blob: cb2cedaa880826222a84541ccea8dfbc611cbcfe [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <asm/uaccess.h>
47
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 Dobriyanf0f37e2f2009-09-27 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);
Kirill Korotaev4e982312006-10-02 02:18:22 -070070static 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 Conge8148f752009-01-06 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
Kirill Korotaev4e982312006-10-02 02:18:22 -070094 if (shp->shm_nattch){
95 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
119void __init shm_init (void)
120{
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
Nadia Derbey023a5352007-10-18 23:40:51 -0700170static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
171 int id)
172{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700173 struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
174
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800175 if (IS_ERR(ipcp))
176 return (struct shmid_kernel *)ipcp;
177
Nadia Derbey03f02c72007-10-18 23:40:51 -0700178 return container_of(ipcp, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700181static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700183 ipc_rmid(&shm_ids(ns), &s->shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184}
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800187/* This is called by fork, once for every shm attach. */
188static void shm_open(struct vm_area_struct *vma)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700189{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800190 struct file *file = vma->vm_file;
191 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 struct shmid_kernel *shp;
193
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800194 shp = shm_lock(sfd->ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700195 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 shp->shm_atim = get_seconds();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700197 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 shp->shm_nattch++;
199 shm_unlock(shp);
200}
201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202/*
203 * shm_destroy - free the struct shmid_kernel
204 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700205 * @ns: namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 * @shp: struct to free
207 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700208 * It has to be called with shp and shm_ids.rwsem (writer) locked,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * but returns with shp unlocked and freed.
210 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700211static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700213 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700214 shm_rmid(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 shm_unlock(shp);
216 if (!is_file_hugepages(shp->shm_file))
217 shmem_lock(shp->shm_file, 0, shp->mlock_user);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100218 else if (shp->mlock_user)
Al Viro496ad9a2013-01-23 17:07:38 -0500219 user_shm_unlock(file_inode(shp->shm_file)->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 shp->mlock_user);
221 fput (shp->shm_file);
222 security_shm_free(shp);
223 ipc_rcu_putref(shp);
224}
225
226/*
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700227 * shm_may_destroy - identifies whether shm segment should be destroyed now
228 *
229 * Returns true if and only if there are no active users of the segment and
230 * one of the following is true:
231 *
232 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
233 *
234 * 2) sysctl kernel.shm_rmid_forced is set to 1.
235 */
236static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
237{
238 return (shp->shm_nattch == 0) &&
239 (ns->shm_rmid_forced ||
240 (shp->shm_perm.mode & SHM_DEST));
241}
242
243/*
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800244 * remove the attach descriptor vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 * free memory for segment if it is marked destroyed.
246 * The descriptor has already been removed from the current->mm->mmap list
247 * and will later be kfree()d.
248 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800249static void shm_close(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800251 struct file * file = vma->vm_file;
252 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 struct shmid_kernel *shp;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800254 struct ipc_namespace *ns = sfd->ns;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700255
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700256 down_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 /* remove from the list of attaches of the shm segment */
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700258 shp = shm_lock(ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700259 BUG_ON(IS_ERR(shp));
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700260 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 shp->shm_dtim = get_seconds();
262 shp->shm_nattch--;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700263 if (shm_may_destroy(ns, shp))
Kirill Korotaev4e982312006-10-02 02:18:22 -0700264 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 else
266 shm_unlock(shp);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700267 up_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268}
269
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700270/* Called with ns->shm_ids(ns).rwsem locked */
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700271static int shm_try_destroy_current(int id, void *p, void *data)
272{
273 struct ipc_namespace *ns = data;
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400274 struct kern_ipc_perm *ipcp = p;
275 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700276
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400277 if (shp->shm_creator != current)
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700278 return 0;
279
Vasiliy Kulikov5774ed02011-07-29 03:55:31 +0400280 /*
281 * Mark it as orphaned to destroy the segment when
282 * kernel.shm_rmid_forced is changed.
283 * It is noop if the following shm_may_destroy() returns true.
284 */
285 shp->shm_creator = NULL;
286
287 /*
288 * Don't even try to destroy it. If shm_rmid_forced=0 and IPC_RMID
289 * is not set, it shouldn't be deleted here.
290 */
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400291 if (!ns->shm_rmid_forced)
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700292 return 0;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700293
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400294 if (shm_may_destroy(ns, shp)) {
295 shm_lock_by_ptr(shp);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700296 shm_destroy(ns, shp);
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400297 }
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700298 return 0;
299}
300
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700301/* Called with ns->shm_ids(ns).rwsem locked */
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700302static int shm_try_destroy_orphaned(int id, void *p, void *data)
303{
304 struct ipc_namespace *ns = data;
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400305 struct kern_ipc_perm *ipcp = p;
306 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700307
308 /*
309 * We want to destroy segments without users and with already
310 * exit'ed originating process.
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400311 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700312 * As shp->* are changed under rwsem, it's safe to skip shp locking.
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700313 */
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400314 if (shp->shm_creator != NULL)
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700315 return 0;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700316
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400317 if (shm_may_destroy(ns, shp)) {
318 shm_lock_by_ptr(shp);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700319 shm_destroy(ns, shp);
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400320 }
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700321 return 0;
322}
323
324void shm_destroy_orphaned(struct ipc_namespace *ns)
325{
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700326 down_write(&shm_ids(ns).rwsem);
Vasiliy Kulikov33a30ed2011-08-03 22:26:55 +0400327 if (shm_ids(ns).in_use)
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400328 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700329 up_write(&shm_ids(ns).rwsem);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700330}
331
332
333void exit_shm(struct task_struct *task)
334{
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400335 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700336
Vasiliy Kulikov298507d2011-08-03 22:28:26 +0400337 if (shm_ids(ns).in_use == 0)
338 return;
339
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700340 /* Destroy all already created segments, but not mapped yet */
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700341 down_write(&shm_ids(ns).rwsem);
Vasiliy Kulikov33a30ed2011-08-03 22:26:55 +0400342 if (shm_ids(ns).in_use)
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400343 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700344 up_write(&shm_ids(ns).rwsem);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700345}
346
Nick Piggind0217ac2007-07-19 01:47:03 -0700347static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800348{
349 struct file *file = vma->vm_file;
350 struct shm_file_data *sfd = shm_file_data(file);
351
Nick Piggind0217ac2007-07-19 01:47:03 -0700352 return sfd->vm_ops->fault(vma, vmf);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800353}
354
355#ifdef CONFIG_NUMA
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700356static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800357{
358 struct file *file = vma->vm_file;
359 struct shm_file_data *sfd = shm_file_data(file);
360 int err = 0;
361 if (sfd->vm_ops->set_policy)
362 err = sfd->vm_ops->set_policy(vma, new);
363 return err;
364}
365
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700366static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
367 unsigned long addr)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800368{
369 struct file *file = vma->vm_file;
370 struct shm_file_data *sfd = shm_file_data(file);
371 struct mempolicy *pol = NULL;
372
373 if (sfd->vm_ops->get_policy)
374 pol = sfd->vm_ops->get_policy(vma, addr);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700375 else if (vma->vm_policy)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800376 pol = vma->vm_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700377
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800378 return pol;
379}
380#endif
381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382static int shm_mmap(struct file * file, struct vm_area_struct * vma)
383{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800384 struct shm_file_data *sfd = shm_file_data(file);
David Howellsb0e15192006-01-06 00:11:42 -0800385 int ret;
386
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800387 ret = sfd->file->f_op->mmap(sfd->file, vma);
388 if (ret != 0)
389 return ret;
390 sfd->vm_ops = vma->vm_ops;
David Howells2e92a3b2007-07-31 00:37:24 -0700391#ifdef CONFIG_MMU
Nick Piggin54cb8822007-07-19 01:46:59 -0700392 BUG_ON(!sfd->vm_ops->fault);
David Howells2e92a3b2007-07-31 00:37:24 -0700393#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800394 vma->vm_ops = &shm_vm_ops;
395 shm_open(vma);
David Howellsb0e15192006-01-06 00:11:42 -0800396
397 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
Kirill Korotaev4e982312006-10-02 02:18:22 -0700400static int shm_release(struct inode *ino, struct file *file)
401{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800402 struct shm_file_data *sfd = shm_file_data(file);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700403
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800404 put_ipc_ns(sfd->ns);
405 shm_file_data(file) = NULL;
406 kfree(sfd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700407 return 0;
408}
409
Josef Bacik02c24a82011-07-16 20:44:56 -0400410static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Adam Litke516dffd2007-03-01 15:46:08 -0800411{
Adam Litke516dffd2007-03-01 15:46:08 -0800412 struct shm_file_data *sfd = shm_file_data(file);
Adam Litke516dffd2007-03-01 15:46:08 -0800413
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200414 if (!sfd->file->f_op->fsync)
415 return -EINVAL;
Josef Bacik02c24a82011-07-16 20:44:56 -0400416 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
Adam Litke516dffd2007-03-01 15:46:08 -0800417}
418
Will Deacon7d8a4562012-06-07 14:21:13 -0700419static long shm_fallocate(struct file *file, int mode, loff_t offset,
420 loff_t len)
421{
422 struct shm_file_data *sfd = shm_file_data(file);
423
424 if (!sfd->file->f_op->fallocate)
425 return -EOPNOTSUPP;
426 return sfd->file->f_op->fallocate(file, mode, offset, len);
427}
428
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800429static unsigned long shm_get_unmapped_area(struct file *file,
430 unsigned long addr, unsigned long len, unsigned long pgoff,
431 unsigned long flags)
432{
433 struct shm_file_data *sfd = shm_file_data(file);
Al Viroc4caa772009-11-30 08:38:43 -0500434 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
435 pgoff, flags);
Adam Litke516dffd2007-03-01 15:46:08 -0800436}
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800437
Arjan van de Ven9a321442007-02-12 00:55:35 -0800438static const struct file_operations shm_file_operations = {
Kirill Korotaev4e982312006-10-02 02:18:22 -0700439 .mmap = shm_mmap,
Adam Litke516dffd2007-03-01 15:46:08 -0800440 .fsync = shm_fsync,
Kirill Korotaev4e982312006-10-02 02:18:22 -0700441 .release = shm_release,
David Howellsed5e5892010-01-15 17:01:32 -0800442#ifndef CONFIG_MMU
443 .get_unmapped_area = shm_get_unmapped_area,
444#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +0200445 .llseek = noop_llseek,
Will Deacon7d8a4562012-06-07 14:21:13 -0700446 .fallocate = shm_fallocate,
Al Viroc4caa772009-11-30 08:38:43 -0500447};
448
449static const struct file_operations shm_file_operations_huge = {
450 .mmap = shm_mmap,
451 .fsync = shm_fsync,
452 .release = shm_release,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800453 .get_unmapped_area = shm_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200454 .llseek = noop_llseek,
Will Deacon7d8a4562012-06-07 14:21:13 -0700455 .fallocate = shm_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456};
457
Al Viroc4caa772009-11-30 08:38:43 -0500458int is_file_shm_hugepages(struct file *file)
459{
460 return file->f_op == &shm_file_operations_huge;
461}
462
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +0400463static const struct vm_operations_struct shm_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 .open = shm_open, /* callback for a new vm-area open */
465 .close = shm_close, /* callback for when the vm-area is released */
Nick Piggin54cb8822007-07-19 01:46:59 -0700466 .fault = shm_fault,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800467#if defined(CONFIG_NUMA)
468 .set_policy = shm_set_policy,
469 .get_policy = shm_get_policy,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470#endif
471};
472
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700473/**
474 * newseg - Create a new shared memory segment
475 * @ns: namespace
476 * @params: ptr to the structure that contains key, size and shmflg
477 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700478 * Called with shm_ids.rwsem held as a writer.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700479 */
480
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700481static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700483 key_t key = params->key;
484 int shmflg = params->flg;
485 size_t size = params->u.size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 int error;
487 struct shmid_kernel *shp;
Robin Holtd69f3ba2013-04-30 19:15:54 -0700488 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 struct file * file;
490 char name[13];
491 int id;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900492 vm_flags_t acctflag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Kirill Korotaev4e982312006-10-02 02:18:22 -0700494 if (size < SHMMIN || size > ns->shm_ctlmax)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 return -EINVAL;
496
Guy Streeterf66d45e2007-01-23 12:20:04 -0600497 if (ns->shm_tot + numpages > ns->shm_ctlall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return -ENOSPC;
499
500 shp = ipc_rcu_alloc(sizeof(*shp));
501 if (!shp)
502 return -ENOMEM;
503
504 shp->shm_perm.key = key;
Andrew Mortonb33291c2006-01-08 01:02:21 -0800505 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 shp->mlock_user = NULL;
507
508 shp->shm_perm.security = NULL;
509 error = security_shm_alloc(shp);
510 if (error) {
511 ipc_rcu_putref(shp);
512 return error;
513 }
514
Eric W. Biederman9d665862007-06-16 10:16:16 -0700515 sprintf (name, "SYSV%08x", key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 if (shmflg & SHM_HUGETLB) {
Andrew Mortonc103a4d2013-07-08 16:01:08 -0700517 struct hstate *hs;
Li Zefan091d0d52013-05-09 15:08:15 +0800518 size_t hugesize;
519
Andrew Mortonc103a4d2013-07-08 16:01:08 -0700520 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +0800521 if (!hs) {
522 error = -EINVAL;
523 goto no_file;
524 }
525 hugesize = ALIGN(size, huge_page_size(hs));
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -0700526
Mel Gorman5a6fe122009-02-10 14:02:27 +0000527 /* hugetlb_file_setup applies strict accounting */
528 if (shmflg & SHM_NORESERVE)
529 acctflag = VM_NORESERVE;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -0700530 file = hugetlb_file_setup(name, hugesize, acctflag,
Andi Kleen42d73952012-12-11 16:01:34 -0800531 &shp->mlock_user, HUGETLB_SHMFS_INODE,
532 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 } else {
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800534 /*
535 * Do not allow no accounting for OVERCOMMIT_NEVER, even
536 * if it's asked for.
537 */
538 if ((shmflg & SHM_NORESERVE) &&
539 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -0800540 acctflag = VM_NORESERVE;
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800541 file = shmem_file_setup(name, size, acctflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543 error = PTR_ERR(file);
544 if (IS_ERR(file))
545 goto no_file;
546
Pierre Peiffer48dea402008-04-29 01:00:35 -0700547 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700548 if (id < 0) {
549 error = id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 goto no_id;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700553 shp->shm_cprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 shp->shm_lprid = 0;
555 shp->shm_atim = shp->shm_dtim = 0;
556 shp->shm_ctim = get_seconds();
557 shp->shm_segsz = size;
558 shp->shm_nattch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 shp->shm_file = file;
Vasiliy Kulikov5774ed02011-07-29 03:55:31 +0400560 shp->shm_creator = current;
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700561
Badari Pulavarty30475cc2007-06-16 10:15:59 -0700562 /*
563 * shmid gets reported as "inode#" in /proc/pid/maps.
564 * proc-ps tools use this. Changing this will break them.
565 */
Al Viro496ad9a2013-01-23 17:07:38 -0500566 file_inode(file)->i_ino = shp->shm_perm.id;
Krishnakumar R551110a2005-10-29 18:16:45 -0700567
Kirill Korotaev4e982312006-10-02 02:18:22 -0700568 ns->shm_tot += numpages;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700569 error = shp->shm_perm.id;
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700570
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700571 ipc_unlock_object(&shp->shm_perm);
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700572 rcu_read_unlock();
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700573 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575no_id:
Hugh Dickins2195d2812009-09-12 12:21:27 +0100576 if (is_file_hugepages(file) && shp->mlock_user)
Hugh Dickins353d5c32009-08-24 16:30:28 +0100577 user_shm_unlock(size, shp->mlock_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 fput(file);
579no_file:
580 security_shm_free(shp);
581 ipc_rcu_putref(shp);
582 return error;
583}
584
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700585/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700586 * Called with shm_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700587 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700588static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700589{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700590 struct shmid_kernel *shp;
591
592 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
593 return security_shm_associate(shp, shmflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700594}
595
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700596/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700597 * Called with shm_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700598 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700599static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
600 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700601{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700602 struct shmid_kernel *shp;
603
604 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
605 if (shp->shm_segsz < params->u.size)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700606 return -EINVAL;
607
608 return 0;
609}
610
Heiko Carstensd5460c92009-01-14 14:14:27 +0100611SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700613 struct ipc_namespace *ns;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700614 struct ipc_ops shm_ops;
615 struct ipc_params shm_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Kirill Korotaev4e982312006-10-02 02:18:22 -0700617 ns = current->nsproxy->ipc_ns;
618
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700619 shm_ops.getnew = newseg;
620 shm_ops.associate = shm_security;
621 shm_ops.more_checks = shm_more_checks;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700622
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700623 shm_params.key = key;
624 shm_params.flg = shmflg;
625 shm_params.u.size = size;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700626
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700627 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
630static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
631{
632 switch(version) {
633 case IPC_64:
634 return copy_to_user(buf, in, sizeof(*in));
635 case IPC_OLD:
636 {
637 struct shmid_ds out;
638
Vasiliy Kulikov3af54c92010-10-30 18:22:49 +0400639 memset(&out, 0, sizeof(out));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
641 out.shm_segsz = in->shm_segsz;
642 out.shm_atime = in->shm_atime;
643 out.shm_dtime = in->shm_dtime;
644 out.shm_ctime = in->shm_ctime;
645 out.shm_cpid = in->shm_cpid;
646 out.shm_lpid = in->shm_lpid;
647 out.shm_nattch = in->shm_nattch;
648
649 return copy_to_user(buf, &out, sizeof(out));
650 }
651 default:
652 return -EINVAL;
653 }
654}
655
Pierre Peiffer016d7132008-04-29 01:00:50 -0700656static inline unsigned long
657copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
659 switch(version) {
660 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -0700661 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 case IPC_OLD:
665 {
666 struct shmid_ds tbuf_old;
667
668 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
669 return -EFAULT;
670
Pierre Peiffer016d7132008-04-29 01:00:50 -0700671 out->shm_perm.uid = tbuf_old.shm_perm.uid;
672 out->shm_perm.gid = tbuf_old.shm_perm.gid;
673 out->shm_perm.mode = tbuf_old.shm_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 return 0;
676 }
677 default:
678 return -EINVAL;
679 }
680}
681
682static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
683{
684 switch(version) {
685 case IPC_64:
686 return copy_to_user(buf, in, sizeof(*in));
687 case IPC_OLD:
688 {
689 struct shminfo out;
690
691 if(in->shmmax > INT_MAX)
692 out.shmmax = INT_MAX;
693 else
694 out.shmmax = (int)in->shmmax;
695
696 out.shmmin = in->shmmin;
697 out.shmmni = in->shmmni;
698 out.shmseg = in->shmseg;
699 out.shmall = in->shmall;
700
701 return copy_to_user(buf, &out, sizeof(out));
702 }
703 default:
704 return -EINVAL;
705 }
706}
707
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700708/*
Helge Dellerb7952182010-10-27 15:34:16 -0700709 * Calculate and add used RSS and swap pages of a shm.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700710 * Called with shm_ids.rwsem held as a reader
Helge Dellerb7952182010-10-27 15:34:16 -0700711 */
712static void shm_add_rss_swap(struct shmid_kernel *shp,
713 unsigned long *rss_add, unsigned long *swp_add)
714{
715 struct inode *inode;
716
Al Viro496ad9a2013-01-23 17:07:38 -0500717 inode = file_inode(shp->shm_file);
Helge Dellerb7952182010-10-27 15:34:16 -0700718
719 if (is_file_hugepages(shp->shm_file)) {
720 struct address_space *mapping = inode->i_mapping;
721 struct hstate *h = hstate_file(shp->shm_file);
722 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
723 } else {
724#ifdef CONFIG_SHMEM
725 struct shmem_inode_info *info = SHMEM_I(inode);
726 spin_lock(&info->lock);
727 *rss_add += inode->i_mapping->nrpages;
728 *swp_add += info->swapped;
729 spin_unlock(&info->lock);
730#else
731 *rss_add += inode->i_mapping->nrpages;
732#endif
733 }
734}
735
736/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700737 * Called with shm_ids.rwsem held as a reader
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700738 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700739static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
740 unsigned long *swp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700742 int next_id;
743 int total, in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
745 *rss = 0;
746 *swp = 0;
747
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700748 in_use = shm_ids(ns).in_use;
749
750 for (total = 0, next_id = 0; total < in_use; next_id++) {
Tony Battersbye562aeb2009-04-02 16:58:26 -0700751 struct kern_ipc_perm *ipc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Tony Battersbye562aeb2009-04-02 16:58:26 -0700754 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
755 if (ipc == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 continue;
Tony Battersbye562aeb2009-04-02 16:58:26 -0700757 shp = container_of(ipc, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Helge Dellerb7952182010-10-27 15:34:16 -0700759 shm_add_rss_swap(shp, rss, swp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700760
761 total++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763}
764
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700765/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700766 * This function handles some shmctl commands which require the rwsem
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700767 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700768 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700769 */
770static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
771 struct shmid_ds __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700773 struct kern_ipc_perm *ipcp;
Pierre Peiffer016d7132008-04-29 01:00:50 -0700774 struct shmid64_ds shmid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 struct shmid_kernel *shp;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700776 int err;
777
778 if (cmd == IPC_SET) {
Pierre Peiffer016d7132008-04-29 01:00:50 -0700779 if (copy_shmid_from_user(&shmid64, buf, version))
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700780 return -EFAULT;
781 }
782
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700783 down_write(&shm_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700784 rcu_read_lock();
785
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700786 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
787 &shmid64.shm_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700788 if (IS_ERR(ipcp)) {
789 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700790 goto out_unlock1;
791 }
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700792
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700793 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700794
795 err = security_shm_shmctl(shp, cmd);
796 if (err)
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700797 goto out_unlock1;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700798
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700799 switch (cmd) {
800 case IPC_RMID:
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700801 ipc_lock_object(&shp->shm_perm);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700802 /* do_shm_rmid unlocks the ipc object and rcu */
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700803 do_shm_rmid(ns, ipcp);
804 goto out_up;
805 case IPC_SET:
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700806 ipc_lock_object(&shp->shm_perm);
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800807 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
808 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700809 goto out_unlock0;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700810 shp->shm_ctim = get_seconds();
811 break;
812 default:
813 err = -EINVAL;
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700814 goto out_unlock1;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700815 }
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700816
817out_unlock0:
818 ipc_unlock_object(&shp->shm_perm);
819out_unlock1:
820 rcu_read_unlock();
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700821out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700822 up_write(&shm_ids(ns).rwsem);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700823 return err;
824}
825
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700826static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
827 int cmd, int version, void __user *buf)
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700828{
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700829 int err;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700830 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700832 /* preliminary security checks for *_INFO */
833 if (cmd == IPC_INFO || cmd == SHM_INFO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 err = security_shm_shmctl(NULL, cmd);
835 if (err)
836 return err;
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700837 }
838
839 switch (cmd) {
840 case IPC_INFO:
841 {
842 struct shminfo64 shminfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
WANG Conge8148f752009-01-06 14:42:49 -0800844 memset(&shminfo, 0, sizeof(shminfo));
Kirill Korotaev4e982312006-10-02 02:18:22 -0700845 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
846 shminfo.shmmax = ns->shm_ctlmax;
847 shminfo.shmall = ns->shm_ctlall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 shminfo.shmmin = SHMMIN;
850 if(copy_shminfo_to_user (buf, &shminfo, version))
851 return -EFAULT;
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700852
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700853 down_read(&shm_ids(ns).rwsem);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700854 err = ipc_get_maxid(&shm_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700855 up_read(&shm_ids(ns).rwsem);
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if(err<0)
858 err = 0;
859 goto out;
860 }
861 case SHM_INFO:
862 {
863 struct shm_info shm_info;
864
WANG Conge8148f752009-01-06 14:42:49 -0800865 memset(&shm_info, 0, sizeof(shm_info));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700866 down_read(&shm_ids(ns).rwsem);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700867 shm_info.used_ids = shm_ids(ns).in_use;
868 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
869 shm_info.shm_tot = ns->shm_tot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 shm_info.swap_attempts = 0;
871 shm_info.swap_successes = 0;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700872 err = ipc_get_maxid(&shm_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700873 up_read(&shm_ids(ns).rwsem);
WANG Conge8148f752009-01-06 14:42:49 -0800874 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 err = -EFAULT;
876 goto out;
877 }
878
879 err = err < 0 ? 0 : err;
880 goto out;
881 }
882 case SHM_STAT:
883 case IPC_STAT:
884 {
885 struct shmid64_ds tbuf;
886 int result;
Nadia Derbey023a5352007-10-18 23:40:51 -0700887
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700888 rcu_read_lock();
Nadia Derbey023a5352007-10-18 23:40:51 -0700889 if (cmd == SHM_STAT) {
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700890 shp = shm_obtain_object(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700891 if (IS_ERR(shp)) {
892 err = PTR_ERR(shp);
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700893 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700894 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700895 result = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 } else {
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700897 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700898 if (IS_ERR(shp)) {
899 err = PTR_ERR(shp);
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700900 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 result = 0;
903 }
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700904
WANG Conge8148f752009-01-06 14:42:49 -0800905 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700906 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 goto out_unlock;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 err = security_shm_shmctl(shp, cmd);
910 if (err)
911 goto out_unlock;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700912
Nadia Derbey023a5352007-10-18 23:40:51 -0700913 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
915 tbuf.shm_segsz = shp->shm_segsz;
916 tbuf.shm_atime = shp->shm_atim;
917 tbuf.shm_dtime = shp->shm_dtim;
918 tbuf.shm_ctime = shp->shm_ctim;
919 tbuf.shm_cpid = shp->shm_cprid;
920 tbuf.shm_lpid = shp->shm_lprid;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800921 tbuf.shm_nattch = shp->shm_nattch;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700922 rcu_read_unlock();
923
924 if (copy_shmid_to_user(buf, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 err = -EFAULT;
926 else
927 err = result;
928 goto out;
929 }
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700930 default:
931 return -EINVAL;
932 }
933
934out_unlock:
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700935 rcu_read_unlock();
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700936out:
937 return err;
938}
939
940SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
941{
942 struct shmid_kernel *shp;
943 int err, version;
944 struct ipc_namespace *ns;
945
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700946 if (cmd < 0 || shmid < 0)
947 return -EINVAL;
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700948
949 version = ipc_parse_version(&cmd);
950 ns = current->nsproxy->ipc_ns;
951
952 switch (cmd) {
953 case IPC_INFO:
954 case SHM_INFO:
955 case SHM_STAT:
956 case IPC_STAT:
957 return shmctl_nolock(ns, shmid, cmd, version, buf);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700958 case IPC_RMID:
959 case IPC_SET:
960 return shmctl_down(ns, shmid, cmd, buf, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 case SHM_LOCK:
962 case SHM_UNLOCK:
963 {
Hugh Dickins850465792012-01-20 14:34:19 -0800964 struct file *shm_file;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -0700965
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700966 rcu_read_lock();
967 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700968 if (IS_ERR(shp)) {
969 err = PTR_ERR(shp);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700970 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Al Viroa33e6752008-12-10 03:40:06 -0500973 audit_ipc_obj(&(shp->shm_perm));
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700974 err = security_shm_shmctl(shp, cmd);
975 if (err)
976 goto out_unlock1;
Steve Grubb073115d2006-04-02 17:07:33 -0400977
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700978 ipc_lock_object(&shp->shm_perm);
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700979 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800980 kuid_t euid = current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 err = -EPERM;
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800982 if (!uid_eq(euid, shp->shm_perm.uid) &&
983 !uid_eq(euid, shp->shm_perm.cuid))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700984 goto out_unlock0;
Jiri Slabyf1eb1332010-03-10 15:23:05 -0800985 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700986 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
988
Hugh Dickins850465792012-01-20 14:34:19 -0800989 shm_file = shp->shm_file;
990 if (is_file_hugepages(shm_file))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700991 goto out_unlock0;
Hugh Dickins850465792012-01-20 14:34:19 -0800992
993 if (cmd == SHM_LOCK) {
David Howells86a264a2008-11-14 10:39:18 +1100994 struct user_struct *user = current_user();
Hugh Dickins850465792012-01-20 14:34:19 -0800995 err = shmem_lock(shm_file, 1, user);
996 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
997 shp->shm_perm.mode |= SHM_LOCKED;
998 shp->mlock_user = user;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 }
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001000 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
Hugh Dickins850465792012-01-20 14:34:19 -08001002
1003 /* SHM_UNLOCK */
1004 if (!(shp->shm_perm.mode & SHM_LOCKED))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001005 goto out_unlock0;
Hugh Dickins850465792012-01-20 14:34:19 -08001006 shmem_lock(shm_file, 0, shp->mlock_user);
1007 shp->shm_perm.mode &= ~SHM_LOCKED;
1008 shp->mlock_user = NULL;
1009 get_file(shm_file);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001010 ipc_unlock_object(&shp->shm_perm);
1011 rcu_read_unlock();
Hugh Dickins24513262012-01-20 14:34:21 -08001012 shmem_unlock_mapping(shm_file->f_mapping);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001013
Hugh Dickins850465792012-01-20 14:34:19 -08001014 fput(shm_file);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -07001015 return err;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 default:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -07001018 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
1020
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001021out_unlock0:
1022 ipc_unlock_object(&shp->shm_perm);
1023out_unlock1:
1024 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 return err;
1026}
1027
1028/*
1029 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1030 *
1031 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1032 * "raddr" thing points to kernel space, and there has to be a wrapper around
1033 * this.
1034 */
Will Deacon079a96a2012-07-30 14:42:38 -07001035long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1036 unsigned long shmlba)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
1038 struct shmid_kernel *shp;
1039 unsigned long addr;
1040 unsigned long size;
1041 struct file * file;
1042 int err;
1043 unsigned long flags;
1044 unsigned long prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 int acc_mode;
Kirill Korotaev4e982312006-10-02 02:18:22 -07001046 struct ipc_namespace *ns;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001047 struct shm_file_data *sfd;
1048 struct path path;
Al Viroaeb5d722008-09-02 15:28:45 -04001049 fmode_t f_mode;
Michel Lespinasse41badc12013-02-22 16:32:47 -08001050 unsigned long populate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001052 err = -EINVAL;
1053 if (shmid < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 goto out;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001055 else if ((addr = (ulong)shmaddr)) {
Will Deacon079a96a2012-07-30 14:42:38 -07001056 if (addr & (shmlba - 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (shmflg & SHM_RND)
Will Deacon079a96a2012-07-30 14:42:38 -07001058 addr &= ~(shmlba - 1); /* round down */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 else
1060#ifndef __ARCH_FORCE_SHMLBA
1061 if (addr & ~PAGE_MASK)
1062#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001063 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
1065 flags = MAP_SHARED | MAP_FIXED;
1066 } else {
1067 if ((shmflg & SHM_REMAP))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001068 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 flags = MAP_SHARED;
1071 }
1072
1073 if (shmflg & SHM_RDONLY) {
1074 prot = PROT_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 acc_mode = S_IRUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001076 f_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 } else {
1078 prot = PROT_READ | PROT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 acc_mode = S_IRUGO | S_IWUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001080 f_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082 if (shmflg & SHM_EXEC) {
1083 prot |= PROT_EXEC;
1084 acc_mode |= S_IXUGO;
1085 }
1086
1087 /*
1088 * We cannot rely on the fs check since SYSV IPC does have an
1089 * additional creator id...
1090 */
Kirill Korotaev4e982312006-10-02 02:18:22 -07001091 ns = current->nsproxy->ipc_ns;
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001092 rcu_read_lock();
1093 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001094 if (IS_ERR(shp)) {
1095 err = PTR_ERR(shp);
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001096 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -07001097 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001098
1099 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001100 if (ipcperms(ns, &shp->shm_perm, acc_mode))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001101 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 err = security_shm_shmat(shp, shmaddr, shmflg);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001104 if (err)
1105 goto out_unlock;
1106
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001107 ipc_lock_object(&shp->shm_perm);
Al Viro2c48b9c2009-08-09 00:52:35 +04001108 path = shp->shm_file->f_path;
1109 path_get(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 shp->shm_nattch++;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001111 size = i_size_read(path.dentry->d_inode);
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001112 ipc_unlock_object(&shp->shm_perm);
1113 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001115 err = -ENOMEM;
1116 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001117 if (!sfd) {
1118 path_put(&path);
1119 goto out_nattch;
1120 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001121
Al Viro2c48b9c2009-08-09 00:52:35 +04001122 file = alloc_file(&path, f_mode,
1123 is_file_hugepages(shp->shm_file) ?
Al Viroc4caa772009-11-30 08:38:43 -05001124 &shm_file_operations_huge :
1125 &shm_file_operations);
Anatol Pomozov39b652522012-09-12 20:11:55 -07001126 err = PTR_ERR(file);
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001127 if (IS_ERR(file)) {
1128 kfree(sfd);
1129 path_put(&path);
1130 goto out_nattch;
1131 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001132
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001133 file->private_data = sfd;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001134 file->f_mapping = shp->shm_file->f_mapping;
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001135 sfd->id = shp->shm_perm.id;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001136 sfd->ns = get_ipc_ns(ns);
1137 sfd->file = shp->shm_file;
1138 sfd->vm_ops = NULL;
1139
Al Viro8b3ec682012-05-30 17:11:23 -04001140 err = security_mmap_file(file, prot, flags);
1141 if (err)
1142 goto out_fput;
1143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 down_write(&current->mm->mmap_sem);
1145 if (addr && !(shmflg & SHM_REMAP)) {
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001146 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 if (find_vma_intersection(current->mm, addr, addr + size))
1148 goto invalid;
1149 /*
1150 * If shm segment goes below stack, make sure there is some
1151 * space left for the stack to grow (at least 4 pages).
1152 */
1153 if (addr < current->mm->start_stack &&
1154 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1155 goto invalid;
1156 }
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001157
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001158 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1159 *raddr = addr;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001160 err = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001161 if (IS_ERR_VALUE(addr))
1162 err = (long)addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163invalid:
1164 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001165 if (populate)
Michel Lespinasse41badc12013-02-22 16:32:47 -08001166 mm_populate(addr, populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
Al Viro8b3ec682012-05-30 17:11:23 -04001168out_fput:
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001169 fput(file);
1170
1171out_nattch:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001172 down_write(&shm_ids(ns).rwsem);
Nadia Derbey00c2bf82008-07-25 01:48:03 -07001173 shp = shm_lock(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001174 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 shp->shm_nattch--;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -07001176 if (shm_may_destroy(ns, shp))
Kirill Korotaev4e982312006-10-02 02:18:22 -07001177 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 else
1179 shm_unlock(shp);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001180 up_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 return err;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001182
1183out_unlock:
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001184 rcu_read_unlock();
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001185out:
1186 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
Heiko Carstensd5460c92009-01-14 14:14:27 +01001189SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -07001190{
1191 unsigned long ret;
1192 long err;
1193
Will Deacon079a96a2012-07-30 14:42:38 -07001194 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -07001195 if (err)
1196 return err;
1197 force_successful_syscall_return();
1198 return (long)ret;
1199}
1200
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201/*
1202 * detach and kill segment if marked destroyed.
1203 * The work is done in shm_close.
1204 */
Heiko Carstensd5460c92009-01-14 14:14:27 +01001205SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
1207 struct mm_struct *mm = current->mm;
Mike Frysinger586c7e62009-06-09 16:26:23 -07001208 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 unsigned long addr = (unsigned long)shmaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 int retval = -EINVAL;
Mike Frysinger586c7e62009-06-09 16:26:23 -07001211#ifdef CONFIG_MMU
1212 loff_t size = 0;
1213 struct vm_area_struct *next;
1214#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Hugh Dickinsdf1e2fb2006-03-24 03:18:06 -08001216 if (addr & ~PAGE_MASK)
1217 return retval;
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 down_write(&mm->mmap_sem);
1220
1221 /*
1222 * This function tries to be smart and unmap shm segments that
1223 * were modified by partial mlock or munmap calls:
1224 * - It first determines the size of the shm segment that should be
1225 * unmapped: It searches for a vma that is backed by shm and that
1226 * started at address shmaddr. It records it's size and then unmaps
1227 * it.
1228 * - Then it unmaps all shm vmas that started at shmaddr and that
1229 * are within the initially determined size.
1230 * Errors from do_munmap are ignored: the function only fails if
1231 * it's called with invalid parameters or if it's called to unmap
1232 * a part of a vma. Both calls in this function are for full vmas,
1233 * the parameters are directly copied from the vma itself and always
1234 * valid - therefore do_munmap cannot fail. (famous last words?)
1235 */
1236 /*
1237 * If it had been mremap()'d, the starting address would not
1238 * match the usual checks anyway. So assume all vma's are
1239 * above the starting address given.
1240 */
1241 vma = find_vma(mm, addr);
1242
David Howells8feae132009-01-08 12:04:47 +00001243#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 while (vma) {
1245 next = vma->vm_next;
1246
1247 /*
1248 * Check if the starting address would match, i.e. it's
1249 * a fragment created by mprotect() and/or munmap(), or it
1250 * otherwise it starts at this address with no hassles.
1251 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001252 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1254
1255
Al Viro496ad9a2013-01-23 17:07:38 -05001256 size = file_inode(vma->vm_file)->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1258 /*
1259 * We discovered the size of the shm segment, so
1260 * break out of here and fall through to the next
1261 * loop that uses the size information to stop
1262 * searching for matching vma's.
1263 */
1264 retval = 0;
1265 vma = next;
1266 break;
1267 }
1268 vma = next;
1269 }
1270
1271 /*
1272 * We need look no further than the maximum address a fragment
1273 * could possibly have landed at. Also cast things to loff_t to
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001274 * prevent overflows and make comparisons vs. equal-width types.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 */
KAMEZAWA Hiroyuki8e367092006-02-10 01:51:12 -08001276 size = PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1278 next = vma->vm_next;
1279
1280 /* finding a matching vma now does not alter retval */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001281 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1283
1284 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1285 vma = next;
1286 }
1287
David Howells8feae132009-01-08 12:04:47 +00001288#else /* CONFIG_MMU */
1289 /* under NOMMU conditions, the exact address to be destroyed must be
1290 * given */
1291 retval = -EINVAL;
1292 if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1293 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1294 retval = 0;
1295 }
1296
1297#endif
1298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 up_write(&mm->mmap_sem);
1300 return retval;
1301}
1302
1303#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07001304static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305{
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001306 struct user_namespace *user_ns = seq_user_ns(s);
Mike Waychison19b49462005-09-06 15:17:10 -07001307 struct shmid_kernel *shp = it;
Helge Dellerb7952182010-10-27 15:34:16 -07001308 unsigned long rss = 0, swp = 0;
1309
1310 shm_add_rss_swap(shp, &rss, &swp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Paul Menage6c826812008-06-12 15:21:49 -07001312#if BITS_PER_LONG <= 32
1313#define SIZE_SPEC "%10lu"
1314#else
1315#define SIZE_SPEC "%21lu"
1316#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Paul Menage6c826812008-06-12 15:21:49 -07001318 return seq_printf(s,
1319 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
Helge Dellerb7952182010-10-27 15:34:16 -07001320 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1321 SIZE_SPEC " " SIZE_SPEC "\n",
Mike Waychison19b49462005-09-06 15:17:10 -07001322 shp->shm_perm.key,
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001323 shp->shm_perm.id,
Andrew Mortonb33291c2006-01-08 01:02:21 -08001324 shp->shm_perm.mode,
Mike Waychison19b49462005-09-06 15:17:10 -07001325 shp->shm_segsz,
1326 shp->shm_cprid,
1327 shp->shm_lprid,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001328 shp->shm_nattch,
Eric W. Biederman1efdb692012-02-07 16:54:11 -08001329 from_kuid_munged(user_ns, shp->shm_perm.uid),
1330 from_kgid_munged(user_ns, shp->shm_perm.gid),
1331 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1332 from_kgid_munged(user_ns, shp->shm_perm.cgid),
Mike Waychison19b49462005-09-06 15:17:10 -07001333 shp->shm_atim,
1334 shp->shm_dtim,
Helge Dellerb7952182010-10-27 15:34:16 -07001335 shp->shm_ctim,
1336 rss * PAGE_SIZE,
1337 swp * PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338}
1339#endif