blob: 05d51d2a792c116630f8d4d1cd28bbdcf754c81f [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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057static struct vm_operations_struct shm_vm_ops;
58
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);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700103}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800104#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
106void __init shm_init (void)
107{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800108 shm_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700109 ipc_init_proc_interface("sysvipc/shm",
110 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
Kirill Korotaev4e982312006-10-02 02:18:22 -0700111 IPC_SHM_IDS, sysvipc_shm_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
Nadia Derbey3e148c72007-10-18 23:40:54 -0700114/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700115 * shm_lock_(check_) routines are called in the paths where the rw_mutex
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700116 * is not necessarily held.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700117 */
Nadia Derbey023a5352007-10-18 23:40:51 -0700118static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700120 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
121
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800122 if (IS_ERR(ipcp))
123 return (struct shmid_kernel *)ipcp;
124
Nadia Derbey03f02c72007-10-18 23:40:51 -0700125 return container_of(ipcp, struct shmid_kernel, shm_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700126}
127
128static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
129 int id)
130{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700131 struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
132
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800133 if (IS_ERR(ipcp))
134 return (struct shmid_kernel *)ipcp;
135
Nadia Derbey03f02c72007-10-18 23:40:51 -0700136 return container_of(ipcp, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700139static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700141 ipc_rmid(&shm_ids(ns), &s->shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800145/* This is called by fork, once for every shm attach. */
146static void shm_open(struct vm_area_struct *vma)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700147{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800148 struct file *file = vma->vm_file;
149 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 struct shmid_kernel *shp;
151
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800152 shp = shm_lock(sfd->ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700153 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 shp->shm_atim = get_seconds();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700155 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 shp->shm_nattch++;
157 shm_unlock(shp);
158}
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160/*
161 * shm_destroy - free the struct shmid_kernel
162 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700163 * @ns: namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 * @shp: struct to free
165 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700166 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 * but returns with shp unlocked and freed.
168 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700169static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700171 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700172 shm_rmid(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 shm_unlock(shp);
174 if (!is_file_hugepages(shp->shm_file))
175 shmem_lock(shp->shm_file, 0, shp->mlock_user);
176 else
Josef Sipek6d630792006-12-08 02:37:11 -0800177 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 shp->mlock_user);
179 fput (shp->shm_file);
180 security_shm_free(shp);
181 ipc_rcu_putref(shp);
182}
183
184/*
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800185 * remove the attach descriptor vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * free memory for segment if it is marked destroyed.
187 * The descriptor has already been removed from the current->mm->mmap list
188 * and will later be kfree()d.
189 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800190static void shm_close(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -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;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800195 struct ipc_namespace *ns = sfd->ns;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700196
Nadia Derbey3e148c72007-10-18 23:40:54 -0700197 down_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 /* remove from the list of attaches of the shm segment */
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700199 shp = shm_lock(ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700200 BUG_ON(IS_ERR(shp));
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700201 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 shp->shm_dtim = get_seconds();
203 shp->shm_nattch--;
204 if(shp->shm_nattch == 0 &&
Andrew Mortonb33291c2006-01-08 01:02:21 -0800205 shp->shm_perm.mode & SHM_DEST)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700206 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 else
208 shm_unlock(shp);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700209 up_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
Nick Piggind0217ac2007-07-19 01:47:03 -0700212static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800213{
214 struct file *file = vma->vm_file;
215 struct shm_file_data *sfd = shm_file_data(file);
216
Nick Piggind0217ac2007-07-19 01:47:03 -0700217 return sfd->vm_ops->fault(vma, vmf);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800218}
219
220#ifdef CONFIG_NUMA
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700221static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800222{
223 struct file *file = vma->vm_file;
224 struct shm_file_data *sfd = shm_file_data(file);
225 int err = 0;
226 if (sfd->vm_ops->set_policy)
227 err = sfd->vm_ops->set_policy(vma, new);
228 return err;
229}
230
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700231static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
232 unsigned long addr)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800233{
234 struct file *file = vma->vm_file;
235 struct shm_file_data *sfd = shm_file_data(file);
236 struct mempolicy *pol = NULL;
237
238 if (sfd->vm_ops->get_policy)
239 pol = sfd->vm_ops->get_policy(vma, addr);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700240 else if (vma->vm_policy)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800241 pol = vma->vm_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700242
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800243 return pol;
244}
245#endif
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247static int shm_mmap(struct file * file, struct vm_area_struct * vma)
248{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800249 struct shm_file_data *sfd = shm_file_data(file);
David Howellsb0e15192006-01-06 00:11:42 -0800250 int ret;
251
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800252 ret = sfd->file->f_op->mmap(sfd->file, vma);
253 if (ret != 0)
254 return ret;
255 sfd->vm_ops = vma->vm_ops;
David Howells2e92a3b2007-07-31 00:37:24 -0700256#ifdef CONFIG_MMU
Nick Piggin54cb8822007-07-19 01:46:59 -0700257 BUG_ON(!sfd->vm_ops->fault);
David Howells2e92a3b2007-07-31 00:37:24 -0700258#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800259 vma->vm_ops = &shm_vm_ops;
260 shm_open(vma);
David Howellsb0e15192006-01-06 00:11:42 -0800261
262 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264
Kirill Korotaev4e982312006-10-02 02:18:22 -0700265static int shm_release(struct inode *ino, struct file *file)
266{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800267 struct shm_file_data *sfd = shm_file_data(file);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700268
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800269 put_ipc_ns(sfd->ns);
270 shm_file_data(file) = NULL;
271 kfree(sfd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700272 return 0;
273}
274
Adam Litke516dffd2007-03-01 15:46:08 -0800275static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
276{
277 int (*fsync) (struct file *, struct dentry *, int datasync);
278 struct shm_file_data *sfd = shm_file_data(file);
279 int ret = -EINVAL;
280
281 fsync = sfd->file->f_op->fsync;
282 if (fsync)
283 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
284 return ret;
285}
286
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800287static unsigned long shm_get_unmapped_area(struct file *file,
288 unsigned long addr, unsigned long len, unsigned long pgoff,
289 unsigned long flags)
290{
291 struct shm_file_data *sfd = shm_file_data(file);
Adam Litke516dffd2007-03-01 15:46:08 -0800292 return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800293}
Adam Litke516dffd2007-03-01 15:46:08 -0800294
295int is_file_shm_hugepages(struct file *file)
296{
297 int ret = 0;
298
299 if (file->f_op == &shm_file_operations) {
300 struct shm_file_data *sfd;
301 sfd = shm_file_data(file);
302 ret = is_file_hugepages(sfd->file);
303 }
304 return ret;
305}
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800306
Arjan van de Ven9a321442007-02-12 00:55:35 -0800307static const struct file_operations shm_file_operations = {
Kirill Korotaev4e982312006-10-02 02:18:22 -0700308 .mmap = shm_mmap,
Adam Litke516dffd2007-03-01 15:46:08 -0800309 .fsync = shm_fsync,
Kirill Korotaev4e982312006-10-02 02:18:22 -0700310 .release = shm_release,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800311 .get_unmapped_area = shm_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312};
313
314static struct vm_operations_struct shm_vm_ops = {
315 .open = shm_open, /* callback for a new vm-area open */
316 .close = shm_close, /* callback for when the vm-area is released */
Nick Piggin54cb8822007-07-19 01:46:59 -0700317 .fault = shm_fault,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800318#if defined(CONFIG_NUMA)
319 .set_policy = shm_set_policy,
320 .get_policy = shm_get_policy,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321#endif
322};
323
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700324/**
325 * newseg - Create a new shared memory segment
326 * @ns: namespace
327 * @params: ptr to the structure that contains key, size and shmflg
328 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700329 * Called with shm_ids.rw_mutex held as a writer.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700330 */
331
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700332static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700334 key_t key = params->key;
335 int shmflg = params->flg;
336 size_t size = params->u.size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 int error;
338 struct shmid_kernel *shp;
339 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
340 struct file * file;
341 char name[13];
342 int id;
Mel Gorman5a6fe122009-02-10 14:02:27 +0000343 int acctflag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Kirill Korotaev4e982312006-10-02 02:18:22 -0700345 if (size < SHMMIN || size > ns->shm_ctlmax)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 return -EINVAL;
347
Guy Streeterf66d45e2007-01-23 12:20:04 -0600348 if (ns->shm_tot + numpages > ns->shm_ctlall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 return -ENOSPC;
350
351 shp = ipc_rcu_alloc(sizeof(*shp));
352 if (!shp)
353 return -ENOMEM;
354
355 shp->shm_perm.key = key;
Andrew Mortonb33291c2006-01-08 01:02:21 -0800356 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 shp->mlock_user = NULL;
358
359 shp->shm_perm.security = NULL;
360 error = security_shm_alloc(shp);
361 if (error) {
362 ipc_rcu_putref(shp);
363 return error;
364 }
365
Eric W. Biederman9d665862007-06-16 10:16:16 -0700366 sprintf (name, "SYSV%08x", key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 if (shmflg & SHM_HUGETLB) {
Mel Gorman5a6fe122009-02-10 14:02:27 +0000368 /* hugetlb_file_setup applies strict accounting */
369 if (shmflg & SHM_NORESERVE)
370 acctflag = VM_NORESERVE;
371 file = hugetlb_file_setup(name, size, acctflag);
David Howells86a264a2008-11-14 10:39:18 +1100372 shp->mlock_user = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 } else {
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800374 /*
375 * Do not allow no accounting for OVERCOMMIT_NEVER, even
376 * if it's asked for.
377 */
378 if ((shmflg & SHM_NORESERVE) &&
379 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -0800380 acctflag = VM_NORESERVE;
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800381 file = shmem_file_setup(name, size, acctflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
383 error = PTR_ERR(file);
384 if (IS_ERR(file))
385 goto no_file;
386
Pierre Peiffer48dea402008-04-29 01:00:35 -0700387 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700388 if (id < 0) {
389 error = id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 goto no_id;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700393 shp->shm_cprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 shp->shm_lprid = 0;
395 shp->shm_atim = shp->shm_dtim = 0;
396 shp->shm_ctim = get_seconds();
397 shp->shm_segsz = size;
398 shp->shm_nattch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 shp->shm_file = file;
Badari Pulavarty30475cc2007-06-16 10:15:59 -0700400 /*
401 * shmid gets reported as "inode#" in /proc/pid/maps.
402 * proc-ps tools use this. Changing this will break them.
403 */
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700404 file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
Krishnakumar R551110a2005-10-29 18:16:45 -0700405
Kirill Korotaev4e982312006-10-02 02:18:22 -0700406 ns->shm_tot += numpages;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700407 error = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 shm_unlock(shp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700409 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411no_id:
412 fput(file);
413no_file:
414 security_shm_free(shp);
415 ipc_rcu_putref(shp);
416 return error;
417}
418
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700419/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700420 * Called with shm_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700421 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700422static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700423{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700424 struct shmid_kernel *shp;
425
426 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
427 return security_shm_associate(shp, shmflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700428}
429
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700430/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700431 * Called with shm_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700432 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700433static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
434 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700435{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700436 struct shmid_kernel *shp;
437
438 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
439 if (shp->shm_segsz < params->u.size)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700440 return -EINVAL;
441
442 return 0;
443}
444
Heiko Carstensd5460c92009-01-14 14:14:27 +0100445SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700447 struct ipc_namespace *ns;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700448 struct ipc_ops shm_ops;
449 struct ipc_params shm_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Kirill Korotaev4e982312006-10-02 02:18:22 -0700451 ns = current->nsproxy->ipc_ns;
452
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700453 shm_ops.getnew = newseg;
454 shm_ops.associate = shm_security;
455 shm_ops.more_checks = shm_more_checks;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700456
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700457 shm_params.key = key;
458 shm_params.flg = shmflg;
459 shm_params.u.size = size;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700460
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700461 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
464static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
465{
466 switch(version) {
467 case IPC_64:
468 return copy_to_user(buf, in, sizeof(*in));
469 case IPC_OLD:
470 {
471 struct shmid_ds out;
472
473 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
474 out.shm_segsz = in->shm_segsz;
475 out.shm_atime = in->shm_atime;
476 out.shm_dtime = in->shm_dtime;
477 out.shm_ctime = in->shm_ctime;
478 out.shm_cpid = in->shm_cpid;
479 out.shm_lpid = in->shm_lpid;
480 out.shm_nattch = in->shm_nattch;
481
482 return copy_to_user(buf, &out, sizeof(out));
483 }
484 default:
485 return -EINVAL;
486 }
487}
488
Pierre Peiffer016d7132008-04-29 01:00:50 -0700489static inline unsigned long
490copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 switch(version) {
493 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -0700494 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 case IPC_OLD:
498 {
499 struct shmid_ds tbuf_old;
500
501 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
502 return -EFAULT;
503
Pierre Peiffer016d7132008-04-29 01:00:50 -0700504 out->shm_perm.uid = tbuf_old.shm_perm.uid;
505 out->shm_perm.gid = tbuf_old.shm_perm.gid;
506 out->shm_perm.mode = tbuf_old.shm_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
508 return 0;
509 }
510 default:
511 return -EINVAL;
512 }
513}
514
515static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
516{
517 switch(version) {
518 case IPC_64:
519 return copy_to_user(buf, in, sizeof(*in));
520 case IPC_OLD:
521 {
522 struct shminfo out;
523
524 if(in->shmmax > INT_MAX)
525 out.shmmax = INT_MAX;
526 else
527 out.shmmax = (int)in->shmmax;
528
529 out.shmmin = in->shmmin;
530 out.shmmni = in->shmmni;
531 out.shmseg = in->shmseg;
532 out.shmall = in->shmall;
533
534 return copy_to_user(buf, &out, sizeof(out));
535 }
536 default:
537 return -EINVAL;
538 }
539}
540
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700541/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700542 * Called with shm_ids.rw_mutex held as a reader
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700543 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700544static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
545 unsigned long *swp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700547 int next_id;
548 int total, in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 *rss = 0;
551 *swp = 0;
552
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700553 in_use = shm_ids(ns).in_use;
554
555 for (total = 0, next_id = 0; total < in_use; next_id++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 struct shmid_kernel *shp;
557 struct inode *inode;
558
Nadia Derbey637c3662007-10-18 23:40:50 -0700559 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700560 if (shp == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 continue;
562
Josef Sipek6d630792006-12-08 02:37:11 -0800563 inode = shp->shm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565 if (is_file_hugepages(shp->shm_file)) {
566 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700567 struct hstate *h = hstate_file(shp->shm_file);
568 *rss += pages_per_huge_page(h) * mapping->nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 } else {
Tony Battersbya68e61e2009-02-04 15:12:04 -0800570#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 struct shmem_inode_info *info = SHMEM_I(inode);
572 spin_lock(&info->lock);
573 *rss += inode->i_mapping->nrpages;
574 *swp += info->swapped;
575 spin_unlock(&info->lock);
Tony Battersbya68e61e2009-02-04 15:12:04 -0800576#else
577 *rss += inode->i_mapping->nrpages;
578#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700580
581 total++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
583}
584
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700585/*
586 * This function handles some shmctl commands which require the rw_mutex
587 * to be held in write mode.
588 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
589 */
590static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
591 struct shmid_ds __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700593 struct kern_ipc_perm *ipcp;
Pierre Peiffer016d7132008-04-29 01:00:50 -0700594 struct shmid64_ds shmid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 struct shmid_kernel *shp;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700596 int err;
597
598 if (cmd == IPC_SET) {
Pierre Peiffer016d7132008-04-29 01:00:50 -0700599 if (copy_shmid_from_user(&shmid64, buf, version))
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700600 return -EFAULT;
601 }
602
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700603 ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
604 if (IS_ERR(ipcp))
605 return PTR_ERR(ipcp);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700606
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700607 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700608
609 err = security_shm_shmctl(shp, cmd);
610 if (err)
611 goto out_unlock;
612 switch (cmd) {
613 case IPC_RMID:
614 do_shm_rmid(ns, ipcp);
615 goto out_up;
616 case IPC_SET:
Pierre Peiffer8f4a3802008-04-29 01:00:51 -0700617 ipc_update_perm(&shmid64.shm_perm, ipcp);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700618 shp->shm_ctim = get_seconds();
619 break;
620 default:
621 err = -EINVAL;
622 }
623out_unlock:
624 shm_unlock(shp);
625out_up:
626 up_write(&shm_ids(ns).rw_mutex);
627 return err;
628}
629
Heiko Carstensd5460c92009-01-14 14:14:27 +0100630SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700631{
632 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 int err, version;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700634 struct ipc_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636 if (cmd < 0 || shmid < 0) {
637 err = -EINVAL;
638 goto out;
639 }
640
641 version = ipc_parse_version(&cmd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700642 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644 switch (cmd) { /* replace with proc interface ? */
645 case IPC_INFO:
646 {
647 struct shminfo64 shminfo;
648
649 err = security_shm_shmctl(NULL, cmd);
650 if (err)
651 return err;
652
WANG Conge8148f72009-01-06 14:42:49 -0800653 memset(&shminfo, 0, sizeof(shminfo));
Kirill Korotaev4e982312006-10-02 02:18:22 -0700654 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
655 shminfo.shmmax = ns->shm_ctlmax;
656 shminfo.shmall = ns->shm_ctlall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658 shminfo.shmmin = SHMMIN;
659 if(copy_shminfo_to_user (buf, &shminfo, version))
660 return -EFAULT;
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700661
Nadia Derbey3e148c72007-10-18 23:40:54 -0700662 down_read(&shm_ids(ns).rw_mutex);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700663 err = ipc_get_maxid(&shm_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700664 up_read(&shm_ids(ns).rw_mutex);
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 if(err<0)
667 err = 0;
668 goto out;
669 }
670 case SHM_INFO:
671 {
672 struct shm_info shm_info;
673
674 err = security_shm_shmctl(NULL, cmd);
675 if (err)
676 return err;
677
WANG Conge8148f72009-01-06 14:42:49 -0800678 memset(&shm_info, 0, sizeof(shm_info));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700679 down_read(&shm_ids(ns).rw_mutex);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700680 shm_info.used_ids = shm_ids(ns).in_use;
681 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
682 shm_info.shm_tot = ns->shm_tot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 shm_info.swap_attempts = 0;
684 shm_info.swap_successes = 0;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700685 err = ipc_get_maxid(&shm_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700686 up_read(&shm_ids(ns).rw_mutex);
WANG Conge8148f72009-01-06 14:42:49 -0800687 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 err = -EFAULT;
689 goto out;
690 }
691
692 err = err < 0 ? 0 : err;
693 goto out;
694 }
695 case SHM_STAT:
696 case IPC_STAT:
697 {
698 struct shmid64_ds tbuf;
699 int result;
Nadia Derbey023a5352007-10-18 23:40:51 -0700700
Nadia Derbey023a5352007-10-18 23:40:51 -0700701 if (cmd == SHM_STAT) {
702 shp = shm_lock(ns, shmid);
703 if (IS_ERR(shp)) {
704 err = PTR_ERR(shp);
705 goto out;
706 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700707 result = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 } else {
Nadia Derbey023a5352007-10-18 23:40:51 -0700709 shp = shm_lock_check(ns, shmid);
710 if (IS_ERR(shp)) {
711 err = PTR_ERR(shp);
712 goto out;
713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 result = 0;
715 }
WANG Conge8148f72009-01-06 14:42:49 -0800716 err = -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (ipcperms (&shp->shm_perm, S_IRUGO))
718 goto out_unlock;
719 err = security_shm_shmctl(shp, cmd);
720 if (err)
721 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700722 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
724 tbuf.shm_segsz = shp->shm_segsz;
725 tbuf.shm_atime = shp->shm_atim;
726 tbuf.shm_dtime = shp->shm_dtim;
727 tbuf.shm_ctime = shp->shm_ctim;
728 tbuf.shm_cpid = shp->shm_cprid;
729 tbuf.shm_lpid = shp->shm_lprid;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800730 tbuf.shm_nattch = shp->shm_nattch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 shm_unlock(shp);
732 if(copy_shmid_to_user (buf, &tbuf, version))
733 err = -EFAULT;
734 else
735 err = result;
736 goto out;
737 }
738 case SHM_LOCK:
739 case SHM_UNLOCK:
740 {
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -0700741 struct file *uninitialized_var(shm_file);
742
743 lru_add_drain_all(); /* drain pagevecs to lru lists */
744
Nadia Derbey023a5352007-10-18 23:40:51 -0700745 shp = shm_lock_check(ns, shmid);
746 if (IS_ERR(shp)) {
747 err = PTR_ERR(shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 goto out;
749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Al Viroa33e6752008-12-10 03:40:06 -0500751 audit_ipc_obj(&(shp->shm_perm));
Steve Grubb073115d2006-04-02 17:07:33 -0400752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (!capable(CAP_IPC_LOCK)) {
David Howells414c0702008-11-14 10:39:06 +1100754 uid_t euid = current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 err = -EPERM;
David Howells414c0702008-11-14 10:39:06 +1100756 if (euid != shp->shm_perm.uid &&
757 euid != shp->shm_perm.cuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 goto out_unlock;
759 if (cmd == SHM_LOCK &&
760 !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
761 goto out_unlock;
762 }
763
764 err = security_shm_shmctl(shp, cmd);
765 if (err)
766 goto out_unlock;
767
768 if(cmd==SHM_LOCK) {
David Howells86a264a2008-11-14 10:39:18 +1100769 struct user_struct *user = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 if (!is_file_hugepages(shp->shm_file)) {
771 err = shmem_lock(shp->shm_file, 1, user);
Pavel Emelianov7be77e22007-07-31 00:38:48 -0700772 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
Andrew Mortonb33291c2006-01-08 01:02:21 -0800773 shp->shm_perm.mode |= SHM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 shp->mlock_user = user;
775 }
776 }
777 } else if (!is_file_hugepages(shp->shm_file)) {
778 shmem_lock(shp->shm_file, 0, shp->mlock_user);
Andrew Mortonb33291c2006-01-08 01:02:21 -0800779 shp->shm_perm.mode &= ~SHM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 shp->mlock_user = NULL;
781 }
782 shm_unlock(shp);
783 goto out;
784 }
785 case IPC_RMID:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 case IPC_SET:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700787 err = shmctl_down(ns, shmid, cmd, buf, version);
788 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 default:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700790 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793out_unlock:
794 shm_unlock(shp);
795out:
796 return err;
797}
798
799/*
800 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
801 *
802 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
803 * "raddr" thing points to kernel space, and there has to be a wrapper around
804 * this.
805 */
806long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
807{
808 struct shmid_kernel *shp;
809 unsigned long addr;
810 unsigned long size;
811 struct file * file;
812 int err;
813 unsigned long flags;
814 unsigned long prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 int acc_mode;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800816 unsigned long user_addr;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700817 struct ipc_namespace *ns;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800818 struct shm_file_data *sfd;
819 struct path path;
Al Viroaeb5d722008-09-02 15:28:45 -0400820 fmode_t f_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800822 err = -EINVAL;
823 if (shmid < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 goto out;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800825 else if ((addr = (ulong)shmaddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (addr & (SHMLBA-1)) {
827 if (shmflg & SHM_RND)
828 addr &= ~(SHMLBA-1); /* round down */
829 else
830#ifndef __ARCH_FORCE_SHMLBA
831 if (addr & ~PAGE_MASK)
832#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800833 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
835 flags = MAP_SHARED | MAP_FIXED;
836 } else {
837 if ((shmflg & SHM_REMAP))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800838 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
840 flags = MAP_SHARED;
841 }
842
843 if (shmflg & SHM_RDONLY) {
844 prot = PROT_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 acc_mode = S_IRUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800846 f_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 } else {
848 prot = PROT_READ | PROT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 acc_mode = S_IRUGO | S_IWUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800850 f_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 }
852 if (shmflg & SHM_EXEC) {
853 prot |= PROT_EXEC;
854 acc_mode |= S_IXUGO;
855 }
856
857 /*
858 * We cannot rely on the fs check since SYSV IPC does have an
859 * additional creator id...
860 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700861 ns = current->nsproxy->ipc_ns;
Nadia Derbey023a5352007-10-18 23:40:51 -0700862 shp = shm_lock_check(ns, shmid);
863 if (IS_ERR(shp)) {
864 err = PTR_ERR(shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 goto out;
Nadia Derbey023a5352007-10-18 23:40:51 -0700866 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800867
868 err = -EACCES;
869 if (ipcperms(&shp->shm_perm, acc_mode))
870 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
872 err = security_shm_shmat(shp, shmaddr, shmflg);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800873 if (err)
874 goto out_unlock;
875
876 path.dentry = dget(shp->shm_file->f_path.dentry);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700877 path.mnt = shp->shm_file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 shp->shm_nattch++;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800879 size = i_size_read(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 shm_unlock(shp);
881
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800882 err = -ENOMEM;
883 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
884 if (!sfd)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700885 goto out_put_dentry;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800886
Dave Hansence8d2cd2007-10-16 23:31:13 -0700887 file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800888 if (!file)
889 goto out_free;
890
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800891 file->private_data = sfd;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800892 file->f_mapping = shp->shm_file->f_mapping;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700893 sfd->id = shp->shm_perm.id;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800894 sfd->ns = get_ipc_ns(ns);
895 sfd->file = shp->shm_file;
896 sfd->vm_ops = NULL;
897
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 down_write(&current->mm->mmap_sem);
899 if (addr && !(shmflg & SHM_REMAP)) {
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800900 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 if (find_vma_intersection(current->mm, addr, addr + size))
902 goto invalid;
903 /*
904 * If shm segment goes below stack, make sure there is some
905 * space left for the stack to grow (at least 4 pages).
906 */
907 if (addr < current->mm->start_stack &&
908 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
909 goto invalid;
910 }
911
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800912 user_addr = do_mmap (file, addr, size, prot, flags, 0);
913 *raddr = user_addr;
914 err = 0;
915 if (IS_ERR_VALUE(user_addr))
916 err = (long)user_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917invalid:
918 up_write(&current->mm->mmap_sem);
919
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800920 fput(file);
921
922out_nattch:
Nadia Derbey3e148c72007-10-18 23:40:54 -0700923 down_write(&shm_ids(ns).rw_mutex);
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700924 shp = shm_lock(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700925 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 shp->shm_nattch--;
927 if(shp->shm_nattch == 0 &&
Andrew Mortonb33291c2006-01-08 01:02:21 -0800928 shp->shm_perm.mode & SHM_DEST)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700929 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 else
931 shm_unlock(shp);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700932 up_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934out:
935 return err;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800936
937out_unlock:
938 shm_unlock(shp);
939 goto out;
940
941out_free:
942 kfree(sfd);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700943out_put_dentry:
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800944 dput(path.dentry);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800945 goto out_nattch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
Heiko Carstensd5460c92009-01-14 14:14:27 +0100948SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
Stephen Rothwell7d87e14c2005-05-01 08:59:12 -0700949{
950 unsigned long ret;
951 long err;
952
953 err = do_shmat(shmid, shmaddr, shmflg, &ret);
954 if (err)
955 return err;
956 force_successful_syscall_return();
957 return (long)ret;
958}
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960/*
961 * detach and kill segment if marked destroyed.
962 * The work is done in shm_close.
963 */
Heiko Carstensd5460c92009-01-14 14:14:27 +0100964SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
966 struct mm_struct *mm = current->mm;
967 struct vm_area_struct *vma, *next;
968 unsigned long addr = (unsigned long)shmaddr;
969 loff_t size = 0;
970 int retval = -EINVAL;
971
Hugh Dickinsdf1e2fb2006-03-24 03:18:06 -0800972 if (addr & ~PAGE_MASK)
973 return retval;
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 down_write(&mm->mmap_sem);
976
977 /*
978 * This function tries to be smart and unmap shm segments that
979 * were modified by partial mlock or munmap calls:
980 * - It first determines the size of the shm segment that should be
981 * unmapped: It searches for a vma that is backed by shm and that
982 * started at address shmaddr. It records it's size and then unmaps
983 * it.
984 * - Then it unmaps all shm vmas that started at shmaddr and that
985 * are within the initially determined size.
986 * Errors from do_munmap are ignored: the function only fails if
987 * it's called with invalid parameters or if it's called to unmap
988 * a part of a vma. Both calls in this function are for full vmas,
989 * the parameters are directly copied from the vma itself and always
990 * valid - therefore do_munmap cannot fail. (famous last words?)
991 */
992 /*
993 * If it had been mremap()'d, the starting address would not
994 * match the usual checks anyway. So assume all vma's are
995 * above the starting address given.
996 */
997 vma = find_vma(mm, addr);
998
David Howells8feae132009-01-08 12:04:47 +0000999#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 while (vma) {
1001 next = vma->vm_next;
1002
1003 /*
1004 * Check if the starting address would match, i.e. it's
1005 * a fragment created by mprotect() and/or munmap(), or it
1006 * otherwise it starts at this address with no hassles.
1007 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001008 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1010
1011
Josef Sipek6d630792006-12-08 02:37:11 -08001012 size = vma->vm_file->f_path.dentry->d_inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1014 /*
1015 * We discovered the size of the shm segment, so
1016 * break out of here and fall through to the next
1017 * loop that uses the size information to stop
1018 * searching for matching vma's.
1019 */
1020 retval = 0;
1021 vma = next;
1022 break;
1023 }
1024 vma = next;
1025 }
1026
1027 /*
1028 * We need look no further than the maximum address a fragment
1029 * could possibly have landed at. Also cast things to loff_t to
1030 * prevent overflows and make comparisions vs. equal-width types.
1031 */
KAMEZAWA Hiroyuki8e367092006-02-10 01:51:12 -08001032 size = PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1034 next = vma->vm_next;
1035
1036 /* finding a matching vma now does not alter retval */
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 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1041 vma = next;
1042 }
1043
David Howells8feae132009-01-08 12:04:47 +00001044#else /* CONFIG_MMU */
1045 /* under NOMMU conditions, the exact address to be destroyed must be
1046 * given */
1047 retval = -EINVAL;
1048 if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1049 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1050 retval = 0;
1051 }
1052
1053#endif
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 up_write(&mm->mmap_sem);
1056 return retval;
1057}
1058
1059#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07001060static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Mike Waychison19b49462005-09-06 15:17:10 -07001062 struct shmid_kernel *shp = it;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Paul Menage6c826812008-06-12 15:21:49 -07001064#if BITS_PER_LONG <= 32
1065#define SIZE_SPEC "%10lu"
1066#else
1067#define SIZE_SPEC "%21lu"
1068#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
Paul Menage6c826812008-06-12 15:21:49 -07001070 return seq_printf(s,
1071 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1072 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
Mike Waychison19b49462005-09-06 15:17:10 -07001073 shp->shm_perm.key,
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001074 shp->shm_perm.id,
Andrew Mortonb33291c2006-01-08 01:02:21 -08001075 shp->shm_perm.mode,
Mike Waychison19b49462005-09-06 15:17:10 -07001076 shp->shm_segsz,
1077 shp->shm_cprid,
1078 shp->shm_lprid,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001079 shp->shm_nattch,
Mike Waychison19b49462005-09-06 15:17:10 -07001080 shp->shm_perm.uid,
1081 shp->shm_perm.gid,
1082 shp->shm_perm.cuid,
1083 shp->shm_perm.cgid,
1084 shp->shm_atim,
1085 shp->shm_dtim,
1086 shp->shm_ctim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087}
1088#endif