mempolicy: fixup Fallback for Default Shmem Policy
[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 #define shm_buildid(id, seq)    ipc_buildid(id, seq)
64
65 static int newseg(struct ipc_namespace *, struct ipc_params *);
66 static void shm_open(struct vm_area_struct *vma);
67 static void shm_close(struct vm_area_struct *vma);
68 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
69 #ifdef CONFIG_PROC_FS
70 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
71 #endif
72
73 void shm_init_ns(struct ipc_namespace *ns)
74 {
75         ns->shm_ctlmax = SHMMAX;
76         ns->shm_ctlall = SHMALL;
77         ns->shm_ctlmni = SHMMNI;
78         ns->shm_tot = 0;
79         ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
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 }
105 #endif
106
107 void __init shm_init (void)
108 {
109         shm_init_ns(&init_ipc_ns);
110         ipc_init_proc_interface("sysvipc/shm",
111                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
112                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
113 }
114
115 /*
116  * shm_lock_(check_)down routines are called in the paths where the rw_mutex
117  * is held to protect access to the idr tree.
118  */
119 static inline struct shmid_kernel *shm_lock_down(struct ipc_namespace *ns,
120                                                 int id)
121 {
122         struct kern_ipc_perm *ipcp = ipc_lock_down(&shm_ids(ns), id);
123
124         if (IS_ERR(ipcp))
125                 return (struct shmid_kernel *)ipcp;
126
127         return container_of(ipcp, struct shmid_kernel, shm_perm);
128 }
129
130 static inline struct shmid_kernel *shm_lock_check_down(
131                                                 struct ipc_namespace *ns,
132                                                 int id)
133 {
134         struct kern_ipc_perm *ipcp = ipc_lock_check_down(&shm_ids(ns), id);
135
136         if (IS_ERR(ipcp))
137                 return (struct shmid_kernel *)ipcp;
138
139         return container_of(ipcp, struct shmid_kernel, shm_perm);
140 }
141
142 /*
143  * shm_lock_(check_) routines are called in the paths where the rw_mutex
144  * is not held.
145  */
146 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
147 {
148         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
149
150         if (IS_ERR(ipcp))
151                 return (struct shmid_kernel *)ipcp;
152
153         return container_of(ipcp, struct shmid_kernel, shm_perm);
154 }
155
156 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
157                                                 int id)
158 {
159         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
160
161         if (IS_ERR(ipcp))
162                 return (struct shmid_kernel *)ipcp;
163
164         return container_of(ipcp, struct shmid_kernel, shm_perm);
165 }
166
167 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
168 {
169         ipc_rmid(&shm_ids(ns), &s->shm_perm);
170 }
171
172 static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
173 {
174         return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
175 }
176
177
178
179 /* This is called by fork, once for every shm attach. */
180 static void shm_open(struct vm_area_struct *vma)
181 {
182         struct file *file = vma->vm_file;
183         struct shm_file_data *sfd = shm_file_data(file);
184         struct shmid_kernel *shp;
185
186         shp = shm_lock(sfd->ns, sfd->id);
187         BUG_ON(IS_ERR(shp));
188         shp->shm_atim = get_seconds();
189         shp->shm_lprid = task_tgid_vnr(current);
190         shp->shm_nattch++;
191         shm_unlock(shp);
192 }
193
194 /*
195  * shm_destroy - free the struct shmid_kernel
196  *
197  * @ns: namespace
198  * @shp: struct to free
199  *
200  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
201  * but returns with shp unlocked and freed.
202  */
203 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
204 {
205         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
206         shm_rmid(ns, shp);
207         shm_unlock(shp);
208         if (!is_file_hugepages(shp->shm_file))
209                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
210         else
211                 user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
212                                                 shp->mlock_user);
213         fput (shp->shm_file);
214         security_shm_free(shp);
215         ipc_rcu_putref(shp);
216 }
217
218 /*
219  * remove the attach descriptor vma.
220  * free memory for segment if it is marked destroyed.
221  * The descriptor has already been removed from the current->mm->mmap list
222  * and will later be kfree()d.
223  */
224 static void shm_close(struct vm_area_struct *vma)
225 {
226         struct file * file = vma->vm_file;
227         struct shm_file_data *sfd = shm_file_data(file);
228         struct shmid_kernel *shp;
229         struct ipc_namespace *ns = sfd->ns;
230
231         down_write(&shm_ids(ns).rw_mutex);
232         /* remove from the list of attaches of the shm segment */
233         shp = shm_lock_down(ns, sfd->id);
234         BUG_ON(IS_ERR(shp));
235         shp->shm_lprid = task_tgid_vnr(current);
236         shp->shm_dtim = get_seconds();
237         shp->shm_nattch--;
238         if(shp->shm_nattch == 0 &&
239            shp->shm_perm.mode & SHM_DEST)
240                 shm_destroy(ns, shp);
241         else
242                 shm_unlock(shp);
243         up_write(&shm_ids(ns).rw_mutex);
244 }
245
246 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
247 {
248         struct file *file = vma->vm_file;
249         struct shm_file_data *sfd = shm_file_data(file);
250
251         return sfd->vm_ops->fault(vma, vmf);
252 }
253
254 #ifdef CONFIG_NUMA
255 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
256 {
257         struct file *file = vma->vm_file;
258         struct shm_file_data *sfd = shm_file_data(file);
259         int err = 0;
260         if (sfd->vm_ops->set_policy)
261                 err = sfd->vm_ops->set_policy(vma, new);
262         return err;
263 }
264
265 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
266                                         unsigned long addr)
267 {
268         struct file *file = vma->vm_file;
269         struct shm_file_data *sfd = shm_file_data(file);
270         struct mempolicy *pol = NULL;
271
272         if (sfd->vm_ops->get_policy)
273                 pol = sfd->vm_ops->get_policy(vma, addr);
274         else if (vma->vm_policy) {
275                 pol = vma->vm_policy;
276                 mpol_get(pol);  /* get_vma_policy() expects this */
277         }
278         return pol;
279 }
280 #endif
281
282 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
283 {
284         struct shm_file_data *sfd = shm_file_data(file);
285         int ret;
286
287         ret = sfd->file->f_op->mmap(sfd->file, vma);
288         if (ret != 0)
289                 return ret;
290         sfd->vm_ops = vma->vm_ops;
291 #ifdef CONFIG_MMU
292         BUG_ON(!sfd->vm_ops->fault);
293 #endif
294         vma->vm_ops = &shm_vm_ops;
295         shm_open(vma);
296
297         return ret;
298 }
299
300 static int shm_release(struct inode *ino, struct file *file)
301 {
302         struct shm_file_data *sfd = shm_file_data(file);
303
304         put_ipc_ns(sfd->ns);
305         shm_file_data(file) = NULL;
306         kfree(sfd);
307         return 0;
308 }
309
310 static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
311 {
312         int (*fsync) (struct file *, struct dentry *, int datasync);
313         struct shm_file_data *sfd = shm_file_data(file);
314         int ret = -EINVAL;
315
316         fsync = sfd->file->f_op->fsync;
317         if (fsync)
318                 ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
319         return ret;
320 }
321
322 static unsigned long shm_get_unmapped_area(struct file *file,
323         unsigned long addr, unsigned long len, unsigned long pgoff,
324         unsigned long flags)
325 {
326         struct shm_file_data *sfd = shm_file_data(file);
327         return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
328 }
329
330 int is_file_shm_hugepages(struct file *file)
331 {
332         int ret = 0;
333
334         if (file->f_op == &shm_file_operations) {
335                 struct shm_file_data *sfd;
336                 sfd = shm_file_data(file);
337                 ret = is_file_hugepages(sfd->file);
338         }
339         return ret;
340 }
341
342 static const struct file_operations shm_file_operations = {
343         .mmap           = shm_mmap,
344         .fsync          = shm_fsync,
345         .release        = shm_release,
346         .get_unmapped_area      = shm_get_unmapped_area,
347 };
348
349 static struct vm_operations_struct shm_vm_ops = {
350         .open   = shm_open,     /* callback for a new vm-area open */
351         .close  = shm_close,    /* callback for when the vm-area is released */
352         .fault  = shm_fault,
353 #if defined(CONFIG_NUMA)
354         .set_policy = shm_set_policy,
355         .get_policy = shm_get_policy,
356 #endif
357 };
358
359 /**
360  * newseg - Create a new shared memory segment
361  * @ns: namespace
362  * @params: ptr to the structure that contains key, size and shmflg
363  *
364  * Called with shm_ids.rw_mutex held as a writer.
365  */
366
367 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
368 {
369         key_t key = params->key;
370         int shmflg = params->flg;
371         size_t size = params->u.size;
372         int error;
373         struct shmid_kernel *shp;
374         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
375         struct file * file;
376         char name[13];
377         int id;
378
379         if (size < SHMMIN || size > ns->shm_ctlmax)
380                 return -EINVAL;
381
382         if (ns->shm_tot + numpages > ns->shm_ctlall)
383                 return -ENOSPC;
384
385         shp = ipc_rcu_alloc(sizeof(*shp));
386         if (!shp)
387                 return -ENOMEM;
388
389         shp->shm_perm.key = key;
390         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
391         shp->mlock_user = NULL;
392
393         shp->shm_perm.security = NULL;
394         error = security_shm_alloc(shp);
395         if (error) {
396                 ipc_rcu_putref(shp);
397                 return error;
398         }
399
400         sprintf (name, "SYSV%08x", key);
401         if (shmflg & SHM_HUGETLB) {
402                 /* hugetlb_file_setup takes care of mlock user accounting */
403                 file = hugetlb_file_setup(name, size);
404                 shp->mlock_user = current->user;
405         } else {
406                 int acctflag = VM_ACCOUNT;
407                 /*
408                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
409                  * if it's asked for.
410                  */
411                 if  ((shmflg & SHM_NORESERVE) &&
412                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
413                         acctflag = 0;
414                 file = shmem_file_setup(name, size, acctflag);
415         }
416         error = PTR_ERR(file);
417         if (IS_ERR(file))
418                 goto no_file;
419
420         id = shm_addid(ns, shp);
421         if (id < 0) {
422                 error = id;
423                 goto no_id;
424         }
425
426         shp->shm_cprid = task_tgid_vnr(current);
427         shp->shm_lprid = 0;
428         shp->shm_atim = shp->shm_dtim = 0;
429         shp->shm_ctim = get_seconds();
430         shp->shm_segsz = size;
431         shp->shm_nattch = 0;
432         shp->shm_perm.id = shm_buildid(id, shp->shm_perm.seq);
433         shp->shm_file = file;
434         /*
435          * shmid gets reported as "inode#" in /proc/pid/maps.
436          * proc-ps tools use this. Changing this will break them.
437          */
438         file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
439
440         ns->shm_tot += numpages;
441         error = shp->shm_perm.id;
442         shm_unlock(shp);
443         return error;
444
445 no_id:
446         fput(file);
447 no_file:
448         security_shm_free(shp);
449         ipc_rcu_putref(shp);
450         return error;
451 }
452
453 /*
454  * Called with shm_ids.rw_mutex and ipcp locked.
455  */
456 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
457 {
458         struct shmid_kernel *shp;
459
460         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
461         return security_shm_associate(shp, shmflg);
462 }
463
464 /*
465  * Called with shm_ids.rw_mutex and ipcp locked.
466  */
467 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
468                                 struct ipc_params *params)
469 {
470         struct shmid_kernel *shp;
471
472         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
473         if (shp->shm_segsz < params->u.size)
474                 return -EINVAL;
475
476         return 0;
477 }
478
479 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
480 {
481         struct ipc_namespace *ns;
482         struct ipc_ops shm_ops;
483         struct ipc_params shm_params;
484
485         ns = current->nsproxy->ipc_ns;
486
487         shm_ops.getnew = newseg;
488         shm_ops.associate = shm_security;
489         shm_ops.more_checks = shm_more_checks;
490
491         shm_params.key = key;
492         shm_params.flg = shmflg;
493         shm_params.u.size = size;
494
495         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
496 }
497
498 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
499 {
500         switch(version) {
501         case IPC_64:
502                 return copy_to_user(buf, in, sizeof(*in));
503         case IPC_OLD:
504             {
505                 struct shmid_ds out;
506
507                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
508                 out.shm_segsz   = in->shm_segsz;
509                 out.shm_atime   = in->shm_atime;
510                 out.shm_dtime   = in->shm_dtime;
511                 out.shm_ctime   = in->shm_ctime;
512                 out.shm_cpid    = in->shm_cpid;
513                 out.shm_lpid    = in->shm_lpid;
514                 out.shm_nattch  = in->shm_nattch;
515
516                 return copy_to_user(buf, &out, sizeof(out));
517             }
518         default:
519                 return -EINVAL;
520         }
521 }
522
523 struct shm_setbuf {
524         uid_t   uid;
525         gid_t   gid;
526         mode_t  mode;
527 };      
528
529 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
530 {
531         switch(version) {
532         case IPC_64:
533             {
534                 struct shmid64_ds tbuf;
535
536                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
537                         return -EFAULT;
538
539                 out->uid        = tbuf.shm_perm.uid;
540                 out->gid        = tbuf.shm_perm.gid;
541                 out->mode       = tbuf.shm_perm.mode;
542
543                 return 0;
544             }
545         case IPC_OLD:
546             {
547                 struct shmid_ds tbuf_old;
548
549                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
550                         return -EFAULT;
551
552                 out->uid        = tbuf_old.shm_perm.uid;
553                 out->gid        = tbuf_old.shm_perm.gid;
554                 out->mode       = tbuf_old.shm_perm.mode;
555
556                 return 0;
557             }
558         default:
559                 return -EINVAL;
560         }
561 }
562
563 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
564 {
565         switch(version) {
566         case IPC_64:
567                 return copy_to_user(buf, in, sizeof(*in));
568         case IPC_OLD:
569             {
570                 struct shminfo out;
571
572                 if(in->shmmax > INT_MAX)
573                         out.shmmax = INT_MAX;
574                 else
575                         out.shmmax = (int)in->shmmax;
576
577                 out.shmmin      = in->shmmin;
578                 out.shmmni      = in->shmmni;
579                 out.shmseg      = in->shmseg;
580                 out.shmall      = in->shmall; 
581
582                 return copy_to_user(buf, &out, sizeof(out));
583             }
584         default:
585                 return -EINVAL;
586         }
587 }
588
589 /*
590  * Called with shm_ids.rw_mutex held as a reader
591  */
592 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
593                 unsigned long *swp)
594 {
595         int next_id;
596         int total, in_use;
597
598         *rss = 0;
599         *swp = 0;
600
601         in_use = shm_ids(ns).in_use;
602
603         for (total = 0, next_id = 0; total < in_use; next_id++) {
604                 struct shmid_kernel *shp;
605                 struct inode *inode;
606
607                 shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
608                 if (shp == NULL)
609                         continue;
610
611                 inode = shp->shm_file->f_path.dentry->d_inode;
612
613                 if (is_file_hugepages(shp->shm_file)) {
614                         struct address_space *mapping = inode->i_mapping;
615                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
616                 } else {
617                         struct shmem_inode_info *info = SHMEM_I(inode);
618                         spin_lock(&info->lock);
619                         *rss += inode->i_mapping->nrpages;
620                         *swp += info->swapped;
621                         spin_unlock(&info->lock);
622                 }
623
624                 total++;
625         }
626 }
627
628 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
629 {
630         struct shm_setbuf setbuf;
631         struct shmid_kernel *shp;
632         int err, version;
633         struct ipc_namespace *ns;
634
635         if (cmd < 0 || shmid < 0) {
636                 err = -EINVAL;
637                 goto out;
638         }
639
640         version = ipc_parse_version(&cmd);
641         ns = current->nsproxy->ipc_ns;
642
643         switch (cmd) { /* replace with proc interface ? */
644         case IPC_INFO:
645         {
646                 struct shminfo64 shminfo;
647
648                 err = security_shm_shmctl(NULL, cmd);
649                 if (err)
650                         return err;
651
652                 memset(&shminfo,0,sizeof(shminfo));
653                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
654                 shminfo.shmmax = ns->shm_ctlmax;
655                 shminfo.shmall = ns->shm_ctlall;
656
657                 shminfo.shmmin = SHMMIN;
658                 if(copy_shminfo_to_user (buf, &shminfo, version))
659                         return -EFAULT;
660
661                 down_read(&shm_ids(ns).rw_mutex);
662                 err = ipc_get_maxid(&shm_ids(ns));
663                 up_read(&shm_ids(ns).rw_mutex);
664
665                 if(err<0)
666                         err = 0;
667                 goto out;
668         }
669         case SHM_INFO:
670         {
671                 struct shm_info shm_info;
672
673                 err = security_shm_shmctl(NULL, cmd);
674                 if (err)
675                         return err;
676
677                 memset(&shm_info,0,sizeof(shm_info));
678                 down_read(&shm_ids(ns).rw_mutex);
679                 shm_info.used_ids = shm_ids(ns).in_use;
680                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
681                 shm_info.shm_tot = ns->shm_tot;
682                 shm_info.swap_attempts = 0;
683                 shm_info.swap_successes = 0;
684                 err = ipc_get_maxid(&shm_ids(ns));
685                 up_read(&shm_ids(ns).rw_mutex);
686                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
687                         err = -EFAULT;
688                         goto out;
689                 }
690
691                 err = err < 0 ? 0 : err;
692                 goto out;
693         }
694         case SHM_STAT:
695         case IPC_STAT:
696         {
697                 struct shmid64_ds tbuf;
698                 int result;
699
700                 if (!buf) {
701                         err = -EFAULT;
702                         goto out;
703                 }
704
705                 if (cmd == SHM_STAT) {
706                         shp = shm_lock(ns, shmid);
707                         if (IS_ERR(shp)) {
708                                 err = PTR_ERR(shp);
709                                 goto out;
710                         }
711                         result = shp->shm_perm.id;
712                 } else {
713                         shp = shm_lock_check(ns, shmid);
714                         if (IS_ERR(shp)) {
715                                 err = PTR_ERR(shp);
716                                 goto out;
717                         }
718                         result = 0;
719                 }
720                 err=-EACCES;
721                 if (ipcperms (&shp->shm_perm, S_IRUGO))
722                         goto out_unlock;
723                 err = security_shm_shmctl(shp, cmd);
724                 if (err)
725                         goto out_unlock;
726                 memset(&tbuf, 0, sizeof(tbuf));
727                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
728                 tbuf.shm_segsz  = shp->shm_segsz;
729                 tbuf.shm_atime  = shp->shm_atim;
730                 tbuf.shm_dtime  = shp->shm_dtim;
731                 tbuf.shm_ctime  = shp->shm_ctim;
732                 tbuf.shm_cpid   = shp->shm_cprid;
733                 tbuf.shm_lpid   = shp->shm_lprid;
734                 tbuf.shm_nattch = shp->shm_nattch;
735                 shm_unlock(shp);
736                 if(copy_shmid_to_user (buf, &tbuf, version))
737                         err = -EFAULT;
738                 else
739                         err = result;
740                 goto out;
741         }
742         case SHM_LOCK:
743         case SHM_UNLOCK:
744         {
745                 shp = shm_lock_check(ns, shmid);
746                 if (IS_ERR(shp)) {
747                         err = PTR_ERR(shp);
748                         goto out;
749                 }
750
751                 err = audit_ipc_obj(&(shp->shm_perm));
752                 if (err)
753                         goto out_unlock;
754
755                 if (!capable(CAP_IPC_LOCK)) {
756                         err = -EPERM;
757                         if (current->euid != shp->shm_perm.uid &&
758                             current->euid != shp->shm_perm.cuid)
759                                 goto out_unlock;
760                         if (cmd == SHM_LOCK &&
761                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
762                                 goto out_unlock;
763                 }
764
765                 err = security_shm_shmctl(shp, cmd);
766                 if (err)
767                         goto out_unlock;
768                 
769                 if(cmd==SHM_LOCK) {
770                         struct user_struct * user = current->user;
771                         if (!is_file_hugepages(shp->shm_file)) {
772                                 err = shmem_lock(shp->shm_file, 1, user);
773                                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
774                                         shp->shm_perm.mode |= SHM_LOCKED;
775                                         shp->mlock_user = user;
776                                 }
777                         }
778                 } else if (!is_file_hugepages(shp->shm_file)) {
779                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
780                         shp->shm_perm.mode &= ~SHM_LOCKED;
781                         shp->mlock_user = NULL;
782                 }
783                 shm_unlock(shp);
784                 goto out;
785         }
786         case IPC_RMID:
787         {
788                 /*
789                  *      We cannot simply remove the file. The SVID states
790                  *      that the block remains until the last person
791                  *      detaches from it, then is deleted. A shmat() on
792                  *      an RMID segment is legal in older Linux and if 
793                  *      we change it apps break...
794                  *
795                  *      Instead we set a destroyed flag, and then blow
796                  *      the name away when the usage hits zero.
797                  */
798                 down_write(&shm_ids(ns).rw_mutex);
799                 shp = shm_lock_check_down(ns, shmid);
800                 if (IS_ERR(shp)) {
801                         err = PTR_ERR(shp);
802                         goto out_up;
803                 }
804
805                 err = audit_ipc_obj(&(shp->shm_perm));
806                 if (err)
807                         goto out_unlock_up;
808
809                 if (current->euid != shp->shm_perm.uid &&
810                     current->euid != shp->shm_perm.cuid && 
811                     !capable(CAP_SYS_ADMIN)) {
812                         err=-EPERM;
813                         goto out_unlock_up;
814                 }
815
816                 err = security_shm_shmctl(shp, cmd);
817                 if (err)
818                         goto out_unlock_up;
819
820                 do_shm_rmid(ns, &shp->shm_perm);
821                 up_write(&shm_ids(ns).rw_mutex);
822                 goto out;
823         }
824
825         case IPC_SET:
826         {
827                 if (!buf) {
828                         err = -EFAULT;
829                         goto out;
830                 }
831
832                 if (copy_shmid_from_user (&setbuf, buf, version)) {
833                         err = -EFAULT;
834                         goto out;
835                 }
836                 down_write(&shm_ids(ns).rw_mutex);
837                 shp = shm_lock_check_down(ns, shmid);
838                 if (IS_ERR(shp)) {
839                         err = PTR_ERR(shp);
840                         goto out_up;
841                 }
842                 err = audit_ipc_obj(&(shp->shm_perm));
843                 if (err)
844                         goto out_unlock_up;
845                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
846                 if (err)
847                         goto out_unlock_up;
848                 err=-EPERM;
849                 if (current->euid != shp->shm_perm.uid &&
850                     current->euid != shp->shm_perm.cuid && 
851                     !capable(CAP_SYS_ADMIN)) {
852                         goto out_unlock_up;
853                 }
854
855                 err = security_shm_shmctl(shp, cmd);
856                 if (err)
857                         goto out_unlock_up;
858                 
859                 shp->shm_perm.uid = setbuf.uid;
860                 shp->shm_perm.gid = setbuf.gid;
861                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
862                         | (setbuf.mode & S_IRWXUGO);
863                 shp->shm_ctim = get_seconds();
864                 break;
865         }
866
867         default:
868                 err = -EINVAL;
869                 goto out;
870         }
871
872         err = 0;
873 out_unlock_up:
874         shm_unlock(shp);
875 out_up:
876         up_write(&shm_ids(ns).rw_mutex);
877         goto out;
878 out_unlock:
879         shm_unlock(shp);
880 out:
881         return err;
882 }
883
884 /*
885  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
886  *
887  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
888  * "raddr" thing points to kernel space, and there has to be a wrapper around
889  * this.
890  */
891 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
892 {
893         struct shmid_kernel *shp;
894         unsigned long addr;
895         unsigned long size;
896         struct file * file;
897         int    err;
898         unsigned long flags;
899         unsigned long prot;
900         int acc_mode;
901         unsigned long user_addr;
902         struct ipc_namespace *ns;
903         struct shm_file_data *sfd;
904         struct path path;
905         mode_t f_mode;
906
907         err = -EINVAL;
908         if (shmid < 0)
909                 goto out;
910         else if ((addr = (ulong)shmaddr)) {
911                 if (addr & (SHMLBA-1)) {
912                         if (shmflg & SHM_RND)
913                                 addr &= ~(SHMLBA-1);       /* round down */
914                         else
915 #ifndef __ARCH_FORCE_SHMLBA
916                                 if (addr & ~PAGE_MASK)
917 #endif
918                                         goto out;
919                 }
920                 flags = MAP_SHARED | MAP_FIXED;
921         } else {
922                 if ((shmflg & SHM_REMAP))
923                         goto out;
924
925                 flags = MAP_SHARED;
926         }
927
928         if (shmflg & SHM_RDONLY) {
929                 prot = PROT_READ;
930                 acc_mode = S_IRUGO;
931                 f_mode = FMODE_READ;
932         } else {
933                 prot = PROT_READ | PROT_WRITE;
934                 acc_mode = S_IRUGO | S_IWUGO;
935                 f_mode = FMODE_READ | FMODE_WRITE;
936         }
937         if (shmflg & SHM_EXEC) {
938                 prot |= PROT_EXEC;
939                 acc_mode |= S_IXUGO;
940         }
941
942         /*
943          * We cannot rely on the fs check since SYSV IPC does have an
944          * additional creator id...
945          */
946         ns = current->nsproxy->ipc_ns;
947         shp = shm_lock_check(ns, shmid);
948         if (IS_ERR(shp)) {
949                 err = PTR_ERR(shp);
950                 goto out;
951         }
952
953         err = -EACCES;
954         if (ipcperms(&shp->shm_perm, acc_mode))
955                 goto out_unlock;
956
957         err = security_shm_shmat(shp, shmaddr, shmflg);
958         if (err)
959                 goto out_unlock;
960
961         path.dentry = dget(shp->shm_file->f_path.dentry);
962         path.mnt    = shp->shm_file->f_path.mnt;
963         shp->shm_nattch++;
964         size = i_size_read(path.dentry->d_inode);
965         shm_unlock(shp);
966
967         err = -ENOMEM;
968         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
969         if (!sfd)
970                 goto out_put_dentry;
971
972         err = -ENOMEM;
973
974         file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
975         if (!file)
976                 goto out_free;
977
978         file->private_data = sfd;
979         file->f_mapping = shp->shm_file->f_mapping;
980         sfd->id = shp->shm_perm.id;
981         sfd->ns = get_ipc_ns(ns);
982         sfd->file = shp->shm_file;
983         sfd->vm_ops = NULL;
984
985         down_write(&current->mm->mmap_sem);
986         if (addr && !(shmflg & SHM_REMAP)) {
987                 err = -EINVAL;
988                 if (find_vma_intersection(current->mm, addr, addr + size))
989                         goto invalid;
990                 /*
991                  * If shm segment goes below stack, make sure there is some
992                  * space left for the stack to grow (at least 4 pages).
993                  */
994                 if (addr < current->mm->start_stack &&
995                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
996                         goto invalid;
997         }
998                 
999         user_addr = do_mmap (file, addr, size, prot, flags, 0);
1000         *raddr = user_addr;
1001         err = 0;
1002         if (IS_ERR_VALUE(user_addr))
1003                 err = (long)user_addr;
1004 invalid:
1005         up_write(&current->mm->mmap_sem);
1006
1007         fput(file);
1008
1009 out_nattch:
1010         down_write(&shm_ids(ns).rw_mutex);
1011         shp = shm_lock_down(ns, shmid);
1012         BUG_ON(IS_ERR(shp));
1013         shp->shm_nattch--;
1014         if(shp->shm_nattch == 0 &&
1015            shp->shm_perm.mode & SHM_DEST)
1016                 shm_destroy(ns, shp);
1017         else
1018                 shm_unlock(shp);
1019         up_write(&shm_ids(ns).rw_mutex);
1020
1021 out:
1022         return err;
1023
1024 out_unlock:
1025         shm_unlock(shp);
1026         goto out;
1027
1028 out_free:
1029         kfree(sfd);
1030 out_put_dentry:
1031         dput(path.dentry);
1032         goto out_nattch;
1033 }
1034
1035 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
1036 {
1037         unsigned long ret;
1038         long err;
1039
1040         err = do_shmat(shmid, shmaddr, shmflg, &ret);
1041         if (err)
1042                 return err;
1043         force_successful_syscall_return();
1044         return (long)ret;
1045 }
1046
1047 /*
1048  * detach and kill segment if marked destroyed.
1049  * The work is done in shm_close.
1050  */
1051 asmlinkage long sys_shmdt(char __user *shmaddr)
1052 {
1053         struct mm_struct *mm = current->mm;
1054         struct vm_area_struct *vma, *next;
1055         unsigned long addr = (unsigned long)shmaddr;
1056         loff_t size = 0;
1057         int retval = -EINVAL;
1058
1059         if (addr & ~PAGE_MASK)
1060                 return retval;
1061
1062         down_write(&mm->mmap_sem);
1063
1064         /*
1065          * This function tries to be smart and unmap shm segments that
1066          * were modified by partial mlock or munmap calls:
1067          * - It first determines the size of the shm segment that should be
1068          *   unmapped: It searches for a vma that is backed by shm and that
1069          *   started at address shmaddr. It records it's size and then unmaps
1070          *   it.
1071          * - Then it unmaps all shm vmas that started at shmaddr and that
1072          *   are within the initially determined size.
1073          * Errors from do_munmap are ignored: the function only fails if
1074          * it's called with invalid parameters or if it's called to unmap
1075          * a part of a vma. Both calls in this function are for full vmas,
1076          * the parameters are directly copied from the vma itself and always
1077          * valid - therefore do_munmap cannot fail. (famous last words?)
1078          */
1079         /*
1080          * If it had been mremap()'d, the starting address would not
1081          * match the usual checks anyway. So assume all vma's are
1082          * above the starting address given.
1083          */
1084         vma = find_vma(mm, addr);
1085
1086         while (vma) {
1087                 next = vma->vm_next;
1088
1089                 /*
1090                  * Check if the starting address would match, i.e. it's
1091                  * a fragment created by mprotect() and/or munmap(), or it
1092                  * otherwise it starts at this address with no hassles.
1093                  */
1094                 if ((vma->vm_ops == &shm_vm_ops) &&
1095                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1096
1097
1098                         size = vma->vm_file->f_path.dentry->d_inode->i_size;
1099                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1100                         /*
1101                          * We discovered the size of the shm segment, so
1102                          * break out of here and fall through to the next
1103                          * loop that uses the size information to stop
1104                          * searching for matching vma's.
1105                          */
1106                         retval = 0;
1107                         vma = next;
1108                         break;
1109                 }
1110                 vma = next;
1111         }
1112
1113         /*
1114          * We need look no further than the maximum address a fragment
1115          * could possibly have landed at. Also cast things to loff_t to
1116          * prevent overflows and make comparisions vs. equal-width types.
1117          */
1118         size = PAGE_ALIGN(size);
1119         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1120                 next = vma->vm_next;
1121
1122                 /* finding a matching vma now does not alter retval */
1123                 if ((vma->vm_ops == &shm_vm_ops) &&
1124                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1125
1126                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1127                 vma = next;
1128         }
1129
1130         up_write(&mm->mmap_sem);
1131         return retval;
1132 }
1133
1134 #ifdef CONFIG_PROC_FS
1135 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1136 {
1137         struct shmid_kernel *shp = it;
1138         char *format;
1139
1140 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1141 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
1142
1143         if (sizeof(size_t) <= sizeof(int))
1144                 format = SMALL_STRING;
1145         else
1146                 format = BIG_STRING;
1147         return seq_printf(s, format,
1148                           shp->shm_perm.key,
1149                           shp->shm_perm.id,
1150                           shp->shm_perm.mode,
1151                           shp->shm_segsz,
1152                           shp->shm_cprid,
1153                           shp->shm_lprid,
1154                           shp->shm_nattch,
1155                           shp->shm_perm.uid,
1156                           shp->shm_perm.gid,
1157                           shp->shm_perm.cuid,
1158                           shp->shm_perm.cgid,
1159                           shp->shm_atim,
1160                           shp->shm_dtim,
1161                           shp->shm_ctim);
1162 }
1163 #endif