xref: /openbmc/linux/fs/ntfs3/file.c (revision 2f56a3f8d824d34525951483e95b0ed04c5954bf)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *
4  * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
5  *
6  *  Regular file handling primitives for NTFS-based filesystems.
7  *
8  */
9 
10 #include <linux/backing-dev.h>
11 #include <linux/blkdev.h>
12 #include <linux/buffer_head.h>
13 #include <linux/compat.h>
14 #include <linux/falloc.h>
15 #include <linux/fiemap.h>
16 
17 #include "debug.h"
18 #include "ntfs.h"
19 #include "ntfs_fs.h"
20 
21 static int ntfs_ioctl_fitrim(struct ntfs_sb_info *sbi, unsigned long arg)
22 {
23 	struct fstrim_range __user *user_range;
24 	struct fstrim_range range;
25 	int err;
26 
27 	if (!capable(CAP_SYS_ADMIN))
28 		return -EPERM;
29 
30 	if (!bdev_max_discard_sectors(sbi->sb->s_bdev))
31 		return -EOPNOTSUPP;
32 
33 	user_range = (struct fstrim_range __user *)arg;
34 	if (copy_from_user(&range, user_range, sizeof(range)))
35 		return -EFAULT;
36 
37 	range.minlen = max_t(u32, range.minlen,
38 			     bdev_discard_granularity(sbi->sb->s_bdev));
39 
40 	err = ntfs_trim_fs(sbi, &range);
41 	if (err < 0)
42 		return err;
43 
44 	if (copy_to_user(user_range, &range, sizeof(range)))
45 		return -EFAULT;
46 
47 	return 0;
48 }
49 
50 static long ntfs_ioctl(struct file *filp, u32 cmd, unsigned long arg)
51 {
52 	struct inode *inode = file_inode(filp);
53 	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
54 
55 	switch (cmd) {
56 	case FITRIM:
57 		return ntfs_ioctl_fitrim(sbi, arg);
58 	}
59 	return -ENOTTY; /* Inappropriate ioctl for device. */
60 }
61 
62 #ifdef CONFIG_COMPAT
63 static long ntfs_compat_ioctl(struct file *filp, u32 cmd, unsigned long arg)
64 
65 {
66 	return ntfs_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
67 }
68 #endif
69 
70 /*
71  * ntfs_getattr - inode_operations::getattr
72  */
73 int ntfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
74 		 struct kstat *stat, u32 request_mask, u32 flags)
75 {
76 	struct inode *inode = d_inode(path->dentry);
77 	struct ntfs_inode *ni = ntfs_i(inode);
78 
79 	if (is_compressed(ni))
80 		stat->attributes |= STATX_ATTR_COMPRESSED;
81 
82 	if (is_encrypted(ni))
83 		stat->attributes |= STATX_ATTR_ENCRYPTED;
84 
85 	stat->attributes_mask |= STATX_ATTR_COMPRESSED | STATX_ATTR_ENCRYPTED;
86 
87 	generic_fillattr(mnt_userns, inode, stat);
88 
89 	stat->result_mask |= STATX_BTIME;
90 	stat->btime = ni->i_crtime;
91 	stat->blksize = ni->mi.sbi->cluster_size; /* 512, 1K, ..., 2M */
92 
93 	return 0;
94 }
95 
96 static int ntfs_extend_initialized_size(struct file *file,
97 					struct ntfs_inode *ni,
98 					const loff_t valid,
99 					const loff_t new_valid)
100 {
101 	struct inode *inode = &ni->vfs_inode;
102 	struct address_space *mapping = inode->i_mapping;
103 	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
104 	loff_t pos = valid;
105 	int err;
106 
107 	if (is_resident(ni)) {
108 		ni->i_valid = new_valid;
109 		return 0;
110 	}
111 
112 	WARN_ON(is_compressed(ni));
113 	WARN_ON(valid >= new_valid);
114 
115 	for (;;) {
116 		u32 zerofrom, len;
117 		struct page *page;
118 		u8 bits;
119 		CLST vcn, lcn, clen;
120 
121 		if (is_sparsed(ni)) {
122 			bits = sbi->cluster_bits;
123 			vcn = pos >> bits;
124 
125 			err = attr_data_get_block(ni, vcn, 0, &lcn, &clen,
126 						  NULL);
127 			if (err)
128 				goto out;
129 
130 			if (lcn == SPARSE_LCN) {
131 				pos = ((loff_t)clen + vcn) << bits;
132 				ni->i_valid = pos;
133 				goto next;
134 			}
135 		}
136 
137 		zerofrom = pos & (PAGE_SIZE - 1);
138 		len = PAGE_SIZE - zerofrom;
139 
140 		if (pos + len > new_valid)
141 			len = new_valid - pos;
142 
143 		err = ntfs_write_begin(file, mapping, pos, len, &page, NULL);
144 		if (err)
145 			goto out;
146 
147 		zero_user_segment(page, zerofrom, PAGE_SIZE);
148 
149 		/* This function in any case puts page. */
150 		err = ntfs_write_end(file, mapping, pos, len, len, page, NULL);
151 		if (err < 0)
152 			goto out;
153 		pos += len;
154 
155 next:
156 		if (pos >= new_valid)
157 			break;
158 
159 		balance_dirty_pages_ratelimited(mapping);
160 		cond_resched();
161 	}
162 
163 	return 0;
164 
165 out:
166 	ni->i_valid = valid;
167 	ntfs_inode_warn(inode, "failed to extend initialized size to %llx.",
168 			new_valid);
169 	return err;
170 }
171 
172 /*
173  * ntfs_zero_range - Helper function for punch_hole.
174  *
175  * It zeroes a range [vbo, vbo_to).
176  */
177 static int ntfs_zero_range(struct inode *inode, u64 vbo, u64 vbo_to)
178 {
179 	int err = 0;
180 	struct address_space *mapping = inode->i_mapping;
181 	u32 blocksize = 1 << inode->i_blkbits;
182 	pgoff_t idx = vbo >> PAGE_SHIFT;
183 	u32 z_start = vbo & (PAGE_SIZE - 1);
184 	pgoff_t idx_end = (vbo_to + PAGE_SIZE - 1) >> PAGE_SHIFT;
185 	loff_t page_off;
186 	struct buffer_head *head, *bh;
187 	u32 bh_next, bh_off, z_end;
188 	sector_t iblock;
189 	struct page *page;
190 
191 	for (; idx < idx_end; idx += 1, z_start = 0) {
192 		page_off = (loff_t)idx << PAGE_SHIFT;
193 		z_end = (page_off + PAGE_SIZE) > vbo_to ? (vbo_to - page_off)
194 							: PAGE_SIZE;
195 		iblock = page_off >> inode->i_blkbits;
196 
197 		page = find_or_create_page(mapping, idx,
198 					   mapping_gfp_constraint(mapping,
199 								  ~__GFP_FS));
200 		if (!page)
201 			return -ENOMEM;
202 
203 		if (!page_has_buffers(page))
204 			create_empty_buffers(page, blocksize, 0);
205 
206 		bh = head = page_buffers(page);
207 		bh_off = 0;
208 		do {
209 			bh_next = bh_off + blocksize;
210 
211 			if (bh_next <= z_start || bh_off >= z_end)
212 				continue;
213 
214 			if (!buffer_mapped(bh)) {
215 				ntfs_get_block(inode, iblock, bh, 0);
216 				/* Unmapped? It's a hole - nothing to do. */
217 				if (!buffer_mapped(bh))
218 					continue;
219 			}
220 
221 			/* Ok, it's mapped. Make sure it's up-to-date. */
222 			if (PageUptodate(page))
223 				set_buffer_uptodate(bh);
224 
225 			if (!buffer_uptodate(bh)) {
226 				lock_buffer(bh);
227 				bh->b_end_io = end_buffer_read_sync;
228 				get_bh(bh);
229 				submit_bh(REQ_OP_READ, bh);
230 
231 				wait_on_buffer(bh);
232 				if (!buffer_uptodate(bh)) {
233 					unlock_page(page);
234 					put_page(page);
235 					err = -EIO;
236 					goto out;
237 				}
238 			}
239 
240 			mark_buffer_dirty(bh);
241 
242 		} while (bh_off = bh_next, iblock += 1,
243 			 head != (bh = bh->b_this_page));
244 
245 		zero_user_segment(page, z_start, z_end);
246 
247 		unlock_page(page);
248 		put_page(page);
249 		cond_resched();
250 	}
251 out:
252 	mark_inode_dirty(inode);
253 	return err;
254 }
255 
256 /*
257  * ntfs_sparse_cluster - Helper function to zero a new allocated clusters.
258  *
259  * NOTE: 512 <= cluster size <= 2M
260  */
261 void ntfs_sparse_cluster(struct inode *inode, struct page *page0, CLST vcn,
262 			 CLST len)
263 {
264 	struct address_space *mapping = inode->i_mapping;
265 	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
266 	u8 cluster_bits = sbi->cluster_bits;
267 	u64 vbo = (u64)vcn << cluster_bits;
268 	u64 bytes = (u64)len << cluster_bits;
269 	u32 blocksize = 1 << inode->i_blkbits;
270 	pgoff_t idx0 = page0 ? page0->index : -1;
271 	loff_t vbo_clst = vbo & sbi->cluster_mask_inv;
272 	loff_t end = ntfs_up_cluster(sbi, vbo + bytes);
273 	pgoff_t idx = vbo_clst >> PAGE_SHIFT;
274 	u32 from = vbo_clst & (PAGE_SIZE - 1);
275 	pgoff_t idx_end = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
276 	loff_t page_off;
277 	u32 to;
278 	bool partial;
279 	struct page *page;
280 
281 	for (; idx < idx_end; idx += 1, from = 0) {
282 		page = idx == idx0 ? page0 : grab_cache_page(mapping, idx);
283 
284 		if (!page)
285 			continue;
286 
287 		page_off = (loff_t)idx << PAGE_SHIFT;
288 		to = (page_off + PAGE_SIZE) > end ? (end - page_off)
289 						  : PAGE_SIZE;
290 		partial = false;
291 
292 		if ((from || PAGE_SIZE != to) &&
293 		    likely(!page_has_buffers(page))) {
294 			create_empty_buffers(page, blocksize, 0);
295 		}
296 
297 		if (page_has_buffers(page)) {
298 			struct buffer_head *head, *bh;
299 			u32 bh_off = 0;
300 
301 			bh = head = page_buffers(page);
302 			do {
303 				u32 bh_next = bh_off + blocksize;
304 
305 				if (from <= bh_off && bh_next <= to) {
306 					set_buffer_uptodate(bh);
307 					mark_buffer_dirty(bh);
308 				} else if (!buffer_uptodate(bh)) {
309 					partial = true;
310 				}
311 				bh_off = bh_next;
312 			} while (head != (bh = bh->b_this_page));
313 		}
314 
315 		zero_user_segment(page, from, to);
316 
317 		if (!partial)
318 			SetPageUptodate(page);
319 		flush_dcache_page(page);
320 		set_page_dirty(page);
321 
322 		if (idx != idx0) {
323 			unlock_page(page);
324 			put_page(page);
325 		}
326 		cond_resched();
327 	}
328 }
329 
330 /*
331  * ntfs_file_mmap - file_operations::mmap
332  */
333 static int ntfs_file_mmap(struct file *file, struct vm_area_struct *vma)
334 {
335 	struct address_space *mapping = file->f_mapping;
336 	struct inode *inode = mapping->host;
337 	struct ntfs_inode *ni = ntfs_i(inode);
338 	u64 from = ((u64)vma->vm_pgoff << PAGE_SHIFT);
339 	bool rw = vma->vm_flags & VM_WRITE;
340 	int err;
341 
342 	if (is_encrypted(ni)) {
343 		ntfs_inode_warn(inode, "mmap encrypted not supported");
344 		return -EOPNOTSUPP;
345 	}
346 
347 	if (is_dedup(ni)) {
348 		ntfs_inode_warn(inode, "mmap deduplicated not supported");
349 		return -EOPNOTSUPP;
350 	}
351 
352 	if (is_compressed(ni) && rw) {
353 		ntfs_inode_warn(inode, "mmap(write) compressed not supported");
354 		return -EOPNOTSUPP;
355 	}
356 
357 	if (rw) {
358 		u64 to = min_t(loff_t, i_size_read(inode),
359 			       from + vma->vm_end - vma->vm_start);
360 
361 		if (is_sparsed(ni)) {
362 			/* Allocate clusters for rw map. */
363 			struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
364 			CLST lcn, len;
365 			CLST vcn = from >> sbi->cluster_bits;
366 			CLST end = bytes_to_cluster(sbi, to);
367 			bool new;
368 
369 			for (; vcn < end; vcn += len) {
370 				err = attr_data_get_block(ni, vcn, 1, &lcn,
371 							  &len, &new);
372 				if (err)
373 					goto out;
374 
375 				if (!new)
376 					continue;
377 				ntfs_sparse_cluster(inode, NULL, vcn, 1);
378 			}
379 		}
380 
381 		if (ni->i_valid < to) {
382 			if (!inode_trylock(inode)) {
383 				err = -EAGAIN;
384 				goto out;
385 			}
386 			err = ntfs_extend_initialized_size(file, ni,
387 							   ni->i_valid, to);
388 			inode_unlock(inode);
389 			if (err)
390 				goto out;
391 		}
392 	}
393 
394 	err = generic_file_mmap(file, vma);
395 out:
396 	return err;
397 }
398 
399 static int ntfs_extend(struct inode *inode, loff_t pos, size_t count,
400 		       struct file *file)
401 {
402 	struct ntfs_inode *ni = ntfs_i(inode);
403 	struct address_space *mapping = inode->i_mapping;
404 	loff_t end = pos + count;
405 	bool extend_init = file && pos > ni->i_valid;
406 	int err;
407 
408 	if (end <= inode->i_size && !extend_init)
409 		return 0;
410 
411 	/* Mark rw ntfs as dirty. It will be cleared at umount. */
412 	ntfs_set_state(ni->mi.sbi, NTFS_DIRTY_DIRTY);
413 
414 	if (end > inode->i_size) {
415 		err = ntfs_set_size(inode, end);
416 		if (err)
417 			goto out;
418 		inode->i_size = end;
419 	}
420 
421 	if (extend_init && !is_compressed(ni)) {
422 		err = ntfs_extend_initialized_size(file, ni, ni->i_valid, pos);
423 		if (err)
424 			goto out;
425 	} else {
426 		err = 0;
427 	}
428 
429 	inode->i_ctime = inode->i_mtime = current_time(inode);
430 	mark_inode_dirty(inode);
431 
432 	if (IS_SYNC(inode)) {
433 		int err2;
434 
435 		err = filemap_fdatawrite_range(mapping, pos, end - 1);
436 		err2 = sync_mapping_buffers(mapping);
437 		if (!err)
438 			err = err2;
439 		err2 = write_inode_now(inode, 1);
440 		if (!err)
441 			err = err2;
442 		if (!err)
443 			err = filemap_fdatawait_range(mapping, pos, end - 1);
444 	}
445 
446 out:
447 	return err;
448 }
449 
450 static int ntfs_truncate(struct inode *inode, loff_t new_size)
451 {
452 	struct super_block *sb = inode->i_sb;
453 	struct ntfs_inode *ni = ntfs_i(inode);
454 	int err, dirty = 0;
455 	u64 new_valid;
456 
457 	if (!S_ISREG(inode->i_mode))
458 		return 0;
459 
460 	if (is_compressed(ni)) {
461 		if (ni->i_valid > new_size)
462 			ni->i_valid = new_size;
463 	} else {
464 		err = block_truncate_page(inode->i_mapping, new_size,
465 					  ntfs_get_block);
466 		if (err)
467 			return err;
468 	}
469 
470 	new_valid = ntfs_up_block(sb, min_t(u64, ni->i_valid, new_size));
471 
472 	ni_lock(ni);
473 
474 	truncate_setsize(inode, new_size);
475 
476 	down_write(&ni->file.run_lock);
477 	err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run, new_size,
478 			    &new_valid, ni->mi.sbi->options->prealloc, NULL);
479 	up_write(&ni->file.run_lock);
480 
481 	if (new_valid < ni->i_valid)
482 		ni->i_valid = new_valid;
483 
484 	ni_unlock(ni);
485 
486 	ni->std_fa |= FILE_ATTRIBUTE_ARCHIVE;
487 	inode->i_ctime = inode->i_mtime = current_time(inode);
488 	if (!IS_DIRSYNC(inode)) {
489 		dirty = 1;
490 	} else {
491 		err = ntfs_sync_inode(inode);
492 		if (err)
493 			return err;
494 	}
495 
496 	if (dirty)
497 		mark_inode_dirty(inode);
498 
499 	/*ntfs_flush_inodes(inode->i_sb, inode, NULL);*/
500 
501 	return 0;
502 }
503 
504 /*
505  * ntfs_fallocate
506  *
507  * Preallocate space for a file. This implements ntfs's fallocate file
508  * operation, which gets called from sys_fallocate system call. User
509  * space requests 'len' bytes at 'vbo'. If FALLOC_FL_KEEP_SIZE is set
510  * we just allocate clusters without zeroing them out. Otherwise we
511  * allocate and zero out clusters via an expanding truncate.
512  */
513 static long ntfs_fallocate(struct file *file, int mode, loff_t vbo, loff_t len)
514 {
515 	struct inode *inode = file->f_mapping->host;
516 	struct address_space *mapping = inode->i_mapping;
517 	struct super_block *sb = inode->i_sb;
518 	struct ntfs_sb_info *sbi = sb->s_fs_info;
519 	struct ntfs_inode *ni = ntfs_i(inode);
520 	loff_t end = vbo + len;
521 	loff_t vbo_down = round_down(vbo, PAGE_SIZE);
522 	bool is_supported_holes = is_sparsed(ni) || is_compressed(ni);
523 	loff_t i_size, new_size;
524 	bool map_locked;
525 	int err;
526 
527 	/* No support for dir. */
528 	if (!S_ISREG(inode->i_mode))
529 		return -EOPNOTSUPP;
530 
531 	/*
532 	 * vfs_fallocate checks all possible combinations of mode.
533 	 * Do additional checks here before ntfs_set_state(dirty).
534 	 */
535 	if (mode & FALLOC_FL_PUNCH_HOLE) {
536 		if (!is_supported_holes)
537 			return -EOPNOTSUPP;
538 	} else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
539 	} else if (mode & FALLOC_FL_INSERT_RANGE) {
540 		if (!is_supported_holes)
541 			return -EOPNOTSUPP;
542 	} else if (mode &
543 		   ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
544 		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)) {
545 		ntfs_inode_warn(inode, "fallocate(0x%x) is not supported",
546 				mode);
547 		return -EOPNOTSUPP;
548 	}
549 
550 	ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
551 
552 	inode_lock(inode);
553 	i_size = inode->i_size;
554 	new_size = max(end, i_size);
555 	map_locked = false;
556 
557 	if (WARN_ON(ni->ni_flags & NI_FLAG_COMPRESSED_MASK)) {
558 		/* Should never be here, see ntfs_file_open. */
559 		err = -EOPNOTSUPP;
560 		goto out;
561 	}
562 
563 	if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE |
564 		    FALLOC_FL_INSERT_RANGE)) {
565 		inode_dio_wait(inode);
566 		filemap_invalidate_lock(mapping);
567 		map_locked = true;
568 	}
569 
570 	if (mode & FALLOC_FL_PUNCH_HOLE) {
571 		u32 frame_size;
572 		loff_t mask, vbo_a, end_a, tmp;
573 
574 		err = filemap_write_and_wait_range(mapping, vbo, LLONG_MAX);
575 		if (err)
576 			goto out;
577 
578 		truncate_pagecache(inode, vbo_down);
579 
580 		ni_lock(ni);
581 		err = attr_punch_hole(ni, vbo, len, &frame_size);
582 		ni_unlock(ni);
583 		if (err != E_NTFS_NOTALIGNED)
584 			goto out;
585 
586 		/* Process not aligned punch. */
587 		mask = frame_size - 1;
588 		vbo_a = (vbo + mask) & ~mask;
589 		end_a = end & ~mask;
590 
591 		tmp = min(vbo_a, end);
592 		if (tmp > vbo) {
593 			err = ntfs_zero_range(inode, vbo, tmp);
594 			if (err)
595 				goto out;
596 		}
597 
598 		if (vbo < end_a && end_a < end) {
599 			err = ntfs_zero_range(inode, end_a, end);
600 			if (err)
601 				goto out;
602 		}
603 
604 		/* Aligned punch_hole */
605 		if (end_a > vbo_a) {
606 			ni_lock(ni);
607 			err = attr_punch_hole(ni, vbo_a, end_a - vbo_a, NULL);
608 			ni_unlock(ni);
609 		}
610 	} else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
611 		/*
612 		 * Write tail of the last page before removed range since
613 		 * it will get removed from the page cache below.
614 		 */
615 		err = filemap_write_and_wait_range(mapping, vbo_down, vbo);
616 		if (err)
617 			goto out;
618 
619 		/*
620 		 * Write data that will be shifted to preserve them
621 		 * when discarding page cache below.
622 		 */
623 		err = filemap_write_and_wait_range(mapping, end, LLONG_MAX);
624 		if (err)
625 			goto out;
626 
627 		truncate_pagecache(inode, vbo_down);
628 
629 		ni_lock(ni);
630 		err = attr_collapse_range(ni, vbo, len);
631 		ni_unlock(ni);
632 	} else if (mode & FALLOC_FL_INSERT_RANGE) {
633 		/* Check new size. */
634 		err = inode_newsize_ok(inode, new_size);
635 		if (err)
636 			goto out;
637 
638 		/* Write out all dirty pages. */
639 		err = filemap_write_and_wait_range(mapping, vbo_down,
640 						   LLONG_MAX);
641 		if (err)
642 			goto out;
643 		truncate_pagecache(inode, vbo_down);
644 
645 		ni_lock(ni);
646 		err = attr_insert_range(ni, vbo, len);
647 		ni_unlock(ni);
648 	} else {
649 		/* Check new size. */
650 
651 		/* generic/213: expected -ENOSPC instead of -EFBIG. */
652 		if (!is_supported_holes) {
653 			loff_t to_alloc = new_size - inode_get_bytes(inode);
654 
655 			if (to_alloc > 0 &&
656 			    (to_alloc >> sbi->cluster_bits) >
657 				    wnd_zeroes(&sbi->used.bitmap)) {
658 				err = -ENOSPC;
659 				goto out;
660 			}
661 		}
662 
663 		err = inode_newsize_ok(inode, new_size);
664 		if (err)
665 			goto out;
666 
667 		/*
668 		 * Allocate clusters, do not change 'valid' size.
669 		 */
670 		err = ntfs_set_size(inode, new_size);
671 		if (err)
672 			goto out;
673 
674 		if (is_supported_holes) {
675 			CLST vcn_v = bytes_to_cluster(sbi, ni->i_valid);
676 			CLST vcn = vbo >> sbi->cluster_bits;
677 			CLST cend = bytes_to_cluster(sbi, end);
678 			CLST lcn, clen;
679 			bool new;
680 
681 			/*
682 			 * Allocate but do not zero new clusters. (see below comments)
683 			 * This breaks security: One can read unused on-disk areas.
684 			 * Zeroing these clusters may be too long.
685 			 * Maybe we should check here for root rights?
686 			 */
687 			for (; vcn < cend; vcn += clen) {
688 				err = attr_data_get_block(ni, vcn, cend - vcn,
689 							  &lcn, &clen, &new);
690 				if (err)
691 					goto out;
692 				if (!new || vcn >= vcn_v)
693 					continue;
694 
695 				/*
696 				 * Unwritten area.
697 				 * NTFS is not able to store several unwritten areas.
698 				 * Activate 'ntfs_sparse_cluster' to zero new allocated clusters.
699 				 *
700 				 * Dangerous in case:
701 				 * 1G of sparsed clusters + 1 cluster of data =>
702 				 * valid_size == 1G + 1 cluster
703 				 * fallocate(1G) will zero 1G and this can be very long
704 				 * xfstest 016/086 will fail without 'ntfs_sparse_cluster'.
705 				 */
706 				ntfs_sparse_cluster(inode, NULL, vcn,
707 						    min(vcn_v - vcn, clen));
708 			}
709 		}
710 
711 		if (mode & FALLOC_FL_KEEP_SIZE) {
712 			ni_lock(ni);
713 			/* True - Keep preallocated. */
714 			err = attr_set_size(ni, ATTR_DATA, NULL, 0,
715 					    &ni->file.run, i_size, &ni->i_valid,
716 					    true, NULL);
717 			ni_unlock(ni);
718 		}
719 	}
720 
721 out:
722 	if (map_locked)
723 		filemap_invalidate_unlock(mapping);
724 
725 	if (!err) {
726 		inode->i_ctime = inode->i_mtime = current_time(inode);
727 		mark_inode_dirty(inode);
728 	}
729 
730 	inode_unlock(inode);
731 	return err;
732 }
733 
734 /*
735  * ntfs3_setattr - inode_operations::setattr
736  */
737 int ntfs3_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
738 		  struct iattr *attr)
739 {
740 	struct super_block *sb = dentry->d_sb;
741 	struct ntfs_sb_info *sbi = sb->s_fs_info;
742 	struct inode *inode = d_inode(dentry);
743 	struct ntfs_inode *ni = ntfs_i(inode);
744 	u32 ia_valid = attr->ia_valid;
745 	umode_t mode = inode->i_mode;
746 	int err;
747 
748 	if (sbi->options->noacsrules) {
749 		/* "No access rules" - Force any changes of time etc. */
750 		attr->ia_valid |= ATTR_FORCE;
751 		/* and disable for editing some attributes. */
752 		attr->ia_valid &= ~(ATTR_UID | ATTR_GID | ATTR_MODE);
753 		ia_valid = attr->ia_valid;
754 	}
755 
756 	err = setattr_prepare(mnt_userns, dentry, attr);
757 	if (err)
758 		goto out;
759 
760 	if (ia_valid & ATTR_SIZE) {
761 		loff_t oldsize = inode->i_size;
762 
763 		if (WARN_ON(ni->ni_flags & NI_FLAG_COMPRESSED_MASK)) {
764 			/* Should never be here, see ntfs_file_open(). */
765 			err = -EOPNOTSUPP;
766 			goto out;
767 		}
768 		inode_dio_wait(inode);
769 
770 		if (attr->ia_size <= oldsize)
771 			err = ntfs_truncate(inode, attr->ia_size);
772 		else if (attr->ia_size > oldsize)
773 			err = ntfs_extend(inode, attr->ia_size, 0, NULL);
774 
775 		if (err)
776 			goto out;
777 
778 		ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
779 	}
780 
781 	setattr_copy(mnt_userns, inode, attr);
782 
783 	if (mode != inode->i_mode) {
784 		err = ntfs_acl_chmod(mnt_userns, inode);
785 		if (err)
786 			goto out;
787 
788 		/* Linux 'w' -> Windows 'ro'. */
789 		if (0222 & inode->i_mode)
790 			ni->std_fa &= ~FILE_ATTRIBUTE_READONLY;
791 		else
792 			ni->std_fa |= FILE_ATTRIBUTE_READONLY;
793 	}
794 
795 	if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE))
796 		ntfs_save_wsl_perm(inode);
797 	mark_inode_dirty(inode);
798 out:
799 	return err;
800 }
801 
802 static ssize_t ntfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
803 {
804 	struct file *file = iocb->ki_filp;
805 	struct inode *inode = file->f_mapping->host;
806 	struct ntfs_inode *ni = ntfs_i(inode);
807 
808 	if (is_encrypted(ni)) {
809 		ntfs_inode_warn(inode, "encrypted i/o not supported");
810 		return -EOPNOTSUPP;
811 	}
812 
813 	if (is_compressed(ni) && (iocb->ki_flags & IOCB_DIRECT)) {
814 		ntfs_inode_warn(inode, "direct i/o + compressed not supported");
815 		return -EOPNOTSUPP;
816 	}
817 
818 #ifndef CONFIG_NTFS3_LZX_XPRESS
819 	if (ni->ni_flags & NI_FLAG_COMPRESSED_MASK) {
820 		ntfs_inode_warn(
821 			inode,
822 			"activate CONFIG_NTFS3_LZX_XPRESS to read external compressed files");
823 		return -EOPNOTSUPP;
824 	}
825 #endif
826 
827 	if (is_dedup(ni)) {
828 		ntfs_inode_warn(inode, "read deduplicated not supported");
829 		return -EOPNOTSUPP;
830 	}
831 
832 	return generic_file_read_iter(iocb, iter);
833 }
834 
835 /*
836  * ntfs_get_frame_pages
837  *
838  * Return: Array of locked pages.
839  */
840 static int ntfs_get_frame_pages(struct address_space *mapping, pgoff_t index,
841 				struct page **pages, u32 pages_per_frame,
842 				bool *frame_uptodate)
843 {
844 	gfp_t gfp_mask = mapping_gfp_mask(mapping);
845 	u32 npages;
846 
847 	*frame_uptodate = true;
848 
849 	for (npages = 0; npages < pages_per_frame; npages++, index++) {
850 		struct page *page;
851 
852 		page = find_or_create_page(mapping, index, gfp_mask);
853 		if (!page) {
854 			while (npages--) {
855 				page = pages[npages];
856 				unlock_page(page);
857 				put_page(page);
858 			}
859 
860 			return -ENOMEM;
861 		}
862 
863 		if (!PageUptodate(page))
864 			*frame_uptodate = false;
865 
866 		pages[npages] = page;
867 	}
868 
869 	return 0;
870 }
871 
872 /*
873  * ntfs_compress_write - Helper for ntfs_file_write_iter() (compressed files).
874  */
875 static ssize_t ntfs_compress_write(struct kiocb *iocb, struct iov_iter *from)
876 {
877 	int err;
878 	struct file *file = iocb->ki_filp;
879 	size_t count = iov_iter_count(from);
880 	loff_t pos = iocb->ki_pos;
881 	struct inode *inode = file_inode(file);
882 	loff_t i_size = inode->i_size;
883 	struct address_space *mapping = inode->i_mapping;
884 	struct ntfs_inode *ni = ntfs_i(inode);
885 	u64 valid = ni->i_valid;
886 	struct ntfs_sb_info *sbi = ni->mi.sbi;
887 	struct page *page, **pages = NULL;
888 	size_t written = 0;
889 	u8 frame_bits = NTFS_LZNT_CUNIT + sbi->cluster_bits;
890 	u32 frame_size = 1u << frame_bits;
891 	u32 pages_per_frame = frame_size >> PAGE_SHIFT;
892 	u32 ip, off;
893 	CLST frame;
894 	u64 frame_vbo;
895 	pgoff_t index;
896 	bool frame_uptodate;
897 
898 	if (frame_size < PAGE_SIZE) {
899 		/*
900 		 * frame_size == 8K if cluster 512
901 		 * frame_size == 64K if cluster 4096
902 		 */
903 		ntfs_inode_warn(inode, "page size is bigger than frame size");
904 		return -EOPNOTSUPP;
905 	}
906 
907 	pages = kmalloc_array(pages_per_frame, sizeof(struct page *), GFP_NOFS);
908 	if (!pages)
909 		return -ENOMEM;
910 
911 	current->backing_dev_info = inode_to_bdi(inode);
912 	err = file_remove_privs(file);
913 	if (err)
914 		goto out;
915 
916 	err = file_update_time(file);
917 	if (err)
918 		goto out;
919 
920 	/* Zero range [valid : pos). */
921 	while (valid < pos) {
922 		CLST lcn, clen;
923 
924 		frame = valid >> frame_bits;
925 		frame_vbo = valid & ~(frame_size - 1);
926 		off = valid & (frame_size - 1);
927 
928 		err = attr_data_get_block(ni, frame << NTFS_LZNT_CUNIT, 0, &lcn,
929 					  &clen, NULL);
930 		if (err)
931 			goto out;
932 
933 		if (lcn == SPARSE_LCN) {
934 			ni->i_valid = valid =
935 				frame_vbo + ((u64)clen << sbi->cluster_bits);
936 			continue;
937 		}
938 
939 		/* Load full frame. */
940 		err = ntfs_get_frame_pages(mapping, frame_vbo >> PAGE_SHIFT,
941 					   pages, pages_per_frame,
942 					   &frame_uptodate);
943 		if (err)
944 			goto out;
945 
946 		if (!frame_uptodate && off) {
947 			err = ni_read_frame(ni, frame_vbo, pages,
948 					    pages_per_frame);
949 			if (err) {
950 				for (ip = 0; ip < pages_per_frame; ip++) {
951 					page = pages[ip];
952 					unlock_page(page);
953 					put_page(page);
954 				}
955 				goto out;
956 			}
957 		}
958 
959 		ip = off >> PAGE_SHIFT;
960 		off = offset_in_page(valid);
961 		for (; ip < pages_per_frame; ip++, off = 0) {
962 			page = pages[ip];
963 			zero_user_segment(page, off, PAGE_SIZE);
964 			flush_dcache_page(page);
965 			SetPageUptodate(page);
966 		}
967 
968 		ni_lock(ni);
969 		err = ni_write_frame(ni, pages, pages_per_frame);
970 		ni_unlock(ni);
971 
972 		for (ip = 0; ip < pages_per_frame; ip++) {
973 			page = pages[ip];
974 			SetPageUptodate(page);
975 			unlock_page(page);
976 			put_page(page);
977 		}
978 
979 		if (err)
980 			goto out;
981 
982 		ni->i_valid = valid = frame_vbo + frame_size;
983 	}
984 
985 	/* Copy user data [pos : pos + count). */
986 	while (count) {
987 		size_t copied, bytes;
988 
989 		off = pos & (frame_size - 1);
990 		bytes = frame_size - off;
991 		if (bytes > count)
992 			bytes = count;
993 
994 		frame_vbo = pos & ~(frame_size - 1);
995 		index = frame_vbo >> PAGE_SHIFT;
996 
997 		if (unlikely(fault_in_iov_iter_readable(from, bytes))) {
998 			err = -EFAULT;
999 			goto out;
1000 		}
1001 
1002 		/* Load full frame. */
1003 		err = ntfs_get_frame_pages(mapping, index, pages,
1004 					   pages_per_frame, &frame_uptodate);
1005 		if (err)
1006 			goto out;
1007 
1008 		if (!frame_uptodate) {
1009 			loff_t to = pos + bytes;
1010 
1011 			if (off || (to < i_size && (to & (frame_size - 1)))) {
1012 				err = ni_read_frame(ni, frame_vbo, pages,
1013 						    pages_per_frame);
1014 				if (err) {
1015 					for (ip = 0; ip < pages_per_frame;
1016 					     ip++) {
1017 						page = pages[ip];
1018 						unlock_page(page);
1019 						put_page(page);
1020 					}
1021 					goto out;
1022 				}
1023 			}
1024 		}
1025 
1026 		WARN_ON(!bytes);
1027 		copied = 0;
1028 		ip = off >> PAGE_SHIFT;
1029 		off = offset_in_page(pos);
1030 
1031 		/* Copy user data to pages. */
1032 		for (;;) {
1033 			size_t cp, tail = PAGE_SIZE - off;
1034 
1035 			page = pages[ip];
1036 			cp = copy_page_from_iter_atomic(page, off,
1037 							min(tail, bytes), from);
1038 			flush_dcache_page(page);
1039 
1040 			copied += cp;
1041 			bytes -= cp;
1042 			if (!bytes || !cp)
1043 				break;
1044 
1045 			if (cp < tail) {
1046 				off += cp;
1047 			} else {
1048 				ip++;
1049 				off = 0;
1050 			}
1051 		}
1052 
1053 		ni_lock(ni);
1054 		err = ni_write_frame(ni, pages, pages_per_frame);
1055 		ni_unlock(ni);
1056 
1057 		for (ip = 0; ip < pages_per_frame; ip++) {
1058 			page = pages[ip];
1059 			ClearPageDirty(page);
1060 			SetPageUptodate(page);
1061 			unlock_page(page);
1062 			put_page(page);
1063 		}
1064 
1065 		if (err)
1066 			goto out;
1067 
1068 		/*
1069 		 * We can loop for a long time in here. Be nice and allow
1070 		 * us to schedule out to avoid softlocking if preempt
1071 		 * is disabled.
1072 		 */
1073 		cond_resched();
1074 
1075 		pos += copied;
1076 		written += copied;
1077 
1078 		count = iov_iter_count(from);
1079 	}
1080 
1081 out:
1082 	kfree(pages);
1083 
1084 	current->backing_dev_info = NULL;
1085 
1086 	if (err < 0)
1087 		return err;
1088 
1089 	iocb->ki_pos += written;
1090 	if (iocb->ki_pos > ni->i_valid)
1091 		ni->i_valid = iocb->ki_pos;
1092 
1093 	return written;
1094 }
1095 
1096 /*
1097  * ntfs_file_write_iter - file_operations::write_iter
1098  */
1099 static ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1100 {
1101 	struct file *file = iocb->ki_filp;
1102 	struct address_space *mapping = file->f_mapping;
1103 	struct inode *inode = mapping->host;
1104 	ssize_t ret;
1105 	struct ntfs_inode *ni = ntfs_i(inode);
1106 
1107 	if (is_encrypted(ni)) {
1108 		ntfs_inode_warn(inode, "encrypted i/o not supported");
1109 		return -EOPNOTSUPP;
1110 	}
1111 
1112 	if (is_compressed(ni) && (iocb->ki_flags & IOCB_DIRECT)) {
1113 		ntfs_inode_warn(inode, "direct i/o + compressed not supported");
1114 		return -EOPNOTSUPP;
1115 	}
1116 
1117 	if (is_dedup(ni)) {
1118 		ntfs_inode_warn(inode, "write into deduplicated not supported");
1119 		return -EOPNOTSUPP;
1120 	}
1121 
1122 	if (!inode_trylock(inode)) {
1123 		if (iocb->ki_flags & IOCB_NOWAIT)
1124 			return -EAGAIN;
1125 		inode_lock(inode);
1126 	}
1127 
1128 	ret = generic_write_checks(iocb, from);
1129 	if (ret <= 0)
1130 		goto out;
1131 
1132 	if (WARN_ON(ni->ni_flags & NI_FLAG_COMPRESSED_MASK)) {
1133 		/* Should never be here, see ntfs_file_open(). */
1134 		ret = -EOPNOTSUPP;
1135 		goto out;
1136 	}
1137 
1138 	ret = ntfs_extend(inode, iocb->ki_pos, ret, file);
1139 	if (ret)
1140 		goto out;
1141 
1142 	ret = is_compressed(ni) ? ntfs_compress_write(iocb, from)
1143 				: __generic_file_write_iter(iocb, from);
1144 
1145 out:
1146 	inode_unlock(inode);
1147 
1148 	if (ret > 0)
1149 		ret = generic_write_sync(iocb, ret);
1150 
1151 	return ret;
1152 }
1153 
1154 /*
1155  * ntfs_file_open - file_operations::open
1156  */
1157 int ntfs_file_open(struct inode *inode, struct file *file)
1158 {
1159 	struct ntfs_inode *ni = ntfs_i(inode);
1160 
1161 	if (unlikely((is_compressed(ni) || is_encrypted(ni)) &&
1162 		     (file->f_flags & O_DIRECT))) {
1163 		return -EOPNOTSUPP;
1164 	}
1165 
1166 	/* Decompress "external compressed" file if opened for rw. */
1167 	if ((ni->ni_flags & NI_FLAG_COMPRESSED_MASK) &&
1168 	    (file->f_flags & (O_WRONLY | O_RDWR | O_TRUNC))) {
1169 #ifdef CONFIG_NTFS3_LZX_XPRESS
1170 		int err = ni_decompress_file(ni);
1171 
1172 		if (err)
1173 			return err;
1174 #else
1175 		ntfs_inode_warn(
1176 			inode,
1177 			"activate CONFIG_NTFS3_LZX_XPRESS to write external compressed files");
1178 		return -EOPNOTSUPP;
1179 #endif
1180 	}
1181 
1182 	return generic_file_open(inode, file);
1183 }
1184 
1185 /*
1186  * ntfs_file_release - file_operations::release
1187  */
1188 static int ntfs_file_release(struct inode *inode, struct file *file)
1189 {
1190 	struct ntfs_inode *ni = ntfs_i(inode);
1191 	struct ntfs_sb_info *sbi = ni->mi.sbi;
1192 	int err = 0;
1193 
1194 	/* If we are last writer on the inode, drop the block reservation. */
1195 	if (sbi->options->prealloc && ((file->f_mode & FMODE_WRITE) &&
1196 				      atomic_read(&inode->i_writecount) == 1)) {
1197 		ni_lock(ni);
1198 		down_write(&ni->file.run_lock);
1199 
1200 		err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run,
1201 				    inode->i_size, &ni->i_valid, false, NULL);
1202 
1203 		up_write(&ni->file.run_lock);
1204 		ni_unlock(ni);
1205 	}
1206 	return err;
1207 }
1208 
1209 /*
1210  * ntfs_fiemap - file_operations::fiemap
1211  */
1212 int ntfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1213 		__u64 start, __u64 len)
1214 {
1215 	int err;
1216 	struct ntfs_inode *ni = ntfs_i(inode);
1217 
1218 	err = fiemap_prep(inode, fieinfo, start, &len, ~FIEMAP_FLAG_XATTR);
1219 	if (err)
1220 		return err;
1221 
1222 	ni_lock(ni);
1223 
1224 	err = ni_fiemap(ni, fieinfo, start, len);
1225 
1226 	ni_unlock(ni);
1227 
1228 	return err;
1229 }
1230 
1231 // clang-format off
1232 const struct inode_operations ntfs_file_inode_operations = {
1233 	.getattr	= ntfs_getattr,
1234 	.setattr	= ntfs3_setattr,
1235 	.listxattr	= ntfs_listxattr,
1236 	.permission	= ntfs_permission,
1237 	.get_acl	= ntfs_get_acl,
1238 	.set_acl	= ntfs_set_acl,
1239 	.fiemap		= ntfs_fiemap,
1240 };
1241 
1242 const struct file_operations ntfs_file_operations = {
1243 	.llseek		= generic_file_llseek,
1244 	.read_iter	= ntfs_file_read_iter,
1245 	.write_iter	= ntfs_file_write_iter,
1246 	.unlocked_ioctl = ntfs_ioctl,
1247 #ifdef CONFIG_COMPAT
1248 	.compat_ioctl	= ntfs_compat_ioctl,
1249 #endif
1250 	.splice_read	= generic_file_splice_read,
1251 	.mmap		= ntfs_file_mmap,
1252 	.open		= ntfs_file_open,
1253 	.fsync		= generic_file_fsync,
1254 	.splice_write	= iter_file_splice_write,
1255 	.fallocate	= ntfs_fallocate,
1256 	.release	= ntfs_file_release,
1257 };
1258 // clang-format on
1259