Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / 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/mutex.h>
39 #include <linux/nsproxy.h>
40
41 #include <asm/uaccess.h>
42
43 #include "util.h"
44
45 static struct file_operations shm_file_operations;
46 static struct vm_operations_struct shm_vm_ops;
47
48 static struct ipc_ids init_shm_ids;
49
50 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
51
52 #define shm_lock(ns, id)                \
53         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
54 #define shm_unlock(shp)                 \
55         ipc_unlock(&(shp)->shm_perm)
56 #define shm_get(ns, id)                 \
57         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
58 #define shm_buildid(ns, id, seq)        \
59         ipc_buildid(&shm_ids(ns), id, seq)
60
61 static int newseg (struct ipc_namespace *ns, key_t key,
62                 int shmflg, size_t size);
63 static void shm_open (struct vm_area_struct *shmd);
64 static void shm_close (struct vm_area_struct *shmd);
65 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
66 #ifdef CONFIG_PROC_FS
67 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
68 #endif
69
70 static void __ipc_init __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
71 {
72         ns->ids[IPC_SHM_IDS] = ids;
73         ns->shm_ctlmax = SHMMAX;
74         ns->shm_ctlall = SHMALL;
75         ns->shm_ctlmni = SHMMNI;
76         ns->shm_tot = 0;
77         ipc_init_ids(ids, 1);
78 }
79
80 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
81 {
82         if (shp->shm_nattch){
83                 shp->shm_perm.mode |= SHM_DEST;
84                 /* Do not find it any more */
85                 shp->shm_perm.key = IPC_PRIVATE;
86                 shm_unlock(shp);
87         } else
88                 shm_destroy(ns, shp);
89 }
90
91 #ifdef CONFIG_IPC_NS
92 int shm_init_ns(struct ipc_namespace *ns)
93 {
94         struct ipc_ids *ids;
95
96         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
97         if (ids == NULL)
98                 return -ENOMEM;
99
100         __shm_init_ns(ns, ids);
101         return 0;
102 }
103
104 void shm_exit_ns(struct ipc_namespace *ns)
105 {
106         int i;
107         struct shmid_kernel *shp;
108
109         mutex_lock(&shm_ids(ns).mutex);
110         for (i = 0; i <= shm_ids(ns).max_id; i++) {
111                 shp = shm_lock(ns, i);
112                 if (shp == NULL)
113                         continue;
114
115                 do_shm_rmid(ns, shp);
116         }
117         mutex_unlock(&shm_ids(ns).mutex);
118
119         ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
120         kfree(ns->ids[IPC_SHM_IDS]);
121         ns->ids[IPC_SHM_IDS] = NULL;
122 }
123 #endif
124
125 void __init shm_init (void)
126 {
127         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
128         ipc_init_proc_interface("sysvipc/shm",
129                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
130                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
131 }
132
133 static inline int shm_checkid(struct ipc_namespace *ns,
134                 struct shmid_kernel *s, int id)
135 {
136         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
137                 return -EIDRM;
138         return 0;
139 }
140
141 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
142 {
143         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
144 }
145
146 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
147 {
148         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
149 }
150
151
152
153 static inline void shm_inc(struct ipc_namespace *ns, int id)
154 {
155         struct shmid_kernel *shp;
156
157         shp = shm_lock(ns, id);
158         BUG_ON(!shp);
159         shp->shm_atim = get_seconds();
160         shp->shm_lprid = current->tgid;
161         shp->shm_nattch++;
162         shm_unlock(shp);
163 }
164
165 #define shm_file_ns(file) (*((struct ipc_namespace **)&(file)->private_data))
166
167 /* This is called by fork, once for every shm attach. */
168 static void shm_open(struct vm_area_struct *shmd)
169 {
170         shm_inc(shm_file_ns(shmd->vm_file),
171                         shmd->vm_file->f_dentry->d_inode->i_ino);
172 }
173
174 /*
175  * shm_destroy - free the struct shmid_kernel
176  *
177  * @shp: struct to free
178  *
179  * It has to be called with shp and shm_ids.mutex locked,
180  * but returns with shp unlocked and freed.
181  */
182 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
183 {
184         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
185         shm_rmid(ns, shp->id);
186         shm_unlock(shp);
187         if (!is_file_hugepages(shp->shm_file))
188                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
189         else
190                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
191                                                 shp->mlock_user);
192         fput (shp->shm_file);
193         security_shm_free(shp);
194         ipc_rcu_putref(shp);
195 }
196
197 /*
198  * remove the attach descriptor shmd.
199  * free memory for segment if it is marked destroyed.
200  * The descriptor has already been removed from the current->mm->mmap list
201  * and will later be kfree()d.
202  */
203 static void shm_close (struct vm_area_struct *shmd)
204 {
205         struct file * file = shmd->vm_file;
206         int id = file->f_dentry->d_inode->i_ino;
207         struct shmid_kernel *shp;
208         struct ipc_namespace *ns;
209
210         ns = shm_file_ns(file);
211
212         mutex_lock(&shm_ids(ns).mutex);
213         /* remove from the list of attaches of the shm segment */
214         shp = shm_lock(ns, id);
215         BUG_ON(!shp);
216         shp->shm_lprid = current->tgid;
217         shp->shm_dtim = get_seconds();
218         shp->shm_nattch--;
219         if(shp->shm_nattch == 0 &&
220            shp->shm_perm.mode & SHM_DEST)
221                 shm_destroy(ns, shp);
222         else
223                 shm_unlock(shp);
224         mutex_unlock(&shm_ids(ns).mutex);
225 }
226
227 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
228 {
229         int ret;
230
231         ret = shmem_mmap(file, vma);
232         if (ret == 0) {
233                 vma->vm_ops = &shm_vm_ops;
234                 if (!(vma->vm_flags & VM_WRITE))
235                         vma->vm_flags &= ~VM_MAYWRITE;
236                 shm_inc(shm_file_ns(file), file->f_dentry->d_inode->i_ino);
237         }
238
239         return ret;
240 }
241
242 static int shm_release(struct inode *ino, struct file *file)
243 {
244         struct ipc_namespace *ns;
245
246         ns = shm_file_ns(file);
247         put_ipc_ns(ns);
248         shm_file_ns(file) = NULL;
249         return 0;
250 }
251
252 static struct file_operations shm_file_operations = {
253         .mmap           = shm_mmap,
254         .release        = shm_release,
255 #ifndef CONFIG_MMU
256         .get_unmapped_area = shmem_get_unmapped_area,
257 #endif
258 };
259
260 static struct vm_operations_struct shm_vm_ops = {
261         .open   = shm_open,     /* callback for a new vm-area open */
262         .close  = shm_close,    /* callback for when the vm-area is released */
263         .nopage = shmem_nopage,
264 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
265         .set_policy = shmem_set_policy,
266         .get_policy = shmem_get_policy,
267 #endif
268 };
269
270 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
271 {
272         int error;
273         struct shmid_kernel *shp;
274         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
275         struct file * file;
276         char name[13];
277         int id;
278
279         if (size < SHMMIN || size > ns->shm_ctlmax)
280                 return -EINVAL;
281
282         if (ns->shm_tot + numpages >= ns->shm_ctlall)
283                 return -ENOSPC;
284
285         shp = ipc_rcu_alloc(sizeof(*shp));
286         if (!shp)
287                 return -ENOMEM;
288
289         shp->shm_perm.key = key;
290         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
291         shp->mlock_user = NULL;
292
293         shp->shm_perm.security = NULL;
294         error = security_shm_alloc(shp);
295         if (error) {
296                 ipc_rcu_putref(shp);
297                 return error;
298         }
299
300         if (shmflg & SHM_HUGETLB) {
301                 /* hugetlb_zero_setup takes care of mlock user accounting */
302                 file = hugetlb_zero_setup(size);
303                 shp->mlock_user = current->user;
304         } else {
305                 int acctflag = VM_ACCOUNT;
306                 /*
307                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
308                  * if it's asked for.
309                  */
310                 if  ((shmflg & SHM_NORESERVE) &&
311                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
312                         acctflag = 0;
313                 sprintf (name, "SYSV%08x", key);
314                 file = shmem_file_setup(name, size, acctflag);
315         }
316         error = PTR_ERR(file);
317         if (IS_ERR(file))
318                 goto no_file;
319
320         error = -ENOSPC;
321         id = shm_addid(ns, shp);
322         if(id == -1) 
323                 goto no_id;
324
325         shp->shm_cprid = current->tgid;
326         shp->shm_lprid = 0;
327         shp->shm_atim = shp->shm_dtim = 0;
328         shp->shm_ctim = get_seconds();
329         shp->shm_segsz = size;
330         shp->shm_nattch = 0;
331         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
332         shp->shm_file = file;
333         file->f_dentry->d_inode->i_ino = shp->id;
334
335         shm_file_ns(file) = get_ipc_ns(ns);
336
337         /* Hugetlb ops would have already been assigned. */
338         if (!(shmflg & SHM_HUGETLB))
339                 file->f_op = &shm_file_operations;
340
341         ns->shm_tot += numpages;
342         shm_unlock(shp);
343         return shp->id;
344
345 no_id:
346         fput(file);
347 no_file:
348         security_shm_free(shp);
349         ipc_rcu_putref(shp);
350         return error;
351 }
352
353 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
354 {
355         struct shmid_kernel *shp;
356         int err, id = 0;
357         struct ipc_namespace *ns;
358
359         ns = current->nsproxy->ipc_ns;
360
361         mutex_lock(&shm_ids(ns).mutex);
362         if (key == IPC_PRIVATE) {
363                 err = newseg(ns, key, shmflg, size);
364         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
365                 if (!(shmflg & IPC_CREAT))
366                         err = -ENOENT;
367                 else
368                         err = newseg(ns, key, shmflg, size);
369         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
370                 err = -EEXIST;
371         } else {
372                 shp = shm_lock(ns, id);
373                 BUG_ON(shp==NULL);
374                 if (shp->shm_segsz < size)
375                         err = -EINVAL;
376                 else if (ipcperms(&shp->shm_perm, shmflg))
377                         err = -EACCES;
378                 else {
379                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
380                         err = security_shm_associate(shp, shmflg);
381                         if (!err)
382                                 err = shmid;
383                 }
384                 shm_unlock(shp);
385         }
386         mutex_unlock(&shm_ids(ns).mutex);
387
388         return err;
389 }
390
391 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
392 {
393         switch(version) {
394         case IPC_64:
395                 return copy_to_user(buf, in, sizeof(*in));
396         case IPC_OLD:
397             {
398                 struct shmid_ds out;
399
400                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
401                 out.shm_segsz   = in->shm_segsz;
402                 out.shm_atime   = in->shm_atime;
403                 out.shm_dtime   = in->shm_dtime;
404                 out.shm_ctime   = in->shm_ctime;
405                 out.shm_cpid    = in->shm_cpid;
406                 out.shm_lpid    = in->shm_lpid;
407                 out.shm_nattch  = in->shm_nattch;
408
409                 return copy_to_user(buf, &out, sizeof(out));
410             }
411         default:
412                 return -EINVAL;
413         }
414 }
415
416 struct shm_setbuf {
417         uid_t   uid;
418         gid_t   gid;
419         mode_t  mode;
420 };      
421
422 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
423 {
424         switch(version) {
425         case IPC_64:
426             {
427                 struct shmid64_ds tbuf;
428
429                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
430                         return -EFAULT;
431
432                 out->uid        = tbuf.shm_perm.uid;
433                 out->gid        = tbuf.shm_perm.gid;
434                 out->mode       = tbuf.shm_perm.mode;
435
436                 return 0;
437             }
438         case IPC_OLD:
439             {
440                 struct shmid_ds tbuf_old;
441
442                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
443                         return -EFAULT;
444
445                 out->uid        = tbuf_old.shm_perm.uid;
446                 out->gid        = tbuf_old.shm_perm.gid;
447                 out->mode       = tbuf_old.shm_perm.mode;
448
449                 return 0;
450             }
451         default:
452                 return -EINVAL;
453         }
454 }
455
456 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
457 {
458         switch(version) {
459         case IPC_64:
460                 return copy_to_user(buf, in, sizeof(*in));
461         case IPC_OLD:
462             {
463                 struct shminfo out;
464
465                 if(in->shmmax > INT_MAX)
466                         out.shmmax = INT_MAX;
467                 else
468                         out.shmmax = (int)in->shmmax;
469
470                 out.shmmin      = in->shmmin;
471                 out.shmmni      = in->shmmni;
472                 out.shmseg      = in->shmseg;
473                 out.shmall      = in->shmall; 
474
475                 return copy_to_user(buf, &out, sizeof(out));
476             }
477         default:
478                 return -EINVAL;
479         }
480 }
481
482 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
483                 unsigned long *swp)
484 {
485         int i;
486
487         *rss = 0;
488         *swp = 0;
489
490         for (i = 0; i <= shm_ids(ns).max_id; i++) {
491                 struct shmid_kernel *shp;
492                 struct inode *inode;
493
494                 shp = shm_get(ns, i);
495                 if(!shp)
496                         continue;
497
498                 inode = shp->shm_file->f_dentry->d_inode;
499
500                 if (is_file_hugepages(shp->shm_file)) {
501                         struct address_space *mapping = inode->i_mapping;
502                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
503                 } else {
504                         struct shmem_inode_info *info = SHMEM_I(inode);
505                         spin_lock(&info->lock);
506                         *rss += inode->i_mapping->nrpages;
507                         *swp += info->swapped;
508                         spin_unlock(&info->lock);
509                 }
510         }
511 }
512
513 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
514 {
515         struct shm_setbuf setbuf;
516         struct shmid_kernel *shp;
517         int err, version;
518         struct ipc_namespace *ns;
519
520         if (cmd < 0 || shmid < 0) {
521                 err = -EINVAL;
522                 goto out;
523         }
524
525         version = ipc_parse_version(&cmd);
526         ns = current->nsproxy->ipc_ns;
527
528         switch (cmd) { /* replace with proc interface ? */
529         case IPC_INFO:
530         {
531                 struct shminfo64 shminfo;
532
533                 err = security_shm_shmctl(NULL, cmd);
534                 if (err)
535                         return err;
536
537                 memset(&shminfo,0,sizeof(shminfo));
538                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
539                 shminfo.shmmax = ns->shm_ctlmax;
540                 shminfo.shmall = ns->shm_ctlall;
541
542                 shminfo.shmmin = SHMMIN;
543                 if(copy_shminfo_to_user (buf, &shminfo, version))
544                         return -EFAULT;
545                 /* reading a integer is always atomic */
546                 err= shm_ids(ns).max_id;
547                 if(err<0)
548                         err = 0;
549                 goto out;
550         }
551         case SHM_INFO:
552         {
553                 struct shm_info shm_info;
554
555                 err = security_shm_shmctl(NULL, cmd);
556                 if (err)
557                         return err;
558
559                 memset(&shm_info,0,sizeof(shm_info));
560                 mutex_lock(&shm_ids(ns).mutex);
561                 shm_info.used_ids = shm_ids(ns).in_use;
562                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
563                 shm_info.shm_tot = ns->shm_tot;
564                 shm_info.swap_attempts = 0;
565                 shm_info.swap_successes = 0;
566                 err = shm_ids(ns).max_id;
567                 mutex_unlock(&shm_ids(ns).mutex);
568                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
569                         err = -EFAULT;
570                         goto out;
571                 }
572
573                 err = err < 0 ? 0 : err;
574                 goto out;
575         }
576         case SHM_STAT:
577         case IPC_STAT:
578         {
579                 struct shmid64_ds tbuf;
580                 int result;
581                 memset(&tbuf, 0, sizeof(tbuf));
582                 shp = shm_lock(ns, shmid);
583                 if(shp==NULL) {
584                         err = -EINVAL;
585                         goto out;
586                 } else if(cmd==SHM_STAT) {
587                         err = -EINVAL;
588                         if (shmid > shm_ids(ns).max_id)
589                                 goto out_unlock;
590                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
591                 } else {
592                         err = shm_checkid(ns, shp,shmid);
593                         if(err)
594                                 goto out_unlock;
595                         result = 0;
596                 }
597                 err=-EACCES;
598                 if (ipcperms (&shp->shm_perm, S_IRUGO))
599                         goto out_unlock;
600                 err = security_shm_shmctl(shp, cmd);
601                 if (err)
602                         goto out_unlock;
603                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
604                 tbuf.shm_segsz  = shp->shm_segsz;
605                 tbuf.shm_atime  = shp->shm_atim;
606                 tbuf.shm_dtime  = shp->shm_dtim;
607                 tbuf.shm_ctime  = shp->shm_ctim;
608                 tbuf.shm_cpid   = shp->shm_cprid;
609                 tbuf.shm_lpid   = shp->shm_lprid;
610                 if (!is_file_hugepages(shp->shm_file))
611                         tbuf.shm_nattch = shp->shm_nattch;
612                 else
613                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
614                 shm_unlock(shp);
615                 if(copy_shmid_to_user (buf, &tbuf, version))
616                         err = -EFAULT;
617                 else
618                         err = result;
619                 goto out;
620         }
621         case SHM_LOCK:
622         case SHM_UNLOCK:
623         {
624                 shp = shm_lock(ns, shmid);
625                 if(shp==NULL) {
626                         err = -EINVAL;
627                         goto out;
628                 }
629                 err = shm_checkid(ns, shp,shmid);
630                 if(err)
631                         goto out_unlock;
632
633                 err = audit_ipc_obj(&(shp->shm_perm));
634                 if (err)
635                         goto out_unlock;
636
637                 if (!capable(CAP_IPC_LOCK)) {
638                         err = -EPERM;
639                         if (current->euid != shp->shm_perm.uid &&
640                             current->euid != shp->shm_perm.cuid)
641                                 goto out_unlock;
642                         if (cmd == SHM_LOCK &&
643                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
644                                 goto out_unlock;
645                 }
646
647                 err = security_shm_shmctl(shp, cmd);
648                 if (err)
649                         goto out_unlock;
650                 
651                 if(cmd==SHM_LOCK) {
652                         struct user_struct * user = current->user;
653                         if (!is_file_hugepages(shp->shm_file)) {
654                                 err = shmem_lock(shp->shm_file, 1, user);
655                                 if (!err) {
656                                         shp->shm_perm.mode |= SHM_LOCKED;
657                                         shp->mlock_user = user;
658                                 }
659                         }
660                 } else if (!is_file_hugepages(shp->shm_file)) {
661                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
662                         shp->shm_perm.mode &= ~SHM_LOCKED;
663                         shp->mlock_user = NULL;
664                 }
665                 shm_unlock(shp);
666                 goto out;
667         }
668         case IPC_RMID:
669         {
670                 /*
671                  *      We cannot simply remove the file. The SVID states
672                  *      that the block remains until the last person
673                  *      detaches from it, then is deleted. A shmat() on
674                  *      an RMID segment is legal in older Linux and if 
675                  *      we change it apps break...
676                  *
677                  *      Instead we set a destroyed flag, and then blow
678                  *      the name away when the usage hits zero.
679                  */
680                 mutex_lock(&shm_ids(ns).mutex);
681                 shp = shm_lock(ns, shmid);
682                 err = -EINVAL;
683                 if (shp == NULL) 
684                         goto out_up;
685                 err = shm_checkid(ns, shp, shmid);
686                 if(err)
687                         goto out_unlock_up;
688
689                 err = audit_ipc_obj(&(shp->shm_perm));
690                 if (err)
691                         goto out_unlock_up;
692
693                 if (current->euid != shp->shm_perm.uid &&
694                     current->euid != shp->shm_perm.cuid && 
695                     !capable(CAP_SYS_ADMIN)) {
696                         err=-EPERM;
697                         goto out_unlock_up;
698                 }
699
700                 err = security_shm_shmctl(shp, cmd);
701                 if (err)
702                         goto out_unlock_up;
703
704                 do_shm_rmid(ns, shp);
705                 mutex_unlock(&shm_ids(ns).mutex);
706                 goto out;
707         }
708
709         case IPC_SET:
710         {
711                 if (copy_shmid_from_user (&setbuf, buf, version)) {
712                         err = -EFAULT;
713                         goto out;
714                 }
715                 mutex_lock(&shm_ids(ns).mutex);
716                 shp = shm_lock(ns, shmid);
717                 err=-EINVAL;
718                 if(shp==NULL)
719                         goto out_up;
720                 err = shm_checkid(ns, shp,shmid);
721                 if(err)
722                         goto out_unlock_up;
723                 err = audit_ipc_obj(&(shp->shm_perm));
724                 if (err)
725                         goto out_unlock_up;
726                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
727                 if (err)
728                         goto out_unlock_up;
729                 err=-EPERM;
730                 if (current->euid != shp->shm_perm.uid &&
731                     current->euid != shp->shm_perm.cuid && 
732                     !capable(CAP_SYS_ADMIN)) {
733                         goto out_unlock_up;
734                 }
735
736                 err = security_shm_shmctl(shp, cmd);
737                 if (err)
738                         goto out_unlock_up;
739                 
740                 shp->shm_perm.uid = setbuf.uid;
741                 shp->shm_perm.gid = setbuf.gid;
742                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
743                         | (setbuf.mode & S_IRWXUGO);
744                 shp->shm_ctim = get_seconds();
745                 break;
746         }
747
748         default:
749                 err = -EINVAL;
750                 goto out;
751         }
752
753         err = 0;
754 out_unlock_up:
755         shm_unlock(shp);
756 out_up:
757         mutex_unlock(&shm_ids(ns).mutex);
758         goto out;
759 out_unlock:
760         shm_unlock(shp);
761 out:
762         return err;
763 }
764
765 /*
766  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
767  *
768  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
769  * "raddr" thing points to kernel space, and there has to be a wrapper around
770  * this.
771  */
772 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
773 {
774         struct shmid_kernel *shp;
775         unsigned long addr;
776         unsigned long size;
777         struct file * file;
778         int    err;
779         unsigned long flags;
780         unsigned long prot;
781         int acc_mode;
782         void *user_addr;
783         struct ipc_namespace *ns;
784
785         if (shmid < 0) {
786                 err = -EINVAL;
787                 goto out;
788         } else if ((addr = (ulong)shmaddr)) {
789                 if (addr & (SHMLBA-1)) {
790                         if (shmflg & SHM_RND)
791                                 addr &= ~(SHMLBA-1);       /* round down */
792                         else
793 #ifndef __ARCH_FORCE_SHMLBA
794                                 if (addr & ~PAGE_MASK)
795 #endif
796                                         return -EINVAL;
797                 }
798                 flags = MAP_SHARED | MAP_FIXED;
799         } else {
800                 if ((shmflg & SHM_REMAP))
801                         return -EINVAL;
802
803                 flags = MAP_SHARED;
804         }
805
806         if (shmflg & SHM_RDONLY) {
807                 prot = PROT_READ;
808                 acc_mode = S_IRUGO;
809         } else {
810                 prot = PROT_READ | PROT_WRITE;
811                 acc_mode = S_IRUGO | S_IWUGO;
812         }
813         if (shmflg & SHM_EXEC) {
814                 prot |= PROT_EXEC;
815                 acc_mode |= S_IXUGO;
816         }
817
818         /*
819          * We cannot rely on the fs check since SYSV IPC does have an
820          * additional creator id...
821          */
822         ns = current->nsproxy->ipc_ns;
823         shp = shm_lock(ns, shmid);
824         if(shp == NULL) {
825                 err = -EINVAL;
826                 goto out;
827         }
828         err = shm_checkid(ns, shp,shmid);
829         if (err) {
830                 shm_unlock(shp);
831                 goto out;
832         }
833         if (ipcperms(&shp->shm_perm, acc_mode)) {
834                 shm_unlock(shp);
835                 err = -EACCES;
836                 goto out;
837         }
838
839         err = security_shm_shmat(shp, shmaddr, shmflg);
840         if (err) {
841                 shm_unlock(shp);
842                 return err;
843         }
844                 
845         file = shp->shm_file;
846         size = i_size_read(file->f_dentry->d_inode);
847         shp->shm_nattch++;
848         shm_unlock(shp);
849
850         down_write(&current->mm->mmap_sem);
851         if (addr && !(shmflg & SHM_REMAP)) {
852                 user_addr = ERR_PTR(-EINVAL);
853                 if (find_vma_intersection(current->mm, addr, addr + size))
854                         goto invalid;
855                 /*
856                  * If shm segment goes below stack, make sure there is some
857                  * space left for the stack to grow (at least 4 pages).
858                  */
859                 if (addr < current->mm->start_stack &&
860                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
861                         goto invalid;
862         }
863                 
864         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
865
866 invalid:
867         up_write(&current->mm->mmap_sem);
868
869         mutex_lock(&shm_ids(ns).mutex);
870         shp = shm_lock(ns, shmid);
871         BUG_ON(!shp);
872         shp->shm_nattch--;
873         if(shp->shm_nattch == 0 &&
874            shp->shm_perm.mode & SHM_DEST)
875                 shm_destroy(ns, shp);
876         else
877                 shm_unlock(shp);
878         mutex_unlock(&shm_ids(ns).mutex);
879
880         *raddr = (unsigned long) user_addr;
881         err = 0;
882         if (IS_ERR(user_addr))
883                 err = PTR_ERR(user_addr);
884 out:
885         return err;
886 }
887
888 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
889 {
890         unsigned long ret;
891         long err;
892
893         err = do_shmat(shmid, shmaddr, shmflg, &ret);
894         if (err)
895                 return err;
896         force_successful_syscall_return();
897         return (long)ret;
898 }
899
900 /*
901  * detach and kill segment if marked destroyed.
902  * The work is done in shm_close.
903  */
904 asmlinkage long sys_shmdt(char __user *shmaddr)
905 {
906         struct mm_struct *mm = current->mm;
907         struct vm_area_struct *vma, *next;
908         unsigned long addr = (unsigned long)shmaddr;
909         loff_t size = 0;
910         int retval = -EINVAL;
911
912         if (addr & ~PAGE_MASK)
913                 return retval;
914
915         down_write(&mm->mmap_sem);
916
917         /*
918          * This function tries to be smart and unmap shm segments that
919          * were modified by partial mlock or munmap calls:
920          * - It first determines the size of the shm segment that should be
921          *   unmapped: It searches for a vma that is backed by shm and that
922          *   started at address shmaddr. It records it's size and then unmaps
923          *   it.
924          * - Then it unmaps all shm vmas that started at shmaddr and that
925          *   are within the initially determined size.
926          * Errors from do_munmap are ignored: the function only fails if
927          * it's called with invalid parameters or if it's called to unmap
928          * a part of a vma. Both calls in this function are for full vmas,
929          * the parameters are directly copied from the vma itself and always
930          * valid - therefore do_munmap cannot fail. (famous last words?)
931          */
932         /*
933          * If it had been mremap()'d, the starting address would not
934          * match the usual checks anyway. So assume all vma's are
935          * above the starting address given.
936          */
937         vma = find_vma(mm, addr);
938
939         while (vma) {
940                 next = vma->vm_next;
941
942                 /*
943                  * Check if the starting address would match, i.e. it's
944                  * a fragment created by mprotect() and/or munmap(), or it
945                  * otherwise it starts at this address with no hassles.
946                  */
947                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
948                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
949
950
951                         size = vma->vm_file->f_dentry->d_inode->i_size;
952                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
953                         /*
954                          * We discovered the size of the shm segment, so
955                          * break out of here and fall through to the next
956                          * loop that uses the size information to stop
957                          * searching for matching vma's.
958                          */
959                         retval = 0;
960                         vma = next;
961                         break;
962                 }
963                 vma = next;
964         }
965
966         /*
967          * We need look no further than the maximum address a fragment
968          * could possibly have landed at. Also cast things to loff_t to
969          * prevent overflows and make comparisions vs. equal-width types.
970          */
971         size = PAGE_ALIGN(size);
972         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
973                 next = vma->vm_next;
974
975                 /* finding a matching vma now does not alter retval */
976                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
977                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
978
979                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
980                 vma = next;
981         }
982
983         up_write(&mm->mmap_sem);
984         return retval;
985 }
986
987 #ifdef CONFIG_PROC_FS
988 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
989 {
990         struct shmid_kernel *shp = it;
991         char *format;
992
993 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
994 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
995
996         if (sizeof(size_t) <= sizeof(int))
997                 format = SMALL_STRING;
998         else
999                 format = BIG_STRING;
1000         return seq_printf(s, format,
1001                           shp->shm_perm.key,
1002                           shp->id,
1003                           shp->shm_perm.mode,
1004                           shp->shm_segsz,
1005                           shp->shm_cprid,
1006                           shp->shm_lprid,
1007                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
1008                           shp->shm_perm.uid,
1009                           shp->shm_perm.gid,
1010                           shp->shm_perm.cuid,
1011                           shp->shm_perm.cgid,
1012                           shp->shm_atim,
1013                           shp->shm_dtim,
1014                           shp->shm_ctim);
1015 }
1016 #endif