xref: /linux/ipc/shm.c (revision 4f1933620f57145212cdbb1ac6ce099eeeb21c5a)
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  */
17 
18 #include <linux/config.h>
19 #include <linux/slab.h>
20 #include <linux/mm.h>
21 #include <linux/hugetlb.h>
22 #include <linux/shm.h>
23 #include <linux/init.h>
24 #include <linux/file.h>
25 #include <linux/mman.h>
26 #include <linux/shmem_fs.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/ptrace.h>
32 #include <linux/seq_file.h>
33 
34 #include <asm/uaccess.h>
35 
36 #include "util.h"
37 
38 static struct file_operations shm_file_operations;
39 static struct vm_operations_struct shm_vm_ops;
40 
41 static struct ipc_ids shm_ids;
42 
43 #define shm_lock(id)	((struct shmid_kernel*)ipc_lock(&shm_ids,id))
44 #define shm_unlock(shp)	ipc_unlock(&(shp)->shm_perm)
45 #define shm_get(id)	((struct shmid_kernel*)ipc_get(&shm_ids,id))
46 #define shm_buildid(id, seq) \
47 	ipc_buildid(&shm_ids, id, seq)
48 
49 static int newseg (key_t key, int shmflg, size_t size);
50 static void shm_open (struct vm_area_struct *shmd);
51 static void shm_close (struct vm_area_struct *shmd);
52 #ifdef CONFIG_PROC_FS
53 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
54 #endif
55 
56 size_t	shm_ctlmax = SHMMAX;
57 size_t 	shm_ctlall = SHMALL;
58 int 	shm_ctlmni = SHMMNI;
59 
60 static int shm_tot; /* total number of shared memory pages */
61 
62 void __init shm_init (void)
63 {
64 	ipc_init_ids(&shm_ids, 1);
65 	ipc_init_proc_interface("sysvipc/shm",
66 				"       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
67 				&shm_ids,
68 				sysvipc_shm_proc_show);
69 }
70 
71 static inline int shm_checkid(struct shmid_kernel *s, int id)
72 {
73 	if (ipc_checkid(&shm_ids,&s->shm_perm,id))
74 		return -EIDRM;
75 	return 0;
76 }
77 
78 static inline struct shmid_kernel *shm_rmid(int id)
79 {
80 	return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
81 }
82 
83 static inline int shm_addid(struct shmid_kernel *shp)
84 {
85 	return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
86 }
87 
88 
89 
90 static inline void shm_inc (int id) {
91 	struct shmid_kernel *shp;
92 
93 	if(!(shp = shm_lock(id)))
94 		BUG();
95 	shp->shm_atim = get_seconds();
96 	shp->shm_lprid = current->tgid;
97 	shp->shm_nattch++;
98 	shm_unlock(shp);
99 }
100 
101 /* This is called by fork, once for every shm attach. */
102 static void shm_open (struct vm_area_struct *shmd)
103 {
104 	shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
105 }
106 
107 /*
108  * shm_destroy - free the struct shmid_kernel
109  *
110  * @shp: struct to free
111  *
112  * It has to be called with shp and shm_ids.sem locked,
113  * but returns with shp unlocked and freed.
114  */
115 static void shm_destroy (struct shmid_kernel *shp)
116 {
117 	shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
118 	shm_rmid (shp->id);
119 	shm_unlock(shp);
120 	if (!is_file_hugepages(shp->shm_file))
121 		shmem_lock(shp->shm_file, 0, shp->mlock_user);
122 	else
123 		user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
124 						shp->mlock_user);
125 	fput (shp->shm_file);
126 	security_shm_free(shp);
127 	ipc_rcu_putref(shp);
128 }
129 
130 /*
131  * remove the attach descriptor shmd.
132  * free memory for segment if it is marked destroyed.
133  * The descriptor has already been removed from the current->mm->mmap list
134  * and will later be kfree()d.
135  */
136 static void shm_close (struct vm_area_struct *shmd)
137 {
138 	struct file * file = shmd->vm_file;
139 	int id = file->f_dentry->d_inode->i_ino;
140 	struct shmid_kernel *shp;
141 
142 	down (&shm_ids.sem);
143 	/* remove from the list of attaches of the shm segment */
144 	if(!(shp = shm_lock(id)))
145 		BUG();
146 	shp->shm_lprid = current->tgid;
147 	shp->shm_dtim = get_seconds();
148 	shp->shm_nattch--;
149 	if(shp->shm_nattch == 0 &&
150 	   shp->shm_perm.mode & SHM_DEST)
151 		shm_destroy (shp);
152 	else
153 		shm_unlock(shp);
154 	up (&shm_ids.sem);
155 }
156 
157 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
158 {
159 	int ret;
160 
161 	ret = shmem_mmap(file, vma);
162 	if (ret == 0) {
163 		vma->vm_ops = &shm_vm_ops;
164 		shm_inc(file->f_dentry->d_inode->i_ino);
165 	}
166 
167 	return ret;
168 }
169 
170 static struct file_operations shm_file_operations = {
171 	.mmap	= shm_mmap,
172 #ifndef CONFIG_MMU
173 	.get_unmapped_area = shmem_get_unmapped_area,
174 #endif
175 };
176 
177 static struct vm_operations_struct shm_vm_ops = {
178 	.open	= shm_open,	/* callback for a new vm-area open */
179 	.close	= shm_close,	/* callback for when the vm-area is released */
180 	.nopage	= shmem_nopage,
181 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
182 	.set_policy = shmem_set_policy,
183 	.get_policy = shmem_get_policy,
184 #endif
185 };
186 
187 static int newseg (key_t key, int shmflg, size_t size)
188 {
189 	int error;
190 	struct shmid_kernel *shp;
191 	int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
192 	struct file * file;
193 	char name[13];
194 	int id;
195 
196 	if (size < SHMMIN || size > shm_ctlmax)
197 		return -EINVAL;
198 
199 	if (shm_tot + numpages >= shm_ctlall)
200 		return -ENOSPC;
201 
202 	shp = ipc_rcu_alloc(sizeof(*shp));
203 	if (!shp)
204 		return -ENOMEM;
205 
206 	shp->shm_perm.key = key;
207 	shp->shm_perm.mode = (shmflg & S_IRWXUGO);
208 	shp->mlock_user = NULL;
209 
210 	shp->shm_perm.security = NULL;
211 	error = security_shm_alloc(shp);
212 	if (error) {
213 		ipc_rcu_putref(shp);
214 		return error;
215 	}
216 
217 	if (shmflg & SHM_HUGETLB) {
218 		/* hugetlb_zero_setup takes care of mlock user accounting */
219 		file = hugetlb_zero_setup(size);
220 		shp->mlock_user = current->user;
221 	} else {
222 		int acctflag = VM_ACCOUNT;
223 		/*
224 		 * Do not allow no accounting for OVERCOMMIT_NEVER, even
225 	 	 * if it's asked for.
226 		 */
227 		if  ((shmflg & SHM_NORESERVE) &&
228 				sysctl_overcommit_memory != OVERCOMMIT_NEVER)
229 			acctflag = 0;
230 		sprintf (name, "SYSV%08x", key);
231 		file = shmem_file_setup(name, size, acctflag);
232 	}
233 	error = PTR_ERR(file);
234 	if (IS_ERR(file))
235 		goto no_file;
236 
237 	error = -ENOSPC;
238 	id = shm_addid(shp);
239 	if(id == -1)
240 		goto no_id;
241 
242 	shp->shm_cprid = current->tgid;
243 	shp->shm_lprid = 0;
244 	shp->shm_atim = shp->shm_dtim = 0;
245 	shp->shm_ctim = get_seconds();
246 	shp->shm_segsz = size;
247 	shp->shm_nattch = 0;
248 	shp->id = shm_buildid(id,shp->shm_perm.seq);
249 	shp->shm_file = file;
250 	file->f_dentry->d_inode->i_ino = shp->id;
251 
252 	/* Hugetlb ops would have already been assigned. */
253 	if (!(shmflg & SHM_HUGETLB))
254 		file->f_op = &shm_file_operations;
255 
256 	shm_tot += numpages;
257 	shm_unlock(shp);
258 	return shp->id;
259 
260 no_id:
261 	fput(file);
262 no_file:
263 	security_shm_free(shp);
264 	ipc_rcu_putref(shp);
265 	return error;
266 }
267 
268 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
269 {
270 	struct shmid_kernel *shp;
271 	int err, id = 0;
272 
273 	down(&shm_ids.sem);
274 	if (key == IPC_PRIVATE) {
275 		err = newseg(key, shmflg, size);
276 	} else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
277 		if (!(shmflg & IPC_CREAT))
278 			err = -ENOENT;
279 		else
280 			err = newseg(key, shmflg, size);
281 	} else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
282 		err = -EEXIST;
283 	} else {
284 		shp = shm_lock(id);
285 		if(shp==NULL)
286 			BUG();
287 		if (shp->shm_segsz < size)
288 			err = -EINVAL;
289 		else if (ipcperms(&shp->shm_perm, shmflg))
290 			err = -EACCES;
291 		else {
292 			int shmid = shm_buildid(id, shp->shm_perm.seq);
293 			err = security_shm_associate(shp, shmflg);
294 			if (!err)
295 				err = shmid;
296 		}
297 		shm_unlock(shp);
298 	}
299 	up(&shm_ids.sem);
300 
301 	return err;
302 }
303 
304 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
305 {
306 	switch(version) {
307 	case IPC_64:
308 		return copy_to_user(buf, in, sizeof(*in));
309 	case IPC_OLD:
310 	    {
311 		struct shmid_ds out;
312 
313 		ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
314 		out.shm_segsz	= in->shm_segsz;
315 		out.shm_atime	= in->shm_atime;
316 		out.shm_dtime	= in->shm_dtime;
317 		out.shm_ctime	= in->shm_ctime;
318 		out.shm_cpid	= in->shm_cpid;
319 		out.shm_lpid	= in->shm_lpid;
320 		out.shm_nattch	= in->shm_nattch;
321 
322 		return copy_to_user(buf, &out, sizeof(out));
323 	    }
324 	default:
325 		return -EINVAL;
326 	}
327 }
328 
329 struct shm_setbuf {
330 	uid_t	uid;
331 	gid_t	gid;
332 	mode_t	mode;
333 };
334 
335 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
336 {
337 	switch(version) {
338 	case IPC_64:
339 	    {
340 		struct shmid64_ds tbuf;
341 
342 		if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
343 			return -EFAULT;
344 
345 		out->uid	= tbuf.shm_perm.uid;
346 		out->gid	= tbuf.shm_perm.gid;
347 		out->mode	= tbuf.shm_perm.mode;
348 
349 		return 0;
350 	    }
351 	case IPC_OLD:
352 	    {
353 		struct shmid_ds tbuf_old;
354 
355 		if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
356 			return -EFAULT;
357 
358 		out->uid	= tbuf_old.shm_perm.uid;
359 		out->gid	= tbuf_old.shm_perm.gid;
360 		out->mode	= tbuf_old.shm_perm.mode;
361 
362 		return 0;
363 	    }
364 	default:
365 		return -EINVAL;
366 	}
367 }
368 
369 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
370 {
371 	switch(version) {
372 	case IPC_64:
373 		return copy_to_user(buf, in, sizeof(*in));
374 	case IPC_OLD:
375 	    {
376 		struct shminfo out;
377 
378 		if(in->shmmax > INT_MAX)
379 			out.shmmax = INT_MAX;
380 		else
381 			out.shmmax = (int)in->shmmax;
382 
383 		out.shmmin	= in->shmmin;
384 		out.shmmni	= in->shmmni;
385 		out.shmseg	= in->shmseg;
386 		out.shmall	= in->shmall;
387 
388 		return copy_to_user(buf, &out, sizeof(out));
389 	    }
390 	default:
391 		return -EINVAL;
392 	}
393 }
394 
395 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
396 {
397 	int i;
398 
399 	*rss = 0;
400 	*swp = 0;
401 
402 	for (i = 0; i <= shm_ids.max_id; i++) {
403 		struct shmid_kernel *shp;
404 		struct inode *inode;
405 
406 		shp = shm_get(i);
407 		if(!shp)
408 			continue;
409 
410 		inode = shp->shm_file->f_dentry->d_inode;
411 
412 		if (is_file_hugepages(shp->shm_file)) {
413 			struct address_space *mapping = inode->i_mapping;
414 			*rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
415 		} else {
416 			struct shmem_inode_info *info = SHMEM_I(inode);
417 			spin_lock(&info->lock);
418 			*rss += inode->i_mapping->nrpages;
419 			*swp += info->swapped;
420 			spin_unlock(&info->lock);
421 		}
422 	}
423 }
424 
425 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
426 {
427 	struct shm_setbuf setbuf;
428 	struct shmid_kernel *shp;
429 	int err, version;
430 
431 	if (cmd < 0 || shmid < 0) {
432 		err = -EINVAL;
433 		goto out;
434 	}
435 
436 	version = ipc_parse_version(&cmd);
437 
438 	switch (cmd) { /* replace with proc interface ? */
439 	case IPC_INFO:
440 	{
441 		struct shminfo64 shminfo;
442 
443 		err = security_shm_shmctl(NULL, cmd);
444 		if (err)
445 			return err;
446 
447 		memset(&shminfo,0,sizeof(shminfo));
448 		shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
449 		shminfo.shmmax = shm_ctlmax;
450 		shminfo.shmall = shm_ctlall;
451 
452 		shminfo.shmmin = SHMMIN;
453 		if(copy_shminfo_to_user (buf, &shminfo, version))
454 			return -EFAULT;
455 		/* reading a integer is always atomic */
456 		err= shm_ids.max_id;
457 		if(err<0)
458 			err = 0;
459 		goto out;
460 	}
461 	case SHM_INFO:
462 	{
463 		struct shm_info shm_info;
464 
465 		err = security_shm_shmctl(NULL, cmd);
466 		if (err)
467 			return err;
468 
469 		memset(&shm_info,0,sizeof(shm_info));
470 		down(&shm_ids.sem);
471 		shm_info.used_ids = shm_ids.in_use;
472 		shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
473 		shm_info.shm_tot = shm_tot;
474 		shm_info.swap_attempts = 0;
475 		shm_info.swap_successes = 0;
476 		err = shm_ids.max_id;
477 		up(&shm_ids.sem);
478 		if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
479 			err = -EFAULT;
480 			goto out;
481 		}
482 
483 		err = err < 0 ? 0 : err;
484 		goto out;
485 	}
486 	case SHM_STAT:
487 	case IPC_STAT:
488 	{
489 		struct shmid64_ds tbuf;
490 		int result;
491 		memset(&tbuf, 0, sizeof(tbuf));
492 		shp = shm_lock(shmid);
493 		if(shp==NULL) {
494 			err = -EINVAL;
495 			goto out;
496 		} else if(cmd==SHM_STAT) {
497 			err = -EINVAL;
498 			if (shmid > shm_ids.max_id)
499 				goto out_unlock;
500 			result = shm_buildid(shmid, shp->shm_perm.seq);
501 		} else {
502 			err = shm_checkid(shp,shmid);
503 			if(err)
504 				goto out_unlock;
505 			result = 0;
506 		}
507 		err=-EACCES;
508 		if (ipcperms (&shp->shm_perm, S_IRUGO))
509 			goto out_unlock;
510 		err = security_shm_shmctl(shp, cmd);
511 		if (err)
512 			goto out_unlock;
513 		kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
514 		tbuf.shm_segsz	= shp->shm_segsz;
515 		tbuf.shm_atime	= shp->shm_atim;
516 		tbuf.shm_dtime	= shp->shm_dtim;
517 		tbuf.shm_ctime	= shp->shm_ctim;
518 		tbuf.shm_cpid	= shp->shm_cprid;
519 		tbuf.shm_lpid	= shp->shm_lprid;
520 		if (!is_file_hugepages(shp->shm_file))
521 			tbuf.shm_nattch	= shp->shm_nattch;
522 		else
523 			tbuf.shm_nattch = file_count(shp->shm_file) - 1;
524 		shm_unlock(shp);
525 		if(copy_shmid_to_user (buf, &tbuf, version))
526 			err = -EFAULT;
527 		else
528 			err = result;
529 		goto out;
530 	}
531 	case SHM_LOCK:
532 	case SHM_UNLOCK:
533 	{
534 		shp = shm_lock(shmid);
535 		if(shp==NULL) {
536 			err = -EINVAL;
537 			goto out;
538 		}
539 		err = shm_checkid(shp,shmid);
540 		if(err)
541 			goto out_unlock;
542 
543 		if (!capable(CAP_IPC_LOCK)) {
544 			err = -EPERM;
545 			if (current->euid != shp->shm_perm.uid &&
546 			    current->euid != shp->shm_perm.cuid)
547 				goto out_unlock;
548 			if (cmd == SHM_LOCK &&
549 			    !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
550 				goto out_unlock;
551 		}
552 
553 		err = security_shm_shmctl(shp, cmd);
554 		if (err)
555 			goto out_unlock;
556 
557 		if(cmd==SHM_LOCK) {
558 			struct user_struct * user = current->user;
559 			if (!is_file_hugepages(shp->shm_file)) {
560 				err = shmem_lock(shp->shm_file, 1, user);
561 				if (!err) {
562 					shp->shm_perm.mode |= SHM_LOCKED;
563 					shp->mlock_user = user;
564 				}
565 			}
566 		} else if (!is_file_hugepages(shp->shm_file)) {
567 			shmem_lock(shp->shm_file, 0, shp->mlock_user);
568 			shp->shm_perm.mode &= ~SHM_LOCKED;
569 			shp->mlock_user = NULL;
570 		}
571 		shm_unlock(shp);
572 		goto out;
573 	}
574 	case IPC_RMID:
575 	{
576 		/*
577 		 *	We cannot simply remove the file. The SVID states
578 		 *	that the block remains until the last person
579 		 *	detaches from it, then is deleted. A shmat() on
580 		 *	an RMID segment is legal in older Linux and if
581 		 *	we change it apps break...
582 		 *
583 		 *	Instead we set a destroyed flag, and then blow
584 		 *	the name away when the usage hits zero.
585 		 */
586 		down(&shm_ids.sem);
587 		shp = shm_lock(shmid);
588 		err = -EINVAL;
589 		if (shp == NULL)
590 			goto out_up;
591 		err = shm_checkid(shp, shmid);
592 		if(err)
593 			goto out_unlock_up;
594 
595 		if (current->euid != shp->shm_perm.uid &&
596 		    current->euid != shp->shm_perm.cuid &&
597 		    !capable(CAP_SYS_ADMIN)) {
598 			err=-EPERM;
599 			goto out_unlock_up;
600 		}
601 
602 		err = security_shm_shmctl(shp, cmd);
603 		if (err)
604 			goto out_unlock_up;
605 
606 		if (shp->shm_nattch){
607 			shp->shm_perm.mode |= SHM_DEST;
608 			/* Do not find it any more */
609 			shp->shm_perm.key = IPC_PRIVATE;
610 			shm_unlock(shp);
611 		} else
612 			shm_destroy (shp);
613 		up(&shm_ids.sem);
614 		goto out;
615 	}
616 
617 	case IPC_SET:
618 	{
619 		if (copy_shmid_from_user (&setbuf, buf, version)) {
620 			err = -EFAULT;
621 			goto out;
622 		}
623 		if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
624 			return err;
625 		down(&shm_ids.sem);
626 		shp = shm_lock(shmid);
627 		err=-EINVAL;
628 		if(shp==NULL)
629 			goto out_up;
630 		err = shm_checkid(shp,shmid);
631 		if(err)
632 			goto out_unlock_up;
633 		err=-EPERM;
634 		if (current->euid != shp->shm_perm.uid &&
635 		    current->euid != shp->shm_perm.cuid &&
636 		    !capable(CAP_SYS_ADMIN)) {
637 			goto out_unlock_up;
638 		}
639 
640 		err = security_shm_shmctl(shp, cmd);
641 		if (err)
642 			goto out_unlock_up;
643 
644 		shp->shm_perm.uid = setbuf.uid;
645 		shp->shm_perm.gid = setbuf.gid;
646 		shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
647 			| (setbuf.mode & S_IRWXUGO);
648 		shp->shm_ctim = get_seconds();
649 		break;
650 	}
651 
652 	default:
653 		err = -EINVAL;
654 		goto out;
655 	}
656 
657 	err = 0;
658 out_unlock_up:
659 	shm_unlock(shp);
660 out_up:
661 	up(&shm_ids.sem);
662 	goto out;
663 out_unlock:
664 	shm_unlock(shp);
665 out:
666 	return err;
667 }
668 
669 /*
670  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
671  *
672  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
673  * "raddr" thing points to kernel space, and there has to be a wrapper around
674  * this.
675  */
676 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
677 {
678 	struct shmid_kernel *shp;
679 	unsigned long addr;
680 	unsigned long size;
681 	struct file * file;
682 	int    err;
683 	unsigned long flags;
684 	unsigned long prot;
685 	unsigned long o_flags;
686 	int acc_mode;
687 	void *user_addr;
688 
689 	if (shmid < 0) {
690 		err = -EINVAL;
691 		goto out;
692 	} else if ((addr = (ulong)shmaddr)) {
693 		if (addr & (SHMLBA-1)) {
694 			if (shmflg & SHM_RND)
695 				addr &= ~(SHMLBA-1);	   /* round down */
696 			else
697 #ifndef __ARCH_FORCE_SHMLBA
698 				if (addr & ~PAGE_MASK)
699 #endif
700 					return -EINVAL;
701 		}
702 		flags = MAP_SHARED | MAP_FIXED;
703 	} else {
704 		if ((shmflg & SHM_REMAP))
705 			return -EINVAL;
706 
707 		flags = MAP_SHARED;
708 	}
709 
710 	if (shmflg & SHM_RDONLY) {
711 		prot = PROT_READ;
712 		o_flags = O_RDONLY;
713 		acc_mode = S_IRUGO;
714 	} else {
715 		prot = PROT_READ | PROT_WRITE;
716 		o_flags = O_RDWR;
717 		acc_mode = S_IRUGO | S_IWUGO;
718 	}
719 	if (shmflg & SHM_EXEC) {
720 		prot |= PROT_EXEC;
721 		acc_mode |= S_IXUGO;
722 	}
723 
724 	/*
725 	 * We cannot rely on the fs check since SYSV IPC does have an
726 	 * additional creator id...
727 	 */
728 	shp = shm_lock(shmid);
729 	if(shp == NULL) {
730 		err = -EINVAL;
731 		goto out;
732 	}
733 	err = shm_checkid(shp,shmid);
734 	if (err) {
735 		shm_unlock(shp);
736 		goto out;
737 	}
738 	if (ipcperms(&shp->shm_perm, acc_mode)) {
739 		shm_unlock(shp);
740 		err = -EACCES;
741 		goto out;
742 	}
743 
744 	err = security_shm_shmat(shp, shmaddr, shmflg);
745 	if (err) {
746 		shm_unlock(shp);
747 		return err;
748 	}
749 
750 	file = shp->shm_file;
751 	size = i_size_read(file->f_dentry->d_inode);
752 	shp->shm_nattch++;
753 	shm_unlock(shp);
754 
755 	down_write(&current->mm->mmap_sem);
756 	if (addr && !(shmflg & SHM_REMAP)) {
757 		user_addr = ERR_PTR(-EINVAL);
758 		if (find_vma_intersection(current->mm, addr, addr + size))
759 			goto invalid;
760 		/*
761 		 * If shm segment goes below stack, make sure there is some
762 		 * space left for the stack to grow (at least 4 pages).
763 		 */
764 		if (addr < current->mm->start_stack &&
765 		    addr > current->mm->start_stack - size - PAGE_SIZE * 5)
766 			goto invalid;
767 	}
768 
769 	user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
770 
771 invalid:
772 	up_write(&current->mm->mmap_sem);
773 
774 	down (&shm_ids.sem);
775 	if(!(shp = shm_lock(shmid)))
776 		BUG();
777 	shp->shm_nattch--;
778 	if(shp->shm_nattch == 0 &&
779 	   shp->shm_perm.mode & SHM_DEST)
780 		shm_destroy (shp);
781 	else
782 		shm_unlock(shp);
783 	up (&shm_ids.sem);
784 
785 	*raddr = (unsigned long) user_addr;
786 	err = 0;
787 	if (IS_ERR(user_addr))
788 		err = PTR_ERR(user_addr);
789 out:
790 	return err;
791 }
792 
793 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
794 {
795 	unsigned long ret;
796 	long err;
797 
798 	err = do_shmat(shmid, shmaddr, shmflg, &ret);
799 	if (err)
800 		return err;
801 	force_successful_syscall_return();
802 	return (long)ret;
803 }
804 
805 /*
806  * detach and kill segment if marked destroyed.
807  * The work is done in shm_close.
808  */
809 asmlinkage long sys_shmdt(char __user *shmaddr)
810 {
811 	struct mm_struct *mm = current->mm;
812 	struct vm_area_struct *vma, *next;
813 	unsigned long addr = (unsigned long)shmaddr;
814 	loff_t size = 0;
815 	int retval = -EINVAL;
816 
817 	down_write(&mm->mmap_sem);
818 
819 	/*
820 	 * This function tries to be smart and unmap shm segments that
821 	 * were modified by partial mlock or munmap calls:
822 	 * - It first determines the size of the shm segment that should be
823 	 *   unmapped: It searches for a vma that is backed by shm and that
824 	 *   started at address shmaddr. It records it's size and then unmaps
825 	 *   it.
826 	 * - Then it unmaps all shm vmas that started at shmaddr and that
827 	 *   are within the initially determined size.
828 	 * Errors from do_munmap are ignored: the function only fails if
829 	 * it's called with invalid parameters or if it's called to unmap
830 	 * a part of a vma. Both calls in this function are for full vmas,
831 	 * the parameters are directly copied from the vma itself and always
832 	 * valid - therefore do_munmap cannot fail. (famous last words?)
833 	 */
834 	/*
835 	 * If it had been mremap()'d, the starting address would not
836 	 * match the usual checks anyway. So assume all vma's are
837 	 * above the starting address given.
838 	 */
839 	vma = find_vma(mm, addr);
840 
841 	while (vma) {
842 		next = vma->vm_next;
843 
844 		/*
845 		 * Check if the starting address would match, i.e. it's
846 		 * a fragment created by mprotect() and/or munmap(), or it
847 		 * otherwise it starts at this address with no hassles.
848 		 */
849 		if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
850 			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
851 
852 
853 			size = vma->vm_file->f_dentry->d_inode->i_size;
854 			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
855 			/*
856 			 * We discovered the size of the shm segment, so
857 			 * break out of here and fall through to the next
858 			 * loop that uses the size information to stop
859 			 * searching for matching vma's.
860 			 */
861 			retval = 0;
862 			vma = next;
863 			break;
864 		}
865 		vma = next;
866 	}
867 
868 	/*
869 	 * We need look no further than the maximum address a fragment
870 	 * could possibly have landed at. Also cast things to loff_t to
871 	 * prevent overflows and make comparisions vs. equal-width types.
872 	 */
873 	size = PAGE_ALIGN(size);
874 	while (vma && (loff_t)(vma->vm_end - addr) <= size) {
875 		next = vma->vm_next;
876 
877 		/* finding a matching vma now does not alter retval */
878 		if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
879 			(vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
880 
881 			do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
882 		vma = next;
883 	}
884 
885 	up_write(&mm->mmap_sem);
886 	return retval;
887 }
888 
889 #ifdef CONFIG_PROC_FS
890 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
891 {
892 	struct shmid_kernel *shp = it;
893 	char *format;
894 
895 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
896 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
897 
898 	if (sizeof(size_t) <= sizeof(int))
899 		format = SMALL_STRING;
900 	else
901 		format = BIG_STRING;
902 	return seq_printf(s, format,
903 			  shp->shm_perm.key,
904 			  shp->id,
905 			  shp->shm_perm.mode,
906 			  shp->shm_segsz,
907 			  shp->shm_cprid,
908 			  shp->shm_lprid,
909 			  is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
910 			  shp->shm_perm.uid,
911 			  shp->shm_perm.gid,
912 			  shp->shm_perm.cuid,
913 			  shp->shm_perm.cgid,
914 			  shp->shm_atim,
915 			  shp->shm_dtim,
916 			  shp->shm_ctim);
917 }
918 #endif
919