Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/scottwood/linux.git
[sfrench/cifs-2.6.git] / ipc / shm.c
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
50 struct 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
59 static const struct file_operations shm_file_operations;
60 static 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
67 static int newseg(struct ipc_namespace *, struct ipc_params *);
68 static void shm_open(struct vm_area_struct *vma);
69 static void shm_close(struct vm_area_struct *vma);
70 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
71 #ifdef CONFIG_PROC_FS
72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
73 #endif
74
75 void 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  */
89 static 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
104 void 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
111 static int __init ipc_ns_init(void)
112 {
113         shm_init_ns(&init_ipc_ns);
114         return 0;
115 }
116
117 pure_initcall(ipc_ns_init);
118
119 void __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
130 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
131 {
132         struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
133
134         if (IS_ERR(ipcp))
135                 return ERR_CAST(ipcp);
136
137         return container_of(ipcp, struct shmid_kernel, shm_perm);
138 }
139
140 static 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  */
154 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
155 {
156         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
157
158         if (IS_ERR(ipcp))
159                 return (struct shmid_kernel *)ipcp;
160
161         return container_of(ipcp, struct shmid_kernel, shm_perm);
162 }
163
164 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
165 {
166         rcu_read_lock();
167         ipc_lock_object(&ipcp->shm_perm);
168 }
169
170 static void shm_rcu_free(struct rcu_head *head)
171 {
172         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
173         struct shmid_kernel *shp = ipc_rcu_to_struct(p);
174
175         security_shm_free(shp);
176         ipc_rcu_free(head);
177 }
178
179 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
180 {
181         list_del(&s->shm_clist);
182         ipc_rmid(&shm_ids(ns), &s->shm_perm);
183 }
184
185
186 /* This is called by fork, once for every shm attach. */
187 static void shm_open(struct vm_area_struct *vma)
188 {
189         struct file *file = vma->vm_file;
190         struct shm_file_data *sfd = shm_file_data(file);
191         struct shmid_kernel *shp;
192
193         shp = shm_lock(sfd->ns, sfd->id);
194         BUG_ON(IS_ERR(shp));
195         shp->shm_atim = get_seconds();
196         shp->shm_lprid = task_tgid_vnr(current);
197         shp->shm_nattch++;
198         shm_unlock(shp);
199 }
200
201 /*
202  * shm_destroy - free the struct shmid_kernel
203  *
204  * @ns: namespace
205  * @shp: struct to free
206  *
207  * It has to be called with shp and shm_ids.rwsem (writer) locked,
208  * but returns with shp unlocked and freed.
209  */
210 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
211 {
212         struct file *shm_file;
213
214         shm_file = shp->shm_file;
215         shp->shm_file = NULL;
216         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
217         shm_rmid(ns, shp);
218         shm_unlock(shp);
219         if (!is_file_hugepages(shm_file))
220                 shmem_lock(shm_file, 0, shp->mlock_user);
221         else if (shp->mlock_user)
222                 user_shm_unlock(file_inode(shm_file)->i_size, shp->mlock_user);
223         fput(shm_file);
224         ipc_rcu_putref(shp, shm_rcu_free);
225 }
226
227 /*
228  * shm_may_destroy - identifies whether shm segment should be destroyed now
229  *
230  * Returns true if and only if there are no active users of the segment and
231  * one of the following is true:
232  *
233  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
234  *
235  * 2) sysctl kernel.shm_rmid_forced is set to 1.
236  */
237 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
238 {
239         return (shp->shm_nattch == 0) &&
240                (ns->shm_rmid_forced ||
241                 (shp->shm_perm.mode & SHM_DEST));
242 }
243
244 /*
245  * remove the attach descriptor vma.
246  * free memory for segment if it is marked destroyed.
247  * The descriptor has already been removed from the current->mm->mmap list
248  * and will later be kfree()d.
249  */
250 static void shm_close(struct vm_area_struct *vma)
251 {
252         struct file *file = vma->vm_file;
253         struct shm_file_data *sfd = shm_file_data(file);
254         struct shmid_kernel *shp;
255         struct ipc_namespace *ns = sfd->ns;
256
257         down_write(&shm_ids(ns).rwsem);
258         /* remove from the list of attaches of the shm segment */
259         shp = shm_lock(ns, sfd->id);
260         BUG_ON(IS_ERR(shp));
261         shp->shm_lprid = task_tgid_vnr(current);
262         shp->shm_dtim = get_seconds();
263         shp->shm_nattch--;
264         if (shm_may_destroy(ns, shp))
265                 shm_destroy(ns, shp);
266         else
267                 shm_unlock(shp);
268         up_write(&shm_ids(ns).rwsem);
269 }
270
271 /* Called with ns->shm_ids(ns).rwsem locked */
272 static int shm_try_destroy_orphaned(int id, void *p, void *data)
273 {
274         struct ipc_namespace *ns = data;
275         struct kern_ipc_perm *ipcp = p;
276         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
277
278         /*
279          * We want to destroy segments without users and with already
280          * exit'ed originating process.
281          *
282          * As shp->* are changed under rwsem, it's safe to skip shp locking.
283          */
284         if (shp->shm_creator != NULL)
285                 return 0;
286
287         if (shm_may_destroy(ns, shp)) {
288                 shm_lock_by_ptr(shp);
289                 shm_destroy(ns, shp);
290         }
291         return 0;
292 }
293
294 void shm_destroy_orphaned(struct ipc_namespace *ns)
295 {
296         down_write(&shm_ids(ns).rwsem);
297         if (shm_ids(ns).in_use)
298                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
299         up_write(&shm_ids(ns).rwsem);
300 }
301
302 /* Locking assumes this will only be called with task == current */
303 void exit_shm(struct task_struct *task)
304 {
305         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
306         struct shmid_kernel *shp, *n;
307
308         if (list_empty(&task->sysvshm.shm_clist))
309                 return;
310
311         /*
312          * If kernel.shm_rmid_forced is not set then only keep track of
313          * which shmids are orphaned, so that a later set of the sysctl
314          * can clean them up.
315          */
316         if (!ns->shm_rmid_forced) {
317                 down_read(&shm_ids(ns).rwsem);
318                 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
319                         shp->shm_creator = NULL;
320                 /*
321                  * Only under read lock but we are only called on current
322                  * so no entry on the list will be shared.
323                  */
324                 list_del(&task->sysvshm.shm_clist);
325                 up_read(&shm_ids(ns).rwsem);
326                 return;
327         }
328
329         /*
330          * Destroy all already created segments, that were not yet mapped,
331          * and mark any mapped as orphan to cover the sysctl toggling.
332          * Destroy is skipped if shm_may_destroy() returns false.
333          */
334         down_write(&shm_ids(ns).rwsem);
335         list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
336                 shp->shm_creator = NULL;
337
338                 if (shm_may_destroy(ns, shp)) {
339                         shm_lock_by_ptr(shp);
340                         shm_destroy(ns, shp);
341                 }
342         }
343
344         /* Remove the list head from any segments still attached. */
345         list_del(&task->sysvshm.shm_clist);
346         up_write(&shm_ids(ns).rwsem);
347 }
348
349 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
350 {
351         struct file *file = vma->vm_file;
352         struct shm_file_data *sfd = shm_file_data(file);
353
354         return sfd->vm_ops->fault(vma, vmf);
355 }
356
357 #ifdef CONFIG_NUMA
358 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
359 {
360         struct file *file = vma->vm_file;
361         struct shm_file_data *sfd = shm_file_data(file);
362         int err = 0;
363         if (sfd->vm_ops->set_policy)
364                 err = sfd->vm_ops->set_policy(vma, new);
365         return err;
366 }
367
368 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
369                                         unsigned long addr)
370 {
371         struct file *file = vma->vm_file;
372         struct shm_file_data *sfd = shm_file_data(file);
373         struct mempolicy *pol = NULL;
374
375         if (sfd->vm_ops->get_policy)
376                 pol = sfd->vm_ops->get_policy(vma, addr);
377         else if (vma->vm_policy)
378                 pol = vma->vm_policy;
379
380         return pol;
381 }
382 #endif
383
384 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
385 {
386         struct shm_file_data *sfd = shm_file_data(file);
387         int ret;
388
389         ret = sfd->file->f_op->mmap(sfd->file, vma);
390         if (ret != 0)
391                 return ret;
392         sfd->vm_ops = vma->vm_ops;
393 #ifdef CONFIG_MMU
394         BUG_ON(!sfd->vm_ops->fault);
395 #endif
396         vma->vm_ops = &shm_vm_ops;
397         shm_open(vma);
398
399         return ret;
400 }
401
402 static int shm_release(struct inode *ino, struct file *file)
403 {
404         struct shm_file_data *sfd = shm_file_data(file);
405
406         put_ipc_ns(sfd->ns);
407         shm_file_data(file) = NULL;
408         kfree(sfd);
409         return 0;
410 }
411
412 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
413 {
414         struct shm_file_data *sfd = shm_file_data(file);
415
416         if (!sfd->file->f_op->fsync)
417                 return -EINVAL;
418         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
419 }
420
421 static long shm_fallocate(struct file *file, int mode, loff_t offset,
422                           loff_t len)
423 {
424         struct shm_file_data *sfd = shm_file_data(file);
425
426         if (!sfd->file->f_op->fallocate)
427                 return -EOPNOTSUPP;
428         return sfd->file->f_op->fallocate(file, mode, offset, len);
429 }
430
431 static unsigned long shm_get_unmapped_area(struct file *file,
432         unsigned long addr, unsigned long len, unsigned long pgoff,
433         unsigned long flags)
434 {
435         struct shm_file_data *sfd = shm_file_data(file);
436         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
437                                                 pgoff, flags);
438 }
439
440 static const struct file_operations shm_file_operations = {
441         .mmap           = shm_mmap,
442         .fsync          = shm_fsync,
443         .release        = shm_release,
444 #ifndef CONFIG_MMU
445         .get_unmapped_area      = shm_get_unmapped_area,
446 #endif
447         .llseek         = noop_llseek,
448         .fallocate      = shm_fallocate,
449 };
450
451 static const struct file_operations shm_file_operations_huge = {
452         .mmap           = shm_mmap,
453         .fsync          = shm_fsync,
454         .release        = shm_release,
455         .get_unmapped_area      = shm_get_unmapped_area,
456         .llseek         = noop_llseek,
457         .fallocate      = shm_fallocate,
458 };
459
460 int is_file_shm_hugepages(struct file *file)
461 {
462         return file->f_op == &shm_file_operations_huge;
463 }
464
465 static const struct vm_operations_struct shm_vm_ops = {
466         .open   = shm_open,     /* callback for a new vm-area open */
467         .close  = shm_close,    /* callback for when the vm-area is released */
468         .fault  = shm_fault,
469 #if defined(CONFIG_NUMA)
470         .set_policy = shm_set_policy,
471         .get_policy = shm_get_policy,
472 #endif
473 };
474
475 /**
476  * newseg - Create a new shared memory segment
477  * @ns: namespace
478  * @params: ptr to the structure that contains key, size and shmflg
479  *
480  * Called with shm_ids.rwsem held as a writer.
481  */
482 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
483 {
484         key_t key = params->key;
485         int shmflg = params->flg;
486         size_t size = params->u.size;
487         int error;
488         struct shmid_kernel *shp;
489         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
490         struct file *file;
491         char name[13];
492         int id;
493         vm_flags_t acctflag = 0;
494
495         if (size < SHMMIN || size > ns->shm_ctlmax)
496                 return -EINVAL;
497
498         if (numpages << PAGE_SHIFT < size)
499                 return -ENOSPC;
500
501         if (ns->shm_tot + numpages < ns->shm_tot ||
502                         ns->shm_tot + numpages > ns->shm_ctlall)
503                 return -ENOSPC;
504
505         shp = ipc_rcu_alloc(sizeof(*shp));
506         if (!shp)
507                 return -ENOMEM;
508
509         shp->shm_perm.key = key;
510         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
511         shp->mlock_user = NULL;
512
513         shp->shm_perm.security = NULL;
514         error = security_shm_alloc(shp);
515         if (error) {
516                 ipc_rcu_putref(shp, ipc_rcu_free);
517                 return error;
518         }
519
520         sprintf(name, "SYSV%08x", key);
521         if (shmflg & SHM_HUGETLB) {
522                 struct hstate *hs;
523                 size_t hugesize;
524
525                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
526                 if (!hs) {
527                         error = -EINVAL;
528                         goto no_file;
529                 }
530                 hugesize = ALIGN(size, huge_page_size(hs));
531
532                 /* hugetlb_file_setup applies strict accounting */
533                 if (shmflg & SHM_NORESERVE)
534                         acctflag = VM_NORESERVE;
535                 file = hugetlb_file_setup(name, hugesize, acctflag,
536                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
537                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
538         } else {
539                 /*
540                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
541                  * if it's asked for.
542                  */
543                 if  ((shmflg & SHM_NORESERVE) &&
544                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
545                         acctflag = VM_NORESERVE;
546                 file = shmem_file_setup(name, size, acctflag);
547         }
548         error = PTR_ERR(file);
549         if (IS_ERR(file))
550                 goto no_file;
551
552         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
553         if (id < 0) {
554                 error = id;
555                 goto no_id;
556         }
557
558         shp->shm_cprid = task_tgid_vnr(current);
559         shp->shm_lprid = 0;
560         shp->shm_atim = shp->shm_dtim = 0;
561         shp->shm_ctim = get_seconds();
562         shp->shm_segsz = size;
563         shp->shm_nattch = 0;
564         shp->shm_file = file;
565         shp->shm_creator = current;
566         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
567
568         /*
569          * shmid gets reported as "inode#" in /proc/pid/maps.
570          * proc-ps tools use this. Changing this will break them.
571          */
572         file_inode(file)->i_ino = shp->shm_perm.id;
573
574         ns->shm_tot += numpages;
575         error = shp->shm_perm.id;
576
577         ipc_unlock_object(&shp->shm_perm);
578         rcu_read_unlock();
579         return error;
580
581 no_id:
582         if (is_file_hugepages(file) && shp->mlock_user)
583                 user_shm_unlock(size, shp->mlock_user);
584         fput(file);
585 no_file:
586         ipc_rcu_putref(shp, shm_rcu_free);
587         return error;
588 }
589
590 /*
591  * Called with shm_ids.rwsem and ipcp locked.
592  */
593 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
594 {
595         struct shmid_kernel *shp;
596
597         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
598         return security_shm_associate(shp, shmflg);
599 }
600
601 /*
602  * Called with shm_ids.rwsem and ipcp locked.
603  */
604 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
605                                 struct ipc_params *params)
606 {
607         struct shmid_kernel *shp;
608
609         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
610         if (shp->shm_segsz < params->u.size)
611                 return -EINVAL;
612
613         return 0;
614 }
615
616 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
617 {
618         struct ipc_namespace *ns;
619         static const struct ipc_ops shm_ops = {
620                 .getnew = newseg,
621                 .associate = shm_security,
622                 .more_checks = shm_more_checks,
623         };
624         struct ipc_params shm_params;
625
626         ns = current->nsproxy->ipc_ns;
627
628         shm_params.key = key;
629         shm_params.flg = shmflg;
630         shm_params.u.size = size;
631
632         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
633 }
634
635 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
636 {
637         switch (version) {
638         case IPC_64:
639                 return copy_to_user(buf, in, sizeof(*in));
640         case IPC_OLD:
641             {
642                 struct shmid_ds out;
643
644                 memset(&out, 0, sizeof(out));
645                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
646                 out.shm_segsz   = in->shm_segsz;
647                 out.shm_atime   = in->shm_atime;
648                 out.shm_dtime   = in->shm_dtime;
649                 out.shm_ctime   = in->shm_ctime;
650                 out.shm_cpid    = in->shm_cpid;
651                 out.shm_lpid    = in->shm_lpid;
652                 out.shm_nattch  = in->shm_nattch;
653
654                 return copy_to_user(buf, &out, sizeof(out));
655             }
656         default:
657                 return -EINVAL;
658         }
659 }
660
661 static inline unsigned long
662 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
663 {
664         switch (version) {
665         case IPC_64:
666                 if (copy_from_user(out, buf, sizeof(*out)))
667                         return -EFAULT;
668                 return 0;
669         case IPC_OLD:
670             {
671                 struct shmid_ds tbuf_old;
672
673                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
674                         return -EFAULT;
675
676                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
677                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
678                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
679
680                 return 0;
681             }
682         default:
683                 return -EINVAL;
684         }
685 }
686
687 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
688 {
689         switch (version) {
690         case IPC_64:
691                 return copy_to_user(buf, in, sizeof(*in));
692         case IPC_OLD:
693             {
694                 struct shminfo out;
695
696                 if (in->shmmax > INT_MAX)
697                         out.shmmax = INT_MAX;
698                 else
699                         out.shmmax = (int)in->shmmax;
700
701                 out.shmmin      = in->shmmin;
702                 out.shmmni      = in->shmmni;
703                 out.shmseg      = in->shmseg;
704                 out.shmall      = in->shmall;
705
706                 return copy_to_user(buf, &out, sizeof(out));
707             }
708         default:
709                 return -EINVAL;
710         }
711 }
712
713 /*
714  * Calculate and add used RSS and swap pages of a shm.
715  * Called with shm_ids.rwsem held as a reader
716  */
717 static void shm_add_rss_swap(struct shmid_kernel *shp,
718         unsigned long *rss_add, unsigned long *swp_add)
719 {
720         struct inode *inode;
721
722         inode = file_inode(shp->shm_file);
723
724         if (is_file_hugepages(shp->shm_file)) {
725                 struct address_space *mapping = inode->i_mapping;
726                 struct hstate *h = hstate_file(shp->shm_file);
727                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
728         } else {
729 #ifdef CONFIG_SHMEM
730                 struct shmem_inode_info *info = SHMEM_I(inode);
731                 spin_lock(&info->lock);
732                 *rss_add += inode->i_mapping->nrpages;
733                 *swp_add += info->swapped;
734                 spin_unlock(&info->lock);
735 #else
736                 *rss_add += inode->i_mapping->nrpages;
737 #endif
738         }
739 }
740
741 /*
742  * Called with shm_ids.rwsem held as a reader
743  */
744 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
745                 unsigned long *swp)
746 {
747         int next_id;
748         int total, in_use;
749
750         *rss = 0;
751         *swp = 0;
752
753         in_use = shm_ids(ns).in_use;
754
755         for (total = 0, next_id = 0; total < in_use; next_id++) {
756                 struct kern_ipc_perm *ipc;
757                 struct shmid_kernel *shp;
758
759                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
760                 if (ipc == NULL)
761                         continue;
762                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
763
764                 shm_add_rss_swap(shp, rss, swp);
765
766                 total++;
767         }
768 }
769
770 /*
771  * This function handles some shmctl commands which require the rwsem
772  * to be held in write mode.
773  * NOTE: no locks must be held, the rwsem is taken inside this function.
774  */
775 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
776                        struct shmid_ds __user *buf, int version)
777 {
778         struct kern_ipc_perm *ipcp;
779         struct shmid64_ds shmid64;
780         struct shmid_kernel *shp;
781         int err;
782
783         if (cmd == IPC_SET) {
784                 if (copy_shmid_from_user(&shmid64, buf, version))
785                         return -EFAULT;
786         }
787
788         down_write(&shm_ids(ns).rwsem);
789         rcu_read_lock();
790
791         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
792                                       &shmid64.shm_perm, 0);
793         if (IS_ERR(ipcp)) {
794                 err = PTR_ERR(ipcp);
795                 goto out_unlock1;
796         }
797
798         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
799
800         err = security_shm_shmctl(shp, cmd);
801         if (err)
802                 goto out_unlock1;
803
804         switch (cmd) {
805         case IPC_RMID:
806                 ipc_lock_object(&shp->shm_perm);
807                 /* do_shm_rmid unlocks the ipc object and rcu */
808                 do_shm_rmid(ns, ipcp);
809                 goto out_up;
810         case IPC_SET:
811                 ipc_lock_object(&shp->shm_perm);
812                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
813                 if (err)
814                         goto out_unlock0;
815                 shp->shm_ctim = get_seconds();
816                 break;
817         default:
818                 err = -EINVAL;
819                 goto out_unlock1;
820         }
821
822 out_unlock0:
823         ipc_unlock_object(&shp->shm_perm);
824 out_unlock1:
825         rcu_read_unlock();
826 out_up:
827         up_write(&shm_ids(ns).rwsem);
828         return err;
829 }
830
831 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
832                          int cmd, int version, void __user *buf)
833 {
834         int err;
835         struct shmid_kernel *shp;
836
837         /* preliminary security checks for *_INFO */
838         if (cmd == IPC_INFO || cmd == SHM_INFO) {
839                 err = security_shm_shmctl(NULL, cmd);
840                 if (err)
841                         return err;
842         }
843
844         switch (cmd) {
845         case IPC_INFO:
846         {
847                 struct shminfo64 shminfo;
848
849                 memset(&shminfo, 0, sizeof(shminfo));
850                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
851                 shminfo.shmmax = ns->shm_ctlmax;
852                 shminfo.shmall = ns->shm_ctlall;
853
854                 shminfo.shmmin = SHMMIN;
855                 if (copy_shminfo_to_user(buf, &shminfo, version))
856                         return -EFAULT;
857
858                 down_read(&shm_ids(ns).rwsem);
859                 err = ipc_get_maxid(&shm_ids(ns));
860                 up_read(&shm_ids(ns).rwsem);
861
862                 if (err < 0)
863                         err = 0;
864                 goto out;
865         }
866         case SHM_INFO:
867         {
868                 struct shm_info shm_info;
869
870                 memset(&shm_info, 0, sizeof(shm_info));
871                 down_read(&shm_ids(ns).rwsem);
872                 shm_info.used_ids = shm_ids(ns).in_use;
873                 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
874                 shm_info.shm_tot = ns->shm_tot;
875                 shm_info.swap_attempts = 0;
876                 shm_info.swap_successes = 0;
877                 err = ipc_get_maxid(&shm_ids(ns));
878                 up_read(&shm_ids(ns).rwsem);
879                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
880                         err = -EFAULT;
881                         goto out;
882                 }
883
884                 err = err < 0 ? 0 : err;
885                 goto out;
886         }
887         case SHM_STAT:
888         case IPC_STAT:
889         {
890                 struct shmid64_ds tbuf;
891                 int result;
892
893                 rcu_read_lock();
894                 if (cmd == SHM_STAT) {
895                         shp = shm_obtain_object(ns, shmid);
896                         if (IS_ERR(shp)) {
897                                 err = PTR_ERR(shp);
898                                 goto out_unlock;
899                         }
900                         result = shp->shm_perm.id;
901                 } else {
902                         shp = shm_obtain_object_check(ns, shmid);
903                         if (IS_ERR(shp)) {
904                                 err = PTR_ERR(shp);
905                                 goto out_unlock;
906                         }
907                         result = 0;
908                 }
909
910                 err = -EACCES;
911                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
912                         goto out_unlock;
913
914                 err = security_shm_shmctl(shp, cmd);
915                 if (err)
916                         goto out_unlock;
917
918                 memset(&tbuf, 0, sizeof(tbuf));
919                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
920                 tbuf.shm_segsz  = shp->shm_segsz;
921                 tbuf.shm_atime  = shp->shm_atim;
922                 tbuf.shm_dtime  = shp->shm_dtim;
923                 tbuf.shm_ctime  = shp->shm_ctim;
924                 tbuf.shm_cpid   = shp->shm_cprid;
925                 tbuf.shm_lpid   = shp->shm_lprid;
926                 tbuf.shm_nattch = shp->shm_nattch;
927                 rcu_read_unlock();
928
929                 if (copy_shmid_to_user(buf, &tbuf, version))
930                         err = -EFAULT;
931                 else
932                         err = result;
933                 goto out;
934         }
935         default:
936                 return -EINVAL;
937         }
938
939 out_unlock:
940         rcu_read_unlock();
941 out:
942         return err;
943 }
944
945 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
946 {
947         struct shmid_kernel *shp;
948         int err, version;
949         struct ipc_namespace *ns;
950
951         if (cmd < 0 || shmid < 0)
952                 return -EINVAL;
953
954         version = ipc_parse_version(&cmd);
955         ns = current->nsproxy->ipc_ns;
956
957         switch (cmd) {
958         case IPC_INFO:
959         case SHM_INFO:
960         case SHM_STAT:
961         case IPC_STAT:
962                 return shmctl_nolock(ns, shmid, cmd, version, buf);
963         case IPC_RMID:
964         case IPC_SET:
965                 return shmctl_down(ns, shmid, cmd, buf, version);
966         case SHM_LOCK:
967         case SHM_UNLOCK:
968         {
969                 struct file *shm_file;
970
971                 rcu_read_lock();
972                 shp = shm_obtain_object_check(ns, shmid);
973                 if (IS_ERR(shp)) {
974                         err = PTR_ERR(shp);
975                         goto out_unlock1;
976                 }
977
978                 audit_ipc_obj(&(shp->shm_perm));
979                 err = security_shm_shmctl(shp, cmd);
980                 if (err)
981                         goto out_unlock1;
982
983                 ipc_lock_object(&shp->shm_perm);
984
985                 /* check if shm_destroy() is tearing down shp */
986                 if (!ipc_valid_object(&shp->shm_perm)) {
987                         err = -EIDRM;
988                         goto out_unlock0;
989                 }
990
991                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
992                         kuid_t euid = current_euid();
993                         if (!uid_eq(euid, shp->shm_perm.uid) &&
994                             !uid_eq(euid, shp->shm_perm.cuid)) {
995                                 err = -EPERM;
996                                 goto out_unlock0;
997                         }
998                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
999                                 err = -EPERM;
1000                                 goto out_unlock0;
1001                         }
1002                 }
1003
1004                 shm_file = shp->shm_file;
1005                 if (is_file_hugepages(shm_file))
1006                         goto out_unlock0;
1007
1008                 if (cmd == SHM_LOCK) {
1009                         struct user_struct *user = current_user();
1010                         err = shmem_lock(shm_file, 1, user);
1011                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1012                                 shp->shm_perm.mode |= SHM_LOCKED;
1013                                 shp->mlock_user = user;
1014                         }
1015                         goto out_unlock0;
1016                 }
1017
1018                 /* SHM_UNLOCK */
1019                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1020                         goto out_unlock0;
1021                 shmem_lock(shm_file, 0, shp->mlock_user);
1022                 shp->shm_perm.mode &= ~SHM_LOCKED;
1023                 shp->mlock_user = NULL;
1024                 get_file(shm_file);
1025                 ipc_unlock_object(&shp->shm_perm);
1026                 rcu_read_unlock();
1027                 shmem_unlock_mapping(shm_file->f_mapping);
1028
1029                 fput(shm_file);
1030                 return err;
1031         }
1032         default:
1033                 return -EINVAL;
1034         }
1035
1036 out_unlock0:
1037         ipc_unlock_object(&shp->shm_perm);
1038 out_unlock1:
1039         rcu_read_unlock();
1040         return err;
1041 }
1042
1043 /*
1044  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1045  *
1046  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1047  * "raddr" thing points to kernel space, and there has to be a wrapper around
1048  * this.
1049  */
1050 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1051               unsigned long shmlba)
1052 {
1053         struct shmid_kernel *shp;
1054         unsigned long addr;
1055         unsigned long size;
1056         struct file *file;
1057         int    err;
1058         unsigned long flags;
1059         unsigned long prot;
1060         int acc_mode;
1061         struct ipc_namespace *ns;
1062         struct shm_file_data *sfd;
1063         struct path path;
1064         fmode_t f_mode;
1065         unsigned long populate = 0;
1066
1067         err = -EINVAL;
1068         if (shmid < 0)
1069                 goto out;
1070         else if ((addr = (ulong)shmaddr)) {
1071                 if (addr & (shmlba - 1)) {
1072                         if (shmflg & SHM_RND)
1073                                 addr &= ~(shmlba - 1);     /* round down */
1074                         else
1075 #ifndef __ARCH_FORCE_SHMLBA
1076                                 if (addr & ~PAGE_MASK)
1077 #endif
1078                                         goto out;
1079                 }
1080                 flags = MAP_SHARED | MAP_FIXED;
1081         } else {
1082                 if ((shmflg & SHM_REMAP))
1083                         goto out;
1084
1085                 flags = MAP_SHARED;
1086         }
1087
1088         if (shmflg & SHM_RDONLY) {
1089                 prot = PROT_READ;
1090                 acc_mode = S_IRUGO;
1091                 f_mode = FMODE_READ;
1092         } else {
1093                 prot = PROT_READ | PROT_WRITE;
1094                 acc_mode = S_IRUGO | S_IWUGO;
1095                 f_mode = FMODE_READ | FMODE_WRITE;
1096         }
1097         if (shmflg & SHM_EXEC) {
1098                 prot |= PROT_EXEC;
1099                 acc_mode |= S_IXUGO;
1100         }
1101
1102         /*
1103          * We cannot rely on the fs check since SYSV IPC does have an
1104          * additional creator id...
1105          */
1106         ns = current->nsproxy->ipc_ns;
1107         rcu_read_lock();
1108         shp = shm_obtain_object_check(ns, shmid);
1109         if (IS_ERR(shp)) {
1110                 err = PTR_ERR(shp);
1111                 goto out_unlock;
1112         }
1113
1114         err = -EACCES;
1115         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1116                 goto out_unlock;
1117
1118         err = security_shm_shmat(shp, shmaddr, shmflg);
1119         if (err)
1120                 goto out_unlock;
1121
1122         ipc_lock_object(&shp->shm_perm);
1123
1124         /* check if shm_destroy() is tearing down shp */
1125         if (!ipc_valid_object(&shp->shm_perm)) {
1126                 ipc_unlock_object(&shp->shm_perm);
1127                 err = -EIDRM;
1128                 goto out_unlock;
1129         }
1130
1131         path = shp->shm_file->f_path;
1132         path_get(&path);
1133         shp->shm_nattch++;
1134         size = i_size_read(path.dentry->d_inode);
1135         ipc_unlock_object(&shp->shm_perm);
1136         rcu_read_unlock();
1137
1138         err = -ENOMEM;
1139         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1140         if (!sfd) {
1141                 path_put(&path);
1142                 goto out_nattch;
1143         }
1144
1145         file = alloc_file(&path, f_mode,
1146                           is_file_hugepages(shp->shm_file) ?
1147                                 &shm_file_operations_huge :
1148                                 &shm_file_operations);
1149         err = PTR_ERR(file);
1150         if (IS_ERR(file)) {
1151                 kfree(sfd);
1152                 path_put(&path);
1153                 goto out_nattch;
1154         }
1155
1156         file->private_data = sfd;
1157         file->f_mapping = shp->shm_file->f_mapping;
1158         sfd->id = shp->shm_perm.id;
1159         sfd->ns = get_ipc_ns(ns);
1160         sfd->file = shp->shm_file;
1161         sfd->vm_ops = NULL;
1162
1163         err = security_mmap_file(file, prot, flags);
1164         if (err)
1165                 goto out_fput;
1166
1167         down_write(&current->mm->mmap_sem);
1168         if (addr && !(shmflg & SHM_REMAP)) {
1169                 err = -EINVAL;
1170                 if (addr + size < addr)
1171                         goto invalid;
1172
1173                 if (find_vma_intersection(current->mm, addr, addr + size))
1174                         goto invalid;
1175                 /*
1176                  * If shm segment goes below stack, make sure there is some
1177                  * space left for the stack to grow (at least 4 pages).
1178                  */
1179                 if (addr < current->mm->start_stack &&
1180                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1181                         goto invalid;
1182         }
1183
1184         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1185         *raddr = addr;
1186         err = 0;
1187         if (IS_ERR_VALUE(addr))
1188                 err = (long)addr;
1189 invalid:
1190         up_write(&current->mm->mmap_sem);
1191         if (populate)
1192                 mm_populate(addr, populate);
1193
1194 out_fput:
1195         fput(file);
1196
1197 out_nattch:
1198         down_write(&shm_ids(ns).rwsem);
1199         shp = shm_lock(ns, shmid);
1200         BUG_ON(IS_ERR(shp));
1201         shp->shm_nattch--;
1202         if (shm_may_destroy(ns, shp))
1203                 shm_destroy(ns, shp);
1204         else
1205                 shm_unlock(shp);
1206         up_write(&shm_ids(ns).rwsem);
1207         return err;
1208
1209 out_unlock:
1210         rcu_read_unlock();
1211 out:
1212         return err;
1213 }
1214
1215 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1216 {
1217         unsigned long ret;
1218         long err;
1219
1220         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1221         if (err)
1222                 return err;
1223         force_successful_syscall_return();
1224         return (long)ret;
1225 }
1226
1227 /*
1228  * detach and kill segment if marked destroyed.
1229  * The work is done in shm_close.
1230  */
1231 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1232 {
1233         struct mm_struct *mm = current->mm;
1234         struct vm_area_struct *vma;
1235         unsigned long addr = (unsigned long)shmaddr;
1236         int retval = -EINVAL;
1237 #ifdef CONFIG_MMU
1238         loff_t size = 0;
1239         struct vm_area_struct *next;
1240 #endif
1241
1242         if (addr & ~PAGE_MASK)
1243                 return retval;
1244
1245         down_write(&mm->mmap_sem);
1246
1247         /*
1248          * This function tries to be smart and unmap shm segments that
1249          * were modified by partial mlock or munmap calls:
1250          * - It first determines the size of the shm segment that should be
1251          *   unmapped: It searches for a vma that is backed by shm and that
1252          *   started at address shmaddr. It records it's size and then unmaps
1253          *   it.
1254          * - Then it unmaps all shm vmas that started at shmaddr and that
1255          *   are within the initially determined size.
1256          * Errors from do_munmap are ignored: the function only fails if
1257          * it's called with invalid parameters or if it's called to unmap
1258          * a part of a vma. Both calls in this function are for full vmas,
1259          * the parameters are directly copied from the vma itself and always
1260          * valid - therefore do_munmap cannot fail. (famous last words?)
1261          */
1262         /*
1263          * If it had been mremap()'d, the starting address would not
1264          * match the usual checks anyway. So assume all vma's are
1265          * above the starting address given.
1266          */
1267         vma = find_vma(mm, addr);
1268
1269 #ifdef CONFIG_MMU
1270         while (vma) {
1271                 next = vma->vm_next;
1272
1273                 /*
1274                  * Check if the starting address would match, i.e. it's
1275                  * a fragment created by mprotect() and/or munmap(), or it
1276                  * otherwise it starts at this address with no hassles.
1277                  */
1278                 if ((vma->vm_ops == &shm_vm_ops) &&
1279                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1280
1281
1282                         size = file_inode(vma->vm_file)->i_size;
1283                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1284                         /*
1285                          * We discovered the size of the shm segment, so
1286                          * break out of here and fall through to the next
1287                          * loop that uses the size information to stop
1288                          * searching for matching vma's.
1289                          */
1290                         retval = 0;
1291                         vma = next;
1292                         break;
1293                 }
1294                 vma = next;
1295         }
1296
1297         /*
1298          * We need look no further than the maximum address a fragment
1299          * could possibly have landed at. Also cast things to loff_t to
1300          * prevent overflows and make comparisons vs. equal-width types.
1301          */
1302         size = PAGE_ALIGN(size);
1303         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1304                 next = vma->vm_next;
1305
1306                 /* finding a matching vma now does not alter retval */
1307                 if ((vma->vm_ops == &shm_vm_ops) &&
1308                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1309
1310                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1311                 vma = next;
1312         }
1313
1314 #else /* CONFIG_MMU */
1315         /* under NOMMU conditions, the exact address to be destroyed must be
1316          * given */
1317         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1318                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1319                 retval = 0;
1320         }
1321
1322 #endif
1323
1324         up_write(&mm->mmap_sem);
1325         return retval;
1326 }
1327
1328 #ifdef CONFIG_PROC_FS
1329 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1330 {
1331         struct user_namespace *user_ns = seq_user_ns(s);
1332         struct shmid_kernel *shp = it;
1333         unsigned long rss = 0, swp = 0;
1334
1335         shm_add_rss_swap(shp, &rss, &swp);
1336
1337 #if BITS_PER_LONG <= 32
1338 #define SIZE_SPEC "%10lu"
1339 #else
1340 #define SIZE_SPEC "%21lu"
1341 #endif
1342
1343         return seq_printf(s,
1344                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1345                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1346                           SIZE_SPEC " " SIZE_SPEC "\n",
1347                           shp->shm_perm.key,
1348                           shp->shm_perm.id,
1349                           shp->shm_perm.mode,
1350                           shp->shm_segsz,
1351                           shp->shm_cprid,
1352                           shp->shm_lprid,
1353                           shp->shm_nattch,
1354                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1355                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1356                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1357                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1358                           shp->shm_atim,
1359                           shp->shm_dtim,
1360                           shp->shm_ctim,
1361                           rss * PAGE_SIZE,
1362                           swp * PAGE_SIZE);
1363 }
1364 #endif