xref: /linux/fs/hugetlbfs/inode.c (revision 25985edcedea6396277003854657b5f3cb31a628)
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 #include <linux/migrate.h>
35 
36 #include <asm/uaccess.h>
37 
38 static const struct super_operations hugetlbfs_ops;
39 static const struct address_space_operations hugetlbfs_aops;
40 const struct file_operations hugetlbfs_file_operations;
41 static const struct inode_operations hugetlbfs_dir_inode_operations;
42 static const struct inode_operations hugetlbfs_inode_operations;
43 
44 static struct backing_dev_info hugetlbfs_backing_dev_info = {
45 	.name		= "hugetlbfs",
46 	.ra_pages	= 0,	/* No readahead */
47 	.capabilities	= BDI_CAP_NO_ACCT_AND_WRITEBACK,
48 };
49 
50 int sysctl_hugetlb_shm_group;
51 
52 enum {
53 	Opt_size, Opt_nr_inodes,
54 	Opt_mode, Opt_uid, Opt_gid,
55 	Opt_pagesize,
56 	Opt_err,
57 };
58 
59 static const 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_pagesize,	"pagesize=%s"},
66 	{Opt_err,	NULL},
67 };
68 
69 static void huge_pagevec_release(struct pagevec *pvec)
70 {
71 	int i;
72 
73 	for (i = 0; i < pagevec_count(pvec); ++i)
74 		put_page(pvec->pages[i]);
75 
76 	pagevec_reinit(pvec);
77 }
78 
79 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
80 {
81 	struct inode *inode = file->f_path.dentry->d_inode;
82 	loff_t len, vma_len;
83 	int ret;
84 	struct hstate *h = hstate_file(file);
85 
86 	/*
87 	 * vma address alignment (but not the pgoff alignment) has
88 	 * already been checked by prepare_hugepage_range.  If you add
89 	 * any error returns here, do so after setting VM_HUGETLB, so
90 	 * is_vm_hugetlb_page tests below unmap_region go the right
91 	 * way when do_mmap_pgoff unwinds (may be important on powerpc
92 	 * and ia64).
93 	 */
94 	vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
95 	vma->vm_ops = &hugetlb_vm_ops;
96 
97 	if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
98 		return -EINVAL;
99 
100 	vma_len = (loff_t)(vma->vm_end - vma->vm_start);
101 
102 	mutex_lock(&inode->i_mutex);
103 	file_accessed(file);
104 
105 	ret = -ENOMEM;
106 	len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
107 
108 	if (hugetlb_reserve_pages(inode,
109 				vma->vm_pgoff >> huge_page_order(h),
110 				len >> huge_page_shift(h), vma,
111 				vma->vm_flags))
112 		goto out;
113 
114 	ret = 0;
115 	hugetlb_prefault_arch_hook(vma->vm_mm);
116 	if (vma->vm_flags & VM_WRITE && inode->i_size < len)
117 		inode->i_size = len;
118 out:
119 	mutex_unlock(&inode->i_mutex);
120 
121 	return ret;
122 }
123 
124 /*
125  * Called under down_write(mmap_sem).
126  */
127 
128 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
129 static unsigned long
130 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
131 		unsigned long len, unsigned long pgoff, unsigned long flags)
132 {
133 	struct mm_struct *mm = current->mm;
134 	struct vm_area_struct *vma;
135 	unsigned long start_addr;
136 	struct hstate *h = hstate_file(file);
137 
138 	if (len & ~huge_page_mask(h))
139 		return -EINVAL;
140 	if (len > TASK_SIZE)
141 		return -ENOMEM;
142 
143 	if (flags & MAP_FIXED) {
144 		if (prepare_hugepage_range(file, addr, len))
145 			return -EINVAL;
146 		return addr;
147 	}
148 
149 	if (addr) {
150 		addr = ALIGN(addr, huge_page_size(h));
151 		vma = find_vma(mm, addr);
152 		if (TASK_SIZE - len >= addr &&
153 		    (!vma || addr + len <= vma->vm_start))
154 			return addr;
155 	}
156 
157 	start_addr = mm->free_area_cache;
158 
159 	if (len <= mm->cached_hole_size)
160 		start_addr = TASK_UNMAPPED_BASE;
161 
162 full_search:
163 	addr = ALIGN(start_addr, huge_page_size(h));
164 
165 	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
166 		/* At this point:  (!vma || addr < vma->vm_end). */
167 		if (TASK_SIZE - len < addr) {
168 			/*
169 			 * Start a new search - just in case we missed
170 			 * some holes.
171 			 */
172 			if (start_addr != TASK_UNMAPPED_BASE) {
173 				start_addr = TASK_UNMAPPED_BASE;
174 				goto full_search;
175 			}
176 			return -ENOMEM;
177 		}
178 
179 		if (!vma || addr + len <= vma->vm_start)
180 			return addr;
181 		addr = ALIGN(vma->vm_end, huge_page_size(h));
182 	}
183 }
184 #endif
185 
186 static int
187 hugetlbfs_read_actor(struct page *page, unsigned long offset,
188 			char __user *buf, unsigned long count,
189 			unsigned long size)
190 {
191 	char *kaddr;
192 	unsigned long left, copied = 0;
193 	int i, chunksize;
194 
195 	if (size > count)
196 		size = count;
197 
198 	/* Find which 4k chunk and offset with in that chunk */
199 	i = offset >> PAGE_CACHE_SHIFT;
200 	offset = offset & ~PAGE_CACHE_MASK;
201 
202 	while (size) {
203 		chunksize = PAGE_CACHE_SIZE;
204 		if (offset)
205 			chunksize -= offset;
206 		if (chunksize > size)
207 			chunksize = size;
208 		kaddr = kmap(&page[i]);
209 		left = __copy_to_user(buf, kaddr + offset, chunksize);
210 		kunmap(&page[i]);
211 		if (left) {
212 			copied += (chunksize - left);
213 			break;
214 		}
215 		offset = 0;
216 		size -= chunksize;
217 		buf += chunksize;
218 		copied += chunksize;
219 		i++;
220 	}
221 	return copied ? copied : -EFAULT;
222 }
223 
224 /*
225  * Support for read() - Find the page attached to f_mapping and copy out the
226  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
227  * since it has PAGE_CACHE_SIZE assumptions.
228  */
229 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
230 			      size_t len, loff_t *ppos)
231 {
232 	struct hstate *h = hstate_file(filp);
233 	struct address_space *mapping = filp->f_mapping;
234 	struct inode *inode = mapping->host;
235 	unsigned long index = *ppos >> huge_page_shift(h);
236 	unsigned long offset = *ppos & ~huge_page_mask(h);
237 	unsigned long end_index;
238 	loff_t isize;
239 	ssize_t retval = 0;
240 
241 	mutex_lock(&inode->i_mutex);
242 
243 	/* validate length */
244 	if (len == 0)
245 		goto out;
246 
247 	isize = i_size_read(inode);
248 	if (!isize)
249 		goto out;
250 
251 	end_index = (isize - 1) >> huge_page_shift(h);
252 	for (;;) {
253 		struct page *page;
254 		unsigned long nr, ret;
255 		int ra;
256 
257 		/* nr is the maximum number of bytes to copy from this page */
258 		nr = huge_page_size(h);
259 		if (index >= end_index) {
260 			if (index > end_index)
261 				goto out;
262 			nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
263 			if (nr <= offset) {
264 				goto out;
265 			}
266 		}
267 		nr = nr - offset;
268 
269 		/* Find the page */
270 		page = find_get_page(mapping, index);
271 		if (unlikely(page == NULL)) {
272 			/*
273 			 * We have a HOLE, zero out the user-buffer for the
274 			 * length of the hole or request.
275 			 */
276 			ret = len < nr ? len : nr;
277 			if (clear_user(buf, ret))
278 				ra = -EFAULT;
279 			else
280 				ra = 0;
281 		} else {
282 			/*
283 			 * We have the page, copy it to user space buffer.
284 			 */
285 			ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
286 			ret = ra;
287 		}
288 		if (ra < 0) {
289 			if (retval == 0)
290 				retval = ra;
291 			if (page)
292 				page_cache_release(page);
293 			goto out;
294 		}
295 
296 		offset += ret;
297 		retval += ret;
298 		len -= ret;
299 		index += offset >> huge_page_shift(h);
300 		offset &= ~huge_page_mask(h);
301 
302 		if (page)
303 			page_cache_release(page);
304 
305 		/* short read or no more work */
306 		if ((ret != nr) || (len == 0))
307 			break;
308 	}
309 out:
310 	*ppos = ((loff_t)index << huge_page_shift(h)) + offset;
311 	mutex_unlock(&inode->i_mutex);
312 	return retval;
313 }
314 
315 static int hugetlbfs_write_begin(struct file *file,
316 			struct address_space *mapping,
317 			loff_t pos, unsigned len, unsigned flags,
318 			struct page **pagep, void **fsdata)
319 {
320 	return -EINVAL;
321 }
322 
323 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
324 			loff_t pos, unsigned len, unsigned copied,
325 			struct page *page, void *fsdata)
326 {
327 	BUG();
328 	return -EINVAL;
329 }
330 
331 static void truncate_huge_page(struct page *page)
332 {
333 	cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
334 	ClearPageUptodate(page);
335 	delete_from_page_cache(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_ino = get_next_ino();
459 		inode->i_mode = mode;
460 		inode->i_uid = uid;
461 		inode->i_gid = gid;
462 		inode->i_mapping->a_ops = &hugetlbfs_aops;
463 		inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
464 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
465 		INIT_LIST_HEAD(&inode->i_mapping->private_list);
466 		info = HUGETLBFS_I(inode);
467 		/*
468 		 * The policy is initialized here even if we are creating a
469 		 * private inode because initialization simply creates an
470 		 * an empty rb tree and calls spin_lock_init(), later when we
471 		 * call mpol_free_shared_policy() it will just return because
472 		 * the rb tree will still be empty.
473 		 */
474 		mpol_shared_policy_init(&info->policy, NULL);
475 		switch (mode & S_IFMT) {
476 		default:
477 			init_special_inode(inode, mode, dev);
478 			break;
479 		case S_IFREG:
480 			inode->i_op = &hugetlbfs_inode_operations;
481 			inode->i_fop = &hugetlbfs_file_operations;
482 			break;
483 		case S_IFDIR:
484 			inode->i_op = &hugetlbfs_dir_inode_operations;
485 			inode->i_fop = &simple_dir_operations;
486 
487 			/* directory inodes start off with i_nlink == 2 (for "." entry) */
488 			inc_nlink(inode);
489 			break;
490 		case S_IFLNK:
491 			inode->i_op = &page_symlink_inode_operations;
492 			break;
493 		}
494 	}
495 	return inode;
496 }
497 
498 /*
499  * File creation. Allocate an inode, and we're done..
500  */
501 static int hugetlbfs_mknod(struct inode *dir,
502 			struct dentry *dentry, int mode, dev_t dev)
503 {
504 	struct inode *inode;
505 	int error = -ENOSPC;
506 	gid_t gid;
507 
508 	if (dir->i_mode & S_ISGID) {
509 		gid = dir->i_gid;
510 		if (S_ISDIR(mode))
511 			mode |= S_ISGID;
512 	} else {
513 		gid = current_fsgid();
514 	}
515 	inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
516 	if (inode) {
517 		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
518 		d_instantiate(dentry, inode);
519 		dget(dentry);	/* Extra count - pin the dentry in core */
520 		error = 0;
521 	}
522 	return error;
523 }
524 
525 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
526 {
527 	int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
528 	if (!retval)
529 		inc_nlink(dir);
530 	return retval;
531 }
532 
533 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
534 {
535 	return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
536 }
537 
538 static int hugetlbfs_symlink(struct inode *dir,
539 			struct dentry *dentry, const char *symname)
540 {
541 	struct inode *inode;
542 	int error = -ENOSPC;
543 	gid_t gid;
544 
545 	if (dir->i_mode & S_ISGID)
546 		gid = dir->i_gid;
547 	else
548 		gid = current_fsgid();
549 
550 	inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
551 					gid, S_IFLNK|S_IRWXUGO, 0);
552 	if (inode) {
553 		int l = strlen(symname)+1;
554 		error = page_symlink(inode, symname, l);
555 		if (!error) {
556 			d_instantiate(dentry, inode);
557 			dget(dentry);
558 		} else
559 			iput(inode);
560 	}
561 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
562 
563 	return error;
564 }
565 
566 /*
567  * mark the head page dirty
568  */
569 static int hugetlbfs_set_page_dirty(struct page *page)
570 {
571 	struct page *head = compound_head(page);
572 
573 	SetPageDirty(head);
574 	return 0;
575 }
576 
577 static int hugetlbfs_migrate_page(struct address_space *mapping,
578 				struct page *newpage, struct page *page)
579 {
580 	int rc;
581 
582 	rc = migrate_huge_page_move_mapping(mapping, newpage, page);
583 	if (rc)
584 		return rc;
585 	migrate_page_copy(newpage, page);
586 
587 	return 0;
588 }
589 
590 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
591 {
592 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
593 	struct hstate *h = hstate_inode(dentry->d_inode);
594 
595 	buf->f_type = HUGETLBFS_MAGIC;
596 	buf->f_bsize = huge_page_size(h);
597 	if (sbinfo) {
598 		spin_lock(&sbinfo->stat_lock);
599 		/* If no limits set, just report 0 for max/free/used
600 		 * blocks, like simple_statfs() */
601 		if (sbinfo->max_blocks >= 0) {
602 			buf->f_blocks = sbinfo->max_blocks;
603 			buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
604 			buf->f_files = sbinfo->max_inodes;
605 			buf->f_ffree = sbinfo->free_inodes;
606 		}
607 		spin_unlock(&sbinfo->stat_lock);
608 	}
609 	buf->f_namelen = NAME_MAX;
610 	return 0;
611 }
612 
613 static void hugetlbfs_put_super(struct super_block *sb)
614 {
615 	struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
616 
617 	if (sbi) {
618 		sb->s_fs_info = NULL;
619 		kfree(sbi);
620 	}
621 }
622 
623 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
624 {
625 	if (sbinfo->free_inodes >= 0) {
626 		spin_lock(&sbinfo->stat_lock);
627 		if (unlikely(!sbinfo->free_inodes)) {
628 			spin_unlock(&sbinfo->stat_lock);
629 			return 0;
630 		}
631 		sbinfo->free_inodes--;
632 		spin_unlock(&sbinfo->stat_lock);
633 	}
634 
635 	return 1;
636 }
637 
638 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
639 {
640 	if (sbinfo->free_inodes >= 0) {
641 		spin_lock(&sbinfo->stat_lock);
642 		sbinfo->free_inodes++;
643 		spin_unlock(&sbinfo->stat_lock);
644 	}
645 }
646 
647 
648 static struct kmem_cache *hugetlbfs_inode_cachep;
649 
650 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
651 {
652 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
653 	struct hugetlbfs_inode_info *p;
654 
655 	if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
656 		return NULL;
657 	p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
658 	if (unlikely(!p)) {
659 		hugetlbfs_inc_free_inodes(sbinfo);
660 		return NULL;
661 	}
662 	return &p->vfs_inode;
663 }
664 
665 static void hugetlbfs_i_callback(struct rcu_head *head)
666 {
667 	struct inode *inode = container_of(head, struct inode, i_rcu);
668 	INIT_LIST_HEAD(&inode->i_dentry);
669 	kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
670 }
671 
672 static void hugetlbfs_destroy_inode(struct inode *inode)
673 {
674 	hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
675 	mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
676 	call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
677 }
678 
679 static const struct address_space_operations hugetlbfs_aops = {
680 	.write_begin	= hugetlbfs_write_begin,
681 	.write_end	= hugetlbfs_write_end,
682 	.set_page_dirty	= hugetlbfs_set_page_dirty,
683 	.migratepage    = hugetlbfs_migrate_page,
684 };
685 
686 
687 static void init_once(void *foo)
688 {
689 	struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
690 
691 	inode_init_once(&ei->vfs_inode);
692 }
693 
694 const struct file_operations hugetlbfs_file_operations = {
695 	.read			= hugetlbfs_read,
696 	.mmap			= hugetlbfs_file_mmap,
697 	.fsync			= noop_fsync,
698 	.get_unmapped_area	= hugetlb_get_unmapped_area,
699 	.llseek		= default_llseek,
700 };
701 
702 static const struct inode_operations hugetlbfs_dir_inode_operations = {
703 	.create		= hugetlbfs_create,
704 	.lookup		= simple_lookup,
705 	.link		= simple_link,
706 	.unlink		= simple_unlink,
707 	.symlink	= hugetlbfs_symlink,
708 	.mkdir		= hugetlbfs_mkdir,
709 	.rmdir		= simple_rmdir,
710 	.mknod		= hugetlbfs_mknod,
711 	.rename		= simple_rename,
712 	.setattr	= hugetlbfs_setattr,
713 };
714 
715 static const struct inode_operations hugetlbfs_inode_operations = {
716 	.setattr	= hugetlbfs_setattr,
717 };
718 
719 static const struct super_operations hugetlbfs_ops = {
720 	.alloc_inode    = hugetlbfs_alloc_inode,
721 	.destroy_inode  = hugetlbfs_destroy_inode,
722 	.evict_inode	= hugetlbfs_evict_inode,
723 	.statfs		= hugetlbfs_statfs,
724 	.put_super	= hugetlbfs_put_super,
725 	.show_options	= generic_show_options,
726 };
727 
728 static int
729 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
730 {
731 	char *p, *rest;
732 	substring_t args[MAX_OPT_ARGS];
733 	int option;
734 	unsigned long long size = 0;
735 	enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
736 
737 	if (!options)
738 		return 0;
739 
740 	while ((p = strsep(&options, ",")) != NULL) {
741 		int token;
742 		if (!*p)
743 			continue;
744 
745 		token = match_token(p, tokens, args);
746 		switch (token) {
747 		case Opt_uid:
748 			if (match_int(&args[0], &option))
749  				goto bad_val;
750 			pconfig->uid = option;
751 			break;
752 
753 		case Opt_gid:
754 			if (match_int(&args[0], &option))
755  				goto bad_val;
756 			pconfig->gid = option;
757 			break;
758 
759 		case Opt_mode:
760 			if (match_octal(&args[0], &option))
761  				goto bad_val;
762 			pconfig->mode = option & 01777U;
763 			break;
764 
765 		case Opt_size: {
766 			/* memparse() will accept a K/M/G without a digit */
767 			if (!isdigit(*args[0].from))
768 				goto bad_val;
769 			size = memparse(args[0].from, &rest);
770 			setsize = SIZE_STD;
771 			if (*rest == '%')
772 				setsize = SIZE_PERCENT;
773 			break;
774 		}
775 
776 		case Opt_nr_inodes:
777 			/* memparse() will accept a K/M/G without a digit */
778 			if (!isdigit(*args[0].from))
779 				goto bad_val;
780 			pconfig->nr_inodes = memparse(args[0].from, &rest);
781 			break;
782 
783 		case Opt_pagesize: {
784 			unsigned long ps;
785 			ps = memparse(args[0].from, &rest);
786 			pconfig->hstate = size_to_hstate(ps);
787 			if (!pconfig->hstate) {
788 				printk(KERN_ERR
789 				"hugetlbfs: Unsupported page size %lu MB\n",
790 					ps >> 20);
791 				return -EINVAL;
792 			}
793 			break;
794 		}
795 
796 		default:
797 			printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
798 				 p);
799 			return -EINVAL;
800 			break;
801 		}
802 	}
803 
804 	/* Do size after hstate is set up */
805 	if (setsize > NO_SIZE) {
806 		struct hstate *h = pconfig->hstate;
807 		if (setsize == SIZE_PERCENT) {
808 			size <<= huge_page_shift(h);
809 			size *= h->max_huge_pages;
810 			do_div(size, 100);
811 		}
812 		pconfig->nr_blocks = (size >> huge_page_shift(h));
813 	}
814 
815 	return 0;
816 
817 bad_val:
818  	printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
819 	       args[0].from, p);
820  	return -EINVAL;
821 }
822 
823 static int
824 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
825 {
826 	struct inode * inode;
827 	struct dentry * root;
828 	int ret;
829 	struct hugetlbfs_config config;
830 	struct hugetlbfs_sb_info *sbinfo;
831 
832 	save_mount_options(sb, data);
833 
834 	config.nr_blocks = -1; /* No limit on size by default */
835 	config.nr_inodes = -1; /* No limit on number of inodes by default */
836 	config.uid = current_fsuid();
837 	config.gid = current_fsgid();
838 	config.mode = 0755;
839 	config.hstate = &default_hstate;
840 	ret = hugetlbfs_parse_options(data, &config);
841 	if (ret)
842 		return ret;
843 
844 	sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
845 	if (!sbinfo)
846 		return -ENOMEM;
847 	sb->s_fs_info = sbinfo;
848 	sbinfo->hstate = config.hstate;
849 	spin_lock_init(&sbinfo->stat_lock);
850 	sbinfo->max_blocks = config.nr_blocks;
851 	sbinfo->free_blocks = config.nr_blocks;
852 	sbinfo->max_inodes = config.nr_inodes;
853 	sbinfo->free_inodes = config.nr_inodes;
854 	sb->s_maxbytes = MAX_LFS_FILESIZE;
855 	sb->s_blocksize = huge_page_size(config.hstate);
856 	sb->s_blocksize_bits = huge_page_shift(config.hstate);
857 	sb->s_magic = HUGETLBFS_MAGIC;
858 	sb->s_op = &hugetlbfs_ops;
859 	sb->s_time_gran = 1;
860 	inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
861 					S_IFDIR | config.mode, 0);
862 	if (!inode)
863 		goto out_free;
864 
865 	root = d_alloc_root(inode);
866 	if (!root) {
867 		iput(inode);
868 		goto out_free;
869 	}
870 	sb->s_root = root;
871 	return 0;
872 out_free:
873 	kfree(sbinfo);
874 	return -ENOMEM;
875 }
876 
877 int hugetlb_get_quota(struct address_space *mapping, long delta)
878 {
879 	int ret = 0;
880 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
881 
882 	if (sbinfo->free_blocks > -1) {
883 		spin_lock(&sbinfo->stat_lock);
884 		if (sbinfo->free_blocks - delta >= 0)
885 			sbinfo->free_blocks -= delta;
886 		else
887 			ret = -ENOMEM;
888 		spin_unlock(&sbinfo->stat_lock);
889 	}
890 
891 	return ret;
892 }
893 
894 void hugetlb_put_quota(struct address_space *mapping, long delta)
895 {
896 	struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
897 
898 	if (sbinfo->free_blocks > -1) {
899 		spin_lock(&sbinfo->stat_lock);
900 		sbinfo->free_blocks += delta;
901 		spin_unlock(&sbinfo->stat_lock);
902 	}
903 }
904 
905 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
906 	int flags, const char *dev_name, void *data)
907 {
908 	return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
909 }
910 
911 static struct file_system_type hugetlbfs_fs_type = {
912 	.name		= "hugetlbfs",
913 	.mount		= hugetlbfs_mount,
914 	.kill_sb	= kill_litter_super,
915 };
916 
917 static struct vfsmount *hugetlbfs_vfsmount;
918 
919 static int can_do_hugetlb_shm(void)
920 {
921 	return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
922 }
923 
924 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
925 				struct user_struct **user, int creat_flags)
926 {
927 	int error = -ENOMEM;
928 	struct file *file;
929 	struct inode *inode;
930 	struct path path;
931 	struct dentry *root;
932 	struct qstr quick_string;
933 
934 	*user = NULL;
935 	if (!hugetlbfs_vfsmount)
936 		return ERR_PTR(-ENOENT);
937 
938 	if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
939 		*user = current_user();
940 		if (user_shm_lock(size, *user)) {
941 			printk_once(KERN_WARNING "Using mlock ulimits for SHM_HUGETLB is deprecated\n");
942 		} else {
943 			*user = NULL;
944 			return ERR_PTR(-EPERM);
945 		}
946 	}
947 
948 	root = hugetlbfs_vfsmount->mnt_root;
949 	quick_string.name = name;
950 	quick_string.len = strlen(quick_string.name);
951 	quick_string.hash = 0;
952 	path.dentry = d_alloc(root, &quick_string);
953 	if (!path.dentry)
954 		goto out_shm_unlock;
955 
956 	path.mnt = mntget(hugetlbfs_vfsmount);
957 	error = -ENOSPC;
958 	inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
959 				current_fsgid(), S_IFREG | S_IRWXUGO, 0);
960 	if (!inode)
961 		goto out_dentry;
962 
963 	error = -ENOMEM;
964 	if (hugetlb_reserve_pages(inode, 0,
965 			size >> huge_page_shift(hstate_inode(inode)), NULL,
966 			acctflag))
967 		goto out_inode;
968 
969 	d_instantiate(path.dentry, inode);
970 	inode->i_size = size;
971 	inode->i_nlink = 0;
972 
973 	error = -ENFILE;
974 	file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
975 			&hugetlbfs_file_operations);
976 	if (!file)
977 		goto out_dentry; /* inode is already attached */
978 
979 	return file;
980 
981 out_inode:
982 	iput(inode);
983 out_dentry:
984 	path_put(&path);
985 out_shm_unlock:
986 	if (*user) {
987 		user_shm_unlock(size, *user);
988 		*user = NULL;
989 	}
990 	return ERR_PTR(error);
991 }
992 
993 static int __init init_hugetlbfs_fs(void)
994 {
995 	int error;
996 	struct vfsmount *vfsmount;
997 
998 	error = bdi_init(&hugetlbfs_backing_dev_info);
999 	if (error)
1000 		return error;
1001 
1002 	hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1003 					sizeof(struct hugetlbfs_inode_info),
1004 					0, 0, init_once);
1005 	if (hugetlbfs_inode_cachep == NULL)
1006 		goto out2;
1007 
1008 	error = register_filesystem(&hugetlbfs_fs_type);
1009 	if (error)
1010 		goto out;
1011 
1012 	vfsmount = kern_mount(&hugetlbfs_fs_type);
1013 
1014 	if (!IS_ERR(vfsmount)) {
1015 		hugetlbfs_vfsmount = vfsmount;
1016 		return 0;
1017 	}
1018 
1019 	error = PTR_ERR(vfsmount);
1020 
1021  out:
1022 	if (error)
1023 		kmem_cache_destroy(hugetlbfs_inode_cachep);
1024  out2:
1025 	bdi_destroy(&hugetlbfs_backing_dev_info);
1026 	return error;
1027 }
1028 
1029 static void __exit exit_hugetlbfs_fs(void)
1030 {
1031 	kmem_cache_destroy(hugetlbfs_inode_cachep);
1032 	unregister_filesystem(&hugetlbfs_fs_type);
1033 	bdi_destroy(&hugetlbfs_backing_dev_info);
1034 }
1035 
1036 module_init(init_hugetlbfs_fs)
1037 module_exit(exit_hugetlbfs_fs)
1038 
1039 MODULE_LICENSE("GPL");
1040