fix idr_find() locking
[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
42 #include <asm/uaccess.h>
43
44 #include "util.h"
45
46 struct shm_file_data {
47         int id;
48         struct ipc_namespace *ns;
49         struct file *file;
50         const struct vm_operations_struct *vm_ops;
51 };
52
53 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
54
55 static const struct file_operations shm_file_operations;
56 static struct vm_operations_struct shm_vm_ops;
57
58 static struct ipc_ids init_shm_ids;
59
60 #define shm_ids(ns)     (*((ns)->ids[IPC_SHM_IDS]))
61
62 #define shm_unlock(shp)                 \
63         ipc_unlock(&(shp)->shm_perm)
64 #define shm_buildid(ns, id, seq)        \
65         ipc_buildid(&shm_ids(ns), 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 int shm_init_ns(struct ipc_namespace *ns)
101 {
102         struct ipc_ids *ids;
103
104         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
105         if (ids == NULL)
106                 return -ENOMEM;
107
108         __shm_init_ns(ns, ids);
109         return 0;
110 }
111
112 void shm_exit_ns(struct ipc_namespace *ns)
113 {
114         struct shmid_kernel *shp;
115         int next_id;
116         int total, in_use;
117
118         down_write(&shm_ids(ns).rw_mutex);
119
120         in_use = shm_ids(ns).in_use;
121
122         for (total = 0, next_id = 0; total < in_use; next_id++) {
123                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
124                 if (shp == NULL)
125                         continue;
126                 ipc_lock_by_ptr(&shp->shm_perm);
127                 do_shm_rmid(ns, shp);
128                 total++;
129         }
130         up_write(&shm_ids(ns).rw_mutex);
131
132         kfree(ns->ids[IPC_SHM_IDS]);
133         ns->ids[IPC_SHM_IDS] = NULL;
134 }
135
136 void __init shm_init (void)
137 {
138         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
139         ipc_init_proc_interface("sysvipc/shm",
140                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
141                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
142 }
143
144 /*
145  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
146  * is held to protect access to the idr tree.
147  */
148 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
149                                                 int id)
150 {
151         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
152
153         return container_of(ipcp, struct shmid_kernel, shm_perm);
154 }
155
156 static inline struct shmid_kernel *shm_lock_check_down(
157                                                 struct ipc_namespace *ns,
158                                                 int id)
159 {
160         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
161
162         return container_of(ipcp, struct shmid_kernel, shm_perm);
163 }
164
165 /*
166  * shm_lock_(check_) routines are called in the paths where the rw_mutex
167  * is not held.
168  */
169 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
170 {
171         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
172
173         return container_of(ipcp, struct shmid_kernel, shm_perm);
174 }
175
176 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
177                                                 int id)
178 {
179         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
180
181         return container_of(ipcp, struct shmid_kernel, shm_perm);
182 }
183
184 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
185 {
186         ipc_rmid(&shm_ids(ns), &s->shm_perm);
187 }
188
189 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
190 {
191         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
192 }
193
194
195
196 /* This is called by fork, once for every shm attach. */
197 static void shm_open(struct vm_area_struct *vma)
198 {
199         struct file *file = vma->vm_file;
200         struct shm_file_data *sfd = shm_file_data(file);
201         struct shmid_kernel *shp;
202
203         shp = shm_lock(sfd->ns, sfd->id);
204         BUG_ON(IS_ERR(shp));
205         shp->shm_atim = get_seconds();
206         shp->shm_lprid = task_tgid_vnr(current);
207         shp->shm_nattch++;
208         shm_unlock(shp);
209 }
210
211 /*
212  * shm_destroy - free the struct shmid_kernel
213  *
214  * @ns: namespace
215  * @shp: struct to free
216  *
217  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
218  * but returns with shp unlocked and freed.
219  */
220 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
221 {
222         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
223         shm_rmid(ns, shp);
224         shm_unlock(shp);
225         if (!is_file_hugepages(shp->shm_file))
226                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
227         else
228                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
229                                                 shp->mlock_user);
230         fput (shp->shm_file);
231         security_shm_free(shp);
232         ipc_rcu_putref(shp);
233 }
234
235 /*
236  * remove the attach descriptor vma.
237  * free memory for segment if it is marked destroyed.
238  * The descriptor has already been removed from the current->mm->mmap list
239  * and will later be kfree()d.
240  */
241 static void shm_close(struct vm_area_struct *vma)
242 {
243         struct file * file = vma->vm_file;
244         struct shm_file_data *sfd = shm_file_data(file);
245         struct shmid_kernel *shp;
246         struct ipc_namespace *ns = sfd->ns;
247
248         down_write(&shm_ids(ns).rw_mutex);
249         /* remove from the list of attaches of the shm segment */
250         shp = shm_lock_down(ns, sfd->id);
251         BUG_ON(IS_ERR(shp));
252         shp->shm_lprid = task_tgid_vnr(current);
253         shp->shm_dtim = get_seconds();
254         shp->shm_nattch--;
255         if(shp->shm_nattch == 0 &&
256            shp->shm_perm.mode & SHM_DEST)
257                 shm_destroy(ns, shp);
258         else
259                 shm_unlock(shp);
260         up_write(&shm_ids(ns).rw_mutex);
261 }
262
263 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
264 {
265         struct file *file = vma->vm_file;
266         struct shm_file_data *sfd = shm_file_data(file);
267
268         return sfd->vm_ops->fault(vma, vmf);
269 }
270
271 #ifdef CONFIG_NUMA
272 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
273 {
274         struct file *file = vma->vm_file;
275         struct shm_file_data *sfd = shm_file_data(file);
276         int err = 0;
277         if (sfd->vm_ops->set_policy)
278                 err = sfd->vm_ops->set_policy(vma, new);
279         return err;
280 }
281
282 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
283                                         unsigned long addr)
284 {
285         struct file *file = vma->vm_file;
286         struct shm_file_data *sfd = shm_file_data(file);
287         struct mempolicy *pol = NULL;
288
289         if (sfd->vm_ops->get_policy)
290                 pol = sfd->vm_ops->get_policy(vma, addr);
291         else if (vma->vm_policy)
292                 pol = vma->vm_policy;
293         else
294                 pol = current->mempolicy;
295         return pol;
296 }
297 #endif
298
299 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
300 {
301         struct shm_file_data *sfd = shm_file_data(file);
302         int ret;
303
304         ret = sfd->file->f_op->mmap(sfd->file, vma);
305         if (ret != 0)
306                 return ret;
307         sfd->vm_ops = vma->vm_ops;
308 #ifdef CONFIG_MMU
309         BUG_ON(!sfd->vm_ops->fault);
310 #endif
311         vma->vm_ops = &shm_vm_ops;
312         shm_open(vma);
313
314         return ret;
315 }
316
317 static int shm_release(struct inode *ino, struct file *file)
318 {
319         struct shm_file_data *sfd = shm_file_data(file);
320
321         put_ipc_ns(sfd->ns);
322         shm_file_data(file) = NULL;
323         kfree(sfd);
324         return 0;
325 }
326
327 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
328 {
329         int (*fsync) (struct file *, struct dentry *, int datasync);
330         struct shm_file_data *sfd = shm_file_data(file);
331         int ret = -EINVAL;
332
333         fsync = sfd->file->f_op->fsync;
334         if (fsync)
335                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
336         return ret;
337 }
338
339 static unsigned long shm_get_unmapped_area(struct file *file,
340         unsigned long addr, unsigned long len, unsigned long pgoff,
341         unsigned long flags)
342 {
343         struct shm_file_data *sfd = shm_file_data(file);
344         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
345 }
346
347 int is_file_shm_hugepages(struct file *file)
348 {
349         int ret = 0;
350
351         if (file->f_op == &shm_file_operations) {
352                 struct shm_file_data *sfd;
353                 sfd = shm_file_data(file);
354                 ret = is_file_hugepages(sfd->file);
355         }
356         return ret;
357 }
358
359 static const struct file_operations shm_file_operations = {
360         .mmap           = shm_mmap,
361         .fsync          = shm_fsync,
362         .release        = shm_release,
363         .get_unmapped_area      = shm_get_unmapped_area,
364 };
365
366 static struct vm_operations_struct shm_vm_ops = {
367         .open   = shm_open,     /* callback for a new vm-area open */
368         .close  = shm_close,    /* callback for when the vm-area is released */
369         .fault  = shm_fault,
370 #if defined(CONFIG_NUMA)
371         .set_policy = shm_set_policy,
372         .get_policy = shm_get_policy,
373 #endif
374 };
375
376 /**
377  * newseg - Create a new shared memory segment
378  * @ns: namespace
379  * @params: ptr to the structure that contains key, size and shmflg
380  *
381  * Called with shm_ids.rw_mutex held as a writer.
382  */
383
384 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
385 {
386         key_t key = params->key;
387         int shmflg = params->flg;
388         size_t size = params->u.size;
389         int error;
390         struct shmid_kernel *shp;
391         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
392         struct file * file;
393         char name[13];
394         int id;
395
396         if (size < SHMMIN || size > ns->shm_ctlmax)
397                 return -EINVAL;
398
399         if (ns->shm_tot + numpages > ns->shm_ctlall)
400                 return -ENOSPC;
401
402         shp = ipc_rcu_alloc(sizeof(*shp));
403         if (!shp)
404                 return -ENOMEM;
405
406         shp->shm_perm.key = key;
407         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
408         shp->mlock_user = NULL;
409
410         shp->shm_perm.security = NULL;
411         error = security_shm_alloc(shp);
412         if (error) {
413                 ipc_rcu_putref(shp);
414                 return error;
415         }
416
417         sprintf (name, "SYSV%08x", key);
418         if (shmflg & SHM_HUGETLB) {
419                 /* hugetlb_file_setup takes care of mlock user accounting */
420                 file = hugetlb_file_setup(name, size);
421                 shp->mlock_user = current->user;
422         } else {
423                 int acctflag = VM_ACCOUNT;
424                 /*
425                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
426                  * if it's asked for.
427                  */
428                 if  ((shmflg & SHM_NORESERVE) &&
429                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
430                         acctflag = 0;
431                 file = shmem_file_setup(name, size, acctflag);
432         }
433         error = PTR_ERR(file);
434         if (IS_ERR(file))
435                 goto no_file;
436
437         error = -ENOSPC;
438         id = shm_addid(ns, shp);
439         if(id == -1) 
440                 goto no_id;
441
442         shp->shm_cprid = task_tgid_vnr(current);
443         shp->shm_lprid = 0;
444         shp->shm_atim = shp->shm_dtim = 0;
445         shp->shm_ctim = get_seconds();
446         shp->shm_segsz = size;
447         shp->shm_nattch = 0;
448         shp->shm_perm.id = shm_buildid(ns, id, shp->shm_perm.seq);
449         shp->shm_file = file;
450         /*
451          * shmid gets reported as "inode#" in /proc/pid/maps.
452          * proc-ps tools use this. Changing this will break them.
453          */
454         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
455
456         ns->shm_tot += numpages;
457         error = shp->shm_perm.id;
458         shm_unlock(shp);
459         return error;
460
461 no_id:
462         fput(file);
463 no_file:
464         security_shm_free(shp);
465         ipc_rcu_putref(shp);
466         return error;
467 }
468
469 /*
470  * Called with shm_ids.rw_mutex and ipcp locked.
471  */
472 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
473 {
474         struct shmid_kernel *shp;
475
476         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
477         return security_shm_associate(shp, shmflg);
478 }
479
480 /*
481  * Called with shm_ids.rw_mutex and ipcp locked.
482  */
483 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
484                                 struct ipc_params *params)
485 {
486         struct shmid_kernel *shp;
487
488         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
489         if (shp->shm_segsz < params->u.size)
490                 return -EINVAL;
491
492         return 0;
493 }
494
495 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
496 {
497         struct ipc_namespace *ns;
498         struct ipc_ops shm_ops;
499         struct ipc_params shm_params;
500
501         ns = current->nsproxy->ipc_ns;
502
503         shm_ops.getnew = newseg;
504         shm_ops.associate = shm_security;
505         shm_ops.more_checks = shm_more_checks;
506
507         shm_params.key = key;
508         shm_params.flg = shmflg;
509         shm_params.u.size = size;
510
511         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
512 }
513
514 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
515 {
516         switch(version) {
517         case IPC_64:
518                 return copy_to_user(buf, in, sizeof(*in));
519         case IPC_OLD:
520             {
521                 struct shmid_ds out;
522
523                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
524                 out.shm_segsz   = in->shm_segsz;
525                 out.shm_atime   = in->shm_atime;
526                 out.shm_dtime   = in->shm_dtime;
527                 out.shm_ctime   = in->shm_ctime;
528                 out.shm_cpid    = in->shm_cpid;
529                 out.shm_lpid    = in->shm_lpid;
530                 out.shm_nattch  = in->shm_nattch;
531
532                 return copy_to_user(buf, &out, sizeof(out));
533             }
534         default:
535                 return -EINVAL;
536         }
537 }
538
539 struct shm_setbuf {
540         uid_t   uid;
541         gid_t   gid;
542         mode_t  mode;
543 };      
544
545 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
546 {
547         switch(version) {
548         case IPC_64:
549             {
550                 struct shmid64_ds tbuf;
551
552                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
553                         return -EFAULT;
554
555                 out->uid        = tbuf.shm_perm.uid;
556                 out->gid        = tbuf.shm_perm.gid;
557                 out->mode       = tbuf.shm_perm.mode;
558
559                 return 0;
560             }
561         case IPC_OLD:
562             {
563                 struct shmid_ds tbuf_old;
564
565                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
566                         return -EFAULT;
567
568                 out->uid        = tbuf_old.shm_perm.uid;
569                 out->gid        = tbuf_old.shm_perm.gid;
570                 out->mode       = tbuf_old.shm_perm.mode;
571
572                 return 0;
573             }
574         default:
575                 return -EINVAL;
576         }
577 }
578
579 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
580 {
581         switch(version) {
582         case IPC_64:
583                 return copy_to_user(buf, in, sizeof(*in));
584         case IPC_OLD:
585             {
586                 struct shminfo out;
587
588                 if(in->shmmax > INT_MAX)
589                         out.shmmax = INT_MAX;
590                 else
591                         out.shmmax = (int)in->shmmax;
592
593                 out.shmmin      = in->shmmin;
594                 out.shmmni      = in->shmmni;
595                 out.shmseg      = in->shmseg;
596                 out.shmall      = in->shmall; 
597
598                 return copy_to_user(buf, &out, sizeof(out));
599             }
600         default:
601                 return -EINVAL;
602         }
603 }
604
605 /*
606  * Called with shm_ids.rw_mutex held as a reader
607  */
608 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
609                 unsigned long *swp)
610 {
611         int next_id;
612         int total, in_use;
613
614         *rss = 0;
615         *swp = 0;
616
617         in_use = shm_ids(ns).in_use;
618
619         for (total = 0, next_id = 0; total < in_use; next_id++) {
620                 struct shmid_kernel *shp;
621                 struct inode *inode;
622
623                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
624                 if (shp == NULL)
625                         continue;
626
627                 inode = shp->shm_file->f_path.dentry->d_inode;
628
629                 if (is_file_hugepages(shp->shm_file)) {
630                         struct address_space *mapping = inode->i_mapping;
631                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
632                 } else {
633                         struct shmem_inode_info *info = SHMEM_I(inode);
634                         spin_lock(&info->lock);
635                         *rss += inode->i_mapping->nrpages;
636                         *swp += info->swapped;
637                         spin_unlock(&info->lock);
638                 }
639
640                 total++;
641         }
642 }
643
644 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
645 {
646         struct shm_setbuf setbuf;
647         struct shmid_kernel *shp;
648         int err, version;
649         struct ipc_namespace *ns;
650
651         if (cmd < 0 || shmid < 0) {
652                 err = -EINVAL;
653                 goto out;
654         }
655
656         version = ipc_parse_version(&cmd);
657         ns = current->nsproxy->ipc_ns;
658
659         switch (cmd) { /* replace with proc interface ? */
660         case IPC_INFO:
661         {
662                 struct shminfo64 shminfo;
663
664                 err = security_shm_shmctl(NULL, cmd);
665                 if (err)
666                         return err;
667
668                 memset(&shminfo,0,sizeof(shminfo));
669                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
670                 shminfo.shmmax = ns->shm_ctlmax;
671                 shminfo.shmall = ns->shm_ctlall;
672
673                 shminfo.shmmin = SHMMIN;
674                 if(copy_shminfo_to_user (buf, &shminfo, version))
675                         return -EFAULT;
676
677                 down_read(&shm_ids(ns).rw_mutex);
678                 err = ipc_get_maxid(&shm_ids(ns));
679                 up_read(&shm_ids(ns).rw_mutex);
680
681                 if(err<0)
682                         err = 0;
683                 goto out;
684         }
685         case SHM_INFO:
686         {
687                 struct shm_info shm_info;
688
689                 err = security_shm_shmctl(NULL, cmd);
690                 if (err)
691                         return err;
692
693                 memset(&shm_info,0,sizeof(shm_info));
694                 down_read(&shm_ids(ns).rw_mutex);
695                 shm_info.used_ids = shm_ids(ns).in_use;
696                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
697                 shm_info.shm_tot = ns->shm_tot;
698                 shm_info.swap_attempts = 0;
699                 shm_info.swap_successes = 0;
700                 err = ipc_get_maxid(&shm_ids(ns));
701                 up_read(&shm_ids(ns).rw_mutex);
702                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
703                         err = -EFAULT;
704                         goto out;
705                 }
706
707                 err = err < 0 ? 0 : err;
708                 goto out;
709         }
710         case SHM_STAT:
711         case IPC_STAT:
712         {
713                 struct shmid64_ds tbuf;
714                 int result;
715
716                 if (!buf) {
717                         err = -EFAULT;
718                         goto out;
719                 }
720
721                 if (cmd == SHM_STAT) {
722                         shp = shm_lock(ns, shmid);
723                         if (IS_ERR(shp)) {
724                                 err = PTR_ERR(shp);
725                                 goto out;
726                         }
727                         result = shp->shm_perm.id;
728                 } else {
729                         shp = shm_lock_check(ns, shmid);
730                         if (IS_ERR(shp)) {
731                                 err = PTR_ERR(shp);
732                                 goto out;
733                         }
734                         result = 0;
735                 }
736                 err=-EACCES;
737                 if (ipcperms (&shp->shm_perm, S_IRUGO))
738                         goto out_unlock;
739                 err = security_shm_shmctl(shp, cmd);
740                 if (err)
741                         goto out_unlock;
742                 memset(&tbuf, 0, sizeof(tbuf));
743                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
744                 tbuf.shm_segsz  = shp->shm_segsz;
745                 tbuf.shm_atime  = shp->shm_atim;
746                 tbuf.shm_dtime  = shp->shm_dtim;
747                 tbuf.shm_ctime  = shp->shm_ctim;
748                 tbuf.shm_cpid   = shp->shm_cprid;
749                 tbuf.shm_lpid   = shp->shm_lprid;
750                 tbuf.shm_nattch = shp->shm_nattch;
751                 shm_unlock(shp);
752                 if(copy_shmid_to_user (buf, &tbuf, version))
753                         err = -EFAULT;
754                 else
755                         err = result;
756                 goto out;
757         }
758         case SHM_LOCK:
759         case SHM_UNLOCK:
760         {
761                 shp = shm_lock_check(ns, shmid);
762                 if (IS_ERR(shp)) {
763                         err = PTR_ERR(shp);
764                         goto out;
765                 }
766
767                 err = audit_ipc_obj(&(shp->shm_perm));
768                 if (err)
769                         goto out_unlock;
770
771                 if (!capable(CAP_IPC_LOCK)) {
772                         err = -EPERM;
773                         if (current->euid != shp->shm_perm.uid &&
774                             current->euid != shp->shm_perm.cuid)
775                                 goto out_unlock;
776                         if (cmd == SHM_LOCK &&
777                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
778                                 goto out_unlock;
779                 }
780
781                 err = security_shm_shmctl(shp, cmd);
782                 if (err)
783                         goto out_unlock;
784                 
785                 if(cmd==SHM_LOCK) {
786                         struct user_struct * user = current->user;
787                         if (!is_file_hugepages(shp->shm_file)) {
788                                 err = shmem_lock(shp->shm_file, 1, user);
789                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
790                                         shp->shm_perm.mode |= SHM_LOCKED;
791                                         shp->mlock_user = user;
792                                 }
793                         }
794                 } else if (!is_file_hugepages(shp->shm_file)) {
795                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
796                         shp->shm_perm.mode &= ~SHM_LOCKED;
797                         shp->mlock_user = NULL;
798                 }
799                 shm_unlock(shp);
800                 goto out;
801         }
802         case IPC_RMID:
803         {
804                 /*
805                  *      We cannot simply remove the file. The SVID states
806                  *      that the block remains until the last person
807                  *      detaches from it, then is deleted. A shmat() on
808                  *      an RMID segment is legal in older Linux and if 
809                  *      we change it apps break...
810                  *
811                  *      Instead we set a destroyed flag, and then blow
812                  *      the name away when the usage hits zero.
813                  */
814                 down_write(&shm_ids(ns).rw_mutex);
815                 shp = shm_lock_check_down(ns, shmid);
816                 if (IS_ERR(shp)) {
817                         err = PTR_ERR(shp);
818                         goto out_up;
819                 }
820
821                 err = audit_ipc_obj(&(shp->shm_perm));
822                 if (err)
823                         goto out_unlock_up;
824
825                 if (current->euid != shp->shm_perm.uid &&
826                     current->euid != shp->shm_perm.cuid && 
827                     !capable(CAP_SYS_ADMIN)) {
828                         err=-EPERM;
829                         goto out_unlock_up;
830                 }
831
832                 err = security_shm_shmctl(shp, cmd);
833                 if (err)
834                         goto out_unlock_up;
835
836                 do_shm_rmid(ns, shp);
837                 up_write(&shm_ids(ns).rw_mutex);
838                 goto out;
839         }
840
841         case IPC_SET:
842         {
843                 if (!buf) {
844                         err = -EFAULT;
845                         goto out;
846                 }
847
848                 if (copy_shmid_from_user (&setbuf, buf, version)) {
849                         err = -EFAULT;
850                         goto out;
851                 }
852                 down_write(&shm_ids(ns).rw_mutex);
853                 shp = shm_lock_check_down(ns, shmid);
854                 if (IS_ERR(shp)) {
855                         err = PTR_ERR(shp);
856                         goto out_up;
857                 }
858                 err = audit_ipc_obj(&(shp->shm_perm));
859                 if (err)
860                         goto out_unlock_up;
861                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
862                 if (err)
863                         goto out_unlock_up;
864                 err=-EPERM;
865                 if (current->euid != shp->shm_perm.uid &&
866                     current->euid != shp->shm_perm.cuid && 
867                     !capable(CAP_SYS_ADMIN)) {
868                         goto out_unlock_up;
869                 }
870
871                 err = security_shm_shmctl(shp, cmd);
872                 if (err)
873                         goto out_unlock_up;
874                 
875                 shp->shm_perm.uid = setbuf.uid;
876                 shp->shm_perm.gid = setbuf.gid;
877                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
878                         | (setbuf.mode & S_IRWXUGO);
879                 shp->shm_ctim = get_seconds();
880                 break;
881         }
882
883         default:
884                 err = -EINVAL;
885                 goto out;
886         }
887
888         err = 0;
889 out_unlock_up:
890         shm_unlock(shp);
891 out_up:
892         up_write(&shm_ids(ns).rw_mutex);
893         goto out;
894 out_unlock:
895         shm_unlock(shp);
896 out:
897         return err;
898 }
899
900 /*
901  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
902  *
903  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
904  * "raddr" thing points to kernel space, and there has to be a wrapper around
905  * this.
906  */
907 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
908 {
909         struct shmid_kernel *shp;
910         unsigned long addr;
911         unsigned long size;
912         struct file * file;
913         int    err;
914         unsigned long flags;
915         unsigned long prot;
916         int acc_mode;
917         unsigned long user_addr;
918         struct ipc_namespace *ns;
919         struct shm_file_data *sfd;
920         struct path path;
921         mode_t f_mode;
922
923         err = -EINVAL;
924         if (shmid < 0)
925                 goto out;
926         else if ((addr = (ulong)shmaddr)) {
927                 if (addr & (SHMLBA-1)) {
928                         if (shmflg & SHM_RND)
929                                 addr &= ~(SHMLBA-1);       /* round down */
930                         else
931 #ifndef __ARCH_FORCE_SHMLBA
932                                 if (addr & ~PAGE_MASK)
933 #endif
934                                         goto out;
935                 }
936                 flags = MAP_SHARED | MAP_FIXED;
937         } else {
938                 if ((shmflg & SHM_REMAP))
939                         goto out;
940
941                 flags = MAP_SHARED;
942         }
943
944         if (shmflg & SHM_RDONLY) {
945                 prot = PROT_READ;
946                 acc_mode = S_IRUGO;
947                 f_mode = FMODE_READ;
948         } else {
949                 prot = PROT_READ | PROT_WRITE;
950                 acc_mode = S_IRUGO | S_IWUGO;
951                 f_mode = FMODE_READ | FMODE_WRITE;
952         }
953         if (shmflg & SHM_EXEC) {
954                 prot |= PROT_EXEC;
955                 acc_mode |= S_IXUGO;
956         }
957
958         /*
959          * We cannot rely on the fs check since SYSV IPC does have an
960          * additional creator id...
961          */
962         ns = current->nsproxy->ipc_ns;
963         shp = shm_lock_check(ns, shmid);
964         if (IS_ERR(shp)) {
965                 err = PTR_ERR(shp);
966                 goto out;
967         }
968
969         err = -EACCES;
970         if (ipcperms(&shp->shm_perm, acc_mode))
971                 goto out_unlock;
972
973         err = security_shm_shmat(shp, shmaddr, shmflg);
974         if (err)
975                 goto out_unlock;
976
977         path.dentry = dget(shp->shm_file->f_path.dentry);
978         path.mnt    = shp->shm_file->f_path.mnt;
979         shp->shm_nattch++;
980         size = i_size_read(path.dentry->d_inode);
981         shm_unlock(shp);
982
983         err = -ENOMEM;
984         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
985         if (!sfd)
986                 goto out_put_dentry;
987
988         err = -ENOMEM;
989
990         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
991         if (!file)
992                 goto out_free;
993
994         file->private_data = sfd;
995         file->f_mapping = shp->shm_file->f_mapping;
996         sfd->id = shp->shm_perm.id;
997         sfd->ns = get_ipc_ns(ns);
998         sfd->file = shp->shm_file;
999         sfd->vm_ops = NULL;
1000
1001         down_write(&current->mm->mmap_sem);
1002         if (addr && !(shmflg & SHM_REMAP)) {
1003                 err = -EINVAL;
1004                 if (find_vma_intersection(current->mm, addr, addr + size))
1005                         goto invalid;
1006                 /*
1007                  * If shm segment goes below stack, make sure there is some
1008                  * space left for the stack to grow (at least 4 pages).
1009                  */
1010                 if (addr < current->mm->start_stack &&
1011                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1012                         goto invalid;
1013         }
1014                 
1015         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1016         *raddr = user_addr;
1017         err = 0;
1018         if (IS_ERR_VALUE(user_addr))
1019                 err = (long)user_addr;
1020 invalid:
1021         up_write(&current->mm->mmap_sem);
1022
1023         fput(file);
1024
1025 out_nattch:
1026         down_write(&shm_ids(ns).rw_mutex);
1027         shp = shm_lock_down(ns, shmid);
1028         BUG_ON(IS_ERR(shp));
1029         shp->shm_nattch--;
1030         if(shp->shm_nattch == 0 &&
1031            shp->shm_perm.mode & SHM_DEST)
1032                 shm_destroy(ns, shp);
1033         else
1034                 shm_unlock(shp);
1035         up_write(&shm_ids(ns).rw_mutex);
1036
1037 out:
1038         return err;
1039
1040 out_unlock:
1041         shm_unlock(shp);
1042         goto out;
1043
1044 out_free:
1045         kfree(sfd);
1046 out_put_dentry:
1047         dput(path.dentry);
1048         goto out_nattch;
1049 }
1050
1051 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1052 {
1053         unsigned long ret;
1054         long err;
1055
1056         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1057         if (err)
1058                 return err;
1059         force_successful_syscall_return();
1060         return (long)ret;
1061 }
1062
1063 /*
1064  * detach and kill segment if marked destroyed.
1065  * The work is done in shm_close.
1066  */
1067 asmlinkage long sys_shmdt(char __user *shmaddr)
1068 {
1069         struct mm_struct *mm = current->mm;
1070         struct vm_area_struct *vma, *next;
1071         unsigned long addr = (unsigned long)shmaddr;
1072         loff_t size = 0;
1073         int retval = -EINVAL;
1074
1075         if (addr & ~PAGE_MASK)
1076                 return retval;
1077
1078         down_write(&mm->mmap_sem);
1079
1080         /*
1081          * This function tries to be smart and unmap shm segments that
1082          * were modified by partial mlock or munmap calls:
1083          * - It first determines the size of the shm segment that should be
1084          *   unmapped: It searches for a vma that is backed by shm and that
1085          *   started at address shmaddr. It records it's size and then unmaps
1086          *   it.
1087          * - Then it unmaps all shm vmas that started at shmaddr and that
1088          *   are within the initially determined size.
1089          * Errors from do_munmap are ignored: the function only fails if
1090          * it's called with invalid parameters or if it's called to unmap
1091          * a part of a vma. Both calls in this function are for full vmas,
1092          * the parameters are directly copied from the vma itself and always
1093          * valid - therefore do_munmap cannot fail. (famous last words?)
1094          */
1095         /*
1096          * If it had been mremap()'d, the starting address would not
1097          * match the usual checks anyway. So assume all vma's are
1098          * above the starting address given.
1099          */
1100         vma = find_vma(mm, addr);
1101
1102         while (vma) {
1103                 next = vma->vm_next;
1104
1105                 /*
1106                  * Check if the starting address would match, i.e. it's
1107                  * a fragment created by mprotect() and/or munmap(), or it
1108                  * otherwise it starts at this address with no hassles.
1109                  */
1110                 if ((vma->vm_ops == &shm_vm_ops) &&
1111                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1112
1113
1114                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1115                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1116                         /*
1117                          * We discovered the size of the shm segment, so
1118                          * break out of here and fall through to the next
1119                          * loop that uses the size information to stop
1120                          * searching for matching vma's.
1121                          */
1122                         retval = 0;
1123                         vma = next;
1124                         break;
1125                 }
1126                 vma = next;
1127         }
1128
1129         /*
1130          * We need look no further than the maximum address a fragment
1131          * could possibly have landed at. Also cast things to loff_t to
1132          * prevent overflows and make comparisions vs. equal-width types.
1133          */
1134         size = PAGE_ALIGN(size);
1135         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1136                 next = vma->vm_next;
1137
1138                 /* finding a matching vma now does not alter retval */
1139                 if ((vma->vm_ops == &shm_vm_ops) &&
1140                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1141
1142                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1143                 vma = next;
1144         }
1145
1146         up_write(&mm->mmap_sem);
1147         return retval;
1148 }
1149
1150 #ifdef CONFIG_PROC_FS
1151 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1152 {
1153         struct shmid_kernel *shp = it;
1154         char *format;
1155
1156 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1157 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1158
1159         if (sizeof(size_t) <= sizeof(int))
1160                 format = SMALL_STRING;
1161         else
1162                 format = BIG_STRING;
1163         return seq_printf(s, format,
1164                           shp->shm_perm.key,
1165                           shp->shm_perm.id,
1166                           shp->shm_perm.mode,
1167                           shp->shm_segsz,
1168                           shp->shm_cprid,
1169                           shp->shm_lprid,
1170                           shp->shm_nattch,
1171                           shp->shm_perm.uid,
1172                           shp->shm_perm.gid,
1173                           shp->shm_perm.cuid,
1174                           shp->shm_perm.cgid,
1175                           shp->shm_atim,
1176                           shp->shm_dtim,
1177                           shp->shm_ctim);
1178 }
1179 #endif