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 *);
74 static void ipc_memory_notifier(struct work_struct *work)
76 ipcns_notify(IPCNS_MEMCHANGED);
79 static int ipc_memory_callback(struct notifier_block *self,
80 unsigned long action, void *arg)
82 static DECLARE_WORK(ipc_memory_wq, ipc_memory_notifier);
85 case MEM_ONLINE: /* memory successfully brought online */
86 case MEM_OFFLINE: /* or offline: it's time to recompute msgmni */
88 * This is done by invoking the ipcns notifier chain with the
89 * IPC_MEMCHANGED event.
90 * In order not to keep the lock on the hotplug memory chain
91 * for too long, queue a work item that will, when waken up,
92 * activate the ipcns notification chain.
94 schedule_work(&ipc_memory_wq);
96 case MEM_GOING_ONLINE:
97 case MEM_GOING_OFFLINE:
98 case MEM_CANCEL_ONLINE:
99 case MEM_CANCEL_OFFLINE:
107 static struct notifier_block ipc_memory_nb = {
108 .notifier_call = ipc_memory_callback,
109 .priority = IPC_CALLBACK_PRI,
113 * ipc_init - initialise IPC subsystem
115 * The various system5 IPC resources (semaphores, messages and shared
116 * memory) are initialised
117 * A callback routine is registered into the memory hotplug notifier
118 * chain: since msgmni scales to lowmem this callback routine will be
119 * called upon successful memory add / remove to recompute msmgni.
122 static int __init ipc_init(void)
127 register_hotmemory_notifier(&ipc_memory_nb);
128 register_ipcns_notifier(&init_ipc_ns);
131 __initcall(ipc_init);
134 * ipc_init_ids - initialise IPC identifiers
135 * @ids: Identifier set
137 * Set up the sequence range to use for the ipc identifier range (limited
138 * below IPCMNI) then initialise the ids idr.
141 void ipc_init_ids(struct ipc_ids *ids)
143 init_rwsem(&ids->rwsem);
149 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
150 if (seq_limit > USHRT_MAX)
151 ids->seq_max = USHRT_MAX;
153 ids->seq_max = seq_limit;
156 idr_init(&ids->ipcs_idr);
159 #ifdef CONFIG_PROC_FS
160 static const struct file_operations sysvipc_proc_fops;
162 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
163 * @path: Path in procfs
164 * @header: Banner to be printed at the beginning of the file.
165 * @ids: ipc id table to iterate.
166 * @show: show routine.
168 void __init ipc_init_proc_interface(const char *path, const char *header,
169 int ids, int (*show)(struct seq_file *, void *))
171 struct proc_dir_entry *pde;
172 struct ipc_proc_iface *iface;
174 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
178 iface->header = header;
182 pde = proc_create_data(path,
183 S_IRUGO, /* world readable */
184 NULL, /* parent dir */
194 * ipc_findkey - find a key in an ipc identifier set
195 * @ids: Identifier set
196 * @key: The key to find
198 * Requires ipc_ids.rwsem locked.
199 * Returns the LOCKED pointer to the ipc structure if found or NULL
201 * If key is found ipc points to the owning ipc structure
204 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
206 struct kern_ipc_perm *ipc;
210 for (total = 0, next_id = 0; total < ids->in_use; next_id++) {
211 ipc = idr_find(&ids->ipcs_idr, next_id);
216 if (ipc->key != key) {
222 ipc_lock_object(ipc);
230 * ipc_get_maxid - get the last assigned id
231 * @ids: IPC identifier set
233 * Called with ipc_ids.rwsem held.
236 int ipc_get_maxid(struct ipc_ids *ids)
238 struct kern_ipc_perm *ipc;
242 if (ids->in_use == 0)
245 if (ids->in_use == IPCMNI)
248 /* Look for the last assigned id */
250 for (id = 0; id < IPCMNI && total < ids->in_use; id++) {
251 ipc = idr_find(&ids->ipcs_idr, id);
261 * ipc_addid - add an IPC identifier
262 * @ids: IPC identifier set
263 * @new: new IPC permission set
264 * @size: limit for the number of used ids
266 * Add an entry 'new' to the IPC ids idr. The permissions object is
267 * initialised and the first free entry is set up and the id assigned
268 * is returned. The 'new' entry is returned in a locked state on success.
269 * On failure the entry is not locked and a negative err-code is returned.
271 * Called with writer ipc_ids.rwsem held.
273 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
278 int next_id = ids->next_id;
283 if (ids->in_use >= size)
286 idr_preload(GFP_KERNEL);
288 spin_lock_init(&new->lock);
291 spin_lock(&new->lock);
293 id = idr_alloc(&ids->ipcs_idr, new,
294 (next_id < 0) ? 0 : ipcid_to_idx(next_id), 0,
298 spin_unlock(&new->lock);
305 current_euid_egid(&euid, &egid);
306 new->cuid = new->uid = euid;
307 new->gid = new->cgid = egid;
310 new->seq = ids->seq++;
311 if (ids->seq > ids->seq_max)
314 new->seq = ipcid_to_seqx(next_id);
318 new->id = ipc_buildid(id, new->seq);
323 * ipcget_new - create a new ipc object
325 * @ids: IPC identifer set
326 * @ops: the actual creation routine to call
327 * @params: its parameters
329 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
330 * when the key is IPC_PRIVATE.
332 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
333 struct ipc_ops *ops, struct ipc_params *params)
337 down_write(&ids->rwsem);
338 err = ops->getnew(ns, params);
339 up_write(&ids->rwsem);
344 * ipc_check_perms - check security and permissions for an IPC
346 * @ipcp: ipc permission set
347 * @ops: the actual security routine to call
348 * @params: its parameters
350 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
351 * when the key is not IPC_PRIVATE and that key already exists in the
354 * On success, the IPC id is returned.
356 * It is called with ipc_ids.rwsem and ipcp->lock held.
358 static int ipc_check_perms(struct ipc_namespace *ns,
359 struct kern_ipc_perm *ipcp,
361 struct ipc_params *params)
365 if (ipcperms(ns, ipcp, params->flg))
368 err = ops->associate(ipcp, params->flg);
377 * ipcget_public - get an ipc object or create a new one
379 * @ids: IPC identifer set
380 * @ops: the actual creation routine to call
381 * @params: its parameters
383 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
384 * when the key is not IPC_PRIVATE.
385 * It adds a new entry if the key is not found and does some permission
386 * / security checkings if the key is found.
388 * On success, the ipc id is returned.
390 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
391 struct ipc_ops *ops, struct ipc_params *params)
393 struct kern_ipc_perm *ipcp;
394 int flg = params->flg;
398 * Take the lock as a writer since we are potentially going to add
399 * a new entry + read locks are not "upgradable"
401 down_write(&ids->rwsem);
402 ipcp = ipc_findkey(ids, params->key);
405 if (!(flg & IPC_CREAT))
408 err = ops->getnew(ns, params);
410 /* ipc object has been locked by ipc_findkey() */
412 if (flg & IPC_CREAT && flg & IPC_EXCL)
416 if (ops->more_checks)
417 err = ops->more_checks(ipcp, params);
420 * ipc_check_perms returns the IPC id on
423 err = ipc_check_perms(ns, ipcp, ops, params);
427 up_write(&ids->rwsem);
434 * ipc_rmid - remove an IPC identifier
435 * @ids: IPC identifier set
436 * @ipcp: ipc perm structure containing the identifier to remove
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.
442 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
444 int lid = ipcid_to_idx(ipcp->id);
446 idr_remove(&ids->ipcs_idr, lid);
456 * ipc_alloc - allocate ipc space
457 * @size: size desired
459 * Allocate memory from the appropriate pools and return a pointer to it.
460 * NULL is returned if the allocation fails
463 void *ipc_alloc(int size)
469 out = kmalloc(size, GFP_KERNEL);
474 * ipc_free - free ipc space
475 * @ptr: pointer returned by ipc_alloc
476 * @size: size of block
478 * Free a block created with ipc_alloc(). The caller must know the size
479 * used in the allocation call.
482 void ipc_free(void* ptr, int size)
491 * ipc_rcu_alloc - allocate ipc and rcu space
492 * @size: size desired
494 * Allocate memory for the rcu header structure + the object.
495 * Returns the pointer to the object or NULL upon failure.
497 void *ipc_rcu_alloc(int size)
500 * We prepend the allocation with the rcu struct
502 struct ipc_rcu *out = ipc_alloc(sizeof(struct ipc_rcu) + size);
505 atomic_set(&out->refcount, 1);
509 int ipc_rcu_getref(void *ptr)
511 struct ipc_rcu *p = ((struct ipc_rcu *)ptr) - 1;
513 return atomic_inc_not_zero(&p->refcount);
516 void ipc_rcu_putref(void *ptr, void (*func)(struct rcu_head *head))
518 struct ipc_rcu *p = ((struct ipc_rcu *)ptr) - 1;
520 if (!atomic_dec_and_test(&p->refcount))
523 call_rcu(&p->rcu, func);
526 void ipc_rcu_free(struct rcu_head *head)
528 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
530 if (is_vmalloc_addr(p))
537 * ipcperms - check IPC permissions
539 * @ipcp: IPC permission set
540 * @flag: desired permission set.
542 * Check user, group, other permissions for access
543 * to ipc resources. return 0 if allowed
545 * @flag will most probably be 0 or S_...UGO from <linux/stat.h>
548 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
550 kuid_t euid = current_euid();
551 int requested_mode, granted_mode;
554 requested_mode = (flag >> 6) | (flag >> 3) | flag;
555 granted_mode = ipcp->mode;
556 if (uid_eq(euid, ipcp->cuid) ||
557 uid_eq(euid, ipcp->uid))
559 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
561 /* is there some bit set in requested_mode but not in granted_mode? */
562 if ((requested_mode & ~granted_mode & 0007) &&
563 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
566 return security_ipc_permission(ipcp, flag);
570 * Functions to convert between the kern_ipc_perm structure and the
571 * old/new ipc_perm structures
575 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
576 * @in: kernel permissions
577 * @out: new style IPC permissions
579 * Turn the kernel object @in into a set of permissions descriptions
580 * for returning to userspace (@out).
584 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
587 out->uid = from_kuid_munged(current_user_ns(), in->uid);
588 out->gid = from_kgid_munged(current_user_ns(), in->gid);
589 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
590 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
591 out->mode = in->mode;
596 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
597 * @in: new style IPC permissions
598 * @out: old style IPC permissions
600 * Turn the new style permissions object @in into a compatibility
601 * object and store it into the @out pointer.
604 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
607 SET_UID(out->uid, in->uid);
608 SET_GID(out->gid, in->gid);
609 SET_UID(out->cuid, in->cuid);
610 SET_GID(out->cgid, in->cgid);
611 out->mode = in->mode;
617 * @ids: ipc identifier set
618 * @id: ipc id to look for
620 * Look for an id in the ipc ids idr and return associated ipc object.
622 * Call inside the RCU critical section.
623 * The ipc object is *not* locked on exit.
625 struct kern_ipc_perm *ipc_obtain_object(struct ipc_ids *ids, int id)
627 struct kern_ipc_perm *out;
628 int lid = ipcid_to_idx(id);
630 out = idr_find(&ids->ipcs_idr, lid);
632 return ERR_PTR(-EINVAL);
638 * ipc_lock - Lock an ipc structure without rwsem held
639 * @ids: IPC identifier set
640 * @id: ipc id to look for
642 * Look for an id in the ipc ids idr and lock the associated ipc object.
644 * The ipc object is locked on successful exit.
646 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
648 struct kern_ipc_perm *out;
651 out = ipc_obtain_object(ids, id);
655 spin_lock(&out->lock);
657 /* ipc_rmid() may have already freed the ID while ipc_lock
658 * was spinning: here verify that the structure is still valid
663 spin_unlock(&out->lock);
664 out = ERR_PTR(-EINVAL);
671 * ipc_obtain_object_check
672 * @ids: ipc identifier set
673 * @id: ipc id to look for
675 * Similar to ipc_obtain_object() but also checks
676 * the ipc object reference counter.
678 * Call inside the RCU critical section.
679 * The ipc object is *not* locked on exit.
681 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
683 struct kern_ipc_perm *out = ipc_obtain_object(ids, id);
688 if (ipc_checkid(out, id))
689 return ERR_PTR(-EIDRM);
695 * ipcget - Common sys_*get() code
697 * @ids : IPC identifier set
698 * @ops : operations to be called on ipc object creation, permission checks
700 * @params : the parameters needed by the previous operations.
702 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
704 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
705 struct ipc_ops *ops, struct ipc_params *params)
707 if (params->key == IPC_PRIVATE)
708 return ipcget_new(ns, ids, ops, params);
710 return ipcget_public(ns, ids, ops, params);
714 * ipc_update_perm - update the permissions of an IPC.
715 * @in: the permission given as input.
716 * @out: the permission of the ipc to set.
718 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
720 kuid_t uid = make_kuid(current_user_ns(), in->uid);
721 kgid_t gid = make_kgid(current_user_ns(), in->gid);
722 if (!uid_valid(uid) || !gid_valid(gid))
727 out->mode = (out->mode & ~S_IRWXUGO)
728 | (in->mode & S_IRWXUGO);
734 * ipcctl_pre_down_nolock - retrieve an ipc and check permissions for some IPC_XXX cmd
735 * @ns: the ipc namespace
736 * @ids: the table of ids where to look for the ipc
737 * @id: the id of the ipc to retrieve
738 * @cmd: the cmd to check
739 * @perm: the permission to set
740 * @extra_perm: one extra permission parameter used by msq
742 * This function does some common audit and permissions check for some IPC_XXX
743 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
744 * It must be called without any lock held and
745 * - retrieves the ipc with the given id in the given table.
746 * - performs some audit and permission check, depending on the given cmd
747 * - returns a pointer to the ipc object or otherwise, the corresponding error.
749 * Call holding the both the rwsem and the rcu read lock.
751 struct kern_ipc_perm *ipcctl_pre_down_nolock(struct ipc_namespace *ns,
752 struct ipc_ids *ids, int id, int cmd,
753 struct ipc64_perm *perm, int extra_perm)
757 struct kern_ipc_perm *ipcp;
759 ipcp = ipc_obtain_object_check(ids, id);
767 audit_ipc_set_perm(extra_perm, perm->uid,
768 perm->gid, perm->mode);
770 euid = current_euid();
771 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
772 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
773 return ipcp; /* successful lookup */
778 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
782 * ipc_parse_version - IPC call version
783 * @cmd: pointer to command
785 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
786 * The @cmd value is turned from an encoding command and version into
787 * just the command code.
790 int ipc_parse_version (int *cmd)
800 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
802 #ifdef CONFIG_PROC_FS
803 struct ipc_proc_iter {
804 struct ipc_namespace *ns;
805 struct ipc_proc_iface *iface;
809 * This routine locks the ipc structure found at least at position pos.
811 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
814 struct kern_ipc_perm *ipc;
818 for (id = 0; id < pos && total < ids->in_use; id++) {
819 ipc = idr_find(&ids->ipcs_idr, id);
824 if (total >= ids->in_use)
827 for ( ; pos < IPCMNI; pos++) {
828 ipc = idr_find(&ids->ipcs_idr, pos);
832 ipc_lock_object(ipc);
837 /* Out of range - return NULL to terminate iteration */
841 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
843 struct ipc_proc_iter *iter = s->private;
844 struct ipc_proc_iface *iface = iter->iface;
845 struct kern_ipc_perm *ipc = it;
847 /* If we had an ipc id locked before, unlock it */
848 if (ipc && ipc != SEQ_START_TOKEN)
851 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
855 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
856 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
858 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
860 struct ipc_proc_iter *iter = s->private;
861 struct ipc_proc_iface *iface = iter->iface;
864 ids = &iter->ns->ids[iface->ids];
867 * Take the lock - this will be released by the corresponding
870 down_read(&ids->rwsem);
872 /* pos < 0 is invalid */
876 /* pos == 0 means header */
878 return SEQ_START_TOKEN;
880 /* Find the (pos-1)th ipc */
881 return sysvipc_find_ipc(ids, *pos - 1, pos);
884 static void sysvipc_proc_stop(struct seq_file *s, void *it)
886 struct kern_ipc_perm *ipc = it;
887 struct ipc_proc_iter *iter = s->private;
888 struct ipc_proc_iface *iface = iter->iface;
891 /* If we had a locked structure, release it */
892 if (ipc && ipc != SEQ_START_TOKEN)
895 ids = &iter->ns->ids[iface->ids];
896 /* Release the lock we took in start() */
897 up_read(&ids->rwsem);
900 static int sysvipc_proc_show(struct seq_file *s, void *it)
902 struct ipc_proc_iter *iter = s->private;
903 struct ipc_proc_iface *iface = iter->iface;
905 if (it == SEQ_START_TOKEN)
906 return seq_puts(s, iface->header);
908 return iface->show(s, it);
911 static const struct seq_operations sysvipc_proc_seqops = {
912 .start = sysvipc_proc_start,
913 .stop = sysvipc_proc_stop,
914 .next = sysvipc_proc_next,
915 .show = sysvipc_proc_show,
918 static int sysvipc_proc_open(struct inode *inode, struct file *file)
921 struct seq_file *seq;
922 struct ipc_proc_iter *iter;
925 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
929 ret = seq_open(file, &sysvipc_proc_seqops);
933 seq = file->private_data;
936 iter->iface = PDE_DATA(inode);
937 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
945 static int sysvipc_proc_release(struct inode *inode, struct file *file)
947 struct seq_file *seq = file->private_data;
948 struct ipc_proc_iter *iter = seq->private;
949 put_ipc_ns(iter->ns);
950 return seq_release_private(inode, file);
953 static const struct file_operations sysvipc_proc_fops = {
954 .open = sysvipc_proc_open,
957 .release = sysvipc_proc_release,
959 #endif /* CONFIG_PROC_FS */