memstick: add support for 8-bit parallel mode
[linux-2.6.git] / ipc / shm.c
1 /*
2  * linux/ipc/shm.c
3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
4  *       Many improvements/fixes by Bruno Haible.
5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7  *
8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15  *
16  * support for audit of ipc object properties and permission changes
17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18  *
19  * namespaces support
20  * OpenVZ, SWsoft Inc.
21  * Pavel Emelianov <xemul@openvz.org>
22  */
23
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42
43 #include <asm/uaccess.h>
44
45 #include "util.h"
46
47 struct shm_file_data {
48         int id;
49         struct ipc_namespace *ns;
50         struct file *file;
51         const struct vm_operations_struct *vm_ops;
52 };
53
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
56 static const struct file_operations shm_file_operations;
57 static struct vm_operations_struct shm_vm_ops;
58
59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
60
61 #define shm_unlock(shp)                 \
62         ipc_unlock(&(shp)->shm_perm)
63 #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 kern_ipc_perm *ipcp)
87 {
88         struct shmid_kernel *shp;
89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
90
91         if (shp->shm_nattch){
92                 shp->shm_perm.mode |= SHM_DEST;
93                 /* Do not find it any more */
94                 shp->shm_perm.key = IPC_PRIVATE;
95                 shm_unlock(shp);
96         } else
97                 shm_destroy(ns, shp);
98 }
99
100 #ifdef CONFIG_IPC_NS
101 void shm_exit_ns(struct ipc_namespace *ns)
102 {
103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
104 }
105 #endif
106
107 void __init shm_init (void)
108 {
109         shm_init_ns(&init_ipc_ns);
110         ipc_init_proc_interface("sysvipc/shm",
111                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
112                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
113 }
114
115 /*
116  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
117  * is held to protect access to the idr tree.
118  */
119 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
120                                                 int id)
121 {
122         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
123
124         if (IS_ERR(ipcp))
125                 return (struct shmid_kernel *)ipcp;
126
127         return container_of(ipcp, struct shmid_kernel, shm_perm);
128 }
129
130 static inline struct shmid_kernel *shm_lock_check_down(
131                                                 struct ipc_namespace *ns,
132                                                 int id)
133 {
134         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
135
136         if (IS_ERR(ipcp))
137                 return (struct shmid_kernel *)ipcp;
138
139         return container_of(ipcp, struct shmid_kernel, shm_perm);
140 }
141
142 /*
143  * shm_lock_(check_) routines are called in the paths where the rw_mutex
144  * is not held.
145  */
146 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
147 {
148         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
149
150         if (IS_ERR(ipcp))
151                 return (struct shmid_kernel *)ipcp;
152
153         return container_of(ipcp, struct shmid_kernel, shm_perm);
154 }
155
156 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
157                                                 int id)
158 {
159         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
160
161         if (IS_ERR(ipcp))
162                 return (struct shmid_kernel *)ipcp;
163
164         return container_of(ipcp, struct shmid_kernel, shm_perm);
165 }
166
167 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
168 {
169         ipc_rmid(&shm_ids(ns), &s->shm_perm);
170 }
171
172 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
173 {
174         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
175 }
176
177
178
179 /* This is called by fork, once for every shm attach. */
180 static void shm_open(struct vm_area_struct *vma)
181 {
182         struct file *file = vma->vm_file;
183         struct shm_file_data *sfd = shm_file_data(file);
184         struct shmid_kernel *shp;
185
186         shp = shm_lock(sfd->ns, sfd->id);
187         BUG_ON(IS_ERR(shp));
188         shp->shm_atim = get_seconds();
189         shp->shm_lprid = task_tgid_vnr(current);
190         shp->shm_nattch++;
191         shm_unlock(shp);
192 }
193
194 /*
195  * shm_destroy - free the struct shmid_kernel
196  *
197  * @ns: namespace
198  * @shp: struct to free
199  *
200  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
201  * but returns with shp unlocked and freed.
202  */
203 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
204 {
205         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
206         shm_rmid(ns, shp);
207         shm_unlock(shp);
208         if (!is_file_hugepages(shp->shm_file))
209                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
210         else
211                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
212                                                 shp->mlock_user);
213         fput (shp->shm_file);
214         security_shm_free(shp);
215         ipc_rcu_putref(shp);
216 }
217
218 /*
219  * remove the attach descriptor vma.
220  * free memory for segment if it is marked destroyed.
221  * The descriptor has already been removed from the current->mm->mmap list
222  * and will later be kfree()d.
223  */
224 static void shm_close(struct vm_area_struct *vma)
225 {
226         struct file * file = vma->vm_file;
227         struct shm_file_data *sfd = shm_file_data(file);
228         struct shmid_kernel *shp;
229         struct ipc_namespace *ns = sfd->ns;
230
231         down_write(&shm_ids(ns).rw_mutex);
232         /* remove from the list of attaches of the shm segment */
233         shp = shm_lock_down(ns, sfd->id);
234         BUG_ON(IS_ERR(shp));
235         shp->shm_lprid = task_tgid_vnr(current);
236         shp->shm_dtim = get_seconds();
237         shp->shm_nattch--;
238         if(shp->shm_nattch == 0 &&
239            shp->shm_perm.mode & SHM_DEST)
240                 shm_destroy(ns, shp);
241         else
242                 shm_unlock(shp);
243         up_write(&shm_ids(ns).rw_mutex);
244 }
245
246 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
247 {
248         struct file *file = vma->vm_file;
249         struct shm_file_data *sfd = shm_file_data(file);
250
251         return sfd->vm_ops->fault(vma, vmf);
252 }
253
254 #ifdef CONFIG_NUMA
255 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
256 {
257         struct file *file = vma->vm_file;
258         struct shm_file_data *sfd = shm_file_data(file);
259         int err = 0;
260         if (sfd->vm_ops->set_policy)
261                 err = sfd->vm_ops->set_policy(vma, new);
262         return err;
263 }
264
265 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
266                                         unsigned long addr)
267 {
268         struct file *file = vma->vm_file;
269         struct shm_file_data *sfd = shm_file_data(file);
270         struct mempolicy *pol = NULL;
271
272         if (sfd->vm_ops->get_policy)
273                 pol = sfd->vm_ops->get_policy(vma, addr);
274         else if (vma->vm_policy) {
275                 pol = vma->vm_policy;
276                 mpol_get(pol);  /* get_vma_policy() expects this */
277         } else
278                 pol = current->mempolicy;
279         return pol;
280 }
281 #endif
282
283 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
284 {
285         struct shm_file_data *sfd = shm_file_data(file);
286         int ret;
287
288         ret = sfd->file->f_op->mmap(sfd->file, vma);
289         if (ret != 0)
290                 return ret;
291         sfd->vm_ops = vma->vm_ops;
292 #ifdef CONFIG_MMU
293         BUG_ON(!sfd->vm_ops->fault);
294 #endif
295         vma->vm_ops = &shm_vm_ops;
296         shm_open(vma);
297
298         return ret;
299 }
300
301 static int shm_release(struct inode *ino, struct file *file)
302 {
303         struct shm_file_data *sfd = shm_file_data(file);
304
305         put_ipc_ns(sfd->ns);
306         shm_file_data(file) = NULL;
307         kfree(sfd);
308         return 0;
309 }
310
311 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
312 {
313         int (*fsync) (struct file *, struct dentry *, int datasync);
314         struct shm_file_data *sfd = shm_file_data(file);
315         int ret = -EINVAL;
316
317         fsync = sfd->file->f_op->fsync;
318         if (fsync)
319                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
320         return ret;
321 }
322
323 static unsigned long shm_get_unmapped_area(struct file *file,
324         unsigned long addr, unsigned long len, unsigned long pgoff,
325         unsigned long flags)
326 {
327         struct shm_file_data *sfd = shm_file_data(file);
328         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
329 }
330
331 int is_file_shm_hugepages(struct file *file)
332 {
333         int ret = 0;
334
335         if (file->f_op == &shm_file_operations) {
336                 struct shm_file_data *sfd;
337                 sfd = shm_file_data(file);
338                 ret = is_file_hugepages(sfd->file);
339         }
340         return ret;
341 }
342
343 static const struct file_operations shm_file_operations = {
344         .mmap           = shm_mmap,
345         .fsync          = shm_fsync,
346         .release        = shm_release,
347         .get_unmapped_area      = shm_get_unmapped_area,
348 };
349
350 static struct vm_operations_struct shm_vm_ops = {
351         .open   = shm_open,     /* callback for a new vm-area open */
352         .close  = shm_close,    /* callback for when the vm-area is released */
353         .fault  = shm_fault,
354 #if defined(CONFIG_NUMA)
355         .set_policy = shm_set_policy,
356         .get_policy = shm_get_policy,
357 #endif
358 };
359
360 /**
361  * newseg - Create a new shared memory segment
362  * @ns: namespace
363  * @params: ptr to the structure that contains key, size and shmflg
364  *
365  * Called with shm_ids.rw_mutex held as a writer.
366  */
367
368 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
369 {
370         key_t key = params->key;
371         int shmflg = params->flg;
372         size_t size = params->u.size;
373         int error;
374         struct shmid_kernel *shp;
375         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
376         struct file * file;
377         char name[13];
378         int id;
379
380         if (size < SHMMIN || size > ns->shm_ctlmax)
381                 return -EINVAL;
382
383         if (ns->shm_tot + numpages > ns->shm_ctlall)
384                 return -ENOSPC;
385
386         shp = ipc_rcu_alloc(sizeof(*shp));
387         if (!shp)
388                 return -ENOMEM;
389
390         shp->shm_perm.key = key;
391         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
392         shp->mlock_user = NULL;
393
394         shp->shm_perm.security = NULL;
395         error = security_shm_alloc(shp);
396         if (error) {
397                 ipc_rcu_putref(shp);
398                 return error;
399         }
400
401         sprintf (name, "SYSV%08x", key);
402         if (shmflg & SHM_HUGETLB) {
403                 /* hugetlb_file_setup takes care of mlock user accounting */
404                 file = hugetlb_file_setup(name, size);
405                 shp->mlock_user = current->user;
406         } else {
407                 int acctflag = VM_ACCOUNT;
408                 /*
409                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
410                  * if it's asked for.
411                  */
412                 if  ((shmflg & SHM_NORESERVE) &&
413                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
414                         acctflag = 0;
415                 file = shmem_file_setup(name, size, acctflag);
416         }
417         error = PTR_ERR(file);
418         if (IS_ERR(file))
419                 goto no_file;
420
421         id = shm_addid(ns, shp);
422         if (id < 0) {
423                 error = id;
424                 goto no_id;
425         }
426
427         shp->shm_cprid = task_tgid_vnr(current);
428         shp->shm_lprid = 0;
429         shp->shm_atim = shp->shm_dtim = 0;
430         shp->shm_ctim = get_seconds();
431         shp->shm_segsz = size;
432         shp->shm_nattch = 0;
433         shp->shm_perm.id = shm_buildid(id, shp->shm_perm.seq);
434         shp->shm_file = file;
435         /*
436          * shmid gets reported as "inode#" in /proc/pid/maps.
437          * proc-ps tools use this. Changing this will break them.
438          */
439         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
440
441         ns->shm_tot += numpages;
442         error = shp->shm_perm.id;
443         shm_unlock(shp);
444         return error;
445
446 no_id:
447         fput(file);
448 no_file:
449         security_shm_free(shp);
450         ipc_rcu_putref(shp);
451         return error;
452 }
453
454 /*
455  * Called with shm_ids.rw_mutex and ipcp locked.
456  */
457 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
458 {
459         struct shmid_kernel *shp;
460
461         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
462         return security_shm_associate(shp, shmflg);
463 }
464
465 /*
466  * Called with shm_ids.rw_mutex and ipcp locked.
467  */
468 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
469                                 struct ipc_params *params)
470 {
471         struct shmid_kernel *shp;
472
473         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
474         if (shp->shm_segsz < params->u.size)
475                 return -EINVAL;
476
477         return 0;
478 }
479
480 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
481 {
482         struct ipc_namespace *ns;
483         struct ipc_ops shm_ops;
484         struct ipc_params shm_params;
485
486         ns = current->nsproxy->ipc_ns;
487
488         shm_ops.getnew = newseg;
489         shm_ops.associate = shm_security;
490         shm_ops.more_checks = shm_more_checks;
491
492         shm_params.key = key;
493         shm_params.flg = shmflg;
494         shm_params.u.size = size;
495
496         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
497 }
498
499 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
500 {
501         switch(version) {
502         case IPC_64:
503                 return copy_to_user(buf, in, sizeof(*in));
504         case IPC_OLD:
505             {
506                 struct shmid_ds out;
507
508                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
509                 out.shm_segsz   = in->shm_segsz;
510                 out.shm_atime   = in->shm_atime;
511                 out.shm_dtime   = in->shm_dtime;
512                 out.shm_ctime   = in->shm_ctime;
513                 out.shm_cpid    = in->shm_cpid;
514                 out.shm_lpid    = in->shm_lpid;
515                 out.shm_nattch  = in->shm_nattch;
516
517                 return copy_to_user(buf, &out, sizeof(out));
518             }
519         default:
520                 return -EINVAL;
521         }
522 }
523
524 struct shm_setbuf {
525         uid_t   uid;
526         gid_t   gid;
527         mode_t  mode;
528 };      
529
530 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
531 {
532         switch(version) {
533         case IPC_64:
534             {
535                 struct shmid64_ds tbuf;
536
537                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
538                         return -EFAULT;
539
540                 out->uid        = tbuf.shm_perm.uid;
541                 out->gid        = tbuf.shm_perm.gid;
542                 out->mode       = tbuf.shm_perm.mode;
543
544                 return 0;
545             }
546         case IPC_OLD:
547             {
548                 struct shmid_ds tbuf_old;
549
550                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
551                         return -EFAULT;
552
553                 out->uid        = tbuf_old.shm_perm.uid;
554                 out->gid        = tbuf_old.shm_perm.gid;
555                 out->mode       = tbuf_old.shm_perm.mode;
556
557                 return 0;
558             }
559         default:
560                 return -EINVAL;
561         }
562 }
563
564 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
565 {
566         switch(version) {
567         case IPC_64:
568                 return copy_to_user(buf, in, sizeof(*in));
569         case IPC_OLD:
570             {
571                 struct shminfo out;
572
573                 if(in->shmmax > INT_MAX)
574                         out.shmmax = INT_MAX;
575                 else
576                         out.shmmax = (int)in->shmmax;
577
578                 out.shmmin      = in->shmmin;
579                 out.shmmni      = in->shmmni;
580                 out.shmseg      = in->shmseg;
581                 out.shmall      = in->shmall; 
582
583                 return copy_to_user(buf, &out, sizeof(out));
584             }
585         default:
586                 return -EINVAL;
587         }
588 }
589
590 /*
591  * Called with shm_ids.rw_mutex held as a reader
592  */
593 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
594                 unsigned long *swp)
595 {
596         int next_id;
597         int total, in_use;
598
599         *rss = 0;
600         *swp = 0;
601
602         in_use = shm_ids(ns).in_use;
603
604         for (total = 0, next_id = 0; total < in_use; next_id++) {
605                 struct shmid_kernel *shp;
606                 struct inode *inode;
607
608                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
609                 if (shp == NULL)
610                         continue;
611
612                 inode = shp->shm_file->f_path.dentry->d_inode;
613
614                 if (is_file_hugepages(shp->shm_file)) {
615                         struct address_space *mapping = inode->i_mapping;
616                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
617                 } else {
618                         struct shmem_inode_info *info = SHMEM_I(inode);
619                         spin_lock(&info->lock);
620                         *rss += inode->i_mapping->nrpages;
621                         *swp += info->swapped;
622                         spin_unlock(&info->lock);
623                 }
624
625                 total++;
626         }
627 }
628
629 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
630 {
631         struct shm_setbuf setbuf;
632         struct shmid_kernel *shp;
633         int err, version;
634         struct ipc_namespace *ns;
635
636         if (cmd < 0 || shmid < 0) {
637                 err = -EINVAL;
638                 goto out;
639         }
640
641         version = ipc_parse_version(&cmd);
642         ns = current->nsproxy->ipc_ns;
643
644         switch (cmd) { /* replace with proc interface ? */
645         case IPC_INFO:
646         {
647                 struct shminfo64 shminfo;
648
649                 err = security_shm_shmctl(NULL, cmd);
650                 if (err)
651                         return err;
652
653                 memset(&shminfo,0,sizeof(shminfo));
654                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
655                 shminfo.shmmax = ns->shm_ctlmax;
656                 shminfo.shmall = ns->shm_ctlall;
657
658                 shminfo.shmmin = SHMMIN;
659                 if(copy_shminfo_to_user (buf, &shminfo, version))
660                         return -EFAULT;
661
662                 down_read(&shm_ids(ns).rw_mutex);
663                 err = ipc_get_maxid(&shm_ids(ns));
664                 up_read(&shm_ids(ns).rw_mutex);
665
666                 if(err<0)
667                         err = 0;
668                 goto out;
669         }
670         case SHM_INFO:
671         {
672                 struct shm_info shm_info;
673
674                 err = security_shm_shmctl(NULL, cmd);
675                 if (err)
676                         return err;
677
678                 memset(&shm_info,0,sizeof(shm_info));
679                 down_read(&shm_ids(ns).rw_mutex);
680                 shm_info.used_ids = shm_ids(ns).in_use;
681                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
682                 shm_info.shm_tot = ns->shm_tot;
683                 shm_info.swap_attempts = 0;
684                 shm_info.swap_successes = 0;
685                 err = ipc_get_maxid(&shm_ids(ns));
686                 up_read(&shm_ids(ns).rw_mutex);
687                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
688                         err = -EFAULT;
689                         goto out;
690                 }
691
692                 err = err < 0 ? 0 : err;
693                 goto out;
694         }
695         case SHM_STAT:
696         case IPC_STAT:
697         {
698                 struct shmid64_ds tbuf;
699                 int result;
700
701                 if (!buf) {
702                         err = -EFAULT;
703                         goto out;
704                 }
705
706                 if (cmd == SHM_STAT) {
707                         shp = shm_lock(ns, shmid);
708                         if (IS_ERR(shp)) {
709                                 err = PTR_ERR(shp);
710                                 goto out;
711                         }
712                         result = shp->shm_perm.id;
713                 } else {
714                         shp = shm_lock_check(ns, shmid);
715                         if (IS_ERR(shp)) {
716                                 err = PTR_ERR(shp);
717                                 goto out;
718                         }
719                         result = 0;
720                 }
721                 err=-EACCES;
722                 if (ipcperms (&shp->shm_perm, S_IRUGO))
723                         goto out_unlock;
724                 err = security_shm_shmctl(shp, cmd);
725                 if (err)
726                         goto out_unlock;
727                 memset(&tbuf, 0, sizeof(tbuf));
728                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
729                 tbuf.shm_segsz  = shp->shm_segsz;
730                 tbuf.shm_atime  = shp->shm_atim;
731                 tbuf.shm_dtime  = shp->shm_dtim;
732                 tbuf.shm_ctime  = shp->shm_ctim;
733                 tbuf.shm_cpid   = shp->shm_cprid;
734                 tbuf.shm_lpid   = shp->shm_lprid;
735                 tbuf.shm_nattch = shp->shm_nattch;
736                 shm_unlock(shp);
737                 if(copy_shmid_to_user (buf, &tbuf, version))
738                         err = -EFAULT;
739                 else
740                         err = result;
741                 goto out;
742         }
743         case SHM_LOCK:
744         case SHM_UNLOCK:
745         {
746                 shp = shm_lock_check(ns, shmid);
747                 if (IS_ERR(shp)) {
748                         err = PTR_ERR(shp);
749                         goto out;
750                 }
751
752                 err = audit_ipc_obj(&(shp->shm_perm));
753                 if (err)
754                         goto out_unlock;
755
756                 if (!capable(CAP_IPC_LOCK)) {
757                         err = -EPERM;
758                         if (current->euid != shp->shm_perm.uid &&
759                             current->euid != shp->shm_perm.cuid)
760                                 goto out_unlock;
761                         if (cmd == SHM_LOCK &&
762                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
763                                 goto out_unlock;
764                 }
765
766                 err = security_shm_shmctl(shp, cmd);
767                 if (err)
768                         goto out_unlock;
769                 
770                 if(cmd==SHM_LOCK) {
771                         struct user_struct * user = current->user;
772                         if (!is_file_hugepages(shp->shm_file)) {
773                                 err = shmem_lock(shp->shm_file, 1, user);
774                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
775                                         shp->shm_perm.mode |= SHM_LOCKED;
776                                         shp->mlock_user = user;
777                                 }
778                         }
779                 } else if (!is_file_hugepages(shp->shm_file)) {
780                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
781                         shp->shm_perm.mode &= ~SHM_LOCKED;
782                         shp->mlock_user = NULL;
783                 }
784                 shm_unlock(shp);
785                 goto out;
786         }
787         case IPC_RMID:
788         {
789                 /*
790                  *      We cannot simply remove the file. The SVID states
791                  *      that the block remains until the last person
792                  *      detaches from it, then is deleted. A shmat() on
793                  *      an RMID segment is legal in older Linux and if 
794                  *      we change it apps break...
795                  *
796                  *      Instead we set a destroyed flag, and then blow
797                  *      the name away when the usage hits zero.
798                  */
799                 down_write(&shm_ids(ns).rw_mutex);
800                 shp = shm_lock_check_down(ns, shmid);
801                 if (IS_ERR(shp)) {
802                         err = PTR_ERR(shp);
803                         goto out_up;
804                 }
805
806                 err = audit_ipc_obj(&(shp->shm_perm));
807                 if (err)
808                         goto out_unlock_up;
809
810                 if (current->euid != shp->shm_perm.uid &&
811                     current->euid != shp->shm_perm.cuid && 
812                     !capable(CAP_SYS_ADMIN)) {
813                         err=-EPERM;
814                         goto out_unlock_up;
815                 }
816
817                 err = security_shm_shmctl(shp, cmd);
818                 if (err)
819                         goto out_unlock_up;
820
821                 do_shm_rmid(ns, &shp->shm_perm);
822                 up_write(&shm_ids(ns).rw_mutex);
823                 goto out;
824         }
825
826         case IPC_SET:
827         {
828                 if (!buf) {
829                         err = -EFAULT;
830                         goto out;
831                 }
832
833                 if (copy_shmid_from_user (&setbuf, buf, version)) {
834                         err = -EFAULT;
835                         goto out;
836                 }
837                 down_write(&shm_ids(ns).rw_mutex);
838                 shp = shm_lock_check_down(ns, shmid);
839                 if (IS_ERR(shp)) {
840                         err = PTR_ERR(shp);
841                         goto out_up;
842                 }
843                 err = audit_ipc_obj(&(shp->shm_perm));
844                 if (err)
845                         goto out_unlock_up;
846                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
847                 if (err)
848                         goto out_unlock_up;
849                 err=-EPERM;
850                 if (current->euid != shp->shm_perm.uid &&
851                     current->euid != shp->shm_perm.cuid && 
852                     !capable(CAP_SYS_ADMIN)) {
853                         goto out_unlock_up;
854                 }
855
856                 err = security_shm_shmctl(shp, cmd);
857                 if (err)
858                         goto out_unlock_up;
859                 
860                 shp->shm_perm.uid = setbuf.uid;
861                 shp->shm_perm.gid = setbuf.gid;
862                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
863                         | (setbuf.mode & S_IRWXUGO);
864                 shp->shm_ctim = get_seconds();
865                 break;
866         }
867
868         default:
869                 err = -EINVAL;
870                 goto out;
871         }
872
873         err = 0;
874 out_unlock_up:
875         shm_unlock(shp);
876 out_up:
877         up_write(&shm_ids(ns).rw_mutex);
878         goto out;
879 out_unlock:
880         shm_unlock(shp);
881 out:
882         return err;
883 }
884
885 /*
886  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
887  *
888  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
889  * "raddr" thing points to kernel space, and there has to be a wrapper around
890  * this.
891  */
892 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
893 {
894         struct shmid_kernel *shp;
895         unsigned long addr;
896         unsigned long size;
897         struct file * file;
898         int    err;
899         unsigned long flags;
900         unsigned long prot;
901         int acc_mode;
902         unsigned long user_addr;
903         struct ipc_namespace *ns;
904         struct shm_file_data *sfd;
905         struct path path;
906         mode_t f_mode;
907
908         err = -EINVAL;
909         if (shmid < 0)
910                 goto out;
911         else if ((addr = (ulong)shmaddr)) {
912                 if (addr & (SHMLBA-1)) {
913                         if (shmflg & SHM_RND)
914                                 addr &= ~(SHMLBA-1);       /* round down */
915                         else
916 #ifndef __ARCH_FORCE_SHMLBA
917                                 if (addr & ~PAGE_MASK)
918 #endif
919                                         goto out;
920                 }
921                 flags = MAP_SHARED | MAP_FIXED;
922         } else {
923                 if ((shmflg & SHM_REMAP))
924                         goto out;
925
926                 flags = MAP_SHARED;
927         }
928
929         if (shmflg & SHM_RDONLY) {
930                 prot = PROT_READ;
931                 acc_mode = S_IRUGO;
932                 f_mode = FMODE_READ;
933         } else {
934                 prot = PROT_READ | PROT_WRITE;
935                 acc_mode = S_IRUGO | S_IWUGO;
936                 f_mode = FMODE_READ | FMODE_WRITE;
937         }
938         if (shmflg & SHM_EXEC) {
939                 prot |= PROT_EXEC;
940                 acc_mode |= S_IXUGO;
941         }
942
943         /*
944          * We cannot rely on the fs check since SYSV IPC does have an
945          * additional creator id...
946          */
947         ns = current->nsproxy->ipc_ns;
948         shp = shm_lock_check(ns, shmid);
949         if (IS_ERR(shp)) {
950                 err = PTR_ERR(shp);
951                 goto out;
952         }
953
954         err = -EACCES;
955         if (ipcperms(&shp->shm_perm, acc_mode))
956                 goto out_unlock;
957
958         err = security_shm_shmat(shp, shmaddr, shmflg);
959         if (err)
960                 goto out_unlock;
961
962         path.dentry = dget(shp->shm_file->f_path.dentry);
963         path.mnt    = shp->shm_file->f_path.mnt;
964         shp->shm_nattch++;
965         size = i_size_read(path.dentry->d_inode);
966         shm_unlock(shp);
967
968         err = -ENOMEM;
969         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
970         if (!sfd)
971                 goto out_put_dentry;
972
973         err = -ENOMEM;
974
975         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
976         if (!file)
977                 goto out_free;
978
979         file->private_data = sfd;
980         file->f_mapping = shp->shm_file->f_mapping;
981         sfd->id = shp->shm_perm.id;
982         sfd->ns = get_ipc_ns(ns);
983         sfd->file = shp->shm_file;
984         sfd->vm_ops = NULL;
985
986         down_write(&current->mm->mmap_sem);
987         if (addr && !(shmflg & SHM_REMAP)) {
988                 err = -EINVAL;
989                 if (find_vma_intersection(current->mm, addr, addr + size))
990                         goto invalid;
991                 /*
992                  * If shm segment goes below stack, make sure there is some
993                  * space left for the stack to grow (at least 4 pages).
994                  */
995                 if (addr < current->mm->start_stack &&
996                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
997                         goto invalid;
998         }
999                 
1000         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1001         *raddr = user_addr;
1002         err = 0;
1003         if (IS_ERR_VALUE(user_addr))
1004                 err = (long)user_addr;
1005 invalid:
1006         up_write(&current->mm->mmap_sem);
1007
1008         fput(file);
1009
1010 out_nattch:
1011         down_write(&shm_ids(ns).rw_mutex);
1012         shp = shm_lock_down(ns, shmid);
1013         BUG_ON(IS_ERR(shp));
1014         shp->shm_nattch--;
1015         if(shp->shm_nattch == 0 &&
1016            shp->shm_perm.mode & SHM_DEST)
1017                 shm_destroy(ns, shp);
1018         else
1019                 shm_unlock(shp);
1020         up_write(&shm_ids(ns).rw_mutex);
1021
1022 out:
1023         return err;
1024
1025 out_unlock:
1026         shm_unlock(shp);
1027         goto out;
1028
1029 out_free:
1030         kfree(sfd);
1031 out_put_dentry:
1032         dput(path.dentry);
1033         goto out_nattch;
1034 }
1035
1036 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1037 {
1038         unsigned long ret;
1039         long err;
1040
1041         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1042         if (err)
1043                 return err;
1044         force_successful_syscall_return();
1045         return (long)ret;
1046 }
1047
1048 /*
1049  * detach and kill segment if marked destroyed.
1050  * The work is done in shm_close.
1051  */
1052 asmlinkage long sys_shmdt(char __user *shmaddr)
1053 {
1054         struct mm_struct *mm = current->mm;
1055         struct vm_area_struct *vma, *next;
1056         unsigned long addr = (unsigned long)shmaddr;
1057         loff_t size = 0;
1058         int retval = -EINVAL;
1059
1060         if (addr & ~PAGE_MASK)
1061                 return retval;
1062
1063         down_write(&mm->mmap_sem);
1064
1065         /*
1066          * This function tries to be smart and unmap shm segments that
1067          * were modified by partial mlock or munmap calls:
1068          * - It first determines the size of the shm segment that should be
1069          *   unmapped: It searches for a vma that is backed by shm and that
1070          *   started at address shmaddr. It records it's size and then unmaps
1071          *   it.
1072          * - Then it unmaps all shm vmas that started at shmaddr and that
1073          *   are within the initially determined size.
1074          * Errors from do_munmap are ignored: the function only fails if
1075          * it's called with invalid parameters or if it's called to unmap
1076          * a part of a vma. Both calls in this function are for full vmas,
1077          * the parameters are directly copied from the vma itself and always
1078          * valid - therefore do_munmap cannot fail. (famous last words?)
1079          */
1080         /*
1081          * If it had been mremap()'d, the starting address would not
1082          * match the usual checks anyway. So assume all vma's are
1083          * above the starting address given.
1084          */
1085         vma = find_vma(mm, addr);
1086
1087         while (vma) {
1088                 next = vma->vm_next;
1089
1090                 /*
1091                  * Check if the starting address would match, i.e. it's
1092                  * a fragment created by mprotect() and/or munmap(), or it
1093                  * otherwise it starts at this address with no hassles.
1094                  */
1095                 if ((vma->vm_ops == &shm_vm_ops) &&
1096                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1097
1098
1099                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1100                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1101                         /*
1102                          * We discovered the size of the shm segment, so
1103                          * break out of here and fall through to the next
1104                          * loop that uses the size information to stop
1105                          * searching for matching vma's.
1106                          */
1107                         retval = 0;
1108                         vma = next;
1109                         break;
1110                 }
1111                 vma = next;
1112         }
1113
1114         /*
1115          * We need look no further than the maximum address a fragment
1116          * could possibly have landed at. Also cast things to loff_t to
1117          * prevent overflows and make comparisions vs. equal-width types.
1118          */
1119         size = PAGE_ALIGN(size);
1120         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1121                 next = vma->vm_next;
1122
1123                 /* finding a matching vma now does not alter retval */
1124                 if ((vma->vm_ops == &shm_vm_ops) &&
1125                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1126
1127                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1128                 vma = next;
1129         }
1130
1131         up_write(&mm->mmap_sem);
1132         return retval;
1133 }
1134
1135 #ifdef CONFIG_PROC_FS
1136 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1137 {
1138         struct shmid_kernel *shp = it;
1139         char *format;
1140
1141 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1142 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1143
1144         if (sizeof(size_t) <= sizeof(int))
1145                 format = SMALL_STRING;
1146         else
1147                 format = BIG_STRING;
1148         return seq_printf(s, format,
1149                           shp->shm_perm.key,
1150                           shp->shm_perm.id,
1151                           shp->shm_perm.mode,
1152                           shp->shm_segsz,
1153                           shp->shm_cprid,
1154                           shp->shm_lprid,
1155                           shp->shm_nattch,
1156                           shp->shm_perm.uid,
1157                           shp->shm_perm.gid,
1158                           shp->shm_perm.cuid,
1159                           shp->shm_perm.cgid,
1160                           shp->shm_atim,
1161                           shp->shm_dtim,
1162                           shp->shm_ctim);
1163 }
1164 #endif