[NETFILTER]: New connection tracking is not EXPERIMENTAL anymore
[linux-2.6/mini2440.git] / ipc / util.c
bloba9b7a227b8d4c7bde79998048206b96fef1f57ed
1 /*
2 * linux/ipc/util.c
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
16 * OpenVZ, SWsoft Inc.
17 * Pavel Emelianov <xemul@openvz.org>
20 #include <linux/mm.h>
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>
39 #include "util.h"
41 struct ipc_proc_iface {
42 const char *path;
43 const char *header;
44 int ids;
45 int (*show)(struct seq_file *, void *);
48 struct ipc_namespace init_ipc_ns = {
49 .kref = {
50 .refcount = ATOMIC_INIT(2),
54 #ifdef CONFIG_IPC_NS
55 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
57 int err;
58 struct ipc_namespace *ns;
60 err = -ENOMEM;
61 ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
62 if (ns == NULL)
63 goto err_mem;
65 err = sem_init_ns(ns);
66 if (err)
67 goto err_sem;
68 err = msg_init_ns(ns);
69 if (err)
70 goto err_msg;
71 err = shm_init_ns(ns);
72 if (err)
73 goto err_shm;
75 kref_init(&ns->kref);
76 return ns;
78 err_shm:
79 msg_exit_ns(ns);
80 err_msg:
81 sem_exit_ns(ns);
82 err_sem:
83 kfree(ns);
84 err_mem:
85 return ERR_PTR(err);
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))
94 return -EPERM;
96 new = clone_ipc_ns(current->nsproxy->ipc_ns);
97 if (IS_ERR(new))
98 return PTR_ERR(new);
100 *new_ipc = new;
103 return 0;
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;
110 int err = 0;
112 if (!old_ns)
113 return 0;
115 get_ipc_ns(old_ns);
117 if (!(flags & CLONE_NEWIPC))
118 return 0;
120 if (!capable(CAP_SYS_ADMIN)) {
121 err = -EPERM;
122 goto out;
125 new_ns = clone_ipc_ns(old_ns);
126 if (!new_ns) {
127 err = -ENOMEM;
128 goto out;
131 tsk->nsproxy->ipc_ns = new_ns;
132 out:
133 put_ipc_ns(old_ns);
134 return err;
137 void free_ipc_ns(struct kref *kref)
139 struct ipc_namespace *ns;
141 ns = container_of(kref, struct ipc_namespace, kref);
142 sem_exit_ns(ns);
143 msg_exit_ns(ns);
144 shm_exit_ns(ns);
145 kfree(ns);
147 #endif
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)
158 sem_init();
159 msg_init();
160 shm_init();
161 return 0;
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
172 * array itself.
175 void __ipc_init ipc_init_ids(struct ipc_ids* ids, int size)
177 int i;
179 mutex_init(&ids->mutex);
181 if(size > IPCMNI)
182 size = IPCMNI;
183 ids->in_use = 0;
184 ids->max_id = -1;
185 ids->seq = 0;
187 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
188 if(seq_limit > USHRT_MAX)
189 ids->seq_max = USHRT_MAX;
190 else
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");
199 size = 0;
200 ids->entries = &ids->nullentry;
202 ids->entries->size = size;
203 for(i=0;i<size;i++)
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);
224 if (!iface)
225 return;
226 iface->path = path;
227 iface->header = header;
228 iface->ids = ids;
229 iface->show = show;
231 pde = create_proc_entry(path,
232 S_IRUGO, /* world readable */
233 NULL /* parent dir */);
234 if (pde) {
235 pde->data = iface;
236 pde->proc_fops = &sysvipc_proc_fops;
237 } else {
238 kfree(iface);
241 #endif
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)
254 int id;
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];
264 if(p==NULL)
265 continue;
266 if (key == p->key)
267 return id;
269 return -1;
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;
279 int i;
280 int size = ids->entries->size;
282 if(newsize > IPCMNI)
283 newsize = IPCMNI;
284 if(newsize <= size)
285 return newsize;
287 new = ipc_rcu_alloc(sizeof(struct kern_ipc_perm *)*newsize +
288 sizeof(struct ipc_id_ary));
289 if(new == NULL)
290 return size;
291 new->size = newsize;
292 memcpy(new->p, ids->entries->p, sizeof(struct kern_ipc_perm *)*size);
293 for(i=size;i<newsize;i++) {
294 new->p[i] = NULL;
296 old = ids->entries;
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);
304 __ipc_fini_ids(ids, old);
305 return newsize;
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)
324 int id;
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)
334 goto found;
336 return -1;
337 found:
338 ids->in_use++;
339 if (id > ids->max_id)
340 ids->max_id = 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)
347 ids->seq = 0;
349 spin_lock_init(&new->lock);
350 new->deleted = 0;
351 rcu_read_lock();
352 spin_lock(&new->lock);
353 ids->entries->p[id] = new;
354 return id;
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
365 * is returned.
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;
382 BUG_ON(p==NULL);
383 ids->in_use--;
385 if (lid == ids->max_id) {
386 do {
387 lid--;
388 if(lid == -1)
389 break;
390 } while (ids->entries->p[lid] == NULL);
391 ids->max_id = lid;
393 p->deleted = 1;
394 return p;
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)
407 void* out;
408 if(size > PAGE_SIZE)
409 out = vmalloc(size);
410 else
411 out = kmalloc(size, GFP_KERNEL);
412 return out;
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)
426 if(size > PAGE_SIZE)
427 vfree(ptr);
428 else
429 kfree(ptr);
433 * rcu allocations:
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:
442 struct ipc_rcu_hdr
444 int refcount;
445 int is_vmalloc;
446 void *data[0];
450 struct ipc_rcu_grace
452 struct rcu_head rcu;
453 /* "void *" makes sure alignment of following data is sane. */
454 void *data[0];
457 struct ipc_rcu_sched
459 struct work_struct work;
460 /* "void *" makes sure alignment of following data is sane. */
461 void *data[0];
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)
473 return 1;
474 return 0;
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)
488 void* out;
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);
495 if (out) {
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;
500 } else {
501 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
502 if (out) {
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;
509 return out;
512 void ipc_rcu_getref(void *ptr)
514 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
517 static void ipc_do_vfree(struct work_struct *work)
519 vfree(container_of(work, struct ipc_rcu_sched, work));
523 * ipc_schedule_free - free ipc + rcu space
524 * @head: RCU callback structure for queued work
526 * Since RCU callback function is called in bh,
527 * we need to defer the vfree to schedule_work
529 static void ipc_schedule_free(struct rcu_head *head)
531 struct ipc_rcu_grace *grace =
532 container_of(head, struct ipc_rcu_grace, rcu);
533 struct ipc_rcu_sched *sched =
534 container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
536 INIT_WORK(&sched->work, ipc_do_vfree);
537 schedule_work(&sched->work);
541 * ipc_immediate_free - free ipc + rcu space
542 * @head: RCU callback structure that contains pointer to be freed
544 * Free from the RCU callback context
546 static void ipc_immediate_free(struct rcu_head *head)
548 struct ipc_rcu_grace *free =
549 container_of(head, struct ipc_rcu_grace, rcu);
550 kfree(free);
553 void ipc_rcu_putref(void *ptr)
555 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
556 return;
558 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
559 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
560 ipc_schedule_free);
561 } else {
562 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
563 ipc_immediate_free);
568 * ipcperms - check IPC permissions
569 * @ipcp: IPC permission set
570 * @flag: desired permission set.
572 * Check user, group, other permissions for access
573 * to ipc resources. return 0 if allowed
576 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
577 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
578 int requested_mode, granted_mode, err;
580 if (unlikely((err = audit_ipc_obj(ipcp))))
581 return err;
582 requested_mode = (flag >> 6) | (flag >> 3) | flag;
583 granted_mode = ipcp->mode;
584 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
585 granted_mode >>= 6;
586 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
587 granted_mode >>= 3;
588 /* is there some bit set in requested_mode but not in granted_mode? */
589 if ((requested_mode & ~granted_mode & 0007) &&
590 !capable(CAP_IPC_OWNER))
591 return -1;
593 return security_ipc_permission(ipcp, flag);
597 * Functions to convert between the kern_ipc_perm structure and the
598 * old/new ipc_perm structures
602 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
603 * @in: kernel permissions
604 * @out: new style IPC permissions
606 * Turn the kernel object 'in' into a set of permissions descriptions
607 * for returning to userspace (out).
611 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
613 out->key = in->key;
614 out->uid = in->uid;
615 out->gid = in->gid;
616 out->cuid = in->cuid;
617 out->cgid = in->cgid;
618 out->mode = in->mode;
619 out->seq = in->seq;
623 * ipc64_perm_to_ipc_perm - convert old ipc permissions to new
624 * @in: new style IPC permissions
625 * @out: old style IPC permissions
627 * Turn the new style permissions object in into a compatibility
628 * object and store it into the 'out' pointer.
631 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
633 out->key = in->key;
634 SET_UID(out->uid, in->uid);
635 SET_GID(out->gid, in->gid);
636 SET_UID(out->cuid, in->cuid);
637 SET_GID(out->cgid, in->cgid);
638 out->mode = in->mode;
639 out->seq = in->seq;
643 * So far only shm_get_stat() calls ipc_get() via shm_get(), so ipc_get()
644 * is called with shm_ids.mutex locked. Since grow_ary() is also called with
645 * shm_ids.mutex down(for Shared Memory), there is no need to add read
646 * barriers here to gurantee the writes in grow_ary() are seen in order
647 * here (for Alpha).
649 * However ipc_get() itself does not necessary require ipc_ids.mutex down. So
650 * if in the future ipc_get() is used by other places without ipc_ids.mutex
651 * down, then ipc_get() needs read memery barriers as ipc_lock() does.
653 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id)
655 struct kern_ipc_perm* out;
656 int lid = id % SEQ_MULTIPLIER;
657 if(lid >= ids->entries->size)
658 return NULL;
659 out = ids->entries->p[lid];
660 return out;
663 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
665 struct kern_ipc_perm* out;
666 int lid = id % SEQ_MULTIPLIER;
667 struct ipc_id_ary* entries;
669 rcu_read_lock();
670 entries = rcu_dereference(ids->entries);
671 if(lid >= entries->size) {
672 rcu_read_unlock();
673 return NULL;
675 out = entries->p[lid];
676 if(out == NULL) {
677 rcu_read_unlock();
678 return NULL;
680 spin_lock(&out->lock);
682 /* ipc_rmid() may have already freed the ID while ipc_lock
683 * was spinning: here verify that the structure is still valid
685 if (out->deleted) {
686 spin_unlock(&out->lock);
687 rcu_read_unlock();
688 return NULL;
690 return out;
693 void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
695 rcu_read_lock();
696 spin_lock(&perm->lock);
699 void ipc_unlock(struct kern_ipc_perm* perm)
701 spin_unlock(&perm->lock);
702 rcu_read_unlock();
705 int ipc_buildid(struct ipc_ids* ids, int id, int seq)
707 return SEQ_MULTIPLIER*seq + id;
710 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid)
712 if(uid/SEQ_MULTIPLIER != ipcp->seq)
713 return 1;
714 return 0;
717 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
721 * ipc_parse_version - IPC call version
722 * @cmd: pointer to command
724 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
725 * The cmd value is turned from an encoding command and version into
726 * just the command code.
729 int ipc_parse_version (int *cmd)
731 if (*cmd & IPC_64) {
732 *cmd ^= IPC_64;
733 return IPC_64;
734 } else {
735 return IPC_OLD;
739 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
741 #ifdef CONFIG_PROC_FS
742 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
744 struct ipc_proc_iface *iface = s->private;
745 struct kern_ipc_perm *ipc = it;
746 loff_t p;
747 struct ipc_ids *ids;
749 ids = current->nsproxy->ipc_ns->ids[iface->ids];
751 /* If we had an ipc id locked before, unlock it */
752 if (ipc && ipc != SEQ_START_TOKEN)
753 ipc_unlock(ipc);
756 * p = *pos - 1 (because id 0 starts at position 1)
757 * + 1 (because we increment the position by one)
759 for (p = *pos; p <= ids->max_id; p++) {
760 if ((ipc = ipc_lock(ids, p)) != NULL) {
761 *pos = p + 1;
762 return ipc;
766 /* Out of range - return NULL to terminate iteration */
767 return NULL;
771 * File positions: pos 0 -> header, pos n -> ipc id + 1.
772 * SeqFile iterator: iterator value locked shp or SEQ_TOKEN_START.
774 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
776 struct ipc_proc_iface *iface = s->private;
777 struct kern_ipc_perm *ipc;
778 loff_t p;
779 struct ipc_ids *ids;
781 ids = current->nsproxy->ipc_ns->ids[iface->ids];
784 * Take the lock - this will be released by the corresponding
785 * call to stop().
787 mutex_lock(&ids->mutex);
789 /* pos < 0 is invalid */
790 if (*pos < 0)
791 return NULL;
793 /* pos == 0 means header */
794 if (*pos == 0)
795 return SEQ_START_TOKEN;
797 /* Find the (pos-1)th ipc */
798 for (p = *pos - 1; p <= ids->max_id; p++) {
799 if ((ipc = ipc_lock(ids, p)) != NULL) {
800 *pos = p + 1;
801 return ipc;
804 return NULL;
807 static void sysvipc_proc_stop(struct seq_file *s, void *it)
809 struct kern_ipc_perm *ipc = it;
810 struct ipc_proc_iface *iface = s->private;
811 struct ipc_ids *ids;
813 /* If we had a locked segment, release it */
814 if (ipc && ipc != SEQ_START_TOKEN)
815 ipc_unlock(ipc);
817 ids = current->nsproxy->ipc_ns->ids[iface->ids];
818 /* Release the lock we took in start() */
819 mutex_unlock(&ids->mutex);
822 static int sysvipc_proc_show(struct seq_file *s, void *it)
824 struct ipc_proc_iface *iface = s->private;
826 if (it == SEQ_START_TOKEN)
827 return seq_puts(s, iface->header);
829 return iface->show(s, it);
832 static struct seq_operations sysvipc_proc_seqops = {
833 .start = sysvipc_proc_start,
834 .stop = sysvipc_proc_stop,
835 .next = sysvipc_proc_next,
836 .show = sysvipc_proc_show,
839 static int sysvipc_proc_open(struct inode *inode, struct file *file) {
840 int ret;
841 struct seq_file *seq;
843 ret = seq_open(file, &sysvipc_proc_seqops);
844 if (!ret) {
845 seq = file->private_data;
846 seq->private = PDE(inode)->data;
848 return ret;
851 static struct file_operations sysvipc_proc_fops = {
852 .open = sysvipc_proc_open,
853 .read = seq_read,
854 .llseek = seq_lseek,
855 .release = seq_release,
857 #endif /* CONFIG_PROC_FS */