]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * POSIX message queues filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2003,2004 Krzysztof Benedyczak (golbi@mat.uni.torun.pl) | |
f66e928b | 5 | * Michal Wronski (michal.wronski@gmail.com) |
1da177e4 LT |
6 | * |
7 | * Spinlocks: Mohamed Abbas (abbas.mohamed@intel.com) | |
8 | * Lockless receive & send, fd based notify: | |
239521f3 | 9 | * Manfred Spraul (manfred@colorfullife.com) |
1da177e4 | 10 | * |
20ca73bc GW |
11 | * Audit: George Wilson (ltcgcw@us.ibm.com) |
12 | * | |
1da177e4 LT |
13 | * This file is released under the GPL. |
14 | */ | |
15 | ||
c59ede7b | 16 | #include <linux/capability.h> |
1da177e4 LT |
17 | #include <linux/init.h> |
18 | #include <linux/pagemap.h> | |
19 | #include <linux/file.h> | |
20 | #include <linux/mount.h> | |
935c6912 | 21 | #include <linux/fs_context.h> |
1da177e4 LT |
22 | #include <linux/namei.h> |
23 | #include <linux/sysctl.h> | |
24 | #include <linux/poll.h> | |
25 | #include <linux/mqueue.h> | |
26 | #include <linux/msg.h> | |
27 | #include <linux/skbuff.h> | |
5b5c4d1a | 28 | #include <linux/vmalloc.h> |
1da177e4 LT |
29 | #include <linux/netlink.h> |
30 | #include <linux/syscalls.h> | |
20ca73bc | 31 | #include <linux/audit.h> |
7ed20e1a | 32 | #include <linux/signal.h> |
5f921ae9 | 33 | #include <linux/mutex.h> |
b488893a PE |
34 | #include <linux/nsproxy.h> |
35 | #include <linux/pid.h> | |
614b84cf | 36 | #include <linux/ipc_namespace.h> |
6b550f94 | 37 | #include <linux/user_namespace.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
84f001e1 | 39 | #include <linux/sched/wake_q.h> |
3f07c014 | 40 | #include <linux/sched/signal.h> |
8703e8a4 | 41 | #include <linux/sched/user.h> |
5f921ae9 | 42 | |
1da177e4 LT |
43 | #include <net/sock.h> |
44 | #include "util.h" | |
45 | ||
935c6912 DH |
46 | struct mqueue_fs_context { |
47 | struct ipc_namespace *ipc_ns; | |
d60c4d01 | 48 | bool newns; /* Set if newly created ipc namespace */ |
935c6912 DH |
49 | }; |
50 | ||
1da177e4 LT |
51 | #define MQUEUE_MAGIC 0x19800202 |
52 | #define DIRENT_SIZE 20 | |
53 | #define FILENT_SIZE 80 | |
54 | ||
55 | #define SEND 0 | |
56 | #define RECV 1 | |
57 | ||
58 | #define STATE_NONE 0 | |
fa6004ad | 59 | #define STATE_READY 1 |
1da177e4 | 60 | |
d6629859 DL |
61 | struct posix_msg_tree_node { |
62 | struct rb_node rb_node; | |
63 | struct list_head msg_list; | |
64 | int priority; | |
65 | }; | |
66 | ||
c5b2cbdb MS |
67 | /* |
68 | * Locking: | |
69 | * | |
70 | * Accesses to a message queue are synchronized by acquiring info->lock. | |
71 | * | |
72 | * There are two notable exceptions: | |
73 | * - The actual wakeup of a sleeping task is performed using the wake_q | |
74 | * framework. info->lock is already released when wake_up_q is called. | |
75 | * - The exit codepaths after sleeping check ext_wait_queue->state without | |
76 | * any locks. If it is STATE_READY, then the syscall is completed without | |
77 | * acquiring info->lock. | |
78 | * | |
79 | * MQ_BARRIER: | |
80 | * To achieve proper release/acquire memory barrier pairing, the state is set to | |
81 | * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed | |
82 | * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used. | |
83 | * | |
84 | * This prevents the following races: | |
85 | * | |
86 | * 1) With the simple wake_q_add(), the task could be gone already before | |
87 | * the increase of the reference happens | |
88 | * Thread A | |
89 | * Thread B | |
90 | * WRITE_ONCE(wait.state, STATE_NONE); | |
91 | * schedule_hrtimeout() | |
92 | * wake_q_add(A) | |
93 | * if (cmpxchg()) // success | |
94 | * ->state = STATE_READY (reordered) | |
95 | * <timeout returns> | |
96 | * if (wait.state == STATE_READY) return; | |
97 | * sysret to user space | |
98 | * sys_exit() | |
99 | * get_task_struct() // UaF | |
100 | * | |
101 | * Solution: Use wake_q_add_safe() and perform the get_task_struct() before | |
102 | * the smp_store_release() that does ->state = STATE_READY. | |
103 | * | |
104 | * 2) Without proper _release/_acquire barriers, the woken up task | |
105 | * could read stale data | |
106 | * | |
107 | * Thread A | |
108 | * Thread B | |
109 | * do_mq_timedreceive | |
110 | * WRITE_ONCE(wait.state, STATE_NONE); | |
111 | * schedule_hrtimeout() | |
112 | * state = STATE_READY; | |
113 | * <timeout returns> | |
114 | * if (wait.state == STATE_READY) return; | |
115 | * msg_ptr = wait.msg; // Access to stale data! | |
116 | * receiver->msg = message; (reordered) | |
117 | * | |
118 | * Solution: use _release and _acquire barriers. | |
119 | * | |
120 | * 3) There is intentionally no barrier when setting current->state | |
121 | * to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the | |
122 | * release memory barrier, and the wakeup is triggered when holding | |
123 | * info->lock, i.e. spin_lock(&info->lock) provided a pairing | |
124 | * acquire memory barrier. | |
125 | */ | |
126 | ||
1da177e4 LT |
127 | struct ext_wait_queue { /* queue of sleeping tasks */ |
128 | struct task_struct *task; | |
129 | struct list_head list; | |
130 | struct msg_msg *msg; /* ptr of loaded message */ | |
131 | int state; /* one of STATE_* values */ | |
132 | }; | |
133 | ||
134 | struct mqueue_inode_info { | |
135 | spinlock_t lock; | |
136 | struct inode vfs_inode; | |
137 | wait_queue_head_t wait_q; | |
138 | ||
d6629859 | 139 | struct rb_root msg_tree; |
a5091fda | 140 | struct rb_node *msg_tree_rightmost; |
ce2d52cc | 141 | struct posix_msg_tree_node *node_cache; |
1da177e4 LT |
142 | struct mq_attr attr; |
143 | ||
144 | struct sigevent notify; | |
239521f3 | 145 | struct pid *notify_owner; |
b5f20061 | 146 | u32 notify_self_exec_id; |
6f9ac6d9 | 147 | struct user_namespace *notify_user_ns; |
6e52a9f0 | 148 | struct ucounts *ucounts; /* user who created, for accounting */ |
1da177e4 LT |
149 | struct sock *notify_sock; |
150 | struct sk_buff *notify_cookie; | |
151 | ||
152 | /* for tasks waiting for free space and messages, respectively */ | |
153 | struct ext_wait_queue e_wait_q[2]; | |
154 | ||
155 | unsigned long qsize; /* size of queue in memory (sum of all msgs) */ | |
156 | }; | |
157 | ||
935c6912 | 158 | static struct file_system_type mqueue_fs_type; |
92e1d5be | 159 | static const struct inode_operations mqueue_dir_inode_operations; |
9a32144e | 160 | static const struct file_operations mqueue_file_operations; |
b87221de | 161 | static const struct super_operations mqueue_super_ops; |
935c6912 | 162 | static const struct fs_context_operations mqueue_fs_context_ops; |
1da177e4 LT |
163 | static void remove_notification(struct mqueue_inode_info *info); |
164 | ||
e18b890b | 165 | static struct kmem_cache *mqueue_inode_cachep; |
1da177e4 | 166 | |
239521f3 | 167 | static struct ctl_table_header *mq_sysctl_table; |
1da177e4 LT |
168 | |
169 | static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode) | |
170 | { | |
171 | return container_of(inode, struct mqueue_inode_info, vfs_inode); | |
172 | } | |
173 | ||
7eafd7c7 SH |
174 | /* |
175 | * This routine should be called with the mq_lock held. | |
176 | */ | |
177 | static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode) | |
614b84cf | 178 | { |
7eafd7c7 | 179 | return get_ipc_ns(inode->i_sb->s_fs_info); |
614b84cf SH |
180 | } |
181 | ||
7eafd7c7 | 182 | static struct ipc_namespace *get_ns_from_inode(struct inode *inode) |
614b84cf | 183 | { |
7eafd7c7 SH |
184 | struct ipc_namespace *ns; |
185 | ||
186 | spin_lock(&mq_lock); | |
187 | ns = __get_ns_from_inode(inode); | |
188 | spin_unlock(&mq_lock); | |
189 | return ns; | |
614b84cf SH |
190 | } |
191 | ||
d6629859 DL |
192 | /* Auxiliary functions to manipulate messages' list */ |
193 | static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info) | |
194 | { | |
195 | struct rb_node **p, *parent = NULL; | |
196 | struct posix_msg_tree_node *leaf; | |
a5091fda | 197 | bool rightmost = true; |
d6629859 DL |
198 | |
199 | p = &info->msg_tree.rb_node; | |
200 | while (*p) { | |
201 | parent = *p; | |
202 | leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); | |
203 | ||
204 | if (likely(leaf->priority == msg->m_type)) | |
205 | goto insert_msg; | |
a5091fda | 206 | else if (msg->m_type < leaf->priority) { |
d6629859 | 207 | p = &(*p)->rb_left; |
a5091fda DB |
208 | rightmost = false; |
209 | } else | |
d6629859 DL |
210 | p = &(*p)->rb_right; |
211 | } | |
ce2d52cc DL |
212 | if (info->node_cache) { |
213 | leaf = info->node_cache; | |
214 | info->node_cache = NULL; | |
215 | } else { | |
216 | leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC); | |
217 | if (!leaf) | |
218 | return -ENOMEM; | |
ce2d52cc | 219 | INIT_LIST_HEAD(&leaf->msg_list); |
ce2d52cc | 220 | } |
d6629859 | 221 | leaf->priority = msg->m_type; |
a5091fda DB |
222 | |
223 | if (rightmost) | |
224 | info->msg_tree_rightmost = &leaf->rb_node; | |
225 | ||
d6629859 DL |
226 | rb_link_node(&leaf->rb_node, parent, p); |
227 | rb_insert_color(&leaf->rb_node, &info->msg_tree); | |
d6629859 DL |
228 | insert_msg: |
229 | info->attr.mq_curmsgs++; | |
230 | info->qsize += msg->m_ts; | |
231 | list_add_tail(&msg->m_list, &leaf->msg_list); | |
232 | return 0; | |
233 | } | |
234 | ||
a5091fda DB |
235 | static inline void msg_tree_erase(struct posix_msg_tree_node *leaf, |
236 | struct mqueue_inode_info *info) | |
237 | { | |
238 | struct rb_node *node = &leaf->rb_node; | |
239 | ||
240 | if (info->msg_tree_rightmost == node) | |
241 | info->msg_tree_rightmost = rb_prev(node); | |
242 | ||
243 | rb_erase(node, &info->msg_tree); | |
43afe4d3 | 244 | if (info->node_cache) |
a5091fda | 245 | kfree(leaf); |
43afe4d3 | 246 | else |
a5091fda | 247 | info->node_cache = leaf; |
a5091fda DB |
248 | } |
249 | ||
d6629859 DL |
250 | static inline struct msg_msg *msg_get(struct mqueue_inode_info *info) |
251 | { | |
a5091fda | 252 | struct rb_node *parent = NULL; |
d6629859 DL |
253 | struct posix_msg_tree_node *leaf; |
254 | struct msg_msg *msg; | |
255 | ||
256 | try_again: | |
a5091fda DB |
257 | /* |
258 | * During insert, low priorities go to the left and high to the | |
259 | * right. On receive, we want the highest priorities first, so | |
260 | * walk all the way to the right. | |
261 | */ | |
262 | parent = info->msg_tree_rightmost; | |
d6629859 DL |
263 | if (!parent) { |
264 | if (info->attr.mq_curmsgs) { | |
265 | pr_warn_once("Inconsistency in POSIX message queue, " | |
266 | "no tree element, but supposedly messages " | |
267 | "should exist!\n"); | |
268 | info->attr.mq_curmsgs = 0; | |
269 | } | |
270 | return NULL; | |
271 | } | |
272 | leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); | |
ce2d52cc | 273 | if (unlikely(list_empty(&leaf->msg_list))) { |
d6629859 DL |
274 | pr_warn_once("Inconsistency in POSIX message queue, " |
275 | "empty leaf node but we haven't implemented " | |
276 | "lazy leaf delete!\n"); | |
a5091fda | 277 | msg_tree_erase(leaf, info); |
d6629859 DL |
278 | goto try_again; |
279 | } else { | |
280 | msg = list_first_entry(&leaf->msg_list, | |
281 | struct msg_msg, m_list); | |
282 | list_del(&msg->m_list); | |
283 | if (list_empty(&leaf->msg_list)) { | |
a5091fda | 284 | msg_tree_erase(leaf, info); |
d6629859 DL |
285 | } |
286 | } | |
287 | info->attr.mq_curmsgs--; | |
288 | info->qsize -= msg->m_ts; | |
289 | return msg; | |
290 | } | |
291 | ||
7eafd7c7 | 292 | static struct inode *mqueue_get_inode(struct super_block *sb, |
1b9d5ff7 | 293 | struct ipc_namespace *ipc_ns, umode_t mode, |
7eafd7c7 | 294 | struct mq_attr *attr) |
1da177e4 LT |
295 | { |
296 | struct inode *inode; | |
d40dcdb0 | 297 | int ret = -ENOMEM; |
1da177e4 LT |
298 | |
299 | inode = new_inode(sb); | |
04715206 JS |
300 | if (!inode) |
301 | goto err; | |
302 | ||
303 | inode->i_ino = get_next_ino(); | |
304 | inode->i_mode = mode; | |
305 | inode->i_uid = current_fsuid(); | |
306 | inode->i_gid = current_fsgid(); | |
078cd827 | 307 | inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode); |
04715206 JS |
308 | |
309 | if (S_ISREG(mode)) { | |
310 | struct mqueue_inode_info *info; | |
d6629859 | 311 | unsigned long mq_bytes, mq_treesize; |
04715206 JS |
312 | |
313 | inode->i_fop = &mqueue_file_operations; | |
314 | inode->i_size = FILENT_SIZE; | |
315 | /* mqueue specific info */ | |
316 | info = MQUEUE_I(inode); | |
317 | spin_lock_init(&info->lock); | |
318 | init_waitqueue_head(&info->wait_q); | |
319 | INIT_LIST_HEAD(&info->e_wait_q[0].list); | |
320 | INIT_LIST_HEAD(&info->e_wait_q[1].list); | |
321 | info->notify_owner = NULL; | |
6f9ac6d9 | 322 | info->notify_user_ns = NULL; |
04715206 | 323 | info->qsize = 0; |
6e52a9f0 | 324 | info->ucounts = NULL; /* set when all is ok */ |
d6629859 | 325 | info->msg_tree = RB_ROOT; |
a5091fda | 326 | info->msg_tree_rightmost = NULL; |
ce2d52cc | 327 | info->node_cache = NULL; |
04715206 | 328 | memset(&info->attr, 0, sizeof(info->attr)); |
cef0184c KM |
329 | info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max, |
330 | ipc_ns->mq_msg_default); | |
331 | info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max, | |
332 | ipc_ns->mq_msgsize_default); | |
04715206 JS |
333 | if (attr) { |
334 | info->attr.mq_maxmsg = attr->mq_maxmsg; | |
335 | info->attr.mq_msgsize = attr->mq_msgsize; | |
336 | } | |
d6629859 DL |
337 | /* |
338 | * We used to allocate a static array of pointers and account | |
339 | * the size of that array as well as one msg_msg struct per | |
340 | * possible message into the queue size. That's no longer | |
341 | * accurate as the queue is now an rbtree and will grow and | |
342 | * shrink depending on usage patterns. We can, however, still | |
343 | * account one msg_msg struct per message, but the nodes are | |
344 | * allocated depending on priority usage, and most programs | |
345 | * only use one, or a handful, of priorities. However, since | |
346 | * this is pinned memory, we need to assume worst case, so | |
347 | * that means the min(mq_maxmsg, max_priorities) * struct | |
348 | * posix_msg_tree_node. | |
349 | */ | |
05c1b290 AV |
350 | |
351 | ret = -EINVAL; | |
352 | if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0) | |
353 | goto out_inode; | |
354 | if (capable(CAP_SYS_RESOURCE)) { | |
355 | if (info->attr.mq_maxmsg > HARD_MSGMAX || | |
356 | info->attr.mq_msgsize > HARD_MSGSIZEMAX) | |
357 | goto out_inode; | |
358 | } else { | |
359 | if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max || | |
360 | info->attr.mq_msgsize > ipc_ns->mq_msgsize_max) | |
361 | goto out_inode; | |
362 | } | |
363 | ret = -EOVERFLOW; | |
364 | /* check for overflow */ | |
365 | if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg) | |
366 | goto out_inode; | |
d6629859 DL |
367 | mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + |
368 | min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * | |
369 | sizeof(struct posix_msg_tree_node); | |
05c1b290 AV |
370 | mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize; |
371 | if (mq_bytes + mq_treesize < mq_bytes) | |
372 | goto out_inode; | |
373 | mq_bytes += mq_treesize; | |
6e52a9f0 AG |
374 | info->ucounts = get_ucounts(current_ucounts()); |
375 | if (info->ucounts) { | |
376 | long msgqueue; | |
377 | ||
378 | spin_lock(&mq_lock); | |
379 | msgqueue = inc_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes); | |
380 | if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) { | |
381 | dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes); | |
382 | spin_unlock(&mq_lock); | |
383 | put_ucounts(info->ucounts); | |
384 | info->ucounts = NULL; | |
385 | /* mqueue_evict_inode() releases info->messages */ | |
386 | ret = -EMFILE; | |
387 | goto out_inode; | |
388 | } | |
04715206 | 389 | spin_unlock(&mq_lock); |
1da177e4 | 390 | } |
04715206 JS |
391 | } else if (S_ISDIR(mode)) { |
392 | inc_nlink(inode); | |
393 | /* Some things misbehave if size == 0 on a directory */ | |
394 | inode->i_size = 2 * DIRENT_SIZE; | |
395 | inode->i_op = &mqueue_dir_inode_operations; | |
396 | inode->i_fop = &simple_dir_operations; | |
1da177e4 | 397 | } |
04715206 | 398 | |
1da177e4 LT |
399 | return inode; |
400 | out_inode: | |
1da177e4 | 401 | iput(inode); |
04715206 | 402 | err: |
d40dcdb0 | 403 | return ERR_PTR(ret); |
1da177e4 LT |
404 | } |
405 | ||
935c6912 | 406 | static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 LT |
407 | { |
408 | struct inode *inode; | |
cfb2f6f6 | 409 | struct ipc_namespace *ns = sb->s_fs_info; |
1da177e4 | 410 | |
a2982cc9 | 411 | sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV; |
09cbfeaf KS |
412 | sb->s_blocksize = PAGE_SIZE; |
413 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
414 | sb->s_magic = MQUEUE_MAGIC; |
415 | sb->s_op = &mqueue_super_ops; | |
416 | ||
48fde701 AV |
417 | inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL); |
418 | if (IS_ERR(inode)) | |
419 | return PTR_ERR(inode); | |
1da177e4 | 420 | |
48fde701 AV |
421 | sb->s_root = d_make_root(inode); |
422 | if (!sb->s_root) | |
423 | return -ENOMEM; | |
424 | return 0; | |
1da177e4 LT |
425 | } |
426 | ||
935c6912 | 427 | static int mqueue_get_tree(struct fs_context *fc) |
1da177e4 | 428 | { |
935c6912 DH |
429 | struct mqueue_fs_context *ctx = fc->fs_private; |
430 | ||
d60c4d01 WL |
431 | /* |
432 | * With a newly created ipc namespace, we don't need to do a search | |
433 | * for an ipc namespace match, but we still need to set s_fs_info. | |
434 | */ | |
435 | if (ctx->newns) { | |
436 | fc->s_fs_info = ctx->ipc_ns; | |
437 | return get_tree_nodev(fc, mqueue_fill_super); | |
438 | } | |
533770cc | 439 | return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns); |
935c6912 DH |
440 | } |
441 | ||
442 | static void mqueue_fs_context_free(struct fs_context *fc) | |
443 | { | |
444 | struct mqueue_fs_context *ctx = fc->fs_private; | |
445 | ||
709a643d | 446 | put_ipc_ns(ctx->ipc_ns); |
935c6912 DH |
447 | kfree(ctx); |
448 | } | |
449 | ||
450 | static int mqueue_init_fs_context(struct fs_context *fc) | |
451 | { | |
452 | struct mqueue_fs_context *ctx; | |
453 | ||
454 | ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL); | |
455 | if (!ctx) | |
456 | return -ENOMEM; | |
457 | ||
458 | ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns); | |
709a643d AV |
459 | put_user_ns(fc->user_ns); |
460 | fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns); | |
935c6912 DH |
461 | fc->fs_private = ctx; |
462 | fc->ops = &mqueue_fs_context_ops; | |
463 | return 0; | |
464 | } | |
465 | ||
d60c4d01 WL |
466 | /* |
467 | * mq_init_ns() is currently the only caller of mq_create_mount(). | |
468 | * So the ns parameter is always a newly created ipc namespace. | |
469 | */ | |
935c6912 DH |
470 | static struct vfsmount *mq_create_mount(struct ipc_namespace *ns) |
471 | { | |
472 | struct mqueue_fs_context *ctx; | |
473 | struct fs_context *fc; | |
474 | struct vfsmount *mnt; | |
475 | ||
476 | fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT); | |
477 | if (IS_ERR(fc)) | |
478 | return ERR_CAST(fc); | |
479 | ||
480 | ctx = fc->fs_private; | |
d60c4d01 | 481 | ctx->newns = true; |
935c6912 DH |
482 | put_ipc_ns(ctx->ipc_ns); |
483 | ctx->ipc_ns = get_ipc_ns(ns); | |
709a643d AV |
484 | put_user_ns(fc->user_ns); |
485 | fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns); | |
935c6912 DH |
486 | |
487 | mnt = fc_mount(fc); | |
488 | put_fs_context(fc); | |
489 | return mnt; | |
1da177e4 LT |
490 | } |
491 | ||
51cc5068 | 492 | static void init_once(void *foo) |
1da177e4 LT |
493 | { |
494 | struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo; | |
495 | ||
a35afb83 | 496 | inode_init_once(&p->vfs_inode); |
1da177e4 LT |
497 | } |
498 | ||
499 | static struct inode *mqueue_alloc_inode(struct super_block *sb) | |
500 | { | |
501 | struct mqueue_inode_info *ei; | |
502 | ||
fd60b288 | 503 | ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL); |
1da177e4 LT |
504 | if (!ei) |
505 | return NULL; | |
506 | return &ei->vfs_inode; | |
507 | } | |
508 | ||
015d7956 | 509 | static void mqueue_free_inode(struct inode *inode) |
1da177e4 LT |
510 | { |
511 | kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode)); | |
512 | } | |
513 | ||
6d8af64c | 514 | static void mqueue_evict_inode(struct inode *inode) |
1da177e4 LT |
515 | { |
516 | struct mqueue_inode_info *info; | |
7eafd7c7 | 517 | struct ipc_namespace *ipc_ns; |
d6a2946a LR |
518 | struct msg_msg *msg, *nmsg; |
519 | LIST_HEAD(tmp_msg); | |
1da177e4 | 520 | |
dbd5768f | 521 | clear_inode(inode); |
6d8af64c AV |
522 | |
523 | if (S_ISDIR(inode->i_mode)) | |
1da177e4 | 524 | return; |
6d8af64c | 525 | |
7eafd7c7 | 526 | ipc_ns = get_ns_from_inode(inode); |
1da177e4 LT |
527 | info = MQUEUE_I(inode); |
528 | spin_lock(&info->lock); | |
d6629859 | 529 | while ((msg = msg_get(info)) != NULL) |
d6a2946a | 530 | list_add_tail(&msg->m_list, &tmp_msg); |
ce2d52cc | 531 | kfree(info->node_cache); |
1da177e4 LT |
532 | spin_unlock(&info->lock); |
533 | ||
d6a2946a LR |
534 | list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) { |
535 | list_del(&msg->m_list); | |
536 | free_msg(msg); | |
537 | } | |
538 | ||
6e52a9f0 | 539 | if (info->ucounts) { |
a318f12e KC |
540 | unsigned long mq_bytes, mq_treesize; |
541 | ||
542 | /* Total amount of bytes accounted for the mqueue */ | |
543 | mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + | |
544 | min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * | |
545 | sizeof(struct posix_msg_tree_node); | |
546 | ||
547 | mq_bytes = mq_treesize + (info->attr.mq_maxmsg * | |
548 | info->attr.mq_msgsize); | |
549 | ||
1da177e4 | 550 | spin_lock(&mq_lock); |
6e52a9f0 | 551 | dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes); |
7eafd7c7 SH |
552 | /* |
553 | * get_ns_from_inode() ensures that the | |
554 | * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns | |
555 | * to which we now hold a reference, or it is NULL. | |
556 | * We can't put it here under mq_lock, though. | |
557 | */ | |
558 | if (ipc_ns) | |
559 | ipc_ns->mq_queues_count--; | |
1da177e4 | 560 | spin_unlock(&mq_lock); |
6e52a9f0 AG |
561 | put_ucounts(info->ucounts); |
562 | info->ucounts = NULL; | |
1da177e4 | 563 | } |
7eafd7c7 SH |
564 | if (ipc_ns) |
565 | put_ipc_ns(ipc_ns); | |
1da177e4 LT |
566 | } |
567 | ||
eecec19d | 568 | static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg) |
1da177e4 | 569 | { |
eecec19d | 570 | struct inode *dir = dentry->d_parent->d_inode; |
1da177e4 | 571 | struct inode *inode; |
eecec19d | 572 | struct mq_attr *attr = arg; |
1da177e4 | 573 | int error; |
7eafd7c7 | 574 | struct ipc_namespace *ipc_ns; |
1da177e4 LT |
575 | |
576 | spin_lock(&mq_lock); | |
7eafd7c7 SH |
577 | ipc_ns = __get_ns_from_inode(dir); |
578 | if (!ipc_ns) { | |
579 | error = -EACCES; | |
580 | goto out_unlock; | |
581 | } | |
f3713fd9 DB |
582 | |
583 | if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max && | |
584 | !capable(CAP_SYS_RESOURCE)) { | |
1da177e4 | 585 | error = -ENOSPC; |
614b84cf | 586 | goto out_unlock; |
1da177e4 | 587 | } |
614b84cf | 588 | ipc_ns->mq_queues_count++; |
1da177e4 LT |
589 | spin_unlock(&mq_lock); |
590 | ||
7eafd7c7 | 591 | inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr); |
d40dcdb0 JS |
592 | if (IS_ERR(inode)) { |
593 | error = PTR_ERR(inode); | |
1da177e4 | 594 | spin_lock(&mq_lock); |
614b84cf SH |
595 | ipc_ns->mq_queues_count--; |
596 | goto out_unlock; | |
1da177e4 LT |
597 | } |
598 | ||
7eafd7c7 | 599 | put_ipc_ns(ipc_ns); |
1da177e4 | 600 | dir->i_size += DIRENT_SIZE; |
078cd827 | 601 | dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir); |
1da177e4 LT |
602 | |
603 | d_instantiate(dentry, inode); | |
604 | dget(dentry); | |
605 | return 0; | |
614b84cf | 606 | out_unlock: |
1da177e4 | 607 | spin_unlock(&mq_lock); |
7eafd7c7 SH |
608 | if (ipc_ns) |
609 | put_ipc_ns(ipc_ns); | |
1da177e4 LT |
610 | return error; |
611 | } | |
612 | ||
549c7297 CB |
613 | static int mqueue_create(struct user_namespace *mnt_userns, struct inode *dir, |
614 | struct dentry *dentry, umode_t mode, bool excl) | |
eecec19d AV |
615 | { |
616 | return mqueue_create_attr(dentry, mode, NULL); | |
617 | } | |
618 | ||
1da177e4 LT |
619 | static int mqueue_unlink(struct inode *dir, struct dentry *dentry) |
620 | { | |
75c3cfa8 | 621 | struct inode *inode = d_inode(dentry); |
1da177e4 | 622 | |
078cd827 | 623 | dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir); |
1da177e4 | 624 | dir->i_size -= DIRENT_SIZE; |
239521f3 MS |
625 | drop_nlink(inode); |
626 | dput(dentry); | |
627 | return 0; | |
1da177e4 LT |
628 | } |
629 | ||
630 | /* | |
631 | * This is routine for system read from queue file. | |
632 | * To avoid mess with doing here some sort of mq_receive we allow | |
633 | * to read only queue size & notification info (the only values | |
634 | * that are interesting from user point of view and aren't accessible | |
635 | * through std routines) | |
636 | */ | |
637 | static ssize_t mqueue_read_file(struct file *filp, char __user *u_data, | |
f1a43f93 | 638 | size_t count, loff_t *off) |
1da177e4 | 639 | { |
496ad9aa | 640 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); |
1da177e4 | 641 | char buffer[FILENT_SIZE]; |
f1a43f93 | 642 | ssize_t ret; |
1da177e4 LT |
643 | |
644 | spin_lock(&info->lock); | |
645 | snprintf(buffer, sizeof(buffer), | |
646 | "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n", | |
647 | info->qsize, | |
648 | info->notify_owner ? info->notify.sigev_notify : 0, | |
649 | (info->notify_owner && | |
650 | info->notify.sigev_notify == SIGEV_SIGNAL) ? | |
651 | info->notify.sigev_signo : 0, | |
6c5f3e7b | 652 | pid_vnr(info->notify_owner)); |
1da177e4 LT |
653 | spin_unlock(&info->lock); |
654 | buffer[sizeof(buffer)-1] = '\0'; | |
1da177e4 | 655 | |
f1a43f93 AM |
656 | ret = simple_read_from_buffer(u_data, count, off, buffer, |
657 | strlen(buffer)); | |
658 | if (ret <= 0) | |
659 | return ret; | |
1da177e4 | 660 | |
078cd827 | 661 | file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp)); |
f1a43f93 | 662 | return ret; |
1da177e4 LT |
663 | } |
664 | ||
75e1fcc0 | 665 | static int mqueue_flush_file(struct file *filp, fl_owner_t id) |
1da177e4 | 666 | { |
496ad9aa | 667 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); |
1da177e4 LT |
668 | |
669 | spin_lock(&info->lock); | |
a03fcb73 | 670 | if (task_tgid(current) == info->notify_owner) |
1da177e4 LT |
671 | remove_notification(info); |
672 | ||
673 | spin_unlock(&info->lock); | |
674 | return 0; | |
675 | } | |
676 | ||
9dd95748 | 677 | static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab) |
1da177e4 | 678 | { |
496ad9aa | 679 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); |
9dd95748 | 680 | __poll_t retval = 0; |
1da177e4 LT |
681 | |
682 | poll_wait(filp, &info->wait_q, poll_tab); | |
683 | ||
684 | spin_lock(&info->lock); | |
685 | if (info->attr.mq_curmsgs) | |
a9a08845 | 686 | retval = EPOLLIN | EPOLLRDNORM; |
1da177e4 LT |
687 | |
688 | if (info->attr.mq_curmsgs < info->attr.mq_maxmsg) | |
a9a08845 | 689 | retval |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 LT |
690 | spin_unlock(&info->lock); |
691 | ||
692 | return retval; | |
693 | } | |
694 | ||
695 | /* Adds current to info->e_wait_q[sr] before element with smaller prio */ | |
696 | static void wq_add(struct mqueue_inode_info *info, int sr, | |
697 | struct ext_wait_queue *ewp) | |
698 | { | |
699 | struct ext_wait_queue *walk; | |
700 | ||
1da177e4 | 701 | list_for_each_entry(walk, &info->e_wait_q[sr].list, list) { |
68e34f4e | 702 | if (walk->task->prio <= current->prio) { |
1da177e4 LT |
703 | list_add_tail(&ewp->list, &walk->list); |
704 | return; | |
705 | } | |
706 | } | |
707 | list_add_tail(&ewp->list, &info->e_wait_q[sr].list); | |
708 | } | |
709 | ||
710 | /* | |
711 | * Puts current task to sleep. Caller must hold queue lock. After return | |
712 | * lock isn't held. | |
713 | * sr: SEND or RECV | |
714 | */ | |
715 | static int wq_sleep(struct mqueue_inode_info *info, int sr, | |
9ca7d8e6 | 716 | ktime_t *timeout, struct ext_wait_queue *ewp) |
eac0b1c3 | 717 | __releases(&info->lock) |
1da177e4 LT |
718 | { |
719 | int retval; | |
720 | signed long time; | |
721 | ||
722 | wq_add(info, sr, ewp); | |
723 | ||
724 | for (;;) { | |
c5b2cbdb | 725 | /* memory barrier not required, we hold info->lock */ |
fa6004ad | 726 | __set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 LT |
727 | |
728 | spin_unlock(&info->lock); | |
32ea845d WG |
729 | time = schedule_hrtimeout_range_clock(timeout, 0, |
730 | HRTIMER_MODE_ABS, CLOCK_REALTIME); | |
1da177e4 | 731 | |
c5b2cbdb MS |
732 | if (READ_ONCE(ewp->state) == STATE_READY) { |
733 | /* see MQ_BARRIER for purpose/pairing */ | |
734 | smp_acquire__after_ctrl_dep(); | |
1da177e4 LT |
735 | retval = 0; |
736 | goto out; | |
737 | } | |
738 | spin_lock(&info->lock); | |
c5b2cbdb MS |
739 | |
740 | /* we hold info->lock, so no memory barrier required */ | |
741 | if (READ_ONCE(ewp->state) == STATE_READY) { | |
1da177e4 LT |
742 | retval = 0; |
743 | goto out_unlock; | |
744 | } | |
745 | if (signal_pending(current)) { | |
746 | retval = -ERESTARTSYS; | |
747 | break; | |
748 | } | |
749 | if (time == 0) { | |
750 | retval = -ETIMEDOUT; | |
751 | break; | |
752 | } | |
753 | } | |
754 | list_del(&ewp->list); | |
755 | out_unlock: | |
756 | spin_unlock(&info->lock); | |
757 | out: | |
758 | return retval; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Returns waiting task that should be serviced first or NULL if none exists | |
763 | */ | |
764 | static struct ext_wait_queue *wq_get_first_waiter( | |
765 | struct mqueue_inode_info *info, int sr) | |
766 | { | |
767 | struct list_head *ptr; | |
768 | ||
769 | ptr = info->e_wait_q[sr].list.prev; | |
770 | if (ptr == &info->e_wait_q[sr].list) | |
771 | return NULL; | |
772 | return list_entry(ptr, struct ext_wait_queue, list); | |
773 | } | |
774 | ||
1da177e4 LT |
775 | |
776 | static inline void set_cookie(struct sk_buff *skb, char code) | |
777 | { | |
239521f3 | 778 | ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code; |
1da177e4 LT |
779 | } |
780 | ||
781 | /* | |
782 | * The next function is only to split too long sys_mq_timedsend | |
783 | */ | |
784 | static void __do_notify(struct mqueue_inode_info *info) | |
785 | { | |
786 | /* notification | |
787 | * invoked when there is registered process and there isn't process | |
788 | * waiting synchronously for message AND state of queue changed from | |
789 | * empty to not empty. Here we are sure that no one is waiting | |
790 | * synchronously. */ | |
791 | if (info->notify_owner && | |
792 | info->attr.mq_curmsgs == 1) { | |
1da177e4 LT |
793 | switch (info->notify.sigev_notify) { |
794 | case SIGEV_NONE: | |
795 | break; | |
b5f20061 ON |
796 | case SIGEV_SIGNAL: { |
797 | struct kernel_siginfo sig_i; | |
798 | struct task_struct *task; | |
799 | ||
800 | /* do_mq_notify() accepts sigev_signo == 0, why?? */ | |
801 | if (!info->notify.sigev_signo) | |
802 | break; | |
1da177e4 | 803 | |
faf1f22b | 804 | clear_siginfo(&sig_i); |
1da177e4 LT |
805 | sig_i.si_signo = info->notify.sigev_signo; |
806 | sig_i.si_errno = 0; | |
807 | sig_i.si_code = SI_MESGQ; | |
808 | sig_i.si_value = info->notify.sigev_value; | |
6b550f94 | 809 | rcu_read_lock(); |
b5f20061 | 810 | /* map current pid/uid into info->owner's namespaces */ |
a6684999 SB |
811 | sig_i.si_pid = task_tgid_nr_ns(current, |
812 | ns_of_pid(info->notify_owner)); | |
b5f20061 ON |
813 | sig_i.si_uid = from_kuid_munged(info->notify_user_ns, |
814 | current_uid()); | |
815 | /* | |
816 | * We can't use kill_pid_info(), this signal should | |
817 | * bypass check_kill_permission(). It is from kernel | |
818 | * but si_fromuser() can't know this. | |
819 | * We do check the self_exec_id, to avoid sending | |
820 | * signals to programs that don't expect them. | |
821 | */ | |
822 | task = pid_task(info->notify_owner, PIDTYPE_TGID); | |
823 | if (task && task->self_exec_id == | |
824 | info->notify_self_exec_id) { | |
825 | do_send_sig_info(info->notify.sigev_signo, | |
826 | &sig_i, task, PIDTYPE_TGID); | |
827 | } | |
6b550f94 | 828 | rcu_read_unlock(); |
1da177e4 | 829 | break; |
b5f20061 | 830 | } |
1da177e4 LT |
831 | case SIGEV_THREAD: |
832 | set_cookie(info->notify_cookie, NOTIFY_WOKENUP); | |
7ee015e0 | 833 | netlink_sendskb(info->notify_sock, info->notify_cookie); |
1da177e4 LT |
834 | break; |
835 | } | |
836 | /* after notification unregisters process */ | |
a03fcb73 | 837 | put_pid(info->notify_owner); |
6f9ac6d9 | 838 | put_user_ns(info->notify_user_ns); |
a03fcb73 | 839 | info->notify_owner = NULL; |
6f9ac6d9 | 840 | info->notify_user_ns = NULL; |
1da177e4 LT |
841 | } |
842 | wake_up(&info->wait_q); | |
843 | } | |
844 | ||
21fc538d | 845 | static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout, |
b9047726 | 846 | struct timespec64 *ts) |
1da177e4 | 847 | { |
b9047726 | 848 | if (get_timespec64(ts, u_abs_timeout)) |
9ca7d8e6 | 849 | return -EFAULT; |
b9047726 | 850 | if (!timespec64_valid(ts)) |
9ca7d8e6 | 851 | return -EINVAL; |
9ca7d8e6 | 852 | return 0; |
1da177e4 LT |
853 | } |
854 | ||
855 | static void remove_notification(struct mqueue_inode_info *info) | |
856 | { | |
a03fcb73 | 857 | if (info->notify_owner != NULL && |
1da177e4 LT |
858 | info->notify.sigev_notify == SIGEV_THREAD) { |
859 | set_cookie(info->notify_cookie, NOTIFY_REMOVED); | |
7ee015e0 | 860 | netlink_sendskb(info->notify_sock, info->notify_cookie); |
1da177e4 | 861 | } |
a03fcb73 | 862 | put_pid(info->notify_owner); |
6f9ac6d9 | 863 | put_user_ns(info->notify_user_ns); |
a03fcb73 | 864 | info->notify_owner = NULL; |
6f9ac6d9 | 865 | info->notify_user_ns = NULL; |
1da177e4 LT |
866 | } |
867 | ||
066cc813 AV |
868 | static int prepare_open(struct dentry *dentry, int oflag, int ro, |
869 | umode_t mode, struct filename *name, | |
614b84cf | 870 | struct mq_attr *attr) |
1da177e4 | 871 | { |
745ca247 DH |
872 | static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE, |
873 | MAY_READ | MAY_WRITE }; | |
765927b2 | 874 | int acc; |
066cc813 | 875 | |
9b20d7fc AV |
876 | if (d_really_is_negative(dentry)) { |
877 | if (!(oflag & O_CREAT)) | |
066cc813 | 878 | return -ENOENT; |
9b20d7fc AV |
879 | if (ro) |
880 | return ro; | |
881 | audit_inode_parent_hidden(name, dentry->d_parent); | |
882 | return vfs_mkobj(dentry, mode & ~current_umask(), | |
883 | mqueue_create_attr, attr); | |
066cc813 | 884 | } |
9b20d7fc AV |
885 | /* it already existed */ |
886 | audit_inode(name, dentry, 0); | |
887 | if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
888 | return -EEXIST; | |
765927b2 | 889 | if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) |
af4a5372 | 890 | return -EINVAL; |
765927b2 | 891 | acc = oflag2acc[oflag & O_ACCMODE]; |
47291baa | 892 | return inode_permission(&init_user_ns, d_inode(dentry), acc); |
1da177e4 LT |
893 | } |
894 | ||
0d060606 AV |
895 | static int do_mq_open(const char __user *u_name, int oflag, umode_t mode, |
896 | struct mq_attr *attr) | |
1da177e4 | 897 | { |
cfb2f6f6 EB |
898 | struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt; |
899 | struct dentry *root = mnt->mnt_root; | |
91a27b2a | 900 | struct filename *name; |
a713fd7f | 901 | struct path path; |
1da177e4 | 902 | int fd, error; |
312b90fb | 903 | int ro; |
1da177e4 | 904 | |
0d060606 | 905 | audit_mq_open(oflag, mode, attr); |
20ca73bc | 906 | |
1da177e4 LT |
907 | if (IS_ERR(name = getname(u_name))) |
908 | return PTR_ERR(name); | |
909 | ||
269f2134 | 910 | fd = get_unused_fd_flags(O_CLOEXEC); |
1da177e4 LT |
911 | if (fd < 0) |
912 | goto out_putname; | |
913 | ||
312b90fb | 914 | ro = mnt_want_write(mnt); /* we'll drop it in any case */ |
5955102c | 915 | inode_lock(d_inode(root)); |
91a27b2a | 916 | path.dentry = lookup_one_len(name->name, root, strlen(name->name)); |
765927b2 AV |
917 | if (IS_ERR(path.dentry)) { |
918 | error = PTR_ERR(path.dentry); | |
4294a8ee | 919 | goto out_putfd; |
1da177e4 | 920 | } |
312b90fb | 921 | path.mnt = mntget(mnt); |
066cc813 AV |
922 | error = prepare_open(path.dentry, oflag, ro, mode, name, attr); |
923 | if (!error) { | |
924 | struct file *file = dentry_open(&path, oflag, current_cred()); | |
925 | if (!IS_ERR(file)) | |
926 | fd_install(fd, file); | |
927 | else | |
928 | error = PTR_ERR(file); | |
7c7dce92 | 929 | } |
765927b2 | 930 | path_put(&path); |
7c7dce92 | 931 | out_putfd: |
765927b2 AV |
932 | if (error) { |
933 | put_unused_fd(fd); | |
934 | fd = error; | |
935 | } | |
5955102c | 936 | inode_unlock(d_inode(root)); |
38d78e58 VD |
937 | if (!ro) |
938 | mnt_drop_write(mnt); | |
1da177e4 LT |
939 | out_putname: |
940 | putname(name); | |
941 | return fd; | |
942 | } | |
943 | ||
0d060606 AV |
944 | SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode, |
945 | struct mq_attr __user *, u_attr) | |
946 | { | |
947 | struct mq_attr attr; | |
948 | if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr))) | |
949 | return -EFAULT; | |
950 | ||
951 | return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL); | |
952 | } | |
953 | ||
d5460c99 | 954 | SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name) |
1da177e4 LT |
955 | { |
956 | int err; | |
91a27b2a | 957 | struct filename *name; |
1da177e4 LT |
958 | struct dentry *dentry; |
959 | struct inode *inode = NULL; | |
7eafd7c7 | 960 | struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns; |
312b90fb | 961 | struct vfsmount *mnt = ipc_ns->mq_mnt; |
1da177e4 LT |
962 | |
963 | name = getname(u_name); | |
964 | if (IS_ERR(name)) | |
965 | return PTR_ERR(name); | |
966 | ||
79f6530c | 967 | audit_inode_parent_hidden(name, mnt->mnt_root); |
312b90fb AV |
968 | err = mnt_want_write(mnt); |
969 | if (err) | |
970 | goto out_name; | |
5955102c | 971 | inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT); |
91a27b2a JL |
972 | dentry = lookup_one_len(name->name, mnt->mnt_root, |
973 | strlen(name->name)); | |
1da177e4 LT |
974 | if (IS_ERR(dentry)) { |
975 | err = PTR_ERR(dentry); | |
976 | goto out_unlock; | |
977 | } | |
978 | ||
75c3cfa8 | 979 | inode = d_inode(dentry); |
312b90fb AV |
980 | if (!inode) { |
981 | err = -ENOENT; | |
982 | } else { | |
7de9c6ee | 983 | ihold(inode); |
6521f891 CB |
984 | err = vfs_unlink(&init_user_ns, d_inode(dentry->d_parent), |
985 | dentry, NULL); | |
312b90fb | 986 | } |
1da177e4 LT |
987 | dput(dentry); |
988 | ||
989 | out_unlock: | |
5955102c | 990 | inode_unlock(d_inode(mnt->mnt_root)); |
1da177e4 LT |
991 | if (inode) |
992 | iput(inode); | |
312b90fb AV |
993 | mnt_drop_write(mnt); |
994 | out_name: | |
995 | putname(name); | |
1da177e4 LT |
996 | |
997 | return err; | |
998 | } | |
999 | ||
1000 | /* Pipelined send and receive functions. | |
1001 | * | |
1002 | * If a receiver finds no waiting message, then it registers itself in the | |
1003 | * list of waiting receivers. A sender checks that list before adding the new | |
1004 | * message into the message array. If there is a waiting receiver, then it | |
1005 | * bypasses the message array and directly hands the message over to the | |
fa6004ad DB |
1006 | * receiver. The receiver accepts the message and returns without grabbing the |
1007 | * queue spinlock: | |
1008 | * | |
1009 | * - Set pointer to message. | |
1010 | * - Queue the receiver task for later wakeup (without the info->lock). | |
1011 | * - Update its state to STATE_READY. Now the receiver can continue. | |
1012 | * - Wake up the process after the lock is dropped. Should the process wake up | |
1013 | * before this wakeup (due to a timeout or a signal) it will either see | |
1014 | * STATE_READY and continue or acquire the lock to check the state again. | |
1da177e4 LT |
1015 | * |
1016 | * The same algorithm is used for senders. | |
1017 | */ | |
1018 | ||
ed29f171 | 1019 | static inline void __pipelined_op(struct wake_q_head *wake_q, |
fa6004ad | 1020 | struct mqueue_inode_info *info, |
ed29f171 | 1021 | struct ext_wait_queue *this) |
1da177e4 | 1022 | { |
a11ddb37 VG |
1023 | struct task_struct *task; |
1024 | ||
ed29f171 | 1025 | list_del(&this->list); |
a11ddb37 | 1026 | task = get_task_struct(this->task); |
c5b2cbdb MS |
1027 | |
1028 | /* see MQ_BARRIER for purpose/pairing */ | |
1029 | smp_store_release(&this->state, STATE_READY); | |
a11ddb37 | 1030 | wake_q_add_safe(wake_q, task); |
ed29f171 DB |
1031 | } |
1032 | ||
1033 | /* pipelined_send() - send a message directly to the task waiting in | |
1034 | * sys_mq_timedreceive() (without inserting message into a queue). | |
1035 | */ | |
1036 | static inline void pipelined_send(struct wake_q_head *wake_q, | |
1037 | struct mqueue_inode_info *info, | |
1038 | struct msg_msg *message, | |
1039 | struct ext_wait_queue *receiver) | |
1040 | { | |
1041 | receiver->msg = message; | |
1042 | __pipelined_op(wake_q, info, receiver); | |
1da177e4 LT |
1043 | } |
1044 | ||
1045 | /* pipelined_receive() - if there is task waiting in sys_mq_timedsend() | |
1046 | * gets its message and put to the queue (we have one free place for sure). */ | |
fa6004ad DB |
1047 | static inline void pipelined_receive(struct wake_q_head *wake_q, |
1048 | struct mqueue_inode_info *info) | |
1da177e4 LT |
1049 | { |
1050 | struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND); | |
1051 | ||
1052 | if (!sender) { | |
1053 | /* for poll */ | |
1054 | wake_up_interruptible(&info->wait_q); | |
1055 | return; | |
1056 | } | |
d6629859 DL |
1057 | if (msg_insert(sender->msg, info)) |
1058 | return; | |
fa6004ad | 1059 | |
ed29f171 | 1060 | __pipelined_op(wake_q, info, sender); |
1da177e4 LT |
1061 | } |
1062 | ||
0d060606 AV |
1063 | static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr, |
1064 | size_t msg_len, unsigned int msg_prio, | |
b9047726 | 1065 | struct timespec64 *ts) |
1da177e4 | 1066 | { |
2903ff01 | 1067 | struct fd f; |
1da177e4 LT |
1068 | struct inode *inode; |
1069 | struct ext_wait_queue wait; | |
1070 | struct ext_wait_queue *receiver; | |
1071 | struct msg_msg *msg_ptr; | |
1072 | struct mqueue_inode_info *info; | |
9ca7d8e6 | 1073 | ktime_t expires, *timeout = NULL; |
ce2d52cc | 1074 | struct posix_msg_tree_node *new_leaf = NULL; |
2903ff01 | 1075 | int ret = 0; |
194a6b5b | 1076 | DEFINE_WAKE_Q(wake_q); |
1da177e4 LT |
1077 | |
1078 | if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX)) | |
1079 | return -EINVAL; | |
1080 | ||
0d060606 | 1081 | if (ts) { |
b9047726 | 1082 | expires = timespec64_to_ktime(*ts); |
0d060606 AV |
1083 | timeout = &expires; |
1084 | } | |
1085 | ||
1086 | audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts); | |
1da177e4 | 1087 | |
2903ff01 AV |
1088 | f = fdget(mqdes); |
1089 | if (unlikely(!f.file)) { | |
8d8ffefa | 1090 | ret = -EBADF; |
1da177e4 | 1091 | goto out; |
8d8ffefa | 1092 | } |
1da177e4 | 1093 | |
496ad9aa | 1094 | inode = file_inode(f.file); |
2903ff01 | 1095 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
8d8ffefa | 1096 | ret = -EBADF; |
1da177e4 | 1097 | goto out_fput; |
8d8ffefa | 1098 | } |
1da177e4 | 1099 | info = MQUEUE_I(inode); |
9f45f5bf | 1100 | audit_file(f.file); |
1da177e4 | 1101 | |
2903ff01 | 1102 | if (unlikely(!(f.file->f_mode & FMODE_WRITE))) { |
8d8ffefa | 1103 | ret = -EBADF; |
1da177e4 | 1104 | goto out_fput; |
8d8ffefa | 1105 | } |
1da177e4 LT |
1106 | |
1107 | if (unlikely(msg_len > info->attr.mq_msgsize)) { | |
1108 | ret = -EMSGSIZE; | |
1109 | goto out_fput; | |
1110 | } | |
1111 | ||
1112 | /* First try to allocate memory, before doing anything with | |
1113 | * existing queues. */ | |
1114 | msg_ptr = load_msg(u_msg_ptr, msg_len); | |
1115 | if (IS_ERR(msg_ptr)) { | |
1116 | ret = PTR_ERR(msg_ptr); | |
1117 | goto out_fput; | |
1118 | } | |
1119 | msg_ptr->m_ts = msg_len; | |
1120 | msg_ptr->m_type = msg_prio; | |
1121 | ||
ce2d52cc DL |
1122 | /* |
1123 | * msg_insert really wants us to have a valid, spare node struct so | |
1124 | * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will | |
1125 | * fall back to that if necessary. | |
1126 | */ | |
1127 | if (!info->node_cache) | |
1128 | new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL); | |
1129 | ||
1da177e4 LT |
1130 | spin_lock(&info->lock); |
1131 | ||
ce2d52cc DL |
1132 | if (!info->node_cache && new_leaf) { |
1133 | /* Save our speculative allocation into the cache */ | |
ce2d52cc DL |
1134 | INIT_LIST_HEAD(&new_leaf->msg_list); |
1135 | info->node_cache = new_leaf; | |
ce2d52cc DL |
1136 | new_leaf = NULL; |
1137 | } else { | |
1138 | kfree(new_leaf); | |
1139 | } | |
1140 | ||
1da177e4 | 1141 | if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) { |
2903ff01 | 1142 | if (f.file->f_flags & O_NONBLOCK) { |
1da177e4 | 1143 | ret = -EAGAIN; |
1da177e4 LT |
1144 | } else { |
1145 | wait.task = current; | |
1146 | wait.msg = (void *) msg_ptr; | |
c5b2cbdb MS |
1147 | |
1148 | /* memory barrier not required, we hold info->lock */ | |
1149 | WRITE_ONCE(wait.state, STATE_NONE); | |
1da177e4 | 1150 | ret = wq_sleep(info, SEND, timeout, &wait); |
ce2d52cc DL |
1151 | /* |
1152 | * wq_sleep must be called with info->lock held, and | |
1153 | * returns with the lock released | |
1154 | */ | |
1155 | goto out_free; | |
1da177e4 | 1156 | } |
1da177e4 LT |
1157 | } else { |
1158 | receiver = wq_get_first_waiter(info, RECV); | |
1159 | if (receiver) { | |
fa6004ad | 1160 | pipelined_send(&wake_q, info, msg_ptr, receiver); |
1da177e4 LT |
1161 | } else { |
1162 | /* adds message to the queue */ | |
ce2d52cc DL |
1163 | ret = msg_insert(msg_ptr, info); |
1164 | if (ret) | |
1165 | goto out_unlock; | |
1da177e4 LT |
1166 | __do_notify(info); |
1167 | } | |
1168 | inode->i_atime = inode->i_mtime = inode->i_ctime = | |
078cd827 | 1169 | current_time(inode); |
1da177e4 | 1170 | } |
ce2d52cc DL |
1171 | out_unlock: |
1172 | spin_unlock(&info->lock); | |
fa6004ad | 1173 | wake_up_q(&wake_q); |
ce2d52cc DL |
1174 | out_free: |
1175 | if (ret) | |
1176 | free_msg(msg_ptr); | |
1da177e4 | 1177 | out_fput: |
2903ff01 | 1178 | fdput(f); |
1da177e4 LT |
1179 | out: |
1180 | return ret; | |
1181 | } | |
1182 | ||
0d060606 AV |
1183 | static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr, |
1184 | size_t msg_len, unsigned int __user *u_msg_prio, | |
b9047726 | 1185 | struct timespec64 *ts) |
1da177e4 | 1186 | { |
1da177e4 LT |
1187 | ssize_t ret; |
1188 | struct msg_msg *msg_ptr; | |
2903ff01 | 1189 | struct fd f; |
1da177e4 LT |
1190 | struct inode *inode; |
1191 | struct mqueue_inode_info *info; | |
1192 | struct ext_wait_queue wait; | |
9ca7d8e6 | 1193 | ktime_t expires, *timeout = NULL; |
ce2d52cc | 1194 | struct posix_msg_tree_node *new_leaf = NULL; |
1da177e4 | 1195 | |
0d060606 | 1196 | if (ts) { |
b9047726 | 1197 | expires = timespec64_to_ktime(*ts); |
9ca7d8e6 | 1198 | timeout = &expires; |
c32c8af4 | 1199 | } |
20ca73bc | 1200 | |
0d060606 | 1201 | audit_mq_sendrecv(mqdes, msg_len, 0, ts); |
1da177e4 | 1202 | |
2903ff01 AV |
1203 | f = fdget(mqdes); |
1204 | if (unlikely(!f.file)) { | |
8d8ffefa | 1205 | ret = -EBADF; |
1da177e4 | 1206 | goto out; |
8d8ffefa | 1207 | } |
1da177e4 | 1208 | |
496ad9aa | 1209 | inode = file_inode(f.file); |
2903ff01 | 1210 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
8d8ffefa | 1211 | ret = -EBADF; |
1da177e4 | 1212 | goto out_fput; |
8d8ffefa | 1213 | } |
1da177e4 | 1214 | info = MQUEUE_I(inode); |
9f45f5bf | 1215 | audit_file(f.file); |
1da177e4 | 1216 | |
2903ff01 | 1217 | if (unlikely(!(f.file->f_mode & FMODE_READ))) { |
8d8ffefa | 1218 | ret = -EBADF; |
1da177e4 | 1219 | goto out_fput; |
8d8ffefa | 1220 | } |
1da177e4 LT |
1221 | |
1222 | /* checks if buffer is big enough */ | |
1223 | if (unlikely(msg_len < info->attr.mq_msgsize)) { | |
1224 | ret = -EMSGSIZE; | |
1225 | goto out_fput; | |
1226 | } | |
1227 | ||
ce2d52cc DL |
1228 | /* |
1229 | * msg_insert really wants us to have a valid, spare node struct so | |
1230 | * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will | |
1231 | * fall back to that if necessary. | |
1232 | */ | |
1233 | if (!info->node_cache) | |
1234 | new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL); | |
1235 | ||
1da177e4 | 1236 | spin_lock(&info->lock); |
ce2d52cc DL |
1237 | |
1238 | if (!info->node_cache && new_leaf) { | |
1239 | /* Save our speculative allocation into the cache */ | |
ce2d52cc DL |
1240 | INIT_LIST_HEAD(&new_leaf->msg_list); |
1241 | info->node_cache = new_leaf; | |
ce2d52cc DL |
1242 | } else { |
1243 | kfree(new_leaf); | |
1244 | } | |
1245 | ||
1da177e4 | 1246 | if (info->attr.mq_curmsgs == 0) { |
2903ff01 | 1247 | if (f.file->f_flags & O_NONBLOCK) { |
1da177e4 LT |
1248 | spin_unlock(&info->lock); |
1249 | ret = -EAGAIN; | |
1da177e4 LT |
1250 | } else { |
1251 | wait.task = current; | |
c5b2cbdb MS |
1252 | |
1253 | /* memory barrier not required, we hold info->lock */ | |
1254 | WRITE_ONCE(wait.state, STATE_NONE); | |
1da177e4 LT |
1255 | ret = wq_sleep(info, RECV, timeout, &wait); |
1256 | msg_ptr = wait.msg; | |
1257 | } | |
1258 | } else { | |
194a6b5b | 1259 | DEFINE_WAKE_Q(wake_q); |
fa6004ad | 1260 | |
1da177e4 LT |
1261 | msg_ptr = msg_get(info); |
1262 | ||
1263 | inode->i_atime = inode->i_mtime = inode->i_ctime = | |
078cd827 | 1264 | current_time(inode); |
1da177e4 LT |
1265 | |
1266 | /* There is now free space in queue. */ | |
fa6004ad | 1267 | pipelined_receive(&wake_q, info); |
1da177e4 | 1268 | spin_unlock(&info->lock); |
fa6004ad | 1269 | wake_up_q(&wake_q); |
1da177e4 LT |
1270 | ret = 0; |
1271 | } | |
1272 | if (ret == 0) { | |
1273 | ret = msg_ptr->m_ts; | |
1274 | ||
1275 | if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) || | |
1276 | store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) { | |
1277 | ret = -EFAULT; | |
1278 | } | |
1279 | free_msg(msg_ptr); | |
1280 | } | |
1281 | out_fput: | |
2903ff01 | 1282 | fdput(f); |
1da177e4 LT |
1283 | out: |
1284 | return ret; | |
1285 | } | |
1286 | ||
0d060606 AV |
1287 | SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr, |
1288 | size_t, msg_len, unsigned int, msg_prio, | |
21fc538d | 1289 | const struct __kernel_timespec __user *, u_abs_timeout) |
0d060606 | 1290 | { |
b9047726 | 1291 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1292 | if (u_abs_timeout) { |
1293 | int res = prepare_timeout(u_abs_timeout, &ts); | |
1294 | if (res) | |
1295 | return res; | |
1296 | p = &ts; | |
1297 | } | |
1298 | return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p); | |
1299 | } | |
1300 | ||
1301 | SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr, | |
1302 | size_t, msg_len, unsigned int __user *, u_msg_prio, | |
21fc538d | 1303 | const struct __kernel_timespec __user *, u_abs_timeout) |
0d060606 | 1304 | { |
b9047726 | 1305 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1306 | if (u_abs_timeout) { |
1307 | int res = prepare_timeout(u_abs_timeout, &ts); | |
1308 | if (res) | |
1309 | return res; | |
1310 | p = &ts; | |
1311 | } | |
1312 | return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p); | |
1313 | } | |
1314 | ||
1da177e4 LT |
1315 | /* |
1316 | * Notes: the case when user wants us to deregister (with NULL as pointer) | |
1317 | * and he isn't currently owner of notification, will be silently discarded. | |
1318 | * It isn't explicitly defined in the POSIX. | |
1319 | */ | |
0d060606 | 1320 | static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification) |
1da177e4 | 1321 | { |
2903ff01 AV |
1322 | int ret; |
1323 | struct fd f; | |
1da177e4 LT |
1324 | struct sock *sock; |
1325 | struct inode *inode; | |
1da177e4 LT |
1326 | struct mqueue_inode_info *info; |
1327 | struct sk_buff *nc; | |
1328 | ||
0d060606 | 1329 | audit_mq_notify(mqdes, notification); |
1da177e4 | 1330 | |
20114f71 AV |
1331 | nc = NULL; |
1332 | sock = NULL; | |
0d060606 AV |
1333 | if (notification != NULL) { |
1334 | if (unlikely(notification->sigev_notify != SIGEV_NONE && | |
1335 | notification->sigev_notify != SIGEV_SIGNAL && | |
1336 | notification->sigev_notify != SIGEV_THREAD)) | |
1da177e4 | 1337 | return -EINVAL; |
0d060606 AV |
1338 | if (notification->sigev_notify == SIGEV_SIGNAL && |
1339 | !valid_signal(notification->sigev_signo)) { | |
1da177e4 LT |
1340 | return -EINVAL; |
1341 | } | |
0d060606 | 1342 | if (notification->sigev_notify == SIGEV_THREAD) { |
c3d8d1e3 PM |
1343 | long timeo; |
1344 | ||
1da177e4 LT |
1345 | /* create the notify skb */ |
1346 | nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL); | |
c231740d ME |
1347 | if (!nc) |
1348 | return -ENOMEM; | |
1349 | ||
1da177e4 | 1350 | if (copy_from_user(nc->data, |
0d060606 | 1351 | notification->sigev_value.sival_ptr, |
1da177e4 | 1352 | NOTIFY_COOKIE_LEN)) { |
8d8ffefa | 1353 | ret = -EFAULT; |
c231740d | 1354 | goto free_skb; |
1da177e4 LT |
1355 | } |
1356 | ||
1357 | /* TODO: add a header? */ | |
1358 | skb_put(nc, NOTIFY_COOKIE_LEN); | |
1359 | /* and attach it to the socket */ | |
1360 | retry: | |
0d060606 | 1361 | f = fdget(notification->sigev_signo); |
2903ff01 | 1362 | if (!f.file) { |
8d8ffefa | 1363 | ret = -EBADF; |
1da177e4 | 1364 | goto out; |
8d8ffefa | 1365 | } |
2903ff01 AV |
1366 | sock = netlink_getsockbyfilp(f.file); |
1367 | fdput(f); | |
1da177e4 LT |
1368 | if (IS_ERR(sock)) { |
1369 | ret = PTR_ERR(sock); | |
c231740d | 1370 | goto free_skb; |
1da177e4 LT |
1371 | } |
1372 | ||
c3d8d1e3 | 1373 | timeo = MAX_SCHEDULE_TIMEOUT; |
9457afee | 1374 | ret = netlink_attachskb(sock, nc, &timeo, NULL); |
f991af3d CW |
1375 | if (ret == 1) { |
1376 | sock = NULL; | |
8d8ffefa | 1377 | goto retry; |
f991af3d | 1378 | } |
c231740d ME |
1379 | if (ret) |
1380 | return ret; | |
1da177e4 LT |
1381 | } |
1382 | } | |
1383 | ||
2903ff01 AV |
1384 | f = fdget(mqdes); |
1385 | if (!f.file) { | |
8d8ffefa | 1386 | ret = -EBADF; |
1da177e4 | 1387 | goto out; |
8d8ffefa | 1388 | } |
1da177e4 | 1389 | |
496ad9aa | 1390 | inode = file_inode(f.file); |
2903ff01 | 1391 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
8d8ffefa | 1392 | ret = -EBADF; |
1da177e4 | 1393 | goto out_fput; |
8d8ffefa | 1394 | } |
1da177e4 LT |
1395 | info = MQUEUE_I(inode); |
1396 | ||
1397 | ret = 0; | |
1398 | spin_lock(&info->lock); | |
0d060606 | 1399 | if (notification == NULL) { |
a03fcb73 | 1400 | if (info->notify_owner == task_tgid(current)) { |
1da177e4 | 1401 | remove_notification(info); |
078cd827 | 1402 | inode->i_atime = inode->i_ctime = current_time(inode); |
1da177e4 | 1403 | } |
a03fcb73 | 1404 | } else if (info->notify_owner != NULL) { |
1da177e4 LT |
1405 | ret = -EBUSY; |
1406 | } else { | |
0d060606 | 1407 | switch (notification->sigev_notify) { |
1da177e4 LT |
1408 | case SIGEV_NONE: |
1409 | info->notify.sigev_notify = SIGEV_NONE; | |
1410 | break; | |
1411 | case SIGEV_THREAD: | |
1412 | info->notify_sock = sock; | |
1413 | info->notify_cookie = nc; | |
1414 | sock = NULL; | |
1415 | nc = NULL; | |
1416 | info->notify.sigev_notify = SIGEV_THREAD; | |
1417 | break; | |
1418 | case SIGEV_SIGNAL: | |
0d060606 AV |
1419 | info->notify.sigev_signo = notification->sigev_signo; |
1420 | info->notify.sigev_value = notification->sigev_value; | |
1da177e4 | 1421 | info->notify.sigev_notify = SIGEV_SIGNAL; |
b5f20061 | 1422 | info->notify_self_exec_id = current->self_exec_id; |
1da177e4 LT |
1423 | break; |
1424 | } | |
a03fcb73 CLG |
1425 | |
1426 | info->notify_owner = get_pid(task_tgid(current)); | |
6f9ac6d9 | 1427 | info->notify_user_ns = get_user_ns(current_user_ns()); |
078cd827 | 1428 | inode->i_atime = inode->i_ctime = current_time(inode); |
1da177e4 LT |
1429 | } |
1430 | spin_unlock(&info->lock); | |
1431 | out_fput: | |
2903ff01 | 1432 | fdput(f); |
1da177e4 | 1433 | out: |
3ab08fe2 | 1434 | if (sock) |
1da177e4 | 1435 | netlink_detachskb(sock, nc); |
97b0b1ad | 1436 | else |
c231740d | 1437 | free_skb: |
1da177e4 | 1438 | dev_kfree_skb(nc); |
3ab08fe2 | 1439 | |
1da177e4 LT |
1440 | return ret; |
1441 | } | |
1442 | ||
0d060606 AV |
1443 | SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes, |
1444 | const struct sigevent __user *, u_notification) | |
1445 | { | |
1446 | struct sigevent n, *p = NULL; | |
1447 | if (u_notification) { | |
1448 | if (copy_from_user(&n, u_notification, sizeof(struct sigevent))) | |
1449 | return -EFAULT; | |
1450 | p = &n; | |
1451 | } | |
1452 | return do_mq_notify(mqdes, p); | |
1453 | } | |
1454 | ||
1455 | static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old) | |
1da177e4 | 1456 | { |
2903ff01 | 1457 | struct fd f; |
1da177e4 LT |
1458 | struct inode *inode; |
1459 | struct mqueue_inode_info *info; | |
1460 | ||
0d060606 AV |
1461 | if (new && (new->mq_flags & (~O_NONBLOCK))) |
1462 | return -EINVAL; | |
1da177e4 | 1463 | |
2903ff01 | 1464 | f = fdget(mqdes); |
0d060606 AV |
1465 | if (!f.file) |
1466 | return -EBADF; | |
1da177e4 | 1467 | |
2903ff01 | 1468 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
0d060606 AV |
1469 | fdput(f); |
1470 | return -EBADF; | |
8d8ffefa | 1471 | } |
0d060606 AV |
1472 | |
1473 | inode = file_inode(f.file); | |
1da177e4 LT |
1474 | info = MQUEUE_I(inode); |
1475 | ||
1476 | spin_lock(&info->lock); | |
1477 | ||
0d060606 AV |
1478 | if (old) { |
1479 | *old = info->attr; | |
1480 | old->mq_flags = f.file->f_flags & O_NONBLOCK; | |
1481 | } | |
1482 | if (new) { | |
1483 | audit_mq_getsetattr(mqdes, new); | |
2903ff01 | 1484 | spin_lock(&f.file->f_lock); |
0d060606 | 1485 | if (new->mq_flags & O_NONBLOCK) |
2903ff01 | 1486 | f.file->f_flags |= O_NONBLOCK; |
1da177e4 | 1487 | else |
2903ff01 AV |
1488 | f.file->f_flags &= ~O_NONBLOCK; |
1489 | spin_unlock(&f.file->f_lock); | |
1da177e4 | 1490 | |
078cd827 | 1491 | inode->i_atime = inode->i_ctime = current_time(inode); |
1da177e4 LT |
1492 | } |
1493 | ||
1494 | spin_unlock(&info->lock); | |
0d060606 AV |
1495 | fdput(f); |
1496 | return 0; | |
1497 | } | |
1da177e4 | 1498 | |
0d060606 AV |
1499 | SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes, |
1500 | const struct mq_attr __user *, u_mqstat, | |
1501 | struct mq_attr __user *, u_omqstat) | |
1502 | { | |
1503 | int ret; | |
1504 | struct mq_attr mqstat, omqstat; | |
1505 | struct mq_attr *new = NULL, *old = NULL; | |
1da177e4 | 1506 | |
0d060606 AV |
1507 | if (u_mqstat) { |
1508 | new = &mqstat; | |
1509 | if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr))) | |
1510 | return -EFAULT; | |
1511 | } | |
1512 | if (u_omqstat) | |
1513 | old = &omqstat; | |
1514 | ||
1515 | ret = do_mq_getsetattr(mqdes, new, old); | |
1516 | if (ret || !old) | |
1517 | return ret; | |
1518 | ||
1519 | if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr))) | |
1520 | return -EFAULT; | |
1521 | return 0; | |
1522 | } | |
1523 | ||
1524 | #ifdef CONFIG_COMPAT | |
1525 | ||
1526 | struct compat_mq_attr { | |
1527 | compat_long_t mq_flags; /* message queue flags */ | |
1528 | compat_long_t mq_maxmsg; /* maximum number of messages */ | |
1529 | compat_long_t mq_msgsize; /* maximum message size */ | |
1530 | compat_long_t mq_curmsgs; /* number of messages currently queued */ | |
1531 | compat_long_t __reserved[4]; /* ignored for input, zeroed for output */ | |
1532 | }; | |
1533 | ||
1534 | static inline int get_compat_mq_attr(struct mq_attr *attr, | |
1535 | const struct compat_mq_attr __user *uattr) | |
1536 | { | |
1537 | struct compat_mq_attr v; | |
1538 | ||
1539 | if (copy_from_user(&v, uattr, sizeof(*uattr))) | |
1540 | return -EFAULT; | |
1541 | ||
1542 | memset(attr, 0, sizeof(*attr)); | |
1543 | attr->mq_flags = v.mq_flags; | |
1544 | attr->mq_maxmsg = v.mq_maxmsg; | |
1545 | attr->mq_msgsize = v.mq_msgsize; | |
1546 | attr->mq_curmsgs = v.mq_curmsgs; | |
1547 | return 0; | |
1548 | } | |
1549 | ||
1550 | static inline int put_compat_mq_attr(const struct mq_attr *attr, | |
1551 | struct compat_mq_attr __user *uattr) | |
1552 | { | |
1553 | struct compat_mq_attr v; | |
1554 | ||
1555 | memset(&v, 0, sizeof(v)); | |
1556 | v.mq_flags = attr->mq_flags; | |
1557 | v.mq_maxmsg = attr->mq_maxmsg; | |
1558 | v.mq_msgsize = attr->mq_msgsize; | |
1559 | v.mq_curmsgs = attr->mq_curmsgs; | |
1560 | if (copy_to_user(uattr, &v, sizeof(*uattr))) | |
1561 | return -EFAULT; | |
1562 | return 0; | |
1563 | } | |
1564 | ||
1565 | COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name, | |
1566 | int, oflag, compat_mode_t, mode, | |
1567 | struct compat_mq_attr __user *, u_attr) | |
1568 | { | |
1569 | struct mq_attr attr, *p = NULL; | |
1570 | if (u_attr && oflag & O_CREAT) { | |
1571 | p = &attr; | |
1572 | if (get_compat_mq_attr(&attr, u_attr)) | |
1573 | return -EFAULT; | |
1574 | } | |
1575 | return do_mq_open(u_name, oflag, mode, p); | |
1576 | } | |
1577 | ||
b0d17578 AB |
1578 | COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes, |
1579 | const struct compat_sigevent __user *, u_notification) | |
1580 | { | |
1581 | struct sigevent n, *p = NULL; | |
1582 | if (u_notification) { | |
1583 | if (get_compat_sigevent(&n, u_notification)) | |
1584 | return -EFAULT; | |
1585 | if (n.sigev_notify == SIGEV_THREAD) | |
1586 | n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int); | |
1587 | p = &n; | |
1588 | } | |
1589 | return do_mq_notify(mqdes, p); | |
1590 | } | |
1591 | ||
1592 | COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes, | |
1593 | const struct compat_mq_attr __user *, u_mqstat, | |
1594 | struct compat_mq_attr __user *, u_omqstat) | |
1595 | { | |
1596 | int ret; | |
1597 | struct mq_attr mqstat, omqstat; | |
1598 | struct mq_attr *new = NULL, *old = NULL; | |
1599 | ||
1600 | if (u_mqstat) { | |
1601 | new = &mqstat; | |
1602 | if (get_compat_mq_attr(new, u_mqstat)) | |
1603 | return -EFAULT; | |
1604 | } | |
1605 | if (u_omqstat) | |
1606 | old = &omqstat; | |
1607 | ||
1608 | ret = do_mq_getsetattr(mqdes, new, old); | |
1609 | if (ret || !old) | |
1610 | return ret; | |
1611 | ||
1612 | if (put_compat_mq_attr(old, u_omqstat)) | |
1613 | return -EFAULT; | |
1614 | return 0; | |
1615 | } | |
1616 | #endif | |
1617 | ||
1618 | #ifdef CONFIG_COMPAT_32BIT_TIME | |
9afc5eee | 1619 | static int compat_prepare_timeout(const struct old_timespec32 __user *p, |
b9047726 | 1620 | struct timespec64 *ts) |
0d060606 | 1621 | { |
9afc5eee | 1622 | if (get_old_timespec32(ts, p)) |
0d060606 | 1623 | return -EFAULT; |
b9047726 | 1624 | if (!timespec64_valid(ts)) |
0d060606 AV |
1625 | return -EINVAL; |
1626 | return 0; | |
1627 | } | |
1628 | ||
8dabe724 AB |
1629 | SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes, |
1630 | const char __user *, u_msg_ptr, | |
1631 | unsigned int, msg_len, unsigned int, msg_prio, | |
1632 | const struct old_timespec32 __user *, u_abs_timeout) | |
0d060606 | 1633 | { |
b9047726 | 1634 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1635 | if (u_abs_timeout) { |
1636 | int res = compat_prepare_timeout(u_abs_timeout, &ts); | |
1637 | if (res) | |
1638 | return res; | |
1639 | p = &ts; | |
1640 | } | |
1641 | return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p); | |
1642 | } | |
1643 | ||
8dabe724 AB |
1644 | SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes, |
1645 | char __user *, u_msg_ptr, | |
1646 | unsigned int, msg_len, unsigned int __user *, u_msg_prio, | |
1647 | const struct old_timespec32 __user *, u_abs_timeout) | |
0d060606 | 1648 | { |
b9047726 | 1649 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1650 | if (u_abs_timeout) { |
1651 | int res = compat_prepare_timeout(u_abs_timeout, &ts); | |
1652 | if (res) | |
1653 | return res; | |
1654 | p = &ts; | |
1655 | } | |
1656 | return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p); | |
1657 | } | |
0d060606 | 1658 | #endif |
1da177e4 | 1659 | |
92e1d5be | 1660 | static const struct inode_operations mqueue_dir_inode_operations = { |
1da177e4 LT |
1661 | .lookup = simple_lookup, |
1662 | .create = mqueue_create, | |
1663 | .unlink = mqueue_unlink, | |
1664 | }; | |
1665 | ||
9a32144e | 1666 | static const struct file_operations mqueue_file_operations = { |
1da177e4 LT |
1667 | .flush = mqueue_flush_file, |
1668 | .poll = mqueue_poll_file, | |
1669 | .read = mqueue_read_file, | |
6038f373 | 1670 | .llseek = default_llseek, |
1da177e4 LT |
1671 | }; |
1672 | ||
b87221de | 1673 | static const struct super_operations mqueue_super_ops = { |
1da177e4 | 1674 | .alloc_inode = mqueue_alloc_inode, |
015d7956 | 1675 | .free_inode = mqueue_free_inode, |
6d8af64c | 1676 | .evict_inode = mqueue_evict_inode, |
1da177e4 | 1677 | .statfs = simple_statfs, |
1da177e4 LT |
1678 | }; |
1679 | ||
935c6912 DH |
1680 | static const struct fs_context_operations mqueue_fs_context_ops = { |
1681 | .free = mqueue_fs_context_free, | |
1682 | .get_tree = mqueue_get_tree, | |
1683 | }; | |
1684 | ||
1da177e4 | 1685 | static struct file_system_type mqueue_fs_type = { |
935c6912 DH |
1686 | .name = "mqueue", |
1687 | .init_fs_context = mqueue_init_fs_context, | |
1688 | .kill_sb = kill_litter_super, | |
1689 | .fs_flags = FS_USERNS_MOUNT, | |
1da177e4 LT |
1690 | }; |
1691 | ||
7eafd7c7 SH |
1692 | int mq_init_ns(struct ipc_namespace *ns) |
1693 | { | |
935c6912 DH |
1694 | struct vfsmount *m; |
1695 | ||
7eafd7c7 SH |
1696 | ns->mq_queues_count = 0; |
1697 | ns->mq_queues_max = DFLT_QUEUESMAX; | |
1698 | ns->mq_msg_max = DFLT_MSGMAX; | |
1699 | ns->mq_msgsize_max = DFLT_MSGSIZEMAX; | |
cef0184c KM |
1700 | ns->mq_msg_default = DFLT_MSG; |
1701 | ns->mq_msgsize_default = DFLT_MSGSIZE; | |
7eafd7c7 | 1702 | |
935c6912 DH |
1703 | m = mq_create_mount(ns); |
1704 | if (IS_ERR(m)) | |
1705 | return PTR_ERR(m); | |
1706 | ns->mq_mnt = m; | |
7eafd7c7 SH |
1707 | return 0; |
1708 | } | |
1709 | ||
1710 | void mq_clear_sbinfo(struct ipc_namespace *ns) | |
1711 | { | |
cfb2f6f6 | 1712 | ns->mq_mnt->mnt_sb->s_fs_info = NULL; |
7eafd7c7 SH |
1713 | } |
1714 | ||
1715 | void mq_put_mnt(struct ipc_namespace *ns) | |
1716 | { | |
cfb2f6f6 | 1717 | kern_unmount(ns->mq_mnt); |
7eafd7c7 SH |
1718 | } |
1719 | ||
1da177e4 LT |
1720 | static int __init init_mqueue_fs(void) |
1721 | { | |
1722 | int error; | |
1723 | ||
1724 | mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache", | |
1725 | sizeof(struct mqueue_inode_info), 0, | |
5d097056 | 1726 | SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once); |
1da177e4 LT |
1727 | if (mqueue_inode_cachep == NULL) |
1728 | return -ENOMEM; | |
1729 | ||
2329e392 | 1730 | /* ignore failures - they are not fatal */ |
bdc8e5f8 | 1731 | mq_sysctl_table = mq_register_sysctl_table(); |
1da177e4 LT |
1732 | |
1733 | error = register_filesystem(&mqueue_fs_type); | |
1734 | if (error) | |
1735 | goto out_sysctl; | |
1736 | ||
7eafd7c7 SH |
1737 | spin_lock_init(&mq_lock); |
1738 | ||
6f686574 AV |
1739 | error = mq_init_ns(&init_ipc_ns); |
1740 | if (error) | |
1da177e4 | 1741 | goto out_filesystem; |
1da177e4 | 1742 | |
1da177e4 LT |
1743 | return 0; |
1744 | ||
1745 | out_filesystem: | |
1746 | unregister_filesystem(&mqueue_fs_type); | |
1747 | out_sysctl: | |
1748 | if (mq_sysctl_table) | |
1749 | unregister_sysctl_table(mq_sysctl_table); | |
1a1d92c1 | 1750 | kmem_cache_destroy(mqueue_inode_cachep); |
1da177e4 LT |
1751 | return error; |
1752 | } | |
1753 | ||
6d08a256 | 1754 | device_initcall(init_mqueue_fs); |