fsmap.c

  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. */
 16void 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. */
 31void 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 */
 42struct 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. */
 57struct 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. */
 65static 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 */
 74static bool ext4_getfsmap_rec_before_low_key(struct ext4_getfsmap_info *info,
 75					     struct ext4_fsmap *rec)
 76{
 77	return rec->fmr_physical + rec->fmr_length <=
 78	       info->gfi_low.fmr_physical;
 79}
 80
 81/*
 82 * Format a reverse mapping for getfsmap, having translated rm_startblock
 83 * into the appropriate daddr units.
 84 */
 85static int ext4_getfsmap_helper(struct super_block *sb,
 86				struct ext4_getfsmap_info *info,
 87				struct ext4_fsmap *rec)
 88{
 89	struct ext4_fsmap fmr;
 90	struct ext4_sb_info *sbi = EXT4_SB(sb);
 91	ext4_fsblk_t rec_fsblk = rec->fmr_physical;
 92	ext4_group_t agno;
 93	ext4_grpblk_t cno;
 94	int error;
 95
 96	if (fatal_signal_pending(current))
 97		return -EINTR;
 98
 99	/*
100	 * Filter out records that start before our startpoint, if the
101	 * caller requested that.
102	 */
103	if (ext4_getfsmap_rec_before_low_key(info, rec)) {
104		rec_fsblk += rec->fmr_length;
105		if (info->gfi_next_fsblk < rec_fsblk)
106			info->gfi_next_fsblk = rec_fsblk;
107		return EXT4_QUERY_RANGE_CONTINUE;
108	}
109
110	/* Are we just counting mappings? */
111	if (info->gfi_head->fmh_count == 0) {
112		if (info->gfi_head->fmh_entries == UINT_MAX)
113			return EXT4_QUERY_RANGE_ABORT;
114
115		if (rec_fsblk > info->gfi_next_fsblk)
116			info->gfi_head->fmh_entries++;
117
118		if (info->gfi_last)
119			return EXT4_QUERY_RANGE_CONTINUE;
120
121		info->gfi_head->fmh_entries++;
122
123		rec_fsblk += rec->fmr_length;
124		if (info->gfi_next_fsblk < rec_fsblk)
125			info->gfi_next_fsblk = rec_fsblk;
126		return EXT4_QUERY_RANGE_CONTINUE;
127	}
128
129	/*
130	 * If the record starts past the last physical block we saw,
131	 * then we've found a gap.  Report the gap as being owned by
132	 * whatever the caller specified is the missing owner.
133	 */
134	if (rec_fsblk > info->gfi_next_fsblk) {
135		if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
136			return EXT4_QUERY_RANGE_ABORT;
137
138		ext4_get_group_no_and_offset(sb, info->gfi_next_fsblk,
139				&agno, &cno);
140		trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno,
141				EXT4_C2B(sbi, cno),
142				rec_fsblk - info->gfi_next_fsblk,
143				EXT4_FMR_OWN_UNKNOWN);
144
145		fmr.fmr_device = info->gfi_dev;
146		fmr.fmr_physical = info->gfi_next_fsblk;
147		fmr.fmr_owner = EXT4_FMR_OWN_UNKNOWN;
148		fmr.fmr_length = rec_fsblk - info->gfi_next_fsblk;
149		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
150		error = info->gfi_formatter(&fmr, info->gfi_format_arg);
151		if (error)
152			return error;
153		info->gfi_head->fmh_entries++;
154	}
155
156	if (info->gfi_last)
157		goto out;
158
159	/* Fill out the extent we found */
160	if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
161		return EXT4_QUERY_RANGE_ABORT;
162
163	ext4_get_group_no_and_offset(sb, rec_fsblk, &agno, &cno);
164	trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno, EXT4_C2B(sbi, cno),
165			rec->fmr_length, rec->fmr_owner);
166
167	fmr.fmr_device = info->gfi_dev;
168	fmr.fmr_physical = rec_fsblk;
169	fmr.fmr_owner = rec->fmr_owner;
170	fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
171	fmr.fmr_length = rec->fmr_length;
172	error = info->gfi_formatter(&fmr, info->gfi_format_arg);
173	if (error)
174		return error;
175	info->gfi_head->fmh_entries++;
176
177out:
178	rec_fsblk += rec->fmr_length;
179	if (info->gfi_next_fsblk < rec_fsblk)
180		info->gfi_next_fsblk = rec_fsblk;
181	return EXT4_QUERY_RANGE_CONTINUE;
182}
183
184static inline ext4_fsblk_t ext4_fsmap_next_pblk(struct ext4_fsmap *fmr)
185{
186	return fmr->fmr_physical + fmr->fmr_length;
187}
188
189static int ext4_getfsmap_meta_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_getfsmap_info *info = priv;
194	struct ext4_fsmap *p;
195	struct ext4_fsmap *tmp;
196	struct ext4_sb_info *sbi = EXT4_SB(sb);
197	ext4_fsblk_t fsb, fs_start, fs_end;
198	int error;
199
200	fs_start = fsb = (EXT4_C2B(sbi, start) +
201			  ext4_group_first_block_no(sb, agno));
202	fs_end = fs_start + EXT4_C2B(sbi, len);
203
204	/*
205	 * Return relevant extents from the meta_list. We emit all extents that
206	 * partially/fully overlap with the query range
207	 */
208	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
209		if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
210			list_del(&p->fmr_list);
211			kfree(p);
212			continue;
213		}
214		if (p->fmr_physical <= fs_end &&
215		    p->fmr_physical + p->fmr_length > fs_start) {
216			/* Emit the retained free extent record if present */
217			if (info->gfi_lastfree.fmr_owner) {
218				error = ext4_getfsmap_helper(sb, info,
219							&info->gfi_lastfree);
220				if (error)
221					return error;
222				info->gfi_lastfree.fmr_owner = 0;
223			}
224			error = ext4_getfsmap_helper(sb, info, p);
225			if (error)
226				return error;
227			fsb = p->fmr_physical + p->fmr_length;
228			if (info->gfi_next_fsblk < fsb)
229				info->gfi_next_fsblk = fsb;
230			list_del(&p->fmr_list);
231			kfree(p);
232			continue;
233		}
234	}
235	if (info->gfi_next_fsblk < fsb)
236		info->gfi_next_fsblk = fsb;
237
238	return 0;
239}
240
241
242/* Transform a blockgroup's free record into a fsmap */
243static int ext4_getfsmap_datadev_helper(struct super_block *sb,
244					ext4_group_t agno, ext4_grpblk_t start,
245					ext4_grpblk_t len, void *priv)
246{
247	struct ext4_fsmap irec;
248	struct ext4_getfsmap_info *info = priv;
249	struct ext4_fsmap *p;
250	struct ext4_fsmap *tmp;
251	struct ext4_sb_info *sbi = EXT4_SB(sb);
252	ext4_fsblk_t fsb;
253	ext4_fsblk_t fslen;
254	int error;
255
256	fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
257	fslen = EXT4_C2B(sbi, len);
258
259	/* If the retained free extent record is set... */
260	if (info->gfi_lastfree.fmr_owner) {
261		/* ...and abuts this one, lengthen it and return. */
262		if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
263			info->gfi_lastfree.fmr_length += fslen;
264			return 0;
265		}
266
267		/*
268		 * There's a gap between the two free extents; emit the
269		 * retained extent prior to merging the meta_list.
270		 */
271		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
272		if (error)
273			return error;
274		info->gfi_lastfree.fmr_owner = 0;
275	}
276
277	/* Merge in any relevant extents from the meta_list */
278	list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
279		if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
280			list_del(&p->fmr_list);
281			kfree(p);
282		} else if (p->fmr_physical < fsb) {
283			error = ext4_getfsmap_helper(sb, info, p);
284			if (error)
285				return error;
286
287			list_del(&p->fmr_list);
288			kfree(p);
289		}
290	}
291
292	irec.fmr_device = 0;
293	irec.fmr_physical = fsb;
294	irec.fmr_length = fslen;
295	irec.fmr_owner = EXT4_FMR_OWN_FREE;
296	irec.fmr_flags = 0;
297
298	/* If this is a free extent at the end of a bg, buffer it. */
299	if (ext4_fsmap_next_pblk(&irec) ==
300			ext4_group_first_block_no(sb, agno + 1)) {
301		info->gfi_lastfree = irec;
302		return 0;
303	}
304
305	/* Otherwise, emit it */
306	return ext4_getfsmap_helper(sb, info, &irec);
307}
308
309/* Execute a getfsmap query against the log device. */
310static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
311				struct ext4_getfsmap_info *info)
312{
313	journal_t *journal = EXT4_SB(sb)->s_journal;
314	struct ext4_fsmap irec;
315
316	/* Set up search keys */
317	info->gfi_low = keys[0];
318	info->gfi_low.fmr_length = 0;
319
320	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
321
322	trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
323			info->gfi_low.fmr_physical,
324			info->gfi_low.fmr_length,
325			info->gfi_low.fmr_owner);
326
327	trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
328			info->gfi_high.fmr_physical,
329			info->gfi_high.fmr_length,
330			info->gfi_high.fmr_owner);
331
332	if (keys[0].fmr_physical > 0)
333		return 0;
334
335	/* Fabricate an rmap entry for the external log device. */
336	irec.fmr_physical = journal->j_blk_offset;
337	irec.fmr_length = journal->j_total_len;
338	irec.fmr_owner = EXT4_FMR_OWN_LOG;
339	irec.fmr_flags = 0;
340
341	return ext4_getfsmap_helper(sb, info, &irec);
342}
343
344/* Helper to fill out an ext4_fsmap. */
345static inline int ext4_getfsmap_fill(struct list_head *meta_list,
346				     ext4_fsblk_t fsb, ext4_fsblk_t len,
347				     uint64_t owner)
348{
349	struct ext4_fsmap *fsm;
350
351	fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
352	if (!fsm)
353		return -ENOMEM;
354	fsm->fmr_device = 0;
355	fsm->fmr_flags = 0;
356	fsm->fmr_physical = fsb;
357	fsm->fmr_owner = owner;
358	fsm->fmr_length = len;
359	list_add_tail(&fsm->fmr_list, meta_list);
360
361	return 0;
362}
363
364/*
365 * This function returns the number of file system metadata blocks at
366 * the beginning of a block group, including the reserved gdt blocks.
367 */
368static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
369					  ext4_group_t agno,
370					  struct list_head *meta_list)
371{
372	struct ext4_sb_info *sbi = EXT4_SB(sb);
373	ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
374	ext4_fsblk_t len;
375	unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
376	unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
377	int error;
378
379	/* Record the superblock. */
380	if (ext4_bg_has_super(sb, agno)) {
381		error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
382		if (error)
383			return error;
384		fsb++;
385	}
386
387	/* Record the group descriptors. */
388	len = ext4_bg_num_gdb(sb, agno);
389	if (!len)
390		return 0;
391	error = ext4_getfsmap_fill(meta_list, fsb, len,
392				   EXT4_FMR_OWN_GDT);
393	if (error)
394		return error;
395	fsb += len;
396
397	/* Reserved GDT blocks */
398	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
399		len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
400
401		/*
402		 * mkfs.ext4 can set s_reserved_gdt_blocks as 0 in some cases,
403		 * check for that.
404		 */
405		if (!len)
406			return 0;
407
408		error = ext4_getfsmap_fill(meta_list, fsb, len,
409					   EXT4_FMR_OWN_RESV_GDT);
410		if (error)
411			return error;
412	}
413
414	return 0;
415}
416
417/* Compare two fsmap items. */
418static int ext4_getfsmap_compare(void *priv,
419				 const struct list_head *a,
420				 const struct list_head *b)
421{
422	struct ext4_fsmap *fa;
423	struct ext4_fsmap *fb;
424
425	fa = container_of(a, struct ext4_fsmap, fmr_list);
426	fb = container_of(b, struct ext4_fsmap, fmr_list);
427	if (fa->fmr_physical < fb->fmr_physical)
428		return -1;
429	else if (fa->fmr_physical > fb->fmr_physical)
430		return 1;
431	return 0;
432}
433
434/* Merge adjacent extents of fixed metadata. */
435static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
436{
437	struct ext4_fsmap *p;
438	struct ext4_fsmap *prev = NULL;
439	struct ext4_fsmap *tmp;
440
441	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
442		if (!prev) {
443			prev = p;
444			continue;
445		}
446
447		if (prev->fmr_owner == p->fmr_owner &&
448		    prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
449			prev->fmr_length += p->fmr_length;
450			list_del(&p->fmr_list);
451			kfree(p);
452		} else
453			prev = p;
454	}
455}
456
457/* Free a list of fixed metadata. */
458static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
459{
460	struct ext4_fsmap *p;
461	struct ext4_fsmap *tmp;
462
463	list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
464		list_del(&p->fmr_list);
465		kfree(p);
466	}
467}
468
469/* Find all the fixed metadata in the filesystem. */
470static int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
471					     struct list_head *meta_list)
472{
473	struct ext4_group_desc *gdp;
474	ext4_group_t agno;
475	int error;
476
477	INIT_LIST_HEAD(meta_list);
478
479	/* Collect everything. */
480	for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
481		gdp = ext4_get_group_desc(sb, agno, NULL);
482		if (!gdp) {
483			error = -EFSCORRUPTED;
484			goto err;
485		}
486
487		/* Superblock & GDT */
488		error = ext4_getfsmap_find_sb(sb, agno, meta_list);
489		if (error)
490			goto err;
491
492		/* Block bitmap */
493		error = ext4_getfsmap_fill(meta_list,
494					   ext4_block_bitmap(sb, gdp), 1,
495					   EXT4_FMR_OWN_BLKBM);
496		if (error)
497			goto err;
498
499		/* Inode bitmap */
500		error = ext4_getfsmap_fill(meta_list,
501					   ext4_inode_bitmap(sb, gdp), 1,
502					   EXT4_FMR_OWN_INOBM);
503		if (error)
504			goto err;
505
506		/* Inodes */
507		error = ext4_getfsmap_fill(meta_list,
508					   ext4_inode_table(sb, gdp),
509					   EXT4_SB(sb)->s_itb_per_group,
510					   EXT4_FMR_OWN_INODES);
511		if (error)
512			goto err;
513	}
514
515	/* Sort the list */
516	list_sort(NULL, meta_list, ext4_getfsmap_compare);
517
518	/* Merge adjacent extents */
519	ext4_getfsmap_merge_fixed_metadata(meta_list);
520
521	return 0;
522err:
523	ext4_getfsmap_free_fixed_metadata(meta_list);
524	return error;
525}
526
527/* Execute a getfsmap query against the buddy bitmaps */
528static int ext4_getfsmap_datadev(struct super_block *sb,
529				 struct ext4_fsmap *keys,
530				 struct ext4_getfsmap_info *info)
531{
532	struct ext4_sb_info *sbi = EXT4_SB(sb);
533	ext4_fsblk_t start_fsb;
534	ext4_fsblk_t end_fsb;
535	ext4_fsblk_t bofs;
536	ext4_fsblk_t eofs;
537	ext4_group_t start_ag;
538	ext4_group_t end_ag;
539	ext4_grpblk_t first_cluster;
540	ext4_grpblk_t last_cluster;
541	struct ext4_fsmap irec;
542	int error = 0;
543
544	bofs = le32_to_cpu(sbi->s_es->s_first_data_block);
545	eofs = ext4_blocks_count(sbi->s_es);
546	if (keys[0].fmr_physical >= eofs)
547		return 0;
548	else if (keys[0].fmr_physical < bofs)
549		keys[0].fmr_physical = bofs;
550	if (keys[1].fmr_physical >= eofs)
551		keys[1].fmr_physical = eofs - 1;
552	if (keys[1].fmr_physical < keys[0].fmr_physical)
553		return 0;
554	start_fsb = keys[0].fmr_physical;
555	end_fsb = keys[1].fmr_physical;
556
557	/* Determine first and last group to examine based on start and end */
558	ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
559	ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
560
561	/*
562	 * Convert the fsmap low/high keys to bg based keys.  Initialize
563	 * low to the fsmap low key and max out the high key to the end
564	 * of the bg.
565	 */
566	info->gfi_low = keys[0];
567	info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
568	info->gfi_low.fmr_length = 0;
569
570	memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
571
572	/* Assemble a list of all the fixed-location metadata. */
573	error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
574	if (error)
575		goto err;
576
577	/* Query each bg */
578	for (info->gfi_agno = start_ag;
579	     info->gfi_agno <= end_ag;
580	     info->gfi_agno++) {
581		/*
582		 * Set the bg high key from the fsmap high key if this
583		 * is the last bg that we're querying.
584		 */
585		if (info->gfi_agno == end_ag) {
586			info->gfi_high = keys[1];
587			info->gfi_high.fmr_physical = EXT4_C2B(sbi,
588					last_cluster);
589			info->gfi_high.fmr_length = 0;
590		}
591
592		trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
593				info->gfi_low.fmr_physical,
594				info->gfi_low.fmr_length,
595				info->gfi_low.fmr_owner);
596
597		trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
598				info->gfi_high.fmr_physical,
599				info->gfi_high.fmr_length,
600				info->gfi_high.fmr_owner);
601
602		error = ext4_mballoc_query_range(sb, info->gfi_agno,
603				EXT4_B2C(sbi, info->gfi_low.fmr_physical),
604				EXT4_B2C(sbi, info->gfi_high.fmr_physical),
605				ext4_getfsmap_meta_helper,
606				ext4_getfsmap_datadev_helper, info);
607		if (error)
608			goto err;
609
610		/*
611		 * Set the bg low key to the start of the bg prior to
612		 * moving on to the next bg.
613		 */
614		if (info->gfi_agno == start_ag)
615			memset(&info->gfi_low, 0, sizeof(info->gfi_low));
616	}
617
618	/* Do we have a retained free extent? */
619	if (info->gfi_lastfree.fmr_owner) {
620		error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
621		if (error)
622			goto err;
623	}
624
625	/*
626	 * The dummy record below will cause ext4_getfsmap_helper() to report
627	 * any allocated blocks at the end of the range.
628	 */
629	irec.fmr_device = 0;
630	irec.fmr_physical = end_fsb + 1;
631	irec.fmr_length = 0;
632	irec.fmr_owner = EXT4_FMR_OWN_FREE;
633	irec.fmr_flags = 0;
634
635	info->gfi_last = true;
636	error = ext4_getfsmap_helper(sb, info, &irec);
637	if (error)
638		goto err;
639
640err:
641	ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
642	return error;
643}
644
645/* Do we recognize the device? */
646static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
647					  struct ext4_fsmap *fm)
648{
649	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
650	    fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
651		return true;
652	if (EXT4_SB(sb)->s_journal_bdev_file &&
653	    fm->fmr_device ==
654	    new_encode_dev(file_bdev(EXT4_SB(sb)->s_journal_bdev_file)->bd_dev))
655		return true;
656	return false;
657}
658
659/* Ensure that the low key is less than the high key. */
660static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
661				     struct ext4_fsmap *high_key)
662{
663	if (low_key->fmr_device > high_key->fmr_device)
664		return false;
665	if (low_key->fmr_device < high_key->fmr_device)
666		return true;
667
668	if (low_key->fmr_physical > high_key->fmr_physical)
669		return false;
670	if (low_key->fmr_physical < high_key->fmr_physical)
671		return true;
672
673	if (low_key->fmr_owner > high_key->fmr_owner)
674		return false;
675	if (low_key->fmr_owner < high_key->fmr_owner)
676		return true;
677
678	return false;
679}
680
681#define EXT4_GETFSMAP_DEVS	2
682/*
683 * Get filesystem's extents as described in head, and format for
684 * output.  Calls formatter to fill the user's buffer until all
685 * extents are mapped, until the passed-in head->fmh_count slots have
686 * been filled, or until the formatter short-circuits the loop, if it
687 * is tracking filled-in extents on its own.
688 *
689 * Key to Confusion
690 * ----------------
691 * There are multiple levels of keys and counters at work here:
692 * _fsmap_head.fmh_keys		-- low and high fsmap keys passed in;
693 * 				   these reflect fs-wide block addrs.
694 * dkeys			-- fmh_keys used to query each device;
695 * 				   these are fmh_keys but w/ the low key
696 * 				   bumped up by fmr_length.
697 * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
698 *				   is how we detect gaps in the fsmap
699 *				   records and report them.
700 * _getfsmap_info.gfi_low/high	-- per-bg low/high keys computed from
701 * 				   dkeys; used to query the free space.
702 */
703int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
704		  ext4_fsmap_format_t formatter, void *arg)
705{
706	struct ext4_fsmap dkeys[2];	/* per-dev keys */
707	struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
708	struct ext4_getfsmap_info info = { NULL };
709	int i;
710	int error = 0;
711
712	if (head->fmh_iflags & ~FMH_IF_VALID)
713		return -EINVAL;
714	if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
715	    !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
716		return -EINVAL;
717
718	head->fmh_entries = 0;
719
720	/* Set up our device handlers. */
721	memset(handlers, 0, sizeof(handlers));
722	handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
723	handlers[0].gfd_fn = ext4_getfsmap_datadev;
724	if (EXT4_SB(sb)->s_journal_bdev_file) {
725		handlers[1].gfd_dev = new_encode_dev(
726			file_bdev(EXT4_SB(sb)->s_journal_bdev_file)->bd_dev);
727		handlers[1].gfd_fn = ext4_getfsmap_logdev;
728	}
729
730	sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
731			ext4_getfsmap_dev_compare, NULL);
732
733	/*
734	 * To continue where we left off, we allow userspace to use the
735	 * last mapping from a previous call as the low key of the next.
736	 * This is identified by a non-zero length in the low key. We
737	 * have to increment the low key in this scenario to ensure we
738	 * don't return the same mapping again, and instead return the
739	 * very next mapping.
740	 *
741	 * Bump the physical offset as there can be no other mapping for
742	 * the same physical block range.
743	 */
744	dkeys[0] = head->fmh_keys[0];
745	dkeys[0].fmr_physical += dkeys[0].fmr_length;
746	dkeys[0].fmr_owner = 0;
747	dkeys[0].fmr_length = 0;
748	memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
749
750	if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
751		return -EINVAL;
752
753	info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
754			  head->fmh_keys[0].fmr_length;
755	info.gfi_formatter = formatter;
756	info.gfi_format_arg = arg;
757	info.gfi_head = head;
758
759	/* For each device we support... */
760	for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
761		/* Is this device within the range the user asked for? */
762		if (!handlers[i].gfd_fn)
763			continue;
764		if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
765			continue;
766		if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
767			break;
768
769		/*
770		 * If this device number matches the high key, we have
771		 * to pass the high key to the handler to limit the
772		 * query results.  If the device number exceeds the
773		 * low key, zero out the low key so that we get
774		 * everything from the beginning.
775		 */
776		if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
777			dkeys[1] = head->fmh_keys[1];
778		if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
779			memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
780
781		info.gfi_dev = handlers[i].gfd_dev;
782		info.gfi_last = false;
783		info.gfi_agno = -1;
784		error = handlers[i].gfd_fn(sb, dkeys, &info);
785		if (error)
786			break;
787		info.gfi_next_fsblk = 0;
788	}
789
790	head->fmh_oflags = FMH_OF_DEV_T;
791	return error;
792}