8c9402d298a0fb6a1ce3f9840670dfa4d9fda5b6
[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         int 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                 /* hugetlb_file_setup applies strict accounting */
495                 if (shmflg & SHM_NORESERVE)
496                         acctflag = VM_NORESERVE;
497                 file = hugetlb_file_setup(name, 0, size, acctflag,
498                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
499                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
500         } else {
501                 /*
502                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
503                  * if it's asked for.
504                  */
505                 if  ((shmflg & SHM_NORESERVE) &&
506                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
507                         acctflag = VM_NORESERVE;
508                 file = shmem_file_setup(name, size, acctflag);
509         }
510         error = PTR_ERR(file);
511         if (IS_ERR(file))
512                 goto no_file;
513
514         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
515         if (id < 0) {
516                 error = id;
517                 goto no_id;
518         }
519
520         shp->shm_cprid = task_tgid_vnr(current);
521         shp->shm_lprid = 0;
522         shp->shm_atim = shp->shm_dtim = 0;
523         shp->shm_ctim = get_seconds();
524         shp->shm_segsz = size;
525         shp->shm_nattch = 0;
526         shp->shm_file = file;
527         shp->shm_creator = current;
528         /*
529          * shmid gets reported as "inode#" in /proc/pid/maps.
530          * proc-ps tools use this. Changing this will break them.
531          */
532         file_inode(file)->i_ino = shp->shm_perm.id;
533
534         ns->shm_tot += numpages;
535         error = shp->shm_perm.id;
536         shm_unlock(shp);
537         return error;
538
539 no_id:
540         if (is_file_hugepages(file) && shp->mlock_user)
541                 user_shm_unlock(size, shp->mlock_user);
542         fput(file);
543 no_file:
544         security_shm_free(shp);
545         ipc_rcu_putref(shp);
546         return error;
547 }
548
549 /*
550  * Called with shm_ids.rw_mutex and ipcp locked.
551  */
552 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
553 {
554         struct shmid_kernel *shp;
555
556         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
557         return security_shm_associate(shp, shmflg);
558 }
559
560 /*
561  * Called with shm_ids.rw_mutex and ipcp locked.
562  */
563 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
564                                 struct ipc_params *params)
565 {
566         struct shmid_kernel *shp;
567
568         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
569         if (shp->shm_segsz < params->u.size)
570                 return -EINVAL;
571
572         return 0;
573 }
574
575 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
576 {
577         struct ipc_namespace *ns;
578         struct ipc_ops shm_ops;
579         struct ipc_params shm_params;
580
581         ns = current->nsproxy->ipc_ns;
582
583         shm_ops.getnew = newseg;
584         shm_ops.associate = shm_security;
585         shm_ops.more_checks = shm_more_checks;
586
587         shm_params.key = key;
588         shm_params.flg = shmflg;
589         shm_params.u.size = size;
590
591         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
592 }
593
594 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
595 {
596         switch(version) {
597         case IPC_64:
598                 return copy_to_user(buf, in, sizeof(*in));
599         case IPC_OLD:
600             {
601                 struct shmid_ds out;
602
603                 memset(&out, 0, sizeof(out));
604                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
605                 out.shm_segsz   = in->shm_segsz;
606                 out.shm_atime   = in->shm_atime;
607                 out.shm_dtime   = in->shm_dtime;
608                 out.shm_ctime   = in->shm_ctime;
609                 out.shm_cpid    = in->shm_cpid;
610                 out.shm_lpid    = in->shm_lpid;
611                 out.shm_nattch  = in->shm_nattch;
612
613                 return copy_to_user(buf, &out, sizeof(out));
614             }
615         default:
616                 return -EINVAL;
617         }
618 }
619
620 static inline unsigned long
621 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
622 {
623         switch(version) {
624         case IPC_64:
625                 if (copy_from_user(out, buf, sizeof(*out)))
626                         return -EFAULT;
627                 return 0;
628         case IPC_OLD:
629             {
630                 struct shmid_ds tbuf_old;
631
632                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
633                         return -EFAULT;
634
635                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
636                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
637                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
638
639                 return 0;
640             }
641         default:
642                 return -EINVAL;
643         }
644 }
645
646 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
647 {
648         switch(version) {
649         case IPC_64:
650                 return copy_to_user(buf, in, sizeof(*in));
651         case IPC_OLD:
652             {
653                 struct shminfo out;
654
655                 if(in->shmmax > INT_MAX)
656                         out.shmmax = INT_MAX;
657                 else
658                         out.shmmax = (int)in->shmmax;
659
660                 out.shmmin      = in->shmmin;
661                 out.shmmni      = in->shmmni;
662                 out.shmseg      = in->shmseg;
663                 out.shmall      = in->shmall; 
664
665                 return copy_to_user(buf, &out, sizeof(out));
666             }
667         default:
668                 return -EINVAL;
669         }
670 }
671
672 /*
673  * Calculate and add used RSS and swap pages of a shm.
674  * Called with shm_ids.rw_mutex held as a reader
675  */
676 static void shm_add_rss_swap(struct shmid_kernel *shp,
677         unsigned long *rss_add, unsigned long *swp_add)
678 {
679         struct inode *inode;
680
681         inode = file_inode(shp->shm_file);
682
683         if (is_file_hugepages(shp->shm_file)) {
684                 struct address_space *mapping = inode->i_mapping;
685                 struct hstate *h = hstate_file(shp->shm_file);
686                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
687         } else {
688 #ifdef CONFIG_SHMEM
689                 struct shmem_inode_info *info = SHMEM_I(inode);
690                 spin_lock(&info->lock);
691                 *rss_add += inode->i_mapping->nrpages;
692                 *swp_add += info->swapped;
693                 spin_unlock(&info->lock);
694 #else
695                 *rss_add += inode->i_mapping->nrpages;
696 #endif
697         }
698 }
699
700 /*
701  * Called with shm_ids.rw_mutex held as a reader
702  */
703 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
704                 unsigned long *swp)
705 {
706         int next_id;
707         int total, in_use;
708
709         *rss = 0;
710         *swp = 0;
711
712         in_use = shm_ids(ns).in_use;
713
714         for (total = 0, next_id = 0; total < in_use; next_id++) {
715                 struct kern_ipc_perm *ipc;
716                 struct shmid_kernel *shp;
717
718                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
719                 if (ipc == NULL)
720                         continue;
721                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
722
723                 shm_add_rss_swap(shp, rss, swp);
724
725                 total++;
726         }
727 }
728
729 /*
730  * This function handles some shmctl commands which require the rw_mutex
731  * to be held in write mode.
732  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
733  */
734 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
735                        struct shmid_ds __user *buf, int version)
736 {
737         struct kern_ipc_perm *ipcp;
738         struct shmid64_ds shmid64;
739         struct shmid_kernel *shp;
740         int err;
741
742         if (cmd == IPC_SET) {
743                 if (copy_shmid_from_user(&shmid64, buf, version))
744                         return -EFAULT;
745         }
746
747         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
748                                &shmid64.shm_perm, 0);
749         if (IS_ERR(ipcp))
750                 return PTR_ERR(ipcp);
751
752         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
753
754         err = security_shm_shmctl(shp, cmd);
755         if (err)
756                 goto out_unlock;
757         switch (cmd) {
758         case IPC_RMID:
759                 do_shm_rmid(ns, ipcp);
760                 goto out_up;
761         case IPC_SET:
762                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
763                 if (err)
764                         goto out_unlock;
765                 shp->shm_ctim = get_seconds();
766                 break;
767         default:
768                 err = -EINVAL;
769         }
770 out_unlock:
771         shm_unlock(shp);
772 out_up:
773         up_write(&shm_ids(ns).rw_mutex);
774         return err;
775 }
776
777 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
778 {
779         struct shmid_kernel *shp;
780         int err, version;
781         struct ipc_namespace *ns;
782
783         if (cmd < 0 || shmid < 0) {
784                 err = -EINVAL;
785                 goto out;
786         }
787
788         version = ipc_parse_version(&cmd);
789         ns = current->nsproxy->ipc_ns;
790
791         switch (cmd) { /* replace with proc interface ? */
792         case IPC_INFO:
793         {
794                 struct shminfo64 shminfo;
795
796                 err = security_shm_shmctl(NULL, cmd);
797                 if (err)
798                         return err;
799
800                 memset(&shminfo, 0, sizeof(shminfo));
801                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
802                 shminfo.shmmax = ns->shm_ctlmax;
803                 shminfo.shmall = ns->shm_ctlall;
804
805                 shminfo.shmmin = SHMMIN;
806                 if(copy_shminfo_to_user (buf, &shminfo, version))
807                         return -EFAULT;
808
809                 down_read(&shm_ids(ns).rw_mutex);
810                 err = ipc_get_maxid(&shm_ids(ns));
811                 up_read(&shm_ids(ns).rw_mutex);
812
813                 if(err<0)
814                         err = 0;
815                 goto out;
816         }
817         case SHM_INFO:
818         {
819                 struct shm_info shm_info;
820
821                 err = security_shm_shmctl(NULL, cmd);
822                 if (err)
823                         return err;
824
825                 memset(&shm_info, 0, sizeof(shm_info));
826                 down_read(&shm_ids(ns).rw_mutex);
827                 shm_info.used_ids = shm_ids(ns).in_use;
828                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
829                 shm_info.shm_tot = ns->shm_tot;
830                 shm_info.swap_attempts = 0;
831                 shm_info.swap_successes = 0;
832                 err = ipc_get_maxid(&shm_ids(ns));
833                 up_read(&shm_ids(ns).rw_mutex);
834                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
835                         err = -EFAULT;
836                         goto out;
837                 }
838
839                 err = err < 0 ? 0 : err;
840                 goto out;
841         }
842         case SHM_STAT:
843         case IPC_STAT:
844         {
845                 struct shmid64_ds tbuf;
846                 int result;
847
848                 if (cmd == SHM_STAT) {
849                         shp = shm_lock(ns, shmid);
850                         if (IS_ERR(shp)) {
851                                 err = PTR_ERR(shp);
852                                 goto out;
853                         }
854                         result = shp->shm_perm.id;
855                 } else {
856                         shp = shm_lock_check(ns, shmid);
857                         if (IS_ERR(shp)) {
858                                 err = PTR_ERR(shp);
859                                 goto out;
860                         }
861                         result = 0;
862                 }
863                 err = -EACCES;
864                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
865                         goto out_unlock;
866                 err = security_shm_shmctl(shp, cmd);
867                 if (err)
868                         goto out_unlock;
869                 memset(&tbuf, 0, sizeof(tbuf));
870                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
871                 tbuf.shm_segsz  = shp->shm_segsz;
872                 tbuf.shm_atime  = shp->shm_atim;
873                 tbuf.shm_dtime  = shp->shm_dtim;
874                 tbuf.shm_ctime  = shp->shm_ctim;
875                 tbuf.shm_cpid   = shp->shm_cprid;
876                 tbuf.shm_lpid   = shp->shm_lprid;
877                 tbuf.shm_nattch = shp->shm_nattch;
878                 shm_unlock(shp);
879                 if(copy_shmid_to_user (buf, &tbuf, version))
880                         err = -EFAULT;
881                 else
882                         err = result;
883                 goto out;
884         }
885         case SHM_LOCK:
886         case SHM_UNLOCK:
887         {
888                 struct file *shm_file;
889
890                 shp = shm_lock_check(ns, shmid);
891                 if (IS_ERR(shp)) {
892                         err = PTR_ERR(shp);
893                         goto out;
894                 }
895
896                 audit_ipc_obj(&(shp->shm_perm));
897
898                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
899                         kuid_t euid = current_euid();
900                         err = -EPERM;
901                         if (!uid_eq(euid, shp->shm_perm.uid) &&
902                             !uid_eq(euid, shp->shm_perm.cuid))
903                                 goto out_unlock;
904                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
905                                 goto out_unlock;
906                 }
907
908                 err = security_shm_shmctl(shp, cmd);
909                 if (err)
910                         goto out_unlock;
911
912                 shm_file = shp->shm_file;
913                 if (is_file_hugepages(shm_file))
914                         goto out_unlock;
915
916                 if (cmd == SHM_LOCK) {
917                         struct user_struct *user = current_user();
918                         err = shmem_lock(shm_file, 1, user);
919                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
920                                 shp->shm_perm.mode |= SHM_LOCKED;
921                                 shp->mlock_user = user;
922                         }
923                         goto out_unlock;
924                 }
925
926                 /* SHM_UNLOCK */
927                 if (!(shp->shm_perm.mode & SHM_LOCKED))
928                         goto out_unlock;
929                 shmem_lock(shm_file, 0, shp->mlock_user);
930                 shp->shm_perm.mode &= ~SHM_LOCKED;
931                 shp->mlock_user = NULL;
932                 get_file(shm_file);
933                 shm_unlock(shp);
934                 shmem_unlock_mapping(shm_file->f_mapping);
935                 fput(shm_file);
936                 goto out;
937         }
938         case IPC_RMID:
939         case IPC_SET:
940                 err = shmctl_down(ns, shmid, cmd, buf, version);
941                 return err;
942         default:
943                 return -EINVAL;
944         }
945
946 out_unlock:
947         shm_unlock(shp);
948 out:
949         return err;
950 }
951
952 /*
953  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
954  *
955  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
956  * "raddr" thing points to kernel space, and there has to be a wrapper around
957  * this.
958  */
959 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
960               unsigned long shmlba)
961 {
962         struct shmid_kernel *shp;
963         unsigned long addr;
964         unsigned long size;
965         struct file * file;
966         int    err;
967         unsigned long flags;
968         unsigned long prot;
969         int acc_mode;
970         unsigned long user_addr;
971         struct ipc_namespace *ns;
972         struct shm_file_data *sfd;
973         struct path path;
974         fmode_t f_mode;
975
976         err = -EINVAL;
977         if (shmid < 0)
978                 goto out;
979         else if ((addr = (ulong)shmaddr)) {
980                 if (addr & (shmlba - 1)) {
981                         if (shmflg & SHM_RND)
982                                 addr &= ~(shmlba - 1);     /* round down */
983                         else
984 #ifndef __ARCH_FORCE_SHMLBA
985                                 if (addr & ~PAGE_MASK)
986 #endif
987                                         goto out;
988                 }
989                 flags = MAP_SHARED | MAP_FIXED;
990         } else {
991                 if ((shmflg & SHM_REMAP))
992                         goto out;
993
994                 flags = MAP_SHARED;
995         }
996
997         if (shmflg & SHM_RDONLY) {
998                 prot = PROT_READ;
999                 acc_mode = S_IRUGO;
1000                 f_mode = FMODE_READ;
1001         } else {
1002                 prot = PROT_READ | PROT_WRITE;
1003                 acc_mode = S_IRUGO | S_IWUGO;
1004                 f_mode = FMODE_READ | FMODE_WRITE;
1005         }
1006         if (shmflg & SHM_EXEC) {
1007                 prot |= PROT_EXEC;
1008                 acc_mode |= S_IXUGO;
1009         }
1010
1011         /*
1012          * We cannot rely on the fs check since SYSV IPC does have an
1013          * additional creator id...
1014          */
1015         ns = current->nsproxy->ipc_ns;
1016         shp = shm_lock_check(ns, shmid);
1017         if (IS_ERR(shp)) {
1018                 err = PTR_ERR(shp);
1019                 goto out;
1020         }
1021
1022         err = -EACCES;
1023         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1024                 goto out_unlock;
1025
1026         err = security_shm_shmat(shp, shmaddr, shmflg);
1027         if (err)
1028                 goto out_unlock;
1029
1030         path = shp->shm_file->f_path;
1031         path_get(&path);
1032         shp->shm_nattch++;
1033         size = i_size_read(path.dentry->d_inode);
1034         shm_unlock(shp);
1035
1036         err = -ENOMEM;
1037         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1038         if (!sfd)
1039                 goto out_put_dentry;
1040
1041         file = alloc_file(&path, f_mode,
1042                           is_file_hugepages(shp->shm_file) ?
1043                                 &shm_file_operations_huge :
1044                                 &shm_file_operations);
1045         if (!file)
1046                 goto out_free;
1047
1048         file->private_data = sfd;
1049         file->f_mapping = shp->shm_file->f_mapping;
1050         sfd->id = shp->shm_perm.id;
1051         sfd->ns = get_ipc_ns(ns);
1052         sfd->file = shp->shm_file;
1053         sfd->vm_ops = NULL;
1054
1055         err = security_mmap_file(file, prot, flags);
1056         if (err)
1057                 goto out_fput;
1058
1059         down_write(&current->mm->mmap_sem);
1060         if (addr && !(shmflg & SHM_REMAP)) {
1061                 err = -EINVAL;
1062                 if (find_vma_intersection(current->mm, addr, addr + size))
1063                         goto invalid;
1064                 /*
1065                  * If shm segment goes below stack, make sure there is some
1066                  * space left for the stack to grow (at least 4 pages).
1067                  */
1068                 if (addr < current->mm->start_stack &&
1069                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1070                         goto invalid;
1071         }
1072                 
1073         user_addr = do_mmap_pgoff(file, addr, size, prot, flags, 0);
1074         *raddr = user_addr;
1075         err = 0;
1076         if (IS_ERR_VALUE(user_addr))
1077                 err = (long)user_addr;
1078 invalid:
1079         up_write(&current->mm->mmap_sem);
1080
1081 out_fput:
1082         fput(file);
1083
1084 out_nattch:
1085         down_write(&shm_ids(ns).rw_mutex);
1086         shp = shm_lock(ns, shmid);
1087         BUG_ON(IS_ERR(shp));
1088         shp->shm_nattch--;
1089         if (shm_may_destroy(ns, shp))
1090                 shm_destroy(ns, shp);
1091         else
1092                 shm_unlock(shp);
1093         up_write(&shm_ids(ns).rw_mutex);
1094
1095 out:
1096         return err;
1097
1098 out_unlock:
1099         shm_unlock(shp);
1100         goto out;
1101
1102 out_free:
1103         kfree(sfd);
1104 out_put_dentry:
1105         path_put(&path);
1106         goto out_nattch;
1107 }
1108
1109 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1110 {
1111         unsigned long ret;
1112         long err;
1113
1114         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1115         if (err)
1116                 return err;
1117         force_successful_syscall_return();
1118         return (long)ret;
1119 }
1120
1121 /*
1122  * detach and kill segment if marked destroyed.
1123  * The work is done in shm_close.
1124  */
1125 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1126 {
1127         struct mm_struct *mm = current->mm;
1128         struct vm_area_struct *vma;
1129         unsigned long addr = (unsigned long)shmaddr;
1130         int retval = -EINVAL;
1131 #ifdef CONFIG_MMU
1132         loff_t size = 0;
1133         struct vm_area_struct *next;
1134 #endif
1135
1136         if (addr & ~PAGE_MASK)
1137                 return retval;
1138
1139         down_write(&mm->mmap_sem);
1140
1141         /*
1142          * This function tries to be smart and unmap shm segments that
1143          * were modified by partial mlock or munmap calls:
1144          * - It first determines the size of the shm segment that should be
1145          *   unmapped: It searches for a vma that is backed by shm and that
1146          *   started at address shmaddr. It records it's size and then unmaps
1147          *   it.
1148          * - Then it unmaps all shm vmas that started at shmaddr and that
1149          *   are within the initially determined size.
1150          * Errors from do_munmap are ignored: the function only fails if
1151          * it's called with invalid parameters or if it's called to unmap
1152          * a part of a vma. Both calls in this function are for full vmas,
1153          * the parameters are directly copied from the vma itself and always
1154          * valid - therefore do_munmap cannot fail. (famous last words?)
1155          */
1156         /*
1157          * If it had been mremap()'d, the starting address would not
1158          * match the usual checks anyway. So assume all vma's are
1159          * above the starting address given.
1160          */
1161         vma = find_vma(mm, addr);
1162
1163 #ifdef CONFIG_MMU
1164         while (vma) {
1165                 next = vma->vm_next;
1166
1167                 /*
1168                  * Check if the starting address would match, i.e. it's
1169                  * a fragment created by mprotect() and/or munmap(), or it
1170                  * otherwise it starts at this address with no hassles.
1171                  */
1172                 if ((vma->vm_ops == &shm_vm_ops) &&
1173                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1174
1175
1176                         size = file_inode(vma->vm_file)->i_size;
1177                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1178                         /*
1179                          * We discovered the size of the shm segment, so
1180                          * break out of here and fall through to the next
1181                          * loop that uses the size information to stop
1182                          * searching for matching vma's.
1183                          */
1184                         retval = 0;
1185                         vma = next;
1186                         break;
1187                 }
1188                 vma = next;
1189         }
1190
1191         /*
1192          * We need look no further than the maximum address a fragment
1193          * could possibly have landed at. Also cast things to loff_t to
1194          * prevent overflows and make comparisons vs. equal-width types.
1195          */
1196         size = PAGE_ALIGN(size);
1197         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1198                 next = vma->vm_next;
1199
1200                 /* finding a matching vma now does not alter retval */
1201                 if ((vma->vm_ops == &shm_vm_ops) &&
1202                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1203
1204                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1205                 vma = next;
1206         }
1207
1208 #else /* CONFIG_MMU */
1209         /* under NOMMU conditions, the exact address to be destroyed must be
1210          * given */
1211         retval = -EINVAL;
1212         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1213                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1214                 retval = 0;
1215         }
1216
1217 #endif
1218
1219         up_write(&mm->mmap_sem);
1220         return retval;
1221 }
1222
1223 #ifdef CONFIG_PROC_FS
1224 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1225 {
1226         struct user_namespace *user_ns = seq_user_ns(s);
1227         struct shmid_kernel *shp = it;
1228         unsigned long rss = 0, swp = 0;
1229
1230         shm_add_rss_swap(shp, &rss, &swp);
1231
1232 #if BITS_PER_LONG <= 32
1233 #define SIZE_SPEC "%10lu"
1234 #else
1235 #define SIZE_SPEC "%21lu"
1236 #endif
1237
1238         return seq_printf(s,
1239                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1240                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1241                           SIZE_SPEC " " SIZE_SPEC "\n",
1242                           shp->shm_perm.key,
1243                           shp->shm_perm.id,
1244                           shp->shm_perm.mode,
1245                           shp->shm_segsz,
1246                           shp->shm_cprid,
1247                           shp->shm_lprid,
1248                           shp->shm_nattch,
1249                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1250                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1251                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1252                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1253                           shp->shm_atim,
1254                           shp->shm_dtim,
1255                           shp->shm_ctim,
1256                           rss * PAGE_SIZE,
1257                           swp * PAGE_SIZE);
1258 }
1259 #endif