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/vmalloc.h>
25 #include <linux/slab.h>
26 #include <linux/capability.h>
27 #include <linux/highuid.h>
28 #include <linux/security.h>
29 #include <linux/rcupdate.h>
30 #include <linux/workqueue.h>
31 #include <linux/seq_file.h>
32 #include <linux/proc_fs.h>
33 #include <linux/audit.h>
34 #include <linux/nsproxy.h>
35 #include <linux/rwsem.h>
36 #include <linux/memory.h>
37 #include <linux/ipc_namespace.h>
39 #include <asm/unistd.h>
43 struct ipc_proc_iface {
47 int (*show)(struct seq_file *, void *);
50 #ifdef CONFIG_MEMORY_HOTPLUG
52 static void ipc_memory_notifier(struct work_struct *work)
54 ipcns_notify(IPCNS_MEMCHANGED);
57 static DECLARE_WORK(ipc_memory_wq, ipc_memory_notifier);
60 static int ipc_memory_callback(struct notifier_block *self,
61 unsigned long action, void *arg)
64 case MEM_ONLINE: /* memory successfully brought online */
65 case MEM_OFFLINE: /* or offline: it's time to recompute msgmni */
67 * This is done by invoking the ipcns notifier chain with the
68 * IPC_MEMCHANGED event.
69 * In order not to keep the lock on the hotplug memory chain
70 * for too long, queue a work item that will, when waken up,
71 * activate the ipcns notification chain.
72 * No need to keep several ipc work items on the queue.
74 if (!work_pending(&ipc_memory_wq))
75 schedule_work(&ipc_memory_wq);
77 case MEM_GOING_ONLINE:
78 case MEM_GOING_OFFLINE:
79 case MEM_CANCEL_ONLINE:
80 case MEM_CANCEL_OFFLINE:
88 #endif /* CONFIG_MEMORY_HOTPLUG */
91 * ipc_init - initialise IPC subsystem
93 * The various system5 IPC resources (semaphores, messages and shared
94 * memory) are initialised
95 * A callback routine is registered into the memory hotplug notifier
96 * chain: since msgmni scales to lowmem this callback routine will be
97 * called upon successful memory add / remove to recompute msmgni.
100 static int __init ipc_init(void)
105 hotplug_memory_notifier(ipc_memory_callback, IPC_CALLBACK_PRI);
106 register_ipcns_notifier(&init_ipc_ns);
109 __initcall(ipc_init);
112 * ipc_init_ids - initialise IPC identifiers
113 * @ids: Identifier set
115 * Set up the sequence range to use for the ipc identifier range (limited
116 * below IPCMNI) then initialise the ids idr.
119 void ipc_init_ids(struct ipc_ids *ids)
121 init_rwsem(&ids->rw_mutex);
126 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
127 if (seq_limit > USHRT_MAX)
128 ids->seq_max = USHRT_MAX;
130 ids->seq_max = seq_limit;
133 idr_init(&ids->ipcs_idr);
136 #ifdef CONFIG_PROC_FS
137 static const struct file_operations sysvipc_proc_fops;
139 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
140 * @path: Path in procfs
141 * @header: Banner to be printed at the beginning of the file.
142 * @ids: ipc id table to iterate.
143 * @show: show routine.
145 void __init ipc_init_proc_interface(const char *path, const char *header,
146 int ids, int (*show)(struct seq_file *, void *))
148 struct proc_dir_entry *pde;
149 struct ipc_proc_iface *iface;
151 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
155 iface->header = header;
159 pde = proc_create_data(path,
160 S_IRUGO, /* world readable */
161 NULL, /* parent dir */
171 * ipc_findkey - find a key in an ipc identifier set
172 * @ids: Identifier set
173 * @key: The key to find
175 * Requires ipc_ids.rw_mutex locked.
176 * Returns the LOCKED pointer to the ipc structure if found or NULL
178 * If key is found ipc points to the owning ipc structure
181 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
183 struct kern_ipc_perm *ipc;
187 for (total = 0, next_id = 0; total < ids->in_use; next_id++) {
188 ipc = idr_find(&ids->ipcs_idr, next_id);
193 if (ipc->key != key) {
198 ipc_lock_by_ptr(ipc);
206 * ipc_get_maxid - get the last assigned id
207 * @ids: IPC identifier set
209 * Called with ipc_ids.rw_mutex held.
212 int ipc_get_maxid(struct ipc_ids *ids)
214 struct kern_ipc_perm *ipc;
218 if (ids->in_use == 0)
221 if (ids->in_use == IPCMNI)
224 /* Look for the last assigned id */
226 for (id = 0; id < IPCMNI && total < ids->in_use; id++) {
227 ipc = idr_find(&ids->ipcs_idr, id);
237 * ipc_addid - add an IPC identifier
238 * @ids: IPC identifier set
239 * @new: new IPC permission set
240 * @size: limit for the number of used ids
242 * Add an entry 'new' to the IPC ids idr. The permissions object is
243 * initialised and the first free entry is set up and the id assigned
244 * is returned. The 'new' entry is returned in a locked state on success.
245 * On failure the entry is not locked and a negative err-code is returned.
247 * Called with ipc_ids.rw_mutex held as a writer.
250 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
259 if (ids->in_use >= size)
262 spin_lock_init(&new->lock);
265 spin_lock(&new->lock);
267 err = idr_get_new(&ids->ipcs_idr, new, &id);
269 spin_unlock(&new->lock);
276 current_euid_egid(&euid, &egid);
277 new->cuid = new->uid = euid;
278 new->gid = new->cgid = egid;
280 new->seq = ids->seq++;
281 if(ids->seq > ids->seq_max)
284 new->id = ipc_buildid(id, new->seq);
289 * ipcget_new - create a new ipc object
291 * @ids: IPC identifer set
292 * @ops: the actual creation routine to call
293 * @params: its parameters
295 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
296 * when the key is IPC_PRIVATE.
298 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
299 struct ipc_ops *ops, struct ipc_params *params)
303 err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL);
308 down_write(&ids->rw_mutex);
309 err = ops->getnew(ns, params);
310 up_write(&ids->rw_mutex);
319 * ipc_check_perms - check security and permissions for an IPC
320 * @ipcp: ipc permission set
321 * @ops: the actual security routine to call
322 * @params: its parameters
324 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
325 * when the key is not IPC_PRIVATE and that key already exists in the
328 * On success, the IPC id is returned.
330 * It is called with ipc_ids.rw_mutex and ipcp->lock held.
332 static int ipc_check_perms(struct kern_ipc_perm *ipcp, struct ipc_ops *ops,
333 struct ipc_params *params)
337 if (ipcperms(ipcp, params->flg))
340 err = ops->associate(ipcp, params->flg);
349 * ipcget_public - get an ipc object or create a new one
351 * @ids: IPC identifer set
352 * @ops: the actual creation routine to call
353 * @params: its parameters
355 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
356 * when the key is not IPC_PRIVATE.
357 * It adds a new entry if the key is not found and does some permission
358 * / security checkings if the key is found.
360 * On success, the ipc id is returned.
362 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
363 struct ipc_ops *ops, struct ipc_params *params)
365 struct kern_ipc_perm *ipcp;
366 int flg = params->flg;
369 err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL);
372 * Take the lock as a writer since we are potentially going to add
373 * a new entry + read locks are not "upgradable"
375 down_write(&ids->rw_mutex);
376 ipcp = ipc_findkey(ids, params->key);
379 if (!(flg & IPC_CREAT))
384 err = ops->getnew(ns, params);
386 /* ipc object has been locked by ipc_findkey() */
388 if (flg & IPC_CREAT && flg & IPC_EXCL)
392 if (ops->more_checks)
393 err = ops->more_checks(ipcp, params);
396 * ipc_check_perms returns the IPC id on
399 err = ipc_check_perms(ipcp, ops, params);
403 up_write(&ids->rw_mutex);
413 * ipc_rmid - remove an IPC identifier
414 * @ids: IPC identifier set
415 * @ipcp: ipc perm structure containing the identifier to remove
417 * ipc_ids.rw_mutex (as a writer) and the spinlock for this ID are held
418 * before this function is called, and remain locked on the exit.
421 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
423 int lid = ipcid_to_idx(ipcp->id);
425 idr_remove(&ids->ipcs_idr, lid);
435 * ipc_alloc - allocate ipc space
436 * @size: size desired
438 * Allocate memory from the appropriate pools and return a pointer to it.
439 * NULL is returned if the allocation fails
442 void* ipc_alloc(int size)
448 out = kmalloc(size, GFP_KERNEL);
453 * ipc_free - free ipc space
454 * @ptr: pointer returned by ipc_alloc
455 * @size: size of block
457 * Free a block created with ipc_alloc(). The caller must know the size
458 * used in the allocation call.
461 void ipc_free(void* ptr, int size)
471 * There are three headers that are prepended to the actual allocation:
472 * - during use: ipc_rcu_hdr.
473 * - during the rcu grace period: ipc_rcu_grace.
474 * - [only if vmalloc]: ipc_rcu_sched.
475 * Their lifetime doesn't overlap, thus the headers share the same memory.
476 * Unlike a normal union, they are right-aligned, thus some container_of
477 * forward/backward casting is necessary:
490 /* "void *" makes sure alignment of following data is sane. */
496 struct work_struct work;
497 /* "void *" makes sure alignment of following data is sane. */
501 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
502 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
503 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
504 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
506 static inline int rcu_use_vmalloc(int size)
508 /* Too big for a single page? */
509 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
515 * ipc_rcu_alloc - allocate ipc and rcu space
516 * @size: size desired
518 * Allocate memory for the rcu header structure + the object.
519 * Returns the pointer to the object.
520 * NULL is returned if the allocation fails.
523 void* ipc_rcu_alloc(int size)
527 * We prepend the allocation with the rcu struct, and
528 * workqueue if necessary (for vmalloc).
530 if (rcu_use_vmalloc(size)) {
531 out = vmalloc(HDRLEN_VMALLOC + size);
533 out += HDRLEN_VMALLOC;
534 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
535 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
538 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
540 out += HDRLEN_KMALLOC;
541 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
542 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
549 void ipc_rcu_getref(void *ptr)
551 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
554 static void ipc_do_vfree(struct work_struct *work)
556 vfree(container_of(work, struct ipc_rcu_sched, work));
560 * ipc_schedule_free - free ipc + rcu space
561 * @head: RCU callback structure for queued work
563 * Since RCU callback function is called in bh,
564 * we need to defer the vfree to schedule_work().
566 static void ipc_schedule_free(struct rcu_head *head)
568 struct ipc_rcu_grace *grace;
569 struct ipc_rcu_sched *sched;
571 grace = container_of(head, struct ipc_rcu_grace, rcu);
572 sched = container_of(&(grace->data[0]), struct ipc_rcu_sched,
575 INIT_WORK(&sched->work, ipc_do_vfree);
576 schedule_work(&sched->work);
580 * ipc_immediate_free - free ipc + rcu space
581 * @head: RCU callback structure that contains pointer to be freed
583 * Free from the RCU callback context.
585 static void ipc_immediate_free(struct rcu_head *head)
587 struct ipc_rcu_grace *free =
588 container_of(head, struct ipc_rcu_grace, rcu);
592 void ipc_rcu_putref(void *ptr)
594 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
597 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
598 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
601 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
607 * ipcperms - check IPC permissions
608 * @ipcp: IPC permission set
609 * @flag: desired permission set.
611 * Check user, group, other permissions for access
612 * to ipc resources. return 0 if allowed
615 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
616 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
617 uid_t euid = current_euid();
618 int requested_mode, granted_mode;
621 requested_mode = (flag >> 6) | (flag >> 3) | flag;
622 granted_mode = ipcp->mode;
623 if (euid == ipcp->cuid ||
626 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
628 /* is there some bit set in requested_mode but not in granted_mode? */
629 if ((requested_mode & ~granted_mode & 0007) &&
630 !capable(CAP_IPC_OWNER))
633 return security_ipc_permission(ipcp, flag);
637 * Functions to convert between the kern_ipc_perm structure and the
638 * old/new ipc_perm structures
642 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
643 * @in: kernel permissions
644 * @out: new style IPC permissions
646 * Turn the kernel object @in into a set of permissions descriptions
647 * for returning to userspace (@out).
651 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
656 out->cuid = in->cuid;
657 out->cgid = in->cgid;
658 out->mode = in->mode;
663 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
664 * @in: new style IPC permissions
665 * @out: old style IPC permissions
667 * Turn the new style permissions object @in into a compatibility
668 * object and store it into the @out pointer.
671 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
674 SET_UID(out->uid, in->uid);
675 SET_GID(out->gid, in->gid);
676 SET_UID(out->cuid, in->cuid);
677 SET_GID(out->cgid, in->cgid);
678 out->mode = in->mode;
683 * ipc_lock - Lock an ipc structure without rw_mutex held
684 * @ids: IPC identifier set
685 * @id: ipc id to look for
687 * Look for an id in the ipc ids idr and lock the associated ipc object.
689 * The ipc object is locked on exit.
692 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
694 struct kern_ipc_perm *out;
695 int lid = ipcid_to_idx(id);
698 out = idr_find(&ids->ipcs_idr, lid);
701 return ERR_PTR(-EINVAL);
704 spin_lock(&out->lock);
706 /* ipc_rmid() may have already freed the ID while ipc_lock
707 * was spinning: here verify that the structure is still valid
710 spin_unlock(&out->lock);
712 return ERR_PTR(-EINVAL);
718 struct kern_ipc_perm *ipc_lock_check(struct ipc_ids *ids, int id)
720 struct kern_ipc_perm *out;
722 out = ipc_lock(ids, id);
726 if (ipc_checkid(out, id)) {
728 return ERR_PTR(-EIDRM);
735 * ipcget - Common sys_*get() code
737 * @ids : IPC identifier set
738 * @ops : operations to be called on ipc object creation, permission checks
740 * @params : the parameters needed by the previous operations.
742 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
744 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
745 struct ipc_ops *ops, struct ipc_params *params)
747 if (params->key == IPC_PRIVATE)
748 return ipcget_new(ns, ids, ops, params);
750 return ipcget_public(ns, ids, ops, params);
754 * ipc_update_perm - update the permissions of an IPC.
755 * @in: the permission given as input.
756 * @out: the permission of the ipc to set.
758 void ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
762 out->mode = (out->mode & ~S_IRWXUGO)
763 | (in->mode & S_IRWXUGO);
767 * ipcctl_pre_down - retrieve an ipc and check permissions for some IPC_XXX cmd
768 * @ids: the table of ids where to look for the ipc
769 * @id: the id of the ipc to retrieve
770 * @cmd: the cmd to check
771 * @perm: the permission to set
772 * @extra_perm: one extra permission parameter used by msq
774 * This function does some common audit and permissions check for some IPC_XXX
775 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
776 * It must be called without any lock held and
777 * - retrieves the ipc with the given id in the given table.
778 * - performs some audit and permission check, depending on the given cmd
779 * - returns the ipc with both ipc and rw_mutex locks held in case of success
780 * or an err-code without any lock held otherwise.
782 struct kern_ipc_perm *ipcctl_pre_down(struct ipc_ids *ids, int id, int cmd,
783 struct ipc64_perm *perm, int extra_perm)
785 struct kern_ipc_perm *ipcp;
789 down_write(&ids->rw_mutex);
790 ipcp = ipc_lock_check(ids, id);
798 audit_ipc_set_perm(extra_perm, perm->uid,
799 perm->gid, perm->mode);
801 euid = current_euid();
802 if (euid == ipcp->cuid ||
803 euid == ipcp->uid || capable(CAP_SYS_ADMIN))
809 up_write(&ids->rw_mutex);
813 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
817 * ipc_parse_version - IPC call version
818 * @cmd: pointer to command
820 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
821 * The @cmd value is turned from an encoding command and version into
822 * just the command code.
825 int ipc_parse_version (int *cmd)
835 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
837 #ifdef CONFIG_PROC_FS
838 struct ipc_proc_iter {
839 struct ipc_namespace *ns;
840 struct ipc_proc_iface *iface;
844 * This routine locks the ipc structure found at least at position pos.
846 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
849 struct kern_ipc_perm *ipc;
853 for (id = 0; id < pos && total < ids->in_use; id++) {
854 ipc = idr_find(&ids->ipcs_idr, id);
859 if (total >= ids->in_use)
862 for ( ; pos < IPCMNI; pos++) {
863 ipc = idr_find(&ids->ipcs_idr, pos);
866 ipc_lock_by_ptr(ipc);
871 /* Out of range - return NULL to terminate iteration */
875 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
877 struct ipc_proc_iter *iter = s->private;
878 struct ipc_proc_iface *iface = iter->iface;
879 struct kern_ipc_perm *ipc = it;
881 /* If we had an ipc id locked before, unlock it */
882 if (ipc && ipc != SEQ_START_TOKEN)
885 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
889 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
890 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
892 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
894 struct ipc_proc_iter *iter = s->private;
895 struct ipc_proc_iface *iface = iter->iface;
898 ids = &iter->ns->ids[iface->ids];
901 * Take the lock - this will be released by the corresponding
904 down_read(&ids->rw_mutex);
906 /* pos < 0 is invalid */
910 /* pos == 0 means header */
912 return SEQ_START_TOKEN;
914 /* Find the (pos-1)th ipc */
915 return sysvipc_find_ipc(ids, *pos - 1, pos);
918 static void sysvipc_proc_stop(struct seq_file *s, void *it)
920 struct kern_ipc_perm *ipc = it;
921 struct ipc_proc_iter *iter = s->private;
922 struct ipc_proc_iface *iface = iter->iface;
925 /* If we had a locked structure, release it */
926 if (ipc && ipc != SEQ_START_TOKEN)
929 ids = &iter->ns->ids[iface->ids];
930 /* Release the lock we took in start() */
931 up_read(&ids->rw_mutex);
934 static int sysvipc_proc_show(struct seq_file *s, void *it)
936 struct ipc_proc_iter *iter = s->private;
937 struct ipc_proc_iface *iface = iter->iface;
939 if (it == SEQ_START_TOKEN)
940 return seq_puts(s, iface->header);
942 return iface->show(s, it);
945 static const struct seq_operations sysvipc_proc_seqops = {
946 .start = sysvipc_proc_start,
947 .stop = sysvipc_proc_stop,
948 .next = sysvipc_proc_next,
949 .show = sysvipc_proc_show,
952 static int sysvipc_proc_open(struct inode *inode, struct file *file)
955 struct seq_file *seq;
956 struct ipc_proc_iter *iter;
959 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
963 ret = seq_open(file, &sysvipc_proc_seqops);
967 seq = file->private_data;
970 iter->iface = PDE(inode)->data;
971 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
979 static int sysvipc_proc_release(struct inode *inode, struct file *file)
981 struct seq_file *seq = file->private_data;
982 struct ipc_proc_iter *iter = seq->private;
983 put_ipc_ns(iter->ns);
984 return seq_release_private(inode, file);
987 static const struct file_operations sysvipc_proc_fops = {
988 .open = sysvipc_proc_open,
991 .release = sysvipc_proc_release,
993 #endif /* CONFIG_PROC_FS */