Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[linux-2.6] / 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 #include <linux/ima.h>
43
44 #include <asm/uaccess.h>
45
46 #include "util.h"
47
48 struct shm_file_data {
49         int id;
50         struct ipc_namespace *ns;
51         struct file *file;
52         const struct vm_operations_struct *vm_ops;
53 };
54
55 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56
57 static const struct file_operations shm_file_operations;
58 static struct vm_operations_struct shm_vm_ops;
59
60 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
61
62 #define shm_unlock(shp)                 \
63         ipc_unlock(&(shp)->shm_perm)
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(&shm_ids(ns));
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_) routines are called in the paths where the rw_mutex
117  * is not necessarily held.
118  */
119 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
120 {
121         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
122
123         if (IS_ERR(ipcp))
124                 return (struct shmid_kernel *)ipcp;
125
126         return container_of(ipcp, struct shmid_kernel, shm_perm);
127 }
128
129 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
130                                                 int id)
131 {
132         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
133
134         if (IS_ERR(ipcp))
135                 return (struct shmid_kernel *)ipcp;
136
137         return container_of(ipcp, struct shmid_kernel, shm_perm);
138 }
139
140 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
141 {
142         ipc_rmid(&shm_ids(ns), &s->shm_perm);
143 }
144
145
146 /* This is called by fork, once for every shm attach. */
147 static void shm_open(struct vm_area_struct *vma)
148 {
149         struct file *file = vma->vm_file;
150         struct shm_file_data *sfd = shm_file_data(file);
151         struct shmid_kernel *shp;
152
153         shp = shm_lock(sfd->ns, sfd->id);
154         BUG_ON(IS_ERR(shp));
155         shp->shm_atim = get_seconds();
156         shp->shm_lprid = task_tgid_vnr(current);
157         shp->shm_nattch++;
158         shm_unlock(shp);
159 }
160
161 /*
162  * shm_destroy - free the struct shmid_kernel
163  *
164  * @ns: namespace
165  * @shp: struct to free
166  *
167  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
168  * but returns with shp unlocked and freed.
169  */
170 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
171 {
172         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
173         shm_rmid(ns, shp);
174         shm_unlock(shp);
175         if (!is_file_hugepages(shp->shm_file))
176                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
177         else
178                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
179                                                 shp->mlock_user);
180         fput (shp->shm_file);
181         security_shm_free(shp);
182         ipc_rcu_putref(shp);
183 }
184
185 /*
186  * remove the attach descriptor vma.
187  * free memory for segment if it is marked destroyed.
188  * The descriptor has already been removed from the current->mm->mmap list
189  * and will later be kfree()d.
190  */
191 static void shm_close(struct vm_area_struct *vma)
192 {
193         struct file * file = vma->vm_file;
194         struct shm_file_data *sfd = shm_file_data(file);
195         struct shmid_kernel *shp;
196         struct ipc_namespace *ns = sfd->ns;
197
198         down_write(&shm_ids(ns).rw_mutex);
199         /* remove from the list of attaches of the shm segment */
200         shp = shm_lock(ns, sfd->id);
201         BUG_ON(IS_ERR(shp));
202         shp->shm_lprid = task_tgid_vnr(current);
203         shp->shm_dtim = get_seconds();
204         shp->shm_nattch--;
205         if(shp->shm_nattch == 0 &&
206            shp->shm_perm.mode & SHM_DEST)
207                 shm_destroy(ns, shp);
208         else
209                 shm_unlock(shp);
210         up_write(&shm_ids(ns).rw_mutex);
211 }
212
213 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
214 {
215         struct file *file = vma->vm_file;
216         struct shm_file_data *sfd = shm_file_data(file);
217
218         return sfd->vm_ops->fault(vma, vmf);
219 }
220
221 #ifdef CONFIG_NUMA
222 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
223 {
224         struct file *file = vma->vm_file;
225         struct shm_file_data *sfd = shm_file_data(file);
226         int err = 0;
227         if (sfd->vm_ops->set_policy)
228                 err = sfd->vm_ops->set_policy(vma, new);
229         return err;
230 }
231
232 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
233                                         unsigned long addr)
234 {
235         struct file *file = vma->vm_file;
236         struct shm_file_data *sfd = shm_file_data(file);
237         struct mempolicy *pol = NULL;
238
239         if (sfd->vm_ops->get_policy)
240                 pol = sfd->vm_ops->get_policy(vma, addr);
241         else if (vma->vm_policy)
242                 pol = vma->vm_policy;
243
244         return pol;
245 }
246 #endif
247
248 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
249 {
250         struct shm_file_data *sfd = shm_file_data(file);
251         int ret;
252
253         ret = sfd->file->f_op->mmap(sfd->file, vma);
254         if (ret != 0)
255                 return ret;
256         sfd->vm_ops = vma->vm_ops;
257 #ifdef CONFIG_MMU
258         BUG_ON(!sfd->vm_ops->fault);
259 #endif
260         vma->vm_ops = &shm_vm_ops;
261         shm_open(vma);
262
263         return ret;
264 }
265
266 static int shm_release(struct inode *ino, struct file *file)
267 {
268         struct shm_file_data *sfd = shm_file_data(file);
269
270         put_ipc_ns(sfd->ns);
271         shm_file_data(file) = NULL;
272         kfree(sfd);
273         return 0;
274 }
275
276 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
277 {
278         int (*fsync) (struct file *, struct dentry *, int datasync);
279         struct shm_file_data *sfd = shm_file_data(file);
280         int ret = -EINVAL;
281
282         fsync = sfd->file->f_op->fsync;
283         if (fsync)
284                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
285         return ret;
286 }
287
288 static unsigned long shm_get_unmapped_area(struct file *file,
289         unsigned long addr, unsigned long len, unsigned long pgoff,
290         unsigned long flags)
291 {
292         struct shm_file_data *sfd = shm_file_data(file);
293         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
294 }
295
296 int is_file_shm_hugepages(struct file *file)
297 {
298         int ret = 0;
299
300         if (file->f_op == &shm_file_operations) {
301                 struct shm_file_data *sfd;
302                 sfd = shm_file_data(file);
303                 ret = is_file_hugepages(sfd->file);
304         }
305         return ret;
306 }
307
308 static const struct file_operations shm_file_operations = {
309         .mmap           = shm_mmap,
310         .fsync          = shm_fsync,
311         .release        = shm_release,
312         .get_unmapped_area      = shm_get_unmapped_area,
313 };
314
315 static struct vm_operations_struct shm_vm_ops = {
316         .open   = shm_open,     /* callback for a new vm-area open */
317         .close  = shm_close,    /* callback for when the vm-area is released */
318         .fault  = shm_fault,
319 #if defined(CONFIG_NUMA)
320         .set_policy = shm_set_policy,
321         .get_policy = shm_get_policy,
322 #endif
323 };
324
325 /**
326  * newseg - Create a new shared memory segment
327  * @ns: namespace
328  * @params: ptr to the structure that contains key, size and shmflg
329  *
330  * Called with shm_ids.rw_mutex held as a writer.
331  */
332
333 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
334 {
335         key_t key = params->key;
336         int shmflg = params->flg;
337         size_t size = params->u.size;
338         int error;
339         struct shmid_kernel *shp;
340         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
341         struct file * file;
342         char name[13];
343         int id;
344         int acctflag = 0;
345
346         if (size < SHMMIN || size > ns->shm_ctlmax)
347                 return -EINVAL;
348
349         if (ns->shm_tot + numpages > ns->shm_ctlall)
350                 return -ENOSPC;
351
352         shp = ipc_rcu_alloc(sizeof(*shp));
353         if (!shp)
354                 return -ENOMEM;
355
356         shp->shm_perm.key = key;
357         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
358         shp->mlock_user = NULL;
359
360         shp->shm_perm.security = NULL;
361         error = security_shm_alloc(shp);
362         if (error) {
363                 ipc_rcu_putref(shp);
364                 return error;
365         }
366
367         sprintf (name, "SYSV%08x", key);
368         if (shmflg & SHM_HUGETLB) {
369                 /* hugetlb_file_setup applies strict accounting */
370                 if (shmflg & SHM_NORESERVE)
371                         acctflag = VM_NORESERVE;
372                 file = hugetlb_file_setup(name, size, acctflag);
373                 shp->mlock_user = current_user();
374         } else {
375                 /*
376                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
377                  * if it's asked for.
378                  */
379                 if  ((shmflg & SHM_NORESERVE) &&
380                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
381                         acctflag = VM_NORESERVE;
382                 file = shmem_file_setup(name, size, acctflag);
383         }
384         error = PTR_ERR(file);
385         if (IS_ERR(file))
386                 goto no_file;
387
388         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
389         if (id < 0) {
390                 error = id;
391                 goto no_id;
392         }
393
394         shp->shm_cprid = task_tgid_vnr(current);
395         shp->shm_lprid = 0;
396         shp->shm_atim = shp->shm_dtim = 0;
397         shp->shm_ctim = get_seconds();
398         shp->shm_segsz = size;
399         shp->shm_nattch = 0;
400         shp->shm_file = file;
401         /*
402          * shmid gets reported as "inode#" in /proc/pid/maps.
403          * proc-ps tools use this. Changing this will break them.
404          */
405         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
406
407         ns->shm_tot += numpages;
408         error = shp->shm_perm.id;
409         shm_unlock(shp);
410         return error;
411
412 no_id:
413         fput(file);
414 no_file:
415         security_shm_free(shp);
416         ipc_rcu_putref(shp);
417         return error;
418 }
419
420 /*
421  * Called with shm_ids.rw_mutex and ipcp locked.
422  */
423 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
424 {
425         struct shmid_kernel *shp;
426
427         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
428         return security_shm_associate(shp, shmflg);
429 }
430
431 /*
432  * Called with shm_ids.rw_mutex and ipcp locked.
433  */
434 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
435                                 struct ipc_params *params)
436 {
437         struct shmid_kernel *shp;
438
439         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
440         if (shp->shm_segsz < params->u.size)
441                 return -EINVAL;
442
443         return 0;
444 }
445
446 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
447 {
448         struct ipc_namespace *ns;
449         struct ipc_ops shm_ops;
450         struct ipc_params shm_params;
451
452         ns = current->nsproxy->ipc_ns;
453
454         shm_ops.getnew = newseg;
455         shm_ops.associate = shm_security;
456         shm_ops.more_checks = shm_more_checks;
457
458         shm_params.key = key;
459         shm_params.flg = shmflg;
460         shm_params.u.size = size;
461
462         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
463 }
464
465 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
466 {
467         switch(version) {
468         case IPC_64:
469                 return copy_to_user(buf, in, sizeof(*in));
470         case IPC_OLD:
471             {
472                 struct shmid_ds out;
473
474                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
475                 out.shm_segsz   = in->shm_segsz;
476                 out.shm_atime   = in->shm_atime;
477                 out.shm_dtime   = in->shm_dtime;
478                 out.shm_ctime   = in->shm_ctime;
479                 out.shm_cpid    = in->shm_cpid;
480                 out.shm_lpid    = in->shm_lpid;
481                 out.shm_nattch  = in->shm_nattch;
482
483                 return copy_to_user(buf, &out, sizeof(out));
484             }
485         default:
486                 return -EINVAL;
487         }
488 }
489
490 static inline unsigned long
491 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
492 {
493         switch(version) {
494         case IPC_64:
495                 if (copy_from_user(out, buf, sizeof(*out)))
496                         return -EFAULT;
497                 return 0;
498         case IPC_OLD:
499             {
500                 struct shmid_ds tbuf_old;
501
502                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
503                         return -EFAULT;
504
505                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
506                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
507                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
508
509                 return 0;
510             }
511         default:
512                 return -EINVAL;
513         }
514 }
515
516 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
517 {
518         switch(version) {
519         case IPC_64:
520                 return copy_to_user(buf, in, sizeof(*in));
521         case IPC_OLD:
522             {
523                 struct shminfo out;
524
525                 if(in->shmmax > INT_MAX)
526                         out.shmmax = INT_MAX;
527                 else
528                         out.shmmax = (int)in->shmmax;
529
530                 out.shmmin      = in->shmmin;
531                 out.shmmni      = in->shmmni;
532                 out.shmseg      = in->shmseg;
533                 out.shmall      = in->shmall; 
534
535                 return copy_to_user(buf, &out, sizeof(out));
536             }
537         default:
538                 return -EINVAL;
539         }
540 }
541
542 /*
543  * Called with shm_ids.rw_mutex held as a reader
544  */
545 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
546                 unsigned long *swp)
547 {
548         int next_id;
549         int total, in_use;
550
551         *rss = 0;
552         *swp = 0;
553
554         in_use = shm_ids(ns).in_use;
555
556         for (total = 0, next_id = 0; total < in_use; next_id++) {
557                 struct kern_ipc_perm *ipc;
558                 struct shmid_kernel *shp;
559                 struct inode *inode;
560
561                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
562                 if (ipc == NULL)
563                         continue;
564                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
565
566                 inode = shp->shm_file->f_path.dentry->d_inode;
567
568                 if (is_file_hugepages(shp->shm_file)) {
569                         struct address_space *mapping = inode->i_mapping;
570                         struct hstate *h = hstate_file(shp->shm_file);
571                         *rss += pages_per_huge_page(h) * mapping->nrpages;
572                 } else {
573 #ifdef CONFIG_SHMEM
574                         struct shmem_inode_info *info = SHMEM_I(inode);
575                         spin_lock(&info->lock);
576                         *rss += inode->i_mapping->nrpages;
577                         *swp += info->swapped;
578                         spin_unlock(&info->lock);
579 #else
580                         *rss += inode->i_mapping->nrpages;
581 #endif
582                 }
583
584                 total++;
585         }
586 }
587
588 /*
589  * This function handles some shmctl commands which require the rw_mutex
590  * to be held in write mode.
591  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
592  */
593 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
594                        struct shmid_ds __user *buf, int version)
595 {
596         struct kern_ipc_perm *ipcp;
597         struct shmid64_ds shmid64;
598         struct shmid_kernel *shp;
599         int err;
600
601         if (cmd == IPC_SET) {
602                 if (copy_shmid_from_user(&shmid64, buf, version))
603                         return -EFAULT;
604         }
605
606         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
607         if (IS_ERR(ipcp))
608                 return PTR_ERR(ipcp);
609
610         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
611
612         err = security_shm_shmctl(shp, cmd);
613         if (err)
614                 goto out_unlock;
615         switch (cmd) {
616         case IPC_RMID:
617                 do_shm_rmid(ns, ipcp);
618                 goto out_up;
619         case IPC_SET:
620                 ipc_update_perm(&shmid64.shm_perm, ipcp);
621                 shp->shm_ctim = get_seconds();
622                 break;
623         default:
624                 err = -EINVAL;
625         }
626 out_unlock:
627         shm_unlock(shp);
628 out_up:
629         up_write(&shm_ids(ns).rw_mutex);
630         return err;
631 }
632
633 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
634 {
635         struct shmid_kernel *shp;
636         int err, version;
637         struct ipc_namespace *ns;
638
639         if (cmd < 0 || shmid < 0) {
640                 err = -EINVAL;
641                 goto out;
642         }
643
644         version = ipc_parse_version(&cmd);
645         ns = current->nsproxy->ipc_ns;
646
647         switch (cmd) { /* replace with proc interface ? */
648         case IPC_INFO:
649         {
650                 struct shminfo64 shminfo;
651
652                 err = security_shm_shmctl(NULL, cmd);
653                 if (err)
654                         return err;
655
656                 memset(&shminfo, 0, sizeof(shminfo));
657                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
658                 shminfo.shmmax = ns->shm_ctlmax;
659                 shminfo.shmall = ns->shm_ctlall;
660
661                 shminfo.shmmin = SHMMIN;
662                 if(copy_shminfo_to_user (buf, &shminfo, version))
663                         return -EFAULT;
664
665                 down_read(&shm_ids(ns).rw_mutex);
666                 err = ipc_get_maxid(&shm_ids(ns));
667                 up_read(&shm_ids(ns).rw_mutex);
668
669                 if(err<0)
670                         err = 0;
671                 goto out;
672         }
673         case SHM_INFO:
674         {
675                 struct shm_info shm_info;
676
677                 err = security_shm_shmctl(NULL, cmd);
678                 if (err)
679                         return err;
680
681                 memset(&shm_info, 0, sizeof(shm_info));
682                 down_read(&shm_ids(ns).rw_mutex);
683                 shm_info.used_ids = shm_ids(ns).in_use;
684                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
685                 shm_info.shm_tot = ns->shm_tot;
686                 shm_info.swap_attempts = 0;
687                 shm_info.swap_successes = 0;
688                 err = ipc_get_maxid(&shm_ids(ns));
689                 up_read(&shm_ids(ns).rw_mutex);
690                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
691                         err = -EFAULT;
692                         goto out;
693                 }
694
695                 err = err < 0 ? 0 : err;
696                 goto out;
697         }
698         case SHM_STAT:
699         case IPC_STAT:
700         {
701                 struct shmid64_ds tbuf;
702                 int result;
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                 struct file *uninitialized_var(shm_file);
745
746                 lru_add_drain_all();  /* drain pagevecs to lru lists */
747
748                 shp = shm_lock_check(ns, shmid);
749                 if (IS_ERR(shp)) {
750                         err = PTR_ERR(shp);
751                         goto out;
752                 }
753
754                 audit_ipc_obj(&(shp->shm_perm));
755
756                 if (!capable(CAP_IPC_LOCK)) {
757                         uid_t euid = current_euid();
758                         err = -EPERM;
759                         if (euid != shp->shm_perm.uid &&
760                             euid != shp->shm_perm.cuid)
761                                 goto out_unlock;
762                         if (cmd == SHM_LOCK &&
763                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
764                                 goto out_unlock;
765                 }
766
767                 err = security_shm_shmctl(shp, cmd);
768                 if (err)
769                         goto out_unlock;
770                 
771                 if(cmd==SHM_LOCK) {
772                         struct user_struct *user = current_user();
773                         if (!is_file_hugepages(shp->shm_file)) {
774                                 err = shmem_lock(shp->shm_file, 1, user);
775                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
776                                         shp->shm_perm.mode |= SHM_LOCKED;
777                                         shp->mlock_user = user;
778                                 }
779                         }
780                 } else if (!is_file_hugepages(shp->shm_file)) {
781                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
782                         shp->shm_perm.mode &= ~SHM_LOCKED;
783                         shp->mlock_user = NULL;
784                 }
785                 shm_unlock(shp);
786                 goto out;
787         }
788         case IPC_RMID:
789         case IPC_SET:
790                 err = shmctl_down(ns, shmid, cmd, buf, version);
791                 return err;
792         default:
793                 return -EINVAL;
794         }
795
796 out_unlock:
797         shm_unlock(shp);
798 out:
799         return err;
800 }
801
802 /*
803  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
804  *
805  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
806  * "raddr" thing points to kernel space, and there has to be a wrapper around
807  * this.
808  */
809 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
810 {
811         struct shmid_kernel *shp;
812         unsigned long addr;
813         unsigned long size;
814         struct file * file;
815         int    err;
816         unsigned long flags;
817         unsigned long prot;
818         int acc_mode;
819         unsigned long user_addr;
820         struct ipc_namespace *ns;
821         struct shm_file_data *sfd;
822         struct path path;
823         fmode_t f_mode;
824
825         err = -EINVAL;
826         if (shmid < 0)
827                 goto out;
828         else if ((addr = (ulong)shmaddr)) {
829                 if (addr & (SHMLBA-1)) {
830                         if (shmflg & SHM_RND)
831                                 addr &= ~(SHMLBA-1);       /* round down */
832                         else
833 #ifndef __ARCH_FORCE_SHMLBA
834                                 if (addr & ~PAGE_MASK)
835 #endif
836                                         goto out;
837                 }
838                 flags = MAP_SHARED | MAP_FIXED;
839         } else {
840                 if ((shmflg & SHM_REMAP))
841                         goto out;
842
843                 flags = MAP_SHARED;
844         }
845
846         if (shmflg & SHM_RDONLY) {
847                 prot = PROT_READ;
848                 acc_mode = S_IRUGO;
849                 f_mode = FMODE_READ;
850         } else {
851                 prot = PROT_READ | PROT_WRITE;
852                 acc_mode = S_IRUGO | S_IWUGO;
853                 f_mode = FMODE_READ | FMODE_WRITE;
854         }
855         if (shmflg & SHM_EXEC) {
856                 prot |= PROT_EXEC;
857                 acc_mode |= S_IXUGO;
858         }
859
860         /*
861          * We cannot rely on the fs check since SYSV IPC does have an
862          * additional creator id...
863          */
864         ns = current->nsproxy->ipc_ns;
865         shp = shm_lock_check(ns, shmid);
866         if (IS_ERR(shp)) {
867                 err = PTR_ERR(shp);
868                 goto out;
869         }
870
871         err = -EACCES;
872         if (ipcperms(&shp->shm_perm, acc_mode))
873                 goto out_unlock;
874
875         err = security_shm_shmat(shp, shmaddr, shmflg);
876         if (err)
877                 goto out_unlock;
878
879         path.dentry = dget(shp->shm_file->f_path.dentry);
880         path.mnt    = shp->shm_file->f_path.mnt;
881         shp->shm_nattch++;
882         size = i_size_read(path.dentry->d_inode);
883         shm_unlock(shp);
884
885         err = -ENOMEM;
886         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
887         if (!sfd)
888                 goto out_put_dentry;
889
890         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
891         if (!file)
892                 goto out_free;
893         ima_counts_get(file);
894
895         file->private_data = sfd;
896         file->f_mapping = shp->shm_file->f_mapping;
897         sfd->id = shp->shm_perm.id;
898         sfd->ns = get_ipc_ns(ns);
899         sfd->file = shp->shm_file;
900         sfd->vm_ops = NULL;
901
902         down_write(&current->mm->mmap_sem);
903         if (addr && !(shmflg & SHM_REMAP)) {
904                 err = -EINVAL;
905                 if (find_vma_intersection(current->mm, addr, addr + size))
906                         goto invalid;
907                 /*
908                  * If shm segment goes below stack, make sure there is some
909                  * space left for the stack to grow (at least 4 pages).
910                  */
911                 if (addr < current->mm->start_stack &&
912                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
913                         goto invalid;
914         }
915                 
916         user_addr = do_mmap (file, addr, size, prot, flags, 0);
917         *raddr = user_addr;
918         err = 0;
919         if (IS_ERR_VALUE(user_addr))
920                 err = (long)user_addr;
921 invalid:
922         up_write(&current->mm->mmap_sem);
923
924         fput(file);
925
926 out_nattch:
927         down_write(&shm_ids(ns).rw_mutex);
928         shp = shm_lock(ns, shmid);
929         BUG_ON(IS_ERR(shp));
930         shp->shm_nattch--;
931         if(shp->shm_nattch == 0 &&
932            shp->shm_perm.mode & SHM_DEST)
933                 shm_destroy(ns, shp);
934         else
935                 shm_unlock(shp);
936         up_write(&shm_ids(ns).rw_mutex);
937
938 out:
939         return err;
940
941 out_unlock:
942         shm_unlock(shp);
943         goto out;
944
945 out_free:
946         kfree(sfd);
947 out_put_dentry:
948         dput(path.dentry);
949         goto out_nattch;
950 }
951
952 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
953 {
954         unsigned long ret;
955         long err;
956
957         err = do_shmat(shmid, shmaddr, shmflg, &ret);
958         if (err)
959                 return err;
960         force_successful_syscall_return();
961         return (long)ret;
962 }
963
964 /*
965  * detach and kill segment if marked destroyed.
966  * The work is done in shm_close.
967  */
968 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
969 {
970         struct mm_struct *mm = current->mm;
971         struct vm_area_struct *vma;
972         unsigned long addr = (unsigned long)shmaddr;
973         int retval = -EINVAL;
974 #ifdef CONFIG_MMU
975         loff_t size = 0;
976         struct vm_area_struct *next;
977 #endif
978
979         if (addr & ~PAGE_MASK)
980                 return retval;
981
982         down_write(&mm->mmap_sem);
983
984         /*
985          * This function tries to be smart and unmap shm segments that
986          * were modified by partial mlock or munmap calls:
987          * - It first determines the size of the shm segment that should be
988          *   unmapped: It searches for a vma that is backed by shm and that
989          *   started at address shmaddr. It records it's size and then unmaps
990          *   it.
991          * - Then it unmaps all shm vmas that started at shmaddr and that
992          *   are within the initially determined size.
993          * Errors from do_munmap are ignored: the function only fails if
994          * it's called with invalid parameters or if it's called to unmap
995          * a part of a vma. Both calls in this function are for full vmas,
996          * the parameters are directly copied from the vma itself and always
997          * valid - therefore do_munmap cannot fail. (famous last words?)
998          */
999         /*
1000          * If it had been mremap()'d, the starting address would not
1001          * match the usual checks anyway. So assume all vma's are
1002          * above the starting address given.
1003          */
1004         vma = find_vma(mm, addr);
1005
1006 #ifdef CONFIG_MMU
1007         while (vma) {
1008                 next = vma->vm_next;
1009
1010                 /*
1011                  * Check if the starting address would match, i.e. it's
1012                  * a fragment created by mprotect() and/or munmap(), or it
1013                  * otherwise it starts at this address with no hassles.
1014                  */
1015                 if ((vma->vm_ops == &shm_vm_ops) &&
1016                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1017
1018
1019                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1020                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1021                         /*
1022                          * We discovered the size of the shm segment, so
1023                          * break out of here and fall through to the next
1024                          * loop that uses the size information to stop
1025                          * searching for matching vma's.
1026                          */
1027                         retval = 0;
1028                         vma = next;
1029                         break;
1030                 }
1031                 vma = next;
1032         }
1033
1034         /*
1035          * We need look no further than the maximum address a fragment
1036          * could possibly have landed at. Also cast things to loff_t to
1037          * prevent overflows and make comparisions vs. equal-width types.
1038          */
1039         size = PAGE_ALIGN(size);
1040         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1041                 next = vma->vm_next;
1042
1043                 /* finding a matching vma now does not alter retval */
1044                 if ((vma->vm_ops == &shm_vm_ops) &&
1045                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1046
1047                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1048                 vma = next;
1049         }
1050
1051 #else /* CONFIG_MMU */
1052         /* under NOMMU conditions, the exact address to be destroyed must be
1053          * given */
1054         retval = -EINVAL;
1055         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1056                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1057                 retval = 0;
1058         }
1059
1060 #endif
1061
1062         up_write(&mm->mmap_sem);
1063         return retval;
1064 }
1065
1066 #ifdef CONFIG_PROC_FS
1067 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1068 {
1069         struct shmid_kernel *shp = it;
1070
1071 #if BITS_PER_LONG <= 32
1072 #define SIZE_SPEC "%10lu"
1073 #else
1074 #define SIZE_SPEC "%21lu"
1075 #endif
1076
1077         return seq_printf(s,
1078                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1079                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1080                           shp->shm_perm.key,
1081                           shp->shm_perm.id,
1082                           shp->shm_perm.mode,
1083                           shp->shm_segsz,
1084                           shp->shm_cprid,
1085                           shp->shm_lprid,
1086                           shp->shm_nattch,
1087                           shp->shm_perm.uid,
1088                           shp->shm_perm.gid,
1089                           shp->shm_perm.cuid,
1090                           shp->shm_perm.cgid,
1091                           shp->shm_atim,
1092                           shp->shm_dtim,
1093                           shp->shm_ctim);
1094 }
1095 #endif