Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6 into for-linus
[linux-2.6] / 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 if (vma->vm_policy)
258                 pol = vma->vm_policy;
259         else
260                 pol = current->mempolicy;
261         return pol;
262 }
263 #endif
264
265 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
266 {
267         struct shm_file_data *sfd = shm_file_data(file);
268         int ret;
269
270         ret = sfd->file->f_op->mmap(sfd->file, vma);
271         if (ret != 0)
272                 return ret;
273         sfd->vm_ops = vma->vm_ops;
274         vma->vm_ops = &shm_vm_ops;
275         shm_open(vma);
276
277         return ret;
278 }
279
280 static int shm_release(struct inode *ino, struct file *file)
281 {
282         struct shm_file_data *sfd = shm_file_data(file);
283
284         put_ipc_ns(sfd->ns);
285         shm_file_data(file) = NULL;
286         kfree(sfd);
287         return 0;
288 }
289
290 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
291 {
292         int (*fsync) (struct file *, struct dentry *, int datasync);
293         struct shm_file_data *sfd = shm_file_data(file);
294         int ret = -EINVAL;
295
296         fsync = sfd->file->f_op->fsync;
297         if (fsync)
298                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
299         return ret;
300 }
301
302 static unsigned long shm_get_unmapped_area(struct file *file,
303         unsigned long addr, unsigned long len, unsigned long pgoff,
304         unsigned long flags)
305 {
306         struct shm_file_data *sfd = shm_file_data(file);
307         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
308 }
309
310 int is_file_shm_hugepages(struct file *file)
311 {
312         int ret = 0;
313
314         if (file->f_op == &shm_file_operations) {
315                 struct shm_file_data *sfd;
316                 sfd = shm_file_data(file);
317                 ret = is_file_hugepages(sfd->file);
318         }
319         return ret;
320 }
321
322 static const struct file_operations shm_file_operations = {
323         .mmap           = shm_mmap,
324         .fsync          = shm_fsync,
325         .release        = shm_release,
326         .get_unmapped_area      = shm_get_unmapped_area,
327 };
328
329 static struct vm_operations_struct shm_vm_ops = {
330         .open   = shm_open,     /* callback for a new vm-area open */
331         .close  = shm_close,    /* callback for when the vm-area is released */
332         .nopage = shm_nopage,
333 #if defined(CONFIG_NUMA)
334         .set_policy = shm_set_policy,
335         .get_policy = shm_get_policy,
336 #endif
337 };
338
339 static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
340 {
341         int error;
342         struct shmid_kernel *shp;
343         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
344         struct file * file;
345         char name[13];
346         int id;
347
348         if (size < SHMMIN || size > ns->shm_ctlmax)
349                 return -EINVAL;
350
351         if (ns->shm_tot + numpages > ns->shm_ctlall)
352                 return -ENOSPC;
353
354         shp = ipc_rcu_alloc(sizeof(*shp));
355         if (!shp)
356                 return -ENOMEM;
357
358         shp->shm_perm.key = key;
359         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
360         shp->mlock_user = NULL;
361
362         shp->shm_perm.security = NULL;
363         error = security_shm_alloc(shp);
364         if (error) {
365                 ipc_rcu_putref(shp);
366                 return error;
367         }
368
369         sprintf (name, "SYSV%08x", key);
370         if (shmflg & SHM_HUGETLB) {
371                 /* hugetlb_file_setup takes care of mlock user accounting */
372                 file = hugetlb_file_setup(name, size);
373                 shp->mlock_user = current->user;
374         } else {
375                 int acctflag = VM_ACCOUNT;
376                 /*
377                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
378                  * if it's asked for.
379                  */
380                 if  ((shmflg & SHM_NORESERVE) &&
381                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
382                         acctflag = 0;
383                 file = shmem_file_setup(name, size, acctflag);
384         }
385         error = PTR_ERR(file);
386         if (IS_ERR(file))
387                 goto no_file;
388
389         error = -ENOSPC;
390         id = shm_addid(ns, shp);
391         if(id == -1) 
392                 goto no_id;
393
394         shp->shm_cprid = current->tgid;
395         shp->shm_lprid = 0;
396         shp->shm_atim = shp->shm_dtim = 0;
397         shp->shm_ctim = get_seconds();
398         shp->shm_segsz = size;
399         shp->shm_nattch = 0;
400         shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
401         shp->shm_file = file;
402         /*
403          * shmid gets reported as "inode#" in /proc/pid/maps.
404          * proc-ps tools use this. Changing this will break them.
405          */
406         file->f_dentry->d_inode->i_ino = shp->id;
407
408         ns->shm_tot += numpages;
409         shm_unlock(shp);
410         return shp->id;
411
412 no_id:
413         fput(file);
414 no_file:
415         security_shm_free(shp);
416         ipc_rcu_putref(shp);
417         return error;
418 }
419
420 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
421 {
422         struct shmid_kernel *shp;
423         int err, id = 0;
424         struct ipc_namespace *ns;
425
426         ns = current->nsproxy->ipc_ns;
427
428         mutex_lock(&shm_ids(ns).mutex);
429         if (key == IPC_PRIVATE) {
430                 err = newseg(ns, key, shmflg, size);
431         } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
432                 if (!(shmflg & IPC_CREAT))
433                         err = -ENOENT;
434                 else
435                         err = newseg(ns, key, shmflg, size);
436         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
437                 err = -EEXIST;
438         } else {
439                 shp = shm_lock(ns, id);
440                 BUG_ON(shp==NULL);
441                 if (shp->shm_segsz < size)
442                         err = -EINVAL;
443                 else if (ipcperms(&shp->shm_perm, shmflg))
444                         err = -EACCES;
445                 else {
446                         int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
447                         err = security_shm_associate(shp, shmflg);
448                         if (!err)
449                                 err = shmid;
450                 }
451                 shm_unlock(shp);
452         }
453         mutex_unlock(&shm_ids(ns).mutex);
454
455         return err;
456 }
457
458 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
459 {
460         switch(version) {
461         case IPC_64:
462                 return copy_to_user(buf, in, sizeof(*in));
463         case IPC_OLD:
464             {
465                 struct shmid_ds out;
466
467                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
468                 out.shm_segsz   = in->shm_segsz;
469                 out.shm_atime   = in->shm_atime;
470                 out.shm_dtime   = in->shm_dtime;
471                 out.shm_ctime   = in->shm_ctime;
472                 out.shm_cpid    = in->shm_cpid;
473                 out.shm_lpid    = in->shm_lpid;
474                 out.shm_nattch  = in->shm_nattch;
475
476                 return copy_to_user(buf, &out, sizeof(out));
477             }
478         default:
479                 return -EINVAL;
480         }
481 }
482
483 struct shm_setbuf {
484         uid_t   uid;
485         gid_t   gid;
486         mode_t  mode;
487 };      
488
489 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
490 {
491         switch(version) {
492         case IPC_64:
493             {
494                 struct shmid64_ds tbuf;
495
496                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
497                         return -EFAULT;
498
499                 out->uid        = tbuf.shm_perm.uid;
500                 out->gid        = tbuf.shm_perm.gid;
501                 out->mode       = tbuf.shm_perm.mode;
502
503                 return 0;
504             }
505         case IPC_OLD:
506             {
507                 struct shmid_ds tbuf_old;
508
509                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
510                         return -EFAULT;
511
512                 out->uid        = tbuf_old.shm_perm.uid;
513                 out->gid        = tbuf_old.shm_perm.gid;
514                 out->mode       = tbuf_old.shm_perm.mode;
515
516                 return 0;
517             }
518         default:
519                 return -EINVAL;
520         }
521 }
522
523 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
524 {
525         switch(version) {
526         case IPC_64:
527                 return copy_to_user(buf, in, sizeof(*in));
528         case IPC_OLD:
529             {
530                 struct shminfo out;
531
532                 if(in->shmmax > INT_MAX)
533                         out.shmmax = INT_MAX;
534                 else
535                         out.shmmax = (int)in->shmmax;
536
537                 out.shmmin      = in->shmmin;
538                 out.shmmni      = in->shmmni;
539                 out.shmseg      = in->shmseg;
540                 out.shmall      = in->shmall; 
541
542                 return copy_to_user(buf, &out, sizeof(out));
543             }
544         default:
545                 return -EINVAL;
546         }
547 }
548
549 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
550                 unsigned long *swp)
551 {
552         int i;
553
554         *rss = 0;
555         *swp = 0;
556
557         for (i = 0; i <= shm_ids(ns).max_id; i++) {
558                 struct shmid_kernel *shp;
559                 struct inode *inode;
560
561                 shp = shm_get(ns, i);
562                 if(!shp)
563                         continue;
564
565                 inode = shp->shm_file->f_path.dentry->d_inode;
566
567                 if (is_file_hugepages(shp->shm_file)) {
568                         struct address_space *mapping = inode->i_mapping;
569                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
570                 } else {
571                         struct shmem_inode_info *info = SHMEM_I(inode);
572                         spin_lock(&info->lock);
573                         *rss += inode->i_mapping->nrpages;
574                         *swp += info->swapped;
575                         spin_unlock(&info->lock);
576                 }
577         }
578 }
579
580 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
581 {
582         struct shm_setbuf setbuf;
583         struct shmid_kernel *shp;
584         int err, version;
585         struct ipc_namespace *ns;
586
587         if (cmd < 0 || shmid < 0) {
588                 err = -EINVAL;
589                 goto out;
590         }
591
592         version = ipc_parse_version(&cmd);
593         ns = current->nsproxy->ipc_ns;
594
595         switch (cmd) { /* replace with proc interface ? */
596         case IPC_INFO:
597         {
598                 struct shminfo64 shminfo;
599
600                 err = security_shm_shmctl(NULL, cmd);
601                 if (err)
602                         return err;
603
604                 memset(&shminfo,0,sizeof(shminfo));
605                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
606                 shminfo.shmmax = ns->shm_ctlmax;
607                 shminfo.shmall = ns->shm_ctlall;
608
609                 shminfo.shmmin = SHMMIN;
610                 if(copy_shminfo_to_user (buf, &shminfo, version))
611                         return -EFAULT;
612                 /* reading a integer is always atomic */
613                 err= shm_ids(ns).max_id;
614                 if(err<0)
615                         err = 0;
616                 goto out;
617         }
618         case SHM_INFO:
619         {
620                 struct shm_info shm_info;
621
622                 err = security_shm_shmctl(NULL, cmd);
623                 if (err)
624                         return err;
625
626                 memset(&shm_info,0,sizeof(shm_info));
627                 mutex_lock(&shm_ids(ns).mutex);
628                 shm_info.used_ids = shm_ids(ns).in_use;
629                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
630                 shm_info.shm_tot = ns->shm_tot;
631                 shm_info.swap_attempts = 0;
632                 shm_info.swap_successes = 0;
633                 err = shm_ids(ns).max_id;
634                 mutex_unlock(&shm_ids(ns).mutex);
635                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
636                         err = -EFAULT;
637                         goto out;
638                 }
639
640                 err = err < 0 ? 0 : err;
641                 goto out;
642         }
643         case SHM_STAT:
644         case IPC_STAT:
645         {
646                 struct shmid64_ds tbuf;
647                 int result;
648                 memset(&tbuf, 0, sizeof(tbuf));
649                 shp = shm_lock(ns, shmid);
650                 if(shp==NULL) {
651                         err = -EINVAL;
652                         goto out;
653                 } else if(cmd==SHM_STAT) {
654                         err = -EINVAL;
655                         if (shmid > shm_ids(ns).max_id)
656                                 goto out_unlock;
657                         result = shm_buildid(ns, shmid, shp->shm_perm.seq);
658                 } else {
659                         err = shm_checkid(ns, shp,shmid);
660                         if(err)
661                                 goto out_unlock;
662                         result = 0;
663                 }
664                 err=-EACCES;
665                 if (ipcperms (&shp->shm_perm, S_IRUGO))
666                         goto out_unlock;
667                 err = security_shm_shmctl(shp, cmd);
668                 if (err)
669                         goto out_unlock;
670                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
671                 tbuf.shm_segsz  = shp->shm_segsz;
672                 tbuf.shm_atime  = shp->shm_atim;
673                 tbuf.shm_dtime  = shp->shm_dtim;
674                 tbuf.shm_ctime  = shp->shm_ctim;
675                 tbuf.shm_cpid   = shp->shm_cprid;
676                 tbuf.shm_lpid   = shp->shm_lprid;
677                 tbuf.shm_nattch = shp->shm_nattch;
678                 shm_unlock(shp);
679                 if(copy_shmid_to_user (buf, &tbuf, version))
680                         err = -EFAULT;
681                 else
682                         err = result;
683                 goto out;
684         }
685         case SHM_LOCK:
686         case SHM_UNLOCK:
687         {
688                 shp = shm_lock(ns, shmid);
689                 if(shp==NULL) {
690                         err = -EINVAL;
691                         goto out;
692                 }
693                 err = shm_checkid(ns, shp,shmid);
694                 if(err)
695                         goto out_unlock;
696
697                 err = audit_ipc_obj(&(shp->shm_perm));
698                 if (err)
699                         goto out_unlock;
700
701                 if (!capable(CAP_IPC_LOCK)) {
702                         err = -EPERM;
703                         if (current->euid != shp->shm_perm.uid &&
704                             current->euid != shp->shm_perm.cuid)
705                                 goto out_unlock;
706                         if (cmd == SHM_LOCK &&
707                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
708                                 goto out_unlock;
709                 }
710
711                 err = security_shm_shmctl(shp, cmd);
712                 if (err)
713                         goto out_unlock;
714                 
715                 if(cmd==SHM_LOCK) {
716                         struct user_struct * user = current->user;
717                         if (!is_file_hugepages(shp->shm_file)) {
718                                 err = shmem_lock(shp->shm_file, 1, user);
719                                 if (!err) {
720                                         shp->shm_perm.mode |= SHM_LOCKED;
721                                         shp->mlock_user = user;
722                                 }
723                         }
724                 } else if (!is_file_hugepages(shp->shm_file)) {
725                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
726                         shp->shm_perm.mode &= ~SHM_LOCKED;
727                         shp->mlock_user = NULL;
728                 }
729                 shm_unlock(shp);
730                 goto out;
731         }
732         case IPC_RMID:
733         {
734                 /*
735                  *      We cannot simply remove the file. The SVID states
736                  *      that the block remains until the last person
737                  *      detaches from it, then is deleted. A shmat() on
738                  *      an RMID segment is legal in older Linux and if 
739                  *      we change it apps break...
740                  *
741                  *      Instead we set a destroyed flag, and then blow
742                  *      the name away when the usage hits zero.
743                  */
744                 mutex_lock(&shm_ids(ns).mutex);
745                 shp = shm_lock(ns, shmid);
746                 err = -EINVAL;
747                 if (shp == NULL) 
748                         goto out_up;
749                 err = shm_checkid(ns, shp, shmid);
750                 if(err)
751                         goto out_unlock_up;
752
753                 err = audit_ipc_obj(&(shp->shm_perm));
754                 if (err)
755                         goto out_unlock_up;
756
757                 if (current->euid != shp->shm_perm.uid &&
758                     current->euid != shp->shm_perm.cuid && 
759                     !capable(CAP_SYS_ADMIN)) {
760                         err=-EPERM;
761                         goto out_unlock_up;
762                 }
763
764                 err = security_shm_shmctl(shp, cmd);
765                 if (err)
766                         goto out_unlock_up;
767
768                 do_shm_rmid(ns, shp);
769                 mutex_unlock(&shm_ids(ns).mutex);
770                 goto out;
771         }
772
773         case IPC_SET:
774         {
775                 if (copy_shmid_from_user (&setbuf, buf, version)) {
776                         err = -EFAULT;
777                         goto out;
778                 }
779                 mutex_lock(&shm_ids(ns).mutex);
780                 shp = shm_lock(ns, shmid);
781                 err=-EINVAL;
782                 if(shp==NULL)
783                         goto out_up;
784                 err = shm_checkid(ns, shp,shmid);
785                 if(err)
786                         goto out_unlock_up;
787                 err = audit_ipc_obj(&(shp->shm_perm));
788                 if (err)
789                         goto out_unlock_up;
790                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
791                 if (err)
792                         goto out_unlock_up;
793                 err=-EPERM;
794                 if (current->euid != shp->shm_perm.uid &&
795                     current->euid != shp->shm_perm.cuid && 
796                     !capable(CAP_SYS_ADMIN)) {
797                         goto out_unlock_up;
798                 }
799
800                 err = security_shm_shmctl(shp, cmd);
801                 if (err)
802                         goto out_unlock_up;
803                 
804                 shp->shm_perm.uid = setbuf.uid;
805                 shp->shm_perm.gid = setbuf.gid;
806                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
807                         | (setbuf.mode & S_IRWXUGO);
808                 shp->shm_ctim = get_seconds();
809                 break;
810         }
811
812         default:
813                 err = -EINVAL;
814                 goto out;
815         }
816
817         err = 0;
818 out_unlock_up:
819         shm_unlock(shp);
820 out_up:
821         mutex_unlock(&shm_ids(ns).mutex);
822         goto out;
823 out_unlock:
824         shm_unlock(shp);
825 out:
826         return err;
827 }
828
829 /*
830  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
831  *
832  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
833  * "raddr" thing points to kernel space, and there has to be a wrapper around
834  * this.
835  */
836 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
837 {
838         struct shmid_kernel *shp;
839         unsigned long addr;
840         unsigned long size;
841         struct file * file;
842         int    err;
843         unsigned long flags;
844         unsigned long prot;
845         int acc_mode;
846         unsigned long user_addr;
847         struct ipc_namespace *ns;
848         struct shm_file_data *sfd;
849         struct path path;
850         mode_t f_mode;
851
852         err = -EINVAL;
853         if (shmid < 0)
854                 goto out;
855         else if ((addr = (ulong)shmaddr)) {
856                 if (addr & (SHMLBA-1)) {
857                         if (shmflg & SHM_RND)
858                                 addr &= ~(SHMLBA-1);       /* round down */
859                         else
860 #ifndef __ARCH_FORCE_SHMLBA
861                                 if (addr & ~PAGE_MASK)
862 #endif
863                                         goto out;
864                 }
865                 flags = MAP_SHARED | MAP_FIXED;
866         } else {
867                 if ((shmflg & SHM_REMAP))
868                         goto out;
869
870                 flags = MAP_SHARED;
871         }
872
873         if (shmflg & SHM_RDONLY) {
874                 prot = PROT_READ;
875                 acc_mode = S_IRUGO;
876                 f_mode = FMODE_READ;
877         } else {
878                 prot = PROT_READ | PROT_WRITE;
879                 acc_mode = S_IRUGO | S_IWUGO;
880                 f_mode = FMODE_READ | FMODE_WRITE;
881         }
882         if (shmflg & SHM_EXEC) {
883                 prot |= PROT_EXEC;
884                 acc_mode |= S_IXUGO;
885         }
886
887         /*
888          * We cannot rely on the fs check since SYSV IPC does have an
889          * additional creator id...
890          */
891         ns = current->nsproxy->ipc_ns;
892         shp = shm_lock(ns, shmid);
893         if(shp == NULL)
894                 goto out;
895
896         err = shm_checkid(ns, shp,shmid);
897         if (err)
898                 goto out_unlock;
899
900         err = -EACCES;
901         if (ipcperms(&shp->shm_perm, acc_mode))
902                 goto out_unlock;
903
904         err = security_shm_shmat(shp, shmaddr, shmflg);
905         if (err)
906                 goto out_unlock;
907
908         path.dentry = dget(shp->shm_file->f_path.dentry);
909         path.mnt    = mntget(shp->shm_file->f_path.mnt);
910         shp->shm_nattch++;
911         size = i_size_read(path.dentry->d_inode);
912         shm_unlock(shp);
913
914         err = -ENOMEM;
915         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
916         if (!sfd)
917                 goto out_put_path;
918
919         err = -ENOMEM;
920         file = get_empty_filp();
921         if (!file)
922                 goto out_free;
923
924         file->f_op = &shm_file_operations;
925         file->private_data = sfd;
926         file->f_path = path;
927         file->f_mapping = shp->shm_file->f_mapping;
928         file->f_mode = f_mode;
929         sfd->id = shp->id;
930         sfd->ns = get_ipc_ns(ns);
931         sfd->file = shp->shm_file;
932         sfd->vm_ops = NULL;
933
934         down_write(&current->mm->mmap_sem);
935         if (addr && !(shmflg & SHM_REMAP)) {
936                 err = -EINVAL;
937                 if (find_vma_intersection(current->mm, addr, addr + size))
938                         goto invalid;
939                 /*
940                  * If shm segment goes below stack, make sure there is some
941                  * space left for the stack to grow (at least 4 pages).
942                  */
943                 if (addr < current->mm->start_stack &&
944                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
945                         goto invalid;
946         }
947                 
948         user_addr = do_mmap (file, addr, size, prot, flags, 0);
949         *raddr = user_addr;
950         err = 0;
951         if (IS_ERR_VALUE(user_addr))
952                 err = (long)user_addr;
953 invalid:
954         up_write(&current->mm->mmap_sem);
955
956         fput(file);
957
958 out_nattch:
959         mutex_lock(&shm_ids(ns).mutex);
960         shp = shm_lock(ns, shmid);
961         BUG_ON(!shp);
962         shp->shm_nattch--;
963         if(shp->shm_nattch == 0 &&
964            shp->shm_perm.mode & SHM_DEST)
965                 shm_destroy(ns, shp);
966         else
967                 shm_unlock(shp);
968         mutex_unlock(&shm_ids(ns).mutex);
969
970 out:
971         return err;
972
973 out_unlock:
974         shm_unlock(shp);
975         goto out;
976
977 out_free:
978         kfree(sfd);
979 out_put_path:
980         dput(path.dentry);
981         mntput(path.mnt);
982         goto out_nattch;
983 }
984
985 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
986 {
987         unsigned long ret;
988         long err;
989
990         err = do_shmat(shmid, shmaddr, shmflg, &ret);
991         if (err)
992                 return err;
993         force_successful_syscall_return();
994         return (long)ret;
995 }
996
997 /*
998  * detach and kill segment if marked destroyed.
999  * The work is done in shm_close.
1000  */
1001 asmlinkage long sys_shmdt(char __user *shmaddr)
1002 {
1003         struct mm_struct *mm = current->mm;
1004         struct vm_area_struct *vma, *next;
1005         unsigned long addr = (unsigned long)shmaddr;
1006         loff_t size = 0;
1007         int retval = -EINVAL;
1008
1009         if (addr & ~PAGE_MASK)
1010                 return retval;
1011
1012         down_write(&mm->mmap_sem);
1013
1014         /*
1015          * This function tries to be smart and unmap shm segments that
1016          * were modified by partial mlock or munmap calls:
1017          * - It first determines the size of the shm segment that should be
1018          *   unmapped: It searches for a vma that is backed by shm and that
1019          *   started at address shmaddr. It records it's size and then unmaps
1020          *   it.
1021          * - Then it unmaps all shm vmas that started at shmaddr and that
1022          *   are within the initially determined size.
1023          * Errors from do_munmap are ignored: the function only fails if
1024          * it's called with invalid parameters or if it's called to unmap
1025          * a part of a vma. Both calls in this function are for full vmas,
1026          * the parameters are directly copied from the vma itself and always
1027          * valid - therefore do_munmap cannot fail. (famous last words?)
1028          */
1029         /*
1030          * If it had been mremap()'d, the starting address would not
1031          * match the usual checks anyway. So assume all vma's are
1032          * above the starting address given.
1033          */
1034         vma = find_vma(mm, addr);
1035
1036         while (vma) {
1037                 next = vma->vm_next;
1038
1039                 /*
1040                  * Check if the starting address would match, i.e. it's
1041                  * a fragment created by mprotect() and/or munmap(), or it
1042                  * otherwise it starts at this address with no hassles.
1043                  */
1044                 if ((vma->vm_ops == &shm_vm_ops) &&
1045                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1046
1047
1048                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1049                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1050                         /*
1051                          * We discovered the size of the shm segment, so
1052                          * break out of here and fall through to the next
1053                          * loop that uses the size information to stop
1054                          * searching for matching vma's.
1055                          */
1056                         retval = 0;
1057                         vma = next;
1058                         break;
1059                 }
1060                 vma = next;
1061         }
1062
1063         /*
1064          * We need look no further than the maximum address a fragment
1065          * could possibly have landed at. Also cast things to loff_t to
1066          * prevent overflows and make comparisions vs. equal-width types.
1067          */
1068         size = PAGE_ALIGN(size);
1069         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1070                 next = vma->vm_next;
1071
1072                 /* finding a matching vma now does not alter retval */
1073                 if ((vma->vm_ops == &shm_vm_ops) &&
1074                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1075
1076                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1077                 vma = next;
1078         }
1079
1080         up_write(&mm->mmap_sem);
1081         return retval;
1082 }
1083
1084 #ifdef CONFIG_PROC_FS
1085 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1086 {
1087         struct shmid_kernel *shp = it;
1088         char *format;
1089
1090 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1091 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1092
1093         if (sizeof(size_t) <= sizeof(int))
1094                 format = SMALL_STRING;
1095         else
1096                 format = BIG_STRING;
1097         return seq_printf(s, format,
1098                           shp->shm_perm.key,
1099                           shp->id,
1100                           shp->shm_perm.mode,
1101                           shp->shm_segsz,
1102                           shp->shm_cprid,
1103                           shp->shm_lprid,
1104                           shp->shm_nattch,
1105                           shp->shm_perm.uid,
1106                           shp->shm_perm.gid,
1107                           shp->shm_perm.cuid,
1108                           shp->shm_perm.cgid,
1109                           shp->shm_atim,
1110                           shp->shm_dtim,
1111                           shp->shm_ctim);
1112 }
1113 #endif