shm.c revision 7f032d6e
1/*
2 * linux/ipc/shm.c
3 * Copyright (C) 1992, 1993 Krishna Balasubramanian
4 *	 Many improvements/fixes by Bruno Haible.
5 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7 *
8 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15 *
16 * support for audit of ipc object properties and permission changes
17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18 *
19 * namespaces support
20 * OpenVZ, SWsoft Inc.
21 * Pavel Emelianov <xemul@openvz.org>
22 *
23 * Better ipc lock (kern_ipc_perm.lock) handling
24 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
25 */
26
27#include <linux/slab.h>
28#include <linux/mm.h>
29#include <linux/hugetlb.h>
30#include <linux/shm.h>
31#include <linux/init.h>
32#include <linux/file.h>
33#include <linux/mman.h>
34#include <linux/shmem_fs.h>
35#include <linux/security.h>
36#include <linux/syscalls.h>
37#include <linux/audit.h>
38#include <linux/capability.h>
39#include <linux/ptrace.h>
40#include <linux/seq_file.h>
41#include <linux/rwsem.h>
42#include <linux/nsproxy.h>
43#include <linux/mount.h>
44#include <linux/ipc_namespace.h>
45
46#include <linux/uaccess.h>
47
48#include "util.h"
49
50struct shm_file_data {
51	int id;
52	struct ipc_namespace *ns;
53	struct file *file;
54	const struct vm_operations_struct *vm_ops;
55};
56
57#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
58
59static const struct file_operations shm_file_operations;
60static const struct vm_operations_struct shm_vm_ops;
61
62#define shm_ids(ns)	((ns)->ids[IPC_SHM_IDS])
63
64#define shm_unlock(shp)			\
65	ipc_unlock(&(shp)->shm_perm)
66
67static int newseg(struct ipc_namespace *, struct ipc_params *);
68static void shm_open(struct vm_area_struct *vma);
69static void shm_close(struct vm_area_struct *vma);
70static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
71#ifdef CONFIG_PROC_FS
72static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
73#endif
74
75void shm_init_ns(struct ipc_namespace *ns)
76{
77	ns->shm_ctlmax = SHMMAX;
78	ns->shm_ctlall = SHMALL;
79	ns->shm_ctlmni = SHMMNI;
80	ns->shm_rmid_forced = 0;
81	ns->shm_tot = 0;
82	ipc_init_ids(&shm_ids(ns));
83}
84
85/*
86 * Called with shm_ids.rwsem (writer) and the shp structure locked.
87 * Only shm_ids.rwsem remains locked on exit.
88 */
89static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
90{
91	struct shmid_kernel *shp;
92	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
93
94	if (shp->shm_nattch) {
95		shp->shm_perm.mode |= SHM_DEST;
96		/* Do not find it any more */
97		shp->shm_perm.key = IPC_PRIVATE;
98		shm_unlock(shp);
99	} else
100		shm_destroy(ns, shp);
101}
102
103#ifdef CONFIG_IPC_NS
104void shm_exit_ns(struct ipc_namespace *ns)
105{
106	free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
107	idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
108}
109#endif
110
111static int __init ipc_ns_init(void)
112{
113	shm_init_ns(&init_ipc_ns);
114	return 0;
115}
116
117pure_initcall(ipc_ns_init);
118
119void __init shm_init(void)
120{
121	ipc_init_proc_interface("sysvipc/shm",
122#if BITS_PER_LONG <= 32
123				"       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
124#else
125				"       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
126#endif
127				IPC_SHM_IDS, sysvipc_shm_proc_show);
128}
129
130static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
131{
132	struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
133
134	if (IS_ERR(ipcp))
135		return ERR_CAST(ipcp);
136
137	return container_of(ipcp, struct shmid_kernel, shm_perm);
138}
139
140static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
141{
142	struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
143
144	if (IS_ERR(ipcp))
145		return ERR_CAST(ipcp);
146
147	return container_of(ipcp, struct shmid_kernel, shm_perm);
148}
149
150/*
151 * shm_lock_(check_) routines are called in the paths where the rwsem
152 * is not necessarily held.
153 */
154static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
155{
156	struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
157
158	if (IS_ERR(ipcp))
159		return (struct shmid_kernel *)ipcp;
160
161	return container_of(ipcp, struct shmid_kernel, shm_perm);
162}
163
164static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
165{
166	rcu_read_lock();
167	ipc_lock_object(&ipcp->shm_perm);
168}
169
170static void shm_rcu_free(struct rcu_head *head)
171{
172	struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
173	struct shmid_kernel *shp = ipc_rcu_to_struct(p);
174
175	security_shm_free(shp);
176	ipc_rcu_free(head);
177}
178
179static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
180{
181	list_del(&s->shm_clist);
182	ipc_rmid(&shm_ids(ns), &s->shm_perm);
183}
184
185
186/* This is called by fork, once for every shm attach. */
187static void shm_open(struct vm_area_struct *vma)
188{
189	struct file *file = vma->vm_file;
190	struct shm_file_data *sfd = shm_file_data(file);
191	struct shmid_kernel *shp;
192
193	shp = shm_lock(sfd->ns, sfd->id);
194	BUG_ON(IS_ERR(shp));
195	shp->shm_atim = get_seconds();
196	shp->shm_lprid = task_tgid_vnr(current);
197	shp->shm_nattch++;
198	shm_unlock(shp);
199}
200
201/*
202 * shm_destroy - free the struct shmid_kernel
203 *
204 * @ns: namespace
205 * @shp: struct to free
206 *
207 * It has to be called with shp and shm_ids.rwsem (writer) locked,
208 * but returns with shp unlocked and freed.
209 */
210static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
211{
212	struct file *shm_file;
213
214	shm_file = shp->shm_file;
215	shp->shm_file = NULL;
216	ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
217	shm_rmid(ns, shp);
218	shm_unlock(shp);
219	if (!is_file_hugepages(shm_file))
220		shmem_lock(shm_file, 0, shp->mlock_user);
221	else if (shp->mlock_user)
222		user_shm_unlock(i_size_read(file_inode(shm_file)),
223				shp->mlock_user);
224	fput(shm_file);
225	ipc_rcu_putref(shp, shm_rcu_free);
226}
227
228/*
229 * shm_may_destroy - identifies whether shm segment should be destroyed now
230 *
231 * Returns true if and only if there are no active users of the segment and
232 * one of the following is true:
233 *
234 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
235 *
236 * 2) sysctl kernel.shm_rmid_forced is set to 1.
237 */
238static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
239{
240	return (shp->shm_nattch == 0) &&
241	       (ns->shm_rmid_forced ||
242		(shp->shm_perm.mode & SHM_DEST));
243}
244
245/*
246 * remove the attach descriptor vma.
247 * free memory for segment if it is marked destroyed.
248 * The descriptor has already been removed from the current->mm->mmap list
249 * and will later be kfree()d.
250 */
251static void shm_close(struct vm_area_struct *vma)
252{
253	struct file *file = vma->vm_file;
254	struct shm_file_data *sfd = shm_file_data(file);
255	struct shmid_kernel *shp;
256	struct ipc_namespace *ns = sfd->ns;
257
258	down_write(&shm_ids(ns).rwsem);
259	/* remove from the list of attaches of the shm segment */
260	shp = shm_lock(ns, sfd->id);
261	BUG_ON(IS_ERR(shp));
262	shp->shm_lprid = task_tgid_vnr(current);
263	shp->shm_dtim = get_seconds();
264	shp->shm_nattch--;
265	if (shm_may_destroy(ns, shp))
266		shm_destroy(ns, shp);
267	else
268		shm_unlock(shp);
269	up_write(&shm_ids(ns).rwsem);
270}
271
272/* Called with ns->shm_ids(ns).rwsem locked */
273static int shm_try_destroy_orphaned(int id, void *p, void *data)
274{
275	struct ipc_namespace *ns = data;
276	struct kern_ipc_perm *ipcp = p;
277	struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
278
279	/*
280	 * We want to destroy segments without users and with already
281	 * exit'ed originating process.
282	 *
283	 * As shp->* are changed under rwsem, it's safe to skip shp locking.
284	 */
285	if (shp->shm_creator != NULL)
286		return 0;
287
288	if (shm_may_destroy(ns, shp)) {
289		shm_lock_by_ptr(shp);
290		shm_destroy(ns, shp);
291	}
292	return 0;
293}
294
295void shm_destroy_orphaned(struct ipc_namespace *ns)
296{
297	down_write(&shm_ids(ns).rwsem);
298	if (shm_ids(ns).in_use)
299		idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
300	up_write(&shm_ids(ns).rwsem);
301}
302
303/* Locking assumes this will only be called with task == current */
304void exit_shm(struct task_struct *task)
305{
306	struct ipc_namespace *ns = task->nsproxy->ipc_ns;
307	struct shmid_kernel *shp, *n;
308
309	if (list_empty(&task->sysvshm.shm_clist))
310		return;
311
312	/*
313	 * If kernel.shm_rmid_forced is not set then only keep track of
314	 * which shmids are orphaned, so that a later set of the sysctl
315	 * can clean them up.
316	 */
317	if (!ns->shm_rmid_forced) {
318		down_read(&shm_ids(ns).rwsem);
319		list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
320			shp->shm_creator = NULL;
321		/*
322		 * Only under read lock but we are only called on current
323		 * so no entry on the list will be shared.
324		 */
325		list_del(&task->sysvshm.shm_clist);
326		up_read(&shm_ids(ns).rwsem);
327		return;
328	}
329
330	/*
331	 * Destroy all already created segments, that were not yet mapped,
332	 * and mark any mapped as orphan to cover the sysctl toggling.
333	 * Destroy is skipped if shm_may_destroy() returns false.
334	 */
335	down_write(&shm_ids(ns).rwsem);
336	list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
337		shp->shm_creator = NULL;
338
339		if (shm_may_destroy(ns, shp)) {
340			shm_lock_by_ptr(shp);
341			shm_destroy(ns, shp);
342		}
343	}
344
345	/* Remove the list head from any segments still attached. */
346	list_del(&task->sysvshm.shm_clist);
347	up_write(&shm_ids(ns).rwsem);
348}
349
350static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
351{
352	struct file *file = vma->vm_file;
353	struct shm_file_data *sfd = shm_file_data(file);
354
355	return sfd->vm_ops->fault(vma, vmf);
356}
357
358#ifdef CONFIG_NUMA
359static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
360{
361	struct file *file = vma->vm_file;
362	struct shm_file_data *sfd = shm_file_data(file);
363	int err = 0;
364	if (sfd->vm_ops->set_policy)
365		err = sfd->vm_ops->set_policy(vma, new);
366	return err;
367}
368
369static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
370					unsigned long addr)
371{
372	struct file *file = vma->vm_file;
373	struct shm_file_data *sfd = shm_file_data(file);
374	struct mempolicy *pol = NULL;
375
376	if (sfd->vm_ops->get_policy)
377		pol = sfd->vm_ops->get_policy(vma, addr);
378	else if (vma->vm_policy)
379		pol = vma->vm_policy;
380
381	return pol;
382}
383#endif
384
385static int shm_mmap(struct file *file, struct vm_area_struct *vma)
386{
387	struct shm_file_data *sfd = shm_file_data(file);
388	int ret;
389
390	ret = sfd->file->f_op->mmap(sfd->file, vma);
391	if (ret != 0)
392		return ret;
393	sfd->vm_ops = vma->vm_ops;
394#ifdef CONFIG_MMU
395	BUG_ON(!sfd->vm_ops->fault);
396#endif
397	vma->vm_ops = &shm_vm_ops;
398	shm_open(vma);
399
400	return ret;
401}
402
403static int shm_release(struct inode *ino, struct file *file)
404{
405	struct shm_file_data *sfd = shm_file_data(file);
406
407	put_ipc_ns(sfd->ns);
408	shm_file_data(file) = NULL;
409	kfree(sfd);
410	return 0;
411}
412
413static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
414{
415	struct shm_file_data *sfd = shm_file_data(file);
416
417	if (!sfd->file->f_op->fsync)
418		return -EINVAL;
419	return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
420}
421
422static long shm_fallocate(struct file *file, int mode, loff_t offset,
423			  loff_t len)
424{
425	struct shm_file_data *sfd = shm_file_data(file);
426
427	if (!sfd->file->f_op->fallocate)
428		return -EOPNOTSUPP;
429	return sfd->file->f_op->fallocate(file, mode, offset, len);
430}
431
432static unsigned long shm_get_unmapped_area(struct file *file,
433	unsigned long addr, unsigned long len, unsigned long pgoff,
434	unsigned long flags)
435{
436	struct shm_file_data *sfd = shm_file_data(file);
437	return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
438						pgoff, flags);
439}
440
441static const struct file_operations shm_file_operations = {
442	.mmap		= shm_mmap,
443	.fsync		= shm_fsync,
444	.release	= shm_release,
445#ifndef CONFIG_MMU
446	.get_unmapped_area	= shm_get_unmapped_area,
447#endif
448	.llseek		= noop_llseek,
449	.fallocate	= shm_fallocate,
450};
451
452static const struct file_operations shm_file_operations_huge = {
453	.mmap		= shm_mmap,
454	.fsync		= shm_fsync,
455	.release	= shm_release,
456	.get_unmapped_area	= shm_get_unmapped_area,
457	.llseek		= noop_llseek,
458	.fallocate	= shm_fallocate,
459};
460
461int is_file_shm_hugepages(struct file *file)
462{
463	return file->f_op == &shm_file_operations_huge;
464}
465
466static const struct vm_operations_struct shm_vm_ops = {
467	.open	= shm_open,	/* callback for a new vm-area open */
468	.close	= shm_close,	/* callback for when the vm-area is released */
469	.fault	= shm_fault,
470#if defined(CONFIG_NUMA)
471	.set_policy = shm_set_policy,
472	.get_policy = shm_get_policy,
473#endif
474};
475
476/**
477 * newseg - Create a new shared memory segment
478 * @ns: namespace
479 * @params: ptr to the structure that contains key, size and shmflg
480 *
481 * Called with shm_ids.rwsem held as a writer.
482 */
483static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
484{
485	key_t key = params->key;
486	int shmflg = params->flg;
487	size_t size = params->u.size;
488	int error;
489	struct shmid_kernel *shp;
490	size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
491	struct file *file;
492	char name[13];
493	int id;
494	vm_flags_t acctflag = 0;
495
496	if (size < SHMMIN || size > ns->shm_ctlmax)
497		return -EINVAL;
498
499	if (numpages << PAGE_SHIFT < size)
500		return -ENOSPC;
501
502	if (ns->shm_tot + numpages < ns->shm_tot ||
503			ns->shm_tot + numpages > ns->shm_ctlall)
504		return -ENOSPC;
505
506	shp = ipc_rcu_alloc(sizeof(*shp));
507	if (!shp)
508		return -ENOMEM;
509
510	shp->shm_perm.key = key;
511	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
512	shp->mlock_user = NULL;
513
514	shp->shm_perm.security = NULL;
515	error = security_shm_alloc(shp);
516	if (error) {
517		ipc_rcu_putref(shp, ipc_rcu_free);
518		return error;
519	}
520
521	sprintf(name, "SYSV%08x", key);
522	if (shmflg & SHM_HUGETLB) {
523		struct hstate *hs;
524		size_t hugesize;
525
526		hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
527		if (!hs) {
528			error = -EINVAL;
529			goto no_file;
530		}
531		hugesize = ALIGN(size, huge_page_size(hs));
532
533		/* hugetlb_file_setup applies strict accounting */
534		if (shmflg & SHM_NORESERVE)
535			acctflag = VM_NORESERVE;
536		file = hugetlb_file_setup(name, hugesize, acctflag,
537				  &shp->mlock_user, HUGETLB_SHMFS_INODE,
538				(shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
539	} else {
540		/*
541		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
542		 * if it's asked for.
543		 */
544		if  ((shmflg & SHM_NORESERVE) &&
545				sysctl_overcommit_memory != OVERCOMMIT_NEVER)
546			acctflag = VM_NORESERVE;
547		file = shmem_file_setup(name, size, acctflag);
548	}
549	error = PTR_ERR(file);
550	if (IS_ERR(file))
551		goto no_file;
552
553	id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
554	if (id < 0) {
555		error = id;
556		goto no_id;
557	}
558
559	shp->shm_cprid = task_tgid_vnr(current);
560	shp->shm_lprid = 0;
561	shp->shm_atim = shp->shm_dtim = 0;
562	shp->shm_ctim = get_seconds();
563	shp->shm_segsz = size;
564	shp->shm_nattch = 0;
565	shp->shm_file = file;
566	shp->shm_creator = current;
567	list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
568
569	/*
570	 * shmid gets reported as "inode#" in /proc/pid/maps.
571	 * proc-ps tools use this. Changing this will break them.
572	 */
573	file_inode(file)->i_ino = shp->shm_perm.id;
574
575	ns->shm_tot += numpages;
576	error = shp->shm_perm.id;
577
578	ipc_unlock_object(&shp->shm_perm);
579	rcu_read_unlock();
580	return error;
581
582no_id:
583	if (is_file_hugepages(file) && shp->mlock_user)
584		user_shm_unlock(size, shp->mlock_user);
585	fput(file);
586no_file:
587	ipc_rcu_putref(shp, shm_rcu_free);
588	return error;
589}
590
591/*
592 * Called with shm_ids.rwsem and ipcp locked.
593 */
594static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
595{
596	struct shmid_kernel *shp;
597
598	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
599	return security_shm_associate(shp, shmflg);
600}
601
602/*
603 * Called with shm_ids.rwsem and ipcp locked.
604 */
605static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
606				struct ipc_params *params)
607{
608	struct shmid_kernel *shp;
609
610	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
611	if (shp->shm_segsz < params->u.size)
612		return -EINVAL;
613
614	return 0;
615}
616
617SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
618{
619	struct ipc_namespace *ns;
620	static const struct ipc_ops shm_ops = {
621		.getnew = newseg,
622		.associate = shm_security,
623		.more_checks = shm_more_checks,
624	};
625	struct ipc_params shm_params;
626
627	ns = current->nsproxy->ipc_ns;
628
629	shm_params.key = key;
630	shm_params.flg = shmflg;
631	shm_params.u.size = size;
632
633	return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
634}
635
636static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
637{
638	switch (version) {
639	case IPC_64:
640		return copy_to_user(buf, in, sizeof(*in));
641	case IPC_OLD:
642	    {
643		struct shmid_ds out;
644
645		memset(&out, 0, sizeof(out));
646		ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
647		out.shm_segsz	= in->shm_segsz;
648		out.shm_atime	= in->shm_atime;
649		out.shm_dtime	= in->shm_dtime;
650		out.shm_ctime	= in->shm_ctime;
651		out.shm_cpid	= in->shm_cpid;
652		out.shm_lpid	= in->shm_lpid;
653		out.shm_nattch	= in->shm_nattch;
654
655		return copy_to_user(buf, &out, sizeof(out));
656	    }
657	default:
658		return -EINVAL;
659	}
660}
661
662static inline unsigned long
663copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
664{
665	switch (version) {
666	case IPC_64:
667		if (copy_from_user(out, buf, sizeof(*out)))
668			return -EFAULT;
669		return 0;
670	case IPC_OLD:
671	    {
672		struct shmid_ds tbuf_old;
673
674		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
675			return -EFAULT;
676
677		out->shm_perm.uid	= tbuf_old.shm_perm.uid;
678		out->shm_perm.gid	= tbuf_old.shm_perm.gid;
679		out->shm_perm.mode	= tbuf_old.shm_perm.mode;
680
681		return 0;
682	    }
683	default:
684		return -EINVAL;
685	}
686}
687
688static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
689{
690	switch (version) {
691	case IPC_64:
692		return copy_to_user(buf, in, sizeof(*in));
693	case IPC_OLD:
694	    {
695		struct shminfo out;
696
697		if (in->shmmax > INT_MAX)
698			out.shmmax = INT_MAX;
699		else
700			out.shmmax = (int)in->shmmax;
701
702		out.shmmin	= in->shmmin;
703		out.shmmni	= in->shmmni;
704		out.shmseg	= in->shmseg;
705		out.shmall	= in->shmall;
706
707		return copy_to_user(buf, &out, sizeof(out));
708	    }
709	default:
710		return -EINVAL;
711	}
712}
713
714/*
715 * Calculate and add used RSS and swap pages of a shm.
716 * Called with shm_ids.rwsem held as a reader
717 */
718static void shm_add_rss_swap(struct shmid_kernel *shp,
719	unsigned long *rss_add, unsigned long *swp_add)
720{
721	struct inode *inode;
722
723	inode = file_inode(shp->shm_file);
724
725	if (is_file_hugepages(shp->shm_file)) {
726		struct address_space *mapping = inode->i_mapping;
727		struct hstate *h = hstate_file(shp->shm_file);
728		*rss_add += pages_per_huge_page(h) * mapping->nrpages;
729	} else {
730#ifdef CONFIG_SHMEM
731		struct shmem_inode_info *info = SHMEM_I(inode);
732		spin_lock(&info->lock);
733		*rss_add += inode->i_mapping->nrpages;
734		*swp_add += info->swapped;
735		spin_unlock(&info->lock);
736#else
737		*rss_add += inode->i_mapping->nrpages;
738#endif
739	}
740}
741
742/*
743 * Called with shm_ids.rwsem held as a reader
744 */
745static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
746		unsigned long *swp)
747{
748	int next_id;
749	int total, in_use;
750
751	*rss = 0;
752	*swp = 0;
753
754	in_use = shm_ids(ns).in_use;
755
756	for (total = 0, next_id = 0; total < in_use; next_id++) {
757		struct kern_ipc_perm *ipc;
758		struct shmid_kernel *shp;
759
760		ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
761		if (ipc == NULL)
762			continue;
763		shp = container_of(ipc, struct shmid_kernel, shm_perm);
764
765		shm_add_rss_swap(shp, rss, swp);
766
767		total++;
768	}
769}
770
771/*
772 * This function handles some shmctl commands which require the rwsem
773 * to be held in write mode.
774 * NOTE: no locks must be held, the rwsem is taken inside this function.
775 */
776static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
777		       struct shmid_ds __user *buf, int version)
778{
779	struct kern_ipc_perm *ipcp;
780	struct shmid64_ds shmid64;
781	struct shmid_kernel *shp;
782	int err;
783
784	if (cmd == IPC_SET) {
785		if (copy_shmid_from_user(&shmid64, buf, version))
786			return -EFAULT;
787	}
788
789	down_write(&shm_ids(ns).rwsem);
790	rcu_read_lock();
791
792	ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
793				      &shmid64.shm_perm, 0);
794	if (IS_ERR(ipcp)) {
795		err = PTR_ERR(ipcp);
796		goto out_unlock1;
797	}
798
799	shp = container_of(ipcp, struct shmid_kernel, shm_perm);
800
801	err = security_shm_shmctl(shp, cmd);
802	if (err)
803		goto out_unlock1;
804
805	switch (cmd) {
806	case IPC_RMID:
807		ipc_lock_object(&shp->shm_perm);
808		/* do_shm_rmid unlocks the ipc object and rcu */
809		do_shm_rmid(ns, ipcp);
810		goto out_up;
811	case IPC_SET:
812		ipc_lock_object(&shp->shm_perm);
813		err = ipc_update_perm(&shmid64.shm_perm, ipcp);
814		if (err)
815			goto out_unlock0;
816		shp->shm_ctim = get_seconds();
817		break;
818	default:
819		err = -EINVAL;
820		goto out_unlock1;
821	}
822
823out_unlock0:
824	ipc_unlock_object(&shp->shm_perm);
825out_unlock1:
826	rcu_read_unlock();
827out_up:
828	up_write(&shm_ids(ns).rwsem);
829	return err;
830}
831
832static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
833			 int cmd, int version, void __user *buf)
834{
835	int err;
836	struct shmid_kernel *shp;
837
838	/* preliminary security checks for *_INFO */
839	if (cmd == IPC_INFO || cmd == SHM_INFO) {
840		err = security_shm_shmctl(NULL, cmd);
841		if (err)
842			return err;
843	}
844
845	switch (cmd) {
846	case IPC_INFO:
847	{
848		struct shminfo64 shminfo;
849
850		memset(&shminfo, 0, sizeof(shminfo));
851		shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
852		shminfo.shmmax = ns->shm_ctlmax;
853		shminfo.shmall = ns->shm_ctlall;
854
855		shminfo.shmmin = SHMMIN;
856		if (copy_shminfo_to_user(buf, &shminfo, version))
857			return -EFAULT;
858
859		down_read(&shm_ids(ns).rwsem);
860		err = ipc_get_maxid(&shm_ids(ns));
861		up_read(&shm_ids(ns).rwsem);
862
863		if (err < 0)
864			err = 0;
865		goto out;
866	}
867	case SHM_INFO:
868	{
869		struct shm_info shm_info;
870
871		memset(&shm_info, 0, sizeof(shm_info));
872		down_read(&shm_ids(ns).rwsem);
873		shm_info.used_ids = shm_ids(ns).in_use;
874		shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
875		shm_info.shm_tot = ns->shm_tot;
876		shm_info.swap_attempts = 0;
877		shm_info.swap_successes = 0;
878		err = ipc_get_maxid(&shm_ids(ns));
879		up_read(&shm_ids(ns).rwsem);
880		if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
881			err = -EFAULT;
882			goto out;
883		}
884
885		err = err < 0 ? 0 : err;
886		goto out;
887	}
888	case SHM_STAT:
889	case IPC_STAT:
890	{
891		struct shmid64_ds tbuf;
892		int result;
893
894		rcu_read_lock();
895		if (cmd == SHM_STAT) {
896			shp = shm_obtain_object(ns, shmid);
897			if (IS_ERR(shp)) {
898				err = PTR_ERR(shp);
899				goto out_unlock;
900			}
901			result = shp->shm_perm.id;
902		} else {
903			shp = shm_obtain_object_check(ns, shmid);
904			if (IS_ERR(shp)) {
905				err = PTR_ERR(shp);
906				goto out_unlock;
907			}
908			result = 0;
909		}
910
911		err = -EACCES;
912		if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
913			goto out_unlock;
914
915		err = security_shm_shmctl(shp, cmd);
916		if (err)
917			goto out_unlock;
918
919		memset(&tbuf, 0, sizeof(tbuf));
920		kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
921		tbuf.shm_segsz	= shp->shm_segsz;
922		tbuf.shm_atime	= shp->shm_atim;
923		tbuf.shm_dtime	= shp->shm_dtim;
924		tbuf.shm_ctime	= shp->shm_ctim;
925		tbuf.shm_cpid	= shp->shm_cprid;
926		tbuf.shm_lpid	= shp->shm_lprid;
927		tbuf.shm_nattch	= shp->shm_nattch;
928		rcu_read_unlock();
929
930		if (copy_shmid_to_user(buf, &tbuf, version))
931			err = -EFAULT;
932		else
933			err = result;
934		goto out;
935	}
936	default:
937		return -EINVAL;
938	}
939
940out_unlock:
941	rcu_read_unlock();
942out:
943	return err;
944}
945
946SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
947{
948	struct shmid_kernel *shp;
949	int err, version;
950	struct ipc_namespace *ns;
951
952	if (cmd < 0 || shmid < 0)
953		return -EINVAL;
954
955	version = ipc_parse_version(&cmd);
956	ns = current->nsproxy->ipc_ns;
957
958	switch (cmd) {
959	case IPC_INFO:
960	case SHM_INFO:
961	case SHM_STAT:
962	case IPC_STAT:
963		return shmctl_nolock(ns, shmid, cmd, version, buf);
964	case IPC_RMID:
965	case IPC_SET:
966		return shmctl_down(ns, shmid, cmd, buf, version);
967	case SHM_LOCK:
968	case SHM_UNLOCK:
969	{
970		struct file *shm_file;
971
972		rcu_read_lock();
973		shp = shm_obtain_object_check(ns, shmid);
974		if (IS_ERR(shp)) {
975			err = PTR_ERR(shp);
976			goto out_unlock1;
977		}
978
979		audit_ipc_obj(&(shp->shm_perm));
980		err = security_shm_shmctl(shp, cmd);
981		if (err)
982			goto out_unlock1;
983
984		ipc_lock_object(&shp->shm_perm);
985
986		/* check if shm_destroy() is tearing down shp */
987		if (!ipc_valid_object(&shp->shm_perm)) {
988			err = -EIDRM;
989			goto out_unlock0;
990		}
991
992		if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
993			kuid_t euid = current_euid();
994			if (!uid_eq(euid, shp->shm_perm.uid) &&
995			    !uid_eq(euid, shp->shm_perm.cuid)) {
996				err = -EPERM;
997				goto out_unlock0;
998			}
999			if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1000				err = -EPERM;
1001				goto out_unlock0;
1002			}
1003		}
1004
1005		shm_file = shp->shm_file;
1006		if (is_file_hugepages(shm_file))
1007			goto out_unlock0;
1008
1009		if (cmd == SHM_LOCK) {
1010			struct user_struct *user = current_user();
1011			err = shmem_lock(shm_file, 1, user);
1012			if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1013				shp->shm_perm.mode |= SHM_LOCKED;
1014				shp->mlock_user = user;
1015			}
1016			goto out_unlock0;
1017		}
1018
1019		/* SHM_UNLOCK */
1020		if (!(shp->shm_perm.mode & SHM_LOCKED))
1021			goto out_unlock0;
1022		shmem_lock(shm_file, 0, shp->mlock_user);
1023		shp->shm_perm.mode &= ~SHM_LOCKED;
1024		shp->mlock_user = NULL;
1025		get_file(shm_file);
1026		ipc_unlock_object(&shp->shm_perm);
1027		rcu_read_unlock();
1028		shmem_unlock_mapping(shm_file->f_mapping);
1029
1030		fput(shm_file);
1031		return err;
1032	}
1033	default:
1034		return -EINVAL;
1035	}
1036
1037out_unlock0:
1038	ipc_unlock_object(&shp->shm_perm);
1039out_unlock1:
1040	rcu_read_unlock();
1041	return err;
1042}
1043
1044/*
1045 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1046 *
1047 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1048 * "raddr" thing points to kernel space, and there has to be a wrapper around
1049 * this.
1050 */
1051long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1052	      unsigned long shmlba)
1053{
1054	struct shmid_kernel *shp;
1055	unsigned long addr;
1056	unsigned long size;
1057	struct file *file;
1058	int    err;
1059	unsigned long flags;
1060	unsigned long prot;
1061	int acc_mode;
1062	struct ipc_namespace *ns;
1063	struct shm_file_data *sfd;
1064	struct path path;
1065	fmode_t f_mode;
1066	unsigned long populate = 0;
1067
1068	err = -EINVAL;
1069	if (shmid < 0)
1070		goto out;
1071	else if ((addr = (ulong)shmaddr)) {
1072		if (addr & (shmlba - 1)) {
1073			if (shmflg & SHM_RND)
1074				addr &= ~(shmlba - 1);	   /* round down */
1075			else
1076#ifndef __ARCH_FORCE_SHMLBA
1077				if (addr & ~PAGE_MASK)
1078#endif
1079					goto out;
1080		}
1081		flags = MAP_SHARED | MAP_FIXED;
1082	} else {
1083		if ((shmflg & SHM_REMAP))
1084			goto out;
1085
1086		flags = MAP_SHARED;
1087	}
1088
1089	if (shmflg & SHM_RDONLY) {
1090		prot = PROT_READ;
1091		acc_mode = S_IRUGO;
1092		f_mode = FMODE_READ;
1093	} else {
1094		prot = PROT_READ | PROT_WRITE;
1095		acc_mode = S_IRUGO | S_IWUGO;
1096		f_mode = FMODE_READ | FMODE_WRITE;
1097	}
1098	if (shmflg & SHM_EXEC) {
1099		prot |= PROT_EXEC;
1100		acc_mode |= S_IXUGO;
1101	}
1102
1103	/*
1104	 * We cannot rely on the fs check since SYSV IPC does have an
1105	 * additional creator id...
1106	 */
1107	ns = current->nsproxy->ipc_ns;
1108	rcu_read_lock();
1109	shp = shm_obtain_object_check(ns, shmid);
1110	if (IS_ERR(shp)) {
1111		err = PTR_ERR(shp);
1112		goto out_unlock;
1113	}
1114
1115	err = -EACCES;
1116	if (ipcperms(ns, &shp->shm_perm, acc_mode))
1117		goto out_unlock;
1118
1119	err = security_shm_shmat(shp, shmaddr, shmflg);
1120	if (err)
1121		goto out_unlock;
1122
1123	ipc_lock_object(&shp->shm_perm);
1124
1125	/* check if shm_destroy() is tearing down shp */
1126	if (!ipc_valid_object(&shp->shm_perm)) {
1127		ipc_unlock_object(&shp->shm_perm);
1128		err = -EIDRM;
1129		goto out_unlock;
1130	}
1131
1132	path = shp->shm_file->f_path;
1133	path_get(&path);
1134	shp->shm_nattch++;
1135	size = i_size_read(path.dentry->d_inode);
1136	ipc_unlock_object(&shp->shm_perm);
1137	rcu_read_unlock();
1138
1139	err = -ENOMEM;
1140	sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1141	if (!sfd) {
1142		path_put(&path);
1143		goto out_nattch;
1144	}
1145
1146	file = alloc_file(&path, f_mode,
1147			  is_file_hugepages(shp->shm_file) ?
1148				&shm_file_operations_huge :
1149				&shm_file_operations);
1150	err = PTR_ERR(file);
1151	if (IS_ERR(file)) {
1152		kfree(sfd);
1153		path_put(&path);
1154		goto out_nattch;
1155	}
1156
1157	file->private_data = sfd;
1158	file->f_mapping = shp->shm_file->f_mapping;
1159	sfd->id = shp->shm_perm.id;
1160	sfd->ns = get_ipc_ns(ns);
1161	sfd->file = shp->shm_file;
1162	sfd->vm_ops = NULL;
1163
1164	err = security_mmap_file(file, prot, flags);
1165	if (err)
1166		goto out_fput;
1167
1168	down_write(&current->mm->mmap_sem);
1169	if (addr && !(shmflg & SHM_REMAP)) {
1170		err = -EINVAL;
1171		if (addr + size < addr)
1172			goto invalid;
1173
1174		if (find_vma_intersection(current->mm, addr, addr + size))
1175			goto invalid;
1176	}
1177
1178	addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1179	*raddr = addr;
1180	err = 0;
1181	if (IS_ERR_VALUE(addr))
1182		err = (long)addr;
1183invalid:
1184	up_write(&current->mm->mmap_sem);
1185	if (populate)
1186		mm_populate(addr, populate);
1187
1188out_fput:
1189	fput(file);
1190
1191out_nattch:
1192	down_write(&shm_ids(ns).rwsem);
1193	shp = shm_lock(ns, shmid);
1194	BUG_ON(IS_ERR(shp));
1195	shp->shm_nattch--;
1196	if (shm_may_destroy(ns, shp))
1197		shm_destroy(ns, shp);
1198	else
1199		shm_unlock(shp);
1200	up_write(&shm_ids(ns).rwsem);
1201	return err;
1202
1203out_unlock:
1204	rcu_read_unlock();
1205out:
1206	return err;
1207}
1208
1209SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1210{
1211	unsigned long ret;
1212	long err;
1213
1214	err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1215	if (err)
1216		return err;
1217	force_successful_syscall_return();
1218	return (long)ret;
1219}
1220
1221/*
1222 * detach and kill segment if marked destroyed.
1223 * The work is done in shm_close.
1224 */
1225SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1226{
1227	struct mm_struct *mm = current->mm;
1228	struct vm_area_struct *vma;
1229	unsigned long addr = (unsigned long)shmaddr;
1230	int retval = -EINVAL;
1231#ifdef CONFIG_MMU
1232	loff_t size = 0;
1233	struct file *file;
1234	struct vm_area_struct *next;
1235#endif
1236
1237	if (addr & ~PAGE_MASK)
1238		return retval;
1239
1240	down_write(&mm->mmap_sem);
1241
1242	/*
1243	 * This function tries to be smart and unmap shm segments that
1244	 * were modified by partial mlock or munmap calls:
1245	 * - It first determines the size of the shm segment that should be
1246	 *   unmapped: It searches for a vma that is backed by shm and that
1247	 *   started at address shmaddr. It records it's size and then unmaps
1248	 *   it.
1249	 * - Then it unmaps all shm vmas that started at shmaddr and that
1250	 *   are within the initially determined size and that are from the
1251	 *   same shm segment from which we determined the size.
1252	 * Errors from do_munmap are ignored: the function only fails if
1253	 * it's called with invalid parameters or if it's called to unmap
1254	 * a part of a vma. Both calls in this function are for full vmas,
1255	 * the parameters are directly copied from the vma itself and always
1256	 * valid - therefore do_munmap cannot fail. (famous last words?)
1257	 */
1258	/*
1259	 * If it had been mremap()'d, the starting address would not
1260	 * match the usual checks anyway. So assume all vma's are
1261	 * above the starting address given.
1262	 */
1263	vma = find_vma(mm, addr);
1264
1265#ifdef CONFIG_MMU
1266	while (vma) {
1267		next = vma->vm_next;
1268
1269		/*
1270		 * Check if the starting address would match, i.e. it's
1271		 * a fragment created by mprotect() and/or munmap(), or it
1272		 * otherwise it starts at this address with no hassles.
1273		 */
1274		if ((vma->vm_ops == &shm_vm_ops) &&
1275			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1276
1277			/*
1278			 * Record the file of the shm segment being
1279			 * unmapped.  With mremap(), someone could place
1280			 * page from another segment but with equal offsets
1281			 * in the range we are unmapping.
1282			 */
1283			file = vma->vm_file;
1284			size = i_size_read(file_inode(vma->vm_file));
1285			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1286			/*
1287			 * We discovered the size of the shm segment, so
1288			 * break out of here and fall through to the next
1289			 * loop that uses the size information to stop
1290			 * searching for matching vma's.
1291			 */
1292			retval = 0;
1293			vma = next;
1294			break;
1295		}
1296		vma = next;
1297	}
1298
1299	/*
1300	 * We need look no further than the maximum address a fragment
1301	 * could possibly have landed at. Also cast things to loff_t to
1302	 * prevent overflows and make comparisons vs. equal-width types.
1303	 */
1304	size = PAGE_ALIGN(size);
1305	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1306		next = vma->vm_next;
1307
1308		/* finding a matching vma now does not alter retval */
1309		if ((vma->vm_ops == &shm_vm_ops) &&
1310		    ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1311		    (vma->vm_file == file))
1312			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1313		vma = next;
1314	}
1315
1316#else /* CONFIG_MMU */
1317	/* under NOMMU conditions, the exact address to be destroyed must be
1318	 * given */
1319	if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1320		do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1321		retval = 0;
1322	}
1323
1324#endif
1325
1326	up_write(&mm->mmap_sem);
1327	return retval;
1328}
1329
1330#ifdef CONFIG_PROC_FS
1331static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1332{
1333	struct user_namespace *user_ns = seq_user_ns(s);
1334	struct shmid_kernel *shp = it;
1335	unsigned long rss = 0, swp = 0;
1336
1337	shm_add_rss_swap(shp, &rss, &swp);
1338
1339#if BITS_PER_LONG <= 32
1340#define SIZE_SPEC "%10lu"
1341#else
1342#define SIZE_SPEC "%21lu"
1343#endif
1344
1345	seq_printf(s,
1346		   "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1347		   "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1348		   SIZE_SPEC " " SIZE_SPEC "\n",
1349		   shp->shm_perm.key,
1350		   shp->shm_perm.id,
1351		   shp->shm_perm.mode,
1352		   shp->shm_segsz,
1353		   shp->shm_cprid,
1354		   shp->shm_lprid,
1355		   shp->shm_nattch,
1356		   from_kuid_munged(user_ns, shp->shm_perm.uid),
1357		   from_kgid_munged(user_ns, shp->shm_perm.gid),
1358		   from_kuid_munged(user_ns, shp->shm_perm.cuid),
1359		   from_kgid_munged(user_ns, shp->shm_perm.cgid),
1360		   shp->shm_atim,
1361		   shp->shm_dtim,
1362		   shp->shm_ctim,
1363		   rss * PAGE_SIZE,
1364		   swp * PAGE_SIZE);
1365
1366	return 0;
1367}
1368#endif
1369