llseek: automatically add .llseek fop
[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 const 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         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
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 if (shp->mlock_user)
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, int datasync)
277 {
278         struct shm_file_data *sfd = shm_file_data(file);
279
280         if (!sfd->file->f_op->fsync)
281                 return -EINVAL;
282         return sfd->file->f_op->fsync(sfd->file, datasync);
283 }
284
285 static unsigned long shm_get_unmapped_area(struct file *file,
286         unsigned long addr, unsigned long len, unsigned long pgoff,
287         unsigned long flags)
288 {
289         struct shm_file_data *sfd = shm_file_data(file);
290         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
291                                                 pgoff, flags);
292 }
293
294 static const struct file_operations shm_file_operations = {
295         .mmap           = shm_mmap,
296         .fsync          = shm_fsync,
297         .release        = shm_release,
298 #ifndef CONFIG_MMU
299         .get_unmapped_area      = shm_get_unmapped_area,
300 #endif
301         .llseek         = noop_llseek,
302 };
303
304 static const struct file_operations shm_file_operations_huge = {
305         .mmap           = shm_mmap,
306         .fsync          = shm_fsync,
307         .release        = shm_release,
308         .get_unmapped_area      = shm_get_unmapped_area,
309         .llseek         = noop_llseek,
310 };
311
312 int is_file_shm_hugepages(struct file *file)
313 {
314         return file->f_op == &shm_file_operations_huge;
315 }
316
317 static const struct vm_operations_struct shm_vm_ops = {
318         .open   = shm_open,     /* callback for a new vm-area open */
319         .close  = shm_close,    /* callback for when the vm-area is released */
320         .fault  = shm_fault,
321 #if defined(CONFIG_NUMA)
322         .set_policy = shm_set_policy,
323         .get_policy = shm_get_policy,
324 #endif
325 };
326
327 /**
328  * newseg - Create a new shared memory segment
329  * @ns: namespace
330  * @params: ptr to the structure that contains key, size and shmflg
331  *
332  * Called with shm_ids.rw_mutex held as a writer.
333  */
334
335 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
336 {
337         key_t key = params->key;
338         int shmflg = params->flg;
339         size_t size = params->u.size;
340         int error;
341         struct shmid_kernel *shp;
342         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
343         struct file * file;
344         char name[13];
345         int id;
346         int acctflag = 0;
347
348         if (size < SHMMIN || size > ns->shm_ctlmax)
349                 return -EINVAL;
350
351         if (ns->shm_tot + numpages > ns->shm_ctlall)
352                 return -ENOSPC;
353
354         shp = ipc_rcu_alloc(sizeof(*shp));
355         if (!shp)
356                 return -ENOMEM;
357
358         shp->shm_perm.key = key;
359         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
360         shp->mlock_user = NULL;
361
362         shp->shm_perm.security = NULL;
363         error = security_shm_alloc(shp);
364         if (error) {
365                 ipc_rcu_putref(shp);
366                 return error;
367         }
368
369         sprintf (name, "SYSV%08x", key);
370         if (shmflg & SHM_HUGETLB) {
371                 /* hugetlb_file_setup applies strict accounting */
372                 if (shmflg & SHM_NORESERVE)
373                         acctflag = VM_NORESERVE;
374                 file = hugetlb_file_setup(name, size, acctflag,
375                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
376         } else {
377                 /*
378                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
379                  * if it's asked for.
380                  */
381                 if  ((shmflg & SHM_NORESERVE) &&
382                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
383                         acctflag = VM_NORESERVE;
384                 file = shmem_file_setup(name, size, acctflag);
385         }
386         error = PTR_ERR(file);
387         if (IS_ERR(file))
388                 goto no_file;
389
390         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
391         if (id < 0) {
392                 error = id;
393                 goto no_id;
394         }
395
396         shp->shm_cprid = task_tgid_vnr(current);
397         shp->shm_lprid = 0;
398         shp->shm_atim = shp->shm_dtim = 0;
399         shp->shm_ctim = get_seconds();
400         shp->shm_segsz = size;
401         shp->shm_nattch = 0;
402         shp->shm_file = file;
403         /*
404          * shmid gets reported as "inode#" in /proc/pid/maps.
405          * proc-ps tools use this. Changing this will break them.
406          */
407         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
408
409         ns->shm_tot += numpages;
410         error = shp->shm_perm.id;
411         shm_unlock(shp);
412         return error;
413
414 no_id:
415         if (is_file_hugepages(file) && shp->mlock_user)
416                 user_shm_unlock(size, shp->mlock_user);
417         fput(file);
418 no_file:
419         security_shm_free(shp);
420         ipc_rcu_putref(shp);
421         return error;
422 }
423
424 /*
425  * Called with shm_ids.rw_mutex and ipcp locked.
426  */
427 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
428 {
429         struct shmid_kernel *shp;
430
431         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
432         return security_shm_associate(shp, shmflg);
433 }
434
435 /*
436  * Called with shm_ids.rw_mutex and ipcp locked.
437  */
438 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
439                                 struct ipc_params *params)
440 {
441         struct shmid_kernel *shp;
442
443         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
444         if (shp->shm_segsz < params->u.size)
445                 return -EINVAL;
446
447         return 0;
448 }
449
450 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
451 {
452         struct ipc_namespace *ns;
453         struct ipc_ops shm_ops;
454         struct ipc_params shm_params;
455
456         ns = current->nsproxy->ipc_ns;
457
458         shm_ops.getnew = newseg;
459         shm_ops.associate = shm_security;
460         shm_ops.more_checks = shm_more_checks;
461
462         shm_params.key = key;
463         shm_params.flg = shmflg;
464         shm_params.u.size = size;
465
466         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
467 }
468
469 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
470 {
471         switch(version) {
472         case IPC_64:
473                 return copy_to_user(buf, in, sizeof(*in));
474         case IPC_OLD:
475             {
476                 struct shmid_ds out;
477
478                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
479                 out.shm_segsz   = in->shm_segsz;
480                 out.shm_atime   = in->shm_atime;
481                 out.shm_dtime   = in->shm_dtime;
482                 out.shm_ctime   = in->shm_ctime;
483                 out.shm_cpid    = in->shm_cpid;
484                 out.shm_lpid    = in->shm_lpid;
485                 out.shm_nattch  = in->shm_nattch;
486
487                 return copy_to_user(buf, &out, sizeof(out));
488             }
489         default:
490                 return -EINVAL;
491         }
492 }
493
494 static inline unsigned long
495 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
496 {
497         switch(version) {
498         case IPC_64:
499                 if (copy_from_user(out, buf, sizeof(*out)))
500                         return -EFAULT;
501                 return 0;
502         case IPC_OLD:
503             {
504                 struct shmid_ds tbuf_old;
505
506                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
507                         return -EFAULT;
508
509                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
510                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
511                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
512
513                 return 0;
514             }
515         default:
516                 return -EINVAL;
517         }
518 }
519
520 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
521 {
522         switch(version) {
523         case IPC_64:
524                 return copy_to_user(buf, in, sizeof(*in));
525         case IPC_OLD:
526             {
527                 struct shminfo out;
528
529                 if(in->shmmax > INT_MAX)
530                         out.shmmax = INT_MAX;
531                 else
532                         out.shmmax = (int)in->shmmax;
533
534                 out.shmmin      = in->shmmin;
535                 out.shmmni      = in->shmmni;
536                 out.shmseg      = in->shmseg;
537                 out.shmall      = in->shmall; 
538
539                 return copy_to_user(buf, &out, sizeof(out));
540             }
541         default:
542                 return -EINVAL;
543         }
544 }
545
546 /*
547  * Called with shm_ids.rw_mutex held as a reader
548  */
549 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
550                 unsigned long *swp)
551 {
552         int next_id;
553         int total, in_use;
554
555         *rss = 0;
556         *swp = 0;
557
558         in_use = shm_ids(ns).in_use;
559
560         for (total = 0, next_id = 0; total < in_use; next_id++) {
561                 struct kern_ipc_perm *ipc;
562                 struct shmid_kernel *shp;
563                 struct inode *inode;
564
565                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
566                 if (ipc == NULL)
567                         continue;
568                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
569
570                 inode = shp->shm_file->f_path.dentry->d_inode;
571
572                 if (is_file_hugepages(shp->shm_file)) {
573                         struct address_space *mapping = inode->i_mapping;
574                         struct hstate *h = hstate_file(shp->shm_file);
575                         *rss += pages_per_huge_page(h) * mapping->nrpages;
576                 } else {
577 #ifdef CONFIG_SHMEM
578                         struct shmem_inode_info *info = SHMEM_I(inode);
579                         spin_lock(&info->lock);
580                         *rss += inode->i_mapping->nrpages;
581                         *swp += info->swapped;
582                         spin_unlock(&info->lock);
583 #else
584                         *rss += inode->i_mapping->nrpages;
585 #endif
586                 }
587
588                 total++;
589         }
590 }
591
592 /*
593  * This function handles some shmctl commands which require the rw_mutex
594  * to be held in write mode.
595  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
596  */
597 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
598                        struct shmid_ds __user *buf, int version)
599 {
600         struct kern_ipc_perm *ipcp;
601         struct shmid64_ds shmid64;
602         struct shmid_kernel *shp;
603         int err;
604
605         if (cmd == IPC_SET) {
606                 if (copy_shmid_from_user(&shmid64, buf, version))
607                         return -EFAULT;
608         }
609
610         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
611         if (IS_ERR(ipcp))
612                 return PTR_ERR(ipcp);
613
614         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
615
616         err = security_shm_shmctl(shp, cmd);
617         if (err)
618                 goto out_unlock;
619         switch (cmd) {
620         case IPC_RMID:
621                 do_shm_rmid(ns, ipcp);
622                 goto out_up;
623         case IPC_SET:
624                 ipc_update_perm(&shmid64.shm_perm, ipcp);
625                 shp->shm_ctim = get_seconds();
626                 break;
627         default:
628                 err = -EINVAL;
629         }
630 out_unlock:
631         shm_unlock(shp);
632 out_up:
633         up_write(&shm_ids(ns).rw_mutex);
634         return err;
635 }
636
637 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
638 {
639         struct shmid_kernel *shp;
640         int err, version;
641         struct ipc_namespace *ns;
642
643         if (cmd < 0 || shmid < 0) {
644                 err = -EINVAL;
645                 goto out;
646         }
647
648         version = ipc_parse_version(&cmd);
649         ns = current->nsproxy->ipc_ns;
650
651         switch (cmd) { /* replace with proc interface ? */
652         case IPC_INFO:
653         {
654                 struct shminfo64 shminfo;
655
656                 err = security_shm_shmctl(NULL, cmd);
657                 if (err)
658                         return err;
659
660                 memset(&shminfo, 0, sizeof(shminfo));
661                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
662                 shminfo.shmmax = ns->shm_ctlmax;
663                 shminfo.shmall = ns->shm_ctlall;
664
665                 shminfo.shmmin = SHMMIN;
666                 if(copy_shminfo_to_user (buf, &shminfo, version))
667                         return -EFAULT;
668
669                 down_read(&shm_ids(ns).rw_mutex);
670                 err = ipc_get_maxid(&shm_ids(ns));
671                 up_read(&shm_ids(ns).rw_mutex);
672
673                 if(err<0)
674                         err = 0;
675                 goto out;
676         }
677         case SHM_INFO:
678         {
679                 struct shm_info shm_info;
680
681                 err = security_shm_shmctl(NULL, cmd);
682                 if (err)
683                         return err;
684
685                 memset(&shm_info, 0, sizeof(shm_info));
686                 down_read(&shm_ids(ns).rw_mutex);
687                 shm_info.used_ids = shm_ids(ns).in_use;
688                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
689                 shm_info.shm_tot = ns->shm_tot;
690                 shm_info.swap_attempts = 0;
691                 shm_info.swap_successes = 0;
692                 err = ipc_get_maxid(&shm_ids(ns));
693                 up_read(&shm_ids(ns).rw_mutex);
694                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
695                         err = -EFAULT;
696                         goto out;
697                 }
698
699                 err = err < 0 ? 0 : err;
700                 goto out;
701         }
702         case SHM_STAT:
703         case IPC_STAT:
704         {
705                 struct shmid64_ds tbuf;
706                 int result;
707
708                 if (cmd == SHM_STAT) {
709                         shp = shm_lock(ns, shmid);
710                         if (IS_ERR(shp)) {
711                                 err = PTR_ERR(shp);
712                                 goto out;
713                         }
714                         result = shp->shm_perm.id;
715                 } else {
716                         shp = shm_lock_check(ns, shmid);
717                         if (IS_ERR(shp)) {
718                                 err = PTR_ERR(shp);
719                                 goto out;
720                         }
721                         result = 0;
722                 }
723                 err = -EACCES;
724                 if (ipcperms (&shp->shm_perm, S_IRUGO))
725                         goto out_unlock;
726                 err = security_shm_shmctl(shp, cmd);
727                 if (err)
728                         goto out_unlock;
729                 memset(&tbuf, 0, sizeof(tbuf));
730                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
731                 tbuf.shm_segsz  = shp->shm_segsz;
732                 tbuf.shm_atime  = shp->shm_atim;
733                 tbuf.shm_dtime  = shp->shm_dtim;
734                 tbuf.shm_ctime  = shp->shm_ctim;
735                 tbuf.shm_cpid   = shp->shm_cprid;
736                 tbuf.shm_lpid   = shp->shm_lprid;
737                 tbuf.shm_nattch = shp->shm_nattch;
738                 shm_unlock(shp);
739                 if(copy_shmid_to_user (buf, &tbuf, version))
740                         err = -EFAULT;
741                 else
742                         err = result;
743                 goto out;
744         }
745         case SHM_LOCK:
746         case SHM_UNLOCK:
747         {
748                 struct file *uninitialized_var(shm_file);
749
750                 lru_add_drain_all();  /* drain pagevecs to lru lists */
751
752                 shp = shm_lock_check(ns, shmid);
753                 if (IS_ERR(shp)) {
754                         err = PTR_ERR(shp);
755                         goto out;
756                 }
757
758                 audit_ipc_obj(&(shp->shm_perm));
759
760                 if (!capable(CAP_IPC_LOCK)) {
761                         uid_t euid = current_euid();
762                         err = -EPERM;
763                         if (euid != shp->shm_perm.uid &&
764                             euid != shp->shm_perm.cuid)
765                                 goto out_unlock;
766                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
767                                 goto out_unlock;
768                 }
769
770                 err = security_shm_shmctl(shp, cmd);
771                 if (err)
772                         goto out_unlock;
773                 
774                 if(cmd==SHM_LOCK) {
775                         struct user_struct *user = current_user();
776                         if (!is_file_hugepages(shp->shm_file)) {
777                                 err = shmem_lock(shp->shm_file, 1, user);
778                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
779                                         shp->shm_perm.mode |= SHM_LOCKED;
780                                         shp->mlock_user = user;
781                                 }
782                         }
783                 } else if (!is_file_hugepages(shp->shm_file)) {
784                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
785                         shp->shm_perm.mode &= ~SHM_LOCKED;
786                         shp->mlock_user = NULL;
787                 }
788                 shm_unlock(shp);
789                 goto out;
790         }
791         case IPC_RMID:
792         case IPC_SET:
793                 err = shmctl_down(ns, shmid, cmd, buf, version);
794                 return err;
795         default:
796                 return -EINVAL;
797         }
798
799 out_unlock:
800         shm_unlock(shp);
801 out:
802         return err;
803 }
804
805 /*
806  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
807  *
808  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
809  * "raddr" thing points to kernel space, and there has to be a wrapper around
810  * this.
811  */
812 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
813 {
814         struct shmid_kernel *shp;
815         unsigned long addr;
816         unsigned long size;
817         struct file * file;
818         int    err;
819         unsigned long flags;
820         unsigned long prot;
821         int acc_mode;
822         unsigned long user_addr;
823         struct ipc_namespace *ns;
824         struct shm_file_data *sfd;
825         struct path path;
826         fmode_t f_mode;
827
828         err = -EINVAL;
829         if (shmid < 0)
830                 goto out;
831         else if ((addr = (ulong)shmaddr)) {
832                 if (addr & (SHMLBA-1)) {
833                         if (shmflg & SHM_RND)
834                                 addr &= ~(SHMLBA-1);       /* round down */
835                         else
836 #ifndef __ARCH_FORCE_SHMLBA
837                                 if (addr & ~PAGE_MASK)
838 #endif
839                                         goto out;
840                 }
841                 flags = MAP_SHARED | MAP_FIXED;
842         } else {
843                 if ((shmflg & SHM_REMAP))
844                         goto out;
845
846                 flags = MAP_SHARED;
847         }
848
849         if (shmflg & SHM_RDONLY) {
850                 prot = PROT_READ;
851                 acc_mode = S_IRUGO;
852                 f_mode = FMODE_READ;
853         } else {
854                 prot = PROT_READ | PROT_WRITE;
855                 acc_mode = S_IRUGO | S_IWUGO;
856                 f_mode = FMODE_READ | FMODE_WRITE;
857         }
858         if (shmflg & SHM_EXEC) {
859                 prot |= PROT_EXEC;
860                 acc_mode |= S_IXUGO;
861         }
862
863         /*
864          * We cannot rely on the fs check since SYSV IPC does have an
865          * additional creator id...
866          */
867         ns = current->nsproxy->ipc_ns;
868         shp = shm_lock_check(ns, shmid);
869         if (IS_ERR(shp)) {
870                 err = PTR_ERR(shp);
871                 goto out;
872         }
873
874         err = -EACCES;
875         if (ipcperms(&shp->shm_perm, acc_mode))
876                 goto out_unlock;
877
878         err = security_shm_shmat(shp, shmaddr, shmflg);
879         if (err)
880                 goto out_unlock;
881
882         path = shp->shm_file->f_path;
883         path_get(&path);
884         shp->shm_nattch++;
885         size = i_size_read(path.dentry->d_inode);
886         shm_unlock(shp);
887
888         err = -ENOMEM;
889         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
890         if (!sfd)
891                 goto out_put_dentry;
892
893         file = alloc_file(&path, f_mode,
894                           is_file_hugepages(shp->shm_file) ?
895                                 &shm_file_operations_huge :
896                                 &shm_file_operations);
897         if (!file)
898                 goto out_free;
899
900         file->private_data = sfd;
901         file->f_mapping = shp->shm_file->f_mapping;
902         sfd->id = shp->shm_perm.id;
903         sfd->ns = get_ipc_ns(ns);
904         sfd->file = shp->shm_file;
905         sfd->vm_ops = NULL;
906
907         down_write(&current->mm->mmap_sem);
908         if (addr && !(shmflg & SHM_REMAP)) {
909                 err = -EINVAL;
910                 if (find_vma_intersection(current->mm, addr, addr + size))
911                         goto invalid;
912                 /*
913                  * If shm segment goes below stack, make sure there is some
914                  * space left for the stack to grow (at least 4 pages).
915                  */
916                 if (addr < current->mm->start_stack &&
917                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
918                         goto invalid;
919         }
920                 
921         user_addr = do_mmap (file, addr, size, prot, flags, 0);
922         *raddr = user_addr;
923         err = 0;
924         if (IS_ERR_VALUE(user_addr))
925                 err = (long)user_addr;
926 invalid:
927         up_write(&current->mm->mmap_sem);
928
929         fput(file);
930
931 out_nattch:
932         down_write(&shm_ids(ns).rw_mutex);
933         shp = shm_lock(ns, shmid);
934         BUG_ON(IS_ERR(shp));
935         shp->shm_nattch--;
936         if(shp->shm_nattch == 0 &&
937            shp->shm_perm.mode & SHM_DEST)
938                 shm_destroy(ns, shp);
939         else
940                 shm_unlock(shp);
941         up_write(&shm_ids(ns).rw_mutex);
942
943 out:
944         return err;
945
946 out_unlock:
947         shm_unlock(shp);
948         goto out;
949
950 out_free:
951         kfree(sfd);
952 out_put_dentry:
953         path_put(&path);
954         goto out_nattch;
955 }
956
957 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
958 {
959         unsigned long ret;
960         long err;
961
962         err = do_shmat(shmid, shmaddr, shmflg, &ret);
963         if (err)
964                 return err;
965         force_successful_syscall_return();
966         return (long)ret;
967 }
968
969 /*
970  * detach and kill segment if marked destroyed.
971  * The work is done in shm_close.
972  */
973 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
974 {
975         struct mm_struct *mm = current->mm;
976         struct vm_area_struct *vma;
977         unsigned long addr = (unsigned long)shmaddr;
978         int retval = -EINVAL;
979 #ifdef CONFIG_MMU
980         loff_t size = 0;
981         struct vm_area_struct *next;
982 #endif
983
984         if (addr & ~PAGE_MASK)
985                 return retval;
986
987         down_write(&mm->mmap_sem);
988
989         /*
990          * This function tries to be smart and unmap shm segments that
991          * were modified by partial mlock or munmap calls:
992          * - It first determines the size of the shm segment that should be
993          *   unmapped: It searches for a vma that is backed by shm and that
994          *   started at address shmaddr. It records it's size and then unmaps
995          *   it.
996          * - Then it unmaps all shm vmas that started at shmaddr and that
997          *   are within the initially determined size.
998          * Errors from do_munmap are ignored: the function only fails if
999          * it's called with invalid parameters or if it's called to unmap
1000          * a part of a vma. Both calls in this function are for full vmas,
1001          * the parameters are directly copied from the vma itself and always
1002          * valid - therefore do_munmap cannot fail. (famous last words?)
1003          */
1004         /*
1005          * If it had been mremap()'d, the starting address would not
1006          * match the usual checks anyway. So assume all vma's are
1007          * above the starting address given.
1008          */
1009         vma = find_vma(mm, addr);
1010
1011 #ifdef CONFIG_MMU
1012         while (vma) {
1013                 next = vma->vm_next;
1014
1015                 /*
1016                  * Check if the starting address would match, i.e. it's
1017                  * a fragment created by mprotect() and/or munmap(), or it
1018                  * otherwise it starts at this address with no hassles.
1019                  */
1020                 if ((vma->vm_ops == &shm_vm_ops) &&
1021                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1022
1023
1024                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1025                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1026                         /*
1027                          * We discovered the size of the shm segment, so
1028                          * break out of here and fall through to the next
1029                          * loop that uses the size information to stop
1030                          * searching for matching vma's.
1031                          */
1032                         retval = 0;
1033                         vma = next;
1034                         break;
1035                 }
1036                 vma = next;
1037         }
1038
1039         /*
1040          * We need look no further than the maximum address a fragment
1041          * could possibly have landed at. Also cast things to loff_t to
1042          * prevent overflows and make comparisions vs. equal-width types.
1043          */
1044         size = PAGE_ALIGN(size);
1045         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1046                 next = vma->vm_next;
1047
1048                 /* finding a matching vma now does not alter retval */
1049                 if ((vma->vm_ops == &shm_vm_ops) &&
1050                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1051
1052                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1053                 vma = next;
1054         }
1055
1056 #else /* CONFIG_MMU */
1057         /* under NOMMU conditions, the exact address to be destroyed must be
1058          * given */
1059         retval = -EINVAL;
1060         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1061                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1062                 retval = 0;
1063         }
1064
1065 #endif
1066
1067         up_write(&mm->mmap_sem);
1068         return retval;
1069 }
1070
1071 #ifdef CONFIG_PROC_FS
1072 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1073 {
1074         struct shmid_kernel *shp = it;
1075
1076 #if BITS_PER_LONG <= 32
1077 #define SIZE_SPEC "%10lu"
1078 #else
1079 #define SIZE_SPEC "%21lu"
1080 #endif
1081
1082         return seq_printf(s,
1083                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1084                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1085                           shp->shm_perm.key,
1086                           shp->shm_perm.id,
1087                           shp->shm_perm.mode,
1088                           shp->shm_segsz,
1089                           shp->shm_cprid,
1090                           shp->shm_lprid,
1091                           shp->shm_nattch,
1092                           shp->shm_perm.uid,
1093                           shp->shm_perm.gid,
1094                           shp->shm_perm.cuid,
1095                           shp->shm_perm.cgid,
1096                           shp->shm_atim,
1097                           shp->shm_dtim,
1098                           shp->shm_ctim);
1099 }
1100 #endif