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)
93 proc_mkdir("sysvipc", NULL);
95 WARN(err_sem, "ipc: sysv sem_init failed: %d\n", err_sem);
97 WARN(err_msg, "ipc: sysv msg_init failed: %d\n", err_msg);
100 return err_msg ? err_msg : err_sem;
102 device_initcall(ipc_init);
104 static const struct rhashtable_params ipc_kht_params = {
105 .head_offset = offsetof(struct kern_ipc_perm, khtnode),
106 .key_offset = offsetof(struct kern_ipc_perm, key),
107 .key_len = FIELD_SIZEOF(struct kern_ipc_perm, key),
109 .automatic_shrinking = true,
113 * ipc_init_ids - initialise ipc identifiers
114 * @ids: ipc identifier set
116 * Set up the sequence range to use for the ipc identifier range (limited
117 * below IPCMNI) then initialise the keys hashtable and ids idr.
119 int ipc_init_ids(struct ipc_ids *ids)
124 init_rwsem(&ids->rwsem);
125 err = rhashtable_init(&ids->key_ht, &ipc_kht_params);
128 idr_init(&ids->ipcs_idr);
129 ids->tables_initialized = true;
131 #ifdef CONFIG_CHECKPOINT_RESTORE
137 #ifdef CONFIG_PROC_FS
138 static const struct file_operations sysvipc_proc_fops;
140 * ipc_init_proc_interface - create a proc interface for sysipc types using a seq_file interface.
141 * @path: Path in procfs
142 * @header: Banner to be printed at the beginning of the file.
143 * @ids: ipc id table to iterate.
144 * @show: show routine.
146 void __init ipc_init_proc_interface(const char *path, const char *header,
147 int ids, int (*show)(struct seq_file *, void *))
149 struct proc_dir_entry *pde;
150 struct ipc_proc_iface *iface;
152 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
156 iface->header = header;
160 pde = proc_create_data(path,
161 S_IRUGO, /* world readable */
162 NULL, /* parent dir */
171 * ipc_findkey - find a key in an ipc identifier set
172 * @ids: ipc identifier set
175 * Returns the locked pointer to the ipc structure if found or NULL
176 * otherwise. If key is found ipc points to the owning ipc structure
178 * Called with writer ipc_ids.rwsem held.
180 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
182 struct kern_ipc_perm *ipcp = NULL;
184 if (likely(ids->tables_initialized))
185 ipcp = rhashtable_lookup_fast(&ids->key_ht, &key,
190 ipc_lock_object(ipcp);
198 * Insert new IPC object into idr tree, and set sequence number and id
199 * in the correct order.
201 * - the sequence number must be set before inserting the object into the idr,
202 * because the sequence number is accessed without a lock.
203 * - the id can/must be set after inserting the object into the idr.
204 * All accesses must be done after getting kern_ipc_perm.lock.
206 * The caller must own kern_ipc_perm.lock.of the new object.
207 * On error, the function returns a (negative) error code.
209 static inline int ipc_idr_alloc(struct ipc_ids *ids, struct kern_ipc_perm *new)
211 int idx, next_id = -1;
213 #ifdef CONFIG_CHECKPOINT_RESTORE
214 next_id = ids->next_id;
219 * As soon as a new object is inserted into the idr,
220 * ipc_obtain_object_idr() or ipc_obtain_object_check() can find it,
221 * and the lockless preparations for ipc operations can start.
222 * This means especially: permission checks, audit calls, allocation
223 * of undo structures, ...
225 * Thus the object must be fully initialized, and if something fails,
226 * then the full tear-down sequence must be followed.
227 * (i.e.: set new->deleted, reduce refcount, call_rcu())
230 if (next_id < 0) { /* !CHECKPOINT_RESTORE or next_id is unset */
231 new->seq = ids->seq++;
232 if (ids->seq > IPCID_SEQ_MAX)
234 idx = idr_alloc(&ids->ipcs_idr, new, 0, 0, GFP_NOWAIT);
236 new->seq = ipcid_to_seqx(next_id);
237 idx = idr_alloc(&ids->ipcs_idr, new, ipcid_to_idx(next_id),
241 new->id = SEQ_MULTIPLIER * new->seq + idx;
246 * ipc_addid - add an ipc identifier
247 * @ids: ipc identifier set
248 * @new: new ipc permission set
249 * @limit: limit for the number of used ids
251 * Add an entry 'new' to the ipc ids idr. The permissions object is
252 * initialised and the first free entry is set up and the id assigned
253 * is returned. The 'new' entry is returned in a locked state on success.
254 * On failure the entry is not locked and a negative err-code is returned.
256 * Called with writer ipc_ids.rwsem held.
258 int ipc_addid(struct ipc_ids *ids, struct kern_ipc_perm *new, int limit)
267 if (!ids->tables_initialized || ids->in_use >= limit)
270 idr_preload(GFP_KERNEL);
272 refcount_set(&new->refcount, 1);
273 spin_lock_init(&new->lock);
274 new->deleted = false;
276 spin_lock(&new->lock);
278 current_euid_egid(&euid, &egid);
279 new->cuid = new->uid = euid;
280 new->gid = new->cgid = egid;
282 idx = ipc_idr_alloc(ids, new);
285 if (idx >= 0 && new->key != IPC_PRIVATE) {
286 err = rhashtable_insert_fast(&ids->key_ht, &new->khtnode,
289 idr_remove(&ids->ipcs_idr, idx);
294 spin_unlock(&new->lock);
300 if (idx > ids->max_id)
306 * ipcget_new - create a new ipc object
308 * @ids: ipc identifier set
309 * @ops: the actual creation routine to call
310 * @params: its parameters
312 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
313 * when the key is IPC_PRIVATE.
315 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
316 const struct ipc_ops *ops, struct ipc_params *params)
320 down_write(&ids->rwsem);
321 err = ops->getnew(ns, params);
322 up_write(&ids->rwsem);
327 * ipc_check_perms - check security and permissions for an ipc object
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.rwsem and ipcp->lock held.
341 static int ipc_check_perms(struct ipc_namespace *ns,
342 struct kern_ipc_perm *ipcp,
343 const struct ipc_ops *ops,
344 struct ipc_params *params)
348 if (ipcperms(ns, ipcp, params->flg))
351 err = ops->associate(ipcp, params->flg);
360 * ipcget_public - get an ipc object or create a new one
362 * @ids: ipc identifier set
363 * @ops: the actual creation routine to call
364 * @params: its parameters
366 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
367 * when the key is not IPC_PRIVATE.
368 * It adds a new entry if the key is not found and does some permission
369 * / security checkings if the key is found.
371 * On success, the ipc id is returned.
373 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
374 const struct ipc_ops *ops, struct ipc_params *params)
376 struct kern_ipc_perm *ipcp;
377 int flg = params->flg;
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->rwsem);
385 ipcp = ipc_findkey(ids, params->key);
388 if (!(flg & IPC_CREAT))
391 err = ops->getnew(ns, params);
393 /* ipc object has been locked by ipc_findkey() */
395 if (flg & IPC_CREAT && flg & IPC_EXCL)
399 if (ops->more_checks)
400 err = ops->more_checks(ipcp, params);
403 * ipc_check_perms returns the IPC id on
406 err = ipc_check_perms(ns, ipcp, ops, params);
410 up_write(&ids->rwsem);
416 * ipc_kht_remove - remove an ipc from the key hashtable
417 * @ids: ipc identifier set
418 * @ipcp: ipc perm structure containing the key to remove
420 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
421 * before this function is called, and remain locked on the exit.
423 static void ipc_kht_remove(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
425 if (ipcp->key != IPC_PRIVATE)
426 rhashtable_remove_fast(&ids->key_ht, &ipcp->khtnode,
431 * ipc_rmid - remove an ipc identifier
432 * @ids: ipc identifier set
433 * @ipcp: ipc perm structure containing the identifier to remove
435 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
436 * before this function is called, and remain locked on the exit.
438 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
440 int lid = ipcid_to_idx(ipcp->id);
442 idr_remove(&ids->ipcs_idr, lid);
443 ipc_kht_remove(ids, ipcp);
445 ipcp->deleted = true;
447 if (unlikely(lid == ids->max_id)) {
452 } while (!idr_find(&ids->ipcs_idr, lid));
458 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
459 * @ids: ipc identifier set
460 * @ipcp: ipc perm structure containing the key to modify
462 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
463 * before this function is called, and remain locked on the exit.
465 void ipc_set_key_private(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
467 ipc_kht_remove(ids, ipcp);
468 ipcp->key = IPC_PRIVATE;
471 int ipc_rcu_getref(struct kern_ipc_perm *ptr)
473 return refcount_inc_not_zero(&ptr->refcount);
476 void ipc_rcu_putref(struct kern_ipc_perm *ptr,
477 void (*func)(struct rcu_head *head))
479 if (!refcount_dec_and_test(&ptr->refcount))
482 call_rcu(&ptr->rcu, func);
486 * ipcperms - check ipc permissions
488 * @ipcp: ipc permission set
489 * @flag: desired permission set
491 * Check user, group, other permissions for access
492 * to ipc resources. return 0 if allowed
494 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
496 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
498 kuid_t euid = current_euid();
499 int requested_mode, granted_mode;
502 requested_mode = (flag >> 6) | (flag >> 3) | flag;
503 granted_mode = ipcp->mode;
504 if (uid_eq(euid, ipcp->cuid) ||
505 uid_eq(euid, ipcp->uid))
507 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
509 /* is there some bit set in requested_mode but not in granted_mode? */
510 if ((requested_mode & ~granted_mode & 0007) &&
511 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
514 return security_ipc_permission(ipcp, flag);
518 * Functions to convert between the kern_ipc_perm structure and the
519 * old/new ipc_perm structures
523 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
524 * @in: kernel permissions
525 * @out: new style ipc permissions
527 * Turn the kernel object @in into a set of permissions descriptions
528 * for returning to userspace (@out).
530 void kernel_to_ipc64_perm(struct kern_ipc_perm *in, struct ipc64_perm *out)
533 out->uid = from_kuid_munged(current_user_ns(), in->uid);
534 out->gid = from_kgid_munged(current_user_ns(), in->gid);
535 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
536 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
537 out->mode = in->mode;
542 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
543 * @in: new style ipc permissions
544 * @out: old style ipc permissions
546 * Turn the new style permissions object @in into a compatibility
547 * object and store it into the @out pointer.
549 void ipc64_perm_to_ipc_perm(struct ipc64_perm *in, struct ipc_perm *out)
552 SET_UID(out->uid, in->uid);
553 SET_GID(out->gid, in->gid);
554 SET_UID(out->cuid, in->cuid);
555 SET_GID(out->cgid, in->cgid);
556 out->mode = in->mode;
561 * ipc_obtain_object_idr
562 * @ids: ipc identifier set
563 * @id: ipc id to look for
565 * Look for an id in the ipc ids idr and return associated ipc object.
567 * Call inside the RCU critical section.
568 * The ipc object is *not* locked on exit.
570 struct kern_ipc_perm *ipc_obtain_object_idr(struct ipc_ids *ids, int id)
572 struct kern_ipc_perm *out;
573 int lid = ipcid_to_idx(id);
575 if (unlikely(!ids->tables_initialized))
576 return ERR_PTR(-EINVAL);
578 out = idr_find(&ids->ipcs_idr, lid);
580 return ERR_PTR(-EINVAL);
586 * ipc_lock - lock an ipc structure without rwsem held
587 * @ids: ipc identifier set
588 * @id: ipc id to look for
590 * Look for an id in the ipc ids idr and lock the associated ipc object.
592 * The ipc object is locked on successful exit.
594 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
596 struct kern_ipc_perm *out;
599 out = ipc_obtain_object_idr(ids, id);
603 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.
608 * Upon races with RMID, return -EIDRM, thus indicating that
609 * the ID points to a removed identifier.
611 if (ipc_valid_object(out))
614 spin_unlock(&out->lock);
615 out = ERR_PTR(-EIDRM);
622 * ipc_obtain_object_check
623 * @ids: ipc identifier set
624 * @id: ipc id to look for
626 * Similar to ipc_obtain_object_idr() but also checks
627 * the ipc object reference counter.
629 * Call inside the RCU critical section.
630 * The ipc object is *not* locked on exit.
632 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
634 struct kern_ipc_perm *out = ipc_obtain_object_idr(ids, id);
639 if (ipc_checkid(out, id))
640 return ERR_PTR(-EINVAL);
646 * ipcget - Common sys_*get() code
648 * @ids: ipc identifier set
649 * @ops: operations to be called on ipc object creation, permission checks
651 * @params: the parameters needed by the previous operations.
653 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
655 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
656 const struct ipc_ops *ops, struct ipc_params *params)
658 if (params->key == IPC_PRIVATE)
659 return ipcget_new(ns, ids, ops, params);
661 return ipcget_public(ns, ids, ops, params);
665 * ipc_update_perm - update the permissions of an ipc object
666 * @in: the permission given as input.
667 * @out: the permission of the ipc to set.
669 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
671 kuid_t uid = make_kuid(current_user_ns(), in->uid);
672 kgid_t gid = make_kgid(current_user_ns(), in->gid);
673 if (!uid_valid(uid) || !gid_valid(gid))
678 out->mode = (out->mode & ~S_IRWXUGO)
679 | (in->mode & S_IRWXUGO);
685 * ipcctl_pre_down_nolock - retrieve an ipc and check permissions for some IPC_XXX cmd
687 * @ids: the table of ids where to look for the ipc
688 * @id: the id of the ipc to retrieve
689 * @cmd: the cmd to check
690 * @perm: the permission to set
691 * @extra_perm: one extra permission parameter used by msq
693 * This function does some common audit and permissions check for some IPC_XXX
694 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
695 * It must be called without any lock held and:
697 * - retrieves the ipc with the given id in the given table.
698 * - performs some audit and permission check, depending on the given cmd
699 * - returns a pointer to the ipc object or otherwise, the corresponding
702 * Call holding the both the rwsem and the rcu read lock.
704 struct kern_ipc_perm *ipcctl_pre_down_nolock(struct ipc_namespace *ns,
705 struct ipc_ids *ids, int id, int cmd,
706 struct ipc64_perm *perm, int extra_perm)
710 struct kern_ipc_perm *ipcp;
712 ipcp = ipc_obtain_object_check(ids, id);
720 audit_ipc_set_perm(extra_perm, perm->uid,
721 perm->gid, perm->mode);
723 euid = current_euid();
724 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
725 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
726 return ipcp; /* successful lookup */
731 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
735 * ipc_parse_version - ipc call version
736 * @cmd: pointer to command
738 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
739 * The @cmd value is turned from an encoding command and version into
740 * just the command code.
742 int ipc_parse_version(int *cmd)
752 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
754 #ifdef CONFIG_PROC_FS
755 struct ipc_proc_iter {
756 struct ipc_namespace *ns;
757 struct pid_namespace *pid_ns;
758 struct ipc_proc_iface *iface;
761 struct pid_namespace *ipc_seq_pid_ns(struct seq_file *s)
763 struct ipc_proc_iter *iter = s->private;
768 * This routine locks the ipc structure found at least at position pos.
770 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
773 struct kern_ipc_perm *ipc;
777 for (id = 0; id < pos && total < ids->in_use; id++) {
778 ipc = idr_find(&ids->ipcs_idr, id);
783 if (total >= ids->in_use)
786 for (; pos < IPCMNI; pos++) {
787 ipc = idr_find(&ids->ipcs_idr, pos);
791 ipc_lock_object(ipc);
796 /* Out of range - return NULL to terminate iteration */
800 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
802 struct ipc_proc_iter *iter = s->private;
803 struct ipc_proc_iface *iface = iter->iface;
804 struct kern_ipc_perm *ipc = it;
806 /* If we had an ipc id locked before, unlock it */
807 if (ipc && ipc != SEQ_START_TOKEN)
810 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
814 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
815 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
817 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
819 struct ipc_proc_iter *iter = s->private;
820 struct ipc_proc_iface *iface = iter->iface;
823 ids = &iter->ns->ids[iface->ids];
826 * Take the lock - this will be released by the corresponding
829 down_read(&ids->rwsem);
831 /* pos < 0 is invalid */
835 /* pos == 0 means header */
837 return SEQ_START_TOKEN;
839 /* Find the (pos-1)th ipc */
840 return sysvipc_find_ipc(ids, *pos - 1, pos);
843 static void sysvipc_proc_stop(struct seq_file *s, void *it)
845 struct kern_ipc_perm *ipc = it;
846 struct ipc_proc_iter *iter = s->private;
847 struct ipc_proc_iface *iface = iter->iface;
850 /* If we had a locked structure, release it */
851 if (ipc && ipc != SEQ_START_TOKEN)
854 ids = &iter->ns->ids[iface->ids];
855 /* Release the lock we took in start() */
856 up_read(&ids->rwsem);
859 static int sysvipc_proc_show(struct seq_file *s, void *it)
861 struct ipc_proc_iter *iter = s->private;
862 struct ipc_proc_iface *iface = iter->iface;
864 if (it == SEQ_START_TOKEN) {
865 seq_puts(s, iface->header);
869 return iface->show(s, it);
872 static const struct seq_operations sysvipc_proc_seqops = {
873 .start = sysvipc_proc_start,
874 .stop = sysvipc_proc_stop,
875 .next = sysvipc_proc_next,
876 .show = sysvipc_proc_show,
879 static int sysvipc_proc_open(struct inode *inode, struct file *file)
881 struct ipc_proc_iter *iter;
883 iter = __seq_open_private(file, &sysvipc_proc_seqops, sizeof(*iter));
887 iter->iface = PDE_DATA(inode);
888 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
889 iter->pid_ns = get_pid_ns(task_active_pid_ns(current));
894 static int sysvipc_proc_release(struct inode *inode, struct file *file)
896 struct seq_file *seq = file->private_data;
897 struct ipc_proc_iter *iter = seq->private;
898 put_ipc_ns(iter->ns);
899 put_pid_ns(iter->pid_ns);
900 return seq_release_private(inode, file);
903 static const struct file_operations sysvipc_proc_fops = {
904 .open = sysvipc_proc_open,
907 .release = sysvipc_proc_release,
909 #endif /* CONFIG_PROC_FS */