xref: /openbmc/linux/fs/xfs/xfs_fsmap.c (revision da60fbe7)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 2017 Oracle.  All Rights Reserved.
4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_btree.h"
16 #include "xfs_rmap_btree.h"
17 #include "xfs_trace.h"
18 #include "xfs_rmap.h"
19 #include "xfs_alloc.h"
20 #include "xfs_bit.h"
21 #include <linux/fsmap.h>
22 #include "xfs_fsmap.h"
23 #include "xfs_refcount.h"
24 #include "xfs_refcount_btree.h"
25 #include "xfs_alloc_btree.h"
26 #include "xfs_rtalloc.h"
27 
28 /* Convert an xfs_fsmap to an fsmap. */
29 void
30 xfs_fsmap_from_internal(
31 	struct fsmap		*dest,
32 	struct xfs_fsmap	*src)
33 {
34 	dest->fmr_device = src->fmr_device;
35 	dest->fmr_flags = src->fmr_flags;
36 	dest->fmr_physical = BBTOB(src->fmr_physical);
37 	dest->fmr_owner = src->fmr_owner;
38 	dest->fmr_offset = BBTOB(src->fmr_offset);
39 	dest->fmr_length = BBTOB(src->fmr_length);
40 	dest->fmr_reserved[0] = 0;
41 	dest->fmr_reserved[1] = 0;
42 	dest->fmr_reserved[2] = 0;
43 }
44 
45 /* Convert an fsmap to an xfs_fsmap. */
46 void
47 xfs_fsmap_to_internal(
48 	struct xfs_fsmap	*dest,
49 	struct fsmap		*src)
50 {
51 	dest->fmr_device = src->fmr_device;
52 	dest->fmr_flags = src->fmr_flags;
53 	dest->fmr_physical = BTOBBT(src->fmr_physical);
54 	dest->fmr_owner = src->fmr_owner;
55 	dest->fmr_offset = BTOBBT(src->fmr_offset);
56 	dest->fmr_length = BTOBBT(src->fmr_length);
57 }
58 
59 /* Convert an fsmap owner into an rmapbt owner. */
60 static int
61 xfs_fsmap_owner_to_rmap(
62 	struct xfs_rmap_irec	*dest,
63 	struct xfs_fsmap	*src)
64 {
65 	if (!(src->fmr_flags & FMR_OF_SPECIAL_OWNER)) {
66 		dest->rm_owner = src->fmr_owner;
67 		return 0;
68 	}
69 
70 	switch (src->fmr_owner) {
71 	case 0:			/* "lowest owner id possible" */
72 	case -1ULL:		/* "highest owner id possible" */
73 		dest->rm_owner = 0;
74 		break;
75 	case XFS_FMR_OWN_FREE:
76 		dest->rm_owner = XFS_RMAP_OWN_NULL;
77 		break;
78 	case XFS_FMR_OWN_UNKNOWN:
79 		dest->rm_owner = XFS_RMAP_OWN_UNKNOWN;
80 		break;
81 	case XFS_FMR_OWN_FS:
82 		dest->rm_owner = XFS_RMAP_OWN_FS;
83 		break;
84 	case XFS_FMR_OWN_LOG:
85 		dest->rm_owner = XFS_RMAP_OWN_LOG;
86 		break;
87 	case XFS_FMR_OWN_AG:
88 		dest->rm_owner = XFS_RMAP_OWN_AG;
89 		break;
90 	case XFS_FMR_OWN_INOBT:
91 		dest->rm_owner = XFS_RMAP_OWN_INOBT;
92 		break;
93 	case XFS_FMR_OWN_INODES:
94 		dest->rm_owner = XFS_RMAP_OWN_INODES;
95 		break;
96 	case XFS_FMR_OWN_REFC:
97 		dest->rm_owner = XFS_RMAP_OWN_REFC;
98 		break;
99 	case XFS_FMR_OWN_COW:
100 		dest->rm_owner = XFS_RMAP_OWN_COW;
101 		break;
102 	case XFS_FMR_OWN_DEFECTIVE:	/* not implemented */
103 		/* fall through */
104 	default:
105 		return -EINVAL;
106 	}
107 	return 0;
108 }
109 
110 /* Convert an rmapbt owner into an fsmap owner. */
111 static int
112 xfs_fsmap_owner_from_rmap(
113 	struct xfs_fsmap	*dest,
114 	struct xfs_rmap_irec	*src)
115 {
116 	dest->fmr_flags = 0;
117 	if (!XFS_RMAP_NON_INODE_OWNER(src->rm_owner)) {
118 		dest->fmr_owner = src->rm_owner;
119 		return 0;
120 	}
121 	dest->fmr_flags |= FMR_OF_SPECIAL_OWNER;
122 
123 	switch (src->rm_owner) {
124 	case XFS_RMAP_OWN_FS:
125 		dest->fmr_owner = XFS_FMR_OWN_FS;
126 		break;
127 	case XFS_RMAP_OWN_LOG:
128 		dest->fmr_owner = XFS_FMR_OWN_LOG;
129 		break;
130 	case XFS_RMAP_OWN_AG:
131 		dest->fmr_owner = XFS_FMR_OWN_AG;
132 		break;
133 	case XFS_RMAP_OWN_INOBT:
134 		dest->fmr_owner = XFS_FMR_OWN_INOBT;
135 		break;
136 	case XFS_RMAP_OWN_INODES:
137 		dest->fmr_owner = XFS_FMR_OWN_INODES;
138 		break;
139 	case XFS_RMAP_OWN_REFC:
140 		dest->fmr_owner = XFS_FMR_OWN_REFC;
141 		break;
142 	case XFS_RMAP_OWN_COW:
143 		dest->fmr_owner = XFS_FMR_OWN_COW;
144 		break;
145 	case XFS_RMAP_OWN_NULL:	/* "free" */
146 		dest->fmr_owner = XFS_FMR_OWN_FREE;
147 		break;
148 	default:
149 		ASSERT(0);
150 		return -EFSCORRUPTED;
151 	}
152 	return 0;
153 }
154 
155 /* getfsmap query state */
156 struct xfs_getfsmap_info {
157 	struct xfs_fsmap_head	*head;
158 	xfs_fsmap_format_t	formatter;	/* formatting fn */
159 	void			*format_arg;	/* format buffer */
160 	struct xfs_buf		*agf_bp;	/* AGF, for refcount queries */
161 	xfs_daddr_t		next_daddr;	/* next daddr we expect */
162 	u64			missing_owner;	/* owner of holes */
163 	u32			dev;		/* device id */
164 	xfs_agnumber_t		agno;		/* AG number, if applicable */
165 	struct xfs_rmap_irec	low;		/* low rmap key */
166 	struct xfs_rmap_irec	high;		/* high rmap key */
167 	bool			last;		/* last extent? */
168 };
169 
170 /* Associate a device with a getfsmap handler. */
171 struct xfs_getfsmap_dev {
172 	u32			dev;
173 	int			(*fn)(struct xfs_trans *tp,
174 				      struct xfs_fsmap *keys,
175 				      struct xfs_getfsmap_info *info);
176 };
177 
178 /* Compare two getfsmap device handlers. */
179 static int
180 xfs_getfsmap_dev_compare(
181 	const void			*p1,
182 	const void			*p2)
183 {
184 	const struct xfs_getfsmap_dev	*d1 = p1;
185 	const struct xfs_getfsmap_dev	*d2 = p2;
186 
187 	return d1->dev - d2->dev;
188 }
189 
190 /* Decide if this mapping is shared. */
191 STATIC int
192 xfs_getfsmap_is_shared(
193 	struct xfs_trans		*tp,
194 	struct xfs_getfsmap_info	*info,
195 	struct xfs_rmap_irec		*rec,
196 	bool				*stat)
197 {
198 	struct xfs_mount		*mp = tp->t_mountp;
199 	struct xfs_btree_cur		*cur;
200 	xfs_agblock_t			fbno;
201 	xfs_extlen_t			flen;
202 	int				error;
203 
204 	*stat = false;
205 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
206 		return 0;
207 	/* rt files will have agno set to NULLAGNUMBER */
208 	if (info->agno == NULLAGNUMBER)
209 		return 0;
210 
211 	/* Are there any shared blocks here? */
212 	flen = 0;
213 	cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp,
214 			info->agno);
215 
216 	error = xfs_refcount_find_shared(cur, rec->rm_startblock,
217 			rec->rm_blockcount, &fbno, &flen, false);
218 
219 	xfs_btree_del_cursor(cur, error);
220 	if (error)
221 		return error;
222 
223 	*stat = flen > 0;
224 	return 0;
225 }
226 
227 /*
228  * Format a reverse mapping for getfsmap, having translated rm_startblock
229  * into the appropriate daddr units.
230  */
231 STATIC int
232 xfs_getfsmap_helper(
233 	struct xfs_trans		*tp,
234 	struct xfs_getfsmap_info	*info,
235 	struct xfs_rmap_irec		*rec,
236 	xfs_daddr_t			rec_daddr)
237 {
238 	struct xfs_fsmap		fmr;
239 	struct xfs_mount		*mp = tp->t_mountp;
240 	bool				shared;
241 	int				error;
242 
243 	if (fatal_signal_pending(current))
244 		return -EINTR;
245 
246 	/*
247 	 * Filter out records that start before our startpoint, if the
248 	 * caller requested that.
249 	 */
250 	if (xfs_rmap_compare(rec, &info->low) < 0) {
251 		rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
252 		if (info->next_daddr < rec_daddr)
253 			info->next_daddr = rec_daddr;
254 		return 0;
255 	}
256 
257 	/* Are we just counting mappings? */
258 	if (info->head->fmh_count == 0) {
259 		if (rec_daddr > info->next_daddr)
260 			info->head->fmh_entries++;
261 
262 		if (info->last)
263 			return 0;
264 
265 		info->head->fmh_entries++;
266 
267 		rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
268 		if (info->next_daddr < rec_daddr)
269 			info->next_daddr = rec_daddr;
270 		return 0;
271 	}
272 
273 	/*
274 	 * If the record starts past the last physical block we saw,
275 	 * then we've found a gap.  Report the gap as being owned by
276 	 * whatever the caller specified is the missing owner.
277 	 */
278 	if (rec_daddr > info->next_daddr) {
279 		if (info->head->fmh_entries >= info->head->fmh_count)
280 			return -ECANCELED;
281 
282 		fmr.fmr_device = info->dev;
283 		fmr.fmr_physical = info->next_daddr;
284 		fmr.fmr_owner = info->missing_owner;
285 		fmr.fmr_offset = 0;
286 		fmr.fmr_length = rec_daddr - info->next_daddr;
287 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
288 		error = info->formatter(&fmr, info->format_arg);
289 		if (error)
290 			return error;
291 		info->head->fmh_entries++;
292 	}
293 
294 	if (info->last)
295 		goto out;
296 
297 	/* Fill out the extent we found */
298 	if (info->head->fmh_entries >= info->head->fmh_count)
299 		return -ECANCELED;
300 
301 	trace_xfs_fsmap_mapping(mp, info->dev, info->agno, rec);
302 
303 	fmr.fmr_device = info->dev;
304 	fmr.fmr_physical = rec_daddr;
305 	error = xfs_fsmap_owner_from_rmap(&fmr, rec);
306 	if (error)
307 		return error;
308 	fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
309 	fmr.fmr_length = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
310 	if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
311 		fmr.fmr_flags |= FMR_OF_PREALLOC;
312 	if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
313 		fmr.fmr_flags |= FMR_OF_ATTR_FORK;
314 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
315 		fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
316 	if (fmr.fmr_flags == 0) {
317 		error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
318 		if (error)
319 			return error;
320 		if (shared)
321 			fmr.fmr_flags |= FMR_OF_SHARED;
322 	}
323 	error = info->formatter(&fmr, info->format_arg);
324 	if (error)
325 		return error;
326 	info->head->fmh_entries++;
327 
328 out:
329 	rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
330 	if (info->next_daddr < rec_daddr)
331 		info->next_daddr = rec_daddr;
332 	return 0;
333 }
334 
335 /* Transform a rmapbt irec into a fsmap */
336 STATIC int
337 xfs_getfsmap_datadev_helper(
338 	struct xfs_btree_cur		*cur,
339 	struct xfs_rmap_irec		*rec,
340 	void				*priv)
341 {
342 	struct xfs_mount		*mp = cur->bc_mp;
343 	struct xfs_getfsmap_info	*info = priv;
344 	xfs_fsblock_t			fsb;
345 	xfs_daddr_t			rec_daddr;
346 
347 	fsb = XFS_AGB_TO_FSB(mp, cur->bc_private.a.agno, rec->rm_startblock);
348 	rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
349 
350 	return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr);
351 }
352 
353 /* Transform a bnobt irec into a fsmap */
354 STATIC int
355 xfs_getfsmap_datadev_bnobt_helper(
356 	struct xfs_btree_cur		*cur,
357 	struct xfs_alloc_rec_incore	*rec,
358 	void				*priv)
359 {
360 	struct xfs_mount		*mp = cur->bc_mp;
361 	struct xfs_getfsmap_info	*info = priv;
362 	struct xfs_rmap_irec		irec;
363 	xfs_daddr_t			rec_daddr;
364 
365 	rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_private.a.agno,
366 			rec->ar_startblock);
367 
368 	irec.rm_startblock = rec->ar_startblock;
369 	irec.rm_blockcount = rec->ar_blockcount;
370 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
371 	irec.rm_offset = 0;
372 	irec.rm_flags = 0;
373 
374 	return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr);
375 }
376 
377 /* Set rmap flags based on the getfsmap flags */
378 static void
379 xfs_getfsmap_set_irec_flags(
380 	struct xfs_rmap_irec	*irec,
381 	struct xfs_fsmap	*fmr)
382 {
383 	irec->rm_flags = 0;
384 	if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
385 		irec->rm_flags |= XFS_RMAP_ATTR_FORK;
386 	if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
387 		irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
388 	if (fmr->fmr_flags & FMR_OF_PREALLOC)
389 		irec->rm_flags |= XFS_RMAP_UNWRITTEN;
390 }
391 
392 /* Execute a getfsmap query against the log device. */
393 STATIC int
394 xfs_getfsmap_logdev(
395 	struct xfs_trans		*tp,
396 	struct xfs_fsmap		*keys,
397 	struct xfs_getfsmap_info	*info)
398 {
399 	struct xfs_mount		*mp = tp->t_mountp;
400 	struct xfs_rmap_irec		rmap;
401 	int				error;
402 
403 	/* Set up search keys */
404 	info->low.rm_startblock = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
405 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
406 	error = xfs_fsmap_owner_to_rmap(&info->low, keys);
407 	if (error)
408 		return error;
409 	info->low.rm_blockcount = 0;
410 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
411 
412 	error = xfs_fsmap_owner_to_rmap(&info->high, keys + 1);
413 	if (error)
414 		return error;
415 	info->high.rm_startblock = -1U;
416 	info->high.rm_owner = ULLONG_MAX;
417 	info->high.rm_offset = ULLONG_MAX;
418 	info->high.rm_blockcount = 0;
419 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
420 	info->missing_owner = XFS_FMR_OWN_FREE;
421 
422 	trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
423 	trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
424 
425 	if (keys[0].fmr_physical > 0)
426 		return 0;
427 
428 	/* Fabricate an rmap entry for the external log device. */
429 	rmap.rm_startblock = 0;
430 	rmap.rm_blockcount = mp->m_sb.sb_logblocks;
431 	rmap.rm_owner = XFS_RMAP_OWN_LOG;
432 	rmap.rm_offset = 0;
433 	rmap.rm_flags = 0;
434 
435 	return xfs_getfsmap_helper(tp, info, &rmap, 0);
436 }
437 
438 #ifdef CONFIG_XFS_RT
439 /* Transform a rtbitmap "record" into a fsmap */
440 STATIC int
441 xfs_getfsmap_rtdev_rtbitmap_helper(
442 	struct xfs_trans		*tp,
443 	struct xfs_rtalloc_rec		*rec,
444 	void				*priv)
445 {
446 	struct xfs_mount		*mp = tp->t_mountp;
447 	struct xfs_getfsmap_info	*info = priv;
448 	struct xfs_rmap_irec		irec;
449 	xfs_daddr_t			rec_daddr;
450 
451 	irec.rm_startblock = rec->ar_startext * mp->m_sb.sb_rextsize;
452 	rec_daddr = XFS_FSB_TO_BB(mp, irec.rm_startblock);
453 	irec.rm_blockcount = rec->ar_extcount * mp->m_sb.sb_rextsize;
454 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
455 	irec.rm_offset = 0;
456 	irec.rm_flags = 0;
457 
458 	return xfs_getfsmap_helper(tp, info, &irec, rec_daddr);
459 }
460 
461 /* Execute a getfsmap query against the realtime device. */
462 STATIC int
463 __xfs_getfsmap_rtdev(
464 	struct xfs_trans		*tp,
465 	struct xfs_fsmap		*keys,
466 	int				(*query_fn)(struct xfs_trans *,
467 						    struct xfs_getfsmap_info *),
468 	struct xfs_getfsmap_info	*info)
469 {
470 	struct xfs_mount		*mp = tp->t_mountp;
471 	xfs_fsblock_t			start_fsb;
472 	xfs_fsblock_t			end_fsb;
473 	xfs_daddr_t			eofs;
474 	int				error = 0;
475 
476 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
477 	if (keys[0].fmr_physical >= eofs)
478 		return 0;
479 	if (keys[1].fmr_physical >= eofs)
480 		keys[1].fmr_physical = eofs - 1;
481 	start_fsb = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
482 	end_fsb = XFS_BB_TO_FSB(mp, keys[1].fmr_physical);
483 
484 	/* Set up search keys */
485 	info->low.rm_startblock = start_fsb;
486 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
487 	if (error)
488 		return error;
489 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
490 	info->low.rm_blockcount = 0;
491 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
492 
493 	info->high.rm_startblock = end_fsb;
494 	error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
495 	if (error)
496 		return error;
497 	info->high.rm_offset = XFS_BB_TO_FSBT(mp, keys[1].fmr_offset);
498 	info->high.rm_blockcount = 0;
499 	xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
500 
501 	trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
502 	trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
503 
504 	return query_fn(tp, info);
505 }
506 
507 /* Actually query the realtime bitmap. */
508 STATIC int
509 xfs_getfsmap_rtdev_rtbitmap_query(
510 	struct xfs_trans		*tp,
511 	struct xfs_getfsmap_info	*info)
512 {
513 	struct xfs_rtalloc_rec		alow = { 0 };
514 	struct xfs_rtalloc_rec		ahigh = { 0 };
515 	int				error;
516 
517 	xfs_ilock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
518 
519 	alow.ar_startext = info->low.rm_startblock;
520 	ahigh.ar_startext = info->high.rm_startblock;
521 	do_div(alow.ar_startext, tp->t_mountp->m_sb.sb_rextsize);
522 	if (do_div(ahigh.ar_startext, tp->t_mountp->m_sb.sb_rextsize))
523 		ahigh.ar_startext++;
524 	error = xfs_rtalloc_query_range(tp, &alow, &ahigh,
525 			xfs_getfsmap_rtdev_rtbitmap_helper, info);
526 	if (error)
527 		goto err;
528 
529 	/* Report any gaps at the end of the rtbitmap */
530 	info->last = true;
531 	error = xfs_getfsmap_rtdev_rtbitmap_helper(tp, &ahigh, info);
532 	if (error)
533 		goto err;
534 err:
535 	xfs_iunlock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
536 	return error;
537 }
538 
539 /* Execute a getfsmap query against the realtime device rtbitmap. */
540 STATIC int
541 xfs_getfsmap_rtdev_rtbitmap(
542 	struct xfs_trans		*tp,
543 	struct xfs_fsmap		*keys,
544 	struct xfs_getfsmap_info	*info)
545 {
546 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
547 	return __xfs_getfsmap_rtdev(tp, keys, xfs_getfsmap_rtdev_rtbitmap_query,
548 			info);
549 }
550 #endif /* CONFIG_XFS_RT */
551 
552 /* Execute a getfsmap query against the regular data device. */
553 STATIC int
554 __xfs_getfsmap_datadev(
555 	struct xfs_trans		*tp,
556 	struct xfs_fsmap		*keys,
557 	struct xfs_getfsmap_info	*info,
558 	int				(*query_fn)(struct xfs_trans *,
559 						    struct xfs_getfsmap_info *,
560 						    struct xfs_btree_cur **,
561 						    void *),
562 	void				*priv)
563 {
564 	struct xfs_mount		*mp = tp->t_mountp;
565 	struct xfs_btree_cur		*bt_cur = NULL;
566 	xfs_fsblock_t			start_fsb;
567 	xfs_fsblock_t			end_fsb;
568 	xfs_agnumber_t			start_ag;
569 	xfs_agnumber_t			end_ag;
570 	xfs_daddr_t			eofs;
571 	int				error = 0;
572 
573 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
574 	if (keys[0].fmr_physical >= eofs)
575 		return 0;
576 	if (keys[1].fmr_physical >= eofs)
577 		keys[1].fmr_physical = eofs - 1;
578 	start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
579 	end_fsb = XFS_DADDR_TO_FSB(mp, keys[1].fmr_physical);
580 
581 	/*
582 	 * Convert the fsmap low/high keys to AG based keys.  Initialize
583 	 * low to the fsmap low key and max out the high key to the end
584 	 * of the AG.
585 	 */
586 	info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
587 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
588 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
589 	if (error)
590 		return error;
591 	info->low.rm_blockcount = 0;
592 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
593 
594 	info->high.rm_startblock = -1U;
595 	info->high.rm_owner = ULLONG_MAX;
596 	info->high.rm_offset = ULLONG_MAX;
597 	info->high.rm_blockcount = 0;
598 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
599 
600 	start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
601 	end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
602 
603 	/* Query each AG */
604 	for (info->agno = start_ag; info->agno <= end_ag; info->agno++) {
605 		/*
606 		 * Set the AG high key from the fsmap high key if this
607 		 * is the last AG that we're querying.
608 		 */
609 		if (info->agno == end_ag) {
610 			info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
611 					end_fsb);
612 			info->high.rm_offset = XFS_BB_TO_FSBT(mp,
613 					keys[1].fmr_offset);
614 			error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
615 			if (error)
616 				goto err;
617 			xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
618 		}
619 
620 		if (bt_cur) {
621 			xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
622 			bt_cur = NULL;
623 			xfs_trans_brelse(tp, info->agf_bp);
624 			info->agf_bp = NULL;
625 		}
626 
627 		error = xfs_alloc_read_agf(mp, tp, info->agno, 0,
628 				&info->agf_bp);
629 		if (error)
630 			goto err;
631 
632 		trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
633 		trace_xfs_fsmap_high_key(mp, info->dev, info->agno,
634 				&info->high);
635 
636 		error = query_fn(tp, info, &bt_cur, priv);
637 		if (error)
638 			goto err;
639 
640 		/*
641 		 * Set the AG low key to the start of the AG prior to
642 		 * moving on to the next AG.
643 		 */
644 		if (info->agno == start_ag) {
645 			info->low.rm_startblock = 0;
646 			info->low.rm_owner = 0;
647 			info->low.rm_offset = 0;
648 			info->low.rm_flags = 0;
649 		}
650 	}
651 
652 	/* Report any gap at the end of the AG */
653 	info->last = true;
654 	error = query_fn(tp, info, &bt_cur, priv);
655 	if (error)
656 		goto err;
657 
658 err:
659 	if (bt_cur)
660 		xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
661 							 XFS_BTREE_NOERROR);
662 	if (info->agf_bp) {
663 		xfs_trans_brelse(tp, info->agf_bp);
664 		info->agf_bp = NULL;
665 	}
666 
667 	return error;
668 }
669 
670 /* Actually query the rmap btree. */
671 STATIC int
672 xfs_getfsmap_datadev_rmapbt_query(
673 	struct xfs_trans		*tp,
674 	struct xfs_getfsmap_info	*info,
675 	struct xfs_btree_cur		**curpp,
676 	void				*priv)
677 {
678 	/* Report any gap at the end of the last AG. */
679 	if (info->last)
680 		return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
681 
682 	/* Allocate cursor for this AG and query_range it. */
683 	*curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
684 			info->agno);
685 	return xfs_rmap_query_range(*curpp, &info->low, &info->high,
686 			xfs_getfsmap_datadev_helper, info);
687 }
688 
689 /* Execute a getfsmap query against the regular data device rmapbt. */
690 STATIC int
691 xfs_getfsmap_datadev_rmapbt(
692 	struct xfs_trans		*tp,
693 	struct xfs_fsmap		*keys,
694 	struct xfs_getfsmap_info	*info)
695 {
696 	info->missing_owner = XFS_FMR_OWN_FREE;
697 	return __xfs_getfsmap_datadev(tp, keys, info,
698 			xfs_getfsmap_datadev_rmapbt_query, NULL);
699 }
700 
701 /* Actually query the bno btree. */
702 STATIC int
703 xfs_getfsmap_datadev_bnobt_query(
704 	struct xfs_trans		*tp,
705 	struct xfs_getfsmap_info	*info,
706 	struct xfs_btree_cur		**curpp,
707 	void				*priv)
708 {
709 	struct xfs_alloc_rec_incore	*key = priv;
710 
711 	/* Report any gap at the end of the last AG. */
712 	if (info->last)
713 		return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
714 
715 	/* Allocate cursor for this AG and query_range it. */
716 	*curpp = xfs_allocbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
717 			info->agno, XFS_BTNUM_BNO);
718 	key->ar_startblock = info->low.rm_startblock;
719 	key[1].ar_startblock = info->high.rm_startblock;
720 	return xfs_alloc_query_range(*curpp, key, &key[1],
721 			xfs_getfsmap_datadev_bnobt_helper, info);
722 }
723 
724 /* Execute a getfsmap query against the regular data device's bnobt. */
725 STATIC int
726 xfs_getfsmap_datadev_bnobt(
727 	struct xfs_trans		*tp,
728 	struct xfs_fsmap		*keys,
729 	struct xfs_getfsmap_info	*info)
730 {
731 	struct xfs_alloc_rec_incore	akeys[2];
732 
733 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
734 	return __xfs_getfsmap_datadev(tp, keys, info,
735 			xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
736 }
737 
738 /* Do we recognize the device? */
739 STATIC bool
740 xfs_getfsmap_is_valid_device(
741 	struct xfs_mount	*mp,
742 	struct xfs_fsmap	*fm)
743 {
744 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
745 	    fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
746 		return true;
747 	if (mp->m_logdev_targp &&
748 	    fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
749 		return true;
750 	if (mp->m_rtdev_targp &&
751 	    fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
752 		return true;
753 	return false;
754 }
755 
756 /* Ensure that the low key is less than the high key. */
757 STATIC bool
758 xfs_getfsmap_check_keys(
759 	struct xfs_fsmap		*low_key,
760 	struct xfs_fsmap		*high_key)
761 {
762 	if (low_key->fmr_device > high_key->fmr_device)
763 		return false;
764 	if (low_key->fmr_device < high_key->fmr_device)
765 		return true;
766 
767 	if (low_key->fmr_physical > high_key->fmr_physical)
768 		return false;
769 	if (low_key->fmr_physical < high_key->fmr_physical)
770 		return true;
771 
772 	if (low_key->fmr_owner > high_key->fmr_owner)
773 		return false;
774 	if (low_key->fmr_owner < high_key->fmr_owner)
775 		return true;
776 
777 	if (low_key->fmr_offset > high_key->fmr_offset)
778 		return false;
779 	if (low_key->fmr_offset < high_key->fmr_offset)
780 		return true;
781 
782 	return false;
783 }
784 
785 /*
786  * There are only two devices if we didn't configure RT devices at build time.
787  */
788 #ifdef CONFIG_XFS_RT
789 #define XFS_GETFSMAP_DEVS	3
790 #else
791 #define XFS_GETFSMAP_DEVS	2
792 #endif /* CONFIG_XFS_RT */
793 
794 /*
795  * Get filesystem's extents as described in head, and format for
796  * output.  Calls formatter to fill the user's buffer until all
797  * extents are mapped, until the passed-in head->fmh_count slots have
798  * been filled, or until the formatter short-circuits the loop, if it
799  * is tracking filled-in extents on its own.
800  *
801  * Key to Confusion
802  * ----------------
803  * There are multiple levels of keys and counters at work here:
804  * xfs_fsmap_head.fmh_keys	-- low and high fsmap keys passed in;
805  * 				   these reflect fs-wide sector addrs.
806  * dkeys			-- fmh_keys used to query each device;
807  * 				   these are fmh_keys but w/ the low key
808  * 				   bumped up by fmr_length.
809  * xfs_getfsmap_info.next_daddr	-- next disk addr we expect to see; this
810  *				   is how we detect gaps in the fsmap
811 				   records and report them.
812  * xfs_getfsmap_info.low/high	-- per-AG low/high keys computed from
813  * 				   dkeys; used to query the metadata.
814  */
815 int
816 xfs_getfsmap(
817 	struct xfs_mount		*mp,
818 	struct xfs_fsmap_head		*head,
819 	xfs_fsmap_format_t		formatter,
820 	void				*arg)
821 {
822 	struct xfs_trans		*tp = NULL;
823 	struct xfs_fsmap		dkeys[2];	/* per-dev keys */
824 	struct xfs_getfsmap_dev		handlers[XFS_GETFSMAP_DEVS];
825 	struct xfs_getfsmap_info	info = { NULL };
826 	bool				use_rmap;
827 	int				i;
828 	int				error = 0;
829 
830 	if (head->fmh_iflags & ~FMH_IF_VALID)
831 		return -EINVAL;
832 	if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
833 	    !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
834 		return -EINVAL;
835 
836 	use_rmap = capable(CAP_SYS_ADMIN) &&
837 		   xfs_sb_version_hasrmapbt(&mp->m_sb);
838 	head->fmh_entries = 0;
839 
840 	/* Set up our device handlers. */
841 	memset(handlers, 0, sizeof(handlers));
842 	handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
843 	if (use_rmap)
844 		handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
845 	else
846 		handlers[0].fn = xfs_getfsmap_datadev_bnobt;
847 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
848 		handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
849 		handlers[1].fn = xfs_getfsmap_logdev;
850 	}
851 #ifdef CONFIG_XFS_RT
852 	if (mp->m_rtdev_targp) {
853 		handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
854 		handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
855 	}
856 #endif /* CONFIG_XFS_RT */
857 
858 	xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
859 			xfs_getfsmap_dev_compare);
860 
861 	/*
862 	 * To continue where we left off, we allow userspace to use the
863 	 * last mapping from a previous call as the low key of the next.
864 	 * This is identified by a non-zero length in the low key. We
865 	 * have to increment the low key in this scenario to ensure we
866 	 * don't return the same mapping again, and instead return the
867 	 * very next mapping.
868 	 *
869 	 * If the low key mapping refers to file data, the same physical
870 	 * blocks could be mapped to several other files/offsets.
871 	 * According to rmapbt record ordering, the minimal next
872 	 * possible record for the block range is the next starting
873 	 * offset in the same inode. Therefore, bump the file offset to
874 	 * continue the search appropriately.  For all other low key
875 	 * mapping types (attr blocks, metadata), bump the physical
876 	 * offset as there can be no other mapping for the same physical
877 	 * block range.
878 	 */
879 	dkeys[0] = head->fmh_keys[0];
880 	if (dkeys[0].fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
881 		dkeys[0].fmr_physical += dkeys[0].fmr_length;
882 		dkeys[0].fmr_owner = 0;
883 		if (dkeys[0].fmr_offset)
884 			return -EINVAL;
885 	} else
886 		dkeys[0].fmr_offset += dkeys[0].fmr_length;
887 	dkeys[0].fmr_length = 0;
888 	memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
889 
890 	if (!xfs_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
891 		return -EINVAL;
892 
893 	info.next_daddr = head->fmh_keys[0].fmr_physical +
894 			  head->fmh_keys[0].fmr_length;
895 	info.formatter = formatter;
896 	info.format_arg = arg;
897 	info.head = head;
898 
899 	/* For each device we support... */
900 	for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
901 		/* Is this device within the range the user asked for? */
902 		if (!handlers[i].fn)
903 			continue;
904 		if (head->fmh_keys[0].fmr_device > handlers[i].dev)
905 			continue;
906 		if (head->fmh_keys[1].fmr_device < handlers[i].dev)
907 			break;
908 
909 		/*
910 		 * If this device number matches the high key, we have
911 		 * to pass the high key to the handler to limit the
912 		 * query results.  If the device number exceeds the
913 		 * low key, zero out the low key so that we get
914 		 * everything from the beginning.
915 		 */
916 		if (handlers[i].dev == head->fmh_keys[1].fmr_device)
917 			dkeys[1] = head->fmh_keys[1];
918 		if (handlers[i].dev > head->fmh_keys[0].fmr_device)
919 			memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
920 
921 		error = xfs_trans_alloc_empty(mp, &tp);
922 		if (error)
923 			break;
924 
925 		info.dev = handlers[i].dev;
926 		info.last = false;
927 		info.agno = NULLAGNUMBER;
928 		error = handlers[i].fn(tp, dkeys, &info);
929 		if (error)
930 			break;
931 		xfs_trans_cancel(tp);
932 		tp = NULL;
933 		info.next_daddr = 0;
934 	}
935 
936 	if (tp)
937 		xfs_trans_cancel(tp);
938 	head->fmh_oflags = FMH_OF_DEV_T;
939 	return error;
940 }
941