fdaf8be65b759369482bb5aaa477346bed8741af
[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_rmid_forced = 0;
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         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
105 }
106 #endif
107
108 void __init shm_init (void)
109 {
110         shm_init_ns(&init_ipc_ns);
111         ipc_init_proc_interface("sysvipc/shm",
112 #if BITS_PER_LONG <= 32
113                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
114 #else
115                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
116 #endif
117                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
118 }
119
120 /*
121  * shm_lock_(check_) routines are called in the paths where the rw_mutex
122  * is not necessarily held.
123  */
124 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
125 {
126         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
127
128         if (IS_ERR(ipcp))
129                 return (struct shmid_kernel *)ipcp;
130
131         return container_of(ipcp, struct shmid_kernel, shm_perm);
132 }
133
134 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
135                                                 int id)
136 {
137         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
138
139         if (IS_ERR(ipcp))
140                 return (struct shmid_kernel *)ipcp;
141
142         return container_of(ipcp, struct shmid_kernel, shm_perm);
143 }
144
145 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
146 {
147         ipc_rmid(&shm_ids(ns), &s->shm_perm);
148 }
149
150
151 /* This is called by fork, once for every shm attach. */
152 static void shm_open(struct vm_area_struct *vma)
153 {
154         struct file *file = vma->vm_file;
155         struct shm_file_data *sfd = shm_file_data(file);
156         struct shmid_kernel *shp;
157
158         shp = shm_lock(sfd->ns, sfd->id);
159         BUG_ON(IS_ERR(shp));
160         shp->shm_atim = get_seconds();
161         shp->shm_lprid = task_tgid_vnr(current);
162         shp->shm_nattch++;
163         shm_unlock(shp);
164 }
165
166 /*
167  * shm_destroy - free the struct shmid_kernel
168  *
169  * @ns: namespace
170  * @shp: struct to free
171  *
172  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
173  * but returns with shp unlocked and freed.
174  */
175 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
176 {
177         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
178         shm_rmid(ns, shp);
179         shm_unlock(shp);
180         if (!is_file_hugepages(shp->shm_file))
181                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
182         else if (shp->mlock_user)
183                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
184                                                 shp->mlock_user);
185         fput (shp->shm_file);
186         security_shm_free(shp);
187         ipc_rcu_putref(shp);
188 }
189
190 /*
191  * shm_may_destroy - identifies whether shm segment should be destroyed now
192  *
193  * Returns true if and only if there are no active users of the segment and
194  * one of the following is true:
195  *
196  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
197  *
198  * 2) sysctl kernel.shm_rmid_forced is set to 1.
199  */
200 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
201 {
202         return (shp->shm_nattch == 0) &&
203                (ns->shm_rmid_forced ||
204                 (shp->shm_perm.mode & SHM_DEST));
205 }
206
207 /*
208  * remove the attach descriptor vma.
209  * free memory for segment if it is marked destroyed.
210  * The descriptor has already been removed from the current->mm->mmap list
211  * and will later be kfree()d.
212  */
213 static void shm_close(struct vm_area_struct *vma)
214 {
215         struct file * file = vma->vm_file;
216         struct shm_file_data *sfd = shm_file_data(file);
217         struct shmid_kernel *shp;
218         struct ipc_namespace *ns = sfd->ns;
219
220         down_write(&shm_ids(ns).rw_mutex);
221         /* remove from the list of attaches of the shm segment */
222         shp = shm_lock(ns, sfd->id);
223         BUG_ON(IS_ERR(shp));
224         shp->shm_lprid = task_tgid_vnr(current);
225         shp->shm_dtim = get_seconds();
226         shp->shm_nattch--;
227         if (shm_may_destroy(ns, shp))
228                 shm_destroy(ns, shp);
229         else
230                 shm_unlock(shp);
231         up_write(&shm_ids(ns).rw_mutex);
232 }
233
234 static int shm_try_destroy_current(int id, void *p, void *data)
235 {
236         struct ipc_namespace *ns = data;
237         struct shmid_kernel *shp = shm_lock(ns, id);
238
239         if (IS_ERR(shp))
240                 return 0;
241
242         if (shp->shm_creator != current) {
243                 shm_unlock(shp);
244                 return 0;
245         }
246
247         /*
248          * Mark it as orphaned to destroy the segment when
249          * kernel.shm_rmid_forced is changed.
250          * It is noop if the following shm_may_destroy() returns true.
251          */
252         shp->shm_creator = NULL;
253
254         /*
255          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
256          * is not set, it shouldn't be deleted here.
257          */
258         if (!ns->shm_rmid_forced) {
259                 shm_unlock(shp);
260                 return 0;
261         }
262
263         if (shm_may_destroy(ns, shp))
264                 shm_destroy(ns, shp);
265         else
266                 shm_unlock(shp);
267         return 0;
268 }
269
270 static int shm_try_destroy_orphaned(int id, void *p, void *data)
271 {
272         struct ipc_namespace *ns = data;
273         struct shmid_kernel *shp = shm_lock(ns, id);
274
275         if (IS_ERR(shp))
276                 return 0;
277
278         /*
279          * We want to destroy segments without users and with already
280          * exit'ed originating process.
281          */
282         if (shp->shm_creator != NULL) {
283                 shm_unlock(shp);
284                 return 0;
285         }
286
287         if (shm_may_destroy(ns, shp))
288                 shm_destroy(ns, shp);
289         else
290                 shm_unlock(shp);
291         return 0;
292 }
293
294 void shm_destroy_orphaned(struct ipc_namespace *ns)
295 {
296         down_write(&shm_ids(ns).rw_mutex);
297         idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
298         up_write(&shm_ids(ns).rw_mutex);
299 }
300
301
302 void exit_shm(struct task_struct *task)
303 {
304         struct nsproxy *nsp = task->nsproxy;
305         struct ipc_namespace *ns;
306
307         if (!nsp)
308                 return;
309         ns = nsp->ipc_ns;
310         if (!ns)
311                 return;
312
313         /* Destroy all already created segments, but not mapped yet */
314         down_write(&shm_ids(ns).rw_mutex);
315         idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
316         up_write(&shm_ids(ns).rw_mutex);
317 }
318
319 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
320 {
321         struct file *file = vma->vm_file;
322         struct shm_file_data *sfd = shm_file_data(file);
323
324         return sfd->vm_ops->fault(vma, vmf);
325 }
326
327 #ifdef CONFIG_NUMA
328 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
329 {
330         struct file *file = vma->vm_file;
331         struct shm_file_data *sfd = shm_file_data(file);
332         int err = 0;
333         if (sfd->vm_ops->set_policy)
334                 err = sfd->vm_ops->set_policy(vma, new);
335         return err;
336 }
337
338 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
339                                         unsigned long addr)
340 {
341         struct file *file = vma->vm_file;
342         struct shm_file_data *sfd = shm_file_data(file);
343         struct mempolicy *pol = NULL;
344
345         if (sfd->vm_ops->get_policy)
346                 pol = sfd->vm_ops->get_policy(vma, addr);
347         else if (vma->vm_policy)
348                 pol = vma->vm_policy;
349
350         return pol;
351 }
352 #endif
353
354 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
355 {
356         struct shm_file_data *sfd = shm_file_data(file);
357         int ret;
358
359         ret = sfd->file->f_op->mmap(sfd->file, vma);
360         if (ret != 0)
361                 return ret;
362         sfd->vm_ops = vma->vm_ops;
363 #ifdef CONFIG_MMU
364         BUG_ON(!sfd->vm_ops->fault);
365 #endif
366         vma->vm_ops = &shm_vm_ops;
367         shm_open(vma);
368
369         return ret;
370 }
371
372 static int shm_release(struct inode *ino, struct file *file)
373 {
374         struct shm_file_data *sfd = shm_file_data(file);
375
376         put_ipc_ns(sfd->ns);
377         shm_file_data(file) = NULL;
378         kfree(sfd);
379         return 0;
380 }
381
382 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
383 {
384         struct shm_file_data *sfd = shm_file_data(file);
385
386         if (!sfd->file->f_op->fsync)
387                 return -EINVAL;
388         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
389 }
390
391 static unsigned long shm_get_unmapped_area(struct file *file,
392         unsigned long addr, unsigned long len, unsigned long pgoff,
393         unsigned long flags)
394 {
395         struct shm_file_data *sfd = shm_file_data(file);
396         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
397                                                 pgoff, flags);
398 }
399
400 static const struct file_operations shm_file_operations = {
401         .mmap           = shm_mmap,
402         .fsync          = shm_fsync,
403         .release        = shm_release,
404 #ifndef CONFIG_MMU
405         .get_unmapped_area      = shm_get_unmapped_area,
406 #endif
407         .llseek         = noop_llseek,
408 };
409
410 static const struct file_operations shm_file_operations_huge = {
411         .mmap           = shm_mmap,
412         .fsync          = shm_fsync,
413         .release        = shm_release,
414         .get_unmapped_area      = shm_get_unmapped_area,
415         .llseek         = noop_llseek,
416 };
417
418 int is_file_shm_hugepages(struct file *file)
419 {
420         return file->f_op == &shm_file_operations_huge;
421 }
422
423 static const struct vm_operations_struct shm_vm_ops = {
424         .open   = shm_open,     /* callback for a new vm-area open */
425         .close  = shm_close,    /* callback for when the vm-area is released */
426         .fault  = shm_fault,
427 #if defined(CONFIG_NUMA)
428         .set_policy = shm_set_policy,
429         .get_policy = shm_get_policy,
430 #endif
431 };
432
433 /**
434  * newseg - Create a new shared memory segment
435  * @ns: namespace
436  * @params: ptr to the structure that contains key, size and shmflg
437  *
438  * Called with shm_ids.rw_mutex held as a writer.
439  */
440
441 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
442 {
443         key_t key = params->key;
444         int shmflg = params->flg;
445         size_t size = params->u.size;
446         int error;
447         struct shmid_kernel *shp;
448         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
449         struct file * file;
450         char name[13];
451         int id;
452         vm_flags_t acctflag = 0;
453
454         if (size < SHMMIN || size > ns->shm_ctlmax)
455                 return -EINVAL;
456
457         if (ns->shm_tot + numpages > ns->shm_ctlall)
458                 return -ENOSPC;
459
460         shp = ipc_rcu_alloc(sizeof(*shp));
461         if (!shp)
462                 return -ENOMEM;
463
464         shp->shm_perm.key = key;
465         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
466         shp->mlock_user = NULL;
467
468         shp->shm_perm.security = NULL;
469         error = security_shm_alloc(shp);
470         if (error) {
471                 ipc_rcu_putref(shp);
472                 return error;
473         }
474
475         sprintf (name, "SYSV%08x", key);
476         if (shmflg & SHM_HUGETLB) {
477                 /* hugetlb_file_setup applies strict accounting */
478                 if (shmflg & SHM_NORESERVE)
479                         acctflag = VM_NORESERVE;
480                 file = hugetlb_file_setup(name, size, acctflag,
481                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
482         } else {
483                 /*
484                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
485                  * if it's asked for.
486                  */
487                 if  ((shmflg & SHM_NORESERVE) &&
488                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
489                         acctflag = VM_NORESERVE;
490                 file = shmem_file_setup(name, size, acctflag);
491         }
492         error = PTR_ERR(file);
493         if (IS_ERR(file))
494                 goto no_file;
495
496         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
497         if (id < 0) {
498                 error = id;
499                 goto no_id;
500         }
501
502         shp->shm_cprid = task_tgid_vnr(current);
503         shp->shm_lprid = 0;
504         shp->shm_atim = shp->shm_dtim = 0;
505         shp->shm_ctim = get_seconds();
506         shp->shm_segsz = size;
507         shp->shm_nattch = 0;
508         shp->shm_file = file;
509         shp->shm_creator = current;
510         /*
511          * shmid gets reported as "inode#" in /proc/pid/maps.
512          * proc-ps tools use this. Changing this will break them.
513          */
514         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
515
516         ns->shm_tot += numpages;
517         error = shp->shm_perm.id;
518         shm_unlock(shp);
519         return error;
520
521 no_id:
522         if (is_file_hugepages(file) && shp->mlock_user)
523                 user_shm_unlock(size, shp->mlock_user);
524         fput(file);
525 no_file:
526         security_shm_free(shp);
527         ipc_rcu_putref(shp);
528         return error;
529 }
530
531 /*
532  * Called with shm_ids.rw_mutex and ipcp locked.
533  */
534 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
535 {
536         struct shmid_kernel *shp;
537
538         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
539         return security_shm_associate(shp, shmflg);
540 }
541
542 /*
543  * Called with shm_ids.rw_mutex and ipcp locked.
544  */
545 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
546                                 struct ipc_params *params)
547 {
548         struct shmid_kernel *shp;
549
550         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
551         if (shp->shm_segsz < params->u.size)
552                 return -EINVAL;
553
554         return 0;
555 }
556
557 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
558 {
559         struct ipc_namespace *ns;
560         struct ipc_ops shm_ops;
561         struct ipc_params shm_params;
562
563         ns = current->nsproxy->ipc_ns;
564
565         shm_ops.getnew = newseg;
566         shm_ops.associate = shm_security;
567         shm_ops.more_checks = shm_more_checks;
568
569         shm_params.key = key;
570         shm_params.flg = shmflg;
571         shm_params.u.size = size;
572
573         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
574 }
575
576 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
577 {
578         switch(version) {
579         case IPC_64:
580                 return copy_to_user(buf, in, sizeof(*in));
581         case IPC_OLD:
582             {
583                 struct shmid_ds out;
584
585                 memset(&out, 0, sizeof(out));
586                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
587                 out.shm_segsz   = in->shm_segsz;
588                 out.shm_atime   = in->shm_atime;
589                 out.shm_dtime   = in->shm_dtime;
590                 out.shm_ctime   = in->shm_ctime;
591                 out.shm_cpid    = in->shm_cpid;
592                 out.shm_lpid    = in->shm_lpid;
593                 out.shm_nattch  = in->shm_nattch;
594
595                 return copy_to_user(buf, &out, sizeof(out));
596             }
597         default:
598                 return -EINVAL;
599         }
600 }
601
602 static inline unsigned long
603 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
604 {
605         switch(version) {
606         case IPC_64:
607                 if (copy_from_user(out, buf, sizeof(*out)))
608                         return -EFAULT;
609                 return 0;
610         case IPC_OLD:
611             {
612                 struct shmid_ds tbuf_old;
613
614                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
615                         return -EFAULT;
616
617                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
618                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
619                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
620
621                 return 0;
622             }
623         default:
624                 return -EINVAL;
625         }
626 }
627
628 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
629 {
630         switch(version) {
631         case IPC_64:
632                 return copy_to_user(buf, in, sizeof(*in));
633         case IPC_OLD:
634             {
635                 struct shminfo out;
636
637                 if(in->shmmax > INT_MAX)
638                         out.shmmax = INT_MAX;
639                 else
640                         out.shmmax = (int)in->shmmax;
641
642                 out.shmmin      = in->shmmin;
643                 out.shmmni      = in->shmmni;
644                 out.shmseg      = in->shmseg;
645                 out.shmall      = in->shmall; 
646
647                 return copy_to_user(buf, &out, sizeof(out));
648             }
649         default:
650                 return -EINVAL;
651         }
652 }
653
654 /*
655  * Calculate and add used RSS and swap pages of a shm.
656  * Called with shm_ids.rw_mutex held as a reader
657  */
658 static void shm_add_rss_swap(struct shmid_kernel *shp,
659         unsigned long *rss_add, unsigned long *swp_add)
660 {
661         struct inode *inode;
662
663         inode = shp->shm_file->f_path.dentry->d_inode;
664
665         if (is_file_hugepages(shp->shm_file)) {
666                 struct address_space *mapping = inode->i_mapping;
667                 struct hstate *h = hstate_file(shp->shm_file);
668                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
669         } else {
670 #ifdef CONFIG_SHMEM
671                 struct shmem_inode_info *info = SHMEM_I(inode);
672                 spin_lock(&info->lock);
673                 *rss_add += inode->i_mapping->nrpages;
674                 *swp_add += info->swapped;
675                 spin_unlock(&info->lock);
676 #else
677                 *rss_add += inode->i_mapping->nrpages;
678 #endif
679         }
680 }
681
682 /*
683  * Called with shm_ids.rw_mutex held as a reader
684  */
685 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
686                 unsigned long *swp)
687 {
688         int next_id;
689         int total, in_use;
690
691         *rss = 0;
692         *swp = 0;
693
694         in_use = shm_ids(ns).in_use;
695
696         for (total = 0, next_id = 0; total < in_use; next_id++) {
697                 struct kern_ipc_perm *ipc;
698                 struct shmid_kernel *shp;
699
700                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
701                 if (ipc == NULL)
702                         continue;
703                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
704
705                 shm_add_rss_swap(shp, rss, swp);
706
707                 total++;
708         }
709 }
710
711 /*
712  * This function handles some shmctl commands which require the rw_mutex
713  * to be held in write mode.
714  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
715  */
716 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
717                        struct shmid_ds __user *buf, int version)
718 {
719         struct kern_ipc_perm *ipcp;
720         struct shmid64_ds shmid64;
721         struct shmid_kernel *shp;
722         int err;
723
724         if (cmd == IPC_SET) {
725                 if (copy_shmid_from_user(&shmid64, buf, version))
726                         return -EFAULT;
727         }
728
729         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
730                                &shmid64.shm_perm, 0);
731         if (IS_ERR(ipcp))
732                 return PTR_ERR(ipcp);
733
734         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
735
736         err = security_shm_shmctl(shp, cmd);
737         if (err)
738                 goto out_unlock;
739         switch (cmd) {
740         case IPC_RMID:
741                 do_shm_rmid(ns, ipcp);
742                 goto out_up;
743         case IPC_SET:
744                 ipc_update_perm(&shmid64.shm_perm, ipcp);
745                 shp->shm_ctim = get_seconds();
746                 break;
747         default:
748                 err = -EINVAL;
749         }
750 out_unlock:
751         shm_unlock(shp);
752 out_up:
753         up_write(&shm_ids(ns).rw_mutex);
754         return err;
755 }
756
757 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
758 {
759         struct shmid_kernel *shp;
760         int err, version;
761         struct ipc_namespace *ns;
762
763         if (cmd < 0 || shmid < 0) {
764                 err = -EINVAL;
765                 goto out;
766         }
767
768         version = ipc_parse_version(&cmd);
769         ns = current->nsproxy->ipc_ns;
770
771         switch (cmd) { /* replace with proc interface ? */
772         case IPC_INFO:
773         {
774                 struct shminfo64 shminfo;
775
776                 err = security_shm_shmctl(NULL, cmd);
777                 if (err)
778                         return err;
779
780                 memset(&shminfo, 0, sizeof(shminfo));
781                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
782                 shminfo.shmmax = ns->shm_ctlmax;
783                 shminfo.shmall = ns->shm_ctlall;
784
785                 shminfo.shmmin = SHMMIN;
786                 if(copy_shminfo_to_user (buf, &shminfo, version))
787                         return -EFAULT;
788
789                 down_read(&shm_ids(ns).rw_mutex);
790                 err = ipc_get_maxid(&shm_ids(ns));
791                 up_read(&shm_ids(ns).rw_mutex);
792
793                 if(err<0)
794                         err = 0;
795                 goto out;
796         }
797         case SHM_INFO:
798         {
799                 struct shm_info shm_info;
800
801                 err = security_shm_shmctl(NULL, cmd);
802                 if (err)
803                         return err;
804
805                 memset(&shm_info, 0, sizeof(shm_info));
806                 down_read(&shm_ids(ns).rw_mutex);
807                 shm_info.used_ids = shm_ids(ns).in_use;
808                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
809                 shm_info.shm_tot = ns->shm_tot;
810                 shm_info.swap_attempts = 0;
811                 shm_info.swap_successes = 0;
812                 err = ipc_get_maxid(&shm_ids(ns));
813                 up_read(&shm_ids(ns).rw_mutex);
814                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
815                         err = -EFAULT;
816                         goto out;
817                 }
818
819                 err = err < 0 ? 0 : err;
820                 goto out;
821         }
822         case SHM_STAT:
823         case IPC_STAT:
824         {
825                 struct shmid64_ds tbuf;
826                 int result;
827
828                 if (cmd == SHM_STAT) {
829                         shp = shm_lock(ns, shmid);
830                         if (IS_ERR(shp)) {
831                                 err = PTR_ERR(shp);
832                                 goto out;
833                         }
834                         result = shp->shm_perm.id;
835                 } else {
836                         shp = shm_lock_check(ns, shmid);
837                         if (IS_ERR(shp)) {
838                                 err = PTR_ERR(shp);
839                                 goto out;
840                         }
841                         result = 0;
842                 }
843                 err = -EACCES;
844                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
845                         goto out_unlock;
846                 err = security_shm_shmctl(shp, cmd);
847                 if (err)
848                         goto out_unlock;
849                 memset(&tbuf, 0, sizeof(tbuf));
850                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
851                 tbuf.shm_segsz  = shp->shm_segsz;
852                 tbuf.shm_atime  = shp->shm_atim;
853                 tbuf.shm_dtime  = shp->shm_dtim;
854                 tbuf.shm_ctime  = shp->shm_ctim;
855                 tbuf.shm_cpid   = shp->shm_cprid;
856                 tbuf.shm_lpid   = shp->shm_lprid;
857                 tbuf.shm_nattch = shp->shm_nattch;
858                 shm_unlock(shp);
859                 if(copy_shmid_to_user (buf, &tbuf, version))
860                         err = -EFAULT;
861                 else
862                         err = result;
863                 goto out;
864         }
865         case SHM_LOCK:
866         case SHM_UNLOCK:
867         {
868                 struct file *uninitialized_var(shm_file);
869
870                 lru_add_drain_all();  /* drain pagevecs to lru lists */
871
872                 shp = shm_lock_check(ns, shmid);
873                 if (IS_ERR(shp)) {
874                         err = PTR_ERR(shp);
875                         goto out;
876                 }
877
878                 audit_ipc_obj(&(shp->shm_perm));
879
880                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
881                         uid_t euid = current_euid();
882                         err = -EPERM;
883                         if (euid != shp->shm_perm.uid &&
884                             euid != shp->shm_perm.cuid)
885                                 goto out_unlock;
886                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
887                                 goto out_unlock;
888                 }
889
890                 err = security_shm_shmctl(shp, cmd);
891                 if (err)
892                         goto out_unlock;
893                 
894                 if(cmd==SHM_LOCK) {
895                         struct user_struct *user = current_user();
896                         if (!is_file_hugepages(shp->shm_file)) {
897                                 err = shmem_lock(shp->shm_file, 1, user);
898                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
899                                         shp->shm_perm.mode |= SHM_LOCKED;
900                                         shp->mlock_user = user;
901                                 }
902                         }
903                 } else if (!is_file_hugepages(shp->shm_file)) {
904                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
905                         shp->shm_perm.mode &= ~SHM_LOCKED;
906                         shp->mlock_user = NULL;
907                 }
908                 shm_unlock(shp);
909                 goto out;
910         }
911         case IPC_RMID:
912         case IPC_SET:
913                 err = shmctl_down(ns, shmid, cmd, buf, version);
914                 return err;
915         default:
916                 return -EINVAL;
917         }
918
919 out_unlock:
920         shm_unlock(shp);
921 out:
922         return err;
923 }
924
925 /*
926  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
927  *
928  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
929  * "raddr" thing points to kernel space, and there has to be a wrapper around
930  * this.
931  */
932 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
933 {
934         struct shmid_kernel *shp;
935         unsigned long addr;
936         unsigned long size;
937         struct file * file;
938         int    err;
939         unsigned long flags;
940         unsigned long prot;
941         int acc_mode;
942         unsigned long user_addr;
943         struct ipc_namespace *ns;
944         struct shm_file_data *sfd;
945         struct path path;
946         fmode_t f_mode;
947
948         err = -EINVAL;
949         if (shmid < 0)
950                 goto out;
951         else if ((addr = (ulong)shmaddr)) {
952                 if (addr & (SHMLBA-1)) {
953                         if (shmflg & SHM_RND)
954                                 addr &= ~(SHMLBA-1);       /* round down */
955                         else
956 #ifndef __ARCH_FORCE_SHMLBA
957                                 if (addr & ~PAGE_MASK)
958 #endif
959                                         goto out;
960                 }
961                 flags = MAP_SHARED | MAP_FIXED;
962         } else {
963                 if ((shmflg & SHM_REMAP))
964                         goto out;
965
966                 flags = MAP_SHARED;
967         }
968
969         if (shmflg & SHM_RDONLY) {
970                 prot = PROT_READ;
971                 acc_mode = S_IRUGO;
972                 f_mode = FMODE_READ;
973         } else {
974                 prot = PROT_READ | PROT_WRITE;
975                 acc_mode = S_IRUGO | S_IWUGO;
976                 f_mode = FMODE_READ | FMODE_WRITE;
977         }
978         if (shmflg & SHM_EXEC) {
979                 prot |= PROT_EXEC;
980                 acc_mode |= S_IXUGO;
981         }
982
983         /*
984          * We cannot rely on the fs check since SYSV IPC does have an
985          * additional creator id...
986          */
987         ns = current->nsproxy->ipc_ns;
988         shp = shm_lock_check(ns, shmid);
989         if (IS_ERR(shp)) {
990                 err = PTR_ERR(shp);
991                 goto out;
992         }
993
994         err = -EACCES;
995         if (ipcperms(ns, &shp->shm_perm, acc_mode))
996                 goto out_unlock;
997
998         err = security_shm_shmat(shp, shmaddr, shmflg);
999         if (err)
1000                 goto out_unlock;
1001
1002         path = shp->shm_file->f_path;
1003         path_get(&path);
1004         shp->shm_nattch++;
1005         size = i_size_read(path.dentry->d_inode);
1006         shm_unlock(shp);
1007
1008         err = -ENOMEM;
1009         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1010         if (!sfd)
1011                 goto out_put_dentry;
1012
1013         file = alloc_file(&path, f_mode,
1014                           is_file_hugepages(shp->shm_file) ?
1015                                 &shm_file_operations_huge :
1016                                 &shm_file_operations);
1017         if (!file)
1018                 goto out_free;
1019
1020         file->private_data = sfd;
1021         file->f_mapping = shp->shm_file->f_mapping;
1022         sfd->id = shp->shm_perm.id;
1023         sfd->ns = get_ipc_ns(ns);
1024         sfd->file = shp->shm_file;
1025         sfd->vm_ops = NULL;
1026
1027         down_write(&current->mm->mmap_sem);
1028         if (addr && !(shmflg & SHM_REMAP)) {
1029                 err = -EINVAL;
1030                 if (find_vma_intersection(current->mm, addr, addr + size))
1031                         goto invalid;
1032                 /*
1033                  * If shm segment goes below stack, make sure there is some
1034                  * space left for the stack to grow (at least 4 pages).
1035                  */
1036                 if (addr < current->mm->start_stack &&
1037                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1038                         goto invalid;
1039         }
1040                 
1041         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1042         *raddr = user_addr;
1043         err = 0;
1044         if (IS_ERR_VALUE(user_addr))
1045                 err = (long)user_addr;
1046 invalid:
1047         up_write(&current->mm->mmap_sem);
1048
1049         fput(file);
1050
1051 out_nattch:
1052         down_write(&shm_ids(ns).rw_mutex);
1053         shp = shm_lock(ns, shmid);
1054         BUG_ON(IS_ERR(shp));
1055         shp->shm_nattch--;
1056         if (shm_may_destroy(ns, shp))
1057                 shm_destroy(ns, shp);
1058         else
1059                 shm_unlock(shp);
1060         up_write(&shm_ids(ns).rw_mutex);
1061
1062 out:
1063         return err;
1064
1065 out_unlock:
1066         shm_unlock(shp);
1067         goto out;
1068
1069 out_free:
1070         kfree(sfd);
1071 out_put_dentry:
1072         path_put(&path);
1073         goto out_nattch;
1074 }
1075
1076 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1077 {
1078         unsigned long ret;
1079         long err;
1080
1081         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1082         if (err)
1083                 return err;
1084         force_successful_syscall_return();
1085         return (long)ret;
1086 }
1087
1088 /*
1089  * detach and kill segment if marked destroyed.
1090  * The work is done in shm_close.
1091  */
1092 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1093 {
1094         struct mm_struct *mm = current->mm;
1095         struct vm_area_struct *vma;
1096         unsigned long addr = (unsigned long)shmaddr;
1097         int retval = -EINVAL;
1098 #ifdef CONFIG_MMU
1099         loff_t size = 0;
1100         struct vm_area_struct *next;
1101 #endif
1102
1103         if (addr & ~PAGE_MASK)
1104                 return retval;
1105
1106         down_write(&mm->mmap_sem);
1107
1108         /*
1109          * This function tries to be smart and unmap shm segments that
1110          * were modified by partial mlock or munmap calls:
1111          * - It first determines the size of the shm segment that should be
1112          *   unmapped: It searches for a vma that is backed by shm and that
1113          *   started at address shmaddr. It records it's size and then unmaps
1114          *   it.
1115          * - Then it unmaps all shm vmas that started at shmaddr and that
1116          *   are within the initially determined size.
1117          * Errors from do_munmap are ignored: the function only fails if
1118          * it's called with invalid parameters or if it's called to unmap
1119          * a part of a vma. Both calls in this function are for full vmas,
1120          * the parameters are directly copied from the vma itself and always
1121          * valid - therefore do_munmap cannot fail. (famous last words?)
1122          */
1123         /*
1124          * If it had been mremap()'d, the starting address would not
1125          * match the usual checks anyway. So assume all vma's are
1126          * above the starting address given.
1127          */
1128         vma = find_vma(mm, addr);
1129
1130 #ifdef CONFIG_MMU
1131         while (vma) {
1132                 next = vma->vm_next;
1133
1134                 /*
1135                  * Check if the starting address would match, i.e. it's
1136                  * a fragment created by mprotect() and/or munmap(), or it
1137                  * otherwise it starts at this address with no hassles.
1138                  */
1139                 if ((vma->vm_ops == &shm_vm_ops) &&
1140                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1141
1142
1143                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1144                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1145                         /*
1146                          * We discovered the size of the shm segment, so
1147                          * break out of here and fall through to the next
1148                          * loop that uses the size information to stop
1149                          * searching for matching vma's.
1150                          */
1151                         retval = 0;
1152                         vma = next;
1153                         break;
1154                 }
1155                 vma = next;
1156         }
1157
1158         /*
1159          * We need look no further than the maximum address a fragment
1160          * could possibly have landed at. Also cast things to loff_t to
1161          * prevent overflows and make comparisons vs. equal-width types.
1162          */
1163         size = PAGE_ALIGN(size);
1164         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1165                 next = vma->vm_next;
1166
1167                 /* finding a matching vma now does not alter retval */
1168                 if ((vma->vm_ops == &shm_vm_ops) &&
1169                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1170
1171                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1172                 vma = next;
1173         }
1174
1175 #else /* CONFIG_MMU */
1176         /* under NOMMU conditions, the exact address to be destroyed must be
1177          * given */
1178         retval = -EINVAL;
1179         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1180                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1181                 retval = 0;
1182         }
1183
1184 #endif
1185
1186         up_write(&mm->mmap_sem);
1187         return retval;
1188 }
1189
1190 #ifdef CONFIG_PROC_FS
1191 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1192 {
1193         struct shmid_kernel *shp = it;
1194         unsigned long rss = 0, swp = 0;
1195
1196         shm_add_rss_swap(shp, &rss, &swp);
1197
1198 #if BITS_PER_LONG <= 32
1199 #define SIZE_SPEC "%10lu"
1200 #else
1201 #define SIZE_SPEC "%21lu"
1202 #endif
1203
1204         return seq_printf(s,
1205                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1206                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1207                           SIZE_SPEC " " SIZE_SPEC "\n",
1208                           shp->shm_perm.key,
1209                           shp->shm_perm.id,
1210                           shp->shm_perm.mode,
1211                           shp->shm_segsz,
1212                           shp->shm_cprid,
1213                           shp->shm_lprid,
1214                           shp->shm_nattch,
1215                           shp->shm_perm.uid,
1216                           shp->shm_perm.gid,
1217                           shp->shm_perm.cuid,
1218                           shp->shm_perm.cgid,
1219                           shp->shm_atim,
1220                           shp->shm_dtim,
1221                           shp->shm_ctim,
1222                           rss * PAGE_SIZE,
1223                           swp * PAGE_SIZE);
1224 }
1225 #endif