[PATCH] Fix get_unmapped_area and fsync for hugetlb shm segments
[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/mutex.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_lock(ns, id)                \
63         ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
64 #define shm_unlock(shp)                 \
65         ipc_unlock(&(shp)->shm_perm)
66 #define shm_get(ns, id)                 \
67         ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
68 #define shm_buildid(ns, id, seq)        \
69         ipc_buildid(&shm_ids(ns), id, seq)
70
71 static int newseg (struct ipc_namespace *ns, key_t key,
72                 int shmflg, size_t size);
73 static void shm_open(struct vm_area_struct *vma);
74 static void shm_close(struct vm_area_struct *vma);
75 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
76 #ifdef CONFIG_PROC_FS
77 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
78 #endif
79
80 static void __ipc_init __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
81 {
82         ns->ids[IPC_SHM_IDS] = ids;
83         ns->shm_ctlmax = SHMMAX;
84         ns->shm_ctlall = SHMALL;
85         ns->shm_ctlmni = SHMMNI;
86         ns->shm_tot = 0;
87         ipc_init_ids(ids, 1);
88 }
89
90 static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
91 {
92         if (shp->shm_nattch){
93                 shp->shm_perm.mode |= SHM_DEST;
94                 /* Do not find it any more */
95                 shp->shm_perm.key = IPC_PRIVATE;
96                 shm_unlock(shp);
97         } else
98                 shm_destroy(ns, shp);
99 }
100
101 #ifdef CONFIG_IPC_NS
102 int shm_init_ns(struct ipc_namespace *ns)
103 {
104         struct ipc_ids *ids;
105
106         ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
107         if (ids == NULL)
108                 return -ENOMEM;
109
110         __shm_init_ns(ns, ids);
111         return 0;
112 }
113
114 void shm_exit_ns(struct ipc_namespace *ns)
115 {
116         int i;
117         struct shmid_kernel *shp;
118
119         mutex_lock(&shm_ids(ns).mutex);
120         for (i = 0; i <= shm_ids(ns).max_id; i++) {
121                 shp = shm_lock(ns, i);
122                 if (shp == NULL)
123                         continue;
124
125                 do_shm_rmid(ns, shp);
126         }
127         mutex_unlock(&shm_ids(ns).mutex);
128
129         ipc_fini_ids(ns->ids[IPC_SHM_IDS]);
130         kfree(ns->ids[IPC_SHM_IDS]);
131         ns->ids[IPC_SHM_IDS] = NULL;
132 }
133 #endif
134
135 void __init shm_init (void)
136 {
137         __shm_init_ns(&init_ipc_ns, &init_shm_ids);
138         ipc_init_proc_interface("sysvipc/shm",
139                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
140                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
141 }
142
143 static inline int shm_checkid(struct ipc_namespace *ns,
144                 struct shmid_kernel *s, int id)
145 {
146         if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
147                 return -EIDRM;
148         return 0;
149 }
150
151 static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
152 {
153         return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
154 }
155
156 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
157 {
158         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
159 }
160
161
162
163 /* This is called by fork, once for every shm attach. */
164 static void shm_open(struct vm_area_struct *vma)
165 {
166         struct file *file = vma->vm_file;
167         struct shm_file_data *sfd = shm_file_data(file);
168         struct shmid_kernel *shp;
169
170         shp = shm_lock(sfd->ns, sfd->id);
171         BUG_ON(!shp);
172         shp->shm_atim = get_seconds();
173         shp->shm_lprid = current->tgid;
174         shp->shm_nattch++;
175         shm_unlock(shp);
176 }
177
178 /*
179  * shm_destroy - free the struct shmid_kernel
180  *
181  * @shp: struct to free
182  *
183  * It has to be called with shp and shm_ids.mutex locked,
184  * but returns with shp unlocked and freed.
185  */
186 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
187 {
188         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
189         shm_rmid(ns, shp->id);
190         shm_unlock(shp);
191         if (!is_file_hugepages(shp->shm_file))
192                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
193         else
194                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
195                                                 shp->mlock_user);
196         fput (shp->shm_file);
197         security_shm_free(shp);
198         ipc_rcu_putref(shp);
199 }
200
201 /*
202  * remove the attach descriptor vma.
203  * free memory for segment if it is marked destroyed.
204  * The descriptor has already been removed from the current->mm->mmap list
205  * and will later be kfree()d.
206  */
207 static void shm_close(struct vm_area_struct *vma)
208 {
209         struct file * file = vma->vm_file;
210         struct shm_file_data *sfd = shm_file_data(file);
211         struct shmid_kernel *shp;
212         struct ipc_namespace *ns = sfd->ns;
213
214         mutex_lock(&shm_ids(ns).mutex);
215         /* remove from the list of attaches of the shm segment */
216         shp = shm_lock(ns, sfd->id);
217         BUG_ON(!shp);
218         shp->shm_lprid = current->tgid;
219         shp->shm_dtim = get_seconds();
220         shp->shm_nattch--;
221         if(shp->shm_nattch == 0 &&
222            shp->shm_perm.mode & SHM_DEST)
223                 shm_destroy(ns, shp);
224         else
225                 shm_unlock(shp);
226         mutex_unlock(&shm_ids(ns).mutex);
227 }
228
229 static struct page *shm_nopage(struct vm_area_struct *vma,
230                                unsigned long address, int *type)
231 {
232         struct file *file = vma->vm_file;
233         struct shm_file_data *sfd = shm_file_data(file);
234
235         return sfd->vm_ops->nopage(vma, address, type);
236 }
237
238 #ifdef CONFIG_NUMA
239 int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
240 {
241         struct file *file = vma->vm_file;
242         struct shm_file_data *sfd = shm_file_data(file);
243         int err = 0;
244         if (sfd->vm_ops->set_policy)
245                 err = sfd->vm_ops->set_policy(vma, new);
246         return err;
247 }
248
249 struct mempolicy *shm_get_policy(struct vm_area_struct *vma, unsigned long addr)
250 {
251         struct file *file = vma->vm_file;
252         struct shm_file_data *sfd = shm_file_data(file);
253         struct mempolicy *pol = NULL;
254
255         if (sfd->vm_ops->get_policy)
256                 pol = sfd->vm_ops->get_policy(vma, addr);
257         else
258                 pol = vma->vm_policy;
259         return pol;
260 }
261 #endif
262
263 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
264 {
265         struct shm_file_data *sfd = shm_file_data(file);
266         int ret;
267
268         ret = sfd->file->f_op->mmap(sfd->file, vma);
269         if (ret != 0)
270                 return ret;
271         sfd->vm_ops = vma->vm_ops;
272         vma->vm_ops = &shm_vm_ops;
273         shm_open(vma);
274
275         return ret;
276 }
277
278 static int shm_release(struct inode *ino, struct file *file)
279 {
280         struct shm_file_data *sfd = shm_file_data(file);
281
282         put_ipc_ns(sfd->ns);
283         shm_file_data(file) = NULL;
284         kfree(sfd);
285         return 0;
286 }
287
288 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
289 {
290         int (*fsync) (struct file *, struct dentry *, int datasync);
291         struct shm_file_data *sfd = shm_file_data(file);
292         int ret = -EINVAL;
293
294         fsync = sfd->file->f_op->fsync;
295         if (fsync)
296                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
297         return ret;
298 }
299
300 static unsigned long shm_get_unmapped_area(struct file *file,
301         unsigned long addr, unsigned long len, unsigned long pgoff,
302         unsigned long flags)
303 {
304         struct shm_file_data *sfd = shm_file_data(file);
305         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
306 }
307
308 int is_file_shm_hugepages(struct file *file)
309 {
310         int ret = 0;
311
312         if (file->f_op == &shm_file_operations) {
313                 struct shm_file_data *sfd;
314                 sfd = shm_file_data(file);
315                 ret = is_file_hugepages(sfd->file);
316         }
317         return ret;
318 }
319
320 static const struct file_operations shm_file_operations = {
321         .mmap           = shm_mmap,
322         .fsync          = shm_fsync,
323         .release        = shm_release,
324         .get_unmapped_area      = shm_get_unmapped_area,
325 };
326
327 static struct vm_operations_struct shm_vm_ops = {
328         .open   = shm_open,     /* callback for a new vm-area open */
329         .close  = shm_close,    /* callback for when the vm-area is released */
330         .nopage = shm_nopage,
331 #if defined(CONFIG_NUMA)
332         .set_policy = shm_set_policy,
333         .get_policy = shm_get_policy,
334 #endif
335 };
336
337 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
338 {
339         int error;
340         struct shmid_kernel *shp;
341         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
342         struct file * file;
343         char name[13];
344         int id;
345
346         if (size < SHMMIN || size > ns->shm_ctlmax)
347                 return -EINVAL;
348
349         if (ns->shm_tot + numpages > ns->shm_ctlall)
350                 return -ENOSPC;
351
352         shp = ipc_rcu_alloc(sizeof(*shp));
353         if (!shp)
354                 return -ENOMEM;
355
356         shp->shm_perm.key = key;
357         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
358         shp->mlock_user = NULL;
359
360         shp->shm_perm.security = NULL;
361         error = security_shm_alloc(shp);
362         if (error) {
363                 ipc_rcu_putref(shp);
364                 return error;
365         }
366
367         if (shmflg & SHM_HUGETLB) {
368                 /* hugetlb_zero_setup takes care of mlock user accounting */
369                 file = hugetlb_zero_setup(size);
370                 shp->mlock_user = current->user;
371         } else {
372                 int acctflag = VM_ACCOUNT;
373                 /*
374                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
375                  * if it's asked for.
376                  */
377                 if  ((shmflg & SHM_NORESERVE) &&
378                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
379                         acctflag = 0;
380                 sprintf (name, "SYSV%08x", key);
381                 file = shmem_file_setup(name, size, acctflag);
382         }
383         error = PTR_ERR(file);
384         if (IS_ERR(file))
385                 goto no_file;
386
387         error = -ENOSPC;
388         id = shm_addid(ns, shp);
389         if(id == -1) 
390                 goto no_id;
391
392         shp->shm_cprid = current->tgid;
393         shp->shm_lprid = 0;
394         shp->shm_atim = shp->shm_dtim = 0;
395         shp->shm_ctim = get_seconds();
396         shp->shm_segsz = size;
397         shp->shm_nattch = 0;
398         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
399         shp->shm_file = file;
400
401         ns->shm_tot += numpages;
402         shm_unlock(shp);
403         return shp->id;
404
405 no_id:
406         fput(file);
407 no_file:
408         security_shm_free(shp);
409         ipc_rcu_putref(shp);
410         return error;
411 }
412
413 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
414 {
415         struct shmid_kernel *shp;
416         int err, id = 0;
417         struct ipc_namespace *ns;
418
419         ns = current->nsproxy->ipc_ns;
420
421         mutex_lock(&shm_ids(ns).mutex);
422         if (key == IPC_PRIVATE) {
423                 err = newseg(ns, key, shmflg, size);
424         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
425                 if (!(shmflg & IPC_CREAT))
426                         err = -ENOENT;
427                 else
428                         err = newseg(ns, key, shmflg, size);
429         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
430                 err = -EEXIST;
431         } else {
432                 shp = shm_lock(ns, id);
433                 BUG_ON(shp==NULL);
434                 if (shp->shm_segsz < size)
435                         err = -EINVAL;
436                 else if (ipcperms(&shp->shm_perm, shmflg))
437                         err = -EACCES;
438                 else {
439                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
440                         err = security_shm_associate(shp, shmflg);
441                         if (!err)
442                                 err = shmid;
443                 }
444                 shm_unlock(shp);
445         }
446         mutex_unlock(&shm_ids(ns).mutex);
447
448         return err;
449 }
450
451 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
452 {
453         switch(version) {
454         case IPC_64:
455                 return copy_to_user(buf, in, sizeof(*in));
456         case IPC_OLD:
457             {
458                 struct shmid_ds out;
459
460                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
461                 out.shm_segsz   = in->shm_segsz;
462                 out.shm_atime   = in->shm_atime;
463                 out.shm_dtime   = in->shm_dtime;
464                 out.shm_ctime   = in->shm_ctime;
465                 out.shm_cpid    = in->shm_cpid;
466                 out.shm_lpid    = in->shm_lpid;
467                 out.shm_nattch  = in->shm_nattch;
468
469                 return copy_to_user(buf, &out, sizeof(out));
470             }
471         default:
472                 return -EINVAL;
473         }
474 }
475
476 struct shm_setbuf {
477         uid_t   uid;
478         gid_t   gid;
479         mode_t  mode;
480 };      
481
482 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
483 {
484         switch(version) {
485         case IPC_64:
486             {
487                 struct shmid64_ds tbuf;
488
489                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
490                         return -EFAULT;
491
492                 out->uid        = tbuf.shm_perm.uid;
493                 out->gid        = tbuf.shm_perm.gid;
494                 out->mode       = tbuf.shm_perm.mode;
495
496                 return 0;
497             }
498         case IPC_OLD:
499             {
500                 struct shmid_ds tbuf_old;
501
502                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
503                         return -EFAULT;
504
505                 out->uid        = tbuf_old.shm_perm.uid;
506                 out->gid        = tbuf_old.shm_perm.gid;
507                 out->mode       = tbuf_old.shm_perm.mode;
508
509                 return 0;
510             }
511         default:
512                 return -EINVAL;
513         }
514 }
515
516 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
517 {
518         switch(version) {
519         case IPC_64:
520                 return copy_to_user(buf, in, sizeof(*in));
521         case IPC_OLD:
522             {
523                 struct shminfo out;
524
525                 if(in->shmmax > INT_MAX)
526                         out.shmmax = INT_MAX;
527                 else
528                         out.shmmax = (int)in->shmmax;
529
530                 out.shmmin      = in->shmmin;
531                 out.shmmni      = in->shmmni;
532                 out.shmseg      = in->shmseg;
533                 out.shmall      = in->shmall; 
534
535                 return copy_to_user(buf, &out, sizeof(out));
536             }
537         default:
538                 return -EINVAL;
539         }
540 }
541
542 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
543                 unsigned long *swp)
544 {
545         int i;
546
547         *rss = 0;
548         *swp = 0;
549
550         for (i = 0; i <= shm_ids(ns).max_id; i++) {
551                 struct shmid_kernel *shp;
552                 struct inode *inode;
553
554                 shp = shm_get(ns, i);
555                 if(!shp)
556                         continue;
557
558                 inode = shp->shm_file->f_path.dentry->d_inode;
559
560                 if (is_file_hugepages(shp->shm_file)) {
561                         struct address_space *mapping = inode->i_mapping;
562                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
563                 } else {
564                         struct shmem_inode_info *info = SHMEM_I(inode);
565                         spin_lock(&info->lock);
566                         *rss += inode->i_mapping->nrpages;
567                         *swp += info->swapped;
568                         spin_unlock(&info->lock);
569                 }
570         }
571 }
572
573 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
574 {
575         struct shm_setbuf setbuf;
576         struct shmid_kernel *shp;
577         int err, version;
578         struct ipc_namespace *ns;
579
580         if (cmd < 0 || shmid < 0) {
581                 err = -EINVAL;
582                 goto out;
583         }
584
585         version = ipc_parse_version(&cmd);
586         ns = current->nsproxy->ipc_ns;
587
588         switch (cmd) { /* replace with proc interface ? */
589         case IPC_INFO:
590         {
591                 struct shminfo64 shminfo;
592
593                 err = security_shm_shmctl(NULL, cmd);
594                 if (err)
595                         return err;
596
597                 memset(&shminfo,0,sizeof(shminfo));
598                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
599                 shminfo.shmmax = ns->shm_ctlmax;
600                 shminfo.shmall = ns->shm_ctlall;
601
602                 shminfo.shmmin = SHMMIN;
603                 if(copy_shminfo_to_user (buf, &shminfo, version))
604                         return -EFAULT;
605                 /* reading a integer is always atomic */
606                 err= shm_ids(ns).max_id;
607                 if(err<0)
608                         err = 0;
609                 goto out;
610         }
611         case SHM_INFO:
612         {
613                 struct shm_info shm_info;
614
615                 err = security_shm_shmctl(NULL, cmd);
616                 if (err)
617                         return err;
618
619                 memset(&shm_info,0,sizeof(shm_info));
620                 mutex_lock(&shm_ids(ns).mutex);
621                 shm_info.used_ids = shm_ids(ns).in_use;
622                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
623                 shm_info.shm_tot = ns->shm_tot;
624                 shm_info.swap_attempts = 0;
625                 shm_info.swap_successes = 0;
626                 err = shm_ids(ns).max_id;
627                 mutex_unlock(&shm_ids(ns).mutex);
628                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
629                         err = -EFAULT;
630                         goto out;
631                 }
632
633                 err = err < 0 ? 0 : err;
634                 goto out;
635         }
636         case SHM_STAT:
637         case IPC_STAT:
638         {
639                 struct shmid64_ds tbuf;
640                 int result;
641                 memset(&tbuf, 0, sizeof(tbuf));
642                 shp = shm_lock(ns, shmid);
643                 if(shp==NULL) {
644                         err = -EINVAL;
645                         goto out;
646                 } else if(cmd==SHM_STAT) {
647                         err = -EINVAL;
648                         if (shmid > shm_ids(ns).max_id)
649                                 goto out_unlock;
650                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
651                 } else {
652                         err = shm_checkid(ns, shp,shmid);
653                         if(err)
654                                 goto out_unlock;
655                         result = 0;
656                 }
657                 err=-EACCES;
658                 if (ipcperms (&shp->shm_perm, S_IRUGO))
659                         goto out_unlock;
660                 err = security_shm_shmctl(shp, cmd);
661                 if (err)
662                         goto out_unlock;
663                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
664                 tbuf.shm_segsz  = shp->shm_segsz;
665                 tbuf.shm_atime  = shp->shm_atim;
666                 tbuf.shm_dtime  = shp->shm_dtim;
667                 tbuf.shm_ctime  = shp->shm_ctim;
668                 tbuf.shm_cpid   = shp->shm_cprid;
669                 tbuf.shm_lpid   = shp->shm_lprid;
670                 tbuf.shm_nattch = shp->shm_nattch;
671                 shm_unlock(shp);
672                 if(copy_shmid_to_user (buf, &tbuf, version))
673                         err = -EFAULT;
674                 else
675                         err = result;
676                 goto out;
677         }
678         case SHM_LOCK:
679         case SHM_UNLOCK:
680         {
681                 shp = shm_lock(ns, shmid);
682                 if(shp==NULL) {
683                         err = -EINVAL;
684                         goto out;
685                 }
686                 err = shm_checkid(ns, shp,shmid);
687                 if(err)
688                         goto out_unlock;
689
690                 err = audit_ipc_obj(&(shp->shm_perm));
691                 if (err)
692                         goto out_unlock;
693
694                 if (!capable(CAP_IPC_LOCK)) {
695                         err = -EPERM;
696                         if (current->euid != shp->shm_perm.uid &&
697                             current->euid != shp->shm_perm.cuid)
698                                 goto out_unlock;
699                         if (cmd == SHM_LOCK &&
700                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
701                                 goto out_unlock;
702                 }
703
704                 err = security_shm_shmctl(shp, cmd);
705                 if (err)
706                         goto out_unlock;
707                 
708                 if(cmd==SHM_LOCK) {
709                         struct user_struct * user = current->user;
710                         if (!is_file_hugepages(shp->shm_file)) {
711                                 err = shmem_lock(shp->shm_file, 1, user);
712                                 if (!err) {
713                                         shp->shm_perm.mode |= SHM_LOCKED;
714                                         shp->mlock_user = user;
715                                 }
716                         }
717                 } else if (!is_file_hugepages(shp->shm_file)) {
718                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
719                         shp->shm_perm.mode &= ~SHM_LOCKED;
720                         shp->mlock_user = NULL;
721                 }
722                 shm_unlock(shp);
723                 goto out;
724         }
725         case IPC_RMID:
726         {
727                 /*
728                  *      We cannot simply remove the file. The SVID states
729                  *      that the block remains until the last person
730                  *      detaches from it, then is deleted. A shmat() on
731                  *      an RMID segment is legal in older Linux and if 
732                  *      we change it apps break...
733                  *
734                  *      Instead we set a destroyed flag, and then blow
735                  *      the name away when the usage hits zero.
736                  */
737                 mutex_lock(&shm_ids(ns).mutex);
738                 shp = shm_lock(ns, shmid);
739                 err = -EINVAL;
740                 if (shp == NULL) 
741                         goto out_up;
742                 err = shm_checkid(ns, shp, shmid);
743                 if(err)
744                         goto out_unlock_up;
745
746                 err = audit_ipc_obj(&(shp->shm_perm));
747                 if (err)
748                         goto out_unlock_up;
749
750                 if (current->euid != shp->shm_perm.uid &&
751                     current->euid != shp->shm_perm.cuid && 
752                     !capable(CAP_SYS_ADMIN)) {
753                         err=-EPERM;
754                         goto out_unlock_up;
755                 }
756
757                 err = security_shm_shmctl(shp, cmd);
758                 if (err)
759                         goto out_unlock_up;
760
761                 do_shm_rmid(ns, shp);
762                 mutex_unlock(&shm_ids(ns).mutex);
763                 goto out;
764         }
765
766         case IPC_SET:
767         {
768                 if (copy_shmid_from_user (&setbuf, buf, version)) {
769                         err = -EFAULT;
770                         goto out;
771                 }
772                 mutex_lock(&shm_ids(ns).mutex);
773                 shp = shm_lock(ns, shmid);
774                 err=-EINVAL;
775                 if(shp==NULL)
776                         goto out_up;
777                 err = shm_checkid(ns, shp,shmid);
778                 if(err)
779                         goto out_unlock_up;
780                 err = audit_ipc_obj(&(shp->shm_perm));
781                 if (err)
782                         goto out_unlock_up;
783                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
784                 if (err)
785                         goto out_unlock_up;
786                 err=-EPERM;
787                 if (current->euid != shp->shm_perm.uid &&
788                     current->euid != shp->shm_perm.cuid && 
789                     !capable(CAP_SYS_ADMIN)) {
790                         goto out_unlock_up;
791                 }
792
793                 err = security_shm_shmctl(shp, cmd);
794                 if (err)
795                         goto out_unlock_up;
796                 
797                 shp->shm_perm.uid = setbuf.uid;
798                 shp->shm_perm.gid = setbuf.gid;
799                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
800                         | (setbuf.mode & S_IRWXUGO);
801                 shp->shm_ctim = get_seconds();
802                 break;
803         }
804
805         default:
806                 err = -EINVAL;
807                 goto out;
808         }
809
810         err = 0;
811 out_unlock_up:
812         shm_unlock(shp);
813 out_up:
814         mutex_unlock(&shm_ids(ns).mutex);
815         goto out;
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         mode_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(ns, shmid);
886         if(shp == NULL)
887                 goto out;
888
889         err = shm_checkid(ns, shp,shmid);
890         if (err)
891                 goto out_unlock;
892
893         err = -EACCES;
894         if (ipcperms(&shp->shm_perm, acc_mode))
895                 goto out_unlock;
896
897         err = security_shm_shmat(shp, shmaddr, shmflg);
898         if (err)
899                 goto out_unlock;
900
901         path.dentry = dget(shp->shm_file->f_path.dentry);
902         path.mnt    = mntget(shp->shm_file->f_path.mnt);
903         shp->shm_nattch++;
904         size = i_size_read(path.dentry->d_inode);
905         shm_unlock(shp);
906
907         err = -ENOMEM;
908         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
909         if (!sfd)
910                 goto out_put_path;
911
912         err = -ENOMEM;
913         file = get_empty_filp();
914         if (!file)
915                 goto out_free;
916
917         file->f_op = &shm_file_operations;
918         file->private_data = sfd;
919         file->f_path = path;
920         file->f_mapping = shp->shm_file->f_mapping;
921         file->f_mode = f_mode;
922         sfd->id = shp->id;
923         sfd->ns = get_ipc_ns(ns);
924         sfd->file = shp->shm_file;
925         sfd->vm_ops = NULL;
926
927         down_write(&current->mm->mmap_sem);
928         if (addr && !(shmflg & SHM_REMAP)) {
929                 err = -EINVAL;
930                 if (find_vma_intersection(current->mm, addr, addr + size))
931                         goto invalid;
932                 /*
933                  * If shm segment goes below stack, make sure there is some
934                  * space left for the stack to grow (at least 4 pages).
935                  */
936                 if (addr < current->mm->start_stack &&
937                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
938                         goto invalid;
939         }
940                 
941         user_addr = do_mmap (file, addr, size, prot, flags, 0);
942         *raddr = user_addr;
943         err = 0;
944         if (IS_ERR_VALUE(user_addr))
945                 err = (long)user_addr;
946 invalid:
947         up_write(&current->mm->mmap_sem);
948
949         fput(file);
950
951 out_nattch:
952         mutex_lock(&shm_ids(ns).mutex);
953         shp = shm_lock(ns, shmid);
954         BUG_ON(!shp);
955         shp->shm_nattch--;
956         if(shp->shm_nattch == 0 &&
957            shp->shm_perm.mode & SHM_DEST)
958                 shm_destroy(ns, shp);
959         else
960                 shm_unlock(shp);
961         mutex_unlock(&shm_ids(ns).mutex);
962
963 out:
964         return err;
965
966 out_unlock:
967         shm_unlock(shp);
968         goto out;
969
970 out_free:
971         kfree(sfd);
972 out_put_path:
973         dput(path.dentry);
974         mntput(path.mnt);
975         goto out_nattch;
976 }
977
978 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
979 {
980         unsigned long ret;
981         long err;
982
983         err = do_shmat(shmid, shmaddr, shmflg, &ret);
984         if (err)
985                 return err;
986         force_successful_syscall_return();
987         return (long)ret;
988 }
989
990 /*
991  * detach and kill segment if marked destroyed.
992  * The work is done in shm_close.
993  */
994 asmlinkage long sys_shmdt(char __user *shmaddr)
995 {
996         struct mm_struct *mm = current->mm;
997         struct vm_area_struct *vma, *next;
998         unsigned long addr = (unsigned long)shmaddr;
999         loff_t size = 0;
1000         int retval = -EINVAL;
1001
1002         if (addr & ~PAGE_MASK)
1003                 return retval;
1004
1005         down_write(&mm->mmap_sem);
1006
1007         /*
1008          * This function tries to be smart and unmap shm segments that
1009          * were modified by partial mlock or munmap calls:
1010          * - It first determines the size of the shm segment that should be
1011          *   unmapped: It searches for a vma that is backed by shm and that
1012          *   started at address shmaddr. It records it's size and then unmaps
1013          *   it.
1014          * - Then it unmaps all shm vmas that started at shmaddr and that
1015          *   are within the initially determined size.
1016          * Errors from do_munmap are ignored: the function only fails if
1017          * it's called with invalid parameters or if it's called to unmap
1018          * a part of a vma. Both calls in this function are for full vmas,
1019          * the parameters are directly copied from the vma itself and always
1020          * valid - therefore do_munmap cannot fail. (famous last words?)
1021          */
1022         /*
1023          * If it had been mremap()'d, the starting address would not
1024          * match the usual checks anyway. So assume all vma's are
1025          * above the starting address given.
1026          */
1027         vma = find_vma(mm, addr);
1028
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 comparisions 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         up_write(&mm->mmap_sem);
1074         return retval;
1075 }
1076
1077 #ifdef CONFIG_PROC_FS
1078 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1079 {
1080         struct shmid_kernel *shp = it;
1081         char *format;
1082
1083 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1084 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1085
1086         if (sizeof(size_t) <= sizeof(int))
1087                 format = SMALL_STRING;
1088         else
1089                 format = BIG_STRING;
1090         return seq_printf(s, format,
1091                           shp->shm_perm.key,
1092                           shp->id,
1093                           shp->shm_perm.mode,
1094                           shp->shm_segsz,
1095                           shp->shm_cprid,
1096                           shp->shm_lprid,
1097                           shp->shm_nattch,
1098                           shp->shm_perm.uid,
1099                           shp->shm_perm.gid,
1100                           shp->shm_perm.cuid,
1101                           shp->shm_perm.cgid,
1102                           shp->shm_atim,
1103                           shp->shm_dtim,
1104                           shp->shm_ctim);
1105 }
1106 #endif