3 * Copyright (C) 1992 Krishna Balasubramanian
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
10 * Manfred Spraul <manfred@colorfullife.com>
11 * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
12 * Mingming Cao <cmm@us.ibm.com>
13 * Mar 2006 - support for audit of ipc object properties
14 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
15 * Jun 2006 - namespaces ssupport
17 * Pavel Emelianov <xemul@openvz.org>
21 #include <linux/shm.h>
22 #include <linux/init.h>
23 #include <linux/msg.h>
24 #include <linux/smp_lock.h>
25 #include <linux/vmalloc.h>
26 #include <linux/slab.h>
27 #include <linux/capability.h>
28 #include <linux/highuid.h>
29 #include <linux/security.h>
30 #include <linux/rcupdate.h>
31 #include <linux/workqueue.h>
32 #include <linux/seq_file.h>
33 #include <linux/proc_fs.h>
34 #include <linux/audit.h>
35 #include <linux/nsproxy.h>
37 #include <asm/unistd.h>
41 struct ipc_proc_iface
{
45 int (*show
)(struct seq_file
*, void *);
48 struct ipc_namespace init_ipc_ns
= {
50 .refcount
= ATOMIC_INIT(2),
55 static struct ipc_namespace
*clone_ipc_ns(struct ipc_namespace
*old_ns
)
58 struct ipc_namespace
*ns
;
61 ns
= kmalloc(sizeof(struct ipc_namespace
), GFP_KERNEL
);
65 err
= sem_init_ns(ns
);
68 err
= msg_init_ns(ns
);
71 err
= shm_init_ns(ns
);
88 int unshare_ipcs(unsigned long unshare_flags
, struct ipc_namespace
**new_ipc
)
90 struct ipc_namespace
*new;
92 if (unshare_flags
& CLONE_NEWIPC
) {
93 if (!capable(CAP_SYS_ADMIN
))
96 new = clone_ipc_ns(current
->nsproxy
->ipc_ns
);
106 int copy_ipcs(unsigned long flags
, struct task_struct
*tsk
)
108 struct ipc_namespace
*old_ns
= tsk
->nsproxy
->ipc_ns
;
109 struct ipc_namespace
*new_ns
;
117 if (!(flags
& CLONE_NEWIPC
))
120 if (!capable(CAP_SYS_ADMIN
)) {
125 new_ns
= clone_ipc_ns(old_ns
);
131 tsk
->nsproxy
->ipc_ns
= new_ns
;
137 void free_ipc_ns(struct kref
*kref
)
139 struct ipc_namespace
*ns
;
141 ns
= container_of(kref
, struct ipc_namespace
, kref
);
148 int copy_ipcs(unsigned long flags
, struct task_struct
*tsk
)
150 if (flags
& CLONE_NEWIPC
)
157 * ipc_init - initialise IPC subsystem
159 * The various system5 IPC resources (semaphores, messages and shared
160 * memory) are initialised
163 static int __init
ipc_init(void)
170 __initcall(ipc_init
);
173 * ipc_init_ids - initialise IPC identifiers
174 * @ids: Identifier set
175 * @size: Number of identifiers
177 * Given a size for the ipc identifier range (limited below IPCMNI)
178 * set up the sequence range to use then allocate and initialise the
182 void __ipc_init
ipc_init_ids(struct ipc_ids
* ids
, int size
)
186 mutex_init(&ids
->mutex
);
194 int seq_limit
= INT_MAX
/SEQ_MULTIPLIER
;
195 if(seq_limit
> USHRT_MAX
)
196 ids
->seq_max
= USHRT_MAX
;
198 ids
->seq_max
= seq_limit
;
201 ids
->entries
= ipc_rcu_alloc(sizeof(struct kern_ipc_perm
*)*size
+
202 sizeof(struct ipc_id_ary
));
204 if(ids
->entries
== NULL
) {
205 printk(KERN_ERR
"ipc_init_ids() failed, ipc service disabled.\n");
207 ids
->entries
= &ids
->nullentry
;
209 ids
->entries
->size
= size
;
211 ids
->entries
->p
[i
] = NULL
;
214 #ifdef CONFIG_PROC_FS
215 static const struct file_operations sysvipc_proc_fops
;
217 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
218 * @path: Path in procfs
219 * @header: Banner to be printed at the beginning of the file.
220 * @ids: ipc id table to iterate.
221 * @show: show routine.
223 void __init
ipc_init_proc_interface(const char *path
, const char *header
,
224 int ids
, int (*show
)(struct seq_file
*, void *))
226 struct proc_dir_entry
*pde
;
227 struct ipc_proc_iface
*iface
;
229 iface
= kmalloc(sizeof(*iface
), GFP_KERNEL
);
233 iface
->header
= header
;
237 pde
= create_proc_entry(path
,
238 S_IRUGO
, /* world readable */
239 NULL
/* parent dir */);
242 pde
->proc_fops
= &sysvipc_proc_fops
;
250 * ipc_findkey - find a key in an ipc identifier set
251 * @ids: Identifier set
252 * @key: The key to find
254 * Requires ipc_ids.mutex locked.
255 * Returns the identifier if found or -1 if not.
258 int ipc_findkey(struct ipc_ids
* ids
, key_t key
)
261 struct kern_ipc_perm
* p
;
262 int max_id
= ids
->max_id
;
265 * rcu_dereference() is not needed here
266 * since ipc_ids.mutex is held
268 for (id
= 0; id
<= max_id
; id
++) {
269 p
= ids
->entries
->p
[id
];
279 * Requires ipc_ids.mutex locked
281 static int grow_ary(struct ipc_ids
* ids
, int newsize
)
283 struct ipc_id_ary
* new;
284 struct ipc_id_ary
* old
;
286 int size
= ids
->entries
->size
;
293 new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm
*)*newsize
+
294 sizeof(struct ipc_id_ary
));
298 memcpy(new->p
, ids
->entries
->p
, sizeof(struct kern_ipc_perm
*)*size
);
299 for(i
=size
;i
<newsize
;i
++) {
305 * Use rcu_assign_pointer() to make sure the memcpyed contents
306 * of the new array are visible before the new array becomes visible.
308 rcu_assign_pointer(ids
->entries
, new);
310 __ipc_fini_ids(ids
, old
);
315 * ipc_addid - add an IPC identifier
316 * @ids: IPC identifier set
317 * @new: new IPC permission set
318 * @size: new size limit for the id array
320 * Add an entry 'new' to the IPC arrays. The permissions object is
321 * initialised and the first free entry is set up and the id assigned
322 * is returned. The list is returned in a locked state on success.
323 * On failure the list is not locked and -1 is returned.
325 * Called with ipc_ids.mutex held.
328 int ipc_addid(struct ipc_ids
* ids
, struct kern_ipc_perm
* new, int size
)
332 size
= grow_ary(ids
,size
);
335 * rcu_dereference()() is not needed here since
336 * ipc_ids.mutex is held
338 for (id
= 0; id
< size
; id
++) {
339 if(ids
->entries
->p
[id
] == NULL
)
345 if (id
> ids
->max_id
)
348 new->cuid
= new->uid
= current
->euid
;
349 new->gid
= new->cgid
= current
->egid
;
351 new->seq
= ids
->seq
++;
352 if(ids
->seq
> ids
->seq_max
)
355 spin_lock_init(&new->lock
);
358 spin_lock(&new->lock
);
359 ids
->entries
->p
[id
] = new;
364 * ipc_rmid - remove an IPC identifier
365 * @ids: identifier set
366 * @id: Identifier to remove
368 * The identifier must be valid, and in use. The kernel will panic if
369 * fed an invalid identifier. The entry is removed and internal
370 * variables recomputed. The object associated with the identifier
372 * ipc_ids.mutex and the spinlock for this ID is hold before this function
373 * is called, and remain locked on the exit.
376 struct kern_ipc_perm
* ipc_rmid(struct ipc_ids
* ids
, int id
)
378 struct kern_ipc_perm
* p
;
379 int lid
= id
% SEQ_MULTIPLIER
;
380 BUG_ON(lid
>= ids
->entries
->size
);
383 * do not need a rcu_dereference()() here to force ordering
384 * on Alpha, since the ipc_ids.mutex is held.
386 p
= ids
->entries
->p
[lid
];
387 ids
->entries
->p
[lid
] = NULL
;
391 if (lid
== ids
->max_id
) {
396 } while (ids
->entries
->p
[lid
] == NULL
);
404 * ipc_alloc - allocate ipc space
405 * @size: size desired
407 * Allocate memory from the appropriate pools and return a pointer to it.
408 * NULL is returned if the allocation fails
411 void* ipc_alloc(int size
)
417 out
= kmalloc(size
, GFP_KERNEL
);
422 * ipc_free - free ipc space
423 * @ptr: pointer returned by ipc_alloc
424 * @size: size of block
426 * Free a block created with ipc_alloc(). The caller must know the size
427 * used in the allocation call.
430 void ipc_free(void* ptr
, int size
)
440 * There are three headers that are prepended to the actual allocation:
441 * - during use: ipc_rcu_hdr.
442 * - during the rcu grace period: ipc_rcu_grace.
443 * - [only if vmalloc]: ipc_rcu_sched.
444 * Their lifetime doesn't overlap, thus the headers share the same memory.
445 * Unlike a normal union, they are right-aligned, thus some container_of
446 * forward/backward casting is necessary:
459 /* "void *" makes sure alignment of following data is sane. */
465 struct work_struct work
;
466 /* "void *" makes sure alignment of following data is sane. */
470 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
471 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
472 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
473 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
475 static inline int rcu_use_vmalloc(int size
)
477 /* Too big for a single page? */
478 if (HDRLEN_KMALLOC
+ size
> PAGE_SIZE
)
484 * ipc_rcu_alloc - allocate ipc and rcu space
485 * @size: size desired
487 * Allocate memory for the rcu header structure + the object.
488 * Returns the pointer to the object.
489 * NULL is returned if the allocation fails.
492 void* ipc_rcu_alloc(int size
)
496 * We prepend the allocation with the rcu struct, and
497 * workqueue if necessary (for vmalloc).
499 if (rcu_use_vmalloc(size
)) {
500 out
= vmalloc(HDRLEN_VMALLOC
+ size
);
502 out
+= HDRLEN_VMALLOC
;
503 container_of(out
, struct ipc_rcu_hdr
, data
)->is_vmalloc
= 1;
504 container_of(out
, struct ipc_rcu_hdr
, data
)->refcount
= 1;
507 out
= kmalloc(HDRLEN_KMALLOC
+ size
, GFP_KERNEL
);
509 out
+= HDRLEN_KMALLOC
;
510 container_of(out
, struct ipc_rcu_hdr
, data
)->is_vmalloc
= 0;
511 container_of(out
, struct ipc_rcu_hdr
, data
)->refcount
= 1;
518 void ipc_rcu_getref(void *ptr
)
520 container_of(ptr
, struct ipc_rcu_hdr
, data
)->refcount
++;
523 static void ipc_do_vfree(struct work_struct
*work
)
525 vfree(container_of(work
, struct ipc_rcu_sched
, work
));
529 * ipc_schedule_free - free ipc + rcu space
530 * @head: RCU callback structure for queued work
532 * Since RCU callback function is called in bh,
533 * we need to defer the vfree to schedule_work().
535 static void ipc_schedule_free(struct rcu_head
*head
)
537 struct ipc_rcu_grace
*grace
=
538 container_of(head
, struct ipc_rcu_grace
, rcu
);
539 struct ipc_rcu_sched
*sched
=
540 container_of(&(grace
->data
[0]), struct ipc_rcu_sched
, data
[0]);
542 INIT_WORK(&sched
->work
, ipc_do_vfree
);
543 schedule_work(&sched
->work
);
547 * ipc_immediate_free - free ipc + rcu space
548 * @head: RCU callback structure that contains pointer to be freed
550 * Free from the RCU callback context.
552 static void ipc_immediate_free(struct rcu_head
*head
)
554 struct ipc_rcu_grace
*free
=
555 container_of(head
, struct ipc_rcu_grace
, rcu
);
559 void ipc_rcu_putref(void *ptr
)
561 if (--container_of(ptr
, struct ipc_rcu_hdr
, data
)->refcount
> 0)
564 if (container_of(ptr
, struct ipc_rcu_hdr
, data
)->is_vmalloc
) {
565 call_rcu(&container_of(ptr
, struct ipc_rcu_grace
, data
)->rcu
,
568 call_rcu(&container_of(ptr
, struct ipc_rcu_grace
, data
)->rcu
,
574 * ipcperms - check IPC permissions
575 * @ipcp: IPC permission set
576 * @flag: desired permission set.
578 * Check user, group, other permissions for access
579 * to ipc resources. return 0 if allowed
582 int ipcperms (struct kern_ipc_perm
*ipcp
, short flag
)
583 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
584 int requested_mode
, granted_mode
, err
;
586 if (unlikely((err
= audit_ipc_obj(ipcp
))))
588 requested_mode
= (flag
>> 6) | (flag
>> 3) | flag
;
589 granted_mode
= ipcp
->mode
;
590 if (current
->euid
== ipcp
->cuid
|| current
->euid
== ipcp
->uid
)
592 else if (in_group_p(ipcp
->cgid
) || in_group_p(ipcp
->gid
))
594 /* is there some bit set in requested_mode but not in granted_mode? */
595 if ((requested_mode
& ~granted_mode
& 0007) &&
596 !capable(CAP_IPC_OWNER
))
599 return security_ipc_permission(ipcp
, flag
);
603 * Functions to convert between the kern_ipc_perm structure and the
604 * old/new ipc_perm structures
608 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
609 * @in: kernel permissions
610 * @out: new style IPC permissions
612 * Turn the kernel object @in into a set of permissions descriptions
613 * for returning to userspace (@out).
617 void kernel_to_ipc64_perm (struct kern_ipc_perm
*in
, struct ipc64_perm
*out
)
622 out
->cuid
= in
->cuid
;
623 out
->cgid
= in
->cgid
;
624 out
->mode
= in
->mode
;
629 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
630 * @in: new style IPC permissions
631 * @out: old style IPC permissions
633 * Turn the new style permissions object @in into a compatibility
634 * object and store it into the @out pointer.
637 void ipc64_perm_to_ipc_perm (struct ipc64_perm
*in
, struct ipc_perm
*out
)
640 SET_UID(out
->uid
, in
->uid
);
641 SET_GID(out
->gid
, in
->gid
);
642 SET_UID(out
->cuid
, in
->cuid
);
643 SET_GID(out
->cgid
, in
->cgid
);
644 out
->mode
= in
->mode
;
649 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
650 * is called with shm_ids.mutex locked. Since grow_ary() is also called with
651 * shm_ids.mutex down(for Shared Memory), there is no need to add read
652 * barriers here to gurantee the writes in grow_ary() are seen in order
655 * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
656 * if in the future ipc_get() is used by other places without ipc_ids.mutex
657 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
659 struct kern_ipc_perm
* ipc_get(struct ipc_ids
* ids
, int id
)
661 struct kern_ipc_perm
* out
;
662 int lid
= id
% SEQ_MULTIPLIER
;
663 if(lid
>= ids
->entries
->size
)
665 out
= ids
->entries
->p
[lid
];
669 struct kern_ipc_perm
* ipc_lock(struct ipc_ids
* ids
, int id
)
671 struct kern_ipc_perm
* out
;
672 int lid
= id
% SEQ_MULTIPLIER
;
673 struct ipc_id_ary
* entries
;
676 entries
= rcu_dereference(ids
->entries
);
677 if(lid
>= entries
->size
) {
681 out
= entries
->p
[lid
];
686 spin_lock(&out
->lock
);
688 /* ipc_rmid() may have already freed the ID while ipc_lock
689 * was spinning: here verify that the structure is still valid
692 spin_unlock(&out
->lock
);
699 void ipc_lock_by_ptr(struct kern_ipc_perm
*perm
)
702 spin_lock(&perm
->lock
);
705 void ipc_unlock(struct kern_ipc_perm
* perm
)
707 spin_unlock(&perm
->lock
);
711 int ipc_buildid(struct ipc_ids
* ids
, int id
, int seq
)
713 return SEQ_MULTIPLIER
*seq
+ id
;
716 int ipc_checkid(struct ipc_ids
* ids
, struct kern_ipc_perm
* ipcp
, int uid
)
718 if(uid
/SEQ_MULTIPLIER
!= ipcp
->seq
)
723 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
727 * ipc_parse_version - IPC call version
728 * @cmd: pointer to command
730 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
731 * The @cmd value is turned from an encoding command and version into
732 * just the command code.
735 int ipc_parse_version (int *cmd
)
745 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
747 #ifdef CONFIG_PROC_FS
748 struct ipc_proc_iter
{
749 struct ipc_namespace
*ns
;
750 struct ipc_proc_iface
*iface
;
753 static void *sysvipc_proc_next(struct seq_file
*s
, void *it
, loff_t
*pos
)
755 struct ipc_proc_iter
*iter
= s
->private;
756 struct ipc_proc_iface
*iface
= iter
->iface
;
757 struct kern_ipc_perm
*ipc
= it
;
761 ids
= iter
->ns
->ids
[iface
->ids
];
763 /* If we had an ipc id locked before, unlock it */
764 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
768 * p = *pos - 1 (because id 0 starts at position 1)
769 * + 1 (because we increment the position by one)
771 for (p
= *pos
; p
<= ids
->max_id
; p
++) {
772 if ((ipc
= ipc_lock(ids
, p
)) != NULL
) {
778 /* Out of range - return NULL to terminate iteration */
783 * File positions: pos 0 -> header, pos n -> ipc id + 1.
784 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
786 static void *sysvipc_proc_start(struct seq_file
*s
, loff_t
*pos
)
788 struct ipc_proc_iter
*iter
= s
->private;
789 struct ipc_proc_iface
*iface
= iter
->iface
;
790 struct kern_ipc_perm
*ipc
;
794 ids
= iter
->ns
->ids
[iface
->ids
];
797 * Take the lock - this will be released by the corresponding
800 mutex_lock(&ids
->mutex
);
802 /* pos < 0 is invalid */
806 /* pos == 0 means header */
808 return SEQ_START_TOKEN
;
810 /* Find the (pos-1)th ipc */
811 for (p
= *pos
- 1; p
<= ids
->max_id
; p
++) {
812 if ((ipc
= ipc_lock(ids
, p
)) != NULL
) {
820 static void sysvipc_proc_stop(struct seq_file
*s
, void *it
)
822 struct kern_ipc_perm
*ipc
= it
;
823 struct ipc_proc_iter
*iter
= s
->private;
824 struct ipc_proc_iface
*iface
= iter
->iface
;
827 /* If we had a locked segment, release it */
828 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
831 ids
= iter
->ns
->ids
[iface
->ids
];
832 /* Release the lock we took in start() */
833 mutex_unlock(&ids
->mutex
);
836 static int sysvipc_proc_show(struct seq_file
*s
, void *it
)
838 struct ipc_proc_iter
*iter
= s
->private;
839 struct ipc_proc_iface
*iface
= iter
->iface
;
841 if (it
== SEQ_START_TOKEN
)
842 return seq_puts(s
, iface
->header
);
844 return iface
->show(s
, it
);
847 static struct seq_operations sysvipc_proc_seqops
= {
848 .start
= sysvipc_proc_start
,
849 .stop
= sysvipc_proc_stop
,
850 .next
= sysvipc_proc_next
,
851 .show
= sysvipc_proc_show
,
854 static int sysvipc_proc_open(struct inode
*inode
, struct file
*file
)
857 struct seq_file
*seq
;
858 struct ipc_proc_iter
*iter
;
861 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
865 ret
= seq_open(file
, &sysvipc_proc_seqops
);
869 seq
= file
->private_data
;
872 iter
->iface
= PDE(inode
)->data
;
873 iter
->ns
= get_ipc_ns(current
->nsproxy
->ipc_ns
);
881 static int sysvipc_proc_release(struct inode
*inode
, struct file
*file
)
883 struct seq_file
*seq
= file
->private_data
;
884 struct ipc_proc_iter
*iter
= seq
->private;
885 put_ipc_ns(iter
->ns
);
886 return seq_release_private(inode
, file
);
889 static const struct file_operations sysvipc_proc_fops
= {
890 .open
= sysvipc_proc_open
,
893 .release
= sysvipc_proc_release
,
895 #endif /* CONFIG_PROC_FS */