]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/ipc/util.c | |
3 | * Copyright (C) 1992 Krishna Balasubramanian | |
4 | * | |
5 | * Sep 1997 - Call suser() last after "normal" permission checks so we | |
6 | * get BSD style process accounting right. | |
7 | * Occurs in several places in the IPC code. | |
8 | * Chris Evans, <[email protected]> | |
9 | * Nov 1999 - ipc helper functions, unified SMP locking | |
624dffcb | 10 | * Manfred Spraul <[email protected]> |
1da177e4 LT |
11 | * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary(). |
12 | * Mingming Cao <[email protected]> | |
073115d6 SG |
13 | * Mar 2006 - support for audit of ipc object properties |
14 | * Dustin Kirkland <[email protected]> | |
73ea4130 KK |
15 | * Jun 2006 - namespaces ssupport |
16 | * OpenVZ, SWsoft Inc. | |
17 | * Pavel Emelianov <[email protected]> | |
1da177e4 LT |
18 | */ |
19 | ||
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/shm.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/msg.h> | |
1da177e4 LT |
24 | #include <linux/vmalloc.h> |
25 | #include <linux/slab.h> | |
c59ede7b | 26 | #include <linux/capability.h> |
1da177e4 LT |
27 | #include <linux/highuid.h> |
28 | #include <linux/security.h> | |
29 | #include <linux/rcupdate.h> | |
30 | #include <linux/workqueue.h> | |
ae781774 MW |
31 | #include <linux/seq_file.h> |
32 | #include <linux/proc_fs.h> | |
073115d6 | 33 | #include <linux/audit.h> |
73ea4130 | 34 | #include <linux/nsproxy.h> |
3e148c79 | 35 | #include <linux/rwsem.h> |
b6b337ad | 36 | #include <linux/memory.h> |
ae5e1b22 | 37 | #include <linux/ipc_namespace.h> |
1da177e4 LT |
38 | |
39 | #include <asm/unistd.h> | |
40 | ||
41 | #include "util.h" | |
42 | ||
ae781774 MW |
43 | struct ipc_proc_iface { |
44 | const char *path; | |
45 | const char *header; | |
73ea4130 | 46 | int ids; |
ae781774 MW |
47 | int (*show)(struct seq_file *, void *); |
48 | }; | |
49 | ||
b6b337ad ND |
50 | #ifdef CONFIG_MEMORY_HOTPLUG |
51 | ||
424450c1 ND |
52 | static void ipc_memory_notifier(struct work_struct *work) |
53 | { | |
54 | ipcns_notify(IPCNS_MEMCHANGED); | |
55 | } | |
56 | ||
57 | static DECLARE_WORK(ipc_memory_wq, ipc_memory_notifier); | |
58 | ||
59 | ||
b6b337ad ND |
60 | static int ipc_memory_callback(struct notifier_block *self, |
61 | unsigned long action, void *arg) | |
62 | { | |
63 | switch (action) { | |
64 | case MEM_ONLINE: /* memory successfully brought online */ | |
65 | case MEM_OFFLINE: /* or offline: it's time to recompute msgmni */ | |
66 | /* | |
67 | * This is done by invoking the ipcns notifier chain with the | |
68 | * IPC_MEMCHANGED event. | |
424450c1 ND |
69 | * In order not to keep the lock on the hotplug memory chain |
70 | * for too long, queue a work item that will, when waken up, | |
71 | * activate the ipcns notification chain. | |
72 | * No need to keep several ipc work items on the queue. | |
b6b337ad | 73 | */ |
424450c1 ND |
74 | if (!work_pending(&ipc_memory_wq)) |
75 | schedule_work(&ipc_memory_wq); | |
b6b337ad ND |
76 | break; |
77 | case MEM_GOING_ONLINE: | |
78 | case MEM_GOING_OFFLINE: | |
79 | case MEM_CANCEL_ONLINE: | |
80 | case MEM_CANCEL_OFFLINE: | |
81 | default: | |
82 | break; | |
83 | } | |
84 | ||
85 | return NOTIFY_OK; | |
86 | } | |
87 | ||
88 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
89 | ||
1da177e4 LT |
90 | /** |
91 | * ipc_init - initialise IPC subsystem | |
92 | * | |
93 | * The various system5 IPC resources (semaphores, messages and shared | |
72fd4a35 | 94 | * memory) are initialised |
b6b337ad ND |
95 | * A callback routine is registered into the memory hotplug notifier |
96 | * chain: since msgmni scales to lowmem this callback routine will be | |
97 | * called upon successful memory add / remove to recompute msmgni. | |
1da177e4 LT |
98 | */ |
99 | ||
100 | static int __init ipc_init(void) | |
101 | { | |
102 | sem_init(); | |
103 | msg_init(); | |
104 | shm_init(); | |
b6b337ad ND |
105 | hotplug_memory_notifier(ipc_memory_callback, IPC_CALLBACK_PRI); |
106 | register_ipcns_notifier(&init_ipc_ns); | |
1da177e4 LT |
107 | return 0; |
108 | } | |
109 | __initcall(ipc_init); | |
110 | ||
111 | /** | |
112 | * ipc_init_ids - initialise IPC identifiers | |
113 | * @ids: Identifier set | |
1da177e4 | 114 | * |
7ca7e564 ND |
115 | * Set up the sequence range to use for the ipc identifier range (limited |
116 | * below IPCMNI) then initialise the ids idr. | |
1da177e4 LT |
117 | */ |
118 | ||
7ca7e564 | 119 | void ipc_init_ids(struct ipc_ids *ids) |
1da177e4 | 120 | { |
3e148c79 | 121 | init_rwsem(&ids->rw_mutex); |
1da177e4 | 122 | |
1da177e4 | 123 | ids->in_use = 0; |
1da177e4 | 124 | ids->seq = 0; |
03f59566 | 125 | ids->next_id = -1; |
1da177e4 LT |
126 | { |
127 | int seq_limit = INT_MAX/SEQ_MULTIPLIER; | |
4be929be AD |
128 | if (seq_limit > USHRT_MAX) |
129 | ids->seq_max = USHRT_MAX; | |
1da177e4 LT |
130 | else |
131 | ids->seq_max = seq_limit; | |
132 | } | |
133 | ||
7ca7e564 | 134 | idr_init(&ids->ipcs_idr); |
1da177e4 LT |
135 | } |
136 | ||
ae781774 | 137 | #ifdef CONFIG_PROC_FS |
9a32144e | 138 | static const struct file_operations sysvipc_proc_fops; |
ae781774 | 139 | /** |
72fd4a35 | 140 | * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface. |
ae781774 MW |
141 | * @path: Path in procfs |
142 | * @header: Banner to be printed at the beginning of the file. | |
143 | * @ids: ipc id table to iterate. | |
144 | * @show: show routine. | |
145 | */ | |
146 | void __init ipc_init_proc_interface(const char *path, const char *header, | |
73ea4130 | 147 | int ids, int (*show)(struct seq_file *, void *)) |
ae781774 MW |
148 | { |
149 | struct proc_dir_entry *pde; | |
150 | struct ipc_proc_iface *iface; | |
151 | ||
152 | iface = kmalloc(sizeof(*iface), GFP_KERNEL); | |
153 | if (!iface) | |
154 | return; | |
155 | iface->path = path; | |
156 | iface->header = header; | |
157 | iface->ids = ids; | |
158 | iface->show = show; | |
159 | ||
6a6375db DL |
160 | pde = proc_create_data(path, |
161 | S_IRUGO, /* world readable */ | |
162 | NULL, /* parent dir */ | |
163 | &sysvipc_proc_fops, | |
164 | iface); | |
165 | if (!pde) { | |
ae781774 MW |
166 | kfree(iface); |
167 | } | |
168 | } | |
169 | #endif | |
170 | ||
1da177e4 LT |
171 | /** |
172 | * ipc_findkey - find a key in an ipc identifier set | |
173 | * @ids: Identifier set | |
174 | * @key: The key to find | |
175 | * | |
3e148c79 | 176 | * Requires ipc_ids.rw_mutex locked. |
7ca7e564 ND |
177 | * Returns the LOCKED pointer to the ipc structure if found or NULL |
178 | * if not. | |
f4566f04 | 179 | * If key is found ipc points to the owning ipc structure |
1da177e4 LT |
180 | */ |
181 | ||
7748dbfa | 182 | static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key) |
1da177e4 | 183 | { |
7ca7e564 ND |
184 | struct kern_ipc_perm *ipc; |
185 | int next_id; | |
186 | int total; | |
1da177e4 | 187 | |
7ca7e564 ND |
188 | for (total = 0, next_id = 0; total < ids->in_use; next_id++) { |
189 | ipc = idr_find(&ids->ipcs_idr, next_id); | |
190 | ||
191 | if (ipc == NULL) | |
1da177e4 | 192 | continue; |
7ca7e564 ND |
193 | |
194 | if (ipc->key != key) { | |
195 | total++; | |
196 | continue; | |
197 | } | |
198 | ||
199 | ipc_lock_by_ptr(ipc); | |
200 | return ipc; | |
1da177e4 | 201 | } |
7ca7e564 ND |
202 | |
203 | return NULL; | |
1da177e4 LT |
204 | } |
205 | ||
7ca7e564 ND |
206 | /** |
207 | * ipc_get_maxid - get the last assigned id | |
208 | * @ids: IPC identifier set | |
209 | * | |
3e148c79 | 210 | * Called with ipc_ids.rw_mutex held. |
1da177e4 | 211 | */ |
1da177e4 | 212 | |
7ca7e564 ND |
213 | int ipc_get_maxid(struct ipc_ids *ids) |
214 | { | |
215 | struct kern_ipc_perm *ipc; | |
216 | int max_id = -1; | |
217 | int total, id; | |
218 | ||
219 | if (ids->in_use == 0) | |
220 | return -1; | |
1da177e4 | 221 | |
7ca7e564 ND |
222 | if (ids->in_use == IPCMNI) |
223 | return IPCMNI - 1; | |
224 | ||
225 | /* Look for the last assigned id */ | |
226 | total = 0; | |
227 | for (id = 0; id < IPCMNI && total < ids->in_use; id++) { | |
228 | ipc = idr_find(&ids->ipcs_idr, id); | |
229 | if (ipc != NULL) { | |
230 | max_id = id; | |
231 | total++; | |
232 | } | |
233 | } | |
234 | return max_id; | |
1da177e4 LT |
235 | } |
236 | ||
237 | /** | |
238 | * ipc_addid - add an IPC identifier | |
239 | * @ids: IPC identifier set | |
240 | * @new: new IPC permission set | |
7ca7e564 | 241 | * @size: limit for the number of used ids |
1da177e4 | 242 | * |
f4566f04 | 243 | * Add an entry 'new' to the IPC ids idr. The permissions object is |
1da177e4 | 244 | * initialised and the first free entry is set up and the id assigned |
f4566f04 | 245 | * is returned. The 'new' entry is returned in a locked state on success. |
283bb7fa | 246 | * On failure the entry is not locked and a negative err-code is returned. |
1da177e4 | 247 | * |
3e148c79 | 248 | * Called with ipc_ids.rw_mutex held as a writer. |
1da177e4 LT |
249 | */ |
250 | ||
251 | int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size) | |
252 | { | |
1efdb69b EB |
253 | kuid_t euid; |
254 | kgid_t egid; | |
7ca7e564 | 255 | int id, err; |
03f59566 | 256 | int next_id = ids->next_id; |
1da177e4 | 257 | |
7ca7e564 ND |
258 | if (size > IPCMNI) |
259 | size = IPCMNI; | |
260 | ||
261 | if (ids->in_use >= size) | |
283bb7fa | 262 | return -ENOSPC; |
7ca7e564 | 263 | |
e00b4ff7 ND |
264 | spin_lock_init(&new->lock); |
265 | new->deleted = 0; | |
266 | rcu_read_lock(); | |
267 | spin_lock(&new->lock); | |
268 | ||
03f59566 SK |
269 | err = idr_get_new_above(&ids->ipcs_idr, new, |
270 | (next_id < 0) ? 0 : ipcid_to_idx(next_id), &id); | |
e00b4ff7 ND |
271 | if (err) { |
272 | spin_unlock(&new->lock); | |
273 | rcu_read_unlock(); | |
283bb7fa | 274 | return err; |
e00b4ff7 | 275 | } |
7ca7e564 | 276 | |
1da177e4 | 277 | ids->in_use++; |
1da177e4 | 278 | |
414c0708 DH |
279 | current_euid_egid(&euid, &egid); |
280 | new->cuid = new->uid = euid; | |
281 | new->gid = new->cgid = egid; | |
1da177e4 | 282 | |
03f59566 SK |
283 | if (next_id < 0) { |
284 | new->seq = ids->seq++; | |
285 | if (ids->seq > ids->seq_max) | |
286 | ids->seq = 0; | |
287 | } else { | |
288 | new->seq = ipcid_to_seqx(next_id); | |
289 | ids->next_id = -1; | |
290 | } | |
1da177e4 | 291 | |
48dea404 | 292 | new->id = ipc_buildid(id, new->seq); |
1da177e4 LT |
293 | return id; |
294 | } | |
295 | ||
7748dbfa ND |
296 | /** |
297 | * ipcget_new - create a new ipc object | |
298 | * @ns: namespace | |
f4566f04 | 299 | * @ids: IPC identifer set |
7748dbfa ND |
300 | * @ops: the actual creation routine to call |
301 | * @params: its parameters | |
302 | * | |
f4566f04 ND |
303 | * This routine is called by sys_msgget, sys_semget() and sys_shmget() |
304 | * when the key is IPC_PRIVATE. | |
7748dbfa | 305 | */ |
b2d75cdd | 306 | static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids, |
7748dbfa ND |
307 | struct ipc_ops *ops, struct ipc_params *params) |
308 | { | |
309 | int err; | |
283bb7fa | 310 | retry: |
7748dbfa ND |
311 | err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL); |
312 | ||
313 | if (!err) | |
314 | return -ENOMEM; | |
315 | ||
3e148c79 | 316 | down_write(&ids->rw_mutex); |
7748dbfa | 317 | err = ops->getnew(ns, params); |
3e148c79 | 318 | up_write(&ids->rw_mutex); |
7748dbfa | 319 | |
283bb7fa PP |
320 | if (err == -EAGAIN) |
321 | goto retry; | |
322 | ||
7748dbfa ND |
323 | return err; |
324 | } | |
325 | ||
326 | /** | |
327 | * ipc_check_perms - check security and permissions for an IPC | |
6213cfe8 | 328 | * @ns: IPC namespace |
7748dbfa | 329 | * @ipcp: ipc permission set |
7748dbfa ND |
330 | * @ops: the actual security routine to call |
331 | * @params: its parameters | |
f4566f04 ND |
332 | * |
333 | * This routine is called by sys_msgget(), sys_semget() and sys_shmget() | |
334 | * when the key is not IPC_PRIVATE and that key already exists in the | |
335 | * ids IDR. | |
336 | * | |
337 | * On success, the IPC id is returned. | |
338 | * | |
3e148c79 | 339 | * It is called with ipc_ids.rw_mutex and ipcp->lock held. |
7748dbfa | 340 | */ |
b0e77598 SH |
341 | static int ipc_check_perms(struct ipc_namespace *ns, |
342 | struct kern_ipc_perm *ipcp, | |
343 | struct ipc_ops *ops, | |
344 | struct ipc_params *params) | |
7748dbfa ND |
345 | { |
346 | int err; | |
347 | ||
b0e77598 | 348 | if (ipcperms(ns, ipcp, params->flg)) |
7748dbfa ND |
349 | err = -EACCES; |
350 | else { | |
351 | err = ops->associate(ipcp, params->flg); | |
352 | if (!err) | |
353 | err = ipcp->id; | |
354 | } | |
355 | ||
356 | return err; | |
357 | } | |
358 | ||
359 | /** | |
360 | * ipcget_public - get an ipc object or create a new one | |
361 | * @ns: namespace | |
f4566f04 | 362 | * @ids: IPC identifer set |
7748dbfa ND |
363 | * @ops: the actual creation routine to call |
364 | * @params: its parameters | |
365 | * | |
f4566f04 ND |
366 | * This routine is called by sys_msgget, sys_semget() and sys_shmget() |
367 | * when the key is not IPC_PRIVATE. | |
368 | * It adds a new entry if the key is not found and does some permission | |
369 | * / security checkings if the key is found. | |
370 | * | |
371 | * On success, the ipc id is returned. | |
7748dbfa | 372 | */ |
b2d75cdd | 373 | static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids, |
7748dbfa ND |
374 | struct ipc_ops *ops, struct ipc_params *params) |
375 | { | |
376 | struct kern_ipc_perm *ipcp; | |
377 | int flg = params->flg; | |
378 | int err; | |
283bb7fa | 379 | retry: |
7748dbfa ND |
380 | err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL); |
381 | ||
3e148c79 ND |
382 | /* |
383 | * Take the lock as a writer since we are potentially going to add | |
384 | * a new entry + read locks are not "upgradable" | |
385 | */ | |
386 | down_write(&ids->rw_mutex); | |
7748dbfa ND |
387 | ipcp = ipc_findkey(ids, params->key); |
388 | if (ipcp == NULL) { | |
389 | /* key not used */ | |
390 | if (!(flg & IPC_CREAT)) | |
391 | err = -ENOENT; | |
392 | else if (!err) | |
393 | err = -ENOMEM; | |
394 | else | |
395 | err = ops->getnew(ns, params); | |
396 | } else { | |
397 | /* ipc object has been locked by ipc_findkey() */ | |
398 | ||
399 | if (flg & IPC_CREAT && flg & IPC_EXCL) | |
400 | err = -EEXIST; | |
401 | else { | |
402 | err = 0; | |
403 | if (ops->more_checks) | |
404 | err = ops->more_checks(ipcp, params); | |
405 | if (!err) | |
f4566f04 ND |
406 | /* |
407 | * ipc_check_perms returns the IPC id on | |
408 | * success | |
409 | */ | |
b0e77598 | 410 | err = ipc_check_perms(ns, ipcp, ops, params); |
7748dbfa ND |
411 | } |
412 | ipc_unlock(ipcp); | |
413 | } | |
3e148c79 | 414 | up_write(&ids->rw_mutex); |
7748dbfa | 415 | |
283bb7fa PP |
416 | if (err == -EAGAIN) |
417 | goto retry; | |
418 | ||
7748dbfa ND |
419 | return err; |
420 | } | |
421 | ||
422 | ||
1da177e4 LT |
423 | /** |
424 | * ipc_rmid - remove an IPC identifier | |
f4566f04 ND |
425 | * @ids: IPC identifier set |
426 | * @ipcp: ipc perm structure containing the identifier to remove | |
1da177e4 | 427 | * |
3e148c79 ND |
428 | * ipc_ids.rw_mutex (as a writer) and the spinlock for this ID are held |
429 | * before this function is called, and remain locked on the exit. | |
1da177e4 LT |
430 | */ |
431 | ||
7ca7e564 | 432 | void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp) |
1da177e4 | 433 | { |
ce621f5b | 434 | int lid = ipcid_to_idx(ipcp->id); |
7ca7e564 ND |
435 | |
436 | idr_remove(&ids->ipcs_idr, lid); | |
1da177e4 | 437 | |
1da177e4 LT |
438 | ids->in_use--; |
439 | ||
7ca7e564 ND |
440 | ipcp->deleted = 1; |
441 | ||
442 | return; | |
1da177e4 LT |
443 | } |
444 | ||
445 | /** | |
446 | * ipc_alloc - allocate ipc space | |
447 | * @size: size desired | |
448 | * | |
449 | * Allocate memory from the appropriate pools and return a pointer to it. | |
450 | * NULL is returned if the allocation fails | |
451 | */ | |
452 | ||
453 | void* ipc_alloc(int size) | |
454 | { | |
455 | void* out; | |
456 | if(size > PAGE_SIZE) | |
457 | out = vmalloc(size); | |
458 | else | |
459 | out = kmalloc(size, GFP_KERNEL); | |
460 | return out; | |
461 | } | |
462 | ||
463 | /** | |
464 | * ipc_free - free ipc space | |
465 | * @ptr: pointer returned by ipc_alloc | |
466 | * @size: size of block | |
467 | * | |
72fd4a35 | 468 | * Free a block created with ipc_alloc(). The caller must know the size |
1da177e4 LT |
469 | * used in the allocation call. |
470 | */ | |
471 | ||
472 | void ipc_free(void* ptr, int size) | |
473 | { | |
474 | if(size > PAGE_SIZE) | |
475 | vfree(ptr); | |
476 | else | |
477 | kfree(ptr); | |
478 | } | |
479 | ||
480 | /* | |
481 | * rcu allocations: | |
482 | * There are three headers that are prepended to the actual allocation: | |
483 | * - during use: ipc_rcu_hdr. | |
484 | * - during the rcu grace period: ipc_rcu_grace. | |
485 | * - [only if vmalloc]: ipc_rcu_sched. | |
486 | * Their lifetime doesn't overlap, thus the headers share the same memory. | |
487 | * Unlike a normal union, they are right-aligned, thus some container_of | |
488 | * forward/backward casting is necessary: | |
489 | */ | |
490 | struct ipc_rcu_hdr | |
491 | { | |
492 | int refcount; | |
493 | int is_vmalloc; | |
494 | void *data[0]; | |
495 | }; | |
496 | ||
497 | ||
498 | struct ipc_rcu_grace | |
499 | { | |
500 | struct rcu_head rcu; | |
501 | /* "void *" makes sure alignment of following data is sane. */ | |
502 | void *data[0]; | |
503 | }; | |
504 | ||
505 | struct ipc_rcu_sched | |
506 | { | |
507 | struct work_struct work; | |
508 | /* "void *" makes sure alignment of following data is sane. */ | |
509 | void *data[0]; | |
510 | }; | |
511 | ||
512 | #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \ | |
513 | sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr)) | |
514 | #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \ | |
515 | sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC) | |
516 | ||
517 | static inline int rcu_use_vmalloc(int size) | |
518 | { | |
519 | /* Too big for a single page? */ | |
520 | if (HDRLEN_KMALLOC + size > PAGE_SIZE) | |
521 | return 1; | |
522 | return 0; | |
523 | } | |
524 | ||
525 | /** | |
526 | * ipc_rcu_alloc - allocate ipc and rcu space | |
527 | * @size: size desired | |
528 | * | |
529 | * Allocate memory for the rcu header structure + the object. | |
530 | * Returns the pointer to the object. | |
531 | * NULL is returned if the allocation fails. | |
532 | */ | |
533 | ||
534 | void* ipc_rcu_alloc(int size) | |
535 | { | |
536 | void* out; | |
537 | /* | |
538 | * We prepend the allocation with the rcu struct, and | |
539 | * workqueue if necessary (for vmalloc). | |
540 | */ | |
541 | if (rcu_use_vmalloc(size)) { | |
542 | out = vmalloc(HDRLEN_VMALLOC + size); | |
543 | if (out) { | |
544 | out += HDRLEN_VMALLOC; | |
545 | container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1; | |
546 | container_of(out, struct ipc_rcu_hdr, data)->refcount = 1; | |
547 | } | |
548 | } else { | |
549 | out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL); | |
550 | if (out) { | |
551 | out += HDRLEN_KMALLOC; | |
552 | container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0; | |
553 | container_of(out, struct ipc_rcu_hdr, data)->refcount = 1; | |
554 | } | |
555 | } | |
556 | ||
557 | return out; | |
558 | } | |
559 | ||
560 | void ipc_rcu_getref(void *ptr) | |
561 | { | |
562 | container_of(ptr, struct ipc_rcu_hdr, data)->refcount++; | |
563 | } | |
564 | ||
65f27f38 DH |
565 | static void ipc_do_vfree(struct work_struct *work) |
566 | { | |
567 | vfree(container_of(work, struct ipc_rcu_sched, work)); | |
568 | } | |
569 | ||
1da177e4 | 570 | /** |
1e5d5331 RD |
571 | * ipc_schedule_free - free ipc + rcu space |
572 | * @head: RCU callback structure for queued work | |
1da177e4 LT |
573 | * |
574 | * Since RCU callback function is called in bh, | |
72fd4a35 | 575 | * we need to defer the vfree to schedule_work(). |
1da177e4 LT |
576 | */ |
577 | static void ipc_schedule_free(struct rcu_head *head) | |
578 | { | |
f4566f04 ND |
579 | struct ipc_rcu_grace *grace; |
580 | struct ipc_rcu_sched *sched; | |
581 | ||
582 | grace = container_of(head, struct ipc_rcu_grace, rcu); | |
583 | sched = container_of(&(grace->data[0]), struct ipc_rcu_sched, | |
584 | data[0]); | |
1da177e4 | 585 | |
65f27f38 | 586 | INIT_WORK(&sched->work, ipc_do_vfree); |
1da177e4 LT |
587 | schedule_work(&sched->work); |
588 | } | |
589 | ||
1da177e4 LT |
590 | void ipc_rcu_putref(void *ptr) |
591 | { | |
592 | if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0) | |
593 | return; | |
594 | ||
595 | if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) { | |
596 | call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu, | |
597 | ipc_schedule_free); | |
598 | } else { | |
d4ee9aa3 | 599 | kfree_rcu(container_of(ptr, struct ipc_rcu_grace, data), rcu); |
1da177e4 LT |
600 | } |
601 | } | |
602 | ||
603 | /** | |
604 | * ipcperms - check IPC permissions | |
6213cfe8 | 605 | * @ns: IPC namespace |
1da177e4 LT |
606 | * @ipcp: IPC permission set |
607 | * @flag: desired permission set. | |
608 | * | |
609 | * Check user, group, other permissions for access | |
610 | * to ipc resources. return 0 if allowed | |
b0e77598 SH |
611 | * |
612 | * @flag will most probably be 0 or S_...UGO from <linux/stat.h> | |
1da177e4 LT |
613 | */ |
614 | ||
b0e77598 SH |
615 | int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag) |
616 | { | |
1efdb69b | 617 | kuid_t euid = current_euid(); |
a33e6751 | 618 | int requested_mode, granted_mode; |
1da177e4 | 619 | |
a33e6751 | 620 | audit_ipc_obj(ipcp); |
1da177e4 LT |
621 | requested_mode = (flag >> 6) | (flag >> 3) | flag; |
622 | granted_mode = ipcp->mode; | |
1efdb69b EB |
623 | if (uid_eq(euid, ipcp->cuid) || |
624 | uid_eq(euid, ipcp->uid)) | |
1da177e4 LT |
625 | granted_mode >>= 6; |
626 | else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid)) | |
627 | granted_mode >>= 3; | |
628 | /* is there some bit set in requested_mode but not in granted_mode? */ | |
629 | if ((requested_mode & ~granted_mode & 0007) && | |
b0e77598 | 630 | !ns_capable(ns->user_ns, CAP_IPC_OWNER)) |
1da177e4 LT |
631 | return -1; |
632 | ||
633 | return security_ipc_permission(ipcp, flag); | |
634 | } | |
635 | ||
636 | /* | |
637 | * Functions to convert between the kern_ipc_perm structure and the | |
638 | * old/new ipc_perm structures | |
639 | */ | |
640 | ||
641 | /** | |
642 | * kernel_to_ipc64_perm - convert kernel ipc permissions to user | |
643 | * @in: kernel permissions | |
644 | * @out: new style IPC permissions | |
645 | * | |
72fd4a35 RD |
646 | * Turn the kernel object @in into a set of permissions descriptions |
647 | * for returning to userspace (@out). | |
1da177e4 LT |
648 | */ |
649 | ||
650 | ||
651 | void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out) | |
652 | { | |
653 | out->key = in->key; | |
1efdb69b EB |
654 | out->uid = from_kuid_munged(current_user_ns(), in->uid); |
655 | out->gid = from_kgid_munged(current_user_ns(), in->gid); | |
656 | out->cuid = from_kuid_munged(current_user_ns(), in->cuid); | |
657 | out->cgid = from_kgid_munged(current_user_ns(), in->cgid); | |
1da177e4 LT |
658 | out->mode = in->mode; |
659 | out->seq = in->seq; | |
660 | } | |
661 | ||
662 | /** | |
f4566f04 | 663 | * ipc64_perm_to_ipc_perm - convert new ipc permissions to old |
1da177e4 LT |
664 | * @in: new style IPC permissions |
665 | * @out: old style IPC permissions | |
666 | * | |
72fd4a35 RD |
667 | * Turn the new style permissions object @in into a compatibility |
668 | * object and store it into the @out pointer. | |
1da177e4 LT |
669 | */ |
670 | ||
671 | void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out) | |
672 | { | |
673 | out->key = in->key; | |
674 | SET_UID(out->uid, in->uid); | |
675 | SET_GID(out->gid, in->gid); | |
676 | SET_UID(out->cuid, in->cuid); | |
677 | SET_GID(out->cgid, in->cgid); | |
678 | out->mode = in->mode; | |
679 | out->seq = in->seq; | |
680 | } | |
681 | ||
f4566f04 | 682 | /** |
3e148c79 | 683 | * ipc_lock - Lock an ipc structure without rw_mutex held |
f4566f04 ND |
684 | * @ids: IPC identifier set |
685 | * @id: ipc id to look for | |
686 | * | |
687 | * Look for an id in the ipc ids idr and lock the associated ipc object. | |
688 | * | |
f4566f04 ND |
689 | * The ipc object is locked on exit. |
690 | */ | |
691 | ||
7ca7e564 | 692 | struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id) |
1da177e4 | 693 | { |
7ca7e564 | 694 | struct kern_ipc_perm *out; |
ce621f5b | 695 | int lid = ipcid_to_idx(id); |
1da177e4 LT |
696 | |
697 | rcu_read_lock(); | |
7ca7e564 ND |
698 | out = idr_find(&ids->ipcs_idr, lid); |
699 | if (out == NULL) { | |
1da177e4 | 700 | rcu_read_unlock(); |
023a5355 | 701 | return ERR_PTR(-EINVAL); |
1da177e4 | 702 | } |
7ca7e564 | 703 | |
1da177e4 LT |
704 | spin_lock(&out->lock); |
705 | ||
706 | /* ipc_rmid() may have already freed the ID while ipc_lock | |
707 | * was spinning: here verify that the structure is still valid | |
708 | */ | |
709 | if (out->deleted) { | |
710 | spin_unlock(&out->lock); | |
711 | rcu_read_unlock(); | |
023a5355 | 712 | return ERR_PTR(-EINVAL); |
1da177e4 | 713 | } |
7ca7e564 | 714 | |
1da177e4 LT |
715 | return out; |
716 | } | |
717 | ||
b2d75cdd PE |
718 | struct kern_ipc_perm *ipc_lock_check(struct ipc_ids *ids, int id) |
719 | { | |
720 | struct kern_ipc_perm *out; | |
721 | ||
722 | out = ipc_lock(ids, id); | |
723 | if (IS_ERR(out)) | |
724 | return out; | |
725 | ||
726 | if (ipc_checkid(out, id)) { | |
727 | ipc_unlock(out); | |
728 | return ERR_PTR(-EIDRM); | |
729 | } | |
730 | ||
731 | return out; | |
732 | } | |
733 | ||
734 | /** | |
735 | * ipcget - Common sys_*get() code | |
736 | * @ns : namsepace | |
737 | * @ids : IPC identifier set | |
738 | * @ops : operations to be called on ipc object creation, permission checks | |
739 | * and further checks | |
740 | * @params : the parameters needed by the previous operations. | |
741 | * | |
742 | * Common routine called by sys_msgget(), sys_semget() and sys_shmget(). | |
743 | */ | |
744 | int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids, | |
745 | struct ipc_ops *ops, struct ipc_params *params) | |
746 | { | |
747 | if (params->key == IPC_PRIVATE) | |
748 | return ipcget_new(ns, ids, ops, params); | |
749 | else | |
750 | return ipcget_public(ns, ids, ops, params); | |
751 | } | |
752 | ||
8f4a3809 PP |
753 | /** |
754 | * ipc_update_perm - update the permissions of an IPC. | |
755 | * @in: the permission given as input. | |
756 | * @out: the permission of the ipc to set. | |
757 | */ | |
1efdb69b | 758 | int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out) |
8f4a3809 | 759 | { |
1efdb69b EB |
760 | kuid_t uid = make_kuid(current_user_ns(), in->uid); |
761 | kgid_t gid = make_kgid(current_user_ns(), in->gid); | |
762 | if (!uid_valid(uid) || !gid_valid(gid)) | |
763 | return -EINVAL; | |
764 | ||
765 | out->uid = uid; | |
766 | out->gid = gid; | |
8f4a3809 PP |
767 | out->mode = (out->mode & ~S_IRWXUGO) |
768 | | (in->mode & S_IRWXUGO); | |
1efdb69b EB |
769 | |
770 | return 0; | |
8f4a3809 PP |
771 | } |
772 | ||
a5f75e7f PP |
773 | /** |
774 | * ipcctl_pre_down - retrieve an ipc and check permissions for some IPC_XXX cmd | |
6213cfe8 | 775 | * @ns: the ipc namespace |
a5f75e7f PP |
776 | * @ids: the table of ids where to look for the ipc |
777 | * @id: the id of the ipc to retrieve | |
778 | * @cmd: the cmd to check | |
779 | * @perm: the permission to set | |
780 | * @extra_perm: one extra permission parameter used by msq | |
781 | * | |
782 | * This function does some common audit and permissions check for some IPC_XXX | |
783 | * cmd and is called from semctl_down, shmctl_down and msgctl_down. | |
784 | * It must be called without any lock held and | |
785 | * - retrieves the ipc with the given id in the given table. | |
786 | * - performs some audit and permission check, depending on the given cmd | |
787 | * - returns the ipc with both ipc and rw_mutex locks held in case of success | |
788 | * or an err-code without any lock held otherwise. | |
789 | */ | |
b0e77598 SH |
790 | struct kern_ipc_perm *ipcctl_pre_down(struct ipc_namespace *ns, |
791 | struct ipc_ids *ids, int id, int cmd, | |
a5f75e7f PP |
792 | struct ipc64_perm *perm, int extra_perm) |
793 | { | |
794 | struct kern_ipc_perm *ipcp; | |
1efdb69b | 795 | kuid_t euid; |
a5f75e7f PP |
796 | int err; |
797 | ||
798 | down_write(&ids->rw_mutex); | |
00c2bf85 | 799 | ipcp = ipc_lock_check(ids, id); |
a5f75e7f PP |
800 | if (IS_ERR(ipcp)) { |
801 | err = PTR_ERR(ipcp); | |
802 | goto out_up; | |
803 | } | |
804 | ||
a33e6751 | 805 | audit_ipc_obj(ipcp); |
e816f370 AV |
806 | if (cmd == IPC_SET) |
807 | audit_ipc_set_perm(extra_perm, perm->uid, | |
a5f75e7f | 808 | perm->gid, perm->mode); |
414c0708 DH |
809 | |
810 | euid = current_euid(); | |
1efdb69b | 811 | if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) || |
b0e77598 | 812 | ns_capable(ns->user_ns, CAP_SYS_ADMIN)) |
a5f75e7f PP |
813 | return ipcp; |
814 | ||
815 | err = -EPERM; | |
a5f75e7f PP |
816 | ipc_unlock(ipcp); |
817 | out_up: | |
818 | up_write(&ids->rw_mutex); | |
819 | return ERR_PTR(err); | |
820 | } | |
821 | ||
c1d7e01d | 822 | #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION |
1da177e4 LT |
823 | |
824 | ||
825 | /** | |
826 | * ipc_parse_version - IPC call version | |
827 | * @cmd: pointer to command | |
828 | * | |
829 | * Return IPC_64 for new style IPC and IPC_OLD for old style IPC. | |
72fd4a35 | 830 | * The @cmd value is turned from an encoding command and version into |
1da177e4 LT |
831 | * just the command code. |
832 | */ | |
833 | ||
834 | int ipc_parse_version (int *cmd) | |
835 | { | |
836 | if (*cmd & IPC_64) { | |
837 | *cmd ^= IPC_64; | |
838 | return IPC_64; | |
839 | } else { | |
840 | return IPC_OLD; | |
841 | } | |
842 | } | |
843 | ||
c1d7e01d | 844 | #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */ |
ae781774 MW |
845 | |
846 | #ifdef CONFIG_PROC_FS | |
bc1fc6d8 EB |
847 | struct ipc_proc_iter { |
848 | struct ipc_namespace *ns; | |
849 | struct ipc_proc_iface *iface; | |
850 | }; | |
851 | ||
7ca7e564 ND |
852 | /* |
853 | * This routine locks the ipc structure found at least at position pos. | |
854 | */ | |
b524b9ad AB |
855 | static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos, |
856 | loff_t *new_pos) | |
ae781774 | 857 | { |
7ca7e564 ND |
858 | struct kern_ipc_perm *ipc; |
859 | int total, id; | |
73ea4130 | 860 | |
7ca7e564 ND |
861 | total = 0; |
862 | for (id = 0; id < pos && total < ids->in_use; id++) { | |
863 | ipc = idr_find(&ids->ipcs_idr, id); | |
864 | if (ipc != NULL) | |
865 | total++; | |
866 | } | |
ae781774 | 867 | |
7ca7e564 ND |
868 | if (total >= ids->in_use) |
869 | return NULL; | |
ae781774 | 870 | |
7ca7e564 ND |
871 | for ( ; pos < IPCMNI; pos++) { |
872 | ipc = idr_find(&ids->ipcs_idr, pos); | |
873 | if (ipc != NULL) { | |
874 | *new_pos = pos + 1; | |
875 | ipc_lock_by_ptr(ipc); | |
ae781774 MW |
876 | return ipc; |
877 | } | |
878 | } | |
879 | ||
880 | /* Out of range - return NULL to terminate iteration */ | |
881 | return NULL; | |
882 | } | |
883 | ||
7ca7e564 ND |
884 | static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos) |
885 | { | |
886 | struct ipc_proc_iter *iter = s->private; | |
887 | struct ipc_proc_iface *iface = iter->iface; | |
888 | struct kern_ipc_perm *ipc = it; | |
889 | ||
890 | /* If we had an ipc id locked before, unlock it */ | |
891 | if (ipc && ipc != SEQ_START_TOKEN) | |
892 | ipc_unlock(ipc); | |
893 | ||
ed2ddbf8 | 894 | return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos); |
7ca7e564 ND |
895 | } |
896 | ||
ae781774 | 897 | /* |
f4566f04 ND |
898 | * File positions: pos 0 -> header, pos n -> ipc id = n - 1. |
899 | * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START. | |
ae781774 MW |
900 | */ |
901 | static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos) | |
902 | { | |
bc1fc6d8 EB |
903 | struct ipc_proc_iter *iter = s->private; |
904 | struct ipc_proc_iface *iface = iter->iface; | |
73ea4130 KK |
905 | struct ipc_ids *ids; |
906 | ||
ed2ddbf8 | 907 | ids = &iter->ns->ids[iface->ids]; |
ae781774 MW |
908 | |
909 | /* | |
910 | * Take the lock - this will be released by the corresponding | |
911 | * call to stop(). | |
912 | */ | |
3e148c79 | 913 | down_read(&ids->rw_mutex); |
ae781774 MW |
914 | |
915 | /* pos < 0 is invalid */ | |
916 | if (*pos < 0) | |
917 | return NULL; | |
918 | ||
919 | /* pos == 0 means header */ | |
920 | if (*pos == 0) | |
921 | return SEQ_START_TOKEN; | |
922 | ||
923 | /* Find the (pos-1)th ipc */ | |
7ca7e564 | 924 | return sysvipc_find_ipc(ids, *pos - 1, pos); |
ae781774 MW |
925 | } |
926 | ||
927 | static void sysvipc_proc_stop(struct seq_file *s, void *it) | |
928 | { | |
929 | struct kern_ipc_perm *ipc = it; | |
bc1fc6d8 EB |
930 | struct ipc_proc_iter *iter = s->private; |
931 | struct ipc_proc_iface *iface = iter->iface; | |
73ea4130 | 932 | struct ipc_ids *ids; |
ae781774 | 933 | |
f4566f04 | 934 | /* If we had a locked structure, release it */ |
ae781774 MW |
935 | if (ipc && ipc != SEQ_START_TOKEN) |
936 | ipc_unlock(ipc); | |
937 | ||
ed2ddbf8 | 938 | ids = &iter->ns->ids[iface->ids]; |
ae781774 | 939 | /* Release the lock we took in start() */ |
3e148c79 | 940 | up_read(&ids->rw_mutex); |
ae781774 MW |
941 | } |
942 | ||
943 | static int sysvipc_proc_show(struct seq_file *s, void *it) | |
944 | { | |
bc1fc6d8 EB |
945 | struct ipc_proc_iter *iter = s->private; |
946 | struct ipc_proc_iface *iface = iter->iface; | |
ae781774 MW |
947 | |
948 | if (it == SEQ_START_TOKEN) | |
949 | return seq_puts(s, iface->header); | |
950 | ||
951 | return iface->show(s, it); | |
952 | } | |
953 | ||
88e9d34c | 954 | static const struct seq_operations sysvipc_proc_seqops = { |
ae781774 MW |
955 | .start = sysvipc_proc_start, |
956 | .stop = sysvipc_proc_stop, | |
957 | .next = sysvipc_proc_next, | |
958 | .show = sysvipc_proc_show, | |
959 | }; | |
960 | ||
bc1fc6d8 EB |
961 | static int sysvipc_proc_open(struct inode *inode, struct file *file) |
962 | { | |
ae781774 MW |
963 | int ret; |
964 | struct seq_file *seq; | |
bc1fc6d8 EB |
965 | struct ipc_proc_iter *iter; |
966 | ||
967 | ret = -ENOMEM; | |
968 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
969 | if (!iter) | |
970 | goto out; | |
ae781774 MW |
971 | |
972 | ret = seq_open(file, &sysvipc_proc_seqops); | |
bc1fc6d8 EB |
973 | if (ret) |
974 | goto out_kfree; | |
975 | ||
976 | seq = file->private_data; | |
977 | seq->private = iter; | |
978 | ||
979 | iter->iface = PDE(inode)->data; | |
980 | iter->ns = get_ipc_ns(current->nsproxy->ipc_ns); | |
981 | out: | |
ae781774 | 982 | return ret; |
bc1fc6d8 EB |
983 | out_kfree: |
984 | kfree(iter); | |
985 | goto out; | |
986 | } | |
987 | ||
988 | static int sysvipc_proc_release(struct inode *inode, struct file *file) | |
989 | { | |
990 | struct seq_file *seq = file->private_data; | |
991 | struct ipc_proc_iter *iter = seq->private; | |
992 | put_ipc_ns(iter->ns); | |
993 | return seq_release_private(inode, file); | |
ae781774 MW |
994 | } |
995 | ||
9a32144e | 996 | static const struct file_operations sysvipc_proc_fops = { |
ae781774 MW |
997 | .open = sysvipc_proc_open, |
998 | .read = seq_read, | |
999 | .llseek = seq_lseek, | |
bc1fc6d8 | 1000 | .release = sysvipc_proc_release, |
ae781774 MW |
1001 | }; |
1002 | #endif /* CONFIG_PROC_FS */ |