struct mq_attr mqstat, omqstat;
1579 struct mq_attr *new = NULL, *old = NULL;
1580
1581 if (u_mqstat) {
1582 new = &mqstat;
1583 if (get_compat_mq_attr(new, u_mqstat))
1584 return -EFAULT;
1585 }
1586 if (u_omqstat)
1587 old = &omqstat;
1588
1589 ret = do_mq_getsetattr(mqdes, new, old);
1590 if (ret || !old)
1591 return ret;
1592
1593 if (put_compat_mq_attr(old, u_omqstat))
1594 return -EFAULT;
1595 return 0;
1596}
1597#endif
1598
1599#ifdef CONFIG_COMPAT_32BIT_TIME
1600static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1601 struct timespec64 *ts)
1602{
1603 if (get_old_timespec32(ts, p))
1604 return -EFAULT;
1605 if (!timespec64_valid(ts))
1606 return -EINVAL;
1607 return 0;
1608}
1609
1610SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1611 const char __user *, u_msg_ptr,
1612 unsigned int, msg_len, unsigned int, msg_prio,
1613 const struct old_timespec32 __user *, u_abs_timeout)
1614{
1615 struct timespec64 ts, *p = NULL;
1616 if (u_abs_timeout) {
1617 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1618 if (res)
1619 return res;
1620 p = &ts;
1621 }
1622 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1623}
1624
1625SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1626 char __user *, u_msg_ptr,
1627 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1628 const struct old_timespec32 __user *, u_abs_timeout)
1629{
1630 struct timespec64 ts, *p = NULL;
1631 if (u_abs_timeout) {
1632 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1633 if (res)
1634 return res;
1635 p = &ts;
1636 }
1637 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1638}
1639#endif
1640
1641static const struct inode_operations mqueue_dir_inode_operations = {
1642 .lookup = simple_lookup,
1643 .create = mqueue_create,
1644 .unlink = mqueue_unlink,
1645};
1646
1647static const struct file_operations mqueue_file_operations = {
1648 .flush = mqueue_flush_file,
1649 .poll = mqueue_poll_file,
1650 .read = mqueue_read_file,
1651 .llseek = default_llseek,
1652};
1653
1654static const struct super_operations mqueue_super_ops = {
1655 .alloc_inode = mqueue_alloc_inode,
1656 .free_inode = mqueue_free_inode,
1657 .evict_inode = mqueue_evict_inode,
1658 .statfs = simple_statfs,
1659};
1660
1661static const struct fs_context_operations mqueue_fs_context_ops = {
1662 .free = mqueue_fs_context_free,
1663 .get_tree = mqueue_get_tree,
1664};
1665
1666static struct file_system_type mqueue_fs_type = {
1667 .name = "mqueue",
1668 .init_fs_context = mqueue_init_fs_context,
1669 .kill_sb = kill_litter_super,
1670 .fs_flags = FS_USERNS_MOUNT,
1671};
1672
1673int mq_init_ns(struct ipc_namespace *ns)
1674{
1675 struct vfsmount *m;
1676
1677 ns->mq_queues_count = 0;
1678 ns->mq_queues_max = DFLT_QUEUESMAX;
1679 ns->mq_msg_max = DFLT_MSGMAX;
1680 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1681 ns->mq_msg_default = DFLT_MSG;
1682 ns->mq_msgsize_default = DFLT_MSGSIZE;
1683
1684 m = mq_create_mount(ns);
1685 if (IS_ERR(m))
1686 return PTR_ERR(m);
1687 ns->mq_mnt = m;
1688 return 0;
1689}
1690
1691void mq_clear_sbinfo(struct ipc_namespace *ns)
1692{
1693 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1694}
1695
1696void mq_put_mnt(struct ipc_namespace *ns)
1697{
1698 kern_unmount(ns->mq_mnt);
1699}
1700
1701static int __init init_mqueue_fs(void)
1702{
1703 int error;
1704
1705 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1706 sizeof(struct mqueue_inode_info), 0,
1707 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1708 if (mqueue_inode_cachep == NULL)
1709 return -ENOMEM;
1710
1711 /* ignore failures - they are not fatal */
1712 mq_sysctl_table = mq_register_sysctl_table();
1713
1714 error = register_filesystem(&mqueue_fs_type);
1715 if (error)
1716 goto out_sysctl;
1717
1718 spin_lock_init(&mq_lock);
1719
1720 error = mq_init_ns(&init_ipc_ns);
1721 if (error)
1722 goto out_filesystem;
1723
1724 return 0;
1725
1726out_filesystem:
1727 unregister_filesystem(&mqueue_fs_type);
1728out_sysctl:
1729 if (mq_sysctl_table)
1730 unregister_sysctl_table(mq_sysctl_table);
1731 kmem_cache_destroy(mqueue_inode_cachep);
1732 return error;
1733}
1734
1735device_initcall(init_mqueue_fs);