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/*
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