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>
23 * Better ipc lock (kern_ipc_perm.lock) handling
24 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
27 #include <linux/slab.h>
29 #include <linux/hugetlb.h>
30 #include <linux/shm.h>
31 #include <linux/init.h>
32 #include <linux/file.h>
33 #include <linux/mman.h>
34 #include <linux/shmem_fs.h>
35 #include <linux/security.h>
36 #include <linux/syscalls.h>
37 #include <linux/audit.h>
38 #include <linux/capability.h>
39 #include <linux/ptrace.h>
40 #include <linux/seq_file.h>
41 #include <linux/rwsem.h>
42 #include <linux/nsproxy.h>
43 #include <linux/mount.h>
44 #include <linux/ipc_namespace.h>
46 #include <asm/uaccess.h>
50 struct shm_file_data
{
52 struct ipc_namespace
*ns
;
54 const struct vm_operations_struct
*vm_ops
;
57 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
59 static const struct file_operations shm_file_operations
;
60 static const struct vm_operations_struct shm_vm_ops
;
62 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
64 #define shm_unlock(shp) \
65 ipc_unlock(&(shp)->shm_perm)
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 void shm_init_ns(struct ipc_namespace
*ns
)
77 ns
->shm_ctlmax
= SHMMAX
;
78 ns
->shm_ctlall
= SHMALL
;
79 ns
->shm_ctlmni
= SHMMNI
;
80 ns
->shm_rmid_forced
= 0;
82 ipc_init_ids(&shm_ids(ns
));
86 * Called with shm_ids.rwsem (writer) and the shp structure locked.
87 * Only shm_ids.rwsem remains locked on exit.
89 static void do_shm_rmid(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
91 struct shmid_kernel
*shp
;
92 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
95 shp
->shm_perm
.mode
|= SHM_DEST
;
96 /* Do not find it any more */
97 shp
->shm_perm
.key
= IPC_PRIVATE
;
100 shm_destroy(ns
, shp
);
104 void shm_exit_ns(struct ipc_namespace
*ns
)
106 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
107 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
111 static int __init
ipc_ns_init(void)
113 shm_init_ns(&init_ipc_ns
);
117 pure_initcall(ipc_ns_init
);
119 void __init
shm_init (void)
121 ipc_init_proc_interface("sysvipc/shm",
122 #if BITS_PER_LONG <= 32
123 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
125 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
127 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
130 static inline struct shmid_kernel
*shm_obtain_object(struct ipc_namespace
*ns
, int id
)
132 struct kern_ipc_perm
*ipcp
= ipc_obtain_object(&shm_ids(ns
), id
);
135 return ERR_CAST(ipcp
);
137 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
140 static inline struct shmid_kernel
*shm_obtain_object_check(struct ipc_namespace
*ns
, int id
)
142 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&shm_ids(ns
), id
);
145 return ERR_CAST(ipcp
);
147 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
151 * shm_lock_(check_) routines are called in the paths where the rwsem
152 * is not necessarily held.
154 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
156 struct kern_ipc_perm
*ipcp
= ipc_lock(&shm_ids(ns
), id
);
159 return (struct shmid_kernel
*)ipcp
;
161 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
164 static inline void shm_lock_by_ptr(struct shmid_kernel
*ipcp
)
167 ipc_lock_object(&ipcp
->shm_perm
);
170 static inline struct shmid_kernel
*shm_lock_check(struct ipc_namespace
*ns
,
173 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&shm_ids(ns
), id
);
176 return (struct shmid_kernel
*)ipcp
;
178 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
181 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
183 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
187 /* This is called by fork, once for every shm attach. */
188 static void shm_open(struct vm_area_struct
*vma
)
190 struct file
*file
= vma
->vm_file
;
191 struct shm_file_data
*sfd
= shm_file_data(file
);
192 struct shmid_kernel
*shp
;
194 shp
= shm_lock(sfd
->ns
, sfd
->id
);
196 shp
->shm_atim
= get_seconds();
197 shp
->shm_lprid
= task_tgid_vnr(current
);
203 * shm_destroy - free the struct shmid_kernel
206 * @shp: struct to free
208 * It has to be called with shp and shm_ids.rwsem (writer) locked,
209 * but returns with shp unlocked and freed.
211 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
213 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
216 if (!is_file_hugepages(shp
->shm_file
))
217 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
218 else if (shp
->mlock_user
)
219 user_shm_unlock(file_inode(shp
->shm_file
)->i_size
,
221 fput (shp
->shm_file
);
222 security_shm_free(shp
);
227 * shm_may_destroy - identifies whether shm segment should be destroyed now
229 * Returns true if and only if there are no active users of the segment and
230 * one of the following is true:
232 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
234 * 2) sysctl kernel.shm_rmid_forced is set to 1.
236 static bool shm_may_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
238 return (shp
->shm_nattch
== 0) &&
239 (ns
->shm_rmid_forced
||
240 (shp
->shm_perm
.mode
& SHM_DEST
));
244 * remove the attach descriptor vma.
245 * free memory for segment if it is marked destroyed.
246 * The descriptor has already been removed from the current->mm->mmap list
247 * and will later be kfree()d.
249 static void shm_close(struct vm_area_struct
*vma
)
251 struct file
* file
= vma
->vm_file
;
252 struct shm_file_data
*sfd
= shm_file_data(file
);
253 struct shmid_kernel
*shp
;
254 struct ipc_namespace
*ns
= sfd
->ns
;
256 down_write(&shm_ids(ns
).rwsem
);
257 /* remove from the list of attaches of the shm segment */
258 shp
= shm_lock(ns
, sfd
->id
);
260 shp
->shm_lprid
= task_tgid_vnr(current
);
261 shp
->shm_dtim
= get_seconds();
263 if (shm_may_destroy(ns
, shp
))
264 shm_destroy(ns
, shp
);
267 up_write(&shm_ids(ns
).rwsem
);
270 /* Called with ns->shm_ids(ns).rwsem locked */
271 static int shm_try_destroy_current(int id
, void *p
, void *data
)
273 struct ipc_namespace
*ns
= data
;
274 struct kern_ipc_perm
*ipcp
= p
;
275 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
277 if (shp
->shm_creator
!= current
)
281 * Mark it as orphaned to destroy the segment when
282 * kernel.shm_rmid_forced is changed.
283 * It is noop if the following shm_may_destroy() returns true.
285 shp
->shm_creator
= NULL
;
288 * Don't even try to destroy it. If shm_rmid_forced=0 and IPC_RMID
289 * is not set, it shouldn't be deleted here.
291 if (!ns
->shm_rmid_forced
)
294 if (shm_may_destroy(ns
, shp
)) {
295 shm_lock_by_ptr(shp
);
296 shm_destroy(ns
, shp
);
301 /* Called with ns->shm_ids(ns).rwsem locked */
302 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
304 struct ipc_namespace
*ns
= data
;
305 struct kern_ipc_perm
*ipcp
= p
;
306 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
309 * We want to destroy segments without users and with already
310 * exit'ed originating process.
312 * As shp->* are changed under rwsem, it's safe to skip shp locking.
314 if (shp
->shm_creator
!= NULL
)
317 if (shm_may_destroy(ns
, shp
)) {
318 shm_lock_by_ptr(shp
);
319 shm_destroy(ns
, shp
);
324 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
326 down_write(&shm_ids(ns
).rwsem
);
327 if (shm_ids(ns
).in_use
)
328 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
329 up_write(&shm_ids(ns
).rwsem
);
333 void exit_shm(struct task_struct
*task
)
335 struct ipc_namespace
*ns
= task
->nsproxy
->ipc_ns
;
337 if (shm_ids(ns
).in_use
== 0)
340 /* Destroy all already created segments, but not mapped yet */
341 down_write(&shm_ids(ns
).rwsem
);
342 if (shm_ids(ns
).in_use
)
343 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_current
, ns
);
344 up_write(&shm_ids(ns
).rwsem
);
347 static int shm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
349 struct file
*file
= vma
->vm_file
;
350 struct shm_file_data
*sfd
= shm_file_data(file
);
352 return sfd
->vm_ops
->fault(vma
, vmf
);
356 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
358 struct file
*file
= vma
->vm_file
;
359 struct shm_file_data
*sfd
= shm_file_data(file
);
361 if (sfd
->vm_ops
->set_policy
)
362 err
= sfd
->vm_ops
->set_policy(vma
, new);
366 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
369 struct file
*file
= vma
->vm_file
;
370 struct shm_file_data
*sfd
= shm_file_data(file
);
371 struct mempolicy
*pol
= NULL
;
373 if (sfd
->vm_ops
->get_policy
)
374 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
375 else if (vma
->vm_policy
)
376 pol
= vma
->vm_policy
;
382 static int shm_mmap(struct file
* file
, struct vm_area_struct
* vma
)
384 struct shm_file_data
*sfd
= shm_file_data(file
);
387 ret
= sfd
->file
->f_op
->mmap(sfd
->file
, vma
);
390 sfd
->vm_ops
= vma
->vm_ops
;
392 BUG_ON(!sfd
->vm_ops
->fault
);
394 vma
->vm_ops
= &shm_vm_ops
;
400 static int shm_release(struct inode
*ino
, struct file
*file
)
402 struct shm_file_data
*sfd
= shm_file_data(file
);
405 shm_file_data(file
) = NULL
;
410 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
412 struct shm_file_data
*sfd
= shm_file_data(file
);
414 if (!sfd
->file
->f_op
->fsync
)
416 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
419 static long shm_fallocate(struct file
*file
, int mode
, loff_t offset
,
422 struct shm_file_data
*sfd
= shm_file_data(file
);
424 if (!sfd
->file
->f_op
->fallocate
)
426 return sfd
->file
->f_op
->fallocate(file
, mode
, offset
, len
);
429 static unsigned long shm_get_unmapped_area(struct file
*file
,
430 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
433 struct shm_file_data
*sfd
= shm_file_data(file
);
434 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
438 static const struct file_operations shm_file_operations
= {
441 .release
= shm_release
,
443 .get_unmapped_area
= shm_get_unmapped_area
,
445 .llseek
= noop_llseek
,
446 .fallocate
= shm_fallocate
,
449 static const struct file_operations shm_file_operations_huge
= {
452 .release
= shm_release
,
453 .get_unmapped_area
= shm_get_unmapped_area
,
454 .llseek
= noop_llseek
,
455 .fallocate
= shm_fallocate
,
458 int is_file_shm_hugepages(struct file
*file
)
460 return file
->f_op
== &shm_file_operations_huge
;
463 static const struct vm_operations_struct shm_vm_ops
= {
464 .open
= shm_open
, /* callback for a new vm-area open */
465 .close
= shm_close
, /* callback for when the vm-area is released */
467 #if defined(CONFIG_NUMA)
468 .set_policy
= shm_set_policy
,
469 .get_policy
= shm_get_policy
,
474 * newseg - Create a new shared memory segment
476 * @params: ptr to the structure that contains key, size and shmflg
478 * Called with shm_ids.rwsem held as a writer.
481 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
483 key_t key
= params
->key
;
484 int shmflg
= params
->flg
;
485 size_t size
= params
->u
.size
;
487 struct shmid_kernel
*shp
;
488 size_t numpages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
492 vm_flags_t acctflag
= 0;
494 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
497 if (ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
500 shp
= ipc_rcu_alloc(sizeof(*shp
));
504 shp
->shm_perm
.key
= key
;
505 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
506 shp
->mlock_user
= NULL
;
508 shp
->shm_perm
.security
= NULL
;
509 error
= security_shm_alloc(shp
);
515 sprintf (name
, "SYSV%08x", key
);
516 if (shmflg
& SHM_HUGETLB
) {
517 struct hstate
*hs
= hstate_sizelog((shmflg
>> SHM_HUGE_SHIFT
)
525 hugesize
= ALIGN(size
, huge_page_size(hs
));
527 /* hugetlb_file_setup applies strict accounting */
528 if (shmflg
& SHM_NORESERVE
)
529 acctflag
= VM_NORESERVE
;
530 file
= hugetlb_file_setup(name
, hugesize
, acctflag
,
531 &shp
->mlock_user
, HUGETLB_SHMFS_INODE
,
532 (shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
535 * Do not allow no accounting for OVERCOMMIT_NEVER, even
538 if ((shmflg
& SHM_NORESERVE
) &&
539 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
540 acctflag
= VM_NORESERVE
;
541 file
= shmem_file_setup(name
, size
, acctflag
);
543 error
= PTR_ERR(file
);
547 id
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
553 shp
->shm_cprid
= task_tgid_vnr(current
);
555 shp
->shm_atim
= shp
->shm_dtim
= 0;
556 shp
->shm_ctim
= get_seconds();
557 shp
->shm_segsz
= size
;
559 shp
->shm_file
= file
;
560 shp
->shm_creator
= current
;
563 * shmid gets reported as "inode#" in /proc/pid/maps.
564 * proc-ps tools use this. Changing this will break them.
566 file_inode(file
)->i_ino
= shp
->shm_perm
.id
;
568 ns
->shm_tot
+= numpages
;
569 error
= shp
->shm_perm
.id
;
571 ipc_unlock_object(&shp
->shm_perm
);
576 if (is_file_hugepages(file
) && shp
->mlock_user
)
577 user_shm_unlock(size
, shp
->mlock_user
);
580 security_shm_free(shp
);
586 * Called with shm_ids.rwsem and ipcp locked.
588 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
590 struct shmid_kernel
*shp
;
592 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
593 return security_shm_associate(shp
, shmflg
);
597 * Called with shm_ids.rwsem and ipcp locked.
599 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
600 struct ipc_params
*params
)
602 struct shmid_kernel
*shp
;
604 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
605 if (shp
->shm_segsz
< params
->u
.size
)
611 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
613 struct ipc_namespace
*ns
;
614 struct ipc_ops shm_ops
;
615 struct ipc_params shm_params
;
617 ns
= current
->nsproxy
->ipc_ns
;
619 shm_ops
.getnew
= newseg
;
620 shm_ops
.associate
= shm_security
;
621 shm_ops
.more_checks
= shm_more_checks
;
623 shm_params
.key
= key
;
624 shm_params
.flg
= shmflg
;
625 shm_params
.u
.size
= size
;
627 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
630 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
634 return copy_to_user(buf
, in
, sizeof(*in
));
639 memset(&out
, 0, sizeof(out
));
640 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
641 out
.shm_segsz
= in
->shm_segsz
;
642 out
.shm_atime
= in
->shm_atime
;
643 out
.shm_dtime
= in
->shm_dtime
;
644 out
.shm_ctime
= in
->shm_ctime
;
645 out
.shm_cpid
= in
->shm_cpid
;
646 out
.shm_lpid
= in
->shm_lpid
;
647 out
.shm_nattch
= in
->shm_nattch
;
649 return copy_to_user(buf
, &out
, sizeof(out
));
656 static inline unsigned long
657 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
661 if (copy_from_user(out
, buf
, sizeof(*out
)))
666 struct shmid_ds tbuf_old
;
668 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
671 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
672 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
673 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
682 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
686 return copy_to_user(buf
, in
, sizeof(*in
));
691 if(in
->shmmax
> INT_MAX
)
692 out
.shmmax
= INT_MAX
;
694 out
.shmmax
= (int)in
->shmmax
;
696 out
.shmmin
= in
->shmmin
;
697 out
.shmmni
= in
->shmmni
;
698 out
.shmseg
= in
->shmseg
;
699 out
.shmall
= in
->shmall
;
701 return copy_to_user(buf
, &out
, sizeof(out
));
709 * Calculate and add used RSS and swap pages of a shm.
710 * Called with shm_ids.rwsem held as a reader
712 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
713 unsigned long *rss_add
, unsigned long *swp_add
)
717 inode
= file_inode(shp
->shm_file
);
719 if (is_file_hugepages(shp
->shm_file
)) {
720 struct address_space
*mapping
= inode
->i_mapping
;
721 struct hstate
*h
= hstate_file(shp
->shm_file
);
722 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
725 struct shmem_inode_info
*info
= SHMEM_I(inode
);
726 spin_lock(&info
->lock
);
727 *rss_add
+= inode
->i_mapping
->nrpages
;
728 *swp_add
+= info
->swapped
;
729 spin_unlock(&info
->lock
);
731 *rss_add
+= inode
->i_mapping
->nrpages
;
737 * Called with shm_ids.rwsem held as a reader
739 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
748 in_use
= shm_ids(ns
).in_use
;
750 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
751 struct kern_ipc_perm
*ipc
;
752 struct shmid_kernel
*shp
;
754 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
757 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
759 shm_add_rss_swap(shp
, rss
, swp
);
766 * This function handles some shmctl commands which require the rwsem
767 * to be held in write mode.
768 * NOTE: no locks must be held, the rwsem is taken inside this function.
770 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
771 struct shmid_ds __user
*buf
, int version
)
773 struct kern_ipc_perm
*ipcp
;
774 struct shmid64_ds shmid64
;
775 struct shmid_kernel
*shp
;
778 if (cmd
== IPC_SET
) {
779 if (copy_shmid_from_user(&shmid64
, buf
, version
))
783 down_write(&shm_ids(ns
).rwsem
);
786 ipcp
= ipcctl_pre_down_nolock(ns
, &shm_ids(ns
), shmid
, cmd
,
787 &shmid64
.shm_perm
, 0);
793 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
795 err
= security_shm_shmctl(shp
, cmd
);
801 ipc_lock_object(&shp
->shm_perm
);
802 /* do_shm_rmid unlocks the ipc object and rcu */
803 do_shm_rmid(ns
, ipcp
);
806 ipc_lock_object(&shp
->shm_perm
);
807 err
= ipc_update_perm(&shmid64
.shm_perm
, ipcp
);
810 shp
->shm_ctim
= get_seconds();
818 ipc_unlock_object(&shp
->shm_perm
);
822 up_write(&shm_ids(ns
).rwsem
);
826 static int shmctl_nolock(struct ipc_namespace
*ns
, int shmid
,
827 int cmd
, int version
, void __user
*buf
)
830 struct shmid_kernel
*shp
;
832 /* preliminary security checks for *_INFO */
833 if (cmd
== IPC_INFO
|| cmd
== SHM_INFO
) {
834 err
= security_shm_shmctl(NULL
, cmd
);
842 struct shminfo64 shminfo
;
844 memset(&shminfo
, 0, sizeof(shminfo
));
845 shminfo
.shmmni
= shminfo
.shmseg
= ns
->shm_ctlmni
;
846 shminfo
.shmmax
= ns
->shm_ctlmax
;
847 shminfo
.shmall
= ns
->shm_ctlall
;
849 shminfo
.shmmin
= SHMMIN
;
850 if(copy_shminfo_to_user (buf
, &shminfo
, version
))
853 down_read(&shm_ids(ns
).rwsem
);
854 err
= ipc_get_maxid(&shm_ids(ns
));
855 up_read(&shm_ids(ns
).rwsem
);
863 struct shm_info shm_info
;
865 memset(&shm_info
, 0, sizeof(shm_info
));
866 down_read(&shm_ids(ns
).rwsem
);
867 shm_info
.used_ids
= shm_ids(ns
).in_use
;
868 shm_get_stat (ns
, &shm_info
.shm_rss
, &shm_info
.shm_swp
);
869 shm_info
.shm_tot
= ns
->shm_tot
;
870 shm_info
.swap_attempts
= 0;
871 shm_info
.swap_successes
= 0;
872 err
= ipc_get_maxid(&shm_ids(ns
));
873 up_read(&shm_ids(ns
).rwsem
);
874 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
))) {
879 err
= err
< 0 ? 0 : err
;
885 struct shmid64_ds tbuf
;
889 if (cmd
== SHM_STAT
) {
890 shp
= shm_obtain_object(ns
, shmid
);
895 result
= shp
->shm_perm
.id
;
897 shp
= shm_obtain_object_check(ns
, shmid
);
906 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
909 err
= security_shm_shmctl(shp
, cmd
);
913 memset(&tbuf
, 0, sizeof(tbuf
));
914 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
.shm_perm
);
915 tbuf
.shm_segsz
= shp
->shm_segsz
;
916 tbuf
.shm_atime
= shp
->shm_atim
;
917 tbuf
.shm_dtime
= shp
->shm_dtim
;
918 tbuf
.shm_ctime
= shp
->shm_ctim
;
919 tbuf
.shm_cpid
= shp
->shm_cprid
;
920 tbuf
.shm_lpid
= shp
->shm_lprid
;
921 tbuf
.shm_nattch
= shp
->shm_nattch
;
924 if (copy_shmid_to_user(buf
, &tbuf
, version
))
940 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
942 struct shmid_kernel
*shp
;
944 struct ipc_namespace
*ns
;
946 if (cmd
< 0 || shmid
< 0)
949 version
= ipc_parse_version(&cmd
);
950 ns
= current
->nsproxy
->ipc_ns
;
957 return shmctl_nolock(ns
, shmid
, cmd
, version
, buf
);
960 return shmctl_down(ns
, shmid
, cmd
, buf
, version
);
964 struct file
*shm_file
;
967 shp
= shm_obtain_object_check(ns
, shmid
);
973 audit_ipc_obj(&(shp
->shm_perm
));
974 err
= security_shm_shmctl(shp
, cmd
);
978 ipc_lock_object(&shp
->shm_perm
);
979 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
980 kuid_t euid
= current_euid();
982 if (!uid_eq(euid
, shp
->shm_perm
.uid
) &&
983 !uid_eq(euid
, shp
->shm_perm
.cuid
))
985 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
))
989 shm_file
= shp
->shm_file
;
990 if (is_file_hugepages(shm_file
))
993 if (cmd
== SHM_LOCK
) {
994 struct user_struct
*user
= current_user();
995 err
= shmem_lock(shm_file
, 1, user
);
996 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)) {
997 shp
->shm_perm
.mode
|= SHM_LOCKED
;
998 shp
->mlock_user
= user
;
1004 if (!(shp
->shm_perm
.mode
& SHM_LOCKED
))
1006 shmem_lock(shm_file
, 0, shp
->mlock_user
);
1007 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
1008 shp
->mlock_user
= NULL
;
1010 ipc_unlock_object(&shp
->shm_perm
);
1012 shmem_unlock_mapping(shm_file
->f_mapping
);
1022 ipc_unlock_object(&shp
->shm_perm
);
1029 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1031 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1032 * "raddr" thing points to kernel space, and there has to be a wrapper around
1035 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
, ulong
*raddr
,
1036 unsigned long shmlba
)
1038 struct shmid_kernel
*shp
;
1043 unsigned long flags
;
1046 struct ipc_namespace
*ns
;
1047 struct shm_file_data
*sfd
;
1050 unsigned long populate
= 0;
1055 else if ((addr
= (ulong
)shmaddr
)) {
1056 if (addr
& (shmlba
- 1)) {
1057 if (shmflg
& SHM_RND
)
1058 addr
&= ~(shmlba
- 1); /* round down */
1060 #ifndef __ARCH_FORCE_SHMLBA
1061 if (addr
& ~PAGE_MASK
)
1065 flags
= MAP_SHARED
| MAP_FIXED
;
1067 if ((shmflg
& SHM_REMAP
))
1073 if (shmflg
& SHM_RDONLY
) {
1076 f_mode
= FMODE_READ
;
1078 prot
= PROT_READ
| PROT_WRITE
;
1079 acc_mode
= S_IRUGO
| S_IWUGO
;
1080 f_mode
= FMODE_READ
| FMODE_WRITE
;
1082 if (shmflg
& SHM_EXEC
) {
1084 acc_mode
|= S_IXUGO
;
1088 * We cannot rely on the fs check since SYSV IPC does have an
1089 * additional creator id...
1091 ns
= current
->nsproxy
->ipc_ns
;
1093 shp
= shm_obtain_object_check(ns
, shmid
);
1100 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
1103 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
1107 ipc_lock_object(&shp
->shm_perm
);
1108 path
= shp
->shm_file
->f_path
;
1111 size
= i_size_read(path
.dentry
->d_inode
);
1112 ipc_unlock_object(&shp
->shm_perm
);
1116 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1122 file
= alloc_file(&path
, f_mode
,
1123 is_file_hugepages(shp
->shm_file
) ?
1124 &shm_file_operations_huge
:
1125 &shm_file_operations
);
1126 err
= PTR_ERR(file
);
1133 file
->private_data
= sfd
;
1134 file
->f_mapping
= shp
->shm_file
->f_mapping
;
1135 sfd
->id
= shp
->shm_perm
.id
;
1136 sfd
->ns
= get_ipc_ns(ns
);
1137 sfd
->file
= shp
->shm_file
;
1140 err
= security_mmap_file(file
, prot
, flags
);
1144 down_write(¤t
->mm
->mmap_sem
);
1145 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1147 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1150 * If shm segment goes below stack, make sure there is some
1151 * space left for the stack to grow (at least 4 pages).
1153 if (addr
< current
->mm
->start_stack
&&
1154 addr
> current
->mm
->start_stack
- size
- PAGE_SIZE
* 5)
1158 addr
= do_mmap_pgoff(file
, addr
, size
, prot
, flags
, 0, &populate
);
1161 if (IS_ERR_VALUE(addr
))
1164 up_write(¤t
->mm
->mmap_sem
);
1166 mm_populate(addr
, populate
);
1172 down_write(&shm_ids(ns
).rwsem
);
1173 shp
= shm_lock(ns
, shmid
);
1174 BUG_ON(IS_ERR(shp
));
1176 if (shm_may_destroy(ns
, shp
))
1177 shm_destroy(ns
, shp
);
1180 up_write(&shm_ids(ns
).rwsem
);
1189 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1194 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
, SHMLBA
);
1197 force_successful_syscall_return();
1202 * detach and kill segment if marked destroyed.
1203 * The work is done in shm_close.
1205 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1207 struct mm_struct
*mm
= current
->mm
;
1208 struct vm_area_struct
*vma
;
1209 unsigned long addr
= (unsigned long)shmaddr
;
1210 int retval
= -EINVAL
;
1213 struct vm_area_struct
*next
;
1216 if (addr
& ~PAGE_MASK
)
1219 down_write(&mm
->mmap_sem
);
1222 * This function tries to be smart and unmap shm segments that
1223 * were modified by partial mlock or munmap calls:
1224 * - It first determines the size of the shm segment that should be
1225 * unmapped: It searches for a vma that is backed by shm and that
1226 * started at address shmaddr. It records it's size and then unmaps
1228 * - Then it unmaps all shm vmas that started at shmaddr and that
1229 * are within the initially determined size.
1230 * Errors from do_munmap are ignored: the function only fails if
1231 * it's called with invalid parameters or if it's called to unmap
1232 * a part of a vma. Both calls in this function are for full vmas,
1233 * the parameters are directly copied from the vma itself and always
1234 * valid - therefore do_munmap cannot fail. (famous last words?)
1237 * If it had been mremap()'d, the starting address would not
1238 * match the usual checks anyway. So assume all vma's are
1239 * above the starting address given.
1241 vma
= find_vma(mm
, addr
);
1245 next
= vma
->vm_next
;
1248 * Check if the starting address would match, i.e. it's
1249 * a fragment created by mprotect() and/or munmap(), or it
1250 * otherwise it starts at this address with no hassles.
1252 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1253 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1256 size
= file_inode(vma
->vm_file
)->i_size
;
1257 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1259 * We discovered the size of the shm segment, so
1260 * break out of here and fall through to the next
1261 * loop that uses the size information to stop
1262 * searching for matching vma's.
1272 * We need look no further than the maximum address a fragment
1273 * could possibly have landed at. Also cast things to loff_t to
1274 * prevent overflows and make comparisons vs. equal-width types.
1276 size
= PAGE_ALIGN(size
);
1277 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1278 next
= vma
->vm_next
;
1280 /* finding a matching vma now does not alter retval */
1281 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1282 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
)
1284 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1288 #else /* CONFIG_MMU */
1289 /* under NOMMU conditions, the exact address to be destroyed must be
1292 if (vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1293 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1299 up_write(&mm
->mmap_sem
);
1303 #ifdef CONFIG_PROC_FS
1304 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1306 struct user_namespace
*user_ns
= seq_user_ns(s
);
1307 struct shmid_kernel
*shp
= it
;
1308 unsigned long rss
= 0, swp
= 0;
1310 shm_add_rss_swap(shp
, &rss
, &swp
);
1312 #if BITS_PER_LONG <= 32
1313 #define SIZE_SPEC "%10lu"
1315 #define SIZE_SPEC "%21lu"
1318 return seq_printf(s
,
1319 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1320 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1321 SIZE_SPEC
" " SIZE_SPEC
"\n",
1329 from_kuid_munged(user_ns
, shp
->shm_perm
.uid
),
1330 from_kgid_munged(user_ns
, shp
->shm_perm
.gid
),
1331 from_kuid_munged(user_ns
, shp
->shm_perm
.cuid
),
1332 from_kgid_munged(user_ns
, shp
->shm_perm
.cgid
),