1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 1992, 1993 Krishna Balasubramanian
5 * Many improvements/fixes by Bruno Haible.
6 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
7 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
9 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
10 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
11 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
12 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
13 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
14 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
15 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Better ipc lock (kern_ipc_perm.lock) handling
25 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
28 #include <linux/slab.h>
30 #include <linux/hugetlb.h>
31 #include <linux/shm.h>
32 #include <linux/init.h>
33 #include <linux/file.h>
34 #include <linux/mman.h>
35 #include <linux/shmem_fs.h>
36 #include <linux/security.h>
37 #include <linux/syscalls.h>
38 #include <linux/audit.h>
39 #include <linux/capability.h>
40 #include <linux/ptrace.h>
41 #include <linux/seq_file.h>
42 #include <linux/rwsem.h>
43 #include <linux/nsproxy.h>
44 #include <linux/mount.h>
45 #include <linux/ipc_namespace.h>
46 #include <linux/rhashtable.h>
48 #include <linux/uaccess.h>
52 struct shmid_kernel
/* private to the kernel */
54 struct kern_ipc_perm shm_perm
;
55 struct file
*shm_file
;
56 unsigned long shm_nattch
;
57 unsigned long shm_segsz
;
61 struct pid
*shm_cprid
;
62 struct pid
*shm_lprid
;
63 struct ucounts
*mlock_ucounts
;
66 * The task created the shm object, for
67 * task_lock(shp->shm_creator)
69 struct task_struct
*shm_creator
;
72 * List by creator. task_lock(->shm_creator) required for read/write.
73 * If list_empty(), then the creator is dead already.
75 struct list_head shm_clist
;
76 struct ipc_namespace
*ns
;
79 /* shm_mode upper byte flags */
80 #define SHM_DEST 01000 /* segment will be destroyed on last detach */
81 #define SHM_LOCKED 02000 /* segment will not be swapped */
83 struct shm_file_data
{
85 struct ipc_namespace
*ns
;
87 const struct vm_operations_struct
*vm_ops
;
90 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
92 static const struct file_operations shm_file_operations
;
93 static const struct vm_operations_struct shm_vm_ops
;
95 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
97 #define shm_unlock(shp) \
98 ipc_unlock(&(shp)->shm_perm)
100 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
101 static void shm_open(struct vm_area_struct
*vma
);
102 static void shm_close(struct vm_area_struct
*vma
);
103 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
104 #ifdef CONFIG_PROC_FS
105 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
108 void shm_init_ns(struct ipc_namespace
*ns
)
110 ns
->shm_ctlmax
= SHMMAX
;
111 ns
->shm_ctlall
= SHMALL
;
112 ns
->shm_ctlmni
= SHMMNI
;
113 ns
->shm_rmid_forced
= 0;
115 ipc_init_ids(&shm_ids(ns
));
119 * Called with shm_ids.rwsem (writer) and the shp structure locked.
120 * Only shm_ids.rwsem remains locked on exit.
122 static void do_shm_rmid(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
124 struct shmid_kernel
*shp
;
126 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
127 WARN_ON(ns
!= shp
->ns
);
129 if (shp
->shm_nattch
) {
130 shp
->shm_perm
.mode
|= SHM_DEST
;
131 /* Do not find it any more */
132 ipc_set_key_private(&shm_ids(ns
), &shp
->shm_perm
);
135 shm_destroy(ns
, shp
);
139 void shm_exit_ns(struct ipc_namespace
*ns
)
141 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
142 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
143 rhashtable_destroy(&ns
->ids
[IPC_SHM_IDS
].key_ht
);
147 static int __init
ipc_ns_init(void)
149 shm_init_ns(&init_ipc_ns
);
153 pure_initcall(ipc_ns_init
);
155 void __init
shm_init(void)
157 ipc_init_proc_interface("sysvipc/shm",
158 #if BITS_PER_LONG <= 32
159 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
161 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
163 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
166 static inline struct shmid_kernel
*shm_obtain_object(struct ipc_namespace
*ns
, int id
)
168 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_idr(&shm_ids(ns
), id
);
171 return ERR_CAST(ipcp
);
173 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
176 static inline struct shmid_kernel
*shm_obtain_object_check(struct ipc_namespace
*ns
, int id
)
178 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&shm_ids(ns
), id
);
181 return ERR_CAST(ipcp
);
183 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
187 * shm_lock_(check_) routines are called in the paths where the rwsem
188 * is not necessarily held.
190 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
192 struct kern_ipc_perm
*ipcp
;
195 ipcp
= ipc_obtain_object_idr(&shm_ids(ns
), id
);
199 ipc_lock_object(ipcp
);
201 * ipc_rmid() may have already freed the ID while ipc_lock_object()
202 * was spinning: here verify that the structure is still valid.
203 * Upon races with RMID, return -EIDRM, thus indicating that
204 * the ID points to a removed identifier.
206 if (ipc_valid_object(ipcp
)) {
207 /* return a locked ipc object upon success */
208 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
211 ipc_unlock_object(ipcp
);
212 ipcp
= ERR_PTR(-EIDRM
);
216 * Callers of shm_lock() must validate the status of the returned ipc
217 * object pointer and error out as appropriate.
219 return ERR_CAST(ipcp
);
222 static inline void shm_lock_by_ptr(struct shmid_kernel
*ipcp
)
225 ipc_lock_object(&ipcp
->shm_perm
);
228 static void shm_rcu_free(struct rcu_head
*head
)
230 struct kern_ipc_perm
*ptr
= container_of(head
, struct kern_ipc_perm
,
232 struct shmid_kernel
*shp
= container_of(ptr
, struct shmid_kernel
,
234 security_shm_free(&shp
->shm_perm
);
239 * It has to be called with shp locked.
240 * It must be called before ipc_rmid()
242 static inline void shm_clist_rm(struct shmid_kernel
*shp
)
244 struct task_struct
*creator
;
246 /* ensure that shm_creator does not disappear */
250 * A concurrent exit_shm may do a list_del_init() as well.
251 * Just do nothing if exit_shm already did the work
253 if (!list_empty(&shp
->shm_clist
)) {
255 * shp->shm_creator is guaranteed to be valid *only*
256 * if shp->shm_clist is not empty.
258 creator
= shp
->shm_creator
;
262 * list_del_init() is a nop if the entry was already removed
265 list_del_init(&shp
->shm_clist
);
266 task_unlock(creator
);
271 static inline void shm_rmid(struct shmid_kernel
*s
)
274 ipc_rmid(&shm_ids(s
->ns
), &s
->shm_perm
);
278 static int __shm_open(struct vm_area_struct
*vma
)
280 struct file
*file
= vma
->vm_file
;
281 struct shm_file_data
*sfd
= shm_file_data(file
);
282 struct shmid_kernel
*shp
;
284 shp
= shm_lock(sfd
->ns
, sfd
->id
);
289 if (shp
->shm_file
!= sfd
->file
) {
295 shp
->shm_atim
= ktime_get_real_seconds();
296 ipc_update_pid(&shp
->shm_lprid
, task_tgid(current
));
302 /* This is called by fork, once for every shm attach. */
303 static void shm_open(struct vm_area_struct
*vma
)
305 int err
= __shm_open(vma
);
307 * We raced in the idr lookup or with shm_destroy().
308 * Either way, the ID is busted.
314 * shm_destroy - free the struct shmid_kernel
317 * @shp: struct to free
319 * It has to be called with shp and shm_ids.rwsem (writer) locked,
320 * but returns with shp unlocked and freed.
322 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
324 struct file
*shm_file
;
326 shm_file
= shp
->shm_file
;
327 shp
->shm_file
= NULL
;
328 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
331 if (!is_file_hugepages(shm_file
))
332 shmem_lock(shm_file
, 0, shp
->mlock_ucounts
);
334 ipc_update_pid(&shp
->shm_cprid
, NULL
);
335 ipc_update_pid(&shp
->shm_lprid
, NULL
);
336 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
340 * shm_may_destroy - identifies whether shm segment should be destroyed now
342 * Returns true if and only if there are no active users of the segment and
343 * one of the following is true:
345 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
347 * 2) sysctl kernel.shm_rmid_forced is set to 1.
349 static bool shm_may_destroy(struct shmid_kernel
*shp
)
351 return (shp
->shm_nattch
== 0) &&
352 (shp
->ns
->shm_rmid_forced
||
353 (shp
->shm_perm
.mode
& SHM_DEST
));
357 * remove the attach descriptor vma.
358 * free memory for segment if it is marked destroyed.
359 * The descriptor has already been removed from the current->mm->mmap list
360 * and will later be kfree()d.
362 static void shm_close(struct vm_area_struct
*vma
)
364 struct file
*file
= vma
->vm_file
;
365 struct shm_file_data
*sfd
= shm_file_data(file
);
366 struct shmid_kernel
*shp
;
367 struct ipc_namespace
*ns
= sfd
->ns
;
369 down_write(&shm_ids(ns
).rwsem
);
370 /* remove from the list of attaches of the shm segment */
371 shp
= shm_lock(ns
, sfd
->id
);
374 * We raced in the idr lookup or with shm_destroy().
375 * Either way, the ID is busted.
377 if (WARN_ON_ONCE(IS_ERR(shp
)))
378 goto done
; /* no-op */
380 ipc_update_pid(&shp
->shm_lprid
, task_tgid(current
));
381 shp
->shm_dtim
= ktime_get_real_seconds();
383 if (shm_may_destroy(shp
))
384 shm_destroy(ns
, shp
);
388 up_write(&shm_ids(ns
).rwsem
);
391 /* Called with ns->shm_ids(ns).rwsem locked */
392 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
394 struct ipc_namespace
*ns
= data
;
395 struct kern_ipc_perm
*ipcp
= p
;
396 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
399 * We want to destroy segments without users and with already
400 * exit'ed originating process.
402 * As shp->* are changed under rwsem, it's safe to skip shp locking.
404 if (!list_empty(&shp
->shm_clist
))
407 if (shm_may_destroy(shp
)) {
408 shm_lock_by_ptr(shp
);
409 shm_destroy(ns
, shp
);
414 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
416 down_write(&shm_ids(ns
).rwsem
);
417 if (shm_ids(ns
).in_use
)
418 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
419 up_write(&shm_ids(ns
).rwsem
);
422 /* Locking assumes this will only be called with task == current */
423 void exit_shm(struct task_struct
*task
)
426 struct shmid_kernel
*shp
;
427 struct ipc_namespace
*ns
;
431 if (list_empty(&task
->sysvshm
.shm_clist
)) {
436 shp
= list_first_entry(&task
->sysvshm
.shm_clist
, struct shmid_kernel
,
440 * 1) Get pointer to the ipc namespace. It is worth to say
441 * that this pointer is guaranteed to be valid because
442 * shp lifetime is always shorter than namespace lifetime
443 * in which shp lives.
444 * We taken task_lock it means that shp won't be freed.
449 * 2) If kernel.shm_rmid_forced is not set then only keep track of
450 * which shmids are orphaned, so that a later set of the sysctl
453 if (!ns
->shm_rmid_forced
)
454 goto unlink_continue
;
457 * 3) get a reference to the namespace.
458 * The refcount could be already 0. If it is 0, then
459 * the shm objects will be free by free_ipc_work().
461 ns
= get_ipc_ns_not_zero(ns
);
464 list_del_init(&shp
->shm_clist
);
470 * 4) get a reference to shp.
471 * This cannot fail: shm_clist_rm() is called before
472 * ipc_rmid(), thus the refcount cannot be 0.
474 WARN_ON(!ipc_rcu_getref(&shp
->shm_perm
));
477 * 5) unlink the shm segment from the list of segments
478 * created by current.
479 * This must be done last. After unlinking,
480 * only the refcounts obtained above prevent IPC_RMID
481 * from destroying the segment or the namespace.
483 list_del_init(&shp
->shm_clist
);
488 * 6) we have all references
489 * Thus lock & if needed destroy shp.
491 down_write(&shm_ids(ns
).rwsem
);
492 shm_lock_by_ptr(shp
);
494 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
495 * safe to call ipc_rcu_putref here
497 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
499 if (ipc_valid_object(&shp
->shm_perm
)) {
500 if (shm_may_destroy(shp
))
501 shm_destroy(ns
, shp
);
506 * Someone else deleted the shp from namespace
507 * idr/kht while we have waited.
508 * Just unlock and continue.
513 up_write(&shm_ids(ns
).rwsem
);
514 put_ipc_ns(ns
); /* paired with get_ipc_ns_not_zero */
518 static vm_fault_t
shm_fault(struct vm_fault
*vmf
)
520 struct file
*file
= vmf
->vma
->vm_file
;
521 struct shm_file_data
*sfd
= shm_file_data(file
);
523 return sfd
->vm_ops
->fault(vmf
);
526 static int shm_may_split(struct vm_area_struct
*vma
, unsigned long addr
)
528 struct file
*file
= vma
->vm_file
;
529 struct shm_file_data
*sfd
= shm_file_data(file
);
531 if (sfd
->vm_ops
->may_split
)
532 return sfd
->vm_ops
->may_split(vma
, addr
);
537 static unsigned long shm_pagesize(struct vm_area_struct
*vma
)
539 struct file
*file
= vma
->vm_file
;
540 struct shm_file_data
*sfd
= shm_file_data(file
);
542 if (sfd
->vm_ops
->pagesize
)
543 return sfd
->vm_ops
->pagesize(vma
);
549 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
551 struct file
*file
= vma
->vm_file
;
552 struct shm_file_data
*sfd
= shm_file_data(file
);
555 if (sfd
->vm_ops
->set_policy
)
556 err
= sfd
->vm_ops
->set_policy(vma
, new);
560 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
563 struct file
*file
= vma
->vm_file
;
564 struct shm_file_data
*sfd
= shm_file_data(file
);
565 struct mempolicy
*pol
= NULL
;
567 if (sfd
->vm_ops
->get_policy
)
568 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
569 else if (vma
->vm_policy
)
570 pol
= vma
->vm_policy
;
576 static int shm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
578 struct shm_file_data
*sfd
= shm_file_data(file
);
582 * In case of remap_file_pages() emulation, the file can represent an
583 * IPC ID that was removed, and possibly even reused by another shm
584 * segment already. Propagate this case as an error to caller.
586 ret
= __shm_open(vma
);
590 ret
= call_mmap(sfd
->file
, vma
);
595 sfd
->vm_ops
= vma
->vm_ops
;
597 WARN_ON(!sfd
->vm_ops
->fault
);
599 vma
->vm_ops
= &shm_vm_ops
;
603 static int shm_release(struct inode
*ino
, struct file
*file
)
605 struct shm_file_data
*sfd
= shm_file_data(file
);
609 shm_file_data(file
) = NULL
;
614 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
616 struct shm_file_data
*sfd
= shm_file_data(file
);
618 if (!sfd
->file
->f_op
->fsync
)
620 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
623 static long shm_fallocate(struct file
*file
, int mode
, loff_t offset
,
626 struct shm_file_data
*sfd
= shm_file_data(file
);
628 if (!sfd
->file
->f_op
->fallocate
)
630 return sfd
->file
->f_op
->fallocate(file
, mode
, offset
, len
);
633 static unsigned long shm_get_unmapped_area(struct file
*file
,
634 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
637 struct shm_file_data
*sfd
= shm_file_data(file
);
639 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
643 static const struct file_operations shm_file_operations
= {
646 .release
= shm_release
,
647 .get_unmapped_area
= shm_get_unmapped_area
,
648 .llseek
= noop_llseek
,
649 .fallocate
= shm_fallocate
,
653 * shm_file_operations_huge is now identical to shm_file_operations,
654 * but we keep it distinct for the sake of is_file_shm_hugepages().
656 static const struct file_operations shm_file_operations_huge
= {
659 .release
= shm_release
,
660 .get_unmapped_area
= shm_get_unmapped_area
,
661 .llseek
= noop_llseek
,
662 .fallocate
= shm_fallocate
,
665 bool is_file_shm_hugepages(struct file
*file
)
667 return file
->f_op
== &shm_file_operations_huge
;
670 static const struct vm_operations_struct shm_vm_ops
= {
671 .open
= shm_open
, /* callback for a new vm-area open */
672 .close
= shm_close
, /* callback for when the vm-area is released */
674 .may_split
= shm_may_split
,
675 .pagesize
= shm_pagesize
,
676 #if defined(CONFIG_NUMA)
677 .set_policy
= shm_set_policy
,
678 .get_policy
= shm_get_policy
,
683 * newseg - Create a new shared memory segment
685 * @params: ptr to the structure that contains key, size and shmflg
687 * Called with shm_ids.rwsem held as a writer.
689 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
691 key_t key
= params
->key
;
692 int shmflg
= params
->flg
;
693 size_t size
= params
->u
.size
;
695 struct shmid_kernel
*shp
;
696 size_t numpages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
699 vm_flags_t acctflag
= 0;
701 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
704 if (numpages
<< PAGE_SHIFT
< size
)
707 if (ns
->shm_tot
+ numpages
< ns
->shm_tot
||
708 ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
711 shp
= kmalloc(sizeof(*shp
), GFP_KERNEL_ACCOUNT
);
715 shp
->shm_perm
.key
= key
;
716 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
717 shp
->mlock_ucounts
= NULL
;
719 shp
->shm_perm
.security
= NULL
;
720 error
= security_shm_alloc(&shp
->shm_perm
);
726 sprintf(name
, "SYSV%08x", key
);
727 if (shmflg
& SHM_HUGETLB
) {
731 hs
= hstate_sizelog((shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
736 hugesize
= ALIGN(size
, huge_page_size(hs
));
738 /* hugetlb_file_setup applies strict accounting */
739 if (shmflg
& SHM_NORESERVE
)
740 acctflag
= VM_NORESERVE
;
741 file
= hugetlb_file_setup(name
, hugesize
, acctflag
,
742 HUGETLB_SHMFS_INODE
, (shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
745 * Do not allow no accounting for OVERCOMMIT_NEVER, even
748 if ((shmflg
& SHM_NORESERVE
) &&
749 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
750 acctflag
= VM_NORESERVE
;
751 file
= shmem_kernel_file_setup(name
, size
, acctflag
);
753 error
= PTR_ERR(file
);
757 shp
->shm_cprid
= get_pid(task_tgid(current
));
758 shp
->shm_lprid
= NULL
;
759 shp
->shm_atim
= shp
->shm_dtim
= 0;
760 shp
->shm_ctim
= ktime_get_real_seconds();
761 shp
->shm_segsz
= size
;
763 shp
->shm_file
= file
;
764 shp
->shm_creator
= current
;
766 /* ipc_addid() locks shp upon success. */
767 error
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
774 list_add(&shp
->shm_clist
, ¤t
->sysvshm
.shm_clist
);
775 task_unlock(current
);
778 * shmid gets reported as "inode#" in /proc/pid/maps.
779 * proc-ps tools use this. Changing this will break them.
781 file_inode(file
)->i_ino
= shp
->shm_perm
.id
;
783 ns
->shm_tot
+= numpages
;
784 error
= shp
->shm_perm
.id
;
786 ipc_unlock_object(&shp
->shm_perm
);
791 ipc_update_pid(&shp
->shm_cprid
, NULL
);
792 ipc_update_pid(&shp
->shm_lprid
, NULL
);
794 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
797 call_rcu(&shp
->shm_perm
.rcu
, shm_rcu_free
);
802 * Called with shm_ids.rwsem and ipcp locked.
804 static int shm_more_checks(struct kern_ipc_perm
*ipcp
, struct ipc_params
*params
)
806 struct shmid_kernel
*shp
;
808 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
809 if (shp
->shm_segsz
< params
->u
.size
)
815 long ksys_shmget(key_t key
, size_t size
, int shmflg
)
817 struct ipc_namespace
*ns
;
818 static const struct ipc_ops shm_ops
= {
820 .associate
= security_shm_associate
,
821 .more_checks
= shm_more_checks
,
823 struct ipc_params shm_params
;
825 ns
= current
->nsproxy
->ipc_ns
;
827 shm_params
.key
= key
;
828 shm_params
.flg
= shmflg
;
829 shm_params
.u
.size
= size
;
831 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
834 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
836 return ksys_shmget(key
, size
, shmflg
);
839 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
843 return copy_to_user(buf
, in
, sizeof(*in
));
848 memset(&out
, 0, sizeof(out
));
849 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
850 out
.shm_segsz
= in
->shm_segsz
;
851 out
.shm_atime
= in
->shm_atime
;
852 out
.shm_dtime
= in
->shm_dtime
;
853 out
.shm_ctime
= in
->shm_ctime
;
854 out
.shm_cpid
= in
->shm_cpid
;
855 out
.shm_lpid
= in
->shm_lpid
;
856 out
.shm_nattch
= in
->shm_nattch
;
858 return copy_to_user(buf
, &out
, sizeof(out
));
865 static inline unsigned long
866 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
870 if (copy_from_user(out
, buf
, sizeof(*out
)))
875 struct shmid_ds tbuf_old
;
877 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
880 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
881 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
882 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
891 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
895 return copy_to_user(buf
, in
, sizeof(*in
));
900 if (in
->shmmax
> INT_MAX
)
901 out
.shmmax
= INT_MAX
;
903 out
.shmmax
= (int)in
->shmmax
;
905 out
.shmmin
= in
->shmmin
;
906 out
.shmmni
= in
->shmmni
;
907 out
.shmseg
= in
->shmseg
;
908 out
.shmall
= in
->shmall
;
910 return copy_to_user(buf
, &out
, sizeof(out
));
918 * Calculate and add used RSS and swap pages of a shm.
919 * Called with shm_ids.rwsem held as a reader
921 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
922 unsigned long *rss_add
, unsigned long *swp_add
)
926 inode
= file_inode(shp
->shm_file
);
928 if (is_file_hugepages(shp
->shm_file
)) {
929 struct address_space
*mapping
= inode
->i_mapping
;
930 struct hstate
*h
= hstate_file(shp
->shm_file
);
931 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
934 struct shmem_inode_info
*info
= SHMEM_I(inode
);
936 spin_lock_irq(&info
->lock
);
937 *rss_add
+= inode
->i_mapping
->nrpages
;
938 *swp_add
+= info
->swapped
;
939 spin_unlock_irq(&info
->lock
);
941 *rss_add
+= inode
->i_mapping
->nrpages
;
947 * Called with shm_ids.rwsem held as a reader
949 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
958 in_use
= shm_ids(ns
).in_use
;
960 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
961 struct kern_ipc_perm
*ipc
;
962 struct shmid_kernel
*shp
;
964 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
967 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
969 shm_add_rss_swap(shp
, rss
, swp
);
976 * This function handles some shmctl commands which require the rwsem
977 * to be held in write mode.
978 * NOTE: no locks must be held, the rwsem is taken inside this function.
980 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
981 struct shmid64_ds
*shmid64
)
983 struct kern_ipc_perm
*ipcp
;
984 struct shmid_kernel
*shp
;
987 down_write(&shm_ids(ns
).rwsem
);
990 ipcp
= ipcctl_obtain_check(ns
, &shm_ids(ns
), shmid
, cmd
,
991 &shmid64
->shm_perm
, 0);
997 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
999 err
= security_shm_shmctl(&shp
->shm_perm
, cmd
);
1005 ipc_lock_object(&shp
->shm_perm
);
1006 /* do_shm_rmid unlocks the ipc object and rcu */
1007 do_shm_rmid(ns
, ipcp
);
1010 ipc_lock_object(&shp
->shm_perm
);
1011 err
= ipc_update_perm(&shmid64
->shm_perm
, ipcp
);
1014 shp
->shm_ctim
= ktime_get_real_seconds();
1022 ipc_unlock_object(&shp
->shm_perm
);
1026 up_write(&shm_ids(ns
).rwsem
);
1030 static int shmctl_ipc_info(struct ipc_namespace
*ns
,
1031 struct shminfo64
*shminfo
)
1033 int err
= security_shm_shmctl(NULL
, IPC_INFO
);
1035 memset(shminfo
, 0, sizeof(*shminfo
));
1036 shminfo
->shmmni
= shminfo
->shmseg
= ns
->shm_ctlmni
;
1037 shminfo
->shmmax
= ns
->shm_ctlmax
;
1038 shminfo
->shmall
= ns
->shm_ctlall
;
1039 shminfo
->shmmin
= SHMMIN
;
1040 down_read(&shm_ids(ns
).rwsem
);
1041 err
= ipc_get_maxidx(&shm_ids(ns
));
1042 up_read(&shm_ids(ns
).rwsem
);
1049 static int shmctl_shm_info(struct ipc_namespace
*ns
,
1050 struct shm_info
*shm_info
)
1052 int err
= security_shm_shmctl(NULL
, SHM_INFO
);
1054 memset(shm_info
, 0, sizeof(*shm_info
));
1055 down_read(&shm_ids(ns
).rwsem
);
1056 shm_info
->used_ids
= shm_ids(ns
).in_use
;
1057 shm_get_stat(ns
, &shm_info
->shm_rss
, &shm_info
->shm_swp
);
1058 shm_info
->shm_tot
= ns
->shm_tot
;
1059 shm_info
->swap_attempts
= 0;
1060 shm_info
->swap_successes
= 0;
1061 err
= ipc_get_maxidx(&shm_ids(ns
));
1062 up_read(&shm_ids(ns
).rwsem
);
1069 static int shmctl_stat(struct ipc_namespace
*ns
, int shmid
,
1070 int cmd
, struct shmid64_ds
*tbuf
)
1072 struct shmid_kernel
*shp
;
1075 memset(tbuf
, 0, sizeof(*tbuf
));
1078 if (cmd
== SHM_STAT
|| cmd
== SHM_STAT_ANY
) {
1079 shp
= shm_obtain_object(ns
, shmid
);
1084 } else { /* IPC_STAT */
1085 shp
= shm_obtain_object_check(ns
, shmid
);
1093 * Semantically SHM_STAT_ANY ought to be identical to
1094 * that functionality provided by the /proc/sysvipc/
1095 * interface. As such, only audit these calls and
1096 * do not do traditional S_IRUGO permission checks on
1099 if (cmd
== SHM_STAT_ANY
)
1100 audit_ipc_obj(&shp
->shm_perm
);
1103 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
1107 err
= security_shm_shmctl(&shp
->shm_perm
, cmd
);
1111 ipc_lock_object(&shp
->shm_perm
);
1113 if (!ipc_valid_object(&shp
->shm_perm
)) {
1114 ipc_unlock_object(&shp
->shm_perm
);
1119 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
->shm_perm
);
1120 tbuf
->shm_segsz
= shp
->shm_segsz
;
1121 tbuf
->shm_atime
= shp
->shm_atim
;
1122 tbuf
->shm_dtime
= shp
->shm_dtim
;
1123 tbuf
->shm_ctime
= shp
->shm_ctim
;
1124 #ifndef CONFIG_64BIT
1125 tbuf
->shm_atime_high
= shp
->shm_atim
>> 32;
1126 tbuf
->shm_dtime_high
= shp
->shm_dtim
>> 32;
1127 tbuf
->shm_ctime_high
= shp
->shm_ctim
>> 32;
1129 tbuf
->shm_cpid
= pid_vnr(shp
->shm_cprid
);
1130 tbuf
->shm_lpid
= pid_vnr(shp
->shm_lprid
);
1131 tbuf
->shm_nattch
= shp
->shm_nattch
;
1133 if (cmd
== IPC_STAT
) {
1135 * As defined in SUS:
1136 * Return 0 on success
1141 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1142 * Return the full id, including the sequence number
1144 err
= shp
->shm_perm
.id
;
1147 ipc_unlock_object(&shp
->shm_perm
);
1153 static int shmctl_do_lock(struct ipc_namespace
*ns
, int shmid
, int cmd
)
1155 struct shmid_kernel
*shp
;
1156 struct file
*shm_file
;
1160 shp
= shm_obtain_object_check(ns
, shmid
);
1166 audit_ipc_obj(&(shp
->shm_perm
));
1167 err
= security_shm_shmctl(&shp
->shm_perm
, cmd
);
1171 ipc_lock_object(&shp
->shm_perm
);
1173 /* check if shm_destroy() is tearing down shp */
1174 if (!ipc_valid_object(&shp
->shm_perm
)) {
1179 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
1180 kuid_t euid
= current_euid();
1182 if (!uid_eq(euid
, shp
->shm_perm
.uid
) &&
1183 !uid_eq(euid
, shp
->shm_perm
.cuid
)) {
1187 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
)) {
1193 shm_file
= shp
->shm_file
;
1194 if (is_file_hugepages(shm_file
))
1197 if (cmd
== SHM_LOCK
) {
1198 struct ucounts
*ucounts
= current_ucounts();
1200 err
= shmem_lock(shm_file
, 1, ucounts
);
1201 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)) {
1202 shp
->shm_perm
.mode
|= SHM_LOCKED
;
1203 shp
->mlock_ucounts
= ucounts
;
1209 if (!(shp
->shm_perm
.mode
& SHM_LOCKED
))
1211 shmem_lock(shm_file
, 0, shp
->mlock_ucounts
);
1212 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
1213 shp
->mlock_ucounts
= NULL
;
1215 ipc_unlock_object(&shp
->shm_perm
);
1217 shmem_unlock_mapping(shm_file
->f_mapping
);
1223 ipc_unlock_object(&shp
->shm_perm
);
1229 static long ksys_shmctl(int shmid
, int cmd
, struct shmid_ds __user
*buf
, int version
)
1232 struct ipc_namespace
*ns
;
1233 struct shmid64_ds sem64
;
1235 if (cmd
< 0 || shmid
< 0)
1238 ns
= current
->nsproxy
->ipc_ns
;
1242 struct shminfo64 shminfo
;
1243 err
= shmctl_ipc_info(ns
, &shminfo
);
1246 if (copy_shminfo_to_user(buf
, &shminfo
, version
))
1251 struct shm_info shm_info
;
1252 err
= shmctl_shm_info(ns
, &shm_info
);
1255 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
)))
1262 err
= shmctl_stat(ns
, shmid
, cmd
, &sem64
);
1265 if (copy_shmid_to_user(buf
, &sem64
, version
))
1270 if (copy_shmid_from_user(&sem64
, buf
, version
))
1274 return shmctl_down(ns
, shmid
, cmd
, &sem64
);
1277 return shmctl_do_lock(ns
, shmid
, cmd
);
1283 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
1285 return ksys_shmctl(shmid
, cmd
, buf
, IPC_64
);
1288 #ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1289 long ksys_old_shmctl(int shmid
, int cmd
, struct shmid_ds __user
*buf
)
1291 int version
= ipc_parse_version(&cmd
);
1293 return ksys_shmctl(shmid
, cmd
, buf
, version
);
1296 SYSCALL_DEFINE3(old_shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
1298 return ksys_old_shmctl(shmid
, cmd
, buf
);
1302 #ifdef CONFIG_COMPAT
1304 struct compat_shmid_ds
{
1305 struct compat_ipc_perm shm_perm
;
1307 old_time32_t shm_atime
;
1308 old_time32_t shm_dtime
;
1309 old_time32_t shm_ctime
;
1310 compat_ipc_pid_t shm_cpid
;
1311 compat_ipc_pid_t shm_lpid
;
1312 unsigned short shm_nattch
;
1313 unsigned short shm_unused
;
1314 compat_uptr_t shm_unused2
;
1315 compat_uptr_t shm_unused3
;
1318 struct compat_shminfo64
{
1319 compat_ulong_t shmmax
;
1320 compat_ulong_t shmmin
;
1321 compat_ulong_t shmmni
;
1322 compat_ulong_t shmseg
;
1323 compat_ulong_t shmall
;
1324 compat_ulong_t __unused1
;
1325 compat_ulong_t __unused2
;
1326 compat_ulong_t __unused3
;
1327 compat_ulong_t __unused4
;
1330 struct compat_shm_info
{
1331 compat_int_t used_ids
;
1332 compat_ulong_t shm_tot
, shm_rss
, shm_swp
;
1333 compat_ulong_t swap_attempts
, swap_successes
;
1336 static int copy_compat_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
,
1339 if (in
->shmmax
> INT_MAX
)
1340 in
->shmmax
= INT_MAX
;
1341 if (version
== IPC_64
) {
1342 struct compat_shminfo64 info
;
1343 memset(&info
, 0, sizeof(info
));
1344 info
.shmmax
= in
->shmmax
;
1345 info
.shmmin
= in
->shmmin
;
1346 info
.shmmni
= in
->shmmni
;
1347 info
.shmseg
= in
->shmseg
;
1348 info
.shmall
= in
->shmall
;
1349 return copy_to_user(buf
, &info
, sizeof(info
));
1351 struct shminfo info
;
1352 memset(&info
, 0, sizeof(info
));
1353 info
.shmmax
= in
->shmmax
;
1354 info
.shmmin
= in
->shmmin
;
1355 info
.shmmni
= in
->shmmni
;
1356 info
.shmseg
= in
->shmseg
;
1357 info
.shmall
= in
->shmall
;
1358 return copy_to_user(buf
, &info
, sizeof(info
));
1362 static int put_compat_shm_info(struct shm_info
*ip
,
1363 struct compat_shm_info __user
*uip
)
1365 struct compat_shm_info info
;
1367 memset(&info
, 0, sizeof(info
));
1368 info
.used_ids
= ip
->used_ids
;
1369 info
.shm_tot
= ip
->shm_tot
;
1370 info
.shm_rss
= ip
->shm_rss
;
1371 info
.shm_swp
= ip
->shm_swp
;
1372 info
.swap_attempts
= ip
->swap_attempts
;
1373 info
.swap_successes
= ip
->swap_successes
;
1374 return copy_to_user(uip
, &info
, sizeof(info
));
1377 static int copy_compat_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
,
1380 if (version
== IPC_64
) {
1381 struct compat_shmid64_ds v
;
1382 memset(&v
, 0, sizeof(v
));
1383 to_compat_ipc64_perm(&v
.shm_perm
, &in
->shm_perm
);
1384 v
.shm_atime
= lower_32_bits(in
->shm_atime
);
1385 v
.shm_atime_high
= upper_32_bits(in
->shm_atime
);
1386 v
.shm_dtime
= lower_32_bits(in
->shm_dtime
);
1387 v
.shm_dtime_high
= upper_32_bits(in
->shm_dtime
);
1388 v
.shm_ctime
= lower_32_bits(in
->shm_ctime
);
1389 v
.shm_ctime_high
= upper_32_bits(in
->shm_ctime
);
1390 v
.shm_segsz
= in
->shm_segsz
;
1391 v
.shm_nattch
= in
->shm_nattch
;
1392 v
.shm_cpid
= in
->shm_cpid
;
1393 v
.shm_lpid
= in
->shm_lpid
;
1394 return copy_to_user(buf
, &v
, sizeof(v
));
1396 struct compat_shmid_ds v
;
1397 memset(&v
, 0, sizeof(v
));
1398 to_compat_ipc_perm(&v
.shm_perm
, &in
->shm_perm
);
1399 v
.shm_perm
.key
= in
->shm_perm
.key
;
1400 v
.shm_atime
= in
->shm_atime
;
1401 v
.shm_dtime
= in
->shm_dtime
;
1402 v
.shm_ctime
= in
->shm_ctime
;
1403 v
.shm_segsz
= in
->shm_segsz
;
1404 v
.shm_nattch
= in
->shm_nattch
;
1405 v
.shm_cpid
= in
->shm_cpid
;
1406 v
.shm_lpid
= in
->shm_lpid
;
1407 return copy_to_user(buf
, &v
, sizeof(v
));
1411 static int copy_compat_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
,
1414 memset(out
, 0, sizeof(*out
));
1415 if (version
== IPC_64
) {
1416 struct compat_shmid64_ds __user
*p
= buf
;
1417 return get_compat_ipc64_perm(&out
->shm_perm
, &p
->shm_perm
);
1419 struct compat_shmid_ds __user
*p
= buf
;
1420 return get_compat_ipc_perm(&out
->shm_perm
, &p
->shm_perm
);
1424 static long compat_ksys_shmctl(int shmid
, int cmd
, void __user
*uptr
, int version
)
1426 struct ipc_namespace
*ns
;
1427 struct shmid64_ds sem64
;
1430 ns
= current
->nsproxy
->ipc_ns
;
1432 if (cmd
< 0 || shmid
< 0)
1437 struct shminfo64 shminfo
;
1438 err
= shmctl_ipc_info(ns
, &shminfo
);
1441 if (copy_compat_shminfo_to_user(uptr
, &shminfo
, version
))
1446 struct shm_info shm_info
;
1447 err
= shmctl_shm_info(ns
, &shm_info
);
1450 if (put_compat_shm_info(&shm_info
, uptr
))
1457 err
= shmctl_stat(ns
, shmid
, cmd
, &sem64
);
1460 if (copy_compat_shmid_to_user(uptr
, &sem64
, version
))
1465 if (copy_compat_shmid_from_user(&sem64
, uptr
, version
))
1469 return shmctl_down(ns
, shmid
, cmd
, &sem64
);
1472 return shmctl_do_lock(ns
, shmid
, cmd
);
1479 COMPAT_SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, void __user
*, uptr
)
1481 return compat_ksys_shmctl(shmid
, cmd
, uptr
, IPC_64
);
1484 #ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1485 long compat_ksys_old_shmctl(int shmid
, int cmd
, void __user
*uptr
)
1487 int version
= compat_ipc_parse_version(&cmd
);
1489 return compat_ksys_shmctl(shmid
, cmd
, uptr
, version
);
1492 COMPAT_SYSCALL_DEFINE3(old_shmctl
, int, shmid
, int, cmd
, void __user
*, uptr
)
1494 return compat_ksys_old_shmctl(shmid
, cmd
, uptr
);
1500 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1502 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1503 * "raddr" thing points to kernel space, and there has to be a wrapper around
1506 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
,
1507 ulong
*raddr
, unsigned long shmlba
)
1509 struct shmid_kernel
*shp
;
1510 unsigned long addr
= (unsigned long)shmaddr
;
1512 struct file
*file
, *base
;
1514 unsigned long flags
= MAP_SHARED
;
1517 struct ipc_namespace
*ns
;
1518 struct shm_file_data
*sfd
;
1520 unsigned long populate
= 0;
1527 if (addr
& (shmlba
- 1)) {
1528 if (shmflg
& SHM_RND
) {
1529 addr
&= ~(shmlba
- 1); /* round down */
1532 * Ensure that the round-down is non-nil
1533 * when remapping. This can happen for
1534 * cases when addr < shmlba.
1536 if (!addr
&& (shmflg
& SHM_REMAP
))
1539 #ifndef __ARCH_FORCE_SHMLBA
1540 if (addr
& ~PAGE_MASK
)
1546 } else if ((shmflg
& SHM_REMAP
))
1549 if (shmflg
& SHM_RDONLY
) {
1554 prot
= PROT_READ
| PROT_WRITE
;
1555 acc_mode
= S_IRUGO
| S_IWUGO
;
1558 if (shmflg
& SHM_EXEC
) {
1560 acc_mode
|= S_IXUGO
;
1564 * We cannot rely on the fs check since SYSV IPC does have an
1565 * additional creator id...
1567 ns
= current
->nsproxy
->ipc_ns
;
1569 shp
= shm_obtain_object_check(ns
, shmid
);
1576 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
1579 err
= security_shm_shmat(&shp
->shm_perm
, shmaddr
, shmflg
);
1583 ipc_lock_object(&shp
->shm_perm
);
1585 /* check if shm_destroy() is tearing down shp */
1586 if (!ipc_valid_object(&shp
->shm_perm
)) {
1587 ipc_unlock_object(&shp
->shm_perm
);
1593 * We need to take a reference to the real shm file to prevent the
1594 * pointer from becoming stale in cases where the lifetime of the outer
1595 * file extends beyond that of the shm segment. It's not usually
1596 * possible, but it can happen during remap_file_pages() emulation as
1597 * that unmaps the memory, then does ->mmap() via file reference only.
1598 * We'll deny the ->mmap() if the shm segment was since removed, but to
1599 * detect shm ID reuse we need to compare the file pointers.
1601 base
= get_file(shp
->shm_file
);
1603 size
= i_size_read(file_inode(base
));
1604 ipc_unlock_object(&shp
->shm_perm
);
1608 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1614 file
= alloc_file_clone(base
, f_flags
,
1615 is_file_hugepages(base
) ?
1616 &shm_file_operations_huge
:
1617 &shm_file_operations
);
1618 err
= PTR_ERR(file
);
1625 sfd
->id
= shp
->shm_perm
.id
;
1626 sfd
->ns
= get_ipc_ns(ns
);
1629 file
->private_data
= sfd
;
1631 err
= security_mmap_file(file
, prot
, flags
);
1635 if (mmap_write_lock_killable(current
->mm
)) {
1640 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1642 if (addr
+ size
< addr
)
1645 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1649 addr
= do_mmap(file
, addr
, size
, prot
, flags
, 0, &populate
, NULL
);
1652 if (IS_ERR_VALUE(addr
))
1655 mmap_write_unlock(current
->mm
);
1657 mm_populate(addr
, populate
);
1663 down_write(&shm_ids(ns
).rwsem
);
1664 shp
= shm_lock(ns
, shmid
);
1667 if (shm_may_destroy(shp
))
1668 shm_destroy(ns
, shp
);
1671 up_write(&shm_ids(ns
).rwsem
);
1680 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1685 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
, SHMLBA
);
1688 force_successful_syscall_return();
1692 #ifdef CONFIG_COMPAT
1694 #ifndef COMPAT_SHMLBA
1695 #define COMPAT_SHMLBA SHMLBA
1698 COMPAT_SYSCALL_DEFINE3(shmat
, int, shmid
, compat_uptr_t
, shmaddr
, int, shmflg
)
1703 err
= do_shmat(shmid
, compat_ptr(shmaddr
), shmflg
, &ret
, COMPAT_SHMLBA
);
1706 force_successful_syscall_return();
1712 * detach and kill segment if marked destroyed.
1713 * The work is done in shm_close.
1715 long ksys_shmdt(char __user
*shmaddr
)
1717 struct mm_struct
*mm
= current
->mm
;
1718 struct vm_area_struct
*vma
;
1719 unsigned long addr
= (unsigned long)shmaddr
;
1720 int retval
= -EINVAL
;
1724 struct vm_area_struct
*next
;
1727 if (addr
& ~PAGE_MASK
)
1730 if (mmap_write_lock_killable(mm
))
1734 * This function tries to be smart and unmap shm segments that
1735 * were modified by partial mlock or munmap calls:
1736 * - It first determines the size of the shm segment that should be
1737 * unmapped: It searches for a vma that is backed by shm and that
1738 * started at address shmaddr. It records it's size and then unmaps
1740 * - Then it unmaps all shm vmas that started at shmaddr and that
1741 * are within the initially determined size and that are from the
1742 * same shm segment from which we determined the size.
1743 * Errors from do_munmap are ignored: the function only fails if
1744 * it's called with invalid parameters or if it's called to unmap
1745 * a part of a vma. Both calls in this function are for full vmas,
1746 * the parameters are directly copied from the vma itself and always
1747 * valid - therefore do_munmap cannot fail. (famous last words?)
1750 * If it had been mremap()'d, the starting address would not
1751 * match the usual checks anyway. So assume all vma's are
1752 * above the starting address given.
1754 vma
= find_vma(mm
, addr
);
1758 next
= vma
->vm_next
;
1761 * Check if the starting address would match, i.e. it's
1762 * a fragment created by mprotect() and/or munmap(), or it
1763 * otherwise it starts at this address with no hassles.
1765 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1766 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1769 * Record the file of the shm segment being
1770 * unmapped. With mremap(), someone could place
1771 * page from another segment but with equal offsets
1772 * in the range we are unmapping.
1774 file
= vma
->vm_file
;
1775 size
= i_size_read(file_inode(vma
->vm_file
));
1776 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1778 * We discovered the size of the shm segment, so
1779 * break out of here and fall through to the next
1780 * loop that uses the size information to stop
1781 * searching for matching vma's.
1791 * We need look no further than the maximum address a fragment
1792 * could possibly have landed at. Also cast things to loff_t to
1793 * prevent overflows and make comparisons vs. equal-width types.
1795 size
= PAGE_ALIGN(size
);
1796 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1797 next
= vma
->vm_next
;
1799 /* finding a matching vma now does not alter retval */
1800 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1801 ((vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) &&
1802 (vma
->vm_file
== file
))
1803 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1807 #else /* CONFIG_MMU */
1808 /* under NOMMU conditions, the exact address to be destroyed must be
1811 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1812 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1818 mmap_write_unlock(mm
);
1822 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1824 return ksys_shmdt(shmaddr
);
1827 #ifdef CONFIG_PROC_FS
1828 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1830 struct pid_namespace
*pid_ns
= ipc_seq_pid_ns(s
);
1831 struct user_namespace
*user_ns
= seq_user_ns(s
);
1832 struct kern_ipc_perm
*ipcp
= it
;
1833 struct shmid_kernel
*shp
;
1834 unsigned long rss
= 0, swp
= 0;
1836 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
1837 shm_add_rss_swap(shp
, &rss
, &swp
);
1839 #if BITS_PER_LONG <= 32
1840 #define SIZE_SPEC "%10lu"
1842 #define SIZE_SPEC "%21lu"
1846 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1847 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1848 SIZE_SPEC
" " SIZE_SPEC
"\n",
1853 pid_nr_ns(shp
->shm_cprid
, pid_ns
),
1854 pid_nr_ns(shp
->shm_lprid
, pid_ns
),
1856 from_kuid_munged(user_ns
, shp
->shm_perm
.uid
),
1857 from_kgid_munged(user_ns
, shp
->shm_perm
.gid
),
1858 from_kuid_munged(user_ns
, shp
->shm_perm
.cuid
),
1859 from_kgid_munged(user_ns
, shp
->shm_perm
.cgid
),