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>
67 #include <linux/log2.h>
69 #include <asm/unistd.h>
73 struct ipc_proc_iface {
77 int (*show)(struct seq_file *, void *);
81 * ipc_init - initialise ipc subsystem
83 * The various sysv ipc resources (semaphores, messages and shared
84 * memory) are initialised.
86 * A callback routine is registered into the memory hotplug notifier
87 * chain: since msgmni scales to lowmem this callback routine will be
88 * called upon successful memory add / remove to recompute msmgni.
90 static int __init ipc_init(void)
92 proc_mkdir("sysvipc", NULL);
99 device_initcall(ipc_init);
101 static const struct rhashtable_params ipc_kht_params = {
102 .head_offset = offsetof(struct kern_ipc_perm, khtnode),
103 .key_offset = offsetof(struct kern_ipc_perm, key),
104 .key_len = sizeof_field(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)
119 init_rwsem(&ids->rwsem);
120 rhashtable_init(&ids->key_ht, &ipc_kht_params);
121 idr_init(&ids->ipcs_idr);
124 #ifdef CONFIG_CHECKPOINT_RESTORE
129 #ifdef CONFIG_PROC_FS
130 static const struct proc_ops sysvipc_proc_ops;
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 * To conserve sequence number space, especially with extended ipc_mni,
199 * the sequence number is incremented only when the returned ID is less than
202 static inline int ipc_idr_alloc(struct ipc_ids *ids, struct kern_ipc_perm *new)
204 int idx, next_id = -1;
206 #ifdef CONFIG_CHECKPOINT_RESTORE
207 next_id = ids->next_id;
212 * As soon as a new object is inserted into the idr,
213 * ipc_obtain_object_idr() or ipc_obtain_object_check() can find it,
214 * and the lockless preparations for ipc operations can start.
215 * This means especially: permission checks, audit calls, allocation
216 * of undo structures, ...
218 * Thus the object must be fully initialized, and if something fails,
219 * then the full tear-down sequence must be followed.
220 * (i.e.: set new->deleted, reduce refcount, call_rcu())
223 if (next_id < 0) { /* !CHECKPOINT_RESTORE or next_id is unset */
226 max_idx = max(ids->in_use*3/2, ipc_min_cycle);
227 max_idx = min(max_idx, ipc_mni);
229 /* allocate the idx, with a NULL struct kern_ipc_perm */
230 idx = idr_alloc_cyclic(&ids->ipcs_idr, NULL, 0, max_idx,
235 * idx got allocated successfully.
236 * Now calculate the sequence number and set the
239 if (idx <= ids->last_idx) {
241 if (ids->seq >= ipcid_seq_max())
247 /* no need for smp_wmb(), this is done
248 * inside idr_replace, as part of
251 idr_replace(&ids->ipcs_idr, new, idx);
254 new->seq = ipcid_to_seqx(next_id);
255 idx = idr_alloc(&ids->ipcs_idr, new, ipcid_to_idx(next_id),
259 new->id = (new->seq << ipcmni_seq_shift()) + idx;
264 * ipc_addid - add an ipc identifier
265 * @ids: ipc identifier set
266 * @new: new ipc permission set
267 * @limit: limit for the number of used ids
269 * Add an entry 'new' to the ipc ids idr. The permissions object is
270 * initialised and the first free entry is set up and the index assigned
271 * is returned. The 'new' entry is returned in a locked state on success.
273 * On failure the entry is not locked and a negative err-code is returned.
274 * The caller must use ipc_rcu_putref() to free the identifier.
276 * Called with writer ipc_ids.rwsem held.
278 int ipc_addid(struct ipc_ids *ids, struct kern_ipc_perm *new, int limit)
284 /* 1) Initialize the refcount so that ipc_rcu_putref works */
285 refcount_set(&new->refcount, 1);
290 if (ids->in_use >= limit)
293 idr_preload(GFP_KERNEL);
295 spin_lock_init(&new->lock);
297 spin_lock(&new->lock);
299 current_euid_egid(&euid, &egid);
300 new->cuid = new->uid = euid;
301 new->gid = new->cgid = egid;
303 new->deleted = false;
305 idx = ipc_idr_alloc(ids, new);
308 if (idx >= 0 && new->key != IPC_PRIVATE) {
309 err = rhashtable_insert_fast(&ids->key_ht, &new->khtnode,
312 idr_remove(&ids->ipcs_idr, idx);
318 spin_unlock(&new->lock);
324 if (idx > ids->max_idx)
330 * ipcget_new - create a new ipc object
332 * @ids: ipc identifier set
333 * @ops: the actual creation routine to call
334 * @params: its parameters
336 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
337 * when the key is IPC_PRIVATE.
339 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
340 const struct ipc_ops *ops, struct ipc_params *params)
344 down_write(&ids->rwsem);
345 err = ops->getnew(ns, params);
346 up_write(&ids->rwsem);
351 * ipc_check_perms - check security and permissions for an ipc object
353 * @ipcp: ipc permission set
354 * @ops: the actual security routine to call
355 * @params: its parameters
357 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
358 * when the key is not IPC_PRIVATE and that key already exists in the
361 * On success, the ipc id is returned.
363 * It is called with ipc_ids.rwsem and ipcp->lock held.
365 static int ipc_check_perms(struct ipc_namespace *ns,
366 struct kern_ipc_perm *ipcp,
367 const struct ipc_ops *ops,
368 struct ipc_params *params)
372 if (ipcperms(ns, ipcp, params->flg))
375 err = ops->associate(ipcp, params->flg);
384 * ipcget_public - get an ipc object or create a new one
386 * @ids: ipc identifier set
387 * @ops: the actual creation routine to call
388 * @params: its parameters
390 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
391 * when the key is not IPC_PRIVATE.
392 * It adds a new entry if the key is not found and does some permission
393 * / security checkings if the key is found.
395 * On success, the ipc id is returned.
397 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
398 const struct ipc_ops *ops, struct ipc_params *params)
400 struct kern_ipc_perm *ipcp;
401 int flg = params->flg;
405 * Take the lock as a writer since we are potentially going to add
406 * a new entry + read locks are not "upgradable"
408 down_write(&ids->rwsem);
409 ipcp = ipc_findkey(ids, params->key);
412 if (!(flg & IPC_CREAT))
415 err = ops->getnew(ns, params);
417 /* ipc object has been locked by ipc_findkey() */
419 if (flg & IPC_CREAT && flg & IPC_EXCL)
423 if (ops->more_checks)
424 err = ops->more_checks(ipcp, params);
427 * ipc_check_perms returns the IPC id on
430 err = ipc_check_perms(ns, ipcp, ops, params);
434 up_write(&ids->rwsem);
440 * ipc_kht_remove - remove an ipc from the key hashtable
441 * @ids: ipc identifier set
442 * @ipcp: ipc perm structure containing the key to remove
444 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
445 * before this function is called, and remain locked on the exit.
447 static void ipc_kht_remove(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
449 if (ipcp->key != IPC_PRIVATE)
450 rhashtable_remove_fast(&ids->key_ht, &ipcp->khtnode,
455 * ipc_search_maxidx - search for the highest assigned index
456 * @ids: ipc identifier set
457 * @limit: known upper limit for highest assigned index
459 * The function determines the highest assigned index in @ids. It is intended
460 * to be called when ids->max_idx needs to be updated.
461 * Updating ids->max_idx is necessary when the current highest index ipc
463 * If no ipc object is allocated, then -1 is returned.
465 * ipc_ids.rwsem needs to be held by the caller.
467 static int ipc_search_maxidx(struct ipc_ids *ids, int limit)
476 for (; i >= 0; i--) {
477 tmpidx = retval | (1<<i);
479 * "0" is a possible index value, thus search using
480 * e.g. 15,7,3,1,0 instead of 16,8,4,2,1.
483 if (idr_get_next(&ids->ipcs_idr, &tmpidx))
490 * ipc_rmid - remove an ipc identifier
491 * @ids: ipc identifier set
492 * @ipcp: ipc perm structure containing the identifier to remove
494 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
495 * before this function is called, and remain locked on the exit.
497 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
499 int idx = ipcid_to_idx(ipcp->id);
501 idr_remove(&ids->ipcs_idr, idx);
502 ipc_kht_remove(ids, ipcp);
504 ipcp->deleted = true;
506 if (unlikely(idx == ids->max_idx)) {
507 idx = ids->max_idx-1;
509 idx = ipc_search_maxidx(ids, idx);
515 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
516 * @ids: ipc identifier set
517 * @ipcp: ipc perm structure containing the key to modify
519 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
520 * before this function is called, and remain locked on the exit.
522 void ipc_set_key_private(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
524 ipc_kht_remove(ids, ipcp);
525 ipcp->key = IPC_PRIVATE;
528 bool ipc_rcu_getref(struct kern_ipc_perm *ptr)
530 return refcount_inc_not_zero(&ptr->refcount);
533 void ipc_rcu_putref(struct kern_ipc_perm *ptr,
534 void (*func)(struct rcu_head *head))
536 if (!refcount_dec_and_test(&ptr->refcount))
539 call_rcu(&ptr->rcu, func);
543 * ipcperms - check ipc permissions
545 * @ipcp: ipc permission set
546 * @flag: desired permission set
548 * Check user, group, other permissions for access
549 * to ipc resources. return 0 if allowed
551 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
553 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
555 kuid_t euid = current_euid();
556 int requested_mode, granted_mode;
559 requested_mode = (flag >> 6) | (flag >> 3) | flag;
560 granted_mode = ipcp->mode;
561 if (uid_eq(euid, ipcp->cuid) ||
562 uid_eq(euid, ipcp->uid))
564 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
566 /* is there some bit set in requested_mode but not in granted_mode? */
567 if ((requested_mode & ~granted_mode & 0007) &&
568 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
571 return security_ipc_permission(ipcp, flag);
575 * Functions to convert between the kern_ipc_perm structure and the
576 * old/new ipc_perm structures
580 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
581 * @in: kernel permissions
582 * @out: new style ipc permissions
584 * Turn the kernel object @in into a set of permissions descriptions
585 * for returning to userspace (@out).
587 void kernel_to_ipc64_perm(struct kern_ipc_perm *in, struct ipc64_perm *out)
590 out->uid = from_kuid_munged(current_user_ns(), in->uid);
591 out->gid = from_kgid_munged(current_user_ns(), in->gid);
592 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
593 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
594 out->mode = in->mode;
599 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
600 * @in: new style ipc permissions
601 * @out: old style ipc permissions
603 * Turn the new style permissions object @in into a compatibility
604 * object and store it into the @out pointer.
606 void ipc64_perm_to_ipc_perm(struct ipc64_perm *in, struct ipc_perm *out)
609 SET_UID(out->uid, in->uid);
610 SET_GID(out->gid, in->gid);
611 SET_UID(out->cuid, in->cuid);
612 SET_GID(out->cgid, in->cgid);
613 out->mode = in->mode;
618 * ipc_obtain_object_idr
619 * @ids: ipc identifier set
620 * @id: ipc id to look for
622 * Look for an id in the ipc ids idr and return associated ipc object.
624 * Call inside the RCU critical section.
625 * The ipc object is *not* locked on exit.
627 struct kern_ipc_perm *ipc_obtain_object_idr(struct ipc_ids *ids, int id)
629 struct kern_ipc_perm *out;
630 int idx = ipcid_to_idx(id);
632 out = idr_find(&ids->ipcs_idr, idx);
634 return ERR_PTR(-EINVAL);
640 * ipc_obtain_object_check
641 * @ids: ipc identifier set
642 * @id: ipc id to look for
644 * Similar to ipc_obtain_object_idr() but also checks the ipc object
647 * Call inside the RCU critical section.
648 * The ipc object is *not* locked on exit.
650 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
652 struct kern_ipc_perm *out = ipc_obtain_object_idr(ids, id);
657 if (ipc_checkid(out, id))
658 return ERR_PTR(-EINVAL);
664 * ipcget - Common sys_*get() code
666 * @ids: ipc identifier set
667 * @ops: operations to be called on ipc object creation, permission checks
669 * @params: the parameters needed by the previous operations.
671 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
673 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
674 const struct ipc_ops *ops, struct ipc_params *params)
676 if (params->key == IPC_PRIVATE)
677 return ipcget_new(ns, ids, ops, params);
679 return ipcget_public(ns, ids, ops, params);
683 * ipc_update_perm - update the permissions of an ipc object
684 * @in: the permission given as input.
685 * @out: the permission of the ipc to set.
687 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
689 kuid_t uid = make_kuid(current_user_ns(), in->uid);
690 kgid_t gid = make_kgid(current_user_ns(), in->gid);
691 if (!uid_valid(uid) || !gid_valid(gid))
696 out->mode = (out->mode & ~S_IRWXUGO)
697 | (in->mode & S_IRWXUGO);
703 * ipcctl_obtain_check - retrieve an ipc object and check permissions
705 * @ids: the table of ids where to look for the ipc
706 * @id: the id of the ipc to retrieve
707 * @cmd: the cmd to check
708 * @perm: the permission to set
709 * @extra_perm: one extra permission parameter used by msq
711 * This function does some common audit and permissions check for some IPC_XXX
712 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
715 * - retrieves the ipc object with the given id in the given table.
716 * - performs some audit and permission check, depending on the given cmd
717 * - returns a pointer to the ipc object or otherwise, the corresponding
720 * Call holding the both the rwsem and the rcu read lock.
722 struct kern_ipc_perm *ipcctl_obtain_check(struct ipc_namespace *ns,
723 struct ipc_ids *ids, int id, int cmd,
724 struct ipc64_perm *perm, int extra_perm)
728 struct kern_ipc_perm *ipcp;
730 ipcp = ipc_obtain_object_check(ids, id);
738 audit_ipc_set_perm(extra_perm, perm->uid,
739 perm->gid, perm->mode);
741 euid = current_euid();
742 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
743 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
744 return ipcp; /* successful lookup */
749 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
753 * ipc_parse_version - ipc call version
754 * @cmd: pointer to command
756 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
757 * The @cmd value is turned from an encoding command and version into
758 * just the command code.
760 int ipc_parse_version(int *cmd)
770 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
772 #ifdef CONFIG_PROC_FS
773 struct ipc_proc_iter {
774 struct ipc_namespace *ns;
775 struct pid_namespace *pid_ns;
776 struct ipc_proc_iface *iface;
779 struct pid_namespace *ipc_seq_pid_ns(struct seq_file *s)
781 struct ipc_proc_iter *iter = s->private;
786 * This routine locks the ipc structure found at least at position pos.
788 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
791 struct kern_ipc_perm *ipc;
795 for (id = 0; id < pos && total < ids->in_use; id++) {
796 ipc = idr_find(&ids->ipcs_idr, id);
802 if (total >= ids->in_use)
805 for (; pos < ipc_mni; pos++) {
806 ipc = idr_find(&ids->ipcs_idr, pos);
809 ipc_lock_object(ipc);
818 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
820 struct ipc_proc_iter *iter = s->private;
821 struct ipc_proc_iface *iface = iter->iface;
822 struct kern_ipc_perm *ipc = it;
824 /* If we had an ipc id locked before, unlock it */
825 if (ipc && ipc != SEQ_START_TOKEN)
828 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
832 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
833 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
835 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
837 struct ipc_proc_iter *iter = s->private;
838 struct ipc_proc_iface *iface = iter->iface;
841 ids = &iter->ns->ids[iface->ids];
844 * Take the lock - this will be released by the corresponding
847 down_read(&ids->rwsem);
849 /* pos < 0 is invalid */
853 /* pos == 0 means header */
855 return SEQ_START_TOKEN;
857 /* Find the (pos-1)th ipc */
858 return sysvipc_find_ipc(ids, *pos - 1, pos);
861 static void sysvipc_proc_stop(struct seq_file *s, void *it)
863 struct kern_ipc_perm *ipc = it;
864 struct ipc_proc_iter *iter = s->private;
865 struct ipc_proc_iface *iface = iter->iface;
868 /* If we had a locked structure, release it */
869 if (ipc && ipc != SEQ_START_TOKEN)
872 ids = &iter->ns->ids[iface->ids];
873 /* Release the lock we took in start() */
874 up_read(&ids->rwsem);
877 static int sysvipc_proc_show(struct seq_file *s, void *it)
879 struct ipc_proc_iter *iter = s->private;
880 struct ipc_proc_iface *iface = iter->iface;
882 if (it == SEQ_START_TOKEN) {
883 seq_puts(s, iface->header);
887 return iface->show(s, it);
890 static const struct seq_operations sysvipc_proc_seqops = {
891 .start = sysvipc_proc_start,
892 .stop = sysvipc_proc_stop,
893 .next = sysvipc_proc_next,
894 .show = sysvipc_proc_show,
897 static int sysvipc_proc_open(struct inode *inode, struct file *file)
899 struct ipc_proc_iter *iter;
901 iter = __seq_open_private(file, &sysvipc_proc_seqops, sizeof(*iter));
905 iter->iface = PDE_DATA(inode);
906 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
907 iter->pid_ns = get_pid_ns(task_active_pid_ns(current));
912 static int sysvipc_proc_release(struct inode *inode, struct file *file)
914 struct seq_file *seq = file->private_data;
915 struct ipc_proc_iter *iter = seq->private;
916 put_ipc_ns(iter->ns);
917 put_pid_ns(iter->pid_ns);
918 return seq_release_private(inode, file);
921 static const struct proc_ops sysvipc_proc_ops = {
922 .proc_flags = PROC_ENTRY_PERMANENT,
923 .proc_open = sysvipc_proc_open,
924 .proc_read = seq_read,
925 .proc_lseek = seq_lseek,
926 .proc_release = sysvipc_proc_release,
928 #endif /* CONFIG_PROC_FS */