]> git.karo-electronics.de Git - karo-tx-linux.git/blob - ipc/shm.c
cciss: fix info leak in cciss_ioctl32_passthru()
[karo-tx-linux.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  * Better ipc lock (kern_ipc_perm.lock) handling
24  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
25  */
26
27 #include <linux/slab.h>
28 #include <linux/mm.h>
29 #include <linux/hugetlb.h>
30 #include <linux/shm.h>
31 #include <linux/init.h>
32 #include <linux/file.h>
33 #include <linux/mman.h>
34 #include <linux/shmem_fs.h>
35 #include <linux/security.h>
36 #include <linux/syscalls.h>
37 #include <linux/audit.h>
38 #include <linux/capability.h>
39 #include <linux/ptrace.h>
40 #include <linux/seq_file.h>
41 #include <linux/rwsem.h>
42 #include <linux/nsproxy.h>
43 #include <linux/mount.h>
44 #include <linux/ipc_namespace.h>
45
46 #include <asm/uaccess.h>
47
48 #include "util.h"
49
50 struct shm_file_data {
51         int id;
52         struct ipc_namespace *ns;
53         struct file *file;
54         const struct vm_operations_struct *vm_ops;
55 };
56
57 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
58
59 static const struct file_operations shm_file_operations;
60 static const struct vm_operations_struct shm_vm_ops;
61
62 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
63
64 #define shm_unlock(shp)                 \
65         ipc_unlock(&(shp)->shm_perm)
66
67 static int newseg(struct ipc_namespace *, struct ipc_params *);
68 static void shm_open(struct vm_area_struct *vma);
69 static void shm_close(struct vm_area_struct *vma);
70 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
71 #ifdef CONFIG_PROC_FS
72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
73 #endif
74
75 void shm_init_ns(struct ipc_namespace *ns)
76 {
77         ns->shm_ctlmax = SHMMAX;
78         ns->shm_ctlall = SHMALL;
79         ns->shm_ctlmni = SHMMNI;
80         ns->shm_rmid_forced = 0;
81         ns->shm_tot = 0;
82         ipc_init_ids(&shm_ids(ns));
83 }
84
85 /*
86  * Called with shm_ids.rwsem (writer) and the shp structure locked.
87  * Only shm_ids.rwsem remains locked on exit.
88  */
89 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
90 {
91         struct shmid_kernel *shp;
92         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
93
94         if (shp->shm_nattch){
95                 shp->shm_perm.mode |= SHM_DEST;
96                 /* Do not find it any more */
97                 shp->shm_perm.key = IPC_PRIVATE;
98                 shm_unlock(shp);
99         } else
100                 shm_destroy(ns, shp);
101 }
102
103 #ifdef CONFIG_IPC_NS
104 void shm_exit_ns(struct ipc_namespace *ns)
105 {
106         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
107         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
108 }
109 #endif
110
111 static int __init ipc_ns_init(void)
112 {
113         shm_init_ns(&init_ipc_ns);
114         return 0;
115 }
116
117 pure_initcall(ipc_ns_init);
118
119 void __init shm_init (void)
120 {
121         ipc_init_proc_interface("sysvipc/shm",
122 #if BITS_PER_LONG <= 32
123                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
124 #else
125                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
126 #endif
127                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
128 }
129
130 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
131 {
132         struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
133
134         if (IS_ERR(ipcp))
135                 return ERR_CAST(ipcp);
136
137         return container_of(ipcp, struct shmid_kernel, shm_perm);
138 }
139
140 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
141 {
142         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
143
144         if (IS_ERR(ipcp))
145                 return ERR_CAST(ipcp);
146
147         return container_of(ipcp, struct shmid_kernel, shm_perm);
148 }
149
150 /*
151  * shm_lock_(check_) routines are called in the paths where the rwsem
152  * is not necessarily held.
153  */
154 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
155 {
156         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
157
158         if (IS_ERR(ipcp))
159                 return (struct shmid_kernel *)ipcp;
160
161         return container_of(ipcp, struct shmid_kernel, shm_perm);
162 }
163
164 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
165 {
166         rcu_read_lock();
167         ipc_lock_object(&ipcp->shm_perm);
168 }
169
170 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
171 {
172         ipc_rmid(&shm_ids(ns), &s->shm_perm);
173 }
174
175
176 /* This is called by fork, once for every shm attach. */
177 static void shm_open(struct vm_area_struct *vma)
178 {
179         struct file *file = vma->vm_file;
180         struct shm_file_data *sfd = shm_file_data(file);
181         struct shmid_kernel *shp;
182
183         shp = shm_lock(sfd->ns, sfd->id);
184         BUG_ON(IS_ERR(shp));
185         shp->shm_atim = get_seconds();
186         shp->shm_lprid = task_tgid_vnr(current);
187         shp->shm_nattch++;
188         shm_unlock(shp);
189 }
190
191 /*
192  * shm_destroy - free the struct shmid_kernel
193  *
194  * @ns: namespace
195  * @shp: struct to free
196  *
197  * It has to be called with shp and shm_ids.rwsem (writer) locked,
198  * but returns with shp unlocked and freed.
199  */
200 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
201 {
202         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
203         shm_rmid(ns, shp);
204         shm_unlock(shp);
205         if (!is_file_hugepages(shp->shm_file))
206                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
207         else if (shp->mlock_user)
208                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
209                                                 shp->mlock_user);
210         fput (shp->shm_file);
211         security_shm_free(shp);
212         ipc_rcu_putref(shp);
213 }
214
215 /*
216  * shm_may_destroy - identifies whether shm segment should be destroyed now
217  *
218  * Returns true if and only if there are no active users of the segment and
219  * one of the following is true:
220  *
221  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
222  *
223  * 2) sysctl kernel.shm_rmid_forced is set to 1.
224  */
225 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
226 {
227         return (shp->shm_nattch == 0) &&
228                (ns->shm_rmid_forced ||
229                 (shp->shm_perm.mode & SHM_DEST));
230 }
231
232 /*
233  * remove the attach descriptor vma.
234  * free memory for segment if it is marked destroyed.
235  * The descriptor has already been removed from the current->mm->mmap list
236  * and will later be kfree()d.
237  */
238 static void shm_close(struct vm_area_struct *vma)
239 {
240         struct file * file = vma->vm_file;
241         struct shm_file_data *sfd = shm_file_data(file);
242         struct shmid_kernel *shp;
243         struct ipc_namespace *ns = sfd->ns;
244
245         down_write(&shm_ids(ns).rwsem);
246         /* remove from the list of attaches of the shm segment */
247         shp = shm_lock(ns, sfd->id);
248         BUG_ON(IS_ERR(shp));
249         shp->shm_lprid = task_tgid_vnr(current);
250         shp->shm_dtim = get_seconds();
251         shp->shm_nattch--;
252         if (shm_may_destroy(ns, shp))
253                 shm_destroy(ns, shp);
254         else
255                 shm_unlock(shp);
256         up_write(&shm_ids(ns).rwsem);
257 }
258
259 /* Called with ns->shm_ids(ns).rwsem locked */
260 static int shm_try_destroy_current(int id, void *p, void *data)
261 {
262         struct ipc_namespace *ns = data;
263         struct kern_ipc_perm *ipcp = p;
264         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
265
266         if (shp->shm_creator != current)
267                 return 0;
268
269         /*
270          * Mark it as orphaned to destroy the segment when
271          * kernel.shm_rmid_forced is changed.
272          * It is noop if the following shm_may_destroy() returns true.
273          */
274         shp->shm_creator = NULL;
275
276         /*
277          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
278          * is not set, it shouldn't be deleted here.
279          */
280         if (!ns->shm_rmid_forced)
281                 return 0;
282
283         if (shm_may_destroy(ns, shp)) {
284                 shm_lock_by_ptr(shp);
285                 shm_destroy(ns, shp);
286         }
287         return 0;
288 }
289
290 /* Called with ns->shm_ids(ns).rwsem locked */
291 static int shm_try_destroy_orphaned(int id, void *p, void *data)
292 {
293         struct ipc_namespace *ns = data;
294         struct kern_ipc_perm *ipcp = p;
295         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
296
297         /*
298          * We want to destroy segments without users and with already
299          * exit'ed originating process.
300          *
301          * As shp->* are changed under rwsem, it's safe to skip shp locking.
302          */
303         if (shp->shm_creator != NULL)
304                 return 0;
305
306         if (shm_may_destroy(ns, shp)) {
307                 shm_lock_by_ptr(shp);
308                 shm_destroy(ns, shp);
309         }
310         return 0;
311 }
312
313 void shm_destroy_orphaned(struct ipc_namespace *ns)
314 {
315         down_write(&shm_ids(ns).rwsem);
316         if (shm_ids(ns).in_use)
317                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
318         up_write(&shm_ids(ns).rwsem);
319 }
320
321
322 void exit_shm(struct task_struct *task)
323 {
324         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
325
326         if (shm_ids(ns).in_use == 0)
327                 return;
328
329         /* Destroy all already created segments, but not mapped yet */
330         down_write(&shm_ids(ns).rwsem);
331         if (shm_ids(ns).in_use)
332                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
333         up_write(&shm_ids(ns).rwsem);
334 }
335
336 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
337 {
338         struct file *file = vma->vm_file;
339         struct shm_file_data *sfd = shm_file_data(file);
340
341         return sfd->vm_ops->fault(vma, vmf);
342 }
343
344 #ifdef CONFIG_NUMA
345 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
346 {
347         struct file *file = vma->vm_file;
348         struct shm_file_data *sfd = shm_file_data(file);
349         int err = 0;
350         if (sfd->vm_ops->set_policy)
351                 err = sfd->vm_ops->set_policy(vma, new);
352         return err;
353 }
354
355 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
356                                         unsigned long addr)
357 {
358         struct file *file = vma->vm_file;
359         struct shm_file_data *sfd = shm_file_data(file);
360         struct mempolicy *pol = NULL;
361
362         if (sfd->vm_ops->get_policy)
363                 pol = sfd->vm_ops->get_policy(vma, addr);
364         else if (vma->vm_policy)
365                 pol = vma->vm_policy;
366
367         return pol;
368 }
369 #endif
370
371 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
372 {
373         struct shm_file_data *sfd = shm_file_data(file);
374         int ret;
375
376         ret = sfd->file->f_op->mmap(sfd->file, vma);
377         if (ret != 0)
378                 return ret;
379         sfd->vm_ops = vma->vm_ops;
380 #ifdef CONFIG_MMU
381         BUG_ON(!sfd->vm_ops->fault);
382 #endif
383         vma->vm_ops = &shm_vm_ops;
384         shm_open(vma);
385
386         return ret;
387 }
388
389 static int shm_release(struct inode *ino, struct file *file)
390 {
391         struct shm_file_data *sfd = shm_file_data(file);
392
393         put_ipc_ns(sfd->ns);
394         shm_file_data(file) = NULL;
395         kfree(sfd);
396         return 0;
397 }
398
399 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
400 {
401         struct shm_file_data *sfd = shm_file_data(file);
402
403         if (!sfd->file->f_op->fsync)
404                 return -EINVAL;
405         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
406 }
407
408 static long shm_fallocate(struct file *file, int mode, loff_t offset,
409                           loff_t len)
410 {
411         struct shm_file_data *sfd = shm_file_data(file);
412
413         if (!sfd->file->f_op->fallocate)
414                 return -EOPNOTSUPP;
415         return sfd->file->f_op->fallocate(file, mode, offset, len);
416 }
417
418 static unsigned long shm_get_unmapped_area(struct file *file,
419         unsigned long addr, unsigned long len, unsigned long pgoff,
420         unsigned long flags)
421 {
422         struct shm_file_data *sfd = shm_file_data(file);
423         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
424                                                 pgoff, flags);
425 }
426
427 static const struct file_operations shm_file_operations = {
428         .mmap           = shm_mmap,
429         .fsync          = shm_fsync,
430         .release        = shm_release,
431 #ifndef CONFIG_MMU
432         .get_unmapped_area      = shm_get_unmapped_area,
433 #endif
434         .llseek         = noop_llseek,
435         .fallocate      = shm_fallocate,
436 };
437
438 static const struct file_operations shm_file_operations_huge = {
439         .mmap           = shm_mmap,
440         .fsync          = shm_fsync,
441         .release        = shm_release,
442         .get_unmapped_area      = shm_get_unmapped_area,
443         .llseek         = noop_llseek,
444         .fallocate      = shm_fallocate,
445 };
446
447 int is_file_shm_hugepages(struct file *file)
448 {
449         return file->f_op == &shm_file_operations_huge;
450 }
451
452 static const struct vm_operations_struct shm_vm_ops = {
453         .open   = shm_open,     /* callback for a new vm-area open */
454         .close  = shm_close,    /* callback for when the vm-area is released */
455         .fault  = shm_fault,
456 #if defined(CONFIG_NUMA)
457         .set_policy = shm_set_policy,
458         .get_policy = shm_get_policy,
459 #endif
460 };
461
462 /**
463  * newseg - Create a new shared memory segment
464  * @ns: namespace
465  * @params: ptr to the structure that contains key, size and shmflg
466  *
467  * Called with shm_ids.rwsem held as a writer.
468  */
469
470 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
471 {
472         key_t key = params->key;
473         int shmflg = params->flg;
474         size_t size = params->u.size;
475         int error;
476         struct shmid_kernel *shp;
477         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
478         struct file * file;
479         char name[13];
480         int id;
481         vm_flags_t acctflag = 0;
482
483         if (size < SHMMIN || size > ns->shm_ctlmax)
484                 return -EINVAL;
485
486         if (ns->shm_tot + numpages > ns->shm_ctlall)
487                 return -ENOSPC;
488
489         shp = ipc_rcu_alloc(sizeof(*shp));
490         if (!shp)
491                 return -ENOMEM;
492
493         shp->shm_perm.key = key;
494         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
495         shp->mlock_user = NULL;
496
497         shp->shm_perm.security = NULL;
498         error = security_shm_alloc(shp);
499         if (error) {
500                 ipc_rcu_putref(shp);
501                 return error;
502         }
503
504         sprintf (name, "SYSV%08x", key);
505         if (shmflg & SHM_HUGETLB) {
506                 struct hstate *hs;
507                 size_t hugesize;
508
509                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
510                 if (!hs) {
511                         error = -EINVAL;
512                         goto no_file;
513                 }
514                 hugesize = ALIGN(size, huge_page_size(hs));
515
516                 /* hugetlb_file_setup applies strict accounting */
517                 if (shmflg & SHM_NORESERVE)
518                         acctflag = VM_NORESERVE;
519                 file = hugetlb_file_setup(name, hugesize, acctflag,
520                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
521                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
522         } else {
523                 /*
524                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
525                  * if it's asked for.
526                  */
527                 if  ((shmflg & SHM_NORESERVE) &&
528                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
529                         acctflag = VM_NORESERVE;
530                 file = shmem_file_setup(name, size, acctflag);
531         }
532         error = PTR_ERR(file);
533         if (IS_ERR(file))
534                 goto no_file;
535
536         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
537         if (id < 0) {
538                 error = id;
539                 goto no_id;
540         }
541
542         shp->shm_cprid = task_tgid_vnr(current);
543         shp->shm_lprid = 0;
544         shp->shm_atim = shp->shm_dtim = 0;
545         shp->shm_ctim = get_seconds();
546         shp->shm_segsz = size;
547         shp->shm_nattch = 0;
548         shp->shm_file = file;
549         shp->shm_creator = current;
550
551         /*
552          * shmid gets reported as "inode#" in /proc/pid/maps.
553          * proc-ps tools use this. Changing this will break them.
554          */
555         file_inode(file)->i_ino = shp->shm_perm.id;
556
557         ns->shm_tot += numpages;
558         error = shp->shm_perm.id;
559
560         ipc_unlock_object(&shp->shm_perm);
561         rcu_read_unlock();
562         return error;
563
564 no_id:
565         if (is_file_hugepages(file) && shp->mlock_user)
566                 user_shm_unlock(size, shp->mlock_user);
567         fput(file);
568 no_file:
569         security_shm_free(shp);
570         ipc_rcu_putref(shp);
571         return error;
572 }
573
574 /*
575  * Called with shm_ids.rwsem and ipcp locked.
576  */
577 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
578 {
579         struct shmid_kernel *shp;
580
581         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
582         return security_shm_associate(shp, shmflg);
583 }
584
585 /*
586  * Called with shm_ids.rwsem and ipcp locked.
587  */
588 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
589                                 struct ipc_params *params)
590 {
591         struct shmid_kernel *shp;
592
593         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
594         if (shp->shm_segsz < params->u.size)
595                 return -EINVAL;
596
597         return 0;
598 }
599
600 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
601 {
602         struct ipc_namespace *ns;
603         struct ipc_ops shm_ops;
604         struct ipc_params shm_params;
605
606         ns = current->nsproxy->ipc_ns;
607
608         shm_ops.getnew = newseg;
609         shm_ops.associate = shm_security;
610         shm_ops.more_checks = shm_more_checks;
611
612         shm_params.key = key;
613         shm_params.flg = shmflg;
614         shm_params.u.size = size;
615
616         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
617 }
618
619 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
620 {
621         switch(version) {
622         case IPC_64:
623                 return copy_to_user(buf, in, sizeof(*in));
624         case IPC_OLD:
625             {
626                 struct shmid_ds out;
627
628                 memset(&out, 0, sizeof(out));
629                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
630                 out.shm_segsz   = in->shm_segsz;
631                 out.shm_atime   = in->shm_atime;
632                 out.shm_dtime   = in->shm_dtime;
633                 out.shm_ctime   = in->shm_ctime;
634                 out.shm_cpid    = in->shm_cpid;
635                 out.shm_lpid    = in->shm_lpid;
636                 out.shm_nattch  = in->shm_nattch;
637
638                 return copy_to_user(buf, &out, sizeof(out));
639             }
640         default:
641                 return -EINVAL;
642         }
643 }
644
645 static inline unsigned long
646 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
647 {
648         switch(version) {
649         case IPC_64:
650                 if (copy_from_user(out, buf, sizeof(*out)))
651                         return -EFAULT;
652                 return 0;
653         case IPC_OLD:
654             {
655                 struct shmid_ds tbuf_old;
656
657                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
658                         return -EFAULT;
659
660                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
661                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
662                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
663
664                 return 0;
665             }
666         default:
667                 return -EINVAL;
668         }
669 }
670
671 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
672 {
673         switch(version) {
674         case IPC_64:
675                 return copy_to_user(buf, in, sizeof(*in));
676         case IPC_OLD:
677             {
678                 struct shminfo out;
679
680                 if(in->shmmax > INT_MAX)
681                         out.shmmax = INT_MAX;
682                 else
683                         out.shmmax = (int)in->shmmax;
684
685                 out.shmmin      = in->shmmin;
686                 out.shmmni      = in->shmmni;
687                 out.shmseg      = in->shmseg;
688                 out.shmall      = in->shmall; 
689
690                 return copy_to_user(buf, &out, sizeof(out));
691             }
692         default:
693                 return -EINVAL;
694         }
695 }
696
697 /*
698  * Calculate and add used RSS and swap pages of a shm.
699  * Called with shm_ids.rwsem held as a reader
700  */
701 static void shm_add_rss_swap(struct shmid_kernel *shp,
702         unsigned long *rss_add, unsigned long *swp_add)
703 {
704         struct inode *inode;
705
706         inode = file_inode(shp->shm_file);
707
708         if (is_file_hugepages(shp->shm_file)) {
709                 struct address_space *mapping = inode->i_mapping;
710                 struct hstate *h = hstate_file(shp->shm_file);
711                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
712         } else {
713 #ifdef CONFIG_SHMEM
714                 struct shmem_inode_info *info = SHMEM_I(inode);
715                 spin_lock(&info->lock);
716                 *rss_add += inode->i_mapping->nrpages;
717                 *swp_add += info->swapped;
718                 spin_unlock(&info->lock);
719 #else
720                 *rss_add += inode->i_mapping->nrpages;
721 #endif
722         }
723 }
724
725 /*
726  * Called with shm_ids.rwsem held as a reader
727  */
728 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
729                 unsigned long *swp)
730 {
731         int next_id;
732         int total, in_use;
733
734         *rss = 0;
735         *swp = 0;
736
737         in_use = shm_ids(ns).in_use;
738
739         for (total = 0, next_id = 0; total < in_use; next_id++) {
740                 struct kern_ipc_perm *ipc;
741                 struct shmid_kernel *shp;
742
743                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
744                 if (ipc == NULL)
745                         continue;
746                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
747
748                 shm_add_rss_swap(shp, rss, swp);
749
750                 total++;
751         }
752 }
753
754 /*
755  * This function handles some shmctl commands which require the rwsem
756  * to be held in write mode.
757  * NOTE: no locks must be held, the rwsem is taken inside this function.
758  */
759 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
760                        struct shmid_ds __user *buf, int version)
761 {
762         struct kern_ipc_perm *ipcp;
763         struct shmid64_ds shmid64;
764         struct shmid_kernel *shp;
765         int err;
766
767         if (cmd == IPC_SET) {
768                 if (copy_shmid_from_user(&shmid64, buf, version))
769                         return -EFAULT;
770         }
771
772         down_write(&shm_ids(ns).rwsem);
773         rcu_read_lock();
774
775         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
776                                       &shmid64.shm_perm, 0);
777         if (IS_ERR(ipcp)) {
778                 err = PTR_ERR(ipcp);
779                 goto out_unlock1;
780         }
781
782         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
783
784         err = security_shm_shmctl(shp, cmd);
785         if (err)
786                 goto out_unlock1;
787
788         switch (cmd) {
789         case IPC_RMID:
790                 ipc_lock_object(&shp->shm_perm);
791                 /* do_shm_rmid unlocks the ipc object and rcu */
792                 do_shm_rmid(ns, ipcp);
793                 goto out_up;
794         case IPC_SET:
795                 ipc_lock_object(&shp->shm_perm);
796                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
797                 if (err)
798                         goto out_unlock0;
799                 shp->shm_ctim = get_seconds();
800                 break;
801         default:
802                 err = -EINVAL;
803                 goto out_unlock1;
804         }
805
806 out_unlock0:
807         ipc_unlock_object(&shp->shm_perm);
808 out_unlock1:
809         rcu_read_unlock();
810 out_up:
811         up_write(&shm_ids(ns).rwsem);
812         return err;
813 }
814
815 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
816                          int cmd, int version, void __user *buf)
817 {
818         int err;
819         struct shmid_kernel *shp;
820
821         /* preliminary security checks for *_INFO */
822         if (cmd == IPC_INFO || cmd == SHM_INFO) {
823                 err = security_shm_shmctl(NULL, cmd);
824                 if (err)
825                         return err;
826         }
827
828         switch (cmd) {
829         case IPC_INFO:
830         {
831                 struct shminfo64 shminfo;
832
833                 memset(&shminfo, 0, sizeof(shminfo));
834                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
835                 shminfo.shmmax = ns->shm_ctlmax;
836                 shminfo.shmall = ns->shm_ctlall;
837
838                 shminfo.shmmin = SHMMIN;
839                 if(copy_shminfo_to_user (buf, &shminfo, version))
840                         return -EFAULT;
841
842                 down_read(&shm_ids(ns).rwsem);
843                 err = ipc_get_maxid(&shm_ids(ns));
844                 up_read(&shm_ids(ns).rwsem);
845
846                 if(err<0)
847                         err = 0;
848                 goto out;
849         }
850         case SHM_INFO:
851         {
852                 struct shm_info shm_info;
853
854                 memset(&shm_info, 0, sizeof(shm_info));
855                 down_read(&shm_ids(ns).rwsem);
856                 shm_info.used_ids = shm_ids(ns).in_use;
857                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
858                 shm_info.shm_tot = ns->shm_tot;
859                 shm_info.swap_attempts = 0;
860                 shm_info.swap_successes = 0;
861                 err = ipc_get_maxid(&shm_ids(ns));
862                 up_read(&shm_ids(ns).rwsem);
863                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
864                         err = -EFAULT;
865                         goto out;
866                 }
867
868                 err = err < 0 ? 0 : err;
869                 goto out;
870         }
871         case SHM_STAT:
872         case IPC_STAT:
873         {
874                 struct shmid64_ds tbuf;
875                 int result;
876
877                 rcu_read_lock();
878                 if (cmd == SHM_STAT) {
879                         shp = shm_obtain_object(ns, shmid);
880                         if (IS_ERR(shp)) {
881                                 err = PTR_ERR(shp);
882                                 goto out_unlock;
883                         }
884                         result = shp->shm_perm.id;
885                 } else {
886                         shp = shm_obtain_object_check(ns, shmid);
887                         if (IS_ERR(shp)) {
888                                 err = PTR_ERR(shp);
889                                 goto out_unlock;
890                         }
891                         result = 0;
892                 }
893
894                 err = -EACCES;
895                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
896                         goto out_unlock;
897
898                 err = security_shm_shmctl(shp, cmd);
899                 if (err)
900                         goto out_unlock;
901
902                 memset(&tbuf, 0, sizeof(tbuf));
903                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
904                 tbuf.shm_segsz  = shp->shm_segsz;
905                 tbuf.shm_atime  = shp->shm_atim;
906                 tbuf.shm_dtime  = shp->shm_dtim;
907                 tbuf.shm_ctime  = shp->shm_ctim;
908                 tbuf.shm_cpid   = shp->shm_cprid;
909                 tbuf.shm_lpid   = shp->shm_lprid;
910                 tbuf.shm_nattch = shp->shm_nattch;
911                 rcu_read_unlock();
912
913                 if (copy_shmid_to_user(buf, &tbuf, version))
914                         err = -EFAULT;
915                 else
916                         err = result;
917                 goto out;
918         }
919         default:
920                 return -EINVAL;
921         }
922
923 out_unlock:
924         rcu_read_unlock();
925 out:
926         return err;
927 }
928
929 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
930 {
931         struct shmid_kernel *shp;
932         int err, version;
933         struct ipc_namespace *ns;
934
935         if (cmd < 0 || shmid < 0)
936                 return -EINVAL;
937
938         version = ipc_parse_version(&cmd);
939         ns = current->nsproxy->ipc_ns;
940
941         switch (cmd) {
942         case IPC_INFO:
943         case SHM_INFO:
944         case SHM_STAT:
945         case IPC_STAT:
946                 return shmctl_nolock(ns, shmid, cmd, version, buf);
947         case IPC_RMID:
948         case IPC_SET:
949                 return shmctl_down(ns, shmid, cmd, buf, version);
950         case SHM_LOCK:
951         case SHM_UNLOCK:
952         {
953                 struct file *shm_file;
954
955                 rcu_read_lock();
956                 shp = shm_obtain_object_check(ns, shmid);
957                 if (IS_ERR(shp)) {
958                         err = PTR_ERR(shp);
959                         goto out_unlock1;
960                 }
961
962                 audit_ipc_obj(&(shp->shm_perm));
963                 err = security_shm_shmctl(shp, cmd);
964                 if (err)
965                         goto out_unlock1;
966
967                 ipc_lock_object(&shp->shm_perm);
968                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
969                         kuid_t euid = current_euid();
970                         err = -EPERM;
971                         if (!uid_eq(euid, shp->shm_perm.uid) &&
972                             !uid_eq(euid, shp->shm_perm.cuid))
973                                 goto out_unlock0;
974                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
975                                 goto out_unlock0;
976                 }
977
978                 shm_file = shp->shm_file;
979                 if (is_file_hugepages(shm_file))
980                         goto out_unlock0;
981
982                 if (cmd == SHM_LOCK) {
983                         struct user_struct *user = current_user();
984                         err = shmem_lock(shm_file, 1, user);
985                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
986                                 shp->shm_perm.mode |= SHM_LOCKED;
987                                 shp->mlock_user = user;
988                         }
989                         goto out_unlock0;
990                 }
991
992                 /* SHM_UNLOCK */
993                 if (!(shp->shm_perm.mode & SHM_LOCKED))
994                         goto out_unlock0;
995                 shmem_lock(shm_file, 0, shp->mlock_user);
996                 shp->shm_perm.mode &= ~SHM_LOCKED;
997                 shp->mlock_user = NULL;
998                 get_file(shm_file);
999                 ipc_unlock_object(&shp->shm_perm);
1000                 rcu_read_unlock();
1001                 shmem_unlock_mapping(shm_file->f_mapping);
1002
1003                 fput(shm_file);
1004                 return err;
1005         }
1006         default:
1007                 return -EINVAL;
1008         }
1009
1010 out_unlock0:
1011         ipc_unlock_object(&shp->shm_perm);
1012 out_unlock1:
1013         rcu_read_unlock();
1014         return err;
1015 }
1016
1017 /*
1018  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1019  *
1020  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1021  * "raddr" thing points to kernel space, and there has to be a wrapper around
1022  * this.
1023  */
1024 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1025               unsigned long shmlba)
1026 {
1027         struct shmid_kernel *shp;
1028         unsigned long addr;
1029         unsigned long size;
1030         struct file * file;
1031         int    err;
1032         unsigned long flags;
1033         unsigned long prot;
1034         int acc_mode;
1035         struct ipc_namespace *ns;
1036         struct shm_file_data *sfd;
1037         struct path path;
1038         fmode_t f_mode;
1039         unsigned long populate = 0;
1040
1041         err = -EINVAL;
1042         if (shmid < 0)
1043                 goto out;
1044         else if ((addr = (ulong)shmaddr)) {
1045                 if (addr & (shmlba - 1)) {
1046                         if (shmflg & SHM_RND)
1047                                 addr &= ~(shmlba - 1);     /* round down */
1048                         else
1049 #ifndef __ARCH_FORCE_SHMLBA
1050                                 if (addr & ~PAGE_MASK)
1051 #endif
1052                                         goto out;
1053                 }
1054                 flags = MAP_SHARED | MAP_FIXED;
1055         } else {
1056                 if ((shmflg & SHM_REMAP))
1057                         goto out;
1058
1059                 flags = MAP_SHARED;
1060         }
1061
1062         if (shmflg & SHM_RDONLY) {
1063                 prot = PROT_READ;
1064                 acc_mode = S_IRUGO;
1065                 f_mode = FMODE_READ;
1066         } else {
1067                 prot = PROT_READ | PROT_WRITE;
1068                 acc_mode = S_IRUGO | S_IWUGO;
1069                 f_mode = FMODE_READ | FMODE_WRITE;
1070         }
1071         if (shmflg & SHM_EXEC) {
1072                 prot |= PROT_EXEC;
1073                 acc_mode |= S_IXUGO;
1074         }
1075
1076         /*
1077          * We cannot rely on the fs check since SYSV IPC does have an
1078          * additional creator id...
1079          */
1080         ns = current->nsproxy->ipc_ns;
1081         rcu_read_lock();
1082         shp = shm_obtain_object_check(ns, shmid);
1083         if (IS_ERR(shp)) {
1084                 err = PTR_ERR(shp);
1085                 goto out_unlock;
1086         }
1087
1088         err = -EACCES;
1089         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1090                 goto out_unlock;
1091
1092         err = security_shm_shmat(shp, shmaddr, shmflg);
1093         if (err)
1094                 goto out_unlock;
1095
1096         ipc_lock_object(&shp->shm_perm);
1097         path = shp->shm_file->f_path;
1098         path_get(&path);
1099         shp->shm_nattch++;
1100         size = i_size_read(path.dentry->d_inode);
1101         ipc_unlock_object(&shp->shm_perm);
1102         rcu_read_unlock();
1103
1104         err = -ENOMEM;
1105         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1106         if (!sfd) {
1107                 path_put(&path);
1108                 goto out_nattch;
1109         }
1110
1111         file = alloc_file(&path, f_mode,
1112                           is_file_hugepages(shp->shm_file) ?
1113                                 &shm_file_operations_huge :
1114                                 &shm_file_operations);
1115         err = PTR_ERR(file);
1116         if (IS_ERR(file)) {
1117                 kfree(sfd);
1118                 path_put(&path);
1119                 goto out_nattch;
1120         }
1121
1122         file->private_data = sfd;
1123         file->f_mapping = shp->shm_file->f_mapping;
1124         sfd->id = shp->shm_perm.id;
1125         sfd->ns = get_ipc_ns(ns);
1126         sfd->file = shp->shm_file;
1127         sfd->vm_ops = NULL;
1128
1129         err = security_mmap_file(file, prot, flags);
1130         if (err)
1131                 goto out_fput;
1132
1133         down_write(&current->mm->mmap_sem);
1134         if (addr && !(shmflg & SHM_REMAP)) {
1135                 err = -EINVAL;
1136                 if (find_vma_intersection(current->mm, addr, addr + size))
1137                         goto invalid;
1138                 /*
1139                  * If shm segment goes below stack, make sure there is some
1140                  * space left for the stack to grow (at least 4 pages).
1141                  */
1142                 if (addr < current->mm->start_stack &&
1143                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1144                         goto invalid;
1145         }
1146
1147         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1148         *raddr = addr;
1149         err = 0;
1150         if (IS_ERR_VALUE(addr))
1151                 err = (long)addr;
1152 invalid:
1153         up_write(&current->mm->mmap_sem);
1154         if (populate)
1155                 mm_populate(addr, populate);
1156
1157 out_fput:
1158         fput(file);
1159
1160 out_nattch:
1161         down_write(&shm_ids(ns).rwsem);
1162         shp = shm_lock(ns, shmid);
1163         BUG_ON(IS_ERR(shp));
1164         shp->shm_nattch--;
1165         if (shm_may_destroy(ns, shp))
1166                 shm_destroy(ns, shp);
1167         else
1168                 shm_unlock(shp);
1169         up_write(&shm_ids(ns).rwsem);
1170         return err;
1171
1172 out_unlock:
1173         rcu_read_unlock();
1174 out:
1175         return err;
1176 }
1177
1178 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1179 {
1180         unsigned long ret;
1181         long err;
1182
1183         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1184         if (err)
1185                 return err;
1186         force_successful_syscall_return();
1187         return (long)ret;
1188 }
1189
1190 /*
1191  * detach and kill segment if marked destroyed.
1192  * The work is done in shm_close.
1193  */
1194 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1195 {
1196         struct mm_struct *mm = current->mm;
1197         struct vm_area_struct *vma;
1198         unsigned long addr = (unsigned long)shmaddr;
1199         int retval = -EINVAL;
1200 #ifdef CONFIG_MMU
1201         loff_t size = 0;
1202         struct vm_area_struct *next;
1203 #endif
1204
1205         if (addr & ~PAGE_MASK)
1206                 return retval;
1207
1208         down_write(&mm->mmap_sem);
1209
1210         /*
1211          * This function tries to be smart and unmap shm segments that
1212          * were modified by partial mlock or munmap calls:
1213          * - It first determines the size of the shm segment that should be
1214          *   unmapped: It searches for a vma that is backed by shm and that
1215          *   started at address shmaddr. It records it's size and then unmaps
1216          *   it.
1217          * - Then it unmaps all shm vmas that started at shmaddr and that
1218          *   are within the initially determined size.
1219          * Errors from do_munmap are ignored: the function only fails if
1220          * it's called with invalid parameters or if it's called to unmap
1221          * a part of a vma. Both calls in this function are for full vmas,
1222          * the parameters are directly copied from the vma itself and always
1223          * valid - therefore do_munmap cannot fail. (famous last words?)
1224          */
1225         /*
1226          * If it had been mremap()'d, the starting address would not
1227          * match the usual checks anyway. So assume all vma's are
1228          * above the starting address given.
1229          */
1230         vma = find_vma(mm, addr);
1231
1232 #ifdef CONFIG_MMU
1233         while (vma) {
1234                 next = vma->vm_next;
1235
1236                 /*
1237                  * Check if the starting address would match, i.e. it's
1238                  * a fragment created by mprotect() and/or munmap(), or it
1239                  * otherwise it starts at this address with no hassles.
1240                  */
1241                 if ((vma->vm_ops == &shm_vm_ops) &&
1242                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1243
1244
1245                         size = file_inode(vma->vm_file)->i_size;
1246                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1247                         /*
1248                          * We discovered the size of the shm segment, so
1249                          * break out of here and fall through to the next
1250                          * loop that uses the size information to stop
1251                          * searching for matching vma's.
1252                          */
1253                         retval = 0;
1254                         vma = next;
1255                         break;
1256                 }
1257                 vma = next;
1258         }
1259
1260         /*
1261          * We need look no further than the maximum address a fragment
1262          * could possibly have landed at. Also cast things to loff_t to
1263          * prevent overflows and make comparisons vs. equal-width types.
1264          */
1265         size = PAGE_ALIGN(size);
1266         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1267                 next = vma->vm_next;
1268
1269                 /* finding a matching vma now does not alter retval */
1270                 if ((vma->vm_ops == &shm_vm_ops) &&
1271                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1272
1273                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1274                 vma = next;
1275         }
1276
1277 #else /* CONFIG_MMU */
1278         /* under NOMMU conditions, the exact address to be destroyed must be
1279          * given */
1280         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1281                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1282                 retval = 0;
1283         }
1284
1285 #endif
1286
1287         up_write(&mm->mmap_sem);
1288         return retval;
1289 }
1290
1291 #ifdef CONFIG_PROC_FS
1292 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1293 {
1294         struct user_namespace *user_ns = seq_user_ns(s);
1295         struct shmid_kernel *shp = it;
1296         unsigned long rss = 0, swp = 0;
1297
1298         shm_add_rss_swap(shp, &rss, &swp);
1299
1300 #if BITS_PER_LONG <= 32
1301 #define SIZE_SPEC "%10lu"
1302 #else
1303 #define SIZE_SPEC "%21lu"
1304 #endif
1305
1306         return seq_printf(s,
1307                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1308                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1309                           SIZE_SPEC " " SIZE_SPEC "\n",
1310                           shp->shm_perm.key,
1311                           shp->shm_perm.id,
1312                           shp->shm_perm.mode,
1313                           shp->shm_segsz,
1314                           shp->shm_cprid,
1315                           shp->shm_lprid,
1316                           shp->shm_nattch,
1317                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1318                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1319                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1320                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1321                           shp->shm_atim,
1322                           shp->shm_dtim,
1323                           shp->shm_ctim,
1324                           rss * PAGE_SIZE,
1325                           swp * PAGE_SIZE);
1326 }
1327 #endif