ipc,shm: shorten critical region for shmat
[platform/adaptation/renesas_rcar/renesas_kernel.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 <asm/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.rw_mutex (writer) and the shp structure locked.
87  * Only shm_ids.rw_mutex 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 rw_mutex
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 inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
171                                                 int id)
172 {
173         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
174
175         if (IS_ERR(ipcp))
176                 return (struct shmid_kernel *)ipcp;
177
178         return container_of(ipcp, struct shmid_kernel, shm_perm);
179 }
180
181 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
182 {
183         ipc_rmid(&shm_ids(ns), &s->shm_perm);
184 }
185
186
187 /* This is called by fork, once for every shm attach. */
188 static void shm_open(struct vm_area_struct *vma)
189 {
190         struct file *file = vma->vm_file;
191         struct shm_file_data *sfd = shm_file_data(file);
192         struct shmid_kernel *shp;
193
194         shp = shm_lock(sfd->ns, sfd->id);
195         BUG_ON(IS_ERR(shp));
196         shp->shm_atim = get_seconds();
197         shp->shm_lprid = task_tgid_vnr(current);
198         shp->shm_nattch++;
199         shm_unlock(shp);
200 }
201
202 /*
203  * shm_destroy - free the struct shmid_kernel
204  *
205  * @ns: namespace
206  * @shp: struct to free
207  *
208  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
209  * but returns with shp unlocked and freed.
210  */
211 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
212 {
213         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
214         shm_rmid(ns, shp);
215         shm_unlock(shp);
216         if (!is_file_hugepages(shp->shm_file))
217                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
218         else if (shp->mlock_user)
219                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
220                                                 shp->mlock_user);
221         fput (shp->shm_file);
222         security_shm_free(shp);
223         ipc_rcu_putref(shp);
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).rw_mutex);
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).rw_mutex);
268 }
269
270 /* Called with ns->shm_ids(ns).rw_mutex 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).rw_mutex 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 rw_mutex, 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).rw_mutex);
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).rw_mutex);
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).rw_mutex);
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).rw_mutex);
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.rw_mutex held as a writer.
479  */
480
481 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
482 {
483         key_t key = params->key;
484         int shmflg = params->flg;
485         size_t size = params->u.size;
486         int error;
487         struct shmid_kernel *shp;
488         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
489         struct file * file;
490         char name[13];
491         int id;
492         vm_flags_t acctflag = 0;
493
494         if (size < SHMMIN || size > ns->shm_ctlmax)
495                 return -EINVAL;
496
497         if (ns->shm_tot + numpages > ns->shm_ctlall)
498                 return -ENOSPC;
499
500         shp = ipc_rcu_alloc(sizeof(*shp));
501         if (!shp)
502                 return -ENOMEM;
503
504         shp->shm_perm.key = key;
505         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
506         shp->mlock_user = NULL;
507
508         shp->shm_perm.security = NULL;
509         error = security_shm_alloc(shp);
510         if (error) {
511                 ipc_rcu_putref(shp);
512                 return error;
513         }
514
515         sprintf (name, "SYSV%08x", key);
516         if (shmflg & SHM_HUGETLB) {
517                 struct hstate *hs;
518                 size_t hugesize;
519
520                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
521                 if (!hs) {
522                         error = -EINVAL;
523                         goto no_file;
524                 }
525                 hugesize = ALIGN(size, huge_page_size(hs));
526
527                 /* hugetlb_file_setup applies strict accounting */
528                 if (shmflg & SHM_NORESERVE)
529                         acctflag = VM_NORESERVE;
530                 file = hugetlb_file_setup(name, hugesize, acctflag,
531                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
532                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
533         } else {
534                 /*
535                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
536                  * if it's asked for.
537                  */
538                 if  ((shmflg & SHM_NORESERVE) &&
539                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
540                         acctflag = VM_NORESERVE;
541                 file = shmem_file_setup(name, size, acctflag);
542         }
543         error = PTR_ERR(file);
544         if (IS_ERR(file))
545                 goto no_file;
546
547         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
548         if (id < 0) {
549                 error = id;
550                 goto no_id;
551         }
552
553         shp->shm_cprid = task_tgid_vnr(current);
554         shp->shm_lprid = 0;
555         shp->shm_atim = shp->shm_dtim = 0;
556         shp->shm_ctim = get_seconds();
557         shp->shm_segsz = size;
558         shp->shm_nattch = 0;
559         shp->shm_file = file;
560         shp->shm_creator = current;
561
562         /*
563          * shmid gets reported as "inode#" in /proc/pid/maps.
564          * proc-ps tools use this. Changing this will break them.
565          */
566         file_inode(file)->i_ino = shp->shm_perm.id;
567
568         ns->shm_tot += numpages;
569         error = shp->shm_perm.id;
570
571         ipc_unlock_object(&shp->shm_perm);
572         rcu_read_unlock();
573         return error;
574
575 no_id:
576         if (is_file_hugepages(file) && shp->mlock_user)
577                 user_shm_unlock(size, shp->mlock_user);
578         fput(file);
579 no_file:
580         security_shm_free(shp);
581         ipc_rcu_putref(shp);
582         return error;
583 }
584
585 /*
586  * Called with shm_ids.rw_mutex and ipcp locked.
587  */
588 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
589 {
590         struct shmid_kernel *shp;
591
592         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
593         return security_shm_associate(shp, shmflg);
594 }
595
596 /*
597  * Called with shm_ids.rw_mutex and ipcp locked.
598  */
599 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
600                                 struct ipc_params *params)
601 {
602         struct shmid_kernel *shp;
603
604         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
605         if (shp->shm_segsz < params->u.size)
606                 return -EINVAL;
607
608         return 0;
609 }
610
611 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
612 {
613         struct ipc_namespace *ns;
614         struct ipc_ops shm_ops;
615         struct ipc_params shm_params;
616
617         ns = current->nsproxy->ipc_ns;
618
619         shm_ops.getnew = newseg;
620         shm_ops.associate = shm_security;
621         shm_ops.more_checks = shm_more_checks;
622
623         shm_params.key = key;
624         shm_params.flg = shmflg;
625         shm_params.u.size = size;
626
627         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
628 }
629
630 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
631 {
632         switch(version) {
633         case IPC_64:
634                 return copy_to_user(buf, in, sizeof(*in));
635         case IPC_OLD:
636             {
637                 struct shmid_ds out;
638
639                 memset(&out, 0, sizeof(out));
640                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
641                 out.shm_segsz   = in->shm_segsz;
642                 out.shm_atime   = in->shm_atime;
643                 out.shm_dtime   = in->shm_dtime;
644                 out.shm_ctime   = in->shm_ctime;
645                 out.shm_cpid    = in->shm_cpid;
646                 out.shm_lpid    = in->shm_lpid;
647                 out.shm_nattch  = in->shm_nattch;
648
649                 return copy_to_user(buf, &out, sizeof(out));
650             }
651         default:
652                 return -EINVAL;
653         }
654 }
655
656 static inline unsigned long
657 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
658 {
659         switch(version) {
660         case IPC_64:
661                 if (copy_from_user(out, buf, sizeof(*out)))
662                         return -EFAULT;
663                 return 0;
664         case IPC_OLD:
665             {
666                 struct shmid_ds tbuf_old;
667
668                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
669                         return -EFAULT;
670
671                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
672                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
673                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
674
675                 return 0;
676             }
677         default:
678                 return -EINVAL;
679         }
680 }
681
682 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
683 {
684         switch(version) {
685         case IPC_64:
686                 return copy_to_user(buf, in, sizeof(*in));
687         case IPC_OLD:
688             {
689                 struct shminfo out;
690
691                 if(in->shmmax > INT_MAX)
692                         out.shmmax = INT_MAX;
693                 else
694                         out.shmmax = (int)in->shmmax;
695
696                 out.shmmin      = in->shmmin;
697                 out.shmmni      = in->shmmni;
698                 out.shmseg      = in->shmseg;
699                 out.shmall      = in->shmall; 
700
701                 return copy_to_user(buf, &out, sizeof(out));
702             }
703         default:
704                 return -EINVAL;
705         }
706 }
707
708 /*
709  * Calculate and add used RSS and swap pages of a shm.
710  * Called with shm_ids.rw_mutex held as a reader
711  */
712 static void shm_add_rss_swap(struct shmid_kernel *shp,
713         unsigned long *rss_add, unsigned long *swp_add)
714 {
715         struct inode *inode;
716
717         inode = file_inode(shp->shm_file);
718
719         if (is_file_hugepages(shp->shm_file)) {
720                 struct address_space *mapping = inode->i_mapping;
721                 struct hstate *h = hstate_file(shp->shm_file);
722                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
723         } else {
724 #ifdef CONFIG_SHMEM
725                 struct shmem_inode_info *info = SHMEM_I(inode);
726                 spin_lock(&info->lock);
727                 *rss_add += inode->i_mapping->nrpages;
728                 *swp_add += info->swapped;
729                 spin_unlock(&info->lock);
730 #else
731                 *rss_add += inode->i_mapping->nrpages;
732 #endif
733         }
734 }
735
736 /*
737  * Called with shm_ids.rw_mutex held as a reader
738  */
739 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
740                 unsigned long *swp)
741 {
742         int next_id;
743         int total, in_use;
744
745         *rss = 0;
746         *swp = 0;
747
748         in_use = shm_ids(ns).in_use;
749
750         for (total = 0, next_id = 0; total < in_use; next_id++) {
751                 struct kern_ipc_perm *ipc;
752                 struct shmid_kernel *shp;
753
754                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
755                 if (ipc == NULL)
756                         continue;
757                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
758
759                 shm_add_rss_swap(shp, rss, swp);
760
761                 total++;
762         }
763 }
764
765 /*
766  * This function handles some shmctl commands which require the rw_mutex
767  * to be held in write mode.
768  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
769  */
770 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
771                        struct shmid_ds __user *buf, int version)
772 {
773         struct kern_ipc_perm *ipcp;
774         struct shmid64_ds shmid64;
775         struct shmid_kernel *shp;
776         int err;
777
778         if (cmd == IPC_SET) {
779                 if (copy_shmid_from_user(&shmid64, buf, version))
780                         return -EFAULT;
781         }
782
783         down_write(&shm_ids(ns).rw_mutex);
784         rcu_read_lock();
785
786         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
787                                       &shmid64.shm_perm, 0);
788         if (IS_ERR(ipcp)) {
789                 err = PTR_ERR(ipcp);
790                 goto out_unlock1;
791         }
792
793         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
794
795         err = security_shm_shmctl(shp, cmd);
796         if (err)
797                 goto out_unlock1;
798
799         switch (cmd) {
800         case IPC_RMID:
801                 ipc_lock_object(&shp->shm_perm);
802                 /* do_shm_rmid unlocks the ipc object and rcu */
803                 do_shm_rmid(ns, ipcp);
804                 goto out_up;
805         case IPC_SET:
806                 ipc_lock_object(&shp->shm_perm);
807                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
808                 if (err)
809                         goto out_unlock0;
810                 shp->shm_ctim = get_seconds();
811                 break;
812         default:
813                 err = -EINVAL;
814                 goto out_unlock1;
815         }
816
817 out_unlock0:
818         ipc_unlock_object(&shp->shm_perm);
819 out_unlock1:
820         rcu_read_unlock();
821 out_up:
822         up_write(&shm_ids(ns).rw_mutex);
823         return err;
824 }
825
826 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
827                          int cmd, int version, void __user *buf)
828 {
829         int err;
830         struct shmid_kernel *shp;
831
832         /* preliminary security checks for *_INFO */
833         if (cmd == IPC_INFO || cmd == SHM_INFO) {
834                 err = security_shm_shmctl(NULL, cmd);
835                 if (err)
836                         return err;
837         }
838
839         switch (cmd) {
840         case IPC_INFO:
841         {
842                 struct shminfo64 shminfo;
843
844                 memset(&shminfo, 0, sizeof(shminfo));
845                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
846                 shminfo.shmmax = ns->shm_ctlmax;
847                 shminfo.shmall = ns->shm_ctlall;
848
849                 shminfo.shmmin = SHMMIN;
850                 if(copy_shminfo_to_user (buf, &shminfo, version))
851                         return -EFAULT;
852
853                 down_read(&shm_ids(ns).rw_mutex);
854                 err = ipc_get_maxid(&shm_ids(ns));
855                 up_read(&shm_ids(ns).rw_mutex);
856
857                 if(err<0)
858                         err = 0;
859                 goto out;
860         }
861         case SHM_INFO:
862         {
863                 struct shm_info shm_info;
864
865                 memset(&shm_info, 0, sizeof(shm_info));
866                 down_read(&shm_ids(ns).rw_mutex);
867                 shm_info.used_ids = shm_ids(ns).in_use;
868                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
869                 shm_info.shm_tot = ns->shm_tot;
870                 shm_info.swap_attempts = 0;
871                 shm_info.swap_successes = 0;
872                 err = ipc_get_maxid(&shm_ids(ns));
873                 up_read(&shm_ids(ns).rw_mutex);
874                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
875                         err = -EFAULT;
876                         goto out;
877                 }
878
879                 err = err < 0 ? 0 : err;
880                 goto out;
881         }
882         case SHM_STAT:
883         case IPC_STAT:
884         {
885                 struct shmid64_ds tbuf;
886                 int result;
887
888                 rcu_read_lock();
889                 if (cmd == SHM_STAT) {
890                         shp = shm_obtain_object(ns, shmid);
891                         if (IS_ERR(shp)) {
892                                 err = PTR_ERR(shp);
893                                 goto out_unlock;
894                         }
895                         result = shp->shm_perm.id;
896                 } else {
897                         shp = shm_obtain_object_check(ns, shmid);
898                         if (IS_ERR(shp)) {
899                                 err = PTR_ERR(shp);
900                                 goto out_unlock;
901                         }
902                         result = 0;
903                 }
904
905                 err = -EACCES;
906                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
907                         goto out_unlock;
908
909                 err = security_shm_shmctl(shp, cmd);
910                 if (err)
911                         goto out_unlock;
912
913                 memset(&tbuf, 0, sizeof(tbuf));
914                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
915                 tbuf.shm_segsz  = shp->shm_segsz;
916                 tbuf.shm_atime  = shp->shm_atim;
917                 tbuf.shm_dtime  = shp->shm_dtim;
918                 tbuf.shm_ctime  = shp->shm_ctim;
919                 tbuf.shm_cpid   = shp->shm_cprid;
920                 tbuf.shm_lpid   = shp->shm_lprid;
921                 tbuf.shm_nattch = shp->shm_nattch;
922                 rcu_read_unlock();
923
924                 if (copy_shmid_to_user(buf, &tbuf, version))
925                         err = -EFAULT;
926                 else
927                         err = result;
928                 goto out;
929         }
930         default:
931                 return -EINVAL;
932         }
933
934 out_unlock:
935         rcu_read_unlock();
936 out:
937         return err;
938 }
939
940 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
941 {
942         struct shmid_kernel *shp;
943         int err, version;
944         struct ipc_namespace *ns;
945
946         if (cmd < 0 || shmid < 0)
947                 return -EINVAL;
948
949         version = ipc_parse_version(&cmd);
950         ns = current->nsproxy->ipc_ns;
951
952         switch (cmd) {
953         case IPC_INFO:
954         case SHM_INFO:
955         case SHM_STAT:
956         case IPC_STAT:
957                 return shmctl_nolock(ns, shmid, cmd, version, buf);
958         case IPC_RMID:
959         case IPC_SET:
960                 return shmctl_down(ns, shmid, cmd, buf, version);
961         case SHM_LOCK:
962         case SHM_UNLOCK:
963         {
964                 struct file *shm_file;
965
966                 rcu_read_lock();
967                 shp = shm_obtain_object_check(ns, shmid);
968                 if (IS_ERR(shp)) {
969                         err = PTR_ERR(shp);
970                         goto out_unlock1;
971                 }
972
973                 audit_ipc_obj(&(shp->shm_perm));
974                 err = security_shm_shmctl(shp, cmd);
975                 if (err)
976                         goto out_unlock1;
977
978                 ipc_lock_object(&shp->shm_perm);
979                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
980                         kuid_t euid = current_euid();
981                         err = -EPERM;
982                         if (!uid_eq(euid, shp->shm_perm.uid) &&
983                             !uid_eq(euid, shp->shm_perm.cuid))
984                                 goto out_unlock0;
985                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
986                                 goto out_unlock0;
987                 }
988
989                 shm_file = shp->shm_file;
990                 if (is_file_hugepages(shm_file))
991                         goto out_unlock0;
992
993                 if (cmd == SHM_LOCK) {
994                         struct user_struct *user = current_user();
995                         err = shmem_lock(shm_file, 1, user);
996                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
997                                 shp->shm_perm.mode |= SHM_LOCKED;
998                                 shp->mlock_user = user;
999                         }
1000                         goto out_unlock0;
1001                 }
1002
1003                 /* SHM_UNLOCK */
1004                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1005                         goto out_unlock0;
1006                 shmem_lock(shm_file, 0, shp->mlock_user);
1007                 shp->shm_perm.mode &= ~SHM_LOCKED;
1008                 shp->mlock_user = NULL;
1009                 get_file(shm_file);
1010                 ipc_unlock_object(&shp->shm_perm);
1011                 rcu_read_unlock();
1012                 shmem_unlock_mapping(shm_file->f_mapping);
1013
1014                 fput(shm_file);
1015                 return err;
1016         }
1017         default:
1018                 return -EINVAL;
1019         }
1020
1021 out_unlock0:
1022         ipc_unlock_object(&shp->shm_perm);
1023 out_unlock1:
1024         rcu_read_unlock();
1025         return err;
1026 }
1027
1028 /*
1029  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1030  *
1031  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1032  * "raddr" thing points to kernel space, and there has to be a wrapper around
1033  * this.
1034  */
1035 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1036               unsigned long shmlba)
1037 {
1038         struct shmid_kernel *shp;
1039         unsigned long addr;
1040         unsigned long size;
1041         struct file * file;
1042         int    err;
1043         unsigned long flags;
1044         unsigned long prot;
1045         int acc_mode;
1046         struct ipc_namespace *ns;
1047         struct shm_file_data *sfd;
1048         struct path path;
1049         fmode_t f_mode;
1050         unsigned long populate = 0;
1051
1052         err = -EINVAL;
1053         if (shmid < 0)
1054                 goto out;
1055         else if ((addr = (ulong)shmaddr)) {
1056                 if (addr & (shmlba - 1)) {
1057                         if (shmflg & SHM_RND)
1058                                 addr &= ~(shmlba - 1);     /* round down */
1059                         else
1060 #ifndef __ARCH_FORCE_SHMLBA
1061                                 if (addr & ~PAGE_MASK)
1062 #endif
1063                                         goto out;
1064                 }
1065                 flags = MAP_SHARED | MAP_FIXED;
1066         } else {
1067                 if ((shmflg & SHM_REMAP))
1068                         goto out;
1069
1070                 flags = MAP_SHARED;
1071         }
1072
1073         if (shmflg & SHM_RDONLY) {
1074                 prot = PROT_READ;
1075                 acc_mode = S_IRUGO;
1076                 f_mode = FMODE_READ;
1077         } else {
1078                 prot = PROT_READ | PROT_WRITE;
1079                 acc_mode = S_IRUGO | S_IWUGO;
1080                 f_mode = FMODE_READ | FMODE_WRITE;
1081         }
1082         if (shmflg & SHM_EXEC) {
1083                 prot |= PROT_EXEC;
1084                 acc_mode |= S_IXUGO;
1085         }
1086
1087         /*
1088          * We cannot rely on the fs check since SYSV IPC does have an
1089          * additional creator id...
1090          */
1091         ns = current->nsproxy->ipc_ns;
1092         rcu_read_lock();
1093         shp = shm_obtain_object_check(ns, shmid);
1094         if (IS_ERR(shp)) {
1095                 err = PTR_ERR(shp);
1096                 goto out_unlock;
1097         }
1098
1099         err = -EACCES;
1100         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1101                 goto out_unlock;
1102
1103         err = security_shm_shmat(shp, shmaddr, shmflg);
1104         if (err)
1105                 goto out_unlock;
1106
1107         ipc_lock_object(&shp->shm_perm);
1108         path = shp->shm_file->f_path;
1109         path_get(&path);
1110         shp->shm_nattch++;
1111         size = i_size_read(path.dentry->d_inode);
1112         ipc_unlock_object(&shp->shm_perm);
1113         rcu_read_unlock();
1114
1115         err = -ENOMEM;
1116         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1117         if (!sfd) {
1118                 path_put(&path);
1119                 goto out_nattch;
1120         }
1121
1122         file = alloc_file(&path, f_mode,
1123                           is_file_hugepages(shp->shm_file) ?
1124                                 &shm_file_operations_huge :
1125                                 &shm_file_operations);
1126         err = PTR_ERR(file);
1127         if (IS_ERR(file)) {
1128                 kfree(sfd);
1129                 path_put(&path);
1130                 goto out_nattch;
1131         }
1132
1133         file->private_data = sfd;
1134         file->f_mapping = shp->shm_file->f_mapping;
1135         sfd->id = shp->shm_perm.id;
1136         sfd->ns = get_ipc_ns(ns);
1137         sfd->file = shp->shm_file;
1138         sfd->vm_ops = NULL;
1139
1140         err = security_mmap_file(file, prot, flags);
1141         if (err)
1142                 goto out_fput;
1143
1144         down_write(&current->mm->mmap_sem);
1145         if (addr && !(shmflg & SHM_REMAP)) {
1146                 err = -EINVAL;
1147                 if (find_vma_intersection(current->mm, addr, addr + size))
1148                         goto invalid;
1149                 /*
1150                  * If shm segment goes below stack, make sure there is some
1151                  * space left for the stack to grow (at least 4 pages).
1152                  */
1153                 if (addr < current->mm->start_stack &&
1154                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1155                         goto invalid;
1156         }
1157
1158         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1159         *raddr = addr;
1160         err = 0;
1161         if (IS_ERR_VALUE(addr))
1162                 err = (long)addr;
1163 invalid:
1164         up_write(&current->mm->mmap_sem);
1165         if (populate)
1166                 mm_populate(addr, populate);
1167
1168 out_fput:
1169         fput(file);
1170
1171 out_nattch:
1172         down_write(&shm_ids(ns).rw_mutex);
1173         shp = shm_lock(ns, shmid);
1174         BUG_ON(IS_ERR(shp));
1175         shp->shm_nattch--;
1176         if (shm_may_destroy(ns, shp))
1177                 shm_destroy(ns, shp);
1178         else
1179                 shm_unlock(shp);
1180         up_write(&shm_ids(ns).rw_mutex);
1181         return err;
1182
1183 out_unlock:
1184         rcu_read_unlock();
1185 out:
1186         return err;
1187 }
1188
1189 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1190 {
1191         unsigned long ret;
1192         long err;
1193
1194         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1195         if (err)
1196                 return err;
1197         force_successful_syscall_return();
1198         return (long)ret;
1199 }
1200
1201 /*
1202  * detach and kill segment if marked destroyed.
1203  * The work is done in shm_close.
1204  */
1205 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1206 {
1207         struct mm_struct *mm = current->mm;
1208         struct vm_area_struct *vma;
1209         unsigned long addr = (unsigned long)shmaddr;
1210         int retval = -EINVAL;
1211 #ifdef CONFIG_MMU
1212         loff_t size = 0;
1213         struct vm_area_struct *next;
1214 #endif
1215
1216         if (addr & ~PAGE_MASK)
1217                 return retval;
1218
1219         down_write(&mm->mmap_sem);
1220
1221         /*
1222          * This function tries to be smart and unmap shm segments that
1223          * were modified by partial mlock or munmap calls:
1224          * - It first determines the size of the shm segment that should be
1225          *   unmapped: It searches for a vma that is backed by shm and that
1226          *   started at address shmaddr. It records it's size and then unmaps
1227          *   it.
1228          * - Then it unmaps all shm vmas that started at shmaddr and that
1229          *   are within the initially determined size.
1230          * Errors from do_munmap are ignored: the function only fails if
1231          * it's called with invalid parameters or if it's called to unmap
1232          * a part of a vma. Both calls in this function are for full vmas,
1233          * the parameters are directly copied from the vma itself and always
1234          * valid - therefore do_munmap cannot fail. (famous last words?)
1235          */
1236         /*
1237          * If it had been mremap()'d, the starting address would not
1238          * match the usual checks anyway. So assume all vma's are
1239          * above the starting address given.
1240          */
1241         vma = find_vma(mm, addr);
1242
1243 #ifdef CONFIG_MMU
1244         while (vma) {
1245                 next = vma->vm_next;
1246
1247                 /*
1248                  * Check if the starting address would match, i.e. it's
1249                  * a fragment created by mprotect() and/or munmap(), or it
1250                  * otherwise it starts at this address with no hassles.
1251                  */
1252                 if ((vma->vm_ops == &shm_vm_ops) &&
1253                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1254
1255
1256                         size = file_inode(vma->vm_file)->i_size;
1257                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1258                         /*
1259                          * We discovered the size of the shm segment, so
1260                          * break out of here and fall through to the next
1261                          * loop that uses the size information to stop
1262                          * searching for matching vma's.
1263                          */
1264                         retval = 0;
1265                         vma = next;
1266                         break;
1267                 }
1268                 vma = next;
1269         }
1270
1271         /*
1272          * We need look no further than the maximum address a fragment
1273          * could possibly have landed at. Also cast things to loff_t to
1274          * prevent overflows and make comparisons vs. equal-width types.
1275          */
1276         size = PAGE_ALIGN(size);
1277         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1278                 next = vma->vm_next;
1279
1280                 /* finding a matching vma now does not alter retval */
1281                 if ((vma->vm_ops == &shm_vm_ops) &&
1282                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1283
1284                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1285                 vma = next;
1286         }
1287
1288 #else /* CONFIG_MMU */
1289         /* under NOMMU conditions, the exact address to be destroyed must be
1290          * given */
1291         retval = -EINVAL;
1292         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1293                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1294                 retval = 0;
1295         }
1296
1297 #endif
1298
1299         up_write(&mm->mmap_sem);
1300         return retval;
1301 }
1302
1303 #ifdef CONFIG_PROC_FS
1304 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1305 {
1306         struct user_namespace *user_ns = seq_user_ns(s);
1307         struct shmid_kernel *shp = it;
1308         unsigned long rss = 0, swp = 0;
1309
1310         shm_add_rss_swap(shp, &rss, &swp);
1311
1312 #if BITS_PER_LONG <= 32
1313 #define SIZE_SPEC "%10lu"
1314 #else
1315 #define SIZE_SPEC "%21lu"
1316 #endif
1317
1318         return seq_printf(s,
1319                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1320                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1321                           SIZE_SPEC " " SIZE_SPEC "\n",
1322                           shp->shm_perm.key,
1323                           shp->shm_perm.id,
1324                           shp->shm_perm.mode,
1325                           shp->shm_segsz,
1326                           shp->shm_cprid,
1327                           shp->shm_lprid,
1328                           shp->shm_nattch,
1329                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1330                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1331                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1332                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1333                           shp->shm_atim,
1334                           shp->shm_dtim,
1335                           shp->shm_ctim,
1336                           rss * PAGE_SIZE,
1337                           swp * PAGE_SIZE);
1338 }
1339 #endif