shm.c revision eae04d25
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);
202err:
203	rcu_read_unlock();
204	/*
205	 * Callers of shm_lock() must validate the status of the returned ipc
206	 * object pointer and error out as appropriate.
207	 */
208	return (void *)ipcp;
209}
210
211static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
212{
213	rcu_read_lock();
214	ipc_lock_object(&ipcp->shm_perm);
215}
216
217static void shm_rcu_free(struct rcu_head *head)
218{
219	struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
220							rcu);
221	struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
222							shm_perm);
223	security_shm_free(&shp->shm_perm);
224	kvfree(shp);
225}
226
227static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
228{
229	list_del(&s->shm_clist);
230	ipc_rmid(&shm_ids(ns), &s->shm_perm);
231}
232
233
234static int __shm_open(struct vm_area_struct *vma)
235{
236	struct file *file = vma->vm_file;
237	struct shm_file_data *sfd = shm_file_data(file);
238	struct shmid_kernel *shp;
239
240	shp = shm_lock(sfd->ns, sfd->id);
241
242	if (IS_ERR(shp))
243		return PTR_ERR(shp);
244
245	if (shp->shm_file != sfd->file) {
246		/* ID was reused */
247		shm_unlock(shp);
248		return -EINVAL;
249	}
250
251	shp->shm_atim = ktime_get_real_seconds();
252	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
253	shp->shm_nattch++;
254	shm_unlock(shp);
255	return 0;
256}
257
258/* This is called by fork, once for every shm attach. */
259static void shm_open(struct vm_area_struct *vma)
260{
261	int err = __shm_open(vma);
262	/*
263	 * We raced in the idr lookup or with shm_destroy().
264	 * Either way, the ID is busted.
265	 */
266	WARN_ON_ONCE(err);
267}
268
269/*
270 * shm_destroy - free the struct shmid_kernel
271 *
272 * @ns: namespace
273 * @shp: struct to free
274 *
275 * It has to be called with shp and shm_ids.rwsem (writer) locked,
276 * but returns with shp unlocked and freed.
277 */
278static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
279{
280	struct file *shm_file;
281
282	shm_file = shp->shm_file;
283	shp->shm_file = NULL;
284	ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
285	shm_rmid(ns, shp);
286	shm_unlock(shp);
287	if (!is_file_hugepages(shm_file))
288		shmem_lock(shm_file, 0, shp->mlock_user);
289	else if (shp->mlock_user)
290		user_shm_unlock(i_size_read(file_inode(shm_file)),
291				shp->mlock_user);
292	fput(shm_file);
293	ipc_update_pid(&shp->shm_cprid, NULL);
294	ipc_update_pid(&shp->shm_lprid, NULL);
295	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
296}
297
298/*
299 * shm_may_destroy - identifies whether shm segment should be destroyed now
300 *
301 * Returns true if and only if there are no active users of the segment and
302 * one of the following is true:
303 *
304 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
305 *
306 * 2) sysctl kernel.shm_rmid_forced is set to 1.
307 */
308static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
309{
310	return (shp->shm_nattch == 0) &&
311	       (ns->shm_rmid_forced ||
312		(shp->shm_perm.mode & SHM_DEST));
313}
314
315/*
316 * remove the attach descriptor vma.
317 * free memory for segment if it is marked destroyed.
318 * The descriptor has already been removed from the current->mm->mmap list
319 * and will later be kfree()d.
320 */
321static void shm_close(struct vm_area_struct *vma)
322{
323	struct file *file = vma->vm_file;
324	struct shm_file_data *sfd = shm_file_data(file);
325	struct shmid_kernel *shp;
326	struct ipc_namespace *ns = sfd->ns;
327
328	down_write(&shm_ids(ns).rwsem);
329	/* remove from the list of attaches of the shm segment */
330	shp = shm_lock(ns, sfd->id);
331
332	/*
333	 * We raced in the idr lookup or with shm_destroy().
334	 * Either way, the ID is busted.
335	 */
336	if (WARN_ON_ONCE(IS_ERR(shp)))
337		goto done; /* no-op */
338
339	ipc_update_pid(&shp->shm_lprid, task_tgid(current));
340	shp->shm_dtim = ktime_get_real_seconds();
341	shp->shm_nattch--;
342	if (shm_may_destroy(ns, shp))
343		shm_destroy(ns, shp);
344	else
345		shm_unlock(shp);
346done:
347	up_write(&shm_ids(ns).rwsem);
348}
349
350/* Called with ns->shm_ids(ns).rwsem locked */
351static int shm_try_destroy_orphaned(int id, void *p, void *data)
352{
353	struct ipc_namespace *ns = data;
354	struct kern_ipc_perm *ipcp = p;
355	struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
356
357	/*
358	 * We want to destroy segments without users and with already
359	 * exit'ed originating process.
360	 *
361	 * As shp->* are changed under rwsem, it's safe to skip shp locking.
362	 */
363	if (shp->shm_creator != NULL)
364		return 0;
365
366	if (shm_may_destroy(ns, shp)) {
367		shm_lock_by_ptr(shp);
368		shm_destroy(ns, shp);
369	}
370	return 0;
371}
372
373void shm_destroy_orphaned(struct ipc_namespace *ns)
374{
375	down_write(&shm_ids(ns).rwsem);
376	if (shm_ids(ns).in_use)
377		idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
378	up_write(&shm_ids(ns).rwsem);
379}
380
381/* Locking assumes this will only be called with task == current */
382void exit_shm(struct task_struct *task)
383{
384	struct ipc_namespace *ns = task->nsproxy->ipc_ns;
385	struct shmid_kernel *shp, *n;
386
387	if (list_empty(&task->sysvshm.shm_clist))
388		return;
389
390	/*
391	 * If kernel.shm_rmid_forced is not set then only keep track of
392	 * which shmids are orphaned, so that a later set of the sysctl
393	 * can clean them up.
394	 */
395	if (!ns->shm_rmid_forced) {
396		down_read(&shm_ids(ns).rwsem);
397		list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
398			shp->shm_creator = NULL;
399		/*
400		 * Only under read lock but we are only called on current
401		 * so no entry on the list will be shared.
402		 */
403		list_del(&task->sysvshm.shm_clist);
404		up_read(&shm_ids(ns).rwsem);
405		return;
406	}
407
408	/*
409	 * Destroy all already created segments, that were not yet mapped,
410	 * and mark any mapped as orphan to cover the sysctl toggling.
411	 * Destroy is skipped if shm_may_destroy() returns false.
412	 */
413	down_write(&shm_ids(ns).rwsem);
414	list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
415		shp->shm_creator = NULL;
416
417		if (shm_may_destroy(ns, shp)) {
418			shm_lock_by_ptr(shp);
419			shm_destroy(ns, shp);
420		}
421	}
422
423	/* Remove the list head from any segments still attached. */
424	list_del(&task->sysvshm.shm_clist);
425	up_write(&shm_ids(ns).rwsem);
426}
427
428static vm_fault_t shm_fault(struct vm_fault *vmf)
429{
430	struct file *file = vmf->vma->vm_file;
431	struct shm_file_data *sfd = shm_file_data(file);
432
433	return sfd->vm_ops->fault(vmf);
434}
435
436static int shm_split(struct vm_area_struct *vma, unsigned long addr)
437{
438	struct file *file = vma->vm_file;
439	struct shm_file_data *sfd = shm_file_data(file);
440
441	if (sfd->vm_ops->split)
442		return sfd->vm_ops->split(vma, addr);
443
444	return 0;
445}
446
447static unsigned long shm_pagesize(struct vm_area_struct *vma)
448{
449	struct file *file = vma->vm_file;
450	struct shm_file_data *sfd = shm_file_data(file);
451
452	if (sfd->vm_ops->pagesize)
453		return sfd->vm_ops->pagesize(vma);
454
455	return PAGE_SIZE;
456}
457
458#ifdef CONFIG_NUMA
459static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
460{
461	struct file *file = vma->vm_file;
462	struct shm_file_data *sfd = shm_file_data(file);
463	int err = 0;
464
465	if (sfd->vm_ops->set_policy)
466		err = sfd->vm_ops->set_policy(vma, new);
467	return err;
468}
469
470static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
471					unsigned long addr)
472{
473	struct file *file = vma->vm_file;
474	struct shm_file_data *sfd = shm_file_data(file);
475	struct mempolicy *pol = NULL;
476
477	if (sfd->vm_ops->get_policy)
478		pol = sfd->vm_ops->get_policy(vma, addr);
479	else if (vma->vm_policy)
480		pol = vma->vm_policy;
481
482	return pol;
483}
484#endif
485
486static int shm_mmap(struct file *file, struct vm_area_struct *vma)
487{
488	struct shm_file_data *sfd = shm_file_data(file);
489	int ret;
490
491	/*
492	 * In case of remap_file_pages() emulation, the file can represent an
493	 * IPC ID that was removed, and possibly even reused by another shm
494	 * segment already.  Propagate this case as an error to caller.
495	 */
496	ret = __shm_open(vma);
497	if (ret)
498		return ret;
499
500	ret = call_mmap(sfd->file, vma);
501	if (ret) {
502		shm_close(vma);
503		return ret;
504	}
505	sfd->vm_ops = vma->vm_ops;
506#ifdef CONFIG_MMU
507	WARN_ON(!sfd->vm_ops->fault);
508#endif
509	vma->vm_ops = &shm_vm_ops;
510	return 0;
511}
512
513static int shm_release(struct inode *ino, struct file *file)
514{
515	struct shm_file_data *sfd = shm_file_data(file);
516
517	put_ipc_ns(sfd->ns);
518	fput(sfd->file);
519	shm_file_data(file) = NULL;
520	kfree(sfd);
521	return 0;
522}
523
524static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
525{
526	struct shm_file_data *sfd = shm_file_data(file);
527
528	if (!sfd->file->f_op->fsync)
529		return -EINVAL;
530	return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
531}
532
533static long shm_fallocate(struct file *file, int mode, loff_t offset,
534			  loff_t len)
535{
536	struct shm_file_data *sfd = shm_file_data(file);
537
538	if (!sfd->file->f_op->fallocate)
539		return -EOPNOTSUPP;
540	return sfd->file->f_op->fallocate(file, mode, offset, len);
541}
542
543static unsigned long shm_get_unmapped_area(struct file *file,
544	unsigned long addr, unsigned long len, unsigned long pgoff,
545	unsigned long flags)
546{
547	struct shm_file_data *sfd = shm_file_data(file);
548
549	return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
550						pgoff, flags);
551}
552
553static const struct file_operations shm_file_operations = {
554	.mmap		= shm_mmap,
555	.fsync		= shm_fsync,
556	.release	= shm_release,
557	.get_unmapped_area	= shm_get_unmapped_area,
558	.llseek		= noop_llseek,
559	.fallocate	= shm_fallocate,
560};
561
562/*
563 * shm_file_operations_huge is now identical to shm_file_operations,
564 * but we keep it distinct for the sake of is_file_shm_hugepages().
565 */
566static const struct file_operations shm_file_operations_huge = {
567	.mmap		= shm_mmap,
568	.fsync		= shm_fsync,
569	.release	= shm_release,
570	.get_unmapped_area	= shm_get_unmapped_area,
571	.llseek		= noop_llseek,
572	.fallocate	= shm_fallocate,
573};
574
575bool is_file_shm_hugepages(struct file *file)
576{
577	return file->f_op == &shm_file_operations_huge;
578}
579
580static const struct vm_operations_struct shm_vm_ops = {
581	.open	= shm_open,	/* callback for a new vm-area open */
582	.close	= shm_close,	/* callback for when the vm-area is released */
583	.fault	= shm_fault,
584	.split	= shm_split,
585	.pagesize = shm_pagesize,
586#if defined(CONFIG_NUMA)
587	.set_policy = shm_set_policy,
588	.get_policy = shm_get_policy,
589#endif
590};
591
592/**
593 * newseg - Create a new shared memory segment
594 * @ns: namespace
595 * @params: ptr to the structure that contains key, size and shmflg
596 *
597 * Called with shm_ids.rwsem held as a writer.
598 */
599static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
600{
601	key_t key = params->key;
602	int shmflg = params->flg;
603	size_t size = params->u.size;
604	int error;
605	struct shmid_kernel *shp;
606	size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
607	struct file *file;
608	char name[13];
609	vm_flags_t acctflag = 0;
610
611	if (size < SHMMIN || size > ns->shm_ctlmax)
612		return -EINVAL;
613
614	if (numpages << PAGE_SHIFT < size)
615		return -ENOSPC;
616
617	if (ns->shm_tot + numpages < ns->shm_tot ||
618			ns->shm_tot + numpages > ns->shm_ctlall)
619		return -ENOSPC;
620
621	shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
622	if (unlikely(!shp))
623		return -ENOMEM;
624
625	shp->shm_perm.key = key;
626	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
627	shp->mlock_user = NULL;
628
629	shp->shm_perm.security = NULL;
630	error = security_shm_alloc(&shp->shm_perm);
631	if (error) {
632		kvfree(shp);
633		return error;
634	}
635
636	sprintf(name, "SYSV%08x", key);
637	if (shmflg & SHM_HUGETLB) {
638		struct hstate *hs;
639		size_t hugesize;
640
641		hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
642		if (!hs) {
643			error = -EINVAL;
644			goto no_file;
645		}
646		hugesize = ALIGN(size, huge_page_size(hs));
647
648		/* hugetlb_file_setup applies strict accounting */
649		if (shmflg & SHM_NORESERVE)
650			acctflag = VM_NORESERVE;
651		file = hugetlb_file_setup(name, hugesize, acctflag,
652				  &shp->mlock_user, HUGETLB_SHMFS_INODE,
653				(shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
654	} else {
655		/*
656		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
657		 * if it's asked for.
658		 */
659		if  ((shmflg & SHM_NORESERVE) &&
660				sysctl_overcommit_memory != OVERCOMMIT_NEVER)
661			acctflag = VM_NORESERVE;
662		file = shmem_kernel_file_setup(name, size, acctflag);
663	}
664	error = PTR_ERR(file);
665	if (IS_ERR(file))
666		goto no_file;
667
668	shp->shm_cprid = get_pid(task_tgid(current));
669	shp->shm_lprid = NULL;
670	shp->shm_atim = shp->shm_dtim = 0;
671	shp->shm_ctim = ktime_get_real_seconds();
672	shp->shm_segsz = size;
673	shp->shm_nattch = 0;
674	shp->shm_file = file;
675	shp->shm_creator = current;
676
677	/* ipc_addid() locks shp upon success. */
678	error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
679	if (error < 0)
680		goto no_id;
681
682	list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
683
684	/*
685	 * shmid gets reported as "inode#" in /proc/pid/maps.
686	 * proc-ps tools use this. Changing this will break them.
687	 */
688	file_inode(file)->i_ino = shp->shm_perm.id;
689
690	ns->shm_tot += numpages;
691	error = shp->shm_perm.id;
692
693	ipc_unlock_object(&shp->shm_perm);
694	rcu_read_unlock();
695	return error;
696
697no_id:
698	ipc_update_pid(&shp->shm_cprid, NULL);
699	ipc_update_pid(&shp->shm_lprid, NULL);
700	if (is_file_hugepages(file) && shp->mlock_user)
701		user_shm_unlock(size, shp->mlock_user);
702	fput(file);
703	ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
704	return error;
705no_file:
706	call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
707	return error;
708}
709
710/*
711 * Called with shm_ids.rwsem and ipcp locked.
712 */
713static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
714				struct ipc_params *params)
715{
716	struct shmid_kernel *shp;
717
718	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
719	if (shp->shm_segsz < params->u.size)
720		return -EINVAL;
721
722	return 0;
723}
724
725long ksys_shmget(key_t key, size_t size, int shmflg)
726{
727	struct ipc_namespace *ns;
728	static const struct ipc_ops shm_ops = {
729		.getnew = newseg,
730		.associate = security_shm_associate,
731		.more_checks = shm_more_checks,
732	};
733	struct ipc_params shm_params;
734
735	ns = current->nsproxy->ipc_ns;
736
737	shm_params.key = key;
738	shm_params.flg = shmflg;
739	shm_params.u.size = size;
740
741	return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
742}
743
744SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
745{
746	return ksys_shmget(key, size, shmflg);
747}
748
749static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
750{
751	switch (version) {
752	case IPC_64:
753		return copy_to_user(buf, in, sizeof(*in));
754	case IPC_OLD:
755	    {
756		struct shmid_ds out;
757
758		memset(&out, 0, sizeof(out));
759		ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
760		out.shm_segsz	= in->shm_segsz;
761		out.shm_atime	= in->shm_atime;
762		out.shm_dtime	= in->shm_dtime;
763		out.shm_ctime	= in->shm_ctime;
764		out.shm_cpid	= in->shm_cpid;
765		out.shm_lpid	= in->shm_lpid;
766		out.shm_nattch	= in->shm_nattch;
767
768		return copy_to_user(buf, &out, sizeof(out));
769	    }
770	default:
771		return -EINVAL;
772	}
773}
774
775static inline unsigned long
776copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
777{
778	switch (version) {
779	case IPC_64:
780		if (copy_from_user(out, buf, sizeof(*out)))
781			return -EFAULT;
782		return 0;
783	case IPC_OLD:
784	    {
785		struct shmid_ds tbuf_old;
786
787		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
788			return -EFAULT;
789
790		out->shm_perm.uid	= tbuf_old.shm_perm.uid;
791		out->shm_perm.gid	= tbuf_old.shm_perm.gid;
792		out->shm_perm.mode	= tbuf_old.shm_perm.mode;
793
794		return 0;
795	    }
796	default:
797		return -EINVAL;
798	}
799}
800
801static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
802{
803	switch (version) {
804	case IPC_64:
805		return copy_to_user(buf, in, sizeof(*in));
806	case IPC_OLD:
807	    {
808		struct shminfo out;
809
810		if (in->shmmax > INT_MAX)
811			out.shmmax = INT_MAX;
812		else
813			out.shmmax = (int)in->shmmax;
814
815		out.shmmin	= in->shmmin;
816		out.shmmni	= in->shmmni;
817		out.shmseg	= in->shmseg;
818		out.shmall	= in->shmall;
819
820		return copy_to_user(buf, &out, sizeof(out));
821	    }
822	default:
823		return -EINVAL;
824	}
825}
826
827/*
828 * Calculate and add used RSS and swap pages of a shm.
829 * Called with shm_ids.rwsem held as a reader
830 */
831static void shm_add_rss_swap(struct shmid_kernel *shp,
832	unsigned long *rss_add, unsigned long *swp_add)
833{
834	struct inode *inode;
835
836	inode = file_inode(shp->shm_file);
837
838	if (is_file_hugepages(shp->shm_file)) {
839		struct address_space *mapping = inode->i_mapping;
840		struct hstate *h = hstate_file(shp->shm_file);
841		*rss_add += pages_per_huge_page(h) * mapping->nrpages;
842	} else {
843#ifdef CONFIG_SHMEM
844		struct shmem_inode_info *info = SHMEM_I(inode);
845
846		spin_lock_irq(&info->lock);
847		*rss_add += inode->i_mapping->nrpages;
848		*swp_add += info->swapped;
849		spin_unlock_irq(&info->lock);
850#else
851		*rss_add += inode->i_mapping->nrpages;
852#endif
853	}
854}
855
856/*
857 * Called with shm_ids.rwsem held as a reader
858 */
859static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
860		unsigned long *swp)
861{
862	int next_id;
863	int total, in_use;
864
865	*rss = 0;
866	*swp = 0;
867
868	in_use = shm_ids(ns).in_use;
869
870	for (total = 0, next_id = 0; total < in_use; next_id++) {
871		struct kern_ipc_perm *ipc;
872		struct shmid_kernel *shp;
873
874		ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
875		if (ipc == NULL)
876			continue;
877		shp = container_of(ipc, struct shmid_kernel, shm_perm);
878
879		shm_add_rss_swap(shp, rss, swp);
880
881		total++;
882	}
883}
884
885/*
886 * This function handles some shmctl commands which require the rwsem
887 * to be held in write mode.
888 * NOTE: no locks must be held, the rwsem is taken inside this function.
889 */
890static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
891		       struct shmid64_ds *shmid64)
892{
893	struct kern_ipc_perm *ipcp;
894	struct shmid_kernel *shp;
895	int err;
896
897	down_write(&shm_ids(ns).rwsem);
898	rcu_read_lock();
899
900	ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
901				      &shmid64->shm_perm, 0);
902	if (IS_ERR(ipcp)) {
903		err = PTR_ERR(ipcp);
904		goto out_unlock1;
905	}
906
907	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
908
909	err = security_shm_shmctl(&shp->shm_perm, cmd);
910	if (err)
911		goto out_unlock1;
912
913	switch (cmd) {
914	case IPC_RMID:
915		ipc_lock_object(&shp->shm_perm);
916		/* do_shm_rmid unlocks the ipc object and rcu */
917		do_shm_rmid(ns, ipcp);
918		goto out_up;
919	case IPC_SET:
920		ipc_lock_object(&shp->shm_perm);
921		err = ipc_update_perm(&shmid64->shm_perm, ipcp);
922		if (err)
923			goto out_unlock0;
924		shp->shm_ctim = ktime_get_real_seconds();
925		break;
926	default:
927		err = -EINVAL;
928		goto out_unlock1;
929	}
930
931out_unlock0:
932	ipc_unlock_object(&shp->shm_perm);
933out_unlock1:
934	rcu_read_unlock();
935out_up:
936	up_write(&shm_ids(ns).rwsem);
937	return err;
938}
939
940static int shmctl_ipc_info(struct ipc_namespace *ns,
941			   struct shminfo64 *shminfo)
942{
943	int err = security_shm_shmctl(NULL, IPC_INFO);
944	if (!err) {
945		memset(shminfo, 0, sizeof(*shminfo));
946		shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
947		shminfo->shmmax = ns->shm_ctlmax;
948		shminfo->shmall = ns->shm_ctlall;
949		shminfo->shmmin = SHMMIN;
950		down_read(&shm_ids(ns).rwsem);
951		err = ipc_get_maxid(&shm_ids(ns));
952		up_read(&shm_ids(ns).rwsem);
953		if (err < 0)
954			err = 0;
955	}
956	return err;
957}
958
959static int shmctl_shm_info(struct ipc_namespace *ns,
960			   struct shm_info *shm_info)
961{
962	int err = security_shm_shmctl(NULL, SHM_INFO);
963	if (!err) {
964		memset(shm_info, 0, sizeof(*shm_info));
965		down_read(&shm_ids(ns).rwsem);
966		shm_info->used_ids = shm_ids(ns).in_use;
967		shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
968		shm_info->shm_tot = ns->shm_tot;
969		shm_info->swap_attempts = 0;
970		shm_info->swap_successes = 0;
971		err = ipc_get_maxid(&shm_ids(ns));
972		up_read(&shm_ids(ns).rwsem);
973		if (err < 0)
974			err = 0;
975	}
976	return err;
977}
978
979static int shmctl_stat(struct ipc_namespace *ns, int shmid,
980			int cmd, struct shmid64_ds *tbuf)
981{
982	struct shmid_kernel *shp;
983	int err;
984
985	memset(tbuf, 0, sizeof(*tbuf));
986
987	rcu_read_lock();
988	if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
989		shp = shm_obtain_object(ns, shmid);
990		if (IS_ERR(shp)) {
991			err = PTR_ERR(shp);
992			goto out_unlock;
993		}
994	} else { /* IPC_STAT */
995		shp = shm_obtain_object_check(ns, shmid);
996		if (IS_ERR(shp)) {
997			err = PTR_ERR(shp);
998			goto out_unlock;
999		}
1000	}
1001
1002	/*
1003	 * Semantically SHM_STAT_ANY ought to be identical to
1004	 * that functionality provided by the /proc/sysvipc/
1005	 * interface. As such, only audit these calls and
1006	 * do not do traditional S_IRUGO permission checks on
1007	 * the ipc object.
1008	 */
1009	if (cmd == SHM_STAT_ANY)
1010		audit_ipc_obj(&shp->shm_perm);
1011	else {
1012		err = -EACCES;
1013		if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1014			goto out_unlock;
1015	}
1016
1017	err = security_shm_shmctl(&shp->shm_perm, cmd);
1018	if (err)
1019		goto out_unlock;
1020
1021	ipc_lock_object(&shp->shm_perm);
1022
1023	if (!ipc_valid_object(&shp->shm_perm)) {
1024		ipc_unlock_object(&shp->shm_perm);
1025		err = -EIDRM;
1026		goto out_unlock;
1027	}
1028
1029	kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1030	tbuf->shm_segsz	= shp->shm_segsz;
1031	tbuf->shm_atime	= shp->shm_atim;
1032	tbuf->shm_dtime	= shp->shm_dtim;
1033	tbuf->shm_ctime	= shp->shm_ctim;
1034#ifndef CONFIG_64BIT
1035	tbuf->shm_atime_high = shp->shm_atim >> 32;
1036	tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1037	tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1038#endif
1039	tbuf->shm_cpid	= pid_vnr(shp->shm_cprid);
1040	tbuf->shm_lpid	= pid_vnr(shp->shm_lprid);
1041	tbuf->shm_nattch = shp->shm_nattch;
1042
1043	if (cmd == IPC_STAT) {
1044		/*
1045		 * As defined in SUS:
1046		 * Return 0 on success
1047		 */
1048		err = 0;
1049	} else {
1050		/*
1051		 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1052		 * Return the full id, including the sequence number
1053		 */
1054		err = shp->shm_perm.id;
1055	}
1056
1057	ipc_unlock_object(&shp->shm_perm);
1058out_unlock:
1059	rcu_read_unlock();
1060	return err;
1061}
1062
1063static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1064{
1065	struct shmid_kernel *shp;
1066	struct file *shm_file;
1067	int err;
1068
1069	rcu_read_lock();
1070	shp = shm_obtain_object_check(ns, shmid);
1071	if (IS_ERR(shp)) {
1072		err = PTR_ERR(shp);
1073		goto out_unlock1;
1074	}
1075
1076	audit_ipc_obj(&(shp->shm_perm));
1077	err = security_shm_shmctl(&shp->shm_perm, cmd);
1078	if (err)
1079		goto out_unlock1;
1080
1081	ipc_lock_object(&shp->shm_perm);
1082
1083	/* check if shm_destroy() is tearing down shp */
1084	if (!ipc_valid_object(&shp->shm_perm)) {
1085		err = -EIDRM;
1086		goto out_unlock0;
1087	}
1088
1089	if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1090		kuid_t euid = current_euid();
1091
1092		if (!uid_eq(euid, shp->shm_perm.uid) &&
1093		    !uid_eq(euid, shp->shm_perm.cuid)) {
1094			err = -EPERM;
1095			goto out_unlock0;
1096		}
1097		if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1098			err = -EPERM;
1099			goto out_unlock0;
1100		}
1101	}
1102
1103	shm_file = shp->shm_file;
1104	if (is_file_hugepages(shm_file))
1105		goto out_unlock0;
1106
1107	if (cmd == SHM_LOCK) {
1108		struct user_struct *user = current_user();
1109
1110		err = shmem_lock(shm_file, 1, user);
1111		if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1112			shp->shm_perm.mode |= SHM_LOCKED;
1113			shp->mlock_user = user;
1114		}
1115		goto out_unlock0;
1116	}
1117
1118	/* SHM_UNLOCK */
1119	if (!(shp->shm_perm.mode & SHM_LOCKED))
1120		goto out_unlock0;
1121	shmem_lock(shm_file, 0, shp->mlock_user);
1122	shp->shm_perm.mode &= ~SHM_LOCKED;
1123	shp->mlock_user = NULL;
1124	get_file(shm_file);
1125	ipc_unlock_object(&shp->shm_perm);
1126	rcu_read_unlock();
1127	shmem_unlock_mapping(shm_file->f_mapping);
1128
1129	fput(shm_file);
1130	return err;
1131
1132out_unlock0:
1133	ipc_unlock_object(&shp->shm_perm);
1134out_unlock1:
1135	rcu_read_unlock();
1136	return err;
1137}
1138
1139long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1140{
1141	int err, version;
1142	struct ipc_namespace *ns;
1143	struct shmid64_ds sem64;
1144
1145	if (cmd < 0 || shmid < 0)
1146		return -EINVAL;
1147
1148	version = ipc_parse_version(&cmd);
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);
1197}
1198
1199#ifdef CONFIG_COMPAT
1200
1201struct compat_shmid_ds {
1202	struct compat_ipc_perm shm_perm;
1203	int shm_segsz;
1204	compat_time_t shm_atime;
1205	compat_time_t shm_dtime;
1206	compat_time_t shm_ctime;
1207	compat_ipc_pid_t shm_cpid;
1208	compat_ipc_pid_t shm_lpid;
1209	unsigned short shm_nattch;
1210	unsigned short shm_unused;
1211	compat_uptr_t shm_unused2;
1212	compat_uptr_t shm_unused3;
1213};
1214
1215struct compat_shminfo64 {
1216	compat_ulong_t shmmax;
1217	compat_ulong_t shmmin;
1218	compat_ulong_t shmmni;
1219	compat_ulong_t shmseg;
1220	compat_ulong_t shmall;
1221	compat_ulong_t __unused1;
1222	compat_ulong_t __unused2;
1223	compat_ulong_t __unused3;
1224	compat_ulong_t __unused4;
1225};
1226
1227struct compat_shm_info {
1228	compat_int_t used_ids;
1229	compat_ulong_t shm_tot, shm_rss, shm_swp;
1230	compat_ulong_t swap_attempts, swap_successes;
1231};
1232
1233static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1234					int version)
1235{
1236	if (in->shmmax > INT_MAX)
1237		in->shmmax = INT_MAX;
1238	if (version == IPC_64) {
1239		struct compat_shminfo64 info;
1240		memset(&info, 0, sizeof(info));
1241		info.shmmax = in->shmmax;
1242		info.shmmin = in->shmmin;
1243		info.shmmni = in->shmmni;
1244		info.shmseg = in->shmseg;
1245		info.shmall = in->shmall;
1246		return copy_to_user(buf, &info, sizeof(info));
1247	} else {
1248		struct shminfo info;
1249		memset(&info, 0, sizeof(info));
1250		info.shmmax = in->shmmax;
1251		info.shmmin = in->shmmin;
1252		info.shmmni = in->shmmni;
1253		info.shmseg = in->shmseg;
1254		info.shmall = in->shmall;
1255		return copy_to_user(buf, &info, sizeof(info));
1256	}
1257}
1258
1259static int put_compat_shm_info(struct shm_info *ip,
1260				struct compat_shm_info __user *uip)
1261{
1262	struct compat_shm_info info;
1263
1264	memset(&info, 0, sizeof(info));
1265	info.used_ids = ip->used_ids;
1266	info.shm_tot = ip->shm_tot;
1267	info.shm_rss = ip->shm_rss;
1268	info.shm_swp = ip->shm_swp;
1269	info.swap_attempts = ip->swap_attempts;
1270	info.swap_successes = ip->swap_successes;
1271	return copy_to_user(uip, &info, sizeof(info));
1272}
1273
1274static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1275					int version)
1276{
1277	if (version == IPC_64) {
1278		struct compat_shmid64_ds v;
1279		memset(&v, 0, sizeof(v));
1280		to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1281		v.shm_atime	 = lower_32_bits(in->shm_atime);
1282		v.shm_atime_high = upper_32_bits(in->shm_atime);
1283		v.shm_dtime	 = lower_32_bits(in->shm_dtime);
1284		v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1285		v.shm_ctime	 = lower_32_bits(in->shm_ctime);
1286		v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1287		v.shm_segsz = in->shm_segsz;
1288		v.shm_nattch = in->shm_nattch;
1289		v.shm_cpid = in->shm_cpid;
1290		v.shm_lpid = in->shm_lpid;
1291		return copy_to_user(buf, &v, sizeof(v));
1292	} else {
1293		struct compat_shmid_ds v;
1294		memset(&v, 0, sizeof(v));
1295		to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1296		v.shm_perm.key = in->shm_perm.key;
1297		v.shm_atime = in->shm_atime;
1298		v.shm_dtime = in->shm_dtime;
1299		v.shm_ctime = in->shm_ctime;
1300		v.shm_segsz = in->shm_segsz;
1301		v.shm_nattch = in->shm_nattch;
1302		v.shm_cpid = in->shm_cpid;
1303		v.shm_lpid = in->shm_lpid;
1304		return copy_to_user(buf, &v, sizeof(v));
1305	}
1306}
1307
1308static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1309					int version)
1310{
1311	memset(out, 0, sizeof(*out));
1312	if (version == IPC_64) {
1313		struct compat_shmid64_ds __user *p = buf;
1314		return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1315	} else {
1316		struct compat_shmid_ds __user *p = buf;
1317		return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1318	}
1319}
1320
1321long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1322{
1323	struct ipc_namespace *ns;
1324	struct shmid64_ds sem64;
1325	int version = compat_ipc_parse_version(&cmd);
1326	int err;
1327
1328	ns = current->nsproxy->ipc_ns;
1329
1330	if (cmd < 0 || shmid < 0)
1331		return -EINVAL;
1332
1333	switch (cmd) {
1334	case IPC_INFO: {
1335		struct shminfo64 shminfo;
1336		err = shmctl_ipc_info(ns, &shminfo);
1337		if (err < 0)
1338			return err;
1339		if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1340			err = -EFAULT;
1341		return err;
1342	}
1343	case SHM_INFO: {
1344		struct shm_info shm_info;
1345		err = shmctl_shm_info(ns, &shm_info);
1346		if (err < 0)
1347			return err;
1348		if (put_compat_shm_info(&shm_info, uptr))
1349			err = -EFAULT;
1350		return err;
1351	}
1352	case IPC_STAT:
1353	case SHM_STAT_ANY:
1354	case SHM_STAT:
1355		err = shmctl_stat(ns, shmid, cmd, &sem64);
1356		if (err < 0)
1357			return err;
1358		if (copy_compat_shmid_to_user(uptr, &sem64, version))
1359			err = -EFAULT;
1360		return err;
1361
1362	case IPC_SET:
1363		if (copy_compat_shmid_from_user(&sem64, uptr, version))
1364			return -EFAULT;
1365		/* fallthru */
1366	case IPC_RMID:
1367		return shmctl_down(ns, shmid, cmd, &sem64);
1368	case SHM_LOCK:
1369	case SHM_UNLOCK:
1370		return shmctl_do_lock(ns, shmid, cmd);
1371		break;
1372	default:
1373		return -EINVAL;
1374	}
1375	return err;
1376}
1377
1378COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1379{
1380	return compat_ksys_shmctl(shmid, cmd, uptr);
1381}
1382#endif
1383
1384/*
1385 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1386 *
1387 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1388 * "raddr" thing points to kernel space, and there has to be a wrapper around
1389 * this.
1390 */
1391long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1392	      ulong *raddr, unsigned long shmlba)
1393{
1394	struct shmid_kernel *shp;
1395	unsigned long addr = (unsigned long)shmaddr;
1396	unsigned long size;
1397	struct file *file, *base;
1398	int    err;
1399	unsigned long flags = MAP_SHARED;
1400	unsigned long prot;
1401	int acc_mode;
1402	struct ipc_namespace *ns;
1403	struct shm_file_data *sfd;
1404	int f_flags;
1405	unsigned long populate = 0;
1406
1407	err = -EINVAL;
1408	if (shmid < 0)
1409		goto out;
1410
1411	if (addr) {
1412		if (addr & (shmlba - 1)) {
1413			if (shmflg & SHM_RND) {
1414				addr &= ~(shmlba - 1);  /* round down */
1415
1416				/*
1417				 * Ensure that the round-down is non-nil
1418				 * when remapping. This can happen for
1419				 * cases when addr < shmlba.
1420				 */
1421				if (!addr && (shmflg & SHM_REMAP))
1422					goto out;
1423			} else
1424#ifndef __ARCH_FORCE_SHMLBA
1425				if (addr & ~PAGE_MASK)
1426#endif
1427					goto out;
1428		}
1429
1430		flags |= MAP_FIXED;
1431	} else if ((shmflg & SHM_REMAP))
1432		goto out;
1433
1434	if (shmflg & SHM_RDONLY) {
1435		prot = PROT_READ;
1436		acc_mode = S_IRUGO;
1437		f_flags = O_RDONLY;
1438	} else {
1439		prot = PROT_READ | PROT_WRITE;
1440		acc_mode = S_IRUGO | S_IWUGO;
1441		f_flags = O_RDWR;
1442	}
1443	if (shmflg & SHM_EXEC) {
1444		prot |= PROT_EXEC;
1445		acc_mode |= S_IXUGO;
1446	}
1447
1448	/*
1449	 * We cannot rely on the fs check since SYSV IPC does have an
1450	 * additional creator id...
1451	 */
1452	ns = current->nsproxy->ipc_ns;
1453	rcu_read_lock();
1454	shp = shm_obtain_object_check(ns, shmid);
1455	if (IS_ERR(shp)) {
1456		err = PTR_ERR(shp);
1457		goto out_unlock;
1458	}
1459
1460	err = -EACCES;
1461	if (ipcperms(ns, &shp->shm_perm, acc_mode))
1462		goto out_unlock;
1463
1464	err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1465	if (err)
1466		goto out_unlock;
1467
1468	ipc_lock_object(&shp->shm_perm);
1469
1470	/* check if shm_destroy() is tearing down shp */
1471	if (!ipc_valid_object(&shp->shm_perm)) {
1472		ipc_unlock_object(&shp->shm_perm);
1473		err = -EIDRM;
1474		goto out_unlock;
1475	}
1476
1477	/*
1478	 * We need to take a reference to the real shm file to prevent the
1479	 * pointer from becoming stale in cases where the lifetime of the outer
1480	 * file extends beyond that of the shm segment.  It's not usually
1481	 * possible, but it can happen during remap_file_pages() emulation as
1482	 * that unmaps the memory, then does ->mmap() via file reference only.
1483	 * We'll deny the ->mmap() if the shm segment was since removed, but to
1484	 * detect shm ID reuse we need to compare the file pointers.
1485	 */
1486	base = get_file(shp->shm_file);
1487	shp->shm_nattch++;
1488	size = i_size_read(file_inode(base));
1489	ipc_unlock_object(&shp->shm_perm);
1490	rcu_read_unlock();
1491
1492	err = -ENOMEM;
1493	sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1494	if (!sfd) {
1495		fput(base);
1496		goto out_nattch;
1497	}
1498
1499	file = alloc_file_clone(base, f_flags,
1500			  is_file_hugepages(base) ?
1501				&shm_file_operations_huge :
1502				&shm_file_operations);
1503	err = PTR_ERR(file);
1504	if (IS_ERR(file)) {
1505		kfree(sfd);
1506		fput(base);
1507		goto out_nattch;
1508	}
1509
1510	sfd->id = shp->shm_perm.id;
1511	sfd->ns = get_ipc_ns(ns);
1512	sfd->file = base;
1513	sfd->vm_ops = NULL;
1514	file->private_data = sfd;
1515
1516	err = security_mmap_file(file, prot, flags);
1517	if (err)
1518		goto out_fput;
1519
1520	if (down_write_killable(&current->mm->mmap_sem)) {
1521		err = -EINTR;
1522		goto out_fput;
1523	}
1524
1525	if (addr && !(shmflg & SHM_REMAP)) {
1526		err = -EINVAL;
1527		if (addr + size < addr)
1528			goto invalid;
1529
1530		if (find_vma_intersection(current->mm, addr, addr + size))
1531			goto invalid;
1532	}
1533
1534	addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1535	*raddr = addr;
1536	err = 0;
1537	if (IS_ERR_VALUE(addr))
1538		err = (long)addr;
1539invalid:
1540	up_write(&current->mm->mmap_sem);
1541	if (populate)
1542		mm_populate(addr, populate);
1543
1544out_fput:
1545	fput(file);
1546
1547out_nattch:
1548	down_write(&shm_ids(ns).rwsem);
1549	shp = shm_lock(ns, shmid);
1550	shp->shm_nattch--;
1551	if (shm_may_destroy(ns, shp))
1552		shm_destroy(ns, shp);
1553	else
1554		shm_unlock(shp);
1555	up_write(&shm_ids(ns).rwsem);
1556	return err;
1557
1558out_unlock:
1559	rcu_read_unlock();
1560out:
1561	return err;
1562}
1563
1564SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1565{
1566	unsigned long ret;
1567	long err;
1568
1569	err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1570	if (err)
1571		return err;
1572	force_successful_syscall_return();
1573	return (long)ret;
1574}
1575
1576#ifdef CONFIG_COMPAT
1577
1578#ifndef COMPAT_SHMLBA
1579#define COMPAT_SHMLBA	SHMLBA
1580#endif
1581
1582COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1583{
1584	unsigned long ret;
1585	long err;
1586
1587	err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1588	if (err)
1589		return err;
1590	force_successful_syscall_return();
1591	return (long)ret;
1592}
1593#endif
1594
1595/*
1596 * detach and kill segment if marked destroyed.
1597 * The work is done in shm_close.
1598 */
1599long ksys_shmdt(char __user *shmaddr)
1600{
1601	struct mm_struct *mm = current->mm;
1602	struct vm_area_struct *vma;
1603	unsigned long addr = (unsigned long)shmaddr;
1604	int retval = -EINVAL;
1605#ifdef CONFIG_MMU
1606	loff_t size = 0;
1607	struct file *file;
1608	struct vm_area_struct *next;
1609#endif
1610
1611	if (addr & ~PAGE_MASK)
1612		return retval;
1613
1614	if (down_write_killable(&mm->mmap_sem))
1615		return -EINTR;
1616
1617	/*
1618	 * This function tries to be smart and unmap shm segments that
1619	 * were modified by partial mlock or munmap calls:
1620	 * - It first determines the size of the shm segment that should be
1621	 *   unmapped: It searches for a vma that is backed by shm and that
1622	 *   started at address shmaddr. It records it's size and then unmaps
1623	 *   it.
1624	 * - Then it unmaps all shm vmas that started at shmaddr and that
1625	 *   are within the initially determined size and that are from the
1626	 *   same shm segment from which we determined the size.
1627	 * Errors from do_munmap are ignored: the function only fails if
1628	 * it's called with invalid parameters or if it's called to unmap
1629	 * a part of a vma. Both calls in this function are for full vmas,
1630	 * the parameters are directly copied from the vma itself and always
1631	 * valid - therefore do_munmap cannot fail. (famous last words?)
1632	 */
1633	/*
1634	 * If it had been mremap()'d, the starting address would not
1635	 * match the usual checks anyway. So assume all vma's are
1636	 * above the starting address given.
1637	 */
1638	vma = find_vma(mm, addr);
1639
1640#ifdef CONFIG_MMU
1641	while (vma) {
1642		next = vma->vm_next;
1643
1644		/*
1645		 * Check if the starting address would match, i.e. it's
1646		 * a fragment created by mprotect() and/or munmap(), or it
1647		 * otherwise it starts at this address with no hassles.
1648		 */
1649		if ((vma->vm_ops == &shm_vm_ops) &&
1650			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1651
1652			/*
1653			 * Record the file of the shm segment being
1654			 * unmapped.  With mremap(), someone could place
1655			 * page from another segment but with equal offsets
1656			 * in the range we are unmapping.
1657			 */
1658			file = vma->vm_file;
1659			size = i_size_read(file_inode(vma->vm_file));
1660			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1661			/*
1662			 * We discovered the size of the shm segment, so
1663			 * break out of here and fall through to the next
1664			 * loop that uses the size information to stop
1665			 * searching for matching vma's.
1666			 */
1667			retval = 0;
1668			vma = next;
1669			break;
1670		}
1671		vma = next;
1672	}
1673
1674	/*
1675	 * We need look no further than the maximum address a fragment
1676	 * could possibly have landed at. Also cast things to loff_t to
1677	 * prevent overflows and make comparisons vs. equal-width types.
1678	 */
1679	size = PAGE_ALIGN(size);
1680	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1681		next = vma->vm_next;
1682
1683		/* finding a matching vma now does not alter retval */
1684		if ((vma->vm_ops == &shm_vm_ops) &&
1685		    ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1686		    (vma->vm_file == file))
1687			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1688		vma = next;
1689	}
1690
1691#else	/* CONFIG_MMU */
1692	/* under NOMMU conditions, the exact address to be destroyed must be
1693	 * given
1694	 */
1695	if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1696		do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1697		retval = 0;
1698	}
1699
1700#endif
1701
1702	up_write(&mm->mmap_sem);
1703	return retval;
1704}
1705
1706SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1707{
1708	return ksys_shmdt(shmaddr);
1709}
1710
1711#ifdef CONFIG_PROC_FS
1712static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1713{
1714	struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1715	struct user_namespace *user_ns = seq_user_ns(s);
1716	struct kern_ipc_perm *ipcp = it;
1717	struct shmid_kernel *shp;
1718	unsigned long rss = 0, swp = 0;
1719
1720	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1721	shm_add_rss_swap(shp, &rss, &swp);
1722
1723#if BITS_PER_LONG <= 32
1724#define SIZE_SPEC "%10lu"
1725#else
1726#define SIZE_SPEC "%21lu"
1727#endif
1728
1729	seq_printf(s,
1730		   "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1731		   "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1732		   SIZE_SPEC " " SIZE_SPEC "\n",
1733		   shp->shm_perm.key,
1734		   shp->shm_perm.id,
1735		   shp->shm_perm.mode,
1736		   shp->shm_segsz,
1737		   pid_nr_ns(shp->shm_cprid, pid_ns),
1738		   pid_nr_ns(shp->shm_lprid, pid_ns),
1739		   shp->shm_nattch,
1740		   from_kuid_munged(user_ns, shp->shm_perm.uid),
1741		   from_kgid_munged(user_ns, shp->shm_perm.gid),
1742		   from_kuid_munged(user_ns, shp->shm_perm.cuid),
1743		   from_kgid_munged(user_ns, shp->shm_perm.cgid),
1744		   shp->shm_atim,
1745		   shp->shm_dtim,
1746		   shp->shm_ctim,
1747		   rss * PAGE_SIZE,
1748		   swp * PAGE_SIZE);
1749
1750	return 0;
1751}
1752#endif
1753