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