xref: /linux/fs/ext4/fsmap.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
4  *
5  * Author: Darrick J. Wong <darrick.wong@oracle.com>
6  */
7 #include "ext4.h"
8 #include <linux/fsmap.h>
9 #include "fsmap.h"
10 #include "mballoc.h"
11 #include <linux/sort.h>
12 #include <linux/list_sort.h>
13 #include <trace/events/ext4.h>
14 
15 /* Convert an ext4_fsmap to an fsmap. */
16 void ext4_fsmap_from_internal(struct super_block *sb, struct fsmap *dest,
17 			      struct ext4_fsmap *src)
18 {
19 	dest->fmr_device = src->fmr_device;
20 	dest->fmr_flags = src->fmr_flags;
21 	dest->fmr_physical = src->fmr_physical << sb->s_blocksize_bits;
22 	dest->fmr_owner = src->fmr_owner;
23 	dest->fmr_offset = 0;
24 	dest->fmr_length = src->fmr_length << sb->s_blocksize_bits;
25 	dest->fmr_reserved[0] = 0;
26 	dest->fmr_reserved[1] = 0;
27 	dest->fmr_reserved[2] = 0;
28 }
29 
30 /* Convert an fsmap to an ext4_fsmap. */
31 void ext4_fsmap_to_internal(struct super_block *sb, struct ext4_fsmap *dest,
32 			    struct fsmap *src)
33 {
34 	dest->fmr_device = src->fmr_device;
35 	dest->fmr_flags = src->fmr_flags;
36 	dest->fmr_physical = src->fmr_physical >> sb->s_blocksize_bits;
37 	dest->fmr_owner = src->fmr_owner;
38 	dest->fmr_length = src->fmr_length >> sb->s_blocksize_bits;
39 }
40 
41 /* getfsmap query state */
42 struct ext4_getfsmap_info {
43 	struct ext4_fsmap_head	*gfi_head;
44 	ext4_fsmap_format_t	gfi_formatter;	/* formatting fn */
45 	void			*gfi_format_arg;/* format buffer */
46 	ext4_fsblk_t		gfi_next_fsblk;	/* next fsblock we expect */
47 	u32			gfi_dev;	/* device id */
48 	ext4_group_t		gfi_agno;	/* bg number, if applicable */
49 	struct ext4_fsmap	gfi_low;	/* low rmap key */
50 	struct ext4_fsmap	gfi_high;	/* high rmap key */
51 	struct ext4_fsmap	gfi_lastfree;	/* free ext at end of last bg */
52 	struct list_head	gfi_meta_list;	/* fixed metadata list */
53 	bool			gfi_last;	/* last extent? */
54 };
55 
56 /* Associate a device with a getfsmap handler. */
57 struct ext4_getfsmap_dev {
58 	int			(*gfd_fn)(struct super_block *sb,
59 				      struct ext4_fsmap *keys,
60 				      struct ext4_getfsmap_info *info);
61 	u32			gfd_dev;
62 };
63 
64 /* Compare two getfsmap device handlers. */
65 static int ext4_getfsmap_dev_compare(const void *p1, const void *p2)
66 {
67 	const struct ext4_getfsmap_dev *d1 = p1;
68 	const struct ext4_getfsmap_dev *d2 = p2;
69 
70 	return d1->gfd_dev - d2->gfd_dev;
71 }
72 
73 /* Compare a record against our starting point */
74 static bool ext4_getfsmap_rec_before_low_key(struct ext4_getfsmap_info *info,
75 					     struct ext4_fsmap *rec)
76 {
77 	return rec->fmr_physical < info->gfi_low.fmr_physical;
78 }
79 
80 /*
81  * Format a reverse mapping for getfsmap, having translated rm_startblock
82  * into the appropriate daddr units.
83  */
84 static int ext4_getfsmap_helper(struct super_block *sb,
85 				struct ext4_getfsmap_info *info,
86 				struct ext4_fsmap *rec)
87 {
88 	struct ext4_fsmap fmr;
89 	struct ext4_sb_info *sbi = EXT4_SB(sb);
90 	ext4_fsblk_t rec_fsblk = rec->fmr_physical;
91 	ext4_group_t agno;
92 	ext4_grpblk_t cno;
93 	int error;
94 
95 	if (fatal_signal_pending(current))
96 		return -EINTR;
97 
98 	/*
99 	 * Filter out records that start before our startpoint, if the
100 	 * caller requested that.
101 	 */
102 	if (ext4_getfsmap_rec_before_low_key(info, rec)) {
103 		rec_fsblk += rec->fmr_length;
104 		if (info->gfi_next_fsblk < rec_fsblk)
105 			info->gfi_next_fsblk = rec_fsblk;
106 		return EXT4_QUERY_RANGE_CONTINUE;
107 	}
108 
109 	/* Are we just counting mappings? */
110 	if (info->gfi_head->fmh_count == 0) {
111 		if (info->gfi_head->fmh_entries == UINT_MAX)
112 			return EXT4_QUERY_RANGE_ABORT;
113 
114 		if (rec_fsblk > info->gfi_next_fsblk)
115 			info->gfi_head->fmh_entries++;
116 
117 		if (info->gfi_last)
118 			return EXT4_QUERY_RANGE_CONTINUE;
119 
120 		info->gfi_head->fmh_entries++;
121 
122 		rec_fsblk += rec->fmr_length;
123 		if (info->gfi_next_fsblk < rec_fsblk)
124 			info->gfi_next_fsblk = rec_fsblk;
125 		return EXT4_QUERY_RANGE_CONTINUE;
126 	}
127 
128 	/*
129 	 * If the record starts past the last physical block we saw,
130 	 * then we've found a gap.  Report the gap as being owned by
131 	 * whatever the caller specified is the missing owner.
132 	 */
133 	if (rec_fsblk > info->gfi_next_fsblk) {
134 		if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
135 			return EXT4_QUERY_RANGE_ABORT;
136 
137 		ext4_get_group_no_and_offset(sb, info->gfi_next_fsblk,
138 				&agno, &cno);
139 		trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno,
140 				EXT4_C2B(sbi, cno),
141 				rec_fsblk - info->gfi_next_fsblk,
142 				EXT4_FMR_OWN_UNKNOWN);
143 
144 		fmr.fmr_device = info->gfi_dev;
145 		fmr.fmr_physical = info->gfi_next_fsblk;
146 		fmr.fmr_owner = EXT4_FMR_OWN_UNKNOWN;
147 		fmr.fmr_length = rec_fsblk - info->gfi_next_fsblk;
148 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
149 		error = info->gfi_formatter(&fmr, info->gfi_format_arg);
150 		if (error)
151 			return error;
152 		info->gfi_head->fmh_entries++;
153 	}
154 
155 	if (info->gfi_last)
156 		goto out;
157 
158 	/* Fill out the extent we found */
159 	if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
160 		return EXT4_QUERY_RANGE_ABORT;
161 
162 	ext4_get_group_no_and_offset(sb, rec_fsblk, &agno, &cno);
163 	trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno, EXT4_C2B(sbi, cno),
164 			rec->fmr_length, rec->fmr_owner);
165 
166 	fmr.fmr_device = info->gfi_dev;
167 	fmr.fmr_physical = rec_fsblk;
168 	fmr.fmr_owner = rec->fmr_owner;
169 	fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
170 	fmr.fmr_length = rec->fmr_length;
171 	error = info->gfi_formatter(&fmr, info->gfi_format_arg);
172 	if (error)
173 		return error;
174 	info->gfi_head->fmh_entries++;
175 
176 out:
177 	rec_fsblk += rec->fmr_length;
178 	if (info->gfi_next_fsblk < rec_fsblk)
179 		info->gfi_next_fsblk = rec_fsblk;
180 	return EXT4_QUERY_RANGE_CONTINUE;
181 }
182 
183 static inline ext4_fsblk_t ext4_fsmap_next_pblk(struct ext4_fsmap *fmr)
184 {
185 	return fmr->fmr_physical + fmr->fmr_length;
186 }
187 
188 static int ext4_getfsmap_meta_helper(struct super_block *sb,
189 				     ext4_group_t agno, ext4_grpblk_t start,
190 				     ext4_grpblk_t len, void *priv)
191 {
192 	struct ext4_getfsmap_info *info = priv;
193 	struct ext4_fsmap *p;
194 	struct ext4_fsmap *tmp;
195 	struct ext4_sb_info *sbi = EXT4_SB(sb);
196 	ext4_fsblk_t fsb, fs_start, fs_end;
197 	int error;
198 
199 	fs_start = fsb = (EXT4_C2B(sbi, start) +
200 			  ext4_group_first_block_no(sb, agno));
201 	fs_end = fs_start + EXT4_C2B(sbi, len);
202 
203 	/* Return relevant extents from the meta_list */
204 	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
205 		if (p->fmr_physical < info->gfi_next_fsblk) {
206 			list_del(&p->fmr_list);
207 			kfree(p);
208 			continue;
209 		}
210 		if (p->fmr_physical <= fs_start ||
211 		    p->fmr_physical + p->fmr_length <= fs_end) {
212 			/* Emit the retained free extent record if present */
213 			if (info->gfi_lastfree.fmr_owner) {
214 				error = ext4_getfsmap_helper(sb, info,
215 							&info->gfi_lastfree);
216 				if (error)
217 					return error;
218 				info->gfi_lastfree.fmr_owner = 0;
219 			}
220 			error = ext4_getfsmap_helper(sb, info, p);
221 			if (error)
222 				return error;
223 			fsb = p->fmr_physical + p->fmr_length;
224 			if (info->gfi_next_fsblk < fsb)
225 				info->gfi_next_fsblk = fsb;
226 			list_del(&p->fmr_list);
227 			kfree(p);
228 			continue;
229 		}
230 	}
231 	if (info->gfi_next_fsblk < fsb)
232 		info->gfi_next_fsblk = fsb;
233 
234 	return 0;
235 }
236 
237 
238 /* Transform a blockgroup's free record into a fsmap */
239 static int ext4_getfsmap_datadev_helper(struct super_block *sb,
240 					ext4_group_t agno, ext4_grpblk_t start,
241 					ext4_grpblk_t len, void *priv)
242 {
243 	struct ext4_fsmap irec;
244 	struct ext4_getfsmap_info *info = priv;
245 	struct ext4_fsmap *p;
246 	struct ext4_fsmap *tmp;
247 	struct ext4_sb_info *sbi = EXT4_SB(sb);
248 	ext4_fsblk_t fsb;
249 	ext4_fsblk_t fslen;
250 	int error;
251 
252 	fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
253 	fslen = EXT4_C2B(sbi, len);
254 
255 	/* If the retained free extent record is set... */
256 	if (info->gfi_lastfree.fmr_owner) {
257 		/* ...and abuts this one, lengthen it and return. */
258 		if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
259 			info->gfi_lastfree.fmr_length += fslen;
260 			return 0;
261 		}
262 
263 		/*
264 		 * There's a gap between the two free extents; emit the
265 		 * retained extent prior to merging the meta_list.
266 		 */
267 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
268 		if (error)
269 			return error;
270 		info->gfi_lastfree.fmr_owner = 0;
271 	}
272 
273 	/* Merge in any relevant extents from the meta_list */
274 	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
275 		if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
276 			list_del(&p->fmr_list);
277 			kfree(p);
278 		} else if (p->fmr_physical < fsb) {
279 			error = ext4_getfsmap_helper(sb, info, p);
280 			if (error)
281 				return error;
282 
283 			list_del(&p->fmr_list);
284 			kfree(p);
285 		}
286 	}
287 
288 	irec.fmr_device = 0;
289 	irec.fmr_physical = fsb;
290 	irec.fmr_length = fslen;
291 	irec.fmr_owner = EXT4_FMR_OWN_FREE;
292 	irec.fmr_flags = 0;
293 
294 	/* If this is a free extent at the end of a bg, buffer it. */
295 	if (ext4_fsmap_next_pblk(&irec) ==
296 			ext4_group_first_block_no(sb, agno + 1)) {
297 		info->gfi_lastfree = irec;
298 		return 0;
299 	}
300 
301 	/* Otherwise, emit it */
302 	return ext4_getfsmap_helper(sb, info, &irec);
303 }
304 
305 /* Execute a getfsmap query against the log device. */
306 static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
307 				struct ext4_getfsmap_info *info)
308 {
309 	journal_t *journal = EXT4_SB(sb)->s_journal;
310 	struct ext4_fsmap irec;
311 
312 	/* Set up search keys */
313 	info->gfi_low = keys[0];
314 	info->gfi_low.fmr_length = 0;
315 
316 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
317 
318 	trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
319 			info->gfi_low.fmr_physical,
320 			info->gfi_low.fmr_length,
321 			info->gfi_low.fmr_owner);
322 
323 	trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
324 			info->gfi_high.fmr_physical,
325 			info->gfi_high.fmr_length,
326 			info->gfi_high.fmr_owner);
327 
328 	if (keys[0].fmr_physical > 0)
329 		return 0;
330 
331 	/* Fabricate an rmap entry for the external log device. */
332 	irec.fmr_physical = journal->j_blk_offset;
333 	irec.fmr_length = journal->j_total_len;
334 	irec.fmr_owner = EXT4_FMR_OWN_LOG;
335 	irec.fmr_flags = 0;
336 
337 	return ext4_getfsmap_helper(sb, info, &irec);
338 }
339 
340 /* Helper to fill out an ext4_fsmap. */
341 static inline int ext4_getfsmap_fill(struct list_head *meta_list,
342 				     ext4_fsblk_t fsb, ext4_fsblk_t len,
343 				     uint64_t owner)
344 {
345 	struct ext4_fsmap *fsm;
346 
347 	fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
348 	if (!fsm)
349 		return -ENOMEM;
350 	fsm->fmr_device = 0;
351 	fsm->fmr_flags = 0;
352 	fsm->fmr_physical = fsb;
353 	fsm->fmr_owner = owner;
354 	fsm->fmr_length = len;
355 	list_add_tail(&fsm->fmr_list, meta_list);
356 
357 	return 0;
358 }
359 
360 /*
361  * This function returns the number of file system metadata blocks at
362  * the beginning of a block group, including the reserved gdt blocks.
363  */
364 static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
365 					  ext4_group_t agno,
366 					  struct list_head *meta_list)
367 {
368 	struct ext4_sb_info *sbi = EXT4_SB(sb);
369 	ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
370 	ext4_fsblk_t len;
371 	unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
372 	unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
373 	int error;
374 
375 	/* Record the superblock. */
376 	if (ext4_bg_has_super(sb, agno)) {
377 		error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
378 		if (error)
379 			return error;
380 		fsb++;
381 	}
382 
383 	/* Record the group descriptors. */
384 	len = ext4_bg_num_gdb(sb, agno);
385 	if (!len)
386 		return 0;
387 	error = ext4_getfsmap_fill(meta_list, fsb, len,
388 				   EXT4_FMR_OWN_GDT);
389 	if (error)
390 		return error;
391 	fsb += len;
392 
393 	/* Reserved GDT blocks */
394 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
395 		len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
396 		error = ext4_getfsmap_fill(meta_list, fsb, len,
397 					   EXT4_FMR_OWN_RESV_GDT);
398 		if (error)
399 			return error;
400 	}
401 
402 	return 0;
403 }
404 
405 /* Compare two fsmap items. */
406 static int ext4_getfsmap_compare(void *priv,
407 				 const struct list_head *a,
408 				 const struct list_head *b)
409 {
410 	struct ext4_fsmap *fa;
411 	struct ext4_fsmap *fb;
412 
413 	fa = container_of(a, struct ext4_fsmap, fmr_list);
414 	fb = container_of(b, struct ext4_fsmap, fmr_list);
415 	if (fa->fmr_physical < fb->fmr_physical)
416 		return -1;
417 	else if (fa->fmr_physical > fb->fmr_physical)
418 		return 1;
419 	return 0;
420 }
421 
422 /* Merge adjacent extents of fixed metadata. */
423 static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
424 {
425 	struct ext4_fsmap *p;
426 	struct ext4_fsmap *prev = NULL;
427 	struct ext4_fsmap *tmp;
428 
429 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
430 		if (!prev) {
431 			prev = p;
432 			continue;
433 		}
434 
435 		if (prev->fmr_owner == p->fmr_owner &&
436 		    prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
437 			prev->fmr_length += p->fmr_length;
438 			list_del(&p->fmr_list);
439 			kfree(p);
440 		} else
441 			prev = p;
442 	}
443 }
444 
445 /* Free a list of fixed metadata. */
446 static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
447 {
448 	struct ext4_fsmap *p;
449 	struct ext4_fsmap *tmp;
450 
451 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
452 		list_del(&p->fmr_list);
453 		kfree(p);
454 	}
455 }
456 
457 /* Find all the fixed metadata in the filesystem. */
458 static int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
459 					     struct list_head *meta_list)
460 {
461 	struct ext4_group_desc *gdp;
462 	ext4_group_t agno;
463 	int error;
464 
465 	INIT_LIST_HEAD(meta_list);
466 
467 	/* Collect everything. */
468 	for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
469 		gdp = ext4_get_group_desc(sb, agno, NULL);
470 		if (!gdp) {
471 			error = -EFSCORRUPTED;
472 			goto err;
473 		}
474 
475 		/* Superblock & GDT */
476 		error = ext4_getfsmap_find_sb(sb, agno, meta_list);
477 		if (error)
478 			goto err;
479 
480 		/* Block bitmap */
481 		error = ext4_getfsmap_fill(meta_list,
482 					   ext4_block_bitmap(sb, gdp), 1,
483 					   EXT4_FMR_OWN_BLKBM);
484 		if (error)
485 			goto err;
486 
487 		/* Inode bitmap */
488 		error = ext4_getfsmap_fill(meta_list,
489 					   ext4_inode_bitmap(sb, gdp), 1,
490 					   EXT4_FMR_OWN_INOBM);
491 		if (error)
492 			goto err;
493 
494 		/* Inodes */
495 		error = ext4_getfsmap_fill(meta_list,
496 					   ext4_inode_table(sb, gdp),
497 					   EXT4_SB(sb)->s_itb_per_group,
498 					   EXT4_FMR_OWN_INODES);
499 		if (error)
500 			goto err;
501 	}
502 
503 	/* Sort the list */
504 	list_sort(NULL, meta_list, ext4_getfsmap_compare);
505 
506 	/* Merge adjacent extents */
507 	ext4_getfsmap_merge_fixed_metadata(meta_list);
508 
509 	return 0;
510 err:
511 	ext4_getfsmap_free_fixed_metadata(meta_list);
512 	return error;
513 }
514 
515 /* Execute a getfsmap query against the buddy bitmaps */
516 static int ext4_getfsmap_datadev(struct super_block *sb,
517 				 struct ext4_fsmap *keys,
518 				 struct ext4_getfsmap_info *info)
519 {
520 	struct ext4_sb_info *sbi = EXT4_SB(sb);
521 	ext4_fsblk_t start_fsb;
522 	ext4_fsblk_t end_fsb;
523 	ext4_fsblk_t bofs;
524 	ext4_fsblk_t eofs;
525 	ext4_group_t start_ag;
526 	ext4_group_t end_ag;
527 	ext4_grpblk_t first_cluster;
528 	ext4_grpblk_t last_cluster;
529 	int error = 0;
530 
531 	bofs = le32_to_cpu(sbi->s_es->s_first_data_block);
532 	eofs = ext4_blocks_count(sbi->s_es);
533 	if (keys[0].fmr_physical >= eofs)
534 		return 0;
535 	else if (keys[0].fmr_physical < bofs)
536 		keys[0].fmr_physical = bofs;
537 	if (keys[1].fmr_physical >= eofs)
538 		keys[1].fmr_physical = eofs - 1;
539 	if (keys[1].fmr_physical < keys[0].fmr_physical)
540 		return 0;
541 	start_fsb = keys[0].fmr_physical;
542 	end_fsb = keys[1].fmr_physical;
543 
544 	/* Determine first and last group to examine based on start and end */
545 	ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
546 	ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
547 
548 	/*
549 	 * Convert the fsmap low/high keys to bg based keys.  Initialize
550 	 * low to the fsmap low key and max out the high key to the end
551 	 * of the bg.
552 	 */
553 	info->gfi_low = keys[0];
554 	info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
555 	info->gfi_low.fmr_length = 0;
556 
557 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
558 
559 	/* Assemble a list of all the fixed-location metadata. */
560 	error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
561 	if (error)
562 		goto err;
563 
564 	/* Query each bg */
565 	for (info->gfi_agno = start_ag;
566 	     info->gfi_agno <= end_ag;
567 	     info->gfi_agno++) {
568 		/*
569 		 * Set the bg high key from the fsmap high key if this
570 		 * is the last bg that we're querying.
571 		 */
572 		if (info->gfi_agno == end_ag) {
573 			info->gfi_high = keys[1];
574 			info->gfi_high.fmr_physical = EXT4_C2B(sbi,
575 					last_cluster);
576 			info->gfi_high.fmr_length = 0;
577 		}
578 
579 		trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
580 				info->gfi_low.fmr_physical,
581 				info->gfi_low.fmr_length,
582 				info->gfi_low.fmr_owner);
583 
584 		trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
585 				info->gfi_high.fmr_physical,
586 				info->gfi_high.fmr_length,
587 				info->gfi_high.fmr_owner);
588 
589 		error = ext4_mballoc_query_range(sb, info->gfi_agno,
590 				EXT4_B2C(sbi, info->gfi_low.fmr_physical),
591 				EXT4_B2C(sbi, info->gfi_high.fmr_physical),
592 				ext4_getfsmap_meta_helper,
593 				ext4_getfsmap_datadev_helper, info);
594 		if (error)
595 			goto err;
596 
597 		/*
598 		 * Set the bg low key to the start of the bg prior to
599 		 * moving on to the next bg.
600 		 */
601 		if (info->gfi_agno == start_ag)
602 			memset(&info->gfi_low, 0, sizeof(info->gfi_low));
603 	}
604 
605 	/* Do we have a retained free extent? */
606 	if (info->gfi_lastfree.fmr_owner) {
607 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
608 		if (error)
609 			goto err;
610 	}
611 
612 	/* Report any gaps at the end of the bg */
613 	info->gfi_last = true;
614 	error = ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster + 1,
615 					     0, info);
616 	if (error)
617 		goto err;
618 
619 err:
620 	ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
621 	return error;
622 }
623 
624 /* Do we recognize the device? */
625 static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
626 					  struct ext4_fsmap *fm)
627 {
628 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
629 	    fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
630 		return true;
631 	if (EXT4_SB(sb)->s_journal_bdev_file &&
632 	    fm->fmr_device ==
633 	    new_encode_dev(file_bdev(EXT4_SB(sb)->s_journal_bdev_file)->bd_dev))
634 		return true;
635 	return false;
636 }
637 
638 /* Ensure that the low key is less than the high key. */
639 static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
640 				     struct ext4_fsmap *high_key)
641 {
642 	if (low_key->fmr_device > high_key->fmr_device)
643 		return false;
644 	if (low_key->fmr_device < high_key->fmr_device)
645 		return true;
646 
647 	if (low_key->fmr_physical > high_key->fmr_physical)
648 		return false;
649 	if (low_key->fmr_physical < high_key->fmr_physical)
650 		return true;
651 
652 	if (low_key->fmr_owner > high_key->fmr_owner)
653 		return false;
654 	if (low_key->fmr_owner < high_key->fmr_owner)
655 		return true;
656 
657 	return false;
658 }
659 
660 #define EXT4_GETFSMAP_DEVS	2
661 /*
662  * Get filesystem's extents as described in head, and format for
663  * output.  Calls formatter to fill the user's buffer until all
664  * extents are mapped, until the passed-in head->fmh_count slots have
665  * been filled, or until the formatter short-circuits the loop, if it
666  * is tracking filled-in extents on its own.
667  *
668  * Key to Confusion
669  * ----------------
670  * There are multiple levels of keys and counters at work here:
671  * _fsmap_head.fmh_keys		-- low and high fsmap keys passed in;
672  * 				   these reflect fs-wide block addrs.
673  * dkeys			-- fmh_keys used to query each device;
674  * 				   these are fmh_keys but w/ the low key
675  * 				   bumped up by fmr_length.
676  * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
677  *				   is how we detect gaps in the fsmap
678  *				   records and report them.
679  * _getfsmap_info.gfi_low/high	-- per-bg low/high keys computed from
680  * 				   dkeys; used to query the free space.
681  */
682 int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
683 		  ext4_fsmap_format_t formatter, void *arg)
684 {
685 	struct ext4_fsmap dkeys[2];	/* per-dev keys */
686 	struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
687 	struct ext4_getfsmap_info info = { NULL };
688 	int i;
689 	int error = 0;
690 
691 	if (head->fmh_iflags & ~FMH_IF_VALID)
692 		return -EINVAL;
693 	if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
694 	    !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
695 		return -EINVAL;
696 
697 	head->fmh_entries = 0;
698 
699 	/* Set up our device handlers. */
700 	memset(handlers, 0, sizeof(handlers));
701 	handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
702 	handlers[0].gfd_fn = ext4_getfsmap_datadev;
703 	if (EXT4_SB(sb)->s_journal_bdev_file) {
704 		handlers[1].gfd_dev = new_encode_dev(
705 			file_bdev(EXT4_SB(sb)->s_journal_bdev_file)->bd_dev);
706 		handlers[1].gfd_fn = ext4_getfsmap_logdev;
707 	}
708 
709 	sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
710 			ext4_getfsmap_dev_compare, NULL);
711 
712 	/*
713 	 * To continue where we left off, we allow userspace to use the
714 	 * last mapping from a previous call as the low key of the next.
715 	 * This is identified by a non-zero length in the low key. We
716 	 * have to increment the low key in this scenario to ensure we
717 	 * don't return the same mapping again, and instead return the
718 	 * very next mapping.
719 	 *
720 	 * Bump the physical offset as there can be no other mapping for
721 	 * the same physical block range.
722 	 */
723 	dkeys[0] = head->fmh_keys[0];
724 	dkeys[0].fmr_physical += dkeys[0].fmr_length;
725 	dkeys[0].fmr_owner = 0;
726 	dkeys[0].fmr_length = 0;
727 	memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
728 
729 	if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
730 		return -EINVAL;
731 
732 	info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
733 			  head->fmh_keys[0].fmr_length;
734 	info.gfi_formatter = formatter;
735 	info.gfi_format_arg = arg;
736 	info.gfi_head = head;
737 
738 	/* For each device we support... */
739 	for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
740 		/* Is this device within the range the user asked for? */
741 		if (!handlers[i].gfd_fn)
742 			continue;
743 		if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
744 			continue;
745 		if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
746 			break;
747 
748 		/*
749 		 * If this device number matches the high key, we have
750 		 * to pass the high key to the handler to limit the
751 		 * query results.  If the device number exceeds the
752 		 * low key, zero out the low key so that we get
753 		 * everything from the beginning.
754 		 */
755 		if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
756 			dkeys[1] = head->fmh_keys[1];
757 		if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
758 			memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
759 
760 		info.gfi_dev = handlers[i].gfd_dev;
761 		info.gfi_last = false;
762 		info.gfi_agno = -1;
763 		error = handlers[i].gfd_fn(sb, dkeys, &info);
764 		if (error)
765 			break;
766 		info.gfi_next_fsblk = 0;
767 	}
768 
769 	head->fmh_oflags = FMH_OF_DEV_T;
770 	return error;
771 }
772