xref: /linux/fs/hugetlbfs/inode.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
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/slab.h>
30 #include <linux/dnotify.h>
31 #include <linux/statfs.h>
32 #include <linux/security.h>
33 #include <linux/magic.h>
34 
35 #include <asm/uaccess.h>
36 
37 static const struct super_operations hugetlbfs_ops;
38 static const struct address_space_operations hugetlbfs_aops;
39 const struct file_operations hugetlbfs_file_operations;
40 static const struct inode_operations hugetlbfs_dir_inode_operations;
41 static const struct inode_operations hugetlbfs_inode_operations;
42 
43 static struct backing_dev_info hugetlbfs_backing_dev_info = {
44 	.name		= "hugetlbfs",
45 	.ra_pages	= 0,	/* No readahead */
46 	.capabilities	= BDI_CAP_NO_ACCT_AND_WRITEBACK,
47 };
48 
49 int sysctl_hugetlb_shm_group;
50 
51 enum {
52 	Opt_size, Opt_nr_inodes,
53 	Opt_mode, Opt_uid, Opt_gid,
54 	Opt_pagesize,
55 	Opt_err,
56 };
57 
58 static const match_table_t tokens = {
59 	{Opt_size,	"size=%s"},
60 	{Opt_nr_inodes,	"nr_inodes=%s"},
61 	{Opt_mode,	"mode=%o"},
62 	{Opt_uid,	"uid=%u"},
63 	{Opt_gid,	"gid=%u"},
64 	{Opt_pagesize,	"pagesize=%s"},
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 	struct hstate *h = hstate_file(file);
84 
85 	/*
86 	 * vma address alignment (but not the pgoff alignment) has
87 	 * already been checked by prepare_hugepage_range.  If you add
88 	 * any error returns here, do so after setting VM_HUGETLB, so
89 	 * is_vm_hugetlb_page tests below unmap_region go the right
90 	 * way when do_mmap_pgoff unwinds (may be important on powerpc
91 	 * and ia64).
92 	 */
93 	vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
94 	vma->vm_ops = &hugetlb_vm_ops;
95 
96 	if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
97 		return -EINVAL;
98 
99 	vma_len = (loff_t)(vma->vm_end - vma->vm_start);
100 
101 	mutex_lock(&inode->i_mutex);
102 	file_accessed(file);
103 
104 	ret = -ENOMEM;
105 	len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
106 
107 	if (hugetlb_reserve_pages(inode,
108 				vma->vm_pgoff >> huge_page_order(h),
109 				len >> huge_page_shift(h), vma,
110 				vma->vm_flags))
111 		goto out;
112 
113 	ret = 0;
114 	hugetlb_prefault_arch_hook(vma->vm_mm);
115 	if (vma->vm_flags & VM_WRITE && inode->i_size < len)
116 		inode->i_size = len;
117 out:
118 	mutex_unlock(&inode->i_mutex);
119 
120 	return ret;
121 }
122 
123 /*
124  * Called under down_write(mmap_sem).
125  */
126 
127 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
128 static unsigned long
129 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
130 		unsigned long len, unsigned long pgoff, unsigned long flags)
131 {
132 	struct mm_struct *mm = current->mm;
133 	struct vm_area_struct *vma;
134 	unsigned long start_addr;
135 	struct hstate *h = hstate_file(file);
136 
137 	if (len & ~huge_page_mask(h))
138 		return -EINVAL;
139 	if (len > TASK_SIZE)
140 		return -ENOMEM;
141 
142 	if (flags & MAP_FIXED) {
143 		if (prepare_hugepage_range(file, addr, len))
144 			return -EINVAL;
145 		return addr;
146 	}
147 
148 	if (addr) {
149 		addr = ALIGN(addr, huge_page_size(h));
150 		vma = find_vma(mm, addr);
151 		if (TASK_SIZE - len >= addr &&
152 		    (!vma || addr + len <= vma->vm_start))
153 			return addr;
154 	}
155 
156 	start_addr = mm->free_area_cache;
157 
158 	if (len <= mm->cached_hole_size)
159 		start_addr = TASK_UNMAPPED_BASE;
160 
161 full_search:
162 	addr = ALIGN(start_addr, huge_page_size(h));
163 
164 	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
165 		/* At this point:  (!vma || addr < vma->vm_end). */
166 		if (TASK_SIZE - len < addr) {
167 			/*
168 			 * Start a new search - just in case we missed
169 			 * some holes.
170 			 */
171 			if (start_addr != TASK_UNMAPPED_BASE) {
172 				start_addr = TASK_UNMAPPED_BASE;
173 				goto full_search;
174 			}
175 			return -ENOMEM;
176 		}
177 
178 		if (!vma || addr + len <= vma->vm_start)
179 			return addr;
180 		addr = ALIGN(vma->vm_end, huge_page_size(h));
181 	}
182 }
183 #endif
184 
185 static int
186 hugetlbfs_read_actor(struct page *page, unsigned long offset,
187 			char __user *buf, unsigned long count,
188 			unsigned long size)
189 {
190 	char *kaddr;
191 	unsigned long left, copied = 0;
192 	int i, chunksize;
193 
194 	if (size > count)
195 		size = count;
196 
197 	/* Find which 4k chunk and offset with in that chunk */
198 	i = offset >> PAGE_CACHE_SHIFT;
199 	offset = offset & ~PAGE_CACHE_MASK;
200 
201 	while (size) {
202 		chunksize = PAGE_CACHE_SIZE;
203 		if (offset)
204 			chunksize -= offset;
205 		if (chunksize > size)
206 			chunksize = size;
207 		kaddr = kmap(&page[i]);
208 		left = __copy_to_user(buf, kaddr + offset, chunksize);
209 		kunmap(&page[i]);
210 		if (left) {
211 			copied += (chunksize - left);
212 			break;
213 		}
214 		offset = 0;
215 		size -= chunksize;
216 		buf += chunksize;
217 		copied += chunksize;
218 		i++;
219 	}
220 	return copied ? copied : -EFAULT;
221 }
222 
223 /*
224  * Support for read() - Find the page attached to f_mapping and copy out the
225  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
226  * since it has PAGE_CACHE_SIZE assumptions.
227  */
228 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
229 			      size_t len, loff_t *ppos)
230 {
231 	struct hstate *h = hstate_file(filp);
232 	struct address_space *mapping = filp->f_mapping;
233 	struct inode *inode = mapping->host;
234 	unsigned long index = *ppos >> huge_page_shift(h);
235 	unsigned long offset = *ppos & ~huge_page_mask(h);
236 	unsigned long end_index;
237 	loff_t isize;
238 	ssize_t retval = 0;
239 
240 	mutex_lock(&inode->i_mutex);
241 
242 	/* validate length */
243 	if (len == 0)
244 		goto out;
245 
246 	isize = i_size_read(inode);
247 	if (!isize)
248 		goto out;
249 
250 	end_index = (isize - 1) >> huge_page_shift(h);
251 	for (;;) {
252 		struct page *page;
253 		unsigned long nr, ret;
254 		int ra;
255 
256 		/* nr is the maximum number of bytes to copy from this page */
257 		nr = huge_page_size(h);
258 		if (index >= end_index) {
259 			if (index > end_index)
260 				goto out;
261 			nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
262 			if (nr <= offset) {
263 				goto out;
264 			}
265 		}
266 		nr = nr - offset;
267 
268 		/* Find the page */
269 		page = find_get_page(mapping, index);
270 		if (unlikely(page == NULL)) {
271 			/*
272 			 * We have a HOLE, zero out the user-buffer for the
273 			 * length of the hole or request.
274 			 */
275 			ret = len < nr ? len : nr;
276 			if (clear_user(buf, ret))
277 				ra = -EFAULT;
278 			else
279 				ra = 0;
280 		} else {
281 			/*
282 			 * We have the page, copy it to user space buffer.
283 			 */
284 			ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
285 			ret = ra;
286 		}
287 		if (ra < 0) {
288 			if (retval == 0)
289 				retval = ra;
290 			if (page)
291 				page_cache_release(page);
292 			goto out;
293 		}
294 
295 		offset += ret;
296 		retval += ret;
297 		len -= ret;
298 		index += offset >> huge_page_shift(h);
299 		offset &= ~huge_page_mask(h);
300 
301 		if (page)
302 			page_cache_release(page);
303 
304 		/* short read or no more work */
305 		if ((ret != nr) || (len == 0))
306 			break;
307 	}
308 out:
309 	*ppos = ((loff_t)index << huge_page_shift(h)) + offset;
310 	mutex_unlock(&inode->i_mutex);
311 	return retval;
312 }
313 
314 static int hugetlbfs_write_begin(struct file *file,
315 			struct address_space *mapping,
316 			loff_t pos, unsigned len, unsigned flags,
317 			struct page **pagep, void **fsdata)
318 {
319 	return -EINVAL;
320 }
321 
322 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
323 			loff_t pos, unsigned len, unsigned copied,
324 			struct page *page, void *fsdata)
325 {
326 	BUG();
327 	return -EINVAL;
328 }
329 
330 static void truncate_huge_page(struct page *page)
331 {
332 	cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
333 	ClearPageUptodate(page);
334 	remove_from_page_cache(page);
335 	put_page(page);
336 }
337 
338 static void truncate_hugepages(struct inode *inode, loff_t lstart)
339 {
340 	struct hstate *h = hstate_inode(inode);
341 	struct address_space *mapping = &inode->i_data;
342 	const pgoff_t start = lstart >> huge_page_shift(h);
343 	struct pagevec pvec;
344 	pgoff_t next;
345 	int i, freed = 0;
346 
347 	pagevec_init(&pvec, 0);
348 	next = start;
349 	while (1) {
350 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
351 			if (next == start)
352 				break;
353 			next = start;
354 			continue;
355 		}
356 
357 		for (i = 0; i < pagevec_count(&pvec); ++i) {
358 			struct page *page = pvec.pages[i];
359 
360 			lock_page(page);
361 			if (page->index > next)
362 				next = page->index;
363 			++next;
364 			truncate_huge_page(page);
365 			unlock_page(page);
366 			freed++;
367 		}
368 		huge_pagevec_release(&pvec);
369 	}
370 	BUG_ON(!lstart && mapping->nrpages);
371 	hugetlb_unreserve_pages(inode, start, freed);
372 }
373 
374 static void hugetlbfs_evict_inode(struct inode *inode)
375 {
376 	truncate_hugepages(inode, 0);
377 	end_writeback(inode);
378 }
379 
380 static inline void
381 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
382 {
383 	struct vm_area_struct *vma;
384 	struct prio_tree_iter iter;
385 
386 	vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
387 		unsigned long v_offset;
388 
389 		/*
390 		 * Can the expression below overflow on 32-bit arches?
391 		 * No, because the prio_tree returns us only those vmas
392 		 * which overlap the truncated area starting at pgoff,
393 		 * and no vma on a 32-bit arch can span beyond the 4GB.
394 		 */
395 		if (vma->vm_pgoff < pgoff)
396 			v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
397 		else
398 			v_offset = 0;
399 
400 		__unmap_hugepage_range(vma,
401 				vma->vm_start + v_offset, vma->vm_end, NULL);
402 	}
403 }
404 
405 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
406 {
407 	pgoff_t pgoff;
408 	struct address_space *mapping = inode->i_mapping;
409 	struct hstate *h = hstate_inode(inode);
410 
411 	BUG_ON(offset & ~huge_page_mask(h));
412 	pgoff = offset >> PAGE_SHIFT;
413 
414 	i_size_write(inode, offset);
415 	spin_lock(&mapping->i_mmap_lock);
416 	if (!prio_tree_empty(&mapping->i_mmap))
417 		hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
418 	spin_unlock(&mapping->i_mmap_lock);
419 	truncate_hugepages(inode, offset);
420 	return 0;
421 }
422 
423 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
424 {
425 	struct inode *inode = dentry->d_inode;
426 	struct hstate *h = hstate_inode(inode);
427 	int error;
428 	unsigned int ia_valid = attr->ia_valid;
429 
430 	BUG_ON(!inode);
431 
432 	error = inode_change_ok(inode, attr);
433 	if (error)
434 		return error;
435 
436 	if (ia_valid & ATTR_SIZE) {
437 		error = -EINVAL;
438 		if (attr->ia_size & ~huge_page_mask(h))
439 			return -EINVAL;
440 		error = hugetlb_vmtruncate(inode, attr->ia_size);
441 		if (error)
442 			return error;
443 	}
444 
445 	setattr_copy(inode, attr);
446 	mark_inode_dirty(inode);
447 	return 0;
448 }
449 
450 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
451 					gid_t gid, int mode, dev_t dev)
452 {
453 	struct inode *inode;
454 
455 	inode = new_inode(sb);
456 	if (inode) {
457 		struct hugetlbfs_inode_info *info;
458 		inode->i_mode = mode;
459 		inode->i_uid = uid;
460 		inode->i_gid = gid;
461 		inode->i_mapping->a_ops = &hugetlbfs_aops;
462 		inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
463 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
464 		INIT_LIST_HEAD(&inode->i_mapping->private_list);
465 		info = HUGETLBFS_I(inode);
466 		/*
467 		 * The policy is initialized here even if we are creating a
468 		 * private inode because initialization simply creates an
469 		 * an empty rb tree and calls spin_lock_init(), later when we
470 		 * call mpol_free_shared_policy() it will just return because
471 		 * the rb tree will still be empty.
472 		 */
473 		mpol_shared_policy_init(&info->policy, NULL);
474 		switch (mode & S_IFMT) {
475 		default:
476 			init_special_inode(inode, mode, dev);
477 			break;
478 		case S_IFREG:
479 			inode->i_op = &hugetlbfs_inode_operations;
480 			inode->i_fop = &hugetlbfs_file_operations;
481 			break;
482 		case S_IFDIR:
483 			inode->i_op = &hugetlbfs_dir_inode_operations;
484 			inode->i_fop = &simple_dir_operations;
485 
486 			/* directory inodes start off with i_nlink == 2 (for "." entry) */
487 			inc_nlink(inode);
488 			break;
489 		case S_IFLNK:
490 			inode->i_op = &page_symlink_inode_operations;
491 			break;
492 		}
493 	}
494 	return inode;
495 }
496 
497 /*
498  * File creation. Allocate an inode, and we're done..
499  */
500 static int hugetlbfs_mknod(struct inode *dir,
501 			struct dentry *dentry, int mode, dev_t dev)
502 {
503 	struct inode *inode;
504 	int error = -ENOSPC;
505 	gid_t gid;
506 
507 	if (dir->i_mode & S_ISGID) {
508 		gid = dir->i_gid;
509 		if (S_ISDIR(mode))
510 			mode |= S_ISGID;
511 	} else {
512 		gid = current_fsgid();
513 	}
514 	inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
515 	if (inode) {
516 		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
517 		d_instantiate(dentry, inode);
518 		dget(dentry);	/* Extra count - pin the dentry in core */
519 		error = 0;
520 	}
521 	return error;
522 }
523 
524 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
525 {
526 	int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
527 	if (!retval)
528 		inc_nlink(dir);
529 	return retval;
530 }
531 
532 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
533 {
534 	return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
535 }
536 
537 static int hugetlbfs_symlink(struct inode *dir,
538 			struct dentry *dentry, const char *symname)
539 {
540 	struct inode *inode;
541 	int error = -ENOSPC;
542 	gid_t gid;
543 
544 	if (dir->i_mode & S_ISGID)
545 		gid = dir->i_gid;
546 	else
547 		gid = current_fsgid();
548 
549 	inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
550 					gid, S_IFLNK|S_IRWXUGO, 0);
551 	if (inode) {
552 		int l = strlen(symname)+1;
553 		error = page_symlink(inode, symname, l);
554 		if (!error) {
555 			d_instantiate(dentry, inode);
556 			dget(dentry);
557 		} else
558 			iput(inode);
559 	}
560 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
561 
562 	return error;
563 }
564 
565 /*
566  * mark the head page dirty
567  */
568 static int hugetlbfs_set_page_dirty(struct page *page)
569 {
570 	struct page *head = compound_head(page);
571 
572 	SetPageDirty(head);
573 	return 0;
574 }
575 
576 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
577 {
578 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
579 	struct hstate *h = hstate_inode(dentry->d_inode);
580 
581 	buf->f_type = HUGETLBFS_MAGIC;
582 	buf->f_bsize = huge_page_size(h);
583 	if (sbinfo) {
584 		spin_lock(&sbinfo->stat_lock);
585 		/* If no limits set, just report 0 for max/free/used
586 		 * blocks, like simple_statfs() */
587 		if (sbinfo->max_blocks >= 0) {
588 			buf->f_blocks = sbinfo->max_blocks;
589 			buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
590 			buf->f_files = sbinfo->max_inodes;
591 			buf->f_ffree = sbinfo->free_inodes;
592 		}
593 		spin_unlock(&sbinfo->stat_lock);
594 	}
595 	buf->f_namelen = NAME_MAX;
596 	return 0;
597 }
598 
599 static void hugetlbfs_put_super(struct super_block *sb)
600 {
601 	struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
602 
603 	if (sbi) {
604 		sb->s_fs_info = NULL;
605 		kfree(sbi);
606 	}
607 }
608 
609 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
610 {
611 	if (sbinfo->free_inodes >= 0) {
612 		spin_lock(&sbinfo->stat_lock);
613 		if (unlikely(!sbinfo->free_inodes)) {
614 			spin_unlock(&sbinfo->stat_lock);
615 			return 0;
616 		}
617 		sbinfo->free_inodes--;
618 		spin_unlock(&sbinfo->stat_lock);
619 	}
620 
621 	return 1;
622 }
623 
624 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
625 {
626 	if (sbinfo->free_inodes >= 0) {
627 		spin_lock(&sbinfo->stat_lock);
628 		sbinfo->free_inodes++;
629 		spin_unlock(&sbinfo->stat_lock);
630 	}
631 }
632 
633 
634 static struct kmem_cache *hugetlbfs_inode_cachep;
635 
636 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
637 {
638 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
639 	struct hugetlbfs_inode_info *p;
640 
641 	if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
642 		return NULL;
643 	p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
644 	if (unlikely(!p)) {
645 		hugetlbfs_inc_free_inodes(sbinfo);
646 		return NULL;
647 	}
648 	return &p->vfs_inode;
649 }
650 
651 static void hugetlbfs_destroy_inode(struct inode *inode)
652 {
653 	hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
654 	mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
655 	kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
656 }
657 
658 static const struct address_space_operations hugetlbfs_aops = {
659 	.write_begin	= hugetlbfs_write_begin,
660 	.write_end	= hugetlbfs_write_end,
661 	.set_page_dirty	= hugetlbfs_set_page_dirty,
662 };
663 
664 
665 static void init_once(void *foo)
666 {
667 	struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
668 
669 	inode_init_once(&ei->vfs_inode);
670 }
671 
672 const struct file_operations hugetlbfs_file_operations = {
673 	.read			= hugetlbfs_read,
674 	.mmap			= hugetlbfs_file_mmap,
675 	.fsync			= noop_fsync,
676 	.get_unmapped_area	= hugetlb_get_unmapped_area,
677 };
678 
679 static const struct inode_operations hugetlbfs_dir_inode_operations = {
680 	.create		= hugetlbfs_create,
681 	.lookup		= simple_lookup,
682 	.link		= simple_link,
683 	.unlink		= simple_unlink,
684 	.symlink	= hugetlbfs_symlink,
685 	.mkdir		= hugetlbfs_mkdir,
686 	.rmdir		= simple_rmdir,
687 	.mknod		= hugetlbfs_mknod,
688 	.rename		= simple_rename,
689 	.setattr	= hugetlbfs_setattr,
690 };
691 
692 static const struct inode_operations hugetlbfs_inode_operations = {
693 	.setattr	= hugetlbfs_setattr,
694 };
695 
696 static const struct super_operations hugetlbfs_ops = {
697 	.alloc_inode    = hugetlbfs_alloc_inode,
698 	.destroy_inode  = hugetlbfs_destroy_inode,
699 	.evict_inode	= hugetlbfs_evict_inode,
700 	.statfs		= hugetlbfs_statfs,
701 	.put_super	= hugetlbfs_put_super,
702 	.show_options	= generic_show_options,
703 };
704 
705 static int
706 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
707 {
708 	char *p, *rest;
709 	substring_t args[MAX_OPT_ARGS];
710 	int option;
711 	unsigned long long size = 0;
712 	enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
713 
714 	if (!options)
715 		return 0;
716 
717 	while ((p = strsep(&options, ",")) != NULL) {
718 		int token;
719 		if (!*p)
720 			continue;
721 
722 		token = match_token(p, tokens, args);
723 		switch (token) {
724 		case Opt_uid:
725 			if (match_int(&args[0], &option))
726  				goto bad_val;
727 			pconfig->uid = option;
728 			break;
729 
730 		case Opt_gid:
731 			if (match_int(&args[0], &option))
732  				goto bad_val;
733 			pconfig->gid = option;
734 			break;
735 
736 		case Opt_mode:
737 			if (match_octal(&args[0], &option))
738  				goto bad_val;
739 			pconfig->mode = option & 01777U;
740 			break;
741 
742 		case Opt_size: {
743 			/* memparse() will accept a K/M/G without a digit */
744 			if (!isdigit(*args[0].from))
745 				goto bad_val;
746 			size = memparse(args[0].from, &rest);
747 			setsize = SIZE_STD;
748 			if (*rest == '%')
749 				setsize = SIZE_PERCENT;
750 			break;
751 		}
752 
753 		case Opt_nr_inodes:
754 			/* memparse() will accept a K/M/G without a digit */
755 			if (!isdigit(*args[0].from))
756 				goto bad_val;
757 			pconfig->nr_inodes = memparse(args[0].from, &rest);
758 			break;
759 
760 		case Opt_pagesize: {
761 			unsigned long ps;
762 			ps = memparse(args[0].from, &rest);
763 			pconfig->hstate = size_to_hstate(ps);
764 			if (!pconfig->hstate) {
765 				printk(KERN_ERR
766 				"hugetlbfs: Unsupported page size %lu MB\n",
767 					ps >> 20);
768 				return -EINVAL;
769 			}
770 			break;
771 		}
772 
773 		default:
774 			printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
775 				 p);
776 			return -EINVAL;
777 			break;
778 		}
779 	}
780 
781 	/* Do size after hstate is set up */
782 	if (setsize > NO_SIZE) {
783 		struct hstate *h = pconfig->hstate;
784 		if (setsize == SIZE_PERCENT) {
785 			size <<= huge_page_shift(h);
786 			size *= h->max_huge_pages;
787 			do_div(size, 100);
788 		}
789 		pconfig->nr_blocks = (size >> huge_page_shift(h));
790 	}
791 
792 	return 0;
793 
794 bad_val:
795  	printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
796 	       args[0].from, p);
797  	return -EINVAL;
798 }
799 
800 static int
801 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
802 {
803 	struct inode * inode;
804 	struct dentry * root;
805 	int ret;
806 	struct hugetlbfs_config config;
807 	struct hugetlbfs_sb_info *sbinfo;
808 
809 	save_mount_options(sb, data);
810 
811 	config.nr_blocks = -1; /* No limit on size by default */
812 	config.nr_inodes = -1; /* No limit on number of inodes by default */
813 	config.uid = current_fsuid();
814 	config.gid = current_fsgid();
815 	config.mode = 0755;
816 	config.hstate = &default_hstate;
817 	ret = hugetlbfs_parse_options(data, &config);
818 	if (ret)
819 		return ret;
820 
821 	sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
822 	if (!sbinfo)
823 		return -ENOMEM;
824 	sb->s_fs_info = sbinfo;
825 	sbinfo->hstate = config.hstate;
826 	spin_lock_init(&sbinfo->stat_lock);
827 	sbinfo->max_blocks = config.nr_blocks;
828 	sbinfo->free_blocks = config.nr_blocks;
829 	sbinfo->max_inodes = config.nr_inodes;
830 	sbinfo->free_inodes = config.nr_inodes;
831 	sb->s_maxbytes = MAX_LFS_FILESIZE;
832 	sb->s_blocksize = huge_page_size(config.hstate);
833 	sb->s_blocksize_bits = huge_page_shift(config.hstate);
834 	sb->s_magic = HUGETLBFS_MAGIC;
835 	sb->s_op = &hugetlbfs_ops;
836 	sb->s_time_gran = 1;
837 	inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
838 					S_IFDIR | config.mode, 0);
839 	if (!inode)
840 		goto out_free;
841 
842 	root = d_alloc_root(inode);
843 	if (!root) {
844 		iput(inode);
845 		goto out_free;
846 	}
847 	sb->s_root = root;
848 	return 0;
849 out_free:
850 	kfree(sbinfo);
851 	return -ENOMEM;
852 }
853 
854 int hugetlb_get_quota(struct address_space *mapping, long delta)
855 {
856 	int ret = 0;
857 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
858 
859 	if (sbinfo->free_blocks > -1) {
860 		spin_lock(&sbinfo->stat_lock);
861 		if (sbinfo->free_blocks - delta >= 0)
862 			sbinfo->free_blocks -= delta;
863 		else
864 			ret = -ENOMEM;
865 		spin_unlock(&sbinfo->stat_lock);
866 	}
867 
868 	return ret;
869 }
870 
871 void hugetlb_put_quota(struct address_space *mapping, long delta)
872 {
873 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
874 
875 	if (sbinfo->free_blocks > -1) {
876 		spin_lock(&sbinfo->stat_lock);
877 		sbinfo->free_blocks += delta;
878 		spin_unlock(&sbinfo->stat_lock);
879 	}
880 }
881 
882 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
883 	int flags, const char *dev_name, void *data, struct vfsmount *mnt)
884 {
885 	return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
886 }
887 
888 static struct file_system_type hugetlbfs_fs_type = {
889 	.name		= "hugetlbfs",
890 	.get_sb		= hugetlbfs_get_sb,
891 	.kill_sb	= kill_litter_super,
892 };
893 
894 static struct vfsmount *hugetlbfs_vfsmount;
895 
896 static int can_do_hugetlb_shm(void)
897 {
898 	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
899 }
900 
901 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
902 				struct user_struct **user, int creat_flags)
903 {
904 	int error = -ENOMEM;
905 	struct file *file;
906 	struct inode *inode;
907 	struct path path;
908 	struct dentry *root;
909 	struct qstr quick_string;
910 
911 	*user = NULL;
912 	if (!hugetlbfs_vfsmount)
913 		return ERR_PTR(-ENOENT);
914 
915 	if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
916 		*user = current_user();
917 		if (user_shm_lock(size, *user)) {
918 			WARN_ONCE(1,
919 			  "Using mlock ulimits for SHM_HUGETLB deprecated\n");
920 		} else {
921 			*user = NULL;
922 			return ERR_PTR(-EPERM);
923 		}
924 	}
925 
926 	root = hugetlbfs_vfsmount->mnt_root;
927 	quick_string.name = name;
928 	quick_string.len = strlen(quick_string.name);
929 	quick_string.hash = 0;
930 	path.dentry = d_alloc(root, &quick_string);
931 	if (!path.dentry)
932 		goto out_shm_unlock;
933 
934 	path.mnt = mntget(hugetlbfs_vfsmount);
935 	error = -ENOSPC;
936 	inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
937 				current_fsgid(), S_IFREG | S_IRWXUGO, 0);
938 	if (!inode)
939 		goto out_dentry;
940 
941 	error = -ENOMEM;
942 	if (hugetlb_reserve_pages(inode, 0,
943 			size >> huge_page_shift(hstate_inode(inode)), NULL,
944 			acctflag))
945 		goto out_inode;
946 
947 	d_instantiate(path.dentry, inode);
948 	inode->i_size = size;
949 	inode->i_nlink = 0;
950 
951 	error = -ENFILE;
952 	file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
953 			&hugetlbfs_file_operations);
954 	if (!file)
955 		goto out_dentry; /* inode is already attached */
956 
957 	return file;
958 
959 out_inode:
960 	iput(inode);
961 out_dentry:
962 	path_put(&path);
963 out_shm_unlock:
964 	if (*user) {
965 		user_shm_unlock(size, *user);
966 		*user = NULL;
967 	}
968 	return ERR_PTR(error);
969 }
970 
971 static int __init init_hugetlbfs_fs(void)
972 {
973 	int error;
974 	struct vfsmount *vfsmount;
975 
976 	error = bdi_init(&hugetlbfs_backing_dev_info);
977 	if (error)
978 		return error;
979 
980 	hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
981 					sizeof(struct hugetlbfs_inode_info),
982 					0, 0, init_once);
983 	if (hugetlbfs_inode_cachep == NULL)
984 		goto out2;
985 
986 	error = register_filesystem(&hugetlbfs_fs_type);
987 	if (error)
988 		goto out;
989 
990 	vfsmount = kern_mount(&hugetlbfs_fs_type);
991 
992 	if (!IS_ERR(vfsmount)) {
993 		hugetlbfs_vfsmount = vfsmount;
994 		return 0;
995 	}
996 
997 	error = PTR_ERR(vfsmount);
998 
999  out:
1000 	if (error)
1001 		kmem_cache_destroy(hugetlbfs_inode_cachep);
1002  out2:
1003 	bdi_destroy(&hugetlbfs_backing_dev_info);
1004 	return error;
1005 }
1006 
1007 static void __exit exit_hugetlbfs_fs(void)
1008 {
1009 	kmem_cache_destroy(hugetlbfs_inode_cachep);
1010 	unregister_filesystem(&hugetlbfs_fs_type);
1011 	bdi_destroy(&hugetlbfs_backing_dev_info);
1012 }
1013 
1014 module_init(init_hugetlbfs_fs)
1015 module_exit(exit_hugetlbfs_fs)
1016 
1017 MODULE_LICENSE("GPL");
1018