KVM: x86: Cleanup of rflags.rf cleaning
[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         ipc_rmid(&shm_ids(ns), &s->shm_perm);
182 }
183
184
185 /* This is called by fork, once for every shm attach. */
186 static void shm_open(struct vm_area_struct *vma)
187 {
188         struct file *file = vma->vm_file;
189         struct shm_file_data *sfd = shm_file_data(file);
190         struct shmid_kernel *shp;
191
192         shp = shm_lock(sfd->ns, sfd->id);
193         BUG_ON(IS_ERR(shp));
194         shp->shm_atim = get_seconds();
195         shp->shm_lprid = task_tgid_vnr(current);
196         shp->shm_nattch++;
197         shm_unlock(shp);
198 }
199
200 /*
201  * shm_destroy - free the struct shmid_kernel
202  *
203  * @ns: namespace
204  * @shp: struct to free
205  *
206  * It has to be called with shp and shm_ids.rwsem (writer) locked,
207  * but returns with shp unlocked and freed.
208  */
209 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
210 {
211         struct file *shm_file;
212
213         shm_file = shp->shm_file;
214         shp->shm_file = NULL;
215         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
216         shm_rmid(ns, shp);
217         shm_unlock(shp);
218         if (!is_file_hugepages(shm_file))
219                 shmem_lock(shm_file, 0, shp->mlock_user);
220         else if (shp->mlock_user)
221                 user_shm_unlock(file_inode(shm_file)->i_size, shp->mlock_user);
222         fput(shm_file);
223         ipc_rcu_putref(shp, shm_rcu_free);
224 }
225
226 /*
227  * shm_may_destroy - identifies whether shm segment should be destroyed now
228  *
229  * Returns true if and only if there are no active users of the segment and
230  * one of the following is true:
231  *
232  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
233  *
234  * 2) sysctl kernel.shm_rmid_forced is set to 1.
235  */
236 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
237 {
238         return (shp->shm_nattch == 0) &&
239                (ns->shm_rmid_forced ||
240                 (shp->shm_perm.mode & SHM_DEST));
241 }
242
243 /*
244  * remove the attach descriptor vma.
245  * free memory for segment if it is marked destroyed.
246  * The descriptor has already been removed from the current->mm->mmap list
247  * and will later be kfree()d.
248  */
249 static void shm_close(struct vm_area_struct *vma)
250 {
251         struct file *file = vma->vm_file;
252         struct shm_file_data *sfd = shm_file_data(file);
253         struct shmid_kernel *shp;
254         struct ipc_namespace *ns = sfd->ns;
255
256         down_write(&shm_ids(ns).rwsem);
257         /* remove from the list of attaches of the shm segment */
258         shp = shm_lock(ns, sfd->id);
259         BUG_ON(IS_ERR(shp));
260         shp->shm_lprid = task_tgid_vnr(current);
261         shp->shm_dtim = get_seconds();
262         shp->shm_nattch--;
263         if (shm_may_destroy(ns, shp))
264                 shm_destroy(ns, shp);
265         else
266                 shm_unlock(shp);
267         up_write(&shm_ids(ns).rwsem);
268 }
269
270 /* Called with ns->shm_ids(ns).rwsem locked */
271 static int shm_try_destroy_current(int id, void *p, void *data)
272 {
273         struct ipc_namespace *ns = data;
274         struct kern_ipc_perm *ipcp = p;
275         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
276
277         if (shp->shm_creator != current)
278                 return 0;
279
280         /*
281          * Mark it as orphaned to destroy the segment when
282          * kernel.shm_rmid_forced is changed.
283          * It is noop if the following shm_may_destroy() returns true.
284          */
285         shp->shm_creator = NULL;
286
287         /*
288          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
289          * is not set, it shouldn't be deleted here.
290          */
291         if (!ns->shm_rmid_forced)
292                 return 0;
293
294         if (shm_may_destroy(ns, shp)) {
295                 shm_lock_by_ptr(shp);
296                 shm_destroy(ns, shp);
297         }
298         return 0;
299 }
300
301 /* Called with ns->shm_ids(ns).rwsem locked */
302 static int shm_try_destroy_orphaned(int id, void *p, void *data)
303 {
304         struct ipc_namespace *ns = data;
305         struct kern_ipc_perm *ipcp = p;
306         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
307
308         /*
309          * We want to destroy segments without users and with already
310          * exit'ed originating process.
311          *
312          * As shp->* are changed under rwsem, it's safe to skip shp locking.
313          */
314         if (shp->shm_creator != NULL)
315                 return 0;
316
317         if (shm_may_destroy(ns, shp)) {
318                 shm_lock_by_ptr(shp);
319                 shm_destroy(ns, shp);
320         }
321         return 0;
322 }
323
324 void shm_destroy_orphaned(struct ipc_namespace *ns)
325 {
326         down_write(&shm_ids(ns).rwsem);
327         if (shm_ids(ns).in_use)
328                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
329         up_write(&shm_ids(ns).rwsem);
330 }
331
332
333 void exit_shm(struct task_struct *task)
334 {
335         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
336
337         if (shm_ids(ns).in_use == 0)
338                 return;
339
340         /* Destroy all already created segments, but not mapped yet */
341         down_write(&shm_ids(ns).rwsem);
342         if (shm_ids(ns).in_use)
343                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
344         up_write(&shm_ids(ns).rwsem);
345 }
346
347 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
348 {
349         struct file *file = vma->vm_file;
350         struct shm_file_data *sfd = shm_file_data(file);
351
352         return sfd->vm_ops->fault(vma, vmf);
353 }
354
355 #ifdef CONFIG_NUMA
356 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
357 {
358         struct file *file = vma->vm_file;
359         struct shm_file_data *sfd = shm_file_data(file);
360         int err = 0;
361         if (sfd->vm_ops->set_policy)
362                 err = sfd->vm_ops->set_policy(vma, new);
363         return err;
364 }
365
366 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
367                                         unsigned long addr)
368 {
369         struct file *file = vma->vm_file;
370         struct shm_file_data *sfd = shm_file_data(file);
371         struct mempolicy *pol = NULL;
372
373         if (sfd->vm_ops->get_policy)
374                 pol = sfd->vm_ops->get_policy(vma, addr);
375         else if (vma->vm_policy)
376                 pol = vma->vm_policy;
377
378         return pol;
379 }
380 #endif
381
382 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
383 {
384         struct shm_file_data *sfd = shm_file_data(file);
385         int ret;
386
387         ret = sfd->file->f_op->mmap(sfd->file, vma);
388         if (ret != 0)
389                 return ret;
390         sfd->vm_ops = vma->vm_ops;
391 #ifdef CONFIG_MMU
392         BUG_ON(!sfd->vm_ops->fault);
393 #endif
394         vma->vm_ops = &shm_vm_ops;
395         shm_open(vma);
396
397         return ret;
398 }
399
400 static int shm_release(struct inode *ino, struct file *file)
401 {
402         struct shm_file_data *sfd = shm_file_data(file);
403
404         put_ipc_ns(sfd->ns);
405         shm_file_data(file) = NULL;
406         kfree(sfd);
407         return 0;
408 }
409
410 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
411 {
412         struct shm_file_data *sfd = shm_file_data(file);
413
414         if (!sfd->file->f_op->fsync)
415                 return -EINVAL;
416         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
417 }
418
419 static long shm_fallocate(struct file *file, int mode, loff_t offset,
420                           loff_t len)
421 {
422         struct shm_file_data *sfd = shm_file_data(file);
423
424         if (!sfd->file->f_op->fallocate)
425                 return -EOPNOTSUPP;
426         return sfd->file->f_op->fallocate(file, mode, offset, len);
427 }
428
429 static unsigned long shm_get_unmapped_area(struct file *file,
430         unsigned long addr, unsigned long len, unsigned long pgoff,
431         unsigned long flags)
432 {
433         struct shm_file_data *sfd = shm_file_data(file);
434         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
435                                                 pgoff, flags);
436 }
437
438 static const struct file_operations shm_file_operations = {
439         .mmap           = shm_mmap,
440         .fsync          = shm_fsync,
441         .release        = shm_release,
442 #ifndef CONFIG_MMU
443         .get_unmapped_area      = shm_get_unmapped_area,
444 #endif
445         .llseek         = noop_llseek,
446         .fallocate      = shm_fallocate,
447 };
448
449 static const struct file_operations shm_file_operations_huge = {
450         .mmap           = shm_mmap,
451         .fsync          = shm_fsync,
452         .release        = shm_release,
453         .get_unmapped_area      = shm_get_unmapped_area,
454         .llseek         = noop_llseek,
455         .fallocate      = shm_fallocate,
456 };
457
458 int is_file_shm_hugepages(struct file *file)
459 {
460         return file->f_op == &shm_file_operations_huge;
461 }
462
463 static const struct vm_operations_struct shm_vm_ops = {
464         .open   = shm_open,     /* callback for a new vm-area open */
465         .close  = shm_close,    /* callback for when the vm-area is released */
466         .fault  = shm_fault,
467 #if defined(CONFIG_NUMA)
468         .set_policy = shm_set_policy,
469         .get_policy = shm_get_policy,
470 #endif
471 };
472
473 /**
474  * newseg - Create a new shared memory segment
475  * @ns: namespace
476  * @params: ptr to the structure that contains key, size and shmflg
477  *
478  * Called with shm_ids.rwsem held as a writer.
479  */
480 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
481 {
482         key_t key = params->key;
483         int shmflg = params->flg;
484         size_t size = params->u.size;
485         int error;
486         struct shmid_kernel *shp;
487         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
488         struct file *file;
489         char name[13];
490         int id;
491         vm_flags_t acctflag = 0;
492
493         if (size < SHMMIN || size > ns->shm_ctlmax)
494                 return -EINVAL;
495
496         if (numpages << PAGE_SHIFT < size)
497                 return -ENOSPC;
498
499         if (ns->shm_tot + numpages < ns->shm_tot ||
500                         ns->shm_tot + numpages > ns->shm_ctlall)
501                 return -ENOSPC;
502
503         shp = ipc_rcu_alloc(sizeof(*shp));
504         if (!shp)
505                 return -ENOMEM;
506
507         shp->shm_perm.key = key;
508         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
509         shp->mlock_user = NULL;
510
511         shp->shm_perm.security = NULL;
512         error = security_shm_alloc(shp);
513         if (error) {
514                 ipc_rcu_putref(shp, ipc_rcu_free);
515                 return error;
516         }
517
518         sprintf(name, "SYSV%08x", key);
519         if (shmflg & SHM_HUGETLB) {
520                 struct hstate *hs;
521                 size_t hugesize;
522
523                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
524                 if (!hs) {
525                         error = -EINVAL;
526                         goto no_file;
527                 }
528                 hugesize = ALIGN(size, huge_page_size(hs));
529
530                 /* hugetlb_file_setup applies strict accounting */
531                 if (shmflg & SHM_NORESERVE)
532                         acctflag = VM_NORESERVE;
533                 file = hugetlb_file_setup(name, hugesize, acctflag,
534                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
535                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
536         } else {
537                 /*
538                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
539                  * if it's asked for.
540                  */
541                 if  ((shmflg & SHM_NORESERVE) &&
542                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
543                         acctflag = VM_NORESERVE;
544                 file = shmem_file_setup(name, size, acctflag);
545         }
546         error = PTR_ERR(file);
547         if (IS_ERR(file))
548                 goto no_file;
549
550         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
551         if (id < 0) {
552                 error = id;
553                 goto no_id;
554         }
555
556         shp->shm_cprid = task_tgid_vnr(current);
557         shp->shm_lprid = 0;
558         shp->shm_atim = shp->shm_dtim = 0;
559         shp->shm_ctim = get_seconds();
560         shp->shm_segsz = size;
561         shp->shm_nattch = 0;
562         shp->shm_file = file;
563         shp->shm_creator = current;
564
565         /*
566          * shmid gets reported as "inode#" in /proc/pid/maps.
567          * proc-ps tools use this. Changing this will break them.
568          */
569         file_inode(file)->i_ino = shp->shm_perm.id;
570
571         ns->shm_tot += numpages;
572         error = shp->shm_perm.id;
573
574         ipc_unlock_object(&shp->shm_perm);
575         rcu_read_unlock();
576         return error;
577
578 no_id:
579         if (is_file_hugepages(file) && shp->mlock_user)
580                 user_shm_unlock(size, shp->mlock_user);
581         fput(file);
582 no_file:
583         ipc_rcu_putref(shp, shm_rcu_free);
584         return error;
585 }
586
587 /*
588  * Called with shm_ids.rwsem and ipcp locked.
589  */
590 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
591 {
592         struct shmid_kernel *shp;
593
594         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
595         return security_shm_associate(shp, shmflg);
596 }
597
598 /*
599  * Called with shm_ids.rwsem and ipcp locked.
600  */
601 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
602                                 struct ipc_params *params)
603 {
604         struct shmid_kernel *shp;
605
606         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
607         if (shp->shm_segsz < params->u.size)
608                 return -EINVAL;
609
610         return 0;
611 }
612
613 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
614 {
615         struct ipc_namespace *ns;
616         static const struct ipc_ops shm_ops = {
617                 .getnew = newseg,
618                 .associate = shm_security,
619                 .more_checks = shm_more_checks,
620         };
621         struct ipc_params shm_params;
622
623         ns = current->nsproxy->ipc_ns;
624
625         shm_params.key = key;
626         shm_params.flg = shmflg;
627         shm_params.u.size = size;
628
629         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
630 }
631
632 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
633 {
634         switch (version) {
635         case IPC_64:
636                 return copy_to_user(buf, in, sizeof(*in));
637         case IPC_OLD:
638             {
639                 struct shmid_ds out;
640
641                 memset(&out, 0, sizeof(out));
642                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
643                 out.shm_segsz   = in->shm_segsz;
644                 out.shm_atime   = in->shm_atime;
645                 out.shm_dtime   = in->shm_dtime;
646                 out.shm_ctime   = in->shm_ctime;
647                 out.shm_cpid    = in->shm_cpid;
648                 out.shm_lpid    = in->shm_lpid;
649                 out.shm_nattch  = in->shm_nattch;
650
651                 return copy_to_user(buf, &out, sizeof(out));
652             }
653         default:
654                 return -EINVAL;
655         }
656 }
657
658 static inline unsigned long
659 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
660 {
661         switch (version) {
662         case IPC_64:
663                 if (copy_from_user(out, buf, sizeof(*out)))
664                         return -EFAULT;
665                 return 0;
666         case IPC_OLD:
667             {
668                 struct shmid_ds tbuf_old;
669
670                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
671                         return -EFAULT;
672
673                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
674                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
675                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
676
677                 return 0;
678             }
679         default:
680                 return -EINVAL;
681         }
682 }
683
684 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
685 {
686         switch (version) {
687         case IPC_64:
688                 return copy_to_user(buf, in, sizeof(*in));
689         case IPC_OLD:
690             {
691                 struct shminfo out;
692
693                 if (in->shmmax > INT_MAX)
694                         out.shmmax = INT_MAX;
695                 else
696                         out.shmmax = (int)in->shmmax;
697
698                 out.shmmin      = in->shmmin;
699                 out.shmmni      = in->shmmni;
700                 out.shmseg      = in->shmseg;
701                 out.shmall      = in->shmall;
702
703                 return copy_to_user(buf, &out, sizeof(out));
704             }
705         default:
706                 return -EINVAL;
707         }
708 }
709
710 /*
711  * Calculate and add used RSS and swap pages of a shm.
712  * Called with shm_ids.rwsem held as a reader
713  */
714 static void shm_add_rss_swap(struct shmid_kernel *shp,
715         unsigned long *rss_add, unsigned long *swp_add)
716 {
717         struct inode *inode;
718
719         inode = file_inode(shp->shm_file);
720
721         if (is_file_hugepages(shp->shm_file)) {
722                 struct address_space *mapping = inode->i_mapping;
723                 struct hstate *h = hstate_file(shp->shm_file);
724                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
725         } else {
726 #ifdef CONFIG_SHMEM
727                 struct shmem_inode_info *info = SHMEM_I(inode);
728                 spin_lock(&info->lock);
729                 *rss_add += inode->i_mapping->nrpages;
730                 *swp_add += info->swapped;
731                 spin_unlock(&info->lock);
732 #else
733                 *rss_add += inode->i_mapping->nrpages;
734 #endif
735         }
736 }
737
738 /*
739  * Called with shm_ids.rwsem held as a reader
740  */
741 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
742                 unsigned long *swp)
743 {
744         int next_id;
745         int total, in_use;
746
747         *rss = 0;
748         *swp = 0;
749
750         in_use = shm_ids(ns).in_use;
751
752         for (total = 0, next_id = 0; total < in_use; next_id++) {
753                 struct kern_ipc_perm *ipc;
754                 struct shmid_kernel *shp;
755
756                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
757                 if (ipc == NULL)
758                         continue;
759                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
760
761                 shm_add_rss_swap(shp, rss, swp);
762
763                 total++;
764         }
765 }
766
767 /*
768  * This function handles some shmctl commands which require the rwsem
769  * to be held in write mode.
770  * NOTE: no locks must be held, the rwsem is taken inside this function.
771  */
772 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
773                        struct shmid_ds __user *buf, int version)
774 {
775         struct kern_ipc_perm *ipcp;
776         struct shmid64_ds shmid64;
777         struct shmid_kernel *shp;
778         int err;
779
780         if (cmd == IPC_SET) {
781                 if (copy_shmid_from_user(&shmid64, buf, version))
782                         return -EFAULT;
783         }
784
785         down_write(&shm_ids(ns).rwsem);
786         rcu_read_lock();
787
788         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
789                                       &shmid64.shm_perm, 0);
790         if (IS_ERR(ipcp)) {
791                 err = PTR_ERR(ipcp);
792                 goto out_unlock1;
793         }
794
795         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
796
797         err = security_shm_shmctl(shp, cmd);
798         if (err)
799                 goto out_unlock1;
800
801         switch (cmd) {
802         case IPC_RMID:
803                 ipc_lock_object(&shp->shm_perm);
804                 /* do_shm_rmid unlocks the ipc object and rcu */
805                 do_shm_rmid(ns, ipcp);
806                 goto out_up;
807         case IPC_SET:
808                 ipc_lock_object(&shp->shm_perm);
809                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
810                 if (err)
811                         goto out_unlock0;
812                 shp->shm_ctim = get_seconds();
813                 break;
814         default:
815                 err = -EINVAL;
816                 goto out_unlock1;
817         }
818
819 out_unlock0:
820         ipc_unlock_object(&shp->shm_perm);
821 out_unlock1:
822         rcu_read_unlock();
823 out_up:
824         up_write(&shm_ids(ns).rwsem);
825         return err;
826 }
827
828 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
829                          int cmd, int version, void __user *buf)
830 {
831         int err;
832         struct shmid_kernel *shp;
833
834         /* preliminary security checks for *_INFO */
835         if (cmd == IPC_INFO || cmd == SHM_INFO) {
836                 err = security_shm_shmctl(NULL, cmd);
837                 if (err)
838                         return err;
839         }
840
841         switch (cmd) {
842         case IPC_INFO:
843         {
844                 struct shminfo64 shminfo;
845
846                 memset(&shminfo, 0, sizeof(shminfo));
847                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
848                 shminfo.shmmax = ns->shm_ctlmax;
849                 shminfo.shmall = ns->shm_ctlall;
850
851                 shminfo.shmmin = SHMMIN;
852                 if (copy_shminfo_to_user(buf, &shminfo, version))
853                         return -EFAULT;
854
855                 down_read(&shm_ids(ns).rwsem);
856                 err = ipc_get_maxid(&shm_ids(ns));
857                 up_read(&shm_ids(ns).rwsem);
858
859                 if (err < 0)
860                         err = 0;
861                 goto out;
862         }
863         case SHM_INFO:
864         {
865                 struct shm_info shm_info;
866
867                 memset(&shm_info, 0, sizeof(shm_info));
868                 down_read(&shm_ids(ns).rwsem);
869                 shm_info.used_ids = shm_ids(ns).in_use;
870                 shm_get_stat(ns, &shm_info.shm_rss, &shm_info.shm_swp);
871                 shm_info.shm_tot = ns->shm_tot;
872                 shm_info.swap_attempts = 0;
873                 shm_info.swap_successes = 0;
874                 err = ipc_get_maxid(&shm_ids(ns));
875                 up_read(&shm_ids(ns).rwsem);
876                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
877                         err = -EFAULT;
878                         goto out;
879                 }
880
881                 err = err < 0 ? 0 : err;
882                 goto out;
883         }
884         case SHM_STAT:
885         case IPC_STAT:
886         {
887                 struct shmid64_ds tbuf;
888                 int result;
889
890                 rcu_read_lock();
891                 if (cmd == SHM_STAT) {
892                         shp = shm_obtain_object(ns, shmid);
893                         if (IS_ERR(shp)) {
894                                 err = PTR_ERR(shp);
895                                 goto out_unlock;
896                         }
897                         result = shp->shm_perm.id;
898                 } else {
899                         shp = shm_obtain_object_check(ns, shmid);
900                         if (IS_ERR(shp)) {
901                                 err = PTR_ERR(shp);
902                                 goto out_unlock;
903                         }
904                         result = 0;
905                 }
906
907                 err = -EACCES;
908                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
909                         goto out_unlock;
910
911                 err = security_shm_shmctl(shp, cmd);
912                 if (err)
913                         goto out_unlock;
914
915                 memset(&tbuf, 0, sizeof(tbuf));
916                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
917                 tbuf.shm_segsz  = shp->shm_segsz;
918                 tbuf.shm_atime  = shp->shm_atim;
919                 tbuf.shm_dtime  = shp->shm_dtim;
920                 tbuf.shm_ctime  = shp->shm_ctim;
921                 tbuf.shm_cpid   = shp->shm_cprid;
922                 tbuf.shm_lpid   = shp->shm_lprid;
923                 tbuf.shm_nattch = shp->shm_nattch;
924                 rcu_read_unlock();
925
926                 if (copy_shmid_to_user(buf, &tbuf, version))
927                         err = -EFAULT;
928                 else
929                         err = result;
930                 goto out;
931         }
932         default:
933                 return -EINVAL;
934         }
935
936 out_unlock:
937         rcu_read_unlock();
938 out:
939         return err;
940 }
941
942 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
943 {
944         struct shmid_kernel *shp;
945         int err, version;
946         struct ipc_namespace *ns;
947
948         if (cmd < 0 || shmid < 0)
949                 return -EINVAL;
950
951         version = ipc_parse_version(&cmd);
952         ns = current->nsproxy->ipc_ns;
953
954         switch (cmd) {
955         case IPC_INFO:
956         case SHM_INFO:
957         case SHM_STAT:
958         case IPC_STAT:
959                 return shmctl_nolock(ns, shmid, cmd, version, buf);
960         case IPC_RMID:
961         case IPC_SET:
962                 return shmctl_down(ns, shmid, cmd, buf, version);
963         case SHM_LOCK:
964         case SHM_UNLOCK:
965         {
966                 struct file *shm_file;
967
968                 rcu_read_lock();
969                 shp = shm_obtain_object_check(ns, shmid);
970                 if (IS_ERR(shp)) {
971                         err = PTR_ERR(shp);
972                         goto out_unlock1;
973                 }
974
975                 audit_ipc_obj(&(shp->shm_perm));
976                 err = security_shm_shmctl(shp, cmd);
977                 if (err)
978                         goto out_unlock1;
979
980                 ipc_lock_object(&shp->shm_perm);
981
982                 /* check if shm_destroy() is tearing down shp */
983                 if (!ipc_valid_object(&shp->shm_perm)) {
984                         err = -EIDRM;
985                         goto out_unlock0;
986                 }
987
988                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
989                         kuid_t euid = current_euid();
990                         if (!uid_eq(euid, shp->shm_perm.uid) &&
991                             !uid_eq(euid, shp->shm_perm.cuid)) {
992                                 err = -EPERM;
993                                 goto out_unlock0;
994                         }
995                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
996                                 err = -EPERM;
997                                 goto out_unlock0;
998                         }
999                 }
1000
1001                 shm_file = shp->shm_file;
1002                 if (is_file_hugepages(shm_file))
1003                         goto out_unlock0;
1004
1005                 if (cmd == SHM_LOCK) {
1006                         struct user_struct *user = current_user();
1007                         err = shmem_lock(shm_file, 1, user);
1008                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1009                                 shp->shm_perm.mode |= SHM_LOCKED;
1010                                 shp->mlock_user = user;
1011                         }
1012                         goto out_unlock0;
1013                 }
1014
1015                 /* SHM_UNLOCK */
1016                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1017                         goto out_unlock0;
1018                 shmem_lock(shm_file, 0, shp->mlock_user);
1019                 shp->shm_perm.mode &= ~SHM_LOCKED;
1020                 shp->mlock_user = NULL;
1021                 get_file(shm_file);
1022                 ipc_unlock_object(&shp->shm_perm);
1023                 rcu_read_unlock();
1024                 shmem_unlock_mapping(shm_file->f_mapping);
1025
1026                 fput(shm_file);
1027                 return err;
1028         }
1029         default:
1030                 return -EINVAL;
1031         }
1032
1033 out_unlock0:
1034         ipc_unlock_object(&shp->shm_perm);
1035 out_unlock1:
1036         rcu_read_unlock();
1037         return err;
1038 }
1039
1040 /*
1041  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1042  *
1043  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1044  * "raddr" thing points to kernel space, and there has to be a wrapper around
1045  * this.
1046  */
1047 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1048               unsigned long shmlba)
1049 {
1050         struct shmid_kernel *shp;
1051         unsigned long addr;
1052         unsigned long size;
1053         struct file *file;
1054         int    err;
1055         unsigned long flags;
1056         unsigned long prot;
1057         int acc_mode;
1058         struct ipc_namespace *ns;
1059         struct shm_file_data *sfd;
1060         struct path path;
1061         fmode_t f_mode;
1062         unsigned long populate = 0;
1063
1064         err = -EINVAL;
1065         if (shmid < 0)
1066                 goto out;
1067         else if ((addr = (ulong)shmaddr)) {
1068                 if (addr & (shmlba - 1)) {
1069                         if (shmflg & SHM_RND)
1070                                 addr &= ~(shmlba - 1);     /* round down */
1071                         else
1072 #ifndef __ARCH_FORCE_SHMLBA
1073                                 if (addr & ~PAGE_MASK)
1074 #endif
1075                                         goto out;
1076                 }
1077                 flags = MAP_SHARED | MAP_FIXED;
1078         } else {
1079                 if ((shmflg & SHM_REMAP))
1080                         goto out;
1081
1082                 flags = MAP_SHARED;
1083         }
1084
1085         if (shmflg & SHM_RDONLY) {
1086                 prot = PROT_READ;
1087                 acc_mode = S_IRUGO;
1088                 f_mode = FMODE_READ;
1089         } else {
1090                 prot = PROT_READ | PROT_WRITE;
1091                 acc_mode = S_IRUGO | S_IWUGO;
1092                 f_mode = FMODE_READ | FMODE_WRITE;
1093         }
1094         if (shmflg & SHM_EXEC) {
1095                 prot |= PROT_EXEC;
1096                 acc_mode |= S_IXUGO;
1097         }
1098
1099         /*
1100          * We cannot rely on the fs check since SYSV IPC does have an
1101          * additional creator id...
1102          */
1103         ns = current->nsproxy->ipc_ns;
1104         rcu_read_lock();
1105         shp = shm_obtain_object_check(ns, shmid);
1106         if (IS_ERR(shp)) {
1107                 err = PTR_ERR(shp);
1108                 goto out_unlock;
1109         }
1110
1111         err = -EACCES;
1112         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1113                 goto out_unlock;
1114
1115         err = security_shm_shmat(shp, shmaddr, shmflg);
1116         if (err)
1117                 goto out_unlock;
1118
1119         ipc_lock_object(&shp->shm_perm);
1120
1121         /* check if shm_destroy() is tearing down shp */
1122         if (!ipc_valid_object(&shp->shm_perm)) {
1123                 ipc_unlock_object(&shp->shm_perm);
1124                 err = -EIDRM;
1125                 goto out_unlock;
1126         }
1127
1128         path = shp->shm_file->f_path;
1129         path_get(&path);
1130         shp->shm_nattch++;
1131         size = i_size_read(path.dentry->d_inode);
1132         ipc_unlock_object(&shp->shm_perm);
1133         rcu_read_unlock();
1134
1135         err = -ENOMEM;
1136         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1137         if (!sfd) {
1138                 path_put(&path);
1139                 goto out_nattch;
1140         }
1141
1142         file = alloc_file(&path, f_mode,
1143                           is_file_hugepages(shp->shm_file) ?
1144                                 &shm_file_operations_huge :
1145                                 &shm_file_operations);
1146         err = PTR_ERR(file);
1147         if (IS_ERR(file)) {
1148                 kfree(sfd);
1149                 path_put(&path);
1150                 goto out_nattch;
1151         }
1152
1153         file->private_data = sfd;
1154         file->f_mapping = shp->shm_file->f_mapping;
1155         sfd->id = shp->shm_perm.id;
1156         sfd->ns = get_ipc_ns(ns);
1157         sfd->file = shp->shm_file;
1158         sfd->vm_ops = NULL;
1159
1160         err = security_mmap_file(file, prot, flags);
1161         if (err)
1162                 goto out_fput;
1163
1164         down_write(&current->mm->mmap_sem);
1165         if (addr && !(shmflg & SHM_REMAP)) {
1166                 err = -EINVAL;
1167                 if (addr + size < addr)
1168                         goto invalid;
1169
1170                 if (find_vma_intersection(current->mm, addr, addr + size))
1171                         goto invalid;
1172                 /*
1173                  * If shm segment goes below stack, make sure there is some
1174                  * space left for the stack to grow (at least 4 pages).
1175                  */
1176                 if (addr < current->mm->start_stack &&
1177                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1178                         goto invalid;
1179         }
1180
1181         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1182         *raddr = addr;
1183         err = 0;
1184         if (IS_ERR_VALUE(addr))
1185                 err = (long)addr;
1186 invalid:
1187         up_write(&current->mm->mmap_sem);
1188         if (populate)
1189                 mm_populate(addr, populate);
1190
1191 out_fput:
1192         fput(file);
1193
1194 out_nattch:
1195         down_write(&shm_ids(ns).rwsem);
1196         shp = shm_lock(ns, shmid);
1197         BUG_ON(IS_ERR(shp));
1198         shp->shm_nattch--;
1199         if (shm_may_destroy(ns, shp))
1200                 shm_destroy(ns, shp);
1201         else
1202                 shm_unlock(shp);
1203         up_write(&shm_ids(ns).rwsem);
1204         return err;
1205
1206 out_unlock:
1207         rcu_read_unlock();
1208 out:
1209         return err;
1210 }
1211
1212 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1213 {
1214         unsigned long ret;
1215         long err;
1216
1217         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1218         if (err)
1219                 return err;
1220         force_successful_syscall_return();
1221         return (long)ret;
1222 }
1223
1224 /*
1225  * detach and kill segment if marked destroyed.
1226  * The work is done in shm_close.
1227  */
1228 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1229 {
1230         struct mm_struct *mm = current->mm;
1231         struct vm_area_struct *vma;
1232         unsigned long addr = (unsigned long)shmaddr;
1233         int retval = -EINVAL;
1234 #ifdef CONFIG_MMU
1235         loff_t size = 0;
1236         struct vm_area_struct *next;
1237 #endif
1238
1239         if (addr & ~PAGE_MASK)
1240                 return retval;
1241
1242         down_write(&mm->mmap_sem);
1243
1244         /*
1245          * This function tries to be smart and unmap shm segments that
1246          * were modified by partial mlock or munmap calls:
1247          * - It first determines the size of the shm segment that should be
1248          *   unmapped: It searches for a vma that is backed by shm and that
1249          *   started at address shmaddr. It records it's size and then unmaps
1250          *   it.
1251          * - Then it unmaps all shm vmas that started at shmaddr and that
1252          *   are within the initially determined size.
1253          * Errors from do_munmap are ignored: the function only fails if
1254          * it's called with invalid parameters or if it's called to unmap
1255          * a part of a vma. Both calls in this function are for full vmas,
1256          * the parameters are directly copied from the vma itself and always
1257          * valid - therefore do_munmap cannot fail. (famous last words?)
1258          */
1259         /*
1260          * If it had been mremap()'d, the starting address would not
1261          * match the usual checks anyway. So assume all vma's are
1262          * above the starting address given.
1263          */
1264         vma = find_vma(mm, addr);
1265
1266 #ifdef CONFIG_MMU
1267         while (vma) {
1268                 next = vma->vm_next;
1269
1270                 /*
1271                  * Check if the starting address would match, i.e. it's
1272                  * a fragment created by mprotect() and/or munmap(), or it
1273                  * otherwise it starts at this address with no hassles.
1274                  */
1275                 if ((vma->vm_ops == &shm_vm_ops) &&
1276                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1277
1278
1279                         size = file_inode(vma->vm_file)->i_size;
1280                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1281                         /*
1282                          * We discovered the size of the shm segment, so
1283                          * break out of here and fall through to the next
1284                          * loop that uses the size information to stop
1285                          * searching for matching vma's.
1286                          */
1287                         retval = 0;
1288                         vma = next;
1289                         break;
1290                 }
1291                 vma = next;
1292         }
1293
1294         /*
1295          * We need look no further than the maximum address a fragment
1296          * could possibly have landed at. Also cast things to loff_t to
1297          * prevent overflows and make comparisons vs. equal-width types.
1298          */
1299         size = PAGE_ALIGN(size);
1300         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1301                 next = vma->vm_next;
1302
1303                 /* finding a matching vma now does not alter retval */
1304                 if ((vma->vm_ops == &shm_vm_ops) &&
1305                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1306
1307                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1308                 vma = next;
1309         }
1310
1311 #else /* CONFIG_MMU */
1312         /* under NOMMU conditions, the exact address to be destroyed must be
1313          * given */
1314         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1315                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1316                 retval = 0;
1317         }
1318
1319 #endif
1320
1321         up_write(&mm->mmap_sem);
1322         return retval;
1323 }
1324
1325 #ifdef CONFIG_PROC_FS
1326 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1327 {
1328         struct user_namespace *user_ns = seq_user_ns(s);
1329         struct shmid_kernel *shp = it;
1330         unsigned long rss = 0, swp = 0;
1331
1332         shm_add_rss_swap(shp, &rss, &swp);
1333
1334 #if BITS_PER_LONG <= 32
1335 #define SIZE_SPEC "%10lu"
1336 #else
1337 #define SIZE_SPEC "%21lu"
1338 #endif
1339
1340         return seq_printf(s,
1341                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1342                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1343                           SIZE_SPEC " " SIZE_SPEC "\n",
1344                           shp->shm_perm.key,
1345                           shp->shm_perm.id,
1346                           shp->shm_perm.mode,
1347                           shp->shm_segsz,
1348                           shp->shm_cprid,
1349                           shp->shm_lprid,
1350                           shp->shm_nattch,
1351                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1352                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1353                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1354                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1355                           shp->shm_atim,
1356                           shp->shm_dtim,
1357                           shp->shm_ctim,
1358                           rss * PAGE_SIZE,
1359                           swp * PAGE_SIZE);
1360 }
1361 #endif