xref: /openbmc/linux/fs/ext4/fsmap.c (revision 55fd7e02)
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 (rec_fsblk > info->gfi_next_fsblk)
112 			info->gfi_head->fmh_entries++;
113 
114 		if (info->gfi_last)
115 			return EXT4_QUERY_RANGE_CONTINUE;
116 
117 		info->gfi_head->fmh_entries++;
118 
119 		rec_fsblk += rec->fmr_length;
120 		if (info->gfi_next_fsblk < rec_fsblk)
121 			info->gfi_next_fsblk = rec_fsblk;
122 		return EXT4_QUERY_RANGE_CONTINUE;
123 	}
124 
125 	/*
126 	 * If the record starts past the last physical block we saw,
127 	 * then we've found a gap.  Report the gap as being owned by
128 	 * whatever the caller specified is the missing owner.
129 	 */
130 	if (rec_fsblk > info->gfi_next_fsblk) {
131 		if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
132 			return EXT4_QUERY_RANGE_ABORT;
133 
134 		ext4_get_group_no_and_offset(sb, info->gfi_next_fsblk,
135 				&agno, &cno);
136 		trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno,
137 				EXT4_C2B(sbi, cno),
138 				rec_fsblk - info->gfi_next_fsblk,
139 				EXT4_FMR_OWN_UNKNOWN);
140 
141 		fmr.fmr_device = info->gfi_dev;
142 		fmr.fmr_physical = info->gfi_next_fsblk;
143 		fmr.fmr_owner = EXT4_FMR_OWN_UNKNOWN;
144 		fmr.fmr_length = rec_fsblk - info->gfi_next_fsblk;
145 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
146 		error = info->gfi_formatter(&fmr, info->gfi_format_arg);
147 		if (error)
148 			return error;
149 		info->gfi_head->fmh_entries++;
150 	}
151 
152 	if (info->gfi_last)
153 		goto out;
154 
155 	/* Fill out the extent we found */
156 	if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
157 		return EXT4_QUERY_RANGE_ABORT;
158 
159 	ext4_get_group_no_and_offset(sb, rec_fsblk, &agno, &cno);
160 	trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno, EXT4_C2B(sbi, cno),
161 			rec->fmr_length, rec->fmr_owner);
162 
163 	fmr.fmr_device = info->gfi_dev;
164 	fmr.fmr_physical = rec_fsblk;
165 	fmr.fmr_owner = rec->fmr_owner;
166 	fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
167 	fmr.fmr_length = rec->fmr_length;
168 	error = info->gfi_formatter(&fmr, info->gfi_format_arg);
169 	if (error)
170 		return error;
171 	info->gfi_head->fmh_entries++;
172 
173 out:
174 	rec_fsblk += rec->fmr_length;
175 	if (info->gfi_next_fsblk < rec_fsblk)
176 		info->gfi_next_fsblk = rec_fsblk;
177 	return EXT4_QUERY_RANGE_CONTINUE;
178 }
179 
180 static inline ext4_fsblk_t ext4_fsmap_next_pblk(struct ext4_fsmap *fmr)
181 {
182 	return fmr->fmr_physical + fmr->fmr_length;
183 }
184 
185 /* Transform a blockgroup's free record into a fsmap */
186 static int ext4_getfsmap_datadev_helper(struct super_block *sb,
187 					ext4_group_t agno, ext4_grpblk_t start,
188 					ext4_grpblk_t len, void *priv)
189 {
190 	struct ext4_fsmap irec;
191 	struct ext4_getfsmap_info *info = priv;
192 	struct ext4_fsmap *p;
193 	struct ext4_fsmap *tmp;
194 	struct ext4_sb_info *sbi = EXT4_SB(sb);
195 	ext4_fsblk_t fsb;
196 	ext4_fsblk_t fslen;
197 	int error;
198 
199 	fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
200 	fslen = EXT4_C2B(sbi, len);
201 
202 	/* If the retained free extent record is set... */
203 	if (info->gfi_lastfree.fmr_owner) {
204 		/* ...and abuts this one, lengthen it and return. */
205 		if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
206 			info->gfi_lastfree.fmr_length += fslen;
207 			return 0;
208 		}
209 
210 		/*
211 		 * There's a gap between the two free extents; emit the
212 		 * retained extent prior to merging the meta_list.
213 		 */
214 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
215 		if (error)
216 			return error;
217 		info->gfi_lastfree.fmr_owner = 0;
218 	}
219 
220 	/* Merge in any relevant extents from the meta_list */
221 	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
222 		if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
223 			list_del(&p->fmr_list);
224 			kfree(p);
225 		} else if (p->fmr_physical < fsb) {
226 			error = ext4_getfsmap_helper(sb, info, p);
227 			if (error)
228 				return error;
229 
230 			list_del(&p->fmr_list);
231 			kfree(p);
232 		}
233 	}
234 
235 	irec.fmr_device = 0;
236 	irec.fmr_physical = fsb;
237 	irec.fmr_length = fslen;
238 	irec.fmr_owner = EXT4_FMR_OWN_FREE;
239 	irec.fmr_flags = 0;
240 
241 	/* If this is a free extent at the end of a bg, buffer it. */
242 	if (ext4_fsmap_next_pblk(&irec) ==
243 			ext4_group_first_block_no(sb, agno + 1)) {
244 		info->gfi_lastfree = irec;
245 		return 0;
246 	}
247 
248 	/* Otherwise, emit it */
249 	return ext4_getfsmap_helper(sb, info, &irec);
250 }
251 
252 /* Execute a getfsmap query against the log device. */
253 static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
254 				struct ext4_getfsmap_info *info)
255 {
256 	journal_t *journal = EXT4_SB(sb)->s_journal;
257 	struct ext4_fsmap irec;
258 
259 	/* Set up search keys */
260 	info->gfi_low = keys[0];
261 	info->gfi_low.fmr_length = 0;
262 
263 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
264 
265 	trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
266 			info->gfi_low.fmr_physical,
267 			info->gfi_low.fmr_length,
268 			info->gfi_low.fmr_owner);
269 
270 	trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
271 			info->gfi_high.fmr_physical,
272 			info->gfi_high.fmr_length,
273 			info->gfi_high.fmr_owner);
274 
275 	if (keys[0].fmr_physical > 0)
276 		return 0;
277 
278 	/* Fabricate an rmap entry for the external log device. */
279 	irec.fmr_physical = journal->j_blk_offset;
280 	irec.fmr_length = journal->j_maxlen;
281 	irec.fmr_owner = EXT4_FMR_OWN_LOG;
282 	irec.fmr_flags = 0;
283 
284 	return ext4_getfsmap_helper(sb, info, &irec);
285 }
286 
287 /* Helper to fill out an ext4_fsmap. */
288 static inline int ext4_getfsmap_fill(struct list_head *meta_list,
289 				     ext4_fsblk_t fsb, ext4_fsblk_t len,
290 				     uint64_t owner)
291 {
292 	struct ext4_fsmap *fsm;
293 
294 	fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
295 	if (!fsm)
296 		return -ENOMEM;
297 	fsm->fmr_device = 0;
298 	fsm->fmr_flags = 0;
299 	fsm->fmr_physical = fsb;
300 	fsm->fmr_owner = owner;
301 	fsm->fmr_length = len;
302 	list_add_tail(&fsm->fmr_list, meta_list);
303 
304 	return 0;
305 }
306 
307 /*
308  * This function returns the number of file system metadata blocks at
309  * the beginning of a block group, including the reserved gdt blocks.
310  */
311 static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
312 					  ext4_group_t agno,
313 					  struct list_head *meta_list)
314 {
315 	struct ext4_sb_info *sbi = EXT4_SB(sb);
316 	ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
317 	ext4_fsblk_t len;
318 	unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
319 	unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
320 	int error;
321 
322 	/* Record the superblock. */
323 	if (ext4_bg_has_super(sb, agno)) {
324 		error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
325 		if (error)
326 			return error;
327 		fsb++;
328 	}
329 
330 	/* Record the group descriptors. */
331 	len = ext4_bg_num_gdb(sb, agno);
332 	if (!len)
333 		return 0;
334 	error = ext4_getfsmap_fill(meta_list, fsb, len,
335 				   EXT4_FMR_OWN_GDT);
336 	if (error)
337 		return error;
338 	fsb += len;
339 
340 	/* Reserved GDT blocks */
341 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
342 		len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
343 		error = ext4_getfsmap_fill(meta_list, fsb, len,
344 					   EXT4_FMR_OWN_RESV_GDT);
345 		if (error)
346 			return error;
347 	}
348 
349 	return 0;
350 }
351 
352 /* Compare two fsmap items. */
353 static int ext4_getfsmap_compare(void *priv,
354 				 struct list_head *a,
355 				 struct list_head *b)
356 {
357 	struct ext4_fsmap *fa;
358 	struct ext4_fsmap *fb;
359 
360 	fa = container_of(a, struct ext4_fsmap, fmr_list);
361 	fb = container_of(b, struct ext4_fsmap, fmr_list);
362 	if (fa->fmr_physical < fb->fmr_physical)
363 		return -1;
364 	else if (fa->fmr_physical > fb->fmr_physical)
365 		return 1;
366 	return 0;
367 }
368 
369 /* Merge adjacent extents of fixed metadata. */
370 static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
371 {
372 	struct ext4_fsmap *p;
373 	struct ext4_fsmap *prev = NULL;
374 	struct ext4_fsmap *tmp;
375 
376 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
377 		if (!prev) {
378 			prev = p;
379 			continue;
380 		}
381 
382 		if (prev->fmr_owner == p->fmr_owner &&
383 		    prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
384 			prev->fmr_length += p->fmr_length;
385 			list_del(&p->fmr_list);
386 			kfree(p);
387 		} else
388 			prev = p;
389 	}
390 }
391 
392 /* Free a list of fixed metadata. */
393 static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
394 {
395 	struct ext4_fsmap *p;
396 	struct ext4_fsmap *tmp;
397 
398 	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
399 		list_del(&p->fmr_list);
400 		kfree(p);
401 	}
402 }
403 
404 /* Find all the fixed metadata in the filesystem. */
405 static int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
406 					     struct list_head *meta_list)
407 {
408 	struct ext4_group_desc *gdp;
409 	ext4_group_t agno;
410 	int error;
411 
412 	INIT_LIST_HEAD(meta_list);
413 
414 	/* Collect everything. */
415 	for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
416 		gdp = ext4_get_group_desc(sb, agno, NULL);
417 		if (!gdp) {
418 			error = -EFSCORRUPTED;
419 			goto err;
420 		}
421 
422 		/* Superblock & GDT */
423 		error = ext4_getfsmap_find_sb(sb, agno, meta_list);
424 		if (error)
425 			goto err;
426 
427 		/* Block bitmap */
428 		error = ext4_getfsmap_fill(meta_list,
429 					   ext4_block_bitmap(sb, gdp), 1,
430 					   EXT4_FMR_OWN_BLKBM);
431 		if (error)
432 			goto err;
433 
434 		/* Inode bitmap */
435 		error = ext4_getfsmap_fill(meta_list,
436 					   ext4_inode_bitmap(sb, gdp), 1,
437 					   EXT4_FMR_OWN_INOBM);
438 		if (error)
439 			goto err;
440 
441 		/* Inodes */
442 		error = ext4_getfsmap_fill(meta_list,
443 					   ext4_inode_table(sb, gdp),
444 					   EXT4_SB(sb)->s_itb_per_group,
445 					   EXT4_FMR_OWN_INODES);
446 		if (error)
447 			goto err;
448 	}
449 
450 	/* Sort the list */
451 	list_sort(NULL, meta_list, ext4_getfsmap_compare);
452 
453 	/* Merge adjacent extents */
454 	ext4_getfsmap_merge_fixed_metadata(meta_list);
455 
456 	return 0;
457 err:
458 	ext4_getfsmap_free_fixed_metadata(meta_list);
459 	return error;
460 }
461 
462 /* Execute a getfsmap query against the buddy bitmaps */
463 static int ext4_getfsmap_datadev(struct super_block *sb,
464 				 struct ext4_fsmap *keys,
465 				 struct ext4_getfsmap_info *info)
466 {
467 	struct ext4_sb_info *sbi = EXT4_SB(sb);
468 	ext4_fsblk_t start_fsb;
469 	ext4_fsblk_t end_fsb;
470 	ext4_fsblk_t bofs;
471 	ext4_fsblk_t eofs;
472 	ext4_group_t start_ag;
473 	ext4_group_t end_ag;
474 	ext4_grpblk_t first_cluster;
475 	ext4_grpblk_t last_cluster;
476 	int error = 0;
477 
478 	bofs = le32_to_cpu(sbi->s_es->s_first_data_block);
479 	eofs = ext4_blocks_count(sbi->s_es);
480 	if (keys[0].fmr_physical >= eofs)
481 		return 0;
482 	else if (keys[0].fmr_physical < bofs)
483 		keys[0].fmr_physical = bofs;
484 	if (keys[1].fmr_physical >= eofs)
485 		keys[1].fmr_physical = eofs - 1;
486 	start_fsb = keys[0].fmr_physical;
487 	end_fsb = keys[1].fmr_physical;
488 
489 	/* Determine first and last group to examine based on start and end */
490 	ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
491 	ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
492 
493 	/*
494 	 * Convert the fsmap low/high keys to bg based keys.  Initialize
495 	 * low to the fsmap low key and max out the high key to the end
496 	 * of the bg.
497 	 */
498 	info->gfi_low = keys[0];
499 	info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
500 	info->gfi_low.fmr_length = 0;
501 
502 	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
503 
504 	/* Assemble a list of all the fixed-location metadata. */
505 	error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
506 	if (error)
507 		goto err;
508 
509 	/* Query each bg */
510 	for (info->gfi_agno = start_ag;
511 	     info->gfi_agno <= end_ag;
512 	     info->gfi_agno++) {
513 		/*
514 		 * Set the bg high key from the fsmap high key if this
515 		 * is the last bg that we're querying.
516 		 */
517 		if (info->gfi_agno == end_ag) {
518 			info->gfi_high = keys[1];
519 			info->gfi_high.fmr_physical = EXT4_C2B(sbi,
520 					last_cluster);
521 			info->gfi_high.fmr_length = 0;
522 		}
523 
524 		trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
525 				info->gfi_low.fmr_physical,
526 				info->gfi_low.fmr_length,
527 				info->gfi_low.fmr_owner);
528 
529 		trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
530 				info->gfi_high.fmr_physical,
531 				info->gfi_high.fmr_length,
532 				info->gfi_high.fmr_owner);
533 
534 		error = ext4_mballoc_query_range(sb, info->gfi_agno,
535 				EXT4_B2C(sbi, info->gfi_low.fmr_physical),
536 				EXT4_B2C(sbi, info->gfi_high.fmr_physical),
537 				ext4_getfsmap_datadev_helper, info);
538 		if (error)
539 			goto err;
540 
541 		/*
542 		 * Set the bg low key to the start of the bg prior to
543 		 * moving on to the next bg.
544 		 */
545 		if (info->gfi_agno == start_ag)
546 			memset(&info->gfi_low, 0, sizeof(info->gfi_low));
547 	}
548 
549 	/* Do we have a retained free extent? */
550 	if (info->gfi_lastfree.fmr_owner) {
551 		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
552 		if (error)
553 			goto err;
554 	}
555 
556 	/* Report any gaps at the end of the bg */
557 	info->gfi_last = true;
558 	error = ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster, 0, info);
559 	if (error)
560 		goto err;
561 
562 err:
563 	ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
564 	return error;
565 }
566 
567 /* Do we recognize the device? */
568 static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
569 					  struct ext4_fsmap *fm)
570 {
571 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
572 	    fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
573 		return true;
574 	if (EXT4_SB(sb)->journal_bdev &&
575 	    fm->fmr_device == new_encode_dev(EXT4_SB(sb)->journal_bdev->bd_dev))
576 		return true;
577 	return false;
578 }
579 
580 /* Ensure that the low key is less than the high key. */
581 static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
582 				     struct ext4_fsmap *high_key)
583 {
584 	if (low_key->fmr_device > high_key->fmr_device)
585 		return false;
586 	if (low_key->fmr_device < high_key->fmr_device)
587 		return true;
588 
589 	if (low_key->fmr_physical > high_key->fmr_physical)
590 		return false;
591 	if (low_key->fmr_physical < high_key->fmr_physical)
592 		return true;
593 
594 	if (low_key->fmr_owner > high_key->fmr_owner)
595 		return false;
596 	if (low_key->fmr_owner < high_key->fmr_owner)
597 		return true;
598 
599 	return false;
600 }
601 
602 #define EXT4_GETFSMAP_DEVS	2
603 /*
604  * Get filesystem's extents as described in head, and format for
605  * output.  Calls formatter to fill the user's buffer until all
606  * extents are mapped, until the passed-in head->fmh_count slots have
607  * been filled, or until the formatter short-circuits the loop, if it
608  * is tracking filled-in extents on its own.
609  *
610  * Key to Confusion
611  * ----------------
612  * There are multiple levels of keys and counters at work here:
613  * _fsmap_head.fmh_keys		-- low and high fsmap keys passed in;
614  * 				   these reflect fs-wide block addrs.
615  * dkeys			-- fmh_keys used to query each device;
616  * 				   these are fmh_keys but w/ the low key
617  * 				   bumped up by fmr_length.
618  * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
619  *				   is how we detect gaps in the fsmap
620  *				   records and report them.
621  * _getfsmap_info.gfi_low/high	-- per-bg low/high keys computed from
622  * 				   dkeys; used to query the free space.
623  */
624 int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
625 		  ext4_fsmap_format_t formatter, void *arg)
626 {
627 	struct ext4_fsmap dkeys[2];	/* per-dev keys */
628 	struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
629 	struct ext4_getfsmap_info info = { NULL };
630 	int i;
631 	int error = 0;
632 
633 	if (head->fmh_iflags & ~FMH_IF_VALID)
634 		return -EINVAL;
635 	if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
636 	    !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
637 		return -EINVAL;
638 
639 	head->fmh_entries = 0;
640 
641 	/* Set up our device handlers. */
642 	memset(handlers, 0, sizeof(handlers));
643 	handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
644 	handlers[0].gfd_fn = ext4_getfsmap_datadev;
645 	if (EXT4_SB(sb)->journal_bdev) {
646 		handlers[1].gfd_dev = new_encode_dev(
647 				EXT4_SB(sb)->journal_bdev->bd_dev);
648 		handlers[1].gfd_fn = ext4_getfsmap_logdev;
649 	}
650 
651 	sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
652 			ext4_getfsmap_dev_compare, NULL);
653 
654 	/*
655 	 * To continue where we left off, we allow userspace to use the
656 	 * last mapping from a previous call as the low key of the next.
657 	 * This is identified by a non-zero length in the low key. We
658 	 * have to increment the low key in this scenario to ensure we
659 	 * don't return the same mapping again, and instead return the
660 	 * very next mapping.
661 	 *
662 	 * Bump the physical offset as there can be no other mapping for
663 	 * the same physical block range.
664 	 */
665 	dkeys[0] = head->fmh_keys[0];
666 	dkeys[0].fmr_physical += dkeys[0].fmr_length;
667 	dkeys[0].fmr_owner = 0;
668 	dkeys[0].fmr_length = 0;
669 	memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
670 
671 	if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
672 		return -EINVAL;
673 
674 	info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
675 			  head->fmh_keys[0].fmr_length;
676 	info.gfi_formatter = formatter;
677 	info.gfi_format_arg = arg;
678 	info.gfi_head = head;
679 
680 	/* For each device we support... */
681 	for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
682 		/* Is this device within the range the user asked for? */
683 		if (!handlers[i].gfd_fn)
684 			continue;
685 		if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
686 			continue;
687 		if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
688 			break;
689 
690 		/*
691 		 * If this device number matches the high key, we have
692 		 * to pass the high key to the handler to limit the
693 		 * query results.  If the device number exceeds the
694 		 * low key, zero out the low key so that we get
695 		 * everything from the beginning.
696 		 */
697 		if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
698 			dkeys[1] = head->fmh_keys[1];
699 		if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
700 			memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
701 
702 		info.gfi_dev = handlers[i].gfd_dev;
703 		info.gfi_last = false;
704 		info.gfi_agno = -1;
705 		error = handlers[i].gfd_fn(sb, dkeys, &info);
706 		if (error)
707 			break;
708 		info.gfi_next_fsblk = 0;
709 	}
710 
711 	head->fmh_oflags = FMH_OF_DEV_T;
712 	return error;
713 }
714