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>
43 #include <asm/uaccess.h>
47 struct shm_file_data
{
49 struct ipc_namespace
*ns
;
51 const struct vm_operations_struct
*vm_ops
;
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56 static const struct file_operations shm_file_operations
;
57 static const struct vm_operations_struct shm_vm_ops
;
59 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
61 #define shm_unlock(shp) \
62 ipc_unlock(&(shp)->shm_perm)
64 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
65 static void shm_open(struct vm_area_struct
*vma
);
66 static void shm_close(struct vm_area_struct
*vma
);
67 static void shm_destroy (struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
69 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
72 void shm_init_ns(struct ipc_namespace
*ns
)
74 ns
->shm_ctlmax
= SHMMAX
;
75 ns
->shm_ctlall
= SHMALL
;
76 ns
->shm_ctlmni
= SHMMNI
;
77 ns
->shm_rmid_forced
= 0;
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 static int __init
ipc_ns_init(void)
110 shm_init_ns(&init_ipc_ns
);
114 pure_initcall(ipc_ns_init
);
116 void __init
shm_init (void)
118 ipc_init_proc_interface("sysvipc/shm",
119 #if BITS_PER_LONG <= 32
120 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
122 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
124 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
128 * shm_lock_(check_) routines are called in the paths where the rw_mutex
129 * is not necessarily held.
131 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
133 struct kern_ipc_perm
*ipcp
= ipc_lock(&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_lock_by_ptr(struct shmid_kernel
*ipcp
)
144 spin_lock(&ipcp
->shm_perm
.lock
);
147 static inline struct shmid_kernel
*shm_lock_check(struct ipc_namespace
*ns
,
150 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&shm_ids(ns
), id
);
153 return (struct shmid_kernel
*)ipcp
;
155 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
158 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
160 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
164 /* This is called by fork, once for every shm attach. */
165 static void shm_open(struct vm_area_struct
*vma
)
167 struct file
*file
= vma
->vm_file
;
168 struct shm_file_data
*sfd
= shm_file_data(file
);
169 struct shmid_kernel
*shp
;
171 shp
= shm_lock(sfd
->ns
, sfd
->id
);
173 shp
->shm_atim
= get_seconds();
174 shp
->shm_lprid
= task_tgid_vnr(current
);
180 * shm_destroy - free the struct shmid_kernel
183 * @shp: struct to free
185 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
186 * but returns with shp unlocked and freed.
188 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
190 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
193 if (!is_file_hugepages(shp
->shm_file
))
194 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
195 else if (shp
->mlock_user
)
196 user_shm_unlock(file_inode(shp
->shm_file
)->i_size
,
198 fput (shp
->shm_file
);
199 security_shm_free(shp
);
204 * shm_may_destroy - identifies whether shm segment should be destroyed now
206 * Returns true if and only if there are no active users of the segment and
207 * one of the following is true:
209 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
211 * 2) sysctl kernel.shm_rmid_forced is set to 1.
213 static bool shm_may_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
215 return (shp
->shm_nattch
== 0) &&
216 (ns
->shm_rmid_forced
||
217 (shp
->shm_perm
.mode
& SHM_DEST
));
221 * remove the attach descriptor vma.
222 * free memory for segment if it is marked destroyed.
223 * The descriptor has already been removed from the current->mm->mmap list
224 * and will later be kfree()d.
226 static void shm_close(struct vm_area_struct
*vma
)
228 struct file
* file
= vma
->vm_file
;
229 struct shm_file_data
*sfd
= shm_file_data(file
);
230 struct shmid_kernel
*shp
;
231 struct ipc_namespace
*ns
= sfd
->ns
;
233 down_write(&shm_ids(ns
).rw_mutex
);
234 /* remove from the list of attaches of the shm segment */
235 shp
= shm_lock(ns
, sfd
->id
);
237 shp
->shm_lprid
= task_tgid_vnr(current
);
238 shp
->shm_dtim
= get_seconds();
240 if (shm_may_destroy(ns
, shp
))
241 shm_destroy(ns
, shp
);
244 up_write(&shm_ids(ns
).rw_mutex
);
247 /* Called with ns->shm_ids(ns).rw_mutex locked */
248 static int shm_try_destroy_current(int id
, void *p
, void *data
)
250 struct ipc_namespace
*ns
= data
;
251 struct kern_ipc_perm
*ipcp
= p
;
252 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
254 if (shp
->shm_creator
!= current
)
258 * Mark it as orphaned to destroy the segment when
259 * kernel.shm_rmid_forced is changed.
260 * It is noop if the following shm_may_destroy() returns true.
262 shp
->shm_creator
= NULL
;
265 * Don't even try to destroy it. If shm_rmid_forced=0 and IPC_RMID
266 * is not set, it shouldn't be deleted here.
268 if (!ns
->shm_rmid_forced
)
271 if (shm_may_destroy(ns
, shp
)) {
272 shm_lock_by_ptr(shp
);
273 shm_destroy(ns
, shp
);
278 /* Called with ns->shm_ids(ns).rw_mutex locked */
279 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
281 struct ipc_namespace
*ns
= data
;
282 struct kern_ipc_perm
*ipcp
= p
;
283 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
286 * We want to destroy segments without users and with already
287 * exit'ed originating process.
289 * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
291 if (shp
->shm_creator
!= NULL
)
294 if (shm_may_destroy(ns
, shp
)) {
295 shm_lock_by_ptr(shp
);
296 shm_destroy(ns
, shp
);
301 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
303 down_write(&shm_ids(ns
).rw_mutex
);
304 if (shm_ids(ns
).in_use
)
305 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
306 up_write(&shm_ids(ns
).rw_mutex
);
310 void exit_shm(struct task_struct
*task
)
312 struct ipc_namespace
*ns
= task
->nsproxy
->ipc_ns
;
314 if (shm_ids(ns
).in_use
== 0)
317 /* Destroy all already created segments, but not mapped yet */
318 down_write(&shm_ids(ns
).rw_mutex
);
319 if (shm_ids(ns
).in_use
)
320 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_current
, ns
);
321 up_write(&shm_ids(ns
).rw_mutex
);
324 static int shm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
326 struct file
*file
= vma
->vm_file
;
327 struct shm_file_data
*sfd
= shm_file_data(file
);
329 return sfd
->vm_ops
->fault(vma
, vmf
);
333 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
335 struct file
*file
= vma
->vm_file
;
336 struct shm_file_data
*sfd
= shm_file_data(file
);
338 if (sfd
->vm_ops
->set_policy
)
339 err
= sfd
->vm_ops
->set_policy(vma
, new);
343 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
346 struct file
*file
= vma
->vm_file
;
347 struct shm_file_data
*sfd
= shm_file_data(file
);
348 struct mempolicy
*pol
= NULL
;
350 if (sfd
->vm_ops
->get_policy
)
351 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
352 else if (vma
->vm_policy
)
353 pol
= vma
->vm_policy
;
359 static int shm_mmap(struct file
* file
, struct vm_area_struct
* vma
)
361 struct shm_file_data
*sfd
= shm_file_data(file
);
364 ret
= sfd
->file
->f_op
->mmap(sfd
->file
, vma
);
367 sfd
->vm_ops
= vma
->vm_ops
;
369 BUG_ON(!sfd
->vm_ops
->fault
);
371 vma
->vm_ops
= &shm_vm_ops
;
377 static int shm_release(struct inode
*ino
, struct file
*file
)
379 struct shm_file_data
*sfd
= shm_file_data(file
);
382 shm_file_data(file
) = NULL
;
387 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
389 struct shm_file_data
*sfd
= shm_file_data(file
);
391 if (!sfd
->file
->f_op
->fsync
)
393 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
396 static long shm_fallocate(struct file
*file
, int mode
, loff_t offset
,
399 struct shm_file_data
*sfd
= shm_file_data(file
);
401 if (!sfd
->file
->f_op
->fallocate
)
403 return sfd
->file
->f_op
->fallocate(file
, mode
, offset
, len
);
406 static unsigned long shm_get_unmapped_area(struct file
*file
,
407 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
410 struct shm_file_data
*sfd
= shm_file_data(file
);
411 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
415 static const struct file_operations shm_file_operations
= {
418 .release
= shm_release
,
420 .get_unmapped_area
= shm_get_unmapped_area
,
422 .llseek
= noop_llseek
,
423 .fallocate
= shm_fallocate
,
426 static const struct file_operations shm_file_operations_huge
= {
429 .release
= shm_release
,
430 .get_unmapped_area
= shm_get_unmapped_area
,
431 .llseek
= noop_llseek
,
432 .fallocate
= shm_fallocate
,
435 int is_file_shm_hugepages(struct file
*file
)
437 return file
->f_op
== &shm_file_operations_huge
;
440 static const struct vm_operations_struct shm_vm_ops
= {
441 .open
= shm_open
, /* callback for a new vm-area open */
442 .close
= shm_close
, /* callback for when the vm-area is released */
444 #if defined(CONFIG_NUMA)
445 .set_policy
= shm_set_policy
,
446 .get_policy
= shm_get_policy
,
451 * newseg - Create a new shared memory segment
453 * @params: ptr to the structure that contains key, size and shmflg
455 * Called with shm_ids.rw_mutex held as a writer.
458 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
460 key_t key
= params
->key
;
461 int shmflg
= params
->flg
;
462 size_t size
= params
->u
.size
;
464 struct shmid_kernel
*shp
;
465 size_t numpages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
469 vm_flags_t acctflag
= 0;
471 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
474 if (ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
477 shp
= ipc_rcu_alloc(sizeof(*shp
));
481 shp
->shm_perm
.key
= key
;
482 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
483 shp
->mlock_user
= NULL
;
485 shp
->shm_perm
.security
= NULL
;
486 error
= security_shm_alloc(shp
);
492 sprintf (name
, "SYSV%08x", key
);
493 if (shmflg
& SHM_HUGETLB
) {
494 struct hstate
*hs
= hstate_sizelog((shmflg
>> SHM_HUGE_SHIFT
)
496 size_t hugesize
= ALIGN(size
, huge_page_size(hs
));
498 /* hugetlb_file_setup applies strict accounting */
499 if (shmflg
& SHM_NORESERVE
)
500 acctflag
= VM_NORESERVE
;
501 file
= hugetlb_file_setup(name
, hugesize
, acctflag
,
502 &shp
->mlock_user
, HUGETLB_SHMFS_INODE
,
503 (shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
506 * Do not allow no accounting for OVERCOMMIT_NEVER, even
509 if ((shmflg
& SHM_NORESERVE
) &&
510 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
511 acctflag
= VM_NORESERVE
;
512 file
= shmem_file_setup(name
, size
, acctflag
);
514 error
= PTR_ERR(file
);
518 id
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
524 shp
->shm_cprid
= task_tgid_vnr(current
);
526 shp
->shm_atim
= shp
->shm_dtim
= 0;
527 shp
->shm_ctim
= get_seconds();
528 shp
->shm_segsz
= size
;
530 shp
->shm_file
= file
;
531 shp
->shm_creator
= current
;
533 * shmid gets reported as "inode#" in /proc/pid/maps.
534 * proc-ps tools use this. Changing this will break them.
536 file_inode(file
)->i_ino
= shp
->shm_perm
.id
;
538 ns
->shm_tot
+= numpages
;
539 error
= shp
->shm_perm
.id
;
544 if (is_file_hugepages(file
) && shp
->mlock_user
)
545 user_shm_unlock(size
, shp
->mlock_user
);
548 security_shm_free(shp
);
554 * Called with shm_ids.rw_mutex and ipcp locked.
556 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
558 struct shmid_kernel
*shp
;
560 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
561 return security_shm_associate(shp
, shmflg
);
565 * Called with shm_ids.rw_mutex and ipcp locked.
567 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
568 struct ipc_params
*params
)
570 struct shmid_kernel
*shp
;
572 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
573 if (shp
->shm_segsz
< params
->u
.size
)
579 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
581 struct ipc_namespace
*ns
;
582 struct ipc_ops shm_ops
;
583 struct ipc_params shm_params
;
585 ns
= current
->nsproxy
->ipc_ns
;
587 shm_ops
.getnew
= newseg
;
588 shm_ops
.associate
= shm_security
;
589 shm_ops
.more_checks
= shm_more_checks
;
591 shm_params
.key
= key
;
592 shm_params
.flg
= shmflg
;
593 shm_params
.u
.size
= size
;
595 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
598 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
602 return copy_to_user(buf
, in
, sizeof(*in
));
607 memset(&out
, 0, sizeof(out
));
608 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
609 out
.shm_segsz
= in
->shm_segsz
;
610 out
.shm_atime
= in
->shm_atime
;
611 out
.shm_dtime
= in
->shm_dtime
;
612 out
.shm_ctime
= in
->shm_ctime
;
613 out
.shm_cpid
= in
->shm_cpid
;
614 out
.shm_lpid
= in
->shm_lpid
;
615 out
.shm_nattch
= in
->shm_nattch
;
617 return copy_to_user(buf
, &out
, sizeof(out
));
624 static inline unsigned long
625 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
629 if (copy_from_user(out
, buf
, sizeof(*out
)))
634 struct shmid_ds tbuf_old
;
636 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
639 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
640 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
641 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
650 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
654 return copy_to_user(buf
, in
, sizeof(*in
));
659 if(in
->shmmax
> INT_MAX
)
660 out
.shmmax
= INT_MAX
;
662 out
.shmmax
= (int)in
->shmmax
;
664 out
.shmmin
= in
->shmmin
;
665 out
.shmmni
= in
->shmmni
;
666 out
.shmseg
= in
->shmseg
;
667 out
.shmall
= in
->shmall
;
669 return copy_to_user(buf
, &out
, sizeof(out
));
677 * Calculate and add used RSS and swap pages of a shm.
678 * Called with shm_ids.rw_mutex held as a reader
680 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
681 unsigned long *rss_add
, unsigned long *swp_add
)
685 inode
= file_inode(shp
->shm_file
);
687 if (is_file_hugepages(shp
->shm_file
)) {
688 struct address_space
*mapping
= inode
->i_mapping
;
689 struct hstate
*h
= hstate_file(shp
->shm_file
);
690 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
693 struct shmem_inode_info
*info
= SHMEM_I(inode
);
694 spin_lock(&info
->lock
);
695 *rss_add
+= inode
->i_mapping
->nrpages
;
696 *swp_add
+= info
->swapped
;
697 spin_unlock(&info
->lock
);
699 *rss_add
+= inode
->i_mapping
->nrpages
;
705 * Called with shm_ids.rw_mutex held as a reader
707 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
716 in_use
= shm_ids(ns
).in_use
;
718 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
719 struct kern_ipc_perm
*ipc
;
720 struct shmid_kernel
*shp
;
722 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
725 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
727 shm_add_rss_swap(shp
, rss
, swp
);
734 * This function handles some shmctl commands which require the rw_mutex
735 * to be held in write mode.
736 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
738 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
739 struct shmid_ds __user
*buf
, int version
)
741 struct kern_ipc_perm
*ipcp
;
742 struct shmid64_ds shmid64
;
743 struct shmid_kernel
*shp
;
746 if (cmd
== IPC_SET
) {
747 if (copy_shmid_from_user(&shmid64
, buf
, version
))
751 ipcp
= ipcctl_pre_down(ns
, &shm_ids(ns
), shmid
, cmd
,
752 &shmid64
.shm_perm
, 0);
754 return PTR_ERR(ipcp
);
756 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
758 err
= security_shm_shmctl(shp
, cmd
);
763 do_shm_rmid(ns
, ipcp
);
766 err
= ipc_update_perm(&shmid64
.shm_perm
, ipcp
);
769 shp
->shm_ctim
= get_seconds();
777 up_write(&shm_ids(ns
).rw_mutex
);
781 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
783 struct shmid_kernel
*shp
;
785 struct ipc_namespace
*ns
;
787 if (cmd
< 0 || shmid
< 0) {
792 version
= ipc_parse_version(&cmd
);
793 ns
= current
->nsproxy
->ipc_ns
;
795 switch (cmd
) { /* replace with proc interface ? */
798 struct shminfo64 shminfo
;
800 err
= security_shm_shmctl(NULL
, cmd
);
804 memset(&shminfo
, 0, sizeof(shminfo
));
805 shminfo
.shmmni
= shminfo
.shmseg
= ns
->shm_ctlmni
;
806 shminfo
.shmmax
= ns
->shm_ctlmax
;
807 shminfo
.shmall
= ns
->shm_ctlall
;
809 shminfo
.shmmin
= SHMMIN
;
810 if(copy_shminfo_to_user (buf
, &shminfo
, version
))
813 down_read(&shm_ids(ns
).rw_mutex
);
814 err
= ipc_get_maxid(&shm_ids(ns
));
815 up_read(&shm_ids(ns
).rw_mutex
);
823 struct shm_info shm_info
;
825 err
= security_shm_shmctl(NULL
, cmd
);
829 memset(&shm_info
, 0, sizeof(shm_info
));
830 down_read(&shm_ids(ns
).rw_mutex
);
831 shm_info
.used_ids
= shm_ids(ns
).in_use
;
832 shm_get_stat (ns
, &shm_info
.shm_rss
, &shm_info
.shm_swp
);
833 shm_info
.shm_tot
= ns
->shm_tot
;
834 shm_info
.swap_attempts
= 0;
835 shm_info
.swap_successes
= 0;
836 err
= ipc_get_maxid(&shm_ids(ns
));
837 up_read(&shm_ids(ns
).rw_mutex
);
838 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
))) {
843 err
= err
< 0 ? 0 : err
;
849 struct shmid64_ds tbuf
;
852 if (cmd
== SHM_STAT
) {
853 shp
= shm_lock(ns
, shmid
);
858 result
= shp
->shm_perm
.id
;
860 shp
= shm_lock_check(ns
, shmid
);
868 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
870 err
= security_shm_shmctl(shp
, cmd
);
873 memset(&tbuf
, 0, sizeof(tbuf
));
874 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
.shm_perm
);
875 tbuf
.shm_segsz
= shp
->shm_segsz
;
876 tbuf
.shm_atime
= shp
->shm_atim
;
877 tbuf
.shm_dtime
= shp
->shm_dtim
;
878 tbuf
.shm_ctime
= shp
->shm_ctim
;
879 tbuf
.shm_cpid
= shp
->shm_cprid
;
880 tbuf
.shm_lpid
= shp
->shm_lprid
;
881 tbuf
.shm_nattch
= shp
->shm_nattch
;
883 if(copy_shmid_to_user (buf
, &tbuf
, version
))
892 struct file
*shm_file
;
894 shp
= shm_lock_check(ns
, shmid
);
900 audit_ipc_obj(&(shp
->shm_perm
));
902 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
903 kuid_t euid
= current_euid();
905 if (!uid_eq(euid
, shp
->shm_perm
.uid
) &&
906 !uid_eq(euid
, shp
->shm_perm
.cuid
))
908 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
))
912 err
= security_shm_shmctl(shp
, cmd
);
916 shm_file
= shp
->shm_file
;
917 if (is_file_hugepages(shm_file
))
920 if (cmd
== SHM_LOCK
) {
921 struct user_struct
*user
= current_user();
922 err
= shmem_lock(shm_file
, 1, user
);
923 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)) {
924 shp
->shm_perm
.mode
|= SHM_LOCKED
;
925 shp
->mlock_user
= user
;
931 if (!(shp
->shm_perm
.mode
& SHM_LOCKED
))
933 shmem_lock(shm_file
, 0, shp
->mlock_user
);
934 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
935 shp
->mlock_user
= NULL
;
938 shmem_unlock_mapping(shm_file
->f_mapping
);
944 err
= shmctl_down(ns
, shmid
, cmd
, buf
, version
);
957 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
959 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
960 * "raddr" thing points to kernel space, and there has to be a wrapper around
963 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
, ulong
*raddr
,
964 unsigned long shmlba
)
966 struct shmid_kernel
*shp
;
974 struct ipc_namespace
*ns
;
975 struct shm_file_data
*sfd
;
978 unsigned long populate
= 0;
983 else if ((addr
= (ulong
)shmaddr
)) {
984 if (addr
& (shmlba
- 1)) {
985 if (shmflg
& SHM_RND
)
986 addr
&= ~(shmlba
- 1); /* round down */
988 #ifndef __ARCH_FORCE_SHMLBA
989 if (addr
& ~PAGE_MASK
)
993 flags
= MAP_SHARED
| MAP_FIXED
;
995 if ((shmflg
& SHM_REMAP
))
1001 if (shmflg
& SHM_RDONLY
) {
1004 f_mode
= FMODE_READ
;
1006 prot
= PROT_READ
| PROT_WRITE
;
1007 acc_mode
= S_IRUGO
| S_IWUGO
;
1008 f_mode
= FMODE_READ
| FMODE_WRITE
;
1010 if (shmflg
& SHM_EXEC
) {
1012 acc_mode
|= S_IXUGO
;
1016 * We cannot rely on the fs check since SYSV IPC does have an
1017 * additional creator id...
1019 ns
= current
->nsproxy
->ipc_ns
;
1020 shp
= shm_lock_check(ns
, shmid
);
1027 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
1030 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
1034 path
= shp
->shm_file
->f_path
;
1037 size
= i_size_read(path
.dentry
->d_inode
);
1041 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1043 goto out_put_dentry
;
1045 file
= alloc_file(&path
, f_mode
,
1046 is_file_hugepages(shp
->shm_file
) ?
1047 &shm_file_operations_huge
:
1048 &shm_file_operations
);
1049 err
= PTR_ERR(file
);
1053 file
->private_data
= sfd
;
1054 file
->f_mapping
= shp
->shm_file
->f_mapping
;
1055 sfd
->id
= shp
->shm_perm
.id
;
1056 sfd
->ns
= get_ipc_ns(ns
);
1057 sfd
->file
= shp
->shm_file
;
1060 err
= security_mmap_file(file
, prot
, flags
);
1064 down_write(¤t
->mm
->mmap_sem
);
1065 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1067 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1070 * If shm segment goes below stack, make sure there is some
1071 * space left for the stack to grow (at least 4 pages).
1073 if (addr
< current
->mm
->start_stack
&&
1074 addr
> current
->mm
->start_stack
- size
- PAGE_SIZE
* 5)
1078 addr
= do_mmap_pgoff(file
, addr
, size
, prot
, flags
, 0, &populate
);
1081 if (IS_ERR_VALUE(addr
))
1084 up_write(¤t
->mm
->mmap_sem
);
1086 mm_populate(addr
, populate
);
1092 down_write(&shm_ids(ns
).rw_mutex
);
1093 shp
= shm_lock(ns
, shmid
);
1094 BUG_ON(IS_ERR(shp
));
1096 if (shm_may_destroy(ns
, shp
))
1097 shm_destroy(ns
, shp
);
1100 up_write(&shm_ids(ns
).rw_mutex
);
1116 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1121 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
, SHMLBA
);
1124 force_successful_syscall_return();
1129 * detach and kill segment if marked destroyed.
1130 * The work is done in shm_close.
1132 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1134 struct mm_struct
*mm
= current
->mm
;
1135 struct vm_area_struct
*vma
;
1136 unsigned long addr
= (unsigned long)shmaddr
;
1137 int retval
= -EINVAL
;
1140 struct vm_area_struct
*next
;
1143 if (addr
& ~PAGE_MASK
)
1146 down_write(&mm
->mmap_sem
);
1149 * This function tries to be smart and unmap shm segments that
1150 * were modified by partial mlock or munmap calls:
1151 * - It first determines the size of the shm segment that should be
1152 * unmapped: It searches for a vma that is backed by shm and that
1153 * started at address shmaddr. It records it's size and then unmaps
1155 * - Then it unmaps all shm vmas that started at shmaddr and that
1156 * are within the initially determined size.
1157 * Errors from do_munmap are ignored: the function only fails if
1158 * it's called with invalid parameters or if it's called to unmap
1159 * a part of a vma. Both calls in this function are for full vmas,
1160 * the parameters are directly copied from the vma itself and always
1161 * valid - therefore do_munmap cannot fail. (famous last words?)
1164 * If it had been mremap()'d, the starting address would not
1165 * match the usual checks anyway. So assume all vma's are
1166 * above the starting address given.
1168 vma
= find_vma(mm
, addr
);
1172 next
= vma
->vm_next
;
1175 * Check if the starting address would match, i.e. it's
1176 * a fragment created by mprotect() and/or munmap(), or it
1177 * otherwise it starts at this address with no hassles.
1179 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1180 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1183 size
= file_inode(vma
->vm_file
)->i_size
;
1184 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1186 * We discovered the size of the shm segment, so
1187 * break out of here and fall through to the next
1188 * loop that uses the size information to stop
1189 * searching for matching vma's.
1199 * We need look no further than the maximum address a fragment
1200 * could possibly have landed at. Also cast things to loff_t to
1201 * prevent overflows and make comparisons vs. equal-width types.
1203 size
= PAGE_ALIGN(size
);
1204 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1205 next
= vma
->vm_next
;
1207 /* finding a matching vma now does not alter retval */
1208 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1209 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
)
1211 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1215 #else /* CONFIG_MMU */
1216 /* under NOMMU conditions, the exact address to be destroyed must be
1219 if (vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1220 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1226 up_write(&mm
->mmap_sem
);
1230 #ifdef CONFIG_PROC_FS
1231 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1233 struct user_namespace
*user_ns
= seq_user_ns(s
);
1234 struct shmid_kernel
*shp
= it
;
1235 unsigned long rss
= 0, swp
= 0;
1237 shm_add_rss_swap(shp
, &rss
, &swp
);
1239 #if BITS_PER_LONG <= 32
1240 #define SIZE_SPEC "%10lu"
1242 #define SIZE_SPEC "%21lu"
1245 return seq_printf(s
,
1246 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1247 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1248 SIZE_SPEC
" " SIZE_SPEC
"\n",
1256 from_kuid_munged(user_ns
, shp
->shm_perm
.uid
),
1257 from_kgid_munged(user_ns
, shp
->shm_perm
.gid
),
1258 from_kuid_munged(user_ns
, shp
->shm_perm
.cuid
),
1259 from_kgid_munged(user_ns
, shp
->shm_perm
.cgid
),