blob: 7a2d8f0c8ae506d631bf3fc455f53b042ef29401 [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);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700145 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700146 leaf->priority = msg->m_type;
147 rb_link_node(&leaf->rb_node, parent, p);
148 rb_insert_color(&leaf->rb_node, &info->msg_tree);
Doug Ledfordd6629852012-05-31 16:26:35 -0700149insert_msg:
150 info->attr.mq_curmsgs++;
151 info->qsize += msg->m_ts;
152 list_add_tail(&msg->m_list, &leaf->msg_list);
153 return 0;
154}
155
156static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
157{
158 struct rb_node **p, *parent = NULL;
159 struct posix_msg_tree_node *leaf;
160 struct msg_msg *msg;
161
162try_again:
163 p = &info->msg_tree.rb_node;
164 while (*p) {
165 parent = *p;
166 /*
167 * During insert, low priorities go to the left and high to the
168 * right. On receive, we want the highest priorities first, so
169 * walk all the way to the right.
170 */
171 p = &(*p)->rb_right;
172 }
173 if (!parent) {
174 if (info->attr.mq_curmsgs) {
175 pr_warn_once("Inconsistency in POSIX message queue, "
176 "no tree element, but supposedly messages "
177 "should exist!\n");
178 info->attr.mq_curmsgs = 0;
179 }
180 return NULL;
181 }
182 leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700183 if (unlikely(list_empty(&leaf->msg_list))) {
Doug Ledfordd6629852012-05-31 16:26:35 -0700184 pr_warn_once("Inconsistency in POSIX message queue, "
185 "empty leaf node but we haven't implemented "
186 "lazy leaf delete!\n");
187 rb_erase(&leaf->rb_node, &info->msg_tree);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700188 if (info->node_cache) {
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700189 kfree(leaf);
190 } else {
191 info->node_cache = leaf;
192 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700193 goto try_again;
194 } else {
195 msg = list_first_entry(&leaf->msg_list,
196 struct msg_msg, m_list);
197 list_del(&msg->m_list);
198 if (list_empty(&leaf->msg_list)) {
199 rb_erase(&leaf->rb_node, &info->msg_tree);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700200 if (info->node_cache) {
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700201 kfree(leaf);
202 } else {
203 info->node_cache = leaf;
204 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700205 }
206 }
207 info->attr.mq_curmsgs--;
208 info->qsize -= msg->m_ts;
209 return msg;
210}
211
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700212static struct inode *mqueue_get_inode(struct super_block *sb,
Al Viro1b9d5ff72011-07-24 14:18:20 -0400213 struct ipc_namespace *ipc_ns, umode_t mode,
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700214 struct mq_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
David Howells86a264a2008-11-14 10:39:18 +1100216 struct user_struct *u = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 struct inode *inode;
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700218 int ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220 inode = new_inode(sb);
Jiri Slaby04715202011-07-26 16:08:46 -0700221 if (!inode)
222 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Jiri Slaby04715202011-07-26 16:08:46 -0700224 inode->i_ino = get_next_ino();
225 inode->i_mode = mode;
226 inode->i_uid = current_fsuid();
227 inode->i_gid = current_fsgid();
Deepa Dinamani078cd822016-09-14 07:48:04 -0700228 inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Jiri Slaby04715202011-07-26 16:08:46 -0700230 if (S_ISREG(mode)) {
231 struct mqueue_inode_info *info;
Doug Ledfordd6629852012-05-31 16:26:35 -0700232 unsigned long mq_bytes, mq_treesize;
André Goddard Rosac8308b12010-02-23 04:04:23 -0300233
Jiri Slaby04715202011-07-26 16:08:46 -0700234 inode->i_fop = &mqueue_file_operations;
235 inode->i_size = FILENT_SIZE;
236 /* mqueue specific info */
237 info = MQUEUE_I(inode);
238 spin_lock_init(&info->lock);
239 init_waitqueue_head(&info->wait_q);
240 INIT_LIST_HEAD(&info->e_wait_q[0].list);
241 INIT_LIST_HEAD(&info->e_wait_q[1].list);
242 info->notify_owner = NULL;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800243 info->notify_user_ns = NULL;
Jiri Slaby04715202011-07-26 16:08:46 -0700244 info->qsize = 0;
245 info->user = NULL; /* set when all is ok */
Doug Ledfordd6629852012-05-31 16:26:35 -0700246 info->msg_tree = RB_ROOT;
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700247 info->node_cache = NULL;
Jiri Slaby04715202011-07-26 16:08:46 -0700248 memset(&info->attr, 0, sizeof(info->attr));
KOSAKI Motohirocef01842012-05-31 16:26:33 -0700249 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
250 ipc_ns->mq_msg_default);
251 info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
252 ipc_ns->mq_msgsize_default);
Jiri Slaby04715202011-07-26 16:08:46 -0700253 if (attr) {
254 info->attr.mq_maxmsg = attr->mq_maxmsg;
255 info->attr.mq_msgsize = attr->mq_msgsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700257 /*
258 * We used to allocate a static array of pointers and account
259 * the size of that array as well as one msg_msg struct per
260 * possible message into the queue size. That's no longer
261 * accurate as the queue is now an rbtree and will grow and
262 * shrink depending on usage patterns. We can, however, still
263 * account one msg_msg struct per message, but the nodes are
264 * allocated depending on priority usage, and most programs
265 * only use one, or a handful, of priorities. However, since
266 * this is pinned memory, we need to assume worst case, so
267 * that means the min(mq_maxmsg, max_priorities) * struct
268 * posix_msg_tree_node.
269 */
270 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
271 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
272 sizeof(struct posix_msg_tree_node);
Jiri Slaby04715202011-07-26 16:08:46 -0700273
Doug Ledfordd6629852012-05-31 16:26:35 -0700274 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
275 info->attr.mq_msgsize);
Jiri Slaby04715202011-07-26 16:08:46 -0700276
277 spin_lock(&mq_lock);
278 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
Davidlohr Bueso2a4e64b2012-01-20 14:34:01 -0800279 u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
Jiri Slaby04715202011-07-26 16:08:46 -0700280 spin_unlock(&mq_lock);
281 /* mqueue_evict_inode() releases info->messages */
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700282 ret = -EMFILE;
Jiri Slaby04715202011-07-26 16:08:46 -0700283 goto out_inode;
284 }
285 u->mq_bytes += mq_bytes;
286 spin_unlock(&mq_lock);
287
288 /* all is ok */
289 info->user = get_uid(u);
290 } else if (S_ISDIR(mode)) {
291 inc_nlink(inode);
292 /* Some things misbehave if size == 0 on a directory */
293 inode->i_size = 2 * DIRENT_SIZE;
294 inode->i_op = &mqueue_dir_inode_operations;
295 inode->i_fop = &simple_dir_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 }
Jiri Slaby04715202011-07-26 16:08:46 -0700297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 return inode;
299out_inode:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 iput(inode);
Jiri Slaby04715202011-07-26 16:08:46 -0700301err:
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700302 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303}
304
305static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
306{
307 struct inode *inode;
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500308 struct ipc_namespace *ns = sb->s_fs_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
Eric W. Biedermana2982cc2016-06-09 15:34:02 -0500310 sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300311 sb->s_blocksize = PAGE_SIZE;
312 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 sb->s_magic = MQUEUE_MAGIC;
314 sb->s_op = &mqueue_super_ops;
315
Al Viro48fde702012-01-08 22:15:13 -0500316 inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
317 if (IS_ERR(inode))
318 return PTR_ERR(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
Al Viro48fde702012-01-08 22:15:13 -0500320 sb->s_root = d_make_root(inode);
321 if (!sb->s_root)
322 return -ENOMEM;
323 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
Al Viroceefda62010-07-26 13:16:50 +0400326static struct dentry *mqueue_mount(struct file_system_type *fs_type,
David Howells454e2392006-06-23 02:02:57 -0700327 int flags, const char *dev_name,
Al Viroceefda62010-07-26 13:16:50 +0400328 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500330 struct ipc_namespace *ns;
331 if (flags & MS_KERNMOUNT) {
332 ns = data;
333 data = NULL;
334 } else {
335 ns = current->nsproxy->ipc_ns;
Eric W. Biedermana636b702013-03-21 18:13:15 -0700336 }
Eric W. Biedermand91ee872016-05-23 14:51:59 -0500337 return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700340static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
342 struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
343
Christoph Lametera35afb82007-05-16 22:10:57 -0700344 inode_init_once(&p->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
347static struct inode *mqueue_alloc_inode(struct super_block *sb)
348{
349 struct mqueue_inode_info *ei;
350
Christoph Lametere94b1762006-12-06 20:33:17 -0800351 ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 if (!ei)
353 return NULL;
354 return &ei->vfs_inode;
355}
356
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100357static void mqueue_i_callback(struct rcu_head *head)
358{
359 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100360 kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
361}
362
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363static void mqueue_destroy_inode(struct inode *inode)
364{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100365 call_rcu(&inode->i_rcu, mqueue_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366}
367
Al Viro6d8af642010-06-05 16:29:45 -0400368static void mqueue_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
370 struct mqueue_inode_info *info;
371 struct user_struct *user;
Doug Ledfordd6629852012-05-31 16:26:35 -0700372 unsigned long mq_bytes, mq_treesize;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700373 struct ipc_namespace *ipc_ns;
Doug Ledfordd6629852012-05-31 16:26:35 -0700374 struct msg_msg *msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Jan Karadbd57682012-05-03 14:48:02 +0200376 clear_inode(inode);
Al Viro6d8af642010-06-05 16:29:45 -0400377
378 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 return;
Al Viro6d8af642010-06-05 16:29:45 -0400380
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700381 ipc_ns = get_ns_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 info = MQUEUE_I(inode);
383 spin_lock(&info->lock);
Doug Ledfordd6629852012-05-31 16:26:35 -0700384 while ((msg = msg_get(info)) != NULL)
385 free_msg(msg);
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700386 kfree(info->node_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 spin_unlock(&info->lock);
388
André Goddard Rosa8834cf72010-02-23 04:04:24 -0300389 /* Total amount of bytes accounted for the mqueue */
Doug Ledfordd6629852012-05-31 16:26:35 -0700390 mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
391 min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
392 sizeof(struct posix_msg_tree_node);
393
394 mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
395 info->attr.mq_msgsize);
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 user = info->user;
398 if (user) {
399 spin_lock(&mq_lock);
400 user->mq_bytes -= mq_bytes;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700401 /*
402 * get_ns_from_inode() ensures that the
403 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
404 * to which we now hold a reference, or it is NULL.
405 * We can't put it here under mq_lock, though.
406 */
407 if (ipc_ns)
408 ipc_ns->mq_queues_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 spin_unlock(&mq_lock);
410 free_uid(user);
411 }
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700412 if (ipc_ns)
413 put_ipc_ns(ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
416static int mqueue_create(struct inode *dir, struct dentry *dentry,
Al Viroebfc3b42012-06-10 18:05:36 -0400417 umode_t mode, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418{
419 struct inode *inode;
420 struct mq_attr *attr = dentry->d_fsdata;
421 int error;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700422 struct ipc_namespace *ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424 spin_lock(&mq_lock);
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700425 ipc_ns = __get_ns_from_inode(dir);
426 if (!ipc_ns) {
427 error = -EACCES;
428 goto out_unlock;
429 }
Davidlohr Buesof3713fd2014-02-25 15:01:45 -0800430
431 if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
432 !capable(CAP_SYS_RESOURCE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 error = -ENOSPC;
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700434 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 }
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700436 ipc_ns->mq_queues_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 spin_unlock(&mq_lock);
438
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700439 inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
Jiri Slabyd40dcdb2011-07-26 16:08:47 -0700440 if (IS_ERR(inode)) {
441 error = PTR_ERR(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 spin_lock(&mq_lock);
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700443 ipc_ns->mq_queues_count--;
444 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 }
446
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700447 put_ipc_ns(ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 dir->i_size += DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -0700449 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 d_instantiate(dentry, inode);
452 dget(dentry);
453 return 0;
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700454out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 spin_unlock(&mq_lock);
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700456 if (ipc_ns)
457 put_ipc_ns(ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 return error;
459}
460
461static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
462{
David Howells75c3cfa2015-03-17 22:26:12 +0000463 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Deepa Dinamani078cd822016-09-14 07:48:04 -0700465 dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 dir->i_size -= DIRENT_SIZE;
Manfred Spraul239521f2014-01-27 17:07:04 -0800467 drop_nlink(inode);
468 dput(dentry);
469 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471
472/*
473* This is routine for system read from queue file.
474* To avoid mess with doing here some sort of mq_receive we allow
475* to read only queue size & notification info (the only values
476* that are interesting from user point of view and aren't accessible
477* through std routines)
478*/
479static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700480 size_t count, loff_t *off)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Al Viro496ad9a2013-01-23 17:07:38 -0500482 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 char buffer[FILENT_SIZE];
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700484 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486 spin_lock(&info->lock);
487 snprintf(buffer, sizeof(buffer),
488 "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
489 info->qsize,
490 info->notify_owner ? info->notify.sigev_notify : 0,
491 (info->notify_owner &&
492 info->notify.sigev_notify == SIGEV_SIGNAL) ?
493 info->notify.sigev_signo : 0,
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800494 pid_vnr(info->notify_owner));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 spin_unlock(&info->lock);
496 buffer[sizeof(buffer)-1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700498 ret = simple_read_from_buffer(u_data, count, off, buffer,
499 strlen(buffer));
500 if (ret <= 0)
501 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Deepa Dinamani078cd822016-09-14 07:48:04 -0700503 file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
Akinobu Mitaf1a43f92008-07-25 01:48:07 -0700504 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700507static int mqueue_flush_file(struct file *filp, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
Al Viro496ad9a2013-01-23 17:07:38 -0500509 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511 spin_lock(&info->lock);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700512 if (task_tgid(current) == info->notify_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 remove_notification(info);
514
515 spin_unlock(&info->lock);
516 return 0;
517}
518
519static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
520{
Al Viro496ad9a2013-01-23 17:07:38 -0500521 struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 int retval = 0;
523
524 poll_wait(filp, &info->wait_q, poll_tab);
525
526 spin_lock(&info->lock);
527 if (info->attr.mq_curmsgs)
528 retval = POLLIN | POLLRDNORM;
529
530 if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
531 retval |= POLLOUT | POLLWRNORM;
532 spin_unlock(&info->lock);
533
534 return retval;
535}
536
537/* Adds current to info->e_wait_q[sr] before element with smaller prio */
538static void wq_add(struct mqueue_inode_info *info, int sr,
539 struct ext_wait_queue *ewp)
540{
541 struct ext_wait_queue *walk;
542
543 ewp->task = current;
544
545 list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
546 if (walk->task->static_prio <= current->static_prio) {
547 list_add_tail(&ewp->list, &walk->list);
548 return;
549 }
550 }
551 list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
552}
553
554/*
555 * Puts current task to sleep. Caller must hold queue lock. After return
556 * lock isn't held.
557 * sr: SEND or RECV
558 */
559static int wq_sleep(struct mqueue_inode_info *info, int sr,
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200560 ktime_t *timeout, struct ext_wait_queue *ewp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
562 int retval;
563 signed long time;
564
565 wq_add(info, sr, ewp);
566
567 for (;;) {
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700568 __set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 spin_unlock(&info->lock);
Wanlong Gao32ea8452011-10-31 17:06:35 -0700571 time = schedule_hrtimeout_range_clock(timeout, 0,
572 HRTIMER_MODE_ABS, CLOCK_REALTIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (ewp->state == STATE_READY) {
575 retval = 0;
576 goto out;
577 }
578 spin_lock(&info->lock);
579 if (ewp->state == STATE_READY) {
580 retval = 0;
581 goto out_unlock;
582 }
583 if (signal_pending(current)) {
584 retval = -ERESTARTSYS;
585 break;
586 }
587 if (time == 0) {
588 retval = -ETIMEDOUT;
589 break;
590 }
591 }
592 list_del(&ewp->list);
593out_unlock:
594 spin_unlock(&info->lock);
595out:
596 return retval;
597}
598
599/*
600 * Returns waiting task that should be serviced first or NULL if none exists
601 */
602static struct ext_wait_queue *wq_get_first_waiter(
603 struct mqueue_inode_info *info, int sr)
604{
605 struct list_head *ptr;
606
607 ptr = info->e_wait_q[sr].list.prev;
608 if (ptr == &info->e_wait_q[sr].list)
609 return NULL;
610 return list_entry(ptr, struct ext_wait_queue, list);
611}
612
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614static inline void set_cookie(struct sk_buff *skb, char code)
615{
Manfred Spraul239521f2014-01-27 17:07:04 -0800616 ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617}
618
619/*
620 * The next function is only to split too long sys_mq_timedsend
621 */
622static void __do_notify(struct mqueue_inode_info *info)
623{
624 /* notification
625 * invoked when there is registered process and there isn't process
626 * waiting synchronously for message AND state of queue changed from
627 * empty to not empty. Here we are sure that no one is waiting
628 * synchronously. */
629 if (info->notify_owner &&
630 info->attr.mq_curmsgs == 1) {
631 struct siginfo sig_i;
632 switch (info->notify.sigev_notify) {
633 case SIGEV_NONE:
634 break;
635 case SIGEV_SIGNAL:
636 /* sends signal */
637
638 sig_i.si_signo = info->notify.sigev_signo;
639 sig_i.si_errno = 0;
640 sig_i.si_code = SI_MESGQ;
641 sig_i.si_value = info->notify.sigev_value;
Serge E. Hallyn6b550f92012-01-10 15:11:37 -0800642 /* map current pid/uid into info->owner's namespaces */
643 rcu_read_lock();
Sukadev Bhattiprolua6684992009-01-07 18:08:50 -0800644 sig_i.si_pid = task_tgid_nr_ns(current,
645 ns_of_pid(info->notify_owner));
Eric W. Biederman76b6db02012-03-14 15:24:19 -0700646 sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
Serge E. Hallyn6b550f92012-01-10 15:11:37 -0800647 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700649 kill_pid_info(info->notify.sigev_signo,
650 &sig_i, info->notify_owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 break;
652 case SIGEV_THREAD:
653 set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
Denis V. Lunev7ee015e2007-10-10 21:14:03 -0700654 netlink_sendskb(info->notify_sock, info->notify_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 break;
656 }
657 /* after notification unregisters process */
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700658 put_pid(info->notify_owner);
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800659 put_user_ns(info->notify_user_ns);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700660 info->notify_owner = NULL;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800661 info->notify_user_ns = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 }
663 wake_up(&info->wait_q);
664}
665
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200666static int prepare_timeout(const struct timespec __user *u_abs_timeout,
667 ktime_t *expires, struct timespec *ts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200669 if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
670 return -EFAULT;
671 if (!timespec_valid(ts))
672 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200674 *expires = timespec_to_ktime(*ts);
675 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678static void remove_notification(struct mqueue_inode_info *info)
679{
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700680 if (info->notify_owner != NULL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 info->notify.sigev_notify == SIGEV_THREAD) {
682 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
Denis V. Lunev7ee015e2007-10-10 21:14:03 -0700683 netlink_sendskb(info->notify_sock, info->notify_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700685 put_pid(info->notify_owner);
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800686 put_user_ns(info->notify_user_ns);
Cedric Le Goatera03fcb72006-10-02 02:17:26 -0700687 info->notify_owner = NULL;
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -0800688 info->notify_user_ns = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700691static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Doug Ledford2c12ea42012-05-31 16:26:36 -0700693 int mq_treesize;
694 unsigned long total_size;
695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
Doug Ledford113289c2012-05-31 16:26:36 -0700697 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 if (capable(CAP_SYS_RESOURCE)) {
Doug Ledford02967ea2012-05-31 16:26:29 -0700699 if (attr->mq_maxmsg > HARD_MSGMAX ||
700 attr->mq_msgsize > HARD_MSGSIZEMAX)
Doug Ledford113289c2012-05-31 16:26:36 -0700701 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 } else {
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700703 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
704 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
Doug Ledford113289c2012-05-31 16:26:36 -0700705 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707 /* check for overflow */
708 if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
Doug Ledford113289c2012-05-31 16:26:36 -0700709 return -EOVERFLOW;
Doug Ledford2c12ea42012-05-31 16:26:36 -0700710 mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
711 min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
712 sizeof(struct posix_msg_tree_node);
713 total_size = attr->mq_maxmsg * attr->mq_msgsize;
714 if (total_size + mq_treesize < total_size)
Doug Ledford113289c2012-05-31 16:26:36 -0700715 return -EOVERFLOW;
716 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
719/*
720 * Invoked when creating a new queue via sys_mq_open
721 */
Al Viro765927b2012-06-26 21:58:53 +0400722static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
723 struct path *path, int oflag, umode_t mode,
Serge E. Hallyn614b84c2009-04-06 19:01:08 -0700724 struct mq_attr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
David Howells745ca242008-11-14 10:39:22 +1100726 const struct cred *cred = current_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 int ret;
728
Al Viro564f6992008-12-14 04:02:26 -0500729 if (attr) {
Doug Ledford113289c2012-05-31 16:26:36 -0700730 ret = mq_attr_ok(ipc_ns, attr);
731 if (ret)
Al Viro765927b2012-06-26 21:58:53 +0400732 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 /* store for use during create */
Al Viro765927b2012-06-26 21:58:53 +0400734 path->dentry->d_fsdata = attr;
Doug Ledford113289c2012-05-31 16:26:36 -0700735 } else {
736 struct mq_attr def_attr;
737
738 def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
739 ipc_ns->mq_msg_default);
740 def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
741 ipc_ns->mq_msgsize_default);
742 ret = mq_attr_ok(ipc_ns, &def_attr);
743 if (ret)
Al Viro765927b2012-06-26 21:58:53 +0400744 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
746
Al Viroce3b0f82009-03-29 19:08:22 -0400747 mode &= ~current_umask();
Al Viro765927b2012-06-26 21:58:53 +0400748 ret = vfs_create(dir, path->dentry, mode, true);
749 path->dentry->d_fsdata = NULL;
Al Viro312b90f2012-08-06 10:18:17 +0400750 if (ret)
751 return ERR_PTR(ret);
752 return dentry_open(path, oflag, cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753}
754
755/* Opens existing queue */
Al Viro765927b2012-06-26 21:58:53 +0400756static struct file *do_open(struct path *path, int oflag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
David Howells745ca242008-11-14 10:39:22 +1100758 static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
759 MAY_READ | MAY_WRITE };
Al Viro765927b2012-06-26 21:58:53 +0400760 int acc;
761 if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
762 return ERR_PTR(-EINVAL);
763 acc = oflag2acc[oflag & O_ACCMODE];
David Howells75c3cfa2015-03-17 22:26:12 +0000764 if (inode_permission(d_inode(path->dentry), acc))
Al Viro765927b2012-06-26 21:58:53 +0400765 return ERR_PTR(-EACCES);
766 return dentry_open(path, oflag, current_cred());
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767}
768
Al Virodf0a4282011-07-26 05:26:10 -0400769SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
Heiko Carstensd5460c92009-01-14 14:14:27 +0100770 struct mq_attr __user *, u_attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
Al Viro765927b2012-06-26 21:58:53 +0400772 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 struct file *filp;
Jeff Layton91a27b22012-10-10 15:25:28 -0400774 struct filename *name;
Al Viro564f6992008-12-14 04:02:26 -0500775 struct mq_attr attr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 int fd, error;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700777 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
Al Viro312b90f2012-08-06 10:18:17 +0400778 struct vfsmount *mnt = ipc_ns->mq_mnt;
779 struct dentry *root = mnt->mnt_root;
780 int ro;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Al Viro564f6992008-12-14 04:02:26 -0500782 if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
783 return -EFAULT;
784
785 audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
George C. Wilson20ca73b2006-05-24 16:09:55 -0500786
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 if (IS_ERR(name = getname(u_name)))
788 return PTR_ERR(name);
789
Ulrich Drepper269f2132008-05-03 15:28:45 -0400790 fd = get_unused_fd_flags(O_CLOEXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (fd < 0)
792 goto out_putname;
793
Al Viro312b90f2012-08-06 10:18:17 +0400794 ro = mnt_want_write(mnt); /* we'll drop it in any case */
Al Viro765927b2012-06-26 21:58:53 +0400795 error = 0;
Al Viro59551022016-01-22 15:40:57 -0500796 inode_lock(d_inode(root));
Jeff Layton91a27b22012-10-10 15:25:28 -0400797 path.dentry = lookup_one_len(name->name, root, strlen(name->name));
Al Viro765927b2012-06-26 21:58:53 +0400798 if (IS_ERR(path.dentry)) {
799 error = PTR_ERR(path.dentry);
André Goddard Rosa4294a8e2010-02-23 04:04:28 -0300800 goto out_putfd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 }
Al Viro312b90f2012-08-06 10:18:17 +0400802 path.mnt = mntget(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 if (oflag & O_CREAT) {
David Howells75c3cfa2015-03-17 22:26:12 +0000805 if (d_really_is_positive(path.dentry)) { /* entry already exists */
Jeff Laytonadb5c242012-10-10 16:43:13 -0400806 audit_inode(name, path.dentry, 0);
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300807 if (oflag & O_EXCL) {
808 error = -EEXIST;
Alexander Viro7c7dce92006-01-14 15:29:55 -0500809 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300810 }
Al Viro765927b2012-06-26 21:58:53 +0400811 filp = do_open(&path, oflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 } else {
Al Viro312b90f2012-08-06 10:18:17 +0400813 if (ro) {
814 error = ro;
815 goto out;
816 }
Jeff Layton79f65302013-07-08 15:59:36 -0700817 audit_inode_parent_hidden(name, root);
David Howells75c3cfa2015-03-17 22:26:12 +0000818 filp = do_create(ipc_ns, d_inode(root),
Al Viro765927b2012-06-26 21:58:53 +0400819 &path, oflag, mode,
Al Viro564f6992008-12-14 04:02:26 -0500820 u_attr ? &attr : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 }
Alexander Viro7c7dce92006-01-14 15:29:55 -0500822 } else {
David Howells75c3cfa2015-03-17 22:26:12 +0000823 if (d_really_is_negative(path.dentry)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300824 error = -ENOENT;
Alexander Viro7c7dce92006-01-14 15:29:55 -0500825 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300826 }
Jeff Laytonadb5c242012-10-10 16:43:13 -0400827 audit_inode(name, path.dentry, 0);
Al Viro765927b2012-06-26 21:58:53 +0400828 filp = do_open(&path, oflag);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Al Viro765927b2012-06-26 21:58:53 +0400831 if (!IS_ERR(filp))
832 fd_install(fd, filp);
833 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 error = PTR_ERR(filp);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500835out:
Al Viro765927b2012-06-26 21:58:53 +0400836 path_put(&path);
Alexander Viro7c7dce92006-01-14 15:29:55 -0500837out_putfd:
Al Viro765927b2012-06-26 21:58:53 +0400838 if (error) {
839 put_unused_fd(fd);
840 fd = error;
841 }
Al Viro59551022016-01-22 15:40:57 -0500842 inode_unlock(d_inode(root));
Vladimir Davydov38d78e52013-03-22 15:04:51 -0700843 if (!ro)
844 mnt_drop_write(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845out_putname:
846 putname(name);
847 return fd;
848}
849
Heiko Carstensd5460c92009-01-14 14:14:27 +0100850SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851{
852 int err;
Jeff Layton91a27b22012-10-10 15:25:28 -0400853 struct filename *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 struct dentry *dentry;
855 struct inode *inode = NULL;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -0700856 struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
Al Viro312b90f2012-08-06 10:18:17 +0400857 struct vfsmount *mnt = ipc_ns->mq_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859 name = getname(u_name);
860 if (IS_ERR(name))
861 return PTR_ERR(name);
862
Jeff Layton79f65302013-07-08 15:59:36 -0700863 audit_inode_parent_hidden(name, mnt->mnt_root);
Al Viro312b90f2012-08-06 10:18:17 +0400864 err = mnt_want_write(mnt);
865 if (err)
866 goto out_name;
Al Viro59551022016-01-22 15:40:57 -0500867 inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
Jeff Layton91a27b22012-10-10 15:25:28 -0400868 dentry = lookup_one_len(name->name, mnt->mnt_root,
869 strlen(name->name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 if (IS_ERR(dentry)) {
871 err = PTR_ERR(dentry);
872 goto out_unlock;
873 }
874
David Howells75c3cfa2015-03-17 22:26:12 +0000875 inode = d_inode(dentry);
Al Viro312b90f2012-08-06 10:18:17 +0400876 if (!inode) {
877 err = -ENOENT;
878 } else {
Al Viro7de9c6ee2010-10-23 11:11:40 -0400879 ihold(inode);
David Howells75c3cfa2015-03-17 22:26:12 +0000880 err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
Al Viro312b90f2012-08-06 10:18:17 +0400881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 dput(dentry);
883
884out_unlock:
Al Viro59551022016-01-22 15:40:57 -0500885 inode_unlock(d_inode(mnt->mnt_root));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 if (inode)
887 iput(inode);
Al Viro312b90f2012-08-06 10:18:17 +0400888 mnt_drop_write(mnt);
889out_name:
890 putname(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 return err;
893}
894
895/* Pipelined send and receive functions.
896 *
897 * If a receiver finds no waiting message, then it registers itself in the
898 * list of waiting receivers. A sender checks that list before adding the new
899 * message into the message array. If there is a waiting receiver, then it
900 * bypasses the message array and directly hands the message over to the
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700901 * receiver. The receiver accepts the message and returns without grabbing the
902 * queue spinlock:
903 *
904 * - Set pointer to message.
905 * - Queue the receiver task for later wakeup (without the info->lock).
906 * - Update its state to STATE_READY. Now the receiver can continue.
907 * - Wake up the process after the lock is dropped. Should the process wake up
908 * before this wakeup (due to a timeout or a signal) it will either see
909 * STATE_READY and continue or acquire the lock to check the state again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 *
911 * The same algorithm is used for senders.
912 */
913
914/* pipelined_send() - send a message directly to the task waiting in
915 * sys_mq_timedreceive() (without inserting message into a queue).
916 */
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700917static inline void pipelined_send(struct wake_q_head *wake_q,
918 struct mqueue_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 struct msg_msg *message,
920 struct ext_wait_queue *receiver)
921{
922 receiver->msg = message;
923 list_del(&receiver->list);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700924 wake_q_add(wake_q, receiver->task);
925 /*
926 * Rely on the implicit cmpxchg barrier from wake_q_add such
927 * that we can ensure that updating receiver->state is the last
928 * write operation: As once set, the receiver can continue,
929 * and if we don't have the reference count from the wake_q,
930 * yet, at that point we can later have a use-after-free
931 * condition and bogus wakeup.
932 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 receiver->state = STATE_READY;
934}
935
936/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
937 * gets its message and put to the queue (we have one free place for sure). */
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700938static inline void pipelined_receive(struct wake_q_head *wake_q,
939 struct mqueue_inode_info *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
942
943 if (!sender) {
944 /* for poll */
945 wake_up_interruptible(&info->wait_q);
946 return;
947 }
Doug Ledfordd6629852012-05-31 16:26:35 -0700948 if (msg_insert(sender->msg, info))
949 return;
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 list_del(&sender->list);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -0700952 wake_q_add(wake_q, sender->task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 sender->state = STATE_READY;
954}
955
Heiko Carstensc4ea37c2009-01-14 14:14:28 +0100956SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
957 size_t, msg_len, unsigned int, msg_prio,
958 const struct timespec __user *, u_abs_timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959{
Al Viro2903ff02012-08-28 12:52:22 -0400960 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 struct inode *inode;
962 struct ext_wait_queue wait;
963 struct ext_wait_queue *receiver;
964 struct msg_msg *msg_ptr;
965 struct mqueue_inode_info *info;
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200966 ktime_t expires, *timeout = NULL;
967 struct timespec ts;
Doug Ledfordce2d52c2012-05-31 16:26:38 -0700968 struct posix_msg_tree_node *new_leaf = NULL;
Al Viro2903ff02012-08-28 12:52:22 -0400969 int ret = 0;
Waiman Long194a6b52016-11-17 11:46:38 -0500970 DEFINE_WAKE_Q(wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Al Viroc32c8af2008-12-14 03:46:48 -0500972 if (u_abs_timeout) {
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200973 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
974 if (res)
975 return res;
976 timeout = &expires;
Al Viroc32c8af2008-12-14 03:46:48 -0500977 }
George C. Wilson20ca73b2006-05-24 16:09:55 -0500978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
980 return -EINVAL;
981
Carsten Emde9ca7d8e2010-04-02 22:40:20 +0200982 audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
Al Viro2903ff02012-08-28 12:52:22 -0400984 f = fdget(mqdes);
985 if (unlikely(!f.file)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300986 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Al Viro496ad9a2013-01-23 17:07:38 -0500990 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -0400991 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300992 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 info = MQUEUE_I(inode);
Al Viro9f45f5b2014-10-31 17:44:57 -0400996 audit_file(f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Al Viro2903ff02012-08-28 12:52:22 -0400998 if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -0300999 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
1003 if (unlikely(msg_len > info->attr.mq_msgsize)) {
1004 ret = -EMSGSIZE;
1005 goto out_fput;
1006 }
1007
1008 /* First try to allocate memory, before doing anything with
1009 * existing queues. */
1010 msg_ptr = load_msg(u_msg_ptr, msg_len);
1011 if (IS_ERR(msg_ptr)) {
1012 ret = PTR_ERR(msg_ptr);
1013 goto out_fput;
1014 }
1015 msg_ptr->m_ts = msg_len;
1016 msg_ptr->m_type = msg_prio;
1017
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001018 /*
1019 * msg_insert really wants us to have a valid, spare node struct so
1020 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1021 * fall back to that if necessary.
1022 */
1023 if (!info->node_cache)
1024 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1025
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 spin_lock(&info->lock);
1027
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001028 if (!info->node_cache && new_leaf) {
1029 /* Save our speculative allocation into the cache */
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001030 INIT_LIST_HEAD(&new_leaf->msg_list);
1031 info->node_cache = new_leaf;
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001032 new_leaf = NULL;
1033 } else {
1034 kfree(new_leaf);
1035 }
1036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
Al Viro2903ff02012-08-28 12:52:22 -04001038 if (f.file->f_flags & O_NONBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 } else {
1041 wait.task = current;
1042 wait.msg = (void *) msg_ptr;
1043 wait.state = STATE_NONE;
1044 ret = wq_sleep(info, SEND, timeout, &wait);
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001045 /*
1046 * wq_sleep must be called with info->lock held, and
1047 * returns with the lock released
1048 */
1049 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 } else {
1052 receiver = wq_get_first_waiter(info, RECV);
1053 if (receiver) {
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001054 pipelined_send(&wake_q, info, msg_ptr, receiver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 } else {
1056 /* adds message to the queue */
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001057 ret = msg_insert(msg_ptr, info);
1058 if (ret)
1059 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 __do_notify(info);
1061 }
1062 inode->i_atime = inode->i_mtime = inode->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07001063 current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001065out_unlock:
1066 spin_unlock(&info->lock);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001067 wake_up_q(&wake_q);
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001068out_free:
1069 if (ret)
1070 free_msg(msg_ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001072 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073out:
1074 return ret;
1075}
1076
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001077SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1078 size_t, msg_len, unsigned int __user *, u_msg_prio,
1079 const struct timespec __user *, u_abs_timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 ssize_t ret;
1082 struct msg_msg *msg_ptr;
Al Viro2903ff02012-08-28 12:52:22 -04001083 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 struct inode *inode;
1085 struct mqueue_inode_info *info;
1086 struct ext_wait_queue wait;
Carsten Emde9ca7d8e2010-04-02 22:40:20 +02001087 ktime_t expires, *timeout = NULL;
1088 struct timespec ts;
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001089 struct posix_msg_tree_node *new_leaf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Al Viroc32c8af2008-12-14 03:46:48 -05001091 if (u_abs_timeout) {
Carsten Emde9ca7d8e2010-04-02 22:40:20 +02001092 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1093 if (res)
1094 return res;
1095 timeout = &expires;
Al Viroc32c8af2008-12-14 03:46:48 -05001096 }
George C. Wilson20ca73b2006-05-24 16:09:55 -05001097
Carsten Emde9ca7d8e2010-04-02 22:40:20 +02001098 audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Al Viro2903ff02012-08-28 12:52:22 -04001100 f = fdget(mqdes);
1101 if (unlikely(!f.file)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001102 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Al Viro496ad9a2013-01-23 17:07:38 -05001106 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -04001107 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001108 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 info = MQUEUE_I(inode);
Al Viro9f45f5b2014-10-31 17:44:57 -04001112 audit_file(f.file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Al Viro2903ff02012-08-28 12:52:22 -04001114 if (unlikely(!(f.file->f_mode & FMODE_READ))) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001115 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 /* checks if buffer is big enough */
1120 if (unlikely(msg_len < info->attr.mq_msgsize)) {
1121 ret = -EMSGSIZE;
1122 goto out_fput;
1123 }
1124
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001125 /*
1126 * msg_insert really wants us to have a valid, spare node struct so
1127 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1128 * fall back to that if necessary.
1129 */
1130 if (!info->node_cache)
1131 new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1132
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 spin_lock(&info->lock);
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001134
1135 if (!info->node_cache && new_leaf) {
1136 /* Save our speculative allocation into the cache */
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001137 INIT_LIST_HEAD(&new_leaf->msg_list);
1138 info->node_cache = new_leaf;
Doug Ledfordce2d52c2012-05-31 16:26:38 -07001139 } else {
1140 kfree(new_leaf);
1141 }
1142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (info->attr.mq_curmsgs == 0) {
Al Viro2903ff02012-08-28 12:52:22 -04001144 if (f.file->f_flags & O_NONBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 spin_unlock(&info->lock);
1146 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 } else {
1148 wait.task = current;
1149 wait.state = STATE_NONE;
1150 ret = wq_sleep(info, RECV, timeout, &wait);
1151 msg_ptr = wait.msg;
1152 }
1153 } else {
Waiman Long194a6b52016-11-17 11:46:38 -05001154 DEFINE_WAKE_Q(wake_q);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 msg_ptr = msg_get(info);
1157
1158 inode->i_atime = inode->i_mtime = inode->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07001159 current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 /* There is now free space in queue. */
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001162 pipelined_receive(&wake_q, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 spin_unlock(&info->lock);
Davidlohr Buesofa6004a2015-05-04 07:02:46 -07001164 wake_up_q(&wake_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 ret = 0;
1166 }
1167 if (ret == 0) {
1168 ret = msg_ptr->m_ts;
1169
1170 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1171 store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1172 ret = -EFAULT;
1173 }
1174 free_msg(msg_ptr);
1175 }
1176out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001177 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178out:
1179 return ret;
1180}
1181
1182/*
1183 * Notes: the case when user wants us to deregister (with NULL as pointer)
1184 * and he isn't currently owner of notification, will be silently discarded.
1185 * It isn't explicitly defined in the POSIX.
1186 */
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001187SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1188 const struct sigevent __user *, u_notification)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Al Viro2903ff02012-08-28 12:52:22 -04001190 int ret;
1191 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 struct sock *sock;
1193 struct inode *inode;
1194 struct sigevent notification;
1195 struct mqueue_inode_info *info;
1196 struct sk_buff *nc;
1197
Al Viro20114f72008-12-10 07:16:12 -05001198 if (u_notification) {
1199 if (copy_from_user(&notification, u_notification,
1200 sizeof(struct sigevent)))
1201 return -EFAULT;
1202 }
1203
1204 audit_mq_notify(mqdes, u_notification ? &notification : NULL);
George C. Wilson20ca73b2006-05-24 16:09:55 -05001205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 nc = NULL;
1207 sock = NULL;
1208 if (u_notification != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1210 notification.sigev_notify != SIGEV_SIGNAL &&
1211 notification.sigev_notify != SIGEV_THREAD))
1212 return -EINVAL;
1213 if (notification.sigev_notify == SIGEV_SIGNAL &&
Jesper Juhl7ed20e12005-05-01 08:59:14 -07001214 !valid_signal(notification.sigev_signo)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 return -EINVAL;
1216 }
1217 if (notification.sigev_notify == SIGEV_THREAD) {
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001218 long timeo;
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 /* create the notify skb */
1221 nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001222 if (!nc) {
1223 ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 if (copy_from_user(nc->data,
1227 notification.sigev_value.sival_ptr,
1228 NOTIFY_COOKIE_LEN)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001229 ret = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 goto out;
1231 }
1232
1233 /* TODO: add a header? */
1234 skb_put(nc, NOTIFY_COOKIE_LEN);
1235 /* and attach it to the socket */
1236retry:
Al Viro2903ff02012-08-28 12:52:22 -04001237 f = fdget(notification.sigev_signo);
1238 if (!f.file) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001239 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001241 }
Al Viro2903ff02012-08-28 12:52:22 -04001242 sock = netlink_getsockbyfilp(f.file);
1243 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 if (IS_ERR(sock)) {
1245 ret = PTR_ERR(sock);
1246 sock = NULL;
1247 goto out;
1248 }
1249
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001250 timeo = MAX_SCHEDULE_TIMEOUT;
Denis V. Lunev9457afe2008-06-05 11:23:39 -07001251 ret = netlink_attachskb(sock, nc, &timeo, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (ret == 1)
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001253 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (ret) {
1255 sock = NULL;
1256 nc = NULL;
1257 goto out;
1258 }
1259 }
1260 }
1261
Al Viro2903ff02012-08-28 12:52:22 -04001262 f = fdget(mqdes);
1263 if (!f.file) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001264 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Al Viro496ad9a2013-01-23 17:07:38 -05001268 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -04001269 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001270 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 info = MQUEUE_I(inode);
1274
1275 ret = 0;
1276 spin_lock(&info->lock);
1277 if (u_notification == NULL) {
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001278 if (info->notify_owner == task_tgid(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 remove_notification(info);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001280 inode->i_atime = inode->i_ctime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001282 } else if (info->notify_owner != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 ret = -EBUSY;
1284 } else {
1285 switch (notification.sigev_notify) {
1286 case SIGEV_NONE:
1287 info->notify.sigev_notify = SIGEV_NONE;
1288 break;
1289 case SIGEV_THREAD:
1290 info->notify_sock = sock;
1291 info->notify_cookie = nc;
1292 sock = NULL;
1293 nc = NULL;
1294 info->notify.sigev_notify = SIGEV_THREAD;
1295 break;
1296 case SIGEV_SIGNAL:
1297 info->notify.sigev_signo = notification.sigev_signo;
1298 info->notify.sigev_value = notification.sigev_value;
1299 info->notify.sigev_notify = SIGEV_SIGNAL;
1300 break;
1301 }
Cedric Le Goatera03fcb72006-10-02 02:17:26 -07001302
1303 info->notify_owner = get_pid(task_tgid(current));
Eric W. Biederman6f9ac6d2011-11-16 22:57:55 -08001304 info->notify_user_ns = get_user_ns(current_user_ns());
Deepa Dinamani078cd822016-09-14 07:48:04 -07001305 inode->i_atime = inode->i_ctime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 }
1307 spin_unlock(&info->lock);
1308out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001309 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310out:
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001311 if (sock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 netlink_detachskb(sock, nc);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001313 else if (nc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 dev_kfree_skb(nc);
Davidlohr Bueso3ab08fe2014-01-27 17:07:06 -08001315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 return ret;
1317}
1318
Heiko Carstensc4ea37c2009-01-14 14:14:28 +01001319SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1320 const struct mq_attr __user *, u_mqstat,
1321 struct mq_attr __user *, u_omqstat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
1323 int ret;
1324 struct mq_attr mqstat, omqstat;
Al Viro2903ff02012-08-28 12:52:22 -04001325 struct fd f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 struct inode *inode;
1327 struct mqueue_inode_info *info;
1328
1329 if (u_mqstat != NULL) {
1330 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1331 return -EFAULT;
1332 if (mqstat.mq_flags & (~O_NONBLOCK))
1333 return -EINVAL;
1334 }
1335
Al Viro2903ff02012-08-28 12:52:22 -04001336 f = fdget(mqdes);
1337 if (!f.file) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001338 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 goto out;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Al Viro496ad9a2013-01-23 17:07:38 -05001342 inode = file_inode(f.file);
Al Viro2903ff02012-08-28 12:52:22 -04001343 if (unlikely(f.file->f_op != &mqueue_file_operations)) {
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001344 ret = -EBADF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 goto out_fput;
André Goddard Rosa8d8ffef2010-02-23 04:04:26 -03001346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 info = MQUEUE_I(inode);
1348
1349 spin_lock(&info->lock);
1350
1351 omqstat = info->attr;
Al Viro2903ff02012-08-28 12:52:22 -04001352 omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 if (u_mqstat) {
Al Viro73929062008-12-10 06:58:59 -05001354 audit_mq_getsetattr(mqdes, &mqstat);
Al Viro2903ff02012-08-28 12:52:22 -04001355 spin_lock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 if (mqstat.mq_flags & O_NONBLOCK)
Al Viro2903ff02012-08-28 12:52:22 -04001357 f.file->f_flags |= O_NONBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 else
Al Viro2903ff02012-08-28 12:52:22 -04001359 f.file->f_flags &= ~O_NONBLOCK;
1360 spin_unlock(&f.file->f_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Deepa Dinamani078cd822016-09-14 07:48:04 -07001362 inode->i_atime = inode->i_ctime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364
1365 spin_unlock(&info->lock);
1366
1367 ret = 0;
1368 if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1369 sizeof(struct mq_attr)))
1370 ret = -EFAULT;
1371
1372out_fput:
Al Viro2903ff02012-08-28 12:52:22 -04001373 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374out:
1375 return ret;
1376}
1377
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001378static const struct inode_operations mqueue_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 .lookup = simple_lookup,
1380 .create = mqueue_create,
1381 .unlink = mqueue_unlink,
1382};
1383
Arjan van de Ven9a321442007-02-12 00:55:35 -08001384static const struct file_operations mqueue_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 .flush = mqueue_flush_file,
1386 .poll = mqueue_poll_file,
1387 .read = mqueue_read_file,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001388 .llseek = default_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389};
1390
Alexey Dobriyanb87221d2009-09-21 17:01:09 -07001391static const struct super_operations mqueue_super_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 .alloc_inode = mqueue_alloc_inode,
1393 .destroy_inode = mqueue_destroy_inode,
Al Viro6d8af642010-06-05 16:29:45 -04001394 .evict_inode = mqueue_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 .statfs = simple_statfs,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396};
1397
1398static struct file_system_type mqueue_fs_type = {
1399 .name = "mqueue",
Al Viroceefda62010-07-26 13:16:50 +04001400 .mount = mqueue_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 .kill_sb = kill_litter_super,
Gao fengbc1b69e2013-01-28 11:09:01 +08001402 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403};
1404
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001405int mq_init_ns(struct ipc_namespace *ns)
1406{
1407 ns->mq_queues_count = 0;
1408 ns->mq_queues_max = DFLT_QUEUESMAX;
1409 ns->mq_msg_max = DFLT_MSGMAX;
1410 ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
KOSAKI Motohirocef01842012-05-31 16:26:33 -07001411 ns->mq_msg_default = DFLT_MSG;
1412 ns->mq_msgsize_default = DFLT_MSGSIZE;
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001413
1414 ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1415 if (IS_ERR(ns->mq_mnt)) {
1416 int err = PTR_ERR(ns->mq_mnt);
1417 ns->mq_mnt = NULL;
1418 return err;
1419 }
1420 return 0;
1421}
1422
1423void mq_clear_sbinfo(struct ipc_namespace *ns)
1424{
1425 ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1426}
1427
1428void mq_put_mnt(struct ipc_namespace *ns)
1429{
Al Viro6f686572011-12-09 00:38:50 -05001430 kern_unmount(ns->mq_mnt);
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001431}
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433static int __init init_mqueue_fs(void)
1434{
1435 int error;
1436
1437 mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1438 sizeof(struct mqueue_inode_info), 0,
Vladimir Davydov5d097052016-01-14 15:18:21 -08001439 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (mqueue_inode_cachep == NULL)
1441 return -ENOMEM;
1442
André Goddard Rosa2329e3922010-02-23 04:04:27 -03001443 /* ignore failures - they are not fatal */
Serge E. Hallynbdc8e5f2009-04-06 19:01:11 -07001444 mq_sysctl_table = mq_register_sysctl_table();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
1446 error = register_filesystem(&mqueue_fs_type);
1447 if (error)
1448 goto out_sysctl;
1449
Serge E. Hallyn7eafd7c2009-04-06 19:01:10 -07001450 spin_lock_init(&mq_lock);
1451
Al Viro6f686572011-12-09 00:38:50 -05001452 error = mq_init_ns(&init_ipc_ns);
1453 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 goto out_filesystem;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 return 0;
1457
1458out_filesystem:
1459 unregister_filesystem(&mqueue_fs_type);
1460out_sysctl:
1461 if (mq_sysctl_table)
1462 unregister_sysctl_table(mq_sysctl_table);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001463 kmem_cache_destroy(mqueue_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return error;
1465}
1466
Davidlohr Bueso6d08a252014-04-07 15:39:18 -07001467device_initcall(init_mqueue_fs);