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.
10 * Nov 1999 - ipc helper functions, unified SMP locking
12 * Oct 2002 - One lock per IPC id. RCU ipc_free for lock-free grow_ary().
14 * Mar 2006 - support for audit of ipc object properties
16 * Jun 2006 - namespaces ssupport
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>
67 #include <asm/unistd.h>
71 struct ipc_proc_iface {
75 int (*show)(struct seq_file *, void *);
79 * ipc_init - initialise ipc subsystem
81 * The various sysv ipc resources (semaphores, messages and shared
82 * memory) are initialised.
84 * A callback routine is registered into the memory hotplug notifier
85 * chain: since msgmni scales to lowmem this callback routine will be
86 * called upon successful memory add / remove to recompute msmgni.
88 static int __init ipc_init(void)
93 WARN(err_sem, "ipc: sysv sem_init failed: %d\n", err_sem);
95 WARN(err_msg, "ipc: sysv msg_init failed: %d\n", err_msg);
98 return err_msg ? err_msg : err_sem;
100 device_initcall(ipc_init);
102 static const struct rhashtable_params ipc_kht_params = {
103 .head_offset = offsetof(struct kern_ipc_perm, khtnode),
104 .key_offset = offsetof(struct kern_ipc_perm, key),
105 .key_len = FIELD_SIZEOF(struct kern_ipc_perm, key),
107 .automatic_shrinking = true,
111 * ipc_init_ids - initialise ipc identifiers
112 * @ids: ipc identifier set
114 * Set up the sequence range to use for the ipc identifier range (limited
115 * below IPCMNI) then initialise the keys hashtable and ids idr.
117 int ipc_init_ids(struct ipc_ids *ids)
122 init_rwsem(&ids->rwsem);
123 err = rhashtable_init(&ids->key_ht, &ipc_kht_params);
126 idr_init(&ids->ipcs_idr);
127 ids->tables_initialized = true;
129 #ifdef CONFIG_CHECKPOINT_RESTORE
135 #ifdef CONFIG_PROC_FS
136 static const struct file_operations sysvipc_proc_fops;
138 * ipc_init_proc_interface - create a proc interface for sysipc types using a seq_file interface.
139 * @path: Path in procfs
140 * @header: Banner to be printed at the beginning of the file.
141 * @ids: ipc id table to iterate.
142 * @show: show routine.
144 void __init ipc_init_proc_interface(const char *path, const char *header,
145 int ids, int (*show)(struct seq_file *, void *))
147 struct proc_dir_entry *pde;
148 struct ipc_proc_iface *iface;
150 iface = kmalloc(sizeof(*iface), GFP_KERNEL);
154 iface->header = header;
158 pde = proc_create_data(path,
159 S_IRUGO, /* world readable */
160 NULL, /* parent dir */
169 * ipc_findkey - find a key in an ipc identifier set
170 * @ids: ipc identifier set
173 * Returns the locked pointer to the ipc structure if found or NULL
174 * otherwise. If key is found ipc points to the owning ipc structure
176 * Called with writer ipc_ids.rwsem held.
178 static struct kern_ipc_perm *ipc_findkey(struct ipc_ids *ids, key_t key)
180 struct kern_ipc_perm *ipcp = NULL;
182 if (likely(ids->tables_initialized))
183 ipcp = rhashtable_lookup_fast(&ids->key_ht, &key,
188 ipc_lock_object(ipcp);
195 #ifdef CONFIG_CHECKPOINT_RESTORE
197 * Specify desired id for next allocated IPC object.
199 #define ipc_idr_alloc(ids, new) \
200 idr_alloc(&(ids)->ipcs_idr, (new), \
201 (ids)->next_id < 0 ? 0 : ipcid_to_idx((ids)->next_id),\
204 static inline int ipc_buildid(int id, struct ipc_ids *ids,
205 struct kern_ipc_perm *new)
207 if (ids->next_id < 0) { /* default, behave as !CHECKPOINT_RESTORE */
208 new->seq = ids->seq++;
209 if (ids->seq > IPCID_SEQ_MAX)
212 new->seq = ipcid_to_seqx(ids->next_id);
216 return SEQ_MULTIPLIER * new->seq + id;
220 #define ipc_idr_alloc(ids, new) \
221 idr_alloc(&(ids)->ipcs_idr, (new), 0, 0, GFP_NOWAIT)
223 static inline int ipc_buildid(int id, struct ipc_ids *ids,
224 struct kern_ipc_perm *new)
226 new->seq = ids->seq++;
227 if (ids->seq > IPCID_SEQ_MAX)
230 return SEQ_MULTIPLIER * new->seq + id;
233 #endif /* CONFIG_CHECKPOINT_RESTORE */
236 * ipc_addid - add an ipc identifier
237 * @ids: ipc identifier set
238 * @new: new ipc permission set
239 * @limit: limit for the number of used ids
241 * Add an entry 'new' to the ipc ids idr. The permissions object is
242 * initialised and the first free entry is set up and the id assigned
243 * is returned. The 'new' entry is returned in a locked state on success.
244 * On failure the entry is not locked and a negative err-code is returned.
246 * Called with writer ipc_ids.rwsem held.
248 int ipc_addid(struct ipc_ids *ids, struct kern_ipc_perm *new, int limit)
257 if (!ids->tables_initialized || ids->in_use >= limit)
260 idr_preload(GFP_KERNEL);
262 refcount_set(&new->refcount, 1);
263 spin_lock_init(&new->lock);
264 new->deleted = false;
266 spin_lock(&new->lock);
268 current_euid_egid(&euid, &egid);
269 new->cuid = new->uid = euid;
270 new->gid = new->cgid = egid;
272 id = ipc_idr_alloc(ids, new);
275 if (id >= 0 && new->key != IPC_PRIVATE) {
276 err = rhashtable_insert_fast(&ids->key_ht, &new->khtnode,
279 idr_remove(&ids->ipcs_idr, id);
284 spin_unlock(&new->lock);
290 if (id > ids->max_id)
293 new->id = ipc_buildid(id, ids, new);
299 * ipcget_new - create a new ipc object
301 * @ids: ipc identifier set
302 * @ops: the actual creation routine to call
303 * @params: its parameters
305 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
306 * when the key is IPC_PRIVATE.
308 static int ipcget_new(struct ipc_namespace *ns, struct ipc_ids *ids,
309 const struct ipc_ops *ops, struct ipc_params *params)
313 down_write(&ids->rwsem);
314 err = ops->getnew(ns, params);
315 up_write(&ids->rwsem);
320 * ipc_check_perms - check security and permissions for an ipc object
322 * @ipcp: ipc permission set
323 * @ops: the actual security routine to call
324 * @params: its parameters
326 * This routine is called by sys_msgget(), sys_semget() and sys_shmget()
327 * when the key is not IPC_PRIVATE and that key already exists in the
330 * On success, the ipc id is returned.
332 * It is called with ipc_ids.rwsem and ipcp->lock held.
334 static int ipc_check_perms(struct ipc_namespace *ns,
335 struct kern_ipc_perm *ipcp,
336 const struct ipc_ops *ops,
337 struct ipc_params *params)
341 if (ipcperms(ns, ipcp, params->flg))
344 err = ops->associate(ipcp, params->flg);
353 * ipcget_public - get an ipc object or create a new one
355 * @ids: ipc identifier set
356 * @ops: the actual creation routine to call
357 * @params: its parameters
359 * This routine is called by sys_msgget, sys_semget() and sys_shmget()
360 * when the key is not IPC_PRIVATE.
361 * It adds a new entry if the key is not found and does some permission
362 * / security checkings if the key is found.
364 * On success, the ipc id is returned.
366 static int ipcget_public(struct ipc_namespace *ns, struct ipc_ids *ids,
367 const struct ipc_ops *ops, struct ipc_params *params)
369 struct kern_ipc_perm *ipcp;
370 int flg = params->flg;
374 * Take the lock as a writer since we are potentially going to add
375 * a new entry + read locks are not "upgradable"
377 down_write(&ids->rwsem);
378 ipcp = ipc_findkey(ids, params->key);
381 if (!(flg & IPC_CREAT))
384 err = ops->getnew(ns, params);
386 /* ipc object has been locked by ipc_findkey() */
388 if (flg & IPC_CREAT && flg & IPC_EXCL)
392 if (ops->more_checks)
393 err = ops->more_checks(ipcp, params);
396 * ipc_check_perms returns the IPC id on
399 err = ipc_check_perms(ns, ipcp, ops, params);
403 up_write(&ids->rwsem);
409 * ipc_kht_remove - remove an ipc from the key hashtable
410 * @ids: ipc identifier set
411 * @ipcp: ipc perm structure containing the key to remove
413 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
414 * before this function is called, and remain locked on the exit.
416 static void ipc_kht_remove(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
418 if (ipcp->key != IPC_PRIVATE)
419 rhashtable_remove_fast(&ids->key_ht, &ipcp->khtnode,
424 * ipc_rmid - remove an ipc identifier
425 * @ids: ipc identifier set
426 * @ipcp: ipc perm structure containing the identifier to remove
428 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
429 * before this function is called, and remain locked on the exit.
431 void ipc_rmid(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
433 int lid = ipcid_to_idx(ipcp->id);
435 idr_remove(&ids->ipcs_idr, lid);
436 ipc_kht_remove(ids, ipcp);
438 ipcp->deleted = true;
440 if (unlikely(lid == ids->max_id)) {
445 } while (!idr_find(&ids->ipcs_idr, lid));
451 * ipc_set_key_private - switch the key of an existing ipc to IPC_PRIVATE
452 * @ids: ipc identifier set
453 * @ipcp: ipc perm structure containing the key to modify
455 * ipc_ids.rwsem (as a writer) and the spinlock for this ID are held
456 * before this function is called, and remain locked on the exit.
458 void ipc_set_key_private(struct ipc_ids *ids, struct kern_ipc_perm *ipcp)
460 ipc_kht_remove(ids, ipcp);
461 ipcp->key = IPC_PRIVATE;
464 int ipc_rcu_getref(struct kern_ipc_perm *ptr)
466 return refcount_inc_not_zero(&ptr->refcount);
469 void ipc_rcu_putref(struct kern_ipc_perm *ptr,
470 void (*func)(struct rcu_head *head))
472 if (!refcount_dec_and_test(&ptr->refcount))
475 call_rcu(&ptr->rcu, func);
479 * ipcperms - check ipc permissions
481 * @ipcp: ipc permission set
482 * @flag: desired permission set
484 * Check user, group, other permissions for access
485 * to ipc resources. return 0 if allowed
487 * @flag will most probably be 0 or ``S_...UGO`` from <linux/stat.h>
489 int ipcperms(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp, short flag)
491 kuid_t euid = current_euid();
492 int requested_mode, granted_mode;
495 requested_mode = (flag >> 6) | (flag >> 3) | flag;
496 granted_mode = ipcp->mode;
497 if (uid_eq(euid, ipcp->cuid) ||
498 uid_eq(euid, ipcp->uid))
500 else if (in_group_p(ipcp->cgid) || in_group_p(ipcp->gid))
502 /* is there some bit set in requested_mode but not in granted_mode? */
503 if ((requested_mode & ~granted_mode & 0007) &&
504 !ns_capable(ns->user_ns, CAP_IPC_OWNER))
507 return security_ipc_permission(ipcp, flag);
511 * Functions to convert between the kern_ipc_perm structure and the
512 * old/new ipc_perm structures
516 * kernel_to_ipc64_perm - convert kernel ipc permissions to user
517 * @in: kernel permissions
518 * @out: new style ipc permissions
520 * Turn the kernel object @in into a set of permissions descriptions
521 * for returning to userspace (@out).
523 void kernel_to_ipc64_perm(struct kern_ipc_perm *in, struct ipc64_perm *out)
526 out->uid = from_kuid_munged(current_user_ns(), in->uid);
527 out->gid = from_kgid_munged(current_user_ns(), in->gid);
528 out->cuid = from_kuid_munged(current_user_ns(), in->cuid);
529 out->cgid = from_kgid_munged(current_user_ns(), in->cgid);
530 out->mode = in->mode;
535 * ipc64_perm_to_ipc_perm - convert new ipc permissions to old
536 * @in: new style ipc permissions
537 * @out: old style ipc permissions
539 * Turn the new style permissions object @in into a compatibility
540 * object and store it into the @out pointer.
542 void ipc64_perm_to_ipc_perm(struct ipc64_perm *in, struct ipc_perm *out)
545 SET_UID(out->uid, in->uid);
546 SET_GID(out->gid, in->gid);
547 SET_UID(out->cuid, in->cuid);
548 SET_GID(out->cgid, in->cgid);
549 out->mode = in->mode;
554 * ipc_obtain_object_idr
555 * @ids: ipc identifier set
556 * @id: ipc id to look for
558 * Look for an id in the ipc ids idr and return associated ipc object.
560 * Call inside the RCU critical section.
561 * The ipc object is *not* locked on exit.
563 struct kern_ipc_perm *ipc_obtain_object_idr(struct ipc_ids *ids, int id)
565 struct kern_ipc_perm *out;
566 int lid = ipcid_to_idx(id);
568 if (unlikely(!ids->tables_initialized))
569 return ERR_PTR(-EINVAL);
571 out = idr_find(&ids->ipcs_idr, lid);
573 return ERR_PTR(-EINVAL);
579 * ipc_lock - lock an ipc structure without rwsem held
580 * @ids: ipc identifier set
581 * @id: ipc id to look for
583 * Look for an id in the ipc ids idr and lock the associated ipc object.
585 * The ipc object is locked on successful exit.
587 struct kern_ipc_perm *ipc_lock(struct ipc_ids *ids, int id)
589 struct kern_ipc_perm *out;
592 out = ipc_obtain_object_idr(ids, id);
596 spin_lock(&out->lock);
599 * ipc_rmid() may have already freed the ID while ipc_lock()
600 * was spinning: here verify that the structure is still valid.
601 * Upon races with RMID, return -EIDRM, thus indicating that
602 * the ID points to a removed identifier.
604 if (ipc_valid_object(out))
607 spin_unlock(&out->lock);
608 out = ERR_PTR(-EIDRM);
615 * ipc_obtain_object_check
616 * @ids: ipc identifier set
617 * @id: ipc id to look for
619 * Similar to ipc_obtain_object_idr() but also checks
620 * the ipc object reference counter.
622 * Call inside the RCU critical section.
623 * The ipc object is *not* locked on exit.
625 struct kern_ipc_perm *ipc_obtain_object_check(struct ipc_ids *ids, int id)
627 struct kern_ipc_perm *out = ipc_obtain_object_idr(ids, id);
632 if (ipc_checkid(out, id))
633 return ERR_PTR(-EINVAL);
639 * ipcget - Common sys_*get() code
641 * @ids: ipc identifier set
642 * @ops: operations to be called on ipc object creation, permission checks
644 * @params: the parameters needed by the previous operations.
646 * Common routine called by sys_msgget(), sys_semget() and sys_shmget().
648 int ipcget(struct ipc_namespace *ns, struct ipc_ids *ids,
649 const struct ipc_ops *ops, struct ipc_params *params)
651 if (params->key == IPC_PRIVATE)
652 return ipcget_new(ns, ids, ops, params);
654 return ipcget_public(ns, ids, ops, params);
658 * ipc_update_perm - update the permissions of an ipc object
659 * @in: the permission given as input.
660 * @out: the permission of the ipc to set.
662 int ipc_update_perm(struct ipc64_perm *in, struct kern_ipc_perm *out)
664 kuid_t uid = make_kuid(current_user_ns(), in->uid);
665 kgid_t gid = make_kgid(current_user_ns(), in->gid);
666 if (!uid_valid(uid) || !gid_valid(gid))
671 out->mode = (out->mode & ~S_IRWXUGO)
672 | (in->mode & S_IRWXUGO);
678 * ipcctl_pre_down_nolock - retrieve an ipc and check permissions for some IPC_XXX cmd
680 * @ids: the table of ids where to look for the ipc
681 * @id: the id of the ipc to retrieve
682 * @cmd: the cmd to check
683 * @perm: the permission to set
684 * @extra_perm: one extra permission parameter used by msq
686 * This function does some common audit and permissions check for some IPC_XXX
687 * cmd and is called from semctl_down, shmctl_down and msgctl_down.
688 * It must be called without any lock held and:
690 * - retrieves the ipc with the given id in the given table.
691 * - performs some audit and permission check, depending on the given cmd
692 * - returns a pointer to the ipc object or otherwise, the corresponding
695 * Call holding the both the rwsem and the rcu read lock.
697 struct kern_ipc_perm *ipcctl_pre_down_nolock(struct ipc_namespace *ns,
698 struct ipc_ids *ids, int id, int cmd,
699 struct ipc64_perm *perm, int extra_perm)
703 struct kern_ipc_perm *ipcp;
705 ipcp = ipc_obtain_object_check(ids, id);
713 audit_ipc_set_perm(extra_perm, perm->uid,
714 perm->gid, perm->mode);
716 euid = current_euid();
717 if (uid_eq(euid, ipcp->cuid) || uid_eq(euid, ipcp->uid) ||
718 ns_capable(ns->user_ns, CAP_SYS_ADMIN))
719 return ipcp; /* successful lookup */
724 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
728 * ipc_parse_version - ipc call version
729 * @cmd: pointer to command
731 * Return IPC_64 for new style IPC and IPC_OLD for old style IPC.
732 * The @cmd value is turned from an encoding command and version into
733 * just the command code.
735 int ipc_parse_version(int *cmd)
745 #endif /* CONFIG_ARCH_WANT_IPC_PARSE_VERSION */
747 #ifdef CONFIG_PROC_FS
748 struct ipc_proc_iter {
749 struct ipc_namespace *ns;
750 struct pid_namespace *pid_ns;
751 struct ipc_proc_iface *iface;
754 struct pid_namespace *ipc_seq_pid_ns(struct seq_file *s)
756 struct ipc_proc_iter *iter = s->private;
761 * This routine locks the ipc structure found at least at position pos.
763 static struct kern_ipc_perm *sysvipc_find_ipc(struct ipc_ids *ids, loff_t pos,
766 struct kern_ipc_perm *ipc;
770 for (id = 0; id < pos && total < ids->in_use; id++) {
771 ipc = idr_find(&ids->ipcs_idr, id);
776 if (total >= ids->in_use)
779 for (; pos < IPCMNI; pos++) {
780 ipc = idr_find(&ids->ipcs_idr, pos);
784 ipc_lock_object(ipc);
789 /* Out of range - return NULL to terminate iteration */
793 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
795 struct ipc_proc_iter *iter = s->private;
796 struct ipc_proc_iface *iface = iter->iface;
797 struct kern_ipc_perm *ipc = it;
799 /* If we had an ipc id locked before, unlock it */
800 if (ipc && ipc != SEQ_START_TOKEN)
803 return sysvipc_find_ipc(&iter->ns->ids[iface->ids], *pos, pos);
807 * File positions: pos 0 -> header, pos n -> ipc id = n - 1.
808 * SeqFile iterator: iterator value locked ipc pointer or SEQ_TOKEN_START.
810 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
812 struct ipc_proc_iter *iter = s->private;
813 struct ipc_proc_iface *iface = iter->iface;
816 ids = &iter->ns->ids[iface->ids];
819 * Take the lock - this will be released by the corresponding
822 down_read(&ids->rwsem);
824 /* pos < 0 is invalid */
828 /* pos == 0 means header */
830 return SEQ_START_TOKEN;
832 /* Find the (pos-1)th ipc */
833 return sysvipc_find_ipc(ids, *pos - 1, pos);
836 static void sysvipc_proc_stop(struct seq_file *s, void *it)
838 struct kern_ipc_perm *ipc = it;
839 struct ipc_proc_iter *iter = s->private;
840 struct ipc_proc_iface *iface = iter->iface;
843 /* If we had a locked structure, release it */
844 if (ipc && ipc != SEQ_START_TOKEN)
847 ids = &iter->ns->ids[iface->ids];
848 /* Release the lock we took in start() */
849 up_read(&ids->rwsem);
852 static int sysvipc_proc_show(struct seq_file *s, void *it)
854 struct ipc_proc_iter *iter = s->private;
855 struct ipc_proc_iface *iface = iter->iface;
857 if (it == SEQ_START_TOKEN) {
858 seq_puts(s, iface->header);
862 return iface->show(s, it);
865 static const struct seq_operations sysvipc_proc_seqops = {
866 .start = sysvipc_proc_start,
867 .stop = sysvipc_proc_stop,
868 .next = sysvipc_proc_next,
869 .show = sysvipc_proc_show,
872 static int sysvipc_proc_open(struct inode *inode, struct file *file)
874 struct ipc_proc_iter *iter;
876 iter = __seq_open_private(file, &sysvipc_proc_seqops, sizeof(*iter));
880 iter->iface = PDE_DATA(inode);
881 iter->ns = get_ipc_ns(current->nsproxy->ipc_ns);
882 iter->pid_ns = get_pid_ns(task_active_pid_ns(current));
887 static int sysvipc_proc_release(struct inode *inode, struct file *file)
889 struct seq_file *seq = file->private_data;
890 struct ipc_proc_iter *iter = seq->private;
891 put_ipc_ns(iter->ns);
892 put_pid_ns(iter->pid_ns);
893 return seq_release_private(inode, file);
896 static const struct file_operations sysvipc_proc_fops = {
897 .open = sysvipc_proc_open,
900 .release = sysvipc_proc_release,
902 #endif /* CONFIG_PROC_FS */