[PATCH] update of IPC audit record cleanup
[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
20 #include <linux/config.h>
21 #include <linux/slab.h>
22 #include <linux/mm.h>
23 #include <linux/hugetlb.h>
24 #include <linux/shm.h>
25 #include <linux/init.h>
26 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/shmem_fs.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/ptrace.h>
34 #include <linux/seq_file.h>
35 #include <linux/mutex.h>
36
37 #include <asm/uaccess.h>
38
39 #include "util.h"
40
41 static struct file_operations shm_file_operations;
42 static struct vm_operations_struct shm_vm_ops;
43
44 static struct ipc_ids shm_ids;
45
46 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
47 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
48 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
49 #define shm_buildid(id, seq) \
50         ipc_buildid(&shm_ids, id, seq)
51
52 static int newseg (key_t key, int shmflg, size_t size);
53 static void shm_open (struct vm_area_struct *shmd);
54 static void shm_close (struct vm_area_struct *shmd);
55 #ifdef CONFIG_PROC_FS
56 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
57 #endif
58
59 size_t  shm_ctlmax = SHMMAX;
60 size_t  shm_ctlall = SHMALL;
61 int     shm_ctlmni = SHMMNI;
62
63 static int shm_tot; /* total number of shared memory pages */
64
65 void __init shm_init (void)
66 {
67         ipc_init_ids(&shm_ids, 1);
68         ipc_init_proc_interface("sysvipc/shm",
69                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
70                                 &shm_ids,
71                                 sysvipc_shm_proc_show);
72 }
73
74 static inline int shm_checkid(struct shmid_kernel *s, int id)
75 {
76         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
77                 return -EIDRM;
78         return 0;
79 }
80
81 static inline struct shmid_kernel *shm_rmid(int id)
82 {
83         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
84 }
85
86 static inline int shm_addid(struct shmid_kernel *shp)
87 {
88         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
89 }
90
91
92
93 static inline void shm_inc (int id) {
94         struct shmid_kernel *shp;
95
96         shp = shm_lock(id);
97         BUG_ON(!shp);
98         shp->shm_atim = get_seconds();
99         shp->shm_lprid = current->tgid;
100         shp->shm_nattch++;
101         shm_unlock(shp);
102 }
103
104 /* This is called by fork, once for every shm attach. */
105 static void shm_open (struct vm_area_struct *shmd)
106 {
107         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
108 }
109
110 /*
111  * shm_destroy - free the struct shmid_kernel
112  *
113  * @shp: struct to free
114  *
115  * It has to be called with shp and shm_ids.mutex locked,
116  * but returns with shp unlocked and freed.
117  */
118 static void shm_destroy (struct shmid_kernel *shp)
119 {
120         shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
121         shm_rmid (shp->id);
122         shm_unlock(shp);
123         if (!is_file_hugepages(shp->shm_file))
124                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
125         else
126                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
127                                                 shp->mlock_user);
128         fput (shp->shm_file);
129         security_shm_free(shp);
130         ipc_rcu_putref(shp);
131 }
132
133 /*
134  * remove the attach descriptor shmd.
135  * free memory for segment if it is marked destroyed.
136  * The descriptor has already been removed from the current->mm->mmap list
137  * and will later be kfree()d.
138  */
139 static void shm_close (struct vm_area_struct *shmd)
140 {
141         struct file * file = shmd->vm_file;
142         int id = file->f_dentry->d_inode->i_ino;
143         struct shmid_kernel *shp;
144
145         mutex_lock(&shm_ids.mutex);
146         /* remove from the list of attaches of the shm segment */
147         shp = shm_lock(id);
148         BUG_ON(!shp);
149         shp->shm_lprid = current->tgid;
150         shp->shm_dtim = get_seconds();
151         shp->shm_nattch--;
152         if(shp->shm_nattch == 0 &&
153            shp->shm_perm.mode & SHM_DEST)
154                 shm_destroy (shp);
155         else
156                 shm_unlock(shp);
157         mutex_unlock(&shm_ids.mutex);
158 }
159
160 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
161 {
162         int ret;
163
164         ret = shmem_mmap(file, vma);
165         if (ret == 0) {
166                 vma->vm_ops = &shm_vm_ops;
167                 if (!(vma->vm_flags & VM_WRITE))
168                         vma->vm_flags &= ~VM_MAYWRITE;
169                 shm_inc(file->f_dentry->d_inode->i_ino);
170         }
171
172         return ret;
173 }
174
175 static struct file_operations shm_file_operations = {
176         .mmap   = shm_mmap,
177 #ifndef CONFIG_MMU
178         .get_unmapped_area = shmem_get_unmapped_area,
179 #endif
180 };
181
182 static struct vm_operations_struct shm_vm_ops = {
183         .open   = shm_open,     /* callback for a new vm-area open */
184         .close  = shm_close,    /* callback for when the vm-area is released */
185         .nopage = shmem_nopage,
186 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
187         .set_policy = shmem_set_policy,
188         .get_policy = shmem_get_policy,
189 #endif
190 };
191
192 static int newseg (key_t key, int shmflg, size_t size)
193 {
194         int error;
195         struct shmid_kernel *shp;
196         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
197         struct file * file;
198         char name[13];
199         int id;
200
201         if (size < SHMMIN || size > shm_ctlmax)
202                 return -EINVAL;
203
204         if (shm_tot + numpages >= shm_ctlall)
205                 return -ENOSPC;
206
207         shp = ipc_rcu_alloc(sizeof(*shp));
208         if (!shp)
209                 return -ENOMEM;
210
211         shp->shm_perm.key = key;
212         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
213         shp->mlock_user = NULL;
214
215         shp->shm_perm.security = NULL;
216         error = security_shm_alloc(shp);
217         if (error) {
218                 ipc_rcu_putref(shp);
219                 return error;
220         }
221
222         if (shmflg & SHM_HUGETLB) {
223                 /* hugetlb_zero_setup takes care of mlock user accounting */
224                 file = hugetlb_zero_setup(size);
225                 shp->mlock_user = current->user;
226         } else {
227                 int acctflag = VM_ACCOUNT;
228                 /*
229                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
230                  * if it's asked for.
231                  */
232                 if  ((shmflg & SHM_NORESERVE) &&
233                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
234                         acctflag = 0;
235                 sprintf (name, "SYSV%08x", key);
236                 file = shmem_file_setup(name, size, acctflag);
237         }
238         error = PTR_ERR(file);
239         if (IS_ERR(file))
240                 goto no_file;
241
242         error = -ENOSPC;
243         id = shm_addid(shp);
244         if(id == -1) 
245                 goto no_id;
246
247         shp->shm_cprid = current->tgid;
248         shp->shm_lprid = 0;
249         shp->shm_atim = shp->shm_dtim = 0;
250         shp->shm_ctim = get_seconds();
251         shp->shm_segsz = size;
252         shp->shm_nattch = 0;
253         shp->id = shm_buildid(id,shp->shm_perm.seq);
254         shp->shm_file = file;
255         file->f_dentry->d_inode->i_ino = shp->id;
256
257         /* Hugetlb ops would have already been assigned. */
258         if (!(shmflg & SHM_HUGETLB))
259                 file->f_op = &shm_file_operations;
260
261         shm_tot += numpages;
262         shm_unlock(shp);
263         return shp->id;
264
265 no_id:
266         fput(file);
267 no_file:
268         security_shm_free(shp);
269         ipc_rcu_putref(shp);
270         return error;
271 }
272
273 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
274 {
275         struct shmid_kernel *shp;
276         int err, id = 0;
277
278         mutex_lock(&shm_ids.mutex);
279         if (key == IPC_PRIVATE) {
280                 err = newseg(key, shmflg, size);
281         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
282                 if (!(shmflg & IPC_CREAT))
283                         err = -ENOENT;
284                 else
285                         err = newseg(key, shmflg, size);
286         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
287                 err = -EEXIST;
288         } else {
289                 shp = shm_lock(id);
290                 BUG_ON(shp==NULL);
291                 if (shp->shm_segsz < size)
292                         err = -EINVAL;
293                 else if (ipcperms(&shp->shm_perm, shmflg))
294                         err = -EACCES;
295                 else {
296                         int shmid = shm_buildid(id, shp->shm_perm.seq);
297                         err = security_shm_associate(shp, shmflg);
298                         if (!err)
299                                 err = shmid;
300                 }
301                 shm_unlock(shp);
302         }
303         mutex_unlock(&shm_ids.mutex);
304
305         return err;
306 }
307
308 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
309 {
310         switch(version) {
311         case IPC_64:
312                 return copy_to_user(buf, in, sizeof(*in));
313         case IPC_OLD:
314             {
315                 struct shmid_ds out;
316
317                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
318                 out.shm_segsz   = in->shm_segsz;
319                 out.shm_atime   = in->shm_atime;
320                 out.shm_dtime   = in->shm_dtime;
321                 out.shm_ctime   = in->shm_ctime;
322                 out.shm_cpid    = in->shm_cpid;
323                 out.shm_lpid    = in->shm_lpid;
324                 out.shm_nattch  = in->shm_nattch;
325
326                 return copy_to_user(buf, &out, sizeof(out));
327             }
328         default:
329                 return -EINVAL;
330         }
331 }
332
333 struct shm_setbuf {
334         uid_t   uid;
335         gid_t   gid;
336         mode_t  mode;
337 };      
338
339 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
340 {
341         switch(version) {
342         case IPC_64:
343             {
344                 struct shmid64_ds tbuf;
345
346                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
347                         return -EFAULT;
348
349                 out->uid        = tbuf.shm_perm.uid;
350                 out->gid        = tbuf.shm_perm.gid;
351                 out->mode       = tbuf.shm_perm.mode;
352
353                 return 0;
354             }
355         case IPC_OLD:
356             {
357                 struct shmid_ds tbuf_old;
358
359                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
360                         return -EFAULT;
361
362                 out->uid        = tbuf_old.shm_perm.uid;
363                 out->gid        = tbuf_old.shm_perm.gid;
364                 out->mode       = tbuf_old.shm_perm.mode;
365
366                 return 0;
367             }
368         default:
369                 return -EINVAL;
370         }
371 }
372
373 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
374 {
375         switch(version) {
376         case IPC_64:
377                 return copy_to_user(buf, in, sizeof(*in));
378         case IPC_OLD:
379             {
380                 struct shminfo out;
381
382                 if(in->shmmax > INT_MAX)
383                         out.shmmax = INT_MAX;
384                 else
385                         out.shmmax = (int)in->shmmax;
386
387                 out.shmmin      = in->shmmin;
388                 out.shmmni      = in->shmmni;
389                 out.shmseg      = in->shmseg;
390                 out.shmall      = in->shmall; 
391
392                 return copy_to_user(buf, &out, sizeof(out));
393             }
394         default:
395                 return -EINVAL;
396         }
397 }
398
399 static void shm_get_stat(unsigned long *rss, unsigned long *swp) 
400 {
401         int i;
402
403         *rss = 0;
404         *swp = 0;
405
406         for (i = 0; i <= shm_ids.max_id; i++) {
407                 struct shmid_kernel *shp;
408                 struct inode *inode;
409
410                 shp = shm_get(i);
411                 if(!shp)
412                         continue;
413
414                 inode = shp->shm_file->f_dentry->d_inode;
415
416                 if (is_file_hugepages(shp->shm_file)) {
417                         struct address_space *mapping = inode->i_mapping;
418                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
419                 } else {
420                         struct shmem_inode_info *info = SHMEM_I(inode);
421                         spin_lock(&info->lock);
422                         *rss += inode->i_mapping->nrpages;
423                         *swp += info->swapped;
424                         spin_unlock(&info->lock);
425                 }
426         }
427 }
428
429 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
430 {
431         struct shm_setbuf setbuf;
432         struct shmid_kernel *shp;
433         int err, version;
434
435         if (cmd < 0 || shmid < 0) {
436                 err = -EINVAL;
437                 goto out;
438         }
439
440         version = ipc_parse_version(&cmd);
441
442         switch (cmd) { /* replace with proc interface ? */
443         case IPC_INFO:
444         {
445                 struct shminfo64 shminfo;
446
447                 err = security_shm_shmctl(NULL, cmd);
448                 if (err)
449                         return err;
450
451                 memset(&shminfo,0,sizeof(shminfo));
452                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
453                 shminfo.shmmax = shm_ctlmax;
454                 shminfo.shmall = shm_ctlall;
455
456                 shminfo.shmmin = SHMMIN;
457                 if(copy_shminfo_to_user (buf, &shminfo, version))
458                         return -EFAULT;
459                 /* reading a integer is always atomic */
460                 err= shm_ids.max_id;
461                 if(err<0)
462                         err = 0;
463                 goto out;
464         }
465         case SHM_INFO:
466         {
467                 struct shm_info shm_info;
468
469                 err = security_shm_shmctl(NULL, cmd);
470                 if (err)
471                         return err;
472
473                 memset(&shm_info,0,sizeof(shm_info));
474                 mutex_lock(&shm_ids.mutex);
475                 shm_info.used_ids = shm_ids.in_use;
476                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
477                 shm_info.shm_tot = shm_tot;
478                 shm_info.swap_attempts = 0;
479                 shm_info.swap_successes = 0;
480                 err = shm_ids.max_id;
481                 mutex_unlock(&shm_ids.mutex);
482                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
483                         err = -EFAULT;
484                         goto out;
485                 }
486
487                 err = err < 0 ? 0 : err;
488                 goto out;
489         }
490         case SHM_STAT:
491         case IPC_STAT:
492         {
493                 struct shmid64_ds tbuf;
494                 int result;
495                 memset(&tbuf, 0, sizeof(tbuf));
496                 shp = shm_lock(shmid);
497                 if(shp==NULL) {
498                         err = -EINVAL;
499                         goto out;
500                 } else if(cmd==SHM_STAT) {
501                         err = -EINVAL;
502                         if (shmid > shm_ids.max_id)
503                                 goto out_unlock;
504                         result = shm_buildid(shmid, shp->shm_perm.seq);
505                 } else {
506                         err = shm_checkid(shp,shmid);
507                         if(err)
508                                 goto out_unlock;
509                         result = 0;
510                 }
511                 err=-EACCES;
512                 if (ipcperms (&shp->shm_perm, S_IRUGO))
513                         goto out_unlock;
514                 err = security_shm_shmctl(shp, cmd);
515                 if (err)
516                         goto out_unlock;
517                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
518                 tbuf.shm_segsz  = shp->shm_segsz;
519                 tbuf.shm_atime  = shp->shm_atim;
520                 tbuf.shm_dtime  = shp->shm_dtim;
521                 tbuf.shm_ctime  = shp->shm_ctim;
522                 tbuf.shm_cpid   = shp->shm_cprid;
523                 tbuf.shm_lpid   = shp->shm_lprid;
524                 if (!is_file_hugepages(shp->shm_file))
525                         tbuf.shm_nattch = shp->shm_nattch;
526                 else
527                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
528                 shm_unlock(shp);
529                 if(copy_shmid_to_user (buf, &tbuf, version))
530                         err = -EFAULT;
531                 else
532                         err = result;
533                 goto out;
534         }
535         case SHM_LOCK:
536         case SHM_UNLOCK:
537         {
538                 shp = shm_lock(shmid);
539                 if(shp==NULL) {
540                         err = -EINVAL;
541                         goto out;
542                 }
543                 err = shm_checkid(shp,shmid);
544                 if(err)
545                         goto out_unlock;
546
547                 err = audit_ipc_obj(&(shp->shm_perm));
548                 if (err)
549                         goto out_unlock;
550
551                 if (!capable(CAP_IPC_LOCK)) {
552                         err = -EPERM;
553                         if (current->euid != shp->shm_perm.uid &&
554                             current->euid != shp->shm_perm.cuid)
555                                 goto out_unlock;
556                         if (cmd == SHM_LOCK &&
557                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
558                                 goto out_unlock;
559                 }
560
561                 err = security_shm_shmctl(shp, cmd);
562                 if (err)
563                         goto out_unlock;
564                 
565                 if(cmd==SHM_LOCK) {
566                         struct user_struct * user = current->user;
567                         if (!is_file_hugepages(shp->shm_file)) {
568                                 err = shmem_lock(shp->shm_file, 1, user);
569                                 if (!err) {
570                                         shp->shm_perm.mode |= SHM_LOCKED;
571                                         shp->mlock_user = user;
572                                 }
573                         }
574                 } else if (!is_file_hugepages(shp->shm_file)) {
575                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
576                         shp->shm_perm.mode &= ~SHM_LOCKED;
577                         shp->mlock_user = NULL;
578                 }
579                 shm_unlock(shp);
580                 goto out;
581         }
582         case IPC_RMID:
583         {
584                 /*
585                  *      We cannot simply remove the file. The SVID states
586                  *      that the block remains until the last person
587                  *      detaches from it, then is deleted. A shmat() on
588                  *      an RMID segment is legal in older Linux and if 
589                  *      we change it apps break...
590                  *
591                  *      Instead we set a destroyed flag, and then blow
592                  *      the name away when the usage hits zero.
593                  */
594                 mutex_lock(&shm_ids.mutex);
595                 shp = shm_lock(shmid);
596                 err = -EINVAL;
597                 if (shp == NULL) 
598                         goto out_up;
599                 err = shm_checkid(shp, shmid);
600                 if(err)
601                         goto out_unlock_up;
602
603                 err = audit_ipc_obj(&(shp->shm_perm));
604                 if (err)
605                         goto out_unlock_up;
606
607                 if (current->euid != shp->shm_perm.uid &&
608                     current->euid != shp->shm_perm.cuid && 
609                     !capable(CAP_SYS_ADMIN)) {
610                         err=-EPERM;
611                         goto out_unlock_up;
612                 }
613
614                 err = security_shm_shmctl(shp, cmd);
615                 if (err)
616                         goto out_unlock_up;
617
618                 if (shp->shm_nattch){
619                         shp->shm_perm.mode |= SHM_DEST;
620                         /* Do not find it any more */
621                         shp->shm_perm.key = IPC_PRIVATE;
622                         shm_unlock(shp);
623                 } else
624                         shm_destroy (shp);
625                 mutex_unlock(&shm_ids.mutex);
626                 goto out;
627         }
628
629         case IPC_SET:
630         {
631                 if (copy_shmid_from_user (&setbuf, buf, version)) {
632                         err = -EFAULT;
633                         goto out;
634                 }
635                 mutex_lock(&shm_ids.mutex);
636                 shp = shm_lock(shmid);
637                 err=-EINVAL;
638                 if(shp==NULL)
639                         goto out_up;
640                 err = shm_checkid(shp,shmid);
641                 if(err)
642                         goto out_unlock_up;
643                 err = audit_ipc_obj(&(shp->shm_perm));
644                 if (err)
645                         goto out_unlock_up;
646                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
647                 if (err)
648                         goto out_unlock_up;
649                 err=-EPERM;
650                 if (current->euid != shp->shm_perm.uid &&
651                     current->euid != shp->shm_perm.cuid && 
652                     !capable(CAP_SYS_ADMIN)) {
653                         goto out_unlock_up;
654                 }
655
656                 err = security_shm_shmctl(shp, cmd);
657                 if (err)
658                         goto out_unlock_up;
659                 
660                 shp->shm_perm.uid = setbuf.uid;
661                 shp->shm_perm.gid = setbuf.gid;
662                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
663                         | (setbuf.mode & S_IRWXUGO);
664                 shp->shm_ctim = get_seconds();
665                 break;
666         }
667
668         default:
669                 err = -EINVAL;
670                 goto out;
671         }
672
673         err = 0;
674 out_unlock_up:
675         shm_unlock(shp);
676 out_up:
677         mutex_unlock(&shm_ids.mutex);
678         goto out;
679 out_unlock:
680         shm_unlock(shp);
681 out:
682         return err;
683 }
684
685 /*
686  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
687  *
688  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
689  * "raddr" thing points to kernel space, and there has to be a wrapper around
690  * this.
691  */
692 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
693 {
694         struct shmid_kernel *shp;
695         unsigned long addr;
696         unsigned long size;
697         struct file * file;
698         int    err;
699         unsigned long flags;
700         unsigned long prot;
701         unsigned long o_flags;
702         int acc_mode;
703         void *user_addr;
704
705         if (shmid < 0) {
706                 err = -EINVAL;
707                 goto out;
708         } else if ((addr = (ulong)shmaddr)) {
709                 if (addr & (SHMLBA-1)) {
710                         if (shmflg & SHM_RND)
711                                 addr &= ~(SHMLBA-1);       /* round down */
712                         else
713 #ifndef __ARCH_FORCE_SHMLBA
714                                 if (addr & ~PAGE_MASK)
715 #endif
716                                         return -EINVAL;
717                 }
718                 flags = MAP_SHARED | MAP_FIXED;
719         } else {
720                 if ((shmflg & SHM_REMAP))
721                         return -EINVAL;
722
723                 flags = MAP_SHARED;
724         }
725
726         if (shmflg & SHM_RDONLY) {
727                 prot = PROT_READ;
728                 o_flags = O_RDONLY;
729                 acc_mode = S_IRUGO;
730         } else {
731                 prot = PROT_READ | PROT_WRITE;
732                 o_flags = O_RDWR;
733                 acc_mode = S_IRUGO | S_IWUGO;
734         }
735         if (shmflg & SHM_EXEC) {
736                 prot |= PROT_EXEC;
737                 acc_mode |= S_IXUGO;
738         }
739
740         /*
741          * We cannot rely on the fs check since SYSV IPC does have an
742          * additional creator id...
743          */
744         shp = shm_lock(shmid);
745         if(shp == NULL) {
746                 err = -EINVAL;
747                 goto out;
748         }
749         err = shm_checkid(shp,shmid);
750         if (err) {
751                 shm_unlock(shp);
752                 goto out;
753         }
754         if (ipcperms(&shp->shm_perm, acc_mode)) {
755                 shm_unlock(shp);
756                 err = -EACCES;
757                 goto out;
758         }
759
760         err = security_shm_shmat(shp, shmaddr, shmflg);
761         if (err) {
762                 shm_unlock(shp);
763                 return err;
764         }
765                 
766         file = shp->shm_file;
767         size = i_size_read(file->f_dentry->d_inode);
768         shp->shm_nattch++;
769         shm_unlock(shp);
770
771         down_write(&current->mm->mmap_sem);
772         if (addr && !(shmflg & SHM_REMAP)) {
773                 user_addr = ERR_PTR(-EINVAL);
774                 if (find_vma_intersection(current->mm, addr, addr + size))
775                         goto invalid;
776                 /*
777                  * If shm segment goes below stack, make sure there is some
778                  * space left for the stack to grow (at least 4 pages).
779                  */
780                 if (addr < current->mm->start_stack &&
781                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
782                         goto invalid;
783         }
784                 
785         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
786
787 invalid:
788         up_write(&current->mm->mmap_sem);
789
790         mutex_lock(&shm_ids.mutex);
791         shp = shm_lock(shmid);
792         BUG_ON(!shp);
793         shp->shm_nattch--;
794         if(shp->shm_nattch == 0 &&
795            shp->shm_perm.mode & SHM_DEST)
796                 shm_destroy (shp);
797         else
798                 shm_unlock(shp);
799         mutex_unlock(&shm_ids.mutex);
800
801         *raddr = (unsigned long) user_addr;
802         err = 0;
803         if (IS_ERR(user_addr))
804                 err = PTR_ERR(user_addr);
805 out:
806         return err;
807 }
808
809 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
810 {
811         unsigned long ret;
812         long err;
813
814         err = do_shmat(shmid, shmaddr, shmflg, &ret);
815         if (err)
816                 return err;
817         force_successful_syscall_return();
818         return (long)ret;
819 }
820
821 /*
822  * detach and kill segment if marked destroyed.
823  * The work is done in shm_close.
824  */
825 asmlinkage long sys_shmdt(char __user *shmaddr)
826 {
827         struct mm_struct *mm = current->mm;
828         struct vm_area_struct *vma, *next;
829         unsigned long addr = (unsigned long)shmaddr;
830         loff_t size = 0;
831         int retval = -EINVAL;
832
833         if (addr & ~PAGE_MASK)
834                 return retval;
835
836         down_write(&mm->mmap_sem);
837
838         /*
839          * This function tries to be smart and unmap shm segments that
840          * were modified by partial mlock or munmap calls:
841          * - It first determines the size of the shm segment that should be
842          *   unmapped: It searches for a vma that is backed by shm and that
843          *   started at address shmaddr. It records it's size and then unmaps
844          *   it.
845          * - Then it unmaps all shm vmas that started at shmaddr and that
846          *   are within the initially determined size.
847          * Errors from do_munmap are ignored: the function only fails if
848          * it's called with invalid parameters or if it's called to unmap
849          * a part of a vma. Both calls in this function are for full vmas,
850          * the parameters are directly copied from the vma itself and always
851          * valid - therefore do_munmap cannot fail. (famous last words?)
852          */
853         /*
854          * If it had been mremap()'d, the starting address would not
855          * match the usual checks anyway. So assume all vma's are
856          * above the starting address given.
857          */
858         vma = find_vma(mm, addr);
859
860         while (vma) {
861                 next = vma->vm_next;
862
863                 /*
864                  * Check if the starting address would match, i.e. it's
865                  * a fragment created by mprotect() and/or munmap(), or it
866                  * otherwise it starts at this address with no hassles.
867                  */
868                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
869                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
870
871
872                         size = vma->vm_file->f_dentry->d_inode->i_size;
873                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
874                         /*
875                          * We discovered the size of the shm segment, so
876                          * break out of here and fall through to the next
877                          * loop that uses the size information to stop
878                          * searching for matching vma's.
879                          */
880                         retval = 0;
881                         vma = next;
882                         break;
883                 }
884                 vma = next;
885         }
886
887         /*
888          * We need look no further than the maximum address a fragment
889          * could possibly have landed at. Also cast things to loff_t to
890          * prevent overflows and make comparisions vs. equal-width types.
891          */
892         size = PAGE_ALIGN(size);
893         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
894                 next = vma->vm_next;
895
896                 /* finding a matching vma now does not alter retval */
897                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
898                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
899
900                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
901                 vma = next;
902         }
903
904         up_write(&mm->mmap_sem);
905         return retval;
906 }
907
908 #ifdef CONFIG_PROC_FS
909 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
910 {
911         struct shmid_kernel *shp = it;
912         char *format;
913
914 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
915 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
916
917         if (sizeof(size_t) <= sizeof(int))
918                 format = SMALL_STRING;
919         else
920                 format = BIG_STRING;
921         return seq_printf(s, format,
922                           shp->shm_perm.key,
923                           shp->id,
924                           shp->shm_perm.mode,
925                           shp->shm_segsz,
926                           shp->shm_cprid,
927                           shp->shm_lprid,
928                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
929                           shp->shm_perm.uid,
930                           shp->shm_perm.gid,
931                           shp->shm_perm.cuid,
932                           shp->shm_perm.cgid,
933                           shp->shm_atim,
934                           shp->shm_dtim,
935                           shp->shm_ctim);
936 }
937 #endif