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 | |
1da177e4 LT |
167 | static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode) |
168 | { | |
169 | return container_of(inode, struct mqueue_inode_info, vfs_inode); | |
170 | } | |
171 | ||
7eafd7c7 SH |
172 | /* |
173 | * This routine should be called with the mq_lock held. | |
174 | */ | |
175 | static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode) | |
614b84cf | 176 | { |
7eafd7c7 | 177 | return get_ipc_ns(inode->i_sb->s_fs_info); |
614b84cf SH |
178 | } |
179 | ||
7eafd7c7 | 180 | static struct ipc_namespace *get_ns_from_inode(struct inode *inode) |
614b84cf | 181 | { |
7eafd7c7 SH |
182 | struct ipc_namespace *ns; |
183 | ||
184 | spin_lock(&mq_lock); | |
185 | ns = __get_ns_from_inode(inode); | |
186 | spin_unlock(&mq_lock); | |
187 | return ns; | |
614b84cf SH |
188 | } |
189 | ||
d6629859 DL |
190 | /* Auxiliary functions to manipulate messages' list */ |
191 | static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info) | |
192 | { | |
193 | struct rb_node **p, *parent = NULL; | |
194 | struct posix_msg_tree_node *leaf; | |
a5091fda | 195 | bool rightmost = true; |
d6629859 DL |
196 | |
197 | p = &info->msg_tree.rb_node; | |
198 | while (*p) { | |
199 | parent = *p; | |
200 | leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); | |
201 | ||
202 | if (likely(leaf->priority == msg->m_type)) | |
203 | goto insert_msg; | |
a5091fda | 204 | else if (msg->m_type < leaf->priority) { |
d6629859 | 205 | p = &(*p)->rb_left; |
a5091fda DB |
206 | rightmost = false; |
207 | } else | |
d6629859 DL |
208 | p = &(*p)->rb_right; |
209 | } | |
ce2d52cc DL |
210 | if (info->node_cache) { |
211 | leaf = info->node_cache; | |
212 | info->node_cache = NULL; | |
213 | } else { | |
214 | leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC); | |
215 | if (!leaf) | |
216 | return -ENOMEM; | |
ce2d52cc | 217 | INIT_LIST_HEAD(&leaf->msg_list); |
ce2d52cc | 218 | } |
d6629859 | 219 | leaf->priority = msg->m_type; |
a5091fda DB |
220 | |
221 | if (rightmost) | |
222 | info->msg_tree_rightmost = &leaf->rb_node; | |
223 | ||
d6629859 DL |
224 | rb_link_node(&leaf->rb_node, parent, p); |
225 | rb_insert_color(&leaf->rb_node, &info->msg_tree); | |
d6629859 DL |
226 | insert_msg: |
227 | info->attr.mq_curmsgs++; | |
228 | info->qsize += msg->m_ts; | |
229 | list_add_tail(&msg->m_list, &leaf->msg_list); | |
230 | return 0; | |
231 | } | |
232 | ||
a5091fda DB |
233 | static inline void msg_tree_erase(struct posix_msg_tree_node *leaf, |
234 | struct mqueue_inode_info *info) | |
235 | { | |
236 | struct rb_node *node = &leaf->rb_node; | |
237 | ||
238 | if (info->msg_tree_rightmost == node) | |
239 | info->msg_tree_rightmost = rb_prev(node); | |
240 | ||
241 | rb_erase(node, &info->msg_tree); | |
43afe4d3 | 242 | if (info->node_cache) |
a5091fda | 243 | kfree(leaf); |
43afe4d3 | 244 | else |
a5091fda | 245 | info->node_cache = leaf; |
a5091fda DB |
246 | } |
247 | ||
d6629859 DL |
248 | static inline struct msg_msg *msg_get(struct mqueue_inode_info *info) |
249 | { | |
a5091fda | 250 | struct rb_node *parent = NULL; |
d6629859 DL |
251 | struct posix_msg_tree_node *leaf; |
252 | struct msg_msg *msg; | |
253 | ||
254 | try_again: | |
a5091fda DB |
255 | /* |
256 | * During insert, low priorities go to the left and high to the | |
257 | * right. On receive, we want the highest priorities first, so | |
258 | * walk all the way to the right. | |
259 | */ | |
260 | parent = info->msg_tree_rightmost; | |
d6629859 DL |
261 | if (!parent) { |
262 | if (info->attr.mq_curmsgs) { | |
263 | pr_warn_once("Inconsistency in POSIX message queue, " | |
264 | "no tree element, but supposedly messages " | |
265 | "should exist!\n"); | |
266 | info->attr.mq_curmsgs = 0; | |
267 | } | |
268 | return NULL; | |
269 | } | |
270 | leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node); | |
ce2d52cc | 271 | if (unlikely(list_empty(&leaf->msg_list))) { |
d6629859 DL |
272 | pr_warn_once("Inconsistency in POSIX message queue, " |
273 | "empty leaf node but we haven't implemented " | |
274 | "lazy leaf delete!\n"); | |
a5091fda | 275 | msg_tree_erase(leaf, info); |
d6629859 DL |
276 | goto try_again; |
277 | } else { | |
278 | msg = list_first_entry(&leaf->msg_list, | |
279 | struct msg_msg, m_list); | |
280 | list_del(&msg->m_list); | |
281 | if (list_empty(&leaf->msg_list)) { | |
a5091fda | 282 | msg_tree_erase(leaf, info); |
d6629859 DL |
283 | } |
284 | } | |
285 | info->attr.mq_curmsgs--; | |
286 | info->qsize -= msg->m_ts; | |
287 | return msg; | |
288 | } | |
289 | ||
7eafd7c7 | 290 | static struct inode *mqueue_get_inode(struct super_block *sb, |
1b9d5ff7 | 291 | struct ipc_namespace *ipc_ns, umode_t mode, |
7eafd7c7 | 292 | struct mq_attr *attr) |
1da177e4 LT |
293 | { |
294 | struct inode *inode; | |
d40dcdb0 | 295 | int ret = -ENOMEM; |
1da177e4 LT |
296 | |
297 | inode = new_inode(sb); | |
04715206 JS |
298 | if (!inode) |
299 | goto err; | |
300 | ||
301 | inode->i_ino = get_next_ino(); | |
302 | inode->i_mode = mode; | |
303 | inode->i_uid = current_fsuid(); | |
304 | inode->i_gid = current_fsgid(); | |
078cd827 | 305 | inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode); |
04715206 JS |
306 | |
307 | if (S_ISREG(mode)) { | |
308 | struct mqueue_inode_info *info; | |
d6629859 | 309 | unsigned long mq_bytes, mq_treesize; |
04715206 JS |
310 | |
311 | inode->i_fop = &mqueue_file_operations; | |
312 | inode->i_size = FILENT_SIZE; | |
313 | /* mqueue specific info */ | |
314 | info = MQUEUE_I(inode); | |
315 | spin_lock_init(&info->lock); | |
316 | init_waitqueue_head(&info->wait_q); | |
317 | INIT_LIST_HEAD(&info->e_wait_q[0].list); | |
318 | INIT_LIST_HEAD(&info->e_wait_q[1].list); | |
319 | info->notify_owner = NULL; | |
6f9ac6d9 | 320 | info->notify_user_ns = NULL; |
04715206 | 321 | info->qsize = 0; |
6e52a9f0 | 322 | info->ucounts = NULL; /* set when all is ok */ |
d6629859 | 323 | info->msg_tree = RB_ROOT; |
a5091fda | 324 | info->msg_tree_rightmost = NULL; |
ce2d52cc | 325 | info->node_cache = NULL; |
04715206 | 326 | memset(&info->attr, 0, sizeof(info->attr)); |
cef0184c KM |
327 | info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max, |
328 | ipc_ns->mq_msg_default); | |
329 | info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max, | |
330 | ipc_ns->mq_msgsize_default); | |
04715206 JS |
331 | if (attr) { |
332 | info->attr.mq_maxmsg = attr->mq_maxmsg; | |
333 | info->attr.mq_msgsize = attr->mq_msgsize; | |
334 | } | |
d6629859 DL |
335 | /* |
336 | * We used to allocate a static array of pointers and account | |
337 | * the size of that array as well as one msg_msg struct per | |
338 | * possible message into the queue size. That's no longer | |
339 | * accurate as the queue is now an rbtree and will grow and | |
340 | * shrink depending on usage patterns. We can, however, still | |
341 | * account one msg_msg struct per message, but the nodes are | |
342 | * allocated depending on priority usage, and most programs | |
343 | * only use one, or a handful, of priorities. However, since | |
344 | * this is pinned memory, we need to assume worst case, so | |
345 | * that means the min(mq_maxmsg, max_priorities) * struct | |
346 | * posix_msg_tree_node. | |
347 | */ | |
05c1b290 AV |
348 | |
349 | ret = -EINVAL; | |
350 | if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0) | |
351 | goto out_inode; | |
352 | if (capable(CAP_SYS_RESOURCE)) { | |
353 | if (info->attr.mq_maxmsg > HARD_MSGMAX || | |
354 | info->attr.mq_msgsize > HARD_MSGSIZEMAX) | |
355 | goto out_inode; | |
356 | } else { | |
357 | if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max || | |
358 | info->attr.mq_msgsize > ipc_ns->mq_msgsize_max) | |
359 | goto out_inode; | |
360 | } | |
361 | ret = -EOVERFLOW; | |
362 | /* check for overflow */ | |
363 | if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg) | |
364 | goto out_inode; | |
d6629859 DL |
365 | mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + |
366 | min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * | |
367 | sizeof(struct posix_msg_tree_node); | |
05c1b290 AV |
368 | mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize; |
369 | if (mq_bytes + mq_treesize < mq_bytes) | |
370 | goto out_inode; | |
371 | mq_bytes += mq_treesize; | |
6e52a9f0 AG |
372 | info->ucounts = get_ucounts(current_ucounts()); |
373 | if (info->ucounts) { | |
374 | long msgqueue; | |
375 | ||
376 | spin_lock(&mq_lock); | |
377 | msgqueue = inc_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes); | |
378 | if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) { | |
379 | dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes); | |
380 | spin_unlock(&mq_lock); | |
381 | put_ucounts(info->ucounts); | |
382 | info->ucounts = NULL; | |
383 | /* mqueue_evict_inode() releases info->messages */ | |
384 | ret = -EMFILE; | |
385 | goto out_inode; | |
386 | } | |
04715206 | 387 | spin_unlock(&mq_lock); |
1da177e4 | 388 | } |
04715206 JS |
389 | } else if (S_ISDIR(mode)) { |
390 | inc_nlink(inode); | |
391 | /* Some things misbehave if size == 0 on a directory */ | |
392 | inode->i_size = 2 * DIRENT_SIZE; | |
393 | inode->i_op = &mqueue_dir_inode_operations; | |
394 | inode->i_fop = &simple_dir_operations; | |
1da177e4 | 395 | } |
04715206 | 396 | |
1da177e4 LT |
397 | return inode; |
398 | out_inode: | |
1da177e4 | 399 | iput(inode); |
04715206 | 400 | err: |
d40dcdb0 | 401 | return ERR_PTR(ret); |
1da177e4 LT |
402 | } |
403 | ||
935c6912 | 404 | static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 LT |
405 | { |
406 | struct inode *inode; | |
cfb2f6f6 | 407 | struct ipc_namespace *ns = sb->s_fs_info; |
1da177e4 | 408 | |
a2982cc9 | 409 | sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV; |
09cbfeaf KS |
410 | sb->s_blocksize = PAGE_SIZE; |
411 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
412 | sb->s_magic = MQUEUE_MAGIC; |
413 | sb->s_op = &mqueue_super_ops; | |
414 | ||
48fde701 AV |
415 | inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL); |
416 | if (IS_ERR(inode)) | |
417 | return PTR_ERR(inode); | |
1da177e4 | 418 | |
48fde701 AV |
419 | sb->s_root = d_make_root(inode); |
420 | if (!sb->s_root) | |
421 | return -ENOMEM; | |
422 | return 0; | |
1da177e4 LT |
423 | } |
424 | ||
935c6912 | 425 | static int mqueue_get_tree(struct fs_context *fc) |
1da177e4 | 426 | { |
935c6912 DH |
427 | struct mqueue_fs_context *ctx = fc->fs_private; |
428 | ||
d60c4d01 WL |
429 | /* |
430 | * With a newly created ipc namespace, we don't need to do a search | |
431 | * for an ipc namespace match, but we still need to set s_fs_info. | |
432 | */ | |
433 | if (ctx->newns) { | |
434 | fc->s_fs_info = ctx->ipc_ns; | |
435 | return get_tree_nodev(fc, mqueue_fill_super); | |
436 | } | |
533770cc | 437 | return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns); |
935c6912 DH |
438 | } |
439 | ||
440 | static void mqueue_fs_context_free(struct fs_context *fc) | |
441 | { | |
442 | struct mqueue_fs_context *ctx = fc->fs_private; | |
443 | ||
709a643d | 444 | put_ipc_ns(ctx->ipc_ns); |
935c6912 DH |
445 | kfree(ctx); |
446 | } | |
447 | ||
448 | static int mqueue_init_fs_context(struct fs_context *fc) | |
449 | { | |
450 | struct mqueue_fs_context *ctx; | |
451 | ||
452 | ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL); | |
453 | if (!ctx) | |
454 | return -ENOMEM; | |
455 | ||
456 | ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns); | |
709a643d AV |
457 | put_user_ns(fc->user_ns); |
458 | fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns); | |
935c6912 DH |
459 | fc->fs_private = ctx; |
460 | fc->ops = &mqueue_fs_context_ops; | |
461 | return 0; | |
462 | } | |
463 | ||
d60c4d01 WL |
464 | /* |
465 | * mq_init_ns() is currently the only caller of mq_create_mount(). | |
466 | * So the ns parameter is always a newly created ipc namespace. | |
467 | */ | |
935c6912 DH |
468 | static struct vfsmount *mq_create_mount(struct ipc_namespace *ns) |
469 | { | |
470 | struct mqueue_fs_context *ctx; | |
471 | struct fs_context *fc; | |
472 | struct vfsmount *mnt; | |
473 | ||
474 | fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT); | |
475 | if (IS_ERR(fc)) | |
476 | return ERR_CAST(fc); | |
477 | ||
478 | ctx = fc->fs_private; | |
d60c4d01 | 479 | ctx->newns = true; |
935c6912 DH |
480 | put_ipc_ns(ctx->ipc_ns); |
481 | ctx->ipc_ns = get_ipc_ns(ns); | |
709a643d AV |
482 | put_user_ns(fc->user_ns); |
483 | fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns); | |
935c6912 DH |
484 | |
485 | mnt = fc_mount(fc); | |
486 | put_fs_context(fc); | |
487 | return mnt; | |
1da177e4 LT |
488 | } |
489 | ||
51cc5068 | 490 | static void init_once(void *foo) |
1da177e4 | 491 | { |
2c795fb0 | 492 | struct mqueue_inode_info *p = foo; |
1da177e4 | 493 | |
a35afb83 | 494 | inode_init_once(&p->vfs_inode); |
1da177e4 LT |
495 | } |
496 | ||
497 | static struct inode *mqueue_alloc_inode(struct super_block *sb) | |
498 | { | |
499 | struct mqueue_inode_info *ei; | |
500 | ||
fd60b288 | 501 | ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL); |
1da177e4 LT |
502 | if (!ei) |
503 | return NULL; | |
504 | return &ei->vfs_inode; | |
505 | } | |
506 | ||
015d7956 | 507 | static void mqueue_free_inode(struct inode *inode) |
1da177e4 LT |
508 | { |
509 | kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode)); | |
510 | } | |
511 | ||
6d8af64c | 512 | static void mqueue_evict_inode(struct inode *inode) |
1da177e4 LT |
513 | { |
514 | struct mqueue_inode_info *info; | |
7eafd7c7 | 515 | struct ipc_namespace *ipc_ns; |
d6a2946a LR |
516 | struct msg_msg *msg, *nmsg; |
517 | LIST_HEAD(tmp_msg); | |
1da177e4 | 518 | |
dbd5768f | 519 | clear_inode(inode); |
6d8af64c AV |
520 | |
521 | if (S_ISDIR(inode->i_mode)) | |
1da177e4 | 522 | return; |
6d8af64c | 523 | |
7eafd7c7 | 524 | ipc_ns = get_ns_from_inode(inode); |
1da177e4 LT |
525 | info = MQUEUE_I(inode); |
526 | spin_lock(&info->lock); | |
d6629859 | 527 | while ((msg = msg_get(info)) != NULL) |
d6a2946a | 528 | list_add_tail(&msg->m_list, &tmp_msg); |
ce2d52cc | 529 | kfree(info->node_cache); |
1da177e4 LT |
530 | spin_unlock(&info->lock); |
531 | ||
d6a2946a LR |
532 | list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) { |
533 | list_del(&msg->m_list); | |
534 | free_msg(msg); | |
535 | } | |
536 | ||
6e52a9f0 | 537 | if (info->ucounts) { |
a318f12e KC |
538 | unsigned long mq_bytes, mq_treesize; |
539 | ||
540 | /* Total amount of bytes accounted for the mqueue */ | |
541 | mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) + | |
542 | min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) * | |
543 | sizeof(struct posix_msg_tree_node); | |
544 | ||
545 | mq_bytes = mq_treesize + (info->attr.mq_maxmsg * | |
546 | info->attr.mq_msgsize); | |
547 | ||
1da177e4 | 548 | spin_lock(&mq_lock); |
6e52a9f0 | 549 | dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes); |
7eafd7c7 SH |
550 | /* |
551 | * get_ns_from_inode() ensures that the | |
552 | * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns | |
553 | * to which we now hold a reference, or it is NULL. | |
554 | * We can't put it here under mq_lock, though. | |
555 | */ | |
556 | if (ipc_ns) | |
557 | ipc_ns->mq_queues_count--; | |
1da177e4 | 558 | spin_unlock(&mq_lock); |
6e52a9f0 AG |
559 | put_ucounts(info->ucounts); |
560 | info->ucounts = NULL; | |
1da177e4 | 561 | } |
7eafd7c7 SH |
562 | if (ipc_ns) |
563 | put_ipc_ns(ipc_ns); | |
1da177e4 LT |
564 | } |
565 | ||
eecec19d | 566 | static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg) |
1da177e4 | 567 | { |
eecec19d | 568 | struct inode *dir = dentry->d_parent->d_inode; |
1da177e4 | 569 | struct inode *inode; |
eecec19d | 570 | struct mq_attr *attr = arg; |
1da177e4 | 571 | int error; |
7eafd7c7 | 572 | struct ipc_namespace *ipc_ns; |
1da177e4 LT |
573 | |
574 | spin_lock(&mq_lock); | |
7eafd7c7 SH |
575 | ipc_ns = __get_ns_from_inode(dir); |
576 | if (!ipc_ns) { | |
577 | error = -EACCES; | |
578 | goto out_unlock; | |
579 | } | |
f3713fd9 DB |
580 | |
581 | if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max && | |
582 | !capable(CAP_SYS_RESOURCE)) { | |
1da177e4 | 583 | error = -ENOSPC; |
614b84cf | 584 | goto out_unlock; |
1da177e4 | 585 | } |
614b84cf | 586 | ipc_ns->mq_queues_count++; |
1da177e4 LT |
587 | spin_unlock(&mq_lock); |
588 | ||
7eafd7c7 | 589 | inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr); |
d40dcdb0 JS |
590 | if (IS_ERR(inode)) { |
591 | error = PTR_ERR(inode); | |
1da177e4 | 592 | spin_lock(&mq_lock); |
614b84cf SH |
593 | ipc_ns->mq_queues_count--; |
594 | goto out_unlock; | |
1da177e4 LT |
595 | } |
596 | ||
7eafd7c7 | 597 | put_ipc_ns(ipc_ns); |
1da177e4 | 598 | dir->i_size += DIRENT_SIZE; |
078cd827 | 599 | dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir); |
1da177e4 LT |
600 | |
601 | d_instantiate(dentry, inode); | |
602 | dget(dentry); | |
603 | return 0; | |
614b84cf | 604 | out_unlock: |
1da177e4 | 605 | spin_unlock(&mq_lock); |
7eafd7c7 SH |
606 | if (ipc_ns) |
607 | put_ipc_ns(ipc_ns); | |
1da177e4 LT |
608 | return error; |
609 | } | |
610 | ||
549c7297 CB |
611 | static int mqueue_create(struct user_namespace *mnt_userns, struct inode *dir, |
612 | struct dentry *dentry, umode_t mode, bool excl) | |
eecec19d AV |
613 | { |
614 | return mqueue_create_attr(dentry, mode, NULL); | |
615 | } | |
616 | ||
1da177e4 LT |
617 | static int mqueue_unlink(struct inode *dir, struct dentry *dentry) |
618 | { | |
75c3cfa8 | 619 | struct inode *inode = d_inode(dentry); |
1da177e4 | 620 | |
078cd827 | 621 | dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir); |
1da177e4 | 622 | dir->i_size -= DIRENT_SIZE; |
239521f3 MS |
623 | drop_nlink(inode); |
624 | dput(dentry); | |
625 | return 0; | |
1da177e4 LT |
626 | } |
627 | ||
628 | /* | |
629 | * This is routine for system read from queue file. | |
630 | * To avoid mess with doing here some sort of mq_receive we allow | |
631 | * to read only queue size & notification info (the only values | |
632 | * that are interesting from user point of view and aren't accessible | |
633 | * through std routines) | |
634 | */ | |
635 | static ssize_t mqueue_read_file(struct file *filp, char __user *u_data, | |
f1a43f93 | 636 | size_t count, loff_t *off) |
1da177e4 | 637 | { |
496ad9aa | 638 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); |
1da177e4 | 639 | char buffer[FILENT_SIZE]; |
f1a43f93 | 640 | ssize_t ret; |
1da177e4 LT |
641 | |
642 | spin_lock(&info->lock); | |
643 | snprintf(buffer, sizeof(buffer), | |
644 | "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n", | |
645 | info->qsize, | |
646 | info->notify_owner ? info->notify.sigev_notify : 0, | |
647 | (info->notify_owner && | |
648 | info->notify.sigev_notify == SIGEV_SIGNAL) ? | |
649 | info->notify.sigev_signo : 0, | |
6c5f3e7b | 650 | pid_vnr(info->notify_owner)); |
1da177e4 LT |
651 | spin_unlock(&info->lock); |
652 | buffer[sizeof(buffer)-1] = '\0'; | |
1da177e4 | 653 | |
f1a43f93 AM |
654 | ret = simple_read_from_buffer(u_data, count, off, buffer, |
655 | strlen(buffer)); | |
656 | if (ret <= 0) | |
657 | return ret; | |
1da177e4 | 658 | |
078cd827 | 659 | file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp)); |
f1a43f93 | 660 | return ret; |
1da177e4 LT |
661 | } |
662 | ||
75e1fcc0 | 663 | static int mqueue_flush_file(struct file *filp, fl_owner_t id) |
1da177e4 | 664 | { |
496ad9aa | 665 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); |
1da177e4 LT |
666 | |
667 | spin_lock(&info->lock); | |
a03fcb73 | 668 | if (task_tgid(current) == info->notify_owner) |
1da177e4 LT |
669 | remove_notification(info); |
670 | ||
671 | spin_unlock(&info->lock); | |
672 | return 0; | |
673 | } | |
674 | ||
9dd95748 | 675 | static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab) |
1da177e4 | 676 | { |
496ad9aa | 677 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp)); |
9dd95748 | 678 | __poll_t retval = 0; |
1da177e4 LT |
679 | |
680 | poll_wait(filp, &info->wait_q, poll_tab); | |
681 | ||
682 | spin_lock(&info->lock); | |
683 | if (info->attr.mq_curmsgs) | |
a9a08845 | 684 | retval = EPOLLIN | EPOLLRDNORM; |
1da177e4 LT |
685 | |
686 | if (info->attr.mq_curmsgs < info->attr.mq_maxmsg) | |
a9a08845 | 687 | retval |= EPOLLOUT | EPOLLWRNORM; |
1da177e4 LT |
688 | spin_unlock(&info->lock); |
689 | ||
690 | return retval; | |
691 | } | |
692 | ||
693 | /* Adds current to info->e_wait_q[sr] before element with smaller prio */ | |
694 | static void wq_add(struct mqueue_inode_info *info, int sr, | |
695 | struct ext_wait_queue *ewp) | |
696 | { | |
697 | struct ext_wait_queue *walk; | |
698 | ||
1da177e4 | 699 | list_for_each_entry(walk, &info->e_wait_q[sr].list, list) { |
68e34f4e | 700 | if (walk->task->prio <= current->prio) { |
1da177e4 LT |
701 | list_add_tail(&ewp->list, &walk->list); |
702 | return; | |
703 | } | |
704 | } | |
705 | list_add_tail(&ewp->list, &info->e_wait_q[sr].list); | |
706 | } | |
707 | ||
708 | /* | |
709 | * Puts current task to sleep. Caller must hold queue lock. After return | |
710 | * lock isn't held. | |
711 | * sr: SEND or RECV | |
712 | */ | |
713 | static int wq_sleep(struct mqueue_inode_info *info, int sr, | |
9ca7d8e6 | 714 | ktime_t *timeout, struct ext_wait_queue *ewp) |
eac0b1c3 | 715 | __releases(&info->lock) |
1da177e4 LT |
716 | { |
717 | int retval; | |
718 | signed long time; | |
719 | ||
720 | wq_add(info, sr, ewp); | |
721 | ||
722 | for (;;) { | |
c5b2cbdb | 723 | /* memory barrier not required, we hold info->lock */ |
fa6004ad | 724 | __set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 LT |
725 | |
726 | spin_unlock(&info->lock); | |
32ea845d WG |
727 | time = schedule_hrtimeout_range_clock(timeout, 0, |
728 | HRTIMER_MODE_ABS, CLOCK_REALTIME); | |
1da177e4 | 729 | |
c5b2cbdb MS |
730 | if (READ_ONCE(ewp->state) == STATE_READY) { |
731 | /* see MQ_BARRIER for purpose/pairing */ | |
732 | smp_acquire__after_ctrl_dep(); | |
1da177e4 LT |
733 | retval = 0; |
734 | goto out; | |
735 | } | |
736 | spin_lock(&info->lock); | |
c5b2cbdb MS |
737 | |
738 | /* we hold info->lock, so no memory barrier required */ | |
739 | if (READ_ONCE(ewp->state) == STATE_READY) { | |
1da177e4 LT |
740 | retval = 0; |
741 | goto out_unlock; | |
742 | } | |
743 | if (signal_pending(current)) { | |
744 | retval = -ERESTARTSYS; | |
745 | break; | |
746 | } | |
747 | if (time == 0) { | |
748 | retval = -ETIMEDOUT; | |
749 | break; | |
750 | } | |
751 | } | |
752 | list_del(&ewp->list); | |
753 | out_unlock: | |
754 | spin_unlock(&info->lock); | |
755 | out: | |
756 | return retval; | |
757 | } | |
758 | ||
759 | /* | |
760 | * Returns waiting task that should be serviced first or NULL if none exists | |
761 | */ | |
762 | static struct ext_wait_queue *wq_get_first_waiter( | |
763 | struct mqueue_inode_info *info, int sr) | |
764 | { | |
765 | struct list_head *ptr; | |
766 | ||
767 | ptr = info->e_wait_q[sr].list.prev; | |
768 | if (ptr == &info->e_wait_q[sr].list) | |
769 | return NULL; | |
770 | return list_entry(ptr, struct ext_wait_queue, list); | |
771 | } | |
772 | ||
1da177e4 LT |
773 | |
774 | static inline void set_cookie(struct sk_buff *skb, char code) | |
775 | { | |
239521f3 | 776 | ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code; |
1da177e4 LT |
777 | } |
778 | ||
779 | /* | |
780 | * The next function is only to split too long sys_mq_timedsend | |
781 | */ | |
782 | static void __do_notify(struct mqueue_inode_info *info) | |
783 | { | |
784 | /* notification | |
785 | * invoked when there is registered process and there isn't process | |
786 | * waiting synchronously for message AND state of queue changed from | |
787 | * empty to not empty. Here we are sure that no one is waiting | |
788 | * synchronously. */ | |
789 | if (info->notify_owner && | |
790 | info->attr.mq_curmsgs == 1) { | |
1da177e4 LT |
791 | switch (info->notify.sigev_notify) { |
792 | case SIGEV_NONE: | |
793 | break; | |
b5f20061 ON |
794 | case SIGEV_SIGNAL: { |
795 | struct kernel_siginfo sig_i; | |
796 | struct task_struct *task; | |
797 | ||
798 | /* do_mq_notify() accepts sigev_signo == 0, why?? */ | |
799 | if (!info->notify.sigev_signo) | |
800 | break; | |
1da177e4 | 801 | |
faf1f22b | 802 | clear_siginfo(&sig_i); |
1da177e4 LT |
803 | sig_i.si_signo = info->notify.sigev_signo; |
804 | sig_i.si_errno = 0; | |
805 | sig_i.si_code = SI_MESGQ; | |
806 | sig_i.si_value = info->notify.sigev_value; | |
6b550f94 | 807 | rcu_read_lock(); |
b5f20061 | 808 | /* map current pid/uid into info->owner's namespaces */ |
a6684999 SB |
809 | sig_i.si_pid = task_tgid_nr_ns(current, |
810 | ns_of_pid(info->notify_owner)); | |
b5f20061 ON |
811 | sig_i.si_uid = from_kuid_munged(info->notify_user_ns, |
812 | current_uid()); | |
813 | /* | |
814 | * We can't use kill_pid_info(), this signal should | |
815 | * bypass check_kill_permission(). It is from kernel | |
816 | * but si_fromuser() can't know this. | |
817 | * We do check the self_exec_id, to avoid sending | |
818 | * signals to programs that don't expect them. | |
819 | */ | |
820 | task = pid_task(info->notify_owner, PIDTYPE_TGID); | |
821 | if (task && task->self_exec_id == | |
822 | info->notify_self_exec_id) { | |
823 | do_send_sig_info(info->notify.sigev_signo, | |
824 | &sig_i, task, PIDTYPE_TGID); | |
825 | } | |
6b550f94 | 826 | rcu_read_unlock(); |
1da177e4 | 827 | break; |
b5f20061 | 828 | } |
1da177e4 LT |
829 | case SIGEV_THREAD: |
830 | set_cookie(info->notify_cookie, NOTIFY_WOKENUP); | |
7ee015e0 | 831 | netlink_sendskb(info->notify_sock, info->notify_cookie); |
1da177e4 LT |
832 | break; |
833 | } | |
834 | /* after notification unregisters process */ | |
a03fcb73 | 835 | put_pid(info->notify_owner); |
6f9ac6d9 | 836 | put_user_ns(info->notify_user_ns); |
a03fcb73 | 837 | info->notify_owner = NULL; |
6f9ac6d9 | 838 | info->notify_user_ns = NULL; |
1da177e4 LT |
839 | } |
840 | wake_up(&info->wait_q); | |
841 | } | |
842 | ||
21fc538d | 843 | static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout, |
b9047726 | 844 | struct timespec64 *ts) |
1da177e4 | 845 | { |
b9047726 | 846 | if (get_timespec64(ts, u_abs_timeout)) |
9ca7d8e6 | 847 | return -EFAULT; |
b9047726 | 848 | if (!timespec64_valid(ts)) |
9ca7d8e6 | 849 | return -EINVAL; |
9ca7d8e6 | 850 | return 0; |
1da177e4 LT |
851 | } |
852 | ||
853 | static void remove_notification(struct mqueue_inode_info *info) | |
854 | { | |
a03fcb73 | 855 | if (info->notify_owner != NULL && |
1da177e4 LT |
856 | info->notify.sigev_notify == SIGEV_THREAD) { |
857 | set_cookie(info->notify_cookie, NOTIFY_REMOVED); | |
7ee015e0 | 858 | netlink_sendskb(info->notify_sock, info->notify_cookie); |
1da177e4 | 859 | } |
a03fcb73 | 860 | put_pid(info->notify_owner); |
6f9ac6d9 | 861 | put_user_ns(info->notify_user_ns); |
a03fcb73 | 862 | info->notify_owner = NULL; |
6f9ac6d9 | 863 | info->notify_user_ns = NULL; |
1da177e4 LT |
864 | } |
865 | ||
066cc813 AV |
866 | static int prepare_open(struct dentry *dentry, int oflag, int ro, |
867 | umode_t mode, struct filename *name, | |
614b84cf | 868 | struct mq_attr *attr) |
1da177e4 | 869 | { |
745ca247 DH |
870 | static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE, |
871 | MAY_READ | MAY_WRITE }; | |
765927b2 | 872 | int acc; |
066cc813 | 873 | |
9b20d7fc AV |
874 | if (d_really_is_negative(dentry)) { |
875 | if (!(oflag & O_CREAT)) | |
066cc813 | 876 | return -ENOENT; |
9b20d7fc AV |
877 | if (ro) |
878 | return ro; | |
879 | audit_inode_parent_hidden(name, dentry->d_parent); | |
880 | return vfs_mkobj(dentry, mode & ~current_umask(), | |
881 | mqueue_create_attr, attr); | |
066cc813 | 882 | } |
9b20d7fc AV |
883 | /* it already existed */ |
884 | audit_inode(name, dentry, 0); | |
885 | if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
886 | return -EEXIST; | |
765927b2 | 887 | if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) |
af4a5372 | 888 | return -EINVAL; |
765927b2 | 889 | acc = oflag2acc[oflag & O_ACCMODE]; |
47291baa | 890 | return inode_permission(&init_user_ns, d_inode(dentry), acc); |
1da177e4 LT |
891 | } |
892 | ||
0d060606 AV |
893 | static int do_mq_open(const char __user *u_name, int oflag, umode_t mode, |
894 | struct mq_attr *attr) | |
1da177e4 | 895 | { |
cfb2f6f6 EB |
896 | struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt; |
897 | struct dentry *root = mnt->mnt_root; | |
91a27b2a | 898 | struct filename *name; |
a713fd7f | 899 | struct path path; |
1da177e4 | 900 | int fd, error; |
312b90fb | 901 | int ro; |
1da177e4 | 902 | |
0d060606 | 903 | audit_mq_open(oflag, mode, attr); |
20ca73bc | 904 | |
1da177e4 LT |
905 | if (IS_ERR(name = getname(u_name))) |
906 | return PTR_ERR(name); | |
907 | ||
269f2134 | 908 | fd = get_unused_fd_flags(O_CLOEXEC); |
1da177e4 LT |
909 | if (fd < 0) |
910 | goto out_putname; | |
911 | ||
312b90fb | 912 | ro = mnt_want_write(mnt); /* we'll drop it in any case */ |
5955102c | 913 | inode_lock(d_inode(root)); |
91a27b2a | 914 | path.dentry = lookup_one_len(name->name, root, strlen(name->name)); |
765927b2 AV |
915 | if (IS_ERR(path.dentry)) { |
916 | error = PTR_ERR(path.dentry); | |
4294a8ee | 917 | goto out_putfd; |
1da177e4 | 918 | } |
312b90fb | 919 | path.mnt = mntget(mnt); |
066cc813 AV |
920 | error = prepare_open(path.dentry, oflag, ro, mode, name, attr); |
921 | if (!error) { | |
922 | struct file *file = dentry_open(&path, oflag, current_cred()); | |
923 | if (!IS_ERR(file)) | |
924 | fd_install(fd, file); | |
925 | else | |
926 | error = PTR_ERR(file); | |
7c7dce92 | 927 | } |
765927b2 | 928 | path_put(&path); |
7c7dce92 | 929 | out_putfd: |
765927b2 AV |
930 | if (error) { |
931 | put_unused_fd(fd); | |
932 | fd = error; | |
933 | } | |
5955102c | 934 | inode_unlock(d_inode(root)); |
38d78e58 VD |
935 | if (!ro) |
936 | mnt_drop_write(mnt); | |
1da177e4 LT |
937 | out_putname: |
938 | putname(name); | |
939 | return fd; | |
940 | } | |
941 | ||
0d060606 AV |
942 | SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode, |
943 | struct mq_attr __user *, u_attr) | |
944 | { | |
945 | struct mq_attr attr; | |
946 | if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr))) | |
947 | return -EFAULT; | |
948 | ||
949 | return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL); | |
950 | } | |
951 | ||
d5460c99 | 952 | SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name) |
1da177e4 LT |
953 | { |
954 | int err; | |
91a27b2a | 955 | struct filename *name; |
1da177e4 LT |
956 | struct dentry *dentry; |
957 | struct inode *inode = NULL; | |
7eafd7c7 | 958 | struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns; |
312b90fb | 959 | struct vfsmount *mnt = ipc_ns->mq_mnt; |
1da177e4 LT |
960 | |
961 | name = getname(u_name); | |
962 | if (IS_ERR(name)) | |
963 | return PTR_ERR(name); | |
964 | ||
79f6530c | 965 | audit_inode_parent_hidden(name, mnt->mnt_root); |
312b90fb AV |
966 | err = mnt_want_write(mnt); |
967 | if (err) | |
968 | goto out_name; | |
5955102c | 969 | inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT); |
91a27b2a JL |
970 | dentry = lookup_one_len(name->name, mnt->mnt_root, |
971 | strlen(name->name)); | |
1da177e4 LT |
972 | if (IS_ERR(dentry)) { |
973 | err = PTR_ERR(dentry); | |
974 | goto out_unlock; | |
975 | } | |
976 | ||
75c3cfa8 | 977 | inode = d_inode(dentry); |
312b90fb AV |
978 | if (!inode) { |
979 | err = -ENOENT; | |
980 | } else { | |
7de9c6ee | 981 | ihold(inode); |
6521f891 CB |
982 | err = vfs_unlink(&init_user_ns, d_inode(dentry->d_parent), |
983 | dentry, NULL); | |
312b90fb | 984 | } |
1da177e4 LT |
985 | dput(dentry); |
986 | ||
987 | out_unlock: | |
5955102c | 988 | inode_unlock(d_inode(mnt->mnt_root)); |
5758478a | 989 | iput(inode); |
312b90fb AV |
990 | mnt_drop_write(mnt); |
991 | out_name: | |
992 | putname(name); | |
1da177e4 LT |
993 | |
994 | return err; | |
995 | } | |
996 | ||
997 | /* Pipelined send and receive functions. | |
998 | * | |
999 | * If a receiver finds no waiting message, then it registers itself in the | |
1000 | * list of waiting receivers. A sender checks that list before adding the new | |
1001 | * message into the message array. If there is a waiting receiver, then it | |
1002 | * bypasses the message array and directly hands the message over to the | |
fa6004ad DB |
1003 | * receiver. The receiver accepts the message and returns without grabbing the |
1004 | * queue spinlock: | |
1005 | * | |
1006 | * - Set pointer to message. | |
1007 | * - Queue the receiver task for later wakeup (without the info->lock). | |
1008 | * - Update its state to STATE_READY. Now the receiver can continue. | |
1009 | * - Wake up the process after the lock is dropped. Should the process wake up | |
1010 | * before this wakeup (due to a timeout or a signal) it will either see | |
1011 | * STATE_READY and continue or acquire the lock to check the state again. | |
1da177e4 LT |
1012 | * |
1013 | * The same algorithm is used for senders. | |
1014 | */ | |
1015 | ||
ed29f171 | 1016 | static inline void __pipelined_op(struct wake_q_head *wake_q, |
fa6004ad | 1017 | struct mqueue_inode_info *info, |
ed29f171 | 1018 | struct ext_wait_queue *this) |
1da177e4 | 1019 | { |
a11ddb37 VG |
1020 | struct task_struct *task; |
1021 | ||
ed29f171 | 1022 | list_del(&this->list); |
a11ddb37 | 1023 | task = get_task_struct(this->task); |
c5b2cbdb MS |
1024 | |
1025 | /* see MQ_BARRIER for purpose/pairing */ | |
1026 | smp_store_release(&this->state, STATE_READY); | |
a11ddb37 | 1027 | wake_q_add_safe(wake_q, task); |
ed29f171 DB |
1028 | } |
1029 | ||
1030 | /* pipelined_send() - send a message directly to the task waiting in | |
1031 | * sys_mq_timedreceive() (without inserting message into a queue). | |
1032 | */ | |
1033 | static inline void pipelined_send(struct wake_q_head *wake_q, | |
1034 | struct mqueue_inode_info *info, | |
1035 | struct msg_msg *message, | |
1036 | struct ext_wait_queue *receiver) | |
1037 | { | |
1038 | receiver->msg = message; | |
1039 | __pipelined_op(wake_q, info, receiver); | |
1da177e4 LT |
1040 | } |
1041 | ||
1042 | /* pipelined_receive() - if there is task waiting in sys_mq_timedsend() | |
1043 | * gets its message and put to the queue (we have one free place for sure). */ | |
fa6004ad DB |
1044 | static inline void pipelined_receive(struct wake_q_head *wake_q, |
1045 | struct mqueue_inode_info *info) | |
1da177e4 LT |
1046 | { |
1047 | struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND); | |
1048 | ||
1049 | if (!sender) { | |
1050 | /* for poll */ | |
1051 | wake_up_interruptible(&info->wait_q); | |
1052 | return; | |
1053 | } | |
d6629859 DL |
1054 | if (msg_insert(sender->msg, info)) |
1055 | return; | |
fa6004ad | 1056 | |
ed29f171 | 1057 | __pipelined_op(wake_q, info, sender); |
1da177e4 LT |
1058 | } |
1059 | ||
0d060606 AV |
1060 | static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr, |
1061 | size_t msg_len, unsigned int msg_prio, | |
b9047726 | 1062 | struct timespec64 *ts) |
1da177e4 | 1063 | { |
2903ff01 | 1064 | struct fd f; |
1da177e4 LT |
1065 | struct inode *inode; |
1066 | struct ext_wait_queue wait; | |
1067 | struct ext_wait_queue *receiver; | |
1068 | struct msg_msg *msg_ptr; | |
1069 | struct mqueue_inode_info *info; | |
9ca7d8e6 | 1070 | ktime_t expires, *timeout = NULL; |
ce2d52cc | 1071 | struct posix_msg_tree_node *new_leaf = NULL; |
2903ff01 | 1072 | int ret = 0; |
194a6b5b | 1073 | DEFINE_WAKE_Q(wake_q); |
1da177e4 LT |
1074 | |
1075 | if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX)) | |
1076 | return -EINVAL; | |
1077 | ||
0d060606 | 1078 | if (ts) { |
b9047726 | 1079 | expires = timespec64_to_ktime(*ts); |
0d060606 AV |
1080 | timeout = &expires; |
1081 | } | |
1082 | ||
1083 | audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts); | |
1da177e4 | 1084 | |
2903ff01 AV |
1085 | f = fdget(mqdes); |
1086 | if (unlikely(!f.file)) { | |
8d8ffefa | 1087 | ret = -EBADF; |
1da177e4 | 1088 | goto out; |
8d8ffefa | 1089 | } |
1da177e4 | 1090 | |
496ad9aa | 1091 | inode = file_inode(f.file); |
2903ff01 | 1092 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
8d8ffefa | 1093 | ret = -EBADF; |
1da177e4 | 1094 | goto out_fput; |
8d8ffefa | 1095 | } |
1da177e4 | 1096 | info = MQUEUE_I(inode); |
9f45f5bf | 1097 | audit_file(f.file); |
1da177e4 | 1098 | |
2903ff01 | 1099 | if (unlikely(!(f.file->f_mode & FMODE_WRITE))) { |
8d8ffefa | 1100 | ret = -EBADF; |
1da177e4 | 1101 | goto out_fput; |
8d8ffefa | 1102 | } |
1da177e4 LT |
1103 | |
1104 | if (unlikely(msg_len > info->attr.mq_msgsize)) { | |
1105 | ret = -EMSGSIZE; | |
1106 | goto out_fput; | |
1107 | } | |
1108 | ||
1109 | /* First try to allocate memory, before doing anything with | |
1110 | * existing queues. */ | |
1111 | msg_ptr = load_msg(u_msg_ptr, msg_len); | |
1112 | if (IS_ERR(msg_ptr)) { | |
1113 | ret = PTR_ERR(msg_ptr); | |
1114 | goto out_fput; | |
1115 | } | |
1116 | msg_ptr->m_ts = msg_len; | |
1117 | msg_ptr->m_type = msg_prio; | |
1118 | ||
ce2d52cc DL |
1119 | /* |
1120 | * msg_insert really wants us to have a valid, spare node struct so | |
1121 | * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will | |
1122 | * fall back to that if necessary. | |
1123 | */ | |
1124 | if (!info->node_cache) | |
1125 | new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL); | |
1126 | ||
1da177e4 LT |
1127 | spin_lock(&info->lock); |
1128 | ||
ce2d52cc DL |
1129 | if (!info->node_cache && new_leaf) { |
1130 | /* Save our speculative allocation into the cache */ | |
ce2d52cc DL |
1131 | INIT_LIST_HEAD(&new_leaf->msg_list); |
1132 | info->node_cache = new_leaf; | |
ce2d52cc DL |
1133 | new_leaf = NULL; |
1134 | } else { | |
1135 | kfree(new_leaf); | |
1136 | } | |
1137 | ||
1da177e4 | 1138 | if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) { |
2903ff01 | 1139 | if (f.file->f_flags & O_NONBLOCK) { |
1da177e4 | 1140 | ret = -EAGAIN; |
1da177e4 LT |
1141 | } else { |
1142 | wait.task = current; | |
1143 | wait.msg = (void *) msg_ptr; | |
c5b2cbdb MS |
1144 | |
1145 | /* memory barrier not required, we hold info->lock */ | |
1146 | WRITE_ONCE(wait.state, STATE_NONE); | |
1da177e4 | 1147 | ret = wq_sleep(info, SEND, timeout, &wait); |
ce2d52cc DL |
1148 | /* |
1149 | * wq_sleep must be called with info->lock held, and | |
1150 | * returns with the lock released | |
1151 | */ | |
1152 | goto out_free; | |
1da177e4 | 1153 | } |
1da177e4 LT |
1154 | } else { |
1155 | receiver = wq_get_first_waiter(info, RECV); | |
1156 | if (receiver) { | |
fa6004ad | 1157 | pipelined_send(&wake_q, info, msg_ptr, receiver); |
1da177e4 LT |
1158 | } else { |
1159 | /* adds message to the queue */ | |
ce2d52cc DL |
1160 | ret = msg_insert(msg_ptr, info); |
1161 | if (ret) | |
1162 | goto out_unlock; | |
1da177e4 LT |
1163 | __do_notify(info); |
1164 | } | |
1165 | inode->i_atime = inode->i_mtime = inode->i_ctime = | |
078cd827 | 1166 | current_time(inode); |
1da177e4 | 1167 | } |
ce2d52cc DL |
1168 | out_unlock: |
1169 | spin_unlock(&info->lock); | |
fa6004ad | 1170 | wake_up_q(&wake_q); |
ce2d52cc DL |
1171 | out_free: |
1172 | if (ret) | |
1173 | free_msg(msg_ptr); | |
1da177e4 | 1174 | out_fput: |
2903ff01 | 1175 | fdput(f); |
1da177e4 LT |
1176 | out: |
1177 | return ret; | |
1178 | } | |
1179 | ||
0d060606 AV |
1180 | static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr, |
1181 | size_t msg_len, unsigned int __user *u_msg_prio, | |
b9047726 | 1182 | struct timespec64 *ts) |
1da177e4 | 1183 | { |
1da177e4 LT |
1184 | ssize_t ret; |
1185 | struct msg_msg *msg_ptr; | |
2903ff01 | 1186 | struct fd f; |
1da177e4 LT |
1187 | struct inode *inode; |
1188 | struct mqueue_inode_info *info; | |
1189 | struct ext_wait_queue wait; | |
9ca7d8e6 | 1190 | ktime_t expires, *timeout = NULL; |
ce2d52cc | 1191 | struct posix_msg_tree_node *new_leaf = NULL; |
1da177e4 | 1192 | |
0d060606 | 1193 | if (ts) { |
b9047726 | 1194 | expires = timespec64_to_ktime(*ts); |
9ca7d8e6 | 1195 | timeout = &expires; |
c32c8af4 | 1196 | } |
20ca73bc | 1197 | |
0d060606 | 1198 | audit_mq_sendrecv(mqdes, msg_len, 0, ts); |
1da177e4 | 1199 | |
2903ff01 AV |
1200 | f = fdget(mqdes); |
1201 | if (unlikely(!f.file)) { | |
8d8ffefa | 1202 | ret = -EBADF; |
1da177e4 | 1203 | goto out; |
8d8ffefa | 1204 | } |
1da177e4 | 1205 | |
496ad9aa | 1206 | inode = file_inode(f.file); |
2903ff01 | 1207 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
8d8ffefa | 1208 | ret = -EBADF; |
1da177e4 | 1209 | goto out_fput; |
8d8ffefa | 1210 | } |
1da177e4 | 1211 | info = MQUEUE_I(inode); |
9f45f5bf | 1212 | audit_file(f.file); |
1da177e4 | 1213 | |
2903ff01 | 1214 | if (unlikely(!(f.file->f_mode & FMODE_READ))) { |
8d8ffefa | 1215 | ret = -EBADF; |
1da177e4 | 1216 | goto out_fput; |
8d8ffefa | 1217 | } |
1da177e4 LT |
1218 | |
1219 | /* checks if buffer is big enough */ | |
1220 | if (unlikely(msg_len < info->attr.mq_msgsize)) { | |
1221 | ret = -EMSGSIZE; | |
1222 | goto out_fput; | |
1223 | } | |
1224 | ||
ce2d52cc DL |
1225 | /* |
1226 | * msg_insert really wants us to have a valid, spare node struct so | |
1227 | * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will | |
1228 | * fall back to that if necessary. | |
1229 | */ | |
1230 | if (!info->node_cache) | |
1231 | new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL); | |
1232 | ||
1da177e4 | 1233 | spin_lock(&info->lock); |
ce2d52cc DL |
1234 | |
1235 | if (!info->node_cache && new_leaf) { | |
1236 | /* Save our speculative allocation into the cache */ | |
ce2d52cc DL |
1237 | INIT_LIST_HEAD(&new_leaf->msg_list); |
1238 | info->node_cache = new_leaf; | |
ce2d52cc DL |
1239 | } else { |
1240 | kfree(new_leaf); | |
1241 | } | |
1242 | ||
1da177e4 | 1243 | if (info->attr.mq_curmsgs == 0) { |
2903ff01 | 1244 | if (f.file->f_flags & O_NONBLOCK) { |
1da177e4 LT |
1245 | spin_unlock(&info->lock); |
1246 | ret = -EAGAIN; | |
1da177e4 LT |
1247 | } else { |
1248 | wait.task = current; | |
c5b2cbdb MS |
1249 | |
1250 | /* memory barrier not required, we hold info->lock */ | |
1251 | WRITE_ONCE(wait.state, STATE_NONE); | |
1da177e4 LT |
1252 | ret = wq_sleep(info, RECV, timeout, &wait); |
1253 | msg_ptr = wait.msg; | |
1254 | } | |
1255 | } else { | |
194a6b5b | 1256 | DEFINE_WAKE_Q(wake_q); |
fa6004ad | 1257 | |
1da177e4 LT |
1258 | msg_ptr = msg_get(info); |
1259 | ||
1260 | inode->i_atime = inode->i_mtime = inode->i_ctime = | |
078cd827 | 1261 | current_time(inode); |
1da177e4 LT |
1262 | |
1263 | /* There is now free space in queue. */ | |
fa6004ad | 1264 | pipelined_receive(&wake_q, info); |
1da177e4 | 1265 | spin_unlock(&info->lock); |
fa6004ad | 1266 | wake_up_q(&wake_q); |
1da177e4 LT |
1267 | ret = 0; |
1268 | } | |
1269 | if (ret == 0) { | |
1270 | ret = msg_ptr->m_ts; | |
1271 | ||
1272 | if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) || | |
1273 | store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) { | |
1274 | ret = -EFAULT; | |
1275 | } | |
1276 | free_msg(msg_ptr); | |
1277 | } | |
1278 | out_fput: | |
2903ff01 | 1279 | fdput(f); |
1da177e4 LT |
1280 | out: |
1281 | return ret; | |
1282 | } | |
1283 | ||
0d060606 AV |
1284 | SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr, |
1285 | size_t, msg_len, unsigned int, msg_prio, | |
21fc538d | 1286 | const struct __kernel_timespec __user *, u_abs_timeout) |
0d060606 | 1287 | { |
b9047726 | 1288 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1289 | if (u_abs_timeout) { |
1290 | int res = prepare_timeout(u_abs_timeout, &ts); | |
1291 | if (res) | |
1292 | return res; | |
1293 | p = &ts; | |
1294 | } | |
1295 | return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p); | |
1296 | } | |
1297 | ||
1298 | SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr, | |
1299 | size_t, msg_len, unsigned int __user *, u_msg_prio, | |
21fc538d | 1300 | const struct __kernel_timespec __user *, u_abs_timeout) |
0d060606 | 1301 | { |
b9047726 | 1302 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1303 | if (u_abs_timeout) { |
1304 | int res = prepare_timeout(u_abs_timeout, &ts); | |
1305 | if (res) | |
1306 | return res; | |
1307 | p = &ts; | |
1308 | } | |
1309 | return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p); | |
1310 | } | |
1311 | ||
1da177e4 LT |
1312 | /* |
1313 | * Notes: the case when user wants us to deregister (with NULL as pointer) | |
1314 | * and he isn't currently owner of notification, will be silently discarded. | |
1315 | * It isn't explicitly defined in the POSIX. | |
1316 | */ | |
0d060606 | 1317 | static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification) |
1da177e4 | 1318 | { |
2903ff01 AV |
1319 | int ret; |
1320 | struct fd f; | |
1da177e4 LT |
1321 | struct sock *sock; |
1322 | struct inode *inode; | |
1da177e4 LT |
1323 | struct mqueue_inode_info *info; |
1324 | struct sk_buff *nc; | |
1325 | ||
0d060606 | 1326 | audit_mq_notify(mqdes, notification); |
1da177e4 | 1327 | |
20114f71 AV |
1328 | nc = NULL; |
1329 | sock = NULL; | |
0d060606 AV |
1330 | if (notification != NULL) { |
1331 | if (unlikely(notification->sigev_notify != SIGEV_NONE && | |
1332 | notification->sigev_notify != SIGEV_SIGNAL && | |
1333 | notification->sigev_notify != SIGEV_THREAD)) | |
1da177e4 | 1334 | return -EINVAL; |
0d060606 AV |
1335 | if (notification->sigev_notify == SIGEV_SIGNAL && |
1336 | !valid_signal(notification->sigev_signo)) { | |
1da177e4 LT |
1337 | return -EINVAL; |
1338 | } | |
0d060606 | 1339 | if (notification->sigev_notify == SIGEV_THREAD) { |
c3d8d1e3 PM |
1340 | long timeo; |
1341 | ||
1da177e4 LT |
1342 | /* create the notify skb */ |
1343 | nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL); | |
c231740d ME |
1344 | if (!nc) |
1345 | return -ENOMEM; | |
1346 | ||
1da177e4 | 1347 | if (copy_from_user(nc->data, |
0d060606 | 1348 | notification->sigev_value.sival_ptr, |
1da177e4 | 1349 | NOTIFY_COOKIE_LEN)) { |
8d8ffefa | 1350 | ret = -EFAULT; |
c231740d | 1351 | goto free_skb; |
1da177e4 LT |
1352 | } |
1353 | ||
1354 | /* TODO: add a header? */ | |
1355 | skb_put(nc, NOTIFY_COOKIE_LEN); | |
1356 | /* and attach it to the socket */ | |
1357 | retry: | |
0d060606 | 1358 | f = fdget(notification->sigev_signo); |
2903ff01 | 1359 | if (!f.file) { |
8d8ffefa | 1360 | ret = -EBADF; |
1da177e4 | 1361 | goto out; |
8d8ffefa | 1362 | } |
2903ff01 AV |
1363 | sock = netlink_getsockbyfilp(f.file); |
1364 | fdput(f); | |
1da177e4 LT |
1365 | if (IS_ERR(sock)) { |
1366 | ret = PTR_ERR(sock); | |
c231740d | 1367 | goto free_skb; |
1da177e4 LT |
1368 | } |
1369 | ||
c3d8d1e3 | 1370 | timeo = MAX_SCHEDULE_TIMEOUT; |
9457afee | 1371 | ret = netlink_attachskb(sock, nc, &timeo, NULL); |
f991af3d CW |
1372 | if (ret == 1) { |
1373 | sock = NULL; | |
8d8ffefa | 1374 | goto retry; |
f991af3d | 1375 | } |
c231740d ME |
1376 | if (ret) |
1377 | return ret; | |
1da177e4 LT |
1378 | } |
1379 | } | |
1380 | ||
2903ff01 AV |
1381 | f = fdget(mqdes); |
1382 | if (!f.file) { | |
8d8ffefa | 1383 | ret = -EBADF; |
1da177e4 | 1384 | goto out; |
8d8ffefa | 1385 | } |
1da177e4 | 1386 | |
496ad9aa | 1387 | inode = file_inode(f.file); |
2903ff01 | 1388 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
8d8ffefa | 1389 | ret = -EBADF; |
1da177e4 | 1390 | goto out_fput; |
8d8ffefa | 1391 | } |
1da177e4 LT |
1392 | info = MQUEUE_I(inode); |
1393 | ||
1394 | ret = 0; | |
1395 | spin_lock(&info->lock); | |
0d060606 | 1396 | if (notification == NULL) { |
a03fcb73 | 1397 | if (info->notify_owner == task_tgid(current)) { |
1da177e4 | 1398 | remove_notification(info); |
078cd827 | 1399 | inode->i_atime = inode->i_ctime = current_time(inode); |
1da177e4 | 1400 | } |
a03fcb73 | 1401 | } else if (info->notify_owner != NULL) { |
1da177e4 LT |
1402 | ret = -EBUSY; |
1403 | } else { | |
0d060606 | 1404 | switch (notification->sigev_notify) { |
1da177e4 LT |
1405 | case SIGEV_NONE: |
1406 | info->notify.sigev_notify = SIGEV_NONE; | |
1407 | break; | |
1408 | case SIGEV_THREAD: | |
1409 | info->notify_sock = sock; | |
1410 | info->notify_cookie = nc; | |
1411 | sock = NULL; | |
1412 | nc = NULL; | |
1413 | info->notify.sigev_notify = SIGEV_THREAD; | |
1414 | break; | |
1415 | case SIGEV_SIGNAL: | |
0d060606 AV |
1416 | info->notify.sigev_signo = notification->sigev_signo; |
1417 | info->notify.sigev_value = notification->sigev_value; | |
1da177e4 | 1418 | info->notify.sigev_notify = SIGEV_SIGNAL; |
b5f20061 | 1419 | info->notify_self_exec_id = current->self_exec_id; |
1da177e4 LT |
1420 | break; |
1421 | } | |
a03fcb73 CLG |
1422 | |
1423 | info->notify_owner = get_pid(task_tgid(current)); | |
6f9ac6d9 | 1424 | info->notify_user_ns = get_user_ns(current_user_ns()); |
078cd827 | 1425 | inode->i_atime = inode->i_ctime = current_time(inode); |
1da177e4 LT |
1426 | } |
1427 | spin_unlock(&info->lock); | |
1428 | out_fput: | |
2903ff01 | 1429 | fdput(f); |
1da177e4 | 1430 | out: |
3ab08fe2 | 1431 | if (sock) |
1da177e4 | 1432 | netlink_detachskb(sock, nc); |
97b0b1ad | 1433 | else |
c231740d | 1434 | free_skb: |
1da177e4 | 1435 | dev_kfree_skb(nc); |
3ab08fe2 | 1436 | |
1da177e4 LT |
1437 | return ret; |
1438 | } | |
1439 | ||
0d060606 AV |
1440 | SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes, |
1441 | const struct sigevent __user *, u_notification) | |
1442 | { | |
1443 | struct sigevent n, *p = NULL; | |
1444 | if (u_notification) { | |
1445 | if (copy_from_user(&n, u_notification, sizeof(struct sigevent))) | |
1446 | return -EFAULT; | |
1447 | p = &n; | |
1448 | } | |
1449 | return do_mq_notify(mqdes, p); | |
1450 | } | |
1451 | ||
1452 | static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old) | |
1da177e4 | 1453 | { |
2903ff01 | 1454 | struct fd f; |
1da177e4 LT |
1455 | struct inode *inode; |
1456 | struct mqueue_inode_info *info; | |
1457 | ||
0d060606 AV |
1458 | if (new && (new->mq_flags & (~O_NONBLOCK))) |
1459 | return -EINVAL; | |
1da177e4 | 1460 | |
2903ff01 | 1461 | f = fdget(mqdes); |
0d060606 AV |
1462 | if (!f.file) |
1463 | return -EBADF; | |
1da177e4 | 1464 | |
2903ff01 | 1465 | if (unlikely(f.file->f_op != &mqueue_file_operations)) { |
0d060606 AV |
1466 | fdput(f); |
1467 | return -EBADF; | |
8d8ffefa | 1468 | } |
0d060606 AV |
1469 | |
1470 | inode = file_inode(f.file); | |
1da177e4 LT |
1471 | info = MQUEUE_I(inode); |
1472 | ||
1473 | spin_lock(&info->lock); | |
1474 | ||
0d060606 AV |
1475 | if (old) { |
1476 | *old = info->attr; | |
1477 | old->mq_flags = f.file->f_flags & O_NONBLOCK; | |
1478 | } | |
1479 | if (new) { | |
1480 | audit_mq_getsetattr(mqdes, new); | |
2903ff01 | 1481 | spin_lock(&f.file->f_lock); |
0d060606 | 1482 | if (new->mq_flags & O_NONBLOCK) |
2903ff01 | 1483 | f.file->f_flags |= O_NONBLOCK; |
1da177e4 | 1484 | else |
2903ff01 AV |
1485 | f.file->f_flags &= ~O_NONBLOCK; |
1486 | spin_unlock(&f.file->f_lock); | |
1da177e4 | 1487 | |
078cd827 | 1488 | inode->i_atime = inode->i_ctime = current_time(inode); |
1da177e4 LT |
1489 | } |
1490 | ||
1491 | spin_unlock(&info->lock); | |
0d060606 AV |
1492 | fdput(f); |
1493 | return 0; | |
1494 | } | |
1da177e4 | 1495 | |
0d060606 AV |
1496 | SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes, |
1497 | const struct mq_attr __user *, u_mqstat, | |
1498 | struct mq_attr __user *, u_omqstat) | |
1499 | { | |
1500 | int ret; | |
1501 | struct mq_attr mqstat, omqstat; | |
1502 | struct mq_attr *new = NULL, *old = NULL; | |
1da177e4 | 1503 | |
0d060606 AV |
1504 | if (u_mqstat) { |
1505 | new = &mqstat; | |
1506 | if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr))) | |
1507 | return -EFAULT; | |
1508 | } | |
1509 | if (u_omqstat) | |
1510 | old = &omqstat; | |
1511 | ||
1512 | ret = do_mq_getsetattr(mqdes, new, old); | |
1513 | if (ret || !old) | |
1514 | return ret; | |
1515 | ||
1516 | if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr))) | |
1517 | return -EFAULT; | |
1518 | return 0; | |
1519 | } | |
1520 | ||
1521 | #ifdef CONFIG_COMPAT | |
1522 | ||
1523 | struct compat_mq_attr { | |
1524 | compat_long_t mq_flags; /* message queue flags */ | |
1525 | compat_long_t mq_maxmsg; /* maximum number of messages */ | |
1526 | compat_long_t mq_msgsize; /* maximum message size */ | |
1527 | compat_long_t mq_curmsgs; /* number of messages currently queued */ | |
1528 | compat_long_t __reserved[4]; /* ignored for input, zeroed for output */ | |
1529 | }; | |
1530 | ||
1531 | static inline int get_compat_mq_attr(struct mq_attr *attr, | |
1532 | const struct compat_mq_attr __user *uattr) | |
1533 | { | |
1534 | struct compat_mq_attr v; | |
1535 | ||
1536 | if (copy_from_user(&v, uattr, sizeof(*uattr))) | |
1537 | return -EFAULT; | |
1538 | ||
1539 | memset(attr, 0, sizeof(*attr)); | |
1540 | attr->mq_flags = v.mq_flags; | |
1541 | attr->mq_maxmsg = v.mq_maxmsg; | |
1542 | attr->mq_msgsize = v.mq_msgsize; | |
1543 | attr->mq_curmsgs = v.mq_curmsgs; | |
1544 | return 0; | |
1545 | } | |
1546 | ||
1547 | static inline int put_compat_mq_attr(const struct mq_attr *attr, | |
1548 | struct compat_mq_attr __user *uattr) | |
1549 | { | |
1550 | struct compat_mq_attr v; | |
1551 | ||
1552 | memset(&v, 0, sizeof(v)); | |
1553 | v.mq_flags = attr->mq_flags; | |
1554 | v.mq_maxmsg = attr->mq_maxmsg; | |
1555 | v.mq_msgsize = attr->mq_msgsize; | |
1556 | v.mq_curmsgs = attr->mq_curmsgs; | |
1557 | if (copy_to_user(uattr, &v, sizeof(*uattr))) | |
1558 | return -EFAULT; | |
1559 | return 0; | |
1560 | } | |
1561 | ||
1562 | COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name, | |
1563 | int, oflag, compat_mode_t, mode, | |
1564 | struct compat_mq_attr __user *, u_attr) | |
1565 | { | |
1566 | struct mq_attr attr, *p = NULL; | |
1567 | if (u_attr && oflag & O_CREAT) { | |
1568 | p = &attr; | |
1569 | if (get_compat_mq_attr(&attr, u_attr)) | |
1570 | return -EFAULT; | |
1571 | } | |
1572 | return do_mq_open(u_name, oflag, mode, p); | |
1573 | } | |
1574 | ||
b0d17578 AB |
1575 | COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes, |
1576 | const struct compat_sigevent __user *, u_notification) | |
1577 | { | |
1578 | struct sigevent n, *p = NULL; | |
1579 | if (u_notification) { | |
1580 | if (get_compat_sigevent(&n, u_notification)) | |
1581 | return -EFAULT; | |
1582 | if (n.sigev_notify == SIGEV_THREAD) | |
1583 | n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int); | |
1584 | p = &n; | |
1585 | } | |
1586 | return do_mq_notify(mqdes, p); | |
1587 | } | |
1588 | ||
1589 | COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes, | |
1590 | const struct compat_mq_attr __user *, u_mqstat, | |
1591 | struct compat_mq_attr __user *, u_omqstat) | |
1592 | { | |
1593 | int ret; | |
1594 | struct mq_attr mqstat, omqstat; | |
1595 | struct mq_attr *new = NULL, *old = NULL; | |
1596 | ||
1597 | if (u_mqstat) { | |
1598 | new = &mqstat; | |
1599 | if (get_compat_mq_attr(new, u_mqstat)) | |
1600 | return -EFAULT; | |
1601 | } | |
1602 | if (u_omqstat) | |
1603 | old = &omqstat; | |
1604 | ||
1605 | ret = do_mq_getsetattr(mqdes, new, old); | |
1606 | if (ret || !old) | |
1607 | return ret; | |
1608 | ||
1609 | if (put_compat_mq_attr(old, u_omqstat)) | |
1610 | return -EFAULT; | |
1611 | return 0; | |
1612 | } | |
1613 | #endif | |
1614 | ||
1615 | #ifdef CONFIG_COMPAT_32BIT_TIME | |
9afc5eee | 1616 | static int compat_prepare_timeout(const struct old_timespec32 __user *p, |
b9047726 | 1617 | struct timespec64 *ts) |
0d060606 | 1618 | { |
9afc5eee | 1619 | if (get_old_timespec32(ts, p)) |
0d060606 | 1620 | return -EFAULT; |
b9047726 | 1621 | if (!timespec64_valid(ts)) |
0d060606 AV |
1622 | return -EINVAL; |
1623 | return 0; | |
1624 | } | |
1625 | ||
8dabe724 AB |
1626 | SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes, |
1627 | const char __user *, u_msg_ptr, | |
1628 | unsigned int, msg_len, unsigned int, msg_prio, | |
1629 | const struct old_timespec32 __user *, u_abs_timeout) | |
0d060606 | 1630 | { |
b9047726 | 1631 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1632 | if (u_abs_timeout) { |
1633 | int res = compat_prepare_timeout(u_abs_timeout, &ts); | |
1634 | if (res) | |
1635 | return res; | |
1636 | p = &ts; | |
1637 | } | |
1638 | return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p); | |
1639 | } | |
1640 | ||
8dabe724 AB |
1641 | SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes, |
1642 | char __user *, u_msg_ptr, | |
1643 | unsigned int, msg_len, unsigned int __user *, u_msg_prio, | |
1644 | const struct old_timespec32 __user *, u_abs_timeout) | |
0d060606 | 1645 | { |
b9047726 | 1646 | struct timespec64 ts, *p = NULL; |
0d060606 AV |
1647 | if (u_abs_timeout) { |
1648 | int res = compat_prepare_timeout(u_abs_timeout, &ts); | |
1649 | if (res) | |
1650 | return res; | |
1651 | p = &ts; | |
1652 | } | |
1653 | return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p); | |
1654 | } | |
0d060606 | 1655 | #endif |
1da177e4 | 1656 | |
92e1d5be | 1657 | static const struct inode_operations mqueue_dir_inode_operations = { |
1da177e4 LT |
1658 | .lookup = simple_lookup, |
1659 | .create = mqueue_create, | |
1660 | .unlink = mqueue_unlink, | |
1661 | }; | |
1662 | ||
9a32144e | 1663 | static const struct file_operations mqueue_file_operations = { |
1da177e4 LT |
1664 | .flush = mqueue_flush_file, |
1665 | .poll = mqueue_poll_file, | |
1666 | .read = mqueue_read_file, | |
6038f373 | 1667 | .llseek = default_llseek, |
1da177e4 LT |
1668 | }; |
1669 | ||
b87221de | 1670 | static const struct super_operations mqueue_super_ops = { |
1da177e4 | 1671 | .alloc_inode = mqueue_alloc_inode, |
015d7956 | 1672 | .free_inode = mqueue_free_inode, |
6d8af64c | 1673 | .evict_inode = mqueue_evict_inode, |
1da177e4 | 1674 | .statfs = simple_statfs, |
1da177e4 LT |
1675 | }; |
1676 | ||
935c6912 DH |
1677 | static const struct fs_context_operations mqueue_fs_context_ops = { |
1678 | .free = mqueue_fs_context_free, | |
1679 | .get_tree = mqueue_get_tree, | |
1680 | }; | |
1681 | ||
1da177e4 | 1682 | static struct file_system_type mqueue_fs_type = { |
935c6912 DH |
1683 | .name = "mqueue", |
1684 | .init_fs_context = mqueue_init_fs_context, | |
1685 | .kill_sb = kill_litter_super, | |
1686 | .fs_flags = FS_USERNS_MOUNT, | |
1da177e4 LT |
1687 | }; |
1688 | ||
7eafd7c7 SH |
1689 | int mq_init_ns(struct ipc_namespace *ns) |
1690 | { | |
935c6912 DH |
1691 | struct vfsmount *m; |
1692 | ||
7eafd7c7 SH |
1693 | ns->mq_queues_count = 0; |
1694 | ns->mq_queues_max = DFLT_QUEUESMAX; | |
1695 | ns->mq_msg_max = DFLT_MSGMAX; | |
1696 | ns->mq_msgsize_max = DFLT_MSGSIZEMAX; | |
cef0184c KM |
1697 | ns->mq_msg_default = DFLT_MSG; |
1698 | ns->mq_msgsize_default = DFLT_MSGSIZE; | |
7eafd7c7 | 1699 | |
935c6912 DH |
1700 | m = mq_create_mount(ns); |
1701 | if (IS_ERR(m)) | |
1702 | return PTR_ERR(m); | |
1703 | ns->mq_mnt = m; | |
7eafd7c7 SH |
1704 | return 0; |
1705 | } | |
1706 | ||
1707 | void mq_clear_sbinfo(struct ipc_namespace *ns) | |
1708 | { | |
cfb2f6f6 | 1709 | ns->mq_mnt->mnt_sb->s_fs_info = NULL; |
7eafd7c7 SH |
1710 | } |
1711 | ||
1712 | void mq_put_mnt(struct ipc_namespace *ns) | |
1713 | { | |
cfb2f6f6 | 1714 | kern_unmount(ns->mq_mnt); |
7eafd7c7 SH |
1715 | } |
1716 | ||
1da177e4 LT |
1717 | static int __init init_mqueue_fs(void) |
1718 | { | |
1719 | int error; | |
1720 | ||
1721 | mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache", | |
1722 | sizeof(struct mqueue_inode_info), 0, | |
5d097056 | 1723 | SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once); |
1da177e4 LT |
1724 | if (mqueue_inode_cachep == NULL) |
1725 | return -ENOMEM; | |
1726 | ||
dc55e35f AG |
1727 | if (!setup_mq_sysctls(&init_ipc_ns)) { |
1728 | pr_warn("sysctl registration failed\n"); | |
12b677f2 ZS |
1729 | error = -ENOMEM; |
1730 | goto out_kmem; | |
dc55e35f | 1731 | } |
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: | |
c579d60f | 1748 | retire_mq_sysctls(&init_ipc_ns); |
12b677f2 ZS |
1749 | out_kmem: |
1750 | kmem_cache_destroy(mqueue_inode_cachep); | |
1da177e4 LT |
1751 | return error; |
1752 | } | |
1753 | ||
6d08a256 | 1754 | device_initcall(init_mqueue_fs); |