1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 1992 Krishna Balasubramanian
6 * Sep 1997 - Call suser() last after "normal" permission checks so we
7 * get BSD style process accounting right.
8 * Occurs in several places in the IPC code.
9 * Chris Evans, <chris@ferret.lmh.ox.ac.uk>
10 * Nov 1999 - ipc helper functions, unified SMP locking
11 * Manfred Spraul <manfred@colorfullife.com>
12 * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
13 * Mingming Cao <cmm@us.ibm.com>
14 * Mar 2006 - support for audit of ipc object properties
15 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
16 * Jun 2006 - namespaces ssupport
18 * Pavel Emelianov <xemul@openvz.org>
20 * General sysv ipc locking scheme:
22 * obtain the ipc object (kern_ipc_perm) by looking up the id in an idr
24 * - perform initial checks (capabilities, auditing and permission,
26 * - perform read-only operations, such as INFO command, that
27 * do not demand atomicity
28 * acquire the ipc lock (kern_ipc_perm.lock) through
30 * - perform read-only operations that demand atomicity,
31 * such as STAT command.
32 * - perform data updates, such as SET, RMID commands and
33 * mechanism-specific operations (semop/semtimedop,
34 * msgsnd/msgrcv, shmat/shmdt).
35 * drop the ipc lock, through ipc_unlock_object().
38 * The ids->rwsem must be taken when:
39 * - creating, removing and iterating the existing entries in ipc
41 * - iterating through files under /proc/sysvipc/
43 * Note that sems have a special fast path that avoids kern_ipc_perm.lock -
48 #include <linux/shm.h>
49 #include <linux/init.h>
50 #include <linux/msg.h>
51 #include <linux/vmalloc.h>
52 #include <linux/slab.h>
53 #include <linux/notifier.h>
54 #include <linux/capability.h>
55 #include <linux/highuid.h>
56 #include <linux/security.h>
57 #include <linux/rcupdate.h>
58 #include <linux/workqueue.h>
59 #include <linux/seq_file.h>
60 #include <linux/proc_fs.h>
61 #include <linux/audit.h>
62 #include <linux/nsproxy.h>
63 #include <linux/rwsem.h>
64 #include <linux/memory.h>
65 #include <linux/ipc_namespace.h>
66 #include <linux/rhashtable.h>
68 #include <asm/unistd.h>
72 struct ipc_proc_iface
{
76 int (*show
)(struct seq_file
*, void *);
80 * ipc_init - initialise ipc subsystem
82 * The various sysv ipc resources (semaphores, messages and shared
83 * memory) are initialised.
85 * A callback routine is registered into the memory hotplug notifier
86 * chain: since msgmni scales to lowmem this callback routine will be
87 * called upon successful memory add / remove to recompute msmgni.
89 static int __init
ipc_init(void)
91 proc_mkdir("sysvipc", NULL
);
98 device_initcall(ipc_init
);
100 static const struct rhashtable_params ipc_kht_params
= {
101 .head_offset
= offsetof(struct kern_ipc_perm
, khtnode
),
102 .key_offset
= offsetof(struct kern_ipc_perm
, key
),
103 .key_len
= FIELD_SIZEOF(struct kern_ipc_perm
, key
),
105 .automatic_shrinking
= true,
109 * ipc_init_ids - initialise ipc identifiers
110 * @ids: ipc identifier set
112 * Set up the sequence range to use for the ipc identifier range (limited
113 * below IPCMNI) then initialise the keys hashtable and ids idr.
115 void ipc_init_ids(struct ipc_ids
*ids
)
119 init_rwsem(&ids
->rwsem
);
120 rhashtable_init(&ids
->key_ht
, &ipc_kht_params
);
121 idr_init(&ids
->ipcs_idr
);
123 #ifdef CONFIG_CHECKPOINT_RESTORE
128 #ifdef CONFIG_PROC_FS
129 static const struct file_operations sysvipc_proc_fops
;
131 * ipc_init_proc_interface - create a proc interface for sysipc types using a seq_file interface.
132 * @path: Path in procfs
133 * @header: Banner to be printed at the beginning of the file.
134 * @ids: ipc id table to iterate.
135 * @show: show routine.
137 void __init
ipc_init_proc_interface(const char *path
, const char *header
,
138 int ids
, int (*show
)(struct seq_file
*, void *))
140 struct proc_dir_entry
*pde
;
141 struct ipc_proc_iface
*iface
;
143 iface
= kmalloc(sizeof(*iface
), GFP_KERNEL
);
147 iface
->header
= header
;
151 pde
= proc_create_data(path
,
152 S_IRUGO
, /* world readable */
153 NULL
, /* parent dir */
162 * ipc_findkey - find a key in an ipc identifier set
163 * @ids: ipc identifier set
166 * Returns the locked pointer to the ipc structure if found or NULL
167 * otherwise. If key is found ipc points to the owning ipc structure
169 * Called with writer ipc_ids.rwsem held.
171 static struct kern_ipc_perm
*ipc_findkey(struct ipc_ids
*ids
, key_t key
)
173 struct kern_ipc_perm
*ipcp
;
175 ipcp
= rhashtable_lookup_fast(&ids
->key_ht
, &key
,
181 ipc_lock_object(ipcp
);
186 * Insert new IPC object into idr tree, and set sequence number and id
187 * in the correct order.
189 * - the sequence number must be set before inserting the object into the idr,
190 * because the sequence number is accessed without a lock.
191 * - the id can/must be set after inserting the object into the idr.
192 * All accesses must be done after getting kern_ipc_perm.lock.
194 * The caller must own kern_ipc_perm.lock.of the new object.
195 * On error, the function returns a (negative) error code.
197 static inline int ipc_idr_alloc(struct ipc_ids
*ids
, struct kern_ipc_perm
*new)
199 int idx
, next_id
= -1;
201 #ifdef CONFIG_CHECKPOINT_RESTORE
202 next_id
= ids
->next_id
;
207 * As soon as a new object is inserted into the idr,
208 * ipc_obtain_object_idr() or ipc_obtain_object_check() can find it,
209 * and the lockless preparations for ipc operations can start.
210 * This means especially: permission checks, audit calls, allocation
211 * of undo structures, ...
213 * Thus the object must be fully initialized, and if something fails,
214 * then the full tear-down sequence must be followed.
215 * (i.e.: set new->deleted, reduce refcount, call_rcu())
218 if (next_id
< 0) { /* !CHECKPOINT_RESTORE or next_id is unset */
219 new->seq
= ids
->seq
++;
220 if (ids
->seq
> IPCID_SEQ_MAX
)
222 idx
= idr_alloc(&ids
->ipcs_idr
, new, 0, 0, GFP_NOWAIT
);
224 new->seq
= ipcid_to_seqx(next_id
);
225 idx
= idr_alloc(&ids
->ipcs_idr
, new, ipcid_to_idx(next_id
),
229 new->id
= SEQ_MULTIPLIER
* new->seq
+ idx
;
234 * ipc_addid - add an ipc identifier
235 * @ids: ipc identifier set
236 * @new: new ipc permission set
237 * @limit: limit for the number of used ids
239 * Add an entry 'new' to the ipc ids idr. The permissions object is
240 * initialised and the first free entry is set up and the index assigned
241 * is returned. The 'new' entry is returned in a locked state on success.
243 * On failure the entry is not locked and a negative err-code is returned.
244 * The caller must use ipc_rcu_putref() to free the identifier.
246 * Called with writer ipc_ids.rwsem held.
248 int ipc_addid(struct ipc_ids
*ids
, struct kern_ipc_perm
*new, int limit
)
254 /* 1) Initialize the refcount so that ipc_rcu_putref works */
255 refcount_set(&new->refcount
, 1);
260 if (ids
->in_use
>= limit
)
263 idr_preload(GFP_KERNEL
);
265 spin_lock_init(&new->lock
);
267 spin_lock(&new->lock
);
269 current_euid_egid(&euid
, &egid
);
270 new->cuid
= new->uid
= euid
;
271 new->gid
= new->cgid
= egid
;
273 new->deleted
= false;
275 idx
= ipc_idr_alloc(ids
, new);
278 if (idx
>= 0 && new->key
!= IPC_PRIVATE
) {
279 err
= rhashtable_insert_fast(&ids
->key_ht
, &new->khtnode
,
282 idr_remove(&ids
->ipcs_idr
, idx
);
288 spin_unlock(&new->lock
);
294 if (idx
> ids
->max_idx
)
300 * ipcget_new - create a new ipc object
302 * @ids: ipc identifier set
303 * @ops: the actual creation routine to call
304 * @params: its parameters
306 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
307 * when the key is IPC_PRIVATE.
309 static int ipcget_new(struct ipc_namespace
*ns
, struct ipc_ids
*ids
,
310 const struct ipc_ops
*ops
, struct ipc_params
*params
)
314 down_write(&ids
->rwsem
);
315 err
= ops
->getnew(ns
, params
);
316 up_write(&ids
->rwsem
);
321 * ipc_check_perms - check security and permissions for an ipc object
323 * @ipcp: ipc permission set
324 * @ops: the actual security routine to call
325 * @params: its parameters
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
331 * On success, the ipc id is returned.
333 * It is called with ipc_ids.rwsem and ipcp->lock held.
335 static int ipc_check_perms(struct ipc_namespace
*ns
,
336 struct kern_ipc_perm
*ipcp
,
337 const struct ipc_ops
*ops
,
338 struct ipc_params
*params
)
342 if (ipcperms(ns
, ipcp
, params
->flg
))
345 err
= ops
->associate(ipcp
, params
->flg
);
354 * ipcget_public - get an ipc object or create a new one
356 * @ids: ipc identifier set
357 * @ops: the actual creation routine to call
358 * @params: its parameters
360 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
361 * when the key is not IPC_PRIVATE.
362 * It adds a new entry if the key is not found and does some permission
363 * / security checkings if the key is found.
365 * On success, the ipc id is returned.
367 static int ipcget_public(struct ipc_namespace
*ns
, struct ipc_ids
*ids
,
368 const struct ipc_ops
*ops
, struct ipc_params
*params
)
370 struct kern_ipc_perm
*ipcp
;
371 int flg
= params
->flg
;
375 * Take the lock as a writer since we are potentially going to add
376 * a new entry + read locks are not "upgradable"
378 down_write(&ids
->rwsem
);
379 ipcp
= ipc_findkey(ids
, params
->key
);
382 if (!(flg
& IPC_CREAT
))
385 err
= ops
->getnew(ns
, params
);
387 /* ipc object has been locked by ipc_findkey() */
389 if (flg
& IPC_CREAT
&& flg
& IPC_EXCL
)
393 if (ops
->more_checks
)
394 err
= ops
->more_checks(ipcp
, params
);
397 * ipc_check_perms returns the IPC id on
400 err
= ipc_check_perms(ns
, ipcp
, ops
, params
);
404 up_write(&ids
->rwsem
);
410 * ipc_kht_remove - remove an ipc from the key hashtable
411 * @ids: ipc identifier set
412 * @ipcp: ipc perm structure containing the key to remove
414 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
415 * before this function is called, and remain locked on the exit.
417 static void ipc_kht_remove(struct ipc_ids
*ids
, struct kern_ipc_perm
*ipcp
)
419 if (ipcp
->key
!= IPC_PRIVATE
)
420 rhashtable_remove_fast(&ids
->key_ht
, &ipcp
->khtnode
,
425 * ipc_rmid - remove an ipc identifier
426 * @ids: ipc identifier set
427 * @ipcp: ipc perm structure containing the identifier to remove
429 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
430 * before this function is called, and remain locked on the exit.
432 void ipc_rmid(struct ipc_ids
*ids
, struct kern_ipc_perm
*ipcp
)
434 int idx
= ipcid_to_idx(ipcp
->id
);
436 idr_remove(&ids
->ipcs_idr
, idx
);
437 ipc_kht_remove(ids
, ipcp
);
439 ipcp
->deleted
= true;
441 if (unlikely(idx
== ids
->max_idx
)) {
446 } while (!idr_find(&ids
->ipcs_idr
, idx
));
452 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
453 * @ids: ipc identifier set
454 * @ipcp: ipc perm structure containing the key to modify
456 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
457 * before this function is called, and remain locked on the exit.
459 void ipc_set_key_private(struct ipc_ids
*ids
, struct kern_ipc_perm
*ipcp
)
461 ipc_kht_remove(ids
, ipcp
);
462 ipcp
->key
= IPC_PRIVATE
;
465 bool ipc_rcu_getref(struct kern_ipc_perm
*ptr
)
467 return refcount_inc_not_zero(&ptr
->refcount
);
470 void ipc_rcu_putref(struct kern_ipc_perm
*ptr
,
471 void (*func
)(struct rcu_head
*head
))
473 if (!refcount_dec_and_test(&ptr
->refcount
))
476 call_rcu(&ptr
->rcu
, func
);
480 * ipcperms - check ipc permissions
482 * @ipcp: ipc permission set
483 * @flag: desired permission set
485 * Check user, group, other permissions for access
486 * to ipc resources. return 0 if allowed
488 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
490 int ipcperms(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
, short flag
)
492 kuid_t euid
= current_euid();
493 int requested_mode
, granted_mode
;
496 requested_mode
= (flag
>> 6) | (flag
>> 3) | flag
;
497 granted_mode
= ipcp
->mode
;
498 if (uid_eq(euid
, ipcp
->cuid
) ||
499 uid_eq(euid
, ipcp
->uid
))
501 else if (in_group_p(ipcp
->cgid
) || in_group_p(ipcp
->gid
))
503 /* is there some bit set in requested_mode but not in granted_mode? */
504 if ((requested_mode
& ~granted_mode
& 0007) &&
505 !ns_capable(ns
->user_ns
, CAP_IPC_OWNER
))
508 return security_ipc_permission(ipcp
, flag
);
512 * Functions to convert between the kern_ipc_perm structure and the
513 * old/new ipc_perm structures
517 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
518 * @in: kernel permissions
519 * @out: new style ipc permissions
521 * Turn the kernel object @in into a set of permissions descriptions
522 * for returning to userspace (@out).
524 void kernel_to_ipc64_perm(struct kern_ipc_perm
*in
, struct ipc64_perm
*out
)
527 out
->uid
= from_kuid_munged(current_user_ns(), in
->uid
);
528 out
->gid
= from_kgid_munged(current_user_ns(), in
->gid
);
529 out
->cuid
= from_kuid_munged(current_user_ns(), in
->cuid
);
530 out
->cgid
= from_kgid_munged(current_user_ns(), in
->cgid
);
531 out
->mode
= in
->mode
;
536 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
537 * @in: new style ipc permissions
538 * @out: old style ipc permissions
540 * Turn the new style permissions object @in into a compatibility
541 * object and store it into the @out pointer.
543 void ipc64_perm_to_ipc_perm(struct ipc64_perm
*in
, struct ipc_perm
*out
)
546 SET_UID(out
->uid
, in
->uid
);
547 SET_GID(out
->gid
, in
->gid
);
548 SET_UID(out
->cuid
, in
->cuid
);
549 SET_GID(out
->cgid
, in
->cgid
);
550 out
->mode
= in
->mode
;
555 * ipc_obtain_object_idr
556 * @ids: ipc identifier set
557 * @id: ipc id to look for
559 * Look for an id in the ipc ids idr and return associated ipc object.
561 * Call inside the RCU critical section.
562 * The ipc object is *not* locked on exit.
564 struct kern_ipc_perm
*ipc_obtain_object_idr(struct ipc_ids
*ids
, int id
)
566 struct kern_ipc_perm
*out
;
567 int idx
= ipcid_to_idx(id
);
569 out
= idr_find(&ids
->ipcs_idr
, idx
);
571 return ERR_PTR(-EINVAL
);
577 * ipc_obtain_object_check
578 * @ids: ipc identifier set
579 * @id: ipc id to look for
581 * Similar to ipc_obtain_object_idr() but also checks the ipc object
584 * Call inside the RCU critical section.
585 * The ipc object is *not* locked on exit.
587 struct kern_ipc_perm
*ipc_obtain_object_check(struct ipc_ids
*ids
, int id
)
589 struct kern_ipc_perm
*out
= ipc_obtain_object_idr(ids
, id
);
594 if (ipc_checkid(out
, id
))
595 return ERR_PTR(-EINVAL
);
601 * ipcget - Common sys_*get() code
603 * @ids: ipc identifier set
604 * @ops: operations to be called on ipc object creation, permission checks
606 * @params: the parameters needed by the previous operations.
608 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
610 int ipcget(struct ipc_namespace
*ns
, struct ipc_ids
*ids
,
611 const struct ipc_ops
*ops
, struct ipc_params
*params
)
613 if (params
->key
== IPC_PRIVATE
)
614 return ipcget_new(ns
, ids
, ops
, params
);
616 return ipcget_public(ns
, ids
, ops
, params
);
620 * ipc_update_perm - update the permissions of an ipc object
621 * @in: the permission given as input.
622 * @out: the permission of the ipc to set.
624 int ipc_update_perm(struct ipc64_perm
*in
, struct kern_ipc_perm
*out
)
626 kuid_t uid
= make_kuid(current_user_ns(), in
->uid
);
627 kgid_t gid
= make_kgid(current_user_ns(), in
->gid
);
628 if (!uid_valid(uid
) || !gid_valid(gid
))
633 out
->mode
= (out
->mode
& ~S_IRWXUGO
)
634 | (in
->mode
& S_IRWXUGO
);
640 * ipcctl_obtain_check - retrieve an ipc object and check permissions
642 * @ids: the table of ids where to look for the ipc
643 * @id: the id of the ipc to retrieve
644 * @cmd: the cmd to check
645 * @perm: the permission to set
646 * @extra_perm: one extra permission parameter used by msq
648 * This function does some common audit and permissions check for some IPC_XXX
649 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
652 * - retrieves the ipc object with the given id in the given table.
653 * - performs some audit and permission check, depending on the given cmd
654 * - returns a pointer to the ipc object or otherwise, the corresponding
657 * Call holding the both the rwsem and the rcu read lock.
659 struct kern_ipc_perm
*ipcctl_obtain_check(struct ipc_namespace
*ns
,
660 struct ipc_ids
*ids
, int id
, int cmd
,
661 struct ipc64_perm
*perm
, int extra_perm
)
665 struct kern_ipc_perm
*ipcp
;
667 ipcp
= ipc_obtain_object_check(ids
, id
);
675 audit_ipc_set_perm(extra_perm
, perm
->uid
,
676 perm
->gid
, perm
->mode
);
678 euid
= current_euid();
679 if (uid_eq(euid
, ipcp
->cuid
) || uid_eq(euid
, ipcp
->uid
) ||
680 ns_capable(ns
->user_ns
, CAP_SYS_ADMIN
))
681 return ipcp
; /* successful lookup */
686 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
690 * ipc_parse_version - ipc call version
691 * @cmd: pointer to command
693 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
694 * The @cmd value is turned from an encoding command and version into
695 * just the command code.
697 int ipc_parse_version(int *cmd
)
707 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
709 #ifdef CONFIG_PROC_FS
710 struct ipc_proc_iter
{
711 struct ipc_namespace
*ns
;
712 struct pid_namespace
*pid_ns
;
713 struct ipc_proc_iface
*iface
;
716 struct pid_namespace
*ipc_seq_pid_ns(struct seq_file
*s
)
718 struct ipc_proc_iter
*iter
= s
->private;
723 * This routine locks the ipc structure found at least at position pos.
725 static struct kern_ipc_perm
*sysvipc_find_ipc(struct ipc_ids
*ids
, loff_t pos
,
728 struct kern_ipc_perm
*ipc
;
732 for (id
= 0; id
< pos
&& total
< ids
->in_use
; id
++) {
733 ipc
= idr_find(&ids
->ipcs_idr
, id
);
738 if (total
>= ids
->in_use
)
741 for (; pos
< IPCMNI
; pos
++) {
742 ipc
= idr_find(&ids
->ipcs_idr
, pos
);
746 ipc_lock_object(ipc
);
751 /* Out of range - return NULL to terminate iteration */
755 static void *sysvipc_proc_next(struct seq_file
*s
, void *it
, loff_t
*pos
)
757 struct ipc_proc_iter
*iter
= s
->private;
758 struct ipc_proc_iface
*iface
= iter
->iface
;
759 struct kern_ipc_perm
*ipc
= it
;
761 /* If we had an ipc id locked before, unlock it */
762 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
765 return sysvipc_find_ipc(&iter
->ns
->ids
[iface
->ids
], *pos
, pos
);
769 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
770 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
772 static void *sysvipc_proc_start(struct seq_file
*s
, loff_t
*pos
)
774 struct ipc_proc_iter
*iter
= s
->private;
775 struct ipc_proc_iface
*iface
= iter
->iface
;
778 ids
= &iter
->ns
->ids
[iface
->ids
];
781 * Take the lock - this will be released by the corresponding
784 down_read(&ids
->rwsem
);
786 /* pos < 0 is invalid */
790 /* pos == 0 means header */
792 return SEQ_START_TOKEN
;
794 /* Find the (pos-1)th ipc */
795 return sysvipc_find_ipc(ids
, *pos
- 1, pos
);
798 static void sysvipc_proc_stop(struct seq_file
*s
, void *it
)
800 struct kern_ipc_perm
*ipc
= it
;
801 struct ipc_proc_iter
*iter
= s
->private;
802 struct ipc_proc_iface
*iface
= iter
->iface
;
805 /* If we had a locked structure, release it */
806 if (ipc
&& ipc
!= SEQ_START_TOKEN
)
809 ids
= &iter
->ns
->ids
[iface
->ids
];
810 /* Release the lock we took in start() */
811 up_read(&ids
->rwsem
);
814 static int sysvipc_proc_show(struct seq_file
*s
, void *it
)
816 struct ipc_proc_iter
*iter
= s
->private;
817 struct ipc_proc_iface
*iface
= iter
->iface
;
819 if (it
== SEQ_START_TOKEN
) {
820 seq_puts(s
, iface
->header
);
824 return iface
->show(s
, it
);
827 static const struct seq_operations sysvipc_proc_seqops
= {
828 .start
= sysvipc_proc_start
,
829 .stop
= sysvipc_proc_stop
,
830 .next
= sysvipc_proc_next
,
831 .show
= sysvipc_proc_show
,
834 static int sysvipc_proc_open(struct inode
*inode
, struct file
*file
)
836 struct ipc_proc_iter
*iter
;
838 iter
= __seq_open_private(file
, &sysvipc_proc_seqops
, sizeof(*iter
));
842 iter
->iface
= PDE_DATA(inode
);
843 iter
->ns
= get_ipc_ns(current
->nsproxy
->ipc_ns
);
844 iter
->pid_ns
= get_pid_ns(task_active_pid_ns(current
));
849 static int sysvipc_proc_release(struct inode
*inode
, struct file
*file
)
851 struct seq_file
*seq
= file
->private_data
;
852 struct ipc_proc_iter
*iter
= seq
->private;
853 put_ipc_ns(iter
->ns
);
854 put_pid_ns(iter
->pid_ns
);
855 return seq_release_private(inode
, file
);
858 static const struct file_operations sysvipc_proc_fops
= {
859 .open
= sysvipc_proc_open
,
862 .release
= sysvipc_proc_release
,
864 #endif /* CONFIG_PROC_FS */