blob: 4aef24d91b633e12275cea64a380df4543fc796b [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{
Davidlohr Bueso55b7ae52015-06-30 14:58:42 -0700132 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
Davidlohr Bueso8b8d52a2013-09-11 14:26:15 -0700133
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
Davidlohr Buesoc5c89752015-06-30 14:58:36 -0700158 /*
159 * We raced in the idr lookup or with shm_destroy(). Either way, the
160 * ID is busted.
161 */
162 BUG_ON(IS_ERR(ipcp));
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800163
Nadia Derbey03f02c72007-10-18 23:40:51 -0700164 return container_of(ipcp, struct shmid_kernel, shm_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700165}
166
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400167static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
168{
169 rcu_read_lock();
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700170 ipc_lock_object(&ipcp->shm_perm);
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400171}
172
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700173static void shm_rcu_free(struct rcu_head *head)
174{
175 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
176 struct shmid_kernel *shp = ipc_rcu_to_struct(p);
177
178 security_shm_free(shp);
179 ipc_rcu_free(head);
180}
181
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700182static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Jack Millerab602f72014-08-08 14:23:19 -0700184 list_del(&s->shm_clist);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700185 ipc_rmid(&shm_ids(ns), &s->shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800189/* This is called by fork, once for every shm attach. */
190static void shm_open(struct vm_area_struct *vma)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700191{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800192 struct file *file = vma->vm_file;
193 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 struct shmid_kernel *shp;
195
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800196 shp = shm_lock(sfd->ns, sfd->id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 shp->shm_atim = get_seconds();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700198 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 shp->shm_nattch++;
200 shm_unlock(shp);
201}
202
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203/*
204 * shm_destroy - free the struct shmid_kernel
205 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700206 * @ns: namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 * @shp: struct to free
208 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700209 * It has to be called with shp and shm_ids.rwsem (writer) locked,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 * but returns with shp unlocked and freed.
211 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700212static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
Greg Thelena399b292013-11-21 14:32:00 -0800214 struct file *shm_file;
215
216 shm_file = shp->shm_file;
217 shp->shm_file = NULL;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700218 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700219 shm_rmid(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 shm_unlock(shp);
Greg Thelena399b292013-11-21 14:32:00 -0800221 if (!is_file_hugepages(shm_file))
222 shmem_lock(shm_file, 0, shp->mlock_user);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100223 else if (shp->mlock_user)
Dave Hansen07a46ed2014-12-12 16:58:22 -0800224 user_shm_unlock(i_size_read(file_inode(shm_file)),
225 shp->mlock_user);
Greg Thelena399b292013-11-21 14:32:00 -0800226 fput(shm_file);
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700227 ipc_rcu_putref(shp, shm_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228}
229
230/*
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700231 * shm_may_destroy - identifies whether shm segment should be destroyed now
232 *
233 * Returns true if and only if there are no active users of the segment and
234 * one of the following is true:
235 *
236 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
237 *
238 * 2) sysctl kernel.shm_rmid_forced is set to 1.
239 */
240static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
241{
242 return (shp->shm_nattch == 0) &&
243 (ns->shm_rmid_forced ||
244 (shp->shm_perm.mode & SHM_DEST));
245}
246
247/*
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800248 * remove the attach descriptor vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 * free memory for segment if it is marked destroyed.
250 * The descriptor has already been removed from the current->mm->mmap list
251 * and will later be kfree()d.
252 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800253static void shm_close(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
Manfred Spraul239521f2014-01-27 17:07:04 -0800255 struct file *file = vma->vm_file;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800256 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 struct shmid_kernel *shp;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800258 struct ipc_namespace *ns = sfd->ns;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700259
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700260 down_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 /* remove from the list of attaches of the shm segment */
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700262 shp = shm_lock(ns, sfd->id);
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700263 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 shp->shm_dtim = get_seconds();
265 shp->shm_nattch--;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700266 if (shm_may_destroy(ns, shp))
Kirill Korotaev4e982312006-10-02 02:18:22 -0700267 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 else
269 shm_unlock(shp);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700270 up_write(&shm_ids(ns).rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700273/* Called with ns->shm_ids(ns).rwsem locked */
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700274static int shm_try_destroy_orphaned(int id, void *p, void *data)
275{
276 struct ipc_namespace *ns = data;
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400277 struct kern_ipc_perm *ipcp = p;
278 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700279
280 /*
281 * We want to destroy segments without users and with already
282 * exit'ed originating process.
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400283 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700284 * As shp->* are changed under rwsem, it's safe to skip shp locking.
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700285 */
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400286 if (shp->shm_creator != NULL)
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700287 return 0;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700288
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400289 if (shm_may_destroy(ns, shp)) {
290 shm_lock_by_ptr(shp);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700291 shm_destroy(ns, shp);
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400292 }
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700293 return 0;
294}
295
296void shm_destroy_orphaned(struct ipc_namespace *ns)
297{
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700298 down_write(&shm_ids(ns).rwsem);
Vasiliy Kulikov33a30ed2011-08-03 22:26:55 +0400299 if (shm_ids(ns).in_use)
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400300 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700301 up_write(&shm_ids(ns).rwsem);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700302}
303
Jack Miller83293c02014-08-08 14:23:21 -0700304/* Locking assumes this will only be called with task == current */
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700305void exit_shm(struct task_struct *task)
306{
Vasiliy Kulikov4c677e22011-07-29 03:56:40 +0400307 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
Jack Millerab602f72014-08-08 14:23:19 -0700308 struct shmid_kernel *shp, *n;
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700309
Jack Miller83293c02014-08-08 14:23:21 -0700310 if (list_empty(&task->sysvshm.shm_clist))
Vasiliy Kulikov298507d2011-08-03 22:28:26 +0400311 return;
312
Jack Miller83293c02014-08-08 14:23:21 -0700313 /*
314 * If kernel.shm_rmid_forced is not set then only keep track of
315 * which shmids are orphaned, so that a later set of the sysctl
316 * can clean them up.
317 */
318 if (!ns->shm_rmid_forced) {
319 down_read(&shm_ids(ns).rwsem);
320 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
321 shp->shm_creator = NULL;
322 /*
323 * Only under read lock but we are only called on current
324 * so no entry on the list will be shared.
325 */
326 list_del(&task->sysvshm.shm_clist);
327 up_read(&shm_ids(ns).rwsem);
328 return;
329 }
330
331 /*
332 * Destroy all already created segments, that were not yet mapped,
333 * and mark any mapped as orphan to cover the sysctl toggling.
334 * Destroy is skipped if shm_may_destroy() returns false.
335 */
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700336 down_write(&shm_ids(ns).rwsem);
Jack Miller83293c02014-08-08 14:23:21 -0700337 list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
338 shp->shm_creator = NULL;
339
340 if (shm_may_destroy(ns, shp)) {
341 shm_lock_by_ptr(shp);
342 shm_destroy(ns, shp);
343 }
344 }
345
346 /* Remove the list head from any segments still attached. */
Jack Millerab602f72014-08-08 14:23:19 -0700347 list_del(&task->sysvshm.shm_clist);
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700348 up_write(&shm_ids(ns).rwsem);
Vasiliy Kulikovb34a6b12011-07-26 16:08:48 -0700349}
350
Nick Piggind0217ac2007-07-19 01:47:03 -0700351static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800352{
353 struct file *file = vma->vm_file;
354 struct shm_file_data *sfd = shm_file_data(file);
355
Nick Piggind0217ac2007-07-19 01:47:03 -0700356 return sfd->vm_ops->fault(vma, vmf);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800357}
358
359#ifdef CONFIG_NUMA
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700360static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800361{
362 struct file *file = vma->vm_file;
363 struct shm_file_data *sfd = shm_file_data(file);
364 int err = 0;
365 if (sfd->vm_ops->set_policy)
366 err = sfd->vm_ops->set_policy(vma, new);
367 return err;
368}
369
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700370static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
371 unsigned long addr)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800372{
373 struct file *file = vma->vm_file;
374 struct shm_file_data *sfd = shm_file_data(file);
375 struct mempolicy *pol = NULL;
376
377 if (sfd->vm_ops->get_policy)
378 pol = sfd->vm_ops->get_policy(vma, addr);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700379 else if (vma->vm_policy)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800380 pol = vma->vm_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700381
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800382 return pol;
383}
384#endif
385
Manfred Spraul239521f2014-01-27 17:07:04 -0800386static int shm_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800388 struct shm_file_data *sfd = shm_file_data(file);
David Howellsb0e15192006-01-06 00:11:42 -0800389 int ret;
390
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800391 ret = sfd->file->f_op->mmap(sfd->file, vma);
392 if (ret != 0)
393 return ret;
394 sfd->vm_ops = vma->vm_ops;
David Howells2e92a3b2007-07-31 00:37:24 -0700395#ifdef CONFIG_MMU
Nick Piggin54cb8822007-07-19 01:46:59 -0700396 BUG_ON(!sfd->vm_ops->fault);
David Howells2e92a3b2007-07-31 00:37:24 -0700397#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800398 vma->vm_ops = &shm_vm_ops;
399 shm_open(vma);
David Howellsb0e15192006-01-06 00:11:42 -0800400
401 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Kirill Korotaev4e982312006-10-02 02:18:22 -0700404static int shm_release(struct inode *ino, struct file *file)
405{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800406 struct shm_file_data *sfd = shm_file_data(file);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700407
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800408 put_ipc_ns(sfd->ns);
409 shm_file_data(file) = NULL;
410 kfree(sfd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700411 return 0;
412}
413
Josef Bacik02c24a82011-07-16 20:44:56 -0400414static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Adam Litke516dffd2007-03-01 15:46:08 -0800415{
Adam Litke516dffd2007-03-01 15:46:08 -0800416 struct shm_file_data *sfd = shm_file_data(file);
Adam Litke516dffd2007-03-01 15:46:08 -0800417
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200418 if (!sfd->file->f_op->fsync)
419 return -EINVAL;
Josef Bacik02c24a82011-07-16 20:44:56 -0400420 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
Adam Litke516dffd2007-03-01 15:46:08 -0800421}
422
Will Deacon7d8a4562012-06-07 14:21:13 -0700423static long shm_fallocate(struct file *file, int mode, loff_t offset,
424 loff_t len)
425{
426 struct shm_file_data *sfd = shm_file_data(file);
427
428 if (!sfd->file->f_op->fallocate)
429 return -EOPNOTSUPP;
430 return sfd->file->f_op->fallocate(file, mode, offset, len);
431}
432
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800433static unsigned long shm_get_unmapped_area(struct file *file,
434 unsigned long addr, unsigned long len, unsigned long pgoff,
435 unsigned long flags)
436{
437 struct shm_file_data *sfd = shm_file_data(file);
Al Viroc4caa772009-11-30 08:38:43 -0500438 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
439 pgoff, flags);
Adam Litke516dffd2007-03-01 15:46:08 -0800440}
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800441
Arjan van de Ven9a321442007-02-12 00:55:35 -0800442static const struct file_operations shm_file_operations = {
Kirill Korotaev4e982312006-10-02 02:18:22 -0700443 .mmap = shm_mmap,
Adam Litke516dffd2007-03-01 15:46:08 -0800444 .fsync = shm_fsync,
Kirill Korotaev4e982312006-10-02 02:18:22 -0700445 .release = shm_release,
David Howellsed5e5892010-01-15 17:01:32 -0800446#ifndef CONFIG_MMU
447 .get_unmapped_area = shm_get_unmapped_area,
448#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +0200449 .llseek = noop_llseek,
Will Deacon7d8a4562012-06-07 14:21:13 -0700450 .fallocate = shm_fallocate,
Al Viroc4caa772009-11-30 08:38:43 -0500451};
452
453static const struct file_operations shm_file_operations_huge = {
454 .mmap = shm_mmap,
455 .fsync = shm_fsync,
456 .release = shm_release,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800457 .get_unmapped_area = shm_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200458 .llseek = noop_llseek,
Will Deacon7d8a4562012-06-07 14:21:13 -0700459 .fallocate = shm_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460};
461
Al Viroc4caa772009-11-30 08:38:43 -0500462int is_file_shm_hugepages(struct file *file)
463{
464 return file->f_op == &shm_file_operations_huge;
465}
466
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400467static const struct vm_operations_struct shm_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 .open = shm_open, /* callback for a new vm-area open */
469 .close = shm_close, /* callback for when the vm-area is released */
Nick Piggin54cb8822007-07-19 01:46:59 -0700470 .fault = shm_fault,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800471#if defined(CONFIG_NUMA)
472 .set_policy = shm_set_policy,
473 .get_policy = shm_get_policy,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474#endif
475};
476
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700477/**
478 * newseg - Create a new shared memory segment
479 * @ns: namespace
480 * @params: ptr to the structure that contains key, size and shmflg
481 *
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700482 * Called with shm_ids.rwsem held as a writer.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700483 */
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700484static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700486 key_t key = params->key;
487 int shmflg = params->flg;
488 size_t size = params->u.size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 int error;
490 struct shmid_kernel *shp;
Robin Holtd69f3ba2013-04-30 19:15:54 -0700491 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Manfred Spraul239521f2014-01-27 17:07:04 -0800492 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 char name[13];
494 int id;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900495 vm_flags_t acctflag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Kirill Korotaev4e982312006-10-02 02:18:22 -0700497 if (size < SHMMIN || size > ns->shm_ctlmax)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return -EINVAL;
499
Manfred Spraul13763272014-06-06 14:37:41 -0700500 if (numpages << PAGE_SHIFT < size)
501 return -ENOSPC;
502
Manfred Spraul09c6eb12014-06-06 14:37:40 -0700503 if (ns->shm_tot + numpages < ns->shm_tot ||
504 ns->shm_tot + numpages > ns->shm_ctlall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 return -ENOSPC;
506
507 shp = ipc_rcu_alloc(sizeof(*shp));
508 if (!shp)
509 return -ENOMEM;
510
511 shp->shm_perm.key = key;
Andrew Mortonb33291c2006-01-08 01:02:21 -0800512 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 shp->mlock_user = NULL;
514
515 shp->shm_perm.security = NULL;
516 error = security_shm_alloc(shp);
517 if (error) {
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700518 ipc_rcu_putref(shp, ipc_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 return error;
520 }
521
Manfred Spraul239521f2014-01-27 17:07:04 -0800522 sprintf(name, "SYSV%08x", key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 if (shmflg & SHM_HUGETLB) {
Andrew Mortonc103a4d2013-07-08 16:01:08 -0700524 struct hstate *hs;
Li Zefan091d0d52013-05-09 15:08:15 +0800525 size_t hugesize;
526
Andrew Mortonc103a4d2013-07-08 16:01:08 -0700527 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
Li Zefan091d0d52013-05-09 15:08:15 +0800528 if (!hs) {
529 error = -EINVAL;
530 goto no_file;
531 }
532 hugesize = ALIGN(size, huge_page_size(hs));
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -0700533
Mel Gorman5a6fe122009-02-10 14:02:27 +0000534 /* hugetlb_file_setup applies strict accounting */
535 if (shmflg & SHM_NORESERVE)
536 acctflag = VM_NORESERVE;
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -0700537 file = hugetlb_file_setup(name, hugesize, acctflag,
Andi Kleen42d73952012-12-11 16:01:34 -0800538 &shp->mlock_user, HUGETLB_SHMFS_INODE,
539 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 } else {
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800541 /*
542 * Do not allow no accounting for OVERCOMMIT_NEVER, even
Manfred Spraul239521f2014-01-27 17:07:04 -0800543 * if it's asked for.
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800544 */
545 if ((shmflg & SHM_NORESERVE) &&
546 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -0800547 acctflag = VM_NORESERVE;
Stephen Smalleye1832f22015-08-06 15:46:55 -0700548 file = shmem_kernel_file_setup(name, size, acctflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
550 error = PTR_ERR(file);
551 if (IS_ERR(file))
552 goto no_file;
553
Pierre Peiffer48dea402008-04-29 01:00:35 -0700554 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700555 if (id < 0) {
556 error = id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 goto no_id;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700560 shp->shm_cprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 shp->shm_lprid = 0;
562 shp->shm_atim = shp->shm_dtim = 0;
563 shp->shm_ctim = get_seconds();
564 shp->shm_segsz = size;
565 shp->shm_nattch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 shp->shm_file = file;
Vasiliy Kulikov5774ed02011-07-29 03:55:31 +0400567 shp->shm_creator = current;
Jack Millerab602f72014-08-08 14:23:19 -0700568 list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700569
Badari Pulavarty30475cc2007-06-16 10:15:59 -0700570 /*
571 * shmid gets reported as "inode#" in /proc/pid/maps.
572 * proc-ps tools use this. Changing this will break them.
573 */
Al Viro496ad9a2013-01-23 17:07:38 -0500574 file_inode(file)->i_ino = shp->shm_perm.id;
Krishnakumar R551110a2005-10-29 18:16:45 -0700575
Kirill Korotaev4e982312006-10-02 02:18:22 -0700576 ns->shm_tot += numpages;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700577 error = shp->shm_perm.id;
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700578
Davidlohr Buesocf9d5d72013-07-08 16:01:11 -0700579 ipc_unlock_object(&shp->shm_perm);
Davidlohr Buesodbfcd912013-07-08 16:01:09 -0700580 rcu_read_unlock();
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700581 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583no_id:
Hugh Dickins2195d282009-09-12 12:21:27 +0100584 if (is_file_hugepages(file) && shp->mlock_user)
Hugh Dickins353d5c32009-08-24 16:30:28 +0100585 user_shm_unlock(size, shp->mlock_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 fput(file);
587no_file:
Davidlohr Bueso53dad6d2013-09-23 17:04:45 -0700588 ipc_rcu_putref(shp, shm_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return error;
590}
591
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700592/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700593 * Called with shm_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700594 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700595static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700596{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700597 struct shmid_kernel *shp;
598
599 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
600 return security_shm_associate(shp, shmflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700601}
602
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700603/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700604 * Called with shm_ids.rwsem and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700605 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700606static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
607 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700608{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700609 struct shmid_kernel *shp;
610
611 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
612 if (shp->shm_segsz < params->u.size)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700613 return -EINVAL;
614
615 return 0;
616}
617
Heiko Carstensd5460c92009-01-14 14:14:27 +0100618SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700620 struct ipc_namespace *ns;
Mathias Krauseeb66ec42014-06-06 14:37:36 -0700621 static const struct ipc_ops shm_ops = {
622 .getnew = newseg,
623 .associate = shm_security,
624 .more_checks = shm_more_checks,
625 };
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700626 struct ipc_params shm_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Kirill Korotaev4e982312006-10-02 02:18:22 -0700628 ns = current->nsproxy->ipc_ns;
629
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700630 shm_params.key = key;
631 shm_params.flg = shmflg;
632 shm_params.u.size = size;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700633
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700634 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
637static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
638{
Manfred Spraul239521f2014-01-27 17:07:04 -0800639 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 case IPC_64:
641 return copy_to_user(buf, in, sizeof(*in));
642 case IPC_OLD:
643 {
644 struct shmid_ds out;
645
Vasiliy Kulikov3af54c92010-10-30 18:22:49 +0400646 memset(&out, 0, sizeof(out));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
648 out.shm_segsz = in->shm_segsz;
649 out.shm_atime = in->shm_atime;
650 out.shm_dtime = in->shm_dtime;
651 out.shm_ctime = in->shm_ctime;
652 out.shm_cpid = in->shm_cpid;
653 out.shm_lpid = in->shm_lpid;
654 out.shm_nattch = in->shm_nattch;
655
656 return copy_to_user(buf, &out, sizeof(out));
657 }
658 default:
659 return -EINVAL;
660 }
661}
662
Pierre Peiffer016d7132008-04-29 01:00:50 -0700663static inline unsigned long
664copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Manfred Spraul239521f2014-01-27 17:07:04 -0800666 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -0700668 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 case IPC_OLD:
672 {
673 struct shmid_ds tbuf_old;
674
675 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
676 return -EFAULT;
677
Pierre Peiffer016d7132008-04-29 01:00:50 -0700678 out->shm_perm.uid = tbuf_old.shm_perm.uid;
679 out->shm_perm.gid = tbuf_old.shm_perm.gid;
680 out->shm_perm.mode = tbuf_old.shm_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 return 0;
683 }
684 default:
685 return -EINVAL;
686 }
687}
688
689static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
690{
Manfred Spraul239521f2014-01-27 17:07:04 -0800691 switch (version) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 case IPC_64:
693 return copy_to_user(buf, in, sizeof(*in));
694 case IPC_OLD:
695 {
696 struct shminfo out;
697
Manfred Spraul239521f2014-01-27 17:07:04 -0800698 if (in->shmmax > INT_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 out.shmmax = INT_MAX;
700 else
701 out.shmmax = (int)in->shmmax;
702
703 out.shmmin = in->shmmin;
704 out.shmmni = in->shmmni;
705 out.shmseg = in->shmseg;
Paul McQuade46c0a8c2014-06-06 14:37:37 -0700706 out.shmall = in->shmall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
708 return copy_to_user(buf, &out, sizeof(out));
709 }
710 default:
711 return -EINVAL;
712 }
713}
714
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700715/*
Helge Dellerb7952182010-10-27 15:34:16 -0700716 * Calculate and add used RSS and swap pages of a shm.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700717 * Called with shm_ids.rwsem held as a reader
Helge Dellerb7952182010-10-27 15:34:16 -0700718 */
719static void shm_add_rss_swap(struct shmid_kernel *shp,
720 unsigned long *rss_add, unsigned long *swp_add)
721{
722 struct inode *inode;
723
Al Viro496ad9a2013-01-23 17:07:38 -0500724 inode = file_inode(shp->shm_file);
Helge Dellerb7952182010-10-27 15:34:16 -0700725
726 if (is_file_hugepages(shp->shm_file)) {
727 struct address_space *mapping = inode->i_mapping;
728 struct hstate *h = hstate_file(shp->shm_file);
729 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
730 } else {
731#ifdef CONFIG_SHMEM
732 struct shmem_inode_info *info = SHMEM_I(inode);
733 spin_lock(&info->lock);
734 *rss_add += inode->i_mapping->nrpages;
735 *swp_add += info->swapped;
736 spin_unlock(&info->lock);
737#else
738 *rss_add += inode->i_mapping->nrpages;
739#endif
740 }
741}
742
743/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700744 * Called with shm_ids.rwsem held as a reader
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700745 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700746static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
747 unsigned long *swp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700749 int next_id;
750 int total, in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 *rss = 0;
753 *swp = 0;
754
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700755 in_use = shm_ids(ns).in_use;
756
757 for (total = 0, next_id = 0; total < in_use; next_id++) {
Tony Battersbye562aeb2009-04-02 16:58:26 -0700758 struct kern_ipc_perm *ipc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Tony Battersbye562aeb2009-04-02 16:58:26 -0700761 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
762 if (ipc == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 continue;
Tony Battersbye562aeb2009-04-02 16:58:26 -0700764 shp = container_of(ipc, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Helge Dellerb7952182010-10-27 15:34:16 -0700766 shm_add_rss_swap(shp, rss, swp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700767
768 total++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 }
770}
771
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700772/*
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700773 * This function handles some shmctl commands which require the rwsem
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700774 * to be held in write mode.
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700775 * NOTE: no locks must be held, the rwsem is taken inside this function.
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700776 */
777static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
778 struct shmid_ds __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700780 struct kern_ipc_perm *ipcp;
Pierre Peiffer016d7132008-04-29 01:00:50 -0700781 struct shmid64_ds shmid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 struct shmid_kernel *shp;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700783 int err;
784
785 if (cmd == IPC_SET) {
Pierre Peiffer016d7132008-04-29 01:00:50 -0700786 if (copy_shmid_from_user(&shmid64, buf, version))
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700787 return -EFAULT;
788 }
789
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700790 down_write(&shm_ids(ns).rwsem);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700791 rcu_read_lock();
792
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700793 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
794 &shmid64.shm_perm, 0);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700795 if (IS_ERR(ipcp)) {
796 err = PTR_ERR(ipcp);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700797 goto out_unlock1;
798 }
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700799
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700800 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700801
802 err = security_shm_shmctl(shp, cmd);
803 if (err)
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700804 goto out_unlock1;
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700805
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700806 switch (cmd) {
807 case IPC_RMID:
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700808 ipc_lock_object(&shp->shm_perm);
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700809 /* do_shm_rmid unlocks the ipc object and rcu */
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700810 do_shm_rmid(ns, ipcp);
811 goto out_up;
812 case IPC_SET:
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700813 ipc_lock_object(&shp->shm_perm);
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800814 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
815 if (err)
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700816 goto out_unlock0;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700817 shp->shm_ctim = get_seconds();
818 break;
819 default:
820 err = -EINVAL;
Davidlohr Bueso79ccf0f2013-09-11 14:26:16 -0700821 goto out_unlock1;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700822 }
Davidlohr Bueso7b4cc5d2013-07-08 16:01:12 -0700823
824out_unlock0:
825 ipc_unlock_object(&shp->shm_perm);
826out_unlock1:
827 rcu_read_unlock();
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700828out_up:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700829 up_write(&shm_ids(ns).rwsem);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700830 return err;
831}
832
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700833static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
834 int cmd, int version, void __user *buf)
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700835{
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700836 int err;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700837 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700839 /* preliminary security checks for *_INFO */
840 if (cmd == IPC_INFO || cmd == SHM_INFO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 err = security_shm_shmctl(NULL, cmd);
842 if (err)
843 return err;
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700844 }
845
846 switch (cmd) {
847 case IPC_INFO:
848 {
849 struct shminfo64 shminfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
WANG Conge8148f72009-01-06 14:42:49 -0800851 memset(&shminfo, 0, sizeof(shminfo));
Kirill Korotaev4e982312006-10-02 02:18:22 -0700852 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
853 shminfo.shmmax = ns->shm_ctlmax;
854 shminfo.shmall = ns->shm_ctlall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 shminfo.shmmin = SHMMIN;
Manfred Spraul239521f2014-01-27 17:07:04 -0800857 if (copy_shminfo_to_user(buf, &shminfo, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 return -EFAULT;
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700859
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700860 down_read(&shm_ids(ns).rwsem);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700861 err = ipc_get_maxid(&shm_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700862 up_read(&shm_ids(ns).rwsem);
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700863
Manfred Spraul239521f2014-01-27 17:07:04 -0800864 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 err = 0;
866 goto out;
867 }
868 case SHM_INFO:
869 {
870 struct shm_info shm_info;
871
WANG Conge8148f72009-01-06 14:42:49 -0800872 memset(&shm_info, 0, sizeof(shm_info));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700873 down_read(&shm_ids(ns).rwsem);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700874 shm_info.used_ids = shm_ids(ns).in_use;
Manfred Spraul239521f2014-01-27 17:07:04 -0800875 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700876 shm_info.shm_tot = ns->shm_tot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 shm_info.swap_attempts = 0;
878 shm_info.swap_successes = 0;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700879 err = ipc_get_maxid(&shm_ids(ns));
Davidlohr Buesod9a605e2013-09-11 14:26:24 -0700880 up_read(&shm_ids(ns).rwsem);
WANG Conge8148f72009-01-06 14:42:49 -0800881 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 err = -EFAULT;
883 goto out;
884 }
885
886 err = err < 0 ? 0 : err;
887 goto out;
888 }
889 case SHM_STAT:
890 case IPC_STAT:
891 {
892 struct shmid64_ds tbuf;
893 int result;
Nadia Derbey023a5352007-10-18 23:40:51 -0700894
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700895 rcu_read_lock();
Nadia Derbey023a5352007-10-18 23:40:51 -0700896 if (cmd == SHM_STAT) {
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700897 shp = shm_obtain_object(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 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700902 result = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 } else {
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700904 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700905 if (IS_ERR(shp)) {
906 err = PTR_ERR(shp);
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700907 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 result = 0;
910 }
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700911
WANG Conge8148f72009-01-06 14:42:49 -0800912 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700913 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 goto out_unlock;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 err = security_shm_shmctl(shp, cmd);
917 if (err)
918 goto out_unlock;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700919
Nadia Derbey023a5352007-10-18 23:40:51 -0700920 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
922 tbuf.shm_segsz = shp->shm_segsz;
923 tbuf.shm_atime = shp->shm_atim;
924 tbuf.shm_dtime = shp->shm_dtim;
925 tbuf.shm_ctime = shp->shm_ctim;
926 tbuf.shm_cpid = shp->shm_cprid;
927 tbuf.shm_lpid = shp->shm_lprid;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800928 tbuf.shm_nattch = shp->shm_nattch;
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700929 rcu_read_unlock();
930
931 if (copy_shmid_to_user(buf, &tbuf, version))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 err = -EFAULT;
933 else
934 err = result;
935 goto out;
936 }
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700937 default:
938 return -EINVAL;
939 }
940
941out_unlock:
Davidlohr Buesoc97cb9c2013-09-11 14:26:20 -0700942 rcu_read_unlock();
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700943out:
944 return err;
945}
946
947SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
948{
949 struct shmid_kernel *shp;
950 int err, version;
951 struct ipc_namespace *ns;
952
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700953 if (cmd < 0 || shmid < 0)
954 return -EINVAL;
Davidlohr Bueso68eccc12013-09-11 14:26:18 -0700955
956 version = ipc_parse_version(&cmd);
957 ns = current->nsproxy->ipc_ns;
958
959 switch (cmd) {
960 case IPC_INFO:
961 case SHM_INFO:
962 case SHM_STAT:
963 case IPC_STAT:
964 return shmctl_nolock(ns, shmid, cmd, version, buf);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700965 case IPC_RMID:
966 case IPC_SET:
967 return shmctl_down(ns, shmid, cmd, buf, version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 case SHM_LOCK:
969 case SHM_UNLOCK:
970 {
Hugh Dickins850465792012-01-20 14:34:19 -0800971 struct file *shm_file;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -0700972
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700973 rcu_read_lock();
974 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700975 if (IS_ERR(shp)) {
976 err = PTR_ERR(shp);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700977 goto out_unlock1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
Al Viroa33e6752008-12-10 03:40:06 -0500980 audit_ipc_obj(&(shp->shm_perm));
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700981 err = security_shm_shmctl(shp, cmd);
982 if (err)
983 goto out_unlock1;
Steve Grubb073115d2006-04-02 17:07:33 -0400984
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700985 ipc_lock_object(&shp->shm_perm);
Rafael Aquini0f3d2b02014-01-27 17:07:01 -0800986
987 /* check if shm_destroy() is tearing down shp */
988 if (!ipc_valid_object(&shp->shm_perm)) {
989 err = -EIDRM;
990 goto out_unlock0;
991 }
992
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700993 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800994 kuid_t euid = current_euid();
Eric W. Biederman1efdb692012-02-07 16:54:11 -0800995 if (!uid_eq(euid, shp->shm_perm.uid) &&
Jesper Nilsson3a726602013-11-21 14:32:08 -0800996 !uid_eq(euid, shp->shm_perm.cuid)) {
997 err = -EPERM;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -0700998 goto out_unlock0;
Jesper Nilsson3a726602013-11-21 14:32:08 -0800999 }
1000 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1001 err = -EPERM;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001002 goto out_unlock0;
Jesper Nilsson3a726602013-11-21 14:32:08 -08001003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005
Hugh Dickins850465792012-01-20 14:34:19 -08001006 shm_file = shp->shm_file;
1007 if (is_file_hugepages(shm_file))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001008 goto out_unlock0;
Hugh Dickins850465792012-01-20 14:34:19 -08001009
1010 if (cmd == SHM_LOCK) {
David Howells86a264a2008-11-14 10:39:18 +11001011 struct user_struct *user = current_user();
Hugh Dickins850465792012-01-20 14:34:19 -08001012 err = shmem_lock(shm_file, 1, user);
1013 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1014 shp->shm_perm.mode |= SHM_LOCKED;
1015 shp->mlock_user = user;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001017 goto out_unlock0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
Hugh Dickins850465792012-01-20 14:34:19 -08001019
1020 /* SHM_UNLOCK */
1021 if (!(shp->shm_perm.mode & SHM_LOCKED))
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001022 goto out_unlock0;
Hugh Dickins850465792012-01-20 14:34:19 -08001023 shmem_lock(shm_file, 0, shp->mlock_user);
1024 shp->shm_perm.mode &= ~SHM_LOCKED;
1025 shp->mlock_user = NULL;
1026 get_file(shm_file);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001027 ipc_unlock_object(&shp->shm_perm);
1028 rcu_read_unlock();
Hugh Dickins24513262012-01-20 14:34:21 -08001029 shmem_unlock_mapping(shm_file->f_mapping);
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001030
Hugh Dickins850465792012-01-20 14:34:19 -08001031 fput(shm_file);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -07001032 return err;
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001033 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 default:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -07001035 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037
Davidlohr Bueso2caacaa2013-09-11 14:26:21 -07001038out_unlock0:
1039 ipc_unlock_object(&shp->shm_perm);
1040out_unlock1:
1041 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 return err;
1043}
1044
1045/*
1046 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1047 *
1048 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1049 * "raddr" thing points to kernel space, and there has to be a wrapper around
1050 * this.
1051 */
Will Deacon079a96a2012-07-30 14:42:38 -07001052long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1053 unsigned long shmlba)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
1055 struct shmid_kernel *shp;
1056 unsigned long addr;
1057 unsigned long size;
Manfred Spraul239521f2014-01-27 17:07:04 -08001058 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 int err;
1060 unsigned long flags;
1061 unsigned long prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 int acc_mode;
Kirill Korotaev4e982312006-10-02 02:18:22 -07001063 struct ipc_namespace *ns;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001064 struct shm_file_data *sfd;
1065 struct path path;
Al Viroaeb5d722008-09-02 15:28:45 -04001066 fmode_t f_mode;
Michel Lespinasse41badc12013-02-22 16:32:47 -08001067 unsigned long populate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001069 err = -EINVAL;
1070 if (shmid < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 goto out;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001072 else if ((addr = (ulong)shmaddr)) {
Will Deacon079a96a2012-07-30 14:42:38 -07001073 if (addr & (shmlba - 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (shmflg & SHM_RND)
Will Deacon079a96a2012-07-30 14:42:38 -07001075 addr &= ~(shmlba - 1); /* round down */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 else
1077#ifndef __ARCH_FORCE_SHMLBA
1078 if (addr & ~PAGE_MASK)
1079#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001080 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082 flags = MAP_SHARED | MAP_FIXED;
1083 } else {
1084 if ((shmflg & SHM_REMAP))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001085 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 flags = MAP_SHARED;
1088 }
1089
1090 if (shmflg & SHM_RDONLY) {
1091 prot = PROT_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 acc_mode = S_IRUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001093 f_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 } else {
1095 prot = PROT_READ | PROT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 acc_mode = S_IRUGO | S_IWUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001097 f_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 }
1099 if (shmflg & SHM_EXEC) {
1100 prot |= PROT_EXEC;
1101 acc_mode |= S_IXUGO;
1102 }
1103
1104 /*
1105 * We cannot rely on the fs check since SYSV IPC does have an
1106 * additional creator id...
1107 */
Kirill Korotaev4e982312006-10-02 02:18:22 -07001108 ns = current->nsproxy->ipc_ns;
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001109 rcu_read_lock();
1110 shp = shm_obtain_object_check(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001111 if (IS_ERR(shp)) {
1112 err = PTR_ERR(shp);
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001113 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -07001114 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001115
1116 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -07001117 if (ipcperms(ns, &shp->shm_perm, acc_mode))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001118 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120 err = security_shm_shmat(shp, shmaddr, shmflg);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001121 if (err)
1122 goto out_unlock;
1123
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001124 ipc_lock_object(&shp->shm_perm);
Greg Thelena399b292013-11-21 14:32:00 -08001125
1126 /* check if shm_destroy() is tearing down shp */
Rafael Aquini0f3d2b02014-01-27 17:07:01 -08001127 if (!ipc_valid_object(&shp->shm_perm)) {
Greg Thelena399b292013-11-21 14:32:00 -08001128 ipc_unlock_object(&shp->shm_perm);
1129 err = -EIDRM;
1130 goto out_unlock;
1131 }
1132
Al Viro2c48b9c2009-08-09 00:52:35 +04001133 path = shp->shm_file->f_path;
1134 path_get(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 shp->shm_nattch++;
David Howells75c3cfa2015-03-17 22:26:12 +00001136 size = i_size_read(d_inode(path.dentry));
Davidlohr Buesoc2c737a2013-09-11 14:26:23 -07001137 ipc_unlock_object(&shp->shm_perm);
1138 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001140 err = -ENOMEM;
1141 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001142 if (!sfd) {
1143 path_put(&path);
1144 goto out_nattch;
1145 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001146
Al Viro2c48b9c2009-08-09 00:52:35 +04001147 file = alloc_file(&path, f_mode,
1148 is_file_hugepages(shp->shm_file) ?
Al Viroc4caa772009-11-30 08:38:43 -05001149 &shm_file_operations_huge :
1150 &shm_file_operations);
Anatol Pomozov39b65252012-09-12 20:11:55 -07001151 err = PTR_ERR(file);
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001152 if (IS_ERR(file)) {
1153 kfree(sfd);
1154 path_put(&path);
1155 goto out_nattch;
1156 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001157
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001158 file->private_data = sfd;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001159 file->f_mapping = shp->shm_file->f_mapping;
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001160 sfd->id = shp->shm_perm.id;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001161 sfd->ns = get_ipc_ns(ns);
1162 sfd->file = shp->shm_file;
1163 sfd->vm_ops = NULL;
1164
Al Viro8b3ec682012-05-30 17:11:23 -04001165 err = security_mmap_file(file, prot, flags);
1166 if (err)
1167 goto out_fput;
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 down_write(&current->mm->mmap_sem);
1170 if (addr && !(shmflg & SHM_REMAP)) {
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001171 err = -EINVAL;
Manfred Spraul247a8ce2014-06-06 14:37:38 -07001172 if (addr + size < addr)
1173 goto invalid;
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (find_vma_intersection(current->mm, addr, addr + size))
1176 goto invalid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
Davidlohr Buesof42569b2013-09-11 14:26:22 -07001178
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001179 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1180 *raddr = addr;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001181 err = 0;
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001182 if (IS_ERR_VALUE(addr))
1183 err = (long)addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184invalid:
1185 up_write(&current->mm->mmap_sem);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -08001186 if (populate)
Michel Lespinasse41badc12013-02-22 16:32:47 -08001187 mm_populate(addr, populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Al Viro8b3ec682012-05-30 17:11:23 -04001189out_fput:
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001190 fput(file);
1191
1192out_nattch:
Davidlohr Buesod9a605e2013-09-11 14:26:24 -07001193 down_write(&shm_ids(ns).rwsem);
Nadia Derbey00c2bf82008-07-25 01:48:03 -07001194 shp = shm_lock(ns, shmid);
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