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
);
150 * ipc_init - initialise IPC subsystem
152 * The various system5 IPC resources (semaphores, messages and shared
153 * memory) are initialised
156 static int __init
ipc_init(void)
163 __initcall(ipc_init
);
166 * ipc_init_ids - initialise IPC identifiers
167 * @ids: Identifier set
168 * @size: Number of identifiers
170 * Given a size for the ipc identifier range (limited below IPCMNI)
171 * set up the sequence range to use then allocate and initialise the
175 void __ipc_init
ipc_init_ids(struct ipc_ids
* ids
, int size
)
179 mutex_init(&ids
->mutex
);
187 int seq_limit
= INT_MAX
/SEQ_MULTIPLIER
;
188 if(seq_limit
> USHRT_MAX
)
189 ids
->seq_max
= USHRT_MAX
;
191 ids
->seq_max
= seq_limit
;
194 ids
->entries
= ipc_rcu_alloc(sizeof(struct kern_ipc_perm
*)*size
+
195 sizeof(struct ipc_id_ary
));
197 if(ids
->entries
== NULL
) {
198 printk(KERN_ERR
"ipc_init_ids() failed, ipc service disabled.\n");
200 ids
->entries
= &ids
->nullentry
;
202 ids
->entries
->size
= size
;
204 ids
->entries
->p
[i
] = NULL
;
207 #ifdef CONFIG_PROC_FS
208 static const struct file_operations sysvipc_proc_fops
;
210 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
211 * @path: Path in procfs
212 * @header: Banner to be printed at the beginning of the file.
213 * @ids: ipc id table to iterate.
214 * @show: show routine.
216 void __init
ipc_init_proc_interface(const char *path
, const char *header
,
217 int ids
, int (*show
)(struct seq_file
*, void *))
219 struct proc_dir_entry
*pde
;
220 struct ipc_proc_iface
*iface
;
222 iface
= kmalloc(sizeof(*iface
), GFP_KERNEL
);
226 iface
->header
= header
;
230 pde
= create_proc_entry(path
,
231 S_IRUGO
, /* world readable */
232 NULL
/* parent dir */);
235 pde
->proc_fops
= &sysvipc_proc_fops
;
243 * ipc_findkey - find a key in an ipc identifier set
244 * @ids: Identifier set
245 * @key: The key to find
247 * Requires ipc_ids.mutex locked.
248 * Returns the identifier if found or -1 if not.
251 int ipc_findkey(struct ipc_ids
* ids
, key_t key
)
254 struct kern_ipc_perm
* p
;
255 int max_id
= ids
->max_id
;
258 * rcu_dereference() is not needed here
259 * since ipc_ids.mutex is held
261 for (id
= 0; id
<= max_id
; id
++) {
262 p
= ids
->entries
->p
[id
];
272 * Requires ipc_ids.mutex locked
274 static int grow_ary(struct ipc_ids
* ids
, int newsize
)
276 struct ipc_id_ary
* new;
277 struct ipc_id_ary
* old
;
279 int size
= ids
->entries
->size
;
286 new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm
*)*newsize
+
287 sizeof(struct ipc_id_ary
));
291 memcpy(new->p
, ids
->entries
->p
, sizeof(struct kern_ipc_perm
*)*size
);
292 for(i
=size
;i
<newsize
;i
++) {
298 * Use rcu_assign_pointer() to make sure the memcpyed contents
299 * of the new array are visible before the new array becomes visible.
301 rcu_assign_pointer(ids
->entries
, new);
303 __ipc_fini_ids(ids
, old
);
308 * ipc_addid - add an IPC identifier
309 * @ids: IPC identifier set
310 * @new: new IPC permission set
311 * @size: new size limit for the id array
313 * Add an entry 'new' to the IPC arrays. The permissions object is
314 * initialised and the first free entry is set up and the id assigned
315 * is returned. The list is returned in a locked state on success.
316 * On failure the list is not locked and -1 is returned.
318 * Called with ipc_ids.mutex held.
321 int ipc_addid(struct ipc_ids
* ids
, struct kern_ipc_perm
* new, int size
)
325 size
= grow_ary(ids
,size
);
328 * rcu_dereference()() is not needed here since
329 * ipc_ids.mutex is held
331 for (id
= 0; id
< size
; id
++) {
332 if(ids
->entries
->p
[id
] == NULL
)
338 if (id
> ids
->max_id
)
341 new->cuid
= new->uid
= current
->euid
;
342 new->gid
= new->cgid
= current
->egid
;
344 new->seq
= ids
->seq
++;
345 if(ids
->seq
> ids
->seq_max
)
348 spin_lock_init(&new->lock
);
351 spin_lock(&new->lock
);
352 ids
->entries
->p
[id
] = new;
357 * ipc_rmid - remove an IPC identifier
358 * @ids: identifier set
359 * @id: Identifier to remove
361 * The identifier must be valid, and in use. The kernel will panic if
362 * fed an invalid identifier. The entry is removed and internal
363 * variables recomputed. The object associated with the identifier
365 * ipc_ids.mutex and the spinlock for this ID is hold before this function
366 * is called, and remain locked on the exit.
369 struct kern_ipc_perm
* ipc_rmid(struct ipc_ids
* ids
, int id
)
371 struct kern_ipc_perm
* p
;
372 int lid
= id
% SEQ_MULTIPLIER
;
373 BUG_ON(lid
>= ids
->entries
->size
);
376 * do not need a rcu_dereference()() here to force ordering
377 * on Alpha, since the ipc_ids.mutex is held.
379 p
= ids
->entries
->p
[lid
];
380 ids
->entries
->p
[lid
] = NULL
;
384 if (lid
== ids
->max_id
) {
389 } while (ids
->entries
->p
[lid
] == NULL
);
397 * ipc_alloc - allocate ipc space
398 * @size: size desired
400 * Allocate memory from the appropriate pools and return a pointer to it.
401 * NULL is returned if the allocation fails
404 void* ipc_alloc(int size
)
410 out
= kmalloc(size
, GFP_KERNEL
);
415 * ipc_free - free ipc space
416 * @ptr: pointer returned by ipc_alloc
417 * @size: size of block
419 * Free a block created with ipc_alloc(). The caller must know the size
420 * used in the allocation call.
423 void ipc_free(void* ptr
, int size
)
433 * There are three headers that are prepended to the actual allocation:
434 * - during use: ipc_rcu_hdr.
435 * - during the rcu grace period: ipc_rcu_grace.
436 * - [only if vmalloc]: ipc_rcu_sched.
437 * Their lifetime doesn't overlap, thus the headers share the same memory.
438 * Unlike a normal union, they are right-aligned, thus some container_of
439 * forward/backward casting is necessary:
452 /* "void *" makes sure alignment of following data is sane. */
458 struct work_struct work
;
459 /* "void *" makes sure alignment of following data is sane. */
463 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
464 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
465 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
466 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
468 static inline int rcu_use_vmalloc(int size
)
470 /* Too big for a single page? */
471 if (HDRLEN_KMALLOC
+ size
> PAGE_SIZE
)
477 * ipc_rcu_alloc - allocate ipc and rcu space
478 * @size: size desired
480 * Allocate memory for the rcu header structure + the object.
481 * Returns the pointer to the object.
482 * NULL is returned if the allocation fails.
485 void* ipc_rcu_alloc(int size
)
489 * We prepend the allocation with the rcu struct, and
490 * workqueue if necessary (for vmalloc).
492 if (rcu_use_vmalloc(size
)) {
493 out
= vmalloc(HDRLEN_VMALLOC
+ size
);
495 out
+= HDRLEN_VMALLOC
;
496 container_of(out
, struct ipc_rcu_hdr
, data
)->is_vmalloc
= 1;
497 container_of(out
, struct ipc_rcu_hdr
, data
)->refcount
= 1;
500 out
= kmalloc(HDRLEN_KMALLOC
+ size
, GFP_KERNEL
);
502 out
+= HDRLEN_KMALLOC
;
503 container_of(out
, struct ipc_rcu_hdr
, data
)->is_vmalloc
= 0;
504 container_of(out
, struct ipc_rcu_hdr
, data
)->refcount
= 1;
511 void ipc_rcu_getref(void *ptr
)
513 container_of(ptr
, struct ipc_rcu_hdr
, data
)->refcount
++;
516 static void ipc_do_vfree(struct work_struct
*work
)
518 vfree(container_of(work
, struct ipc_rcu_sched
, work
));
522 * ipc_schedule_free - free ipc + rcu space
523 * @head: RCU callback structure for queued work
525 * Since RCU callback function is called in bh,
526 * we need to defer the vfree to schedule_work().
528 static void ipc_schedule_free(struct rcu_head
*head
)
530 struct ipc_rcu_grace
*grace
=
531 container_of(head
, struct ipc_rcu_grace
, rcu
);
532 struct ipc_rcu_sched
*sched
=
533 container_of(&(grace
->data
[0]), struct ipc_rcu_sched
, data
[0]);
535 INIT_WORK(&sched
->work
, ipc_do_vfree
);
536 schedule_work(&sched
->work
);
540 * ipc_immediate_free - free ipc + rcu space
541 * @head: RCU callback structure that contains pointer to be freed
543 * Free from the RCU callback context.
545 static void ipc_immediate_free(struct rcu_head
*head
)
547 struct ipc_rcu_grace
*free
=
548 container_of(head
, struct ipc_rcu_grace
, rcu
);
552 void ipc_rcu_putref(void *ptr
)
554 if (--container_of(ptr
, struct ipc_rcu_hdr
, data
)->refcount
> 0)
557 if (container_of(ptr
, struct ipc_rcu_hdr
, data
)->is_vmalloc
) {
558 call_rcu(&container_of(ptr
, struct ipc_rcu_grace
, data
)->rcu
,
561 call_rcu(&container_of(ptr
, struct ipc_rcu_grace
, data
)->rcu
,
567 * ipcperms - check IPC permissions
568 * @ipcp: IPC permission set
569 * @flag: desired permission set.
571 * Check user, group, other permissions for access
572 * to ipc resources. return 0 if allowed
575 int ipcperms (struct kern_ipc_perm
*ipcp
, short flag
)
576 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
577 int requested_mode
, granted_mode
, err
;
579 if (unlikely((err
= audit_ipc_obj(ipcp
))))
581 requested_mode
= (flag
>> 6) | (flag
>> 3) | flag
;
582 granted_mode
= ipcp
->mode
;
583 if (current
->euid
== ipcp
->cuid
|| current
->euid
== ipcp
->uid
)
585 else if (in_group_p(ipcp
->cgid
) || in_group_p(ipcp
->gid
))
587 /* is there some bit set in requested_mode but not in granted_mode? */
588 if ((requested_mode
& ~granted_mode
& 0007) &&
589 !capable(CAP_IPC_OWNER
))
592 return security_ipc_permission(ipcp
, flag
);
596 * Functions to convert between the kern_ipc_perm structure and the
597 * old/new ipc_perm structures
601 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
602 * @in: kernel permissions
603 * @out: new style IPC permissions
605 * Turn the kernel object @in into a set of permissions descriptions
606 * for returning to userspace (@out).
610 void kernel_to_ipc64_perm (struct kern_ipc_perm
*in
, struct ipc64_perm
*out
)
615 out
->cuid
= in
->cuid
;
616 out
->cgid
= in
->cgid
;
617 out
->mode
= in
->mode
;
622 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
623 * @in: new style IPC permissions
624 * @out: old style IPC permissions
626 * Turn the new style permissions object @in into a compatibility
627 * object and store it into the @out pointer.
630 void ipc64_perm_to_ipc_perm (struct ipc64_perm
*in
, struct ipc_perm
*out
)
633 SET_UID(out
->uid
, in
->uid
);
634 SET_GID(out
->gid
, in
->gid
);
635 SET_UID(out
->cuid
, in
->cuid
);
636 SET_GID(out
->cgid
, in
->cgid
);
637 out
->mode
= in
->mode
;
642 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
643 * is called with shm_ids.mutex locked. Since grow_ary() is also called with
644 * shm_ids.mutex down(for Shared Memory), there is no need to add read
645 * barriers here to gurantee the writes in grow_ary() are seen in order
648 * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
649 * if in the future ipc_get() is used by other places without ipc_ids.mutex
650 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
652 struct kern_ipc_perm
* ipc_get(struct ipc_ids
* ids
, int id
)
654 struct kern_ipc_perm
* out
;
655 int lid
= id
% SEQ_MULTIPLIER
;
656 if(lid
>= ids
->entries
->size
)
658 out
= ids
->entries
->p
[lid
];
662 struct kern_ipc_perm
* ipc_lock(struct ipc_ids
* ids
, int id
)
664 struct kern_ipc_perm
* out
;
665 int lid
= id
% SEQ_MULTIPLIER
;
666 struct ipc_id_ary
* entries
;
669 entries
= rcu_dereference(ids
->entries
);
670 if(lid
>= entries
->size
) {
674 out
= entries
->p
[lid
];
679 spin_lock(&out
->lock
);
681 /* ipc_rmid() may have already freed the ID while ipc_lock
682 * was spinning: here verify that the structure is still valid
685 spin_unlock(&out
->lock
);
692 void ipc_lock_by_ptr(struct kern_ipc_perm
*perm
)
695 spin_lock(&perm
->lock
);
698 void ipc_unlock(struct kern_ipc_perm
* perm
)
700 spin_unlock(&perm
->lock
);
704 int ipc_buildid(struct ipc_ids
* ids
, int id
, int seq
)
706 return SEQ_MULTIPLIER
*seq
+ id
;
709 int ipc_checkid(struct ipc_ids
* ids
, struct kern_ipc_perm
* ipcp
, int uid
)
711 if(uid
/SEQ_MULTIPLIER
!= ipcp
->seq
)
716 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
720 * ipc_parse_version - IPC call version
721 * @cmd: pointer to command
723 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
724 * The @cmd value is turned from an encoding command and version into
725 * just the command code.
728 int ipc_parse_version (int *cmd
)
738 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
740 #ifdef CONFIG_PROC_FS
741 struct ipc_proc_iter
{
742 struct ipc_namespace
*ns
;
743 struct ipc_proc_iface
*iface
;
746 static void *sysvipc_proc_next(struct seq_file
*s
, void *it
, loff_t
*pos
)
748 struct ipc_proc_iter
*iter
= s
->private;
749 struct ipc_proc_iface
*iface
= iter
->iface
;
750 struct kern_ipc_perm
*ipc
= it
;
754 ids
= iter
->ns
->ids
[iface
->ids
];
756 /* If we had an ipc id locked before, unlock it */
757 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
761 * p = *pos - 1 (because id 0 starts at position 1)
762 * + 1 (because we increment the position by one)
764 for (p
= *pos
; p
<= ids
->max_id
; p
++) {
765 if ((ipc
= ipc_lock(ids
, p
)) != NULL
) {
771 /* Out of range - return NULL to terminate iteration */
776 * File positions: pos 0 -> header, pos n -> ipc id + 1.
777 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
779 static void *sysvipc_proc_start(struct seq_file
*s
, loff_t
*pos
)
781 struct ipc_proc_iter
*iter
= s
->private;
782 struct ipc_proc_iface
*iface
= iter
->iface
;
783 struct kern_ipc_perm
*ipc
;
787 ids
= iter
->ns
->ids
[iface
->ids
];
790 * Take the lock - this will be released by the corresponding
793 mutex_lock(&ids
->mutex
);
795 /* pos < 0 is invalid */
799 /* pos == 0 means header */
801 return SEQ_START_TOKEN
;
803 /* Find the (pos-1)th ipc */
804 for (p
= *pos
- 1; p
<= ids
->max_id
; p
++) {
805 if ((ipc
= ipc_lock(ids
, p
)) != NULL
) {
813 static void sysvipc_proc_stop(struct seq_file
*s
, void *it
)
815 struct kern_ipc_perm
*ipc
= it
;
816 struct ipc_proc_iter
*iter
= s
->private;
817 struct ipc_proc_iface
*iface
= iter
->iface
;
820 /* If we had a locked segment, release it */
821 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
824 ids
= iter
->ns
->ids
[iface
->ids
];
825 /* Release the lock we took in start() */
826 mutex_unlock(&ids
->mutex
);
829 static int sysvipc_proc_show(struct seq_file
*s
, void *it
)
831 struct ipc_proc_iter
*iter
= s
->private;
832 struct ipc_proc_iface
*iface
= iter
->iface
;
834 if (it
== SEQ_START_TOKEN
)
835 return seq_puts(s
, iface
->header
);
837 return iface
->show(s
, it
);
840 static struct seq_operations sysvipc_proc_seqops
= {
841 .start
= sysvipc_proc_start
,
842 .stop
= sysvipc_proc_stop
,
843 .next
= sysvipc_proc_next
,
844 .show
= sysvipc_proc_show
,
847 static int sysvipc_proc_open(struct inode
*inode
, struct file
*file
)
850 struct seq_file
*seq
;
851 struct ipc_proc_iter
*iter
;
854 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
858 ret
= seq_open(file
, &sysvipc_proc_seqops
);
862 seq
= file
->private_data
;
865 iter
->iface
= PDE(inode
)->data
;
866 iter
->ns
= get_ipc_ns(current
->nsproxy
->ipc_ns
);
874 static int sysvipc_proc_release(struct inode
*inode
, struct file
*file
)
876 struct seq_file
*seq
= file
->private_data
;
877 struct ipc_proc_iter
*iter
= seq
->private;
878 put_ipc_ns(iter
->ns
);
879 return seq_release_private(inode
, file
);
882 static const struct file_operations sysvipc_proc_fops
= {
883 .open
= sysvipc_proc_open
,
886 .release
= sysvipc_proc_release
,
888 #endif /* CONFIG_PROC_FS */