xref: /openbmc/linux/fs/hugetlbfs/inode.c (revision c21b37f6)
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/kernel.h>
17 #include <linux/writeback.h>
18 #include <linux/pagemap.h>
19 #include <linux/highmem.h>
20 #include <linux/init.h>
21 #include <linux/string.h>
22 #include <linux/capability.h>
23 #include <linux/ctype.h>
24 #include <linux/backing-dev.h>
25 #include <linux/hugetlb.h>
26 #include <linux/pagevec.h>
27 #include <linux/parser.h>
28 #include <linux/mman.h>
29 #include <linux/quotaops.h>
30 #include <linux/slab.h>
31 #include <linux/dnotify.h>
32 #include <linux/statfs.h>
33 #include <linux/security.h>
34 
35 #include <asm/uaccess.h>
36 
37 /* some random number */
38 #define HUGETLBFS_MAGIC	0x958458f6
39 
40 static const struct super_operations hugetlbfs_ops;
41 static const struct address_space_operations hugetlbfs_aops;
42 const struct file_operations hugetlbfs_file_operations;
43 static const struct inode_operations hugetlbfs_dir_inode_operations;
44 static const struct inode_operations hugetlbfs_inode_operations;
45 
46 static struct backing_dev_info hugetlbfs_backing_dev_info = {
47 	.ra_pages	= 0,	/* No readahead */
48 	.capabilities	= BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
49 };
50 
51 int sysctl_hugetlb_shm_group;
52 
53 enum {
54 	Opt_size, Opt_nr_inodes,
55 	Opt_mode, Opt_uid, Opt_gid,
56 	Opt_err,
57 };
58 
59 static match_table_t tokens = {
60 	{Opt_size,	"size=%s"},
61 	{Opt_nr_inodes,	"nr_inodes=%s"},
62 	{Opt_mode,	"mode=%o"},
63 	{Opt_uid,	"uid=%u"},
64 	{Opt_gid,	"gid=%u"},
65 	{Opt_err,	NULL},
66 };
67 
68 static void huge_pagevec_release(struct pagevec *pvec)
69 {
70 	int i;
71 
72 	for (i = 0; i < pagevec_count(pvec); ++i)
73 		put_page(pvec->pages[i]);
74 
75 	pagevec_reinit(pvec);
76 }
77 
78 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
79 {
80 	struct inode *inode = file->f_path.dentry->d_inode;
81 	loff_t len, vma_len;
82 	int ret;
83 
84 	/*
85 	 * vma alignment has already been checked by prepare_hugepage_range.
86 	 * If you add any error returns here, do so after setting VM_HUGETLB,
87 	 * so is_vm_hugetlb_page tests below unmap_region go the right way
88 	 * when do_mmap_pgoff unwinds (may be important on powerpc and ia64).
89 	 */
90 	vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
91 	vma->vm_ops = &hugetlb_vm_ops;
92 
93 	vma_len = (loff_t)(vma->vm_end - vma->vm_start);
94 
95 	mutex_lock(&inode->i_mutex);
96 	file_accessed(file);
97 
98 	ret = -ENOMEM;
99 	len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
100 
101 	if (vma->vm_flags & VM_MAYSHARE &&
102 	    hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
103 				  len >> HPAGE_SHIFT))
104 		goto out;
105 
106 	ret = 0;
107 	hugetlb_prefault_arch_hook(vma->vm_mm);
108 	if (vma->vm_flags & VM_WRITE && inode->i_size < len)
109 		inode->i_size = len;
110 out:
111 	mutex_unlock(&inode->i_mutex);
112 
113 	return ret;
114 }
115 
116 /*
117  * Called under down_write(mmap_sem).
118  */
119 
120 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
121 static unsigned long
122 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
123 		unsigned long len, unsigned long pgoff, unsigned long flags)
124 {
125 	struct mm_struct *mm = current->mm;
126 	struct vm_area_struct *vma;
127 	unsigned long start_addr;
128 
129 	if (len & ~HPAGE_MASK)
130 		return -EINVAL;
131 	if (len > TASK_SIZE)
132 		return -ENOMEM;
133 
134 	if (flags & MAP_FIXED) {
135 		if (prepare_hugepage_range(addr, len, pgoff))
136 			return -EINVAL;
137 		return addr;
138 	}
139 
140 	if (addr) {
141 		addr = ALIGN(addr, HPAGE_SIZE);
142 		vma = find_vma(mm, addr);
143 		if (TASK_SIZE - len >= addr &&
144 		    (!vma || addr + len <= vma->vm_start))
145 			return addr;
146 	}
147 
148 	start_addr = mm->free_area_cache;
149 
150 	if (len <= mm->cached_hole_size)
151 		start_addr = TASK_UNMAPPED_BASE;
152 
153 full_search:
154 	addr = ALIGN(start_addr, HPAGE_SIZE);
155 
156 	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
157 		/* At this point:  (!vma || addr < vma->vm_end). */
158 		if (TASK_SIZE - len < addr) {
159 			/*
160 			 * Start a new search - just in case we missed
161 			 * some holes.
162 			 */
163 			if (start_addr != TASK_UNMAPPED_BASE) {
164 				start_addr = TASK_UNMAPPED_BASE;
165 				goto full_search;
166 			}
167 			return -ENOMEM;
168 		}
169 
170 		if (!vma || addr + len <= vma->vm_start)
171 			return addr;
172 		addr = ALIGN(vma->vm_end, HPAGE_SIZE);
173 	}
174 }
175 #endif
176 
177 /*
178  * Read a page. Again trivial. If it didn't already exist
179  * in the page cache, it is zero-filled.
180  */
181 static int hugetlbfs_readpage(struct file *file, struct page * page)
182 {
183 	unlock_page(page);
184 	return -EINVAL;
185 }
186 
187 static int hugetlbfs_prepare_write(struct file *file,
188 			struct page *page, unsigned offset, unsigned to)
189 {
190 	return -EINVAL;
191 }
192 
193 static int hugetlbfs_commit_write(struct file *file,
194 			struct page *page, unsigned offset, unsigned to)
195 {
196 	return -EINVAL;
197 }
198 
199 static void truncate_huge_page(struct page *page)
200 {
201 	cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
202 	ClearPageUptodate(page);
203 	remove_from_page_cache(page);
204 	put_page(page);
205 }
206 
207 static void truncate_hugepages(struct inode *inode, loff_t lstart)
208 {
209 	struct address_space *mapping = &inode->i_data;
210 	const pgoff_t start = lstart >> HPAGE_SHIFT;
211 	struct pagevec pvec;
212 	pgoff_t next;
213 	int i, freed = 0;
214 
215 	pagevec_init(&pvec, 0);
216 	next = start;
217 	while (1) {
218 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
219 			if (next == start)
220 				break;
221 			next = start;
222 			continue;
223 		}
224 
225 		for (i = 0; i < pagevec_count(&pvec); ++i) {
226 			struct page *page = pvec.pages[i];
227 
228 			lock_page(page);
229 			if (page->index > next)
230 				next = page->index;
231 			++next;
232 			truncate_huge_page(page);
233 			unlock_page(page);
234 			hugetlb_put_quota(mapping);
235 			freed++;
236 		}
237 		huge_pagevec_release(&pvec);
238 	}
239 	BUG_ON(!lstart && mapping->nrpages);
240 	hugetlb_unreserve_pages(inode, start, freed);
241 }
242 
243 static void hugetlbfs_delete_inode(struct inode *inode)
244 {
245 	truncate_hugepages(inode, 0);
246 	clear_inode(inode);
247 }
248 
249 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
250 {
251 	struct super_block *sb = inode->i_sb;
252 
253 	if (!hlist_unhashed(&inode->i_hash)) {
254 		if (!(inode->i_state & (I_DIRTY|I_LOCK)))
255 			list_move(&inode->i_list, &inode_unused);
256 		inodes_stat.nr_unused++;
257 		if (!sb || (sb->s_flags & MS_ACTIVE)) {
258 			spin_unlock(&inode_lock);
259 			return;
260 		}
261 		inode->i_state |= I_WILL_FREE;
262 		spin_unlock(&inode_lock);
263 		/*
264 		 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
265 		 * in our backing_dev_info.
266 		 */
267 		write_inode_now(inode, 1);
268 		spin_lock(&inode_lock);
269 		inode->i_state &= ~I_WILL_FREE;
270 		inodes_stat.nr_unused--;
271 		hlist_del_init(&inode->i_hash);
272 	}
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 	truncate_hugepages(inode, 0);
279 	clear_inode(inode);
280 	destroy_inode(inode);
281 }
282 
283 static void hugetlbfs_drop_inode(struct inode *inode)
284 {
285 	if (!inode->i_nlink)
286 		generic_delete_inode(inode);
287 	else
288 		hugetlbfs_forget_inode(inode);
289 }
290 
291 static inline void
292 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
293 {
294 	struct vm_area_struct *vma;
295 	struct prio_tree_iter iter;
296 
297 	vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
298 		unsigned long v_offset;
299 
300 		/*
301 		 * Can the expression below overflow on 32-bit arches?
302 		 * No, because the prio_tree returns us only those vmas
303 		 * which overlap the truncated area starting at pgoff,
304 		 * and no vma on a 32-bit arch can span beyond the 4GB.
305 		 */
306 		if (vma->vm_pgoff < pgoff)
307 			v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
308 		else
309 			v_offset = 0;
310 
311 		__unmap_hugepage_range(vma,
312 				vma->vm_start + v_offset, vma->vm_end);
313 	}
314 }
315 
316 /*
317  * Expanding truncates are not allowed.
318  */
319 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
320 {
321 	pgoff_t pgoff;
322 	struct address_space *mapping = inode->i_mapping;
323 
324 	if (offset > inode->i_size)
325 		return -EINVAL;
326 
327 	BUG_ON(offset & ~HPAGE_MASK);
328 	pgoff = offset >> PAGE_SHIFT;
329 
330 	inode->i_size = offset;
331 	spin_lock(&mapping->i_mmap_lock);
332 	if (!prio_tree_empty(&mapping->i_mmap))
333 		hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
334 	spin_unlock(&mapping->i_mmap_lock);
335 	truncate_hugepages(inode, offset);
336 	return 0;
337 }
338 
339 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
340 {
341 	struct inode *inode = dentry->d_inode;
342 	int error;
343 	unsigned int ia_valid = attr->ia_valid;
344 
345 	BUG_ON(!inode);
346 
347 	error = inode_change_ok(inode, attr);
348 	if (error)
349 		goto out;
350 
351 	if (ia_valid & ATTR_SIZE) {
352 		error = -EINVAL;
353 		if (!(attr->ia_size & ~HPAGE_MASK))
354 			error = hugetlb_vmtruncate(inode, attr->ia_size);
355 		if (error)
356 			goto out;
357 		attr->ia_valid &= ~ATTR_SIZE;
358 	}
359 	error = inode_setattr(inode, attr);
360 out:
361 	return error;
362 }
363 
364 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
365 					gid_t gid, int mode, dev_t dev)
366 {
367 	struct inode *inode;
368 
369 	inode = new_inode(sb);
370 	if (inode) {
371 		struct hugetlbfs_inode_info *info;
372 		inode->i_mode = mode;
373 		inode->i_uid = uid;
374 		inode->i_gid = gid;
375 		inode->i_blocks = 0;
376 		inode->i_mapping->a_ops = &hugetlbfs_aops;
377 		inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
378 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
379 		INIT_LIST_HEAD(&inode->i_mapping->private_list);
380 		info = HUGETLBFS_I(inode);
381 		mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
382 		switch (mode & S_IFMT) {
383 		default:
384 			init_special_inode(inode, mode, dev);
385 			break;
386 		case S_IFREG:
387 			inode->i_op = &hugetlbfs_inode_operations;
388 			inode->i_fop = &hugetlbfs_file_operations;
389 			break;
390 		case S_IFDIR:
391 			inode->i_op = &hugetlbfs_dir_inode_operations;
392 			inode->i_fop = &simple_dir_operations;
393 
394 			/* directory inodes start off with i_nlink == 2 (for "." entry) */
395 			inc_nlink(inode);
396 			break;
397 		case S_IFLNK:
398 			inode->i_op = &page_symlink_inode_operations;
399 			break;
400 		}
401 	}
402 	return inode;
403 }
404 
405 /*
406  * File creation. Allocate an inode, and we're done..
407  */
408 static int hugetlbfs_mknod(struct inode *dir,
409 			struct dentry *dentry, int mode, dev_t dev)
410 {
411 	struct inode *inode;
412 	int error = -ENOSPC;
413 	gid_t gid;
414 
415 	if (dir->i_mode & S_ISGID) {
416 		gid = dir->i_gid;
417 		if (S_ISDIR(mode))
418 			mode |= S_ISGID;
419 	} else {
420 		gid = current->fsgid;
421 	}
422 	inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
423 	if (inode) {
424 		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
425 		d_instantiate(dentry, inode);
426 		dget(dentry);	/* Extra count - pin the dentry in core */
427 		error = 0;
428 	}
429 	return error;
430 }
431 
432 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
433 {
434 	int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
435 	if (!retval)
436 		inc_nlink(dir);
437 	return retval;
438 }
439 
440 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
441 {
442 	return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
443 }
444 
445 static int hugetlbfs_symlink(struct inode *dir,
446 			struct dentry *dentry, const char *symname)
447 {
448 	struct inode *inode;
449 	int error = -ENOSPC;
450 	gid_t gid;
451 
452 	if (dir->i_mode & S_ISGID)
453 		gid = dir->i_gid;
454 	else
455 		gid = current->fsgid;
456 
457 	inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
458 					gid, S_IFLNK|S_IRWXUGO, 0);
459 	if (inode) {
460 		int l = strlen(symname)+1;
461 		error = page_symlink(inode, symname, l);
462 		if (!error) {
463 			d_instantiate(dentry, inode);
464 			dget(dentry);
465 		} else
466 			iput(inode);
467 	}
468 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
469 
470 	return error;
471 }
472 
473 /*
474  * mark the head page dirty
475  */
476 static int hugetlbfs_set_page_dirty(struct page *page)
477 {
478 	struct page *head = compound_head(page);
479 
480 	SetPageDirty(head);
481 	return 0;
482 }
483 
484 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
485 {
486 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
487 
488 	buf->f_type = HUGETLBFS_MAGIC;
489 	buf->f_bsize = HPAGE_SIZE;
490 	if (sbinfo) {
491 		spin_lock(&sbinfo->stat_lock);
492 		/* If no limits set, just report 0 for max/free/used
493 		 * blocks, like simple_statfs() */
494 		if (sbinfo->max_blocks >= 0) {
495 			buf->f_blocks = sbinfo->max_blocks;
496 			buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
497 			buf->f_files = sbinfo->max_inodes;
498 			buf->f_ffree = sbinfo->free_inodes;
499 		}
500 		spin_unlock(&sbinfo->stat_lock);
501 	}
502 	buf->f_namelen = NAME_MAX;
503 	return 0;
504 }
505 
506 static void hugetlbfs_put_super(struct super_block *sb)
507 {
508 	struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
509 
510 	if (sbi) {
511 		sb->s_fs_info = NULL;
512 		kfree(sbi);
513 	}
514 }
515 
516 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
517 {
518 	if (sbinfo->free_inodes >= 0) {
519 		spin_lock(&sbinfo->stat_lock);
520 		if (unlikely(!sbinfo->free_inodes)) {
521 			spin_unlock(&sbinfo->stat_lock);
522 			return 0;
523 		}
524 		sbinfo->free_inodes--;
525 		spin_unlock(&sbinfo->stat_lock);
526 	}
527 
528 	return 1;
529 }
530 
531 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
532 {
533 	if (sbinfo->free_inodes >= 0) {
534 		spin_lock(&sbinfo->stat_lock);
535 		sbinfo->free_inodes++;
536 		spin_unlock(&sbinfo->stat_lock);
537 	}
538 }
539 
540 
541 static struct kmem_cache *hugetlbfs_inode_cachep;
542 
543 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
544 {
545 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
546 	struct hugetlbfs_inode_info *p;
547 
548 	if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
549 		return NULL;
550 	p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
551 	if (unlikely(!p)) {
552 		hugetlbfs_inc_free_inodes(sbinfo);
553 		return NULL;
554 	}
555 	return &p->vfs_inode;
556 }
557 
558 static void hugetlbfs_destroy_inode(struct inode *inode)
559 {
560 	hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
561 	mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
562 	kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
563 }
564 
565 static const struct address_space_operations hugetlbfs_aops = {
566 	.readpage	= hugetlbfs_readpage,
567 	.prepare_write	= hugetlbfs_prepare_write,
568 	.commit_write	= hugetlbfs_commit_write,
569 	.set_page_dirty	= hugetlbfs_set_page_dirty,
570 };
571 
572 
573 static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
574 {
575 	struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
576 
577 	inode_init_once(&ei->vfs_inode);
578 }
579 
580 const struct file_operations hugetlbfs_file_operations = {
581 	.mmap			= hugetlbfs_file_mmap,
582 	.fsync			= simple_sync_file,
583 	.get_unmapped_area	= hugetlb_get_unmapped_area,
584 };
585 
586 static const struct inode_operations hugetlbfs_dir_inode_operations = {
587 	.create		= hugetlbfs_create,
588 	.lookup		= simple_lookup,
589 	.link		= simple_link,
590 	.unlink		= simple_unlink,
591 	.symlink	= hugetlbfs_symlink,
592 	.mkdir		= hugetlbfs_mkdir,
593 	.rmdir		= simple_rmdir,
594 	.mknod		= hugetlbfs_mknod,
595 	.rename		= simple_rename,
596 	.setattr	= hugetlbfs_setattr,
597 };
598 
599 static const struct inode_operations hugetlbfs_inode_operations = {
600 	.setattr	= hugetlbfs_setattr,
601 };
602 
603 static const struct super_operations hugetlbfs_ops = {
604 	.alloc_inode    = hugetlbfs_alloc_inode,
605 	.destroy_inode  = hugetlbfs_destroy_inode,
606 	.statfs		= hugetlbfs_statfs,
607 	.delete_inode	= hugetlbfs_delete_inode,
608 	.drop_inode	= hugetlbfs_drop_inode,
609 	.put_super	= hugetlbfs_put_super,
610 };
611 
612 static int
613 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
614 {
615 	char *p, *rest;
616 	substring_t args[MAX_OPT_ARGS];
617 	int option;
618 
619 	if (!options)
620 		return 0;
621 
622 	while ((p = strsep(&options, ",")) != NULL) {
623 		int token;
624 		if (!*p)
625 			continue;
626 
627 		token = match_token(p, tokens, args);
628 		switch (token) {
629 		case Opt_uid:
630 			if (match_int(&args[0], &option))
631  				goto bad_val;
632 			pconfig->uid = option;
633 			break;
634 
635 		case Opt_gid:
636 			if (match_int(&args[0], &option))
637  				goto bad_val;
638 			pconfig->gid = option;
639 			break;
640 
641 		case Opt_mode:
642 			if (match_octal(&args[0], &option))
643  				goto bad_val;
644 			pconfig->mode = option & 0777U;
645 			break;
646 
647 		case Opt_size: {
648  			unsigned long long size;
649 			/* memparse() will accept a K/M/G without a digit */
650 			if (!isdigit(*args[0].from))
651 				goto bad_val;
652 			size = memparse(args[0].from, &rest);
653 			if (*rest == '%') {
654 				size <<= HPAGE_SHIFT;
655 				size *= max_huge_pages;
656 				do_div(size, 100);
657 			}
658 			pconfig->nr_blocks = (size >> HPAGE_SHIFT);
659 			break;
660 		}
661 
662 		case Opt_nr_inodes:
663 			/* memparse() will accept a K/M/G without a digit */
664 			if (!isdigit(*args[0].from))
665 				goto bad_val;
666 			pconfig->nr_inodes = memparse(args[0].from, &rest);
667 			break;
668 
669 		default:
670 			printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
671 				 p);
672 			return -EINVAL;
673 			break;
674 		}
675 	}
676 	return 0;
677 
678 bad_val:
679  	printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
680 	       args[0].from, p);
681  	return 1;
682 }
683 
684 static int
685 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
686 {
687 	struct inode * inode;
688 	struct dentry * root;
689 	int ret;
690 	struct hugetlbfs_config config;
691 	struct hugetlbfs_sb_info *sbinfo;
692 
693 	config.nr_blocks = -1; /* No limit on size by default */
694 	config.nr_inodes = -1; /* No limit on number of inodes by default */
695 	config.uid = current->fsuid;
696 	config.gid = current->fsgid;
697 	config.mode = 0755;
698 	ret = hugetlbfs_parse_options(data, &config);
699 	if (ret)
700 		return ret;
701 
702 	sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
703 	if (!sbinfo)
704 		return -ENOMEM;
705 	sb->s_fs_info = sbinfo;
706 	spin_lock_init(&sbinfo->stat_lock);
707 	sbinfo->max_blocks = config.nr_blocks;
708 	sbinfo->free_blocks = config.nr_blocks;
709 	sbinfo->max_inodes = config.nr_inodes;
710 	sbinfo->free_inodes = config.nr_inodes;
711 	sb->s_maxbytes = MAX_LFS_FILESIZE;
712 	sb->s_blocksize = HPAGE_SIZE;
713 	sb->s_blocksize_bits = HPAGE_SHIFT;
714 	sb->s_magic = HUGETLBFS_MAGIC;
715 	sb->s_op = &hugetlbfs_ops;
716 	sb->s_time_gran = 1;
717 	inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
718 					S_IFDIR | config.mode, 0);
719 	if (!inode)
720 		goto out_free;
721 
722 	root = d_alloc_root(inode);
723 	if (!root) {
724 		iput(inode);
725 		goto out_free;
726 	}
727 	sb->s_root = root;
728 	return 0;
729 out_free:
730 	kfree(sbinfo);
731 	return -ENOMEM;
732 }
733 
734 int hugetlb_get_quota(struct address_space *mapping)
735 {
736 	int ret = 0;
737 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
738 
739 	if (sbinfo->free_blocks > -1) {
740 		spin_lock(&sbinfo->stat_lock);
741 		if (sbinfo->free_blocks > 0)
742 			sbinfo->free_blocks--;
743 		else
744 			ret = -ENOMEM;
745 		spin_unlock(&sbinfo->stat_lock);
746 	}
747 
748 	return ret;
749 }
750 
751 void hugetlb_put_quota(struct address_space *mapping)
752 {
753 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
754 
755 	if (sbinfo->free_blocks > -1) {
756 		spin_lock(&sbinfo->stat_lock);
757 		sbinfo->free_blocks++;
758 		spin_unlock(&sbinfo->stat_lock);
759 	}
760 }
761 
762 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
763 	int flags, const char *dev_name, void *data, struct vfsmount *mnt)
764 {
765 	return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
766 }
767 
768 static struct file_system_type hugetlbfs_fs_type = {
769 	.name		= "hugetlbfs",
770 	.get_sb		= hugetlbfs_get_sb,
771 	.kill_sb	= kill_litter_super,
772 };
773 
774 static struct vfsmount *hugetlbfs_vfsmount;
775 
776 static int can_do_hugetlb_shm(void)
777 {
778 	return likely(capable(CAP_IPC_LOCK) ||
779 			in_group_p(sysctl_hugetlb_shm_group) ||
780 			can_do_mlock());
781 }
782 
783 struct file *hugetlb_file_setup(const char *name, size_t size)
784 {
785 	int error = -ENOMEM;
786 	struct file *file;
787 	struct inode *inode;
788 	struct dentry *dentry, *root;
789 	struct qstr quick_string;
790 
791 	if (!hugetlbfs_vfsmount)
792 		return ERR_PTR(-ENOENT);
793 
794 	if (!can_do_hugetlb_shm())
795 		return ERR_PTR(-EPERM);
796 
797 	if (!user_shm_lock(size, current->user))
798 		return ERR_PTR(-ENOMEM);
799 
800 	root = hugetlbfs_vfsmount->mnt_root;
801 	quick_string.name = name;
802 	quick_string.len = strlen(quick_string.name);
803 	quick_string.hash = 0;
804 	dentry = d_alloc(root, &quick_string);
805 	if (!dentry)
806 		goto out_shm_unlock;
807 
808 	error = -ENFILE;
809 	file = get_empty_filp();
810 	if (!file)
811 		goto out_dentry;
812 
813 	error = -ENOSPC;
814 	inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
815 				current->fsgid, S_IFREG | S_IRWXUGO, 0);
816 	if (!inode)
817 		goto out_file;
818 
819 	error = -ENOMEM;
820 	if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
821 		goto out_inode;
822 
823 	d_instantiate(dentry, inode);
824 	inode->i_size = size;
825 	inode->i_nlink = 0;
826 	file->f_path.mnt = mntget(hugetlbfs_vfsmount);
827 	file->f_path.dentry = dentry;
828 	file->f_mapping = inode->i_mapping;
829 	file->f_op = &hugetlbfs_file_operations;
830 	file->f_mode = FMODE_WRITE | FMODE_READ;
831 	return file;
832 
833 out_inode:
834 	iput(inode);
835 out_file:
836 	put_filp(file);
837 out_dentry:
838 	dput(dentry);
839 out_shm_unlock:
840 	user_shm_unlock(size, current->user);
841 	return ERR_PTR(error);
842 }
843 
844 static int __init init_hugetlbfs_fs(void)
845 {
846 	int error;
847 	struct vfsmount *vfsmount;
848 
849 	hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
850 					sizeof(struct hugetlbfs_inode_info),
851 					0, 0, init_once);
852 	if (hugetlbfs_inode_cachep == NULL)
853 		return -ENOMEM;
854 
855 	error = register_filesystem(&hugetlbfs_fs_type);
856 	if (error)
857 		goto out;
858 
859 	vfsmount = kern_mount(&hugetlbfs_fs_type);
860 
861 	if (!IS_ERR(vfsmount)) {
862 		hugetlbfs_vfsmount = vfsmount;
863 		return 0;
864 	}
865 
866 	error = PTR_ERR(vfsmount);
867 
868  out:
869 	if (error)
870 		kmem_cache_destroy(hugetlbfs_inode_cachep);
871 	return error;
872 }
873 
874 static void __exit exit_hugetlbfs_fs(void)
875 {
876 	kmem_cache_destroy(hugetlbfs_inode_cachep);
877 	unregister_filesystem(&hugetlbfs_fs_type);
878 }
879 
880 module_init(init_hugetlbfs_fs)
881 module_exit(exit_hugetlbfs_fs)
882 
883 MODULE_LICENSE("GPL");
884