115dccebc63e77956b08d3cdc001df4c1170bd48
[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
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
43 #include <asm/uaccess.h>
44
45 #include "util.h"
46
47 struct shm_file_data {
48         int id;
49         struct ipc_namespace *ns;
50         struct file *file;
51         const struct vm_operations_struct *vm_ops;
52 };
53
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
56 static const struct file_operations shm_file_operations;
57 static const struct vm_operations_struct shm_vm_ops;
58
59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
60
61 #define shm_unlock(shp)                 \
62         ipc_unlock(&(shp)->shm_perm)
63
64 static int newseg(struct ipc_namespace *, struct ipc_params *);
65 static void shm_open(struct vm_area_struct *vma);
66 static void shm_close(struct vm_area_struct *vma);
67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
68 #ifdef CONFIG_PROC_FS
69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
70 #endif
71
72 void shm_init_ns(struct ipc_namespace *ns)
73 {
74         ns->shm_ctlmax = SHMMAX;
75         ns->shm_ctlall = SHMALL;
76         ns->shm_ctlmni = SHMMNI;
77         ns->shm_rmid_forced = 0;
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 static int __init ipc_ns_init(void)
109 {
110         shm_init_ns(&init_ipc_ns);
111         return 0;
112 }
113
114 pure_initcall(ipc_ns_init);
115
116 void __init shm_init (void)
117 {
118         ipc_init_proc_interface("sysvipc/shm",
119 #if BITS_PER_LONG <= 32
120                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
121 #else
122                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
123 #endif
124                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
125 }
126
127 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
128 {
129         struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
130
131         if (IS_ERR(ipcp))
132                 return ERR_CAST(ipcp);
133
134         return container_of(ipcp, struct shmid_kernel, shm_perm);
135 }
136
137 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
138 {
139         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
140
141         if (IS_ERR(ipcp))
142                 return ERR_CAST(ipcp);
143
144         return container_of(ipcp, struct shmid_kernel, shm_perm);
145 }
146
147 /*
148  * shm_lock_(check_) routines are called in the paths where the rw_mutex
149  * is not necessarily held.
150  */
151 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
152 {
153         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
154
155         if (IS_ERR(ipcp))
156                 return (struct shmid_kernel *)ipcp;
157
158         return container_of(ipcp, struct shmid_kernel, shm_perm);
159 }
160
161 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
162 {
163         rcu_read_lock();
164         ipc_lock_object(&ipcp->shm_perm);
165 }
166
167 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
168                                                 int id)
169 {
170         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
171
172         if (IS_ERR(ipcp))
173                 return (struct shmid_kernel *)ipcp;
174
175         return container_of(ipcp, struct shmid_kernel, shm_perm);
176 }
177
178 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
179 {
180         ipc_rmid(&shm_ids(ns), &s->shm_perm);
181 }
182
183
184 /* This is called by fork, once for every shm attach. */
185 static void shm_open(struct vm_area_struct *vma)
186 {
187         struct file *file = vma->vm_file;
188         struct shm_file_data *sfd = shm_file_data(file);
189         struct shmid_kernel *shp;
190
191         shp = shm_lock(sfd->ns, sfd->id);
192         BUG_ON(IS_ERR(shp));
193         shp->shm_atim = get_seconds();
194         shp->shm_lprid = task_tgid_vnr(current);
195         shp->shm_nattch++;
196         shm_unlock(shp);
197 }
198
199 /*
200  * shm_destroy - free the struct shmid_kernel
201  *
202  * @ns: namespace
203  * @shp: struct to free
204  *
205  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
206  * but returns with shp unlocked and freed.
207  */
208 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
209 {
210         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
211         shm_rmid(ns, shp);
212         shm_unlock(shp);
213         if (!is_file_hugepages(shp->shm_file))
214                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
215         else if (shp->mlock_user)
216                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
217                                                 shp->mlock_user);
218         fput (shp->shm_file);
219         security_shm_free(shp);
220         ipc_rcu_putref(shp);
221 }
222
223 /*
224  * shm_may_destroy - identifies whether shm segment should be destroyed now
225  *
226  * Returns true if and only if there are no active users of the segment and
227  * one of the following is true:
228  *
229  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
230  *
231  * 2) sysctl kernel.shm_rmid_forced is set to 1.
232  */
233 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
234 {
235         return (shp->shm_nattch == 0) &&
236                (ns->shm_rmid_forced ||
237                 (shp->shm_perm.mode & SHM_DEST));
238 }
239
240 /*
241  * remove the attach descriptor vma.
242  * free memory for segment if it is marked destroyed.
243  * The descriptor has already been removed from the current->mm->mmap list
244  * and will later be kfree()d.
245  */
246 static void shm_close(struct vm_area_struct *vma)
247 {
248         struct file * file = vma->vm_file;
249         struct shm_file_data *sfd = shm_file_data(file);
250         struct shmid_kernel *shp;
251         struct ipc_namespace *ns = sfd->ns;
252
253         down_write(&shm_ids(ns).rw_mutex);
254         /* remove from the list of attaches of the shm segment */
255         shp = shm_lock(ns, sfd->id);
256         BUG_ON(IS_ERR(shp));
257         shp->shm_lprid = task_tgid_vnr(current);
258         shp->shm_dtim = get_seconds();
259         shp->shm_nattch--;
260         if (shm_may_destroy(ns, shp))
261                 shm_destroy(ns, shp);
262         else
263                 shm_unlock(shp);
264         up_write(&shm_ids(ns).rw_mutex);
265 }
266
267 /* Called with ns->shm_ids(ns).rw_mutex locked */
268 static int shm_try_destroy_current(int id, void *p, void *data)
269 {
270         struct ipc_namespace *ns = data;
271         struct kern_ipc_perm *ipcp = p;
272         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
273
274         if (shp->shm_creator != current)
275                 return 0;
276
277         /*
278          * Mark it as orphaned to destroy the segment when
279          * kernel.shm_rmid_forced is changed.
280          * It is noop if the following shm_may_destroy() returns true.
281          */
282         shp->shm_creator = NULL;
283
284         /*
285          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
286          * is not set, it shouldn't be deleted here.
287          */
288         if (!ns->shm_rmid_forced)
289                 return 0;
290
291         if (shm_may_destroy(ns, shp)) {
292                 shm_lock_by_ptr(shp);
293                 shm_destroy(ns, shp);
294         }
295         return 0;
296 }
297
298 /* Called with ns->shm_ids(ns).rw_mutex locked */
299 static int shm_try_destroy_orphaned(int id, void *p, void *data)
300 {
301         struct ipc_namespace *ns = data;
302         struct kern_ipc_perm *ipcp = p;
303         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
304
305         /*
306          * We want to destroy segments without users and with already
307          * exit'ed originating process.
308          *
309          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
310          */
311         if (shp->shm_creator != NULL)
312                 return 0;
313
314         if (shm_may_destroy(ns, shp)) {
315                 shm_lock_by_ptr(shp);
316                 shm_destroy(ns, shp);
317         }
318         return 0;
319 }
320
321 void shm_destroy_orphaned(struct ipc_namespace *ns)
322 {
323         down_write(&shm_ids(ns).rw_mutex);
324         if (shm_ids(ns).in_use)
325                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
326         up_write(&shm_ids(ns).rw_mutex);
327 }
328
329
330 void exit_shm(struct task_struct *task)
331 {
332         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
333
334         if (shm_ids(ns).in_use == 0)
335                 return;
336
337         /* Destroy all already created segments, but not mapped yet */
338         down_write(&shm_ids(ns).rw_mutex);
339         if (shm_ids(ns).in_use)
340                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
341         up_write(&shm_ids(ns).rw_mutex);
342 }
343
344 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
345 {
346         struct file *file = vma->vm_file;
347         struct shm_file_data *sfd = shm_file_data(file);
348
349         return sfd->vm_ops->fault(vma, vmf);
350 }
351
352 #ifdef CONFIG_NUMA
353 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
354 {
355         struct file *file = vma->vm_file;
356         struct shm_file_data *sfd = shm_file_data(file);
357         int err = 0;
358         if (sfd->vm_ops->set_policy)
359                 err = sfd->vm_ops->set_policy(vma, new);
360         return err;
361 }
362
363 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
364                                         unsigned long addr)
365 {
366         struct file *file = vma->vm_file;
367         struct shm_file_data *sfd = shm_file_data(file);
368         struct mempolicy *pol = NULL;
369
370         if (sfd->vm_ops->get_policy)
371                 pol = sfd->vm_ops->get_policy(vma, addr);
372         else if (vma->vm_policy)
373                 pol = vma->vm_policy;
374
375         return pol;
376 }
377 #endif
378
379 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
380 {
381         struct shm_file_data *sfd = shm_file_data(file);
382         int ret;
383
384         ret = sfd->file->f_op->mmap(sfd->file, vma);
385         if (ret != 0)
386                 return ret;
387         sfd->vm_ops = vma->vm_ops;
388 #ifdef CONFIG_MMU
389         BUG_ON(!sfd->vm_ops->fault);
390 #endif
391         vma->vm_ops = &shm_vm_ops;
392         shm_open(vma);
393
394         return ret;
395 }
396
397 static int shm_release(struct inode *ino, struct file *file)
398 {
399         struct shm_file_data *sfd = shm_file_data(file);
400
401         put_ipc_ns(sfd->ns);
402         shm_file_data(file) = NULL;
403         kfree(sfd);
404         return 0;
405 }
406
407 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
408 {
409         struct shm_file_data *sfd = shm_file_data(file);
410
411         if (!sfd->file->f_op->fsync)
412                 return -EINVAL;
413         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
414 }
415
416 static long shm_fallocate(struct file *file, int mode, loff_t offset,
417                           loff_t len)
418 {
419         struct shm_file_data *sfd = shm_file_data(file);
420
421         if (!sfd->file->f_op->fallocate)
422                 return -EOPNOTSUPP;
423         return sfd->file->f_op->fallocate(file, mode, offset, len);
424 }
425
426 static unsigned long shm_get_unmapped_area(struct file *file,
427         unsigned long addr, unsigned long len, unsigned long pgoff,
428         unsigned long flags)
429 {
430         struct shm_file_data *sfd = shm_file_data(file);
431         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
432                                                 pgoff, flags);
433 }
434
435 static const struct file_operations shm_file_operations = {
436         .mmap           = shm_mmap,
437         .fsync          = shm_fsync,
438         .release        = shm_release,
439 #ifndef CONFIG_MMU
440         .get_unmapped_area      = shm_get_unmapped_area,
441 #endif
442         .llseek         = noop_llseek,
443         .fallocate      = shm_fallocate,
444 };
445
446 static const struct file_operations shm_file_operations_huge = {
447         .mmap           = shm_mmap,
448         .fsync          = shm_fsync,
449         .release        = shm_release,
450         .get_unmapped_area      = shm_get_unmapped_area,
451         .llseek         = noop_llseek,
452         .fallocate      = shm_fallocate,
453 };
454
455 int is_file_shm_hugepages(struct file *file)
456 {
457         return file->f_op == &shm_file_operations_huge;
458 }
459
460 static const struct vm_operations_struct shm_vm_ops = {
461         .open   = shm_open,     /* callback for a new vm-area open */
462         .close  = shm_close,    /* callback for when the vm-area is released */
463         .fault  = shm_fault,
464 #if defined(CONFIG_NUMA)
465         .set_policy = shm_set_policy,
466         .get_policy = shm_get_policy,
467 #endif
468 };
469
470 /**
471  * newseg - Create a new shared memory segment
472  * @ns: namespace
473  * @params: ptr to the structure that contains key, size and shmflg
474  *
475  * Called with shm_ids.rw_mutex held as a writer.
476  */
477
478 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
479 {
480         key_t key = params->key;
481         int shmflg = params->flg;
482         size_t size = params->u.size;
483         int error;
484         struct shmid_kernel *shp;
485         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
486         struct file * file;
487         char name[13];
488         int id;
489         vm_flags_t acctflag = 0;
490
491         if (size < SHMMIN || size > ns->shm_ctlmax)
492                 return -EINVAL;
493
494         if (ns->shm_tot + numpages > ns->shm_ctlall)
495                 return -ENOSPC;
496
497         shp = ipc_rcu_alloc(sizeof(*shp));
498         if (!shp)
499                 return -ENOMEM;
500
501         shp->shm_perm.key = key;
502         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
503         shp->mlock_user = NULL;
504
505         shp->shm_perm.security = NULL;
506         error = security_shm_alloc(shp);
507         if (error) {
508                 ipc_rcu_putref(shp);
509                 return error;
510         }
511
512         sprintf (name, "SYSV%08x", key);
513         if (shmflg & SHM_HUGETLB) {
514                 struct hstate *hs;
515                 size_t hugesize;
516
517                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
518                 if (!hs) {
519                         error = -EINVAL;
520                         goto no_file;
521                 }
522                 hugesize = ALIGN(size, huge_page_size(hs));
523
524                 /* hugetlb_file_setup applies strict accounting */
525                 if (shmflg & SHM_NORESERVE)
526                         acctflag = VM_NORESERVE;
527                 file = hugetlb_file_setup(name, hugesize, acctflag,
528                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
529                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
530         } else {
531                 /*
532                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
533                  * if it's asked for.
534                  */
535                 if  ((shmflg & SHM_NORESERVE) &&
536                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
537                         acctflag = VM_NORESERVE;
538                 file = shmem_file_setup(name, size, acctflag);
539         }
540         error = PTR_ERR(file);
541         if (IS_ERR(file))
542                 goto no_file;
543
544         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
545         if (id < 0) {
546                 error = id;
547                 goto no_id;
548         }
549
550         shp->shm_cprid = task_tgid_vnr(current);
551         shp->shm_lprid = 0;
552         shp->shm_atim = shp->shm_dtim = 0;
553         shp->shm_ctim = get_seconds();
554         shp->shm_segsz = size;
555         shp->shm_nattch = 0;
556         shp->shm_file = file;
557         shp->shm_creator = current;
558
559         /*
560          * shmid gets reported as "inode#" in /proc/pid/maps.
561          * proc-ps tools use this. Changing this will break them.
562          */
563         file_inode(file)->i_ino = shp->shm_perm.id;
564
565         ns->shm_tot += numpages;
566         error = shp->shm_perm.id;
567
568         ipc_unlock_object(&shp->shm_perm);
569         rcu_read_unlock();
570         return error;
571
572 no_id:
573         if (is_file_hugepages(file) && shp->mlock_user)
574                 user_shm_unlock(size, shp->mlock_user);
575         fput(file);
576 no_file:
577         security_shm_free(shp);
578         ipc_rcu_putref(shp);
579         return error;
580 }
581
582 /*
583  * Called with shm_ids.rw_mutex and ipcp locked.
584  */
585 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
586 {
587         struct shmid_kernel *shp;
588
589         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
590         return security_shm_associate(shp, shmflg);
591 }
592
593 /*
594  * Called with shm_ids.rw_mutex and ipcp locked.
595  */
596 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
597                                 struct ipc_params *params)
598 {
599         struct shmid_kernel *shp;
600
601         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
602         if (shp->shm_segsz < params->u.size)
603                 return -EINVAL;
604
605         return 0;
606 }
607
608 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
609 {
610         struct ipc_namespace *ns;
611         struct ipc_ops shm_ops;
612         struct ipc_params shm_params;
613
614         ns = current->nsproxy->ipc_ns;
615
616         shm_ops.getnew = newseg;
617         shm_ops.associate = shm_security;
618         shm_ops.more_checks = shm_more_checks;
619
620         shm_params.key = key;
621         shm_params.flg = shmflg;
622         shm_params.u.size = size;
623
624         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
625 }
626
627 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
628 {
629         switch(version) {
630         case IPC_64:
631                 return copy_to_user(buf, in, sizeof(*in));
632         case IPC_OLD:
633             {
634                 struct shmid_ds out;
635
636                 memset(&out, 0, sizeof(out));
637                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
638                 out.shm_segsz   = in->shm_segsz;
639                 out.shm_atime   = in->shm_atime;
640                 out.shm_dtime   = in->shm_dtime;
641                 out.shm_ctime   = in->shm_ctime;
642                 out.shm_cpid    = in->shm_cpid;
643                 out.shm_lpid    = in->shm_lpid;
644                 out.shm_nattch  = in->shm_nattch;
645
646                 return copy_to_user(buf, &out, sizeof(out));
647             }
648         default:
649                 return -EINVAL;
650         }
651 }
652
653 static inline unsigned long
654 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
655 {
656         switch(version) {
657         case IPC_64:
658                 if (copy_from_user(out, buf, sizeof(*out)))
659                         return -EFAULT;
660                 return 0;
661         case IPC_OLD:
662             {
663                 struct shmid_ds tbuf_old;
664
665                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
666                         return -EFAULT;
667
668                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
669                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
670                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
671
672                 return 0;
673             }
674         default:
675                 return -EINVAL;
676         }
677 }
678
679 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
680 {
681         switch(version) {
682         case IPC_64:
683                 return copy_to_user(buf, in, sizeof(*in));
684         case IPC_OLD:
685             {
686                 struct shminfo out;
687
688                 if(in->shmmax > INT_MAX)
689                         out.shmmax = INT_MAX;
690                 else
691                         out.shmmax = (int)in->shmmax;
692
693                 out.shmmin      = in->shmmin;
694                 out.shmmni      = in->shmmni;
695                 out.shmseg      = in->shmseg;
696                 out.shmall      = in->shmall; 
697
698                 return copy_to_user(buf, &out, sizeof(out));
699             }
700         default:
701                 return -EINVAL;
702         }
703 }
704
705 /*
706  * Calculate and add used RSS and swap pages of a shm.
707  * Called with shm_ids.rw_mutex held as a reader
708  */
709 static void shm_add_rss_swap(struct shmid_kernel *shp,
710         unsigned long *rss_add, unsigned long *swp_add)
711 {
712         struct inode *inode;
713
714         inode = file_inode(shp->shm_file);
715
716         if (is_file_hugepages(shp->shm_file)) {
717                 struct address_space *mapping = inode->i_mapping;
718                 struct hstate *h = hstate_file(shp->shm_file);
719                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
720         } else {
721 #ifdef CONFIG_SHMEM
722                 struct shmem_inode_info *info = SHMEM_I(inode);
723                 spin_lock(&info->lock);
724                 *rss_add += inode->i_mapping->nrpages;
725                 *swp_add += info->swapped;
726                 spin_unlock(&info->lock);
727 #else
728                 *rss_add += inode->i_mapping->nrpages;
729 #endif
730         }
731 }
732
733 /*
734  * Called with shm_ids.rw_mutex held as a reader
735  */
736 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
737                 unsigned long *swp)
738 {
739         int next_id;
740         int total, in_use;
741
742         *rss = 0;
743         *swp = 0;
744
745         in_use = shm_ids(ns).in_use;
746
747         for (total = 0, next_id = 0; total < in_use; next_id++) {
748                 struct kern_ipc_perm *ipc;
749                 struct shmid_kernel *shp;
750
751                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
752                 if (ipc == NULL)
753                         continue;
754                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
755
756                 shm_add_rss_swap(shp, rss, swp);
757
758                 total++;
759         }
760 }
761
762 /*
763  * This function handles some shmctl commands which require the rw_mutex
764  * to be held in write mode.
765  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
766  */
767 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
768                        struct shmid_ds __user *buf, int version)
769 {
770         struct kern_ipc_perm *ipcp;
771         struct shmid64_ds shmid64;
772         struct shmid_kernel *shp;
773         int err;
774
775         if (cmd == IPC_SET) {
776                 if (copy_shmid_from_user(&shmid64, buf, version))
777                         return -EFAULT;
778         }
779
780         down_write(&shm_ids(ns).rw_mutex);
781         rcu_read_lock();
782
783         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
784                                       &shmid64.shm_perm, 0);
785         if (IS_ERR(ipcp)) {
786                 err = PTR_ERR(ipcp);
787                 goto out_unlock1;
788         }
789
790         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
791
792         err = security_shm_shmctl(shp, cmd);
793         if (err)
794                 goto out_unlock1;
795
796         switch (cmd) {
797         case IPC_RMID:
798                 ipc_lock_object(&shp->shm_perm);
799                 /* do_shm_rmid unlocks the ipc object and rcu */
800                 do_shm_rmid(ns, ipcp);
801                 goto out_up;
802         case IPC_SET:
803                 ipc_lock_object(&shp->shm_perm);
804                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
805                 if (err)
806                         goto out_unlock0;
807                 shp->shm_ctim = get_seconds();
808                 break;
809         default:
810                 err = -EINVAL;
811                 goto out_unlock1;
812         }
813
814 out_unlock0:
815         ipc_unlock_object(&shp->shm_perm);
816 out_unlock1:
817         rcu_read_unlock();
818 out_up:
819         up_write(&shm_ids(ns).rw_mutex);
820         return err;
821 }
822
823 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
824                          int cmd, int version, void __user *buf)
825 {
826         int err;
827         struct shmid_kernel *shp;
828
829         /* preliminary security checks for *_INFO */
830         if (cmd == IPC_INFO || cmd == SHM_INFO) {
831                 err = security_shm_shmctl(NULL, cmd);
832                 if (err)
833                         return err;
834         }
835
836         switch (cmd) {
837         case IPC_INFO:
838         {
839                 struct shminfo64 shminfo;
840
841                 memset(&shminfo, 0, sizeof(shminfo));
842                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
843                 shminfo.shmmax = ns->shm_ctlmax;
844                 shminfo.shmall = ns->shm_ctlall;
845
846                 shminfo.shmmin = SHMMIN;
847                 if(copy_shminfo_to_user (buf, &shminfo, version))
848                         return -EFAULT;
849
850                 down_read(&shm_ids(ns).rw_mutex);
851                 err = ipc_get_maxid(&shm_ids(ns));
852                 up_read(&shm_ids(ns).rw_mutex);
853
854                 if(err<0)
855                         err = 0;
856                 goto out;
857         }
858         case SHM_INFO:
859         {
860                 struct shm_info shm_info;
861
862                 memset(&shm_info, 0, sizeof(shm_info));
863                 down_read(&shm_ids(ns).rw_mutex);
864                 shm_info.used_ids = shm_ids(ns).in_use;
865                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
866                 shm_info.shm_tot = ns->shm_tot;
867                 shm_info.swap_attempts = 0;
868                 shm_info.swap_successes = 0;
869                 err = ipc_get_maxid(&shm_ids(ns));
870                 up_read(&shm_ids(ns).rw_mutex);
871                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
872                         err = -EFAULT;
873                         goto out;
874                 }
875
876                 err = err < 0 ? 0 : err;
877                 goto out;
878         }
879         case SHM_STAT:
880         case IPC_STAT:
881         {
882                 struct shmid64_ds tbuf;
883                 int result;
884
885                 rcu_read_lock();
886                 if (cmd == SHM_STAT) {
887                         shp = shm_obtain_object(ns, shmid);
888                         if (IS_ERR(shp)) {
889                                 err = PTR_ERR(shp);
890                                 goto out_unlock;
891                         }
892                         result = shp->shm_perm.id;
893                 } else {
894                         shp = shm_obtain_object_check(ns, shmid);
895                         if (IS_ERR(shp)) {
896                                 err = PTR_ERR(shp);
897                                 goto out_unlock;
898                         }
899                         result = 0;
900                 }
901
902                 err = -EACCES;
903                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
904                         goto out_unlock;
905
906                 err = security_shm_shmctl(shp, cmd);
907                 if (err)
908                         goto out_unlock;
909
910                 memset(&tbuf, 0, sizeof(tbuf));
911                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
912                 tbuf.shm_segsz  = shp->shm_segsz;
913                 tbuf.shm_atime  = shp->shm_atim;
914                 tbuf.shm_dtime  = shp->shm_dtim;
915                 tbuf.shm_ctime  = shp->shm_ctim;
916                 tbuf.shm_cpid   = shp->shm_cprid;
917                 tbuf.shm_lpid   = shp->shm_lprid;
918                 tbuf.shm_nattch = shp->shm_nattch;
919                 rcu_read_unlock();
920
921                 if (copy_shmid_to_user(buf, &tbuf, version))
922                         err = -EFAULT;
923                 else
924                         err = result;
925                 goto out;
926         }
927         default:
928                 return -EINVAL;
929         }
930
931 out_unlock:
932         rcu_read_unlock();
933 out:
934         return err;
935 }
936
937 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
938 {
939         struct shmid_kernel *shp;
940         int err, version;
941         struct ipc_namespace *ns;
942
943         if (cmd < 0 || shmid < 0)
944                 return -EINVAL;
945
946         version = ipc_parse_version(&cmd);
947         ns = current->nsproxy->ipc_ns;
948
949         switch (cmd) {
950         case IPC_INFO:
951         case SHM_INFO:
952         case SHM_STAT:
953         case IPC_STAT:
954                 return shmctl_nolock(ns, shmid, cmd, version, buf);
955         case IPC_RMID:
956         case IPC_SET:
957                 return shmctl_down(ns, shmid, cmd, buf, version);
958         case SHM_LOCK:
959         case SHM_UNLOCK:
960         {
961                 struct file *shm_file;
962
963                 rcu_read_lock();
964                 shp = shm_obtain_object_check(ns, shmid);
965                 if (IS_ERR(shp)) {
966                         err = PTR_ERR(shp);
967                         goto out_unlock1;
968                 }
969
970                 audit_ipc_obj(&(shp->shm_perm));
971                 err = security_shm_shmctl(shp, cmd);
972                 if (err)
973                         goto out_unlock1;
974
975                 ipc_lock_object(&shp->shm_perm);
976                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
977                         kuid_t euid = current_euid();
978                         err = -EPERM;
979                         if (!uid_eq(euid, shp->shm_perm.uid) &&
980                             !uid_eq(euid, shp->shm_perm.cuid))
981                                 goto out_unlock0;
982                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
983                                 goto out_unlock0;
984                 }
985
986                 shm_file = shp->shm_file;
987                 if (is_file_hugepages(shm_file))
988                         goto out_unlock0;
989
990                 if (cmd == SHM_LOCK) {
991                         struct user_struct *user = current_user();
992                         err = shmem_lock(shm_file, 1, user);
993                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
994                                 shp->shm_perm.mode |= SHM_LOCKED;
995                                 shp->mlock_user = user;
996                         }
997                         goto out_unlock0;
998                 }
999
1000                 /* SHM_UNLOCK */
1001                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1002                         goto out_unlock0;
1003                 shmem_lock(shm_file, 0, shp->mlock_user);
1004                 shp->shm_perm.mode &= ~SHM_LOCKED;
1005                 shp->mlock_user = NULL;
1006                 get_file(shm_file);
1007                 ipc_unlock_object(&shp->shm_perm);
1008                 rcu_read_unlock();
1009                 shmem_unlock_mapping(shm_file->f_mapping);
1010
1011                 fput(shm_file);
1012                 return err;
1013         }
1014         default:
1015                 return -EINVAL;
1016         }
1017
1018 out_unlock0:
1019         ipc_unlock_object(&shp->shm_perm);
1020 out_unlock1:
1021         rcu_read_unlock();
1022         return err;
1023 }
1024
1025 /*
1026  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1027  *
1028  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1029  * "raddr" thing points to kernel space, and there has to be a wrapper around
1030  * this.
1031  */
1032 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1033               unsigned long shmlba)
1034 {
1035         struct shmid_kernel *shp;
1036         unsigned long addr;
1037         unsigned long size;
1038         struct file * file;
1039         int    err;
1040         unsigned long flags;
1041         unsigned long prot;
1042         int acc_mode;
1043         struct ipc_namespace *ns;
1044         struct shm_file_data *sfd;
1045         struct path path;
1046         fmode_t f_mode;
1047         unsigned long populate = 0;
1048
1049         err = -EINVAL;
1050         if (shmid < 0)
1051                 goto out;
1052         else if ((addr = (ulong)shmaddr)) {
1053                 if (addr & (shmlba - 1)) {
1054                         if (shmflg & SHM_RND)
1055                                 addr &= ~(shmlba - 1);     /* round down */
1056                         else
1057 #ifndef __ARCH_FORCE_SHMLBA
1058                                 if (addr & ~PAGE_MASK)
1059 #endif
1060                                         goto out;
1061                 }
1062                 flags = MAP_SHARED | MAP_FIXED;
1063         } else {
1064                 if ((shmflg & SHM_REMAP))
1065                         goto out;
1066
1067                 flags = MAP_SHARED;
1068         }
1069
1070         if (shmflg & SHM_RDONLY) {
1071                 prot = PROT_READ;
1072                 acc_mode = S_IRUGO;
1073                 f_mode = FMODE_READ;
1074         } else {
1075                 prot = PROT_READ | PROT_WRITE;
1076                 acc_mode = S_IRUGO | S_IWUGO;
1077                 f_mode = FMODE_READ | FMODE_WRITE;
1078         }
1079         if (shmflg & SHM_EXEC) {
1080                 prot |= PROT_EXEC;
1081                 acc_mode |= S_IXUGO;
1082         }
1083
1084         /*
1085          * We cannot rely on the fs check since SYSV IPC does have an
1086          * additional creator id...
1087          */
1088         ns = current->nsproxy->ipc_ns;
1089         shp = shm_lock_check(ns, shmid);
1090         if (IS_ERR(shp)) {
1091                 err = PTR_ERR(shp);
1092                 goto out;
1093         }
1094
1095         err = -EACCES;
1096         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1097                 goto out_unlock;
1098
1099         err = security_shm_shmat(shp, shmaddr, shmflg);
1100         if (err)
1101                 goto out_unlock;
1102
1103         path = shp->shm_file->f_path;
1104         path_get(&path);
1105         shp->shm_nattch++;
1106         size = i_size_read(path.dentry->d_inode);
1107         shm_unlock(shp);
1108
1109         err = -ENOMEM;
1110         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1111         if (!sfd) {
1112                 path_put(&path);
1113                 goto out_nattch;
1114         }
1115
1116         file = alloc_file(&path, f_mode,
1117                           is_file_hugepages(shp->shm_file) ?
1118                                 &shm_file_operations_huge :
1119                                 &shm_file_operations);
1120         err = PTR_ERR(file);
1121         if (IS_ERR(file)) {
1122                 kfree(sfd);
1123                 path_put(&path);
1124                 goto out_nattch;
1125         }
1126
1127         file->private_data = sfd;
1128         file->f_mapping = shp->shm_file->f_mapping;
1129         sfd->id = shp->shm_perm.id;
1130         sfd->ns = get_ipc_ns(ns);
1131         sfd->file = shp->shm_file;
1132         sfd->vm_ops = NULL;
1133
1134         err = security_mmap_file(file, prot, flags);
1135         if (err)
1136                 goto out_fput;
1137
1138         down_write(&current->mm->mmap_sem);
1139         if (addr && !(shmflg & SHM_REMAP)) {
1140                 err = -EINVAL;
1141                 if (find_vma_intersection(current->mm, addr, addr + size))
1142                         goto invalid;
1143                 /*
1144                  * If shm segment goes below stack, make sure there is some
1145                  * space left for the stack to grow (at least 4 pages).
1146                  */
1147                 if (addr < current->mm->start_stack &&
1148                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1149                         goto invalid;
1150         }
1151
1152         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1153         *raddr = addr;
1154         err = 0;
1155         if (IS_ERR_VALUE(addr))
1156                 err = (long)addr;
1157 invalid:
1158         up_write(&current->mm->mmap_sem);
1159         if (populate)
1160                 mm_populate(addr, populate);
1161
1162 out_fput:
1163         fput(file);
1164
1165 out_nattch:
1166         down_write(&shm_ids(ns).rw_mutex);
1167         shp = shm_lock(ns, shmid);
1168         BUG_ON(IS_ERR(shp));
1169         shp->shm_nattch--;
1170         if (shm_may_destroy(ns, shp))
1171                 shm_destroy(ns, shp);
1172         else
1173                 shm_unlock(shp);
1174         up_write(&shm_ids(ns).rw_mutex);
1175         return err;
1176
1177 out_unlock:
1178         shm_unlock(shp);
1179 out:
1180         return err;
1181 }
1182
1183 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1184 {
1185         unsigned long ret;
1186         long err;
1187
1188         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1189         if (err)
1190                 return err;
1191         force_successful_syscall_return();
1192         return (long)ret;
1193 }
1194
1195 /*
1196  * detach and kill segment if marked destroyed.
1197  * The work is done in shm_close.
1198  */
1199 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1200 {
1201         struct mm_struct *mm = current->mm;
1202         struct vm_area_struct *vma;
1203         unsigned long addr = (unsigned long)shmaddr;
1204         int retval = -EINVAL;
1205 #ifdef CONFIG_MMU
1206         loff_t size = 0;
1207         struct vm_area_struct *next;
1208 #endif
1209
1210         if (addr & ~PAGE_MASK)
1211                 return retval;
1212
1213         down_write(&mm->mmap_sem);
1214
1215         /*
1216          * This function tries to be smart and unmap shm segments that
1217          * were modified by partial mlock or munmap calls:
1218          * - It first determines the size of the shm segment that should be
1219          *   unmapped: It searches for a vma that is backed by shm and that
1220          *   started at address shmaddr. It records it's size and then unmaps
1221          *   it.
1222          * - Then it unmaps all shm vmas that started at shmaddr and that
1223          *   are within the initially determined size.
1224          * Errors from do_munmap are ignored: the function only fails if
1225          * it's called with invalid parameters or if it's called to unmap
1226          * a part of a vma. Both calls in this function are for full vmas,
1227          * the parameters are directly copied from the vma itself and always
1228          * valid - therefore do_munmap cannot fail. (famous last words?)
1229          */
1230         /*
1231          * If it had been mremap()'d, the starting address would not
1232          * match the usual checks anyway. So assume all vma's are
1233          * above the starting address given.
1234          */
1235         vma = find_vma(mm, addr);
1236
1237 #ifdef CONFIG_MMU
1238         while (vma) {
1239                 next = vma->vm_next;
1240
1241                 /*
1242                  * Check if the starting address would match, i.e. it's
1243                  * a fragment created by mprotect() and/or munmap(), or it
1244                  * otherwise it starts at this address with no hassles.
1245                  */
1246                 if ((vma->vm_ops == &shm_vm_ops) &&
1247                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1248
1249
1250                         size = file_inode(vma->vm_file)->i_size;
1251                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1252                         /*
1253                          * We discovered the size of the shm segment, so
1254                          * break out of here and fall through to the next
1255                          * loop that uses the size information to stop
1256                          * searching for matching vma's.
1257                          */
1258                         retval = 0;
1259                         vma = next;
1260                         break;
1261                 }
1262                 vma = next;
1263         }
1264
1265         /*
1266          * We need look no further than the maximum address a fragment
1267          * could possibly have landed at. Also cast things to loff_t to
1268          * prevent overflows and make comparisons vs. equal-width types.
1269          */
1270         size = PAGE_ALIGN(size);
1271         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1272                 next = vma->vm_next;
1273
1274                 /* finding a matching vma now does not alter retval */
1275                 if ((vma->vm_ops == &shm_vm_ops) &&
1276                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1277
1278                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1279                 vma = next;
1280         }
1281
1282 #else /* CONFIG_MMU */
1283         /* under NOMMU conditions, the exact address to be destroyed must be
1284          * given */
1285         retval = -EINVAL;
1286         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1287                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1288                 retval = 0;
1289         }
1290
1291 #endif
1292
1293         up_write(&mm->mmap_sem);
1294         return retval;
1295 }
1296
1297 #ifdef CONFIG_PROC_FS
1298 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1299 {
1300         struct user_namespace *user_ns = seq_user_ns(s);
1301         struct shmid_kernel *shp = it;
1302         unsigned long rss = 0, swp = 0;
1303
1304         shm_add_rss_swap(shp, &rss, &swp);
1305
1306 #if BITS_PER_LONG <= 32
1307 #define SIZE_SPEC "%10lu"
1308 #else
1309 #define SIZE_SPEC "%21lu"
1310 #endif
1311
1312         return seq_printf(s,
1313                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1314                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1315                           SIZE_SPEC " " SIZE_SPEC "\n",
1316                           shp->shm_perm.key,
1317                           shp->shm_perm.id,
1318                           shp->shm_perm.mode,
1319                           shp->shm_segsz,
1320                           shp->shm_cprid,
1321                           shp->shm_lprid,
1322                           shp->shm_nattch,
1323                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1324                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1325                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1326                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1327                           shp->shm_atim,
1328                           shp->shm_dtim,
1329                           shp->shm_ctim,
1330                           rss * PAGE_SIZE,
1331                           swp * PAGE_SIZE);
1332 }
1333 #endif