xref: /linux/fs/ext4/fsmap.c (revision 72bea132f3680ee51e7ed2cee62892b6f5121909)
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 /* Transform a blockgroup's free record into a fsmap */
189 static int ext4_getfsmap_datadev_helper(struct super_block *sb,
190 					ext4_group_t agno, ext4_grpblk_t start,
191 					ext4_grpblk_t len, void *priv)
192 {
193 	struct ext4_fsmap irec;
194 	struct ext4_getfsmap_info *info = priv;
195 	struct ext4_fsmap *p;
196 	struct ext4_fsmap *tmp;
197 	struct ext4_sb_info *sbi = EXT4_SB(sb);
198 	ext4_fsblk_t fsb;
199 	ext4_fsblk_t fslen;
200 	int error;
201 
202 	fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
203 	fslen = EXT4_C2B(sbi, len);
204 
205 	/* If the retained free extent record is set... */
206 	if (info->gfi_lastfree.fmr_owner) {
207 		/* ...and abuts this one, lengthen it and return. */
208 		if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
209 			info->gfi_lastfree.fmr_length += fslen;
210 			return 0;
211 		}
212 
213 		/*
214 		 * There's a gap between the two free extents; emit the
215 		 * retained extent prior to merging the meta_list.
216 		 */
217 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
218 		if (error)
219 			return error;
220 		info->gfi_lastfree.fmr_owner = 0;
221 	}
222 
223 	/* Merge in any relevant extents from the meta_list */
224 	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
225 		if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
226 			list_del(&p->fmr_list);
227 			kfree(p);
228 		} else if (p->fmr_physical < fsb) {
229 			error = ext4_getfsmap_helper(sb, info, p);
230 			if (error)
231 				return error;
232 
233 			list_del(&p->fmr_list);
234 			kfree(p);
235 		}
236 	}
237 
238 	irec.fmr_device = 0;
239 	irec.fmr_physical = fsb;
240 	irec.fmr_length = fslen;
241 	irec.fmr_owner = EXT4_FMR_OWN_FREE;
242 	irec.fmr_flags = 0;
243 
244 	/* If this is a free extent at the end of a bg, buffer it. */
245 	if (ext4_fsmap_next_pblk(&irec) ==
246 			ext4_group_first_block_no(sb, agno + 1)) {
247 		info->gfi_lastfree = irec;
248 		return 0;
249 	}
250 
251 	/* Otherwise, emit it */
252 	return ext4_getfsmap_helper(sb, info, &irec);
253 }
254 
255 /* Execute a getfsmap query against the log device. */
256 static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
257 				struct ext4_getfsmap_info *info)
258 {
259 	journal_t *journal = EXT4_SB(sb)->s_journal;
260 	struct ext4_fsmap irec;
261 
262 	/* Set up search keys */
263 	info->gfi_low = keys[0];
264 	info->gfi_low.fmr_length = 0;
265 
266 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
267 
268 	trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
269 			info->gfi_low.fmr_physical,
270 			info->gfi_low.fmr_length,
271 			info->gfi_low.fmr_owner);
272 
273 	trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
274 			info->gfi_high.fmr_physical,
275 			info->gfi_high.fmr_length,
276 			info->gfi_high.fmr_owner);
277 
278 	if (keys[0].fmr_physical > 0)
279 		return 0;
280 
281 	/* Fabricate an rmap entry for the external log device. */
282 	irec.fmr_physical = journal->j_blk_offset;
283 	irec.fmr_length = journal->j_total_len;
284 	irec.fmr_owner = EXT4_FMR_OWN_LOG;
285 	irec.fmr_flags = 0;
286 
287 	return ext4_getfsmap_helper(sb, info, &irec);
288 }
289 
290 /* Helper to fill out an ext4_fsmap. */
291 static inline int ext4_getfsmap_fill(struct list_head *meta_list,
292 				     ext4_fsblk_t fsb, ext4_fsblk_t len,
293 				     uint64_t owner)
294 {
295 	struct ext4_fsmap *fsm;
296 
297 	fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
298 	if (!fsm)
299 		return -ENOMEM;
300 	fsm->fmr_device = 0;
301 	fsm->fmr_flags = 0;
302 	fsm->fmr_physical = fsb;
303 	fsm->fmr_owner = owner;
304 	fsm->fmr_length = len;
305 	list_add_tail(&fsm->fmr_list, meta_list);
306 
307 	return 0;
308 }
309 
310 /*
311  * This function returns the number of file system metadata blocks at
312  * the beginning of a block group, including the reserved gdt blocks.
313  */
314 static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
315 					  ext4_group_t agno,
316 					  struct list_head *meta_list)
317 {
318 	struct ext4_sb_info *sbi = EXT4_SB(sb);
319 	ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
320 	ext4_fsblk_t len;
321 	unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
322 	unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
323 	int error;
324 
325 	/* Record the superblock. */
326 	if (ext4_bg_has_super(sb, agno)) {
327 		error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
328 		if (error)
329 			return error;
330 		fsb++;
331 	}
332 
333 	/* Record the group descriptors. */
334 	len = ext4_bg_num_gdb(sb, agno);
335 	if (!len)
336 		return 0;
337 	error = ext4_getfsmap_fill(meta_list, fsb, len,
338 				   EXT4_FMR_OWN_GDT);
339 	if (error)
340 		return error;
341 	fsb += len;
342 
343 	/* Reserved GDT blocks */
344 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
345 		len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
346 		error = ext4_getfsmap_fill(meta_list, fsb, len,
347 					   EXT4_FMR_OWN_RESV_GDT);
348 		if (error)
349 			return error;
350 	}
351 
352 	return 0;
353 }
354 
355 /* Compare two fsmap items. */
356 static int ext4_getfsmap_compare(void *priv,
357 				 const struct list_head *a,
358 				 const struct list_head *b)
359 {
360 	struct ext4_fsmap *fa;
361 	struct ext4_fsmap *fb;
362 
363 	fa = container_of(a, struct ext4_fsmap, fmr_list);
364 	fb = container_of(b, struct ext4_fsmap, fmr_list);
365 	if (fa->fmr_physical < fb->fmr_physical)
366 		return -1;
367 	else if (fa->fmr_physical > fb->fmr_physical)
368 		return 1;
369 	return 0;
370 }
371 
372 /* Merge adjacent extents of fixed metadata. */
373 static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
374 {
375 	struct ext4_fsmap *p;
376 	struct ext4_fsmap *prev = NULL;
377 	struct ext4_fsmap *tmp;
378 
379 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
380 		if (!prev) {
381 			prev = p;
382 			continue;
383 		}
384 
385 		if (prev->fmr_owner == p->fmr_owner &&
386 		    prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
387 			prev->fmr_length += p->fmr_length;
388 			list_del(&p->fmr_list);
389 			kfree(p);
390 		} else
391 			prev = p;
392 	}
393 }
394 
395 /* Free a list of fixed metadata. */
396 static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
397 {
398 	struct ext4_fsmap *p;
399 	struct ext4_fsmap *tmp;
400 
401 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
402 		list_del(&p->fmr_list);
403 		kfree(p);
404 	}
405 }
406 
407 /* Find all the fixed metadata in the filesystem. */
408 static int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
409 					     struct list_head *meta_list)
410 {
411 	struct ext4_group_desc *gdp;
412 	ext4_group_t agno;
413 	int error;
414 
415 	INIT_LIST_HEAD(meta_list);
416 
417 	/* Collect everything. */
418 	for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
419 		gdp = ext4_get_group_desc(sb, agno, NULL);
420 		if (!gdp) {
421 			error = -EFSCORRUPTED;
422 			goto err;
423 		}
424 
425 		/* Superblock & GDT */
426 		error = ext4_getfsmap_find_sb(sb, agno, meta_list);
427 		if (error)
428 			goto err;
429 
430 		/* Block bitmap */
431 		error = ext4_getfsmap_fill(meta_list,
432 					   ext4_block_bitmap(sb, gdp), 1,
433 					   EXT4_FMR_OWN_BLKBM);
434 		if (error)
435 			goto err;
436 
437 		/* Inode bitmap */
438 		error = ext4_getfsmap_fill(meta_list,
439 					   ext4_inode_bitmap(sb, gdp), 1,
440 					   EXT4_FMR_OWN_INOBM);
441 		if (error)
442 			goto err;
443 
444 		/* Inodes */
445 		error = ext4_getfsmap_fill(meta_list,
446 					   ext4_inode_table(sb, gdp),
447 					   EXT4_SB(sb)->s_itb_per_group,
448 					   EXT4_FMR_OWN_INODES);
449 		if (error)
450 			goto err;
451 	}
452 
453 	/* Sort the list */
454 	list_sort(NULL, meta_list, ext4_getfsmap_compare);
455 
456 	/* Merge adjacent extents */
457 	ext4_getfsmap_merge_fixed_metadata(meta_list);
458 
459 	return 0;
460 err:
461 	ext4_getfsmap_free_fixed_metadata(meta_list);
462 	return error;
463 }
464 
465 /* Execute a getfsmap query against the buddy bitmaps */
466 static int ext4_getfsmap_datadev(struct super_block *sb,
467 				 struct ext4_fsmap *keys,
468 				 struct ext4_getfsmap_info *info)
469 {
470 	struct ext4_sb_info *sbi = EXT4_SB(sb);
471 	ext4_fsblk_t start_fsb;
472 	ext4_fsblk_t end_fsb;
473 	ext4_fsblk_t bofs;
474 	ext4_fsblk_t eofs;
475 	ext4_group_t start_ag;
476 	ext4_group_t end_ag;
477 	ext4_grpblk_t first_cluster;
478 	ext4_grpblk_t last_cluster;
479 	int error = 0;
480 
481 	bofs = le32_to_cpu(sbi->s_es->s_first_data_block);
482 	eofs = ext4_blocks_count(sbi->s_es);
483 	if (keys[0].fmr_physical >= eofs)
484 		return 0;
485 	else if (keys[0].fmr_physical < bofs)
486 		keys[0].fmr_physical = bofs;
487 	if (keys[1].fmr_physical >= eofs)
488 		keys[1].fmr_physical = eofs - 1;
489 	if (keys[1].fmr_physical < keys[0].fmr_physical)
490 		return 0;
491 	start_fsb = keys[0].fmr_physical;
492 	end_fsb = keys[1].fmr_physical;
493 
494 	/* Determine first and last group to examine based on start and end */
495 	ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
496 	ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
497 
498 	/*
499 	 * Convert the fsmap low/high keys to bg based keys.  Initialize
500 	 * low to the fsmap low key and max out the high key to the end
501 	 * of the bg.
502 	 */
503 	info->gfi_low = keys[0];
504 	info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
505 	info->gfi_low.fmr_length = 0;
506 
507 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
508 
509 	/* Assemble a list of all the fixed-location metadata. */
510 	error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
511 	if (error)
512 		goto err;
513 
514 	/* Query each bg */
515 	for (info->gfi_agno = start_ag;
516 	     info->gfi_agno <= end_ag;
517 	     info->gfi_agno++) {
518 		/*
519 		 * Set the bg high key from the fsmap high key if this
520 		 * is the last bg that we're querying.
521 		 */
522 		if (info->gfi_agno == end_ag) {
523 			info->gfi_high = keys[1];
524 			info->gfi_high.fmr_physical = EXT4_C2B(sbi,
525 					last_cluster);
526 			info->gfi_high.fmr_length = 0;
527 		}
528 
529 		trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
530 				info->gfi_low.fmr_physical,
531 				info->gfi_low.fmr_length,
532 				info->gfi_low.fmr_owner);
533 
534 		trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
535 				info->gfi_high.fmr_physical,
536 				info->gfi_high.fmr_length,
537 				info->gfi_high.fmr_owner);
538 
539 		error = ext4_mballoc_query_range(sb, info->gfi_agno,
540 				EXT4_B2C(sbi, info->gfi_low.fmr_physical),
541 				EXT4_B2C(sbi, info->gfi_high.fmr_physical),
542 				ext4_getfsmap_datadev_helper, info);
543 		if (error)
544 			goto err;
545 
546 		/*
547 		 * Set the bg low key to the start of the bg prior to
548 		 * moving on to the next bg.
549 		 */
550 		if (info->gfi_agno == start_ag)
551 			memset(&info->gfi_low, 0, sizeof(info->gfi_low));
552 	}
553 
554 	/* Do we have a retained free extent? */
555 	if (info->gfi_lastfree.fmr_owner) {
556 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
557 		if (error)
558 			goto err;
559 	}
560 
561 	/* Report any gaps at the end of the bg */
562 	info->gfi_last = true;
563 	error = ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster, 0, info);
564 	if (error)
565 		goto err;
566 
567 err:
568 	ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
569 	return error;
570 }
571 
572 /* Do we recognize the device? */
573 static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
574 					  struct ext4_fsmap *fm)
575 {
576 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
577 	    fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
578 		return true;
579 	if (EXT4_SB(sb)->s_journal_bdev_file &&
580 	    fm->fmr_device ==
581 	    new_encode_dev(file_bdev(EXT4_SB(sb)->s_journal_bdev_file)->bd_dev))
582 		return true;
583 	return false;
584 }
585 
586 /* Ensure that the low key is less than the high key. */
587 static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
588 				     struct ext4_fsmap *high_key)
589 {
590 	if (low_key->fmr_device > high_key->fmr_device)
591 		return false;
592 	if (low_key->fmr_device < high_key->fmr_device)
593 		return true;
594 
595 	if (low_key->fmr_physical > high_key->fmr_physical)
596 		return false;
597 	if (low_key->fmr_physical < high_key->fmr_physical)
598 		return true;
599 
600 	if (low_key->fmr_owner > high_key->fmr_owner)
601 		return false;
602 	if (low_key->fmr_owner < high_key->fmr_owner)
603 		return true;
604 
605 	return false;
606 }
607 
608 #define EXT4_GETFSMAP_DEVS	2
609 /*
610  * Get filesystem's extents as described in head, and format for
611  * output.  Calls formatter to fill the user's buffer until all
612  * extents are mapped, until the passed-in head->fmh_count slots have
613  * been filled, or until the formatter short-circuits the loop, if it
614  * is tracking filled-in extents on its own.
615  *
616  * Key to Confusion
617  * ----------------
618  * There are multiple levels of keys and counters at work here:
619  * _fsmap_head.fmh_keys		-- low and high fsmap keys passed in;
620  * 				   these reflect fs-wide block addrs.
621  * dkeys			-- fmh_keys used to query each device;
622  * 				   these are fmh_keys but w/ the low key
623  * 				   bumped up by fmr_length.
624  * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
625  *				   is how we detect gaps in the fsmap
626  *				   records and report them.
627  * _getfsmap_info.gfi_low/high	-- per-bg low/high keys computed from
628  * 				   dkeys; used to query the free space.
629  */
630 int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
631 		  ext4_fsmap_format_t formatter, void *arg)
632 {
633 	struct ext4_fsmap dkeys[2];	/* per-dev keys */
634 	struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
635 	struct ext4_getfsmap_info info = { NULL };
636 	int i;
637 	int error = 0;
638 
639 	if (head->fmh_iflags & ~FMH_IF_VALID)
640 		return -EINVAL;
641 	if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
642 	    !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
643 		return -EINVAL;
644 
645 	head->fmh_entries = 0;
646 
647 	/* Set up our device handlers. */
648 	memset(handlers, 0, sizeof(handlers));
649 	handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
650 	handlers[0].gfd_fn = ext4_getfsmap_datadev;
651 	if (EXT4_SB(sb)->s_journal_bdev_file) {
652 		handlers[1].gfd_dev = new_encode_dev(
653 			file_bdev(EXT4_SB(sb)->s_journal_bdev_file)->bd_dev);
654 		handlers[1].gfd_fn = ext4_getfsmap_logdev;
655 	}
656 
657 	sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
658 			ext4_getfsmap_dev_compare, NULL);
659 
660 	/*
661 	 * To continue where we left off, we allow userspace to use the
662 	 * last mapping from a previous call as the low key of the next.
663 	 * This is identified by a non-zero length in the low key. We
664 	 * have to increment the low key in this scenario to ensure we
665 	 * don't return the same mapping again, and instead return the
666 	 * very next mapping.
667 	 *
668 	 * Bump the physical offset as there can be no other mapping for
669 	 * the same physical block range.
670 	 */
671 	dkeys[0] = head->fmh_keys[0];
672 	dkeys[0].fmr_physical += dkeys[0].fmr_length;
673 	dkeys[0].fmr_owner = 0;
674 	dkeys[0].fmr_length = 0;
675 	memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
676 
677 	if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
678 		return -EINVAL;
679 
680 	info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
681 			  head->fmh_keys[0].fmr_length;
682 	info.gfi_formatter = formatter;
683 	info.gfi_format_arg = arg;
684 	info.gfi_head = head;
685 
686 	/* For each device we support... */
687 	for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
688 		/* Is this device within the range the user asked for? */
689 		if (!handlers[i].gfd_fn)
690 			continue;
691 		if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
692 			continue;
693 		if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
694 			break;
695 
696 		/*
697 		 * If this device number matches the high key, we have
698 		 * to pass the high key to the handler to limit the
699 		 * query results.  If the device number exceeds the
700 		 * low key, zero out the low key so that we get
701 		 * everything from the beginning.
702 		 */
703 		if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
704 			dkeys[1] = head->fmh_keys[1];
705 		if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
706 			memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
707 
708 		info.gfi_dev = handlers[i].gfd_dev;
709 		info.gfi_last = false;
710 		info.gfi_agno = -1;
711 		error = handlers[i].gfd_fn(sb, dkeys, &info);
712 		if (error)
713 			break;
714 		info.gfi_next_fsblk = 0;
715 	}
716 
717 	head->fmh_oflags = FMH_OF_DEV_T;
718 	return error;
719 }
720