IPC/shared memory: introduce shmctl_down
[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
43 #include <asm/uaccess.h>
44
45 #include "util.h"
46
47 struct shm_file_data {
48         int id;
49         struct ipc_namespace *ns;
50         struct file *file;
51         const struct vm_operations_struct *vm_ops;
52 };
53
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
56 static const struct file_operations shm_file_operations;
57 static struct vm_operations_struct shm_vm_ops;
58
59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
60
61 #define shm_unlock(shp)                 \
62         ipc_unlock(&(shp)->shm_perm)
63
64 static int newseg(struct ipc_namespace *, struct ipc_params *);
65 static void shm_open(struct vm_area_struct *vma);
66 static void shm_close(struct vm_area_struct *vma);
67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
68 #ifdef CONFIG_PROC_FS
69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
70 #endif
71
72 void shm_init_ns(struct ipc_namespace *ns)
73 {
74         ns->shm_ctlmax = SHMMAX;
75         ns->shm_ctlall = SHMALL;
76         ns->shm_ctlmni = SHMMNI;
77         ns->shm_tot = 0;
78         ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
79 }
80
81 /*
82  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
83  * Only shm_ids.rw_mutex remains locked on exit.
84  */
85 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
86 {
87         struct shmid_kernel *shp;
88         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
89
90         if (shp->shm_nattch){
91                 shp->shm_perm.mode |= SHM_DEST;
92                 /* Do not find it any more */
93                 shp->shm_perm.key = IPC_PRIVATE;
94                 shm_unlock(shp);
95         } else
96                 shm_destroy(ns, shp);
97 }
98
99 #ifdef CONFIG_IPC_NS
100 void shm_exit_ns(struct ipc_namespace *ns)
101 {
102         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
103 }
104 #endif
105
106 void __init shm_init (void)
107 {
108         shm_init_ns(&init_ipc_ns);
109         ipc_init_proc_interface("sysvipc/shm",
110                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
111                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
112 }
113
114 /*
115  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
116  * is held to protect access to the idr tree.
117  */
118 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
119                                                 int id)
120 {
121         struct kern_ipc_perm *ipcp = ipc_lock_down(&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_down(
130                                                 struct ipc_namespace *ns,
131                                                 int id)
132 {
133         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
134
135         if (IS_ERR(ipcp))
136                 return (struct shmid_kernel *)ipcp;
137
138         return container_of(ipcp, struct shmid_kernel, shm_perm);
139 }
140
141 /*
142  * shm_lock_(check_) routines are called in the paths where the rw_mutex
143  * is not held.
144  */
145 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
146 {
147         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
148
149         if (IS_ERR(ipcp))
150                 return (struct shmid_kernel *)ipcp;
151
152         return container_of(ipcp, struct shmid_kernel, shm_perm);
153 }
154
155 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
156                                                 int id)
157 {
158         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
159
160         if (IS_ERR(ipcp))
161                 return (struct shmid_kernel *)ipcp;
162
163         return container_of(ipcp, struct shmid_kernel, shm_perm);
164 }
165
166 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
167 {
168         ipc_rmid(&shm_ids(ns), &s->shm_perm);
169 }
170
171
172 /* This is called by fork, once for every shm attach. */
173 static void shm_open(struct vm_area_struct *vma)
174 {
175         struct file *file = vma->vm_file;
176         struct shm_file_data *sfd = shm_file_data(file);
177         struct shmid_kernel *shp;
178
179         shp = shm_lock(sfd->ns, sfd->id);
180         BUG_ON(IS_ERR(shp));
181         shp->shm_atim = get_seconds();
182         shp->shm_lprid = task_tgid_vnr(current);
183         shp->shm_nattch++;
184         shm_unlock(shp);
185 }
186
187 /*
188  * shm_destroy - free the struct shmid_kernel
189  *
190  * @ns: namespace
191  * @shp: struct to free
192  *
193  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
194  * but returns with shp unlocked and freed.
195  */
196 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
197 {
198         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
199         shm_rmid(ns, shp);
200         shm_unlock(shp);
201         if (!is_file_hugepages(shp->shm_file))
202                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
203         else
204                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
205                                                 shp->mlock_user);
206         fput (shp->shm_file);
207         security_shm_free(shp);
208         ipc_rcu_putref(shp);
209 }
210
211 /*
212  * remove the attach descriptor vma.
213  * free memory for segment if it is marked destroyed.
214  * The descriptor has already been removed from the current->mm->mmap list
215  * and will later be kfree()d.
216  */
217 static void shm_close(struct vm_area_struct *vma)
218 {
219         struct file * file = vma->vm_file;
220         struct shm_file_data *sfd = shm_file_data(file);
221         struct shmid_kernel *shp;
222         struct ipc_namespace *ns = sfd->ns;
223
224         down_write(&shm_ids(ns).rw_mutex);
225         /* remove from the list of attaches of the shm segment */
226         shp = shm_lock_down(ns, sfd->id);
227         BUG_ON(IS_ERR(shp));
228         shp->shm_lprid = task_tgid_vnr(current);
229         shp->shm_dtim = get_seconds();
230         shp->shm_nattch--;
231         if(shp->shm_nattch == 0 &&
232            shp->shm_perm.mode & SHM_DEST)
233                 shm_destroy(ns, shp);
234         else
235                 shm_unlock(shp);
236         up_write(&shm_ids(ns).rw_mutex);
237 }
238
239 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
240 {
241         struct file *file = vma->vm_file;
242         struct shm_file_data *sfd = shm_file_data(file);
243
244         return sfd->vm_ops->fault(vma, vmf);
245 }
246
247 #ifdef CONFIG_NUMA
248 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
249 {
250         struct file *file = vma->vm_file;
251         struct shm_file_data *sfd = shm_file_data(file);
252         int err = 0;
253         if (sfd->vm_ops->set_policy)
254                 err = sfd->vm_ops->set_policy(vma, new);
255         return err;
256 }
257
258 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
259                                         unsigned long addr)
260 {
261         struct file *file = vma->vm_file;
262         struct shm_file_data *sfd = shm_file_data(file);
263         struct mempolicy *pol = NULL;
264
265         if (sfd->vm_ops->get_policy)
266                 pol = sfd->vm_ops->get_policy(vma, addr);
267         else if (vma->vm_policy)
268                 pol = vma->vm_policy;
269
270         return pol;
271 }
272 #endif
273
274 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
275 {
276         struct shm_file_data *sfd = shm_file_data(file);
277         int ret;
278
279         ret = sfd->file->f_op->mmap(sfd->file, vma);
280         if (ret != 0)
281                 return ret;
282         sfd->vm_ops = vma->vm_ops;
283 #ifdef CONFIG_MMU
284         BUG_ON(!sfd->vm_ops->fault);
285 #endif
286         vma->vm_ops = &shm_vm_ops;
287         shm_open(vma);
288
289         return ret;
290 }
291
292 static int shm_release(struct inode *ino, struct file *file)
293 {
294         struct shm_file_data *sfd = shm_file_data(file);
295
296         put_ipc_ns(sfd->ns);
297         shm_file_data(file) = NULL;
298         kfree(sfd);
299         return 0;
300 }
301
302 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
303 {
304         int (*fsync) (struct file *, struct dentry *, int datasync);
305         struct shm_file_data *sfd = shm_file_data(file);
306         int ret = -EINVAL;
307
308         fsync = sfd->file->f_op->fsync;
309         if (fsync)
310                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
311         return ret;
312 }
313
314 static unsigned long shm_get_unmapped_area(struct file *file,
315         unsigned long addr, unsigned long len, unsigned long pgoff,
316         unsigned long flags)
317 {
318         struct shm_file_data *sfd = shm_file_data(file);
319         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
320 }
321
322 int is_file_shm_hugepages(struct file *file)
323 {
324         int ret = 0;
325
326         if (file->f_op == &shm_file_operations) {
327                 struct shm_file_data *sfd;
328                 sfd = shm_file_data(file);
329                 ret = is_file_hugepages(sfd->file);
330         }
331         return ret;
332 }
333
334 static const struct file_operations shm_file_operations = {
335         .mmap           = shm_mmap,
336         .fsync          = shm_fsync,
337         .release        = shm_release,
338         .get_unmapped_area      = shm_get_unmapped_area,
339 };
340
341 static struct vm_operations_struct shm_vm_ops = {
342         .open   = shm_open,     /* callback for a new vm-area open */
343         .close  = shm_close,    /* callback for when the vm-area is released */
344         .fault  = shm_fault,
345 #if defined(CONFIG_NUMA)
346         .set_policy = shm_set_policy,
347         .get_policy = shm_get_policy,
348 #endif
349 };
350
351 /**
352  * newseg - Create a new shared memory segment
353  * @ns: namespace
354  * @params: ptr to the structure that contains key, size and shmflg
355  *
356  * Called with shm_ids.rw_mutex held as a writer.
357  */
358
359 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
360 {
361         key_t key = params->key;
362         int shmflg = params->flg;
363         size_t size = params->u.size;
364         int error;
365         struct shmid_kernel *shp;
366         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
367         struct file * file;
368         char name[13];
369         int id;
370
371         if (size < SHMMIN || size > ns->shm_ctlmax)
372                 return -EINVAL;
373
374         if (ns->shm_tot + numpages > ns->shm_ctlall)
375                 return -ENOSPC;
376
377         shp = ipc_rcu_alloc(sizeof(*shp));
378         if (!shp)
379                 return -ENOMEM;
380
381         shp->shm_perm.key = key;
382         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
383         shp->mlock_user = NULL;
384
385         shp->shm_perm.security = NULL;
386         error = security_shm_alloc(shp);
387         if (error) {
388                 ipc_rcu_putref(shp);
389                 return error;
390         }
391
392         sprintf (name, "SYSV%08x", key);
393         if (shmflg & SHM_HUGETLB) {
394                 /* hugetlb_file_setup takes care of mlock user accounting */
395                 file = hugetlb_file_setup(name, size);
396                 shp->mlock_user = current->user;
397         } else {
398                 int acctflag = VM_ACCOUNT;
399                 /*
400                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
401                  * if it's asked for.
402                  */
403                 if  ((shmflg & SHM_NORESERVE) &&
404                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
405                         acctflag = 0;
406                 file = shmem_file_setup(name, size, acctflag);
407         }
408         error = PTR_ERR(file);
409         if (IS_ERR(file))
410                 goto no_file;
411
412         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
413         if (id < 0) {
414                 error = id;
415                 goto no_id;
416         }
417
418         shp->shm_cprid = task_tgid_vnr(current);
419         shp->shm_lprid = 0;
420         shp->shm_atim = shp->shm_dtim = 0;
421         shp->shm_ctim = get_seconds();
422         shp->shm_segsz = size;
423         shp->shm_nattch = 0;
424         shp->shm_file = file;
425         /*
426          * shmid gets reported as "inode#" in /proc/pid/maps.
427          * proc-ps tools use this. Changing this will break them.
428          */
429         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
430
431         ns->shm_tot += numpages;
432         error = shp->shm_perm.id;
433         shm_unlock(shp);
434         return error;
435
436 no_id:
437         fput(file);
438 no_file:
439         security_shm_free(shp);
440         ipc_rcu_putref(shp);
441         return error;
442 }
443
444 /*
445  * Called with shm_ids.rw_mutex and ipcp locked.
446  */
447 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
448 {
449         struct shmid_kernel *shp;
450
451         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
452         return security_shm_associate(shp, shmflg);
453 }
454
455 /*
456  * Called with shm_ids.rw_mutex and ipcp locked.
457  */
458 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
459                                 struct ipc_params *params)
460 {
461         struct shmid_kernel *shp;
462
463         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
464         if (shp->shm_segsz < params->u.size)
465                 return -EINVAL;
466
467         return 0;
468 }
469
470 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
471 {
472         struct ipc_namespace *ns;
473         struct ipc_ops shm_ops;
474         struct ipc_params shm_params;
475
476         ns = current->nsproxy->ipc_ns;
477
478         shm_ops.getnew = newseg;
479         shm_ops.associate = shm_security;
480         shm_ops.more_checks = shm_more_checks;
481
482         shm_params.key = key;
483         shm_params.flg = shmflg;
484         shm_params.u.size = size;
485
486         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
487 }
488
489 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
490 {
491         switch(version) {
492         case IPC_64:
493                 return copy_to_user(buf, in, sizeof(*in));
494         case IPC_OLD:
495             {
496                 struct shmid_ds out;
497
498                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
499                 out.shm_segsz   = in->shm_segsz;
500                 out.shm_atime   = in->shm_atime;
501                 out.shm_dtime   = in->shm_dtime;
502                 out.shm_ctime   = in->shm_ctime;
503                 out.shm_cpid    = in->shm_cpid;
504                 out.shm_lpid    = in->shm_lpid;
505                 out.shm_nattch  = in->shm_nattch;
506
507                 return copy_to_user(buf, &out, sizeof(out));
508             }
509         default:
510                 return -EINVAL;
511         }
512 }
513
514 struct shm_setbuf {
515         uid_t   uid;
516         gid_t   gid;
517         mode_t  mode;
518 };      
519
520 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
521 {
522         switch(version) {
523         case IPC_64:
524             {
525                 struct shmid64_ds tbuf;
526
527                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
528                         return -EFAULT;
529
530                 out->uid        = tbuf.shm_perm.uid;
531                 out->gid        = tbuf.shm_perm.gid;
532                 out->mode       = tbuf.shm_perm.mode;
533
534                 return 0;
535             }
536         case IPC_OLD:
537             {
538                 struct shmid_ds tbuf_old;
539
540                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
541                         return -EFAULT;
542
543                 out->uid        = tbuf_old.shm_perm.uid;
544                 out->gid        = tbuf_old.shm_perm.gid;
545                 out->mode       = tbuf_old.shm_perm.mode;
546
547                 return 0;
548             }
549         default:
550                 return -EINVAL;
551         }
552 }
553
554 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
555 {
556         switch(version) {
557         case IPC_64:
558                 return copy_to_user(buf, in, sizeof(*in));
559         case IPC_OLD:
560             {
561                 struct shminfo out;
562
563                 if(in->shmmax > INT_MAX)
564                         out.shmmax = INT_MAX;
565                 else
566                         out.shmmax = (int)in->shmmax;
567
568                 out.shmmin      = in->shmmin;
569                 out.shmmni      = in->shmmni;
570                 out.shmseg      = in->shmseg;
571                 out.shmall      = in->shmall; 
572
573                 return copy_to_user(buf, &out, sizeof(out));
574             }
575         default:
576                 return -EINVAL;
577         }
578 }
579
580 /*
581  * Called with shm_ids.rw_mutex held as a reader
582  */
583 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
584                 unsigned long *swp)
585 {
586         int next_id;
587         int total, in_use;
588
589         *rss = 0;
590         *swp = 0;
591
592         in_use = shm_ids(ns).in_use;
593
594         for (total = 0, next_id = 0; total < in_use; next_id++) {
595                 struct shmid_kernel *shp;
596                 struct inode *inode;
597
598                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
599                 if (shp == NULL)
600                         continue;
601
602                 inode = shp->shm_file->f_path.dentry->d_inode;
603
604                 if (is_file_hugepages(shp->shm_file)) {
605                         struct address_space *mapping = inode->i_mapping;
606                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
607                 } else {
608                         struct shmem_inode_info *info = SHMEM_I(inode);
609                         spin_lock(&info->lock);
610                         *rss += inode->i_mapping->nrpages;
611                         *swp += info->swapped;
612                         spin_unlock(&info->lock);
613                 }
614
615                 total++;
616         }
617 }
618
619 /*
620  * This function handles some shmctl commands which require the rw_mutex
621  * to be held in write mode.
622  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
623  */
624 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
625                        struct shmid_ds __user *buf, int version)
626 {
627         struct kern_ipc_perm *ipcp;
628         struct shm_setbuf setbuf;
629         struct shmid_kernel *shp;
630         int err;
631
632         if (cmd == IPC_SET) {
633                 if (copy_shmid_from_user(&setbuf, buf, version))
634                         return -EFAULT;
635         }
636
637         down_write(&shm_ids(ns).rw_mutex);
638         shp = shm_lock_check_down(ns, shmid);
639         if (IS_ERR(shp)) {
640                 err = PTR_ERR(shp);
641                 goto out_up;
642         }
643
644         ipcp = &shp->shm_perm;
645
646         err = audit_ipc_obj(ipcp);
647         if (err)
648                 goto out_unlock;
649
650         if (cmd == IPC_SET) {
651                 err = audit_ipc_set_perm(0, setbuf.uid,
652                                          setbuf.gid, setbuf.mode);
653                 if (err)
654                         goto out_unlock;
655         }
656
657         if (current->euid != ipcp->uid &&
658             current->euid != ipcp->cuid &&
659             !capable(CAP_SYS_ADMIN)) {
660                 err = -EPERM;
661                 goto out_unlock;
662         }
663
664         err = security_shm_shmctl(shp, cmd);
665         if (err)
666                 goto out_unlock;
667         switch (cmd) {
668         case IPC_RMID:
669                 do_shm_rmid(ns, ipcp);
670                 goto out_up;
671         case IPC_SET:
672                 ipcp->uid = setbuf.uid;
673                 ipcp->gid = setbuf.gid;
674                 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
675                         | (setbuf.mode & S_IRWXUGO);
676                 shp->shm_ctim = get_seconds();
677                 break;
678         default:
679                 err = -EINVAL;
680         }
681 out_unlock:
682         shm_unlock(shp);
683 out_up:
684         up_write(&shm_ids(ns).rw_mutex);
685         return err;
686 }
687
688 asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
689 {
690         struct shmid_kernel *shp;
691         int err, version;
692         struct ipc_namespace *ns;
693
694         if (cmd < 0 || shmid < 0) {
695                 err = -EINVAL;
696                 goto out;
697         }
698
699         version = ipc_parse_version(&cmd);
700         ns = current->nsproxy->ipc_ns;
701
702         switch (cmd) { /* replace with proc interface ? */
703         case IPC_INFO:
704         {
705                 struct shminfo64 shminfo;
706
707                 err = security_shm_shmctl(NULL, cmd);
708                 if (err)
709                         return err;
710
711                 memset(&shminfo,0,sizeof(shminfo));
712                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
713                 shminfo.shmmax = ns->shm_ctlmax;
714                 shminfo.shmall = ns->shm_ctlall;
715
716                 shminfo.shmmin = SHMMIN;
717                 if(copy_shminfo_to_user (buf, &shminfo, version))
718                         return -EFAULT;
719
720                 down_read(&shm_ids(ns).rw_mutex);
721                 err = ipc_get_maxid(&shm_ids(ns));
722                 up_read(&shm_ids(ns).rw_mutex);
723
724                 if(err<0)
725                         err = 0;
726                 goto out;
727         }
728         case SHM_INFO:
729         {
730                 struct shm_info shm_info;
731
732                 err = security_shm_shmctl(NULL, cmd);
733                 if (err)
734                         return err;
735
736                 memset(&shm_info,0,sizeof(shm_info));
737                 down_read(&shm_ids(ns).rw_mutex);
738                 shm_info.used_ids = shm_ids(ns).in_use;
739                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
740                 shm_info.shm_tot = ns->shm_tot;
741                 shm_info.swap_attempts = 0;
742                 shm_info.swap_successes = 0;
743                 err = ipc_get_maxid(&shm_ids(ns));
744                 up_read(&shm_ids(ns).rw_mutex);
745                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
746                         err = -EFAULT;
747                         goto out;
748                 }
749
750                 err = err < 0 ? 0 : err;
751                 goto out;
752         }
753         case SHM_STAT:
754         case IPC_STAT:
755         {
756                 struct shmid64_ds tbuf;
757                 int result;
758
759                 if (!buf) {
760                         err = -EFAULT;
761                         goto out;
762                 }
763
764                 if (cmd == SHM_STAT) {
765                         shp = shm_lock(ns, shmid);
766                         if (IS_ERR(shp)) {
767                                 err = PTR_ERR(shp);
768                                 goto out;
769                         }
770                         result = shp->shm_perm.id;
771                 } else {
772                         shp = shm_lock_check(ns, shmid);
773                         if (IS_ERR(shp)) {
774                                 err = PTR_ERR(shp);
775                                 goto out;
776                         }
777                         result = 0;
778                 }
779                 err=-EACCES;
780                 if (ipcperms (&shp->shm_perm, S_IRUGO))
781                         goto out_unlock;
782                 err = security_shm_shmctl(shp, cmd);
783                 if (err)
784                         goto out_unlock;
785                 memset(&tbuf, 0, sizeof(tbuf));
786                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
787                 tbuf.shm_segsz  = shp->shm_segsz;
788                 tbuf.shm_atime  = shp->shm_atim;
789                 tbuf.shm_dtime  = shp->shm_dtim;
790                 tbuf.shm_ctime  = shp->shm_ctim;
791                 tbuf.shm_cpid   = shp->shm_cprid;
792                 tbuf.shm_lpid   = shp->shm_lprid;
793                 tbuf.shm_nattch = shp->shm_nattch;
794                 shm_unlock(shp);
795                 if(copy_shmid_to_user (buf, &tbuf, version))
796                         err = -EFAULT;
797                 else
798                         err = result;
799                 goto out;
800         }
801         case SHM_LOCK:
802         case SHM_UNLOCK:
803         {
804                 shp = shm_lock_check(ns, shmid);
805                 if (IS_ERR(shp)) {
806                         err = PTR_ERR(shp);
807                         goto out;
808                 }
809
810                 err = audit_ipc_obj(&(shp->shm_perm));
811                 if (err)
812                         goto out_unlock;
813
814                 if (!capable(CAP_IPC_LOCK)) {
815                         err = -EPERM;
816                         if (current->euid != shp->shm_perm.uid &&
817                             current->euid != shp->shm_perm.cuid)
818                                 goto out_unlock;
819                         if (cmd == SHM_LOCK &&
820                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
821                                 goto out_unlock;
822                 }
823
824                 err = security_shm_shmctl(shp, cmd);
825                 if (err)
826                         goto out_unlock;
827                 
828                 if(cmd==SHM_LOCK) {
829                         struct user_struct * user = current->user;
830                         if (!is_file_hugepages(shp->shm_file)) {
831                                 err = shmem_lock(shp->shm_file, 1, user);
832                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
833                                         shp->shm_perm.mode |= SHM_LOCKED;
834                                         shp->mlock_user = user;
835                                 }
836                         }
837                 } else if (!is_file_hugepages(shp->shm_file)) {
838                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
839                         shp->shm_perm.mode &= ~SHM_LOCKED;
840                         shp->mlock_user = NULL;
841                 }
842                 shm_unlock(shp);
843                 goto out;
844         }
845         case IPC_RMID:
846         case IPC_SET:
847                 err = shmctl_down(ns, shmid, cmd, buf, version);
848                 return err;
849         default:
850                 return -EINVAL;
851         }
852
853 out_unlock:
854         shm_unlock(shp);
855 out:
856         return err;
857 }
858
859 /*
860  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
861  *
862  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
863  * "raddr" thing points to kernel space, and there has to be a wrapper around
864  * this.
865  */
866 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
867 {
868         struct shmid_kernel *shp;
869         unsigned long addr;
870         unsigned long size;
871         struct file * file;
872         int    err;
873         unsigned long flags;
874         unsigned long prot;
875         int acc_mode;
876         unsigned long user_addr;
877         struct ipc_namespace *ns;
878         struct shm_file_data *sfd;
879         struct path path;
880         mode_t f_mode;
881
882         err = -EINVAL;
883         if (shmid < 0)
884                 goto out;
885         else if ((addr = (ulong)shmaddr)) {
886                 if (addr & (SHMLBA-1)) {
887                         if (shmflg & SHM_RND)
888                                 addr &= ~(SHMLBA-1);       /* round down */
889                         else
890 #ifndef __ARCH_FORCE_SHMLBA
891                                 if (addr & ~PAGE_MASK)
892 #endif
893                                         goto out;
894                 }
895                 flags = MAP_SHARED | MAP_FIXED;
896         } else {
897                 if ((shmflg & SHM_REMAP))
898                         goto out;
899
900                 flags = MAP_SHARED;
901         }
902
903         if (shmflg & SHM_RDONLY) {
904                 prot = PROT_READ;
905                 acc_mode = S_IRUGO;
906                 f_mode = FMODE_READ;
907         } else {
908                 prot = PROT_READ | PROT_WRITE;
909                 acc_mode = S_IRUGO | S_IWUGO;
910                 f_mode = FMODE_READ | FMODE_WRITE;
911         }
912         if (shmflg & SHM_EXEC) {
913                 prot |= PROT_EXEC;
914                 acc_mode |= S_IXUGO;
915         }
916
917         /*
918          * We cannot rely on the fs check since SYSV IPC does have an
919          * additional creator id...
920          */
921         ns = current->nsproxy->ipc_ns;
922         shp = shm_lock_check(ns, shmid);
923         if (IS_ERR(shp)) {
924                 err = PTR_ERR(shp);
925                 goto out;
926         }
927
928         err = -EACCES;
929         if (ipcperms(&shp->shm_perm, acc_mode))
930                 goto out_unlock;
931
932         err = security_shm_shmat(shp, shmaddr, shmflg);
933         if (err)
934                 goto out_unlock;
935
936         path.dentry = dget(shp->shm_file->f_path.dentry);
937         path.mnt    = shp->shm_file->f_path.mnt;
938         shp->shm_nattch++;
939         size = i_size_read(path.dentry->d_inode);
940         shm_unlock(shp);
941
942         err = -ENOMEM;
943         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
944         if (!sfd)
945                 goto out_put_dentry;
946
947         err = -ENOMEM;
948
949         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
950         if (!file)
951                 goto out_free;
952
953         file->private_data = sfd;
954         file->f_mapping = shp->shm_file->f_mapping;
955         sfd->id = shp->shm_perm.id;
956         sfd->ns = get_ipc_ns(ns);
957         sfd->file = shp->shm_file;
958         sfd->vm_ops = NULL;
959
960         down_write(&current->mm->mmap_sem);
961         if (addr && !(shmflg & SHM_REMAP)) {
962                 err = -EINVAL;
963                 if (find_vma_intersection(current->mm, addr, addr + size))
964                         goto invalid;
965                 /*
966                  * If shm segment goes below stack, make sure there is some
967                  * space left for the stack to grow (at least 4 pages).
968                  */
969                 if (addr < current->mm->start_stack &&
970                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
971                         goto invalid;
972         }
973                 
974         user_addr = do_mmap (file, addr, size, prot, flags, 0);
975         *raddr = user_addr;
976         err = 0;
977         if (IS_ERR_VALUE(user_addr))
978                 err = (long)user_addr;
979 invalid:
980         up_write(&current->mm->mmap_sem);
981
982         fput(file);
983
984 out_nattch:
985         down_write(&shm_ids(ns).rw_mutex);
986         shp = shm_lock_down(ns, shmid);
987         BUG_ON(IS_ERR(shp));
988         shp->shm_nattch--;
989         if(shp->shm_nattch == 0 &&
990            shp->shm_perm.mode & SHM_DEST)
991                 shm_destroy(ns, shp);
992         else
993                 shm_unlock(shp);
994         up_write(&shm_ids(ns).rw_mutex);
995
996 out:
997         return err;
998
999 out_unlock:
1000         shm_unlock(shp);
1001         goto out;
1002
1003 out_free:
1004         kfree(sfd);
1005 out_put_dentry:
1006         dput(path.dentry);
1007         goto out_nattch;
1008 }
1009
1010 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1011 {
1012         unsigned long ret;
1013         long err;
1014
1015         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1016         if (err)
1017                 return err;
1018         force_successful_syscall_return();
1019         return (long)ret;
1020 }
1021
1022 /*
1023  * detach and kill segment if marked destroyed.
1024  * The work is done in shm_close.
1025  */
1026 asmlinkage long sys_shmdt(char __user *shmaddr)
1027 {
1028         struct mm_struct *mm = current->mm;
1029         struct vm_area_struct *vma, *next;
1030         unsigned long addr = (unsigned long)shmaddr;
1031         loff_t size = 0;
1032         int retval = -EINVAL;
1033
1034         if (addr & ~PAGE_MASK)
1035                 return retval;
1036
1037         down_write(&mm->mmap_sem);
1038
1039         /*
1040          * This function tries to be smart and unmap shm segments that
1041          * were modified by partial mlock or munmap calls:
1042          * - It first determines the size of the shm segment that should be
1043          *   unmapped: It searches for a vma that is backed by shm and that
1044          *   started at address shmaddr. It records it's size and then unmaps
1045          *   it.
1046          * - Then it unmaps all shm vmas that started at shmaddr and that
1047          *   are within the initially determined size.
1048          * Errors from do_munmap are ignored: the function only fails if
1049          * it's called with invalid parameters or if it's called to unmap
1050          * a part of a vma. Both calls in this function are for full vmas,
1051          * the parameters are directly copied from the vma itself and always
1052          * valid - therefore do_munmap cannot fail. (famous last words?)
1053          */
1054         /*
1055          * If it had been mremap()'d, the starting address would not
1056          * match the usual checks anyway. So assume all vma's are
1057          * above the starting address given.
1058          */
1059         vma = find_vma(mm, addr);
1060
1061         while (vma) {
1062                 next = vma->vm_next;
1063
1064                 /*
1065                  * Check if the starting address would match, i.e. it's
1066                  * a fragment created by mprotect() and/or munmap(), or it
1067                  * otherwise it starts at this address with no hassles.
1068                  */
1069                 if ((vma->vm_ops == &shm_vm_ops) &&
1070                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1071
1072
1073                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1074                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1075                         /*
1076                          * We discovered the size of the shm segment, so
1077                          * break out of here and fall through to the next
1078                          * loop that uses the size information to stop
1079                          * searching for matching vma's.
1080                          */
1081                         retval = 0;
1082                         vma = next;
1083                         break;
1084                 }
1085                 vma = next;
1086         }
1087
1088         /*
1089          * We need look no further than the maximum address a fragment
1090          * could possibly have landed at. Also cast things to loff_t to
1091          * prevent overflows and make comparisions vs. equal-width types.
1092          */
1093         size = PAGE_ALIGN(size);
1094         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1095                 next = vma->vm_next;
1096
1097                 /* finding a matching vma now does not alter retval */
1098                 if ((vma->vm_ops == &shm_vm_ops) &&
1099                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1100
1101                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1102                 vma = next;
1103         }
1104
1105         up_write(&mm->mmap_sem);
1106         return retval;
1107 }
1108
1109 #ifdef CONFIG_PROC_FS
1110 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1111 {
1112         struct shmid_kernel *shp = it;
1113         char *format;
1114
1115 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1116 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1117
1118         if (sizeof(size_t) <= sizeof(int))
1119                 format = SMALL_STRING;
1120         else
1121                 format = BIG_STRING;
1122         return seq_printf(s, format,
1123                           shp->shm_perm.key,
1124                           shp->shm_perm.id,
1125                           shp->shm_perm.mode,
1126                           shp->shm_segsz,
1127                           shp->shm_cprid,
1128                           shp->shm_lprid,
1129                           shp->shm_nattch,
1130                           shp->shm_perm.uid,
1131                           shp->shm_perm.gid,
1132                           shp->shm_perm.cuid,
1133                           shp->shm_perm.cgid,
1134                           shp->shm_atim,
1135                           shp->shm_dtim,
1136                           shp->shm_ctim);
1137 }
1138 #endif