blob: a24ba9fe5bb8892dfaa7452fe78f9ef68d1d97fc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * POSIX message queues filesystem for Linux.
3 *
4 * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl)
Michal Wronskif66e9282006-10-03 23:23:27 +02005 * Michal Wronski (michal.wronski@gmail.com)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com)
8 * Lockless receive & send, fd based notify:
Manfred Spraul239521f2014-01-27 17:07:04 -08009 * Manfred Spraul (manfred@colorfullife.com)
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 *
George C. Wilson20ca73b2006-05-24 16:09:55 -050011 * Audit: George Wilson (ltcgcw@us.ibm.com)
12 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * This file is released under the GPL.
14 */
15
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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>
Doug Ledford5b5c4d12012-05-31 16:26:30 -070027#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/netlink.h>
29#include <linux/syscalls.h>
George C. Wilson20ca73b2006-05-24 16:09:55 -050030#include <linux/audit.h>
Jesper Juhl7ed20e12005-05-01 08:59:14 -070031#include <linux/signal.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080032#include <linux/mutex.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070033#include <linux/nsproxy.h>
34#include <linux/pid.h>
Serge E. Hallyn614b84c2009-04-06 19:01:08 -070035#include <linux/ipc_namespace.h>
Serge E. Hallyn6b550f92012-01-10 15:11:37 -080036#include <linux/user_namespace.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/slab.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080038
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#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
Davidlohr Buesofa6004a2015-05-04 07:02:46 -070050#define STATE_READY 1
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Doug Ledfordd6629852012-05-31 16:26:35 -070052struct posix_msg_tree_node {
53 struct rb_node rb_node;
54 struct list_head msg_list;
55 int priority;
56};
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058struct 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
Doug Ledfordd6629852012-05-31 16:26:35 -070070 struct rb_root msg_tree;
Doug Ledfordce2d52c2012-05-31 16:26:38 -070071 struct posix_msg_tree_node *node_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 struct mq_attr attr;
73
74 struct sigevent notify;
Manfred Spraul239521f2014-01-27 17:07:04 -080075 struct pid *notify_owner;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -080076 struct user_namespace *notify_user_ns;
Adrian Bunk338cec32005-09-10 00:26:54 -070077 struct user_struct *user; /* user who created, for accounting */
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 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
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080087static const struct inode_operations mqueue_dir_inode_operations;
Arjan van de Ven9a321442007-02-12 00:55:35 -080088static const struct file_operations mqueue_file_operations;
Alexey Dobriyanb87221d2009-09-21 17:01:09 -070089static const struct super_operations mqueue_super_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090static void remove_notification(struct mqueue_inode_info *info);
91
Christoph Lametere18b8902006-12-06 20:33:20 -080092static struct kmem_cache *mqueue_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Manfred Spraul239521f2014-01-27 17:07:04 -080094static struct ctl_table_header *mq_sysctl_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
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
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700101/*
102 * This routine should be called with the mq_lock held.
103 */
104static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700105{
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700106 return get_ipc_ns(inode->i_sb->s_fs_info);
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700107}
108
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700109static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700110{
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700111 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;
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700117}
118
Doug Ledfordd6629852012-05-31 16:26:35 -0700119/* 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 }
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700137 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;
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700144 INIT_LIST_HEAD(&leaf->msg_list);
145 info->qsize += sizeof(*leaf);
146 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700147 leaf->priority = msg->m_type;
148 rb_link_node(&leaf->rb_node, parent, p);
149 rb_insert_color(&leaf->rb_node, &info->msg_tree);
Doug Ledfordd6629852012-05-31 16:26:35 -0700150insert_msg:
151 info->attr.mq_curmsgs++;
152 info->qsize += msg->m_ts;
153 list_add_tail(&msg->m_list, &leaf->msg_list);
154 return 0;
155}
156
157static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
158{
159 struct rb_node **p, *parent = NULL;
160 struct posix_msg_tree_node *leaf;
161 struct msg_msg *msg;
162
163try_again:
164 p = &info->msg_tree.rb_node;
165 while (*p) {
166 parent = *p;
167 /*
168 * During insert, low priorities go to the left and high to the
169 * right. On receive, we want the highest priorities first, so
170 * walk all the way to the right.
171 */
172 p = &(*p)->rb_right;
173 }
174 if (!parent) {
175 if (info->attr.mq_curmsgs) {
176 pr_warn_once("Inconsistency in POSIX message queue, "
177 "no tree element, but supposedly messages "
178 "should exist!\n");
179 info->attr.mq_curmsgs = 0;
180 }
181 return NULL;
182 }
183 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700184 if (unlikely(list_empty(&leaf->msg_list))) {
Doug Ledfordd6629852012-05-31 16:26:35 -0700185 pr_warn_once("Inconsistency in POSIX message queue, "
186 "empty leaf node but we haven't implemented "
187 "lazy leaf delete!\n");
188 rb_erase(&leaf->rb_node, &info->msg_tree);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700189 if (info->node_cache) {
190 info->qsize -= sizeof(*leaf);
191 kfree(leaf);
192 } else {
193 info->node_cache = leaf;
194 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700195 goto try_again;
196 } else {
197 msg = list_first_entry(&leaf->msg_list,
198 struct msg_msg, m_list);
199 list_del(&msg->m_list);
200 if (list_empty(&leaf->msg_list)) {
201 rb_erase(&leaf->rb_node, &info->msg_tree);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700202 if (info->node_cache) {
203 info->qsize -= sizeof(*leaf);
204 kfree(leaf);
205 } else {
206 info->node_cache = leaf;
207 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700208 }
209 }
210 info->attr.mq_curmsgs--;
211 info->qsize -= msg->m_ts;
212 return msg;
213}
214
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700215static struct inode *mqueue_get_inode(struct super_block *sb,
Al Viro1b9d5ff72011-07-24 14:18:20 -0400216 struct ipc_namespace *ipc_ns, umode_t mode,
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700217 struct mq_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
David Howells86a264a2008-11-14 10:39:18 +1100219 struct user_struct *u = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 struct inode *inode;
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700221 int ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223 inode = new_inode(sb);
Jiri Slaby04715202011-07-26 16:08:46 -0700224 if (!inode)
225 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Jiri Slaby04715202011-07-26 16:08:46 -0700227 inode->i_ino = get_next_ino();
228 inode->i_mode = mode;
229 inode->i_uid = current_fsuid();
230 inode->i_gid = current_fsgid();
231 inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Jiri Slaby04715202011-07-26 16:08:46 -0700233 if (S_ISREG(mode)) {
234 struct mqueue_inode_info *info;
Doug Ledfordd6629852012-05-31 16:26:35 -0700235 unsigned long mq_bytes, mq_treesize;
André Goddard Rosac8308b12010-02-23 04:04:23 -0300236
Jiri Slaby04715202011-07-26 16:08:46 -0700237 inode->i_fop = &mqueue_file_operations;
238 inode->i_size = FILENT_SIZE;
239 /* mqueue specific info */
240 info = MQUEUE_I(inode);
241 spin_lock_init(&info->lock);
242 init_waitqueue_head(&info->wait_q);
243 INIT_LIST_HEAD(&info->e_wait_q[0].list);
244 INIT_LIST_HEAD(&info->e_wait_q[1].list);
245 info->notify_owner = NULL;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800246 info->notify_user_ns = NULL;
Jiri Slaby04715202011-07-26 16:08:46 -0700247 info->qsize = 0;
248 info->user = NULL; /* set when all is ok */
Doug Ledfordd6629852012-05-31 16:26:35 -0700249 info->msg_tree = RB_ROOT;
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700250 info->node_cache = NULL;
Jiri Slaby04715202011-07-26 16:08:46 -0700251 memset(&info->attr, 0, sizeof(info->attr));
KOSAKI Motohirocef01842012-05-31 16:26:33 -0700252 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
253 ipc_ns->mq_msg_default);
254 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
255 ipc_ns->mq_msgsize_default);
Jiri Slaby04715202011-07-26 16:08:46 -0700256 if (attr) {
257 info->attr.mq_maxmsg = attr->mq_maxmsg;
258 info->attr.mq_msgsize = attr->mq_msgsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700260 /*
261 * We used to allocate a static array of pointers and account
262 * the size of that array as well as one msg_msg struct per
263 * possible message into the queue size. That's no longer
264 * accurate as the queue is now an rbtree and will grow and
265 * shrink depending on usage patterns. We can, however, still
266 * account one msg_msg struct per message, but the nodes are
267 * allocated depending on priority usage, and most programs
268 * only use one, or a handful, of priorities. However, since
269 * this is pinned memory, we need to assume worst case, so
270 * that means the min(mq_maxmsg, max_priorities) * struct
271 * posix_msg_tree_node.
272 */
273 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
274 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
275 sizeof(struct posix_msg_tree_node);
Jiri Slaby04715202011-07-26 16:08:46 -0700276
Doug Ledfordd6629852012-05-31 16:26:35 -0700277 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
278 info->attr.mq_msgsize);
Jiri Slaby04715202011-07-26 16:08:46 -0700279
280 spin_lock(&mq_lock);
281 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
Davidlohr Bueso2a4e64b2012-01-20 14:34:01 -0800282 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
Jiri Slaby04715202011-07-26 16:08:46 -0700283 spin_unlock(&mq_lock);
284 /* mqueue_evict_inode() releases info->messages */
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700285 ret = -EMFILE;
Jiri Slaby04715202011-07-26 16:08:46 -0700286 goto out_inode;
287 }
288 u->mq_bytes += mq_bytes;
289 spin_unlock(&mq_lock);
290
291 /* all is ok */
292 info->user = get_uid(u);
293 } else if (S_ISDIR(mode)) {
294 inc_nlink(inode);
295 /* Some things misbehave if size == 0 on a directory */
296 inode->i_size = 2 * DIRENT_SIZE;
297 inode->i_op = &mqueue_dir_inode_operations;
298 inode->i_fop = &simple_dir_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 }
Jiri Slaby04715202011-07-26 16:08:46 -0700300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 return inode;
302out_inode:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 iput(inode);
Jiri Slaby04715202011-07-26 16:08:46 -0700304err:
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700305 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
308static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
309{
310 struct inode *inode;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700311 struct ipc_namespace *ns = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313 sb->s_blocksize = PAGE_CACHE_SIZE;
314 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
315 sb->s_magic = MQUEUE_MAGIC;
316 sb->s_op = &mqueue_super_ops;
317
Al Viro48fde702012-01-08 22:15:13 -0500318 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
319 if (IS_ERR(inode))
320 return PTR_ERR(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Al Viro48fde702012-01-08 22:15:13 -0500322 sb->s_root = d_make_root(inode);
323 if (!sb->s_root)
324 return -ENOMEM;
325 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
Al Viroceefda62010-07-26 13:16:50 +0400328static struct dentry *mqueue_mount(struct file_system_type *fs_type,
David Howells454e2392006-06-23 02:02:57 -0700329 int flags, const char *dev_name,
Al Viroceefda62010-07-26 13:16:50 +0400330 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Eric W. Biedermana636b702013-03-21 18:13:15 -0700332 if (!(flags & MS_KERNMOUNT)) {
333 struct ipc_namespace *ns = current->nsproxy->ipc_ns;
334 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
335 * over the ipc namespace.
336 */
337 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN))
338 return ERR_PTR(-EPERM);
339
340 data = ns;
341 }
Al Viroceefda62010-07-26 13:16:50 +0400342 return mount_ns(fs_type, flags, data, mqueue_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
344
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700345static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
347 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
348
Christoph Lametera35afb82007-05-16 22:10:57 -0700349 inode_init_once(&p->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
352static struct inode *mqueue_alloc_inode(struct super_block *sb)
353{
354 struct mqueue_inode_info *ei;
355
Christoph Lametere94b1762006-12-06 20:33:17 -0800356 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 if (!ei)
358 return NULL;
359 return &ei->vfs_inode;
360}
361
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100362static void mqueue_i_callback(struct rcu_head *head)
363{
364 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100365 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368static void mqueue_destroy_inode(struct inode *inode)
369{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100370 call_rcu(&inode->i_rcu, mqueue_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Al Viro6d8af642010-06-05 16:29:45 -0400373static void mqueue_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
375 struct mqueue_inode_info *info;
376 struct user_struct *user;
Doug Ledfordd6629852012-05-31 16:26:35 -0700377 unsigned long mq_bytes, mq_treesize;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700378 struct ipc_namespace *ipc_ns;
Doug Ledfordd6629852012-05-31 16:26:35 -0700379 struct msg_msg *msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Jan Karadbd57682012-05-03 14:48:02 +0200381 clear_inode(inode);
Al Viro6d8af642010-06-05 16:29:45 -0400382
383 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return;
Al Viro6d8af642010-06-05 16:29:45 -0400385
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700386 ipc_ns = get_ns_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 info = MQUEUE_I(inode);
388 spin_lock(&info->lock);
Doug Ledfordd6629852012-05-31 16:26:35 -0700389 while ((msg = msg_get(info)) != NULL)
390 free_msg(msg);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700391 kfree(info->node_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 spin_unlock(&info->lock);
393
André Goddard Rosa8834cf72010-02-23 04:04:24 -0300394 /* Total amount of bytes accounted for the mqueue */
Doug Ledfordd6629852012-05-31 16:26:35 -0700395 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
396 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
397 sizeof(struct posix_msg_tree_node);
398
399 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
400 info->attr.mq_msgsize);
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 user = info->user;
403 if (user) {
404 spin_lock(&mq_lock);
405 user->mq_bytes -= mq_bytes;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700406 /*
407 * get_ns_from_inode() ensures that the
408 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
409 * to which we now hold a reference, or it is NULL.
410 * We can't put it here under mq_lock, though.
411 */
412 if (ipc_ns)
413 ipc_ns->mq_queues_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 spin_unlock(&mq_lock);
415 free_uid(user);
416 }
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700417 if (ipc_ns)
418 put_ipc_ns(ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419}
420
421static int mqueue_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -0400422 umode_t mode, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
424 struct inode *inode;
425 struct mq_attr *attr = dentry->d_fsdata;
426 int error;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700427 struct ipc_namespace *ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 spin_lock(&mq_lock);
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700430 ipc_ns = __get_ns_from_inode(dir);
431 if (!ipc_ns) {
432 error = -EACCES;
433 goto out_unlock;
434 }
Davidlohr Buesof3713fd2014-02-25 15:01:45 -0800435
436 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
437 !capable(CAP_SYS_RESOURCE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 error = -ENOSPC;
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700439 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 }
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700441 ipc_ns->mq_queues_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 spin_unlock(&mq_lock);
443
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700444 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700445 if (IS_ERR(inode)) {
446 error = PTR_ERR(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 spin_lock(&mq_lock);
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700448 ipc_ns->mq_queues_count--;
449 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 }
451
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700452 put_ipc_ns(ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 dir->i_size += DIRENT_SIZE;
454 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
455
456 d_instantiate(dentry, inode);
457 dget(dentry);
458 return 0;
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700459out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 spin_unlock(&mq_lock);
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700461 if (ipc_ns)
462 put_ipc_ns(ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return error;
464}
465
466static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
467{
David Howells75c3cfa2015-03-17 22:26:12 +0000468 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
471 dir->i_size -= DIRENT_SIZE;
Manfred Spraul239521f2014-01-27 17:07:04 -0800472 drop_nlink(inode);
473 dput(dentry);
474 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477/*
478* This is routine for system read from queue file.
479* To avoid mess with doing here some sort of mq_receive we allow
480* to read only queue size & notification info (the only values
481* that are interesting from user point of view and aren't accessible
482* through std routines)
483*/
484static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700485 size_t count, loff_t *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486{
Al Viro496ad9a2013-01-23 17:07:38 -0500487 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 char buffer[FILENT_SIZE];
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700489 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 spin_lock(&info->lock);
492 snprintf(buffer, sizeof(buffer),
493 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
494 info->qsize,
495 info->notify_owner ? info->notify.sigev_notify : 0,
496 (info->notify_owner &&
497 info->notify.sigev_notify == SIGEV_SIGNAL) ?
498 info->notify.sigev_signo : 0,
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800499 pid_vnr(info->notify_owner));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 spin_unlock(&info->lock);
501 buffer[sizeof(buffer)-1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700503 ret = simple_read_from_buffer(u_data, count, off, buffer,
504 strlen(buffer));
505 if (ret <= 0)
506 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Al Viro496ad9a2013-01-23 17:07:38 -0500508 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = CURRENT_TIME;
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700509 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700512static int mqueue_flush_file(struct file *filp, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
Al Viro496ad9a2013-01-23 17:07:38 -0500514 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
516 spin_lock(&info->lock);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700517 if (task_tgid(current) == info->notify_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 remove_notification(info);
519
520 spin_unlock(&info->lock);
521 return 0;
522}
523
524static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
525{
Al Viro496ad9a2013-01-23 17:07:38 -0500526 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 int retval = 0;
528
529 poll_wait(filp, &info->wait_q, poll_tab);
530
531 spin_lock(&info->lock);
532 if (info->attr.mq_curmsgs)
533 retval = POLLIN | POLLRDNORM;
534
535 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
536 retval |= POLLOUT | POLLWRNORM;
537 spin_unlock(&info->lock);
538
539 return retval;
540}
541
542/* Adds current to info->e_wait_q[sr] before element with smaller prio */
543static void wq_add(struct mqueue_inode_info *info, int sr,
544 struct ext_wait_queue *ewp)
545{
546 struct ext_wait_queue *walk;
547
548 ewp->task = current;
549
550 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
551 if (walk->task->static_prio <= current->static_prio) {
552 list_add_tail(&ewp->list, &walk->list);
553 return;
554 }
555 }
556 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
557}
558
559/*
560 * Puts current task to sleep. Caller must hold queue lock. After return
561 * lock isn't held.
562 * sr: SEND or RECV
563 */
564static int wq_sleep(struct mqueue_inode_info *info, int sr,
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200565 ktime_t *timeout, struct ext_wait_queue *ewp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
567 int retval;
568 signed long time;
569
570 wq_add(info, sr, ewp);
571
572 for (;;) {
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700573 __set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 spin_unlock(&info->lock);
Wanlong Gao32ea8452011-10-31 17:06:35 -0700576 time = schedule_hrtimeout_range_clock(timeout, 0,
577 HRTIMER_MODE_ABS, CLOCK_REALTIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (ewp->state == STATE_READY) {
580 retval = 0;
581 goto out;
582 }
583 spin_lock(&info->lock);
584 if (ewp->state == STATE_READY) {
585 retval = 0;
586 goto out_unlock;
587 }
588 if (signal_pending(current)) {
589 retval = -ERESTARTSYS;
590 break;
591 }
592 if (time == 0) {
593 retval = -ETIMEDOUT;
594 break;
595 }
596 }
597 list_del(&ewp->list);
598out_unlock:
599 spin_unlock(&info->lock);
600out:
601 return retval;
602}
603
604/*
605 * Returns waiting task that should be serviced first or NULL if none exists
606 */
607static struct ext_wait_queue *wq_get_first_waiter(
608 struct mqueue_inode_info *info, int sr)
609{
610 struct list_head *ptr;
611
612 ptr = info->e_wait_q[sr].list.prev;
613 if (ptr == &info->e_wait_q[sr].list)
614 return NULL;
615 return list_entry(ptr, struct ext_wait_queue, list);
616}
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619static inline void set_cookie(struct sk_buff *skb, char code)
620{
Manfred Spraul239521f2014-01-27 17:07:04 -0800621 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
624/*
625 * The next function is only to split too long sys_mq_timedsend
626 */
627static void __do_notify(struct mqueue_inode_info *info)
628{
629 /* notification
630 * invoked when there is registered process and there isn't process
631 * waiting synchronously for message AND state of queue changed from
632 * empty to not empty. Here we are sure that no one is waiting
633 * synchronously. */
634 if (info->notify_owner &&
635 info->attr.mq_curmsgs == 1) {
636 struct siginfo sig_i;
637 switch (info->notify.sigev_notify) {
638 case SIGEV_NONE:
639 break;
640 case SIGEV_SIGNAL:
641 /* sends signal */
642
643 sig_i.si_signo = info->notify.sigev_signo;
644 sig_i.si_errno = 0;
645 sig_i.si_code = SI_MESGQ;
646 sig_i.si_value = info->notify.sigev_value;
Serge E. Hallyn6b550f92012-01-10 15:11:37 -0800647 /* map current pid/uid into info->owner's namespaces */
648 rcu_read_lock();
Sukadev Bhattiprolua6684992009-01-07 18:08:50 -0800649 sig_i.si_pid = task_tgid_nr_ns(current,
650 ns_of_pid(info->notify_owner));
Eric W. Biederman76b6db02012-03-14 15:24:19 -0700651 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
Serge E. Hallyn6b550f92012-01-10 15:11:37 -0800652 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700654 kill_pid_info(info->notify.sigev_signo,
655 &sig_i, info->notify_owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 break;
657 case SIGEV_THREAD:
658 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
Denis V. Lunev7ee015e2007-10-10 21:14:03 -0700659 netlink_sendskb(info->notify_sock, info->notify_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 break;
661 }
662 /* after notification unregisters process */
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700663 put_pid(info->notify_owner);
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800664 put_user_ns(info->notify_user_ns);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700665 info->notify_owner = NULL;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800666 info->notify_user_ns = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 }
668 wake_up(&info->wait_q);
669}
670
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200671static int prepare_timeout(const struct timespec __user *u_abs_timeout,
672 ktime_t *expires, struct timespec *ts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200674 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
675 return -EFAULT;
676 if (!timespec_valid(ts))
677 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200679 *expires = timespec_to_ktime(*ts);
680 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
683static void remove_notification(struct mqueue_inode_info *info)
684{
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700685 if (info->notify_owner != NULL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 info->notify.sigev_notify == SIGEV_THREAD) {
687 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
Denis V. Lunev7ee015e2007-10-10 21:14:03 -0700688 netlink_sendskb(info->notify_sock, info->notify_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700690 put_pid(info->notify_owner);
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800691 put_user_ns(info->notify_user_ns);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700692 info->notify_owner = NULL;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800693 info->notify_user_ns = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700696static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
Doug Ledford2c12ea42012-05-31 16:26:36 -0700698 int mq_treesize;
699 unsigned long total_size;
700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
Doug Ledford113289c2012-05-31 16:26:36 -0700702 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 if (capable(CAP_SYS_RESOURCE)) {
Doug Ledford02967ea2012-05-31 16:26:29 -0700704 if (attr->mq_maxmsg > HARD_MSGMAX ||
705 attr->mq_msgsize > HARD_MSGSIZEMAX)
Doug Ledford113289c2012-05-31 16:26:36 -0700706 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 } else {
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700708 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
709 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
Doug Ledford113289c2012-05-31 16:26:36 -0700710 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
712 /* check for overflow */
713 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
Doug Ledford113289c2012-05-31 16:26:36 -0700714 return -EOVERFLOW;
Doug Ledford2c12ea42012-05-31 16:26:36 -0700715 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
716 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
717 sizeof(struct posix_msg_tree_node);
718 total_size = attr->mq_maxmsg * attr->mq_msgsize;
719 if (total_size + mq_treesize < total_size)
Doug Ledford113289c2012-05-31 16:26:36 -0700720 return -EOVERFLOW;
721 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
724/*
725 * Invoked when creating a new queue via sys_mq_open
726 */
Al Viro765927b2012-06-26 21:58:53 +0400727static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
728 struct path *path, int oflag, umode_t mode,
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700729 struct mq_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
David Howells745ca242008-11-14 10:39:22 +1100731 const struct cred *cred = current_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 int ret;
733
Al Viro564f6992008-12-14 04:02:26 -0500734 if (attr) {
Doug Ledford113289c2012-05-31 16:26:36 -0700735 ret = mq_attr_ok(ipc_ns, attr);
736 if (ret)
Al Viro765927b2012-06-26 21:58:53 +0400737 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 /* store for use during create */
Al Viro765927b2012-06-26 21:58:53 +0400739 path->dentry->d_fsdata = attr;
Doug Ledford113289c2012-05-31 16:26:36 -0700740 } else {
741 struct mq_attr def_attr;
742
743 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
744 ipc_ns->mq_msg_default);
745 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
746 ipc_ns->mq_msgsize_default);
747 ret = mq_attr_ok(ipc_ns, &def_attr);
748 if (ret)
Al Viro765927b2012-06-26 21:58:53 +0400749 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751
Al Viroce3b0f82009-03-29 19:08:22 -0400752 mode &= ~current_umask();
Al Viro765927b2012-06-26 21:58:53 +0400753 ret = vfs_create(dir, path->dentry, mode, true);
754 path->dentry->d_fsdata = NULL;
Al Viro312b90f2012-08-06 10:18:17 +0400755 if (ret)
756 return ERR_PTR(ret);
757 return dentry_open(path, oflag, cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
759
760/* Opens existing queue */
Al Viro765927b2012-06-26 21:58:53 +0400761static struct file *do_open(struct path *path, int oflag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
David Howells745ca242008-11-14 10:39:22 +1100763 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
764 MAY_READ | MAY_WRITE };
Al Viro765927b2012-06-26 21:58:53 +0400765 int acc;
766 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
767 return ERR_PTR(-EINVAL);
768 acc = oflag2acc[oflag & O_ACCMODE];
David Howells75c3cfa2015-03-17 22:26:12 +0000769 if (inode_permission(d_inode(path->dentry), acc))
Al Viro765927b2012-06-26 21:58:53 +0400770 return ERR_PTR(-EACCES);
771 return dentry_open(path, oflag, current_cred());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
Al Virodf0a4282011-07-26 05:26:10 -0400774SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
Heiko Carstensd5460c92009-01-14 14:14:27 +0100775 struct mq_attr __user *, u_attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
Al Viro765927b2012-06-26 21:58:53 +0400777 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 struct file *filp;
Jeff Layton91a27b22012-10-10 15:25:28 -0400779 struct filename *name;
Al Viro564f6992008-12-14 04:02:26 -0500780 struct mq_attr attr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 int fd, error;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700782 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
Al Viro312b90f2012-08-06 10:18:17 +0400783 struct vfsmount *mnt = ipc_ns->mq_mnt;
784 struct dentry *root = mnt->mnt_root;
785 int ro;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Al Viro564f6992008-12-14 04:02:26 -0500787 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
788 return -EFAULT;
789
790 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
George C. Wilson20ca73b2006-05-24 16:09:55 -0500791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (IS_ERR(name = getname(u_name)))
793 return PTR_ERR(name);
794
Ulrich Drepper269f2132008-05-03 15:28:45 -0400795 fd = get_unused_fd_flags(O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 if (fd < 0)
797 goto out_putname;
798
Al Viro312b90f2012-08-06 10:18:17 +0400799 ro = mnt_want_write(mnt); /* we'll drop it in any case */
Al Viro765927b2012-06-26 21:58:53 +0400800 error = 0;
David Howells75c3cfa2015-03-17 22:26:12 +0000801 mutex_lock(&d_inode(root)->i_mutex);
Jeff Layton91a27b22012-10-10 15:25:28 -0400802 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
Al Viro765927b2012-06-26 21:58:53 +0400803 if (IS_ERR(path.dentry)) {
804 error = PTR_ERR(path.dentry);
André Goddard Rosa4294a8e2010-02-23 04:04:28 -0300805 goto out_putfd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
Al Viro312b90f2012-08-06 10:18:17 +0400807 path.mnt = mntget(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 if (oflag & O_CREAT) {
David Howells75c3cfa2015-03-17 22:26:12 +0000810 if (d_really_is_positive(path.dentry)) { /* entry already exists */
Jeff Laytonadb5c242012-10-10 16:43:13 -0400811 audit_inode(name, path.dentry, 0);
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300812 if (oflag & O_EXCL) {
813 error = -EEXIST;
Alexander Viro7c7dce92006-01-14 15:29:55 -0500814 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300815 }
Al Viro765927b2012-06-26 21:58:53 +0400816 filp = do_open(&path, oflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 } else {
Al Viro312b90f2012-08-06 10:18:17 +0400818 if (ro) {
819 error = ro;
820 goto out;
821 }
Jeff Layton79f65302013-07-08 15:59:36 -0700822 audit_inode_parent_hidden(name, root);
David Howells75c3cfa2015-03-17 22:26:12 +0000823 filp = do_create(ipc_ns, d_inode(root),
Al Viro765927b2012-06-26 21:58:53 +0400824 &path, oflag, mode,
Al Viro564f6992008-12-14 04:02:26 -0500825 u_attr ? &attr : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
Alexander Viro7c7dce92006-01-14 15:29:55 -0500827 } else {
David Howells75c3cfa2015-03-17 22:26:12 +0000828 if (d_really_is_negative(path.dentry)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300829 error = -ENOENT;
Alexander Viro7c7dce92006-01-14 15:29:55 -0500830 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300831 }
Jeff Laytonadb5c242012-10-10 16:43:13 -0400832 audit_inode(name, path.dentry, 0);
Al Viro765927b2012-06-26 21:58:53 +0400833 filp = do_open(&path, oflag);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Al Viro765927b2012-06-26 21:58:53 +0400836 if (!IS_ERR(filp))
837 fd_install(fd, filp);
838 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 error = PTR_ERR(filp);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500840out:
Al Viro765927b2012-06-26 21:58:53 +0400841 path_put(&path);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500842out_putfd:
Al Viro765927b2012-06-26 21:58:53 +0400843 if (error) {
844 put_unused_fd(fd);
845 fd = error;
846 }
David Howells75c3cfa2015-03-17 22:26:12 +0000847 mutex_unlock(&d_inode(root)->i_mutex);
Vladimir Davydov38d78e52013-03-22 15:04:51 -0700848 if (!ro)
849 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850out_putname:
851 putname(name);
852 return fd;
853}
854
Heiko Carstensd5460c92009-01-14 14:14:27 +0100855SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856{
857 int err;
Jeff Layton91a27b22012-10-10 15:25:28 -0400858 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 struct dentry *dentry;
860 struct inode *inode = NULL;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700861 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
Al Viro312b90f2012-08-06 10:18:17 +0400862 struct vfsmount *mnt = ipc_ns->mq_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 name = getname(u_name);
865 if (IS_ERR(name))
866 return PTR_ERR(name);
867
Jeff Layton79f65302013-07-08 15:59:36 -0700868 audit_inode_parent_hidden(name, mnt->mnt_root);
Al Viro312b90f2012-08-06 10:18:17 +0400869 err = mnt_want_write(mnt);
870 if (err)
871 goto out_name;
David Howells75c3cfa2015-03-17 22:26:12 +0000872 mutex_lock_nested(&d_inode(mnt->mnt_root)->i_mutex, I_MUTEX_PARENT);
Jeff Layton91a27b22012-10-10 15:25:28 -0400873 dentry = lookup_one_len(name->name, mnt->mnt_root,
874 strlen(name->name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (IS_ERR(dentry)) {
876 err = PTR_ERR(dentry);
877 goto out_unlock;
878 }
879
David Howells75c3cfa2015-03-17 22:26:12 +0000880 inode = d_inode(dentry);
Al Viro312b90f2012-08-06 10:18:17 +0400881 if (!inode) {
882 err = -ENOENT;
883 } else {
Al Viro7de9c6ee2010-10-23 11:11:40 -0400884 ihold(inode);
David Howells75c3cfa2015-03-17 22:26:12 +0000885 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
Al Viro312b90f2012-08-06 10:18:17 +0400886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 dput(dentry);
888
889out_unlock:
David Howells75c3cfa2015-03-17 22:26:12 +0000890 mutex_unlock(&d_inode(mnt->mnt_root)->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if (inode)
892 iput(inode);
Al Viro312b90f2012-08-06 10:18:17 +0400893 mnt_drop_write(mnt);
894out_name:
895 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 return err;
898}
899
900/* Pipelined send and receive functions.
901 *
902 * If a receiver finds no waiting message, then it registers itself in the
903 * list of waiting receivers. A sender checks that list before adding the new
904 * message into the message array. If there is a waiting receiver, then it
905 * bypasses the message array and directly hands the message over to the
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700906 * receiver. The receiver accepts the message and returns without grabbing the
907 * queue spinlock:
908 *
909 * - Set pointer to message.
910 * - Queue the receiver task for later wakeup (without the info->lock).
911 * - Update its state to STATE_READY. Now the receiver can continue.
912 * - Wake up the process after the lock is dropped. Should the process wake up
913 * before this wakeup (due to a timeout or a signal) it will either see
914 * STATE_READY and continue or acquire the lock to check the state again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 *
916 * The same algorithm is used for senders.
917 */
918
919/* pipelined_send() - send a message directly to the task waiting in
920 * sys_mq_timedreceive() (without inserting message into a queue).
921 */
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700922static inline void pipelined_send(struct wake_q_head *wake_q,
923 struct mqueue_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 struct msg_msg *message,
925 struct ext_wait_queue *receiver)
926{
927 receiver->msg = message;
928 list_del(&receiver->list);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700929 wake_q_add(wake_q, receiver->task);
930 /*
931 * Rely on the implicit cmpxchg barrier from wake_q_add such
932 * that we can ensure that updating receiver->state is the last
933 * write operation: As once set, the receiver can continue,
934 * and if we don't have the reference count from the wake_q,
935 * yet, at that point we can later have a use-after-free
936 * condition and bogus wakeup.
937 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 receiver->state = STATE_READY;
939}
940
941/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
942 * gets its message and put to the queue (we have one free place for sure). */
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700943static inline void pipelined_receive(struct wake_q_head *wake_q,
944 struct mqueue_inode_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945{
946 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
947
948 if (!sender) {
949 /* for poll */
950 wake_up_interruptible(&info->wait_q);
951 return;
952 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700953 if (msg_insert(sender->msg, info))
954 return;
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 list_del(&sender->list);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700957 wake_q_add(wake_q, sender->task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 sender->state = STATE_READY;
959}
960
Heiko Carstensc4ea37c2009-01-14 14:14:28 +0100961SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
962 size_t, msg_len, unsigned int, msg_prio,
963 const struct timespec __user *, u_abs_timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
Al Viro2903ff02012-08-28 12:52:22 -0400965 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 struct inode *inode;
967 struct ext_wait_queue wait;
968 struct ext_wait_queue *receiver;
969 struct msg_msg *msg_ptr;
970 struct mqueue_inode_info *info;
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200971 ktime_t expires, *timeout = NULL;
972 struct timespec ts;
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700973 struct posix_msg_tree_node *new_leaf = NULL;
Al Viro2903ff02012-08-28 12:52:22 -0400974 int ret = 0;
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700975 WAKE_Q(wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
Al Viroc32c8af2008-12-14 03:46:48 -0500977 if (u_abs_timeout) {
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200978 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
979 if (res)
980 return res;
981 timeout = &expires;
Al Viroc32c8af2008-12-14 03:46:48 -0500982 }
George C. Wilson20ca73b2006-05-24 16:09:55 -0500983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
985 return -EINVAL;
986
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200987 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Al Viro2903ff02012-08-28 12:52:22 -0400989 f = fdget(mqdes);
990 if (unlikely(!f.file)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300991 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Al Viro496ad9a2013-01-23 17:07:38 -0500995 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -0400996 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300997 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 info = MQUEUE_I(inode);
Al Viro9f45f5b2014-10-31 17:44:57 -04001001 audit_file(f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Al Viro2903ff02012-08-28 12:52:22 -04001003 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001004 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1009 ret = -EMSGSIZE;
1010 goto out_fput;
1011 }
1012
1013 /* First try to allocate memory, before doing anything with
1014 * existing queues. */
1015 msg_ptr = load_msg(u_msg_ptr, msg_len);
1016 if (IS_ERR(msg_ptr)) {
1017 ret = PTR_ERR(msg_ptr);
1018 goto out_fput;
1019 }
1020 msg_ptr->m_ts = msg_len;
1021 msg_ptr->m_type = msg_prio;
1022
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001023 /*
1024 * msg_insert really wants us to have a valid, spare node struct so
1025 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1026 * fall back to that if necessary.
1027 */
1028 if (!info->node_cache)
1029 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 spin_lock(&info->lock);
1032
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001033 if (!info->node_cache && new_leaf) {
1034 /* Save our speculative allocation into the cache */
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001035 INIT_LIST_HEAD(&new_leaf->msg_list);
1036 info->node_cache = new_leaf;
1037 info->qsize += sizeof(*new_leaf);
1038 new_leaf = NULL;
1039 } else {
1040 kfree(new_leaf);
1041 }
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
Al Viro2903ff02012-08-28 12:52:22 -04001044 if (f.file->f_flags & O_NONBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 } else {
1047 wait.task = current;
1048 wait.msg = (void *) msg_ptr;
1049 wait.state = STATE_NONE;
1050 ret = wq_sleep(info, SEND, timeout, &wait);
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001051 /*
1052 * wq_sleep must be called with info->lock held, and
1053 * returns with the lock released
1054 */
1055 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 } else {
1058 receiver = wq_get_first_waiter(info, RECV);
1059 if (receiver) {
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001060 pipelined_send(&wake_q, info, msg_ptr, receiver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 } else {
1062 /* adds message to the queue */
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001063 ret = msg_insert(msg_ptr, info);
1064 if (ret)
1065 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 __do_notify(info);
1067 }
1068 inode->i_atime = inode->i_mtime = inode->i_ctime =
1069 CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001071out_unlock:
1072 spin_unlock(&info->lock);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001073 wake_up_q(&wake_q);
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001074out_free:
1075 if (ret)
1076 free_msg(msg_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001078 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079out:
1080 return ret;
1081}
1082
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001083SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1084 size_t, msg_len, unsigned int __user *, u_msg_prio,
1085 const struct timespec __user *, u_abs_timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 ssize_t ret;
1088 struct msg_msg *msg_ptr;
Al Viro2903ff02012-08-28 12:52:22 -04001089 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 struct inode *inode;
1091 struct mqueue_inode_info *info;
1092 struct ext_wait_queue wait;
Carsten Emde9ca7d8e2010-04-02 22:40:20 +02001093 ktime_t expires, *timeout = NULL;
1094 struct timespec ts;
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001095 struct posix_msg_tree_node *new_leaf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Al Viroc32c8af2008-12-14 03:46:48 -05001097 if (u_abs_timeout) {
Carsten Emde9ca7d8e2010-04-02 22:40:20 +02001098 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1099 if (res)
1100 return res;
1101 timeout = &expires;
Al Viroc32c8af2008-12-14 03:46:48 -05001102 }
George C. Wilson20ca73b2006-05-24 16:09:55 -05001103
Carsten Emde9ca7d8e2010-04-02 22:40:20 +02001104 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Al Viro2903ff02012-08-28 12:52:22 -04001106 f = fdget(mqdes);
1107 if (unlikely(!f.file)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001108 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Al Viro496ad9a2013-01-23 17:07:38 -05001112 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -04001113 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001114 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 info = MQUEUE_I(inode);
Al Viro9f45f5b2014-10-31 17:44:57 -04001118 audit_file(f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Al Viro2903ff02012-08-28 12:52:22 -04001120 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001121 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 /* checks if buffer is big enough */
1126 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1127 ret = -EMSGSIZE;
1128 goto out_fput;
1129 }
1130
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001131 /*
1132 * msg_insert really wants us to have a valid, spare node struct so
1133 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1134 * fall back to that if necessary.
1135 */
1136 if (!info->node_cache)
1137 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 spin_lock(&info->lock);
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001140
1141 if (!info->node_cache && new_leaf) {
1142 /* Save our speculative allocation into the cache */
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001143 INIT_LIST_HEAD(&new_leaf->msg_list);
1144 info->node_cache = new_leaf;
1145 info->qsize += sizeof(*new_leaf);
1146 } else {
1147 kfree(new_leaf);
1148 }
1149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (info->attr.mq_curmsgs == 0) {
Al Viro2903ff02012-08-28 12:52:22 -04001151 if (f.file->f_flags & O_NONBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 spin_unlock(&info->lock);
1153 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 } else {
1155 wait.task = current;
1156 wait.state = STATE_NONE;
1157 ret = wq_sleep(info, RECV, timeout, &wait);
1158 msg_ptr = wait.msg;
1159 }
1160 } else {
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001161 WAKE_Q(wake_q);
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 msg_ptr = msg_get(info);
1164
1165 inode->i_atime = inode->i_mtime = inode->i_ctime =
1166 CURRENT_TIME;
1167
1168 /* There is now free space in queue. */
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001169 pipelined_receive(&wake_q, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 spin_unlock(&info->lock);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001171 wake_up_q(&wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 ret = 0;
1173 }
1174 if (ret == 0) {
1175 ret = msg_ptr->m_ts;
1176
1177 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1178 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1179 ret = -EFAULT;
1180 }
1181 free_msg(msg_ptr);
1182 }
1183out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001184 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185out:
1186 return ret;
1187}
1188
1189/*
1190 * Notes: the case when user wants us to deregister (with NULL as pointer)
1191 * and he isn't currently owner of notification, will be silently discarded.
1192 * It isn't explicitly defined in the POSIX.
1193 */
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001194SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1195 const struct sigevent __user *, u_notification)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196{
Al Viro2903ff02012-08-28 12:52:22 -04001197 int ret;
1198 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 struct sock *sock;
1200 struct inode *inode;
1201 struct sigevent notification;
1202 struct mqueue_inode_info *info;
1203 struct sk_buff *nc;
1204
Al Viro20114f72008-12-10 07:16:12 -05001205 if (u_notification) {
1206 if (copy_from_user(&notification, u_notification,
1207 sizeof(struct sigevent)))
1208 return -EFAULT;
1209 }
1210
1211 audit_mq_notify(mqdes, u_notification ? &notification : NULL);
George C. Wilson20ca73b2006-05-24 16:09:55 -05001212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 nc = NULL;
1214 sock = NULL;
1215 if (u_notification != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1217 notification.sigev_notify != SIGEV_SIGNAL &&
1218 notification.sigev_notify != SIGEV_THREAD))
1219 return -EINVAL;
1220 if (notification.sigev_notify == SIGEV_SIGNAL &&
Jesper Juhl7ed20e12005-05-01 08:59:14 -07001221 !valid_signal(notification.sigev_signo)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 return -EINVAL;
1223 }
1224 if (notification.sigev_notify == SIGEV_THREAD) {
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001225 long timeo;
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 /* create the notify skb */
1228 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001229 if (!nc) {
1230 ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 if (copy_from_user(nc->data,
1234 notification.sigev_value.sival_ptr,
1235 NOTIFY_COOKIE_LEN)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001236 ret = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 goto out;
1238 }
1239
1240 /* TODO: add a header? */
1241 skb_put(nc, NOTIFY_COOKIE_LEN);
1242 /* and attach it to the socket */
1243retry:
Al Viro2903ff02012-08-28 12:52:22 -04001244 f = fdget(notification.sigev_signo);
1245 if (!f.file) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001246 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001248 }
Al Viro2903ff02012-08-28 12:52:22 -04001249 sock = netlink_getsockbyfilp(f.file);
1250 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 if (IS_ERR(sock)) {
1252 ret = PTR_ERR(sock);
1253 sock = NULL;
1254 goto out;
1255 }
1256
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001257 timeo = MAX_SCHEDULE_TIMEOUT;
Denis V. Lunev9457afe2008-06-05 11:23:39 -07001258 ret = netlink_attachskb(sock, nc, &timeo, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (ret == 1)
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001260 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (ret) {
1262 sock = NULL;
1263 nc = NULL;
1264 goto out;
1265 }
1266 }
1267 }
1268
Al Viro2903ff02012-08-28 12:52:22 -04001269 f = fdget(mqdes);
1270 if (!f.file) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001271 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Al Viro496ad9a2013-01-23 17:07:38 -05001275 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -04001276 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001277 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 info = MQUEUE_I(inode);
1281
1282 ret = 0;
1283 spin_lock(&info->lock);
1284 if (u_notification == NULL) {
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001285 if (info->notify_owner == task_tgid(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 remove_notification(info);
1287 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1288 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001289 } else if (info->notify_owner != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 ret = -EBUSY;
1291 } else {
1292 switch (notification.sigev_notify) {
1293 case SIGEV_NONE:
1294 info->notify.sigev_notify = SIGEV_NONE;
1295 break;
1296 case SIGEV_THREAD:
1297 info->notify_sock = sock;
1298 info->notify_cookie = nc;
1299 sock = NULL;
1300 nc = NULL;
1301 info->notify.sigev_notify = SIGEV_THREAD;
1302 break;
1303 case SIGEV_SIGNAL:
1304 info->notify.sigev_signo = notification.sigev_signo;
1305 info->notify.sigev_value = notification.sigev_value;
1306 info->notify.sigev_notify = SIGEV_SIGNAL;
1307 break;
1308 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001309
1310 info->notify_owner = get_pid(task_tgid(current));
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -08001311 info->notify_user_ns = get_user_ns(current_user_ns());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1313 }
1314 spin_unlock(&info->lock);
1315out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001316 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317out:
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001318 if (sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 netlink_detachskb(sock, nc);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001320 else if (nc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 dev_kfree_skb(nc);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 return ret;
1324}
1325
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001326SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1327 const struct mq_attr __user *, u_mqstat,
1328 struct mq_attr __user *, u_omqstat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
1330 int ret;
1331 struct mq_attr mqstat, omqstat;
Al Viro2903ff02012-08-28 12:52:22 -04001332 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 struct inode *inode;
1334 struct mqueue_inode_info *info;
1335
1336 if (u_mqstat != NULL) {
1337 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1338 return -EFAULT;
1339 if (mqstat.mq_flags & (~O_NONBLOCK))
1340 return -EINVAL;
1341 }
1342
Al Viro2903ff02012-08-28 12:52:22 -04001343 f = fdget(mqdes);
1344 if (!f.file) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001345 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Al Viro496ad9a2013-01-23 17:07:38 -05001349 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -04001350 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001351 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 info = MQUEUE_I(inode);
1355
1356 spin_lock(&info->lock);
1357
1358 omqstat = info->attr;
Al Viro2903ff02012-08-28 12:52:22 -04001359 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 if (u_mqstat) {
Al Viro73929062008-12-10 06:58:59 -05001361 audit_mq_getsetattr(mqdes, &mqstat);
Al Viro2903ff02012-08-28 12:52:22 -04001362 spin_lock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (mqstat.mq_flags & O_NONBLOCK)
Al Viro2903ff02012-08-28 12:52:22 -04001364 f.file->f_flags |= O_NONBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 else
Al Viro2903ff02012-08-28 12:52:22 -04001366 f.file->f_flags &= ~O_NONBLOCK;
1367 spin_unlock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
1369 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1370 }
1371
1372 spin_unlock(&info->lock);
1373
1374 ret = 0;
1375 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1376 sizeof(struct mq_attr)))
1377 ret = -EFAULT;
1378
1379out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001380 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381out:
1382 return ret;
1383}
1384
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001385static const struct inode_operations mqueue_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 .lookup = simple_lookup,
1387 .create = mqueue_create,
1388 .unlink = mqueue_unlink,
1389};
1390
Arjan van de Ven9a321442007-02-12 00:55:35 -08001391static const struct file_operations mqueue_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 .flush = mqueue_flush_file,
1393 .poll = mqueue_poll_file,
1394 .read = mqueue_read_file,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001395 .llseek = default_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396};
1397
Alexey Dobriyanb87221d2009-09-21 17:01:09 -07001398static const struct super_operations mqueue_super_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 .alloc_inode = mqueue_alloc_inode,
1400 .destroy_inode = mqueue_destroy_inode,
Al Viro6d8af642010-06-05 16:29:45 -04001401 .evict_inode = mqueue_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 .statfs = simple_statfs,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403};
1404
1405static struct file_system_type mqueue_fs_type = {
1406 .name = "mqueue",
Al Viroceefda62010-07-26 13:16:50 +04001407 .mount = mqueue_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 .kill_sb = kill_litter_super,
Gao fengbc1b69e2013-01-28 11:09:01 +08001409 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410};
1411
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001412int mq_init_ns(struct ipc_namespace *ns)
1413{
1414 ns->mq_queues_count = 0;
1415 ns->mq_queues_max = DFLT_QUEUESMAX;
1416 ns->mq_msg_max = DFLT_MSGMAX;
1417 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
KOSAKI Motohirocef01842012-05-31 16:26:33 -07001418 ns->mq_msg_default = DFLT_MSG;
1419 ns->mq_msgsize_default = DFLT_MSGSIZE;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001420
1421 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1422 if (IS_ERR(ns->mq_mnt)) {
1423 int err = PTR_ERR(ns->mq_mnt);
1424 ns->mq_mnt = NULL;
1425 return err;
1426 }
1427 return 0;
1428}
1429
1430void mq_clear_sbinfo(struct ipc_namespace *ns)
1431{
1432 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1433}
1434
1435void mq_put_mnt(struct ipc_namespace *ns)
1436{
Al Viro6f686572011-12-09 00:38:50 -05001437 kern_unmount(ns->mq_mnt);
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001438}
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440static int __init init_mqueue_fs(void)
1441{
1442 int error;
1443
1444 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1445 sizeof(struct mqueue_inode_info), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001446 SLAB_HWCACHE_ALIGN, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 if (mqueue_inode_cachep == NULL)
1448 return -ENOMEM;
1449
André Goddard Rosa2329e3922010-02-23 04:04:27 -03001450 /* ignore failures - they are not fatal */
Serge E. Hallynbdc8e5f2009-04-06 19:01:11 -07001451 mq_sysctl_table = mq_register_sysctl_table();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 error = register_filesystem(&mqueue_fs_type);
1454 if (error)
1455 goto out_sysctl;
1456
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001457 spin_lock_init(&mq_lock);
1458
Al Viro6f686572011-12-09 00:38:50 -05001459 error = mq_init_ns(&init_ipc_ns);
1460 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 goto out_filesystem;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 return 0;
1464
1465out_filesystem:
1466 unregister_filesystem(&mqueue_fs_type);
1467out_sysctl:
1468 if (mq_sysctl_table)
1469 unregister_sysctl_table(mq_sysctl_table);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001470 kmem_cache_destroy(mqueue_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return error;
1472}
1473
Davidlohr Bueso6d08a252014-04-07 15:39:18 -07001474device_initcall(init_mqueue_fs);