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>
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),
54 static struct ipc_namespace *clone_ipc_ns(struct ipc_namespace *old_ns)
57 struct ipc_namespace *ns;
60 ns = kmalloc(sizeof(struct ipc_namespace), GFP_KERNEL);
64 err = sem_init_ns(ns);
67 err = msg_init_ns(ns);
70 err = shm_init_ns(ns);
87 struct ipc_namespace *copy_ipcs(unsigned long flags, struct ipc_namespace *ns)
89 struct ipc_namespace *new_ns;
94 if (!(flags & CLONE_NEWIPC))
97 new_ns = clone_ipc_ns(ns);
103 void free_ipc_ns(struct kref *kref)
105 struct ipc_namespace *ns;
107 ns = container_of(kref, struct ipc_namespace, kref);
115 * ipc_init - initialise IPC subsystem
117 * The various system5 IPC resources (semaphores, messages and shared
118 * memory) are initialised
121 static int __init ipc_init(void)
128 __initcall(ipc_init);
131 * ipc_init_ids - initialise IPC identifiers
132 * @ids: Identifier set
134 * Set up the sequence range to use for the ipc identifier range (limited
135 * below IPCMNI) then initialise the ids idr.
138 void ipc_init_ids(struct ipc_ids *ids)
140 init_rwsem(&ids->rw_mutex);
145 int seq_limit = INT_MAX/SEQ_MULTIPLIER;
146 if(seq_limit > USHRT_MAX)
147 ids->seq_max = USHRT_MAX;
149 ids->seq_max = seq_limit;
152 idr_init(&ids->ipcs_idr);
155 #ifdef CONFIG_PROC_FS
156 static const struct file_operations sysvipc_proc_fops;
158 * ipc_init_proc_interface - Create a proc interface for sysipc types using a seq_file interface.
159 * @path: Path in procfs
160 * @header: Banner to be printed at the beginning of the file.
161 * @ids: ipc id table to iterate.
162 * @show: show routine.
164 void __init ipc_init_proc_interface(const char *path, const char *header,
165 int ids, int (*show)(struct seq_file *, void *))
167 struct proc_dir_entry *pde;
168 struct ipc_proc_iface *iface;
170 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
174 iface->header = header;
178 pde = create_proc_entry(path,
179 S_IRUGO, /* world readable */
180 NULL /* parent dir */);
183 pde->proc_fops = &sysvipc_proc_fops;
191 * ipc_findkey - find a key in an ipc identifier set
192 * @ids: Identifier set
193 * @key: The key to find
195 * Requires ipc_ids.rw_mutex locked.
196 * Returns the LOCKED pointer to the ipc structure if found or NULL
198 * If key is found ipc points to the owning ipc structure
201 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
203 struct kern_ipc_perm *ipc;
207 for (total = 0, next_id = 0; total < ids->in_use; next_id++) {
208 ipc = idr_find(&ids->ipcs_idr, next_id);
213 if (ipc->key != key) {
218 ipc_lock_by_ptr(ipc);
226 * ipc_get_maxid - get the last assigned id
227 * @ids: IPC identifier set
229 * Called with ipc_ids.rw_mutex held.
232 int ipc_get_maxid(struct ipc_ids *ids)
234 struct kern_ipc_perm *ipc;
238 if (ids->in_use == 0)
241 if (ids->in_use == IPCMNI)
244 /* Look for the last assigned id */
246 for (id = 0; id < IPCMNI && total < ids->in_use; id++) {
247 ipc = idr_find(&ids->ipcs_idr, id);
257 * ipc_addid - add an IPC identifier
258 * @ids: IPC identifier set
259 * @new: new IPC permission set
260 * @size: limit for the number of used ids
262 * Add an entry 'new' to the IPC ids idr. The permissions object is
263 * initialised and the first free entry is set up and the id assigned
264 * is returned. The 'new' entry is returned in a locked state on success.
265 * On failure the entry is not locked and -1 is returned.
267 * Called with ipc_ids.rw_mutex held as a writer.
270 int ipc_addid(struct ipc_ids* ids, struct kern_ipc_perm* new, int size)
277 if (ids->in_use >= size)
280 err = idr_get_new(&ids->ipcs_idr, new, &id);
286 new->cuid = new->uid = current->euid;
287 new->gid = new->cgid = current->egid;
289 new->seq = ids->seq++;
290 if(ids->seq > ids->seq_max)
293 spin_lock_init(&new->lock);
296 spin_lock(&new->lock);
301 * ipcget_new - create a new ipc object
303 * @ids: IPC identifer set
304 * @ops: the actual creation routine to call
305 * @params: its parameters
307 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
308 * when the key is IPC_PRIVATE.
310 int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
311 struct ipc_ops *ops, struct ipc_params *params)
315 err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL);
320 down_write(&ids->rw_mutex);
321 err = ops->getnew(ns, params);
322 up_write(&ids->rw_mutex);
328 * ipc_check_perms - check security and permissions for an IPC
329 * @ipcp: ipc permission set
330 * @ops: the actual security routine to call
331 * @params: its parameters
333 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
334 * when the key is not IPC_PRIVATE and that key already exists in the
337 * On success, the IPC id is returned.
339 * It is called with ipc_ids.rw_mutex and ipcp->lock held.
341 static int ipc_check_perms(struct kern_ipc_perm *ipcp, struct ipc_ops *ops,
342 struct ipc_params *params)
346 if (ipcperms(ipcp, params->flg))
349 err = ops->associate(ipcp, params->flg);
358 * ipcget_public - get an ipc object or create a new one
360 * @ids: IPC identifer set
361 * @ops: the actual creation routine to call
362 * @params: its parameters
364 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
365 * when the key is not IPC_PRIVATE.
366 * It adds a new entry if the key is not found and does some permission
367 * / security checkings if the key is found.
369 * On success, the ipc id is returned.
371 int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
372 struct ipc_ops *ops, struct ipc_params *params)
374 struct kern_ipc_perm *ipcp;
375 int flg = params->flg;
378 err = idr_pre_get(&ids->ipcs_idr, GFP_KERNEL);
381 * Take the lock as a writer since we are potentially going to add
382 * a new entry + read locks are not "upgradable"
384 down_write(&ids->rw_mutex);
385 ipcp = ipc_findkey(ids, params->key);
388 if (!(flg & IPC_CREAT))
393 err = ops->getnew(ns, params);
395 /* ipc object has been locked by ipc_findkey() */
397 if (flg & IPC_CREAT && flg & IPC_EXCL)
401 if (ops->more_checks)
402 err = ops->more_checks(ipcp, params);
405 * ipc_check_perms returns the IPC id on
408 err = ipc_check_perms(ipcp, ops, params);
412 up_write(&ids->rw_mutex);
419 * ipc_rmid - remove an IPC identifier
420 * @ids: IPC identifier set
421 * @ipcp: ipc perm structure containing the identifier to remove
423 * ipc_ids.rw_mutex (as a writer) and the spinlock for this ID are held
424 * before this function is called, and remain locked on the exit.
427 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
429 int lid = ipcid_to_idx(ipcp->id);
431 idr_remove(&ids->ipcs_idr, lid);
441 * ipc_alloc - allocate ipc space
442 * @size: size desired
444 * Allocate memory from the appropriate pools and return a pointer to it.
445 * NULL is returned if the allocation fails
448 void* ipc_alloc(int size)
454 out = kmalloc(size, GFP_KERNEL);
459 * ipc_free - free ipc space
460 * @ptr: pointer returned by ipc_alloc
461 * @size: size of block
463 * Free a block created with ipc_alloc(). The caller must know the size
464 * used in the allocation call.
467 void ipc_free(void* ptr, int size)
477 * There are three headers that are prepended to the actual allocation:
478 * - during use: ipc_rcu_hdr.
479 * - during the rcu grace period: ipc_rcu_grace.
480 * - [only if vmalloc]: ipc_rcu_sched.
481 * Their lifetime doesn't overlap, thus the headers share the same memory.
482 * Unlike a normal union, they are right-aligned, thus some container_of
483 * forward/backward casting is necessary:
496 /* "void *" makes sure alignment of following data is sane. */
502 struct work_struct work;
503 /* "void *" makes sure alignment of following data is sane. */
507 #define HDRLEN_KMALLOC (sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
508 sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
509 #define HDRLEN_VMALLOC (sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
510 sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
512 static inline int rcu_use_vmalloc(int size)
514 /* Too big for a single page? */
515 if (HDRLEN_KMALLOC + size > PAGE_SIZE)
521 * ipc_rcu_alloc - allocate ipc and rcu space
522 * @size: size desired
524 * Allocate memory for the rcu header structure + the object.
525 * Returns the pointer to the object.
526 * NULL is returned if the allocation fails.
529 void* ipc_rcu_alloc(int size)
533 * We prepend the allocation with the rcu struct, and
534 * workqueue if necessary (for vmalloc).
536 if (rcu_use_vmalloc(size)) {
537 out = vmalloc(HDRLEN_VMALLOC + size);
539 out += HDRLEN_VMALLOC;
540 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
541 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
544 out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
546 out += HDRLEN_KMALLOC;
547 container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
548 container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
555 void ipc_rcu_getref(void *ptr)
557 container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
560 static void ipc_do_vfree(struct work_struct *work)
562 vfree(container_of(work, struct ipc_rcu_sched, work));
566 * ipc_schedule_free - free ipc + rcu space
567 * @head: RCU callback structure for queued work
569 * Since RCU callback function is called in bh,
570 * we need to defer the vfree to schedule_work().
572 static void ipc_schedule_free(struct rcu_head *head)
574 struct ipc_rcu_grace *grace;
575 struct ipc_rcu_sched *sched;
577 grace = container_of(head, struct ipc_rcu_grace, rcu);
578 sched = container_of(&(grace->data[0]), struct ipc_rcu_sched,
581 INIT_WORK(&sched->work, ipc_do_vfree);
582 schedule_work(&sched->work);
586 * ipc_immediate_free - free ipc + rcu space
587 * @head: RCU callback structure that contains pointer to be freed
589 * Free from the RCU callback context.
591 static void ipc_immediate_free(struct rcu_head *head)
593 struct ipc_rcu_grace *free =
594 container_of(head, struct ipc_rcu_grace, rcu);
598 void ipc_rcu_putref(void *ptr)
600 if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
603 if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
604 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
607 call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
613 * ipcperms - check IPC permissions
614 * @ipcp: IPC permission set
615 * @flag: desired permission set.
617 * Check user, group, other permissions for access
618 * to ipc resources. return 0 if allowed
621 int ipcperms (struct kern_ipc_perm *ipcp, short flag)
622 { /* flag will most probably be 0 or S_...UGO from <linux/stat.h> */
623 int requested_mode, granted_mode, err;
625 if (unlikely((err = audit_ipc_obj(ipcp))))
627 requested_mode = (flag >> 6) | (flag >> 3) | flag;
628 granted_mode = ipcp->mode;
629 if (current->euid == ipcp->cuid || current->euid == ipcp->uid)
631 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
633 /* is there some bit set in requested_mode but not in granted_mode? */
634 if ((requested_mode & ~granted_mode & 0007) &&
635 !capable(CAP_IPC_OWNER))
638 return security_ipc_permission(ipcp, flag);
642 * Functions to convert between the kern_ipc_perm structure and the
643 * old/new ipc_perm structures
647 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
648 * @in: kernel permissions
649 * @out: new style IPC permissions
651 * Turn the kernel object @in into a set of permissions descriptions
652 * for returning to userspace (@out).
656 void kernel_to_ipc64_perm (struct kern_ipc_perm *in, struct ipc64_perm *out)
661 out->cuid = in->cuid;
662 out->cgid = in->cgid;
663 out->mode = in->mode;
668 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
669 * @in: new style IPC permissions
670 * @out: old style IPC permissions
672 * Turn the new style permissions object @in into a compatibility
673 * object and store it into the @out pointer.
676 void ipc64_perm_to_ipc_perm (struct ipc64_perm *in, struct ipc_perm *out)
679 SET_UID(out->uid, in->uid);
680 SET_GID(out->gid, in->gid);
681 SET_UID(out->cuid, in->cuid);
682 SET_GID(out->cgid, in->cgid);
683 out->mode = in->mode;
688 * ipc_lock - Lock an ipc structure without rw_mutex held
689 * @ids: IPC identifier set
690 * @id: ipc id to look for
692 * Look for an id in the ipc ids idr and lock the associated ipc object.
694 * The ipc object is locked on exit.
696 * This is the routine that should be called when the rw_mutex is not already
697 * held, i.e. idr tree not protected: it protects the idr tree in read mode
698 * during the idr_find().
701 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
703 struct kern_ipc_perm *out;
704 int lid = ipcid_to_idx(id);
706 down_read(&ids->rw_mutex);
709 out = idr_find(&ids->ipcs_idr, lid);
712 up_read(&ids->rw_mutex);
713 return ERR_PTR(-EINVAL);
716 up_read(&ids->rw_mutex);
718 spin_lock(&out->lock);
720 /* ipc_rmid() may have already freed the ID while ipc_lock
721 * was spinning: here verify that the structure is still valid
724 spin_unlock(&out->lock);
726 return ERR_PTR(-EINVAL);
733 * ipc_lock_down - Lock an ipc structure with rw_sem held
734 * @ids: IPC identifier set
735 * @id: ipc id to look for
737 * Look for an id in the ipc ids idr and lock the associated ipc object.
739 * The ipc object is locked on exit.
741 * This is the routine that should be called when the rw_mutex is already
742 * held, i.e. idr tree protected.
745 struct kern_ipc_perm *ipc_lock_down(struct ipc_ids *ids, int id)
747 struct kern_ipc_perm *out;
748 int lid = ipcid_to_idx(id);
751 out = idr_find(&ids->ipcs_idr, lid);
754 return ERR_PTR(-EINVAL);
757 spin_lock(&out->lock);
760 * No need to verify that the structure is still valid since the
766 #ifdef __ARCH_WANT_IPC_PARSE_VERSION
770 * ipc_parse_version - IPC call version
771 * @cmd: pointer to command
773 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
774 * The @cmd value is turned from an encoding command and version into
775 * just the command code.
778 int ipc_parse_version (int *cmd)
788 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
790 #ifdef CONFIG_PROC_FS
791 struct ipc_proc_iter {
792 struct ipc_namespace *ns;
793 struct ipc_proc_iface *iface;
797 * This routine locks the ipc structure found at least at position pos.
799 struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
802 struct kern_ipc_perm *ipc;
806 for (id = 0; id < pos && total < ids->in_use; id++) {
807 ipc = idr_find(&ids->ipcs_idr, id);
812 if (total >= ids->in_use)
815 for ( ; pos < IPCMNI; pos++) {
816 ipc = idr_find(&ids->ipcs_idr, pos);
819 ipc_lock_by_ptr(ipc);
824 /* Out of range - return NULL to terminate iteration */
828 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
830 struct ipc_proc_iter *iter = s->private;
831 struct ipc_proc_iface *iface = iter->iface;
832 struct kern_ipc_perm *ipc = it;
834 /* If we had an ipc id locked before, unlock it */
835 if (ipc && ipc != SEQ_START_TOKEN)
838 return sysvipc_find_ipc(iter->ns->ids[iface->ids], *pos, pos);
842 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
843 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
845 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
847 struct ipc_proc_iter *iter = s->private;
848 struct ipc_proc_iface *iface = iter->iface;
851 ids = iter->ns->ids[iface->ids];
854 * Take the lock - this will be released by the corresponding
857 down_read(&ids->rw_mutex);
859 /* pos < 0 is invalid */
863 /* pos == 0 means header */
865 return SEQ_START_TOKEN;
867 /* Find the (pos-1)th ipc */
868 return sysvipc_find_ipc(ids, *pos - 1, pos);
871 static void sysvipc_proc_stop(struct seq_file *s, void *it)
873 struct kern_ipc_perm *ipc = it;
874 struct ipc_proc_iter *iter = s->private;
875 struct ipc_proc_iface *iface = iter->iface;
878 /* If we had a locked structure, release it */
879 if (ipc && ipc != SEQ_START_TOKEN)
882 ids = iter->ns->ids[iface->ids];
883 /* Release the lock we took in start() */
884 up_read(&ids->rw_mutex);
887 static int sysvipc_proc_show(struct seq_file *s, void *it)
889 struct ipc_proc_iter *iter = s->private;
890 struct ipc_proc_iface *iface = iter->iface;
892 if (it == SEQ_START_TOKEN)
893 return seq_puts(s, iface->header);
895 return iface->show(s, it);
898 static struct seq_operations sysvipc_proc_seqops = {
899 .start = sysvipc_proc_start,
900 .stop = sysvipc_proc_stop,
901 .next = sysvipc_proc_next,
902 .show = sysvipc_proc_show,
905 static int sysvipc_proc_open(struct inode *inode, struct file *file)
908 struct seq_file *seq;
909 struct ipc_proc_iter *iter;
912 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
916 ret = seq_open(file, &sysvipc_proc_seqops);
920 seq = file->private_data;
923 iter->iface = PDE(inode)->data;
924 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
932 static int sysvipc_proc_release(struct inode *inode, struct file *file)
934 struct seq_file *seq = file->private_data;
935 struct ipc_proc_iter *iter = seq->private;
936 put_ipc_ns(iter->ns);
937 return seq_release_private(inode, file);
940 static const struct file_operations sysvipc_proc_fops = {
941 .open = sysvipc_proc_open,
944 .release = sysvipc_proc_release,
946 #endif /* CONFIG_PROC_FS */