shm: optimize exit_shm()
[linux-2.6.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 void __init shm_init (void)
109 {
110         shm_init_ns(&init_ipc_ns);
111         ipc_init_proc_interface("sysvipc/shm",
112 #if BITS_PER_LONG <= 32
113                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
114 #else
115                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
116 #endif
117                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
118 }
119
120 /*
121  * shm_lock_(check_) routines are called in the paths where the rw_mutex
122  * is not necessarily held.
123  */
124 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
125 {
126         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
127
128         if (IS_ERR(ipcp))
129                 return (struct shmid_kernel *)ipcp;
130
131         return container_of(ipcp, struct shmid_kernel, shm_perm);
132 }
133
134 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
135 {
136         rcu_read_lock();
137         spin_lock(&ipcp->shm_perm.lock);
138 }
139
140 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
141                                                 int id)
142 {
143         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
144
145         if (IS_ERR(ipcp))
146                 return (struct shmid_kernel *)ipcp;
147
148         return container_of(ipcp, struct shmid_kernel, shm_perm);
149 }
150
151 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
152 {
153         ipc_rmid(&shm_ids(ns), &s->shm_perm);
154 }
155
156
157 /* This is called by fork, once for every shm attach. */
158 static void shm_open(struct vm_area_struct *vma)
159 {
160         struct file *file = vma->vm_file;
161         struct shm_file_data *sfd = shm_file_data(file);
162         struct shmid_kernel *shp;
163
164         shp = shm_lock(sfd->ns, sfd->id);
165         BUG_ON(IS_ERR(shp));
166         shp->shm_atim = get_seconds();
167         shp->shm_lprid = task_tgid_vnr(current);
168         shp->shm_nattch++;
169         shm_unlock(shp);
170 }
171
172 /*
173  * shm_destroy - free the struct shmid_kernel
174  *
175  * @ns: namespace
176  * @shp: struct to free
177  *
178  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
179  * but returns with shp unlocked and freed.
180  */
181 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
182 {
183         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
184         shm_rmid(ns, shp);
185         shm_unlock(shp);
186         if (!is_file_hugepages(shp->shm_file))
187                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
188         else if (shp->mlock_user)
189                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
190                                                 shp->mlock_user);
191         fput (shp->shm_file);
192         security_shm_free(shp);
193         ipc_rcu_putref(shp);
194 }
195
196 /*
197  * shm_may_destroy - identifies whether shm segment should be destroyed now
198  *
199  * Returns true if and only if there are no active users of the segment and
200  * one of the following is true:
201  *
202  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
203  *
204  * 2) sysctl kernel.shm_rmid_forced is set to 1.
205  */
206 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
207 {
208         return (shp->shm_nattch == 0) &&
209                (ns->shm_rmid_forced ||
210                 (shp->shm_perm.mode & SHM_DEST));
211 }
212
213 /*
214  * remove the attach descriptor vma.
215  * free memory for segment if it is marked destroyed.
216  * The descriptor has already been removed from the current->mm->mmap list
217  * and will later be kfree()d.
218  */
219 static void shm_close(struct vm_area_struct *vma)
220 {
221         struct file * file = vma->vm_file;
222         struct shm_file_data *sfd = shm_file_data(file);
223         struct shmid_kernel *shp;
224         struct ipc_namespace *ns = sfd->ns;
225
226         down_write(&shm_ids(ns).rw_mutex);
227         /* remove from the list of attaches of the shm segment */
228         shp = shm_lock(ns, sfd->id);
229         BUG_ON(IS_ERR(shp));
230         shp->shm_lprid = task_tgid_vnr(current);
231         shp->shm_dtim = get_seconds();
232         shp->shm_nattch--;
233         if (shm_may_destroy(ns, shp))
234                 shm_destroy(ns, shp);
235         else
236                 shm_unlock(shp);
237         up_write(&shm_ids(ns).rw_mutex);
238 }
239
240 /* Called with ns->shm_ids(ns).rw_mutex locked */
241 static int shm_try_destroy_current(int id, void *p, void *data)
242 {
243         struct ipc_namespace *ns = data;
244         struct kern_ipc_perm *ipcp = p;
245         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
246
247         if (shp->shm_creator != current)
248                 return 0;
249
250         /*
251          * Mark it as orphaned to destroy the segment when
252          * kernel.shm_rmid_forced is changed.
253          * It is noop if the following shm_may_destroy() returns true.
254          */
255         shp->shm_creator = NULL;
256
257         /*
258          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
259          * is not set, it shouldn't be deleted here.
260          */
261         if (!ns->shm_rmid_forced)
262                 return 0;
263
264         if (shm_may_destroy(ns, shp)) {
265                 shm_lock_by_ptr(shp);
266                 shm_destroy(ns, shp);
267         }
268         return 0;
269 }
270
271 /* Called with ns->shm_ids(ns).rw_mutex locked */
272 static int shm_try_destroy_orphaned(int id, void *p, void *data)
273 {
274         struct ipc_namespace *ns = data;
275         struct kern_ipc_perm *ipcp = p;
276         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
277
278         /*
279          * We want to destroy segments without users and with already
280          * exit'ed originating process.
281          *
282          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
283          */
284         if (shp->shm_creator != NULL)
285                 return 0;
286
287         if (shm_may_destroy(ns, shp)) {
288                 shm_lock_by_ptr(shp);
289                 shm_destroy(ns, shp);
290         }
291         return 0;
292 }
293
294 void shm_destroy_orphaned(struct ipc_namespace *ns)
295 {
296         down_write(&shm_ids(ns).rw_mutex);
297         if (shm_ids(ns).in_use)
298                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
299         up_write(&shm_ids(ns).rw_mutex);
300 }
301
302
303 void exit_shm(struct task_struct *task)
304 {
305         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
306
307         if (shm_ids(ns).in_use == 0)
308                 return;
309
310         /* Destroy all already created segments, but not mapped yet */
311         down_write(&shm_ids(ns).rw_mutex);
312         if (shm_ids(ns).in_use)
313                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
314         up_write(&shm_ids(ns).rw_mutex);
315 }
316
317 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
318 {
319         struct file *file = vma->vm_file;
320         struct shm_file_data *sfd = shm_file_data(file);
321
322         return sfd->vm_ops->fault(vma, vmf);
323 }
324
325 #ifdef CONFIG_NUMA
326 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
327 {
328         struct file *file = vma->vm_file;
329         struct shm_file_data *sfd = shm_file_data(file);
330         int err = 0;
331         if (sfd->vm_ops->set_policy)
332                 err = sfd->vm_ops->set_policy(vma, new);
333         return err;
334 }
335
336 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
337                                         unsigned long addr)
338 {
339         struct file *file = vma->vm_file;
340         struct shm_file_data *sfd = shm_file_data(file);
341         struct mempolicy *pol = NULL;
342
343         if (sfd->vm_ops->get_policy)
344                 pol = sfd->vm_ops->get_policy(vma, addr);
345         else if (vma->vm_policy)
346                 pol = vma->vm_policy;
347
348         return pol;
349 }
350 #endif
351
352 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
353 {
354         struct shm_file_data *sfd = shm_file_data(file);
355         int ret;
356
357         ret = sfd->file->f_op->mmap(sfd->file, vma);
358         if (ret != 0)
359                 return ret;
360         sfd->vm_ops = vma->vm_ops;
361 #ifdef CONFIG_MMU
362         BUG_ON(!sfd->vm_ops->fault);
363 #endif
364         vma->vm_ops = &shm_vm_ops;
365         shm_open(vma);
366
367         return ret;
368 }
369
370 static int shm_release(struct inode *ino, struct file *file)
371 {
372         struct shm_file_data *sfd = shm_file_data(file);
373
374         put_ipc_ns(sfd->ns);
375         shm_file_data(file) = NULL;
376         kfree(sfd);
377         return 0;
378 }
379
380 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
381 {
382         struct shm_file_data *sfd = shm_file_data(file);
383
384         if (!sfd->file->f_op->fsync)
385                 return -EINVAL;
386         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
387 }
388
389 static unsigned long shm_get_unmapped_area(struct file *file,
390         unsigned long addr, unsigned long len, unsigned long pgoff,
391         unsigned long flags)
392 {
393         struct shm_file_data *sfd = shm_file_data(file);
394         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
395                                                 pgoff, flags);
396 }
397
398 static const struct file_operations shm_file_operations = {
399         .mmap           = shm_mmap,
400         .fsync          = shm_fsync,
401         .release        = shm_release,
402 #ifndef CONFIG_MMU
403         .get_unmapped_area      = shm_get_unmapped_area,
404 #endif
405         .llseek         = noop_llseek,
406 };
407
408 static const struct file_operations shm_file_operations_huge = {
409         .mmap           = shm_mmap,
410         .fsync          = shm_fsync,
411         .release        = shm_release,
412         .get_unmapped_area      = shm_get_unmapped_area,
413         .llseek         = noop_llseek,
414 };
415
416 int is_file_shm_hugepages(struct file *file)
417 {
418         return file->f_op == &shm_file_operations_huge;
419 }
420
421 static const struct vm_operations_struct shm_vm_ops = {
422         .open   = shm_open,     /* callback for a new vm-area open */
423         .close  = shm_close,    /* callback for when the vm-area is released */
424         .fault  = shm_fault,
425 #if defined(CONFIG_NUMA)
426         .set_policy = shm_set_policy,
427         .get_policy = shm_get_policy,
428 #endif
429 };
430
431 /**
432  * newseg - Create a new shared memory segment
433  * @ns: namespace
434  * @params: ptr to the structure that contains key, size and shmflg
435  *
436  * Called with shm_ids.rw_mutex held as a writer.
437  */
438
439 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
440 {
441         key_t key = params->key;
442         int shmflg = params->flg;
443         size_t size = params->u.size;
444         int error;
445         struct shmid_kernel *shp;
446         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
447         struct file * file;
448         char name[13];
449         int id;
450         vm_flags_t acctflag = 0;
451
452         if (size < SHMMIN || size > ns->shm_ctlmax)
453                 return -EINVAL;
454
455         if (ns->shm_tot + numpages > ns->shm_ctlall)
456                 return -ENOSPC;
457
458         shp = ipc_rcu_alloc(sizeof(*shp));
459         if (!shp)
460                 return -ENOMEM;
461
462         shp->shm_perm.key = key;
463         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
464         shp->mlock_user = NULL;
465
466         shp->shm_perm.security = NULL;
467         error = security_shm_alloc(shp);
468         if (error) {
469                 ipc_rcu_putref(shp);
470                 return error;
471         }
472
473         sprintf (name, "SYSV%08x", key);
474         if (shmflg & SHM_HUGETLB) {
475                 /* hugetlb_file_setup applies strict accounting */
476                 if (shmflg & SHM_NORESERVE)
477                         acctflag = VM_NORESERVE;
478                 file = hugetlb_file_setup(name, size, acctflag,
479                                         &shp->mlock_user, HUGETLB_SHMFS_INODE);
480         } else {
481                 /*
482                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
483                  * if it's asked for.
484                  */
485                 if  ((shmflg & SHM_NORESERVE) &&
486                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
487                         acctflag = VM_NORESERVE;
488                 file = shmem_file_setup(name, size, acctflag);
489         }
490         error = PTR_ERR(file);
491         if (IS_ERR(file))
492                 goto no_file;
493
494         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
495         if (id < 0) {
496                 error = id;
497                 goto no_id;
498         }
499
500         shp->shm_cprid = task_tgid_vnr(current);
501         shp->shm_lprid = 0;
502         shp->shm_atim = shp->shm_dtim = 0;
503         shp->shm_ctim = get_seconds();
504         shp->shm_segsz = size;
505         shp->shm_nattch = 0;
506         shp->shm_file = file;
507         shp->shm_creator = current;
508         /*
509          * shmid gets reported as "inode#" in /proc/pid/maps.
510          * proc-ps tools use this. Changing this will break them.
511          */
512         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
513
514         ns->shm_tot += numpages;
515         error = shp->shm_perm.id;
516         shm_unlock(shp);
517         return error;
518
519 no_id:
520         if (is_file_hugepages(file) && shp->mlock_user)
521                 user_shm_unlock(size, shp->mlock_user);
522         fput(file);
523 no_file:
524         security_shm_free(shp);
525         ipc_rcu_putref(shp);
526         return error;
527 }
528
529 /*
530  * Called with shm_ids.rw_mutex and ipcp locked.
531  */
532 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
533 {
534         struct shmid_kernel *shp;
535
536         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
537         return security_shm_associate(shp, shmflg);
538 }
539
540 /*
541  * Called with shm_ids.rw_mutex and ipcp locked.
542  */
543 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
544                                 struct ipc_params *params)
545 {
546         struct shmid_kernel *shp;
547
548         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
549         if (shp->shm_segsz < params->u.size)
550                 return -EINVAL;
551
552         return 0;
553 }
554
555 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
556 {
557         struct ipc_namespace *ns;
558         struct ipc_ops shm_ops;
559         struct ipc_params shm_params;
560
561         ns = current->nsproxy->ipc_ns;
562
563         shm_ops.getnew = newseg;
564         shm_ops.associate = shm_security;
565         shm_ops.more_checks = shm_more_checks;
566
567         shm_params.key = key;
568         shm_params.flg = shmflg;
569         shm_params.u.size = size;
570
571         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
572 }
573
574 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
575 {
576         switch(version) {
577         case IPC_64:
578                 return copy_to_user(buf, in, sizeof(*in));
579         case IPC_OLD:
580             {
581                 struct shmid_ds out;
582
583                 memset(&out, 0, sizeof(out));
584                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
585                 out.shm_segsz   = in->shm_segsz;
586                 out.shm_atime   = in->shm_atime;
587                 out.shm_dtime   = in->shm_dtime;
588                 out.shm_ctime   = in->shm_ctime;
589                 out.shm_cpid    = in->shm_cpid;
590                 out.shm_lpid    = in->shm_lpid;
591                 out.shm_nattch  = in->shm_nattch;
592
593                 return copy_to_user(buf, &out, sizeof(out));
594             }
595         default:
596                 return -EINVAL;
597         }
598 }
599
600 static inline unsigned long
601 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
602 {
603         switch(version) {
604         case IPC_64:
605                 if (copy_from_user(out, buf, sizeof(*out)))
606                         return -EFAULT;
607                 return 0;
608         case IPC_OLD:
609             {
610                 struct shmid_ds tbuf_old;
611
612                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
613                         return -EFAULT;
614
615                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
616                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
617                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
618
619                 return 0;
620             }
621         default:
622                 return -EINVAL;
623         }
624 }
625
626 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
627 {
628         switch(version) {
629         case IPC_64:
630                 return copy_to_user(buf, in, sizeof(*in));
631         case IPC_OLD:
632             {
633                 struct shminfo out;
634
635                 if(in->shmmax > INT_MAX)
636                         out.shmmax = INT_MAX;
637                 else
638                         out.shmmax = (int)in->shmmax;
639
640                 out.shmmin      = in->shmmin;
641                 out.shmmni      = in->shmmni;
642                 out.shmseg      = in->shmseg;
643                 out.shmall      = in->shmall; 
644
645                 return copy_to_user(buf, &out, sizeof(out));
646             }
647         default:
648                 return -EINVAL;
649         }
650 }
651
652 /*
653  * Calculate and add used RSS and swap pages of a shm.
654  * Called with shm_ids.rw_mutex held as a reader
655  */
656 static void shm_add_rss_swap(struct shmid_kernel *shp,
657         unsigned long *rss_add, unsigned long *swp_add)
658 {
659         struct inode *inode;
660
661         inode = shp->shm_file->f_path.dentry->d_inode;
662
663         if (is_file_hugepages(shp->shm_file)) {
664                 struct address_space *mapping = inode->i_mapping;
665                 struct hstate *h = hstate_file(shp->shm_file);
666                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
667         } else {
668 #ifdef CONFIG_SHMEM
669                 struct shmem_inode_info *info = SHMEM_I(inode);
670                 spin_lock(&info->lock);
671                 *rss_add += inode->i_mapping->nrpages;
672                 *swp_add += info->swapped;
673                 spin_unlock(&info->lock);
674 #else
675                 *rss_add += inode->i_mapping->nrpages;
676 #endif
677         }
678 }
679
680 /*
681  * Called with shm_ids.rw_mutex held as a reader
682  */
683 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
684                 unsigned long *swp)
685 {
686         int next_id;
687         int total, in_use;
688
689         *rss = 0;
690         *swp = 0;
691
692         in_use = shm_ids(ns).in_use;
693
694         for (total = 0, next_id = 0; total < in_use; next_id++) {
695                 struct kern_ipc_perm *ipc;
696                 struct shmid_kernel *shp;
697
698                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
699                 if (ipc == NULL)
700                         continue;
701                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
702
703                 shm_add_rss_swap(shp, rss, swp);
704
705                 total++;
706         }
707 }
708
709 /*
710  * This function handles some shmctl commands which require the rw_mutex
711  * to be held in write mode.
712  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
713  */
714 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
715                        struct shmid_ds __user *buf, int version)
716 {
717         struct kern_ipc_perm *ipcp;
718         struct shmid64_ds shmid64;
719         struct shmid_kernel *shp;
720         int err;
721
722         if (cmd == IPC_SET) {
723                 if (copy_shmid_from_user(&shmid64, buf, version))
724                         return -EFAULT;
725         }
726
727         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
728                                &shmid64.shm_perm, 0);
729         if (IS_ERR(ipcp))
730                 return PTR_ERR(ipcp);
731
732         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
733
734         err = security_shm_shmctl(shp, cmd);
735         if (err)
736                 goto out_unlock;
737         switch (cmd) {
738         case IPC_RMID:
739                 do_shm_rmid(ns, ipcp);
740                 goto out_up;
741         case IPC_SET:
742                 ipc_update_perm(&shmid64.shm_perm, ipcp);
743                 shp->shm_ctim = get_seconds();
744                 break;
745         default:
746                 err = -EINVAL;
747         }
748 out_unlock:
749         shm_unlock(shp);
750 out_up:
751         up_write(&shm_ids(ns).rw_mutex);
752         return err;
753 }
754
755 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
756 {
757         struct shmid_kernel *shp;
758         int err, version;
759         struct ipc_namespace *ns;
760
761         if (cmd < 0 || shmid < 0) {
762                 err = -EINVAL;
763                 goto out;
764         }
765
766         version = ipc_parse_version(&cmd);
767         ns = current->nsproxy->ipc_ns;
768
769         switch (cmd) { /* replace with proc interface ? */
770         case IPC_INFO:
771         {
772                 struct shminfo64 shminfo;
773
774                 err = security_shm_shmctl(NULL, cmd);
775                 if (err)
776                         return err;
777
778                 memset(&shminfo, 0, sizeof(shminfo));
779                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
780                 shminfo.shmmax = ns->shm_ctlmax;
781                 shminfo.shmall = ns->shm_ctlall;
782
783                 shminfo.shmmin = SHMMIN;
784                 if(copy_shminfo_to_user (buf, &shminfo, version))
785                         return -EFAULT;
786
787                 down_read(&shm_ids(ns).rw_mutex);
788                 err = ipc_get_maxid(&shm_ids(ns));
789                 up_read(&shm_ids(ns).rw_mutex);
790
791                 if(err<0)
792                         err = 0;
793                 goto out;
794         }
795         case SHM_INFO:
796         {
797                 struct shm_info shm_info;
798
799                 err = security_shm_shmctl(NULL, cmd);
800                 if (err)
801                         return err;
802
803                 memset(&shm_info, 0, sizeof(shm_info));
804                 down_read(&shm_ids(ns).rw_mutex);
805                 shm_info.used_ids = shm_ids(ns).in_use;
806                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
807                 shm_info.shm_tot = ns->shm_tot;
808                 shm_info.swap_attempts = 0;
809                 shm_info.swap_successes = 0;
810                 err = ipc_get_maxid(&shm_ids(ns));
811                 up_read(&shm_ids(ns).rw_mutex);
812                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
813                         err = -EFAULT;
814                         goto out;
815                 }
816
817                 err = err < 0 ? 0 : err;
818                 goto out;
819         }
820         case SHM_STAT:
821         case IPC_STAT:
822         {
823                 struct shmid64_ds tbuf;
824                 int result;
825
826                 if (cmd == SHM_STAT) {
827                         shp = shm_lock(ns, shmid);
828                         if (IS_ERR(shp)) {
829                                 err = PTR_ERR(shp);
830                                 goto out;
831                         }
832                         result = shp->shm_perm.id;
833                 } else {
834                         shp = shm_lock_check(ns, shmid);
835                         if (IS_ERR(shp)) {
836                                 err = PTR_ERR(shp);
837                                 goto out;
838                         }
839                         result = 0;
840                 }
841                 err = -EACCES;
842                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
843                         goto out_unlock;
844                 err = security_shm_shmctl(shp, cmd);
845                 if (err)
846                         goto out_unlock;
847                 memset(&tbuf, 0, sizeof(tbuf));
848                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
849                 tbuf.shm_segsz  = shp->shm_segsz;
850                 tbuf.shm_atime  = shp->shm_atim;
851                 tbuf.shm_dtime  = shp->shm_dtim;
852                 tbuf.shm_ctime  = shp->shm_ctim;
853                 tbuf.shm_cpid   = shp->shm_cprid;
854                 tbuf.shm_lpid   = shp->shm_lprid;
855                 tbuf.shm_nattch = shp->shm_nattch;
856                 shm_unlock(shp);
857                 if(copy_shmid_to_user (buf, &tbuf, version))
858                         err = -EFAULT;
859                 else
860                         err = result;
861                 goto out;
862         }
863         case SHM_LOCK:
864         case SHM_UNLOCK:
865         {
866                 struct file *uninitialized_var(shm_file);
867
868                 lru_add_drain_all();  /* drain pagevecs to lru lists */
869
870                 shp = shm_lock_check(ns, shmid);
871                 if (IS_ERR(shp)) {
872                         err = PTR_ERR(shp);
873                         goto out;
874                 }
875
876                 audit_ipc_obj(&(shp->shm_perm));
877
878                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
879                         uid_t euid = current_euid();
880                         err = -EPERM;
881                         if (euid != shp->shm_perm.uid &&
882                             euid != shp->shm_perm.cuid)
883                                 goto out_unlock;
884                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
885                                 goto out_unlock;
886                 }
887
888                 err = security_shm_shmctl(shp, cmd);
889                 if (err)
890                         goto out_unlock;
891                 
892                 if(cmd==SHM_LOCK) {
893                         struct user_struct *user = current_user();
894                         if (!is_file_hugepages(shp->shm_file)) {
895                                 err = shmem_lock(shp->shm_file, 1, user);
896                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
897                                         shp->shm_perm.mode |= SHM_LOCKED;
898                                         shp->mlock_user = user;
899                                 }
900                         }
901                 } else if (!is_file_hugepages(shp->shm_file)) {
902                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
903                         shp->shm_perm.mode &= ~SHM_LOCKED;
904                         shp->mlock_user = NULL;
905                 }
906                 shm_unlock(shp);
907                 goto out;
908         }
909         case IPC_RMID:
910         case IPC_SET:
911                 err = shmctl_down(ns, shmid, cmd, buf, version);
912                 return err;
913         default:
914                 return -EINVAL;
915         }
916
917 out_unlock:
918         shm_unlock(shp);
919 out:
920         return err;
921 }
922
923 /*
924  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
925  *
926  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
927  * "raddr" thing points to kernel space, and there has to be a wrapper around
928  * this.
929  */
930 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
931 {
932         struct shmid_kernel *shp;
933         unsigned long addr;
934         unsigned long size;
935         struct file * file;
936         int    err;
937         unsigned long flags;
938         unsigned long prot;
939         int acc_mode;
940         unsigned long user_addr;
941         struct ipc_namespace *ns;
942         struct shm_file_data *sfd;
943         struct path path;
944         fmode_t f_mode;
945
946         err = -EINVAL;
947         if (shmid < 0)
948                 goto out;
949         else if ((addr = (ulong)shmaddr)) {
950                 if (addr & (SHMLBA-1)) {
951                         if (shmflg & SHM_RND)
952                                 addr &= ~(SHMLBA-1);       /* round down */
953                         else
954 #ifndef __ARCH_FORCE_SHMLBA
955                                 if (addr & ~PAGE_MASK)
956 #endif
957                                         goto out;
958                 }
959                 flags = MAP_SHARED | MAP_FIXED;
960         } else {
961                 if ((shmflg & SHM_REMAP))
962                         goto out;
963
964                 flags = MAP_SHARED;
965         }
966
967         if (shmflg & SHM_RDONLY) {
968                 prot = PROT_READ;
969                 acc_mode = S_IRUGO;
970                 f_mode = FMODE_READ;
971         } else {
972                 prot = PROT_READ | PROT_WRITE;
973                 acc_mode = S_IRUGO | S_IWUGO;
974                 f_mode = FMODE_READ | FMODE_WRITE;
975         }
976         if (shmflg & SHM_EXEC) {
977                 prot |= PROT_EXEC;
978                 acc_mode |= S_IXUGO;
979         }
980
981         /*
982          * We cannot rely on the fs check since SYSV IPC does have an
983          * additional creator id...
984          */
985         ns = current->nsproxy->ipc_ns;
986         shp = shm_lock_check(ns, shmid);
987         if (IS_ERR(shp)) {
988                 err = PTR_ERR(shp);
989                 goto out;
990         }
991
992         err = -EACCES;
993         if (ipcperms(ns, &shp->shm_perm, acc_mode))
994                 goto out_unlock;
995
996         err = security_shm_shmat(shp, shmaddr, shmflg);
997         if (err)
998                 goto out_unlock;
999
1000         path = shp->shm_file->f_path;
1001         path_get(&path);
1002         shp->shm_nattch++;
1003         size = i_size_read(path.dentry->d_inode);
1004         shm_unlock(shp);
1005
1006         err = -ENOMEM;
1007         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1008         if (!sfd)
1009                 goto out_put_dentry;
1010
1011         file = alloc_file(&path, f_mode,
1012                           is_file_hugepages(shp->shm_file) ?
1013                                 &shm_file_operations_huge :
1014                                 &shm_file_operations);
1015         if (!file)
1016                 goto out_free;
1017
1018         file->private_data = sfd;
1019         file->f_mapping = shp->shm_file->f_mapping;
1020         sfd->id = shp->shm_perm.id;
1021         sfd->ns = get_ipc_ns(ns);
1022         sfd->file = shp->shm_file;
1023         sfd->vm_ops = NULL;
1024
1025         down_write(&current->mm->mmap_sem);
1026         if (addr && !(shmflg & SHM_REMAP)) {
1027                 err = -EINVAL;
1028                 if (find_vma_intersection(current->mm, addr, addr + size))
1029                         goto invalid;
1030                 /*
1031                  * If shm segment goes below stack, make sure there is some
1032                  * space left for the stack to grow (at least 4 pages).
1033                  */
1034                 if (addr < current->mm->start_stack &&
1035                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1036                         goto invalid;
1037         }
1038                 
1039         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1040         *raddr = user_addr;
1041         err = 0;
1042         if (IS_ERR_VALUE(user_addr))
1043                 err = (long)user_addr;
1044 invalid:
1045         up_write(&current->mm->mmap_sem);
1046
1047         fput(file);
1048
1049 out_nattch:
1050         down_write(&shm_ids(ns).rw_mutex);
1051         shp = shm_lock(ns, shmid);
1052         BUG_ON(IS_ERR(shp));
1053         shp->shm_nattch--;
1054         if (shm_may_destroy(ns, shp))
1055                 shm_destroy(ns, shp);
1056         else
1057                 shm_unlock(shp);
1058         up_write(&shm_ids(ns).rw_mutex);
1059
1060 out:
1061         return err;
1062
1063 out_unlock:
1064         shm_unlock(shp);
1065         goto out;
1066
1067 out_free:
1068         kfree(sfd);
1069 out_put_dentry:
1070         path_put(&path);
1071         goto out_nattch;
1072 }
1073
1074 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1075 {
1076         unsigned long ret;
1077         long err;
1078
1079         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1080         if (err)
1081                 return err;
1082         force_successful_syscall_return();
1083         return (long)ret;
1084 }
1085
1086 /*
1087  * detach and kill segment if marked destroyed.
1088  * The work is done in shm_close.
1089  */
1090 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1091 {
1092         struct mm_struct *mm = current->mm;
1093         struct vm_area_struct *vma;
1094         unsigned long addr = (unsigned long)shmaddr;
1095         int retval = -EINVAL;
1096 #ifdef CONFIG_MMU
1097         loff_t size = 0;
1098         struct vm_area_struct *next;
1099 #endif
1100
1101         if (addr & ~PAGE_MASK)
1102                 return retval;
1103
1104         down_write(&mm->mmap_sem);
1105
1106         /*
1107          * This function tries to be smart and unmap shm segments that
1108          * were modified by partial mlock or munmap calls:
1109          * - It first determines the size of the shm segment that should be
1110          *   unmapped: It searches for a vma that is backed by shm and that
1111          *   started at address shmaddr. It records it's size and then unmaps
1112          *   it.
1113          * - Then it unmaps all shm vmas that started at shmaddr and that
1114          *   are within the initially determined size.
1115          * Errors from do_munmap are ignored: the function only fails if
1116          * it's called with invalid parameters or if it's called to unmap
1117          * a part of a vma. Both calls in this function are for full vmas,
1118          * the parameters are directly copied from the vma itself and always
1119          * valid - therefore do_munmap cannot fail. (famous last words?)
1120          */
1121         /*
1122          * If it had been mremap()'d, the starting address would not
1123          * match the usual checks anyway. So assume all vma's are
1124          * above the starting address given.
1125          */
1126         vma = find_vma(mm, addr);
1127
1128 #ifdef CONFIG_MMU
1129         while (vma) {
1130                 next = vma->vm_next;
1131
1132                 /*
1133                  * Check if the starting address would match, i.e. it's
1134                  * a fragment created by mprotect() and/or munmap(), or it
1135                  * otherwise it starts at this address with no hassles.
1136                  */
1137                 if ((vma->vm_ops == &shm_vm_ops) &&
1138                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1139
1140
1141                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1142                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1143                         /*
1144                          * We discovered the size of the shm segment, so
1145                          * break out of here and fall through to the next
1146                          * loop that uses the size information to stop
1147                          * searching for matching vma's.
1148                          */
1149                         retval = 0;
1150                         vma = next;
1151                         break;
1152                 }
1153                 vma = next;
1154         }
1155
1156         /*
1157          * We need look no further than the maximum address a fragment
1158          * could possibly have landed at. Also cast things to loff_t to
1159          * prevent overflows and make comparisons vs. equal-width types.
1160          */
1161         size = PAGE_ALIGN(size);
1162         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1163                 next = vma->vm_next;
1164
1165                 /* finding a matching vma now does not alter retval */
1166                 if ((vma->vm_ops == &shm_vm_ops) &&
1167                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1168
1169                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1170                 vma = next;
1171         }
1172
1173 #else /* CONFIG_MMU */
1174         /* under NOMMU conditions, the exact address to be destroyed must be
1175          * given */
1176         retval = -EINVAL;
1177         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1178                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1179                 retval = 0;
1180         }
1181
1182 #endif
1183
1184         up_write(&mm->mmap_sem);
1185         return retval;
1186 }
1187
1188 #ifdef CONFIG_PROC_FS
1189 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1190 {
1191         struct shmid_kernel *shp = it;
1192         unsigned long rss = 0, swp = 0;
1193
1194         shm_add_rss_swap(shp, &rss, &swp);
1195
1196 #if BITS_PER_LONG <= 32
1197 #define SIZE_SPEC "%10lu"
1198 #else
1199 #define SIZE_SPEC "%21lu"
1200 #endif
1201
1202         return seq_printf(s,
1203                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1204                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1205                           SIZE_SPEC " " SIZE_SPEC "\n",
1206                           shp->shm_perm.key,
1207                           shp->shm_perm.id,
1208                           shp->shm_perm.mode,
1209                           shp->shm_segsz,
1210                           shp->shm_cprid,
1211                           shp->shm_lprid,
1212                           shp->shm_nattch,
1213                           shp->shm_perm.uid,
1214                           shp->shm_perm.gid,
1215                           shp->shm_perm.cuid,
1216                           shp->shm_perm.cgid,
1217                           shp->shm_atim,
1218                           shp->shm_dtim,
1219                           shp->shm_ctim,
1220                           rss * PAGE_SIZE,
1221                           swp * PAGE_SIZE);
1222 }
1223 #endif