Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 sb->s_d_flags = DCACHE_DONTCACHE;
415
416 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
417 if (IS_ERR(inode))
418 return PTR_ERR(inode);
419
420 sb->s_root = d_make_root(inode);
421 if (!sb->s_root)
422 return -ENOMEM;
423 return 0;
424}
425
426static int mqueue_get_tree(struct fs_context *fc)
427{
428 struct mqueue_fs_context *ctx = fc->fs_private;
429
430 /*
431 * With a newly created ipc namespace, we don't need to do a search
432 * for an ipc namespace match, but we still need to set s_fs_info.
433 */
434 if (ctx->newns) {
435 fc->s_fs_info = ctx->ipc_ns;
436 return get_tree_nodev(fc, mqueue_fill_super);
437 }
438 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
439}
440
441static void mqueue_fs_context_free(struct fs_context *fc)
442{
443 struct mqueue_fs_context *ctx = fc->fs_private;
444
445 put_ipc_ns(ctx->ipc_ns);
446 kfree(ctx);
447}
448
449static int mqueue_init_fs_context(struct fs_context *fc)
450{
451 struct mqueue_fs_context *ctx;
452
453 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
454 if (!ctx)
455 return -ENOMEM;
456
457 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
458 put_user_ns(fc->user_ns);
459 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
460 fc->fs_private = ctx;
461 fc->ops = &mqueue_fs_context_ops;
462 return 0;
463}
464
465/*
466 * mq_init_ns() is currently the only caller of mq_create_mount().
467 * So the ns parameter is always a newly created ipc namespace.
468 */
469static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
470{
471 struct mqueue_fs_context *ctx;
472 struct fs_context *fc;
473 struct vfsmount *mnt;
474
475 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
476 if (IS_ERR(fc))
477 return ERR_CAST(fc);
478
479 ctx = fc->fs_private;
480 ctx->newns = true;
481 put_ipc_ns(ctx->ipc_ns);
482 ctx->ipc_ns = get_ipc_ns(ns);
483 put_user_ns(fc->user_ns);
484 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
485
486 mnt = fc_mount_longterm(fc);
487 put_fs_context(fc);
488 return mnt;
489}
490
491static void init_once(void *foo)
492{
493 struct mqueue_inode_info *p = foo;
494
495 inode_init_once(&p->vfs_inode);
496}
497
498static struct inode *mqueue_alloc_inode(struct super_block *sb)
499{
500 struct mqueue_inode_info *ei;
501
502 ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL);
503 if (!ei)
504 return NULL;
505 return &ei->vfs_inode;
506}
507
508static void mqueue_free_inode(struct inode *inode)
509{
510 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
511}
512
513static void mqueue_evict_inode(struct inode *inode)
514{
515 struct mqueue_inode_info *info;
516 struct ipc_namespace *ipc_ns;
517 struct msg_msg *msg, *nmsg;
518 LIST_HEAD(tmp_msg);
519
520 clear_inode(inode);
521
522 if (S_ISDIR(inode->i_mode))
523 return;
524
525 ipc_ns = get_ns_from_inode(inode);
526 info = MQUEUE_I(inode);
527 spin_lock(&info->lock);
528 while ((msg = msg_get(info)) != NULL)
529 list_add_tail(&msg->m_list, &tmp_msg);
530 kfree(info->node_cache);
531 spin_unlock(&info->lock);
532
533 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
534 list_del(&msg->m_list);
535 free_msg(msg);
536 }
537
538 if (info->ucounts) {
539 unsigned long mq_bytes, mq_treesize;
540
541 /* Total amount of bytes accounted for the mqueue */
542 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
543 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
544 sizeof(struct posix_msg_tree_node);
545
546 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
547 info->attr.mq_msgsize);
548
549 spin_lock(&mq_lock);
550 dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
551 /*
552 * get_ns_from_inode() ensures that the
553 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
554 * to which we now hold a reference, or it is NULL.
555 * We can't put it here under mq_lock, though.
556 */
557 if (ipc_ns)
558 ipc_ns->mq_queues_count--;
559 spin_unlock(&mq_lock);
560 put_ucounts(info->ucounts);
561 info->ucounts = NULL;
562 }
563 if (ipc_ns)
564 put_ipc_ns(ipc_ns);
565}
566
567static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
568{
569 struct inode *dir = dentry->d_parent->d_inode;
570 struct inode *inode;
571 struct mq_attr *attr = arg;
572 int error;
573 struct ipc_namespace *ipc_ns;
574
575 spin_lock(&mq_lock);
576 ipc_ns = __get_ns_from_inode(dir);
577 if (!ipc_ns) {
578 error = -EACCES;
579 goto out_unlock;
580 }
581
582 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
583 !capable(CAP_SYS_RESOURCE)) {
584 error = -ENOSPC;
585 goto out_unlock;
586 }
587 ipc_ns->mq_queues_count++;
588 spin_unlock(&mq_lock);
589
590 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
591 if (IS_ERR(inode)) {
592 error = PTR_ERR(inode);
593 spin_lock(&mq_lock);
594 ipc_ns->mq_queues_count--;
595 goto out_unlock;
596 }
597
598 put_ipc_ns(ipc_ns);
599 dir->i_size += DIRENT_SIZE;
600 simple_inode_init_ts(dir);
601
602 d_make_persistent(dentry, inode);
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 dir->i_size -= DIRENT_SIZE;
620 return simple_unlink(dir, dentry);
621}
622
623/*
624* This is routine for system read from queue file.
625* To avoid mess with doing here some sort of mq_receive we allow
626* to read only queue size & notification info (the only values
627* that are interesting from user point of view and aren't accessible
628* through std routines)
629*/
630static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
631 size_t count, loff_t *off)
632{
633 struct inode *inode = file_inode(filp);
634 struct mqueue_inode_info *info = MQUEUE_I(inode);
635 char buffer[FILENT_SIZE];
636 ssize_t ret;
637
638 spin_lock(&info->lock);
639 snprintf(buffer, sizeof(buffer),
640 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
641 info->qsize,
642 info->notify_owner ? info->notify.sigev_notify : 0,
643 (info->notify_owner &&
644 info->notify.sigev_notify == SIGEV_SIGNAL) ?
645 info->notify.sigev_signo : 0,
646 pid_vnr(info->notify_owner));
647 spin_unlock(&info->lock);
648 buffer[sizeof(buffer)-1] = '\0';
649
650 ret = simple_read_from_buffer(u_data, count, off, buffer,
651 strlen(buffer));
652 if (ret <= 0)
653 return ret;
654
655 inode_set_atime_to_ts(inode, inode_set_ctime_current(inode));
656 return ret;
657}
658
659static int mqueue_flush_file(struct file *filp, fl_owner_t id)
660{
661 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
662
663 spin_lock(&info->lock);
664 if (task_tgid(current) == info->notify_owner)
665 remove_notification(info);
666
667 spin_unlock(&info->lock);
668 return 0;
669}
670
671static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
672{
673 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
674 __poll_t retval = 0;
675
676 poll_wait(filp, &info->wait_q, poll_tab);
677
678 spin_lock(&info->lock);
679 if (info->attr.mq_curmsgs)
680 retval = EPOLLIN | EPOLLRDNORM;
681
682 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
683 retval |= EPOLLOUT | EPOLLWRNORM;
684 spin_unlock(&info->lock);
685
686 return retval;
687}
688
689/* Adds current to info->e_wait_q[sr] before element with smaller prio */
690static void wq_add(struct mqueue_inode_info *info, int sr,
691 struct ext_wait_queue *ewp)
692{
693 struct ext_wait_queue *walk;
694
695 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
696 if (walk->task->prio <= current->prio) {
697 list_add_tail(&ewp->list, &walk->list);
698 return;
699 }
700 }
701 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
702}
703
704/*
705 * Puts current task to sleep. Caller must hold queue lock. After return
706 * lock isn't held.
707 * sr: SEND or RECV
708 */
709static int wq_sleep(struct mqueue_inode_info *info, int sr,
710 ktime_t *timeout, struct ext_wait_queue *ewp)
711 __releases(&info->lock)
712{
713 int retval;
714 signed long time;
715
716 wq_add(info, sr, ewp);
717
718 for (;;) {
719 /* memory barrier not required, we hold info->lock */
720 __set_current_state(TASK_INTERRUPTIBLE);
721
722 spin_unlock(&info->lock);
723 time = schedule_hrtimeout_range_clock(timeout, 0,
724 HRTIMER_MODE_ABS, CLOCK_REALTIME);
725
726 if (READ_ONCE(ewp->state) == STATE_READY) {
727 /* see MQ_BARRIER for purpose/pairing */
728 smp_acquire__after_ctrl_dep();
729 retval = 0;
730 goto out;
731 }
732 spin_lock(&info->lock);
733
734 /* we hold info->lock, so no memory barrier required */
735 if (READ_ONCE(ewp->state) == STATE_READY) {
736 retval = 0;
737 goto out_unlock;
738 }
739 if (signal_pending(current)) {
740 retval = -ERESTARTSYS;
741 break;
742 }
743 if (time == 0) {
744 retval = -ETIMEDOUT;
745 break;
746 }
747 }
748 list_del(&ewp->list);
749out_unlock:
750 spin_unlock(&info->lock);
751out:
752 return retval;
753}
754
755/*
756 * Returns waiting task that should be serviced first or NULL if none exists
757 */
758static struct ext_wait_queue *wq_get_first_waiter(
759 struct mqueue_inode_info *info, int sr)
760{
761 struct list_head *ptr;
762
763 ptr = info->e_wait_q[sr].list.prev;
764 if (ptr == &info->e_wait_q[sr].list)
765 return NULL;
766 return list_entry(ptr, struct ext_wait_queue, list);
767}
768
769
770static inline void set_cookie(struct sk_buff *skb, char code)
771{
772 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
773}
774
775/*
776 * The next function is only to split too long sys_mq_timedsend
777 */
778static void __do_notify(struct mqueue_inode_info *info)
779{
780 /* notification
781 * invoked when there is registered process and there isn't process
782 * waiting synchronously for message AND state of queue changed from
783 * empty to not empty. Here we are sure that no one is waiting
784 * synchronously. */
785 if (info->notify_owner &&
786 info->attr.mq_curmsgs == 1) {
787 switch (info->notify.sigev_notify) {
788 case SIGEV_NONE:
789 break;
790 case SIGEV_SIGNAL: {
791 struct kernel_siginfo sig_i;
792 struct task_struct *task;
793
794 /* do_mq_notify() accepts sigev_signo == 0, why?? */
795 if (!info->notify.sigev_signo)
796 break;
797
798 clear_siginfo(&sig_i);
799 sig_i.si_signo = info->notify.sigev_signo;
800 sig_i.si_errno = 0;
801 sig_i.si_code = SI_MESGQ;
802 sig_i.si_value = info->notify.sigev_value;
803 rcu_read_lock();
804 /* map current pid/uid into info->owner's namespaces */
805 sig_i.si_pid = task_tgid_nr_ns(current,
806 ns_of_pid(info->notify_owner));
807 sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
808 current_uid());
809 /*
810 * We can't use kill_pid_info(), this signal should
811 * bypass check_kill_permission(). It is from kernel
812 * but si_fromuser() can't know this.
813 * We do check the self_exec_id, to avoid sending
814 * signals to programs that don't expect them.
815 */
816 task = pid_task(info->notify_owner, PIDTYPE_TGID);
817 if (task && task->self_exec_id ==
818 info->notify_self_exec_id) {
819 do_send_sig_info(info->notify.sigev_signo,
820 &sig_i, task, PIDTYPE_TGID);
821 }
822 rcu_read_unlock();
823 break;
824 }
825 case SIGEV_THREAD:
826 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
827 netlink_sendskb(info->notify_sock, info->notify_cookie);
828 break;
829 }
830 /* after notification unregisters process */
831 put_pid(info->notify_owner);
832 put_user_ns(info->notify_user_ns);
833 info->notify_owner = NULL;
834 info->notify_user_ns = NULL;
835 }
836 wake_up(&info->wait_q);
837}
838
839static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
840 struct timespec64 *ts)
841{
842 if (get_timespec64(ts, u_abs_timeout))
843 return -EFAULT;
844 if (!timespec64_valid(ts))
845 return -EINVAL;
846 return 0;
847}
848
849static void remove_notification(struct mqueue_inode_info *info)
850{
851 if (info->notify_owner != NULL &&
852 info->notify.sigev_notify == SIGEV_THREAD) {
853 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
854 netlink_sendskb(info->notify_sock, info->notify_cookie);
855 }
856 put_pid(info->notify_owner);
857 put_user_ns(info->notify_user_ns);
858 info->notify_owner = NULL;
859 info->notify_user_ns = NULL;
860}
861
862static int prepare_open(struct dentry *dentry, int oflag, int ro,
863 umode_t mode, struct filename *name,
864 struct mq_attr *attr)
865{
866 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
867 MAY_READ | MAY_WRITE };
868 int acc;
869
870 if (d_really_is_negative(dentry)) {
871 if (!(oflag & O_CREAT))
872 return -ENOENT;
873 if (ro)
874 return ro;
875 audit_inode_parent_hidden(name, dentry->d_parent);
876 return vfs_mkobj(dentry, mode & ~current_umask(),
877 mqueue_create_attr, attr);
878 }
879 /* it already existed */
880 audit_inode(name, dentry, 0);
881 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
882 return -EEXIST;
883 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
884 return -EINVAL;
885 acc = oflag2acc[oflag & O_ACCMODE];
886 return inode_permission(&nop_mnt_idmap, d_inode(dentry), acc);
887}
888
889static struct file *mqueue_file_open(struct filename *name,
890 struct vfsmount *mnt, int oflag, int ro,
891 umode_t mode, struct mq_attr *attr)
892{
893 struct dentry *dentry;
894 struct file *file;
895 int ret;
896
897 dentry = start_creating_noperm(mnt->mnt_root, &QSTR(name->name));
898 if (IS_ERR(dentry))
899 return ERR_CAST(dentry);
900
901 ret = prepare_open(dentry, oflag, ro, mode, name, attr);
902 file = ERR_PTR(ret);
903 if (!ret) {
904 const struct path path = { .mnt = mnt, .dentry = dentry };
905 file = dentry_open(&path, oflag, current_cred());
906 }
907
908 end_creating(dentry);
909 return file;
910}
911
912static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
913 struct mq_attr *attr)
914{
915 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
916 int fd, ro;
917
918 audit_mq_open(oflag, mode, attr);
919
920 CLASS(filename, name)(u_name);
921 if (IS_ERR(name))
922 return PTR_ERR(name);
923
924 ro = mnt_want_write(mnt); /* we'll drop it in any case */
925 fd = FD_ADD(O_CLOEXEC, mqueue_file_open(name, mnt, oflag, ro, mode, attr));
926 if (!ro)
927 mnt_drop_write(mnt);
928 return fd;
929}
930
931SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
932 struct mq_attr __user *, u_attr)
933{
934 struct mq_attr attr;
935 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
936 return -EFAULT;
937
938 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
939}
940
941SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
942{
943 int err;
944 struct dentry *dentry;
945 struct inode *inode;
946 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
947 struct vfsmount *mnt = ipc_ns->mq_mnt;
948 CLASS(filename, name)(u_name);
949
950 if (IS_ERR(name))
951 return PTR_ERR(name);
952
953 audit_inode_parent_hidden(name, mnt->mnt_root);
954 err = mnt_want_write(mnt);
955 if (err)
956 return err;
957 dentry = start_removing_noperm(mnt->mnt_root, &QSTR(name->name));
958 if (IS_ERR(dentry)) {
959 err = PTR_ERR(dentry);
960 goto out_drop_write;
961 }
962
963 inode = d_inode(dentry);
964 ihold(inode);
965 err = vfs_unlink(&nop_mnt_idmap, d_inode(mnt->mnt_root),
966 dentry, NULL);
967 end_removing(dentry);
968 iput(inode);
969
970out_drop_write:
971 mnt_drop_write(mnt);
972 return err;
973}
974
975/* Pipelined send and receive functions.
976 *
977 * If a receiver finds no waiting message, then it registers itself in the
978 * list of waiting receivers. A sender checks that list before adding the new
979 * message into the message array. If there is a waiting receiver, then it
980 * bypasses the message array and directly hands the message over to the
981 * receiver. The receiver accepts the message and returns without grabbing the
982 * queue spinlock:
983 *
984 * - Set pointer to message.
985 * - Queue the receiver task for later wakeup (without the info->lock).
986 * - Update its state to STATE_READY. Now the receiver can continue.
987 * - Wake up the process after the lock is dropped. Should the process wake up
988 * before this wakeup (due to a timeout or a signal) it will either see
989 * STATE_READY and continue or acquire the lock to check the state again.
990 *
991 * The same algorithm is used for senders.
992 */
993
994static inline void __pipelined_op(struct wake_q_head *wake_q,
995 struct mqueue_inode_info *info,
996 struct ext_wait_queue *this)
997{
998 struct task_struct *task;
999
1000 list_del(&this->list);
1001 task = get_task_struct(this->task);
1002
1003 /* see MQ_BARRIER for purpose/pairing */
1004 smp_store_release(&this->state, STATE_READY);
1005 wake_q_add_safe(wake_q, task);
1006}
1007
1008/* pipelined_send() - send a message directly to the task waiting in
1009 * sys_mq_timedreceive() (without inserting message into a queue).
1010 */
1011static inline void pipelined_send(struct wake_q_head *wake_q,
1012 struct mqueue_inode_info *info,
1013 struct msg_msg *message,
1014 struct ext_wait_queue *receiver)
1015{
1016 receiver->msg = message;
1017 __pipelined_op(wake_q, info, receiver);
1018}
1019
1020/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1021 * gets its message and put to the queue (we have one free place for sure). */
1022static inline void pipelined_receive(struct wake_q_head *wake_q,
1023 struct mqueue_inode_info *info)
1024{
1025 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1026
1027 if (!sender) {
1028 /* for poll */
1029 wake_up_interruptible(&info->wait_q);
1030 return;
1031 }
1032 if (msg_insert(sender->msg, info))
1033 return;
1034
1035 __pipelined_op(wake_q, info, sender);
1036}
1037
1038static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1039 size_t msg_len, unsigned int msg_prio,
1040 struct timespec64 *ts)
1041{
1042 struct inode *inode;
1043 struct ext_wait_queue wait;
1044 struct ext_wait_queue *receiver;
1045 struct msg_msg *msg_ptr;
1046 struct mqueue_inode_info *info;
1047 ktime_t expires, *timeout = NULL;
1048 struct posix_msg_tree_node *new_leaf = NULL;
1049 int ret = 0;
1050 DEFINE_WAKE_Q(wake_q);
1051
1052 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1053 return -EINVAL;
1054
1055 if (ts) {
1056 expires = timespec64_to_ktime(*ts);
1057 timeout = &expires;
1058 }
1059
1060 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1061
1062 CLASS(fd, f)(mqdes);
1063 if (fd_empty(f))
1064 return -EBADF;
1065
1066 inode = file_inode(fd_file(f));
1067 if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1068 return -EBADF;
1069 info = MQUEUE_I(inode);
1070 audit_file(fd_file(f));
1071
1072 if (unlikely(!(fd_file(f)->f_mode & FMODE_WRITE)))
1073 return -EBADF;
1074
1075 if (unlikely(msg_len > info->attr.mq_msgsize))
1076 return -EMSGSIZE;
1077
1078 /* First try to allocate memory, before doing anything with
1079 * existing queues. */
1080 msg_ptr = load_msg(u_msg_ptr, msg_len);
1081 if (IS_ERR(msg_ptr))
1082 return PTR_ERR(msg_ptr);
1083 msg_ptr->m_ts = msg_len;
1084 msg_ptr->m_type = msg_prio;
1085
1086 /*
1087 * msg_insert really wants us to have a valid, spare node struct so
1088 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1089 * fall back to that if necessary.
1090 */
1091 if (!info->node_cache)
1092 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1093
1094 spin_lock(&info->lock);
1095
1096 if (!info->node_cache && new_leaf) {
1097 /* Save our speculative allocation into the cache */
1098 INIT_LIST_HEAD(&new_leaf->msg_list);
1099 info->node_cache = new_leaf;
1100 new_leaf = NULL;
1101 } else {
1102 kfree(new_leaf);
1103 }
1104
1105 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1106 if (fd_file(f)->f_flags & O_NONBLOCK) {
1107 ret = -EAGAIN;
1108 } else {
1109 wait.task = current;
1110 wait.msg = (void *) msg_ptr;
1111
1112 /* memory barrier not required, we hold info->lock */
1113 WRITE_ONCE(wait.state, STATE_NONE);
1114 ret = wq_sleep(info, SEND, timeout, &wait);
1115 /*
1116 * wq_sleep must be called with info->lock held, and
1117 * returns with the lock released
1118 */
1119 goto out_free;
1120 }
1121 } else {
1122 receiver = wq_get_first_waiter(info, RECV);
1123 if (receiver) {
1124 pipelined_send(&wake_q, info, msg_ptr, receiver);
1125 } else {
1126 /* adds message to the queue */
1127 ret = msg_insert(msg_ptr, info);
1128 if (ret)
1129 goto out_unlock;
1130 __do_notify(info);
1131 }
1132 simple_inode_init_ts(inode);
1133 }
1134out_unlock:
1135 spin_unlock(&info->lock);
1136 wake_up_q(&wake_q);
1137out_free:
1138 if (ret)
1139 free_msg(msg_ptr);
1140 return ret;
1141}
1142
1143static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1144 size_t msg_len, unsigned int __user *u_msg_prio,
1145 struct timespec64 *ts)
1146{
1147 ssize_t ret;
1148 struct msg_msg *msg_ptr;
1149 struct inode *inode;
1150 struct mqueue_inode_info *info;
1151 struct ext_wait_queue wait;
1152 ktime_t expires, *timeout = NULL;
1153 struct posix_msg_tree_node *new_leaf = NULL;
1154
1155 if (ts) {
1156 expires = timespec64_to_ktime(*ts);
1157 timeout = &expires;
1158 }
1159
1160 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1161
1162 CLASS(fd, f)(mqdes);
1163 if (fd_empty(f))
1164 return -EBADF;
1165
1166 inode = file_inode(fd_file(f));
1167 if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1168 return -EBADF;
1169 info = MQUEUE_I(inode);
1170 audit_file(fd_file(f));
1171
1172 if (unlikely(!(fd_file(f)->f_mode & FMODE_READ)))
1173 return -EBADF;
1174
1175 /* checks if buffer is big enough */
1176 if (unlikely(msg_len < info->attr.mq_msgsize))
1177 return -EMSGSIZE;
1178
1179 /*
1180 * msg_insert really wants us to have a valid, spare node struct so
1181 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1182 * fall back to that if necessary.
1183 */
1184 if (!info->node_cache)
1185 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1186
1187 spin_lock(&info->lock);
1188
1189 if (!info->node_cache && new_leaf) {
1190 /* Save our speculative allocation into the cache */
1191 INIT_LIST_HEAD(&new_leaf->msg_list);
1192 info->node_cache = new_leaf;
1193 } else {
1194 kfree(new_leaf);
1195 }
1196
1197 if (info->attr.mq_curmsgs == 0) {
1198 if (fd_file(f)->f_flags & O_NONBLOCK) {
1199 spin_unlock(&info->lock);
1200 ret = -EAGAIN;
1201 } else {
1202 wait.task = current;
1203
1204 /* memory barrier not required, we hold info->lock */
1205 WRITE_ONCE(wait.state, STATE_NONE);
1206 ret = wq_sleep(info, RECV, timeout, &wait);
1207 msg_ptr = wait.msg;
1208 }
1209 } else {
1210 DEFINE_WAKE_Q(wake_q);
1211
1212 msg_ptr = msg_get(info);
1213
1214 simple_inode_init_ts(inode);
1215
1216 /* There is now free space in queue. */
1217 pipelined_receive(&wake_q, info);
1218 spin_unlock(&info->lock);
1219 wake_up_q(&wake_q);
1220 ret = 0;
1221 }
1222 if (ret == 0) {
1223 ret = msg_ptr->m_ts;
1224
1225 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1226 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1227 ret = -EFAULT;
1228 }
1229 free_msg(msg_ptr);
1230 }
1231 return ret;
1232}
1233
1234SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1235 size_t, msg_len, unsigned int, msg_prio,
1236 const struct __kernel_timespec __user *, u_abs_timeout)
1237{
1238 struct timespec64 ts, *p = NULL;
1239 if (u_abs_timeout) {
1240 int res = prepare_timeout(u_abs_timeout, &ts);
1241 if (res)
1242 return res;
1243 p = &ts;
1244 }
1245 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1246}
1247
1248SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1249 size_t, msg_len, unsigned int __user *, u_msg_prio,
1250 const struct __kernel_timespec __user *, u_abs_timeout)
1251{
1252 struct timespec64 ts, *p = NULL;
1253 if (u_abs_timeout) {
1254 int res = prepare_timeout(u_abs_timeout, &ts);
1255 if (res)
1256 return res;
1257 p = &ts;
1258 }
1259 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1260}
1261
1262/*
1263 * Notes: the case when user wants us to deregister (with NULL as pointer)
1264 * and he isn't currently owner of notification, will be silently discarded.
1265 * It isn't explicitly defined in the POSIX.
1266 */
1267static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1268{
1269 int ret;
1270 struct sock *sock;
1271 struct inode *inode;
1272 struct mqueue_inode_info *info;
1273 struct sk_buff *nc;
1274
1275 audit_mq_notify(mqdes, notification);
1276
1277 nc = NULL;
1278 sock = NULL;
1279 if (notification != NULL) {
1280 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1281 notification->sigev_notify != SIGEV_SIGNAL &&
1282 notification->sigev_notify != SIGEV_THREAD))
1283 return -EINVAL;
1284 if (notification->sigev_notify == SIGEV_SIGNAL &&
1285 !valid_signal(notification->sigev_signo)) {
1286 return -EINVAL;
1287 }
1288 if (notification->sigev_notify == SIGEV_THREAD) {
1289 long timeo;
1290
1291 /* create the notify skb */
1292 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1293 if (!nc)
1294 return -ENOMEM;
1295
1296 if (copy_from_user(nc->data,
1297 notification->sigev_value.sival_ptr,
1298 NOTIFY_COOKIE_LEN)) {
1299 kfree_skb(nc);
1300 return -EFAULT;
1301 }
1302
1303 /* TODO: add a header? */
1304 skb_put(nc, NOTIFY_COOKIE_LEN);
1305 /* and attach it to the socket */
1306retry:
1307 sock = netlink_getsockbyfd(notification->sigev_signo);
1308 if (IS_ERR(sock)) {
1309 kfree_skb(nc);
1310 return PTR_ERR(sock);
1311 }
1312
1313 timeo = MAX_SCHEDULE_TIMEOUT;
1314 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1315 if (ret == 1)
1316 goto retry;
1317 if (ret)
1318 return ret;
1319 }
1320 }
1321
1322 CLASS(fd, f)(mqdes);
1323 if (fd_empty(f)) {
1324 ret = -EBADF;
1325 goto out;
1326 }
1327
1328 inode = file_inode(fd_file(f));
1329 if (unlikely(fd_file(f)->f_op != &mqueue_file_operations)) {
1330 ret = -EBADF;
1331 goto out;
1332 }
1333 info = MQUEUE_I(inode);
1334
1335 ret = 0;
1336 spin_lock(&info->lock);
1337 if (notification == NULL) {
1338 if (info->notify_owner == task_tgid(current)) {
1339 remove_notification(info);
1340 inode_set_atime_to_ts(inode,
1341 inode_set_ctime_current(inode));
1342 }
1343 } else if (info->notify_owner != NULL) {
1344 ret = -EBUSY;
1345 } else {
1346 switch (notification->sigev_notify) {
1347 case SIGEV_NONE:
1348 info->notify.sigev_notify = SIGEV_NONE;
1349 break;
1350 case SIGEV_THREAD:
1351 info->notify_sock = sock;
1352 info->notify_cookie = nc;
1353 sock = NULL;
1354 nc = NULL;
1355 info->notify.sigev_notify = SIGEV_THREAD;
1356 break;
1357 case SIGEV_SIGNAL:
1358 info->notify.sigev_signo = notification->sigev_signo;
1359 info->notify.sigev_value = notification->sigev_value;
1360 info->notify.sigev_notify = SIGEV_SIGNAL;
1361 info->notify_self_exec_id = current->self_exec_id;
1362 break;
1363 }
1364
1365 info->notify_owner = get_pid(task_tgid(current));
1366 info->notify_user_ns = get_user_ns(current_user_ns());
1367 inode_set_atime_to_ts(inode, inode_set_ctime_current(inode));
1368 }
1369 spin_unlock(&info->lock);
1370out:
1371 if (sock)
1372 netlink_detachskb(sock, nc);
1373 return ret;
1374}
1375
1376SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1377 const struct sigevent __user *, u_notification)
1378{
1379 struct sigevent n, *p = NULL;
1380 if (u_notification) {
1381 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1382 return -EFAULT;
1383 p = &n;
1384 }
1385 return do_mq_notify(mqdes, p);
1386}
1387
1388static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1389{
1390 struct inode *inode;
1391 struct mqueue_inode_info *info;
1392
1393 if (new && (new->mq_flags & (~O_NONBLOCK)))
1394 return -EINVAL;
1395
1396 CLASS(fd, f)(mqdes);
1397 if (fd_empty(f))
1398 return -EBADF;
1399
1400 if (unlikely(fd_file(f)->f_op != &mqueue_file_operations))
1401 return -EBADF;
1402
1403 inode = file_inode(fd_file(f));
1404 info = MQUEUE_I(inode);
1405
1406 spin_lock(&info->lock);
1407
1408 if (old) {
1409 *old = info->attr;
1410 old->mq_flags = fd_file(f)->f_flags & O_NONBLOCK;
1411 }
1412 if (new) {
1413 audit_mq_getsetattr(mqdes, new);
1414 spin_lock(&fd_file(f)->f_lock);
1415 if (new->mq_flags & O_NONBLOCK)
1416 fd_file(f)->f_flags |= O_NONBLOCK;
1417 else
1418 fd_file(f)->f_flags &= ~O_NONBLOCK;
1419 spin_unlock(&fd_file(f)->f_lock);
1420
1421 inode_set_atime_to_ts(inode, inode_set_ctime_current(inode));
1422 }
1423
1424 spin_unlock(&info->lock);
1425 return 0;
1426}
1427
1428SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1429 const struct mq_attr __user *, u_mqstat,
1430 struct mq_attr __user *, u_omqstat)
1431{
1432 int ret;
1433 struct mq_attr mqstat, omqstat;
1434 struct mq_attr *new = NULL, *old = NULL;
1435
1436 if (u_mqstat) {
1437 new = &mqstat;
1438 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1439 return -EFAULT;
1440 }
1441 if (u_omqstat)
1442 old = &omqstat;
1443
1444 ret = do_mq_getsetattr(mqdes, new, old);
1445 if (ret || !old)
1446 return ret;
1447
1448 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1449 return -EFAULT;
1450 return 0;
1451}
1452
1453#ifdef CONFIG_COMPAT
1454
1455struct compat_mq_attr {
1456 compat_long_t mq_flags; /* message queue flags */
1457 compat_long_t mq_maxmsg; /* maximum number of messages */
1458 compat_long_t mq_msgsize; /* maximum message size */
1459 compat_long_t mq_curmsgs; /* number of messages currently queued */
1460 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1461};
1462
1463static inline int get_compat_mq_attr(struct mq_attr *attr,
1464 const struct compat_mq_attr __user *uattr)
1465{
1466 struct compat_mq_attr v;
1467
1468 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1469 return -EFAULT;
1470
1471 memset(attr, 0, sizeof(*attr));
1472 attr->mq_flags = v.mq_flags;
1473 attr->mq_maxmsg = v.mq_maxmsg;
1474 attr->mq_msgsize = v.mq_msgsize;
1475 attr->mq_curmsgs = v.mq_curmsgs;
1476 return 0;
1477}
1478
1479static inline int put_compat_mq_attr(const struct mq_attr *attr,
1480 struct compat_mq_attr __user *uattr)
1481{
1482 struct compat_mq_attr v;
1483
1484 memset(&v, 0, sizeof(v));
1485 v.mq_flags = attr->mq_flags;
1486 v.mq_maxmsg = attr->mq_maxmsg;
1487 v.mq_msgsize = attr->mq_msgsize;
1488 v.mq_curmsgs = attr->mq_curmsgs;
1489 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1490 return -EFAULT;
1491 return 0;
1492}
1493
1494COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1495 int, oflag, compat_mode_t, mode,
1496 struct compat_mq_attr __user *, u_attr)
1497{
1498 struct mq_attr attr, *p = NULL;
1499 if (u_attr && oflag & O_CREAT) {
1500 p = &attr;
1501 if (get_compat_mq_attr(&attr, u_attr))
1502 return -EFAULT;
1503 }
1504 return do_mq_open(u_name, oflag, mode, p);
1505}
1506
1507COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1508 const struct compat_sigevent __user *, u_notification)
1509{
1510 struct sigevent n, *p = NULL;
1511 if (u_notification) {
1512 if (get_compat_sigevent(&n, u_notification))
1513 return -EFAULT;
1514 if (n.sigev_notify == SIGEV_THREAD)
1515 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1516 p = &n;
1517 }
1518 return do_mq_notify(mqdes, p);
1519}
1520
1521COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1522 const struct compat_mq_attr __user *, u_mqstat,
1523 struct compat_mq_attr __user *, u_omqstat)
1524{
1525 int ret;
1526 struct mq_attr mqstat, omqstat;
1527 struct mq_attr *new = NULL, *old = NULL;
1528
1529 if (u_mqstat) {
1530 new = &mqstat;
1531 if (get_compat_mq_attr(new, u_mqstat))
1532 return -EFAULT;
1533 }
1534 if (u_omqstat)
1535 old = &omqstat;
1536
1537 ret = do_mq_getsetattr(mqdes, new, old);
1538 if (ret || !old)
1539 return ret;
1540
1541 if (put_compat_mq_attr(old, u_omqstat))
1542 return -EFAULT;
1543 return 0;
1544}
1545#endif
1546
1547#ifdef CONFIG_COMPAT_32BIT_TIME
1548static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1549 struct timespec64 *ts)
1550{
1551 if (get_old_timespec32(ts, p))
1552 return -EFAULT;
1553 if (!timespec64_valid(ts))
1554 return -EINVAL;
1555 return 0;
1556}
1557
1558SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1559 const char __user *, u_msg_ptr,
1560 unsigned int, msg_len, unsigned int, msg_prio,
1561 const struct old_timespec32 __user *, u_abs_timeout)
1562{
1563 struct timespec64 ts, *p = NULL;
1564 if (u_abs_timeout) {
1565 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1566 if (res)
1567 return res;
1568 p = &ts;
1569 }
1570 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1571}
1572
1573SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1574 char __user *, u_msg_ptr,
1575 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1576 const struct old_timespec32 __user *, u_abs_timeout)
1577{
1578 struct timespec64 ts, *p = NULL;
1579 if (u_abs_timeout) {
1580 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1581 if (res)
1582 return res;
1583 p = &ts;
1584 }
1585 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1586}
1587#endif
1588
1589static const struct inode_operations mqueue_dir_inode_operations = {
1590 .lookup = simple_lookup,
1591 .create = mqueue_create,
1592 .unlink = mqueue_unlink,
1593};
1594
1595static const struct file_operations mqueue_file_operations = {
1596 .flush = mqueue_flush_file,
1597 .poll = mqueue_poll_file,
1598 .read = mqueue_read_file,
1599 .llseek = default_llseek,
1600};
1601
1602static const struct super_operations mqueue_super_ops = {
1603 .alloc_inode = mqueue_alloc_inode,
1604 .free_inode = mqueue_free_inode,
1605 .evict_inode = mqueue_evict_inode,
1606 .statfs = simple_statfs,
1607};
1608
1609static const struct fs_context_operations mqueue_fs_context_ops = {
1610 .free = mqueue_fs_context_free,
1611 .get_tree = mqueue_get_tree,
1612};
1613
1614static struct file_system_type mqueue_fs_type = {
1615 .name = "mqueue",
1616 .init_fs_context = mqueue_init_fs_context,
1617 .kill_sb = kill_anon_super,
1618 .fs_flags = FS_USERNS_MOUNT,
1619};
1620
1621int mq_init_ns(struct ipc_namespace *ns)
1622{
1623 struct vfsmount *m;
1624
1625 ns->mq_queues_count = 0;
1626 ns->mq_queues_max = DFLT_QUEUESMAX;
1627 ns->mq_msg_max = DFLT_MSGMAX;
1628 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1629 ns->mq_msg_default = DFLT_MSG;
1630 ns->mq_msgsize_default = DFLT_MSGSIZE;
1631
1632 m = mq_create_mount(ns);
1633 if (IS_ERR(m))
1634 return PTR_ERR(m);
1635 ns->mq_mnt = m;
1636 return 0;
1637}
1638
1639void mq_clear_sbinfo(struct ipc_namespace *ns)
1640{
1641 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1642}
1643
1644static int __init init_mqueue_fs(void)
1645{
1646 int error;
1647
1648 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1649 sizeof(struct mqueue_inode_info), 0,
1650 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1651 if (mqueue_inode_cachep == NULL)
1652 return -ENOMEM;
1653
1654 if (!setup_mq_sysctls(&init_ipc_ns)) {
1655 pr_warn("sysctl registration failed\n");
1656 error = -ENOMEM;
1657 goto out_kmem;
1658 }
1659
1660 error = register_filesystem(&mqueue_fs_type);
1661 if (error)
1662 goto out_sysctl;
1663
1664 spin_lock_init(&mq_lock);
1665
1666 error = mq_init_ns(&init_ipc_ns);
1667 if (error)
1668 goto out_filesystem;
1669
1670 return 0;
1671
1672out_filesystem:
1673 unregister_filesystem(&mqueue_fs_type);
1674out_sysctl:
1675 retire_mq_sysctls(&init_ipc_ns);
1676out_kmem:
1677 kmem_cache_destroy(mqueue_inode_cachep);
1678 return error;
1679}
1680
1681device_initcall(init_mqueue_fs);