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 <linux/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
;
93 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
95 if (shp
->shm_nattch
) {
96 shp
->shm_perm
.mode
|= SHM_DEST
;
97 /* Do not find it any more */
98 shp
->shm_perm
.key
= IPC_PRIVATE
;
101 shm_destroy(ns
, shp
);
105 void shm_exit_ns(struct ipc_namespace
*ns
)
107 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
108 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
112 static int __init
ipc_ns_init(void)
114 shm_init_ns(&init_ipc_ns
);
118 pure_initcall(ipc_ns_init
);
120 void __init
shm_init(void)
122 ipc_init_proc_interface("sysvipc/shm",
123 #if BITS_PER_LONG <= 32
124 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
126 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
128 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
131 static inline struct shmid_kernel
*shm_obtain_object(struct ipc_namespace
*ns
, int id
)
133 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_idr(&shm_ids(ns
), id
);
136 return ERR_CAST(ipcp
);
138 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
141 static inline struct shmid_kernel
*shm_obtain_object_check(struct ipc_namespace
*ns
, int id
)
143 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&shm_ids(ns
), id
);
146 return ERR_CAST(ipcp
);
148 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
152 * shm_lock_(check_) routines are called in the paths where the rwsem
153 * is not necessarily held.
155 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
157 struct kern_ipc_perm
*ipcp
= ipc_lock(&shm_ids(ns
), id
);
160 * Callers of shm_lock() must validate the status of the returned ipc
161 * object pointer (as returned by ipc_lock()), and error out as
166 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
169 static inline void shm_lock_by_ptr(struct shmid_kernel
*ipcp
)
172 ipc_lock_object(&ipcp
->shm_perm
);
175 static void shm_rcu_free(struct rcu_head
*head
)
177 struct kern_ipc_perm
*ptr
= container_of(head
, struct kern_ipc_perm
,
179 struct shmid_kernel
*shp
= container_of(ptr
, struct shmid_kernel
,
181 security_shm_free(shp
);
185 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
187 list_del(&s
->shm_clist
);
188 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
192 static int __shm_open(struct vm_area_struct
*vma
)
194 struct file
*file
= vma
->vm_file
;
195 struct shm_file_data
*sfd
= shm_file_data(file
);
196 struct shmid_kernel
*shp
;
198 shp
= shm_lock(sfd
->ns
, sfd
->id
);
203 shp
->shm_atim
= get_seconds();
204 shp
->shm_lprid
= task_tgid_vnr(current
);
210 /* This is called by fork, once for every shm attach. */
211 static void shm_open(struct vm_area_struct
*vma
)
213 int err
= __shm_open(vma
);
215 * We raced in the idr lookup or with shm_destroy().
216 * Either way, the ID is busted.
222 * shm_destroy - free the struct shmid_kernel
225 * @shp: struct to free
227 * It has to be called with shp and shm_ids.rwsem (writer) locked,
228 * but returns with shp unlocked and freed.
230 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
232 struct file
*shm_file
;
234 shm_file
= shp
->shm_file
;
235 shp
->shm_file
= NULL
;
236 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
239 if (!is_file_hugepages(shm_file
))
240 shmem_lock(shm_file
, 0, shp
->mlock_user
);
241 else if (shp
->mlock_user
)
242 user_shm_unlock(i_size_read(file_inode(shm_file
)),
245 ipc_rcu_putref(&shp
->shm_perm
, shm_rcu_free
);
249 * shm_may_destroy - identifies whether shm segment should be destroyed now
251 * Returns true if and only if there are no active users of the segment and
252 * one of the following is true:
254 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
256 * 2) sysctl kernel.shm_rmid_forced is set to 1.
258 static bool shm_may_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
260 return (shp
->shm_nattch
== 0) &&
261 (ns
->shm_rmid_forced
||
262 (shp
->shm_perm
.mode
& SHM_DEST
));
266 * remove the attach descriptor vma.
267 * free memory for segment if it is marked destroyed.
268 * The descriptor has already been removed from the current->mm->mmap list
269 * and will later be kfree()d.
271 static void shm_close(struct vm_area_struct
*vma
)
273 struct file
*file
= vma
->vm_file
;
274 struct shm_file_data
*sfd
= shm_file_data(file
);
275 struct shmid_kernel
*shp
;
276 struct ipc_namespace
*ns
= sfd
->ns
;
278 down_write(&shm_ids(ns
).rwsem
);
279 /* remove from the list of attaches of the shm segment */
280 shp
= shm_lock(ns
, sfd
->id
);
283 * We raced in the idr lookup or with shm_destroy().
284 * Either way, the ID is busted.
286 if (WARN_ON_ONCE(IS_ERR(shp
)))
287 goto done
; /* no-op */
289 shp
->shm_lprid
= task_tgid_vnr(current
);
290 shp
->shm_dtim
= get_seconds();
292 if (shm_may_destroy(ns
, shp
))
293 shm_destroy(ns
, shp
);
297 up_write(&shm_ids(ns
).rwsem
);
300 /* Called with ns->shm_ids(ns).rwsem locked */
301 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
303 struct ipc_namespace
*ns
= data
;
304 struct kern_ipc_perm
*ipcp
= p
;
305 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
308 * We want to destroy segments without users and with already
309 * exit'ed originating process.
311 * As shp->* are changed under rwsem, it's safe to skip shp locking.
313 if (shp
->shm_creator
!= NULL
)
316 if (shm_may_destroy(ns
, shp
)) {
317 shm_lock_by_ptr(shp
);
318 shm_destroy(ns
, shp
);
323 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
325 down_write(&shm_ids(ns
).rwsem
);
326 if (shm_ids(ns
).in_use
)
327 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
328 up_write(&shm_ids(ns
).rwsem
);
331 /* Locking assumes this will only be called with task == current */
332 void exit_shm(struct task_struct
*task
)
334 struct ipc_namespace
*ns
= task
->nsproxy
->ipc_ns
;
335 struct shmid_kernel
*shp
, *n
;
337 if (list_empty(&task
->sysvshm
.shm_clist
))
341 * If kernel.shm_rmid_forced is not set then only keep track of
342 * which shmids are orphaned, so that a later set of the sysctl
345 if (!ns
->shm_rmid_forced
) {
346 down_read(&shm_ids(ns
).rwsem
);
347 list_for_each_entry(shp
, &task
->sysvshm
.shm_clist
, shm_clist
)
348 shp
->shm_creator
= NULL
;
350 * Only under read lock but we are only called on current
351 * so no entry on the list will be shared.
353 list_del(&task
->sysvshm
.shm_clist
);
354 up_read(&shm_ids(ns
).rwsem
);
359 * Destroy all already created segments, that were not yet mapped,
360 * and mark any mapped as orphan to cover the sysctl toggling.
361 * Destroy is skipped if shm_may_destroy() returns false.
363 down_write(&shm_ids(ns
).rwsem
);
364 list_for_each_entry_safe(shp
, n
, &task
->sysvshm
.shm_clist
, shm_clist
) {
365 shp
->shm_creator
= NULL
;
367 if (shm_may_destroy(ns
, shp
)) {
368 shm_lock_by_ptr(shp
);
369 shm_destroy(ns
, shp
);
373 /* Remove the list head from any segments still attached. */
374 list_del(&task
->sysvshm
.shm_clist
);
375 up_write(&shm_ids(ns
).rwsem
);
378 static int shm_fault(struct vm_fault
*vmf
)
380 struct file
*file
= vmf
->vma
->vm_file
;
381 struct shm_file_data
*sfd
= shm_file_data(file
);
383 return sfd
->vm_ops
->fault(vmf
);
387 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
389 struct file
*file
= vma
->vm_file
;
390 struct shm_file_data
*sfd
= shm_file_data(file
);
393 if (sfd
->vm_ops
->set_policy
)
394 err
= sfd
->vm_ops
->set_policy(vma
, new);
398 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
401 struct file
*file
= vma
->vm_file
;
402 struct shm_file_data
*sfd
= shm_file_data(file
);
403 struct mempolicy
*pol
= NULL
;
405 if (sfd
->vm_ops
->get_policy
)
406 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
407 else if (vma
->vm_policy
)
408 pol
= vma
->vm_policy
;
414 static int shm_mmap(struct file
*file
, struct vm_area_struct
*vma
)
416 struct shm_file_data
*sfd
= shm_file_data(file
);
420 * In case of remap_file_pages() emulation, the file can represent
421 * removed IPC ID: propogate shm_lock() error to caller.
423 ret
= __shm_open(vma
);
427 ret
= call_mmap(sfd
->file
, vma
);
432 sfd
->vm_ops
= vma
->vm_ops
;
434 WARN_ON(!sfd
->vm_ops
->fault
);
436 vma
->vm_ops
= &shm_vm_ops
;
440 static int shm_release(struct inode
*ino
, struct file
*file
)
442 struct shm_file_data
*sfd
= shm_file_data(file
);
445 shm_file_data(file
) = NULL
;
450 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
452 struct shm_file_data
*sfd
= shm_file_data(file
);
454 if (!sfd
->file
->f_op
->fsync
)
456 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
459 static long shm_fallocate(struct file
*file
, int mode
, loff_t offset
,
462 struct shm_file_data
*sfd
= shm_file_data(file
);
464 if (!sfd
->file
->f_op
->fallocate
)
466 return sfd
->file
->f_op
->fallocate(file
, mode
, offset
, len
);
469 static unsigned long shm_get_unmapped_area(struct file
*file
,
470 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
473 struct shm_file_data
*sfd
= shm_file_data(file
);
475 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
479 static const struct file_operations shm_file_operations
= {
482 .release
= shm_release
,
483 .get_unmapped_area
= shm_get_unmapped_area
,
484 .llseek
= noop_llseek
,
485 .fallocate
= shm_fallocate
,
489 * shm_file_operations_huge is now identical to shm_file_operations,
490 * but we keep it distinct for the sake of is_file_shm_hugepages().
492 static const struct file_operations shm_file_operations_huge
= {
495 .release
= shm_release
,
496 .get_unmapped_area
= shm_get_unmapped_area
,
497 .llseek
= noop_llseek
,
498 .fallocate
= shm_fallocate
,
501 bool is_file_shm_hugepages(struct file
*file
)
503 return file
->f_op
== &shm_file_operations_huge
;
506 static const struct vm_operations_struct shm_vm_ops
= {
507 .open
= shm_open
, /* callback for a new vm-area open */
508 .close
= shm_close
, /* callback for when the vm-area is released */
510 #if defined(CONFIG_NUMA)
511 .set_policy
= shm_set_policy
,
512 .get_policy
= shm_get_policy
,
517 * newseg - Create a new shared memory segment
519 * @params: ptr to the structure that contains key, size and shmflg
521 * Called with shm_ids.rwsem held as a writer.
523 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
525 key_t key
= params
->key
;
526 int shmflg
= params
->flg
;
527 size_t size
= params
->u
.size
;
529 struct shmid_kernel
*shp
;
530 size_t numpages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
533 vm_flags_t acctflag
= 0;
535 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
538 if (numpages
<< PAGE_SHIFT
< size
)
541 if (ns
->shm_tot
+ numpages
< ns
->shm_tot
||
542 ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
545 shp
= kvmalloc(sizeof(*shp
), GFP_KERNEL
);
549 shp
->shm_perm
.key
= key
;
550 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
551 shp
->mlock_user
= NULL
;
553 shp
->shm_perm
.security
= NULL
;
554 error
= security_shm_alloc(shp
);
560 sprintf(name
, "SYSV%08x", key
);
561 if (shmflg
& SHM_HUGETLB
) {
565 hs
= hstate_sizelog((shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
570 hugesize
= ALIGN(size
, huge_page_size(hs
));
572 /* hugetlb_file_setup applies strict accounting */
573 if (shmflg
& SHM_NORESERVE
)
574 acctflag
= VM_NORESERVE
;
575 file
= hugetlb_file_setup(name
, hugesize
, acctflag
,
576 &shp
->mlock_user
, HUGETLB_SHMFS_INODE
,
577 (shmflg
>> SHM_HUGE_SHIFT
) & SHM_HUGE_MASK
);
580 * Do not allow no accounting for OVERCOMMIT_NEVER, even
583 if ((shmflg
& SHM_NORESERVE
) &&
584 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
585 acctflag
= VM_NORESERVE
;
586 file
= shmem_kernel_file_setup(name
, size
, acctflag
);
588 error
= PTR_ERR(file
);
592 shp
->shm_cprid
= task_tgid_vnr(current
);
594 shp
->shm_atim
= shp
->shm_dtim
= 0;
595 shp
->shm_ctim
= get_seconds();
596 shp
->shm_segsz
= size
;
598 shp
->shm_file
= file
;
599 shp
->shm_creator
= current
;
601 error
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
605 list_add(&shp
->shm_clist
, ¤t
->sysvshm
.shm_clist
);
608 * shmid gets reported as "inode#" in /proc/pid/maps.
609 * proc-ps tools use this. Changing this will break them.
611 file_inode(file
)->i_ino
= shp
->shm_perm
.id
;
613 ns
->shm_tot
+= numpages
;
614 error
= shp
->shm_perm
.id
;
616 ipc_unlock_object(&shp
->shm_perm
);
621 if (is_file_hugepages(file
) && shp
->mlock_user
)
622 user_shm_unlock(size
, shp
->mlock_user
);
625 call_rcu(&shp
->shm_perm
.rcu
, shm_rcu_free
);
630 * Called with shm_ids.rwsem and ipcp locked.
632 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
634 struct shmid_kernel
*shp
;
636 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
637 return security_shm_associate(shp
, shmflg
);
641 * Called with shm_ids.rwsem and ipcp locked.
643 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
644 struct ipc_params
*params
)
646 struct shmid_kernel
*shp
;
648 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
649 if (shp
->shm_segsz
< params
->u
.size
)
655 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
657 struct ipc_namespace
*ns
;
658 static const struct ipc_ops shm_ops
= {
660 .associate
= shm_security
,
661 .more_checks
= shm_more_checks
,
663 struct ipc_params shm_params
;
665 ns
= current
->nsproxy
->ipc_ns
;
667 shm_params
.key
= key
;
668 shm_params
.flg
= shmflg
;
669 shm_params
.u
.size
= size
;
671 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
674 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
678 return copy_to_user(buf
, in
, sizeof(*in
));
683 memset(&out
, 0, sizeof(out
));
684 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
685 out
.shm_segsz
= in
->shm_segsz
;
686 out
.shm_atime
= in
->shm_atime
;
687 out
.shm_dtime
= in
->shm_dtime
;
688 out
.shm_ctime
= in
->shm_ctime
;
689 out
.shm_cpid
= in
->shm_cpid
;
690 out
.shm_lpid
= in
->shm_lpid
;
691 out
.shm_nattch
= in
->shm_nattch
;
693 return copy_to_user(buf
, &out
, sizeof(out
));
700 static inline unsigned long
701 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
705 if (copy_from_user(out
, buf
, sizeof(*out
)))
710 struct shmid_ds tbuf_old
;
712 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
715 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
716 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
717 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
726 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
730 return copy_to_user(buf
, in
, sizeof(*in
));
735 if (in
->shmmax
> INT_MAX
)
736 out
.shmmax
= INT_MAX
;
738 out
.shmmax
= (int)in
->shmmax
;
740 out
.shmmin
= in
->shmmin
;
741 out
.shmmni
= in
->shmmni
;
742 out
.shmseg
= in
->shmseg
;
743 out
.shmall
= in
->shmall
;
745 return copy_to_user(buf
, &out
, sizeof(out
));
753 * Calculate and add used RSS and swap pages of a shm.
754 * Called with shm_ids.rwsem held as a reader
756 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
757 unsigned long *rss_add
, unsigned long *swp_add
)
761 inode
= file_inode(shp
->shm_file
);
763 if (is_file_hugepages(shp
->shm_file
)) {
764 struct address_space
*mapping
= inode
->i_mapping
;
765 struct hstate
*h
= hstate_file(shp
->shm_file
);
766 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
769 struct shmem_inode_info
*info
= SHMEM_I(inode
);
771 spin_lock_irq(&info
->lock
);
772 *rss_add
+= inode
->i_mapping
->nrpages
;
773 *swp_add
+= info
->swapped
;
774 spin_unlock_irq(&info
->lock
);
776 *rss_add
+= inode
->i_mapping
->nrpages
;
782 * Called with shm_ids.rwsem held as a reader
784 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
793 in_use
= shm_ids(ns
).in_use
;
795 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
796 struct kern_ipc_perm
*ipc
;
797 struct shmid_kernel
*shp
;
799 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
802 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
804 shm_add_rss_swap(shp
, rss
, swp
);
811 * This function handles some shmctl commands which require the rwsem
812 * to be held in write mode.
813 * NOTE: no locks must be held, the rwsem is taken inside this function.
815 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
816 struct shmid_ds __user
*buf
, int version
)
818 struct kern_ipc_perm
*ipcp
;
819 struct shmid64_ds shmid64
;
820 struct shmid_kernel
*shp
;
823 if (cmd
== IPC_SET
) {
824 if (copy_shmid_from_user(&shmid64
, buf
, version
))
828 down_write(&shm_ids(ns
).rwsem
);
831 ipcp
= ipcctl_pre_down_nolock(ns
, &shm_ids(ns
), shmid
, cmd
,
832 &shmid64
.shm_perm
, 0);
838 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
840 err
= security_shm_shmctl(shp
, cmd
);
846 ipc_lock_object(&shp
->shm_perm
);
847 /* do_shm_rmid unlocks the ipc object and rcu */
848 do_shm_rmid(ns
, ipcp
);
851 ipc_lock_object(&shp
->shm_perm
);
852 err
= ipc_update_perm(&shmid64
.shm_perm
, ipcp
);
855 shp
->shm_ctim
= get_seconds();
863 ipc_unlock_object(&shp
->shm_perm
);
867 up_write(&shm_ids(ns
).rwsem
);
871 static int shmctl_nolock(struct ipc_namespace
*ns
, int shmid
,
872 int cmd
, int version
, void __user
*buf
)
875 struct shmid_kernel
*shp
;
877 /* preliminary security checks for *_INFO */
878 if (cmd
== IPC_INFO
|| cmd
== SHM_INFO
) {
879 err
= security_shm_shmctl(NULL
, cmd
);
887 struct shminfo64 shminfo
;
889 memset(&shminfo
, 0, sizeof(shminfo
));
890 shminfo
.shmmni
= shminfo
.shmseg
= ns
->shm_ctlmni
;
891 shminfo
.shmmax
= ns
->shm_ctlmax
;
892 shminfo
.shmall
= ns
->shm_ctlall
;
894 shminfo
.shmmin
= SHMMIN
;
895 if (copy_shminfo_to_user(buf
, &shminfo
, version
))
898 down_read(&shm_ids(ns
).rwsem
);
899 err
= ipc_get_maxid(&shm_ids(ns
));
900 up_read(&shm_ids(ns
).rwsem
);
908 struct shm_info shm_info
;
910 memset(&shm_info
, 0, sizeof(shm_info
));
911 down_read(&shm_ids(ns
).rwsem
);
912 shm_info
.used_ids
= shm_ids(ns
).in_use
;
913 shm_get_stat(ns
, &shm_info
.shm_rss
, &shm_info
.shm_swp
);
914 shm_info
.shm_tot
= ns
->shm_tot
;
915 shm_info
.swap_attempts
= 0;
916 shm_info
.swap_successes
= 0;
917 err
= ipc_get_maxid(&shm_ids(ns
));
918 up_read(&shm_ids(ns
).rwsem
);
919 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
))) {
924 err
= err
< 0 ? 0 : err
;
930 struct shmid64_ds tbuf
;
934 if (cmd
== SHM_STAT
) {
935 shp
= shm_obtain_object(ns
, shmid
);
940 result
= shp
->shm_perm
.id
;
942 shp
= shm_obtain_object_check(ns
, shmid
);
951 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
954 err
= security_shm_shmctl(shp
, cmd
);
958 memset(&tbuf
, 0, sizeof(tbuf
));
959 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
.shm_perm
);
960 tbuf
.shm_segsz
= shp
->shm_segsz
;
961 tbuf
.shm_atime
= shp
->shm_atim
;
962 tbuf
.shm_dtime
= shp
->shm_dtim
;
963 tbuf
.shm_ctime
= shp
->shm_ctim
;
964 tbuf
.shm_cpid
= shp
->shm_cprid
;
965 tbuf
.shm_lpid
= shp
->shm_lprid
;
966 tbuf
.shm_nattch
= shp
->shm_nattch
;
969 if (copy_shmid_to_user(buf
, &tbuf
, version
))
985 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
987 struct shmid_kernel
*shp
;
989 struct ipc_namespace
*ns
;
991 if (cmd
< 0 || shmid
< 0)
994 version
= ipc_parse_version(&cmd
);
995 ns
= current
->nsproxy
->ipc_ns
;
1002 return shmctl_nolock(ns
, shmid
, cmd
, version
, buf
);
1005 return shmctl_down(ns
, shmid
, cmd
, buf
, version
);
1009 struct file
*shm_file
;
1012 shp
= shm_obtain_object_check(ns
, shmid
);
1018 audit_ipc_obj(&(shp
->shm_perm
));
1019 err
= security_shm_shmctl(shp
, cmd
);
1023 ipc_lock_object(&shp
->shm_perm
);
1025 /* check if shm_destroy() is tearing down shp */
1026 if (!ipc_valid_object(&shp
->shm_perm
)) {
1031 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
1032 kuid_t euid
= current_euid();
1034 if (!uid_eq(euid
, shp
->shm_perm
.uid
) &&
1035 !uid_eq(euid
, shp
->shm_perm
.cuid
)) {
1039 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
)) {
1045 shm_file
= shp
->shm_file
;
1046 if (is_file_hugepages(shm_file
))
1049 if (cmd
== SHM_LOCK
) {
1050 struct user_struct
*user
= current_user();
1052 err
= shmem_lock(shm_file
, 1, user
);
1053 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)) {
1054 shp
->shm_perm
.mode
|= SHM_LOCKED
;
1055 shp
->mlock_user
= user
;
1061 if (!(shp
->shm_perm
.mode
& SHM_LOCKED
))
1063 shmem_lock(shm_file
, 0, shp
->mlock_user
);
1064 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
1065 shp
->mlock_user
= NULL
;
1067 ipc_unlock_object(&shp
->shm_perm
);
1069 shmem_unlock_mapping(shm_file
->f_mapping
);
1079 ipc_unlock_object(&shp
->shm_perm
);
1086 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1088 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1089 * "raddr" thing points to kernel space, and there has to be a wrapper around
1092 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
,
1093 ulong
*raddr
, unsigned long shmlba
)
1095 struct shmid_kernel
*shp
;
1096 unsigned long addr
= (unsigned long)shmaddr
;
1100 unsigned long flags
= MAP_SHARED
;
1103 struct ipc_namespace
*ns
;
1104 struct shm_file_data
*sfd
;
1107 unsigned long populate
= 0;
1114 if (addr
& (shmlba
- 1)) {
1116 * Round down to the nearest multiple of shmlba.
1117 * For sane do_mmap_pgoff() parameters, avoid
1118 * round downs that trigger nil-page and MAP_FIXED.
1120 if ((shmflg
& SHM_RND
) && addr
>= shmlba
)
1121 addr
&= ~(shmlba
- 1);
1123 #ifndef __ARCH_FORCE_SHMLBA
1124 if (addr
& ~PAGE_MASK
)
1130 } else if ((shmflg
& SHM_REMAP
))
1133 if (shmflg
& SHM_RDONLY
) {
1136 f_mode
= FMODE_READ
;
1138 prot
= PROT_READ
| PROT_WRITE
;
1139 acc_mode
= S_IRUGO
| S_IWUGO
;
1140 f_mode
= FMODE_READ
| FMODE_WRITE
;
1142 if (shmflg
& SHM_EXEC
) {
1144 acc_mode
|= S_IXUGO
;
1148 * We cannot rely on the fs check since SYSV IPC does have an
1149 * additional creator id...
1151 ns
= current
->nsproxy
->ipc_ns
;
1153 shp
= shm_obtain_object_check(ns
, shmid
);
1160 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
1163 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
1167 ipc_lock_object(&shp
->shm_perm
);
1169 /* check if shm_destroy() is tearing down shp */
1170 if (!ipc_valid_object(&shp
->shm_perm
)) {
1171 ipc_unlock_object(&shp
->shm_perm
);
1176 path
= shp
->shm_file
->f_path
;
1179 size
= i_size_read(d_inode(path
.dentry
));
1180 ipc_unlock_object(&shp
->shm_perm
);
1184 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1190 file
= alloc_file(&path
, f_mode
,
1191 is_file_hugepages(shp
->shm_file
) ?
1192 &shm_file_operations_huge
:
1193 &shm_file_operations
);
1194 err
= PTR_ERR(file
);
1201 file
->private_data
= sfd
;
1202 file
->f_mapping
= shp
->shm_file
->f_mapping
;
1203 sfd
->id
= shp
->shm_perm
.id
;
1204 sfd
->ns
= get_ipc_ns(ns
);
1205 sfd
->file
= shp
->shm_file
;
1208 err
= security_mmap_file(file
, prot
, flags
);
1212 if (down_write_killable(¤t
->mm
->mmap_sem
)) {
1217 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1219 if (addr
+ size
< addr
)
1222 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1226 addr
= do_mmap_pgoff(file
, addr
, size
, prot
, flags
, 0, &populate
, NULL
);
1229 if (IS_ERR_VALUE(addr
))
1232 up_write(¤t
->mm
->mmap_sem
);
1234 mm_populate(addr
, populate
);
1240 down_write(&shm_ids(ns
).rwsem
);
1241 shp
= shm_lock(ns
, shmid
);
1243 if (shm_may_destroy(ns
, shp
))
1244 shm_destroy(ns
, shp
);
1247 up_write(&shm_ids(ns
).rwsem
);
1256 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1261 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
, SHMLBA
);
1264 force_successful_syscall_return();
1269 * detach and kill segment if marked destroyed.
1270 * The work is done in shm_close.
1272 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1274 struct mm_struct
*mm
= current
->mm
;
1275 struct vm_area_struct
*vma
;
1276 unsigned long addr
= (unsigned long)shmaddr
;
1277 int retval
= -EINVAL
;
1281 struct vm_area_struct
*next
;
1284 if (addr
& ~PAGE_MASK
)
1287 if (down_write_killable(&mm
->mmap_sem
))
1291 * This function tries to be smart and unmap shm segments that
1292 * were modified by partial mlock or munmap calls:
1293 * - It first determines the size of the shm segment that should be
1294 * unmapped: It searches for a vma that is backed by shm and that
1295 * started at address shmaddr. It records it's size and then unmaps
1297 * - Then it unmaps all shm vmas that started at shmaddr and that
1298 * are within the initially determined size and that are from the
1299 * same shm segment from which we determined the size.
1300 * Errors from do_munmap are ignored: the function only fails if
1301 * it's called with invalid parameters or if it's called to unmap
1302 * a part of a vma. Both calls in this function are for full vmas,
1303 * the parameters are directly copied from the vma itself and always
1304 * valid - therefore do_munmap cannot fail. (famous last words?)
1307 * If it had been mremap()'d, the starting address would not
1308 * match the usual checks anyway. So assume all vma's are
1309 * above the starting address given.
1311 vma
= find_vma(mm
, addr
);
1315 next
= vma
->vm_next
;
1318 * Check if the starting address would match, i.e. it's
1319 * a fragment created by mprotect() and/or munmap(), or it
1320 * otherwise it starts at this address with no hassles.
1322 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1323 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1326 * Record the file of the shm segment being
1327 * unmapped. With mremap(), someone could place
1328 * page from another segment but with equal offsets
1329 * in the range we are unmapping.
1331 file
= vma
->vm_file
;
1332 size
= i_size_read(file_inode(vma
->vm_file
));
1333 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1335 * We discovered the size of the shm segment, so
1336 * break out of here and fall through to the next
1337 * loop that uses the size information to stop
1338 * searching for matching vma's.
1348 * We need look no further than the maximum address a fragment
1349 * could possibly have landed at. Also cast things to loff_t to
1350 * prevent overflows and make comparisons vs. equal-width types.
1352 size
= PAGE_ALIGN(size
);
1353 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1354 next
= vma
->vm_next
;
1356 /* finding a matching vma now does not alter retval */
1357 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1358 ((vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) &&
1359 (vma
->vm_file
== file
))
1360 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1364 #else /* CONFIG_MMU */
1365 /* under NOMMU conditions, the exact address to be destroyed must be
1368 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1369 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, NULL
);
1375 up_write(&mm
->mmap_sem
);
1379 #ifdef CONFIG_PROC_FS
1380 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1382 struct user_namespace
*user_ns
= seq_user_ns(s
);
1383 struct kern_ipc_perm
*ipcp
= it
;
1384 struct shmid_kernel
*shp
;
1385 unsigned long rss
= 0, swp
= 0;
1387 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
1388 shm_add_rss_swap(shp
, &rss
, &swp
);
1390 #if BITS_PER_LONG <= 32
1391 #define SIZE_SPEC "%10lu"
1393 #define SIZE_SPEC "%21lu"
1397 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1398 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1399 SIZE_SPEC
" " SIZE_SPEC
"\n",
1407 from_kuid_munged(user_ns
, shp
->shm_perm
.uid
),
1408 from_kgid_munged(user_ns
, shp
->shm_perm
.gid
),
1409 from_kuid_munged(user_ns
, shp
->shm_perm
.cuid
),
1410 from_kgid_munged(user_ns
, shp
->shm_perm
.cgid
),