3 * Copyright (C) 1992, 1993 Krishna Balasubramanian
4 * Many improvements/fixes by Bruno Haible.
5 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
8 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
16 * support for audit of ipc object properties and permission changes
17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 * Pavel Emelianov <xemul@openvz.org>
24 #include <linux/slab.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
43 #include <asm/uaccess.h>
47 struct shm_file_data
{
49 struct ipc_namespace
*ns
;
51 const struct vm_operations_struct
*vm_ops
;
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56 static const struct file_operations shm_file_operations
;
57 static const struct vm_operations_struct shm_vm_ops
;
59 #define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
61 #define shm_unlock(shp) \
62 ipc_unlock(&(shp)->shm_perm)
64 static int newseg(struct ipc_namespace
*, struct ipc_params
*);
65 static void shm_open(struct vm_area_struct
*vma
);
66 static void shm_close(struct vm_area_struct
*vma
);
67 static void shm_destroy (struct ipc_namespace
*ns
, struct shmid_kernel
*shp
);
69 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
);
72 void shm_init_ns(struct ipc_namespace
*ns
)
74 ns
->shm_ctlmax
= SHMMAX
;
75 ns
->shm_ctlall
= SHMALL
;
76 ns
->shm_ctlmni
= SHMMNI
;
77 ns
->shm_rmid_forced
= 0;
79 ipc_init_ids(&shm_ids(ns
));
83 * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84 * Only shm_ids.rw_mutex remains locked on exit.
86 static void do_shm_rmid(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
88 struct shmid_kernel
*shp
;
89 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
92 shp
->shm_perm
.mode
|= SHM_DEST
;
93 /* Do not find it any more */
94 shp
->shm_perm
.key
= IPC_PRIVATE
;
101 void shm_exit_ns(struct ipc_namespace
*ns
)
103 free_ipcs(ns
, &shm_ids(ns
), do_shm_rmid
);
104 idr_destroy(&ns
->ids
[IPC_SHM_IDS
].ipcs_idr
);
108 void __init
shm_init (void)
110 shm_init_ns(&init_ipc_ns
);
111 ipc_init_proc_interface("sysvipc/shm",
112 #if BITS_PER_LONG <= 32
113 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
115 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
117 IPC_SHM_IDS
, sysvipc_shm_proc_show
);
121 * shm_lock_(check_) routines are called in the paths where the rw_mutex
122 * is not necessarily held.
124 static inline struct shmid_kernel
*shm_lock(struct ipc_namespace
*ns
, int id
)
126 struct kern_ipc_perm
*ipcp
= ipc_lock(&shm_ids(ns
), id
);
129 return (struct shmid_kernel
*)ipcp
;
131 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
134 static inline void shm_lock_by_ptr(struct shmid_kernel
*ipcp
)
137 spin_lock(&ipcp
->shm_perm
.lock
);
140 static inline struct shmid_kernel
*shm_lock_check(struct ipc_namespace
*ns
,
143 struct kern_ipc_perm
*ipcp
= ipc_lock_check(&shm_ids(ns
), id
);
146 return (struct shmid_kernel
*)ipcp
;
148 return container_of(ipcp
, struct shmid_kernel
, shm_perm
);
151 static inline void shm_rmid(struct ipc_namespace
*ns
, struct shmid_kernel
*s
)
153 ipc_rmid(&shm_ids(ns
), &s
->shm_perm
);
157 /* This is called by fork, once for every shm attach. */
158 static void shm_open(struct vm_area_struct
*vma
)
160 struct file
*file
= vma
->vm_file
;
161 struct shm_file_data
*sfd
= shm_file_data(file
);
162 struct shmid_kernel
*shp
;
164 shp
= shm_lock(sfd
->ns
, sfd
->id
);
166 shp
->shm_atim
= get_seconds();
167 shp
->shm_lprid
= task_tgid_vnr(current
);
173 * shm_destroy - free the struct shmid_kernel
176 * @shp: struct to free
178 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
179 * but returns with shp unlocked and freed.
181 static void shm_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
183 ns
->shm_tot
-= (shp
->shm_segsz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
186 if (!is_file_hugepages(shp
->shm_file
))
187 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
188 else if (shp
->mlock_user
)
189 user_shm_unlock(shp
->shm_file
->f_path
.dentry
->d_inode
->i_size
,
191 fput (shp
->shm_file
);
192 security_shm_free(shp
);
197 * shm_may_destroy - identifies whether shm segment should be destroyed now
199 * Returns true if and only if there are no active users of the segment and
200 * one of the following is true:
202 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
204 * 2) sysctl kernel.shm_rmid_forced is set to 1.
206 static bool shm_may_destroy(struct ipc_namespace
*ns
, struct shmid_kernel
*shp
)
208 return (shp
->shm_nattch
== 0) &&
209 (ns
->shm_rmid_forced
||
210 (shp
->shm_perm
.mode
& SHM_DEST
));
214 * remove the attach descriptor vma.
215 * free memory for segment if it is marked destroyed.
216 * The descriptor has already been removed from the current->mm->mmap list
217 * and will later be kfree()d.
219 static void shm_close(struct vm_area_struct
*vma
)
221 struct file
* file
= vma
->vm_file
;
222 struct shm_file_data
*sfd
= shm_file_data(file
);
223 struct shmid_kernel
*shp
;
224 struct ipc_namespace
*ns
= sfd
->ns
;
226 down_write(&shm_ids(ns
).rw_mutex
);
227 /* remove from the list of attaches of the shm segment */
228 shp
= shm_lock(ns
, sfd
->id
);
230 shp
->shm_lprid
= task_tgid_vnr(current
);
231 shp
->shm_dtim
= get_seconds();
233 if (shm_may_destroy(ns
, shp
))
234 shm_destroy(ns
, shp
);
237 up_write(&shm_ids(ns
).rw_mutex
);
240 /* Called with ns->shm_ids(ns).rw_mutex locked */
241 static int shm_try_destroy_current(int id
, void *p
, void *data
)
243 struct ipc_namespace
*ns
= data
;
244 struct kern_ipc_perm
*ipcp
= p
;
245 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
247 if (shp
->shm_creator
!= current
)
251 * Mark it as orphaned to destroy the segment when
252 * kernel.shm_rmid_forced is changed.
253 * It is noop if the following shm_may_destroy() returns true.
255 shp
->shm_creator
= NULL
;
258 * Don't even try to destroy it. If shm_rmid_forced=0 and IPC_RMID
259 * is not set, it shouldn't be deleted here.
261 if (!ns
->shm_rmid_forced
)
264 if (shm_may_destroy(ns
, shp
)) {
265 shm_lock_by_ptr(shp
);
266 shm_destroy(ns
, shp
);
271 /* Called with ns->shm_ids(ns).rw_mutex locked */
272 static int shm_try_destroy_orphaned(int id
, void *p
, void *data
)
274 struct ipc_namespace
*ns
= data
;
275 struct kern_ipc_perm
*ipcp
= p
;
276 struct shmid_kernel
*shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
279 * We want to destroy segments without users and with already
280 * exit'ed originating process.
282 * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
284 if (shp
->shm_creator
!= NULL
)
287 if (shm_may_destroy(ns
, shp
)) {
288 shm_lock_by_ptr(shp
);
289 shm_destroy(ns
, shp
);
294 void shm_destroy_orphaned(struct ipc_namespace
*ns
)
296 down_write(&shm_ids(ns
).rw_mutex
);
297 if (shm_ids(ns
).in_use
)
298 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_orphaned
, ns
);
299 up_write(&shm_ids(ns
).rw_mutex
);
303 void exit_shm(struct task_struct
*task
)
305 struct ipc_namespace
*ns
= task
->nsproxy
->ipc_ns
;
307 if (shm_ids(ns
).in_use
== 0)
310 /* Destroy all already created segments, but not mapped yet */
311 down_write(&shm_ids(ns
).rw_mutex
);
312 if (shm_ids(ns
).in_use
)
313 idr_for_each(&shm_ids(ns
).ipcs_idr
, &shm_try_destroy_current
, ns
);
314 up_write(&shm_ids(ns
).rw_mutex
);
317 static int shm_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
319 struct file
*file
= vma
->vm_file
;
320 struct shm_file_data
*sfd
= shm_file_data(file
);
322 return sfd
->vm_ops
->fault(vma
, vmf
);
326 static int shm_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
328 struct file
*file
= vma
->vm_file
;
329 struct shm_file_data
*sfd
= shm_file_data(file
);
331 if (sfd
->vm_ops
->set_policy
)
332 err
= sfd
->vm_ops
->set_policy(vma
, new);
336 static struct mempolicy
*shm_get_policy(struct vm_area_struct
*vma
,
339 struct file
*file
= vma
->vm_file
;
340 struct shm_file_data
*sfd
= shm_file_data(file
);
341 struct mempolicy
*pol
= NULL
;
343 if (sfd
->vm_ops
->get_policy
)
344 pol
= sfd
->vm_ops
->get_policy(vma
, addr
);
345 else if (vma
->vm_policy
)
346 pol
= vma
->vm_policy
;
352 static int shm_mmap(struct file
* file
, struct vm_area_struct
* vma
)
354 struct shm_file_data
*sfd
= shm_file_data(file
);
357 ret
= sfd
->file
->f_op
->mmap(sfd
->file
, vma
);
360 sfd
->vm_ops
= vma
->vm_ops
;
362 BUG_ON(!sfd
->vm_ops
->fault
);
364 vma
->vm_ops
= &shm_vm_ops
;
370 static int shm_release(struct inode
*ino
, struct file
*file
)
372 struct shm_file_data
*sfd
= shm_file_data(file
);
375 shm_file_data(file
) = NULL
;
380 static int shm_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
382 struct shm_file_data
*sfd
= shm_file_data(file
);
384 if (!sfd
->file
->f_op
->fsync
)
386 return sfd
->file
->f_op
->fsync(sfd
->file
, start
, end
, datasync
);
389 static unsigned long shm_get_unmapped_area(struct file
*file
,
390 unsigned long addr
, unsigned long len
, unsigned long pgoff
,
393 struct shm_file_data
*sfd
= shm_file_data(file
);
394 return sfd
->file
->f_op
->get_unmapped_area(sfd
->file
, addr
, len
,
398 static const struct file_operations shm_file_operations
= {
401 .release
= shm_release
,
403 .get_unmapped_area
= shm_get_unmapped_area
,
405 .llseek
= noop_llseek
,
408 static const struct file_operations shm_file_operations_huge
= {
411 .release
= shm_release
,
412 .get_unmapped_area
= shm_get_unmapped_area
,
413 .llseek
= noop_llseek
,
416 int is_file_shm_hugepages(struct file
*file
)
418 return file
->f_op
== &shm_file_operations_huge
;
421 static const struct vm_operations_struct shm_vm_ops
= {
422 .open
= shm_open
, /* callback for a new vm-area open */
423 .close
= shm_close
, /* callback for when the vm-area is released */
425 #if defined(CONFIG_NUMA)
426 .set_policy
= shm_set_policy
,
427 .get_policy
= shm_get_policy
,
432 * newseg - Create a new shared memory segment
434 * @params: ptr to the structure that contains key, size and shmflg
436 * Called with shm_ids.rw_mutex held as a writer.
439 static int newseg(struct ipc_namespace
*ns
, struct ipc_params
*params
)
441 key_t key
= params
->key
;
442 int shmflg
= params
->flg
;
443 size_t size
= params
->u
.size
;
445 struct shmid_kernel
*shp
;
446 int numpages
= (size
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
450 vm_flags_t acctflag
= 0;
452 if (size
< SHMMIN
|| size
> ns
->shm_ctlmax
)
455 if (ns
->shm_tot
+ numpages
> ns
->shm_ctlall
)
458 shp
= ipc_rcu_alloc(sizeof(*shp
));
462 shp
->shm_perm
.key
= key
;
463 shp
->shm_perm
.mode
= (shmflg
& S_IRWXUGO
);
464 shp
->mlock_user
= NULL
;
466 shp
->shm_perm
.security
= NULL
;
467 error
= security_shm_alloc(shp
);
473 sprintf (name
, "SYSV%08x", key
);
474 if (shmflg
& SHM_HUGETLB
) {
475 /* hugetlb_file_setup applies strict accounting */
476 if (shmflg
& SHM_NORESERVE
)
477 acctflag
= VM_NORESERVE
;
478 file
= hugetlb_file_setup(name
, size
, acctflag
,
479 &shp
->mlock_user
, HUGETLB_SHMFS_INODE
);
482 * Do not allow no accounting for OVERCOMMIT_NEVER, even
485 if ((shmflg
& SHM_NORESERVE
) &&
486 sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
487 acctflag
= VM_NORESERVE
;
488 file
= shmem_file_setup(name
, size
, acctflag
);
490 error
= PTR_ERR(file
);
494 id
= ipc_addid(&shm_ids(ns
), &shp
->shm_perm
, ns
->shm_ctlmni
);
500 shp
->shm_cprid
= task_tgid_vnr(current
);
502 shp
->shm_atim
= shp
->shm_dtim
= 0;
503 shp
->shm_ctim
= get_seconds();
504 shp
->shm_segsz
= size
;
506 shp
->shm_file
= file
;
507 shp
->shm_creator
= current
;
509 * shmid gets reported as "inode#" in /proc/pid/maps.
510 * proc-ps tools use this. Changing this will break them.
512 file
->f_dentry
->d_inode
->i_ino
= shp
->shm_perm
.id
;
514 ns
->shm_tot
+= numpages
;
515 error
= shp
->shm_perm
.id
;
520 if (is_file_hugepages(file
) && shp
->mlock_user
)
521 user_shm_unlock(size
, shp
->mlock_user
);
524 security_shm_free(shp
);
530 * Called with shm_ids.rw_mutex and ipcp locked.
532 static inline int shm_security(struct kern_ipc_perm
*ipcp
, int shmflg
)
534 struct shmid_kernel
*shp
;
536 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
537 return security_shm_associate(shp
, shmflg
);
541 * Called with shm_ids.rw_mutex and ipcp locked.
543 static inline int shm_more_checks(struct kern_ipc_perm
*ipcp
,
544 struct ipc_params
*params
)
546 struct shmid_kernel
*shp
;
548 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
549 if (shp
->shm_segsz
< params
->u
.size
)
555 SYSCALL_DEFINE3(shmget
, key_t
, key
, size_t, size
, int, shmflg
)
557 struct ipc_namespace
*ns
;
558 struct ipc_ops shm_ops
;
559 struct ipc_params shm_params
;
561 ns
= current
->nsproxy
->ipc_ns
;
563 shm_ops
.getnew
= newseg
;
564 shm_ops
.associate
= shm_security
;
565 shm_ops
.more_checks
= shm_more_checks
;
567 shm_params
.key
= key
;
568 shm_params
.flg
= shmflg
;
569 shm_params
.u
.size
= size
;
571 return ipcget(ns
, &shm_ids(ns
), &shm_ops
, &shm_params
);
574 static inline unsigned long copy_shmid_to_user(void __user
*buf
, struct shmid64_ds
*in
, int version
)
578 return copy_to_user(buf
, in
, sizeof(*in
));
583 memset(&out
, 0, sizeof(out
));
584 ipc64_perm_to_ipc_perm(&in
->shm_perm
, &out
.shm_perm
);
585 out
.shm_segsz
= in
->shm_segsz
;
586 out
.shm_atime
= in
->shm_atime
;
587 out
.shm_dtime
= in
->shm_dtime
;
588 out
.shm_ctime
= in
->shm_ctime
;
589 out
.shm_cpid
= in
->shm_cpid
;
590 out
.shm_lpid
= in
->shm_lpid
;
591 out
.shm_nattch
= in
->shm_nattch
;
593 return copy_to_user(buf
, &out
, sizeof(out
));
600 static inline unsigned long
601 copy_shmid_from_user(struct shmid64_ds
*out
, void __user
*buf
, int version
)
605 if (copy_from_user(out
, buf
, sizeof(*out
)))
610 struct shmid_ds tbuf_old
;
612 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
615 out
->shm_perm
.uid
= tbuf_old
.shm_perm
.uid
;
616 out
->shm_perm
.gid
= tbuf_old
.shm_perm
.gid
;
617 out
->shm_perm
.mode
= tbuf_old
.shm_perm
.mode
;
626 static inline unsigned long copy_shminfo_to_user(void __user
*buf
, struct shminfo64
*in
, int version
)
630 return copy_to_user(buf
, in
, sizeof(*in
));
635 if(in
->shmmax
> INT_MAX
)
636 out
.shmmax
= INT_MAX
;
638 out
.shmmax
= (int)in
->shmmax
;
640 out
.shmmin
= in
->shmmin
;
641 out
.shmmni
= in
->shmmni
;
642 out
.shmseg
= in
->shmseg
;
643 out
.shmall
= in
->shmall
;
645 return copy_to_user(buf
, &out
, sizeof(out
));
653 * Calculate and add used RSS and swap pages of a shm.
654 * Called with shm_ids.rw_mutex held as a reader
656 static void shm_add_rss_swap(struct shmid_kernel
*shp
,
657 unsigned long *rss_add
, unsigned long *swp_add
)
661 inode
= shp
->shm_file
->f_path
.dentry
->d_inode
;
663 if (is_file_hugepages(shp
->shm_file
)) {
664 struct address_space
*mapping
= inode
->i_mapping
;
665 struct hstate
*h
= hstate_file(shp
->shm_file
);
666 *rss_add
+= pages_per_huge_page(h
) * mapping
->nrpages
;
669 struct shmem_inode_info
*info
= SHMEM_I(inode
);
670 spin_lock(&info
->lock
);
671 *rss_add
+= inode
->i_mapping
->nrpages
;
672 *swp_add
+= info
->swapped
;
673 spin_unlock(&info
->lock
);
675 *rss_add
+= inode
->i_mapping
->nrpages
;
681 * Called with shm_ids.rw_mutex held as a reader
683 static void shm_get_stat(struct ipc_namespace
*ns
, unsigned long *rss
,
692 in_use
= shm_ids(ns
).in_use
;
694 for (total
= 0, next_id
= 0; total
< in_use
; next_id
++) {
695 struct kern_ipc_perm
*ipc
;
696 struct shmid_kernel
*shp
;
698 ipc
= idr_find(&shm_ids(ns
).ipcs_idr
, next_id
);
701 shp
= container_of(ipc
, struct shmid_kernel
, shm_perm
);
703 shm_add_rss_swap(shp
, rss
, swp
);
710 * This function handles some shmctl commands which require the rw_mutex
711 * to be held in write mode.
712 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
714 static int shmctl_down(struct ipc_namespace
*ns
, int shmid
, int cmd
,
715 struct shmid_ds __user
*buf
, int version
)
717 struct kern_ipc_perm
*ipcp
;
718 struct shmid64_ds shmid64
;
719 struct shmid_kernel
*shp
;
722 if (cmd
== IPC_SET
) {
723 if (copy_shmid_from_user(&shmid64
, buf
, version
))
727 ipcp
= ipcctl_pre_down(ns
, &shm_ids(ns
), shmid
, cmd
,
728 &shmid64
.shm_perm
, 0);
730 return PTR_ERR(ipcp
);
732 shp
= container_of(ipcp
, struct shmid_kernel
, shm_perm
);
734 err
= security_shm_shmctl(shp
, cmd
);
739 do_shm_rmid(ns
, ipcp
);
742 ipc_update_perm(&shmid64
.shm_perm
, ipcp
);
743 shp
->shm_ctim
= get_seconds();
751 up_write(&shm_ids(ns
).rw_mutex
);
755 SYSCALL_DEFINE3(shmctl
, int, shmid
, int, cmd
, struct shmid_ds __user
*, buf
)
757 struct shmid_kernel
*shp
;
759 struct ipc_namespace
*ns
;
761 if (cmd
< 0 || shmid
< 0) {
766 version
= ipc_parse_version(&cmd
);
767 ns
= current
->nsproxy
->ipc_ns
;
769 switch (cmd
) { /* replace with proc interface ? */
772 struct shminfo64 shminfo
;
774 err
= security_shm_shmctl(NULL
, cmd
);
778 memset(&shminfo
, 0, sizeof(shminfo
));
779 shminfo
.shmmni
= shminfo
.shmseg
= ns
->shm_ctlmni
;
780 shminfo
.shmmax
= ns
->shm_ctlmax
;
781 shminfo
.shmall
= ns
->shm_ctlall
;
783 shminfo
.shmmin
= SHMMIN
;
784 if(copy_shminfo_to_user (buf
, &shminfo
, version
))
787 down_read(&shm_ids(ns
).rw_mutex
);
788 err
= ipc_get_maxid(&shm_ids(ns
));
789 up_read(&shm_ids(ns
).rw_mutex
);
797 struct shm_info shm_info
;
799 err
= security_shm_shmctl(NULL
, cmd
);
803 memset(&shm_info
, 0, sizeof(shm_info
));
804 down_read(&shm_ids(ns
).rw_mutex
);
805 shm_info
.used_ids
= shm_ids(ns
).in_use
;
806 shm_get_stat (ns
, &shm_info
.shm_rss
, &shm_info
.shm_swp
);
807 shm_info
.shm_tot
= ns
->shm_tot
;
808 shm_info
.swap_attempts
= 0;
809 shm_info
.swap_successes
= 0;
810 err
= ipc_get_maxid(&shm_ids(ns
));
811 up_read(&shm_ids(ns
).rw_mutex
);
812 if (copy_to_user(buf
, &shm_info
, sizeof(shm_info
))) {
817 err
= err
< 0 ? 0 : err
;
823 struct shmid64_ds tbuf
;
826 if (cmd
== SHM_STAT
) {
827 shp
= shm_lock(ns
, shmid
);
832 result
= shp
->shm_perm
.id
;
834 shp
= shm_lock_check(ns
, shmid
);
842 if (ipcperms(ns
, &shp
->shm_perm
, S_IRUGO
))
844 err
= security_shm_shmctl(shp
, cmd
);
847 memset(&tbuf
, 0, sizeof(tbuf
));
848 kernel_to_ipc64_perm(&shp
->shm_perm
, &tbuf
.shm_perm
);
849 tbuf
.shm_segsz
= shp
->shm_segsz
;
850 tbuf
.shm_atime
= shp
->shm_atim
;
851 tbuf
.shm_dtime
= shp
->shm_dtim
;
852 tbuf
.shm_ctime
= shp
->shm_ctim
;
853 tbuf
.shm_cpid
= shp
->shm_cprid
;
854 tbuf
.shm_lpid
= shp
->shm_lprid
;
855 tbuf
.shm_nattch
= shp
->shm_nattch
;
857 if(copy_shmid_to_user (buf
, &tbuf
, version
))
866 struct file
*uninitialized_var(shm_file
);
868 lru_add_drain_all(); /* drain pagevecs to lru lists */
870 shp
= shm_lock_check(ns
, shmid
);
876 audit_ipc_obj(&(shp
->shm_perm
));
878 if (!ns_capable(ns
->user_ns
, CAP_IPC_LOCK
)) {
879 uid_t euid
= current_euid();
881 if (euid
!= shp
->shm_perm
.uid
&&
882 euid
!= shp
->shm_perm
.cuid
)
884 if (cmd
== SHM_LOCK
&& !rlimit(RLIMIT_MEMLOCK
))
888 err
= security_shm_shmctl(shp
, cmd
);
893 struct user_struct
*user
= current_user();
894 if (!is_file_hugepages(shp
->shm_file
)) {
895 err
= shmem_lock(shp
->shm_file
, 1, user
);
896 if (!err
&& !(shp
->shm_perm
.mode
& SHM_LOCKED
)){
897 shp
->shm_perm
.mode
|= SHM_LOCKED
;
898 shp
->mlock_user
= user
;
901 } else if (!is_file_hugepages(shp
->shm_file
)) {
902 shmem_lock(shp
->shm_file
, 0, shp
->mlock_user
);
903 shp
->shm_perm
.mode
&= ~SHM_LOCKED
;
904 shp
->mlock_user
= NULL
;
911 err
= shmctl_down(ns
, shmid
, cmd
, buf
, version
);
924 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
926 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
927 * "raddr" thing points to kernel space, and there has to be a wrapper around
930 long do_shmat(int shmid
, char __user
*shmaddr
, int shmflg
, ulong
*raddr
)
932 struct shmid_kernel
*shp
;
940 unsigned long user_addr
;
941 struct ipc_namespace
*ns
;
942 struct shm_file_data
*sfd
;
949 else if ((addr
= (ulong
)shmaddr
)) {
950 if (addr
& (SHMLBA
-1)) {
951 if (shmflg
& SHM_RND
)
952 addr
&= ~(SHMLBA
-1); /* round down */
954 #ifndef __ARCH_FORCE_SHMLBA
955 if (addr
& ~PAGE_MASK
)
959 flags
= MAP_SHARED
| MAP_FIXED
;
961 if ((shmflg
& SHM_REMAP
))
967 if (shmflg
& SHM_RDONLY
) {
972 prot
= PROT_READ
| PROT_WRITE
;
973 acc_mode
= S_IRUGO
| S_IWUGO
;
974 f_mode
= FMODE_READ
| FMODE_WRITE
;
976 if (shmflg
& SHM_EXEC
) {
982 * We cannot rely on the fs check since SYSV IPC does have an
983 * additional creator id...
985 ns
= current
->nsproxy
->ipc_ns
;
986 shp
= shm_lock_check(ns
, shmid
);
993 if (ipcperms(ns
, &shp
->shm_perm
, acc_mode
))
996 err
= security_shm_shmat(shp
, shmaddr
, shmflg
);
1000 path
= shp
->shm_file
->f_path
;
1003 size
= i_size_read(path
.dentry
->d_inode
);
1007 sfd
= kzalloc(sizeof(*sfd
), GFP_KERNEL
);
1009 goto out_put_dentry
;
1011 file
= alloc_file(&path
, f_mode
,
1012 is_file_hugepages(shp
->shm_file
) ?
1013 &shm_file_operations_huge
:
1014 &shm_file_operations
);
1018 file
->private_data
= sfd
;
1019 file
->f_mapping
= shp
->shm_file
->f_mapping
;
1020 sfd
->id
= shp
->shm_perm
.id
;
1021 sfd
->ns
= get_ipc_ns(ns
);
1022 sfd
->file
= shp
->shm_file
;
1025 down_write(¤t
->mm
->mmap_sem
);
1026 if (addr
&& !(shmflg
& SHM_REMAP
)) {
1028 if (find_vma_intersection(current
->mm
, addr
, addr
+ size
))
1031 * If shm segment goes below stack, make sure there is some
1032 * space left for the stack to grow (at least 4 pages).
1034 if (addr
< current
->mm
->start_stack
&&
1035 addr
> current
->mm
->start_stack
- size
- PAGE_SIZE
* 5)
1039 user_addr
= do_mmap (file
, addr
, size
, prot
, flags
, 0);
1042 if (IS_ERR_VALUE(user_addr
))
1043 err
= (long)user_addr
;
1045 up_write(¤t
->mm
->mmap_sem
);
1050 down_write(&shm_ids(ns
).rw_mutex
);
1051 shp
= shm_lock(ns
, shmid
);
1052 BUG_ON(IS_ERR(shp
));
1054 if (shm_may_destroy(ns
, shp
))
1055 shm_destroy(ns
, shp
);
1058 up_write(&shm_ids(ns
).rw_mutex
);
1074 SYSCALL_DEFINE3(shmat
, int, shmid
, char __user
*, shmaddr
, int, shmflg
)
1079 err
= do_shmat(shmid
, shmaddr
, shmflg
, &ret
);
1082 force_successful_syscall_return();
1087 * detach and kill segment if marked destroyed.
1088 * The work is done in shm_close.
1090 SYSCALL_DEFINE1(shmdt
, char __user
*, shmaddr
)
1092 struct mm_struct
*mm
= current
->mm
;
1093 struct vm_area_struct
*vma
;
1094 unsigned long addr
= (unsigned long)shmaddr
;
1095 int retval
= -EINVAL
;
1098 struct vm_area_struct
*next
;
1101 if (addr
& ~PAGE_MASK
)
1104 down_write(&mm
->mmap_sem
);
1107 * This function tries to be smart and unmap shm segments that
1108 * were modified by partial mlock or munmap calls:
1109 * - It first determines the size of the shm segment that should be
1110 * unmapped: It searches for a vma that is backed by shm and that
1111 * started at address shmaddr. It records it's size and then unmaps
1113 * - Then it unmaps all shm vmas that started at shmaddr and that
1114 * are within the initially determined size.
1115 * Errors from do_munmap are ignored: the function only fails if
1116 * it's called with invalid parameters or if it's called to unmap
1117 * a part of a vma. Both calls in this function are for full vmas,
1118 * the parameters are directly copied from the vma itself and always
1119 * valid - therefore do_munmap cannot fail. (famous last words?)
1122 * If it had been mremap()'d, the starting address would not
1123 * match the usual checks anyway. So assume all vma's are
1124 * above the starting address given.
1126 vma
= find_vma(mm
, addr
);
1130 next
= vma
->vm_next
;
1133 * Check if the starting address would match, i.e. it's
1134 * a fragment created by mprotect() and/or munmap(), or it
1135 * otherwise it starts at this address with no hassles.
1137 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1138 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
) {
1141 size
= vma
->vm_file
->f_path
.dentry
->d_inode
->i_size
;
1142 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1144 * We discovered the size of the shm segment, so
1145 * break out of here and fall through to the next
1146 * loop that uses the size information to stop
1147 * searching for matching vma's.
1157 * We need look no further than the maximum address a fragment
1158 * could possibly have landed at. Also cast things to loff_t to
1159 * prevent overflows and make comparisons vs. equal-width types.
1161 size
= PAGE_ALIGN(size
);
1162 while (vma
&& (loff_t
)(vma
->vm_end
- addr
) <= size
) {
1163 next
= vma
->vm_next
;
1165 /* finding a matching vma now does not alter retval */
1166 if ((vma
->vm_ops
== &shm_vm_ops
) &&
1167 (vma
->vm_start
- addr
)/PAGE_SIZE
== vma
->vm_pgoff
)
1169 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1173 #else /* CONFIG_MMU */
1174 /* under NOMMU conditions, the exact address to be destroyed must be
1177 if (vma
->vm_start
== addr
&& vma
->vm_ops
== &shm_vm_ops
) {
1178 do_munmap(mm
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
);
1184 up_write(&mm
->mmap_sem
);
1188 #ifdef CONFIG_PROC_FS
1189 static int sysvipc_shm_proc_show(struct seq_file
*s
, void *it
)
1191 struct shmid_kernel
*shp
= it
;
1192 unsigned long rss
= 0, swp
= 0;
1194 shm_add_rss_swap(shp
, &rss
, &swp
);
1196 #if BITS_PER_LONG <= 32
1197 #define SIZE_SPEC "%10lu"
1199 #define SIZE_SPEC "%21lu"
1202 return seq_printf(s
,
1203 "%10d %10d %4o " SIZE_SPEC
" %5u %5u "
1204 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1205 SIZE_SPEC
" " SIZE_SPEC
"\n",