shm.c revision 45e55300
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
52struct 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 user_struct	*mlock_user;
64
65	/* The task created the shm object.  NULL if the task is dead. */
66	struct task_struct	*shm_creator;
67	struct list_head	shm_clist;	/* list by creator */
68} __randomize_layout;
69
70/* shm_mode upper byte flags */
71#define SHM_DEST	01000	/* segment will be destroyed on last detach */
72#define SHM_LOCKED	02000   /* segment will not be swapped */
73
74struct shm_file_data {
75	int id;
76	struct ipc_namespace *ns;
77	struct file *file;
78	const struct vm_operations_struct *vm_ops;
79};
80
81#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
82
83static const struct file_operations shm_file_operations;
84static const struct vm_operations_struct shm_vm_ops;
85
86#define shm_ids(ns)	((ns)->ids[IPC_SHM_IDS])
87
88#define shm_unlock(shp)			\
89	ipc_unlock(&(shp)->shm_perm)
90
91static int newseg(struct ipc_namespace *, struct ipc_params *);
92static void shm_open(struct vm_area_struct *vma);
93static void shm_close(struct vm_area_struct *vma);
94static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
95#ifdef CONFIG_PROC_FS
96static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
97#endif
98
99void shm_init_ns(struct ipc_namespace *ns)
100{
101	ns->shm_ctlmax = SHMMAX;
102	ns->shm_ctlall = SHMALL;
103	ns->shm_ctlmni = SHMMNI;
104	ns->shm_rmid_forced = 0;
105	ns->shm_tot = 0;
106	ipc_init_ids(&shm_ids(ns));
107}
108
109/*
110 * Called with shm_ids.rwsem (writer) and the shp structure locked.
111 * Only shm_ids.rwsem remains locked on exit.
112 */
113static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
114{
115	struct shmid_kernel *shp;
116
117	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
118
119	if (shp->shm_nattch) {
120		shp->shm_perm.mode |= SHM_DEST;
121		/* Do not find it any more */
122		ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
123		shm_unlock(shp);
124	} else
125		shm_destroy(ns, shp);
126}
127
128#ifdef CONFIG_IPC_NS
129void shm_exit_ns(struct ipc_namespace *ns)
130{
131	free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
132	idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
133	rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
134}
135#endif
136
137static int __init ipc_ns_init(void)
138{
139	shm_init_ns(&init_ipc_ns);
140	return 0;
141}
142
143pure_initcall(ipc_ns_init);
144
145void __init shm_init(void)
146{
147	ipc_init_proc_interface("sysvipc/shm",
148#if BITS_PER_LONG <= 32
149				"       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
150#else
151				"       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
152#endif
153				IPC_SHM_IDS, sysvipc_shm_proc_show);
154}
155
156static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
157{
158	struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
159
160	if (IS_ERR(ipcp))
161		return ERR_CAST(ipcp);
162
163	return container_of(ipcp, struct shmid_kernel, shm_perm);
164}
165
166static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
167{
168	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
169
170	if (IS_ERR(ipcp))
171		return ERR_CAST(ipcp);
172
173	return container_of(ipcp, struct shmid_kernel, shm_perm);
174}
175
176/*
177 * shm_lock_(check_) routines are called in the paths where the rwsem
178 * is not necessarily held.
179 */
180static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
181{
182	struct kern_ipc_perm *ipcp;
183
184	rcu_read_lock();
185	ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
186	if (IS_ERR(ipcp))
187		goto err;
188
189	ipc_lock_object(ipcp);
190	/*
191	 * ipc_rmid() may have already freed the ID while ipc_lock_object()
192	 * was spinning: here verify that the structure is still valid.
193	 * Upon races with RMID, return -EIDRM, thus indicating that
194	 * the ID points to a removed identifier.
195	 */
196	if (ipc_valid_object(ipcp)) {
197		/* return a locked ipc object upon success */
198		return container_of(ipcp, struct shmid_kernel, shm_perm);
199	}
200
201	ipc_unlock_object(ipcp);
202	ipcp = ERR_PTR(-EIDRM);
203err:
204	rcu_read_unlock();
205	/*
206	 * Callers of shm_lock() must validate the status of the returned ipc
207	 * object pointer and error out as appropriate.
208	 */
209	return ERR_CAST(ipcp);
210}
211
212static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
213{
214	rcu_read_lock();
215	ipc_lock_object(&ipcp->shm_perm);
216}
217
218static void shm_rcu_free(struct rcu_head *head)
219{
220	struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
221							rcu);
222	struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
223							shm_perm);
224	security_shm_free(&shp->shm_perm);
225	kvfree(shp);
226}
227
228static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
229{
230	list_del(&s->shm_clist);
231	ipc_rmid(&shm_ids(ns), &s->shm_perm);
232}
233
234
235static int __shm_open(struct vm_area_struct *vma)
236{
237	struct file *file = vma->vm_file;
238	struct shm_file_data *sfd = shm_file_data(file);
239	struct shmid_kernel *shp;
240
241	shp = shm_lock(sfd->ns, sfd->id);
242
243	if (IS_ERR(shp))
244		return PTR_ERR(shp);
245
246	if (shp->shm_file != sfd->file) {
247		/* ID was reused */
248		shm_unlock(shp);
249		return -EINVAL;
250	}
251
252	shp->shm_atim = ktime_get_real_seconds();
253	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
254	shp->shm_nattch++;
255	shm_unlock(shp);
256	return 0;
257}
258
259/* This is called by fork, once for every shm attach. */
260static void shm_open(struct vm_area_struct *vma)
261{
262	int err = __shm_open(vma);
263	/*
264	 * We raced in the idr lookup or with shm_destroy().
265	 * Either way, the ID is busted.
266	 */
267	WARN_ON_ONCE(err);
268}
269
270/*
271 * shm_destroy - free the struct shmid_kernel
272 *
273 * @ns: namespace
274 * @shp: struct to free
275 *
276 * It has to be called with shp and shm_ids.rwsem (writer) locked,
277 * but returns with shp unlocked and freed.
278 */
279static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
280{
281	struct file *shm_file;
282
283	shm_file = shp->shm_file;
284	shp->shm_file = NULL;
285	ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
286	shm_rmid(ns, shp);
287	shm_unlock(shp);
288	if (!is_file_hugepages(shm_file))
289		shmem_lock(shm_file, 0, shp->mlock_user);
290	else if (shp->mlock_user)
291		user_shm_unlock(i_size_read(file_inode(shm_file)),
292				shp->mlock_user);
293	fput(shm_file);
294	ipc_update_pid(&shp->shm_cprid, NULL);
295	ipc_update_pid(&shp->shm_lprid, NULL);
296	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
297}
298
299/*
300 * shm_may_destroy - identifies whether shm segment should be destroyed now
301 *
302 * Returns true if and only if there are no active users of the segment and
303 * one of the following is true:
304 *
305 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
306 *
307 * 2) sysctl kernel.shm_rmid_forced is set to 1.
308 */
309static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
310{
311	return (shp->shm_nattch == 0) &&
312	       (ns->shm_rmid_forced ||
313		(shp->shm_perm.mode & SHM_DEST));
314}
315
316/*
317 * remove the attach descriptor vma.
318 * free memory for segment if it is marked destroyed.
319 * The descriptor has already been removed from the current->mm->mmap list
320 * and will later be kfree()d.
321 */
322static void shm_close(struct vm_area_struct *vma)
323{
324	struct file *file = vma->vm_file;
325	struct shm_file_data *sfd = shm_file_data(file);
326	struct shmid_kernel *shp;
327	struct ipc_namespace *ns = sfd->ns;
328
329	down_write(&shm_ids(ns).rwsem);
330	/* remove from the list of attaches of the shm segment */
331	shp = shm_lock(ns, sfd->id);
332
333	/*
334	 * We raced in the idr lookup or with shm_destroy().
335	 * Either way, the ID is busted.
336	 */
337	if (WARN_ON_ONCE(IS_ERR(shp)))
338		goto done; /* no-op */
339
340	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
341	shp->shm_dtim = ktime_get_real_seconds();
342	shp->shm_nattch--;
343	if (shm_may_destroy(ns, shp))
344		shm_destroy(ns, shp);
345	else
346		shm_unlock(shp);
347done:
348	up_write(&shm_ids(ns).rwsem);
349}
350
351/* Called with ns->shm_ids(ns).rwsem locked */
352static int shm_try_destroy_orphaned(int id, void *p, void *data)
353{
354	struct ipc_namespace *ns = data;
355	struct kern_ipc_perm *ipcp = p;
356	struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
357
358	/*
359	 * We want to destroy segments without users and with already
360	 * exit'ed originating process.
361	 *
362	 * As shp->* are changed under rwsem, it's safe to skip shp locking.
363	 */
364	if (shp->shm_creator != NULL)
365		return 0;
366
367	if (shm_may_destroy(ns, shp)) {
368		shm_lock_by_ptr(shp);
369		shm_destroy(ns, shp);
370	}
371	return 0;
372}
373
374void shm_destroy_orphaned(struct ipc_namespace *ns)
375{
376	down_write(&shm_ids(ns).rwsem);
377	if (shm_ids(ns).in_use)
378		idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
379	up_write(&shm_ids(ns).rwsem);
380}
381
382/* Locking assumes this will only be called with task == current */
383void exit_shm(struct task_struct *task)
384{
385	struct ipc_namespace *ns = task->nsproxy->ipc_ns;
386	struct shmid_kernel *shp, *n;
387
388	if (list_empty(&task->sysvshm.shm_clist))
389		return;
390
391	/*
392	 * If kernel.shm_rmid_forced is not set then only keep track of
393	 * which shmids are orphaned, so that a later set of the sysctl
394	 * can clean them up.
395	 */
396	if (!ns->shm_rmid_forced) {
397		down_read(&shm_ids(ns).rwsem);
398		list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
399			shp->shm_creator = NULL;
400		/*
401		 * Only under read lock but we are only called on current
402		 * so no entry on the list will be shared.
403		 */
404		list_del(&task->sysvshm.shm_clist);
405		up_read(&shm_ids(ns).rwsem);
406		return;
407	}
408
409	/*
410	 * Destroy all already created segments, that were not yet mapped,
411	 * and mark any mapped as orphan to cover the sysctl toggling.
412	 * Destroy is skipped if shm_may_destroy() returns false.
413	 */
414	down_write(&shm_ids(ns).rwsem);
415	list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
416		shp->shm_creator = NULL;
417
418		if (shm_may_destroy(ns, shp)) {
419			shm_lock_by_ptr(shp);
420			shm_destroy(ns, shp);
421		}
422	}
423
424	/* Remove the list head from any segments still attached. */
425	list_del(&task->sysvshm.shm_clist);
426	up_write(&shm_ids(ns).rwsem);
427}
428
429static vm_fault_t shm_fault(struct vm_fault *vmf)
430{
431	struct file *file = vmf->vma->vm_file;
432	struct shm_file_data *sfd = shm_file_data(file);
433
434	return sfd->vm_ops->fault(vmf);
435}
436
437static int shm_split(struct vm_area_struct *vma, unsigned long addr)
438{
439	struct file *file = vma->vm_file;
440	struct shm_file_data *sfd = shm_file_data(file);
441
442	if (sfd->vm_ops->split)
443		return sfd->vm_ops->split(vma, addr);
444
445	return 0;
446}
447
448static unsigned long shm_pagesize(struct vm_area_struct *vma)
449{
450	struct file *file = vma->vm_file;
451	struct shm_file_data *sfd = shm_file_data(file);
452
453	if (sfd->vm_ops->pagesize)
454		return sfd->vm_ops->pagesize(vma);
455
456	return PAGE_SIZE;
457}
458
459#ifdef CONFIG_NUMA
460static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
461{
462	struct file *file = vma->vm_file;
463	struct shm_file_data *sfd = shm_file_data(file);
464	int err = 0;
465
466	if (sfd->vm_ops->set_policy)
467		err = sfd->vm_ops->set_policy(vma, new);
468	return err;
469}
470
471static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
472					unsigned long addr)
473{
474	struct file *file = vma->vm_file;
475	struct shm_file_data *sfd = shm_file_data(file);
476	struct mempolicy *pol = NULL;
477
478	if (sfd->vm_ops->get_policy)
479		pol = sfd->vm_ops->get_policy(vma, addr);
480	else if (vma->vm_policy)
481		pol = vma->vm_policy;
482
483	return pol;
484}
485#endif
486
487static int shm_mmap(struct file *file, struct vm_area_struct *vma)
488{
489	struct shm_file_data *sfd = shm_file_data(file);
490	int ret;
491
492	/*
493	 * In case of remap_file_pages() emulation, the file can represent an
494	 * IPC ID that was removed, and possibly even reused by another shm
495	 * segment already.  Propagate this case as an error to caller.
496	 */
497	ret = __shm_open(vma);
498	if (ret)
499		return ret;
500
501	ret = call_mmap(sfd->file, vma);
502	if (ret) {
503		shm_close(vma);
504		return ret;
505	}
506	sfd->vm_ops = vma->vm_ops;
507#ifdef CONFIG_MMU
508	WARN_ON(!sfd->vm_ops->fault);
509#endif
510	vma->vm_ops = &shm_vm_ops;
511	return 0;
512}
513
514static int shm_release(struct inode *ino, struct file *file)
515{
516	struct shm_file_data *sfd = shm_file_data(file);
517
518	put_ipc_ns(sfd->ns);
519	fput(sfd->file);
520	shm_file_data(file) = NULL;
521	kfree(sfd);
522	return 0;
523}
524
525static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
526{
527	struct shm_file_data *sfd = shm_file_data(file);
528
529	if (!sfd->file->f_op->fsync)
530		return -EINVAL;
531	return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
532}
533
534static long shm_fallocate(struct file *file, int mode, loff_t offset,
535			  loff_t len)
536{
537	struct shm_file_data *sfd = shm_file_data(file);
538
539	if (!sfd->file->f_op->fallocate)
540		return -EOPNOTSUPP;
541	return sfd->file->f_op->fallocate(file, mode, offset, len);
542}
543
544static unsigned long shm_get_unmapped_area(struct file *file,
545	unsigned long addr, unsigned long len, unsigned long pgoff,
546	unsigned long flags)
547{
548	struct shm_file_data *sfd = shm_file_data(file);
549
550	return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
551						pgoff, flags);
552}
553
554static const struct file_operations shm_file_operations = {
555	.mmap		= shm_mmap,
556	.fsync		= shm_fsync,
557	.release	= shm_release,
558	.get_unmapped_area	= shm_get_unmapped_area,
559	.llseek		= noop_llseek,
560	.fallocate	= shm_fallocate,
561};
562
563/*
564 * shm_file_operations_huge is now identical to shm_file_operations,
565 * but we keep it distinct for the sake of is_file_shm_hugepages().
566 */
567static const struct file_operations shm_file_operations_huge = {
568	.mmap		= shm_mmap,
569	.fsync		= shm_fsync,
570	.release	= shm_release,
571	.get_unmapped_area	= shm_get_unmapped_area,
572	.llseek		= noop_llseek,
573	.fallocate	= shm_fallocate,
574};
575
576bool is_file_shm_hugepages(struct file *file)
577{
578	return file->f_op == &shm_file_operations_huge;
579}
580
581static const struct vm_operations_struct shm_vm_ops = {
582	.open	= shm_open,	/* callback for a new vm-area open */
583	.close	= shm_close,	/* callback for when the vm-area is released */
584	.fault	= shm_fault,
585	.split	= shm_split,
586	.pagesize = shm_pagesize,
587#if defined(CONFIG_NUMA)
588	.set_policy = shm_set_policy,
589	.get_policy = shm_get_policy,
590#endif
591};
592
593/**
594 * newseg - Create a new shared memory segment
595 * @ns: namespace
596 * @params: ptr to the structure that contains key, size and shmflg
597 *
598 * Called with shm_ids.rwsem held as a writer.
599 */
600static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
601{
602	key_t key = params->key;
603	int shmflg = params->flg;
604	size_t size = params->u.size;
605	int error;
606	struct shmid_kernel *shp;
607	size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
608	struct file *file;
609	char name[13];
610	vm_flags_t acctflag = 0;
611
612	if (size < SHMMIN || size > ns->shm_ctlmax)
613		return -EINVAL;
614
615	if (numpages << PAGE_SHIFT < size)
616		return -ENOSPC;
617
618	if (ns->shm_tot + numpages < ns->shm_tot ||
619			ns->shm_tot + numpages > ns->shm_ctlall)
620		return -ENOSPC;
621
622	shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
623	if (unlikely(!shp))
624		return -ENOMEM;
625
626	shp->shm_perm.key = key;
627	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
628	shp->mlock_user = NULL;
629
630	shp->shm_perm.security = NULL;
631	error = security_shm_alloc(&shp->shm_perm);
632	if (error) {
633		kvfree(shp);
634		return error;
635	}
636
637	sprintf(name, "SYSV%08x", key);
638	if (shmflg & SHM_HUGETLB) {
639		struct hstate *hs;
640		size_t hugesize;
641
642		hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
643		if (!hs) {
644			error = -EINVAL;
645			goto no_file;
646		}
647		hugesize = ALIGN(size, huge_page_size(hs));
648
649		/* hugetlb_file_setup applies strict accounting */
650		if (shmflg & SHM_NORESERVE)
651			acctflag = VM_NORESERVE;
652		file = hugetlb_file_setup(name, hugesize, acctflag,
653				  &shp->mlock_user, HUGETLB_SHMFS_INODE,
654				(shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
655	} else {
656		/*
657		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
658		 * if it's asked for.
659		 */
660		if  ((shmflg & SHM_NORESERVE) &&
661				sysctl_overcommit_memory != OVERCOMMIT_NEVER)
662			acctflag = VM_NORESERVE;
663		file = shmem_kernel_file_setup(name, size, acctflag);
664	}
665	error = PTR_ERR(file);
666	if (IS_ERR(file))
667		goto no_file;
668
669	shp->shm_cprid = get_pid(task_tgid(current));
670	shp->shm_lprid = NULL;
671	shp->shm_atim = shp->shm_dtim = 0;
672	shp->shm_ctim = ktime_get_real_seconds();
673	shp->shm_segsz = size;
674	shp->shm_nattch = 0;
675	shp->shm_file = file;
676	shp->shm_creator = current;
677
678	/* ipc_addid() locks shp upon success. */
679	error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
680	if (error < 0)
681		goto no_id;
682
683	list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
684
685	/*
686	 * shmid gets reported as "inode#" in /proc/pid/maps.
687	 * proc-ps tools use this. Changing this will break them.
688	 */
689	file_inode(file)->i_ino = shp->shm_perm.id;
690
691	ns->shm_tot += numpages;
692	error = shp->shm_perm.id;
693
694	ipc_unlock_object(&shp->shm_perm);
695	rcu_read_unlock();
696	return error;
697
698no_id:
699	ipc_update_pid(&shp->shm_cprid, NULL);
700	ipc_update_pid(&shp->shm_lprid, NULL);
701	if (is_file_hugepages(file) && shp->mlock_user)
702		user_shm_unlock(size, shp->mlock_user);
703	fput(file);
704	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
705	return error;
706no_file:
707	call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
708	return error;
709}
710
711/*
712 * Called with shm_ids.rwsem and ipcp locked.
713 */
714static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
715				struct ipc_params *params)
716{
717	struct shmid_kernel *shp;
718
719	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
720	if (shp->shm_segsz < params->u.size)
721		return -EINVAL;
722
723	return 0;
724}
725
726long ksys_shmget(key_t key, size_t size, int shmflg)
727{
728	struct ipc_namespace *ns;
729	static const struct ipc_ops shm_ops = {
730		.getnew = newseg,
731		.associate = security_shm_associate,
732		.more_checks = shm_more_checks,
733	};
734	struct ipc_params shm_params;
735
736	ns = current->nsproxy->ipc_ns;
737
738	shm_params.key = key;
739	shm_params.flg = shmflg;
740	shm_params.u.size = size;
741
742	return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
743}
744
745SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
746{
747	return ksys_shmget(key, size, shmflg);
748}
749
750static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
751{
752	switch (version) {
753	case IPC_64:
754		return copy_to_user(buf, in, sizeof(*in));
755	case IPC_OLD:
756	    {
757		struct shmid_ds out;
758
759		memset(&out, 0, sizeof(out));
760		ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
761		out.shm_segsz	= in->shm_segsz;
762		out.shm_atime	= in->shm_atime;
763		out.shm_dtime	= in->shm_dtime;
764		out.shm_ctime	= in->shm_ctime;
765		out.shm_cpid	= in->shm_cpid;
766		out.shm_lpid	= in->shm_lpid;
767		out.shm_nattch	= in->shm_nattch;
768
769		return copy_to_user(buf, &out, sizeof(out));
770	    }
771	default:
772		return -EINVAL;
773	}
774}
775
776static inline unsigned long
777copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
778{
779	switch (version) {
780	case IPC_64:
781		if (copy_from_user(out, buf, sizeof(*out)))
782			return -EFAULT;
783		return 0;
784	case IPC_OLD:
785	    {
786		struct shmid_ds tbuf_old;
787
788		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
789			return -EFAULT;
790
791		out->shm_perm.uid	= tbuf_old.shm_perm.uid;
792		out->shm_perm.gid	= tbuf_old.shm_perm.gid;
793		out->shm_perm.mode	= tbuf_old.shm_perm.mode;
794
795		return 0;
796	    }
797	default:
798		return -EINVAL;
799	}
800}
801
802static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
803{
804	switch (version) {
805	case IPC_64:
806		return copy_to_user(buf, in, sizeof(*in));
807	case IPC_OLD:
808	    {
809		struct shminfo out;
810
811		if (in->shmmax > INT_MAX)
812			out.shmmax = INT_MAX;
813		else
814			out.shmmax = (int)in->shmmax;
815
816		out.shmmin	= in->shmmin;
817		out.shmmni	= in->shmmni;
818		out.shmseg	= in->shmseg;
819		out.shmall	= in->shmall;
820
821		return copy_to_user(buf, &out, sizeof(out));
822	    }
823	default:
824		return -EINVAL;
825	}
826}
827
828/*
829 * Calculate and add used RSS and swap pages of a shm.
830 * Called with shm_ids.rwsem held as a reader
831 */
832static void shm_add_rss_swap(struct shmid_kernel *shp,
833	unsigned long *rss_add, unsigned long *swp_add)
834{
835	struct inode *inode;
836
837	inode = file_inode(shp->shm_file);
838
839	if (is_file_hugepages(shp->shm_file)) {
840		struct address_space *mapping = inode->i_mapping;
841		struct hstate *h = hstate_file(shp->shm_file);
842		*rss_add += pages_per_huge_page(h) * mapping->nrpages;
843	} else {
844#ifdef CONFIG_SHMEM
845		struct shmem_inode_info *info = SHMEM_I(inode);
846
847		spin_lock_irq(&info->lock);
848		*rss_add += inode->i_mapping->nrpages;
849		*swp_add += info->swapped;
850		spin_unlock_irq(&info->lock);
851#else
852		*rss_add += inode->i_mapping->nrpages;
853#endif
854	}
855}
856
857/*
858 * Called with shm_ids.rwsem held as a reader
859 */
860static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
861		unsigned long *swp)
862{
863	int next_id;
864	int total, in_use;
865
866	*rss = 0;
867	*swp = 0;
868
869	in_use = shm_ids(ns).in_use;
870
871	for (total = 0, next_id = 0; total < in_use; next_id++) {
872		struct kern_ipc_perm *ipc;
873		struct shmid_kernel *shp;
874
875		ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
876		if (ipc == NULL)
877			continue;
878		shp = container_of(ipc, struct shmid_kernel, shm_perm);
879
880		shm_add_rss_swap(shp, rss, swp);
881
882		total++;
883	}
884}
885
886/*
887 * This function handles some shmctl commands which require the rwsem
888 * to be held in write mode.
889 * NOTE: no locks must be held, the rwsem is taken inside this function.
890 */
891static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
892		       struct shmid64_ds *shmid64)
893{
894	struct kern_ipc_perm *ipcp;
895	struct shmid_kernel *shp;
896	int err;
897
898	down_write(&shm_ids(ns).rwsem);
899	rcu_read_lock();
900
901	ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
902				      &shmid64->shm_perm, 0);
903	if (IS_ERR(ipcp)) {
904		err = PTR_ERR(ipcp);
905		goto out_unlock1;
906	}
907
908	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
909
910	err = security_shm_shmctl(&shp->shm_perm, cmd);
911	if (err)
912		goto out_unlock1;
913
914	switch (cmd) {
915	case IPC_RMID:
916		ipc_lock_object(&shp->shm_perm);
917		/* do_shm_rmid unlocks the ipc object and rcu */
918		do_shm_rmid(ns, ipcp);
919		goto out_up;
920	case IPC_SET:
921		ipc_lock_object(&shp->shm_perm);
922		err = ipc_update_perm(&shmid64->shm_perm, ipcp);
923		if (err)
924			goto out_unlock0;
925		shp->shm_ctim = ktime_get_real_seconds();
926		break;
927	default:
928		err = -EINVAL;
929		goto out_unlock1;
930	}
931
932out_unlock0:
933	ipc_unlock_object(&shp->shm_perm);
934out_unlock1:
935	rcu_read_unlock();
936out_up:
937	up_write(&shm_ids(ns).rwsem);
938	return err;
939}
940
941static int shmctl_ipc_info(struct ipc_namespace *ns,
942			   struct shminfo64 *shminfo)
943{
944	int err = security_shm_shmctl(NULL, IPC_INFO);
945	if (!err) {
946		memset(shminfo, 0, sizeof(*shminfo));
947		shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
948		shminfo->shmmax = ns->shm_ctlmax;
949		shminfo->shmall = ns->shm_ctlall;
950		shminfo->shmmin = SHMMIN;
951		down_read(&shm_ids(ns).rwsem);
952		err = ipc_get_maxidx(&shm_ids(ns));
953		up_read(&shm_ids(ns).rwsem);
954		if (err < 0)
955			err = 0;
956	}
957	return err;
958}
959
960static int shmctl_shm_info(struct ipc_namespace *ns,
961			   struct shm_info *shm_info)
962{
963	int err = security_shm_shmctl(NULL, SHM_INFO);
964	if (!err) {
965		memset(shm_info, 0, sizeof(*shm_info));
966		down_read(&shm_ids(ns).rwsem);
967		shm_info->used_ids = shm_ids(ns).in_use;
968		shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
969		shm_info->shm_tot = ns->shm_tot;
970		shm_info->swap_attempts = 0;
971		shm_info->swap_successes = 0;
972		err = ipc_get_maxidx(&shm_ids(ns));
973		up_read(&shm_ids(ns).rwsem);
974		if (err < 0)
975			err = 0;
976	}
977	return err;
978}
979
980static int shmctl_stat(struct ipc_namespace *ns, int shmid,
981			int cmd, struct shmid64_ds *tbuf)
982{
983	struct shmid_kernel *shp;
984	int err;
985
986	memset(tbuf, 0, sizeof(*tbuf));
987
988	rcu_read_lock();
989	if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
990		shp = shm_obtain_object(ns, shmid);
991		if (IS_ERR(shp)) {
992			err = PTR_ERR(shp);
993			goto out_unlock;
994		}
995	} else { /* IPC_STAT */
996		shp = shm_obtain_object_check(ns, shmid);
997		if (IS_ERR(shp)) {
998			err = PTR_ERR(shp);
999			goto out_unlock;
1000		}
1001	}
1002
1003	/*
1004	 * Semantically SHM_STAT_ANY ought to be identical to
1005	 * that functionality provided by the /proc/sysvipc/
1006	 * interface. As such, only audit these calls and
1007	 * do not do traditional S_IRUGO permission checks on
1008	 * the ipc object.
1009	 */
1010	if (cmd == SHM_STAT_ANY)
1011		audit_ipc_obj(&shp->shm_perm);
1012	else {
1013		err = -EACCES;
1014		if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1015			goto out_unlock;
1016	}
1017
1018	err = security_shm_shmctl(&shp->shm_perm, cmd);
1019	if (err)
1020		goto out_unlock;
1021
1022	ipc_lock_object(&shp->shm_perm);
1023
1024	if (!ipc_valid_object(&shp->shm_perm)) {
1025		ipc_unlock_object(&shp->shm_perm);
1026		err = -EIDRM;
1027		goto out_unlock;
1028	}
1029
1030	kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1031	tbuf->shm_segsz	= shp->shm_segsz;
1032	tbuf->shm_atime	= shp->shm_atim;
1033	tbuf->shm_dtime	= shp->shm_dtim;
1034	tbuf->shm_ctime	= shp->shm_ctim;
1035#ifndef CONFIG_64BIT
1036	tbuf->shm_atime_high = shp->shm_atim >> 32;
1037	tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1038	tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1039#endif
1040	tbuf->shm_cpid	= pid_vnr(shp->shm_cprid);
1041	tbuf->shm_lpid	= pid_vnr(shp->shm_lprid);
1042	tbuf->shm_nattch = shp->shm_nattch;
1043
1044	if (cmd == IPC_STAT) {
1045		/*
1046		 * As defined in SUS:
1047		 * Return 0 on success
1048		 */
1049		err = 0;
1050	} else {
1051		/*
1052		 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1053		 * Return the full id, including the sequence number
1054		 */
1055		err = shp->shm_perm.id;
1056	}
1057
1058	ipc_unlock_object(&shp->shm_perm);
1059out_unlock:
1060	rcu_read_unlock();
1061	return err;
1062}
1063
1064static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1065{
1066	struct shmid_kernel *shp;
1067	struct file *shm_file;
1068	int err;
1069
1070	rcu_read_lock();
1071	shp = shm_obtain_object_check(ns, shmid);
1072	if (IS_ERR(shp)) {
1073		err = PTR_ERR(shp);
1074		goto out_unlock1;
1075	}
1076
1077	audit_ipc_obj(&(shp->shm_perm));
1078	err = security_shm_shmctl(&shp->shm_perm, cmd);
1079	if (err)
1080		goto out_unlock1;
1081
1082	ipc_lock_object(&shp->shm_perm);
1083
1084	/* check if shm_destroy() is tearing down shp */
1085	if (!ipc_valid_object(&shp->shm_perm)) {
1086		err = -EIDRM;
1087		goto out_unlock0;
1088	}
1089
1090	if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1091		kuid_t euid = current_euid();
1092
1093		if (!uid_eq(euid, shp->shm_perm.uid) &&
1094		    !uid_eq(euid, shp->shm_perm.cuid)) {
1095			err = -EPERM;
1096			goto out_unlock0;
1097		}
1098		if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1099			err = -EPERM;
1100			goto out_unlock0;
1101		}
1102	}
1103
1104	shm_file = shp->shm_file;
1105	if (is_file_hugepages(shm_file))
1106		goto out_unlock0;
1107
1108	if (cmd == SHM_LOCK) {
1109		struct user_struct *user = current_user();
1110
1111		err = shmem_lock(shm_file, 1, user);
1112		if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1113			shp->shm_perm.mode |= SHM_LOCKED;
1114			shp->mlock_user = user;
1115		}
1116		goto out_unlock0;
1117	}
1118
1119	/* SHM_UNLOCK */
1120	if (!(shp->shm_perm.mode & SHM_LOCKED))
1121		goto out_unlock0;
1122	shmem_lock(shm_file, 0, shp->mlock_user);
1123	shp->shm_perm.mode &= ~SHM_LOCKED;
1124	shp->mlock_user = NULL;
1125	get_file(shm_file);
1126	ipc_unlock_object(&shp->shm_perm);
1127	rcu_read_unlock();
1128	shmem_unlock_mapping(shm_file->f_mapping);
1129
1130	fput(shm_file);
1131	return err;
1132
1133out_unlock0:
1134	ipc_unlock_object(&shp->shm_perm);
1135out_unlock1:
1136	rcu_read_unlock();
1137	return err;
1138}
1139
1140static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1141{
1142	int err;
1143	struct ipc_namespace *ns;
1144	struct shmid64_ds sem64;
1145
1146	if (cmd < 0 || shmid < 0)
1147		return -EINVAL;
1148
1149	ns = current->nsproxy->ipc_ns;
1150
1151	switch (cmd) {
1152	case IPC_INFO: {
1153		struct shminfo64 shminfo;
1154		err = shmctl_ipc_info(ns, &shminfo);
1155		if (err < 0)
1156			return err;
1157		if (copy_shminfo_to_user(buf, &shminfo, version))
1158			err = -EFAULT;
1159		return err;
1160	}
1161	case SHM_INFO: {
1162		struct shm_info shm_info;
1163		err = shmctl_shm_info(ns, &shm_info);
1164		if (err < 0)
1165			return err;
1166		if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1167			err = -EFAULT;
1168		return err;
1169	}
1170	case SHM_STAT:
1171	case SHM_STAT_ANY:
1172	case IPC_STAT: {
1173		err = shmctl_stat(ns, shmid, cmd, &sem64);
1174		if (err < 0)
1175			return err;
1176		if (copy_shmid_to_user(buf, &sem64, version))
1177			err = -EFAULT;
1178		return err;
1179	}
1180	case IPC_SET:
1181		if (copy_shmid_from_user(&sem64, buf, version))
1182			return -EFAULT;
1183		/* fallthru */
1184	case IPC_RMID:
1185		return shmctl_down(ns, shmid, cmd, &sem64);
1186	case SHM_LOCK:
1187	case SHM_UNLOCK:
1188		return shmctl_do_lock(ns, shmid, cmd);
1189	default:
1190		return -EINVAL;
1191	}
1192}
1193
1194SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1195{
1196	return ksys_shmctl(shmid, cmd, buf, IPC_64);
1197}
1198
1199#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1200long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1201{
1202	int version = ipc_parse_version(&cmd);
1203
1204	return ksys_shmctl(shmid, cmd, buf, version);
1205}
1206
1207SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1208{
1209	return ksys_old_shmctl(shmid, cmd, buf);
1210}
1211#endif
1212
1213#ifdef CONFIG_COMPAT
1214
1215struct compat_shmid_ds {
1216	struct compat_ipc_perm shm_perm;
1217	int shm_segsz;
1218	old_time32_t shm_atime;
1219	old_time32_t shm_dtime;
1220	old_time32_t shm_ctime;
1221	compat_ipc_pid_t shm_cpid;
1222	compat_ipc_pid_t shm_lpid;
1223	unsigned short shm_nattch;
1224	unsigned short shm_unused;
1225	compat_uptr_t shm_unused2;
1226	compat_uptr_t shm_unused3;
1227};
1228
1229struct compat_shminfo64 {
1230	compat_ulong_t shmmax;
1231	compat_ulong_t shmmin;
1232	compat_ulong_t shmmni;
1233	compat_ulong_t shmseg;
1234	compat_ulong_t shmall;
1235	compat_ulong_t __unused1;
1236	compat_ulong_t __unused2;
1237	compat_ulong_t __unused3;
1238	compat_ulong_t __unused4;
1239};
1240
1241struct compat_shm_info {
1242	compat_int_t used_ids;
1243	compat_ulong_t shm_tot, shm_rss, shm_swp;
1244	compat_ulong_t swap_attempts, swap_successes;
1245};
1246
1247static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1248					int version)
1249{
1250	if (in->shmmax > INT_MAX)
1251		in->shmmax = INT_MAX;
1252	if (version == IPC_64) {
1253		struct compat_shminfo64 info;
1254		memset(&info, 0, sizeof(info));
1255		info.shmmax = in->shmmax;
1256		info.shmmin = in->shmmin;
1257		info.shmmni = in->shmmni;
1258		info.shmseg = in->shmseg;
1259		info.shmall = in->shmall;
1260		return copy_to_user(buf, &info, sizeof(info));
1261	} else {
1262		struct shminfo info;
1263		memset(&info, 0, sizeof(info));
1264		info.shmmax = in->shmmax;
1265		info.shmmin = in->shmmin;
1266		info.shmmni = in->shmmni;
1267		info.shmseg = in->shmseg;
1268		info.shmall = in->shmall;
1269		return copy_to_user(buf, &info, sizeof(info));
1270	}
1271}
1272
1273static int put_compat_shm_info(struct shm_info *ip,
1274				struct compat_shm_info __user *uip)
1275{
1276	struct compat_shm_info info;
1277
1278	memset(&info, 0, sizeof(info));
1279	info.used_ids = ip->used_ids;
1280	info.shm_tot = ip->shm_tot;
1281	info.shm_rss = ip->shm_rss;
1282	info.shm_swp = ip->shm_swp;
1283	info.swap_attempts = ip->swap_attempts;
1284	info.swap_successes = ip->swap_successes;
1285	return copy_to_user(uip, &info, sizeof(info));
1286}
1287
1288static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1289					int version)
1290{
1291	if (version == IPC_64) {
1292		struct compat_shmid64_ds v;
1293		memset(&v, 0, sizeof(v));
1294		to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1295		v.shm_atime	 = lower_32_bits(in->shm_atime);
1296		v.shm_atime_high = upper_32_bits(in->shm_atime);
1297		v.shm_dtime	 = lower_32_bits(in->shm_dtime);
1298		v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1299		v.shm_ctime	 = lower_32_bits(in->shm_ctime);
1300		v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1301		v.shm_segsz = in->shm_segsz;
1302		v.shm_nattch = in->shm_nattch;
1303		v.shm_cpid = in->shm_cpid;
1304		v.shm_lpid = in->shm_lpid;
1305		return copy_to_user(buf, &v, sizeof(v));
1306	} else {
1307		struct compat_shmid_ds v;
1308		memset(&v, 0, sizeof(v));
1309		to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1310		v.shm_perm.key = in->shm_perm.key;
1311		v.shm_atime = in->shm_atime;
1312		v.shm_dtime = in->shm_dtime;
1313		v.shm_ctime = in->shm_ctime;
1314		v.shm_segsz = in->shm_segsz;
1315		v.shm_nattch = in->shm_nattch;
1316		v.shm_cpid = in->shm_cpid;
1317		v.shm_lpid = in->shm_lpid;
1318		return copy_to_user(buf, &v, sizeof(v));
1319	}
1320}
1321
1322static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1323					int version)
1324{
1325	memset(out, 0, sizeof(*out));
1326	if (version == IPC_64) {
1327		struct compat_shmid64_ds __user *p = buf;
1328		return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1329	} else {
1330		struct compat_shmid_ds __user *p = buf;
1331		return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1332	}
1333}
1334
1335static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1336{
1337	struct ipc_namespace *ns;
1338	struct shmid64_ds sem64;
1339	int err;
1340
1341	ns = current->nsproxy->ipc_ns;
1342
1343	if (cmd < 0 || shmid < 0)
1344		return -EINVAL;
1345
1346	switch (cmd) {
1347	case IPC_INFO: {
1348		struct shminfo64 shminfo;
1349		err = shmctl_ipc_info(ns, &shminfo);
1350		if (err < 0)
1351			return err;
1352		if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1353			err = -EFAULT;
1354		return err;
1355	}
1356	case SHM_INFO: {
1357		struct shm_info shm_info;
1358		err = shmctl_shm_info(ns, &shm_info);
1359		if (err < 0)
1360			return err;
1361		if (put_compat_shm_info(&shm_info, uptr))
1362			err = -EFAULT;
1363		return err;
1364	}
1365	case IPC_STAT:
1366	case SHM_STAT_ANY:
1367	case SHM_STAT:
1368		err = shmctl_stat(ns, shmid, cmd, &sem64);
1369		if (err < 0)
1370			return err;
1371		if (copy_compat_shmid_to_user(uptr, &sem64, version))
1372			err = -EFAULT;
1373		return err;
1374
1375	case IPC_SET:
1376		if (copy_compat_shmid_from_user(&sem64, uptr, version))
1377			return -EFAULT;
1378		/* fallthru */
1379	case IPC_RMID:
1380		return shmctl_down(ns, shmid, cmd, &sem64);
1381	case SHM_LOCK:
1382	case SHM_UNLOCK:
1383		return shmctl_do_lock(ns, shmid, cmd);
1384		break;
1385	default:
1386		return -EINVAL;
1387	}
1388	return err;
1389}
1390
1391COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1392{
1393	return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1394}
1395
1396#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1397long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1398{
1399	int version = compat_ipc_parse_version(&cmd);
1400
1401	return compat_ksys_shmctl(shmid, cmd, uptr, version);
1402}
1403
1404COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1405{
1406	return compat_ksys_old_shmctl(shmid, cmd, uptr);
1407}
1408#endif
1409#endif
1410
1411/*
1412 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1413 *
1414 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1415 * "raddr" thing points to kernel space, and there has to be a wrapper around
1416 * this.
1417 */
1418long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1419	      ulong *raddr, unsigned long shmlba)
1420{
1421	struct shmid_kernel *shp;
1422	unsigned long addr = (unsigned long)shmaddr;
1423	unsigned long size;
1424	struct file *file, *base;
1425	int    err;
1426	unsigned long flags = MAP_SHARED;
1427	unsigned long prot;
1428	int acc_mode;
1429	struct ipc_namespace *ns;
1430	struct shm_file_data *sfd;
1431	int f_flags;
1432	unsigned long populate = 0;
1433
1434	err = -EINVAL;
1435	if (shmid < 0)
1436		goto out;
1437
1438	if (addr) {
1439		if (addr & (shmlba - 1)) {
1440			if (shmflg & SHM_RND) {
1441				addr &= ~(shmlba - 1);  /* round down */
1442
1443				/*
1444				 * Ensure that the round-down is non-nil
1445				 * when remapping. This can happen for
1446				 * cases when addr < shmlba.
1447				 */
1448				if (!addr && (shmflg & SHM_REMAP))
1449					goto out;
1450			} else
1451#ifndef __ARCH_FORCE_SHMLBA
1452				if (addr & ~PAGE_MASK)
1453#endif
1454					goto out;
1455		}
1456
1457		flags |= MAP_FIXED;
1458	} else if ((shmflg & SHM_REMAP))
1459		goto out;
1460
1461	if (shmflg & SHM_RDONLY) {
1462		prot = PROT_READ;
1463		acc_mode = S_IRUGO;
1464		f_flags = O_RDONLY;
1465	} else {
1466		prot = PROT_READ | PROT_WRITE;
1467		acc_mode = S_IRUGO | S_IWUGO;
1468		f_flags = O_RDWR;
1469	}
1470	if (shmflg & SHM_EXEC) {
1471		prot |= PROT_EXEC;
1472		acc_mode |= S_IXUGO;
1473	}
1474
1475	/*
1476	 * We cannot rely on the fs check since SYSV IPC does have an
1477	 * additional creator id...
1478	 */
1479	ns = current->nsproxy->ipc_ns;
1480	rcu_read_lock();
1481	shp = shm_obtain_object_check(ns, shmid);
1482	if (IS_ERR(shp)) {
1483		err = PTR_ERR(shp);
1484		goto out_unlock;
1485	}
1486
1487	err = -EACCES;
1488	if (ipcperms(ns, &shp->shm_perm, acc_mode))
1489		goto out_unlock;
1490
1491	err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1492	if (err)
1493		goto out_unlock;
1494
1495	ipc_lock_object(&shp->shm_perm);
1496
1497	/* check if shm_destroy() is tearing down shp */
1498	if (!ipc_valid_object(&shp->shm_perm)) {
1499		ipc_unlock_object(&shp->shm_perm);
1500		err = -EIDRM;
1501		goto out_unlock;
1502	}
1503
1504	/*
1505	 * We need to take a reference to the real shm file to prevent the
1506	 * pointer from becoming stale in cases where the lifetime of the outer
1507	 * file extends beyond that of the shm segment.  It's not usually
1508	 * possible, but it can happen during remap_file_pages() emulation as
1509	 * that unmaps the memory, then does ->mmap() via file reference only.
1510	 * We'll deny the ->mmap() if the shm segment was since removed, but to
1511	 * detect shm ID reuse we need to compare the file pointers.
1512	 */
1513	base = get_file(shp->shm_file);
1514	shp->shm_nattch++;
1515	size = i_size_read(file_inode(base));
1516	ipc_unlock_object(&shp->shm_perm);
1517	rcu_read_unlock();
1518
1519	err = -ENOMEM;
1520	sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1521	if (!sfd) {
1522		fput(base);
1523		goto out_nattch;
1524	}
1525
1526	file = alloc_file_clone(base, f_flags,
1527			  is_file_hugepages(base) ?
1528				&shm_file_operations_huge :
1529				&shm_file_operations);
1530	err = PTR_ERR(file);
1531	if (IS_ERR(file)) {
1532		kfree(sfd);
1533		fput(base);
1534		goto out_nattch;
1535	}
1536
1537	sfd->id = shp->shm_perm.id;
1538	sfd->ns = get_ipc_ns(ns);
1539	sfd->file = base;
1540	sfd->vm_ops = NULL;
1541	file->private_data = sfd;
1542
1543	err = security_mmap_file(file, prot, flags);
1544	if (err)
1545		goto out_fput;
1546
1547	if (mmap_write_lock_killable(current->mm)) {
1548		err = -EINTR;
1549		goto out_fput;
1550	}
1551
1552	if (addr && !(shmflg & SHM_REMAP)) {
1553		err = -EINVAL;
1554		if (addr + size < addr)
1555			goto invalid;
1556
1557		if (find_vma_intersection(current->mm, addr, addr + size))
1558			goto invalid;
1559	}
1560
1561	addr = do_mmap(file, addr, size, prot, flags, 0, &populate, NULL);
1562	*raddr = addr;
1563	err = 0;
1564	if (IS_ERR_VALUE(addr))
1565		err = (long)addr;
1566invalid:
1567	mmap_write_unlock(current->mm);
1568	if (populate)
1569		mm_populate(addr, populate);
1570
1571out_fput:
1572	fput(file);
1573
1574out_nattch:
1575	down_write(&shm_ids(ns).rwsem);
1576	shp = shm_lock(ns, shmid);
1577	shp->shm_nattch--;
1578	if (shm_may_destroy(ns, shp))
1579		shm_destroy(ns, shp);
1580	else
1581		shm_unlock(shp);
1582	up_write(&shm_ids(ns).rwsem);
1583	return err;
1584
1585out_unlock:
1586	rcu_read_unlock();
1587out:
1588	return err;
1589}
1590
1591SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1592{
1593	unsigned long ret;
1594	long err;
1595
1596	err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1597	if (err)
1598		return err;
1599	force_successful_syscall_return();
1600	return (long)ret;
1601}
1602
1603#ifdef CONFIG_COMPAT
1604
1605#ifndef COMPAT_SHMLBA
1606#define COMPAT_SHMLBA	SHMLBA
1607#endif
1608
1609COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1610{
1611	unsigned long ret;
1612	long err;
1613
1614	err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1615	if (err)
1616		return err;
1617	force_successful_syscall_return();
1618	return (long)ret;
1619}
1620#endif
1621
1622/*
1623 * detach and kill segment if marked destroyed.
1624 * The work is done in shm_close.
1625 */
1626long ksys_shmdt(char __user *shmaddr)
1627{
1628	struct mm_struct *mm = current->mm;
1629	struct vm_area_struct *vma;
1630	unsigned long addr = (unsigned long)shmaddr;
1631	int retval = -EINVAL;
1632#ifdef CONFIG_MMU
1633	loff_t size = 0;
1634	struct file *file;
1635	struct vm_area_struct *next;
1636#endif
1637
1638	if (addr & ~PAGE_MASK)
1639		return retval;
1640
1641	if (mmap_write_lock_killable(mm))
1642		return -EINTR;
1643
1644	/*
1645	 * This function tries to be smart and unmap shm segments that
1646	 * were modified by partial mlock or munmap calls:
1647	 * - It first determines the size of the shm segment that should be
1648	 *   unmapped: It searches for a vma that is backed by shm and that
1649	 *   started at address shmaddr. It records it's size and then unmaps
1650	 *   it.
1651	 * - Then it unmaps all shm vmas that started at shmaddr and that
1652	 *   are within the initially determined size and that are from the
1653	 *   same shm segment from which we determined the size.
1654	 * Errors from do_munmap are ignored: the function only fails if
1655	 * it's called with invalid parameters or if it's called to unmap
1656	 * a part of a vma. Both calls in this function are for full vmas,
1657	 * the parameters are directly copied from the vma itself and always
1658	 * valid - therefore do_munmap cannot fail. (famous last words?)
1659	 */
1660	/*
1661	 * If it had been mremap()'d, the starting address would not
1662	 * match the usual checks anyway. So assume all vma's are
1663	 * above the starting address given.
1664	 */
1665	vma = find_vma(mm, addr);
1666
1667#ifdef CONFIG_MMU
1668	while (vma) {
1669		next = vma->vm_next;
1670
1671		/*
1672		 * Check if the starting address would match, i.e. it's
1673		 * a fragment created by mprotect() and/or munmap(), or it
1674		 * otherwise it starts at this address with no hassles.
1675		 */
1676		if ((vma->vm_ops == &shm_vm_ops) &&
1677			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1678
1679			/*
1680			 * Record the file of the shm segment being
1681			 * unmapped.  With mremap(), someone could place
1682			 * page from another segment but with equal offsets
1683			 * in the range we are unmapping.
1684			 */
1685			file = vma->vm_file;
1686			size = i_size_read(file_inode(vma->vm_file));
1687			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1688			/*
1689			 * We discovered the size of the shm segment, so
1690			 * break out of here and fall through to the next
1691			 * loop that uses the size information to stop
1692			 * searching for matching vma's.
1693			 */
1694			retval = 0;
1695			vma = next;
1696			break;
1697		}
1698		vma = next;
1699	}
1700
1701	/*
1702	 * We need look no further than the maximum address a fragment
1703	 * could possibly have landed at. Also cast things to loff_t to
1704	 * prevent overflows and make comparisons vs. equal-width types.
1705	 */
1706	size = PAGE_ALIGN(size);
1707	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1708		next = vma->vm_next;
1709
1710		/* finding a matching vma now does not alter retval */
1711		if ((vma->vm_ops == &shm_vm_ops) &&
1712		    ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1713		    (vma->vm_file == file))
1714			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1715		vma = next;
1716	}
1717
1718#else	/* CONFIG_MMU */
1719	/* under NOMMU conditions, the exact address to be destroyed must be
1720	 * given
1721	 */
1722	if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1723		do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1724		retval = 0;
1725	}
1726
1727#endif
1728
1729	mmap_write_unlock(mm);
1730	return retval;
1731}
1732
1733SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1734{
1735	return ksys_shmdt(shmaddr);
1736}
1737
1738#ifdef CONFIG_PROC_FS
1739static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1740{
1741	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1742	struct user_namespace *user_ns = seq_user_ns(s);
1743	struct kern_ipc_perm *ipcp = it;
1744	struct shmid_kernel *shp;
1745	unsigned long rss = 0, swp = 0;
1746
1747	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1748	shm_add_rss_swap(shp, &rss, &swp);
1749
1750#if BITS_PER_LONG <= 32
1751#define SIZE_SPEC "%10lu"
1752#else
1753#define SIZE_SPEC "%21lu"
1754#endif
1755
1756	seq_printf(s,
1757		   "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1758		   "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1759		   SIZE_SPEC " " SIZE_SPEC "\n",
1760		   shp->shm_perm.key,
1761		   shp->shm_perm.id,
1762		   shp->shm_perm.mode,
1763		   shp->shm_segsz,
1764		   pid_nr_ns(shp->shm_cprid, pid_ns),
1765		   pid_nr_ns(shp->shm_lprid, pid_ns),
1766		   shp->shm_nattch,
1767		   from_kuid_munged(user_ns, shp->shm_perm.uid),
1768		   from_kgid_munged(user_ns, shp->shm_perm.gid),
1769		   from_kuid_munged(user_ns, shp->shm_perm.cuid),
1770		   from_kgid_munged(user_ns, shp->shm_perm.cgid),
1771		   shp->shm_atim,
1772		   shp->shm_dtim,
1773		   shp->shm_ctim,
1774		   rss * PAGE_SIZE,
1775		   swp * PAGE_SIZE);
1776
1777	return 0;
1778}
1779#endif
1780