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.
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>
16 * support for audit of ipc object properties and permission changes
17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 * Pavel Emelianov <xemul@openvz.org>
24 #include <linux/slab.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>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42 #include <linux/ima.h>
44 #include <asm/uaccess.h>
48 struct shm_file_data
{
50 struct ipc_namespace
*ns
;
52 const struct vm_operations_struct
*vm_ops
;
55 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
57 static const struct file_operations shm_file_operations
;
58 static const struct vm_operations_struct shm_vm_ops
;
60 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
62 #define shm_unlock(shp) \
63 ipc_unlock(&(shp)->shm_perm)
65 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
66 static void shm_open(struct vm_area_struct
*vma
);
67 static void shm_close(struct vm_area_struct
*vma
);
68 static void shm_destroy (struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
70 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
73 void shm_init_ns(struct ipc_namespace
*ns
)
75 ns
->shm_ctlmax
= SHMMAX
;
76 ns
->shm_ctlall
= SHMALL
;
77 ns
->shm_ctlmni
= SHMMNI
;
79 ipc_init_ids(&shm_ids(ns
));
83 * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84 * Only shm_ids.rw_mutex remains locked on exit.
86 static void do_shm_rmid(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
88 struct shmid_kernel
*shp
;
89 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
92 shp
->shm_perm
.mode
|= SHM_DEST
;
93 /* Do not find it any more */
94 shp
->shm_perm
.key
= IPC_PRIVATE
;
101 void shm_exit_ns(struct ipc_namespace
*ns
)
103 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
104 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
108 void __init
shm_init (void)
110 shm_init_ns(&init_ipc_ns
);
111 ipc_init_proc_interface("sysvipc/shm",
112 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
113 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
117 * shm_lock_(check_) routines are called in the paths where the rw_mutex
118 * is not necessarily held.
120 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
122 struct kern_ipc_perm
*ipcp
= ipc_lock(&shm_ids(ns
), id
);
125 return (struct shmid_kernel
*)ipcp
;
127 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
130 static inline struct shmid_kernel
*shm_lock_check(struct ipc_namespace
*ns
,
133 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&shm_ids(ns
), id
);
136 return (struct shmid_kernel
*)ipcp
;
138 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
141 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
143 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
147 /* This is called by fork, once for every shm attach. */
148 static void shm_open(struct vm_area_struct
*vma
)
150 struct file
*file
= vma
->vm_file
;
151 struct shm_file_data
*sfd
= shm_file_data(file
);
152 struct shmid_kernel
*shp
;
154 shp
= shm_lock(sfd
->ns
, sfd
->id
);
156 shp
->shm_atim
= get_seconds();
157 shp
->shm_lprid
= task_tgid_vnr(current
);
163 * shm_destroy - free the struct shmid_kernel
166 * @shp: struct to free
168 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
169 * but returns with shp unlocked and freed.
171 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
173 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
176 if (!is_file_hugepages(shp
->shm_file
))
177 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
178 else if (shp
->mlock_user
)
179 user_shm_unlock(shp
->shm_file
->f_path
.dentry
->d_inode
->i_size
,
181 fput (shp
->shm_file
);
182 security_shm_free(shp
);
187 * remove the attach descriptor vma.
188 * free memory for segment if it is marked destroyed.
189 * The descriptor has already been removed from the current->mm->mmap list
190 * and will later be kfree()d.
192 static void shm_close(struct vm_area_struct
*vma
)
194 struct file
* file
= vma
->vm_file
;
195 struct shm_file_data
*sfd
= shm_file_data(file
);
196 struct shmid_kernel
*shp
;
197 struct ipc_namespace
*ns
= sfd
->ns
;
199 down_write(&shm_ids(ns
).rw_mutex
);
200 /* remove from the list of attaches of the shm segment */
201 shp
= shm_lock(ns
, sfd
->id
);
203 shp
->shm_lprid
= task_tgid_vnr(current
);
204 shp
->shm_dtim
= get_seconds();
206 if(shp
->shm_nattch
== 0 &&
207 shp
->shm_perm
.mode
& SHM_DEST
)
208 shm_destroy(ns
, shp
);
211 up_write(&shm_ids(ns
).rw_mutex
);
214 static int shm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
216 struct file
*file
= vma
->vm_file
;
217 struct shm_file_data
*sfd
= shm_file_data(file
);
219 return sfd
->vm_ops
->fault(vma
, vmf
);
223 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
225 struct file
*file
= vma
->vm_file
;
226 struct shm_file_data
*sfd
= shm_file_data(file
);
228 if (sfd
->vm_ops
->set_policy
)
229 err
= sfd
->vm_ops
->set_policy(vma
, new);
233 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
236 struct file
*file
= vma
->vm_file
;
237 struct shm_file_data
*sfd
= shm_file_data(file
);
238 struct mempolicy
*pol
= NULL
;
240 if (sfd
->vm_ops
->get_policy
)
241 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
242 else if (vma
->vm_policy
)
243 pol
= vma
->vm_policy
;
249 static int shm_mmap(struct file
* file
, struct vm_area_struct
* vma
)
251 struct shm_file_data
*sfd
= shm_file_data(file
);
254 ret
= sfd
->file
->f_op
->mmap(sfd
->file
, vma
);
257 sfd
->vm_ops
= vma
->vm_ops
;
259 BUG_ON(!sfd
->vm_ops
->fault
);
261 vma
->vm_ops
= &shm_vm_ops
;
267 static int shm_release(struct inode
*ino
, struct file
*file
)
269 struct shm_file_data
*sfd
= shm_file_data(file
);
272 shm_file_data(file
) = NULL
;
277 static int shm_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
279 int (*fsync
) (struct file
*, struct dentry
*, int datasync
);
280 struct shm_file_data
*sfd
= shm_file_data(file
);
283 fsync
= sfd
->file
->f_op
->fsync
;
285 ret
= fsync(sfd
->file
, sfd
->file
->f_path
.dentry
, datasync
);
289 static unsigned long shm_get_unmapped_area(struct file
*file
,
290 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
293 struct shm_file_data
*sfd
= shm_file_data(file
);
294 return get_unmapped_area(sfd
->file
, addr
, len
, pgoff
, flags
);
297 int is_file_shm_hugepages(struct file
*file
)
301 if (file
->f_op
== &shm_file_operations
) {
302 struct shm_file_data
*sfd
;
303 sfd
= shm_file_data(file
);
304 ret
= is_file_hugepages(sfd
->file
);
309 static const struct file_operations shm_file_operations
= {
312 .release
= shm_release
,
313 .get_unmapped_area
= shm_get_unmapped_area
,
316 static const struct vm_operations_struct shm_vm_ops
= {
317 .open
= shm_open
, /* callback for a new vm-area open */
318 .close
= shm_close
, /* callback for when the vm-area is released */
320 #if defined(CONFIG_NUMA)
321 .set_policy
= shm_set_policy
,
322 .get_policy
= shm_get_policy
,
327 * newseg - Create a new shared memory segment
329 * @params: ptr to the structure that contains key, size and shmflg
331 * Called with shm_ids.rw_mutex held as a writer.
334 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
336 key_t key
= params
->key
;
337 int shmflg
= params
->flg
;
338 size_t size
= params
->u
.size
;
340 struct shmid_kernel
*shp
;
341 int numpages
= (size
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
347 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
350 if (ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
353 shp
= ipc_rcu_alloc(sizeof(*shp
));
357 shp
->shm_perm
.key
= key
;
358 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
359 shp
->mlock_user
= NULL
;
361 shp
->shm_perm
.security
= NULL
;
362 error
= security_shm_alloc(shp
);
368 sprintf (name
, "SYSV%08x", key
);
369 if (shmflg
& SHM_HUGETLB
) {
370 /* hugetlb_file_setup applies strict accounting */
371 if (shmflg
& SHM_NORESERVE
)
372 acctflag
= VM_NORESERVE
;
373 file
= hugetlb_file_setup(name
, size
, acctflag
,
374 &shp
->mlock_user
, HUGETLB_SHMFS_INODE
);
377 * Do not allow no accounting for OVERCOMMIT_NEVER, even
380 if ((shmflg
& SHM_NORESERVE
) &&
381 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
382 acctflag
= VM_NORESERVE
;
383 file
= shmem_file_setup(name
, size
, acctflag
);
385 error
= PTR_ERR(file
);
389 id
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
395 shp
->shm_cprid
= task_tgid_vnr(current
);
397 shp
->shm_atim
= shp
->shm_dtim
= 0;
398 shp
->shm_ctim
= get_seconds();
399 shp
->shm_segsz
= size
;
401 shp
->shm_file
= file
;
403 * shmid gets reported as "inode#" in /proc/pid/maps.
404 * proc-ps tools use this. Changing this will break them.
406 file
->f_dentry
->d_inode
->i_ino
= shp
->shm_perm
.id
;
408 ns
->shm_tot
+= numpages
;
409 error
= shp
->shm_perm
.id
;
414 if (is_file_hugepages(file
) && shp
->mlock_user
)
415 user_shm_unlock(size
, shp
->mlock_user
);
418 security_shm_free(shp
);
424 * Called with shm_ids.rw_mutex and ipcp locked.
426 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
428 struct shmid_kernel
*shp
;
430 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
431 return security_shm_associate(shp
, shmflg
);
435 * Called with shm_ids.rw_mutex and ipcp locked.
437 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
438 struct ipc_params
*params
)
440 struct shmid_kernel
*shp
;
442 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
443 if (shp
->shm_segsz
< params
->u
.size
)
449 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
451 struct ipc_namespace
*ns
;
452 struct ipc_ops shm_ops
;
453 struct ipc_params shm_params
;
455 ns
= current
->nsproxy
->ipc_ns
;
457 shm_ops
.getnew
= newseg
;
458 shm_ops
.associate
= shm_security
;
459 shm_ops
.more_checks
= shm_more_checks
;
461 shm_params
.key
= key
;
462 shm_params
.flg
= shmflg
;
463 shm_params
.u
.size
= size
;
465 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
468 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
472 return copy_to_user(buf
, in
, sizeof(*in
));
477 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
478 out
.shm_segsz
= in
->shm_segsz
;
479 out
.shm_atime
= in
->shm_atime
;
480 out
.shm_dtime
= in
->shm_dtime
;
481 out
.shm_ctime
= in
->shm_ctime
;
482 out
.shm_cpid
= in
->shm_cpid
;
483 out
.shm_lpid
= in
->shm_lpid
;
484 out
.shm_nattch
= in
->shm_nattch
;
486 return copy_to_user(buf
, &out
, sizeof(out
));
493 static inline unsigned long
494 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
498 if (copy_from_user(out
, buf
, sizeof(*out
)))
503 struct shmid_ds tbuf_old
;
505 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
508 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
509 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
510 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
519 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
523 return copy_to_user(buf
, in
, sizeof(*in
));
528 if(in
->shmmax
> INT_MAX
)
529 out
.shmmax
= INT_MAX
;
531 out
.shmmax
= (int)in
->shmmax
;
533 out
.shmmin
= in
->shmmin
;
534 out
.shmmni
= in
->shmmni
;
535 out
.shmseg
= in
->shmseg
;
536 out
.shmall
= in
->shmall
;
538 return copy_to_user(buf
, &out
, sizeof(out
));
546 * Called with shm_ids.rw_mutex held as a reader
548 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
557 in_use
= shm_ids(ns
).in_use
;
559 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
560 struct kern_ipc_perm
*ipc
;
561 struct shmid_kernel
*shp
;
564 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
567 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
569 inode
= shp
->shm_file
->f_path
.dentry
->d_inode
;
571 if (is_file_hugepages(shp
->shm_file
)) {
572 struct address_space
*mapping
= inode
->i_mapping
;
573 struct hstate
*h
= hstate_file(shp
->shm_file
);
574 *rss
+= pages_per_huge_page(h
) * mapping
->nrpages
;
577 struct shmem_inode_info
*info
= SHMEM_I(inode
);
578 spin_lock(&info
->lock
);
579 *rss
+= inode
->i_mapping
->nrpages
;
580 *swp
+= info
->swapped
;
581 spin_unlock(&info
->lock
);
583 *rss
+= inode
->i_mapping
->nrpages
;
592 * This function handles some shmctl commands which require the rw_mutex
593 * to be held in write mode.
594 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
596 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
597 struct shmid_ds __user
*buf
, int version
)
599 struct kern_ipc_perm
*ipcp
;
600 struct shmid64_ds shmid64
;
601 struct shmid_kernel
*shp
;
604 if (cmd
== IPC_SET
) {
605 if (copy_shmid_from_user(&shmid64
, buf
, version
))
609 ipcp
= ipcctl_pre_down(&shm_ids(ns
), shmid
, cmd
, &shmid64
.shm_perm
, 0);
611 return PTR_ERR(ipcp
);
613 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
615 err
= security_shm_shmctl(shp
, cmd
);
620 do_shm_rmid(ns
, ipcp
);
623 ipc_update_perm(&shmid64
.shm_perm
, ipcp
);
624 shp
->shm_ctim
= get_seconds();
632 up_write(&shm_ids(ns
).rw_mutex
);
636 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
638 struct shmid_kernel
*shp
;
640 struct ipc_namespace
*ns
;
642 if (cmd
< 0 || shmid
< 0) {
647 version
= ipc_parse_version(&cmd
);
648 ns
= current
->nsproxy
->ipc_ns
;
650 switch (cmd
) { /* replace with proc interface ? */
653 struct shminfo64 shminfo
;
655 err
= security_shm_shmctl(NULL
, cmd
);
659 memset(&shminfo
, 0, sizeof(shminfo
));
660 shminfo
.shmmni
= shminfo
.shmseg
= ns
->shm_ctlmni
;
661 shminfo
.shmmax
= ns
->shm_ctlmax
;
662 shminfo
.shmall
= ns
->shm_ctlall
;
664 shminfo
.shmmin
= SHMMIN
;
665 if(copy_shminfo_to_user (buf
, &shminfo
, version
))
668 down_read(&shm_ids(ns
).rw_mutex
);
669 err
= ipc_get_maxid(&shm_ids(ns
));
670 up_read(&shm_ids(ns
).rw_mutex
);
678 struct shm_info shm_info
;
680 err
= security_shm_shmctl(NULL
, cmd
);
684 memset(&shm_info
, 0, sizeof(shm_info
));
685 down_read(&shm_ids(ns
).rw_mutex
);
686 shm_info
.used_ids
= shm_ids(ns
).in_use
;
687 shm_get_stat (ns
, &shm_info
.shm_rss
, &shm_info
.shm_swp
);
688 shm_info
.shm_tot
= ns
->shm_tot
;
689 shm_info
.swap_attempts
= 0;
690 shm_info
.swap_successes
= 0;
691 err
= ipc_get_maxid(&shm_ids(ns
));
692 up_read(&shm_ids(ns
).rw_mutex
);
693 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
))) {
698 err
= err
< 0 ? 0 : err
;
704 struct shmid64_ds tbuf
;
707 if (cmd
== SHM_STAT
) {
708 shp
= shm_lock(ns
, shmid
);
713 result
= shp
->shm_perm
.id
;
715 shp
= shm_lock_check(ns
, shmid
);
723 if (ipcperms (&shp
->shm_perm
, S_IRUGO
))
725 err
= security_shm_shmctl(shp
, cmd
);
728 memset(&tbuf
, 0, sizeof(tbuf
));
729 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
.shm_perm
);
730 tbuf
.shm_segsz
= shp
->shm_segsz
;
731 tbuf
.shm_atime
= shp
->shm_atim
;
732 tbuf
.shm_dtime
= shp
->shm_dtim
;
733 tbuf
.shm_ctime
= shp
->shm_ctim
;
734 tbuf
.shm_cpid
= shp
->shm_cprid
;
735 tbuf
.shm_lpid
= shp
->shm_lprid
;
736 tbuf
.shm_nattch
= shp
->shm_nattch
;
738 if(copy_shmid_to_user (buf
, &tbuf
, version
))
747 struct file
*uninitialized_var(shm_file
);
749 lru_add_drain_all(); /* drain pagevecs to lru lists */
751 shp
= shm_lock_check(ns
, shmid
);
757 audit_ipc_obj(&(shp
->shm_perm
));
759 if (!capable(CAP_IPC_LOCK
)) {
760 uid_t euid
= current_euid();
762 if (euid
!= shp
->shm_perm
.uid
&&
763 euid
!= shp
->shm_perm
.cuid
)
765 if (cmd
== SHM_LOCK
&&
766 !current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
)
770 err
= security_shm_shmctl(shp
, cmd
);
775 struct user_struct
*user
= current_user();
776 if (!is_file_hugepages(shp
->shm_file
)) {
777 err
= shmem_lock(shp
->shm_file
, 1, user
);
778 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)){
779 shp
->shm_perm
.mode
|= SHM_LOCKED
;
780 shp
->mlock_user
= user
;
783 } else if (!is_file_hugepages(shp
->shm_file
)) {
784 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
785 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
786 shp
->mlock_user
= NULL
;
793 err
= shmctl_down(ns
, shmid
, cmd
, buf
, version
);
806 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
808 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
809 * "raddr" thing points to kernel space, and there has to be a wrapper around
812 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
, ulong
*raddr
)
814 struct shmid_kernel
*shp
;
822 unsigned long user_addr
;
823 struct ipc_namespace
*ns
;
824 struct shm_file_data
*sfd
;
831 else if ((addr
= (ulong
)shmaddr
)) {
832 if (addr
& (SHMLBA
-1)) {
833 if (shmflg
& SHM_RND
)
834 addr
&= ~(SHMLBA
-1); /* round down */
836 #ifndef __ARCH_FORCE_SHMLBA
837 if (addr
& ~PAGE_MASK
)
841 flags
= MAP_SHARED
| MAP_FIXED
;
843 if ((shmflg
& SHM_REMAP
))
849 if (shmflg
& SHM_RDONLY
) {
854 prot
= PROT_READ
| PROT_WRITE
;
855 acc_mode
= S_IRUGO
| S_IWUGO
;
856 f_mode
= FMODE_READ
| FMODE_WRITE
;
858 if (shmflg
& SHM_EXEC
) {
864 * We cannot rely on the fs check since SYSV IPC does have an
865 * additional creator id...
867 ns
= current
->nsproxy
->ipc_ns
;
868 shp
= shm_lock_check(ns
, shmid
);
875 if (ipcperms(&shp
->shm_perm
, acc_mode
))
878 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
882 path
.dentry
= dget(shp
->shm_file
->f_path
.dentry
);
883 path
.mnt
= shp
->shm_file
->f_path
.mnt
;
885 size
= i_size_read(path
.dentry
->d_inode
);
889 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
893 file
= alloc_file(path
.mnt
, path
.dentry
, f_mode
, &shm_file_operations
);
896 ima_counts_get(file
);
898 file
->private_data
= sfd
;
899 file
->f_mapping
= shp
->shm_file
->f_mapping
;
900 sfd
->id
= shp
->shm_perm
.id
;
901 sfd
->ns
= get_ipc_ns(ns
);
902 sfd
->file
= shp
->shm_file
;
905 down_write(¤t
->mm
->mmap_sem
);
906 if (addr
&& !(shmflg
& SHM_REMAP
)) {
908 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
911 * If shm segment goes below stack, make sure there is some
912 * space left for the stack to grow (at least 4 pages).
914 if (addr
< current
->mm
->start_stack
&&
915 addr
> current
->mm
->start_stack
- size
- PAGE_SIZE
* 5)
919 user_addr
= do_mmap (file
, addr
, size
, prot
, flags
, 0);
922 if (IS_ERR_VALUE(user_addr
))
923 err
= (long)user_addr
;
925 up_write(¤t
->mm
->mmap_sem
);
930 down_write(&shm_ids(ns
).rw_mutex
);
931 shp
= shm_lock(ns
, shmid
);
934 if(shp
->shm_nattch
== 0 &&
935 shp
->shm_perm
.mode
& SHM_DEST
)
936 shm_destroy(ns
, shp
);
939 up_write(&shm_ids(ns
).rw_mutex
);
955 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
960 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
);
963 force_successful_syscall_return();
968 * detach and kill segment if marked destroyed.
969 * The work is done in shm_close.
971 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
973 struct mm_struct
*mm
= current
->mm
;
974 struct vm_area_struct
*vma
;
975 unsigned long addr
= (unsigned long)shmaddr
;
976 int retval
= -EINVAL
;
979 struct vm_area_struct
*next
;
982 if (addr
& ~PAGE_MASK
)
985 down_write(&mm
->mmap_sem
);
988 * This function tries to be smart and unmap shm segments that
989 * were modified by partial mlock or munmap calls:
990 * - It first determines the size of the shm segment that should be
991 * unmapped: It searches for a vma that is backed by shm and that
992 * started at address shmaddr. It records it's size and then unmaps
994 * - Then it unmaps all shm vmas that started at shmaddr and that
995 * are within the initially determined size.
996 * Errors from do_munmap are ignored: the function only fails if
997 * it's called with invalid parameters or if it's called to unmap
998 * a part of a vma. Both calls in this function are for full vmas,
999 * the parameters are directly copied from the vma itself and always
1000 * valid - therefore do_munmap cannot fail. (famous last words?)
1003 * If it had been mremap()'d, the starting address would not
1004 * match the usual checks anyway. So assume all vma's are
1005 * above the starting address given.
1007 vma
= find_vma(mm
, addr
);
1011 next
= vma
->vm_next
;
1014 * Check if the starting address would match, i.e. it's
1015 * a fragment created by mprotect() and/or munmap(), or it
1016 * otherwise it starts at this address with no hassles.
1018 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1019 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1022 size
= vma
->vm_file
->f_path
.dentry
->d_inode
->i_size
;
1023 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1025 * We discovered the size of the shm segment, so
1026 * break out of here and fall through to the next
1027 * loop that uses the size information to stop
1028 * searching for matching vma's.
1038 * We need look no further than the maximum address a fragment
1039 * could possibly have landed at. Also cast things to loff_t to
1040 * prevent overflows and make comparisions vs. equal-width types.
1042 size
= PAGE_ALIGN(size
);
1043 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1044 next
= vma
->vm_next
;
1046 /* finding a matching vma now does not alter retval */
1047 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1048 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
)
1050 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1054 #else /* CONFIG_MMU */
1055 /* under NOMMU conditions, the exact address to be destroyed must be
1058 if (vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1059 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1065 up_write(&mm
->mmap_sem
);
1069 #ifdef CONFIG_PROC_FS
1070 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1072 struct shmid_kernel
*shp
= it
;
1074 #if BITS_PER_LONG <= 32
1075 #define SIZE_SPEC "%10lu"
1077 #define SIZE_SPEC "%21lu"
1080 return seq_printf(s
,
1081 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1082 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",