9fb044f3b345e5e61ea072e255ab3b67b5ce3539
[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 void shm_lock_by_ptr(struct shmid_kernel *ipcp)
135 {
136         rcu_read_lock();
137         spin_lock(&ipcp->shm_perm.lock);
138 }
139
140 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
141                                                 int id)
142 {
143         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
144
145         if (IS_ERR(ipcp))
146                 return (struct shmid_kernel *)ipcp;
147
148         return container_of(ipcp, struct shmid_kernel, shm_perm);
149 }
150
151 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
152 {
153         ipc_rmid(&shm_ids(ns), &s->shm_perm);
154 }
155
156
157 /* This is called by fork, once for every shm attach. */
158 static void shm_open(struct vm_area_struct *vma)
159 {
160         struct file *file = vma->vm_file;
161         struct shm_file_data *sfd = shm_file_data(file);
162         struct shmid_kernel *shp;
163
164         shp = shm_lock(sfd->ns, sfd->id);
165         BUG_ON(IS_ERR(shp));
166         shp->shm_atim = get_seconds();
167         shp->shm_lprid = task_tgid_vnr(current);
168         shp->shm_nattch++;
169         shm_unlock(shp);
170 }
171
172 /*
173  * shm_destroy - free the struct shmid_kernel
174  *
175  * @ns: namespace
176  * @shp: struct to free
177  *
178  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
179  * but returns with shp unlocked and freed.
180  */
181 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
182 {
183         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
184         shm_rmid(ns, shp);
185         shm_unlock(shp);
186         if (!is_file_hugepages(shp->shm_file))
187                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
188         else if (shp->mlock_user)
189                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
190                                                 shp->mlock_user);
191         fput (shp->shm_file);
192         security_shm_free(shp);
193         ipc_rcu_putref(shp);
194 }
195
196 /*
197  * shm_may_destroy - identifies whether shm segment should be destroyed now
198  *
199  * Returns true if and only if there are no active users of the segment and
200  * one of the following is true:
201  *
202  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
203  *
204  * 2) sysctl kernel.shm_rmid_forced is set to 1.
205  */
206 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
207 {
208         return (shp->shm_nattch == 0) &&
209                (ns->shm_rmid_forced ||
210                 (shp->shm_perm.mode & SHM_DEST));
211 }
212
213 /*
214  * remove the attach descriptor vma.
215  * free memory for segment if it is marked destroyed.
216  * The descriptor has already been removed from the current->mm->mmap list
217  * and will later be kfree()d.
218  */
219 static void shm_close(struct vm_area_struct *vma)
220 {
221         struct file * file = vma->vm_file;
222         struct shm_file_data *sfd = shm_file_data(file);
223         struct shmid_kernel *shp;
224         struct ipc_namespace *ns = sfd->ns;
225
226         down_write(&shm_ids(ns).rw_mutex);
227         /* remove from the list of attaches of the shm segment */
228         shp = shm_lock(ns, sfd->id);
229         BUG_ON(IS_ERR(shp));
230         shp->shm_lprid = task_tgid_vnr(current);
231         shp->shm_dtim = get_seconds();
232         shp->shm_nattch--;
233         if (shm_may_destroy(ns, shp))
234                 shm_destroy(ns, shp);
235         else
236                 shm_unlock(shp);
237         up_write(&shm_ids(ns).rw_mutex);
238 }
239
240 /* Called with ns->shm_ids(ns).rw_mutex locked */
241 static int shm_try_destroy_current(int id, void *p, void *data)
242 {
243         struct ipc_namespace *ns = data;
244         struct kern_ipc_perm *ipcp = p;
245         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
246
247         if (shp->shm_creator != current)
248                 return 0;
249
250         /*
251          * Mark it as orphaned to destroy the segment when
252          * kernel.shm_rmid_forced is changed.
253          * It is noop if the following shm_may_destroy() returns true.
254          */
255         shp->shm_creator = NULL;
256
257         /*
258          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
259          * is not set, it shouldn't be deleted here.
260          */
261         if (!ns->shm_rmid_forced)
262                 return 0;
263
264         if (shm_may_destroy(ns, shp)) {
265                 shm_lock_by_ptr(shp);
266                 shm_destroy(ns, shp);
267         }
268         return 0;
269 }
270
271 /* Called with ns->shm_ids(ns).rw_mutex locked */
272 static int shm_try_destroy_orphaned(int id, void *p, void *data)
273 {
274         struct ipc_namespace *ns = data;
275         struct kern_ipc_perm *ipcp = p;
276         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
277
278         /*
279          * We want to destroy segments without users and with already
280          * exit'ed originating process.
281          *
282          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
283          */
284         if (shp->shm_creator != NULL)
285                 return 0;
286
287         if (shm_may_destroy(ns, shp)) {
288                 shm_lock_by_ptr(shp);
289                 shm_destroy(ns, shp);
290         }
291         return 0;
292 }
293
294 void shm_destroy_orphaned(struct ipc_namespace *ns)
295 {
296         down_write(&shm_ids(ns).rw_mutex);
297         if (&shm_ids(ns).in_use)
298                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
299         up_write(&shm_ids(ns).rw_mutex);
300 }
301
302
303 void exit_shm(struct task_struct *task)
304 {
305         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
306
307         /* Destroy all already created segments, but not mapped yet */
308         down_write(&shm_ids(ns).rw_mutex);
309         if (&shm_ids(ns).in_use)
310                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
311         up_write(&shm_ids(ns).rw_mutex);
312 }
313
314 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
315 {
316         struct file *file = vma->vm_file;
317         struct shm_file_data *sfd = shm_file_data(file);
318
319         return sfd->vm_ops->fault(vma, vmf);
320 }
321
322 #ifdef CONFIG_NUMA
323 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
324 {
325         struct file *file = vma->vm_file;
326         struct shm_file_data *sfd = shm_file_data(file);
327         int err = 0;
328         if (sfd->vm_ops->set_policy)
329                 err = sfd->vm_ops->set_policy(vma, new);
330         return err;
331 }
332
333 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
334                                         unsigned long addr)
335 {
336         struct file *file = vma->vm_file;
337         struct shm_file_data *sfd = shm_file_data(file);
338         struct mempolicy *pol = NULL;
339
340         if (sfd->vm_ops->get_policy)
341                 pol = sfd->vm_ops->get_policy(vma, addr);
342         else if (vma->vm_policy)
343                 pol = vma->vm_policy;
344
345         return pol;
346 }
347 #endif
348
349 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
350 {
351         struct shm_file_data *sfd = shm_file_data(file);
352         int ret;
353
354         ret = sfd->file->f_op->mmap(sfd->file, vma);
355         if (ret != 0)
356                 return ret;
357         sfd->vm_ops = vma->vm_ops;
358 #ifdef CONFIG_MMU
359         BUG_ON(!sfd->vm_ops->fault);
360 #endif
361         vma->vm_ops = &shm_vm_ops;
362         shm_open(vma);
363
364         return ret;
365 }
366
367 static int shm_release(struct inode *ino, struct file *file)
368 {
369         struct shm_file_data *sfd = shm_file_data(file);
370
371         put_ipc_ns(sfd->ns);
372         shm_file_data(file) = NULL;
373         kfree(sfd);
374         return 0;
375 }
376
377 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
378 {
379         struct shm_file_data *sfd = shm_file_data(file);
380
381         if (!sfd->file->f_op->fsync)
382                 return -EINVAL;
383         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
384 }
385
386 static unsigned long shm_get_unmapped_area(struct file *file,
387         unsigned long addr, unsigned long len, unsigned long pgoff,
388         unsigned long flags)
389 {
390         struct shm_file_data *sfd = shm_file_data(file);
391         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
392                                                 pgoff, flags);
393 }
394
395 static const struct file_operations shm_file_operations = {
396         .mmap           = shm_mmap,
397         .fsync          = shm_fsync,
398         .release        = shm_release,
399 #ifndef CONFIG_MMU
400         .get_unmapped_area      = shm_get_unmapped_area,
401 #endif
402         .llseek         = noop_llseek,
403 };
404
405 static const struct file_operations shm_file_operations_huge = {
406         .mmap           = shm_mmap,
407         .fsync          = shm_fsync,
408         .release        = shm_release,
409         .get_unmapped_area      = shm_get_unmapped_area,
410         .llseek         = noop_llseek,
411 };
412
413 int is_file_shm_hugepages(struct file *file)
414 {
415         return file->f_op == &shm_file_operations_huge;
416 }
417
418 static const struct vm_operations_struct shm_vm_ops = {
419         .open   = shm_open,     /* callback for a new vm-area open */
420         .close  = shm_close,    /* callback for when the vm-area is released */
421         .fault  = shm_fault,
422 #if defined(CONFIG_NUMA)
423         .set_policy = shm_set_policy,
424         .get_policy = shm_get_policy,
425 #endif
426 };
427
428 /**
429  * newseg - Create a new shared memory segment
430  * @ns: namespace
431  * @params: ptr to the structure that contains key, size and shmflg
432  *
433  * Called with shm_ids.rw_mutex held as a writer.
434  */
435
436 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
437 {
438         key_t key = params->key;
439         int shmflg = params->flg;
440         size_t size = params->u.size;
441         int error;
442         struct shmid_kernel *shp;
443         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
444         struct file * file;
445         char name[13];
446         int id;
447         vm_flags_t acctflag = 0;
448
449         if (size < SHMMIN || size > ns->shm_ctlmax)
450                 return -EINVAL;
451
452         if (ns->shm_tot + numpages > ns->shm_ctlall)
453                 return -ENOSPC;
454
455         shp = ipc_rcu_alloc(sizeof(*shp));
456         if (!shp)
457                 return -ENOMEM;
458
459         shp->shm_perm.key = key;
460         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
461         shp->mlock_user = NULL;
462
463         shp->shm_perm.security = NULL;
464         error = security_shm_alloc(shp);
465         if (error) {
466                 ipc_rcu_putref(shp);
467                 return error;
468         }
469
470         sprintf (name, "SYSV%08x", key);
471         if (shmflg & SHM_HUGETLB) {
472                 /* hugetlb_file_setup applies strict accounting */
473                 if (shmflg & SHM_NORESERVE)
474                         acctflag = VM_NORESERVE;
475                 file = hugetlb_file_setup(name, size, acctflag,
476                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
477         } else {
478                 /*
479                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
480                  * if it's asked for.
481                  */
482                 if  ((shmflg & SHM_NORESERVE) &&
483                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
484                         acctflag = VM_NORESERVE;
485                 file = shmem_file_setup(name, size, acctflag);
486         }
487         error = PTR_ERR(file);
488         if (IS_ERR(file))
489                 goto no_file;
490
491         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
492         if (id < 0) {
493                 error = id;
494                 goto no_id;
495         }
496
497         shp->shm_cprid = task_tgid_vnr(current);
498         shp->shm_lprid = 0;
499         shp->shm_atim = shp->shm_dtim = 0;
500         shp->shm_ctim = get_seconds();
501         shp->shm_segsz = size;
502         shp->shm_nattch = 0;
503         shp->shm_file = file;
504         shp->shm_creator = current;
505         /*
506          * shmid gets reported as "inode#" in /proc/pid/maps.
507          * proc-ps tools use this. Changing this will break them.
508          */
509         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
510
511         ns->shm_tot += numpages;
512         error = shp->shm_perm.id;
513         shm_unlock(shp);
514         return error;
515
516 no_id:
517         if (is_file_hugepages(file) && shp->mlock_user)
518                 user_shm_unlock(size, shp->mlock_user);
519         fput(file);
520 no_file:
521         security_shm_free(shp);
522         ipc_rcu_putref(shp);
523         return error;
524 }
525
526 /*
527  * Called with shm_ids.rw_mutex and ipcp locked.
528  */
529 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
530 {
531         struct shmid_kernel *shp;
532
533         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
534         return security_shm_associate(shp, shmflg);
535 }
536
537 /*
538  * Called with shm_ids.rw_mutex and ipcp locked.
539  */
540 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
541                                 struct ipc_params *params)
542 {
543         struct shmid_kernel *shp;
544
545         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
546         if (shp->shm_segsz < params->u.size)
547                 return -EINVAL;
548
549         return 0;
550 }
551
552 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
553 {
554         struct ipc_namespace *ns;
555         struct ipc_ops shm_ops;
556         struct ipc_params shm_params;
557
558         ns = current->nsproxy->ipc_ns;
559
560         shm_ops.getnew = newseg;
561         shm_ops.associate = shm_security;
562         shm_ops.more_checks = shm_more_checks;
563
564         shm_params.key = key;
565         shm_params.flg = shmflg;
566         shm_params.u.size = size;
567
568         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
569 }
570
571 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
572 {
573         switch(version) {
574         case IPC_64:
575                 return copy_to_user(buf, in, sizeof(*in));
576         case IPC_OLD:
577             {
578                 struct shmid_ds out;
579
580                 memset(&out, 0, sizeof(out));
581                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
582                 out.shm_segsz   = in->shm_segsz;
583                 out.shm_atime   = in->shm_atime;
584                 out.shm_dtime   = in->shm_dtime;
585                 out.shm_ctime   = in->shm_ctime;
586                 out.shm_cpid    = in->shm_cpid;
587                 out.shm_lpid    = in->shm_lpid;
588                 out.shm_nattch  = in->shm_nattch;
589
590                 return copy_to_user(buf, &out, sizeof(out));
591             }
592         default:
593                 return -EINVAL;
594         }
595 }
596
597 static inline unsigned long
598 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
599 {
600         switch(version) {
601         case IPC_64:
602                 if (copy_from_user(out, buf, sizeof(*out)))
603                         return -EFAULT;
604                 return 0;
605         case IPC_OLD:
606             {
607                 struct shmid_ds tbuf_old;
608
609                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
610                         return -EFAULT;
611
612                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
613                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
614                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
615
616                 return 0;
617             }
618         default:
619                 return -EINVAL;
620         }
621 }
622
623 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
624 {
625         switch(version) {
626         case IPC_64:
627                 return copy_to_user(buf, in, sizeof(*in));
628         case IPC_OLD:
629             {
630                 struct shminfo out;
631
632                 if(in->shmmax > INT_MAX)
633                         out.shmmax = INT_MAX;
634                 else
635                         out.shmmax = (int)in->shmmax;
636
637                 out.shmmin      = in->shmmin;
638                 out.shmmni      = in->shmmni;
639                 out.shmseg      = in->shmseg;
640                 out.shmall      = in->shmall; 
641
642                 return copy_to_user(buf, &out, sizeof(out));
643             }
644         default:
645                 return -EINVAL;
646         }
647 }
648
649 /*
650  * Calculate and add used RSS and swap pages of a shm.
651  * Called with shm_ids.rw_mutex held as a reader
652  */
653 static void shm_add_rss_swap(struct shmid_kernel *shp,
654         unsigned long *rss_add, unsigned long *swp_add)
655 {
656         struct inode *inode;
657
658         inode = shp->shm_file->f_path.dentry->d_inode;
659
660         if (is_file_hugepages(shp->shm_file)) {
661                 struct address_space *mapping = inode->i_mapping;
662                 struct hstate *h = hstate_file(shp->shm_file);
663                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
664         } else {
665 #ifdef CONFIG_SHMEM
666                 struct shmem_inode_info *info = SHMEM_I(inode);
667                 spin_lock(&info->lock);
668                 *rss_add += inode->i_mapping->nrpages;
669                 *swp_add += info->swapped;
670                 spin_unlock(&info->lock);
671 #else
672                 *rss_add += inode->i_mapping->nrpages;
673 #endif
674         }
675 }
676
677 /*
678  * Called with shm_ids.rw_mutex held as a reader
679  */
680 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
681                 unsigned long *swp)
682 {
683         int next_id;
684         int total, in_use;
685
686         *rss = 0;
687         *swp = 0;
688
689         in_use = shm_ids(ns).in_use;
690
691         for (total = 0, next_id = 0; total < in_use; next_id++) {
692                 struct kern_ipc_perm *ipc;
693                 struct shmid_kernel *shp;
694
695                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
696                 if (ipc == NULL)
697                         continue;
698                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
699
700                 shm_add_rss_swap(shp, rss, swp);
701
702                 total++;
703         }
704 }
705
706 /*
707  * This function handles some shmctl commands which require the rw_mutex
708  * to be held in write mode.
709  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
710  */
711 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
712                        struct shmid_ds __user *buf, int version)
713 {
714         struct kern_ipc_perm *ipcp;
715         struct shmid64_ds shmid64;
716         struct shmid_kernel *shp;
717         int err;
718
719         if (cmd == IPC_SET) {
720                 if (copy_shmid_from_user(&shmid64, buf, version))
721                         return -EFAULT;
722         }
723
724         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
725                                &shmid64.shm_perm, 0);
726         if (IS_ERR(ipcp))
727                 return PTR_ERR(ipcp);
728
729         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
730
731         err = security_shm_shmctl(shp, cmd);
732         if (err)
733                 goto out_unlock;
734         switch (cmd) {
735         case IPC_RMID:
736                 do_shm_rmid(ns, ipcp);
737                 goto out_up;
738         case IPC_SET:
739                 ipc_update_perm(&shmid64.shm_perm, ipcp);
740                 shp->shm_ctim = get_seconds();
741                 break;
742         default:
743                 err = -EINVAL;
744         }
745 out_unlock:
746         shm_unlock(shp);
747 out_up:
748         up_write(&shm_ids(ns).rw_mutex);
749         return err;
750 }
751
752 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
753 {
754         struct shmid_kernel *shp;
755         int err, version;
756         struct ipc_namespace *ns;
757
758         if (cmd < 0 || shmid < 0) {
759                 err = -EINVAL;
760                 goto out;
761         }
762
763         version = ipc_parse_version(&cmd);
764         ns = current->nsproxy->ipc_ns;
765
766         switch (cmd) { /* replace with proc interface ? */
767         case IPC_INFO:
768         {
769                 struct shminfo64 shminfo;
770
771                 err = security_shm_shmctl(NULL, cmd);
772                 if (err)
773                         return err;
774
775                 memset(&shminfo, 0, sizeof(shminfo));
776                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
777                 shminfo.shmmax = ns->shm_ctlmax;
778                 shminfo.shmall = ns->shm_ctlall;
779
780                 shminfo.shmmin = SHMMIN;
781                 if(copy_shminfo_to_user (buf, &shminfo, version))
782                         return -EFAULT;
783
784                 down_read(&shm_ids(ns).rw_mutex);
785                 err = ipc_get_maxid(&shm_ids(ns));
786                 up_read(&shm_ids(ns).rw_mutex);
787
788                 if(err<0)
789                         err = 0;
790                 goto out;
791         }
792         case SHM_INFO:
793         {
794                 struct shm_info shm_info;
795
796                 err = security_shm_shmctl(NULL, cmd);
797                 if (err)
798                         return err;
799
800                 memset(&shm_info, 0, sizeof(shm_info));
801                 down_read(&shm_ids(ns).rw_mutex);
802                 shm_info.used_ids = shm_ids(ns).in_use;
803                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
804                 shm_info.shm_tot = ns->shm_tot;
805                 shm_info.swap_attempts = 0;
806                 shm_info.swap_successes = 0;
807                 err = ipc_get_maxid(&shm_ids(ns));
808                 up_read(&shm_ids(ns).rw_mutex);
809                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
810                         err = -EFAULT;
811                         goto out;
812                 }
813
814                 err = err < 0 ? 0 : err;
815                 goto out;
816         }
817         case SHM_STAT:
818         case IPC_STAT:
819         {
820                 struct shmid64_ds tbuf;
821                 int result;
822
823                 if (cmd == SHM_STAT) {
824                         shp = shm_lock(ns, shmid);
825                         if (IS_ERR(shp)) {
826                                 err = PTR_ERR(shp);
827                                 goto out;
828                         }
829                         result = shp->shm_perm.id;
830                 } else {
831                         shp = shm_lock_check(ns, shmid);
832                         if (IS_ERR(shp)) {
833                                 err = PTR_ERR(shp);
834                                 goto out;
835                         }
836                         result = 0;
837                 }
838                 err = -EACCES;
839                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
840                         goto out_unlock;
841                 err = security_shm_shmctl(shp, cmd);
842                 if (err)
843                         goto out_unlock;
844                 memset(&tbuf, 0, sizeof(tbuf));
845                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
846                 tbuf.shm_segsz  = shp->shm_segsz;
847                 tbuf.shm_atime  = shp->shm_atim;
848                 tbuf.shm_dtime  = shp->shm_dtim;
849                 tbuf.shm_ctime  = shp->shm_ctim;
850                 tbuf.shm_cpid   = shp->shm_cprid;
851                 tbuf.shm_lpid   = shp->shm_lprid;
852                 tbuf.shm_nattch = shp->shm_nattch;
853                 shm_unlock(shp);
854                 if(copy_shmid_to_user (buf, &tbuf, version))
855                         err = -EFAULT;
856                 else
857                         err = result;
858                 goto out;
859         }
860         case SHM_LOCK:
861         case SHM_UNLOCK:
862         {
863                 struct file *uninitialized_var(shm_file);
864
865                 lru_add_drain_all();  /* drain pagevecs to lru lists */
866
867                 shp = shm_lock_check(ns, shmid);
868                 if (IS_ERR(shp)) {
869                         err = PTR_ERR(shp);
870                         goto out;
871                 }
872
873                 audit_ipc_obj(&(shp->shm_perm));
874
875                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
876                         uid_t euid = current_euid();
877                         err = -EPERM;
878                         if (euid != shp->shm_perm.uid &&
879                             euid != shp->shm_perm.cuid)
880                                 goto out_unlock;
881                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
882                                 goto out_unlock;
883                 }
884
885                 err = security_shm_shmctl(shp, cmd);
886                 if (err)
887                         goto out_unlock;
888                 
889                 if(cmd==SHM_LOCK) {
890                         struct user_struct *user = current_user();
891                         if (!is_file_hugepages(shp->shm_file)) {
892                                 err = shmem_lock(shp->shm_file, 1, user);
893                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
894                                         shp->shm_perm.mode |= SHM_LOCKED;
895                                         shp->mlock_user = user;
896                                 }
897                         }
898                 } else if (!is_file_hugepages(shp->shm_file)) {
899                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
900                         shp->shm_perm.mode &= ~SHM_LOCKED;
901                         shp->mlock_user = NULL;
902                 }
903                 shm_unlock(shp);
904                 goto out;
905         }
906         case IPC_RMID:
907         case IPC_SET:
908                 err = shmctl_down(ns, shmid, cmd, buf, version);
909                 return err;
910         default:
911                 return -EINVAL;
912         }
913
914 out_unlock:
915         shm_unlock(shp);
916 out:
917         return err;
918 }
919
920 /*
921  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
922  *
923  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
924  * "raddr" thing points to kernel space, and there has to be a wrapper around
925  * this.
926  */
927 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
928 {
929         struct shmid_kernel *shp;
930         unsigned long addr;
931         unsigned long size;
932         struct file * file;
933         int    err;
934         unsigned long flags;
935         unsigned long prot;
936         int acc_mode;
937         unsigned long user_addr;
938         struct ipc_namespace *ns;
939         struct shm_file_data *sfd;
940         struct path path;
941         fmode_t f_mode;
942
943         err = -EINVAL;
944         if (shmid < 0)
945                 goto out;
946         else if ((addr = (ulong)shmaddr)) {
947                 if (addr & (SHMLBA-1)) {
948                         if (shmflg & SHM_RND)
949                                 addr &= ~(SHMLBA-1);       /* round down */
950                         else
951 #ifndef __ARCH_FORCE_SHMLBA
952                                 if (addr & ~PAGE_MASK)
953 #endif
954                                         goto out;
955                 }
956                 flags = MAP_SHARED | MAP_FIXED;
957         } else {
958                 if ((shmflg & SHM_REMAP))
959                         goto out;
960
961                 flags = MAP_SHARED;
962         }
963
964         if (shmflg & SHM_RDONLY) {
965                 prot = PROT_READ;
966                 acc_mode = S_IRUGO;
967                 f_mode = FMODE_READ;
968         } else {
969                 prot = PROT_READ | PROT_WRITE;
970                 acc_mode = S_IRUGO | S_IWUGO;
971                 f_mode = FMODE_READ | FMODE_WRITE;
972         }
973         if (shmflg & SHM_EXEC) {
974                 prot |= PROT_EXEC;
975                 acc_mode |= S_IXUGO;
976         }
977
978         /*
979          * We cannot rely on the fs check since SYSV IPC does have an
980          * additional creator id...
981          */
982         ns = current->nsproxy->ipc_ns;
983         shp = shm_lock_check(ns, shmid);
984         if (IS_ERR(shp)) {
985                 err = PTR_ERR(shp);
986                 goto out;
987         }
988
989         err = -EACCES;
990         if (ipcperms(ns, &shp->shm_perm, acc_mode))
991                 goto out_unlock;
992
993         err = security_shm_shmat(shp, shmaddr, shmflg);
994         if (err)
995                 goto out_unlock;
996
997         path = shp->shm_file->f_path;
998         path_get(&path);
999         shp->shm_nattch++;
1000         size = i_size_read(path.dentry->d_inode);
1001         shm_unlock(shp);
1002
1003         err = -ENOMEM;
1004         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1005         if (!sfd)
1006                 goto out_put_dentry;
1007
1008         file = alloc_file(&path, f_mode,
1009                           is_file_hugepages(shp->shm_file) ?
1010                                 &shm_file_operations_huge :
1011                                 &shm_file_operations);
1012         if (!file)
1013                 goto out_free;
1014
1015         file->private_data = sfd;
1016         file->f_mapping = shp->shm_file->f_mapping;
1017         sfd->id = shp->shm_perm.id;
1018         sfd->ns = get_ipc_ns(ns);
1019         sfd->file = shp->shm_file;
1020         sfd->vm_ops = NULL;
1021
1022         down_write(&current->mm->mmap_sem);
1023         if (addr && !(shmflg & SHM_REMAP)) {
1024                 err = -EINVAL;
1025                 if (find_vma_intersection(current->mm, addr, addr + size))
1026                         goto invalid;
1027                 /*
1028                  * If shm segment goes below stack, make sure there is some
1029                  * space left for the stack to grow (at least 4 pages).
1030                  */
1031                 if (addr < current->mm->start_stack &&
1032                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1033                         goto invalid;
1034         }
1035                 
1036         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1037         *raddr = user_addr;
1038         err = 0;
1039         if (IS_ERR_VALUE(user_addr))
1040                 err = (long)user_addr;
1041 invalid:
1042         up_write(&current->mm->mmap_sem);
1043
1044         fput(file);
1045
1046 out_nattch:
1047         down_write(&shm_ids(ns).rw_mutex);
1048         shp = shm_lock(ns, shmid);
1049         BUG_ON(IS_ERR(shp));
1050         shp->shm_nattch--;
1051         if (shm_may_destroy(ns, shp))
1052                 shm_destroy(ns, shp);
1053         else
1054                 shm_unlock(shp);
1055         up_write(&shm_ids(ns).rw_mutex);
1056
1057 out:
1058         return err;
1059
1060 out_unlock:
1061         shm_unlock(shp);
1062         goto out;
1063
1064 out_free:
1065         kfree(sfd);
1066 out_put_dentry:
1067         path_put(&path);
1068         goto out_nattch;
1069 }
1070
1071 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1072 {
1073         unsigned long ret;
1074         long err;
1075
1076         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1077         if (err)
1078                 return err;
1079         force_successful_syscall_return();
1080         return (long)ret;
1081 }
1082
1083 /*
1084  * detach and kill segment if marked destroyed.
1085  * The work is done in shm_close.
1086  */
1087 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1088 {
1089         struct mm_struct *mm = current->mm;
1090         struct vm_area_struct *vma;
1091         unsigned long addr = (unsigned long)shmaddr;
1092         int retval = -EINVAL;
1093 #ifdef CONFIG_MMU
1094         loff_t size = 0;
1095         struct vm_area_struct *next;
1096 #endif
1097
1098         if (addr & ~PAGE_MASK)
1099                 return retval;
1100
1101         down_write(&mm->mmap_sem);
1102
1103         /*
1104          * This function tries to be smart and unmap shm segments that
1105          * were modified by partial mlock or munmap calls:
1106          * - It first determines the size of the shm segment that should be
1107          *   unmapped: It searches for a vma that is backed by shm and that
1108          *   started at address shmaddr. It records it's size and then unmaps
1109          *   it.
1110          * - Then it unmaps all shm vmas that started at shmaddr and that
1111          *   are within the initially determined size.
1112          * Errors from do_munmap are ignored: the function only fails if
1113          * it's called with invalid parameters or if it's called to unmap
1114          * a part of a vma. Both calls in this function are for full vmas,
1115          * the parameters are directly copied from the vma itself and always
1116          * valid - therefore do_munmap cannot fail. (famous last words?)
1117          */
1118         /*
1119          * If it had been mremap()'d, the starting address would not
1120          * match the usual checks anyway. So assume all vma's are
1121          * above the starting address given.
1122          */
1123         vma = find_vma(mm, addr);
1124
1125 #ifdef CONFIG_MMU
1126         while (vma) {
1127                 next = vma->vm_next;
1128
1129                 /*
1130                  * Check if the starting address would match, i.e. it's
1131                  * a fragment created by mprotect() and/or munmap(), or it
1132                  * otherwise it starts at this address with no hassles.
1133                  */
1134                 if ((vma->vm_ops == &shm_vm_ops) &&
1135                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1136
1137
1138                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1139                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1140                         /*
1141                          * We discovered the size of the shm segment, so
1142                          * break out of here and fall through to the next
1143                          * loop that uses the size information to stop
1144                          * searching for matching vma's.
1145                          */
1146                         retval = 0;
1147                         vma = next;
1148                         break;
1149                 }
1150                 vma = next;
1151         }
1152
1153         /*
1154          * We need look no further than the maximum address a fragment
1155          * could possibly have landed at. Also cast things to loff_t to
1156          * prevent overflows and make comparisons vs. equal-width types.
1157          */
1158         size = PAGE_ALIGN(size);
1159         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1160                 next = vma->vm_next;
1161
1162                 /* finding a matching vma now does not alter retval */
1163                 if ((vma->vm_ops == &shm_vm_ops) &&
1164                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1165
1166                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1167                 vma = next;
1168         }
1169
1170 #else /* CONFIG_MMU */
1171         /* under NOMMU conditions, the exact address to be destroyed must be
1172          * given */
1173         retval = -EINVAL;
1174         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1175                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1176                 retval = 0;
1177         }
1178
1179 #endif
1180
1181         up_write(&mm->mmap_sem);
1182         return retval;
1183 }
1184
1185 #ifdef CONFIG_PROC_FS
1186 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1187 {
1188         struct shmid_kernel *shp = it;
1189         unsigned long rss = 0, swp = 0;
1190
1191         shm_add_rss_swap(shp, &rss, &swp);
1192
1193 #if BITS_PER_LONG <= 32
1194 #define SIZE_SPEC "%10lu"
1195 #else
1196 #define SIZE_SPEC "%21lu"
1197 #endif
1198
1199         return seq_printf(s,
1200                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1201                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1202                           SIZE_SPEC " " SIZE_SPEC "\n",
1203                           shp->shm_perm.key,
1204                           shp->shm_perm.id,
1205                           shp->shm_perm.mode,
1206                           shp->shm_segsz,
1207                           shp->shm_cprid,
1208                           shp->shm_lprid,
1209                           shp->shm_nattch,
1210                           shp->shm_perm.uid,
1211                           shp->shm_perm.gid,
1212                           shp->shm_perm.cuid,
1213                           shp->shm_perm.cgid,
1214                           shp->shm_atim,
1215                           shp->shm_dtim,
1216                           shp->shm_ctim,
1217                           rss * PAGE_SIZE,
1218                           swp * PAGE_SIZE);
1219 }
1220 #endif