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