blob: 6d767071c3673dce09984933c3a444d78880bd10 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/ipc/shm.c
3 * Copyright (C) 1992, 1993 Krishna Balasubramanian
4 * Many improvements/fixes by Bruno Haible.
5 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7 *
8 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15 *
Steve Grubb073115d2006-04-02 17:07:33 -040016 * support for audit of ipc object properties and permission changes
17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaev4e982312006-10-02 02:18:22 -070018 *
19 * namespaces support
20 * OpenVZ, SWsoft Inc.
21 * Pavel Emelianov <xemul@openvz.org>
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -070022 *
23 * Better ipc lock (kern_ipc_perm.lock) handling
24 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 */
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/slab.h>
28#include <linux/mm.h>
29#include <linux/hugetlb.h>
30#include <linux/shm.h>
31#include <linux/init.h>
32#include <linux/file.h>
33#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/shmem_fs.h>
35#include <linux/security.h>
36#include <linux/syscalls.h>
37#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080038#include <linux/capability.h>
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -070039#include <linux/ptrace.h>
Mike Waychison19b49462005-09-06 15:17:10 -070040#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070041#include <linux/rwsem.h>
Kirill Korotaev4e982312006-10-02 02:18:22 -070042#include <linux/nsproxy.h>
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080043#include <linux/mount.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080044#include <linux/ipc_namespace.h>
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -070045
Paul McQuade7153e402014-06-06 14:37:37 -070046#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#include "util.h"
49
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080050struct shm_file_data {
51 int id;
52 struct ipc_namespace *ns;
53 struct file *file;
54 const struct vm_operations_struct *vm_ops;
55};
56
57#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
58
Arjan van de Ven9a321442007-02-12 00:55:35 -080059static const struct file_operations shm_file_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040060static const struct vm_operations_struct shm_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080062#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Kirill Korotaev4e982312006-10-02 02:18:22 -070064#define shm_unlock(shp) \
65 ipc_unlock(&(shp)->shm_perm)
Kirill Korotaev4e982312006-10-02 02:18:22 -070066
Nadia Derbey7748dbf2007-10-18 23:40:49 -070067static int newseg(struct ipc_namespace *, struct ipc_params *);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080068static void shm_open(struct vm_area_struct *vma);
69static void shm_close(struct vm_area_struct *vma);
Manfred Spraul239521f2014-01-27 17:07:04 -080070static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -070072static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#endif
74
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080075void shm_init_ns(struct ipc_namespace *ns)
Kirill Korotaev4e982312006-10-02 02:18:22 -070076{
Kirill Korotaev4e982312006-10-02 02:18:22 -070077 ns->shm_ctlmax = SHMMAX;
78 ns->shm_ctlall = SHMALL;
79 ns->shm_ctlmni = SHMMNI;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -070080 ns->shm_rmid_forced = 0;
Kirill Korotaev4e982312006-10-02 02:18:22 -070081 ns->shm_tot = 0;
WANG Conge8148f72009-01-06 14:42:49 -080082 ipc_init_ids(&shm_ids(ns));
Kirill Korotaev4e982312006-10-02 02:18:22 -070083}
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Nadia Derbeyf4566f02007-10-18 23:40:53 -070085/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -070086 * Called with shm_ids.rwsem (writer) and the shp structure locked.
87 * Only shm_ids.rwsem remains locked on exit.
Nadia Derbeyf4566f02007-10-18 23:40:53 -070088 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -080089static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Kirill Korotaev4e982312006-10-02 02:18:22 -070090{
Pierre Peiffer01b8b072008-02-08 04:18:57 -080091 struct shmid_kernel *shp;
92 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
93
Manfred Spraul239521f2014-01-27 17:07:04 -080094 if (shp->shm_nattch) {
Kirill Korotaev4e982312006-10-02 02:18:22 -070095 shp->shm_perm.mode |= SHM_DEST;
96 /* Do not find it any more */
97 shp->shm_perm.key = IPC_PRIVATE;
98 shm_unlock(shp);
99 } else
100 shm_destroy(ns, shp);
101}
102
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800103#ifdef CONFIG_IPC_NS
Kirill Korotaev4e982312006-10-02 02:18:22 -0700104void shm_exit_ns(struct ipc_namespace *ns)
105{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800106 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800107 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700108}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800109#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Linus Torvalds140d0b22011-08-04 19:35:59 -1000111static int __init ipc_ns_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800113 shm_init_ns(&init_ipc_ns);
Linus Torvalds140d0b22011-08-04 19:35:59 -1000114 return 0;
115}
116
117pure_initcall(ipc_ns_init);
118
Manfred Spraul239521f2014-01-27 17:07:04 -0800119void __init shm_init(void)
Linus Torvalds140d0b22011-08-04 19:35:59 -1000120{
Mike Waychison19b49462005-09-06 15:17:10 -0700121 ipc_init_proc_interface("sysvipc/shm",
Helge Dellerb7952182010-10-27 15:34:16 -0700122#if BITS_PER_LONG <= 32
123 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
124#else
125 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
126#endif
Kirill Korotaev4e982312006-10-02 02:18:22 -0700127 IPC_SHM_IDS, sysvipc_shm_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128}
129
Davidlohr Bueso8b8d52a2013-09-11 14:26:15 -0700130static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
131{
132 struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
133
134 if (IS_ERR(ipcp))
135 return ERR_CAST(ipcp);
136
137 return container_of(ipcp, struct shmid_kernel, shm_perm);
138}
139
140static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
141{
142 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
143
144 if (IS_ERR(ipcp))
145 return ERR_CAST(ipcp);
146
147 return container_of(ipcp, struct shmid_kernel, shm_perm);
148}
149
Nadia Derbey3e148c72007-10-18 23:40:54 -0700150/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700151 * shm_lock_(check_) routines are called in the paths where the rwsem
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700152 * is not necessarily held.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700153 */
Nadia Derbey023a5352007-10-18 23:40:51 -0700154static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700156 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
157
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800158 if (IS_ERR(ipcp))
159 return (struct shmid_kernel *)ipcp;
160
Nadia Derbey03f02c72007-10-18 23:40:51 -0700161 return container_of(ipcp, struct shmid_kernel, shm_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700162}
163
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400164static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
165{
166 rcu_read_lock();
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700167 ipc_lock_object(&ipcp->shm_perm);
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400168}
169
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700170static void shm_rcu_free(struct rcu_head *head)
171{
172 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
173 struct shmid_kernel *shp = ipc_rcu_to_struct(p);
174
175 security_shm_free(shp);
176 ipc_rcu_free(head);
177}
178
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700179static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Jack Millerab602f72014-08-08 14:23:19 -0700181 list_del(&s->shm_clist);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700182 ipc_rmid(&shm_ids(ns), &s->shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800186/* This is called by fork, once for every shm attach. */
187static void shm_open(struct vm_area_struct *vma)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700188{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800189 struct file *file = vma->vm_file;
190 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 struct shmid_kernel *shp;
192
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800193 shp = shm_lock(sfd->ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700194 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 shp->shm_atim = get_seconds();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700196 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 shp->shm_nattch++;
198 shm_unlock(shp);
199}
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201/*
202 * shm_destroy - free the struct shmid_kernel
203 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700204 * @ns: namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 * @shp: struct to free
206 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700207 * It has to be called with shp and shm_ids.rwsem (writer) locked,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 * but returns with shp unlocked and freed.
209 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700210static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Greg Thelena399b292013-11-21 14:32:00 -0800212 struct file *shm_file;
213
214 shm_file = shp->shm_file;
215 shp->shm_file = NULL;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700216 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700217 shm_rmid(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 shm_unlock(shp);
Greg Thelena399b292013-11-21 14:32:00 -0800219 if (!is_file_hugepages(shm_file))
220 shmem_lock(shm_file, 0, shp->mlock_user);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100221 else if (shp->mlock_user)
Dave Hansen07a46ed2014-12-12 16:58:22 -0800222 user_shm_unlock(i_size_read(file_inode(shm_file)),
223 shp->mlock_user);
Greg Thelena399b292013-11-21 14:32:00 -0800224 fput(shm_file);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700225 ipc_rcu_putref(shp, shm_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
228/*
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700229 * shm_may_destroy - identifies whether shm segment should be destroyed now
230 *
231 * Returns true if and only if there are no active users of the segment and
232 * one of the following is true:
233 *
234 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
235 *
236 * 2) sysctl kernel.shm_rmid_forced is set to 1.
237 */
238static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
239{
240 return (shp->shm_nattch == 0) &&
241 (ns->shm_rmid_forced ||
242 (shp->shm_perm.mode & SHM_DEST));
243}
244
245/*
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800246 * remove the attach descriptor vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 * free memory for segment if it is marked destroyed.
248 * The descriptor has already been removed from the current->mm->mmap list
249 * and will later be kfree()d.
250 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800251static void shm_close(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Manfred Spraul239521f2014-01-27 17:07:04 -0800253 struct file *file = vma->vm_file;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800254 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 struct shmid_kernel *shp;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800256 struct ipc_namespace *ns = sfd->ns;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700257
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700258 down_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 /* remove from the list of attaches of the shm segment */
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700260 shp = shm_lock(ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700261 BUG_ON(IS_ERR(shp));
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700262 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 shp->shm_dtim = get_seconds();
264 shp->shm_nattch--;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700265 if (shm_may_destroy(ns, shp))
Kirill Korotaev4e982312006-10-02 02:18:22 -0700266 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 else
268 shm_unlock(shp);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700269 up_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700272/* Called with ns->shm_ids(ns).rwsem locked */
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700273static int shm_try_destroy_orphaned(int id, void *p, void *data)
274{
275 struct ipc_namespace *ns = data;
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400276 struct kern_ipc_perm *ipcp = p;
277 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700278
279 /*
280 * We want to destroy segments without users and with already
281 * exit'ed originating process.
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400282 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700283 * As shp->* are changed under rwsem, it's safe to skip shp locking.
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700284 */
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400285 if (shp->shm_creator != NULL)
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700286 return 0;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700287
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400288 if (shm_may_destroy(ns, shp)) {
289 shm_lock_by_ptr(shp);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700290 shm_destroy(ns, shp);
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400291 }
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700292 return 0;
293}
294
295void shm_destroy_orphaned(struct ipc_namespace *ns)
296{
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700297 down_write(&shm_ids(ns).rwsem);
Vasiliy Kulikov33a30ed2011-08-03 22:26:55 +0400298 if (shm_ids(ns).in_use)
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400299 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700300 up_write(&shm_ids(ns).rwsem);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700301}
302
Jack Miller83293c02014-08-08 14:23:21 -0700303/* Locking assumes this will only be called with task == current */
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700304void exit_shm(struct task_struct *task)
305{
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400306 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
Jack Millerab602f72014-08-08 14:23:19 -0700307 struct shmid_kernel *shp, *n;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700308
Jack Miller83293c02014-08-08 14:23:21 -0700309 if (list_empty(&task->sysvshm.shm_clist))
Vasiliy Kulikov298507d2011-08-03 22:28:26 +0400310 return;
311
Jack Miller83293c02014-08-08 14:23:21 -0700312 /*
313 * If kernel.shm_rmid_forced is not set then only keep track of
314 * which shmids are orphaned, so that a later set of the sysctl
315 * can clean them up.
316 */
317 if (!ns->shm_rmid_forced) {
318 down_read(&shm_ids(ns).rwsem);
319 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
320 shp->shm_creator = NULL;
321 /*
322 * Only under read lock but we are only called on current
323 * so no entry on the list will be shared.
324 */
325 list_del(&task->sysvshm.shm_clist);
326 up_read(&shm_ids(ns).rwsem);
327 return;
328 }
329
330 /*
331 * Destroy all already created segments, that were not yet mapped,
332 * and mark any mapped as orphan to cover the sysctl toggling.
333 * Destroy is skipped if shm_may_destroy() returns false.
334 */
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700335 down_write(&shm_ids(ns).rwsem);
Jack Miller83293c02014-08-08 14:23:21 -0700336 list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
337 shp->shm_creator = NULL;
338
339 if (shm_may_destroy(ns, shp)) {
340 shm_lock_by_ptr(shp);
341 shm_destroy(ns, shp);
342 }
343 }
344
345 /* Remove the list head from any segments still attached. */
Jack Millerab602f72014-08-08 14:23:19 -0700346 list_del(&task->sysvshm.shm_clist);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700347 up_write(&shm_ids(ns).rwsem);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700348}
349
Nick Piggind0217ac2007-07-19 01:47:03 -0700350static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800351{
352 struct file *file = vma->vm_file;
353 struct shm_file_data *sfd = shm_file_data(file);
354
Nick Piggind0217ac2007-07-19 01:47:03 -0700355 return sfd->vm_ops->fault(vma, vmf);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800356}
357
358#ifdef CONFIG_NUMA
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700359static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800360{
361 struct file *file = vma->vm_file;
362 struct shm_file_data *sfd = shm_file_data(file);
363 int err = 0;
364 if (sfd->vm_ops->set_policy)
365 err = sfd->vm_ops->set_policy(vma, new);
366 return err;
367}
368
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700369static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
370 unsigned long addr)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800371{
372 struct file *file = vma->vm_file;
373 struct shm_file_data *sfd = shm_file_data(file);
374 struct mempolicy *pol = NULL;
375
376 if (sfd->vm_ops->get_policy)
377 pol = sfd->vm_ops->get_policy(vma, addr);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700378 else if (vma->vm_policy)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800379 pol = vma->vm_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700380
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800381 return pol;
382}
383#endif
384
Manfred Spraul239521f2014-01-27 17:07:04 -0800385static int shm_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800387 struct shm_file_data *sfd = shm_file_data(file);
David Howellsb0e15192006-01-06 00:11:42 -0800388 int ret;
389
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800390 ret = sfd->file->f_op->mmap(sfd->file, vma);
391 if (ret != 0)
392 return ret;
393 sfd->vm_ops = vma->vm_ops;
David Howells2e92a3b2007-07-31 00:37:24 -0700394#ifdef CONFIG_MMU
Nick Piggin54cb8822007-07-19 01:46:59 -0700395 BUG_ON(!sfd->vm_ops->fault);
David Howells2e92a3b2007-07-31 00:37:24 -0700396#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800397 vma->vm_ops = &shm_vm_ops;
398 shm_open(vma);
David Howellsb0e15192006-01-06 00:11:42 -0800399
400 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
Kirill Korotaev4e982312006-10-02 02:18:22 -0700403static int shm_release(struct inode *ino, struct file *file)
404{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800405 struct shm_file_data *sfd = shm_file_data(file);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700406
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800407 put_ipc_ns(sfd->ns);
408 shm_file_data(file) = NULL;
409 kfree(sfd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700410 return 0;
411}
412
Josef Bacik02c24a82011-07-16 20:44:56 -0400413static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Adam Litke516dffd2007-03-01 15:46:08 -0800414{
Adam Litke516dffd2007-03-01 15:46:08 -0800415 struct shm_file_data *sfd = shm_file_data(file);
Adam Litke516dffd2007-03-01 15:46:08 -0800416
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200417 if (!sfd->file->f_op->fsync)
418 return -EINVAL;
Josef Bacik02c24a82011-07-16 20:44:56 -0400419 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
Adam Litke516dffd2007-03-01 15:46:08 -0800420}
421
Will Deacon7d8a4562012-06-07 14:21:13 -0700422static long shm_fallocate(struct file *file, int mode, loff_t offset,
423 loff_t len)
424{
425 struct shm_file_data *sfd = shm_file_data(file);
426
427 if (!sfd->file->f_op->fallocate)
428 return -EOPNOTSUPP;
429 return sfd->file->f_op->fallocate(file, mode, offset, len);
430}
431
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800432static unsigned long shm_get_unmapped_area(struct file *file,
433 unsigned long addr, unsigned long len, unsigned long pgoff,
434 unsigned long flags)
435{
436 struct shm_file_data *sfd = shm_file_data(file);
Al Viroc4caa772009-11-30 08:38:43 -0500437 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
438 pgoff, flags);
Adam Litke516dffd2007-03-01 15:46:08 -0800439}
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800440
Arjan van de Ven9a321442007-02-12 00:55:35 -0800441static const struct file_operations shm_file_operations = {
Kirill Korotaev4e982312006-10-02 02:18:22 -0700442 .mmap = shm_mmap,
Adam Litke516dffd2007-03-01 15:46:08 -0800443 .fsync = shm_fsync,
Kirill Korotaev4e982312006-10-02 02:18:22 -0700444 .release = shm_release,
David Howellsed5e5892010-01-15 17:01:32 -0800445#ifndef CONFIG_MMU
446 .get_unmapped_area = shm_get_unmapped_area,
447#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +0200448 .llseek = noop_llseek,
Will Deacon7d8a4562012-06-07 14:21:13 -0700449 .fallocate = shm_fallocate,
Al Viroc4caa772009-11-30 08:38:43 -0500450};
451
452static const struct file_operations shm_file_operations_huge = {
453 .mmap = shm_mmap,
454 .fsync = shm_fsync,
455 .release = shm_release,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800456 .get_unmapped_area = shm_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200457 .llseek = noop_llseek,
Will Deacon7d8a4562012-06-07 14:21:13 -0700458 .fallocate = shm_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459};
460
Al Viroc4caa772009-11-30 08:38:43 -0500461int is_file_shm_hugepages(struct file *file)
462{
463 return file->f_op == &shm_file_operations_huge;
464}
465
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400466static const struct vm_operations_struct shm_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 .open = shm_open, /* callback for a new vm-area open */
468 .close = shm_close, /* callback for when the vm-area is released */
Nick Piggin54cb8822007-07-19 01:46:59 -0700469 .fault = shm_fault,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800470#if defined(CONFIG_NUMA)
471 .set_policy = shm_set_policy,
472 .get_policy = shm_get_policy,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473#endif
474};
475
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700476/**
477 * newseg - Create a new shared memory segment
478 * @ns: namespace
479 * @params: ptr to the structure that contains key, size and shmflg
480 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700481 * Called with shm_ids.rwsem held as a writer.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700482 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700483static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700485 key_t key = params->key;
486 int shmflg = params->flg;
487 size_t size = params->u.size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 int error;
489 struct shmid_kernel *shp;
Robin Holtd69f3ba2013-04-30 19:15:54 -0700490 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Manfred Spraul239521f2014-01-27 17:07:04 -0800491 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 char name[13];
493 int id;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900494 vm_flags_t acctflag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Kirill Korotaev4e982312006-10-02 02:18:22 -0700496 if (size < SHMMIN || size > ns->shm_ctlmax)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 return -EINVAL;
498
Manfred Spraul13763272014-06-06 14:37:41 -0700499 if (numpages << PAGE_SHIFT < size)
500 return -ENOSPC;
501
Manfred Spraul09c6eb12014-06-06 14:37:40 -0700502 if (ns->shm_tot + numpages < ns->shm_tot ||
503 ns->shm_tot + numpages > ns->shm_ctlall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 return -ENOSPC;
505
506 shp = ipc_rcu_alloc(sizeof(*shp));
507 if (!shp)
508 return -ENOMEM;
509
510 shp->shm_perm.key = key;
Andrew Mortonb33291c2006-01-08 01:02:21 -0800511 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 shp->mlock_user = NULL;
513
514 shp->shm_perm.security = NULL;
515 error = security_shm_alloc(shp);
516 if (error) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700517 ipc_rcu_putref(shp, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return error;
519 }
520
Manfred Spraul239521f2014-01-27 17:07:04 -0800521 sprintf(name, "SYSV%08x", key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (shmflg & SHM_HUGETLB) {
Andrew Mortonc103a4d2013-07-08 16:01:08 -0700523 struct hstate *hs;
Li Zefan091d0d52013-05-09 15:08:15 +0800524 size_t hugesize;
525
Andrew Mortonc103a4d2013-07-08 16:01:08 -0700526 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +0800527 if (!hs) {
528 error = -EINVAL;
529 goto no_file;
530 }
531 hugesize = ALIGN(size, huge_page_size(hs));
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -0700532
Mel Gorman5a6fe122009-02-10 14:02:27 +0000533 /* hugetlb_file_setup applies strict accounting */
534 if (shmflg & SHM_NORESERVE)
535 acctflag = VM_NORESERVE;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -0700536 file = hugetlb_file_setup(name, hugesize, acctflag,
Andi Kleen42d73952012-12-11 16:01:34 -0800537 &shp->mlock_user, HUGETLB_SHMFS_INODE,
538 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 } else {
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800540 /*
541 * Do not allow no accounting for OVERCOMMIT_NEVER, even
Manfred Spraul239521f2014-01-27 17:07:04 -0800542 * if it's asked for.
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800543 */
544 if ((shmflg & SHM_NORESERVE) &&
545 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -0800546 acctflag = VM_NORESERVE;
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800547 file = shmem_file_setup(name, size, acctflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
549 error = PTR_ERR(file);
550 if (IS_ERR(file))
551 goto no_file;
552
Pierre Peiffer48dea402008-04-29 01:00:35 -0700553 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700554 if (id < 0) {
555 error = id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 goto no_id;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700559 shp->shm_cprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 shp->shm_lprid = 0;
561 shp->shm_atim = shp->shm_dtim = 0;
562 shp->shm_ctim = get_seconds();
563 shp->shm_segsz = size;
564 shp->shm_nattch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 shp->shm_file = file;
Vasiliy Kulikov5774ed02011-07-29 03:55:31 +0400566 shp->shm_creator = current;
Jack Millerab602f72014-08-08 14:23:19 -0700567 list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700568
Badari Pulavarty30475cc2007-06-16 10:15:59 -0700569 /*
570 * shmid gets reported as "inode#" in /proc/pid/maps.
571 * proc-ps tools use this. Changing this will break them.
572 */
Al Viro496ad9a2013-01-23 17:07:38 -0500573 file_inode(file)->i_ino = shp->shm_perm.id;
Krishnakumar R551110a2005-10-29 18:16:45 -0700574
Kirill Korotaev4e982312006-10-02 02:18:22 -0700575 ns->shm_tot += numpages;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700576 error = shp->shm_perm.id;
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700577
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700578 ipc_unlock_object(&shp->shm_perm);
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700579 rcu_read_unlock();
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700580 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582no_id:
Hugh Dickins2195d282009-09-12 12:21:27 +0100583 if (is_file_hugepages(file) && shp->mlock_user)
Hugh Dickins353d5c32009-08-24 16:30:28 +0100584 user_shm_unlock(size, shp->mlock_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 fput(file);
586no_file:
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700587 ipc_rcu_putref(shp, shm_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return error;
589}
590
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700591/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700592 * Called with shm_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700593 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700594static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700595{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700596 struct shmid_kernel *shp;
597
598 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
599 return security_shm_associate(shp, shmflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700600}
601
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700602/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700603 * Called with shm_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700604 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700605static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
606 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700607{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700608 struct shmid_kernel *shp;
609
610 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
611 if (shp->shm_segsz < params->u.size)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700612 return -EINVAL;
613
614 return 0;
615}
616
Heiko Carstensd5460c92009-01-14 14:14:27 +0100617SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700619 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700620 static const struct ipc_ops shm_ops = {
621 .getnew = newseg,
622 .associate = shm_security,
623 .more_checks = shm_more_checks,
624 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700625 struct ipc_params shm_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Kirill Korotaev4e982312006-10-02 02:18:22 -0700627 ns = current->nsproxy->ipc_ns;
628
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700629 shm_params.key = key;
630 shm_params.flg = shmflg;
631 shm_params.u.size = size;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700632
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700633 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
635
636static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
637{
Manfred Spraul239521f2014-01-27 17:07:04 -0800638 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 case IPC_64:
640 return copy_to_user(buf, in, sizeof(*in));
641 case IPC_OLD:
642 {
643 struct shmid_ds out;
644
Vasiliy Kulikov3af54c92010-10-30 18:22:49 +0400645 memset(&out, 0, sizeof(out));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
647 out.shm_segsz = in->shm_segsz;
648 out.shm_atime = in->shm_atime;
649 out.shm_dtime = in->shm_dtime;
650 out.shm_ctime = in->shm_ctime;
651 out.shm_cpid = in->shm_cpid;
652 out.shm_lpid = in->shm_lpid;
653 out.shm_nattch = in->shm_nattch;
654
655 return copy_to_user(buf, &out, sizeof(out));
656 }
657 default:
658 return -EINVAL;
659 }
660}
661
Pierre Peiffer016d7132008-04-29 01:00:50 -0700662static inline unsigned long
663copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Manfred Spraul239521f2014-01-27 17:07:04 -0800665 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -0700667 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 case IPC_OLD:
671 {
672 struct shmid_ds tbuf_old;
673
674 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
675 return -EFAULT;
676
Pierre Peiffer016d7132008-04-29 01:00:50 -0700677 out->shm_perm.uid = tbuf_old.shm_perm.uid;
678 out->shm_perm.gid = tbuf_old.shm_perm.gid;
679 out->shm_perm.mode = tbuf_old.shm_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 return 0;
682 }
683 default:
684 return -EINVAL;
685 }
686}
687
688static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
689{
Manfred Spraul239521f2014-01-27 17:07:04 -0800690 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 case IPC_64:
692 return copy_to_user(buf, in, sizeof(*in));
693 case IPC_OLD:
694 {
695 struct shminfo out;
696
Manfred Spraul239521f2014-01-27 17:07:04 -0800697 if (in->shmmax > INT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 out.shmmax = INT_MAX;
699 else
700 out.shmmax = (int)in->shmmax;
701
702 out.shmmin = in->shmmin;
703 out.shmmni = in->shmmni;
704 out.shmseg = in->shmseg;
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700705 out.shmall = in->shmall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 return copy_to_user(buf, &out, sizeof(out));
708 }
709 default:
710 return -EINVAL;
711 }
712}
713
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700714/*
Helge Dellerb7952182010-10-27 15:34:16 -0700715 * Calculate and add used RSS and swap pages of a shm.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700716 * Called with shm_ids.rwsem held as a reader
Helge Dellerb7952182010-10-27 15:34:16 -0700717 */
718static void shm_add_rss_swap(struct shmid_kernel *shp,
719 unsigned long *rss_add, unsigned long *swp_add)
720{
721 struct inode *inode;
722
Al Viro496ad9a2013-01-23 17:07:38 -0500723 inode = file_inode(shp->shm_file);
Helge Dellerb7952182010-10-27 15:34:16 -0700724
725 if (is_file_hugepages(shp->shm_file)) {
726 struct address_space *mapping = inode->i_mapping;
727 struct hstate *h = hstate_file(shp->shm_file);
728 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
729 } else {
730#ifdef CONFIG_SHMEM
731 struct shmem_inode_info *info = SHMEM_I(inode);
732 spin_lock(&info->lock);
733 *rss_add += inode->i_mapping->nrpages;
734 *swp_add += info->swapped;
735 spin_unlock(&info->lock);
736#else
737 *rss_add += inode->i_mapping->nrpages;
738#endif
739 }
740}
741
742/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700743 * Called with shm_ids.rwsem held as a reader
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700744 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700745static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
746 unsigned long *swp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700748 int next_id;
749 int total, in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751 *rss = 0;
752 *swp = 0;
753
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700754 in_use = shm_ids(ns).in_use;
755
756 for (total = 0, next_id = 0; total < in_use; next_id++) {
Tony Battersbye562aeb2009-04-02 16:58:26 -0700757 struct kern_ipc_perm *ipc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Tony Battersbye562aeb2009-04-02 16:58:26 -0700760 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
761 if (ipc == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 continue;
Tony Battersbye562aeb2009-04-02 16:58:26 -0700763 shp = container_of(ipc, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Helge Dellerb7952182010-10-27 15:34:16 -0700765 shm_add_rss_swap(shp, rss, swp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700766
767 total++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
769}
770
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700771/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700772 * This function handles some shmctl commands which require the rwsem
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700773 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700774 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700775 */
776static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
777 struct shmid_ds __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700779 struct kern_ipc_perm *ipcp;
Pierre Peiffer016d7132008-04-29 01:00:50 -0700780 struct shmid64_ds shmid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 struct shmid_kernel *shp;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700782 int err;
783
784 if (cmd == IPC_SET) {
Pierre Peiffer016d7132008-04-29 01:00:50 -0700785 if (copy_shmid_from_user(&shmid64, buf, version))
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700786 return -EFAULT;
787 }
788
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700789 down_write(&shm_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700790 rcu_read_lock();
791
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700792 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
793 &shmid64.shm_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700794 if (IS_ERR(ipcp)) {
795 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700796 goto out_unlock1;
797 }
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700798
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700799 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700800
801 err = security_shm_shmctl(shp, cmd);
802 if (err)
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700803 goto out_unlock1;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700804
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700805 switch (cmd) {
806 case IPC_RMID:
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700807 ipc_lock_object(&shp->shm_perm);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700808 /* do_shm_rmid unlocks the ipc object and rcu */
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700809 do_shm_rmid(ns, ipcp);
810 goto out_up;
811 case IPC_SET:
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700812 ipc_lock_object(&shp->shm_perm);
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800813 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
814 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700815 goto out_unlock0;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700816 shp->shm_ctim = get_seconds();
817 break;
818 default:
819 err = -EINVAL;
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700820 goto out_unlock1;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700821 }
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700822
823out_unlock0:
824 ipc_unlock_object(&shp->shm_perm);
825out_unlock1:
826 rcu_read_unlock();
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700827out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700828 up_write(&shm_ids(ns).rwsem);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700829 return err;
830}
831
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700832static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
833 int cmd, int version, void __user *buf)
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700834{
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700835 int err;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700836 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700838 /* preliminary security checks for *_INFO */
839 if (cmd == IPC_INFO || cmd == SHM_INFO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 err = security_shm_shmctl(NULL, cmd);
841 if (err)
842 return err;
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700843 }
844
845 switch (cmd) {
846 case IPC_INFO:
847 {
848 struct shminfo64 shminfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
WANG Conge8148f72009-01-06 14:42:49 -0800850 memset(&shminfo, 0, sizeof(shminfo));
Kirill Korotaev4e982312006-10-02 02:18:22 -0700851 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
852 shminfo.shmmax = ns->shm_ctlmax;
853 shminfo.shmall = ns->shm_ctlall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 shminfo.shmmin = SHMMIN;
Manfred Spraul239521f2014-01-27 17:07:04 -0800856 if (copy_shminfo_to_user(buf, &shminfo, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return -EFAULT;
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700858
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700859 down_read(&shm_ids(ns).rwsem);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700860 err = ipc_get_maxid(&shm_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700861 up_read(&shm_ids(ns).rwsem);
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700862
Manfred Spraul239521f2014-01-27 17:07:04 -0800863 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 err = 0;
865 goto out;
866 }
867 case SHM_INFO:
868 {
869 struct shm_info shm_info;
870
WANG Conge8148f72009-01-06 14:42:49 -0800871 memset(&shm_info, 0, sizeof(shm_info));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700872 down_read(&shm_ids(ns).rwsem);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700873 shm_info.used_ids = shm_ids(ns).in_use;
Manfred Spraul239521f2014-01-27 17:07:04 -0800874 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700875 shm_info.shm_tot = ns->shm_tot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 shm_info.swap_attempts = 0;
877 shm_info.swap_successes = 0;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700878 err = ipc_get_maxid(&shm_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700879 up_read(&shm_ids(ns).rwsem);
WANG Conge8148f72009-01-06 14:42:49 -0800880 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 err = -EFAULT;
882 goto out;
883 }
884
885 err = err < 0 ? 0 : err;
886 goto out;
887 }
888 case SHM_STAT:
889 case IPC_STAT:
890 {
891 struct shmid64_ds tbuf;
892 int result;
Nadia Derbey023a5352007-10-18 23:40:51 -0700893
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700894 rcu_read_lock();
Nadia Derbey023a5352007-10-18 23:40:51 -0700895 if (cmd == SHM_STAT) {
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700896 shp = shm_obtain_object(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700897 if (IS_ERR(shp)) {
898 err = PTR_ERR(shp);
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700899 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700900 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700901 result = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 } else {
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700903 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700904 if (IS_ERR(shp)) {
905 err = PTR_ERR(shp);
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700906 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 result = 0;
909 }
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700910
WANG Conge8148f72009-01-06 14:42:49 -0800911 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700912 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 goto out_unlock;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 err = security_shm_shmctl(shp, cmd);
916 if (err)
917 goto out_unlock;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700918
Nadia Derbey023a5352007-10-18 23:40:51 -0700919 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
921 tbuf.shm_segsz = shp->shm_segsz;
922 tbuf.shm_atime = shp->shm_atim;
923 tbuf.shm_dtime = shp->shm_dtim;
924 tbuf.shm_ctime = shp->shm_ctim;
925 tbuf.shm_cpid = shp->shm_cprid;
926 tbuf.shm_lpid = shp->shm_lprid;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800927 tbuf.shm_nattch = shp->shm_nattch;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700928 rcu_read_unlock();
929
930 if (copy_shmid_to_user(buf, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 err = -EFAULT;
932 else
933 err = result;
934 goto out;
935 }
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700936 default:
937 return -EINVAL;
938 }
939
940out_unlock:
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700941 rcu_read_unlock();
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700942out:
943 return err;
944}
945
946SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
947{
948 struct shmid_kernel *shp;
949 int err, version;
950 struct ipc_namespace *ns;
951
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700952 if (cmd < 0 || shmid < 0)
953 return -EINVAL;
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700954
955 version = ipc_parse_version(&cmd);
956 ns = current->nsproxy->ipc_ns;
957
958 switch (cmd) {
959 case IPC_INFO:
960 case SHM_INFO:
961 case SHM_STAT:
962 case IPC_STAT:
963 return shmctl_nolock(ns, shmid, cmd, version, buf);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700964 case IPC_RMID:
965 case IPC_SET:
966 return shmctl_down(ns, shmid, cmd, buf, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 case SHM_LOCK:
968 case SHM_UNLOCK:
969 {
Hugh Dickins850465792012-01-20 14:34:19 -0800970 struct file *shm_file;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -0700971
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700972 rcu_read_lock();
973 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700974 if (IS_ERR(shp)) {
975 err = PTR_ERR(shp);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700976 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
Al Viroa33e6752008-12-10 03:40:06 -0500979 audit_ipc_obj(&(shp->shm_perm));
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700980 err = security_shm_shmctl(shp, cmd);
981 if (err)
982 goto out_unlock1;
Steve Grubb073115d2006-04-02 17:07:33 -0400983
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700984 ipc_lock_object(&shp->shm_perm);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -0800985
986 /* check if shm_destroy() is tearing down shp */
987 if (!ipc_valid_object(&shp->shm_perm)) {
988 err = -EIDRM;
989 goto out_unlock0;
990 }
991
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700992 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800993 kuid_t euid = current_euid();
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800994 if (!uid_eq(euid, shp->shm_perm.uid) &&
Jesper Nilsson3a726602013-11-21 14:32:08 -0800995 !uid_eq(euid, shp->shm_perm.cuid)) {
996 err = -EPERM;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700997 goto out_unlock0;
Jesper Nilsson3a726602013-11-21 14:32:08 -0800998 }
999 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1000 err = -EPERM;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001001 goto out_unlock0;
Jesper Nilsson3a726602013-11-21 14:32:08 -08001002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004
Hugh Dickins850465792012-01-20 14:34:19 -08001005 shm_file = shp->shm_file;
1006 if (is_file_hugepages(shm_file))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001007 goto out_unlock0;
Hugh Dickins850465792012-01-20 14:34:19 -08001008
1009 if (cmd == SHM_LOCK) {
David Howells86a264a2008-11-14 10:39:18 +11001010 struct user_struct *user = current_user();
Hugh Dickins850465792012-01-20 14:34:19 -08001011 err = shmem_lock(shm_file, 1, user);
1012 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1013 shp->shm_perm.mode |= SHM_LOCKED;
1014 shp->mlock_user = user;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 }
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001016 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 }
Hugh Dickins850465792012-01-20 14:34:19 -08001018
1019 /* SHM_UNLOCK */
1020 if (!(shp->shm_perm.mode & SHM_LOCKED))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001021 goto out_unlock0;
Hugh Dickins850465792012-01-20 14:34:19 -08001022 shmem_lock(shm_file, 0, shp->mlock_user);
1023 shp->shm_perm.mode &= ~SHM_LOCKED;
1024 shp->mlock_user = NULL;
1025 get_file(shm_file);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001026 ipc_unlock_object(&shp->shm_perm);
1027 rcu_read_unlock();
Hugh Dickins24513262012-01-20 14:34:21 -08001028 shmem_unlock_mapping(shm_file->f_mapping);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001029
Hugh Dickins850465792012-01-20 14:34:19 -08001030 fput(shm_file);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -07001031 return err;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 default:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -07001034 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
1036
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001037out_unlock0:
1038 ipc_unlock_object(&shp->shm_perm);
1039out_unlock1:
1040 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 return err;
1042}
1043
1044/*
1045 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1046 *
1047 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1048 * "raddr" thing points to kernel space, and there has to be a wrapper around
1049 * this.
1050 */
Will Deacon079a96a2012-07-30 14:42:38 -07001051long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1052 unsigned long shmlba)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
1054 struct shmid_kernel *shp;
1055 unsigned long addr;
1056 unsigned long size;
Manfred Spraul239521f2014-01-27 17:07:04 -08001057 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 int err;
1059 unsigned long flags;
1060 unsigned long prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 int acc_mode;
Kirill Korotaev4e982312006-10-02 02:18:22 -07001062 struct ipc_namespace *ns;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001063 struct shm_file_data *sfd;
1064 struct path path;
Al Viroaeb5d722008-09-02 15:28:45 -04001065 fmode_t f_mode;
Michel Lespinasse41badc12013-02-22 16:32:47 -08001066 unsigned long populate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001068 err = -EINVAL;
1069 if (shmid < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 goto out;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001071 else if ((addr = (ulong)shmaddr)) {
Will Deacon079a96a2012-07-30 14:42:38 -07001072 if (addr & (shmlba - 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (shmflg & SHM_RND)
Will Deacon079a96a2012-07-30 14:42:38 -07001074 addr &= ~(shmlba - 1); /* round down */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 else
1076#ifndef __ARCH_FORCE_SHMLBA
1077 if (addr & ~PAGE_MASK)
1078#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001079 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
1081 flags = MAP_SHARED | MAP_FIXED;
1082 } else {
1083 if ((shmflg & SHM_REMAP))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001084 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086 flags = MAP_SHARED;
1087 }
1088
1089 if (shmflg & SHM_RDONLY) {
1090 prot = PROT_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 acc_mode = S_IRUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001092 f_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 } else {
1094 prot = PROT_READ | PROT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 acc_mode = S_IRUGO | S_IWUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001096 f_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 }
1098 if (shmflg & SHM_EXEC) {
1099 prot |= PROT_EXEC;
1100 acc_mode |= S_IXUGO;
1101 }
1102
1103 /*
1104 * We cannot rely on the fs check since SYSV IPC does have an
1105 * additional creator id...
1106 */
Kirill Korotaev4e982312006-10-02 02:18:22 -07001107 ns = current->nsproxy->ipc_ns;
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001108 rcu_read_lock();
1109 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001110 if (IS_ERR(shp)) {
1111 err = PTR_ERR(shp);
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001112 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -07001113 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001114
1115 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001116 if (ipcperms(ns, &shp->shm_perm, acc_mode))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001117 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 err = security_shm_shmat(shp, shmaddr, shmflg);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001120 if (err)
1121 goto out_unlock;
1122
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001123 ipc_lock_object(&shp->shm_perm);
Greg Thelena399b292013-11-21 14:32:00 -08001124
1125 /* check if shm_destroy() is tearing down shp */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001126 if (!ipc_valid_object(&shp->shm_perm)) {
Greg Thelena399b292013-11-21 14:32:00 -08001127 ipc_unlock_object(&shp->shm_perm);
1128 err = -EIDRM;
1129 goto out_unlock;
1130 }
1131
Al Viro2c48b9c2009-08-09 00:52:35 +04001132 path = shp->shm_file->f_path;
1133 path_get(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 shp->shm_nattch++;
David Howells75c3cfa2015-03-17 22:26:12 +00001135 size = i_size_read(d_inode(path.dentry));
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001136 ipc_unlock_object(&shp->shm_perm);
1137 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001139 err = -ENOMEM;
1140 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001141 if (!sfd) {
1142 path_put(&path);
1143 goto out_nattch;
1144 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001145
Al Viro2c48b9c2009-08-09 00:52:35 +04001146 file = alloc_file(&path, f_mode,
1147 is_file_hugepages(shp->shm_file) ?
Al Viroc4caa772009-11-30 08:38:43 -05001148 &shm_file_operations_huge :
1149 &shm_file_operations);
Anatol Pomozov39b65252012-09-12 20:11:55 -07001150 err = PTR_ERR(file);
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001151 if (IS_ERR(file)) {
1152 kfree(sfd);
1153 path_put(&path);
1154 goto out_nattch;
1155 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001156
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001157 file->private_data = sfd;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001158 file->f_mapping = shp->shm_file->f_mapping;
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001159 sfd->id = shp->shm_perm.id;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001160 sfd->ns = get_ipc_ns(ns);
1161 sfd->file = shp->shm_file;
1162 sfd->vm_ops = NULL;
1163
Al Viro8b3ec682012-05-30 17:11:23 -04001164 err = security_mmap_file(file, prot, flags);
1165 if (err)
1166 goto out_fput;
1167
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 down_write(&current->mm->mmap_sem);
1169 if (addr && !(shmflg & SHM_REMAP)) {
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001170 err = -EINVAL;
Manfred Spraul247a8ce2014-06-06 14:37:38 -07001171 if (addr + size < addr)
1172 goto invalid;
1173
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 if (find_vma_intersection(current->mm, addr, addr + size))
1175 goto invalid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001177
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001178 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1179 *raddr = addr;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001180 err = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001181 if (IS_ERR_VALUE(addr))
1182 err = (long)addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183invalid:
1184 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001185 if (populate)
Michel Lespinasse41badc12013-02-22 16:32:47 -08001186 mm_populate(addr, populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Al Viro8b3ec682012-05-30 17:11:23 -04001188out_fput:
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001189 fput(file);
1190
1191out_nattch:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001192 down_write(&shm_ids(ns).rwsem);
Nadia Derbey00c2bf82008-07-25 01:48:03 -07001193 shp = shm_lock(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001194 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 shp->shm_nattch--;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -07001196 if (shm_may_destroy(ns, shp))
Kirill Korotaev4e982312006-10-02 02:18:22 -07001197 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 else
1199 shm_unlock(shp);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001200 up_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 return err;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001202
1203out_unlock:
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001204 rcu_read_unlock();
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001205out:
1206 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
Heiko Carstensd5460c92009-01-14 14:14:27 +01001209SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -07001210{
1211 unsigned long ret;
1212 long err;
1213
Will Deacon079a96a2012-07-30 14:42:38 -07001214 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -07001215 if (err)
1216 return err;
1217 force_successful_syscall_return();
1218 return (long)ret;
1219}
1220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221/*
1222 * detach and kill segment if marked destroyed.
1223 * The work is done in shm_close.
1224 */
Heiko Carstensd5460c92009-01-14 14:14:27 +01001225SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226{
1227 struct mm_struct *mm = current->mm;
Mike Frysinger586c7e62009-06-09 16:26:23 -07001228 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 unsigned long addr = (unsigned long)shmaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 int retval = -EINVAL;
Mike Frysinger586c7e62009-06-09 16:26:23 -07001231#ifdef CONFIG_MMU
1232 loff_t size = 0;
Dave Hansend3c97902014-12-12 16:58:19 -08001233 struct file *file;
Mike Frysinger586c7e62009-06-09 16:26:23 -07001234 struct vm_area_struct *next;
1235#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Hugh Dickinsdf1e2fb2006-03-24 03:18:06 -08001237 if (addr & ~PAGE_MASK)
1238 return retval;
1239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 down_write(&mm->mmap_sem);
1241
1242 /*
1243 * This function tries to be smart and unmap shm segments that
1244 * were modified by partial mlock or munmap calls:
1245 * - It first determines the size of the shm segment that should be
1246 * unmapped: It searches for a vma that is backed by shm and that
1247 * started at address shmaddr. It records it's size and then unmaps
1248 * it.
1249 * - Then it unmaps all shm vmas that started at shmaddr and that
Dave Hansend3c97902014-12-12 16:58:19 -08001250 * are within the initially determined size and that are from the
1251 * same shm segment from which we determined the size.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 * Errors from do_munmap are ignored: the function only fails if
1253 * it's called with invalid parameters or if it's called to unmap
1254 * a part of a vma. Both calls in this function are for full vmas,
1255 * the parameters are directly copied from the vma itself and always
1256 * valid - therefore do_munmap cannot fail. (famous last words?)
1257 */
1258 /*
1259 * If it had been mremap()'d, the starting address would not
1260 * match the usual checks anyway. So assume all vma's are
1261 * above the starting address given.
1262 */
1263 vma = find_vma(mm, addr);
1264
David Howells8feae132009-01-08 12:04:47 +00001265#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 while (vma) {
1267 next = vma->vm_next;
1268
1269 /*
1270 * Check if the starting address would match, i.e. it's
1271 * a fragment created by mprotect() and/or munmap(), or it
1272 * otherwise it starts at this address with no hassles.
1273 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001274 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1276
Dave Hansend3c97902014-12-12 16:58:19 -08001277 /*
1278 * Record the file of the shm segment being
1279 * unmapped. With mremap(), someone could place
1280 * page from another segment but with equal offsets
1281 * in the range we are unmapping.
1282 */
1283 file = vma->vm_file;
Dave Hansen07a46ed2014-12-12 16:58:22 -08001284 size = i_size_read(file_inode(vma->vm_file));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1286 /*
1287 * We discovered the size of the shm segment, so
1288 * break out of here and fall through to the next
1289 * loop that uses the size information to stop
1290 * searching for matching vma's.
1291 */
1292 retval = 0;
1293 vma = next;
1294 break;
1295 }
1296 vma = next;
1297 }
1298
1299 /*
1300 * We need look no further than the maximum address a fragment
1301 * could possibly have landed at. Also cast things to loff_t to
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001302 * prevent overflows and make comparisons vs. equal-width types.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 */
KAMEZAWA Hiroyuki8e367092006-02-10 01:51:12 -08001304 size = PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1306 next = vma->vm_next;
1307
1308 /* finding a matching vma now does not alter retval */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001309 if ((vma->vm_ops == &shm_vm_ops) &&
Dave Hansend3c97902014-12-12 16:58:19 -08001310 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1311 (vma->vm_file == file))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1313 vma = next;
1314 }
1315
David Howells8feae132009-01-08 12:04:47 +00001316#else /* CONFIG_MMU */
1317 /* under NOMMU conditions, the exact address to be destroyed must be
1318 * given */
Davidlohr Bueso530fcd162013-09-11 14:26:28 -07001319 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
David Howells8feae132009-01-08 12:04:47 +00001320 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1321 retval = 0;
1322 }
1323
1324#endif
1325
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 up_write(&mm->mmap_sem);
1327 return retval;
1328}
1329
1330#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07001331static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001333 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07001334 struct shmid_kernel *shp = it;
Helge Dellerb7952182010-10-27 15:34:16 -07001335 unsigned long rss = 0, swp = 0;
1336
1337 shm_add_rss_swap(shp, &rss, &swp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Paul Menage6c826812008-06-12 15:21:49 -07001339#if BITS_PER_LONG <= 32
1340#define SIZE_SPEC "%10lu"
1341#else
1342#define SIZE_SPEC "%21lu"
1343#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Joe Perches7f032d62015-04-15 16:17:54 -07001345 seq_printf(s,
1346 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1347 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1348 SIZE_SPEC " " SIZE_SPEC "\n",
1349 shp->shm_perm.key,
1350 shp->shm_perm.id,
1351 shp->shm_perm.mode,
1352 shp->shm_segsz,
1353 shp->shm_cprid,
1354 shp->shm_lprid,
1355 shp->shm_nattch,
1356 from_kuid_munged(user_ns, shp->shm_perm.uid),
1357 from_kgid_munged(user_ns, shp->shm_perm.gid),
1358 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1359 from_kgid_munged(user_ns, shp->shm_perm.cgid),
1360 shp->shm_atim,
1361 shp->shm_dtim,
1362 shp->shm_ctim,
1363 rss * PAGE_SIZE,
1364 swp * PAGE_SIZE);
1365
1366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367}
1368#endif