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>
19 * General sysv ipc locking scheme:
21 * obtain the ipc object (kern_ipc_perm) by looking up the id in an idr
23 * - perform initial checks (capabilities, auditing and permission,
25 * - perform read-only operations, such as STAT, INFO commands.
26 * acquire the ipc lock (kern_ipc_perm.lock) through
28 * - perform data updates, such as SET, RMID commands and
29 * mechanism-specific operations (semop/semtimedop,
30 * msgsnd/msgrcv, shmat/shmdt).
31 * drop the ipc lock, through ipc_unlock_object().
34 * The ids->rwsem must be taken when:
35 * - creating, removing and iterating the existing entries in ipc
37 * - iterating through files under /proc/sysvipc/
39 * Note that sems have a special fast path that avoids kern_ipc_perm.lock -
44 #include <linux/shm.h>
45 #include <linux/init.h>
46 #include <linux/msg.h>
47 #include <linux/vmalloc.h>
48 #include <linux/slab.h>
49 #include <linux/notifier.h>
50 #include <linux/capability.h>
51 #include <linux/highuid.h>
52 #include <linux/security.h>
53 #include <linux/rcupdate.h>
54 #include <linux/workqueue.h>
55 #include <linux/seq_file.h>
56 #include <linux/proc_fs.h>
57 #include <linux/audit.h>
58 #include <linux/nsproxy.h>
59 #include <linux/rwsem.h>
60 #include <linux/memory.h>
61 #include <linux/ipc_namespace.h>
63 #include <asm/unistd.h>
67 struct ipc_proc_iface {
71 int (*show)(struct seq_file *, void *);
75 * ipc_init - initialise ipc subsystem
77 * The various sysv ipc resources (semaphores, messages and shared
78 * memory) are initialised.
80 * A callback routine is registered into the memory hotplug notifier
81 * chain: since msgmni scales to lowmem this callback routine will be
82 * called upon successful memory add / remove to recompute msmgni.
84 static int __init ipc_init(void)
89 WARN(err_sem, "ipc: sysv sem_init failed: %d\n", err_sem);
91 WARN(err_msg, "ipc: sysv msg_init failed: %d\n", err_msg);
94 return err_msg ? err_msg : err_sem;
96 device_initcall(ipc_init);
98 static const struct rhashtable_params ipc_kht_params = {
99 .head_offset = offsetof(struct kern_ipc_perm, khtnode),
100 .key_offset = offsetof(struct kern_ipc_perm, key),
101 .key_len = FIELD_SIZEOF(struct kern_ipc_perm, key),
103 .automatic_shrinking = true,
107 * ipc_init_ids - initialise ipc identifiers
108 * @ids: ipc identifier set
110 * Set up the sequence range to use for the ipc identifier range (limited
111 * below IPCMNI) then initialise the keys hashtable and ids idr.
113 int ipc_init_ids(struct ipc_ids *ids)
119 init_rwsem(&ids->rwsem);
120 err = rhashtable_init(&ids->key_ht, &ipc_kht_params);
123 idr_init(&ids->ipcs_idr);
124 ids->tables_initialized = true;
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 = NULL;
175 if (likely(ids->tables_initialized))
176 ipcp = rhashtable_lookup_fast(&ids->key_ht, &key,
181 ipc_lock_object(ipcp);
189 * ipc_get_maxid - get the last assigned id
190 * @ids: ipc identifier set
192 * Called with ipc_ids.rwsem held.
194 int ipc_get_maxid(struct ipc_ids *ids)
196 struct kern_ipc_perm *ipc;
200 if (ids->in_use == 0)
203 if (ids->in_use == IPCMNI)
206 /* Look for the last assigned id */
208 for (id = 0; id < IPCMNI && total < ids->in_use; id++) {
209 ipc = idr_find(&ids->ipcs_idr, id);
219 * ipc_addid - add an ipc identifier
220 * @ids: ipc identifier set
221 * @new: new ipc permission set
222 * @size: limit for the number of used ids
224 * Add an entry 'new' to the ipc ids idr. The permissions object is
225 * initialised and the first free entry is set up and the id assigned
226 * is returned. The 'new' entry is returned in a locked state on success.
227 * On failure the entry is not locked and a negative err-code is returned.
229 * Called with writer ipc_ids.rwsem held.
231 int ipc_addid(struct ipc_ids *ids, struct kern_ipc_perm *new, int size)
236 int next_id = ids->next_id;
241 if (!ids->tables_initialized || ids->in_use >= size)
244 idr_preload(GFP_KERNEL);
246 refcount_set(&new->refcount, 1);
247 spin_lock_init(&new->lock);
248 new->deleted = false;
250 spin_lock(&new->lock);
252 current_euid_egid(&euid, &egid);
253 new->cuid = new->uid = euid;
254 new->gid = new->cgid = egid;
256 id = idr_alloc(&ids->ipcs_idr, new,
257 (next_id < 0) ? 0 : ipcid_to_idx(next_id), 0,
261 if (id >= 0 && new->key != IPC_PRIVATE) {
262 err = rhashtable_insert_fast(&ids->key_ht, &new->khtnode,
265 idr_remove(&ids->ipcs_idr, id);
270 spin_unlock(&new->lock);
278 new->seq = ids->seq++;
279 if (ids->seq > IPCID_SEQ_MAX)
282 new->seq = ipcid_to_seqx(next_id);
286 new->id = ipc_buildid(id, new->seq);
291 * ipcget_new - create a new ipc object
293 * @ids: ipc identifier set
294 * @ops: the actual creation routine to call
295 * @params: its parameters
297 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
298 * when the key is IPC_PRIVATE.
300 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
301 const struct ipc_ops *ops, struct ipc_params *params)
305 down_write(&ids->rwsem);
306 err = ops->getnew(ns, params);
307 up_write(&ids->rwsem);
312 * ipc_check_perms - check security and permissions for an ipc object
314 * @ipcp: ipc permission set
315 * @ops: the actual security routine to call
316 * @params: its parameters
318 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
319 * when the key is not IPC_PRIVATE and that key already exists in the
322 * On success, the ipc id is returned.
324 * It is called with ipc_ids.rwsem and ipcp->lock held.
326 static int ipc_check_perms(struct ipc_namespace *ns,
327 struct kern_ipc_perm *ipcp,
328 const struct ipc_ops *ops,
329 struct ipc_params *params)
333 if (ipcperms(ns, ipcp, params->flg))
336 err = ops->associate(ipcp, params->flg);
345 * ipcget_public - get an ipc object or create a new one
347 * @ids: ipc identifier set
348 * @ops: the actual creation routine to call
349 * @params: its parameters
351 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
352 * when the key is not IPC_PRIVATE.
353 * It adds a new entry if the key is not found and does some permission
354 * / security checkings if the key is found.
356 * On success, the ipc id is returned.
358 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
359 const struct ipc_ops *ops, struct ipc_params *params)
361 struct kern_ipc_perm *ipcp;
362 int flg = params->flg;
366 * Take the lock as a writer since we are potentially going to add
367 * a new entry + read locks are not "upgradable"
369 down_write(&ids->rwsem);
370 ipcp = ipc_findkey(ids, params->key);
373 if (!(flg & IPC_CREAT))
376 err = ops->getnew(ns, params);
378 /* ipc object has been locked by ipc_findkey() */
380 if (flg & IPC_CREAT && flg & IPC_EXCL)
384 if (ops->more_checks)
385 err = ops->more_checks(ipcp, params);
388 * ipc_check_perms returns the IPC id on
391 err = ipc_check_perms(ns, ipcp, ops, params);
395 up_write(&ids->rwsem);
401 * ipc_kht_remove - remove an ipc from the key hashtable
402 * @ids: ipc identifier set
403 * @ipcp: ipc perm structure containing the key to remove
405 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
406 * before this function is called, and remain locked on the exit.
408 static void ipc_kht_remove(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
410 if (ipcp->key != IPC_PRIVATE)
411 rhashtable_remove_fast(&ids->key_ht, &ipcp->khtnode,
416 * ipc_rmid - remove an ipc identifier
417 * @ids: ipc identifier set
418 * @ipcp: ipc perm structure containing the identifier to remove
420 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
421 * before this function is called, and remain locked on the exit.
423 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
425 int lid = ipcid_to_idx(ipcp->id);
427 idr_remove(&ids->ipcs_idr, lid);
428 ipc_kht_remove(ids, ipcp);
430 ipcp->deleted = true;
434 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
435 * @ids: ipc identifier set
436 * @ipcp: ipc perm structure containing the key to modify
438 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
439 * before this function is called, and remain locked on the exit.
441 void ipc_set_key_private(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
443 ipc_kht_remove(ids, ipcp);
444 ipcp->key = IPC_PRIVATE;
447 int ipc_rcu_getref(struct kern_ipc_perm *ptr)
449 return refcount_inc_not_zero(&ptr->refcount);
452 void ipc_rcu_putref(struct kern_ipc_perm *ptr,
453 void (*func)(struct rcu_head *head))
455 if (!refcount_dec_and_test(&ptr->refcount))
458 call_rcu(&ptr->rcu, func);
462 * ipcperms - check ipc permissions
464 * @ipcp: ipc permission set
465 * @flag: desired permission set
467 * Check user, group, other permissions for access
468 * to ipc resources. return 0 if allowed
470 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
472 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
474 kuid_t euid = current_euid();
475 int requested_mode, granted_mode;
478 requested_mode = (flag >> 6) | (flag >> 3) | flag;
479 granted_mode = ipcp->mode;
480 if (uid_eq(euid, ipcp->cuid) ||
481 uid_eq(euid, ipcp->uid))
483 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
485 /* is there some bit set in requested_mode but not in granted_mode? */
486 if ((requested_mode & ~granted_mode & 0007) &&
487 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
490 return security_ipc_permission(ipcp, flag);
494 * Functions to convert between the kern_ipc_perm structure and the
495 * old/new ipc_perm structures
499 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
500 * @in: kernel permissions
501 * @out: new style ipc permissions
503 * Turn the kernel object @in into a set of permissions descriptions
504 * for returning to userspace (@out).
506 void kernel_to_ipc64_perm(struct kern_ipc_perm *in, struct ipc64_perm *out)
509 out->uid = from_kuid_munged(current_user_ns(), in->uid);
510 out->gid = from_kgid_munged(current_user_ns(), in->gid);
511 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
512 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
513 out->mode = in->mode;
518 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
519 * @in: new style ipc permissions
520 * @out: old style ipc permissions
522 * Turn the new style permissions object @in into a compatibility
523 * object and store it into the @out pointer.
525 void ipc64_perm_to_ipc_perm(struct ipc64_perm *in, struct ipc_perm *out)
528 SET_UID(out->uid, in->uid);
529 SET_GID(out->gid, in->gid);
530 SET_UID(out->cuid, in->cuid);
531 SET_GID(out->cgid, in->cgid);
532 out->mode = in->mode;
537 * ipc_obtain_object_idr
538 * @ids: ipc identifier set
539 * @id: ipc id to look for
541 * Look for an id in the ipc ids idr and return associated ipc object.
543 * Call inside the RCU critical section.
544 * The ipc object is *not* locked on exit.
546 struct kern_ipc_perm *ipc_obtain_object_idr(struct ipc_ids *ids, int id)
548 struct kern_ipc_perm *out;
549 int lid = ipcid_to_idx(id);
551 if (unlikely(!ids->tables_initialized))
552 return ERR_PTR(-EINVAL);
554 out = idr_find(&ids->ipcs_idr, lid);
556 return ERR_PTR(-EINVAL);
562 * ipc_lock - lock an ipc structure without rwsem held
563 * @ids: ipc identifier set
564 * @id: ipc id to look for
566 * Look for an id in the ipc ids idr and lock the associated ipc object.
568 * The ipc object is locked on successful exit.
570 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
572 struct kern_ipc_perm *out;
575 out = ipc_obtain_object_idr(ids, id);
579 spin_lock(&out->lock);
582 * ipc_rmid() may have already freed the ID while ipc_lock()
583 * was spinning: here verify that the structure is still valid.
584 * Upon races with RMID, return -EIDRM, thus indicating that
585 * the ID points to a removed identifier.
587 if (ipc_valid_object(out))
590 spin_unlock(&out->lock);
591 out = ERR_PTR(-EIDRM);
598 * ipc_obtain_object_check
599 * @ids: ipc identifier set
600 * @id: ipc id to look for
602 * Similar to ipc_obtain_object_idr() but also checks
603 * the ipc object reference counter.
605 * Call inside the RCU critical section.
606 * The ipc object is *not* locked on exit.
608 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
610 struct kern_ipc_perm *out = ipc_obtain_object_idr(ids, id);
615 if (ipc_checkid(out, id))
616 return ERR_PTR(-EINVAL);
622 * ipcget - Common sys_*get() code
624 * @ids: ipc identifier set
625 * @ops: operations to be called on ipc object creation, permission checks
627 * @params: the parameters needed by the previous operations.
629 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
631 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
632 const struct ipc_ops *ops, struct ipc_params *params)
634 if (params->key == IPC_PRIVATE)
635 return ipcget_new(ns, ids, ops, params);
637 return ipcget_public(ns, ids, ops, params);
641 * ipc_update_perm - update the permissions of an ipc object
642 * @in: the permission given as input.
643 * @out: the permission of the ipc to set.
645 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
647 kuid_t uid = make_kuid(current_user_ns(), in->uid);
648 kgid_t gid = make_kgid(current_user_ns(), in->gid);
649 if (!uid_valid(uid) || !gid_valid(gid))
654 out->mode = (out->mode & ~S_IRWXUGO)
655 | (in->mode & S_IRWXUGO);
661 * ipcctl_pre_down_nolock - retrieve an ipc and check permissions for some IPC_XXX cmd
663 * @ids: the table of ids where to look for the ipc
664 * @id: the id of the ipc to retrieve
665 * @cmd: the cmd to check
666 * @perm: the permission to set
667 * @extra_perm: one extra permission parameter used by msq
669 * This function does some common audit and permissions check for some IPC_XXX
670 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
671 * It must be called without any lock held and:
673 * - retrieves the ipc with the given id in the given table.
674 * - performs some audit and permission check, depending on the given cmd
675 * - returns a pointer to the ipc object or otherwise, the corresponding
678 * Call holding the both the rwsem and the rcu read lock.
680 struct kern_ipc_perm *ipcctl_pre_down_nolock(struct ipc_namespace *ns,
681 struct ipc_ids *ids, int id, int cmd,
682 struct ipc64_perm *perm, int extra_perm)
686 struct kern_ipc_perm *ipcp;
688 ipcp = ipc_obtain_object_check(ids, id);
696 audit_ipc_set_perm(extra_perm, perm->uid,
697 perm->gid, perm->mode);
699 euid = current_euid();
700 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
701 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
702 return ipcp; /* successful lookup */
707 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
711 * ipc_parse_version - ipc call version
712 * @cmd: pointer to command
714 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
715 * The @cmd value is turned from an encoding command and version into
716 * just the command code.
718 int ipc_parse_version(int *cmd)
728 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
730 #ifdef CONFIG_PROC_FS
731 struct ipc_proc_iter {
732 struct ipc_namespace *ns;
733 struct ipc_proc_iface *iface;
737 * This routine locks the ipc structure found at least at position pos.
739 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
742 struct kern_ipc_perm *ipc;
746 for (id = 0; id < pos && total < ids->in_use; id++) {
747 ipc = idr_find(&ids->ipcs_idr, id);
752 if (total >= ids->in_use)
755 for (; pos < IPCMNI; pos++) {
756 ipc = idr_find(&ids->ipcs_idr, pos);
760 ipc_lock_object(ipc);
765 /* Out of range - return NULL to terminate iteration */
769 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
771 struct ipc_proc_iter *iter = s->private;
772 struct ipc_proc_iface *iface = iter->iface;
773 struct kern_ipc_perm *ipc = it;
775 /* If we had an ipc id locked before, unlock it */
776 if (ipc && ipc != SEQ_START_TOKEN)
779 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
783 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
784 * SeqFile iterator: iterator value locked ipc pointer 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;
792 ids = &iter->ns->ids[iface->ids];
795 * Take the lock - this will be released by the corresponding
798 down_read(&ids->rwsem);
800 /* pos < 0 is invalid */
804 /* pos == 0 means header */
806 return SEQ_START_TOKEN;
808 /* Find the (pos-1)th ipc */
809 return sysvipc_find_ipc(ids, *pos - 1, pos);
812 static void sysvipc_proc_stop(struct seq_file *s, void *it)
814 struct kern_ipc_perm *ipc = it;
815 struct ipc_proc_iter *iter = s->private;
816 struct ipc_proc_iface *iface = iter->iface;
819 /* If we had a locked structure, release it */
820 if (ipc && ipc != SEQ_START_TOKEN)
823 ids = &iter->ns->ids[iface->ids];
824 /* Release the lock we took in start() */
825 up_read(&ids->rwsem);
828 static int sysvipc_proc_show(struct seq_file *s, void *it)
830 struct ipc_proc_iter *iter = s->private;
831 struct ipc_proc_iface *iface = iter->iface;
833 if (it == SEQ_START_TOKEN) {
834 seq_puts(s, iface->header);
838 return iface->show(s, it);
841 static const struct seq_operations sysvipc_proc_seqops = {
842 .start = sysvipc_proc_start,
843 .stop = sysvipc_proc_stop,
844 .next = sysvipc_proc_next,
845 .show = sysvipc_proc_show,
848 static int sysvipc_proc_open(struct inode *inode, struct file *file)
850 struct ipc_proc_iter *iter;
852 iter = __seq_open_private(file, &sysvipc_proc_seqops, sizeof(*iter));
856 iter->iface = PDE_DATA(inode);
857 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
862 static int sysvipc_proc_release(struct inode *inode, struct file *file)
864 struct seq_file *seq = file->private_data;
865 struct ipc_proc_iter *iter = seq->private;
866 put_ipc_ns(iter->ns);
867 return seq_release_private(inode, file);
870 static const struct file_operations sysvipc_proc_fops = {
871 .open = sysvipc_proc_open,
874 .release = sysvipc_proc_release,
876 #endif /* CONFIG_PROC_FS */