const: mark remaining super_operations const
[linux-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
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42 #include <linux/ima.h>
43
44 #include <asm/uaccess.h>
45
46 #include "util.h"
47
48 struct shm_file_data {
49         int id;
50         struct ipc_namespace *ns;
51         struct file *file;
52         const struct vm_operations_struct *vm_ops;
53 };
54
55 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56
57 static const struct file_operations shm_file_operations;
58 static struct vm_operations_struct shm_vm_ops;
59
60 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
61
62 #define shm_unlock(shp)                 \
63         ipc_unlock(&(shp)->shm_perm)
64
65 static int newseg(struct ipc_namespace *, struct ipc_params *);
66 static void shm_open(struct vm_area_struct *vma);
67 static void shm_close(struct vm_area_struct *vma);
68 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
69 #ifdef CONFIG_PROC_FS
70 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
71 #endif
72
73 void shm_init_ns(struct ipc_namespace *ns)
74 {
75         ns->shm_ctlmax = SHMMAX;
76         ns->shm_ctlall = SHMALL;
77         ns->shm_ctlmni = SHMMNI;
78         ns->shm_tot = 0;
79         ipc_init_ids(&shm_ids(ns));
80 }
81
82 /*
83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84  * Only shm_ids.rw_mutex remains locked on exit.
85  */
86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
87 {
88         struct shmid_kernel *shp;
89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
90
91         if (shp->shm_nattch){
92                 shp->shm_perm.mode |= SHM_DEST;
93                 /* Do not find it any more */
94                 shp->shm_perm.key = IPC_PRIVATE;
95                 shm_unlock(shp);
96         } else
97                 shm_destroy(ns, shp);
98 }
99
100 #ifdef CONFIG_IPC_NS
101 void shm_exit_ns(struct ipc_namespace *ns)
102 {
103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
104 }
105 #endif
106
107 void __init shm_init (void)
108 {
109         shm_init_ns(&init_ipc_ns);
110         ipc_init_proc_interface("sysvipc/shm",
111                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
112                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
113 }
114
115 /*
116  * shm_lock_(check_) routines are called in the paths where the rw_mutex
117  * is not necessarily held.
118  */
119 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
120 {
121         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
122
123         if (IS_ERR(ipcp))
124                 return (struct shmid_kernel *)ipcp;
125
126         return container_of(ipcp, struct shmid_kernel, shm_perm);
127 }
128
129 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
130                                                 int id)
131 {
132         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
133
134         if (IS_ERR(ipcp))
135                 return (struct shmid_kernel *)ipcp;
136
137         return container_of(ipcp, struct shmid_kernel, shm_perm);
138 }
139
140 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
141 {
142         ipc_rmid(&shm_ids(ns), &s->shm_perm);
143 }
144
145
146 /* This is called by fork, once for every shm attach. */
147 static void shm_open(struct vm_area_struct *vma)
148 {
149         struct file *file = vma->vm_file;
150         struct shm_file_data *sfd = shm_file_data(file);
151         struct shmid_kernel *shp;
152
153         shp = shm_lock(sfd->ns, sfd->id);
154         BUG_ON(IS_ERR(shp));
155         shp->shm_atim = get_seconds();
156         shp->shm_lprid = task_tgid_vnr(current);
157         shp->shm_nattch++;
158         shm_unlock(shp);
159 }
160
161 /*
162  * shm_destroy - free the struct shmid_kernel
163  *
164  * @ns: namespace
165  * @shp: struct to free
166  *
167  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
168  * but returns with shp unlocked and freed.
169  */
170 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
171 {
172         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
173         shm_rmid(ns, shp);
174         shm_unlock(shp);
175         if (!is_file_hugepages(shp->shm_file))
176                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
177         else if (shp->mlock_user)
178                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
179                                                 shp->mlock_user);
180         fput (shp->shm_file);
181         security_shm_free(shp);
182         ipc_rcu_putref(shp);
183 }
184
185 /*
186  * remove the attach descriptor vma.
187  * free memory for segment if it is marked destroyed.
188  * The descriptor has already been removed from the current->mm->mmap list
189  * and will later be kfree()d.
190  */
191 static void shm_close(struct vm_area_struct *vma)
192 {
193         struct file * file = vma->vm_file;
194         struct shm_file_data *sfd = shm_file_data(file);
195         struct shmid_kernel *shp;
196         struct ipc_namespace *ns = sfd->ns;
197
198         down_write(&shm_ids(ns).rw_mutex);
199         /* remove from the list of attaches of the shm segment */
200         shp = shm_lock(ns, sfd->id);
201         BUG_ON(IS_ERR(shp));
202         shp->shm_lprid = task_tgid_vnr(current);
203         shp->shm_dtim = get_seconds();
204         shp->shm_nattch--;
205         if(shp->shm_nattch == 0 &&
206            shp->shm_perm.mode & SHM_DEST)
207                 shm_destroy(ns, shp);
208         else
209                 shm_unlock(shp);
210         up_write(&shm_ids(ns).rw_mutex);
211 }
212
213 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
214 {
215         struct file *file = vma->vm_file;
216         struct shm_file_data *sfd = shm_file_data(file);
217
218         return sfd->vm_ops->fault(vma, vmf);
219 }
220
221 #ifdef CONFIG_NUMA
222 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
223 {
224         struct file *file = vma->vm_file;
225         struct shm_file_data *sfd = shm_file_data(file);
226         int err = 0;
227         if (sfd->vm_ops->set_policy)
228                 err = sfd->vm_ops->set_policy(vma, new);
229         return err;
230 }
231
232 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
233                                         unsigned long addr)
234 {
235         struct file *file = vma->vm_file;
236         struct shm_file_data *sfd = shm_file_data(file);
237         struct mempolicy *pol = NULL;
238
239         if (sfd->vm_ops->get_policy)
240                 pol = sfd->vm_ops->get_policy(vma, addr);
241         else if (vma->vm_policy)
242                 pol = vma->vm_policy;
243
244         return pol;
245 }
246 #endif
247
248 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
249 {
250         struct shm_file_data *sfd = shm_file_data(file);
251         int ret;
252
253         ret = sfd->file->f_op->mmap(sfd->file, vma);
254         if (ret != 0)
255                 return ret;
256         sfd->vm_ops = vma->vm_ops;
257 #ifdef CONFIG_MMU
258         BUG_ON(!sfd->vm_ops->fault);
259 #endif
260         vma->vm_ops = &shm_vm_ops;
261         shm_open(vma);
262
263         return ret;
264 }
265
266 static int shm_release(struct inode *ino, struct file *file)
267 {
268         struct shm_file_data *sfd = shm_file_data(file);
269
270         put_ipc_ns(sfd->ns);
271         shm_file_data(file) = NULL;
272         kfree(sfd);
273         return 0;
274 }
275
276 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
277 {
278         int (*fsync) (struct file *, struct dentry *, int datasync);
279         struct shm_file_data *sfd = shm_file_data(file);
280         int ret = -EINVAL;
281
282         fsync = sfd->file->f_op->fsync;
283         if (fsync)
284                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
285         return ret;
286 }
287
288 static unsigned long shm_get_unmapped_area(struct file *file,
289         unsigned long addr, unsigned long len, unsigned long pgoff,
290         unsigned long flags)
291 {
292         struct shm_file_data *sfd = shm_file_data(file);
293         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
294 }
295
296 int is_file_shm_hugepages(struct file *file)
297 {
298         int ret = 0;
299
300         if (file->f_op == &shm_file_operations) {
301                 struct shm_file_data *sfd;
302                 sfd = shm_file_data(file);
303                 ret = is_file_hugepages(sfd->file);
304         }
305         return ret;
306 }
307
308 static const struct file_operations shm_file_operations = {
309         .mmap           = shm_mmap,
310         .fsync          = shm_fsync,
311         .release        = shm_release,
312         .get_unmapped_area      = shm_get_unmapped_area,
313 };
314
315 static struct vm_operations_struct shm_vm_ops = {
316         .open   = shm_open,     /* callback for a new vm-area open */
317         .close  = shm_close,    /* callback for when the vm-area is released */
318         .fault  = shm_fault,
319 #if defined(CONFIG_NUMA)
320         .set_policy = shm_set_policy,
321         .get_policy = shm_get_policy,
322 #endif
323 };
324
325 /**
326  * newseg - Create a new shared memory segment
327  * @ns: namespace
328  * @params: ptr to the structure that contains key, size and shmflg
329  *
330  * Called with shm_ids.rw_mutex held as a writer.
331  */
332
333 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
334 {
335         key_t key = params->key;
336         int shmflg = params->flg;
337         size_t size = params->u.size;
338         int error;
339         struct shmid_kernel *shp;
340         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
341         struct file * file;
342         char name[13];
343         int id;
344         int acctflag = 0;
345
346         if (size < SHMMIN || size > ns->shm_ctlmax)
347                 return -EINVAL;
348
349         if (ns->shm_tot + numpages > ns->shm_ctlall)
350                 return -ENOSPC;
351
352         shp = ipc_rcu_alloc(sizeof(*shp));
353         if (!shp)
354                 return -ENOMEM;
355
356         shp->shm_perm.key = key;
357         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
358         shp->mlock_user = NULL;
359
360         shp->shm_perm.security = NULL;
361         error = security_shm_alloc(shp);
362         if (error) {
363                 ipc_rcu_putref(shp);
364                 return error;
365         }
366
367         sprintf (name, "SYSV%08x", key);
368         if (shmflg & SHM_HUGETLB) {
369                 /* hugetlb_file_setup applies strict accounting */
370                 if (shmflg & SHM_NORESERVE)
371                         acctflag = VM_NORESERVE;
372                 file = hugetlb_file_setup(name, size, acctflag,
373                                                         &shp->mlock_user);
374         } else {
375                 /*
376                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
377                  * if it's asked for.
378                  */
379                 if  ((shmflg & SHM_NORESERVE) &&
380                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
381                         acctflag = VM_NORESERVE;
382                 file = shmem_file_setup(name, size, acctflag);
383         }
384         error = PTR_ERR(file);
385         if (IS_ERR(file))
386                 goto no_file;
387
388         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
389         if (id < 0) {
390                 error = id;
391                 goto no_id;
392         }
393
394         shp->shm_cprid = task_tgid_vnr(current);
395         shp->shm_lprid = 0;
396         shp->shm_atim = shp->shm_dtim = 0;
397         shp->shm_ctim = get_seconds();
398         shp->shm_segsz = size;
399         shp->shm_nattch = 0;
400         shp->shm_file = file;
401         /*
402          * shmid gets reported as "inode#" in /proc/pid/maps.
403          * proc-ps tools use this. Changing this will break them.
404          */
405         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
406
407         ns->shm_tot += numpages;
408         error = shp->shm_perm.id;
409         shm_unlock(shp);
410         return error;
411
412 no_id:
413         if (is_file_hugepages(file) && shp->mlock_user)
414                 user_shm_unlock(size, shp->mlock_user);
415         fput(file);
416 no_file:
417         security_shm_free(shp);
418         ipc_rcu_putref(shp);
419         return error;
420 }
421
422 /*
423  * Called with shm_ids.rw_mutex and ipcp locked.
424  */
425 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
426 {
427         struct shmid_kernel *shp;
428
429         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
430         return security_shm_associate(shp, shmflg);
431 }
432
433 /*
434  * Called with shm_ids.rw_mutex and ipcp locked.
435  */
436 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
437                                 struct ipc_params *params)
438 {
439         struct shmid_kernel *shp;
440
441         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
442         if (shp->shm_segsz < params->u.size)
443                 return -EINVAL;
444
445         return 0;
446 }
447
448 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
449 {
450         struct ipc_namespace *ns;
451         struct ipc_ops shm_ops;
452         struct ipc_params shm_params;
453
454         ns = current->nsproxy->ipc_ns;
455
456         shm_ops.getnew = newseg;
457         shm_ops.associate = shm_security;
458         shm_ops.more_checks = shm_more_checks;
459
460         shm_params.key = key;
461         shm_params.flg = shmflg;
462         shm_params.u.size = size;
463
464         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
465 }
466
467 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
468 {
469         switch(version) {
470         case IPC_64:
471                 return copy_to_user(buf, in, sizeof(*in));
472         case IPC_OLD:
473             {
474                 struct shmid_ds out;
475
476                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
477                 out.shm_segsz   = in->shm_segsz;
478                 out.shm_atime   = in->shm_atime;
479                 out.shm_dtime   = in->shm_dtime;
480                 out.shm_ctime   = in->shm_ctime;
481                 out.shm_cpid    = in->shm_cpid;
482                 out.shm_lpid    = in->shm_lpid;
483                 out.shm_nattch  = in->shm_nattch;
484
485                 return copy_to_user(buf, &out, sizeof(out));
486             }
487         default:
488                 return -EINVAL;
489         }
490 }
491
492 static inline unsigned long
493 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
494 {
495         switch(version) {
496         case IPC_64:
497                 if (copy_from_user(out, buf, sizeof(*out)))
498                         return -EFAULT;
499                 return 0;
500         case IPC_OLD:
501             {
502                 struct shmid_ds tbuf_old;
503
504                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
505                         return -EFAULT;
506
507                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
508                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
509                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
510
511                 return 0;
512             }
513         default:
514                 return -EINVAL;
515         }
516 }
517
518 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
519 {
520         switch(version) {
521         case IPC_64:
522                 return copy_to_user(buf, in, sizeof(*in));
523         case IPC_OLD:
524             {
525                 struct shminfo out;
526
527                 if(in->shmmax > INT_MAX)
528                         out.shmmax = INT_MAX;
529                 else
530                         out.shmmax = (int)in->shmmax;
531
532                 out.shmmin      = in->shmmin;
533                 out.shmmni      = in->shmmni;
534                 out.shmseg      = in->shmseg;
535                 out.shmall      = in->shmall; 
536
537                 return copy_to_user(buf, &out, sizeof(out));
538             }
539         default:
540                 return -EINVAL;
541         }
542 }
543
544 /*
545  * Called with shm_ids.rw_mutex held as a reader
546  */
547 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
548                 unsigned long *swp)
549 {
550         int next_id;
551         int total, in_use;
552
553         *rss = 0;
554         *swp = 0;
555
556         in_use = shm_ids(ns).in_use;
557
558         for (total = 0, next_id = 0; total < in_use; next_id++) {
559                 struct kern_ipc_perm *ipc;
560                 struct shmid_kernel *shp;
561                 struct inode *inode;
562
563                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
564                 if (ipc == NULL)
565                         continue;
566                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
567
568                 inode = shp->shm_file->f_path.dentry->d_inode;
569
570                 if (is_file_hugepages(shp->shm_file)) {
571                         struct address_space *mapping = inode->i_mapping;
572                         struct hstate *h = hstate_file(shp->shm_file);
573                         *rss += pages_per_huge_page(h) * mapping->nrpages;
574                 } else {
575 #ifdef CONFIG_SHMEM
576                         struct shmem_inode_info *info = SHMEM_I(inode);
577                         spin_lock(&info->lock);
578                         *rss += inode->i_mapping->nrpages;
579                         *swp += info->swapped;
580                         spin_unlock(&info->lock);
581 #else
582                         *rss += inode->i_mapping->nrpages;
583 #endif
584                 }
585
586                 total++;
587         }
588 }
589
590 /*
591  * This function handles some shmctl commands which require the rw_mutex
592  * to be held in write mode.
593  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
594  */
595 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
596                        struct shmid_ds __user *buf, int version)
597 {
598         struct kern_ipc_perm *ipcp;
599         struct shmid64_ds shmid64;
600         struct shmid_kernel *shp;
601         int err;
602
603         if (cmd == IPC_SET) {
604                 if (copy_shmid_from_user(&shmid64, buf, version))
605                         return -EFAULT;
606         }
607
608         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
609         if (IS_ERR(ipcp))
610                 return PTR_ERR(ipcp);
611
612         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
613
614         err = security_shm_shmctl(shp, cmd);
615         if (err)
616                 goto out_unlock;
617         switch (cmd) {
618         case IPC_RMID:
619                 do_shm_rmid(ns, ipcp);
620                 goto out_up;
621         case IPC_SET:
622                 ipc_update_perm(&shmid64.shm_perm, ipcp);
623                 shp->shm_ctim = get_seconds();
624                 break;
625         default:
626                 err = -EINVAL;
627         }
628 out_unlock:
629         shm_unlock(shp);
630 out_up:
631         up_write(&shm_ids(ns).rw_mutex);
632         return err;
633 }
634
635 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
636 {
637         struct shmid_kernel *shp;
638         int err, version;
639         struct ipc_namespace *ns;
640
641         if (cmd < 0 || shmid < 0) {
642                 err = -EINVAL;
643                 goto out;
644         }
645
646         version = ipc_parse_version(&cmd);
647         ns = current->nsproxy->ipc_ns;
648
649         switch (cmd) { /* replace with proc interface ? */
650         case IPC_INFO:
651         {
652                 struct shminfo64 shminfo;
653
654                 err = security_shm_shmctl(NULL, cmd);
655                 if (err)
656                         return err;
657
658                 memset(&shminfo, 0, sizeof(shminfo));
659                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
660                 shminfo.shmmax = ns->shm_ctlmax;
661                 shminfo.shmall = ns->shm_ctlall;
662
663                 shminfo.shmmin = SHMMIN;
664                 if(copy_shminfo_to_user (buf, &shminfo, version))
665                         return -EFAULT;
666
667                 down_read(&shm_ids(ns).rw_mutex);
668                 err = ipc_get_maxid(&shm_ids(ns));
669                 up_read(&shm_ids(ns).rw_mutex);
670
671                 if(err<0)
672                         err = 0;
673                 goto out;
674         }
675         case SHM_INFO:
676         {
677                 struct shm_info shm_info;
678
679                 err = security_shm_shmctl(NULL, cmd);
680                 if (err)
681                         return err;
682
683                 memset(&shm_info, 0, sizeof(shm_info));
684                 down_read(&shm_ids(ns).rw_mutex);
685                 shm_info.used_ids = shm_ids(ns).in_use;
686                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
687                 shm_info.shm_tot = ns->shm_tot;
688                 shm_info.swap_attempts = 0;
689                 shm_info.swap_successes = 0;
690                 err = ipc_get_maxid(&shm_ids(ns));
691                 up_read(&shm_ids(ns).rw_mutex);
692                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
693                         err = -EFAULT;
694                         goto out;
695                 }
696
697                 err = err < 0 ? 0 : err;
698                 goto out;
699         }
700         case SHM_STAT:
701         case IPC_STAT:
702         {
703                 struct shmid64_ds tbuf;
704                 int result;
705
706                 if (cmd == SHM_STAT) {
707                         shp = shm_lock(ns, shmid);
708                         if (IS_ERR(shp)) {
709                                 err = PTR_ERR(shp);
710                                 goto out;
711                         }
712                         result = shp->shm_perm.id;
713                 } else {
714                         shp = shm_lock_check(ns, shmid);
715                         if (IS_ERR(shp)) {
716                                 err = PTR_ERR(shp);
717                                 goto out;
718                         }
719                         result = 0;
720                 }
721                 err = -EACCES;
722                 if (ipcperms (&shp->shm_perm, S_IRUGO))
723                         goto out_unlock;
724                 err = security_shm_shmctl(shp, cmd);
725                 if (err)
726                         goto out_unlock;
727                 memset(&tbuf, 0, sizeof(tbuf));
728                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
729                 tbuf.shm_segsz  = shp->shm_segsz;
730                 tbuf.shm_atime  = shp->shm_atim;
731                 tbuf.shm_dtime  = shp->shm_dtim;
732                 tbuf.shm_ctime  = shp->shm_ctim;
733                 tbuf.shm_cpid   = shp->shm_cprid;
734                 tbuf.shm_lpid   = shp->shm_lprid;
735                 tbuf.shm_nattch = shp->shm_nattch;
736                 shm_unlock(shp);
737                 if(copy_shmid_to_user (buf, &tbuf, version))
738                         err = -EFAULT;
739                 else
740                         err = result;
741                 goto out;
742         }
743         case SHM_LOCK:
744         case SHM_UNLOCK:
745         {
746                 struct file *uninitialized_var(shm_file);
747
748                 lru_add_drain_all();  /* drain pagevecs to lru lists */
749
750                 shp = shm_lock_check(ns, shmid);
751                 if (IS_ERR(shp)) {
752                         err = PTR_ERR(shp);
753                         goto out;
754                 }
755
756                 audit_ipc_obj(&(shp->shm_perm));
757
758                 if (!capable(CAP_IPC_LOCK)) {
759                         uid_t euid = current_euid();
760                         err = -EPERM;
761                         if (euid != shp->shm_perm.uid &&
762                             euid != shp->shm_perm.cuid)
763                                 goto out_unlock;
764                         if (cmd == SHM_LOCK &&
765                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
766                                 goto out_unlock;
767                 }
768
769                 err = security_shm_shmctl(shp, cmd);
770                 if (err)
771                         goto out_unlock;
772                 
773                 if(cmd==SHM_LOCK) {
774                         struct user_struct *user = current_user();
775                         if (!is_file_hugepages(shp->shm_file)) {
776                                 err = shmem_lock(shp->shm_file, 1, user);
777                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
778                                         shp->shm_perm.mode |= SHM_LOCKED;
779                                         shp->mlock_user = user;
780                                 }
781                         }
782                 } else if (!is_file_hugepages(shp->shm_file)) {
783                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
784                         shp->shm_perm.mode &= ~SHM_LOCKED;
785                         shp->mlock_user = NULL;
786                 }
787                 shm_unlock(shp);
788                 goto out;
789         }
790         case IPC_RMID:
791         case IPC_SET:
792                 err = shmctl_down(ns, shmid, cmd, buf, version);
793                 return err;
794         default:
795                 return -EINVAL;
796         }
797
798 out_unlock:
799         shm_unlock(shp);
800 out:
801         return err;
802 }
803
804 /*
805  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
806  *
807  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
808  * "raddr" thing points to kernel space, and there has to be a wrapper around
809  * this.
810  */
811 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
812 {
813         struct shmid_kernel *shp;
814         unsigned long addr;
815         unsigned long size;
816         struct file * file;
817         int    err;
818         unsigned long flags;
819         unsigned long prot;
820         int acc_mode;
821         unsigned long user_addr;
822         struct ipc_namespace *ns;
823         struct shm_file_data *sfd;
824         struct path path;
825         fmode_t f_mode;
826
827         err = -EINVAL;
828         if (shmid < 0)
829                 goto out;
830         else if ((addr = (ulong)shmaddr)) {
831                 if (addr & (SHMLBA-1)) {
832                         if (shmflg & SHM_RND)
833                                 addr &= ~(SHMLBA-1);       /* round down */
834                         else
835 #ifndef __ARCH_FORCE_SHMLBA
836                                 if (addr & ~PAGE_MASK)
837 #endif
838                                         goto out;
839                 }
840                 flags = MAP_SHARED | MAP_FIXED;
841         } else {
842                 if ((shmflg & SHM_REMAP))
843                         goto out;
844
845                 flags = MAP_SHARED;
846         }
847
848         if (shmflg & SHM_RDONLY) {
849                 prot = PROT_READ;
850                 acc_mode = S_IRUGO;
851                 f_mode = FMODE_READ;
852         } else {
853                 prot = PROT_READ | PROT_WRITE;
854                 acc_mode = S_IRUGO | S_IWUGO;
855                 f_mode = FMODE_READ | FMODE_WRITE;
856         }
857         if (shmflg & SHM_EXEC) {
858                 prot |= PROT_EXEC;
859                 acc_mode |= S_IXUGO;
860         }
861
862         /*
863          * We cannot rely on the fs check since SYSV IPC does have an
864          * additional creator id...
865          */
866         ns = current->nsproxy->ipc_ns;
867         shp = shm_lock_check(ns, shmid);
868         if (IS_ERR(shp)) {
869                 err = PTR_ERR(shp);
870                 goto out;
871         }
872
873         err = -EACCES;
874         if (ipcperms(&shp->shm_perm, acc_mode))
875                 goto out_unlock;
876
877         err = security_shm_shmat(shp, shmaddr, shmflg);
878         if (err)
879                 goto out_unlock;
880
881         path.dentry = dget(shp->shm_file->f_path.dentry);
882         path.mnt    = shp->shm_file->f_path.mnt;
883         shp->shm_nattch++;
884         size = i_size_read(path.dentry->d_inode);
885         shm_unlock(shp);
886
887         err = -ENOMEM;
888         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
889         if (!sfd)
890                 goto out_put_dentry;
891
892         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
893         if (!file)
894                 goto out_free;
895         ima_counts_get(file);
896
897         file->private_data = sfd;
898         file->f_mapping = shp->shm_file->f_mapping;
899         sfd->id = shp->shm_perm.id;
900         sfd->ns = get_ipc_ns(ns);
901         sfd->file = shp->shm_file;
902         sfd->vm_ops = NULL;
903
904         down_write(&current->mm->mmap_sem);
905         if (addr && !(shmflg & SHM_REMAP)) {
906                 err = -EINVAL;
907                 if (find_vma_intersection(current->mm, addr, addr + size))
908                         goto invalid;
909                 /*
910                  * If shm segment goes below stack, make sure there is some
911                  * space left for the stack to grow (at least 4 pages).
912                  */
913                 if (addr < current->mm->start_stack &&
914                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
915                         goto invalid;
916         }
917                 
918         user_addr = do_mmap (file, addr, size, prot, flags, 0);
919         *raddr = user_addr;
920         err = 0;
921         if (IS_ERR_VALUE(user_addr))
922                 err = (long)user_addr;
923 invalid:
924         up_write(&current->mm->mmap_sem);
925
926         fput(file);
927
928 out_nattch:
929         down_write(&shm_ids(ns).rw_mutex);
930         shp = shm_lock(ns, shmid);
931         BUG_ON(IS_ERR(shp));
932         shp->shm_nattch--;
933         if(shp->shm_nattch == 0 &&
934            shp->shm_perm.mode & SHM_DEST)
935                 shm_destroy(ns, shp);
936         else
937                 shm_unlock(shp);
938         up_write(&shm_ids(ns).rw_mutex);
939
940 out:
941         return err;
942
943 out_unlock:
944         shm_unlock(shp);
945         goto out;
946
947 out_free:
948         kfree(sfd);
949 out_put_dentry:
950         dput(path.dentry);
951         goto out_nattch;
952 }
953
954 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
955 {
956         unsigned long ret;
957         long err;
958
959         err = do_shmat(shmid, shmaddr, shmflg, &ret);
960         if (err)
961                 return err;
962         force_successful_syscall_return();
963         return (long)ret;
964 }
965
966 /*
967  * detach and kill segment if marked destroyed.
968  * The work is done in shm_close.
969  */
970 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
971 {
972         struct mm_struct *mm = current->mm;
973         struct vm_area_struct *vma;
974         unsigned long addr = (unsigned long)shmaddr;
975         int retval = -EINVAL;
976 #ifdef CONFIG_MMU
977         loff_t size = 0;
978         struct vm_area_struct *next;
979 #endif
980
981         if (addr & ~PAGE_MASK)
982                 return retval;
983
984         down_write(&mm->mmap_sem);
985
986         /*
987          * This function tries to be smart and unmap shm segments that
988          * were modified by partial mlock or munmap calls:
989          * - It first determines the size of the shm segment that should be
990          *   unmapped: It searches for a vma that is backed by shm and that
991          *   started at address shmaddr. It records it's size and then unmaps
992          *   it.
993          * - Then it unmaps all shm vmas that started at shmaddr and that
994          *   are within the initially determined size.
995          * Errors from do_munmap are ignored: the function only fails if
996          * it's called with invalid parameters or if it's called to unmap
997          * a part of a vma. Both calls in this function are for full vmas,
998          * the parameters are directly copied from the vma itself and always
999          * valid - therefore do_munmap cannot fail. (famous last words?)
1000          */
1001         /*
1002          * If it had been mremap()'d, the starting address would not
1003          * match the usual checks anyway. So assume all vma's are
1004          * above the starting address given.
1005          */
1006         vma = find_vma(mm, addr);
1007
1008 #ifdef CONFIG_MMU
1009         while (vma) {
1010                 next = vma->vm_next;
1011
1012                 /*
1013                  * Check if the starting address would match, i.e. it's
1014                  * a fragment created by mprotect() and/or munmap(), or it
1015                  * otherwise it starts at this address with no hassles.
1016                  */
1017                 if ((vma->vm_ops == &shm_vm_ops) &&
1018                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1019
1020
1021                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1022                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1023                         /*
1024                          * We discovered the size of the shm segment, so
1025                          * break out of here and fall through to the next
1026                          * loop that uses the size information to stop
1027                          * searching for matching vma's.
1028                          */
1029                         retval = 0;
1030                         vma = next;
1031                         break;
1032                 }
1033                 vma = next;
1034         }
1035
1036         /*
1037          * We need look no further than the maximum address a fragment
1038          * could possibly have landed at. Also cast things to loff_t to
1039          * prevent overflows and make comparisions vs. equal-width types.
1040          */
1041         size = PAGE_ALIGN(size);
1042         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1043                 next = vma->vm_next;
1044
1045                 /* finding a matching vma now does not alter retval */
1046                 if ((vma->vm_ops == &shm_vm_ops) &&
1047                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1048
1049                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1050                 vma = next;
1051         }
1052
1053 #else /* CONFIG_MMU */
1054         /* under NOMMU conditions, the exact address to be destroyed must be
1055          * given */
1056         retval = -EINVAL;
1057         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1058                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1059                 retval = 0;
1060         }
1061
1062 #endif
1063
1064         up_write(&mm->mmap_sem);
1065         return retval;
1066 }
1067
1068 #ifdef CONFIG_PROC_FS
1069 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1070 {
1071         struct shmid_kernel *shp = it;
1072
1073 #if BITS_PER_LONG <= 32
1074 #define SIZE_SPEC "%10lu"
1075 #else
1076 #define SIZE_SPEC "%21lu"
1077 #endif
1078
1079         return seq_printf(s,
1080                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1081                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1082                           shp->shm_perm.key,
1083                           shp->shm_perm.id,
1084                           shp->shm_perm.mode,
1085                           shp->shm_segsz,
1086                           shp->shm_cprid,
1087                           shp->shm_lprid,
1088                           shp->shm_nattch,
1089                           shp->shm_perm.uid,
1090                           shp->shm_perm.gid,
1091                           shp->shm_perm.cuid,
1092                           shp->shm_perm.cgid,
1093                           shp->shm_atim,
1094                           shp->shm_dtim,
1095                           shp->shm_ctim);
1096 }
1097 #endif