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#include <linux/sched/wake_q.h>
39#include <linux/sched/signal.h>
40#include <linux/sched/user.h>
41
42#include <net/sock.h>
43#include "util.h"
44
45#define MQUEUE_MAGIC 0x19800202
46#define DIRENT_SIZE 20
47#define FILENT_SIZE 80
48
49#define SEND 0
50#define RECV 1
51
52#define STATE_NONE 0
53#define STATE_READY 1
54
55struct posix_msg_tree_node {
56 struct rb_node rb_node;
57 struct list_head msg_list;
58 int priority;
59};
60
61struct ext_wait_queue { /* queue of sleeping tasks */
62 struct task_struct *task;
63 struct list_head list;
64 struct msg_msg *msg; /* ptr of loaded message */
65 int state; /* one of STATE_* values */
66};
67
68struct mqueue_inode_info {
69 spinlock_t lock;
70 struct inode vfs_inode;
71 wait_queue_head_t wait_q;
72
73 struct rb_root msg_tree;
74 struct posix_msg_tree_node *node_cache;
75 struct mq_attr attr;
76
77 struct sigevent notify;
78 struct pid *notify_owner;
79 struct user_namespace *notify_user_ns;
80 struct user_struct *user; /* user who created, for accounting */
81 struct sock *notify_sock;
82 struct sk_buff *notify_cookie;
83
84 /* for tasks waiting for free space and messages, respectively */
85 struct ext_wait_queue e_wait_q[2];
86
87 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
88};
89
90static const struct inode_operations mqueue_dir_inode_operations;
91static const struct file_operations mqueue_file_operations;
92static const struct super_operations mqueue_super_ops;
93static void remove_notification(struct mqueue_inode_info *info);
94
95static struct kmem_cache *mqueue_inode_cachep;
96
97static struct ctl_table_header *mq_sysctl_table;
98
99static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
100{
101 return container_of(inode, struct mqueue_inode_info, vfs_inode);
102}
103
104/*
105 * This routine should be called with the mq_lock held.
106 */
107static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
108{
109 return get_ipc_ns(inode->i_sb->s_fs_info);
110}
111
112static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
113{
114 struct ipc_namespace *ns;
115
116 spin_lock(&mq_lock);
117 ns = __get_ns_from_inode(inode);
118 spin_unlock(&mq_lock);
119 return ns;
120}
121
122/* Auxiliary functions to manipulate messages' list */
123static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
124{
125 struct rb_node **p, *parent = NULL;
126 struct posix_msg_tree_node *leaf;
127
128 p = &info->msg_tree.rb_node;
129 while (*p) {
130 parent = *p;
131 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
132
133 if (likely(leaf->priority == msg->m_type))
134 goto insert_msg;
135 else if (msg->m_type < leaf->priority)
136 p = &(*p)->rb_left;
137 else
138 p = &(*p)->rb_right;
139 }
140 if (info->node_cache) {
141 leaf = info->node_cache;
142 info->node_cache = NULL;
143 } else {
144 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
145 if (!leaf)
146 return -ENOMEM;
147 INIT_LIST_HEAD(&leaf->msg_list);
148 }
149 leaf->priority = msg->m_type;
150 rb_link_node(&leaf->rb_node, parent, p);
151 rb_insert_color(&leaf->rb_node, &info->msg_tree);
152insert_msg:
153 info->attr.mq_curmsgs++;
154 info->qsize += msg->m_ts;
155 list_add_tail(&msg->m_list, &leaf->msg_list);
156 return 0;
157}
158
159static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
160{
161 struct rb_node **p, *parent = NULL;
162 struct posix_msg_tree_node *leaf;
163 struct msg_msg *msg;
164
165try_again:
166 p = &info->msg_tree.rb_node;
167 while (*p) {
168 parent = *p;
169 /*
170 * During insert, low priorities go to the left and high to the
171 * right. On receive, we want the highest priorities first, so
172 * walk all the way to the right.
173 */
174 p = &(*p)->rb_right;
175 }
176 if (!parent) {
177 if (info->attr.mq_curmsgs) {
178 pr_warn_once("Inconsistency in POSIX message queue, "
179 "no tree element, but supposedly messages "
180 "should exist!\n");
181 info->attr.mq_curmsgs = 0;
182 }
183 return NULL;
184 }
185 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
186 if (unlikely(list_empty(&leaf->msg_list))) {
187 pr_warn_once("Inconsistency in POSIX message queue, "
188 "empty leaf node but we haven't implemented "
189 "lazy leaf delete!\n");
190 rb_erase(&leaf->rb_node, &info->msg_tree);
191 if (info->node_cache) {
192 kfree(leaf);
193 } else {
194 info->node_cache = leaf;
195 }
196 goto try_again;
197 } else {
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
203 if (info->node_cache) {
204 kfree(leaf);
205 } else {
206 info->node_cache = leaf;
207 }
208 }
209 }
210 info->attr.mq_curmsgs--;
211 info->qsize -= msg->m_ts;
212 return msg;
213}
214
215static struct inode *mqueue_get_inode(struct super_block *sb,
216 struct ipc_namespace *ipc_ns, umode_t mode,
217 struct mq_attr *attr)
218{
219 struct user_struct *u = current_user();
220 struct inode *inode;
221 int ret = -ENOMEM;
222
223 inode = new_inode(sb);
224 if (!inode)
225 goto err;
226
227 inode->i_ino = get_next_ino();
228 inode->i_mode = mode;
229 inode->i_uid = current_fsuid();
230 inode->i_gid = current_fsgid();
231 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
232
233 if (S_ISREG(mode)) {
234 struct mqueue_inode_info *info;
235 unsigned long mq_bytes, mq_treesize;
236
237 inode->i_fop = &mqueue_file_operations;
238 inode->i_size = FILENT_SIZE;
239 /* mqueue specific info */
240 info = MQUEUE_I(inode);
241 spin_lock_init(&info->lock);
242 init_waitqueue_head(&info->wait_q);
243 INIT_LIST_HEAD(&info->e_wait_q[0].list);
244 INIT_LIST_HEAD(&info->e_wait_q[1].list);
245 info->notify_owner = NULL;
246 info->notify_user_ns = NULL;
247 info->qsize = 0;
248 info->user = NULL; /* set when all is ok */
249 info->msg_tree = RB_ROOT;
250 info->node_cache = NULL;
251 memset(&info->attr, 0, sizeof(info->attr));
252 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253 ipc_ns->mq_msg_default);
254 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255 ipc_ns->mq_msgsize_default);
256 if (attr) {
257 info->attr.mq_maxmsg = attr->mq_maxmsg;
258 info->attr.mq_msgsize = attr->mq_msgsize;
259 }
260 /*
261 * We used to allocate a static array of pointers and account
262 * the size of that array as well as one msg_msg struct per
263 * possible message into the queue size. That's no longer
264 * accurate as the queue is now an rbtree and will grow and
265 * shrink depending on usage patterns. We can, however, still
266 * account one msg_msg struct per message, but the nodes are
267 * allocated depending on priority usage, and most programs
268 * only use one, or a handful, of priorities. However, since
269 * this is pinned memory, we need to assume worst case, so
270 * that means the min(mq_maxmsg, max_priorities) * struct
271 * posix_msg_tree_node.
272 */
273
274 ret = -EINVAL;
275 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
276 goto out_inode;
277 if (capable(CAP_SYS_RESOURCE)) {
278 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
279 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
280 goto out_inode;
281 } else {
282 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
283 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
284 goto out_inode;
285 }
286 ret = -EOVERFLOW;
287 /* check for overflow */
288 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
289 goto out_inode;
290 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
291 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
292 sizeof(struct posix_msg_tree_node);
293 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
294 if (mq_bytes + mq_treesize < mq_bytes)
295 goto out_inode;
296 mq_bytes += mq_treesize;
297 spin_lock(&mq_lock);
298 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
299 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
300 spin_unlock(&mq_lock);
301 /* mqueue_evict_inode() releases info->messages */
302 ret = -EMFILE;
303 goto out_inode;
304 }
305 u->mq_bytes += mq_bytes;
306 spin_unlock(&mq_lock);
307
308 /* all is ok */
309 info->user = get_uid(u);
310 } else if (S_ISDIR(mode)) {
311 inc_nlink(inode);
312 /* Some things misbehave if size == 0 on a directory */
313 inode->i_size = 2 * DIRENT_SIZE;
314 inode->i_op = &mqueue_dir_inode_operations;
315 inode->i_fop = &simple_dir_operations;
316 }
317
318 return inode;
319out_inode:
320 iput(inode);
321err:
322 return ERR_PTR(ret);
323}
324
325static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
326{
327 struct inode *inode;
328 struct ipc_namespace *ns = sb->s_fs_info;
329
330 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
331 sb->s_blocksize = PAGE_SIZE;
332 sb->s_blocksize_bits = PAGE_SHIFT;
333 sb->s_magic = MQUEUE_MAGIC;
334 sb->s_op = &mqueue_super_ops;
335
336 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
337 if (IS_ERR(inode))
338 return PTR_ERR(inode);
339
340 sb->s_root = d_make_root(inode);
341 if (!sb->s_root)
342 return -ENOMEM;
343 return 0;
344}
345
346static struct dentry *mqueue_mount(struct file_system_type *fs_type,
347 int flags, const char *dev_name,
348 void *data)
349{
350 struct ipc_namespace *ns;
351 if (flags & SB_KERNMOUNT) {
352 ns = data;
353 data = NULL;
354 } else {
355 ns = current->nsproxy->ipc_ns;
356 }
357 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
358}
359
360static void init_once(void *foo)
361{
362 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
363
364 inode_init_once(&p->vfs_inode);
365}
366
367static struct inode *mqueue_alloc_inode(struct super_block *sb)
368{
369 struct mqueue_inode_info *ei;
370
371 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
372 if (!ei)
373 return NULL;
374 return &ei->vfs_inode;
375}
376
377static void mqueue_i_callback(struct rcu_head *head)
378{
379 struct inode *inode = container_of(head, struct inode, i_rcu);
380 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
381}
382
383static void mqueue_destroy_inode(struct inode *inode)
384{
385 call_rcu(&inode->i_rcu, mqueue_i_callback);
386}
387
388static void mqueue_evict_inode(struct inode *inode)
389{
390 struct mqueue_inode_info *info;
391 struct user_struct *user;
392 unsigned long mq_bytes, mq_treesize;
393 struct ipc_namespace *ipc_ns;
394 struct msg_msg *msg;
395
396 clear_inode(inode);
397
398 if (S_ISDIR(inode->i_mode))
399 return;
400
401 ipc_ns = get_ns_from_inode(inode);
402 info = MQUEUE_I(inode);
403 spin_lock(&info->lock);
404 while ((msg = msg_get(info)) != NULL)
405 free_msg(msg);
406 kfree(info->node_cache);
407 spin_unlock(&info->lock);
408
409 /* Total amount of bytes accounted for the mqueue */
410 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
411 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
412 sizeof(struct posix_msg_tree_node);
413
414 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
415 info->attr.mq_msgsize);
416
417 user = info->user;
418 if (user) {
419 spin_lock(&mq_lock);
420 user->mq_bytes -= mq_bytes;
421 /*
422 * get_ns_from_inode() ensures that the
423 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
424 * to which we now hold a reference, or it is NULL.
425 * We can't put it here under mq_lock, though.
426 */
427 if (ipc_ns)
428 ipc_ns->mq_queues_count--;
429 spin_unlock(&mq_lock);
430 free_uid(user);
431 }
432 if (ipc_ns)
433 put_ipc_ns(ipc_ns);
434}
435
436static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
437{
438 struct inode *dir = dentry->d_parent->d_inode;
439 struct inode *inode;
440 struct mq_attr *attr = arg;
441 int error;
442 struct ipc_namespace *ipc_ns;
443
444 spin_lock(&mq_lock);
445 ipc_ns = __get_ns_from_inode(dir);
446 if (!ipc_ns) {
447 error = -EACCES;
448 goto out_unlock;
449 }
450
451 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
452 !capable(CAP_SYS_RESOURCE)) {
453 error = -ENOSPC;
454 goto out_unlock;
455 }
456 ipc_ns->mq_queues_count++;
457 spin_unlock(&mq_lock);
458
459 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
460 if (IS_ERR(inode)) {
461 error = PTR_ERR(inode);
462 spin_lock(&mq_lock);
463 ipc_ns->mq_queues_count--;
464 goto out_unlock;
465 }
466
467 put_ipc_ns(ipc_ns);
468 dir->i_size += DIRENT_SIZE;
469 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
470
471 d_instantiate(dentry, inode);
472 dget(dentry);
473 return 0;
474out_unlock:
475 spin_unlock(&mq_lock);
476 if (ipc_ns)
477 put_ipc_ns(ipc_ns);
478 return error;
479}
480
481static int mqueue_create(struct inode *dir, struct dentry *dentry,
482 umode_t mode, bool excl)
483{
484 return mqueue_create_attr(dentry, mode, NULL);
485}
486
487static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
488{
489 struct inode *inode = d_inode(dentry);
490
491 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
492 dir->i_size -= DIRENT_SIZE;
493 drop_nlink(inode);
494 dput(dentry);
495 return 0;
496}
497
498/*
499* This is routine for system read from queue file.
500* To avoid mess with doing here some sort of mq_receive we allow
501* to read only queue size & notification info (the only values
502* that are interesting from user point of view and aren't accessible
503* through std routines)
504*/
505static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
506 size_t count, loff_t *off)
507{
508 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
509 char buffer[FILENT_SIZE];
510 ssize_t ret;
511
512 spin_lock(&info->lock);
513 snprintf(buffer, sizeof(buffer),
514 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
515 info->qsize,
516 info->notify_owner ? info->notify.sigev_notify : 0,
517 (info->notify_owner &&
518 info->notify.sigev_notify == SIGEV_SIGNAL) ?
519 info->notify.sigev_signo : 0,
520 pid_vnr(info->notify_owner));
521 spin_unlock(&info->lock);
522 buffer[sizeof(buffer)-1] = '\0';
523
524 ret = simple_read_from_buffer(u_data, count, off, buffer,
525 strlen(buffer));
526 if (ret <= 0)
527 return ret;
528
529 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
530 return ret;
531}
532
533static int mqueue_flush_file(struct file *filp, fl_owner_t id)
534{
535 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
536
537 spin_lock(&info->lock);
538 if (task_tgid(current) == info->notify_owner)
539 remove_notification(info);
540
541 spin_unlock(&info->lock);
542 return 0;
543}
544
545static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
546{
547 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
548 __poll_t retval = 0;
549
550 poll_wait(filp, &info->wait_q, poll_tab);
551
552 spin_lock(&info->lock);
553 if (info->attr.mq_curmsgs)
554 retval = EPOLLIN | EPOLLRDNORM;
555
556 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
557 retval |= EPOLLOUT | EPOLLWRNORM;
558 spin_unlock(&info->lock);
559
560 return retval;
561}
562
563/* Adds current to info->e_wait_q[sr] before element with smaller prio */
564static void wq_add(struct mqueue_inode_info *info, int sr,
565 struct ext_wait_queue *ewp)
566{
567 struct ext_wait_queue *walk;
568
569 ewp->task = current;
570
571 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
572 if (walk->task->prio <= current->prio) {
573 list_add_tail(&ewp->list, &walk->list);
574 return;
575 }
576 }
577 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
578}
579
580/*
581 * Puts current task to sleep. Caller must hold queue lock. After return
582 * lock isn't held.
583 * sr: SEND or RECV
584 */
585static int wq_sleep(struct mqueue_inode_info *info, int sr,
586 ktime_t *timeout, struct ext_wait_queue *ewp)
587 __releases(&info->lock)
588{
589 int retval;
590 signed long time;
591
592 wq_add(info, sr, ewp);
593
594 for (;;) {
595 __set_current_state(TASK_INTERRUPTIBLE);
596
597 spin_unlock(&info->lock);
598 time = schedule_hrtimeout_range_clock(timeout, 0,
599 HRTIMER_MODE_ABS, CLOCK_REALTIME);
600
601 if (ewp->state == STATE_READY) {
602 retval = 0;
603 goto out;
604 }
605 spin_lock(&info->lock);
606 if (ewp->state == STATE_READY) {
607 retval = 0;
608 goto out_unlock;
609 }
610 if (signal_pending(current)) {
611 retval = -ERESTARTSYS;
612 break;
613 }
614 if (time == 0) {
615 retval = -ETIMEDOUT;
616 break;
617 }
618 }
619 list_del(&ewp->list);
620out_unlock:
621 spin_unlock(&info->lock);
622out:
623 return retval;
624}
625
626/*
627 * Returns waiting task that should be serviced first or NULL if none exists
628 */
629static struct ext_wait_queue *wq_get_first_waiter(
630 struct mqueue_inode_info *info, int sr)
631{
632 struct list_head *ptr;
633
634 ptr = info->e_wait_q[sr].list.prev;
635 if (ptr == &info->e_wait_q[sr].list)
636 return NULL;
637 return list_entry(ptr, struct ext_wait_queue, list);
638}
639
640
641static inline void set_cookie(struct sk_buff *skb, char code)
642{
643 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
644}
645
646/*
647 * The next function is only to split too long sys_mq_timedsend
648 */
649static void __do_notify(struct mqueue_inode_info *info)
650{
651 /* notification
652 * invoked when there is registered process and there isn't process
653 * waiting synchronously for message AND state of queue changed from
654 * empty to not empty. Here we are sure that no one is waiting
655 * synchronously. */
656 if (info->notify_owner &&
657 info->attr.mq_curmsgs == 1) {
658 struct siginfo sig_i;
659 switch (info->notify.sigev_notify) {
660 case SIGEV_NONE:
661 break;
662 case SIGEV_SIGNAL:
663 /* sends signal */
664
665 clear_siginfo(&sig_i);
666 sig_i.si_signo = info->notify.sigev_signo;
667 sig_i.si_errno = 0;
668 sig_i.si_code = SI_MESGQ;
669 sig_i.si_value = info->notify.sigev_value;
670 /* map current pid/uid into info->owner's namespaces */
671 rcu_read_lock();
672 sig_i.si_pid = task_tgid_nr_ns(current,
673 ns_of_pid(info->notify_owner));
674 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
675 rcu_read_unlock();
676
677 kill_pid_info(info->notify.sigev_signo,
678 &sig_i, info->notify_owner);
679 break;
680 case SIGEV_THREAD:
681 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
682 netlink_sendskb(info->notify_sock, info->notify_cookie);
683 break;
684 }
685 /* after notification unregisters process */
686 put_pid(info->notify_owner);
687 put_user_ns(info->notify_user_ns);
688 info->notify_owner = NULL;
689 info->notify_user_ns = NULL;
690 }
691 wake_up(&info->wait_q);
692}
693
694static int prepare_timeout(const struct timespec __user *u_abs_timeout,
695 struct timespec64 *ts)
696{
697 if (get_timespec64(ts, u_abs_timeout))
698 return -EFAULT;
699 if (!timespec64_valid(ts))
700 return -EINVAL;
701 return 0;
702}
703
704static void remove_notification(struct mqueue_inode_info *info)
705{
706 if (info->notify_owner != NULL &&
707 info->notify.sigev_notify == SIGEV_THREAD) {
708 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
709 netlink_sendskb(info->notify_sock, info->notify_cookie);
710 }
711 put_pid(info->notify_owner);
712 put_user_ns(info->notify_user_ns);
713 info->notify_owner = NULL;
714 info->notify_user_ns = NULL;
715}
716
717static int prepare_open(struct dentry *dentry, int oflag, int ro,
718 umode_t mode, struct filename *name,
719 struct mq_attr *attr)
720{
721 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
722 MAY_READ | MAY_WRITE };
723 int acc;
724
725 if (d_really_is_negative(dentry)) {
726 if (!(oflag & O_CREAT))
727 return -ENOENT;
728 if (ro)
729 return ro;
730 audit_inode_parent_hidden(name, dentry->d_parent);
731 return vfs_mkobj(dentry, mode & ~current_umask(),
732 mqueue_create_attr, attr);
733 }
734 /* it already existed */
735 audit_inode(name, dentry, 0);
736 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
737 return -EEXIST;
738 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
739 return -EINVAL;
740 acc = oflag2acc[oflag & O_ACCMODE];
741 return inode_permission(d_inode(dentry), acc);
742}
743
744static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
745 struct mq_attr *attr)
746{
747 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
748 struct dentry *root = mnt->mnt_root;
749 struct filename *name;
750 struct path path;
751 int fd, error;
752 int ro;
753
754 audit_mq_open(oflag, mode, attr);
755
756 if (IS_ERR(name = getname(u_name)))
757 return PTR_ERR(name);
758
759 fd = get_unused_fd_flags(O_CLOEXEC);
760 if (fd < 0)
761 goto out_putname;
762
763 ro = mnt_want_write(mnt); /* we'll drop it in any case */
764 inode_lock(d_inode(root));
765 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
766 if (IS_ERR(path.dentry)) {
767 error = PTR_ERR(path.dentry);
768 goto out_putfd;
769 }
770 path.mnt = mntget(mnt);
771 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
772 if (!error) {
773 struct file *file = dentry_open(&path, oflag, current_cred());
774 if (!IS_ERR(file))
775 fd_install(fd, file);
776 else
777 error = PTR_ERR(file);
778 }
779 path_put(&path);
780out_putfd:
781 if (error) {
782 put_unused_fd(fd);
783 fd = error;
784 }
785 inode_unlock(d_inode(root));
786 if (!ro)
787 mnt_drop_write(mnt);
788out_putname:
789 putname(name);
790 return fd;
791}
792
793SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
794 struct mq_attr __user *, u_attr)
795{
796 struct mq_attr attr;
797 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
798 return -EFAULT;
799
800 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
801}
802
803SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
804{
805 int err;
806 struct filename *name;
807 struct dentry *dentry;
808 struct inode *inode = NULL;
809 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
810 struct vfsmount *mnt = ipc_ns->mq_mnt;
811
812 name = getname(u_name);
813 if (IS_ERR(name))
814 return PTR_ERR(name);
815
816 audit_inode_parent_hidden(name, mnt->mnt_root);
817 err = mnt_want_write(mnt);
818 if (err)
819 goto out_name;
820 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
821 dentry = lookup_one_len(name->name, mnt->mnt_root,
822 strlen(name->name));
823 if (IS_ERR(dentry)) {
824 err = PTR_ERR(dentry);
825 goto out_unlock;
826 }
827
828 inode = d_inode(dentry);
829 if (!inode) {
830 err = -ENOENT;
831 } else {
832 ihold(inode);
833 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
834 }
835 dput(dentry);
836
837out_unlock:
838 inode_unlock(d_inode(mnt->mnt_root));
839 if (inode)
840 iput(inode);
841 mnt_drop_write(mnt);
842out_name:
843 putname(name);
844
845 return err;
846}
847
848/* Pipelined send and receive functions.
849 *
850 * If a receiver finds no waiting message, then it registers itself in the
851 * list of waiting receivers. A sender checks that list before adding the new
852 * message into the message array. If there is a waiting receiver, then it
853 * bypasses the message array and directly hands the message over to the
854 * receiver. The receiver accepts the message and returns without grabbing the
855 * queue spinlock:
856 *
857 * - Set pointer to message.
858 * - Queue the receiver task for later wakeup (without the info->lock).
859 * - Update its state to STATE_READY. Now the receiver can continue.
860 * - Wake up the process after the lock is dropped. Should the process wake up
861 * before this wakeup (due to a timeout or a signal) it will either see
862 * STATE_READY and continue or acquire the lock to check the state again.
863 *
864 * The same algorithm is used for senders.
865 */
866
867/* pipelined_send() - send a message directly to the task waiting in
868 * sys_mq_timedreceive() (without inserting message into a queue).
869 */
870static inline void pipelined_send(struct wake_q_head *wake_q,
871 struct mqueue_inode_info *info,
872 struct msg_msg *message,
873 struct ext_wait_queue *receiver)
874{
875 receiver->msg = message;
876 list_del(&receiver->list);
877 wake_q_add(wake_q, receiver->task);
878 /*
879 * Rely on the implicit cmpxchg barrier from wake_q_add such
880 * that we can ensure that updating receiver->state is the last
881 * write operation: As once set, the receiver can continue,
882 * and if we don't have the reference count from the wake_q,
883 * yet, at that point we can later have a use-after-free
884 * condition and bogus wakeup.
885 */
886 receiver->state = STATE_READY;
887}
888
889/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
890 * gets its message and put to the queue (we have one free place for sure). */
891static inline void pipelined_receive(struct wake_q_head *wake_q,
892 struct mqueue_inode_info *info)
893{
894 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
895
896 if (!sender) {
897 /* for poll */
898 wake_up_interruptible(&info->wait_q);
899 return;
900 }
901 if (msg_insert(sender->msg, info))
902 return;
903
904 list_del(&sender->list);
905 wake_q_add(wake_q, sender->task);
906 sender->state = STATE_READY;
907}
908
909static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
910 size_t msg_len, unsigned int msg_prio,
911 struct timespec64 *ts)
912{
913 struct fd f;
914 struct inode *inode;
915 struct ext_wait_queue wait;
916 struct ext_wait_queue *receiver;
917 struct msg_msg *msg_ptr;
918 struct mqueue_inode_info *info;
919 ktime_t expires, *timeout = NULL;
920 struct posix_msg_tree_node *new_leaf = NULL;
921 int ret = 0;
922 DEFINE_WAKE_Q(wake_q);
923
924 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
925 return -EINVAL;
926
927 if (ts) {
928 expires = timespec64_to_ktime(*ts);
929 timeout = &expires;
930 }
931
932 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
933
934 f = fdget(mqdes);
935 if (unlikely(!f.file)) {
936 ret = -EBADF;
937 goto out;
938 }
939
940 inode = file_inode(f.file);
941 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
942 ret = -EBADF;
943 goto out_fput;
944 }
945 info = MQUEUE_I(inode);
946 audit_file(f.file);
947
948 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
949 ret = -EBADF;
950 goto out_fput;
951 }
952
953 if (unlikely(msg_len > info->attr.mq_msgsize)) {
954 ret = -EMSGSIZE;
955 goto out_fput;
956 }
957
958 /* First try to allocate memory, before doing anything with
959 * existing queues. */
960 msg_ptr = load_msg(u_msg_ptr, msg_len);
961 if (IS_ERR(msg_ptr)) {
962 ret = PTR_ERR(msg_ptr);
963 goto out_fput;
964 }
965 msg_ptr->m_ts = msg_len;
966 msg_ptr->m_type = msg_prio;
967
968 /*
969 * msg_insert really wants us to have a valid, spare node struct so
970 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
971 * fall back to that if necessary.
972 */
973 if (!info->node_cache)
974 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
975
976 spin_lock(&info->lock);
977
978 if (!info->node_cache && new_leaf) {
979 /* Save our speculative allocation into the cache */
980 INIT_LIST_HEAD(&new_leaf->msg_list);
981 info->node_cache = new_leaf;
982 new_leaf = NULL;
983 } else {
984 kfree(new_leaf);
985 }
986
987 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
988 if (f.file->f_flags & O_NONBLOCK) {
989 ret = -EAGAIN;
990 } else {
991 wait.task = current;
992 wait.msg = (void *) msg_ptr;
993 wait.state = STATE_NONE;
994 ret = wq_sleep(info, SEND, timeout, &wait);
995 /*
996 * wq_sleep must be called with info->lock held, and
997 * returns with the lock released
998 */
999 goto out_free;
1000 }
1001 } else {
1002 receiver = wq_get_first_waiter(info, RECV);
1003 if (receiver) {
1004 pipelined_send(&wake_q, info, msg_ptr, receiver);
1005 } else {
1006 /* adds message to the queue */
1007 ret = msg_insert(msg_ptr, info);
1008 if (ret)
1009 goto out_unlock;
1010 __do_notify(info);
1011 }
1012 inode->i_atime = inode->i_mtime = inode->i_ctime =
1013 current_time(inode);
1014 }
1015out_unlock:
1016 spin_unlock(&info->lock);
1017 wake_up_q(&wake_q);
1018out_free:
1019 if (ret)
1020 free_msg(msg_ptr);
1021out_fput:
1022 fdput(f);
1023out:
1024 return ret;
1025}
1026
1027static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1028 size_t msg_len, unsigned int __user *u_msg_prio,
1029 struct timespec64 *ts)
1030{
1031 ssize_t ret;
1032 struct msg_msg *msg_ptr;
1033 struct fd f;
1034 struct inode *inode;
1035 struct mqueue_inode_info *info;
1036 struct ext_wait_queue wait;
1037 ktime_t expires, *timeout = NULL;
1038 struct posix_msg_tree_node *new_leaf = NULL;
1039
1040 if (ts) {
1041 expires = timespec64_to_ktime(*ts);
1042 timeout = &expires;
1043 }
1044
1045 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1046
1047 f = fdget(mqdes);
1048 if (unlikely(!f.file)) {
1049 ret = -EBADF;
1050 goto out;
1051 }
1052
1053 inode = file_inode(f.file);
1054 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1055 ret = -EBADF;
1056 goto out_fput;
1057 }
1058 info = MQUEUE_I(inode);
1059 audit_file(f.file);
1060
1061 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1062 ret = -EBADF;
1063 goto out_fput;
1064 }
1065
1066 /* checks if buffer is big enough */
1067 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1068 ret = -EMSGSIZE;
1069 goto out_fput;
1070 }
1071
1072 /*
1073 * msg_insert really wants us to have a valid, spare node struct so
1074 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1075 * fall back to that if necessary.
1076 */
1077 if (!info->node_cache)
1078 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1079
1080 spin_lock(&info->lock);
1081
1082 if (!info->node_cache && new_leaf) {
1083 /* Save our speculative allocation into the cache */
1084 INIT_LIST_HEAD(&new_leaf->msg_list);
1085 info->node_cache = new_leaf;
1086 } else {
1087 kfree(new_leaf);
1088 }
1089
1090 if (info->attr.mq_curmsgs == 0) {
1091 if (f.file->f_flags & O_NONBLOCK) {
1092 spin_unlock(&info->lock);
1093 ret = -EAGAIN;
1094 } else {
1095 wait.task = current;
1096 wait.state = STATE_NONE;
1097 ret = wq_sleep(info, RECV, timeout, &wait);
1098 msg_ptr = wait.msg;
1099 }
1100 } else {
1101 DEFINE_WAKE_Q(wake_q);
1102
1103 msg_ptr = msg_get(info);
1104
1105 inode->i_atime = inode->i_mtime = inode->i_ctime =
1106 current_time(inode);
1107
1108 /* There is now free space in queue. */
1109 pipelined_receive(&wake_q, info);
1110 spin_unlock(&info->lock);
1111 wake_up_q(&wake_q);
1112 ret = 0;
1113 }
1114 if (ret == 0) {
1115 ret = msg_ptr->m_ts;
1116
1117 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1118 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1119 ret = -EFAULT;
1120 }
1121 free_msg(msg_ptr);
1122 }
1123out_fput:
1124 fdput(f);
1125out:
1126 return ret;
1127}
1128
1129SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1130 size_t, msg_len, unsigned int, msg_prio,
1131 const struct timespec __user *, u_abs_timeout)
1132{
1133 struct timespec64 ts, *p = NULL;
1134 if (u_abs_timeout) {
1135 int res = prepare_timeout(u_abs_timeout, &ts);
1136 if (res)
1137 return res;
1138 p = &ts;
1139 }
1140 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1141}
1142
1143SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1144 size_t, msg_len, unsigned int __user *, u_msg_prio,
1145 const struct timespec __user *, u_abs_timeout)
1146{
1147 struct timespec64 ts, *p = NULL;
1148 if (u_abs_timeout) {
1149 int res = prepare_timeout(u_abs_timeout, &ts);
1150 if (res)
1151 return res;
1152 p = &ts;
1153 }
1154 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1155}
1156
1157/*
1158 * Notes: the case when user wants us to deregister (with NULL as pointer)
1159 * and he isn't currently owner of notification, will be silently discarded.
1160 * It isn't explicitly defined in the POSIX.
1161 */
1162static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1163{
1164 int ret;
1165 struct fd f;
1166 struct sock *sock;
1167 struct inode *inode;
1168 struct mqueue_inode_info *info;
1169 struct sk_buff *nc;
1170
1171 audit_mq_notify(mqdes, notification);
1172
1173 nc = NULL;
1174 sock = NULL;
1175 if (notification != NULL) {
1176 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1177 notification->sigev_notify != SIGEV_SIGNAL &&
1178 notification->sigev_notify != SIGEV_THREAD))
1179 return -EINVAL;
1180 if (notification->sigev_notify == SIGEV_SIGNAL &&
1181 !valid_signal(notification->sigev_signo)) {
1182 return -EINVAL;
1183 }
1184 if (notification->sigev_notify == SIGEV_THREAD) {
1185 long timeo;
1186
1187 /* create the notify skb */
1188 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1189 if (!nc) {
1190 ret = -ENOMEM;
1191 goto out;
1192 }
1193 if (copy_from_user(nc->data,
1194 notification->sigev_value.sival_ptr,
1195 NOTIFY_COOKIE_LEN)) {
1196 ret = -EFAULT;
1197 goto out;
1198 }
1199
1200 /* TODO: add a header? */
1201 skb_put(nc, NOTIFY_COOKIE_LEN);
1202 /* and attach it to the socket */
1203retry:
1204 f = fdget(notification->sigev_signo);
1205 if (!f.file) {
1206 ret = -EBADF;
1207 goto out;
1208 }
1209 sock = netlink_getsockbyfilp(f.file);
1210 fdput(f);
1211 if (IS_ERR(sock)) {
1212 ret = PTR_ERR(sock);
1213 sock = NULL;
1214 goto out;
1215 }
1216
1217 timeo = MAX_SCHEDULE_TIMEOUT;
1218 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1219 if (ret == 1) {
1220 sock = NULL;
1221 goto retry;
1222 }
1223 if (ret) {
1224 sock = NULL;
1225 nc = NULL;
1226 goto out;
1227 }
1228 }
1229 }
1230
1231 f = fdget(mqdes);
1232 if (!f.file) {
1233 ret = -EBADF;
1234 goto out;
1235 }
1236
1237 inode = file_inode(f.file);
1238 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1239 ret = -EBADF;
1240 goto out_fput;
1241 }
1242 info = MQUEUE_I(inode);
1243
1244 ret = 0;
1245 spin_lock(&info->lock);
1246 if (notification == NULL) {
1247 if (info->notify_owner == task_tgid(current)) {
1248 remove_notification(info);
1249 inode->i_atime = inode->i_ctime = current_time(inode);
1250 }
1251 } else if (info->notify_owner != NULL) {
1252 ret = -EBUSY;
1253 } else {
1254 switch (notification->sigev_notify) {
1255 case SIGEV_NONE:
1256 info->notify.sigev_notify = SIGEV_NONE;
1257 break;
1258 case SIGEV_THREAD:
1259 info->notify_sock = sock;
1260 info->notify_cookie = nc;
1261 sock = NULL;
1262 nc = NULL;
1263 info->notify.sigev_notify = SIGEV_THREAD;
1264 break;
1265 case SIGEV_SIGNAL:
1266 info->notify.sigev_signo = notification->sigev_signo;
1267 info->notify.sigev_value = notification->sigev_value;
1268 info->notify.sigev_notify = SIGEV_SIGNAL;
1269 break;
1270 }
1271
1272 info->notify_owner = get_pid(task_tgid(current));
1273 info->notify_user_ns = get_user_ns(current_user_ns());
1274 inode->i_atime = inode->i_ctime = current_time(inode);
1275 }
1276 spin_unlock(&info->lock);
1277out_fput:
1278 fdput(f);
1279out:
1280 if (sock)
1281 netlink_detachskb(sock, nc);
1282 else if (nc)
1283 dev_kfree_skb(nc);
1284
1285 return ret;
1286}
1287
1288SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1289 const struct sigevent __user *, u_notification)
1290{
1291 struct sigevent n, *p = NULL;
1292 if (u_notification) {
1293 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1294 return -EFAULT;
1295 p = &n;
1296 }
1297 return do_mq_notify(mqdes, p);
1298}
1299
1300static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1301{
1302 struct fd f;
1303 struct inode *inode;
1304 struct mqueue_inode_info *info;
1305
1306 if (new && (new->mq_flags & (~O_NONBLOCK)))
1307 return -EINVAL;
1308
1309 f = fdget(mqdes);
1310 if (!f.file)
1311 return -EBADF;
1312
1313 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1314 fdput(f);
1315 return -EBADF;
1316 }
1317
1318 inode = file_inode(f.file);
1319 info = MQUEUE_I(inode);
1320
1321 spin_lock(&info->lock);
1322
1323 if (old) {
1324 *old = info->attr;
1325 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1326 }
1327 if (new) {
1328 audit_mq_getsetattr(mqdes, new);
1329 spin_lock(&f.file->f_lock);
1330 if (new->mq_flags & O_NONBLOCK)
1331 f.file->f_flags |= O_NONBLOCK;
1332 else
1333 f.file->f_flags &= ~O_NONBLOCK;
1334 spin_unlock(&f.file->f_lock);
1335
1336 inode->i_atime = inode->i_ctime = current_time(inode);
1337 }
1338
1339 spin_unlock(&info->lock);
1340 fdput(f);
1341 return 0;
1342}
1343
1344SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1345 const struct mq_attr __user *, u_mqstat,
1346 struct mq_attr __user *, u_omqstat)
1347{
1348 int ret;
1349 struct mq_attr mqstat, omqstat;
1350 struct mq_attr *new = NULL, *old = NULL;
1351
1352 if (u_mqstat) {
1353 new = &mqstat;
1354 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1355 return -EFAULT;
1356 }
1357 if (u_omqstat)
1358 old = &omqstat;
1359
1360 ret = do_mq_getsetattr(mqdes, new, old);
1361 if (ret || !old)
1362 return ret;
1363
1364 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1365 return -EFAULT;
1366 return 0;
1367}
1368
1369#ifdef CONFIG_COMPAT
1370
1371struct compat_mq_attr {
1372 compat_long_t mq_flags; /* message queue flags */
1373 compat_long_t mq_maxmsg; /* maximum number of messages */
1374 compat_long_t mq_msgsize; /* maximum message size */
1375 compat_long_t mq_curmsgs; /* number of messages currently queued */
1376 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1377};
1378
1379static inline int get_compat_mq_attr(struct mq_attr *attr,
1380 const struct compat_mq_attr __user *uattr)
1381{
1382 struct compat_mq_attr v;
1383
1384 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1385 return -EFAULT;
1386
1387 memset(attr, 0, sizeof(*attr));
1388 attr->mq_flags = v.mq_flags;
1389 attr->mq_maxmsg = v.mq_maxmsg;
1390 attr->mq_msgsize = v.mq_msgsize;
1391 attr->mq_curmsgs = v.mq_curmsgs;
1392 return 0;
1393}
1394
1395static inline int put_compat_mq_attr(const struct mq_attr *attr,
1396 struct compat_mq_attr __user *uattr)
1397{
1398 struct compat_mq_attr v;
1399
1400 memset(&v, 0, sizeof(v));
1401 v.mq_flags = attr->mq_flags;
1402 v.mq_maxmsg = attr->mq_maxmsg;
1403 v.mq_msgsize = attr->mq_msgsize;
1404 v.mq_curmsgs = attr->mq_curmsgs;
1405 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1406 return -EFAULT;
1407 return 0;
1408}
1409
1410COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1411 int, oflag, compat_mode_t, mode,
1412 struct compat_mq_attr __user *, u_attr)
1413{
1414 struct mq_attr attr, *p = NULL;
1415 if (u_attr && oflag & O_CREAT) {
1416 p = &attr;
1417 if (get_compat_mq_attr(&attr, u_attr))
1418 return -EFAULT;
1419 }
1420 return do_mq_open(u_name, oflag, mode, p);
1421}
1422
1423static int compat_prepare_timeout(const struct compat_timespec __user *p,
1424 struct timespec64 *ts)
1425{
1426 if (compat_get_timespec64(ts, p))
1427 return -EFAULT;
1428 if (!timespec64_valid(ts))
1429 return -EINVAL;
1430 return 0;
1431}
1432
1433COMPAT_SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes,
1434 const char __user *, u_msg_ptr,
1435 compat_size_t, msg_len, unsigned int, msg_prio,
1436 const struct compat_timespec __user *, u_abs_timeout)
1437{
1438 struct timespec64 ts, *p = NULL;
1439 if (u_abs_timeout) {
1440 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1441 if (res)
1442 return res;
1443 p = &ts;
1444 }
1445 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1446}
1447
1448COMPAT_SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes,
1449 char __user *, u_msg_ptr,
1450 compat_size_t, msg_len, unsigned int __user *, u_msg_prio,
1451 const struct compat_timespec __user *, u_abs_timeout)
1452{
1453 struct timespec64 ts, *p = NULL;
1454 if (u_abs_timeout) {
1455 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1456 if (res)
1457 return res;
1458 p = &ts;
1459 }
1460 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1461}
1462
1463COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1464 const struct compat_sigevent __user *, u_notification)
1465{
1466 struct sigevent n, *p = NULL;
1467 if (u_notification) {
1468 if (get_compat_sigevent(&n, u_notification))
1469 return -EFAULT;
1470 if (n.sigev_notify == SIGEV_THREAD)
1471 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1472 p = &n;
1473 }
1474 return do_mq_notify(mqdes, p);
1475}
1476
1477COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1478 const struct compat_mq_attr __user *, u_mqstat,
1479 struct compat_mq_attr __user *, u_omqstat)
1480{
1481 int ret;
1482 struct mq_attr mqstat, omqstat;
1483 struct mq_attr *new = NULL, *old = NULL;
1484
1485 if (u_mqstat) {
1486 new = &mqstat;
1487 if (get_compat_mq_attr(new, u_mqstat))
1488 return -EFAULT;
1489 }
1490 if (u_omqstat)
1491 old = &omqstat;
1492
1493 ret = do_mq_getsetattr(mqdes, new, old);
1494 if (ret || !old)
1495 return ret;
1496
1497 if (put_compat_mq_attr(old, u_omqstat))
1498 return -EFAULT;
1499 return 0;
1500}
1501#endif
1502
1503static const struct inode_operations mqueue_dir_inode_operations = {
1504 .lookup = simple_lookup,
1505 .create = mqueue_create,
1506 .unlink = mqueue_unlink,
1507};
1508
1509static const struct file_operations mqueue_file_operations = {
1510 .flush = mqueue_flush_file,
1511 .poll = mqueue_poll_file,
1512 .read = mqueue_read_file,
1513 .llseek = default_llseek,
1514};
1515
1516static const struct super_operations mqueue_super_ops = {
1517 .alloc_inode = mqueue_alloc_inode,
1518 .destroy_inode = mqueue_destroy_inode,
1519 .evict_inode = mqueue_evict_inode,
1520 .statfs = simple_statfs,
1521};
1522
1523static struct file_system_type mqueue_fs_type = {
1524 .name = "mqueue",
1525 .mount = mqueue_mount,
1526 .kill_sb = kill_litter_super,
1527 .fs_flags = FS_USERNS_MOUNT,
1528};
1529
1530int mq_init_ns(struct ipc_namespace *ns)
1531{
1532 ns->mq_queues_count = 0;
1533 ns->mq_queues_max = DFLT_QUEUESMAX;
1534 ns->mq_msg_max = DFLT_MSGMAX;
1535 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1536 ns->mq_msg_default = DFLT_MSG;
1537 ns->mq_msgsize_default = DFLT_MSGSIZE;
1538
1539 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1540 if (IS_ERR(ns->mq_mnt)) {
1541 int err = PTR_ERR(ns->mq_mnt);
1542 ns->mq_mnt = NULL;
1543 return err;
1544 }
1545 return 0;
1546}
1547
1548void mq_clear_sbinfo(struct ipc_namespace *ns)
1549{
1550 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1551}
1552
1553void mq_put_mnt(struct ipc_namespace *ns)
1554{
1555 kern_unmount(ns->mq_mnt);
1556}
1557
1558static int __init init_mqueue_fs(void)
1559{
1560 int error;
1561
1562 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1563 sizeof(struct mqueue_inode_info), 0,
1564 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1565 if (mqueue_inode_cachep == NULL)
1566 return -ENOMEM;
1567
1568 /* ignore failures - they are not fatal */
1569 mq_sysctl_table = mq_register_sysctl_table();
1570
1571 error = register_filesystem(&mqueue_fs_type);
1572 if (error)
1573 goto out_sysctl;
1574
1575 spin_lock_init(&mq_lock);
1576
1577 error = mq_init_ns(&init_ipc_ns);
1578 if (error)
1579 goto out_filesystem;
1580
1581 return 0;
1582
1583out_filesystem:
1584 unregister_filesystem(&mqueue_fs_type);
1585out_sysctl:
1586 if (mq_sysctl_table)
1587 unregister_sysctl_table(mq_sysctl_table);
1588 kmem_cache_destroy(mqueue_inode_cachep);
1589 return error;
1590}
1591
1592device_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
66struct ext_wait_queue { /* queue of sleeping tasks */
67 struct task_struct *task;
68 struct list_head list;
69 struct msg_msg *msg; /* ptr of loaded message */
70 int state; /* one of STATE_* values */
71};
72
73struct mqueue_inode_info {
74 spinlock_t lock;
75 struct inode vfs_inode;
76 wait_queue_head_t wait_q;
77
78 struct rb_root msg_tree;
79 struct rb_node *msg_tree_rightmost;
80 struct posix_msg_tree_node *node_cache;
81 struct mq_attr attr;
82
83 struct sigevent notify;
84 struct pid *notify_owner;
85 struct user_namespace *notify_user_ns;
86 struct user_struct *user; /* user who created, for accounting */
87 struct sock *notify_sock;
88 struct sk_buff *notify_cookie;
89
90 /* for tasks waiting for free space and messages, respectively */
91 struct ext_wait_queue e_wait_q[2];
92
93 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
94};
95
96static struct file_system_type mqueue_fs_type;
97static const struct inode_operations mqueue_dir_inode_operations;
98static const struct file_operations mqueue_file_operations;
99static const struct super_operations mqueue_super_ops;
100static const struct fs_context_operations mqueue_fs_context_ops;
101static void remove_notification(struct mqueue_inode_info *info);
102
103static struct kmem_cache *mqueue_inode_cachep;
104
105static struct ctl_table_header *mq_sysctl_table;
106
107static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
108{
109 return container_of(inode, struct mqueue_inode_info, vfs_inode);
110}
111
112/*
113 * This routine should be called with the mq_lock held.
114 */
115static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
116{
117 return get_ipc_ns(inode->i_sb->s_fs_info);
118}
119
120static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
121{
122 struct ipc_namespace *ns;
123
124 spin_lock(&mq_lock);
125 ns = __get_ns_from_inode(inode);
126 spin_unlock(&mq_lock);
127 return ns;
128}
129
130/* Auxiliary functions to manipulate messages' list */
131static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
132{
133 struct rb_node **p, *parent = NULL;
134 struct posix_msg_tree_node *leaf;
135 bool rightmost = true;
136
137 p = &info->msg_tree.rb_node;
138 while (*p) {
139 parent = *p;
140 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
141
142 if (likely(leaf->priority == msg->m_type))
143 goto insert_msg;
144 else if (msg->m_type < leaf->priority) {
145 p = &(*p)->rb_left;
146 rightmost = false;
147 } else
148 p = &(*p)->rb_right;
149 }
150 if (info->node_cache) {
151 leaf = info->node_cache;
152 info->node_cache = NULL;
153 } else {
154 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
155 if (!leaf)
156 return -ENOMEM;
157 INIT_LIST_HEAD(&leaf->msg_list);
158 }
159 leaf->priority = msg->m_type;
160
161 if (rightmost)
162 info->msg_tree_rightmost = &leaf->rb_node;
163
164 rb_link_node(&leaf->rb_node, parent, p);
165 rb_insert_color(&leaf->rb_node, &info->msg_tree);
166insert_msg:
167 info->attr.mq_curmsgs++;
168 info->qsize += msg->m_ts;
169 list_add_tail(&msg->m_list, &leaf->msg_list);
170 return 0;
171}
172
173static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
174 struct mqueue_inode_info *info)
175{
176 struct rb_node *node = &leaf->rb_node;
177
178 if (info->msg_tree_rightmost == node)
179 info->msg_tree_rightmost = rb_prev(node);
180
181 rb_erase(node, &info->msg_tree);
182 if (info->node_cache) {
183 kfree(leaf);
184 } else {
185 info->node_cache = leaf;
186 }
187}
188
189static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
190{
191 struct rb_node *parent = NULL;
192 struct posix_msg_tree_node *leaf;
193 struct msg_msg *msg;
194
195try_again:
196 /*
197 * During insert, low priorities go to the left and high to the
198 * right. On receive, we want the highest priorities first, so
199 * walk all the way to the right.
200 */
201 parent = info->msg_tree_rightmost;
202 if (!parent) {
203 if (info->attr.mq_curmsgs) {
204 pr_warn_once("Inconsistency in POSIX message queue, "
205 "no tree element, but supposedly messages "
206 "should exist!\n");
207 info->attr.mq_curmsgs = 0;
208 }
209 return NULL;
210 }
211 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
212 if (unlikely(list_empty(&leaf->msg_list))) {
213 pr_warn_once("Inconsistency in POSIX message queue, "
214 "empty leaf node but we haven't implemented "
215 "lazy leaf delete!\n");
216 msg_tree_erase(leaf, info);
217 goto try_again;
218 } else {
219 msg = list_first_entry(&leaf->msg_list,
220 struct msg_msg, m_list);
221 list_del(&msg->m_list);
222 if (list_empty(&leaf->msg_list)) {
223 msg_tree_erase(leaf, info);
224 }
225 }
226 info->attr.mq_curmsgs--;
227 info->qsize -= msg->m_ts;
228 return msg;
229}
230
231static struct inode *mqueue_get_inode(struct super_block *sb,
232 struct ipc_namespace *ipc_ns, umode_t mode,
233 struct mq_attr *attr)
234{
235 struct user_struct *u = current_user();
236 struct inode *inode;
237 int ret = -ENOMEM;
238
239 inode = new_inode(sb);
240 if (!inode)
241 goto err;
242
243 inode->i_ino = get_next_ino();
244 inode->i_mode = mode;
245 inode->i_uid = current_fsuid();
246 inode->i_gid = current_fsgid();
247 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
248
249 if (S_ISREG(mode)) {
250 struct mqueue_inode_info *info;
251 unsigned long mq_bytes, mq_treesize;
252
253 inode->i_fop = &mqueue_file_operations;
254 inode->i_size = FILENT_SIZE;
255 /* mqueue specific info */
256 info = MQUEUE_I(inode);
257 spin_lock_init(&info->lock);
258 init_waitqueue_head(&info->wait_q);
259 INIT_LIST_HEAD(&info->e_wait_q[0].list);
260 INIT_LIST_HEAD(&info->e_wait_q[1].list);
261 info->notify_owner = NULL;
262 info->notify_user_ns = NULL;
263 info->qsize = 0;
264 info->user = NULL; /* set when all is ok */
265 info->msg_tree = RB_ROOT;
266 info->msg_tree_rightmost = NULL;
267 info->node_cache = NULL;
268 memset(&info->attr, 0, sizeof(info->attr));
269 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
270 ipc_ns->mq_msg_default);
271 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
272 ipc_ns->mq_msgsize_default);
273 if (attr) {
274 info->attr.mq_maxmsg = attr->mq_maxmsg;
275 info->attr.mq_msgsize = attr->mq_msgsize;
276 }
277 /*
278 * We used to allocate a static array of pointers and account
279 * the size of that array as well as one msg_msg struct per
280 * possible message into the queue size. That's no longer
281 * accurate as the queue is now an rbtree and will grow and
282 * shrink depending on usage patterns. We can, however, still
283 * account one msg_msg struct per message, but the nodes are
284 * allocated depending on priority usage, and most programs
285 * only use one, or a handful, of priorities. However, since
286 * this is pinned memory, we need to assume worst case, so
287 * that means the min(mq_maxmsg, max_priorities) * struct
288 * posix_msg_tree_node.
289 */
290
291 ret = -EINVAL;
292 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
293 goto out_inode;
294 if (capable(CAP_SYS_RESOURCE)) {
295 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
296 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
297 goto out_inode;
298 } else {
299 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
300 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
301 goto out_inode;
302 }
303 ret = -EOVERFLOW;
304 /* check for overflow */
305 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
306 goto out_inode;
307 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
308 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
309 sizeof(struct posix_msg_tree_node);
310 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
311 if (mq_bytes + mq_treesize < mq_bytes)
312 goto out_inode;
313 mq_bytes += mq_treesize;
314 spin_lock(&mq_lock);
315 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
316 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
317 spin_unlock(&mq_lock);
318 /* mqueue_evict_inode() releases info->messages */
319 ret = -EMFILE;
320 goto out_inode;
321 }
322 u->mq_bytes += mq_bytes;
323 spin_unlock(&mq_lock);
324
325 /* all is ok */
326 info->user = get_uid(u);
327 } else if (S_ISDIR(mode)) {
328 inc_nlink(inode);
329 /* Some things misbehave if size == 0 on a directory */
330 inode->i_size = 2 * DIRENT_SIZE;
331 inode->i_op = &mqueue_dir_inode_operations;
332 inode->i_fop = &simple_dir_operations;
333 }
334
335 return inode;
336out_inode:
337 iput(inode);
338err:
339 return ERR_PTR(ret);
340}
341
342static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
343{
344 struct inode *inode;
345 struct ipc_namespace *ns = sb->s_fs_info;
346
347 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
348 sb->s_blocksize = PAGE_SIZE;
349 sb->s_blocksize_bits = PAGE_SHIFT;
350 sb->s_magic = MQUEUE_MAGIC;
351 sb->s_op = &mqueue_super_ops;
352
353 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
354 if (IS_ERR(inode))
355 return PTR_ERR(inode);
356
357 sb->s_root = d_make_root(inode);
358 if (!sb->s_root)
359 return -ENOMEM;
360 return 0;
361}
362
363static int mqueue_get_tree(struct fs_context *fc)
364{
365 struct mqueue_fs_context *ctx = fc->fs_private;
366
367 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
368}
369
370static void mqueue_fs_context_free(struct fs_context *fc)
371{
372 struct mqueue_fs_context *ctx = fc->fs_private;
373
374 put_ipc_ns(ctx->ipc_ns);
375 kfree(ctx);
376}
377
378static int mqueue_init_fs_context(struct fs_context *fc)
379{
380 struct mqueue_fs_context *ctx;
381
382 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
383 if (!ctx)
384 return -ENOMEM;
385
386 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
387 put_user_ns(fc->user_ns);
388 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
389 fc->fs_private = ctx;
390 fc->ops = &mqueue_fs_context_ops;
391 return 0;
392}
393
394static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
395{
396 struct mqueue_fs_context *ctx;
397 struct fs_context *fc;
398 struct vfsmount *mnt;
399
400 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
401 if (IS_ERR(fc))
402 return ERR_CAST(fc);
403
404 ctx = fc->fs_private;
405 put_ipc_ns(ctx->ipc_ns);
406 ctx->ipc_ns = get_ipc_ns(ns);
407 put_user_ns(fc->user_ns);
408 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
409
410 mnt = fc_mount(fc);
411 put_fs_context(fc);
412 return mnt;
413}
414
415static void init_once(void *foo)
416{
417 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
418
419 inode_init_once(&p->vfs_inode);
420}
421
422static struct inode *mqueue_alloc_inode(struct super_block *sb)
423{
424 struct mqueue_inode_info *ei;
425
426 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
427 if (!ei)
428 return NULL;
429 return &ei->vfs_inode;
430}
431
432static void mqueue_free_inode(struct inode *inode)
433{
434 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
435}
436
437static void mqueue_evict_inode(struct inode *inode)
438{
439 struct mqueue_inode_info *info;
440 struct user_struct *user;
441 struct ipc_namespace *ipc_ns;
442 struct msg_msg *msg, *nmsg;
443 LIST_HEAD(tmp_msg);
444
445 clear_inode(inode);
446
447 if (S_ISDIR(inode->i_mode))
448 return;
449
450 ipc_ns = get_ns_from_inode(inode);
451 info = MQUEUE_I(inode);
452 spin_lock(&info->lock);
453 while ((msg = msg_get(info)) != NULL)
454 list_add_tail(&msg->m_list, &tmp_msg);
455 kfree(info->node_cache);
456 spin_unlock(&info->lock);
457
458 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
459 list_del(&msg->m_list);
460 free_msg(msg);
461 }
462
463 user = info->user;
464 if (user) {
465 unsigned long mq_bytes, mq_treesize;
466
467 /* Total amount of bytes accounted for the mqueue */
468 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
469 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
470 sizeof(struct posix_msg_tree_node);
471
472 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
473 info->attr.mq_msgsize);
474
475 spin_lock(&mq_lock);
476 user->mq_bytes -= mq_bytes;
477 /*
478 * get_ns_from_inode() ensures that the
479 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
480 * to which we now hold a reference, or it is NULL.
481 * We can't put it here under mq_lock, though.
482 */
483 if (ipc_ns)
484 ipc_ns->mq_queues_count--;
485 spin_unlock(&mq_lock);
486 free_uid(user);
487 }
488 if (ipc_ns)
489 put_ipc_ns(ipc_ns);
490}
491
492static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
493{
494 struct inode *dir = dentry->d_parent->d_inode;
495 struct inode *inode;
496 struct mq_attr *attr = arg;
497 int error;
498 struct ipc_namespace *ipc_ns;
499
500 spin_lock(&mq_lock);
501 ipc_ns = __get_ns_from_inode(dir);
502 if (!ipc_ns) {
503 error = -EACCES;
504 goto out_unlock;
505 }
506
507 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
508 !capable(CAP_SYS_RESOURCE)) {
509 error = -ENOSPC;
510 goto out_unlock;
511 }
512 ipc_ns->mq_queues_count++;
513 spin_unlock(&mq_lock);
514
515 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
516 if (IS_ERR(inode)) {
517 error = PTR_ERR(inode);
518 spin_lock(&mq_lock);
519 ipc_ns->mq_queues_count--;
520 goto out_unlock;
521 }
522
523 put_ipc_ns(ipc_ns);
524 dir->i_size += DIRENT_SIZE;
525 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
526
527 d_instantiate(dentry, inode);
528 dget(dentry);
529 return 0;
530out_unlock:
531 spin_unlock(&mq_lock);
532 if (ipc_ns)
533 put_ipc_ns(ipc_ns);
534 return error;
535}
536
537static int mqueue_create(struct inode *dir, struct dentry *dentry,
538 umode_t mode, bool excl)
539{
540 return mqueue_create_attr(dentry, mode, NULL);
541}
542
543static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
544{
545 struct inode *inode = d_inode(dentry);
546
547 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
548 dir->i_size -= DIRENT_SIZE;
549 drop_nlink(inode);
550 dput(dentry);
551 return 0;
552}
553
554/*
555* This is routine for system read from queue file.
556* To avoid mess with doing here some sort of mq_receive we allow
557* to read only queue size & notification info (the only values
558* that are interesting from user point of view and aren't accessible
559* through std routines)
560*/
561static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
562 size_t count, loff_t *off)
563{
564 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
565 char buffer[FILENT_SIZE];
566 ssize_t ret;
567
568 spin_lock(&info->lock);
569 snprintf(buffer, sizeof(buffer),
570 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
571 info->qsize,
572 info->notify_owner ? info->notify.sigev_notify : 0,
573 (info->notify_owner &&
574 info->notify.sigev_notify == SIGEV_SIGNAL) ?
575 info->notify.sigev_signo : 0,
576 pid_vnr(info->notify_owner));
577 spin_unlock(&info->lock);
578 buffer[sizeof(buffer)-1] = '\0';
579
580 ret = simple_read_from_buffer(u_data, count, off, buffer,
581 strlen(buffer));
582 if (ret <= 0)
583 return ret;
584
585 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
586 return ret;
587}
588
589static int mqueue_flush_file(struct file *filp, fl_owner_t id)
590{
591 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
592
593 spin_lock(&info->lock);
594 if (task_tgid(current) == info->notify_owner)
595 remove_notification(info);
596
597 spin_unlock(&info->lock);
598 return 0;
599}
600
601static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
602{
603 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
604 __poll_t retval = 0;
605
606 poll_wait(filp, &info->wait_q, poll_tab);
607
608 spin_lock(&info->lock);
609 if (info->attr.mq_curmsgs)
610 retval = EPOLLIN | EPOLLRDNORM;
611
612 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
613 retval |= EPOLLOUT | EPOLLWRNORM;
614 spin_unlock(&info->lock);
615
616 return retval;
617}
618
619/* Adds current to info->e_wait_q[sr] before element with smaller prio */
620static void wq_add(struct mqueue_inode_info *info, int sr,
621 struct ext_wait_queue *ewp)
622{
623 struct ext_wait_queue *walk;
624
625 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
626 if (walk->task->prio <= current->prio) {
627 list_add_tail(&ewp->list, &walk->list);
628 return;
629 }
630 }
631 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
632}
633
634/*
635 * Puts current task to sleep. Caller must hold queue lock. After return
636 * lock isn't held.
637 * sr: SEND or RECV
638 */
639static int wq_sleep(struct mqueue_inode_info *info, int sr,
640 ktime_t *timeout, struct ext_wait_queue *ewp)
641 __releases(&info->lock)
642{
643 int retval;
644 signed long time;
645
646 wq_add(info, sr, ewp);
647
648 for (;;) {
649 __set_current_state(TASK_INTERRUPTIBLE);
650
651 spin_unlock(&info->lock);
652 time = schedule_hrtimeout_range_clock(timeout, 0,
653 HRTIMER_MODE_ABS, CLOCK_REALTIME);
654
655 if (ewp->state == STATE_READY) {
656 retval = 0;
657 goto out;
658 }
659 spin_lock(&info->lock);
660 if (ewp->state == STATE_READY) {
661 retval = 0;
662 goto out_unlock;
663 }
664 if (signal_pending(current)) {
665 retval = -ERESTARTSYS;
666 break;
667 }
668 if (time == 0) {
669 retval = -ETIMEDOUT;
670 break;
671 }
672 }
673 list_del(&ewp->list);
674out_unlock:
675 spin_unlock(&info->lock);
676out:
677 return retval;
678}
679
680/*
681 * Returns waiting task that should be serviced first or NULL if none exists
682 */
683static struct ext_wait_queue *wq_get_first_waiter(
684 struct mqueue_inode_info *info, int sr)
685{
686 struct list_head *ptr;
687
688 ptr = info->e_wait_q[sr].list.prev;
689 if (ptr == &info->e_wait_q[sr].list)
690 return NULL;
691 return list_entry(ptr, struct ext_wait_queue, list);
692}
693
694
695static inline void set_cookie(struct sk_buff *skb, char code)
696{
697 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
698}
699
700/*
701 * The next function is only to split too long sys_mq_timedsend
702 */
703static void __do_notify(struct mqueue_inode_info *info)
704{
705 /* notification
706 * invoked when there is registered process and there isn't process
707 * waiting synchronously for message AND state of queue changed from
708 * empty to not empty. Here we are sure that no one is waiting
709 * synchronously. */
710 if (info->notify_owner &&
711 info->attr.mq_curmsgs == 1) {
712 struct kernel_siginfo sig_i;
713 switch (info->notify.sigev_notify) {
714 case SIGEV_NONE:
715 break;
716 case SIGEV_SIGNAL:
717 /* sends signal */
718
719 clear_siginfo(&sig_i);
720 sig_i.si_signo = info->notify.sigev_signo;
721 sig_i.si_errno = 0;
722 sig_i.si_code = SI_MESGQ;
723 sig_i.si_value = info->notify.sigev_value;
724 /* map current pid/uid into info->owner's namespaces */
725 rcu_read_lock();
726 sig_i.si_pid = task_tgid_nr_ns(current,
727 ns_of_pid(info->notify_owner));
728 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
729 rcu_read_unlock();
730
731 kill_pid_info(info->notify.sigev_signo,
732 &sig_i, info->notify_owner);
733 break;
734 case SIGEV_THREAD:
735 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
736 netlink_sendskb(info->notify_sock, info->notify_cookie);
737 break;
738 }
739 /* after notification unregisters process */
740 put_pid(info->notify_owner);
741 put_user_ns(info->notify_user_ns);
742 info->notify_owner = NULL;
743 info->notify_user_ns = NULL;
744 }
745 wake_up(&info->wait_q);
746}
747
748static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
749 struct timespec64 *ts)
750{
751 if (get_timespec64(ts, u_abs_timeout))
752 return -EFAULT;
753 if (!timespec64_valid(ts))
754 return -EINVAL;
755 return 0;
756}
757
758static void remove_notification(struct mqueue_inode_info *info)
759{
760 if (info->notify_owner != NULL &&
761 info->notify.sigev_notify == SIGEV_THREAD) {
762 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
763 netlink_sendskb(info->notify_sock, info->notify_cookie);
764 }
765 put_pid(info->notify_owner);
766 put_user_ns(info->notify_user_ns);
767 info->notify_owner = NULL;
768 info->notify_user_ns = NULL;
769}
770
771static int prepare_open(struct dentry *dentry, int oflag, int ro,
772 umode_t mode, struct filename *name,
773 struct mq_attr *attr)
774{
775 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
776 MAY_READ | MAY_WRITE };
777 int acc;
778
779 if (d_really_is_negative(dentry)) {
780 if (!(oflag & O_CREAT))
781 return -ENOENT;
782 if (ro)
783 return ro;
784 audit_inode_parent_hidden(name, dentry->d_parent);
785 return vfs_mkobj(dentry, mode & ~current_umask(),
786 mqueue_create_attr, attr);
787 }
788 /* it already existed */
789 audit_inode(name, dentry, 0);
790 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
791 return -EEXIST;
792 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
793 return -EINVAL;
794 acc = oflag2acc[oflag & O_ACCMODE];
795 return inode_permission(d_inode(dentry), acc);
796}
797
798static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
799 struct mq_attr *attr)
800{
801 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
802 struct dentry *root = mnt->mnt_root;
803 struct filename *name;
804 struct path path;
805 int fd, error;
806 int ro;
807
808 audit_mq_open(oflag, mode, attr);
809
810 if (IS_ERR(name = getname(u_name)))
811 return PTR_ERR(name);
812
813 fd = get_unused_fd_flags(O_CLOEXEC);
814 if (fd < 0)
815 goto out_putname;
816
817 ro = mnt_want_write(mnt); /* we'll drop it in any case */
818 inode_lock(d_inode(root));
819 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
820 if (IS_ERR(path.dentry)) {
821 error = PTR_ERR(path.dentry);
822 goto out_putfd;
823 }
824 path.mnt = mntget(mnt);
825 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
826 if (!error) {
827 struct file *file = dentry_open(&path, oflag, current_cred());
828 if (!IS_ERR(file))
829 fd_install(fd, file);
830 else
831 error = PTR_ERR(file);
832 }
833 path_put(&path);
834out_putfd:
835 if (error) {
836 put_unused_fd(fd);
837 fd = error;
838 }
839 inode_unlock(d_inode(root));
840 if (!ro)
841 mnt_drop_write(mnt);
842out_putname:
843 putname(name);
844 return fd;
845}
846
847SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
848 struct mq_attr __user *, u_attr)
849{
850 struct mq_attr attr;
851 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
852 return -EFAULT;
853
854 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
855}
856
857SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
858{
859 int err;
860 struct filename *name;
861 struct dentry *dentry;
862 struct inode *inode = NULL;
863 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
864 struct vfsmount *mnt = ipc_ns->mq_mnt;
865
866 name = getname(u_name);
867 if (IS_ERR(name))
868 return PTR_ERR(name);
869
870 audit_inode_parent_hidden(name, mnt->mnt_root);
871 err = mnt_want_write(mnt);
872 if (err)
873 goto out_name;
874 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
875 dentry = lookup_one_len(name->name, mnt->mnt_root,
876 strlen(name->name));
877 if (IS_ERR(dentry)) {
878 err = PTR_ERR(dentry);
879 goto out_unlock;
880 }
881
882 inode = d_inode(dentry);
883 if (!inode) {
884 err = -ENOENT;
885 } else {
886 ihold(inode);
887 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
888 }
889 dput(dentry);
890
891out_unlock:
892 inode_unlock(d_inode(mnt->mnt_root));
893 if (inode)
894 iput(inode);
895 mnt_drop_write(mnt);
896out_name:
897 putname(name);
898
899 return err;
900}
901
902/* Pipelined send and receive functions.
903 *
904 * If a receiver finds no waiting message, then it registers itself in the
905 * list of waiting receivers. A sender checks that list before adding the new
906 * message into the message array. If there is a waiting receiver, then it
907 * bypasses the message array and directly hands the message over to the
908 * receiver. The receiver accepts the message and returns without grabbing the
909 * queue spinlock:
910 *
911 * - Set pointer to message.
912 * - Queue the receiver task for later wakeup (without the info->lock).
913 * - Update its state to STATE_READY. Now the receiver can continue.
914 * - Wake up the process after the lock is dropped. Should the process wake up
915 * before this wakeup (due to a timeout or a signal) it will either see
916 * STATE_READY and continue or acquire the lock to check the state again.
917 *
918 * The same algorithm is used for senders.
919 */
920
921/* pipelined_send() - send a message directly to the task waiting in
922 * sys_mq_timedreceive() (without inserting message into a queue).
923 */
924static inline void pipelined_send(struct wake_q_head *wake_q,
925 struct mqueue_inode_info *info,
926 struct msg_msg *message,
927 struct ext_wait_queue *receiver)
928{
929 receiver->msg = message;
930 list_del(&receiver->list);
931 wake_q_add(wake_q, receiver->task);
932 /*
933 * Rely on the implicit cmpxchg barrier from wake_q_add such
934 * that we can ensure that updating receiver->state is the last
935 * write operation: As once set, the receiver can continue,
936 * and if we don't have the reference count from the wake_q,
937 * yet, at that point we can later have a use-after-free
938 * condition and bogus wakeup.
939 */
940 receiver->state = STATE_READY;
941}
942
943/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
944 * gets its message and put to the queue (we have one free place for sure). */
945static inline void pipelined_receive(struct wake_q_head *wake_q,
946 struct mqueue_inode_info *info)
947{
948 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
949
950 if (!sender) {
951 /* for poll */
952 wake_up_interruptible(&info->wait_q);
953 return;
954 }
955 if (msg_insert(sender->msg, info))
956 return;
957
958 list_del(&sender->list);
959 wake_q_add(wake_q, sender->task);
960 sender->state = STATE_READY;
961}
962
963static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
964 size_t msg_len, unsigned int msg_prio,
965 struct timespec64 *ts)
966{
967 struct fd f;
968 struct inode *inode;
969 struct ext_wait_queue wait;
970 struct ext_wait_queue *receiver;
971 struct msg_msg *msg_ptr;
972 struct mqueue_inode_info *info;
973 ktime_t expires, *timeout = NULL;
974 struct posix_msg_tree_node *new_leaf = NULL;
975 int ret = 0;
976 DEFINE_WAKE_Q(wake_q);
977
978 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
979 return -EINVAL;
980
981 if (ts) {
982 expires = timespec64_to_ktime(*ts);
983 timeout = &expires;
984 }
985
986 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
987
988 f = fdget(mqdes);
989 if (unlikely(!f.file)) {
990 ret = -EBADF;
991 goto out;
992 }
993
994 inode = file_inode(f.file);
995 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
996 ret = -EBADF;
997 goto out_fput;
998 }
999 info = MQUEUE_I(inode);
1000 audit_file(f.file);
1001
1002 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1003 ret = -EBADF;
1004 goto out_fput;
1005 }
1006
1007 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1008 ret = -EMSGSIZE;
1009 goto out_fput;
1010 }
1011
1012 /* First try to allocate memory, before doing anything with
1013 * existing queues. */
1014 msg_ptr = load_msg(u_msg_ptr, msg_len);
1015 if (IS_ERR(msg_ptr)) {
1016 ret = PTR_ERR(msg_ptr);
1017 goto out_fput;
1018 }
1019 msg_ptr->m_ts = msg_len;
1020 msg_ptr->m_type = msg_prio;
1021
1022 /*
1023 * msg_insert really wants us to have a valid, spare node struct so
1024 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1025 * fall back to that if necessary.
1026 */
1027 if (!info->node_cache)
1028 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1029
1030 spin_lock(&info->lock);
1031
1032 if (!info->node_cache && new_leaf) {
1033 /* Save our speculative allocation into the cache */
1034 INIT_LIST_HEAD(&new_leaf->msg_list);
1035 info->node_cache = new_leaf;
1036 new_leaf = NULL;
1037 } else {
1038 kfree(new_leaf);
1039 }
1040
1041 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1042 if (f.file->f_flags & O_NONBLOCK) {
1043 ret = -EAGAIN;
1044 } else {
1045 wait.task = current;
1046 wait.msg = (void *) msg_ptr;
1047 wait.state = STATE_NONE;
1048 ret = wq_sleep(info, SEND, timeout, &wait);
1049 /*
1050 * wq_sleep must be called with info->lock held, and
1051 * returns with the lock released
1052 */
1053 goto out_free;
1054 }
1055 } else {
1056 receiver = wq_get_first_waiter(info, RECV);
1057 if (receiver) {
1058 pipelined_send(&wake_q, info, msg_ptr, receiver);
1059 } else {
1060 /* adds message to the queue */
1061 ret = msg_insert(msg_ptr, info);
1062 if (ret)
1063 goto out_unlock;
1064 __do_notify(info);
1065 }
1066 inode->i_atime = inode->i_mtime = inode->i_ctime =
1067 current_time(inode);
1068 }
1069out_unlock:
1070 spin_unlock(&info->lock);
1071 wake_up_q(&wake_q);
1072out_free:
1073 if (ret)
1074 free_msg(msg_ptr);
1075out_fput:
1076 fdput(f);
1077out:
1078 return ret;
1079}
1080
1081static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1082 size_t msg_len, unsigned int __user *u_msg_prio,
1083 struct timespec64 *ts)
1084{
1085 ssize_t ret;
1086 struct msg_msg *msg_ptr;
1087 struct fd f;
1088 struct inode *inode;
1089 struct mqueue_inode_info *info;
1090 struct ext_wait_queue wait;
1091 ktime_t expires, *timeout = NULL;
1092 struct posix_msg_tree_node *new_leaf = NULL;
1093
1094 if (ts) {
1095 expires = timespec64_to_ktime(*ts);
1096 timeout = &expires;
1097 }
1098
1099 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1100
1101 f = fdget(mqdes);
1102 if (unlikely(!f.file)) {
1103 ret = -EBADF;
1104 goto out;
1105 }
1106
1107 inode = file_inode(f.file);
1108 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1109 ret = -EBADF;
1110 goto out_fput;
1111 }
1112 info = MQUEUE_I(inode);
1113 audit_file(f.file);
1114
1115 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1116 ret = -EBADF;
1117 goto out_fput;
1118 }
1119
1120 /* checks if buffer is big enough */
1121 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1122 ret = -EMSGSIZE;
1123 goto out_fput;
1124 }
1125
1126 /*
1127 * msg_insert really wants us to have a valid, spare node struct so
1128 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1129 * fall back to that if necessary.
1130 */
1131 if (!info->node_cache)
1132 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1133
1134 spin_lock(&info->lock);
1135
1136 if (!info->node_cache && new_leaf) {
1137 /* Save our speculative allocation into the cache */
1138 INIT_LIST_HEAD(&new_leaf->msg_list);
1139 info->node_cache = new_leaf;
1140 } else {
1141 kfree(new_leaf);
1142 }
1143
1144 if (info->attr.mq_curmsgs == 0) {
1145 if (f.file->f_flags & O_NONBLOCK) {
1146 spin_unlock(&info->lock);
1147 ret = -EAGAIN;
1148 } else {
1149 wait.task = current;
1150 wait.state = STATE_NONE;
1151 ret = wq_sleep(info, RECV, timeout, &wait);
1152 msg_ptr = wait.msg;
1153 }
1154 } else {
1155 DEFINE_WAKE_Q(wake_q);
1156
1157 msg_ptr = msg_get(info);
1158
1159 inode->i_atime = inode->i_mtime = inode->i_ctime =
1160 current_time(inode);
1161
1162 /* There is now free space in queue. */
1163 pipelined_receive(&wake_q, info);
1164 spin_unlock(&info->lock);
1165 wake_up_q(&wake_q);
1166 ret = 0;
1167 }
1168 if (ret == 0) {
1169 ret = msg_ptr->m_ts;
1170
1171 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1172 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1173 ret = -EFAULT;
1174 }
1175 free_msg(msg_ptr);
1176 }
1177out_fput:
1178 fdput(f);
1179out:
1180 return ret;
1181}
1182
1183SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1184 size_t, msg_len, unsigned int, msg_prio,
1185 const struct __kernel_timespec __user *, u_abs_timeout)
1186{
1187 struct timespec64 ts, *p = NULL;
1188 if (u_abs_timeout) {
1189 int res = prepare_timeout(u_abs_timeout, &ts);
1190 if (res)
1191 return res;
1192 p = &ts;
1193 }
1194 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1195}
1196
1197SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1198 size_t, msg_len, unsigned int __user *, u_msg_prio,
1199 const struct __kernel_timespec __user *, u_abs_timeout)
1200{
1201 struct timespec64 ts, *p = NULL;
1202 if (u_abs_timeout) {
1203 int res = prepare_timeout(u_abs_timeout, &ts);
1204 if (res)
1205 return res;
1206 p = &ts;
1207 }
1208 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1209}
1210
1211/*
1212 * Notes: the case when user wants us to deregister (with NULL as pointer)
1213 * and he isn't currently owner of notification, will be silently discarded.
1214 * It isn't explicitly defined in the POSIX.
1215 */
1216static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1217{
1218 int ret;
1219 struct fd f;
1220 struct sock *sock;
1221 struct inode *inode;
1222 struct mqueue_inode_info *info;
1223 struct sk_buff *nc;
1224
1225 audit_mq_notify(mqdes, notification);
1226
1227 nc = NULL;
1228 sock = NULL;
1229 if (notification != NULL) {
1230 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1231 notification->sigev_notify != SIGEV_SIGNAL &&
1232 notification->sigev_notify != SIGEV_THREAD))
1233 return -EINVAL;
1234 if (notification->sigev_notify == SIGEV_SIGNAL &&
1235 !valid_signal(notification->sigev_signo)) {
1236 return -EINVAL;
1237 }
1238 if (notification->sigev_notify == SIGEV_THREAD) {
1239 long timeo;
1240
1241 /* create the notify skb */
1242 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1243 if (!nc)
1244 return -ENOMEM;
1245
1246 if (copy_from_user(nc->data,
1247 notification->sigev_value.sival_ptr,
1248 NOTIFY_COOKIE_LEN)) {
1249 ret = -EFAULT;
1250 goto free_skb;
1251 }
1252
1253 /* TODO: add a header? */
1254 skb_put(nc, NOTIFY_COOKIE_LEN);
1255 /* and attach it to the socket */
1256retry:
1257 f = fdget(notification->sigev_signo);
1258 if (!f.file) {
1259 ret = -EBADF;
1260 goto out;
1261 }
1262 sock = netlink_getsockbyfilp(f.file);
1263 fdput(f);
1264 if (IS_ERR(sock)) {
1265 ret = PTR_ERR(sock);
1266 goto free_skb;
1267 }
1268
1269 timeo = MAX_SCHEDULE_TIMEOUT;
1270 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1271 if (ret == 1) {
1272 sock = NULL;
1273 goto retry;
1274 }
1275 if (ret)
1276 return ret;
1277 }
1278 }
1279
1280 f = fdget(mqdes);
1281 if (!f.file) {
1282 ret = -EBADF;
1283 goto out;
1284 }
1285
1286 inode = file_inode(f.file);
1287 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1288 ret = -EBADF;
1289 goto out_fput;
1290 }
1291 info = MQUEUE_I(inode);
1292
1293 ret = 0;
1294 spin_lock(&info->lock);
1295 if (notification == NULL) {
1296 if (info->notify_owner == task_tgid(current)) {
1297 remove_notification(info);
1298 inode->i_atime = inode->i_ctime = current_time(inode);
1299 }
1300 } else if (info->notify_owner != NULL) {
1301 ret = -EBUSY;
1302 } else {
1303 switch (notification->sigev_notify) {
1304 case SIGEV_NONE:
1305 info->notify.sigev_notify = SIGEV_NONE;
1306 break;
1307 case SIGEV_THREAD:
1308 info->notify_sock = sock;
1309 info->notify_cookie = nc;
1310 sock = NULL;
1311 nc = NULL;
1312 info->notify.sigev_notify = SIGEV_THREAD;
1313 break;
1314 case SIGEV_SIGNAL:
1315 info->notify.sigev_signo = notification->sigev_signo;
1316 info->notify.sigev_value = notification->sigev_value;
1317 info->notify.sigev_notify = SIGEV_SIGNAL;
1318 break;
1319 }
1320
1321 info->notify_owner = get_pid(task_tgid(current));
1322 info->notify_user_ns = get_user_ns(current_user_ns());
1323 inode->i_atime = inode->i_ctime = current_time(inode);
1324 }
1325 spin_unlock(&info->lock);
1326out_fput:
1327 fdput(f);
1328out:
1329 if (sock)
1330 netlink_detachskb(sock, nc);
1331 else
1332free_skb:
1333 dev_kfree_skb(nc);
1334
1335 return ret;
1336}
1337
1338SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1339 const struct sigevent __user *, u_notification)
1340{
1341 struct sigevent n, *p = NULL;
1342 if (u_notification) {
1343 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1344 return -EFAULT;
1345 p = &n;
1346 }
1347 return do_mq_notify(mqdes, p);
1348}
1349
1350static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1351{
1352 struct fd f;
1353 struct inode *inode;
1354 struct mqueue_inode_info *info;
1355
1356 if (new && (new->mq_flags & (~O_NONBLOCK)))
1357 return -EINVAL;
1358
1359 f = fdget(mqdes);
1360 if (!f.file)
1361 return -EBADF;
1362
1363 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1364 fdput(f);
1365 return -EBADF;
1366 }
1367
1368 inode = file_inode(f.file);
1369 info = MQUEUE_I(inode);
1370
1371 spin_lock(&info->lock);
1372
1373 if (old) {
1374 *old = info->attr;
1375 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1376 }
1377 if (new) {
1378 audit_mq_getsetattr(mqdes, new);
1379 spin_lock(&f.file->f_lock);
1380 if (new->mq_flags & O_NONBLOCK)
1381 f.file->f_flags |= O_NONBLOCK;
1382 else
1383 f.file->f_flags &= ~O_NONBLOCK;
1384 spin_unlock(&f.file->f_lock);
1385
1386 inode->i_atime = inode->i_ctime = current_time(inode);
1387 }
1388
1389 spin_unlock(&info->lock);
1390 fdput(f);
1391 return 0;
1392}
1393
1394SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1395 const struct mq_attr __user *, u_mqstat,
1396 struct mq_attr __user *, u_omqstat)
1397{
1398 int ret;
1399 struct mq_attr mqstat, omqstat;
1400 struct mq_attr *new = NULL, *old = NULL;
1401
1402 if (u_mqstat) {
1403 new = &mqstat;
1404 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1405 return -EFAULT;
1406 }
1407 if (u_omqstat)
1408 old = &omqstat;
1409
1410 ret = do_mq_getsetattr(mqdes, new, old);
1411 if (ret || !old)
1412 return ret;
1413
1414 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1415 return -EFAULT;
1416 return 0;
1417}
1418
1419#ifdef CONFIG_COMPAT
1420
1421struct compat_mq_attr {
1422 compat_long_t mq_flags; /* message queue flags */
1423 compat_long_t mq_maxmsg; /* maximum number of messages */
1424 compat_long_t mq_msgsize; /* maximum message size */
1425 compat_long_t mq_curmsgs; /* number of messages currently queued */
1426 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1427};
1428
1429static inline int get_compat_mq_attr(struct mq_attr *attr,
1430 const struct compat_mq_attr __user *uattr)
1431{
1432 struct compat_mq_attr v;
1433
1434 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1435 return -EFAULT;
1436
1437 memset(attr, 0, sizeof(*attr));
1438 attr->mq_flags = v.mq_flags;
1439 attr->mq_maxmsg = v.mq_maxmsg;
1440 attr->mq_msgsize = v.mq_msgsize;
1441 attr->mq_curmsgs = v.mq_curmsgs;
1442 return 0;
1443}
1444
1445static inline int put_compat_mq_attr(const struct mq_attr *attr,
1446 struct compat_mq_attr __user *uattr)
1447{
1448 struct compat_mq_attr v;
1449
1450 memset(&v, 0, sizeof(v));
1451 v.mq_flags = attr->mq_flags;
1452 v.mq_maxmsg = attr->mq_maxmsg;
1453 v.mq_msgsize = attr->mq_msgsize;
1454 v.mq_curmsgs = attr->mq_curmsgs;
1455 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1456 return -EFAULT;
1457 return 0;
1458}
1459
1460COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1461 int, oflag, compat_mode_t, mode,
1462 struct compat_mq_attr __user *, u_attr)
1463{
1464 struct mq_attr attr, *p = NULL;
1465 if (u_attr && oflag & O_CREAT) {
1466 p = &attr;
1467 if (get_compat_mq_attr(&attr, u_attr))
1468 return -EFAULT;
1469 }
1470 return do_mq_open(u_name, oflag, mode, p);
1471}
1472
1473COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1474 const struct compat_sigevent __user *, u_notification)
1475{
1476 struct sigevent n, *p = NULL;
1477 if (u_notification) {
1478 if (get_compat_sigevent(&n, u_notification))
1479 return -EFAULT;
1480 if (n.sigev_notify == SIGEV_THREAD)
1481 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1482 p = &n;
1483 }
1484 return do_mq_notify(mqdes, p);
1485}
1486
1487COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1488 const struct compat_mq_attr __user *, u_mqstat,
1489 struct compat_mq_attr __user *, u_omqstat)
1490{
1491 int ret;
1492 struct mq_attr mqstat, omqstat;
1493 struct mq_attr *new = NULL, *old = NULL;
1494
1495 if (u_mqstat) {
1496 new = &mqstat;
1497 if (get_compat_mq_attr(new, u_mqstat))
1498 return -EFAULT;
1499 }
1500 if (u_omqstat)
1501 old = &omqstat;
1502
1503 ret = do_mq_getsetattr(mqdes, new, old);
1504 if (ret || !old)
1505 return ret;
1506
1507 if (put_compat_mq_attr(old, u_omqstat))
1508 return -EFAULT;
1509 return 0;
1510}
1511#endif
1512
1513#ifdef CONFIG_COMPAT_32BIT_TIME
1514static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1515 struct timespec64 *ts)
1516{
1517 if (get_old_timespec32(ts, p))
1518 return -EFAULT;
1519 if (!timespec64_valid(ts))
1520 return -EINVAL;
1521 return 0;
1522}
1523
1524SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1525 const char __user *, u_msg_ptr,
1526 unsigned int, msg_len, unsigned int, msg_prio,
1527 const struct old_timespec32 __user *, u_abs_timeout)
1528{
1529 struct timespec64 ts, *p = NULL;
1530 if (u_abs_timeout) {
1531 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1532 if (res)
1533 return res;
1534 p = &ts;
1535 }
1536 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1537}
1538
1539SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1540 char __user *, u_msg_ptr,
1541 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1542 const struct old_timespec32 __user *, u_abs_timeout)
1543{
1544 struct timespec64 ts, *p = NULL;
1545 if (u_abs_timeout) {
1546 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1547 if (res)
1548 return res;
1549 p = &ts;
1550 }
1551 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1552}
1553#endif
1554
1555static const struct inode_operations mqueue_dir_inode_operations = {
1556 .lookup = simple_lookup,
1557 .create = mqueue_create,
1558 .unlink = mqueue_unlink,
1559};
1560
1561static const struct file_operations mqueue_file_operations = {
1562 .flush = mqueue_flush_file,
1563 .poll = mqueue_poll_file,
1564 .read = mqueue_read_file,
1565 .llseek = default_llseek,
1566};
1567
1568static const struct super_operations mqueue_super_ops = {
1569 .alloc_inode = mqueue_alloc_inode,
1570 .free_inode = mqueue_free_inode,
1571 .evict_inode = mqueue_evict_inode,
1572 .statfs = simple_statfs,
1573};
1574
1575static const struct fs_context_operations mqueue_fs_context_ops = {
1576 .free = mqueue_fs_context_free,
1577 .get_tree = mqueue_get_tree,
1578};
1579
1580static struct file_system_type mqueue_fs_type = {
1581 .name = "mqueue",
1582 .init_fs_context = mqueue_init_fs_context,
1583 .kill_sb = kill_litter_super,
1584 .fs_flags = FS_USERNS_MOUNT,
1585};
1586
1587int mq_init_ns(struct ipc_namespace *ns)
1588{
1589 struct vfsmount *m;
1590
1591 ns->mq_queues_count = 0;
1592 ns->mq_queues_max = DFLT_QUEUESMAX;
1593 ns->mq_msg_max = DFLT_MSGMAX;
1594 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1595 ns->mq_msg_default = DFLT_MSG;
1596 ns->mq_msgsize_default = DFLT_MSGSIZE;
1597
1598 m = mq_create_mount(ns);
1599 if (IS_ERR(m))
1600 return PTR_ERR(m);
1601 ns->mq_mnt = m;
1602 return 0;
1603}
1604
1605void mq_clear_sbinfo(struct ipc_namespace *ns)
1606{
1607 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1608}
1609
1610void mq_put_mnt(struct ipc_namespace *ns)
1611{
1612 kern_unmount(ns->mq_mnt);
1613}
1614
1615static int __init init_mqueue_fs(void)
1616{
1617 int error;
1618
1619 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1620 sizeof(struct mqueue_inode_info), 0,
1621 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1622 if (mqueue_inode_cachep == NULL)
1623 return -ENOMEM;
1624
1625 /* ignore failures - they are not fatal */
1626 mq_sysctl_table = mq_register_sysctl_table();
1627
1628 error = register_filesystem(&mqueue_fs_type);
1629 if (error)
1630 goto out_sysctl;
1631
1632 spin_lock_init(&mq_lock);
1633
1634 error = mq_init_ns(&init_ipc_ns);
1635 if (error)
1636 goto out_filesystem;
1637
1638 return 0;
1639
1640out_filesystem:
1641 unregister_filesystem(&mqueue_fs_type);
1642out_sysctl:
1643 if (mq_sysctl_table)
1644 unregister_sysctl_table(mq_sysctl_table);
1645 kmem_cache_destroy(mqueue_inode_cachep);
1646 return error;
1647}
1648
1649device_initcall(init_mqueue_fs);