Merge commit 'v2.6.29-rc1' into timers/urgent
[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
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
344         if (size < SHMMIN || size > ns->shm_ctlmax)
345                 return -EINVAL;
346
347         if (ns->shm_tot + numpages > ns->shm_ctlall)
348                 return -ENOSPC;
349
350         shp = ipc_rcu_alloc(sizeof(*shp));
351         if (!shp)
352                 return -ENOMEM;
353
354         shp->shm_perm.key = key;
355         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
356         shp->mlock_user = NULL;
357
358         shp->shm_perm.security = NULL;
359         error = security_shm_alloc(shp);
360         if (error) {
361                 ipc_rcu_putref(shp);
362                 return error;
363         }
364
365         sprintf (name, "SYSV%08x", key);
366         if (shmflg & SHM_HUGETLB) {
367                 /* hugetlb_file_setup takes care of mlock user accounting */
368                 file = hugetlb_file_setup(name, size);
369                 shp->mlock_user = current_user();
370         } else {
371                 int acctflag = VM_ACCOUNT;
372                 /*
373                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
374                  * if it's asked for.
375                  */
376                 if  ((shmflg & SHM_NORESERVE) &&
377                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
378                         acctflag = 0;
379                 file = shmem_file_setup(name, size, acctflag);
380         }
381         error = PTR_ERR(file);
382         if (IS_ERR(file))
383                 goto no_file;
384
385         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
386         if (id < 0) {
387                 error = id;
388                 goto no_id;
389         }
390
391         shp->shm_cprid = task_tgid_vnr(current);
392         shp->shm_lprid = 0;
393         shp->shm_atim = shp->shm_dtim = 0;
394         shp->shm_ctim = get_seconds();
395         shp->shm_segsz = size;
396         shp->shm_nattch = 0;
397         shp->shm_file = file;
398         /*
399          * shmid gets reported as "inode#" in /proc/pid/maps.
400          * proc-ps tools use this. Changing this will break them.
401          */
402         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
403
404         ns->shm_tot += numpages;
405         error = shp->shm_perm.id;
406         shm_unlock(shp);
407         return error;
408
409 no_id:
410         fput(file);
411 no_file:
412         security_shm_free(shp);
413         ipc_rcu_putref(shp);
414         return error;
415 }
416
417 /*
418  * Called with shm_ids.rw_mutex and ipcp locked.
419  */
420 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
421 {
422         struct shmid_kernel *shp;
423
424         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
425         return security_shm_associate(shp, shmflg);
426 }
427
428 /*
429  * Called with shm_ids.rw_mutex and ipcp locked.
430  */
431 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
432                                 struct ipc_params *params)
433 {
434         struct shmid_kernel *shp;
435
436         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
437         if (shp->shm_segsz < params->u.size)
438                 return -EINVAL;
439
440         return 0;
441 }
442
443 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
444 {
445         struct ipc_namespace *ns;
446         struct ipc_ops shm_ops;
447         struct ipc_params shm_params;
448
449         ns = current->nsproxy->ipc_ns;
450
451         shm_ops.getnew = newseg;
452         shm_ops.associate = shm_security;
453         shm_ops.more_checks = shm_more_checks;
454
455         shm_params.key = key;
456         shm_params.flg = shmflg;
457         shm_params.u.size = size;
458
459         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
460 }
461
462 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
463 {
464         switch(version) {
465         case IPC_64:
466                 return copy_to_user(buf, in, sizeof(*in));
467         case IPC_OLD:
468             {
469                 struct shmid_ds out;
470
471                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
472                 out.shm_segsz   = in->shm_segsz;
473                 out.shm_atime   = in->shm_atime;
474                 out.shm_dtime   = in->shm_dtime;
475                 out.shm_ctime   = in->shm_ctime;
476                 out.shm_cpid    = in->shm_cpid;
477                 out.shm_lpid    = in->shm_lpid;
478                 out.shm_nattch  = in->shm_nattch;
479
480                 return copy_to_user(buf, &out, sizeof(out));
481             }
482         default:
483                 return -EINVAL;
484         }
485 }
486
487 static inline unsigned long
488 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
489 {
490         switch(version) {
491         case IPC_64:
492                 if (copy_from_user(out, buf, sizeof(*out)))
493                         return -EFAULT;
494                 return 0;
495         case IPC_OLD:
496             {
497                 struct shmid_ds tbuf_old;
498
499                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
500                         return -EFAULT;
501
502                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
503                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
504                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
505
506                 return 0;
507             }
508         default:
509                 return -EINVAL;
510         }
511 }
512
513 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
514 {
515         switch(version) {
516         case IPC_64:
517                 return copy_to_user(buf, in, sizeof(*in));
518         case IPC_OLD:
519             {
520                 struct shminfo out;
521
522                 if(in->shmmax > INT_MAX)
523                         out.shmmax = INT_MAX;
524                 else
525                         out.shmmax = (int)in->shmmax;
526
527                 out.shmmin      = in->shmmin;
528                 out.shmmni      = in->shmmni;
529                 out.shmseg      = in->shmseg;
530                 out.shmall      = in->shmall; 
531
532                 return copy_to_user(buf, &out, sizeof(out));
533             }
534         default:
535                 return -EINVAL;
536         }
537 }
538
539 /*
540  * Called with shm_ids.rw_mutex held as a reader
541  */
542 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
543                 unsigned long *swp)
544 {
545         int next_id;
546         int total, in_use;
547
548         *rss = 0;
549         *swp = 0;
550
551         in_use = shm_ids(ns).in_use;
552
553         for (total = 0, next_id = 0; total < in_use; next_id++) {
554                 struct shmid_kernel *shp;
555                 struct inode *inode;
556
557                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
558                 if (shp == NULL)
559                         continue;
560
561                 inode = shp->shm_file->f_path.dentry->d_inode;
562
563                 if (is_file_hugepages(shp->shm_file)) {
564                         struct address_space *mapping = inode->i_mapping;
565                         struct hstate *h = hstate_file(shp->shm_file);
566                         *rss += pages_per_huge_page(h) * mapping->nrpages;
567                 } else {
568                         struct shmem_inode_info *info = SHMEM_I(inode);
569                         spin_lock(&info->lock);
570                         *rss += inode->i_mapping->nrpages;
571                         *swp += info->swapped;
572                         spin_unlock(&info->lock);
573                 }
574
575                 total++;
576         }
577 }
578
579 /*
580  * This function handles some shmctl commands which require the rw_mutex
581  * to be held in write mode.
582  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
583  */
584 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
585                        struct shmid_ds __user *buf, int version)
586 {
587         struct kern_ipc_perm *ipcp;
588         struct shmid64_ds shmid64;
589         struct shmid_kernel *shp;
590         int err;
591
592         if (cmd == IPC_SET) {
593                 if (copy_shmid_from_user(&shmid64, buf, version))
594                         return -EFAULT;
595         }
596
597         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
598         if (IS_ERR(ipcp))
599                 return PTR_ERR(ipcp);
600
601         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
602
603         err = security_shm_shmctl(shp, cmd);
604         if (err)
605                 goto out_unlock;
606         switch (cmd) {
607         case IPC_RMID:
608                 do_shm_rmid(ns, ipcp);
609                 goto out_up;
610         case IPC_SET:
611                 ipc_update_perm(&shmid64.shm_perm, ipcp);
612                 shp->shm_ctim = get_seconds();
613                 break;
614         default:
615                 err = -EINVAL;
616         }
617 out_unlock:
618         shm_unlock(shp);
619 out_up:
620         up_write(&shm_ids(ns).rw_mutex);
621         return err;
622 }
623
624 asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
625 {
626         struct shmid_kernel *shp;
627         int err, version;
628         struct ipc_namespace *ns;
629
630         if (cmd < 0 || shmid < 0) {
631                 err = -EINVAL;
632                 goto out;
633         }
634
635         version = ipc_parse_version(&cmd);
636         ns = current->nsproxy->ipc_ns;
637
638         switch (cmd) { /* replace with proc interface ? */
639         case IPC_INFO:
640         {
641                 struct shminfo64 shminfo;
642
643                 err = security_shm_shmctl(NULL, cmd);
644                 if (err)
645                         return err;
646
647                 memset(&shminfo, 0, sizeof(shminfo));
648                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
649                 shminfo.shmmax = ns->shm_ctlmax;
650                 shminfo.shmall = ns->shm_ctlall;
651
652                 shminfo.shmmin = SHMMIN;
653                 if(copy_shminfo_to_user (buf, &shminfo, version))
654                         return -EFAULT;
655
656                 down_read(&shm_ids(ns).rw_mutex);
657                 err = ipc_get_maxid(&shm_ids(ns));
658                 up_read(&shm_ids(ns).rw_mutex);
659
660                 if(err<0)
661                         err = 0;
662                 goto out;
663         }
664         case SHM_INFO:
665         {
666                 struct shm_info shm_info;
667
668                 err = security_shm_shmctl(NULL, cmd);
669                 if (err)
670                         return err;
671
672                 memset(&shm_info, 0, sizeof(shm_info));
673                 down_read(&shm_ids(ns).rw_mutex);
674                 shm_info.used_ids = shm_ids(ns).in_use;
675                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
676                 shm_info.shm_tot = ns->shm_tot;
677                 shm_info.swap_attempts = 0;
678                 shm_info.swap_successes = 0;
679                 err = ipc_get_maxid(&shm_ids(ns));
680                 up_read(&shm_ids(ns).rw_mutex);
681                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
682                         err = -EFAULT;
683                         goto out;
684                 }
685
686                 err = err < 0 ? 0 : err;
687                 goto out;
688         }
689         case SHM_STAT:
690         case IPC_STAT:
691         {
692                 struct shmid64_ds tbuf;
693                 int result;
694
695                 if (cmd == SHM_STAT) {
696                         shp = shm_lock(ns, shmid);
697                         if (IS_ERR(shp)) {
698                                 err = PTR_ERR(shp);
699                                 goto out;
700                         }
701                         result = shp->shm_perm.id;
702                 } else {
703                         shp = shm_lock_check(ns, shmid);
704                         if (IS_ERR(shp)) {
705                                 err = PTR_ERR(shp);
706                                 goto out;
707                         }
708                         result = 0;
709                 }
710                 err = -EACCES;
711                 if (ipcperms (&shp->shm_perm, S_IRUGO))
712                         goto out_unlock;
713                 err = security_shm_shmctl(shp, cmd);
714                 if (err)
715                         goto out_unlock;
716                 memset(&tbuf, 0, sizeof(tbuf));
717                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
718                 tbuf.shm_segsz  = shp->shm_segsz;
719                 tbuf.shm_atime  = shp->shm_atim;
720                 tbuf.shm_dtime  = shp->shm_dtim;
721                 tbuf.shm_ctime  = shp->shm_ctim;
722                 tbuf.shm_cpid   = shp->shm_cprid;
723                 tbuf.shm_lpid   = shp->shm_lprid;
724                 tbuf.shm_nattch = shp->shm_nattch;
725                 shm_unlock(shp);
726                 if(copy_shmid_to_user (buf, &tbuf, version))
727                         err = -EFAULT;
728                 else
729                         err = result;
730                 goto out;
731         }
732         case SHM_LOCK:
733         case SHM_UNLOCK:
734         {
735                 struct file *uninitialized_var(shm_file);
736
737                 lru_add_drain_all();  /* drain pagevecs to lru lists */
738
739                 shp = shm_lock_check(ns, shmid);
740                 if (IS_ERR(shp)) {
741                         err = PTR_ERR(shp);
742                         goto out;
743                 }
744
745                 audit_ipc_obj(&(shp->shm_perm));
746
747                 if (!capable(CAP_IPC_LOCK)) {
748                         uid_t euid = current_euid();
749                         err = -EPERM;
750                         if (euid != shp->shm_perm.uid &&
751                             euid != shp->shm_perm.cuid)
752                                 goto out_unlock;
753                         if (cmd == SHM_LOCK &&
754                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
755                                 goto out_unlock;
756                 }
757
758                 err = security_shm_shmctl(shp, cmd);
759                 if (err)
760                         goto out_unlock;
761                 
762                 if(cmd==SHM_LOCK) {
763                         struct user_struct *user = current_user();
764                         if (!is_file_hugepages(shp->shm_file)) {
765                                 err = shmem_lock(shp->shm_file, 1, user);
766                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
767                                         shp->shm_perm.mode |= SHM_LOCKED;
768                                         shp->mlock_user = user;
769                                 }
770                         }
771                 } else if (!is_file_hugepages(shp->shm_file)) {
772                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
773                         shp->shm_perm.mode &= ~SHM_LOCKED;
774                         shp->mlock_user = NULL;
775                 }
776                 shm_unlock(shp);
777                 goto out;
778         }
779         case IPC_RMID:
780         case IPC_SET:
781                 err = shmctl_down(ns, shmid, cmd, buf, version);
782                 return err;
783         default:
784                 return -EINVAL;
785         }
786
787 out_unlock:
788         shm_unlock(shp);
789 out:
790         return err;
791 }
792
793 /*
794  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
795  *
796  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
797  * "raddr" thing points to kernel space, and there has to be a wrapper around
798  * this.
799  */
800 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
801 {
802         struct shmid_kernel *shp;
803         unsigned long addr;
804         unsigned long size;
805         struct file * file;
806         int    err;
807         unsigned long flags;
808         unsigned long prot;
809         int acc_mode;
810         unsigned long user_addr;
811         struct ipc_namespace *ns;
812         struct shm_file_data *sfd;
813         struct path path;
814         fmode_t f_mode;
815
816         err = -EINVAL;
817         if (shmid < 0)
818                 goto out;
819         else if ((addr = (ulong)shmaddr)) {
820                 if (addr & (SHMLBA-1)) {
821                         if (shmflg & SHM_RND)
822                                 addr &= ~(SHMLBA-1);       /* round down */
823                         else
824 #ifndef __ARCH_FORCE_SHMLBA
825                                 if (addr & ~PAGE_MASK)
826 #endif
827                                         goto out;
828                 }
829                 flags = MAP_SHARED | MAP_FIXED;
830         } else {
831                 if ((shmflg & SHM_REMAP))
832                         goto out;
833
834                 flags = MAP_SHARED;
835         }
836
837         if (shmflg & SHM_RDONLY) {
838                 prot = PROT_READ;
839                 acc_mode = S_IRUGO;
840                 f_mode = FMODE_READ;
841         } else {
842                 prot = PROT_READ | PROT_WRITE;
843                 acc_mode = S_IRUGO | S_IWUGO;
844                 f_mode = FMODE_READ | FMODE_WRITE;
845         }
846         if (shmflg & SHM_EXEC) {
847                 prot |= PROT_EXEC;
848                 acc_mode |= S_IXUGO;
849         }
850
851         /*
852          * We cannot rely on the fs check since SYSV IPC does have an
853          * additional creator id...
854          */
855         ns = current->nsproxy->ipc_ns;
856         shp = shm_lock_check(ns, shmid);
857         if (IS_ERR(shp)) {
858                 err = PTR_ERR(shp);
859                 goto out;
860         }
861
862         err = -EACCES;
863         if (ipcperms(&shp->shm_perm, acc_mode))
864                 goto out_unlock;
865
866         err = security_shm_shmat(shp, shmaddr, shmflg);
867         if (err)
868                 goto out_unlock;
869
870         path.dentry = dget(shp->shm_file->f_path.dentry);
871         path.mnt    = shp->shm_file->f_path.mnt;
872         shp->shm_nattch++;
873         size = i_size_read(path.dentry->d_inode);
874         shm_unlock(shp);
875
876         err = -ENOMEM;
877         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
878         if (!sfd)
879                 goto out_put_dentry;
880
881         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
882         if (!file)
883                 goto out_free;
884
885         file->private_data = sfd;
886         file->f_mapping = shp->shm_file->f_mapping;
887         sfd->id = shp->shm_perm.id;
888         sfd->ns = get_ipc_ns(ns);
889         sfd->file = shp->shm_file;
890         sfd->vm_ops = NULL;
891
892         down_write(&current->mm->mmap_sem);
893         if (addr && !(shmflg & SHM_REMAP)) {
894                 err = -EINVAL;
895                 if (find_vma_intersection(current->mm, addr, addr + size))
896                         goto invalid;
897                 /*
898                  * If shm segment goes below stack, make sure there is some
899                  * space left for the stack to grow (at least 4 pages).
900                  */
901                 if (addr < current->mm->start_stack &&
902                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
903                         goto invalid;
904         }
905                 
906         user_addr = do_mmap (file, addr, size, prot, flags, 0);
907         *raddr = user_addr;
908         err = 0;
909         if (IS_ERR_VALUE(user_addr))
910                 err = (long)user_addr;
911 invalid:
912         up_write(&current->mm->mmap_sem);
913
914         fput(file);
915
916 out_nattch:
917         down_write(&shm_ids(ns).rw_mutex);
918         shp = shm_lock(ns, shmid);
919         BUG_ON(IS_ERR(shp));
920         shp->shm_nattch--;
921         if(shp->shm_nattch == 0 &&
922            shp->shm_perm.mode & SHM_DEST)
923                 shm_destroy(ns, shp);
924         else
925                 shm_unlock(shp);
926         up_write(&shm_ids(ns).rw_mutex);
927
928 out:
929         return err;
930
931 out_unlock:
932         shm_unlock(shp);
933         goto out;
934
935 out_free:
936         kfree(sfd);
937 out_put_dentry:
938         dput(path.dentry);
939         goto out_nattch;
940 }
941
942 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
943 {
944         unsigned long ret;
945         long err;
946
947         err = do_shmat(shmid, shmaddr, shmflg, &ret);
948         if (err)
949                 return err;
950         force_successful_syscall_return();
951         return (long)ret;
952 }
953
954 /*
955  * detach and kill segment if marked destroyed.
956  * The work is done in shm_close.
957  */
958 asmlinkage long sys_shmdt(char __user *shmaddr)
959 {
960         struct mm_struct *mm = current->mm;
961         struct vm_area_struct *vma, *next;
962         unsigned long addr = (unsigned long)shmaddr;
963         loff_t size = 0;
964         int retval = -EINVAL;
965
966         if (addr & ~PAGE_MASK)
967                 return retval;
968
969         down_write(&mm->mmap_sem);
970
971         /*
972          * This function tries to be smart and unmap shm segments that
973          * were modified by partial mlock or munmap calls:
974          * - It first determines the size of the shm segment that should be
975          *   unmapped: It searches for a vma that is backed by shm and that
976          *   started at address shmaddr. It records it's size and then unmaps
977          *   it.
978          * - Then it unmaps all shm vmas that started at shmaddr and that
979          *   are within the initially determined size.
980          * Errors from do_munmap are ignored: the function only fails if
981          * it's called with invalid parameters or if it's called to unmap
982          * a part of a vma. Both calls in this function are for full vmas,
983          * the parameters are directly copied from the vma itself and always
984          * valid - therefore do_munmap cannot fail. (famous last words?)
985          */
986         /*
987          * If it had been mremap()'d, the starting address would not
988          * match the usual checks anyway. So assume all vma's are
989          * above the starting address given.
990          */
991         vma = find_vma(mm, addr);
992
993 #ifdef CONFIG_MMU
994         while (vma) {
995                 next = vma->vm_next;
996
997                 /*
998                  * Check if the starting address would match, i.e. it's
999                  * a fragment created by mprotect() and/or munmap(), or it
1000                  * otherwise it starts at this address with no hassles.
1001                  */
1002                 if ((vma->vm_ops == &shm_vm_ops) &&
1003                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1004
1005
1006                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1007                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1008                         /*
1009                          * We discovered the size of the shm segment, so
1010                          * break out of here and fall through to the next
1011                          * loop that uses the size information to stop
1012                          * searching for matching vma's.
1013                          */
1014                         retval = 0;
1015                         vma = next;
1016                         break;
1017                 }
1018                 vma = next;
1019         }
1020
1021         /*
1022          * We need look no further than the maximum address a fragment
1023          * could possibly have landed at. Also cast things to loff_t to
1024          * prevent overflows and make comparisions vs. equal-width types.
1025          */
1026         size = PAGE_ALIGN(size);
1027         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1028                 next = vma->vm_next;
1029
1030                 /* finding a matching vma now does not alter retval */
1031                 if ((vma->vm_ops == &shm_vm_ops) &&
1032                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1033
1034                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1035                 vma = next;
1036         }
1037
1038 #else /* CONFIG_MMU */
1039         /* under NOMMU conditions, the exact address to be destroyed must be
1040          * given */
1041         retval = -EINVAL;
1042         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1043                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1044                 retval = 0;
1045         }
1046
1047 #endif
1048
1049         up_write(&mm->mmap_sem);
1050         return retval;
1051 }
1052
1053 #ifdef CONFIG_PROC_FS
1054 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1055 {
1056         struct shmid_kernel *shp = it;
1057
1058 #if BITS_PER_LONG <= 32
1059 #define SIZE_SPEC "%10lu"
1060 #else
1061 #define SIZE_SPEC "%21lu"
1062 #endif
1063
1064         return seq_printf(s,
1065                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1066                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1067                           shp->shm_perm.key,
1068                           shp->shm_perm.id,
1069                           shp->shm_perm.mode,
1070                           shp->shm_segsz,
1071                           shp->shm_cprid,
1072                           shp->shm_lprid,
1073                           shp->shm_nattch,
1074                           shp->shm_perm.uid,
1075                           shp->shm_perm.gid,
1076                           shp->shm_perm.cuid,
1077                           shp->shm_perm.cgid,
1078                           shp->shm_atim,
1079                           shp->shm_dtim,
1080                           shp->shm_ctim);
1081 }
1082 #endif