1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* linux/ipc/shm.c
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.
8	*
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>
16	*
17	* support for audit of ipc object properties and permission changes
18	* Dustin Kirkland <dustin.kirkland@us.ibm.com>
19	*
20	* namespaces support
21	* OpenVZ, SWsoft Inc.
22	* Pavel Emelianov <xemul@openvz.org>
23	*
24	* Better ipc lock (kern_ipc_perm.lock) handling
25	* Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
26	*/
27
28	#include <linux/slab.h>
29	#include <linux/mm.h>
30	#include <linux/hugetlb.h>
31	#include <linux/shm.h>
32	#include <linux/init.h>
33	#include <linux/file.h>
34	#include <linux/mman.h>
35	#include <linux/shmem_fs.h>
36	#include <linux/security.h>
37	#include <linux/syscalls.h>
38	#include <linux/audit.h>
39	#include <linux/capability.h>
40	#include <linux/ptrace.h>
41	#include <linux/seq_file.h>
42	#include <linux/rwsem.h>
43	#include <linux/nsproxy.h>
44	#include <linux/mount.h>
45	#include <linux/ipc_namespace.h>
46	#include <linux/rhashtable.h>
47
48	#include <linux/uaccess.h>
49
50	#include "util.h"
51
52	struct shmid_kernel /* private to the kernel */
53	{
54	struct kern_ipc_perm shm_perm;
55	struct file *shm_file;
56	unsigned long shm_nattch;
57	unsigned long shm_segsz;
58	time64_t shm_atim;
59	time64_t shm_dtim;
60	time64_t shm_ctim;
61	struct pid *shm_cprid;
62	struct pid *shm_lprid;
63	struct ucounts *mlock_ucounts;
64
65	/*
66	* The task created the shm object, for
67	* task_lock(shp->shm_creator)
68	*/
69	struct task_struct *shm_creator;
70
71	/*
72	* List by creator. task_lock(->shm_creator) required for read/write.
73	* If list_empty(), then the creator is dead already.
74	*/
75	struct list_head shm_clist;
76	struct ipc_namespace *ns;
77	} __randomize_layout;
78
79	/* shm_mode upper byte flags */
80	#define SHM_DEST 01000 /* segment will be destroyed on last detach */
81	#define SHM_LOCKED 02000 /* segment will not be swapped */
82
83	struct shm_file_data {
84	int id;
85	struct ipc_namespace *ns;
86	struct file *file;
87	const struct vm_operations_struct *vm_ops;
88	};
89
90	#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
91
92	static const struct file_operations shm_file_operations;
93	static const struct vm_operations_struct shm_vm_ops;
94
95	#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
96
97	#define shm_unlock(shp) \
98	ipc_unlock(&(shp)->shm_perm)
99
100	static int newseg(struct ipc_namespace *, struct ipc_params *);
101	static void shm_open(struct vm_area_struct *vma);
102	static void shm_close(struct vm_area_struct *vma);
103	static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
104	#ifdef CONFIG_PROC_FS
105	static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
106	#endif
107
108	void shm_init_ns(struct ipc_namespace *ns)
109	{
110	ns->shm_ctlmax = SHMMAX;
111	ns->shm_ctlall = SHMALL;
112	ns->shm_ctlmni = SHMMNI;
113	ns->shm_rmid_forced = 0;
114	ns->shm_tot = 0;
115	ipc_init_ids(ids: &shm_ids(ns));
116	}
117
118	/*
119	* Called with shm_ids.rwsem (writer) and the shp structure locked.
120	* Only shm_ids.rwsem remains locked on exit.
121	*/
122	static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
123	{
124	struct shmid_kernel *shp;
125
126	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
127	WARN_ON(ns != shp->ns);
128
129	if (shp->shm_nattch) {
130	shp->shm_perm.mode |= SHM_DEST;
131	/* Do not find it any more */
132	ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
133	shm_unlock(shp);
134	} else
135	shm_destroy(ns, shp);
136	}
137
138	#ifdef CONFIG_IPC_NS
139	void shm_exit_ns(struct ipc_namespace *ns)
140	{
141	free_ipcs(ns, ids: &shm_ids(ns), free: do_shm_rmid);
142	idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
143	rhashtable_destroy(ht: &ns->ids[IPC_SHM_IDS].key_ht);
144	}
145	#endif
146
147	static int __init ipc_ns_init(void)
148	{
149	shm_init_ns(ns: &init_ipc_ns);
150	return 0;
151	}
152
153	pure_initcall(ipc_ns_init);
154
155	void __init shm_init(void)
156	{
157	ipc_init_proc_interface(path: "sysvipc/shm",
158	#if BITS_PER_LONG <= 32
159	" key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
160	#else
161	header: " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
162	#endif
163	IPC_SHM_IDS, show: sysvipc_shm_proc_show);
164	}
165
166	static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
167	{
168	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(ids: &shm_ids(ns), id);
169
170	if (IS_ERR(ptr: ipcp))
171	return ERR_CAST(ptr: ipcp);
172
173	return container_of(ipcp, struct shmid_kernel, shm_perm);
174	}
175
176	static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
177	{
178	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(ids: &shm_ids(ns), id);
179
180	if (IS_ERR(ptr: ipcp))
181	return ERR_CAST(ptr: ipcp);
182
183	return container_of(ipcp, struct shmid_kernel, shm_perm);
184	}
185
186	/*
187	* shm_lock_(check_) routines are called in the paths where the rwsem
188	* is not necessarily held.
189	*/
190	static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
191	{
192	struct kern_ipc_perm *ipcp;
193
194	rcu_read_lock();
195	ipcp = ipc_obtain_object_idr(ids: &shm_ids(ns), id);
196	if (IS_ERR(ptr: ipcp))
197	goto err;
198
199	ipc_lock_object(perm: ipcp);
200	/*
201	* ipc_rmid() may have already freed the ID while ipc_lock_object()
202	* was spinning: here verify that the structure is still valid.
203	* Upon races with RMID, return -EIDRM, thus indicating that
204	* the ID points to a removed identifier.
205	*/
206	if (ipc_valid_object(perm: ipcp)) {
207	/* return a locked ipc object upon success */
208	return container_of(ipcp, struct shmid_kernel, shm_perm);
209	}
210
211	ipc_unlock_object(perm: ipcp);
212	ipcp = ERR_PTR(error: -EIDRM);
213	err:
214	rcu_read_unlock();
215	/*
216	* Callers of shm_lock() must validate the status of the returned ipc
217	* object pointer and error out as appropriate.
218	*/
219	return ERR_CAST(ptr: ipcp);
220	}
221
222	static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
223	{
224	rcu_read_lock();
225	ipc_lock_object(perm: &ipcp->shm_perm);
226	}
227
228	static void shm_rcu_free(struct rcu_head *head)
229	{
230	struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
231	rcu);
232	struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
233	shm_perm);
234	security_shm_free(shp: &shp->shm_perm);
235	kfree(objp: shp);
236	}
237
238	/*
239	* It has to be called with shp locked.
240	* It must be called before ipc_rmid()
241	*/
242	static inline void shm_clist_rm(struct shmid_kernel *shp)
243	{
244	struct task_struct *creator;
245
246	/* ensure that shm_creator does not disappear */
247	rcu_read_lock();
248
249	/*
250	* A concurrent exit_shm may do a list_del_init() as well.
251	* Just do nothing if exit_shm already did the work
252	*/
253	if (!list_empty(head: &shp->shm_clist)) {
254	/*
255	* shp->shm_creator is guaranteed to be valid *only*
256	* if shp->shm_clist is not empty.
257	*/
258	creator = shp->shm_creator;
259
260	task_lock(p: creator);
261	/*
262	* list_del_init() is a nop if the entry was already removed
263	* from the list.
264	*/
265	list_del_init(entry: &shp->shm_clist);
266	task_unlock(p: creator);
267	}
268	rcu_read_unlock();
269	}
270
271	static inline void shm_rmid(struct shmid_kernel *s)
272	{
273	shm_clist_rm(shp: s);
274	ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
275	}
276
277
278	static int __shm_open(struct shm_file_data *sfd)
279	{
280	struct shmid_kernel *shp;
281
282	shp = shm_lock(ns: sfd->ns, id: sfd->id);
283
284	if (IS_ERR(ptr: shp))
285	return PTR_ERR(ptr: shp);
286
287	if (shp->shm_file != sfd->file) {
288	/* ID was reused */
289	shm_unlock(shp);
290	return -EINVAL;
291	}
292
293	shp->shm_atim = ktime_get_real_seconds();
294	ipc_update_pid(pos: &shp->shm_lprid, pid: task_tgid(current));
295	shp->shm_nattch++;
296	shm_unlock(shp);
297	return 0;
298	}
299
300	/* This is called by fork, once for every shm attach. */
301	static void shm_open(struct vm_area_struct *vma)
302	{
303	struct file *file = vma->vm_file;
304	struct shm_file_data *sfd = shm_file_data(file);
305	int err;
306
307	/* Always call underlying open if present */
308	if (sfd->vm_ops->open)
309	sfd->vm_ops->open(vma);
310
311	err = __shm_open(sfd);
312	/*
313	* We raced in the idr lookup or with shm_destroy().
314	* Either way, the ID is busted.
315	*/
316	WARN_ON_ONCE(err);
317	}
318
319	/*
320	* shm_destroy - free the struct shmid_kernel
321	*
322	* @ns: namespace
323	* @shp: struct to free
324	*
325	* It has to be called with shp and shm_ids.rwsem (writer) locked,
326	* but returns with shp unlocked and freed.
327	*/
328	static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
329	{
330	struct file *shm_file;
331
332	shm_file = shp->shm_file;
333	shp->shm_file = NULL;
334	ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
335	shm_rmid(s: shp);
336	shm_unlock(shp);
337	if (!is_file_hugepages(file: shm_file))
338	shmem_lock(file: shm_file, lock: 0, ucounts: shp->mlock_ucounts);
339	fput(shm_file);
340	ipc_update_pid(pos: &shp->shm_cprid, NULL);
341	ipc_update_pid(pos: &shp->shm_lprid, NULL);
342	ipc_rcu_putref(ptr: &shp->shm_perm, func: shm_rcu_free);
343	}
344
345	/*
346	* shm_may_destroy - identifies whether shm segment should be destroyed now
347	*
348	* Returns true if and only if there are no active users of the segment and
349	* one of the following is true:
350	*
351	* 1) shmctl(id, IPC_RMID, NULL) was called for this shp
352	*
353	* 2) sysctl kernel.shm_rmid_forced is set to 1.
354	*/
355	static bool shm_may_destroy(struct shmid_kernel *shp)
356	{
357	return (shp->shm_nattch == 0) &&
358	(shp->ns->shm_rmid_forced ||
359	(shp->shm_perm.mode & SHM_DEST));
360	}
361
362	/*
363	* remove the attach descriptor vma.
364	* free memory for segment if it is marked destroyed.
365	* The descriptor has already been removed from the current->mm->mmap list
366	* and will later be kfree()d.
367	*/
368	static void __shm_close(struct shm_file_data *sfd)
369	{
370	struct shmid_kernel *shp;
371	struct ipc_namespace *ns = sfd->ns;
372
373	down_write(sem: &shm_ids(ns).rwsem);
374	/* remove from the list of attaches of the shm segment */
375	shp = shm_lock(ns, id: sfd->id);
376
377	/*
378	* We raced in the idr lookup or with shm_destroy().
379	* Either way, the ID is busted.
380	*/
381	if (WARN_ON_ONCE(IS_ERR(shp)))
382	goto done; /* no-op */
383
384	ipc_update_pid(pos: &shp->shm_lprid, pid: task_tgid(current));
385	shp->shm_dtim = ktime_get_real_seconds();
386	shp->shm_nattch--;
387	if (shm_may_destroy(shp))
388	shm_destroy(ns, shp);
389	else
390	shm_unlock(shp);
391	done:
392	up_write(sem: &shm_ids(ns).rwsem);
393	}
394
395	static void shm_close(struct vm_area_struct *vma)
396	{
397	struct file *file = vma->vm_file;
398	struct shm_file_data *sfd = shm_file_data(file);
399
400	/* Always call underlying close if present */
401	if (sfd->vm_ops->close)
402	sfd->vm_ops->close(vma);
403
404	__shm_close(sfd);
405	}
406
407	/* Called with ns->shm_ids(ns).rwsem locked */
408	static int shm_try_destroy_orphaned(int id, void *p, void *data)
409	{
410	struct ipc_namespace *ns = data;
411	struct kern_ipc_perm *ipcp = p;
412	struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
413
414	/*
415	* We want to destroy segments without users and with already
416	* exit'ed originating process.
417	*
418	* As shp->* are changed under rwsem, it's safe to skip shp locking.
419	*/
420	if (!list_empty(head: &shp->shm_clist))
421	return 0;
422
423	if (shm_may_destroy(shp)) {
424	shm_lock_by_ptr(ipcp: shp);
425	shm_destroy(ns, shp);
426	}
427	return 0;
428	}
429
430	void shm_destroy_orphaned(struct ipc_namespace *ns)
431	{
432	down_write(sem: &shm_ids(ns).rwsem);
433	if (shm_ids(ns).in_use)
434	idr_for_each(&shm_ids(ns).ipcs_idr, fn: &shm_try_destroy_orphaned, data: ns);
435	up_write(sem: &shm_ids(ns).rwsem);
436	}
437
438	/* Locking assumes this will only be called with task == current */
439	void exit_shm(struct task_struct *task)
440	{
441	for (;;) {
442	struct shmid_kernel *shp;
443	struct ipc_namespace *ns;
444
445	task_lock(p: task);
446
447	if (list_empty(head: &task->sysvshm.shm_clist)) {
448	task_unlock(p: task);
449	break;
450	}
451
452	shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
453	shm_clist);
454
455	/*
456	* 1) Get pointer to the ipc namespace. It is worth to say
457	* that this pointer is guaranteed to be valid because
458	* shp lifetime is always shorter than namespace lifetime
459	* in which shp lives.
460	* We taken task_lock it means that shp won't be freed.
461	*/
462	ns = shp->ns;
463
464	/*
465	* 2) If kernel.shm_rmid_forced is not set then only keep track of
466	* which shmids are orphaned, so that a later set of the sysctl
467	* can clean them up.
468	*/
469	if (!ns->shm_rmid_forced)
470	goto unlink_continue;
471
472	/*
473	* 3) get a reference to the namespace.
474	* The refcount could be already 0. If it is 0, then
475	* the shm objects will be free by free_ipc_work().
476	*/
477	ns = get_ipc_ns_not_zero(ns);
478	if (!ns) {
479	unlink_continue:
480	list_del_init(entry: &shp->shm_clist);
481	task_unlock(p: task);
482	continue;
483	}
484
485	/*
486	* 4) get a reference to shp.
487	* This cannot fail: shm_clist_rm() is called before
488	* ipc_rmid(), thus the refcount cannot be 0.
489	*/
490	WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
491
492	/*
493	* 5) unlink the shm segment from the list of segments
494	* created by current.
495	* This must be done last. After unlinking,
496	* only the refcounts obtained above prevent IPC_RMID
497	* from destroying the segment or the namespace.
498	*/
499	list_del_init(entry: &shp->shm_clist);
500
501	task_unlock(p: task);
502
503	/*
504	* 6) we have all references
505	* Thus lock & if needed destroy shp.
506	*/
507	down_write(sem: &shm_ids(ns).rwsem);
508	shm_lock_by_ptr(ipcp: shp);
509	/*
510	* rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
511	* safe to call ipc_rcu_putref here
512	*/
513	ipc_rcu_putref(ptr: &shp->shm_perm, func: shm_rcu_free);
514
515	if (ipc_valid_object(perm: &shp->shm_perm)) {
516	if (shm_may_destroy(shp))
517	shm_destroy(ns, shp);
518	else
519	shm_unlock(shp);
520	} else {
521	/*
522	* Someone else deleted the shp from namespace
523	* idr/kht while we have waited.
524	* Just unlock and continue.
525	*/
526	shm_unlock(shp);
527	}
528
529	up_write(sem: &shm_ids(ns).rwsem);
530	put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
531	}
532	}
533
534	static vm_fault_t shm_fault(struct vm_fault *vmf)
535	{
536	struct file *file = vmf->vma->vm_file;
537	struct shm_file_data *sfd = shm_file_data(file);
538
539	return sfd->vm_ops->fault(vmf);
540	}
541
542	static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
543	{
544	struct file *file = vma->vm_file;
545	struct shm_file_data *sfd = shm_file_data(file);
546
547	if (sfd->vm_ops->may_split)
548	return sfd->vm_ops->may_split(vma, addr);
549
550	return 0;
551	}
552
553	static unsigned long shm_pagesize(struct vm_area_struct *vma)
554	{
555	struct file *file = vma->vm_file;
556	struct shm_file_data *sfd = shm_file_data(file);
557
558	if (sfd->vm_ops->pagesize)
559	return sfd->vm_ops->pagesize(vma);
560
561	return PAGE_SIZE;
562	}
563
564	#ifdef CONFIG_NUMA
565	static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
566	{
567	struct shm_file_data *sfd = shm_file_data(vma->vm_file);
568	int err = 0;
569
570	if (sfd->vm_ops->set_policy)
571	err = sfd->vm_ops->set_policy(vma, mpol);
572	return err;
573	}
574
575	static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
576	unsigned long addr, pgoff_t *ilx)
577	{
578	struct shm_file_data *sfd = shm_file_data(vma->vm_file);
579	struct mempolicy *mpol = vma->vm_policy;
580
581	if (sfd->vm_ops->get_policy)
582	mpol = sfd->vm_ops->get_policy(vma, addr, ilx);
583	return mpol;
584	}
585	#endif
586
587	static int shm_mmap(struct file *file, struct vm_area_struct *vma)
588	{
589	struct shm_file_data *sfd = shm_file_data(file);
590	int ret;
591
592	/*
593	* In case of remap_file_pages() emulation, the file can represent an
594	* IPC ID that was removed, and possibly even reused by another shm
595	* segment already. Propagate this case as an error to caller.
596	*/
597	ret = __shm_open(sfd);
598	if (ret)
599	return ret;
600
601	ret = call_mmap(file: sfd->file, vma);
602	if (ret) {
603	__shm_close(sfd);
604	return ret;
605	}
606	sfd->vm_ops = vma->vm_ops;
607	#ifdef CONFIG_MMU
608	WARN_ON(!sfd->vm_ops->fault);
609	#endif
610	vma->vm_ops = &shm_vm_ops;
611	return 0;
612	}
613
614	static int shm_release(struct inode *ino, struct file *file)
615	{
616	struct shm_file_data *sfd = shm_file_data(file);
617
618	put_ipc_ns(ns: sfd->ns);
619	fput(sfd->file);
620	shm_file_data(file) = NULL;
621	kfree(objp: sfd);
622	return 0;
623	}
624
625	static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
626	{
627	struct shm_file_data *sfd = shm_file_data(file);
628
629	if (!sfd->file->f_op->fsync)
630	return -EINVAL;
631	return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
632	}
633
634	static long shm_fallocate(struct file *file, int mode, loff_t offset,
635	loff_t len)
636	{
637	struct shm_file_data *sfd = shm_file_data(file);
638
639	if (!sfd->file->f_op->fallocate)
640	return -EOPNOTSUPP;
641	return sfd->file->f_op->fallocate(file, mode, offset, len);
642	}
643
644	static unsigned long shm_get_unmapped_area(struct file *file,
645	unsigned long addr, unsigned long len, unsigned long pgoff,
646	unsigned long flags)
647	{
648	struct shm_file_data *sfd = shm_file_data(file);
649
650	return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
651	pgoff, flags);
652	}
653
654	static const struct file_operations shm_file_operations = {
655	.mmap = shm_mmap,
656	.fsync = shm_fsync,
657	.release = shm_release,
658	.get_unmapped_area = shm_get_unmapped_area,
659	.llseek = noop_llseek,
660	.fallocate = shm_fallocate,
661	};
662
663	/*
664	* shm_file_operations_huge is now identical to shm_file_operations,
665	* but we keep it distinct for the sake of is_file_shm_hugepages().
666	*/
667	static const struct file_operations shm_file_operations_huge = {
668	.mmap = shm_mmap,
669	.fsync = shm_fsync,
670	.release = shm_release,
671	.get_unmapped_area = shm_get_unmapped_area,
672	.llseek = noop_llseek,
673	.fallocate = shm_fallocate,
674	};
675
676	bool is_file_shm_hugepages(struct file *file)
677	{
678	return file->f_op == &shm_file_operations_huge;
679	}
680
681	static const struct vm_operations_struct shm_vm_ops = {
682	.open = shm_open, /* callback for a new vm-area open */
683	.close = shm_close, /* callback for when the vm-area is released */
684	.fault = shm_fault,
685	.may_split = shm_may_split,
686	.pagesize = shm_pagesize,
687	#if defined(CONFIG_NUMA)
688	.set_policy = shm_set_policy,
689	.get_policy = shm_get_policy,
690	#endif
691	};
692
693	/**
694	* newseg - Create a new shared memory segment
695	* @ns: namespace
696	* @params: ptr to the structure that contains key, size and shmflg
697	*
698	* Called with shm_ids.rwsem held as a writer.
699	*/
700	static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
701	{
702	key_t key = params->key;
703	int shmflg = params->flg;
704	size_t size = params->u.size;
705	int error;
706	struct shmid_kernel *shp;
707	size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
708	struct file *file;
709	char name[13];
710	vm_flags_t acctflag = 0;
711
712	if (size < SHMMIN || size > ns->shm_ctlmax)
713	return -EINVAL;
714
715	if (numpages << PAGE_SHIFT < size)
716	return -ENOSPC;
717
718	if (ns->shm_tot + numpages < ns->shm_tot ||
719	ns->shm_tot + numpages > ns->shm_ctlall)
720	return -ENOSPC;
721
722	shp = kmalloc(size: sizeof(*shp), GFP_KERNEL_ACCOUNT);
723	if (unlikely(!shp))
724	return -ENOMEM;
725
726	shp->shm_perm.key = key;
727	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
728	shp->mlock_ucounts = NULL;
729
730	shp->shm_perm.security = NULL;
731	error = security_shm_alloc(shp: &shp->shm_perm);
732	if (error) {
733	kfree(objp: shp);
734	return error;
735	}
736
737	sprintf(buf: name, fmt: "SYSV%08x", key);
738	if (shmflg & SHM_HUGETLB) {
739	struct hstate *hs;
740	size_t hugesize;
741
742	hs = hstate_sizelog(page_size_log: (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
743	if (!hs) {
744	error = -EINVAL;
745	goto no_file;
746	}
747	hugesize = ALIGN(size, huge_page_size(hs));
748
749	/* hugetlb_file_setup applies strict accounting */
750	if (shmflg & SHM_NORESERVE)
751	acctflag = VM_NORESERVE;
752	file = hugetlb_file_setup(name, size: hugesize, acct: acctflag,
753	creat_flags: HUGETLB_SHMFS_INODE, page_size_log: (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
754	} else {
755	/*
756	* Do not allow no accounting for OVERCOMMIT_NEVER, even
757	* if it's asked for.
758	*/
759	if ((shmflg & SHM_NORESERVE) &&
760	sysctl_overcommit_memory != OVERCOMMIT_NEVER)
761	acctflag = VM_NORESERVE;
762	file = shmem_kernel_file_setup(name, size, flags: acctflag);
763	}
764	error = PTR_ERR(ptr: file);
765	if (IS_ERR(ptr: file))
766	goto no_file;
767
768	shp->shm_cprid = get_pid(pid: task_tgid(current));
769	shp->shm_lprid = NULL;
770	shp->shm_atim = shp->shm_dtim = 0;
771	shp->shm_ctim = ktime_get_real_seconds();
772	shp->shm_segsz = size;
773	shp->shm_nattch = 0;
774	shp->shm_file = file;
775	shp->shm_creator = current;
776
777	/* ipc_addid() locks shp upon success. */
778	error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
779	if (error < 0)
780	goto no_id;
781
782	shp->ns = ns;
783
784	task_lock(current);
785	list_add(new: &shp->shm_clist, head: &current->sysvshm.shm_clist);
786	task_unlock(current);
787
788	/*
789	* shmid gets reported as "inode#" in /proc/pid/maps.
790	* proc-ps tools use this. Changing this will break them.
791	*/
792	file_inode(f: file)->i_ino = shp->shm_perm.id;
793
794	ns->shm_tot += numpages;
795	error = shp->shm_perm.id;
796
797	ipc_unlock_object(perm: &shp->shm_perm);
798	rcu_read_unlock();
799	return error;
800
801	no_id:
802	ipc_update_pid(pos: &shp->shm_cprid, NULL);
803	ipc_update_pid(pos: &shp->shm_lprid, NULL);
804	fput(file);
805	ipc_rcu_putref(ptr: &shp->shm_perm, func: shm_rcu_free);
806	return error;
807	no_file:
808	call_rcu(head: &shp->shm_perm.rcu, func: shm_rcu_free);
809	return error;
810	}
811
812	/*
813	* Called with shm_ids.rwsem and ipcp locked.
814	*/
815	static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
816	{
817	struct shmid_kernel *shp;
818
819	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
820	if (shp->shm_segsz < params->u.size)
821	return -EINVAL;
822
823	return 0;
824	}
825
826	long ksys_shmget(key_t key, size_t size, int shmflg)
827	{
828	struct ipc_namespace *ns;
829	static const struct ipc_ops shm_ops = {
830	.getnew = newseg,
831	.associate = security_shm_associate,
832	.more_checks = shm_more_checks,
833	};
834	struct ipc_params shm_params;
835
836	ns = current->nsproxy->ipc_ns;
837
838	shm_params.key = key;
839	shm_params.flg = shmflg;
840	shm_params.u.size = size;
841
842	return ipcget(ns, ids: &shm_ids(ns), ops: &shm_ops, params: &shm_params);
843	}
844
845	SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
846	{
847	return ksys_shmget(key, size, shmflg);
848	}
849
850	static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
851	{
852	switch (version) {
853	case IPC_64:
854	return copy_to_user(to: buf, from: in, n: sizeof(*in));
855	case IPC_OLD:
856	{
857	struct shmid_ds out;
858
859	memset(&out, 0, sizeof(out));
860	ipc64_perm_to_ipc_perm(in: &in->shm_perm, out: &out.shm_perm);
861	out.shm_segsz = in->shm_segsz;
862	out.shm_atime = in->shm_atime;
863	out.shm_dtime = in->shm_dtime;
864	out.shm_ctime = in->shm_ctime;
865	out.shm_cpid = in->shm_cpid;
866	out.shm_lpid = in->shm_lpid;
867	out.shm_nattch = in->shm_nattch;
868
869	return copy_to_user(to: buf, from: &out, n: sizeof(out));
870	}
871	default:
872	return -EINVAL;
873	}
874	}
875
876	static inline unsigned long
877	copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
878	{
879	switch (version) {
880	case IPC_64:
881	if (copy_from_user(to: out, from: buf, n: sizeof(*out)))
882	return -EFAULT;
883	return 0;
884	case IPC_OLD:
885	{
886	struct shmid_ds tbuf_old;
887
888	if (copy_from_user(to: &tbuf_old, from: buf, n: sizeof(tbuf_old)))
889	return -EFAULT;
890
891	out->shm_perm.uid = tbuf_old.shm_perm.uid;
892	out->shm_perm.gid = tbuf_old.shm_perm.gid;
893	out->shm_perm.mode = tbuf_old.shm_perm.mode;
894
895	return 0;
896	}
897	default:
898	return -EINVAL;
899	}
900	}
901
902	static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
903	{
904	switch (version) {
905	case IPC_64:
906	return copy_to_user(to: buf, from: in, n: sizeof(*in));
907	case IPC_OLD:
908	{
909	struct shminfo out;
910
911	if (in->shmmax > INT_MAX)
912	out.shmmax = INT_MAX;
913	else
914	out.shmmax = (int)in->shmmax;
915
916	out.shmmin = in->shmmin;
917	out.shmmni = in->shmmni;
918	out.shmseg = in->shmseg;
919	out.shmall = in->shmall;
920
921	return copy_to_user(to: buf, from: &out, n: sizeof(out));
922	}
923	default:
924	return -EINVAL;
925	}
926	}
927
928	/*
929	* Calculate and add used RSS and swap pages of a shm.
930	* Called with shm_ids.rwsem held as a reader
931	*/
932	static void shm_add_rss_swap(struct shmid_kernel *shp,
933	unsigned long *rss_add, unsigned long *swp_add)
934	{
935	struct inode *inode;
936
937	inode = file_inode(f: shp->shm_file);
938
939	if (is_file_hugepages(file: shp->shm_file)) {
940	struct address_space *mapping = inode->i_mapping;
941	struct hstate *h = hstate_file(f: shp->shm_file);
942	*rss_add += pages_per_huge_page(h) * mapping->nrpages;
943	} else {
944	#ifdef CONFIG_SHMEM
945	struct shmem_inode_info *info = SHMEM_I(inode);
946
947	spin_lock_irq(lock: &info->lock);
948	*rss_add += inode->i_mapping->nrpages;
949	*swp_add += info->swapped;
950	spin_unlock_irq(lock: &info->lock);
951	#else
952	*rss_add += inode->i_mapping->nrpages;
953	#endif
954	}
955	}
956
957	/*
958	* Called with shm_ids.rwsem held as a reader
959	*/
960	static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
961	unsigned long *swp)
962	{
963	int next_id;
964	int total, in_use;
965
966	*rss = 0;
967	*swp = 0;
968
969	in_use = shm_ids(ns).in_use;
970
971	for (total = 0, next_id = 0; total < in_use; next_id++) {
972	struct kern_ipc_perm *ipc;
973	struct shmid_kernel *shp;
974
975	ipc = idr_find(&shm_ids(ns).ipcs_idr, id: next_id);
976	if (ipc == NULL)
977	continue;
978	shp = container_of(ipc, struct shmid_kernel, shm_perm);
979
980	shm_add_rss_swap(shp, rss_add: rss, swp_add: swp);
981
982	total++;
983	}
984	}
985
986	/*
987	* This function handles some shmctl commands which require the rwsem
988	* to be held in write mode.
989	* NOTE: no locks must be held, the rwsem is taken inside this function.
990	*/
991	static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
992	struct shmid64_ds *shmid64)
993	{
994	struct kern_ipc_perm *ipcp;
995	struct shmid_kernel *shp;
996	int err;
997
998	down_write(sem: &shm_ids(ns).rwsem);
999	rcu_read_lock();
1000
1001	ipcp = ipcctl_obtain_check(ns, ids: &shm_ids(ns), id: shmid, cmd,
1002	perm: &shmid64->shm_perm, extra_perm: 0);
1003	if (IS_ERR(ptr: ipcp)) {
1004	err = PTR_ERR(ptr: ipcp);
1005	goto out_unlock1;
1006	}
1007
1008	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1009
1010	err = security_shm_shmctl(shp: &shp->shm_perm, cmd);
1011	if (err)
1012	goto out_unlock1;
1013
1014	switch (cmd) {
1015	case IPC_RMID:
1016	ipc_lock_object(perm: &shp->shm_perm);
1017	/* do_shm_rmid unlocks the ipc object and rcu */
1018	do_shm_rmid(ns, ipcp);
1019	goto out_up;
1020	case IPC_SET:
1021	ipc_lock_object(perm: &shp->shm_perm);
1022	err = ipc_update_perm(in: &shmid64->shm_perm, out: ipcp);
1023	if (err)
1024	goto out_unlock0;
1025	shp->shm_ctim = ktime_get_real_seconds();
1026	break;
1027	default:
1028	err = -EINVAL;
1029	goto out_unlock1;
1030	}
1031
1032	out_unlock0:
1033	ipc_unlock_object(perm: &shp->shm_perm);
1034	out_unlock1:
1035	rcu_read_unlock();
1036	out_up:
1037	up_write(sem: &shm_ids(ns).rwsem);
1038	return err;
1039	}
1040
1041	static int shmctl_ipc_info(struct ipc_namespace *ns,
1042	struct shminfo64 *shminfo)
1043	{
1044	int err = security_shm_shmctl(NULL, IPC_INFO);
1045	if (!err) {
1046	memset(shminfo, 0, sizeof(*shminfo));
1047	shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1048	shminfo->shmmax = ns->shm_ctlmax;
1049	shminfo->shmall = ns->shm_ctlall;
1050	shminfo->shmmin = SHMMIN;
1051	down_read(sem: &shm_ids(ns).rwsem);
1052	err = ipc_get_maxidx(ids: &shm_ids(ns));
1053	up_read(sem: &shm_ids(ns).rwsem);
1054	if (err < 0)
1055	err = 0;
1056	}
1057	return err;
1058	}
1059
1060	static int shmctl_shm_info(struct ipc_namespace *ns,
1061	struct shm_info *shm_info)
1062	{
1063	int err = security_shm_shmctl(NULL, SHM_INFO);
1064	if (!err) {
1065	memset(shm_info, 0, sizeof(*shm_info));
1066	down_read(sem: &shm_ids(ns).rwsem);
1067	shm_info->used_ids = shm_ids(ns).in_use;
1068	shm_get_stat(ns, rss: &shm_info->shm_rss, swp: &shm_info->shm_swp);
1069	shm_info->shm_tot = ns->shm_tot;
1070	shm_info->swap_attempts = 0;
1071	shm_info->swap_successes = 0;
1072	err = ipc_get_maxidx(ids: &shm_ids(ns));
1073	up_read(sem: &shm_ids(ns).rwsem);
1074	if (err < 0)
1075	err = 0;
1076	}
1077	return err;
1078	}
1079
1080	static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1081	int cmd, struct shmid64_ds *tbuf)
1082	{
1083	struct shmid_kernel *shp;
1084	int err;
1085
1086	memset(tbuf, 0, sizeof(*tbuf));
1087
1088	rcu_read_lock();
1089	if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1090	shp = shm_obtain_object(ns, id: shmid);
1091	if (IS_ERR(ptr: shp)) {
1092	err = PTR_ERR(ptr: shp);
1093	goto out_unlock;
1094	}
1095	} else { /* IPC_STAT */
1096	shp = shm_obtain_object_check(ns, id: shmid);
1097	if (IS_ERR(ptr: shp)) {
1098	err = PTR_ERR(ptr: shp);
1099	goto out_unlock;
1100	}
1101	}
1102
1103	/*
1104	* Semantically SHM_STAT_ANY ought to be identical to
1105	* that functionality provided by the /proc/sysvipc/
1106	* interface. As such, only audit these calls and
1107	* do not do traditional S_IRUGO permission checks on
1108	* the ipc object.
1109	*/
1110	if (cmd == SHM_STAT_ANY)
1111	audit_ipc_obj(ipcp: &shp->shm_perm);
1112	else {
1113	err = -EACCES;
1114	if (ipcperms(ns, ipcp: &shp->shm_perm, S_IRUGO))
1115	goto out_unlock;
1116	}
1117
1118	err = security_shm_shmctl(shp: &shp->shm_perm, cmd);
1119	if (err)
1120	goto out_unlock;
1121
1122	ipc_lock_object(perm: &shp->shm_perm);
1123
1124	if (!ipc_valid_object(perm: &shp->shm_perm)) {
1125	ipc_unlock_object(perm: &shp->shm_perm);
1126	err = -EIDRM;
1127	goto out_unlock;
1128	}
1129
1130	kernel_to_ipc64_perm(in: &shp->shm_perm, out: &tbuf->shm_perm);
1131	tbuf->shm_segsz = shp->shm_segsz;
1132	tbuf->shm_atime = shp->shm_atim;
1133	tbuf->shm_dtime = shp->shm_dtim;
1134	tbuf->shm_ctime = shp->shm_ctim;
1135	#ifndef CONFIG_64BIT
1136	tbuf->shm_atime_high = shp->shm_atim >> 32;
1137	tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1138	tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1139	#endif
1140	tbuf->shm_cpid = pid_vnr(pid: shp->shm_cprid);
1141	tbuf->shm_lpid = pid_vnr(pid: shp->shm_lprid);
1142	tbuf->shm_nattch = shp->shm_nattch;
1143
1144	if (cmd == IPC_STAT) {
1145	/*
1146	* As defined in SUS:
1147	* Return 0 on success
1148	*/
1149	err = 0;
1150	} else {
1151	/*
1152	* SHM_STAT and SHM_STAT_ANY (both Linux specific)
1153	* Return the full id, including the sequence number
1154	*/
1155	err = shp->shm_perm.id;
1156	}
1157
1158	ipc_unlock_object(perm: &shp->shm_perm);
1159	out_unlock:
1160	rcu_read_unlock();
1161	return err;
1162	}
1163
1164	static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1165	{
1166	struct shmid_kernel *shp;
1167	struct file *shm_file;
1168	int err;
1169
1170	rcu_read_lock();
1171	shp = shm_obtain_object_check(ns, id: shmid);
1172	if (IS_ERR(ptr: shp)) {
1173	err = PTR_ERR(ptr: shp);
1174	goto out_unlock1;
1175	}
1176
1177	audit_ipc_obj(ipcp: &(shp->shm_perm));
1178	err = security_shm_shmctl(shp: &shp->shm_perm, cmd);
1179	if (err)
1180	goto out_unlock1;
1181
1182	ipc_lock_object(perm: &shp->shm_perm);
1183
1184	/* check if shm_destroy() is tearing down shp */
1185	if (!ipc_valid_object(perm: &shp->shm_perm)) {
1186	err = -EIDRM;
1187	goto out_unlock0;
1188	}
1189
1190	if (!ns_capable(ns: ns->user_ns, CAP_IPC_LOCK)) {
1191	kuid_t euid = current_euid();
1192
1193	if (!uid_eq(left: euid, right: shp->shm_perm.uid) &&
1194	!uid_eq(left: euid, right: shp->shm_perm.cuid)) {
1195	err = -EPERM;
1196	goto out_unlock0;
1197	}
1198	if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1199	err = -EPERM;
1200	goto out_unlock0;
1201	}
1202	}
1203
1204	shm_file = shp->shm_file;
1205	if (is_file_hugepages(file: shm_file))
1206	goto out_unlock0;
1207
1208	if (cmd == SHM_LOCK) {
1209	struct ucounts *ucounts = current_ucounts();
1210
1211	err = shmem_lock(file: shm_file, lock: 1, ucounts);
1212	if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1213	shp->shm_perm.mode |= SHM_LOCKED;
1214	shp->mlock_ucounts = ucounts;
1215	}
1216	goto out_unlock0;
1217	}
1218
1219	/* SHM_UNLOCK */
1220	if (!(shp->shm_perm.mode & SHM_LOCKED))
1221	goto out_unlock0;
1222	shmem_lock(file: shm_file, lock: 0, ucounts: shp->mlock_ucounts);
1223	shp->shm_perm.mode &= ~SHM_LOCKED;
1224	shp->mlock_ucounts = NULL;
1225	get_file(f: shm_file);
1226	ipc_unlock_object(perm: &shp->shm_perm);
1227	rcu_read_unlock();
1228	shmem_unlock_mapping(mapping: shm_file->f_mapping);
1229
1230	fput(shm_file);
1231	return err;
1232
1233	out_unlock0:
1234	ipc_unlock_object(perm: &shp->shm_perm);
1235	out_unlock1:
1236	rcu_read_unlock();
1237	return err;
1238	}
1239
1240	static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1241	{
1242	int err;
1243	struct ipc_namespace *ns;
1244	struct shmid64_ds sem64;
1245
1246	if (cmd < 0 || shmid < 0)
1247	return -EINVAL;
1248
1249	ns = current->nsproxy->ipc_ns;
1250
1251	switch (cmd) {
1252	case IPC_INFO: {
1253	struct shminfo64 shminfo;
1254	err = shmctl_ipc_info(ns, shminfo: &shminfo);
1255	if (err < 0)
1256	return err;
1257	if (copy_shminfo_to_user(buf, in: &shminfo, version))
1258	err = -EFAULT;
1259	return err;
1260	}
1261	case SHM_INFO: {
1262	struct shm_info shm_info;
1263	err = shmctl_shm_info(ns, shm_info: &shm_info);
1264	if (err < 0)
1265	return err;
1266	if (copy_to_user(to: buf, from: &shm_info, n: sizeof(shm_info)))
1267	err = -EFAULT;
1268	return err;
1269	}
1270	case SHM_STAT:
1271	case SHM_STAT_ANY:
1272	case IPC_STAT: {
1273	err = shmctl_stat(ns, shmid, cmd, tbuf: &sem64);
1274	if (err < 0)
1275	return err;
1276	if (copy_shmid_to_user(buf, in: &sem64, version))
1277	err = -EFAULT;
1278	return err;
1279	}
1280	case IPC_SET:
1281	if (copy_shmid_from_user(out: &sem64, buf, version))
1282	return -EFAULT;
1283	fallthrough;
1284	case IPC_RMID:
1285	return shmctl_down(ns, shmid, cmd, shmid64: &sem64);
1286	case SHM_LOCK:
1287	case SHM_UNLOCK:
1288	return shmctl_do_lock(ns, shmid, cmd);
1289	default:
1290	return -EINVAL;
1291	}
1292	}
1293
1294	SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1295	{
1296	return ksys_shmctl(shmid, cmd, buf, IPC_64);
1297	}
1298
1299	#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1300	long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1301	{
1302	int version = ipc_parse_version(&cmd);
1303
1304	return ksys_shmctl(shmid, cmd, buf, version);
1305	}
1306
1307	SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1308	{
1309	return ksys_old_shmctl(shmid, cmd, buf);
1310	}
1311	#endif
1312
1313	#ifdef CONFIG_COMPAT
1314
1315	struct compat_shmid_ds {
1316	struct compat_ipc_perm shm_perm;
1317	int shm_segsz;
1318	old_time32_t shm_atime;
1319	old_time32_t shm_dtime;
1320	old_time32_t shm_ctime;
1321	compat_ipc_pid_t shm_cpid;
1322	compat_ipc_pid_t shm_lpid;
1323	unsigned short shm_nattch;
1324	unsigned short shm_unused;
1325	compat_uptr_t shm_unused2;
1326	compat_uptr_t shm_unused3;
1327	};
1328
1329	struct compat_shminfo64 {
1330	compat_ulong_t shmmax;
1331	compat_ulong_t shmmin;
1332	compat_ulong_t shmmni;
1333	compat_ulong_t shmseg;
1334	compat_ulong_t shmall;
1335	compat_ulong_t __unused1;
1336	compat_ulong_t __unused2;
1337	compat_ulong_t __unused3;
1338	compat_ulong_t __unused4;
1339	};
1340
1341	struct compat_shm_info {
1342	compat_int_t used_ids;
1343	compat_ulong_t shm_tot, shm_rss, shm_swp;
1344	compat_ulong_t swap_attempts, swap_successes;
1345	};
1346
1347	static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1348	int version)
1349	{
1350	if (in->shmmax > INT_MAX)
1351	in->shmmax = INT_MAX;
1352	if (version == IPC_64) {
1353	struct compat_shminfo64 info;
1354	memset(&info, 0, sizeof(info));
1355	info.shmmax = in->shmmax;
1356	info.shmmin = in->shmmin;
1357	info.shmmni = in->shmmni;
1358	info.shmseg = in->shmseg;
1359	info.shmall = in->shmall;
1360	return copy_to_user(to: buf, from: &info, n: sizeof(info));
1361	} else {
1362	struct shminfo info;
1363	memset(&info, 0, sizeof(info));
1364	info.shmmax = in->shmmax;
1365	info.shmmin = in->shmmin;
1366	info.shmmni = in->shmmni;
1367	info.shmseg = in->shmseg;
1368	info.shmall = in->shmall;
1369	return copy_to_user(to: buf, from: &info, n: sizeof(info));
1370	}
1371	}
1372
1373	static int put_compat_shm_info(struct shm_info *ip,
1374	struct compat_shm_info __user *uip)
1375	{
1376	struct compat_shm_info info;
1377
1378	memset(&info, 0, sizeof(info));
1379	info.used_ids = ip->used_ids;
1380	info.shm_tot = ip->shm_tot;
1381	info.shm_rss = ip->shm_rss;
1382	info.shm_swp = ip->shm_swp;
1383	info.swap_attempts = ip->swap_attempts;
1384	info.swap_successes = ip->swap_successes;
1385	return copy_to_user(to: uip, from: &info, n: sizeof(info));
1386	}
1387
1388	static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1389	int version)
1390	{
1391	if (version == IPC_64) {
1392	struct compat_shmid64_ds v;
1393	memset(&v, 0, sizeof(v));
1394	to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1395	v.shm_atime = lower_32_bits(in->shm_atime);
1396	v.shm_atime_high = upper_32_bits(in->shm_atime);
1397	v.shm_dtime = lower_32_bits(in->shm_dtime);
1398	v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1399	v.shm_ctime = lower_32_bits(in->shm_ctime);
1400	v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1401	v.shm_segsz = in->shm_segsz;
1402	v.shm_nattch = in->shm_nattch;
1403	v.shm_cpid = in->shm_cpid;
1404	v.shm_lpid = in->shm_lpid;
1405	return copy_to_user(to: buf, from: &v, n: sizeof(v));
1406	} else {
1407	struct compat_shmid_ds v;
1408	memset(&v, 0, sizeof(v));
1409	to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1410	v.shm_perm.key = in->shm_perm.key;
1411	v.shm_atime = in->shm_atime;
1412	v.shm_dtime = in->shm_dtime;
1413	v.shm_ctime = in->shm_ctime;
1414	v.shm_segsz = in->shm_segsz;
1415	v.shm_nattch = in->shm_nattch;
1416	v.shm_cpid = in->shm_cpid;
1417	v.shm_lpid = in->shm_lpid;
1418	return copy_to_user(to: buf, from: &v, n: sizeof(v));
1419	}
1420	}
1421
1422	static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1423	int version)
1424	{
1425	memset(out, 0, sizeof(*out));
1426	if (version == IPC_64) {
1427	struct compat_shmid64_ds __user *p = buf;
1428	return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1429	} else {
1430	struct compat_shmid_ds __user *p = buf;
1431	return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1432	}
1433	}
1434
1435	static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1436	{
1437	struct ipc_namespace *ns;
1438	struct shmid64_ds sem64;
1439	int err;
1440
1441	ns = current->nsproxy->ipc_ns;
1442
1443	if (cmd < 0 || shmid < 0)
1444	return -EINVAL;
1445
1446	switch (cmd) {
1447	case IPC_INFO: {
1448	struct shminfo64 shminfo;
1449	err = shmctl_ipc_info(ns, shminfo: &shminfo);
1450	if (err < 0)
1451	return err;
1452	if (copy_compat_shminfo_to_user(buf: uptr, in: &shminfo, version))
1453	err = -EFAULT;
1454	return err;
1455	}
1456	case SHM_INFO: {
1457	struct shm_info shm_info;
1458	err = shmctl_shm_info(ns, shm_info: &shm_info);
1459	if (err < 0)
1460	return err;
1461	if (put_compat_shm_info(ip: &shm_info, uip: uptr))
1462	err = -EFAULT;
1463	return err;
1464	}
1465	case IPC_STAT:
1466	case SHM_STAT_ANY:
1467	case SHM_STAT:
1468	err = shmctl_stat(ns, shmid, cmd, tbuf: &sem64);
1469	if (err < 0)
1470	return err;
1471	if (copy_compat_shmid_to_user(buf: uptr, in: &sem64, version))
1472	err = -EFAULT;
1473	return err;
1474
1475	case IPC_SET:
1476	if (copy_compat_shmid_from_user(out: &sem64, buf: uptr, version))
1477	return -EFAULT;
1478	fallthrough;
1479	case IPC_RMID:
1480	return shmctl_down(ns, shmid, cmd, shmid64: &sem64);
1481	case SHM_LOCK:
1482	case SHM_UNLOCK:
1483	return shmctl_do_lock(ns, shmid, cmd);
1484	default:
1485	return -EINVAL;
1486	}
1487	return err;
1488	}
1489
1490	COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1491	{
1492	return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1493	}
1494
1495	#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1496	long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1497	{
1498	int version = compat_ipc_parse_version(cmd: &cmd);
1499
1500	return compat_ksys_shmctl(shmid, cmd, uptr, version);
1501	}
1502
1503	COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1504	{
1505	return compat_ksys_old_shmctl(shmid, cmd, uptr);
1506	}
1507	#endif
1508	#endif
1509
1510	/*
1511	* Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1512	*
1513	* NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1514	* "raddr" thing points to kernel space, and there has to be a wrapper around
1515	* this.
1516	*/
1517	long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1518	ulong *raddr, unsigned long shmlba)
1519	{
1520	struct shmid_kernel *shp;
1521	unsigned long addr = (unsigned long)shmaddr;
1522	unsigned long size;
1523	struct file *file, *base;
1524	int err;
1525	unsigned long flags = MAP_SHARED;
1526	unsigned long prot;
1527	int acc_mode;
1528	struct ipc_namespace *ns;
1529	struct shm_file_data *sfd;
1530	int f_flags;
1531	unsigned long populate = 0;
1532
1533	err = -EINVAL;
1534	if (shmid < 0)
1535	goto out;
1536
1537	if (addr) {
1538	if (addr & (shmlba - 1)) {
1539	if (shmflg & SHM_RND) {
1540	addr &= ~(shmlba - 1); /* round down */
1541
1542	/*
1543	* Ensure that the round-down is non-nil
1544	* when remapping. This can happen for
1545	* cases when addr < shmlba.
1546	*/
1547	if (!addr && (shmflg & SHM_REMAP))
1548	goto out;
1549	} else
1550	#ifndef __ARCH_FORCE_SHMLBA
1551	if (addr & ~PAGE_MASK)
1552	#endif
1553	goto out;
1554	}
1555
1556	flags |= MAP_FIXED;
1557	} else if ((shmflg & SHM_REMAP))
1558	goto out;
1559
1560	if (shmflg & SHM_RDONLY) {
1561	prot = PROT_READ;
1562	acc_mode = S_IRUGO;
1563	f_flags = O_RDONLY;
1564	} else {
1565	prot = PROT_READ | PROT_WRITE;
1566	acc_mode = S_IRUGO | S_IWUGO;
1567	f_flags = O_RDWR;
1568	}
1569	if (shmflg & SHM_EXEC) {
1570	prot |= PROT_EXEC;
1571	acc_mode |= S_IXUGO;
1572	}
1573
1574	/*
1575	* We cannot rely on the fs check since SYSV IPC does have an
1576	* additional creator id...
1577	*/
1578	ns = current->nsproxy->ipc_ns;
1579	rcu_read_lock();
1580	shp = shm_obtain_object_check(ns, id: shmid);
1581	if (IS_ERR(ptr: shp)) {
1582	err = PTR_ERR(ptr: shp);
1583	goto out_unlock;
1584	}
1585
1586	err = -EACCES;
1587	if (ipcperms(ns, ipcp: &shp->shm_perm, flg: acc_mode))
1588	goto out_unlock;
1589
1590	err = security_shm_shmat(shp: &shp->shm_perm, shmaddr, shmflg);
1591	if (err)
1592	goto out_unlock;
1593
1594	ipc_lock_object(perm: &shp->shm_perm);
1595
1596	/* check if shm_destroy() is tearing down shp */
1597	if (!ipc_valid_object(perm: &shp->shm_perm)) {
1598	ipc_unlock_object(perm: &shp->shm_perm);
1599	err = -EIDRM;
1600	goto out_unlock;
1601	}
1602
1603	/*
1604	* We need to take a reference to the real shm file to prevent the
1605	* pointer from becoming stale in cases where the lifetime of the outer
1606	* file extends beyond that of the shm segment. It's not usually
1607	* possible, but it can happen during remap_file_pages() emulation as
1608	* that unmaps the memory, then does ->mmap() via file reference only.
1609	* We'll deny the ->mmap() if the shm segment was since removed, but to
1610	* detect shm ID reuse we need to compare the file pointers.
1611	*/
1612	base = get_file(f: shp->shm_file);
1613	shp->shm_nattch++;
1614	size = i_size_read(inode: file_inode(f: base));
1615	ipc_unlock_object(perm: &shp->shm_perm);
1616	rcu_read_unlock();
1617
1618	err = -ENOMEM;
1619	sfd = kzalloc(size: sizeof(*sfd), GFP_KERNEL);
1620	if (!sfd) {
1621	fput(base);
1622	goto out_nattch;
1623	}
1624
1625	file = alloc_file_clone(base, flags: f_flags,
1626	is_file_hugepages(file: base) ?
1627	&shm_file_operations_huge :
1628	&shm_file_operations);
1629	err = PTR_ERR(ptr: file);
1630	if (IS_ERR(ptr: file)) {
1631	kfree(objp: sfd);
1632	fput(base);
1633	goto out_nattch;
1634	}
1635
1636	sfd->id = shp->shm_perm.id;
1637	sfd->ns = get_ipc_ns(ns);
1638	sfd->file = base;
1639	sfd->vm_ops = NULL;
1640	file->private_data = sfd;
1641
1642	err = security_mmap_file(file, prot, flags);
1643	if (err)
1644	goto out_fput;
1645
1646	if (mmap_write_lock_killable(current->mm)) {
1647	err = -EINTR;
1648	goto out_fput;
1649	}
1650
1651	if (addr && !(shmflg & SHM_REMAP)) {
1652	err = -EINVAL;
1653	if (addr + size < addr)
1654	goto invalid;
1655
1656	if (find_vma_intersection(current->mm, start_addr: addr, end_addr: addr + size))
1657	goto invalid;
1658	}
1659
1660	addr = do_mmap(file, addr, len: size, prot, flags, vm_flags: 0, pgoff: 0, populate: &populate, NULL);
1661	*raddr = addr;
1662	err = 0;
1663	if (IS_ERR_VALUE(addr))
1664	err = (long)addr;
1665	invalid:
1666	mmap_write_unlock(current->mm);
1667	if (populate)
1668	mm_populate(addr, len: populate);
1669
1670	out_fput:
1671	fput(file);
1672
1673	out_nattch:
1674	down_write(sem: &shm_ids(ns).rwsem);
1675	shp = shm_lock(ns, id: shmid);
1676	shp->shm_nattch--;
1677
1678	if (shm_may_destroy(shp))
1679	shm_destroy(ns, shp);
1680	else
1681	shm_unlock(shp);
1682	up_write(sem: &shm_ids(ns).rwsem);
1683	return err;
1684
1685	out_unlock:
1686	rcu_read_unlock();
1687	out:
1688	return err;
1689	}
1690
1691	SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1692	{
1693	unsigned long ret;
1694	long err;
1695
1696	err = do_shmat(shmid, shmaddr, shmflg, raddr: &ret, SHMLBA);
1697	if (err)
1698	return err;
1699	force_successful_syscall_return();
1700	return (long)ret;
1701	}
1702
1703	#ifdef CONFIG_COMPAT
1704
1705	#ifndef COMPAT_SHMLBA
1706	#define COMPAT_SHMLBA SHMLBA
1707	#endif
1708
1709	COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1710	{
1711	unsigned long ret;
1712	long err;
1713
1714	err = do_shmat(shmid, shmaddr: compat_ptr(uptr: shmaddr), shmflg, raddr: &ret, COMPAT_SHMLBA);
1715	if (err)
1716	return err;
1717	force_successful_syscall_return();
1718	return (long)ret;
1719	}
1720	#endif
1721
1722	/*
1723	* detach and kill segment if marked destroyed.
1724	* The work is done in shm_close.
1725	*/
1726	long ksys_shmdt(char __user *shmaddr)
1727	{
1728	struct mm_struct *mm = current->mm;
1729	struct vm_area_struct *vma;
1730	unsigned long addr = (unsigned long)shmaddr;
1731	int retval = -EINVAL;
1732	#ifdef CONFIG_MMU
1733	loff_t size = 0;
1734	struct file *file;
1735	VMA_ITERATOR(vmi, mm, addr);
1736	#endif
1737
1738	if (addr & ~PAGE_MASK)
1739	return retval;
1740
1741	if (mmap_write_lock_killable(mm))
1742	return -EINTR;
1743
1744	/*
1745	* This function tries to be smart and unmap shm segments that
1746	* were modified by partial mlock or munmap calls:
1747	* - It first determines the size of the shm segment that should be
1748	* unmapped: It searches for a vma that is backed by shm and that
1749	* started at address shmaddr. It records it's size and then unmaps
1750	* it.
1751	* - Then it unmaps all shm vmas that started at shmaddr and that
1752	* are within the initially determined size and that are from the
1753	* same shm segment from which we determined the size.
1754	* Errors from do_munmap are ignored: the function only fails if
1755	* it's called with invalid parameters or if it's called to unmap
1756	* a part of a vma. Both calls in this function are for full vmas,
1757	* the parameters are directly copied from the vma itself and always
1758	* valid - therefore do_munmap cannot fail. (famous last words?)
1759	*/
1760	/*
1761	* If it had been mremap()'d, the starting address would not
1762	* match the usual checks anyway. So assume all vma's are
1763	* above the starting address given.
1764	*/
1765
1766	#ifdef CONFIG_MMU
1767	for_each_vma(vmi, vma) {
1768	/*
1769	* Check if the starting address would match, i.e. it's
1770	* a fragment created by mprotect() and/or munmap(), or it
1771	* otherwise it starts at this address with no hassles.
1772	*/
1773	if ((vma->vm_ops == &shm_vm_ops) &&
1774	(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1775
1776	/*
1777	* Record the file of the shm segment being
1778	* unmapped. With mremap(), someone could place
1779	* page from another segment but with equal offsets
1780	* in the range we are unmapping.
1781	*/
1782	file = vma->vm_file;
1783	size = i_size_read(inode: file_inode(f: vma->vm_file));
1784	do_vma_munmap(vmi: &vmi, vma, start: vma->vm_start, end: vma->vm_end,
1785	NULL, unlock: false);
1786	/*
1787	* We discovered the size of the shm segment, so
1788	* break out of here and fall through to the next
1789	* loop that uses the size information to stop
1790	* searching for matching vma's.
1791	*/
1792	retval = 0;
1793	vma = vma_next(vmi: &vmi);
1794	break;
1795	}
1796	}
1797
1798	/*
1799	* We need look no further than the maximum address a fragment
1800	* could possibly have landed at. Also cast things to loff_t to
1801	* prevent overflows and make comparisons vs. equal-width types.
1802	*/
1803	size = PAGE_ALIGN(size);
1804	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1805	/* finding a matching vma now does not alter retval */
1806	if ((vma->vm_ops == &shm_vm_ops) &&
1807	((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1808	(vma->vm_file == file)) {
1809	do_vma_munmap(vmi: &vmi, vma, start: vma->vm_start, end: vma->vm_end,
1810	NULL, unlock: false);
1811	}
1812
1813	vma = vma_next(vmi: &vmi);
1814	}
1815
1816	#else /* CONFIG_MMU */
1817	vma = vma_lookup(mm, addr);
1818	/* under NOMMU conditions, the exact address to be destroyed must be
1819	* given
1820	*/
1821	if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1822	do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1823	retval = 0;
1824	}
1825
1826	#endif
1827
1828	mmap_write_unlock(mm);
1829	return retval;
1830	}
1831
1832	SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1833	{
1834	return ksys_shmdt(shmaddr);
1835	}
1836
1837	#ifdef CONFIG_PROC_FS
1838	static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1839	{
1840	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1841	struct user_namespace *user_ns = seq_user_ns(seq: s);
1842	struct kern_ipc_perm *ipcp = it;
1843	struct shmid_kernel *shp;
1844	unsigned long rss = 0, swp = 0;
1845
1846	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1847	shm_add_rss_swap(shp, rss_add: &rss, swp_add: &swp);
1848
1849	#if BITS_PER_LONG <= 32
1850	#define SIZE_SPEC "%10lu"
1851	#else
1852	#define SIZE_SPEC "%21lu"
1853	#endif
1854
1855	seq_printf(m: s,
1856	fmt: "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1857	"%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1858	SIZE_SPEC " " SIZE_SPEC "\n",
1859	shp->shm_perm.key,
1860	shp->shm_perm.id,
1861	shp->shm_perm.mode,
1862	shp->shm_segsz,
1863	pid_nr_ns(pid: shp->shm_cprid, ns: pid_ns),
1864	pid_nr_ns(pid: shp->shm_lprid, ns: pid_ns),
1865	shp->shm_nattch,
1866	from_kuid_munged(to: user_ns, uid: shp->shm_perm.uid),
1867	from_kgid_munged(to: user_ns, gid: shp->shm_perm.gid),
1868	from_kuid_munged(to: user_ns, uid: shp->shm_perm.cuid),
1869	from_kgid_munged(to: user_ns, gid: shp->shm_perm.cgid),
1870	shp->shm_atim,
1871	shp->shm_dtim,
1872	shp->shm_ctim,
1873	rss * PAGE_SIZE,
1874	swp * PAGE_SIZE);
1875
1876	return 0;
1877	}
1878	#endif
1879