xref: /openbmc/linux/fs/hugetlbfs/inode.c (revision 508034a3)
1 /*
2  * hugetlbpage-backed filesystem.  Based on ramfs.
3  *
4  * William Irwin, 2002
5  *
6  * Copyright (C) 2002 Linus Torvalds.
7  */
8 
9 #include <linux/module.h>
10 #include <linux/thread_info.h>
11 #include <asm/current.h>
12 #include <linux/sched.h>		/* remove ASAP */
13 #include <linux/fs.h>
14 #include <linux/mount.h>
15 #include <linux/file.h>
16 #include <linux/writeback.h>
17 #include <linux/pagemap.h>
18 #include <linux/highmem.h>
19 #include <linux/init.h>
20 #include <linux/string.h>
21 #include <linux/backing-dev.h>
22 #include <linux/hugetlb.h>
23 #include <linux/pagevec.h>
24 #include <linux/quotaops.h>
25 #include <linux/slab.h>
26 #include <linux/dnotify.h>
27 #include <linux/statfs.h>
28 #include <linux/security.h>
29 
30 #include <asm/uaccess.h>
31 
32 /* some random number */
33 #define HUGETLBFS_MAGIC	0x958458f6
34 
35 static struct super_operations hugetlbfs_ops;
36 static struct address_space_operations hugetlbfs_aops;
37 struct file_operations hugetlbfs_file_operations;
38 static struct inode_operations hugetlbfs_dir_inode_operations;
39 static struct inode_operations hugetlbfs_inode_operations;
40 
41 static struct backing_dev_info hugetlbfs_backing_dev_info = {
42 	.ra_pages	= 0,	/* No readahead */
43 	.capabilities	= BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
44 };
45 
46 int sysctl_hugetlb_shm_group;
47 
48 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
49 {
50 	struct inode *inode = file->f_dentry->d_inode;
51 	struct address_space *mapping = inode->i_mapping;
52 	loff_t len, vma_len;
53 	int ret;
54 
55 	if ((vma->vm_flags & (VM_MAYSHARE | VM_WRITE)) == VM_WRITE)
56 		return -EINVAL;
57 
58 	if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
59 		return -EINVAL;
60 
61 	if (vma->vm_start & ~HPAGE_MASK)
62 		return -EINVAL;
63 
64 	if (vma->vm_end & ~HPAGE_MASK)
65 		return -EINVAL;
66 
67 	if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
68 		return -EINVAL;
69 
70 	vma_len = (loff_t)(vma->vm_end - vma->vm_start);
71 
72 	down(&inode->i_sem);
73 	file_accessed(file);
74 	vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
75 	vma->vm_ops = &hugetlb_vm_ops;
76 
77 	ret = -ENOMEM;
78 	len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
79 	if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
80 		goto out;
81 
82 	ret = hugetlb_prefault(mapping, vma);
83 	if (ret)
84 		goto out;
85 
86 	if (inode->i_size < len)
87 		inode->i_size = len;
88 out:
89 	up(&inode->i_sem);
90 
91 	return ret;
92 }
93 
94 /*
95  * Called under down_write(mmap_sem).
96  */
97 
98 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
99 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
100 		unsigned long len, unsigned long pgoff, unsigned long flags);
101 #else
102 static unsigned long
103 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
104 		unsigned long len, unsigned long pgoff, unsigned long flags)
105 {
106 	struct mm_struct *mm = current->mm;
107 	struct vm_area_struct *vma;
108 	unsigned long start_addr;
109 
110 	if (len & ~HPAGE_MASK)
111 		return -EINVAL;
112 	if (len > TASK_SIZE)
113 		return -ENOMEM;
114 
115 	if (addr) {
116 		addr = ALIGN(addr, HPAGE_SIZE);
117 		vma = find_vma(mm, addr);
118 		if (TASK_SIZE - len >= addr &&
119 		    (!vma || addr + len <= vma->vm_start))
120 			return addr;
121 	}
122 
123 	start_addr = mm->free_area_cache;
124 
125 	if (len <= mm->cached_hole_size)
126 		start_addr = TASK_UNMAPPED_BASE;
127 
128 full_search:
129 	addr = ALIGN(start_addr, HPAGE_SIZE);
130 
131 	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
132 		/* At this point:  (!vma || addr < vma->vm_end). */
133 		if (TASK_SIZE - len < addr) {
134 			/*
135 			 * Start a new search - just in case we missed
136 			 * some holes.
137 			 */
138 			if (start_addr != TASK_UNMAPPED_BASE) {
139 				start_addr = TASK_UNMAPPED_BASE;
140 				goto full_search;
141 			}
142 			return -ENOMEM;
143 		}
144 
145 		if (!vma || addr + len <= vma->vm_start)
146 			return addr;
147 		addr = ALIGN(vma->vm_end, HPAGE_SIZE);
148 	}
149 }
150 #endif
151 
152 /*
153  * Read a page. Again trivial. If it didn't already exist
154  * in the page cache, it is zero-filled.
155  */
156 static int hugetlbfs_readpage(struct file *file, struct page * page)
157 {
158 	unlock_page(page);
159 	return -EINVAL;
160 }
161 
162 static int hugetlbfs_prepare_write(struct file *file,
163 			struct page *page, unsigned offset, unsigned to)
164 {
165 	return -EINVAL;
166 }
167 
168 static int hugetlbfs_commit_write(struct file *file,
169 			struct page *page, unsigned offset, unsigned to)
170 {
171 	return -EINVAL;
172 }
173 
174 static void huge_pagevec_release(struct pagevec *pvec)
175 {
176 	int i;
177 
178 	for (i = 0; i < pagevec_count(pvec); ++i)
179 		put_page(pvec->pages[i]);
180 
181 	pagevec_reinit(pvec);
182 }
183 
184 static void truncate_huge_page(struct page *page)
185 {
186 	clear_page_dirty(page);
187 	ClearPageUptodate(page);
188 	remove_from_page_cache(page);
189 	put_page(page);
190 }
191 
192 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
193 {
194 	const pgoff_t start = lstart >> HPAGE_SHIFT;
195 	struct pagevec pvec;
196 	pgoff_t next;
197 	int i;
198 
199 	pagevec_init(&pvec, 0);
200 	next = start;
201 	while (1) {
202 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
203 			if (next == start)
204 				break;
205 			next = start;
206 			continue;
207 		}
208 
209 		for (i = 0; i < pagevec_count(&pvec); ++i) {
210 			struct page *page = pvec.pages[i];
211 
212 			lock_page(page);
213 			if (page->index > next)
214 				next = page->index;
215 			++next;
216 			truncate_huge_page(page);
217 			unlock_page(page);
218 			hugetlb_put_quota(mapping);
219 		}
220 		huge_pagevec_release(&pvec);
221 	}
222 	BUG_ON(!lstart && mapping->nrpages);
223 }
224 
225 static void hugetlbfs_delete_inode(struct inode *inode)
226 {
227 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
228 
229 	hlist_del_init(&inode->i_hash);
230 	list_del_init(&inode->i_list);
231 	list_del_init(&inode->i_sb_list);
232 	inode->i_state |= I_FREEING;
233 	inodes_stat.nr_inodes--;
234 	spin_unlock(&inode_lock);
235 
236 	if (inode->i_data.nrpages)
237 		truncate_hugepages(&inode->i_data, 0);
238 
239 	security_inode_delete(inode);
240 
241 	if (sbinfo->free_inodes >= 0) {
242 		spin_lock(&sbinfo->stat_lock);
243 		sbinfo->free_inodes++;
244 		spin_unlock(&sbinfo->stat_lock);
245 	}
246 
247 	clear_inode(inode);
248 	destroy_inode(inode);
249 }
250 
251 static void hugetlbfs_forget_inode(struct inode *inode)
252 {
253 	struct super_block *super_block = inode->i_sb;
254 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
255 
256 	if (hlist_unhashed(&inode->i_hash))
257 		goto out_truncate;
258 
259 	if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
260 		list_del(&inode->i_list);
261 		list_add(&inode->i_list, &inode_unused);
262 	}
263 	inodes_stat.nr_unused++;
264 	if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
265 		spin_unlock(&inode_lock);
266 		return;
267 	}
268 
269 	/* write_inode_now() ? */
270 	inodes_stat.nr_unused--;
271 	hlist_del_init(&inode->i_hash);
272 out_truncate:
273 	list_del_init(&inode->i_list);
274 	list_del_init(&inode->i_sb_list);
275 	inode->i_state |= I_FREEING;
276 	inodes_stat.nr_inodes--;
277 	spin_unlock(&inode_lock);
278 	if (inode->i_data.nrpages)
279 		truncate_hugepages(&inode->i_data, 0);
280 
281 	if (sbinfo->free_inodes >= 0) {
282 		spin_lock(&sbinfo->stat_lock);
283 		sbinfo->free_inodes++;
284 		spin_unlock(&sbinfo->stat_lock);
285 	}
286 
287 	clear_inode(inode);
288 	destroy_inode(inode);
289 }
290 
291 static void hugetlbfs_drop_inode(struct inode *inode)
292 {
293 	if (!inode->i_nlink)
294 		hugetlbfs_delete_inode(inode);
295 	else
296 		hugetlbfs_forget_inode(inode);
297 }
298 
299 /*
300  * h_pgoff is in HPAGE_SIZE units.
301  * vma->vm_pgoff is in PAGE_SIZE units.
302  */
303 static inline void
304 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
305 {
306 	struct vm_area_struct *vma;
307 	struct prio_tree_iter iter;
308 
309 	vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
310 		unsigned long h_vm_pgoff;
311 		unsigned long v_offset;
312 
313 		h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
314 		v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
315 		/*
316 		 * Is this VMA fully outside the truncation point?
317 		 */
318 		if (h_vm_pgoff >= h_pgoff)
319 			v_offset = 0;
320 
321 		unmap_hugepage_range(vma,
322 				vma->vm_start + v_offset, vma->vm_end);
323 	}
324 }
325 
326 /*
327  * Expanding truncates are not allowed.
328  */
329 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
330 {
331 	unsigned long pgoff;
332 	struct address_space *mapping = inode->i_mapping;
333 
334 	if (offset > inode->i_size)
335 		return -EINVAL;
336 
337 	BUG_ON(offset & ~HPAGE_MASK);
338 	pgoff = offset >> HPAGE_SHIFT;
339 
340 	inode->i_size = offset;
341 	spin_lock(&mapping->i_mmap_lock);
342 	if (!prio_tree_empty(&mapping->i_mmap))
343 		hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
344 	spin_unlock(&mapping->i_mmap_lock);
345 	truncate_hugepages(mapping, offset);
346 	return 0;
347 }
348 
349 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
350 {
351 	struct inode *inode = dentry->d_inode;
352 	int error;
353 	unsigned int ia_valid = attr->ia_valid;
354 
355 	BUG_ON(!inode);
356 
357 	error = inode_change_ok(inode, attr);
358 	if (error)
359 		goto out;
360 
361 	if (ia_valid & ATTR_SIZE) {
362 		error = -EINVAL;
363 		if (!(attr->ia_size & ~HPAGE_MASK))
364 			error = hugetlb_vmtruncate(inode, attr->ia_size);
365 		if (error)
366 			goto out;
367 		attr->ia_valid &= ~ATTR_SIZE;
368 	}
369 	error = inode_setattr(inode, attr);
370 out:
371 	return error;
372 }
373 
374 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
375 					gid_t gid, int mode, dev_t dev)
376 {
377 	struct inode *inode;
378 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
379 
380 	if (sbinfo->free_inodes >= 0) {
381 		spin_lock(&sbinfo->stat_lock);
382 		if (!sbinfo->free_inodes) {
383 			spin_unlock(&sbinfo->stat_lock);
384 			return NULL;
385 		}
386 		sbinfo->free_inodes--;
387 		spin_unlock(&sbinfo->stat_lock);
388 	}
389 
390 	inode = new_inode(sb);
391 	if (inode) {
392 		struct hugetlbfs_inode_info *info;
393 		inode->i_mode = mode;
394 		inode->i_uid = uid;
395 		inode->i_gid = gid;
396 		inode->i_blksize = HPAGE_SIZE;
397 		inode->i_blocks = 0;
398 		inode->i_mapping->a_ops = &hugetlbfs_aops;
399 		inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
400 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
401 		info = HUGETLBFS_I(inode);
402 		mpol_shared_policy_init(&info->policy);
403 		switch (mode & S_IFMT) {
404 		default:
405 			init_special_inode(inode, mode, dev);
406 			break;
407 		case S_IFREG:
408 			inode->i_op = &hugetlbfs_inode_operations;
409 			inode->i_fop = &hugetlbfs_file_operations;
410 			break;
411 		case S_IFDIR:
412 			inode->i_op = &hugetlbfs_dir_inode_operations;
413 			inode->i_fop = &simple_dir_operations;
414 
415 			/* directory inodes start off with i_nlink == 2 (for "." entry) */
416 			inode->i_nlink++;
417 			break;
418 		case S_IFLNK:
419 			inode->i_op = &page_symlink_inode_operations;
420 			break;
421 		}
422 	}
423 	return inode;
424 }
425 
426 /*
427  * File creation. Allocate an inode, and we're done..
428  */
429 static int hugetlbfs_mknod(struct inode *dir,
430 			struct dentry *dentry, int mode, dev_t dev)
431 {
432 	struct inode *inode;
433 	int error = -ENOSPC;
434 	gid_t gid;
435 
436 	if (dir->i_mode & S_ISGID) {
437 		gid = dir->i_gid;
438 		if (S_ISDIR(mode))
439 			mode |= S_ISGID;
440 	} else {
441 		gid = current->fsgid;
442 	}
443 	inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
444 	if (inode) {
445 		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
446 		d_instantiate(dentry, inode);
447 		dget(dentry);	/* Extra count - pin the dentry in core */
448 		error = 0;
449 	}
450 	return error;
451 }
452 
453 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
454 {
455 	int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
456 	if (!retval)
457 		dir->i_nlink++;
458 	return retval;
459 }
460 
461 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
462 {
463 	return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
464 }
465 
466 static int hugetlbfs_symlink(struct inode *dir,
467 			struct dentry *dentry, const char *symname)
468 {
469 	struct inode *inode;
470 	int error = -ENOSPC;
471 	gid_t gid;
472 
473 	if (dir->i_mode & S_ISGID)
474 		gid = dir->i_gid;
475 	else
476 		gid = current->fsgid;
477 
478 	inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
479 					gid, S_IFLNK|S_IRWXUGO, 0);
480 	if (inode) {
481 		int l = strlen(symname)+1;
482 		error = page_symlink(inode, symname, l);
483 		if (!error) {
484 			d_instantiate(dentry, inode);
485 			dget(dentry);
486 		} else
487 			iput(inode);
488 	}
489 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
490 
491 	return error;
492 }
493 
494 /*
495  * For direct-IO reads into hugetlb pages
496  */
497 static int hugetlbfs_set_page_dirty(struct page *page)
498 {
499 	return 0;
500 }
501 
502 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
503 {
504 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
505 
506 	buf->f_type = HUGETLBFS_MAGIC;
507 	buf->f_bsize = HPAGE_SIZE;
508 	if (sbinfo) {
509 		spin_lock(&sbinfo->stat_lock);
510 		buf->f_blocks = sbinfo->max_blocks;
511 		buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
512 		buf->f_files = sbinfo->max_inodes;
513 		buf->f_ffree = sbinfo->free_inodes;
514 		spin_unlock(&sbinfo->stat_lock);
515 	}
516 	buf->f_namelen = NAME_MAX;
517 	return 0;
518 }
519 
520 static void hugetlbfs_put_super(struct super_block *sb)
521 {
522 	struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
523 
524 	if (sbi) {
525 		sb->s_fs_info = NULL;
526 		kfree(sbi);
527 	}
528 }
529 
530 static kmem_cache_t *hugetlbfs_inode_cachep;
531 
532 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
533 {
534 	struct hugetlbfs_inode_info *p;
535 
536 	p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
537 	if (!p)
538 		return NULL;
539 	return &p->vfs_inode;
540 }
541 
542 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
543 {
544 	struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
545 
546 	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
547 	    SLAB_CTOR_CONSTRUCTOR)
548 		inode_init_once(&ei->vfs_inode);
549 }
550 
551 static void hugetlbfs_destroy_inode(struct inode *inode)
552 {
553 	mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
554 	kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
555 }
556 
557 static struct address_space_operations hugetlbfs_aops = {
558 	.readpage	= hugetlbfs_readpage,
559 	.prepare_write	= hugetlbfs_prepare_write,
560 	.commit_write	= hugetlbfs_commit_write,
561 	.set_page_dirty	= hugetlbfs_set_page_dirty,
562 };
563 
564 struct file_operations hugetlbfs_file_operations = {
565 	.mmap			= hugetlbfs_file_mmap,
566 	.fsync			= simple_sync_file,
567 	.get_unmapped_area	= hugetlb_get_unmapped_area,
568 };
569 
570 static struct inode_operations hugetlbfs_dir_inode_operations = {
571 	.create		= hugetlbfs_create,
572 	.lookup		= simple_lookup,
573 	.link		= simple_link,
574 	.unlink		= simple_unlink,
575 	.symlink	= hugetlbfs_symlink,
576 	.mkdir		= hugetlbfs_mkdir,
577 	.rmdir		= simple_rmdir,
578 	.mknod		= hugetlbfs_mknod,
579 	.rename		= simple_rename,
580 	.setattr	= hugetlbfs_setattr,
581 };
582 
583 static struct inode_operations hugetlbfs_inode_operations = {
584 	.setattr	= hugetlbfs_setattr,
585 };
586 
587 static struct super_operations hugetlbfs_ops = {
588 	.alloc_inode    = hugetlbfs_alloc_inode,
589 	.destroy_inode  = hugetlbfs_destroy_inode,
590 	.statfs		= hugetlbfs_statfs,
591 	.drop_inode	= hugetlbfs_drop_inode,
592 	.put_super	= hugetlbfs_put_super,
593 };
594 
595 static int
596 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
597 {
598 	char *opt, *value, *rest;
599 
600 	if (!options)
601 		return 0;
602 	while ((opt = strsep(&options, ",")) != NULL) {
603 		if (!*opt)
604 			continue;
605 
606 		value = strchr(opt, '=');
607 		if (!value || !*value)
608 			return -EINVAL;
609 		else
610 			*value++ = '\0';
611 
612 		if (!strcmp(opt, "uid"))
613 			pconfig->uid = simple_strtoul(value, &value, 0);
614 		else if (!strcmp(opt, "gid"))
615 			pconfig->gid = simple_strtoul(value, &value, 0);
616 		else if (!strcmp(opt, "mode"))
617 			pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
618 		else if (!strcmp(opt, "size")) {
619 			unsigned long long size = memparse(value, &rest);
620 			if (*rest == '%') {
621 				size <<= HPAGE_SHIFT;
622 				size *= max_huge_pages;
623 				do_div(size, 100);
624 				rest++;
625 			}
626 			size &= HPAGE_MASK;
627 			pconfig->nr_blocks = (size >> HPAGE_SHIFT);
628 			value = rest;
629 		} else if (!strcmp(opt,"nr_inodes")) {
630 			pconfig->nr_inodes = memparse(value, &rest);
631 			value = rest;
632 		} else
633 			return -EINVAL;
634 
635 		if (*value)
636 			return -EINVAL;
637 	}
638 	return 0;
639 }
640 
641 static int
642 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
643 {
644 	struct inode * inode;
645 	struct dentry * root;
646 	int ret;
647 	struct hugetlbfs_config config;
648 	struct hugetlbfs_sb_info *sbinfo;
649 
650 	config.nr_blocks = -1; /* No limit on size by default */
651 	config.nr_inodes = -1; /* No limit on number of inodes by default */
652 	config.uid = current->fsuid;
653 	config.gid = current->fsgid;
654 	config.mode = 0755;
655 	ret = hugetlbfs_parse_options(data, &config);
656 
657 	if (ret)
658 		return ret;
659 
660 	sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
661 	if (!sbinfo)
662 		return -ENOMEM;
663 	sb->s_fs_info = sbinfo;
664 	spin_lock_init(&sbinfo->stat_lock);
665 	sbinfo->max_blocks = config.nr_blocks;
666 	sbinfo->free_blocks = config.nr_blocks;
667 	sbinfo->max_inodes = config.nr_inodes;
668 	sbinfo->free_inodes = config.nr_inodes;
669 	sb->s_maxbytes = MAX_LFS_FILESIZE;
670 	sb->s_blocksize = HPAGE_SIZE;
671 	sb->s_blocksize_bits = HPAGE_SHIFT;
672 	sb->s_magic = HUGETLBFS_MAGIC;
673 	sb->s_op = &hugetlbfs_ops;
674 	sb->s_time_gran = 1;
675 	inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
676 					S_IFDIR | config.mode, 0);
677 	if (!inode)
678 		goto out_free;
679 
680 	root = d_alloc_root(inode);
681 	if (!root) {
682 		iput(inode);
683 		goto out_free;
684 	}
685 	sb->s_root = root;
686 	return 0;
687 out_free:
688 	kfree(sbinfo);
689 	return -ENOMEM;
690 }
691 
692 int hugetlb_get_quota(struct address_space *mapping)
693 {
694 	int ret = 0;
695 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
696 
697 	if (sbinfo->free_blocks > -1) {
698 		spin_lock(&sbinfo->stat_lock);
699 		if (sbinfo->free_blocks > 0)
700 			sbinfo->free_blocks--;
701 		else
702 			ret = -ENOMEM;
703 		spin_unlock(&sbinfo->stat_lock);
704 	}
705 
706 	return ret;
707 }
708 
709 void hugetlb_put_quota(struct address_space *mapping)
710 {
711 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
712 
713 	if (sbinfo->free_blocks > -1) {
714 		spin_lock(&sbinfo->stat_lock);
715 		sbinfo->free_blocks++;
716 		spin_unlock(&sbinfo->stat_lock);
717 	}
718 }
719 
720 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
721 	int flags, const char *dev_name, void *data)
722 {
723 	return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
724 }
725 
726 static struct file_system_type hugetlbfs_fs_type = {
727 	.name		= "hugetlbfs",
728 	.get_sb		= hugetlbfs_get_sb,
729 	.kill_sb	= kill_litter_super,
730 };
731 
732 static struct vfsmount *hugetlbfs_vfsmount;
733 
734 /*
735  * Return the next identifier for a shm file
736  */
737 static unsigned long hugetlbfs_counter(void)
738 {
739 	static DEFINE_SPINLOCK(lock);
740 	static unsigned long counter;
741 	unsigned long ret;
742 
743 	spin_lock(&lock);
744 	ret = ++counter;
745 	spin_unlock(&lock);
746 	return ret;
747 }
748 
749 static int can_do_hugetlb_shm(void)
750 {
751 	return likely(capable(CAP_IPC_LOCK) ||
752 			in_group_p(sysctl_hugetlb_shm_group) ||
753 			can_do_mlock());
754 }
755 
756 struct file *hugetlb_zero_setup(size_t size)
757 {
758 	int error = -ENOMEM;
759 	struct file *file;
760 	struct inode *inode;
761 	struct dentry *dentry, *root;
762 	struct qstr quick_string;
763 	char buf[16];
764 
765 	if (!can_do_hugetlb_shm())
766 		return ERR_PTR(-EPERM);
767 
768 	if (!is_hugepage_mem_enough(size))
769 		return ERR_PTR(-ENOMEM);
770 
771 	if (!user_shm_lock(size, current->user))
772 		return ERR_PTR(-ENOMEM);
773 
774 	root = hugetlbfs_vfsmount->mnt_root;
775 	snprintf(buf, 16, "%lu", hugetlbfs_counter());
776 	quick_string.name = buf;
777 	quick_string.len = strlen(quick_string.name);
778 	quick_string.hash = 0;
779 	dentry = d_alloc(root, &quick_string);
780 	if (!dentry)
781 		goto out_shm_unlock;
782 
783 	error = -ENFILE;
784 	file = get_empty_filp();
785 	if (!file)
786 		goto out_dentry;
787 
788 	error = -ENOSPC;
789 	inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
790 				current->fsgid, S_IFREG | S_IRWXUGO, 0);
791 	if (!inode)
792 		goto out_file;
793 
794 	d_instantiate(dentry, inode);
795 	inode->i_size = size;
796 	inode->i_nlink = 0;
797 	file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
798 	file->f_dentry = dentry;
799 	file->f_mapping = inode->i_mapping;
800 	file->f_op = &hugetlbfs_file_operations;
801 	file->f_mode = FMODE_WRITE | FMODE_READ;
802 	return file;
803 
804 out_file:
805 	put_filp(file);
806 out_dentry:
807 	dput(dentry);
808 out_shm_unlock:
809 	user_shm_unlock(size, current->user);
810 	return ERR_PTR(error);
811 }
812 
813 static int __init init_hugetlbfs_fs(void)
814 {
815 	int error;
816 	struct vfsmount *vfsmount;
817 
818 	hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
819 					sizeof(struct hugetlbfs_inode_info),
820 					0, 0, init_once, NULL);
821 	if (hugetlbfs_inode_cachep == NULL)
822 		return -ENOMEM;
823 
824 	error = register_filesystem(&hugetlbfs_fs_type);
825 	if (error)
826 		goto out;
827 
828 	vfsmount = kern_mount(&hugetlbfs_fs_type);
829 
830 	if (!IS_ERR(vfsmount)) {
831 		hugetlbfs_vfsmount = vfsmount;
832 		return 0;
833 	}
834 
835 	error = PTR_ERR(vfsmount);
836 
837  out:
838 	if (error)
839 		kmem_cache_destroy(hugetlbfs_inode_cachep);
840 	return error;
841 }
842 
843 static void __exit exit_hugetlbfs_fs(void)
844 {
845 	kmem_cache_destroy(hugetlbfs_inode_cachep);
846 	unregister_filesystem(&hugetlbfs_fs_type);
847 }
848 
849 module_init(init_hugetlbfs_fs)
850 module_exit(exit_hugetlbfs_fs)
851 
852 MODULE_LICENSE("GPL");
853