blob: faa46da99ebed7884ebb632f70ec7d79799d7f87 [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 Rothwell7d87e142005-05-01 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>
Mimi Zohar1df9f0a2009-02-04 09:07:02 -050042#include <linux/ima.h>
Stephen Rothwell7d87e142005-05-01 08:59:12 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/uaccess.h>
45
46#include "util.h"
47
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080048struct shm_file_data {
49 int id;
50 struct ipc_namespace *ns;
51 struct file *file;
52 const struct vm_operations_struct *vm_ops;
53};
54
55#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56
Arjan van de Ven9a321442007-02-12 00:55:35 -080057static const struct file_operations shm_file_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static struct vm_operations_struct shm_vm_ops;
59
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080060#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Kirill Korotaev4e982312006-10-02 02:18:22 -070062#define shm_unlock(shp) \
63 ipc_unlock(&(shp)->shm_perm)
Kirill Korotaev4e982312006-10-02 02:18:22 -070064
Nadia Derbey7748dbf2007-10-18 23:40:49 -070065static int newseg(struct ipc_namespace *, struct ipc_params *);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -080066static void shm_open(struct vm_area_struct *vma);
67static void shm_close(struct vm_area_struct *vma);
Kirill Korotaev4e982312006-10-02 02:18:22 -070068static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -070070static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#endif
72
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080073void shm_init_ns(struct ipc_namespace *ns)
Kirill Korotaev4e982312006-10-02 02:18:22 -070074{
Kirill Korotaev4e982312006-10-02 02:18:22 -070075 ns->shm_ctlmax = SHMMAX;
76 ns->shm_ctlall = SHMALL;
77 ns->shm_ctlmni = SHMMNI;
78 ns->shm_tot = 0;
WANG Conge8148f72009-01-06 14:42:49 -080079 ipc_init_ids(&shm_ids(ns));
Kirill Korotaev4e982312006-10-02 02:18:22 -070080}
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Nadia Derbeyf4566f02007-10-18 23:40:53 -070082/*
Nadia Derbey3e148c72007-10-18 23:40:54 -070083 * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84 * Only shm_ids.rw_mutex remains locked on exit.
Nadia Derbeyf4566f02007-10-18 23:40:53 -070085 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -080086static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Kirill Korotaev4e982312006-10-02 02:18:22 -070087{
Pierre Peiffer01b8b072008-02-08 04:18:57 -080088 struct shmid_kernel *shp;
89 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
90
Kirill Korotaev4e982312006-10-02 02:18:22 -070091 if (shp->shm_nattch){
92 shp->shm_perm.mode |= SHM_DEST;
93 /* Do not find it any more */
94 shp->shm_perm.key = IPC_PRIVATE;
95 shm_unlock(shp);
96 } else
97 shm_destroy(ns, shp);
98}
99
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800100#ifdef CONFIG_IPC_NS
Kirill Korotaev4e982312006-10-02 02:18:22 -0700101void shm_exit_ns(struct ipc_namespace *ns)
102{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800103 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
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",
111 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
Kirill Korotaev4e982312006-10-02 02:18:22 -0700112 IPC_SHM_IDS, sysvipc_shm_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114
Nadia Derbey3e148c72007-10-18 23:40:54 -0700115/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700116 * shm_lock_(check_) routines are called in the paths where the rw_mutex
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700117 * is not necessarily held.
Nadia Derbey3e148c72007-10-18 23:40:54 -0700118 */
Nadia Derbey023a5352007-10-18 23:40:51 -0700119static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700121 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
122
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800123 if (IS_ERR(ipcp))
124 return (struct shmid_kernel *)ipcp;
125
Nadia Derbey03f02c72007-10-18 23:40:51 -0700126 return container_of(ipcp, struct shmid_kernel, shm_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700127}
128
129static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
130 int id)
131{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700132 struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
133
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800134 if (IS_ERR(ipcp))
135 return (struct shmid_kernel *)ipcp;
136
Nadia Derbey03f02c72007-10-18 23:40:51 -0700137 return container_of(ipcp, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138}
139
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700140static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700142 ipc_rmid(&shm_ids(ns), &s->shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800146/* This is called by fork, once for every shm attach. */
147static void shm_open(struct vm_area_struct *vma)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700148{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800149 struct file *file = vma->vm_file;
150 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 struct shmid_kernel *shp;
152
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800153 shp = shm_lock(sfd->ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700154 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 shp->shm_atim = get_seconds();
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700156 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 shp->shm_nattch++;
158 shm_unlock(shp);
159}
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161/*
162 * shm_destroy - free the struct shmid_kernel
163 *
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700164 * @ns: namespace
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 * @shp: struct to free
166 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700167 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 * but returns with shp unlocked and freed.
169 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700170static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700172 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700173 shm_rmid(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 shm_unlock(shp);
175 if (!is_file_hugepages(shp->shm_file))
176 shmem_lock(shp->shm_file, 0, shp->mlock_user);
177 else
Josef Sipek6d630792006-12-08 02:37:11 -0800178 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 shp->mlock_user);
180 fput (shp->shm_file);
181 security_shm_free(shp);
182 ipc_rcu_putref(shp);
183}
184
185/*
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800186 * remove the attach descriptor vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 * free memory for segment if it is marked destroyed.
188 * The descriptor has already been removed from the current->mm->mmap list
189 * and will later be kfree()d.
190 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800191static void shm_close(struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800193 struct file * file = vma->vm_file;
194 struct shm_file_data *sfd = shm_file_data(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 struct shmid_kernel *shp;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800196 struct ipc_namespace *ns = sfd->ns;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700197
Nadia Derbey3e148c72007-10-18 23:40:54 -0700198 down_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 /* remove from the list of attaches of the shm segment */
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700200 shp = shm_lock(ns, sfd->id);
Nadia Derbey023a5352007-10-18 23:40:51 -0700201 BUG_ON(IS_ERR(shp));
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700202 shp->shm_lprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 shp->shm_dtim = get_seconds();
204 shp->shm_nattch--;
205 if(shp->shm_nattch == 0 &&
Andrew Mortonb33291c2006-01-08 01:02:21 -0800206 shp->shm_perm.mode & SHM_DEST)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700207 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 else
209 shm_unlock(shp);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700210 up_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
Nick Piggind0217ac2007-07-19 01:47:03 -0700213static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800214{
215 struct file *file = vma->vm_file;
216 struct shm_file_data *sfd = shm_file_data(file);
217
Nick Piggind0217ac2007-07-19 01:47:03 -0700218 return sfd->vm_ops->fault(vma, vmf);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800219}
220
221#ifdef CONFIG_NUMA
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700222static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800223{
224 struct file *file = vma->vm_file;
225 struct shm_file_data *sfd = shm_file_data(file);
226 int err = 0;
227 if (sfd->vm_ops->set_policy)
228 err = sfd->vm_ops->set_policy(vma, new);
229 return err;
230}
231
Adrian Bunkd823e3e2007-10-16 23:26:42 -0700232static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
233 unsigned long addr)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800234{
235 struct file *file = vma->vm_file;
236 struct shm_file_data *sfd = shm_file_data(file);
237 struct mempolicy *pol = NULL;
238
239 if (sfd->vm_ops->get_policy)
240 pol = sfd->vm_ops->get_policy(vma, addr);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700241 else if (vma->vm_policy)
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800242 pol = vma->vm_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700243
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800244 return pol;
245}
246#endif
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248static int shm_mmap(struct file * file, struct vm_area_struct * vma)
249{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800250 struct shm_file_data *sfd = shm_file_data(file);
David Howellsb0e15192006-01-06 00:11:42 -0800251 int ret;
252
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800253 ret = sfd->file->f_op->mmap(sfd->file, vma);
254 if (ret != 0)
255 return ret;
256 sfd->vm_ops = vma->vm_ops;
David Howells2e92a3b2007-07-31 00:37:24 -0700257#ifdef CONFIG_MMU
Nick Piggin54cb8822007-07-19 01:46:59 -0700258 BUG_ON(!sfd->vm_ops->fault);
David Howells2e92a3b2007-07-31 00:37:24 -0700259#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800260 vma->vm_ops = &shm_vm_ops;
261 shm_open(vma);
David Howellsb0e15192006-01-06 00:11:42 -0800262
263 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
Kirill Korotaev4e982312006-10-02 02:18:22 -0700266static int shm_release(struct inode *ino, struct file *file)
267{
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800268 struct shm_file_data *sfd = shm_file_data(file);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700269
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800270 put_ipc_ns(sfd->ns);
271 shm_file_data(file) = NULL;
272 kfree(sfd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700273 return 0;
274}
275
Adam Litke516dffd2007-03-01 15:46:08 -0800276static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
277{
278 int (*fsync) (struct file *, struct dentry *, int datasync);
279 struct shm_file_data *sfd = shm_file_data(file);
280 int ret = -EINVAL;
281
282 fsync = sfd->file->f_op->fsync;
283 if (fsync)
284 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
285 return ret;
286}
287
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800288static unsigned long shm_get_unmapped_area(struct file *file,
289 unsigned long addr, unsigned long len, unsigned long pgoff,
290 unsigned long flags)
291{
292 struct shm_file_data *sfd = shm_file_data(file);
Adam Litke516dffd2007-03-01 15:46:08 -0800293 return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800294}
Adam Litke516dffd2007-03-01 15:46:08 -0800295
296int is_file_shm_hugepages(struct file *file)
297{
298 int ret = 0;
299
300 if (file->f_op == &shm_file_operations) {
301 struct shm_file_data *sfd;
302 sfd = shm_file_data(file);
303 ret = is_file_hugepages(sfd->file);
304 }
305 return ret;
306}
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800307
Arjan van de Ven9a321442007-02-12 00:55:35 -0800308static const struct file_operations shm_file_operations = {
Kirill Korotaev4e982312006-10-02 02:18:22 -0700309 .mmap = shm_mmap,
Adam Litke516dffd2007-03-01 15:46:08 -0800310 .fsync = shm_fsync,
Kirill Korotaev4e982312006-10-02 02:18:22 -0700311 .release = shm_release,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800312 .get_unmapped_area = shm_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313};
314
315static struct vm_operations_struct shm_vm_ops = {
316 .open = shm_open, /* callback for a new vm-area open */
317 .close = shm_close, /* callback for when the vm-area is released */
Nick Piggin54cb8822007-07-19 01:46:59 -0700318 .fault = shm_fault,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800319#if defined(CONFIG_NUMA)
320 .set_policy = shm_set_policy,
321 .get_policy = shm_get_policy,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322#endif
323};
324
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700325/**
326 * newseg - Create a new shared memory segment
327 * @ns: namespace
328 * @params: ptr to the structure that contains key, size and shmflg
329 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700330 * Called with shm_ids.rw_mutex held as a writer.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700331 */
332
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700333static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700335 key_t key = params->key;
336 int shmflg = params->flg;
337 size_t size = params->u.size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 int error;
339 struct shmid_kernel *shp;
340 int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
341 struct file * file;
342 char name[13];
343 int id;
Mel Gorman5a6fe122009-02-10 14:02:27 +0000344 int acctflag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Kirill Korotaev4e982312006-10-02 02:18:22 -0700346 if (size < SHMMIN || size > ns->shm_ctlmax)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return -EINVAL;
348
Guy Streeterf66d45e2007-01-23 12:20:04 -0600349 if (ns->shm_tot + numpages > ns->shm_ctlall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return -ENOSPC;
351
352 shp = ipc_rcu_alloc(sizeof(*shp));
353 if (!shp)
354 return -ENOMEM;
355
356 shp->shm_perm.key = key;
Andrew Mortonb33291c2006-01-08 01:02:21 -0800357 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 shp->mlock_user = NULL;
359
360 shp->shm_perm.security = NULL;
361 error = security_shm_alloc(shp);
362 if (error) {
363 ipc_rcu_putref(shp);
364 return error;
365 }
366
Eric W. Biederman9d665862007-06-16 10:16:16 -0700367 sprintf (name, "SYSV%08x", key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 if (shmflg & SHM_HUGETLB) {
Mel Gorman5a6fe122009-02-10 14:02:27 +0000369 /* hugetlb_file_setup applies strict accounting */
370 if (shmflg & SHM_NORESERVE)
371 acctflag = VM_NORESERVE;
372 file = hugetlb_file_setup(name, size, acctflag);
David Howells86a264a2008-11-14 10:39:18 +1100373 shp->mlock_user = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 } else {
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800375 /*
376 * Do not allow no accounting for OVERCOMMIT_NEVER, even
377 * if it's asked for.
378 */
379 if ((shmflg & SHM_NORESERVE) &&
380 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
Linus Torvaldsfc8744a2009-01-31 15:08:56 -0800381 acctflag = VM_NORESERVE;
Badari Pulavartybf8f9722005-11-07 00:59:27 -0800382 file = shmem_file_setup(name, size, acctflag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
384 error = PTR_ERR(file);
385 if (IS_ERR(file))
386 goto no_file;
Mimi Zohar1df9f0a2009-02-04 09:07:02 -0500387 ima_shm_check(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Pierre Peiffer48dea402008-04-29 01:00:35 -0700389 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700390 if (id < 0) {
391 error = id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 goto no_id;
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700395 shp->shm_cprid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 shp->shm_lprid = 0;
397 shp->shm_atim = shp->shm_dtim = 0;
398 shp->shm_ctim = get_seconds();
399 shp->shm_segsz = size;
400 shp->shm_nattch = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 shp->shm_file = file;
Badari Pulavarty30475cc2007-06-16 10:15:59 -0700402 /*
403 * shmid gets reported as "inode#" in /proc/pid/maps.
404 * proc-ps tools use this. Changing this will break them.
405 */
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700406 file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
Krishnakumar R551110a2005-10-29 18:16:45 -0700407
Kirill Korotaev4e982312006-10-02 02:18:22 -0700408 ns->shm_tot += numpages;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700409 error = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 shm_unlock(shp);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700411 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413no_id:
414 fput(file);
415no_file:
416 security_shm_free(shp);
417 ipc_rcu_putref(shp);
418 return error;
419}
420
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700421/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700422 * Called with shm_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700423 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700424static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700425{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700426 struct shmid_kernel *shp;
427
428 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
429 return security_shm_associate(shp, shmflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700430}
431
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700432/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700433 * Called with shm_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700434 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700435static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
436 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700437{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700438 struct shmid_kernel *shp;
439
440 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
441 if (shp->shm_segsz < params->u.size)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700442 return -EINVAL;
443
444 return 0;
445}
446
Heiko Carstensd5460c92009-01-14 14:14:27 +0100447SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Kirill Korotaev4e982312006-10-02 02:18:22 -0700449 struct ipc_namespace *ns;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700450 struct ipc_ops shm_ops;
451 struct ipc_params shm_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
Kirill Korotaev4e982312006-10-02 02:18:22 -0700453 ns = current->nsproxy->ipc_ns;
454
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700455 shm_ops.getnew = newseg;
456 shm_ops.associate = shm_security;
457 shm_ops.more_checks = shm_more_checks;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700458
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700459 shm_params.key = key;
460 shm_params.flg = shmflg;
461 shm_params.u.size = size;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700462
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700463 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
466static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
467{
468 switch(version) {
469 case IPC_64:
470 return copy_to_user(buf, in, sizeof(*in));
471 case IPC_OLD:
472 {
473 struct shmid_ds out;
474
475 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
476 out.shm_segsz = in->shm_segsz;
477 out.shm_atime = in->shm_atime;
478 out.shm_dtime = in->shm_dtime;
479 out.shm_ctime = in->shm_ctime;
480 out.shm_cpid = in->shm_cpid;
481 out.shm_lpid = in->shm_lpid;
482 out.shm_nattch = in->shm_nattch;
483
484 return copy_to_user(buf, &out, sizeof(out));
485 }
486 default:
487 return -EINVAL;
488 }
489}
490
Pierre Peiffer016d7132008-04-29 01:00:50 -0700491static inline unsigned long
492copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 switch(version) {
495 case IPC_64:
Pierre Peiffer016d7132008-04-29 01:00:50 -0700496 if (copy_from_user(out, buf, sizeof(*out)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 case IPC_OLD:
500 {
501 struct shmid_ds tbuf_old;
502
503 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
504 return -EFAULT;
505
Pierre Peiffer016d7132008-04-29 01:00:50 -0700506 out->shm_perm.uid = tbuf_old.shm_perm.uid;
507 out->shm_perm.gid = tbuf_old.shm_perm.gid;
508 out->shm_perm.mode = tbuf_old.shm_perm.mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 return 0;
511 }
512 default:
513 return -EINVAL;
514 }
515}
516
517static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
518{
519 switch(version) {
520 case IPC_64:
521 return copy_to_user(buf, in, sizeof(*in));
522 case IPC_OLD:
523 {
524 struct shminfo out;
525
526 if(in->shmmax > INT_MAX)
527 out.shmmax = INT_MAX;
528 else
529 out.shmmax = (int)in->shmmax;
530
531 out.shmmin = in->shmmin;
532 out.shmmni = in->shmmni;
533 out.shmseg = in->shmseg;
534 out.shmall = in->shmall;
535
536 return copy_to_user(buf, &out, sizeof(out));
537 }
538 default:
539 return -EINVAL;
540 }
541}
542
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700543/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700544 * Called with shm_ids.rw_mutex held as a reader
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700545 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700546static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
547 unsigned long *swp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700549 int next_id;
550 int total, in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552 *rss = 0;
553 *swp = 0;
554
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700555 in_use = shm_ids(ns).in_use;
556
557 for (total = 0, next_id = 0; total < in_use; next_id++) {
Tony Battersbye562aeb2009-04-02 16:58:26 -0700558 struct kern_ipc_perm *ipc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 struct shmid_kernel *shp;
560 struct inode *inode;
561
Tony Battersbye562aeb2009-04-02 16:58:26 -0700562 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
563 if (ipc == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 continue;
Tony Battersbye562aeb2009-04-02 16:58:26 -0700565 shp = container_of(ipc, struct shmid_kernel, shm_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Josef Sipek6d630792006-12-08 02:37:11 -0800567 inode = shp->shm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
569 if (is_file_hugepages(shp->shm_file)) {
570 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700571 struct hstate *h = hstate_file(shp->shm_file);
572 *rss += pages_per_huge_page(h) * mapping->nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 } else {
Tony Battersbya68e61e2009-02-04 15:12:04 -0800574#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 struct shmem_inode_info *info = SHMEM_I(inode);
576 spin_lock(&info->lock);
577 *rss += inode->i_mapping->nrpages;
578 *swp += info->swapped;
579 spin_unlock(&info->lock);
Tony Battersbya68e61e2009-02-04 15:12:04 -0800580#else
581 *rss += inode->i_mapping->nrpages;
582#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700584
585 total++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
587}
588
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700589/*
590 * This function handles some shmctl commands which require the rw_mutex
591 * to be held in write mode.
592 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
593 */
594static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
595 struct shmid_ds __user *buf, int version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700597 struct kern_ipc_perm *ipcp;
Pierre Peiffer016d7132008-04-29 01:00:50 -0700598 struct shmid64_ds shmid64;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 struct shmid_kernel *shp;
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700600 int err;
601
602 if (cmd == IPC_SET) {
Pierre Peiffer016d7132008-04-29 01:00:50 -0700603 if (copy_shmid_from_user(&shmid64, buf, version))
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700604 return -EFAULT;
605 }
606
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700607 ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
608 if (IS_ERR(ipcp))
609 return PTR_ERR(ipcp);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700610
Pierre Peiffera5f75e72008-04-29 01:00:54 -0700611 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700612
613 err = security_shm_shmctl(shp, cmd);
614 if (err)
615 goto out_unlock;
616 switch (cmd) {
617 case IPC_RMID:
618 do_shm_rmid(ns, ipcp);
619 goto out_up;
620 case IPC_SET:
Pierre Peiffer8f4a3802008-04-29 01:00:51 -0700621 ipc_update_perm(&shmid64.shm_perm, ipcp);
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700622 shp->shm_ctim = get_seconds();
623 break;
624 default:
625 err = -EINVAL;
626 }
627out_unlock:
628 shm_unlock(shp);
629out_up:
630 up_write(&shm_ids(ns).rw_mutex);
631 return err;
632}
633
Heiko Carstensd5460c92009-01-14 14:14:27 +0100634SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700635{
636 struct shmid_kernel *shp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 int err, version;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700638 struct ipc_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
640 if (cmd < 0 || shmid < 0) {
641 err = -EINVAL;
642 goto out;
643 }
644
645 version = ipc_parse_version(&cmd);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700646 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648 switch (cmd) { /* replace with proc interface ? */
649 case IPC_INFO:
650 {
651 struct shminfo64 shminfo;
652
653 err = security_shm_shmctl(NULL, cmd);
654 if (err)
655 return err;
656
WANG Conge8148f72009-01-06 14:42:49 -0800657 memset(&shminfo, 0, sizeof(shminfo));
Kirill Korotaev4e982312006-10-02 02:18:22 -0700658 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
659 shminfo.shmmax = ns->shm_ctlmax;
660 shminfo.shmall = ns->shm_ctlall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662 shminfo.shmmin = SHMMIN;
663 if(copy_shminfo_to_user (buf, &shminfo, version))
664 return -EFAULT;
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700665
Nadia Derbey3e148c72007-10-18 23:40:54 -0700666 down_read(&shm_ids(ns).rw_mutex);
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700667 err = ipc_get_maxid(&shm_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700668 up_read(&shm_ids(ns).rw_mutex);
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 if(err<0)
671 err = 0;
672 goto out;
673 }
674 case SHM_INFO:
675 {
676 struct shm_info shm_info;
677
678 err = security_shm_shmctl(NULL, cmd);
679 if (err)
680 return err;
681
WANG Conge8148f72009-01-06 14:42:49 -0800682 memset(&shm_info, 0, sizeof(shm_info));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700683 down_read(&shm_ids(ns).rw_mutex);
Kirill Korotaev4e982312006-10-02 02:18:22 -0700684 shm_info.used_ids = shm_ids(ns).in_use;
685 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
686 shm_info.shm_tot = ns->shm_tot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 shm_info.swap_attempts = 0;
688 shm_info.swap_successes = 0;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700689 err = ipc_get_maxid(&shm_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700690 up_read(&shm_ids(ns).rw_mutex);
WANG Conge8148f72009-01-06 14:42:49 -0800691 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 err = -EFAULT;
693 goto out;
694 }
695
696 err = err < 0 ? 0 : err;
697 goto out;
698 }
699 case SHM_STAT:
700 case IPC_STAT:
701 {
702 struct shmid64_ds tbuf;
703 int result;
Nadia Derbey023a5352007-10-18 23:40:51 -0700704
Nadia Derbey023a5352007-10-18 23:40:51 -0700705 if (cmd == SHM_STAT) {
706 shp = shm_lock(ns, shmid);
707 if (IS_ERR(shp)) {
708 err = PTR_ERR(shp);
709 goto out;
710 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700711 result = shp->shm_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 } else {
Nadia Derbey023a5352007-10-18 23:40:51 -0700713 shp = shm_lock_check(ns, shmid);
714 if (IS_ERR(shp)) {
715 err = PTR_ERR(shp);
716 goto out;
717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 result = 0;
719 }
WANG Conge8148f72009-01-06 14:42:49 -0800720 err = -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 if (ipcperms (&shp->shm_perm, S_IRUGO))
722 goto out_unlock;
723 err = security_shm_shmctl(shp, cmd);
724 if (err)
725 goto out_unlock;
Nadia Derbey023a5352007-10-18 23:40:51 -0700726 memset(&tbuf, 0, sizeof(tbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
728 tbuf.shm_segsz = shp->shm_segsz;
729 tbuf.shm_atime = shp->shm_atim;
730 tbuf.shm_dtime = shp->shm_dtim;
731 tbuf.shm_ctime = shp->shm_ctim;
732 tbuf.shm_cpid = shp->shm_cprid;
733 tbuf.shm_lpid = shp->shm_lprid;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800734 tbuf.shm_nattch = shp->shm_nattch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 shm_unlock(shp);
736 if(copy_shmid_to_user (buf, &tbuf, version))
737 err = -EFAULT;
738 else
739 err = result;
740 goto out;
741 }
742 case SHM_LOCK:
743 case SHM_UNLOCK:
744 {
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -0700745 struct file *uninitialized_var(shm_file);
746
747 lru_add_drain_all(); /* drain pagevecs to lru lists */
748
Nadia Derbey023a5352007-10-18 23:40:51 -0700749 shp = shm_lock_check(ns, shmid);
750 if (IS_ERR(shp)) {
751 err = PTR_ERR(shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 goto out;
753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Al Viroa33e6752008-12-10 03:40:06 -0500755 audit_ipc_obj(&(shp->shm_perm));
Steve Grubb073115d2006-04-02 17:07:33 -0400756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (!capable(CAP_IPC_LOCK)) {
David Howells414c0702008-11-14 10:39:06 +1100758 uid_t euid = current_euid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 err = -EPERM;
David Howells414c0702008-11-14 10:39:06 +1100760 if (euid != shp->shm_perm.uid &&
761 euid != shp->shm_perm.cuid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 goto out_unlock;
763 if (cmd == SHM_LOCK &&
764 !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
765 goto out_unlock;
766 }
767
768 err = security_shm_shmctl(shp, cmd);
769 if (err)
770 goto out_unlock;
771
772 if(cmd==SHM_LOCK) {
David Howells86a264a2008-11-14 10:39:18 +1100773 struct user_struct *user = current_user();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 if (!is_file_hugepages(shp->shm_file)) {
775 err = shmem_lock(shp->shm_file, 1, user);
Pavel Emelianov7be77e22007-07-31 00:38:48 -0700776 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
Andrew Mortonb33291c2006-01-08 01:02:21 -0800777 shp->shm_perm.mode |= SHM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 shp->mlock_user = user;
779 }
780 }
781 } else if (!is_file_hugepages(shp->shm_file)) {
782 shmem_lock(shp->shm_file, 0, shp->mlock_user);
Andrew Mortonb33291c2006-01-08 01:02:21 -0800783 shp->shm_perm.mode &= ~SHM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 shp->mlock_user = NULL;
785 }
786 shm_unlock(shp);
787 goto out;
788 }
789 case IPC_RMID:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 case IPC_SET:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700791 err = shmctl_down(ns, shmid, cmd, buf, version);
792 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 default:
Pierre Peiffer8d4cc8b2008-04-29 01:00:47 -0700794 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797out_unlock:
798 shm_unlock(shp);
799out:
800 return err;
801}
802
803/*
804 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
805 *
806 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
807 * "raddr" thing points to kernel space, and there has to be a wrapper around
808 * this.
809 */
810long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
811{
812 struct shmid_kernel *shp;
813 unsigned long addr;
814 unsigned long size;
815 struct file * file;
816 int err;
817 unsigned long flags;
818 unsigned long prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 int acc_mode;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800820 unsigned long user_addr;
Kirill Korotaev4e982312006-10-02 02:18:22 -0700821 struct ipc_namespace *ns;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800822 struct shm_file_data *sfd;
823 struct path path;
Al Viroaeb5d722008-09-02 15:28:45 -0400824 fmode_t f_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800826 err = -EINVAL;
827 if (shmid < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 goto out;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800829 else if ((addr = (ulong)shmaddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (addr & (SHMLBA-1)) {
831 if (shmflg & SHM_RND)
832 addr &= ~(SHMLBA-1); /* round down */
833 else
834#ifndef __ARCH_FORCE_SHMLBA
835 if (addr & ~PAGE_MASK)
836#endif
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800837 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 }
839 flags = MAP_SHARED | MAP_FIXED;
840 } else {
841 if ((shmflg & SHM_REMAP))
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800842 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 flags = MAP_SHARED;
845 }
846
847 if (shmflg & SHM_RDONLY) {
848 prot = PROT_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 acc_mode = S_IRUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800850 f_mode = FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 } else {
852 prot = PROT_READ | PROT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 acc_mode = S_IRUGO | S_IWUGO;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800854 f_mode = FMODE_READ | FMODE_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
856 if (shmflg & SHM_EXEC) {
857 prot |= PROT_EXEC;
858 acc_mode |= S_IXUGO;
859 }
860
861 /*
862 * We cannot rely on the fs check since SYSV IPC does have an
863 * additional creator id...
864 */
Kirill Korotaev4e982312006-10-02 02:18:22 -0700865 ns = current->nsproxy->ipc_ns;
Nadia Derbey023a5352007-10-18 23:40:51 -0700866 shp = shm_lock_check(ns, shmid);
867 if (IS_ERR(shp)) {
868 err = PTR_ERR(shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 goto out;
Nadia Derbey023a5352007-10-18 23:40:51 -0700870 }
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800871
872 err = -EACCES;
873 if (ipcperms(&shp->shm_perm, acc_mode))
874 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 err = security_shm_shmat(shp, shmaddr, shmflg);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800877 if (err)
878 goto out_unlock;
879
880 path.dentry = dget(shp->shm_file->f_path.dentry);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700881 path.mnt = shp->shm_file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 shp->shm_nattch++;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800883 size = i_size_read(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 shm_unlock(shp);
885
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800886 err = -ENOMEM;
887 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
888 if (!sfd)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700889 goto out_put_dentry;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800890
Dave Hansence8d2cd2007-10-16 23:31:13 -0700891 file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800892 if (!file)
893 goto out_free;
Mimi Zohar1df9f0a2009-02-04 09:07:02 -0500894 ima_shm_check(file);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800895
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800896 file->private_data = sfd;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800897 file->f_mapping = shp->shm_file->f_mapping;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700898 sfd->id = shp->shm_perm.id;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800899 sfd->ns = get_ipc_ns(ns);
900 sfd->file = shp->shm_file;
901 sfd->vm_ops = NULL;
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 down_write(&current->mm->mmap_sem);
904 if (addr && !(shmflg & SHM_REMAP)) {
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800905 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 if (find_vma_intersection(current->mm, addr, addr + size))
907 goto invalid;
908 /*
909 * If shm segment goes below stack, make sure there is some
910 * space left for the stack to grow (at least 4 pages).
911 */
912 if (addr < current->mm->start_stack &&
913 addr > current->mm->start_stack - size - PAGE_SIZE * 5)
914 goto invalid;
915 }
916
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800917 user_addr = do_mmap (file, addr, size, prot, flags, 0);
918 *raddr = user_addr;
919 err = 0;
920 if (IS_ERR_VALUE(user_addr))
921 err = (long)user_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922invalid:
923 up_write(&current->mm->mmap_sem);
924
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800925 fput(file);
926
927out_nattch:
Nadia Derbey3e148c72007-10-18 23:40:54 -0700928 down_write(&shm_ids(ns).rw_mutex);
Nadia Derbey00c2bf82008-07-25 01:48:03 -0700929 shp = shm_lock(ns, shmid);
Nadia Derbey023a5352007-10-18 23:40:51 -0700930 BUG_ON(IS_ERR(shp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 shp->shm_nattch--;
932 if(shp->shm_nattch == 0 &&
Andrew Mortonb33291c2006-01-08 01:02:21 -0800933 shp->shm_perm.mode & SHM_DEST)
Kirill Korotaev4e982312006-10-02 02:18:22 -0700934 shm_destroy(ns, shp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 else
936 shm_unlock(shp);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700937 up_write(&shm_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939out:
940 return err;
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800941
942out_unlock:
943 shm_unlock(shp);
944 goto out;
945
946out_free:
947 kfree(sfd);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700948out_put_dentry:
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800949 dput(path.dentry);
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -0800950 goto out_nattch;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
Heiko Carstensd5460c92009-01-14 14:14:27 +0100953SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
Stephen Rothwell7d87e142005-05-01 08:59:12 -0700954{
955 unsigned long ret;
956 long err;
957
958 err = do_shmat(shmid, shmaddr, shmflg, &ret);
959 if (err)
960 return err;
961 force_successful_syscall_return();
962 return (long)ret;
963}
964
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965/*
966 * detach and kill segment if marked destroyed.
967 * The work is done in shm_close.
968 */
Heiko Carstensd5460c92009-01-14 14:14:27 +0100969SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
971 struct mm_struct *mm = current->mm;
972 struct vm_area_struct *vma, *next;
973 unsigned long addr = (unsigned long)shmaddr;
974 loff_t size = 0;
975 int retval = -EINVAL;
976
Hugh Dickinsdf1e2fb2006-03-24 03:18:06 -0800977 if (addr & ~PAGE_MASK)
978 return retval;
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 down_write(&mm->mmap_sem);
981
982 /*
983 * This function tries to be smart and unmap shm segments that
984 * were modified by partial mlock or munmap calls:
985 * - It first determines the size of the shm segment that should be
986 * unmapped: It searches for a vma that is backed by shm and that
987 * started at address shmaddr. It records it's size and then unmaps
988 * it.
989 * - Then it unmaps all shm vmas that started at shmaddr and that
990 * are within the initially determined size.
991 * Errors from do_munmap are ignored: the function only fails if
992 * it's called with invalid parameters or if it's called to unmap
993 * a part of a vma. Both calls in this function are for full vmas,
994 * the parameters are directly copied from the vma itself and always
995 * valid - therefore do_munmap cannot fail. (famous last words?)
996 */
997 /*
998 * If it had been mremap()'d, the starting address would not
999 * match the usual checks anyway. So assume all vma's are
1000 * above the starting address given.
1001 */
1002 vma = find_vma(mm, addr);
1003
David Howells8feae132009-01-08 12:04:47 +00001004#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 while (vma) {
1006 next = vma->vm_next;
1007
1008 /*
1009 * Check if the starting address would match, i.e. it's
1010 * a fragment created by mprotect() and/or munmap(), or it
1011 * otherwise it starts at this address with no hassles.
1012 */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001013 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1015
1016
Josef Sipek6d630792006-12-08 02:37:11 -08001017 size = vma->vm_file->f_path.dentry->d_inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1019 /*
1020 * We discovered the size of the shm segment, so
1021 * break out of here and fall through to the next
1022 * loop that uses the size information to stop
1023 * searching for matching vma's.
1024 */
1025 retval = 0;
1026 vma = next;
1027 break;
1028 }
1029 vma = next;
1030 }
1031
1032 /*
1033 * We need look no further than the maximum address a fragment
1034 * could possibly have landed at. Also cast things to loff_t to
1035 * prevent overflows and make comparisions vs. equal-width types.
1036 */
KAMEZAWA Hiroyuki8e367092006-02-10 01:51:12 -08001037 size = PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1039 next = vma->vm_next;
1040
1041 /* finding a matching vma now does not alter retval */
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001042 if ((vma->vm_ops == &shm_vm_ops) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1044
1045 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1046 vma = next;
1047 }
1048
David Howells8feae132009-01-08 12:04:47 +00001049#else /* CONFIG_MMU */
1050 /* under NOMMU conditions, the exact address to be destroyed must be
1051 * given */
1052 retval = -EINVAL;
1053 if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1054 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1055 retval = 0;
1056 }
1057
1058#endif
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 up_write(&mm->mmap_sem);
1061 return retval;
1062}
1063
1064#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07001065static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
Mike Waychison19b49462005-09-06 15:17:10 -07001067 struct shmid_kernel *shp = it;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
Paul Menage6c826812008-06-12 15:21:49 -07001069#if BITS_PER_LONG <= 32
1070#define SIZE_SPEC "%10lu"
1071#else
1072#define SIZE_SPEC "%21lu"
1073#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Paul Menage6c826812008-06-12 15:21:49 -07001075 return seq_printf(s,
1076 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1077 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
Mike Waychison19b49462005-09-06 15:17:10 -07001078 shp->shm_perm.key,
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001079 shp->shm_perm.id,
Andrew Mortonb33291c2006-01-08 01:02:21 -08001080 shp->shm_perm.mode,
Mike Waychison19b49462005-09-06 15:17:10 -07001081 shp->shm_segsz,
1082 shp->shm_cprid,
1083 shp->shm_lprid,
Eric W. Biedermanbc56bba2007-02-20 13:57:53 -08001084 shp->shm_nattch,
Mike Waychison19b49462005-09-06 15:17:10 -07001085 shp->shm_perm.uid,
1086 shp->shm_perm.gid,
1087 shp->shm_perm.cuid,
1088 shp->shm_perm.cgid,
1089 shp->shm_atim,
1090 shp->shm_dtim,
1091 shp->shm_ctim);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092}
1093#endif