hugetlb: modular state for hugetlb page size
[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 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_tot = 0;
78         ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
79 }
80
81 /*
82  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
83  * Only shm_ids.rw_mutex remains locked on exit.
84  */
85 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
86 {
87         struct shmid_kernel *shp;
88         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
89
90         if (shp->shm_nattch){
91                 shp->shm_perm.mode |= SHM_DEST;
92                 /* Do not find it any more */
93                 shp->shm_perm.key = IPC_PRIVATE;
94                 shm_unlock(shp);
95         } else
96                 shm_destroy(ns, shp);
97 }
98
99 #ifdef CONFIG_IPC_NS
100 void shm_exit_ns(struct ipc_namespace *ns)
101 {
102         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
103 }
104 #endif
105
106 void __init shm_init (void)
107 {
108         shm_init_ns(&init_ipc_ns);
109         ipc_init_proc_interface("sysvipc/shm",
110                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
111                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
112 }
113
114 /*
115  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
116  * is held to protect access to the idr tree.
117  */
118 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
119                                                 int id)
120 {
121         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
122
123         if (IS_ERR(ipcp))
124                 return (struct shmid_kernel *)ipcp;
125
126         return container_of(ipcp, struct shmid_kernel, shm_perm);
127 }
128
129 /*
130  * shm_lock_(check_) routines are called in the paths where the rw_mutex
131  * is not held.
132  */
133 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
134 {
135         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
136
137         if (IS_ERR(ipcp))
138                 return (struct shmid_kernel *)ipcp;
139
140         return container_of(ipcp, struct shmid_kernel, shm_perm);
141 }
142
143 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
144                                                 int id)
145 {
146         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
147
148         if (IS_ERR(ipcp))
149                 return (struct shmid_kernel *)ipcp;
150
151         return container_of(ipcp, struct shmid_kernel, shm_perm);
152 }
153
154 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
155 {
156         ipc_rmid(&shm_ids(ns), &s->shm_perm);
157 }
158
159
160 /* This is called by fork, once for every shm attach. */
161 static void shm_open(struct vm_area_struct *vma)
162 {
163         struct file *file = vma->vm_file;
164         struct shm_file_data *sfd = shm_file_data(file);
165         struct shmid_kernel *shp;
166
167         shp = shm_lock(sfd->ns, sfd->id);
168         BUG_ON(IS_ERR(shp));
169         shp->shm_atim = get_seconds();
170         shp->shm_lprid = task_tgid_vnr(current);
171         shp->shm_nattch++;
172         shm_unlock(shp);
173 }
174
175 /*
176  * shm_destroy - free the struct shmid_kernel
177  *
178  * @ns: namespace
179  * @shp: struct to free
180  *
181  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
182  * but returns with shp unlocked and freed.
183  */
184 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
185 {
186         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
187         shm_rmid(ns, shp);
188         shm_unlock(shp);
189         if (!is_file_hugepages(shp->shm_file))
190                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
191         else
192                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
193                                                 shp->mlock_user);
194         fput (shp->shm_file);
195         security_shm_free(shp);
196         ipc_rcu_putref(shp);
197 }
198
199 /*
200  * remove the attach descriptor vma.
201  * free memory for segment if it is marked destroyed.
202  * The descriptor has already been removed from the current->mm->mmap list
203  * and will later be kfree()d.
204  */
205 static void shm_close(struct vm_area_struct *vma)
206 {
207         struct file * file = vma->vm_file;
208         struct shm_file_data *sfd = shm_file_data(file);
209         struct shmid_kernel *shp;
210         struct ipc_namespace *ns = sfd->ns;
211
212         down_write(&shm_ids(ns).rw_mutex);
213         /* remove from the list of attaches of the shm segment */
214         shp = shm_lock_down(ns, sfd->id);
215         BUG_ON(IS_ERR(shp));
216         shp->shm_lprid = task_tgid_vnr(current);
217         shp->shm_dtim = get_seconds();
218         shp->shm_nattch--;
219         if(shp->shm_nattch == 0 &&
220            shp->shm_perm.mode & SHM_DEST)
221                 shm_destroy(ns, shp);
222         else
223                 shm_unlock(shp);
224         up_write(&shm_ids(ns).rw_mutex);
225 }
226
227 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
228 {
229         struct file *file = vma->vm_file;
230         struct shm_file_data *sfd = shm_file_data(file);
231
232         return sfd->vm_ops->fault(vma, vmf);
233 }
234
235 #ifdef CONFIG_NUMA
236 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
237 {
238         struct file *file = vma->vm_file;
239         struct shm_file_data *sfd = shm_file_data(file);
240         int err = 0;
241         if (sfd->vm_ops->set_policy)
242                 err = sfd->vm_ops->set_policy(vma, new);
243         return err;
244 }
245
246 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
247                                         unsigned long addr)
248 {
249         struct file *file = vma->vm_file;
250         struct shm_file_data *sfd = shm_file_data(file);
251         struct mempolicy *pol = NULL;
252
253         if (sfd->vm_ops->get_policy)
254                 pol = sfd->vm_ops->get_policy(vma, addr);
255         else if (vma->vm_policy)
256                 pol = vma->vm_policy;
257
258         return pol;
259 }
260 #endif
261
262 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
263 {
264         struct shm_file_data *sfd = shm_file_data(file);
265         int ret;
266
267         ret = sfd->file->f_op->mmap(sfd->file, vma);
268         if (ret != 0)
269                 return ret;
270         sfd->vm_ops = vma->vm_ops;
271 #ifdef CONFIG_MMU
272         BUG_ON(!sfd->vm_ops->fault);
273 #endif
274         vma->vm_ops = &shm_vm_ops;
275         shm_open(vma);
276
277         return ret;
278 }
279
280 static int shm_release(struct inode *ino, struct file *file)
281 {
282         struct shm_file_data *sfd = shm_file_data(file);
283
284         put_ipc_ns(sfd->ns);
285         shm_file_data(file) = NULL;
286         kfree(sfd);
287         return 0;
288 }
289
290 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
291 {
292         int (*fsync) (struct file *, struct dentry *, int datasync);
293         struct shm_file_data *sfd = shm_file_data(file);
294         int ret = -EINVAL;
295
296         fsync = sfd->file->f_op->fsync;
297         if (fsync)
298                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
299         return ret;
300 }
301
302 static unsigned long shm_get_unmapped_area(struct file *file,
303         unsigned long addr, unsigned long len, unsigned long pgoff,
304         unsigned long flags)
305 {
306         struct shm_file_data *sfd = shm_file_data(file);
307         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
308 }
309
310 int is_file_shm_hugepages(struct file *file)
311 {
312         int ret = 0;
313
314         if (file->f_op == &shm_file_operations) {
315                 struct shm_file_data *sfd;
316                 sfd = shm_file_data(file);
317                 ret = is_file_hugepages(sfd->file);
318         }
319         return ret;
320 }
321
322 static const struct file_operations shm_file_operations = {
323         .mmap           = shm_mmap,
324         .fsync          = shm_fsync,
325         .release        = shm_release,
326         .get_unmapped_area      = shm_get_unmapped_area,
327 };
328
329 static struct vm_operations_struct shm_vm_ops = {
330         .open   = shm_open,     /* callback for a new vm-area open */
331         .close  = shm_close,    /* callback for when the vm-area is released */
332         .fault  = shm_fault,
333 #if defined(CONFIG_NUMA)
334         .set_policy = shm_set_policy,
335         .get_policy = shm_get_policy,
336 #endif
337 };
338
339 /**
340  * newseg - Create a new shared memory segment
341  * @ns: namespace
342  * @params: ptr to the structure that contains key, size and shmflg
343  *
344  * Called with shm_ids.rw_mutex held as a writer.
345  */
346
347 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
348 {
349         key_t key = params->key;
350         int shmflg = params->flg;
351         size_t size = params->u.size;
352         int error;
353         struct shmid_kernel *shp;
354         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
355         struct file * file;
356         char name[13];
357         int id;
358
359         if (size < SHMMIN || size > ns->shm_ctlmax)
360                 return -EINVAL;
361
362         if (ns->shm_tot + numpages > ns->shm_ctlall)
363                 return -ENOSPC;
364
365         shp = ipc_rcu_alloc(sizeof(*shp));
366         if (!shp)
367                 return -ENOMEM;
368
369         shp->shm_perm.key = key;
370         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
371         shp->mlock_user = NULL;
372
373         shp->shm_perm.security = NULL;
374         error = security_shm_alloc(shp);
375         if (error) {
376                 ipc_rcu_putref(shp);
377                 return error;
378         }
379
380         sprintf (name, "SYSV%08x", key);
381         if (shmflg & SHM_HUGETLB) {
382                 /* hugetlb_file_setup takes care of mlock user accounting */
383                 file = hugetlb_file_setup(name, size);
384                 shp->mlock_user = current->user;
385         } else {
386                 int acctflag = VM_ACCOUNT;
387                 /*
388                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
389                  * if it's asked for.
390                  */
391                 if  ((shmflg & SHM_NORESERVE) &&
392                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
393                         acctflag = 0;
394                 file = shmem_file_setup(name, size, acctflag);
395         }
396         error = PTR_ERR(file);
397         if (IS_ERR(file))
398                 goto no_file;
399
400         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
401         if (id < 0) {
402                 error = id;
403                 goto no_id;
404         }
405
406         shp->shm_cprid = task_tgid_vnr(current);
407         shp->shm_lprid = 0;
408         shp->shm_atim = shp->shm_dtim = 0;
409         shp->shm_ctim = get_seconds();
410         shp->shm_segsz = size;
411         shp->shm_nattch = 0;
412         shp->shm_file = file;
413         /*
414          * shmid gets reported as "inode#" in /proc/pid/maps.
415          * proc-ps tools use this. Changing this will break them.
416          */
417         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
418
419         ns->shm_tot += numpages;
420         error = shp->shm_perm.id;
421         shm_unlock(shp);
422         return error;
423
424 no_id:
425         fput(file);
426 no_file:
427         security_shm_free(shp);
428         ipc_rcu_putref(shp);
429         return error;
430 }
431
432 /*
433  * Called with shm_ids.rw_mutex and ipcp locked.
434  */
435 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
436 {
437         struct shmid_kernel *shp;
438
439         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
440         return security_shm_associate(shp, shmflg);
441 }
442
443 /*
444  * Called with shm_ids.rw_mutex and ipcp locked.
445  */
446 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
447                                 struct ipc_params *params)
448 {
449         struct shmid_kernel *shp;
450
451         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
452         if (shp->shm_segsz < params->u.size)
453                 return -EINVAL;
454
455         return 0;
456 }
457
458 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
459 {
460         struct ipc_namespace *ns;
461         struct ipc_ops shm_ops;
462         struct ipc_params shm_params;
463
464         ns = current->nsproxy->ipc_ns;
465
466         shm_ops.getnew = newseg;
467         shm_ops.associate = shm_security;
468         shm_ops.more_checks = shm_more_checks;
469
470         shm_params.key = key;
471         shm_params.flg = shmflg;
472         shm_params.u.size = size;
473
474         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
475 }
476
477 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
478 {
479         switch(version) {
480         case IPC_64:
481                 return copy_to_user(buf, in, sizeof(*in));
482         case IPC_OLD:
483             {
484                 struct shmid_ds out;
485
486                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
487                 out.shm_segsz   = in->shm_segsz;
488                 out.shm_atime   = in->shm_atime;
489                 out.shm_dtime   = in->shm_dtime;
490                 out.shm_ctime   = in->shm_ctime;
491                 out.shm_cpid    = in->shm_cpid;
492                 out.shm_lpid    = in->shm_lpid;
493                 out.shm_nattch  = in->shm_nattch;
494
495                 return copy_to_user(buf, &out, sizeof(out));
496             }
497         default:
498                 return -EINVAL;
499         }
500 }
501
502 static inline unsigned long
503 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
504 {
505         switch(version) {
506         case IPC_64:
507                 if (copy_from_user(out, buf, sizeof(*out)))
508                         return -EFAULT;
509                 return 0;
510         case IPC_OLD:
511             {
512                 struct shmid_ds tbuf_old;
513
514                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
515                         return -EFAULT;
516
517                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
518                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
519                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
520
521                 return 0;
522             }
523         default:
524                 return -EINVAL;
525         }
526 }
527
528 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
529 {
530         switch(version) {
531         case IPC_64:
532                 return copy_to_user(buf, in, sizeof(*in));
533         case IPC_OLD:
534             {
535                 struct shminfo out;
536
537                 if(in->shmmax > INT_MAX)
538                         out.shmmax = INT_MAX;
539                 else
540                         out.shmmax = (int)in->shmmax;
541
542                 out.shmmin      = in->shmmin;
543                 out.shmmni      = in->shmmni;
544                 out.shmseg      = in->shmseg;
545                 out.shmall      = in->shmall; 
546
547                 return copy_to_user(buf, &out, sizeof(out));
548             }
549         default:
550                 return -EINVAL;
551         }
552 }
553
554 /*
555  * Called with shm_ids.rw_mutex held as a reader
556  */
557 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
558                 unsigned long *swp)
559 {
560         int next_id;
561         int total, in_use;
562
563         *rss = 0;
564         *swp = 0;
565
566         in_use = shm_ids(ns).in_use;
567
568         for (total = 0, next_id = 0; total < in_use; next_id++) {
569                 struct shmid_kernel *shp;
570                 struct inode *inode;
571
572                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
573                 if (shp == NULL)
574                         continue;
575
576                 inode = shp->shm_file->f_path.dentry->d_inode;
577
578                 if (is_file_hugepages(shp->shm_file)) {
579                         struct address_space *mapping = inode->i_mapping;
580                         struct hstate *h = hstate_file(shp->shm_file);
581                         *rss += pages_per_huge_page(h) * mapping->nrpages;
582                 } else {
583                         struct shmem_inode_info *info = SHMEM_I(inode);
584                         spin_lock(&info->lock);
585                         *rss += inode->i_mapping->nrpages;
586                         *swp += info->swapped;
587                         spin_unlock(&info->lock);
588                 }
589
590                 total++;
591         }
592 }
593
594 /*
595  * This function handles some shmctl commands which require the rw_mutex
596  * to be held in write mode.
597  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
598  */
599 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
600                        struct shmid_ds __user *buf, int version)
601 {
602         struct kern_ipc_perm *ipcp;
603         struct shmid64_ds shmid64;
604         struct shmid_kernel *shp;
605         int err;
606
607         if (cmd == IPC_SET) {
608                 if (copy_shmid_from_user(&shmid64, buf, version))
609                         return -EFAULT;
610         }
611
612         ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
613         if (IS_ERR(ipcp))
614                 return PTR_ERR(ipcp);
615
616         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
617
618         err = security_shm_shmctl(shp, cmd);
619         if (err)
620                 goto out_unlock;
621         switch (cmd) {
622         case IPC_RMID:
623                 do_shm_rmid(ns, ipcp);
624                 goto out_up;
625         case IPC_SET:
626                 ipc_update_perm(&shmid64.shm_perm, ipcp);
627                 shp->shm_ctim = get_seconds();
628                 break;
629         default:
630                 err = -EINVAL;
631         }
632 out_unlock:
633         shm_unlock(shp);
634 out_up:
635         up_write(&shm_ids(ns).rw_mutex);
636         return err;
637 }
638
639 asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
640 {
641         struct shmid_kernel *shp;
642         int err, version;
643         struct ipc_namespace *ns;
644
645         if (cmd < 0 || shmid < 0) {
646                 err = -EINVAL;
647                 goto out;
648         }
649
650         version = ipc_parse_version(&cmd);
651         ns = current->nsproxy->ipc_ns;
652
653         switch (cmd) { /* replace with proc interface ? */
654         case IPC_INFO:
655         {
656                 struct shminfo64 shminfo;
657
658                 err = security_shm_shmctl(NULL, cmd);
659                 if (err)
660                         return err;
661
662                 memset(&shminfo,0,sizeof(shminfo));
663                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
664                 shminfo.shmmax = ns->shm_ctlmax;
665                 shminfo.shmall = ns->shm_ctlall;
666
667                 shminfo.shmmin = SHMMIN;
668                 if(copy_shminfo_to_user (buf, &shminfo, version))
669                         return -EFAULT;
670
671                 down_read(&shm_ids(ns).rw_mutex);
672                 err = ipc_get_maxid(&shm_ids(ns));
673                 up_read(&shm_ids(ns).rw_mutex);
674
675                 if(err<0)
676                         err = 0;
677                 goto out;
678         }
679         case SHM_INFO:
680         {
681                 struct shm_info shm_info;
682
683                 err = security_shm_shmctl(NULL, cmd);
684                 if (err)
685                         return err;
686
687                 memset(&shm_info,0,sizeof(shm_info));
688                 down_read(&shm_ids(ns).rw_mutex);
689                 shm_info.used_ids = shm_ids(ns).in_use;
690                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
691                 shm_info.shm_tot = ns->shm_tot;
692                 shm_info.swap_attempts = 0;
693                 shm_info.swap_successes = 0;
694                 err = ipc_get_maxid(&shm_ids(ns));
695                 up_read(&shm_ids(ns).rw_mutex);
696                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
697                         err = -EFAULT;
698                         goto out;
699                 }
700
701                 err = err < 0 ? 0 : err;
702                 goto out;
703         }
704         case SHM_STAT:
705         case IPC_STAT:
706         {
707                 struct shmid64_ds tbuf;
708                 int result;
709
710                 if (!buf) {
711                         err = -EFAULT;
712                         goto out;
713                 }
714
715                 if (cmd == SHM_STAT) {
716                         shp = shm_lock(ns, shmid);
717                         if (IS_ERR(shp)) {
718                                 err = PTR_ERR(shp);
719                                 goto out;
720                         }
721                         result = shp->shm_perm.id;
722                 } else {
723                         shp = shm_lock_check(ns, shmid);
724                         if (IS_ERR(shp)) {
725                                 err = PTR_ERR(shp);
726                                 goto out;
727                         }
728                         result = 0;
729                 }
730                 err=-EACCES;
731                 if (ipcperms (&shp->shm_perm, S_IRUGO))
732                         goto out_unlock;
733                 err = security_shm_shmctl(shp, cmd);
734                 if (err)
735                         goto out_unlock;
736                 memset(&tbuf, 0, sizeof(tbuf));
737                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
738                 tbuf.shm_segsz  = shp->shm_segsz;
739                 tbuf.shm_atime  = shp->shm_atim;
740                 tbuf.shm_dtime  = shp->shm_dtim;
741                 tbuf.shm_ctime  = shp->shm_ctim;
742                 tbuf.shm_cpid   = shp->shm_cprid;
743                 tbuf.shm_lpid   = shp->shm_lprid;
744                 tbuf.shm_nattch = shp->shm_nattch;
745                 shm_unlock(shp);
746                 if(copy_shmid_to_user (buf, &tbuf, version))
747                         err = -EFAULT;
748                 else
749                         err = result;
750                 goto out;
751         }
752         case SHM_LOCK:
753         case SHM_UNLOCK:
754         {
755                 shp = shm_lock_check(ns, shmid);
756                 if (IS_ERR(shp)) {
757                         err = PTR_ERR(shp);
758                         goto out;
759                 }
760
761                 err = audit_ipc_obj(&(shp->shm_perm));
762                 if (err)
763                         goto out_unlock;
764
765                 if (!capable(CAP_IPC_LOCK)) {
766                         err = -EPERM;
767                         if (current->euid != shp->shm_perm.uid &&
768                             current->euid != shp->shm_perm.cuid)
769                                 goto out_unlock;
770                         if (cmd == SHM_LOCK &&
771                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
772                                 goto out_unlock;
773                 }
774
775                 err = security_shm_shmctl(shp, cmd);
776                 if (err)
777                         goto out_unlock;
778                 
779                 if(cmd==SHM_LOCK) {
780                         struct user_struct * user = current->user;
781                         if (!is_file_hugepages(shp->shm_file)) {
782                                 err = shmem_lock(shp->shm_file, 1, user);
783                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
784                                         shp->shm_perm.mode |= SHM_LOCKED;
785                                         shp->mlock_user = user;
786                                 }
787                         }
788                 } else if (!is_file_hugepages(shp->shm_file)) {
789                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
790                         shp->shm_perm.mode &= ~SHM_LOCKED;
791                         shp->mlock_user = NULL;
792                 }
793                 shm_unlock(shp);
794                 goto out;
795         }
796         case IPC_RMID:
797         case IPC_SET:
798                 err = shmctl_down(ns, shmid, cmd, buf, version);
799                 return err;
800         default:
801                 return -EINVAL;
802         }
803
804 out_unlock:
805         shm_unlock(shp);
806 out:
807         return err;
808 }
809
810 /*
811  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
812  *
813  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
814  * "raddr" thing points to kernel space, and there has to be a wrapper around
815  * this.
816  */
817 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
818 {
819         struct shmid_kernel *shp;
820         unsigned long addr;
821         unsigned long size;
822         struct file * file;
823         int    err;
824         unsigned long flags;
825         unsigned long prot;
826         int acc_mode;
827         unsigned long user_addr;
828         struct ipc_namespace *ns;
829         struct shm_file_data *sfd;
830         struct path path;
831         mode_t f_mode;
832
833         err = -EINVAL;
834         if (shmid < 0)
835                 goto out;
836         else if ((addr = (ulong)shmaddr)) {
837                 if (addr & (SHMLBA-1)) {
838                         if (shmflg & SHM_RND)
839                                 addr &= ~(SHMLBA-1);       /* round down */
840                         else
841 #ifndef __ARCH_FORCE_SHMLBA
842                                 if (addr & ~PAGE_MASK)
843 #endif
844                                         goto out;
845                 }
846                 flags = MAP_SHARED | MAP_FIXED;
847         } else {
848                 if ((shmflg & SHM_REMAP))
849                         goto out;
850
851                 flags = MAP_SHARED;
852         }
853
854         if (shmflg & SHM_RDONLY) {
855                 prot = PROT_READ;
856                 acc_mode = S_IRUGO;
857                 f_mode = FMODE_READ;
858         } else {
859                 prot = PROT_READ | PROT_WRITE;
860                 acc_mode = S_IRUGO | S_IWUGO;
861                 f_mode = FMODE_READ | FMODE_WRITE;
862         }
863         if (shmflg & SHM_EXEC) {
864                 prot |= PROT_EXEC;
865                 acc_mode |= S_IXUGO;
866         }
867
868         /*
869          * We cannot rely on the fs check since SYSV IPC does have an
870          * additional creator id...
871          */
872         ns = current->nsproxy->ipc_ns;
873         shp = shm_lock_check(ns, shmid);
874         if (IS_ERR(shp)) {
875                 err = PTR_ERR(shp);
876                 goto out;
877         }
878
879         err = -EACCES;
880         if (ipcperms(&shp->shm_perm, acc_mode))
881                 goto out_unlock;
882
883         err = security_shm_shmat(shp, shmaddr, shmflg);
884         if (err)
885                 goto out_unlock;
886
887         path.dentry = dget(shp->shm_file->f_path.dentry);
888         path.mnt    = shp->shm_file->f_path.mnt;
889         shp->shm_nattch++;
890         size = i_size_read(path.dentry->d_inode);
891         shm_unlock(shp);
892
893         err = -ENOMEM;
894         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
895         if (!sfd)
896                 goto out_put_dentry;
897
898         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
899         if (!file)
900                 goto out_free;
901
902         file->private_data = sfd;
903         file->f_mapping = shp->shm_file->f_mapping;
904         sfd->id = shp->shm_perm.id;
905         sfd->ns = get_ipc_ns(ns);
906         sfd->file = shp->shm_file;
907         sfd->vm_ops = NULL;
908
909         down_write(&current->mm->mmap_sem);
910         if (addr && !(shmflg & SHM_REMAP)) {
911                 err = -EINVAL;
912                 if (find_vma_intersection(current->mm, addr, addr + size))
913                         goto invalid;
914                 /*
915                  * If shm segment goes below stack, make sure there is some
916                  * space left for the stack to grow (at least 4 pages).
917                  */
918                 if (addr < current->mm->start_stack &&
919                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
920                         goto invalid;
921         }
922                 
923         user_addr = do_mmap (file, addr, size, prot, flags, 0);
924         *raddr = user_addr;
925         err = 0;
926         if (IS_ERR_VALUE(user_addr))
927                 err = (long)user_addr;
928 invalid:
929         up_write(&current->mm->mmap_sem);
930
931         fput(file);
932
933 out_nattch:
934         down_write(&shm_ids(ns).rw_mutex);
935         shp = shm_lock_down(ns, shmid);
936         BUG_ON(IS_ERR(shp));
937         shp->shm_nattch--;
938         if(shp->shm_nattch == 0 &&
939            shp->shm_perm.mode & SHM_DEST)
940                 shm_destroy(ns, shp);
941         else
942                 shm_unlock(shp);
943         up_write(&shm_ids(ns).rw_mutex);
944
945 out:
946         return err;
947
948 out_unlock:
949         shm_unlock(shp);
950         goto out;
951
952 out_free:
953         kfree(sfd);
954 out_put_dentry:
955         dput(path.dentry);
956         goto out_nattch;
957 }
958
959 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
960 {
961         unsigned long ret;
962         long err;
963
964         err = do_shmat(shmid, shmaddr, shmflg, &ret);
965         if (err)
966                 return err;
967         force_successful_syscall_return();
968         return (long)ret;
969 }
970
971 /*
972  * detach and kill segment if marked destroyed.
973  * The work is done in shm_close.
974  */
975 asmlinkage long sys_shmdt(char __user *shmaddr)
976 {
977         struct mm_struct *mm = current->mm;
978         struct vm_area_struct *vma, *next;
979         unsigned long addr = (unsigned long)shmaddr;
980         loff_t size = 0;
981         int retval = -EINVAL;
982
983         if (addr & ~PAGE_MASK)
984                 return retval;
985
986         down_write(&mm->mmap_sem);
987
988         /*
989          * This function tries to be smart and unmap shm segments that
990          * were modified by partial mlock or munmap calls:
991          * - It first determines the size of the shm segment that should be
992          *   unmapped: It searches for a vma that is backed by shm and that
993          *   started at address shmaddr. It records it's size and then unmaps
994          *   it.
995          * - Then it unmaps all shm vmas that started at shmaddr and that
996          *   are within the initially determined size.
997          * Errors from do_munmap are ignored: the function only fails if
998          * it's called with invalid parameters or if it's called to unmap
999          * a part of a vma. Both calls in this function are for full vmas,
1000          * the parameters are directly copied from the vma itself and always
1001          * valid - therefore do_munmap cannot fail. (famous last words?)
1002          */
1003         /*
1004          * If it had been mremap()'d, the starting address would not
1005          * match the usual checks anyway. So assume all vma's are
1006          * above the starting address given.
1007          */
1008         vma = find_vma(mm, addr);
1009
1010         while (vma) {
1011                 next = vma->vm_next;
1012
1013                 /*
1014                  * Check if the starting address would match, i.e. it's
1015                  * a fragment created by mprotect() and/or munmap(), or it
1016                  * otherwise it starts at this address with no hassles.
1017                  */
1018                 if ((vma->vm_ops == &shm_vm_ops) &&
1019                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1020
1021
1022                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1023                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1024                         /*
1025                          * We discovered the size of the shm segment, so
1026                          * break out of here and fall through to the next
1027                          * loop that uses the size information to stop
1028                          * searching for matching vma's.
1029                          */
1030                         retval = 0;
1031                         vma = next;
1032                         break;
1033                 }
1034                 vma = next;
1035         }
1036
1037         /*
1038          * We need look no further than the maximum address a fragment
1039          * could possibly have landed at. Also cast things to loff_t to
1040          * prevent overflows and make comparisions vs. equal-width types.
1041          */
1042         size = PAGE_ALIGN(size);
1043         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1044                 next = vma->vm_next;
1045
1046                 /* finding a matching vma now does not alter retval */
1047                 if ((vma->vm_ops == &shm_vm_ops) &&
1048                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1049
1050                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1051                 vma = next;
1052         }
1053
1054         up_write(&mm->mmap_sem);
1055         return retval;
1056 }
1057
1058 #ifdef CONFIG_PROC_FS
1059 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1060 {
1061         struct shmid_kernel *shp = it;
1062
1063 #if BITS_PER_LONG <= 32
1064 #define SIZE_SPEC "%10lu"
1065 #else
1066 #define SIZE_SPEC "%21lu"
1067 #endif
1068
1069         return seq_printf(s,
1070                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1071                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1072                           shp->shm_perm.key,
1073                           shp->shm_perm.id,
1074                           shp->shm_perm.mode,
1075                           shp->shm_segsz,
1076                           shp->shm_cprid,
1077                           shp->shm_lprid,
1078                           shp->shm_nattch,
1079                           shp->shm_perm.uid,
1080                           shp->shm_perm.gid,
1081                           shp->shm_perm.cuid,
1082                           shp->shm_perm.cgid,
1083                           shp->shm_atim,
1084                           shp->shm_dtim,
1085                           shp->shm_ctim);
1086 }
1087 #endif