Linux 3.2.102
[pandora-kernel.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(shp->shm_file->f_path.dentry->d_inode->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 unsigned long shm_get_unmapped_area(struct file *file,
397         unsigned long addr, unsigned long len, unsigned long pgoff,
398         unsigned long flags)
399 {
400         struct shm_file_data *sfd = shm_file_data(file);
401         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
402                                                 pgoff, flags);
403 }
404
405 static const struct file_operations shm_file_operations = {
406         .mmap           = shm_mmap,
407         .fsync          = shm_fsync,
408         .release        = shm_release,
409 #ifndef CONFIG_MMU
410         .get_unmapped_area      = shm_get_unmapped_area,
411 #endif
412         .llseek         = noop_llseek,
413 };
414
415 static const struct file_operations shm_file_operations_huge = {
416         .mmap           = shm_mmap,
417         .fsync          = shm_fsync,
418         .release        = shm_release,
419         .get_unmapped_area      = shm_get_unmapped_area,
420         .llseek         = noop_llseek,
421 };
422
423 int is_file_shm_hugepages(struct file *file)
424 {
425         return file->f_op == &shm_file_operations_huge;
426 }
427
428 static const struct vm_operations_struct shm_vm_ops = {
429         .open   = shm_open,     /* callback for a new vm-area open */
430         .close  = shm_close,    /* callback for when the vm-area is released */
431         .fault  = shm_fault,
432 #if defined(CONFIG_NUMA)
433         .set_policy = shm_set_policy,
434         .get_policy = shm_get_policy,
435 #endif
436 };
437
438 /**
439  * newseg - Create a new shared memory segment
440  * @ns: namespace
441  * @params: ptr to the structure that contains key, size and shmflg
442  *
443  * Called with shm_ids.rw_mutex held as a writer.
444  */
445
446 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
447 {
448         key_t key = params->key;
449         int shmflg = params->flg;
450         size_t size = params->u.size;
451         int error;
452         struct shmid_kernel *shp;
453         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
454         struct file * file;
455         char name[13];
456         int id;
457         vm_flags_t acctflag = 0;
458
459         if (size < SHMMIN || size > ns->shm_ctlmax)
460                 return -EINVAL;
461
462         if (ns->shm_tot + numpages > ns->shm_ctlall)
463                 return -ENOSPC;
464
465         shp = ipc_rcu_alloc(sizeof(*shp));
466         if (!shp)
467                 return -ENOMEM;
468
469         shp->shm_perm.key = key;
470         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
471         shp->mlock_user = NULL;
472
473         shp->shm_perm.security = NULL;
474         error = security_shm_alloc(shp);
475         if (error) {
476                 ipc_rcu_putref(shp);
477                 return error;
478         }
479
480         sprintf (name, "SYSV%08x", key);
481         if (shmflg & SHM_HUGETLB) {
482                 /* hugetlb_file_setup applies strict accounting */
483                 if (shmflg & SHM_NORESERVE)
484                         acctflag = VM_NORESERVE;
485                 file = hugetlb_file_setup(name, size, acctflag,
486                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
487         } else {
488                 /*
489                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
490                  * if it's asked for.
491                  */
492                 if  ((shmflg & SHM_NORESERVE) &&
493                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
494                         acctflag = VM_NORESERVE;
495                 file = shmem_file_setup(name, size, acctflag);
496         }
497         error = PTR_ERR(file);
498         if (IS_ERR(file))
499                 goto no_file;
500
501         shp->shm_cprid = task_tgid_vnr(current);
502         shp->shm_lprid = 0;
503         shp->shm_atim = shp->shm_dtim = 0;
504         shp->shm_ctim = get_seconds();
505         shp->shm_segsz = size;
506         shp->shm_nattch = 0;
507         shp->shm_file = file;
508         shp->shm_creator = current;
509
510         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
511         if (id < 0) {
512                 error = id;
513                 goto no_id;
514         }
515
516         /*
517          * shmid gets reported as "inode#" in /proc/pid/maps.
518          * proc-ps tools use this. Changing this will break them.
519          */
520         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
521
522         ns->shm_tot += numpages;
523         error = shp->shm_perm.id;
524         shm_unlock(shp);
525         return error;
526
527 no_id:
528         if (is_file_hugepages(file) && shp->mlock_user)
529                 user_shm_unlock(size, shp->mlock_user);
530         fput(file);
531 no_file:
532         security_shm_free(shp);
533         ipc_rcu_putref(shp);
534         return error;
535 }
536
537 /*
538  * Called with shm_ids.rw_mutex and ipcp locked.
539  */
540 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
541 {
542         struct shmid_kernel *shp;
543
544         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
545         return security_shm_associate(shp, shmflg);
546 }
547
548 /*
549  * Called with shm_ids.rw_mutex and ipcp locked.
550  */
551 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
552                                 struct ipc_params *params)
553 {
554         struct shmid_kernel *shp;
555
556         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
557         if (shp->shm_segsz < params->u.size)
558                 return -EINVAL;
559
560         return 0;
561 }
562
563 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
564 {
565         struct ipc_namespace *ns;
566         struct ipc_ops shm_ops;
567         struct ipc_params shm_params;
568
569         ns = current->nsproxy->ipc_ns;
570
571         shm_ops.getnew = newseg;
572         shm_ops.associate = shm_security;
573         shm_ops.more_checks = shm_more_checks;
574
575         shm_params.key = key;
576         shm_params.flg = shmflg;
577         shm_params.u.size = size;
578
579         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
580 }
581
582 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
583 {
584         switch(version) {
585         case IPC_64:
586                 return copy_to_user(buf, in, sizeof(*in));
587         case IPC_OLD:
588             {
589                 struct shmid_ds out;
590
591                 memset(&out, 0, sizeof(out));
592                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
593                 out.shm_segsz   = in->shm_segsz;
594                 out.shm_atime   = in->shm_atime;
595                 out.shm_dtime   = in->shm_dtime;
596                 out.shm_ctime   = in->shm_ctime;
597                 out.shm_cpid    = in->shm_cpid;
598                 out.shm_lpid    = in->shm_lpid;
599                 out.shm_nattch  = in->shm_nattch;
600
601                 return copy_to_user(buf, &out, sizeof(out));
602             }
603         default:
604                 return -EINVAL;
605         }
606 }
607
608 static inline unsigned long
609 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
610 {
611         switch(version) {
612         case IPC_64:
613                 if (copy_from_user(out, buf, sizeof(*out)))
614                         return -EFAULT;
615                 return 0;
616         case IPC_OLD:
617             {
618                 struct shmid_ds tbuf_old;
619
620                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
621                         return -EFAULT;
622
623                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
624                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
625                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
626
627                 return 0;
628             }
629         default:
630                 return -EINVAL;
631         }
632 }
633
634 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
635 {
636         switch(version) {
637         case IPC_64:
638                 return copy_to_user(buf, in, sizeof(*in));
639         case IPC_OLD:
640             {
641                 struct shminfo out;
642
643                 if(in->shmmax > INT_MAX)
644                         out.shmmax = INT_MAX;
645                 else
646                         out.shmmax = (int)in->shmmax;
647
648                 out.shmmin      = in->shmmin;
649                 out.shmmni      = in->shmmni;
650                 out.shmseg      = in->shmseg;
651                 out.shmall      = in->shmall; 
652
653                 return copy_to_user(buf, &out, sizeof(out));
654             }
655         default:
656                 return -EINVAL;
657         }
658 }
659
660 /*
661  * Calculate and add used RSS and swap pages of a shm.
662  * Called with shm_ids.rw_mutex held as a reader
663  */
664 static void shm_add_rss_swap(struct shmid_kernel *shp,
665         unsigned long *rss_add, unsigned long *swp_add)
666 {
667         struct inode *inode;
668
669         inode = shp->shm_file->f_path.dentry->d_inode;
670
671         if (is_file_hugepages(shp->shm_file)) {
672                 struct address_space *mapping = inode->i_mapping;
673                 struct hstate *h = hstate_file(shp->shm_file);
674                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
675         } else {
676 #ifdef CONFIG_SHMEM
677                 struct shmem_inode_info *info = SHMEM_I(inode);
678                 spin_lock(&info->lock);
679                 *rss_add += inode->i_mapping->nrpages;
680                 *swp_add += info->swapped;
681                 spin_unlock(&info->lock);
682 #else
683                 *rss_add += inode->i_mapping->nrpages;
684 #endif
685         }
686 }
687
688 /*
689  * Called with shm_ids.rw_mutex held as a reader
690  */
691 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
692                 unsigned long *swp)
693 {
694         int next_id;
695         int total, in_use;
696
697         *rss = 0;
698         *swp = 0;
699
700         in_use = shm_ids(ns).in_use;
701
702         for (total = 0, next_id = 0; total < in_use; next_id++) {
703                 struct kern_ipc_perm *ipc;
704                 struct shmid_kernel *shp;
705
706                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
707                 if (ipc == NULL)
708                         continue;
709                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
710
711                 shm_add_rss_swap(shp, rss, swp);
712
713                 total++;
714         }
715 }
716
717 /*
718  * This function handles some shmctl commands which require the rw_mutex
719  * to be held in write mode.
720  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
721  */
722 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
723                        struct shmid_ds __user *buf, int version)
724 {
725         struct kern_ipc_perm *ipcp;
726         struct shmid64_ds shmid64;
727         struct shmid_kernel *shp;
728         int err;
729
730         if (cmd == IPC_SET) {
731                 if (copy_shmid_from_user(&shmid64, buf, version))
732                         return -EFAULT;
733         }
734
735         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
736                                &shmid64.shm_perm, 0);
737         if (IS_ERR(ipcp))
738                 return PTR_ERR(ipcp);
739
740         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
741
742         err = security_shm_shmctl(shp, cmd);
743         if (err)
744                 goto out_unlock;
745         switch (cmd) {
746         case IPC_RMID:
747                 do_shm_rmid(ns, ipcp);
748                 goto out_up;
749         case IPC_SET:
750                 ipc_update_perm(&shmid64.shm_perm, ipcp);
751                 shp->shm_ctim = get_seconds();
752                 break;
753         default:
754                 err = -EINVAL;
755         }
756 out_unlock:
757         shm_unlock(shp);
758 out_up:
759         up_write(&shm_ids(ns).rw_mutex);
760         return err;
761 }
762
763 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
764 {
765         struct shmid_kernel *shp;
766         int err, version;
767         struct ipc_namespace *ns;
768
769         if (cmd < 0 || shmid < 0) {
770                 err = -EINVAL;
771                 goto out;
772         }
773
774         version = ipc_parse_version(&cmd);
775         ns = current->nsproxy->ipc_ns;
776
777         switch (cmd) { /* replace with proc interface ? */
778         case IPC_INFO:
779         {
780                 struct shminfo64 shminfo;
781
782                 err = security_shm_shmctl(NULL, cmd);
783                 if (err)
784                         return err;
785
786                 memset(&shminfo, 0, sizeof(shminfo));
787                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
788                 shminfo.shmmax = ns->shm_ctlmax;
789                 shminfo.shmall = ns->shm_ctlall;
790
791                 shminfo.shmmin = SHMMIN;
792                 if(copy_shminfo_to_user (buf, &shminfo, version))
793                         return -EFAULT;
794
795                 down_read(&shm_ids(ns).rw_mutex);
796                 err = ipc_get_maxid(&shm_ids(ns));
797                 up_read(&shm_ids(ns).rw_mutex);
798
799                 if(err<0)
800                         err = 0;
801                 goto out;
802         }
803         case SHM_INFO:
804         {
805                 struct shm_info shm_info;
806
807                 err = security_shm_shmctl(NULL, cmd);
808                 if (err)
809                         return err;
810
811                 memset(&shm_info, 0, sizeof(shm_info));
812                 down_read(&shm_ids(ns).rw_mutex);
813                 shm_info.used_ids = shm_ids(ns).in_use;
814                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
815                 shm_info.shm_tot = ns->shm_tot;
816                 shm_info.swap_attempts = 0;
817                 shm_info.swap_successes = 0;
818                 err = ipc_get_maxid(&shm_ids(ns));
819                 up_read(&shm_ids(ns).rw_mutex);
820                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
821                         err = -EFAULT;
822                         goto out;
823                 }
824
825                 err = err < 0 ? 0 : err;
826                 goto out;
827         }
828         case SHM_STAT:
829         case IPC_STAT:
830         {
831                 struct shmid64_ds tbuf;
832                 int result;
833
834                 if (cmd == SHM_STAT) {
835                         shp = shm_lock(ns, shmid);
836                         if (IS_ERR(shp)) {
837                                 err = PTR_ERR(shp);
838                                 goto out;
839                         }
840                         result = shp->shm_perm.id;
841                 } else {
842                         shp = shm_lock_check(ns, shmid);
843                         if (IS_ERR(shp)) {
844                                 err = PTR_ERR(shp);
845                                 goto out;
846                         }
847                         result = 0;
848                 }
849                 err = -EACCES;
850                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
851                         goto out_unlock;
852                 err = security_shm_shmctl(shp, cmd);
853                 if (err)
854                         goto out_unlock;
855                 memset(&tbuf, 0, sizeof(tbuf));
856                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
857                 tbuf.shm_segsz  = shp->shm_segsz;
858                 tbuf.shm_atime  = shp->shm_atim;
859                 tbuf.shm_dtime  = shp->shm_dtim;
860                 tbuf.shm_ctime  = shp->shm_ctim;
861                 tbuf.shm_cpid   = shp->shm_cprid;
862                 tbuf.shm_lpid   = shp->shm_lprid;
863                 tbuf.shm_nattch = shp->shm_nattch;
864                 shm_unlock(shp);
865                 if(copy_shmid_to_user (buf, &tbuf, version))
866                         err = -EFAULT;
867                 else
868                         err = result;
869                 goto out;
870         }
871         case SHM_LOCK:
872         case SHM_UNLOCK:
873         {
874                 struct file *shm_file;
875
876                 shp = shm_lock_check(ns, shmid);
877                 if (IS_ERR(shp)) {
878                         err = PTR_ERR(shp);
879                         goto out;
880                 }
881
882                 audit_ipc_obj(&(shp->shm_perm));
883
884                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
885                         uid_t euid = current_euid();
886                         err = -EPERM;
887                         if (euid != shp->shm_perm.uid &&
888                             euid != shp->shm_perm.cuid)
889                                 goto out_unlock;
890                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
891                                 goto out_unlock;
892                 }
893
894                 err = security_shm_shmctl(shp, cmd);
895                 if (err)
896                         goto out_unlock;
897
898                 shm_file = shp->shm_file;
899                 if (is_file_hugepages(shm_file))
900                         goto out_unlock;
901
902                 if (cmd == SHM_LOCK) {
903                         struct user_struct *user = current_user();
904                         err = shmem_lock(shm_file, 1, user);
905                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
906                                 shp->shm_perm.mode |= SHM_LOCKED;
907                                 shp->mlock_user = user;
908                         }
909                         goto out_unlock;
910                 }
911
912                 /* SHM_UNLOCK */
913                 if (!(shp->shm_perm.mode & SHM_LOCKED))
914                         goto out_unlock;
915                 shmem_lock(shm_file, 0, shp->mlock_user);
916                 shp->shm_perm.mode &= ~SHM_LOCKED;
917                 shp->mlock_user = NULL;
918                 get_file(shm_file);
919                 shm_unlock(shp);
920                 shmem_unlock_mapping(shm_file->f_mapping);
921                 fput(shm_file);
922                 goto out;
923         }
924         case IPC_RMID:
925         case IPC_SET:
926                 err = shmctl_down(ns, shmid, cmd, buf, version);
927                 return err;
928         default:
929                 return -EINVAL;
930         }
931
932 out_unlock:
933         shm_unlock(shp);
934 out:
935         return err;
936 }
937
938 /*
939  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
940  *
941  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
942  * "raddr" thing points to kernel space, and there has to be a wrapper around
943  * this.
944  */
945 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
946 {
947         struct shmid_kernel *shp;
948         unsigned long addr;
949         unsigned long size;
950         struct file * file;
951         int    err;
952         unsigned long flags;
953         unsigned long prot;
954         int acc_mode;
955         unsigned long user_addr;
956         struct ipc_namespace *ns;
957         struct shm_file_data *sfd;
958         struct path path;
959         fmode_t f_mode;
960
961         err = -EINVAL;
962         if (shmid < 0)
963                 goto out;
964         else if ((addr = (ulong)shmaddr)) {
965                 if (addr & (SHMLBA-1)) {
966                         /*
967                          * Round down to the nearest multiple of shmlba.
968                          * For sane do_mmap_pgoff() parameters, avoid
969                          * round downs that trigger nil-page and MAP_FIXED.
970                          */
971                         if ((shmflg & SHM_RND) && addr >= SHMLBA)
972                                 addr &= ~(SHMLBA - 1);
973                         else
974 #ifndef __ARCH_FORCE_SHMLBA
975                                 if (addr & ~PAGE_MASK)
976 #endif
977                                         goto out;
978                 }
979                 flags = MAP_SHARED | MAP_FIXED;
980         } else {
981                 if ((shmflg & SHM_REMAP))
982                         goto out;
983
984                 flags = MAP_SHARED;
985         }
986
987         if (shmflg & SHM_RDONLY) {
988                 prot = PROT_READ;
989                 acc_mode = S_IRUGO;
990                 f_mode = FMODE_READ;
991         } else {
992                 prot = PROT_READ | PROT_WRITE;
993                 acc_mode = S_IRUGO | S_IWUGO;
994                 f_mode = FMODE_READ | FMODE_WRITE;
995         }
996         if (shmflg & SHM_EXEC) {
997                 prot |= PROT_EXEC;
998                 acc_mode |= S_IXUGO;
999         }
1000
1001         /*
1002          * We cannot rely on the fs check since SYSV IPC does have an
1003          * additional creator id...
1004          */
1005         ns = current->nsproxy->ipc_ns;
1006         shp = shm_lock_check(ns, shmid);
1007         if (IS_ERR(shp)) {
1008                 err = PTR_ERR(shp);
1009                 goto out;
1010         }
1011
1012         err = -EACCES;
1013         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1014                 goto out_unlock;
1015
1016         err = security_shm_shmat(shp, shmaddr, shmflg);
1017         if (err)
1018                 goto out_unlock;
1019
1020         path = shp->shm_file->f_path;
1021         path_get(&path);
1022         shp->shm_nattch++;
1023         size = i_size_read(path.dentry->d_inode);
1024         shm_unlock(shp);
1025
1026         err = -ENOMEM;
1027         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1028         if (!sfd)
1029                 goto out_put_dentry;
1030
1031         file = alloc_file(&path, f_mode,
1032                           is_file_hugepages(shp->shm_file) ?
1033                                 &shm_file_operations_huge :
1034                                 &shm_file_operations);
1035         if (!file)
1036                 goto out_free;
1037
1038         file->private_data = sfd;
1039         file->f_mapping = shp->shm_file->f_mapping;
1040         sfd->id = shp->shm_perm.id;
1041         sfd->ns = get_ipc_ns(ns);
1042         sfd->file = shp->shm_file;
1043         sfd->vm_ops = NULL;
1044
1045         down_write(&current->mm->mmap_sem);
1046         if (addr && !(shmflg & SHM_REMAP)) {
1047                 err = -EINVAL;
1048                 if (find_vma_intersection(current->mm, addr, addr + size))
1049                         goto invalid;
1050                 /*
1051                  * If shm segment goes below stack, make sure there is some
1052                  * space left for the stack to grow (at least 4 pages).
1053                  */
1054                 if (addr < current->mm->start_stack &&
1055                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1056                         goto invalid;
1057         }
1058                 
1059         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1060         *raddr = user_addr;
1061         err = 0;
1062         if (IS_ERR_VALUE(user_addr))
1063                 err = (long)user_addr;
1064 invalid:
1065         up_write(&current->mm->mmap_sem);
1066
1067         fput(file);
1068
1069 out_nattch:
1070         down_write(&shm_ids(ns).rw_mutex);
1071         shp = shm_lock(ns, shmid);
1072         BUG_ON(IS_ERR(shp));
1073         shp->shm_nattch--;
1074         if (shm_may_destroy(ns, shp))
1075                 shm_destroy(ns, shp);
1076         else
1077                 shm_unlock(shp);
1078         up_write(&shm_ids(ns).rw_mutex);
1079
1080 out:
1081         return err;
1082
1083 out_unlock:
1084         shm_unlock(shp);
1085         goto out;
1086
1087 out_free:
1088         kfree(sfd);
1089 out_put_dentry:
1090         path_put(&path);
1091         goto out_nattch;
1092 }
1093
1094 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1095 {
1096         unsigned long ret;
1097         long err;
1098
1099         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1100         if (err)
1101                 return err;
1102         force_successful_syscall_return();
1103         return (long)ret;
1104 }
1105
1106 /*
1107  * detach and kill segment if marked destroyed.
1108  * The work is done in shm_close.
1109  */
1110 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1111 {
1112         struct mm_struct *mm = current->mm;
1113         struct vm_area_struct *vma;
1114         unsigned long addr = (unsigned long)shmaddr;
1115         int retval = -EINVAL;
1116 #ifdef CONFIG_MMU
1117         loff_t size = 0;
1118         struct vm_area_struct *next;
1119 #endif
1120
1121         if (addr & ~PAGE_MASK)
1122                 return retval;
1123
1124         down_write(&mm->mmap_sem);
1125
1126         /*
1127          * This function tries to be smart and unmap shm segments that
1128          * were modified by partial mlock or munmap calls:
1129          * - It first determines the size of the shm segment that should be
1130          *   unmapped: It searches for a vma that is backed by shm and that
1131          *   started at address shmaddr. It records it's size and then unmaps
1132          *   it.
1133          * - Then it unmaps all shm vmas that started at shmaddr and that
1134          *   are within the initially determined size.
1135          * Errors from do_munmap are ignored: the function only fails if
1136          * it's called with invalid parameters or if it's called to unmap
1137          * a part of a vma. Both calls in this function are for full vmas,
1138          * the parameters are directly copied from the vma itself and always
1139          * valid - therefore do_munmap cannot fail. (famous last words?)
1140          */
1141         /*
1142          * If it had been mremap()'d, the starting address would not
1143          * match the usual checks anyway. So assume all vma's are
1144          * above the starting address given.
1145          */
1146         vma = find_vma(mm, addr);
1147
1148 #ifdef CONFIG_MMU
1149         while (vma) {
1150                 next = vma->vm_next;
1151
1152                 /*
1153                  * Check if the starting address would match, i.e. it's
1154                  * a fragment created by mprotect() and/or munmap(), or it
1155                  * otherwise it starts at this address with no hassles.
1156                  */
1157                 if ((vma->vm_ops == &shm_vm_ops) &&
1158                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1159
1160
1161                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1162                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1163                         /*
1164                          * We discovered the size of the shm segment, so
1165                          * break out of here and fall through to the next
1166                          * loop that uses the size information to stop
1167                          * searching for matching vma's.
1168                          */
1169                         retval = 0;
1170                         vma = next;
1171                         break;
1172                 }
1173                 vma = next;
1174         }
1175
1176         /*
1177          * We need look no further than the maximum address a fragment
1178          * could possibly have landed at. Also cast things to loff_t to
1179          * prevent overflows and make comparisons vs. equal-width types.
1180          */
1181         size = PAGE_ALIGN(size);
1182         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1183                 next = vma->vm_next;
1184
1185                 /* finding a matching vma now does not alter retval */
1186                 if ((vma->vm_ops == &shm_vm_ops) &&
1187                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1188
1189                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1190                 vma = next;
1191         }
1192
1193 #else /* CONFIG_MMU */
1194         /* under NOMMU conditions, the exact address to be destroyed must be
1195          * given */
1196         retval = -EINVAL;
1197         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1198                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1199                 retval = 0;
1200         }
1201
1202 #endif
1203
1204         up_write(&mm->mmap_sem);
1205         return retval;
1206 }
1207
1208 #ifdef CONFIG_PROC_FS
1209 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1210 {
1211         struct shmid_kernel *shp = it;
1212         unsigned long rss = 0, swp = 0;
1213
1214         shm_add_rss_swap(shp, &rss, &swp);
1215
1216 #if BITS_PER_LONG <= 32
1217 #define SIZE_SPEC "%10lu"
1218 #else
1219 #define SIZE_SPEC "%21lu"
1220 #endif
1221
1222         return seq_printf(s,
1223                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1224                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1225                           SIZE_SPEC " " SIZE_SPEC "\n",
1226                           shp->shm_perm.key,
1227                           shp->shm_perm.id,
1228                           shp->shm_perm.mode,
1229                           shp->shm_segsz,
1230                           shp->shm_cprid,
1231                           shp->shm_lprid,
1232                           shp->shm_nattch,
1233                           shp->shm_perm.uid,
1234                           shp->shm_perm.gid,
1235                           shp->shm_perm.cuid,
1236                           shp->shm_perm.cgid,
1237                           shp->shm_atim,
1238                           shp->shm_dtim,
1239                           shp->shm_ctim,
1240                           rss * PAGE_SIZE,
1241                           swp * PAGE_SIZE);
1242 }
1243 #endif