blob: 729acb7e31487f67660a3202d818f1f55b4fd6e5 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 */
23
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
25#include <linux/mm.h>
26#include <linux/hugetlb.h>
27#include <linux/shm.h>
28#include <linux/init.h>
29#include <linux/file.h>
30#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/shmem_fs.h>
32#include <linux/security.h>
33#include <linux/syscalls.h>
34#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080035#include <linux/capability.h>
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -070036#include <linux/ptrace.h>
Mike Waychison19b49462005-09-06 15:17:10 -070037#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070038#include <linux/rwsem.h>
Kirill Korotaev4e982312006-10-02 02:18:22 -070039#include <linux/nsproxy.h>
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080040#include <linux/mount.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080041#include <linux/ipc_namespace.h>
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <asm/uaccess.h>
44
45#include "util.h"
46
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080047struct shm_file_data {
48 int id;
49 struct ipc_namespace *ns;
50 struct file *file;
51 const struct vm_operations_struct *vm_ops;
52};
53
54#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
Arjan van de Ven9a321442007-02-12 00:55:35 -080056static const struct file_operations shm_file_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040057static const struct vm_operations_struct shm_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080059#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Kirill Korotaev4e982312006-10-02 02:18:22 -070061#define shm_unlock(shp) \
62 ipc_unlock(&(shp)->shm_perm)
Kirill Korotaev4e982312006-10-02 02:18:22 -070063
Nadia Derbey7748dbf2007-10-18 23:40:49 -070064static int newseg(struct ipc_namespace *, struct ipc_params *);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080065static void shm_open(struct vm_area_struct *vma);
66static void shm_close(struct vm_area_struct *vma);
Kirill Korotaev4e982312006-10-02 02:18:22 -070067static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -070069static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#endif
71
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080072void shm_init_ns(struct ipc_namespace *ns)
Kirill Korotaev4e982312006-10-02 02:18:22 -070073{
Kirill Korotaev4e982312006-10-02 02:18:22 -070074 ns->shm_ctlmax = SHMMAX;
75 ns->shm_ctlall = SHMALL;
76 ns->shm_ctlmni = SHMMNI;
77 ns->shm_tot = 0;
WANG Conge8148f72009-01-06 14:42:49 -080078 ipc_init_ids(&shm_ids(ns));
Kirill Korotaev4e982312006-10-02 02:18:22 -070079}
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Nadia Derbeyf4566f02007-10-18 23:40:53 -070081/*
Nadia Derbey3e148c72007-10-18 23:40:54 -070082 * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
83 * Only shm_ids.rw_mutex remains locked on exit.
Nadia Derbeyf4566f02007-10-18 23:40:53 -070084 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -080085static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Kirill Korotaev4e982312006-10-02 02:18:22 -070086{
Pierre Peiffer01b8b072008-02-08 04:18:57 -080087 struct shmid_kernel *shp;
88 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
89
Kirill Korotaev4e982312006-10-02 02:18:22 -070090 if (shp->shm_nattch){
91 shp->shm_perm.mode |= SHM_DEST;
92 /* Do not find it any more */
93 shp->shm_perm.key = IPC_PRIVATE;
94 shm_unlock(shp);
95 } else
96 shm_destroy(ns, shp);
97}
98
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080099#ifdef CONFIG_IPC_NS
Kirill Korotaev4e982312006-10-02 02:18:22 -0700100void shm_exit_ns(struct ipc_namespace *ns)
101{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800102 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
Serge E. Hallyn7d6feeb2009-12-15 16:47:27 -0800103 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700104}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800105#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107void __init shm_init (void)
108{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800109 shm_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700110 ipc_init_proc_interface("sysvipc/shm",
Helge Dellerb7952182010-10-27 15:34:16 -0700111#if BITS_PER_LONG <= 32
112 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
113#else
114 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
115#endif
Kirill Korotaev4e982312006-10-02 02:18:22 -0700116 IPC_SHM_IDS, sysvipc_shm_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117}
118
Nadia Derbey3e148c72007-10-18 23:40:54 -0700119/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700120 * shm_lock_(check_) routines are called in the paths where the rw_mutex
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700121 * is not necessarily held.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700122 */
Nadia Derbey023a5352007-10-18 23:40:51 -0700123static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700125 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
126
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800127 if (IS_ERR(ipcp))
128 return (struct shmid_kernel *)ipcp;
129
Nadia Derbey03f02c72007-10-18 23:40:51 -0700130 return container_of(ipcp, struct shmid_kernel, shm_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700131}
132
133static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
134 int id)
135{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700136 struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
137
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800138 if (IS_ERR(ipcp))
139 return (struct shmid_kernel *)ipcp;
140
Nadia Derbey03f02c72007-10-18 23:40:51 -0700141 return container_of(ipcp, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700144static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700146 ipc_rmid(&shm_ids(ns), &s->shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800150/* This is called by fork, once for every shm attach. */
151static void shm_open(struct vm_area_struct *vma)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700152{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800153 struct file *file = vma->vm_file;
154 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 struct shmid_kernel *shp;
156
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800157 shp = shm_lock(sfd->ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700158 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 shp->shm_atim = get_seconds();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700160 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 shp->shm_nattch++;
162 shm_unlock(shp);
163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165/*
166 * shm_destroy - free the struct shmid_kernel
167 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700168 * @ns: namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 * @shp: struct to free
170 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700171 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 * but returns with shp unlocked and freed.
173 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700174static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700176 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700177 shm_rmid(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 shm_unlock(shp);
179 if (!is_file_hugepages(shp->shm_file))
180 shmem_lock(shp->shm_file, 0, shp->mlock_user);
Hugh Dickins353d5c32009-08-24 16:30:28 +0100181 else if (shp->mlock_user)
Josef Sipek6d630792006-12-08 02:37:11 -0800182 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 shp->mlock_user);
184 fput (shp->shm_file);
185 security_shm_free(shp);
186 ipc_rcu_putref(shp);
187}
188
189/*
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800190 * remove the attach descriptor vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 * free memory for segment if it is marked destroyed.
192 * The descriptor has already been removed from the current->mm->mmap list
193 * and will later be kfree()d.
194 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800195static void shm_close(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800197 struct file * file = vma->vm_file;
198 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 struct shmid_kernel *shp;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800200 struct ipc_namespace *ns = sfd->ns;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700201
Nadia Derbey3e148c72007-10-18 23:40:54 -0700202 down_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 /* remove from the list of attaches of the shm segment */
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700204 shp = shm_lock(ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700205 BUG_ON(IS_ERR(shp));
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700206 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 shp->shm_dtim = get_seconds();
208 shp->shm_nattch--;
209 if(shp->shm_nattch == 0 &&
Andrew Mortonb33291c2006-01-08 01:02:21 -0800210 shp->shm_perm.mode & SHM_DEST)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700211 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 else
213 shm_unlock(shp);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700214 up_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
Nick Piggind0217ac2007-07-19 01:47:03 -0700217static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800218{
219 struct file *file = vma->vm_file;
220 struct shm_file_data *sfd = shm_file_data(file);
221
Nick Piggind0217ac2007-07-19 01:47:03 -0700222 return sfd->vm_ops->fault(vma, vmf);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800223}
224
225#ifdef CONFIG_NUMA
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700226static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800227{
228 struct file *file = vma->vm_file;
229 struct shm_file_data *sfd = shm_file_data(file);
230 int err = 0;
231 if (sfd->vm_ops->set_policy)
232 err = sfd->vm_ops->set_policy(vma, new);
233 return err;
234}
235
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700236static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
237 unsigned long addr)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800238{
239 struct file *file = vma->vm_file;
240 struct shm_file_data *sfd = shm_file_data(file);
241 struct mempolicy *pol = NULL;
242
243 if (sfd->vm_ops->get_policy)
244 pol = sfd->vm_ops->get_policy(vma, addr);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700245 else if (vma->vm_policy)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800246 pol = vma->vm_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700247
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800248 return pol;
249}
250#endif
251
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252static int shm_mmap(struct file * file, struct vm_area_struct * vma)
253{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800254 struct shm_file_data *sfd = shm_file_data(file);
David Howellsb0e15192006-01-06 00:11:42 -0800255 int ret;
256
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800257 ret = sfd->file->f_op->mmap(sfd->file, vma);
258 if (ret != 0)
259 return ret;
260 sfd->vm_ops = vma->vm_ops;
David Howells2e92a3b2007-07-31 00:37:24 -0700261#ifdef CONFIG_MMU
Nick Piggin54cb8822007-07-19 01:46:59 -0700262 BUG_ON(!sfd->vm_ops->fault);
David Howells2e92a3b2007-07-31 00:37:24 -0700263#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800264 vma->vm_ops = &shm_vm_ops;
265 shm_open(vma);
David Howellsb0e15192006-01-06 00:11:42 -0800266
267 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268}
269
Kirill Korotaev4e982312006-10-02 02:18:22 -0700270static int shm_release(struct inode *ino, struct file *file)
271{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800272 struct shm_file_data *sfd = shm_file_data(file);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700273
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800274 put_ipc_ns(sfd->ns);
275 shm_file_data(file) = NULL;
276 kfree(sfd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700277 return 0;
278}
279
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200280static int shm_fsync(struct file *file, int datasync)
Adam Litke516dffd2007-03-01 15:46:08 -0800281{
Adam Litke516dffd2007-03-01 15:46:08 -0800282 struct shm_file_data *sfd = shm_file_data(file);
Adam Litke516dffd2007-03-01 15:46:08 -0800283
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200284 if (!sfd->file->f_op->fsync)
285 return -EINVAL;
286 return sfd->file->f_op->fsync(sfd->file, datasync);
Adam Litke516dffd2007-03-01 15:46:08 -0800287}
288
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800289static unsigned long shm_get_unmapped_area(struct file *file,
290 unsigned long addr, unsigned long len, unsigned long pgoff,
291 unsigned long flags)
292{
293 struct shm_file_data *sfd = shm_file_data(file);
Al Viroc4caa772009-11-30 08:38:43 -0500294 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
295 pgoff, flags);
Adam Litke516dffd2007-03-01 15:46:08 -0800296}
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800297
Arjan van de Ven9a321442007-02-12 00:55:35 -0800298static const struct file_operations shm_file_operations = {
Kirill Korotaev4e982312006-10-02 02:18:22 -0700299 .mmap = shm_mmap,
Adam Litke516dffd2007-03-01 15:46:08 -0800300 .fsync = shm_fsync,
Kirill Korotaev4e982312006-10-02 02:18:22 -0700301 .release = shm_release,
David Howellsed5e5892010-01-15 17:01:32 -0800302#ifndef CONFIG_MMU
303 .get_unmapped_area = shm_get_unmapped_area,
304#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +0200305 .llseek = noop_llseek,
Al Viroc4caa772009-11-30 08:38:43 -0500306};
307
308static const struct file_operations shm_file_operations_huge = {
309 .mmap = shm_mmap,
310 .fsync = shm_fsync,
311 .release = shm_release,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800312 .get_unmapped_area = shm_get_unmapped_area,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200313 .llseek = noop_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314};
315
Al Viroc4caa772009-11-30 08:38:43 -0500316int is_file_shm_hugepages(struct file *file)
317{
318 return file->f_op == &shm_file_operations_huge;
319}
320
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400321static const struct vm_operations_struct shm_vm_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 .open = shm_open, /* callback for a new vm-area open */
323 .close = shm_close, /* callback for when the vm-area is released */
Nick Piggin54cb8822007-07-19 01:46:59 -0700324 .fault = shm_fault,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800325#if defined(CONFIG_NUMA)
326 .set_policy = shm_set_policy,
327 .get_policy = shm_get_policy,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328#endif
329};
330
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700331/**
332 * newseg - Create a new shared memory segment
333 * @ns: namespace
334 * @params: ptr to the structure that contains key, size and shmflg
335 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700336 * Called with shm_ids.rw_mutex held as a writer.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700337 */
338
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700339static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700341 key_t key = params->key;
342 int shmflg = params->flg;
343 size_t size = params->u.size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 int error;
345 struct shmid_kernel *shp;
346 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
347 struct file * file;
348 char name[13];
349 int id;
Mel Gorman5a6fe122009-02-10 14:02:27 +0000350 int acctflag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Kirill Korotaev4e982312006-10-02 02:18:22 -0700352 if (size < SHMMIN || size > ns->shm_ctlmax)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return -EINVAL;
354
Guy Streeterf66d45e2007-01-23 12:20:04 -0600355 if (ns->shm_tot + numpages > ns->shm_ctlall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return -ENOSPC;
357
358 shp = ipc_rcu_alloc(sizeof(*shp));
359 if (!shp)
360 return -ENOMEM;
361
362 shp->shm_perm.key = key;
Andrew Mortonb33291c2006-01-08 01:02:21 -0800363 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 shp->mlock_user = NULL;
365
366 shp->shm_perm.security = NULL;
367 error = security_shm_alloc(shp);
368 if (error) {
369 ipc_rcu_putref(shp);
370 return error;
371 }
372
Eric W. Biederman9d665862007-06-16 10:16:16 -0700373 sprintf (name, "SYSV%08x", key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 if (shmflg & SHM_HUGETLB) {
Mel Gorman5a6fe122009-02-10 14:02:27 +0000375 /* hugetlb_file_setup applies strict accounting */
376 if (shmflg & SHM_NORESERVE)
377 acctflag = VM_NORESERVE;
Hugh Dickins353d5c32009-08-24 16:30:28 +0100378 file = hugetlb_file_setup(name, size, acctflag,
Eric B Munson6bfde052009-09-21 17:03:43 -0700379 &shp->mlock_user, HUGETLB_SHMFS_INODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 } else {
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800381 /*
382 * Do not allow no accounting for OVERCOMMIT_NEVER, even
383 * if it's asked for.
384 */
385 if ((shmflg & SHM_NORESERVE) &&
386 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -0800387 acctflag = VM_NORESERVE;
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800388 file = shmem_file_setup(name, size, acctflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 }
390 error = PTR_ERR(file);
391 if (IS_ERR(file))
392 goto no_file;
393
Pierre Peiffer48dea402008-04-29 01:00:35 -0700394 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700395 if (id < 0) {
396 error = id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 goto no_id;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700400 shp->shm_cprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 shp->shm_lprid = 0;
402 shp->shm_atim = shp->shm_dtim = 0;
403 shp->shm_ctim = get_seconds();
404 shp->shm_segsz = size;
405 shp->shm_nattch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 shp->shm_file = file;
Badari Pulavarty30475cc2007-06-16 10:15:59 -0700407 /*
408 * shmid gets reported as "inode#" in /proc/pid/maps.
409 * proc-ps tools use this. Changing this will break them.
410 */
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700411 file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
Krishnakumar R551110a2005-10-29 18:16:45 -0700412
Kirill Korotaev4e982312006-10-02 02:18:22 -0700413 ns->shm_tot += numpages;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700414 error = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 shm_unlock(shp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700416 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418no_id:
Hugh Dickins2195d282009-09-12 12:21:27 +0100419 if (is_file_hugepages(file) && shp->mlock_user)
Hugh Dickins353d5c32009-08-24 16:30:28 +0100420 user_shm_unlock(size, shp->mlock_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 fput(file);
422no_file:
423 security_shm_free(shp);
424 ipc_rcu_putref(shp);
425 return error;
426}
427
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700428/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700429 * Called with shm_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700430 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700431static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700432{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700433 struct shmid_kernel *shp;
434
435 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
436 return security_shm_associate(shp, shmflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700437}
438
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700439/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700440 * Called with shm_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700441 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700442static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
443 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700444{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700445 struct shmid_kernel *shp;
446
447 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
448 if (shp->shm_segsz < params->u.size)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700449 return -EINVAL;
450
451 return 0;
452}
453
Heiko Carstensd5460c92009-01-14 14:14:27 +0100454SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700456 struct ipc_namespace *ns;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700457 struct ipc_ops shm_ops;
458 struct ipc_params shm_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Kirill Korotaev4e982312006-10-02 02:18:22 -0700460 ns = current->nsproxy->ipc_ns;
461
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700462 shm_ops.getnew = newseg;
463 shm_ops.associate = shm_security;
464 shm_ops.more_checks = shm_more_checks;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700465
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700466 shm_params.key = key;
467 shm_params.flg = shmflg;
468 shm_params.u.size = size;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700469
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700470 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
472
473static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
474{
475 switch(version) {
476 case IPC_64:
477 return copy_to_user(buf, in, sizeof(*in));
478 case IPC_OLD:
479 {
480 struct shmid_ds out;
481
Vasiliy Kulikov3af54c92010-10-30 18:22:49 +0400482 memset(&out, 0, sizeof(out));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
484 out.shm_segsz = in->shm_segsz;
485 out.shm_atime = in->shm_atime;
486 out.shm_dtime = in->shm_dtime;
487 out.shm_ctime = in->shm_ctime;
488 out.shm_cpid = in->shm_cpid;
489 out.shm_lpid = in->shm_lpid;
490 out.shm_nattch = in->shm_nattch;
491
492 return copy_to_user(buf, &out, sizeof(out));
493 }
494 default:
495 return -EINVAL;
496 }
497}
498
Pierre Peiffer016d7132008-04-29 01:00:50 -0700499static inline unsigned long
500copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
502 switch(version) {
503 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -0700504 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 case IPC_OLD:
508 {
509 struct shmid_ds tbuf_old;
510
511 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
512 return -EFAULT;
513
Pierre Peiffer016d7132008-04-29 01:00:50 -0700514 out->shm_perm.uid = tbuf_old.shm_perm.uid;
515 out->shm_perm.gid = tbuf_old.shm_perm.gid;
516 out->shm_perm.mode = tbuf_old.shm_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
518 return 0;
519 }
520 default:
521 return -EINVAL;
522 }
523}
524
525static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
526{
527 switch(version) {
528 case IPC_64:
529 return copy_to_user(buf, in, sizeof(*in));
530 case IPC_OLD:
531 {
532 struct shminfo out;
533
534 if(in->shmmax > INT_MAX)
535 out.shmmax = INT_MAX;
536 else
537 out.shmmax = (int)in->shmmax;
538
539 out.shmmin = in->shmmin;
540 out.shmmni = in->shmmni;
541 out.shmseg = in->shmseg;
542 out.shmall = in->shmall;
543
544 return copy_to_user(buf, &out, sizeof(out));
545 }
546 default:
547 return -EINVAL;
548 }
549}
550
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700551/*
Helge Dellerb7952182010-10-27 15:34:16 -0700552 * Calculate and add used RSS and swap pages of a shm.
553 * Called with shm_ids.rw_mutex held as a reader
554 */
555static void shm_add_rss_swap(struct shmid_kernel *shp,
556 unsigned long *rss_add, unsigned long *swp_add)
557{
558 struct inode *inode;
559
560 inode = shp->shm_file->f_path.dentry->d_inode;
561
562 if (is_file_hugepages(shp->shm_file)) {
563 struct address_space *mapping = inode->i_mapping;
564 struct hstate *h = hstate_file(shp->shm_file);
565 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
566 } else {
567#ifdef CONFIG_SHMEM
568 struct shmem_inode_info *info = SHMEM_I(inode);
569 spin_lock(&info->lock);
570 *rss_add += inode->i_mapping->nrpages;
571 *swp_add += info->swapped;
572 spin_unlock(&info->lock);
573#else
574 *rss_add += inode->i_mapping->nrpages;
575#endif
576 }
577}
578
579/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700580 * Called with shm_ids.rw_mutex held as a reader
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700581 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700582static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
583 unsigned long *swp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700585 int next_id;
586 int total, in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 *rss = 0;
589 *swp = 0;
590
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700591 in_use = shm_ids(ns).in_use;
592
593 for (total = 0, next_id = 0; total < in_use; next_id++) {
Tony Battersbye562aeb2009-04-02 16:58:26 -0700594 struct kern_ipc_perm *ipc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Tony Battersbye562aeb2009-04-02 16:58:26 -0700597 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
598 if (ipc == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 continue;
Tony Battersbye562aeb2009-04-02 16:58:26 -0700600 shp = container_of(ipc, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Helge Dellerb7952182010-10-27 15:34:16 -0700602 shm_add_rss_swap(shp, rss, swp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700603
604 total++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606}
607
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700608/*
609 * This function handles some shmctl commands which require the rw_mutex
610 * to be held in write mode.
611 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
612 */
613static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
614 struct shmid_ds __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700616 struct kern_ipc_perm *ipcp;
Pierre Peiffer016d7132008-04-29 01:00:50 -0700617 struct shmid64_ds shmid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 struct shmid_kernel *shp;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700619 int err;
620
621 if (cmd == IPC_SET) {
Pierre Peiffer016d7132008-04-29 01:00:50 -0700622 if (copy_shmid_from_user(&shmid64, buf, version))
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700623 return -EFAULT;
624 }
625
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700626 ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
627 &shmid64.shm_perm, 0);
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700628 if (IS_ERR(ipcp))
629 return PTR_ERR(ipcp);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700630
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700631 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700632
633 err = security_shm_shmctl(shp, cmd);
634 if (err)
635 goto out_unlock;
636 switch (cmd) {
637 case IPC_RMID:
638 do_shm_rmid(ns, ipcp);
639 goto out_up;
640 case IPC_SET:
Pierre Peiffer8f4a3802008-04-29 01:00:51 -0700641 ipc_update_perm(&shmid64.shm_perm, ipcp);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700642 shp->shm_ctim = get_seconds();
643 break;
644 default:
645 err = -EINVAL;
646 }
647out_unlock:
648 shm_unlock(shp);
649out_up:
650 up_write(&shm_ids(ns).rw_mutex);
651 return err;
652}
653
Heiko Carstensd5460c92009-01-14 14:14:27 +0100654SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700655{
656 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 int err, version;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700658 struct ipc_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 if (cmd < 0 || shmid < 0) {
661 err = -EINVAL;
662 goto out;
663 }
664
665 version = ipc_parse_version(&cmd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700666 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 switch (cmd) { /* replace with proc interface ? */
669 case IPC_INFO:
670 {
671 struct shminfo64 shminfo;
672
673 err = security_shm_shmctl(NULL, cmd);
674 if (err)
675 return err;
676
WANG Conge8148f72009-01-06 14:42:49 -0800677 memset(&shminfo, 0, sizeof(shminfo));
Kirill Korotaev4e982312006-10-02 02:18:22 -0700678 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
679 shminfo.shmmax = ns->shm_ctlmax;
680 shminfo.shmall = ns->shm_ctlall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 shminfo.shmmin = SHMMIN;
683 if(copy_shminfo_to_user (buf, &shminfo, version))
684 return -EFAULT;
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700685
Nadia Derbey3e148c72007-10-18 23:40:54 -0700686 down_read(&shm_ids(ns).rw_mutex);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700687 err = ipc_get_maxid(&shm_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700688 up_read(&shm_ids(ns).rw_mutex);
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if(err<0)
691 err = 0;
692 goto out;
693 }
694 case SHM_INFO:
695 {
696 struct shm_info shm_info;
697
698 err = security_shm_shmctl(NULL, cmd);
699 if (err)
700 return err;
701
WANG Conge8148f72009-01-06 14:42:49 -0800702 memset(&shm_info, 0, sizeof(shm_info));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700703 down_read(&shm_ids(ns).rw_mutex);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700704 shm_info.used_ids = shm_ids(ns).in_use;
705 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
706 shm_info.shm_tot = ns->shm_tot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 shm_info.swap_attempts = 0;
708 shm_info.swap_successes = 0;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700709 err = ipc_get_maxid(&shm_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700710 up_read(&shm_ids(ns).rw_mutex);
WANG Conge8148f72009-01-06 14:42:49 -0800711 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 err = -EFAULT;
713 goto out;
714 }
715
716 err = err < 0 ? 0 : err;
717 goto out;
718 }
719 case SHM_STAT:
720 case IPC_STAT:
721 {
722 struct shmid64_ds tbuf;
723 int result;
Nadia Derbey023a5352007-10-18 23:40:51 -0700724
Nadia Derbey023a5352007-10-18 23:40:51 -0700725 if (cmd == SHM_STAT) {
726 shp = shm_lock(ns, shmid);
727 if (IS_ERR(shp)) {
728 err = PTR_ERR(shp);
729 goto out;
730 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700731 result = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 } else {
Nadia Derbey023a5352007-10-18 23:40:51 -0700733 shp = shm_lock_check(ns, shmid);
734 if (IS_ERR(shp)) {
735 err = PTR_ERR(shp);
736 goto out;
737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 result = 0;
739 }
WANG Conge8148f72009-01-06 14:42:49 -0800740 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700741 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 goto out_unlock;
743 err = security_shm_shmctl(shp, cmd);
744 if (err)
745 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700746 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
748 tbuf.shm_segsz = shp->shm_segsz;
749 tbuf.shm_atime = shp->shm_atim;
750 tbuf.shm_dtime = shp->shm_dtim;
751 tbuf.shm_ctime = shp->shm_ctim;
752 tbuf.shm_cpid = shp->shm_cprid;
753 tbuf.shm_lpid = shp->shm_lprid;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800754 tbuf.shm_nattch = shp->shm_nattch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 shm_unlock(shp);
756 if(copy_shmid_to_user (buf, &tbuf, version))
757 err = -EFAULT;
758 else
759 err = result;
760 goto out;
761 }
762 case SHM_LOCK:
763 case SHM_UNLOCK:
764 {
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -0700765 struct file *uninitialized_var(shm_file);
766
767 lru_add_drain_all(); /* drain pagevecs to lru lists */
768
Nadia Derbey023a5352007-10-18 23:40:51 -0700769 shp = shm_lock_check(ns, shmid);
770 if (IS_ERR(shp)) {
771 err = PTR_ERR(shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 goto out;
773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Al Viroa33e6752008-12-10 03:40:06 -0500775 audit_ipc_obj(&(shp->shm_perm));
Steve Grubb073115d2006-04-02 17:07:33 -0400776
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700777 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
David Howells414c0702008-11-14 10:39:06 +1100778 uid_t euid = current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 err = -EPERM;
David Howells414c0702008-11-14 10:39:06 +1100780 if (euid != shp->shm_perm.uid &&
781 euid != shp->shm_perm.cuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 goto out_unlock;
Jiri Slabyf1eb1332010-03-10 15:23:05 -0800783 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 goto out_unlock;
785 }
786
787 err = security_shm_shmctl(shp, cmd);
788 if (err)
789 goto out_unlock;
790
791 if(cmd==SHM_LOCK) {
David Howells86a264a2008-11-14 10:39:18 +1100792 struct user_struct *user = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 if (!is_file_hugepages(shp->shm_file)) {
794 err = shmem_lock(shp->shm_file, 1, user);
Pavel Emelianov7be77e22007-07-31 00:38:48 -0700795 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
Andrew Mortonb33291c2006-01-08 01:02:21 -0800796 shp->shm_perm.mode |= SHM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 shp->mlock_user = user;
798 }
799 }
800 } else if (!is_file_hugepages(shp->shm_file)) {
801 shmem_lock(shp->shm_file, 0, shp->mlock_user);
Andrew Mortonb33291c2006-01-08 01:02:21 -0800802 shp->shm_perm.mode &= ~SHM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 shp->mlock_user = NULL;
804 }
805 shm_unlock(shp);
806 goto out;
807 }
808 case IPC_RMID:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 case IPC_SET:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700810 err = shmctl_down(ns, shmid, cmd, buf, version);
811 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 default:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700813 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 }
815
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816out_unlock:
817 shm_unlock(shp);
818out:
819 return err;
820}
821
822/*
823 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
824 *
825 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
826 * "raddr" thing points to kernel space, and there has to be a wrapper around
827 * this.
828 */
829long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
830{
831 struct shmid_kernel *shp;
832 unsigned long addr;
833 unsigned long size;
834 struct file * file;
835 int err;
836 unsigned long flags;
837 unsigned long prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 int acc_mode;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800839 unsigned long user_addr;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700840 struct ipc_namespace *ns;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800841 struct shm_file_data *sfd;
842 struct path path;
Al Viroaeb5d722008-09-02 15:28:45 -0400843 fmode_t f_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800845 err = -EINVAL;
846 if (shmid < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 goto out;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800848 else if ((addr = (ulong)shmaddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 if (addr & (SHMLBA-1)) {
850 if (shmflg & SHM_RND)
851 addr &= ~(SHMLBA-1); /* round down */
852 else
853#ifndef __ARCH_FORCE_SHMLBA
854 if (addr & ~PAGE_MASK)
855#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800856 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 }
858 flags = MAP_SHARED | MAP_FIXED;
859 } else {
860 if ((shmflg & SHM_REMAP))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800861 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 flags = MAP_SHARED;
864 }
865
866 if (shmflg & SHM_RDONLY) {
867 prot = PROT_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 acc_mode = S_IRUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800869 f_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 } else {
871 prot = PROT_READ | PROT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 acc_mode = S_IRUGO | S_IWUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800873 f_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 }
875 if (shmflg & SHM_EXEC) {
876 prot |= PROT_EXEC;
877 acc_mode |= S_IXUGO;
878 }
879
880 /*
881 * We cannot rely on the fs check since SYSV IPC does have an
882 * additional creator id...
883 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700884 ns = current->nsproxy->ipc_ns;
Nadia Derbey023a5352007-10-18 23:40:51 -0700885 shp = shm_lock_check(ns, shmid);
886 if (IS_ERR(shp)) {
887 err = PTR_ERR(shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 goto out;
Nadia Derbey023a5352007-10-18 23:40:51 -0700889 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800890
891 err = -EACCES;
Serge E. Hallynb0e77592011-03-23 16:43:24 -0700892 if (ipcperms(ns, &shp->shm_perm, acc_mode))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800893 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 err = security_shm_shmat(shp, shmaddr, shmflg);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800896 if (err)
897 goto out_unlock;
898
Al Viro2c48b9c2009-08-09 00:52:35 +0400899 path = shp->shm_file->f_path;
900 path_get(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 shp->shm_nattch++;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800902 size = i_size_read(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 shm_unlock(shp);
904
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800905 err = -ENOMEM;
906 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
907 if (!sfd)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700908 goto out_put_dentry;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800909
Al Viro2c48b9c2009-08-09 00:52:35 +0400910 file = alloc_file(&path, f_mode,
911 is_file_hugepages(shp->shm_file) ?
Al Viroc4caa772009-11-30 08:38:43 -0500912 &shm_file_operations_huge :
913 &shm_file_operations);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800914 if (!file)
915 goto out_free;
916
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800917 file->private_data = sfd;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800918 file->f_mapping = shp->shm_file->f_mapping;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700919 sfd->id = shp->shm_perm.id;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800920 sfd->ns = get_ipc_ns(ns);
921 sfd->file = shp->shm_file;
922 sfd->vm_ops = NULL;
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 down_write(&current->mm->mmap_sem);
925 if (addr && !(shmflg & SHM_REMAP)) {
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800926 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (find_vma_intersection(current->mm, addr, addr + size))
928 goto invalid;
929 /*
930 * If shm segment goes below stack, make sure there is some
931 * space left for the stack to grow (at least 4 pages).
932 */
933 if (addr < current->mm->start_stack &&
934 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
935 goto invalid;
936 }
937
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800938 user_addr = do_mmap (file, addr, size, prot, flags, 0);
939 *raddr = user_addr;
940 err = 0;
941 if (IS_ERR_VALUE(user_addr))
942 err = (long)user_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943invalid:
944 up_write(&current->mm->mmap_sem);
945
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800946 fput(file);
947
948out_nattch:
Nadia Derbey3e148c72007-10-18 23:40:54 -0700949 down_write(&shm_ids(ns).rw_mutex);
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700950 shp = shm_lock(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700951 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 shp->shm_nattch--;
953 if(shp->shm_nattch == 0 &&
Andrew Mortonb33291c2006-01-08 01:02:21 -0800954 shp->shm_perm.mode & SHM_DEST)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700955 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 else
957 shm_unlock(shp);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700958 up_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960out:
961 return err;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800962
963out_unlock:
964 shm_unlock(shp);
965 goto out;
966
967out_free:
968 kfree(sfd);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700969out_put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +0400970 path_put(&path);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800971 goto out_nattch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972}
973
Heiko Carstensd5460c92009-01-14 14:14:27 +0100974SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -0700975{
976 unsigned long ret;
977 long err;
978
979 err = do_shmat(shmid, shmaddr, shmflg, &ret);
980 if (err)
981 return err;
982 force_successful_syscall_return();
983 return (long)ret;
984}
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986/*
987 * detach and kill segment if marked destroyed.
988 * The work is done in shm_close.
989 */
Heiko Carstensd5460c92009-01-14 14:14:27 +0100990SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
992 struct mm_struct *mm = current->mm;
Mike Frysinger586c7e62009-06-09 16:26:23 -0700993 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 unsigned long addr = (unsigned long)shmaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 int retval = -EINVAL;
Mike Frysinger586c7e62009-06-09 16:26:23 -0700996#ifdef CONFIG_MMU
997 loff_t size = 0;
998 struct vm_area_struct *next;
999#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Hugh Dickinsdf1e2fb2006-03-24 03:18:06 -08001001 if (addr & ~PAGE_MASK)
1002 return retval;
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 down_write(&mm->mmap_sem);
1005
1006 /*
1007 * This function tries to be smart and unmap shm segments that
1008 * were modified by partial mlock or munmap calls:
1009 * - It first determines the size of the shm segment that should be
1010 * unmapped: It searches for a vma that is backed by shm and that
1011 * started at address shmaddr. It records it's size and then unmaps
1012 * it.
1013 * - Then it unmaps all shm vmas that started at shmaddr and that
1014 * are within the initially determined size.
1015 * Errors from do_munmap are ignored: the function only fails if
1016 * it's called with invalid parameters or if it's called to unmap
1017 * a part of a vma. Both calls in this function are for full vmas,
1018 * the parameters are directly copied from the vma itself and always
1019 * valid - therefore do_munmap cannot fail. (famous last words?)
1020 */
1021 /*
1022 * If it had been mremap()'d, the starting address would not
1023 * match the usual checks anyway. So assume all vma's are
1024 * above the starting address given.
1025 */
1026 vma = find_vma(mm, addr);
1027
David Howells8feae132009-01-08 12:04:47 +00001028#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 while (vma) {
1030 next = vma->vm_next;
1031
1032 /*
1033 * Check if the starting address would match, i.e. it's
1034 * a fragment created by mprotect() and/or munmap(), or it
1035 * otherwise it starts at this address with no hassles.
1036 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001037 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1039
1040
Josef Sipek6d630792006-12-08 02:37:11 -08001041 size = vma->vm_file->f_path.dentry->d_inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1043 /*
1044 * We discovered the size of the shm segment, so
1045 * break out of here and fall through to the next
1046 * loop that uses the size information to stop
1047 * searching for matching vma's.
1048 */
1049 retval = 0;
1050 vma = next;
1051 break;
1052 }
1053 vma = next;
1054 }
1055
1056 /*
1057 * We need look no further than the maximum address a fragment
1058 * could possibly have landed at. Also cast things to loff_t to
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001059 * prevent overflows and make comparisons vs. equal-width types.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 */
KAMEZAWA Hiroyuki8e367092006-02-10 01:51:12 -08001061 size = PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1063 next = vma->vm_next;
1064
1065 /* finding a matching vma now does not alter retval */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001066 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1068
1069 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1070 vma = next;
1071 }
1072
David Howells8feae132009-01-08 12:04:47 +00001073#else /* CONFIG_MMU */
1074 /* under NOMMU conditions, the exact address to be destroyed must be
1075 * given */
1076 retval = -EINVAL;
1077 if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1078 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1079 retval = 0;
1080 }
1081
1082#endif
1083
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 up_write(&mm->mmap_sem);
1085 return retval;
1086}
1087
1088#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07001089static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
Mike Waychison19b49462005-09-06 15:17:10 -07001091 struct shmid_kernel *shp = it;
Helge Dellerb7952182010-10-27 15:34:16 -07001092 unsigned long rss = 0, swp = 0;
1093
1094 shm_add_rss_swap(shp, &rss, &swp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Paul Menage6c826812008-06-12 15:21:49 -07001096#if BITS_PER_LONG <= 32
1097#define SIZE_SPEC "%10lu"
1098#else
1099#define SIZE_SPEC "%21lu"
1100#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Paul Menage6c826812008-06-12 15:21:49 -07001102 return seq_printf(s,
1103 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
Helge Dellerb7952182010-10-27 15:34:16 -07001104 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1105 SIZE_SPEC " " SIZE_SPEC "\n",
Mike Waychison19b49462005-09-06 15:17:10 -07001106 shp->shm_perm.key,
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001107 shp->shm_perm.id,
Andrew Mortonb33291c2006-01-08 01:02:21 -08001108 shp->shm_perm.mode,
Mike Waychison19b49462005-09-06 15:17:10 -07001109 shp->shm_segsz,
1110 shp->shm_cprid,
1111 shp->shm_lprid,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001112 shp->shm_nattch,
Mike Waychison19b49462005-09-06 15:17:10 -07001113 shp->shm_perm.uid,
1114 shp->shm_perm.gid,
1115 shp->shm_perm.cuid,
1116 shp->shm_perm.cgid,
1117 shp->shm_atim,
1118 shp->shm_dtim,
Helge Dellerb7952182010-10-27 15:34:16 -07001119 shp->shm_ctim,
1120 rss * PAGE_SIZE,
1121 swp * PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122}
1123#endif