Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
[pandora-kernel.git] / ipc / shm.c
1 /*
2  * linux/ipc/shm.c
3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
4  *       Many improvements/fixes by Bruno Haible.
5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7  *
8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15  *
16  * support for audit of ipc object properties and permission changes
17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18  *
19  * namespaces support
20  * OpenVZ, SWsoft Inc.
21  * Pavel Emelianov <xemul@openvz.org>
22  */
23
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 #include <linux/mount.h>
41
42 #include <asm/uaccess.h>
43
44 #include "util.h"
45
46 struct shm_file_data {
47         int id;
48         struct ipc_namespace *ns;
49         struct file *file;
50         const struct vm_operations_struct *vm_ops;
51 };
52
53 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
54
55 static const struct file_operations shm_file_operations;
56 static struct vm_operations_struct shm_vm_ops;
57
58 static struct ipc_ids init_shm_ids;
59
60 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
61
62 #define shm_lock(ns, id)                \
63         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
64 #define shm_unlock(shp)                 \
65         ipc_unlock(&(shp)->shm_perm)
66 #define shm_get(ns, id)                 \
67         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
68 #define shm_buildid(ns, id, seq)        \
69         ipc_buildid(&shm_ids(ns), id, seq)
70
71 static int newseg (struct ipc_namespace *ns, key_t key,
72                 int shmflg, size_t size);
73 static void shm_open(struct vm_area_struct *vma);
74 static void shm_close(struct vm_area_struct *vma);
75 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
76 #ifdef CONFIG_PROC_FS
77 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
78 #endif
79
80 static void __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
81 {
82         ns->ids[IPC_SHM_IDS] = ids;
83         ns->shm_ctlmax = SHMMAX;
84         ns->shm_ctlall = SHMALL;
85         ns->shm_ctlmni = SHMMNI;
86         ns->shm_tot = 0;
87         ipc_init_ids(ids, 1);
88 }
89
90 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
91 {
92         if (shp->shm_nattch){
93                 shp->shm_perm.mode |= SHM_DEST;
94                 /* Do not find it any more */
95                 shp->shm_perm.key = IPC_PRIVATE;
96                 shm_unlock(shp);
97         } else
98                 shm_destroy(ns, shp);
99 }
100
101 int shm_init_ns(struct ipc_namespace *ns)
102 {
103         struct ipc_ids *ids;
104
105         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
106         if (ids == NULL)
107                 return -ENOMEM;
108
109         __shm_init_ns(ns, ids);
110         return 0;
111 }
112
113 void shm_exit_ns(struct ipc_namespace *ns)
114 {
115         int i;
116         struct shmid_kernel *shp;
117
118         mutex_lock(&shm_ids(ns).mutex);
119         for (i = 0; i <= shm_ids(ns).max_id; i++) {
120                 shp = shm_lock(ns, i);
121                 if (shp == NULL)
122                         continue;
123
124                 do_shm_rmid(ns, shp);
125         }
126         mutex_unlock(&shm_ids(ns).mutex);
127
128         ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
129         kfree(ns->ids[IPC_SHM_IDS]);
130         ns->ids[IPC_SHM_IDS] = NULL;
131 }
132
133 void __init shm_init (void)
134 {
135         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
136         ipc_init_proc_interface("sysvipc/shm",
137                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
138                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
139 }
140
141 static inline int shm_checkid(struct ipc_namespace *ns,
142                 struct shmid_kernel *s, int id)
143 {
144         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
145                 return -EIDRM;
146         return 0;
147 }
148
149 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
150 {
151         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
152 }
153
154 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
155 {
156         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
157 }
158
159
160
161 /* This is called by fork, once for every shm attach. */
162 static void shm_open(struct vm_area_struct *vma)
163 {
164         struct file *file = vma->vm_file;
165         struct shm_file_data *sfd = shm_file_data(file);
166         struct shmid_kernel *shp;
167
168         shp = shm_lock(sfd->ns, sfd->id);
169         BUG_ON(!shp);
170         shp->shm_atim = get_seconds();
171         shp->shm_lprid = current->tgid;
172         shp->shm_nattch++;
173         shm_unlock(shp);
174 }
175
176 /*
177  * shm_destroy - free the struct shmid_kernel
178  *
179  * @shp: struct to free
180  *
181  * It has to be called with shp and shm_ids.mutex locked,
182  * but returns with shp unlocked and freed.
183  */
184 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
185 {
186         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
187         shm_rmid(ns, shp->id);
188         shm_unlock(shp);
189         if (!is_file_hugepages(shp->shm_file))
190                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
191         else
192                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
193                                                 shp->mlock_user);
194         fput (shp->shm_file);
195         security_shm_free(shp);
196         ipc_rcu_putref(shp);
197 }
198
199 /*
200  * remove the attach descriptor vma.
201  * free memory for segment if it is marked destroyed.
202  * The descriptor has already been removed from the current->mm->mmap list
203  * and will later be kfree()d.
204  */
205 static void shm_close(struct vm_area_struct *vma)
206 {
207         struct file * file = vma->vm_file;
208         struct shm_file_data *sfd = shm_file_data(file);
209         struct shmid_kernel *shp;
210         struct ipc_namespace *ns = sfd->ns;
211
212         mutex_lock(&shm_ids(ns).mutex);
213         /* remove from the list of attaches of the shm segment */
214         shp = shm_lock(ns, sfd->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_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
228 {
229         struct file *file = vma->vm_file;
230         struct shm_file_data *sfd = shm_file_data(file);
231
232         return sfd->vm_ops->fault(vma, vmf);
233 }
234
235 #ifdef CONFIG_NUMA
236 int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
237 {
238         struct file *file = vma->vm_file;
239         struct shm_file_data *sfd = shm_file_data(file);
240         int err = 0;
241         if (sfd->vm_ops->set_policy)
242                 err = sfd->vm_ops->set_policy(vma, new);
243         return err;
244 }
245
246 struct mempolicy *shm_get_policy(struct vm_area_struct *vma, unsigned long addr)
247 {
248         struct file *file = vma->vm_file;
249         struct shm_file_data *sfd = shm_file_data(file);
250         struct mempolicy *pol = NULL;
251
252         if (sfd->vm_ops->get_policy)
253                 pol = sfd->vm_ops->get_policy(vma, addr);
254         else if (vma->vm_policy)
255                 pol = vma->vm_policy;
256         else
257                 pol = current->mempolicy;
258         return pol;
259 }
260 #endif
261
262 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
263 {
264         struct shm_file_data *sfd = shm_file_data(file);
265         int ret;
266
267         ret = sfd->file->f_op->mmap(sfd->file, vma);
268         if (ret != 0)
269                 return ret;
270         sfd->vm_ops = vma->vm_ops;
271         BUG_ON(!sfd->vm_ops->fault);
272         vma->vm_ops = &shm_vm_ops;
273         shm_open(vma);
274
275         return ret;
276 }
277
278 static int shm_release(struct inode *ino, struct file *file)
279 {
280         struct shm_file_data *sfd = shm_file_data(file);
281
282         put_ipc_ns(sfd->ns);
283         shm_file_data(file) = NULL;
284         kfree(sfd);
285         return 0;
286 }
287
288 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
289 {
290         int (*fsync) (struct file *, struct dentry *, int datasync);
291         struct shm_file_data *sfd = shm_file_data(file);
292         int ret = -EINVAL;
293
294         fsync = sfd->file->f_op->fsync;
295         if (fsync)
296                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
297         return ret;
298 }
299
300 static unsigned long shm_get_unmapped_area(struct file *file,
301         unsigned long addr, unsigned long len, unsigned long pgoff,
302         unsigned long flags)
303 {
304         struct shm_file_data *sfd = shm_file_data(file);
305         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
306 }
307
308 int is_file_shm_hugepages(struct file *file)
309 {
310         int ret = 0;
311
312         if (file->f_op == &shm_file_operations) {
313                 struct shm_file_data *sfd;
314                 sfd = shm_file_data(file);
315                 ret = is_file_hugepages(sfd->file);
316         }
317         return ret;
318 }
319
320 static const struct file_operations shm_file_operations = {
321         .mmap           = shm_mmap,
322         .fsync          = shm_fsync,
323         .release        = shm_release,
324         .get_unmapped_area      = shm_get_unmapped_area,
325 };
326
327 static struct vm_operations_struct shm_vm_ops = {
328         .open   = shm_open,     /* callback for a new vm-area open */
329         .close  = shm_close,    /* callback for when the vm-area is released */
330         .fault  = shm_fault,
331 #if defined(CONFIG_NUMA)
332         .set_policy = shm_set_policy,
333         .get_policy = shm_get_policy,
334 #endif
335 };
336
337 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
338 {
339         int error;
340         struct shmid_kernel *shp;
341         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
342         struct file * file;
343         char name[13];
344         int id;
345
346         if (size < SHMMIN || size > ns->shm_ctlmax)
347                 return -EINVAL;
348
349         if (ns->shm_tot + numpages > ns->shm_ctlall)
350                 return -ENOSPC;
351
352         shp = ipc_rcu_alloc(sizeof(*shp));
353         if (!shp)
354                 return -ENOMEM;
355
356         shp->shm_perm.key = key;
357         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
358         shp->mlock_user = NULL;
359
360         shp->shm_perm.security = NULL;
361         error = security_shm_alloc(shp);
362         if (error) {
363                 ipc_rcu_putref(shp);
364                 return error;
365         }
366
367         sprintf (name, "SYSV%08x", key);
368         if (shmflg & SHM_HUGETLB) {
369                 /* hugetlb_file_setup takes care of mlock user accounting */
370                 file = hugetlb_file_setup(name, size);
371                 shp->mlock_user = current->user;
372         } else {
373                 int acctflag = VM_ACCOUNT;
374                 /*
375                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
376                  * if it's asked for.
377                  */
378                 if  ((shmflg & SHM_NORESERVE) &&
379                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
380                         acctflag = 0;
381                 file = shmem_file_setup(name, size, acctflag);
382         }
383         error = PTR_ERR(file);
384         if (IS_ERR(file))
385                 goto no_file;
386
387         error = -ENOSPC;
388         id = shm_addid(ns, shp);
389         if(id == -1) 
390                 goto no_id;
391
392         shp->shm_cprid = current->tgid;
393         shp->shm_lprid = 0;
394         shp->shm_atim = shp->shm_dtim = 0;
395         shp->shm_ctim = get_seconds();
396         shp->shm_segsz = size;
397         shp->shm_nattch = 0;
398         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
399         shp->shm_file = file;
400         /*
401          * shmid gets reported as "inode#" in /proc/pid/maps.
402          * proc-ps tools use this. Changing this will break them.
403          */
404         file->f_dentry->d_inode->i_ino = shp->id;
405
406         ns->shm_tot += numpages;
407         shm_unlock(shp);
408         return shp->id;
409
410 no_id:
411         fput(file);
412 no_file:
413         security_shm_free(shp);
414         ipc_rcu_putref(shp);
415         return error;
416 }
417
418 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
419 {
420         struct shmid_kernel *shp;
421         int err, id = 0;
422         struct ipc_namespace *ns;
423
424         ns = current->nsproxy->ipc_ns;
425
426         mutex_lock(&shm_ids(ns).mutex);
427         if (key == IPC_PRIVATE) {
428                 err = newseg(ns, key, shmflg, size);
429         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
430                 if (!(shmflg & IPC_CREAT))
431                         err = -ENOENT;
432                 else
433                         err = newseg(ns, key, shmflg, size);
434         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
435                 err = -EEXIST;
436         } else {
437                 shp = shm_lock(ns, id);
438                 BUG_ON(shp==NULL);
439                 if (shp->shm_segsz < size)
440                         err = -EINVAL;
441                 else if (ipcperms(&shp->shm_perm, shmflg))
442                         err = -EACCES;
443                 else {
444                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
445                         err = security_shm_associate(shp, shmflg);
446                         if (!err)
447                                 err = shmid;
448                 }
449                 shm_unlock(shp);
450         }
451         mutex_unlock(&shm_ids(ns).mutex);
452
453         return err;
454 }
455
456 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *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 shmid_ds out;
464
465                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
466                 out.shm_segsz   = in->shm_segsz;
467                 out.shm_atime   = in->shm_atime;
468                 out.shm_dtime   = in->shm_dtime;
469                 out.shm_ctime   = in->shm_ctime;
470                 out.shm_cpid    = in->shm_cpid;
471                 out.shm_lpid    = in->shm_lpid;
472                 out.shm_nattch  = in->shm_nattch;
473
474                 return copy_to_user(buf, &out, sizeof(out));
475             }
476         default:
477                 return -EINVAL;
478         }
479 }
480
481 struct shm_setbuf {
482         uid_t   uid;
483         gid_t   gid;
484         mode_t  mode;
485 };      
486
487 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
488 {
489         switch(version) {
490         case IPC_64:
491             {
492                 struct shmid64_ds tbuf;
493
494                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
495                         return -EFAULT;
496
497                 out->uid        = tbuf.shm_perm.uid;
498                 out->gid        = tbuf.shm_perm.gid;
499                 out->mode       = tbuf.shm_perm.mode;
500
501                 return 0;
502             }
503         case IPC_OLD:
504             {
505                 struct shmid_ds tbuf_old;
506
507                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
508                         return -EFAULT;
509
510                 out->uid        = tbuf_old.shm_perm.uid;
511                 out->gid        = tbuf_old.shm_perm.gid;
512                 out->mode       = tbuf_old.shm_perm.mode;
513
514                 return 0;
515             }
516         default:
517                 return -EINVAL;
518         }
519 }
520
521 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
522 {
523         switch(version) {
524         case IPC_64:
525                 return copy_to_user(buf, in, sizeof(*in));
526         case IPC_OLD:
527             {
528                 struct shminfo out;
529
530                 if(in->shmmax > INT_MAX)
531                         out.shmmax = INT_MAX;
532                 else
533                         out.shmmax = (int)in->shmmax;
534
535                 out.shmmin      = in->shmmin;
536                 out.shmmni      = in->shmmni;
537                 out.shmseg      = in->shmseg;
538                 out.shmall      = in->shmall; 
539
540                 return copy_to_user(buf, &out, sizeof(out));
541             }
542         default:
543                 return -EINVAL;
544         }
545 }
546
547 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
548                 unsigned long *swp)
549 {
550         int i;
551
552         *rss = 0;
553         *swp = 0;
554
555         for (i = 0; i <= shm_ids(ns).max_id; i++) {
556                 struct shmid_kernel *shp;
557                 struct inode *inode;
558
559                 shp = shm_get(ns, i);
560                 if(!shp)
561                         continue;
562
563                 inode = shp->shm_file->f_path.dentry->d_inode;
564
565                 if (is_file_hugepages(shp->shm_file)) {
566                         struct address_space *mapping = inode->i_mapping;
567                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
568                 } else {
569                         struct shmem_inode_info *info = SHMEM_I(inode);
570                         spin_lock(&info->lock);
571                         *rss += inode->i_mapping->nrpages;
572                         *swp += info->swapped;
573                         spin_unlock(&info->lock);
574                 }
575         }
576 }
577
578 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
579 {
580         struct shm_setbuf setbuf;
581         struct shmid_kernel *shp;
582         int err, version;
583         struct ipc_namespace *ns;
584
585         if (cmd < 0 || shmid < 0) {
586                 err = -EINVAL;
587                 goto out;
588         }
589
590         version = ipc_parse_version(&cmd);
591         ns = current->nsproxy->ipc_ns;
592
593         switch (cmd) { /* replace with proc interface ? */
594         case IPC_INFO:
595         {
596                 struct shminfo64 shminfo;
597
598                 err = security_shm_shmctl(NULL, cmd);
599                 if (err)
600                         return err;
601
602                 memset(&shminfo,0,sizeof(shminfo));
603                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
604                 shminfo.shmmax = ns->shm_ctlmax;
605                 shminfo.shmall = ns->shm_ctlall;
606
607                 shminfo.shmmin = SHMMIN;
608                 if(copy_shminfo_to_user (buf, &shminfo, version))
609                         return -EFAULT;
610                 /* reading a integer is always atomic */
611                 err= shm_ids(ns).max_id;
612                 if(err<0)
613                         err = 0;
614                 goto out;
615         }
616         case SHM_INFO:
617         {
618                 struct shm_info shm_info;
619
620                 err = security_shm_shmctl(NULL, cmd);
621                 if (err)
622                         return err;
623
624                 memset(&shm_info,0,sizeof(shm_info));
625                 mutex_lock(&shm_ids(ns).mutex);
626                 shm_info.used_ids = shm_ids(ns).in_use;
627                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
628                 shm_info.shm_tot = ns->shm_tot;
629                 shm_info.swap_attempts = 0;
630                 shm_info.swap_successes = 0;
631                 err = shm_ids(ns).max_id;
632                 mutex_unlock(&shm_ids(ns).mutex);
633                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
634                         err = -EFAULT;
635                         goto out;
636                 }
637
638                 err = err < 0 ? 0 : err;
639                 goto out;
640         }
641         case SHM_STAT:
642         case IPC_STAT:
643         {
644                 struct shmid64_ds tbuf;
645                 int result;
646                 memset(&tbuf, 0, sizeof(tbuf));
647                 shp = shm_lock(ns, shmid);
648                 if(shp==NULL) {
649                         err = -EINVAL;
650                         goto out;
651                 } else if(cmd==SHM_STAT) {
652                         err = -EINVAL;
653                         if (shmid > shm_ids(ns).max_id)
654                                 goto out_unlock;
655                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
656                 } else {
657                         err = shm_checkid(ns, shp,shmid);
658                         if(err)
659                                 goto out_unlock;
660                         result = 0;
661                 }
662                 err=-EACCES;
663                 if (ipcperms (&shp->shm_perm, S_IRUGO))
664                         goto out_unlock;
665                 err = security_shm_shmctl(shp, cmd);
666                 if (err)
667                         goto out_unlock;
668                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
669                 tbuf.shm_segsz  = shp->shm_segsz;
670                 tbuf.shm_atime  = shp->shm_atim;
671                 tbuf.shm_dtime  = shp->shm_dtim;
672                 tbuf.shm_ctime  = shp->shm_ctim;
673                 tbuf.shm_cpid   = shp->shm_cprid;
674                 tbuf.shm_lpid   = shp->shm_lprid;
675                 tbuf.shm_nattch = shp->shm_nattch;
676                 shm_unlock(shp);
677                 if(copy_shmid_to_user (buf, &tbuf, version))
678                         err = -EFAULT;
679                 else
680                         err = result;
681                 goto out;
682         }
683         case SHM_LOCK:
684         case SHM_UNLOCK:
685         {
686                 shp = shm_lock(ns, shmid);
687                 if(shp==NULL) {
688                         err = -EINVAL;
689                         goto out;
690                 }
691                 err = shm_checkid(ns, shp,shmid);
692                 if(err)
693                         goto out_unlock;
694
695                 err = audit_ipc_obj(&(shp->shm_perm));
696                 if (err)
697                         goto out_unlock;
698
699                 if (!capable(CAP_IPC_LOCK)) {
700                         err = -EPERM;
701                         if (current->euid != shp->shm_perm.uid &&
702                             current->euid != shp->shm_perm.cuid)
703                                 goto out_unlock;
704                         if (cmd == SHM_LOCK &&
705                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
706                                 goto out_unlock;
707                 }
708
709                 err = security_shm_shmctl(shp, cmd);
710                 if (err)
711                         goto out_unlock;
712                 
713                 if(cmd==SHM_LOCK) {
714                         struct user_struct * user = current->user;
715                         if (!is_file_hugepages(shp->shm_file)) {
716                                 err = shmem_lock(shp->shm_file, 1, user);
717                                 if (!err) {
718                                         shp->shm_perm.mode |= SHM_LOCKED;
719                                         shp->mlock_user = user;
720                                 }
721                         }
722                 } else if (!is_file_hugepages(shp->shm_file)) {
723                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
724                         shp->shm_perm.mode &= ~SHM_LOCKED;
725                         shp->mlock_user = NULL;
726                 }
727                 shm_unlock(shp);
728                 goto out;
729         }
730         case IPC_RMID:
731         {
732                 /*
733                  *      We cannot simply remove the file. The SVID states
734                  *      that the block remains until the last person
735                  *      detaches from it, then is deleted. A shmat() on
736                  *      an RMID segment is legal in older Linux and if 
737                  *      we change it apps break...
738                  *
739                  *      Instead we set a destroyed flag, and then blow
740                  *      the name away when the usage hits zero.
741                  */
742                 mutex_lock(&shm_ids(ns).mutex);
743                 shp = shm_lock(ns, shmid);
744                 err = -EINVAL;
745                 if (shp == NULL) 
746                         goto out_up;
747                 err = shm_checkid(ns, shp, shmid);
748                 if(err)
749                         goto out_unlock_up;
750
751                 err = audit_ipc_obj(&(shp->shm_perm));
752                 if (err)
753                         goto out_unlock_up;
754
755                 if (current->euid != shp->shm_perm.uid &&
756                     current->euid != shp->shm_perm.cuid && 
757                     !capable(CAP_SYS_ADMIN)) {
758                         err=-EPERM;
759                         goto out_unlock_up;
760                 }
761
762                 err = security_shm_shmctl(shp, cmd);
763                 if (err)
764                         goto out_unlock_up;
765
766                 do_shm_rmid(ns, shp);
767                 mutex_unlock(&shm_ids(ns).mutex);
768                 goto out;
769         }
770
771         case IPC_SET:
772         {
773                 if (copy_shmid_from_user (&setbuf, buf, version)) {
774                         err = -EFAULT;
775                         goto out;
776                 }
777                 mutex_lock(&shm_ids(ns).mutex);
778                 shp = shm_lock(ns, shmid);
779                 err=-EINVAL;
780                 if(shp==NULL)
781                         goto out_up;
782                 err = shm_checkid(ns, shp,shmid);
783                 if(err)
784                         goto out_unlock_up;
785                 err = audit_ipc_obj(&(shp->shm_perm));
786                 if (err)
787                         goto out_unlock_up;
788                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
789                 if (err)
790                         goto out_unlock_up;
791                 err=-EPERM;
792                 if (current->euid != shp->shm_perm.uid &&
793                     current->euid != shp->shm_perm.cuid && 
794                     !capable(CAP_SYS_ADMIN)) {
795                         goto out_unlock_up;
796                 }
797
798                 err = security_shm_shmctl(shp, cmd);
799                 if (err)
800                         goto out_unlock_up;
801                 
802                 shp->shm_perm.uid = setbuf.uid;
803                 shp->shm_perm.gid = setbuf.gid;
804                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
805                         | (setbuf.mode & S_IRWXUGO);
806                 shp->shm_ctim = get_seconds();
807                 break;
808         }
809
810         default:
811                 err = -EINVAL;
812                 goto out;
813         }
814
815         err = 0;
816 out_unlock_up:
817         shm_unlock(shp);
818 out_up:
819         mutex_unlock(&shm_ids(ns).mutex);
820         goto out;
821 out_unlock:
822         shm_unlock(shp);
823 out:
824         return err;
825 }
826
827 /*
828  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
829  *
830  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
831  * "raddr" thing points to kernel space, and there has to be a wrapper around
832  * this.
833  */
834 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
835 {
836         struct shmid_kernel *shp;
837         unsigned long addr;
838         unsigned long size;
839         struct file * file;
840         int    err;
841         unsigned long flags;
842         unsigned long prot;
843         int acc_mode;
844         unsigned long user_addr;
845         struct ipc_namespace *ns;
846         struct shm_file_data *sfd;
847         struct path path;
848         mode_t f_mode;
849
850         err = -EINVAL;
851         if (shmid < 0)
852                 goto out;
853         else if ((addr = (ulong)shmaddr)) {
854                 if (addr & (SHMLBA-1)) {
855                         if (shmflg & SHM_RND)
856                                 addr &= ~(SHMLBA-1);       /* round down */
857                         else
858 #ifndef __ARCH_FORCE_SHMLBA
859                                 if (addr & ~PAGE_MASK)
860 #endif
861                                         goto out;
862                 }
863                 flags = MAP_SHARED | MAP_FIXED;
864         } else {
865                 if ((shmflg & SHM_REMAP))
866                         goto out;
867
868                 flags = MAP_SHARED;
869         }
870
871         if (shmflg & SHM_RDONLY) {
872                 prot = PROT_READ;
873                 acc_mode = S_IRUGO;
874                 f_mode = FMODE_READ;
875         } else {
876                 prot = PROT_READ | PROT_WRITE;
877                 acc_mode = S_IRUGO | S_IWUGO;
878                 f_mode = FMODE_READ | FMODE_WRITE;
879         }
880         if (shmflg & SHM_EXEC) {
881                 prot |= PROT_EXEC;
882                 acc_mode |= S_IXUGO;
883         }
884
885         /*
886          * We cannot rely on the fs check since SYSV IPC does have an
887          * additional creator id...
888          */
889         ns = current->nsproxy->ipc_ns;
890         shp = shm_lock(ns, shmid);
891         if(shp == NULL)
892                 goto out;
893
894         err = shm_checkid(ns, shp,shmid);
895         if (err)
896                 goto out_unlock;
897
898         err = -EACCES;
899         if (ipcperms(&shp->shm_perm, acc_mode))
900                 goto out_unlock;
901
902         err = security_shm_shmat(shp, shmaddr, shmflg);
903         if (err)
904                 goto out_unlock;
905
906         path.dentry = dget(shp->shm_file->f_path.dentry);
907         path.mnt    = mntget(shp->shm_file->f_path.mnt);
908         shp->shm_nattch++;
909         size = i_size_read(path.dentry->d_inode);
910         shm_unlock(shp);
911
912         err = -ENOMEM;
913         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
914         if (!sfd)
915                 goto out_put_path;
916
917         err = -ENOMEM;
918         file = get_empty_filp();
919         if (!file)
920                 goto out_free;
921
922         file->f_op = &shm_file_operations;
923         file->private_data = sfd;
924         file->f_path = path;
925         file->f_mapping = shp->shm_file->f_mapping;
926         file->f_mode = f_mode;
927         sfd->id = shp->id;
928         sfd->ns = get_ipc_ns(ns);
929         sfd->file = shp->shm_file;
930         sfd->vm_ops = NULL;
931
932         down_write(&current->mm->mmap_sem);
933         if (addr && !(shmflg & SHM_REMAP)) {
934                 err = -EINVAL;
935                 if (find_vma_intersection(current->mm, addr, addr + size))
936                         goto invalid;
937                 /*
938                  * If shm segment goes below stack, make sure there is some
939                  * space left for the stack to grow (at least 4 pages).
940                  */
941                 if (addr < current->mm->start_stack &&
942                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
943                         goto invalid;
944         }
945                 
946         user_addr = do_mmap (file, addr, size, prot, flags, 0);
947         *raddr = user_addr;
948         err = 0;
949         if (IS_ERR_VALUE(user_addr))
950                 err = (long)user_addr;
951 invalid:
952         up_write(&current->mm->mmap_sem);
953
954         fput(file);
955
956 out_nattch:
957         mutex_lock(&shm_ids(ns).mutex);
958         shp = shm_lock(ns, shmid);
959         BUG_ON(!shp);
960         shp->shm_nattch--;
961         if(shp->shm_nattch == 0 &&
962            shp->shm_perm.mode & SHM_DEST)
963                 shm_destroy(ns, shp);
964         else
965                 shm_unlock(shp);
966         mutex_unlock(&shm_ids(ns).mutex);
967
968 out:
969         return err;
970
971 out_unlock:
972         shm_unlock(shp);
973         goto out;
974
975 out_free:
976         kfree(sfd);
977 out_put_path:
978         dput(path.dentry);
979         mntput(path.mnt);
980         goto out_nattch;
981 }
982
983 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
984 {
985         unsigned long ret;
986         long err;
987
988         err = do_shmat(shmid, shmaddr, shmflg, &ret);
989         if (err)
990                 return err;
991         force_successful_syscall_return();
992         return (long)ret;
993 }
994
995 /*
996  * detach and kill segment if marked destroyed.
997  * The work is done in shm_close.
998  */
999 asmlinkage long sys_shmdt(char __user *shmaddr)
1000 {
1001         struct mm_struct *mm = current->mm;
1002         struct vm_area_struct *vma, *next;
1003         unsigned long addr = (unsigned long)shmaddr;
1004         loff_t size = 0;
1005         int retval = -EINVAL;
1006
1007         if (addr & ~PAGE_MASK)
1008                 return retval;
1009
1010         down_write(&mm->mmap_sem);
1011
1012         /*
1013          * This function tries to be smart and unmap shm segments that
1014          * were modified by partial mlock or munmap calls:
1015          * - It first determines the size of the shm segment that should be
1016          *   unmapped: It searches for a vma that is backed by shm and that
1017          *   started at address shmaddr. It records it's size and then unmaps
1018          *   it.
1019          * - Then it unmaps all shm vmas that started at shmaddr and that
1020          *   are within the initially determined size.
1021          * Errors from do_munmap are ignored: the function only fails if
1022          * it's called with invalid parameters or if it's called to unmap
1023          * a part of a vma. Both calls in this function are for full vmas,
1024          * the parameters are directly copied from the vma itself and always
1025          * valid - therefore do_munmap cannot fail. (famous last words?)
1026          */
1027         /*
1028          * If it had been mremap()'d, the starting address would not
1029          * match the usual checks anyway. So assume all vma's are
1030          * above the starting address given.
1031          */
1032         vma = find_vma(mm, addr);
1033
1034         while (vma) {
1035                 next = vma->vm_next;
1036
1037                 /*
1038                  * Check if the starting address would match, i.e. it's
1039                  * a fragment created by mprotect() and/or munmap(), or it
1040                  * otherwise it starts at this address with no hassles.
1041                  */
1042                 if ((vma->vm_ops == &shm_vm_ops) &&
1043                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1044
1045
1046                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1047                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1048                         /*
1049                          * We discovered the size of the shm segment, so
1050                          * break out of here and fall through to the next
1051                          * loop that uses the size information to stop
1052                          * searching for matching vma's.
1053                          */
1054                         retval = 0;
1055                         vma = next;
1056                         break;
1057                 }
1058                 vma = next;
1059         }
1060
1061         /*
1062          * We need look no further than the maximum address a fragment
1063          * could possibly have landed at. Also cast things to loff_t to
1064          * prevent overflows and make comparisions vs. equal-width types.
1065          */
1066         size = PAGE_ALIGN(size);
1067         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1068                 next = vma->vm_next;
1069
1070                 /* finding a matching vma now does not alter retval */
1071                 if ((vma->vm_ops == &shm_vm_ops) &&
1072                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1073
1074                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1075                 vma = next;
1076         }
1077
1078         up_write(&mm->mmap_sem);
1079         return retval;
1080 }
1081
1082 #ifdef CONFIG_PROC_FS
1083 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1084 {
1085         struct shmid_kernel *shp = it;
1086         char *format;
1087
1088 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1089 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1090
1091         if (sizeof(size_t) <= sizeof(int))
1092                 format = SMALL_STRING;
1093         else
1094                 format = BIG_STRING;
1095         return seq_printf(s, format,
1096                           shp->shm_perm.key,
1097                           shp->id,
1098                           shp->shm_perm.mode,
1099                           shp->shm_segsz,
1100                           shp->shm_cprid,
1101                           shp->shm_lprid,
1102                           shp->shm_nattch,
1103                           shp->shm_perm.uid,
1104                           shp->shm_perm.gid,
1105                           shp->shm_perm.cuid,
1106                           shp->shm_perm.cgid,
1107                           shp->shm_atim,
1108                           shp->shm_dtim,
1109                           shp->shm_ctim);
1110 }
1111 #endif