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>
47 #include <linux/uaccess.h>
51 struct shm_file_data
{
53 struct ipc_namespace
*ns
;
55 const struct vm_operations_struct
*vm_ops
;
58 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
60 static const struct file_operations shm_file_operations
;
61 static const struct vm_operations_struct shm_vm_ops
;
63 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
65 #define shm_unlock(shp) \
66 ipc_unlock(&(shp)->shm_perm)
68 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
69 static void shm_open(struct vm_area_struct
*vma
);
70 static void shm_close(struct vm_area_struct
*vma
);
71 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
73 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
76 int shm_init_ns(struct ipc_namespace
*ns
)
78 ns
->shm_ctlmax
= SHMMAX
;
79 ns
->shm_ctlall
= SHMALL
;
80 ns
->shm_ctlmni
= SHMMNI
;
81 ns
->shm_rmid_forced
= 0;
83 return ipc_init_ids(&shm_ids(ns
));
87 * Called with shm_ids.rwsem (writer) and the shp structure locked.
88 * Only shm_ids.rwsem remains locked on exit.
90 static void do_shm_rmid(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
92 struct shmid_kernel
*shp
;
94 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
96 if (shp
->shm_nattch
) {
97 shp
->shm_perm
.mode
|= SHM_DEST
;
98 /* Do not find it any more */
99 ipc_set_key_private(&shm_ids(ns
), &shp
->shm_perm
);
102 shm_destroy(ns
, shp
);
106 void shm_exit_ns(struct ipc_namespace
*ns
)
108 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
109 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
110 rhashtable_destroy(&ns
->ids
[IPC_SHM_IDS
].key_ht
);
114 static int __init
ipc_ns_init(void)
116 const int err
= shm_init_ns(&init_ipc_ns
);
117 WARN(err
, "ipc: sysv shm_init_ns failed: %d\n", err
);
121 pure_initcall(ipc_ns_init
);
123 void __init
shm_init(void)
125 ipc_init_proc_interface("sysvipc/shm",
126 #if BITS_PER_LONG <= 32
127 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
129 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
131 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
134 static inline struct shmid_kernel
*shm_obtain_object(struct ipc_namespace
*ns
, int id
)
136 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_idr(&shm_ids(ns
), id
);
139 return ERR_CAST(ipcp
);
141 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
144 static inline struct shmid_kernel
*shm_obtain_object_check(struct ipc_namespace
*ns
, int id
)
146 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&shm_ids(ns
), id
);
149 return ERR_CAST(ipcp
);
151 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
155 * shm_lock_(check_) routines are called in the paths where the rwsem
156 * is not necessarily held.
158 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
160 struct kern_ipc_perm
*ipcp
= ipc_lock(&shm_ids(ns
), id
);
163 * Callers of shm_lock() must validate the status of the returned ipc
164 * object pointer (as returned by ipc_lock()), and error out as
169 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
172 static inline void shm_lock_by_ptr(struct shmid_kernel
*ipcp
)
175 ipc_lock_object(&ipcp
->shm_perm
);
178 static void shm_rcu_free(struct rcu_head
*head
)
180 struct kern_ipc_perm
*ptr
= container_of(head
, struct kern_ipc_perm
,
182 struct shmid_kernel
*shp
= container_of(ptr
, struct shmid_kernel
,
184 security_shm_free(shp
);
188 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
190 list_del(&s
->shm_clist
);
191 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
195 static int __shm_open(struct vm_area_struct
*vma
)
197 struct file
*file
= vma
->vm_file
;
198 struct shm_file_data
*sfd
= shm_file_data(file
);
199 struct shmid_kernel
*shp
;
201 shp
= shm_lock(sfd
->ns
, sfd
->id
);
206 if (shp
->shm_file
!= sfd
->file
) {
212 shp
->shm_atim
= ktime_get_real_seconds();
213 shp
->shm_lprid
= task_tgid_vnr(current
);
219 /* This is called by fork, once for every shm attach. */
220 static void shm_open(struct vm_area_struct
*vma
)
222 int err
= __shm_open(vma
);
224 * We raced in the idr lookup or with shm_destroy().
225 * Either way, the ID is busted.
231 * shm_destroy - free the struct shmid_kernel
234 * @shp: struct to free
236 * It has to be called with shp and shm_ids.rwsem (writer) locked,
237 * but returns with shp unlocked and freed.
239 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
241 struct file
*shm_file
;
243 shm_file
= shp
->shm_file
;
244 shp
->shm_file
= NULL
;
245 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
248 if (!is_file_hugepages(shm_file
))
249 shmem_lock(shm_file
, 0, shp
->mlock_user
);
250 else if (shp
->mlock_user
)
251 user_shm_unlock(i_size_read(file_inode(shm_file
)),
254 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
258 * shm_may_destroy - identifies whether shm segment should be destroyed now
260 * Returns true if and only if there are no active users of the segment and
261 * one of the following is true:
263 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
265 * 2) sysctl kernel.shm_rmid_forced is set to 1.
267 static bool shm_may_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
269 return (shp
->shm_nattch
== 0) &&
270 (ns
->shm_rmid_forced
||
271 (shp
->shm_perm
.mode
& SHM_DEST
));
275 * remove the attach descriptor vma.
276 * free memory for segment if it is marked destroyed.
277 * The descriptor has already been removed from the current->mm->mmap list
278 * and will later be kfree()d.
280 static void shm_close(struct vm_area_struct
*vma
)
282 struct file
*file
= vma
->vm_file
;
283 struct shm_file_data
*sfd
= shm_file_data(file
);
284 struct shmid_kernel
*shp
;
285 struct ipc_namespace
*ns
= sfd
->ns
;
287 down_write(&shm_ids(ns
).rwsem
);
288 /* remove from the list of attaches of the shm segment */
289 shp
= shm_lock(ns
, sfd
->id
);
292 * We raced in the idr lookup or with shm_destroy().
293 * Either way, the ID is busted.
295 if (WARN_ON_ONCE(IS_ERR(shp
)))
296 goto done
; /* no-op */
298 shp
->shm_lprid
= task_tgid_vnr(current
);
299 shp
->shm_dtim
= ktime_get_real_seconds();
301 if (shm_may_destroy(ns
, shp
))
302 shm_destroy(ns
, shp
);
306 up_write(&shm_ids(ns
).rwsem
);
309 /* Called with ns->shm_ids(ns).rwsem locked */
310 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
312 struct ipc_namespace
*ns
= data
;
313 struct kern_ipc_perm
*ipcp
= p
;
314 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
317 * We want to destroy segments without users and with already
318 * exit'ed originating process.
320 * As shp->* are changed under rwsem, it's safe to skip shp locking.
322 if (shp
->shm_creator
!= NULL
)
325 if (shm_may_destroy(ns
, shp
)) {
326 shm_lock_by_ptr(shp
);
327 shm_destroy(ns
, shp
);
332 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
334 down_write(&shm_ids(ns
).rwsem
);
335 if (shm_ids(ns
).in_use
)
336 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
337 up_write(&shm_ids(ns
).rwsem
);
340 /* Locking assumes this will only be called with task == current */
341 void exit_shm(struct task_struct
*task
)
343 struct ipc_namespace
*ns
= task
->nsproxy
->ipc_ns
;
344 struct shmid_kernel
*shp
, *n
;
346 if (list_empty(&task
->sysvshm
.shm_clist
))
350 * If kernel.shm_rmid_forced is not set then only keep track of
351 * which shmids are orphaned, so that a later set of the sysctl
354 if (!ns
->shm_rmid_forced
) {
355 down_read(&shm_ids(ns
).rwsem
);
356 list_for_each_entry(shp
, &task
->sysvshm
.shm_clist
, shm_clist
)
357 shp
->shm_creator
= NULL
;
359 * Only under read lock but we are only called on current
360 * so no entry on the list will be shared.
362 list_del(&task
->sysvshm
.shm_clist
);
363 up_read(&shm_ids(ns
).rwsem
);
368 * Destroy all already created segments, that were not yet mapped,
369 * and mark any mapped as orphan to cover the sysctl toggling.
370 * Destroy is skipped if shm_may_destroy() returns false.
372 down_write(&shm_ids(ns
).rwsem
);
373 list_for_each_entry_safe(shp
, n
, &task
->sysvshm
.shm_clist
, shm_clist
) {
374 shp
->shm_creator
= NULL
;
376 if (shm_may_destroy(ns
, shp
)) {
377 shm_lock_by_ptr(shp
);
378 shm_destroy(ns
, shp
);
382 /* Remove the list head from any segments still attached. */
383 list_del(&task
->sysvshm
.shm_clist
);
384 up_write(&shm_ids(ns
).rwsem
);
387 static int shm_fault(struct vm_fault
*vmf
)
389 struct file
*file
= vmf
->vma
->vm_file
;
390 struct shm_file_data
*sfd
= shm_file_data(file
);
392 return sfd
->vm_ops
->fault(vmf
);
395 static int shm_split(struct vm_area_struct
*vma
, unsigned long addr
)
397 struct file
*file
= vma
->vm_file
;
398 struct shm_file_data
*sfd
= shm_file_data(file
);
400 if (sfd
->vm_ops
&& sfd
->vm_ops
->split
)
401 return sfd
->vm_ops
->split(vma
, addr
);
407 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
409 struct file
*file
= vma
->vm_file
;
410 struct shm_file_data
*sfd
= shm_file_data(file
);
413 if (sfd
->vm_ops
->set_policy
)
414 err
= sfd
->vm_ops
->set_policy(vma
, new);
418 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
421 struct file
*file
= vma
->vm_file
;
422 struct shm_file_data
*sfd
= shm_file_data(file
);
423 struct mempolicy
*pol
= NULL
;
425 if (sfd
->vm_ops
->get_policy
)
426 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
427 else if (vma
->vm_policy
)
428 pol
= vma
->vm_policy
;
434 static int shm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
436 struct shm_file_data
*sfd
= shm_file_data(file
);
440 * In case of remap_file_pages() emulation, the file can represent an
441 * IPC ID that was removed, and possibly even reused by another shm
442 * segment already. Propagate this case as an error to caller.
444 ret
= __shm_open(vma
);
448 ret
= call_mmap(sfd
->file
, vma
);
453 sfd
->vm_ops
= vma
->vm_ops
;
455 WARN_ON(!sfd
->vm_ops
->fault
);
457 vma
->vm_ops
= &shm_vm_ops
;
461 static int shm_release(struct inode
*ino
, struct file
*file
)
463 struct shm_file_data
*sfd
= shm_file_data(file
);
467 shm_file_data(file
) = NULL
;
472 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
474 struct shm_file_data
*sfd
= shm_file_data(file
);
476 if (!sfd
->file
->f_op
->fsync
)
478 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
481 static long shm_fallocate(struct file
*file
, int mode
, loff_t offset
,
484 struct shm_file_data
*sfd
= shm_file_data(file
);
486 if (!sfd
->file
->f_op
->fallocate
)
488 return sfd
->file
->f_op
->fallocate(file
, mode
, offset
, len
);
491 static unsigned long shm_get_unmapped_area(struct file
*file
,
492 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
495 struct shm_file_data
*sfd
= shm_file_data(file
);
497 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
501 static const struct file_operations shm_file_operations
= {
504 .release
= shm_release
,
505 .get_unmapped_area
= shm_get_unmapped_area
,
506 .llseek
= noop_llseek
,
507 .fallocate
= shm_fallocate
,
511 * shm_file_operations_huge is now identical to shm_file_operations,
512 * but we keep it distinct for the sake of is_file_shm_hugepages().
514 static const struct file_operations shm_file_operations_huge
= {
517 .release
= shm_release
,
518 .get_unmapped_area
= shm_get_unmapped_area
,
519 .llseek
= noop_llseek
,
520 .fallocate
= shm_fallocate
,
523 bool is_file_shm_hugepages(struct file
*file
)
525 return file
->f_op
== &shm_file_operations_huge
;
528 static const struct vm_operations_struct shm_vm_ops
= {
529 .open
= shm_open
, /* callback for a new vm-area open */
530 .close
= shm_close
, /* callback for when the vm-area is released */
533 #if defined(CONFIG_NUMA)
534 .set_policy
= shm_set_policy
,
535 .get_policy
= shm_get_policy
,
540 * newseg - Create a new shared memory segment
542 * @params: ptr to the structure that contains key, size and shmflg
544 * Called with shm_ids.rwsem held as a writer.
546 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
548 key_t key
= params
->key
;
549 int shmflg
= params
->flg
;
550 size_t size
= params
->u
.size
;
552 struct shmid_kernel
*shp
;
553 size_t numpages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
556 vm_flags_t acctflag
= 0;
558 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
561 if (numpages
<< PAGE_SHIFT
< size
)
564 if (ns
->shm_tot
+ numpages
< ns
->shm_tot
||
565 ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
568 shp
= kvmalloc(sizeof(*shp
), GFP_KERNEL
);
572 shp
->shm_perm
.key
= key
;
573 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
574 shp
->mlock_user
= NULL
;
576 shp
->shm_perm
.security
= NULL
;
577 error
= security_shm_alloc(shp
);
583 sprintf(name
, "SYSV%08x", key
);
584 if (shmflg
& SHM_HUGETLB
) {
588 hs
= hstate_sizelog((shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
593 hugesize
= ALIGN(size
, huge_page_size(hs
));
595 /* hugetlb_file_setup applies strict accounting */
596 if (shmflg
& SHM_NORESERVE
)
597 acctflag
= VM_NORESERVE
;
598 file
= hugetlb_file_setup(name
, hugesize
, acctflag
,
599 &shp
->mlock_user
, HUGETLB_SHMFS_INODE
,
600 (shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
603 * Do not allow no accounting for OVERCOMMIT_NEVER, even
606 if ((shmflg
& SHM_NORESERVE
) &&
607 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
608 acctflag
= VM_NORESERVE
;
609 file
= shmem_kernel_file_setup(name
, size
, acctflag
);
611 error
= PTR_ERR(file
);
615 shp
->shm_cprid
= task_tgid_vnr(current
);
617 shp
->shm_atim
= shp
->shm_dtim
= 0;
618 shp
->shm_ctim
= ktime_get_real_seconds();
619 shp
->shm_segsz
= size
;
621 shp
->shm_file
= file
;
622 shp
->shm_creator
= current
;
624 error
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
628 list_add(&shp
->shm_clist
, ¤t
->sysvshm
.shm_clist
);
631 * shmid gets reported as "inode#" in /proc/pid/maps.
632 * proc-ps tools use this. Changing this will break them.
634 file_inode(file
)->i_ino
= shp
->shm_perm
.id
;
636 ns
->shm_tot
+= numpages
;
637 error
= shp
->shm_perm
.id
;
639 ipc_unlock_object(&shp
->shm_perm
);
644 if (is_file_hugepages(file
) && shp
->mlock_user
)
645 user_shm_unlock(size
, shp
->mlock_user
);
648 call_rcu(&shp
->shm_perm
.rcu
, shm_rcu_free
);
653 * Called with shm_ids.rwsem and ipcp locked.
655 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
657 struct shmid_kernel
*shp
;
659 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
660 return security_shm_associate(shp
, shmflg
);
664 * Called with shm_ids.rwsem and ipcp locked.
666 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
667 struct ipc_params
*params
)
669 struct shmid_kernel
*shp
;
671 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
672 if (shp
->shm_segsz
< params
->u
.size
)
678 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
680 struct ipc_namespace
*ns
;
681 static const struct ipc_ops shm_ops
= {
683 .associate
= shm_security
,
684 .more_checks
= shm_more_checks
,
686 struct ipc_params shm_params
;
688 ns
= current
->nsproxy
->ipc_ns
;
690 shm_params
.key
= key
;
691 shm_params
.flg
= shmflg
;
692 shm_params
.u
.size
= size
;
694 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
697 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
701 return copy_to_user(buf
, in
, sizeof(*in
));
706 memset(&out
, 0, sizeof(out
));
707 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
708 out
.shm_segsz
= in
->shm_segsz
;
709 out
.shm_atime
= in
->shm_atime
;
710 out
.shm_dtime
= in
->shm_dtime
;
711 out
.shm_ctime
= in
->shm_ctime
;
712 out
.shm_cpid
= in
->shm_cpid
;
713 out
.shm_lpid
= in
->shm_lpid
;
714 out
.shm_nattch
= in
->shm_nattch
;
716 return copy_to_user(buf
, &out
, sizeof(out
));
723 static inline unsigned long
724 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
728 if (copy_from_user(out
, buf
, sizeof(*out
)))
733 struct shmid_ds tbuf_old
;
735 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
738 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
739 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
740 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
749 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
753 return copy_to_user(buf
, in
, sizeof(*in
));
758 if (in
->shmmax
> INT_MAX
)
759 out
.shmmax
= INT_MAX
;
761 out
.shmmax
= (int)in
->shmmax
;
763 out
.shmmin
= in
->shmmin
;
764 out
.shmmni
= in
->shmmni
;
765 out
.shmseg
= in
->shmseg
;
766 out
.shmall
= in
->shmall
;
768 return copy_to_user(buf
, &out
, sizeof(out
));
776 * Calculate and add used RSS and swap pages of a shm.
777 * Called with shm_ids.rwsem held as a reader
779 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
780 unsigned long *rss_add
, unsigned long *swp_add
)
784 inode
= file_inode(shp
->shm_file
);
786 if (is_file_hugepages(shp
->shm_file
)) {
787 struct address_space
*mapping
= inode
->i_mapping
;
788 struct hstate
*h
= hstate_file(shp
->shm_file
);
789 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
792 struct shmem_inode_info
*info
= SHMEM_I(inode
);
794 spin_lock_irq(&info
->lock
);
795 *rss_add
+= inode
->i_mapping
->nrpages
;
796 *swp_add
+= info
->swapped
;
797 spin_unlock_irq(&info
->lock
);
799 *rss_add
+= inode
->i_mapping
->nrpages
;
805 * Called with shm_ids.rwsem held as a reader
807 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
816 in_use
= shm_ids(ns
).in_use
;
818 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
819 struct kern_ipc_perm
*ipc
;
820 struct shmid_kernel
*shp
;
822 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
825 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
827 shm_add_rss_swap(shp
, rss
, swp
);
834 * This function handles some shmctl commands which require the rwsem
835 * to be held in write mode.
836 * NOTE: no locks must be held, the rwsem is taken inside this function.
838 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
839 struct shmid64_ds
*shmid64
)
841 struct kern_ipc_perm
*ipcp
;
842 struct shmid_kernel
*shp
;
845 down_write(&shm_ids(ns
).rwsem
);
848 ipcp
= ipcctl_pre_down_nolock(ns
, &shm_ids(ns
), shmid
, cmd
,
849 &shmid64
->shm_perm
, 0);
855 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
857 err
= security_shm_shmctl(shp
, cmd
);
863 ipc_lock_object(&shp
->shm_perm
);
864 /* do_shm_rmid unlocks the ipc object and rcu */
865 do_shm_rmid(ns
, ipcp
);
868 ipc_lock_object(&shp
->shm_perm
);
869 err
= ipc_update_perm(&shmid64
->shm_perm
, ipcp
);
872 shp
->shm_ctim
= ktime_get_real_seconds();
880 ipc_unlock_object(&shp
->shm_perm
);
884 up_write(&shm_ids(ns
).rwsem
);
888 static int shmctl_ipc_info(struct ipc_namespace
*ns
,
889 struct shminfo64
*shminfo
)
891 int err
= security_shm_shmctl(NULL
, IPC_INFO
);
893 memset(shminfo
, 0, sizeof(*shminfo
));
894 shminfo
->shmmni
= shminfo
->shmseg
= ns
->shm_ctlmni
;
895 shminfo
->shmmax
= ns
->shm_ctlmax
;
896 shminfo
->shmall
= ns
->shm_ctlall
;
897 shminfo
->shmmin
= SHMMIN
;
898 down_read(&shm_ids(ns
).rwsem
);
899 err
= ipc_get_maxid(&shm_ids(ns
));
900 up_read(&shm_ids(ns
).rwsem
);
907 static int shmctl_shm_info(struct ipc_namespace
*ns
,
908 struct shm_info
*shm_info
)
910 int err
= security_shm_shmctl(NULL
, SHM_INFO
);
912 memset(shm_info
, 0, sizeof(*shm_info
));
913 down_read(&shm_ids(ns
).rwsem
);
914 shm_info
->used_ids
= shm_ids(ns
).in_use
;
915 shm_get_stat(ns
, &shm_info
->shm_rss
, &shm_info
->shm_swp
);
916 shm_info
->shm_tot
= ns
->shm_tot
;
917 shm_info
->swap_attempts
= 0;
918 shm_info
->swap_successes
= 0;
919 err
= ipc_get_maxid(&shm_ids(ns
));
920 up_read(&shm_ids(ns
).rwsem
);
927 static int shmctl_stat(struct ipc_namespace
*ns
, int shmid
,
928 int cmd
, struct shmid64_ds
*tbuf
)
930 struct shmid_kernel
*shp
;
935 if (cmd
== SHM_STAT
) {
936 shp
= shm_obtain_object(ns
, shmid
);
941 result
= shp
->shm_perm
.id
;
943 shp
= shm_obtain_object_check(ns
, shmid
);
952 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
955 err
= security_shm_shmctl(shp
, cmd
);
959 memset(tbuf
, 0, sizeof(*tbuf
));
960 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
->shm_perm
);
961 tbuf
->shm_segsz
= shp
->shm_segsz
;
962 tbuf
->shm_atime
= shp
->shm_atim
;
963 tbuf
->shm_dtime
= shp
->shm_dtim
;
964 tbuf
->shm_ctime
= shp
->shm_ctim
;
965 tbuf
->shm_cpid
= shp
->shm_cprid
;
966 tbuf
->shm_lpid
= shp
->shm_lprid
;
967 tbuf
->shm_nattch
= shp
->shm_nattch
;
976 static int shmctl_do_lock(struct ipc_namespace
*ns
, int shmid
, int cmd
)
978 struct shmid_kernel
*shp
;
979 struct file
*shm_file
;
983 shp
= shm_obtain_object_check(ns
, shmid
);
989 audit_ipc_obj(&(shp
->shm_perm
));
990 err
= security_shm_shmctl(shp
, cmd
);
994 ipc_lock_object(&shp
->shm_perm
);
996 /* check if shm_destroy() is tearing down shp */
997 if (!ipc_valid_object(&shp
->shm_perm
)) {
1002 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
1003 kuid_t euid
= current_euid();
1005 if (!uid_eq(euid
, shp
->shm_perm
.uid
) &&
1006 !uid_eq(euid
, shp
->shm_perm
.cuid
)) {
1010 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
)) {
1016 shm_file
= shp
->shm_file
;
1017 if (is_file_hugepages(shm_file
))
1020 if (cmd
== SHM_LOCK
) {
1021 struct user_struct
*user
= current_user();
1023 err
= shmem_lock(shm_file
, 1, user
);
1024 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)) {
1025 shp
->shm_perm
.mode
|= SHM_LOCKED
;
1026 shp
->mlock_user
= user
;
1032 if (!(shp
->shm_perm
.mode
& SHM_LOCKED
))
1034 shmem_lock(shm_file
, 0, shp
->mlock_user
);
1035 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
1036 shp
->mlock_user
= NULL
;
1038 ipc_unlock_object(&shp
->shm_perm
);
1040 shmem_unlock_mapping(shm_file
->f_mapping
);
1046 ipc_unlock_object(&shp
->shm_perm
);
1052 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
1055 struct ipc_namespace
*ns
;
1056 struct shmid64_ds sem64
;
1058 if (cmd
< 0 || shmid
< 0)
1061 version
= ipc_parse_version(&cmd
);
1062 ns
= current
->nsproxy
->ipc_ns
;
1066 struct shminfo64 shminfo
;
1067 err
= shmctl_ipc_info(ns
, &shminfo
);
1070 if (copy_shminfo_to_user(buf
, &shminfo
, version
))
1075 struct shm_info shm_info
;
1076 err
= shmctl_shm_info(ns
, &shm_info
);
1079 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
)))
1085 err
= shmctl_stat(ns
, shmid
, cmd
, &sem64
);
1088 if (copy_shmid_to_user(buf
, &sem64
, version
))
1093 if (copy_shmid_from_user(&sem64
, buf
, version
))
1097 return shmctl_down(ns
, shmid
, cmd
, &sem64
);
1100 return shmctl_do_lock(ns
, shmid
, cmd
);
1106 #ifdef CONFIG_COMPAT
1108 struct compat_shmid_ds
{
1109 struct compat_ipc_perm shm_perm
;
1111 compat_time_t shm_atime
;
1112 compat_time_t shm_dtime
;
1113 compat_time_t shm_ctime
;
1114 compat_ipc_pid_t shm_cpid
;
1115 compat_ipc_pid_t shm_lpid
;
1116 unsigned short shm_nattch
;
1117 unsigned short shm_unused
;
1118 compat_uptr_t shm_unused2
;
1119 compat_uptr_t shm_unused3
;
1122 struct compat_shminfo64
{
1123 compat_ulong_t shmmax
;
1124 compat_ulong_t shmmin
;
1125 compat_ulong_t shmmni
;
1126 compat_ulong_t shmseg
;
1127 compat_ulong_t shmall
;
1128 compat_ulong_t __unused1
;
1129 compat_ulong_t __unused2
;
1130 compat_ulong_t __unused3
;
1131 compat_ulong_t __unused4
;
1134 struct compat_shm_info
{
1135 compat_int_t used_ids
;
1136 compat_ulong_t shm_tot
, shm_rss
, shm_swp
;
1137 compat_ulong_t swap_attempts
, swap_successes
;
1140 static int copy_compat_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
,
1143 if (in
->shmmax
> INT_MAX
)
1144 in
->shmmax
= INT_MAX
;
1145 if (version
== IPC_64
) {
1146 struct compat_shminfo64 info
;
1147 memset(&info
, 0, sizeof(info
));
1148 info
.shmmax
= in
->shmmax
;
1149 info
.shmmin
= in
->shmmin
;
1150 info
.shmmni
= in
->shmmni
;
1151 info
.shmseg
= in
->shmseg
;
1152 info
.shmall
= in
->shmall
;
1153 return copy_to_user(buf
, &info
, sizeof(info
));
1155 struct shminfo info
;
1156 memset(&info
, 0, sizeof(info
));
1157 info
.shmmax
= in
->shmmax
;
1158 info
.shmmin
= in
->shmmin
;
1159 info
.shmmni
= in
->shmmni
;
1160 info
.shmseg
= in
->shmseg
;
1161 info
.shmall
= in
->shmall
;
1162 return copy_to_user(buf
, &info
, sizeof(info
));
1166 static int put_compat_shm_info(struct shm_info
*ip
,
1167 struct compat_shm_info __user
*uip
)
1169 struct compat_shm_info info
;
1171 memset(&info
, 0, sizeof(info
));
1172 info
.used_ids
= ip
->used_ids
;
1173 info
.shm_tot
= ip
->shm_tot
;
1174 info
.shm_rss
= ip
->shm_rss
;
1175 info
.shm_swp
= ip
->shm_swp
;
1176 info
.swap_attempts
= ip
->swap_attempts
;
1177 info
.swap_successes
= ip
->swap_successes
;
1178 return copy_to_user(uip
, &info
, sizeof(info
));
1181 static int copy_compat_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
,
1184 if (version
== IPC_64
) {
1185 struct compat_shmid64_ds v
;
1186 memset(&v
, 0, sizeof(v
));
1187 to_compat_ipc64_perm(&v
.shm_perm
, &in
->shm_perm
);
1188 v
.shm_atime
= in
->shm_atime
;
1189 v
.shm_dtime
= in
->shm_dtime
;
1190 v
.shm_ctime
= in
->shm_ctime
;
1191 v
.shm_segsz
= in
->shm_segsz
;
1192 v
.shm_nattch
= in
->shm_nattch
;
1193 v
.shm_cpid
= in
->shm_cpid
;
1194 v
.shm_lpid
= in
->shm_lpid
;
1195 return copy_to_user(buf
, &v
, sizeof(v
));
1197 struct compat_shmid_ds v
;
1198 memset(&v
, 0, sizeof(v
));
1199 to_compat_ipc_perm(&v
.shm_perm
, &in
->shm_perm
);
1200 v
.shm_perm
.key
= in
->shm_perm
.key
;
1201 v
.shm_atime
= in
->shm_atime
;
1202 v
.shm_dtime
= in
->shm_dtime
;
1203 v
.shm_ctime
= in
->shm_ctime
;
1204 v
.shm_segsz
= in
->shm_segsz
;
1205 v
.shm_nattch
= in
->shm_nattch
;
1206 v
.shm_cpid
= in
->shm_cpid
;
1207 v
.shm_lpid
= in
->shm_lpid
;
1208 return copy_to_user(buf
, &v
, sizeof(v
));
1212 static int copy_compat_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
,
1215 memset(out
, 0, sizeof(*out
));
1216 if (version
== IPC_64
) {
1217 struct compat_shmid64_ds
*p
= buf
;
1218 return get_compat_ipc64_perm(&out
->shm_perm
, &p
->shm_perm
);
1220 struct compat_shmid_ds
*p
= buf
;
1221 return get_compat_ipc_perm(&out
->shm_perm
, &p
->shm_perm
);
1225 COMPAT_SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, void __user
*, uptr
)
1227 struct ipc_namespace
*ns
;
1228 struct shmid64_ds sem64
;
1229 int version
= compat_ipc_parse_version(&cmd
);
1232 ns
= current
->nsproxy
->ipc_ns
;
1234 if (cmd
< 0 || shmid
< 0)
1239 struct shminfo64 shminfo
;
1240 err
= shmctl_ipc_info(ns
, &shminfo
);
1243 if (copy_compat_shminfo_to_user(uptr
, &shminfo
, version
))
1248 struct shm_info shm_info
;
1249 err
= shmctl_shm_info(ns
, &shm_info
);
1252 if (put_compat_shm_info(&shm_info
, uptr
))
1258 err
= shmctl_stat(ns
, shmid
, cmd
, &sem64
);
1261 if (copy_compat_shmid_to_user(uptr
, &sem64
, version
))
1266 if (copy_compat_shmid_from_user(&sem64
, uptr
, version
))
1270 return shmctl_down(ns
, shmid
, cmd
, &sem64
);
1273 return shmctl_do_lock(ns
, shmid
, cmd
);
1283 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1285 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1286 * "raddr" thing points to kernel space, and there has to be a wrapper around
1289 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
,
1290 ulong
*raddr
, unsigned long shmlba
)
1292 struct shmid_kernel
*shp
;
1293 unsigned long addr
= (unsigned long)shmaddr
;
1297 unsigned long flags
= MAP_SHARED
;
1300 struct ipc_namespace
*ns
;
1301 struct shm_file_data
*sfd
;
1304 unsigned long populate
= 0;
1311 if (addr
& (shmlba
- 1)) {
1312 if (shmflg
& SHM_RND
) {
1313 addr
&= ~(shmlba
- 1); /* round down */
1316 * Ensure that the round-down is non-nil
1317 * when remapping. This can happen for
1318 * cases when addr < shmlba.
1320 if (!addr
&& (shmflg
& SHM_REMAP
))
1323 #ifndef __ARCH_FORCE_SHMLBA
1324 if (addr
& ~PAGE_MASK
)
1330 } else if ((shmflg
& SHM_REMAP
))
1333 if (shmflg
& SHM_RDONLY
) {
1336 f_mode
= FMODE_READ
;
1338 prot
= PROT_READ
| PROT_WRITE
;
1339 acc_mode
= S_IRUGO
| S_IWUGO
;
1340 f_mode
= FMODE_READ
| FMODE_WRITE
;
1342 if (shmflg
& SHM_EXEC
) {
1344 acc_mode
|= S_IXUGO
;
1348 * We cannot rely on the fs check since SYSV IPC does have an
1349 * additional creator id...
1351 ns
= current
->nsproxy
->ipc_ns
;
1353 shp
= shm_obtain_object_check(ns
, shmid
);
1360 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
1363 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
1367 ipc_lock_object(&shp
->shm_perm
);
1369 /* check if shm_destroy() is tearing down shp */
1370 if (!ipc_valid_object(&shp
->shm_perm
)) {
1371 ipc_unlock_object(&shp
->shm_perm
);
1376 path
= shp
->shm_file
->f_path
;
1379 size
= i_size_read(d_inode(path
.dentry
));
1380 ipc_unlock_object(&shp
->shm_perm
);
1384 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1390 file
= alloc_file(&path
, f_mode
,
1391 is_file_hugepages(shp
->shm_file
) ?
1392 &shm_file_operations_huge
:
1393 &shm_file_operations
);
1394 err
= PTR_ERR(file
);
1401 file
->private_data
= sfd
;
1402 file
->f_mapping
= shp
->shm_file
->f_mapping
;
1403 sfd
->id
= shp
->shm_perm
.id
;
1404 sfd
->ns
= get_ipc_ns(ns
);
1406 * We need to take a reference to the real shm file to prevent the
1407 * pointer from becoming stale in cases where the lifetime of the outer
1408 * file extends beyond that of the shm segment. It's not usually
1409 * possible, but it can happen during remap_file_pages() emulation as
1410 * that unmaps the memory, then does ->mmap() via file reference only.
1411 * We'll deny the ->mmap() if the shm segment was since removed, but to
1412 * detect shm ID reuse we need to compare the file pointers.
1414 sfd
->file
= get_file(shp
->shm_file
);
1417 err
= security_mmap_file(file
, prot
, flags
);
1421 if (down_write_killable(¤t
->mm
->mmap_sem
)) {
1426 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1428 if (addr
+ size
< addr
)
1431 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1435 addr
= do_mmap_pgoff(file
, addr
, size
, prot
, flags
, 0, &populate
, NULL
);
1438 if (IS_ERR_VALUE(addr
))
1441 up_write(¤t
->mm
->mmap_sem
);
1443 mm_populate(addr
, populate
);
1449 down_write(&shm_ids(ns
).rwsem
);
1450 shp
= shm_lock(ns
, shmid
);
1452 if (shm_may_destroy(ns
, shp
))
1453 shm_destroy(ns
, shp
);
1456 up_write(&shm_ids(ns
).rwsem
);
1465 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1470 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
, SHMLBA
);
1473 force_successful_syscall_return();
1477 #ifdef CONFIG_COMPAT
1479 #ifndef COMPAT_SHMLBA
1480 #define COMPAT_SHMLBA SHMLBA
1483 COMPAT_SYSCALL_DEFINE3(shmat
, int, shmid
, compat_uptr_t
, shmaddr
, int, shmflg
)
1488 err
= do_shmat(shmid
, compat_ptr(shmaddr
), shmflg
, &ret
, COMPAT_SHMLBA
);
1491 force_successful_syscall_return();
1497 * detach and kill segment if marked destroyed.
1498 * The work is done in shm_close.
1500 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1502 struct mm_struct
*mm
= current
->mm
;
1503 struct vm_area_struct
*vma
;
1504 unsigned long addr
= (unsigned long)shmaddr
;
1505 int retval
= -EINVAL
;
1509 struct vm_area_struct
*next
;
1512 if (addr
& ~PAGE_MASK
)
1515 if (down_write_killable(&mm
->mmap_sem
))
1519 * This function tries to be smart and unmap shm segments that
1520 * were modified by partial mlock or munmap calls:
1521 * - It first determines the size of the shm segment that should be
1522 * unmapped: It searches for a vma that is backed by shm and that
1523 * started at address shmaddr. It records it's size and then unmaps
1525 * - Then it unmaps all shm vmas that started at shmaddr and that
1526 * are within the initially determined size and that are from the
1527 * same shm segment from which we determined the size.
1528 * Errors from do_munmap are ignored: the function only fails if
1529 * it's called with invalid parameters or if it's called to unmap
1530 * a part of a vma. Both calls in this function are for full vmas,
1531 * the parameters are directly copied from the vma itself and always
1532 * valid - therefore do_munmap cannot fail. (famous last words?)
1535 * If it had been mremap()'d, the starting address would not
1536 * match the usual checks anyway. So assume all vma's are
1537 * above the starting address given.
1539 vma
= find_vma(mm
, addr
);
1543 next
= vma
->vm_next
;
1546 * Check if the starting address would match, i.e. it's
1547 * a fragment created by mprotect() and/or munmap(), or it
1548 * otherwise it starts at this address with no hassles.
1550 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1551 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1554 * Record the file of the shm segment being
1555 * unmapped. With mremap(), someone could place
1556 * page from another segment but with equal offsets
1557 * in the range we are unmapping.
1559 file
= vma
->vm_file
;
1560 size
= i_size_read(file_inode(vma
->vm_file
));
1561 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1563 * We discovered the size of the shm segment, so
1564 * break out of here and fall through to the next
1565 * loop that uses the size information to stop
1566 * searching for matching vma's.
1576 * We need look no further than the maximum address a fragment
1577 * could possibly have landed at. Also cast things to loff_t to
1578 * prevent overflows and make comparisons vs. equal-width types.
1580 size
= PAGE_ALIGN(size
);
1581 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1582 next
= vma
->vm_next
;
1584 /* finding a matching vma now does not alter retval */
1585 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1586 ((vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) &&
1587 (vma
->vm_file
== file
))
1588 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1592 #else /* CONFIG_MMU */
1593 /* under NOMMU conditions, the exact address to be destroyed must be
1596 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1597 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1603 up_write(&mm
->mmap_sem
);
1607 #ifdef CONFIG_PROC_FS
1608 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1610 struct user_namespace
*user_ns
= seq_user_ns(s
);
1611 struct kern_ipc_perm
*ipcp
= it
;
1612 struct shmid_kernel
*shp
;
1613 unsigned long rss
= 0, swp
= 0;
1615 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
1616 shm_add_rss_swap(shp
, &rss
, &swp
);
1618 #if BITS_PER_LONG <= 32
1619 #define SIZE_SPEC "%10lu"
1621 #define SIZE_SPEC "%21lu"
1625 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1626 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1627 SIZE_SPEC
" " SIZE_SPEC
"\n",
1635 from_kuid_munged(user_ns
, shp
->shm_perm
.uid
),
1636 from_kgid_munged(user_ns
, shp
->shm_perm
.gid
),
1637 from_kuid_munged(user_ns
, shp
->shm_perm
.cuid
),
1638 from_kgid_munged(user_ns
, shp
->shm_perm
.cgid
),