acefe44fefefa187c838c3830d944b5d3a4c047d
[muen/linux.git] / ipc / shm.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/ipc/shm.c
4  * Copyright (C) 1992, 1993 Krishna Balasubramanian
5  *       Many improvements/fixes by Bruno Haible.
6  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
7  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
8  *
9  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
10  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
11  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
12  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
13  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
14  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
15  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
16  *
17  * support for audit of ipc object properties and permission changes
18  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19  *
20  * namespaces support
21  * OpenVZ, SWsoft Inc.
22  * Pavel Emelianov <xemul@openvz.org>
23  *
24  * Better ipc lock (kern_ipc_perm.lock) handling
25  * Davidlohr Bueso <davidlohr.bueso@hp.com>, June 2013.
26  */
27
28 #include <linux/slab.h>
29 #include <linux/mm.h>
30 #include <linux/hugetlb.h>
31 #include <linux/shm.h>
32 #include <linux/init.h>
33 #include <linux/file.h>
34 #include <linux/mman.h>
35 #include <linux/shmem_fs.h>
36 #include <linux/security.h>
37 #include <linux/syscalls.h>
38 #include <linux/audit.h>
39 #include <linux/capability.h>
40 #include <linux/ptrace.h>
41 #include <linux/seq_file.h>
42 #include <linux/rwsem.h>
43 #include <linux/nsproxy.h>
44 #include <linux/mount.h>
45 #include <linux/ipc_namespace.h>
46
47 #include <linux/uaccess.h>
48
49 #include "util.h"
50
51 struct shmid_kernel /* private to the kernel */
52 {
53         struct kern_ipc_perm    shm_perm;
54         struct file             *shm_file;
55         unsigned long           shm_nattch;
56         unsigned long           shm_segsz;
57         time64_t                shm_atim;
58         time64_t                shm_dtim;
59         time64_t                shm_ctim;
60         struct pid              *shm_cprid;
61         struct pid              *shm_lprid;
62         struct user_struct      *mlock_user;
63
64         /* The task created the shm object.  NULL if the task is dead. */
65         struct task_struct      *shm_creator;
66         struct list_head        shm_clist;      /* list by creator */
67 } __randomize_layout;
68
69 /* shm_mode upper byte flags */
70 #define SHM_DEST        01000   /* segment will be destroyed on last detach */
71 #define SHM_LOCKED      02000   /* segment will not be swapped */
72
73 struct shm_file_data {
74         int id;
75         struct ipc_namespace *ns;
76         struct file *file;
77         const struct vm_operations_struct *vm_ops;
78 };
79
80 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
81
82 static const struct file_operations shm_file_operations;
83 static const struct vm_operations_struct shm_vm_ops;
84
85 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
86
87 #define shm_unlock(shp)                 \
88         ipc_unlock(&(shp)->shm_perm)
89
90 static int newseg(struct ipc_namespace *, struct ipc_params *);
91 static void shm_open(struct vm_area_struct *vma);
92 static void shm_close(struct vm_area_struct *vma);
93 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp);
94 #ifdef CONFIG_PROC_FS
95 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
96 #endif
97
98 int shm_init_ns(struct ipc_namespace *ns)
99 {
100         ns->shm_ctlmax = SHMMAX;
101         ns->shm_ctlall = SHMALL;
102         ns->shm_ctlmni = SHMMNI;
103         ns->shm_rmid_forced = 0;
104         ns->shm_tot = 0;
105         return ipc_init_ids(&shm_ids(ns));
106 }
107
108 /*
109  * Called with shm_ids.rwsem (writer) and the shp structure locked.
110  * Only shm_ids.rwsem remains locked on exit.
111  */
112 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
113 {
114         struct shmid_kernel *shp;
115
116         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
117
118         if (shp->shm_nattch) {
119                 shp->shm_perm.mode |= SHM_DEST;
120                 /* Do not find it any more */
121                 ipc_set_key_private(&shm_ids(ns), &shp->shm_perm);
122                 shm_unlock(shp);
123         } else
124                 shm_destroy(ns, shp);
125 }
126
127 #ifdef CONFIG_IPC_NS
128 void shm_exit_ns(struct ipc_namespace *ns)
129 {
130         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
131         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
132         rhashtable_destroy(&ns->ids[IPC_SHM_IDS].key_ht);
133 }
134 #endif
135
136 static int __init ipc_ns_init(void)
137 {
138         const int err = shm_init_ns(&init_ipc_ns);
139         WARN(err, "ipc: sysv shm_init_ns failed: %d\n", err);
140         return err;
141 }
142
143 pure_initcall(ipc_ns_init);
144
145 void __init shm_init(void)
146 {
147         ipc_init_proc_interface("sysvipc/shm",
148 #if BITS_PER_LONG <= 32
149                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
150 #else
151                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
152 #endif
153                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
154 }
155
156 static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
157 {
158         struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&shm_ids(ns), id);
159
160         if (IS_ERR(ipcp))
161                 return ERR_CAST(ipcp);
162
163         return container_of(ipcp, struct shmid_kernel, shm_perm);
164 }
165
166 static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
167 {
168         struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
169
170         if (IS_ERR(ipcp))
171                 return ERR_CAST(ipcp);
172
173         return container_of(ipcp, struct shmid_kernel, shm_perm);
174 }
175
176 /*
177  * shm_lock_(check_) routines are called in the paths where the rwsem
178  * is not necessarily held.
179  */
180 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
181 {
182         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
183
184         /*
185          * Callers of shm_lock() must validate the status of the returned ipc
186          * object pointer (as returned by ipc_lock()), and error out as
187          * appropriate.
188          */
189         if (IS_ERR(ipcp))
190                 return (void *)ipcp;
191         return container_of(ipcp, struct shmid_kernel, shm_perm);
192 }
193
194 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
195 {
196         rcu_read_lock();
197         ipc_lock_object(&ipcp->shm_perm);
198 }
199
200 static void shm_rcu_free(struct rcu_head *head)
201 {
202         struct kern_ipc_perm *ptr = container_of(head, struct kern_ipc_perm,
203                                                         rcu);
204         struct shmid_kernel *shp = container_of(ptr, struct shmid_kernel,
205                                                         shm_perm);
206         security_shm_free(&shp->shm_perm);
207         kvfree(shp);
208 }
209
210 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
211 {
212         list_del(&s->shm_clist);
213         ipc_rmid(&shm_ids(ns), &s->shm_perm);
214 }
215
216
217 static int __shm_open(struct vm_area_struct *vma)
218 {
219         struct file *file = vma->vm_file;
220         struct shm_file_data *sfd = shm_file_data(file);
221         struct shmid_kernel *shp;
222
223         shp = shm_lock(sfd->ns, sfd->id);
224
225         if (IS_ERR(shp))
226                 return PTR_ERR(shp);
227
228         shp->shm_atim = ktime_get_real_seconds();
229         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
230         shp->shm_nattch++;
231         shm_unlock(shp);
232         return 0;
233 }
234
235 /* This is called by fork, once for every shm attach. */
236 static void shm_open(struct vm_area_struct *vma)
237 {
238         int err = __shm_open(vma);
239         /*
240          * We raced in the idr lookup or with shm_destroy().
241          * Either way, the ID is busted.
242          */
243         WARN_ON_ONCE(err);
244 }
245
246 /*
247  * shm_destroy - free the struct shmid_kernel
248  *
249  * @ns: namespace
250  * @shp: struct to free
251  *
252  * It has to be called with shp and shm_ids.rwsem (writer) locked,
253  * but returns with shp unlocked and freed.
254  */
255 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
256 {
257         struct file *shm_file;
258
259         shm_file = shp->shm_file;
260         shp->shm_file = NULL;
261         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
262         shm_rmid(ns, shp);
263         shm_unlock(shp);
264         if (!is_file_hugepages(shm_file))
265                 shmem_lock(shm_file, 0, shp->mlock_user);
266         else if (shp->mlock_user)
267                 user_shm_unlock(i_size_read(file_inode(shm_file)),
268                                 shp->mlock_user);
269         fput(shm_file);
270         ipc_update_pid(&shp->shm_cprid, NULL);
271         ipc_update_pid(&shp->shm_lprid, NULL);
272         ipc_rcu_putref(&shp->shm_perm, shm_rcu_free);
273 }
274
275 /*
276  * shm_may_destroy - identifies whether shm segment should be destroyed now
277  *
278  * Returns true if and only if there are no active users of the segment and
279  * one of the following is true:
280  *
281  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
282  *
283  * 2) sysctl kernel.shm_rmid_forced is set to 1.
284  */
285 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
286 {
287         return (shp->shm_nattch == 0) &&
288                (ns->shm_rmid_forced ||
289                 (shp->shm_perm.mode & SHM_DEST));
290 }
291
292 /*
293  * remove the attach descriptor vma.
294  * free memory for segment if it is marked destroyed.
295  * The descriptor has already been removed from the current->mm->mmap list
296  * and will later be kfree()d.
297  */
298 static void shm_close(struct vm_area_struct *vma)
299 {
300         struct file *file = vma->vm_file;
301         struct shm_file_data *sfd = shm_file_data(file);
302         struct shmid_kernel *shp;
303         struct ipc_namespace *ns = sfd->ns;
304
305         down_write(&shm_ids(ns).rwsem);
306         /* remove from the list of attaches of the shm segment */
307         shp = shm_lock(ns, sfd->id);
308
309         /*
310          * We raced in the idr lookup or with shm_destroy().
311          * Either way, the ID is busted.
312          */
313         if (WARN_ON_ONCE(IS_ERR(shp)))
314                 goto done; /* no-op */
315
316         ipc_update_pid(&shp->shm_lprid, task_tgid(current));
317         shp->shm_dtim = ktime_get_real_seconds();
318         shp->shm_nattch--;
319         if (shm_may_destroy(ns, shp))
320                 shm_destroy(ns, shp);
321         else
322                 shm_unlock(shp);
323 done:
324         up_write(&shm_ids(ns).rwsem);
325 }
326
327 /* Called with ns->shm_ids(ns).rwsem locked */
328 static int shm_try_destroy_orphaned(int id, void *p, void *data)
329 {
330         struct ipc_namespace *ns = data;
331         struct kern_ipc_perm *ipcp = p;
332         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
333
334         /*
335          * We want to destroy segments without users and with already
336          * exit'ed originating process.
337          *
338          * As shp->* are changed under rwsem, it's safe to skip shp locking.
339          */
340         if (shp->shm_creator != NULL)
341                 return 0;
342
343         if (shm_may_destroy(ns, shp)) {
344                 shm_lock_by_ptr(shp);
345                 shm_destroy(ns, shp);
346         }
347         return 0;
348 }
349
350 void shm_destroy_orphaned(struct ipc_namespace *ns)
351 {
352         down_write(&shm_ids(ns).rwsem);
353         if (shm_ids(ns).in_use)
354                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
355         up_write(&shm_ids(ns).rwsem);
356 }
357
358 /* Locking assumes this will only be called with task == current */
359 void exit_shm(struct task_struct *task)
360 {
361         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
362         struct shmid_kernel *shp, *n;
363
364         if (list_empty(&task->sysvshm.shm_clist))
365                 return;
366
367         /*
368          * If kernel.shm_rmid_forced is not set then only keep track of
369          * which shmids are orphaned, so that a later set of the sysctl
370          * can clean them up.
371          */
372         if (!ns->shm_rmid_forced) {
373                 down_read(&shm_ids(ns).rwsem);
374                 list_for_each_entry(shp, &task->sysvshm.shm_clist, shm_clist)
375                         shp->shm_creator = NULL;
376                 /*
377                  * Only under read lock but we are only called on current
378                  * so no entry on the list will be shared.
379                  */
380                 list_del(&task->sysvshm.shm_clist);
381                 up_read(&shm_ids(ns).rwsem);
382                 return;
383         }
384
385         /*
386          * Destroy all already created segments, that were not yet mapped,
387          * and mark any mapped as orphan to cover the sysctl toggling.
388          * Destroy is skipped if shm_may_destroy() returns false.
389          */
390         down_write(&shm_ids(ns).rwsem);
391         list_for_each_entry_safe(shp, n, &task->sysvshm.shm_clist, shm_clist) {
392                 shp->shm_creator = NULL;
393
394                 if (shm_may_destroy(ns, shp)) {
395                         shm_lock_by_ptr(shp);
396                         shm_destroy(ns, shp);
397                 }
398         }
399
400         /* Remove the list head from any segments still attached. */
401         list_del(&task->sysvshm.shm_clist);
402         up_write(&shm_ids(ns).rwsem);
403 }
404
405 static int shm_fault(struct vm_fault *vmf)
406 {
407         struct file *file = vmf->vma->vm_file;
408         struct shm_file_data *sfd = shm_file_data(file);
409
410         return sfd->vm_ops->fault(vmf);
411 }
412
413 static int shm_split(struct vm_area_struct *vma, unsigned long addr)
414 {
415         struct file *file = vma->vm_file;
416         struct shm_file_data *sfd = shm_file_data(file);
417
418         if (sfd->vm_ops && sfd->vm_ops->split)
419                 return sfd->vm_ops->split(vma, addr);
420
421         return 0;
422 }
423
424 #ifdef CONFIG_NUMA
425 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
426 {
427         struct file *file = vma->vm_file;
428         struct shm_file_data *sfd = shm_file_data(file);
429         int err = 0;
430
431         if (sfd->vm_ops->set_policy)
432                 err = sfd->vm_ops->set_policy(vma, new);
433         return err;
434 }
435
436 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
437                                         unsigned long addr)
438 {
439         struct file *file = vma->vm_file;
440         struct shm_file_data *sfd = shm_file_data(file);
441         struct mempolicy *pol = NULL;
442
443         if (sfd->vm_ops->get_policy)
444                 pol = sfd->vm_ops->get_policy(vma, addr);
445         else if (vma->vm_policy)
446                 pol = vma->vm_policy;
447
448         return pol;
449 }
450 #endif
451
452 static int shm_mmap(struct file *file, struct vm_area_struct *vma)
453 {
454         struct shm_file_data *sfd = shm_file_data(file);
455         int ret;
456
457         /*
458          * In case of remap_file_pages() emulation, the file can represent
459          * removed IPC ID: propogate shm_lock() error to caller.
460          */
461         ret = __shm_open(vma);
462         if (ret)
463                 return ret;
464
465         ret = call_mmap(sfd->file, vma);
466         if (ret) {
467                 shm_close(vma);
468                 return ret;
469         }
470         sfd->vm_ops = vma->vm_ops;
471 #ifdef CONFIG_MMU
472         WARN_ON(!sfd->vm_ops->fault);
473 #endif
474         vma->vm_ops = &shm_vm_ops;
475         return 0;
476 }
477
478 static int shm_release(struct inode *ino, struct file *file)
479 {
480         struct shm_file_data *sfd = shm_file_data(file);
481
482         put_ipc_ns(sfd->ns);
483         shm_file_data(file) = NULL;
484         kfree(sfd);
485         return 0;
486 }
487
488 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
489 {
490         struct shm_file_data *sfd = shm_file_data(file);
491
492         if (!sfd->file->f_op->fsync)
493                 return -EINVAL;
494         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
495 }
496
497 static long shm_fallocate(struct file *file, int mode, loff_t offset,
498                           loff_t len)
499 {
500         struct shm_file_data *sfd = shm_file_data(file);
501
502         if (!sfd->file->f_op->fallocate)
503                 return -EOPNOTSUPP;
504         return sfd->file->f_op->fallocate(file, mode, offset, len);
505 }
506
507 static unsigned long shm_get_unmapped_area(struct file *file,
508         unsigned long addr, unsigned long len, unsigned long pgoff,
509         unsigned long flags)
510 {
511         struct shm_file_data *sfd = shm_file_data(file);
512
513         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
514                                                 pgoff, flags);
515 }
516
517 static const struct file_operations shm_file_operations = {
518         .mmap           = shm_mmap,
519         .fsync          = shm_fsync,
520         .release        = shm_release,
521         .get_unmapped_area      = shm_get_unmapped_area,
522         .llseek         = noop_llseek,
523         .fallocate      = shm_fallocate,
524 };
525
526 /*
527  * shm_file_operations_huge is now identical to shm_file_operations,
528  * but we keep it distinct for the sake of is_file_shm_hugepages().
529  */
530 static const struct file_operations shm_file_operations_huge = {
531         .mmap           = shm_mmap,
532         .fsync          = shm_fsync,
533         .release        = shm_release,
534         .get_unmapped_area      = shm_get_unmapped_area,
535         .llseek         = noop_llseek,
536         .fallocate      = shm_fallocate,
537 };
538
539 bool is_file_shm_hugepages(struct file *file)
540 {
541         return file->f_op == &shm_file_operations_huge;
542 }
543
544 static const struct vm_operations_struct shm_vm_ops = {
545         .open   = shm_open,     /* callback for a new vm-area open */
546         .close  = shm_close,    /* callback for when the vm-area is released */
547         .fault  = shm_fault,
548         .split  = shm_split,
549 #if defined(CONFIG_NUMA)
550         .set_policy = shm_set_policy,
551         .get_policy = shm_get_policy,
552 #endif
553 };
554
555 /**
556  * newseg - Create a new shared memory segment
557  * @ns: namespace
558  * @params: ptr to the structure that contains key, size and shmflg
559  *
560  * Called with shm_ids.rwsem held as a writer.
561  */
562 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
563 {
564         key_t key = params->key;
565         int shmflg = params->flg;
566         size_t size = params->u.size;
567         int error;
568         struct shmid_kernel *shp;
569         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
570         struct file *file;
571         char name[13];
572         vm_flags_t acctflag = 0;
573
574         if (size < SHMMIN || size > ns->shm_ctlmax)
575                 return -EINVAL;
576
577         if (numpages << PAGE_SHIFT < size)
578                 return -ENOSPC;
579
580         if (ns->shm_tot + numpages < ns->shm_tot ||
581                         ns->shm_tot + numpages > ns->shm_ctlall)
582                 return -ENOSPC;
583
584         shp = kvmalloc(sizeof(*shp), GFP_KERNEL);
585         if (unlikely(!shp))
586                 return -ENOMEM;
587
588         shp->shm_perm.key = key;
589         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
590         shp->mlock_user = NULL;
591
592         shp->shm_perm.security = NULL;
593         error = security_shm_alloc(&shp->shm_perm);
594         if (error) {
595                 kvfree(shp);
596                 return error;
597         }
598
599         sprintf(name, "SYSV%08x", key);
600         if (shmflg & SHM_HUGETLB) {
601                 struct hstate *hs;
602                 size_t hugesize;
603
604                 hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
605                 if (!hs) {
606                         error = -EINVAL;
607                         goto no_file;
608                 }
609                 hugesize = ALIGN(size, huge_page_size(hs));
610
611                 /* hugetlb_file_setup applies strict accounting */
612                 if (shmflg & SHM_NORESERVE)
613                         acctflag = VM_NORESERVE;
614                 file = hugetlb_file_setup(name, hugesize, acctflag,
615                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
616                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
617         } else {
618                 /*
619                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
620                  * if it's asked for.
621                  */
622                 if  ((shmflg & SHM_NORESERVE) &&
623                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
624                         acctflag = VM_NORESERVE;
625                 file = shmem_kernel_file_setup(name, size, acctflag);
626         }
627         error = PTR_ERR(file);
628         if (IS_ERR(file))
629                 goto no_file;
630
631         shp->shm_cprid = get_pid(task_tgid(current));
632         shp->shm_lprid = NULL;
633         shp->shm_atim = shp->shm_dtim = 0;
634         shp->shm_ctim = ktime_get_real_seconds();
635         shp->shm_segsz = size;
636         shp->shm_nattch = 0;
637         shp->shm_file = file;
638         shp->shm_creator = current;
639
640         /* ipc_addid() locks shp upon success. */
641         error = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
642         if (error < 0)
643                 goto no_id;
644
645         list_add(&shp->shm_clist, &current->sysvshm.shm_clist);
646
647         /*
648          * shmid gets reported as "inode#" in /proc/pid/maps.
649          * proc-ps tools use this. Changing this will break them.
650          */
651         file_inode(file)->i_ino = shp->shm_perm.id;
652
653         ns->shm_tot += numpages;
654         error = shp->shm_perm.id;
655
656         ipc_unlock_object(&shp->shm_perm);
657         rcu_read_unlock();
658         return error;
659
660 no_id:
661         ipc_update_pid(&shp->shm_cprid, NULL);
662         ipc_update_pid(&shp->shm_lprid, NULL);
663         if (is_file_hugepages(file) && shp->mlock_user)
664                 user_shm_unlock(size, shp->mlock_user);
665         fput(file);
666 no_file:
667         call_rcu(&shp->shm_perm.rcu, shm_rcu_free);
668         return error;
669 }
670
671 /*
672  * Called with shm_ids.rwsem and ipcp locked.
673  */
674 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
675                                 struct ipc_params *params)
676 {
677         struct shmid_kernel *shp;
678
679         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
680         if (shp->shm_segsz < params->u.size)
681                 return -EINVAL;
682
683         return 0;
684 }
685
686 long ksys_shmget(key_t key, size_t size, int shmflg)
687 {
688         struct ipc_namespace *ns;
689         static const struct ipc_ops shm_ops = {
690                 .getnew = newseg,
691                 .associate = security_shm_associate,
692                 .more_checks = shm_more_checks,
693         };
694         struct ipc_params shm_params;
695
696         ns = current->nsproxy->ipc_ns;
697
698         shm_params.key = key;
699         shm_params.flg = shmflg;
700         shm_params.u.size = size;
701
702         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
703 }
704
705 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
706 {
707         return ksys_shmget(key, size, shmflg);
708 }
709
710 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
711 {
712         switch (version) {
713         case IPC_64:
714                 return copy_to_user(buf, in, sizeof(*in));
715         case IPC_OLD:
716             {
717                 struct shmid_ds out;
718
719                 memset(&out, 0, sizeof(out));
720                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
721                 out.shm_segsz   = in->shm_segsz;
722                 out.shm_atime   = in->shm_atime;
723                 out.shm_dtime   = in->shm_dtime;
724                 out.shm_ctime   = in->shm_ctime;
725                 out.shm_cpid    = in->shm_cpid;
726                 out.shm_lpid    = in->shm_lpid;
727                 out.shm_nattch  = in->shm_nattch;
728
729                 return copy_to_user(buf, &out, sizeof(out));
730             }
731         default:
732                 return -EINVAL;
733         }
734 }
735
736 static inline unsigned long
737 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
738 {
739         switch (version) {
740         case IPC_64:
741                 if (copy_from_user(out, buf, sizeof(*out)))
742                         return -EFAULT;
743                 return 0;
744         case IPC_OLD:
745             {
746                 struct shmid_ds tbuf_old;
747
748                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
749                         return -EFAULT;
750
751                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
752                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
753                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
754
755                 return 0;
756             }
757         default:
758                 return -EINVAL;
759         }
760 }
761
762 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
763 {
764         switch (version) {
765         case IPC_64:
766                 return copy_to_user(buf, in, sizeof(*in));
767         case IPC_OLD:
768             {
769                 struct shminfo out;
770
771                 if (in->shmmax > INT_MAX)
772                         out.shmmax = INT_MAX;
773                 else
774                         out.shmmax = (int)in->shmmax;
775
776                 out.shmmin      = in->shmmin;
777                 out.shmmni      = in->shmmni;
778                 out.shmseg      = in->shmseg;
779                 out.shmall      = in->shmall;
780
781                 return copy_to_user(buf, &out, sizeof(out));
782             }
783         default:
784                 return -EINVAL;
785         }
786 }
787
788 /*
789  * Calculate and add used RSS and swap pages of a shm.
790  * Called with shm_ids.rwsem held as a reader
791  */
792 static void shm_add_rss_swap(struct shmid_kernel *shp,
793         unsigned long *rss_add, unsigned long *swp_add)
794 {
795         struct inode *inode;
796
797         inode = file_inode(shp->shm_file);
798
799         if (is_file_hugepages(shp->shm_file)) {
800                 struct address_space *mapping = inode->i_mapping;
801                 struct hstate *h = hstate_file(shp->shm_file);
802                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
803         } else {
804 #ifdef CONFIG_SHMEM
805                 struct shmem_inode_info *info = SHMEM_I(inode);
806
807                 spin_lock_irq(&info->lock);
808                 *rss_add += inode->i_mapping->nrpages;
809                 *swp_add += info->swapped;
810                 spin_unlock_irq(&info->lock);
811 #else
812                 *rss_add += inode->i_mapping->nrpages;
813 #endif
814         }
815 }
816
817 /*
818  * Called with shm_ids.rwsem held as a reader
819  */
820 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
821                 unsigned long *swp)
822 {
823         int next_id;
824         int total, in_use;
825
826         *rss = 0;
827         *swp = 0;
828
829         in_use = shm_ids(ns).in_use;
830
831         for (total = 0, next_id = 0; total < in_use; next_id++) {
832                 struct kern_ipc_perm *ipc;
833                 struct shmid_kernel *shp;
834
835                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
836                 if (ipc == NULL)
837                         continue;
838                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
839
840                 shm_add_rss_swap(shp, rss, swp);
841
842                 total++;
843         }
844 }
845
846 /*
847  * This function handles some shmctl commands which require the rwsem
848  * to be held in write mode.
849  * NOTE: no locks must be held, the rwsem is taken inside this function.
850  */
851 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
852                        struct shmid64_ds *shmid64)
853 {
854         struct kern_ipc_perm *ipcp;
855         struct shmid_kernel *shp;
856         int err;
857
858         down_write(&shm_ids(ns).rwsem);
859         rcu_read_lock();
860
861         ipcp = ipcctl_pre_down_nolock(ns, &shm_ids(ns), shmid, cmd,
862                                       &shmid64->shm_perm, 0);
863         if (IS_ERR(ipcp)) {
864                 err = PTR_ERR(ipcp);
865                 goto out_unlock1;
866         }
867
868         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
869
870         err = security_shm_shmctl(&shp->shm_perm, cmd);
871         if (err)
872                 goto out_unlock1;
873
874         switch (cmd) {
875         case IPC_RMID:
876                 ipc_lock_object(&shp->shm_perm);
877                 /* do_shm_rmid unlocks the ipc object and rcu */
878                 do_shm_rmid(ns, ipcp);
879                 goto out_up;
880         case IPC_SET:
881                 ipc_lock_object(&shp->shm_perm);
882                 err = ipc_update_perm(&shmid64->shm_perm, ipcp);
883                 if (err)
884                         goto out_unlock0;
885                 shp->shm_ctim = ktime_get_real_seconds();
886                 break;
887         default:
888                 err = -EINVAL;
889                 goto out_unlock1;
890         }
891
892 out_unlock0:
893         ipc_unlock_object(&shp->shm_perm);
894 out_unlock1:
895         rcu_read_unlock();
896 out_up:
897         up_write(&shm_ids(ns).rwsem);
898         return err;
899 }
900
901 static int shmctl_ipc_info(struct ipc_namespace *ns,
902                            struct shminfo64 *shminfo)
903 {
904         int err = security_shm_shmctl(NULL, IPC_INFO);
905         if (!err) {
906                 memset(shminfo, 0, sizeof(*shminfo));
907                 shminfo->shmmni = shminfo->shmseg = ns->shm_ctlmni;
908                 shminfo->shmmax = ns->shm_ctlmax;
909                 shminfo->shmall = ns->shm_ctlall;
910                 shminfo->shmmin = SHMMIN;
911                 down_read(&shm_ids(ns).rwsem);
912                 err = ipc_get_maxid(&shm_ids(ns));
913                 up_read(&shm_ids(ns).rwsem);
914                 if (err < 0)
915                         err = 0;
916         }
917         return err;
918 }
919
920 static int shmctl_shm_info(struct ipc_namespace *ns,
921                            struct shm_info *shm_info)
922 {
923         int err = security_shm_shmctl(NULL, SHM_INFO);
924         if (!err) {
925                 memset(shm_info, 0, sizeof(*shm_info));
926                 down_read(&shm_ids(ns).rwsem);
927                 shm_info->used_ids = shm_ids(ns).in_use;
928                 shm_get_stat(ns, &shm_info->shm_rss, &shm_info->shm_swp);
929                 shm_info->shm_tot = ns->shm_tot;
930                 shm_info->swap_attempts = 0;
931                 shm_info->swap_successes = 0;
932                 err = ipc_get_maxid(&shm_ids(ns));
933                 up_read(&shm_ids(ns).rwsem);
934                 if (err < 0)
935                         err = 0;
936         }
937         return err;
938 }
939
940 static int shmctl_stat(struct ipc_namespace *ns, int shmid,
941                         int cmd, struct shmid64_ds *tbuf)
942 {
943         struct shmid_kernel *shp;
944         int id = 0;
945         int err;
946
947         memset(tbuf, 0, sizeof(*tbuf));
948
949         rcu_read_lock();
950         if (cmd == SHM_STAT) {
951                 shp = shm_obtain_object(ns, shmid);
952                 if (IS_ERR(shp)) {
953                         err = PTR_ERR(shp);
954                         goto out_unlock;
955                 }
956                 id = shp->shm_perm.id;
957         } else {
958                 shp = shm_obtain_object_check(ns, shmid);
959                 if (IS_ERR(shp)) {
960                         err = PTR_ERR(shp);
961                         goto out_unlock;
962                 }
963         }
964
965         err = -EACCES;
966         if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
967                 goto out_unlock;
968
969         err = security_shm_shmctl(&shp->shm_perm, cmd);
970         if (err)
971                 goto out_unlock;
972
973         ipc_lock_object(&shp->shm_perm);
974
975         if (!ipc_valid_object(&shp->shm_perm)) {
976                 ipc_unlock_object(&shp->shm_perm);
977                 err = -EIDRM;
978                 goto out_unlock;
979         }
980
981         kernel_to_ipc64_perm(&shp->shm_perm, &tbuf->shm_perm);
982         tbuf->shm_segsz = shp->shm_segsz;
983         tbuf->shm_atime = shp->shm_atim;
984         tbuf->shm_dtime = shp->shm_dtim;
985         tbuf->shm_ctime = shp->shm_ctim;
986         tbuf->shm_cpid  = pid_vnr(shp->shm_cprid);
987         tbuf->shm_lpid  = pid_vnr(shp->shm_lprid);
988         tbuf->shm_nattch = shp->shm_nattch;
989
990         ipc_unlock_object(&shp->shm_perm);
991         rcu_read_unlock();
992         return id;
993
994 out_unlock:
995         rcu_read_unlock();
996         return err;
997 }
998
999 static int shmctl_do_lock(struct ipc_namespace *ns, int shmid, int cmd)
1000 {
1001         struct shmid_kernel *shp;
1002         struct file *shm_file;
1003         int err;
1004
1005         rcu_read_lock();
1006         shp = shm_obtain_object_check(ns, shmid);
1007         if (IS_ERR(shp)) {
1008                 err = PTR_ERR(shp);
1009                 goto out_unlock1;
1010         }
1011
1012         audit_ipc_obj(&(shp->shm_perm));
1013         err = security_shm_shmctl(&shp->shm_perm, cmd);
1014         if (err)
1015                 goto out_unlock1;
1016
1017         ipc_lock_object(&shp->shm_perm);
1018
1019         /* check if shm_destroy() is tearing down shp */
1020         if (!ipc_valid_object(&shp->shm_perm)) {
1021                 err = -EIDRM;
1022                 goto out_unlock0;
1023         }
1024
1025         if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
1026                 kuid_t euid = current_euid();
1027
1028                 if (!uid_eq(euid, shp->shm_perm.uid) &&
1029                     !uid_eq(euid, shp->shm_perm.cuid)) {
1030                         err = -EPERM;
1031                         goto out_unlock0;
1032                 }
1033                 if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK)) {
1034                         err = -EPERM;
1035                         goto out_unlock0;
1036                 }
1037         }
1038
1039         shm_file = shp->shm_file;
1040         if (is_file_hugepages(shm_file))
1041                 goto out_unlock0;
1042
1043         if (cmd == SHM_LOCK) {
1044                 struct user_struct *user = current_user();
1045
1046                 err = shmem_lock(shm_file, 1, user);
1047                 if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
1048                         shp->shm_perm.mode |= SHM_LOCKED;
1049                         shp->mlock_user = user;
1050                 }
1051                 goto out_unlock0;
1052         }
1053
1054         /* SHM_UNLOCK */
1055         if (!(shp->shm_perm.mode & SHM_LOCKED))
1056                 goto out_unlock0;
1057         shmem_lock(shm_file, 0, shp->mlock_user);
1058         shp->shm_perm.mode &= ~SHM_LOCKED;
1059         shp->mlock_user = NULL;
1060         get_file(shm_file);
1061         ipc_unlock_object(&shp->shm_perm);
1062         rcu_read_unlock();
1063         shmem_unlock_mapping(shm_file->f_mapping);
1064
1065         fput(shm_file);
1066         return err;
1067
1068 out_unlock0:
1069         ipc_unlock_object(&shp->shm_perm);
1070 out_unlock1:
1071         rcu_read_unlock();
1072         return err;
1073 }
1074
1075 long ksys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
1076 {
1077         int err, version;
1078         struct ipc_namespace *ns;
1079         struct shmid64_ds sem64;
1080
1081         if (cmd < 0 || shmid < 0)
1082                 return -EINVAL;
1083
1084         version = ipc_parse_version(&cmd);
1085         ns = current->nsproxy->ipc_ns;
1086
1087         switch (cmd) {
1088         case IPC_INFO: {
1089                 struct shminfo64 shminfo;
1090                 err = shmctl_ipc_info(ns, &shminfo);
1091                 if (err < 0)
1092                         return err;
1093                 if (copy_shminfo_to_user(buf, &shminfo, version))
1094                         err = -EFAULT;
1095                 return err;
1096         }
1097         case SHM_INFO: {
1098                 struct shm_info shm_info;
1099                 err = shmctl_shm_info(ns, &shm_info);
1100                 if (err < 0)
1101                         return err;
1102                 if (copy_to_user(buf, &shm_info, sizeof(shm_info)))
1103                         err = -EFAULT;
1104                 return err;
1105         }
1106         case SHM_STAT:
1107         case IPC_STAT: {
1108                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1109                 if (err < 0)
1110                         return err;
1111                 if (copy_shmid_to_user(buf, &sem64, version))
1112                         err = -EFAULT;
1113                 return err;
1114         }
1115         case IPC_SET:
1116                 if (copy_shmid_from_user(&sem64, buf, version))
1117                         return -EFAULT;
1118                 /* fallthru */
1119         case IPC_RMID:
1120                 return shmctl_down(ns, shmid, cmd, &sem64);
1121         case SHM_LOCK:
1122         case SHM_UNLOCK:
1123                 return shmctl_do_lock(ns, shmid, cmd);
1124         default:
1125                 return -EINVAL;
1126         }
1127 }
1128
1129 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
1130 {
1131         return ksys_shmctl(shmid, cmd, buf);
1132 }
1133
1134 #ifdef CONFIG_COMPAT
1135
1136 struct compat_shmid_ds {
1137         struct compat_ipc_perm shm_perm;
1138         int shm_segsz;
1139         compat_time_t shm_atime;
1140         compat_time_t shm_dtime;
1141         compat_time_t shm_ctime;
1142         compat_ipc_pid_t shm_cpid;
1143         compat_ipc_pid_t shm_lpid;
1144         unsigned short shm_nattch;
1145         unsigned short shm_unused;
1146         compat_uptr_t shm_unused2;
1147         compat_uptr_t shm_unused3;
1148 };
1149
1150 struct compat_shminfo64 {
1151         compat_ulong_t shmmax;
1152         compat_ulong_t shmmin;
1153         compat_ulong_t shmmni;
1154         compat_ulong_t shmseg;
1155         compat_ulong_t shmall;
1156         compat_ulong_t __unused1;
1157         compat_ulong_t __unused2;
1158         compat_ulong_t __unused3;
1159         compat_ulong_t __unused4;
1160 };
1161
1162 struct compat_shm_info {
1163         compat_int_t used_ids;
1164         compat_ulong_t shm_tot, shm_rss, shm_swp;
1165         compat_ulong_t swap_attempts, swap_successes;
1166 };
1167
1168 static int copy_compat_shminfo_to_user(void __user *buf, struct shminfo64 *in,
1169                                         int version)
1170 {
1171         if (in->shmmax > INT_MAX)
1172                 in->shmmax = INT_MAX;
1173         if (version == IPC_64) {
1174                 struct compat_shminfo64 info;
1175                 memset(&info, 0, sizeof(info));
1176                 info.shmmax = in->shmmax;
1177                 info.shmmin = in->shmmin;
1178                 info.shmmni = in->shmmni;
1179                 info.shmseg = in->shmseg;
1180                 info.shmall = in->shmall;
1181                 return copy_to_user(buf, &info, sizeof(info));
1182         } else {
1183                 struct shminfo info;
1184                 memset(&info, 0, sizeof(info));
1185                 info.shmmax = in->shmmax;
1186                 info.shmmin = in->shmmin;
1187                 info.shmmni = in->shmmni;
1188                 info.shmseg = in->shmseg;
1189                 info.shmall = in->shmall;
1190                 return copy_to_user(buf, &info, sizeof(info));
1191         }
1192 }
1193
1194 static int put_compat_shm_info(struct shm_info *ip,
1195                                 struct compat_shm_info __user *uip)
1196 {
1197         struct compat_shm_info info;
1198
1199         memset(&info, 0, sizeof(info));
1200         info.used_ids = ip->used_ids;
1201         info.shm_tot = ip->shm_tot;
1202         info.shm_rss = ip->shm_rss;
1203         info.shm_swp = ip->shm_swp;
1204         info.swap_attempts = ip->swap_attempts;
1205         info.swap_successes = ip->swap_successes;
1206         return copy_to_user(uip, &info, sizeof(info));
1207 }
1208
1209 static int copy_compat_shmid_to_user(void __user *buf, struct shmid64_ds *in,
1210                                         int version)
1211 {
1212         if (version == IPC_64) {
1213                 struct compat_shmid64_ds v;
1214                 memset(&v, 0, sizeof(v));
1215                 to_compat_ipc64_perm(&v.shm_perm, &in->shm_perm);
1216                 v.shm_atime = in->shm_atime;
1217                 v.shm_dtime = in->shm_dtime;
1218                 v.shm_ctime = in->shm_ctime;
1219                 v.shm_segsz = in->shm_segsz;
1220                 v.shm_nattch = in->shm_nattch;
1221                 v.shm_cpid = in->shm_cpid;
1222                 v.shm_lpid = in->shm_lpid;
1223                 return copy_to_user(buf, &v, sizeof(v));
1224         } else {
1225                 struct compat_shmid_ds v;
1226                 memset(&v, 0, sizeof(v));
1227                 to_compat_ipc_perm(&v.shm_perm, &in->shm_perm);
1228                 v.shm_perm.key = in->shm_perm.key;
1229                 v.shm_atime = in->shm_atime;
1230                 v.shm_dtime = in->shm_dtime;
1231                 v.shm_ctime = in->shm_ctime;
1232                 v.shm_segsz = in->shm_segsz;
1233                 v.shm_nattch = in->shm_nattch;
1234                 v.shm_cpid = in->shm_cpid;
1235                 v.shm_lpid = in->shm_lpid;
1236                 return copy_to_user(buf, &v, sizeof(v));
1237         }
1238 }
1239
1240 static int copy_compat_shmid_from_user(struct shmid64_ds *out, void __user *buf,
1241                                         int version)
1242 {
1243         memset(out, 0, sizeof(*out));
1244         if (version == IPC_64) {
1245                 struct compat_shmid64_ds __user *p = buf;
1246                 return get_compat_ipc64_perm(&out->shm_perm, &p->shm_perm);
1247         } else {
1248                 struct compat_shmid_ds __user *p = buf;
1249                 return get_compat_ipc_perm(&out->shm_perm, &p->shm_perm);
1250         }
1251 }
1252
1253 long compat_ksys_shmctl(int shmid, int cmd, void __user *uptr)
1254 {
1255         struct ipc_namespace *ns;
1256         struct shmid64_ds sem64;
1257         int version = compat_ipc_parse_version(&cmd);
1258         int err;
1259
1260         ns = current->nsproxy->ipc_ns;
1261
1262         if (cmd < 0 || shmid < 0)
1263                 return -EINVAL;
1264
1265         switch (cmd) {
1266         case IPC_INFO: {
1267                 struct shminfo64 shminfo;
1268                 err = shmctl_ipc_info(ns, &shminfo);
1269                 if (err < 0)
1270                         return err;
1271                 if (copy_compat_shminfo_to_user(uptr, &shminfo, version))
1272                         err = -EFAULT;
1273                 return err;
1274         }
1275         case SHM_INFO: {
1276                 struct shm_info shm_info;
1277                 err = shmctl_shm_info(ns, &shm_info);
1278                 if (err < 0)
1279                         return err;
1280                 if (put_compat_shm_info(&shm_info, uptr))
1281                         err = -EFAULT;
1282                 return err;
1283         }
1284         case IPC_STAT:
1285         case SHM_STAT:
1286                 err = shmctl_stat(ns, shmid, cmd, &sem64);
1287                 if (err < 0)
1288                         return err;
1289                 if (copy_compat_shmid_to_user(uptr, &sem64, version))
1290                         err = -EFAULT;
1291                 return err;
1292
1293         case IPC_SET:
1294                 if (copy_compat_shmid_from_user(&sem64, uptr, version))
1295                         return -EFAULT;
1296                 /* fallthru */
1297         case IPC_RMID:
1298                 return shmctl_down(ns, shmid, cmd, &sem64);
1299         case SHM_LOCK:
1300         case SHM_UNLOCK:
1301                 return shmctl_do_lock(ns, shmid, cmd);
1302                 break;
1303         default:
1304                 return -EINVAL;
1305         }
1306         return err;
1307 }
1308
1309 COMPAT_SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, void __user *, uptr)
1310 {
1311         return compat_ksys_shmctl(shmid, cmd, uptr);
1312 }
1313 #endif
1314
1315 /*
1316  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
1317  *
1318  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
1319  * "raddr" thing points to kernel space, and there has to be a wrapper around
1320  * this.
1321  */
1322 long do_shmat(int shmid, char __user *shmaddr, int shmflg,
1323               ulong *raddr, unsigned long shmlba)
1324 {
1325         struct shmid_kernel *shp;
1326         unsigned long addr = (unsigned long)shmaddr;
1327         unsigned long size;
1328         struct file *file;
1329         int    err;
1330         unsigned long flags = MAP_SHARED;
1331         unsigned long prot;
1332         int acc_mode;
1333         struct ipc_namespace *ns;
1334         struct shm_file_data *sfd;
1335         struct path path;
1336         fmode_t f_mode;
1337         unsigned long populate = 0;
1338
1339         err = -EINVAL;
1340         if (shmid < 0)
1341                 goto out;
1342
1343         if (addr) {
1344                 if (addr & (shmlba - 1)) {
1345                         /*
1346                          * Round down to the nearest multiple of shmlba.
1347                          * For sane do_mmap_pgoff() parameters, avoid
1348                          * round downs that trigger nil-page and MAP_FIXED.
1349                          */
1350                         if ((shmflg & SHM_RND) && addr >= shmlba)
1351                                 addr &= ~(shmlba - 1);
1352                         else
1353 #ifndef __ARCH_FORCE_SHMLBA
1354                                 if (addr & ~PAGE_MASK)
1355 #endif
1356                                         goto out;
1357                 }
1358
1359                 flags |= MAP_FIXED;
1360         } else if ((shmflg & SHM_REMAP))
1361                 goto out;
1362
1363         if (shmflg & SHM_RDONLY) {
1364                 prot = PROT_READ;
1365                 acc_mode = S_IRUGO;
1366                 f_mode = FMODE_READ;
1367         } else {
1368                 prot = PROT_READ | PROT_WRITE;
1369                 acc_mode = S_IRUGO | S_IWUGO;
1370                 f_mode = FMODE_READ | FMODE_WRITE;
1371         }
1372         if (shmflg & SHM_EXEC) {
1373                 prot |= PROT_EXEC;
1374                 acc_mode |= S_IXUGO;
1375         }
1376
1377         /*
1378          * We cannot rely on the fs check since SYSV IPC does have an
1379          * additional creator id...
1380          */
1381         ns = current->nsproxy->ipc_ns;
1382         rcu_read_lock();
1383         shp = shm_obtain_object_check(ns, shmid);
1384         if (IS_ERR(shp)) {
1385                 err = PTR_ERR(shp);
1386                 goto out_unlock;
1387         }
1388
1389         err = -EACCES;
1390         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1391                 goto out_unlock;
1392
1393         err = security_shm_shmat(&shp->shm_perm, shmaddr, shmflg);
1394         if (err)
1395                 goto out_unlock;
1396
1397         ipc_lock_object(&shp->shm_perm);
1398
1399         /* check if shm_destroy() is tearing down shp */
1400         if (!ipc_valid_object(&shp->shm_perm)) {
1401                 ipc_unlock_object(&shp->shm_perm);
1402                 err = -EIDRM;
1403                 goto out_unlock;
1404         }
1405
1406         path = shp->shm_file->f_path;
1407         path_get(&path);
1408         shp->shm_nattch++;
1409         size = i_size_read(d_inode(path.dentry));
1410         ipc_unlock_object(&shp->shm_perm);
1411         rcu_read_unlock();
1412
1413         err = -ENOMEM;
1414         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1415         if (!sfd) {
1416                 path_put(&path);
1417                 goto out_nattch;
1418         }
1419
1420         file = alloc_file(&path, f_mode,
1421                           is_file_hugepages(shp->shm_file) ?
1422                                 &shm_file_operations_huge :
1423                                 &shm_file_operations);
1424         err = PTR_ERR(file);
1425         if (IS_ERR(file)) {
1426                 kfree(sfd);
1427                 path_put(&path);
1428                 goto out_nattch;
1429         }
1430
1431         file->private_data = sfd;
1432         file->f_mapping = shp->shm_file->f_mapping;
1433         sfd->id = shp->shm_perm.id;
1434         sfd->ns = get_ipc_ns(ns);
1435         sfd->file = shp->shm_file;
1436         sfd->vm_ops = NULL;
1437
1438         err = security_mmap_file(file, prot, flags);
1439         if (err)
1440                 goto out_fput;
1441
1442         if (down_write_killable(&current->mm->mmap_sem)) {
1443                 err = -EINTR;
1444                 goto out_fput;
1445         }
1446
1447         if (addr && !(shmflg & SHM_REMAP)) {
1448                 err = -EINVAL;
1449                 if (addr + size < addr)
1450                         goto invalid;
1451
1452                 if (find_vma_intersection(current->mm, addr, addr + size))
1453                         goto invalid;
1454         }
1455
1456         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate, NULL);
1457         *raddr = addr;
1458         err = 0;
1459         if (IS_ERR_VALUE(addr))
1460                 err = (long)addr;
1461 invalid:
1462         up_write(&current->mm->mmap_sem);
1463         if (populate)
1464                 mm_populate(addr, populate);
1465
1466 out_fput:
1467         fput(file);
1468
1469 out_nattch:
1470         down_write(&shm_ids(ns).rwsem);
1471         shp = shm_lock(ns, shmid);
1472         shp->shm_nattch--;
1473         if (shm_may_destroy(ns, shp))
1474                 shm_destroy(ns, shp);
1475         else
1476                 shm_unlock(shp);
1477         up_write(&shm_ids(ns).rwsem);
1478         return err;
1479
1480 out_unlock:
1481         rcu_read_unlock();
1482 out:
1483         return err;
1484 }
1485
1486 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1487 {
1488         unsigned long ret;
1489         long err;
1490
1491         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1492         if (err)
1493                 return err;
1494         force_successful_syscall_return();
1495         return (long)ret;
1496 }
1497
1498 #ifdef CONFIG_COMPAT
1499
1500 #ifndef COMPAT_SHMLBA
1501 #define COMPAT_SHMLBA   SHMLBA
1502 #endif
1503
1504 COMPAT_SYSCALL_DEFINE3(shmat, int, shmid, compat_uptr_t, shmaddr, int, shmflg)
1505 {
1506         unsigned long ret;
1507         long err;
1508
1509         err = do_shmat(shmid, compat_ptr(shmaddr), shmflg, &ret, COMPAT_SHMLBA);
1510         if (err)
1511                 return err;
1512         force_successful_syscall_return();
1513         return (long)ret;
1514 }
1515 #endif
1516
1517 /*
1518  * detach and kill segment if marked destroyed.
1519  * The work is done in shm_close.
1520  */
1521 long ksys_shmdt(char __user *shmaddr)
1522 {
1523         struct mm_struct *mm = current->mm;
1524         struct vm_area_struct *vma;
1525         unsigned long addr = (unsigned long)shmaddr;
1526         int retval = -EINVAL;
1527 #ifdef CONFIG_MMU
1528         loff_t size = 0;
1529         struct file *file;
1530         struct vm_area_struct *next;
1531 #endif
1532
1533         if (addr & ~PAGE_MASK)
1534                 return retval;
1535
1536         if (down_write_killable(&mm->mmap_sem))
1537                 return -EINTR;
1538
1539         /*
1540          * This function tries to be smart and unmap shm segments that
1541          * were modified by partial mlock or munmap calls:
1542          * - It first determines the size of the shm segment that should be
1543          *   unmapped: It searches for a vma that is backed by shm and that
1544          *   started at address shmaddr. It records it's size and then unmaps
1545          *   it.
1546          * - Then it unmaps all shm vmas that started at shmaddr and that
1547          *   are within the initially determined size and that are from the
1548          *   same shm segment from which we determined the size.
1549          * Errors from do_munmap are ignored: the function only fails if
1550          * it's called with invalid parameters or if it's called to unmap
1551          * a part of a vma. Both calls in this function are for full vmas,
1552          * the parameters are directly copied from the vma itself and always
1553          * valid - therefore do_munmap cannot fail. (famous last words?)
1554          */
1555         /*
1556          * If it had been mremap()'d, the starting address would not
1557          * match the usual checks anyway. So assume all vma's are
1558          * above the starting address given.
1559          */
1560         vma = find_vma(mm, addr);
1561
1562 #ifdef CONFIG_MMU
1563         while (vma) {
1564                 next = vma->vm_next;
1565
1566                 /*
1567                  * Check if the starting address would match, i.e. it's
1568                  * a fragment created by mprotect() and/or munmap(), or it
1569                  * otherwise it starts at this address with no hassles.
1570                  */
1571                 if ((vma->vm_ops == &shm_vm_ops) &&
1572                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1573
1574                         /*
1575                          * Record the file of the shm segment being
1576                          * unmapped.  With mremap(), someone could place
1577                          * page from another segment but with equal offsets
1578                          * in the range we are unmapping.
1579                          */
1580                         file = vma->vm_file;
1581                         size = i_size_read(file_inode(vma->vm_file));
1582                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1583                         /*
1584                          * We discovered the size of the shm segment, so
1585                          * break out of here and fall through to the next
1586                          * loop that uses the size information to stop
1587                          * searching for matching vma's.
1588                          */
1589                         retval = 0;
1590                         vma = next;
1591                         break;
1592                 }
1593                 vma = next;
1594         }
1595
1596         /*
1597          * We need look no further than the maximum address a fragment
1598          * could possibly have landed at. Also cast things to loff_t to
1599          * prevent overflows and make comparisons vs. equal-width types.
1600          */
1601         size = PAGE_ALIGN(size);
1602         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1603                 next = vma->vm_next;
1604
1605                 /* finding a matching vma now does not alter retval */
1606                 if ((vma->vm_ops == &shm_vm_ops) &&
1607                     ((vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) &&
1608                     (vma->vm_file == file))
1609                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1610                 vma = next;
1611         }
1612
1613 #else   /* CONFIG_MMU */
1614         /* under NOMMU conditions, the exact address to be destroyed must be
1615          * given
1616          */
1617         if (vma && vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1618                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
1619                 retval = 0;
1620         }
1621
1622 #endif
1623
1624         up_write(&mm->mmap_sem);
1625         return retval;
1626 }
1627
1628 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1629 {
1630         return ksys_shmdt(shmaddr);
1631 }
1632
1633 #ifdef CONFIG_PROC_FS
1634 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1635 {
1636         struct pid_namespace *pid_ns = ipc_seq_pid_ns(s);
1637         struct user_namespace *user_ns = seq_user_ns(s);
1638         struct kern_ipc_perm *ipcp = it;
1639         struct shmid_kernel *shp;
1640         unsigned long rss = 0, swp = 0;
1641
1642         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
1643         shm_add_rss_swap(shp, &rss, &swp);
1644
1645 #if BITS_PER_LONG <= 32
1646 #define SIZE_SPEC "%10lu"
1647 #else
1648 #define SIZE_SPEC "%21lu"
1649 #endif
1650
1651         seq_printf(s,
1652                    "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1653                    "%5lu %5u %5u %5u %5u %10llu %10llu %10llu "
1654                    SIZE_SPEC " " SIZE_SPEC "\n",
1655                    shp->shm_perm.key,
1656                    shp->shm_perm.id,
1657                    shp->shm_perm.mode,
1658                    shp->shm_segsz,
1659                    pid_nr_ns(shp->shm_cprid, pid_ns),
1660                    pid_nr_ns(shp->shm_lprid, pid_ns),
1661                    shp->shm_nattch,
1662                    from_kuid_munged(user_ns, shp->shm_perm.uid),
1663                    from_kgid_munged(user_ns, shp->shm_perm.gid),
1664                    from_kuid_munged(user_ns, shp->shm_perm.cuid),
1665                    from_kgid_munged(user_ns, shp->shm_perm.cgid),
1666                    shp->shm_atim,
1667                    shp->shm_dtim,
1668                    shp->shm_ctim,
1669                    rss * PAGE_SIZE,
1670                    swp * PAGE_SIZE);
1671
1672         return 0;
1673 }
1674 #endif