Loading...
1/*
2 * linux/ipc/shm.c
3 * Copyright (C) 1992, 1993 Krishna Balasubramanian
4 * Many improvements/fixes by Bruno Haible.
5 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7 *
8 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15 *
16 * support for audit of ipc object properties and permission changes
17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18 *
19 * namespaces support
20 * OpenVZ, SWsoft Inc.
21 * Pavel Emelianov <xemul@openvz.org>
22 *
23 * Better ipc lock (kern_ipc_perm.lock) handling
24 * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
25 */
26
27#include <linux/slab.h>
28#include <linux/mm.h>
29#include <linux/hugetlb.h>
30#include <linux/shm.h>
31#include <linux/init.h>
32#include <linux/file.h>
33#include <linux/mman.h>
34#include <linux/shmem_fs.h>
35#include <linux/security.h>
36#include <linux/syscalls.h>
37#include <linux/audit.h>
38#include <linux/capability.h>
39#include <linux/ptrace.h>
40#include <linux/seq_file.h>
41#include <linux/rwsem.h>
42#include <linux/nsproxy.h>
43#include <linux/mount.h>
44#include <linux/ipc_namespace.h>
45
46#include <linux/uaccess.h>
47
48#include "util.h"
49
50struct shm_file_data {
51 int id;
52 struct ipc_namespace *ns;
53 struct file *file;
54 const struct vm_operations_struct *vm_ops;
55};
56
57#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
58
59static const struct file_operations shm_file_operations;
60static const struct vm_operations_struct shm_vm_ops;
61
62#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
63
64#define shm_unlock(shp) \
65 ipc_unlock(&(shp)->shm_perm)
66
67static int newseg(struct ipc_namespace *, struct ipc_params *);
68static void shm_open(struct vm_area_struct *vma);
69static void shm_close(struct vm_area_struct *vma);
70static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
71#ifdef CONFIG_PROC_FS
72static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
73#endif
74
75void shm_init_ns(struct ipc_namespace *ns)
76{
77 ns->shm_ctlmax = SHMMAX;
78 ns->shm_ctlall = SHMALL;
79 ns->shm_ctlmni = SHMMNI;
80 ns->shm_rmid_forced = 0;
81 ns->shm_tot = 0;
82 ipc_init_ids(&shm_ids(ns));
83}
84
85/*
86 * Called with shm_ids.rwsem (writer) and the shp structure locked.
87 * Only shm_ids.rwsem remains locked on exit.
88 */
89static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
90{
91 struct shmid_kernel *shp;
92 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
93
94 if (shp->shm_nattch) {
95 shp->shm_perm.mode |= SHM_DEST;
96 /* Do not find it any more */
97 shp->shm_perm.key = IPC_PRIVATE;
98 shm_unlock(shp);
99 } else
100 shm_destroy(ns, shp);
101}
102
103#ifdef CONFIG_IPC_NS
104void shm_exit_ns(struct ipc_namespace *ns)
105{
106 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
107 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
108}
109#endif
110
111static int __init ipc_ns_init(void)
112{
113 shm_init_ns(&init_ipc_ns);
114 return 0;
115}
116
117pure_initcall(ipc_ns_init);
118
119void __init shm_init(void)
120{
121 ipc_init_proc_interface("sysvipc/shm",
122#if BITS_PER_LONG <= 32
123 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
124#else
125 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
126#endif
127 IPC_SHM_IDS, sysvipc_shm_proc_show);
128}
129
130static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
131{
132 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
133
134 if (IS_ERR(ipcp))
135 return ERR_CAST(ipcp);
136
137 return container_of(ipcp, struct shmid_kernel, shm_perm);
138}
139
140static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
141{
142 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
143
144 if (IS_ERR(ipcp))
145 return ERR_CAST(ipcp);
146
147 return container_of(ipcp, struct shmid_kernel, shm_perm);
148}
149
150/*
151 * shm_lock_(check_) routines are called in the paths where the rwsem
152 * is not necessarily held.
153 */
154static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
155{
156 struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
157
158 /*
159 * Callers of shm_lock() must validate the status of the returned ipc
160 * object pointer (as returned by ipc_lock()), and error out as
161 * appropriate.
162 */
163 if (IS_ERR(ipcp))
164 return (void *)ipcp;
165 return container_of(ipcp, struct shmid_kernel, shm_perm);
166}
167
168static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
169{
170 rcu_read_lock();
171 ipc_lock_object(&ipcp->shm_perm);
172}
173
174static void shm_rcu_free(struct rcu_head *head)
175{
176 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
177 struct shmid_kernel *shp = ipc_rcu_to_struct(p);
178
179 security_shm_free(shp);
180 ipc_rcu_free(head);
181}
182
183static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
184{
185 list_del(&s->shm_clist);
186 ipc_rmid(&shm_ids(ns), &s->shm_perm);
187}
188
189
190static int __shm_open(struct vm_area_struct *vma)
191{
192 struct file *file = vma->vm_file;
193 struct shm_file_data *sfd = shm_file_data(file);
194 struct shmid_kernel *shp;
195
196 shp = shm_lock(sfd->ns, sfd->id);
197
198 if (IS_ERR(shp))
199 return PTR_ERR(shp);
200
201 shp->shm_atim = get_seconds();
202 shp->shm_lprid = task_tgid_vnr(current);
203 shp->shm_nattch++;
204 shm_unlock(shp);
205 return 0;
206}
207
208/* This is called by fork, once for every shm attach. */
209static void shm_open(struct vm_area_struct *vma)
210{
211 int err = __shm_open(vma);
212 /*
213 * We raced in the idr lookup or with shm_destroy().
214 * Either way, the ID is busted.
215 */
216 WARN_ON_ONCE(err);
217}
218
219/*
220 * shm_destroy - free the struct shmid_kernel
221 *
222 * @ns: namespace
223 * @shp: struct to free
224 *
225 * It has to be called with shp and shm_ids.rwsem (writer) locked,
226 * but returns with shp unlocked and freed.
227 */
228static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
229{
230 struct file *shm_file;
231
232 shm_file = shp->shm_file;
233 shp->shm_file = NULL;
234 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
235 shm_rmid(ns, shp);
236 shm_unlock(shp);
237 if (!is_file_hugepages(shm_file))
238 shmem_lock(shm_file, 0, shp->mlock_user);
239 else if (shp->mlock_user)
240 user_shm_unlock(i_size_read(file_inode(shm_file)),
241 shp->mlock_user);
242 fput(shm_file);
243 ipc_rcu_putref(shp, shm_rcu_free);
244}
245
246/*
247 * shm_may_destroy - identifies whether shm segment should be destroyed now
248 *
249 * Returns true if and only if there are no active users of the segment and
250 * one of the following is true:
251 *
252 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
253 *
254 * 2) sysctl kernel.shm_rmid_forced is set to 1.
255 */
256static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
257{
258 return (shp->shm_nattch == 0) &&
259 (ns->shm_rmid_forced ||
260 (shp->shm_perm.mode & SHM_DEST));
261}
262
263/*
264 * remove the attach descriptor vma.
265 * free memory for segment if it is marked destroyed.
266 * The descriptor has already been removed from the current->mm->mmap list
267 * and will later be kfree()d.
268 */
269static void shm_close(struct vm_area_struct *vma)
270{
271 struct file *file = vma->vm_file;
272 struct shm_file_data *sfd = shm_file_data(file);
273 struct shmid_kernel *shp;
274 struct ipc_namespace *ns = sfd->ns;
275
276 down_write(&shm_ids(ns).rwsem);
277 /* remove from the list of attaches of the shm segment */
278 shp = shm_lock(ns, sfd->id);
279
280 /*
281 * We raced in the idr lookup or with shm_destroy().
282 * Either way, the ID is busted.
283 */
284 if (WARN_ON_ONCE(IS_ERR(shp)))
285 goto done; /* no-op */
286
287 shp->shm_lprid = task_tgid_vnr(current);
288 shp->shm_dtim = get_seconds();
289 shp->shm_nattch--;
290 if (shm_may_destroy(ns, shp))
291 shm_destroy(ns, shp);
292 else
293 shm_unlock(shp);
294done:
295 up_write(&shm_ids(ns).rwsem);
296}
297
298/* Called with ns->shm_ids(ns).rwsem locked */
299static int shm_try_destroy_orphaned(int id, void *p, void *data)
300{
301 struct ipc_namespace *ns = data;
302 struct kern_ipc_perm *ipcp = p;
303 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
304
305 /*
306 * We want to destroy segments without users and with already
307 * exit'ed originating process.
308 *
309 * As shp->* are changed under rwsem, it's safe to skip shp locking.
310 */
311 if (shp->shm_creator != NULL)
312 return 0;
313
314 if (shm_may_destroy(ns, shp)) {
315 shm_lock_by_ptr(shp);
316 shm_destroy(ns, shp);
317 }
318 return 0;
319}
320
321void shm_destroy_orphaned(struct ipc_namespace *ns)
322{
323 down_write(&shm_ids(ns).rwsem);
324 if (shm_ids(ns).in_use)
325 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
326 up_write(&shm_ids(ns).rwsem);
327}
328
329/* Locking assumes this will only be called with task == current */
330void exit_shm(struct task_struct *task)
331{
332 struct ipc_namespace *ns = task->nsproxy->ipc_ns;
333 struct shmid_kernel *shp, *n;
334
335 if (list_empty(&task->sysvshm.shm_clist))
336 return;
337
338 /*
339 * If kernel.shm_rmid_forced is not set then only keep track of
340 * which shmids are orphaned, so that a later set of the sysctl
341 * can clean them up.
342 */
343 if (!ns->shm_rmid_forced) {
344 down_read(&shm_ids(ns).rwsem);
345 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
346 shp->shm_creator = NULL;
347 /*
348 * Only under read lock but we are only called on current
349 * so no entry on the list will be shared.
350 */
351 list_del(&task->sysvshm.shm_clist);
352 up_read(&shm_ids(ns).rwsem);
353 return;
354 }
355
356 /*
357 * Destroy all already created segments, that were not yet mapped,
358 * and mark any mapped as orphan to cover the sysctl toggling.
359 * Destroy is skipped if shm_may_destroy() returns false.
360 */
361 down_write(&shm_ids(ns).rwsem);
362 list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
363 shp->shm_creator = NULL;
364
365 if (shm_may_destroy(ns, shp)) {
366 shm_lock_by_ptr(shp);
367 shm_destroy(ns, shp);
368 }
369 }
370
371 /* Remove the list head from any segments still attached. */
372 list_del(&task->sysvshm.shm_clist);
373 up_write(&shm_ids(ns).rwsem);
374}
375
376static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
377{
378 struct file *file = vma->vm_file;
379 struct shm_file_data *sfd = shm_file_data(file);
380
381 return sfd->vm_ops->fault(vma, vmf);
382}
383
384#ifdef CONFIG_NUMA
385static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
386{
387 struct file *file = vma->vm_file;
388 struct shm_file_data *sfd = shm_file_data(file);
389 int err = 0;
390 if (sfd->vm_ops->set_policy)
391 err = sfd->vm_ops->set_policy(vma, new);
392 return err;
393}
394
395static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
396 unsigned long addr)
397{
398 struct file *file = vma->vm_file;
399 struct shm_file_data *sfd = shm_file_data(file);
400 struct mempolicy *pol = NULL;
401
402 if (sfd->vm_ops->get_policy)
403 pol = sfd->vm_ops->get_policy(vma, addr);
404 else if (vma->vm_policy)
405 pol = vma->vm_policy;
406
407 return pol;
408}
409#endif
410
411static int shm_mmap(struct file *file, struct vm_area_struct *vma)
412{
413 struct shm_file_data *sfd = shm_file_data(file);
414 int ret;
415
416 /*
417 * In case of remap_file_pages() emulation, the file can represent
418 * removed IPC ID: propogate shm_lock() error to caller.
419 */
420 ret =__shm_open(vma);
421 if (ret)
422 return ret;
423
424 ret = sfd->file->f_op->mmap(sfd->file, vma);
425 if (ret) {
426 shm_close(vma);
427 return ret;
428 }
429 sfd->vm_ops = vma->vm_ops;
430#ifdef CONFIG_MMU
431 WARN_ON(!sfd->vm_ops->fault);
432#endif
433 vma->vm_ops = &shm_vm_ops;
434 return 0;
435}
436
437static int shm_release(struct inode *ino, struct file *file)
438{
439 struct shm_file_data *sfd = shm_file_data(file);
440
441 put_ipc_ns(sfd->ns);
442 shm_file_data(file) = NULL;
443 kfree(sfd);
444 return 0;
445}
446
447static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
448{
449 struct shm_file_data *sfd = shm_file_data(file);
450
451 if (!sfd->file->f_op->fsync)
452 return -EINVAL;
453 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
454}
455
456static long shm_fallocate(struct file *file, int mode, loff_t offset,
457 loff_t len)
458{
459 struct shm_file_data *sfd = shm_file_data(file);
460
461 if (!sfd->file->f_op->fallocate)
462 return -EOPNOTSUPP;
463 return sfd->file->f_op->fallocate(file, mode, offset, len);
464}
465
466static unsigned long shm_get_unmapped_area(struct file *file,
467 unsigned long addr, unsigned long len, unsigned long pgoff,
468 unsigned long flags)
469{
470 struct shm_file_data *sfd = shm_file_data(file);
471 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
472 pgoff, flags);
473}
474
475static const struct file_operations shm_file_operations = {
476 .mmap = shm_mmap,
477 .fsync = shm_fsync,
478 .release = shm_release,
479#ifndef CONFIG_MMU
480 .get_unmapped_area = shm_get_unmapped_area,
481#endif
482 .llseek = noop_llseek,
483 .fallocate = shm_fallocate,
484};
485
486static const struct file_operations shm_file_operations_huge = {
487 .mmap = shm_mmap,
488 .fsync = shm_fsync,
489 .release = shm_release,
490 .get_unmapped_area = shm_get_unmapped_area,
491 .llseek = noop_llseek,
492 .fallocate = shm_fallocate,
493};
494
495bool is_file_shm_hugepages(struct file *file)
496{
497 return file->f_op == &shm_file_operations_huge;
498}
499
500static const struct vm_operations_struct shm_vm_ops = {
501 .open = shm_open, /* callback for a new vm-area open */
502 .close = shm_close, /* callback for when the vm-area is released */
503 .fault = shm_fault,
504#if defined(CONFIG_NUMA)
505 .set_policy = shm_set_policy,
506 .get_policy = shm_get_policy,
507#endif
508};
509
510/**
511 * newseg - Create a new shared memory segment
512 * @ns: namespace
513 * @params: ptr to the structure that contains key, size and shmflg
514 *
515 * Called with shm_ids.rwsem held as a writer.
516 */
517static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
518{
519 key_t key = params->key;
520 int shmflg = params->flg;
521 size_t size = params->u.size;
522 int error;
523 struct shmid_kernel *shp;
524 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
525 struct file *file;
526 char name[13];
527 int id;
528 vm_flags_t acctflag = 0;
529
530 if (size < SHMMIN || size > ns->shm_ctlmax)
531 return -EINVAL;
532
533 if (numpages << PAGE_SHIFT < size)
534 return -ENOSPC;
535
536 if (ns->shm_tot + numpages < ns->shm_tot ||
537 ns->shm_tot + numpages > ns->shm_ctlall)
538 return -ENOSPC;
539
540 shp = ipc_rcu_alloc(sizeof(*shp));
541 if (!shp)
542 return -ENOMEM;
543
544 shp->shm_perm.key = key;
545 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
546 shp->mlock_user = NULL;
547
548 shp->shm_perm.security = NULL;
549 error = security_shm_alloc(shp);
550 if (error) {
551 ipc_rcu_putref(shp, ipc_rcu_free);
552 return error;
553 }
554
555 sprintf(name, "SYSV%08x", key);
556 if (shmflg & SHM_HUGETLB) {
557 struct hstate *hs;
558 size_t hugesize;
559
560 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
561 if (!hs) {
562 error = -EINVAL;
563 goto no_file;
564 }
565 hugesize = ALIGN(size, huge_page_size(hs));
566
567 /* hugetlb_file_setup applies strict accounting */
568 if (shmflg & SHM_NORESERVE)
569 acctflag = VM_NORESERVE;
570 file = hugetlb_file_setup(name, hugesize, acctflag,
571 &shp->mlock_user, HUGETLB_SHMFS_INODE,
572 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
573 } else {
574 /*
575 * Do not allow no accounting for OVERCOMMIT_NEVER, even
576 * if it's asked for.
577 */
578 if ((shmflg & SHM_NORESERVE) &&
579 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
580 acctflag = VM_NORESERVE;
581 file = shmem_kernel_file_setup(name, size, acctflag);
582 }
583 error = PTR_ERR(file);
584 if (IS_ERR(file))
585 goto no_file;
586
587 shp->shm_cprid = task_tgid_vnr(current);
588 shp->shm_lprid = 0;
589 shp->shm_atim = shp->shm_dtim = 0;
590 shp->shm_ctim = get_seconds();
591 shp->shm_segsz = size;
592 shp->shm_nattch = 0;
593 shp->shm_file = file;
594 shp->shm_creator = current;
595
596 id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
597 if (id < 0) {
598 error = id;
599 goto no_id;
600 }
601
602 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
603
604 /*
605 * shmid gets reported as "inode#" in /proc/pid/maps.
606 * proc-ps tools use this. Changing this will break them.
607 */
608 file_inode(file)->i_ino = shp->shm_perm.id;
609
610 ns->shm_tot += numpages;
611 error = shp->shm_perm.id;
612
613 ipc_unlock_object(&shp->shm_perm);
614 rcu_read_unlock();
615 return error;
616
617no_id:
618 if (is_file_hugepages(file) && shp->mlock_user)
619 user_shm_unlock(size, shp->mlock_user);
620 fput(file);
621no_file:
622 ipc_rcu_putref(shp, shm_rcu_free);
623 return error;
624}
625
626/*
627 * Called with shm_ids.rwsem and ipcp locked.
628 */
629static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
630{
631 struct shmid_kernel *shp;
632
633 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
634 return security_shm_associate(shp, shmflg);
635}
636
637/*
638 * Called with shm_ids.rwsem and ipcp locked.
639 */
640static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
641 struct ipc_params *params)
642{
643 struct shmid_kernel *shp;
644
645 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
646 if (shp->shm_segsz < params->u.size)
647 return -EINVAL;
648
649 return 0;
650}
651
652SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
653{
654 struct ipc_namespace *ns;
655 static const struct ipc_ops shm_ops = {
656 .getnew = newseg,
657 .associate = shm_security,
658 .more_checks = shm_more_checks,
659 };
660 struct ipc_params shm_params;
661
662 ns = current->nsproxy->ipc_ns;
663
664 shm_params.key = key;
665 shm_params.flg = shmflg;
666 shm_params.u.size = size;
667
668 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
669}
670
671static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
672{
673 switch (version) {
674 case IPC_64:
675 return copy_to_user(buf, in, sizeof(*in));
676 case IPC_OLD:
677 {
678 struct shmid_ds out;
679
680 memset(&out, 0, sizeof(out));
681 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
682 out.shm_segsz = in->shm_segsz;
683 out.shm_atime = in->shm_atime;
684 out.shm_dtime = in->shm_dtime;
685 out.shm_ctime = in->shm_ctime;
686 out.shm_cpid = in->shm_cpid;
687 out.shm_lpid = in->shm_lpid;
688 out.shm_nattch = in->shm_nattch;
689
690 return copy_to_user(buf, &out, sizeof(out));
691 }
692 default:
693 return -EINVAL;
694 }
695}
696
697static inline unsigned long
698copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
699{
700 switch (version) {
701 case IPC_64:
702 if (copy_from_user(out, buf, sizeof(*out)))
703 return -EFAULT;
704 return 0;
705 case IPC_OLD:
706 {
707 struct shmid_ds tbuf_old;
708
709 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
710 return -EFAULT;
711
712 out->shm_perm.uid = tbuf_old.shm_perm.uid;
713 out->shm_perm.gid = tbuf_old.shm_perm.gid;
714 out->shm_perm.mode = tbuf_old.shm_perm.mode;
715
716 return 0;
717 }
718 default:
719 return -EINVAL;
720 }
721}
722
723static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
724{
725 switch (version) {
726 case IPC_64:
727 return copy_to_user(buf, in, sizeof(*in));
728 case IPC_OLD:
729 {
730 struct shminfo out;
731
732 if (in->shmmax > INT_MAX)
733 out.shmmax = INT_MAX;
734 else
735 out.shmmax = (int)in->shmmax;
736
737 out.shmmin = in->shmmin;
738 out.shmmni = in->shmmni;
739 out.shmseg = in->shmseg;
740 out.shmall = in->shmall;
741
742 return copy_to_user(buf, &out, sizeof(out));
743 }
744 default:
745 return -EINVAL;
746 }
747}
748
749/*
750 * Calculate and add used RSS and swap pages of a shm.
751 * Called with shm_ids.rwsem held as a reader
752 */
753static void shm_add_rss_swap(struct shmid_kernel *shp,
754 unsigned long *rss_add, unsigned long *swp_add)
755{
756 struct inode *inode;
757
758 inode = file_inode(shp->shm_file);
759
760 if (is_file_hugepages(shp->shm_file)) {
761 struct address_space *mapping = inode->i_mapping;
762 struct hstate *h = hstate_file(shp->shm_file);
763 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
764 } else {
765#ifdef CONFIG_SHMEM
766 struct shmem_inode_info *info = SHMEM_I(inode);
767 spin_lock(&info->lock);
768 *rss_add += inode->i_mapping->nrpages;
769 *swp_add += info->swapped;
770 spin_unlock(&info->lock);
771#else
772 *rss_add += inode->i_mapping->nrpages;
773#endif
774 }
775}
776
777/*
778 * Called with shm_ids.rwsem held as a reader
779 */
780static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
781 unsigned long *swp)
782{
783 int next_id;
784 int total, in_use;
785
786 *rss = 0;
787 *swp = 0;
788
789 in_use = shm_ids(ns).in_use;
790
791 for (total = 0, next_id = 0; total < in_use; next_id++) {
792 struct kern_ipc_perm *ipc;
793 struct shmid_kernel *shp;
794
795 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
796 if (ipc == NULL)
797 continue;
798 shp = container_of(ipc, struct shmid_kernel, shm_perm);
799
800 shm_add_rss_swap(shp, rss, swp);
801
802 total++;
803 }
804}
805
806/*
807 * This function handles some shmctl commands which require the rwsem
808 * to be held in write mode.
809 * NOTE: no locks must be held, the rwsem is taken inside this function.
810 */
811static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
812 struct shmid_ds __user *buf, int version)
813{
814 struct kern_ipc_perm *ipcp;
815 struct shmid64_ds shmid64;
816 struct shmid_kernel *shp;
817 int err;
818
819 if (cmd == IPC_SET) {
820 if (copy_shmid_from_user(&shmid64, buf, version))
821 return -EFAULT;
822 }
823
824 down_write(&shm_ids(ns).rwsem);
825 rcu_read_lock();
826
827 ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
828 &shmid64.shm_perm, 0);
829 if (IS_ERR(ipcp)) {
830 err = PTR_ERR(ipcp);
831 goto out_unlock1;
832 }
833
834 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
835
836 err = security_shm_shmctl(shp, cmd);
837 if (err)
838 goto out_unlock1;
839
840 switch (cmd) {
841 case IPC_RMID:
842 ipc_lock_object(&shp->shm_perm);
843 /* do_shm_rmid unlocks the ipc object and rcu */
844 do_shm_rmid(ns, ipcp);
845 goto out_up;
846 case IPC_SET:
847 ipc_lock_object(&shp->shm_perm);
848 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
849 if (err)
850 goto out_unlock0;
851 shp->shm_ctim = get_seconds();
852 break;
853 default:
854 err = -EINVAL;
855 goto out_unlock1;
856 }
857
858out_unlock0:
859 ipc_unlock_object(&shp->shm_perm);
860out_unlock1:
861 rcu_read_unlock();
862out_up:
863 up_write(&shm_ids(ns).rwsem);
864 return err;
865}
866
867static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
868 int cmd, int version, void __user *buf)
869{
870 int err;
871 struct shmid_kernel *shp;
872
873 /* preliminary security checks for *_INFO */
874 if (cmd == IPC_INFO || cmd == SHM_INFO) {
875 err = security_shm_shmctl(NULL, cmd);
876 if (err)
877 return err;
878 }
879
880 switch (cmd) {
881 case IPC_INFO:
882 {
883 struct shminfo64 shminfo;
884
885 memset(&shminfo, 0, sizeof(shminfo));
886 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
887 shminfo.shmmax = ns->shm_ctlmax;
888 shminfo.shmall = ns->shm_ctlall;
889
890 shminfo.shmmin = SHMMIN;
891 if (copy_shminfo_to_user(buf, &shminfo, version))
892 return -EFAULT;
893
894 down_read(&shm_ids(ns).rwsem);
895 err = ipc_get_maxid(&shm_ids(ns));
896 up_read(&shm_ids(ns).rwsem);
897
898 if (err < 0)
899 err = 0;
900 goto out;
901 }
902 case SHM_INFO:
903 {
904 struct shm_info shm_info;
905
906 memset(&shm_info, 0, sizeof(shm_info));
907 down_read(&shm_ids(ns).rwsem);
908 shm_info.used_ids = shm_ids(ns).in_use;
909 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
910 shm_info.shm_tot = ns->shm_tot;
911 shm_info.swap_attempts = 0;
912 shm_info.swap_successes = 0;
913 err = ipc_get_maxid(&shm_ids(ns));
914 up_read(&shm_ids(ns).rwsem);
915 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
916 err = -EFAULT;
917 goto out;
918 }
919
920 err = err < 0 ? 0 : err;
921 goto out;
922 }
923 case SHM_STAT:
924 case IPC_STAT:
925 {
926 struct shmid64_ds tbuf;
927 int result;
928
929 rcu_read_lock();
930 if (cmd == SHM_STAT) {
931 shp = shm_obtain_object(ns, shmid);
932 if (IS_ERR(shp)) {
933 err = PTR_ERR(shp);
934 goto out_unlock;
935 }
936 result = shp->shm_perm.id;
937 } else {
938 shp = shm_obtain_object_check(ns, shmid);
939 if (IS_ERR(shp)) {
940 err = PTR_ERR(shp);
941 goto out_unlock;
942 }
943 result = 0;
944 }
945
946 err = -EACCES;
947 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
948 goto out_unlock;
949
950 err = security_shm_shmctl(shp, cmd);
951 if (err)
952 goto out_unlock;
953
954 memset(&tbuf, 0, sizeof(tbuf));
955 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
956 tbuf.shm_segsz = shp->shm_segsz;
957 tbuf.shm_atime = shp->shm_atim;
958 tbuf.shm_dtime = shp->shm_dtim;
959 tbuf.shm_ctime = shp->shm_ctim;
960 tbuf.shm_cpid = shp->shm_cprid;
961 tbuf.shm_lpid = shp->shm_lprid;
962 tbuf.shm_nattch = shp->shm_nattch;
963 rcu_read_unlock();
964
965 if (copy_shmid_to_user(buf, &tbuf, version))
966 err = -EFAULT;
967 else
968 err = result;
969 goto out;
970 }
971 default:
972 return -EINVAL;
973 }
974
975out_unlock:
976 rcu_read_unlock();
977out:
978 return err;
979}
980
981SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
982{
983 struct shmid_kernel *shp;
984 int err, version;
985 struct ipc_namespace *ns;
986
987 if (cmd < 0 || shmid < 0)
988 return -EINVAL;
989
990 version = ipc_parse_version(&cmd);
991 ns = current->nsproxy->ipc_ns;
992
993 switch (cmd) {
994 case IPC_INFO:
995 case SHM_INFO:
996 case SHM_STAT:
997 case IPC_STAT:
998 return shmctl_nolock(ns, shmid, cmd, version, buf);
999 case IPC_RMID:
1000 case IPC_SET:
1001 return shmctl_down(ns, shmid, cmd, buf, version);
1002 case SHM_LOCK:
1003 case SHM_UNLOCK:
1004 {
1005 struct file *shm_file;
1006
1007 rcu_read_lock();
1008 shp = shm_obtain_object_check(ns, shmid);
1009 if (IS_ERR(shp)) {
1010 err = PTR_ERR(shp);
1011 goto out_unlock1;
1012 }
1013
1014 audit_ipc_obj(&(shp->shm_perm));
1015 err = security_shm_shmctl(shp, cmd);
1016 if (err)
1017 goto out_unlock1;
1018
1019 ipc_lock_object(&shp->shm_perm);
1020
1021 /* check if shm_destroy() is tearing down shp */
1022 if (!ipc_valid_object(&shp->shm_perm)) {
1023 err = -EIDRM;
1024 goto out_unlock0;
1025 }
1026
1027 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1028 kuid_t euid = current_euid();
1029 if (!uid_eq(euid, shp->shm_perm.uid) &&
1030 !uid_eq(euid, shp->shm_perm.cuid)) {
1031 err = -EPERM;
1032 goto out_unlock0;
1033 }
1034 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1035 err = -EPERM;
1036 goto out_unlock0;
1037 }
1038 }
1039
1040 shm_file = shp->shm_file;
1041 if (is_file_hugepages(shm_file))
1042 goto out_unlock0;
1043
1044 if (cmd == SHM_LOCK) {
1045 struct user_struct *user = current_user();
1046 err = shmem_lock(shm_file, 1, user);
1047 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1048 shp->shm_perm.mode |= SHM_LOCKED;
1049 shp->mlock_user = user;
1050 }
1051 goto out_unlock0;
1052 }
1053
1054 /* SHM_UNLOCK */
1055 if (!(shp->shm_perm.mode & SHM_LOCKED))
1056 goto out_unlock0;
1057 shmem_lock(shm_file, 0, shp->mlock_user);
1058 shp->shm_perm.mode &= ~SHM_LOCKED;
1059 shp->mlock_user = NULL;
1060 get_file(shm_file);
1061 ipc_unlock_object(&shp->shm_perm);
1062 rcu_read_unlock();
1063 shmem_unlock_mapping(shm_file->f_mapping);
1064
1065 fput(shm_file);
1066 return err;
1067 }
1068 default:
1069 return -EINVAL;
1070 }
1071
1072out_unlock0:
1073 ipc_unlock_object(&shp->shm_perm);
1074out_unlock1:
1075 rcu_read_unlock();
1076 return err;
1077}
1078
1079/*
1080 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1081 *
1082 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1083 * "raddr" thing points to kernel space, and there has to be a wrapper around
1084 * this.
1085 */
1086long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1087 unsigned long shmlba)
1088{
1089 struct shmid_kernel *shp;
1090 unsigned long addr;
1091 unsigned long size;
1092 struct file *file;
1093 int err;
1094 unsigned long flags;
1095 unsigned long prot;
1096 int acc_mode;
1097 struct ipc_namespace *ns;
1098 struct shm_file_data *sfd;
1099 struct path path;
1100 fmode_t f_mode;
1101 unsigned long populate = 0;
1102
1103 err = -EINVAL;
1104 if (shmid < 0)
1105 goto out;
1106 else if ((addr = (ulong)shmaddr)) {
1107 if (addr & (shmlba - 1)) {
1108 if (shmflg & SHM_RND)
1109 addr &= ~(shmlba - 1); /* round down */
1110 else
1111#ifndef __ARCH_FORCE_SHMLBA
1112 if (addr & ~PAGE_MASK)
1113#endif
1114 goto out;
1115 }
1116 flags = MAP_SHARED | MAP_FIXED;
1117 } else {
1118 if ((shmflg & SHM_REMAP))
1119 goto out;
1120
1121 flags = MAP_SHARED;
1122 }
1123
1124 if (shmflg & SHM_RDONLY) {
1125 prot = PROT_READ;
1126 acc_mode = S_IRUGO;
1127 f_mode = FMODE_READ;
1128 } else {
1129 prot = PROT_READ | PROT_WRITE;
1130 acc_mode = S_IRUGO | S_IWUGO;
1131 f_mode = FMODE_READ | FMODE_WRITE;
1132 }
1133 if (shmflg & SHM_EXEC) {
1134 prot |= PROT_EXEC;
1135 acc_mode |= S_IXUGO;
1136 }
1137
1138 /*
1139 * We cannot rely on the fs check since SYSV IPC does have an
1140 * additional creator id...
1141 */
1142 ns = current->nsproxy->ipc_ns;
1143 rcu_read_lock();
1144 shp = shm_obtain_object_check(ns, shmid);
1145 if (IS_ERR(shp)) {
1146 err = PTR_ERR(shp);
1147 goto out_unlock;
1148 }
1149
1150 err = -EACCES;
1151 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1152 goto out_unlock;
1153
1154 err = security_shm_shmat(shp, shmaddr, shmflg);
1155 if (err)
1156 goto out_unlock;
1157
1158 ipc_lock_object(&shp->shm_perm);
1159
1160 /* check if shm_destroy() is tearing down shp */
1161 if (!ipc_valid_object(&shp->shm_perm)) {
1162 ipc_unlock_object(&shp->shm_perm);
1163 err = -EIDRM;
1164 goto out_unlock;
1165 }
1166
1167 path = shp->shm_file->f_path;
1168 path_get(&path);
1169 shp->shm_nattch++;
1170 size = i_size_read(d_inode(path.dentry));
1171 ipc_unlock_object(&shp->shm_perm);
1172 rcu_read_unlock();
1173
1174 err = -ENOMEM;
1175 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1176 if (!sfd) {
1177 path_put(&path);
1178 goto out_nattch;
1179 }
1180
1181 file = alloc_file(&path, f_mode,
1182 is_file_hugepages(shp->shm_file) ?
1183 &shm_file_operations_huge :
1184 &shm_file_operations);
1185 err = PTR_ERR(file);
1186 if (IS_ERR(file)) {
1187 kfree(sfd);
1188 path_put(&path);
1189 goto out_nattch;
1190 }
1191
1192 file->private_data = sfd;
1193 file->f_mapping = shp->shm_file->f_mapping;
1194 sfd->id = shp->shm_perm.id;
1195 sfd->ns = get_ipc_ns(ns);
1196 sfd->file = shp->shm_file;
1197 sfd->vm_ops = NULL;
1198
1199 err = security_mmap_file(file, prot, flags);
1200 if (err)
1201 goto out_fput;
1202
1203 down_write(¤t->mm->mmap_sem);
1204 if (addr && !(shmflg & SHM_REMAP)) {
1205 err = -EINVAL;
1206 if (addr + size < addr)
1207 goto invalid;
1208
1209 if (find_vma_intersection(current->mm, addr, addr + size))
1210 goto invalid;
1211 }
1212
1213 addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1214 *raddr = addr;
1215 err = 0;
1216 if (IS_ERR_VALUE(addr))
1217 err = (long)addr;
1218invalid:
1219 up_write(¤t->mm->mmap_sem);
1220 if (populate)
1221 mm_populate(addr, populate);
1222
1223out_fput:
1224 fput(file);
1225
1226out_nattch:
1227 down_write(&shm_ids(ns).rwsem);
1228 shp = shm_lock(ns, shmid);
1229 shp->shm_nattch--;
1230 if (shm_may_destroy(ns, shp))
1231 shm_destroy(ns, shp);
1232 else
1233 shm_unlock(shp);
1234 up_write(&shm_ids(ns).rwsem);
1235 return err;
1236
1237out_unlock:
1238 rcu_read_unlock();
1239out:
1240 return err;
1241}
1242
1243SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1244{
1245 unsigned long ret;
1246 long err;
1247
1248 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1249 if (err)
1250 return err;
1251 force_successful_syscall_return();
1252 return (long)ret;
1253}
1254
1255/*
1256 * detach and kill segment if marked destroyed.
1257 * The work is done in shm_close.
1258 */
1259SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1260{
1261 struct mm_struct *mm = current->mm;
1262 struct vm_area_struct *vma;
1263 unsigned long addr = (unsigned long)shmaddr;
1264 int retval = -EINVAL;
1265#ifdef CONFIG_MMU
1266 loff_t size = 0;
1267 struct file *file;
1268 struct vm_area_struct *next;
1269#endif
1270
1271 if (addr & ~PAGE_MASK)
1272 return retval;
1273
1274 down_write(&mm->mmap_sem);
1275
1276 /*
1277 * This function tries to be smart and unmap shm segments that
1278 * were modified by partial mlock or munmap calls:
1279 * - It first determines the size of the shm segment that should be
1280 * unmapped: It searches for a vma that is backed by shm and that
1281 * started at address shmaddr. It records it's size and then unmaps
1282 * it.
1283 * - Then it unmaps all shm vmas that started at shmaddr and that
1284 * are within the initially determined size and that are from the
1285 * same shm segment from which we determined the size.
1286 * Errors from do_munmap are ignored: the function only fails if
1287 * it's called with invalid parameters or if it's called to unmap
1288 * a part of a vma. Both calls in this function are for full vmas,
1289 * the parameters are directly copied from the vma itself and always
1290 * valid - therefore do_munmap cannot fail. (famous last words?)
1291 */
1292 /*
1293 * If it had been mremap()'d, the starting address would not
1294 * match the usual checks anyway. So assume all vma's are
1295 * above the starting address given.
1296 */
1297 vma = find_vma(mm, addr);
1298
1299#ifdef CONFIG_MMU
1300 while (vma) {
1301 next = vma->vm_next;
1302
1303 /*
1304 * Check if the starting address would match, i.e. it's
1305 * a fragment created by mprotect() and/or munmap(), or it
1306 * otherwise it starts at this address with no hassles.
1307 */
1308 if ((vma->vm_ops == &shm_vm_ops) &&
1309 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1310
1311 /*
1312 * Record the file of the shm segment being
1313 * unmapped. With mremap(), someone could place
1314 * page from another segment but with equal offsets
1315 * in the range we are unmapping.
1316 */
1317 file = vma->vm_file;
1318 size = i_size_read(file_inode(vma->vm_file));
1319 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1320 /*
1321 * We discovered the size of the shm segment, so
1322 * break out of here and fall through to the next
1323 * loop that uses the size information to stop
1324 * searching for matching vma's.
1325 */
1326 retval = 0;
1327 vma = next;
1328 break;
1329 }
1330 vma = next;
1331 }
1332
1333 /*
1334 * We need look no further than the maximum address a fragment
1335 * could possibly have landed at. Also cast things to loff_t to
1336 * prevent overflows and make comparisons vs. equal-width types.
1337 */
1338 size = PAGE_ALIGN(size);
1339 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1340 next = vma->vm_next;
1341
1342 /* finding a matching vma now does not alter retval */
1343 if ((vma->vm_ops == &shm_vm_ops) &&
1344 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1345 (vma->vm_file == file))
1346 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1347 vma = next;
1348 }
1349
1350#else /* CONFIG_MMU */
1351 /* under NOMMU conditions, the exact address to be destroyed must be
1352 * given */
1353 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1354 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1355 retval = 0;
1356 }
1357
1358#endif
1359
1360 up_write(&mm->mmap_sem);
1361 return retval;
1362}
1363
1364#ifdef CONFIG_PROC_FS
1365static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1366{
1367 struct user_namespace *user_ns = seq_user_ns(s);
1368 struct shmid_kernel *shp = it;
1369 unsigned long rss = 0, swp = 0;
1370
1371 shm_add_rss_swap(shp, &rss, &swp);
1372
1373#if BITS_PER_LONG <= 32
1374#define SIZE_SPEC "%10lu"
1375#else
1376#define SIZE_SPEC "%21lu"
1377#endif
1378
1379 seq_printf(s,
1380 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1381 "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1382 SIZE_SPEC " " SIZE_SPEC "\n",
1383 shp->shm_perm.key,
1384 shp->shm_perm.id,
1385 shp->shm_perm.mode,
1386 shp->shm_segsz,
1387 shp->shm_cprid,
1388 shp->shm_lprid,
1389 shp->shm_nattch,
1390 from_kuid_munged(user_ns, shp->shm_perm.uid),
1391 from_kgid_munged(user_ns, shp->shm_perm.gid),
1392 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1393 from_kgid_munged(user_ns, shp->shm_perm.cgid),
1394 shp->shm_atim,
1395 shp->shm_dtim,
1396 shp->shm_ctim,
1397 rss * PAGE_SIZE,
1398 swp * PAGE_SIZE);
1399
1400 return 0;
1401}
1402#endif
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 <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
49#include <linux/uaccess.h>
50
51#include "util.h"
52
53struct shmid_kernel /* private to the kernel */
54{
55 struct kern_ipc_perm shm_perm;
56 struct file *shm_file;
57 unsigned long shm_nattch;
58 unsigned long shm_segsz;
59 time64_t shm_atim;
60 time64_t shm_dtim;
61 time64_t shm_ctim;
62 struct pid *shm_cprid;
63 struct pid *shm_lprid;
64 struct ucounts *mlock_ucounts;
65
66 /*
67 * The task created the shm object, for
68 * task_lock(shp->shm_creator)
69 */
70 struct task_struct *shm_creator;
71
72 /*
73 * List by creator. task_lock(->shm_creator) required for read/write.
74 * If list_empty(), then the creator is dead already.
75 */
76 struct list_head shm_clist;
77 struct ipc_namespace *ns;
78} __randomize_layout;
79
80/* shm_mode upper byte flags */
81#define SHM_DEST 01000 /* segment will be destroyed on last detach */
82#define SHM_LOCKED 02000 /* segment will not be swapped */
83
84struct shm_file_data {
85 int id;
86 struct ipc_namespace *ns;
87 struct file *file;
88 const struct vm_operations_struct *vm_ops;
89};
90
91#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
92
93static const struct file_operations shm_file_operations;
94static const struct vm_operations_struct shm_vm_ops;
95
96#define shm_ids(ns) ((ns)->ids[IPC_SHM_IDS])
97
98#define shm_unlock(shp) \
99 ipc_unlock(&(shp)->shm_perm)
100
101static int newseg(struct ipc_namespace *, struct ipc_params *);
102static void shm_open(struct vm_area_struct *vma);
103static void shm_close(struct vm_area_struct *vma);
104static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
105#ifdef CONFIG_PROC_FS
106static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
107#endif
108
109void shm_init_ns(struct ipc_namespace *ns)
110{
111 ns->shm_ctlmax = SHMMAX;
112 ns->shm_ctlall = SHMALL;
113 ns->shm_ctlmni = SHMMNI;
114 ns->shm_rmid_forced = 0;
115 ns->shm_tot = 0;
116 ipc_init_ids(&shm_ids(ns));
117}
118
119/*
120 * Called with shm_ids.rwsem (writer) and the shp structure locked.
121 * Only shm_ids.rwsem remains locked on exit.
122 */
123static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
124{
125 struct shmid_kernel *shp;
126
127 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
128 WARN_ON(ns != shp->ns);
129
130 if (shp->shm_nattch) {
131 shp->shm_perm.mode |= SHM_DEST;
132 /* Do not find it any more */
133 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
134 shm_unlock(shp);
135 } else
136 shm_destroy(ns, shp);
137}
138
139#ifdef CONFIG_IPC_NS
140void shm_exit_ns(struct ipc_namespace *ns)
141{
142 free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
143 idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
144 rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
145}
146#endif
147
148static int __init ipc_ns_init(void)
149{
150 shm_init_ns(&init_ipc_ns);
151 return 0;
152}
153
154pure_initcall(ipc_ns_init);
155
156void __init shm_init(void)
157{
158 ipc_init_proc_interface("sysvipc/shm",
159#if BITS_PER_LONG <= 32
160 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
161#else
162 " key shmid perms size cpid lpid nattch uid gid cuid cgid atime dtime ctime rss swap\n",
163#endif
164 IPC_SHM_IDS, sysvipc_shm_proc_show);
165}
166
167static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
168{
169 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
170
171 if (IS_ERR(ipcp))
172 return ERR_CAST(ipcp);
173
174 return container_of(ipcp, struct shmid_kernel, shm_perm);
175}
176
177static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
178{
179 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
180
181 if (IS_ERR(ipcp))
182 return ERR_CAST(ipcp);
183
184 return container_of(ipcp, struct shmid_kernel, shm_perm);
185}
186
187/*
188 * shm_lock_(check_) routines are called in the paths where the rwsem
189 * is not necessarily held.
190 */
191static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
192{
193 struct kern_ipc_perm *ipcp;
194
195 rcu_read_lock();
196 ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
197 if (IS_ERR(ipcp))
198 goto err;
199
200 ipc_lock_object(ipcp);
201 /*
202 * ipc_rmid() may have already freed the ID while ipc_lock_object()
203 * was spinning: here verify that the structure is still valid.
204 * Upon races with RMID, return -EIDRM, thus indicating that
205 * the ID points to a removed identifier.
206 */
207 if (ipc_valid_object(ipcp)) {
208 /* return a locked ipc object upon success */
209 return container_of(ipcp, struct shmid_kernel, shm_perm);
210 }
211
212 ipc_unlock_object(ipcp);
213 ipcp = ERR_PTR(-EIDRM);
214err:
215 rcu_read_unlock();
216 /*
217 * Callers of shm_lock() must validate the status of the returned ipc
218 * object pointer and error out as appropriate.
219 */
220 return ERR_CAST(ipcp);
221}
222
223static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
224{
225 rcu_read_lock();
226 ipc_lock_object(&ipcp->shm_perm);
227}
228
229static void shm_rcu_free(struct rcu_head *head)
230{
231 struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
232 rcu);
233 struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
234 shm_perm);
235 security_shm_free(&shp->shm_perm);
236 kfree(shp);
237}
238
239/*
240 * It has to be called with shp locked.
241 * It must be called before ipc_rmid()
242 */
243static inline void shm_clist_rm(struct shmid_kernel *shp)
244{
245 struct task_struct *creator;
246
247 /* ensure that shm_creator does not disappear */
248 rcu_read_lock();
249
250 /*
251 * A concurrent exit_shm may do a list_del_init() as well.
252 * Just do nothing if exit_shm already did the work
253 */
254 if (!list_empty(&shp->shm_clist)) {
255 /*
256 * shp->shm_creator is guaranteed to be valid *only*
257 * if shp->shm_clist is not empty.
258 */
259 creator = shp->shm_creator;
260
261 task_lock(creator);
262 /*
263 * list_del_init() is a nop if the entry was already removed
264 * from the list.
265 */
266 list_del_init(&shp->shm_clist);
267 task_unlock(creator);
268 }
269 rcu_read_unlock();
270}
271
272static inline void shm_rmid(struct shmid_kernel *s)
273{
274 shm_clist_rm(s);
275 ipc_rmid(&shm_ids(s->ns), &s->shm_perm);
276}
277
278
279static int __shm_open(struct shm_file_data *sfd)
280{
281 struct shmid_kernel *shp;
282
283 shp = shm_lock(sfd->ns, sfd->id);
284
285 if (IS_ERR(shp))
286 return PTR_ERR(shp);
287
288 if (shp->shm_file != sfd->file) {
289 /* ID was reused */
290 shm_unlock(shp);
291 return -EINVAL;
292 }
293
294 shp->shm_atim = ktime_get_real_seconds();
295 ipc_update_pid(&shp->shm_lprid, task_tgid(current));
296 shp->shm_nattch++;
297 shm_unlock(shp);
298 return 0;
299}
300
301/* This is called by fork, once for every shm attach. */
302static void shm_open(struct vm_area_struct *vma)
303{
304 struct file *file = vma->vm_file;
305 struct shm_file_data *sfd = shm_file_data(file);
306 int err;
307
308 /* Always call underlying open if present */
309 if (sfd->vm_ops->open)
310 sfd->vm_ops->open(vma);
311
312 err = __shm_open(sfd);
313 /*
314 * We raced in the idr lookup or with shm_destroy().
315 * Either way, the ID is busted.
316 */
317 WARN_ON_ONCE(err);
318}
319
320/*
321 * shm_destroy - free the struct shmid_kernel
322 *
323 * @ns: namespace
324 * @shp: struct to free
325 *
326 * It has to be called with shp and shm_ids.rwsem (writer) locked,
327 * but returns with shp unlocked and freed.
328 */
329static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
330{
331 struct file *shm_file;
332
333 shm_file = shp->shm_file;
334 shp->shm_file = NULL;
335 ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
336 shm_rmid(shp);
337 shm_unlock(shp);
338 if (!is_file_hugepages(shm_file))
339 shmem_lock(shm_file, 0, shp->mlock_ucounts);
340 fput(shm_file);
341 ipc_update_pid(&shp->shm_cprid, NULL);
342 ipc_update_pid(&shp->shm_lprid, NULL);
343 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
344}
345
346/*
347 * shm_may_destroy - identifies whether shm segment should be destroyed now
348 *
349 * Returns true if and only if there are no active users of the segment and
350 * one of the following is true:
351 *
352 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
353 *
354 * 2) sysctl kernel.shm_rmid_forced is set to 1.
355 */
356static bool shm_may_destroy(struct shmid_kernel *shp)
357{
358 return (shp->shm_nattch == 0) &&
359 (shp->ns->shm_rmid_forced ||
360 (shp->shm_perm.mode & SHM_DEST));
361}
362
363/*
364 * remove the attach descriptor vma.
365 * free memory for segment if it is marked destroyed.
366 * The descriptor has already been removed from the current->mm->mmap list
367 * and will later be kfree()d.
368 */
369static void __shm_close(struct shm_file_data *sfd)
370{
371 struct shmid_kernel *shp;
372 struct ipc_namespace *ns = sfd->ns;
373
374 down_write(&shm_ids(ns).rwsem);
375 /* remove from the list of attaches of the shm segment */
376 shp = shm_lock(ns, sfd->id);
377
378 /*
379 * We raced in the idr lookup or with shm_destroy().
380 * Either way, the ID is busted.
381 */
382 if (WARN_ON_ONCE(IS_ERR(shp)))
383 goto done; /* no-op */
384
385 ipc_update_pid(&shp->shm_lprid, task_tgid(current));
386 shp->shm_dtim = ktime_get_real_seconds();
387 shp->shm_nattch--;
388 if (shm_may_destroy(shp))
389 shm_destroy(ns, shp);
390 else
391 shm_unlock(shp);
392done:
393 up_write(&shm_ids(ns).rwsem);
394}
395
396static void shm_close(struct vm_area_struct *vma)
397{
398 struct file *file = vma->vm_file;
399 struct shm_file_data *sfd = shm_file_data(file);
400
401 /* Always call underlying close if present */
402 if (sfd->vm_ops->close)
403 sfd->vm_ops->close(vma);
404
405 __shm_close(sfd);
406}
407
408/* Called with ns->shm_ids(ns).rwsem locked */
409static int shm_try_destroy_orphaned(int id, void *p, void *data)
410{
411 struct ipc_namespace *ns = data;
412 struct kern_ipc_perm *ipcp = p;
413 struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
414
415 /*
416 * We want to destroy segments without users and with already
417 * exit'ed originating process.
418 *
419 * As shp->* are changed under rwsem, it's safe to skip shp locking.
420 */
421 if (!list_empty(&shp->shm_clist))
422 return 0;
423
424 if (shm_may_destroy(shp)) {
425 shm_lock_by_ptr(shp);
426 shm_destroy(ns, shp);
427 }
428 return 0;
429}
430
431void shm_destroy_orphaned(struct ipc_namespace *ns)
432{
433 down_write(&shm_ids(ns).rwsem);
434 if (shm_ids(ns).in_use)
435 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
436 up_write(&shm_ids(ns).rwsem);
437}
438
439/* Locking assumes this will only be called with task == current */
440void exit_shm(struct task_struct *task)
441{
442 for (;;) {
443 struct shmid_kernel *shp;
444 struct ipc_namespace *ns;
445
446 task_lock(task);
447
448 if (list_empty(&task->sysvshm.shm_clist)) {
449 task_unlock(task);
450 break;
451 }
452
453 shp = list_first_entry(&task->sysvshm.shm_clist, struct shmid_kernel,
454 shm_clist);
455
456 /*
457 * 1) Get pointer to the ipc namespace. It is worth to say
458 * that this pointer is guaranteed to be valid because
459 * shp lifetime is always shorter than namespace lifetime
460 * in which shp lives.
461 * We taken task_lock it means that shp won't be freed.
462 */
463 ns = shp->ns;
464
465 /*
466 * 2) If kernel.shm_rmid_forced is not set then only keep track of
467 * which shmids are orphaned, so that a later set of the sysctl
468 * can clean them up.
469 */
470 if (!ns->shm_rmid_forced)
471 goto unlink_continue;
472
473 /*
474 * 3) get a reference to the namespace.
475 * The refcount could be already 0. If it is 0, then
476 * the shm objects will be free by free_ipc_work().
477 */
478 ns = get_ipc_ns_not_zero(ns);
479 if (!ns) {
480unlink_continue:
481 list_del_init(&shp->shm_clist);
482 task_unlock(task);
483 continue;
484 }
485
486 /*
487 * 4) get a reference to shp.
488 * This cannot fail: shm_clist_rm() is called before
489 * ipc_rmid(), thus the refcount cannot be 0.
490 */
491 WARN_ON(!ipc_rcu_getref(&shp->shm_perm));
492
493 /*
494 * 5) unlink the shm segment from the list of segments
495 * created by current.
496 * This must be done last. After unlinking,
497 * only the refcounts obtained above prevent IPC_RMID
498 * from destroying the segment or the namespace.
499 */
500 list_del_init(&shp->shm_clist);
501
502 task_unlock(task);
503
504 /*
505 * 6) we have all references
506 * Thus lock & if needed destroy shp.
507 */
508 down_write(&shm_ids(ns).rwsem);
509 shm_lock_by_ptr(shp);
510 /*
511 * rcu_read_lock was implicitly taken in shm_lock_by_ptr, it's
512 * safe to call ipc_rcu_putref here
513 */
514 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
515
516 if (ipc_valid_object(&shp->shm_perm)) {
517 if (shm_may_destroy(shp))
518 shm_destroy(ns, shp);
519 else
520 shm_unlock(shp);
521 } else {
522 /*
523 * Someone else deleted the shp from namespace
524 * idr/kht while we have waited.
525 * Just unlock and continue.
526 */
527 shm_unlock(shp);
528 }
529
530 up_write(&shm_ids(ns).rwsem);
531 put_ipc_ns(ns); /* paired with get_ipc_ns_not_zero */
532 }
533}
534
535static vm_fault_t shm_fault(struct vm_fault *vmf)
536{
537 struct file *file = vmf->vma->vm_file;
538 struct shm_file_data *sfd = shm_file_data(file);
539
540 return sfd->vm_ops->fault(vmf);
541}
542
543static int shm_may_split(struct vm_area_struct *vma, unsigned long addr)
544{
545 struct file *file = vma->vm_file;
546 struct shm_file_data *sfd = shm_file_data(file);
547
548 if (sfd->vm_ops->may_split)
549 return sfd->vm_ops->may_split(vma, addr);
550
551 return 0;
552}
553
554static unsigned long shm_pagesize(struct vm_area_struct *vma)
555{
556 struct file *file = vma->vm_file;
557 struct shm_file_data *sfd = shm_file_data(file);
558
559 if (sfd->vm_ops->pagesize)
560 return sfd->vm_ops->pagesize(vma);
561
562 return PAGE_SIZE;
563}
564
565#ifdef CONFIG_NUMA
566static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
567{
568 struct shm_file_data *sfd = shm_file_data(vma->vm_file);
569 int err = 0;
570
571 if (sfd->vm_ops->set_policy)
572 err = sfd->vm_ops->set_policy(vma, mpol);
573 return err;
574}
575
576static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
577 unsigned long addr, pgoff_t *ilx)
578{
579 struct shm_file_data *sfd = shm_file_data(vma->vm_file);
580 struct mempolicy *mpol = vma->vm_policy;
581
582 if (sfd->vm_ops->get_policy)
583 mpol = sfd->vm_ops->get_policy(vma, addr, ilx);
584 return mpol;
585}
586#endif
587
588static int shm_mmap(struct file *file, struct vm_area_struct *vma)
589{
590 struct shm_file_data *sfd = shm_file_data(file);
591 int ret;
592
593 /*
594 * In case of remap_file_pages() emulation, the file can represent an
595 * IPC ID that was removed, and possibly even reused by another shm
596 * segment already. Propagate this case as an error to caller.
597 */
598 ret = __shm_open(sfd);
599 if (ret)
600 return ret;
601
602 ret = call_mmap(sfd->file, vma);
603 if (ret) {
604 __shm_close(sfd);
605 return ret;
606 }
607 sfd->vm_ops = vma->vm_ops;
608#ifdef CONFIG_MMU
609 WARN_ON(!sfd->vm_ops->fault);
610#endif
611 vma->vm_ops = &shm_vm_ops;
612 return 0;
613}
614
615static int shm_release(struct inode *ino, struct file *file)
616{
617 struct shm_file_data *sfd = shm_file_data(file);
618
619 put_ipc_ns(sfd->ns);
620 fput(sfd->file);
621 shm_file_data(file) = NULL;
622 kfree(sfd);
623 return 0;
624}
625
626static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
627{
628 struct shm_file_data *sfd = shm_file_data(file);
629
630 if (!sfd->file->f_op->fsync)
631 return -EINVAL;
632 return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
633}
634
635static long shm_fallocate(struct file *file, int mode, loff_t offset,
636 loff_t len)
637{
638 struct shm_file_data *sfd = shm_file_data(file);
639
640 if (!sfd->file->f_op->fallocate)
641 return -EOPNOTSUPP;
642 return sfd->file->f_op->fallocate(file, mode, offset, len);
643}
644
645static unsigned long shm_get_unmapped_area(struct file *file,
646 unsigned long addr, unsigned long len, unsigned long pgoff,
647 unsigned long flags)
648{
649 struct shm_file_data *sfd = shm_file_data(file);
650
651 return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
652 pgoff, flags);
653}
654
655static const struct file_operations shm_file_operations = {
656 .mmap = shm_mmap,
657 .fsync = shm_fsync,
658 .release = shm_release,
659 .get_unmapped_area = shm_get_unmapped_area,
660 .llseek = noop_llseek,
661 .fallocate = shm_fallocate,
662};
663
664/*
665 * shm_file_operations_huge is now identical to shm_file_operations
666 * except for fop_flags
667 */
668static const struct file_operations shm_file_operations_huge = {
669 .mmap = shm_mmap,
670 .fsync = shm_fsync,
671 .release = shm_release,
672 .get_unmapped_area = shm_get_unmapped_area,
673 .llseek = noop_llseek,
674 .fallocate = shm_fallocate,
675 .fop_flags = FOP_HUGE_PAGES,
676};
677
678static const struct vm_operations_struct shm_vm_ops = {
679 .open = shm_open, /* callback for a new vm-area open */
680 .close = shm_close, /* callback for when the vm-area is released */
681 .fault = shm_fault,
682 .may_split = shm_may_split,
683 .pagesize = shm_pagesize,
684#if defined(CONFIG_NUMA)
685 .set_policy = shm_set_policy,
686 .get_policy = shm_get_policy,
687#endif
688};
689
690/**
691 * newseg - Create a new shared memory segment
692 * @ns: namespace
693 * @params: ptr to the structure that contains key, size and shmflg
694 *
695 * Called with shm_ids.rwsem held as a writer.
696 */
697static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
698{
699 key_t key = params->key;
700 int shmflg = params->flg;
701 size_t size = params->u.size;
702 int error;
703 struct shmid_kernel *shp;
704 size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
705 struct file *file;
706 char name[13];
707 vm_flags_t acctflag = 0;
708
709 if (size < SHMMIN || size > ns->shm_ctlmax)
710 return -EINVAL;
711
712 if (numpages << PAGE_SHIFT < size)
713 return -ENOSPC;
714
715 if (ns->shm_tot + numpages < ns->shm_tot ||
716 ns->shm_tot + numpages > ns->shm_ctlall)
717 return -ENOSPC;
718
719 shp = kmalloc(sizeof(*shp), GFP_KERNEL_ACCOUNT);
720 if (unlikely(!shp))
721 return -ENOMEM;
722
723 shp->shm_perm.key = key;
724 shp->shm_perm.mode = (shmflg & S_IRWXUGO);
725 shp->mlock_ucounts = NULL;
726
727 shp->shm_perm.security = NULL;
728 error = security_shm_alloc(&shp->shm_perm);
729 if (error) {
730 kfree(shp);
731 return error;
732 }
733
734 sprintf(name, "SYSV%08x", key);
735 if (shmflg & SHM_HUGETLB) {
736 struct hstate *hs;
737 size_t hugesize;
738
739 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
740 if (!hs) {
741 error = -EINVAL;
742 goto no_file;
743 }
744 hugesize = ALIGN(size, huge_page_size(hs));
745
746 /* hugetlb_file_setup applies strict accounting */
747 if (shmflg & SHM_NORESERVE)
748 acctflag = VM_NORESERVE;
749 file = hugetlb_file_setup(name, hugesize, acctflag,
750 HUGETLB_SHMFS_INODE, (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
751 } else {
752 /*
753 * Do not allow no accounting for OVERCOMMIT_NEVER, even
754 * if it's asked for.
755 */
756 if ((shmflg & SHM_NORESERVE) &&
757 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
758 acctflag = VM_NORESERVE;
759 file = shmem_kernel_file_setup(name, size, acctflag);
760 }
761 error = PTR_ERR(file);
762 if (IS_ERR(file))
763 goto no_file;
764
765 shp->shm_cprid = get_pid(task_tgid(current));
766 shp->shm_lprid = NULL;
767 shp->shm_atim = shp->shm_dtim = 0;
768 shp->shm_ctim = ktime_get_real_seconds();
769 shp->shm_segsz = size;
770 shp->shm_nattch = 0;
771 shp->shm_file = file;
772 shp->shm_creator = current;
773
774 /* ipc_addid() locks shp upon success. */
775 error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
776 if (error < 0)
777 goto no_id;
778
779 shp->ns = ns;
780
781 task_lock(current);
782 list_add(&shp->shm_clist, ¤t->sysvshm.shm_clist);
783 task_unlock(current);
784
785 /*
786 * shmid gets reported as "inode#" in /proc/pid/maps.
787 * proc-ps tools use this. Changing this will break them.
788 */
789 file_inode(file)->i_ino = shp->shm_perm.id;
790
791 ns->shm_tot += numpages;
792 error = shp->shm_perm.id;
793
794 ipc_unlock_object(&shp->shm_perm);
795 rcu_read_unlock();
796 return error;
797
798no_id:
799 ipc_update_pid(&shp->shm_cprid, NULL);
800 ipc_update_pid(&shp->shm_lprid, NULL);
801 fput(file);
802 ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
803 return error;
804no_file:
805 call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
806 return error;
807}
808
809/*
810 * Called with shm_ids.rwsem and ipcp locked.
811 */
812static int shm_more_checks(struct kern_ipc_perm *ipcp, struct ipc_params *params)
813{
814 struct shmid_kernel *shp;
815
816 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
817 if (shp->shm_segsz < params->u.size)
818 return -EINVAL;
819
820 return 0;
821}
822
823long ksys_shmget(key_t key, size_t size, int shmflg)
824{
825 struct ipc_namespace *ns;
826 static const struct ipc_ops shm_ops = {
827 .getnew = newseg,
828 .associate = security_shm_associate,
829 .more_checks = shm_more_checks,
830 };
831 struct ipc_params shm_params;
832
833 ns = current->nsproxy->ipc_ns;
834
835 shm_params.key = key;
836 shm_params.flg = shmflg;
837 shm_params.u.size = size;
838
839 return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
840}
841
842SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
843{
844 return ksys_shmget(key, size, shmflg);
845}
846
847static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
848{
849 switch (version) {
850 case IPC_64:
851 return copy_to_user(buf, in, sizeof(*in));
852 case IPC_OLD:
853 {
854 struct shmid_ds out;
855
856 memset(&out, 0, sizeof(out));
857 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
858 out.shm_segsz = in->shm_segsz;
859 out.shm_atime = in->shm_atime;
860 out.shm_dtime = in->shm_dtime;
861 out.shm_ctime = in->shm_ctime;
862 out.shm_cpid = in->shm_cpid;
863 out.shm_lpid = in->shm_lpid;
864 out.shm_nattch = in->shm_nattch;
865
866 return copy_to_user(buf, &out, sizeof(out));
867 }
868 default:
869 return -EINVAL;
870 }
871}
872
873static inline unsigned long
874copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
875{
876 switch (version) {
877 case IPC_64:
878 if (copy_from_user(out, buf, sizeof(*out)))
879 return -EFAULT;
880 return 0;
881 case IPC_OLD:
882 {
883 struct shmid_ds tbuf_old;
884
885 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
886 return -EFAULT;
887
888 out->shm_perm.uid = tbuf_old.shm_perm.uid;
889 out->shm_perm.gid = tbuf_old.shm_perm.gid;
890 out->shm_perm.mode = tbuf_old.shm_perm.mode;
891
892 return 0;
893 }
894 default:
895 return -EINVAL;
896 }
897}
898
899static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
900{
901 switch (version) {
902 case IPC_64:
903 return copy_to_user(buf, in, sizeof(*in));
904 case IPC_OLD:
905 {
906 struct shminfo out;
907
908 if (in->shmmax > INT_MAX)
909 out.shmmax = INT_MAX;
910 else
911 out.shmmax = (int)in->shmmax;
912
913 out.shmmin = in->shmmin;
914 out.shmmni = in->shmmni;
915 out.shmseg = in->shmseg;
916 out.shmall = in->shmall;
917
918 return copy_to_user(buf, &out, sizeof(out));
919 }
920 default:
921 return -EINVAL;
922 }
923}
924
925/*
926 * Calculate and add used RSS and swap pages of a shm.
927 * Called with shm_ids.rwsem held as a reader
928 */
929static void shm_add_rss_swap(struct shmid_kernel *shp,
930 unsigned long *rss_add, unsigned long *swp_add)
931{
932 struct inode *inode;
933
934 inode = file_inode(shp->shm_file);
935
936 if (is_file_hugepages(shp->shm_file)) {
937 struct address_space *mapping = inode->i_mapping;
938 struct hstate *h = hstate_file(shp->shm_file);
939 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
940 } else {
941#ifdef CONFIG_SHMEM
942 struct shmem_inode_info *info = SHMEM_I(inode);
943
944 spin_lock_irq(&info->lock);
945 *rss_add += inode->i_mapping->nrpages;
946 *swp_add += info->swapped;
947 spin_unlock_irq(&info->lock);
948#else
949 *rss_add += inode->i_mapping->nrpages;
950#endif
951 }
952}
953
954/*
955 * Called with shm_ids.rwsem held as a reader
956 */
957static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
958 unsigned long *swp)
959{
960 int next_id;
961 int total, in_use;
962
963 *rss = 0;
964 *swp = 0;
965
966 in_use = shm_ids(ns).in_use;
967
968 for (total = 0, next_id = 0; total < in_use; next_id++) {
969 struct kern_ipc_perm *ipc;
970 struct shmid_kernel *shp;
971
972 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
973 if (ipc == NULL)
974 continue;
975 shp = container_of(ipc, struct shmid_kernel, shm_perm);
976
977 shm_add_rss_swap(shp, rss, swp);
978
979 total++;
980 }
981}
982
983/*
984 * This function handles some shmctl commands which require the rwsem
985 * to be held in write mode.
986 * NOTE: no locks must be held, the rwsem is taken inside this function.
987 */
988static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
989 struct shmid64_ds *shmid64)
990{
991 struct kern_ipc_perm *ipcp;
992 struct shmid_kernel *shp;
993 int err;
994
995 down_write(&shm_ids(ns).rwsem);
996 rcu_read_lock();
997
998 ipcp = ipcctl_obtain_check(ns, &shm_ids(ns), shmid, cmd,
999 &shmid64->shm_perm, 0);
1000 if (IS_ERR(ipcp)) {
1001 err = PTR_ERR(ipcp);
1002 goto out_unlock1;
1003 }
1004
1005 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1006
1007 err = security_shm_shmctl(&shp->shm_perm, cmd);
1008 if (err)
1009 goto out_unlock1;
1010
1011 switch (cmd) {
1012 case IPC_RMID:
1013 ipc_lock_object(&shp->shm_perm);
1014 /* do_shm_rmid unlocks the ipc object and rcu */
1015 do_shm_rmid(ns, ipcp);
1016 goto out_up;
1017 case IPC_SET:
1018 ipc_lock_object(&shp->shm_perm);
1019 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
1020 if (err)
1021 goto out_unlock0;
1022 shp->shm_ctim = ktime_get_real_seconds();
1023 break;
1024 default:
1025 err = -EINVAL;
1026 goto out_unlock1;
1027 }
1028
1029out_unlock0:
1030 ipc_unlock_object(&shp->shm_perm);
1031out_unlock1:
1032 rcu_read_unlock();
1033out_up:
1034 up_write(&shm_ids(ns).rwsem);
1035 return err;
1036}
1037
1038static int shmctl_ipc_info(struct ipc_namespace *ns,
1039 struct shminfo64 *shminfo)
1040{
1041 int err = security_shm_shmctl(NULL, IPC_INFO);
1042 if (!err) {
1043 memset(shminfo, 0, sizeof(*shminfo));
1044 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
1045 shminfo->shmmax = ns->shm_ctlmax;
1046 shminfo->shmall = ns->shm_ctlall;
1047 shminfo->shmmin = SHMMIN;
1048 down_read(&shm_ids(ns).rwsem);
1049 err = ipc_get_maxidx(&shm_ids(ns));
1050 up_read(&shm_ids(ns).rwsem);
1051 if (err < 0)
1052 err = 0;
1053 }
1054 return err;
1055}
1056
1057static int shmctl_shm_info(struct ipc_namespace *ns,
1058 struct shm_info *shm_info)
1059{
1060 int err = security_shm_shmctl(NULL, SHM_INFO);
1061 if (!err) {
1062 memset(shm_info, 0, sizeof(*shm_info));
1063 down_read(&shm_ids(ns).rwsem);
1064 shm_info->used_ids = shm_ids(ns).in_use;
1065 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
1066 shm_info->shm_tot = ns->shm_tot;
1067 shm_info->swap_attempts = 0;
1068 shm_info->swap_successes = 0;
1069 err = ipc_get_maxidx(&shm_ids(ns));
1070 up_read(&shm_ids(ns).rwsem);
1071 if (err < 0)
1072 err = 0;
1073 }
1074 return err;
1075}
1076
1077static int shmctl_stat(struct ipc_namespace *ns, int shmid,
1078 int cmd, struct shmid64_ds *tbuf)
1079{
1080 struct shmid_kernel *shp;
1081 int err;
1082
1083 memset(tbuf, 0, sizeof(*tbuf));
1084
1085 rcu_read_lock();
1086 if (cmd == SHM_STAT || cmd == SHM_STAT_ANY) {
1087 shp = shm_obtain_object(ns, shmid);
1088 if (IS_ERR(shp)) {
1089 err = PTR_ERR(shp);
1090 goto out_unlock;
1091 }
1092 } else { /* IPC_STAT */
1093 shp = shm_obtain_object_check(ns, shmid);
1094 if (IS_ERR(shp)) {
1095 err = PTR_ERR(shp);
1096 goto out_unlock;
1097 }
1098 }
1099
1100 /*
1101 * Semantically SHM_STAT_ANY ought to be identical to
1102 * that functionality provided by the /proc/sysvipc/
1103 * interface. As such, only audit these calls and
1104 * do not do traditional S_IRUGO permission checks on
1105 * the ipc object.
1106 */
1107 if (cmd == SHM_STAT_ANY)
1108 audit_ipc_obj(&shp->shm_perm);
1109 else {
1110 err = -EACCES;
1111 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
1112 goto out_unlock;
1113 }
1114
1115 err = security_shm_shmctl(&shp->shm_perm, cmd);
1116 if (err)
1117 goto out_unlock;
1118
1119 ipc_lock_object(&shp->shm_perm);
1120
1121 if (!ipc_valid_object(&shp->shm_perm)) {
1122 ipc_unlock_object(&shp->shm_perm);
1123 err = -EIDRM;
1124 goto out_unlock;
1125 }
1126
1127 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
1128 tbuf->shm_segsz = shp->shm_segsz;
1129 tbuf->shm_atime = shp->shm_atim;
1130 tbuf->shm_dtime = shp->shm_dtim;
1131 tbuf->shm_ctime = shp->shm_ctim;
1132#ifndef CONFIG_64BIT
1133 tbuf->shm_atime_high = shp->shm_atim >> 32;
1134 tbuf->shm_dtime_high = shp->shm_dtim >> 32;
1135 tbuf->shm_ctime_high = shp->shm_ctim >> 32;
1136#endif
1137 tbuf->shm_cpid = pid_vnr(shp->shm_cprid);
1138 tbuf->shm_lpid = pid_vnr(shp->shm_lprid);
1139 tbuf->shm_nattch = shp->shm_nattch;
1140
1141 if (cmd == IPC_STAT) {
1142 /*
1143 * As defined in SUS:
1144 * Return 0 on success
1145 */
1146 err = 0;
1147 } else {
1148 /*
1149 * SHM_STAT and SHM_STAT_ANY (both Linux specific)
1150 * Return the full id, including the sequence number
1151 */
1152 err = shp->shm_perm.id;
1153 }
1154
1155 ipc_unlock_object(&shp->shm_perm);
1156out_unlock:
1157 rcu_read_unlock();
1158 return err;
1159}
1160
1161static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1162{
1163 struct shmid_kernel *shp;
1164 struct file *shm_file;
1165 int err;
1166
1167 rcu_read_lock();
1168 shp = shm_obtain_object_check(ns, shmid);
1169 if (IS_ERR(shp)) {
1170 err = PTR_ERR(shp);
1171 goto out_unlock1;
1172 }
1173
1174 audit_ipc_obj(&(shp->shm_perm));
1175 err = security_shm_shmctl(&shp->shm_perm, cmd);
1176 if (err)
1177 goto out_unlock1;
1178
1179 ipc_lock_object(&shp->shm_perm);
1180
1181 /* check if shm_destroy() is tearing down shp */
1182 if (!ipc_valid_object(&shp->shm_perm)) {
1183 err = -EIDRM;
1184 goto out_unlock0;
1185 }
1186
1187 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1188 kuid_t euid = current_euid();
1189
1190 if (!uid_eq(euid, shp->shm_perm.uid) &&
1191 !uid_eq(euid, shp->shm_perm.cuid)) {
1192 err = -EPERM;
1193 goto out_unlock0;
1194 }
1195 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1196 err = -EPERM;
1197 goto out_unlock0;
1198 }
1199 }
1200
1201 shm_file = shp->shm_file;
1202 if (is_file_hugepages(shm_file))
1203 goto out_unlock0;
1204
1205 if (cmd == SHM_LOCK) {
1206 struct ucounts *ucounts = current_ucounts();
1207
1208 err = shmem_lock(shm_file, 1, ucounts);
1209 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1210 shp->shm_perm.mode |= SHM_LOCKED;
1211 shp->mlock_ucounts = ucounts;
1212 }
1213 goto out_unlock0;
1214 }
1215
1216 /* SHM_UNLOCK */
1217 if (!(shp->shm_perm.mode & SHM_LOCKED))
1218 goto out_unlock0;
1219 shmem_lock(shm_file, 0, shp->mlock_ucounts);
1220 shp->shm_perm.mode &= ~SHM_LOCKED;
1221 shp->mlock_ucounts = NULL;
1222 get_file(shm_file);
1223 ipc_unlock_object(&shp->shm_perm);
1224 rcu_read_unlock();
1225 shmem_unlock_mapping(shm_file->f_mapping);
1226
1227 fput(shm_file);
1228 return err;
1229
1230out_unlock0:
1231 ipc_unlock_object(&shp->shm_perm);
1232out_unlock1:
1233 rcu_read_unlock();
1234 return err;
1235}
1236
1237static long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf, int version)
1238{
1239 int err;
1240 struct ipc_namespace *ns;
1241 struct shmid64_ds sem64;
1242
1243 if (cmd < 0 || shmid < 0)
1244 return -EINVAL;
1245
1246 ns = current->nsproxy->ipc_ns;
1247
1248 switch (cmd) {
1249 case IPC_INFO: {
1250 struct shminfo64 shminfo;
1251 err = shmctl_ipc_info(ns, &shminfo);
1252 if (err < 0)
1253 return err;
1254 if (copy_shminfo_to_user(buf, &shminfo, version))
1255 err = -EFAULT;
1256 return err;
1257 }
1258 case SHM_INFO: {
1259 struct shm_info shm_info;
1260 err = shmctl_shm_info(ns, &shm_info);
1261 if (err < 0)
1262 return err;
1263 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1264 err = -EFAULT;
1265 return err;
1266 }
1267 case SHM_STAT:
1268 case SHM_STAT_ANY:
1269 case IPC_STAT: {
1270 err = shmctl_stat(ns, shmid, cmd, &sem64);
1271 if (err < 0)
1272 return err;
1273 if (copy_shmid_to_user(buf, &sem64, version))
1274 err = -EFAULT;
1275 return err;
1276 }
1277 case IPC_SET:
1278 if (copy_shmid_from_user(&sem64, buf, version))
1279 return -EFAULT;
1280 fallthrough;
1281 case IPC_RMID:
1282 return shmctl_down(ns, shmid, cmd, &sem64);
1283 case SHM_LOCK:
1284 case SHM_UNLOCK:
1285 return shmctl_do_lock(ns, shmid, cmd);
1286 default:
1287 return -EINVAL;
1288 }
1289}
1290
1291SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1292{
1293 return ksys_shmctl(shmid, cmd, buf, IPC_64);
1294}
1295
1296#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1297long ksys_old_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1298{
1299 int version = ipc_parse_version(&cmd);
1300
1301 return ksys_shmctl(shmid, cmd, buf, version);
1302}
1303
1304SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1305{
1306 return ksys_old_shmctl(shmid, cmd, buf);
1307}
1308#endif
1309
1310#ifdef CONFIG_COMPAT
1311
1312struct compat_shmid_ds {
1313 struct compat_ipc_perm shm_perm;
1314 int shm_segsz;
1315 old_time32_t shm_atime;
1316 old_time32_t shm_dtime;
1317 old_time32_t shm_ctime;
1318 compat_ipc_pid_t shm_cpid;
1319 compat_ipc_pid_t shm_lpid;
1320 unsigned short shm_nattch;
1321 unsigned short shm_unused;
1322 compat_uptr_t shm_unused2;
1323 compat_uptr_t shm_unused3;
1324};
1325
1326struct compat_shminfo64 {
1327 compat_ulong_t shmmax;
1328 compat_ulong_t shmmin;
1329 compat_ulong_t shmmni;
1330 compat_ulong_t shmseg;
1331 compat_ulong_t shmall;
1332 compat_ulong_t __unused1;
1333 compat_ulong_t __unused2;
1334 compat_ulong_t __unused3;
1335 compat_ulong_t __unused4;
1336};
1337
1338struct compat_shm_info {
1339 compat_int_t used_ids;
1340 compat_ulong_t shm_tot, shm_rss, shm_swp;
1341 compat_ulong_t swap_attempts, swap_successes;
1342};
1343
1344static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1345 int version)
1346{
1347 if (in->shmmax > INT_MAX)
1348 in->shmmax = INT_MAX;
1349 if (version == IPC_64) {
1350 struct compat_shminfo64 info;
1351 memset(&info, 0, sizeof(info));
1352 info.shmmax = in->shmmax;
1353 info.shmmin = in->shmmin;
1354 info.shmmni = in->shmmni;
1355 info.shmseg = in->shmseg;
1356 info.shmall = in->shmall;
1357 return copy_to_user(buf, &info, sizeof(info));
1358 } else {
1359 struct shminfo info;
1360 memset(&info, 0, sizeof(info));
1361 info.shmmax = in->shmmax;
1362 info.shmmin = in->shmmin;
1363 info.shmmni = in->shmmni;
1364 info.shmseg = in->shmseg;
1365 info.shmall = in->shmall;
1366 return copy_to_user(buf, &info, sizeof(info));
1367 }
1368}
1369
1370static int put_compat_shm_info(struct shm_info *ip,
1371 struct compat_shm_info __user *uip)
1372{
1373 struct compat_shm_info info;
1374
1375 memset(&info, 0, sizeof(info));
1376 info.used_ids = ip->used_ids;
1377 info.shm_tot = ip->shm_tot;
1378 info.shm_rss = ip->shm_rss;
1379 info.shm_swp = ip->shm_swp;
1380 info.swap_attempts = ip->swap_attempts;
1381 info.swap_successes = ip->swap_successes;
1382 return copy_to_user(uip, &info, sizeof(info));
1383}
1384
1385static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1386 int version)
1387{
1388 if (version == IPC_64) {
1389 struct compat_shmid64_ds v;
1390 memset(&v, 0, sizeof(v));
1391 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1392 v.shm_atime = lower_32_bits(in->shm_atime);
1393 v.shm_atime_high = upper_32_bits(in->shm_atime);
1394 v.shm_dtime = lower_32_bits(in->shm_dtime);
1395 v.shm_dtime_high = upper_32_bits(in->shm_dtime);
1396 v.shm_ctime = lower_32_bits(in->shm_ctime);
1397 v.shm_ctime_high = upper_32_bits(in->shm_ctime);
1398 v.shm_segsz = in->shm_segsz;
1399 v.shm_nattch = in->shm_nattch;
1400 v.shm_cpid = in->shm_cpid;
1401 v.shm_lpid = in->shm_lpid;
1402 return copy_to_user(buf, &v, sizeof(v));
1403 } else {
1404 struct compat_shmid_ds v;
1405 memset(&v, 0, sizeof(v));
1406 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1407 v.shm_perm.key = in->shm_perm.key;
1408 v.shm_atime = in->shm_atime;
1409 v.shm_dtime = in->shm_dtime;
1410 v.shm_ctime = in->shm_ctime;
1411 v.shm_segsz = in->shm_segsz;
1412 v.shm_nattch = in->shm_nattch;
1413 v.shm_cpid = in->shm_cpid;
1414 v.shm_lpid = in->shm_lpid;
1415 return copy_to_user(buf, &v, sizeof(v));
1416 }
1417}
1418
1419static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1420 int version)
1421{
1422 memset(out, 0, sizeof(*out));
1423 if (version == IPC_64) {
1424 struct compat_shmid64_ds __user *p = buf;
1425 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1426 } else {
1427 struct compat_shmid_ds __user *p = buf;
1428 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1429 }
1430}
1431
1432static long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr, int version)
1433{
1434 struct ipc_namespace *ns;
1435 struct shmid64_ds sem64;
1436 int err;
1437
1438 ns = current->nsproxy->ipc_ns;
1439
1440 if (cmd < 0 || shmid < 0)
1441 return -EINVAL;
1442
1443 switch (cmd) {
1444 case IPC_INFO: {
1445 struct shminfo64 shminfo;
1446 err = shmctl_ipc_info(ns, &shminfo);
1447 if (err < 0)
1448 return err;
1449 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1450 err = -EFAULT;
1451 return err;
1452 }
1453 case SHM_INFO: {
1454 struct shm_info shm_info;
1455 err = shmctl_shm_info(ns, &shm_info);
1456 if (err < 0)
1457 return err;
1458 if (put_compat_shm_info(&shm_info, uptr))
1459 err = -EFAULT;
1460 return err;
1461 }
1462 case IPC_STAT:
1463 case SHM_STAT_ANY:
1464 case SHM_STAT:
1465 err = shmctl_stat(ns, shmid, cmd, &sem64);
1466 if (err < 0)
1467 return err;
1468 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1469 err = -EFAULT;
1470 return err;
1471
1472 case IPC_SET:
1473 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1474 return -EFAULT;
1475 fallthrough;
1476 case IPC_RMID:
1477 return shmctl_down(ns, shmid, cmd, &sem64);
1478 case SHM_LOCK:
1479 case SHM_UNLOCK:
1480 return shmctl_do_lock(ns, shmid, cmd);
1481 default:
1482 return -EINVAL;
1483 }
1484 return err;
1485}
1486
1487COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1488{
1489 return compat_ksys_shmctl(shmid, cmd, uptr, IPC_64);
1490}
1491
1492#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1493long compat_ksys_old_shmctl(int shmid, int cmd, void __user *uptr)
1494{
1495 int version = compat_ipc_parse_version(&cmd);
1496
1497 return compat_ksys_shmctl(shmid, cmd, uptr, version);
1498}
1499
1500COMPAT_SYSCALL_DEFINE3(old_shmctl, int, shmid, int, cmd, void __user *, uptr)
1501{
1502 return compat_ksys_old_shmctl(shmid, cmd, uptr);
1503}
1504#endif
1505#endif
1506
1507/*
1508 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1509 *
1510 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1511 * "raddr" thing points to kernel space, and there has to be a wrapper around
1512 * this.
1513 */
1514long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1515 ulong *raddr, unsigned long shmlba)
1516{
1517 struct shmid_kernel *shp;
1518 unsigned long addr = (unsigned long)shmaddr;
1519 unsigned long size;
1520 struct file *file, *base;
1521 int err;
1522 unsigned long flags = MAP_SHARED;
1523 unsigned long prot;
1524 int acc_mode;
1525 struct ipc_namespace *ns;
1526 struct shm_file_data *sfd;
1527 int f_flags;
1528 unsigned long populate = 0;
1529
1530 err = -EINVAL;
1531 if (shmid < 0)
1532 goto out;
1533
1534 if (addr) {
1535 if (addr & (shmlba - 1)) {
1536 if (shmflg & SHM_RND) {
1537 addr &= ~(shmlba - 1); /* round down */
1538
1539 /*
1540 * Ensure that the round-down is non-nil
1541 * when remapping. This can happen for
1542 * cases when addr < shmlba.
1543 */
1544 if (!addr && (shmflg & SHM_REMAP))
1545 goto out;
1546 } else
1547#ifndef __ARCH_FORCE_SHMLBA
1548 if (addr & ~PAGE_MASK)
1549#endif
1550 goto out;
1551 }
1552
1553 flags |= MAP_FIXED;
1554 } else if ((shmflg & SHM_REMAP))
1555 goto out;
1556
1557 if (shmflg & SHM_RDONLY) {
1558 prot = PROT_READ;
1559 acc_mode = S_IRUGO;
1560 f_flags = O_RDONLY;
1561 } else {
1562 prot = PROT_READ | PROT_WRITE;
1563 acc_mode = S_IRUGO | S_IWUGO;
1564 f_flags = O_RDWR;
1565 }
1566 if (shmflg & SHM_EXEC) {
1567 prot |= PROT_EXEC;
1568 acc_mode |= S_IXUGO;
1569 }
1570
1571 /*
1572 * We cannot rely on the fs check since SYSV IPC does have an
1573 * additional creator id...
1574 */
1575 ns = current->nsproxy->ipc_ns;
1576 rcu_read_lock();
1577 shp = shm_obtain_object_check(ns, shmid);
1578 if (IS_ERR(shp)) {
1579 err = PTR_ERR(shp);
1580 goto out_unlock;
1581 }
1582
1583 err = -EACCES;
1584 if (ipcperms(ns, &shp->shm_perm, acc_mode))
1585 goto out_unlock;
1586
1587 err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1588 if (err)
1589 goto out_unlock;
1590
1591 ipc_lock_object(&shp->shm_perm);
1592
1593 /* check if shm_destroy() is tearing down shp */
1594 if (!ipc_valid_object(&shp->shm_perm)) {
1595 ipc_unlock_object(&shp->shm_perm);
1596 err = -EIDRM;
1597 goto out_unlock;
1598 }
1599
1600 /*
1601 * We need to take a reference to the real shm file to prevent the
1602 * pointer from becoming stale in cases where the lifetime of the outer
1603 * file extends beyond that of the shm segment. It's not usually
1604 * possible, but it can happen during remap_file_pages() emulation as
1605 * that unmaps the memory, then does ->mmap() via file reference only.
1606 * We'll deny the ->mmap() if the shm segment was since removed, but to
1607 * detect shm ID reuse we need to compare the file pointers.
1608 */
1609 base = get_file(shp->shm_file);
1610 shp->shm_nattch++;
1611 size = i_size_read(file_inode(base));
1612 ipc_unlock_object(&shp->shm_perm);
1613 rcu_read_unlock();
1614
1615 err = -ENOMEM;
1616 sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1617 if (!sfd) {
1618 fput(base);
1619 goto out_nattch;
1620 }
1621
1622 file = alloc_file_clone(base, f_flags,
1623 is_file_hugepages(base) ?
1624 &shm_file_operations_huge :
1625 &shm_file_operations);
1626 err = PTR_ERR(file);
1627 if (IS_ERR(file)) {
1628 kfree(sfd);
1629 fput(base);
1630 goto out_nattch;
1631 }
1632
1633 sfd->id = shp->shm_perm.id;
1634 sfd->ns = get_ipc_ns(ns);
1635 sfd->file = base;
1636 sfd->vm_ops = NULL;
1637 file->private_data = sfd;
1638
1639 err = security_mmap_file(file, prot, flags);
1640 if (err)
1641 goto out_fput;
1642
1643 if (mmap_write_lock_killable(current->mm)) {
1644 err = -EINTR;
1645 goto out_fput;
1646 }
1647
1648 if (addr && !(shmflg & SHM_REMAP)) {
1649 err = -EINVAL;
1650 if (addr + size < addr)
1651 goto invalid;
1652
1653 if (find_vma_intersection(current->mm, addr, addr + size))
1654 goto invalid;
1655 }
1656
1657 addr = do_mmap(file, addr, size, prot, flags, 0, 0, &populate, NULL);
1658 *raddr = addr;
1659 err = 0;
1660 if (IS_ERR_VALUE(addr))
1661 err = (long)addr;
1662invalid:
1663 mmap_write_unlock(current->mm);
1664 if (populate)
1665 mm_populate(addr, populate);
1666
1667out_fput:
1668 fput(file);
1669
1670out_nattch:
1671 down_write(&shm_ids(ns).rwsem);
1672 shp = shm_lock(ns, shmid);
1673 shp->shm_nattch--;
1674
1675 if (shm_may_destroy(shp))
1676 shm_destroy(ns, shp);
1677 else
1678 shm_unlock(shp);
1679 up_write(&shm_ids(ns).rwsem);
1680 return err;
1681
1682out_unlock:
1683 rcu_read_unlock();
1684out:
1685 return err;
1686}
1687
1688SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1689{
1690 unsigned long ret;
1691 long err;
1692
1693 err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1694 if (err)
1695 return err;
1696 force_successful_syscall_return();
1697 return (long)ret;
1698}
1699
1700#ifdef CONFIG_COMPAT
1701
1702#ifndef COMPAT_SHMLBA
1703#define COMPAT_SHMLBA SHMLBA
1704#endif
1705
1706COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1707{
1708 unsigned long ret;
1709 long err;
1710
1711 err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1712 if (err)
1713 return err;
1714 force_successful_syscall_return();
1715 return (long)ret;
1716}
1717#endif
1718
1719/*
1720 * detach and kill segment if marked destroyed.
1721 * The work is done in shm_close.
1722 */
1723long ksys_shmdt(char __user *shmaddr)
1724{
1725 struct mm_struct *mm = current->mm;
1726 struct vm_area_struct *vma;
1727 unsigned long addr = (unsigned long)shmaddr;
1728 int retval = -EINVAL;
1729#ifdef CONFIG_MMU
1730 loff_t size = 0;
1731 struct file *file;
1732 VMA_ITERATOR(vmi, mm, addr);
1733#endif
1734
1735 if (addr & ~PAGE_MASK)
1736 return retval;
1737
1738 if (mmap_write_lock_killable(mm))
1739 return -EINTR;
1740
1741 /*
1742 * This function tries to be smart and unmap shm segments that
1743 * were modified by partial mlock or munmap calls:
1744 * - It first determines the size of the shm segment that should be
1745 * unmapped: It searches for a vma that is backed by shm and that
1746 * started at address shmaddr. It records it's size and then unmaps
1747 * it.
1748 * - Then it unmaps all shm vmas that started at shmaddr and that
1749 * are within the initially determined size and that are from the
1750 * same shm segment from which we determined the size.
1751 * Errors from do_munmap are ignored: the function only fails if
1752 * it's called with invalid parameters or if it's called to unmap
1753 * a part of a vma. Both calls in this function are for full vmas,
1754 * the parameters are directly copied from the vma itself and always
1755 * valid - therefore do_munmap cannot fail. (famous last words?)
1756 */
1757 /*
1758 * If it had been mremap()'d, the starting address would not
1759 * match the usual checks anyway. So assume all vma's are
1760 * above the starting address given.
1761 */
1762
1763#ifdef CONFIG_MMU
1764 for_each_vma(vmi, vma) {
1765 /*
1766 * Check if the starting address would match, i.e. it's
1767 * a fragment created by mprotect() and/or munmap(), or it
1768 * otherwise it starts at this address with no hassles.
1769 */
1770 if ((vma->vm_ops == &shm_vm_ops) &&
1771 (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1772
1773 /*
1774 * Record the file of the shm segment being
1775 * unmapped. With mremap(), someone could place
1776 * page from another segment but with equal offsets
1777 * in the range we are unmapping.
1778 */
1779 file = vma->vm_file;
1780 size = i_size_read(file_inode(vma->vm_file));
1781 do_vmi_align_munmap(&vmi, vma, mm, vma->vm_start,
1782 vma->vm_end, NULL, false);
1783 /*
1784 * We discovered the size of the shm segment, so
1785 * break out of here and fall through to the next
1786 * loop that uses the size information to stop
1787 * searching for matching vma's.
1788 */
1789 retval = 0;
1790 vma = vma_next(&vmi);
1791 break;
1792 }
1793 }
1794
1795 /*
1796 * We need look no further than the maximum address a fragment
1797 * could possibly have landed at. Also cast things to loff_t to
1798 * prevent overflows and make comparisons vs. equal-width types.
1799 */
1800 size = PAGE_ALIGN(size);
1801 while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1802 /* finding a matching vma now does not alter retval */
1803 if ((vma->vm_ops == &shm_vm_ops) &&
1804 ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1805 (vma->vm_file == file)) {
1806 do_vmi_align_munmap(&vmi, vma, mm, vma->vm_start,
1807 vma->vm_end, NULL, false);
1808 }
1809
1810 vma = vma_next(&vmi);
1811 }
1812
1813#else /* CONFIG_MMU */
1814 vma = vma_lookup(mm, addr);
1815 /* under NOMMU conditions, the exact address to be destroyed must be
1816 * given
1817 */
1818 if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1819 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1820 retval = 0;
1821 }
1822
1823#endif
1824
1825 mmap_write_unlock(mm);
1826 return retval;
1827}
1828
1829SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1830{
1831 return ksys_shmdt(shmaddr);
1832}
1833
1834#ifdef CONFIG_PROC_FS
1835static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1836{
1837 struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1838 struct user_namespace *user_ns = seq_user_ns(s);
1839 struct kern_ipc_perm *ipcp = it;
1840 struct shmid_kernel *shp;
1841 unsigned long rss = 0, swp = 0;
1842
1843 shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1844 shm_add_rss_swap(shp, &rss, &swp);
1845
1846#if BITS_PER_LONG <= 32
1847#define SIZE_SPEC "%10lu"
1848#else
1849#define SIZE_SPEC "%21lu"
1850#endif
1851
1852 seq_printf(s,
1853 "%10d %10d %4o " SIZE_SPEC " %5u %5u "
1854 "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1855 SIZE_SPEC " " SIZE_SPEC "\n",
1856 shp->shm_perm.key,
1857 shp->shm_perm.id,
1858 shp->shm_perm.mode,
1859 shp->shm_segsz,
1860 pid_nr_ns(shp->shm_cprid, pid_ns),
1861 pid_nr_ns(shp->shm_lprid, pid_ns),
1862 shp->shm_nattch,
1863 from_kuid_munged(user_ns, shp->shm_perm.uid),
1864 from_kgid_munged(user_ns, shp->shm_perm.gid),
1865 from_kuid_munged(user_ns, shp->shm_perm.cuid),
1866 from_kgid_munged(user_ns, shp->shm_perm.cgid),
1867 shp->shm_atim,
1868 shp->shm_dtim,
1869 shp->shm_ctim,
1870 rss * PAGE_SIZE,
1871 swp * PAGE_SIZE);
1872
1873 return 0;
1874}
1875#endif