xref: /linux/fs/xfs/xfs_fsmap.c (revision 56fb34d86e875dbb0d3e6a81c5d3d035db373031)
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 		return -EFSCORRUPTED;
150 	}
151 	return 0;
152 }
153 
154 /* getfsmap query state */
155 struct xfs_getfsmap_info {
156 	struct xfs_fsmap_head	*head;
157 	xfs_fsmap_format_t	formatter;	/* formatting fn */
158 	void			*format_arg;	/* format buffer */
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 /*
227  * Format a reverse mapping for getfsmap, having translated rm_startblock
228  * into the appropriate daddr units.
229  */
230 STATIC int
231 xfs_getfsmap_helper(
232 	struct xfs_trans		*tp,
233 	struct xfs_getfsmap_info	*info,
234 	struct xfs_rmap_irec		*rec,
235 	xfs_daddr_t			rec_daddr)
236 {
237 	struct xfs_fsmap		fmr;
238 	struct xfs_mount		*mp = tp->t_mountp;
239 	bool				shared;
240 	int				error;
241 
242 	if (fatal_signal_pending(current))
243 		return -EINTR;
244 
245 	/*
246 	 * Filter out records that start before our startpoint, if the
247 	 * caller requested that.
248 	 */
249 	if (xfs_rmap_compare(rec, &info->low) < 0) {
250 		rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
251 		if (info->next_daddr < rec_daddr)
252 			info->next_daddr = rec_daddr;
253 		return 0;
254 	}
255 
256 	/* Are we just counting mappings? */
257 	if (info->head->fmh_count == 0) {
258 		if (rec_daddr > info->next_daddr)
259 			info->head->fmh_entries++;
260 
261 		if (info->last)
262 			return 0;
263 
264 		info->head->fmh_entries++;
265 
266 		rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
267 		if (info->next_daddr < rec_daddr)
268 			info->next_daddr = rec_daddr;
269 		return 0;
270 	}
271 
272 	/*
273 	 * If the record starts past the last physical block we saw,
274 	 * then we've found a gap.  Report the gap as being owned by
275 	 * whatever the caller specified is the missing owner.
276 	 */
277 	if (rec_daddr > info->next_daddr) {
278 		if (info->head->fmh_entries >= info->head->fmh_count)
279 			return -ECANCELED;
280 
281 		fmr.fmr_device = info->dev;
282 		fmr.fmr_physical = info->next_daddr;
283 		fmr.fmr_owner = info->missing_owner;
284 		fmr.fmr_offset = 0;
285 		fmr.fmr_length = rec_daddr - info->next_daddr;
286 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
287 		error = info->formatter(&fmr, info->format_arg);
288 		if (error)
289 			return error;
290 		info->head->fmh_entries++;
291 	}
292 
293 	if (info->last)
294 		goto out;
295 
296 	/* Fill out the extent we found */
297 	if (info->head->fmh_entries >= info->head->fmh_count)
298 		return -ECANCELED;
299 
300 	trace_xfs_fsmap_mapping(mp, info->dev, info->agno, rec);
301 
302 	fmr.fmr_device = info->dev;
303 	fmr.fmr_physical = rec_daddr;
304 	error = xfs_fsmap_owner_from_rmap(&fmr, rec);
305 	if (error)
306 		return error;
307 	fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
308 	fmr.fmr_length = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
309 	if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
310 		fmr.fmr_flags |= FMR_OF_PREALLOC;
311 	if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
312 		fmr.fmr_flags |= FMR_OF_ATTR_FORK;
313 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
314 		fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
315 	if (fmr.fmr_flags == 0) {
316 		error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
317 		if (error)
318 			return error;
319 		if (shared)
320 			fmr.fmr_flags |= FMR_OF_SHARED;
321 	}
322 	error = info->formatter(&fmr, info->format_arg);
323 	if (error)
324 		return error;
325 	info->head->fmh_entries++;
326 
327 out:
328 	rec_daddr += XFS_FSB_TO_BB(mp, rec->rm_blockcount);
329 	if (info->next_daddr < rec_daddr)
330 		info->next_daddr = rec_daddr;
331 	return 0;
332 }
333 
334 /* Transform a rmapbt irec into a fsmap */
335 STATIC int
336 xfs_getfsmap_datadev_helper(
337 	struct xfs_btree_cur		*cur,
338 	struct xfs_rmap_irec		*rec,
339 	void				*priv)
340 {
341 	struct xfs_mount		*mp = cur->bc_mp;
342 	struct xfs_getfsmap_info	*info = priv;
343 	xfs_fsblock_t			fsb;
344 	xfs_daddr_t			rec_daddr;
345 
346 	fsb = XFS_AGB_TO_FSB(mp, cur->bc_private.a.agno, rec->rm_startblock);
347 	rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
348 
349 	return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr);
350 }
351 
352 /* Transform a bnobt irec into a fsmap */
353 STATIC int
354 xfs_getfsmap_datadev_bnobt_helper(
355 	struct xfs_btree_cur		*cur,
356 	struct xfs_alloc_rec_incore	*rec,
357 	void				*priv)
358 {
359 	struct xfs_mount		*mp = cur->bc_mp;
360 	struct xfs_getfsmap_info	*info = priv;
361 	struct xfs_rmap_irec		irec;
362 	xfs_daddr_t			rec_daddr;
363 
364 	rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_private.a.agno,
365 			rec->ar_startblock);
366 
367 	irec.rm_startblock = rec->ar_startblock;
368 	irec.rm_blockcount = rec->ar_blockcount;
369 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
370 	irec.rm_offset = 0;
371 	irec.rm_flags = 0;
372 
373 	return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr);
374 }
375 
376 /* Set rmap flags based on the getfsmap flags */
377 static void
378 xfs_getfsmap_set_irec_flags(
379 	struct xfs_rmap_irec	*irec,
380 	struct xfs_fsmap	*fmr)
381 {
382 	irec->rm_flags = 0;
383 	if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
384 		irec->rm_flags |= XFS_RMAP_ATTR_FORK;
385 	if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
386 		irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
387 	if (fmr->fmr_flags & FMR_OF_PREALLOC)
388 		irec->rm_flags |= XFS_RMAP_UNWRITTEN;
389 }
390 
391 /* Execute a getfsmap query against the log device. */
392 STATIC int
393 xfs_getfsmap_logdev(
394 	struct xfs_trans		*tp,
395 	struct xfs_fsmap		*keys,
396 	struct xfs_getfsmap_info	*info)
397 {
398 	struct xfs_mount		*mp = tp->t_mountp;
399 	struct xfs_rmap_irec		rmap;
400 	int				error;
401 
402 	/* Set up search keys */
403 	info->low.rm_startblock = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
404 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
405 	error = xfs_fsmap_owner_to_rmap(&info->low, keys);
406 	if (error)
407 		return error;
408 	info->low.rm_blockcount = 0;
409 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
410 
411 	error = xfs_fsmap_owner_to_rmap(&info->high, keys + 1);
412 	if (error)
413 		return error;
414 	info->high.rm_startblock = -1U;
415 	info->high.rm_owner = ULLONG_MAX;
416 	info->high.rm_offset = ULLONG_MAX;
417 	info->high.rm_blockcount = 0;
418 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
419 	info->missing_owner = XFS_FMR_OWN_FREE;
420 
421 	trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
422 	trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
423 
424 	if (keys[0].fmr_physical > 0)
425 		return 0;
426 
427 	/* Fabricate an rmap entry for the external log device. */
428 	rmap.rm_startblock = 0;
429 	rmap.rm_blockcount = mp->m_sb.sb_logblocks;
430 	rmap.rm_owner = XFS_RMAP_OWN_LOG;
431 	rmap.rm_offset = 0;
432 	rmap.rm_flags = 0;
433 
434 	return xfs_getfsmap_helper(tp, info, &rmap, 0);
435 }
436 
437 #ifdef CONFIG_XFS_RT
438 /* Transform a rtbitmap "record" into a fsmap */
439 STATIC int
440 xfs_getfsmap_rtdev_rtbitmap_helper(
441 	struct xfs_trans		*tp,
442 	struct xfs_rtalloc_rec		*rec,
443 	void				*priv)
444 {
445 	struct xfs_mount		*mp = tp->t_mountp;
446 	struct xfs_getfsmap_info	*info = priv;
447 	struct xfs_rmap_irec		irec;
448 	xfs_daddr_t			rec_daddr;
449 
450 	irec.rm_startblock = rec->ar_startext * mp->m_sb.sb_rextsize;
451 	rec_daddr = XFS_FSB_TO_BB(mp, irec.rm_startblock);
452 	irec.rm_blockcount = rec->ar_extcount * mp->m_sb.sb_rextsize;
453 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
454 	irec.rm_offset = 0;
455 	irec.rm_flags = 0;
456 
457 	return xfs_getfsmap_helper(tp, info, &irec, rec_daddr);
458 }
459 
460 /* Execute a getfsmap query against the realtime device. */
461 STATIC int
462 __xfs_getfsmap_rtdev(
463 	struct xfs_trans		*tp,
464 	struct xfs_fsmap		*keys,
465 	int				(*query_fn)(struct xfs_trans *,
466 						    struct xfs_getfsmap_info *),
467 	struct xfs_getfsmap_info	*info)
468 {
469 	struct xfs_mount		*mp = tp->t_mountp;
470 	xfs_fsblock_t			start_fsb;
471 	xfs_fsblock_t			end_fsb;
472 	xfs_daddr_t			eofs;
473 	int				error = 0;
474 
475 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
476 	if (keys[0].fmr_physical >= eofs)
477 		return 0;
478 	if (keys[1].fmr_physical >= eofs)
479 		keys[1].fmr_physical = eofs - 1;
480 	start_fsb = XFS_BB_TO_FSBT(mp, keys[0].fmr_physical);
481 	end_fsb = XFS_BB_TO_FSB(mp, keys[1].fmr_physical);
482 
483 	/* Set up search keys */
484 	info->low.rm_startblock = start_fsb;
485 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
486 	if (error)
487 		return error;
488 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
489 	info->low.rm_blockcount = 0;
490 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
491 
492 	info->high.rm_startblock = end_fsb;
493 	error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
494 	if (error)
495 		return error;
496 	info->high.rm_offset = XFS_BB_TO_FSBT(mp, keys[1].fmr_offset);
497 	info->high.rm_blockcount = 0;
498 	xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
499 
500 	trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
501 	trace_xfs_fsmap_high_key(mp, info->dev, info->agno, &info->high);
502 
503 	return query_fn(tp, info);
504 }
505 
506 /* Actually query the realtime bitmap. */
507 STATIC int
508 xfs_getfsmap_rtdev_rtbitmap_query(
509 	struct xfs_trans		*tp,
510 	struct xfs_getfsmap_info	*info)
511 {
512 	struct xfs_rtalloc_rec		alow = { 0 };
513 	struct xfs_rtalloc_rec		ahigh = { 0 };
514 	int				error;
515 
516 	xfs_ilock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
517 
518 	alow.ar_startext = info->low.rm_startblock;
519 	ahigh.ar_startext = info->high.rm_startblock;
520 	do_div(alow.ar_startext, tp->t_mountp->m_sb.sb_rextsize);
521 	if (do_div(ahigh.ar_startext, tp->t_mountp->m_sb.sb_rextsize))
522 		ahigh.ar_startext++;
523 	error = xfs_rtalloc_query_range(tp, &alow, &ahigh,
524 			xfs_getfsmap_rtdev_rtbitmap_helper, info);
525 	if (error)
526 		goto err;
527 
528 	/* Report any gaps at the end of the rtbitmap */
529 	info->last = true;
530 	error = xfs_getfsmap_rtdev_rtbitmap_helper(tp, &ahigh, info);
531 	if (error)
532 		goto err;
533 err:
534 	xfs_iunlock(tp->t_mountp->m_rbmip, XFS_ILOCK_SHARED);
535 	return error;
536 }
537 
538 /* Execute a getfsmap query against the realtime device rtbitmap. */
539 STATIC int
540 xfs_getfsmap_rtdev_rtbitmap(
541 	struct xfs_trans		*tp,
542 	struct xfs_fsmap		*keys,
543 	struct xfs_getfsmap_info	*info)
544 {
545 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
546 	return __xfs_getfsmap_rtdev(tp, keys, xfs_getfsmap_rtdev_rtbitmap_query,
547 			info);
548 }
549 #endif /* CONFIG_XFS_RT */
550 
551 /* Execute a getfsmap query against the regular data device. */
552 STATIC int
553 __xfs_getfsmap_datadev(
554 	struct xfs_trans		*tp,
555 	struct xfs_fsmap		*keys,
556 	struct xfs_getfsmap_info	*info,
557 	int				(*query_fn)(struct xfs_trans *,
558 						    struct xfs_getfsmap_info *,
559 						    struct xfs_btree_cur **,
560 						    void *),
561 	void				*priv)
562 {
563 	struct xfs_mount		*mp = tp->t_mountp;
564 	struct xfs_btree_cur		*bt_cur = NULL;
565 	xfs_fsblock_t			start_fsb;
566 	xfs_fsblock_t			end_fsb;
567 	xfs_agnumber_t			start_ag;
568 	xfs_agnumber_t			end_ag;
569 	xfs_daddr_t			eofs;
570 	int				error = 0;
571 
572 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
573 	if (keys[0].fmr_physical >= eofs)
574 		return 0;
575 	if (keys[1].fmr_physical >= eofs)
576 		keys[1].fmr_physical = eofs - 1;
577 	start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
578 	end_fsb = XFS_DADDR_TO_FSB(mp, keys[1].fmr_physical);
579 
580 	/*
581 	 * Convert the fsmap low/high keys to AG based keys.  Initialize
582 	 * low to the fsmap low key and max out the high key to the end
583 	 * of the AG.
584 	 */
585 	info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
586 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
587 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
588 	if (error)
589 		return error;
590 	info->low.rm_blockcount = 0;
591 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
592 
593 	info->high.rm_startblock = -1U;
594 	info->high.rm_owner = ULLONG_MAX;
595 	info->high.rm_offset = ULLONG_MAX;
596 	info->high.rm_blockcount = 0;
597 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
598 
599 	start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
600 	end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
601 
602 	/* Query each AG */
603 	for (info->agno = start_ag; info->agno <= end_ag; info->agno++) {
604 		/*
605 		 * Set the AG high key from the fsmap high key if this
606 		 * is the last AG that we're querying.
607 		 */
608 		if (info->agno == end_ag) {
609 			info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
610 					end_fsb);
611 			info->high.rm_offset = XFS_BB_TO_FSBT(mp,
612 					keys[1].fmr_offset);
613 			error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
614 			if (error)
615 				goto err;
616 			xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
617 		}
618 
619 		if (bt_cur) {
620 			xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
621 			bt_cur = NULL;
622 			xfs_trans_brelse(tp, info->agf_bp);
623 			info->agf_bp = NULL;
624 		}
625 
626 		error = xfs_alloc_read_agf(mp, tp, info->agno, 0,
627 				&info->agf_bp);
628 		if (error)
629 			goto err;
630 
631 		trace_xfs_fsmap_low_key(mp, info->dev, info->agno, &info->low);
632 		trace_xfs_fsmap_high_key(mp, info->dev, info->agno,
633 				&info->high);
634 
635 		error = query_fn(tp, info, &bt_cur, priv);
636 		if (error)
637 			goto err;
638 
639 		/*
640 		 * Set the AG low key to the start of the AG prior to
641 		 * moving on to the next AG.
642 		 */
643 		if (info->agno == start_ag) {
644 			info->low.rm_startblock = 0;
645 			info->low.rm_owner = 0;
646 			info->low.rm_offset = 0;
647 			info->low.rm_flags = 0;
648 		}
649 	}
650 
651 	/* Report any gap at the end of the AG */
652 	info->last = true;
653 	error = query_fn(tp, info, &bt_cur, priv);
654 	if (error)
655 		goto err;
656 
657 err:
658 	if (bt_cur)
659 		xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
660 							 XFS_BTREE_NOERROR);
661 	if (info->agf_bp) {
662 		xfs_trans_brelse(tp, info->agf_bp);
663 		info->agf_bp = NULL;
664 	}
665 
666 	return error;
667 }
668 
669 /* Actually query the rmap btree. */
670 STATIC int
671 xfs_getfsmap_datadev_rmapbt_query(
672 	struct xfs_trans		*tp,
673 	struct xfs_getfsmap_info	*info,
674 	struct xfs_btree_cur		**curpp,
675 	void				*priv)
676 {
677 	/* Report any gap at the end of the last AG. */
678 	if (info->last)
679 		return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
680 
681 	/* Allocate cursor for this AG and query_range it. */
682 	*curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
683 			info->agno);
684 	return xfs_rmap_query_range(*curpp, &info->low, &info->high,
685 			xfs_getfsmap_datadev_helper, info);
686 }
687 
688 /* Execute a getfsmap query against the regular data device rmapbt. */
689 STATIC int
690 xfs_getfsmap_datadev_rmapbt(
691 	struct xfs_trans		*tp,
692 	struct xfs_fsmap		*keys,
693 	struct xfs_getfsmap_info	*info)
694 {
695 	info->missing_owner = XFS_FMR_OWN_FREE;
696 	return __xfs_getfsmap_datadev(tp, keys, info,
697 			xfs_getfsmap_datadev_rmapbt_query, NULL);
698 }
699 
700 /* Actually query the bno btree. */
701 STATIC int
702 xfs_getfsmap_datadev_bnobt_query(
703 	struct xfs_trans		*tp,
704 	struct xfs_getfsmap_info	*info,
705 	struct xfs_btree_cur		**curpp,
706 	void				*priv)
707 {
708 	struct xfs_alloc_rec_incore	*key = priv;
709 
710 	/* Report any gap at the end of the last AG. */
711 	if (info->last)
712 		return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
713 
714 	/* Allocate cursor for this AG and query_range it. */
715 	*curpp = xfs_allocbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
716 			info->agno, XFS_BTNUM_BNO);
717 	key->ar_startblock = info->low.rm_startblock;
718 	key[1].ar_startblock = info->high.rm_startblock;
719 	return xfs_alloc_query_range(*curpp, key, &key[1],
720 			xfs_getfsmap_datadev_bnobt_helper, info);
721 }
722 
723 /* Execute a getfsmap query against the regular data device's bnobt. */
724 STATIC int
725 xfs_getfsmap_datadev_bnobt(
726 	struct xfs_trans		*tp,
727 	struct xfs_fsmap		*keys,
728 	struct xfs_getfsmap_info	*info)
729 {
730 	struct xfs_alloc_rec_incore	akeys[2];
731 
732 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
733 	return __xfs_getfsmap_datadev(tp, keys, info,
734 			xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
735 }
736 
737 /* Do we recognize the device? */
738 STATIC bool
739 xfs_getfsmap_is_valid_device(
740 	struct xfs_mount	*mp,
741 	struct xfs_fsmap	*fm)
742 {
743 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
744 	    fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
745 		return true;
746 	if (mp->m_logdev_targp &&
747 	    fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
748 		return true;
749 	if (mp->m_rtdev_targp &&
750 	    fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
751 		return true;
752 	return false;
753 }
754 
755 /* Ensure that the low key is less than the high key. */
756 STATIC bool
757 xfs_getfsmap_check_keys(
758 	struct xfs_fsmap		*low_key,
759 	struct xfs_fsmap		*high_key)
760 {
761 	if (low_key->fmr_device > high_key->fmr_device)
762 		return false;
763 	if (low_key->fmr_device < high_key->fmr_device)
764 		return true;
765 
766 	if (low_key->fmr_physical > high_key->fmr_physical)
767 		return false;
768 	if (low_key->fmr_physical < high_key->fmr_physical)
769 		return true;
770 
771 	if (low_key->fmr_owner > high_key->fmr_owner)
772 		return false;
773 	if (low_key->fmr_owner < high_key->fmr_owner)
774 		return true;
775 
776 	if (low_key->fmr_offset > high_key->fmr_offset)
777 		return false;
778 	if (low_key->fmr_offset < high_key->fmr_offset)
779 		return true;
780 
781 	return false;
782 }
783 
784 /*
785  * There are only two devices if we didn't configure RT devices at build time.
786  */
787 #ifdef CONFIG_XFS_RT
788 #define XFS_GETFSMAP_DEVS	3
789 #else
790 #define XFS_GETFSMAP_DEVS	2
791 #endif /* CONFIG_XFS_RT */
792 
793 /*
794  * Get filesystem's extents as described in head, and format for
795  * output.  Calls formatter to fill the user's buffer until all
796  * extents are mapped, until the passed-in head->fmh_count slots have
797  * been filled, or until the formatter short-circuits the loop, if it
798  * is tracking filled-in extents on its own.
799  *
800  * Key to Confusion
801  * ----------------
802  * There are multiple levels of keys and counters at work here:
803  * xfs_fsmap_head.fmh_keys	-- low and high fsmap keys passed in;
804  * 				   these reflect fs-wide sector addrs.
805  * dkeys			-- fmh_keys used to query each device;
806  * 				   these are fmh_keys but w/ the low key
807  * 				   bumped up by fmr_length.
808  * xfs_getfsmap_info.next_daddr	-- next disk addr we expect to see; this
809  *				   is how we detect gaps in the fsmap
810 				   records and report them.
811  * xfs_getfsmap_info.low/high	-- per-AG low/high keys computed from
812  * 				   dkeys; used to query the metadata.
813  */
814 int
815 xfs_getfsmap(
816 	struct xfs_mount		*mp,
817 	struct xfs_fsmap_head		*head,
818 	xfs_fsmap_format_t		formatter,
819 	void				*arg)
820 {
821 	struct xfs_trans		*tp = NULL;
822 	struct xfs_fsmap		dkeys[2];	/* per-dev keys */
823 	struct xfs_getfsmap_dev		handlers[XFS_GETFSMAP_DEVS];
824 	struct xfs_getfsmap_info	info = { NULL };
825 	bool				use_rmap;
826 	int				i;
827 	int				error = 0;
828 
829 	if (head->fmh_iflags & ~FMH_IF_VALID)
830 		return -EINVAL;
831 	if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
832 	    !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
833 		return -EINVAL;
834 
835 	use_rmap = capable(CAP_SYS_ADMIN) &&
836 		   xfs_sb_version_hasrmapbt(&mp->m_sb);
837 	head->fmh_entries = 0;
838 
839 	/* Set up our device handlers. */
840 	memset(handlers, 0, sizeof(handlers));
841 	handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
842 	if (use_rmap)
843 		handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
844 	else
845 		handlers[0].fn = xfs_getfsmap_datadev_bnobt;
846 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
847 		handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
848 		handlers[1].fn = xfs_getfsmap_logdev;
849 	}
850 #ifdef CONFIG_XFS_RT
851 	if (mp->m_rtdev_targp) {
852 		handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
853 		handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
854 	}
855 #endif /* CONFIG_XFS_RT */
856 
857 	xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
858 			xfs_getfsmap_dev_compare);
859 
860 	/*
861 	 * To continue where we left off, we allow userspace to use the
862 	 * last mapping from a previous call as the low key of the next.
863 	 * This is identified by a non-zero length in the low key. We
864 	 * have to increment the low key in this scenario to ensure we
865 	 * don't return the same mapping again, and instead return the
866 	 * very next mapping.
867 	 *
868 	 * If the low key mapping refers to file data, the same physical
869 	 * blocks could be mapped to several other files/offsets.
870 	 * According to rmapbt record ordering, the minimal next
871 	 * possible record for the block range is the next starting
872 	 * offset in the same inode. Therefore, bump the file offset to
873 	 * continue the search appropriately.  For all other low key
874 	 * mapping types (attr blocks, metadata), bump the physical
875 	 * offset as there can be no other mapping for the same physical
876 	 * block range.
877 	 */
878 	dkeys[0] = head->fmh_keys[0];
879 	if (dkeys[0].fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
880 		dkeys[0].fmr_physical += dkeys[0].fmr_length;
881 		dkeys[0].fmr_owner = 0;
882 		if (dkeys[0].fmr_offset)
883 			return -EINVAL;
884 	} else
885 		dkeys[0].fmr_offset += dkeys[0].fmr_length;
886 	dkeys[0].fmr_length = 0;
887 	memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
888 
889 	if (!xfs_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
890 		return -EINVAL;
891 
892 	info.next_daddr = head->fmh_keys[0].fmr_physical +
893 			  head->fmh_keys[0].fmr_length;
894 	info.formatter = formatter;
895 	info.format_arg = arg;
896 	info.head = head;
897 
898 	/* For each device we support... */
899 	for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
900 		/* Is this device within the range the user asked for? */
901 		if (!handlers[i].fn)
902 			continue;
903 		if (head->fmh_keys[0].fmr_device > handlers[i].dev)
904 			continue;
905 		if (head->fmh_keys[1].fmr_device < handlers[i].dev)
906 			break;
907 
908 		/*
909 		 * If this device number matches the high key, we have
910 		 * to pass the high key to the handler to limit the
911 		 * query results.  If the device number exceeds the
912 		 * low key, zero out the low key so that we get
913 		 * everything from the beginning.
914 		 */
915 		if (handlers[i].dev == head->fmh_keys[1].fmr_device)
916 			dkeys[1] = head->fmh_keys[1];
917 		if (handlers[i].dev > head->fmh_keys[0].fmr_device)
918 			memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
919 
920 		error = xfs_trans_alloc_empty(mp, &tp);
921 		if (error)
922 			break;
923 
924 		info.dev = handlers[i].dev;
925 		info.last = false;
926 		info.agno = NULLAGNUMBER;
927 		error = handlers[i].fn(tp, dkeys, &info);
928 		if (error)
929 			break;
930 		xfs_trans_cancel(tp);
931 		tp = NULL;
932 		info.next_daddr = 0;
933 	}
934 
935 	if (tp)
936 		xfs_trans_cancel(tp);
937 	head->fmh_oflags = FMH_OF_DEV_T;
938 	return error;
939 }
940