IPC: make struct ipc_ids static in ipc_namespace
[linux-3.10.git] / ipc / shm.c
1 /*
2  * linux/ipc/shm.c
3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
4  *       Many improvements/fixes by Bruno Haible.
5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7  *
8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15  *
16  * support for audit of ipc object properties and permission changes
17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18  *
19  * namespaces support
20  * OpenVZ, SWsoft Inc.
21  * Pavel Emelianov <xemul@openvz.org>
22  */
23
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42
43 #include <asm/uaccess.h>
44
45 #include "util.h"
46
47 struct shm_file_data {
48         int id;
49         struct ipc_namespace *ns;
50         struct file *file;
51         const struct vm_operations_struct *vm_ops;
52 };
53
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
56 static const struct file_operations shm_file_operations;
57 static struct vm_operations_struct shm_vm_ops;
58
59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
60
61 #define shm_unlock(shp)                 \
62         ipc_unlock(&(shp)->shm_perm)
63 #define shm_buildid(id, seq)    ipc_buildid(id, seq)
64
65 static int newseg(struct ipc_namespace *, struct ipc_params *);
66 static void shm_open(struct vm_area_struct *vma);
67 static void shm_close(struct vm_area_struct *vma);
68 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
69 #ifdef CONFIG_PROC_FS
70 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
71 #endif
72
73 void shm_init_ns(struct ipc_namespace *ns)
74 {
75         ns->shm_ctlmax = SHMMAX;
76         ns->shm_ctlall = SHMALL;
77         ns->shm_ctlmni = SHMMNI;
78         ns->shm_tot = 0;
79         ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
80 }
81
82 /*
83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84  * Only shm_ids.rw_mutex remains locked on exit.
85  */
86 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
87 {
88         if (shp->shm_nattch){
89                 shp->shm_perm.mode |= SHM_DEST;
90                 /* Do not find it any more */
91                 shp->shm_perm.key = IPC_PRIVATE;
92                 shm_unlock(shp);
93         } else
94                 shm_destroy(ns, shp);
95 }
96
97 #ifdef CONFIG_IPC_NS
98 void shm_exit_ns(struct ipc_namespace *ns)
99 {
100         struct shmid_kernel *shp;
101         struct kern_ipc_perm *perm;
102         int next_id;
103         int total, in_use;
104
105         down_write(&shm_ids(ns).rw_mutex);
106
107         in_use = shm_ids(ns).in_use;
108
109         for (total = 0, next_id = 0; total < in_use; next_id++) {
110                 perm = idr_find(&shm_ids(ns).ipcs_idr, next_id);
111                 if (perm == NULL)
112                         continue;
113                 ipc_lock_by_ptr(perm);
114                 shp = container_of(perm, struct shmid_kernel, shm_perm);
115                 do_shm_rmid(ns, shp);
116                 total++;
117         }
118         up_write(&shm_ids(ns).rw_mutex);
119 }
120 #endif
121
122 void __init shm_init (void)
123 {
124         shm_init_ns(&init_ipc_ns);
125         ipc_init_proc_interface("sysvipc/shm",
126                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
127                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
128 }
129
130 /*
131  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
132  * is held to protect access to the idr tree.
133  */
134 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
135                                                 int id)
136 {
137         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
138
139         if (IS_ERR(ipcp))
140                 return (struct shmid_kernel *)ipcp;
141
142         return container_of(ipcp, struct shmid_kernel, shm_perm);
143 }
144
145 static inline struct shmid_kernel *shm_lock_check_down(
146                                                 struct ipc_namespace *ns,
147                                                 int id)
148 {
149         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
150
151         if (IS_ERR(ipcp))
152                 return (struct shmid_kernel *)ipcp;
153
154         return container_of(ipcp, struct shmid_kernel, shm_perm);
155 }
156
157 /*
158  * shm_lock_(check_) routines are called in the paths where the rw_mutex
159  * is not held.
160  */
161 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
162 {
163         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
164
165         if (IS_ERR(ipcp))
166                 return (struct shmid_kernel *)ipcp;
167
168         return container_of(ipcp, struct shmid_kernel, shm_perm);
169 }
170
171 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
172                                                 int id)
173 {
174         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
175
176         if (IS_ERR(ipcp))
177                 return (struct shmid_kernel *)ipcp;
178
179         return container_of(ipcp, struct shmid_kernel, shm_perm);
180 }
181
182 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
183 {
184         ipc_rmid(&shm_ids(ns), &s->shm_perm);
185 }
186
187 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
188 {
189         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
190 }
191
192
193
194 /* This is called by fork, once for every shm attach. */
195 static void shm_open(struct vm_area_struct *vma)
196 {
197         struct file *file = vma->vm_file;
198         struct shm_file_data *sfd = shm_file_data(file);
199         struct shmid_kernel *shp;
200
201         shp = shm_lock(sfd->ns, sfd->id);
202         BUG_ON(IS_ERR(shp));
203         shp->shm_atim = get_seconds();
204         shp->shm_lprid = task_tgid_vnr(current);
205         shp->shm_nattch++;
206         shm_unlock(shp);
207 }
208
209 /*
210  * shm_destroy - free the struct shmid_kernel
211  *
212  * @ns: namespace
213  * @shp: struct to free
214  *
215  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
216  * but returns with shp unlocked and freed.
217  */
218 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
219 {
220         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
221         shm_rmid(ns, shp);
222         shm_unlock(shp);
223         if (!is_file_hugepages(shp->shm_file))
224                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
225         else
226                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
227                                                 shp->mlock_user);
228         fput (shp->shm_file);
229         security_shm_free(shp);
230         ipc_rcu_putref(shp);
231 }
232
233 /*
234  * remove the attach descriptor vma.
235  * free memory for segment if it is marked destroyed.
236  * The descriptor has already been removed from the current->mm->mmap list
237  * and will later be kfree()d.
238  */
239 static void shm_close(struct vm_area_struct *vma)
240 {
241         struct file * file = vma->vm_file;
242         struct shm_file_data *sfd = shm_file_data(file);
243         struct shmid_kernel *shp;
244         struct ipc_namespace *ns = sfd->ns;
245
246         down_write(&shm_ids(ns).rw_mutex);
247         /* remove from the list of attaches of the shm segment */
248         shp = shm_lock_down(ns, sfd->id);
249         BUG_ON(IS_ERR(shp));
250         shp->shm_lprid = task_tgid_vnr(current);
251         shp->shm_dtim = get_seconds();
252         shp->shm_nattch--;
253         if(shp->shm_nattch == 0 &&
254            shp->shm_perm.mode & SHM_DEST)
255                 shm_destroy(ns, shp);
256         else
257                 shm_unlock(shp);
258         up_write(&shm_ids(ns).rw_mutex);
259 }
260
261 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
262 {
263         struct file *file = vma->vm_file;
264         struct shm_file_data *sfd = shm_file_data(file);
265
266         return sfd->vm_ops->fault(vma, vmf);
267 }
268
269 #ifdef CONFIG_NUMA
270 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
271 {
272         struct file *file = vma->vm_file;
273         struct shm_file_data *sfd = shm_file_data(file);
274         int err = 0;
275         if (sfd->vm_ops->set_policy)
276                 err = sfd->vm_ops->set_policy(vma, new);
277         return err;
278 }
279
280 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
281                                         unsigned long addr)
282 {
283         struct file *file = vma->vm_file;
284         struct shm_file_data *sfd = shm_file_data(file);
285         struct mempolicy *pol = NULL;
286
287         if (sfd->vm_ops->get_policy)
288                 pol = sfd->vm_ops->get_policy(vma, addr);
289         else if (vma->vm_policy)
290                 pol = vma->vm_policy;
291         else
292                 pol = current->mempolicy;
293         return pol;
294 }
295 #endif
296
297 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
298 {
299         struct shm_file_data *sfd = shm_file_data(file);
300         int ret;
301
302         ret = sfd->file->f_op->mmap(sfd->file, vma);
303         if (ret != 0)
304                 return ret;
305         sfd->vm_ops = vma->vm_ops;
306 #ifdef CONFIG_MMU
307         BUG_ON(!sfd->vm_ops->fault);
308 #endif
309         vma->vm_ops = &shm_vm_ops;
310         shm_open(vma);
311
312         return ret;
313 }
314
315 static int shm_release(struct inode *ino, struct file *file)
316 {
317         struct shm_file_data *sfd = shm_file_data(file);
318
319         put_ipc_ns(sfd->ns);
320         shm_file_data(file) = NULL;
321         kfree(sfd);
322         return 0;
323 }
324
325 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
326 {
327         int (*fsync) (struct file *, struct dentry *, int datasync);
328         struct shm_file_data *sfd = shm_file_data(file);
329         int ret = -EINVAL;
330
331         fsync = sfd->file->f_op->fsync;
332         if (fsync)
333                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
334         return ret;
335 }
336
337 static unsigned long shm_get_unmapped_area(struct file *file,
338         unsigned long addr, unsigned long len, unsigned long pgoff,
339         unsigned long flags)
340 {
341         struct shm_file_data *sfd = shm_file_data(file);
342         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
343 }
344
345 int is_file_shm_hugepages(struct file *file)
346 {
347         int ret = 0;
348
349         if (file->f_op == &shm_file_operations) {
350                 struct shm_file_data *sfd;
351                 sfd = shm_file_data(file);
352                 ret = is_file_hugepages(sfd->file);
353         }
354         return ret;
355 }
356
357 static const struct file_operations shm_file_operations = {
358         .mmap           = shm_mmap,
359         .fsync          = shm_fsync,
360         .release        = shm_release,
361         .get_unmapped_area      = shm_get_unmapped_area,
362 };
363
364 static struct vm_operations_struct shm_vm_ops = {
365         .open   = shm_open,     /* callback for a new vm-area open */
366         .close  = shm_close,    /* callback for when the vm-area is released */
367         .fault  = shm_fault,
368 #if defined(CONFIG_NUMA)
369         .set_policy = shm_set_policy,
370         .get_policy = shm_get_policy,
371 #endif
372 };
373
374 /**
375  * newseg - Create a new shared memory segment
376  * @ns: namespace
377  * @params: ptr to the structure that contains key, size and shmflg
378  *
379  * Called with shm_ids.rw_mutex held as a writer.
380  */
381
382 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
383 {
384         key_t key = params->key;
385         int shmflg = params->flg;
386         size_t size = params->u.size;
387         int error;
388         struct shmid_kernel *shp;
389         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
390         struct file * file;
391         char name[13];
392         int id;
393
394         if (size < SHMMIN || size > ns->shm_ctlmax)
395                 return -EINVAL;
396
397         if (ns->shm_tot + numpages > ns->shm_ctlall)
398                 return -ENOSPC;
399
400         shp = ipc_rcu_alloc(sizeof(*shp));
401         if (!shp)
402                 return -ENOMEM;
403
404         shp->shm_perm.key = key;
405         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
406         shp->mlock_user = NULL;
407
408         shp->shm_perm.security = NULL;
409         error = security_shm_alloc(shp);
410         if (error) {
411                 ipc_rcu_putref(shp);
412                 return error;
413         }
414
415         sprintf (name, "SYSV%08x", key);
416         if (shmflg & SHM_HUGETLB) {
417                 /* hugetlb_file_setup takes care of mlock user accounting */
418                 file = hugetlb_file_setup(name, size);
419                 shp->mlock_user = current->user;
420         } else {
421                 int acctflag = VM_ACCOUNT;
422                 /*
423                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
424                  * if it's asked for.
425                  */
426                 if  ((shmflg & SHM_NORESERVE) &&
427                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
428                         acctflag = 0;
429                 file = shmem_file_setup(name, size, acctflag);
430         }
431         error = PTR_ERR(file);
432         if (IS_ERR(file))
433                 goto no_file;
434
435         id = shm_addid(ns, shp);
436         if (id < 0) {
437                 error = id;
438                 goto no_id;
439         }
440
441         shp->shm_cprid = task_tgid_vnr(current);
442         shp->shm_lprid = 0;
443         shp->shm_atim = shp->shm_dtim = 0;
444         shp->shm_ctim = get_seconds();
445         shp->shm_segsz = size;
446         shp->shm_nattch = 0;
447         shp->shm_perm.id = shm_buildid(id, shp->shm_perm.seq);
448         shp->shm_file = file;
449         /*
450          * shmid gets reported as "inode#" in /proc/pid/maps.
451          * proc-ps tools use this. Changing this will break them.
452          */
453         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
454
455         ns->shm_tot += numpages;
456         error = shp->shm_perm.id;
457         shm_unlock(shp);
458         return error;
459
460 no_id:
461         fput(file);
462 no_file:
463         security_shm_free(shp);
464         ipc_rcu_putref(shp);
465         return error;
466 }
467
468 /*
469  * Called with shm_ids.rw_mutex and ipcp locked.
470  */
471 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
472 {
473         struct shmid_kernel *shp;
474
475         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
476         return security_shm_associate(shp, shmflg);
477 }
478
479 /*
480  * Called with shm_ids.rw_mutex and ipcp locked.
481  */
482 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
483                                 struct ipc_params *params)
484 {
485         struct shmid_kernel *shp;
486
487         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
488         if (shp->shm_segsz < params->u.size)
489                 return -EINVAL;
490
491         return 0;
492 }
493
494 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
495 {
496         struct ipc_namespace *ns;
497         struct ipc_ops shm_ops;
498         struct ipc_params shm_params;
499
500         ns = current->nsproxy->ipc_ns;
501
502         shm_ops.getnew = newseg;
503         shm_ops.associate = shm_security;
504         shm_ops.more_checks = shm_more_checks;
505
506         shm_params.key = key;
507         shm_params.flg = shmflg;
508         shm_params.u.size = size;
509
510         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
511 }
512
513 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
514 {
515         switch(version) {
516         case IPC_64:
517                 return copy_to_user(buf, in, sizeof(*in));
518         case IPC_OLD:
519             {
520                 struct shmid_ds out;
521
522                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
523                 out.shm_segsz   = in->shm_segsz;
524                 out.shm_atime   = in->shm_atime;
525                 out.shm_dtime   = in->shm_dtime;
526                 out.shm_ctime   = in->shm_ctime;
527                 out.shm_cpid    = in->shm_cpid;
528                 out.shm_lpid    = in->shm_lpid;
529                 out.shm_nattch  = in->shm_nattch;
530
531                 return copy_to_user(buf, &out, sizeof(out));
532             }
533         default:
534                 return -EINVAL;
535         }
536 }
537
538 struct shm_setbuf {
539         uid_t   uid;
540         gid_t   gid;
541         mode_t  mode;
542 };      
543
544 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
545 {
546         switch(version) {
547         case IPC_64:
548             {
549                 struct shmid64_ds tbuf;
550
551                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
552                         return -EFAULT;
553
554                 out->uid        = tbuf.shm_perm.uid;
555                 out->gid        = tbuf.shm_perm.gid;
556                 out->mode       = tbuf.shm_perm.mode;
557
558                 return 0;
559             }
560         case IPC_OLD:
561             {
562                 struct shmid_ds tbuf_old;
563
564                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
565                         return -EFAULT;
566
567                 out->uid        = tbuf_old.shm_perm.uid;
568                 out->gid        = tbuf_old.shm_perm.gid;
569                 out->mode       = tbuf_old.shm_perm.mode;
570
571                 return 0;
572             }
573         default:
574                 return -EINVAL;
575         }
576 }
577
578 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
579 {
580         switch(version) {
581         case IPC_64:
582                 return copy_to_user(buf, in, sizeof(*in));
583         case IPC_OLD:
584             {
585                 struct shminfo out;
586
587                 if(in->shmmax > INT_MAX)
588                         out.shmmax = INT_MAX;
589                 else
590                         out.shmmax = (int)in->shmmax;
591
592                 out.shmmin      = in->shmmin;
593                 out.shmmni      = in->shmmni;
594                 out.shmseg      = in->shmseg;
595                 out.shmall      = in->shmall; 
596
597                 return copy_to_user(buf, &out, sizeof(out));
598             }
599         default:
600                 return -EINVAL;
601         }
602 }
603
604 /*
605  * Called with shm_ids.rw_mutex held as a reader
606  */
607 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
608                 unsigned long *swp)
609 {
610         int next_id;
611         int total, in_use;
612
613         *rss = 0;
614         *swp = 0;
615
616         in_use = shm_ids(ns).in_use;
617
618         for (total = 0, next_id = 0; total < in_use; next_id++) {
619                 struct shmid_kernel *shp;
620                 struct inode *inode;
621
622                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
623                 if (shp == NULL)
624                         continue;
625
626                 inode = shp->shm_file->f_path.dentry->d_inode;
627
628                 if (is_file_hugepages(shp->shm_file)) {
629                         struct address_space *mapping = inode->i_mapping;
630                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
631                 } else {
632                         struct shmem_inode_info *info = SHMEM_I(inode);
633                         spin_lock(&info->lock);
634                         *rss += inode->i_mapping->nrpages;
635                         *swp += info->swapped;
636                         spin_unlock(&info->lock);
637                 }
638
639                 total++;
640         }
641 }
642
643 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
644 {
645         struct shm_setbuf setbuf;
646         struct shmid_kernel *shp;
647         int err, version;
648         struct ipc_namespace *ns;
649
650         if (cmd < 0 || shmid < 0) {
651                 err = -EINVAL;
652                 goto out;
653         }
654
655         version = ipc_parse_version(&cmd);
656         ns = current->nsproxy->ipc_ns;
657
658         switch (cmd) { /* replace with proc interface ? */
659         case IPC_INFO:
660         {
661                 struct shminfo64 shminfo;
662
663                 err = security_shm_shmctl(NULL, cmd);
664                 if (err)
665                         return err;
666
667                 memset(&shminfo,0,sizeof(shminfo));
668                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
669                 shminfo.shmmax = ns->shm_ctlmax;
670                 shminfo.shmall = ns->shm_ctlall;
671
672                 shminfo.shmmin = SHMMIN;
673                 if(copy_shminfo_to_user (buf, &shminfo, version))
674                         return -EFAULT;
675
676                 down_read(&shm_ids(ns).rw_mutex);
677                 err = ipc_get_maxid(&shm_ids(ns));
678                 up_read(&shm_ids(ns).rw_mutex);
679
680                 if(err<0)
681                         err = 0;
682                 goto out;
683         }
684         case SHM_INFO:
685         {
686                 struct shm_info shm_info;
687
688                 err = security_shm_shmctl(NULL, cmd);
689                 if (err)
690                         return err;
691
692                 memset(&shm_info,0,sizeof(shm_info));
693                 down_read(&shm_ids(ns).rw_mutex);
694                 shm_info.used_ids = shm_ids(ns).in_use;
695                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
696                 shm_info.shm_tot = ns->shm_tot;
697                 shm_info.swap_attempts = 0;
698                 shm_info.swap_successes = 0;
699                 err = ipc_get_maxid(&shm_ids(ns));
700                 up_read(&shm_ids(ns).rw_mutex);
701                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
702                         err = -EFAULT;
703                         goto out;
704                 }
705
706                 err = err < 0 ? 0 : err;
707                 goto out;
708         }
709         case SHM_STAT:
710         case IPC_STAT:
711         {
712                 struct shmid64_ds tbuf;
713                 int result;
714
715                 if (!buf) {
716                         err = -EFAULT;
717                         goto out;
718                 }
719
720                 if (cmd == SHM_STAT) {
721                         shp = shm_lock(ns, shmid);
722                         if (IS_ERR(shp)) {
723                                 err = PTR_ERR(shp);
724                                 goto out;
725                         }
726                         result = shp->shm_perm.id;
727                 } else {
728                         shp = shm_lock_check(ns, shmid);
729                         if (IS_ERR(shp)) {
730                                 err = PTR_ERR(shp);
731                                 goto out;
732                         }
733                         result = 0;
734                 }
735                 err=-EACCES;
736                 if (ipcperms (&shp->shm_perm, S_IRUGO))
737                         goto out_unlock;
738                 err = security_shm_shmctl(shp, cmd);
739                 if (err)
740                         goto out_unlock;
741                 memset(&tbuf, 0, sizeof(tbuf));
742                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
743                 tbuf.shm_segsz  = shp->shm_segsz;
744                 tbuf.shm_atime  = shp->shm_atim;
745                 tbuf.shm_dtime  = shp->shm_dtim;
746                 tbuf.shm_ctime  = shp->shm_ctim;
747                 tbuf.shm_cpid   = shp->shm_cprid;
748                 tbuf.shm_lpid   = shp->shm_lprid;
749                 tbuf.shm_nattch = shp->shm_nattch;
750                 shm_unlock(shp);
751                 if(copy_shmid_to_user (buf, &tbuf, version))
752                         err = -EFAULT;
753                 else
754                         err = result;
755                 goto out;
756         }
757         case SHM_LOCK:
758         case SHM_UNLOCK:
759         {
760                 shp = shm_lock_check(ns, shmid);
761                 if (IS_ERR(shp)) {
762                         err = PTR_ERR(shp);
763                         goto out;
764                 }
765
766                 err = audit_ipc_obj(&(shp->shm_perm));
767                 if (err)
768                         goto out_unlock;
769
770                 if (!capable(CAP_IPC_LOCK)) {
771                         err = -EPERM;
772                         if (current->euid != shp->shm_perm.uid &&
773                             current->euid != shp->shm_perm.cuid)
774                                 goto out_unlock;
775                         if (cmd == SHM_LOCK &&
776                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
777                                 goto out_unlock;
778                 }
779
780                 err = security_shm_shmctl(shp, cmd);
781                 if (err)
782                         goto out_unlock;
783                 
784                 if(cmd==SHM_LOCK) {
785                         struct user_struct * user = current->user;
786                         if (!is_file_hugepages(shp->shm_file)) {
787                                 err = shmem_lock(shp->shm_file, 1, user);
788                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
789                                         shp->shm_perm.mode |= SHM_LOCKED;
790                                         shp->mlock_user = user;
791                                 }
792                         }
793                 } else if (!is_file_hugepages(shp->shm_file)) {
794                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
795                         shp->shm_perm.mode &= ~SHM_LOCKED;
796                         shp->mlock_user = NULL;
797                 }
798                 shm_unlock(shp);
799                 goto out;
800         }
801         case IPC_RMID:
802         {
803                 /*
804                  *      We cannot simply remove the file. The SVID states
805                  *      that the block remains until the last person
806                  *      detaches from it, then is deleted. A shmat() on
807                  *      an RMID segment is legal in older Linux and if 
808                  *      we change it apps break...
809                  *
810                  *      Instead we set a destroyed flag, and then blow
811                  *      the name away when the usage hits zero.
812                  */
813                 down_write(&shm_ids(ns).rw_mutex);
814                 shp = shm_lock_check_down(ns, shmid);
815                 if (IS_ERR(shp)) {
816                         err = PTR_ERR(shp);
817                         goto out_up;
818                 }
819
820                 err = audit_ipc_obj(&(shp->shm_perm));
821                 if (err)
822                         goto out_unlock_up;
823
824                 if (current->euid != shp->shm_perm.uid &&
825                     current->euid != shp->shm_perm.cuid && 
826                     !capable(CAP_SYS_ADMIN)) {
827                         err=-EPERM;
828                         goto out_unlock_up;
829                 }
830
831                 err = security_shm_shmctl(shp, cmd);
832                 if (err)
833                         goto out_unlock_up;
834
835                 do_shm_rmid(ns, shp);
836                 up_write(&shm_ids(ns).rw_mutex);
837                 goto out;
838         }
839
840         case IPC_SET:
841         {
842                 if (!buf) {
843                         err = -EFAULT;
844                         goto out;
845                 }
846
847                 if (copy_shmid_from_user (&setbuf, buf, version)) {
848                         err = -EFAULT;
849                         goto out;
850                 }
851                 down_write(&shm_ids(ns).rw_mutex);
852                 shp = shm_lock_check_down(ns, shmid);
853                 if (IS_ERR(shp)) {
854                         err = PTR_ERR(shp);
855                         goto out_up;
856                 }
857                 err = audit_ipc_obj(&(shp->shm_perm));
858                 if (err)
859                         goto out_unlock_up;
860                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
861                 if (err)
862                         goto out_unlock_up;
863                 err=-EPERM;
864                 if (current->euid != shp->shm_perm.uid &&
865                     current->euid != shp->shm_perm.cuid && 
866                     !capable(CAP_SYS_ADMIN)) {
867                         goto out_unlock_up;
868                 }
869
870                 err = security_shm_shmctl(shp, cmd);
871                 if (err)
872                         goto out_unlock_up;
873                 
874                 shp->shm_perm.uid = setbuf.uid;
875                 shp->shm_perm.gid = setbuf.gid;
876                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
877                         | (setbuf.mode & S_IRWXUGO);
878                 shp->shm_ctim = get_seconds();
879                 break;
880         }
881
882         default:
883                 err = -EINVAL;
884                 goto out;
885         }
886
887         err = 0;
888 out_unlock_up:
889         shm_unlock(shp);
890 out_up:
891         up_write(&shm_ids(ns).rw_mutex);
892         goto out;
893 out_unlock:
894         shm_unlock(shp);
895 out:
896         return err;
897 }
898
899 /*
900  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
901  *
902  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
903  * "raddr" thing points to kernel space, and there has to be a wrapper around
904  * this.
905  */
906 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
907 {
908         struct shmid_kernel *shp;
909         unsigned long addr;
910         unsigned long size;
911         struct file * file;
912         int    err;
913         unsigned long flags;
914         unsigned long prot;
915         int acc_mode;
916         unsigned long user_addr;
917         struct ipc_namespace *ns;
918         struct shm_file_data *sfd;
919         struct path path;
920         mode_t f_mode;
921
922         err = -EINVAL;
923         if (shmid < 0)
924                 goto out;
925         else if ((addr = (ulong)shmaddr)) {
926                 if (addr & (SHMLBA-1)) {
927                         if (shmflg & SHM_RND)
928                                 addr &= ~(SHMLBA-1);       /* round down */
929                         else
930 #ifndef __ARCH_FORCE_SHMLBA
931                                 if (addr & ~PAGE_MASK)
932 #endif
933                                         goto out;
934                 }
935                 flags = MAP_SHARED | MAP_FIXED;
936         } else {
937                 if ((shmflg & SHM_REMAP))
938                         goto out;
939
940                 flags = MAP_SHARED;
941         }
942
943         if (shmflg & SHM_RDONLY) {
944                 prot = PROT_READ;
945                 acc_mode = S_IRUGO;
946                 f_mode = FMODE_READ;
947         } else {
948                 prot = PROT_READ | PROT_WRITE;
949                 acc_mode = S_IRUGO | S_IWUGO;
950                 f_mode = FMODE_READ | FMODE_WRITE;
951         }
952         if (shmflg & SHM_EXEC) {
953                 prot |= PROT_EXEC;
954                 acc_mode |= S_IXUGO;
955         }
956
957         /*
958          * We cannot rely on the fs check since SYSV IPC does have an
959          * additional creator id...
960          */
961         ns = current->nsproxy->ipc_ns;
962         shp = shm_lock_check(ns, shmid);
963         if (IS_ERR(shp)) {
964                 err = PTR_ERR(shp);
965                 goto out;
966         }
967
968         err = -EACCES;
969         if (ipcperms(&shp->shm_perm, acc_mode))
970                 goto out_unlock;
971
972         err = security_shm_shmat(shp, shmaddr, shmflg);
973         if (err)
974                 goto out_unlock;
975
976         path.dentry = dget(shp->shm_file->f_path.dentry);
977         path.mnt    = shp->shm_file->f_path.mnt;
978         shp->shm_nattch++;
979         size = i_size_read(path.dentry->d_inode);
980         shm_unlock(shp);
981
982         err = -ENOMEM;
983         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
984         if (!sfd)
985                 goto out_put_dentry;
986
987         err = -ENOMEM;
988
989         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
990         if (!file)
991                 goto out_free;
992
993         file->private_data = sfd;
994         file->f_mapping = shp->shm_file->f_mapping;
995         sfd->id = shp->shm_perm.id;
996         sfd->ns = get_ipc_ns(ns);
997         sfd->file = shp->shm_file;
998         sfd->vm_ops = NULL;
999
1000         down_write(&current->mm->mmap_sem);
1001         if (addr && !(shmflg & SHM_REMAP)) {
1002                 err = -EINVAL;
1003                 if (find_vma_intersection(current->mm, addr, addr + size))
1004                         goto invalid;
1005                 /*
1006                  * If shm segment goes below stack, make sure there is some
1007                  * space left for the stack to grow (at least 4 pages).
1008                  */
1009                 if (addr < current->mm->start_stack &&
1010                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1011                         goto invalid;
1012         }
1013                 
1014         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1015         *raddr = user_addr;
1016         err = 0;
1017         if (IS_ERR_VALUE(user_addr))
1018                 err = (long)user_addr;
1019 invalid:
1020         up_write(&current->mm->mmap_sem);
1021
1022         fput(file);
1023
1024 out_nattch:
1025         down_write(&shm_ids(ns).rw_mutex);
1026         shp = shm_lock_down(ns, shmid);
1027         BUG_ON(IS_ERR(shp));
1028         shp->shm_nattch--;
1029         if(shp->shm_nattch == 0 &&
1030            shp->shm_perm.mode & SHM_DEST)
1031                 shm_destroy(ns, shp);
1032         else
1033                 shm_unlock(shp);
1034         up_write(&shm_ids(ns).rw_mutex);
1035
1036 out:
1037         return err;
1038
1039 out_unlock:
1040         shm_unlock(shp);
1041         goto out;
1042
1043 out_free:
1044         kfree(sfd);
1045 out_put_dentry:
1046         dput(path.dentry);
1047         goto out_nattch;
1048 }
1049
1050 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1051 {
1052         unsigned long ret;
1053         long err;
1054
1055         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1056         if (err)
1057                 return err;
1058         force_successful_syscall_return();
1059         return (long)ret;
1060 }
1061
1062 /*
1063  * detach and kill segment if marked destroyed.
1064  * The work is done in shm_close.
1065  */
1066 asmlinkage long sys_shmdt(char __user *shmaddr)
1067 {
1068         struct mm_struct *mm = current->mm;
1069         struct vm_area_struct *vma, *next;
1070         unsigned long addr = (unsigned long)shmaddr;
1071         loff_t size = 0;
1072         int retval = -EINVAL;
1073
1074         if (addr & ~PAGE_MASK)
1075                 return retval;
1076
1077         down_write(&mm->mmap_sem);
1078
1079         /*
1080          * This function tries to be smart and unmap shm segments that
1081          * were modified by partial mlock or munmap calls:
1082          * - It first determines the size of the shm segment that should be
1083          *   unmapped: It searches for a vma that is backed by shm and that
1084          *   started at address shmaddr. It records it's size and then unmaps
1085          *   it.
1086          * - Then it unmaps all shm vmas that started at shmaddr and that
1087          *   are within the initially determined size.
1088          * Errors from do_munmap are ignored: the function only fails if
1089          * it's called with invalid parameters or if it's called to unmap
1090          * a part of a vma. Both calls in this function are for full vmas,
1091          * the parameters are directly copied from the vma itself and always
1092          * valid - therefore do_munmap cannot fail. (famous last words?)
1093          */
1094         /*
1095          * If it had been mremap()'d, the starting address would not
1096          * match the usual checks anyway. So assume all vma's are
1097          * above the starting address given.
1098          */
1099         vma = find_vma(mm, addr);
1100
1101         while (vma) {
1102                 next = vma->vm_next;
1103
1104                 /*
1105                  * Check if the starting address would match, i.e. it's
1106                  * a fragment created by mprotect() and/or munmap(), or it
1107                  * otherwise it starts at this address with no hassles.
1108                  */
1109                 if ((vma->vm_ops == &shm_vm_ops) &&
1110                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1111
1112
1113                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1114                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1115                         /*
1116                          * We discovered the size of the shm segment, so
1117                          * break out of here and fall through to the next
1118                          * loop that uses the size information to stop
1119                          * searching for matching vma's.
1120                          */
1121                         retval = 0;
1122                         vma = next;
1123                         break;
1124                 }
1125                 vma = next;
1126         }
1127
1128         /*
1129          * We need look no further than the maximum address a fragment
1130          * could possibly have landed at. Also cast things to loff_t to
1131          * prevent overflows and make comparisions vs. equal-width types.
1132          */
1133         size = PAGE_ALIGN(size);
1134         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1135                 next = vma->vm_next;
1136
1137                 /* finding a matching vma now does not alter retval */
1138                 if ((vma->vm_ops == &shm_vm_ops) &&
1139                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1140
1141                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1142                 vma = next;
1143         }
1144
1145         up_write(&mm->mmap_sem);
1146         return retval;
1147 }
1148
1149 #ifdef CONFIG_PROC_FS
1150 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1151 {
1152         struct shmid_kernel *shp = it;
1153         char *format;
1154
1155 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1156 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1157
1158         if (sizeof(size_t) <= sizeof(int))
1159                 format = SMALL_STRING;
1160         else
1161                 format = BIG_STRING;
1162         return seq_printf(s, format,
1163                           shp->shm_perm.key,
1164                           shp->shm_perm.id,
1165                           shp->shm_perm.mode,
1166                           shp->shm_segsz,
1167                           shp->shm_cprid,
1168                           shp->shm_lprid,
1169                           shp->shm_nattch,
1170                           shp->shm_perm.uid,
1171                           shp->shm_perm.gid,
1172                           shp->shm_perm.cuid,
1173                           shp->shm_perm.cgid,
1174                           shp->shm_atim,
1175                           shp->shm_dtim,
1176                           shp->shm_ctim);
1177 }
1178 #endif