ipc: sysv shared memory limited to 8TiB
[opensuse:kernel.git] / ipc / shm.c
1 /*
2  * linux/ipc/shm.c
3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
4  *       Many improvements/fixes by Bruno Haible.
5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7  *
8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15  *
16  * support for audit of ipc object properties and permission changes
17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18  *
19  * namespaces support
20  * OpenVZ, SWsoft Inc.
21  * Pavel Emelianov <xemul@openvz.org>
22  */
23
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42
43 #include <asm/uaccess.h>
44
45 #include "util.h"
46
47 struct shm_file_data {
48         int id;
49         struct ipc_namespace *ns;
50         struct file *file;
51         const struct vm_operations_struct *vm_ops;
52 };
53
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
56 static const struct file_operations shm_file_operations;
57 static const struct vm_operations_struct shm_vm_ops;
58
59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
60
61 #define shm_unlock(shp)                 \
62         ipc_unlock(&(shp)->shm_perm)
63
64 static int newseg(struct ipc_namespace *, struct ipc_params *);
65 static void shm_open(struct vm_area_struct *vma);
66 static void shm_close(struct vm_area_struct *vma);
67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
68 #ifdef CONFIG_PROC_FS
69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
70 #endif
71
72 void shm_init_ns(struct ipc_namespace *ns)
73 {
74         ns->shm_ctlmax = SHMMAX;
75         ns->shm_ctlall = SHMALL;
76         ns->shm_ctlmni = SHMMNI;
77         ns->shm_tot = 0;
78         ipc_init_ids(&shm_ids(ns));
79 }
80
81 /*
82  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
83  * Only shm_ids.rw_mutex remains locked on exit.
84  */
85 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
86 {
87         struct shmid_kernel *shp;
88         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
89
90         if (shp->shm_nattch){
91                 shp->shm_perm.mode |= SHM_DEST;
92                 /* Do not find it any more */
93                 shp->shm_perm.key = IPC_PRIVATE;
94                 shm_unlock(shp);
95         } else
96                 shm_destroy(ns, shp);
97 }
98
99 #ifdef CONFIG_IPC_NS
100 void shm_exit_ns(struct ipc_namespace *ns)
101 {
102         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
103         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
104 }
105 #endif
106
107 void __init shm_init (void)
108 {
109         shm_init_ns(&init_ipc_ns);
110         ipc_init_proc_interface("sysvipc/shm",
111 #if BITS_PER_LONG <= 32
112                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
113 #else
114                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
115 #endif
116                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
117 }
118
119 /*
120  * shm_lock_(check_) routines are called in the paths where the rw_mutex
121  * is not necessarily held.
122  */
123 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
124 {
125         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
126
127         if (IS_ERR(ipcp))
128                 return (struct shmid_kernel *)ipcp;
129
130         return container_of(ipcp, struct shmid_kernel, shm_perm);
131 }
132
133 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
134                                                 int id)
135 {
136         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
137
138         if (IS_ERR(ipcp))
139                 return (struct shmid_kernel *)ipcp;
140
141         return container_of(ipcp, struct shmid_kernel, shm_perm);
142 }
143
144 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
145 {
146         ipc_rmid(&shm_ids(ns), &s->shm_perm);
147 }
148
149
150 /* This is called by fork, once for every shm attach. */
151 static void shm_open(struct vm_area_struct *vma)
152 {
153         struct file *file = vma->vm_file;
154         struct shm_file_data *sfd = shm_file_data(file);
155         struct shmid_kernel *shp;
156
157         shp = shm_lock(sfd->ns, sfd->id);
158         BUG_ON(IS_ERR(shp));
159         shp->shm_atim = get_seconds();
160         shp->shm_lprid = task_tgid_vnr(current);
161         shp->shm_nattch++;
162         shm_unlock(shp);
163 }
164
165 /*
166  * shm_destroy - free the struct shmid_kernel
167  *
168  * @ns: namespace
169  * @shp: struct to free
170  *
171  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
172  * but returns with shp unlocked and freed.
173  */
174 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
175 {
176         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
177         shm_rmid(ns, shp);
178         shm_unlock(shp);
179         if (!is_file_hugepages(shp->shm_file))
180                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
181         else if (shp->mlock_user)
182                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
183                                                 shp->mlock_user);
184         fput (shp->shm_file);
185         security_shm_free(shp);
186         ipc_rcu_putref(shp);
187 }
188
189 /*
190  * remove the attach descriptor vma.
191  * free memory for segment if it is marked destroyed.
192  * The descriptor has already been removed from the current->mm->mmap list
193  * and will later be kfree()d.
194  */
195 static void shm_close(struct vm_area_struct *vma)
196 {
197         struct file * file = vma->vm_file;
198         struct shm_file_data *sfd = shm_file_data(file);
199         struct shmid_kernel *shp;
200         struct ipc_namespace *ns = sfd->ns;
201
202         down_write(&shm_ids(ns).rw_mutex);
203         /* remove from the list of attaches of the shm segment */
204         shp = shm_lock(ns, sfd->id);
205         BUG_ON(IS_ERR(shp));
206         shp->shm_lprid = task_tgid_vnr(current);
207         shp->shm_dtim = get_seconds();
208         shp->shm_nattch--;
209         if(shp->shm_nattch == 0 &&
210            shp->shm_perm.mode & SHM_DEST)
211                 shm_destroy(ns, shp);
212         else
213                 shm_unlock(shp);
214         up_write(&shm_ids(ns).rw_mutex);
215 }
216
217 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
218 {
219         struct file *file = vma->vm_file;
220         struct shm_file_data *sfd = shm_file_data(file);
221
222         return sfd->vm_ops->fault(vma, vmf);
223 }
224
225 #ifdef CONFIG_NUMA
226 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
227 {
228         struct file *file = vma->vm_file;
229         struct shm_file_data *sfd = shm_file_data(file);
230         int err = 0;
231         if (sfd->vm_ops->set_policy)
232                 err = sfd->vm_ops->set_policy(vma, new);
233         return err;
234 }
235
236 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
237                                         unsigned long addr)
238 {
239         struct file *file = vma->vm_file;
240         struct shm_file_data *sfd = shm_file_data(file);
241         struct mempolicy *pol = NULL;
242
243         if (sfd->vm_ops->get_policy)
244                 pol = sfd->vm_ops->get_policy(vma, addr);
245         else if (vma->vm_policy)
246                 pol = vma->vm_policy;
247
248         return pol;
249 }
250 #endif
251
252 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
253 {
254         struct shm_file_data *sfd = shm_file_data(file);
255         int ret;
256
257         ret = sfd->file->f_op->mmap(sfd->file, vma);
258         if (ret != 0)
259                 return ret;
260         sfd->vm_ops = vma->vm_ops;
261 #ifdef CONFIG_MMU
262         BUG_ON(!sfd->vm_ops->fault);
263 #endif
264         vma->vm_ops = &shm_vm_ops;
265         shm_open(vma);
266
267         return ret;
268 }
269
270 static int shm_release(struct inode *ino, struct file *file)
271 {
272         struct shm_file_data *sfd = shm_file_data(file);
273
274         put_ipc_ns(sfd->ns);
275         shm_file_data(file) = NULL;
276         kfree(sfd);
277         return 0;
278 }
279
280 static int shm_fsync(struct file *file, int datasync)
281 {
282         struct shm_file_data *sfd = shm_file_data(file);
283
284         if (!sfd->file->f_op->fsync)
285                 return -EINVAL;
286         return sfd->file->f_op->fsync(sfd->file, datasync);
287 }
288
289 static unsigned long shm_get_unmapped_area(struct file *file,
290         unsigned long addr, unsigned long len, unsigned long pgoff,
291         unsigned long flags)
292 {
293         struct shm_file_data *sfd = shm_file_data(file);
294         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
295                                                 pgoff, flags);
296 }
297
298 static const struct file_operations shm_file_operations = {
299         .mmap           = shm_mmap,
300         .fsync          = shm_fsync,
301         .release        = shm_release,
302 #ifndef CONFIG_MMU
303         .get_unmapped_area      = shm_get_unmapped_area,
304 #endif
305         .llseek         = noop_llseek,
306 };
307
308 static const struct file_operations shm_file_operations_huge = {
309         .mmap           = shm_mmap,
310         .fsync          = shm_fsync,
311         .release        = shm_release,
312         .get_unmapped_area      = shm_get_unmapped_area,
313         .llseek         = noop_llseek,
314 };
315
316 int is_file_shm_hugepages(struct file *file)
317 {
318         return file->f_op == &shm_file_operations_huge;
319 }
320
321 static const struct vm_operations_struct shm_vm_ops = {
322         .open   = shm_open,     /* callback for a new vm-area open */
323         .close  = shm_close,    /* callback for when the vm-area is released */
324         .fault  = shm_fault,
325 #if defined(CONFIG_NUMA)
326         .set_policy = shm_set_policy,
327         .get_policy = shm_get_policy,
328 #endif
329 };
330
331 /**
332  * newseg - Create a new shared memory segment
333  * @ns: namespace
334  * @params: ptr to the structure that contains key, size and shmflg
335  *
336  * Called with shm_ids.rw_mutex held as a writer.
337  */
338
339 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
340 {
341         key_t key = params->key;
342         int shmflg = params->flg;
343         size_t size = params->u.size;
344         int error;
345         struct shmid_kernel *shp;
346         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
347         struct file * file;
348         char name[13];
349         int id;
350         vm_flags_t acctflag = 0;
351
352         if (size < SHMMIN || size > ns->shm_ctlmax)
353                 return -EINVAL;
354
355         if (ns->shm_tot + numpages > ns->shm_ctlall)
356                 return -ENOSPC;
357
358         shp = ipc_rcu_alloc(sizeof(*shp));
359         if (!shp)
360                 return -ENOMEM;
361
362         shp->shm_perm.key = key;
363         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
364         shp->mlock_user = NULL;
365
366         shp->shm_perm.security = NULL;
367         error = security_shm_alloc(shp);
368         if (error) {
369                 ipc_rcu_putref(shp);
370                 return error;
371         }
372
373         sprintf (name, "SYSV%08x", key);
374         if (shmflg & SHM_HUGETLB) {
375                 /* hugetlb_file_setup applies strict accounting */
376                 if (shmflg & SHM_NORESERVE)
377                         acctflag = VM_NORESERVE;
378                 file = hugetlb_file_setup(name, size, acctflag,
379                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
380         } else {
381                 /*
382                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
383                  * if it's asked for.
384                  */
385                 if  ((shmflg & SHM_NORESERVE) &&
386                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
387                         acctflag = VM_NORESERVE;
388                 file = shmem_file_setup(name, size, acctflag);
389         }
390         error = PTR_ERR(file);
391         if (IS_ERR(file))
392                 goto no_file;
393
394         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
395         if (id < 0) {
396                 error = id;
397                 goto no_id;
398         }
399
400         shp->shm_cprid = task_tgid_vnr(current);
401         shp->shm_lprid = 0;
402         shp->shm_atim = shp->shm_dtim = 0;
403         shp->shm_ctim = get_seconds();
404         shp->shm_segsz = size;
405         shp->shm_nattch = 0;
406         shp->shm_file = file;
407         /*
408          * shmid gets reported as "inode#" in /proc/pid/maps.
409          * proc-ps tools use this. Changing this will break them.
410          */
411         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
412
413         ns->shm_tot += numpages;
414         error = shp->shm_perm.id;
415         shm_unlock(shp);
416         return error;
417
418 no_id:
419         if (is_file_hugepages(file) && shp->mlock_user)
420                 user_shm_unlock(size, shp->mlock_user);
421         fput(file);
422 no_file:
423         security_shm_free(shp);
424         ipc_rcu_putref(shp);
425         return error;
426 }
427
428 /*
429  * Called with shm_ids.rw_mutex and ipcp locked.
430  */
431 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
432 {
433         struct shmid_kernel *shp;
434
435         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
436         return security_shm_associate(shp, shmflg);
437 }
438
439 /*
440  * Called with shm_ids.rw_mutex and ipcp locked.
441  */
442 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
443                                 struct ipc_params *params)
444 {
445         struct shmid_kernel *shp;
446
447         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
448         if (shp->shm_segsz < params->u.size)
449                 return -EINVAL;
450
451         return 0;
452 }
453
454 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
455 {
456         struct ipc_namespace *ns;
457         struct ipc_ops shm_ops;
458         struct ipc_params shm_params;
459
460         ns = current->nsproxy->ipc_ns;
461
462         shm_ops.getnew = newseg;
463         shm_ops.associate = shm_security;
464         shm_ops.more_checks = shm_more_checks;
465
466         shm_params.key = key;
467         shm_params.flg = shmflg;
468         shm_params.u.size = size;
469
470         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
471 }
472
473 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
474 {
475         switch(version) {
476         case IPC_64:
477                 return copy_to_user(buf, in, sizeof(*in));
478         case IPC_OLD:
479             {
480                 struct shmid_ds out;
481
482                 memset(&out, 0, sizeof(out));
483                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
484                 out.shm_segsz   = in->shm_segsz;
485                 out.shm_atime   = in->shm_atime;
486                 out.shm_dtime   = in->shm_dtime;
487                 out.shm_ctime   = in->shm_ctime;
488                 out.shm_cpid    = in->shm_cpid;
489                 out.shm_lpid    = in->shm_lpid;
490                 out.shm_nattch  = in->shm_nattch;
491
492                 return copy_to_user(buf, &out, sizeof(out));
493             }
494         default:
495                 return -EINVAL;
496         }
497 }
498
499 static inline unsigned long
500 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
501 {
502         switch(version) {
503         case IPC_64:
504                 if (copy_from_user(out, buf, sizeof(*out)))
505                         return -EFAULT;
506                 return 0;
507         case IPC_OLD:
508             {
509                 struct shmid_ds tbuf_old;
510
511                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
512                         return -EFAULT;
513
514                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
515                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
516                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
517
518                 return 0;
519             }
520         default:
521                 return -EINVAL;
522         }
523 }
524
525 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
526 {
527         switch(version) {
528         case IPC_64:
529                 return copy_to_user(buf, in, sizeof(*in));
530         case IPC_OLD:
531             {
532                 struct shminfo out;
533
534                 if(in->shmmax > INT_MAX)
535                         out.shmmax = INT_MAX;
536                 else
537                         out.shmmax = (int)in->shmmax;
538
539                 out.shmmin      = in->shmmin;
540                 out.shmmni      = in->shmmni;
541                 out.shmseg      = in->shmseg;
542                 out.shmall      = in->shmall; 
543
544                 return copy_to_user(buf, &out, sizeof(out));
545             }
546         default:
547                 return -EINVAL;
548         }
549 }
550
551 /*
552  * Calculate and add used RSS and swap pages of a shm.
553  * Called with shm_ids.rw_mutex held as a reader
554  */
555 static void shm_add_rss_swap(struct shmid_kernel *shp,
556         unsigned long *rss_add, unsigned long *swp_add)
557 {
558         struct inode *inode;
559
560         inode = shp->shm_file->f_path.dentry->d_inode;
561
562         if (is_file_hugepages(shp->shm_file)) {
563                 struct address_space *mapping = inode->i_mapping;
564                 struct hstate *h = hstate_file(shp->shm_file);
565                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
566         } else {
567 #ifdef CONFIG_SHMEM
568                 struct shmem_inode_info *info = SHMEM_I(inode);
569                 spin_lock(&info->lock);
570                 *rss_add += inode->i_mapping->nrpages;
571                 *swp_add += info->swapped;
572                 spin_unlock(&info->lock);
573 #else
574                 *rss_add += inode->i_mapping->nrpages;
575 #endif
576         }
577 }
578
579 /*
580  * Called with shm_ids.rw_mutex held as a reader
581  */
582 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
583                 unsigned long *swp)
584 {
585         int next_id;
586         int total, in_use;
587
588         *rss = 0;
589         *swp = 0;
590
591         in_use = shm_ids(ns).in_use;
592
593         for (total = 0, next_id = 0; total < in_use; next_id++) {
594                 struct kern_ipc_perm *ipc;
595                 struct shmid_kernel *shp;
596
597                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
598                 if (ipc == NULL)
599                         continue;
600                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
601
602                 shm_add_rss_swap(shp, rss, swp);
603
604                 total++;
605         }
606 }
607
608 /*
609  * This function handles some shmctl commands which require the rw_mutex
610  * to be held in write mode.
611  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
612  */
613 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
614                        struct shmid_ds __user *buf, int version)
615 {
616         struct kern_ipc_perm *ipcp;
617         struct shmid64_ds shmid64;
618         struct shmid_kernel *shp;
619         int err;
620
621         if (cmd == IPC_SET) {
622                 if (copy_shmid_from_user(&shmid64, buf, version))
623                         return -EFAULT;
624         }
625
626         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
627                                &shmid64.shm_perm, 0);
628         if (IS_ERR(ipcp))
629                 return PTR_ERR(ipcp);
630
631         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
632
633         err = security_shm_shmctl(shp, cmd);
634         if (err)
635                 goto out_unlock;
636         switch (cmd) {
637         case IPC_RMID:
638                 do_shm_rmid(ns, ipcp);
639                 goto out_up;
640         case IPC_SET:
641                 ipc_update_perm(&shmid64.shm_perm, ipcp);
642                 shp->shm_ctim = get_seconds();
643                 break;
644         default:
645                 err = -EINVAL;
646         }
647 out_unlock:
648         shm_unlock(shp);
649 out_up:
650         up_write(&shm_ids(ns).rw_mutex);
651         return err;
652 }
653
654 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
655 {
656         struct shmid_kernel *shp;
657         int err, version;
658         struct ipc_namespace *ns;
659
660         if (cmd < 0 || shmid < 0) {
661                 err = -EINVAL;
662                 goto out;
663         }
664
665         version = ipc_parse_version(&cmd);
666         ns = current->nsproxy->ipc_ns;
667
668         switch (cmd) { /* replace with proc interface ? */
669         case IPC_INFO:
670         {
671                 struct shminfo64 shminfo;
672
673                 err = security_shm_shmctl(NULL, cmd);
674                 if (err)
675                         return err;
676
677                 memset(&shminfo, 0, sizeof(shminfo));
678                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
679                 shminfo.shmmax = ns->shm_ctlmax;
680                 shminfo.shmall = ns->shm_ctlall;
681
682                 shminfo.shmmin = SHMMIN;
683                 if(copy_shminfo_to_user (buf, &shminfo, version))
684                         return -EFAULT;
685
686                 down_read(&shm_ids(ns).rw_mutex);
687                 err = ipc_get_maxid(&shm_ids(ns));
688                 up_read(&shm_ids(ns).rw_mutex);
689
690                 if(err<0)
691                         err = 0;
692                 goto out;
693         }
694         case SHM_INFO:
695         {
696                 struct shm_info shm_info;
697
698                 err = security_shm_shmctl(NULL, cmd);
699                 if (err)
700                         return err;
701
702                 memset(&shm_info, 0, sizeof(shm_info));
703                 down_read(&shm_ids(ns).rw_mutex);
704                 shm_info.used_ids = shm_ids(ns).in_use;
705                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
706                 shm_info.shm_tot = ns->shm_tot;
707                 shm_info.swap_attempts = 0;
708                 shm_info.swap_successes = 0;
709                 err = ipc_get_maxid(&shm_ids(ns));
710                 up_read(&shm_ids(ns).rw_mutex);
711                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
712                         err = -EFAULT;
713                         goto out;
714                 }
715
716                 err = err < 0 ? 0 : err;
717                 goto out;
718         }
719         case SHM_STAT:
720         case IPC_STAT:
721         {
722                 struct shmid64_ds tbuf;
723                 int result;
724
725                 if (cmd == SHM_STAT) {
726                         shp = shm_lock(ns, shmid);
727                         if (IS_ERR(shp)) {
728                                 err = PTR_ERR(shp);
729                                 goto out;
730                         }
731                         result = shp->shm_perm.id;
732                 } else {
733                         shp = shm_lock_check(ns, shmid);
734                         if (IS_ERR(shp)) {
735                                 err = PTR_ERR(shp);
736                                 goto out;
737                         }
738                         result = 0;
739                 }
740                 err = -EACCES;
741                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
742                         goto out_unlock;
743                 err = security_shm_shmctl(shp, cmd);
744                 if (err)
745                         goto out_unlock;
746                 memset(&tbuf, 0, sizeof(tbuf));
747                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
748                 tbuf.shm_segsz  = shp->shm_segsz;
749                 tbuf.shm_atime  = shp->shm_atim;
750                 tbuf.shm_dtime  = shp->shm_dtim;
751                 tbuf.shm_ctime  = shp->shm_ctim;
752                 tbuf.shm_cpid   = shp->shm_cprid;
753                 tbuf.shm_lpid   = shp->shm_lprid;
754                 tbuf.shm_nattch = shp->shm_nattch;
755                 shm_unlock(shp);
756                 if(copy_shmid_to_user (buf, &tbuf, version))
757                         err = -EFAULT;
758                 else
759                         err = result;
760                 goto out;
761         }
762         case SHM_LOCK:
763         case SHM_UNLOCK:
764         {
765                 struct file *uninitialized_var(shm_file);
766
767                 lru_add_drain_all();  /* drain pagevecs to lru lists */
768
769                 shp = shm_lock_check(ns, shmid);
770                 if (IS_ERR(shp)) {
771                         err = PTR_ERR(shp);
772                         goto out;
773                 }
774
775                 audit_ipc_obj(&(shp->shm_perm));
776
777                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
778                         uid_t euid = current_euid();
779                         err = -EPERM;
780                         if (euid != shp->shm_perm.uid &&
781                             euid != shp->shm_perm.cuid)
782                                 goto out_unlock;
783                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
784                                 goto out_unlock;
785                 }
786
787                 err = security_shm_shmctl(shp, cmd);
788                 if (err)
789                         goto out_unlock;
790                 
791                 if(cmd==SHM_LOCK) {
792                         struct user_struct *user = current_user();
793                         if (!is_file_hugepages(shp->shm_file)) {
794                                 err = shmem_lock(shp->shm_file, 1, user);
795                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
796                                         shp->shm_perm.mode |= SHM_LOCKED;
797                                         shp->mlock_user = user;
798                                 }
799                         }
800                 } else if (!is_file_hugepages(shp->shm_file)) {
801                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
802                         shp->shm_perm.mode &= ~SHM_LOCKED;
803                         shp->mlock_user = NULL;
804                 }
805                 shm_unlock(shp);
806                 goto out;
807         }
808         case IPC_RMID:
809         case IPC_SET:
810                 err = shmctl_down(ns, shmid, cmd, buf, version);
811                 return err;
812         default:
813                 return -EINVAL;
814         }
815
816 out_unlock:
817         shm_unlock(shp);
818 out:
819         return err;
820 }
821
822 /*
823  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
824  *
825  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
826  * "raddr" thing points to kernel space, and there has to be a wrapper around
827  * this.
828  */
829 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
830 {
831         struct shmid_kernel *shp;
832         unsigned long addr;
833         unsigned long size;
834         struct file * file;
835         int    err;
836         unsigned long flags;
837         unsigned long prot;
838         int acc_mode;
839         unsigned long user_addr;
840         struct ipc_namespace *ns;
841         struct shm_file_data *sfd;
842         struct path path;
843         fmode_t f_mode;
844
845         err = -EINVAL;
846         if (shmid < 0)
847                 goto out;
848         else if ((addr = (ulong)shmaddr)) {
849                 if (addr & (SHMLBA-1)) {
850                         if (shmflg & SHM_RND)
851                                 addr &= ~(SHMLBA-1);       /* round down */
852                         else
853 #ifndef __ARCH_FORCE_SHMLBA
854                                 if (addr & ~PAGE_MASK)
855 #endif
856                                         goto out;
857                 }
858                 flags = MAP_SHARED | MAP_FIXED;
859         } else {
860                 if ((shmflg & SHM_REMAP))
861                         goto out;
862
863                 flags = MAP_SHARED;
864         }
865
866         if (shmflg & SHM_RDONLY) {
867                 prot = PROT_READ;
868                 acc_mode = S_IRUGO;
869                 f_mode = FMODE_READ;
870         } else {
871                 prot = PROT_READ | PROT_WRITE;
872                 acc_mode = S_IRUGO | S_IWUGO;
873                 f_mode = FMODE_READ | FMODE_WRITE;
874         }
875         if (shmflg & SHM_EXEC) {
876                 prot |= PROT_EXEC;
877                 acc_mode |= S_IXUGO;
878         }
879
880         /*
881          * We cannot rely on the fs check since SYSV IPC does have an
882          * additional creator id...
883          */
884         ns = current->nsproxy->ipc_ns;
885         shp = shm_lock_check(ns, shmid);
886         if (IS_ERR(shp)) {
887                 err = PTR_ERR(shp);
888                 goto out;
889         }
890
891         err = -EACCES;
892         if (ipcperms(ns, &shp->shm_perm, acc_mode))
893                 goto out_unlock;
894
895         err = security_shm_shmat(shp, shmaddr, shmflg);
896         if (err)
897                 goto out_unlock;
898
899         path = shp->shm_file->f_path;
900         path_get(&path);
901         shp->shm_nattch++;
902         size = i_size_read(path.dentry->d_inode);
903         shm_unlock(shp);
904
905         err = -ENOMEM;
906         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
907         if (!sfd)
908                 goto out_put_dentry;
909
910         file = alloc_file(&path, f_mode,
911                           is_file_hugepages(shp->shm_file) ?
912                                 &shm_file_operations_huge :
913                                 &shm_file_operations);
914         if (!file)
915                 goto out_free;
916
917         file->private_data = sfd;
918         file->f_mapping = shp->shm_file->f_mapping;
919         sfd->id = shp->shm_perm.id;
920         sfd->ns = get_ipc_ns(ns);
921         sfd->file = shp->shm_file;
922         sfd->vm_ops = NULL;
923
924         down_write(&current->mm->mmap_sem);
925         if (addr && !(shmflg & SHM_REMAP)) {
926                 err = -EINVAL;
927                 if (find_vma_intersection(current->mm, addr, addr + size))
928                         goto invalid;
929                 /*
930                  * If shm segment goes below stack, make sure there is some
931                  * space left for the stack to grow (at least 4 pages).
932                  */
933                 if (addr < current->mm->start_stack &&
934                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
935                         goto invalid;
936         }
937                 
938         user_addr = do_mmap (file, addr, size, prot, flags, 0);
939         *raddr = user_addr;
940         err = 0;
941         if (IS_ERR_VALUE(user_addr))
942                 err = (long)user_addr;
943 invalid:
944         up_write(&current->mm->mmap_sem);
945
946         fput(file);
947
948 out_nattch:
949         down_write(&shm_ids(ns).rw_mutex);
950         shp = shm_lock(ns, shmid);
951         BUG_ON(IS_ERR(shp));
952         shp->shm_nattch--;
953         if(shp->shm_nattch == 0 &&
954            shp->shm_perm.mode & SHM_DEST)
955                 shm_destroy(ns, shp);
956         else
957                 shm_unlock(shp);
958         up_write(&shm_ids(ns).rw_mutex);
959
960 out:
961         return err;
962
963 out_unlock:
964         shm_unlock(shp);
965         goto out;
966
967 out_free:
968         kfree(sfd);
969 out_put_dentry:
970         path_put(&path);
971         goto out_nattch;
972 }
973
974 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
975 {
976         unsigned long ret;
977         long err;
978
979         err = do_shmat(shmid, shmaddr, shmflg, &ret);
980         if (err)
981                 return err;
982         force_successful_syscall_return();
983         return (long)ret;
984 }
985
986 /*
987  * detach and kill segment if marked destroyed.
988  * The work is done in shm_close.
989  */
990 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
991 {
992         struct mm_struct *mm = current->mm;
993         struct vm_area_struct *vma;
994         unsigned long addr = (unsigned long)shmaddr;
995         int retval = -EINVAL;
996 #ifdef CONFIG_MMU
997         loff_t size = 0;
998         struct vm_area_struct *next;
999 #endif
1000
1001         if (addr & ~PAGE_MASK)
1002                 return retval;
1003
1004         down_write(&mm->mmap_sem);
1005
1006         /*
1007          * This function tries to be smart and unmap shm segments that
1008          * were modified by partial mlock or munmap calls:
1009          * - It first determines the size of the shm segment that should be
1010          *   unmapped: It searches for a vma that is backed by shm and that
1011          *   started at address shmaddr. It records it's size and then unmaps
1012          *   it.
1013          * - Then it unmaps all shm vmas that started at shmaddr and that
1014          *   are within the initially determined size.
1015          * Errors from do_munmap are ignored: the function only fails if
1016          * it's called with invalid parameters or if it's called to unmap
1017          * a part of a vma. Both calls in this function are for full vmas,
1018          * the parameters are directly copied from the vma itself and always
1019          * valid - therefore do_munmap cannot fail. (famous last words?)
1020          */
1021         /*
1022          * If it had been mremap()'d, the starting address would not
1023          * match the usual checks anyway. So assume all vma's are
1024          * above the starting address given.
1025          */
1026         vma = find_vma(mm, addr);
1027
1028 #ifdef CONFIG_MMU
1029         while (vma) {
1030                 next = vma->vm_next;
1031
1032                 /*
1033                  * Check if the starting address would match, i.e. it's
1034                  * a fragment created by mprotect() and/or munmap(), or it
1035                  * otherwise it starts at this address with no hassles.
1036                  */
1037                 if ((vma->vm_ops == &shm_vm_ops) &&
1038                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1039
1040
1041                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1042                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1043                         /*
1044                          * We discovered the size of the shm segment, so
1045                          * break out of here and fall through to the next
1046                          * loop that uses the size information to stop
1047                          * searching for matching vma's.
1048                          */
1049                         retval = 0;
1050                         vma = next;
1051                         break;
1052                 }
1053                 vma = next;
1054         }
1055
1056         /*
1057          * We need look no further than the maximum address a fragment
1058          * could possibly have landed at. Also cast things to loff_t to
1059          * prevent overflows and make comparisons vs. equal-width types.
1060          */
1061         size = PAGE_ALIGN(size);
1062         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1063                 next = vma->vm_next;
1064
1065                 /* finding a matching vma now does not alter retval */
1066                 if ((vma->vm_ops == &shm_vm_ops) &&
1067                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1068
1069                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1070                 vma = next;
1071         }
1072
1073 #else /* CONFIG_MMU */
1074         /* under NOMMU conditions, the exact address to be destroyed must be
1075          * given */
1076         retval = -EINVAL;
1077         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1078                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1079                 retval = 0;
1080         }
1081
1082 #endif
1083
1084         up_write(&mm->mmap_sem);
1085         return retval;
1086 }
1087
1088 #ifdef CONFIG_PROC_FS
1089 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1090 {
1091         struct shmid_kernel *shp = it;
1092         unsigned long rss = 0, swp = 0;
1093
1094         shm_add_rss_swap(shp, &rss, &swp);
1095
1096 #if BITS_PER_LONG <= 32
1097 #define SIZE_SPEC "%10lu"
1098 #else
1099 #define SIZE_SPEC "%21lu"
1100 #endif
1101
1102         return seq_printf(s,
1103                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1104                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1105                           SIZE_SPEC " " SIZE_SPEC "\n",
1106                           shp->shm_perm.key,
1107                           shp->shm_perm.id,
1108                           shp->shm_perm.mode,
1109                           shp->shm_segsz,
1110                           shp->shm_cprid,
1111                           shp->shm_lprid,
1112                           shp->shm_nattch,
1113                           shp->shm_perm.uid,
1114                           shp->shm_perm.gid,
1115                           shp->shm_perm.cuid,
1116                           shp->shm_perm.cgid,
1117                           shp->shm_atim,
1118                           shp->shm_dtim,
1119                           shp->shm_ctim,
1120                           rss * PAGE_SIZE,
1121                           swp * PAGE_SIZE);
1122 }
1123 #endif