ipc,sem: fix semctl(..., GETZCNT)
[linux-3.10.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 /*
128  * shm_lock_(check_) routines are called in the paths where the rw_mutex
129  * is not necessarily held.
130  */
131 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
132 {
133         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
134
135         if (IS_ERR(ipcp))
136                 return (struct shmid_kernel *)ipcp;
137
138         return container_of(ipcp, struct shmid_kernel, shm_perm);
139 }
140
141 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
142 {
143         rcu_read_lock();
144         spin_lock(&ipcp->shm_perm.lock);
145 }
146
147 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
148                                                 int id)
149 {
150         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
151
152         if (IS_ERR(ipcp))
153                 return (struct shmid_kernel *)ipcp;
154
155         return container_of(ipcp, struct shmid_kernel, shm_perm);
156 }
157
158 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
159 {
160         ipc_rmid(&shm_ids(ns), &s->shm_perm);
161 }
162
163
164 /* This is called by fork, once for every shm attach. */
165 static void shm_open(struct vm_area_struct *vma)
166 {
167         struct file *file = vma->vm_file;
168         struct shm_file_data *sfd = shm_file_data(file);
169         struct shmid_kernel *shp;
170
171         shp = shm_lock(sfd->ns, sfd->id);
172         BUG_ON(IS_ERR(shp));
173         shp->shm_atim = get_seconds();
174         shp->shm_lprid = task_tgid_vnr(current);
175         shp->shm_nattch++;
176         shm_unlock(shp);
177 }
178
179 /*
180  * shm_destroy - free the struct shmid_kernel
181  *
182  * @ns: namespace
183  * @shp: struct to free
184  *
185  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
186  * but returns with shp unlocked and freed.
187  */
188 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
189 {
190         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
191         shm_rmid(ns, shp);
192         shm_unlock(shp);
193         if (!is_file_hugepages(shp->shm_file))
194                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
195         else if (shp->mlock_user)
196                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
197                                                 shp->mlock_user);
198         fput (shp->shm_file);
199         security_shm_free(shp);
200         ipc_rcu_putref(shp);
201 }
202
203 /*
204  * shm_may_destroy - identifies whether shm segment should be destroyed now
205  *
206  * Returns true if and only if there are no active users of the segment and
207  * one of the following is true:
208  *
209  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
210  *
211  * 2) sysctl kernel.shm_rmid_forced is set to 1.
212  */
213 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
214 {
215         return (shp->shm_nattch == 0) &&
216                (ns->shm_rmid_forced ||
217                 (shp->shm_perm.mode & SHM_DEST));
218 }
219
220 /*
221  * remove the attach descriptor vma.
222  * free memory for segment if it is marked destroyed.
223  * The descriptor has already been removed from the current->mm->mmap list
224  * and will later be kfree()d.
225  */
226 static void shm_close(struct vm_area_struct *vma)
227 {
228         struct file * file = vma->vm_file;
229         struct shm_file_data *sfd = shm_file_data(file);
230         struct shmid_kernel *shp;
231         struct ipc_namespace *ns = sfd->ns;
232
233         down_write(&shm_ids(ns).rw_mutex);
234         /* remove from the list of attaches of the shm segment */
235         shp = shm_lock(ns, sfd->id);
236         BUG_ON(IS_ERR(shp));
237         shp->shm_lprid = task_tgid_vnr(current);
238         shp->shm_dtim = get_seconds();
239         shp->shm_nattch--;
240         if (shm_may_destroy(ns, shp))
241                 shm_destroy(ns, shp);
242         else
243                 shm_unlock(shp);
244         up_write(&shm_ids(ns).rw_mutex);
245 }
246
247 /* Called with ns->shm_ids(ns).rw_mutex locked */
248 static int shm_try_destroy_current(int id, void *p, void *data)
249 {
250         struct ipc_namespace *ns = data;
251         struct kern_ipc_perm *ipcp = p;
252         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
253
254         if (shp->shm_creator != current)
255                 return 0;
256
257         /*
258          * Mark it as orphaned to destroy the segment when
259          * kernel.shm_rmid_forced is changed.
260          * It is noop if the following shm_may_destroy() returns true.
261          */
262         shp->shm_creator = NULL;
263
264         /*
265          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
266          * is not set, it shouldn't be deleted here.
267          */
268         if (!ns->shm_rmid_forced)
269                 return 0;
270
271         if (shm_may_destroy(ns, shp)) {
272                 shm_lock_by_ptr(shp);
273                 shm_destroy(ns, shp);
274         }
275         return 0;
276 }
277
278 /* Called with ns->shm_ids(ns).rw_mutex locked */
279 static int shm_try_destroy_orphaned(int id, void *p, void *data)
280 {
281         struct ipc_namespace *ns = data;
282         struct kern_ipc_perm *ipcp = p;
283         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
284
285         /*
286          * We want to destroy segments without users and with already
287          * exit'ed originating process.
288          *
289          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
290          */
291         if (shp->shm_creator != NULL)
292                 return 0;
293
294         if (shm_may_destroy(ns, shp)) {
295                 shm_lock_by_ptr(shp);
296                 shm_destroy(ns, shp);
297         }
298         return 0;
299 }
300
301 void shm_destroy_orphaned(struct ipc_namespace *ns)
302 {
303         down_write(&shm_ids(ns).rw_mutex);
304         if (shm_ids(ns).in_use)
305                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
306         up_write(&shm_ids(ns).rw_mutex);
307 }
308
309
310 void exit_shm(struct task_struct *task)
311 {
312         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
313
314         if (shm_ids(ns).in_use == 0)
315                 return;
316
317         /* Destroy all already created segments, but not mapped yet */
318         down_write(&shm_ids(ns).rw_mutex);
319         if (shm_ids(ns).in_use)
320                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
321         up_write(&shm_ids(ns).rw_mutex);
322 }
323
324 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
325 {
326         struct file *file = vma->vm_file;
327         struct shm_file_data *sfd = shm_file_data(file);
328
329         return sfd->vm_ops->fault(vma, vmf);
330 }
331
332 #ifdef CONFIG_NUMA
333 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
334 {
335         struct file *file = vma->vm_file;
336         struct shm_file_data *sfd = shm_file_data(file);
337         int err = 0;
338         if (sfd->vm_ops->set_policy)
339                 err = sfd->vm_ops->set_policy(vma, new);
340         return err;
341 }
342
343 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
344                                         unsigned long addr)
345 {
346         struct file *file = vma->vm_file;
347         struct shm_file_data *sfd = shm_file_data(file);
348         struct mempolicy *pol = NULL;
349
350         if (sfd->vm_ops->get_policy)
351                 pol = sfd->vm_ops->get_policy(vma, addr);
352         else if (vma->vm_policy)
353                 pol = vma->vm_policy;
354
355         return pol;
356 }
357 #endif
358
359 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
360 {
361         struct shm_file_data *sfd = shm_file_data(file);
362         int ret;
363
364         ret = sfd->file->f_op->mmap(sfd->file, vma);
365         if (ret != 0)
366                 return ret;
367         sfd->vm_ops = vma->vm_ops;
368 #ifdef CONFIG_MMU
369         BUG_ON(!sfd->vm_ops->fault);
370 #endif
371         vma->vm_ops = &shm_vm_ops;
372         shm_open(vma);
373
374         return ret;
375 }
376
377 static int shm_release(struct inode *ino, struct file *file)
378 {
379         struct shm_file_data *sfd = shm_file_data(file);
380
381         put_ipc_ns(sfd->ns);
382         shm_file_data(file) = NULL;
383         kfree(sfd);
384         return 0;
385 }
386
387 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
388 {
389         struct shm_file_data *sfd = shm_file_data(file);
390
391         if (!sfd->file->f_op->fsync)
392                 return -EINVAL;
393         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
394 }
395
396 static long shm_fallocate(struct file *file, int mode, loff_t offset,
397                           loff_t len)
398 {
399         struct shm_file_data *sfd = shm_file_data(file);
400
401         if (!sfd->file->f_op->fallocate)
402                 return -EOPNOTSUPP;
403         return sfd->file->f_op->fallocate(file, mode, offset, len);
404 }
405
406 static unsigned long shm_get_unmapped_area(struct file *file,
407         unsigned long addr, unsigned long len, unsigned long pgoff,
408         unsigned long flags)
409 {
410         struct shm_file_data *sfd = shm_file_data(file);
411         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
412                                                 pgoff, flags);
413 }
414
415 static const struct file_operations shm_file_operations = {
416         .mmap           = shm_mmap,
417         .fsync          = shm_fsync,
418         .release        = shm_release,
419 #ifndef CONFIG_MMU
420         .get_unmapped_area      = shm_get_unmapped_area,
421 #endif
422         .llseek         = noop_llseek,
423         .fallocate      = shm_fallocate,
424 };
425
426 static const struct file_operations shm_file_operations_huge = {
427         .mmap           = shm_mmap,
428         .fsync          = shm_fsync,
429         .release        = shm_release,
430         .get_unmapped_area      = shm_get_unmapped_area,
431         .llseek         = noop_llseek,
432         .fallocate      = shm_fallocate,
433 };
434
435 int is_file_shm_hugepages(struct file *file)
436 {
437         return file->f_op == &shm_file_operations_huge;
438 }
439
440 static const struct vm_operations_struct shm_vm_ops = {
441         .open   = shm_open,     /* callback for a new vm-area open */
442         .close  = shm_close,    /* callback for when the vm-area is released */
443         .fault  = shm_fault,
444 #if defined(CONFIG_NUMA)
445         .set_policy = shm_set_policy,
446         .get_policy = shm_get_policy,
447 #endif
448 };
449
450 /**
451  * newseg - Create a new shared memory segment
452  * @ns: namespace
453  * @params: ptr to the structure that contains key, size and shmflg
454  *
455  * Called with shm_ids.rw_mutex held as a writer.
456  */
457
458 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
459 {
460         key_t key = params->key;
461         int shmflg = params->flg;
462         size_t size = params->u.size;
463         int error;
464         struct shmid_kernel *shp;
465         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
466         struct file * file;
467         char name[13];
468         int id;
469         vm_flags_t acctflag = 0;
470
471         if (size < SHMMIN || size > ns->shm_ctlmax)
472                 return -EINVAL;
473
474         if (ns->shm_tot + numpages > ns->shm_ctlall)
475                 return -ENOSPC;
476
477         shp = ipc_rcu_alloc(sizeof(*shp));
478         if (!shp)
479                 return -ENOMEM;
480
481         shp->shm_perm.key = key;
482         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
483         shp->mlock_user = NULL;
484
485         shp->shm_perm.security = NULL;
486         error = security_shm_alloc(shp);
487         if (error) {
488                 ipc_rcu_putref(shp);
489                 return error;
490         }
491
492         sprintf (name, "SYSV%08x", key);
493         if (shmflg & SHM_HUGETLB) {
494                 struct hstate *hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT)
495                                                 & SHM_HUGE_MASK);
496                 size_t hugesize = ALIGN(size, huge_page_size(hs));
497
498                 /* hugetlb_file_setup applies strict accounting */
499                 if (shmflg & SHM_NORESERVE)
500                         acctflag = VM_NORESERVE;
501                 file = hugetlb_file_setup(name, hugesize, acctflag,
502                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
503                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
504         } else {
505                 /*
506                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
507                  * if it's asked for.
508                  */
509                 if  ((shmflg & SHM_NORESERVE) &&
510                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
511                         acctflag = VM_NORESERVE;
512                 file = shmem_file_setup(name, size, acctflag);
513         }
514         error = PTR_ERR(file);
515         if (IS_ERR(file))
516                 goto no_file;
517
518         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
519         if (id < 0) {
520                 error = id;
521                 goto no_id;
522         }
523
524         shp->shm_cprid = task_tgid_vnr(current);
525         shp->shm_lprid = 0;
526         shp->shm_atim = shp->shm_dtim = 0;
527         shp->shm_ctim = get_seconds();
528         shp->shm_segsz = size;
529         shp->shm_nattch = 0;
530         shp->shm_file = file;
531         shp->shm_creator = current;
532         /*
533          * shmid gets reported as "inode#" in /proc/pid/maps.
534          * proc-ps tools use this. Changing this will break them.
535          */
536         file_inode(file)->i_ino = shp->shm_perm.id;
537
538         ns->shm_tot += numpages;
539         error = shp->shm_perm.id;
540         shm_unlock(shp);
541         return error;
542
543 no_id:
544         if (is_file_hugepages(file) && shp->mlock_user)
545                 user_shm_unlock(size, shp->mlock_user);
546         fput(file);
547 no_file:
548         security_shm_free(shp);
549         ipc_rcu_putref(shp);
550         return error;
551 }
552
553 /*
554  * Called with shm_ids.rw_mutex and ipcp locked.
555  */
556 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
557 {
558         struct shmid_kernel *shp;
559
560         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
561         return security_shm_associate(shp, shmflg);
562 }
563
564 /*
565  * Called with shm_ids.rw_mutex and ipcp locked.
566  */
567 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
568                                 struct ipc_params *params)
569 {
570         struct shmid_kernel *shp;
571
572         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
573         if (shp->shm_segsz < params->u.size)
574                 return -EINVAL;
575
576         return 0;
577 }
578
579 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
580 {
581         struct ipc_namespace *ns;
582         struct ipc_ops shm_ops;
583         struct ipc_params shm_params;
584
585         ns = current->nsproxy->ipc_ns;
586
587         shm_ops.getnew = newseg;
588         shm_ops.associate = shm_security;
589         shm_ops.more_checks = shm_more_checks;
590
591         shm_params.key = key;
592         shm_params.flg = shmflg;
593         shm_params.u.size = size;
594
595         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
596 }
597
598 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
599 {
600         switch(version) {
601         case IPC_64:
602                 return copy_to_user(buf, in, sizeof(*in));
603         case IPC_OLD:
604             {
605                 struct shmid_ds out;
606
607                 memset(&out, 0, sizeof(out));
608                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
609                 out.shm_segsz   = in->shm_segsz;
610                 out.shm_atime   = in->shm_atime;
611                 out.shm_dtime   = in->shm_dtime;
612                 out.shm_ctime   = in->shm_ctime;
613                 out.shm_cpid    = in->shm_cpid;
614                 out.shm_lpid    = in->shm_lpid;
615                 out.shm_nattch  = in->shm_nattch;
616
617                 return copy_to_user(buf, &out, sizeof(out));
618             }
619         default:
620                 return -EINVAL;
621         }
622 }
623
624 static inline unsigned long
625 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
626 {
627         switch(version) {
628         case IPC_64:
629                 if (copy_from_user(out, buf, sizeof(*out)))
630                         return -EFAULT;
631                 return 0;
632         case IPC_OLD:
633             {
634                 struct shmid_ds tbuf_old;
635
636                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
637                         return -EFAULT;
638
639                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
640                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
641                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
642
643                 return 0;
644             }
645         default:
646                 return -EINVAL;
647         }
648 }
649
650 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
651 {
652         switch(version) {
653         case IPC_64:
654                 return copy_to_user(buf, in, sizeof(*in));
655         case IPC_OLD:
656             {
657                 struct shminfo out;
658
659                 if(in->shmmax > INT_MAX)
660                         out.shmmax = INT_MAX;
661                 else
662                         out.shmmax = (int)in->shmmax;
663
664                 out.shmmin      = in->shmmin;
665                 out.shmmni      = in->shmmni;
666                 out.shmseg      = in->shmseg;
667                 out.shmall      = in->shmall; 
668
669                 return copy_to_user(buf, &out, sizeof(out));
670             }
671         default:
672                 return -EINVAL;
673         }
674 }
675
676 /*
677  * Calculate and add used RSS and swap pages of a shm.
678  * Called with shm_ids.rw_mutex held as a reader
679  */
680 static void shm_add_rss_swap(struct shmid_kernel *shp,
681         unsigned long *rss_add, unsigned long *swp_add)
682 {
683         struct inode *inode;
684
685         inode = file_inode(shp->shm_file);
686
687         if (is_file_hugepages(shp->shm_file)) {
688                 struct address_space *mapping = inode->i_mapping;
689                 struct hstate *h = hstate_file(shp->shm_file);
690                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
691         } else {
692 #ifdef CONFIG_SHMEM
693                 struct shmem_inode_info *info = SHMEM_I(inode);
694                 spin_lock(&info->lock);
695                 *rss_add += inode->i_mapping->nrpages;
696                 *swp_add += info->swapped;
697                 spin_unlock(&info->lock);
698 #else
699                 *rss_add += inode->i_mapping->nrpages;
700 #endif
701         }
702 }
703
704 /*
705  * Called with shm_ids.rw_mutex held as a reader
706  */
707 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
708                 unsigned long *swp)
709 {
710         int next_id;
711         int total, in_use;
712
713         *rss = 0;
714         *swp = 0;
715
716         in_use = shm_ids(ns).in_use;
717
718         for (total = 0, next_id = 0; total < in_use; next_id++) {
719                 struct kern_ipc_perm *ipc;
720                 struct shmid_kernel *shp;
721
722                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
723                 if (ipc == NULL)
724                         continue;
725                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
726
727                 shm_add_rss_swap(shp, rss, swp);
728
729                 total++;
730         }
731 }
732
733 /*
734  * This function handles some shmctl commands which require the rw_mutex
735  * to be held in write mode.
736  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
737  */
738 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
739                        struct shmid_ds __user *buf, int version)
740 {
741         struct kern_ipc_perm *ipcp;
742         struct shmid64_ds shmid64;
743         struct shmid_kernel *shp;
744         int err;
745
746         if (cmd == IPC_SET) {
747                 if (copy_shmid_from_user(&shmid64, buf, version))
748                         return -EFAULT;
749         }
750
751         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
752                                &shmid64.shm_perm, 0);
753         if (IS_ERR(ipcp))
754                 return PTR_ERR(ipcp);
755
756         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
757
758         err = security_shm_shmctl(shp, cmd);
759         if (err)
760                 goto out_unlock;
761         switch (cmd) {
762         case IPC_RMID:
763                 do_shm_rmid(ns, ipcp);
764                 goto out_up;
765         case IPC_SET:
766                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
767                 if (err)
768                         goto out_unlock;
769                 shp->shm_ctim = get_seconds();
770                 break;
771         default:
772                 err = -EINVAL;
773         }
774 out_unlock:
775         shm_unlock(shp);
776 out_up:
777         up_write(&shm_ids(ns).rw_mutex);
778         return err;
779 }
780
781 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
782 {
783         struct shmid_kernel *shp;
784         int err, version;
785         struct ipc_namespace *ns;
786
787         if (cmd < 0 || shmid < 0) {
788                 err = -EINVAL;
789                 goto out;
790         }
791
792         version = ipc_parse_version(&cmd);
793         ns = current->nsproxy->ipc_ns;
794
795         switch (cmd) { /* replace with proc interface ? */
796         case IPC_INFO:
797         {
798                 struct shminfo64 shminfo;
799
800                 err = security_shm_shmctl(NULL, cmd);
801                 if (err)
802                         return err;
803
804                 memset(&shminfo, 0, sizeof(shminfo));
805                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
806                 shminfo.shmmax = ns->shm_ctlmax;
807                 shminfo.shmall = ns->shm_ctlall;
808
809                 shminfo.shmmin = SHMMIN;
810                 if(copy_shminfo_to_user (buf, &shminfo, version))
811                         return -EFAULT;
812
813                 down_read(&shm_ids(ns).rw_mutex);
814                 err = ipc_get_maxid(&shm_ids(ns));
815                 up_read(&shm_ids(ns).rw_mutex);
816
817                 if(err<0)
818                         err = 0;
819                 goto out;
820         }
821         case SHM_INFO:
822         {
823                 struct shm_info shm_info;
824
825                 err = security_shm_shmctl(NULL, cmd);
826                 if (err)
827                         return err;
828
829                 memset(&shm_info, 0, sizeof(shm_info));
830                 down_read(&shm_ids(ns).rw_mutex);
831                 shm_info.used_ids = shm_ids(ns).in_use;
832                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
833                 shm_info.shm_tot = ns->shm_tot;
834                 shm_info.swap_attempts = 0;
835                 shm_info.swap_successes = 0;
836                 err = ipc_get_maxid(&shm_ids(ns));
837                 up_read(&shm_ids(ns).rw_mutex);
838                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
839                         err = -EFAULT;
840                         goto out;
841                 }
842
843                 err = err < 0 ? 0 : err;
844                 goto out;
845         }
846         case SHM_STAT:
847         case IPC_STAT:
848         {
849                 struct shmid64_ds tbuf;
850                 int result;
851
852                 if (cmd == SHM_STAT) {
853                         shp = shm_lock(ns, shmid);
854                         if (IS_ERR(shp)) {
855                                 err = PTR_ERR(shp);
856                                 goto out;
857                         }
858                         result = shp->shm_perm.id;
859                 } else {
860                         shp = shm_lock_check(ns, shmid);
861                         if (IS_ERR(shp)) {
862                                 err = PTR_ERR(shp);
863                                 goto out;
864                         }
865                         result = 0;
866                 }
867                 err = -EACCES;
868                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
869                         goto out_unlock;
870                 err = security_shm_shmctl(shp, cmd);
871                 if (err)
872                         goto out_unlock;
873                 memset(&tbuf, 0, sizeof(tbuf));
874                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
875                 tbuf.shm_segsz  = shp->shm_segsz;
876                 tbuf.shm_atime  = shp->shm_atim;
877                 tbuf.shm_dtime  = shp->shm_dtim;
878                 tbuf.shm_ctime  = shp->shm_ctim;
879                 tbuf.shm_cpid   = shp->shm_cprid;
880                 tbuf.shm_lpid   = shp->shm_lprid;
881                 tbuf.shm_nattch = shp->shm_nattch;
882                 shm_unlock(shp);
883                 if(copy_shmid_to_user (buf, &tbuf, version))
884                         err = -EFAULT;
885                 else
886                         err = result;
887                 goto out;
888         }
889         case SHM_LOCK:
890         case SHM_UNLOCK:
891         {
892                 struct file *shm_file;
893
894                 shp = shm_lock_check(ns, shmid);
895                 if (IS_ERR(shp)) {
896                         err = PTR_ERR(shp);
897                         goto out;
898                 }
899
900                 audit_ipc_obj(&(shp->shm_perm));
901
902                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
903                         kuid_t euid = current_euid();
904                         err = -EPERM;
905                         if (!uid_eq(euid, shp->shm_perm.uid) &&
906                             !uid_eq(euid, shp->shm_perm.cuid))
907                                 goto out_unlock;
908                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
909                                 goto out_unlock;
910                 }
911
912                 err = security_shm_shmctl(shp, cmd);
913                 if (err)
914                         goto out_unlock;
915
916                 shm_file = shp->shm_file;
917                 if (is_file_hugepages(shm_file))
918                         goto out_unlock;
919
920                 if (cmd == SHM_LOCK) {
921                         struct user_struct *user = current_user();
922                         err = shmem_lock(shm_file, 1, user);
923                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
924                                 shp->shm_perm.mode |= SHM_LOCKED;
925                                 shp->mlock_user = user;
926                         }
927                         goto out_unlock;
928                 }
929
930                 /* SHM_UNLOCK */
931                 if (!(shp->shm_perm.mode & SHM_LOCKED))
932                         goto out_unlock;
933                 shmem_lock(shm_file, 0, shp->mlock_user);
934                 shp->shm_perm.mode &= ~SHM_LOCKED;
935                 shp->mlock_user = NULL;
936                 get_file(shm_file);
937                 shm_unlock(shp);
938                 shmem_unlock_mapping(shm_file->f_mapping);
939                 fput(shm_file);
940                 goto out;
941         }
942         case IPC_RMID:
943         case IPC_SET:
944                 err = shmctl_down(ns, shmid, cmd, buf, version);
945                 return err;
946         default:
947                 return -EINVAL;
948         }
949
950 out_unlock:
951         shm_unlock(shp);
952 out:
953         return err;
954 }
955
956 /*
957  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
958  *
959  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
960  * "raddr" thing points to kernel space, and there has to be a wrapper around
961  * this.
962  */
963 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
964               unsigned long shmlba)
965 {
966         struct shmid_kernel *shp;
967         unsigned long addr;
968         unsigned long size;
969         struct file * file;
970         int    err;
971         unsigned long flags;
972         unsigned long prot;
973         int acc_mode;
974         struct ipc_namespace *ns;
975         struct shm_file_data *sfd;
976         struct path path;
977         fmode_t f_mode;
978         unsigned long populate = 0;
979
980         err = -EINVAL;
981         if (shmid < 0)
982                 goto out;
983         else if ((addr = (ulong)shmaddr)) {
984                 if (addr & (shmlba - 1)) {
985                         if (shmflg & SHM_RND)
986                                 addr &= ~(shmlba - 1);     /* round down */
987                         else
988 #ifndef __ARCH_FORCE_SHMLBA
989                                 if (addr & ~PAGE_MASK)
990 #endif
991                                         goto out;
992                 }
993                 flags = MAP_SHARED | MAP_FIXED;
994         } else {
995                 if ((shmflg & SHM_REMAP))
996                         goto out;
997
998                 flags = MAP_SHARED;
999         }
1000
1001         if (shmflg & SHM_RDONLY) {
1002                 prot = PROT_READ;
1003                 acc_mode = S_IRUGO;
1004                 f_mode = FMODE_READ;
1005         } else {
1006                 prot = PROT_READ | PROT_WRITE;
1007                 acc_mode = S_IRUGO | S_IWUGO;
1008                 f_mode = FMODE_READ | FMODE_WRITE;
1009         }
1010         if (shmflg & SHM_EXEC) {
1011                 prot |= PROT_EXEC;
1012                 acc_mode |= S_IXUGO;
1013         }
1014
1015         /*
1016          * We cannot rely on the fs check since SYSV IPC does have an
1017          * additional creator id...
1018          */
1019         ns = current->nsproxy->ipc_ns;
1020         shp = shm_lock_check(ns, shmid);
1021         if (IS_ERR(shp)) {
1022                 err = PTR_ERR(shp);
1023                 goto out;
1024         }
1025
1026         err = -EACCES;
1027         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1028                 goto out_unlock;
1029
1030         err = security_shm_shmat(shp, shmaddr, shmflg);
1031         if (err)
1032                 goto out_unlock;
1033
1034         path = shp->shm_file->f_path;
1035         path_get(&path);
1036         shp->shm_nattch++;
1037         size = i_size_read(path.dentry->d_inode);
1038         shm_unlock(shp);
1039
1040         err = -ENOMEM;
1041         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1042         if (!sfd)
1043                 goto out_put_dentry;
1044
1045         file = alloc_file(&path, f_mode,
1046                           is_file_hugepages(shp->shm_file) ?
1047                                 &shm_file_operations_huge :
1048                                 &shm_file_operations);
1049         err = PTR_ERR(file);
1050         if (IS_ERR(file))
1051                 goto out_free;
1052
1053         file->private_data = sfd;
1054         file->f_mapping = shp->shm_file->f_mapping;
1055         sfd->id = shp->shm_perm.id;
1056         sfd->ns = get_ipc_ns(ns);
1057         sfd->file = shp->shm_file;
1058         sfd->vm_ops = NULL;
1059
1060         err = security_mmap_file(file, prot, flags);
1061         if (err)
1062                 goto out_fput;
1063
1064         down_write(&current->mm->mmap_sem);
1065         if (addr && !(shmflg & SHM_REMAP)) {
1066                 err = -EINVAL;
1067                 if (find_vma_intersection(current->mm, addr, addr + size))
1068                         goto invalid;
1069                 /*
1070                  * If shm segment goes below stack, make sure there is some
1071                  * space left for the stack to grow (at least 4 pages).
1072                  */
1073                 if (addr < current->mm->start_stack &&
1074                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1075                         goto invalid;
1076         }
1077                 
1078         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1079         *raddr = addr;
1080         err = 0;
1081         if (IS_ERR_VALUE(addr))
1082                 err = (long)addr;
1083 invalid:
1084         up_write(&current->mm->mmap_sem);
1085         if (populate)
1086                 mm_populate(addr, populate);
1087
1088 out_fput:
1089         fput(file);
1090
1091 out_nattch:
1092         down_write(&shm_ids(ns).rw_mutex);
1093         shp = shm_lock(ns, shmid);
1094         BUG_ON(IS_ERR(shp));
1095         shp->shm_nattch--;
1096         if (shm_may_destroy(ns, shp))
1097                 shm_destroy(ns, shp);
1098         else
1099                 shm_unlock(shp);
1100         up_write(&shm_ids(ns).rw_mutex);
1101
1102 out:
1103         return err;
1104
1105 out_unlock:
1106         shm_unlock(shp);
1107         goto out;
1108
1109 out_free:
1110         kfree(sfd);
1111 out_put_dentry:
1112         path_put(&path);
1113         goto out_nattch;
1114 }
1115
1116 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1117 {
1118         unsigned long ret;
1119         long err;
1120
1121         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1122         if (err)
1123                 return err;
1124         force_successful_syscall_return();
1125         return (long)ret;
1126 }
1127
1128 /*
1129  * detach and kill segment if marked destroyed.
1130  * The work is done in shm_close.
1131  */
1132 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1133 {
1134         struct mm_struct *mm = current->mm;
1135         struct vm_area_struct *vma;
1136         unsigned long addr = (unsigned long)shmaddr;
1137         int retval = -EINVAL;
1138 #ifdef CONFIG_MMU
1139         loff_t size = 0;
1140         struct vm_area_struct *next;
1141 #endif
1142
1143         if (addr & ~PAGE_MASK)
1144                 return retval;
1145
1146         down_write(&mm->mmap_sem);
1147
1148         /*
1149          * This function tries to be smart and unmap shm segments that
1150          * were modified by partial mlock or munmap calls:
1151          * - It first determines the size of the shm segment that should be
1152          *   unmapped: It searches for a vma that is backed by shm and that
1153          *   started at address shmaddr. It records it's size and then unmaps
1154          *   it.
1155          * - Then it unmaps all shm vmas that started at shmaddr and that
1156          *   are within the initially determined size.
1157          * Errors from do_munmap are ignored: the function only fails if
1158          * it's called with invalid parameters or if it's called to unmap
1159          * a part of a vma. Both calls in this function are for full vmas,
1160          * the parameters are directly copied from the vma itself and always
1161          * valid - therefore do_munmap cannot fail. (famous last words?)
1162          */
1163         /*
1164          * If it had been mremap()'d, the starting address would not
1165          * match the usual checks anyway. So assume all vma's are
1166          * above the starting address given.
1167          */
1168         vma = find_vma(mm, addr);
1169
1170 #ifdef CONFIG_MMU
1171         while (vma) {
1172                 next = vma->vm_next;
1173
1174                 /*
1175                  * Check if the starting address would match, i.e. it's
1176                  * a fragment created by mprotect() and/or munmap(), or it
1177                  * otherwise it starts at this address with no hassles.
1178                  */
1179                 if ((vma->vm_ops == &shm_vm_ops) &&
1180                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1181
1182
1183                         size = file_inode(vma->vm_file)->i_size;
1184                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1185                         /*
1186                          * We discovered the size of the shm segment, so
1187                          * break out of here and fall through to the next
1188                          * loop that uses the size information to stop
1189                          * searching for matching vma's.
1190                          */
1191                         retval = 0;
1192                         vma = next;
1193                         break;
1194                 }
1195                 vma = next;
1196         }
1197
1198         /*
1199          * We need look no further than the maximum address a fragment
1200          * could possibly have landed at. Also cast things to loff_t to
1201          * prevent overflows and make comparisons vs. equal-width types.
1202          */
1203         size = PAGE_ALIGN(size);
1204         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1205                 next = vma->vm_next;
1206
1207                 /* finding a matching vma now does not alter retval */
1208                 if ((vma->vm_ops == &shm_vm_ops) &&
1209                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1210
1211                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1212                 vma = next;
1213         }
1214
1215 #else /* CONFIG_MMU */
1216         /* under NOMMU conditions, the exact address to be destroyed must be
1217          * given */
1218         retval = -EINVAL;
1219         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1220                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1221                 retval = 0;
1222         }
1223
1224 #endif
1225
1226         up_write(&mm->mmap_sem);
1227         return retval;
1228 }
1229
1230 #ifdef CONFIG_PROC_FS
1231 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1232 {
1233         struct user_namespace *user_ns = seq_user_ns(s);
1234         struct shmid_kernel *shp = it;
1235         unsigned long rss = 0, swp = 0;
1236
1237         shm_add_rss_swap(shp, &rss, &swp);
1238
1239 #if BITS_PER_LONG <= 32
1240 #define SIZE_SPEC "%10lu"
1241 #else
1242 #define SIZE_SPEC "%21lu"
1243 #endif
1244
1245         return seq_printf(s,
1246                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1247                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1248                           SIZE_SPEC " " SIZE_SPEC "\n",
1249                           shp->shm_perm.key,
1250                           shp->shm_perm.id,
1251                           shp->shm_perm.mode,
1252                           shp->shm_segsz,
1253                           shp->shm_cprid,
1254                           shp->shm_lprid,
1255                           shp->shm_nattch,
1256                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1257                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1258                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1259                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1260                           shp->shm_atim,
1261                           shp->shm_dtim,
1262                           shp->shm_ctim,
1263                           rss * PAGE_SIZE,
1264                           swp * PAGE_SIZE);
1265 }
1266 #endif