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