]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/ipc/msg.c | |
5a06a363 | 3 | * Copyright (C) 1992 Krishna Balasubramanian |
1da177e4 LT |
4 | * |
5 | * Removed all the remaining kerneld mess | |
6 | * Catch the -EFAULT stuff properly | |
7 | * Use GFP_KERNEL for messages as in 1.2 | |
8 | * Fixed up the unchecked user space derefs | |
9 | * Copyright (C) 1998 Alan Cox & Andi Kleen | |
10 | * | |
11 | * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com> | |
12 | * | |
13 | * mostly rewritten, threaded and wake-one semantics added | |
14 | * MSGMAX limit removed, sysctl's added | |
624dffcb | 15 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
073115d6 SG |
16 | * |
17 | * support for audit of ipc object properties and permission changes | |
18 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
1e786937 KK |
19 | * |
20 | * namespaces support | |
21 | * OpenVZ, SWsoft Inc. | |
22 | * Pavel Emelianov <xemul@openvz.org> | |
1da177e4 LT |
23 | */ |
24 | ||
c59ede7b | 25 | #include <linux/capability.h> |
1da177e4 LT |
26 | #include <linux/slab.h> |
27 | #include <linux/msg.h> | |
28 | #include <linux/spinlock.h> | |
29 | #include <linux/init.h> | |
f7bf3df8 | 30 | #include <linux/mm.h> |
1da177e4 LT |
31 | #include <linux/proc_fs.h> |
32 | #include <linux/list.h> | |
33 | #include <linux/security.h> | |
34 | #include <linux/sched.h> | |
35 | #include <linux/syscalls.h> | |
36 | #include <linux/audit.h> | |
19b4946c | 37 | #include <linux/seq_file.h> |
3e148c79 | 38 | #include <linux/rwsem.h> |
1e786937 | 39 | #include <linux/nsproxy.h> |
ae5e1b22 | 40 | #include <linux/ipc_namespace.h> |
5f921ae9 | 41 | |
1da177e4 LT |
42 | #include <asm/current.h> |
43 | #include <asm/uaccess.h> | |
44 | #include "util.h" | |
45 | ||
5a06a363 IM |
46 | /* |
47 | * one msg_receiver structure for each sleeping receiver: | |
48 | */ | |
1da177e4 | 49 | struct msg_receiver { |
5a06a363 IM |
50 | struct list_head r_list; |
51 | struct task_struct *r_tsk; | |
1da177e4 | 52 | |
5a06a363 IM |
53 | int r_mode; |
54 | long r_msgtype; | |
55 | long r_maxsize; | |
1da177e4 | 56 | |
80491eb9 | 57 | struct msg_msg *volatile r_msg; |
1da177e4 LT |
58 | }; |
59 | ||
60 | /* one msg_sender for each sleeping sender */ | |
61 | struct msg_sender { | |
5a06a363 IM |
62 | struct list_head list; |
63 | struct task_struct *tsk; | |
1da177e4 LT |
64 | }; |
65 | ||
66 | #define SEARCH_ANY 1 | |
67 | #define SEARCH_EQUAL 2 | |
68 | #define SEARCH_NOTEQUAL 3 | |
69 | #define SEARCH_LESSEQUAL 4 | |
70 | ||
ed2ddbf8 | 71 | #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS]) |
1da177e4 | 72 | |
1e786937 | 73 | #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm) |
1e786937 | 74 | |
01b8b07a | 75 | static void freeque(struct ipc_namespace *, struct kern_ipc_perm *); |
7748dbfa | 76 | static int newque(struct ipc_namespace *, struct ipc_params *); |
1da177e4 | 77 | #ifdef CONFIG_PROC_FS |
19b4946c | 78 | static int sysvipc_msg_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
79 | #endif |
80 | ||
f7bf3df8 ND |
81 | /* |
82 | * Scale msgmni with the available lowmem size: the memory dedicated to msg | |
83 | * queues should occupy at most 1/MSG_MEM_SCALE of lowmem. | |
4d89dc6a ND |
84 | * Also take into account the number of nsproxies created so far. |
85 | * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range. | |
f7bf3df8 | 86 | */ |
b6b337ad | 87 | void recompute_msgmni(struct ipc_namespace *ns) |
f7bf3df8 ND |
88 | { |
89 | struct sysinfo i; | |
90 | unsigned long allowed; | |
4d89dc6a | 91 | int nb_ns; |
f7bf3df8 ND |
92 | |
93 | si_meminfo(&i); | |
94 | allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit) | |
95 | / MSGMNB; | |
4d89dc6a ND |
96 | nb_ns = atomic_read(&nr_ipc_ns); |
97 | allowed /= nb_ns; | |
f7bf3df8 ND |
98 | |
99 | if (allowed < MSGMNI) { | |
100 | ns->msg_ctlmni = MSGMNI; | |
101 | goto out_callback; | |
102 | } | |
103 | ||
4d89dc6a ND |
104 | if (allowed > IPCMNI / nb_ns) { |
105 | ns->msg_ctlmni = IPCMNI / nb_ns; | |
f7bf3df8 ND |
106 | goto out_callback; |
107 | } | |
108 | ||
109 | ns->msg_ctlmni = allowed; | |
110 | ||
111 | out_callback: | |
112 | ||
113 | printk(KERN_INFO "msgmni has been set to %d for ipc namespace %p\n", | |
114 | ns->msg_ctlmni, ns); | |
115 | } | |
116 | ||
ed2ddbf8 | 117 | void msg_init_ns(struct ipc_namespace *ns) |
1e786937 | 118 | { |
1e786937 KK |
119 | ns->msg_ctlmax = MSGMAX; |
120 | ns->msg_ctlmnb = MSGMNB; | |
f7bf3df8 ND |
121 | |
122 | recompute_msgmni(ns); | |
123 | ||
3ac88a41 KK |
124 | atomic_set(&ns->msg_bytes, 0); |
125 | atomic_set(&ns->msg_hdrs, 0); | |
ed2ddbf8 | 126 | ipc_init_ids(&ns->ids[IPC_MSG_IDS]); |
1e786937 KK |
127 | } |
128 | ||
ae5e1b22 | 129 | #ifdef CONFIG_IPC_NS |
1e786937 KK |
130 | void msg_exit_ns(struct ipc_namespace *ns) |
131 | { | |
01b8b07a | 132 | free_ipcs(ns, &msg_ids(ns), freeque); |
1e786937 | 133 | } |
ae5e1b22 | 134 | #endif |
1e786937 | 135 | |
5a06a363 | 136 | void __init msg_init(void) |
1da177e4 | 137 | { |
ed2ddbf8 | 138 | msg_init_ns(&init_ipc_ns); |
19b4946c MW |
139 | ipc_init_proc_interface("sysvipc/msg", |
140 | " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n", | |
1e786937 | 141 | IPC_MSG_IDS, sysvipc_msg_proc_show); |
1da177e4 LT |
142 | } |
143 | ||
3e148c79 ND |
144 | /* |
145 | * This routine is called in the paths where the rw_mutex is held to protect | |
146 | * access to the idr tree. | |
147 | */ | |
148 | static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns, | |
149 | int id) | |
150 | { | |
151 | struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id); | |
152 | ||
b1ed88b4 PP |
153 | if (IS_ERR(ipcp)) |
154 | return (struct msg_queue *)ipcp; | |
155 | ||
3e148c79 ND |
156 | return container_of(ipcp, struct msg_queue, q_perm); |
157 | } | |
158 | ||
159 | /* | |
160 | * msg_lock_(check_) routines are called in the paths where the rw_mutex | |
161 | * is not held. | |
162 | */ | |
023a5355 ND |
163 | static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id) |
164 | { | |
03f02c76 ND |
165 | struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id); |
166 | ||
b1ed88b4 PP |
167 | if (IS_ERR(ipcp)) |
168 | return (struct msg_queue *)ipcp; | |
169 | ||
03f02c76 | 170 | return container_of(ipcp, struct msg_queue, q_perm); |
023a5355 ND |
171 | } |
172 | ||
173 | static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns, | |
174 | int id) | |
175 | { | |
03f02c76 ND |
176 | struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id); |
177 | ||
b1ed88b4 PP |
178 | if (IS_ERR(ipcp)) |
179 | return (struct msg_queue *)ipcp; | |
180 | ||
03f02c76 | 181 | return container_of(ipcp, struct msg_queue, q_perm); |
023a5355 ND |
182 | } |
183 | ||
7ca7e564 ND |
184 | static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s) |
185 | { | |
186 | ipc_rmid(&msg_ids(ns), &s->q_perm); | |
187 | } | |
188 | ||
f4566f04 ND |
189 | /** |
190 | * newque - Create a new msg queue | |
191 | * @ns: namespace | |
192 | * @params: ptr to the structure that contains the key and msgflg | |
193 | * | |
3e148c79 | 194 | * Called with msg_ids.rw_mutex held (writer) |
f4566f04 | 195 | */ |
7748dbfa | 196 | static int newque(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 | 197 | { |
1da177e4 | 198 | struct msg_queue *msq; |
5a06a363 | 199 | int id, retval; |
7748dbfa ND |
200 | key_t key = params->key; |
201 | int msgflg = params->flg; | |
1da177e4 | 202 | |
5a06a363 IM |
203 | msq = ipc_rcu_alloc(sizeof(*msq)); |
204 | if (!msq) | |
1da177e4 LT |
205 | return -ENOMEM; |
206 | ||
5a06a363 | 207 | msq->q_perm.mode = msgflg & S_IRWXUGO; |
1da177e4 LT |
208 | msq->q_perm.key = key; |
209 | ||
210 | msq->q_perm.security = NULL; | |
211 | retval = security_msg_queue_alloc(msq); | |
212 | if (retval) { | |
213 | ipc_rcu_putref(msq); | |
214 | return retval; | |
215 | } | |
216 | ||
7ca7e564 ND |
217 | /* |
218 | * ipc_addid() locks msq | |
219 | */ | |
1e786937 | 220 | id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni); |
283bb7fa | 221 | if (id < 0) { |
1da177e4 LT |
222 | security_msg_queue_free(msq); |
223 | ipc_rcu_putref(msq); | |
283bb7fa | 224 | return id; |
1da177e4 LT |
225 | } |
226 | ||
227 | msq->q_stime = msq->q_rtime = 0; | |
228 | msq->q_ctime = get_seconds(); | |
229 | msq->q_cbytes = msq->q_qnum = 0; | |
1e786937 | 230 | msq->q_qbytes = ns->msg_ctlmnb; |
1da177e4 LT |
231 | msq->q_lspid = msq->q_lrpid = 0; |
232 | INIT_LIST_HEAD(&msq->q_messages); | |
233 | INIT_LIST_HEAD(&msq->q_receivers); | |
234 | INIT_LIST_HEAD(&msq->q_senders); | |
7ca7e564 | 235 | |
1da177e4 LT |
236 | msg_unlock(msq); |
237 | ||
7ca7e564 | 238 | return msq->q_perm.id; |
1da177e4 LT |
239 | } |
240 | ||
5a06a363 | 241 | static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss) |
1da177e4 | 242 | { |
5a06a363 IM |
243 | mss->tsk = current; |
244 | current->state = TASK_INTERRUPTIBLE; | |
245 | list_add_tail(&mss->list, &msq->q_senders); | |
1da177e4 LT |
246 | } |
247 | ||
5a06a363 | 248 | static inline void ss_del(struct msg_sender *mss) |
1da177e4 | 249 | { |
5a06a363 | 250 | if (mss->list.next != NULL) |
1da177e4 LT |
251 | list_del(&mss->list); |
252 | } | |
253 | ||
5a06a363 | 254 | static void ss_wakeup(struct list_head *h, int kill) |
1da177e4 LT |
255 | { |
256 | struct list_head *tmp; | |
257 | ||
258 | tmp = h->next; | |
259 | while (tmp != h) { | |
5a06a363 IM |
260 | struct msg_sender *mss; |
261 | ||
262 | mss = list_entry(tmp, struct msg_sender, list); | |
1da177e4 | 263 | tmp = tmp->next; |
5a06a363 IM |
264 | if (kill) |
265 | mss->list.next = NULL; | |
1da177e4 LT |
266 | wake_up_process(mss->tsk); |
267 | } | |
268 | } | |
269 | ||
5a06a363 | 270 | static void expunge_all(struct msg_queue *msq, int res) |
1da177e4 LT |
271 | { |
272 | struct list_head *tmp; | |
273 | ||
274 | tmp = msq->q_receivers.next; | |
275 | while (tmp != &msq->q_receivers) { | |
5a06a363 IM |
276 | struct msg_receiver *msr; |
277 | ||
278 | msr = list_entry(tmp, struct msg_receiver, r_list); | |
1da177e4 LT |
279 | tmp = tmp->next; |
280 | msr->r_msg = NULL; | |
281 | wake_up_process(msr->r_tsk); | |
282 | smp_mb(); | |
283 | msr->r_msg = ERR_PTR(res); | |
284 | } | |
285 | } | |
5a06a363 IM |
286 | |
287 | /* | |
288 | * freeque() wakes up waiters on the sender and receiver waiting queue, | |
f4566f04 ND |
289 | * removes the message queue from message queue ID IDR, and cleans up all the |
290 | * messages associated with this queue. | |
1da177e4 | 291 | * |
3e148c79 ND |
292 | * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held |
293 | * before freeque() is called. msg_ids.rw_mutex remains locked on exit. | |
1da177e4 | 294 | */ |
01b8b07a | 295 | static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 LT |
296 | { |
297 | struct list_head *tmp; | |
01b8b07a | 298 | struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm); |
1da177e4 | 299 | |
5a06a363 IM |
300 | expunge_all(msq, -EIDRM); |
301 | ss_wakeup(&msq->q_senders, 1); | |
7ca7e564 | 302 | msg_rmid(ns, msq); |
1da177e4 | 303 | msg_unlock(msq); |
5a06a363 | 304 | |
1da177e4 | 305 | tmp = msq->q_messages.next; |
5a06a363 IM |
306 | while (tmp != &msq->q_messages) { |
307 | struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list); | |
308 | ||
1da177e4 | 309 | tmp = tmp->next; |
3ac88a41 | 310 | atomic_dec(&ns->msg_hdrs); |
1da177e4 LT |
311 | free_msg(msg); |
312 | } | |
3ac88a41 | 313 | atomic_sub(msq->q_cbytes, &ns->msg_bytes); |
1da177e4 LT |
314 | security_msg_queue_free(msq); |
315 | ipc_rcu_putref(msq); | |
316 | } | |
317 | ||
f4566f04 | 318 | /* |
3e148c79 | 319 | * Called with msg_ids.rw_mutex and ipcp locked. |
f4566f04 | 320 | */ |
03f02c76 | 321 | static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg) |
7748dbfa | 322 | { |
03f02c76 ND |
323 | struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm); |
324 | ||
325 | return security_msg_queue_associate(msq, msgflg); | |
7748dbfa ND |
326 | } |
327 | ||
5a06a363 | 328 | asmlinkage long sys_msgget(key_t key, int msgflg) |
1da177e4 | 329 | { |
1e786937 | 330 | struct ipc_namespace *ns; |
7748dbfa ND |
331 | struct ipc_ops msg_ops; |
332 | struct ipc_params msg_params; | |
1e786937 KK |
333 | |
334 | ns = current->nsproxy->ipc_ns; | |
7ca7e564 | 335 | |
7748dbfa ND |
336 | msg_ops.getnew = newque; |
337 | msg_ops.associate = msg_security; | |
338 | msg_ops.more_checks = NULL; | |
339 | ||
340 | msg_params.key = key; | |
341 | msg_params.flg = msgflg; | |
5a06a363 | 342 | |
7748dbfa | 343 | return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params); |
1da177e4 LT |
344 | } |
345 | ||
5a06a363 IM |
346 | static inline unsigned long |
347 | copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version) | |
1da177e4 LT |
348 | { |
349 | switch(version) { | |
350 | case IPC_64: | |
5a06a363 | 351 | return copy_to_user(buf, in, sizeof(*in)); |
1da177e4 | 352 | case IPC_OLD: |
5a06a363 | 353 | { |
1da177e4 LT |
354 | struct msqid_ds out; |
355 | ||
5a06a363 | 356 | memset(&out, 0, sizeof(out)); |
1da177e4 LT |
357 | |
358 | ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm); | |
359 | ||
360 | out.msg_stime = in->msg_stime; | |
361 | out.msg_rtime = in->msg_rtime; | |
362 | out.msg_ctime = in->msg_ctime; | |
363 | ||
5a06a363 | 364 | if (in->msg_cbytes > USHRT_MAX) |
1da177e4 LT |
365 | out.msg_cbytes = USHRT_MAX; |
366 | else | |
367 | out.msg_cbytes = in->msg_cbytes; | |
368 | out.msg_lcbytes = in->msg_cbytes; | |
369 | ||
5a06a363 | 370 | if (in->msg_qnum > USHRT_MAX) |
1da177e4 LT |
371 | out.msg_qnum = USHRT_MAX; |
372 | else | |
373 | out.msg_qnum = in->msg_qnum; | |
374 | ||
5a06a363 | 375 | if (in->msg_qbytes > USHRT_MAX) |
1da177e4 LT |
376 | out.msg_qbytes = USHRT_MAX; |
377 | else | |
378 | out.msg_qbytes = in->msg_qbytes; | |
379 | out.msg_lqbytes = in->msg_qbytes; | |
380 | ||
381 | out.msg_lspid = in->msg_lspid; | |
382 | out.msg_lrpid = in->msg_lrpid; | |
383 | ||
5a06a363 IM |
384 | return copy_to_user(buf, &out, sizeof(out)); |
385 | } | |
1da177e4 LT |
386 | default: |
387 | return -EINVAL; | |
388 | } | |
389 | } | |
390 | ||
391 | struct msq_setbuf { | |
392 | unsigned long qbytes; | |
393 | uid_t uid; | |
394 | gid_t gid; | |
395 | mode_t mode; | |
396 | }; | |
397 | ||
5a06a363 IM |
398 | static inline unsigned long |
399 | copy_msqid_from_user(struct msq_setbuf *out, void __user *buf, int version) | |
1da177e4 LT |
400 | { |
401 | switch(version) { | |
402 | case IPC_64: | |
5a06a363 | 403 | { |
1da177e4 LT |
404 | struct msqid64_ds tbuf; |
405 | ||
5a06a363 | 406 | if (copy_from_user(&tbuf, buf, sizeof(tbuf))) |
1da177e4 LT |
407 | return -EFAULT; |
408 | ||
409 | out->qbytes = tbuf.msg_qbytes; | |
410 | out->uid = tbuf.msg_perm.uid; | |
411 | out->gid = tbuf.msg_perm.gid; | |
412 | out->mode = tbuf.msg_perm.mode; | |
413 | ||
414 | return 0; | |
5a06a363 | 415 | } |
1da177e4 | 416 | case IPC_OLD: |
5a06a363 | 417 | { |
1da177e4 LT |
418 | struct msqid_ds tbuf_old; |
419 | ||
5a06a363 | 420 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
421 | return -EFAULT; |
422 | ||
423 | out->uid = tbuf_old.msg_perm.uid; | |
424 | out->gid = tbuf_old.msg_perm.gid; | |
425 | out->mode = tbuf_old.msg_perm.mode; | |
426 | ||
5a06a363 | 427 | if (tbuf_old.msg_qbytes == 0) |
1da177e4 LT |
428 | out->qbytes = tbuf_old.msg_lqbytes; |
429 | else | |
430 | out->qbytes = tbuf_old.msg_qbytes; | |
431 | ||
432 | return 0; | |
5a06a363 | 433 | } |
1da177e4 LT |
434 | default: |
435 | return -EINVAL; | |
436 | } | |
437 | } | |
438 | ||
a0d092fc PP |
439 | /* |
440 | * This function handles some msgctl commands which require the rw_mutex | |
441 | * to be held in write mode. | |
442 | * NOTE: no locks must be held, the rw_mutex is taken inside this function. | |
443 | */ | |
444 | static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd, | |
445 | struct msqid_ds __user *buf, int version) | |
1da177e4 | 446 | { |
1da177e4 | 447 | struct kern_ipc_perm *ipcp; |
a0d092fc PP |
448 | struct msq_setbuf setbuf; |
449 | struct msg_queue *msq; | |
450 | int err; | |
451 | ||
452 | if (cmd == IPC_SET) { | |
453 | if (copy_msqid_from_user(&setbuf, buf, version)) | |
454 | return -EFAULT; | |
455 | } | |
456 | ||
457 | down_write(&msg_ids(ns).rw_mutex); | |
458 | msq = msg_lock_check_down(ns, msqid); | |
459 | if (IS_ERR(msq)) { | |
460 | err = PTR_ERR(msq); | |
461 | goto out_up; | |
462 | } | |
463 | ||
464 | ipcp = &msq->q_perm; | |
465 | ||
466 | err = audit_ipc_obj(ipcp); | |
467 | if (err) | |
468 | goto out_unlock; | |
469 | ||
470 | if (cmd == IPC_SET) { | |
471 | err = audit_ipc_set_perm(setbuf.qbytes, setbuf.uid, setbuf.gid, | |
472 | setbuf.mode); | |
473 | if (err) | |
474 | goto out_unlock; | |
475 | } | |
476 | ||
477 | if (current->euid != ipcp->cuid && | |
478 | current->euid != ipcp->uid && | |
479 | !capable(CAP_SYS_ADMIN)) { | |
480 | /* We _could_ check for CAP_CHOWN above, but we don't */ | |
481 | err = -EPERM; | |
482 | goto out_unlock; | |
483 | } | |
484 | ||
485 | err = security_msg_queue_msgctl(msq, cmd); | |
486 | if (err) | |
487 | goto out_unlock; | |
488 | ||
489 | switch (cmd) { | |
490 | case IPC_RMID: | |
491 | freeque(ns, ipcp); | |
492 | goto out_up; | |
493 | case IPC_SET: | |
494 | if (setbuf.qbytes > ns->msg_ctlmnb && | |
495 | !capable(CAP_SYS_RESOURCE)) { | |
496 | err = -EPERM; | |
497 | goto out_unlock; | |
498 | } | |
499 | ||
500 | msq->q_qbytes = setbuf.qbytes; | |
501 | ||
502 | ipcp->uid = setbuf.uid; | |
503 | ipcp->gid = setbuf.gid; | |
504 | ipcp->mode = (ipcp->mode & ~S_IRWXUGO) | | |
505 | (S_IRWXUGO & setbuf.mode); | |
506 | msq->q_ctime = get_seconds(); | |
507 | /* sleeping receivers might be excluded by | |
508 | * stricter permissions. | |
509 | */ | |
510 | expunge_all(msq, -EAGAIN); | |
511 | /* sleeping senders might be able to send | |
512 | * due to a larger queue size. | |
513 | */ | |
514 | ss_wakeup(&msq->q_senders, 0); | |
515 | break; | |
516 | default: | |
517 | err = -EINVAL; | |
518 | } | |
519 | out_unlock: | |
520 | msg_unlock(msq); | |
521 | out_up: | |
522 | up_write(&msg_ids(ns).rw_mutex); | |
523 | return err; | |
524 | } | |
525 | ||
526 | asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf) | |
527 | { | |
5a06a363 IM |
528 | struct msg_queue *msq; |
529 | int err, version; | |
1e786937 | 530 | struct ipc_namespace *ns; |
5a06a363 | 531 | |
1da177e4 LT |
532 | if (msqid < 0 || cmd < 0) |
533 | return -EINVAL; | |
534 | ||
535 | version = ipc_parse_version(&cmd); | |
1e786937 | 536 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
537 | |
538 | switch (cmd) { | |
5a06a363 IM |
539 | case IPC_INFO: |
540 | case MSG_INFO: | |
541 | { | |
1da177e4 LT |
542 | struct msginfo msginfo; |
543 | int max_id; | |
5a06a363 | 544 | |
1da177e4 LT |
545 | if (!buf) |
546 | return -EFAULT; | |
5a06a363 IM |
547 | /* |
548 | * We must not return kernel stack data. | |
1da177e4 LT |
549 | * due to padding, it's not enough |
550 | * to set all member fields. | |
551 | */ | |
1da177e4 LT |
552 | err = security_msg_queue_msgctl(NULL, cmd); |
553 | if (err) | |
554 | return err; | |
555 | ||
5a06a363 | 556 | memset(&msginfo, 0, sizeof(msginfo)); |
1e786937 KK |
557 | msginfo.msgmni = ns->msg_ctlmni; |
558 | msginfo.msgmax = ns->msg_ctlmax; | |
559 | msginfo.msgmnb = ns->msg_ctlmnb; | |
1da177e4 LT |
560 | msginfo.msgssz = MSGSSZ; |
561 | msginfo.msgseg = MSGSEG; | |
3e148c79 | 562 | down_read(&msg_ids(ns).rw_mutex); |
1da177e4 | 563 | if (cmd == MSG_INFO) { |
1e786937 | 564 | msginfo.msgpool = msg_ids(ns).in_use; |
3ac88a41 KK |
565 | msginfo.msgmap = atomic_read(&ns->msg_hdrs); |
566 | msginfo.msgtql = atomic_read(&ns->msg_bytes); | |
1da177e4 LT |
567 | } else { |
568 | msginfo.msgmap = MSGMAP; | |
569 | msginfo.msgpool = MSGPOOL; | |
570 | msginfo.msgtql = MSGTQL; | |
571 | } | |
7ca7e564 | 572 | max_id = ipc_get_maxid(&msg_ids(ns)); |
3e148c79 | 573 | up_read(&msg_ids(ns).rw_mutex); |
5a06a363 | 574 | if (copy_to_user(buf, &msginfo, sizeof(struct msginfo))) |
1da177e4 | 575 | return -EFAULT; |
5a06a363 | 576 | return (max_id < 0) ? 0 : max_id; |
1da177e4 | 577 | } |
7ca7e564 | 578 | case MSG_STAT: /* msqid is an index rather than a msg queue id */ |
1da177e4 LT |
579 | case IPC_STAT: |
580 | { | |
581 | struct msqid64_ds tbuf; | |
582 | int success_return; | |
5a06a363 | 583 | |
1da177e4 LT |
584 | if (!buf) |
585 | return -EFAULT; | |
1da177e4 | 586 | |
5a06a363 | 587 | if (cmd == MSG_STAT) { |
023a5355 ND |
588 | msq = msg_lock(ns, msqid); |
589 | if (IS_ERR(msq)) | |
590 | return PTR_ERR(msq); | |
7ca7e564 | 591 | success_return = msq->q_perm.id; |
1da177e4 | 592 | } else { |
023a5355 ND |
593 | msq = msg_lock_check(ns, msqid); |
594 | if (IS_ERR(msq)) | |
595 | return PTR_ERR(msq); | |
1da177e4 LT |
596 | success_return = 0; |
597 | } | |
598 | err = -EACCES; | |
5a06a363 | 599 | if (ipcperms(&msq->q_perm, S_IRUGO)) |
1da177e4 LT |
600 | goto out_unlock; |
601 | ||
602 | err = security_msg_queue_msgctl(msq, cmd); | |
603 | if (err) | |
604 | goto out_unlock; | |
605 | ||
023a5355 ND |
606 | memset(&tbuf, 0, sizeof(tbuf)); |
607 | ||
1da177e4 LT |
608 | kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm); |
609 | tbuf.msg_stime = msq->q_stime; | |
610 | tbuf.msg_rtime = msq->q_rtime; | |
611 | tbuf.msg_ctime = msq->q_ctime; | |
612 | tbuf.msg_cbytes = msq->q_cbytes; | |
613 | tbuf.msg_qnum = msq->q_qnum; | |
614 | tbuf.msg_qbytes = msq->q_qbytes; | |
615 | tbuf.msg_lspid = msq->q_lspid; | |
616 | tbuf.msg_lrpid = msq->q_lrpid; | |
617 | msg_unlock(msq); | |
618 | if (copy_msqid_to_user(buf, &tbuf, version)) | |
619 | return -EFAULT; | |
620 | return success_return; | |
621 | } | |
622 | case IPC_SET: | |
1da177e4 | 623 | case IPC_RMID: |
a0d092fc PP |
624 | err = msgctl_down(ns, msqid, cmd, buf, version); |
625 | return err; | |
1da177e4 LT |
626 | default: |
627 | return -EINVAL; | |
628 | } | |
629 | ||
1da177e4 LT |
630 | out_unlock: |
631 | msg_unlock(msq); | |
632 | return err; | |
633 | } | |
634 | ||
5a06a363 | 635 | static int testmsg(struct msg_msg *msg, long type, int mode) |
1da177e4 LT |
636 | { |
637 | switch(mode) | |
638 | { | |
639 | case SEARCH_ANY: | |
640 | return 1; | |
641 | case SEARCH_LESSEQUAL: | |
5a06a363 | 642 | if (msg->m_type <=type) |
1da177e4 LT |
643 | return 1; |
644 | break; | |
645 | case SEARCH_EQUAL: | |
5a06a363 | 646 | if (msg->m_type == type) |
1da177e4 LT |
647 | return 1; |
648 | break; | |
649 | case SEARCH_NOTEQUAL: | |
5a06a363 | 650 | if (msg->m_type != type) |
1da177e4 LT |
651 | return 1; |
652 | break; | |
653 | } | |
654 | return 0; | |
655 | } | |
656 | ||
5a06a363 | 657 | static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg) |
1da177e4 | 658 | { |
5a06a363 | 659 | struct list_head *tmp; |
1da177e4 LT |
660 | |
661 | tmp = msq->q_receivers.next; | |
662 | while (tmp != &msq->q_receivers) { | |
5a06a363 IM |
663 | struct msg_receiver *msr; |
664 | ||
665 | msr = list_entry(tmp, struct msg_receiver, r_list); | |
1da177e4 | 666 | tmp = tmp->next; |
5a06a363 IM |
667 | if (testmsg(msg, msr->r_msgtype, msr->r_mode) && |
668 | !security_msg_queue_msgrcv(msq, msg, msr->r_tsk, | |
669 | msr->r_msgtype, msr->r_mode)) { | |
670 | ||
1da177e4 | 671 | list_del(&msr->r_list); |
5a06a363 | 672 | if (msr->r_maxsize < msg->m_ts) { |
1da177e4 LT |
673 | msr->r_msg = NULL; |
674 | wake_up_process(msr->r_tsk); | |
675 | smp_mb(); | |
676 | msr->r_msg = ERR_PTR(-E2BIG); | |
677 | } else { | |
678 | msr->r_msg = NULL; | |
b488893a | 679 | msq->q_lrpid = task_pid_vnr(msr->r_tsk); |
1da177e4 LT |
680 | msq->q_rtime = get_seconds(); |
681 | wake_up_process(msr->r_tsk); | |
682 | smp_mb(); | |
683 | msr->r_msg = msg; | |
5a06a363 | 684 | |
1da177e4 LT |
685 | return 1; |
686 | } | |
687 | } | |
688 | } | |
689 | return 0; | |
690 | } | |
691 | ||
651971cb | 692 | long do_msgsnd(int msqid, long mtype, void __user *mtext, |
693 | size_t msgsz, int msgflg) | |
1da177e4 LT |
694 | { |
695 | struct msg_queue *msq; | |
696 | struct msg_msg *msg; | |
1da177e4 | 697 | int err; |
1e786937 KK |
698 | struct ipc_namespace *ns; |
699 | ||
700 | ns = current->nsproxy->ipc_ns; | |
5a06a363 | 701 | |
1e786937 | 702 | if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0) |
1da177e4 | 703 | return -EINVAL; |
1da177e4 LT |
704 | if (mtype < 1) |
705 | return -EINVAL; | |
706 | ||
651971cb | 707 | msg = load_msg(mtext, msgsz); |
5a06a363 | 708 | if (IS_ERR(msg)) |
1da177e4 LT |
709 | return PTR_ERR(msg); |
710 | ||
711 | msg->m_type = mtype; | |
712 | msg->m_ts = msgsz; | |
713 | ||
023a5355 ND |
714 | msq = msg_lock_check(ns, msqid); |
715 | if (IS_ERR(msq)) { | |
716 | err = PTR_ERR(msq); | |
1da177e4 | 717 | goto out_free; |
023a5355 | 718 | } |
1da177e4 LT |
719 | |
720 | for (;;) { | |
721 | struct msg_sender s; | |
722 | ||
5a06a363 | 723 | err = -EACCES; |
1da177e4 LT |
724 | if (ipcperms(&msq->q_perm, S_IWUGO)) |
725 | goto out_unlock_free; | |
726 | ||
727 | err = security_msg_queue_msgsnd(msq, msg, msgflg); | |
728 | if (err) | |
729 | goto out_unlock_free; | |
730 | ||
5a06a363 | 731 | if (msgsz + msq->q_cbytes <= msq->q_qbytes && |
1da177e4 LT |
732 | 1 + msq->q_qnum <= msq->q_qbytes) { |
733 | break; | |
734 | } | |
735 | ||
736 | /* queue full, wait: */ | |
5a06a363 IM |
737 | if (msgflg & IPC_NOWAIT) { |
738 | err = -EAGAIN; | |
1da177e4 LT |
739 | goto out_unlock_free; |
740 | } | |
741 | ss_add(msq, &s); | |
742 | ipc_rcu_getref(msq); | |
743 | msg_unlock(msq); | |
744 | schedule(); | |
745 | ||
746 | ipc_lock_by_ptr(&msq->q_perm); | |
747 | ipc_rcu_putref(msq); | |
748 | if (msq->q_perm.deleted) { | |
749 | err = -EIDRM; | |
750 | goto out_unlock_free; | |
751 | } | |
752 | ss_del(&s); | |
5a06a363 | 753 | |
1da177e4 | 754 | if (signal_pending(current)) { |
5a06a363 | 755 | err = -ERESTARTNOHAND; |
1da177e4 LT |
756 | goto out_unlock_free; |
757 | } | |
758 | } | |
759 | ||
b488893a | 760 | msq->q_lspid = task_tgid_vnr(current); |
1da177e4 LT |
761 | msq->q_stime = get_seconds(); |
762 | ||
5a06a363 | 763 | if (!pipelined_send(msq, msg)) { |
1da177e4 | 764 | /* noone is waiting for this message, enqueue it */ |
5a06a363 | 765 | list_add_tail(&msg->m_list, &msq->q_messages); |
1da177e4 LT |
766 | msq->q_cbytes += msgsz; |
767 | msq->q_qnum++; | |
3ac88a41 KK |
768 | atomic_add(msgsz, &ns->msg_bytes); |
769 | atomic_inc(&ns->msg_hdrs); | |
1da177e4 | 770 | } |
5a06a363 | 771 | |
1da177e4 LT |
772 | err = 0; |
773 | msg = NULL; | |
774 | ||
775 | out_unlock_free: | |
776 | msg_unlock(msq); | |
777 | out_free: | |
5a06a363 | 778 | if (msg != NULL) |
1da177e4 LT |
779 | free_msg(msg); |
780 | return err; | |
781 | } | |
782 | ||
651971cb | 783 | asmlinkage long |
784 | sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg) | |
785 | { | |
786 | long mtype; | |
787 | ||
788 | if (get_user(mtype, &msgp->mtype)) | |
789 | return -EFAULT; | |
790 | return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg); | |
791 | } | |
792 | ||
5a06a363 | 793 | static inline int convert_mode(long *msgtyp, int msgflg) |
1da177e4 | 794 | { |
5a06a363 | 795 | /* |
1da177e4 LT |
796 | * find message of correct type. |
797 | * msgtyp = 0 => get first. | |
798 | * msgtyp > 0 => get first message of matching type. | |
5a06a363 | 799 | * msgtyp < 0 => get message with least type must be < abs(msgtype). |
1da177e4 | 800 | */ |
5a06a363 | 801 | if (*msgtyp == 0) |
1da177e4 | 802 | return SEARCH_ANY; |
5a06a363 IM |
803 | if (*msgtyp < 0) { |
804 | *msgtyp = -*msgtyp; | |
1da177e4 LT |
805 | return SEARCH_LESSEQUAL; |
806 | } | |
5a06a363 | 807 | if (msgflg & MSG_EXCEPT) |
1da177e4 LT |
808 | return SEARCH_NOTEQUAL; |
809 | return SEARCH_EQUAL; | |
810 | } | |
811 | ||
651971cb | 812 | long do_msgrcv(int msqid, long *pmtype, void __user *mtext, |
813 | size_t msgsz, long msgtyp, int msgflg) | |
1da177e4 LT |
814 | { |
815 | struct msg_queue *msq; | |
816 | struct msg_msg *msg; | |
817 | int mode; | |
1e786937 | 818 | struct ipc_namespace *ns; |
1da177e4 LT |
819 | |
820 | if (msqid < 0 || (long) msgsz < 0) | |
821 | return -EINVAL; | |
5a06a363 | 822 | mode = convert_mode(&msgtyp, msgflg); |
1e786937 | 823 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 824 | |
023a5355 ND |
825 | msq = msg_lock_check(ns, msqid); |
826 | if (IS_ERR(msq)) | |
827 | return PTR_ERR(msq); | |
1da177e4 LT |
828 | |
829 | for (;;) { | |
830 | struct msg_receiver msr_d; | |
5a06a363 | 831 | struct list_head *tmp; |
1da177e4 LT |
832 | |
833 | msg = ERR_PTR(-EACCES); | |
5a06a363 | 834 | if (ipcperms(&msq->q_perm, S_IRUGO)) |
1da177e4 LT |
835 | goto out_unlock; |
836 | ||
837 | msg = ERR_PTR(-EAGAIN); | |
838 | tmp = msq->q_messages.next; | |
839 | while (tmp != &msq->q_messages) { | |
840 | struct msg_msg *walk_msg; | |
5a06a363 IM |
841 | |
842 | walk_msg = list_entry(tmp, struct msg_msg, m_list); | |
843 | if (testmsg(walk_msg, msgtyp, mode) && | |
844 | !security_msg_queue_msgrcv(msq, walk_msg, current, | |
845 | msgtyp, mode)) { | |
846 | ||
1da177e4 | 847 | msg = walk_msg; |
5a06a363 IM |
848 | if (mode == SEARCH_LESSEQUAL && |
849 | walk_msg->m_type != 1) { | |
850 | msg = walk_msg; | |
851 | msgtyp = walk_msg->m_type - 1; | |
1da177e4 | 852 | } else { |
5a06a363 | 853 | msg = walk_msg; |
1da177e4 LT |
854 | break; |
855 | } | |
856 | } | |
857 | tmp = tmp->next; | |
858 | } | |
5a06a363 IM |
859 | if (!IS_ERR(msg)) { |
860 | /* | |
861 | * Found a suitable message. | |
862 | * Unlink it from the queue. | |
863 | */ | |
1da177e4 LT |
864 | if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) { |
865 | msg = ERR_PTR(-E2BIG); | |
866 | goto out_unlock; | |
867 | } | |
868 | list_del(&msg->m_list); | |
869 | msq->q_qnum--; | |
870 | msq->q_rtime = get_seconds(); | |
b488893a | 871 | msq->q_lrpid = task_tgid_vnr(current); |
1da177e4 | 872 | msq->q_cbytes -= msg->m_ts; |
3ac88a41 KK |
873 | atomic_sub(msg->m_ts, &ns->msg_bytes); |
874 | atomic_dec(&ns->msg_hdrs); | |
5a06a363 | 875 | ss_wakeup(&msq->q_senders, 0); |
1da177e4 LT |
876 | msg_unlock(msq); |
877 | break; | |
878 | } | |
879 | /* No message waiting. Wait for a message */ | |
880 | if (msgflg & IPC_NOWAIT) { | |
881 | msg = ERR_PTR(-ENOMSG); | |
882 | goto out_unlock; | |
883 | } | |
5a06a363 | 884 | list_add_tail(&msr_d.r_list, &msq->q_receivers); |
1da177e4 LT |
885 | msr_d.r_tsk = current; |
886 | msr_d.r_msgtype = msgtyp; | |
887 | msr_d.r_mode = mode; | |
5a06a363 | 888 | if (msgflg & MSG_NOERROR) |
1da177e4 | 889 | msr_d.r_maxsize = INT_MAX; |
5a06a363 | 890 | else |
1da177e4 LT |
891 | msr_d.r_maxsize = msgsz; |
892 | msr_d.r_msg = ERR_PTR(-EAGAIN); | |
893 | current->state = TASK_INTERRUPTIBLE; | |
894 | msg_unlock(msq); | |
895 | ||
896 | schedule(); | |
897 | ||
898 | /* Lockless receive, part 1: | |
899 | * Disable preemption. We don't hold a reference to the queue | |
900 | * and getting a reference would defeat the idea of a lockless | |
901 | * operation, thus the code relies on rcu to guarantee the | |
902 | * existance of msq: | |
903 | * Prior to destruction, expunge_all(-EIRDM) changes r_msg. | |
904 | * Thus if r_msg is -EAGAIN, then the queue not yet destroyed. | |
905 | * rcu_read_lock() prevents preemption between reading r_msg | |
906 | * and the spin_lock() inside ipc_lock_by_ptr(). | |
907 | */ | |
908 | rcu_read_lock(); | |
909 | ||
910 | /* Lockless receive, part 2: | |
911 | * Wait until pipelined_send or expunge_all are outside of | |
912 | * wake_up_process(). There is a race with exit(), see | |
913 | * ipc/mqueue.c for the details. | |
914 | */ | |
5a06a363 | 915 | msg = (struct msg_msg*)msr_d.r_msg; |
1da177e4 LT |
916 | while (msg == NULL) { |
917 | cpu_relax(); | |
5a06a363 | 918 | msg = (struct msg_msg *)msr_d.r_msg; |
1da177e4 LT |
919 | } |
920 | ||
921 | /* Lockless receive, part 3: | |
922 | * If there is a message or an error then accept it without | |
923 | * locking. | |
924 | */ | |
5a06a363 | 925 | if (msg != ERR_PTR(-EAGAIN)) { |
1da177e4 LT |
926 | rcu_read_unlock(); |
927 | break; | |
928 | } | |
929 | ||
930 | /* Lockless receive, part 3: | |
931 | * Acquire the queue spinlock. | |
932 | */ | |
933 | ipc_lock_by_ptr(&msq->q_perm); | |
934 | rcu_read_unlock(); | |
935 | ||
936 | /* Lockless receive, part 4: | |
937 | * Repeat test after acquiring the spinlock. | |
938 | */ | |
939 | msg = (struct msg_msg*)msr_d.r_msg; | |
5a06a363 | 940 | if (msg != ERR_PTR(-EAGAIN)) |
1da177e4 LT |
941 | goto out_unlock; |
942 | ||
943 | list_del(&msr_d.r_list); | |
944 | if (signal_pending(current)) { | |
945 | msg = ERR_PTR(-ERESTARTNOHAND); | |
946 | out_unlock: | |
947 | msg_unlock(msq); | |
948 | break; | |
949 | } | |
950 | } | |
951 | if (IS_ERR(msg)) | |
5a06a363 | 952 | return PTR_ERR(msg); |
1da177e4 LT |
953 | |
954 | msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz; | |
651971cb | 955 | *pmtype = msg->m_type; |
956 | if (store_msg(mtext, msg, msgsz)) | |
5a06a363 | 957 | msgsz = -EFAULT; |
651971cb | 958 | |
1da177e4 | 959 | free_msg(msg); |
5a06a363 | 960 | |
1da177e4 LT |
961 | return msgsz; |
962 | } | |
963 | ||
651971cb | 964 | asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz, |
965 | long msgtyp, int msgflg) | |
966 | { | |
967 | long err, mtype; | |
968 | ||
969 | err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg); | |
970 | if (err < 0) | |
971 | goto out; | |
972 | ||
973 | if (put_user(mtype, &msgp->mtype)) | |
974 | err = -EFAULT; | |
975 | out: | |
976 | return err; | |
977 | } | |
978 | ||
1da177e4 | 979 | #ifdef CONFIG_PROC_FS |
19b4946c | 980 | static int sysvipc_msg_proc_show(struct seq_file *s, void *it) |
1da177e4 | 981 | { |
19b4946c MW |
982 | struct msg_queue *msq = it; |
983 | ||
984 | return seq_printf(s, | |
5a06a363 IM |
985 | "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n", |
986 | msq->q_perm.key, | |
7ca7e564 | 987 | msq->q_perm.id, |
5a06a363 IM |
988 | msq->q_perm.mode, |
989 | msq->q_cbytes, | |
990 | msq->q_qnum, | |
991 | msq->q_lspid, | |
992 | msq->q_lrpid, | |
993 | msq->q_perm.uid, | |
994 | msq->q_perm.gid, | |
995 | msq->q_perm.cuid, | |
996 | msq->q_perm.cgid, | |
997 | msq->q_stime, | |
998 | msq->q_rtime, | |
999 | msq->q_ctime); | |
1da177e4 LT |
1000 | } |
1001 | #endif |