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