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/mutex.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
42 #include <asm/uaccess.h>
46 struct shm_file_data
{
48 struct ipc_namespace
*ns
;
50 const struct vm_operations_struct
*vm_ops
;
53 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55 static const struct file_operations shm_file_operations
;
56 static struct vm_operations_struct shm_vm_ops
;
58 static struct ipc_ids init_shm_ids
;
60 #define shm_ids(ns) (*((ns)->ids[IPC_SHM_IDS]))
62 #define shm_unlock(shp) \
63 ipc_unlock(&(shp)->shm_perm)
64 #define shm_buildid(ns, id, seq) \
65 ipc_buildid(&shm_ids(ns), id, seq)
67 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
68 static void shm_open(struct vm_area_struct
*vma
);
69 static void shm_close(struct vm_area_struct
*vma
);
70 static void shm_destroy (struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
72 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
75 static void __shm_init_ns(struct ipc_namespace
*ns
, struct ipc_ids
*ids
)
77 ns
->ids
[IPC_SHM_IDS
] = ids
;
78 ns
->shm_ctlmax
= SHMMAX
;
79 ns
->shm_ctlall
= SHMALL
;
80 ns
->shm_ctlmni
= SHMMNI
;
86 * Called with shm_ids.mutex and the shp structure locked.
87 * Only shm_ids.mutex remains locked on exit.
89 static void do_shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
92 shp
->shm_perm
.mode
|= SHM_DEST
;
93 /* Do not find it any more */
94 shp
->shm_perm
.key
= IPC_PRIVATE
;
100 int shm_init_ns(struct ipc_namespace
*ns
)
104 ids
= kmalloc(sizeof(struct ipc_ids
), GFP_KERNEL
);
108 __shm_init_ns(ns
, ids
);
112 void shm_exit_ns(struct ipc_namespace
*ns
)
114 struct shmid_kernel
*shp
;
118 mutex_lock(&shm_ids(ns
).mutex
);
120 in_use
= shm_ids(ns
).in_use
;
122 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
123 shp
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
126 ipc_lock_by_ptr(&shp
->shm_perm
);
127 do_shm_rmid(ns
, shp
);
130 mutex_unlock(&shm_ids(ns
).mutex
);
132 kfree(ns
->ids
[IPC_SHM_IDS
]);
133 ns
->ids
[IPC_SHM_IDS
] = NULL
;
136 void __init
shm_init (void)
138 __shm_init_ns(&init_ipc_ns
, &init_shm_ids
);
139 ipc_init_proc_interface("sysvipc/shm",
140 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime\n",
141 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
144 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
146 struct kern_ipc_perm
*ipcp
= ipc_lock(&shm_ids(ns
), id
);
148 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
151 static inline struct shmid_kernel
*shm_lock_check(struct ipc_namespace
*ns
,
154 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&shm_ids(ns
), id
);
156 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
159 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
161 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
164 static inline int shm_addid(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
166 return ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
171 /* This is called by fork, once for every shm attach. */
172 static void shm_open(struct vm_area_struct
*vma
)
174 struct file
*file
= vma
->vm_file
;
175 struct shm_file_data
*sfd
= shm_file_data(file
);
176 struct shmid_kernel
*shp
;
178 shp
= shm_lock(sfd
->ns
, sfd
->id
);
180 shp
->shm_atim
= get_seconds();
181 shp
->shm_lprid
= task_tgid_vnr(current
);
187 * shm_destroy - free the struct shmid_kernel
190 * @shp: struct to free
192 * It has to be called with shp and shm_ids.mutex locked,
193 * but returns with shp unlocked and freed.
195 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
197 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
200 if (!is_file_hugepages(shp
->shm_file
))
201 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
203 user_shm_unlock(shp
->shm_file
->f_path
.dentry
->d_inode
->i_size
,
205 fput (shp
->shm_file
);
206 security_shm_free(shp
);
211 * remove the attach descriptor vma.
212 * free memory for segment if it is marked destroyed.
213 * The descriptor has already been removed from the current->mm->mmap list
214 * and will later be kfree()d.
216 static void shm_close(struct vm_area_struct
*vma
)
218 struct file
* file
= vma
->vm_file
;
219 struct shm_file_data
*sfd
= shm_file_data(file
);
220 struct shmid_kernel
*shp
;
221 struct ipc_namespace
*ns
= sfd
->ns
;
223 mutex_lock(&shm_ids(ns
).mutex
);
224 /* remove from the list of attaches of the shm segment */
225 shp
= shm_lock(ns
, sfd
->id
);
227 shp
->shm_lprid
= task_tgid_vnr(current
);
228 shp
->shm_dtim
= get_seconds();
230 if(shp
->shm_nattch
== 0 &&
231 shp
->shm_perm
.mode
& SHM_DEST
)
232 shm_destroy(ns
, shp
);
235 mutex_unlock(&shm_ids(ns
).mutex
);
238 static int shm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
240 struct file
*file
= vma
->vm_file
;
241 struct shm_file_data
*sfd
= shm_file_data(file
);
243 return sfd
->vm_ops
->fault(vma
, vmf
);
247 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
249 struct file
*file
= vma
->vm_file
;
250 struct shm_file_data
*sfd
= shm_file_data(file
);
252 if (sfd
->vm_ops
->set_policy
)
253 err
= sfd
->vm_ops
->set_policy(vma
, new);
257 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
260 struct file
*file
= vma
->vm_file
;
261 struct shm_file_data
*sfd
= shm_file_data(file
);
262 struct mempolicy
*pol
= NULL
;
264 if (sfd
->vm_ops
->get_policy
)
265 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
266 else if (vma
->vm_policy
)
267 pol
= vma
->vm_policy
;
269 pol
= current
->mempolicy
;
274 static int shm_mmap(struct file
* file
, struct vm_area_struct
* vma
)
276 struct shm_file_data
*sfd
= shm_file_data(file
);
279 ret
= sfd
->file
->f_op
->mmap(sfd
->file
, vma
);
282 sfd
->vm_ops
= vma
->vm_ops
;
284 BUG_ON(!sfd
->vm_ops
->fault
);
286 vma
->vm_ops
= &shm_vm_ops
;
292 static int shm_release(struct inode
*ino
, struct file
*file
)
294 struct shm_file_data
*sfd
= shm_file_data(file
);
297 shm_file_data(file
) = NULL
;
302 static int shm_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
304 int (*fsync
) (struct file
*, struct dentry
*, int datasync
);
305 struct shm_file_data
*sfd
= shm_file_data(file
);
308 fsync
= sfd
->file
->f_op
->fsync
;
310 ret
= fsync(sfd
->file
, sfd
->file
->f_path
.dentry
, datasync
);
314 static unsigned long shm_get_unmapped_area(struct file
*file
,
315 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
318 struct shm_file_data
*sfd
= shm_file_data(file
);
319 return get_unmapped_area(sfd
->file
, addr
, len
, pgoff
, flags
);
322 int is_file_shm_hugepages(struct file
*file
)
326 if (file
->f_op
== &shm_file_operations
) {
327 struct shm_file_data
*sfd
;
328 sfd
= shm_file_data(file
);
329 ret
= is_file_hugepages(sfd
->file
);
334 static const struct file_operations shm_file_operations
= {
337 .release
= shm_release
,
338 .get_unmapped_area
= shm_get_unmapped_area
,
341 static struct vm_operations_struct shm_vm_ops
= {
342 .open
= shm_open
, /* callback for a new vm-area open */
343 .close
= shm_close
, /* callback for when the vm-area is released */
345 #if defined(CONFIG_NUMA)
346 .set_policy
= shm_set_policy
,
347 .get_policy
= shm_get_policy
,
352 * newseg - Create a new shared memory segment
354 * @params: ptr to the structure that contains key, size and shmflg
356 * Called with shm_ids.mutex held
359 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
361 key_t key
= params
->key
;
362 int shmflg
= params
->flg
;
363 size_t size
= params
->u
.size
;
365 struct shmid_kernel
*shp
;
366 int numpages
= (size
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
371 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
374 if (ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
377 shp
= ipc_rcu_alloc(sizeof(*shp
));
381 shp
->shm_perm
.key
= key
;
382 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
383 shp
->mlock_user
= NULL
;
385 shp
->shm_perm
.security
= NULL
;
386 error
= security_shm_alloc(shp
);
392 sprintf (name
, "SYSV%08x", key
);
393 if (shmflg
& SHM_HUGETLB
) {
394 /* hugetlb_file_setup takes care of mlock user accounting */
395 file
= hugetlb_file_setup(name
, size
);
396 shp
->mlock_user
= current
->user
;
398 int acctflag
= VM_ACCOUNT
;
400 * Do not allow no accounting for OVERCOMMIT_NEVER, even
403 if ((shmflg
& SHM_NORESERVE
) &&
404 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
406 file
= shmem_file_setup(name
, size
, acctflag
);
408 error
= PTR_ERR(file
);
413 id
= shm_addid(ns
, shp
);
417 shp
->shm_cprid
= task_tgid_vnr(current
);
419 shp
->shm_atim
= shp
->shm_dtim
= 0;
420 shp
->shm_ctim
= get_seconds();
421 shp
->shm_segsz
= size
;
423 shp
->shm_perm
.id
= shm_buildid(ns
, id
, shp
->shm_perm
.seq
);
424 shp
->shm_file
= file
;
426 * shmid gets reported as "inode#" in /proc/pid/maps.
427 * proc-ps tools use this. Changing this will break them.
429 file
->f_dentry
->d_inode
->i_ino
= shp
->shm_perm
.id
;
431 ns
->shm_tot
+= numpages
;
432 error
= shp
->shm_perm
.id
;
439 security_shm_free(shp
);
445 * Called with shm_ids.mutex and ipcp locked.
447 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
449 struct shmid_kernel
*shp
;
451 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
452 return security_shm_associate(shp
, shmflg
);
456 * Called with shm_ids.mutex and ipcp locked.
458 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
459 struct ipc_params
*params
)
461 struct shmid_kernel
*shp
;
463 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
464 if (shp
->shm_segsz
< params
->u
.size
)
470 asmlinkage
long sys_shmget (key_t key
, size_t size
, int shmflg
)
472 struct ipc_namespace
*ns
;
473 struct ipc_ops shm_ops
;
474 struct ipc_params shm_params
;
476 ns
= current
->nsproxy
->ipc_ns
;
478 shm_ops
.getnew
= newseg
;
479 shm_ops
.associate
= shm_security
;
480 shm_ops
.more_checks
= shm_more_checks
;
482 shm_params
.key
= key
;
483 shm_params
.flg
= shmflg
;
484 shm_params
.u
.size
= size
;
486 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
489 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
493 return copy_to_user(buf
, in
, sizeof(*in
));
498 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
499 out
.shm_segsz
= in
->shm_segsz
;
500 out
.shm_atime
= in
->shm_atime
;
501 out
.shm_dtime
= in
->shm_dtime
;
502 out
.shm_ctime
= in
->shm_ctime
;
503 out
.shm_cpid
= in
->shm_cpid
;
504 out
.shm_lpid
= in
->shm_lpid
;
505 out
.shm_nattch
= in
->shm_nattch
;
507 return copy_to_user(buf
, &out
, sizeof(out
));
520 static inline unsigned long copy_shmid_from_user(struct shm_setbuf
*out
, void __user
*buf
, int version
)
525 struct shmid64_ds tbuf
;
527 if (copy_from_user(&tbuf
, buf
, sizeof(tbuf
)))
530 out
->uid
= tbuf
.shm_perm
.uid
;
531 out
->gid
= tbuf
.shm_perm
.gid
;
532 out
->mode
= tbuf
.shm_perm
.mode
;
538 struct shmid_ds tbuf_old
;
540 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
543 out
->uid
= tbuf_old
.shm_perm
.uid
;
544 out
->gid
= tbuf_old
.shm_perm
.gid
;
545 out
->mode
= tbuf_old
.shm_perm
.mode
;
554 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
558 return copy_to_user(buf
, in
, sizeof(*in
));
563 if(in
->shmmax
> INT_MAX
)
564 out
.shmmax
= INT_MAX
;
566 out
.shmmax
= (int)in
->shmmax
;
568 out
.shmmin
= in
->shmmin
;
569 out
.shmmni
= in
->shmmni
;
570 out
.shmseg
= in
->shmseg
;
571 out
.shmall
= in
->shmall
;
573 return copy_to_user(buf
, &out
, sizeof(out
));
581 * Called with shm_ids.mutex held
583 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
592 in_use
= shm_ids(ns
).in_use
;
594 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
595 struct shmid_kernel
*shp
;
598 shp
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
602 inode
= shp
->shm_file
->f_path
.dentry
->d_inode
;
604 if (is_file_hugepages(shp
->shm_file
)) {
605 struct address_space
*mapping
= inode
->i_mapping
;
606 *rss
+= (HPAGE_SIZE
/PAGE_SIZE
)*mapping
->nrpages
;
608 struct shmem_inode_info
*info
= SHMEM_I(inode
);
609 spin_lock(&info
->lock
);
610 *rss
+= inode
->i_mapping
->nrpages
;
611 *swp
+= info
->swapped
;
612 spin_unlock(&info
->lock
);
619 asmlinkage
long sys_shmctl (int shmid
, int cmd
, struct shmid_ds __user
*buf
)
621 struct shm_setbuf setbuf
;
622 struct shmid_kernel
*shp
;
624 struct ipc_namespace
*ns
;
626 if (cmd
< 0 || shmid
< 0) {
631 version
= ipc_parse_version(&cmd
);
632 ns
= current
->nsproxy
->ipc_ns
;
634 switch (cmd
) { /* replace with proc interface ? */
637 struct shminfo64 shminfo
;
639 err
= security_shm_shmctl(NULL
, cmd
);
643 memset(&shminfo
,0,sizeof(shminfo
));
644 shminfo
.shmmni
= shminfo
.shmseg
= ns
->shm_ctlmni
;
645 shminfo
.shmmax
= ns
->shm_ctlmax
;
646 shminfo
.shmall
= ns
->shm_ctlall
;
648 shminfo
.shmmin
= SHMMIN
;
649 if(copy_shminfo_to_user (buf
, &shminfo
, version
))
652 mutex_lock(&shm_ids(ns
).mutex
);
653 err
= ipc_get_maxid(&shm_ids(ns
));
654 mutex_unlock(&shm_ids(ns
).mutex
);
662 struct shm_info shm_info
;
664 err
= security_shm_shmctl(NULL
, cmd
);
668 memset(&shm_info
,0,sizeof(shm_info
));
669 mutex_lock(&shm_ids(ns
).mutex
);
670 shm_info
.used_ids
= shm_ids(ns
).in_use
;
671 shm_get_stat (ns
, &shm_info
.shm_rss
, &shm_info
.shm_swp
);
672 shm_info
.shm_tot
= ns
->shm_tot
;
673 shm_info
.swap_attempts
= 0;
674 shm_info
.swap_successes
= 0;
675 err
= ipc_get_maxid(&shm_ids(ns
));
676 mutex_unlock(&shm_ids(ns
).mutex
);
677 if(copy_to_user (buf
, &shm_info
, sizeof(shm_info
))) {
682 err
= err
< 0 ? 0 : err
;
688 struct shmid64_ds tbuf
;
696 if (cmd
== SHM_STAT
) {
697 shp
= shm_lock(ns
, shmid
);
702 result
= shp
->shm_perm
.id
;
704 shp
= shm_lock_check(ns
, shmid
);
712 if (ipcperms (&shp
->shm_perm
, S_IRUGO
))
714 err
= security_shm_shmctl(shp
, cmd
);
717 memset(&tbuf
, 0, sizeof(tbuf
));
718 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
.shm_perm
);
719 tbuf
.shm_segsz
= shp
->shm_segsz
;
720 tbuf
.shm_atime
= shp
->shm_atim
;
721 tbuf
.shm_dtime
= shp
->shm_dtim
;
722 tbuf
.shm_ctime
= shp
->shm_ctim
;
723 tbuf
.shm_cpid
= shp
->shm_cprid
;
724 tbuf
.shm_lpid
= shp
->shm_lprid
;
725 tbuf
.shm_nattch
= shp
->shm_nattch
;
727 if(copy_shmid_to_user (buf
, &tbuf
, version
))
736 shp
= shm_lock_check(ns
, shmid
);
742 err
= audit_ipc_obj(&(shp
->shm_perm
));
746 if (!capable(CAP_IPC_LOCK
)) {
748 if (current
->euid
!= shp
->shm_perm
.uid
&&
749 current
->euid
!= shp
->shm_perm
.cuid
)
751 if (cmd
== SHM_LOCK
&&
752 !current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
)
756 err
= security_shm_shmctl(shp
, cmd
);
761 struct user_struct
* user
= current
->user
;
762 if (!is_file_hugepages(shp
->shm_file
)) {
763 err
= shmem_lock(shp
->shm_file
, 1, user
);
764 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)){
765 shp
->shm_perm
.mode
|= SHM_LOCKED
;
766 shp
->mlock_user
= user
;
769 } else if (!is_file_hugepages(shp
->shm_file
)) {
770 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
771 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
772 shp
->mlock_user
= NULL
;
780 * We cannot simply remove the file. The SVID states
781 * that the block remains until the last person
782 * detaches from it, then is deleted. A shmat() on
783 * an RMID segment is legal in older Linux and if
784 * we change it apps break...
786 * Instead we set a destroyed flag, and then blow
787 * the name away when the usage hits zero.
789 mutex_lock(&shm_ids(ns
).mutex
);
790 shp
= shm_lock_check(ns
, shmid
);
796 err
= audit_ipc_obj(&(shp
->shm_perm
));
800 if (current
->euid
!= shp
->shm_perm
.uid
&&
801 current
->euid
!= shp
->shm_perm
.cuid
&&
802 !capable(CAP_SYS_ADMIN
)) {
807 err
= security_shm_shmctl(shp
, cmd
);
811 do_shm_rmid(ns
, shp
);
812 mutex_unlock(&shm_ids(ns
).mutex
);
823 if (copy_shmid_from_user (&setbuf
, buf
, version
)) {
827 mutex_lock(&shm_ids(ns
).mutex
);
828 shp
= shm_lock_check(ns
, shmid
);
833 err
= audit_ipc_obj(&(shp
->shm_perm
));
836 err
= audit_ipc_set_perm(0, setbuf
.uid
, setbuf
.gid
, setbuf
.mode
);
840 if (current
->euid
!= shp
->shm_perm
.uid
&&
841 current
->euid
!= shp
->shm_perm
.cuid
&&
842 !capable(CAP_SYS_ADMIN
)) {
846 err
= security_shm_shmctl(shp
, cmd
);
850 shp
->shm_perm
.uid
= setbuf
.uid
;
851 shp
->shm_perm
.gid
= setbuf
.gid
;
852 shp
->shm_perm
.mode
= (shp
->shm_perm
.mode
& ~S_IRWXUGO
)
853 | (setbuf
.mode
& S_IRWXUGO
);
854 shp
->shm_ctim
= get_seconds();
867 mutex_unlock(&shm_ids(ns
).mutex
);
876 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
878 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
879 * "raddr" thing points to kernel space, and there has to be a wrapper around
882 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
, ulong
*raddr
)
884 struct shmid_kernel
*shp
;
892 unsigned long user_addr
;
893 struct ipc_namespace
*ns
;
894 struct shm_file_data
*sfd
;
901 else if ((addr
= (ulong
)shmaddr
)) {
902 if (addr
& (SHMLBA
-1)) {
903 if (shmflg
& SHM_RND
)
904 addr
&= ~(SHMLBA
-1); /* round down */
906 #ifndef __ARCH_FORCE_SHMLBA
907 if (addr
& ~PAGE_MASK
)
911 flags
= MAP_SHARED
| MAP_FIXED
;
913 if ((shmflg
& SHM_REMAP
))
919 if (shmflg
& SHM_RDONLY
) {
924 prot
= PROT_READ
| PROT_WRITE
;
925 acc_mode
= S_IRUGO
| S_IWUGO
;
926 f_mode
= FMODE_READ
| FMODE_WRITE
;
928 if (shmflg
& SHM_EXEC
) {
934 * We cannot rely on the fs check since SYSV IPC does have an
935 * additional creator id...
937 ns
= current
->nsproxy
->ipc_ns
;
938 shp
= shm_lock_check(ns
, shmid
);
945 if (ipcperms(&shp
->shm_perm
, acc_mode
))
948 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
952 path
.dentry
= dget(shp
->shm_file
->f_path
.dentry
);
953 path
.mnt
= shp
->shm_file
->f_path
.mnt
;
955 size
= i_size_read(path
.dentry
->d_inode
);
959 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
965 file
= alloc_file(path
.mnt
, path
.dentry
, f_mode
, &shm_file_operations
);
969 file
->private_data
= sfd
;
970 file
->f_mapping
= shp
->shm_file
->f_mapping
;
971 sfd
->id
= shp
->shm_perm
.id
;
972 sfd
->ns
= get_ipc_ns(ns
);
973 sfd
->file
= shp
->shm_file
;
976 down_write(¤t
->mm
->mmap_sem
);
977 if (addr
&& !(shmflg
& SHM_REMAP
)) {
979 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
982 * If shm segment goes below stack, make sure there is some
983 * space left for the stack to grow (at least 4 pages).
985 if (addr
< current
->mm
->start_stack
&&
986 addr
> current
->mm
->start_stack
- size
- PAGE_SIZE
* 5)
990 user_addr
= do_mmap (file
, addr
, size
, prot
, flags
, 0);
993 if (IS_ERR_VALUE(user_addr
))
994 err
= (long)user_addr
;
996 up_write(¤t
->mm
->mmap_sem
);
1001 mutex_lock(&shm_ids(ns
).mutex
);
1002 shp
= shm_lock(ns
, shmid
);
1003 BUG_ON(IS_ERR(shp
));
1005 if(shp
->shm_nattch
== 0 &&
1006 shp
->shm_perm
.mode
& SHM_DEST
)
1007 shm_destroy(ns
, shp
);
1010 mutex_unlock(&shm_ids(ns
).mutex
);
1026 asmlinkage
long sys_shmat(int shmid
, char __user
*shmaddr
, int shmflg
)
1031 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
);
1034 force_successful_syscall_return();
1039 * detach and kill segment if marked destroyed.
1040 * The work is done in shm_close.
1042 asmlinkage
long sys_shmdt(char __user
*shmaddr
)
1044 struct mm_struct
*mm
= current
->mm
;
1045 struct vm_area_struct
*vma
, *next
;
1046 unsigned long addr
= (unsigned long)shmaddr
;
1048 int retval
= -EINVAL
;
1050 if (addr
& ~PAGE_MASK
)
1053 down_write(&mm
->mmap_sem
);
1056 * This function tries to be smart and unmap shm segments that
1057 * were modified by partial mlock or munmap calls:
1058 * - It first determines the size of the shm segment that should be
1059 * unmapped: It searches for a vma that is backed by shm and that
1060 * started at address shmaddr. It records it's size and then unmaps
1062 * - Then it unmaps all shm vmas that started at shmaddr and that
1063 * are within the initially determined size.
1064 * Errors from do_munmap are ignored: the function only fails if
1065 * it's called with invalid parameters or if it's called to unmap
1066 * a part of a vma. Both calls in this function are for full vmas,
1067 * the parameters are directly copied from the vma itself and always
1068 * valid - therefore do_munmap cannot fail. (famous last words?)
1071 * If it had been mremap()'d, the starting address would not
1072 * match the usual checks anyway. So assume all vma's are
1073 * above the starting address given.
1075 vma
= find_vma(mm
, addr
);
1078 next
= vma
->vm_next
;
1081 * Check if the starting address would match, i.e. it's
1082 * a fragment created by mprotect() and/or munmap(), or it
1083 * otherwise it starts at this address with no hassles.
1085 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1086 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1089 size
= vma
->vm_file
->f_path
.dentry
->d_inode
->i_size
;
1090 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1092 * We discovered the size of the shm segment, so
1093 * break out of here and fall through to the next
1094 * loop that uses the size information to stop
1095 * searching for matching vma's.
1105 * We need look no further than the maximum address a fragment
1106 * could possibly have landed at. Also cast things to loff_t to
1107 * prevent overflows and make comparisions vs. equal-width types.
1109 size
= PAGE_ALIGN(size
);
1110 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1111 next
= vma
->vm_next
;
1113 /* finding a matching vma now does not alter retval */
1114 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1115 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
)
1117 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1121 up_write(&mm
->mmap_sem
);
1125 #ifdef CONFIG_PROC_FS
1126 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1128 struct shmid_kernel
*shp
= it
;
1131 #define SMALL_STRING "%10d %10d %4o %10u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1132 #define BIG_STRING "%10d %10d %4o %21u %5u %5u %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1134 if (sizeof(size_t) <= sizeof(int))
1135 format
= SMALL_STRING
;
1137 format
= BIG_STRING
;
1138 return seq_printf(s
, format
,