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),
105 .automatic_shrinking = true,
109 * ipc_init_ids - initialise ipc identifiers
110 * @ids: ipc identifier set
112 * Set up the sequence range to use for the ipc identifier range (limited
113 * below ipc_mni) then initialise the keys hashtable and ids idr.
115 void ipc_init_ids(struct ipc_ids *ids)
118 ids->deleted = false;
119 ids->seq = ipc_mni_extended ? 0 : -1; /* seq # is pre-incremented */
120 init_rwsem(&ids->rwsem);
121 rhashtable_init(&ids->key_ht, &ipc_kht_params);
122 idr_init(&ids->ipcs_idr);
124 #ifdef CONFIG_CHECKPOINT_RESTORE
129 #ifdef CONFIG_PROC_FS
130 static const struct file_operations sysvipc_proc_fops;
132 * ipc_init_proc_interface - create a proc interface for sysipc types using a seq_file interface.
133 * @path: Path in procfs
134 * @header: Banner to be printed at the beginning of the file.
135 * @ids: ipc id table to iterate.
136 * @show: show routine.
138 void __init ipc_init_proc_interface(const char *path, const char *header,
139 int ids, int (*show)(struct seq_file *, void *))
141 struct proc_dir_entry *pde;
142 struct ipc_proc_iface *iface;
144 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
148 iface->header = header;
152 pde = proc_create_data(path,
153 S_IRUGO, /* world readable */
154 NULL, /* parent dir */
163 * ipc_findkey - find a key in an ipc identifier set
164 * @ids: ipc identifier set
167 * Returns the locked pointer to the ipc structure if found or NULL
168 * otherwise. If key is found ipc points to the owning ipc structure
170 * Called with writer ipc_ids.rwsem held.
172 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
174 struct kern_ipc_perm *ipcp;
176 ipcp = rhashtable_lookup_fast(&ids->key_ht, &key,
182 ipc_lock_object(ipcp);
187 * Insert new IPC object into idr tree, and set sequence number and id
188 * in the correct order.
190 * - the sequence number must be set before inserting the object into the idr,
191 * because the sequence number is accessed without a lock.
192 * - the id can/must be set after inserting the object into the idr.
193 * All accesses must be done after getting kern_ipc_perm.lock.
195 * The caller must own kern_ipc_perm.lock.of the new object.
196 * On error, the function returns a (negative) error code.
198 static inline int ipc_idr_alloc(struct ipc_ids *ids, struct kern_ipc_perm *new)
200 int idx, next_id = -1;
203 * To conserve sequence number space with extended ipc_mni when new ID
204 * is built, the sequence number is incremented only when one or more
205 * IDs have been removed previously.
207 #ifdef CONFIG_CHECKPOINT_RESTORE
208 next_id = ids->next_id;
213 * As soon as a new object is inserted into the idr,
214 * ipc_obtain_object_idr() or ipc_obtain_object_check() can find it,
215 * and the lockless preparations for ipc operations can start.
216 * This means especially: permission checks, audit calls, allocation
217 * of undo structures, ...
219 * Thus the object must be fully initialized, and if something fails,
220 * then the full tear-down sequence must be followed.
221 * (i.e.: set new->deleted, reduce refcount, call_rcu())
224 if (next_id < 0) { /* !CHECKPOINT_RESTORE or next_id is unset */
225 if (!ipc_mni_extended || ids->deleted) {
227 if (ids->seq > IPCID_SEQ_MAX)
229 ids->deleted = false;
232 idx = idr_alloc(&ids->ipcs_idr, new, 0, 0, GFP_NOWAIT);
234 new->seq = ipcid_to_seqx(next_id);
235 idx = idr_alloc(&ids->ipcs_idr, new, ipcid_to_idx(next_id),
239 new->id = (new->seq << IPCMNI_SEQ_SHIFT) + idx;
244 * ipc_addid - add an ipc identifier
245 * @ids: ipc identifier set
246 * @new: new ipc permission set
247 * @limit: limit for the number of used ids
249 * Add an entry 'new' to the ipc ids idr. The permissions object is
250 * initialised and the first free entry is set up and the index assigned
251 * is returned. The 'new' entry is returned in a locked state on success.
253 * On failure the entry is not locked and a negative err-code is returned.
254 * The caller must use ipc_rcu_putref() to free the identifier.
256 * Called with writer ipc_ids.rwsem held.
258 int ipc_addid(struct ipc_ids *ids, struct kern_ipc_perm *new, int limit)
264 /* 1) Initialize the refcount so that ipc_rcu_putref works */
265 refcount_set(&new->refcount, 1);
270 if (ids->in_use >= limit)
273 idr_preload(GFP_KERNEL);
275 spin_lock_init(&new->lock);
277 spin_lock(&new->lock);
279 current_euid_egid(&euid, &egid);
280 new->cuid = new->uid = euid;
281 new->gid = new->cgid = egid;
283 new->deleted = false;
285 idx = ipc_idr_alloc(ids, new);
288 if (idx >= 0 && new->key != IPC_PRIVATE) {
289 err = rhashtable_insert_fast(&ids->key_ht, &new->khtnode,
292 idr_remove(&ids->ipcs_idr, idx);
298 spin_unlock(&new->lock);
304 if (idx > ids->max_idx)
310 * ipcget_new - create a new ipc object
312 * @ids: ipc identifier set
313 * @ops: the actual creation routine to call
314 * @params: its parameters
316 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
317 * when the key is IPC_PRIVATE.
319 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
320 const struct ipc_ops *ops, struct ipc_params *params)
324 down_write(&ids->rwsem);
325 err = ops->getnew(ns, params);
326 up_write(&ids->rwsem);
331 * ipc_check_perms - check security and permissions for an ipc object
333 * @ipcp: ipc permission set
334 * @ops: the actual security routine to call
335 * @params: its parameters
337 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
338 * when the key is not IPC_PRIVATE and that key already exists in the
341 * On success, the ipc id is returned.
343 * It is called with ipc_ids.rwsem and ipcp->lock held.
345 static int ipc_check_perms(struct ipc_namespace *ns,
346 struct kern_ipc_perm *ipcp,
347 const struct ipc_ops *ops,
348 struct ipc_params *params)
352 if (ipcperms(ns, ipcp, params->flg))
355 err = ops->associate(ipcp, params->flg);
364 * ipcget_public - get an ipc object or create a new one
366 * @ids: ipc identifier set
367 * @ops: the actual creation routine to call
368 * @params: its parameters
370 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
371 * when the key is not IPC_PRIVATE.
372 * It adds a new entry if the key is not found and does some permission
373 * / security checkings if the key is found.
375 * On success, the ipc id is returned.
377 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
378 const struct ipc_ops *ops, struct ipc_params *params)
380 struct kern_ipc_perm *ipcp;
381 int flg = params->flg;
385 * Take the lock as a writer since we are potentially going to add
386 * a new entry + read locks are not "upgradable"
388 down_write(&ids->rwsem);
389 ipcp = ipc_findkey(ids, params->key);
392 if (!(flg & IPC_CREAT))
395 err = ops->getnew(ns, params);
397 /* ipc object has been locked by ipc_findkey() */
399 if (flg & IPC_CREAT && flg & IPC_EXCL)
403 if (ops->more_checks)
404 err = ops->more_checks(ipcp, params);
407 * ipc_check_perms returns the IPC id on
410 err = ipc_check_perms(ns, ipcp, ops, params);
414 up_write(&ids->rwsem);
420 * ipc_kht_remove - remove an ipc from the key hashtable
421 * @ids: ipc identifier set
422 * @ipcp: ipc perm structure containing the key to remove
424 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
425 * before this function is called, and remain locked on the exit.
427 static void ipc_kht_remove(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
429 if (ipcp->key != IPC_PRIVATE)
430 rhashtable_remove_fast(&ids->key_ht, &ipcp->khtnode,
435 * ipc_rmid - remove an ipc identifier
436 * @ids: ipc identifier set
437 * @ipcp: ipc perm structure containing the identifier to remove
439 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
440 * before this function is called, and remain locked on the exit.
442 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
444 int idx = ipcid_to_idx(ipcp->id);
446 idr_remove(&ids->ipcs_idr, idx);
447 ipc_kht_remove(ids, ipcp);
450 ipcp->deleted = true;
452 if (unlikely(idx == ids->max_idx)) {
457 } while (!idr_find(&ids->ipcs_idr, idx));
463 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
464 * @ids: ipc identifier set
465 * @ipcp: ipc perm structure containing the key to modify
467 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
468 * before this function is called, and remain locked on the exit.
470 void ipc_set_key_private(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
472 ipc_kht_remove(ids, ipcp);
473 ipcp->key = IPC_PRIVATE;
476 bool ipc_rcu_getref(struct kern_ipc_perm *ptr)
478 return refcount_inc_not_zero(&ptr->refcount);
481 void ipc_rcu_putref(struct kern_ipc_perm *ptr,
482 void (*func)(struct rcu_head *head))
484 if (!refcount_dec_and_test(&ptr->refcount))
487 call_rcu(&ptr->rcu, func);
491 * ipcperms - check ipc permissions
493 * @ipcp: ipc permission set
494 * @flag: desired permission set
496 * Check user, group, other permissions for access
497 * to ipc resources. return 0 if allowed
499 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
501 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
503 kuid_t euid = current_euid();
504 int requested_mode, granted_mode;
507 requested_mode = (flag >> 6) | (flag >> 3) | flag;
508 granted_mode = ipcp->mode;
509 if (uid_eq(euid, ipcp->cuid) ||
510 uid_eq(euid, ipcp->uid))
512 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
514 /* is there some bit set in requested_mode but not in granted_mode? */
515 if ((requested_mode & ~granted_mode & 0007) &&
516 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
519 return security_ipc_permission(ipcp, flag);
523 * Functions to convert between the kern_ipc_perm structure and the
524 * old/new ipc_perm structures
528 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
529 * @in: kernel permissions
530 * @out: new style ipc permissions
532 * Turn the kernel object @in into a set of permissions descriptions
533 * for returning to userspace (@out).
535 void kernel_to_ipc64_perm(struct kern_ipc_perm *in, struct ipc64_perm *out)
538 out->uid = from_kuid_munged(current_user_ns(), in->uid);
539 out->gid = from_kgid_munged(current_user_ns(), in->gid);
540 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
541 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
542 out->mode = in->mode;
547 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
548 * @in: new style ipc permissions
549 * @out: old style ipc permissions
551 * Turn the new style permissions object @in into a compatibility
552 * object and store it into the @out pointer.
554 void ipc64_perm_to_ipc_perm(struct ipc64_perm *in, struct ipc_perm *out)
557 SET_UID(out->uid, in->uid);
558 SET_GID(out->gid, in->gid);
559 SET_UID(out->cuid, in->cuid);
560 SET_GID(out->cgid, in->cgid);
561 out->mode = in->mode;
566 * ipc_obtain_object_idr
567 * @ids: ipc identifier set
568 * @id: ipc id to look for
570 * Look for an id in the ipc ids idr and return associated ipc object.
572 * Call inside the RCU critical section.
573 * The ipc object is *not* locked on exit.
575 struct kern_ipc_perm *ipc_obtain_object_idr(struct ipc_ids *ids, int id)
577 struct kern_ipc_perm *out;
578 int idx = ipcid_to_idx(id);
580 out = idr_find(&ids->ipcs_idr, idx);
582 return ERR_PTR(-EINVAL);
588 * ipc_obtain_object_check
589 * @ids: ipc identifier set
590 * @id: ipc id to look for
592 * Similar to ipc_obtain_object_idr() but also checks the ipc object
595 * Call inside the RCU critical section.
596 * The ipc object is *not* locked on exit.
598 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
600 struct kern_ipc_perm *out = ipc_obtain_object_idr(ids, id);
605 if (ipc_checkid(out, id))
606 return ERR_PTR(-EINVAL);
612 * ipcget - Common sys_*get() code
614 * @ids: ipc identifier set
615 * @ops: operations to be called on ipc object creation, permission checks
617 * @params: the parameters needed by the previous operations.
619 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
621 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
622 const struct ipc_ops *ops, struct ipc_params *params)
624 if (params->key == IPC_PRIVATE)
625 return ipcget_new(ns, ids, ops, params);
627 return ipcget_public(ns, ids, ops, params);
631 * ipc_update_perm - update the permissions of an ipc object
632 * @in: the permission given as input.
633 * @out: the permission of the ipc to set.
635 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
637 kuid_t uid = make_kuid(current_user_ns(), in->uid);
638 kgid_t gid = make_kgid(current_user_ns(), in->gid);
639 if (!uid_valid(uid) || !gid_valid(gid))
644 out->mode = (out->mode & ~S_IRWXUGO)
645 | (in->mode & S_IRWXUGO);
651 * ipcctl_obtain_check - retrieve an ipc object and check permissions
653 * @ids: the table of ids where to look for the ipc
654 * @id: the id of the ipc to retrieve
655 * @cmd: the cmd to check
656 * @perm: the permission to set
657 * @extra_perm: one extra permission parameter used by msq
659 * This function does some common audit and permissions check for some IPC_XXX
660 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
663 * - retrieves the ipc object with the given id in the given table.
664 * - performs some audit and permission check, depending on the given cmd
665 * - returns a pointer to the ipc object or otherwise, the corresponding
668 * Call holding the both the rwsem and the rcu read lock.
670 struct kern_ipc_perm *ipcctl_obtain_check(struct ipc_namespace *ns,
671 struct ipc_ids *ids, int id, int cmd,
672 struct ipc64_perm *perm, int extra_perm)
676 struct kern_ipc_perm *ipcp;
678 ipcp = ipc_obtain_object_check(ids, id);
686 audit_ipc_set_perm(extra_perm, perm->uid,
687 perm->gid, perm->mode);
689 euid = current_euid();
690 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
691 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
692 return ipcp; /* successful lookup */
697 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
701 * ipc_parse_version - ipc call version
702 * @cmd: pointer to command
704 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
705 * The @cmd value is turned from an encoding command and version into
706 * just the command code.
708 int ipc_parse_version(int *cmd)
718 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
720 #ifdef CONFIG_PROC_FS
721 struct ipc_proc_iter {
722 struct ipc_namespace *ns;
723 struct pid_namespace *pid_ns;
724 struct ipc_proc_iface *iface;
727 struct pid_namespace *ipc_seq_pid_ns(struct seq_file *s)
729 struct ipc_proc_iter *iter = s->private;
734 * This routine locks the ipc structure found at least at position pos.
736 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
739 struct kern_ipc_perm *ipc;
743 for (id = 0; id < pos && total < ids->in_use; id++) {
744 ipc = idr_find(&ids->ipcs_idr, id);
749 if (total >= ids->in_use)
752 for (; pos < ipc_mni; pos++) {
753 ipc = idr_find(&ids->ipcs_idr, pos);
757 ipc_lock_object(ipc);
762 /* Out of range - return NULL to terminate iteration */
766 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
768 struct ipc_proc_iter *iter = s->private;
769 struct ipc_proc_iface *iface = iter->iface;
770 struct kern_ipc_perm *ipc = it;
772 /* If we had an ipc id locked before, unlock it */
773 if (ipc && ipc != SEQ_START_TOKEN)
776 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
780 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
781 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
783 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
785 struct ipc_proc_iter *iter = s->private;
786 struct ipc_proc_iface *iface = iter->iface;
789 ids = &iter->ns->ids[iface->ids];
792 * Take the lock - this will be released by the corresponding
795 down_read(&ids->rwsem);
797 /* pos < 0 is invalid */
801 /* pos == 0 means header */
803 return SEQ_START_TOKEN;
805 /* Find the (pos-1)th ipc */
806 return sysvipc_find_ipc(ids, *pos - 1, pos);
809 static void sysvipc_proc_stop(struct seq_file *s, void *it)
811 struct kern_ipc_perm *ipc = it;
812 struct ipc_proc_iter *iter = s->private;
813 struct ipc_proc_iface *iface = iter->iface;
816 /* If we had a locked structure, release it */
817 if (ipc && ipc != SEQ_START_TOKEN)
820 ids = &iter->ns->ids[iface->ids];
821 /* Release the lock we took in start() */
822 up_read(&ids->rwsem);
825 static int sysvipc_proc_show(struct seq_file *s, void *it)
827 struct ipc_proc_iter *iter = s->private;
828 struct ipc_proc_iface *iface = iter->iface;
830 if (it == SEQ_START_TOKEN) {
831 seq_puts(s, iface->header);
835 return iface->show(s, it);
838 static const struct seq_operations sysvipc_proc_seqops = {
839 .start = sysvipc_proc_start,
840 .stop = sysvipc_proc_stop,
841 .next = sysvipc_proc_next,
842 .show = sysvipc_proc_show,
845 static int sysvipc_proc_open(struct inode *inode, struct file *file)
847 struct ipc_proc_iter *iter;
849 iter = __seq_open_private(file, &sysvipc_proc_seqops, sizeof(*iter));
853 iter->iface = PDE_DATA(inode);
854 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
855 iter->pid_ns = get_pid_ns(task_active_pid_ns(current));
860 static int sysvipc_proc_release(struct inode *inode, struct file *file)
862 struct seq_file *seq = file->private_data;
863 struct ipc_proc_iter *iter = seq->private;
864 put_ipc_ns(iter->ns);
865 put_pid_ns(iter->pid_ns);
866 return seq_release_private(inode, file);
869 static const struct file_operations sysvipc_proc_fops = {
870 .open = sysvipc_proc_open,
873 .release = sysvipc_proc_release,
875 #endif /* CONFIG_PROC_FS */