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/smp_lock.h>
25 #include <linux/vmalloc.h>
26 #include <linux/slab.h>
27 #include <linux/capability.h>
28 #include <linux/highuid.h>
29 #include <linux/security.h>
30 #include <linux/rcupdate.h>
31 #include <linux/workqueue.h>
32 #include <linux/seq_file.h>
33 #include <linux/proc_fs.h>
34 #include <linux/audit.h>
35 #include <linux/nsproxy.h>
37 #include <asm/unistd.h>
41 struct ipc_proc_iface {
45 int (*show)(struct seq_file *, void *);
48 struct ipc_namespace init_ipc_ns = {
50 .refcount = ATOMIC_INIT(2),
55 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
58 struct ipc_namespace *ns;
61 ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
65 err = sem_init_ns(ns);
68 err = msg_init_ns(ns);
71 err = shm_init_ns(ns);
88 int unshare_ipcs(unsigned long unshare_flags, struct ipc_namespace **new_ipc)
90 struct ipc_namespace *new;
92 if (unshare_flags & CLONE_NEWIPC) {
93 if (!capable(CAP_SYS_ADMIN))
96 new = clone_ipc_ns(current->nsproxy->ipc_ns);
106 int copy_ipcs(unsigned long flags, struct task_struct *tsk)
108 struct ipc_namespace *old_ns = tsk->nsproxy->ipc_ns;
109 struct ipc_namespace *new_ns;
117 if (!(flags & CLONE_NEWIPC))
120 if (!capable(CAP_SYS_ADMIN)) {
125 new_ns = clone_ipc_ns(old_ns);
131 tsk->nsproxy->ipc_ns = new_ns;
137 void free_ipc_ns(struct kref *kref)
139 struct ipc_namespace *ns;
141 ns = container_of(kref, struct ipc_namespace, kref);
150 * ipc_init - initialise IPC subsystem
152 * The various system5 IPC resources (semaphores, messages and shared
153 * memory are initialised
156 static int __init ipc_init(void)
163 __initcall(ipc_init);
166 * ipc_init_ids - initialise IPC identifiers
167 * @ids: Identifier set
168 * @size: Number of identifiers
170 * Given a size for the ipc identifier range (limited below IPCMNI)
171 * set up the sequence range to use then allocate and initialise the
175 void __ipc_init ipc_init_ids(struct ipc_ids* ids, int size)
179 mutex_init(&ids->mutex);
187 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
188 if(seq_limit > USHRT_MAX)
189 ids->seq_max = USHRT_MAX;
191 ids->seq_max = seq_limit;
194 ids->entries = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*size +
195 sizeof(struct ipc_id_ary));
197 if(ids->entries == NULL) {
198 printk(KERN_ERR "ipc_init_ids() failed, ipc service disabled.\n");
200 ids->entries = &ids->nullentry;
202 ids->entries->size = size;
204 ids->entries->p[i] = NULL;
207 #ifdef CONFIG_PROC_FS
208 static struct file_operations sysvipc_proc_fops;
210 * ipc_init_proc_interface - Create a proc interface for sysipc types
211 * using a seq_file interface.
212 * @path: Path in procfs
213 * @header: Banner to be printed at the beginning of the file.
214 * @ids: ipc id table to iterate.
215 * @show: show routine.
217 void __init ipc_init_proc_interface(const char *path, const char *header,
218 int ids, int (*show)(struct seq_file *, void *))
220 struct proc_dir_entry *pde;
221 struct ipc_proc_iface *iface;
223 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
227 iface->header = header;
231 pde = create_proc_entry(path,
232 S_IRUGO, /* world readable */
233 NULL /* parent dir */);
236 pde->proc_fops = &sysvipc_proc_fops;
244 * ipc_findkey - find a key in an ipc identifier set
245 * @ids: Identifier set
246 * @key: The key to find
248 * Requires ipc_ids.mutex locked.
249 * Returns the identifier if found or -1 if not.
252 int ipc_findkey(struct ipc_ids* ids, key_t key)
255 struct kern_ipc_perm* p;
256 int max_id = ids->max_id;
259 * rcu_dereference() is not needed here
260 * since ipc_ids.mutex is held
262 for (id = 0; id <= max_id; id++) {
263 p = ids->entries->p[id];
273 * Requires ipc_ids.mutex locked
275 static int grow_ary(struct ipc_ids* ids, int newsize)
277 struct ipc_id_ary* new;
278 struct ipc_id_ary* old;
280 int size = ids->entries->size;
287 new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*newsize +
288 sizeof(struct ipc_id_ary));
292 memcpy(new->p, ids->entries->p, sizeof(struct kern_ipc_perm *)*size);
293 for(i=size;i<newsize;i++) {
299 * Use rcu_assign_pointer() to make sure the memcpyed contents
300 * of the new array are visible before the new array becomes visible.
302 rcu_assign_pointer(ids->entries, new);
309 * ipc_addid - add an IPC identifier
310 * @ids: IPC identifier set
311 * @new: new IPC permission set
312 * @size: new size limit for the id array
314 * Add an entry 'new' to the IPC arrays. The permissions object is
315 * initialised and the first free entry is set up and the id assigned
316 * is returned. The list is returned in a locked state on success.
317 * On failure the list is not locked and -1 is returned.
319 * Called with ipc_ids.mutex held.
322 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
326 size = grow_ary(ids,size);
329 * rcu_dereference()() is not needed here since
330 * ipc_ids.mutex is held
332 for (id = 0; id < size; id++) {
333 if(ids->entries->p[id] == NULL)
339 if (id > ids->max_id)
342 new->cuid = new->uid = current->euid;
343 new->gid = new->cgid = current->egid;
345 new->seq = ids->seq++;
346 if(ids->seq > ids->seq_max)
349 spin_lock_init(&new->lock);
352 spin_lock(&new->lock);
353 ids->entries->p[id] = new;
358 * ipc_rmid - remove an IPC identifier
359 * @ids: identifier set
360 * @id: Identifier to remove
362 * The identifier must be valid, and in use. The kernel will panic if
363 * fed an invalid identifier. The entry is removed and internal
364 * variables recomputed. The object associated with the identifier
366 * ipc_ids.mutex and the spinlock for this ID is hold before this function
367 * is called, and remain locked on the exit.
370 struct kern_ipc_perm* ipc_rmid(struct ipc_ids* ids, int id)
372 struct kern_ipc_perm* p;
373 int lid = id % SEQ_MULTIPLIER;
374 BUG_ON(lid >= ids->entries->size);
377 * do not need a rcu_dereference()() here to force ordering
378 * on Alpha, since the ipc_ids.mutex is held.
380 p = ids->entries->p[lid];
381 ids->entries->p[lid] = NULL;
385 if (lid == ids->max_id) {
390 } while (ids->entries->p[lid] == NULL);
398 * ipc_alloc - allocate ipc space
399 * @size: size desired
401 * Allocate memory from the appropriate pools and return a pointer to it.
402 * NULL is returned if the allocation fails
405 void* ipc_alloc(int size)
411 out = kmalloc(size, GFP_KERNEL);
416 * ipc_free - free ipc space
417 * @ptr: pointer returned by ipc_alloc
418 * @size: size of block
420 * Free a block created with ipc_alloc. The caller must know the size
421 * used in the allocation call.
424 void ipc_free(void* ptr, int size)
434 * There are three headers that are prepended to the actual allocation:
435 * - during use: ipc_rcu_hdr.
436 * - during the rcu grace period: ipc_rcu_grace.
437 * - [only if vmalloc]: ipc_rcu_sched.
438 * Their lifetime doesn't overlap, thus the headers share the same memory.
439 * Unlike a normal union, they are right-aligned, thus some container_of
440 * forward/backward casting is necessary:
453 /* "void *" makes sure alignment of following data is sane. */
459 struct work_struct work;
460 /* "void *" makes sure alignment of following data is sane. */
464 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
465 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
466 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
467 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
469 static inline int rcu_use_vmalloc(int size)
471 /* Too big for a single page? */
472 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
478 * ipc_rcu_alloc - allocate ipc and rcu space
479 * @size: size desired
481 * Allocate memory for the rcu header structure + the object.
482 * Returns the pointer to the object.
483 * NULL is returned if the allocation fails.
486 void* ipc_rcu_alloc(int size)
490 * We prepend the allocation with the rcu struct, and
491 * workqueue if necessary (for vmalloc).
493 if (rcu_use_vmalloc(size)) {
494 out = vmalloc(HDRLEN_VMALLOC + size);
496 out += HDRLEN_VMALLOC;
497 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
498 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
501 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
503 out += HDRLEN_KMALLOC;
504 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
505 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
512 void ipc_rcu_getref(void *ptr)
514 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
518 * ipc_schedule_free - free ipc + rcu space
519 * @head: RCU callback structure for queued work
521 * Since RCU callback function is called in bh,
522 * we need to defer the vfree to schedule_work
524 static void ipc_schedule_free(struct rcu_head *head)
526 struct ipc_rcu_grace *grace =
527 container_of(head, struct ipc_rcu_grace, rcu);
528 struct ipc_rcu_sched *sched =
529 container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
531 INIT_WORK(&sched->work, vfree, sched);
532 schedule_work(&sched->work);
536 * ipc_immediate_free - free ipc + rcu space
537 * @head: RCU callback structure that contains pointer to be freed
539 * Free from the RCU callback context
541 static void ipc_immediate_free(struct rcu_head *head)
543 struct ipc_rcu_grace *free =
544 container_of(head, struct ipc_rcu_grace, rcu);
548 void ipc_rcu_putref(void *ptr)
550 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
553 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
554 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
557 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
563 * ipcperms - check IPC permissions
564 * @ipcp: IPC permission set
565 * @flag: desired permission set.
567 * Check user, group, other permissions for access
568 * to ipc resources. return 0 if allowed
571 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
572 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
573 int requested_mode, granted_mode, err;
575 if (unlikely((err = audit_ipc_obj(ipcp))))
577 requested_mode = (flag >> 6) | (flag >> 3) | flag;
578 granted_mode = ipcp->mode;
579 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
581 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
583 /* is there some bit set in requested_mode but not in granted_mode? */
584 if ((requested_mode & ~granted_mode & 0007) &&
585 !capable(CAP_IPC_OWNER))
588 return security_ipc_permission(ipcp, flag);
592 * Functions to convert between the kern_ipc_perm structure and the
593 * old/new ipc_perm structures
597 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
598 * @in: kernel permissions
599 * @out: new style IPC permissions
601 * Turn the kernel object 'in' into a set of permissions descriptions
602 * for returning to userspace (out).
606 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
611 out->cuid = in->cuid;
612 out->cgid = in->cgid;
613 out->mode = in->mode;
618 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
619 * @in: new style IPC permissions
620 * @out: old style IPC permissions
622 * Turn the new style permissions object in into a compatibility
623 * object and store it into the 'out' pointer.
626 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
629 SET_UID(out->uid, in->uid);
630 SET_GID(out->gid, in->gid);
631 SET_UID(out->cuid, in->cuid);
632 SET_GID(out->cgid, in->cgid);
633 out->mode = in->mode;
638 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
639 * is called with shm_ids.mutex locked. Since grow_ary() is also called with
640 * shm_ids.mutex down(for Shared Memory), there is no need to add read
641 * barriers here to gurantee the writes in grow_ary() are seen in order
644 * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
645 * if in the future ipc_get() is used by other places without ipc_ids.mutex
646 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
648 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
650 struct kern_ipc_perm* out;
651 int lid = id % SEQ_MULTIPLIER;
652 if(lid >= ids->entries->size)
654 out = ids->entries->p[lid];
658 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
660 struct kern_ipc_perm* out;
661 int lid = id % SEQ_MULTIPLIER;
662 struct ipc_id_ary* entries;
665 entries = rcu_dereference(ids->entries);
666 if(lid >= entries->size) {
670 out = entries->p[lid];
675 spin_lock(&out->lock);
677 /* ipc_rmid() may have already freed the ID while ipc_lock
678 * was spinning: here verify that the structure is still valid
681 spin_unlock(&out->lock);
688 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
691 spin_lock(&perm->lock);
694 void ipc_unlock(struct kern_ipc_perm* perm)
696 spin_unlock(&perm->lock);
700 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
702 return SEQ_MULTIPLIER*seq + id;
705 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
707 if(uid/SEQ_MULTIPLIER != ipcp->seq)
712 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
716 * ipc_parse_version - IPC call version
717 * @cmd: pointer to command
719 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
720 * The cmd value is turned from an encoding command and version into
721 * just the command code.
724 int ipc_parse_version (int *cmd)
734 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
736 #ifdef CONFIG_PROC_FS
737 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
739 struct ipc_proc_iface *iface = s->private;
740 struct kern_ipc_perm *ipc = it;
744 ids = current->nsproxy->ipc_ns->ids[iface->ids];
746 /* If we had an ipc id locked before, unlock it */
747 if (ipc && ipc != SEQ_START_TOKEN)
751 * p = *pos - 1 (because id 0 starts at position 1)
752 * + 1 (because we increment the position by one)
754 for (p = *pos; p <= ids->max_id; p++) {
755 if ((ipc = ipc_lock(ids, p)) != NULL) {
761 /* Out of range - return NULL to terminate iteration */
766 * File positions: pos 0 -> header, pos n -> ipc id + 1.
767 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
769 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
771 struct ipc_proc_iface *iface = s->private;
772 struct kern_ipc_perm *ipc;
776 ids = current->nsproxy->ipc_ns->ids[iface->ids];
779 * Take the lock - this will be released by the corresponding
782 mutex_lock(&ids->mutex);
784 /* pos < 0 is invalid */
788 /* pos == 0 means header */
790 return SEQ_START_TOKEN;
792 /* Find the (pos-1)th ipc */
793 for (p = *pos - 1; p <= ids->max_id; p++) {
794 if ((ipc = ipc_lock(ids, p)) != NULL) {
802 static void sysvipc_proc_stop(struct seq_file *s, void *it)
804 struct kern_ipc_perm *ipc = it;
805 struct ipc_proc_iface *iface = s->private;
808 /* If we had a locked segment, release it */
809 if (ipc && ipc != SEQ_START_TOKEN)
812 ids = current->nsproxy->ipc_ns->ids[iface->ids];
813 /* Release the lock we took in start() */
814 mutex_unlock(&ids->mutex);
817 static int sysvipc_proc_show(struct seq_file *s, void *it)
819 struct ipc_proc_iface *iface = s->private;
821 if (it == SEQ_START_TOKEN)
822 return seq_puts(s, iface->header);
824 return iface->show(s, it);
827 static 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 seq_file *seq;
838 ret = seq_open(file, &sysvipc_proc_seqops);
840 seq = file->private_data;
841 seq->private = PDE(inode)->data;
846 static struct file_operations sysvipc_proc_fops = {
847 .open = sysvipc_proc_open,
850 .release = seq_release,
852 #endif /* CONFIG_PROC_FS */