07f4b7abc80a41024774c5ccd842a264f9159d8e
[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 struct vm_operations_struct shm_vm_ops;
58
59 static struct ipc_ids init_shm_ids;
60
61 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
62
63 #define shm_unlock(shp)                 \
64         ipc_unlock(&(shp)->shm_perm)
65 #define shm_buildid(id, seq)    ipc_buildid(id, seq)
66
67 static int newseg(struct ipc_namespace *, struct ipc_params *);
68 static void shm_open(struct vm_area_struct *vma);
69 static void shm_close(struct vm_area_struct *vma);
70 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
71 #ifdef CONFIG_PROC_FS
72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
73 #endif
74
75 static void __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
76 {
77         ns->ids[IPC_SHM_IDS] = ids;
78         ns->shm_ctlmax = SHMMAX;
79         ns->shm_ctlall = SHMALL;
80         ns->shm_ctlmni = SHMMNI;
81         ns->shm_tot = 0;
82         ipc_init_ids(ids);
83 }
84
85 /*
86  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
87  * Only shm_ids.rw_mutex remains locked on exit.
88  */
89 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
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 int shm_init_ns(struct ipc_namespace *ns)
102 {
103         struct ipc_ids *ids;
104
105         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
106         if (ids == NULL)
107                 return -ENOMEM;
108
109         __shm_init_ns(ns, ids);
110         return 0;
111 }
112
113 void shm_exit_ns(struct ipc_namespace *ns)
114 {
115         struct shmid_kernel *shp;
116         struct kern_ipc_perm *perm;
117         int next_id;
118         int total, in_use;
119
120         down_write(&shm_ids(ns).rw_mutex);
121
122         in_use = shm_ids(ns).in_use;
123
124         for (total = 0, next_id = 0; total < in_use; next_id++) {
125                 perm = idr_find(&shm_ids(ns).ipcs_idr, next_id);
126                 if (perm == NULL)
127                         continue;
128                 ipc_lock_by_ptr(perm);
129                 shp = container_of(perm, struct shmid_kernel, shm_perm);
130                 do_shm_rmid(ns, shp);
131                 total++;
132         }
133         up_write(&shm_ids(ns).rw_mutex);
134
135         kfree(ns->ids[IPC_SHM_IDS]);
136         ns->ids[IPC_SHM_IDS] = NULL;
137 }
138 #endif
139
140 void __init shm_init (void)
141 {
142         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
143         ipc_init_proc_interface("sysvipc/shm",
144                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
145                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
146 }
147
148 /*
149  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
150  * is held to protect access to the idr tree.
151  */
152 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
153                                                 int id)
154 {
155         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
156
157         if (IS_ERR(ipcp))
158                 return (struct shmid_kernel *)ipcp;
159
160         return container_of(ipcp, struct shmid_kernel, shm_perm);
161 }
162
163 static inline struct shmid_kernel *shm_lock_check_down(
164                                                 struct ipc_namespace *ns,
165                                                 int id)
166 {
167         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
168
169         if (IS_ERR(ipcp))
170                 return (struct shmid_kernel *)ipcp;
171
172         return container_of(ipcp, struct shmid_kernel, shm_perm);
173 }
174
175 /*
176  * shm_lock_(check_) routines are called in the paths where the rw_mutex
177  * is not held.
178  */
179 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
180 {
181         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
182
183         if (IS_ERR(ipcp))
184                 return (struct shmid_kernel *)ipcp;
185
186         return container_of(ipcp, struct shmid_kernel, shm_perm);
187 }
188
189 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
190                                                 int id)
191 {
192         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
193
194         if (IS_ERR(ipcp))
195                 return (struct shmid_kernel *)ipcp;
196
197         return container_of(ipcp, struct shmid_kernel, shm_perm);
198 }
199
200 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
201 {
202         ipc_rmid(&shm_ids(ns), &s->shm_perm);
203 }
204
205 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
206 {
207         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
208 }
209
210
211
212 /* This is called by fork, once for every shm attach. */
213 static void shm_open(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
219         shp = shm_lock(sfd->ns, sfd->id);
220         BUG_ON(IS_ERR(shp));
221         shp->shm_atim = get_seconds();
222         shp->shm_lprid = task_tgid_vnr(current);
223         shp->shm_nattch++;
224         shm_unlock(shp);
225 }
226
227 /*
228  * shm_destroy - free the struct shmid_kernel
229  *
230  * @ns: namespace
231  * @shp: struct to free
232  *
233  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
234  * but returns with shp unlocked and freed.
235  */
236 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
237 {
238         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
239         shm_rmid(ns, shp);
240         shm_unlock(shp);
241         if (!is_file_hugepages(shp->shm_file))
242                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
243         else
244                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
245                                                 shp->mlock_user);
246         fput (shp->shm_file);
247         security_shm_free(shp);
248         ipc_rcu_putref(shp);
249 }
250
251 /*
252  * remove the attach descriptor vma.
253  * free memory for segment if it is marked destroyed.
254  * The descriptor has already been removed from the current->mm->mmap list
255  * and will later be kfree()d.
256  */
257 static void shm_close(struct vm_area_struct *vma)
258 {
259         struct file * file = vma->vm_file;
260         struct shm_file_data *sfd = shm_file_data(file);
261         struct shmid_kernel *shp;
262         struct ipc_namespace *ns = sfd->ns;
263
264         down_write(&shm_ids(ns).rw_mutex);
265         /* remove from the list of attaches of the shm segment */
266         shp = shm_lock_down(ns, sfd->id);
267         BUG_ON(IS_ERR(shp));
268         shp->shm_lprid = task_tgid_vnr(current);
269         shp->shm_dtim = get_seconds();
270         shp->shm_nattch--;
271         if(shp->shm_nattch == 0 &&
272            shp->shm_perm.mode & SHM_DEST)
273                 shm_destroy(ns, shp);
274         else
275                 shm_unlock(shp);
276         up_write(&shm_ids(ns).rw_mutex);
277 }
278
279 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
280 {
281         struct file *file = vma->vm_file;
282         struct shm_file_data *sfd = shm_file_data(file);
283
284         return sfd->vm_ops->fault(vma, vmf);
285 }
286
287 #ifdef CONFIG_NUMA
288 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
289 {
290         struct file *file = vma->vm_file;
291         struct shm_file_data *sfd = shm_file_data(file);
292         int err = 0;
293         if (sfd->vm_ops->set_policy)
294                 err = sfd->vm_ops->set_policy(vma, new);
295         return err;
296 }
297
298 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
299                                         unsigned long addr)
300 {
301         struct file *file = vma->vm_file;
302         struct shm_file_data *sfd = shm_file_data(file);
303         struct mempolicy *pol = NULL;
304
305         if (sfd->vm_ops->get_policy)
306                 pol = sfd->vm_ops->get_policy(vma, addr);
307         else if (vma->vm_policy)
308                 pol = vma->vm_policy;
309         else
310                 pol = current->mempolicy;
311         return pol;
312 }
313 #endif
314
315 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
316 {
317         struct shm_file_data *sfd = shm_file_data(file);
318         int ret;
319
320         ret = sfd->file->f_op->mmap(sfd->file, vma);
321         if (ret != 0)
322                 return ret;
323         sfd->vm_ops = vma->vm_ops;
324 #ifdef CONFIG_MMU
325         BUG_ON(!sfd->vm_ops->fault);
326 #endif
327         vma->vm_ops = &shm_vm_ops;
328         shm_open(vma);
329
330         return ret;
331 }
332
333 static int shm_release(struct inode *ino, struct file *file)
334 {
335         struct shm_file_data *sfd = shm_file_data(file);
336
337         put_ipc_ns(sfd->ns);
338         shm_file_data(file) = NULL;
339         kfree(sfd);
340         return 0;
341 }
342
343 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
344 {
345         int (*fsync) (struct file *, struct dentry *, int datasync);
346         struct shm_file_data *sfd = shm_file_data(file);
347         int ret = -EINVAL;
348
349         fsync = sfd->file->f_op->fsync;
350         if (fsync)
351                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
352         return ret;
353 }
354
355 static unsigned long shm_get_unmapped_area(struct file *file,
356         unsigned long addr, unsigned long len, unsigned long pgoff,
357         unsigned long flags)
358 {
359         struct shm_file_data *sfd = shm_file_data(file);
360         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
361 }
362
363 int is_file_shm_hugepages(struct file *file)
364 {
365         int ret = 0;
366
367         if (file->f_op == &shm_file_operations) {
368                 struct shm_file_data *sfd;
369                 sfd = shm_file_data(file);
370                 ret = is_file_hugepages(sfd->file);
371         }
372         return ret;
373 }
374
375 static const struct file_operations shm_file_operations = {
376         .mmap           = shm_mmap,
377         .fsync          = shm_fsync,
378         .release        = shm_release,
379         .get_unmapped_area      = shm_get_unmapped_area,
380 };
381
382 static struct vm_operations_struct shm_vm_ops = {
383         .open   = shm_open,     /* callback for a new vm-area open */
384         .close  = shm_close,    /* callback for when the vm-area is released */
385         .fault  = shm_fault,
386 #if defined(CONFIG_NUMA)
387         .set_policy = shm_set_policy,
388         .get_policy = shm_get_policy,
389 #endif
390 };
391
392 /**
393  * newseg - Create a new shared memory segment
394  * @ns: namespace
395  * @params: ptr to the structure that contains key, size and shmflg
396  *
397  * Called with shm_ids.rw_mutex held as a writer.
398  */
399
400 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
401 {
402         key_t key = params->key;
403         int shmflg = params->flg;
404         size_t size = params->u.size;
405         int error;
406         struct shmid_kernel *shp;
407         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
408         struct file * file;
409         char name[13];
410         int id;
411
412         if (size < SHMMIN || size > ns->shm_ctlmax)
413                 return -EINVAL;
414
415         if (ns->shm_tot + numpages > ns->shm_ctlall)
416                 return -ENOSPC;
417
418         shp = ipc_rcu_alloc(sizeof(*shp));
419         if (!shp)
420                 return -ENOMEM;
421
422         shp->shm_perm.key = key;
423         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
424         shp->mlock_user = NULL;
425
426         shp->shm_perm.security = NULL;
427         error = security_shm_alloc(shp);
428         if (error) {
429                 ipc_rcu_putref(shp);
430                 return error;
431         }
432
433         sprintf (name, "SYSV%08x", key);
434         if (shmflg & SHM_HUGETLB) {
435                 /* hugetlb_file_setup takes care of mlock user accounting */
436                 file = hugetlb_file_setup(name, size);
437                 shp->mlock_user = current->user;
438         } else {
439                 int acctflag = VM_ACCOUNT;
440                 /*
441                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
442                  * if it's asked for.
443                  */
444                 if  ((shmflg & SHM_NORESERVE) &&
445                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
446                         acctflag = 0;
447                 file = shmem_file_setup(name, size, acctflag);
448         }
449         error = PTR_ERR(file);
450         if (IS_ERR(file))
451                 goto no_file;
452
453         id = shm_addid(ns, shp);
454         if (id < 0) {
455                 error = id;
456                 goto no_id;
457         }
458
459         shp->shm_cprid = task_tgid_vnr(current);
460         shp->shm_lprid = 0;
461         shp->shm_atim = shp->shm_dtim = 0;
462         shp->shm_ctim = get_seconds();
463         shp->shm_segsz = size;
464         shp->shm_nattch = 0;
465         shp->shm_perm.id = shm_buildid(id, shp->shm_perm.seq);
466         shp->shm_file = file;
467         /*
468          * shmid gets reported as "inode#" in /proc/pid/maps.
469          * proc-ps tools use this. Changing this will break them.
470          */
471         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
472
473         ns->shm_tot += numpages;
474         error = shp->shm_perm.id;
475         shm_unlock(shp);
476         return error;
477
478 no_id:
479         fput(file);
480 no_file:
481         security_shm_free(shp);
482         ipc_rcu_putref(shp);
483         return error;
484 }
485
486 /*
487  * Called with shm_ids.rw_mutex and ipcp locked.
488  */
489 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
490 {
491         struct shmid_kernel *shp;
492
493         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
494         return security_shm_associate(shp, shmflg);
495 }
496
497 /*
498  * Called with shm_ids.rw_mutex and ipcp locked.
499  */
500 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
501                                 struct ipc_params *params)
502 {
503         struct shmid_kernel *shp;
504
505         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
506         if (shp->shm_segsz < params->u.size)
507                 return -EINVAL;
508
509         return 0;
510 }
511
512 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
513 {
514         struct ipc_namespace *ns;
515         struct ipc_ops shm_ops;
516         struct ipc_params shm_params;
517
518         ns = current->nsproxy->ipc_ns;
519
520         shm_ops.getnew = newseg;
521         shm_ops.associate = shm_security;
522         shm_ops.more_checks = shm_more_checks;
523
524         shm_params.key = key;
525         shm_params.flg = shmflg;
526         shm_params.u.size = size;
527
528         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
529 }
530
531 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
532 {
533         switch(version) {
534         case IPC_64:
535                 return copy_to_user(buf, in, sizeof(*in));
536         case IPC_OLD:
537             {
538                 struct shmid_ds out;
539
540                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
541                 out.shm_segsz   = in->shm_segsz;
542                 out.shm_atime   = in->shm_atime;
543                 out.shm_dtime   = in->shm_dtime;
544                 out.shm_ctime   = in->shm_ctime;
545                 out.shm_cpid    = in->shm_cpid;
546                 out.shm_lpid    = in->shm_lpid;
547                 out.shm_nattch  = in->shm_nattch;
548
549                 return copy_to_user(buf, &out, sizeof(out));
550             }
551         default:
552                 return -EINVAL;
553         }
554 }
555
556 struct shm_setbuf {
557         uid_t   uid;
558         gid_t   gid;
559         mode_t  mode;
560 };      
561
562 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
563 {
564         switch(version) {
565         case IPC_64:
566             {
567                 struct shmid64_ds tbuf;
568
569                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
570                         return -EFAULT;
571
572                 out->uid        = tbuf.shm_perm.uid;
573                 out->gid        = tbuf.shm_perm.gid;
574                 out->mode       = tbuf.shm_perm.mode;
575
576                 return 0;
577             }
578         case IPC_OLD:
579             {
580                 struct shmid_ds tbuf_old;
581
582                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
583                         return -EFAULT;
584
585                 out->uid        = tbuf_old.shm_perm.uid;
586                 out->gid        = tbuf_old.shm_perm.gid;
587                 out->mode       = tbuf_old.shm_perm.mode;
588
589                 return 0;
590             }
591         default:
592                 return -EINVAL;
593         }
594 }
595
596 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
597 {
598         switch(version) {
599         case IPC_64:
600                 return copy_to_user(buf, in, sizeof(*in));
601         case IPC_OLD:
602             {
603                 struct shminfo out;
604
605                 if(in->shmmax > INT_MAX)
606                         out.shmmax = INT_MAX;
607                 else
608                         out.shmmax = (int)in->shmmax;
609
610                 out.shmmin      = in->shmmin;
611                 out.shmmni      = in->shmmni;
612                 out.shmseg      = in->shmseg;
613                 out.shmall      = in->shmall; 
614
615                 return copy_to_user(buf, &out, sizeof(out));
616             }
617         default:
618                 return -EINVAL;
619         }
620 }
621
622 /*
623  * Called with shm_ids.rw_mutex held as a reader
624  */
625 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
626                 unsigned long *swp)
627 {
628         int next_id;
629         int total, in_use;
630
631         *rss = 0;
632         *swp = 0;
633
634         in_use = shm_ids(ns).in_use;
635
636         for (total = 0, next_id = 0; total < in_use; next_id++) {
637                 struct shmid_kernel *shp;
638                 struct inode *inode;
639
640                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
641                 if (shp == NULL)
642                         continue;
643
644                 inode = shp->shm_file->f_path.dentry->d_inode;
645
646                 if (is_file_hugepages(shp->shm_file)) {
647                         struct address_space *mapping = inode->i_mapping;
648                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
649                 } else {
650                         struct shmem_inode_info *info = SHMEM_I(inode);
651                         spin_lock(&info->lock);
652                         *rss += inode->i_mapping->nrpages;
653                         *swp += info->swapped;
654                         spin_unlock(&info->lock);
655                 }
656
657                 total++;
658         }
659 }
660
661 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
662 {
663         struct shm_setbuf setbuf;
664         struct shmid_kernel *shp;
665         int err, version;
666         struct ipc_namespace *ns;
667
668         if (cmd < 0 || shmid < 0) {
669                 err = -EINVAL;
670                 goto out;
671         }
672
673         version = ipc_parse_version(&cmd);
674         ns = current->nsproxy->ipc_ns;
675
676         switch (cmd) { /* replace with proc interface ? */
677         case IPC_INFO:
678         {
679                 struct shminfo64 shminfo;
680
681                 err = security_shm_shmctl(NULL, cmd);
682                 if (err)
683                         return err;
684
685                 memset(&shminfo,0,sizeof(shminfo));
686                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
687                 shminfo.shmmax = ns->shm_ctlmax;
688                 shminfo.shmall = ns->shm_ctlall;
689
690                 shminfo.shmmin = SHMMIN;
691                 if(copy_shminfo_to_user (buf, &shminfo, version))
692                         return -EFAULT;
693
694                 down_read(&shm_ids(ns).rw_mutex);
695                 err = ipc_get_maxid(&shm_ids(ns));
696                 up_read(&shm_ids(ns).rw_mutex);
697
698                 if(err<0)
699                         err = 0;
700                 goto out;
701         }
702         case SHM_INFO:
703         {
704                 struct shm_info shm_info;
705
706                 err = security_shm_shmctl(NULL, cmd);
707                 if (err)
708                         return err;
709
710                 memset(&shm_info,0,sizeof(shm_info));
711                 down_read(&shm_ids(ns).rw_mutex);
712                 shm_info.used_ids = shm_ids(ns).in_use;
713                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
714                 shm_info.shm_tot = ns->shm_tot;
715                 shm_info.swap_attempts = 0;
716                 shm_info.swap_successes = 0;
717                 err = ipc_get_maxid(&shm_ids(ns));
718                 up_read(&shm_ids(ns).rw_mutex);
719                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
720                         err = -EFAULT;
721                         goto out;
722                 }
723
724                 err = err < 0 ? 0 : err;
725                 goto out;
726         }
727         case SHM_STAT:
728         case IPC_STAT:
729         {
730                 struct shmid64_ds tbuf;
731                 int result;
732
733                 if (!buf) {
734                         err = -EFAULT;
735                         goto out;
736                 }
737
738                 if (cmd == SHM_STAT) {
739                         shp = shm_lock(ns, shmid);
740                         if (IS_ERR(shp)) {
741                                 err = PTR_ERR(shp);
742                                 goto out;
743                         }
744                         result = shp->shm_perm.id;
745                 } else {
746                         shp = shm_lock_check(ns, shmid);
747                         if (IS_ERR(shp)) {
748                                 err = PTR_ERR(shp);
749                                 goto out;
750                         }
751                         result = 0;
752                 }
753                 err=-EACCES;
754                 if (ipcperms (&shp->shm_perm, S_IRUGO))
755                         goto out_unlock;
756                 err = security_shm_shmctl(shp, cmd);
757                 if (err)
758                         goto out_unlock;
759                 memset(&tbuf, 0, sizeof(tbuf));
760                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
761                 tbuf.shm_segsz  = shp->shm_segsz;
762                 tbuf.shm_atime  = shp->shm_atim;
763                 tbuf.shm_dtime  = shp->shm_dtim;
764                 tbuf.shm_ctime  = shp->shm_ctim;
765                 tbuf.shm_cpid   = shp->shm_cprid;
766                 tbuf.shm_lpid   = shp->shm_lprid;
767                 tbuf.shm_nattch = shp->shm_nattch;
768                 shm_unlock(shp);
769                 if(copy_shmid_to_user (buf, &tbuf, version))
770                         err = -EFAULT;
771                 else
772                         err = result;
773                 goto out;
774         }
775         case SHM_LOCK:
776         case SHM_UNLOCK:
777         {
778                 shp = shm_lock_check(ns, shmid);
779                 if (IS_ERR(shp)) {
780                         err = PTR_ERR(shp);
781                         goto out;
782                 }
783
784                 err = audit_ipc_obj(&(shp->shm_perm));
785                 if (err)
786                         goto out_unlock;
787
788                 if (!capable(CAP_IPC_LOCK)) {
789                         err = -EPERM;
790                         if (current->euid != shp->shm_perm.uid &&
791                             current->euid != shp->shm_perm.cuid)
792                                 goto out_unlock;
793                         if (cmd == SHM_LOCK &&
794                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
795                                 goto out_unlock;
796                 }
797
798                 err = security_shm_shmctl(shp, cmd);
799                 if (err)
800                         goto out_unlock;
801                 
802                 if(cmd==SHM_LOCK) {
803                         struct user_struct * user = current->user;
804                         if (!is_file_hugepages(shp->shm_file)) {
805                                 err = shmem_lock(shp->shm_file, 1, user);
806                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
807                                         shp->shm_perm.mode |= SHM_LOCKED;
808                                         shp->mlock_user = user;
809                                 }
810                         }
811                 } else if (!is_file_hugepages(shp->shm_file)) {
812                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
813                         shp->shm_perm.mode &= ~SHM_LOCKED;
814                         shp->mlock_user = NULL;
815                 }
816                 shm_unlock(shp);
817                 goto out;
818         }
819         case IPC_RMID:
820         {
821                 /*
822                  *      We cannot simply remove the file. The SVID states
823                  *      that the block remains until the last person
824                  *      detaches from it, then is deleted. A shmat() on
825                  *      an RMID segment is legal in older Linux and if 
826                  *      we change it apps break...
827                  *
828                  *      Instead we set a destroyed flag, and then blow
829                  *      the name away when the usage hits zero.
830                  */
831                 down_write(&shm_ids(ns).rw_mutex);
832                 shp = shm_lock_check_down(ns, shmid);
833                 if (IS_ERR(shp)) {
834                         err = PTR_ERR(shp);
835                         goto out_up;
836                 }
837
838                 err = audit_ipc_obj(&(shp->shm_perm));
839                 if (err)
840                         goto out_unlock_up;
841
842                 if (current->euid != shp->shm_perm.uid &&
843                     current->euid != shp->shm_perm.cuid && 
844                     !capable(CAP_SYS_ADMIN)) {
845                         err=-EPERM;
846                         goto out_unlock_up;
847                 }
848
849                 err = security_shm_shmctl(shp, cmd);
850                 if (err)
851                         goto out_unlock_up;
852
853                 do_shm_rmid(ns, shp);
854                 up_write(&shm_ids(ns).rw_mutex);
855                 goto out;
856         }
857
858         case IPC_SET:
859         {
860                 if (!buf) {
861                         err = -EFAULT;
862                         goto out;
863                 }
864
865                 if (copy_shmid_from_user (&setbuf, buf, version)) {
866                         err = -EFAULT;
867                         goto out;
868                 }
869                 down_write(&shm_ids(ns).rw_mutex);
870                 shp = shm_lock_check_down(ns, shmid);
871                 if (IS_ERR(shp)) {
872                         err = PTR_ERR(shp);
873                         goto out_up;
874                 }
875                 err = audit_ipc_obj(&(shp->shm_perm));
876                 if (err)
877                         goto out_unlock_up;
878                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
879                 if (err)
880                         goto out_unlock_up;
881                 err=-EPERM;
882                 if (current->euid != shp->shm_perm.uid &&
883                     current->euid != shp->shm_perm.cuid && 
884                     !capable(CAP_SYS_ADMIN)) {
885                         goto out_unlock_up;
886                 }
887
888                 err = security_shm_shmctl(shp, cmd);
889                 if (err)
890                         goto out_unlock_up;
891                 
892                 shp->shm_perm.uid = setbuf.uid;
893                 shp->shm_perm.gid = setbuf.gid;
894                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
895                         | (setbuf.mode & S_IRWXUGO);
896                 shp->shm_ctim = get_seconds();
897                 break;
898         }
899
900         default:
901                 err = -EINVAL;
902                 goto out;
903         }
904
905         err = 0;
906 out_unlock_up:
907         shm_unlock(shp);
908 out_up:
909         up_write(&shm_ids(ns).rw_mutex);
910         goto out;
911 out_unlock:
912         shm_unlock(shp);
913 out:
914         return err;
915 }
916
917 /*
918  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
919  *
920  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
921  * "raddr" thing points to kernel space, and there has to be a wrapper around
922  * this.
923  */
924 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
925 {
926         struct shmid_kernel *shp;
927         unsigned long addr;
928         unsigned long size;
929         struct file * file;
930         int    err;
931         unsigned long flags;
932         unsigned long prot;
933         int acc_mode;
934         unsigned long user_addr;
935         struct ipc_namespace *ns;
936         struct shm_file_data *sfd;
937         struct path path;
938         mode_t f_mode;
939
940         err = -EINVAL;
941         if (shmid < 0)
942                 goto out;
943         else if ((addr = (ulong)shmaddr)) {
944                 if (addr & (SHMLBA-1)) {
945                         if (shmflg & SHM_RND)
946                                 addr &= ~(SHMLBA-1);       /* round down */
947                         else
948 #ifndef __ARCH_FORCE_SHMLBA
949                                 if (addr & ~PAGE_MASK)
950 #endif
951                                         goto out;
952                 }
953                 flags = MAP_SHARED | MAP_FIXED;
954         } else {
955                 if ((shmflg & SHM_REMAP))
956                         goto out;
957
958                 flags = MAP_SHARED;
959         }
960
961         if (shmflg & SHM_RDONLY) {
962                 prot = PROT_READ;
963                 acc_mode = S_IRUGO;
964                 f_mode = FMODE_READ;
965         } else {
966                 prot = PROT_READ | PROT_WRITE;
967                 acc_mode = S_IRUGO | S_IWUGO;
968                 f_mode = FMODE_READ | FMODE_WRITE;
969         }
970         if (shmflg & SHM_EXEC) {
971                 prot |= PROT_EXEC;
972                 acc_mode |= S_IXUGO;
973         }
974
975         /*
976          * We cannot rely on the fs check since SYSV IPC does have an
977          * additional creator id...
978          */
979         ns = current->nsproxy->ipc_ns;
980         shp = shm_lock_check(ns, shmid);
981         if (IS_ERR(shp)) {
982                 err = PTR_ERR(shp);
983                 goto out;
984         }
985
986         err = -EACCES;
987         if (ipcperms(&shp->shm_perm, acc_mode))
988                 goto out_unlock;
989
990         err = security_shm_shmat(shp, shmaddr, shmflg);
991         if (err)
992                 goto out_unlock;
993
994         path.dentry = dget(shp->shm_file->f_path.dentry);
995         path.mnt    = shp->shm_file->f_path.mnt;
996         shp->shm_nattch++;
997         size = i_size_read(path.dentry->d_inode);
998         shm_unlock(shp);
999
1000         err = -ENOMEM;
1001         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1002         if (!sfd)
1003                 goto out_put_dentry;
1004
1005         err = -ENOMEM;
1006
1007         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
1008         if (!file)
1009                 goto out_free;
1010
1011         file->private_data = sfd;
1012         file->f_mapping = shp->shm_file->f_mapping;
1013         sfd->id = shp->shm_perm.id;
1014         sfd->ns = get_ipc_ns(ns);
1015         sfd->file = shp->shm_file;
1016         sfd->vm_ops = NULL;
1017
1018         down_write(&current->mm->mmap_sem);
1019         if (addr && !(shmflg & SHM_REMAP)) {
1020                 err = -EINVAL;
1021                 if (find_vma_intersection(current->mm, addr, addr + size))
1022                         goto invalid;
1023                 /*
1024                  * If shm segment goes below stack, make sure there is some
1025                  * space left for the stack to grow (at least 4 pages).
1026                  */
1027                 if (addr < current->mm->start_stack &&
1028                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1029                         goto invalid;
1030         }
1031                 
1032         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1033         *raddr = user_addr;
1034         err = 0;
1035         if (IS_ERR_VALUE(user_addr))
1036                 err = (long)user_addr;
1037 invalid:
1038         up_write(&current->mm->mmap_sem);
1039
1040         fput(file);
1041
1042 out_nattch:
1043         down_write(&shm_ids(ns).rw_mutex);
1044         shp = shm_lock_down(ns, shmid);
1045         BUG_ON(IS_ERR(shp));
1046         shp->shm_nattch--;
1047         if(shp->shm_nattch == 0 &&
1048            shp->shm_perm.mode & SHM_DEST)
1049                 shm_destroy(ns, shp);
1050         else
1051                 shm_unlock(shp);
1052         up_write(&shm_ids(ns).rw_mutex);
1053
1054 out:
1055         return err;
1056
1057 out_unlock:
1058         shm_unlock(shp);
1059         goto out;
1060
1061 out_free:
1062         kfree(sfd);
1063 out_put_dentry:
1064         dput(path.dentry);
1065         goto out_nattch;
1066 }
1067
1068 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1069 {
1070         unsigned long ret;
1071         long err;
1072
1073         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1074         if (err)
1075                 return err;
1076         force_successful_syscall_return();
1077         return (long)ret;
1078 }
1079
1080 /*
1081  * detach and kill segment if marked destroyed.
1082  * The work is done in shm_close.
1083  */
1084 asmlinkage long sys_shmdt(char __user *shmaddr)
1085 {
1086         struct mm_struct *mm = current->mm;
1087         struct vm_area_struct *vma, *next;
1088         unsigned long addr = (unsigned long)shmaddr;
1089         loff_t size = 0;
1090         int retval = -EINVAL;
1091
1092         if (addr & ~PAGE_MASK)
1093                 return retval;
1094
1095         down_write(&mm->mmap_sem);
1096
1097         /*
1098          * This function tries to be smart and unmap shm segments that
1099          * were modified by partial mlock or munmap calls:
1100          * - It first determines the size of the shm segment that should be
1101          *   unmapped: It searches for a vma that is backed by shm and that
1102          *   started at address shmaddr. It records it's size and then unmaps
1103          *   it.
1104          * - Then it unmaps all shm vmas that started at shmaddr and that
1105          *   are within the initially determined size.
1106          * Errors from do_munmap are ignored: the function only fails if
1107          * it's called with invalid parameters or if it's called to unmap
1108          * a part of a vma. Both calls in this function are for full vmas,
1109          * the parameters are directly copied from the vma itself and always
1110          * valid - therefore do_munmap cannot fail. (famous last words?)
1111          */
1112         /*
1113          * If it had been mremap()'d, the starting address would not
1114          * match the usual checks anyway. So assume all vma's are
1115          * above the starting address given.
1116          */
1117         vma = find_vma(mm, addr);
1118
1119         while (vma) {
1120                 next = vma->vm_next;
1121
1122                 /*
1123                  * Check if the starting address would match, i.e. it's
1124                  * a fragment created by mprotect() and/or munmap(), or it
1125                  * otherwise it starts at this address with no hassles.
1126                  */
1127                 if ((vma->vm_ops == &shm_vm_ops) &&
1128                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1129
1130
1131                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1132                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1133                         /*
1134                          * We discovered the size of the shm segment, so
1135                          * break out of here and fall through to the next
1136                          * loop that uses the size information to stop
1137                          * searching for matching vma's.
1138                          */
1139                         retval = 0;
1140                         vma = next;
1141                         break;
1142                 }
1143                 vma = next;
1144         }
1145
1146         /*
1147          * We need look no further than the maximum address a fragment
1148          * could possibly have landed at. Also cast things to loff_t to
1149          * prevent overflows and make comparisions vs. equal-width types.
1150          */
1151         size = PAGE_ALIGN(size);
1152         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1153                 next = vma->vm_next;
1154
1155                 /* finding a matching vma now does not alter retval */
1156                 if ((vma->vm_ops == &shm_vm_ops) &&
1157                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1158
1159                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1160                 vma = next;
1161         }
1162
1163         up_write(&mm->mmap_sem);
1164         return retval;
1165 }
1166
1167 #ifdef CONFIG_PROC_FS
1168 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1169 {
1170         struct shmid_kernel *shp = it;
1171         char *format;
1172
1173 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1174 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1175
1176         if (sizeof(size_t) <= sizeof(int))
1177                 format = SMALL_STRING;
1178         else
1179                 format = BIG_STRING;
1180         return seq_printf(s, format,
1181                           shp->shm_perm.key,
1182                           shp->shm_perm.id,
1183                           shp->shm_perm.mode,
1184                           shp->shm_segsz,
1185                           shp->shm_cprid,
1186                           shp->shm_lprid,
1187                           shp->shm_nattch,
1188                           shp->shm_perm.uid,
1189                           shp->shm_perm.gid,
1190                           shp->shm_perm.cuid,
1191                           shp->shm_perm.cgid,
1192                           shp->shm_atim,
1193                           shp->shm_dtim,
1194                           shp->shm_ctim);
1195 }
1196 #endif