]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - ipc/shm.c
pwm: make PWM_POLARITY_INVERTED flag optional
[karo-tx-linux.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  * Better ipc lock (kern_ipc_perm.lock) handling
24  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
25  */
26
27 #include <linux/slab.h>
28 #include <linux/mm.h>
29 #include <linux/hugetlb.h>
30 #include <linux/shm.h>
31 #include <linux/init.h>
32 #include <linux/file.h>
33 #include <linux/mman.h>
34 #include <linux/shmem_fs.h>
35 #include <linux/security.h>
36 #include <linux/syscalls.h>
37 #include <linux/audit.h>
38 #include <linux/capability.h>
39 #include <linux/ptrace.h>
40 #include <linux/seq_file.h>
41 #include <linux/rwsem.h>
42 #include <linux/nsproxy.h>
43 #include <linux/mount.h>
44 #include <linux/ipc_namespace.h>
45
46 #include <asm/uaccess.h>
47
48 #include "util.h"
49
50 struct shm_file_data {
51         int id;
52         struct ipc_namespace *ns;
53         struct file *file;
54         const struct vm_operations_struct *vm_ops;
55 };
56
57 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
58
59 static const struct file_operations shm_file_operations;
60 static const struct vm_operations_struct shm_vm_ops;
61
62 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
63
64 #define shm_unlock(shp)                 \
65         ipc_unlock(&(shp)->shm_perm)
66
67 static int newseg(struct ipc_namespace *, struct ipc_params *);
68 static void shm_open(struct vm_area_struct *vma);
69 static void shm_close(struct vm_area_struct *vma);
70 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
71 #ifdef CONFIG_PROC_FS
72 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
73 #endif
74
75 void shm_init_ns(struct ipc_namespace *ns)
76 {
77         ns->shm_ctlmax = SHMMAX;
78         ns->shm_ctlall = SHMALL;
79         ns->shm_ctlmni = SHMMNI;
80         ns->shm_rmid_forced = 0;
81         ns->shm_tot = 0;
82         ipc_init_ids(&shm_ids(ns));
83 }
84
85 /*
86  * Called with shm_ids.rwsem (writer) and the shp structure locked.
87  * Only shm_ids.rwsem remains locked on exit.
88  */
89 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
90 {
91         struct shmid_kernel *shp;
92         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
93
94         if (shp->shm_nattch){
95                 shp->shm_perm.mode |= SHM_DEST;
96                 /* Do not find it any more */
97                 shp->shm_perm.key = IPC_PRIVATE;
98                 shm_unlock(shp);
99         } else
100                 shm_destroy(ns, shp);
101 }
102
103 #ifdef CONFIG_IPC_NS
104 void shm_exit_ns(struct ipc_namespace *ns)
105 {
106         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
107         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
108 }
109 #endif
110
111 static int __init ipc_ns_init(void)
112 {
113         shm_init_ns(&init_ipc_ns);
114         return 0;
115 }
116
117 pure_initcall(ipc_ns_init);
118
119 void __init shm_init (void)
120 {
121         ipc_init_proc_interface("sysvipc/shm",
122 #if BITS_PER_LONG <= 32
123                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
124 #else
125                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
126 #endif
127                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
128 }
129
130 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
131 {
132         struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
133
134         if (IS_ERR(ipcp))
135                 return ERR_CAST(ipcp);
136
137         return container_of(ipcp, struct shmid_kernel, shm_perm);
138 }
139
140 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
141 {
142         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
143
144         if (IS_ERR(ipcp))
145                 return ERR_CAST(ipcp);
146
147         return container_of(ipcp, struct shmid_kernel, shm_perm);
148 }
149
150 /*
151  * shm_lock_(check_) routines are called in the paths where the rwsem
152  * is not necessarily held.
153  */
154 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
155 {
156         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
157
158         if (IS_ERR(ipcp))
159                 return (struct shmid_kernel *)ipcp;
160
161         return container_of(ipcp, struct shmid_kernel, shm_perm);
162 }
163
164 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
165 {
166         rcu_read_lock();
167         ipc_lock_object(&ipcp->shm_perm);
168 }
169
170 static void shm_rcu_free(struct rcu_head *head)
171 {
172         struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
173         struct shmid_kernel *shp = ipc_rcu_to_struct(p);
174
175         security_shm_free(shp);
176         ipc_rcu_free(head);
177 }
178
179 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
180 {
181         ipc_rmid(&shm_ids(ns), &s->shm_perm);
182 }
183
184
185 /* This is called by fork, once for every shm attach. */
186 static void shm_open(struct vm_area_struct *vma)
187 {
188         struct file *file = vma->vm_file;
189         struct shm_file_data *sfd = shm_file_data(file);
190         struct shmid_kernel *shp;
191
192         shp = shm_lock(sfd->ns, sfd->id);
193         BUG_ON(IS_ERR(shp));
194         shp->shm_atim = get_seconds();
195         shp->shm_lprid = task_tgid_vnr(current);
196         shp->shm_nattch++;
197         shm_unlock(shp);
198 }
199
200 /*
201  * shm_destroy - free the struct shmid_kernel
202  *
203  * @ns: namespace
204  * @shp: struct to free
205  *
206  * It has to be called with shp and shm_ids.rwsem (writer) locked,
207  * but returns with shp unlocked and freed.
208  */
209 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
210 {
211         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
212         shm_rmid(ns, shp);
213         shm_unlock(shp);
214         if (!is_file_hugepages(shp->shm_file))
215                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
216         else if (shp->mlock_user)
217                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
218                                                 shp->mlock_user);
219         fput (shp->shm_file);
220         ipc_rcu_putref(shp, shm_rcu_free);
221 }
222
223 /*
224  * shm_may_destroy - identifies whether shm segment should be destroyed now
225  *
226  * Returns true if and only if there are no active users of the segment and
227  * one of the following is true:
228  *
229  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
230  *
231  * 2) sysctl kernel.shm_rmid_forced is set to 1.
232  */
233 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
234 {
235         return (shp->shm_nattch == 0) &&
236                (ns->shm_rmid_forced ||
237                 (shp->shm_perm.mode & SHM_DEST));
238 }
239
240 /*
241  * remove the attach descriptor vma.
242  * free memory for segment if it is marked destroyed.
243  * The descriptor has already been removed from the current->mm->mmap list
244  * and will later be kfree()d.
245  */
246 static void shm_close(struct vm_area_struct *vma)
247 {
248         struct file * file = vma->vm_file;
249         struct shm_file_data *sfd = shm_file_data(file);
250         struct shmid_kernel *shp;
251         struct ipc_namespace *ns = sfd->ns;
252
253         down_write(&shm_ids(ns).rwsem);
254         /* remove from the list of attaches of the shm segment */
255         shp = shm_lock(ns, sfd->id);
256         BUG_ON(IS_ERR(shp));
257         shp->shm_lprid = task_tgid_vnr(current);
258         shp->shm_dtim = get_seconds();
259         shp->shm_nattch--;
260         if (shm_may_destroy(ns, shp))
261                 shm_destroy(ns, shp);
262         else
263                 shm_unlock(shp);
264         up_write(&shm_ids(ns).rwsem);
265 }
266
267 /* Called with ns->shm_ids(ns).rwsem locked */
268 static int shm_try_destroy_current(int id, void *p, void *data)
269 {
270         struct ipc_namespace *ns = data;
271         struct kern_ipc_perm *ipcp = p;
272         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
273
274         if (shp->shm_creator != current)
275                 return 0;
276
277         /*
278          * Mark it as orphaned to destroy the segment when
279          * kernel.shm_rmid_forced is changed.
280          * It is noop if the following shm_may_destroy() returns true.
281          */
282         shp->shm_creator = NULL;
283
284         /*
285          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
286          * is not set, it shouldn't be deleted here.
287          */
288         if (!ns->shm_rmid_forced)
289                 return 0;
290
291         if (shm_may_destroy(ns, shp)) {
292                 shm_lock_by_ptr(shp);
293                 shm_destroy(ns, shp);
294         }
295         return 0;
296 }
297
298 /* Called with ns->shm_ids(ns).rwsem locked */
299 static int shm_try_destroy_orphaned(int id, void *p, void *data)
300 {
301         struct ipc_namespace *ns = data;
302         struct kern_ipc_perm *ipcp = p;
303         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
304
305         /*
306          * We want to destroy segments without users and with already
307          * exit'ed originating process.
308          *
309          * As shp->* are changed under rwsem, it's safe to skip shp locking.
310          */
311         if (shp->shm_creator != NULL)
312                 return 0;
313
314         if (shm_may_destroy(ns, shp)) {
315                 shm_lock_by_ptr(shp);
316                 shm_destroy(ns, shp);
317         }
318         return 0;
319 }
320
321 void shm_destroy_orphaned(struct ipc_namespace *ns)
322 {
323         down_write(&shm_ids(ns).rwsem);
324         if (shm_ids(ns).in_use)
325                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
326         up_write(&shm_ids(ns).rwsem);
327 }
328
329
330 void exit_shm(struct task_struct *task)
331 {
332         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
333
334         if (shm_ids(ns).in_use == 0)
335                 return;
336
337         /* Destroy all already created segments, but not mapped yet */
338         down_write(&shm_ids(ns).rwsem);
339         if (shm_ids(ns).in_use)
340                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
341         up_write(&shm_ids(ns).rwsem);
342 }
343
344 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
345 {
346         struct file *file = vma->vm_file;
347         struct shm_file_data *sfd = shm_file_data(file);
348
349         return sfd->vm_ops->fault(vma, vmf);
350 }
351
352 #ifdef CONFIG_NUMA
353 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
354 {
355         struct file *file = vma->vm_file;
356         struct shm_file_data *sfd = shm_file_data(file);
357         int err = 0;
358         if (sfd->vm_ops->set_policy)
359                 err = sfd->vm_ops->set_policy(vma, new);
360         return err;
361 }
362
363 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
364                                         unsigned long addr)
365 {
366         struct file *file = vma->vm_file;
367         struct shm_file_data *sfd = shm_file_data(file);
368         struct mempolicy *pol = NULL;
369
370         if (sfd->vm_ops->get_policy)
371                 pol = sfd->vm_ops->get_policy(vma, addr);
372         else if (vma->vm_policy)
373                 pol = vma->vm_policy;
374
375         return pol;
376 }
377 #endif
378
379 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
380 {
381         struct shm_file_data *sfd = shm_file_data(file);
382         int ret;
383
384         ret = sfd->file->f_op->mmap(sfd->file, vma);
385         if (ret != 0)
386                 return ret;
387         sfd->vm_ops = vma->vm_ops;
388 #ifdef CONFIG_MMU
389         BUG_ON(!sfd->vm_ops->fault);
390 #endif
391         vma->vm_ops = &shm_vm_ops;
392         shm_open(vma);
393
394         return ret;
395 }
396
397 static int shm_release(struct inode *ino, struct file *file)
398 {
399         struct shm_file_data *sfd = shm_file_data(file);
400
401         put_ipc_ns(sfd->ns);
402         shm_file_data(file) = NULL;
403         kfree(sfd);
404         return 0;
405 }
406
407 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
408 {
409         struct shm_file_data *sfd = shm_file_data(file);
410
411         if (!sfd->file->f_op->fsync)
412                 return -EINVAL;
413         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
414 }
415
416 static long shm_fallocate(struct file *file, int mode, loff_t offset,
417                           loff_t len)
418 {
419         struct shm_file_data *sfd = shm_file_data(file);
420
421         if (!sfd->file->f_op->fallocate)
422                 return -EOPNOTSUPP;
423         return sfd->file->f_op->fallocate(file, mode, offset, len);
424 }
425
426 static unsigned long shm_get_unmapped_area(struct file *file,
427         unsigned long addr, unsigned long len, unsigned long pgoff,
428         unsigned long flags)
429 {
430         struct shm_file_data *sfd = shm_file_data(file);
431         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
432                                                 pgoff, flags);
433 }
434
435 static const struct file_operations shm_file_operations = {
436         .mmap           = shm_mmap,
437         .fsync          = shm_fsync,
438         .release        = shm_release,
439 #ifndef CONFIG_MMU
440         .get_unmapped_area      = shm_get_unmapped_area,
441 #endif
442         .llseek         = noop_llseek,
443         .fallocate      = shm_fallocate,
444 };
445
446 static const struct file_operations shm_file_operations_huge = {
447         .mmap           = shm_mmap,
448         .fsync          = shm_fsync,
449         .release        = shm_release,
450         .get_unmapped_area      = shm_get_unmapped_area,
451         .llseek         = noop_llseek,
452         .fallocate      = shm_fallocate,
453 };
454
455 int is_file_shm_hugepages(struct file *file)
456 {
457         return file->f_op == &shm_file_operations_huge;
458 }
459
460 static const struct vm_operations_struct shm_vm_ops = {
461         .open   = shm_open,     /* callback for a new vm-area open */
462         .close  = shm_close,    /* callback for when the vm-area is released */
463         .fault  = shm_fault,
464 #if defined(CONFIG_NUMA)
465         .set_policy = shm_set_policy,
466         .get_policy = shm_get_policy,
467 #endif
468 };
469
470 /**
471  * newseg - Create a new shared memory segment
472  * @ns: namespace
473  * @params: ptr to the structure that contains key, size and shmflg
474  *
475  * Called with shm_ids.rwsem held as a writer.
476  */
477
478 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
479 {
480         key_t key = params->key;
481         int shmflg = params->flg;
482         size_t size = params->u.size;
483         int error;
484         struct shmid_kernel *shp;
485         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
486         struct file * file;
487         char name[13];
488         int id;
489         vm_flags_t acctflag = 0;
490
491         if (size < SHMMIN || size > ns->shm_ctlmax)
492                 return -EINVAL;
493
494         if (ns->shm_tot + numpages > ns->shm_ctlall)
495                 return -ENOSPC;
496
497         shp = ipc_rcu_alloc(sizeof(*shp));
498         if (!shp)
499                 return -ENOMEM;
500
501         shp->shm_perm.key = key;
502         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
503         shp->mlock_user = NULL;
504
505         shp->shm_perm.security = NULL;
506         error = security_shm_alloc(shp);
507         if (error) {
508                 ipc_rcu_putref(shp, ipc_rcu_free);
509                 return error;
510         }
511
512         sprintf (name, "SYSV%08x", key);
513         if (shmflg & SHM_HUGETLB) {
514                 struct hstate *hs;
515                 size_t hugesize;
516
517                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
518                 if (!hs) {
519                         error = -EINVAL;
520                         goto no_file;
521                 }
522                 hugesize = ALIGN(size, huge_page_size(hs));
523
524                 /* hugetlb_file_setup applies strict accounting */
525                 if (shmflg & SHM_NORESERVE)
526                         acctflag = VM_NORESERVE;
527                 file = hugetlb_file_setup(name, hugesize, acctflag,
528                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
529                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
530         } else {
531                 /*
532                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
533                  * if it's asked for.
534                  */
535                 if  ((shmflg & SHM_NORESERVE) &&
536                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
537                         acctflag = VM_NORESERVE;
538                 file = shmem_file_setup(name, size, acctflag);
539         }
540         error = PTR_ERR(file);
541         if (IS_ERR(file))
542                 goto no_file;
543
544         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
545         if (id < 0) {
546                 error = id;
547                 goto no_id;
548         }
549
550         shp->shm_cprid = task_tgid_vnr(current);
551         shp->shm_lprid = 0;
552         shp->shm_atim = shp->shm_dtim = 0;
553         shp->shm_ctim = get_seconds();
554         shp->shm_segsz = size;
555         shp->shm_nattch = 0;
556         shp->shm_file = file;
557         shp->shm_creator = current;
558
559         /*
560          * shmid gets reported as "inode#" in /proc/pid/maps.
561          * proc-ps tools use this. Changing this will break them.
562          */
563         file_inode(file)->i_ino = shp->shm_perm.id;
564
565         ns->shm_tot += numpages;
566         error = shp->shm_perm.id;
567
568         ipc_unlock_object(&shp->shm_perm);
569         rcu_read_unlock();
570         return error;
571
572 no_id:
573         if (is_file_hugepages(file) && shp->mlock_user)
574                 user_shm_unlock(size, shp->mlock_user);
575         fput(file);
576 no_file:
577         ipc_rcu_putref(shp, shm_rcu_free);
578         return error;
579 }
580
581 /*
582  * Called with shm_ids.rwsem and ipcp locked.
583  */
584 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
585 {
586         struct shmid_kernel *shp;
587
588         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
589         return security_shm_associate(shp, shmflg);
590 }
591
592 /*
593  * Called with shm_ids.rwsem and ipcp locked.
594  */
595 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
596                                 struct ipc_params *params)
597 {
598         struct shmid_kernel *shp;
599
600         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
601         if (shp->shm_segsz < params->u.size)
602                 return -EINVAL;
603
604         return 0;
605 }
606
607 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
608 {
609         struct ipc_namespace *ns;
610         struct ipc_ops shm_ops;
611         struct ipc_params shm_params;
612
613         ns = current->nsproxy->ipc_ns;
614
615         shm_ops.getnew = newseg;
616         shm_ops.associate = shm_security;
617         shm_ops.more_checks = shm_more_checks;
618
619         shm_params.key = key;
620         shm_params.flg = shmflg;
621         shm_params.u.size = size;
622
623         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
624 }
625
626 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *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 shmid_ds out;
634
635                 memset(&out, 0, sizeof(out));
636                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
637                 out.shm_segsz   = in->shm_segsz;
638                 out.shm_atime   = in->shm_atime;
639                 out.shm_dtime   = in->shm_dtime;
640                 out.shm_ctime   = in->shm_ctime;
641                 out.shm_cpid    = in->shm_cpid;
642                 out.shm_lpid    = in->shm_lpid;
643                 out.shm_nattch  = in->shm_nattch;
644
645                 return copy_to_user(buf, &out, sizeof(out));
646             }
647         default:
648                 return -EINVAL;
649         }
650 }
651
652 static inline unsigned long
653 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
654 {
655         switch(version) {
656         case IPC_64:
657                 if (copy_from_user(out, buf, sizeof(*out)))
658                         return -EFAULT;
659                 return 0;
660         case IPC_OLD:
661             {
662                 struct shmid_ds tbuf_old;
663
664                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
665                         return -EFAULT;
666
667                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
668                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
669                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
670
671                 return 0;
672             }
673         default:
674                 return -EINVAL;
675         }
676 }
677
678 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
679 {
680         switch(version) {
681         case IPC_64:
682                 return copy_to_user(buf, in, sizeof(*in));
683         case IPC_OLD:
684             {
685                 struct shminfo out;
686
687                 if(in->shmmax > INT_MAX)
688                         out.shmmax = INT_MAX;
689                 else
690                         out.shmmax = (int)in->shmmax;
691
692                 out.shmmin      = in->shmmin;
693                 out.shmmni      = in->shmmni;
694                 out.shmseg      = in->shmseg;
695                 out.shmall      = in->shmall; 
696
697                 return copy_to_user(buf, &out, sizeof(out));
698             }
699         default:
700                 return -EINVAL;
701         }
702 }
703
704 /*
705  * Calculate and add used RSS and swap pages of a shm.
706  * Called with shm_ids.rwsem held as a reader
707  */
708 static void shm_add_rss_swap(struct shmid_kernel *shp,
709         unsigned long *rss_add, unsigned long *swp_add)
710 {
711         struct inode *inode;
712
713         inode = file_inode(shp->shm_file);
714
715         if (is_file_hugepages(shp->shm_file)) {
716                 struct address_space *mapping = inode->i_mapping;
717                 struct hstate *h = hstate_file(shp->shm_file);
718                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
719         } else {
720 #ifdef CONFIG_SHMEM
721                 struct shmem_inode_info *info = SHMEM_I(inode);
722                 spin_lock(&info->lock);
723                 *rss_add += inode->i_mapping->nrpages;
724                 *swp_add += info->swapped;
725                 spin_unlock(&info->lock);
726 #else
727                 *rss_add += inode->i_mapping->nrpages;
728 #endif
729         }
730 }
731
732 /*
733  * Called with shm_ids.rwsem held as a reader
734  */
735 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
736                 unsigned long *swp)
737 {
738         int next_id;
739         int total, in_use;
740
741         *rss = 0;
742         *swp = 0;
743
744         in_use = shm_ids(ns).in_use;
745
746         for (total = 0, next_id = 0; total < in_use; next_id++) {
747                 struct kern_ipc_perm *ipc;
748                 struct shmid_kernel *shp;
749
750                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
751                 if (ipc == NULL)
752                         continue;
753                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
754
755                 shm_add_rss_swap(shp, rss, swp);
756
757                 total++;
758         }
759 }
760
761 /*
762  * This function handles some shmctl commands which require the rwsem
763  * to be held in write mode.
764  * NOTE: no locks must be held, the rwsem is taken inside this function.
765  */
766 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
767                        struct shmid_ds __user *buf, int version)
768 {
769         struct kern_ipc_perm *ipcp;
770         struct shmid64_ds shmid64;
771         struct shmid_kernel *shp;
772         int err;
773
774         if (cmd == IPC_SET) {
775                 if (copy_shmid_from_user(&shmid64, buf, version))
776                         return -EFAULT;
777         }
778
779         down_write(&shm_ids(ns).rwsem);
780         rcu_read_lock();
781
782         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
783                                       &shmid64.shm_perm, 0);
784         if (IS_ERR(ipcp)) {
785                 err = PTR_ERR(ipcp);
786                 goto out_unlock1;
787         }
788
789         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
790
791         err = security_shm_shmctl(shp, cmd);
792         if (err)
793                 goto out_unlock1;
794
795         switch (cmd) {
796         case IPC_RMID:
797                 ipc_lock_object(&shp->shm_perm);
798                 /* do_shm_rmid unlocks the ipc object and rcu */
799                 do_shm_rmid(ns, ipcp);
800                 goto out_up;
801         case IPC_SET:
802                 ipc_lock_object(&shp->shm_perm);
803                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
804                 if (err)
805                         goto out_unlock0;
806                 shp->shm_ctim = get_seconds();
807                 break;
808         default:
809                 err = -EINVAL;
810                 goto out_unlock1;
811         }
812
813 out_unlock0:
814         ipc_unlock_object(&shp->shm_perm);
815 out_unlock1:
816         rcu_read_unlock();
817 out_up:
818         up_write(&shm_ids(ns).rwsem);
819         return err;
820 }
821
822 static int shmctl_nolock(struct ipc_namespace *ns, int shmid,
823                          int cmd, int version, void __user *buf)
824 {
825         int err;
826         struct shmid_kernel *shp;
827
828         /* preliminary security checks for *_INFO */
829         if (cmd == IPC_INFO || cmd == SHM_INFO) {
830                 err = security_shm_shmctl(NULL, cmd);
831                 if (err)
832                         return err;
833         }
834
835         switch (cmd) {
836         case IPC_INFO:
837         {
838                 struct shminfo64 shminfo;
839
840                 memset(&shminfo, 0, sizeof(shminfo));
841                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
842                 shminfo.shmmax = ns->shm_ctlmax;
843                 shminfo.shmall = ns->shm_ctlall;
844
845                 shminfo.shmmin = SHMMIN;
846                 if(copy_shminfo_to_user (buf, &shminfo, version))
847                         return -EFAULT;
848
849                 down_read(&shm_ids(ns).rwsem);
850                 err = ipc_get_maxid(&shm_ids(ns));
851                 up_read(&shm_ids(ns).rwsem);
852
853                 if(err<0)
854                         err = 0;
855                 goto out;
856         }
857         case SHM_INFO:
858         {
859                 struct shm_info shm_info;
860
861                 memset(&shm_info, 0, sizeof(shm_info));
862                 down_read(&shm_ids(ns).rwsem);
863                 shm_info.used_ids = shm_ids(ns).in_use;
864                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
865                 shm_info.shm_tot = ns->shm_tot;
866                 shm_info.swap_attempts = 0;
867                 shm_info.swap_successes = 0;
868                 err = ipc_get_maxid(&shm_ids(ns));
869                 up_read(&shm_ids(ns).rwsem);
870                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
871                         err = -EFAULT;
872                         goto out;
873                 }
874
875                 err = err < 0 ? 0 : err;
876                 goto out;
877         }
878         case SHM_STAT:
879         case IPC_STAT:
880         {
881                 struct shmid64_ds tbuf;
882                 int result;
883
884                 rcu_read_lock();
885                 if (cmd == SHM_STAT) {
886                         shp = shm_obtain_object(ns, shmid);
887                         if (IS_ERR(shp)) {
888                                 err = PTR_ERR(shp);
889                                 goto out_unlock;
890                         }
891                         result = shp->shm_perm.id;
892                 } else {
893                         shp = shm_obtain_object_check(ns, shmid);
894                         if (IS_ERR(shp)) {
895                                 err = PTR_ERR(shp);
896                                 goto out_unlock;
897                         }
898                         result = 0;
899                 }
900
901                 err = -EACCES;
902                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
903                         goto out_unlock;
904
905                 err = security_shm_shmctl(shp, cmd);
906                 if (err)
907                         goto out_unlock;
908
909                 memset(&tbuf, 0, sizeof(tbuf));
910                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
911                 tbuf.shm_segsz  = shp->shm_segsz;
912                 tbuf.shm_atime  = shp->shm_atim;
913                 tbuf.shm_dtime  = shp->shm_dtim;
914                 tbuf.shm_ctime  = shp->shm_ctim;
915                 tbuf.shm_cpid   = shp->shm_cprid;
916                 tbuf.shm_lpid   = shp->shm_lprid;
917                 tbuf.shm_nattch = shp->shm_nattch;
918                 rcu_read_unlock();
919
920                 if (copy_shmid_to_user(buf, &tbuf, version))
921                         err = -EFAULT;
922                 else
923                         err = result;
924                 goto out;
925         }
926         default:
927                 return -EINVAL;
928         }
929
930 out_unlock:
931         rcu_read_unlock();
932 out:
933         return err;
934 }
935
936 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
937 {
938         struct shmid_kernel *shp;
939         int err, version;
940         struct ipc_namespace *ns;
941
942         if (cmd < 0 || shmid < 0)
943                 return -EINVAL;
944
945         version = ipc_parse_version(&cmd);
946         ns = current->nsproxy->ipc_ns;
947
948         switch (cmd) {
949         case IPC_INFO:
950         case SHM_INFO:
951         case SHM_STAT:
952         case IPC_STAT:
953                 return shmctl_nolock(ns, shmid, cmd, version, buf);
954         case IPC_RMID:
955         case IPC_SET:
956                 return shmctl_down(ns, shmid, cmd, buf, version);
957         case SHM_LOCK:
958         case SHM_UNLOCK:
959         {
960                 struct file *shm_file;
961
962                 rcu_read_lock();
963                 shp = shm_obtain_object_check(ns, shmid);
964                 if (IS_ERR(shp)) {
965                         err = PTR_ERR(shp);
966                         goto out_unlock1;
967                 }
968
969                 audit_ipc_obj(&(shp->shm_perm));
970                 err = security_shm_shmctl(shp, cmd);
971                 if (err)
972                         goto out_unlock1;
973
974                 ipc_lock_object(&shp->shm_perm);
975                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
976                         kuid_t euid = current_euid();
977                         err = -EPERM;
978                         if (!uid_eq(euid, shp->shm_perm.uid) &&
979                             !uid_eq(euid, shp->shm_perm.cuid))
980                                 goto out_unlock0;
981                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
982                                 goto out_unlock0;
983                 }
984
985                 shm_file = shp->shm_file;
986                 if (is_file_hugepages(shm_file))
987                         goto out_unlock0;
988
989                 if (cmd == SHM_LOCK) {
990                         struct user_struct *user = current_user();
991                         err = shmem_lock(shm_file, 1, user);
992                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
993                                 shp->shm_perm.mode |= SHM_LOCKED;
994                                 shp->mlock_user = user;
995                         }
996                         goto out_unlock0;
997                 }
998
999                 /* SHM_UNLOCK */
1000                 if (!(shp->shm_perm.mode & SHM_LOCKED))
1001                         goto out_unlock0;
1002                 shmem_lock(shm_file, 0, shp->mlock_user);
1003                 shp->shm_perm.mode &= ~SHM_LOCKED;
1004                 shp->mlock_user = NULL;
1005                 get_file(shm_file);
1006                 ipc_unlock_object(&shp->shm_perm);
1007                 rcu_read_unlock();
1008                 shmem_unlock_mapping(shm_file->f_mapping);
1009
1010                 fput(shm_file);
1011                 return err;
1012         }
1013         default:
1014                 return -EINVAL;
1015         }
1016
1017 out_unlock0:
1018         ipc_unlock_object(&shp->shm_perm);
1019 out_unlock1:
1020         rcu_read_unlock();
1021         return err;
1022 }
1023
1024 /*
1025  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1026  *
1027  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1028  * "raddr" thing points to kernel space, and there has to be a wrapper around
1029  * this.
1030  */
1031 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
1032               unsigned long shmlba)
1033 {
1034         struct shmid_kernel *shp;
1035         unsigned long addr;
1036         unsigned long size;
1037         struct file * file;
1038         int    err;
1039         unsigned long flags;
1040         unsigned long prot;
1041         int acc_mode;
1042         struct ipc_namespace *ns;
1043         struct shm_file_data *sfd;
1044         struct path path;
1045         fmode_t f_mode;
1046         unsigned long populate = 0;
1047
1048         err = -EINVAL;
1049         if (shmid < 0)
1050                 goto out;
1051         else if ((addr = (ulong)shmaddr)) {
1052                 if (addr & (shmlba - 1)) {
1053                         if (shmflg & SHM_RND)
1054                                 addr &= ~(shmlba - 1);     /* round down */
1055                         else
1056 #ifndef __ARCH_FORCE_SHMLBA
1057                                 if (addr & ~PAGE_MASK)
1058 #endif
1059                                         goto out;
1060                 }
1061                 flags = MAP_SHARED | MAP_FIXED;
1062         } else {
1063                 if ((shmflg & SHM_REMAP))
1064                         goto out;
1065
1066                 flags = MAP_SHARED;
1067         }
1068
1069         if (shmflg & SHM_RDONLY) {
1070                 prot = PROT_READ;
1071                 acc_mode = S_IRUGO;
1072                 f_mode = FMODE_READ;
1073         } else {
1074                 prot = PROT_READ | PROT_WRITE;
1075                 acc_mode = S_IRUGO | S_IWUGO;
1076                 f_mode = FMODE_READ | FMODE_WRITE;
1077         }
1078         if (shmflg & SHM_EXEC) {
1079                 prot |= PROT_EXEC;
1080                 acc_mode |= S_IXUGO;
1081         }
1082
1083         /*
1084          * We cannot rely on the fs check since SYSV IPC does have an
1085          * additional creator id...
1086          */
1087         ns = current->nsproxy->ipc_ns;
1088         rcu_read_lock();
1089         shp = shm_obtain_object_check(ns, shmid);
1090         if (IS_ERR(shp)) {
1091                 err = PTR_ERR(shp);
1092                 goto out_unlock;
1093         }
1094
1095         err = -EACCES;
1096         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1097                 goto out_unlock;
1098
1099         err = security_shm_shmat(shp, shmaddr, shmflg);
1100         if (err)
1101                 goto out_unlock;
1102
1103         ipc_lock_object(&shp->shm_perm);
1104         path = shp->shm_file->f_path;
1105         path_get(&path);
1106         shp->shm_nattch++;
1107         size = i_size_read(path.dentry->d_inode);
1108         ipc_unlock_object(&shp->shm_perm);
1109         rcu_read_unlock();
1110
1111         err = -ENOMEM;
1112         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1113         if (!sfd) {
1114                 path_put(&path);
1115                 goto out_nattch;
1116         }
1117
1118         file = alloc_file(&path, f_mode,
1119                           is_file_hugepages(shp->shm_file) ?
1120                                 &shm_file_operations_huge :
1121                                 &shm_file_operations);
1122         err = PTR_ERR(file);
1123         if (IS_ERR(file)) {
1124                 kfree(sfd);
1125                 path_put(&path);
1126                 goto out_nattch;
1127         }
1128
1129         file->private_data = sfd;
1130         file->f_mapping = shp->shm_file->f_mapping;
1131         sfd->id = shp->shm_perm.id;
1132         sfd->ns = get_ipc_ns(ns);
1133         sfd->file = shp->shm_file;
1134         sfd->vm_ops = NULL;
1135
1136         err = security_mmap_file(file, prot, flags);
1137         if (err)
1138                 goto out_fput;
1139
1140         down_write(&current->mm->mmap_sem);
1141         if (addr && !(shmflg & SHM_REMAP)) {
1142                 err = -EINVAL;
1143                 if (find_vma_intersection(current->mm, addr, addr + size))
1144                         goto invalid;
1145                 /*
1146                  * If shm segment goes below stack, make sure there is some
1147                  * space left for the stack to grow (at least 4 pages).
1148                  */
1149                 if (addr < current->mm->start_stack &&
1150                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1151                         goto invalid;
1152         }
1153
1154         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1155         *raddr = addr;
1156         err = 0;
1157         if (IS_ERR_VALUE(addr))
1158                 err = (long)addr;
1159 invalid:
1160         up_write(&current->mm->mmap_sem);
1161         if (populate)
1162                 mm_populate(addr, populate);
1163
1164 out_fput:
1165         fput(file);
1166
1167 out_nattch:
1168         down_write(&shm_ids(ns).rwsem);
1169         shp = shm_lock(ns, shmid);
1170         BUG_ON(IS_ERR(shp));
1171         shp->shm_nattch--;
1172         if (shm_may_destroy(ns, shp))
1173                 shm_destroy(ns, shp);
1174         else
1175                 shm_unlock(shp);
1176         up_write(&shm_ids(ns).rwsem);
1177         return err;
1178
1179 out_unlock:
1180         rcu_read_unlock();
1181 out:
1182         return err;
1183 }
1184
1185 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1186 {
1187         unsigned long ret;
1188         long err;
1189
1190         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1191         if (err)
1192                 return err;
1193         force_successful_syscall_return();
1194         return (long)ret;
1195 }
1196
1197 /*
1198  * detach and kill segment if marked destroyed.
1199  * The work is done in shm_close.
1200  */
1201 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1202 {
1203         struct mm_struct *mm = current->mm;
1204         struct vm_area_struct *vma;
1205         unsigned long addr = (unsigned long)shmaddr;
1206         int retval = -EINVAL;
1207 #ifdef CONFIG_MMU
1208         loff_t size = 0;
1209         struct vm_area_struct *next;
1210 #endif
1211
1212         if (addr & ~PAGE_MASK)
1213                 return retval;
1214
1215         down_write(&mm->mmap_sem);
1216
1217         /*
1218          * This function tries to be smart and unmap shm segments that
1219          * were modified by partial mlock or munmap calls:
1220          * - It first determines the size of the shm segment that should be
1221          *   unmapped: It searches for a vma that is backed by shm and that
1222          *   started at address shmaddr. It records it's size and then unmaps
1223          *   it.
1224          * - Then it unmaps all shm vmas that started at shmaddr and that
1225          *   are within the initially determined size.
1226          * Errors from do_munmap are ignored: the function only fails if
1227          * it's called with invalid parameters or if it's called to unmap
1228          * a part of a vma. Both calls in this function are for full vmas,
1229          * the parameters are directly copied from the vma itself and always
1230          * valid - therefore do_munmap cannot fail. (famous last words?)
1231          */
1232         /*
1233          * If it had been mremap()'d, the starting address would not
1234          * match the usual checks anyway. So assume all vma's are
1235          * above the starting address given.
1236          */
1237         vma = find_vma(mm, addr);
1238
1239 #ifdef CONFIG_MMU
1240         while (vma) {
1241                 next = vma->vm_next;
1242
1243                 /*
1244                  * Check if the starting address would match, i.e. it's
1245                  * a fragment created by mprotect() and/or munmap(), or it
1246                  * otherwise it starts at this address with no hassles.
1247                  */
1248                 if ((vma->vm_ops == &shm_vm_ops) &&
1249                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1250
1251
1252                         size = file_inode(vma->vm_file)->i_size;
1253                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1254                         /*
1255                          * We discovered the size of the shm segment, so
1256                          * break out of here and fall through to the next
1257                          * loop that uses the size information to stop
1258                          * searching for matching vma's.
1259                          */
1260                         retval = 0;
1261                         vma = next;
1262                         break;
1263                 }
1264                 vma = next;
1265         }
1266
1267         /*
1268          * We need look no further than the maximum address a fragment
1269          * could possibly have landed at. Also cast things to loff_t to
1270          * prevent overflows and make comparisons vs. equal-width types.
1271          */
1272         size = PAGE_ALIGN(size);
1273         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1274                 next = vma->vm_next;
1275
1276                 /* finding a matching vma now does not alter retval */
1277                 if ((vma->vm_ops == &shm_vm_ops) &&
1278                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1279
1280                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1281                 vma = next;
1282         }
1283
1284 #else /* CONFIG_MMU */
1285         /* under NOMMU conditions, the exact address to be destroyed must be
1286          * given */
1287         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1288                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1289                 retval = 0;
1290         }
1291
1292 #endif
1293
1294         up_write(&mm->mmap_sem);
1295         return retval;
1296 }
1297
1298 #ifdef CONFIG_PROC_FS
1299 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1300 {
1301         struct user_namespace *user_ns = seq_user_ns(s);
1302         struct shmid_kernel *shp = it;
1303         unsigned long rss = 0, swp = 0;
1304
1305         shm_add_rss_swap(shp, &rss, &swp);
1306
1307 #if BITS_PER_LONG <= 32
1308 #define SIZE_SPEC "%10lu"
1309 #else
1310 #define SIZE_SPEC "%21lu"
1311 #endif
1312
1313         return seq_printf(s,
1314                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1315                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1316                           SIZE_SPEC " " SIZE_SPEC "\n",
1317                           shp->shm_perm.key,
1318                           shp->shm_perm.id,
1319                           shp->shm_perm.mode,
1320                           shp->shm_segsz,
1321                           shp->shm_cprid,
1322                           shp->shm_lprid,
1323                           shp->shm_nattch,
1324                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1325                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1326                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1327                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1328                           shp->shm_atim,
1329                           shp->shm_dtim,
1330                           shp->shm_ctim,
1331                           rss * PAGE_SIZE,
1332                           swp * PAGE_SIZE);
1333 }
1334 #endif