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>
36 #include <asm/unistd.h>
40 struct ipc_proc_iface {
44 int (*show)(struct seq_file *, void *);
47 struct ipc_namespace init_ipc_ns = {
49 .refcount = ATOMIC_INIT(2),
53 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
56 struct ipc_namespace *ns;
59 ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
63 err = sem_init_ns(ns);
66 err = msg_init_ns(ns);
69 err = shm_init_ns(ns);
86 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
88 struct ipc_namespace *new_ns;
93 if (!(flags & CLONE_NEWIPC))
96 new_ns = clone_ipc_ns(ns);
102 void free_ipc_ns(struct kref *kref)
104 struct ipc_namespace *ns;
106 ns = container_of(kref, struct ipc_namespace, kref);
114 * ipc_init - initialise IPC subsystem
116 * The various system5 IPC resources (semaphores, messages and shared
117 * memory) are initialised
120 static int __init ipc_init(void)
127 __initcall(ipc_init);
130 * ipc_init_ids - initialise IPC identifiers
131 * @ids: Identifier set
133 * Set up the sequence range to use for the ipc identifier range (limited
134 * below IPCMNI) then initialise the ids idr.
137 void ipc_init_ids(struct ipc_ids *ids)
139 mutex_init(&ids->mutex);
144 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
145 if(seq_limit > USHRT_MAX)
146 ids->seq_max = USHRT_MAX;
148 ids->seq_max = seq_limit;
151 idr_init(&ids->ipcs_idr);
154 #ifdef CONFIG_PROC_FS
155 static const struct file_operations sysvipc_proc_fops;
157 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
158 * @path: Path in procfs
159 * @header: Banner to be printed at the beginning of the file.
160 * @ids: ipc id table to iterate.
161 * @show: show routine.
163 void __init ipc_init_proc_interface(const char *path, const char *header,
164 int ids, int (*show)(struct seq_file *, void *))
166 struct proc_dir_entry *pde;
167 struct ipc_proc_iface *iface;
169 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
173 iface->header = header;
177 pde = create_proc_entry(path,
178 S_IRUGO, /* world readable */
179 NULL /* parent dir */);
182 pde->proc_fops = &sysvipc_proc_fops;
190 * ipc_findkey - find a key in an ipc identifier set
191 * @ids: Identifier set
192 * @key: The key to find
194 * Requires ipc_ids.mutex locked.
195 * Returns the LOCKED pointer to the ipc structure if found or NULL
197 * If key is found ipc contains its ipc structure
200 struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
202 struct kern_ipc_perm *ipc;
206 for (total = 0, next_id = 0; total < ids->in_use; next_id++) {
207 ipc = idr_find(&ids->ipcs_idr, next_id);
212 if (ipc->key != key) {
217 ipc_lock_by_ptr(ipc);
225 * ipc_get_maxid - get the last assigned id
226 * @ids: IPC identifier set
228 * Called with ipc_ids.mutex held.
231 int ipc_get_maxid(struct ipc_ids *ids)
233 struct kern_ipc_perm *ipc;
237 if (ids->in_use == 0)
240 if (ids->in_use == IPCMNI)
243 /* Look for the last assigned id */
245 for (id = 0; id < IPCMNI && total < ids->in_use; id++) {
246 ipc = idr_find(&ids->ipcs_idr, id);
256 * ipc_addid - add an IPC identifier
257 * @ids: IPC identifier set
258 * @new: new IPC permission set
259 * @size: limit for the number of used ids
261 * Add an entry 'new' to the IPC idr. The permissions object is
262 * initialised and the first free entry is set up and the id assigned
263 * is returned. The list is returned in a locked state on success.
264 * On failure the list is not locked and -1 is returned.
266 * Called with ipc_ids.mutex held.
269 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
274 * rcu_dereference()() is not needed here since
275 * ipc_ids.mutex is held
280 if (ids->in_use >= size)
283 err = idr_get_new(&ids->ipcs_idr, new, &id);
289 new->cuid = new->uid = current->euid;
290 new->gid = new->cgid = current->egid;
292 new->seq = ids->seq++;
293 if(ids->seq > ids->seq_max)
296 spin_lock_init(&new->lock);
299 spin_lock(&new->lock);
304 * ipc_rmid - remove an IPC identifier
305 * @ids: identifier set
306 * @id: ipc perm structure containing the identifier to remove
308 * The identifier must be valid, and in use. The kernel will panic if
309 * fed an invalid identifier. The entry is removed and internal
310 * variables recomputed.
311 * ipc_ids.mutex and the spinlock for this ID are held before this
312 * function is called, and remain locked on the exit.
315 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
317 int lid = ipcp->id % SEQ_MULTIPLIER;
319 idr_remove(&ids->ipcs_idr, lid);
329 * ipc_alloc - allocate ipc space
330 * @size: size desired
332 * Allocate memory from the appropriate pools and return a pointer to it.
333 * NULL is returned if the allocation fails
336 void* ipc_alloc(int size)
342 out = kmalloc(size, GFP_KERNEL);
347 * ipc_free - free ipc space
348 * @ptr: pointer returned by ipc_alloc
349 * @size: size of block
351 * Free a block created with ipc_alloc(). The caller must know the size
352 * used in the allocation call.
355 void ipc_free(void* ptr, int size)
365 * There are three headers that are prepended to the actual allocation:
366 * - during use: ipc_rcu_hdr.
367 * - during the rcu grace period: ipc_rcu_grace.
368 * - [only if vmalloc]: ipc_rcu_sched.
369 * Their lifetime doesn't overlap, thus the headers share the same memory.
370 * Unlike a normal union, they are right-aligned, thus some container_of
371 * forward/backward casting is necessary:
384 /* "void *" makes sure alignment of following data is sane. */
390 struct work_struct work;
391 /* "void *" makes sure alignment of following data is sane. */
395 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
396 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
397 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
398 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
400 static inline int rcu_use_vmalloc(int size)
402 /* Too big for a single page? */
403 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
409 * ipc_rcu_alloc - allocate ipc and rcu space
410 * @size: size desired
412 * Allocate memory for the rcu header structure + the object.
413 * Returns the pointer to the object.
414 * NULL is returned if the allocation fails.
417 void* ipc_rcu_alloc(int size)
421 * We prepend the allocation with the rcu struct, and
422 * workqueue if necessary (for vmalloc).
424 if (rcu_use_vmalloc(size)) {
425 out = vmalloc(HDRLEN_VMALLOC + size);
427 out += HDRLEN_VMALLOC;
428 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
429 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
432 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
434 out += HDRLEN_KMALLOC;
435 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
436 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
443 void ipc_rcu_getref(void *ptr)
445 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
448 static void ipc_do_vfree(struct work_struct *work)
450 vfree(container_of(work, struct ipc_rcu_sched, work));
454 * ipc_schedule_free - free ipc + rcu space
455 * @head: RCU callback structure for queued work
457 * Since RCU callback function is called in bh,
458 * we need to defer the vfree to schedule_work().
460 static void ipc_schedule_free(struct rcu_head *head)
462 struct ipc_rcu_grace *grace =
463 container_of(head, struct ipc_rcu_grace, rcu);
464 struct ipc_rcu_sched *sched =
465 container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
467 INIT_WORK(&sched->work, ipc_do_vfree);
468 schedule_work(&sched->work);
472 * ipc_immediate_free - free ipc + rcu space
473 * @head: RCU callback structure that contains pointer to be freed
475 * Free from the RCU callback context.
477 static void ipc_immediate_free(struct rcu_head *head)
479 struct ipc_rcu_grace *free =
480 container_of(head, struct ipc_rcu_grace, rcu);
484 void ipc_rcu_putref(void *ptr)
486 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
489 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
490 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
493 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
499 * ipcperms - check IPC permissions
500 * @ipcp: IPC permission set
501 * @flag: desired permission set.
503 * Check user, group, other permissions for access
504 * to ipc resources. return 0 if allowed
507 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
508 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
509 int requested_mode, granted_mode, err;
511 if (unlikely((err = audit_ipc_obj(ipcp))))
513 requested_mode = (flag >> 6) | (flag >> 3) | flag;
514 granted_mode = ipcp->mode;
515 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
517 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
519 /* is there some bit set in requested_mode but not in granted_mode? */
520 if ((requested_mode & ~granted_mode & 0007) &&
521 !capable(CAP_IPC_OWNER))
524 return security_ipc_permission(ipcp, flag);
528 * Functions to convert between the kern_ipc_perm structure and the
529 * old/new ipc_perm structures
533 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
534 * @in: kernel permissions
535 * @out: new style IPC permissions
537 * Turn the kernel object @in into a set of permissions descriptions
538 * for returning to userspace (@out).
542 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
547 out->cuid = in->cuid;
548 out->cgid = in->cgid;
549 out->mode = in->mode;
554 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
555 * @in: new style IPC permissions
556 * @out: old style IPC permissions
558 * Turn the new style permissions object @in into a compatibility
559 * object and store it into the @out pointer.
562 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
565 SET_UID(out->uid, in->uid);
566 SET_GID(out->gid, in->gid);
567 SET_UID(out->cuid, in->cuid);
568 SET_GID(out->cgid, in->cgid);
569 out->mode = in->mode;
574 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
575 * is called with shm_ids.mutex locked. Since grow_ary() is also called with
576 * shm_ids.mutex down(for Shared Memory), there is no need to add read
577 * barriers here to gurantee the writes in grow_ary() are seen in order
580 * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
581 * if in the future ipc_get() is used by other places without ipc_ids.mutex
582 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
584 struct kern_ipc_perm *ipc_get(struct ipc_ids *ids, int id)
586 struct kern_ipc_perm *out;
587 int lid = id % SEQ_MULTIPLIER;
588 out = idr_find(&ids->ipcs_idr, lid);
592 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
594 struct kern_ipc_perm *out;
595 int lid = id % SEQ_MULTIPLIER;
598 out = idr_find(&ids->ipcs_idr, lid);
604 spin_lock(&out->lock);
606 /* ipc_rmid() may have already freed the ID while ipc_lock
607 * was spinning: here verify that the structure is still valid
610 spin_unlock(&out->lock);
618 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
621 spin_lock(&perm->lock);
624 void ipc_unlock(struct kern_ipc_perm* perm)
626 spin_unlock(&perm->lock);
630 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
632 return SEQ_MULTIPLIER*seq + id;
635 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
637 if(uid/SEQ_MULTIPLIER != ipcp->seq)
642 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
646 * ipc_parse_version - IPC call version
647 * @cmd: pointer to command
649 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
650 * The @cmd value is turned from an encoding command and version into
651 * just the command code.
654 int ipc_parse_version (int *cmd)
664 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
666 #ifdef CONFIG_PROC_FS
667 struct ipc_proc_iter {
668 struct ipc_namespace *ns;
669 struct ipc_proc_iface *iface;
673 * This routine locks the ipc structure found at least at position pos.
675 struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
678 struct kern_ipc_perm *ipc;
682 for (id = 0; id < pos && total < ids->in_use; id++) {
683 ipc = idr_find(&ids->ipcs_idr, id);
688 if (total >= ids->in_use)
691 for ( ; pos < IPCMNI; pos++) {
692 ipc = idr_find(&ids->ipcs_idr, pos);
695 ipc_lock_by_ptr(ipc);
700 /* Out of range - return NULL to terminate iteration */
704 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
706 struct ipc_proc_iter *iter = s->private;
707 struct ipc_proc_iface *iface = iter->iface;
708 struct kern_ipc_perm *ipc = it;
710 /* If we had an ipc id locked before, unlock it */
711 if (ipc && ipc != SEQ_START_TOKEN)
714 return sysvipc_find_ipc(iter->ns->ids[iface->ids], *pos, pos);
718 * File positions: pos 0 -> header, pos n -> ipc id + 1.
719 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
721 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
723 struct ipc_proc_iter *iter = s->private;
724 struct ipc_proc_iface *iface = iter->iface;
727 ids = iter->ns->ids[iface->ids];
730 * Take the lock - this will be released by the corresponding
733 mutex_lock(&ids->mutex);
735 /* pos < 0 is invalid */
739 /* pos == 0 means header */
741 return SEQ_START_TOKEN;
743 /* Find the (pos-1)th ipc */
744 return sysvipc_find_ipc(ids, *pos - 1, pos);
747 static void sysvipc_proc_stop(struct seq_file *s, void *it)
749 struct kern_ipc_perm *ipc = it;
750 struct ipc_proc_iter *iter = s->private;
751 struct ipc_proc_iface *iface = iter->iface;
754 /* If we had a locked segment, release it */
755 if (ipc && ipc != SEQ_START_TOKEN)
758 ids = iter->ns->ids[iface->ids];
759 /* Release the lock we took in start() */
760 mutex_unlock(&ids->mutex);
763 static int sysvipc_proc_show(struct seq_file *s, void *it)
765 struct ipc_proc_iter *iter = s->private;
766 struct ipc_proc_iface *iface = iter->iface;
768 if (it == SEQ_START_TOKEN)
769 return seq_puts(s, iface->header);
771 return iface->show(s, it);
774 static struct seq_operations sysvipc_proc_seqops = {
775 .start = sysvipc_proc_start,
776 .stop = sysvipc_proc_stop,
777 .next = sysvipc_proc_next,
778 .show = sysvipc_proc_show,
781 static int sysvipc_proc_open(struct inode *inode, struct file *file)
784 struct seq_file *seq;
785 struct ipc_proc_iter *iter;
788 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
792 ret = seq_open(file, &sysvipc_proc_seqops);
796 seq = file->private_data;
799 iter->iface = PDE(inode)->data;
800 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
808 static int sysvipc_proc_release(struct inode *inode, struct file *file)
810 struct seq_file *seq = file->private_data;
811 struct ipc_proc_iter *iter = seq->private;
812 put_ipc_ns(iter->ns);
813 return seq_release_private(inode, file);
816 static const struct file_operations sysvipc_proc_fops = {
817 .open = sysvipc_proc_open,
820 .release = sysvipc_proc_release,
822 #endif /* CONFIG_PROC_FS */