Loading...
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
5 * Michal Wronski (michal.wronski@gmail.com)
6 *
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
10 *
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
12 *
13 * This file is released under the GPL.
14 */
15
16#include <linux/capability.h>
17#include <linux/init.h>
18#include <linux/pagemap.h>
19#include <linux/file.h>
20#include <linux/mount.h>
21#include <linux/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 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
306
307 if (S_ISREG(mode)) {
308 struct mqueue_inode_info *info;
309 unsigned long mq_bytes, mq_treesize;
310
311 inode->i_fop = &mqueue_file_operations;
312 inode->i_size = FILENT_SIZE;
313 /* mqueue specific info */
314 info = MQUEUE_I(inode);
315 spin_lock_init(&info->lock);
316 init_waitqueue_head(&info->wait_q);
317 INIT_LIST_HEAD(&info->e_wait_q[0].list);
318 INIT_LIST_HEAD(&info->e_wait_q[1].list);
319 info->notify_owner = NULL;
320 info->notify_user_ns = NULL;
321 info->qsize = 0;
322 info->ucounts = NULL; /* set when all is ok */
323 info->msg_tree = RB_ROOT;
324 info->msg_tree_rightmost = NULL;
325 info->node_cache = NULL;
326 memset(&info->attr, 0, sizeof(info->attr));
327 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
328 ipc_ns->mq_msg_default);
329 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
330 ipc_ns->mq_msgsize_default);
331 if (attr) {
332 info->attr.mq_maxmsg = attr->mq_maxmsg;
333 info->attr.mq_msgsize = attr->mq_msgsize;
334 }
335 /*
336 * We used to allocate a static array of pointers and account
337 * the size of that array as well as one msg_msg struct per
338 * possible message into the queue size. That's no longer
339 * accurate as the queue is now an rbtree and will grow and
340 * shrink depending on usage patterns. We can, however, still
341 * account one msg_msg struct per message, but the nodes are
342 * allocated depending on priority usage, and most programs
343 * only use one, or a handful, of priorities. However, since
344 * this is pinned memory, we need to assume worst case, so
345 * that means the min(mq_maxmsg, max_priorities) * struct
346 * posix_msg_tree_node.
347 */
348
349 ret = -EINVAL;
350 if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
351 goto out_inode;
352 if (capable(CAP_SYS_RESOURCE)) {
353 if (info->attr.mq_maxmsg > HARD_MSGMAX ||
354 info->attr.mq_msgsize > HARD_MSGSIZEMAX)
355 goto out_inode;
356 } else {
357 if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
358 info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
359 goto out_inode;
360 }
361 ret = -EOVERFLOW;
362 /* check for overflow */
363 if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
364 goto out_inode;
365 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
366 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
367 sizeof(struct posix_msg_tree_node);
368 mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
369 if (mq_bytes + mq_treesize < mq_bytes)
370 goto out_inode;
371 mq_bytes += mq_treesize;
372 info->ucounts = get_ucounts(current_ucounts());
373 if (info->ucounts) {
374 long msgqueue;
375
376 spin_lock(&mq_lock);
377 msgqueue = inc_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
378 if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) {
379 dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
380 spin_unlock(&mq_lock);
381 put_ucounts(info->ucounts);
382 info->ucounts = NULL;
383 /* mqueue_evict_inode() releases info->messages */
384 ret = -EMFILE;
385 goto out_inode;
386 }
387 spin_unlock(&mq_lock);
388 }
389 } else if (S_ISDIR(mode)) {
390 inc_nlink(inode);
391 /* Some things misbehave if size == 0 on a directory */
392 inode->i_size = 2 * DIRENT_SIZE;
393 inode->i_op = &mqueue_dir_inode_operations;
394 inode->i_fop = &simple_dir_operations;
395 }
396
397 return inode;
398out_inode:
399 iput(inode);
400err:
401 return ERR_PTR(ret);
402}
403
404static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
405{
406 struct inode *inode;
407 struct ipc_namespace *ns = sb->s_fs_info;
408
409 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
410 sb->s_blocksize = PAGE_SIZE;
411 sb->s_blocksize_bits = PAGE_SHIFT;
412 sb->s_magic = MQUEUE_MAGIC;
413 sb->s_op = &mqueue_super_ops;
414
415 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
416 if (IS_ERR(inode))
417 return PTR_ERR(inode);
418
419 sb->s_root = d_make_root(inode);
420 if (!sb->s_root)
421 return -ENOMEM;
422 return 0;
423}
424
425static int mqueue_get_tree(struct fs_context *fc)
426{
427 struct mqueue_fs_context *ctx = fc->fs_private;
428
429 /*
430 * With a newly created ipc namespace, we don't need to do a search
431 * for an ipc namespace match, but we still need to set s_fs_info.
432 */
433 if (ctx->newns) {
434 fc->s_fs_info = ctx->ipc_ns;
435 return get_tree_nodev(fc, mqueue_fill_super);
436 }
437 return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
438}
439
440static void mqueue_fs_context_free(struct fs_context *fc)
441{
442 struct mqueue_fs_context *ctx = fc->fs_private;
443
444 put_ipc_ns(ctx->ipc_ns);
445 kfree(ctx);
446}
447
448static int mqueue_init_fs_context(struct fs_context *fc)
449{
450 struct mqueue_fs_context *ctx;
451
452 ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
453 if (!ctx)
454 return -ENOMEM;
455
456 ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
457 put_user_ns(fc->user_ns);
458 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
459 fc->fs_private = ctx;
460 fc->ops = &mqueue_fs_context_ops;
461 return 0;
462}
463
464/*
465 * mq_init_ns() is currently the only caller of mq_create_mount().
466 * So the ns parameter is always a newly created ipc namespace.
467 */
468static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
469{
470 struct mqueue_fs_context *ctx;
471 struct fs_context *fc;
472 struct vfsmount *mnt;
473
474 fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
475 if (IS_ERR(fc))
476 return ERR_CAST(fc);
477
478 ctx = fc->fs_private;
479 ctx->newns = true;
480 put_ipc_ns(ctx->ipc_ns);
481 ctx->ipc_ns = get_ipc_ns(ns);
482 put_user_ns(fc->user_ns);
483 fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
484
485 mnt = fc_mount(fc);
486 put_fs_context(fc);
487 return mnt;
488}
489
490static void init_once(void *foo)
491{
492 struct mqueue_inode_info *p = foo;
493
494 inode_init_once(&p->vfs_inode);
495}
496
497static struct inode *mqueue_alloc_inode(struct super_block *sb)
498{
499 struct mqueue_inode_info *ei;
500
501 ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL);
502 if (!ei)
503 return NULL;
504 return &ei->vfs_inode;
505}
506
507static void mqueue_free_inode(struct inode *inode)
508{
509 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
510}
511
512static void mqueue_evict_inode(struct inode *inode)
513{
514 struct mqueue_inode_info *info;
515 struct ipc_namespace *ipc_ns;
516 struct msg_msg *msg, *nmsg;
517 LIST_HEAD(tmp_msg);
518
519 clear_inode(inode);
520
521 if (S_ISDIR(inode->i_mode))
522 return;
523
524 ipc_ns = get_ns_from_inode(inode);
525 info = MQUEUE_I(inode);
526 spin_lock(&info->lock);
527 while ((msg = msg_get(info)) != NULL)
528 list_add_tail(&msg->m_list, &tmp_msg);
529 kfree(info->node_cache);
530 spin_unlock(&info->lock);
531
532 list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
533 list_del(&msg->m_list);
534 free_msg(msg);
535 }
536
537 if (info->ucounts) {
538 unsigned long mq_bytes, mq_treesize;
539
540 /* Total amount of bytes accounted for the mqueue */
541 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
542 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
543 sizeof(struct posix_msg_tree_node);
544
545 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
546 info->attr.mq_msgsize);
547
548 spin_lock(&mq_lock);
549 dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
550 /*
551 * get_ns_from_inode() ensures that the
552 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
553 * to which we now hold a reference, or it is NULL.
554 * We can't put it here under mq_lock, though.
555 */
556 if (ipc_ns)
557 ipc_ns->mq_queues_count--;
558 spin_unlock(&mq_lock);
559 put_ucounts(info->ucounts);
560 info->ucounts = NULL;
561 }
562 if (ipc_ns)
563 put_ipc_ns(ipc_ns);
564}
565
566static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
567{
568 struct inode *dir = dentry->d_parent->d_inode;
569 struct inode *inode;
570 struct mq_attr *attr = arg;
571 int error;
572 struct ipc_namespace *ipc_ns;
573
574 spin_lock(&mq_lock);
575 ipc_ns = __get_ns_from_inode(dir);
576 if (!ipc_ns) {
577 error = -EACCES;
578 goto out_unlock;
579 }
580
581 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
582 !capable(CAP_SYS_RESOURCE)) {
583 error = -ENOSPC;
584 goto out_unlock;
585 }
586 ipc_ns->mq_queues_count++;
587 spin_unlock(&mq_lock);
588
589 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
590 if (IS_ERR(inode)) {
591 error = PTR_ERR(inode);
592 spin_lock(&mq_lock);
593 ipc_ns->mq_queues_count--;
594 goto out_unlock;
595 }
596
597 put_ipc_ns(ipc_ns);
598 dir->i_size += DIRENT_SIZE;
599 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
600
601 d_instantiate(dentry, inode);
602 dget(dentry);
603 return 0;
604out_unlock:
605 spin_unlock(&mq_lock);
606 if (ipc_ns)
607 put_ipc_ns(ipc_ns);
608 return error;
609}
610
611static int mqueue_create(struct user_namespace *mnt_userns, struct inode *dir,
612 struct dentry *dentry, umode_t mode, bool excl)
613{
614 return mqueue_create_attr(dentry, mode, NULL);
615}
616
617static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
618{
619 struct inode *inode = d_inode(dentry);
620
621 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
622 dir->i_size -= DIRENT_SIZE;
623 drop_nlink(inode);
624 dput(dentry);
625 return 0;
626}
627
628/*
629* This is routine for system read from queue file.
630* To avoid mess with doing here some sort of mq_receive we allow
631* to read only queue size & notification info (the only values
632* that are interesting from user point of view and aren't accessible
633* through std routines)
634*/
635static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
636 size_t count, loff_t *off)
637{
638 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
639 char buffer[FILENT_SIZE];
640 ssize_t ret;
641
642 spin_lock(&info->lock);
643 snprintf(buffer, sizeof(buffer),
644 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
645 info->qsize,
646 info->notify_owner ? info->notify.sigev_notify : 0,
647 (info->notify_owner &&
648 info->notify.sigev_notify == SIGEV_SIGNAL) ?
649 info->notify.sigev_signo : 0,
650 pid_vnr(info->notify_owner));
651 spin_unlock(&info->lock);
652 buffer[sizeof(buffer)-1] = '\0';
653
654 ret = simple_read_from_buffer(u_data, count, off, buffer,
655 strlen(buffer));
656 if (ret <= 0)
657 return ret;
658
659 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
660 return ret;
661}
662
663static int mqueue_flush_file(struct file *filp, fl_owner_t id)
664{
665 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
666
667 spin_lock(&info->lock);
668 if (task_tgid(current) == info->notify_owner)
669 remove_notification(info);
670
671 spin_unlock(&info->lock);
672 return 0;
673}
674
675static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
676{
677 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
678 __poll_t retval = 0;
679
680 poll_wait(filp, &info->wait_q, poll_tab);
681
682 spin_lock(&info->lock);
683 if (info->attr.mq_curmsgs)
684 retval = EPOLLIN | EPOLLRDNORM;
685
686 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
687 retval |= EPOLLOUT | EPOLLWRNORM;
688 spin_unlock(&info->lock);
689
690 return retval;
691}
692
693/* Adds current to info->e_wait_q[sr] before element with smaller prio */
694static void wq_add(struct mqueue_inode_info *info, int sr,
695 struct ext_wait_queue *ewp)
696{
697 struct ext_wait_queue *walk;
698
699 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
700 if (walk->task->prio <= current->prio) {
701 list_add_tail(&ewp->list, &walk->list);
702 return;
703 }
704 }
705 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
706}
707
708/*
709 * Puts current task to sleep. Caller must hold queue lock. After return
710 * lock isn't held.
711 * sr: SEND or RECV
712 */
713static int wq_sleep(struct mqueue_inode_info *info, int sr,
714 ktime_t *timeout, struct ext_wait_queue *ewp)
715 __releases(&info->lock)
716{
717 int retval;
718 signed long time;
719
720 wq_add(info, sr, ewp);
721
722 for (;;) {
723 /* memory barrier not required, we hold info->lock */
724 __set_current_state(TASK_INTERRUPTIBLE);
725
726 spin_unlock(&info->lock);
727 time = schedule_hrtimeout_range_clock(timeout, 0,
728 HRTIMER_MODE_ABS, CLOCK_REALTIME);
729
730 if (READ_ONCE(ewp->state) == STATE_READY) {
731 /* see MQ_BARRIER for purpose/pairing */
732 smp_acquire__after_ctrl_dep();
733 retval = 0;
734 goto out;
735 }
736 spin_lock(&info->lock);
737
738 /* we hold info->lock, so no memory barrier required */
739 if (READ_ONCE(ewp->state) == STATE_READY) {
740 retval = 0;
741 goto out_unlock;
742 }
743 if (signal_pending(current)) {
744 retval = -ERESTARTSYS;
745 break;
746 }
747 if (time == 0) {
748 retval = -ETIMEDOUT;
749 break;
750 }
751 }
752 list_del(&ewp->list);
753out_unlock:
754 spin_unlock(&info->lock);
755out:
756 return retval;
757}
758
759/*
760 * Returns waiting task that should be serviced first or NULL if none exists
761 */
762static struct ext_wait_queue *wq_get_first_waiter(
763 struct mqueue_inode_info *info, int sr)
764{
765 struct list_head *ptr;
766
767 ptr = info->e_wait_q[sr].list.prev;
768 if (ptr == &info->e_wait_q[sr].list)
769 return NULL;
770 return list_entry(ptr, struct ext_wait_queue, list);
771}
772
773
774static inline void set_cookie(struct sk_buff *skb, char code)
775{
776 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
777}
778
779/*
780 * The next function is only to split too long sys_mq_timedsend
781 */
782static void __do_notify(struct mqueue_inode_info *info)
783{
784 /* notification
785 * invoked when there is registered process and there isn't process
786 * waiting synchronously for message AND state of queue changed from
787 * empty to not empty. Here we are sure that no one is waiting
788 * synchronously. */
789 if (info->notify_owner &&
790 info->attr.mq_curmsgs == 1) {
791 switch (info->notify.sigev_notify) {
792 case SIGEV_NONE:
793 break;
794 case SIGEV_SIGNAL: {
795 struct kernel_siginfo sig_i;
796 struct task_struct *task;
797
798 /* do_mq_notify() accepts sigev_signo == 0, why?? */
799 if (!info->notify.sigev_signo)
800 break;
801
802 clear_siginfo(&sig_i);
803 sig_i.si_signo = info->notify.sigev_signo;
804 sig_i.si_errno = 0;
805 sig_i.si_code = SI_MESGQ;
806 sig_i.si_value = info->notify.sigev_value;
807 rcu_read_lock();
808 /* map current pid/uid into info->owner's namespaces */
809 sig_i.si_pid = task_tgid_nr_ns(current,
810 ns_of_pid(info->notify_owner));
811 sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
812 current_uid());
813 /*
814 * We can't use kill_pid_info(), this signal should
815 * bypass check_kill_permission(). It is from kernel
816 * but si_fromuser() can't know this.
817 * We do check the self_exec_id, to avoid sending
818 * signals to programs that don't expect them.
819 */
820 task = pid_task(info->notify_owner, PIDTYPE_TGID);
821 if (task && task->self_exec_id ==
822 info->notify_self_exec_id) {
823 do_send_sig_info(info->notify.sigev_signo,
824 &sig_i, task, PIDTYPE_TGID);
825 }
826 rcu_read_unlock();
827 break;
828 }
829 case SIGEV_THREAD:
830 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
831 netlink_sendskb(info->notify_sock, info->notify_cookie);
832 break;
833 }
834 /* after notification unregisters process */
835 put_pid(info->notify_owner);
836 put_user_ns(info->notify_user_ns);
837 info->notify_owner = NULL;
838 info->notify_user_ns = NULL;
839 }
840 wake_up(&info->wait_q);
841}
842
843static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
844 struct timespec64 *ts)
845{
846 if (get_timespec64(ts, u_abs_timeout))
847 return -EFAULT;
848 if (!timespec64_valid(ts))
849 return -EINVAL;
850 return 0;
851}
852
853static void remove_notification(struct mqueue_inode_info *info)
854{
855 if (info->notify_owner != NULL &&
856 info->notify.sigev_notify == SIGEV_THREAD) {
857 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
858 netlink_sendskb(info->notify_sock, info->notify_cookie);
859 }
860 put_pid(info->notify_owner);
861 put_user_ns(info->notify_user_ns);
862 info->notify_owner = NULL;
863 info->notify_user_ns = NULL;
864}
865
866static int prepare_open(struct dentry *dentry, int oflag, int ro,
867 umode_t mode, struct filename *name,
868 struct mq_attr *attr)
869{
870 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
871 MAY_READ | MAY_WRITE };
872 int acc;
873
874 if (d_really_is_negative(dentry)) {
875 if (!(oflag & O_CREAT))
876 return -ENOENT;
877 if (ro)
878 return ro;
879 audit_inode_parent_hidden(name, dentry->d_parent);
880 return vfs_mkobj(dentry, mode & ~current_umask(),
881 mqueue_create_attr, attr);
882 }
883 /* it already existed */
884 audit_inode(name, dentry, 0);
885 if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
886 return -EEXIST;
887 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
888 return -EINVAL;
889 acc = oflag2acc[oflag & O_ACCMODE];
890 return inode_permission(&init_user_ns, d_inode(dentry), acc);
891}
892
893static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
894 struct mq_attr *attr)
895{
896 struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
897 struct dentry *root = mnt->mnt_root;
898 struct filename *name;
899 struct path path;
900 int fd, error;
901 int ro;
902
903 audit_mq_open(oflag, mode, attr);
904
905 if (IS_ERR(name = getname(u_name)))
906 return PTR_ERR(name);
907
908 fd = get_unused_fd_flags(O_CLOEXEC);
909 if (fd < 0)
910 goto out_putname;
911
912 ro = mnt_want_write(mnt); /* we'll drop it in any case */
913 inode_lock(d_inode(root));
914 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
915 if (IS_ERR(path.dentry)) {
916 error = PTR_ERR(path.dentry);
917 goto out_putfd;
918 }
919 path.mnt = mntget(mnt);
920 error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
921 if (!error) {
922 struct file *file = dentry_open(&path, oflag, current_cred());
923 if (!IS_ERR(file))
924 fd_install(fd, file);
925 else
926 error = PTR_ERR(file);
927 }
928 path_put(&path);
929out_putfd:
930 if (error) {
931 put_unused_fd(fd);
932 fd = error;
933 }
934 inode_unlock(d_inode(root));
935 if (!ro)
936 mnt_drop_write(mnt);
937out_putname:
938 putname(name);
939 return fd;
940}
941
942SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
943 struct mq_attr __user *, u_attr)
944{
945 struct mq_attr attr;
946 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
947 return -EFAULT;
948
949 return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
950}
951
952SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
953{
954 int err;
955 struct filename *name;
956 struct dentry *dentry;
957 struct inode *inode = NULL;
958 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
959 struct vfsmount *mnt = ipc_ns->mq_mnt;
960
961 name = getname(u_name);
962 if (IS_ERR(name))
963 return PTR_ERR(name);
964
965 audit_inode_parent_hidden(name, mnt->mnt_root);
966 err = mnt_want_write(mnt);
967 if (err)
968 goto out_name;
969 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
970 dentry = lookup_one_len(name->name, mnt->mnt_root,
971 strlen(name->name));
972 if (IS_ERR(dentry)) {
973 err = PTR_ERR(dentry);
974 goto out_unlock;
975 }
976
977 inode = d_inode(dentry);
978 if (!inode) {
979 err = -ENOENT;
980 } else {
981 ihold(inode);
982 err = vfs_unlink(&init_user_ns, d_inode(dentry->d_parent),
983 dentry, NULL);
984 }
985 dput(dentry);
986
987out_unlock:
988 inode_unlock(d_inode(mnt->mnt_root));
989 iput(inode);
990 mnt_drop_write(mnt);
991out_name:
992 putname(name);
993
994 return err;
995}
996
997/* Pipelined send and receive functions.
998 *
999 * If a receiver finds no waiting message, then it registers itself in the
1000 * list of waiting receivers. A sender checks that list before adding the new
1001 * message into the message array. If there is a waiting receiver, then it
1002 * bypasses the message array and directly hands the message over to the
1003 * receiver. The receiver accepts the message and returns without grabbing the
1004 * queue spinlock:
1005 *
1006 * - Set pointer to message.
1007 * - Queue the receiver task for later wakeup (without the info->lock).
1008 * - Update its state to STATE_READY. Now the receiver can continue.
1009 * - Wake up the process after the lock is dropped. Should the process wake up
1010 * before this wakeup (due to a timeout or a signal) it will either see
1011 * STATE_READY and continue or acquire the lock to check the state again.
1012 *
1013 * The same algorithm is used for senders.
1014 */
1015
1016static inline void __pipelined_op(struct wake_q_head *wake_q,
1017 struct mqueue_inode_info *info,
1018 struct ext_wait_queue *this)
1019{
1020 struct task_struct *task;
1021
1022 list_del(&this->list);
1023 task = get_task_struct(this->task);
1024
1025 /* see MQ_BARRIER for purpose/pairing */
1026 smp_store_release(&this->state, STATE_READY);
1027 wake_q_add_safe(wake_q, task);
1028}
1029
1030/* pipelined_send() - send a message directly to the task waiting in
1031 * sys_mq_timedreceive() (without inserting message into a queue).
1032 */
1033static inline void pipelined_send(struct wake_q_head *wake_q,
1034 struct mqueue_inode_info *info,
1035 struct msg_msg *message,
1036 struct ext_wait_queue *receiver)
1037{
1038 receiver->msg = message;
1039 __pipelined_op(wake_q, info, receiver);
1040}
1041
1042/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1043 * gets its message and put to the queue (we have one free place for sure). */
1044static inline void pipelined_receive(struct wake_q_head *wake_q,
1045 struct mqueue_inode_info *info)
1046{
1047 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1048
1049 if (!sender) {
1050 /* for poll */
1051 wake_up_interruptible(&info->wait_q);
1052 return;
1053 }
1054 if (msg_insert(sender->msg, info))
1055 return;
1056
1057 __pipelined_op(wake_q, info, sender);
1058}
1059
1060static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1061 size_t msg_len, unsigned int msg_prio,
1062 struct timespec64 *ts)
1063{
1064 struct fd f;
1065 struct inode *inode;
1066 struct ext_wait_queue wait;
1067 struct ext_wait_queue *receiver;
1068 struct msg_msg *msg_ptr;
1069 struct mqueue_inode_info *info;
1070 ktime_t expires, *timeout = NULL;
1071 struct posix_msg_tree_node *new_leaf = NULL;
1072 int ret = 0;
1073 DEFINE_WAKE_Q(wake_q);
1074
1075 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1076 return -EINVAL;
1077
1078 if (ts) {
1079 expires = timespec64_to_ktime(*ts);
1080 timeout = &expires;
1081 }
1082
1083 audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1084
1085 f = fdget(mqdes);
1086 if (unlikely(!f.file)) {
1087 ret = -EBADF;
1088 goto out;
1089 }
1090
1091 inode = file_inode(f.file);
1092 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1093 ret = -EBADF;
1094 goto out_fput;
1095 }
1096 info = MQUEUE_I(inode);
1097 audit_file(f.file);
1098
1099 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1100 ret = -EBADF;
1101 goto out_fput;
1102 }
1103
1104 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1105 ret = -EMSGSIZE;
1106 goto out_fput;
1107 }
1108
1109 /* First try to allocate memory, before doing anything with
1110 * existing queues. */
1111 msg_ptr = load_msg(u_msg_ptr, msg_len);
1112 if (IS_ERR(msg_ptr)) {
1113 ret = PTR_ERR(msg_ptr);
1114 goto out_fput;
1115 }
1116 msg_ptr->m_ts = msg_len;
1117 msg_ptr->m_type = msg_prio;
1118
1119 /*
1120 * msg_insert really wants us to have a valid, spare node struct so
1121 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1122 * fall back to that if necessary.
1123 */
1124 if (!info->node_cache)
1125 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1126
1127 spin_lock(&info->lock);
1128
1129 if (!info->node_cache && new_leaf) {
1130 /* Save our speculative allocation into the cache */
1131 INIT_LIST_HEAD(&new_leaf->msg_list);
1132 info->node_cache = new_leaf;
1133 new_leaf = NULL;
1134 } else {
1135 kfree(new_leaf);
1136 }
1137
1138 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1139 if (f.file->f_flags & O_NONBLOCK) {
1140 ret = -EAGAIN;
1141 } else {
1142 wait.task = current;
1143 wait.msg = (void *) msg_ptr;
1144
1145 /* memory barrier not required, we hold info->lock */
1146 WRITE_ONCE(wait.state, STATE_NONE);
1147 ret = wq_sleep(info, SEND, timeout, &wait);
1148 /*
1149 * wq_sleep must be called with info->lock held, and
1150 * returns with the lock released
1151 */
1152 goto out_free;
1153 }
1154 } else {
1155 receiver = wq_get_first_waiter(info, RECV);
1156 if (receiver) {
1157 pipelined_send(&wake_q, info, msg_ptr, receiver);
1158 } else {
1159 /* adds message to the queue */
1160 ret = msg_insert(msg_ptr, info);
1161 if (ret)
1162 goto out_unlock;
1163 __do_notify(info);
1164 }
1165 inode->i_atime = inode->i_mtime = inode->i_ctime =
1166 current_time(inode);
1167 }
1168out_unlock:
1169 spin_unlock(&info->lock);
1170 wake_up_q(&wake_q);
1171out_free:
1172 if (ret)
1173 free_msg(msg_ptr);
1174out_fput:
1175 fdput(f);
1176out:
1177 return ret;
1178}
1179
1180static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1181 size_t msg_len, unsigned int __user *u_msg_prio,
1182 struct timespec64 *ts)
1183{
1184 ssize_t ret;
1185 struct msg_msg *msg_ptr;
1186 struct fd f;
1187 struct inode *inode;
1188 struct mqueue_inode_info *info;
1189 struct ext_wait_queue wait;
1190 ktime_t expires, *timeout = NULL;
1191 struct posix_msg_tree_node *new_leaf = NULL;
1192
1193 if (ts) {
1194 expires = timespec64_to_ktime(*ts);
1195 timeout = &expires;
1196 }
1197
1198 audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1199
1200 f = fdget(mqdes);
1201 if (unlikely(!f.file)) {
1202 ret = -EBADF;
1203 goto out;
1204 }
1205
1206 inode = file_inode(f.file);
1207 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1208 ret = -EBADF;
1209 goto out_fput;
1210 }
1211 info = MQUEUE_I(inode);
1212 audit_file(f.file);
1213
1214 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1215 ret = -EBADF;
1216 goto out_fput;
1217 }
1218
1219 /* checks if buffer is big enough */
1220 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1221 ret = -EMSGSIZE;
1222 goto out_fput;
1223 }
1224
1225 /*
1226 * msg_insert really wants us to have a valid, spare node struct so
1227 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1228 * fall back to that if necessary.
1229 */
1230 if (!info->node_cache)
1231 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1232
1233 spin_lock(&info->lock);
1234
1235 if (!info->node_cache && new_leaf) {
1236 /* Save our speculative allocation into the cache */
1237 INIT_LIST_HEAD(&new_leaf->msg_list);
1238 info->node_cache = new_leaf;
1239 } else {
1240 kfree(new_leaf);
1241 }
1242
1243 if (info->attr.mq_curmsgs == 0) {
1244 if (f.file->f_flags & O_NONBLOCK) {
1245 spin_unlock(&info->lock);
1246 ret = -EAGAIN;
1247 } else {
1248 wait.task = current;
1249
1250 /* memory barrier not required, we hold info->lock */
1251 WRITE_ONCE(wait.state, STATE_NONE);
1252 ret = wq_sleep(info, RECV, timeout, &wait);
1253 msg_ptr = wait.msg;
1254 }
1255 } else {
1256 DEFINE_WAKE_Q(wake_q);
1257
1258 msg_ptr = msg_get(info);
1259
1260 inode->i_atime = inode->i_mtime = inode->i_ctime =
1261 current_time(inode);
1262
1263 /* There is now free space in queue. */
1264 pipelined_receive(&wake_q, info);
1265 spin_unlock(&info->lock);
1266 wake_up_q(&wake_q);
1267 ret = 0;
1268 }
1269 if (ret == 0) {
1270 ret = msg_ptr->m_ts;
1271
1272 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1273 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1274 ret = -EFAULT;
1275 }
1276 free_msg(msg_ptr);
1277 }
1278out_fput:
1279 fdput(f);
1280out:
1281 return ret;
1282}
1283
1284SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1285 size_t, msg_len, unsigned int, msg_prio,
1286 const struct __kernel_timespec __user *, u_abs_timeout)
1287{
1288 struct timespec64 ts, *p = NULL;
1289 if (u_abs_timeout) {
1290 int res = prepare_timeout(u_abs_timeout, &ts);
1291 if (res)
1292 return res;
1293 p = &ts;
1294 }
1295 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1296}
1297
1298SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1299 size_t, msg_len, unsigned int __user *, u_msg_prio,
1300 const struct __kernel_timespec __user *, u_abs_timeout)
1301{
1302 struct timespec64 ts, *p = NULL;
1303 if (u_abs_timeout) {
1304 int res = prepare_timeout(u_abs_timeout, &ts);
1305 if (res)
1306 return res;
1307 p = &ts;
1308 }
1309 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1310}
1311
1312/*
1313 * Notes: the case when user wants us to deregister (with NULL as pointer)
1314 * and he isn't currently owner of notification, will be silently discarded.
1315 * It isn't explicitly defined in the POSIX.
1316 */
1317static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1318{
1319 int ret;
1320 struct fd f;
1321 struct sock *sock;
1322 struct inode *inode;
1323 struct mqueue_inode_info *info;
1324 struct sk_buff *nc;
1325
1326 audit_mq_notify(mqdes, notification);
1327
1328 nc = NULL;
1329 sock = NULL;
1330 if (notification != NULL) {
1331 if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1332 notification->sigev_notify != SIGEV_SIGNAL &&
1333 notification->sigev_notify != SIGEV_THREAD))
1334 return -EINVAL;
1335 if (notification->sigev_notify == SIGEV_SIGNAL &&
1336 !valid_signal(notification->sigev_signo)) {
1337 return -EINVAL;
1338 }
1339 if (notification->sigev_notify == SIGEV_THREAD) {
1340 long timeo;
1341
1342 /* create the notify skb */
1343 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1344 if (!nc)
1345 return -ENOMEM;
1346
1347 if (copy_from_user(nc->data,
1348 notification->sigev_value.sival_ptr,
1349 NOTIFY_COOKIE_LEN)) {
1350 ret = -EFAULT;
1351 goto free_skb;
1352 }
1353
1354 /* TODO: add a header? */
1355 skb_put(nc, NOTIFY_COOKIE_LEN);
1356 /* and attach it to the socket */
1357retry:
1358 f = fdget(notification->sigev_signo);
1359 if (!f.file) {
1360 ret = -EBADF;
1361 goto out;
1362 }
1363 sock = netlink_getsockbyfilp(f.file);
1364 fdput(f);
1365 if (IS_ERR(sock)) {
1366 ret = PTR_ERR(sock);
1367 goto free_skb;
1368 }
1369
1370 timeo = MAX_SCHEDULE_TIMEOUT;
1371 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1372 if (ret == 1) {
1373 sock = NULL;
1374 goto retry;
1375 }
1376 if (ret)
1377 return ret;
1378 }
1379 }
1380
1381 f = fdget(mqdes);
1382 if (!f.file) {
1383 ret = -EBADF;
1384 goto out;
1385 }
1386
1387 inode = file_inode(f.file);
1388 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1389 ret = -EBADF;
1390 goto out_fput;
1391 }
1392 info = MQUEUE_I(inode);
1393
1394 ret = 0;
1395 spin_lock(&info->lock);
1396 if (notification == NULL) {
1397 if (info->notify_owner == task_tgid(current)) {
1398 remove_notification(info);
1399 inode->i_atime = inode->i_ctime = current_time(inode);
1400 }
1401 } else if (info->notify_owner != NULL) {
1402 ret = -EBUSY;
1403 } else {
1404 switch (notification->sigev_notify) {
1405 case SIGEV_NONE:
1406 info->notify.sigev_notify = SIGEV_NONE;
1407 break;
1408 case SIGEV_THREAD:
1409 info->notify_sock = sock;
1410 info->notify_cookie = nc;
1411 sock = NULL;
1412 nc = NULL;
1413 info->notify.sigev_notify = SIGEV_THREAD;
1414 break;
1415 case SIGEV_SIGNAL:
1416 info->notify.sigev_signo = notification->sigev_signo;
1417 info->notify.sigev_value = notification->sigev_value;
1418 info->notify.sigev_notify = SIGEV_SIGNAL;
1419 info->notify_self_exec_id = current->self_exec_id;
1420 break;
1421 }
1422
1423 info->notify_owner = get_pid(task_tgid(current));
1424 info->notify_user_ns = get_user_ns(current_user_ns());
1425 inode->i_atime = inode->i_ctime = current_time(inode);
1426 }
1427 spin_unlock(&info->lock);
1428out_fput:
1429 fdput(f);
1430out:
1431 if (sock)
1432 netlink_detachskb(sock, nc);
1433 else
1434free_skb:
1435 dev_kfree_skb(nc);
1436
1437 return ret;
1438}
1439
1440SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1441 const struct sigevent __user *, u_notification)
1442{
1443 struct sigevent n, *p = NULL;
1444 if (u_notification) {
1445 if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1446 return -EFAULT;
1447 p = &n;
1448 }
1449 return do_mq_notify(mqdes, p);
1450}
1451
1452static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1453{
1454 struct fd f;
1455 struct inode *inode;
1456 struct mqueue_inode_info *info;
1457
1458 if (new && (new->mq_flags & (~O_NONBLOCK)))
1459 return -EINVAL;
1460
1461 f = fdget(mqdes);
1462 if (!f.file)
1463 return -EBADF;
1464
1465 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1466 fdput(f);
1467 return -EBADF;
1468 }
1469
1470 inode = file_inode(f.file);
1471 info = MQUEUE_I(inode);
1472
1473 spin_lock(&info->lock);
1474
1475 if (old) {
1476 *old = info->attr;
1477 old->mq_flags = f.file->f_flags & O_NONBLOCK;
1478 }
1479 if (new) {
1480 audit_mq_getsetattr(mqdes, new);
1481 spin_lock(&f.file->f_lock);
1482 if (new->mq_flags & O_NONBLOCK)
1483 f.file->f_flags |= O_NONBLOCK;
1484 else
1485 f.file->f_flags &= ~O_NONBLOCK;
1486 spin_unlock(&f.file->f_lock);
1487
1488 inode->i_atime = inode->i_ctime = current_time(inode);
1489 }
1490
1491 spin_unlock(&info->lock);
1492 fdput(f);
1493 return 0;
1494}
1495
1496SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1497 const struct mq_attr __user *, u_mqstat,
1498 struct mq_attr __user *, u_omqstat)
1499{
1500 int ret;
1501 struct mq_attr mqstat, omqstat;
1502 struct mq_attr *new = NULL, *old = NULL;
1503
1504 if (u_mqstat) {
1505 new = &mqstat;
1506 if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1507 return -EFAULT;
1508 }
1509 if (u_omqstat)
1510 old = &omqstat;
1511
1512 ret = do_mq_getsetattr(mqdes, new, old);
1513 if (ret || !old)
1514 return ret;
1515
1516 if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1517 return -EFAULT;
1518 return 0;
1519}
1520
1521#ifdef CONFIG_COMPAT
1522
1523struct compat_mq_attr {
1524 compat_long_t mq_flags; /* message queue flags */
1525 compat_long_t mq_maxmsg; /* maximum number of messages */
1526 compat_long_t mq_msgsize; /* maximum message size */
1527 compat_long_t mq_curmsgs; /* number of messages currently queued */
1528 compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1529};
1530
1531static inline int get_compat_mq_attr(struct mq_attr *attr,
1532 const struct compat_mq_attr __user *uattr)
1533{
1534 struct compat_mq_attr v;
1535
1536 if (copy_from_user(&v, uattr, sizeof(*uattr)))
1537 return -EFAULT;
1538
1539 memset(attr, 0, sizeof(*attr));
1540 attr->mq_flags = v.mq_flags;
1541 attr->mq_maxmsg = v.mq_maxmsg;
1542 attr->mq_msgsize = v.mq_msgsize;
1543 attr->mq_curmsgs = v.mq_curmsgs;
1544 return 0;
1545}
1546
1547static inline int put_compat_mq_attr(const struct mq_attr *attr,
1548 struct compat_mq_attr __user *uattr)
1549{
1550 struct compat_mq_attr v;
1551
1552 memset(&v, 0, sizeof(v));
1553 v.mq_flags = attr->mq_flags;
1554 v.mq_maxmsg = attr->mq_maxmsg;
1555 v.mq_msgsize = attr->mq_msgsize;
1556 v.mq_curmsgs = attr->mq_curmsgs;
1557 if (copy_to_user(uattr, &v, sizeof(*uattr)))
1558 return -EFAULT;
1559 return 0;
1560}
1561
1562COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1563 int, oflag, compat_mode_t, mode,
1564 struct compat_mq_attr __user *, u_attr)
1565{
1566 struct mq_attr attr, *p = NULL;
1567 if (u_attr && oflag & O_CREAT) {
1568 p = &attr;
1569 if (get_compat_mq_attr(&attr, u_attr))
1570 return -EFAULT;
1571 }
1572 return do_mq_open(u_name, oflag, mode, p);
1573}
1574
1575COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1576 const struct compat_sigevent __user *, u_notification)
1577{
1578 struct sigevent n, *p = NULL;
1579 if (u_notification) {
1580 if (get_compat_sigevent(&n, u_notification))
1581 return -EFAULT;
1582 if (n.sigev_notify == SIGEV_THREAD)
1583 n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1584 p = &n;
1585 }
1586 return do_mq_notify(mqdes, p);
1587}
1588
1589COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1590 const struct compat_mq_attr __user *, u_mqstat,
1591 struct compat_mq_attr __user *, u_omqstat)
1592{
1593 int ret;
1594 struct mq_attr mqstat, omqstat;
1595 struct mq_attr *new = NULL, *old = NULL;
1596
1597 if (u_mqstat) {
1598 new = &mqstat;
1599 if (get_compat_mq_attr(new, u_mqstat))
1600 return -EFAULT;
1601 }
1602 if (u_omqstat)
1603 old = &omqstat;
1604
1605 ret = do_mq_getsetattr(mqdes, new, old);
1606 if (ret || !old)
1607 return ret;
1608
1609 if (put_compat_mq_attr(old, u_omqstat))
1610 return -EFAULT;
1611 return 0;
1612}
1613#endif
1614
1615#ifdef CONFIG_COMPAT_32BIT_TIME
1616static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1617 struct timespec64 *ts)
1618{
1619 if (get_old_timespec32(ts, p))
1620 return -EFAULT;
1621 if (!timespec64_valid(ts))
1622 return -EINVAL;
1623 return 0;
1624}
1625
1626SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1627 const char __user *, u_msg_ptr,
1628 unsigned int, msg_len, unsigned int, msg_prio,
1629 const struct old_timespec32 __user *, u_abs_timeout)
1630{
1631 struct timespec64 ts, *p = NULL;
1632 if (u_abs_timeout) {
1633 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1634 if (res)
1635 return res;
1636 p = &ts;
1637 }
1638 return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1639}
1640
1641SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1642 char __user *, u_msg_ptr,
1643 unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1644 const struct old_timespec32 __user *, u_abs_timeout)
1645{
1646 struct timespec64 ts, *p = NULL;
1647 if (u_abs_timeout) {
1648 int res = compat_prepare_timeout(u_abs_timeout, &ts);
1649 if (res)
1650 return res;
1651 p = &ts;
1652 }
1653 return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1654}
1655#endif
1656
1657static const struct inode_operations mqueue_dir_inode_operations = {
1658 .lookup = simple_lookup,
1659 .create = mqueue_create,
1660 .unlink = mqueue_unlink,
1661};
1662
1663static const struct file_operations mqueue_file_operations = {
1664 .flush = mqueue_flush_file,
1665 .poll = mqueue_poll_file,
1666 .read = mqueue_read_file,
1667 .llseek = default_llseek,
1668};
1669
1670static const struct super_operations mqueue_super_ops = {
1671 .alloc_inode = mqueue_alloc_inode,
1672 .free_inode = mqueue_free_inode,
1673 .evict_inode = mqueue_evict_inode,
1674 .statfs = simple_statfs,
1675};
1676
1677static const struct fs_context_operations mqueue_fs_context_ops = {
1678 .free = mqueue_fs_context_free,
1679 .get_tree = mqueue_get_tree,
1680};
1681
1682static struct file_system_type mqueue_fs_type = {
1683 .name = "mqueue",
1684 .init_fs_context = mqueue_init_fs_context,
1685 .kill_sb = kill_litter_super,
1686 .fs_flags = FS_USERNS_MOUNT,
1687};
1688
1689int mq_init_ns(struct ipc_namespace *ns)
1690{
1691 struct vfsmount *m;
1692
1693 ns->mq_queues_count = 0;
1694 ns->mq_queues_max = DFLT_QUEUESMAX;
1695 ns->mq_msg_max = DFLT_MSGMAX;
1696 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1697 ns->mq_msg_default = DFLT_MSG;
1698 ns->mq_msgsize_default = DFLT_MSGSIZE;
1699
1700 m = mq_create_mount(ns);
1701 if (IS_ERR(m))
1702 return PTR_ERR(m);
1703 ns->mq_mnt = m;
1704 return 0;
1705}
1706
1707void mq_clear_sbinfo(struct ipc_namespace *ns)
1708{
1709 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1710}
1711
1712void mq_put_mnt(struct ipc_namespace *ns)
1713{
1714 kern_unmount(ns->mq_mnt);
1715}
1716
1717static int __init init_mqueue_fs(void)
1718{
1719 int error;
1720
1721 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1722 sizeof(struct mqueue_inode_info), 0,
1723 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1724 if (mqueue_inode_cachep == NULL)
1725 return -ENOMEM;
1726
1727 if (!setup_mq_sysctls(&init_ipc_ns)) {
1728 pr_warn("sysctl registration failed\n");
1729 error = -ENOMEM;
1730 goto out_kmem;
1731 }
1732
1733 error = register_filesystem(&mqueue_fs_type);
1734 if (error)
1735 goto out_sysctl;
1736
1737 spin_lock_init(&mq_lock);
1738
1739 error = mq_init_ns(&init_ipc_ns);
1740 if (error)
1741 goto out_filesystem;
1742
1743 return 0;
1744
1745out_filesystem:
1746 unregister_filesystem(&mqueue_fs_type);
1747out_sysctl:
1748 retire_mq_sysctls(&init_ipc_ns);
1749out_kmem:
1750 kmem_cache_destroy(mqueue_inode_cachep);
1751 return error;
1752}
1753
1754device_initcall(init_mqueue_fs);
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
5 * Michal Wronski (michal.wronski@gmail.com)
6 *
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
9 * Manfred Spraul (manfred@colorfullife.com)
10 *
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
12 *
13 * This file is released under the GPL.
14 */
15
16#include <linux/capability.h>
17#include <linux/init.h>
18#include <linux/pagemap.h>
19#include <linux/file.h>
20#include <linux/mount.h>
21#include <linux/namei.h>
22#include <linux/sysctl.h>
23#include <linux/poll.h>
24#include <linux/mqueue.h>
25#include <linux/msg.h>
26#include <linux/skbuff.h>
27#include <linux/vmalloc.h>
28#include <linux/netlink.h>
29#include <linux/syscalls.h>
30#include <linux/audit.h>
31#include <linux/signal.h>
32#include <linux/mutex.h>
33#include <linux/nsproxy.h>
34#include <linux/pid.h>
35#include <linux/ipc_namespace.h>
36#include <linux/user_namespace.h>
37#include <linux/slab.h>
38
39#include <net/sock.h>
40#include "util.h"
41
42#define MQUEUE_MAGIC 0x19800202
43#define DIRENT_SIZE 20
44#define FILENT_SIZE 80
45
46#define SEND 0
47#define RECV 1
48
49#define STATE_NONE 0
50#define STATE_PENDING 1
51#define STATE_READY 2
52
53struct posix_msg_tree_node {
54 struct rb_node rb_node;
55 struct list_head msg_list;
56 int priority;
57};
58
59struct ext_wait_queue { /* queue of sleeping tasks */
60 struct task_struct *task;
61 struct list_head list;
62 struct msg_msg *msg; /* ptr of loaded message */
63 int state; /* one of STATE_* values */
64};
65
66struct mqueue_inode_info {
67 spinlock_t lock;
68 struct inode vfs_inode;
69 wait_queue_head_t wait_q;
70
71 struct rb_root msg_tree;
72 struct posix_msg_tree_node *node_cache;
73 struct mq_attr attr;
74
75 struct sigevent notify;
76 struct pid *notify_owner;
77 struct user_namespace *notify_user_ns;
78 struct user_struct *user; /* user who created, for accounting */
79 struct sock *notify_sock;
80 struct sk_buff *notify_cookie;
81
82 /* for tasks waiting for free space and messages, respectively */
83 struct ext_wait_queue e_wait_q[2];
84
85 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
86};
87
88static const struct inode_operations mqueue_dir_inode_operations;
89static const struct file_operations mqueue_file_operations;
90static const struct super_operations mqueue_super_ops;
91static void remove_notification(struct mqueue_inode_info *info);
92
93static struct kmem_cache *mqueue_inode_cachep;
94
95static struct ctl_table_header *mq_sysctl_table;
96
97static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98{
99 return container_of(inode, struct mqueue_inode_info, vfs_inode);
100}
101
102/*
103 * This routine should be called with the mq_lock held.
104 */
105static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
106{
107 return get_ipc_ns(inode->i_sb->s_fs_info);
108}
109
110static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
111{
112 struct ipc_namespace *ns;
113
114 spin_lock(&mq_lock);
115 ns = __get_ns_from_inode(inode);
116 spin_unlock(&mq_lock);
117 return ns;
118}
119
120/* Auxiliary functions to manipulate messages' list */
121static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122{
123 struct rb_node **p, *parent = NULL;
124 struct posix_msg_tree_node *leaf;
125
126 p = &info->msg_tree.rb_node;
127 while (*p) {
128 parent = *p;
129 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130
131 if (likely(leaf->priority == msg->m_type))
132 goto insert_msg;
133 else if (msg->m_type < leaf->priority)
134 p = &(*p)->rb_left;
135 else
136 p = &(*p)->rb_right;
137 }
138 if (info->node_cache) {
139 leaf = info->node_cache;
140 info->node_cache = NULL;
141 } else {
142 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
143 if (!leaf)
144 return -ENOMEM;
145 INIT_LIST_HEAD(&leaf->msg_list);
146 info->qsize += sizeof(*leaf);
147 }
148 leaf->priority = msg->m_type;
149 rb_link_node(&leaf->rb_node, parent, p);
150 rb_insert_color(&leaf->rb_node, &info->msg_tree);
151insert_msg:
152 info->attr.mq_curmsgs++;
153 info->qsize += msg->m_ts;
154 list_add_tail(&msg->m_list, &leaf->msg_list);
155 return 0;
156}
157
158static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
159{
160 struct rb_node **p, *parent = NULL;
161 struct posix_msg_tree_node *leaf;
162 struct msg_msg *msg;
163
164try_again:
165 p = &info->msg_tree.rb_node;
166 while (*p) {
167 parent = *p;
168 /*
169 * During insert, low priorities go to the left and high to the
170 * right. On receive, we want the highest priorities first, so
171 * walk all the way to the right.
172 */
173 p = &(*p)->rb_right;
174 }
175 if (!parent) {
176 if (info->attr.mq_curmsgs) {
177 pr_warn_once("Inconsistency in POSIX message queue, "
178 "no tree element, but supposedly messages "
179 "should exist!\n");
180 info->attr.mq_curmsgs = 0;
181 }
182 return NULL;
183 }
184 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
185 if (unlikely(list_empty(&leaf->msg_list))) {
186 pr_warn_once("Inconsistency in POSIX message queue, "
187 "empty leaf node but we haven't implemented "
188 "lazy leaf delete!\n");
189 rb_erase(&leaf->rb_node, &info->msg_tree);
190 if (info->node_cache) {
191 info->qsize -= sizeof(*leaf);
192 kfree(leaf);
193 } else {
194 info->node_cache = leaf;
195 }
196 goto try_again;
197 } else {
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
203 if (info->node_cache) {
204 info->qsize -= sizeof(*leaf);
205 kfree(leaf);
206 } else {
207 info->node_cache = leaf;
208 }
209 }
210 }
211 info->attr.mq_curmsgs--;
212 info->qsize -= msg->m_ts;
213 return msg;
214}
215
216static struct inode *mqueue_get_inode(struct super_block *sb,
217 struct ipc_namespace *ipc_ns, umode_t mode,
218 struct mq_attr *attr)
219{
220 struct user_struct *u = current_user();
221 struct inode *inode;
222 int ret = -ENOMEM;
223
224 inode = new_inode(sb);
225 if (!inode)
226 goto err;
227
228 inode->i_ino = get_next_ino();
229 inode->i_mode = mode;
230 inode->i_uid = current_fsuid();
231 inode->i_gid = current_fsgid();
232 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
233
234 if (S_ISREG(mode)) {
235 struct mqueue_inode_info *info;
236 unsigned long mq_bytes, mq_treesize;
237
238 inode->i_fop = &mqueue_file_operations;
239 inode->i_size = FILENT_SIZE;
240 /* mqueue specific info */
241 info = MQUEUE_I(inode);
242 spin_lock_init(&info->lock);
243 init_waitqueue_head(&info->wait_q);
244 INIT_LIST_HEAD(&info->e_wait_q[0].list);
245 INIT_LIST_HEAD(&info->e_wait_q[1].list);
246 info->notify_owner = NULL;
247 info->notify_user_ns = NULL;
248 info->qsize = 0;
249 info->user = NULL; /* set when all is ok */
250 info->msg_tree = RB_ROOT;
251 info->node_cache = NULL;
252 memset(&info->attr, 0, sizeof(info->attr));
253 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
254 ipc_ns->mq_msg_default);
255 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
256 ipc_ns->mq_msgsize_default);
257 if (attr) {
258 info->attr.mq_maxmsg = attr->mq_maxmsg;
259 info->attr.mq_msgsize = attr->mq_msgsize;
260 }
261 /*
262 * We used to allocate a static array of pointers and account
263 * the size of that array as well as one msg_msg struct per
264 * possible message into the queue size. That's no longer
265 * accurate as the queue is now an rbtree and will grow and
266 * shrink depending on usage patterns. We can, however, still
267 * account one msg_msg struct per message, but the nodes are
268 * allocated depending on priority usage, and most programs
269 * only use one, or a handful, of priorities. However, since
270 * this is pinned memory, we need to assume worst case, so
271 * that means the min(mq_maxmsg, max_priorities) * struct
272 * posix_msg_tree_node.
273 */
274 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
275 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
276 sizeof(struct posix_msg_tree_node);
277
278 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
279 info->attr.mq_msgsize);
280
281 spin_lock(&mq_lock);
282 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
283 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
284 spin_unlock(&mq_lock);
285 /* mqueue_evict_inode() releases info->messages */
286 ret = -EMFILE;
287 goto out_inode;
288 }
289 u->mq_bytes += mq_bytes;
290 spin_unlock(&mq_lock);
291
292 /* all is ok */
293 info->user = get_uid(u);
294 } else if (S_ISDIR(mode)) {
295 inc_nlink(inode);
296 /* Some things misbehave if size == 0 on a directory */
297 inode->i_size = 2 * DIRENT_SIZE;
298 inode->i_op = &mqueue_dir_inode_operations;
299 inode->i_fop = &simple_dir_operations;
300 }
301
302 return inode;
303out_inode:
304 iput(inode);
305err:
306 return ERR_PTR(ret);
307}
308
309static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
310{
311 struct inode *inode;
312 struct ipc_namespace *ns = data;
313
314 sb->s_blocksize = PAGE_CACHE_SIZE;
315 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
316 sb->s_magic = MQUEUE_MAGIC;
317 sb->s_op = &mqueue_super_ops;
318
319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
320 if (IS_ERR(inode))
321 return PTR_ERR(inode);
322
323 sb->s_root = d_make_root(inode);
324 if (!sb->s_root)
325 return -ENOMEM;
326 return 0;
327}
328
329static struct dentry *mqueue_mount(struct file_system_type *fs_type,
330 int flags, const char *dev_name,
331 void *data)
332{
333 if (!(flags & MS_KERNMOUNT)) {
334 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
335 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
336 * over the ipc namespace.
337 */
338 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
339 return ERR_PTR(-EPERM);
340
341 data = ns;
342 }
343 return mount_ns(fs_type, flags, data, mqueue_fill_super);
344}
345
346static void init_once(void *foo)
347{
348 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
349
350 inode_init_once(&p->vfs_inode);
351}
352
353static struct inode *mqueue_alloc_inode(struct super_block *sb)
354{
355 struct mqueue_inode_info *ei;
356
357 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
358 if (!ei)
359 return NULL;
360 return &ei->vfs_inode;
361}
362
363static void mqueue_i_callback(struct rcu_head *head)
364{
365 struct inode *inode = container_of(head, struct inode, i_rcu);
366 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
367}
368
369static void mqueue_destroy_inode(struct inode *inode)
370{
371 call_rcu(&inode->i_rcu, mqueue_i_callback);
372}
373
374static void mqueue_evict_inode(struct inode *inode)
375{
376 struct mqueue_inode_info *info;
377 struct user_struct *user;
378 unsigned long mq_bytes, mq_treesize;
379 struct ipc_namespace *ipc_ns;
380 struct msg_msg *msg;
381
382 clear_inode(inode);
383
384 if (S_ISDIR(inode->i_mode))
385 return;
386
387 ipc_ns = get_ns_from_inode(inode);
388 info = MQUEUE_I(inode);
389 spin_lock(&info->lock);
390 while ((msg = msg_get(info)) != NULL)
391 free_msg(msg);
392 kfree(info->node_cache);
393 spin_unlock(&info->lock);
394
395 /* Total amount of bytes accounted for the mqueue */
396 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
397 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
398 sizeof(struct posix_msg_tree_node);
399
400 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
401 info->attr.mq_msgsize);
402
403 user = info->user;
404 if (user) {
405 spin_lock(&mq_lock);
406 user->mq_bytes -= mq_bytes;
407 /*
408 * get_ns_from_inode() ensures that the
409 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
410 * to which we now hold a reference, or it is NULL.
411 * We can't put it here under mq_lock, though.
412 */
413 if (ipc_ns)
414 ipc_ns->mq_queues_count--;
415 spin_unlock(&mq_lock);
416 free_uid(user);
417 }
418 if (ipc_ns)
419 put_ipc_ns(ipc_ns);
420}
421
422static int mqueue_create(struct inode *dir, struct dentry *dentry,
423 umode_t mode, bool excl)
424{
425 struct inode *inode;
426 struct mq_attr *attr = dentry->d_fsdata;
427 int error;
428 struct ipc_namespace *ipc_ns;
429
430 spin_lock(&mq_lock);
431 ipc_ns = __get_ns_from_inode(dir);
432 if (!ipc_ns) {
433 error = -EACCES;
434 goto out_unlock;
435 }
436
437 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
438 !capable(CAP_SYS_RESOURCE)) {
439 error = -ENOSPC;
440 goto out_unlock;
441 }
442 ipc_ns->mq_queues_count++;
443 spin_unlock(&mq_lock);
444
445 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
446 if (IS_ERR(inode)) {
447 error = PTR_ERR(inode);
448 spin_lock(&mq_lock);
449 ipc_ns->mq_queues_count--;
450 goto out_unlock;
451 }
452
453 put_ipc_ns(ipc_ns);
454 dir->i_size += DIRENT_SIZE;
455 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
456
457 d_instantiate(dentry, inode);
458 dget(dentry);
459 return 0;
460out_unlock:
461 spin_unlock(&mq_lock);
462 if (ipc_ns)
463 put_ipc_ns(ipc_ns);
464 return error;
465}
466
467static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
468{
469 struct inode *inode = dentry->d_inode;
470
471 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
472 dir->i_size -= DIRENT_SIZE;
473 drop_nlink(inode);
474 dput(dentry);
475 return 0;
476}
477
478/*
479* This is routine for system read from queue file.
480* To avoid mess with doing here some sort of mq_receive we allow
481* to read only queue size & notification info (the only values
482* that are interesting from user point of view and aren't accessible
483* through std routines)
484*/
485static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
486 size_t count, loff_t *off)
487{
488 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
489 char buffer[FILENT_SIZE];
490 ssize_t ret;
491
492 spin_lock(&info->lock);
493 snprintf(buffer, sizeof(buffer),
494 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
495 info->qsize,
496 info->notify_owner ? info->notify.sigev_notify : 0,
497 (info->notify_owner &&
498 info->notify.sigev_notify == SIGEV_SIGNAL) ?
499 info->notify.sigev_signo : 0,
500 pid_vnr(info->notify_owner));
501 spin_unlock(&info->lock);
502 buffer[sizeof(buffer)-1] = '\0';
503
504 ret = simple_read_from_buffer(u_data, count, off, buffer,
505 strlen(buffer));
506 if (ret <= 0)
507 return ret;
508
509 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
510 return ret;
511}
512
513static int mqueue_flush_file(struct file *filp, fl_owner_t id)
514{
515 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
516
517 spin_lock(&info->lock);
518 if (task_tgid(current) == info->notify_owner)
519 remove_notification(info);
520
521 spin_unlock(&info->lock);
522 return 0;
523}
524
525static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
526{
527 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
528 int retval = 0;
529
530 poll_wait(filp, &info->wait_q, poll_tab);
531
532 spin_lock(&info->lock);
533 if (info->attr.mq_curmsgs)
534 retval = POLLIN | POLLRDNORM;
535
536 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
537 retval |= POLLOUT | POLLWRNORM;
538 spin_unlock(&info->lock);
539
540 return retval;
541}
542
543/* Adds current to info->e_wait_q[sr] before element with smaller prio */
544static void wq_add(struct mqueue_inode_info *info, int sr,
545 struct ext_wait_queue *ewp)
546{
547 struct ext_wait_queue *walk;
548
549 ewp->task = current;
550
551 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
552 if (walk->task->static_prio <= current->static_prio) {
553 list_add_tail(&ewp->list, &walk->list);
554 return;
555 }
556 }
557 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
558}
559
560/*
561 * Puts current task to sleep. Caller must hold queue lock. After return
562 * lock isn't held.
563 * sr: SEND or RECV
564 */
565static int wq_sleep(struct mqueue_inode_info *info, int sr,
566 ktime_t *timeout, struct ext_wait_queue *ewp)
567{
568 int retval;
569 signed long time;
570
571 wq_add(info, sr, ewp);
572
573 for (;;) {
574 set_current_state(TASK_INTERRUPTIBLE);
575
576 spin_unlock(&info->lock);
577 time = schedule_hrtimeout_range_clock(timeout, 0,
578 HRTIMER_MODE_ABS, CLOCK_REALTIME);
579
580 while (ewp->state == STATE_PENDING)
581 cpu_relax();
582
583 if (ewp->state == STATE_READY) {
584 retval = 0;
585 goto out;
586 }
587 spin_lock(&info->lock);
588 if (ewp->state == STATE_READY) {
589 retval = 0;
590 goto out_unlock;
591 }
592 if (signal_pending(current)) {
593 retval = -ERESTARTSYS;
594 break;
595 }
596 if (time == 0) {
597 retval = -ETIMEDOUT;
598 break;
599 }
600 }
601 list_del(&ewp->list);
602out_unlock:
603 spin_unlock(&info->lock);
604out:
605 return retval;
606}
607
608/*
609 * Returns waiting task that should be serviced first or NULL if none exists
610 */
611static struct ext_wait_queue *wq_get_first_waiter(
612 struct mqueue_inode_info *info, int sr)
613{
614 struct list_head *ptr;
615
616 ptr = info->e_wait_q[sr].list.prev;
617 if (ptr == &info->e_wait_q[sr].list)
618 return NULL;
619 return list_entry(ptr, struct ext_wait_queue, list);
620}
621
622
623static inline void set_cookie(struct sk_buff *skb, char code)
624{
625 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
626}
627
628/*
629 * The next function is only to split too long sys_mq_timedsend
630 */
631static void __do_notify(struct mqueue_inode_info *info)
632{
633 /* notification
634 * invoked when there is registered process and there isn't process
635 * waiting synchronously for message AND state of queue changed from
636 * empty to not empty. Here we are sure that no one is waiting
637 * synchronously. */
638 if (info->notify_owner &&
639 info->attr.mq_curmsgs == 1) {
640 struct siginfo sig_i;
641 switch (info->notify.sigev_notify) {
642 case SIGEV_NONE:
643 break;
644 case SIGEV_SIGNAL:
645 /* sends signal */
646
647 sig_i.si_signo = info->notify.sigev_signo;
648 sig_i.si_errno = 0;
649 sig_i.si_code = SI_MESGQ;
650 sig_i.si_value = info->notify.sigev_value;
651 /* map current pid/uid into info->owner's namespaces */
652 rcu_read_lock();
653 sig_i.si_pid = task_tgid_nr_ns(current,
654 ns_of_pid(info->notify_owner));
655 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
656 rcu_read_unlock();
657
658 kill_pid_info(info->notify.sigev_signo,
659 &sig_i, info->notify_owner);
660 break;
661 case SIGEV_THREAD:
662 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
663 netlink_sendskb(info->notify_sock, info->notify_cookie);
664 break;
665 }
666 /* after notification unregisters process */
667 put_pid(info->notify_owner);
668 put_user_ns(info->notify_user_ns);
669 info->notify_owner = NULL;
670 info->notify_user_ns = NULL;
671 }
672 wake_up(&info->wait_q);
673}
674
675static int prepare_timeout(const struct timespec __user *u_abs_timeout,
676 ktime_t *expires, struct timespec *ts)
677{
678 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
679 return -EFAULT;
680 if (!timespec_valid(ts))
681 return -EINVAL;
682
683 *expires = timespec_to_ktime(*ts);
684 return 0;
685}
686
687static void remove_notification(struct mqueue_inode_info *info)
688{
689 if (info->notify_owner != NULL &&
690 info->notify.sigev_notify == SIGEV_THREAD) {
691 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
692 netlink_sendskb(info->notify_sock, info->notify_cookie);
693 }
694 put_pid(info->notify_owner);
695 put_user_ns(info->notify_user_ns);
696 info->notify_owner = NULL;
697 info->notify_user_ns = NULL;
698}
699
700static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
701{
702 int mq_treesize;
703 unsigned long total_size;
704
705 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
706 return -EINVAL;
707 if (capable(CAP_SYS_RESOURCE)) {
708 if (attr->mq_maxmsg > HARD_MSGMAX ||
709 attr->mq_msgsize > HARD_MSGSIZEMAX)
710 return -EINVAL;
711 } else {
712 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
713 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
714 return -EINVAL;
715 }
716 /* check for overflow */
717 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
718 return -EOVERFLOW;
719 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
720 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
721 sizeof(struct posix_msg_tree_node);
722 total_size = attr->mq_maxmsg * attr->mq_msgsize;
723 if (total_size + mq_treesize < total_size)
724 return -EOVERFLOW;
725 return 0;
726}
727
728/*
729 * Invoked when creating a new queue via sys_mq_open
730 */
731static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
732 struct path *path, int oflag, umode_t mode,
733 struct mq_attr *attr)
734{
735 const struct cred *cred = current_cred();
736 int ret;
737
738 if (attr) {
739 ret = mq_attr_ok(ipc_ns, attr);
740 if (ret)
741 return ERR_PTR(ret);
742 /* store for use during create */
743 path->dentry->d_fsdata = attr;
744 } else {
745 struct mq_attr def_attr;
746
747 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
748 ipc_ns->mq_msg_default);
749 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
750 ipc_ns->mq_msgsize_default);
751 ret = mq_attr_ok(ipc_ns, &def_attr);
752 if (ret)
753 return ERR_PTR(ret);
754 }
755
756 mode &= ~current_umask();
757 ret = vfs_create(dir, path->dentry, mode, true);
758 path->dentry->d_fsdata = NULL;
759 if (ret)
760 return ERR_PTR(ret);
761 return dentry_open(path, oflag, cred);
762}
763
764/* Opens existing queue */
765static struct file *do_open(struct path *path, int oflag)
766{
767 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
768 MAY_READ | MAY_WRITE };
769 int acc;
770 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
771 return ERR_PTR(-EINVAL);
772 acc = oflag2acc[oflag & O_ACCMODE];
773 if (inode_permission(path->dentry->d_inode, acc))
774 return ERR_PTR(-EACCES);
775 return dentry_open(path, oflag, current_cred());
776}
777
778SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
779 struct mq_attr __user *, u_attr)
780{
781 struct path path;
782 struct file *filp;
783 struct filename *name;
784 struct mq_attr attr;
785 int fd, error;
786 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
787 struct vfsmount *mnt = ipc_ns->mq_mnt;
788 struct dentry *root = mnt->mnt_root;
789 int ro;
790
791 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
792 return -EFAULT;
793
794 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
795
796 if (IS_ERR(name = getname(u_name)))
797 return PTR_ERR(name);
798
799 fd = get_unused_fd_flags(O_CLOEXEC);
800 if (fd < 0)
801 goto out_putname;
802
803 ro = mnt_want_write(mnt); /* we'll drop it in any case */
804 error = 0;
805 mutex_lock(&root->d_inode->i_mutex);
806 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
807 if (IS_ERR(path.dentry)) {
808 error = PTR_ERR(path.dentry);
809 goto out_putfd;
810 }
811 path.mnt = mntget(mnt);
812
813 if (oflag & O_CREAT) {
814 if (path.dentry->d_inode) { /* entry already exists */
815 audit_inode(name, path.dentry, 0);
816 if (oflag & O_EXCL) {
817 error = -EEXIST;
818 goto out;
819 }
820 filp = do_open(&path, oflag);
821 } else {
822 if (ro) {
823 error = ro;
824 goto out;
825 }
826 audit_inode_parent_hidden(name, root);
827 filp = do_create(ipc_ns, root->d_inode,
828 &path, oflag, mode,
829 u_attr ? &attr : NULL);
830 }
831 } else {
832 if (!path.dentry->d_inode) {
833 error = -ENOENT;
834 goto out;
835 }
836 audit_inode(name, path.dentry, 0);
837 filp = do_open(&path, oflag);
838 }
839
840 if (!IS_ERR(filp))
841 fd_install(fd, filp);
842 else
843 error = PTR_ERR(filp);
844out:
845 path_put(&path);
846out_putfd:
847 if (error) {
848 put_unused_fd(fd);
849 fd = error;
850 }
851 mutex_unlock(&root->d_inode->i_mutex);
852 if (!ro)
853 mnt_drop_write(mnt);
854out_putname:
855 putname(name);
856 return fd;
857}
858
859SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
860{
861 int err;
862 struct filename *name;
863 struct dentry *dentry;
864 struct inode *inode = NULL;
865 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
866 struct vfsmount *mnt = ipc_ns->mq_mnt;
867
868 name = getname(u_name);
869 if (IS_ERR(name))
870 return PTR_ERR(name);
871
872 audit_inode_parent_hidden(name, mnt->mnt_root);
873 err = mnt_want_write(mnt);
874 if (err)
875 goto out_name;
876 mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
877 dentry = lookup_one_len(name->name, mnt->mnt_root,
878 strlen(name->name));
879 if (IS_ERR(dentry)) {
880 err = PTR_ERR(dentry);
881 goto out_unlock;
882 }
883
884 inode = dentry->d_inode;
885 if (!inode) {
886 err = -ENOENT;
887 } else {
888 ihold(inode);
889 err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
890 }
891 dput(dentry);
892
893out_unlock:
894 mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
895 if (inode)
896 iput(inode);
897 mnt_drop_write(mnt);
898out_name:
899 putname(name);
900
901 return err;
902}
903
904/* Pipelined send and receive functions.
905 *
906 * If a receiver finds no waiting message, then it registers itself in the
907 * list of waiting receivers. A sender checks that list before adding the new
908 * message into the message array. If there is a waiting receiver, then it
909 * bypasses the message array and directly hands the message over to the
910 * receiver.
911 * The receiver accepts the message and returns without grabbing the queue
912 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
913 * are necessary. The same algorithm is used for sysv semaphores, see
914 * ipc/sem.c for more details.
915 *
916 * The same algorithm is used for senders.
917 */
918
919/* pipelined_send() - send a message directly to the task waiting in
920 * sys_mq_timedreceive() (without inserting message into a queue).
921 */
922static inline void pipelined_send(struct mqueue_inode_info *info,
923 struct msg_msg *message,
924 struct ext_wait_queue *receiver)
925{
926 receiver->msg = message;
927 list_del(&receiver->list);
928 receiver->state = STATE_PENDING;
929 wake_up_process(receiver->task);
930 smp_wmb();
931 receiver->state = STATE_READY;
932}
933
934/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
935 * gets its message and put to the queue (we have one free place for sure). */
936static inline void pipelined_receive(struct mqueue_inode_info *info)
937{
938 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
939
940 if (!sender) {
941 /* for poll */
942 wake_up_interruptible(&info->wait_q);
943 return;
944 }
945 if (msg_insert(sender->msg, info))
946 return;
947 list_del(&sender->list);
948 sender->state = STATE_PENDING;
949 wake_up_process(sender->task);
950 smp_wmb();
951 sender->state = STATE_READY;
952}
953
954SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
955 size_t, msg_len, unsigned int, msg_prio,
956 const struct timespec __user *, u_abs_timeout)
957{
958 struct fd f;
959 struct inode *inode;
960 struct ext_wait_queue wait;
961 struct ext_wait_queue *receiver;
962 struct msg_msg *msg_ptr;
963 struct mqueue_inode_info *info;
964 ktime_t expires, *timeout = NULL;
965 struct timespec ts;
966 struct posix_msg_tree_node *new_leaf = NULL;
967 int ret = 0;
968
969 if (u_abs_timeout) {
970 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
971 if (res)
972 return res;
973 timeout = &expires;
974 }
975
976 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
977 return -EINVAL;
978
979 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
980
981 f = fdget(mqdes);
982 if (unlikely(!f.file)) {
983 ret = -EBADF;
984 goto out;
985 }
986
987 inode = file_inode(f.file);
988 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
989 ret = -EBADF;
990 goto out_fput;
991 }
992 info = MQUEUE_I(inode);
993 audit_inode(NULL, f.file->f_path.dentry, 0);
994
995 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
996 ret = -EBADF;
997 goto out_fput;
998 }
999
1000 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1001 ret = -EMSGSIZE;
1002 goto out_fput;
1003 }
1004
1005 /* First try to allocate memory, before doing anything with
1006 * existing queues. */
1007 msg_ptr = load_msg(u_msg_ptr, msg_len);
1008 if (IS_ERR(msg_ptr)) {
1009 ret = PTR_ERR(msg_ptr);
1010 goto out_fput;
1011 }
1012 msg_ptr->m_ts = msg_len;
1013 msg_ptr->m_type = msg_prio;
1014
1015 /*
1016 * msg_insert really wants us to have a valid, spare node struct so
1017 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1018 * fall back to that if necessary.
1019 */
1020 if (!info->node_cache)
1021 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1022
1023 spin_lock(&info->lock);
1024
1025 if (!info->node_cache && new_leaf) {
1026 /* Save our speculative allocation into the cache */
1027 INIT_LIST_HEAD(&new_leaf->msg_list);
1028 info->node_cache = new_leaf;
1029 info->qsize += sizeof(*new_leaf);
1030 new_leaf = NULL;
1031 } else {
1032 kfree(new_leaf);
1033 }
1034
1035 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1036 if (f.file->f_flags & O_NONBLOCK) {
1037 ret = -EAGAIN;
1038 } else {
1039 wait.task = current;
1040 wait.msg = (void *) msg_ptr;
1041 wait.state = STATE_NONE;
1042 ret = wq_sleep(info, SEND, timeout, &wait);
1043 /*
1044 * wq_sleep must be called with info->lock held, and
1045 * returns with the lock released
1046 */
1047 goto out_free;
1048 }
1049 } else {
1050 receiver = wq_get_first_waiter(info, RECV);
1051 if (receiver) {
1052 pipelined_send(info, msg_ptr, receiver);
1053 } else {
1054 /* adds message to the queue */
1055 ret = msg_insert(msg_ptr, info);
1056 if (ret)
1057 goto out_unlock;
1058 __do_notify(info);
1059 }
1060 inode->i_atime = inode->i_mtime = inode->i_ctime =
1061 CURRENT_TIME;
1062 }
1063out_unlock:
1064 spin_unlock(&info->lock);
1065out_free:
1066 if (ret)
1067 free_msg(msg_ptr);
1068out_fput:
1069 fdput(f);
1070out:
1071 return ret;
1072}
1073
1074SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1075 size_t, msg_len, unsigned int __user *, u_msg_prio,
1076 const struct timespec __user *, u_abs_timeout)
1077{
1078 ssize_t ret;
1079 struct msg_msg *msg_ptr;
1080 struct fd f;
1081 struct inode *inode;
1082 struct mqueue_inode_info *info;
1083 struct ext_wait_queue wait;
1084 ktime_t expires, *timeout = NULL;
1085 struct timespec ts;
1086 struct posix_msg_tree_node *new_leaf = NULL;
1087
1088 if (u_abs_timeout) {
1089 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1090 if (res)
1091 return res;
1092 timeout = &expires;
1093 }
1094
1095 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1096
1097 f = fdget(mqdes);
1098 if (unlikely(!f.file)) {
1099 ret = -EBADF;
1100 goto out;
1101 }
1102
1103 inode = file_inode(f.file);
1104 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1105 ret = -EBADF;
1106 goto out_fput;
1107 }
1108 info = MQUEUE_I(inode);
1109 audit_inode(NULL, f.file->f_path.dentry, 0);
1110
1111 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1112 ret = -EBADF;
1113 goto out_fput;
1114 }
1115
1116 /* checks if buffer is big enough */
1117 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1118 ret = -EMSGSIZE;
1119 goto out_fput;
1120 }
1121
1122 /*
1123 * msg_insert really wants us to have a valid, spare node struct so
1124 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1125 * fall back to that if necessary.
1126 */
1127 if (!info->node_cache)
1128 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1129
1130 spin_lock(&info->lock);
1131
1132 if (!info->node_cache && new_leaf) {
1133 /* Save our speculative allocation into the cache */
1134 INIT_LIST_HEAD(&new_leaf->msg_list);
1135 info->node_cache = new_leaf;
1136 info->qsize += sizeof(*new_leaf);
1137 } else {
1138 kfree(new_leaf);
1139 }
1140
1141 if (info->attr.mq_curmsgs == 0) {
1142 if (f.file->f_flags & O_NONBLOCK) {
1143 spin_unlock(&info->lock);
1144 ret = -EAGAIN;
1145 } else {
1146 wait.task = current;
1147 wait.state = STATE_NONE;
1148 ret = wq_sleep(info, RECV, timeout, &wait);
1149 msg_ptr = wait.msg;
1150 }
1151 } else {
1152 msg_ptr = msg_get(info);
1153
1154 inode->i_atime = inode->i_mtime = inode->i_ctime =
1155 CURRENT_TIME;
1156
1157 /* There is now free space in queue. */
1158 pipelined_receive(info);
1159 spin_unlock(&info->lock);
1160 ret = 0;
1161 }
1162 if (ret == 0) {
1163 ret = msg_ptr->m_ts;
1164
1165 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1166 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1167 ret = -EFAULT;
1168 }
1169 free_msg(msg_ptr);
1170 }
1171out_fput:
1172 fdput(f);
1173out:
1174 return ret;
1175}
1176
1177/*
1178 * Notes: the case when user wants us to deregister (with NULL as pointer)
1179 * and he isn't currently owner of notification, will be silently discarded.
1180 * It isn't explicitly defined in the POSIX.
1181 */
1182SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1183 const struct sigevent __user *, u_notification)
1184{
1185 int ret;
1186 struct fd f;
1187 struct sock *sock;
1188 struct inode *inode;
1189 struct sigevent notification;
1190 struct mqueue_inode_info *info;
1191 struct sk_buff *nc;
1192
1193 if (u_notification) {
1194 if (copy_from_user(¬ification, u_notification,
1195 sizeof(struct sigevent)))
1196 return -EFAULT;
1197 }
1198
1199 audit_mq_notify(mqdes, u_notification ? ¬ification : NULL);
1200
1201 nc = NULL;
1202 sock = NULL;
1203 if (u_notification != NULL) {
1204 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1205 notification.sigev_notify != SIGEV_SIGNAL &&
1206 notification.sigev_notify != SIGEV_THREAD))
1207 return -EINVAL;
1208 if (notification.sigev_notify == SIGEV_SIGNAL &&
1209 !valid_signal(notification.sigev_signo)) {
1210 return -EINVAL;
1211 }
1212 if (notification.sigev_notify == SIGEV_THREAD) {
1213 long timeo;
1214
1215 /* create the notify skb */
1216 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1217 if (!nc) {
1218 ret = -ENOMEM;
1219 goto out;
1220 }
1221 if (copy_from_user(nc->data,
1222 notification.sigev_value.sival_ptr,
1223 NOTIFY_COOKIE_LEN)) {
1224 ret = -EFAULT;
1225 goto out;
1226 }
1227
1228 /* TODO: add a header? */
1229 skb_put(nc, NOTIFY_COOKIE_LEN);
1230 /* and attach it to the socket */
1231retry:
1232 f = fdget(notification.sigev_signo);
1233 if (!f.file) {
1234 ret = -EBADF;
1235 goto out;
1236 }
1237 sock = netlink_getsockbyfilp(f.file);
1238 fdput(f);
1239 if (IS_ERR(sock)) {
1240 ret = PTR_ERR(sock);
1241 sock = NULL;
1242 goto out;
1243 }
1244
1245 timeo = MAX_SCHEDULE_TIMEOUT;
1246 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1247 if (ret == 1)
1248 goto retry;
1249 if (ret) {
1250 sock = NULL;
1251 nc = NULL;
1252 goto out;
1253 }
1254 }
1255 }
1256
1257 f = fdget(mqdes);
1258 if (!f.file) {
1259 ret = -EBADF;
1260 goto out;
1261 }
1262
1263 inode = file_inode(f.file);
1264 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1265 ret = -EBADF;
1266 goto out_fput;
1267 }
1268 info = MQUEUE_I(inode);
1269
1270 ret = 0;
1271 spin_lock(&info->lock);
1272 if (u_notification == NULL) {
1273 if (info->notify_owner == task_tgid(current)) {
1274 remove_notification(info);
1275 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1276 }
1277 } else if (info->notify_owner != NULL) {
1278 ret = -EBUSY;
1279 } else {
1280 switch (notification.sigev_notify) {
1281 case SIGEV_NONE:
1282 info->notify.sigev_notify = SIGEV_NONE;
1283 break;
1284 case SIGEV_THREAD:
1285 info->notify_sock = sock;
1286 info->notify_cookie = nc;
1287 sock = NULL;
1288 nc = NULL;
1289 info->notify.sigev_notify = SIGEV_THREAD;
1290 break;
1291 case SIGEV_SIGNAL:
1292 info->notify.sigev_signo = notification.sigev_signo;
1293 info->notify.sigev_value = notification.sigev_value;
1294 info->notify.sigev_notify = SIGEV_SIGNAL;
1295 break;
1296 }
1297
1298 info->notify_owner = get_pid(task_tgid(current));
1299 info->notify_user_ns = get_user_ns(current_user_ns());
1300 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1301 }
1302 spin_unlock(&info->lock);
1303out_fput:
1304 fdput(f);
1305out:
1306 if (sock)
1307 netlink_detachskb(sock, nc);
1308 else if (nc)
1309 dev_kfree_skb(nc);
1310
1311 return ret;
1312}
1313
1314SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1315 const struct mq_attr __user *, u_mqstat,
1316 struct mq_attr __user *, u_omqstat)
1317{
1318 int ret;
1319 struct mq_attr mqstat, omqstat;
1320 struct fd f;
1321 struct inode *inode;
1322 struct mqueue_inode_info *info;
1323
1324 if (u_mqstat != NULL) {
1325 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1326 return -EFAULT;
1327 if (mqstat.mq_flags & (~O_NONBLOCK))
1328 return -EINVAL;
1329 }
1330
1331 f = fdget(mqdes);
1332 if (!f.file) {
1333 ret = -EBADF;
1334 goto out;
1335 }
1336
1337 inode = file_inode(f.file);
1338 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1339 ret = -EBADF;
1340 goto out_fput;
1341 }
1342 info = MQUEUE_I(inode);
1343
1344 spin_lock(&info->lock);
1345
1346 omqstat = info->attr;
1347 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1348 if (u_mqstat) {
1349 audit_mq_getsetattr(mqdes, &mqstat);
1350 spin_lock(&f.file->f_lock);
1351 if (mqstat.mq_flags & O_NONBLOCK)
1352 f.file->f_flags |= O_NONBLOCK;
1353 else
1354 f.file->f_flags &= ~O_NONBLOCK;
1355 spin_unlock(&f.file->f_lock);
1356
1357 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1358 }
1359
1360 spin_unlock(&info->lock);
1361
1362 ret = 0;
1363 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1364 sizeof(struct mq_attr)))
1365 ret = -EFAULT;
1366
1367out_fput:
1368 fdput(f);
1369out:
1370 return ret;
1371}
1372
1373static const struct inode_operations mqueue_dir_inode_operations = {
1374 .lookup = simple_lookup,
1375 .create = mqueue_create,
1376 .unlink = mqueue_unlink,
1377};
1378
1379static const struct file_operations mqueue_file_operations = {
1380 .flush = mqueue_flush_file,
1381 .poll = mqueue_poll_file,
1382 .read = mqueue_read_file,
1383 .llseek = default_llseek,
1384};
1385
1386static const struct super_operations mqueue_super_ops = {
1387 .alloc_inode = mqueue_alloc_inode,
1388 .destroy_inode = mqueue_destroy_inode,
1389 .evict_inode = mqueue_evict_inode,
1390 .statfs = simple_statfs,
1391};
1392
1393static struct file_system_type mqueue_fs_type = {
1394 .name = "mqueue",
1395 .mount = mqueue_mount,
1396 .kill_sb = kill_litter_super,
1397 .fs_flags = FS_USERNS_MOUNT,
1398};
1399
1400int mq_init_ns(struct ipc_namespace *ns)
1401{
1402 ns->mq_queues_count = 0;
1403 ns->mq_queues_max = DFLT_QUEUESMAX;
1404 ns->mq_msg_max = DFLT_MSGMAX;
1405 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
1406 ns->mq_msg_default = DFLT_MSG;
1407 ns->mq_msgsize_default = DFLT_MSGSIZE;
1408
1409 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1410 if (IS_ERR(ns->mq_mnt)) {
1411 int err = PTR_ERR(ns->mq_mnt);
1412 ns->mq_mnt = NULL;
1413 return err;
1414 }
1415 return 0;
1416}
1417
1418void mq_clear_sbinfo(struct ipc_namespace *ns)
1419{
1420 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1421}
1422
1423void mq_put_mnt(struct ipc_namespace *ns)
1424{
1425 kern_unmount(ns->mq_mnt);
1426}
1427
1428static int __init init_mqueue_fs(void)
1429{
1430 int error;
1431
1432 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1433 sizeof(struct mqueue_inode_info), 0,
1434 SLAB_HWCACHE_ALIGN, init_once);
1435 if (mqueue_inode_cachep == NULL)
1436 return -ENOMEM;
1437
1438 /* ignore failures - they are not fatal */
1439 mq_sysctl_table = mq_register_sysctl_table();
1440
1441 error = register_filesystem(&mqueue_fs_type);
1442 if (error)
1443 goto out_sysctl;
1444
1445 spin_lock_init(&mq_lock);
1446
1447 error = mq_init_ns(&init_ipc_ns);
1448 if (error)
1449 goto out_filesystem;
1450
1451 return 0;
1452
1453out_filesystem:
1454 unregister_filesystem(&mqueue_fs_type);
1455out_sysctl:
1456 if (mq_sysctl_table)
1457 unregister_sysctl_table(mq_sysctl_table);
1458 kmem_cache_destroy(mqueue_inode_cachep);
1459 return error;
1460}
1461
1462device_initcall(init_mqueue_fs);