ipc/sem.c: optimize single semop operations
[linux-2.6.git] / ipc / shm.c
1 /*
2  * linux/ipc/shm.c
3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
4  *       Many improvements/fixes by Bruno Haible.
5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7  *
8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15  *
16  * support for audit of ipc object properties and permission changes
17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18  *
19  * namespaces support
20  * OpenVZ, SWsoft Inc.
21  * Pavel Emelianov <xemul@openvz.org>
22  */
23
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42 #include <linux/ima.h>
43
44 #include <asm/uaccess.h>
45
46 #include "util.h"
47
48 struct shm_file_data {
49         int id;
50         struct ipc_namespace *ns;
51         struct file *file;
52         const struct vm_operations_struct *vm_ops;
53 };
54
55 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
56
57 static const struct file_operations shm_file_operations;
58 static const struct vm_operations_struct shm_vm_ops;
59
60 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
61
62 #define shm_unlock(shp)                 \
63         ipc_unlock(&(shp)->shm_perm)
64
65 static int newseg(struct ipc_namespace *, struct ipc_params *);
66 static void shm_open(struct vm_area_struct *vma);
67 static void shm_close(struct vm_area_struct *vma);
68 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
69 #ifdef CONFIG_PROC_FS
70 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
71 #endif
72
73 void shm_init_ns(struct ipc_namespace *ns)
74 {
75         ns->shm_ctlmax = SHMMAX;
76         ns->shm_ctlall = SHMALL;
77         ns->shm_ctlmni = SHMMNI;
78         ns->shm_tot = 0;
79         ipc_init_ids(&shm_ids(ns));
80 }
81
82 /*
83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84  * Only shm_ids.rw_mutex remains locked on exit.
85  */
86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
87 {
88         struct shmid_kernel *shp;
89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
90
91         if (shp->shm_nattch){
92                 shp->shm_perm.mode |= SHM_DEST;
93                 /* Do not find it any more */
94                 shp->shm_perm.key = IPC_PRIVATE;
95                 shm_unlock(shp);
96         } else
97                 shm_destroy(ns, shp);
98 }
99
100 #ifdef CONFIG_IPC_NS
101 void shm_exit_ns(struct ipc_namespace *ns)
102 {
103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
104         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
105 }
106 #endif
107
108 void __init shm_init (void)
109 {
110         shm_init_ns(&init_ipc_ns);
111         ipc_init_proc_interface("sysvipc/shm",
112                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
113                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
114 }
115
116 /*
117  * shm_lock_(check_) routines are called in the paths where the rw_mutex
118  * is not necessarily held.
119  */
120 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
121 {
122         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
123
124         if (IS_ERR(ipcp))
125                 return (struct shmid_kernel *)ipcp;
126
127         return container_of(ipcp, struct shmid_kernel, shm_perm);
128 }
129
130 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
131                                                 int id)
132 {
133         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
134
135         if (IS_ERR(ipcp))
136                 return (struct shmid_kernel *)ipcp;
137
138         return container_of(ipcp, struct shmid_kernel, shm_perm);
139 }
140
141 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
142 {
143         ipc_rmid(&shm_ids(ns), &s->shm_perm);
144 }
145
146
147 /* This is called by fork, once for every shm attach. */
148 static void shm_open(struct vm_area_struct *vma)
149 {
150         struct file *file = vma->vm_file;
151         struct shm_file_data *sfd = shm_file_data(file);
152         struct shmid_kernel *shp;
153
154         shp = shm_lock(sfd->ns, sfd->id);
155         BUG_ON(IS_ERR(shp));
156         shp->shm_atim = get_seconds();
157         shp->shm_lprid = task_tgid_vnr(current);
158         shp->shm_nattch++;
159         shm_unlock(shp);
160 }
161
162 /*
163  * shm_destroy - free the struct shmid_kernel
164  *
165  * @ns: namespace
166  * @shp: struct to free
167  *
168  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
169  * but returns with shp unlocked and freed.
170  */
171 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
172 {
173         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
174         shm_rmid(ns, shp);
175         shm_unlock(shp);
176         if (!is_file_hugepages(shp->shm_file))
177                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
178         else if (shp->mlock_user)
179                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
180                                                 shp->mlock_user);
181         fput (shp->shm_file);
182         security_shm_free(shp);
183         ipc_rcu_putref(shp);
184 }
185
186 /*
187  * remove the attach descriptor vma.
188  * free memory for segment if it is marked destroyed.
189  * The descriptor has already been removed from the current->mm->mmap list
190  * and will later be kfree()d.
191  */
192 static void shm_close(struct vm_area_struct *vma)
193 {
194         struct file * file = vma->vm_file;
195         struct shm_file_data *sfd = shm_file_data(file);
196         struct shmid_kernel *shp;
197         struct ipc_namespace *ns = sfd->ns;
198
199         down_write(&shm_ids(ns).rw_mutex);
200         /* remove from the list of attaches of the shm segment */
201         shp = shm_lock(ns, sfd->id);
202         BUG_ON(IS_ERR(shp));
203         shp->shm_lprid = task_tgid_vnr(current);
204         shp->shm_dtim = get_seconds();
205         shp->shm_nattch--;
206         if(shp->shm_nattch == 0 &&
207            shp->shm_perm.mode & SHM_DEST)
208                 shm_destroy(ns, shp);
209         else
210                 shm_unlock(shp);
211         up_write(&shm_ids(ns).rw_mutex);
212 }
213
214 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
215 {
216         struct file *file = vma->vm_file;
217         struct shm_file_data *sfd = shm_file_data(file);
218
219         return sfd->vm_ops->fault(vma, vmf);
220 }
221
222 #ifdef CONFIG_NUMA
223 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
224 {
225         struct file *file = vma->vm_file;
226         struct shm_file_data *sfd = shm_file_data(file);
227         int err = 0;
228         if (sfd->vm_ops->set_policy)
229                 err = sfd->vm_ops->set_policy(vma, new);
230         return err;
231 }
232
233 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
234                                         unsigned long addr)
235 {
236         struct file *file = vma->vm_file;
237         struct shm_file_data *sfd = shm_file_data(file);
238         struct mempolicy *pol = NULL;
239
240         if (sfd->vm_ops->get_policy)
241                 pol = sfd->vm_ops->get_policy(vma, addr);
242         else if (vma->vm_policy)
243                 pol = vma->vm_policy;
244
245         return pol;
246 }
247 #endif
248
249 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
250 {
251         struct shm_file_data *sfd = shm_file_data(file);
252         int ret;
253
254         ret = sfd->file->f_op->mmap(sfd->file, vma);
255         if (ret != 0)
256                 return ret;
257         sfd->vm_ops = vma->vm_ops;
258 #ifdef CONFIG_MMU
259         BUG_ON(!sfd->vm_ops->fault);
260 #endif
261         vma->vm_ops = &shm_vm_ops;
262         shm_open(vma);
263
264         return ret;
265 }
266
267 static int shm_release(struct inode *ino, struct file *file)
268 {
269         struct shm_file_data *sfd = shm_file_data(file);
270
271         put_ipc_ns(sfd->ns);
272         shm_file_data(file) = NULL;
273         kfree(sfd);
274         return 0;
275 }
276
277 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
278 {
279         int (*fsync) (struct file *, struct dentry *, int datasync);
280         struct shm_file_data *sfd = shm_file_data(file);
281         int ret = -EINVAL;
282
283         fsync = sfd->file->f_op->fsync;
284         if (fsync)
285                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
286         return ret;
287 }
288
289 static unsigned long shm_get_unmapped_area(struct file *file,
290         unsigned long addr, unsigned long len, unsigned long pgoff,
291         unsigned long flags)
292 {
293         struct shm_file_data *sfd = shm_file_data(file);
294         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
295                                                 pgoff, flags);
296 }
297
298 static const struct file_operations shm_file_operations = {
299         .mmap           = shm_mmap,
300         .fsync          = shm_fsync,
301         .release        = shm_release,
302 };
303
304 static const struct file_operations shm_file_operations_huge = {
305         .mmap           = shm_mmap,
306         .fsync          = shm_fsync,
307         .release        = shm_release,
308         .get_unmapped_area      = shm_get_unmapped_area,
309 };
310
311 int is_file_shm_hugepages(struct file *file)
312 {
313         return file->f_op == &shm_file_operations_huge;
314 }
315
316 static const struct vm_operations_struct shm_vm_ops = {
317         .open   = shm_open,     /* callback for a new vm-area open */
318         .close  = shm_close,    /* callback for when the vm-area is released */
319         .fault  = shm_fault,
320 #if defined(CONFIG_NUMA)
321         .set_policy = shm_set_policy,
322         .get_policy = shm_get_policy,
323 #endif
324 };
325
326 /**
327  * newseg - Create a new shared memory segment
328  * @ns: namespace
329  * @params: ptr to the structure that contains key, size and shmflg
330  *
331  * Called with shm_ids.rw_mutex held as a writer.
332  */
333
334 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
335 {
336         key_t key = params->key;
337         int shmflg = params->flg;
338         size_t size = params->u.size;
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         int acctflag = 0;
346
347         if (size < SHMMIN || size > ns->shm_ctlmax)
348                 return -EINVAL;
349
350         if (ns->shm_tot + numpages > ns->shm_ctlall)
351                 return -ENOSPC;
352
353         shp = ipc_rcu_alloc(sizeof(*shp));
354         if (!shp)
355                 return -ENOMEM;
356
357         shp->shm_perm.key = key;
358         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
359         shp->mlock_user = NULL;
360
361         shp->shm_perm.security = NULL;
362         error = security_shm_alloc(shp);
363         if (error) {
364                 ipc_rcu_putref(shp);
365                 return error;
366         }
367
368         sprintf (name, "SYSV%08x", key);
369         if (shmflg & SHM_HUGETLB) {
370                 /* hugetlb_file_setup applies strict accounting */
371                 if (shmflg & SHM_NORESERVE)
372                         acctflag = VM_NORESERVE;
373                 file = hugetlb_file_setup(name, size, acctflag,
374                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
375         } else {
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 = VM_NORESERVE;
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         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
390         if (id < 0) {
391                 error = id;
392                 goto no_id;
393         }
394
395         shp->shm_cprid = task_tgid_vnr(current);
396         shp->shm_lprid = 0;
397         shp->shm_atim = shp->shm_dtim = 0;
398         shp->shm_ctim = get_seconds();
399         shp->shm_segsz = size;
400         shp->shm_nattch = 0;
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->shm_perm.id;
407
408         ns->shm_tot += numpages;
409         error = shp->shm_perm.id;
410         shm_unlock(shp);
411         return error;
412
413 no_id:
414         if (is_file_hugepages(file) && shp->mlock_user)
415                 user_shm_unlock(size, shp->mlock_user);
416         fput(file);
417 no_file:
418         security_shm_free(shp);
419         ipc_rcu_putref(shp);
420         return error;
421 }
422
423 /*
424  * Called with shm_ids.rw_mutex and ipcp locked.
425  */
426 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
427 {
428         struct shmid_kernel *shp;
429
430         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
431         return security_shm_associate(shp, shmflg);
432 }
433
434 /*
435  * Called with shm_ids.rw_mutex and ipcp locked.
436  */
437 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
438                                 struct ipc_params *params)
439 {
440         struct shmid_kernel *shp;
441
442         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
443         if (shp->shm_segsz < params->u.size)
444                 return -EINVAL;
445
446         return 0;
447 }
448
449 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
450 {
451         struct ipc_namespace *ns;
452         struct ipc_ops shm_ops;
453         struct ipc_params shm_params;
454
455         ns = current->nsproxy->ipc_ns;
456
457         shm_ops.getnew = newseg;
458         shm_ops.associate = shm_security;
459         shm_ops.more_checks = shm_more_checks;
460
461         shm_params.key = key;
462         shm_params.flg = shmflg;
463         shm_params.u.size = size;
464
465         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
466 }
467
468 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
469 {
470         switch(version) {
471         case IPC_64:
472                 return copy_to_user(buf, in, sizeof(*in));
473         case IPC_OLD:
474             {
475                 struct shmid_ds out;
476
477                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
478                 out.shm_segsz   = in->shm_segsz;
479                 out.shm_atime   = in->shm_atime;
480                 out.shm_dtime   = in->shm_dtime;
481                 out.shm_ctime   = in->shm_ctime;
482                 out.shm_cpid    = in->shm_cpid;
483                 out.shm_lpid    = in->shm_lpid;
484                 out.shm_nattch  = in->shm_nattch;
485
486                 return copy_to_user(buf, &out, sizeof(out));
487             }
488         default:
489                 return -EINVAL;
490         }
491 }
492
493 static inline unsigned long
494 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
495 {
496         switch(version) {
497         case IPC_64:
498                 if (copy_from_user(out, buf, sizeof(*out)))
499                         return -EFAULT;
500                 return 0;
501         case IPC_OLD:
502             {
503                 struct shmid_ds tbuf_old;
504
505                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
506                         return -EFAULT;
507
508                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
509                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
510                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
511
512                 return 0;
513             }
514         default:
515                 return -EINVAL;
516         }
517 }
518
519 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
520 {
521         switch(version) {
522         case IPC_64:
523                 return copy_to_user(buf, in, sizeof(*in));
524         case IPC_OLD:
525             {
526                 struct shminfo out;
527
528                 if(in->shmmax > INT_MAX)
529                         out.shmmax = INT_MAX;
530                 else
531                         out.shmmax = (int)in->shmmax;
532
533                 out.shmmin      = in->shmmin;
534                 out.shmmni      = in->shmmni;
535                 out.shmseg      = in->shmseg;
536                 out.shmall      = in->shmall; 
537
538                 return copy_to_user(buf, &out, sizeof(out));
539             }
540         default:
541                 return -EINVAL;
542         }
543 }
544
545 /*
546  * Called with shm_ids.rw_mutex held as a reader
547  */
548 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
549                 unsigned long *swp)
550 {
551         int next_id;
552         int total, in_use;
553
554         *rss = 0;
555         *swp = 0;
556
557         in_use = shm_ids(ns).in_use;
558
559         for (total = 0, next_id = 0; total < in_use; next_id++) {
560                 struct kern_ipc_perm *ipc;
561                 struct shmid_kernel *shp;
562                 struct inode *inode;
563
564                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
565                 if (ipc == NULL)
566                         continue;
567                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
568
569                 inode = shp->shm_file->f_path.dentry->d_inode;
570
571                 if (is_file_hugepages(shp->shm_file)) {
572                         struct address_space *mapping = inode->i_mapping;
573                         struct hstate *h = hstate_file(shp->shm_file);
574                         *rss += pages_per_huge_page(h) * mapping->nrpages;
575                 } else {
576 #ifdef CONFIG_SHMEM
577                         struct shmem_inode_info *info = SHMEM_I(inode);
578                         spin_lock(&info->lock);
579                         *rss += inode->i_mapping->nrpages;
580                         *swp += info->swapped;
581                         spin_unlock(&info->lock);
582 #else
583                         *rss += inode->i_mapping->nrpages;
584 #endif
585                 }
586
587                 total++;
588         }
589 }
590
591 /*
592  * This function handles some shmctl commands which require the rw_mutex
593  * to be held in write mode.
594  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
595  */
596 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
597                        struct shmid_ds __user *buf, int version)
598 {
599         struct kern_ipc_perm *ipcp;
600         struct shmid64_ds shmid64;
601         struct shmid_kernel *shp;
602         int err;
603
604         if (cmd == IPC_SET) {
605                 if (copy_shmid_from_user(&shmid64, buf, version))
606                         return -EFAULT;
607         }
608
609         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
610         if (IS_ERR(ipcp))
611                 return PTR_ERR(ipcp);
612
613         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
614
615         err = security_shm_shmctl(shp, cmd);
616         if (err)
617                 goto out_unlock;
618         switch (cmd) {
619         case IPC_RMID:
620                 do_shm_rmid(ns, ipcp);
621                 goto out_up;
622         case IPC_SET:
623                 ipc_update_perm(&shmid64.shm_perm, ipcp);
624                 shp->shm_ctim = get_seconds();
625                 break;
626         default:
627                 err = -EINVAL;
628         }
629 out_unlock:
630         shm_unlock(shp);
631 out_up:
632         up_write(&shm_ids(ns).rw_mutex);
633         return err;
634 }
635
636 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
637 {
638         struct shmid_kernel *shp;
639         int err, version;
640         struct ipc_namespace *ns;
641
642         if (cmd < 0 || shmid < 0) {
643                 err = -EINVAL;
644                 goto out;
645         }
646
647         version = ipc_parse_version(&cmd);
648         ns = current->nsproxy->ipc_ns;
649
650         switch (cmd) { /* replace with proc interface ? */
651         case IPC_INFO:
652         {
653                 struct shminfo64 shminfo;
654
655                 err = security_shm_shmctl(NULL, cmd);
656                 if (err)
657                         return err;
658
659                 memset(&shminfo, 0, sizeof(shminfo));
660                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
661                 shminfo.shmmax = ns->shm_ctlmax;
662                 shminfo.shmall = ns->shm_ctlall;
663
664                 shminfo.shmmin = SHMMIN;
665                 if(copy_shminfo_to_user (buf, &shminfo, version))
666                         return -EFAULT;
667
668                 down_read(&shm_ids(ns).rw_mutex);
669                 err = ipc_get_maxid(&shm_ids(ns));
670                 up_read(&shm_ids(ns).rw_mutex);
671
672                 if(err<0)
673                         err = 0;
674                 goto out;
675         }
676         case SHM_INFO:
677         {
678                 struct shm_info shm_info;
679
680                 err = security_shm_shmctl(NULL, cmd);
681                 if (err)
682                         return err;
683
684                 memset(&shm_info, 0, sizeof(shm_info));
685                 down_read(&shm_ids(ns).rw_mutex);
686                 shm_info.used_ids = shm_ids(ns).in_use;
687                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
688                 shm_info.shm_tot = ns->shm_tot;
689                 shm_info.swap_attempts = 0;
690                 shm_info.swap_successes = 0;
691                 err = ipc_get_maxid(&shm_ids(ns));
692                 up_read(&shm_ids(ns).rw_mutex);
693                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
694                         err = -EFAULT;
695                         goto out;
696                 }
697
698                 err = err < 0 ? 0 : err;
699                 goto out;
700         }
701         case SHM_STAT:
702         case IPC_STAT:
703         {
704                 struct shmid64_ds tbuf;
705                 int result;
706
707                 if (cmd == SHM_STAT) {
708                         shp = shm_lock(ns, shmid);
709                         if (IS_ERR(shp)) {
710                                 err = PTR_ERR(shp);
711                                 goto out;
712                         }
713                         result = shp->shm_perm.id;
714                 } else {
715                         shp = shm_lock_check(ns, shmid);
716                         if (IS_ERR(shp)) {
717                                 err = PTR_ERR(shp);
718                                 goto out;
719                         }
720                         result = 0;
721                 }
722                 err = -EACCES;
723                 if (ipcperms (&shp->shm_perm, S_IRUGO))
724                         goto out_unlock;
725                 err = security_shm_shmctl(shp, cmd);
726                 if (err)
727                         goto out_unlock;
728                 memset(&tbuf, 0, sizeof(tbuf));
729                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
730                 tbuf.shm_segsz  = shp->shm_segsz;
731                 tbuf.shm_atime  = shp->shm_atim;
732                 tbuf.shm_dtime  = shp->shm_dtim;
733                 tbuf.shm_ctime  = shp->shm_ctim;
734                 tbuf.shm_cpid   = shp->shm_cprid;
735                 tbuf.shm_lpid   = shp->shm_lprid;
736                 tbuf.shm_nattch = shp->shm_nattch;
737                 shm_unlock(shp);
738                 if(copy_shmid_to_user (buf, &tbuf, version))
739                         err = -EFAULT;
740                 else
741                         err = result;
742                 goto out;
743         }
744         case SHM_LOCK:
745         case SHM_UNLOCK:
746         {
747                 struct file *uninitialized_var(shm_file);
748
749                 lru_add_drain_all();  /* drain pagevecs to lru lists */
750
751                 shp = shm_lock_check(ns, shmid);
752                 if (IS_ERR(shp)) {
753                         err = PTR_ERR(shp);
754                         goto out;
755                 }
756
757                 audit_ipc_obj(&(shp->shm_perm));
758
759                 if (!capable(CAP_IPC_LOCK)) {
760                         uid_t euid = current_euid();
761                         err = -EPERM;
762                         if (euid != shp->shm_perm.uid &&
763                             euid != shp->shm_perm.cuid)
764                                 goto out_unlock;
765                         if (cmd == SHM_LOCK &&
766                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
767                                 goto out_unlock;
768                 }
769
770                 err = security_shm_shmctl(shp, cmd);
771                 if (err)
772                         goto out_unlock;
773                 
774                 if(cmd==SHM_LOCK) {
775                         struct user_struct *user = current_user();
776                         if (!is_file_hugepages(shp->shm_file)) {
777                                 err = shmem_lock(shp->shm_file, 1, user);
778                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
779                                         shp->shm_perm.mode |= SHM_LOCKED;
780                                         shp->mlock_user = user;
781                                 }
782                         }
783                 } else if (!is_file_hugepages(shp->shm_file)) {
784                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
785                         shp->shm_perm.mode &= ~SHM_LOCKED;
786                         shp->mlock_user = NULL;
787                 }
788                 shm_unlock(shp);
789                 goto out;
790         }
791         case IPC_RMID:
792         case IPC_SET:
793                 err = shmctl_down(ns, shmid, cmd, buf, version);
794                 return err;
795         default:
796                 return -EINVAL;
797         }
798
799 out_unlock:
800         shm_unlock(shp);
801 out:
802         return err;
803 }
804
805 /*
806  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
807  *
808  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
809  * "raddr" thing points to kernel space, and there has to be a wrapper around
810  * this.
811  */
812 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
813 {
814         struct shmid_kernel *shp;
815         unsigned long addr;
816         unsigned long size;
817         struct file * file;
818         int    err;
819         unsigned long flags;
820         unsigned long prot;
821         int acc_mode;
822         unsigned long user_addr;
823         struct ipc_namespace *ns;
824         struct shm_file_data *sfd;
825         struct path path;
826         fmode_t f_mode;
827
828         err = -EINVAL;
829         if (shmid < 0)
830                 goto out;
831         else if ((addr = (ulong)shmaddr)) {
832                 if (addr & (SHMLBA-1)) {
833                         if (shmflg & SHM_RND)
834                                 addr &= ~(SHMLBA-1);       /* round down */
835                         else
836 #ifndef __ARCH_FORCE_SHMLBA
837                                 if (addr & ~PAGE_MASK)
838 #endif
839                                         goto out;
840                 }
841                 flags = MAP_SHARED | MAP_FIXED;
842         } else {
843                 if ((shmflg & SHM_REMAP))
844                         goto out;
845
846                 flags = MAP_SHARED;
847         }
848
849         if (shmflg & SHM_RDONLY) {
850                 prot = PROT_READ;
851                 acc_mode = S_IRUGO;
852                 f_mode = FMODE_READ;
853         } else {
854                 prot = PROT_READ | PROT_WRITE;
855                 acc_mode = S_IRUGO | S_IWUGO;
856                 f_mode = FMODE_READ | FMODE_WRITE;
857         }
858         if (shmflg & SHM_EXEC) {
859                 prot |= PROT_EXEC;
860                 acc_mode |= S_IXUGO;
861         }
862
863         /*
864          * We cannot rely on the fs check since SYSV IPC does have an
865          * additional creator id...
866          */
867         ns = current->nsproxy->ipc_ns;
868         shp = shm_lock_check(ns, shmid);
869         if (IS_ERR(shp)) {
870                 err = PTR_ERR(shp);
871                 goto out;
872         }
873
874         err = -EACCES;
875         if (ipcperms(&shp->shm_perm, acc_mode))
876                 goto out_unlock;
877
878         err = security_shm_shmat(shp, shmaddr, shmflg);
879         if (err)
880                 goto out_unlock;
881
882         path.dentry = dget(shp->shm_file->f_path.dentry);
883         path.mnt    = shp->shm_file->f_path.mnt;
884         shp->shm_nattch++;
885         size = i_size_read(path.dentry->d_inode);
886         shm_unlock(shp);
887
888         err = -ENOMEM;
889         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
890         if (!sfd)
891                 goto out_put_dentry;
892
893         file = alloc_file(path.mnt, path.dentry, f_mode,
894                         is_file_hugepages(shp->shm_file) ?
895                                 &shm_file_operations_huge :
896                                 &shm_file_operations);
897         if (!file)
898                 goto out_free;
899         ima_counts_get(file);
900
901         file->private_data = sfd;
902         file->f_mapping = shp->shm_file->f_mapping;
903         sfd->id = shp->shm_perm.id;
904         sfd->ns = get_ipc_ns(ns);
905         sfd->file = shp->shm_file;
906         sfd->vm_ops = NULL;
907
908         down_write(&current->mm->mmap_sem);
909         if (addr && !(shmflg & SHM_REMAP)) {
910                 err = -EINVAL;
911                 if (find_vma_intersection(current->mm, addr, addr + size))
912                         goto invalid;
913                 /*
914                  * If shm segment goes below stack, make sure there is some
915                  * space left for the stack to grow (at least 4 pages).
916                  */
917                 if (addr < current->mm->start_stack &&
918                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
919                         goto invalid;
920         }
921                 
922         user_addr = do_mmap (file, addr, size, prot, flags, 0);
923         *raddr = user_addr;
924         err = 0;
925         if (IS_ERR_VALUE(user_addr))
926                 err = (long)user_addr;
927 invalid:
928         up_write(&current->mm->mmap_sem);
929
930         fput(file);
931
932 out_nattch:
933         down_write(&shm_ids(ns).rw_mutex);
934         shp = shm_lock(ns, shmid);
935         BUG_ON(IS_ERR(shp));
936         shp->shm_nattch--;
937         if(shp->shm_nattch == 0 &&
938            shp->shm_perm.mode & SHM_DEST)
939                 shm_destroy(ns, shp);
940         else
941                 shm_unlock(shp);
942         up_write(&shm_ids(ns).rw_mutex);
943
944 out:
945         return err;
946
947 out_unlock:
948         shm_unlock(shp);
949         goto out;
950
951 out_free:
952         kfree(sfd);
953 out_put_dentry:
954         dput(path.dentry);
955         goto out_nattch;
956 }
957
958 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
959 {
960         unsigned long ret;
961         long err;
962
963         err = do_shmat(shmid, shmaddr, shmflg, &ret);
964         if (err)
965                 return err;
966         force_successful_syscall_return();
967         return (long)ret;
968 }
969
970 /*
971  * detach and kill segment if marked destroyed.
972  * The work is done in shm_close.
973  */
974 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
975 {
976         struct mm_struct *mm = current->mm;
977         struct vm_area_struct *vma;
978         unsigned long addr = (unsigned long)shmaddr;
979         int retval = -EINVAL;
980 #ifdef CONFIG_MMU
981         loff_t size = 0;
982         struct vm_area_struct *next;
983 #endif
984
985         if (addr & ~PAGE_MASK)
986                 return retval;
987
988         down_write(&mm->mmap_sem);
989
990         /*
991          * This function tries to be smart and unmap shm segments that
992          * were modified by partial mlock or munmap calls:
993          * - It first determines the size of the shm segment that should be
994          *   unmapped: It searches for a vma that is backed by shm and that
995          *   started at address shmaddr. It records it's size and then unmaps
996          *   it.
997          * - Then it unmaps all shm vmas that started at shmaddr and that
998          *   are within the initially determined size.
999          * Errors from do_munmap are ignored: the function only fails if
1000          * it's called with invalid parameters or if it's called to unmap
1001          * a part of a vma. Both calls in this function are for full vmas,
1002          * the parameters are directly copied from the vma itself and always
1003          * valid - therefore do_munmap cannot fail. (famous last words?)
1004          */
1005         /*
1006          * If it had been mremap()'d, the starting address would not
1007          * match the usual checks anyway. So assume all vma's are
1008          * above the starting address given.
1009          */
1010         vma = find_vma(mm, addr);
1011
1012 #ifdef CONFIG_MMU
1013         while (vma) {
1014                 next = vma->vm_next;
1015
1016                 /*
1017                  * Check if the starting address would match, i.e. it's
1018                  * a fragment created by mprotect() and/or munmap(), or it
1019                  * otherwise it starts at this address with no hassles.
1020                  */
1021                 if ((vma->vm_ops == &shm_vm_ops) &&
1022                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1023
1024
1025                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1026                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1027                         /*
1028                          * We discovered the size of the shm segment, so
1029                          * break out of here and fall through to the next
1030                          * loop that uses the size information to stop
1031                          * searching for matching vma's.
1032                          */
1033                         retval = 0;
1034                         vma = next;
1035                         break;
1036                 }
1037                 vma = next;
1038         }
1039
1040         /*
1041          * We need look no further than the maximum address a fragment
1042          * could possibly have landed at. Also cast things to loff_t to
1043          * prevent overflows and make comparisions vs. equal-width types.
1044          */
1045         size = PAGE_ALIGN(size);
1046         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1047                 next = vma->vm_next;
1048
1049                 /* finding a matching vma now does not alter retval */
1050                 if ((vma->vm_ops == &shm_vm_ops) &&
1051                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1052
1053                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1054                 vma = next;
1055         }
1056
1057 #else /* CONFIG_MMU */
1058         /* under NOMMU conditions, the exact address to be destroyed must be
1059          * given */
1060         retval = -EINVAL;
1061         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1062                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1063                 retval = 0;
1064         }
1065
1066 #endif
1067
1068         up_write(&mm->mmap_sem);
1069         return retval;
1070 }
1071
1072 #ifdef CONFIG_PROC_FS
1073 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1074 {
1075         struct shmid_kernel *shp = it;
1076
1077 #if BITS_PER_LONG <= 32
1078 #define SIZE_SPEC "%10lu"
1079 #else
1080 #define SIZE_SPEC "%21lu"
1081 #endif
1082
1083         return seq_printf(s,
1084                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1085                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1086                           shp->shm_perm.key,
1087                           shp->shm_perm.id,
1088                           shp->shm_perm.mode,
1089                           shp->shm_segsz,
1090                           shp->shm_cprid,
1091                           shp->shm_lprid,
1092                           shp->shm_nattch,
1093                           shp->shm_perm.uid,
1094                           shp->shm_perm.gid,
1095                           shp->shm_perm.cuid,
1096                           shp->shm_perm.cgid,
1097                           shp->shm_atim,
1098                           shp->shm_dtim,
1099                           shp->shm_ctim);
1100 }
1101 #endif