projects/linux·blob master · refs

1883 lines · 45.2 KiB

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 <[email protected]>
10 * BIGMEM support, Andrea Arcangeli <[email protected]>
11 * SMP thread shm, Jean-Luc Boyard <[email protected]>
12 * HIGHMEM support, Ingo Molnar <[email protected]>
13 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <[email protected]>
14 * Shared /dev/zero support, Kanoj Sarcar <[email protected]>
15 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <[email protected]>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <[email protected]>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <[email protected]>
23 *
24 * Better ipc lock (kern_ipc_perm.lock) handling
25 * Davidlohr Bueso <[email protected]>, 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 <uapi/linux/shm.h>
33#include <linux/init.h>
34#include <linux/file.h>
35#include <linux/mman.h>
36#include <linux/shmem_fs.h>
37#include <linux/security.h>
38#include <linux/syscalls.h>
39#include <linux/audit.h>
40#include <linux/capability.h>
41#include <linux/ptrace.h>
42#include <linux/seq_file.h>
43#include <linux/rwsem.h>
44#include <linux/nsproxy.h>
45#include <linux/mount.h>
46#include <linux/ipc_namespace.h>
47#include <linux/rhashtable.h>
48#include <linux/nstree.h>
49
50#include <linux/uaccess.h>
51
52#include "util.h"
53
54struct shmid_kernel /* private to the kernel */
55{
56 struct kern_ipc_perm shm_perm;
57 struct file *shm_file;
58 unsigned long shm_nattch;
59 unsigned long shm_segsz;
60 time64_t shm_atim;
61 time64_t shm_dtim;
62 time64_t shm_ctim;
63 struct pid *shm_cprid;
64 struct pid *shm_lprid;
65 struct ucounts *mlock_ucounts;
66
67 /*
68 * The task created the shm object, for
69 * task_lock(shp->shm_creator)
70 */
71 struct task_struct *shm_creator;
72
73 /*
74 * List by creator. task_lock(->shm_creator) required for read/write.
75 * If list_empty(), then the creator is dead already.
76 */
77 struct list_head shm_clist;
78 struct ipc_namespace *ns;
79} __randomize_layout;
80
81/* shm_mode upper byte flags */
82#define SHM_DEST 01000 /* segment will be destroyed on last detach */
83#define SHM_LOCKED 02000 /* segment will not be swapped */
84
85struct shm_file_data {
86 int id;
87 struct ipc_namespace *ns;
88 struct file *file;
89 const struct vm_operations_struct *vm_ops;
90};
91
92#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
93
94static const struct file_operations shm_file_operations;
95static const struct vm_operations_struct shm_vm_ops;
96
97#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
98
99#define shm_unlock(shp) \
100 ipc_unlock(&(shp)->shm_perm)
101
102static int newseg(struct ipc_namespace *, struct ipc_params *);
103static void shm_open(struct vm_area_struct *vma);
104static void shm_close(struct vm_area_struct *vma);
105static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
106#ifdef CONFIG_PROC_FS
107static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
108#endif
109
110void shm_init_ns(struct ipc_namespace *ns)
111{
112 ns->shm_ctlmax = SHMMAX;
113 ns->shm_ctlall = SHMALL;
114 ns->shm_ctlmni = SHMMNI;
115 ns->shm_rmid_forced = 0;
116 ns->shm_tot = 0;
117 ipc_init_ids(&shm_ids(ns));
118}
119
120/*
121 * Called with shm_ids.rwsem (writer) and the shp structure locked.
122 * Only shm_ids.rwsem remains locked on exit.
123 */
124static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
125{
126 struct shmid_kernel *shp;
127
128 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
129 WARN_ON(ns != shp->ns);
130
131 if (shp->shm_nattch) {
132 shp->shm_perm.mode |= SHM_DEST;
133 /* Do not find it any more */
134 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
135 shm_unlock(shp);
136 } else
137 shm_destroy(ns, shp);
138}
139
140#ifdef CONFIG_IPC_NS
141void shm_exit_ns(struct ipc_namespace *ns)
142{
143 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
144 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
145 rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
146}
147#endif
148
149static int __init ipc_ns_init(void)
150{
151 shm_init_ns(&init_ipc_ns);
152 ns_tree_add(&init_ipc_ns);
153 return 0;
154}
155
156pure_initcall(ipc_ns_init);
157
158void __init shm_init(void)
159{
160 ipc_init_proc_interface("sysvipc/shm",
161#if BITS_PER_LONG <= 32
162 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
163#else
164 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
165#endif
166 IPC_SHM_IDS, sysvipc_shm_proc_show);
167}
168
169static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
170{
171 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
172
173 if (IS_ERR(ipcp))
174 return ERR_CAST(ipcp);
175
176 return container_of(ipcp, struct shmid_kernel, shm_perm);
177}
178
179static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
180{
181 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
182
183 if (IS_ERR(ipcp))
184 return ERR_CAST(ipcp);
185
186 return container_of(ipcp, struct shmid_kernel, shm_perm);
187}
188
189/*
190 * shm_lock_(check_) routines are called in the paths where the rwsem
191 * is not necessarily held.
192 */
193static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
194{
195 struct kern_ipc_perm *ipcp;
196
197 rcu_read_lock();
198 ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
199 if (IS_ERR(ipcp))
200 goto err;
201
202 ipc_lock_object(ipcp);
203 /*
204 * ipc_rmid() may have already freed the ID while ipc_lock_object()
205 * was spinning: here verify that the structure is still valid.
206 * Upon races with RMID, return -EIDRM, thus indicating that
207 * the ID points to a removed identifier.
208 */
209 if (ipc_valid_object(ipcp)) {
210 /* return a locked ipc object upon success */
211 return container_of(ipcp, struct shmid_kernel, shm_perm);
212 }
213
214 ipc_unlock_object(ipcp);
215 ipcp = ERR_PTR(-EIDRM);
216err:
217 rcu_read_unlock();
218 /*
219 * Callers of shm_lock() must validate the status of the returned ipc
220 * object pointer and error out as appropriate.
221 */
222 return ERR_CAST(ipcp);
223}
224
225static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
226{
227 rcu_read_lock();
228 ipc_lock_object(&ipcp->shm_perm);
229}
230
231static void shm_rcu_free(struct rcu_head *head)
232{
233 struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
234 rcu);
235 struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
236 shm_perm);
237 security_shm_free(&shp->shm_perm);
238 kfree(shp);
239}
240
241/*
242 * It has to be called with shp locked.
243 * It must be called before ipc_rmid()
244 */
245static inline void shm_clist_rm(struct shmid_kernel *shp)
246{
247 struct task_struct *creator;
248
249 /* ensure that shm_creator does not disappear */
250 rcu_read_lock();
251
252 /*
253 * A concurrent exit_shm may do a list_del_init() as well.
254 * Just do nothing if exit_shm already did the work
255 */
256 if (!list_empty(&shp->shm_clist)) {
257 /*
258 * shp->shm_creator is guaranteed to be valid *only*
259 * if shp->shm_clist is not empty.
260 */
261 creator = shp->shm_creator;
262
263 task_lock(creator);
264 /*
265 * list_del_init() is a nop if the entry was already removed
266 * from the list.
267 */
268 list_del_init(&shp->shm_clist);
269 task_unlock(creator);
270 }
271 rcu_read_unlock();
272}
273
274static inline void shm_rmid(struct shmid_kernel *s)
275{
276 shm_clist_rm(s);
277 ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
278}
279
280
281static int __shm_open(struct shm_file_data *sfd)
282{
283 struct shmid_kernel *shp;
284
285 shp = shm_lock(sfd->ns, sfd->id);
286
287 if (IS_ERR(shp))
288 return PTR_ERR(shp);
289
290 if (shp->shm_file != sfd->file) {
291 /* ID was reused */
292 shm_unlock(shp);
293 return -EINVAL;
294 }
295
296 shp->shm_atim = ktime_get_real_seconds();
297 ipc_update_pid(&shp->shm_lprid, task_tgid(current));
298 shp->shm_nattch++;
299 shm_unlock(shp);
300 return 0;
301}
302
303/* This is called by fork, once for every shm attach. */
304static void shm_open(struct vm_area_struct *vma)
305{
306 struct file *file = vma->vm_file;
307 struct shm_file_data *sfd = shm_file_data(file);
308 int err;
309
310 /* Always call underlying open if present */
311 if (sfd->vm_ops->open)
312 sfd->vm_ops->open(vma);
313
314 err = __shm_open(sfd);
315 /*
316 * We raced in the idr lookup or with shm_destroy().
317 * Either way, the ID is busted.
318 */
319 WARN_ON_ONCE(err);
320}
321
322/*
323 * shm_destroy - free the struct shmid_kernel
324 *
325 * @ns: namespace
326 * @shp: struct to free
327 *
328 * It has to be called with shp and shm_ids.rwsem (writer) locked,
329 * but returns with shp unlocked and freed.
330 */
331static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
332{
333 struct file *shm_file;
334
335 shm_file = shp->shm_file;
336 shp->shm_file = NULL;
337 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
338 shm_rmid(shp);
339 shm_unlock(shp);
340 if (!is_file_hugepages(shm_file))
341 shmem_lock(shm_file, 0, shp->mlock_ucounts);
342 fput(shm_file);
343 ipc_update_pid(&shp->shm_cprid, NULL);
344 ipc_update_pid(&shp->shm_lprid, NULL);
345 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
346}
347
348/*
349 * shm_may_destroy - identifies whether shm segment should be destroyed now
350 *
351 * Returns true if and only if there are no active users of the segment and
352 * one of the following is true:
353 *
354 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
355 *
356 * 2) sysctl kernel.shm_rmid_forced is set to 1.
357 */
358static bool shm_may_destroy(struct shmid_kernel *shp)
359{
360 return (shp->shm_nattch == 0) &&
361 (shp->ns->shm_rmid_forced ||
362 (shp->shm_perm.mode & SHM_DEST));
363}
364
365/*
366 * remove the attach descriptor vma.
367 * free memory for segment if it is marked destroyed.
368 * The descriptor has already been removed from the current->mm->mmap list
369 * and will later be kfree()d.
370 */
371static void __shm_close(struct shm_file_data *sfd)
372{
373 struct shmid_kernel *shp;
374 struct ipc_namespace *ns = sfd->ns;
375
376 down_write(&shm_ids(ns).rwsem);
377 /* remove from the list of attaches of the shm segment */
378 shp = shm_lock(ns, sfd->id);
379
380 /*
381 * We raced in the idr lookup or with shm_destroy().
382 * Either way, the ID is busted.
383 */
384 if (WARN_ON_ONCE(IS_ERR(shp)))
385 goto done; /* no-op */
386
387 ipc_update_pid(&shp->shm_lprid, task_tgid(current));
388 shp->shm_dtim = ktime_get_real_seconds();
389 shp->shm_nattch--;
390 if (shm_may_destroy(shp))
391 shm_destroy(ns, shp);
392 else
393 shm_unlock(shp);
394done:
395 up_write(&shm_ids(ns).rwsem);
396}
397
398static void shm_close(struct vm_area_struct *vma)
399{
400 struct file *file = vma->vm_file;
401 struct shm_file_data *sfd = shm_file_data(file);
402
403 /* Always call underlying close if present */
404 if (sfd->vm_ops->close)
405 sfd->vm_ops->close(vma);
406
407 __shm_close(sfd);
408}
409
410/* Called with ns->shm_ids(ns).rwsem locked */
411static int shm_try_destroy_orphaned(int id, void *p, void *data)
412{
413 struct ipc_namespace *ns = data;
414 struct kern_ipc_perm *ipcp = p;
415 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
416
417 /*
418 * We want to destroy segments without users and with already
419 * exit'ed originating process.
420 *
421 * shm_nattch can be changed under shm_perm.lock without holding the
422 * rwsem, so take the object lock before checking shm_may_destroy().
423 */
424 if (!list_empty(&shp->shm_clist))
425 return 0;
426
427 shm_lock_by_ptr(shp);
428 if (shm_may_destroy(shp))
429 shm_destroy(ns, shp);
430 else
431 shm_unlock(shp);
432 return 0;
433}
434
435void shm_destroy_orphaned(struct ipc_namespace *ns)
436{
437 down_write(&shm_ids(ns).rwsem);
438 if (shm_ids(ns).in_use) {
439 rcu_read_lock();
440 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
441 rcu_read_unlock();
442 }
443 up_write(&shm_ids(ns).rwsem);
444}
445
446/* Locking assumes this will only be called with task == current */
447void exit_shm(struct task_struct *task)
448{
449 for (;;) {
450 struct shmid_kernel *shp;
451 struct ipc_namespace *ns;
452
453 task_lock(task);
454
455 if (list_empty(&task->sysvshm.shm_clist)) {
456 task_unlock(task);
457 break;
458 }
459
460 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
461 shm_clist);
462
463 /*
464 * 1) Get pointer to the ipc namespace. It is worth to say
465 * that this pointer is guaranteed to be valid because
466 * shp lifetime is always shorter than namespace lifetime
467 * in which shp lives.
468 * We taken task_lock it means that shp won't be freed.
469 */
470 ns = shp->ns;
471
472 /*
473 * 2) If kernel.shm_rmid_forced is not set then only keep track of
474 * which shmids are orphaned, so that a later set of the sysctl
475 * can clean them up.
476 */
477 if (!ns->shm_rmid_forced)
478 goto unlink_continue;
479
480 /*
481 * 3) get a reference to the namespace.
482 * The refcount could be already 0. If it is 0, then
483 * the shm objects will be free by free_ipc_work().
484 */
485 ns = get_ipc_ns_not_zero(ns);
486 if (!ns) {
487unlink_continue:
488 list_del_init(&shp->shm_clist);
489 task_unlock(task);
490 continue;
491 }
492
493 /*
494 * 4) get a reference to shp.
495 * This cannot fail: shm_clist_rm() is called before
496 * ipc_rmid(), thus the refcount cannot be 0.
497 */
498 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
499
500 /*
501 * 5) unlink the shm segment from the list of segments
502 * created by current.
503 * This must be done last. After unlinking,
504 * only the refcounts obtained above prevent IPC_RMID
505 * from destroying the segment or the namespace.
506 */
507 list_del_init(&shp->shm_clist);
508
509 task_unlock(task);
510
511 /*
512 * 6) we have all references
513 * Thus lock & if needed destroy shp.
514 */
515 down_write(&shm_ids(ns).rwsem);
516 shm_lock_by_ptr(shp);
517 /*
518 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
519 * safe to call ipc_rcu_putref here
520 */
521 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
522
523 if (ipc_valid_object(&shp->shm_perm)) {
524 if (shm_may_destroy(shp))
525 shm_destroy(ns, shp);
526 else
527 shm_unlock(shp);
528 } else {
529 /*
530 * Someone else deleted the shp from namespace
531 * idr/kht while we have waited.
532 * Just unlock and continue.
533 */
534 shm_unlock(shp);
535 }
536
537 up_write(&shm_ids(ns).rwsem);
538 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
539 }
540}
541
542static vm_fault_t shm_fault(struct vm_fault *vmf)
543{
544 struct file *file = vmf->vma->vm_file;
545 struct shm_file_data *sfd = shm_file_data(file);
546
547 return sfd->vm_ops->fault(vmf);
548}
549
550static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
551{
552 struct file *file = vma->vm_file;
553 struct shm_file_data *sfd = shm_file_data(file);
554
555 if (sfd->vm_ops->may_split)
556 return sfd->vm_ops->may_split(vma, addr);
557
558 return 0;
559}
560
561static unsigned long shm_pagesize(struct vm_area_struct *vma)
562{
563 struct file *file = vma->vm_file;
564 struct shm_file_data *sfd = shm_file_data(file);
565
566 if (sfd->vm_ops->pagesize)
567 return sfd->vm_ops->pagesize(vma);
568
569 return PAGE_SIZE;
570}
571
572#ifdef CONFIG_NUMA
573static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
574{
575 struct shm_file_data *sfd = shm_file_data(vma->vm_file);
576 int err = 0;
577
578 if (sfd->vm_ops->set_policy)
579 err = sfd->vm_ops->set_policy(vma, mpol);
580 return err;
581}
582
583static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
584 unsigned long addr, pgoff_t *ilx)
585{
586 struct shm_file_data *sfd = shm_file_data(vma->vm_file);
587 struct mempolicy *mpol = vma->vm_policy;
588
589 if (sfd->vm_ops->get_policy)
590 mpol = sfd->vm_ops->get_policy(vma, addr, ilx);
591 return mpol;
592}
593#endif
594
595static int shm_mmap(struct file *file, struct vm_area_struct *vma)
596{
597 struct shm_file_data *sfd = shm_file_data(file);
598 int ret;
599
600 /*
601 * In case of remap_file_pages() emulation, the file can represent an
602 * IPC ID that was removed, and possibly even reused by another shm
603 * segment already. Propagate this case as an error to caller.
604 */
605 ret = __shm_open(sfd);
606 if (ret)
607 return ret;
608
609 ret = vfs_mmap(sfd->file, vma);
610 if (ret) {
611 __shm_close(sfd);
612 return ret;
613 }
614 sfd->vm_ops = vma->vm_ops;
615#ifdef CONFIG_MMU
616 WARN_ON(!sfd->vm_ops->fault);
617#endif
618 vma->vm_ops = &shm_vm_ops;
619 return 0;
620}
621
622static int shm_release(struct inode *ino, struct file *file)
623{
624 struct shm_file_data *sfd = shm_file_data(file);
625
626 put_ipc_ns(sfd->ns);
627 fput(sfd->file);
628 shm_file_data(file) = NULL;
629 kfree(sfd);
630 return 0;
631}
632
633static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
634{
635 struct shm_file_data *sfd = shm_file_data(file);
636
637 if (!sfd->file->f_op->fsync)
638 return -EINVAL;
639 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
640}
641
642static long shm_fallocate(struct file *file, int mode, loff_t offset,
643 loff_t len)
644{
645 struct shm_file_data *sfd = shm_file_data(file);
646
647 if (!sfd->file->f_op->fallocate)
648 return -EOPNOTSUPP;
649 return sfd->file->f_op->fallocate(file, mode, offset, len);
650}
651
652static unsigned long shm_get_unmapped_area(struct file *file,
653 unsigned long addr, unsigned long len, unsigned long pgoff,
654 unsigned long flags)
655{
656 struct shm_file_data *sfd = shm_file_data(file);
657
658 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
659 pgoff, flags);
660}
661
662static const struct file_operations shm_file_operations = {
663 .mmap = shm_mmap,
664 .fsync = shm_fsync,
665 .release = shm_release,
666 .get_unmapped_area = shm_get_unmapped_area,
667 .llseek = noop_llseek,
668 .fallocate = shm_fallocate,
669};
670
671/*
672 * shm_file_operations_huge is now identical to shm_file_operations
673 * except for fop_flags
674 */
675static const struct file_operations shm_file_operations_huge = {
676 .mmap = shm_mmap,
677 .fsync = shm_fsync,
678 .release = shm_release,
679 .get_unmapped_area = shm_get_unmapped_area,
680 .llseek = noop_llseek,
681 .fallocate = shm_fallocate,
682 .fop_flags = FOP_HUGE_PAGES,
683};
684
685static const struct vm_operations_struct shm_vm_ops = {
686 .open = shm_open, /* callback for a new vm-area open */
687 .close = shm_close, /* callback for when the vm-area is released */
688 .fault = shm_fault,
689 .may_split = shm_may_split,
690 .pagesize = shm_pagesize,
691#if defined(CONFIG_NUMA)
692 .set_policy = shm_set_policy,
693 .get_policy = shm_get_policy,
694#endif
695};
696
697/**
698 * newseg - Create a new shared memory segment
699 * @ns: namespace
700 * @params: ptr to the structure that contains key, size and shmflg
701 *
702 * Called with shm_ids.rwsem held as a writer.
703 */
704static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
705{
706 key_t key = params->key;
707 int shmflg = params->flg;
708 size_t size = params->u.size;
709 int error;
710 struct shmid_kernel *shp;
711 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
712 const bool has_no_reserve = shmflg & SHM_NORESERVE;
713 vma_flags_t acctflag = EMPTY_VMA_FLAGS;
714 struct file *file;
715 char name[13];
716
717 if (size < SHMMIN || size > ns->shm_ctlmax)
718 return -EINVAL;
719
720 if (numpages << PAGE_SHIFT < size)
721 return -ENOSPC;
722
723 if (ns->shm_tot + numpages < ns->shm_tot ||
724 ns->shm_tot + numpages > ns->shm_ctlall)
725 return -ENOSPC;
726
727 shp = kmalloc_obj(*shp, GFP_KERNEL_ACCOUNT);
728 if (unlikely(!shp))
729 return -ENOMEM;
730
731 shp->shm_perm.key = key;
732 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
733 shp->mlock_ucounts = NULL;
734
735 shp->shm_perm.security = NULL;
736 error = security_shm_alloc(&shp->shm_perm);
737 if (error) {
738 kfree(shp);
739 return error;
740 }
741
742 sprintf(name, "SYSV%08x", key);
743 if (shmflg & SHM_HUGETLB) {
744 struct hstate *hs;
745 size_t hugesize;
746
747 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
748 if (!hs) {
749 error = -EINVAL;
750 goto no_file;
751 }
752 hugesize = ALIGN(size, huge_page_size(hs));
753
754 /* hugetlb_file_setup applies strict accounting */
755 if (has_no_reserve)
756 vma_flags_set(&acctflag, VMA_NORESERVE_BIT);
757 file = hugetlb_file_setup(name, hugesize, acctflag,
758 HUGETLB_SHMFS_INODE, (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
759 } else {
760 /*
761 * Do not allow no accounting for OVERCOMMIT_NEVER, even
762 * if it's asked for.
763 */
764 if (has_no_reserve && sysctl_overcommit_memory != OVERCOMMIT_NEVER)
765 vma_flags_set(&acctflag, VMA_NORESERVE_BIT);
766 file = shmem_kernel_file_setup(name, size, acctflag);
767 }
768 error = PTR_ERR(file);
769 if (IS_ERR(file))
770 goto no_file;
771
772 shp->shm_cprid = get_pid(task_tgid(current));
773 shp->shm_lprid = NULL;
774 shp->shm_atim = shp->shm_dtim = 0;
775 shp->shm_ctim = ktime_get_real_seconds();
776 shp->shm_segsz = size;
777 shp->shm_nattch = 0;
778 shp->shm_file = file;
779 shp->shm_creator = current;
780
781 /* ipc_addid() locks shp upon success. */
782 error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
783 if (error < 0)
784 goto no_id;
785
786 shp->ns = ns;
787
788 task_lock(current);
789 list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
790 task_unlock(current);
791
792 /*
793 * shmid gets reported as "inode#" in /proc/pid/maps.
794 * proc-ps tools use this. Changing this will break them.
795 */
796 file_inode(file)->i_ino = shp->shm_perm.id;
797
798 ns->shm_tot += numpages;
799 error = shp->shm_perm.id;
800
801 ipc_unlock_object(&shp->shm_perm);
802 rcu_read_unlock();
803 return error;
804
805no_id:
806 ipc_update_pid(&shp->shm_cprid, NULL);
807 ipc_update_pid(&shp->shm_lprid, NULL);
808 fput(file);
809 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
810 return error;
811no_file:
812 call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
813 return error;
814}
815
816/*
817 * Called with shm_ids.rwsem and ipcp locked.
818 */
819static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
820{
821 struct shmid_kernel *shp;
822
823 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
824 if (shp->shm_segsz < params->u.size)
825 return -EINVAL;
826
827 return 0;
828}
829
830long ksys_shmget(key_t key, size_t size, int shmflg)
831{
832 struct ipc_namespace *ns;
833 static const struct ipc_ops shm_ops = {
834 .getnew = newseg,
835 .associate = security_shm_associate,
836 .more_checks = shm_more_checks,
837 };
838 struct ipc_params shm_params;
839
840 ns = current->nsproxy->ipc_ns;
841
842 shm_params.key = key;
843 shm_params.flg = shmflg;
844 shm_params.u.size = size;
845
846 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
847}
848
849SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
850{
851 return ksys_shmget(key, size, shmflg);
852}
853
854static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
855{
856 switch (version) {
857 case IPC_64:
858 return copy_to_user(buf, in, sizeof(*in));
859 case IPC_OLD:
860 {
861 struct shmid_ds out;
862
863 memset(&out, 0, sizeof(out));
864 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
865 out.shm_segsz = in->shm_segsz;
866 out.shm_atime = in->shm_atime;
867 out.shm_dtime = in->shm_dtime;
868 out.shm_ctime = in->shm_ctime;
869 out.shm_cpid = in->shm_cpid;
870 out.shm_lpid = in->shm_lpid;
871 out.shm_nattch = in->shm_nattch;
872
873 return copy_to_user(buf, &out, sizeof(out));
874 }
875 default:
876 return -EINVAL;
877 }
878}
879
880static inline unsigned long
881copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
882{
883 switch (version) {
884 case IPC_64:
885 if (copy_from_user(out, buf, sizeof(*out)))
886 return -EFAULT;
887 return 0;
888 case IPC_OLD:
889 {
890 struct shmid_ds tbuf_old;
891
892 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
893 return -EFAULT;
894
895 out->shm_perm.uid = tbuf_old.shm_perm.uid;
896 out->shm_perm.gid = tbuf_old.shm_perm.gid;
897 out->shm_perm.mode = tbuf_old.shm_perm.mode;
898
899 return 0;
900 }
901 default:
902 return -EINVAL;
903 }
904}
905
906static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
907{
908 switch (version) {
909 case IPC_64:
910 return copy_to_user(buf, in, sizeof(*in));
911 case IPC_OLD:
912 {
913 struct shminfo out;
914
915 if (in->shmmax > INT_MAX)
916 out.shmmax = INT_MAX;
917 else
918 out.shmmax = (int)in->shmmax;
919
920 out.shmmin = in->shmmin;
921 out.shmmni = in->shmmni;
922 out.shmseg = in->shmseg;
923 out.shmall = in->shmall;
924
925 return copy_to_user(buf, &out, sizeof(out));
926 }
927 default:
928 return -EINVAL;
929 }
930}
931
932/*
933 * Calculate and add used RSS and swap pages of a shm.
934 * Called with shm_ids.rwsem held as a reader
935 */
936static void shm_add_rss_swap(struct shmid_kernel *shp,
937 unsigned long *rss_add, unsigned long *swp_add)
938{
939 struct inode *inode;
940
941 inode = file_inode(shp->shm_file);
942
943 if (is_file_hugepages(shp->shm_file)) {
944 struct address_space *mapping = inode->i_mapping;
945 struct hstate *h = hstate_file(shp->shm_file);
946 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
947 } else {
948#ifdef CONFIG_SHMEM
949 struct shmem_inode_info *info = SHMEM_I(inode);
950
951 spin_lock_irq(&info->lock);
952 *rss_add += inode->i_mapping->nrpages;
953 *swp_add += info->swapped;
954 spin_unlock_irq(&info->lock);
955#else
956 *rss_add += inode->i_mapping->nrpages;
957#endif
958 }
959}
960
961/*
962 * Called with shm_ids.rwsem held as a reader
963 */
964static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
965 unsigned long *swp)
966{
967 int next_id;
968 int total, in_use;
969
970 *rss = 0;
971 *swp = 0;
972
973 in_use = shm_ids(ns).in_use;
974
975 for (total = 0, next_id = 0; total < in_use; next_id++) {
976 struct kern_ipc_perm *ipc;
977 struct shmid_kernel *shp;
978
979 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
980 if (ipc == NULL)
981 continue;
982 shp = container_of(ipc, struct shmid_kernel, shm_perm);
983
984 shm_add_rss_swap(shp, rss, swp);
985
986 total++;
987 }
988}
989
990/*
991 * This function handles some shmctl commands which require the rwsem
992 * to be held in write mode.
993 * NOTE: no locks must be held, the rwsem is taken inside this function.
994 */
995static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
996 struct shmid64_ds *shmid64)
997{
998 struct kern_ipc_perm *ipcp;
999 struct shmid_kernel *shp;
1000 int err;
1001
1002 down_write(&shm_ids(ns).rwsem);
1003 rcu_read_lock();
1004
1005 ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
1006 &shmid64->shm_perm, 0);
1007 if (IS_ERR(ipcp)) {
1008 err = PTR_ERR(ipcp);
1009 goto out_unlock1;
1010 }
1011
1012 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1013
1014 err = security_shm_shmctl(&shp->shm_perm, cmd);
1015 if (err)
1016 goto out_unlock1;
1017
1018 switch (cmd) {
1019 case IPC_RMID:
1020 ipc_lock_object(&shp->shm_perm);
1021 /* do_shm_rmid unlocks the ipc object and rcu */
1022 do_shm_rmid(ns, ipcp);
1023 goto out_up;
1024 case IPC_SET:
1025 ipc_lock_object(&shp->shm_perm);
1026 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1027 if (err)
1028 goto out_unlock0;
1029 shp->shm_ctim = ktime_get_real_seconds();
1030 break;
1031 default:
1032 err = -EINVAL;
1033 goto out_unlock1;
1034 }
1035
1036out_unlock0:
1037 ipc_unlock_object(&shp->shm_perm);
1038out_unlock1:
1039 rcu_read_unlock();
1040out_up:
1041 up_write(&shm_ids(ns).rwsem);
1042 return err;
1043}
1044
1045static int shmctl_ipc_info(struct ipc_namespace *ns,
1046 struct shminfo64 *shminfo)
1047{
1048 int err = security_shm_shmctl(NULL, IPC_INFO);
1049 if (!err) {
1050 memset(shminfo, 0, sizeof(*shminfo));
1051 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1052 shminfo->shmmax = ns->shm_ctlmax;
1053 shminfo->shmall = ns->shm_ctlall;
1054 shminfo->shmmin = SHMMIN;
1055 down_read(&shm_ids(ns).rwsem);
1056 err = ipc_get_maxidx(&shm_ids(ns));
1057 up_read(&shm_ids(ns).rwsem);
1058 if (err < 0)
1059 err = 0;
1060 }
1061 return err;
1062}
1063
1064static int shmctl_shm_info(struct ipc_namespace *ns,
1065 struct shm_info *shm_info)
1066{
1067 int err = security_shm_shmctl(NULL, SHM_INFO);
1068 if (!err) {
1069 memset(shm_info, 0, sizeof(*shm_info));
1070 down_read(&shm_ids(ns).rwsem);
1071 shm_info->used_ids = shm_ids(ns).in_use;
1072 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1073 shm_info->shm_tot = ns->shm_tot;
1074 shm_info->swap_attempts = 0;
1075 shm_info->swap_successes = 0;
1076 err = ipc_get_maxidx(&shm_ids(ns));
1077 up_read(&shm_ids(ns).rwsem);
1078 if (err < 0)
1079 err = 0;
1080 }
1081 return err;
1082}
1083
1084static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1085 int cmd, struct shmid64_ds *tbuf)
1086{
1087 struct shmid_kernel *shp;
1088 int err;
1089
1090 memset(tbuf, 0, sizeof(*tbuf));
1091
1092 rcu_read_lock();
1093 if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1094 shp = shm_obtain_object(ns, shmid);
1095 if (IS_ERR(shp)) {
1096 err = PTR_ERR(shp);
1097 goto out_unlock;
1098 }
1099 } else { /* IPC_STAT */
1100 shp = shm_obtain_object_check(ns, shmid);
1101 if (IS_ERR(shp)) {
1102 err = PTR_ERR(shp);
1103 goto out_unlock;
1104 }
1105 }
1106
1107 /*
1108 * Semantically SHM_STAT_ANY ought to be identical to
1109 * that functionality provided by the /proc/sysvipc/
1110 * interface. As such, only audit these calls and
1111 * do not do traditional S_IRUGO permission checks on
1112 * the ipc object.
1113 */
1114 if (cmd == SHM_STAT_ANY)
1115 audit_ipc_obj(&shp->shm_perm);
1116 else {
1117 err = -EACCES;
1118 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1119 goto out_unlock;
1120 }
1121
1122 err = security_shm_shmctl(&shp->shm_perm, cmd);
1123 if (err)
1124 goto out_unlock;
1125
1126 ipc_lock_object(&shp->shm_perm);
1127
1128 if (!ipc_valid_object(&shp->shm_perm)) {
1129 ipc_unlock_object(&shp->shm_perm);
1130 err = -EIDRM;
1131 goto out_unlock;
1132 }
1133
1134 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1135 tbuf->shm_segsz = shp->shm_segsz;
1136 tbuf->shm_atime = shp->shm_atim;
1137 tbuf->shm_dtime = shp->shm_dtim;
1138 tbuf->shm_ctime = shp->shm_ctim;
1139#ifndef CONFIG_64BIT
1140 tbuf->shm_atime_high = shp->shm_atim >> 32;
1141 tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1142 tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1143#endif
1144 tbuf->shm_cpid = pid_vnr(shp->shm_cprid);
1145 tbuf->shm_lpid = pid_vnr(shp->shm_lprid);
1146 tbuf->shm_nattch = shp->shm_nattch;
1147
1148 if (cmd == IPC_STAT) {
1149 /*
1150 * As defined in SUS:
1151 * Return 0 on success
1152 */
1153 err = 0;
1154 } else {
1155 /*
1156 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1157 * Return the full id, including the sequence number
1158 */
1159 err = shp->shm_perm.id;
1160 }
1161
1162 ipc_unlock_object(&shp->shm_perm);
1163out_unlock:
1164 rcu_read_unlock();
1165 return err;
1166}
1167
1168static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1169{
1170 struct shmid_kernel *shp;
1171 struct file *shm_file;
1172 int err;
1173
1174 rcu_read_lock();
1175 shp = shm_obtain_object_check(ns, shmid);
1176 if (IS_ERR(shp)) {
1177 err = PTR_ERR(shp);
1178 goto out_unlock1;
1179 }
1180
1181 audit_ipc_obj(&(shp->shm_perm));
1182 err = security_shm_shmctl(&shp->shm_perm, cmd);
1183 if (err)
1184 goto out_unlock1;
1185
1186 ipc_lock_object(&shp->shm_perm);
1187
1188 /* check if shm_destroy() is tearing down shp */
1189 if (!ipc_valid_object(&shp->shm_perm)) {
1190 err = -EIDRM;
1191 goto out_unlock0;
1192 }
1193
1194 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1195 kuid_t euid = current_euid();
1196
1197 if (!uid_eq(euid, shp->shm_perm.uid) &&
1198 !uid_eq(euid, shp->shm_perm.cuid)) {
1199 err = -EPERM;
1200 goto out_unlock0;
1201 }
1202 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1203 err = -EPERM;
1204 goto out_unlock0;
1205 }
1206 }
1207
1208 shm_file = shp->shm_file;
1209 if (is_file_hugepages(shm_file))
1210 goto out_unlock0;
1211
1212 if (cmd == SHM_LOCK) {
1213 struct ucounts *ucounts = current_ucounts();
1214
1215 err = shmem_lock(shm_file, 1, ucounts);
1216 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1217 shp->shm_perm.mode |= SHM_LOCKED;
1218 shp->mlock_ucounts = ucounts;
1219 }
1220 goto out_unlock0;
1221 }
1222
1223 /* SHM_UNLOCK */
1224 if (!(shp->shm_perm.mode & SHM_LOCKED))
1225 goto out_unlock0;
1226 shmem_lock(shm_file, 0, shp->mlock_ucounts);
1227 shp->shm_perm.mode &= ~SHM_LOCKED;
1228 shp->mlock_ucounts = NULL;
1229 get_file(shm_file);
1230 ipc_unlock_object(&shp->shm_perm);
1231 rcu_read_unlock();
1232 shmem_unlock_mapping(shm_file->f_mapping);
1233
1234 fput(shm_file);
1235 return err;
1236
1237out_unlock0:
1238 ipc_unlock_object(&shp->shm_perm);
1239out_unlock1:
1240 rcu_read_unlock();
1241 return err;
1242}
1243
1244static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1245{
1246 int err;
1247 struct ipc_namespace *ns;
1248 struct shmid64_ds sem64;
1249
1250 if (cmd < 0 || shmid < 0)
1251 return -EINVAL;
1252
1253 ns = current->nsproxy->ipc_ns;
1254
1255 switch (cmd) {
1256 case IPC_INFO: {
1257 struct shminfo64 shminfo;
1258 err = shmctl_ipc_info(ns, &shminfo);
1259 if (err < 0)
1260 return err;
1261 if (copy_shminfo_to_user(buf, &shminfo, version))
1262 err = -EFAULT;
1263 return err;
1264 }
1265 case SHM_INFO: {
1266 struct shm_info shm_info;
1267 err = shmctl_shm_info(ns, &shm_info);
1268 if (err < 0)
1269 return err;
1270 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1271 err = -EFAULT;
1272 return err;
1273 }
1274 case SHM_STAT:
1275 case SHM_STAT_ANY:
1276 case IPC_STAT: {
1277 err = shmctl_stat(ns, shmid, cmd, &sem64);
1278 if (err < 0)
1279 return err;
1280 if (copy_shmid_to_user(buf, &sem64, version))
1281 err = -EFAULT;
1282 return err;
1283 }
1284 case IPC_SET:
1285 if (copy_shmid_from_user(&sem64, buf, version))
1286 return -EFAULT;
1287 fallthrough;
1288 case IPC_RMID:
1289 return shmctl_down(ns, shmid, cmd, &sem64);
1290 case SHM_LOCK:
1291 case SHM_UNLOCK:
1292 return shmctl_do_lock(ns, shmid, cmd);
1293 default:
1294 return -EINVAL;
1295 }
1296}
1297
1298SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1299{
1300 return ksys_shmctl(shmid, cmd, buf, IPC_64);
1301}
1302
1303#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1304long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1305{
1306 int version = ipc_parse_version(&cmd);
1307
1308 return ksys_shmctl(shmid, cmd, buf, version);
1309}
1310
1311SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1312{
1313 return ksys_old_shmctl(shmid, cmd, buf);
1314}
1315#endif
1316
1317#ifdef CONFIG_COMPAT
1318
1319struct compat_shmid_ds {
1320 struct compat_ipc_perm shm_perm;
1321 int shm_segsz;
1322 old_time32_t shm_atime;
1323 old_time32_t shm_dtime;
1324 old_time32_t shm_ctime;
1325 compat_ipc_pid_t shm_cpid;
1326 compat_ipc_pid_t shm_lpid;
1327 unsigned short shm_nattch;
1328 unsigned short shm_unused;
1329 compat_uptr_t shm_unused2;
1330 compat_uptr_t shm_unused3;
1331};
1332
1333struct compat_shminfo64 {
1334 compat_ulong_t shmmax;
1335 compat_ulong_t shmmin;
1336 compat_ulong_t shmmni;
1337 compat_ulong_t shmseg;
1338 compat_ulong_t shmall;
1339 compat_ulong_t __unused1;
1340 compat_ulong_t __unused2;
1341 compat_ulong_t __unused3;
1342 compat_ulong_t __unused4;
1343};
1344
1345struct compat_shm_info {
1346 compat_int_t used_ids;
1347 compat_ulong_t shm_tot, shm_rss, shm_swp;
1348 compat_ulong_t swap_attempts, swap_successes;
1349};
1350
1351static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1352 int version)
1353{
1354 if (in->shmmax > INT_MAX)
1355 in->shmmax = INT_MAX;
1356 if (version == IPC_64) {
1357 struct compat_shminfo64 info;
1358 memset(&info, 0, sizeof(info));
1359 info.shmmax = in->shmmax;
1360 info.shmmin = in->shmmin;
1361 info.shmmni = in->shmmni;
1362 info.shmseg = in->shmseg;
1363 info.shmall = in->shmall;
1364 return copy_to_user(buf, &info, sizeof(info));
1365 } else {
1366 struct shminfo info;
1367 memset(&info, 0, sizeof(info));
1368 info.shmmax = in->shmmax;
1369 info.shmmin = in->shmmin;
1370 info.shmmni = in->shmmni;
1371 info.shmseg = in->shmseg;
1372 info.shmall = in->shmall;
1373 return copy_to_user(buf, &info, sizeof(info));
1374 }
1375}
1376
1377static int put_compat_shm_info(struct shm_info *ip,
1378 struct compat_shm_info __user *uip)
1379{
1380 struct compat_shm_info info;
1381
1382 memset(&info, 0, sizeof(info));
1383 info.used_ids = ip->used_ids;
1384 info.shm_tot = ip->shm_tot;
1385 info.shm_rss = ip->shm_rss;
1386 info.shm_swp = ip->shm_swp;
1387 info.swap_attempts = ip->swap_attempts;
1388 info.swap_successes = ip->swap_successes;
1389 return copy_to_user(uip, &info, sizeof(info));
1390}
1391
1392static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1393 int version)
1394{
1395 if (version == IPC_64) {
1396 struct compat_shmid64_ds v;
1397 memset(&v, 0, sizeof(v));
1398 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1399 v.shm_atime = lower_32_bits(in->shm_atime);
1400 v.shm_atime_high = upper_32_bits(in->shm_atime);
1401 v.shm_dtime = lower_32_bits(in->shm_dtime);
1402 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1403 v.shm_ctime = lower_32_bits(in->shm_ctime);
1404 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1405 v.shm_segsz = in->shm_segsz;
1406 v.shm_nattch = in->shm_nattch;
1407 v.shm_cpid = in->shm_cpid;
1408 v.shm_lpid = in->shm_lpid;
1409 return copy_to_user(buf, &v, sizeof(v));
1410 } else {
1411 struct compat_shmid_ds v;
1412 memset(&v, 0, sizeof(v));
1413 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1414 v.shm_perm.key = in->shm_perm.key;
1415 v.shm_atime = in->shm_atime;
1416 v.shm_dtime = in->shm_dtime;
1417 v.shm_ctime = in->shm_ctime;
1418 v.shm_segsz = in->shm_segsz;
1419 v.shm_nattch = in->shm_nattch;
1420 v.shm_cpid = in->shm_cpid;
1421 v.shm_lpid = in->shm_lpid;
1422 return copy_to_user(buf, &v, sizeof(v));
1423 }
1424}
1425
1426static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1427 int version)
1428{
1429 memset(out, 0, sizeof(*out));
1430 if (version == IPC_64) {
1431 struct compat_shmid64_ds __user *p = buf;
1432 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1433 } else {
1434 struct compat_shmid_ds __user *p = buf;
1435 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1436 }
1437}
1438
1439static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1440{
1441 struct ipc_namespace *ns;
1442 struct shmid64_ds sem64;
1443 int err;
1444
1445 ns = current->nsproxy->ipc_ns;
1446
1447 if (cmd < 0 || shmid < 0)
1448 return -EINVAL;
1449
1450 switch (cmd) {
1451 case IPC_INFO: {
1452 struct shminfo64 shminfo;
1453 err = shmctl_ipc_info(ns, &shminfo);
1454 if (err < 0)
1455 return err;
1456 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1457 err = -EFAULT;
1458 return err;
1459 }
1460 case SHM_INFO: {
1461 struct shm_info shm_info;
1462 err = shmctl_shm_info(ns, &shm_info);
1463 if (err < 0)
1464 return err;
1465 if (put_compat_shm_info(&shm_info, uptr))
1466 err = -EFAULT;
1467 return err;
1468 }
1469 case IPC_STAT:
1470 case SHM_STAT_ANY:
1471 case SHM_STAT:
1472 err = shmctl_stat(ns, shmid, cmd, &sem64);
1473 if (err < 0)
1474 return err;
1475 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1476 err = -EFAULT;
1477 return err;
1478
1479 case IPC_SET:
1480 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1481 return -EFAULT;
1482 fallthrough;
1483 case IPC_RMID:
1484 return shmctl_down(ns, shmid, cmd, &sem64);
1485 case SHM_LOCK:
1486 case SHM_UNLOCK:
1487 return shmctl_do_lock(ns, shmid, cmd);
1488 default:
1489 return -EINVAL;
1490 }
1491 return err;
1492}
1493
1494COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1495{
1496 return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1497}
1498
1499#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1500long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1501{
1502 int version = compat_ipc_parse_version(&cmd);
1503
1504 return compat_ksys_shmctl(shmid, cmd, uptr, version);
1505}
1506
1507COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1508{
1509 return compat_ksys_old_shmctl(shmid, cmd, uptr);
1510}
1511#endif
1512#endif
1513
1514/*
1515 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1516 *
1517 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1518 * "raddr" thing points to kernel space, and there has to be a wrapper around
1519 * this.
1520 */
1521long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1522 ulong *raddr, unsigned long shmlba)
1523{
1524 struct shmid_kernel *shp;
1525 unsigned long addr = (unsigned long)shmaddr;
1526 unsigned long size;
1527 struct file *file, *base;
1528 int err;
1529 unsigned long flags = MAP_SHARED;
1530 unsigned long prot;
1531 int acc_mode;
1532 struct ipc_namespace *ns;
1533 struct shm_file_data *sfd;
1534 int f_flags;
1535 unsigned long populate = 0;
1536
1537 err = -EINVAL;
1538 if (shmid < 0)
1539 goto out;
1540
1541 if (addr) {
1542 if (addr & (shmlba - 1)) {
1543 if (shmflg & SHM_RND) {
1544 addr &= ~(shmlba - 1); /* round down */
1545
1546 /*
1547 * Ensure that the round-down is non-nil
1548 * when remapping. This can happen for
1549 * cases when addr < shmlba.
1550 */
1551 if (!addr && (shmflg & SHM_REMAP))
1552 goto out;
1553 } else
1554#ifndef __ARCH_FORCE_SHMLBA
1555 if (addr & ~PAGE_MASK)
1556#endif
1557 goto out;
1558 }
1559
1560 flags |= MAP_FIXED;
1561 } else if ((shmflg & SHM_REMAP))
1562 goto out;
1563
1564 if (shmflg & SHM_RDONLY) {
1565 prot = PROT_READ;
1566 acc_mode = S_IRUGO;
1567 f_flags = O_RDONLY;
1568 } else {
1569 prot = PROT_READ | PROT_WRITE;
1570 acc_mode = S_IRUGO | S_IWUGO;
1571 f_flags = O_RDWR;
1572 }
1573 if (shmflg & SHM_EXEC) {
1574 prot |= PROT_EXEC;
1575 acc_mode |= S_IXUGO;
1576 }
1577
1578 /*
1579 * We cannot rely on the fs check since SYSV IPC does have an
1580 * additional creator id...
1581 */
1582 ns = current->nsproxy->ipc_ns;
1583 rcu_read_lock();
1584 shp = shm_obtain_object_check(ns, shmid);
1585 if (IS_ERR(shp)) {
1586 err = PTR_ERR(shp);
1587 goto out_unlock;
1588 }
1589
1590 err = -EACCES;
1591 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1592 goto out_unlock;
1593
1594 err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1595 if (err)
1596 goto out_unlock;
1597
1598 ipc_lock_object(&shp->shm_perm);
1599
1600 /* check if shm_destroy() is tearing down shp */
1601 if (!ipc_valid_object(&shp->shm_perm)) {
1602 ipc_unlock_object(&shp->shm_perm);
1603 err = -EIDRM;
1604 goto out_unlock;
1605 }
1606
1607 /*
1608 * We need to take a reference to the real shm file to prevent the
1609 * pointer from becoming stale in cases where the lifetime of the outer
1610 * file extends beyond that of the shm segment. It's not usually
1611 * possible, but it can happen during remap_file_pages() emulation as
1612 * that unmaps the memory, then does ->mmap() via file reference only.
1613 * We'll deny the ->mmap() if the shm segment was since removed, but to
1614 * detect shm ID reuse we need to compare the file pointers.
1615 */
1616 base = get_file(shp->shm_file);
1617 shp->shm_nattch++;
1618 size = i_size_read(file_inode(base));
1619 ipc_unlock_object(&shp->shm_perm);
1620 rcu_read_unlock();
1621
1622 err = -ENOMEM;
1623 sfd = kzalloc_obj(*sfd);
1624 if (!sfd) {
1625 fput(base);
1626 goto out_nattch;
1627 }
1628
1629 file = alloc_file_clone(base, f_flags,
1630 is_file_hugepages(base) ?
1631 &shm_file_operations_huge :
1632 &shm_file_operations);
1633 err = PTR_ERR(file);
1634 if (IS_ERR(file)) {
1635 kfree(sfd);
1636 fput(base);
1637 goto out_nattch;
1638 }
1639
1640 sfd->id = shp->shm_perm.id;
1641 sfd->ns = get_ipc_ns(ns);
1642 sfd->file = base;
1643 sfd->vm_ops = NULL;
1644 file->private_data = sfd;
1645
1646 err = security_mmap_file(file, prot, flags);
1647 if (err)
1648 goto out_fput;
1649
1650 if (mmap_write_lock_killable(current->mm)) {
1651 err = -EINTR;
1652 goto out_fput;
1653 }
1654
1655 if (addr && !(shmflg & SHM_REMAP)) {
1656 err = -EINVAL;
1657 if (addr + size < addr)
1658 goto invalid;
1659
1660 if (find_vma_intersection(current->mm, addr, addr + size))
1661 goto invalid;
1662 }
1663
1664 addr = do_mmap(file, addr, size, prot, flags, EMPTY_VMA_FLAGS, 0,
1665 &populate, NULL);
1666 *raddr = addr;
1667 err = 0;
1668 if (IS_ERR_VALUE(addr))
1669 err = (long)addr;
1670invalid:
1671 mmap_write_unlock(current->mm);
1672 if (populate)
1673 mm_populate(addr, populate);
1674
1675out_fput:
1676 fput(file);
1677
1678out_nattch:
1679 down_write(&shm_ids(ns).rwsem);
1680 shp = shm_lock(ns, shmid);
1681 shp->shm_nattch--;
1682
1683 if (shm_may_destroy(shp))
1684 shm_destroy(ns, shp);
1685 else
1686 shm_unlock(shp);
1687 up_write(&shm_ids(ns).rwsem);
1688 return err;
1689
1690out_unlock:
1691 rcu_read_unlock();
1692out:
1693 return err;
1694}
1695
1696SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1697{
1698 unsigned long ret;
1699 long err;
1700
1701 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1702 if (err)
1703 return err;
1704 force_successful_syscall_return();
1705 return (long)ret;
1706}
1707
1708#ifdef CONFIG_COMPAT
1709
1710#ifndef COMPAT_SHMLBA
1711#define COMPAT_SHMLBA SHMLBA
1712#endif
1713
1714COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1715{
1716 unsigned long ret;
1717 long err;
1718
1719 err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1720 if (err)
1721 return err;
1722 force_successful_syscall_return();
1723 return (long)ret;
1724}
1725#endif
1726
1727/*
1728 * detach and kill segment if marked destroyed.
1729 * The work is done in shm_close.
1730 */
1731long ksys_shmdt(char __user *shmaddr)
1732{
1733 struct mm_struct *mm = current->mm;
1734 struct vm_area_struct *vma;
1735 unsigned long addr = (unsigned long)shmaddr;
1736 int retval = -EINVAL;
1737#ifdef CONFIG_MMU
1738 loff_t size = 0;
1739 struct file *file;
1740 VMA_ITERATOR(vmi, mm, addr);
1741#endif
1742
1743 if (addr & ~PAGE_MASK)
1744 return retval;
1745
1746 if (mmap_write_lock_killable(mm))
1747 return -EINTR;
1748
1749 /*
1750 * This function tries to be smart and unmap shm segments that
1751 * were modified by partial mlock or munmap calls:
1752 * - It first determines the size of the shm segment that should be
1753 * unmapped: It searches for a vma that is backed by shm and that
1754 * started at address shmaddr. It records it's size and then unmaps
1755 * it.
1756 * - Then it unmaps all shm vmas that started at shmaddr and that
1757 * are within the initially determined size and that are from the
1758 * same shm segment from which we determined the size.
1759 * Errors from do_munmap are ignored: the function only fails if
1760 * it's called with invalid parameters or if it's called to unmap
1761 * a part of a vma. Both calls in this function are for full vmas,
1762 * the parameters are directly copied from the vma itself and always
1763 * valid - therefore do_munmap cannot fail. (famous last words?)
1764 */
1765 /*
1766 * If it had been mremap()'d, the starting address would not
1767 * match the usual checks anyway. So assume all vma's are
1768 * above the starting address given.
1769 */
1770
1771#ifdef CONFIG_MMU
1772 for_each_vma(vmi, vma) {
1773 /*
1774 * Check if the starting address would match, i.e. it's
1775 * a fragment created by mprotect() and/or munmap(), or it
1776 * otherwise it starts at this address with no hassles.
1777 */
1778 if ((vma->vm_ops == &shm_vm_ops) &&
1779 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1780
1781 /*
1782 * Record the file of the shm segment being
1783 * unmapped. With mremap(), someone could place
1784 * page from another segment but with equal offsets
1785 * in the range we are unmapping.
1786 */
1787 file = vma->vm_file;
1788 size = i_size_read(file_inode(vma->vm_file));
1789 do_vmi_align_munmap(&vmi, vma, mm, vma->vm_start,
1790 vma->vm_end, NULL, false);
1791 /*
1792 * We discovered the size of the shm segment, so
1793 * break out of here and fall through to the next
1794 * loop that uses the size information to stop
1795 * searching for matching vma's.
1796 */
1797 retval = 0;
1798 vma = vma_next(&vmi);
1799 break;
1800 }
1801 }
1802
1803 /*
1804 * We need look no further than the maximum address a fragment
1805 * could possibly have landed at. Also cast things to loff_t to
1806 * prevent overflows and make comparisons vs. equal-width types.
1807 */
1808 size = PAGE_ALIGN(size);
1809 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1810 /* finding a matching vma now does not alter retval */
1811 if ((vma->vm_ops == &shm_vm_ops) &&
1812 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1813 (vma->vm_file == file)) {
1814 do_vmi_align_munmap(&vmi, vma, mm, vma->vm_start,
1815 vma->vm_end, NULL, false);
1816 }
1817
1818 vma = vma_next(&vmi);
1819 }
1820
1821#else /* CONFIG_MMU */
1822 vma = vma_lookup(mm, addr);
1823 /* under NOMMU conditions, the exact address to be destroyed must be
1824 * given
1825 */
1826 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1827 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1828 retval = 0;
1829 }
1830
1831#endif
1832
1833 mmap_write_unlock(mm);
1834 return retval;
1835}
1836
1837SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1838{
1839 return ksys_shmdt(shmaddr);
1840}
1841
1842#ifdef CONFIG_PROC_FS
1843static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1844{
1845 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1846 struct user_namespace *user_ns = seq_user_ns(s);
1847 struct kern_ipc_perm *ipcp = it;
1848 struct shmid_kernel *shp;
1849 unsigned long rss = 0, swp = 0;
1850
1851 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1852 shm_add_rss_swap(shp, &rss, &swp);
1853
1854#if BITS_PER_LONG <= 32
1855#define SIZE_SPEC "%10lu"
1856#else
1857#define SIZE_SPEC "%21lu"
1858#endif
1859
1860 seq_printf(s,
1861 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1862 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1863 SIZE_SPEC " " SIZE_SPEC "\n",
1864 shp->shm_perm.key,
1865 shp->shm_perm.id,
1866 shp->shm_perm.mode,
1867 shp->shm_segsz,
1868 pid_nr_ns(shp->shm_cprid, pid_ns),
1869 pid_nr_ns(shp->shm_lprid, pid_ns),
1870 shp->shm_nattch,
1871 from_kuid_munged(user_ns, shp->shm_perm.uid),
1872 from_kgid_munged(user_ns, shp->shm_perm.gid),
1873 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1874 from_kgid_munged(user_ns, shp->shm_perm.cgid),
1875 shp->shm_atim,
1876 shp->shm_dtim,
1877 shp->shm_ctim,
1878 rss * PAGE_SIZE,
1879 swp * PAGE_SIZE);
1880
1881 return 0;
1882}
1883#endif