xref: /openbmc/linux/fs/xfs/xfs_fsmap.c (revision d9fd5a71)
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 static 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 	struct fsmap		*fsmap_recs;	/* mapping records */
159 	struct xfs_buf		*agf_bp;	/* AGF, for refcount queries */
160 	xfs_daddr_t		next_daddr;	/* next daddr we expect */
161 	u64			missing_owner;	/* owner of holes */
162 	u32			dev;		/* device id */
163 	xfs_agnumber_t		agno;		/* AG number, if applicable */
164 	struct xfs_rmap_irec	low;		/* low rmap key */
165 	struct xfs_rmap_irec	high;		/* high rmap key */
166 	bool			last;		/* last extent? */
167 };
168 
169 /* Associate a device with a getfsmap handler. */
170 struct xfs_getfsmap_dev {
171 	u32			dev;
172 	int			(*fn)(struct xfs_trans *tp,
173 				      struct xfs_fsmap *keys,
174 				      struct xfs_getfsmap_info *info);
175 };
176 
177 /* Compare two getfsmap device handlers. */
178 static int
179 xfs_getfsmap_dev_compare(
180 	const void			*p1,
181 	const void			*p2)
182 {
183 	const struct xfs_getfsmap_dev	*d1 = p1;
184 	const struct xfs_getfsmap_dev	*d2 = p2;
185 
186 	return d1->dev - d2->dev;
187 }
188 
189 /* Decide if this mapping is shared. */
190 STATIC int
191 xfs_getfsmap_is_shared(
192 	struct xfs_trans		*tp,
193 	struct xfs_getfsmap_info	*info,
194 	struct xfs_rmap_irec		*rec,
195 	bool				*stat)
196 {
197 	struct xfs_mount		*mp = tp->t_mountp;
198 	struct xfs_btree_cur		*cur;
199 	xfs_agblock_t			fbno;
200 	xfs_extlen_t			flen;
201 	int				error;
202 
203 	*stat = false;
204 	if (!xfs_sb_version_hasreflink(&mp->m_sb))
205 		return 0;
206 	/* rt files will have agno set to NULLAGNUMBER */
207 	if (info->agno == NULLAGNUMBER)
208 		return 0;
209 
210 	/* Are there any shared blocks here? */
211 	flen = 0;
212 	cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp,
213 			info->agno);
214 
215 	error = xfs_refcount_find_shared(cur, rec->rm_startblock,
216 			rec->rm_blockcount, &fbno, &flen, false);
217 
218 	xfs_btree_del_cursor(cur, error);
219 	if (error)
220 		return error;
221 
222 	*stat = flen > 0;
223 	return 0;
224 }
225 
226 static inline void
227 xfs_getfsmap_format(
228 	struct xfs_mount		*mp,
229 	struct xfs_fsmap		*xfm,
230 	struct xfs_getfsmap_info	*info)
231 {
232 	struct fsmap			*rec;
233 
234 	trace_xfs_getfsmap_mapping(mp, xfm);
235 
236 	rec = &info->fsmap_recs[info->head->fmh_entries++];
237 	xfs_fsmap_from_internal(rec, xfm);
238 }
239 
240 /*
241  * Format a reverse mapping for getfsmap, having translated rm_startblock
242  * into the appropriate daddr units.
243  */
244 STATIC int
245 xfs_getfsmap_helper(
246 	struct xfs_trans		*tp,
247 	struct xfs_getfsmap_info	*info,
248 	struct xfs_rmap_irec		*rec,
249 	xfs_daddr_t			rec_daddr)
250 {
251 	struct xfs_fsmap		fmr;
252 	struct xfs_mount		*mp = tp->t_mountp;
253 	bool				shared;
254 	int				error;
255 
256 	if (fatal_signal_pending(current))
257 		return -EINTR;
258 
259 	/*
260 	 * Filter out records that start before our startpoint, if the
261 	 * caller requested that.
262 	 */
263 	if (xfs_rmap_compare(rec, &info->low) < 0) {
264 		rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
265 		if (info->next_daddr < rec_daddr)
266 			info->next_daddr = rec_daddr;
267 		return 0;
268 	}
269 
270 	/* Are we just counting mappings? */
271 	if (info->head->fmh_count == 0) {
272 		if (info->head->fmh_entries == UINT_MAX)
273 			return -ECANCELED;
274 
275 		if (rec_daddr > info->next_daddr)
276 			info->head->fmh_entries++;
277 
278 		if (info->last)
279 			return 0;
280 
281 		info->head->fmh_entries++;
282 
283 		rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
284 		if (info->next_daddr < rec_daddr)
285 			info->next_daddr = rec_daddr;
286 		return 0;
287 	}
288 
289 	/*
290 	 * If the record starts past the last physical block we saw,
291 	 * then we've found a gap.  Report the gap as being owned by
292 	 * whatever the caller specified is the missing owner.
293 	 */
294 	if (rec_daddr > info->next_daddr) {
295 		if (info->head->fmh_entries >= info->head->fmh_count)
296 			return -ECANCELED;
297 
298 		fmr.fmr_device = info->dev;
299 		fmr.fmr_physical = info->next_daddr;
300 		fmr.fmr_owner = info->missing_owner;
301 		fmr.fmr_offset = 0;
302 		fmr.fmr_length = rec_daddr - info->next_daddr;
303 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
304 		xfs_getfsmap_format(mp, &fmr, info);
305 	}
306 
307 	if (info->last)
308 		goto out;
309 
310 	/* Fill out the extent we found */
311 	if (info->head->fmh_entries >= info->head->fmh_count)
312 		return -ECANCELED;
313 
314 	trace_xfs_fsmap_mapping(mp, info->dev, info->agno, rec);
315 
316 	fmr.fmr_device = info->dev;
317 	fmr.fmr_physical = rec_daddr;
318 	error = xfs_fsmap_owner_from_rmap(&fmr, rec);
319 	if (error)
320 		return error;
321 	fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
322 	fmr.fmr_length = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
323 	if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
324 		fmr.fmr_flags |= FMR_OF_PREALLOC;
325 	if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
326 		fmr.fmr_flags |= FMR_OF_ATTR_FORK;
327 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
328 		fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
329 	if (fmr.fmr_flags == 0) {
330 		error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
331 		if (error)
332 			return error;
333 		if (shared)
334 			fmr.fmr_flags |= FMR_OF_SHARED;
335 	}
336 
337 	xfs_getfsmap_format(mp, &fmr, info);
338 out:
339 	rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
340 	if (info->next_daddr < rec_daddr)
341 		info->next_daddr = rec_daddr;
342 	return 0;
343 }
344 
345 /* Transform a rmapbt irec into a fsmap */
346 STATIC int
347 xfs_getfsmap_datadev_helper(
348 	struct xfs_btree_cur		*cur,
349 	struct xfs_rmap_irec		*rec,
350 	void				*priv)
351 {
352 	struct xfs_mount		*mp = cur->bc_mp;
353 	struct xfs_getfsmap_info	*info = priv;
354 	xfs_fsblock_t			fsb;
355 	xfs_daddr_t			rec_daddr;
356 
357 	fsb = XFS_AGB_TO_FSB(mp, cur->bc_ag.agno, rec->rm_startblock);
358 	rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
359 
360 	return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr);
361 }
362 
363 /* Transform a bnobt irec into a fsmap */
364 STATIC int
365 xfs_getfsmap_datadev_bnobt_helper(
366 	struct xfs_btree_cur		*cur,
367 	struct xfs_alloc_rec_incore	*rec,
368 	void				*priv)
369 {
370 	struct xfs_mount		*mp = cur->bc_mp;
371 	struct xfs_getfsmap_info	*info = priv;
372 	struct xfs_rmap_irec		irec;
373 	xfs_daddr_t			rec_daddr;
374 
375 	rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_ag.agno,
376 			rec->ar_startblock);
377 
378 	irec.rm_startblock = rec->ar_startblock;
379 	irec.rm_blockcount = rec->ar_blockcount;
380 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
381 	irec.rm_offset = 0;
382 	irec.rm_flags = 0;
383 
384 	return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr);
385 }
386 
387 /* Set rmap flags based on the getfsmap flags */
388 static void
389 xfs_getfsmap_set_irec_flags(
390 	struct xfs_rmap_irec	*irec,
391 	struct xfs_fsmap	*fmr)
392 {
393 	irec->rm_flags = 0;
394 	if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
395 		irec->rm_flags |= XFS_RMAP_ATTR_FORK;
396 	if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
397 		irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
398 	if (fmr->fmr_flags & FMR_OF_PREALLOC)
399 		irec->rm_flags |= XFS_RMAP_UNWRITTEN;
400 }
401 
402 /* Execute a getfsmap query against the log device. */
403 STATIC int
404 xfs_getfsmap_logdev(
405 	struct xfs_trans		*tp,
406 	struct xfs_fsmap		*keys,
407 	struct xfs_getfsmap_info	*info)
408 {
409 	struct xfs_mount		*mp = tp->t_mountp;
410 	struct xfs_rmap_irec		rmap;
411 	int				error;
412 
413 	/* Set up search keys */
414 	info->low.rm_startblock = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
415 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
416 	error = xfs_fsmap_owner_to_rmap(&info->low, keys);
417 	if (error)
418 		return error;
419 	info->low.rm_blockcount = 0;
420 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
421 
422 	error = xfs_fsmap_owner_to_rmap(&info->high, keys + 1);
423 	if (error)
424 		return error;
425 	info->high.rm_startblock = -1U;
426 	info->high.rm_owner = ULLONG_MAX;
427 	info->high.rm_offset = ULLONG_MAX;
428 	info->high.rm_blockcount = 0;
429 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
430 	info->missing_owner = XFS_FMR_OWN_FREE;
431 
432 	trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
433 	trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
434 
435 	if (keys[0].fmr_physical > 0)
436 		return 0;
437 
438 	/* Fabricate an rmap entry for the external log device. */
439 	rmap.rm_startblock = 0;
440 	rmap.rm_blockcount = mp->m_sb.sb_logblocks;
441 	rmap.rm_owner = XFS_RMAP_OWN_LOG;
442 	rmap.rm_offset = 0;
443 	rmap.rm_flags = 0;
444 
445 	return xfs_getfsmap_helper(tp, info, &rmap, 0);
446 }
447 
448 #ifdef CONFIG_XFS_RT
449 /* Transform a rtbitmap "record" into a fsmap */
450 STATIC int
451 xfs_getfsmap_rtdev_rtbitmap_helper(
452 	struct xfs_trans		*tp,
453 	struct xfs_rtalloc_rec		*rec,
454 	void				*priv)
455 {
456 	struct xfs_mount		*mp = tp->t_mountp;
457 	struct xfs_getfsmap_info	*info = priv;
458 	struct xfs_rmap_irec		irec;
459 	xfs_daddr_t			rec_daddr;
460 
461 	irec.rm_startblock = rec->ar_startext * mp->m_sb.sb_rextsize;
462 	rec_daddr = XFS_FSB_TO_BB(mp, irec.rm_startblock);
463 	irec.rm_blockcount = rec->ar_extcount * mp->m_sb.sb_rextsize;
464 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
465 	irec.rm_offset = 0;
466 	irec.rm_flags = 0;
467 
468 	return xfs_getfsmap_helper(tp, info, &irec, rec_daddr);
469 }
470 
471 /* Execute a getfsmap query against the realtime device. */
472 STATIC int
473 __xfs_getfsmap_rtdev(
474 	struct xfs_trans		*tp,
475 	struct xfs_fsmap		*keys,
476 	int				(*query_fn)(struct xfs_trans *,
477 						    struct xfs_getfsmap_info *),
478 	struct xfs_getfsmap_info	*info)
479 {
480 	struct xfs_mount		*mp = tp->t_mountp;
481 	xfs_fsblock_t			start_fsb;
482 	xfs_fsblock_t			end_fsb;
483 	xfs_daddr_t			eofs;
484 	int				error = 0;
485 
486 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
487 	if (keys[0].fmr_physical >= eofs)
488 		return 0;
489 	if (keys[1].fmr_physical >= eofs)
490 		keys[1].fmr_physical = eofs - 1;
491 	start_fsb = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
492 	end_fsb = XFS_BB_TO_FSB(mp, keys[1].fmr_physical);
493 
494 	/* Set up search keys */
495 	info->low.rm_startblock = start_fsb;
496 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
497 	if (error)
498 		return error;
499 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
500 	info->low.rm_blockcount = 0;
501 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
502 
503 	info->high.rm_startblock = end_fsb;
504 	error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
505 	if (error)
506 		return error;
507 	info->high.rm_offset = XFS_BB_TO_FSBT(mp, keys[1].fmr_offset);
508 	info->high.rm_blockcount = 0;
509 	xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
510 
511 	trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
512 	trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
513 
514 	return query_fn(tp, info);
515 }
516 
517 /* Actually query the realtime bitmap. */
518 STATIC int
519 xfs_getfsmap_rtdev_rtbitmap_query(
520 	struct xfs_trans		*tp,
521 	struct xfs_getfsmap_info	*info)
522 {
523 	struct xfs_rtalloc_rec		alow = { 0 };
524 	struct xfs_rtalloc_rec		ahigh = { 0 };
525 	int				error;
526 
527 	xfs_ilock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
528 
529 	alow.ar_startext = info->low.rm_startblock;
530 	ahigh.ar_startext = info->high.rm_startblock;
531 	do_div(alow.ar_startext, tp->t_mountp->m_sb.sb_rextsize);
532 	if (do_div(ahigh.ar_startext, tp->t_mountp->m_sb.sb_rextsize))
533 		ahigh.ar_startext++;
534 	error = xfs_rtalloc_query_range(tp, &alow, &ahigh,
535 			xfs_getfsmap_rtdev_rtbitmap_helper, info);
536 	if (error)
537 		goto err;
538 
539 	/* Report any gaps at the end of the rtbitmap */
540 	info->last = true;
541 	error = xfs_getfsmap_rtdev_rtbitmap_helper(tp, &ahigh, info);
542 	if (error)
543 		goto err;
544 err:
545 	xfs_iunlock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
546 	return error;
547 }
548 
549 /* Execute a getfsmap query against the realtime device rtbitmap. */
550 STATIC int
551 xfs_getfsmap_rtdev_rtbitmap(
552 	struct xfs_trans		*tp,
553 	struct xfs_fsmap		*keys,
554 	struct xfs_getfsmap_info	*info)
555 {
556 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
557 	return __xfs_getfsmap_rtdev(tp, keys, xfs_getfsmap_rtdev_rtbitmap_query,
558 			info);
559 }
560 #endif /* CONFIG_XFS_RT */
561 
562 /* Execute a getfsmap query against the regular data device. */
563 STATIC int
564 __xfs_getfsmap_datadev(
565 	struct xfs_trans		*tp,
566 	struct xfs_fsmap		*keys,
567 	struct xfs_getfsmap_info	*info,
568 	int				(*query_fn)(struct xfs_trans *,
569 						    struct xfs_getfsmap_info *,
570 						    struct xfs_btree_cur **,
571 						    void *),
572 	void				*priv)
573 {
574 	struct xfs_mount		*mp = tp->t_mountp;
575 	struct xfs_btree_cur		*bt_cur = NULL;
576 	xfs_fsblock_t			start_fsb;
577 	xfs_fsblock_t			end_fsb;
578 	xfs_agnumber_t			start_ag;
579 	xfs_agnumber_t			end_ag;
580 	xfs_daddr_t			eofs;
581 	int				error = 0;
582 
583 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
584 	if (keys[0].fmr_physical >= eofs)
585 		return 0;
586 	if (keys[1].fmr_physical >= eofs)
587 		keys[1].fmr_physical = eofs - 1;
588 	start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
589 	end_fsb = XFS_DADDR_TO_FSB(mp, keys[1].fmr_physical);
590 
591 	/*
592 	 * Convert the fsmap low/high keys to AG based keys.  Initialize
593 	 * low to the fsmap low key and max out the high key to the end
594 	 * of the AG.
595 	 */
596 	info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
597 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
598 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
599 	if (error)
600 		return error;
601 	info->low.rm_blockcount = 0;
602 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
603 
604 	info->high.rm_startblock = -1U;
605 	info->high.rm_owner = ULLONG_MAX;
606 	info->high.rm_offset = ULLONG_MAX;
607 	info->high.rm_blockcount = 0;
608 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
609 
610 	start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
611 	end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
612 
613 	/* Query each AG */
614 	for (info->agno = start_ag; info->agno <= end_ag; info->agno++) {
615 		/*
616 		 * Set the AG high key from the fsmap high key if this
617 		 * is the last AG that we're querying.
618 		 */
619 		if (info->agno == end_ag) {
620 			info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
621 					end_fsb);
622 			info->high.rm_offset = XFS_BB_TO_FSBT(mp,
623 					keys[1].fmr_offset);
624 			error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
625 			if (error)
626 				goto err;
627 			xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
628 		}
629 
630 		if (bt_cur) {
631 			xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
632 			bt_cur = NULL;
633 			xfs_trans_brelse(tp, info->agf_bp);
634 			info->agf_bp = NULL;
635 		}
636 
637 		error = xfs_alloc_read_agf(mp, tp, info->agno, 0,
638 				&info->agf_bp);
639 		if (error)
640 			goto err;
641 
642 		trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
643 		trace_xfs_fsmap_high_key(mp, info->dev, info->agno,
644 				&info->high);
645 
646 		error = query_fn(tp, info, &bt_cur, priv);
647 		if (error)
648 			goto err;
649 
650 		/*
651 		 * Set the AG low key to the start of the AG prior to
652 		 * moving on to the next AG.
653 		 */
654 		if (info->agno == start_ag) {
655 			info->low.rm_startblock = 0;
656 			info->low.rm_owner = 0;
657 			info->low.rm_offset = 0;
658 			info->low.rm_flags = 0;
659 		}
660 	}
661 
662 	/* Report any gap at the end of the AG */
663 	info->last = true;
664 	error = query_fn(tp, info, &bt_cur, priv);
665 	if (error)
666 		goto err;
667 
668 err:
669 	if (bt_cur)
670 		xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
671 							 XFS_BTREE_NOERROR);
672 	if (info->agf_bp) {
673 		xfs_trans_brelse(tp, info->agf_bp);
674 		info->agf_bp = NULL;
675 	}
676 
677 	return error;
678 }
679 
680 /* Actually query the rmap btree. */
681 STATIC int
682 xfs_getfsmap_datadev_rmapbt_query(
683 	struct xfs_trans		*tp,
684 	struct xfs_getfsmap_info	*info,
685 	struct xfs_btree_cur		**curpp,
686 	void				*priv)
687 {
688 	/* Report any gap at the end of the last AG. */
689 	if (info->last)
690 		return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
691 
692 	/* Allocate cursor for this AG and query_range it. */
693 	*curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
694 			info->agno);
695 	return xfs_rmap_query_range(*curpp, &info->low, &info->high,
696 			xfs_getfsmap_datadev_helper, info);
697 }
698 
699 /* Execute a getfsmap query against the regular data device rmapbt. */
700 STATIC int
701 xfs_getfsmap_datadev_rmapbt(
702 	struct xfs_trans		*tp,
703 	struct xfs_fsmap		*keys,
704 	struct xfs_getfsmap_info	*info)
705 {
706 	info->missing_owner = XFS_FMR_OWN_FREE;
707 	return __xfs_getfsmap_datadev(tp, keys, info,
708 			xfs_getfsmap_datadev_rmapbt_query, NULL);
709 }
710 
711 /* Actually query the bno btree. */
712 STATIC int
713 xfs_getfsmap_datadev_bnobt_query(
714 	struct xfs_trans		*tp,
715 	struct xfs_getfsmap_info	*info,
716 	struct xfs_btree_cur		**curpp,
717 	void				*priv)
718 {
719 	struct xfs_alloc_rec_incore	*key = priv;
720 
721 	/* Report any gap at the end of the last AG. */
722 	if (info->last)
723 		return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
724 
725 	/* Allocate cursor for this AG and query_range it. */
726 	*curpp = xfs_allocbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
727 			info->agno, XFS_BTNUM_BNO);
728 	key->ar_startblock = info->low.rm_startblock;
729 	key[1].ar_startblock = info->high.rm_startblock;
730 	return xfs_alloc_query_range(*curpp, key, &key[1],
731 			xfs_getfsmap_datadev_bnobt_helper, info);
732 }
733 
734 /* Execute a getfsmap query against the regular data device's bnobt. */
735 STATIC int
736 xfs_getfsmap_datadev_bnobt(
737 	struct xfs_trans		*tp,
738 	struct xfs_fsmap		*keys,
739 	struct xfs_getfsmap_info	*info)
740 {
741 	struct xfs_alloc_rec_incore	akeys[2];
742 
743 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
744 	return __xfs_getfsmap_datadev(tp, keys, info,
745 			xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
746 }
747 
748 /* Do we recognize the device? */
749 STATIC bool
750 xfs_getfsmap_is_valid_device(
751 	struct xfs_mount	*mp,
752 	struct xfs_fsmap	*fm)
753 {
754 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
755 	    fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
756 		return true;
757 	if (mp->m_logdev_targp &&
758 	    fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
759 		return true;
760 	if (mp->m_rtdev_targp &&
761 	    fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
762 		return true;
763 	return false;
764 }
765 
766 /* Ensure that the low key is less than the high key. */
767 STATIC bool
768 xfs_getfsmap_check_keys(
769 	struct xfs_fsmap		*low_key,
770 	struct xfs_fsmap		*high_key)
771 {
772 	if (low_key->fmr_device > high_key->fmr_device)
773 		return false;
774 	if (low_key->fmr_device < high_key->fmr_device)
775 		return true;
776 
777 	if (low_key->fmr_physical > high_key->fmr_physical)
778 		return false;
779 	if (low_key->fmr_physical < high_key->fmr_physical)
780 		return true;
781 
782 	if (low_key->fmr_owner > high_key->fmr_owner)
783 		return false;
784 	if (low_key->fmr_owner < high_key->fmr_owner)
785 		return true;
786 
787 	if (low_key->fmr_offset > high_key->fmr_offset)
788 		return false;
789 	if (low_key->fmr_offset < high_key->fmr_offset)
790 		return true;
791 
792 	return false;
793 }
794 
795 /*
796  * There are only two devices if we didn't configure RT devices at build time.
797  */
798 #ifdef CONFIG_XFS_RT
799 #define XFS_GETFSMAP_DEVS	3
800 #else
801 #define XFS_GETFSMAP_DEVS	2
802 #endif /* CONFIG_XFS_RT */
803 
804 /*
805  * Get filesystem's extents as described in head, and format for output. Fills
806  * in the supplied records array until there are no more reverse mappings to
807  * return or head.fmh_entries == head.fmh_count.  In the second case, this
808  * function returns -ECANCELED to indicate that more records would have been
809  * returned.
810  *
811  * Key to Confusion
812  * ----------------
813  * There are multiple levels of keys and counters at work here:
814  * xfs_fsmap_head.fmh_keys	-- low and high fsmap keys passed in;
815  * 				   these reflect fs-wide sector addrs.
816  * dkeys			-- fmh_keys used to query each device;
817  * 				   these are fmh_keys but w/ the low key
818  * 				   bumped up by fmr_length.
819  * xfs_getfsmap_info.next_daddr	-- next disk addr we expect to see; this
820  *				   is how we detect gaps in the fsmap
821 				   records and report them.
822  * xfs_getfsmap_info.low/high	-- per-AG low/high keys computed from
823  * 				   dkeys; used to query the metadata.
824  */
825 int
826 xfs_getfsmap(
827 	struct xfs_mount		*mp,
828 	struct xfs_fsmap_head		*head,
829 	struct fsmap			*fsmap_recs)
830 {
831 	struct xfs_trans		*tp = NULL;
832 	struct xfs_fsmap		dkeys[2];	/* per-dev keys */
833 	struct xfs_getfsmap_dev		handlers[XFS_GETFSMAP_DEVS];
834 	struct xfs_getfsmap_info	info = { NULL };
835 	bool				use_rmap;
836 	int				i;
837 	int				error = 0;
838 
839 	if (head->fmh_iflags & ~FMH_IF_VALID)
840 		return -EINVAL;
841 	if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
842 	    !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
843 		return -EINVAL;
844 
845 	use_rmap = capable(CAP_SYS_ADMIN) &&
846 		   xfs_sb_version_hasrmapbt(&mp->m_sb);
847 	head->fmh_entries = 0;
848 
849 	/* Set up our device handlers. */
850 	memset(handlers, 0, sizeof(handlers));
851 	handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
852 	if (use_rmap)
853 		handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
854 	else
855 		handlers[0].fn = xfs_getfsmap_datadev_bnobt;
856 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
857 		handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
858 		handlers[1].fn = xfs_getfsmap_logdev;
859 	}
860 #ifdef CONFIG_XFS_RT
861 	if (mp->m_rtdev_targp) {
862 		handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
863 		handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
864 	}
865 #endif /* CONFIG_XFS_RT */
866 
867 	xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
868 			xfs_getfsmap_dev_compare);
869 
870 	/*
871 	 * To continue where we left off, we allow userspace to use the
872 	 * last mapping from a previous call as the low key of the next.
873 	 * This is identified by a non-zero length in the low key. We
874 	 * have to increment the low key in this scenario to ensure we
875 	 * don't return the same mapping again, and instead return the
876 	 * very next mapping.
877 	 *
878 	 * If the low key mapping refers to file data, the same physical
879 	 * blocks could be mapped to several other files/offsets.
880 	 * According to rmapbt record ordering, the minimal next
881 	 * possible record for the block range is the next starting
882 	 * offset in the same inode. Therefore, bump the file offset to
883 	 * continue the search appropriately.  For all other low key
884 	 * mapping types (attr blocks, metadata), bump the physical
885 	 * offset as there can be no other mapping for the same physical
886 	 * block range.
887 	 */
888 	dkeys[0] = head->fmh_keys[0];
889 	if (dkeys[0].fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
890 		dkeys[0].fmr_physical += dkeys[0].fmr_length;
891 		dkeys[0].fmr_owner = 0;
892 		if (dkeys[0].fmr_offset)
893 			return -EINVAL;
894 	} else
895 		dkeys[0].fmr_offset += dkeys[0].fmr_length;
896 	dkeys[0].fmr_length = 0;
897 	memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
898 
899 	if (!xfs_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
900 		return -EINVAL;
901 
902 	info.next_daddr = head->fmh_keys[0].fmr_physical +
903 			  head->fmh_keys[0].fmr_length;
904 	info.fsmap_recs = fsmap_recs;
905 	info.head = head;
906 
907 	/*
908 	 * If fsmap runs concurrently with a scrub, the freeze can be delayed
909 	 * indefinitely as we walk the rmapbt and iterate over metadata
910 	 * buffers.  Freeze quiesces the log (which waits for the buffer LRU to
911 	 * be emptied) and that won't happen while we're reading buffers.
912 	 */
913 	sb_start_write(mp->m_super);
914 
915 	/* For each device we support... */
916 	for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
917 		/* Is this device within the range the user asked for? */
918 		if (!handlers[i].fn)
919 			continue;
920 		if (head->fmh_keys[0].fmr_device > handlers[i].dev)
921 			continue;
922 		if (head->fmh_keys[1].fmr_device < handlers[i].dev)
923 			break;
924 
925 		/*
926 		 * If this device number matches the high key, we have
927 		 * to pass the high key to the handler to limit the
928 		 * query results.  If the device number exceeds the
929 		 * low key, zero out the low key so that we get
930 		 * everything from the beginning.
931 		 */
932 		if (handlers[i].dev == head->fmh_keys[1].fmr_device)
933 			dkeys[1] = head->fmh_keys[1];
934 		if (handlers[i].dev > head->fmh_keys[0].fmr_device)
935 			memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
936 
937 		error = xfs_trans_alloc_empty(mp, &tp);
938 		if (error)
939 			break;
940 
941 		info.dev = handlers[i].dev;
942 		info.last = false;
943 		info.agno = NULLAGNUMBER;
944 		error = handlers[i].fn(tp, dkeys, &info);
945 		if (error)
946 			break;
947 		xfs_trans_cancel(tp);
948 		tp = NULL;
949 		info.next_daddr = 0;
950 	}
951 
952 	if (tp)
953 		xfs_trans_cancel(tp);
954 	sb_end_write(mp->m_super);
955 	head->fmh_oflags = FMH_OF_DEV_T;
956 	return error;
957 }
958