1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 1992 Krishna Balasubramanian
6 * Sep 1997 - Call suser() last after "normal" permission checks so we
7 * get BSD style process accounting right.
8 * Occurs in several places in the IPC code.
9 * Chris Evans, <chris@ferret.lmh.ox.ac.uk>
10 * Nov 1999 - ipc helper functions, unified SMP locking
11 * Manfred Spraul <manfred@colorfullife.com>
12 * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
13 * Mingming Cao <cmm@us.ibm.com>
14 * Mar 2006 - support for audit of ipc object properties
15 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
16 * Jun 2006 - namespaces ssupport
18 * Pavel Emelianov <xemul@openvz.org>
20 * General sysv ipc locking scheme:
22 * obtain the ipc object (kern_ipc_perm) by looking up the id in an idr
24 * - perform initial checks (capabilities, auditing and permission,
26 * - perform read-only operations, such as INFO command, that
27 * do not demand atomicity
28 * acquire the ipc lock (kern_ipc_perm.lock) through
30 * - perform read-only operations that demand atomicity,
31 * such as STAT command.
32 * - perform data updates, such as SET, RMID commands and
33 * mechanism-specific operations (semop/semtimedop,
34 * msgsnd/msgrcv, shmat/shmdt).
35 * drop the ipc lock, through ipc_unlock_object().
38 * The ids->rwsem must be taken when:
39 * - creating, removing and iterating the existing entries in ipc
41 * - iterating through files under /proc/sysvipc/
43 * Note that sems have a special fast path that avoids kern_ipc_perm.lock -
48 #include <linux/shm.h>
49 #include <linux/init.h>
50 #include <linux/msg.h>
51 #include <linux/vmalloc.h>
52 #include <linux/slab.h>
53 #include <linux/notifier.h>
54 #include <linux/capability.h>
55 #include <linux/highuid.h>
56 #include <linux/security.h>
57 #include <linux/rcupdate.h>
58 #include <linux/workqueue.h>
59 #include <linux/seq_file.h>
60 #include <linux/proc_fs.h>
61 #include <linux/audit.h>
62 #include <linux/nsproxy.h>
63 #include <linux/rwsem.h>
64 #include <linux/memory.h>
65 #include <linux/ipc_namespace.h>
66 #include <linux/rhashtable.h>
68 #include <asm/unistd.h>
72 struct ipc_proc_iface {
76 int (*show)(struct seq_file *, void *);
80 * ipc_init - initialise ipc subsystem
82 * The various sysv ipc resources (semaphores, messages and shared
83 * memory) are initialised.
85 * A callback routine is registered into the memory hotplug notifier
86 * chain: since msgmni scales to lowmem this callback routine will be
87 * called upon successful memory add / remove to recompute msmgni.
89 static int __init ipc_init(void)
91 proc_mkdir("sysvipc", NULL);
98 device_initcall(ipc_init);
100 static const struct rhashtable_params ipc_kht_params = {
101 .head_offset = offsetof(struct kern_ipc_perm, khtnode),
102 .key_offset = offsetof(struct kern_ipc_perm, key),
103 .key_len = FIELD_SIZEOF(struct kern_ipc_perm, key),
104 .automatic_shrinking = true,
108 * ipc_init_ids - initialise ipc identifiers
109 * @ids: ipc identifier set
111 * Set up the sequence range to use for the ipc identifier range (limited
112 * below ipc_mni) then initialise the keys hashtable and ids idr.
114 void ipc_init_ids(struct ipc_ids *ids)
118 init_rwsem(&ids->rwsem);
119 rhashtable_init(&ids->key_ht, &ipc_kht_params);
120 idr_init(&ids->ipcs_idr);
123 #ifdef CONFIG_CHECKPOINT_RESTORE
128 #ifdef CONFIG_PROC_FS
129 static const struct file_operations sysvipc_proc_fops;
131 * ipc_init_proc_interface - create a proc interface for sysipc types using a seq_file interface.
132 * @path: Path in procfs
133 * @header: Banner to be printed at the beginning of the file.
134 * @ids: ipc id table to iterate.
135 * @show: show routine.
137 void __init ipc_init_proc_interface(const char *path, const char *header,
138 int ids, int (*show)(struct seq_file *, void *))
140 struct proc_dir_entry *pde;
141 struct ipc_proc_iface *iface;
143 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
147 iface->header = header;
151 pde = proc_create_data(path,
152 S_IRUGO, /* world readable */
153 NULL, /* parent dir */
162 * ipc_findkey - find a key in an ipc identifier set
163 * @ids: ipc identifier set
166 * Returns the locked pointer to the ipc structure if found or NULL
167 * otherwise. If key is found ipc points to the owning ipc structure
169 * Called with writer ipc_ids.rwsem held.
171 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
173 struct kern_ipc_perm *ipcp;
175 ipcp = rhashtable_lookup_fast(&ids->key_ht, &key,
181 ipc_lock_object(ipcp);
186 * Insert new IPC object into idr tree, and set sequence number and id
187 * in the correct order.
189 * - the sequence number must be set before inserting the object into the idr,
190 * because the sequence number is accessed without a lock.
191 * - the id can/must be set after inserting the object into the idr.
192 * All accesses must be done after getting kern_ipc_perm.lock.
194 * The caller must own kern_ipc_perm.lock.of the new object.
195 * On error, the function returns a (negative) error code.
197 * To conserve sequence number space, especially with extended ipc_mni,
198 * the sequence number is incremented only when the returned ID is less than
201 static inline int ipc_idr_alloc(struct ipc_ids *ids, struct kern_ipc_perm *new)
203 int idx, next_id = -1;
205 #ifdef CONFIG_CHECKPOINT_RESTORE
206 next_id = ids->next_id;
211 * As soon as a new object is inserted into the idr,
212 * ipc_obtain_object_idr() or ipc_obtain_object_check() can find it,
213 * and the lockless preparations for ipc operations can start.
214 * This means especially: permission checks, audit calls, allocation
215 * of undo structures, ...
217 * Thus the object must be fully initialized, and if something fails,
218 * then the full tear-down sequence must be followed.
219 * (i.e.: set new->deleted, reduce refcount, call_rcu())
222 if (next_id < 0) { /* !CHECKPOINT_RESTORE or next_id is unset */
224 /* allocate the idx, with a NULL struct kern_ipc_perm */
225 idx = idr_alloc(&ids->ipcs_idr, NULL, 0, 0, GFP_NOWAIT);
229 * idx got allocated successfully.
230 * Now calculate the sequence number and set the
233 if (idx <= ids->last_idx) {
235 if (ids->seq >= ipcid_seq_max())
241 /* no need for smp_wmb(), this is done
242 * inside idr_replace, as part of
245 idr_replace(&ids->ipcs_idr, new, idx);
248 new->seq = ipcid_to_seqx(next_id);
249 idx = idr_alloc(&ids->ipcs_idr, new, ipcid_to_idx(next_id),
253 new->id = (new->seq << ipcmni_seq_shift()) + idx;
258 * ipc_addid - add an ipc identifier
259 * @ids: ipc identifier set
260 * @new: new ipc permission set
261 * @limit: limit for the number of used ids
263 * Add an entry 'new' to the ipc ids idr. The permissions object is
264 * initialised and the first free entry is set up and the index assigned
265 * is returned. The 'new' entry is returned in a locked state on success.
267 * On failure the entry is not locked and a negative err-code is returned.
268 * The caller must use ipc_rcu_putref() to free the identifier.
270 * Called with writer ipc_ids.rwsem held.
272 int ipc_addid(struct ipc_ids *ids, struct kern_ipc_perm *new, int limit)
278 /* 1) Initialize the refcount so that ipc_rcu_putref works */
279 refcount_set(&new->refcount, 1);
284 if (ids->in_use >= limit)
287 idr_preload(GFP_KERNEL);
289 spin_lock_init(&new->lock);
291 spin_lock(&new->lock);
293 current_euid_egid(&euid, &egid);
294 new->cuid = new->uid = euid;
295 new->gid = new->cgid = egid;
297 new->deleted = false;
299 idx = ipc_idr_alloc(ids, new);
302 if (idx >= 0 && new->key != IPC_PRIVATE) {
303 err = rhashtable_insert_fast(&ids->key_ht, &new->khtnode,
306 idr_remove(&ids->ipcs_idr, idx);
312 spin_unlock(&new->lock);
318 if (idx > ids->max_idx)
324 * ipcget_new - create a new ipc object
326 * @ids: ipc identifier set
327 * @ops: the actual creation routine to call
328 * @params: its parameters
330 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
331 * when the key is IPC_PRIVATE.
333 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
334 const struct ipc_ops *ops, struct ipc_params *params)
338 down_write(&ids->rwsem);
339 err = ops->getnew(ns, params);
340 up_write(&ids->rwsem);
345 * ipc_check_perms - check security and permissions for an ipc object
347 * @ipcp: ipc permission set
348 * @ops: the actual security routine to call
349 * @params: its parameters
351 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
352 * when the key is not IPC_PRIVATE and that key already exists in the
355 * On success, the ipc id is returned.
357 * It is called with ipc_ids.rwsem and ipcp->lock held.
359 static int ipc_check_perms(struct ipc_namespace *ns,
360 struct kern_ipc_perm *ipcp,
361 const struct ipc_ops *ops,
362 struct ipc_params *params)
366 if (ipcperms(ns, ipcp, params->flg))
369 err = ops->associate(ipcp, params->flg);
378 * ipcget_public - get an ipc object or create a new one
380 * @ids: ipc identifier set
381 * @ops: the actual creation routine to call
382 * @params: its parameters
384 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
385 * when the key is not IPC_PRIVATE.
386 * It adds a new entry if the key is not found and does some permission
387 * / security checkings if the key is found.
389 * On success, the ipc id is returned.
391 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
392 const struct ipc_ops *ops, struct ipc_params *params)
394 struct kern_ipc_perm *ipcp;
395 int flg = params->flg;
399 * Take the lock as a writer since we are potentially going to add
400 * a new entry + read locks are not "upgradable"
402 down_write(&ids->rwsem);
403 ipcp = ipc_findkey(ids, params->key);
406 if (!(flg & IPC_CREAT))
409 err = ops->getnew(ns, params);
411 /* ipc object has been locked by ipc_findkey() */
413 if (flg & IPC_CREAT && flg & IPC_EXCL)
417 if (ops->more_checks)
418 err = ops->more_checks(ipcp, params);
421 * ipc_check_perms returns the IPC id on
424 err = ipc_check_perms(ns, ipcp, ops, params);
428 up_write(&ids->rwsem);
434 * ipc_kht_remove - remove an ipc from the key hashtable
435 * @ids: ipc identifier set
436 * @ipcp: ipc perm structure containing the key to remove
438 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
439 * before this function is called, and remain locked on the exit.
441 static void ipc_kht_remove(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
443 if (ipcp->key != IPC_PRIVATE)
444 rhashtable_remove_fast(&ids->key_ht, &ipcp->khtnode,
449 * ipc_rmid - remove an ipc identifier
450 * @ids: ipc identifier set
451 * @ipcp: ipc perm structure containing the identifier to remove
453 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
454 * before this function is called, and remain locked on the exit.
456 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
458 int idx = ipcid_to_idx(ipcp->id);
460 idr_remove(&ids->ipcs_idr, idx);
461 ipc_kht_remove(ids, ipcp);
463 ipcp->deleted = true;
465 if (unlikely(idx == ids->max_idx)) {
470 } while (!idr_find(&ids->ipcs_idr, idx));
476 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
477 * @ids: ipc identifier set
478 * @ipcp: ipc perm structure containing the key to modify
480 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
481 * before this function is called, and remain locked on the exit.
483 void ipc_set_key_private(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
485 ipc_kht_remove(ids, ipcp);
486 ipcp->key = IPC_PRIVATE;
489 bool ipc_rcu_getref(struct kern_ipc_perm *ptr)
491 return refcount_inc_not_zero(&ptr->refcount);
494 void ipc_rcu_putref(struct kern_ipc_perm *ptr,
495 void (*func)(struct rcu_head *head))
497 if (!refcount_dec_and_test(&ptr->refcount))
500 call_rcu(&ptr->rcu, func);
504 * ipcperms - check ipc permissions
506 * @ipcp: ipc permission set
507 * @flag: desired permission set
509 * Check user, group, other permissions for access
510 * to ipc resources. return 0 if allowed
512 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
514 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
516 kuid_t euid = current_euid();
517 int requested_mode, granted_mode;
520 requested_mode = (flag >> 6) | (flag >> 3) | flag;
521 granted_mode = ipcp->mode;
522 if (uid_eq(euid, ipcp->cuid) ||
523 uid_eq(euid, ipcp->uid))
525 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
527 /* is there some bit set in requested_mode but not in granted_mode? */
528 if ((requested_mode & ~granted_mode & 0007) &&
529 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
532 return security_ipc_permission(ipcp, flag);
536 * Functions to convert between the kern_ipc_perm structure and the
537 * old/new ipc_perm structures
541 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
542 * @in: kernel permissions
543 * @out: new style ipc permissions
545 * Turn the kernel object @in into a set of permissions descriptions
546 * for returning to userspace (@out).
548 void kernel_to_ipc64_perm(struct kern_ipc_perm *in, struct ipc64_perm *out)
551 out->uid = from_kuid_munged(current_user_ns(), in->uid);
552 out->gid = from_kgid_munged(current_user_ns(), in->gid);
553 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
554 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
555 out->mode = in->mode;
560 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
561 * @in: new style ipc permissions
562 * @out: old style ipc permissions
564 * Turn the new style permissions object @in into a compatibility
565 * object and store it into the @out pointer.
567 void ipc64_perm_to_ipc_perm(struct ipc64_perm *in, struct ipc_perm *out)
570 SET_UID(out->uid, in->uid);
571 SET_GID(out->gid, in->gid);
572 SET_UID(out->cuid, in->cuid);
573 SET_GID(out->cgid, in->cgid);
574 out->mode = in->mode;
579 * ipc_obtain_object_idr
580 * @ids: ipc identifier set
581 * @id: ipc id to look for
583 * Look for an id in the ipc ids idr and return associated ipc object.
585 * Call inside the RCU critical section.
586 * The ipc object is *not* locked on exit.
588 struct kern_ipc_perm *ipc_obtain_object_idr(struct ipc_ids *ids, int id)
590 struct kern_ipc_perm *out;
591 int idx = ipcid_to_idx(id);
593 out = idr_find(&ids->ipcs_idr, idx);
595 return ERR_PTR(-EINVAL);
601 * ipc_obtain_object_check
602 * @ids: ipc identifier set
603 * @id: ipc id to look for
605 * Similar to ipc_obtain_object_idr() but also checks the ipc object
608 * Call inside the RCU critical section.
609 * The ipc object is *not* locked on exit.
611 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
613 struct kern_ipc_perm *out = ipc_obtain_object_idr(ids, id);
618 if (ipc_checkid(out, id))
619 return ERR_PTR(-EINVAL);
625 * ipcget - Common sys_*get() code
627 * @ids: ipc identifier set
628 * @ops: operations to be called on ipc object creation, permission checks
630 * @params: the parameters needed by the previous operations.
632 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
634 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
635 const struct ipc_ops *ops, struct ipc_params *params)
637 if (params->key == IPC_PRIVATE)
638 return ipcget_new(ns, ids, ops, params);
640 return ipcget_public(ns, ids, ops, params);
644 * ipc_update_perm - update the permissions of an ipc object
645 * @in: the permission given as input.
646 * @out: the permission of the ipc to set.
648 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
650 kuid_t uid = make_kuid(current_user_ns(), in->uid);
651 kgid_t gid = make_kgid(current_user_ns(), in->gid);
652 if (!uid_valid(uid) || !gid_valid(gid))
657 out->mode = (out->mode & ~S_IRWXUGO)
658 | (in->mode & S_IRWXUGO);
664 * ipcctl_obtain_check - retrieve an ipc object and check permissions
666 * @ids: the table of ids where to look for the ipc
667 * @id: the id of the ipc to retrieve
668 * @cmd: the cmd to check
669 * @perm: the permission to set
670 * @extra_perm: one extra permission parameter used by msq
672 * This function does some common audit and permissions check for some IPC_XXX
673 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
676 * - retrieves the ipc object with the given id in the given table.
677 * - performs some audit and permission check, depending on the given cmd
678 * - returns a pointer to the ipc object or otherwise, the corresponding
681 * Call holding the both the rwsem and the rcu read lock.
683 struct kern_ipc_perm *ipcctl_obtain_check(struct ipc_namespace *ns,
684 struct ipc_ids *ids, int id, int cmd,
685 struct ipc64_perm *perm, int extra_perm)
689 struct kern_ipc_perm *ipcp;
691 ipcp = ipc_obtain_object_check(ids, id);
699 audit_ipc_set_perm(extra_perm, perm->uid,
700 perm->gid, perm->mode);
702 euid = current_euid();
703 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
704 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
705 return ipcp; /* successful lookup */
710 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
714 * ipc_parse_version - ipc call version
715 * @cmd: pointer to command
717 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
718 * The @cmd value is turned from an encoding command and version into
719 * just the command code.
721 int ipc_parse_version(int *cmd)
731 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
733 #ifdef CONFIG_PROC_FS
734 struct ipc_proc_iter {
735 struct ipc_namespace *ns;
736 struct pid_namespace *pid_ns;
737 struct ipc_proc_iface *iface;
740 struct pid_namespace *ipc_seq_pid_ns(struct seq_file *s)
742 struct ipc_proc_iter *iter = s->private;
747 * This routine locks the ipc structure found at least at position pos.
749 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
752 struct kern_ipc_perm *ipc;
756 for (id = 0; id < pos && total < ids->in_use; id++) {
757 ipc = idr_find(&ids->ipcs_idr, id);
762 if (total >= ids->in_use)
765 for (; pos < ipc_mni; pos++) {
766 ipc = idr_find(&ids->ipcs_idr, pos);
770 ipc_lock_object(ipc);
775 /* Out of range - return NULL to terminate iteration */
779 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
781 struct ipc_proc_iter *iter = s->private;
782 struct ipc_proc_iface *iface = iter->iface;
783 struct kern_ipc_perm *ipc = it;
785 /* If we had an ipc id locked before, unlock it */
786 if (ipc && ipc != SEQ_START_TOKEN)
789 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
793 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
794 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
796 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
798 struct ipc_proc_iter *iter = s->private;
799 struct ipc_proc_iface *iface = iter->iface;
802 ids = &iter->ns->ids[iface->ids];
805 * Take the lock - this will be released by the corresponding
808 down_read(&ids->rwsem);
810 /* pos < 0 is invalid */
814 /* pos == 0 means header */
816 return SEQ_START_TOKEN;
818 /* Find the (pos-1)th ipc */
819 return sysvipc_find_ipc(ids, *pos - 1, pos);
822 static void sysvipc_proc_stop(struct seq_file *s, void *it)
824 struct kern_ipc_perm *ipc = it;
825 struct ipc_proc_iter *iter = s->private;
826 struct ipc_proc_iface *iface = iter->iface;
829 /* If we had a locked structure, release it */
830 if (ipc && ipc != SEQ_START_TOKEN)
833 ids = &iter->ns->ids[iface->ids];
834 /* Release the lock we took in start() */
835 up_read(&ids->rwsem);
838 static int sysvipc_proc_show(struct seq_file *s, void *it)
840 struct ipc_proc_iter *iter = s->private;
841 struct ipc_proc_iface *iface = iter->iface;
843 if (it == SEQ_START_TOKEN) {
844 seq_puts(s, iface->header);
848 return iface->show(s, it);
851 static const struct seq_operations sysvipc_proc_seqops = {
852 .start = sysvipc_proc_start,
853 .stop = sysvipc_proc_stop,
854 .next = sysvipc_proc_next,
855 .show = sysvipc_proc_show,
858 static int sysvipc_proc_open(struct inode *inode, struct file *file)
860 struct ipc_proc_iter *iter;
862 iter = __seq_open_private(file, &sysvipc_proc_seqops, sizeof(*iter));
866 iter->iface = PDE_DATA(inode);
867 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
868 iter->pid_ns = get_pid_ns(task_active_pid_ns(current));
873 static int sysvipc_proc_release(struct inode *inode, struct file *file)
875 struct seq_file *seq = file->private_data;
876 struct ipc_proc_iter *iter = seq->private;
877 put_ipc_ns(iter->ns);
878 put_pid_ns(iter->pid_ns);
879 return seq_release_private(inode, file);
882 static const struct file_operations sysvipc_proc_fops = {
883 .open = sysvipc_proc_open,
886 .release = sysvipc_proc_release,
888 #endif /* CONFIG_PROC_FS */