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(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(lock: &mq_lock);
185 ns = __get_ns_from_inode(inode);
186 spin_unlock(lock: &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(list: &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(node: &leaf->rb_node, parent, rb_link: 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(new: &msg->m_list, head: &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(objp: 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(entry: &msg->m_list);
281 if (list_empty(head: &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(list: &info->e_wait_q[0].list);
318 INIT_LIST_HEAD(list: &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(s: &info->attr, c: 0, n: 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(lock: &mq_lock);
377 msgqueue = inc_rlimit_ucounts(ucounts: info->ucounts, type: UCOUNT_RLIMIT_MSGQUEUE, v: mq_bytes);
378 if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) {
379 dec_rlimit_ucounts(ucounts: info->ucounts, type: UCOUNT_RLIMIT_MSGQUEUE, v: mq_bytes);
380 spin_unlock(lock: &mq_lock);
381 put_ucounts(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(lock: &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(error: 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, ipc_ns: ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
417 if (IS_ERR(ptr: inode))
418 return PTR_ERR(ptr: 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, fill_super: mqueue_fill_super);
437 }
438 return get_tree_keyed(fc, fill_super: mqueue_fill_super, key: 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(ns: ctx->ipc_ns);
446 kfree(objp: 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(ns: fc->user_ns);
459 fc->user_ns = get_user_ns(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(fs_type: &mqueue_fs_type, SB_KERNMOUNT);
476 if (IS_ERR(ptr: fc))
477 return ERR_CAST(ptr: fc);
478
479 ctx = fc->fs_private;
480 ctx->newns = true;
481 put_ipc_ns(ns: ctx->ipc_ns);
482 ctx->ipc_ns = get_ipc_ns(ns);
483 put_user_ns(ns: fc->user_ns);
484 fc->user_ns = get_user_ns(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(s: mqueue_inode_cachep, objp: 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(lock: &info->lock);
528 while ((msg = msg_get(info)) != NULL)
529 list_add_tail(new: &msg->m_list, head: &tmp_msg);
530 kfree(objp: info->node_cache);
531 spin_unlock(lock: &info->lock);
532
533 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
534 list_del(entry: &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(lock: &mq_lock);
550 dec_rlimit_ucounts(ucounts: info->ucounts, type: UCOUNT_RLIMIT_MSGQUEUE, v: 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(lock: &mq_lock);
560 put_ucounts(ucounts: info->ucounts);
561 info->ucounts = NULL;
562 }
563 if (ipc_ns)
564 put_ipc_ns(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(lock: &mq_lock);
576 ipc_ns = __get_ns_from_inode(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(lock: &mq_lock);
589
590 inode = mqueue_get_inode(sb: dir->i_sb, ipc_ns, mode, attr);
591 if (IS_ERR(ptr: inode)) {
592 error = PTR_ERR(ptr: inode);
593 spin_lock(lock: &mq_lock);
594 ipc_ns->mq_queues_count--;
595 goto out_unlock;
596 }
597
598 put_ipc_ns(ns: ipc_ns);
599 dir->i_size += DIRENT_SIZE;
600 simple_inode_init_ts(inode: dir);
601
602 d_instantiate(dentry, inode);
603 dget(dentry);
604 return 0;
605out_unlock:
606 spin_unlock(lock: &mq_lock);
607 if (ipc_ns)
608 put_ipc_ns(ns: ipc_ns);
609 return error;
610}
611
612static int mqueue_create(struct mnt_idmap *idmap, struct inode *dir,
613 struct dentry *dentry, umode_t mode, bool excl)
614{
615 return mqueue_create_attr(dentry, mode, NULL);
616}
617
618static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
619{
620 struct inode *inode = d_inode(dentry);
621
622 simple_inode_init_ts(inode: dir);
623 dir->i_size -= DIRENT_SIZE;
624 drop_nlink(inode);
625 dput(dentry);
626 return 0;
627}
628
629/*
630* This is routine for system read from queue file.
631* To avoid mess with doing here some sort of mq_receive we allow
632* to read only queue size & notification info (the only values
633* that are interesting from user point of view and aren't accessible
634* through std routines)
635*/
636static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
637 size_t count, loff_t *off)
638{
639 struct inode *inode = file_inode(f: filp);
640 struct mqueue_inode_info *info = MQUEUE_I(inode);
641 char buffer[FILENT_SIZE];
642 ssize_t ret;
643
644 spin_lock(lock: &info->lock);
645 snprintf(buf: buffer, size: sizeof(buffer),
646 fmt: "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
647 info->qsize,
648 info->notify_owner ? info->notify.sigev_notify : 0,
649 (info->notify_owner &&
650 info->notify.sigev_notify == SIGEV_SIGNAL) ?
651 info->notify.sigev_signo : 0,
652 pid_vnr(pid: info->notify_owner));
653 spin_unlock(lock: &info->lock);
654 buffer[sizeof(buffer)-1] = '\0';
655
656 ret = simple_read_from_buffer(to: u_data, count, ppos: off, from: buffer,
657 available: strlen(buffer));
658 if (ret <= 0)
659 return ret;
660
661 inode_set_atime_to_ts(inode, ts: inode_set_ctime_current(inode));
662 return ret;
663}
664
665static int mqueue_flush_file(struct file *filp, fl_owner_t id)
666{
667 struct mqueue_inode_info *info = MQUEUE_I(inode: file_inode(f: filp));
668
669 spin_lock(lock: &info->lock);
670 if (task_tgid(current) == info->notify_owner)
671 remove_notification(info);
672
673 spin_unlock(lock: &info->lock);
674 return 0;
675}
676
677static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
678{
679 struct mqueue_inode_info *info = MQUEUE_I(inode: file_inode(f: filp));
680 __poll_t retval = 0;
681
682 poll_wait(filp, wait_address: &info->wait_q, p: poll_tab);
683
684 spin_lock(lock: &info->lock);
685 if (info->attr.mq_curmsgs)
686 retval = EPOLLIN | EPOLLRDNORM;
687
688 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
689 retval |= EPOLLOUT | EPOLLWRNORM;
690 spin_unlock(lock: &info->lock);
691
692 return retval;
693}
694
695/* Adds current to info->e_wait_q[sr] before element with smaller prio */
696static void wq_add(struct mqueue_inode_info *info, int sr,
697 struct ext_wait_queue *ewp)
698{
699 struct ext_wait_queue *walk;
700
701 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
702 if (walk->task->prio <= current->prio) {
703 list_add_tail(new: &ewp->list, head: &walk->list);
704 return;
705 }
706 }
707 list_add_tail(new: &ewp->list, head: &info->e_wait_q[sr].list);
708}
709
710/*
711 * Puts current task to sleep. Caller must hold queue lock. After return
712 * lock isn't held.
713 * sr: SEND or RECV
714 */
715static int wq_sleep(struct mqueue_inode_info *info, int sr,
716 ktime_t *timeout, struct ext_wait_queue *ewp)
717 __releases(&info->lock)
718{
719 int retval;
720 signed long time;
721
722 wq_add(info, sr, ewp);
723
724 for (;;) {
725 /* memory barrier not required, we hold info->lock */
726 __set_current_state(TASK_INTERRUPTIBLE);
727
728 spin_unlock(lock: &info->lock);
729 time = schedule_hrtimeout_range_clock(expires: timeout, delta: 0,
730 mode: HRTIMER_MODE_ABS, CLOCK_REALTIME);
731
732 if (READ_ONCE(ewp->state) == STATE_READY) {
733 /* see MQ_BARRIER for purpose/pairing */
734 smp_acquire__after_ctrl_dep();
735 retval = 0;
736 goto out;
737 }
738 spin_lock(lock: &info->lock);
739
740 /* we hold info->lock, so no memory barrier required */
741 if (READ_ONCE(ewp->state) == STATE_READY) {
742 retval = 0;
743 goto out_unlock;
744 }
745 if (signal_pending(current)) {
746 retval = -ERESTARTSYS;
747 break;
748 }
749 if (time == 0) {
750 retval = -ETIMEDOUT;
751 break;
752 }
753 }
754 list_del(entry: &ewp->list);
755out_unlock:
756 spin_unlock(lock: &info->lock);
757out:
758 return retval;
759}
760
761/*
762 * Returns waiting task that should be serviced first or NULL if none exists
763 */
764static struct ext_wait_queue *wq_get_first_waiter(
765 struct mqueue_inode_info *info, int sr)
766{
767 struct list_head *ptr;
768
769 ptr = info->e_wait_q[sr].list.prev;
770 if (ptr == &info->e_wait_q[sr].list)
771 return NULL;
772 return list_entry(ptr, struct ext_wait_queue, list);
773}
774
775
776static inline void set_cookie(struct sk_buff *skb, char code)
777{
778 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
779}
780
781/*
782 * The next function is only to split too long sys_mq_timedsend
783 */
784static void __do_notify(struct mqueue_inode_info *info)
785{
786 /* notification
787 * invoked when there is registered process and there isn't process
788 * waiting synchronously for message AND state of queue changed from
789 * empty to not empty. Here we are sure that no one is waiting
790 * synchronously. */
791 if (info->notify_owner &&
792 info->attr.mq_curmsgs == 1) {
793 switch (info->notify.sigev_notify) {
794 case SIGEV_NONE:
795 break;
796 case SIGEV_SIGNAL: {
797 struct kernel_siginfo sig_i;
798 struct task_struct *task;
799
800 /* do_mq_notify() accepts sigev_signo == 0, why?? */
801 if (!info->notify.sigev_signo)
802 break;
803
804 clear_siginfo(info: &sig_i);
805 sig_i.si_signo = info->notify.sigev_signo;
806 sig_i.si_errno = 0;
807 sig_i.si_code = SI_MESGQ;
808 sig_i.si_value = info->notify.sigev_value;
809 rcu_read_lock();
810 /* map current pid/uid into info->owner's namespaces */
811 sig_i.si_pid = task_tgid_nr_ns(current,
812 ns: ns_of_pid(pid: info->notify_owner));
813 sig_i.si_uid = from_kuid_munged(to: info->notify_user_ns,
814 current_uid());
815 /*
816 * We can't use kill_pid_info(), this signal should
817 * bypass check_kill_permission(). It is from kernel
818 * but si_fromuser() can't know this.
819 * We do check the self_exec_id, to avoid sending
820 * signals to programs that don't expect them.
821 */
822 task = pid_task(pid: info->notify_owner, PIDTYPE_TGID);
823 if (task && task->self_exec_id ==
824 info->notify_self_exec_id) {
825 do_send_sig_info(sig: info->notify.sigev_signo,
826 info: &sig_i, p: task, type: PIDTYPE_TGID);
827 }
828 rcu_read_unlock();
829 break;
830 }
831 case SIGEV_THREAD:
832 set_cookie(skb: info->notify_cookie, NOTIFY_WOKENUP);
833 netlink_sendskb(sk: info->notify_sock, skb: info->notify_cookie);
834 break;
835 }
836 /* after notification unregisters process */
837 put_pid(pid: info->notify_owner);
838 put_user_ns(ns: info->notify_user_ns);
839 info->notify_owner = NULL;
840 info->notify_user_ns = NULL;
841 }
842 wake_up(&info->wait_q);
843}
844
845static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
846 struct timespec64 *ts)
847{
848 if (get_timespec64(ts, uts: u_abs_timeout))
849 return -EFAULT;
850 if (!timespec64_valid(ts))
851 return -EINVAL;
852 return 0;
853}
854
855static void remove_notification(struct mqueue_inode_info *info)
856{
857 if (info->notify_owner != NULL &&
858 info->notify.sigev_notify == SIGEV_THREAD) {
859 set_cookie(skb: info->notify_cookie, NOTIFY_REMOVED);
860 netlink_sendskb(sk: info->notify_sock, skb: info->notify_cookie);
861 }
862 put_pid(pid: info->notify_owner);
863 put_user_ns(ns: info->notify_user_ns);
864 info->notify_owner = NULL;
865 info->notify_user_ns = NULL;
866}
867
868static int prepare_open(struct dentry *dentry, int oflag, int ro,
869 umode_t mode, struct filename *name,
870 struct mq_attr *attr)
871{
872 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
873 MAY_READ | MAY_WRITE };
874 int acc;
875
876 if (d_really_is_negative(dentry)) {
877 if (!(oflag & O_CREAT))
878 return -ENOENT;
879 if (ro)
880 return ro;
881 audit_inode_parent_hidden(name, dentry: dentry->d_parent);
882 return vfs_mkobj(dentry, mode & ~current_umask(),
883 f: mqueue_create_attr, attr);
884 }
885 /* it already existed */
886 audit_inode(name, dentry, aflags: 0);
887 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
888 return -EEXIST;
889 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
890 return -EINVAL;
891 acc = oflag2acc[oflag & O_ACCMODE];
892 return inode_permission(&nop_mnt_idmap, d_inode(dentry), acc);
893}
894
895static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
896 struct mq_attr *attr)
897{
898 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
899 struct dentry *root = mnt->mnt_root;
900 struct filename *name;
901 struct path path;
902 int fd, error;
903 int ro;
904
905 audit_mq_open(oflag, mode, attr);
906
907 name = getname(name: u_name);
908 if (IS_ERR(ptr: name))
909 return PTR_ERR(ptr: name);
910
911 fd = get_unused_fd_flags(O_CLOEXEC);
912 if (fd < 0)
913 goto out_putname;
914
915 ro = mnt_want_write(mnt); /* we'll drop it in any case */
916 inode_lock(inode: d_inode(dentry: root));
917 path.dentry = lookup_noperm(&QSTR(name->name), root);
918 if (IS_ERR(ptr: path.dentry)) {
919 error = PTR_ERR(ptr: path.dentry);
920 goto out_putfd;
921 }
922 path.mnt = mntget(mnt);
923 error = prepare_open(dentry: path.dentry, oflag, ro, mode, name, attr);
924 if (!error) {
925 struct file *file = dentry_open(path: &path, flags: oflag, current_cred());
926 if (!IS_ERR(ptr: file))
927 fd_install(fd, file);
928 else
929 error = PTR_ERR(ptr: file);
930 }
931 path_put(&path);
932out_putfd:
933 if (error) {
934 put_unused_fd(fd);
935 fd = error;
936 }
937 inode_unlock(inode: d_inode(dentry: root));
938 if (!ro)
939 mnt_drop_write(mnt);
940out_putname:
941 putname(name);
942 return fd;
943}
944
945SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
946 struct mq_attr __user *, u_attr)
947{
948 struct mq_attr attr;
949 if (u_attr && copy_from_user(to: &attr, from: u_attr, n: sizeof(struct mq_attr)))
950 return -EFAULT;
951
952 return do_mq_open(u_name, oflag, mode, attr: u_attr ? &attr : NULL);
953}
954
955SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
956{
957 int err;
958 struct filename *name;
959 struct dentry *dentry;
960 struct inode *inode = NULL;
961 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
962 struct vfsmount *mnt = ipc_ns->mq_mnt;
963
964 name = getname(name: u_name);
965 if (IS_ERR(ptr: name))
966 return PTR_ERR(ptr: name);
967
968 audit_inode_parent_hidden(name, dentry: mnt->mnt_root);
969 err = mnt_want_write(mnt);
970 if (err)
971 goto out_name;
972 inode_lock_nested(inode: d_inode(dentry: mnt->mnt_root), subclass: I_MUTEX_PARENT);
973 dentry = lookup_noperm(&QSTR(name->name), mnt->mnt_root);
974 if (IS_ERR(ptr: dentry)) {
975 err = PTR_ERR(ptr: 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: dentry->d_parent),
985 dentry, NULL);
986 }
987 dput(dentry);
988
989out_unlock:
990 inode_unlock(inode: d_inode(dentry: 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(entry: &this->list);
1025 task = get_task_struct(t: this->task);
1026
1027 /* see MQ_BARRIER for purpose/pairing */
1028 smp_store_release(&this->state, STATE_READY);
1029 wake_q_add_safe(head: 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, this: 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(msg: sender->msg, info))
1057 return;
1058
1059 __pipelined_op(wake_q, info, this: 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: *ts);
1081 timeout = &expires;
1082 }
1083
1084 audit_mq_sendrecv(mqdes, msg_len, msg_prio, abs_timeout: ts);
1085
1086 CLASS(fd, f)(fd: 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(src: u_msg_ptr, len: msg_len);
1105 if (IS_ERR(ptr: msg_ptr))
1106 return PTR_ERR(ptr: 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(lock: &info->lock);
1119
1120 if (!info->node_cache && new_leaf) {
1121 /* Save our speculative allocation into the cache */
1122 INIT_LIST_HEAD(list: &new_leaf->msg_list);
1123 info->node_cache = new_leaf;
1124 new_leaf = NULL;
1125 } else {
1126 kfree(objp: 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, ewp: &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: &wake_q, info, message: msg_ptr, receiver);
1149 } else {
1150 /* adds message to the queue */
1151 ret = msg_insert(msg: 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(lock: &info->lock);
1160 wake_up_q(head: &wake_q);
1161out_free:
1162 if (ret)
1163 free_msg(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: *ts);
1181 timeout = &expires;
1182 }
1183
1184 audit_mq_sendrecv(mqdes, msg_len, msg_prio: 0, abs_timeout: ts);
1185
1186 CLASS(fd, f)(fd: 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(lock: &info->lock);
1212
1213 if (!info->node_cache && new_leaf) {
1214 /* Save our speculative allocation into the cache */
1215 INIT_LIST_HEAD(list: &new_leaf->msg_list);
1216 info->node_cache = new_leaf;
1217 } else {
1218 kfree(objp: new_leaf);
1219 }
1220
1221 if (info->attr.mq_curmsgs == 0) {
1222 if (fd_file(f)->f_flags & O_NONBLOCK) {
1223 spin_unlock(lock: &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, ewp: &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: &wake_q, info);
1242 spin_unlock(lock: &info->lock);
1243 wake_up_q(head: &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(dest: u_msg_ptr, msg: msg_ptr, len: msg_ptr->m_ts)) {
1251 ret = -EFAULT;
1252 }
1253 free_msg(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: &ts);
1265 if (res)
1266 return res;
1267 p = &ts;
1268 }
1269 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, ts: 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: &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, ts: 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(sig: 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(to: nc->data,
1321 from: notification->sigev_value.sival_ptr,
1322 NOTIFY_COOKIE_LEN)) {
1323 kfree_skb(skb: nc);
1324 return -EFAULT;
1325 }
1326
1327 /* TODO: add a header? */
1328 skb_put(skb: nc, NOTIFY_COOKIE_LEN);
1329 /* and attach it to the socket */
1330retry:
1331 sock = netlink_getsockbyfd(fd: notification->sigev_signo);
1332 if (IS_ERR(ptr: sock)) {
1333 kfree_skb(skb: nc);
1334 return PTR_ERR(ptr: sock);
1335 }
1336
1337 timeo = MAX_SCHEDULE_TIMEOUT;
1338 ret = netlink_attachskb(sk: sock, skb: nc, timeo: &timeo, NULL);
1339 if (ret == 1)
1340 goto retry;
1341 if (ret)
1342 return ret;
1343 }
1344 }
1345
1346 CLASS(fd, f)(fd: 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(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 ts: 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(pid: task_tgid(current));
1390 info->notify_user_ns = get_user_ns(ns: current_user_ns());
1391 inode_set_atime_to_ts(inode, ts: inode_set_ctime_current(inode));
1392 }
1393 spin_unlock(lock: &info->lock);
1394out:
1395 if (sock)
1396 netlink_detachskb(sk: sock, skb: 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(to: &n, from: u_notification, n: sizeof(struct sigevent)))
1406 return -EFAULT;
1407 p = &n;
1408 }
1409 return do_mq_notify(mqdes, notification: 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)(fd: 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(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, mqstat: new);
1438 spin_lock(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(lock: &fd_file(f)->f_lock);
1444
1445 inode_set_atime_to_ts(inode, ts: inode_set_ctime_current(inode));
1446 }
1447
1448 spin_unlock(lock: &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(to: new, from: u_mqstat, n: 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(to: u_omqstat, from: old, n: 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(to: &v, from: uattr, n: sizeof(*uattr)))
1493 return -EFAULT;
1494
1495 memset(s: attr, c: 0, n: 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(s: &v, c: 0, n: 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(to: uattr, from: &v, n: 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: &attr, uattr: u_attr))
1526 return -EFAULT;
1527 }
1528 return do_mq_open(u_name, oflag, mode, attr: 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(event: &n, u_event: u_notification))
1537 return -EFAULT;
1538 if (n.sigev_notify == SIGEV_THREAD)
1539 n.sigev_value.sival_ptr = compat_ptr(uptr: n.sigev_value.sival_int);
1540 p = &n;
1541 }
1542 return do_mq_notify(mqdes, notification: 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(attr: new, uattr: 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(attr: old, uattr: 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(p: u_abs_timeout, ts: &ts);
1590 if (res)
1591 return res;
1592 p = &ts;
1593 }
1594 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, ts: 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(p: u_abs_timeout, ts: &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, ts: 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(ptr: m))
1658 return PTR_ERR(ptr: 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(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(ns: &init_ipc_ns);
1700out_kmem:
1701 kmem_cache_destroy(s: mqueue_inode_cachep);
1702 return error;
1703}
1704
1705device_initcall(init_mqueue_fs);
1706