ipc: make shm_get_stat() more robust
[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 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 }
105 #endif
106
107 void __init shm_init (void)
108 {
109         shm_init_ns(&init_ipc_ns);
110         ipc_init_proc_interface("sysvipc/shm",
111                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
112                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
113 }
114
115 /*
116  * shm_lock_(check_) routines are called in the paths where the rw_mutex
117  * is not necessarily held.
118  */
119 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
120 {
121         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
122
123         if (IS_ERR(ipcp))
124                 return (struct shmid_kernel *)ipcp;
125
126         return container_of(ipcp, struct shmid_kernel, shm_perm);
127 }
128
129 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
130                                                 int id)
131 {
132         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
133
134         if (IS_ERR(ipcp))
135                 return (struct shmid_kernel *)ipcp;
136
137         return container_of(ipcp, struct shmid_kernel, shm_perm);
138 }
139
140 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
141 {
142         ipc_rmid(&shm_ids(ns), &s->shm_perm);
143 }
144
145
146 /* This is called by fork, once for every shm attach. */
147 static void shm_open(struct vm_area_struct *vma)
148 {
149         struct file *file = vma->vm_file;
150         struct shm_file_data *sfd = shm_file_data(file);
151         struct shmid_kernel *shp;
152
153         shp = shm_lock(sfd->ns, sfd->id);
154         BUG_ON(IS_ERR(shp));
155         shp->shm_atim = get_seconds();
156         shp->shm_lprid = task_tgid_vnr(current);
157         shp->shm_nattch++;
158         shm_unlock(shp);
159 }
160
161 /*
162  * shm_destroy - free the struct shmid_kernel
163  *
164  * @ns: namespace
165  * @shp: struct to free
166  *
167  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
168  * but returns with shp unlocked and freed.
169  */
170 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
171 {
172         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
173         shm_rmid(ns, shp);
174         shm_unlock(shp);
175         if (!is_file_hugepages(shp->shm_file))
176                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
177         else
178                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
179                                                 shp->mlock_user);
180         fput (shp->shm_file);
181         security_shm_free(shp);
182         ipc_rcu_putref(shp);
183 }
184
185 /*
186  * remove the attach descriptor vma.
187  * free memory for segment if it is marked destroyed.
188  * The descriptor has already been removed from the current->mm->mmap list
189  * and will later be kfree()d.
190  */
191 static void shm_close(struct vm_area_struct *vma)
192 {
193         struct file * file = vma->vm_file;
194         struct shm_file_data *sfd = shm_file_data(file);
195         struct shmid_kernel *shp;
196         struct ipc_namespace *ns = sfd->ns;
197
198         down_write(&shm_ids(ns).rw_mutex);
199         /* remove from the list of attaches of the shm segment */
200         shp = shm_lock(ns, sfd->id);
201         BUG_ON(IS_ERR(shp));
202         shp->shm_lprid = task_tgid_vnr(current);
203         shp->shm_dtim = get_seconds();
204         shp->shm_nattch--;
205         if(shp->shm_nattch == 0 &&
206            shp->shm_perm.mode & SHM_DEST)
207                 shm_destroy(ns, shp);
208         else
209                 shm_unlock(shp);
210         up_write(&shm_ids(ns).rw_mutex);
211 }
212
213 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
214 {
215         struct file *file = vma->vm_file;
216         struct shm_file_data *sfd = shm_file_data(file);
217
218         return sfd->vm_ops->fault(vma, vmf);
219 }
220
221 #ifdef CONFIG_NUMA
222 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
223 {
224         struct file *file = vma->vm_file;
225         struct shm_file_data *sfd = shm_file_data(file);
226         int err = 0;
227         if (sfd->vm_ops->set_policy)
228                 err = sfd->vm_ops->set_policy(vma, new);
229         return err;
230 }
231
232 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
233                                         unsigned long addr)
234 {
235         struct file *file = vma->vm_file;
236         struct shm_file_data *sfd = shm_file_data(file);
237         struct mempolicy *pol = NULL;
238
239         if (sfd->vm_ops->get_policy)
240                 pol = sfd->vm_ops->get_policy(vma, addr);
241         else if (vma->vm_policy)
242                 pol = vma->vm_policy;
243
244         return pol;
245 }
246 #endif
247
248 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
249 {
250         struct shm_file_data *sfd = shm_file_data(file);
251         int ret;
252
253         ret = sfd->file->f_op->mmap(sfd->file, vma);
254         if (ret != 0)
255                 return ret;
256         sfd->vm_ops = vma->vm_ops;
257 #ifdef CONFIG_MMU
258         BUG_ON(!sfd->vm_ops->fault);
259 #endif
260         vma->vm_ops = &shm_vm_ops;
261         shm_open(vma);
262
263         return ret;
264 }
265
266 static int shm_release(struct inode *ino, struct file *file)
267 {
268         struct shm_file_data *sfd = shm_file_data(file);
269
270         put_ipc_ns(sfd->ns);
271         shm_file_data(file) = NULL;
272         kfree(sfd);
273         return 0;
274 }
275
276 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
277 {
278         int (*fsync) (struct file *, struct dentry *, int datasync);
279         struct shm_file_data *sfd = shm_file_data(file);
280         int ret = -EINVAL;
281
282         fsync = sfd->file->f_op->fsync;
283         if (fsync)
284                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
285         return ret;
286 }
287
288 static unsigned long shm_get_unmapped_area(struct file *file,
289         unsigned long addr, unsigned long len, unsigned long pgoff,
290         unsigned long flags)
291 {
292         struct shm_file_data *sfd = shm_file_data(file);
293         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
294 }
295
296 int is_file_shm_hugepages(struct file *file)
297 {
298         int ret = 0;
299
300         if (file->f_op == &shm_file_operations) {
301                 struct shm_file_data *sfd;
302                 sfd = shm_file_data(file);
303                 ret = is_file_hugepages(sfd->file);
304         }
305         return ret;
306 }
307
308 static const struct file_operations shm_file_operations = {
309         .mmap           = shm_mmap,
310         .fsync          = shm_fsync,
311         .release        = shm_release,
312         .get_unmapped_area      = shm_get_unmapped_area,
313 };
314
315 static struct vm_operations_struct shm_vm_ops = {
316         .open   = shm_open,     /* callback for a new vm-area open */
317         .close  = shm_close,    /* callback for when the vm-area is released */
318         .fault  = shm_fault,
319 #if defined(CONFIG_NUMA)
320         .set_policy = shm_set_policy,
321         .get_policy = shm_get_policy,
322 #endif
323 };
324
325 /**
326  * newseg - Create a new shared memory segment
327  * @ns: namespace
328  * @params: ptr to the structure that contains key, size and shmflg
329  *
330  * Called with shm_ids.rw_mutex held as a writer.
331  */
332
333 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
334 {
335         key_t key = params->key;
336         int shmflg = params->flg;
337         size_t size = params->u.size;
338         int error;
339         struct shmid_kernel *shp;
340         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
341         struct file * file;
342         char name[13];
343         int id;
344         int acctflag = 0;
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         sprintf (name, "SYSV%08x", key);
368         if (shmflg & SHM_HUGETLB) {
369                 /* hugetlb_file_setup applies strict accounting */
370                 if (shmflg & SHM_NORESERVE)
371                         acctflag = VM_NORESERVE;
372                 file = hugetlb_file_setup(name, size, acctflag);
373                 shp->mlock_user = current_user();
374         } else {
375                 /*
376                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
377                  * if it's asked for.
378                  */
379                 if  ((shmflg & SHM_NORESERVE) &&
380                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
381                         acctflag = VM_NORESERVE;
382                 file = shmem_file_setup(name, size, acctflag);
383         }
384         error = PTR_ERR(file);
385         if (IS_ERR(file))
386                 goto no_file;
387         ima_shm_check(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         fput(file);
415 no_file:
416         security_shm_free(shp);
417         ipc_rcu_putref(shp);
418         return error;
419 }
420
421 /*
422  * Called with shm_ids.rw_mutex and ipcp locked.
423  */
424 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
425 {
426         struct shmid_kernel *shp;
427
428         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
429         return security_shm_associate(shp, shmflg);
430 }
431
432 /*
433  * Called with shm_ids.rw_mutex and ipcp locked.
434  */
435 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
436                                 struct ipc_params *params)
437 {
438         struct shmid_kernel *shp;
439
440         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
441         if (shp->shm_segsz < params->u.size)
442                 return -EINVAL;
443
444         return 0;
445 }
446
447 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
448 {
449         struct ipc_namespace *ns;
450         struct ipc_ops shm_ops;
451         struct ipc_params shm_params;
452
453         ns = current->nsproxy->ipc_ns;
454
455         shm_ops.getnew = newseg;
456         shm_ops.associate = shm_security;
457         shm_ops.more_checks = shm_more_checks;
458
459         shm_params.key = key;
460         shm_params.flg = shmflg;
461         shm_params.u.size = size;
462
463         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
464 }
465
466 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
467 {
468         switch(version) {
469         case IPC_64:
470                 return copy_to_user(buf, in, sizeof(*in));
471         case IPC_OLD:
472             {
473                 struct shmid_ds out;
474
475                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
476                 out.shm_segsz   = in->shm_segsz;
477                 out.shm_atime   = in->shm_atime;
478                 out.shm_dtime   = in->shm_dtime;
479                 out.shm_ctime   = in->shm_ctime;
480                 out.shm_cpid    = in->shm_cpid;
481                 out.shm_lpid    = in->shm_lpid;
482                 out.shm_nattch  = in->shm_nattch;
483
484                 return copy_to_user(buf, &out, sizeof(out));
485             }
486         default:
487                 return -EINVAL;
488         }
489 }
490
491 static inline unsigned long
492 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
493 {
494         switch(version) {
495         case IPC_64:
496                 if (copy_from_user(out, buf, sizeof(*out)))
497                         return -EFAULT;
498                 return 0;
499         case IPC_OLD:
500             {
501                 struct shmid_ds tbuf_old;
502
503                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
504                         return -EFAULT;
505
506                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
507                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
508                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
509
510                 return 0;
511             }
512         default:
513                 return -EINVAL;
514         }
515 }
516
517 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
518 {
519         switch(version) {
520         case IPC_64:
521                 return copy_to_user(buf, in, sizeof(*in));
522         case IPC_OLD:
523             {
524                 struct shminfo out;
525
526                 if(in->shmmax > INT_MAX)
527                         out.shmmax = INT_MAX;
528                 else
529                         out.shmmax = (int)in->shmmax;
530
531                 out.shmmin      = in->shmmin;
532                 out.shmmni      = in->shmmni;
533                 out.shmseg      = in->shmseg;
534                 out.shmall      = in->shmall; 
535
536                 return copy_to_user(buf, &out, sizeof(out));
537             }
538         default:
539                 return -EINVAL;
540         }
541 }
542
543 /*
544  * Called with shm_ids.rw_mutex held as a reader
545  */
546 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
547                 unsigned long *swp)
548 {
549         int next_id;
550         int total, in_use;
551
552         *rss = 0;
553         *swp = 0;
554
555         in_use = shm_ids(ns).in_use;
556
557         for (total = 0, next_id = 0; total < in_use; next_id++) {
558                 struct kern_ipc_perm *ipc;
559                 struct shmid_kernel *shp;
560                 struct inode *inode;
561
562                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
563                 if (ipc == NULL)
564                         continue;
565                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
566
567                 inode = shp->shm_file->f_path.dentry->d_inode;
568
569                 if (is_file_hugepages(shp->shm_file)) {
570                         struct address_space *mapping = inode->i_mapping;
571                         struct hstate *h = hstate_file(shp->shm_file);
572                         *rss += pages_per_huge_page(h) * mapping->nrpages;
573                 } else {
574 #ifdef CONFIG_SHMEM
575                         struct shmem_inode_info *info = SHMEM_I(inode);
576                         spin_lock(&info->lock);
577                         *rss += inode->i_mapping->nrpages;
578                         *swp += info->swapped;
579                         spin_unlock(&info->lock);
580 #else
581                         *rss += inode->i_mapping->nrpages;
582 #endif
583                 }
584
585                 total++;
586         }
587 }
588
589 /*
590  * This function handles some shmctl commands which require the rw_mutex
591  * to be held in write mode.
592  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
593  */
594 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
595                        struct shmid_ds __user *buf, int version)
596 {
597         struct kern_ipc_perm *ipcp;
598         struct shmid64_ds shmid64;
599         struct shmid_kernel *shp;
600         int err;
601
602         if (cmd == IPC_SET) {
603                 if (copy_shmid_from_user(&shmid64, buf, version))
604                         return -EFAULT;
605         }
606
607         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
608         if (IS_ERR(ipcp))
609                 return PTR_ERR(ipcp);
610
611         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
612
613         err = security_shm_shmctl(shp, cmd);
614         if (err)
615                 goto out_unlock;
616         switch (cmd) {
617         case IPC_RMID:
618                 do_shm_rmid(ns, ipcp);
619                 goto out_up;
620         case IPC_SET:
621                 ipc_update_perm(&shmid64.shm_perm, ipcp);
622                 shp->shm_ctim = get_seconds();
623                 break;
624         default:
625                 err = -EINVAL;
626         }
627 out_unlock:
628         shm_unlock(shp);
629 out_up:
630         up_write(&shm_ids(ns).rw_mutex);
631         return err;
632 }
633
634 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
635 {
636         struct shmid_kernel *shp;
637         int err, version;
638         struct ipc_namespace *ns;
639
640         if (cmd < 0 || shmid < 0) {
641                 err = -EINVAL;
642                 goto out;
643         }
644
645         version = ipc_parse_version(&cmd);
646         ns = current->nsproxy->ipc_ns;
647
648         switch (cmd) { /* replace with proc interface ? */
649         case IPC_INFO:
650         {
651                 struct shminfo64 shminfo;
652
653                 err = security_shm_shmctl(NULL, cmd);
654                 if (err)
655                         return err;
656
657                 memset(&shminfo, 0, sizeof(shminfo));
658                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
659                 shminfo.shmmax = ns->shm_ctlmax;
660                 shminfo.shmall = ns->shm_ctlall;
661
662                 shminfo.shmmin = SHMMIN;
663                 if(copy_shminfo_to_user (buf, &shminfo, version))
664                         return -EFAULT;
665
666                 down_read(&shm_ids(ns).rw_mutex);
667                 err = ipc_get_maxid(&shm_ids(ns));
668                 up_read(&shm_ids(ns).rw_mutex);
669
670                 if(err<0)
671                         err = 0;
672                 goto out;
673         }
674         case SHM_INFO:
675         {
676                 struct shm_info shm_info;
677
678                 err = security_shm_shmctl(NULL, cmd);
679                 if (err)
680                         return err;
681
682                 memset(&shm_info, 0, sizeof(shm_info));
683                 down_read(&shm_ids(ns).rw_mutex);
684                 shm_info.used_ids = shm_ids(ns).in_use;
685                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
686                 shm_info.shm_tot = ns->shm_tot;
687                 shm_info.swap_attempts = 0;
688                 shm_info.swap_successes = 0;
689                 err = ipc_get_maxid(&shm_ids(ns));
690                 up_read(&shm_ids(ns).rw_mutex);
691                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
692                         err = -EFAULT;
693                         goto out;
694                 }
695
696                 err = err < 0 ? 0 : err;
697                 goto out;
698         }
699         case SHM_STAT:
700         case IPC_STAT:
701         {
702                 struct shmid64_ds tbuf;
703                 int result;
704
705                 if (cmd == SHM_STAT) {
706                         shp = shm_lock(ns, shmid);
707                         if (IS_ERR(shp)) {
708                                 err = PTR_ERR(shp);
709                                 goto out;
710                         }
711                         result = shp->shm_perm.id;
712                 } else {
713                         shp = shm_lock_check(ns, shmid);
714                         if (IS_ERR(shp)) {
715                                 err = PTR_ERR(shp);
716                                 goto out;
717                         }
718                         result = 0;
719                 }
720                 err = -EACCES;
721                 if (ipcperms (&shp->shm_perm, S_IRUGO))
722                         goto out_unlock;
723                 err = security_shm_shmctl(shp, cmd);
724                 if (err)
725                         goto out_unlock;
726                 memset(&tbuf, 0, sizeof(tbuf));
727                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
728                 tbuf.shm_segsz  = shp->shm_segsz;
729                 tbuf.shm_atime  = shp->shm_atim;
730                 tbuf.shm_dtime  = shp->shm_dtim;
731                 tbuf.shm_ctime  = shp->shm_ctim;
732                 tbuf.shm_cpid   = shp->shm_cprid;
733                 tbuf.shm_lpid   = shp->shm_lprid;
734                 tbuf.shm_nattch = shp->shm_nattch;
735                 shm_unlock(shp);
736                 if(copy_shmid_to_user (buf, &tbuf, version))
737                         err = -EFAULT;
738                 else
739                         err = result;
740                 goto out;
741         }
742         case SHM_LOCK:
743         case SHM_UNLOCK:
744         {
745                 struct file *uninitialized_var(shm_file);
746
747                 lru_add_drain_all();  /* drain pagevecs to lru lists */
748
749                 shp = shm_lock_check(ns, shmid);
750                 if (IS_ERR(shp)) {
751                         err = PTR_ERR(shp);
752                         goto out;
753                 }
754
755                 audit_ipc_obj(&(shp->shm_perm));
756
757                 if (!capable(CAP_IPC_LOCK)) {
758                         uid_t euid = current_euid();
759                         err = -EPERM;
760                         if (euid != shp->shm_perm.uid &&
761                             euid != shp->shm_perm.cuid)
762                                 goto out_unlock;
763                         if (cmd == SHM_LOCK &&
764                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
765                                 goto out_unlock;
766                 }
767
768                 err = security_shm_shmctl(shp, cmd);
769                 if (err)
770                         goto out_unlock;
771                 
772                 if(cmd==SHM_LOCK) {
773                         struct user_struct *user = current_user();
774                         if (!is_file_hugepages(shp->shm_file)) {
775                                 err = shmem_lock(shp->shm_file, 1, user);
776                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
777                                         shp->shm_perm.mode |= SHM_LOCKED;
778                                         shp->mlock_user = user;
779                                 }
780                         }
781                 } else if (!is_file_hugepages(shp->shm_file)) {
782                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
783                         shp->shm_perm.mode &= ~SHM_LOCKED;
784                         shp->mlock_user = NULL;
785                 }
786                 shm_unlock(shp);
787                 goto out;
788         }
789         case IPC_RMID:
790         case IPC_SET:
791                 err = shmctl_down(ns, shmid, cmd, buf, version);
792                 return err;
793         default:
794                 return -EINVAL;
795         }
796
797 out_unlock:
798         shm_unlock(shp);
799 out:
800         return err;
801 }
802
803 /*
804  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
805  *
806  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
807  * "raddr" thing points to kernel space, and there has to be a wrapper around
808  * this.
809  */
810 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
811 {
812         struct shmid_kernel *shp;
813         unsigned long addr;
814         unsigned long size;
815         struct file * file;
816         int    err;
817         unsigned long flags;
818         unsigned long prot;
819         int acc_mode;
820         unsigned long user_addr;
821         struct ipc_namespace *ns;
822         struct shm_file_data *sfd;
823         struct path path;
824         fmode_t f_mode;
825
826         err = -EINVAL;
827         if (shmid < 0)
828                 goto out;
829         else if ((addr = (ulong)shmaddr)) {
830                 if (addr & (SHMLBA-1)) {
831                         if (shmflg & SHM_RND)
832                                 addr &= ~(SHMLBA-1);       /* round down */
833                         else
834 #ifndef __ARCH_FORCE_SHMLBA
835                                 if (addr & ~PAGE_MASK)
836 #endif
837                                         goto out;
838                 }
839                 flags = MAP_SHARED | MAP_FIXED;
840         } else {
841                 if ((shmflg & SHM_REMAP))
842                         goto out;
843
844                 flags = MAP_SHARED;
845         }
846
847         if (shmflg & SHM_RDONLY) {
848                 prot = PROT_READ;
849                 acc_mode = S_IRUGO;
850                 f_mode = FMODE_READ;
851         } else {
852                 prot = PROT_READ | PROT_WRITE;
853                 acc_mode = S_IRUGO | S_IWUGO;
854                 f_mode = FMODE_READ | FMODE_WRITE;
855         }
856         if (shmflg & SHM_EXEC) {
857                 prot |= PROT_EXEC;
858                 acc_mode |= S_IXUGO;
859         }
860
861         /*
862          * We cannot rely on the fs check since SYSV IPC does have an
863          * additional creator id...
864          */
865         ns = current->nsproxy->ipc_ns;
866         shp = shm_lock_check(ns, shmid);
867         if (IS_ERR(shp)) {
868                 err = PTR_ERR(shp);
869                 goto out;
870         }
871
872         err = -EACCES;
873         if (ipcperms(&shp->shm_perm, acc_mode))
874                 goto out_unlock;
875
876         err = security_shm_shmat(shp, shmaddr, shmflg);
877         if (err)
878                 goto out_unlock;
879
880         path.dentry = dget(shp->shm_file->f_path.dentry);
881         path.mnt    = shp->shm_file->f_path.mnt;
882         shp->shm_nattch++;
883         size = i_size_read(path.dentry->d_inode);
884         shm_unlock(shp);
885
886         err = -ENOMEM;
887         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
888         if (!sfd)
889                 goto out_put_dentry;
890
891         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
892         if (!file)
893                 goto out_free;
894         ima_shm_check(file);
895
896         file->private_data = sfd;
897         file->f_mapping = shp->shm_file->f_mapping;
898         sfd->id = shp->shm_perm.id;
899         sfd->ns = get_ipc_ns(ns);
900         sfd->file = shp->shm_file;
901         sfd->vm_ops = NULL;
902
903         down_write(&current->mm->mmap_sem);
904         if (addr && !(shmflg & SHM_REMAP)) {
905                 err = -EINVAL;
906                 if (find_vma_intersection(current->mm, addr, addr + size))
907                         goto invalid;
908                 /*
909                  * If shm segment goes below stack, make sure there is some
910                  * space left for the stack to grow (at least 4 pages).
911                  */
912                 if (addr < current->mm->start_stack &&
913                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
914                         goto invalid;
915         }
916                 
917         user_addr = do_mmap (file, addr, size, prot, flags, 0);
918         *raddr = user_addr;
919         err = 0;
920         if (IS_ERR_VALUE(user_addr))
921                 err = (long)user_addr;
922 invalid:
923         up_write(&current->mm->mmap_sem);
924
925         fput(file);
926
927 out_nattch:
928         down_write(&shm_ids(ns).rw_mutex);
929         shp = shm_lock(ns, shmid);
930         BUG_ON(IS_ERR(shp));
931         shp->shm_nattch--;
932         if(shp->shm_nattch == 0 &&
933            shp->shm_perm.mode & SHM_DEST)
934                 shm_destroy(ns, shp);
935         else
936                 shm_unlock(shp);
937         up_write(&shm_ids(ns).rw_mutex);
938
939 out:
940         return err;
941
942 out_unlock:
943         shm_unlock(shp);
944         goto out;
945
946 out_free:
947         kfree(sfd);
948 out_put_dentry:
949         dput(path.dentry);
950         goto out_nattch;
951 }
952
953 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
954 {
955         unsigned long ret;
956         long err;
957
958         err = do_shmat(shmid, shmaddr, shmflg, &ret);
959         if (err)
960                 return err;
961         force_successful_syscall_return();
962         return (long)ret;
963 }
964
965 /*
966  * detach and kill segment if marked destroyed.
967  * The work is done in shm_close.
968  */
969 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
970 {
971         struct mm_struct *mm = current->mm;
972         struct vm_area_struct *vma, *next;
973         unsigned long addr = (unsigned long)shmaddr;
974         loff_t size = 0;
975         int retval = -EINVAL;
976
977         if (addr & ~PAGE_MASK)
978                 return retval;
979
980         down_write(&mm->mmap_sem);
981
982         /*
983          * This function tries to be smart and unmap shm segments that
984          * were modified by partial mlock or munmap calls:
985          * - It first determines the size of the shm segment that should be
986          *   unmapped: It searches for a vma that is backed by shm and that
987          *   started at address shmaddr. It records it's size and then unmaps
988          *   it.
989          * - Then it unmaps all shm vmas that started at shmaddr and that
990          *   are within the initially determined size.
991          * Errors from do_munmap are ignored: the function only fails if
992          * it's called with invalid parameters or if it's called to unmap
993          * a part of a vma. Both calls in this function are for full vmas,
994          * the parameters are directly copied from the vma itself and always
995          * valid - therefore do_munmap cannot fail. (famous last words?)
996          */
997         /*
998          * If it had been mremap()'d, the starting address would not
999          * match the usual checks anyway. So assume all vma's are
1000          * above the starting address given.
1001          */
1002         vma = find_vma(mm, addr);
1003
1004 #ifdef CONFIG_MMU
1005         while (vma) {
1006                 next = vma->vm_next;
1007
1008                 /*
1009                  * Check if the starting address would match, i.e. it's
1010                  * a fragment created by mprotect() and/or munmap(), or it
1011                  * otherwise it starts at this address with no hassles.
1012                  */
1013                 if ((vma->vm_ops == &shm_vm_ops) &&
1014                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1015
1016
1017                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1018                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1019                         /*
1020                          * We discovered the size of the shm segment, so
1021                          * break out of here and fall through to the next
1022                          * loop that uses the size information to stop
1023                          * searching for matching vma's.
1024                          */
1025                         retval = 0;
1026                         vma = next;
1027                         break;
1028                 }
1029                 vma = next;
1030         }
1031
1032         /*
1033          * We need look no further than the maximum address a fragment
1034          * could possibly have landed at. Also cast things to loff_t to
1035          * prevent overflows and make comparisions vs. equal-width types.
1036          */
1037         size = PAGE_ALIGN(size);
1038         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1039                 next = vma->vm_next;
1040
1041                 /* finding a matching vma now does not alter retval */
1042                 if ((vma->vm_ops == &shm_vm_ops) &&
1043                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1044
1045                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1046                 vma = next;
1047         }
1048
1049 #else /* CONFIG_MMU */
1050         /* under NOMMU conditions, the exact address to be destroyed must be
1051          * given */
1052         retval = -EINVAL;
1053         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1054                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1055                 retval = 0;
1056         }
1057
1058 #endif
1059
1060         up_write(&mm->mmap_sem);
1061         return retval;
1062 }
1063
1064 #ifdef CONFIG_PROC_FS
1065 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1066 {
1067         struct shmid_kernel *shp = it;
1068
1069 #if BITS_PER_LONG <= 32
1070 #define SIZE_SPEC "%10lu"
1071 #else
1072 #define SIZE_SPEC "%21lu"
1073 #endif
1074
1075         return seq_printf(s,
1076                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1077                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1078                           shp->shm_perm.key,
1079                           shp->shm_perm.id,
1080                           shp->shm_perm.mode,
1081                           shp->shm_segsz,
1082                           shp->shm_cprid,
1083                           shp->shm_lprid,
1084                           shp->shm_nattch,
1085                           shp->shm_perm.uid,
1086                           shp->shm_perm.gid,
1087                           shp->shm_perm.cuid,
1088                           shp->shm_perm.cgid,
1089                           shp->shm_atim,
1090                           shp->shm_dtim,
1091                           shp->shm_ctim);
1092 }
1093 #endif