]>
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, <chris@ferret.lmh.ox.ac.uk> | |
9 | * Nov 1999 - ipc helper functions, unified SMP locking | |
624dffcb | 10 | * Manfred Spraul <manfred@colorfullife.com> |
1da177e4 LT |
11 | * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary(). |
12 | * Mingming Cao <cmm@us.ibm.com> | |
073115d6 SG |
13 | * Mar 2006 - support for audit of ipc object properties |
14 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
73ea4130 KK |
15 | * Jun 2006 - namespaces ssupport |
16 | * OpenVZ, SWsoft Inc. | |
17 | * Pavel Emelianov <xemul@openvz.org> | |
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 | ||
73ea4130 KK |
50 | struct ipc_namespace init_ipc_ns = { |
51 | .kref = { | |
52 | .refcount = ATOMIC_INIT(2), | |
53 | }, | |
54 | }; | |
55 | ||
4d89dc6a ND |
56 | atomic_t nr_ipc_ns = ATOMIC_INIT(1); |
57 | ||
58 | ||
b6b337ad ND |
59 | #ifdef CONFIG_MEMORY_HOTPLUG |
60 | ||
61 | static int ipc_memory_callback(struct notifier_block *self, | |
62 | unsigned long action, void *arg) | |
63 | { | |
64 | switch (action) { | |
65 | case MEM_ONLINE: /* memory successfully brought online */ | |
66 | case MEM_OFFLINE: /* or offline: it's time to recompute msgmni */ | |
67 | /* | |
68 | * This is done by invoking the ipcns notifier chain with the | |
69 | * IPC_MEMCHANGED event. | |
70 | */ | |
71 | ipcns_notify(IPCNS_MEMCHANGED); | |
72 | break; | |
73 | case MEM_GOING_ONLINE: | |
74 | case MEM_GOING_OFFLINE: | |
75 | case MEM_CANCEL_ONLINE: | |
76 | case MEM_CANCEL_OFFLINE: | |
77 | default: | |
78 | break; | |
79 | } | |
80 | ||
81 | return NOTIFY_OK; | |
82 | } | |
83 | ||
84 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
85 | ||
1da177e4 LT |
86 | /** |
87 | * ipc_init - initialise IPC subsystem | |
88 | * | |
89 | * The various system5 IPC resources (semaphores, messages and shared | |
72fd4a35 | 90 | * memory) are initialised |
b6b337ad ND |
91 | * A callback routine is registered into the memory hotplug notifier |
92 | * chain: since msgmni scales to lowmem this callback routine will be | |
93 | * called upon successful memory add / remove to recompute msmgni. | |
1da177e4 LT |
94 | */ |
95 | ||
96 | static int __init ipc_init(void) | |
97 | { | |
98 | sem_init(); | |
99 | msg_init(); | |
100 | shm_init(); | |
b6b337ad ND |
101 | hotplug_memory_notifier(ipc_memory_callback, IPC_CALLBACK_PRI); |
102 | register_ipcns_notifier(&init_ipc_ns); | |
1da177e4 LT |
103 | return 0; |
104 | } | |
105 | __initcall(ipc_init); | |
106 | ||
107 | /** | |
108 | * ipc_init_ids - initialise IPC identifiers | |
109 | * @ids: Identifier set | |
1da177e4 | 110 | * |
7ca7e564 ND |
111 | * Set up the sequence range to use for the ipc identifier range (limited |
112 | * below IPCMNI) then initialise the ids idr. | |
1da177e4 LT |
113 | */ |
114 | ||
7ca7e564 | 115 | void ipc_init_ids(struct ipc_ids *ids) |
1da177e4 | 116 | { |
3e148c79 | 117 | init_rwsem(&ids->rw_mutex); |
1da177e4 | 118 | |
1da177e4 | 119 | ids->in_use = 0; |
1da177e4 LT |
120 | ids->seq = 0; |
121 | { | |
122 | int seq_limit = INT_MAX/SEQ_MULTIPLIER; | |
123 | if(seq_limit > USHRT_MAX) | |
124 | ids->seq_max = USHRT_MAX; | |
125 | else | |
126 | ids->seq_max = seq_limit; | |
127 | } | |
128 | ||
7ca7e564 | 129 | idr_init(&ids->ipcs_idr); |
1da177e4 LT |
130 | } |
131 | ||
ae781774 | 132 | #ifdef CONFIG_PROC_FS |
9a32144e | 133 | static const struct file_operations sysvipc_proc_fops; |
ae781774 | 134 | /** |
72fd4a35 | 135 | * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface. |
ae781774 MW |
136 | * @path: Path in procfs |
137 | * @header: Banner to be printed at the beginning of the file. | |
138 | * @ids: ipc id table to iterate. | |
139 | * @show: show routine. | |
140 | */ | |
141 | void __init ipc_init_proc_interface(const char *path, const char *header, | |
73ea4130 | 142 | int ids, int (*show)(struct seq_file *, void *)) |
ae781774 MW |
143 | { |
144 | struct proc_dir_entry *pde; | |
145 | struct ipc_proc_iface *iface; | |
146 | ||
147 | iface = kmalloc(sizeof(*iface), GFP_KERNEL); | |
148 | if (!iface) | |
149 | return; | |
150 | iface->path = path; | |
151 | iface->header = header; | |
152 | iface->ids = ids; | |
153 | iface->show = show; | |
154 | ||
155 | pde = create_proc_entry(path, | |
156 | S_IRUGO, /* world readable */ | |
157 | NULL /* parent dir */); | |
158 | if (pde) { | |
159 | pde->data = iface; | |
160 | pde->proc_fops = &sysvipc_proc_fops; | |
161 | } else { | |
162 | kfree(iface); | |
163 | } | |
164 | } | |
165 | #endif | |
166 | ||
1da177e4 LT |
167 | /** |
168 | * ipc_findkey - find a key in an ipc identifier set | |
169 | * @ids: Identifier set | |
170 | * @key: The key to find | |
171 | * | |
3e148c79 | 172 | * Requires ipc_ids.rw_mutex locked. |
7ca7e564 ND |
173 | * Returns the LOCKED pointer to the ipc structure if found or NULL |
174 | * if not. | |
f4566f04 | 175 | * If key is found ipc points to the owning ipc structure |
1da177e4 LT |
176 | */ |
177 | ||
7748dbfa | 178 | static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key) |
1da177e4 | 179 | { |
7ca7e564 ND |
180 | struct kern_ipc_perm *ipc; |
181 | int next_id; | |
182 | int total; | |
1da177e4 | 183 | |
7ca7e564 ND |
184 | for (total = 0, next_id = 0; total < ids->in_use; next_id++) { |
185 | ipc = idr_find(&ids->ipcs_idr, next_id); | |
186 | ||
187 | if (ipc == NULL) | |
1da177e4 | 188 | continue; |
7ca7e564 ND |
189 | |
190 | if (ipc->key != key) { | |
191 | total++; | |
192 | continue; | |
193 | } | |
194 | ||
195 | ipc_lock_by_ptr(ipc); | |
196 | return ipc; | |
1da177e4 | 197 | } |
7ca7e564 ND |
198 | |
199 | return NULL; | |
1da177e4 LT |
200 | } |
201 | ||
7ca7e564 ND |
202 | /** |
203 | * ipc_get_maxid - get the last assigned id | |
204 | * @ids: IPC identifier set | |
205 | * | |
3e148c79 | 206 | * Called with ipc_ids.rw_mutex held. |
1da177e4 | 207 | */ |
1da177e4 | 208 | |
7ca7e564 ND |
209 | int ipc_get_maxid(struct ipc_ids *ids) |
210 | { | |
211 | struct kern_ipc_perm *ipc; | |
212 | int max_id = -1; | |
213 | int total, id; | |
214 | ||
215 | if (ids->in_use == 0) | |
216 | return -1; | |
1da177e4 | 217 | |
7ca7e564 ND |
218 | if (ids->in_use == IPCMNI) |
219 | return IPCMNI - 1; | |
220 | ||
221 | /* Look for the last assigned id */ | |
222 | total = 0; | |
223 | for (id = 0; id < IPCMNI && total < ids->in_use; id++) { | |
224 | ipc = idr_find(&ids->ipcs_idr, id); | |
225 | if (ipc != NULL) { | |
226 | max_id = id; | |
227 | total++; | |
228 | } | |
229 | } | |
230 | return max_id; | |
1da177e4 LT |
231 | } |
232 | ||
233 | /** | |
234 | * ipc_addid - add an IPC identifier | |
235 | * @ids: IPC identifier set | |
236 | * @new: new IPC permission set | |
7ca7e564 | 237 | * @size: limit for the number of used ids |
1da177e4 | 238 | * |
f4566f04 | 239 | * Add an entry 'new' to the IPC ids idr. The permissions object is |
1da177e4 | 240 | * initialised and the first free entry is set up and the id assigned |
f4566f04 | 241 | * is returned. The 'new' entry is returned in a locked state on success. |
283bb7fa | 242 | * On failure the entry is not locked and a negative err-code is returned. |
1da177e4 | 243 | * |
3e148c79 | 244 | * Called with ipc_ids.rw_mutex held as a writer. |
1da177e4 LT |
245 | */ |
246 | ||
247 | int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size) | |
248 | { | |
7ca7e564 | 249 | int id, err; |
1da177e4 | 250 | |
7ca7e564 ND |
251 | if (size > IPCMNI) |
252 | size = IPCMNI; | |
253 | ||
254 | if (ids->in_use >= size) | |
283bb7fa | 255 | return -ENOSPC; |
7ca7e564 ND |
256 | |
257 | err = idr_get_new(&ids->ipcs_idr, new, &id); | |
258 | if (err) | |
283bb7fa | 259 | return err; |
7ca7e564 | 260 | |
1da177e4 | 261 | ids->in_use++; |
1da177e4 LT |
262 | |
263 | new->cuid = new->uid = current->euid; | |
264 | new->gid = new->cgid = current->egid; | |
265 | ||
266 | new->seq = ids->seq++; | |
267 | if(ids->seq > ids->seq_max) | |
268 | ids->seq = 0; | |
269 | ||
48dea404 | 270 | new->id = ipc_buildid(id, new->seq); |
1da177e4 LT |
271 | spin_lock_init(&new->lock); |
272 | new->deleted = 0; | |
273 | rcu_read_lock(); | |
274 | spin_lock(&new->lock); | |
1da177e4 LT |
275 | return id; |
276 | } | |
277 | ||
7748dbfa ND |
278 | /** |
279 | * ipcget_new - create a new ipc object | |
280 | * @ns: namespace | |
f4566f04 | 281 | * @ids: IPC identifer set |
7748dbfa ND |
282 | * @ops: the actual creation routine to call |
283 | * @params: its parameters | |
284 | * | |
f4566f04 ND |
285 | * This routine is called by sys_msgget, sys_semget() and sys_shmget() |
286 | * when the key is IPC_PRIVATE. | |
7748dbfa | 287 | */ |
b2d75cdd | 288 | static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids, |
7748dbfa ND |
289 | struct ipc_ops *ops, struct ipc_params *params) |
290 | { | |
291 | int err; | |
283bb7fa | 292 | retry: |
7748dbfa ND |
293 | err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL); |
294 | ||
295 | if (!err) | |
296 | return -ENOMEM; | |
297 | ||
3e148c79 | 298 | down_write(&ids->rw_mutex); |
7748dbfa | 299 | err = ops->getnew(ns, params); |
3e148c79 | 300 | up_write(&ids->rw_mutex); |
7748dbfa | 301 | |
283bb7fa PP |
302 | if (err == -EAGAIN) |
303 | goto retry; | |
304 | ||
7748dbfa ND |
305 | return err; |
306 | } | |
307 | ||
308 | /** | |
309 | * ipc_check_perms - check security and permissions for an IPC | |
310 | * @ipcp: ipc permission set | |
7748dbfa ND |
311 | * @ops: the actual security routine to call |
312 | * @params: its parameters | |
f4566f04 ND |
313 | * |
314 | * This routine is called by sys_msgget(), sys_semget() and sys_shmget() | |
315 | * when the key is not IPC_PRIVATE and that key already exists in the | |
316 | * ids IDR. | |
317 | * | |
318 | * On success, the IPC id is returned. | |
319 | * | |
3e148c79 | 320 | * It is called with ipc_ids.rw_mutex and ipcp->lock held. |
7748dbfa ND |
321 | */ |
322 | static int ipc_check_perms(struct kern_ipc_perm *ipcp, struct ipc_ops *ops, | |
323 | struct ipc_params *params) | |
324 | { | |
325 | int err; | |
326 | ||
327 | if (ipcperms(ipcp, params->flg)) | |
328 | err = -EACCES; | |
329 | else { | |
330 | err = ops->associate(ipcp, params->flg); | |
331 | if (!err) | |
332 | err = ipcp->id; | |
333 | } | |
334 | ||
335 | return err; | |
336 | } | |
337 | ||
338 | /** | |
339 | * ipcget_public - get an ipc object or create a new one | |
340 | * @ns: namespace | |
f4566f04 | 341 | * @ids: IPC identifer set |
7748dbfa ND |
342 | * @ops: the actual creation routine to call |
343 | * @params: its parameters | |
344 | * | |
f4566f04 ND |
345 | * This routine is called by sys_msgget, sys_semget() and sys_shmget() |
346 | * when the key is not IPC_PRIVATE. | |
347 | * It adds a new entry if the key is not found and does some permission | |
348 | * / security checkings if the key is found. | |
349 | * | |
350 | * On success, the ipc id is returned. | |
7748dbfa | 351 | */ |
b2d75cdd | 352 | static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids, |
7748dbfa ND |
353 | struct ipc_ops *ops, struct ipc_params *params) |
354 | { | |
355 | struct kern_ipc_perm *ipcp; | |
356 | int flg = params->flg; | |
357 | int err; | |
283bb7fa | 358 | retry: |
7748dbfa ND |
359 | err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL); |
360 | ||
3e148c79 ND |
361 | /* |
362 | * Take the lock as a writer since we are potentially going to add | |
363 | * a new entry + read locks are not "upgradable" | |
364 | */ | |
365 | down_write(&ids->rw_mutex); | |
7748dbfa ND |
366 | ipcp = ipc_findkey(ids, params->key); |
367 | if (ipcp == NULL) { | |
368 | /* key not used */ | |
369 | if (!(flg & IPC_CREAT)) | |
370 | err = -ENOENT; | |
371 | else if (!err) | |
372 | err = -ENOMEM; | |
373 | else | |
374 | err = ops->getnew(ns, params); | |
375 | } else { | |
376 | /* ipc object has been locked by ipc_findkey() */ | |
377 | ||
378 | if (flg & IPC_CREAT && flg & IPC_EXCL) | |
379 | err = -EEXIST; | |
380 | else { | |
381 | err = 0; | |
382 | if (ops->more_checks) | |
383 | err = ops->more_checks(ipcp, params); | |
384 | if (!err) | |
f4566f04 ND |
385 | /* |
386 | * ipc_check_perms returns the IPC id on | |
387 | * success | |
388 | */ | |
7748dbfa ND |
389 | err = ipc_check_perms(ipcp, ops, params); |
390 | } | |
391 | ipc_unlock(ipcp); | |
392 | } | |
3e148c79 | 393 | up_write(&ids->rw_mutex); |
7748dbfa | 394 | |
283bb7fa PP |
395 | if (err == -EAGAIN) |
396 | goto retry; | |
397 | ||
7748dbfa ND |
398 | return err; |
399 | } | |
400 | ||
401 | ||
1da177e4 LT |
402 | /** |
403 | * ipc_rmid - remove an IPC identifier | |
f4566f04 ND |
404 | * @ids: IPC identifier set |
405 | * @ipcp: ipc perm structure containing the identifier to remove | |
1da177e4 | 406 | * |
3e148c79 ND |
407 | * ipc_ids.rw_mutex (as a writer) and the spinlock for this ID are held |
408 | * before this function is called, and remain locked on the exit. | |
1da177e4 LT |
409 | */ |
410 | ||
7ca7e564 | 411 | void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp) |
1da177e4 | 412 | { |
ce621f5b | 413 | int lid = ipcid_to_idx(ipcp->id); |
7ca7e564 ND |
414 | |
415 | idr_remove(&ids->ipcs_idr, lid); | |
1da177e4 | 416 | |
1da177e4 LT |
417 | ids->in_use--; |
418 | ||
7ca7e564 ND |
419 | ipcp->deleted = 1; |
420 | ||
421 | return; | |
1da177e4 LT |
422 | } |
423 | ||
424 | /** | |
425 | * ipc_alloc - allocate ipc space | |
426 | * @size: size desired | |
427 | * | |
428 | * Allocate memory from the appropriate pools and return a pointer to it. | |
429 | * NULL is returned if the allocation fails | |
430 | */ | |
431 | ||
432 | void* ipc_alloc(int size) | |
433 | { | |
434 | void* out; | |
435 | if(size > PAGE_SIZE) | |
436 | out = vmalloc(size); | |
437 | else | |
438 | out = kmalloc(size, GFP_KERNEL); | |
439 | return out; | |
440 | } | |
441 | ||
442 | /** | |
443 | * ipc_free - free ipc space | |
444 | * @ptr: pointer returned by ipc_alloc | |
445 | * @size: size of block | |
446 | * | |
72fd4a35 | 447 | * Free a block created with ipc_alloc(). The caller must know the size |
1da177e4 LT |
448 | * used in the allocation call. |
449 | */ | |
450 | ||
451 | void ipc_free(void* ptr, int size) | |
452 | { | |
453 | if(size > PAGE_SIZE) | |
454 | vfree(ptr); | |
455 | else | |
456 | kfree(ptr); | |
457 | } | |
458 | ||
459 | /* | |
460 | * rcu allocations: | |
461 | * There are three headers that are prepended to the actual allocation: | |
462 | * - during use: ipc_rcu_hdr. | |
463 | * - during the rcu grace period: ipc_rcu_grace. | |
464 | * - [only if vmalloc]: ipc_rcu_sched. | |
465 | * Their lifetime doesn't overlap, thus the headers share the same memory. | |
466 | * Unlike a normal union, they are right-aligned, thus some container_of | |
467 | * forward/backward casting is necessary: | |
468 | */ | |
469 | struct ipc_rcu_hdr | |
470 | { | |
471 | int refcount; | |
472 | int is_vmalloc; | |
473 | void *data[0]; | |
474 | }; | |
475 | ||
476 | ||
477 | struct ipc_rcu_grace | |
478 | { | |
479 | struct rcu_head rcu; | |
480 | /* "void *" makes sure alignment of following data is sane. */ | |
481 | void *data[0]; | |
482 | }; | |
483 | ||
484 | struct ipc_rcu_sched | |
485 | { | |
486 | struct work_struct work; | |
487 | /* "void *" makes sure alignment of following data is sane. */ | |
488 | void *data[0]; | |
489 | }; | |
490 | ||
491 | #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \ | |
492 | sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr)) | |
493 | #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \ | |
494 | sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC) | |
495 | ||
496 | static inline int rcu_use_vmalloc(int size) | |
497 | { | |
498 | /* Too big for a single page? */ | |
499 | if (HDRLEN_KMALLOC + size > PAGE_SIZE) | |
500 | return 1; | |
501 | return 0; | |
502 | } | |
503 | ||
504 | /** | |
505 | * ipc_rcu_alloc - allocate ipc and rcu space | |
506 | * @size: size desired | |
507 | * | |
508 | * Allocate memory for the rcu header structure + the object. | |
509 | * Returns the pointer to the object. | |
510 | * NULL is returned if the allocation fails. | |
511 | */ | |
512 | ||
513 | void* ipc_rcu_alloc(int size) | |
514 | { | |
515 | void* out; | |
516 | /* | |
517 | * We prepend the allocation with the rcu struct, and | |
518 | * workqueue if necessary (for vmalloc). | |
519 | */ | |
520 | if (rcu_use_vmalloc(size)) { | |
521 | out = vmalloc(HDRLEN_VMALLOC + size); | |
522 | if (out) { | |
523 | out += HDRLEN_VMALLOC; | |
524 | container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1; | |
525 | container_of(out, struct ipc_rcu_hdr, data)->refcount = 1; | |
526 | } | |
527 | } else { | |
528 | out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL); | |
529 | if (out) { | |
530 | out += HDRLEN_KMALLOC; | |
531 | container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0; | |
532 | container_of(out, struct ipc_rcu_hdr, data)->refcount = 1; | |
533 | } | |
534 | } | |
535 | ||
536 | return out; | |
537 | } | |
538 | ||
539 | void ipc_rcu_getref(void *ptr) | |
540 | { | |
541 | container_of(ptr, struct ipc_rcu_hdr, data)->refcount++; | |
542 | } | |
543 | ||
65f27f38 DH |
544 | static void ipc_do_vfree(struct work_struct *work) |
545 | { | |
546 | vfree(container_of(work, struct ipc_rcu_sched, work)); | |
547 | } | |
548 | ||
1da177e4 | 549 | /** |
1e5d5331 RD |
550 | * ipc_schedule_free - free ipc + rcu space |
551 | * @head: RCU callback structure for queued work | |
1da177e4 LT |
552 | * |
553 | * Since RCU callback function is called in bh, | |
72fd4a35 | 554 | * we need to defer the vfree to schedule_work(). |
1da177e4 LT |
555 | */ |
556 | static void ipc_schedule_free(struct rcu_head *head) | |
557 | { | |
f4566f04 ND |
558 | struct ipc_rcu_grace *grace; |
559 | struct ipc_rcu_sched *sched; | |
560 | ||
561 | grace = container_of(head, struct ipc_rcu_grace, rcu); | |
562 | sched = container_of(&(grace->data[0]), struct ipc_rcu_sched, | |
563 | data[0]); | |
1da177e4 | 564 | |
65f27f38 | 565 | INIT_WORK(&sched->work, ipc_do_vfree); |
1da177e4 LT |
566 | schedule_work(&sched->work); |
567 | } | |
568 | ||
569 | /** | |
1e5d5331 RD |
570 | * ipc_immediate_free - free ipc + rcu space |
571 | * @head: RCU callback structure that contains pointer to be freed | |
1da177e4 | 572 | * |
72fd4a35 | 573 | * Free from the RCU callback context. |
1da177e4 LT |
574 | */ |
575 | static void ipc_immediate_free(struct rcu_head *head) | |
576 | { | |
577 | struct ipc_rcu_grace *free = | |
578 | container_of(head, struct ipc_rcu_grace, rcu); | |
579 | kfree(free); | |
580 | } | |
581 | ||
582 | void ipc_rcu_putref(void *ptr) | |
583 | { | |
584 | if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0) | |
585 | return; | |
586 | ||
587 | if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) { | |
588 | call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu, | |
589 | ipc_schedule_free); | |
590 | } else { | |
591 | call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu, | |
592 | ipc_immediate_free); | |
593 | } | |
594 | } | |
595 | ||
596 | /** | |
597 | * ipcperms - check IPC permissions | |
598 | * @ipcp: IPC permission set | |
599 | * @flag: desired permission set. | |
600 | * | |
601 | * Check user, group, other permissions for access | |
602 | * to ipc resources. return 0 if allowed | |
603 | */ | |
604 | ||
605 | int ipcperms (struct kern_ipc_perm *ipcp, short flag) | |
606 | { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */ | |
073115d6 | 607 | int requested_mode, granted_mode, err; |
1da177e4 | 608 | |
073115d6 SG |
609 | if (unlikely((err = audit_ipc_obj(ipcp)))) |
610 | return err; | |
1da177e4 LT |
611 | requested_mode = (flag >> 6) | (flag >> 3) | flag; |
612 | granted_mode = ipcp->mode; | |
613 | if (current->euid == ipcp->cuid || current->euid == ipcp->uid) | |
614 | granted_mode >>= 6; | |
615 | else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid)) | |
616 | granted_mode >>= 3; | |
617 | /* is there some bit set in requested_mode but not in granted_mode? */ | |
618 | if ((requested_mode & ~granted_mode & 0007) && | |
619 | !capable(CAP_IPC_OWNER)) | |
620 | return -1; | |
621 | ||
622 | return security_ipc_permission(ipcp, flag); | |
623 | } | |
624 | ||
625 | /* | |
626 | * Functions to convert between the kern_ipc_perm structure and the | |
627 | * old/new ipc_perm structures | |
628 | */ | |
629 | ||
630 | /** | |
631 | * kernel_to_ipc64_perm - convert kernel ipc permissions to user | |
632 | * @in: kernel permissions | |
633 | * @out: new style IPC permissions | |
634 | * | |
72fd4a35 RD |
635 | * Turn the kernel object @in into a set of permissions descriptions |
636 | * for returning to userspace (@out). | |
1da177e4 LT |
637 | */ |
638 | ||
639 | ||
640 | void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out) | |
641 | { | |
642 | out->key = in->key; | |
643 | out->uid = in->uid; | |
644 | out->gid = in->gid; | |
645 | out->cuid = in->cuid; | |
646 | out->cgid = in->cgid; | |
647 | out->mode = in->mode; | |
648 | out->seq = in->seq; | |
649 | } | |
650 | ||
651 | /** | |
f4566f04 | 652 | * ipc64_perm_to_ipc_perm - convert new ipc permissions to old |
1da177e4 LT |
653 | * @in: new style IPC permissions |
654 | * @out: old style IPC permissions | |
655 | * | |
72fd4a35 RD |
656 | * Turn the new style permissions object @in into a compatibility |
657 | * object and store it into the @out pointer. | |
1da177e4 LT |
658 | */ |
659 | ||
660 | void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out) | |
661 | { | |
662 | out->key = in->key; | |
663 | SET_UID(out->uid, in->uid); | |
664 | SET_GID(out->gid, in->gid); | |
665 | SET_UID(out->cuid, in->cuid); | |
666 | SET_GID(out->cgid, in->cgid); | |
667 | out->mode = in->mode; | |
668 | out->seq = in->seq; | |
669 | } | |
670 | ||
f4566f04 | 671 | /** |
3e148c79 | 672 | * ipc_lock - Lock an ipc structure without rw_mutex held |
f4566f04 ND |
673 | * @ids: IPC identifier set |
674 | * @id: ipc id to look for | |
675 | * | |
676 | * Look for an id in the ipc ids idr and lock the associated ipc object. | |
677 | * | |
f4566f04 | 678 | * The ipc object is locked on exit. |
3e148c79 ND |
679 | * |
680 | * This is the routine that should be called when the rw_mutex is not already | |
681 | * held, i.e. idr tree not protected: it protects the idr tree in read mode | |
682 | * during the idr_find(). | |
f4566f04 ND |
683 | */ |
684 | ||
7ca7e564 | 685 | struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id) |
1da177e4 | 686 | { |
7ca7e564 | 687 | struct kern_ipc_perm *out; |
ce621f5b | 688 | int lid = ipcid_to_idx(id); |
1da177e4 | 689 | |
3e148c79 ND |
690 | down_read(&ids->rw_mutex); |
691 | ||
1da177e4 | 692 | rcu_read_lock(); |
7ca7e564 ND |
693 | out = idr_find(&ids->ipcs_idr, lid); |
694 | if (out == NULL) { | |
1da177e4 | 695 | rcu_read_unlock(); |
3e148c79 | 696 | up_read(&ids->rw_mutex); |
023a5355 | 697 | return ERR_PTR(-EINVAL); |
1da177e4 | 698 | } |
7ca7e564 | 699 | |
3e148c79 ND |
700 | up_read(&ids->rw_mutex); |
701 | ||
1da177e4 LT |
702 | spin_lock(&out->lock); |
703 | ||
704 | /* ipc_rmid() may have already freed the ID while ipc_lock | |
705 | * was spinning: here verify that the structure is still valid | |
706 | */ | |
707 | if (out->deleted) { | |
708 | spin_unlock(&out->lock); | |
709 | rcu_read_unlock(); | |
023a5355 | 710 | return ERR_PTR(-EINVAL); |
1da177e4 | 711 | } |
7ca7e564 | 712 | |
1da177e4 LT |
713 | return out; |
714 | } | |
715 | ||
3e148c79 ND |
716 | /** |
717 | * ipc_lock_down - Lock an ipc structure with rw_sem held | |
718 | * @ids: IPC identifier set | |
719 | * @id: ipc id to look for | |
720 | * | |
721 | * Look for an id in the ipc ids idr and lock the associated ipc object. | |
722 | * | |
723 | * The ipc object is locked on exit. | |
724 | * | |
725 | * This is the routine that should be called when the rw_mutex is already | |
726 | * held, i.e. idr tree protected. | |
727 | */ | |
728 | ||
729 | struct kern_ipc_perm *ipc_lock_down(struct ipc_ids *ids, int id) | |
730 | { | |
731 | struct kern_ipc_perm *out; | |
732 | int lid = ipcid_to_idx(id); | |
733 | ||
734 | rcu_read_lock(); | |
735 | out = idr_find(&ids->ipcs_idr, lid); | |
736 | if (out == NULL) { | |
737 | rcu_read_unlock(); | |
738 | return ERR_PTR(-EINVAL); | |
739 | } | |
740 | ||
741 | spin_lock(&out->lock); | |
742 | ||
743 | /* | |
744 | * No need to verify that the structure is still valid since the | |
745 | * rw_mutex is held. | |
746 | */ | |
747 | return out; | |
748 | } | |
749 | ||
b2d75cdd PE |
750 | struct kern_ipc_perm *ipc_lock_check_down(struct ipc_ids *ids, int id) |
751 | { | |
752 | struct kern_ipc_perm *out; | |
753 | ||
754 | out = ipc_lock_down(ids, id); | |
755 | if (IS_ERR(out)) | |
756 | return out; | |
757 | ||
758 | if (ipc_checkid(out, id)) { | |
759 | ipc_unlock(out); | |
760 | return ERR_PTR(-EIDRM); | |
761 | } | |
762 | ||
763 | return out; | |
764 | } | |
765 | ||
766 | struct kern_ipc_perm *ipc_lock_check(struct ipc_ids *ids, int id) | |
767 | { | |
768 | struct kern_ipc_perm *out; | |
769 | ||
770 | out = ipc_lock(ids, id); | |
771 | if (IS_ERR(out)) | |
772 | return out; | |
773 | ||
774 | if (ipc_checkid(out, id)) { | |
775 | ipc_unlock(out); | |
776 | return ERR_PTR(-EIDRM); | |
777 | } | |
778 | ||
779 | return out; | |
780 | } | |
781 | ||
782 | /** | |
783 | * ipcget - Common sys_*get() code | |
784 | * @ns : namsepace | |
785 | * @ids : IPC identifier set | |
786 | * @ops : operations to be called on ipc object creation, permission checks | |
787 | * and further checks | |
788 | * @params : the parameters needed by the previous operations. | |
789 | * | |
790 | * Common routine called by sys_msgget(), sys_semget() and sys_shmget(). | |
791 | */ | |
792 | int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids, | |
793 | struct ipc_ops *ops, struct ipc_params *params) | |
794 | { | |
795 | if (params->key == IPC_PRIVATE) | |
796 | return ipcget_new(ns, ids, ops, params); | |
797 | else | |
798 | return ipcget_public(ns, ids, ops, params); | |
799 | } | |
800 | ||
1da177e4 LT |
801 | #ifdef __ARCH_WANT_IPC_PARSE_VERSION |
802 | ||
803 | ||
804 | /** | |
805 | * ipc_parse_version - IPC call version | |
806 | * @cmd: pointer to command | |
807 | * | |
808 | * Return IPC_64 for new style IPC and IPC_OLD for old style IPC. | |
72fd4a35 | 809 | * The @cmd value is turned from an encoding command and version into |
1da177e4 LT |
810 | * just the command code. |
811 | */ | |
812 | ||
813 | int ipc_parse_version (int *cmd) | |
814 | { | |
815 | if (*cmd & IPC_64) { | |
816 | *cmd ^= IPC_64; | |
817 | return IPC_64; | |
818 | } else { | |
819 | return IPC_OLD; | |
820 | } | |
821 | } | |
822 | ||
823 | #endif /* __ARCH_WANT_IPC_PARSE_VERSION */ | |
ae781774 MW |
824 | |
825 | #ifdef CONFIG_PROC_FS | |
bc1fc6d8 EB |
826 | struct ipc_proc_iter { |
827 | struct ipc_namespace *ns; | |
828 | struct ipc_proc_iface *iface; | |
829 | }; | |
830 | ||
7ca7e564 ND |
831 | /* |
832 | * This routine locks the ipc structure found at least at position pos. | |
833 | */ | |
b524b9ad AB |
834 | static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos, |
835 | loff_t *new_pos) | |
ae781774 | 836 | { |
7ca7e564 ND |
837 | struct kern_ipc_perm *ipc; |
838 | int total, id; | |
73ea4130 | 839 | |
7ca7e564 ND |
840 | total = 0; |
841 | for (id = 0; id < pos && total < ids->in_use; id++) { | |
842 | ipc = idr_find(&ids->ipcs_idr, id); | |
843 | if (ipc != NULL) | |
844 | total++; | |
845 | } | |
ae781774 | 846 | |
7ca7e564 ND |
847 | if (total >= ids->in_use) |
848 | return NULL; | |
ae781774 | 849 | |
7ca7e564 ND |
850 | for ( ; pos < IPCMNI; pos++) { |
851 | ipc = idr_find(&ids->ipcs_idr, pos); | |
852 | if (ipc != NULL) { | |
853 | *new_pos = pos + 1; | |
854 | ipc_lock_by_ptr(ipc); | |
ae781774 MW |
855 | return ipc; |
856 | } | |
857 | } | |
858 | ||
859 | /* Out of range - return NULL to terminate iteration */ | |
860 | return NULL; | |
861 | } | |
862 | ||
7ca7e564 ND |
863 | static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos) |
864 | { | |
865 | struct ipc_proc_iter *iter = s->private; | |
866 | struct ipc_proc_iface *iface = iter->iface; | |
867 | struct kern_ipc_perm *ipc = it; | |
868 | ||
869 | /* If we had an ipc id locked before, unlock it */ | |
870 | if (ipc && ipc != SEQ_START_TOKEN) | |
871 | ipc_unlock(ipc); | |
872 | ||
ed2ddbf8 | 873 | return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos); |
7ca7e564 ND |
874 | } |
875 | ||
ae781774 | 876 | /* |
f4566f04 ND |
877 | * File positions: pos 0 -> header, pos n -> ipc id = n - 1. |
878 | * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START. | |
ae781774 MW |
879 | */ |
880 | static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos) | |
881 | { | |
bc1fc6d8 EB |
882 | struct ipc_proc_iter *iter = s->private; |
883 | struct ipc_proc_iface *iface = iter->iface; | |
73ea4130 KK |
884 | struct ipc_ids *ids; |
885 | ||
ed2ddbf8 | 886 | ids = &iter->ns->ids[iface->ids]; |
ae781774 MW |
887 | |
888 | /* | |
889 | * Take the lock - this will be released by the corresponding | |
890 | * call to stop(). | |
891 | */ | |
3e148c79 | 892 | down_read(&ids->rw_mutex); |
ae781774 MW |
893 | |
894 | /* pos < 0 is invalid */ | |
895 | if (*pos < 0) | |
896 | return NULL; | |
897 | ||
898 | /* pos == 0 means header */ | |
899 | if (*pos == 0) | |
900 | return SEQ_START_TOKEN; | |
901 | ||
902 | /* Find the (pos-1)th ipc */ | |
7ca7e564 | 903 | return sysvipc_find_ipc(ids, *pos - 1, pos); |
ae781774 MW |
904 | } |
905 | ||
906 | static void sysvipc_proc_stop(struct seq_file *s, void *it) | |
907 | { | |
908 | struct kern_ipc_perm *ipc = it; | |
bc1fc6d8 EB |
909 | struct ipc_proc_iter *iter = s->private; |
910 | struct ipc_proc_iface *iface = iter->iface; | |
73ea4130 | 911 | struct ipc_ids *ids; |
ae781774 | 912 | |
f4566f04 | 913 | /* If we had a locked structure, release it */ |
ae781774 MW |
914 | if (ipc && ipc != SEQ_START_TOKEN) |
915 | ipc_unlock(ipc); | |
916 | ||
ed2ddbf8 | 917 | ids = &iter->ns->ids[iface->ids]; |
ae781774 | 918 | /* Release the lock we took in start() */ |
3e148c79 | 919 | up_read(&ids->rw_mutex); |
ae781774 MW |
920 | } |
921 | ||
922 | static int sysvipc_proc_show(struct seq_file *s, void *it) | |
923 | { | |
bc1fc6d8 EB |
924 | struct ipc_proc_iter *iter = s->private; |
925 | struct ipc_proc_iface *iface = iter->iface; | |
ae781774 MW |
926 | |
927 | if (it == SEQ_START_TOKEN) | |
928 | return seq_puts(s, iface->header); | |
929 | ||
930 | return iface->show(s, it); | |
931 | } | |
932 | ||
933 | static struct seq_operations sysvipc_proc_seqops = { | |
934 | .start = sysvipc_proc_start, | |
935 | .stop = sysvipc_proc_stop, | |
936 | .next = sysvipc_proc_next, | |
937 | .show = sysvipc_proc_show, | |
938 | }; | |
939 | ||
bc1fc6d8 EB |
940 | static int sysvipc_proc_open(struct inode *inode, struct file *file) |
941 | { | |
ae781774 MW |
942 | int ret; |
943 | struct seq_file *seq; | |
bc1fc6d8 EB |
944 | struct ipc_proc_iter *iter; |
945 | ||
946 | ret = -ENOMEM; | |
947 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
948 | if (!iter) | |
949 | goto out; | |
ae781774 MW |
950 | |
951 | ret = seq_open(file, &sysvipc_proc_seqops); | |
bc1fc6d8 EB |
952 | if (ret) |
953 | goto out_kfree; | |
954 | ||
955 | seq = file->private_data; | |
956 | seq->private = iter; | |
957 | ||
958 | iter->iface = PDE(inode)->data; | |
959 | iter->ns = get_ipc_ns(current->nsproxy->ipc_ns); | |
960 | out: | |
ae781774 | 961 | return ret; |
bc1fc6d8 EB |
962 | out_kfree: |
963 | kfree(iter); | |
964 | goto out; | |
965 | } | |
966 | ||
967 | static int sysvipc_proc_release(struct inode *inode, struct file *file) | |
968 | { | |
969 | struct seq_file *seq = file->private_data; | |
970 | struct ipc_proc_iter *iter = seq->private; | |
971 | put_ipc_ns(iter->ns); | |
972 | return seq_release_private(inode, file); | |
ae781774 MW |
973 | } |
974 | ||
9a32144e | 975 | static const struct file_operations sysvipc_proc_fops = { |
ae781774 MW |
976 | .open = sysvipc_proc_open, |
977 | .read = seq_read, | |
978 | .llseek = seq_lseek, | |
bc1fc6d8 | 979 | .release = sysvipc_proc_release, |
ae781774 MW |
980 | }; |
981 | #endif /* CONFIG_PROC_FS */ |