Do not account for the address space used by hugetlbfs using VM_ACCOUNT
[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
64 static int newseg(struct ipc_namespace *, struct ipc_params *);
65 static void shm_open(struct vm_area_struct *vma);
66 static void shm_close(struct vm_area_struct *vma);
67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
68 #ifdef CONFIG_PROC_FS
69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
70 #endif
71
72 void shm_init_ns(struct ipc_namespace *ns)
73 {
74         ns->shm_ctlmax = SHMMAX;
75         ns->shm_ctlall = SHMALL;
76         ns->shm_ctlmni = SHMMNI;
77         ns->shm_tot = 0;
78         ipc_init_ids(&shm_ids(ns));
79 }
80
81 /*
82  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
83  * Only shm_ids.rw_mutex remains locked on exit.
84  */
85 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
86 {
87         struct shmid_kernel *shp;
88         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
89
90         if (shp->shm_nattch){
91                 shp->shm_perm.mode |= SHM_DEST;
92                 /* Do not find it any more */
93                 shp->shm_perm.key = IPC_PRIVATE;
94                 shm_unlock(shp);
95         } else
96                 shm_destroy(ns, shp);
97 }
98
99 #ifdef CONFIG_IPC_NS
100 void shm_exit_ns(struct ipc_namespace *ns)
101 {
102         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
103 }
104 #endif
105
106 void __init shm_init (void)
107 {
108         shm_init_ns(&init_ipc_ns);
109         ipc_init_proc_interface("sysvipc/shm",
110                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
111                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
112 }
113
114 /*
115  * shm_lock_(check_) routines are called in the paths where the rw_mutex
116  * is not necessarily held.
117  */
118 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
119 {
120         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
121
122         if (IS_ERR(ipcp))
123                 return (struct shmid_kernel *)ipcp;
124
125         return container_of(ipcp, struct shmid_kernel, shm_perm);
126 }
127
128 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
129                                                 int id)
130 {
131         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
132
133         if (IS_ERR(ipcp))
134                 return (struct shmid_kernel *)ipcp;
135
136         return container_of(ipcp, struct shmid_kernel, shm_perm);
137 }
138
139 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
140 {
141         ipc_rmid(&shm_ids(ns), &s->shm_perm);
142 }
143
144
145 /* This is called by fork, once for every shm attach. */
146 static void shm_open(struct vm_area_struct *vma)
147 {
148         struct file *file = vma->vm_file;
149         struct shm_file_data *sfd = shm_file_data(file);
150         struct shmid_kernel *shp;
151
152         shp = shm_lock(sfd->ns, sfd->id);
153         BUG_ON(IS_ERR(shp));
154         shp->shm_atim = get_seconds();
155         shp->shm_lprid = task_tgid_vnr(current);
156         shp->shm_nattch++;
157         shm_unlock(shp);
158 }
159
160 /*
161  * shm_destroy - free the struct shmid_kernel
162  *
163  * @ns: namespace
164  * @shp: struct to free
165  *
166  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
167  * but returns with shp unlocked and freed.
168  */
169 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
170 {
171         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
172         shm_rmid(ns, shp);
173         shm_unlock(shp);
174         if (!is_file_hugepages(shp->shm_file))
175                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
176         else
177                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
178                                                 shp->mlock_user);
179         fput (shp->shm_file);
180         security_shm_free(shp);
181         ipc_rcu_putref(shp);
182 }
183
184 /*
185  * remove the attach descriptor vma.
186  * free memory for segment if it is marked destroyed.
187  * The descriptor has already been removed from the current->mm->mmap list
188  * and will later be kfree()d.
189  */
190 static void shm_close(struct vm_area_struct *vma)
191 {
192         struct file * file = vma->vm_file;
193         struct shm_file_data *sfd = shm_file_data(file);
194         struct shmid_kernel *shp;
195         struct ipc_namespace *ns = sfd->ns;
196
197         down_write(&shm_ids(ns).rw_mutex);
198         /* remove from the list of attaches of the shm segment */
199         shp = shm_lock(ns, sfd->id);
200         BUG_ON(IS_ERR(shp));
201         shp->shm_lprid = task_tgid_vnr(current);
202         shp->shm_dtim = get_seconds();
203         shp->shm_nattch--;
204         if(shp->shm_nattch == 0 &&
205            shp->shm_perm.mode & SHM_DEST)
206                 shm_destroy(ns, shp);
207         else
208                 shm_unlock(shp);
209         up_write(&shm_ids(ns).rw_mutex);
210 }
211
212 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
213 {
214         struct file *file = vma->vm_file;
215         struct shm_file_data *sfd = shm_file_data(file);
216
217         return sfd->vm_ops->fault(vma, vmf);
218 }
219
220 #ifdef CONFIG_NUMA
221 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
222 {
223         struct file *file = vma->vm_file;
224         struct shm_file_data *sfd = shm_file_data(file);
225         int err = 0;
226         if (sfd->vm_ops->set_policy)
227                 err = sfd->vm_ops->set_policy(vma, new);
228         return err;
229 }
230
231 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
232                                         unsigned long addr)
233 {
234         struct file *file = vma->vm_file;
235         struct shm_file_data *sfd = shm_file_data(file);
236         struct mempolicy *pol = NULL;
237
238         if (sfd->vm_ops->get_policy)
239                 pol = sfd->vm_ops->get_policy(vma, addr);
240         else if (vma->vm_policy)
241                 pol = vma->vm_policy;
242
243         return pol;
244 }
245 #endif
246
247 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
248 {
249         struct shm_file_data *sfd = shm_file_data(file);
250         int ret;
251
252         ret = sfd->file->f_op->mmap(sfd->file, vma);
253         if (ret != 0)
254                 return ret;
255         sfd->vm_ops = vma->vm_ops;
256 #ifdef CONFIG_MMU
257         BUG_ON(!sfd->vm_ops->fault);
258 #endif
259         vma->vm_ops = &shm_vm_ops;
260         shm_open(vma);
261
262         return ret;
263 }
264
265 static int shm_release(struct inode *ino, struct file *file)
266 {
267         struct shm_file_data *sfd = shm_file_data(file);
268
269         put_ipc_ns(sfd->ns);
270         shm_file_data(file) = NULL;
271         kfree(sfd);
272         return 0;
273 }
274
275 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
276 {
277         int (*fsync) (struct file *, struct dentry *, int datasync);
278         struct shm_file_data *sfd = shm_file_data(file);
279         int ret = -EINVAL;
280
281         fsync = sfd->file->f_op->fsync;
282         if (fsync)
283                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
284         return ret;
285 }
286
287 static unsigned long shm_get_unmapped_area(struct file *file,
288         unsigned long addr, unsigned long len, unsigned long pgoff,
289         unsigned long flags)
290 {
291         struct shm_file_data *sfd = shm_file_data(file);
292         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
293 }
294
295 int is_file_shm_hugepages(struct file *file)
296 {
297         int ret = 0;
298
299         if (file->f_op == &shm_file_operations) {
300                 struct shm_file_data *sfd;
301                 sfd = shm_file_data(file);
302                 ret = is_file_hugepages(sfd->file);
303         }
304         return ret;
305 }
306
307 static const struct file_operations shm_file_operations = {
308         .mmap           = shm_mmap,
309         .fsync          = shm_fsync,
310         .release        = shm_release,
311         .get_unmapped_area      = shm_get_unmapped_area,
312 };
313
314 static struct vm_operations_struct shm_vm_ops = {
315         .open   = shm_open,     /* callback for a new vm-area open */
316         .close  = shm_close,    /* callback for when the vm-area is released */
317         .fault  = shm_fault,
318 #if defined(CONFIG_NUMA)
319         .set_policy = shm_set_policy,
320         .get_policy = shm_get_policy,
321 #endif
322 };
323
324 /**
325  * newseg - Create a new shared memory segment
326  * @ns: namespace
327  * @params: ptr to the structure that contains key, size and shmflg
328  *
329  * Called with shm_ids.rw_mutex held as a writer.
330  */
331
332 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
333 {
334         key_t key = params->key;
335         int shmflg = params->flg;
336         size_t size = params->u.size;
337         int error;
338         struct shmid_kernel *shp;
339         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
340         struct file * file;
341         char name[13];
342         int id;
343         int acctflag = 0;
344
345         if (size < SHMMIN || size > ns->shm_ctlmax)
346                 return -EINVAL;
347
348         if (ns->shm_tot + numpages > ns->shm_ctlall)
349                 return -ENOSPC;
350
351         shp = ipc_rcu_alloc(sizeof(*shp));
352         if (!shp)
353                 return -ENOMEM;
354
355         shp->shm_perm.key = key;
356         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
357         shp->mlock_user = NULL;
358
359         shp->shm_perm.security = NULL;
360         error = security_shm_alloc(shp);
361         if (error) {
362                 ipc_rcu_putref(shp);
363                 return error;
364         }
365
366         sprintf (name, "SYSV%08x", key);
367         if (shmflg & SHM_HUGETLB) {
368                 /* hugetlb_file_setup applies strict accounting */
369                 if (shmflg & SHM_NORESERVE)
370                         acctflag = VM_NORESERVE;
371                 file = hugetlb_file_setup(name, size, acctflag);
372                 shp->mlock_user = current_user();
373         } else {
374                 /*
375                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
376                  * if it's asked for.
377                  */
378                 if  ((shmflg & SHM_NORESERVE) &&
379                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
380                         acctflag = VM_NORESERVE;
381                 file = shmem_file_setup(name, size, acctflag);
382         }
383         error = PTR_ERR(file);
384         if (IS_ERR(file))
385                 goto no_file;
386
387         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
388         if (id < 0) {
389                 error = id;
390                 goto no_id;
391         }
392
393         shp->shm_cprid = task_tgid_vnr(current);
394         shp->shm_lprid = 0;
395         shp->shm_atim = shp->shm_dtim = 0;
396         shp->shm_ctim = get_seconds();
397         shp->shm_segsz = size;
398         shp->shm_nattch = 0;
399         shp->shm_file = file;
400         /*
401          * shmid gets reported as "inode#" in /proc/pid/maps.
402          * proc-ps tools use this. Changing this will break them.
403          */
404         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
405
406         ns->shm_tot += numpages;
407         error = shp->shm_perm.id;
408         shm_unlock(shp);
409         return error;
410
411 no_id:
412         fput(file);
413 no_file:
414         security_shm_free(shp);
415         ipc_rcu_putref(shp);
416         return error;
417 }
418
419 /*
420  * Called with shm_ids.rw_mutex and ipcp locked.
421  */
422 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
423 {
424         struct shmid_kernel *shp;
425
426         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
427         return security_shm_associate(shp, shmflg);
428 }
429
430 /*
431  * Called with shm_ids.rw_mutex and ipcp locked.
432  */
433 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
434                                 struct ipc_params *params)
435 {
436         struct shmid_kernel *shp;
437
438         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
439         if (shp->shm_segsz < params->u.size)
440                 return -EINVAL;
441
442         return 0;
443 }
444
445 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
446 {
447         struct ipc_namespace *ns;
448         struct ipc_ops shm_ops;
449         struct ipc_params shm_params;
450
451         ns = current->nsproxy->ipc_ns;
452
453         shm_ops.getnew = newseg;
454         shm_ops.associate = shm_security;
455         shm_ops.more_checks = shm_more_checks;
456
457         shm_params.key = key;
458         shm_params.flg = shmflg;
459         shm_params.u.size = size;
460
461         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
462 }
463
464 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
465 {
466         switch(version) {
467         case IPC_64:
468                 return copy_to_user(buf, in, sizeof(*in));
469         case IPC_OLD:
470             {
471                 struct shmid_ds out;
472
473                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
474                 out.shm_segsz   = in->shm_segsz;
475                 out.shm_atime   = in->shm_atime;
476                 out.shm_dtime   = in->shm_dtime;
477                 out.shm_ctime   = in->shm_ctime;
478                 out.shm_cpid    = in->shm_cpid;
479                 out.shm_lpid    = in->shm_lpid;
480                 out.shm_nattch  = in->shm_nattch;
481
482                 return copy_to_user(buf, &out, sizeof(out));
483             }
484         default:
485                 return -EINVAL;
486         }
487 }
488
489 static inline unsigned long
490 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
491 {
492         switch(version) {
493         case IPC_64:
494                 if (copy_from_user(out, buf, sizeof(*out)))
495                         return -EFAULT;
496                 return 0;
497         case IPC_OLD:
498             {
499                 struct shmid_ds tbuf_old;
500
501                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
502                         return -EFAULT;
503
504                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
505                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
506                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
507
508                 return 0;
509             }
510         default:
511                 return -EINVAL;
512         }
513 }
514
515 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
516 {
517         switch(version) {
518         case IPC_64:
519                 return copy_to_user(buf, in, sizeof(*in));
520         case IPC_OLD:
521             {
522                 struct shminfo out;
523
524                 if(in->shmmax > INT_MAX)
525                         out.shmmax = INT_MAX;
526                 else
527                         out.shmmax = (int)in->shmmax;
528
529                 out.shmmin      = in->shmmin;
530                 out.shmmni      = in->shmmni;
531                 out.shmseg      = in->shmseg;
532                 out.shmall      = in->shmall; 
533
534                 return copy_to_user(buf, &out, sizeof(out));
535             }
536         default:
537                 return -EINVAL;
538         }
539 }
540
541 /*
542  * Called with shm_ids.rw_mutex held as a reader
543  */
544 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
545                 unsigned long *swp)
546 {
547         int next_id;
548         int total, in_use;
549
550         *rss = 0;
551         *swp = 0;
552
553         in_use = shm_ids(ns).in_use;
554
555         for (total = 0, next_id = 0; total < in_use; next_id++) {
556                 struct shmid_kernel *shp;
557                 struct inode *inode;
558
559                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
560                 if (shp == NULL)
561                         continue;
562
563                 inode = shp->shm_file->f_path.dentry->d_inode;
564
565                 if (is_file_hugepages(shp->shm_file)) {
566                         struct address_space *mapping = inode->i_mapping;
567                         struct hstate *h = hstate_file(shp->shm_file);
568                         *rss += pages_per_huge_page(h) * mapping->nrpages;
569                 } else {
570 #ifdef CONFIG_SHMEM
571                         struct shmem_inode_info *info = SHMEM_I(inode);
572                         spin_lock(&info->lock);
573                         *rss += inode->i_mapping->nrpages;
574                         *swp += info->swapped;
575                         spin_unlock(&info->lock);
576 #else
577                         *rss += inode->i_mapping->nrpages;
578 #endif
579                 }
580
581                 total++;
582         }
583 }
584
585 /*
586  * This function handles some shmctl commands which require the rw_mutex
587  * to be held in write mode.
588  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
589  */
590 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
591                        struct shmid_ds __user *buf, int version)
592 {
593         struct kern_ipc_perm *ipcp;
594         struct shmid64_ds shmid64;
595         struct shmid_kernel *shp;
596         int err;
597
598         if (cmd == IPC_SET) {
599                 if (copy_shmid_from_user(&shmid64, buf, version))
600                         return -EFAULT;
601         }
602
603         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
604         if (IS_ERR(ipcp))
605                 return PTR_ERR(ipcp);
606
607         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
608
609         err = security_shm_shmctl(shp, cmd);
610         if (err)
611                 goto out_unlock;
612         switch (cmd) {
613         case IPC_RMID:
614                 do_shm_rmid(ns, ipcp);
615                 goto out_up;
616         case IPC_SET:
617                 ipc_update_perm(&shmid64.shm_perm, ipcp);
618                 shp->shm_ctim = get_seconds();
619                 break;
620         default:
621                 err = -EINVAL;
622         }
623 out_unlock:
624         shm_unlock(shp);
625 out_up:
626         up_write(&shm_ids(ns).rw_mutex);
627         return err;
628 }
629
630 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
631 {
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 (cmd == SHM_STAT) {
702                         shp = shm_lock(ns, shmid);
703                         if (IS_ERR(shp)) {
704                                 err = PTR_ERR(shp);
705                                 goto out;
706                         }
707                         result = shp->shm_perm.id;
708                 } else {
709                         shp = shm_lock_check(ns, shmid);
710                         if (IS_ERR(shp)) {
711                                 err = PTR_ERR(shp);
712                                 goto out;
713                         }
714                         result = 0;
715                 }
716                 err = -EACCES;
717                 if (ipcperms (&shp->shm_perm, S_IRUGO))
718                         goto out_unlock;
719                 err = security_shm_shmctl(shp, cmd);
720                 if (err)
721                         goto out_unlock;
722                 memset(&tbuf, 0, sizeof(tbuf));
723                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
724                 tbuf.shm_segsz  = shp->shm_segsz;
725                 tbuf.shm_atime  = shp->shm_atim;
726                 tbuf.shm_dtime  = shp->shm_dtim;
727                 tbuf.shm_ctime  = shp->shm_ctim;
728                 tbuf.shm_cpid   = shp->shm_cprid;
729                 tbuf.shm_lpid   = shp->shm_lprid;
730                 tbuf.shm_nattch = shp->shm_nattch;
731                 shm_unlock(shp);
732                 if(copy_shmid_to_user (buf, &tbuf, version))
733                         err = -EFAULT;
734                 else
735                         err = result;
736                 goto out;
737         }
738         case SHM_LOCK:
739         case SHM_UNLOCK:
740         {
741                 struct file *uninitialized_var(shm_file);
742
743                 lru_add_drain_all();  /* drain pagevecs to lru lists */
744
745                 shp = shm_lock_check(ns, shmid);
746                 if (IS_ERR(shp)) {
747                         err = PTR_ERR(shp);
748                         goto out;
749                 }
750
751                 audit_ipc_obj(&(shp->shm_perm));
752
753                 if (!capable(CAP_IPC_LOCK)) {
754                         uid_t euid = current_euid();
755                         err = -EPERM;
756                         if (euid != shp->shm_perm.uid &&
757                             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         case IPC_SET:
787                 err = shmctl_down(ns, shmid, cmd, buf, version);
788                 return err;
789         default:
790                 return -EINVAL;
791         }
792
793 out_unlock:
794         shm_unlock(shp);
795 out:
796         return err;
797 }
798
799 /*
800  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
801  *
802  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
803  * "raddr" thing points to kernel space, and there has to be a wrapper around
804  * this.
805  */
806 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
807 {
808         struct shmid_kernel *shp;
809         unsigned long addr;
810         unsigned long size;
811         struct file * file;
812         int    err;
813         unsigned long flags;
814         unsigned long prot;
815         int acc_mode;
816         unsigned long user_addr;
817         struct ipc_namespace *ns;
818         struct shm_file_data *sfd;
819         struct path path;
820         fmode_t f_mode;
821
822         err = -EINVAL;
823         if (shmid < 0)
824                 goto out;
825         else if ((addr = (ulong)shmaddr)) {
826                 if (addr & (SHMLBA-1)) {
827                         if (shmflg & SHM_RND)
828                                 addr &= ~(SHMLBA-1);       /* round down */
829                         else
830 #ifndef __ARCH_FORCE_SHMLBA
831                                 if (addr & ~PAGE_MASK)
832 #endif
833                                         goto out;
834                 }
835                 flags = MAP_SHARED | MAP_FIXED;
836         } else {
837                 if ((shmflg & SHM_REMAP))
838                         goto out;
839
840                 flags = MAP_SHARED;
841         }
842
843         if (shmflg & SHM_RDONLY) {
844                 prot = PROT_READ;
845                 acc_mode = S_IRUGO;
846                 f_mode = FMODE_READ;
847         } else {
848                 prot = PROT_READ | PROT_WRITE;
849                 acc_mode = S_IRUGO | S_IWUGO;
850                 f_mode = FMODE_READ | FMODE_WRITE;
851         }
852         if (shmflg & SHM_EXEC) {
853                 prot |= PROT_EXEC;
854                 acc_mode |= S_IXUGO;
855         }
856
857         /*
858          * We cannot rely on the fs check since SYSV IPC does have an
859          * additional creator id...
860          */
861         ns = current->nsproxy->ipc_ns;
862         shp = shm_lock_check(ns, shmid);
863         if (IS_ERR(shp)) {
864                 err = PTR_ERR(shp);
865                 goto out;
866         }
867
868         err = -EACCES;
869         if (ipcperms(&shp->shm_perm, acc_mode))
870                 goto out_unlock;
871
872         err = security_shm_shmat(shp, shmaddr, shmflg);
873         if (err)
874                 goto out_unlock;
875
876         path.dentry = dget(shp->shm_file->f_path.dentry);
877         path.mnt    = shp->shm_file->f_path.mnt;
878         shp->shm_nattch++;
879         size = i_size_read(path.dentry->d_inode);
880         shm_unlock(shp);
881
882         err = -ENOMEM;
883         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
884         if (!sfd)
885                 goto out_put_dentry;
886
887         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
888         if (!file)
889                 goto out_free;
890
891         file->private_data = sfd;
892         file->f_mapping = shp->shm_file->f_mapping;
893         sfd->id = shp->shm_perm.id;
894         sfd->ns = get_ipc_ns(ns);
895         sfd->file = shp->shm_file;
896         sfd->vm_ops = NULL;
897
898         down_write(&current->mm->mmap_sem);
899         if (addr && !(shmflg & SHM_REMAP)) {
900                 err = -EINVAL;
901                 if (find_vma_intersection(current->mm, addr, addr + size))
902                         goto invalid;
903                 /*
904                  * If shm segment goes below stack, make sure there is some
905                  * space left for the stack to grow (at least 4 pages).
906                  */
907                 if (addr < current->mm->start_stack &&
908                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
909                         goto invalid;
910         }
911                 
912         user_addr = do_mmap (file, addr, size, prot, flags, 0);
913         *raddr = user_addr;
914         err = 0;
915         if (IS_ERR_VALUE(user_addr))
916                 err = (long)user_addr;
917 invalid:
918         up_write(&current->mm->mmap_sem);
919
920         fput(file);
921
922 out_nattch:
923         down_write(&shm_ids(ns).rw_mutex);
924         shp = shm_lock(ns, shmid);
925         BUG_ON(IS_ERR(shp));
926         shp->shm_nattch--;
927         if(shp->shm_nattch == 0 &&
928            shp->shm_perm.mode & SHM_DEST)
929                 shm_destroy(ns, shp);
930         else
931                 shm_unlock(shp);
932         up_write(&shm_ids(ns).rw_mutex);
933
934 out:
935         return err;
936
937 out_unlock:
938         shm_unlock(shp);
939         goto out;
940
941 out_free:
942         kfree(sfd);
943 out_put_dentry:
944         dput(path.dentry);
945         goto out_nattch;
946 }
947
948 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
949 {
950         unsigned long ret;
951         long err;
952
953         err = do_shmat(shmid, shmaddr, shmflg, &ret);
954         if (err)
955                 return err;
956         force_successful_syscall_return();
957         return (long)ret;
958 }
959
960 /*
961  * detach and kill segment if marked destroyed.
962  * The work is done in shm_close.
963  */
964 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
965 {
966         struct mm_struct *mm = current->mm;
967         struct vm_area_struct *vma, *next;
968         unsigned long addr = (unsigned long)shmaddr;
969         loff_t size = 0;
970         int retval = -EINVAL;
971
972         if (addr & ~PAGE_MASK)
973                 return retval;
974
975         down_write(&mm->mmap_sem);
976
977         /*
978          * This function tries to be smart and unmap shm segments that
979          * were modified by partial mlock or munmap calls:
980          * - It first determines the size of the shm segment that should be
981          *   unmapped: It searches for a vma that is backed by shm and that
982          *   started at address shmaddr. It records it's size and then unmaps
983          *   it.
984          * - Then it unmaps all shm vmas that started at shmaddr and that
985          *   are within the initially determined size.
986          * Errors from do_munmap are ignored: the function only fails if
987          * it's called with invalid parameters or if it's called to unmap
988          * a part of a vma. Both calls in this function are for full vmas,
989          * the parameters are directly copied from the vma itself and always
990          * valid - therefore do_munmap cannot fail. (famous last words?)
991          */
992         /*
993          * If it had been mremap()'d, the starting address would not
994          * match the usual checks anyway. So assume all vma's are
995          * above the starting address given.
996          */
997         vma = find_vma(mm, addr);
998
999 #ifdef CONFIG_MMU
1000         while (vma) {
1001                 next = vma->vm_next;
1002
1003                 /*
1004                  * Check if the starting address would match, i.e. it's
1005                  * a fragment created by mprotect() and/or munmap(), or it
1006                  * otherwise it starts at this address with no hassles.
1007                  */
1008                 if ((vma->vm_ops == &shm_vm_ops) &&
1009                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1010
1011
1012                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1013                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1014                         /*
1015                          * We discovered the size of the shm segment, so
1016                          * break out of here and fall through to the next
1017                          * loop that uses the size information to stop
1018                          * searching for matching vma's.
1019                          */
1020                         retval = 0;
1021                         vma = next;
1022                         break;
1023                 }
1024                 vma = next;
1025         }
1026
1027         /*
1028          * We need look no further than the maximum address a fragment
1029          * could possibly have landed at. Also cast things to loff_t to
1030          * prevent overflows and make comparisions vs. equal-width types.
1031          */
1032         size = PAGE_ALIGN(size);
1033         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1034                 next = vma->vm_next;
1035
1036                 /* finding a matching vma now does not alter retval */
1037                 if ((vma->vm_ops == &shm_vm_ops) &&
1038                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1039
1040                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1041                 vma = next;
1042         }
1043
1044 #else /* CONFIG_MMU */
1045         /* under NOMMU conditions, the exact address to be destroyed must be
1046          * given */
1047         retval = -EINVAL;
1048         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1049                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1050                 retval = 0;
1051         }
1052
1053 #endif
1054
1055         up_write(&mm->mmap_sem);
1056         return retval;
1057 }
1058
1059 #ifdef CONFIG_PROC_FS
1060 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1061 {
1062         struct shmid_kernel *shp = it;
1063
1064 #if BITS_PER_LONG <= 32
1065 #define SIZE_SPEC "%10lu"
1066 #else
1067 #define SIZE_SPEC "%21lu"
1068 #endif
1069
1070         return seq_printf(s,
1071                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1072                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1073                           shp->shm_perm.key,
1074                           shp->shm_perm.id,
1075                           shp->shm_perm.mode,
1076                           shp->shm_segsz,
1077                           shp->shm_cprid,
1078                           shp->shm_lprid,
1079                           shp->shm_nattch,
1080                           shp->shm_perm.uid,
1081                           shp->shm_perm.gid,
1082                           shp->shm_perm.cuid,
1083                           shp->shm_perm.cgid,
1084                           shp->shm_atim,
1085                           shp->shm_dtim,
1086                           shp->shm_ctim);
1087 }
1088 #endif