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 <manfreds@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>
15 #include <linux/config.h>
17 #include <linux/shm.h>
18 #include <linux/init.h>
19 #include <linux/msg.h>
20 #include <linux/smp_lock.h>
21 #include <linux/vmalloc.h>
22 #include <linux/slab.h>
23 #include <linux/highuid.h>
24 #include <linux/security.h>
25 #include <linux/rcupdate.h>
26 #include <linux/workqueue.h>
27 #include <linux/seq_file.h>
28 #include <linux/proc_fs.h>
30 #include <asm/unistd.h>
34 struct ipc_proc_iface
{
38 int (*show
)(struct seq_file
*, void *);
42 * ipc_init - initialise IPC subsystem
44 * The various system5 IPC resources (semaphores, messages and shared
45 * memory are initialised
48 static int __init
ipc_init(void)
58 * ipc_init_ids - initialise IPC identifiers
59 * @ids: Identifier set
60 * @size: Number of identifiers
62 * Given a size for the ipc identifier range (limited below IPCMNI)
63 * set up the sequence range to use then allocate and initialise the
67 void __init
ipc_init_ids(struct ipc_ids
* ids
, int size
)
70 sema_init(&ids
->sem
,1);
78 int seq_limit
= INT_MAX
/SEQ_MULTIPLIER
;
79 if(seq_limit
> USHRT_MAX
)
80 ids
->seq_max
= USHRT_MAX
;
82 ids
->seq_max
= seq_limit
;
85 ids
->entries
= ipc_rcu_alloc(sizeof(struct kern_ipc_perm
*)*size
+
86 sizeof(struct ipc_id_ary
));
88 if(ids
->entries
== NULL
) {
89 printk(KERN_ERR
"ipc_init_ids() failed, ipc service disabled.\n");
91 ids
->entries
= &ids
->nullentry
;
93 ids
->entries
->size
= size
;
95 ids
->entries
->p
[i
] = NULL
;
99 static struct file_operations sysvipc_proc_fops
;
101 * ipc_init_proc_interface - Create a proc interface for sysipc types
102 * using a seq_file interface.
103 * @path: Path in procfs
104 * @header: Banner to be printed at the beginning of the file.
105 * @ids: ipc id table to iterate.
106 * @show: show routine.
108 void __init
ipc_init_proc_interface(const char *path
, const char *header
,
110 int (*show
)(struct seq_file
*, void *))
112 struct proc_dir_entry
*pde
;
113 struct ipc_proc_iface
*iface
;
115 iface
= kmalloc(sizeof(*iface
), GFP_KERNEL
);
119 iface
->header
= header
;
123 pde
= create_proc_entry(path
,
124 S_IRUGO
, /* world readable */
125 NULL
/* parent dir */);
128 pde
->proc_fops
= &sysvipc_proc_fops
;
136 * ipc_findkey - find a key in an ipc identifier set
137 * @ids: Identifier set
138 * @key: The key to find
140 * Requires ipc_ids.sem locked.
141 * Returns the identifier if found or -1 if not.
144 int ipc_findkey(struct ipc_ids
* ids
, key_t key
)
147 struct kern_ipc_perm
* p
;
148 int max_id
= ids
->max_id
;
151 * rcu_dereference() is not needed here
152 * since ipc_ids.sem is held
154 for (id
= 0; id
<= max_id
; id
++) {
155 p
= ids
->entries
->p
[id
];
165 * Requires ipc_ids.sem locked
167 static int grow_ary(struct ipc_ids
* ids
, int newsize
)
169 struct ipc_id_ary
* new;
170 struct ipc_id_ary
* old
;
172 int size
= ids
->entries
->size
;
179 new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm
*)*newsize
+
180 sizeof(struct ipc_id_ary
));
184 memcpy(new->p
, ids
->entries
->p
, sizeof(struct kern_ipc_perm
*)*size
+
185 sizeof(struct ipc_id_ary
));
186 for(i
=size
;i
<newsize
;i
++) {
192 * Use rcu_assign_pointer() to make sure the memcpyed contents
193 * of the new array are visible before the new array becomes visible.
195 rcu_assign_pointer(ids
->entries
, new);
202 * ipc_addid - add an IPC identifier
203 * @ids: IPC identifier set
204 * @new: new IPC permission set
205 * @size: new size limit for the id array
207 * Add an entry 'new' to the IPC arrays. The permissions object is
208 * initialised and the first free entry is set up and the id assigned
209 * is returned. The list is returned in a locked state on success.
210 * On failure the list is not locked and -1 is returned.
212 * Called with ipc_ids.sem held.
215 int ipc_addid(struct ipc_ids
* ids
, struct kern_ipc_perm
* new, int size
)
219 size
= grow_ary(ids
,size
);
222 * rcu_dereference()() is not needed here since
223 * ipc_ids.sem is held
225 for (id
= 0; id
< size
; id
++) {
226 if(ids
->entries
->p
[id
] == NULL
)
232 if (id
> ids
->max_id
)
235 new->cuid
= new->uid
= current
->euid
;
236 new->gid
= new->cgid
= current
->egid
;
238 new->seq
= ids
->seq
++;
239 if(ids
->seq
> ids
->seq_max
)
242 spin_lock_init(&new->lock
);
245 spin_lock(&new->lock
);
246 ids
->entries
->p
[id
] = new;
251 * ipc_rmid - remove an IPC identifier
252 * @ids: identifier set
253 * @id: Identifier to remove
255 * The identifier must be valid, and in use. The kernel will panic if
256 * fed an invalid identifier. The entry is removed and internal
257 * variables recomputed. The object associated with the identifier
259 * ipc_ids.sem and the spinlock for this ID is hold before this function
260 * is called, and remain locked on the exit.
263 struct kern_ipc_perm
* ipc_rmid(struct ipc_ids
* ids
, int id
)
265 struct kern_ipc_perm
* p
;
266 int lid
= id
% SEQ_MULTIPLIER
;
267 if(lid
>= ids
->entries
->size
)
271 * do not need a rcu_dereference()() here to force ordering
272 * on Alpha, since the ipc_ids.sem is held.
274 p
= ids
->entries
->p
[lid
];
275 ids
->entries
->p
[lid
] = NULL
;
280 if (lid
== ids
->max_id
) {
285 } while (ids
->entries
->p
[lid
] == NULL
);
293 * ipc_alloc - allocate ipc space
294 * @size: size desired
296 * Allocate memory from the appropriate pools and return a pointer to it.
297 * NULL is returned if the allocation fails
300 void* ipc_alloc(int size
)
306 out
= kmalloc(size
, GFP_KERNEL
);
311 * ipc_free - free ipc space
312 * @ptr: pointer returned by ipc_alloc
313 * @size: size of block
315 * Free a block created with ipc_alloc. The caller must know the size
316 * used in the allocation call.
319 void ipc_free(void* ptr
, int size
)
329 * There are three headers that are prepended to the actual allocation:
330 * - during use: ipc_rcu_hdr.
331 * - during the rcu grace period: ipc_rcu_grace.
332 * - [only if vmalloc]: ipc_rcu_sched.
333 * Their lifetime doesn't overlap, thus the headers share the same memory.
334 * Unlike a normal union, they are right-aligned, thus some container_of
335 * forward/backward casting is necessary:
348 /* "void *" makes sure alignment of following data is sane. */
354 struct work_struct work
;
355 /* "void *" makes sure alignment of following data is sane. */
359 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
360 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
361 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
362 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
364 static inline int rcu_use_vmalloc(int size
)
366 /* Too big for a single page? */
367 if (HDRLEN_KMALLOC
+ size
> PAGE_SIZE
)
373 * ipc_rcu_alloc - allocate ipc and rcu space
374 * @size: size desired
376 * Allocate memory for the rcu header structure + the object.
377 * Returns the pointer to the object.
378 * NULL is returned if the allocation fails.
381 void* ipc_rcu_alloc(int size
)
385 * We prepend the allocation with the rcu struct, and
386 * workqueue if necessary (for vmalloc).
388 if (rcu_use_vmalloc(size
)) {
389 out
= vmalloc(HDRLEN_VMALLOC
+ size
);
391 out
+= HDRLEN_VMALLOC
;
392 container_of(out
, struct ipc_rcu_hdr
, data
)->is_vmalloc
= 1;
393 container_of(out
, struct ipc_rcu_hdr
, data
)->refcount
= 1;
396 out
= kmalloc(HDRLEN_KMALLOC
+ size
, GFP_KERNEL
);
398 out
+= HDRLEN_KMALLOC
;
399 container_of(out
, struct ipc_rcu_hdr
, data
)->is_vmalloc
= 0;
400 container_of(out
, struct ipc_rcu_hdr
, data
)->refcount
= 1;
407 void ipc_rcu_getref(void *ptr
)
409 container_of(ptr
, struct ipc_rcu_hdr
, data
)->refcount
++;
413 * ipc_schedule_free - free ipc + rcu space
414 * @head: RCU callback structure for queued work
416 * Since RCU callback function is called in bh,
417 * we need to defer the vfree to schedule_work
419 static void ipc_schedule_free(struct rcu_head
*head
)
421 struct ipc_rcu_grace
*grace
=
422 container_of(head
, struct ipc_rcu_grace
, rcu
);
423 struct ipc_rcu_sched
*sched
=
424 container_of(&(grace
->data
[0]), struct ipc_rcu_sched
, data
[0]);
426 INIT_WORK(&sched
->work
, vfree
, sched
);
427 schedule_work(&sched
->work
);
431 * ipc_immediate_free - free ipc + rcu space
432 * @head: RCU callback structure that contains pointer to be freed
434 * Free from the RCU callback context
436 static void ipc_immediate_free(struct rcu_head
*head
)
438 struct ipc_rcu_grace
*free
=
439 container_of(head
, struct ipc_rcu_grace
, rcu
);
443 void ipc_rcu_putref(void *ptr
)
445 if (--container_of(ptr
, struct ipc_rcu_hdr
, data
)->refcount
> 0)
448 if (container_of(ptr
, struct ipc_rcu_hdr
, data
)->is_vmalloc
) {
449 call_rcu(&container_of(ptr
, struct ipc_rcu_grace
, data
)->rcu
,
452 call_rcu(&container_of(ptr
, struct ipc_rcu_grace
, data
)->rcu
,
458 * ipcperms - check IPC permissions
459 * @ipcp: IPC permission set
460 * @flag: desired permission set.
462 * Check user, group, other permissions for access
463 * to ipc resources. return 0 if allowed
466 int ipcperms (struct kern_ipc_perm
*ipcp
, short flag
)
467 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
468 int requested_mode
, granted_mode
;
470 requested_mode
= (flag
>> 6) | (flag
>> 3) | flag
;
471 granted_mode
= ipcp
->mode
;
472 if (current
->euid
== ipcp
->cuid
|| current
->euid
== ipcp
->uid
)
474 else if (in_group_p(ipcp
->cgid
) || in_group_p(ipcp
->gid
))
476 /* is there some bit set in requested_mode but not in granted_mode? */
477 if ((requested_mode
& ~granted_mode
& 0007) &&
478 !capable(CAP_IPC_OWNER
))
481 return security_ipc_permission(ipcp
, flag
);
485 * Functions to convert between the kern_ipc_perm structure and the
486 * old/new ipc_perm structures
490 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
491 * @in: kernel permissions
492 * @out: new style IPC permissions
494 * Turn the kernel object 'in' into a set of permissions descriptions
495 * for returning to userspace (out).
499 void kernel_to_ipc64_perm (struct kern_ipc_perm
*in
, struct ipc64_perm
*out
)
504 out
->cuid
= in
->cuid
;
505 out
->cgid
= in
->cgid
;
506 out
->mode
= in
->mode
;
511 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
512 * @in: new style IPC permissions
513 * @out: old style IPC permissions
515 * Turn the new style permissions object in into a compatibility
516 * object and store it into the 'out' pointer.
519 void ipc64_perm_to_ipc_perm (struct ipc64_perm
*in
, struct ipc_perm
*out
)
522 SET_UID(out
->uid
, in
->uid
);
523 SET_GID(out
->gid
, in
->gid
);
524 SET_UID(out
->cuid
, in
->cuid
);
525 SET_GID(out
->cgid
, in
->cgid
);
526 out
->mode
= in
->mode
;
531 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
532 * is called with shm_ids.sem locked. Since grow_ary() is also called with
533 * shm_ids.sem down(for Shared Memory), there is no need to add read
534 * barriers here to gurantee the writes in grow_ary() are seen in order
537 * However ipc_get() itself does not necessary require ipc_ids.sem down. So
538 * if in the future ipc_get() is used by other places without ipc_ids.sem
539 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
541 struct kern_ipc_perm
* ipc_get(struct ipc_ids
* ids
, int id
)
543 struct kern_ipc_perm
* out
;
544 int lid
= id
% SEQ_MULTIPLIER
;
545 if(lid
>= ids
->entries
->size
)
547 out
= ids
->entries
->p
[lid
];
551 struct kern_ipc_perm
* ipc_lock(struct ipc_ids
* ids
, int id
)
553 struct kern_ipc_perm
* out
;
554 int lid
= id
% SEQ_MULTIPLIER
;
555 struct ipc_id_ary
* entries
;
558 entries
= rcu_dereference(ids
->entries
);
559 if(lid
>= entries
->size
) {
563 out
= entries
->p
[lid
];
568 spin_lock(&out
->lock
);
570 /* ipc_rmid() may have already freed the ID while ipc_lock
571 * was spinning: here verify that the structure is still valid
574 spin_unlock(&out
->lock
);
581 void ipc_lock_by_ptr(struct kern_ipc_perm
*perm
)
584 spin_lock(&perm
->lock
);
587 void ipc_unlock(struct kern_ipc_perm
* perm
)
589 spin_unlock(&perm
->lock
);
593 int ipc_buildid(struct ipc_ids
* ids
, int id
, int seq
)
595 return SEQ_MULTIPLIER
*seq
+ id
;
598 int ipc_checkid(struct ipc_ids
* ids
, struct kern_ipc_perm
* ipcp
, int uid
)
600 if(uid
/SEQ_MULTIPLIER
!= ipcp
->seq
)
605 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
609 * ipc_parse_version - IPC call version
610 * @cmd: pointer to command
612 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
613 * The cmd value is turned from an encoding command and version into
614 * just the command code.
617 int ipc_parse_version (int *cmd
)
627 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
629 #ifdef CONFIG_PROC_FS
630 static void *sysvipc_proc_next(struct seq_file
*s
, void *it
, loff_t
*pos
)
632 struct ipc_proc_iface
*iface
= s
->private;
633 struct kern_ipc_perm
*ipc
= it
;
636 /* If we had an ipc id locked before, unlock it */
637 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
641 * p = *pos - 1 (because id 0 starts at position 1)
642 * + 1 (because we increment the position by one)
644 for (p
= *pos
; p
<= iface
->ids
->max_id
; p
++) {
645 if ((ipc
= ipc_lock(iface
->ids
, p
)) != NULL
) {
651 /* Out of range - return NULL to terminate iteration */
656 * File positions: pos 0 -> header, pos n -> ipc id + 1.
657 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
659 static void *sysvipc_proc_start(struct seq_file
*s
, loff_t
*pos
)
661 struct ipc_proc_iface
*iface
= s
->private;
662 struct kern_ipc_perm
*ipc
;
666 * Take the lock - this will be released by the corresponding
669 down(&iface
->ids
->sem
);
671 /* pos < 0 is invalid */
675 /* pos == 0 means header */
677 return SEQ_START_TOKEN
;
679 /* Find the (pos-1)th ipc */
680 for (p
= *pos
- 1; p
<= iface
->ids
->max_id
; p
++) {
681 if ((ipc
= ipc_lock(iface
->ids
, p
)) != NULL
) {
689 static void sysvipc_proc_stop(struct seq_file
*s
, void *it
)
691 struct kern_ipc_perm
*ipc
= it
;
692 struct ipc_proc_iface
*iface
= s
->private;
694 /* If we had a locked segment, release it */
695 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
698 /* Release the lock we took in start() */
699 up(&iface
->ids
->sem
);
702 static int sysvipc_proc_show(struct seq_file
*s
, void *it
)
704 struct ipc_proc_iface
*iface
= s
->private;
706 if (it
== SEQ_START_TOKEN
)
707 return seq_puts(s
, iface
->header
);
709 return iface
->show(s
, it
);
712 static struct seq_operations sysvipc_proc_seqops
= {
713 .start
= sysvipc_proc_start
,
714 .stop
= sysvipc_proc_stop
,
715 .next
= sysvipc_proc_next
,
716 .show
= sysvipc_proc_show
,
719 static int sysvipc_proc_open(struct inode
*inode
, struct file
*file
) {
721 struct seq_file
*seq
;
723 ret
= seq_open(file
, &sysvipc_proc_seqops
);
725 seq
= file
->private_data
;
726 seq
->private = PDE(inode
)->data
;
731 static struct file_operations sysvipc_proc_fops
= {
732 .open
= sysvipc_proc_open
,
735 .release
= seq_release
,
737 #endif /* CONFIG_PROC_FS */