shm.c revision c2ab975c
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
47#include <linux/uaccess.h>
48
49#include "util.h"
50
51struct shmid_kernel /* private to the kernel */
52{
53	struct kern_ipc_perm	shm_perm;
54	struct file		*shm_file;
55	unsigned long		shm_nattch;
56	unsigned long		shm_segsz;
57	time64_t		shm_atim;
58	time64_t		shm_dtim;
59	time64_t		shm_ctim;
60	struct pid		*shm_cprid;
61	struct pid		*shm_lprid;
62	struct user_struct	*mlock_user;
63
64	/* The task created the shm object.  NULL if the task is dead. */
65	struct task_struct	*shm_creator;
66	struct list_head	shm_clist;	/* list by creator */
67} __randomize_layout;
68
69/* shm_mode upper byte flags */
70#define SHM_DEST	01000	/* segment will be destroyed on last detach */
71#define SHM_LOCKED	02000   /* segment will not be swapped */
72
73struct shm_file_data {
74	int id;
75	struct ipc_namespace *ns;
76	struct file *file;
77	const struct vm_operations_struct *vm_ops;
78};
79
80#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
81
82static const struct file_operations shm_file_operations;
83static const struct vm_operations_struct shm_vm_ops;
84
85#define shm_ids(ns)	((ns)->ids[IPC_SHM_IDS])
86
87#define shm_unlock(shp)			\
88	ipc_unlock(&(shp)->shm_perm)
89
90static int newseg(struct ipc_namespace *, struct ipc_params *);
91static void shm_open(struct vm_area_struct *vma);
92static void shm_close(struct vm_area_struct *vma);
93static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
94#ifdef CONFIG_PROC_FS
95static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
96#endif
97
98int shm_init_ns(struct ipc_namespace *ns)
99{
100	ns->shm_ctlmax = SHMMAX;
101	ns->shm_ctlall = SHMALL;
102	ns->shm_ctlmni = SHMMNI;
103	ns->shm_rmid_forced = 0;
104	ns->shm_tot = 0;
105	return ipc_init_ids(&shm_ids(ns));
106}
107
108/*
109 * Called with shm_ids.rwsem (writer) and the shp structure locked.
110 * Only shm_ids.rwsem remains locked on exit.
111 */
112static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
113{
114	struct shmid_kernel *shp;
115
116	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
117
118	if (shp->shm_nattch) {
119		shp->shm_perm.mode |= SHM_DEST;
120		/* Do not find it any more */
121		ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
122		shm_unlock(shp);
123	} else
124		shm_destroy(ns, shp);
125}
126
127#ifdef CONFIG_IPC_NS
128void shm_exit_ns(struct ipc_namespace *ns)
129{
130	free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
131	idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
132	rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
133}
134#endif
135
136static int __init ipc_ns_init(void)
137{
138	const int err = shm_init_ns(&init_ipc_ns);
139	WARN(err, "ipc: sysv shm_init_ns failed: %d\n", err);
140	return err;
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 = ipc_lock(&shm_ids(ns), id);
183
184	/*
185	 * Callers of shm_lock() must validate the status of the returned ipc
186	 * object pointer (as returned by ipc_lock()), and error out as
187	 * appropriate.
188	 */
189	if (IS_ERR(ipcp))
190		return (void *)ipcp;
191	return container_of(ipcp, struct shmid_kernel, shm_perm);
192}
193
194static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
195{
196	rcu_read_lock();
197	ipc_lock_object(&ipcp->shm_perm);
198}
199
200static void shm_rcu_free(struct rcu_head *head)
201{
202	struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
203							rcu);
204	struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
205							shm_perm);
206	security_shm_free(&shp->shm_perm);
207	kvfree(shp);
208}
209
210static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
211{
212	list_del(&s->shm_clist);
213	ipc_rmid(&shm_ids(ns), &s->shm_perm);
214}
215
216
217static int __shm_open(struct vm_area_struct *vma)
218{
219	struct file *file = vma->vm_file;
220	struct shm_file_data *sfd = shm_file_data(file);
221	struct shmid_kernel *shp;
222
223	shp = shm_lock(sfd->ns, sfd->id);
224
225	if (IS_ERR(shp))
226		return PTR_ERR(shp);
227
228	if (shp->shm_file != sfd->file) {
229		/* ID was reused */
230		shm_unlock(shp);
231		return -EINVAL;
232	}
233
234	shp->shm_atim = ktime_get_real_seconds();
235	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
236	shp->shm_nattch++;
237	shm_unlock(shp);
238	return 0;
239}
240
241/* This is called by fork, once for every shm attach. */
242static void shm_open(struct vm_area_struct *vma)
243{
244	int err = __shm_open(vma);
245	/*
246	 * We raced in the idr lookup or with shm_destroy().
247	 * Either way, the ID is busted.
248	 */
249	WARN_ON_ONCE(err);
250}
251
252/*
253 * shm_destroy - free the struct shmid_kernel
254 *
255 * @ns: namespace
256 * @shp: struct to free
257 *
258 * It has to be called with shp and shm_ids.rwsem (writer) locked,
259 * but returns with shp unlocked and freed.
260 */
261static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
262{
263	struct file *shm_file;
264
265	shm_file = shp->shm_file;
266	shp->shm_file = NULL;
267	ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
268	shm_rmid(ns, shp);
269	shm_unlock(shp);
270	if (!is_file_hugepages(shm_file))
271		shmem_lock(shm_file, 0, shp->mlock_user);
272	else if (shp->mlock_user)
273		user_shm_unlock(i_size_read(file_inode(shm_file)),
274				shp->mlock_user);
275	fput(shm_file);
276	ipc_update_pid(&shp->shm_cprid, NULL);
277	ipc_update_pid(&shp->shm_lprid, NULL);
278	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
279}
280
281/*
282 * shm_may_destroy - identifies whether shm segment should be destroyed now
283 *
284 * Returns true if and only if there are no active users of the segment and
285 * one of the following is true:
286 *
287 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
288 *
289 * 2) sysctl kernel.shm_rmid_forced is set to 1.
290 */
291static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
292{
293	return (shp->shm_nattch == 0) &&
294	       (ns->shm_rmid_forced ||
295		(shp->shm_perm.mode & SHM_DEST));
296}
297
298/*
299 * remove the attach descriptor vma.
300 * free memory for segment if it is marked destroyed.
301 * The descriptor has already been removed from the current->mm->mmap list
302 * and will later be kfree()d.
303 */
304static void shm_close(struct vm_area_struct *vma)
305{
306	struct file *file = vma->vm_file;
307	struct shm_file_data *sfd = shm_file_data(file);
308	struct shmid_kernel *shp;
309	struct ipc_namespace *ns = sfd->ns;
310
311	down_write(&shm_ids(ns).rwsem);
312	/* remove from the list of attaches of the shm segment */
313	shp = shm_lock(ns, sfd->id);
314
315	/*
316	 * We raced in the idr lookup or with shm_destroy().
317	 * Either way, the ID is busted.
318	 */
319	if (WARN_ON_ONCE(IS_ERR(shp)))
320		goto done; /* no-op */
321
322	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
323	shp->shm_dtim = ktime_get_real_seconds();
324	shp->shm_nattch--;
325	if (shm_may_destroy(ns, shp))
326		shm_destroy(ns, shp);
327	else
328		shm_unlock(shp);
329done:
330	up_write(&shm_ids(ns).rwsem);
331}
332
333/* Called with ns->shm_ids(ns).rwsem locked */
334static int shm_try_destroy_orphaned(int id, void *p, void *data)
335{
336	struct ipc_namespace *ns = data;
337	struct kern_ipc_perm *ipcp = p;
338	struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
339
340	/*
341	 * We want to destroy segments without users and with already
342	 * exit'ed originating process.
343	 *
344	 * As shp->* are changed under rwsem, it's safe to skip shp locking.
345	 */
346	if (shp->shm_creator != NULL)
347		return 0;
348
349	if (shm_may_destroy(ns, shp)) {
350		shm_lock_by_ptr(shp);
351		shm_destroy(ns, shp);
352	}
353	return 0;
354}
355
356void shm_destroy_orphaned(struct ipc_namespace *ns)
357{
358	down_write(&shm_ids(ns).rwsem);
359	if (shm_ids(ns).in_use)
360		idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
361	up_write(&shm_ids(ns).rwsem);
362}
363
364/* Locking assumes this will only be called with task == current */
365void exit_shm(struct task_struct *task)
366{
367	struct ipc_namespace *ns = task->nsproxy->ipc_ns;
368	struct shmid_kernel *shp, *n;
369
370	if (list_empty(&task->sysvshm.shm_clist))
371		return;
372
373	/*
374	 * If kernel.shm_rmid_forced is not set then only keep track of
375	 * which shmids are orphaned, so that a later set of the sysctl
376	 * can clean them up.
377	 */
378	if (!ns->shm_rmid_forced) {
379		down_read(&shm_ids(ns).rwsem);
380		list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
381			shp->shm_creator = NULL;
382		/*
383		 * Only under read lock but we are only called on current
384		 * so no entry on the list will be shared.
385		 */
386		list_del(&task->sysvshm.shm_clist);
387		up_read(&shm_ids(ns).rwsem);
388		return;
389	}
390
391	/*
392	 * Destroy all already created segments, that were not yet mapped,
393	 * and mark any mapped as orphan to cover the sysctl toggling.
394	 * Destroy is skipped if shm_may_destroy() returns false.
395	 */
396	down_write(&shm_ids(ns).rwsem);
397	list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
398		shp->shm_creator = NULL;
399
400		if (shm_may_destroy(ns, shp)) {
401			shm_lock_by_ptr(shp);
402			shm_destroy(ns, shp);
403		}
404	}
405
406	/* Remove the list head from any segments still attached. */
407	list_del(&task->sysvshm.shm_clist);
408	up_write(&shm_ids(ns).rwsem);
409}
410
411static int shm_fault(struct vm_fault *vmf)
412{
413	struct file *file = vmf->vma->vm_file;
414	struct shm_file_data *sfd = shm_file_data(file);
415
416	return sfd->vm_ops->fault(vmf);
417}
418
419static int shm_split(struct vm_area_struct *vma, unsigned long addr)
420{
421	struct file *file = vma->vm_file;
422	struct shm_file_data *sfd = shm_file_data(file);
423
424	if (sfd->vm_ops->split)
425		return sfd->vm_ops->split(vma, addr);
426
427	return 0;
428}
429
430#ifdef CONFIG_NUMA
431static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
432{
433	struct file *file = vma->vm_file;
434	struct shm_file_data *sfd = shm_file_data(file);
435	int err = 0;
436
437	if (sfd->vm_ops->set_policy)
438		err = sfd->vm_ops->set_policy(vma, new);
439	return err;
440}
441
442static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
443					unsigned long addr)
444{
445	struct file *file = vma->vm_file;
446	struct shm_file_data *sfd = shm_file_data(file);
447	struct mempolicy *pol = NULL;
448
449	if (sfd->vm_ops->get_policy)
450		pol = sfd->vm_ops->get_policy(vma, addr);
451	else if (vma->vm_policy)
452		pol = vma->vm_policy;
453
454	return pol;
455}
456#endif
457
458static int shm_mmap(struct file *file, struct vm_area_struct *vma)
459{
460	struct shm_file_data *sfd = shm_file_data(file);
461	int ret;
462
463	/*
464	 * In case of remap_file_pages() emulation, the file can represent an
465	 * IPC ID that was removed, and possibly even reused by another shm
466	 * segment already.  Propagate this case as an error to caller.
467	 */
468	ret = __shm_open(vma);
469	if (ret)
470		return ret;
471
472	ret = call_mmap(sfd->file, vma);
473	if (ret) {
474		shm_close(vma);
475		return ret;
476	}
477	sfd->vm_ops = vma->vm_ops;
478#ifdef CONFIG_MMU
479	WARN_ON(!sfd->vm_ops->fault);
480#endif
481	vma->vm_ops = &shm_vm_ops;
482	return 0;
483}
484
485static int shm_release(struct inode *ino, struct file *file)
486{
487	struct shm_file_data *sfd = shm_file_data(file);
488
489	put_ipc_ns(sfd->ns);
490	fput(sfd->file);
491	shm_file_data(file) = NULL;
492	kfree(sfd);
493	return 0;
494}
495
496static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
497{
498	struct shm_file_data *sfd = shm_file_data(file);
499
500	if (!sfd->file->f_op->fsync)
501		return -EINVAL;
502	return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
503}
504
505static long shm_fallocate(struct file *file, int mode, loff_t offset,
506			  loff_t len)
507{
508	struct shm_file_data *sfd = shm_file_data(file);
509
510	if (!sfd->file->f_op->fallocate)
511		return -EOPNOTSUPP;
512	return sfd->file->f_op->fallocate(file, mode, offset, len);
513}
514
515static unsigned long shm_get_unmapped_area(struct file *file,
516	unsigned long addr, unsigned long len, unsigned long pgoff,
517	unsigned long flags)
518{
519	struct shm_file_data *sfd = shm_file_data(file);
520
521	return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
522						pgoff, flags);
523}
524
525static const struct file_operations shm_file_operations = {
526	.mmap		= shm_mmap,
527	.fsync		= shm_fsync,
528	.release	= shm_release,
529	.get_unmapped_area	= shm_get_unmapped_area,
530	.llseek		= noop_llseek,
531	.fallocate	= shm_fallocate,
532};
533
534/*
535 * shm_file_operations_huge is now identical to shm_file_operations,
536 * but we keep it distinct for the sake of is_file_shm_hugepages().
537 */
538static const struct file_operations shm_file_operations_huge = {
539	.mmap		= shm_mmap,
540	.fsync		= shm_fsync,
541	.release	= shm_release,
542	.get_unmapped_area	= shm_get_unmapped_area,
543	.llseek		= noop_llseek,
544	.fallocate	= shm_fallocate,
545};
546
547bool is_file_shm_hugepages(struct file *file)
548{
549	return file->f_op == &shm_file_operations_huge;
550}
551
552static const struct vm_operations_struct shm_vm_ops = {
553	.open	= shm_open,	/* callback for a new vm-area open */
554	.close	= shm_close,	/* callback for when the vm-area is released */
555	.fault	= shm_fault,
556	.split	= shm_split,
557#if defined(CONFIG_NUMA)
558	.set_policy = shm_set_policy,
559	.get_policy = shm_get_policy,
560#endif
561};
562
563/**
564 * newseg - Create a new shared memory segment
565 * @ns: namespace
566 * @params: ptr to the structure that contains key, size and shmflg
567 *
568 * Called with shm_ids.rwsem held as a writer.
569 */
570static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
571{
572	key_t key = params->key;
573	int shmflg = params->flg;
574	size_t size = params->u.size;
575	int error;
576	struct shmid_kernel *shp;
577	size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
578	struct file *file;
579	char name[13];
580	vm_flags_t acctflag = 0;
581
582	if (size < SHMMIN || size > ns->shm_ctlmax)
583		return -EINVAL;
584
585	if (numpages << PAGE_SHIFT < size)
586		return -ENOSPC;
587
588	if (ns->shm_tot + numpages < ns->shm_tot ||
589			ns->shm_tot + numpages > ns->shm_ctlall)
590		return -ENOSPC;
591
592	shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
593	if (unlikely(!shp))
594		return -ENOMEM;
595
596	shp->shm_perm.key = key;
597	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
598	shp->mlock_user = NULL;
599
600	shp->shm_perm.security = NULL;
601	error = security_shm_alloc(&shp->shm_perm);
602	if (error) {
603		kvfree(shp);
604		return error;
605	}
606
607	sprintf(name, "SYSV%08x", key);
608	if (shmflg & SHM_HUGETLB) {
609		struct hstate *hs;
610		size_t hugesize;
611
612		hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
613		if (!hs) {
614			error = -EINVAL;
615			goto no_file;
616		}
617		hugesize = ALIGN(size, huge_page_size(hs));
618
619		/* hugetlb_file_setup applies strict accounting */
620		if (shmflg & SHM_NORESERVE)
621			acctflag = VM_NORESERVE;
622		file = hugetlb_file_setup(name, hugesize, acctflag,
623				  &shp->mlock_user, HUGETLB_SHMFS_INODE,
624				(shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
625	} else {
626		/*
627		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
628		 * if it's asked for.
629		 */
630		if  ((shmflg & SHM_NORESERVE) &&
631				sysctl_overcommit_memory != OVERCOMMIT_NEVER)
632			acctflag = VM_NORESERVE;
633		file = shmem_kernel_file_setup(name, size, acctflag);
634	}
635	error = PTR_ERR(file);
636	if (IS_ERR(file))
637		goto no_file;
638
639	shp->shm_cprid = get_pid(task_tgid(current));
640	shp->shm_lprid = NULL;
641	shp->shm_atim = shp->shm_dtim = 0;
642	shp->shm_ctim = ktime_get_real_seconds();
643	shp->shm_segsz = size;
644	shp->shm_nattch = 0;
645	shp->shm_file = file;
646	shp->shm_creator = current;
647
648	/* ipc_addid() locks shp upon success. */
649	error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
650	if (error < 0)
651		goto no_id;
652
653	list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
654
655	/*
656	 * shmid gets reported as "inode#" in /proc/pid/maps.
657	 * proc-ps tools use this. Changing this will break them.
658	 */
659	file_inode(file)->i_ino = shp->shm_perm.id;
660
661	ns->shm_tot += numpages;
662	error = shp->shm_perm.id;
663
664	ipc_unlock_object(&shp->shm_perm);
665	rcu_read_unlock();
666	return error;
667
668no_id:
669	ipc_update_pid(&shp->shm_cprid, NULL);
670	ipc_update_pid(&shp->shm_lprid, NULL);
671	if (is_file_hugepages(file) && shp->mlock_user)
672		user_shm_unlock(size, shp->mlock_user);
673	fput(file);
674no_file:
675	call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
676	return error;
677}
678
679/*
680 * Called with shm_ids.rwsem and ipcp locked.
681 */
682static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
683				struct ipc_params *params)
684{
685	struct shmid_kernel *shp;
686
687	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
688	if (shp->shm_segsz < params->u.size)
689		return -EINVAL;
690
691	return 0;
692}
693
694long ksys_shmget(key_t key, size_t size, int shmflg)
695{
696	struct ipc_namespace *ns;
697	static const struct ipc_ops shm_ops = {
698		.getnew = newseg,
699		.associate = security_shm_associate,
700		.more_checks = shm_more_checks,
701	};
702	struct ipc_params shm_params;
703
704	ns = current->nsproxy->ipc_ns;
705
706	shm_params.key = key;
707	shm_params.flg = shmflg;
708	shm_params.u.size = size;
709
710	return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
711}
712
713SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
714{
715	return ksys_shmget(key, size, shmflg);
716}
717
718static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
719{
720	switch (version) {
721	case IPC_64:
722		return copy_to_user(buf, in, sizeof(*in));
723	case IPC_OLD:
724	    {
725		struct shmid_ds out;
726
727		memset(&out, 0, sizeof(out));
728		ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
729		out.shm_segsz	= in->shm_segsz;
730		out.shm_atime	= in->shm_atime;
731		out.shm_dtime	= in->shm_dtime;
732		out.shm_ctime	= in->shm_ctime;
733		out.shm_cpid	= in->shm_cpid;
734		out.shm_lpid	= in->shm_lpid;
735		out.shm_nattch	= in->shm_nattch;
736
737		return copy_to_user(buf, &out, sizeof(out));
738	    }
739	default:
740		return -EINVAL;
741	}
742}
743
744static inline unsigned long
745copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
746{
747	switch (version) {
748	case IPC_64:
749		if (copy_from_user(out, buf, sizeof(*out)))
750			return -EFAULT;
751		return 0;
752	case IPC_OLD:
753	    {
754		struct shmid_ds tbuf_old;
755
756		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
757			return -EFAULT;
758
759		out->shm_perm.uid	= tbuf_old.shm_perm.uid;
760		out->shm_perm.gid	= tbuf_old.shm_perm.gid;
761		out->shm_perm.mode	= tbuf_old.shm_perm.mode;
762
763		return 0;
764	    }
765	default:
766		return -EINVAL;
767	}
768}
769
770static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
771{
772	switch (version) {
773	case IPC_64:
774		return copy_to_user(buf, in, sizeof(*in));
775	case IPC_OLD:
776	    {
777		struct shminfo out;
778
779		if (in->shmmax > INT_MAX)
780			out.shmmax = INT_MAX;
781		else
782			out.shmmax = (int)in->shmmax;
783
784		out.shmmin	= in->shmmin;
785		out.shmmni	= in->shmmni;
786		out.shmseg	= in->shmseg;
787		out.shmall	= in->shmall;
788
789		return copy_to_user(buf, &out, sizeof(out));
790	    }
791	default:
792		return -EINVAL;
793	}
794}
795
796/*
797 * Calculate and add used RSS and swap pages of a shm.
798 * Called with shm_ids.rwsem held as a reader
799 */
800static void shm_add_rss_swap(struct shmid_kernel *shp,
801	unsigned long *rss_add, unsigned long *swp_add)
802{
803	struct inode *inode;
804
805	inode = file_inode(shp->shm_file);
806
807	if (is_file_hugepages(shp->shm_file)) {
808		struct address_space *mapping = inode->i_mapping;
809		struct hstate *h = hstate_file(shp->shm_file);
810		*rss_add += pages_per_huge_page(h) * mapping->nrpages;
811	} else {
812#ifdef CONFIG_SHMEM
813		struct shmem_inode_info *info = SHMEM_I(inode);
814
815		spin_lock_irq(&info->lock);
816		*rss_add += inode->i_mapping->nrpages;
817		*swp_add += info->swapped;
818		spin_unlock_irq(&info->lock);
819#else
820		*rss_add += inode->i_mapping->nrpages;
821#endif
822	}
823}
824
825/*
826 * Called with shm_ids.rwsem held as a reader
827 */
828static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
829		unsigned long *swp)
830{
831	int next_id;
832	int total, in_use;
833
834	*rss = 0;
835	*swp = 0;
836
837	in_use = shm_ids(ns).in_use;
838
839	for (total = 0, next_id = 0; total < in_use; next_id++) {
840		struct kern_ipc_perm *ipc;
841		struct shmid_kernel *shp;
842
843		ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
844		if (ipc == NULL)
845			continue;
846		shp = container_of(ipc, struct shmid_kernel, shm_perm);
847
848		shm_add_rss_swap(shp, rss, swp);
849
850		total++;
851	}
852}
853
854/*
855 * This function handles some shmctl commands which require the rwsem
856 * to be held in write mode.
857 * NOTE: no locks must be held, the rwsem is taken inside this function.
858 */
859static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
860		       struct shmid64_ds *shmid64)
861{
862	struct kern_ipc_perm *ipcp;
863	struct shmid_kernel *shp;
864	int err;
865
866	down_write(&shm_ids(ns).rwsem);
867	rcu_read_lock();
868
869	ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
870				      &shmid64->shm_perm, 0);
871	if (IS_ERR(ipcp)) {
872		err = PTR_ERR(ipcp);
873		goto out_unlock1;
874	}
875
876	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
877
878	err = security_shm_shmctl(&shp->shm_perm, cmd);
879	if (err)
880		goto out_unlock1;
881
882	switch (cmd) {
883	case IPC_RMID:
884		ipc_lock_object(&shp->shm_perm);
885		/* do_shm_rmid unlocks the ipc object and rcu */
886		do_shm_rmid(ns, ipcp);
887		goto out_up;
888	case IPC_SET:
889		ipc_lock_object(&shp->shm_perm);
890		err = ipc_update_perm(&shmid64->shm_perm, ipcp);
891		if (err)
892			goto out_unlock0;
893		shp->shm_ctim = ktime_get_real_seconds();
894		break;
895	default:
896		err = -EINVAL;
897		goto out_unlock1;
898	}
899
900out_unlock0:
901	ipc_unlock_object(&shp->shm_perm);
902out_unlock1:
903	rcu_read_unlock();
904out_up:
905	up_write(&shm_ids(ns).rwsem);
906	return err;
907}
908
909static int shmctl_ipc_info(struct ipc_namespace *ns,
910			   struct shminfo64 *shminfo)
911{
912	int err = security_shm_shmctl(NULL, IPC_INFO);
913	if (!err) {
914		memset(shminfo, 0, sizeof(*shminfo));
915		shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
916		shminfo->shmmax = ns->shm_ctlmax;
917		shminfo->shmall = ns->shm_ctlall;
918		shminfo->shmmin = SHMMIN;
919		down_read(&shm_ids(ns).rwsem);
920		err = ipc_get_maxid(&shm_ids(ns));
921		up_read(&shm_ids(ns).rwsem);
922		if (err < 0)
923			err = 0;
924	}
925	return err;
926}
927
928static int shmctl_shm_info(struct ipc_namespace *ns,
929			   struct shm_info *shm_info)
930{
931	int err = security_shm_shmctl(NULL, SHM_INFO);
932	if (!err) {
933		memset(shm_info, 0, sizeof(*shm_info));
934		down_read(&shm_ids(ns).rwsem);
935		shm_info->used_ids = shm_ids(ns).in_use;
936		shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
937		shm_info->shm_tot = ns->shm_tot;
938		shm_info->swap_attempts = 0;
939		shm_info->swap_successes = 0;
940		err = ipc_get_maxid(&shm_ids(ns));
941		up_read(&shm_ids(ns).rwsem);
942		if (err < 0)
943			err = 0;
944	}
945	return err;
946}
947
948static int shmctl_stat(struct ipc_namespace *ns, int shmid,
949			int cmd, struct shmid64_ds *tbuf)
950{
951	struct shmid_kernel *shp;
952	int id = 0;
953	int err;
954
955	memset(tbuf, 0, sizeof(*tbuf));
956
957	rcu_read_lock();
958	if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
959		shp = shm_obtain_object(ns, shmid);
960		if (IS_ERR(shp)) {
961			err = PTR_ERR(shp);
962			goto out_unlock;
963		}
964		id = shp->shm_perm.id;
965	} else { /* IPC_STAT */
966		shp = shm_obtain_object_check(ns, shmid);
967		if (IS_ERR(shp)) {
968			err = PTR_ERR(shp);
969			goto out_unlock;
970		}
971	}
972
973	/*
974	 * Semantically SHM_STAT_ANY ought to be identical to
975	 * that functionality provided by the /proc/sysvipc/
976	 * interface. As such, only audit these calls and
977	 * do not do traditional S_IRUGO permission checks on
978	 * the ipc object.
979	 */
980	if (cmd == SHM_STAT_ANY)
981		audit_ipc_obj(&shp->shm_perm);
982	else {
983		err = -EACCES;
984		if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
985			goto out_unlock;
986	}
987
988	err = security_shm_shmctl(&shp->shm_perm, cmd);
989	if (err)
990		goto out_unlock;
991
992	ipc_lock_object(&shp->shm_perm);
993
994	if (!ipc_valid_object(&shp->shm_perm)) {
995		ipc_unlock_object(&shp->shm_perm);
996		err = -EIDRM;
997		goto out_unlock;
998	}
999
1000	kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1001	tbuf->shm_segsz	= shp->shm_segsz;
1002	tbuf->shm_atime	= shp->shm_atim;
1003	tbuf->shm_dtime	= shp->shm_dtim;
1004	tbuf->shm_ctime	= shp->shm_ctim;
1005#ifndef CONFIG_64BIT
1006	tbuf->shm_atime_high = shp->shm_atim >> 32;
1007	tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1008	tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1009#endif
1010	tbuf->shm_cpid	= pid_vnr(shp->shm_cprid);
1011	tbuf->shm_lpid	= pid_vnr(shp->shm_lprid);
1012	tbuf->shm_nattch = shp->shm_nattch;
1013
1014	ipc_unlock_object(&shp->shm_perm);
1015	rcu_read_unlock();
1016	return id;
1017
1018out_unlock:
1019	rcu_read_unlock();
1020	return err;
1021}
1022
1023static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1024{
1025	struct shmid_kernel *shp;
1026	struct file *shm_file;
1027	int err;
1028
1029	rcu_read_lock();
1030	shp = shm_obtain_object_check(ns, shmid);
1031	if (IS_ERR(shp)) {
1032		err = PTR_ERR(shp);
1033		goto out_unlock1;
1034	}
1035
1036	audit_ipc_obj(&(shp->shm_perm));
1037	err = security_shm_shmctl(&shp->shm_perm, cmd);
1038	if (err)
1039		goto out_unlock1;
1040
1041	ipc_lock_object(&shp->shm_perm);
1042
1043	/* check if shm_destroy() is tearing down shp */
1044	if (!ipc_valid_object(&shp->shm_perm)) {
1045		err = -EIDRM;
1046		goto out_unlock0;
1047	}
1048
1049	if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1050		kuid_t euid = current_euid();
1051
1052		if (!uid_eq(euid, shp->shm_perm.uid) &&
1053		    !uid_eq(euid, shp->shm_perm.cuid)) {
1054			err = -EPERM;
1055			goto out_unlock0;
1056		}
1057		if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1058			err = -EPERM;
1059			goto out_unlock0;
1060		}
1061	}
1062
1063	shm_file = shp->shm_file;
1064	if (is_file_hugepages(shm_file))
1065		goto out_unlock0;
1066
1067	if (cmd == SHM_LOCK) {
1068		struct user_struct *user = current_user();
1069
1070		err = shmem_lock(shm_file, 1, user);
1071		if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1072			shp->shm_perm.mode |= SHM_LOCKED;
1073			shp->mlock_user = user;
1074		}
1075		goto out_unlock0;
1076	}
1077
1078	/* SHM_UNLOCK */
1079	if (!(shp->shm_perm.mode & SHM_LOCKED))
1080		goto out_unlock0;
1081	shmem_lock(shm_file, 0, shp->mlock_user);
1082	shp->shm_perm.mode &= ~SHM_LOCKED;
1083	shp->mlock_user = NULL;
1084	get_file(shm_file);
1085	ipc_unlock_object(&shp->shm_perm);
1086	rcu_read_unlock();
1087	shmem_unlock_mapping(shm_file->f_mapping);
1088
1089	fput(shm_file);
1090	return err;
1091
1092out_unlock0:
1093	ipc_unlock_object(&shp->shm_perm);
1094out_unlock1:
1095	rcu_read_unlock();
1096	return err;
1097}
1098
1099long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1100{
1101	int err, version;
1102	struct ipc_namespace *ns;
1103	struct shmid64_ds sem64;
1104
1105	if (cmd < 0 || shmid < 0)
1106		return -EINVAL;
1107
1108	version = ipc_parse_version(&cmd);
1109	ns = current->nsproxy->ipc_ns;
1110
1111	switch (cmd) {
1112	case IPC_INFO: {
1113		struct shminfo64 shminfo;
1114		err = shmctl_ipc_info(ns, &shminfo);
1115		if (err < 0)
1116			return err;
1117		if (copy_shminfo_to_user(buf, &shminfo, version))
1118			err = -EFAULT;
1119		return err;
1120	}
1121	case SHM_INFO: {
1122		struct shm_info shm_info;
1123		err = shmctl_shm_info(ns, &shm_info);
1124		if (err < 0)
1125			return err;
1126		if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1127			err = -EFAULT;
1128		return err;
1129	}
1130	case SHM_STAT:
1131	case SHM_STAT_ANY:
1132	case IPC_STAT: {
1133		err = shmctl_stat(ns, shmid, cmd, &sem64);
1134		if (err < 0)
1135			return err;
1136		if (copy_shmid_to_user(buf, &sem64, version))
1137			err = -EFAULT;
1138		return err;
1139	}
1140	case IPC_SET:
1141		if (copy_shmid_from_user(&sem64, buf, version))
1142			return -EFAULT;
1143		/* fallthru */
1144	case IPC_RMID:
1145		return shmctl_down(ns, shmid, cmd, &sem64);
1146	case SHM_LOCK:
1147	case SHM_UNLOCK:
1148		return shmctl_do_lock(ns, shmid, cmd);
1149	default:
1150		return -EINVAL;
1151	}
1152}
1153
1154SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1155{
1156	return ksys_shmctl(shmid, cmd, buf);
1157}
1158
1159#ifdef CONFIG_COMPAT
1160
1161struct compat_shmid_ds {
1162	struct compat_ipc_perm shm_perm;
1163	int shm_segsz;
1164	compat_time_t shm_atime;
1165	compat_time_t shm_dtime;
1166	compat_time_t shm_ctime;
1167	compat_ipc_pid_t shm_cpid;
1168	compat_ipc_pid_t shm_lpid;
1169	unsigned short shm_nattch;
1170	unsigned short shm_unused;
1171	compat_uptr_t shm_unused2;
1172	compat_uptr_t shm_unused3;
1173};
1174
1175struct compat_shminfo64 {
1176	compat_ulong_t shmmax;
1177	compat_ulong_t shmmin;
1178	compat_ulong_t shmmni;
1179	compat_ulong_t shmseg;
1180	compat_ulong_t shmall;
1181	compat_ulong_t __unused1;
1182	compat_ulong_t __unused2;
1183	compat_ulong_t __unused3;
1184	compat_ulong_t __unused4;
1185};
1186
1187struct compat_shm_info {
1188	compat_int_t used_ids;
1189	compat_ulong_t shm_tot, shm_rss, shm_swp;
1190	compat_ulong_t swap_attempts, swap_successes;
1191};
1192
1193static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1194					int version)
1195{
1196	if (in->shmmax > INT_MAX)
1197		in->shmmax = INT_MAX;
1198	if (version == IPC_64) {
1199		struct compat_shminfo64 info;
1200		memset(&info, 0, sizeof(info));
1201		info.shmmax = in->shmmax;
1202		info.shmmin = in->shmmin;
1203		info.shmmni = in->shmmni;
1204		info.shmseg = in->shmseg;
1205		info.shmall = in->shmall;
1206		return copy_to_user(buf, &info, sizeof(info));
1207	} else {
1208		struct shminfo info;
1209		memset(&info, 0, sizeof(info));
1210		info.shmmax = in->shmmax;
1211		info.shmmin = in->shmmin;
1212		info.shmmni = in->shmmni;
1213		info.shmseg = in->shmseg;
1214		info.shmall = in->shmall;
1215		return copy_to_user(buf, &info, sizeof(info));
1216	}
1217}
1218
1219static int put_compat_shm_info(struct shm_info *ip,
1220				struct compat_shm_info __user *uip)
1221{
1222	struct compat_shm_info info;
1223
1224	memset(&info, 0, sizeof(info));
1225	info.used_ids = ip->used_ids;
1226	info.shm_tot = ip->shm_tot;
1227	info.shm_rss = ip->shm_rss;
1228	info.shm_swp = ip->shm_swp;
1229	info.swap_attempts = ip->swap_attempts;
1230	info.swap_successes = ip->swap_successes;
1231	return copy_to_user(uip, &info, sizeof(info));
1232}
1233
1234static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1235					int version)
1236{
1237	if (version == IPC_64) {
1238		struct compat_shmid64_ds v;
1239		memset(&v, 0, sizeof(v));
1240		to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1241		v.shm_atime	 = lower_32_bits(in->shm_atime);
1242		v.shm_atime_high = upper_32_bits(in->shm_atime);
1243		v.shm_dtime	 = lower_32_bits(in->shm_dtime);
1244		v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1245		v.shm_ctime	 = lower_32_bits(in->shm_ctime);
1246		v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1247		v.shm_segsz = in->shm_segsz;
1248		v.shm_nattch = in->shm_nattch;
1249		v.shm_cpid = in->shm_cpid;
1250		v.shm_lpid = in->shm_lpid;
1251		return copy_to_user(buf, &v, sizeof(v));
1252	} else {
1253		struct compat_shmid_ds v;
1254		memset(&v, 0, sizeof(v));
1255		to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1256		v.shm_perm.key = in->shm_perm.key;
1257		v.shm_atime = in->shm_atime;
1258		v.shm_dtime = in->shm_dtime;
1259		v.shm_ctime = in->shm_ctime;
1260		v.shm_segsz = in->shm_segsz;
1261		v.shm_nattch = in->shm_nattch;
1262		v.shm_cpid = in->shm_cpid;
1263		v.shm_lpid = in->shm_lpid;
1264		return copy_to_user(buf, &v, sizeof(v));
1265	}
1266}
1267
1268static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1269					int version)
1270{
1271	memset(out, 0, sizeof(*out));
1272	if (version == IPC_64) {
1273		struct compat_shmid64_ds __user *p = buf;
1274		return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1275	} else {
1276		struct compat_shmid_ds __user *p = buf;
1277		return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1278	}
1279}
1280
1281long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1282{
1283	struct ipc_namespace *ns;
1284	struct shmid64_ds sem64;
1285	int version = compat_ipc_parse_version(&cmd);
1286	int err;
1287
1288	ns = current->nsproxy->ipc_ns;
1289
1290	if (cmd < 0 || shmid < 0)
1291		return -EINVAL;
1292
1293	switch (cmd) {
1294	case IPC_INFO: {
1295		struct shminfo64 shminfo;
1296		err = shmctl_ipc_info(ns, &shminfo);
1297		if (err < 0)
1298			return err;
1299		if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1300			err = -EFAULT;
1301		return err;
1302	}
1303	case SHM_INFO: {
1304		struct shm_info shm_info;
1305		err = shmctl_shm_info(ns, &shm_info);
1306		if (err < 0)
1307			return err;
1308		if (put_compat_shm_info(&shm_info, uptr))
1309			err = -EFAULT;
1310		return err;
1311	}
1312	case IPC_STAT:
1313	case SHM_STAT_ANY:
1314	case SHM_STAT:
1315		err = shmctl_stat(ns, shmid, cmd, &sem64);
1316		if (err < 0)
1317			return err;
1318		if (copy_compat_shmid_to_user(uptr, &sem64, version))
1319			err = -EFAULT;
1320		return err;
1321
1322	case IPC_SET:
1323		if (copy_compat_shmid_from_user(&sem64, uptr, version))
1324			return -EFAULT;
1325		/* fallthru */
1326	case IPC_RMID:
1327		return shmctl_down(ns, shmid, cmd, &sem64);
1328	case SHM_LOCK:
1329	case SHM_UNLOCK:
1330		return shmctl_do_lock(ns, shmid, cmd);
1331		break;
1332	default:
1333		return -EINVAL;
1334	}
1335	return err;
1336}
1337
1338COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1339{
1340	return compat_ksys_shmctl(shmid, cmd, uptr);
1341}
1342#endif
1343
1344/*
1345 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1346 *
1347 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1348 * "raddr" thing points to kernel space, and there has to be a wrapper around
1349 * this.
1350 */
1351long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1352	      ulong *raddr, unsigned long shmlba)
1353{
1354	struct shmid_kernel *shp;
1355	unsigned long addr = (unsigned long)shmaddr;
1356	unsigned long size;
1357	struct file *file;
1358	int    err;
1359	unsigned long flags = MAP_SHARED;
1360	unsigned long prot;
1361	int acc_mode;
1362	struct ipc_namespace *ns;
1363	struct shm_file_data *sfd;
1364	struct path path;
1365	fmode_t f_mode;
1366	unsigned long populate = 0;
1367
1368	err = -EINVAL;
1369	if (shmid < 0)
1370		goto out;
1371
1372	if (addr) {
1373		if (addr & (shmlba - 1)) {
1374			/*
1375			 * Round down to the nearest multiple of shmlba.
1376			 * For sane do_mmap_pgoff() parameters, avoid
1377			 * round downs that trigger nil-page and MAP_FIXED.
1378			 */
1379			if ((shmflg & SHM_RND) && addr >= shmlba)
1380				addr &= ~(shmlba - 1);
1381			else
1382#ifndef __ARCH_FORCE_SHMLBA
1383				if (addr & ~PAGE_MASK)
1384#endif
1385					goto out;
1386		}
1387
1388		flags |= MAP_FIXED;
1389	} else if ((shmflg & SHM_REMAP))
1390		goto out;
1391
1392	if (shmflg & SHM_RDONLY) {
1393		prot = PROT_READ;
1394		acc_mode = S_IRUGO;
1395		f_mode = FMODE_READ;
1396	} else {
1397		prot = PROT_READ | PROT_WRITE;
1398		acc_mode = S_IRUGO | S_IWUGO;
1399		f_mode = FMODE_READ | FMODE_WRITE;
1400	}
1401	if (shmflg & SHM_EXEC) {
1402		prot |= PROT_EXEC;
1403		acc_mode |= S_IXUGO;
1404	}
1405
1406	/*
1407	 * We cannot rely on the fs check since SYSV IPC does have an
1408	 * additional creator id...
1409	 */
1410	ns = current->nsproxy->ipc_ns;
1411	rcu_read_lock();
1412	shp = shm_obtain_object_check(ns, shmid);
1413	if (IS_ERR(shp)) {
1414		err = PTR_ERR(shp);
1415		goto out_unlock;
1416	}
1417
1418	err = -EACCES;
1419	if (ipcperms(ns, &shp->shm_perm, acc_mode))
1420		goto out_unlock;
1421
1422	err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1423	if (err)
1424		goto out_unlock;
1425
1426	ipc_lock_object(&shp->shm_perm);
1427
1428	/* check if shm_destroy() is tearing down shp */
1429	if (!ipc_valid_object(&shp->shm_perm)) {
1430		ipc_unlock_object(&shp->shm_perm);
1431		err = -EIDRM;
1432		goto out_unlock;
1433	}
1434
1435	path = shp->shm_file->f_path;
1436	path_get(&path);
1437	shp->shm_nattch++;
1438	size = i_size_read(d_inode(path.dentry));
1439	ipc_unlock_object(&shp->shm_perm);
1440	rcu_read_unlock();
1441
1442	err = -ENOMEM;
1443	sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1444	if (!sfd) {
1445		path_put(&path);
1446		goto out_nattch;
1447	}
1448
1449	file = alloc_file(&path, f_mode,
1450			  is_file_hugepages(shp->shm_file) ?
1451				&shm_file_operations_huge :
1452				&shm_file_operations);
1453	err = PTR_ERR(file);
1454	if (IS_ERR(file)) {
1455		kfree(sfd);
1456		path_put(&path);
1457		goto out_nattch;
1458	}
1459
1460	file->private_data = sfd;
1461	file->f_mapping = shp->shm_file->f_mapping;
1462	sfd->id = shp->shm_perm.id;
1463	sfd->ns = get_ipc_ns(ns);
1464	/*
1465	 * We need to take a reference to the real shm file to prevent the
1466	 * pointer from becoming stale in cases where the lifetime of the outer
1467	 * file extends beyond that of the shm segment.  It's not usually
1468	 * possible, but it can happen during remap_file_pages() emulation as
1469	 * that unmaps the memory, then does ->mmap() via file reference only.
1470	 * We'll deny the ->mmap() if the shm segment was since removed, but to
1471	 * detect shm ID reuse we need to compare the file pointers.
1472	 */
1473	sfd->file = get_file(shp->shm_file);
1474	sfd->vm_ops = NULL;
1475
1476	err = security_mmap_file(file, prot, flags);
1477	if (err)
1478		goto out_fput;
1479
1480	if (down_write_killable(&current->mm->mmap_sem)) {
1481		err = -EINTR;
1482		goto out_fput;
1483	}
1484
1485	if (addr && !(shmflg & SHM_REMAP)) {
1486		err = -EINVAL;
1487		if (addr + size < addr)
1488			goto invalid;
1489
1490		if (find_vma_intersection(current->mm, addr, addr + size))
1491			goto invalid;
1492	}
1493
1494	addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1495	*raddr = addr;
1496	err = 0;
1497	if (IS_ERR_VALUE(addr))
1498		err = (long)addr;
1499invalid:
1500	up_write(&current->mm->mmap_sem);
1501	if (populate)
1502		mm_populate(addr, populate);
1503
1504out_fput:
1505	fput(file);
1506
1507out_nattch:
1508	down_write(&shm_ids(ns).rwsem);
1509	shp = shm_lock(ns, shmid);
1510	shp->shm_nattch--;
1511	if (shm_may_destroy(ns, shp))
1512		shm_destroy(ns, shp);
1513	else
1514		shm_unlock(shp);
1515	up_write(&shm_ids(ns).rwsem);
1516	return err;
1517
1518out_unlock:
1519	rcu_read_unlock();
1520out:
1521	return err;
1522}
1523
1524SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1525{
1526	unsigned long ret;
1527	long err;
1528
1529	err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1530	if (err)
1531		return err;
1532	force_successful_syscall_return();
1533	return (long)ret;
1534}
1535
1536#ifdef CONFIG_COMPAT
1537
1538#ifndef COMPAT_SHMLBA
1539#define COMPAT_SHMLBA	SHMLBA
1540#endif
1541
1542COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1543{
1544	unsigned long ret;
1545	long err;
1546
1547	err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1548	if (err)
1549		return err;
1550	force_successful_syscall_return();
1551	return (long)ret;
1552}
1553#endif
1554
1555/*
1556 * detach and kill segment if marked destroyed.
1557 * The work is done in shm_close.
1558 */
1559long ksys_shmdt(char __user *shmaddr)
1560{
1561	struct mm_struct *mm = current->mm;
1562	struct vm_area_struct *vma;
1563	unsigned long addr = (unsigned long)shmaddr;
1564	int retval = -EINVAL;
1565#ifdef CONFIG_MMU
1566	loff_t size = 0;
1567	struct file *file;
1568	struct vm_area_struct *next;
1569#endif
1570
1571	if (addr & ~PAGE_MASK)
1572		return retval;
1573
1574	if (down_write_killable(&mm->mmap_sem))
1575		return -EINTR;
1576
1577	/*
1578	 * This function tries to be smart and unmap shm segments that
1579	 * were modified by partial mlock or munmap calls:
1580	 * - It first determines the size of the shm segment that should be
1581	 *   unmapped: It searches for a vma that is backed by shm and that
1582	 *   started at address shmaddr. It records it's size and then unmaps
1583	 *   it.
1584	 * - Then it unmaps all shm vmas that started at shmaddr and that
1585	 *   are within the initially determined size and that are from the
1586	 *   same shm segment from which we determined the size.
1587	 * Errors from do_munmap are ignored: the function only fails if
1588	 * it's called with invalid parameters or if it's called to unmap
1589	 * a part of a vma. Both calls in this function are for full vmas,
1590	 * the parameters are directly copied from the vma itself and always
1591	 * valid - therefore do_munmap cannot fail. (famous last words?)
1592	 */
1593	/*
1594	 * If it had been mremap()'d, the starting address would not
1595	 * match the usual checks anyway. So assume all vma's are
1596	 * above the starting address given.
1597	 */
1598	vma = find_vma(mm, addr);
1599
1600#ifdef CONFIG_MMU
1601	while (vma) {
1602		next = vma->vm_next;
1603
1604		/*
1605		 * Check if the starting address would match, i.e. it's
1606		 * a fragment created by mprotect() and/or munmap(), or it
1607		 * otherwise it starts at this address with no hassles.
1608		 */
1609		if ((vma->vm_ops == &shm_vm_ops) &&
1610			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1611
1612			/*
1613			 * Record the file of the shm segment being
1614			 * unmapped.  With mremap(), someone could place
1615			 * page from another segment but with equal offsets
1616			 * in the range we are unmapping.
1617			 */
1618			file = vma->vm_file;
1619			size = i_size_read(file_inode(vma->vm_file));
1620			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1621			/*
1622			 * We discovered the size of the shm segment, so
1623			 * break out of here and fall through to the next
1624			 * loop that uses the size information to stop
1625			 * searching for matching vma's.
1626			 */
1627			retval = 0;
1628			vma = next;
1629			break;
1630		}
1631		vma = next;
1632	}
1633
1634	/*
1635	 * We need look no further than the maximum address a fragment
1636	 * could possibly have landed at. Also cast things to loff_t to
1637	 * prevent overflows and make comparisons vs. equal-width types.
1638	 */
1639	size = PAGE_ALIGN(size);
1640	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1641		next = vma->vm_next;
1642
1643		/* finding a matching vma now does not alter retval */
1644		if ((vma->vm_ops == &shm_vm_ops) &&
1645		    ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1646		    (vma->vm_file == file))
1647			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1648		vma = next;
1649	}
1650
1651#else	/* CONFIG_MMU */
1652	/* under NOMMU conditions, the exact address to be destroyed must be
1653	 * given
1654	 */
1655	if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1656		do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1657		retval = 0;
1658	}
1659
1660#endif
1661
1662	up_write(&mm->mmap_sem);
1663	return retval;
1664}
1665
1666SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1667{
1668	return ksys_shmdt(shmaddr);
1669}
1670
1671#ifdef CONFIG_PROC_FS
1672static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1673{
1674	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1675	struct user_namespace *user_ns = seq_user_ns(s);
1676	struct kern_ipc_perm *ipcp = it;
1677	struct shmid_kernel *shp;
1678	unsigned long rss = 0, swp = 0;
1679
1680	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1681	shm_add_rss_swap(shp, &rss, &swp);
1682
1683#if BITS_PER_LONG <= 32
1684#define SIZE_SPEC "%10lu"
1685#else
1686#define SIZE_SPEC "%21lu"
1687#endif
1688
1689	seq_printf(s,
1690		   "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1691		   "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1692		   SIZE_SPEC " " SIZE_SPEC "\n",
1693		   shp->shm_perm.key,
1694		   shp->shm_perm.id,
1695		   shp->shm_perm.mode,
1696		   shp->shm_segsz,
1697		   pid_nr_ns(shp->shm_cprid, pid_ns),
1698		   pid_nr_ns(shp->shm_lprid, pid_ns),
1699		   shp->shm_nattch,
1700		   from_kuid_munged(user_ns, shp->shm_perm.uid),
1701		   from_kgid_munged(user_ns, shp->shm_perm.gid),
1702		   from_kuid_munged(user_ns, shp->shm_perm.cuid),
1703		   from_kgid_munged(user_ns, shp->shm_perm.cgid),
1704		   shp->shm_atim,
1705		   shp->shm_dtim,
1706		   shp->shm_ctim,
1707		   rss * PAGE_SIZE,
1708		   swp * PAGE_SIZE);
1709
1710	return 0;
1711}
1712#endif
1713