xref: /openbmc/linux/fs/ext4/fsmap.c (revision fcc8487d)
1 /*
2  * Copyright (C) 2017 Oracle.  All Rights Reserved.
3  *
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version 2
9  * of the License, or (at your option) any later version.
10  *
11  * This program is distributed in the hope that it would be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write the Free Software Foundation,
18  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
19  */
20 #include "ext4.h"
21 #include <linux/fsmap.h>
22 #include "fsmap.h"
23 #include "mballoc.h"
24 #include <linux/sort.h>
25 #include <linux/list_sort.h>
26 #include <trace/events/ext4.h>
27 
28 /* Convert an ext4_fsmap to an fsmap. */
29 void ext4_fsmap_from_internal(struct super_block *sb, struct fsmap *dest,
30 			      struct ext4_fsmap *src)
31 {
32 	dest->fmr_device = src->fmr_device;
33 	dest->fmr_flags = src->fmr_flags;
34 	dest->fmr_physical = src->fmr_physical << sb->s_blocksize_bits;
35 	dest->fmr_owner = src->fmr_owner;
36 	dest->fmr_offset = 0;
37 	dest->fmr_length = src->fmr_length << sb->s_blocksize_bits;
38 	dest->fmr_reserved[0] = 0;
39 	dest->fmr_reserved[1] = 0;
40 	dest->fmr_reserved[2] = 0;
41 }
42 
43 /* Convert an fsmap to an ext4_fsmap. */
44 void ext4_fsmap_to_internal(struct super_block *sb, struct ext4_fsmap *dest,
45 			    struct fsmap *src)
46 {
47 	dest->fmr_device = src->fmr_device;
48 	dest->fmr_flags = src->fmr_flags;
49 	dest->fmr_physical = src->fmr_physical >> sb->s_blocksize_bits;
50 	dest->fmr_owner = src->fmr_owner;
51 	dest->fmr_length = src->fmr_length >> sb->s_blocksize_bits;
52 }
53 
54 /* getfsmap query state */
55 struct ext4_getfsmap_info {
56 	struct ext4_fsmap_head	*gfi_head;
57 	ext4_fsmap_format_t	gfi_formatter;	/* formatting fn */
58 	void			*gfi_format_arg;/* format buffer */
59 	ext4_fsblk_t		gfi_next_fsblk;	/* next fsblock we expect */
60 	u32			gfi_dev;	/* device id */
61 	ext4_group_t		gfi_agno;	/* bg number, if applicable */
62 	struct ext4_fsmap	gfi_low;	/* low rmap key */
63 	struct ext4_fsmap	gfi_high;	/* high rmap key */
64 	struct ext4_fsmap	gfi_lastfree;	/* free ext at end of last bg */
65 	struct list_head	gfi_meta_list;	/* fixed metadata list */
66 	bool			gfi_last;	/* last extent? */
67 };
68 
69 /* Associate a device with a getfsmap handler. */
70 struct ext4_getfsmap_dev {
71 	int			(*gfd_fn)(struct super_block *sb,
72 				      struct ext4_fsmap *keys,
73 				      struct ext4_getfsmap_info *info);
74 	u32			gfd_dev;
75 };
76 
77 /* Compare two getfsmap device handlers. */
78 static int ext4_getfsmap_dev_compare(const void *p1, const void *p2)
79 {
80 	const struct ext4_getfsmap_dev *d1 = p1;
81 	const struct ext4_getfsmap_dev *d2 = p2;
82 
83 	return d1->gfd_dev - d2->gfd_dev;
84 }
85 
86 /* Compare a record against our starting point */
87 static bool ext4_getfsmap_rec_before_low_key(struct ext4_getfsmap_info *info,
88 					     struct ext4_fsmap *rec)
89 {
90 	return rec->fmr_physical < info->gfi_low.fmr_physical;
91 }
92 
93 /*
94  * Format a reverse mapping for getfsmap, having translated rm_startblock
95  * into the appropriate daddr units.
96  */
97 static int ext4_getfsmap_helper(struct super_block *sb,
98 				struct ext4_getfsmap_info *info,
99 				struct ext4_fsmap *rec)
100 {
101 	struct ext4_fsmap fmr;
102 	struct ext4_sb_info *sbi = EXT4_SB(sb);
103 	ext4_fsblk_t rec_fsblk = rec->fmr_physical;
104 	ext4_group_t agno;
105 	ext4_grpblk_t cno;
106 	int error;
107 
108 	if (fatal_signal_pending(current))
109 		return -EINTR;
110 
111 	/*
112 	 * Filter out records that start before our startpoint, if the
113 	 * caller requested that.
114 	 */
115 	if (ext4_getfsmap_rec_before_low_key(info, rec)) {
116 		rec_fsblk += rec->fmr_length;
117 		if (info->gfi_next_fsblk < rec_fsblk)
118 			info->gfi_next_fsblk = rec_fsblk;
119 		return EXT4_QUERY_RANGE_CONTINUE;
120 	}
121 
122 	/* Are we just counting mappings? */
123 	if (info->gfi_head->fmh_count == 0) {
124 		if (rec_fsblk > info->gfi_next_fsblk)
125 			info->gfi_head->fmh_entries++;
126 
127 		if (info->gfi_last)
128 			return EXT4_QUERY_RANGE_CONTINUE;
129 
130 		info->gfi_head->fmh_entries++;
131 
132 		rec_fsblk += rec->fmr_length;
133 		if (info->gfi_next_fsblk < rec_fsblk)
134 			info->gfi_next_fsblk = rec_fsblk;
135 		return EXT4_QUERY_RANGE_CONTINUE;
136 	}
137 
138 	/*
139 	 * If the record starts past the last physical block we saw,
140 	 * then we've found a gap.  Report the gap as being owned by
141 	 * whatever the caller specified is the missing owner.
142 	 */
143 	if (rec_fsblk > info->gfi_next_fsblk) {
144 		if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
145 			return EXT4_QUERY_RANGE_ABORT;
146 
147 		ext4_get_group_no_and_offset(sb, info->gfi_next_fsblk,
148 				&agno, &cno);
149 		trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno,
150 				EXT4_C2B(sbi, cno),
151 				rec_fsblk - info->gfi_next_fsblk,
152 				EXT4_FMR_OWN_UNKNOWN);
153 
154 		fmr.fmr_device = info->gfi_dev;
155 		fmr.fmr_physical = info->gfi_next_fsblk;
156 		fmr.fmr_owner = EXT4_FMR_OWN_UNKNOWN;
157 		fmr.fmr_length = rec_fsblk - info->gfi_next_fsblk;
158 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
159 		error = info->gfi_formatter(&fmr, info->gfi_format_arg);
160 		if (error)
161 			return error;
162 		info->gfi_head->fmh_entries++;
163 	}
164 
165 	if (info->gfi_last)
166 		goto out;
167 
168 	/* Fill out the extent we found */
169 	if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
170 		return EXT4_QUERY_RANGE_ABORT;
171 
172 	ext4_get_group_no_and_offset(sb, rec_fsblk, &agno, &cno);
173 	trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno, EXT4_C2B(sbi, cno),
174 			rec->fmr_length, rec->fmr_owner);
175 
176 	fmr.fmr_device = info->gfi_dev;
177 	fmr.fmr_physical = rec_fsblk;
178 	fmr.fmr_owner = rec->fmr_owner;
179 	fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
180 	fmr.fmr_length = rec->fmr_length;
181 	error = info->gfi_formatter(&fmr, info->gfi_format_arg);
182 	if (error)
183 		return error;
184 	info->gfi_head->fmh_entries++;
185 
186 out:
187 	rec_fsblk += rec->fmr_length;
188 	if (info->gfi_next_fsblk < rec_fsblk)
189 		info->gfi_next_fsblk = rec_fsblk;
190 	return EXT4_QUERY_RANGE_CONTINUE;
191 }
192 
193 static inline ext4_fsblk_t ext4_fsmap_next_pblk(struct ext4_fsmap *fmr)
194 {
195 	return fmr->fmr_physical + fmr->fmr_length;
196 }
197 
198 /* Transform a blockgroup's free record into a fsmap */
199 static int ext4_getfsmap_datadev_helper(struct super_block *sb,
200 					ext4_group_t agno, ext4_grpblk_t start,
201 					ext4_grpblk_t len, void *priv)
202 {
203 	struct ext4_fsmap irec;
204 	struct ext4_getfsmap_info *info = priv;
205 	struct ext4_fsmap *p;
206 	struct ext4_fsmap *tmp;
207 	struct ext4_sb_info *sbi = EXT4_SB(sb);
208 	ext4_fsblk_t fsb;
209 	ext4_fsblk_t fslen;
210 	int error;
211 
212 	fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
213 	fslen = EXT4_C2B(sbi, len);
214 
215 	/* If the retained free extent record is set... */
216 	if (info->gfi_lastfree.fmr_owner) {
217 		/* ...and abuts this one, lengthen it and return. */
218 		if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
219 			info->gfi_lastfree.fmr_length += fslen;
220 			return 0;
221 		}
222 
223 		/*
224 		 * There's a gap between the two free extents; emit the
225 		 * retained extent prior to merging the meta_list.
226 		 */
227 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
228 		if (error)
229 			return error;
230 		info->gfi_lastfree.fmr_owner = 0;
231 	}
232 
233 	/* Merge in any relevant extents from the meta_list */
234 	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
235 		if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
236 			list_del(&p->fmr_list);
237 			kfree(p);
238 		} else if (p->fmr_physical < fsb) {
239 			error = ext4_getfsmap_helper(sb, info, p);
240 			if (error)
241 				return error;
242 
243 			list_del(&p->fmr_list);
244 			kfree(p);
245 		}
246 	}
247 
248 	irec.fmr_device = 0;
249 	irec.fmr_physical = fsb;
250 	irec.fmr_length = fslen;
251 	irec.fmr_owner = EXT4_FMR_OWN_FREE;
252 	irec.fmr_flags = 0;
253 
254 	/* If this is a free extent at the end of a bg, buffer it. */
255 	if (ext4_fsmap_next_pblk(&irec) ==
256 			ext4_group_first_block_no(sb, agno + 1)) {
257 		info->gfi_lastfree = irec;
258 		return 0;
259 	}
260 
261 	/* Otherwise, emit it */
262 	return ext4_getfsmap_helper(sb, info, &irec);
263 }
264 
265 /* Execute a getfsmap query against the log device. */
266 static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
267 				struct ext4_getfsmap_info *info)
268 {
269 	journal_t *journal = EXT4_SB(sb)->s_journal;
270 	struct ext4_fsmap irec;
271 
272 	/* Set up search keys */
273 	info->gfi_low = keys[0];
274 	info->gfi_low.fmr_length = 0;
275 
276 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
277 
278 	trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
279 			info->gfi_low.fmr_physical,
280 			info->gfi_low.fmr_length,
281 			info->gfi_low.fmr_owner);
282 
283 	trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
284 			info->gfi_high.fmr_physical,
285 			info->gfi_high.fmr_length,
286 			info->gfi_high.fmr_owner);
287 
288 	if (keys[0].fmr_physical > 0)
289 		return 0;
290 
291 	/* Fabricate an rmap entry for the external log device. */
292 	irec.fmr_physical = journal->j_blk_offset;
293 	irec.fmr_length = journal->j_maxlen;
294 	irec.fmr_owner = EXT4_FMR_OWN_LOG;
295 	irec.fmr_flags = 0;
296 
297 	return ext4_getfsmap_helper(sb, info, &irec);
298 }
299 
300 /* Helper to fill out an ext4_fsmap. */
301 static inline int ext4_getfsmap_fill(struct list_head *meta_list,
302 				     ext4_fsblk_t fsb, ext4_fsblk_t len,
303 				     uint64_t owner)
304 {
305 	struct ext4_fsmap *fsm;
306 
307 	fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
308 	if (!fsm)
309 		return -ENOMEM;
310 	fsm->fmr_device = 0;
311 	fsm->fmr_flags = 0;
312 	fsm->fmr_physical = fsb;
313 	fsm->fmr_owner = owner;
314 	fsm->fmr_length = len;
315 	list_add_tail(&fsm->fmr_list, meta_list);
316 
317 	return 0;
318 }
319 
320 /*
321  * This function returns the number of file system metadata blocks at
322  * the beginning of a block group, including the reserved gdt blocks.
323  */
324 static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
325 					  ext4_group_t agno,
326 					  struct list_head *meta_list)
327 {
328 	struct ext4_sb_info *sbi = EXT4_SB(sb);
329 	ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
330 	ext4_fsblk_t len;
331 	unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
332 	unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
333 	int error;
334 
335 	/* Record the superblock. */
336 	if (ext4_bg_has_super(sb, agno)) {
337 		error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
338 		if (error)
339 			return error;
340 		fsb++;
341 	}
342 
343 	/* Record the group descriptors. */
344 	len = ext4_bg_num_gdb(sb, agno);
345 	if (!len)
346 		return 0;
347 	error = ext4_getfsmap_fill(meta_list, fsb, len,
348 				   EXT4_FMR_OWN_GDT);
349 	if (error)
350 		return error;
351 	fsb += len;
352 
353 	/* Reserved GDT blocks */
354 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
355 		len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
356 		error = ext4_getfsmap_fill(meta_list, fsb, len,
357 					   EXT4_FMR_OWN_RESV_GDT);
358 		if (error)
359 			return error;
360 	}
361 
362 	return 0;
363 }
364 
365 /* Compare two fsmap items. */
366 static int ext4_getfsmap_compare(void *priv,
367 				 struct list_head *a,
368 				 struct list_head *b)
369 {
370 	struct ext4_fsmap *fa;
371 	struct ext4_fsmap *fb;
372 
373 	fa = container_of(a, struct ext4_fsmap, fmr_list);
374 	fb = container_of(b, struct ext4_fsmap, fmr_list);
375 	if (fa->fmr_physical < fb->fmr_physical)
376 		return -1;
377 	else if (fa->fmr_physical > fb->fmr_physical)
378 		return 1;
379 	return 0;
380 }
381 
382 /* Merge adjacent extents of fixed metadata. */
383 static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
384 {
385 	struct ext4_fsmap *p;
386 	struct ext4_fsmap *prev = NULL;
387 	struct ext4_fsmap *tmp;
388 
389 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
390 		if (!prev) {
391 			prev = p;
392 			continue;
393 		}
394 
395 		if (prev->fmr_owner == p->fmr_owner &&
396 		    prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
397 			prev->fmr_length += p->fmr_length;
398 			list_del(&p->fmr_list);
399 			kfree(p);
400 		} else
401 			prev = p;
402 	}
403 }
404 
405 /* Free a list of fixed metadata. */
406 static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
407 {
408 	struct ext4_fsmap *p;
409 	struct ext4_fsmap *tmp;
410 
411 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
412 		list_del(&p->fmr_list);
413 		kfree(p);
414 	}
415 }
416 
417 /* Find all the fixed metadata in the filesystem. */
418 int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
419 				      struct list_head *meta_list)
420 {
421 	struct ext4_group_desc *gdp;
422 	ext4_group_t agno;
423 	int error;
424 
425 	INIT_LIST_HEAD(meta_list);
426 
427 	/* Collect everything. */
428 	for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
429 		gdp = ext4_get_group_desc(sb, agno, NULL);
430 		if (!gdp) {
431 			error = -EFSCORRUPTED;
432 			goto err;
433 		}
434 
435 		/* Superblock & GDT */
436 		error = ext4_getfsmap_find_sb(sb, agno, meta_list);
437 		if (error)
438 			goto err;
439 
440 		/* Block bitmap */
441 		error = ext4_getfsmap_fill(meta_list,
442 					   ext4_block_bitmap(sb, gdp), 1,
443 					   EXT4_FMR_OWN_BLKBM);
444 		if (error)
445 			goto err;
446 
447 		/* Inode bitmap */
448 		error = ext4_getfsmap_fill(meta_list,
449 					   ext4_inode_bitmap(sb, gdp), 1,
450 					   EXT4_FMR_OWN_INOBM);
451 		if (error)
452 			goto err;
453 
454 		/* Inodes */
455 		error = ext4_getfsmap_fill(meta_list,
456 					   ext4_inode_table(sb, gdp),
457 					   EXT4_SB(sb)->s_itb_per_group,
458 					   EXT4_FMR_OWN_INODES);
459 		if (error)
460 			goto err;
461 	}
462 
463 	/* Sort the list */
464 	list_sort(NULL, meta_list, ext4_getfsmap_compare);
465 
466 	/* Merge adjacent extents */
467 	ext4_getfsmap_merge_fixed_metadata(meta_list);
468 
469 	return 0;
470 err:
471 	ext4_getfsmap_free_fixed_metadata(meta_list);
472 	return error;
473 }
474 
475 /* Execute a getfsmap query against the buddy bitmaps */
476 static int ext4_getfsmap_datadev(struct super_block *sb,
477 				 struct ext4_fsmap *keys,
478 				 struct ext4_getfsmap_info *info)
479 {
480 	struct ext4_sb_info *sbi = EXT4_SB(sb);
481 	ext4_fsblk_t start_fsb;
482 	ext4_fsblk_t end_fsb;
483 	ext4_fsblk_t eofs;
484 	ext4_group_t start_ag;
485 	ext4_group_t end_ag;
486 	ext4_grpblk_t first_cluster;
487 	ext4_grpblk_t last_cluster;
488 	int error = 0;
489 
490 	eofs = ext4_blocks_count(sbi->s_es);
491 	if (keys[0].fmr_physical >= eofs)
492 		return 0;
493 	if (keys[1].fmr_physical >= eofs)
494 		keys[1].fmr_physical = eofs - 1;
495 	start_fsb = keys[0].fmr_physical;
496 	end_fsb = keys[1].fmr_physical;
497 
498 	/* Determine first and last group to examine based on start and end */
499 	ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
500 	ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
501 
502 	/*
503 	 * Convert the fsmap low/high keys to bg based keys.  Initialize
504 	 * low to the fsmap low key and max out the high key to the end
505 	 * of the bg.
506 	 */
507 	info->gfi_low = keys[0];
508 	info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
509 	info->gfi_low.fmr_length = 0;
510 
511 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
512 
513 	/* Assemble a list of all the fixed-location metadata. */
514 	error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
515 	if (error)
516 		goto err;
517 
518 	/* Query each bg */
519 	for (info->gfi_agno = start_ag;
520 	     info->gfi_agno <= end_ag;
521 	     info->gfi_agno++) {
522 		/*
523 		 * Set the bg high key from the fsmap high key if this
524 		 * is the last bg that we're querying.
525 		 */
526 		if (info->gfi_agno == end_ag) {
527 			info->gfi_high = keys[1];
528 			info->gfi_high.fmr_physical = EXT4_C2B(sbi,
529 					last_cluster);
530 			info->gfi_high.fmr_length = 0;
531 		}
532 
533 		trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
534 				info->gfi_low.fmr_physical,
535 				info->gfi_low.fmr_length,
536 				info->gfi_low.fmr_owner);
537 
538 		trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
539 				info->gfi_high.fmr_physical,
540 				info->gfi_high.fmr_length,
541 				info->gfi_high.fmr_owner);
542 
543 		error = ext4_mballoc_query_range(sb, info->gfi_agno,
544 				EXT4_B2C(sbi, info->gfi_low.fmr_physical),
545 				EXT4_B2C(sbi, info->gfi_high.fmr_physical),
546 				ext4_getfsmap_datadev_helper, info);
547 		if (error)
548 			goto err;
549 
550 		/*
551 		 * Set the bg low key to the start of the bg prior to
552 		 * moving on to the next bg.
553 		 */
554 		if (info->gfi_agno == start_ag)
555 			memset(&info->gfi_low, 0, sizeof(info->gfi_low));
556 	}
557 
558 	/* Do we have a retained free extent? */
559 	if (info->gfi_lastfree.fmr_owner) {
560 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
561 		if (error)
562 			goto err;
563 	}
564 
565 	/* Report any gaps at the end of the bg */
566 	info->gfi_last = true;
567 	error = ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster, 0, info);
568 	if (error)
569 		goto err;
570 
571 err:
572 	ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
573 	return error;
574 }
575 
576 /* Do we recognize the device? */
577 static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
578 					  struct ext4_fsmap *fm)
579 {
580 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
581 	    fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
582 		return true;
583 	if (EXT4_SB(sb)->journal_bdev &&
584 	    fm->fmr_device == new_encode_dev(EXT4_SB(sb)->journal_bdev->bd_dev))
585 		return true;
586 	return false;
587 }
588 
589 /* Ensure that the low key is less than the high key. */
590 static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
591 				     struct ext4_fsmap *high_key)
592 {
593 	if (low_key->fmr_device > high_key->fmr_device)
594 		return false;
595 	if (low_key->fmr_device < high_key->fmr_device)
596 		return true;
597 
598 	if (low_key->fmr_physical > high_key->fmr_physical)
599 		return false;
600 	if (low_key->fmr_physical < high_key->fmr_physical)
601 		return true;
602 
603 	if (low_key->fmr_owner > high_key->fmr_owner)
604 		return false;
605 	if (low_key->fmr_owner < high_key->fmr_owner)
606 		return true;
607 
608 	return false;
609 }
610 
611 #define EXT4_GETFSMAP_DEVS	2
612 /*
613  * Get filesystem's extents as described in head, and format for
614  * output.  Calls formatter to fill the user's buffer until all
615  * extents are mapped, until the passed-in head->fmh_count slots have
616  * been filled, or until the formatter short-circuits the loop, if it
617  * is tracking filled-in extents on its own.
618  *
619  * Key to Confusion
620  * ----------------
621  * There are multiple levels of keys and counters at work here:
622  * _fsmap_head.fmh_keys		-- low and high fsmap keys passed in;
623  * 				   these reflect fs-wide block addrs.
624  * dkeys			-- fmh_keys used to query each device;
625  * 				   these are fmh_keys but w/ the low key
626  * 				   bumped up by fmr_length.
627  * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
628  *				   is how we detect gaps in the fsmap
629  *				   records and report them.
630  * _getfsmap_info.gfi_low/high	-- per-bg low/high keys computed from
631  * 				   dkeys; used to query the free space.
632  */
633 int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
634 		  ext4_fsmap_format_t formatter, void *arg)
635 {
636 	struct ext4_fsmap dkeys[2];	/* per-dev keys */
637 	struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
638 	struct ext4_getfsmap_info info = {0};
639 	int i;
640 	int error = 0;
641 
642 	if (head->fmh_iflags & ~FMH_IF_VALID)
643 		return -EINVAL;
644 	if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
645 	    !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
646 		return -EINVAL;
647 
648 	head->fmh_entries = 0;
649 
650 	/* Set up our device handlers. */
651 	memset(handlers, 0, sizeof(handlers));
652 	handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
653 	handlers[0].gfd_fn = ext4_getfsmap_datadev;
654 	if (EXT4_SB(sb)->journal_bdev) {
655 		handlers[1].gfd_dev = new_encode_dev(
656 				EXT4_SB(sb)->journal_bdev->bd_dev);
657 		handlers[1].gfd_fn = ext4_getfsmap_logdev;
658 	}
659 
660 	sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
661 			ext4_getfsmap_dev_compare, NULL);
662 
663 	/*
664 	 * To continue where we left off, we allow userspace to use the
665 	 * last mapping from a previous call as the low key of the next.
666 	 * This is identified by a non-zero length in the low key. We
667 	 * have to increment the low key in this scenario to ensure we
668 	 * don't return the same mapping again, and instead return the
669 	 * very next mapping.
670 	 *
671 	 * Bump the physical offset as there can be no other mapping for
672 	 * the same physical block range.
673 	 */
674 	dkeys[0] = head->fmh_keys[0];
675 	dkeys[0].fmr_physical += dkeys[0].fmr_length;
676 	dkeys[0].fmr_owner = 0;
677 	dkeys[0].fmr_length = 0;
678 	memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
679 
680 	if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
681 		return -EINVAL;
682 
683 	info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
684 			  head->fmh_keys[0].fmr_length;
685 	info.gfi_formatter = formatter;
686 	info.gfi_format_arg = arg;
687 	info.gfi_head = head;
688 
689 	/* For each device we support... */
690 	for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
691 		/* Is this device within the range the user asked for? */
692 		if (!handlers[i].gfd_fn)
693 			continue;
694 		if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
695 			continue;
696 		if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
697 			break;
698 
699 		/*
700 		 * If this device number matches the high key, we have
701 		 * to pass the high key to the handler to limit the
702 		 * query results.  If the device number exceeds the
703 		 * low key, zero out the low key so that we get
704 		 * everything from the beginning.
705 		 */
706 		if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
707 			dkeys[1] = head->fmh_keys[1];
708 		if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
709 			memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
710 
711 		info.gfi_dev = handlers[i].gfd_dev;
712 		info.gfi_last = false;
713 		info.gfi_agno = -1;
714 		error = handlers[i].gfd_fn(sb, dkeys, &info);
715 		if (error)
716 			break;
717 		info.gfi_next_fsblk = 0;
718 	}
719 
720 	head->fmh_oflags = FMH_OF_DEV_T;
721 	return error;
722 }
723