xref: /openbmc/linux/fs/ext4/inline.c (revision 2a6579ef)
1 // SPDX-License-Identifier: LGPL-2.1
2 /*
3  * Copyright (c) 2012 Taobao.
4  * Written by Tao Ma <boyu.mt@taobao.com>
5  */
6 
7 #include <linux/iomap.h>
8 #include <linux/fiemap.h>
9 #include <linux/namei.h>
10 #include <linux/iversion.h>
11 #include <linux/sched/mm.h>
12 
13 #include "ext4_jbd2.h"
14 #include "ext4.h"
15 #include "xattr.h"
16 #include "truncate.h"
17 
18 #define EXT4_XATTR_SYSTEM_DATA	"data"
19 #define EXT4_MIN_INLINE_DATA_SIZE	((sizeof(__le32) * EXT4_N_BLOCKS))
20 #define EXT4_INLINE_DOTDOT_OFFSET	2
21 #define EXT4_INLINE_DOTDOT_SIZE		4
22 
ext4_get_inline_size(struct inode * inode)23 static int ext4_get_inline_size(struct inode *inode)
24 {
25 	if (EXT4_I(inode)->i_inline_off)
26 		return EXT4_I(inode)->i_inline_size;
27 
28 	return 0;
29 }
30 
get_max_inline_xattr_value_size(struct inode * inode,struct ext4_iloc * iloc)31 static int get_max_inline_xattr_value_size(struct inode *inode,
32 					   struct ext4_iloc *iloc)
33 {
34 	struct ext4_xattr_ibody_header *header;
35 	struct ext4_xattr_entry *entry;
36 	struct ext4_inode *raw_inode;
37 	void *end;
38 	int free, min_offs;
39 
40 	if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
41 		return 0;
42 
43 	min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
44 			EXT4_GOOD_OLD_INODE_SIZE -
45 			EXT4_I(inode)->i_extra_isize -
46 			sizeof(struct ext4_xattr_ibody_header);
47 
48 	/*
49 	 * We need to subtract another sizeof(__u32) since an in-inode xattr
50 	 * needs an empty 4 bytes to indicate the gap between the xattr entry
51 	 * and the name/value pair.
52 	 */
53 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
54 		return EXT4_XATTR_SIZE(min_offs -
55 			EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
56 			EXT4_XATTR_ROUND - sizeof(__u32));
57 
58 	raw_inode = ext4_raw_inode(iloc);
59 	header = IHDR(inode, raw_inode);
60 	entry = IFIRST(header);
61 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
62 
63 	/* Compute min_offs. */
64 	while (!IS_LAST_ENTRY(entry)) {
65 		void *next = EXT4_XATTR_NEXT(entry);
66 
67 		if (next >= end) {
68 			EXT4_ERROR_INODE(inode,
69 					 "corrupt xattr in inline inode");
70 			return 0;
71 		}
72 		if (!entry->e_value_inum && entry->e_value_size) {
73 			size_t offs = le16_to_cpu(entry->e_value_offs);
74 			if (offs < min_offs)
75 				min_offs = offs;
76 		}
77 		entry = next;
78 	}
79 	free = min_offs -
80 		((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
81 
82 	if (EXT4_I(inode)->i_inline_off) {
83 		entry = (struct ext4_xattr_entry *)
84 			((void *)raw_inode + EXT4_I(inode)->i_inline_off);
85 
86 		free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
87 		goto out;
88 	}
89 
90 	free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
91 
92 	if (free > EXT4_XATTR_ROUND)
93 		free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
94 	else
95 		free = 0;
96 
97 out:
98 	return free;
99 }
100 
101 /*
102  * Get the maximum size we now can store in an inode.
103  * If we can't find the space for a xattr entry, don't use the space
104  * of the extents since we have no space to indicate the inline data.
105  */
ext4_get_max_inline_size(struct inode * inode)106 int ext4_get_max_inline_size(struct inode *inode)
107 {
108 	int error, max_inline_size;
109 	struct ext4_iloc iloc;
110 
111 	if (EXT4_I(inode)->i_extra_isize == 0)
112 		return 0;
113 
114 	error = ext4_get_inode_loc(inode, &iloc);
115 	if (error) {
116 		ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
117 				     "can't get inode location %lu",
118 				     inode->i_ino);
119 		return 0;
120 	}
121 
122 	down_read(&EXT4_I(inode)->xattr_sem);
123 	max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
124 	up_read(&EXT4_I(inode)->xattr_sem);
125 
126 	brelse(iloc.bh);
127 
128 	if (!max_inline_size)
129 		return 0;
130 
131 	return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
132 }
133 
134 /*
135  * this function does not take xattr_sem, which is OK because it is
136  * currently only used in a code path coming form ext4_iget, before
137  * the new inode has been unlocked
138  */
ext4_find_inline_data_nolock(struct inode * inode)139 int ext4_find_inline_data_nolock(struct inode *inode)
140 {
141 	struct ext4_xattr_ibody_find is = {
142 		.s = { .not_found = -ENODATA, },
143 	};
144 	struct ext4_xattr_info i = {
145 		.name_index = EXT4_XATTR_INDEX_SYSTEM,
146 		.name = EXT4_XATTR_SYSTEM_DATA,
147 	};
148 	int error;
149 
150 	if (EXT4_I(inode)->i_extra_isize == 0)
151 		return 0;
152 
153 	error = ext4_get_inode_loc(inode, &is.iloc);
154 	if (error)
155 		return error;
156 
157 	error = ext4_xattr_ibody_find(inode, &i, &is);
158 	if (error)
159 		goto out;
160 
161 	if (!is.s.not_found) {
162 		if (is.s.here->e_value_inum) {
163 			EXT4_ERROR_INODE(inode, "inline data xattr refers "
164 					 "to an external xattr inode");
165 			error = -EFSCORRUPTED;
166 			goto out;
167 		}
168 		EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
169 					(void *)ext4_raw_inode(&is.iloc));
170 		EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
171 				le32_to_cpu(is.s.here->e_value_size);
172 	}
173 out:
174 	brelse(is.iloc.bh);
175 	return error;
176 }
177 
ext4_read_inline_data(struct inode * inode,void * buffer,unsigned int len,struct ext4_iloc * iloc)178 static int ext4_read_inline_data(struct inode *inode, void *buffer,
179 				 unsigned int len,
180 				 struct ext4_iloc *iloc)
181 {
182 	struct ext4_xattr_entry *entry;
183 	struct ext4_xattr_ibody_header *header;
184 	int cp_len = 0;
185 	struct ext4_inode *raw_inode;
186 
187 	if (!len)
188 		return 0;
189 
190 	BUG_ON(len > EXT4_I(inode)->i_inline_size);
191 
192 	cp_len = min_t(unsigned int, len, EXT4_MIN_INLINE_DATA_SIZE);
193 
194 	raw_inode = ext4_raw_inode(iloc);
195 	memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
196 
197 	len -= cp_len;
198 	buffer += cp_len;
199 
200 	if (!len)
201 		goto out;
202 
203 	header = IHDR(inode, raw_inode);
204 	entry = (struct ext4_xattr_entry *)((void *)raw_inode +
205 					    EXT4_I(inode)->i_inline_off);
206 	len = min_t(unsigned int, len,
207 		    (unsigned int)le32_to_cpu(entry->e_value_size));
208 
209 	memcpy(buffer,
210 	       (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
211 	cp_len += len;
212 
213 out:
214 	return cp_len;
215 }
216 
217 /*
218  * write the buffer to the inline inode.
219  * If 'create' is set, we don't need to do the extra copy in the xattr
220  * value since it is already handled by ext4_xattr_ibody_set.
221  * That saves us one memcpy.
222  */
ext4_write_inline_data(struct inode * inode,struct ext4_iloc * iloc,void * buffer,loff_t pos,unsigned int len)223 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
224 				   void *buffer, loff_t pos, unsigned int len)
225 {
226 	struct ext4_xattr_entry *entry;
227 	struct ext4_xattr_ibody_header *header;
228 	struct ext4_inode *raw_inode;
229 	int cp_len = 0;
230 
231 	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
232 		return;
233 
234 	BUG_ON(!EXT4_I(inode)->i_inline_off);
235 	BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
236 
237 	raw_inode = ext4_raw_inode(iloc);
238 	buffer += pos;
239 
240 	if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
241 		cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
242 			 EXT4_MIN_INLINE_DATA_SIZE - pos : len;
243 		memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
244 
245 		len -= cp_len;
246 		buffer += cp_len;
247 		pos += cp_len;
248 	}
249 
250 	if (!len)
251 		return;
252 
253 	pos -= EXT4_MIN_INLINE_DATA_SIZE;
254 	header = IHDR(inode, raw_inode);
255 	entry = (struct ext4_xattr_entry *)((void *)raw_inode +
256 					    EXT4_I(inode)->i_inline_off);
257 
258 	memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
259 	       buffer, len);
260 }
261 
ext4_create_inline_data(handle_t * handle,struct inode * inode,unsigned len)262 static int ext4_create_inline_data(handle_t *handle,
263 				   struct inode *inode, unsigned len)
264 {
265 	int error;
266 	void *value = NULL;
267 	struct ext4_xattr_ibody_find is = {
268 		.s = { .not_found = -ENODATA, },
269 	};
270 	struct ext4_xattr_info i = {
271 		.name_index = EXT4_XATTR_INDEX_SYSTEM,
272 		.name = EXT4_XATTR_SYSTEM_DATA,
273 	};
274 
275 	error = ext4_get_inode_loc(inode, &is.iloc);
276 	if (error)
277 		return error;
278 
279 	BUFFER_TRACE(is.iloc.bh, "get_write_access");
280 	error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
281 					      EXT4_JTR_NONE);
282 	if (error)
283 		goto out;
284 
285 	if (len > EXT4_MIN_INLINE_DATA_SIZE) {
286 		value = EXT4_ZERO_XATTR_VALUE;
287 		len -= EXT4_MIN_INLINE_DATA_SIZE;
288 	} else {
289 		value = "";
290 		len = 0;
291 	}
292 
293 	/* Insert the xttr entry. */
294 	i.value = value;
295 	i.value_len = len;
296 
297 	error = ext4_xattr_ibody_find(inode, &i, &is);
298 	if (error)
299 		goto out;
300 
301 	BUG_ON(!is.s.not_found);
302 
303 	error = ext4_xattr_ibody_set(handle, inode, &i, &is);
304 	if (error) {
305 		if (error == -ENOSPC)
306 			ext4_clear_inode_state(inode,
307 					       EXT4_STATE_MAY_INLINE_DATA);
308 		goto out;
309 	}
310 
311 	memset((void *)ext4_raw_inode(&is.iloc)->i_block,
312 		0, EXT4_MIN_INLINE_DATA_SIZE);
313 
314 	EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
315 				      (void *)ext4_raw_inode(&is.iloc));
316 	EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
317 	ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
318 	ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
319 	get_bh(is.iloc.bh);
320 	error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
321 
322 out:
323 	brelse(is.iloc.bh);
324 	return error;
325 }
326 
ext4_update_inline_data(handle_t * handle,struct inode * inode,unsigned int len)327 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
328 				   unsigned int len)
329 {
330 	int error;
331 	void *value = NULL;
332 	struct ext4_xattr_ibody_find is = {
333 		.s = { .not_found = -ENODATA, },
334 	};
335 	struct ext4_xattr_info i = {
336 		.name_index = EXT4_XATTR_INDEX_SYSTEM,
337 		.name = EXT4_XATTR_SYSTEM_DATA,
338 	};
339 
340 	/* If the old space is ok, write the data directly. */
341 	if (len <= EXT4_I(inode)->i_inline_size)
342 		return 0;
343 
344 	error = ext4_get_inode_loc(inode, &is.iloc);
345 	if (error)
346 		return error;
347 
348 	error = ext4_xattr_ibody_find(inode, &i, &is);
349 	if (error)
350 		goto out;
351 
352 	BUG_ON(is.s.not_found);
353 
354 	len -= EXT4_MIN_INLINE_DATA_SIZE;
355 	value = kzalloc(len, GFP_NOFS);
356 	if (!value) {
357 		error = -ENOMEM;
358 		goto out;
359 	}
360 
361 	error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
362 				     value, len);
363 	if (error < 0)
364 		goto out;
365 
366 	BUFFER_TRACE(is.iloc.bh, "get_write_access");
367 	error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
368 					      EXT4_JTR_NONE);
369 	if (error)
370 		goto out;
371 
372 	/* Update the xattr entry. */
373 	i.value = value;
374 	i.value_len = len;
375 
376 	error = ext4_xattr_ibody_set(handle, inode, &i, &is);
377 	if (error)
378 		goto out;
379 
380 	EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
381 				      (void *)ext4_raw_inode(&is.iloc));
382 	EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
383 				le32_to_cpu(is.s.here->e_value_size);
384 	ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
385 	get_bh(is.iloc.bh);
386 	error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
387 
388 out:
389 	kfree(value);
390 	brelse(is.iloc.bh);
391 	return error;
392 }
393 
ext4_prepare_inline_data(handle_t * handle,struct inode * inode,unsigned int len)394 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
395 				    unsigned int len)
396 {
397 	int ret, size, no_expand;
398 	struct ext4_inode_info *ei = EXT4_I(inode);
399 
400 	if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
401 		return -ENOSPC;
402 
403 	size = ext4_get_max_inline_size(inode);
404 	if (size < len)
405 		return -ENOSPC;
406 
407 	ext4_write_lock_xattr(inode, &no_expand);
408 
409 	if (ei->i_inline_off)
410 		ret = ext4_update_inline_data(handle, inode, len);
411 	else
412 		ret = ext4_create_inline_data(handle, inode, len);
413 
414 	ext4_write_unlock_xattr(inode, &no_expand);
415 	return ret;
416 }
417 
ext4_destroy_inline_data_nolock(handle_t * handle,struct inode * inode)418 static int ext4_destroy_inline_data_nolock(handle_t *handle,
419 					   struct inode *inode)
420 {
421 	struct ext4_inode_info *ei = EXT4_I(inode);
422 	struct ext4_xattr_ibody_find is = {
423 		.s = { .not_found = 0, },
424 	};
425 	struct ext4_xattr_info i = {
426 		.name_index = EXT4_XATTR_INDEX_SYSTEM,
427 		.name = EXT4_XATTR_SYSTEM_DATA,
428 		.value = NULL,
429 		.value_len = 0,
430 	};
431 	int error;
432 
433 	if (!ei->i_inline_off)
434 		return 0;
435 
436 	error = ext4_get_inode_loc(inode, &is.iloc);
437 	if (error)
438 		return error;
439 
440 	error = ext4_xattr_ibody_find(inode, &i, &is);
441 	if (error)
442 		goto out;
443 
444 	BUFFER_TRACE(is.iloc.bh, "get_write_access");
445 	error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
446 					      EXT4_JTR_NONE);
447 	if (error)
448 		goto out;
449 
450 	error = ext4_xattr_ibody_set(handle, inode, &i, &is);
451 	if (error)
452 		goto out;
453 
454 	memset((void *)ext4_raw_inode(&is.iloc)->i_block,
455 		0, EXT4_MIN_INLINE_DATA_SIZE);
456 	memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
457 
458 	if (ext4_has_feature_extents(inode->i_sb)) {
459 		if (S_ISDIR(inode->i_mode) ||
460 		    S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
461 			ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
462 			ext4_ext_tree_init(handle, inode);
463 		}
464 	}
465 	ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
466 
467 	get_bh(is.iloc.bh);
468 	error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
469 
470 	EXT4_I(inode)->i_inline_off = 0;
471 	EXT4_I(inode)->i_inline_size = 0;
472 	ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
473 out:
474 	brelse(is.iloc.bh);
475 	if (error == -ENODATA)
476 		error = 0;
477 	return error;
478 }
479 
ext4_read_inline_folio(struct inode * inode,struct folio * folio)480 static int ext4_read_inline_folio(struct inode *inode, struct folio *folio)
481 {
482 	void *kaddr;
483 	int ret = 0;
484 	size_t len;
485 	struct ext4_iloc iloc;
486 
487 	BUG_ON(!folio_test_locked(folio));
488 	BUG_ON(!ext4_has_inline_data(inode));
489 	BUG_ON(folio->index);
490 
491 	if (!EXT4_I(inode)->i_inline_off) {
492 		ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
493 			     inode->i_ino);
494 		goto out;
495 	}
496 
497 	ret = ext4_get_inode_loc(inode, &iloc);
498 	if (ret)
499 		goto out;
500 
501 	len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
502 	BUG_ON(len > PAGE_SIZE);
503 	kaddr = kmap_local_folio(folio, 0);
504 	ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
505 	flush_dcache_folio(folio);
506 	kunmap_local(kaddr);
507 	folio_zero_segment(folio, len, folio_size(folio));
508 	folio_mark_uptodate(folio);
509 	brelse(iloc.bh);
510 
511 out:
512 	return ret;
513 }
514 
ext4_readpage_inline(struct inode * inode,struct folio * folio)515 int ext4_readpage_inline(struct inode *inode, struct folio *folio)
516 {
517 	int ret = 0;
518 
519 	down_read(&EXT4_I(inode)->xattr_sem);
520 	if (!ext4_has_inline_data(inode)) {
521 		up_read(&EXT4_I(inode)->xattr_sem);
522 		return -EAGAIN;
523 	}
524 
525 	/*
526 	 * Current inline data can only exist in the 1st page,
527 	 * So for all the other pages, just set them uptodate.
528 	 */
529 	if (!folio->index)
530 		ret = ext4_read_inline_folio(inode, folio);
531 	else if (!folio_test_uptodate(folio)) {
532 		folio_zero_segment(folio, 0, folio_size(folio));
533 		folio_mark_uptodate(folio);
534 	}
535 
536 	up_read(&EXT4_I(inode)->xattr_sem);
537 
538 	folio_unlock(folio);
539 	return ret >= 0 ? 0 : ret;
540 }
541 
ext4_convert_inline_data_to_extent(struct address_space * mapping,struct inode * inode)542 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
543 					      struct inode *inode)
544 {
545 	int ret, needed_blocks, no_expand;
546 	handle_t *handle = NULL;
547 	int retries = 0, sem_held = 0;
548 	struct folio *folio = NULL;
549 	unsigned from, to;
550 	struct ext4_iloc iloc;
551 
552 	if (!ext4_has_inline_data(inode)) {
553 		/*
554 		 * clear the flag so that no new write
555 		 * will trap here again.
556 		 */
557 		ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
558 		return 0;
559 	}
560 
561 	needed_blocks = ext4_writepage_trans_blocks(inode);
562 
563 	ret = ext4_get_inode_loc(inode, &iloc);
564 	if (ret)
565 		return ret;
566 
567 retry:
568 	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
569 	if (IS_ERR(handle)) {
570 		ret = PTR_ERR(handle);
571 		handle = NULL;
572 		goto out;
573 	}
574 
575 	/* We cannot recurse into the filesystem as the transaction is already
576 	 * started */
577 	folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
578 			mapping_gfp_mask(mapping));
579 	if (IS_ERR(folio)) {
580 		ret = PTR_ERR(folio);
581 		goto out_nofolio;
582 	}
583 
584 	ext4_write_lock_xattr(inode, &no_expand);
585 	sem_held = 1;
586 	/* If some one has already done this for us, just exit. */
587 	if (!ext4_has_inline_data(inode)) {
588 		ret = 0;
589 		goto out;
590 	}
591 
592 	from = 0;
593 	to = ext4_get_inline_size(inode);
594 	if (!folio_test_uptodate(folio)) {
595 		ret = ext4_read_inline_folio(inode, folio);
596 		if (ret < 0)
597 			goto out;
598 	}
599 
600 	ret = ext4_destroy_inline_data_nolock(handle, inode);
601 	if (ret)
602 		goto out;
603 
604 	if (ext4_should_dioread_nolock(inode)) {
605 		ret = __block_write_begin(&folio->page, from, to,
606 					  ext4_get_block_unwritten);
607 	} else
608 		ret = __block_write_begin(&folio->page, from, to, ext4_get_block);
609 
610 	if (!ret && ext4_should_journal_data(inode)) {
611 		ret = ext4_walk_page_buffers(handle, inode,
612 					     folio_buffers(folio), from, to,
613 					     NULL, do_journal_get_write_access);
614 	}
615 
616 	if (ret) {
617 		folio_unlock(folio);
618 		folio_put(folio);
619 		folio = NULL;
620 		ext4_orphan_add(handle, inode);
621 		ext4_write_unlock_xattr(inode, &no_expand);
622 		sem_held = 0;
623 		ext4_journal_stop(handle);
624 		handle = NULL;
625 		ext4_truncate_failed_write(inode);
626 		/*
627 		 * If truncate failed early the inode might
628 		 * still be on the orphan list; we need to
629 		 * make sure the inode is removed from the
630 		 * orphan list in that case.
631 		 */
632 		if (inode->i_nlink)
633 			ext4_orphan_del(NULL, inode);
634 	}
635 
636 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
637 		goto retry;
638 
639 	if (folio)
640 		block_commit_write(&folio->page, from, to);
641 out:
642 	if (folio) {
643 		folio_unlock(folio);
644 		folio_put(folio);
645 	}
646 out_nofolio:
647 	if (sem_held)
648 		ext4_write_unlock_xattr(inode, &no_expand);
649 	if (handle)
650 		ext4_journal_stop(handle);
651 	brelse(iloc.bh);
652 	return ret;
653 }
654 
655 /*
656  * Try to write data in the inode.
657  * If the inode has inline data, check whether the new write can be
658  * in the inode also. If not, create the page the handle, move the data
659  * to the page make it update and let the later codes create extent for it.
660  */
ext4_try_to_write_inline_data(struct address_space * mapping,struct inode * inode,loff_t pos,unsigned len,struct page ** pagep)661 int ext4_try_to_write_inline_data(struct address_space *mapping,
662 				  struct inode *inode,
663 				  loff_t pos, unsigned len,
664 				  struct page **pagep)
665 {
666 	int ret;
667 	handle_t *handle;
668 	struct folio *folio;
669 	struct ext4_iloc iloc;
670 
671 	if (pos + len > ext4_get_max_inline_size(inode))
672 		goto convert;
673 
674 	ret = ext4_get_inode_loc(inode, &iloc);
675 	if (ret)
676 		return ret;
677 
678 	/*
679 	 * The possible write could happen in the inode,
680 	 * so try to reserve the space in inode first.
681 	 */
682 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
683 	if (IS_ERR(handle)) {
684 		ret = PTR_ERR(handle);
685 		handle = NULL;
686 		goto out;
687 	}
688 
689 	ret = ext4_prepare_inline_data(handle, inode, pos + len);
690 	if (ret && ret != -ENOSPC)
691 		goto out;
692 
693 	/* We don't have space in inline inode, so convert it to extent. */
694 	if (ret == -ENOSPC) {
695 		ext4_journal_stop(handle);
696 		brelse(iloc.bh);
697 		goto convert;
698 	}
699 
700 	ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
701 					    EXT4_JTR_NONE);
702 	if (ret)
703 		goto out;
704 
705 	folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
706 					mapping_gfp_mask(mapping));
707 	if (IS_ERR(folio)) {
708 		ret = PTR_ERR(folio);
709 		goto out;
710 	}
711 
712 	*pagep = &folio->page;
713 	down_read(&EXT4_I(inode)->xattr_sem);
714 	if (!ext4_has_inline_data(inode)) {
715 		ret = 0;
716 		folio_unlock(folio);
717 		folio_put(folio);
718 		goto out_up_read;
719 	}
720 
721 	if (!folio_test_uptodate(folio)) {
722 		ret = ext4_read_inline_folio(inode, folio);
723 		if (ret < 0) {
724 			folio_unlock(folio);
725 			folio_put(folio);
726 			goto out_up_read;
727 		}
728 	}
729 
730 	ret = 1;
731 	handle = NULL;
732 out_up_read:
733 	up_read(&EXT4_I(inode)->xattr_sem);
734 out:
735 	if (handle && (ret != 1))
736 		ext4_journal_stop(handle);
737 	brelse(iloc.bh);
738 	return ret;
739 convert:
740 	return ext4_convert_inline_data_to_extent(mapping, inode);
741 }
742 
ext4_write_inline_data_end(struct inode * inode,loff_t pos,unsigned len,unsigned copied,struct folio * folio)743 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
744 			       unsigned copied, struct folio *folio)
745 {
746 	handle_t *handle = ext4_journal_current_handle();
747 	int no_expand;
748 	void *kaddr;
749 	struct ext4_iloc iloc;
750 	int ret = 0, ret2;
751 
752 	if (unlikely(copied < len) && !folio_test_uptodate(folio))
753 		copied = 0;
754 
755 	if (likely(copied)) {
756 		ret = ext4_get_inode_loc(inode, &iloc);
757 		if (ret) {
758 			folio_unlock(folio);
759 			folio_put(folio);
760 			ext4_std_error(inode->i_sb, ret);
761 			goto out;
762 		}
763 		ext4_write_lock_xattr(inode, &no_expand);
764 		BUG_ON(!ext4_has_inline_data(inode));
765 
766 		/*
767 		 * ei->i_inline_off may have changed since
768 		 * ext4_write_begin() called
769 		 * ext4_try_to_write_inline_data()
770 		 */
771 		(void) ext4_find_inline_data_nolock(inode);
772 
773 		kaddr = kmap_local_folio(folio, 0);
774 		ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
775 		kunmap_local(kaddr);
776 		folio_mark_uptodate(folio);
777 		/* clear dirty flag so that writepages wouldn't work for us. */
778 		folio_clear_dirty(folio);
779 
780 		ext4_write_unlock_xattr(inode, &no_expand);
781 		brelse(iloc.bh);
782 
783 		/*
784 		 * It's important to update i_size while still holding folio
785 		 * lock: page writeout could otherwise come in and zero
786 		 * beyond i_size.
787 		 */
788 		ext4_update_inode_size(inode, pos + copied);
789 	}
790 	folio_unlock(folio);
791 	folio_put(folio);
792 
793 	/*
794 	 * Don't mark the inode dirty under folio lock. First, it unnecessarily
795 	 * makes the holding time of folio lock longer. Second, it forces lock
796 	 * ordering of folio lock and transaction start for journaling
797 	 * filesystems.
798 	 */
799 	if (likely(copied))
800 		mark_inode_dirty(inode);
801 out:
802 	/*
803 	 * If we didn't copy as much data as expected, we need to trim back
804 	 * size of xattr containing inline data.
805 	 */
806 	if (pos + len > inode->i_size && ext4_can_truncate(inode))
807 		ext4_orphan_add(handle, inode);
808 
809 	ret2 = ext4_journal_stop(handle);
810 	if (!ret)
811 		ret = ret2;
812 	if (pos + len > inode->i_size) {
813 		ext4_truncate_failed_write(inode);
814 		/*
815 		 * If truncate failed early the inode might still be
816 		 * on the orphan list; we need to make sure the inode
817 		 * is removed from the orphan list in that case.
818 		 */
819 		if (inode->i_nlink)
820 			ext4_orphan_del(NULL, inode);
821 	}
822 	return ret ? ret : copied;
823 }
824 
825 /*
826  * Try to make the page cache and handle ready for the inline data case.
827  * We can call this function in 2 cases:
828  * 1. The inode is created and the first write exceeds inline size. We can
829  *    clear the inode state safely.
830  * 2. The inode has inline data, then we need to read the data, make it
831  *    update and dirty so that ext4_da_writepages can handle it. We don't
832  *    need to start the journal since the file's metadata isn't changed now.
833  */
ext4_da_convert_inline_data_to_extent(struct address_space * mapping,struct inode * inode,void ** fsdata)834 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
835 						 struct inode *inode,
836 						 void **fsdata)
837 {
838 	int ret = 0, inline_size;
839 	struct folio *folio;
840 
841 	folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN,
842 					mapping_gfp_mask(mapping));
843 	if (IS_ERR(folio))
844 		return PTR_ERR(folio);
845 
846 	down_read(&EXT4_I(inode)->xattr_sem);
847 	if (!ext4_has_inline_data(inode)) {
848 		ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
849 		goto out;
850 	}
851 
852 	inline_size = ext4_get_inline_size(inode);
853 
854 	if (!folio_test_uptodate(folio)) {
855 		ret = ext4_read_inline_folio(inode, folio);
856 		if (ret < 0)
857 			goto out;
858 	}
859 
860 	ret = __block_write_begin(&folio->page, 0, inline_size,
861 				  ext4_da_get_block_prep);
862 	if (ret) {
863 		up_read(&EXT4_I(inode)->xattr_sem);
864 		folio_unlock(folio);
865 		folio_put(folio);
866 		ext4_truncate_failed_write(inode);
867 		return ret;
868 	}
869 
870 	folio_mark_dirty(folio);
871 	folio_mark_uptodate(folio);
872 	ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
873 	*fsdata = (void *)CONVERT_INLINE_DATA;
874 
875 out:
876 	up_read(&EXT4_I(inode)->xattr_sem);
877 	if (folio) {
878 		folio_unlock(folio);
879 		folio_put(folio);
880 	}
881 	return ret;
882 }
883 
884 /*
885  * Prepare the write for the inline data.
886  * If the data can be written into the inode, we just read
887  * the page and make it uptodate, and start the journal.
888  * Otherwise read the page, makes it dirty so that it can be
889  * handle in writepages(the i_disksize update is left to the
890  * normal ext4_da_write_end).
891  */
ext4_da_write_inline_data_begin(struct address_space * mapping,struct inode * inode,loff_t pos,unsigned len,struct page ** pagep,void ** fsdata)892 int ext4_da_write_inline_data_begin(struct address_space *mapping,
893 				    struct inode *inode,
894 				    loff_t pos, unsigned len,
895 				    struct page **pagep,
896 				    void **fsdata)
897 {
898 	int ret;
899 	handle_t *handle;
900 	struct folio *folio;
901 	struct ext4_iloc iloc;
902 	int retries = 0;
903 
904 	ret = ext4_get_inode_loc(inode, &iloc);
905 	if (ret)
906 		return ret;
907 
908 retry_journal:
909 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
910 	if (IS_ERR(handle)) {
911 		ret = PTR_ERR(handle);
912 		goto out;
913 	}
914 
915 	ret = ext4_prepare_inline_data(handle, inode, pos + len);
916 	if (ret && ret != -ENOSPC)
917 		goto out_journal;
918 
919 	if (ret == -ENOSPC) {
920 		ext4_journal_stop(handle);
921 		ret = ext4_da_convert_inline_data_to_extent(mapping,
922 							    inode,
923 							    fsdata);
924 		if (ret == -ENOSPC &&
925 		    ext4_should_retry_alloc(inode->i_sb, &retries))
926 			goto retry_journal;
927 		goto out;
928 	}
929 
930 	/*
931 	 * We cannot recurse into the filesystem as the transaction
932 	 * is already started.
933 	 */
934 	folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
935 					mapping_gfp_mask(mapping));
936 	if (IS_ERR(folio)) {
937 		ret = PTR_ERR(folio);
938 		goto out_journal;
939 	}
940 
941 	down_read(&EXT4_I(inode)->xattr_sem);
942 	if (!ext4_has_inline_data(inode)) {
943 		ret = 0;
944 		goto out_release_page;
945 	}
946 
947 	if (!folio_test_uptodate(folio)) {
948 		ret = ext4_read_inline_folio(inode, folio);
949 		if (ret < 0)
950 			goto out_release_page;
951 	}
952 	ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
953 					    EXT4_JTR_NONE);
954 	if (ret)
955 		goto out_release_page;
956 
957 	up_read(&EXT4_I(inode)->xattr_sem);
958 	*pagep = &folio->page;
959 	brelse(iloc.bh);
960 	return 1;
961 out_release_page:
962 	up_read(&EXT4_I(inode)->xattr_sem);
963 	folio_unlock(folio);
964 	folio_put(folio);
965 out_journal:
966 	ext4_journal_stop(handle);
967 out:
968 	brelse(iloc.bh);
969 	return ret;
970 }
971 
972 #ifdef INLINE_DIR_DEBUG
ext4_show_inline_dir(struct inode * dir,struct buffer_head * bh,void * inline_start,int inline_size)973 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
974 			  void *inline_start, int inline_size)
975 {
976 	int offset;
977 	unsigned short de_len;
978 	struct ext4_dir_entry_2 *de = inline_start;
979 	void *dlimit = inline_start + inline_size;
980 
981 	trace_printk("inode %lu\n", dir->i_ino);
982 	offset = 0;
983 	while ((void *)de < dlimit) {
984 		de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
985 		trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
986 			     offset, de_len, de->name_len, de->name,
987 			     de->name_len, le32_to_cpu(de->inode));
988 		if (ext4_check_dir_entry(dir, NULL, de, bh,
989 					 inline_start, inline_size, offset))
990 			BUG();
991 
992 		offset += de_len;
993 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
994 	}
995 }
996 #else
997 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
998 #endif
999 
1000 /*
1001  * Add a new entry into a inline dir.
1002  * It will return -ENOSPC if no space is available, and -EIO
1003  * and -EEXIST if directory entry already exists.
1004  */
ext4_add_dirent_to_inline(handle_t * handle,struct ext4_filename * fname,struct inode * dir,struct inode * inode,struct ext4_iloc * iloc,void * inline_start,int inline_size)1005 static int ext4_add_dirent_to_inline(handle_t *handle,
1006 				     struct ext4_filename *fname,
1007 				     struct inode *dir,
1008 				     struct inode *inode,
1009 				     struct ext4_iloc *iloc,
1010 				     void *inline_start, int inline_size)
1011 {
1012 	int		err;
1013 	struct ext4_dir_entry_2 *de;
1014 
1015 	err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1016 				inline_size, fname, &de);
1017 	if (err)
1018 		return err;
1019 
1020 	BUFFER_TRACE(iloc->bh, "get_write_access");
1021 	err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1022 					    EXT4_JTR_NONE);
1023 	if (err)
1024 		return err;
1025 	ext4_insert_dentry(dir, inode, de, inline_size, fname);
1026 
1027 	ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1028 
1029 	/*
1030 	 * XXX shouldn't update any times until successful
1031 	 * completion of syscall, but too many callers depend
1032 	 * on this.
1033 	 *
1034 	 * XXX similarly, too many callers depend on
1035 	 * ext4_new_inode() setting the times, but error
1036 	 * recovery deletes the inode, so the worst that can
1037 	 * happen is that the times are slightly out of date
1038 	 * and/or different from the directory change time.
1039 	 */
1040 	dir->i_mtime = inode_set_ctime_current(dir);
1041 	ext4_update_dx_flag(dir);
1042 	inode_inc_iversion(dir);
1043 	return 1;
1044 }
1045 
ext4_get_inline_xattr_pos(struct inode * inode,struct ext4_iloc * iloc)1046 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1047 				       struct ext4_iloc *iloc)
1048 {
1049 	struct ext4_xattr_entry *entry;
1050 	struct ext4_xattr_ibody_header *header;
1051 
1052 	BUG_ON(!EXT4_I(inode)->i_inline_off);
1053 
1054 	header = IHDR(inode, ext4_raw_inode(iloc));
1055 	entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1056 					    EXT4_I(inode)->i_inline_off);
1057 
1058 	return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1059 }
1060 
1061 /* Set the final de to cover the whole block. */
ext4_update_final_de(void * de_buf,int old_size,int new_size)1062 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1063 {
1064 	struct ext4_dir_entry_2 *de, *prev_de;
1065 	void *limit;
1066 	int de_len;
1067 
1068 	de = de_buf;
1069 	if (old_size) {
1070 		limit = de_buf + old_size;
1071 		do {
1072 			prev_de = de;
1073 			de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1074 			de_buf += de_len;
1075 			de = de_buf;
1076 		} while (de_buf < limit);
1077 
1078 		prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1079 							old_size, new_size);
1080 	} else {
1081 		/* this is just created, so create an empty entry. */
1082 		de->inode = 0;
1083 		de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1084 	}
1085 }
1086 
ext4_update_inline_dir(handle_t * handle,struct inode * dir,struct ext4_iloc * iloc)1087 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1088 				  struct ext4_iloc *iloc)
1089 {
1090 	int ret;
1091 	int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1092 	int new_size = get_max_inline_xattr_value_size(dir, iloc);
1093 
1094 	if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1095 		return -ENOSPC;
1096 
1097 	ret = ext4_update_inline_data(handle, dir,
1098 				      new_size + EXT4_MIN_INLINE_DATA_SIZE);
1099 	if (ret)
1100 		return ret;
1101 
1102 	ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1103 			     EXT4_I(dir)->i_inline_size -
1104 						EXT4_MIN_INLINE_DATA_SIZE);
1105 	dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1106 	return 0;
1107 }
1108 
ext4_restore_inline_data(handle_t * handle,struct inode * inode,struct ext4_iloc * iloc,void * buf,int inline_size)1109 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1110 				     struct ext4_iloc *iloc,
1111 				     void *buf, int inline_size)
1112 {
1113 	int ret;
1114 
1115 	ret = ext4_create_inline_data(handle, inode, inline_size);
1116 	if (ret) {
1117 		ext4_msg(inode->i_sb, KERN_EMERG,
1118 			"error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1119 			inode->i_ino, ret);
1120 		return;
1121 	}
1122 	ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1123 	ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1124 }
1125 
ext4_finish_convert_inline_dir(handle_t * handle,struct inode * inode,struct buffer_head * dir_block,void * buf,int inline_size)1126 static int ext4_finish_convert_inline_dir(handle_t *handle,
1127 					  struct inode *inode,
1128 					  struct buffer_head *dir_block,
1129 					  void *buf,
1130 					  int inline_size)
1131 {
1132 	int err, csum_size = 0, header_size = 0;
1133 	struct ext4_dir_entry_2 *de;
1134 	void *target = dir_block->b_data;
1135 
1136 	/*
1137 	 * First create "." and ".." and then copy the dir information
1138 	 * back to the block.
1139 	 */
1140 	de = target;
1141 	de = ext4_init_dot_dotdot(inode, de,
1142 		inode->i_sb->s_blocksize, csum_size,
1143 		le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1144 	header_size = (void *)de - target;
1145 
1146 	memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1147 		inline_size - EXT4_INLINE_DOTDOT_SIZE);
1148 
1149 	if (ext4_has_metadata_csum(inode->i_sb))
1150 		csum_size = sizeof(struct ext4_dir_entry_tail);
1151 
1152 	inode->i_size = inode->i_sb->s_blocksize;
1153 	i_size_write(inode, inode->i_sb->s_blocksize);
1154 	EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1155 	ext4_update_final_de(dir_block->b_data,
1156 			inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1157 			inode->i_sb->s_blocksize - csum_size);
1158 
1159 	if (csum_size)
1160 		ext4_initialize_dirent_tail(dir_block,
1161 					    inode->i_sb->s_blocksize);
1162 	set_buffer_uptodate(dir_block);
1163 	unlock_buffer(dir_block);
1164 	err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1165 	if (err)
1166 		return err;
1167 	set_buffer_verified(dir_block);
1168 	return ext4_mark_inode_dirty(handle, inode);
1169 }
1170 
ext4_convert_inline_data_nolock(handle_t * handle,struct inode * inode,struct ext4_iloc * iloc)1171 static int ext4_convert_inline_data_nolock(handle_t *handle,
1172 					   struct inode *inode,
1173 					   struct ext4_iloc *iloc)
1174 {
1175 	int error;
1176 	void *buf = NULL;
1177 	struct buffer_head *data_bh = NULL;
1178 	struct ext4_map_blocks map;
1179 	int inline_size;
1180 
1181 	inline_size = ext4_get_inline_size(inode);
1182 	buf = kmalloc(inline_size, GFP_NOFS);
1183 	if (!buf) {
1184 		error = -ENOMEM;
1185 		goto out;
1186 	}
1187 
1188 	error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1189 	if (error < 0)
1190 		goto out;
1191 
1192 	/*
1193 	 * Make sure the inline directory entries pass checks before we try to
1194 	 * convert them, so that we avoid touching stuff that needs fsck.
1195 	 */
1196 	if (S_ISDIR(inode->i_mode)) {
1197 		error = ext4_check_all_de(inode, iloc->bh,
1198 					buf + EXT4_INLINE_DOTDOT_SIZE,
1199 					inline_size - EXT4_INLINE_DOTDOT_SIZE);
1200 		if (error)
1201 			goto out;
1202 	}
1203 
1204 	error = ext4_destroy_inline_data_nolock(handle, inode);
1205 	if (error)
1206 		goto out;
1207 
1208 	map.m_lblk = 0;
1209 	map.m_len = 1;
1210 	map.m_flags = 0;
1211 	error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1212 	if (error < 0)
1213 		goto out_restore;
1214 	if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1215 		error = -EIO;
1216 		goto out_restore;
1217 	}
1218 
1219 	data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1220 	if (!data_bh) {
1221 		error = -ENOMEM;
1222 		goto out_restore;
1223 	}
1224 
1225 	lock_buffer(data_bh);
1226 	error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1227 					       EXT4_JTR_NONE);
1228 	if (error) {
1229 		unlock_buffer(data_bh);
1230 		error = -EIO;
1231 		goto out_restore;
1232 	}
1233 	memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1234 
1235 	if (!S_ISDIR(inode->i_mode)) {
1236 		memcpy(data_bh->b_data, buf, inline_size);
1237 		set_buffer_uptodate(data_bh);
1238 		unlock_buffer(data_bh);
1239 		error = ext4_handle_dirty_metadata(handle,
1240 						   inode, data_bh);
1241 	} else {
1242 		error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1243 						       buf, inline_size);
1244 	}
1245 
1246 out_restore:
1247 	if (error)
1248 		ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1249 
1250 out:
1251 	brelse(data_bh);
1252 	kfree(buf);
1253 	return error;
1254 }
1255 
1256 /*
1257  * Try to add the new entry to the inline data.
1258  * If succeeds, return 0. If not, extended the inline dir and copied data to
1259  * the new created block.
1260  */
ext4_try_add_inline_entry(handle_t * handle,struct ext4_filename * fname,struct inode * dir,struct inode * inode)1261 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1262 			      struct inode *dir, struct inode *inode)
1263 {
1264 	int ret, ret2, inline_size, no_expand;
1265 	void *inline_start;
1266 	struct ext4_iloc iloc;
1267 
1268 	ret = ext4_get_inode_loc(dir, &iloc);
1269 	if (ret)
1270 		return ret;
1271 
1272 	ext4_write_lock_xattr(dir, &no_expand);
1273 	if (!ext4_has_inline_data(dir))
1274 		goto out;
1275 
1276 	inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1277 						 EXT4_INLINE_DOTDOT_SIZE;
1278 	inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1279 
1280 	ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1281 					inline_start, inline_size);
1282 	if (ret != -ENOSPC)
1283 		goto out;
1284 
1285 	/* check whether it can be inserted to inline xattr space. */
1286 	inline_size = EXT4_I(dir)->i_inline_size -
1287 			EXT4_MIN_INLINE_DATA_SIZE;
1288 	if (!inline_size) {
1289 		/* Try to use the xattr space.*/
1290 		ret = ext4_update_inline_dir(handle, dir, &iloc);
1291 		if (ret && ret != -ENOSPC)
1292 			goto out;
1293 
1294 		inline_size = EXT4_I(dir)->i_inline_size -
1295 				EXT4_MIN_INLINE_DATA_SIZE;
1296 	}
1297 
1298 	if (inline_size) {
1299 		inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1300 
1301 		ret = ext4_add_dirent_to_inline(handle, fname, dir,
1302 						inode, &iloc, inline_start,
1303 						inline_size);
1304 
1305 		if (ret != -ENOSPC)
1306 			goto out;
1307 	}
1308 
1309 	/*
1310 	 * The inline space is filled up, so create a new block for it.
1311 	 * As the extent tree will be created, we have to save the inline
1312 	 * dir first.
1313 	 */
1314 	ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1315 
1316 out:
1317 	ext4_write_unlock_xattr(dir, &no_expand);
1318 	ret2 = ext4_mark_inode_dirty(handle, dir);
1319 	if (unlikely(ret2 && !ret))
1320 		ret = ret2;
1321 	brelse(iloc.bh);
1322 	return ret;
1323 }
1324 
1325 /*
1326  * This function fills a red-black tree with information from an
1327  * inlined dir.  It returns the number directory entries loaded
1328  * into the tree.  If there is an error it is returned in err.
1329  */
ext4_inlinedir_to_tree(struct file * dir_file,struct inode * dir,ext4_lblk_t block,struct dx_hash_info * hinfo,__u32 start_hash,__u32 start_minor_hash,int * has_inline_data)1330 int ext4_inlinedir_to_tree(struct file *dir_file,
1331 			   struct inode *dir, ext4_lblk_t block,
1332 			   struct dx_hash_info *hinfo,
1333 			   __u32 start_hash, __u32 start_minor_hash,
1334 			   int *has_inline_data)
1335 {
1336 	int err = 0, count = 0;
1337 	unsigned int parent_ino;
1338 	int pos;
1339 	struct ext4_dir_entry_2 *de;
1340 	struct inode *inode = file_inode(dir_file);
1341 	int ret, inline_size = 0;
1342 	struct ext4_iloc iloc;
1343 	void *dir_buf = NULL;
1344 	struct ext4_dir_entry_2 fake;
1345 	struct fscrypt_str tmp_str;
1346 
1347 	ret = ext4_get_inode_loc(inode, &iloc);
1348 	if (ret)
1349 		return ret;
1350 
1351 	down_read(&EXT4_I(inode)->xattr_sem);
1352 	if (!ext4_has_inline_data(inode)) {
1353 		up_read(&EXT4_I(inode)->xattr_sem);
1354 		*has_inline_data = 0;
1355 		goto out;
1356 	}
1357 
1358 	inline_size = ext4_get_inline_size(inode);
1359 	dir_buf = kmalloc(inline_size, GFP_NOFS);
1360 	if (!dir_buf) {
1361 		ret = -ENOMEM;
1362 		up_read(&EXT4_I(inode)->xattr_sem);
1363 		goto out;
1364 	}
1365 
1366 	ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1367 	up_read(&EXT4_I(inode)->xattr_sem);
1368 	if (ret < 0)
1369 		goto out;
1370 
1371 	pos = 0;
1372 	parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1373 	while (pos < inline_size) {
1374 		/*
1375 		 * As inlined dir doesn't store any information about '.' and
1376 		 * only the inode number of '..' is stored, we have to handle
1377 		 * them differently.
1378 		 */
1379 		if (pos == 0) {
1380 			fake.inode = cpu_to_le32(inode->i_ino);
1381 			fake.name_len = 1;
1382 			strcpy(fake.name, ".");
1383 			fake.rec_len = ext4_rec_len_to_disk(
1384 					  ext4_dir_rec_len(fake.name_len, NULL),
1385 					  inline_size);
1386 			ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1387 			de = &fake;
1388 			pos = EXT4_INLINE_DOTDOT_OFFSET;
1389 		} else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1390 			fake.inode = cpu_to_le32(parent_ino);
1391 			fake.name_len = 2;
1392 			strcpy(fake.name, "..");
1393 			fake.rec_len = ext4_rec_len_to_disk(
1394 					  ext4_dir_rec_len(fake.name_len, NULL),
1395 					  inline_size);
1396 			ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1397 			de = &fake;
1398 			pos = EXT4_INLINE_DOTDOT_SIZE;
1399 		} else {
1400 			de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1401 			pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1402 			if (ext4_check_dir_entry(inode, dir_file, de,
1403 					 iloc.bh, dir_buf,
1404 					 inline_size, pos)) {
1405 				ret = count;
1406 				goto out;
1407 			}
1408 		}
1409 
1410 		if (ext4_hash_in_dirent(dir)) {
1411 			hinfo->hash = EXT4_DIRENT_HASH(de);
1412 			hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1413 		} else {
1414 			err = ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1415 			if (err) {
1416 				ret = err;
1417 				goto out;
1418 			}
1419 		}
1420 		if ((hinfo->hash < start_hash) ||
1421 		    ((hinfo->hash == start_hash) &&
1422 		     (hinfo->minor_hash < start_minor_hash)))
1423 			continue;
1424 		if (de->inode == 0)
1425 			continue;
1426 		tmp_str.name = de->name;
1427 		tmp_str.len = de->name_len;
1428 		err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1429 					      hinfo->minor_hash, de, &tmp_str);
1430 		if (err) {
1431 			ret = err;
1432 			goto out;
1433 		}
1434 		count++;
1435 	}
1436 	ret = count;
1437 out:
1438 	kfree(dir_buf);
1439 	brelse(iloc.bh);
1440 	return ret;
1441 }
1442 
1443 /*
1444  * So this function is called when the volume is mkfsed with
1445  * dir_index disabled. In order to keep f_pos persistent
1446  * after we convert from an inlined dir to a blocked based,
1447  * we just pretend that we are a normal dir and return the
1448  * offset as if '.' and '..' really take place.
1449  *
1450  */
ext4_read_inline_dir(struct file * file,struct dir_context * ctx,int * has_inline_data)1451 int ext4_read_inline_dir(struct file *file,
1452 			 struct dir_context *ctx,
1453 			 int *has_inline_data)
1454 {
1455 	unsigned int offset, parent_ino;
1456 	int i;
1457 	struct ext4_dir_entry_2 *de;
1458 	struct super_block *sb;
1459 	struct inode *inode = file_inode(file);
1460 	int ret, inline_size = 0;
1461 	struct ext4_iloc iloc;
1462 	void *dir_buf = NULL;
1463 	int dotdot_offset, dotdot_size, extra_offset, extra_size;
1464 
1465 	ret = ext4_get_inode_loc(inode, &iloc);
1466 	if (ret)
1467 		return ret;
1468 
1469 	down_read(&EXT4_I(inode)->xattr_sem);
1470 	if (!ext4_has_inline_data(inode)) {
1471 		up_read(&EXT4_I(inode)->xattr_sem);
1472 		*has_inline_data = 0;
1473 		goto out;
1474 	}
1475 
1476 	inline_size = ext4_get_inline_size(inode);
1477 	dir_buf = kmalloc(inline_size, GFP_NOFS);
1478 	if (!dir_buf) {
1479 		ret = -ENOMEM;
1480 		up_read(&EXT4_I(inode)->xattr_sem);
1481 		goto out;
1482 	}
1483 
1484 	ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1485 	up_read(&EXT4_I(inode)->xattr_sem);
1486 	if (ret < 0)
1487 		goto out;
1488 
1489 	ret = 0;
1490 	sb = inode->i_sb;
1491 	parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1492 	offset = ctx->pos;
1493 
1494 	/*
1495 	 * dotdot_offset and dotdot_size is the real offset and
1496 	 * size for ".." and "." if the dir is block based while
1497 	 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1498 	 * So we will use extra_offset and extra_size to indicate them
1499 	 * during the inline dir iteration.
1500 	 */
1501 	dotdot_offset = ext4_dir_rec_len(1, NULL);
1502 	dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1503 	extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1504 	extra_size = extra_offset + inline_size;
1505 
1506 	/*
1507 	 * If the version has changed since the last call to
1508 	 * readdir(2), then we might be pointing to an invalid
1509 	 * dirent right now.  Scan from the start of the inline
1510 	 * dir to make sure.
1511 	 */
1512 	if (!inode_eq_iversion(inode, file->f_version)) {
1513 		for (i = 0; i < extra_size && i < offset;) {
1514 			/*
1515 			 * "." is with offset 0 and
1516 			 * ".." is dotdot_offset.
1517 			 */
1518 			if (!i) {
1519 				i = dotdot_offset;
1520 				continue;
1521 			} else if (i == dotdot_offset) {
1522 				i = dotdot_size;
1523 				continue;
1524 			}
1525 			/* for other entry, the real offset in
1526 			 * the buf has to be tuned accordingly.
1527 			 */
1528 			de = (struct ext4_dir_entry_2 *)
1529 				(dir_buf + i - extra_offset);
1530 			/* It's too expensive to do a full
1531 			 * dirent test each time round this
1532 			 * loop, but we do have to test at
1533 			 * least that it is non-zero.  A
1534 			 * failure will be detected in the
1535 			 * dirent test below. */
1536 			if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1537 				< ext4_dir_rec_len(1, NULL))
1538 				break;
1539 			i += ext4_rec_len_from_disk(de->rec_len,
1540 						    extra_size);
1541 		}
1542 		offset = i;
1543 		ctx->pos = offset;
1544 		file->f_version = inode_query_iversion(inode);
1545 	}
1546 
1547 	while (ctx->pos < extra_size) {
1548 		if (ctx->pos == 0) {
1549 			if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1550 				goto out;
1551 			ctx->pos = dotdot_offset;
1552 			continue;
1553 		}
1554 
1555 		if (ctx->pos == dotdot_offset) {
1556 			if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1557 				goto out;
1558 			ctx->pos = dotdot_size;
1559 			continue;
1560 		}
1561 
1562 		de = (struct ext4_dir_entry_2 *)
1563 			(dir_buf + ctx->pos - extra_offset);
1564 		if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1565 					 extra_size, ctx->pos))
1566 			goto out;
1567 		if (le32_to_cpu(de->inode)) {
1568 			if (!dir_emit(ctx, de->name, de->name_len,
1569 				      le32_to_cpu(de->inode),
1570 				      get_dtype(sb, de->file_type)))
1571 				goto out;
1572 		}
1573 		ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1574 	}
1575 out:
1576 	kfree(dir_buf);
1577 	brelse(iloc.bh);
1578 	return ret;
1579 }
1580 
ext4_read_inline_link(struct inode * inode)1581 void *ext4_read_inline_link(struct inode *inode)
1582 {
1583 	struct ext4_iloc iloc;
1584 	int ret, inline_size;
1585 	void *link;
1586 
1587 	ret = ext4_get_inode_loc(inode, &iloc);
1588 	if (ret)
1589 		return ERR_PTR(ret);
1590 
1591 	ret = -ENOMEM;
1592 	inline_size = ext4_get_inline_size(inode);
1593 	link = kmalloc(inline_size + 1, GFP_NOFS);
1594 	if (!link)
1595 		goto out;
1596 
1597 	ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1598 	if (ret < 0) {
1599 		kfree(link);
1600 		goto out;
1601 	}
1602 	nd_terminate_link(link, inode->i_size, ret);
1603 out:
1604 	if (ret < 0)
1605 		link = ERR_PTR(ret);
1606 	brelse(iloc.bh);
1607 	return link;
1608 }
1609 
ext4_get_first_inline_block(struct inode * inode,struct ext4_dir_entry_2 ** parent_de,int * retval)1610 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1611 					struct ext4_dir_entry_2 **parent_de,
1612 					int *retval)
1613 {
1614 	struct ext4_iloc iloc;
1615 
1616 	*retval = ext4_get_inode_loc(inode, &iloc);
1617 	if (*retval)
1618 		return NULL;
1619 
1620 	*parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1621 
1622 	return iloc.bh;
1623 }
1624 
1625 /*
1626  * Try to create the inline data for the new dir.
1627  * If it succeeds, return 0, otherwise return the error.
1628  * In case of ENOSPC, the caller should create the normal disk layout dir.
1629  */
ext4_try_create_inline_dir(handle_t * handle,struct inode * parent,struct inode * inode)1630 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1631 			       struct inode *inode)
1632 {
1633 	int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1634 	struct ext4_iloc iloc;
1635 	struct ext4_dir_entry_2 *de;
1636 
1637 	ret = ext4_get_inode_loc(inode, &iloc);
1638 	if (ret)
1639 		return ret;
1640 
1641 	ret = ext4_prepare_inline_data(handle, inode, inline_size);
1642 	if (ret)
1643 		goto out;
1644 
1645 	/*
1646 	 * For inline dir, we only save the inode information for the ".."
1647 	 * and create a fake dentry to cover the left space.
1648 	 */
1649 	de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1650 	de->inode = cpu_to_le32(parent->i_ino);
1651 	de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1652 	de->inode = 0;
1653 	de->rec_len = ext4_rec_len_to_disk(
1654 				inline_size - EXT4_INLINE_DOTDOT_SIZE,
1655 				inline_size);
1656 	set_nlink(inode, 2);
1657 	inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1658 out:
1659 	brelse(iloc.bh);
1660 	return ret;
1661 }
1662 
ext4_find_inline_entry(struct inode * dir,struct ext4_filename * fname,struct ext4_dir_entry_2 ** res_dir,int * has_inline_data)1663 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1664 					struct ext4_filename *fname,
1665 					struct ext4_dir_entry_2 **res_dir,
1666 					int *has_inline_data)
1667 {
1668 	struct ext4_xattr_ibody_find is = {
1669 		.s = { .not_found = -ENODATA, },
1670 	};
1671 	struct ext4_xattr_info i = {
1672 		.name_index = EXT4_XATTR_INDEX_SYSTEM,
1673 		.name = EXT4_XATTR_SYSTEM_DATA,
1674 	};
1675 	int ret;
1676 	void *inline_start;
1677 	int inline_size;
1678 
1679 	ret = ext4_get_inode_loc(dir, &is.iloc);
1680 	if (ret)
1681 		return ERR_PTR(ret);
1682 
1683 	down_read(&EXT4_I(dir)->xattr_sem);
1684 
1685 	ret = ext4_xattr_ibody_find(dir, &i, &is);
1686 	if (ret)
1687 		goto out;
1688 
1689 	if (!ext4_has_inline_data(dir)) {
1690 		*has_inline_data = 0;
1691 		goto out;
1692 	}
1693 
1694 	inline_start = (void *)ext4_raw_inode(&is.iloc)->i_block +
1695 						EXT4_INLINE_DOTDOT_SIZE;
1696 	inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1697 	ret = ext4_search_dir(is.iloc.bh, inline_start, inline_size,
1698 			      dir, fname, 0, res_dir);
1699 	if (ret == 1)
1700 		goto out_find;
1701 	if (ret < 0)
1702 		goto out;
1703 
1704 	if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1705 		goto out;
1706 
1707 	inline_start = ext4_get_inline_xattr_pos(dir, &is.iloc);
1708 	inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1709 
1710 	ret = ext4_search_dir(is.iloc.bh, inline_start, inline_size,
1711 			      dir, fname, 0, res_dir);
1712 	if (ret == 1)
1713 		goto out_find;
1714 
1715 out:
1716 	brelse(is.iloc.bh);
1717 	if (ret < 0)
1718 		is.iloc.bh = ERR_PTR(ret);
1719 	else
1720 		is.iloc.bh = NULL;
1721 out_find:
1722 	up_read(&EXT4_I(dir)->xattr_sem);
1723 	return is.iloc.bh;
1724 }
1725 
ext4_delete_inline_entry(handle_t * handle,struct inode * dir,struct ext4_dir_entry_2 * de_del,struct buffer_head * bh,int * has_inline_data)1726 int ext4_delete_inline_entry(handle_t *handle,
1727 			     struct inode *dir,
1728 			     struct ext4_dir_entry_2 *de_del,
1729 			     struct buffer_head *bh,
1730 			     int *has_inline_data)
1731 {
1732 	int err, inline_size, no_expand;
1733 	struct ext4_iloc iloc;
1734 	void *inline_start;
1735 
1736 	err = ext4_get_inode_loc(dir, &iloc);
1737 	if (err)
1738 		return err;
1739 
1740 	ext4_write_lock_xattr(dir, &no_expand);
1741 	if (!ext4_has_inline_data(dir)) {
1742 		*has_inline_data = 0;
1743 		goto out;
1744 	}
1745 
1746 	if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1747 		EXT4_MIN_INLINE_DATA_SIZE) {
1748 		inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1749 					EXT4_INLINE_DOTDOT_SIZE;
1750 		inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1751 				EXT4_INLINE_DOTDOT_SIZE;
1752 	} else {
1753 		inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1754 		inline_size = ext4_get_inline_size(dir) -
1755 				EXT4_MIN_INLINE_DATA_SIZE;
1756 	}
1757 
1758 	BUFFER_TRACE(bh, "get_write_access");
1759 	err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1760 					    EXT4_JTR_NONE);
1761 	if (err)
1762 		goto out;
1763 
1764 	err = ext4_generic_delete_entry(dir, de_del, bh,
1765 					inline_start, inline_size, 0);
1766 	if (err)
1767 		goto out;
1768 
1769 	ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1770 out:
1771 	ext4_write_unlock_xattr(dir, &no_expand);
1772 	if (likely(err == 0))
1773 		err = ext4_mark_inode_dirty(handle, dir);
1774 	brelse(iloc.bh);
1775 	if (err != -ENOENT)
1776 		ext4_std_error(dir->i_sb, err);
1777 	return err;
1778 }
1779 
1780 /*
1781  * Get the inline dentry at offset.
1782  */
1783 static inline struct ext4_dir_entry_2 *
ext4_get_inline_entry(struct inode * inode,struct ext4_iloc * iloc,unsigned int offset,void ** inline_start,int * inline_size)1784 ext4_get_inline_entry(struct inode *inode,
1785 		      struct ext4_iloc *iloc,
1786 		      unsigned int offset,
1787 		      void **inline_start,
1788 		      int *inline_size)
1789 {
1790 	void *inline_pos;
1791 
1792 	BUG_ON(offset > ext4_get_inline_size(inode));
1793 
1794 	if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1795 		inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1796 		*inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1797 	} else {
1798 		inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1799 		offset -= EXT4_MIN_INLINE_DATA_SIZE;
1800 		*inline_size = ext4_get_inline_size(inode) -
1801 				EXT4_MIN_INLINE_DATA_SIZE;
1802 	}
1803 
1804 	if (inline_start)
1805 		*inline_start = inline_pos;
1806 	return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1807 }
1808 
empty_inline_dir(struct inode * dir,int * has_inline_data)1809 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1810 {
1811 	int err, inline_size;
1812 	struct ext4_iloc iloc;
1813 	size_t inline_len;
1814 	void *inline_pos;
1815 	unsigned int offset;
1816 	struct ext4_dir_entry_2 *de;
1817 	bool ret = false;
1818 
1819 	err = ext4_get_inode_loc(dir, &iloc);
1820 	if (err) {
1821 		EXT4_ERROR_INODE_ERR(dir, -err,
1822 				     "error %d getting inode %lu block",
1823 				     err, dir->i_ino);
1824 		return false;
1825 	}
1826 
1827 	down_read(&EXT4_I(dir)->xattr_sem);
1828 	if (!ext4_has_inline_data(dir)) {
1829 		*has_inline_data = 0;
1830 		ret = true;
1831 		goto out;
1832 	}
1833 
1834 	de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1835 	if (!le32_to_cpu(de->inode)) {
1836 		ext4_warning(dir->i_sb,
1837 			     "bad inline directory (dir #%lu) - no `..'",
1838 			     dir->i_ino);
1839 		goto out;
1840 	}
1841 
1842 	inline_len = ext4_get_inline_size(dir);
1843 	offset = EXT4_INLINE_DOTDOT_SIZE;
1844 	while (offset < inline_len) {
1845 		de = ext4_get_inline_entry(dir, &iloc, offset,
1846 					   &inline_pos, &inline_size);
1847 		if (ext4_check_dir_entry(dir, NULL, de,
1848 					 iloc.bh, inline_pos,
1849 					 inline_size, offset)) {
1850 			ext4_warning(dir->i_sb,
1851 				     "bad inline directory (dir #%lu) - "
1852 				     "inode %u, rec_len %u, name_len %d"
1853 				     "inline size %d",
1854 				     dir->i_ino, le32_to_cpu(de->inode),
1855 				     le16_to_cpu(de->rec_len), de->name_len,
1856 				     inline_size);
1857 			goto out;
1858 		}
1859 		if (le32_to_cpu(de->inode)) {
1860 			goto out;
1861 		}
1862 		offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1863 	}
1864 
1865 	ret = true;
1866 out:
1867 	up_read(&EXT4_I(dir)->xattr_sem);
1868 	brelse(iloc.bh);
1869 	return ret;
1870 }
1871 
ext4_destroy_inline_data(handle_t * handle,struct inode * inode)1872 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1873 {
1874 	int ret, no_expand;
1875 
1876 	ext4_write_lock_xattr(inode, &no_expand);
1877 	ret = ext4_destroy_inline_data_nolock(handle, inode);
1878 	ext4_write_unlock_xattr(inode, &no_expand);
1879 
1880 	return ret;
1881 }
1882 
ext4_inline_data_iomap(struct inode * inode,struct iomap * iomap)1883 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1884 {
1885 	__u64 addr;
1886 	int error = -EAGAIN;
1887 	struct ext4_iloc iloc;
1888 
1889 	down_read(&EXT4_I(inode)->xattr_sem);
1890 	if (!ext4_has_inline_data(inode))
1891 		goto out;
1892 
1893 	error = ext4_get_inode_loc(inode, &iloc);
1894 	if (error)
1895 		goto out;
1896 
1897 	addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1898 	addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1899 	addr += offsetof(struct ext4_inode, i_block);
1900 
1901 	brelse(iloc.bh);
1902 
1903 	iomap->addr = addr;
1904 	iomap->offset = 0;
1905 	iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1906 			      i_size_read(inode));
1907 	iomap->type = IOMAP_INLINE;
1908 	iomap->flags = 0;
1909 
1910 out:
1911 	up_read(&EXT4_I(inode)->xattr_sem);
1912 	return error;
1913 }
1914 
ext4_inline_data_truncate(struct inode * inode,int * has_inline)1915 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1916 {
1917 	handle_t *handle;
1918 	int inline_size, value_len, needed_blocks, no_expand, err = 0;
1919 	size_t i_size;
1920 	void *value = NULL;
1921 	struct ext4_xattr_ibody_find is = {
1922 		.s = { .not_found = -ENODATA, },
1923 	};
1924 	struct ext4_xattr_info i = {
1925 		.name_index = EXT4_XATTR_INDEX_SYSTEM,
1926 		.name = EXT4_XATTR_SYSTEM_DATA,
1927 	};
1928 
1929 
1930 	needed_blocks = ext4_writepage_trans_blocks(inode);
1931 	handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1932 	if (IS_ERR(handle))
1933 		return PTR_ERR(handle);
1934 
1935 	ext4_write_lock_xattr(inode, &no_expand);
1936 	if (!ext4_has_inline_data(inode)) {
1937 		ext4_write_unlock_xattr(inode, &no_expand);
1938 		*has_inline = 0;
1939 		ext4_journal_stop(handle);
1940 		return 0;
1941 	}
1942 
1943 	if ((err = ext4_orphan_add(handle, inode)) != 0)
1944 		goto out;
1945 
1946 	if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1947 		goto out;
1948 
1949 	down_write(&EXT4_I(inode)->i_data_sem);
1950 	i_size = inode->i_size;
1951 	inline_size = ext4_get_inline_size(inode);
1952 	EXT4_I(inode)->i_disksize = i_size;
1953 
1954 	if (i_size < inline_size) {
1955 		/*
1956 		 * if there's inline data to truncate and this file was
1957 		 * converted to extents after that inline data was written,
1958 		 * the extent status cache must be cleared to avoid leaving
1959 		 * behind stale delayed allocated extent entries
1960 		 */
1961 		if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1962 			ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1963 
1964 		/* Clear the content in the xattr space. */
1965 		if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1966 			if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1967 				goto out_error;
1968 
1969 			BUG_ON(is.s.not_found);
1970 
1971 			value_len = le32_to_cpu(is.s.here->e_value_size);
1972 			value = kmalloc(value_len, GFP_NOFS);
1973 			if (!value) {
1974 				err = -ENOMEM;
1975 				goto out_error;
1976 			}
1977 
1978 			err = ext4_xattr_ibody_get(inode, i.name_index,
1979 						   i.name, value, value_len);
1980 			if (err <= 0)
1981 				goto out_error;
1982 
1983 			i.value = value;
1984 			i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1985 					i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1986 			err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1987 			if (err)
1988 				goto out_error;
1989 		}
1990 
1991 		/* Clear the content within i_blocks. */
1992 		if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1993 			void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1994 			memset(p + i_size, 0,
1995 			       EXT4_MIN_INLINE_DATA_SIZE - i_size);
1996 		}
1997 
1998 		EXT4_I(inode)->i_inline_size = i_size <
1999 					EXT4_MIN_INLINE_DATA_SIZE ?
2000 					EXT4_MIN_INLINE_DATA_SIZE : i_size;
2001 	}
2002 
2003 out_error:
2004 	up_write(&EXT4_I(inode)->i_data_sem);
2005 out:
2006 	brelse(is.iloc.bh);
2007 	ext4_write_unlock_xattr(inode, &no_expand);
2008 	kfree(value);
2009 	if (inode->i_nlink)
2010 		ext4_orphan_del(handle, inode);
2011 
2012 	if (err == 0) {
2013 		inode->i_mtime = inode_set_ctime_current(inode);
2014 		err = ext4_mark_inode_dirty(handle, inode);
2015 		if (IS_SYNC(inode))
2016 			ext4_handle_sync(handle);
2017 	}
2018 	ext4_journal_stop(handle);
2019 	return err;
2020 }
2021 
ext4_convert_inline_data(struct inode * inode)2022 int ext4_convert_inline_data(struct inode *inode)
2023 {
2024 	int error, needed_blocks, no_expand;
2025 	handle_t *handle;
2026 	struct ext4_iloc iloc;
2027 
2028 	if (!ext4_has_inline_data(inode)) {
2029 		ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2030 		return 0;
2031 	} else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2032 		/*
2033 		 * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2034 		 * cleared. This means we are in the middle of moving of
2035 		 * inline data to delay allocated block. Just force writeout
2036 		 * here to finish conversion.
2037 		 */
2038 		error = filemap_flush(inode->i_mapping);
2039 		if (error)
2040 			return error;
2041 		if (!ext4_has_inline_data(inode))
2042 			return 0;
2043 	}
2044 
2045 	needed_blocks = ext4_writepage_trans_blocks(inode);
2046 
2047 	iloc.bh = NULL;
2048 	error = ext4_get_inode_loc(inode, &iloc);
2049 	if (error)
2050 		return error;
2051 
2052 	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2053 	if (IS_ERR(handle)) {
2054 		error = PTR_ERR(handle);
2055 		goto out_free;
2056 	}
2057 
2058 	ext4_write_lock_xattr(inode, &no_expand);
2059 	if (ext4_has_inline_data(inode))
2060 		error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2061 	ext4_write_unlock_xattr(inode, &no_expand);
2062 	ext4_journal_stop(handle);
2063 out_free:
2064 	brelse(iloc.bh);
2065 	return error;
2066 }
2067