xref: /openbmc/linux/fs/hugetlbfs/inode.c (revision b627b4ed)
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_AND_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_pagesize,
57 	Opt_err,
58 };
59 
60 static const match_table_t tokens = {
61 	{Opt_size,	"size=%s"},
62 	{Opt_nr_inodes,	"nr_inodes=%s"},
63 	{Opt_mode,	"mode=%o"},
64 	{Opt_uid,	"uid=%u"},
65 	{Opt_gid,	"gid=%u"},
66 	{Opt_pagesize,	"pagesize=%s"},
67 	{Opt_err,	NULL},
68 };
69 
70 static void huge_pagevec_release(struct pagevec *pvec)
71 {
72 	int i;
73 
74 	for (i = 0; i < pagevec_count(pvec); ++i)
75 		put_page(pvec->pages[i]);
76 
77 	pagevec_reinit(pvec);
78 }
79 
80 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
81 {
82 	struct inode *inode = file->f_path.dentry->d_inode;
83 	loff_t len, vma_len;
84 	int ret;
85 	struct hstate *h = hstate_file(file);
86 
87 	/*
88 	 * vma address alignment (but not the pgoff alignment) has
89 	 * already been checked by prepare_hugepage_range.  If you add
90 	 * any error returns here, do so after setting VM_HUGETLB, so
91 	 * is_vm_hugetlb_page tests below unmap_region go the right
92 	 * way when do_mmap_pgoff unwinds (may be important on powerpc
93 	 * and ia64).
94 	 */
95 	vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
96 	vma->vm_ops = &hugetlb_vm_ops;
97 
98 	if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
99 		return -EINVAL;
100 
101 	vma_len = (loff_t)(vma->vm_end - vma->vm_start);
102 
103 	mutex_lock(&inode->i_mutex);
104 	file_accessed(file);
105 
106 	ret = -ENOMEM;
107 	len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
108 
109 	if (hugetlb_reserve_pages(inode,
110 				vma->vm_pgoff >> huge_page_order(h),
111 				len >> huge_page_shift(h), vma,
112 				vma->vm_flags))
113 		goto out;
114 
115 	ret = 0;
116 	hugetlb_prefault_arch_hook(vma->vm_mm);
117 	if (vma->vm_flags & VM_WRITE && inode->i_size < len)
118 		inode->i_size = len;
119 out:
120 	mutex_unlock(&inode->i_mutex);
121 
122 	return ret;
123 }
124 
125 /*
126  * Called under down_write(mmap_sem).
127  */
128 
129 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
130 static unsigned long
131 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
132 		unsigned long len, unsigned long pgoff, unsigned long flags)
133 {
134 	struct mm_struct *mm = current->mm;
135 	struct vm_area_struct *vma;
136 	unsigned long start_addr;
137 	struct hstate *h = hstate_file(file);
138 
139 	if (len & ~huge_page_mask(h))
140 		return -EINVAL;
141 	if (len > TASK_SIZE)
142 		return -ENOMEM;
143 
144 	if (flags & MAP_FIXED) {
145 		if (prepare_hugepage_range(file, addr, len))
146 			return -EINVAL;
147 		return addr;
148 	}
149 
150 	if (addr) {
151 		addr = ALIGN(addr, huge_page_size(h));
152 		vma = find_vma(mm, addr);
153 		if (TASK_SIZE - len >= addr &&
154 		    (!vma || addr + len <= vma->vm_start))
155 			return addr;
156 	}
157 
158 	start_addr = mm->free_area_cache;
159 
160 	if (len <= mm->cached_hole_size)
161 		start_addr = TASK_UNMAPPED_BASE;
162 
163 full_search:
164 	addr = ALIGN(start_addr, huge_page_size(h));
165 
166 	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
167 		/* At this point:  (!vma || addr < vma->vm_end). */
168 		if (TASK_SIZE - len < addr) {
169 			/*
170 			 * Start a new search - just in case we missed
171 			 * some holes.
172 			 */
173 			if (start_addr != TASK_UNMAPPED_BASE) {
174 				start_addr = TASK_UNMAPPED_BASE;
175 				goto full_search;
176 			}
177 			return -ENOMEM;
178 		}
179 
180 		if (!vma || addr + len <= vma->vm_start)
181 			return addr;
182 		addr = ALIGN(vma->vm_end, huge_page_size(h));
183 	}
184 }
185 #endif
186 
187 static int
188 hugetlbfs_read_actor(struct page *page, unsigned long offset,
189 			char __user *buf, unsigned long count,
190 			unsigned long size)
191 {
192 	char *kaddr;
193 	unsigned long left, copied = 0;
194 	int i, chunksize;
195 
196 	if (size > count)
197 		size = count;
198 
199 	/* Find which 4k chunk and offset with in that chunk */
200 	i = offset >> PAGE_CACHE_SHIFT;
201 	offset = offset & ~PAGE_CACHE_MASK;
202 
203 	while (size) {
204 		chunksize = PAGE_CACHE_SIZE;
205 		if (offset)
206 			chunksize -= offset;
207 		if (chunksize > size)
208 			chunksize = size;
209 		kaddr = kmap(&page[i]);
210 		left = __copy_to_user(buf, kaddr + offset, chunksize);
211 		kunmap(&page[i]);
212 		if (left) {
213 			copied += (chunksize - left);
214 			break;
215 		}
216 		offset = 0;
217 		size -= chunksize;
218 		buf += chunksize;
219 		copied += chunksize;
220 		i++;
221 	}
222 	return copied ? copied : -EFAULT;
223 }
224 
225 /*
226  * Support for read() - Find the page attached to f_mapping and copy out the
227  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
228  * since it has PAGE_CACHE_SIZE assumptions.
229  */
230 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
231 			      size_t len, loff_t *ppos)
232 {
233 	struct hstate *h = hstate_file(filp);
234 	struct address_space *mapping = filp->f_mapping;
235 	struct inode *inode = mapping->host;
236 	unsigned long index = *ppos >> huge_page_shift(h);
237 	unsigned long offset = *ppos & ~huge_page_mask(h);
238 	unsigned long end_index;
239 	loff_t isize;
240 	ssize_t retval = 0;
241 
242 	mutex_lock(&inode->i_mutex);
243 
244 	/* validate length */
245 	if (len == 0)
246 		goto out;
247 
248 	isize = i_size_read(inode);
249 	if (!isize)
250 		goto out;
251 
252 	end_index = (isize - 1) >> huge_page_shift(h);
253 	for (;;) {
254 		struct page *page;
255 		unsigned long nr, ret;
256 		int ra;
257 
258 		/* nr is the maximum number of bytes to copy from this page */
259 		nr = huge_page_size(h);
260 		if (index >= end_index) {
261 			if (index > end_index)
262 				goto out;
263 			nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
264 			if (nr <= offset) {
265 				goto out;
266 			}
267 		}
268 		nr = nr - offset;
269 
270 		/* Find the page */
271 		page = find_get_page(mapping, index);
272 		if (unlikely(page == NULL)) {
273 			/*
274 			 * We have a HOLE, zero out the user-buffer for the
275 			 * length of the hole or request.
276 			 */
277 			ret = len < nr ? len : nr;
278 			if (clear_user(buf, ret))
279 				ra = -EFAULT;
280 			else
281 				ra = 0;
282 		} else {
283 			/*
284 			 * We have the page, copy it to user space buffer.
285 			 */
286 			ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
287 			ret = ra;
288 		}
289 		if (ra < 0) {
290 			if (retval == 0)
291 				retval = ra;
292 			if (page)
293 				page_cache_release(page);
294 			goto out;
295 		}
296 
297 		offset += ret;
298 		retval += ret;
299 		len -= ret;
300 		index += offset >> huge_page_shift(h);
301 		offset &= ~huge_page_mask(h);
302 
303 		if (page)
304 			page_cache_release(page);
305 
306 		/* short read or no more work */
307 		if ((ret != nr) || (len == 0))
308 			break;
309 	}
310 out:
311 	*ppos = ((loff_t)index << huge_page_shift(h)) + offset;
312 	mutex_unlock(&inode->i_mutex);
313 	return retval;
314 }
315 
316 /*
317  * Read a page. Again trivial. If it didn't already exist
318  * in the page cache, it is zero-filled.
319  */
320 static int hugetlbfs_readpage(struct file *file, struct page * page)
321 {
322 	unlock_page(page);
323 	return -EINVAL;
324 }
325 
326 static int hugetlbfs_write_begin(struct file *file,
327 			struct address_space *mapping,
328 			loff_t pos, unsigned len, unsigned flags,
329 			struct page **pagep, void **fsdata)
330 {
331 	return -EINVAL;
332 }
333 
334 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
335 			loff_t pos, unsigned len, unsigned copied,
336 			struct page *page, void *fsdata)
337 {
338 	BUG();
339 	return -EINVAL;
340 }
341 
342 static void truncate_huge_page(struct page *page)
343 {
344 	cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
345 	ClearPageUptodate(page);
346 	remove_from_page_cache(page);
347 	put_page(page);
348 }
349 
350 static void truncate_hugepages(struct inode *inode, loff_t lstart)
351 {
352 	struct hstate *h = hstate_inode(inode);
353 	struct address_space *mapping = &inode->i_data;
354 	const pgoff_t start = lstart >> huge_page_shift(h);
355 	struct pagevec pvec;
356 	pgoff_t next;
357 	int i, freed = 0;
358 
359 	pagevec_init(&pvec, 0);
360 	next = start;
361 	while (1) {
362 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
363 			if (next == start)
364 				break;
365 			next = start;
366 			continue;
367 		}
368 
369 		for (i = 0; i < pagevec_count(&pvec); ++i) {
370 			struct page *page = pvec.pages[i];
371 
372 			lock_page(page);
373 			if (page->index > next)
374 				next = page->index;
375 			++next;
376 			truncate_huge_page(page);
377 			unlock_page(page);
378 			freed++;
379 		}
380 		huge_pagevec_release(&pvec);
381 	}
382 	BUG_ON(!lstart && mapping->nrpages);
383 	hugetlb_unreserve_pages(inode, start, freed);
384 }
385 
386 static void hugetlbfs_delete_inode(struct inode *inode)
387 {
388 	truncate_hugepages(inode, 0);
389 	clear_inode(inode);
390 }
391 
392 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
393 {
394 	struct super_block *sb = inode->i_sb;
395 
396 	if (!hlist_unhashed(&inode->i_hash)) {
397 		if (!(inode->i_state & (I_DIRTY|I_SYNC)))
398 			list_move(&inode->i_list, &inode_unused);
399 		inodes_stat.nr_unused++;
400 		if (!sb || (sb->s_flags & MS_ACTIVE)) {
401 			spin_unlock(&inode_lock);
402 			return;
403 		}
404 		inode->i_state |= I_WILL_FREE;
405 		spin_unlock(&inode_lock);
406 		/*
407 		 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
408 		 * in our backing_dev_info.
409 		 */
410 		write_inode_now(inode, 1);
411 		spin_lock(&inode_lock);
412 		inode->i_state &= ~I_WILL_FREE;
413 		inodes_stat.nr_unused--;
414 		hlist_del_init(&inode->i_hash);
415 	}
416 	list_del_init(&inode->i_list);
417 	list_del_init(&inode->i_sb_list);
418 	inode->i_state |= I_FREEING;
419 	inodes_stat.nr_inodes--;
420 	spin_unlock(&inode_lock);
421 	truncate_hugepages(inode, 0);
422 	clear_inode(inode);
423 	destroy_inode(inode);
424 }
425 
426 static void hugetlbfs_drop_inode(struct inode *inode)
427 {
428 	if (!inode->i_nlink)
429 		generic_delete_inode(inode);
430 	else
431 		hugetlbfs_forget_inode(inode);
432 }
433 
434 static inline void
435 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
436 {
437 	struct vm_area_struct *vma;
438 	struct prio_tree_iter iter;
439 
440 	vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
441 		unsigned long v_offset;
442 
443 		/*
444 		 * Can the expression below overflow on 32-bit arches?
445 		 * No, because the prio_tree returns us only those vmas
446 		 * which overlap the truncated area starting at pgoff,
447 		 * and no vma on a 32-bit arch can span beyond the 4GB.
448 		 */
449 		if (vma->vm_pgoff < pgoff)
450 			v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
451 		else
452 			v_offset = 0;
453 
454 		__unmap_hugepage_range(vma,
455 				vma->vm_start + v_offset, vma->vm_end, NULL);
456 	}
457 }
458 
459 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
460 {
461 	pgoff_t pgoff;
462 	struct address_space *mapping = inode->i_mapping;
463 	struct hstate *h = hstate_inode(inode);
464 
465 	BUG_ON(offset & ~huge_page_mask(h));
466 	pgoff = offset >> PAGE_SHIFT;
467 
468 	i_size_write(inode, offset);
469 	spin_lock(&mapping->i_mmap_lock);
470 	if (!prio_tree_empty(&mapping->i_mmap))
471 		hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
472 	spin_unlock(&mapping->i_mmap_lock);
473 	truncate_hugepages(inode, offset);
474 	return 0;
475 }
476 
477 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
478 {
479 	struct inode *inode = dentry->d_inode;
480 	struct hstate *h = hstate_inode(inode);
481 	int error;
482 	unsigned int ia_valid = attr->ia_valid;
483 
484 	BUG_ON(!inode);
485 
486 	error = inode_change_ok(inode, attr);
487 	if (error)
488 		goto out;
489 
490 	if (ia_valid & ATTR_SIZE) {
491 		error = -EINVAL;
492 		if (!(attr->ia_size & ~huge_page_mask(h)))
493 			error = hugetlb_vmtruncate(inode, attr->ia_size);
494 		if (error)
495 			goto out;
496 		attr->ia_valid &= ~ATTR_SIZE;
497 	}
498 	error = inode_setattr(inode, attr);
499 out:
500 	return error;
501 }
502 
503 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
504 					gid_t gid, int mode, dev_t dev)
505 {
506 	struct inode *inode;
507 
508 	inode = new_inode(sb);
509 	if (inode) {
510 		struct hugetlbfs_inode_info *info;
511 		inode->i_mode = mode;
512 		inode->i_uid = uid;
513 		inode->i_gid = gid;
514 		inode->i_mapping->a_ops = &hugetlbfs_aops;
515 		inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
516 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
517 		INIT_LIST_HEAD(&inode->i_mapping->private_list);
518 		info = HUGETLBFS_I(inode);
519 		mpol_shared_policy_init(&info->policy, NULL);
520 		switch (mode & S_IFMT) {
521 		default:
522 			init_special_inode(inode, mode, dev);
523 			break;
524 		case S_IFREG:
525 			inode->i_op = &hugetlbfs_inode_operations;
526 			inode->i_fop = &hugetlbfs_file_operations;
527 			break;
528 		case S_IFDIR:
529 			inode->i_op = &hugetlbfs_dir_inode_operations;
530 			inode->i_fop = &simple_dir_operations;
531 
532 			/* directory inodes start off with i_nlink == 2 (for "." entry) */
533 			inc_nlink(inode);
534 			break;
535 		case S_IFLNK:
536 			inode->i_op = &page_symlink_inode_operations;
537 			break;
538 		}
539 	}
540 	return inode;
541 }
542 
543 /*
544  * File creation. Allocate an inode, and we're done..
545  */
546 static int hugetlbfs_mknod(struct inode *dir,
547 			struct dentry *dentry, int mode, dev_t dev)
548 {
549 	struct inode *inode;
550 	int error = -ENOSPC;
551 	gid_t gid;
552 
553 	if (dir->i_mode & S_ISGID) {
554 		gid = dir->i_gid;
555 		if (S_ISDIR(mode))
556 			mode |= S_ISGID;
557 	} else {
558 		gid = current_fsgid();
559 	}
560 	inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
561 	if (inode) {
562 		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
563 		d_instantiate(dentry, inode);
564 		dget(dentry);	/* Extra count - pin the dentry in core */
565 		error = 0;
566 	}
567 	return error;
568 }
569 
570 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
571 {
572 	int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
573 	if (!retval)
574 		inc_nlink(dir);
575 	return retval;
576 }
577 
578 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
579 {
580 	return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
581 }
582 
583 static int hugetlbfs_symlink(struct inode *dir,
584 			struct dentry *dentry, const char *symname)
585 {
586 	struct inode *inode;
587 	int error = -ENOSPC;
588 	gid_t gid;
589 
590 	if (dir->i_mode & S_ISGID)
591 		gid = dir->i_gid;
592 	else
593 		gid = current_fsgid();
594 
595 	inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
596 					gid, S_IFLNK|S_IRWXUGO, 0);
597 	if (inode) {
598 		int l = strlen(symname)+1;
599 		error = page_symlink(inode, symname, l);
600 		if (!error) {
601 			d_instantiate(dentry, inode);
602 			dget(dentry);
603 		} else
604 			iput(inode);
605 	}
606 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
607 
608 	return error;
609 }
610 
611 /*
612  * mark the head page dirty
613  */
614 static int hugetlbfs_set_page_dirty(struct page *page)
615 {
616 	struct page *head = compound_head(page);
617 
618 	SetPageDirty(head);
619 	return 0;
620 }
621 
622 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
623 {
624 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
625 	struct hstate *h = hstate_inode(dentry->d_inode);
626 
627 	buf->f_type = HUGETLBFS_MAGIC;
628 	buf->f_bsize = huge_page_size(h);
629 	if (sbinfo) {
630 		spin_lock(&sbinfo->stat_lock);
631 		/* If no limits set, just report 0 for max/free/used
632 		 * blocks, like simple_statfs() */
633 		if (sbinfo->max_blocks >= 0) {
634 			buf->f_blocks = sbinfo->max_blocks;
635 			buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
636 			buf->f_files = sbinfo->max_inodes;
637 			buf->f_ffree = sbinfo->free_inodes;
638 		}
639 		spin_unlock(&sbinfo->stat_lock);
640 	}
641 	buf->f_namelen = NAME_MAX;
642 	return 0;
643 }
644 
645 static void hugetlbfs_put_super(struct super_block *sb)
646 {
647 	struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
648 
649 	if (sbi) {
650 		sb->s_fs_info = NULL;
651 		kfree(sbi);
652 	}
653 }
654 
655 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
656 {
657 	if (sbinfo->free_inodes >= 0) {
658 		spin_lock(&sbinfo->stat_lock);
659 		if (unlikely(!sbinfo->free_inodes)) {
660 			spin_unlock(&sbinfo->stat_lock);
661 			return 0;
662 		}
663 		sbinfo->free_inodes--;
664 		spin_unlock(&sbinfo->stat_lock);
665 	}
666 
667 	return 1;
668 }
669 
670 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
671 {
672 	if (sbinfo->free_inodes >= 0) {
673 		spin_lock(&sbinfo->stat_lock);
674 		sbinfo->free_inodes++;
675 		spin_unlock(&sbinfo->stat_lock);
676 	}
677 }
678 
679 
680 static struct kmem_cache *hugetlbfs_inode_cachep;
681 
682 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
683 {
684 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
685 	struct hugetlbfs_inode_info *p;
686 
687 	if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
688 		return NULL;
689 	p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
690 	if (unlikely(!p)) {
691 		hugetlbfs_inc_free_inodes(sbinfo);
692 		return NULL;
693 	}
694 	return &p->vfs_inode;
695 }
696 
697 static void hugetlbfs_destroy_inode(struct inode *inode)
698 {
699 	hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
700 	mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
701 	kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
702 }
703 
704 static const struct address_space_operations hugetlbfs_aops = {
705 	.readpage	= hugetlbfs_readpage,
706 	.write_begin	= hugetlbfs_write_begin,
707 	.write_end	= hugetlbfs_write_end,
708 	.set_page_dirty	= hugetlbfs_set_page_dirty,
709 };
710 
711 
712 static void init_once(void *foo)
713 {
714 	struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
715 
716 	inode_init_once(&ei->vfs_inode);
717 }
718 
719 const struct file_operations hugetlbfs_file_operations = {
720 	.read			= hugetlbfs_read,
721 	.mmap			= hugetlbfs_file_mmap,
722 	.fsync			= simple_sync_file,
723 	.get_unmapped_area	= hugetlb_get_unmapped_area,
724 };
725 
726 static const struct inode_operations hugetlbfs_dir_inode_operations = {
727 	.create		= hugetlbfs_create,
728 	.lookup		= simple_lookup,
729 	.link		= simple_link,
730 	.unlink		= simple_unlink,
731 	.symlink	= hugetlbfs_symlink,
732 	.mkdir		= hugetlbfs_mkdir,
733 	.rmdir		= simple_rmdir,
734 	.mknod		= hugetlbfs_mknod,
735 	.rename		= simple_rename,
736 	.setattr	= hugetlbfs_setattr,
737 };
738 
739 static const struct inode_operations hugetlbfs_inode_operations = {
740 	.setattr	= hugetlbfs_setattr,
741 };
742 
743 static const struct super_operations hugetlbfs_ops = {
744 	.alloc_inode    = hugetlbfs_alloc_inode,
745 	.destroy_inode  = hugetlbfs_destroy_inode,
746 	.statfs		= hugetlbfs_statfs,
747 	.delete_inode	= hugetlbfs_delete_inode,
748 	.drop_inode	= hugetlbfs_drop_inode,
749 	.put_super	= hugetlbfs_put_super,
750 	.show_options	= generic_show_options,
751 };
752 
753 static int
754 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
755 {
756 	char *p, *rest;
757 	substring_t args[MAX_OPT_ARGS];
758 	int option;
759 	unsigned long long size = 0;
760 	enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
761 
762 	if (!options)
763 		return 0;
764 
765 	while ((p = strsep(&options, ",")) != NULL) {
766 		int token;
767 		if (!*p)
768 			continue;
769 
770 		token = match_token(p, tokens, args);
771 		switch (token) {
772 		case Opt_uid:
773 			if (match_int(&args[0], &option))
774  				goto bad_val;
775 			pconfig->uid = option;
776 			break;
777 
778 		case Opt_gid:
779 			if (match_int(&args[0], &option))
780  				goto bad_val;
781 			pconfig->gid = option;
782 			break;
783 
784 		case Opt_mode:
785 			if (match_octal(&args[0], &option))
786  				goto bad_val;
787 			pconfig->mode = option & 01777U;
788 			break;
789 
790 		case Opt_size: {
791 			/* memparse() will accept a K/M/G without a digit */
792 			if (!isdigit(*args[0].from))
793 				goto bad_val;
794 			size = memparse(args[0].from, &rest);
795 			setsize = SIZE_STD;
796 			if (*rest == '%')
797 				setsize = SIZE_PERCENT;
798 			break;
799 		}
800 
801 		case Opt_nr_inodes:
802 			/* memparse() will accept a K/M/G without a digit */
803 			if (!isdigit(*args[0].from))
804 				goto bad_val;
805 			pconfig->nr_inodes = memparse(args[0].from, &rest);
806 			break;
807 
808 		case Opt_pagesize: {
809 			unsigned long ps;
810 			ps = memparse(args[0].from, &rest);
811 			pconfig->hstate = size_to_hstate(ps);
812 			if (!pconfig->hstate) {
813 				printk(KERN_ERR
814 				"hugetlbfs: Unsupported page size %lu MB\n",
815 					ps >> 20);
816 				return -EINVAL;
817 			}
818 			break;
819 		}
820 
821 		default:
822 			printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
823 				 p);
824 			return -EINVAL;
825 			break;
826 		}
827 	}
828 
829 	/* Do size after hstate is set up */
830 	if (setsize > NO_SIZE) {
831 		struct hstate *h = pconfig->hstate;
832 		if (setsize == SIZE_PERCENT) {
833 			size <<= huge_page_shift(h);
834 			size *= h->max_huge_pages;
835 			do_div(size, 100);
836 		}
837 		pconfig->nr_blocks = (size >> huge_page_shift(h));
838 	}
839 
840 	return 0;
841 
842 bad_val:
843  	printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
844 	       args[0].from, p);
845  	return 1;
846 }
847 
848 static int
849 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
850 {
851 	struct inode * inode;
852 	struct dentry * root;
853 	int ret;
854 	struct hugetlbfs_config config;
855 	struct hugetlbfs_sb_info *sbinfo;
856 
857 	save_mount_options(sb, data);
858 
859 	config.nr_blocks = -1; /* No limit on size by default */
860 	config.nr_inodes = -1; /* No limit on number of inodes by default */
861 	config.uid = current_fsuid();
862 	config.gid = current_fsgid();
863 	config.mode = 0755;
864 	config.hstate = &default_hstate;
865 	ret = hugetlbfs_parse_options(data, &config);
866 	if (ret)
867 		return ret;
868 
869 	sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
870 	if (!sbinfo)
871 		return -ENOMEM;
872 	sb->s_fs_info = sbinfo;
873 	sbinfo->hstate = config.hstate;
874 	spin_lock_init(&sbinfo->stat_lock);
875 	sbinfo->max_blocks = config.nr_blocks;
876 	sbinfo->free_blocks = config.nr_blocks;
877 	sbinfo->max_inodes = config.nr_inodes;
878 	sbinfo->free_inodes = config.nr_inodes;
879 	sb->s_maxbytes = MAX_LFS_FILESIZE;
880 	sb->s_blocksize = huge_page_size(config.hstate);
881 	sb->s_blocksize_bits = huge_page_shift(config.hstate);
882 	sb->s_magic = HUGETLBFS_MAGIC;
883 	sb->s_op = &hugetlbfs_ops;
884 	sb->s_time_gran = 1;
885 	inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
886 					S_IFDIR | config.mode, 0);
887 	if (!inode)
888 		goto out_free;
889 
890 	root = d_alloc_root(inode);
891 	if (!root) {
892 		iput(inode);
893 		goto out_free;
894 	}
895 	sb->s_root = root;
896 	return 0;
897 out_free:
898 	kfree(sbinfo);
899 	return -ENOMEM;
900 }
901 
902 int hugetlb_get_quota(struct address_space *mapping, long delta)
903 {
904 	int ret = 0;
905 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
906 
907 	if (sbinfo->free_blocks > -1) {
908 		spin_lock(&sbinfo->stat_lock);
909 		if (sbinfo->free_blocks - delta >= 0)
910 			sbinfo->free_blocks -= delta;
911 		else
912 			ret = -ENOMEM;
913 		spin_unlock(&sbinfo->stat_lock);
914 	}
915 
916 	return ret;
917 }
918 
919 void hugetlb_put_quota(struct address_space *mapping, long delta)
920 {
921 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
922 
923 	if (sbinfo->free_blocks > -1) {
924 		spin_lock(&sbinfo->stat_lock);
925 		sbinfo->free_blocks += delta;
926 		spin_unlock(&sbinfo->stat_lock);
927 	}
928 }
929 
930 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
931 	int flags, const char *dev_name, void *data, struct vfsmount *mnt)
932 {
933 	return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
934 }
935 
936 static struct file_system_type hugetlbfs_fs_type = {
937 	.name		= "hugetlbfs",
938 	.get_sb		= hugetlbfs_get_sb,
939 	.kill_sb	= kill_litter_super,
940 };
941 
942 static struct vfsmount *hugetlbfs_vfsmount;
943 
944 static int can_do_hugetlb_shm(void)
945 {
946 	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
947 }
948 
949 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)
950 {
951 	int error = -ENOMEM;
952 	int unlock_shm = 0;
953 	struct file *file;
954 	struct inode *inode;
955 	struct dentry *dentry, *root;
956 	struct qstr quick_string;
957 	struct user_struct *user = current_user();
958 
959 	if (!hugetlbfs_vfsmount)
960 		return ERR_PTR(-ENOENT);
961 
962 	if (!can_do_hugetlb_shm()) {
963 		if (user_shm_lock(size, user)) {
964 			unlock_shm = 1;
965 			WARN_ONCE(1,
966 			  "Using mlock ulimits for SHM_HUGETLB deprecated\n");
967 		} else
968 			return ERR_PTR(-EPERM);
969 	}
970 
971 	root = hugetlbfs_vfsmount->mnt_root;
972 	quick_string.name = name;
973 	quick_string.len = strlen(quick_string.name);
974 	quick_string.hash = 0;
975 	dentry = d_alloc(root, &quick_string);
976 	if (!dentry)
977 		goto out_shm_unlock;
978 
979 	error = -ENOSPC;
980 	inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
981 				current_fsgid(), S_IFREG | S_IRWXUGO, 0);
982 	if (!inode)
983 		goto out_dentry;
984 
985 	error = -ENOMEM;
986 	if (hugetlb_reserve_pages(inode, 0,
987 			size >> huge_page_shift(hstate_inode(inode)), NULL,
988 			acctflag))
989 		goto out_inode;
990 
991 	d_instantiate(dentry, inode);
992 	inode->i_size = size;
993 	inode->i_nlink = 0;
994 
995 	error = -ENFILE;
996 	file = alloc_file(hugetlbfs_vfsmount, dentry,
997 			FMODE_WRITE | FMODE_READ,
998 			&hugetlbfs_file_operations);
999 	if (!file)
1000 		goto out_dentry; /* inode is already attached */
1001 
1002 	return file;
1003 
1004 out_inode:
1005 	iput(inode);
1006 out_dentry:
1007 	dput(dentry);
1008 out_shm_unlock:
1009 	if (unlock_shm)
1010 		user_shm_unlock(size, user);
1011 	return ERR_PTR(error);
1012 }
1013 
1014 static int __init init_hugetlbfs_fs(void)
1015 {
1016 	int error;
1017 	struct vfsmount *vfsmount;
1018 
1019 	error = bdi_init(&hugetlbfs_backing_dev_info);
1020 	if (error)
1021 		return error;
1022 
1023 	hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1024 					sizeof(struct hugetlbfs_inode_info),
1025 					0, 0, init_once);
1026 	if (hugetlbfs_inode_cachep == NULL)
1027 		goto out2;
1028 
1029 	error = register_filesystem(&hugetlbfs_fs_type);
1030 	if (error)
1031 		goto out;
1032 
1033 	vfsmount = kern_mount(&hugetlbfs_fs_type);
1034 
1035 	if (!IS_ERR(vfsmount)) {
1036 		hugetlbfs_vfsmount = vfsmount;
1037 		return 0;
1038 	}
1039 
1040 	error = PTR_ERR(vfsmount);
1041 
1042  out:
1043 	if (error)
1044 		kmem_cache_destroy(hugetlbfs_inode_cachep);
1045  out2:
1046 	bdi_destroy(&hugetlbfs_backing_dev_info);
1047 	return error;
1048 }
1049 
1050 static void __exit exit_hugetlbfs_fs(void)
1051 {
1052 	kmem_cache_destroy(hugetlbfs_inode_cachep);
1053 	unregister_filesystem(&hugetlbfs_fs_type);
1054 	bdi_destroy(&hugetlbfs_backing_dev_info);
1055 }
1056 
1057 module_init(init_hugetlbfs_fs)
1058 module_exit(exit_hugetlbfs_fs)
1059 
1060 MODULE_LICENSE("GPL");
1061