ipc,msg: drop msg_unlock
[GitHub/mt8127/android_kernel_alcatel_ttab.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:
9 * Manfred Spraul (manfred@colorfullife.com)
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
50#define STATE_PENDING 1
51#define STATE_READY 2
52
d6629859
DL
53struct posix_msg_tree_node {
54 struct rb_node rb_node;
55 struct list_head msg_list;
56 int priority;
57};
58
1da177e4
LT
59struct ext_wait_queue { /* queue of sleeping tasks */
60 struct task_struct *task;
61 struct list_head list;
62 struct msg_msg *msg; /* ptr of loaded message */
63 int state; /* one of STATE_* values */
64};
65
66struct mqueue_inode_info {
67 spinlock_t lock;
68 struct inode vfs_inode;
69 wait_queue_head_t wait_q;
70
d6629859 71 struct rb_root msg_tree;
ce2d52cc 72 struct posix_msg_tree_node *node_cache;
1da177e4
LT
73 struct mq_attr attr;
74
75 struct sigevent notify;
a03fcb73 76 struct pid* notify_owner;
6f9ac6d9 77 struct user_namespace *notify_user_ns;
338cec32 78 struct user_struct *user; /* user who created, for accounting */
1da177e4
LT
79 struct sock *notify_sock;
80 struct sk_buff *notify_cookie;
81
82 /* for tasks waiting for free space and messages, respectively */
83 struct ext_wait_queue e_wait_q[2];
84
85 unsigned long qsize; /* size of queue in memory (sum of all msgs) */
86};
87
92e1d5be 88static const struct inode_operations mqueue_dir_inode_operations;
9a32144e 89static const struct file_operations mqueue_file_operations;
b87221de 90static const struct super_operations mqueue_super_ops;
1da177e4
LT
91static void remove_notification(struct mqueue_inode_info *info);
92
e18b890b 93static struct kmem_cache *mqueue_inode_cachep;
1da177e4
LT
94
95static struct ctl_table_header * mq_sysctl_table;
96
97static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
98{
99 return container_of(inode, struct mqueue_inode_info, vfs_inode);
100}
101
7eafd7c7
SH
102/*
103 * This routine should be called with the mq_lock held.
104 */
105static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
614b84cf 106{
7eafd7c7 107 return get_ipc_ns(inode->i_sb->s_fs_info);
614b84cf
SH
108}
109
7eafd7c7 110static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
614b84cf 111{
7eafd7c7
SH
112 struct ipc_namespace *ns;
113
114 spin_lock(&mq_lock);
115 ns = __get_ns_from_inode(inode);
116 spin_unlock(&mq_lock);
117 return ns;
614b84cf
SH
118}
119
d6629859
DL
120/* Auxiliary functions to manipulate messages' list */
121static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
122{
123 struct rb_node **p, *parent = NULL;
124 struct posix_msg_tree_node *leaf;
125
126 p = &info->msg_tree.rb_node;
127 while (*p) {
128 parent = *p;
129 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
130
131 if (likely(leaf->priority == msg->m_type))
132 goto insert_msg;
133 else if (msg->m_type < leaf->priority)
134 p = &(*p)->rb_left;
135 else
136 p = &(*p)->rb_right;
137 }
ce2d52cc
DL
138 if (info->node_cache) {
139 leaf = info->node_cache;
140 info->node_cache = NULL;
141 } else {
142 leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
143 if (!leaf)
144 return -ENOMEM;
ce2d52cc
DL
145 INIT_LIST_HEAD(&leaf->msg_list);
146 info->qsize += sizeof(*leaf);
147 }
d6629859
DL
148 leaf->priority = msg->m_type;
149 rb_link_node(&leaf->rb_node, parent, p);
150 rb_insert_color(&leaf->rb_node, &info->msg_tree);
d6629859
DL
151insert_msg:
152 info->attr.mq_curmsgs++;
153 info->qsize += msg->m_ts;
154 list_add_tail(&msg->m_list, &leaf->msg_list);
155 return 0;
156}
157
158static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
159{
160 struct rb_node **p, *parent = NULL;
161 struct posix_msg_tree_node *leaf;
162 struct msg_msg *msg;
163
164try_again:
165 p = &info->msg_tree.rb_node;
166 while (*p) {
167 parent = *p;
168 /*
169 * During insert, low priorities go to the left and high to the
170 * right. On receive, we want the highest priorities first, so
171 * walk all the way to the right.
172 */
173 p = &(*p)->rb_right;
174 }
175 if (!parent) {
176 if (info->attr.mq_curmsgs) {
177 pr_warn_once("Inconsistency in POSIX message queue, "
178 "no tree element, but supposedly messages "
179 "should exist!\n");
180 info->attr.mq_curmsgs = 0;
181 }
182 return NULL;
183 }
184 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
ce2d52cc 185 if (unlikely(list_empty(&leaf->msg_list))) {
d6629859
DL
186 pr_warn_once("Inconsistency in POSIX message queue, "
187 "empty leaf node but we haven't implemented "
188 "lazy leaf delete!\n");
189 rb_erase(&leaf->rb_node, &info->msg_tree);
ce2d52cc
DL
190 if (info->node_cache) {
191 info->qsize -= sizeof(*leaf);
192 kfree(leaf);
193 } else {
194 info->node_cache = leaf;
195 }
d6629859
DL
196 goto try_again;
197 } else {
198 msg = list_first_entry(&leaf->msg_list,
199 struct msg_msg, m_list);
200 list_del(&msg->m_list);
201 if (list_empty(&leaf->msg_list)) {
202 rb_erase(&leaf->rb_node, &info->msg_tree);
ce2d52cc
DL
203 if (info->node_cache) {
204 info->qsize -= sizeof(*leaf);
205 kfree(leaf);
206 } else {
207 info->node_cache = leaf;
208 }
d6629859
DL
209 }
210 }
211 info->attr.mq_curmsgs--;
212 info->qsize -= msg->m_ts;
213 return msg;
214}
215
7eafd7c7 216static struct inode *mqueue_get_inode(struct super_block *sb,
1b9d5ff7 217 struct ipc_namespace *ipc_ns, umode_t mode,
7eafd7c7 218 struct mq_attr *attr)
1da177e4 219{
86a264ab 220 struct user_struct *u = current_user();
1da177e4 221 struct inode *inode;
d40dcdb0 222 int ret = -ENOMEM;
1da177e4
LT
223
224 inode = new_inode(sb);
04715206
JS
225 if (!inode)
226 goto err;
227
228 inode->i_ino = get_next_ino();
229 inode->i_mode = mode;
230 inode->i_uid = current_fsuid();
231 inode->i_gid = current_fsgid();
232 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
233
234 if (S_ISREG(mode)) {
235 struct mqueue_inode_info *info;
d6629859 236 unsigned long mq_bytes, mq_treesize;
04715206
JS
237
238 inode->i_fop = &mqueue_file_operations;
239 inode->i_size = FILENT_SIZE;
240 /* mqueue specific info */
241 info = MQUEUE_I(inode);
242 spin_lock_init(&info->lock);
243 init_waitqueue_head(&info->wait_q);
244 INIT_LIST_HEAD(&info->e_wait_q[0].list);
245 INIT_LIST_HEAD(&info->e_wait_q[1].list);
246 info->notify_owner = NULL;
6f9ac6d9 247 info->notify_user_ns = NULL;
04715206
JS
248 info->qsize = 0;
249 info->user = NULL; /* set when all is ok */
d6629859 250 info->msg_tree = RB_ROOT;
ce2d52cc 251 info->node_cache = NULL;
04715206 252 memset(&info->attr, 0, sizeof(info->attr));
cef0184c
KM
253 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
254 ipc_ns->mq_msg_default);
255 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
256 ipc_ns->mq_msgsize_default);
04715206
JS
257 if (attr) {
258 info->attr.mq_maxmsg = attr->mq_maxmsg;
259 info->attr.mq_msgsize = attr->mq_msgsize;
260 }
d6629859
DL
261 /*
262 * We used to allocate a static array of pointers and account
263 * the size of that array as well as one msg_msg struct per
264 * possible message into the queue size. That's no longer
265 * accurate as the queue is now an rbtree and will grow and
266 * shrink depending on usage patterns. We can, however, still
267 * account one msg_msg struct per message, but the nodes are
268 * allocated depending on priority usage, and most programs
269 * only use one, or a handful, of priorities. However, since
270 * this is pinned memory, we need to assume worst case, so
271 * that means the min(mq_maxmsg, max_priorities) * struct
272 * posix_msg_tree_node.
273 */
274 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
275 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
276 sizeof(struct posix_msg_tree_node);
04715206 277
d6629859
DL
278 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
279 info->attr.mq_msgsize);
1da177e4 280
04715206
JS
281 spin_lock(&mq_lock);
282 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
2a4e64b8 283 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
04715206
JS
284 spin_unlock(&mq_lock);
285 /* mqueue_evict_inode() releases info->messages */
d40dcdb0 286 ret = -EMFILE;
04715206 287 goto out_inode;
1da177e4 288 }
04715206
JS
289 u->mq_bytes += mq_bytes;
290 spin_unlock(&mq_lock);
291
292 /* all is ok */
293 info->user = get_uid(u);
294 } else if (S_ISDIR(mode)) {
295 inc_nlink(inode);
296 /* Some things misbehave if size == 0 on a directory */
297 inode->i_size = 2 * DIRENT_SIZE;
298 inode->i_op = &mqueue_dir_inode_operations;
299 inode->i_fop = &simple_dir_operations;
1da177e4 300 }
04715206 301
1da177e4
LT
302 return inode;
303out_inode:
1da177e4 304 iput(inode);
04715206 305err:
d40dcdb0 306 return ERR_PTR(ret);
1da177e4
LT
307}
308
309static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
310{
311 struct inode *inode;
7eafd7c7 312 struct ipc_namespace *ns = data;
1da177e4
LT
313
314 sb->s_blocksize = PAGE_CACHE_SIZE;
315 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
316 sb->s_magic = MQUEUE_MAGIC;
317 sb->s_op = &mqueue_super_ops;
318
48fde701
AV
319 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
320 if (IS_ERR(inode))
321 return PTR_ERR(inode);
1da177e4 322
48fde701
AV
323 sb->s_root = d_make_root(inode);
324 if (!sb->s_root)
325 return -ENOMEM;
326 return 0;
1da177e4
LT
327}
328
ceefda69 329static struct dentry *mqueue_mount(struct file_system_type *fs_type,
454e2398 330 int flags, const char *dev_name,
ceefda69 331 void *data)
1da177e4 332{
a636b702
EB
333 if (!(flags & MS_KERNMOUNT)) {
334 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
335 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
336 * over the ipc namespace.
337 */
338 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
339 return ERR_PTR(-EPERM);
340
341 data = ns;
342 }
ceefda69 343 return mount_ns(fs_type, flags, data, mqueue_fill_super);
1da177e4
LT
344}
345
51cc5068 346static void init_once(void *foo)
1da177e4
LT
347{
348 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
349
a35afb83 350 inode_init_once(&p->vfs_inode);
1da177e4
LT
351}
352
353static struct inode *mqueue_alloc_inode(struct super_block *sb)
354{
355 struct mqueue_inode_info *ei;
356
e94b1766 357 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
1da177e4
LT
358 if (!ei)
359 return NULL;
360 return &ei->vfs_inode;
361}
362
fa0d7e3d 363static void mqueue_i_callback(struct rcu_head *head)
1da177e4 364{
fa0d7e3d 365 struct inode *inode = container_of(head, struct inode, i_rcu);
1da177e4
LT
366 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
367}
368
fa0d7e3d
NP
369static void mqueue_destroy_inode(struct inode *inode)
370{
371 call_rcu(&inode->i_rcu, mqueue_i_callback);
372}
373
6d8af64c 374static void mqueue_evict_inode(struct inode *inode)
1da177e4
LT
375{
376 struct mqueue_inode_info *info;
377 struct user_struct *user;
d6629859 378 unsigned long mq_bytes, mq_treesize;
7eafd7c7 379 struct ipc_namespace *ipc_ns;
d6629859 380 struct msg_msg *msg;
1da177e4 381
dbd5768f 382 clear_inode(inode);
6d8af64c
AV
383
384 if (S_ISDIR(inode->i_mode))
1da177e4 385 return;
6d8af64c 386
7eafd7c7 387 ipc_ns = get_ns_from_inode(inode);
1da177e4
LT
388 info = MQUEUE_I(inode);
389 spin_lock(&info->lock);
d6629859
DL
390 while ((msg = msg_get(info)) != NULL)
391 free_msg(msg);
ce2d52cc 392 kfree(info->node_cache);
1da177e4
LT
393 spin_unlock(&info->lock);
394
8834cf79 395 /* Total amount of bytes accounted for the mqueue */
d6629859
DL
396 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
397 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
398 sizeof(struct posix_msg_tree_node);
399
400 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
401 info->attr.mq_msgsize);
402
1da177e4
LT
403 user = info->user;
404 if (user) {
405 spin_lock(&mq_lock);
406 user->mq_bytes -= mq_bytes;
7eafd7c7
SH
407 /*
408 * get_ns_from_inode() ensures that the
409 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
410 * to which we now hold a reference, or it is NULL.
411 * We can't put it here under mq_lock, though.
412 */
413 if (ipc_ns)
414 ipc_ns->mq_queues_count--;
1da177e4
LT
415 spin_unlock(&mq_lock);
416 free_uid(user);
417 }
7eafd7c7
SH
418 if (ipc_ns)
419 put_ipc_ns(ipc_ns);
1da177e4
LT
420}
421
422static int mqueue_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 423 umode_t mode, bool excl)
1da177e4
LT
424{
425 struct inode *inode;
426 struct mq_attr *attr = dentry->d_fsdata;
427 int error;
7eafd7c7 428 struct ipc_namespace *ipc_ns;
1da177e4
LT
429
430 spin_lock(&mq_lock);
7eafd7c7
SH
431 ipc_ns = __get_ns_from_inode(dir);
432 if (!ipc_ns) {
433 error = -EACCES;
434 goto out_unlock;
435 }
02967ea0
DL
436 if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
437 (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
438 !capable(CAP_SYS_RESOURCE))) {
1da177e4 439 error = -ENOSPC;
614b84cf 440 goto out_unlock;
1da177e4 441 }
614b84cf 442 ipc_ns->mq_queues_count++;
1da177e4
LT
443 spin_unlock(&mq_lock);
444
7eafd7c7 445 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
d40dcdb0
JS
446 if (IS_ERR(inode)) {
447 error = PTR_ERR(inode);
1da177e4 448 spin_lock(&mq_lock);
614b84cf
SH
449 ipc_ns->mq_queues_count--;
450 goto out_unlock;
1da177e4
LT
451 }
452
7eafd7c7 453 put_ipc_ns(ipc_ns);
1da177e4
LT
454 dir->i_size += DIRENT_SIZE;
455 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
456
457 d_instantiate(dentry, inode);
458 dget(dentry);
459 return 0;
614b84cf 460out_unlock:
1da177e4 461 spin_unlock(&mq_lock);
7eafd7c7
SH
462 if (ipc_ns)
463 put_ipc_ns(ipc_ns);
1da177e4
LT
464 return error;
465}
466
467static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
468{
469 struct inode *inode = dentry->d_inode;
470
471 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
472 dir->i_size -= DIRENT_SIZE;
9a53c3a7 473 drop_nlink(inode);
1da177e4
LT
474 dput(dentry);
475 return 0;
476}
477
478/*
479* This is routine for system read from queue file.
480* To avoid mess with doing here some sort of mq_receive we allow
481* to read only queue size & notification info (the only values
482* that are interesting from user point of view and aren't accessible
483* through std routines)
484*/
485static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
f1a43f93 486 size_t count, loff_t *off)
1da177e4 487{
496ad9aa 488 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4 489 char buffer[FILENT_SIZE];
f1a43f93 490 ssize_t ret;
1da177e4
LT
491
492 spin_lock(&info->lock);
493 snprintf(buffer, sizeof(buffer),
494 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
495 info->qsize,
496 info->notify_owner ? info->notify.sigev_notify : 0,
497 (info->notify_owner &&
498 info->notify.sigev_notify == SIGEV_SIGNAL) ?
499 info->notify.sigev_signo : 0,
6c5f3e7b 500 pid_vnr(info->notify_owner));
1da177e4
LT
501 spin_unlock(&info->lock);
502 buffer[sizeof(buffer)-1] = '\0';
1da177e4 503
f1a43f93
AM
504 ret = simple_read_from_buffer(u_data, count, off, buffer,
505 strlen(buffer));
506 if (ret <= 0)
507 return ret;
1da177e4 508
496ad9aa 509 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
f1a43f93 510 return ret;
1da177e4
LT
511}
512
75e1fcc0 513static int mqueue_flush_file(struct file *filp, fl_owner_t id)
1da177e4 514{
496ad9aa 515 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4
LT
516
517 spin_lock(&info->lock);
a03fcb73 518 if (task_tgid(current) == info->notify_owner)
1da177e4
LT
519 remove_notification(info);
520
521 spin_unlock(&info->lock);
522 return 0;
523}
524
525static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
526{
496ad9aa 527 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
1da177e4
LT
528 int retval = 0;
529
530 poll_wait(filp, &info->wait_q, poll_tab);
531
532 spin_lock(&info->lock);
533 if (info->attr.mq_curmsgs)
534 retval = POLLIN | POLLRDNORM;
535
536 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
537 retval |= POLLOUT | POLLWRNORM;
538 spin_unlock(&info->lock);
539
540 return retval;
541}
542
543/* Adds current to info->e_wait_q[sr] before element with smaller prio */
544static void wq_add(struct mqueue_inode_info *info, int sr,
545 struct ext_wait_queue *ewp)
546{
547 struct ext_wait_queue *walk;
548
549 ewp->task = current;
550
551 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
552 if (walk->task->static_prio <= current->static_prio) {
553 list_add_tail(&ewp->list, &walk->list);
554 return;
555 }
556 }
557 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
558}
559
560/*
561 * Puts current task to sleep. Caller must hold queue lock. After return
562 * lock isn't held.
563 * sr: SEND or RECV
564 */
565static int wq_sleep(struct mqueue_inode_info *info, int sr,
9ca7d8e6 566 ktime_t *timeout, struct ext_wait_queue *ewp)
1da177e4
LT
567{
568 int retval;
569 signed long time;
570
571 wq_add(info, sr, ewp);
572
573 for (;;) {
574 set_current_state(TASK_INTERRUPTIBLE);
575
576 spin_unlock(&info->lock);
32ea845d
WG
577 time = schedule_hrtimeout_range_clock(timeout, 0,
578 HRTIMER_MODE_ABS, CLOCK_REALTIME);
1da177e4
LT
579
580 while (ewp->state == STATE_PENDING)
581 cpu_relax();
582
583 if (ewp->state == STATE_READY) {
584 retval = 0;
585 goto out;
586 }
587 spin_lock(&info->lock);
588 if (ewp->state == STATE_READY) {
589 retval = 0;
590 goto out_unlock;
591 }
592 if (signal_pending(current)) {
593 retval = -ERESTARTSYS;
594 break;
595 }
596 if (time == 0) {
597 retval = -ETIMEDOUT;
598 break;
599 }
600 }
601 list_del(&ewp->list);
602out_unlock:
603 spin_unlock(&info->lock);
604out:
605 return retval;
606}
607
608/*
609 * Returns waiting task that should be serviced first or NULL if none exists
610 */
611static struct ext_wait_queue *wq_get_first_waiter(
612 struct mqueue_inode_info *info, int sr)
613{
614 struct list_head *ptr;
615
616 ptr = info->e_wait_q[sr].list.prev;
617 if (ptr == &info->e_wait_q[sr].list)
618 return NULL;
619 return list_entry(ptr, struct ext_wait_queue, list);
620}
621
1da177e4
LT
622
623static inline void set_cookie(struct sk_buff *skb, char code)
624{
625 ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
626}
627
628/*
629 * The next function is only to split too long sys_mq_timedsend
630 */
631static void __do_notify(struct mqueue_inode_info *info)
632{
633 /* notification
634 * invoked when there is registered process and there isn't process
635 * waiting synchronously for message AND state of queue changed from
636 * empty to not empty. Here we are sure that no one is waiting
637 * synchronously. */
638 if (info->notify_owner &&
639 info->attr.mq_curmsgs == 1) {
640 struct siginfo sig_i;
641 switch (info->notify.sigev_notify) {
642 case SIGEV_NONE:
643 break;
644 case SIGEV_SIGNAL:
645 /* sends signal */
646
647 sig_i.si_signo = info->notify.sigev_signo;
648 sig_i.si_errno = 0;
649 sig_i.si_code = SI_MESGQ;
650 sig_i.si_value = info->notify.sigev_value;
6b550f94
SH
651 /* map current pid/uid into info->owner's namespaces */
652 rcu_read_lock();
a6684999
SB
653 sig_i.si_pid = task_tgid_nr_ns(current,
654 ns_of_pid(info->notify_owner));
76b6db01 655 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
6b550f94 656 rcu_read_unlock();
1da177e4 657
a03fcb73
CLG
658 kill_pid_info(info->notify.sigev_signo,
659 &sig_i, info->notify_owner);
1da177e4
LT
660 break;
661 case SIGEV_THREAD:
662 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
7ee015e0 663 netlink_sendskb(info->notify_sock, info->notify_cookie);
1da177e4
LT
664 break;
665 }
666 /* after notification unregisters process */
a03fcb73 667 put_pid(info->notify_owner);
6f9ac6d9 668 put_user_ns(info->notify_user_ns);
a03fcb73 669 info->notify_owner = NULL;
6f9ac6d9 670 info->notify_user_ns = NULL;
1da177e4
LT
671 }
672 wake_up(&info->wait_q);
673}
674
9ca7d8e6
CE
675static int prepare_timeout(const struct timespec __user *u_abs_timeout,
676 ktime_t *expires, struct timespec *ts)
1da177e4 677{
9ca7d8e6
CE
678 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
679 return -EFAULT;
680 if (!timespec_valid(ts))
681 return -EINVAL;
1da177e4 682
9ca7d8e6
CE
683 *expires = timespec_to_ktime(*ts);
684 return 0;
1da177e4
LT
685}
686
687static void remove_notification(struct mqueue_inode_info *info)
688{
a03fcb73 689 if (info->notify_owner != NULL &&
1da177e4
LT
690 info->notify.sigev_notify == SIGEV_THREAD) {
691 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
7ee015e0 692 netlink_sendskb(info->notify_sock, info->notify_cookie);
1da177e4 693 }
a03fcb73 694 put_pid(info->notify_owner);
6f9ac6d9 695 put_user_ns(info->notify_user_ns);
a03fcb73 696 info->notify_owner = NULL;
6f9ac6d9 697 info->notify_user_ns = NULL;
1da177e4
LT
698}
699
614b84cf 700static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
1da177e4 701{
2c12ea49
DL
702 int mq_treesize;
703 unsigned long total_size;
704
1da177e4 705 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
113289cc 706 return -EINVAL;
1da177e4 707 if (capable(CAP_SYS_RESOURCE)) {
02967ea0
DL
708 if (attr->mq_maxmsg > HARD_MSGMAX ||
709 attr->mq_msgsize > HARD_MSGSIZEMAX)
113289cc 710 return -EINVAL;
1da177e4 711 } else {
614b84cf
SH
712 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
713 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
113289cc 714 return -EINVAL;
1da177e4
LT
715 }
716 /* check for overflow */
717 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
113289cc 718 return -EOVERFLOW;
2c12ea49
DL
719 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
720 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
721 sizeof(struct posix_msg_tree_node);
722 total_size = attr->mq_maxmsg * attr->mq_msgsize;
723 if (total_size + mq_treesize < total_size)
113289cc
DL
724 return -EOVERFLOW;
725 return 0;
1da177e4
LT
726}
727
728/*
729 * Invoked when creating a new queue via sys_mq_open
730 */
765927b2
AV
731static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
732 struct path *path, int oflag, umode_t mode,
614b84cf 733 struct mq_attr *attr)
1da177e4 734{
745ca247 735 const struct cred *cred = current_cred();
1da177e4
LT
736 int ret;
737
564f6993 738 if (attr) {
113289cc
DL
739 ret = mq_attr_ok(ipc_ns, attr);
740 if (ret)
765927b2 741 return ERR_PTR(ret);
1da177e4 742 /* store for use during create */
765927b2 743 path->dentry->d_fsdata = attr;
113289cc
DL
744 } else {
745 struct mq_attr def_attr;
746
747 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
748 ipc_ns->mq_msg_default);
749 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
750 ipc_ns->mq_msgsize_default);
751 ret = mq_attr_ok(ipc_ns, &def_attr);
752 if (ret)
765927b2 753 return ERR_PTR(ret);
1da177e4
LT
754 }
755
ce3b0f8d 756 mode &= ~current_umask();
765927b2
AV
757 ret = vfs_create(dir, path->dentry, mode, true);
758 path->dentry->d_fsdata = NULL;
312b90fb
AV
759 if (ret)
760 return ERR_PTR(ret);
761 return dentry_open(path, oflag, cred);
1da177e4
LT
762}
763
764/* Opens existing queue */
765927b2 765static struct file *do_open(struct path *path, int oflag)
1da177e4 766{
745ca247
DH
767 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
768 MAY_READ | MAY_WRITE };
765927b2
AV
769 int acc;
770 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
771 return ERR_PTR(-EINVAL);
772 acc = oflag2acc[oflag & O_ACCMODE];
773 if (inode_permission(path->dentry->d_inode, acc))
774 return ERR_PTR(-EACCES);
775 return dentry_open(path, oflag, current_cred());
1da177e4
LT
776}
777
df0a4283 778SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
d5460c99 779 struct mq_attr __user *, u_attr)
1da177e4 780{
765927b2 781 struct path path;
1da177e4 782 struct file *filp;
91a27b2a 783 struct filename *name;
564f6993 784 struct mq_attr attr;
1da177e4 785 int fd, error;
7eafd7c7 786 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
312b90fb
AV
787 struct vfsmount *mnt = ipc_ns->mq_mnt;
788 struct dentry *root = mnt->mnt_root;
789 int ro;
1da177e4 790
564f6993
AV
791 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
792 return -EFAULT;
793
794 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
20ca73bc 795
1da177e4
LT
796 if (IS_ERR(name = getname(u_name)))
797 return PTR_ERR(name);
798
269f2134 799 fd = get_unused_fd_flags(O_CLOEXEC);
1da177e4
LT
800 if (fd < 0)
801 goto out_putname;
802
312b90fb 803 ro = mnt_want_write(mnt); /* we'll drop it in any case */
765927b2
AV
804 error = 0;
805 mutex_lock(&root->d_inode->i_mutex);
91a27b2a 806 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
765927b2
AV
807 if (IS_ERR(path.dentry)) {
808 error = PTR_ERR(path.dentry);
4294a8ee 809 goto out_putfd;
1da177e4 810 }
312b90fb 811 path.mnt = mntget(mnt);
1da177e4
LT
812
813 if (oflag & O_CREAT) {
765927b2 814 if (path.dentry->d_inode) { /* entry already exists */
adb5c247 815 audit_inode(name, path.dentry, 0);
8d8ffefa
AGR
816 if (oflag & O_EXCL) {
817 error = -EEXIST;
7c7dce92 818 goto out;
8d8ffefa 819 }
765927b2 820 filp = do_open(&path, oflag);
1da177e4 821 } else {
312b90fb
AV
822 if (ro) {
823 error = ro;
824 goto out;
825 }
765927b2
AV
826 filp = do_create(ipc_ns, root->d_inode,
827 &path, oflag, mode,
564f6993 828 u_attr ? &attr : NULL);
1da177e4 829 }
7c7dce92 830 } else {
765927b2 831 if (!path.dentry->d_inode) {
8d8ffefa 832 error = -ENOENT;
7c7dce92 833 goto out;
8d8ffefa 834 }
adb5c247 835 audit_inode(name, path.dentry, 0);
765927b2 836 filp = do_open(&path, oflag);
7c7dce92 837 }
1da177e4 838
765927b2
AV
839 if (!IS_ERR(filp))
840 fd_install(fd, filp);
841 else
1da177e4 842 error = PTR_ERR(filp);
7c7dce92 843out:
765927b2 844 path_put(&path);
7c7dce92 845out_putfd:
765927b2
AV
846 if (error) {
847 put_unused_fd(fd);
848 fd = error;
849 }
850 mutex_unlock(&root->d_inode->i_mutex);
38d78e58
VD
851 if (!ro)
852 mnt_drop_write(mnt);
1da177e4
LT
853out_putname:
854 putname(name);
855 return fd;
856}
857
d5460c99 858SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
1da177e4
LT
859{
860 int err;
91a27b2a 861 struct filename *name;
1da177e4
LT
862 struct dentry *dentry;
863 struct inode *inode = NULL;
7eafd7c7 864 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
312b90fb 865 struct vfsmount *mnt = ipc_ns->mq_mnt;
1da177e4
LT
866
867 name = getname(u_name);
868 if (IS_ERR(name))
869 return PTR_ERR(name);
870
312b90fb
AV
871 err = mnt_want_write(mnt);
872 if (err)
873 goto out_name;
874 mutex_lock_nested(&mnt->mnt_root->d_inode->i_mutex, I_MUTEX_PARENT);
91a27b2a
JL
875 dentry = lookup_one_len(name->name, mnt->mnt_root,
876 strlen(name->name));
1da177e4
LT
877 if (IS_ERR(dentry)) {
878 err = PTR_ERR(dentry);
879 goto out_unlock;
880 }
881
1da177e4 882 inode = dentry->d_inode;
312b90fb
AV
883 if (!inode) {
884 err = -ENOENT;
885 } else {
7de9c6ee 886 ihold(inode);
312b90fb
AV
887 err = vfs_unlink(dentry->d_parent->d_inode, dentry);
888 }
1da177e4
LT
889 dput(dentry);
890
891out_unlock:
312b90fb 892 mutex_unlock(&mnt->mnt_root->d_inode->i_mutex);
1da177e4
LT
893 if (inode)
894 iput(inode);
312b90fb
AV
895 mnt_drop_write(mnt);
896out_name:
897 putname(name);
1da177e4
LT
898
899 return err;
900}
901
902/* Pipelined send and receive functions.
903 *
904 * If a receiver finds no waiting message, then it registers itself in the
905 * list of waiting receivers. A sender checks that list before adding the new
906 * message into the message array. If there is a waiting receiver, then it
907 * bypasses the message array and directly hands the message over to the
908 * receiver.
909 * The receiver accepts the message and returns without grabbing the queue
910 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
911 * are necessary. The same algorithm is used for sysv semaphores, see
7b7a317c 912 * ipc/sem.c for more details.
1da177e4
LT
913 *
914 * The same algorithm is used for senders.
915 */
916
917/* pipelined_send() - send a message directly to the task waiting in
918 * sys_mq_timedreceive() (without inserting message into a queue).
919 */
920static inline void pipelined_send(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 receiver->state = STATE_PENDING;
927 wake_up_process(receiver->task);
d59dd462 928 smp_wmb();
1da177e4
LT
929 receiver->state = STATE_READY;
930}
931
932/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
933 * gets its message and put to the queue (we have one free place for sure). */
934static inline void pipelined_receive(struct mqueue_inode_info *info)
935{
936 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
937
938 if (!sender) {
939 /* for poll */
940 wake_up_interruptible(&info->wait_q);
941 return;
942 }
d6629859
DL
943 if (msg_insert(sender->msg, info))
944 return;
1da177e4
LT
945 list_del(&sender->list);
946 sender->state = STATE_PENDING;
947 wake_up_process(sender->task);
d59dd462 948 smp_wmb();
1da177e4
LT
949 sender->state = STATE_READY;
950}
951
c4ea37c2
HC
952SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
953 size_t, msg_len, unsigned int, msg_prio,
954 const struct timespec __user *, u_abs_timeout)
1da177e4 955{
2903ff01 956 struct fd f;
1da177e4
LT
957 struct inode *inode;
958 struct ext_wait_queue wait;
959 struct ext_wait_queue *receiver;
960 struct msg_msg *msg_ptr;
961 struct mqueue_inode_info *info;
9ca7d8e6
CE
962 ktime_t expires, *timeout = NULL;
963 struct timespec ts;
ce2d52cc 964 struct posix_msg_tree_node *new_leaf = NULL;
2903ff01 965 int ret = 0;
1da177e4 966
c32c8af4 967 if (u_abs_timeout) {
9ca7d8e6
CE
968 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
969 if (res)
970 return res;
971 timeout = &expires;
c32c8af4 972 }
20ca73bc 973
1da177e4
LT
974 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
975 return -EINVAL;
976
9ca7d8e6 977 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
1da177e4 978
2903ff01
AV
979 f = fdget(mqdes);
980 if (unlikely(!f.file)) {
8d8ffefa 981 ret = -EBADF;
1da177e4 982 goto out;
8d8ffefa 983 }
1da177e4 984
496ad9aa 985 inode = file_inode(f.file);
2903ff01 986 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 987 ret = -EBADF;
1da177e4 988 goto out_fput;
8d8ffefa 989 }
1da177e4 990 info = MQUEUE_I(inode);
bfcec708 991 audit_inode(NULL, f.file->f_path.dentry, 0);
1da177e4 992
2903ff01 993 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
8d8ffefa 994 ret = -EBADF;
1da177e4 995 goto out_fput;
8d8ffefa 996 }
1da177e4
LT
997
998 if (unlikely(msg_len > info->attr.mq_msgsize)) {
999 ret = -EMSGSIZE;
1000 goto out_fput;
1001 }
1002
1003 /* First try to allocate memory, before doing anything with
1004 * existing queues. */
1005 msg_ptr = load_msg(u_msg_ptr, msg_len);
1006 if (IS_ERR(msg_ptr)) {
1007 ret = PTR_ERR(msg_ptr);
1008 goto out_fput;
1009 }
1010 msg_ptr->m_ts = msg_len;
1011 msg_ptr->m_type = msg_prio;
1012
ce2d52cc
DL
1013 /*
1014 * msg_insert really wants us to have a valid, spare node struct so
1015 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1016 * fall back to that if necessary.
1017 */
1018 if (!info->node_cache)
1019 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1020
1da177e4
LT
1021 spin_lock(&info->lock);
1022
ce2d52cc
DL
1023 if (!info->node_cache && new_leaf) {
1024 /* Save our speculative allocation into the cache */
ce2d52cc
DL
1025 INIT_LIST_HEAD(&new_leaf->msg_list);
1026 info->node_cache = new_leaf;
1027 info->qsize += sizeof(*new_leaf);
1028 new_leaf = NULL;
1029 } else {
1030 kfree(new_leaf);
1031 }
1032
1da177e4 1033 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
2903ff01 1034 if (f.file->f_flags & O_NONBLOCK) {
1da177e4 1035 ret = -EAGAIN;
1da177e4
LT
1036 } else {
1037 wait.task = current;
1038 wait.msg = (void *) msg_ptr;
1039 wait.state = STATE_NONE;
1040 ret = wq_sleep(info, SEND, timeout, &wait);
ce2d52cc
DL
1041 /*
1042 * wq_sleep must be called with info->lock held, and
1043 * returns with the lock released
1044 */
1045 goto out_free;
1da177e4 1046 }
1da177e4
LT
1047 } else {
1048 receiver = wq_get_first_waiter(info, RECV);
1049 if (receiver) {
1050 pipelined_send(info, msg_ptr, receiver);
1051 } else {
1052 /* adds message to the queue */
ce2d52cc
DL
1053 ret = msg_insert(msg_ptr, info);
1054 if (ret)
1055 goto out_unlock;
1da177e4
LT
1056 __do_notify(info);
1057 }
1058 inode->i_atime = inode->i_mtime = inode->i_ctime =
1059 CURRENT_TIME;
1da177e4 1060 }
ce2d52cc
DL
1061out_unlock:
1062 spin_unlock(&info->lock);
1063out_free:
1064 if (ret)
1065 free_msg(msg_ptr);
1da177e4 1066out_fput:
2903ff01 1067 fdput(f);
1da177e4
LT
1068out:
1069 return ret;
1070}
1071
c4ea37c2
HC
1072SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1073 size_t, msg_len, unsigned int __user *, u_msg_prio,
1074 const struct timespec __user *, u_abs_timeout)
1da177e4 1075{
1da177e4
LT
1076 ssize_t ret;
1077 struct msg_msg *msg_ptr;
2903ff01 1078 struct fd f;
1da177e4
LT
1079 struct inode *inode;
1080 struct mqueue_inode_info *info;
1081 struct ext_wait_queue wait;
9ca7d8e6
CE
1082 ktime_t expires, *timeout = NULL;
1083 struct timespec ts;
ce2d52cc 1084 struct posix_msg_tree_node *new_leaf = NULL;
1da177e4 1085
c32c8af4 1086 if (u_abs_timeout) {
9ca7d8e6
CE
1087 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1088 if (res)
1089 return res;
1090 timeout = &expires;
c32c8af4 1091 }
20ca73bc 1092
9ca7d8e6 1093 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1da177e4 1094
2903ff01
AV
1095 f = fdget(mqdes);
1096 if (unlikely(!f.file)) {
8d8ffefa 1097 ret = -EBADF;
1da177e4 1098 goto out;
8d8ffefa 1099 }
1da177e4 1100
496ad9aa 1101 inode = file_inode(f.file);
2903ff01 1102 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1103 ret = -EBADF;
1da177e4 1104 goto out_fput;
8d8ffefa 1105 }
1da177e4 1106 info = MQUEUE_I(inode);
bfcec708 1107 audit_inode(NULL, f.file->f_path.dentry, 0);
1da177e4 1108
2903ff01 1109 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
8d8ffefa 1110 ret = -EBADF;
1da177e4 1111 goto out_fput;
8d8ffefa 1112 }
1da177e4
LT
1113
1114 /* checks if buffer is big enough */
1115 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1116 ret = -EMSGSIZE;
1117 goto out_fput;
1118 }
1119
ce2d52cc
DL
1120 /*
1121 * msg_insert really wants us to have a valid, spare node struct so
1122 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1123 * fall back to that if necessary.
1124 */
1125 if (!info->node_cache)
1126 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1127
1da177e4 1128 spin_lock(&info->lock);
ce2d52cc
DL
1129
1130 if (!info->node_cache && new_leaf) {
1131 /* Save our speculative allocation into the cache */
ce2d52cc
DL
1132 INIT_LIST_HEAD(&new_leaf->msg_list);
1133 info->node_cache = new_leaf;
1134 info->qsize += sizeof(*new_leaf);
1135 } else {
1136 kfree(new_leaf);
1137 }
1138
1da177e4 1139 if (info->attr.mq_curmsgs == 0) {
2903ff01 1140 if (f.file->f_flags & O_NONBLOCK) {
1da177e4
LT
1141 spin_unlock(&info->lock);
1142 ret = -EAGAIN;
1da177e4
LT
1143 } else {
1144 wait.task = current;
1145 wait.state = STATE_NONE;
1146 ret = wq_sleep(info, RECV, timeout, &wait);
1147 msg_ptr = wait.msg;
1148 }
1149 } else {
1150 msg_ptr = msg_get(info);
1151
1152 inode->i_atime = inode->i_mtime = inode->i_ctime =
1153 CURRENT_TIME;
1154
1155 /* There is now free space in queue. */
1156 pipelined_receive(info);
1157 spin_unlock(&info->lock);
1158 ret = 0;
1159 }
1160 if (ret == 0) {
1161 ret = msg_ptr->m_ts;
1162
1163 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1164 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1165 ret = -EFAULT;
1166 }
1167 free_msg(msg_ptr);
1168 }
1169out_fput:
2903ff01 1170 fdput(f);
1da177e4
LT
1171out:
1172 return ret;
1173}
1174
1175/*
1176 * Notes: the case when user wants us to deregister (with NULL as pointer)
1177 * and he isn't currently owner of notification, will be silently discarded.
1178 * It isn't explicitly defined in the POSIX.
1179 */
c4ea37c2
HC
1180SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1181 const struct sigevent __user *, u_notification)
1da177e4 1182{
2903ff01
AV
1183 int ret;
1184 struct fd f;
1da177e4
LT
1185 struct sock *sock;
1186 struct inode *inode;
1187 struct sigevent notification;
1188 struct mqueue_inode_info *info;
1189 struct sk_buff *nc;
1190
20114f71 1191 if (u_notification) {
1da177e4
LT
1192 if (copy_from_user(&notification, u_notification,
1193 sizeof(struct sigevent)))
1194 return -EFAULT;
20114f71
AV
1195 }
1196
1197 audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1da177e4 1198
20114f71
AV
1199 nc = NULL;
1200 sock = NULL;
1201 if (u_notification != NULL) {
1da177e4
LT
1202 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1203 notification.sigev_notify != SIGEV_SIGNAL &&
1204 notification.sigev_notify != SIGEV_THREAD))
1205 return -EINVAL;
1206 if (notification.sigev_notify == SIGEV_SIGNAL &&
7ed20e1a 1207 !valid_signal(notification.sigev_signo)) {
1da177e4
LT
1208 return -EINVAL;
1209 }
1210 if (notification.sigev_notify == SIGEV_THREAD) {
c3d8d1e3
PM
1211 long timeo;
1212
1da177e4
LT
1213 /* create the notify skb */
1214 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
8d8ffefa
AGR
1215 if (!nc) {
1216 ret = -ENOMEM;
1da177e4 1217 goto out;
8d8ffefa 1218 }
1da177e4
LT
1219 if (copy_from_user(nc->data,
1220 notification.sigev_value.sival_ptr,
1221 NOTIFY_COOKIE_LEN)) {
8d8ffefa 1222 ret = -EFAULT;
1da177e4
LT
1223 goto out;
1224 }
1225
1226 /* TODO: add a header? */
1227 skb_put(nc, NOTIFY_COOKIE_LEN);
1228 /* and attach it to the socket */
1229retry:
2903ff01
AV
1230 f = fdget(notification.sigev_signo);
1231 if (!f.file) {
8d8ffefa 1232 ret = -EBADF;
1da177e4 1233 goto out;
8d8ffefa 1234 }
2903ff01
AV
1235 sock = netlink_getsockbyfilp(f.file);
1236 fdput(f);
1da177e4
LT
1237 if (IS_ERR(sock)) {
1238 ret = PTR_ERR(sock);
1239 sock = NULL;
1240 goto out;
1241 }
1242
c3d8d1e3 1243 timeo = MAX_SCHEDULE_TIMEOUT;
9457afee 1244 ret = netlink_attachskb(sock, nc, &timeo, NULL);
1da177e4 1245 if (ret == 1)
8d8ffefa 1246 goto retry;
1da177e4
LT
1247 if (ret) {
1248 sock = NULL;
1249 nc = NULL;
1250 goto out;
1251 }
1252 }
1253 }
1254
2903ff01
AV
1255 f = fdget(mqdes);
1256 if (!f.file) {
8d8ffefa 1257 ret = -EBADF;
1da177e4 1258 goto out;
8d8ffefa 1259 }
1da177e4 1260
496ad9aa 1261 inode = file_inode(f.file);
2903ff01 1262 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1263 ret = -EBADF;
1da177e4 1264 goto out_fput;
8d8ffefa 1265 }
1da177e4
LT
1266 info = MQUEUE_I(inode);
1267
1268 ret = 0;
1269 spin_lock(&info->lock);
1270 if (u_notification == NULL) {
a03fcb73 1271 if (info->notify_owner == task_tgid(current)) {
1da177e4
LT
1272 remove_notification(info);
1273 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1274 }
a03fcb73 1275 } else if (info->notify_owner != NULL) {
1da177e4
LT
1276 ret = -EBUSY;
1277 } else {
1278 switch (notification.sigev_notify) {
1279 case SIGEV_NONE:
1280 info->notify.sigev_notify = SIGEV_NONE;
1281 break;
1282 case SIGEV_THREAD:
1283 info->notify_sock = sock;
1284 info->notify_cookie = nc;
1285 sock = NULL;
1286 nc = NULL;
1287 info->notify.sigev_notify = SIGEV_THREAD;
1288 break;
1289 case SIGEV_SIGNAL:
1290 info->notify.sigev_signo = notification.sigev_signo;
1291 info->notify.sigev_value = notification.sigev_value;
1292 info->notify.sigev_notify = SIGEV_SIGNAL;
1293 break;
1294 }
a03fcb73
CLG
1295
1296 info->notify_owner = get_pid(task_tgid(current));
6f9ac6d9 1297 info->notify_user_ns = get_user_ns(current_user_ns());
1da177e4
LT
1298 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1299 }
1300 spin_unlock(&info->lock);
1301out_fput:
2903ff01 1302 fdput(f);
1da177e4
LT
1303out:
1304 if (sock) {
1305 netlink_detachskb(sock, nc);
1306 } else if (nc) {
1307 dev_kfree_skb(nc);
1308 }
1309 return ret;
1310}
1311
c4ea37c2
HC
1312SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1313 const struct mq_attr __user *, u_mqstat,
1314 struct mq_attr __user *, u_omqstat)
1da177e4
LT
1315{
1316 int ret;
1317 struct mq_attr mqstat, omqstat;
2903ff01 1318 struct fd f;
1da177e4
LT
1319 struct inode *inode;
1320 struct mqueue_inode_info *info;
1321
1322 if (u_mqstat != NULL) {
1323 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1324 return -EFAULT;
1325 if (mqstat.mq_flags & (~O_NONBLOCK))
1326 return -EINVAL;
1327 }
1328
2903ff01
AV
1329 f = fdget(mqdes);
1330 if (!f.file) {
8d8ffefa 1331 ret = -EBADF;
1da177e4 1332 goto out;
8d8ffefa 1333 }
1da177e4 1334
496ad9aa 1335 inode = file_inode(f.file);
2903ff01 1336 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
8d8ffefa 1337 ret = -EBADF;
1da177e4 1338 goto out_fput;
8d8ffefa 1339 }
1da177e4
LT
1340 info = MQUEUE_I(inode);
1341
1342 spin_lock(&info->lock);
1343
1344 omqstat = info->attr;
2903ff01 1345 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1da177e4 1346 if (u_mqstat) {
7392906e 1347 audit_mq_getsetattr(mqdes, &mqstat);
2903ff01 1348 spin_lock(&f.file->f_lock);
1da177e4 1349 if (mqstat.mq_flags & O_NONBLOCK)
2903ff01 1350 f.file->f_flags |= O_NONBLOCK;
1da177e4 1351 else
2903ff01
AV
1352 f.file->f_flags &= ~O_NONBLOCK;
1353 spin_unlock(&f.file->f_lock);
1da177e4
LT
1354
1355 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1356 }
1357
1358 spin_unlock(&info->lock);
1359
1360 ret = 0;
1361 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1362 sizeof(struct mq_attr)))
1363 ret = -EFAULT;
1364
1365out_fput:
2903ff01 1366 fdput(f);
1da177e4
LT
1367out:
1368 return ret;
1369}
1370
92e1d5be 1371static const struct inode_operations mqueue_dir_inode_operations = {
1da177e4
LT
1372 .lookup = simple_lookup,
1373 .create = mqueue_create,
1374 .unlink = mqueue_unlink,
1375};
1376
9a32144e 1377static const struct file_operations mqueue_file_operations = {
1da177e4
LT
1378 .flush = mqueue_flush_file,
1379 .poll = mqueue_poll_file,
1380 .read = mqueue_read_file,
6038f373 1381 .llseek = default_llseek,
1da177e4
LT
1382};
1383
b87221de 1384static const struct super_operations mqueue_super_ops = {
1da177e4
LT
1385 .alloc_inode = mqueue_alloc_inode,
1386 .destroy_inode = mqueue_destroy_inode,
6d8af64c 1387 .evict_inode = mqueue_evict_inode,
1da177e4 1388 .statfs = simple_statfs,
1da177e4
LT
1389};
1390
1391static struct file_system_type mqueue_fs_type = {
1392 .name = "mqueue",
ceefda69 1393 .mount = mqueue_mount,
1da177e4 1394 .kill_sb = kill_litter_super,
bc1b69ed 1395 .fs_flags = FS_USERNS_MOUNT,
1da177e4
LT
1396};
1397
7eafd7c7
SH
1398int mq_init_ns(struct ipc_namespace *ns)
1399{
1400 ns->mq_queues_count = 0;
1401 ns->mq_queues_max = DFLT_QUEUESMAX;
1402 ns->mq_msg_max = DFLT_MSGMAX;
1403 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
cef0184c
KM
1404 ns->mq_msg_default = DFLT_MSG;
1405 ns->mq_msgsize_default = DFLT_MSGSIZE;
7eafd7c7
SH
1406
1407 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1408 if (IS_ERR(ns->mq_mnt)) {
1409 int err = PTR_ERR(ns->mq_mnt);
1410 ns->mq_mnt = NULL;
1411 return err;
1412 }
1413 return 0;
1414}
1415
1416void mq_clear_sbinfo(struct ipc_namespace *ns)
1417{
1418 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1419}
1420
1421void mq_put_mnt(struct ipc_namespace *ns)
1422{
6f686574 1423 kern_unmount(ns->mq_mnt);
7eafd7c7
SH
1424}
1425
1da177e4
LT
1426static int __init init_mqueue_fs(void)
1427{
1428 int error;
1429
1430 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1431 sizeof(struct mqueue_inode_info), 0,
20c2df83 1432 SLAB_HWCACHE_ALIGN, init_once);
1da177e4
LT
1433 if (mqueue_inode_cachep == NULL)
1434 return -ENOMEM;
1435
2329e392 1436 /* ignore failures - they are not fatal */
bdc8e5f8 1437 mq_sysctl_table = mq_register_sysctl_table();
1da177e4
LT
1438
1439 error = register_filesystem(&mqueue_fs_type);
1440 if (error)
1441 goto out_sysctl;
1442
7eafd7c7
SH
1443 spin_lock_init(&mq_lock);
1444
6f686574
AV
1445 error = mq_init_ns(&init_ipc_ns);
1446 if (error)
1da177e4 1447 goto out_filesystem;
1da177e4 1448
1da177e4
LT
1449 return 0;
1450
1451out_filesystem:
1452 unregister_filesystem(&mqueue_fs_type);
1453out_sysctl:
1454 if (mq_sysctl_table)
1455 unregister_sysctl_table(mq_sysctl_table);
1a1d92c1 1456 kmem_cache_destroy(mqueue_inode_cachep);
1da177e4
LT
1457 return error;
1458}
1459
1460__initcall(init_mqueue_fs);