e636910454a9b80ab9216498b09002c86d93e088
[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
277         return pol;
278 }
279 #endif
280
281 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
282 {
283         struct shm_file_data *sfd = shm_file_data(file);
284         int ret;
285
286         ret = sfd->file->f_op->mmap(sfd->file, vma);
287         if (ret != 0)
288                 return ret;
289         sfd->vm_ops = vma->vm_ops;
290 #ifdef CONFIG_MMU
291         BUG_ON(!sfd->vm_ops->fault);
292 #endif
293         vma->vm_ops = &shm_vm_ops;
294         shm_open(vma);
295
296         return ret;
297 }
298
299 static int shm_release(struct inode *ino, struct file *file)
300 {
301         struct shm_file_data *sfd = shm_file_data(file);
302
303         put_ipc_ns(sfd->ns);
304         shm_file_data(file) = NULL;
305         kfree(sfd);
306         return 0;
307 }
308
309 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
310 {
311         int (*fsync) (struct file *, struct dentry *, int datasync);
312         struct shm_file_data *sfd = shm_file_data(file);
313         int ret = -EINVAL;
314
315         fsync = sfd->file->f_op->fsync;
316         if (fsync)
317                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
318         return ret;
319 }
320
321 static unsigned long shm_get_unmapped_area(struct file *file,
322         unsigned long addr, unsigned long len, unsigned long pgoff,
323         unsigned long flags)
324 {
325         struct shm_file_data *sfd = shm_file_data(file);
326         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
327 }
328
329 int is_file_shm_hugepages(struct file *file)
330 {
331         int ret = 0;
332
333         if (file->f_op == &shm_file_operations) {
334                 struct shm_file_data *sfd;
335                 sfd = shm_file_data(file);
336                 ret = is_file_hugepages(sfd->file);
337         }
338         return ret;
339 }
340
341 static const struct file_operations shm_file_operations = {
342         .mmap           = shm_mmap,
343         .fsync          = shm_fsync,
344         .release        = shm_release,
345         .get_unmapped_area      = shm_get_unmapped_area,
346 };
347
348 static struct vm_operations_struct shm_vm_ops = {
349         .open   = shm_open,     /* callback for a new vm-area open */
350         .close  = shm_close,    /* callback for when the vm-area is released */
351         .fault  = shm_fault,
352 #if defined(CONFIG_NUMA)
353         .set_policy = shm_set_policy,
354         .get_policy = shm_get_policy,
355 #endif
356 };
357
358 /**
359  * newseg - Create a new shared memory segment
360  * @ns: namespace
361  * @params: ptr to the structure that contains key, size and shmflg
362  *
363  * Called with shm_ids.rw_mutex held as a writer.
364  */
365
366 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
367 {
368         key_t key = params->key;
369         int shmflg = params->flg;
370         size_t size = params->u.size;
371         int error;
372         struct shmid_kernel *shp;
373         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
374         struct file * file;
375         char name[13];
376         int id;
377
378         if (size < SHMMIN || size > ns->shm_ctlmax)
379                 return -EINVAL;
380
381         if (ns->shm_tot + numpages > ns->shm_ctlall)
382                 return -ENOSPC;
383
384         shp = ipc_rcu_alloc(sizeof(*shp));
385         if (!shp)
386                 return -ENOMEM;
387
388         shp->shm_perm.key = key;
389         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
390         shp->mlock_user = NULL;
391
392         shp->shm_perm.security = NULL;
393         error = security_shm_alloc(shp);
394         if (error) {
395                 ipc_rcu_putref(shp);
396                 return error;
397         }
398
399         sprintf (name, "SYSV%08x", key);
400         if (shmflg & SHM_HUGETLB) {
401                 /* hugetlb_file_setup takes care of mlock user accounting */
402                 file = hugetlb_file_setup(name, size);
403                 shp->mlock_user = current->user;
404         } else {
405                 int acctflag = VM_ACCOUNT;
406                 /*
407                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
408                  * if it's asked for.
409                  */
410                 if  ((shmflg & SHM_NORESERVE) &&
411                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
412                         acctflag = 0;
413                 file = shmem_file_setup(name, size, acctflag);
414         }
415         error = PTR_ERR(file);
416         if (IS_ERR(file))
417                 goto no_file;
418
419         id = shm_addid(ns, shp);
420         if (id < 0) {
421                 error = id;
422                 goto no_id;
423         }
424
425         shp->shm_cprid = task_tgid_vnr(current);
426         shp->shm_lprid = 0;
427         shp->shm_atim = shp->shm_dtim = 0;
428         shp->shm_ctim = get_seconds();
429         shp->shm_segsz = size;
430         shp->shm_nattch = 0;
431         shp->shm_perm.id = shm_buildid(id, shp->shm_perm.seq);
432         shp->shm_file = file;
433         /*
434          * shmid gets reported as "inode#" in /proc/pid/maps.
435          * proc-ps tools use this. Changing this will break them.
436          */
437         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
438
439         ns->shm_tot += numpages;
440         error = shp->shm_perm.id;
441         shm_unlock(shp);
442         return error;
443
444 no_id:
445         fput(file);
446 no_file:
447         security_shm_free(shp);
448         ipc_rcu_putref(shp);
449         return error;
450 }
451
452 /*
453  * Called with shm_ids.rw_mutex and ipcp locked.
454  */
455 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
456 {
457         struct shmid_kernel *shp;
458
459         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
460         return security_shm_associate(shp, shmflg);
461 }
462
463 /*
464  * Called with shm_ids.rw_mutex and ipcp locked.
465  */
466 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
467                                 struct ipc_params *params)
468 {
469         struct shmid_kernel *shp;
470
471         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
472         if (shp->shm_segsz < params->u.size)
473                 return -EINVAL;
474
475         return 0;
476 }
477
478 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
479 {
480         struct ipc_namespace *ns;
481         struct ipc_ops shm_ops;
482         struct ipc_params shm_params;
483
484         ns = current->nsproxy->ipc_ns;
485
486         shm_ops.getnew = newseg;
487         shm_ops.associate = shm_security;
488         shm_ops.more_checks = shm_more_checks;
489
490         shm_params.key = key;
491         shm_params.flg = shmflg;
492         shm_params.u.size = size;
493
494         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
495 }
496
497 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
498 {
499         switch(version) {
500         case IPC_64:
501                 return copy_to_user(buf, in, sizeof(*in));
502         case IPC_OLD:
503             {
504                 struct shmid_ds out;
505
506                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
507                 out.shm_segsz   = in->shm_segsz;
508                 out.shm_atime   = in->shm_atime;
509                 out.shm_dtime   = in->shm_dtime;
510                 out.shm_ctime   = in->shm_ctime;
511                 out.shm_cpid    = in->shm_cpid;
512                 out.shm_lpid    = in->shm_lpid;
513                 out.shm_nattch  = in->shm_nattch;
514
515                 return copy_to_user(buf, &out, sizeof(out));
516             }
517         default:
518                 return -EINVAL;
519         }
520 }
521
522 struct shm_setbuf {
523         uid_t   uid;
524         gid_t   gid;
525         mode_t  mode;
526 };      
527
528 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
529 {
530         switch(version) {
531         case IPC_64:
532             {
533                 struct shmid64_ds tbuf;
534
535                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
536                         return -EFAULT;
537
538                 out->uid        = tbuf.shm_perm.uid;
539                 out->gid        = tbuf.shm_perm.gid;
540                 out->mode       = tbuf.shm_perm.mode;
541
542                 return 0;
543             }
544         case IPC_OLD:
545             {
546                 struct shmid_ds tbuf_old;
547
548                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
549                         return -EFAULT;
550
551                 out->uid        = tbuf_old.shm_perm.uid;
552                 out->gid        = tbuf_old.shm_perm.gid;
553                 out->mode       = tbuf_old.shm_perm.mode;
554
555                 return 0;
556             }
557         default:
558                 return -EINVAL;
559         }
560 }
561
562 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
563 {
564         switch(version) {
565         case IPC_64:
566                 return copy_to_user(buf, in, sizeof(*in));
567         case IPC_OLD:
568             {
569                 struct shminfo out;
570
571                 if(in->shmmax > INT_MAX)
572                         out.shmmax = INT_MAX;
573                 else
574                         out.shmmax = (int)in->shmmax;
575
576                 out.shmmin      = in->shmmin;
577                 out.shmmni      = in->shmmni;
578                 out.shmseg      = in->shmseg;
579                 out.shmall      = in->shmall; 
580
581                 return copy_to_user(buf, &out, sizeof(out));
582             }
583         default:
584                 return -EINVAL;
585         }
586 }
587
588 /*
589  * Called with shm_ids.rw_mutex held as a reader
590  */
591 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
592                 unsigned long *swp)
593 {
594         int next_id;
595         int total, in_use;
596
597         *rss = 0;
598         *swp = 0;
599
600         in_use = shm_ids(ns).in_use;
601
602         for (total = 0, next_id = 0; total < in_use; next_id++) {
603                 struct shmid_kernel *shp;
604                 struct inode *inode;
605
606                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
607                 if (shp == NULL)
608                         continue;
609
610                 inode = shp->shm_file->f_path.dentry->d_inode;
611
612                 if (is_file_hugepages(shp->shm_file)) {
613                         struct address_space *mapping = inode->i_mapping;
614                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
615                 } else {
616                         struct shmem_inode_info *info = SHMEM_I(inode);
617                         spin_lock(&info->lock);
618                         *rss += inode->i_mapping->nrpages;
619                         *swp += info->swapped;
620                         spin_unlock(&info->lock);
621                 }
622
623                 total++;
624         }
625 }
626
627 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
628 {
629         struct shm_setbuf setbuf;
630         struct shmid_kernel *shp;
631         int err, version;
632         struct ipc_namespace *ns;
633
634         if (cmd < 0 || shmid < 0) {
635                 err = -EINVAL;
636                 goto out;
637         }
638
639         version = ipc_parse_version(&cmd);
640         ns = current->nsproxy->ipc_ns;
641
642         switch (cmd) { /* replace with proc interface ? */
643         case IPC_INFO:
644         {
645                 struct shminfo64 shminfo;
646
647                 err = security_shm_shmctl(NULL, cmd);
648                 if (err)
649                         return err;
650
651                 memset(&shminfo,0,sizeof(shminfo));
652                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
653                 shminfo.shmmax = ns->shm_ctlmax;
654                 shminfo.shmall = ns->shm_ctlall;
655
656                 shminfo.shmmin = SHMMIN;
657                 if(copy_shminfo_to_user (buf, &shminfo, version))
658                         return -EFAULT;
659
660                 down_read(&shm_ids(ns).rw_mutex);
661                 err = ipc_get_maxid(&shm_ids(ns));
662                 up_read(&shm_ids(ns).rw_mutex);
663
664                 if(err<0)
665                         err = 0;
666                 goto out;
667         }
668         case SHM_INFO:
669         {
670                 struct shm_info shm_info;
671
672                 err = security_shm_shmctl(NULL, cmd);
673                 if (err)
674                         return err;
675
676                 memset(&shm_info,0,sizeof(shm_info));
677                 down_read(&shm_ids(ns).rw_mutex);
678                 shm_info.used_ids = shm_ids(ns).in_use;
679                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
680                 shm_info.shm_tot = ns->shm_tot;
681                 shm_info.swap_attempts = 0;
682                 shm_info.swap_successes = 0;
683                 err = ipc_get_maxid(&shm_ids(ns));
684                 up_read(&shm_ids(ns).rw_mutex);
685                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
686                         err = -EFAULT;
687                         goto out;
688                 }
689
690                 err = err < 0 ? 0 : err;
691                 goto out;
692         }
693         case SHM_STAT:
694         case IPC_STAT:
695         {
696                 struct shmid64_ds tbuf;
697                 int result;
698
699                 if (!buf) {
700                         err = -EFAULT;
701                         goto out;
702                 }
703
704                 if (cmd == SHM_STAT) {
705                         shp = shm_lock(ns, shmid);
706                         if (IS_ERR(shp)) {
707                                 err = PTR_ERR(shp);
708                                 goto out;
709                         }
710                         result = shp->shm_perm.id;
711                 } else {
712                         shp = shm_lock_check(ns, shmid);
713                         if (IS_ERR(shp)) {
714                                 err = PTR_ERR(shp);
715                                 goto out;
716                         }
717                         result = 0;
718                 }
719                 err=-EACCES;
720                 if (ipcperms (&shp->shm_perm, S_IRUGO))
721                         goto out_unlock;
722                 err = security_shm_shmctl(shp, cmd);
723                 if (err)
724                         goto out_unlock;
725                 memset(&tbuf, 0, sizeof(tbuf));
726                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
727                 tbuf.shm_segsz  = shp->shm_segsz;
728                 tbuf.shm_atime  = shp->shm_atim;
729                 tbuf.shm_dtime  = shp->shm_dtim;
730                 tbuf.shm_ctime  = shp->shm_ctim;
731                 tbuf.shm_cpid   = shp->shm_cprid;
732                 tbuf.shm_lpid   = shp->shm_lprid;
733                 tbuf.shm_nattch = shp->shm_nattch;
734                 shm_unlock(shp);
735                 if(copy_shmid_to_user (buf, &tbuf, version))
736                         err = -EFAULT;
737                 else
738                         err = result;
739                 goto out;
740         }
741         case SHM_LOCK:
742         case SHM_UNLOCK:
743         {
744                 shp = shm_lock_check(ns, shmid);
745                 if (IS_ERR(shp)) {
746                         err = PTR_ERR(shp);
747                         goto out;
748                 }
749
750                 err = audit_ipc_obj(&(shp->shm_perm));
751                 if (err)
752                         goto out_unlock;
753
754                 if (!capable(CAP_IPC_LOCK)) {
755                         err = -EPERM;
756                         if (current->euid != shp->shm_perm.uid &&
757                             current->euid != shp->shm_perm.cuid)
758                                 goto out_unlock;
759                         if (cmd == SHM_LOCK &&
760                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
761                                 goto out_unlock;
762                 }
763
764                 err = security_shm_shmctl(shp, cmd);
765                 if (err)
766                         goto out_unlock;
767                 
768                 if(cmd==SHM_LOCK) {
769                         struct user_struct * user = current->user;
770                         if (!is_file_hugepages(shp->shm_file)) {
771                                 err = shmem_lock(shp->shm_file, 1, user);
772                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
773                                         shp->shm_perm.mode |= SHM_LOCKED;
774                                         shp->mlock_user = user;
775                                 }
776                         }
777                 } else if (!is_file_hugepages(shp->shm_file)) {
778                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
779                         shp->shm_perm.mode &= ~SHM_LOCKED;
780                         shp->mlock_user = NULL;
781                 }
782                 shm_unlock(shp);
783                 goto out;
784         }
785         case IPC_RMID:
786         {
787                 /*
788                  *      We cannot simply remove the file. The SVID states
789                  *      that the block remains until the last person
790                  *      detaches from it, then is deleted. A shmat() on
791                  *      an RMID segment is legal in older Linux and if 
792                  *      we change it apps break...
793                  *
794                  *      Instead we set a destroyed flag, and then blow
795                  *      the name away when the usage hits zero.
796                  */
797                 down_write(&shm_ids(ns).rw_mutex);
798                 shp = shm_lock_check_down(ns, shmid);
799                 if (IS_ERR(shp)) {
800                         err = PTR_ERR(shp);
801                         goto out_up;
802                 }
803
804                 err = audit_ipc_obj(&(shp->shm_perm));
805                 if (err)
806                         goto out_unlock_up;
807
808                 if (current->euid != shp->shm_perm.uid &&
809                     current->euid != shp->shm_perm.cuid && 
810                     !capable(CAP_SYS_ADMIN)) {
811                         err=-EPERM;
812                         goto out_unlock_up;
813                 }
814
815                 err = security_shm_shmctl(shp, cmd);
816                 if (err)
817                         goto out_unlock_up;
818
819                 do_shm_rmid(ns, &shp->shm_perm);
820                 up_write(&shm_ids(ns).rw_mutex);
821                 goto out;
822         }
823
824         case IPC_SET:
825         {
826                 if (!buf) {
827                         err = -EFAULT;
828                         goto out;
829                 }
830
831                 if (copy_shmid_from_user (&setbuf, buf, version)) {
832                         err = -EFAULT;
833                         goto out;
834                 }
835                 down_write(&shm_ids(ns).rw_mutex);
836                 shp = shm_lock_check_down(ns, shmid);
837                 if (IS_ERR(shp)) {
838                         err = PTR_ERR(shp);
839                         goto out_up;
840                 }
841                 err = audit_ipc_obj(&(shp->shm_perm));
842                 if (err)
843                         goto out_unlock_up;
844                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
845                 if (err)
846                         goto out_unlock_up;
847                 err=-EPERM;
848                 if (current->euid != shp->shm_perm.uid &&
849                     current->euid != shp->shm_perm.cuid && 
850                     !capable(CAP_SYS_ADMIN)) {
851                         goto out_unlock_up;
852                 }
853
854                 err = security_shm_shmctl(shp, cmd);
855                 if (err)
856                         goto out_unlock_up;
857                 
858                 shp->shm_perm.uid = setbuf.uid;
859                 shp->shm_perm.gid = setbuf.gid;
860                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
861                         | (setbuf.mode & S_IRWXUGO);
862                 shp->shm_ctim = get_seconds();
863                 break;
864         }
865
866         default:
867                 err = -EINVAL;
868                 goto out;
869         }
870
871         err = 0;
872 out_unlock_up:
873         shm_unlock(shp);
874 out_up:
875         up_write(&shm_ids(ns).rw_mutex);
876         goto out;
877 out_unlock:
878         shm_unlock(shp);
879 out:
880         return err;
881 }
882
883 /*
884  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
885  *
886  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
887  * "raddr" thing points to kernel space, and there has to be a wrapper around
888  * this.
889  */
890 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
891 {
892         struct shmid_kernel *shp;
893         unsigned long addr;
894         unsigned long size;
895         struct file * file;
896         int    err;
897         unsigned long flags;
898         unsigned long prot;
899         int acc_mode;
900         unsigned long user_addr;
901         struct ipc_namespace *ns;
902         struct shm_file_data *sfd;
903         struct path path;
904         mode_t f_mode;
905
906         err = -EINVAL;
907         if (shmid < 0)
908                 goto out;
909         else if ((addr = (ulong)shmaddr)) {
910                 if (addr & (SHMLBA-1)) {
911                         if (shmflg & SHM_RND)
912                                 addr &= ~(SHMLBA-1);       /* round down */
913                         else
914 #ifndef __ARCH_FORCE_SHMLBA
915                                 if (addr & ~PAGE_MASK)
916 #endif
917                                         goto out;
918                 }
919                 flags = MAP_SHARED | MAP_FIXED;
920         } else {
921                 if ((shmflg & SHM_REMAP))
922                         goto out;
923
924                 flags = MAP_SHARED;
925         }
926
927         if (shmflg & SHM_RDONLY) {
928                 prot = PROT_READ;
929                 acc_mode = S_IRUGO;
930                 f_mode = FMODE_READ;
931         } else {
932                 prot = PROT_READ | PROT_WRITE;
933                 acc_mode = S_IRUGO | S_IWUGO;
934                 f_mode = FMODE_READ | FMODE_WRITE;
935         }
936         if (shmflg & SHM_EXEC) {
937                 prot |= PROT_EXEC;
938                 acc_mode |= S_IXUGO;
939         }
940
941         /*
942          * We cannot rely on the fs check since SYSV IPC does have an
943          * additional creator id...
944          */
945         ns = current->nsproxy->ipc_ns;
946         shp = shm_lock_check(ns, shmid);
947         if (IS_ERR(shp)) {
948                 err = PTR_ERR(shp);
949                 goto out;
950         }
951
952         err = -EACCES;
953         if (ipcperms(&shp->shm_perm, acc_mode))
954                 goto out_unlock;
955
956         err = security_shm_shmat(shp, shmaddr, shmflg);
957         if (err)
958                 goto out_unlock;
959
960         path.dentry = dget(shp->shm_file->f_path.dentry);
961         path.mnt    = shp->shm_file->f_path.mnt;
962         shp->shm_nattch++;
963         size = i_size_read(path.dentry->d_inode);
964         shm_unlock(shp);
965
966         err = -ENOMEM;
967         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
968         if (!sfd)
969                 goto out_put_dentry;
970
971         err = -ENOMEM;
972
973         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
974         if (!file)
975                 goto out_free;
976
977         file->private_data = sfd;
978         file->f_mapping = shp->shm_file->f_mapping;
979         sfd->id = shp->shm_perm.id;
980         sfd->ns = get_ipc_ns(ns);
981         sfd->file = shp->shm_file;
982         sfd->vm_ops = NULL;
983
984         down_write(&current->mm->mmap_sem);
985         if (addr && !(shmflg & SHM_REMAP)) {
986                 err = -EINVAL;
987                 if (find_vma_intersection(current->mm, addr, addr + size))
988                         goto invalid;
989                 /*
990                  * If shm segment goes below stack, make sure there is some
991                  * space left for the stack to grow (at least 4 pages).
992                  */
993                 if (addr < current->mm->start_stack &&
994                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
995                         goto invalid;
996         }
997                 
998         user_addr = do_mmap (file, addr, size, prot, flags, 0);
999         *raddr = user_addr;
1000         err = 0;
1001         if (IS_ERR_VALUE(user_addr))
1002                 err = (long)user_addr;
1003 invalid:
1004         up_write(&current->mm->mmap_sem);
1005
1006         fput(file);
1007
1008 out_nattch:
1009         down_write(&shm_ids(ns).rw_mutex);
1010         shp = shm_lock_down(ns, shmid);
1011         BUG_ON(IS_ERR(shp));
1012         shp->shm_nattch--;
1013         if(shp->shm_nattch == 0 &&
1014            shp->shm_perm.mode & SHM_DEST)
1015                 shm_destroy(ns, shp);
1016         else
1017                 shm_unlock(shp);
1018         up_write(&shm_ids(ns).rw_mutex);
1019
1020 out:
1021         return err;
1022
1023 out_unlock:
1024         shm_unlock(shp);
1025         goto out;
1026
1027 out_free:
1028         kfree(sfd);
1029 out_put_dentry:
1030         dput(path.dentry);
1031         goto out_nattch;
1032 }
1033
1034 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1035 {
1036         unsigned long ret;
1037         long err;
1038
1039         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1040         if (err)
1041                 return err;
1042         force_successful_syscall_return();
1043         return (long)ret;
1044 }
1045
1046 /*
1047  * detach and kill segment if marked destroyed.
1048  * The work is done in shm_close.
1049  */
1050 asmlinkage long sys_shmdt(char __user *shmaddr)
1051 {
1052         struct mm_struct *mm = current->mm;
1053         struct vm_area_struct *vma, *next;
1054         unsigned long addr = (unsigned long)shmaddr;
1055         loff_t size = 0;
1056         int retval = -EINVAL;
1057
1058         if (addr & ~PAGE_MASK)
1059                 return retval;
1060
1061         down_write(&mm->mmap_sem);
1062
1063         /*
1064          * This function tries to be smart and unmap shm segments that
1065          * were modified by partial mlock or munmap calls:
1066          * - It first determines the size of the shm segment that should be
1067          *   unmapped: It searches for a vma that is backed by shm and that
1068          *   started at address shmaddr. It records it's size and then unmaps
1069          *   it.
1070          * - Then it unmaps all shm vmas that started at shmaddr and that
1071          *   are within the initially determined size.
1072          * Errors from do_munmap are ignored: the function only fails if
1073          * it's called with invalid parameters or if it's called to unmap
1074          * a part of a vma. Both calls in this function are for full vmas,
1075          * the parameters are directly copied from the vma itself and always
1076          * valid - therefore do_munmap cannot fail. (famous last words?)
1077          */
1078         /*
1079          * If it had been mremap()'d, the starting address would not
1080          * match the usual checks anyway. So assume all vma's are
1081          * above the starting address given.
1082          */
1083         vma = find_vma(mm, addr);
1084
1085         while (vma) {
1086                 next = vma->vm_next;
1087
1088                 /*
1089                  * Check if the starting address would match, i.e. it's
1090                  * a fragment created by mprotect() and/or munmap(), or it
1091                  * otherwise it starts at this address with no hassles.
1092                  */
1093                 if ((vma->vm_ops == &shm_vm_ops) &&
1094                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1095
1096
1097                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1098                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1099                         /*
1100                          * We discovered the size of the shm segment, so
1101                          * break out of here and fall through to the next
1102                          * loop that uses the size information to stop
1103                          * searching for matching vma's.
1104                          */
1105                         retval = 0;
1106                         vma = next;
1107                         break;
1108                 }
1109                 vma = next;
1110         }
1111
1112         /*
1113          * We need look no further than the maximum address a fragment
1114          * could possibly have landed at. Also cast things to loff_t to
1115          * prevent overflows and make comparisions vs. equal-width types.
1116          */
1117         size = PAGE_ALIGN(size);
1118         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1119                 next = vma->vm_next;
1120
1121                 /* finding a matching vma now does not alter retval */
1122                 if ((vma->vm_ops == &shm_vm_ops) &&
1123                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1124
1125                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1126                 vma = next;
1127         }
1128
1129         up_write(&mm->mmap_sem);
1130         return retval;
1131 }
1132
1133 #ifdef CONFIG_PROC_FS
1134 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1135 {
1136         struct shmid_kernel *shp = it;
1137         char *format;
1138
1139 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1140 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1141
1142         if (sizeof(size_t) <= sizeof(int))
1143                 format = SMALL_STRING;
1144         else
1145                 format = BIG_STRING;
1146         return seq_printf(s, format,
1147                           shp->shm_perm.key,
1148                           shp->shm_perm.id,
1149                           shp->shm_perm.mode,
1150                           shp->shm_segsz,
1151                           shp->shm_cprid,
1152                           shp->shm_lprid,
1153                           shp->shm_nattch,
1154                           shp->shm_perm.uid,
1155                           shp->shm_perm.gid,
1156                           shp->shm_perm.cuid,
1157                           shp->shm_perm.cgid,
1158                           shp->shm_atim,
1159                           shp->shm_dtim,
1160                           shp->shm_ctim);
1161 }
1162 #endif