blob: 7a5d5e393c4bce64df495242a3bf1437a1793f66 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Christian Kujau624dffc2006-01-15 02:43:54 +010010 * Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
12 * Mingming Cao <cmm@us.ibm.com>
Steve Grubb073115d2006-04-02 17:07:33 -040013 * Mar 2006 - support for audit of ipc object properties
14 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaev73ea4132006-10-02 02:18:20 -070015 * Jun 2006 - namespaces ssupport
16 * OpenVZ, SWsoft Inc.
17 * Pavel Emelianov <xemul@openvz.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/shm.h>
22#include <linux/init.h>
23#include <linux/msg.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/vmalloc.h>
25#include <linux/slab.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080026#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/highuid.h>
28#include <linux/security.h>
29#include <linux/rcupdate.h>
30#include <linux/workqueue.h>
Mike Waychisonae781772005-09-06 15:17:09 -070031#include <linux/seq_file.h>
32#include <linux/proc_fs.h>
Steve Grubb073115d2006-04-02 17:07:33 -040033#include <linux/audit.h>
Kirill Korotaev73ea4132006-10-02 02:18:20 -070034#include <linux/nsproxy.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070035#include <linux/rwsem.h>
Nadia Derbeyb6b337a2008-04-29 01:00:42 -070036#include <linux/memory.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080037#include <linux/ipc_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include <asm/unistd.h>
40
41#include "util.h"
42
Mike Waychisonae781772005-09-06 15:17:09 -070043struct ipc_proc_iface {
44 const char *path;
45 const char *header;
Kirill Korotaev73ea4132006-10-02 02:18:20 -070046 int ids;
Mike Waychisonae781772005-09-06 15:17:09 -070047 int (*show)(struct seq_file *, void *);
48};
49
Kirill Korotaev73ea4132006-10-02 02:18:20 -070050struct ipc_namespace init_ipc_ns = {
51 .kref = {
52 .refcount = ATOMIC_INIT(2),
53 },
54};
55
Nadia Derbey4d89dc62008-04-29 01:00:40 -070056atomic_t nr_ipc_ns = ATOMIC_INIT(1);
57
58
Nadia Derbeyb6b337a2008-04-29 01:00:42 -070059#ifdef CONFIG_MEMORY_HOTPLUG
60
Nadia Derbey424450c2008-04-29 01:00:43 -070061static void ipc_memory_notifier(struct work_struct *work)
62{
63 ipcns_notify(IPCNS_MEMCHANGED);
64}
65
66static DECLARE_WORK(ipc_memory_wq, ipc_memory_notifier);
67
68
Nadia Derbeyb6b337a2008-04-29 01:00:42 -070069static int ipc_memory_callback(struct notifier_block *self,
70 unsigned long action, void *arg)
71{
72 switch (action) {
73 case MEM_ONLINE: /* memory successfully brought online */
74 case MEM_OFFLINE: /* or offline: it's time to recompute msgmni */
75 /*
76 * This is done by invoking the ipcns notifier chain with the
77 * IPC_MEMCHANGED event.
Nadia Derbey424450c2008-04-29 01:00:43 -070078 * In order not to keep the lock on the hotplug memory chain
79 * for too long, queue a work item that will, when waken up,
80 * activate the ipcns notification chain.
81 * No need to keep several ipc work items on the queue.
Nadia Derbeyb6b337a2008-04-29 01:00:42 -070082 */
Nadia Derbey424450c2008-04-29 01:00:43 -070083 if (!work_pending(&ipc_memory_wq))
84 schedule_work(&ipc_memory_wq);
Nadia Derbeyb6b337a2008-04-29 01:00:42 -070085 break;
86 case MEM_GOING_ONLINE:
87 case MEM_GOING_OFFLINE:
88 case MEM_CANCEL_ONLINE:
89 case MEM_CANCEL_OFFLINE:
90 default:
91 break;
92 }
93
94 return NOTIFY_OK;
95}
96
97#endif /* CONFIG_MEMORY_HOTPLUG */
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099/**
100 * ipc_init - initialise IPC subsystem
101 *
102 * The various system5 IPC resources (semaphores, messages and shared
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800103 * memory) are initialised
Nadia Derbeyb6b337a2008-04-29 01:00:42 -0700104 * A callback routine is registered into the memory hotplug notifier
105 * chain: since msgmni scales to lowmem this callback routine will be
106 * called upon successful memory add / remove to recompute msmgni.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 */
108
109static int __init ipc_init(void)
110{
111 sem_init();
112 msg_init();
113 shm_init();
Nadia Derbeyb6b337a2008-04-29 01:00:42 -0700114 hotplug_memory_notifier(ipc_memory_callback, IPC_CALLBACK_PRI);
115 register_ipcns_notifier(&init_ipc_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 return 0;
117}
118__initcall(ipc_init);
119
120/**
121 * ipc_init_ids - initialise IPC identifiers
122 * @ids: Identifier set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 *
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700124 * Set up the sequence range to use for the ipc identifier range (limited
125 * below IPCMNI) then initialise the ids idr.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 */
127
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700128void ipc_init_ids(struct ipc_ids *ids)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
Nadia Derbey3e148c72007-10-18 23:40:54 -0700130 init_rwsem(&ids->rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 ids->in_use = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 ids->seq = 0;
134 {
135 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
136 if(seq_limit > USHRT_MAX)
137 ids->seq_max = USHRT_MAX;
138 else
139 ids->seq_max = seq_limit;
140 }
141
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700142 idr_init(&ids->ipcs_idr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
Mike Waychisonae781772005-09-06 15:17:09 -0700145#ifdef CONFIG_PROC_FS
Arjan van de Ven9a321442007-02-12 00:55:35 -0800146static const struct file_operations sysvipc_proc_fops;
Mike Waychisonae781772005-09-06 15:17:09 -0700147/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800148 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
Mike Waychisonae781772005-09-06 15:17:09 -0700149 * @path: Path in procfs
150 * @header: Banner to be printed at the beginning of the file.
151 * @ids: ipc id table to iterate.
152 * @show: show routine.
153 */
154void __init ipc_init_proc_interface(const char *path, const char *header,
Kirill Korotaev73ea4132006-10-02 02:18:20 -0700155 int ids, int (*show)(struct seq_file *, void *))
Mike Waychisonae781772005-09-06 15:17:09 -0700156{
157 struct proc_dir_entry *pde;
158 struct ipc_proc_iface *iface;
159
160 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
161 if (!iface)
162 return;
163 iface->path = path;
164 iface->header = header;
165 iface->ids = ids;
166 iface->show = show;
167
168 pde = create_proc_entry(path,
169 S_IRUGO, /* world readable */
170 NULL /* parent dir */);
171 if (pde) {
172 pde->data = iface;
173 pde->proc_fops = &sysvipc_proc_fops;
174 } else {
175 kfree(iface);
176 }
177}
178#endif
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180/**
181 * ipc_findkey - find a key in an ipc identifier set
182 * @ids: Identifier set
183 * @key: The key to find
184 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700185 * Requires ipc_ids.rw_mutex locked.
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700186 * Returns the LOCKED pointer to the ipc structure if found or NULL
187 * if not.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700188 * If key is found ipc points to the owning ipc structure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 */
190
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700191static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700193 struct kern_ipc_perm *ipc;
194 int next_id;
195 int total;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700197 for (total = 0, next_id = 0; total < ids->in_use; next_id++) {
198 ipc = idr_find(&ids->ipcs_idr, next_id);
199
200 if (ipc == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 continue;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700202
203 if (ipc->key != key) {
204 total++;
205 continue;
206 }
207
208 ipc_lock_by_ptr(ipc);
209 return ipc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700211
212 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700215/**
216 * ipc_get_maxid - get the last assigned id
217 * @ids: IPC identifier set
218 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700219 * Called with ipc_ids.rw_mutex held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 */
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700221
222int ipc_get_maxid(struct ipc_ids *ids)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700224 struct kern_ipc_perm *ipc;
225 int max_id = -1;
226 int total, id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700228 if (ids->in_use == 0)
229 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700231 if (ids->in_use == IPCMNI)
232 return IPCMNI - 1;
233
234 /* Look for the last assigned id */
235 total = 0;
236 for (id = 0; id < IPCMNI && total < ids->in_use; id++) {
237 ipc = idr_find(&ids->ipcs_idr, id);
238 if (ipc != NULL) {
239 max_id = id;
240 total++;
241 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700243 return max_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
245
246/**
247 * ipc_addid - add an IPC identifier
248 * @ids: IPC identifier set
249 * @new: new IPC permission set
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700250 * @size: limit for the number of used ids
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700252 * Add an entry 'new' to the IPC ids idr. The permissions object is
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 * initialised and the first free entry is set up and the id assigned
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700254 * is returned. The 'new' entry is returned in a locked state on success.
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700255 * On failure the entry is not locked and a negative err-code is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700257 * Called with ipc_ids.rw_mutex held as a writer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 */
259
260int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
261{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700262 int id, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700264 if (size > IPCMNI)
265 size = IPCMNI;
266
267 if (ids->in_use >= size)
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700268 return -ENOSPC;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700269
270 err = idr_get_new(&ids->ipcs_idr, new, &id);
271 if (err)
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700272 return err;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 ids->in_use++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276 new->cuid = new->uid = current->euid;
277 new->gid = new->cgid = current->egid;
278
279 new->seq = ids->seq++;
280 if(ids->seq > ids->seq_max)
281 ids->seq = 0;
282
Pierre Peiffer48dea402008-04-29 01:00:35 -0700283 new->id = ipc_buildid(id, new->seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 spin_lock_init(&new->lock);
285 new->deleted = 0;
286 rcu_read_lock();
287 spin_lock(&new->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return id;
289}
290
291/**
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700292 * ipcget_new - create a new ipc object
293 * @ns: namespace
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700294 * @ids: IPC identifer set
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700295 * @ops: the actual creation routine to call
296 * @params: its parameters
297 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700298 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
299 * when the key is IPC_PRIVATE.
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700300 */
Pavel Emelyanovb2d75cd2008-02-08 04:18:54 -0800301static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700302 struct ipc_ops *ops, struct ipc_params *params)
303{
304 int err;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700305retry:
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700306 err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL);
307
308 if (!err)
309 return -ENOMEM;
310
Nadia Derbey3e148c72007-10-18 23:40:54 -0700311 down_write(&ids->rw_mutex);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700312 err = ops->getnew(ns, params);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700313 up_write(&ids->rw_mutex);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700314
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700315 if (err == -EAGAIN)
316 goto retry;
317
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700318 return err;
319}
320
321/**
322 * ipc_check_perms - check security and permissions for an IPC
323 * @ipcp: ipc permission set
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700324 * @ops: the actual security routine to call
325 * @params: its parameters
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700326 *
327 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
328 * when the key is not IPC_PRIVATE and that key already exists in the
329 * ids IDR.
330 *
331 * On success, the IPC id is returned.
332 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700333 * It is called with ipc_ids.rw_mutex and ipcp->lock held.
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700334 */
335static int ipc_check_perms(struct kern_ipc_perm *ipcp, struct ipc_ops *ops,
336 struct ipc_params *params)
337{
338 int err;
339
340 if (ipcperms(ipcp, params->flg))
341 err = -EACCES;
342 else {
343 err = ops->associate(ipcp, params->flg);
344 if (!err)
345 err = ipcp->id;
346 }
347
348 return err;
349}
350
351/**
352 * ipcget_public - get an ipc object or create a new one
353 * @ns: namespace
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700354 * @ids: IPC identifer set
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700355 * @ops: the actual creation routine to call
356 * @params: its parameters
357 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700358 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
359 * when the key is not IPC_PRIVATE.
360 * It adds a new entry if the key is not found and does some permission
361 * / security checkings if the key is found.
362 *
363 * On success, the ipc id is returned.
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700364 */
Pavel Emelyanovb2d75cd2008-02-08 04:18:54 -0800365static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700366 struct ipc_ops *ops, struct ipc_params *params)
367{
368 struct kern_ipc_perm *ipcp;
369 int flg = params->flg;
370 int err;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700371retry:
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700372 err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL);
373
Nadia Derbey3e148c72007-10-18 23:40:54 -0700374 /*
375 * Take the lock as a writer since we are potentially going to add
376 * a new entry + read locks are not "upgradable"
377 */
378 down_write(&ids->rw_mutex);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700379 ipcp = ipc_findkey(ids, params->key);
380 if (ipcp == NULL) {
381 /* key not used */
382 if (!(flg & IPC_CREAT))
383 err = -ENOENT;
384 else if (!err)
385 err = -ENOMEM;
386 else
387 err = ops->getnew(ns, params);
388 } else {
389 /* ipc object has been locked by ipc_findkey() */
390
391 if (flg & IPC_CREAT && flg & IPC_EXCL)
392 err = -EEXIST;
393 else {
394 err = 0;
395 if (ops->more_checks)
396 err = ops->more_checks(ipcp, params);
397 if (!err)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700398 /*
399 * ipc_check_perms returns the IPC id on
400 * success
401 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700402 err = ipc_check_perms(ipcp, ops, params);
403 }
404 ipc_unlock(ipcp);
405 }
Nadia Derbey3e148c72007-10-18 23:40:54 -0700406 up_write(&ids->rw_mutex);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700407
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700408 if (err == -EAGAIN)
409 goto retry;
410
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700411 return err;
412}
413
414
415/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 * ipc_rmid - remove an IPC identifier
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700417 * @ids: IPC identifier set
418 * @ipcp: ipc perm structure containing the identifier to remove
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700420 * ipc_ids.rw_mutex (as a writer) and the spinlock for this ID are held
421 * before this function is called, and remain locked on the exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 */
423
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700424void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
Nadia Derbeyce621f52007-10-18 23:40:52 -0700426 int lid = ipcid_to_idx(ipcp->id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700428 idr_remove(&ids->ipcs_idr, lid);
429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 ids->in_use--;
431
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700432 ipcp->deleted = 1;
433
434 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
437/**
438 * ipc_alloc - allocate ipc space
439 * @size: size desired
440 *
441 * Allocate memory from the appropriate pools and return a pointer to it.
442 * NULL is returned if the allocation fails
443 */
444
445void* ipc_alloc(int size)
446{
447 void* out;
448 if(size > PAGE_SIZE)
449 out = vmalloc(size);
450 else
451 out = kmalloc(size, GFP_KERNEL);
452 return out;
453}
454
455/**
456 * ipc_free - free ipc space
457 * @ptr: pointer returned by ipc_alloc
458 * @size: size of block
459 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800460 * Free a block created with ipc_alloc(). The caller must know the size
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * used in the allocation call.
462 */
463
464void ipc_free(void* ptr, int size)
465{
466 if(size > PAGE_SIZE)
467 vfree(ptr);
468 else
469 kfree(ptr);
470}
471
472/*
473 * rcu allocations:
474 * There are three headers that are prepended to the actual allocation:
475 * - during use: ipc_rcu_hdr.
476 * - during the rcu grace period: ipc_rcu_grace.
477 * - [only if vmalloc]: ipc_rcu_sched.
478 * Their lifetime doesn't overlap, thus the headers share the same memory.
479 * Unlike a normal union, they are right-aligned, thus some container_of
480 * forward/backward casting is necessary:
481 */
482struct ipc_rcu_hdr
483{
484 int refcount;
485 int is_vmalloc;
486 void *data[0];
487};
488
489
490struct ipc_rcu_grace
491{
492 struct rcu_head rcu;
493 /* "void *" makes sure alignment of following data is sane. */
494 void *data[0];
495};
496
497struct ipc_rcu_sched
498{
499 struct work_struct work;
500 /* "void *" makes sure alignment of following data is sane. */
501 void *data[0];
502};
503
504#define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
505 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
506#define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
507 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
508
509static inline int rcu_use_vmalloc(int size)
510{
511 /* Too big for a single page? */
512 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
513 return 1;
514 return 0;
515}
516
517/**
518 * ipc_rcu_alloc - allocate ipc and rcu space
519 * @size: size desired
520 *
521 * Allocate memory for the rcu header structure + the object.
522 * Returns the pointer to the object.
523 * NULL is returned if the allocation fails.
524 */
525
526void* ipc_rcu_alloc(int size)
527{
528 void* out;
529 /*
530 * We prepend the allocation with the rcu struct, and
531 * workqueue if necessary (for vmalloc).
532 */
533 if (rcu_use_vmalloc(size)) {
534 out = vmalloc(HDRLEN_VMALLOC + size);
535 if (out) {
536 out += HDRLEN_VMALLOC;
537 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
538 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
539 }
540 } else {
541 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
542 if (out) {
543 out += HDRLEN_KMALLOC;
544 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
545 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
546 }
547 }
548
549 return out;
550}
551
552void ipc_rcu_getref(void *ptr)
553{
554 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
555}
556
David Howells65f27f32006-11-22 14:55:48 +0000557static void ipc_do_vfree(struct work_struct *work)
558{
559 vfree(container_of(work, struct ipc_rcu_sched, work));
560}
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/**
Randy Dunlap1e5d5332005-11-07 01:01:06 -0800563 * ipc_schedule_free - free ipc + rcu space
564 * @head: RCU callback structure for queued work
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 *
566 * Since RCU callback function is called in bh,
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800567 * we need to defer the vfree to schedule_work().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 */
569static void ipc_schedule_free(struct rcu_head *head)
570{
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700571 struct ipc_rcu_grace *grace;
572 struct ipc_rcu_sched *sched;
573
574 grace = container_of(head, struct ipc_rcu_grace, rcu);
575 sched = container_of(&(grace->data[0]), struct ipc_rcu_sched,
576 data[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
David Howells65f27f32006-11-22 14:55:48 +0000578 INIT_WORK(&sched->work, ipc_do_vfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 schedule_work(&sched->work);
580}
581
582/**
Randy Dunlap1e5d5332005-11-07 01:01:06 -0800583 * ipc_immediate_free - free ipc + rcu space
584 * @head: RCU callback structure that contains pointer to be freed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800586 * Free from the RCU callback context.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
588static void ipc_immediate_free(struct rcu_head *head)
589{
590 struct ipc_rcu_grace *free =
591 container_of(head, struct ipc_rcu_grace, rcu);
592 kfree(free);
593}
594
595void ipc_rcu_putref(void *ptr)
596{
597 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
598 return;
599
600 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
601 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
602 ipc_schedule_free);
603 } else {
604 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
605 ipc_immediate_free);
606 }
607}
608
609/**
610 * ipcperms - check IPC permissions
611 * @ipcp: IPC permission set
612 * @flag: desired permission set.
613 *
614 * Check user, group, other permissions for access
615 * to ipc resources. return 0 if allowed
616 */
617
618int ipcperms (struct kern_ipc_perm *ipcp, short flag)
619{ /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
Steve Grubb073115d2006-04-02 17:07:33 -0400620 int requested_mode, granted_mode, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Steve Grubb073115d2006-04-02 17:07:33 -0400622 if (unlikely((err = audit_ipc_obj(ipcp))))
623 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 requested_mode = (flag >> 6) | (flag >> 3) | flag;
625 granted_mode = ipcp->mode;
626 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
627 granted_mode >>= 6;
628 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
629 granted_mode >>= 3;
630 /* is there some bit set in requested_mode but not in granted_mode? */
631 if ((requested_mode & ~granted_mode & 0007) &&
632 !capable(CAP_IPC_OWNER))
633 return -1;
634
635 return security_ipc_permission(ipcp, flag);
636}
637
638/*
639 * Functions to convert between the kern_ipc_perm structure and the
640 * old/new ipc_perm structures
641 */
642
643/**
644 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
645 * @in: kernel permissions
646 * @out: new style IPC permissions
647 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800648 * Turn the kernel object @in into a set of permissions descriptions
649 * for returning to userspace (@out).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 */
651
652
653void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
654{
655 out->key = in->key;
656 out->uid = in->uid;
657 out->gid = in->gid;
658 out->cuid = in->cuid;
659 out->cgid = in->cgid;
660 out->mode = in->mode;
661 out->seq = in->seq;
662}
663
664/**
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700665 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 * @in: new style IPC permissions
667 * @out: old style IPC permissions
668 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800669 * Turn the new style permissions object @in into a compatibility
670 * object and store it into the @out pointer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 */
672
673void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
674{
675 out->key = in->key;
676 SET_UID(out->uid, in->uid);
677 SET_GID(out->gid, in->gid);
678 SET_UID(out->cuid, in->cuid);
679 SET_GID(out->cgid, in->cgid);
680 out->mode = in->mode;
681 out->seq = in->seq;
682}
683
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700684/**
Nadia Derbey3e148c72007-10-18 23:40:54 -0700685 * ipc_lock - Lock an ipc structure without rw_mutex held
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700686 * @ids: IPC identifier set
687 * @id: ipc id to look for
688 *
689 * Look for an id in the ipc ids idr and lock the associated ipc object.
690 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700691 * The ipc object is locked on exit.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700692 *
693 * This is the routine that should be called when the rw_mutex is not already
694 * held, i.e. idr tree not protected: it protects the idr tree in read mode
695 * during the idr_find().
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700696 */
697
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700698struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700700 struct kern_ipc_perm *out;
Nadia Derbeyce621f52007-10-18 23:40:52 -0700701 int lid = ipcid_to_idx(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Nadia Derbey3e148c72007-10-18 23:40:54 -0700703 down_read(&ids->rw_mutex);
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 rcu_read_lock();
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700706 out = idr_find(&ids->ipcs_idr, lid);
707 if (out == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 rcu_read_unlock();
Nadia Derbey3e148c72007-10-18 23:40:54 -0700709 up_read(&ids->rw_mutex);
Nadia Derbey023a5352007-10-18 23:40:51 -0700710 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700712
Nadia Derbey3e148c72007-10-18 23:40:54 -0700713 up_read(&ids->rw_mutex);
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 spin_lock(&out->lock);
716
717 /* ipc_rmid() may have already freed the ID while ipc_lock
718 * was spinning: here verify that the structure is still valid
719 */
720 if (out->deleted) {
721 spin_unlock(&out->lock);
722 rcu_read_unlock();
Nadia Derbey023a5352007-10-18 23:40:51 -0700723 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 return out;
727}
728
Nadia Derbey3e148c72007-10-18 23:40:54 -0700729/**
730 * ipc_lock_down - Lock an ipc structure with rw_sem held
731 * @ids: IPC identifier set
732 * @id: ipc id to look for
733 *
734 * Look for an id in the ipc ids idr and lock the associated ipc object.
735 *
736 * The ipc object is locked on exit.
737 *
738 * This is the routine that should be called when the rw_mutex is already
739 * held, i.e. idr tree protected.
740 */
741
742struct kern_ipc_perm *ipc_lock_down(struct ipc_ids *ids, int id)
743{
744 struct kern_ipc_perm *out;
745 int lid = ipcid_to_idx(id);
746
747 rcu_read_lock();
748 out = idr_find(&ids->ipcs_idr, lid);
749 if (out == NULL) {
750 rcu_read_unlock();
751 return ERR_PTR(-EINVAL);
752 }
753
754 spin_lock(&out->lock);
755
756 /*
757 * No need to verify that the structure is still valid since the
758 * rw_mutex is held.
759 */
760 return out;
761}
762
Pavel Emelyanovb2d75cd2008-02-08 04:18:54 -0800763struct kern_ipc_perm *ipc_lock_check_down(struct ipc_ids *ids, int id)
764{
765 struct kern_ipc_perm *out;
766
767 out = ipc_lock_down(ids, id);
768 if (IS_ERR(out))
769 return out;
770
771 if (ipc_checkid(out, id)) {
772 ipc_unlock(out);
773 return ERR_PTR(-EIDRM);
774 }
775
776 return out;
777}
778
779struct kern_ipc_perm *ipc_lock_check(struct ipc_ids *ids, int id)
780{
781 struct kern_ipc_perm *out;
782
783 out = ipc_lock(ids, id);
784 if (IS_ERR(out))
785 return out;
786
787 if (ipc_checkid(out, id)) {
788 ipc_unlock(out);
789 return ERR_PTR(-EIDRM);
790 }
791
792 return out;
793}
794
795/**
796 * ipcget - Common sys_*get() code
797 * @ns : namsepace
798 * @ids : IPC identifier set
799 * @ops : operations to be called on ipc object creation, permission checks
800 * and further checks
801 * @params : the parameters needed by the previous operations.
802 *
803 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
804 */
805int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
806 struct ipc_ops *ops, struct ipc_params *params)
807{
808 if (params->key == IPC_PRIVATE)
809 return ipcget_new(ns, ids, ops, params);
810 else
811 return ipcget_public(ns, ids, ops, params);
812}
813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814#ifdef __ARCH_WANT_IPC_PARSE_VERSION
815
816
817/**
818 * ipc_parse_version - IPC call version
819 * @cmd: pointer to command
820 *
821 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800822 * The @cmd value is turned from an encoding command and version into
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 * just the command code.
824 */
825
826int ipc_parse_version (int *cmd)
827{
828 if (*cmd & IPC_64) {
829 *cmd ^= IPC_64;
830 return IPC_64;
831 } else {
832 return IPC_OLD;
833 }
834}
835
836#endif /* __ARCH_WANT_IPC_PARSE_VERSION */
Mike Waychisonae781772005-09-06 15:17:09 -0700837
838#ifdef CONFIG_PROC_FS
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800839struct ipc_proc_iter {
840 struct ipc_namespace *ns;
841 struct ipc_proc_iface *iface;
842};
843
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700844/*
845 * This routine locks the ipc structure found at least at position pos.
846 */
Adrian Bunkb524b9a2008-02-06 01:36:28 -0800847static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
848 loff_t *new_pos)
Mike Waychisonae781772005-09-06 15:17:09 -0700849{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700850 struct kern_ipc_perm *ipc;
851 int total, id;
Kirill Korotaev73ea4132006-10-02 02:18:20 -0700852
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700853 total = 0;
854 for (id = 0; id < pos && total < ids->in_use; id++) {
855 ipc = idr_find(&ids->ipcs_idr, id);
856 if (ipc != NULL)
857 total++;
858 }
Mike Waychisonae781772005-09-06 15:17:09 -0700859
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700860 if (total >= ids->in_use)
861 return NULL;
Mike Waychisonae781772005-09-06 15:17:09 -0700862
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700863 for ( ; pos < IPCMNI; pos++) {
864 ipc = idr_find(&ids->ipcs_idr, pos);
865 if (ipc != NULL) {
866 *new_pos = pos + 1;
867 ipc_lock_by_ptr(ipc);
Mike Waychisonae781772005-09-06 15:17:09 -0700868 return ipc;
869 }
870 }
871
872 /* Out of range - return NULL to terminate iteration */
873 return NULL;
874}
875
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700876static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
877{
878 struct ipc_proc_iter *iter = s->private;
879 struct ipc_proc_iface *iface = iter->iface;
880 struct kern_ipc_perm *ipc = it;
881
882 /* If we had an ipc id locked before, unlock it */
883 if (ipc && ipc != SEQ_START_TOKEN)
884 ipc_unlock(ipc);
885
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800886 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700887}
888
Mike Waychisonae781772005-09-06 15:17:09 -0700889/*
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700890 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
891 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
Mike Waychisonae781772005-09-06 15:17:09 -0700892 */
893static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
894{
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800895 struct ipc_proc_iter *iter = s->private;
896 struct ipc_proc_iface *iface = iter->iface;
Kirill Korotaev73ea4132006-10-02 02:18:20 -0700897 struct ipc_ids *ids;
898
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800899 ids = &iter->ns->ids[iface->ids];
Mike Waychisonae781772005-09-06 15:17:09 -0700900
901 /*
902 * Take the lock - this will be released by the corresponding
903 * call to stop().
904 */
Nadia Derbey3e148c72007-10-18 23:40:54 -0700905 down_read(&ids->rw_mutex);
Mike Waychisonae781772005-09-06 15:17:09 -0700906
907 /* pos < 0 is invalid */
908 if (*pos < 0)
909 return NULL;
910
911 /* pos == 0 means header */
912 if (*pos == 0)
913 return SEQ_START_TOKEN;
914
915 /* Find the (pos-1)th ipc */
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700916 return sysvipc_find_ipc(ids, *pos - 1, pos);
Mike Waychisonae781772005-09-06 15:17:09 -0700917}
918
919static void sysvipc_proc_stop(struct seq_file *s, void *it)
920{
921 struct kern_ipc_perm *ipc = it;
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800922 struct ipc_proc_iter *iter = s->private;
923 struct ipc_proc_iface *iface = iter->iface;
Kirill Korotaev73ea4132006-10-02 02:18:20 -0700924 struct ipc_ids *ids;
Mike Waychisonae781772005-09-06 15:17:09 -0700925
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700926 /* If we had a locked structure, release it */
Mike Waychisonae781772005-09-06 15:17:09 -0700927 if (ipc && ipc != SEQ_START_TOKEN)
928 ipc_unlock(ipc);
929
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800930 ids = &iter->ns->ids[iface->ids];
Mike Waychisonae781772005-09-06 15:17:09 -0700931 /* Release the lock we took in start() */
Nadia Derbey3e148c72007-10-18 23:40:54 -0700932 up_read(&ids->rw_mutex);
Mike Waychisonae781772005-09-06 15:17:09 -0700933}
934
935static int sysvipc_proc_show(struct seq_file *s, void *it)
936{
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800937 struct ipc_proc_iter *iter = s->private;
938 struct ipc_proc_iface *iface = iter->iface;
Mike Waychisonae781772005-09-06 15:17:09 -0700939
940 if (it == SEQ_START_TOKEN)
941 return seq_puts(s, iface->header);
942
943 return iface->show(s, it);
944}
945
946static struct seq_operations sysvipc_proc_seqops = {
947 .start = sysvipc_proc_start,
948 .stop = sysvipc_proc_stop,
949 .next = sysvipc_proc_next,
950 .show = sysvipc_proc_show,
951};
952
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800953static int sysvipc_proc_open(struct inode *inode, struct file *file)
954{
Mike Waychisonae781772005-09-06 15:17:09 -0700955 int ret;
956 struct seq_file *seq;
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800957 struct ipc_proc_iter *iter;
958
959 ret = -ENOMEM;
960 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
961 if (!iter)
962 goto out;
Mike Waychisonae781772005-09-06 15:17:09 -0700963
964 ret = seq_open(file, &sysvipc_proc_seqops);
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800965 if (ret)
966 goto out_kfree;
967
968 seq = file->private_data;
969 seq->private = iter;
970
971 iter->iface = PDE(inode)->data;
972 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
973out:
Mike Waychisonae781772005-09-06 15:17:09 -0700974 return ret;
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800975out_kfree:
976 kfree(iter);
977 goto out;
978}
979
980static int sysvipc_proc_release(struct inode *inode, struct file *file)
981{
982 struct seq_file *seq = file->private_data;
983 struct ipc_proc_iter *iter = seq->private;
984 put_ipc_ns(iter->ns);
985 return seq_release_private(inode, file);
Mike Waychisonae781772005-09-06 15:17:09 -0700986}
987
Arjan van de Ven9a321442007-02-12 00:55:35 -0800988static const struct file_operations sysvipc_proc_fops = {
Mike Waychisonae781772005-09-06 15:17:09 -0700989 .open = sysvipc_proc_open,
990 .read = seq_read,
991 .llseek = seq_lseek,
Eric W. Biedermanbc1fc6d2007-02-12 00:52:10 -0800992 .release = sysvipc_proc_release,
Mike Waychisonae781772005-09-06 15:17:09 -0700993};
994#endif /* CONFIG_PROC_FS */