Merge remote-tracking branch 'asoc/topic/core' into asoc-next
[linux-2.6-block.git] / ipc / mqueue.c
CommitLineData
1da177e4
LT
1/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
f66e928b 5 * Michal Wronski (michal.wronski@gmail.com)
1da177e4
LT
6 *
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
239521f3 9 * Manfred Spraul (manfred@colorfullife.com)
1da177e4 10 *
20ca73bc
GW
11 * Audit: George Wilson (ltcgcw@us.ibm.com)
12 *
1da177e4
LT
13 * This file is released under the GPL.
14 */
15
c59ede7b 16#include <linux/capability.h>
1da177e4
LT
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>
5b5c4d1a 27#include <linux/vmalloc.h>
1da177e4
LT
28#include <linux/netlink.h>
29#include <linux/syscalls.h>
20ca73bc 30#include <linux/audit.h>
7ed20e1a 31#include <linux/signal.h>
5f921ae9 32#include <linux/mutex.h>
b488893a
PE
33#include <linux/nsproxy.h>
34#include <linux/pid.h>
614b84cf 35#include <linux/ipc_namespace.h>
6b550f94 36#include <linux/user_namespace.h>
5a0e3ad6 37#include <linux/slab.h>
5f921ae9 38
1da177e4
LT
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
fa6004ad 50#define STATE_READY 1
1da177e4 51
d6629859
DL
52struct posix_msg_tree_node {
53 struct rb_node rb_node;
54 struct list_head msg_list;
55 int priority;
56};
57
1da177e4
LT
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
d6629859 70 struct rb_root msg_tree;
ce2d52cc 71 struct posix_msg_tree_node *node_cache;
1da177e4
LT
72 struct mq_attr attr;
73
74 struct sigevent notify;
239521f3 75 struct pid *notify_owner;
6f9ac6d9 76 struct user_namespace *notify_user_ns;
338cec32 77 struct user_struct *user; /* user who created, for accounting */
1da177e4
LT
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
92e1d5be 87static const struct inode_operations mqueue_dir_inode_operations;
9a32144e 88static const struct file_operations mqueue_file_operations;
b87221de 89static const struct super_operations mqueue_super_ops;
1da177e4
LT
90static void remove_notification(struct mqueue_inode_info *info);
91
e18b890b 92static struct kmem_cache *mqueue_inode_cachep;
1da177e4 93
239521f3 94static struct ctl_table_header *mq_sysctl_table;
1da177e4
LT
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
7eafd7c7
SH
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)
614b84cf 105{
7eafd7c7 106 return get_ipc_ns(inode->i_sb->s_fs_info);
614b84cf
SH
107}
108
7eafd7c7 109static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
614b84cf 110{
7eafd7c7
SH
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;
614b84cf
SH
117}
118
d6629859
DL
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 }
ce2d52cc
DL
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;
ce2d52cc 144 INIT_LIST_HEAD(&leaf->msg_list);
ce2d52cc 145 }
d6629859
DL
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);
d6629859
DL
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);
ce2d52cc 183 if (unlikely(list_empty(&leaf->msg_list))) {
d6629859
DL
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);
ce2d52cc 188 if (info->node_cache) {
ce2d52cc
DL
189 kfree(leaf);
190 } else {
191 info->node_cache = leaf;
192 }
d6629859
DL
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);
ce2d52cc 200 if (info->node_cache) {
ce2d52cc
DL
201 kfree(leaf);
202 } else {
203 info->node_cache = leaf;
204 }
d6629859
DL
205 }
206 }
207 info->attr.mq_curmsgs--;
208 info->qsize -= msg->m_ts;
209 return msg;
210}
211
7eafd7c7 212static struct inode *mqueue_get_inode(struct super_block *sb,
1b9d5ff7 213 struct ipc_namespace *ipc_ns, umode_t mode,
7eafd7c7 214 struct mq_attr *attr)
1da177e4 215{
86a264ab 216 struct user_struct *u = current_user();
1da177e4 217 struct inode *inode;
d40dcdb0 218 int ret = -ENOMEM;
1da177e4
LT
219
220 inode = new_inode(sb);
04715206
JS
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;
d6629859 232 unsigned long mq_bytes, mq_treesize;
04715206
JS
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;
6f9ac6d9 243 info->notify_user_ns = NULL;
04715206
JS
244 info->qsize = 0;
245 info->user = NULL; /* set when all is ok */
d6629859 246 info->msg_tree = RB_ROOT;
ce2d52cc 247 info->node_cache = NULL;
04715206 248 memset(&info->attr, 0, sizeof(info->attr));
cef0184c
KM
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);
04715206
JS
253 if (attr) {
254 info->attr.mq_maxmsg = attr->mq_maxmsg;
255 info->attr.mq_msgsize = attr->mq_msgsize;
256 }
d6629859
DL
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);
04715206 273
d6629859
DL
274 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
275 info->attr.mq_msgsize);
1da177e4 276
04715206
JS
277 spin_lock(&mq_lock);
278 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
2a4e64b8 279 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
04715206
JS
280 spin_unlock(&mq_lock);
281 /* mqueue_evict_inode() releases info->messages */
d40dcdb0 282 ret = -EMFILE;
04715206 283 goto out_inode;
1da177e4 284 }
04715206
JS
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;
1da177e4 296 }
04715206 297
1da177e4
LT
298 return inode;
299out_inode:
1da177e4 300 iput(inode);
04715206 301err:
d40dcdb0 302 return ERR_PTR(ret);
1da177e4
LT
303}
304
305static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
306{
307 struct inode *inode;
7eafd7c7 308 struct ipc_namespace *ns = data;
1da177e4
LT
309
310 sb->s_blocksize = PAGE_CACHE_SIZE;
311 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
312 sb->s_magic = MQUEUE_MAGIC;
313 sb->s_op = &mqueue_super_ops;
314
48fde701
AV
315 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
316 if (IS_ERR(inode))
317 return PTR_ERR(inode);
1da177e4 318
48fde701
AV
319 sb->s_root = d_make_root(inode);
320 if (!sb->s_root)
321 return -ENOMEM;
322 return 0;
1da177e4
LT
323}
324
ceefda69 325static struct dentry *mqueue_mount(struct file_system_type *fs_type,
454e2398 326 int flags, const char *dev_name,
ceefda69 327 void *data)
1da177e4 328{
a636b702
EB
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 }
ceefda69 339 return mount_ns(fs_type, flags, data, mqueue_fill_super);
1da177e4
LT
340}
341
51cc5068 342static void init_once(void *foo)
1da177e4
LT
343{
344 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
345
a35afb83 346 inode_init_once(&p->vfs_inode);
1da177e4
LT
347}
348
349static struct inode *mqueue_alloc_inode(struct super_block *sb)
350{
351 struct mqueue_inode_info *ei;
352
e94b1766 353 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
1da177e4
LT
354 if (!ei)
355 return NULL;
356 return &ei->vfs_inode;
357}
358
fa0d7e3d 359static void mqueue_i_callback(struct rcu_head *head)
1da177e4 360{
fa0d7e3d 361 struct inode *inode = container_of(head, struct inode, i_rcu);
1da177e4
LT
362 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
363}
364
fa0d7e3d
NP
365static void mqueue_destroy_inode(struct inode *inode)
366{
367 call_rcu(&inode->i_rcu, mqueue_i_callback);
368}
369
6d8af64c 370static void mqueue_evict_inode(struct inode *inode)
1da177e4
LT
371{
372 struct mqueue_inode_info *info;
373 struct user_struct *user;
d6629859 374 unsigned long mq_bytes, mq_treesize;
7eafd7c7 375 struct ipc_namespace *ipc_ns;
d6629859 376 struct msg_msg *msg;
1da177e4 377
dbd5768f 378 clear_inode(inode);
6d8af64c
AV
379
380 if (S_ISDIR(inode->i_mode))
1da177e4 381 return;
6d8af64c 382
7eafd7c7 383 ipc_ns = get_ns_from_inode(inode);
1da177e4
LT
384 info = MQUEUE_I(inode);
385 spin_lock(&info->lock);
d6629859
DL
386 while ((msg = msg_get(info)) != NULL)
387 free_msg(msg);
ce2d52cc 388 kfree(info->node_cache);
1da177e4
LT
389 spin_unlock(&info->lock);
390
8834cf79 391 /* Total amount of bytes accounted for the mqueue */
d6629859
DL
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
1da177e4
LT
399 user = info->user;
400 if (user) {
401 spin_lock(&mq_lock);
402 user->mq_bytes -= mq_bytes;
7eafd7c7
SH
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--;
1da177e4
LT
411 spin_unlock(&mq_lock);
412 free_uid(user);
413 }
7eafd7c7
SH
414 if (ipc_ns)
415 put_ipc_ns(ipc_ns);
1da177e4
LT
416}
417
418static int mqueue_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 419 umode_t mode, bool excl)
1da177e4
LT
420{
421 struct inode *inode;
422 struct mq_attr *attr = dentry->d_fsdata;
423 int error;
7eafd7c7 424 struct ipc_namespace *ipc_ns;
1da177e4
LT
425
426 spin_lock(&mq_lock);
7eafd7c7
SH
427 ipc_ns = __get_ns_from_inode(dir);
428 if (!ipc_ns) {
429 error = -EACCES;
430 goto out_unlock;
431 }
f3713fd9
DB
432
433 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
434 !capable(CAP_SYS_RESOURCE)) {
1da177e4 435 error = -ENOSPC;
614b84cf 436 goto out_unlock;
1da177e4 437 }
614b84cf 438 ipc_ns->mq_queues_count++;
1da177e4
LT
439 spin_unlock(&mq_lock);
440
7eafd7c7 441 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
d40dcdb0
JS
442 if (IS_ERR(inode)) {
443 error = PTR_ERR(inode);
1da177e4 444 spin_lock(&mq_lock);
614b84cf
SH
445 ipc_ns->mq_queues_count--;
446 goto out_unlock;
1da177e4
LT
447 }
448
7eafd7c7 449 put_ipc_ns(ipc_ns);
1da177e4
LT
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;
614b84cf 456out_unlock:
1da177e4 457 spin_unlock(&mq_lock);
7eafd7c7
SH
458 if (ipc_ns)
459 put_ipc_ns(ipc_ns);
1da177e4
LT
460 return error;
461}
462
463static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
464{
75c3cfa8 465 struct inode *inode = d_inode(dentry);
1da177e4
LT
466
467 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
468 dir->i_size -= DIRENT_SIZE;
239521f3
MS
469 drop_nlink(inode);
470 dput(dentry);
471 return 0;
1da177e4
LT
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,
f1a43f93 482 size_t count, loff_t *off)
1da177e4 483{
496ad9aa 484 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4 485 char buffer[FILENT_SIZE];
f1a43f93 486 ssize_t ret;
1da177e4
LT
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,
6c5f3e7b 496 pid_vnr(info->notify_owner));
1da177e4
LT
497 spin_unlock(&info->lock);
498 buffer[sizeof(buffer)-1] = '\0';
1da177e4 499
f1a43f93
AM
500 ret = simple_read_from_buffer(u_data, count, off, buffer,
501 strlen(buffer));
502 if (ret <= 0)
503 return ret;
1da177e4 504
496ad9aa 505 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
f1a43f93 506 return ret;
1da177e4
LT
507}
508
75e1fcc0 509static int mqueue_flush_file(struct file *filp, fl_owner_t id)
1da177e4 510{
496ad9aa 511 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4
LT
512
513 spin_lock(&info->lock);
a03fcb73 514 if (task_tgid(current) == info->notify_owner)
1da177e4
LT
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{
496ad9aa 523 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4
LT
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,
9ca7d8e6 562 ktime_t *timeout, struct ext_wait_queue *ewp)
1da177e4
LT
563{
564 int retval;
565 signed long time;
566
567 wq_add(info, sr, ewp);
568
569 for (;;) {
fa6004ad 570 __set_current_state(TASK_INTERRUPTIBLE);
1da177e4
LT
571
572 spin_unlock(&info->lock);
32ea845d
WG
573 time = schedule_hrtimeout_range_clock(timeout, 0,
574 HRTIMER_MODE_ABS, CLOCK_REALTIME);
1da177e4 575
1da177e4
LT
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
1da177e4
LT
615
616static inline void set_cookie(struct sk_buff *skb, char code)
617{
239521f3 618 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
1da177e4
LT
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;
6b550f94
SH
644 /* map current pid/uid into info->owner's namespaces */
645 rcu_read_lock();
a6684999
SB
646 sig_i.si_pid = task_tgid_nr_ns(current,
647 ns_of_pid(info->notify_owner));
76b6db01 648 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
6b550f94 649 rcu_read_unlock();
1da177e4 650
a03fcb73
CLG
651 kill_pid_info(info->notify.sigev_signo,
652 &sig_i, info->notify_owner);
1da177e4
LT
653 break;
654 case SIGEV_THREAD:
655 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
7ee015e0 656 netlink_sendskb(info->notify_sock, info->notify_cookie);
1da177e4
LT
657 break;
658 }
659 /* after notification unregisters process */
a03fcb73 660 put_pid(info->notify_owner);
6f9ac6d9 661 put_user_ns(info->notify_user_ns);
a03fcb73 662 info->notify_owner = NULL;
6f9ac6d9 663 info->notify_user_ns = NULL;
1da177e4
LT
664 }
665 wake_up(&info->wait_q);
666}
667
9ca7d8e6
CE
668static int prepare_timeout(const struct timespec __user *u_abs_timeout,
669 ktime_t *expires, struct timespec *ts)
1da177e4 670{
9ca7d8e6
CE
671 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
672 return -EFAULT;
673 if (!timespec_valid(ts))
674 return -EINVAL;
1da177e4 675
9ca7d8e6
CE
676 *expires = timespec_to_ktime(*ts);
677 return 0;
1da177e4
LT
678}
679
680static void remove_notification(struct mqueue_inode_info *info)
681{
a03fcb73 682 if (info->notify_owner != NULL &&
1da177e4
LT
683 info->notify.sigev_notify == SIGEV_THREAD) {
684 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
7ee015e0 685 netlink_sendskb(info->notify_sock, info->notify_cookie);
1da177e4 686 }
a03fcb73 687 put_pid(info->notify_owner);
6f9ac6d9 688 put_user_ns(info->notify_user_ns);
a03fcb73 689 info->notify_owner = NULL;
6f9ac6d9 690 info->notify_user_ns = NULL;
1da177e4
LT
691}
692
614b84cf 693static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
1da177e4 694{
2c12ea49
DL
695 int mq_treesize;
696 unsigned long total_size;
697
1da177e4 698 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
113289cc 699 return -EINVAL;
1da177e4 700 if (capable(CAP_SYS_RESOURCE)) {
02967ea0
DL
701 if (attr->mq_maxmsg > HARD_MSGMAX ||
702 attr->mq_msgsize > HARD_MSGSIZEMAX)
113289cc 703 return -EINVAL;
1da177e4 704 } else {
614b84cf
SH
705 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
706 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
113289cc 707 return -EINVAL;
1da177e4
LT
708 }
709 /* check for overflow */
710 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
113289cc 711 return -EOVERFLOW;
2c12ea49
DL
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)
113289cc
DL
717 return -EOVERFLOW;
718 return 0;
1da177e4
LT
719}
720
721/*
722 * Invoked when creating a new queue via sys_mq_open
723 */
765927b2
AV
724static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
725 struct path *path, int oflag, umode_t mode,
614b84cf 726 struct mq_attr *attr)
1da177e4 727{
745ca247 728 const struct cred *cred = current_cred();
1da177e4
LT
729 int ret;
730
564f6993 731 if (attr) {
113289cc
DL
732 ret = mq_attr_ok(ipc_ns, attr);
733 if (ret)
765927b2 734 return ERR_PTR(ret);
1da177e4 735 /* store for use during create */
765927b2 736 path->dentry->d_fsdata = attr;
113289cc
DL
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)
765927b2 746 return ERR_PTR(ret);
1da177e4
LT
747 }
748
ce3b0f8d 749 mode &= ~current_umask();
765927b2
AV
750 ret = vfs_create(dir, path->dentry, mode, true);
751 path->dentry->d_fsdata = NULL;
312b90fb
AV
752 if (ret)
753 return ERR_PTR(ret);
754 return dentry_open(path, oflag, cred);
1da177e4
LT
755}
756
757/* Opens existing queue */
765927b2 758static struct file *do_open(struct path *path, int oflag)
1da177e4 759{
745ca247
DH
760 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
761 MAY_READ | MAY_WRITE };
765927b2
AV
762 int acc;
763 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
764 return ERR_PTR(-EINVAL);
765 acc = oflag2acc[oflag & O_ACCMODE];
75c3cfa8 766 if (inode_permission(d_inode(path->dentry), acc))
765927b2
AV
767 return ERR_PTR(-EACCES);
768 return dentry_open(path, oflag, current_cred());
1da177e4
LT
769}
770
df0a4283 771SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
d5460c99 772 struct mq_attr __user *, u_attr)
1da177e4 773{
765927b2 774 struct path path;
1da177e4 775 struct file *filp;
91a27b2a 776 struct filename *name;
564f6993 777 struct mq_attr attr;
1da177e4 778 int fd, error;
7eafd7c7 779 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
312b90fb
AV
780 struct vfsmount *mnt = ipc_ns->mq_mnt;
781 struct dentry *root = mnt->mnt_root;
782 int ro;
1da177e4 783
564f6993
AV
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);
20ca73bc 788
1da177e4
LT
789 if (IS_ERR(name = getname(u_name)))
790 return PTR_ERR(name);
791
269f2134 792 fd = get_unused_fd_flags(O_CLOEXEC);
1da177e4
LT
793 if (fd < 0)
794 goto out_putname;
795
312b90fb 796 ro = mnt_want_write(mnt); /* we'll drop it in any case */
765927b2 797 error = 0;
75c3cfa8 798 mutex_lock(&d_inode(root)->i_mutex);
91a27b2a 799 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
765927b2
AV
800 if (IS_ERR(path.dentry)) {
801 error = PTR_ERR(path.dentry);
4294a8ee 802 goto out_putfd;
1da177e4 803 }
312b90fb 804 path.mnt = mntget(mnt);
1da177e4
LT
805
806 if (oflag & O_CREAT) {
75c3cfa8 807 if (d_really_is_positive(path.dentry)) { /* entry already exists */
adb5c247 808 audit_inode(name, path.dentry, 0);
8d8ffefa
AGR
809 if (oflag & O_EXCL) {
810 error = -EEXIST;
7c7dce92 811 goto out;
8d8ffefa 812 }
765927b2 813 filp = do_open(&path, oflag);
1da177e4 814 } else {
312b90fb
AV
815 if (ro) {
816 error = ro;
817 goto out;
818 }
79f6530c 819 audit_inode_parent_hidden(name, root);
75c3cfa8 820 filp = do_create(ipc_ns, d_inode(root),
765927b2 821 &path, oflag, mode,
564f6993 822 u_attr ? &attr : NULL);
1da177e4 823 }
7c7dce92 824 } else {
75c3cfa8 825 if (d_really_is_negative(path.dentry)) {
8d8ffefa 826 error = -ENOENT;
7c7dce92 827 goto out;
8d8ffefa 828 }
adb5c247 829 audit_inode(name, path.dentry, 0);
765927b2 830 filp = do_open(&path, oflag);
7c7dce92 831 }
1da177e4 832
765927b2
AV
833 if (!IS_ERR(filp))
834 fd_install(fd, filp);
835 else
1da177e4 836 error = PTR_ERR(filp);
7c7dce92 837out:
765927b2 838 path_put(&path);
7c7dce92 839out_putfd:
765927b2
AV
840 if (error) {
841 put_unused_fd(fd);
842 fd = error;
843 }
75c3cfa8 844 mutex_unlock(&d_inode(root)->i_mutex);
38d78e58
VD
845 if (!ro)
846 mnt_drop_write(mnt);
1da177e4
LT
847out_putname:
848 putname(name);
849 return fd;
850}
851
d5460c99 852SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
1da177e4
LT
853{
854 int err;
91a27b2a 855 struct filename *name;
1da177e4
LT
856 struct dentry *dentry;
857 struct inode *inode = NULL;
7eafd7c7 858 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
312b90fb 859 struct vfsmount *mnt = ipc_ns->mq_mnt;
1da177e4
LT
860
861 name = getname(u_name);
862 if (IS_ERR(name))
863 return PTR_ERR(name);
864
79f6530c 865 audit_inode_parent_hidden(name, mnt->mnt_root);
312b90fb
AV
866 err = mnt_want_write(mnt);
867 if (err)
868 goto out_name;
75c3cfa8 869 mutex_lock_nested(&d_inode(mnt->mnt_root)->i_mutex, I_MUTEX_PARENT);
91a27b2a
JL
870 dentry = lookup_one_len(name->name, mnt->mnt_root,
871 strlen(name->name));
1da177e4
LT
872 if (IS_ERR(dentry)) {
873 err = PTR_ERR(dentry);
874 goto out_unlock;
875 }
876
75c3cfa8 877 inode = d_inode(dentry);
312b90fb
AV
878 if (!inode) {
879 err = -ENOENT;
880 } else {
7de9c6ee 881 ihold(inode);
75c3cfa8 882 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
312b90fb 883 }
1da177e4
LT
884 dput(dentry);
885
886out_unlock:
75c3cfa8 887 mutex_unlock(&d_inode(mnt->mnt_root)->i_mutex);
1da177e4
LT
888 if (inode)
889 iput(inode);
312b90fb
AV
890 mnt_drop_write(mnt);
891out_name:
892 putname(name);
1da177e4
LT
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
fa6004ad
DB
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.
1da177e4
LT
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 */
fa6004ad
DB
919static inline void pipelined_send(struct wake_q_head *wake_q,
920 struct mqueue_inode_info *info,
1da177e4
LT
921 struct msg_msg *message,
922 struct ext_wait_queue *receiver)
923{
924 receiver->msg = message;
925 list_del(&receiver->list);
fa6004ad
DB
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 */
1da177e4
LT
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). */
fa6004ad
DB
940static inline void pipelined_receive(struct wake_q_head *wake_q,
941 struct mqueue_inode_info *info)
1da177e4
LT
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 }
d6629859
DL
950 if (msg_insert(sender->msg, info))
951 return;
fa6004ad 952
1da177e4 953 list_del(&sender->list);
fa6004ad 954 wake_q_add(wake_q, sender->task);
1da177e4
LT
955 sender->state = STATE_READY;
956}
957
c4ea37c2
HC
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)
1da177e4 961{
2903ff01 962 struct fd f;
1da177e4
LT
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;
9ca7d8e6
CE
968 ktime_t expires, *timeout = NULL;
969 struct timespec ts;
ce2d52cc 970 struct posix_msg_tree_node *new_leaf = NULL;
2903ff01 971 int ret = 0;
fa6004ad 972 WAKE_Q(wake_q);
1da177e4 973
c32c8af4 974 if (u_abs_timeout) {
9ca7d8e6
CE
975 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
976 if (res)
977 return res;
978 timeout = &expires;
c32c8af4 979 }
20ca73bc 980
1da177e4
LT
981 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
982 return -EINVAL;
983
9ca7d8e6 984 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
1da177e4 985
2903ff01
AV
986 f = fdget(mqdes);
987 if (unlikely(!f.file)) {
8d8ffefa 988 ret = -EBADF;
1da177e4 989 goto out;
8d8ffefa 990 }
1da177e4 991
496ad9aa 992 inode = file_inode(f.file);
2903ff01 993 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 994 ret = -EBADF;
1da177e4 995 goto out_fput;
8d8ffefa 996 }
1da177e4 997 info = MQUEUE_I(inode);
9f45f5bf 998 audit_file(f.file);
1da177e4 999
2903ff01 1000 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
8d8ffefa 1001 ret = -EBADF;
1da177e4 1002 goto out_fput;
8d8ffefa 1003 }
1da177e4
LT
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
ce2d52cc
DL
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
1da177e4
LT
1028 spin_lock(&info->lock);
1029
ce2d52cc
DL
1030 if (!info->node_cache && new_leaf) {
1031 /* Save our speculative allocation into the cache */
ce2d52cc
DL
1032 INIT_LIST_HEAD(&new_leaf->msg_list);
1033 info->node_cache = new_leaf;
ce2d52cc
DL
1034 new_leaf = NULL;
1035 } else {
1036 kfree(new_leaf);
1037 }
1038
1da177e4 1039 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
2903ff01 1040 if (f.file->f_flags & O_NONBLOCK) {
1da177e4 1041 ret = -EAGAIN;
1da177e4
LT
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);
ce2d52cc
DL
1047 /*
1048 * wq_sleep must be called with info->lock held, and
1049 * returns with the lock released
1050 */
1051 goto out_free;
1da177e4 1052 }
1da177e4
LT
1053 } else {
1054 receiver = wq_get_first_waiter(info, RECV);
1055 if (receiver) {
fa6004ad 1056 pipelined_send(&wake_q, info, msg_ptr, receiver);
1da177e4
LT
1057 } else {
1058 /* adds message to the queue */
ce2d52cc
DL
1059 ret = msg_insert(msg_ptr, info);
1060 if (ret)
1061 goto out_unlock;
1da177e4
LT
1062 __do_notify(info);
1063 }
1064 inode->i_atime = inode->i_mtime = inode->i_ctime =
1065 CURRENT_TIME;
1da177e4 1066 }
ce2d52cc
DL
1067out_unlock:
1068 spin_unlock(&info->lock);
fa6004ad 1069 wake_up_q(&wake_q);
ce2d52cc
DL
1070out_free:
1071 if (ret)
1072 free_msg(msg_ptr);
1da177e4 1073out_fput:
2903ff01 1074 fdput(f);
1da177e4
LT
1075out:
1076 return ret;
1077}
1078
c4ea37c2
HC
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)
1da177e4 1082{
1da177e4
LT
1083 ssize_t ret;
1084 struct msg_msg *msg_ptr;
2903ff01 1085 struct fd f;
1da177e4
LT
1086 struct inode *inode;
1087 struct mqueue_inode_info *info;
1088 struct ext_wait_queue wait;
9ca7d8e6
CE
1089 ktime_t expires, *timeout = NULL;
1090 struct timespec ts;
ce2d52cc 1091 struct posix_msg_tree_node *new_leaf = NULL;
1da177e4 1092
c32c8af4 1093 if (u_abs_timeout) {
9ca7d8e6
CE
1094 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1095 if (res)
1096 return res;
1097 timeout = &expires;
c32c8af4 1098 }
20ca73bc 1099
9ca7d8e6 1100 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1da177e4 1101
2903ff01
AV
1102 f = fdget(mqdes);
1103 if (unlikely(!f.file)) {
8d8ffefa 1104 ret = -EBADF;
1da177e4 1105 goto out;
8d8ffefa 1106 }
1da177e4 1107
496ad9aa 1108 inode = file_inode(f.file);
2903ff01 1109 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1110 ret = -EBADF;
1da177e4 1111 goto out_fput;
8d8ffefa 1112 }
1da177e4 1113 info = MQUEUE_I(inode);
9f45f5bf 1114 audit_file(f.file);
1da177e4 1115
2903ff01 1116 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
8d8ffefa 1117 ret = -EBADF;
1da177e4 1118 goto out_fput;
8d8ffefa 1119 }
1da177e4
LT
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
ce2d52cc
DL
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
1da177e4 1135 spin_lock(&info->lock);
ce2d52cc
DL
1136
1137 if (!info->node_cache && new_leaf) {
1138 /* Save our speculative allocation into the cache */
ce2d52cc
DL
1139 INIT_LIST_HEAD(&new_leaf->msg_list);
1140 info->node_cache = new_leaf;
ce2d52cc
DL
1141 } else {
1142 kfree(new_leaf);
1143 }
1144
1da177e4 1145 if (info->attr.mq_curmsgs == 0) {
2903ff01 1146 if (f.file->f_flags & O_NONBLOCK) {
1da177e4
LT
1147 spin_unlock(&info->lock);
1148 ret = -EAGAIN;
1da177e4
LT
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 {
fa6004ad
DB
1156 WAKE_Q(wake_q);
1157
1da177e4
LT
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. */
fa6004ad 1164 pipelined_receive(&wake_q, info);
1da177e4 1165 spin_unlock(&info->lock);
fa6004ad 1166 wake_up_q(&wake_q);
1da177e4
LT
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:
2903ff01 1179 fdput(f);
1da177e4
LT
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 */
c4ea37c2
HC
1189SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1190 const struct sigevent __user *, u_notification)
1da177e4 1191{
2903ff01
AV
1192 int ret;
1193 struct fd f;
1da177e4
LT
1194 struct sock *sock;
1195 struct inode *inode;
1196 struct sigevent notification;
1197 struct mqueue_inode_info *info;
1198 struct sk_buff *nc;
1199
20114f71 1200 if (u_notification) {
1da177e4
LT
1201 if (copy_from_user(&notification, u_notification,
1202 sizeof(struct sigevent)))
1203 return -EFAULT;
20114f71
AV
1204 }
1205
1206 audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1da177e4 1207
20114f71
AV
1208 nc = NULL;
1209 sock = NULL;
1210 if (u_notification != NULL) {
1da177e4
LT
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 &&
7ed20e1a 1216 !valid_signal(notification.sigev_signo)) {
1da177e4
LT
1217 return -EINVAL;
1218 }
1219 if (notification.sigev_notify == SIGEV_THREAD) {
c3d8d1e3
PM
1220 long timeo;
1221
1da177e4
LT
1222 /* create the notify skb */
1223 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
8d8ffefa
AGR
1224 if (!nc) {
1225 ret = -ENOMEM;
1da177e4 1226 goto out;
8d8ffefa 1227 }
1da177e4
LT
1228 if (copy_from_user(nc->data,
1229 notification.sigev_value.sival_ptr,
1230 NOTIFY_COOKIE_LEN)) {
8d8ffefa 1231 ret = -EFAULT;
1da177e4
LT
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:
2903ff01
AV
1239 f = fdget(notification.sigev_signo);
1240 if (!f.file) {
8d8ffefa 1241 ret = -EBADF;
1da177e4 1242 goto out;
8d8ffefa 1243 }
2903ff01
AV
1244 sock = netlink_getsockbyfilp(f.file);
1245 fdput(f);
1da177e4
LT
1246 if (IS_ERR(sock)) {
1247 ret = PTR_ERR(sock);
1248 sock = NULL;
1249 goto out;
1250 }
1251
c3d8d1e3 1252 timeo = MAX_SCHEDULE_TIMEOUT;
9457afee 1253 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1da177e4 1254 if (ret == 1)
8d8ffefa 1255 goto retry;
1da177e4
LT
1256 if (ret) {
1257 sock = NULL;
1258 nc = NULL;
1259 goto out;
1260 }
1261 }
1262 }
1263
2903ff01
AV
1264 f = fdget(mqdes);
1265 if (!f.file) {
8d8ffefa 1266 ret = -EBADF;
1da177e4 1267 goto out;
8d8ffefa 1268 }
1da177e4 1269
496ad9aa 1270 inode = file_inode(f.file);
2903ff01 1271 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1272 ret = -EBADF;
1da177e4 1273 goto out_fput;
8d8ffefa 1274 }
1da177e4
LT
1275 info = MQUEUE_I(inode);
1276
1277 ret = 0;
1278 spin_lock(&info->lock);
1279 if (u_notification == NULL) {
a03fcb73 1280 if (info->notify_owner == task_tgid(current)) {
1da177e4
LT
1281 remove_notification(info);
1282 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1283 }
a03fcb73 1284 } else if (info->notify_owner != NULL) {
1da177e4
LT
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 }
a03fcb73
CLG
1304
1305 info->notify_owner = get_pid(task_tgid(current));
6f9ac6d9 1306 info->notify_user_ns = get_user_ns(current_user_ns());
1da177e4
LT
1307 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1308 }
1309 spin_unlock(&info->lock);
1310out_fput:
2903ff01 1311 fdput(f);
1da177e4 1312out:
3ab08fe2 1313 if (sock)
1da177e4 1314 netlink_detachskb(sock, nc);
3ab08fe2 1315 else if (nc)
1da177e4 1316 dev_kfree_skb(nc);
3ab08fe2 1317
1da177e4
LT
1318 return ret;
1319}
1320
c4ea37c2
HC
1321SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1322 const struct mq_attr __user *, u_mqstat,
1323 struct mq_attr __user *, u_omqstat)
1da177e4
LT
1324{
1325 int ret;
1326 struct mq_attr mqstat, omqstat;
2903ff01 1327 struct fd f;
1da177e4
LT
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
2903ff01
AV
1338 f = fdget(mqdes);
1339 if (!f.file) {
8d8ffefa 1340 ret = -EBADF;
1da177e4 1341 goto out;
8d8ffefa 1342 }
1da177e4 1343
496ad9aa 1344 inode = file_inode(f.file);
2903ff01 1345 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1346 ret = -EBADF;
1da177e4 1347 goto out_fput;
8d8ffefa 1348 }
1da177e4
LT
1349 info = MQUEUE_I(inode);
1350
1351 spin_lock(&info->lock);
1352
1353 omqstat = info->attr;
2903ff01 1354 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1da177e4 1355 if (u_mqstat) {
7392906e 1356 audit_mq_getsetattr(mqdes, &mqstat);
2903ff01 1357 spin_lock(&f.file->f_lock);
1da177e4 1358 if (mqstat.mq_flags & O_NONBLOCK)
2903ff01 1359 f.file->f_flags |= O_NONBLOCK;
1da177e4 1360 else
2903ff01
AV
1361 f.file->f_flags &= ~O_NONBLOCK;
1362 spin_unlock(&f.file->f_lock);
1da177e4
LT
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:
2903ff01 1375 fdput(f);
1da177e4
LT
1376out:
1377 return ret;
1378}
1379
92e1d5be 1380static const struct inode_operations mqueue_dir_inode_operations = {
1da177e4
LT
1381 .lookup = simple_lookup,
1382 .create = mqueue_create,
1383 .unlink = mqueue_unlink,
1384};
1385
9a32144e 1386static const struct file_operations mqueue_file_operations = {
1da177e4
LT
1387 .flush = mqueue_flush_file,
1388 .poll = mqueue_poll_file,
1389 .read = mqueue_read_file,
6038f373 1390 .llseek = default_llseek,
1da177e4
LT
1391};
1392
b87221de 1393static const struct super_operations mqueue_super_ops = {
1da177e4
LT
1394 .alloc_inode = mqueue_alloc_inode,
1395 .destroy_inode = mqueue_destroy_inode,
6d8af64c 1396 .evict_inode = mqueue_evict_inode,
1da177e4 1397 .statfs = simple_statfs,
1da177e4
LT
1398};
1399
1400static struct file_system_type mqueue_fs_type = {
1401 .name = "mqueue",
ceefda69 1402 .mount = mqueue_mount,
1da177e4 1403 .kill_sb = kill_litter_super,
bc1b69ed 1404 .fs_flags = FS_USERNS_MOUNT,
1da177e4
LT
1405};
1406
7eafd7c7
SH
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;
cef0184c
KM
1413 ns->mq_msg_default = DFLT_MSG;
1414 ns->mq_msgsize_default = DFLT_MSGSIZE;
7eafd7c7
SH
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{
6f686574 1432 kern_unmount(ns->mq_mnt);
7eafd7c7
SH
1433}
1434
1da177e4
LT
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,
20c2df83 1441 SLAB_HWCACHE_ALIGN, init_once);
1da177e4
LT
1442 if (mqueue_inode_cachep == NULL)
1443 return -ENOMEM;
1444
2329e392 1445 /* ignore failures - they are not fatal */
bdc8e5f8 1446 mq_sysctl_table = mq_register_sysctl_table();
1da177e4
LT
1447
1448 error = register_filesystem(&mqueue_fs_type);
1449 if (error)
1450 goto out_sysctl;
1451
7eafd7c7
SH
1452 spin_lock_init(&mq_lock);
1453
6f686574
AV
1454 error = mq_init_ns(&init_ipc_ns);
1455 if (error)
1da177e4 1456 goto out_filesystem;
1da177e4 1457
1da177e4
LT
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);
1a1d92c1 1465 kmem_cache_destroy(mqueue_inode_cachep);
1da177e4
LT
1466 return error;
1467}
1468
6d08a256 1469device_initcall(init_mqueue_fs);