xref: /linux/fs/xfs/xfs_fsmap.c (revision 63ba5b0fb4f54db256ec43b3062b2606b383055d)
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_rtbitmap.h"
27 #include "xfs_ag.h"
28 
29 /* Convert an xfs_fsmap to an fsmap. */
30 static void
31 xfs_fsmap_from_internal(
32 	struct fsmap		*dest,
33 	struct xfs_fsmap	*src)
34 {
35 	dest->fmr_device = src->fmr_device;
36 	dest->fmr_flags = src->fmr_flags;
37 	dest->fmr_physical = BBTOB(src->fmr_physical);
38 	dest->fmr_owner = src->fmr_owner;
39 	dest->fmr_offset = BBTOB(src->fmr_offset);
40 	dest->fmr_length = BBTOB(src->fmr_length);
41 	dest->fmr_reserved[0] = 0;
42 	dest->fmr_reserved[1] = 0;
43 	dest->fmr_reserved[2] = 0;
44 }
45 
46 /* Convert an fsmap to an xfs_fsmap. */
47 void
48 xfs_fsmap_to_internal(
49 	struct xfs_fsmap	*dest,
50 	struct fsmap		*src)
51 {
52 	dest->fmr_device = src->fmr_device;
53 	dest->fmr_flags = src->fmr_flags;
54 	dest->fmr_physical = BTOBBT(src->fmr_physical);
55 	dest->fmr_owner = src->fmr_owner;
56 	dest->fmr_offset = BTOBBT(src->fmr_offset);
57 	dest->fmr_length = BTOBBT(src->fmr_length);
58 }
59 
60 /* Convert an fsmap owner into an rmapbt owner. */
61 static int
62 xfs_fsmap_owner_to_rmap(
63 	struct xfs_rmap_irec	*dest,
64 	const struct xfs_fsmap	*src)
65 {
66 	if (!(src->fmr_flags & FMR_OF_SPECIAL_OWNER)) {
67 		dest->rm_owner = src->fmr_owner;
68 		return 0;
69 	}
70 
71 	switch (src->fmr_owner) {
72 	case 0:			/* "lowest owner id possible" */
73 	case -1ULL:		/* "highest owner id possible" */
74 		dest->rm_owner = 0;
75 		break;
76 	case XFS_FMR_OWN_FREE:
77 		dest->rm_owner = XFS_RMAP_OWN_NULL;
78 		break;
79 	case XFS_FMR_OWN_UNKNOWN:
80 		dest->rm_owner = XFS_RMAP_OWN_UNKNOWN;
81 		break;
82 	case XFS_FMR_OWN_FS:
83 		dest->rm_owner = XFS_RMAP_OWN_FS;
84 		break;
85 	case XFS_FMR_OWN_LOG:
86 		dest->rm_owner = XFS_RMAP_OWN_LOG;
87 		break;
88 	case XFS_FMR_OWN_AG:
89 		dest->rm_owner = XFS_RMAP_OWN_AG;
90 		break;
91 	case XFS_FMR_OWN_INOBT:
92 		dest->rm_owner = XFS_RMAP_OWN_INOBT;
93 		break;
94 	case XFS_FMR_OWN_INODES:
95 		dest->rm_owner = XFS_RMAP_OWN_INODES;
96 		break;
97 	case XFS_FMR_OWN_REFC:
98 		dest->rm_owner = XFS_RMAP_OWN_REFC;
99 		break;
100 	case XFS_FMR_OWN_COW:
101 		dest->rm_owner = XFS_RMAP_OWN_COW;
102 		break;
103 	case XFS_FMR_OWN_DEFECTIVE:	/* not implemented */
104 		/* fall through */
105 	default:
106 		return -EINVAL;
107 	}
108 	return 0;
109 }
110 
111 /* Convert an rmapbt owner into an fsmap owner. */
112 static int
113 xfs_fsmap_owner_from_rmap(
114 	struct xfs_fsmap		*dest,
115 	const struct xfs_rmap_irec	*src)
116 {
117 	dest->fmr_flags = 0;
118 	if (!XFS_RMAP_NON_INODE_OWNER(src->rm_owner)) {
119 		dest->fmr_owner = src->rm_owner;
120 		return 0;
121 	}
122 	dest->fmr_flags |= FMR_OF_SPECIAL_OWNER;
123 
124 	switch (src->rm_owner) {
125 	case XFS_RMAP_OWN_FS:
126 		dest->fmr_owner = XFS_FMR_OWN_FS;
127 		break;
128 	case XFS_RMAP_OWN_LOG:
129 		dest->fmr_owner = XFS_FMR_OWN_LOG;
130 		break;
131 	case XFS_RMAP_OWN_AG:
132 		dest->fmr_owner = XFS_FMR_OWN_AG;
133 		break;
134 	case XFS_RMAP_OWN_INOBT:
135 		dest->fmr_owner = XFS_FMR_OWN_INOBT;
136 		break;
137 	case XFS_RMAP_OWN_INODES:
138 		dest->fmr_owner = XFS_FMR_OWN_INODES;
139 		break;
140 	case XFS_RMAP_OWN_REFC:
141 		dest->fmr_owner = XFS_FMR_OWN_REFC;
142 		break;
143 	case XFS_RMAP_OWN_COW:
144 		dest->fmr_owner = XFS_FMR_OWN_COW;
145 		break;
146 	case XFS_RMAP_OWN_NULL:	/* "free" */
147 		dest->fmr_owner = XFS_FMR_OWN_FREE;
148 		break;
149 	default:
150 		ASSERT(0);
151 		return -EFSCORRUPTED;
152 	}
153 	return 0;
154 }
155 
156 /* getfsmap query state */
157 struct xfs_getfsmap_info {
158 	struct xfs_fsmap_head	*head;
159 	struct fsmap		*fsmap_recs;	/* mapping records */
160 	struct xfs_buf		*agf_bp;	/* AGF, for refcount queries */
161 	struct xfs_perag	*pag;		/* AG info, if applicable */
162 	xfs_daddr_t		next_daddr;	/* next daddr we expect */
163 	/* daddr of low fsmap key when we're using the rtbitmap */
164 	xfs_daddr_t		low_daddr;
165 	u64			missing_owner;	/* owner of holes */
166 	u32			dev;		/* device id */
167 	/*
168 	 * Low rmap key for the query.  If low.rm_blockcount is nonzero, this
169 	 * is the second (or later) call to retrieve the recordset in pieces.
170 	 * xfs_getfsmap_rec_before_start will compare all records retrieved
171 	 * by the rmapbt query to filter out any records that start before
172 	 * the last record.
173 	 */
174 	struct xfs_rmap_irec	low;
175 	struct xfs_rmap_irec	high;		/* high rmap key */
176 	bool			last;		/* last extent? */
177 };
178 
179 /* Associate a device with a getfsmap handler. */
180 struct xfs_getfsmap_dev {
181 	u32			dev;
182 	int			(*fn)(struct xfs_trans *tp,
183 				      const struct xfs_fsmap *keys,
184 				      struct xfs_getfsmap_info *info);
185 };
186 
187 /* Compare two getfsmap device handlers. */
188 static int
189 xfs_getfsmap_dev_compare(
190 	const void			*p1,
191 	const void			*p2)
192 {
193 	const struct xfs_getfsmap_dev	*d1 = p1;
194 	const struct xfs_getfsmap_dev	*d2 = p2;
195 
196 	return d1->dev - d2->dev;
197 }
198 
199 /* Decide if this mapping is shared. */
200 STATIC int
201 xfs_getfsmap_is_shared(
202 	struct xfs_trans		*tp,
203 	struct xfs_getfsmap_info	*info,
204 	const struct xfs_rmap_irec	*rec,
205 	bool				*stat)
206 {
207 	struct xfs_mount		*mp = tp->t_mountp;
208 	struct xfs_btree_cur		*cur;
209 	xfs_agblock_t			fbno;
210 	xfs_extlen_t			flen;
211 	int				error;
212 
213 	*stat = false;
214 	if (!xfs_has_reflink(mp))
215 		return 0;
216 	/* rt files will have no perag structure */
217 	if (!info->pag)
218 		return 0;
219 
220 	/* Are there any shared blocks here? */
221 	flen = 0;
222 	cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp, info->pag);
223 
224 	error = xfs_refcount_find_shared(cur, rec->rm_startblock,
225 			rec->rm_blockcount, &fbno, &flen, false);
226 
227 	xfs_btree_del_cursor(cur, error);
228 	if (error)
229 		return error;
230 
231 	*stat = flen > 0;
232 	return 0;
233 }
234 
235 static inline void
236 xfs_getfsmap_format(
237 	struct xfs_mount		*mp,
238 	struct xfs_fsmap		*xfm,
239 	struct xfs_getfsmap_info	*info)
240 {
241 	struct fsmap			*rec;
242 
243 	trace_xfs_getfsmap_mapping(mp, xfm);
244 
245 	rec = &info->fsmap_recs[info->head->fmh_entries++];
246 	xfs_fsmap_from_internal(rec, xfm);
247 }
248 
249 static inline bool
250 xfs_getfsmap_rec_before_start(
251 	struct xfs_getfsmap_info	*info,
252 	const struct xfs_rmap_irec	*rec,
253 	xfs_daddr_t			rec_daddr)
254 {
255 	if (info->low_daddr != -1ULL)
256 		return rec_daddr < info->low_daddr;
257 	if (info->low.rm_blockcount)
258 		return xfs_rmap_compare(rec, &info->low) < 0;
259 	return false;
260 }
261 
262 /*
263  * Format a reverse mapping for getfsmap, having translated rm_startblock
264  * into the appropriate daddr units.  Pass in a nonzero @len_daddr if the
265  * length could be larger than rm_blockcount in struct xfs_rmap_irec.
266  */
267 STATIC int
268 xfs_getfsmap_helper(
269 	struct xfs_trans		*tp,
270 	struct xfs_getfsmap_info	*info,
271 	const struct xfs_rmap_irec	*rec,
272 	xfs_daddr_t			rec_daddr,
273 	xfs_daddr_t			len_daddr)
274 {
275 	struct xfs_fsmap		fmr;
276 	struct xfs_mount		*mp = tp->t_mountp;
277 	bool				shared;
278 	int				error;
279 
280 	if (fatal_signal_pending(current))
281 		return -EINTR;
282 
283 	if (len_daddr == 0)
284 		len_daddr = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
285 
286 	/*
287 	 * Filter out records that start before our startpoint, if the
288 	 * caller requested that.
289 	 */
290 	if (xfs_getfsmap_rec_before_start(info, rec, rec_daddr)) {
291 		rec_daddr += len_daddr;
292 		if (info->next_daddr < rec_daddr)
293 			info->next_daddr = rec_daddr;
294 		return 0;
295 	}
296 
297 	/* Are we just counting mappings? */
298 	if (info->head->fmh_count == 0) {
299 		if (info->head->fmh_entries == UINT_MAX)
300 			return -ECANCELED;
301 
302 		if (rec_daddr > info->next_daddr)
303 			info->head->fmh_entries++;
304 
305 		if (info->last)
306 			return 0;
307 
308 		info->head->fmh_entries++;
309 
310 		rec_daddr += len_daddr;
311 		if (info->next_daddr < rec_daddr)
312 			info->next_daddr = rec_daddr;
313 		return 0;
314 	}
315 
316 	/*
317 	 * If the record starts past the last physical block we saw,
318 	 * then we've found a gap.  Report the gap as being owned by
319 	 * whatever the caller specified is the missing owner.
320 	 */
321 	if (rec_daddr > info->next_daddr) {
322 		if (info->head->fmh_entries >= info->head->fmh_count)
323 			return -ECANCELED;
324 
325 		fmr.fmr_device = info->dev;
326 		fmr.fmr_physical = info->next_daddr;
327 		fmr.fmr_owner = info->missing_owner;
328 		fmr.fmr_offset = 0;
329 		fmr.fmr_length = rec_daddr - info->next_daddr;
330 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
331 		xfs_getfsmap_format(mp, &fmr, info);
332 	}
333 
334 	if (info->last)
335 		goto out;
336 
337 	/* Fill out the extent we found */
338 	if (info->head->fmh_entries >= info->head->fmh_count)
339 		return -ECANCELED;
340 
341 	trace_xfs_fsmap_mapping(mp, info->dev,
342 			info->pag ? info->pag->pag_agno : NULLAGNUMBER, rec);
343 
344 	fmr.fmr_device = info->dev;
345 	fmr.fmr_physical = rec_daddr;
346 	error = xfs_fsmap_owner_from_rmap(&fmr, rec);
347 	if (error)
348 		return error;
349 	fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
350 	fmr.fmr_length = len_daddr;
351 	if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
352 		fmr.fmr_flags |= FMR_OF_PREALLOC;
353 	if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
354 		fmr.fmr_flags |= FMR_OF_ATTR_FORK;
355 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
356 		fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
357 	if (fmr.fmr_flags == 0) {
358 		error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
359 		if (error)
360 			return error;
361 		if (shared)
362 			fmr.fmr_flags |= FMR_OF_SHARED;
363 	}
364 
365 	xfs_getfsmap_format(mp, &fmr, info);
366 out:
367 	rec_daddr += len_daddr;
368 	if (info->next_daddr < rec_daddr)
369 		info->next_daddr = rec_daddr;
370 	return 0;
371 }
372 
373 /* Transform a rmapbt irec into a fsmap */
374 STATIC int
375 xfs_getfsmap_datadev_helper(
376 	struct xfs_btree_cur		*cur,
377 	const struct xfs_rmap_irec	*rec,
378 	void				*priv)
379 {
380 	struct xfs_mount		*mp = cur->bc_mp;
381 	struct xfs_getfsmap_info	*info = priv;
382 	xfs_fsblock_t			fsb;
383 	xfs_daddr_t			rec_daddr;
384 
385 	fsb = XFS_AGB_TO_FSB(mp, cur->bc_ag.pag->pag_agno, rec->rm_startblock);
386 	rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
387 
388 	return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr, 0);
389 }
390 
391 /* Transform a bnobt irec into a fsmap */
392 STATIC int
393 xfs_getfsmap_datadev_bnobt_helper(
394 	struct xfs_btree_cur		*cur,
395 	const struct xfs_alloc_rec_incore *rec,
396 	void				*priv)
397 {
398 	struct xfs_mount		*mp = cur->bc_mp;
399 	struct xfs_getfsmap_info	*info = priv;
400 	struct xfs_rmap_irec		irec;
401 	xfs_daddr_t			rec_daddr;
402 
403 	rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_ag.pag->pag_agno,
404 			rec->ar_startblock);
405 
406 	irec.rm_startblock = rec->ar_startblock;
407 	irec.rm_blockcount = rec->ar_blockcount;
408 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
409 	irec.rm_offset = 0;
410 	irec.rm_flags = 0;
411 
412 	return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr, 0);
413 }
414 
415 /* Set rmap flags based on the getfsmap flags */
416 static void
417 xfs_getfsmap_set_irec_flags(
418 	struct xfs_rmap_irec	*irec,
419 	const struct xfs_fsmap	*fmr)
420 {
421 	irec->rm_flags = 0;
422 	if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
423 		irec->rm_flags |= XFS_RMAP_ATTR_FORK;
424 	if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
425 		irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
426 	if (fmr->fmr_flags & FMR_OF_PREALLOC)
427 		irec->rm_flags |= XFS_RMAP_UNWRITTEN;
428 }
429 
430 /* Execute a getfsmap query against the log device. */
431 STATIC int
432 xfs_getfsmap_logdev(
433 	struct xfs_trans		*tp,
434 	const struct xfs_fsmap		*keys,
435 	struct xfs_getfsmap_info	*info)
436 {
437 	struct xfs_mount		*mp = tp->t_mountp;
438 	struct xfs_rmap_irec		rmap;
439 	xfs_daddr_t			rec_daddr, len_daddr;
440 	xfs_fsblock_t			start_fsb, end_fsb;
441 	uint64_t			eofs;
442 
443 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
444 	if (keys[0].fmr_physical >= eofs)
445 		return 0;
446 	start_fsb = XFS_BB_TO_FSBT(mp,
447 				keys[0].fmr_physical + keys[0].fmr_length);
448 	end_fsb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
449 
450 	/* Adjust the low key if we are continuing from where we left off. */
451 	if (keys[0].fmr_length > 0)
452 		info->low_daddr = XFS_FSB_TO_BB(mp, start_fsb);
453 
454 	trace_xfs_fsmap_low_key_linear(mp, info->dev, start_fsb);
455 	trace_xfs_fsmap_high_key_linear(mp, info->dev, end_fsb);
456 
457 	if (start_fsb > 0)
458 		return 0;
459 
460 	/* Fabricate an rmap entry for the external log device. */
461 	rmap.rm_startblock = 0;
462 	rmap.rm_blockcount = mp->m_sb.sb_logblocks;
463 	rmap.rm_owner = XFS_RMAP_OWN_LOG;
464 	rmap.rm_offset = 0;
465 	rmap.rm_flags = 0;
466 
467 	rec_daddr = XFS_FSB_TO_BB(mp, rmap.rm_startblock);
468 	len_daddr = XFS_FSB_TO_BB(mp, rmap.rm_blockcount);
469 	return xfs_getfsmap_helper(tp, info, &rmap, rec_daddr, len_daddr);
470 }
471 
472 #ifdef CONFIG_XFS_RT
473 /* Transform a rtbitmap "record" into a fsmap */
474 STATIC int
475 xfs_getfsmap_rtdev_rtbitmap_helper(
476 	struct xfs_mount		*mp,
477 	struct xfs_trans		*tp,
478 	const struct xfs_rtalloc_rec	*rec,
479 	void				*priv)
480 {
481 	struct xfs_getfsmap_info	*info = priv;
482 	struct xfs_rmap_irec		irec;
483 	xfs_rtblock_t			rtbno;
484 	xfs_daddr_t			rec_daddr, len_daddr;
485 
486 	rtbno = xfs_rtx_to_rtb(mp, rec->ar_startext);
487 	rec_daddr = XFS_FSB_TO_BB(mp, rtbno);
488 	irec.rm_startblock = rtbno;
489 
490 	rtbno = xfs_rtx_to_rtb(mp, rec->ar_extcount);
491 	len_daddr = XFS_FSB_TO_BB(mp, rtbno);
492 	irec.rm_blockcount = rtbno;
493 
494 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
495 	irec.rm_offset = 0;
496 	irec.rm_flags = 0;
497 
498 	return xfs_getfsmap_helper(tp, info, &irec, rec_daddr, len_daddr);
499 }
500 
501 /* Execute a getfsmap query against the realtime device rtbitmap. */
502 STATIC int
503 xfs_getfsmap_rtdev_rtbitmap(
504 	struct xfs_trans		*tp,
505 	const struct xfs_fsmap		*keys,
506 	struct xfs_getfsmap_info	*info)
507 {
508 
509 	struct xfs_rtalloc_rec		alow = { 0 };
510 	struct xfs_rtalloc_rec		ahigh = { 0 };
511 	struct xfs_mount		*mp = tp->t_mountp;
512 	xfs_rtblock_t			start_rtb;
513 	xfs_rtblock_t			end_rtb;
514 	uint64_t			eofs;
515 	int				error;
516 
517 	eofs = XFS_FSB_TO_BB(mp, xfs_rtx_to_rtb(mp, mp->m_sb.sb_rextents));
518 	if (keys[0].fmr_physical >= eofs)
519 		return 0;
520 	start_rtb = XFS_BB_TO_FSBT(mp,
521 				keys[0].fmr_physical + keys[0].fmr_length);
522 	end_rtb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
523 
524 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
525 
526 	/* Adjust the low key if we are continuing from where we left off. */
527 	if (keys[0].fmr_length > 0) {
528 		info->low_daddr = XFS_FSB_TO_BB(mp, start_rtb);
529 		if (info->low_daddr >= eofs)
530 			return 0;
531 	}
532 
533 	trace_xfs_fsmap_low_key_linear(mp, info->dev, start_rtb);
534 	trace_xfs_fsmap_high_key_linear(mp, info->dev, end_rtb);
535 
536 	xfs_rtbitmap_lock_shared(mp, XFS_RBMLOCK_BITMAP);
537 
538 	/*
539 	 * Set up query parameters to return free rtextents covering the range
540 	 * we want.
541 	 */
542 	alow.ar_startext = xfs_rtb_to_rtx(mp, start_rtb);
543 	ahigh.ar_startext = xfs_rtb_to_rtxup(mp, end_rtb);
544 	error = xfs_rtalloc_query_range(mp, tp, &alow, &ahigh,
545 			xfs_getfsmap_rtdev_rtbitmap_helper, info);
546 	if (error)
547 		goto err;
548 
549 	/*
550 	 * Report any gaps at the end of the rtbitmap by simulating a null
551 	 * rmap starting at the block after the end of the query range.
552 	 */
553 	info->last = true;
554 	ahigh.ar_startext = min(mp->m_sb.sb_rextents, ahigh.ar_startext);
555 
556 	error = xfs_getfsmap_rtdev_rtbitmap_helper(mp, tp, &ahigh, info);
557 	if (error)
558 		goto err;
559 err:
560 	xfs_rtbitmap_unlock_shared(mp, XFS_RBMLOCK_BITMAP);
561 	return error;
562 }
563 #endif /* CONFIG_XFS_RT */
564 
565 static inline bool
566 rmap_not_shareable(struct xfs_mount *mp, const struct xfs_rmap_irec *r)
567 {
568 	if (!xfs_has_reflink(mp))
569 		return true;
570 	if (XFS_RMAP_NON_INODE_OWNER(r->rm_owner))
571 		return true;
572 	if (r->rm_flags & (XFS_RMAP_ATTR_FORK | XFS_RMAP_BMBT_BLOCK |
573 			   XFS_RMAP_UNWRITTEN))
574 		return true;
575 	return false;
576 }
577 
578 /* Execute a getfsmap query against the regular data device. */
579 STATIC int
580 __xfs_getfsmap_datadev(
581 	struct xfs_trans		*tp,
582 	const struct xfs_fsmap		*keys,
583 	struct xfs_getfsmap_info	*info,
584 	int				(*query_fn)(struct xfs_trans *,
585 						    struct xfs_getfsmap_info *,
586 						    struct xfs_btree_cur **,
587 						    void *),
588 	void				*priv)
589 {
590 	struct xfs_mount		*mp = tp->t_mountp;
591 	struct xfs_perag		*pag;
592 	struct xfs_btree_cur		*bt_cur = NULL;
593 	xfs_fsblock_t			start_fsb;
594 	xfs_fsblock_t			end_fsb;
595 	xfs_agnumber_t			start_ag;
596 	xfs_agnumber_t			end_ag;
597 	uint64_t			eofs;
598 	int				error = 0;
599 
600 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
601 	if (keys[0].fmr_physical >= eofs)
602 		return 0;
603 	start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
604 	end_fsb = XFS_DADDR_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
605 
606 	/*
607 	 * Convert the fsmap low/high keys to AG based keys.  Initialize
608 	 * low to the fsmap low key and max out the high key to the end
609 	 * of the AG.
610 	 */
611 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
612 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
613 	if (error)
614 		return error;
615 	info->low.rm_blockcount = XFS_BB_TO_FSBT(mp, keys[0].fmr_length);
616 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
617 
618 	/* Adjust the low key if we are continuing from where we left off. */
619 	if (info->low.rm_blockcount == 0) {
620 		/* No previous record from which to continue */
621 	} else if (rmap_not_shareable(mp, &info->low)) {
622 		/* Last record seen was an unshareable extent */
623 		info->low.rm_owner = 0;
624 		info->low.rm_offset = 0;
625 
626 		start_fsb += info->low.rm_blockcount;
627 		if (XFS_FSB_TO_DADDR(mp, start_fsb) >= eofs)
628 			return 0;
629 	} else {
630 		/* Last record seen was a shareable file data extent */
631 		info->low.rm_offset += info->low.rm_blockcount;
632 	}
633 	info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
634 
635 	info->high.rm_startblock = -1U;
636 	info->high.rm_owner = ULLONG_MAX;
637 	info->high.rm_offset = ULLONG_MAX;
638 	info->high.rm_blockcount = 0;
639 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
640 
641 	start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
642 	end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
643 
644 	for_each_perag_range(mp, start_ag, end_ag, pag) {
645 		/*
646 		 * Set the AG high key from the fsmap high key if this
647 		 * is the last AG that we're querying.
648 		 */
649 		info->pag = pag;
650 		if (pag->pag_agno == end_ag) {
651 			info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
652 					end_fsb);
653 			info->high.rm_offset = XFS_BB_TO_FSBT(mp,
654 					keys[1].fmr_offset);
655 			error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
656 			if (error)
657 				break;
658 			xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
659 		}
660 
661 		if (bt_cur) {
662 			xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
663 			bt_cur = NULL;
664 			xfs_trans_brelse(tp, info->agf_bp);
665 			info->agf_bp = NULL;
666 		}
667 
668 		error = xfs_alloc_read_agf(pag, tp, 0, &info->agf_bp);
669 		if (error)
670 			break;
671 
672 		trace_xfs_fsmap_low_key(mp, info->dev, pag->pag_agno,
673 				&info->low);
674 		trace_xfs_fsmap_high_key(mp, info->dev, pag->pag_agno,
675 				&info->high);
676 
677 		error = query_fn(tp, info, &bt_cur, priv);
678 		if (error)
679 			break;
680 
681 		/*
682 		 * Set the AG low key to the start of the AG prior to
683 		 * moving on to the next AG.
684 		 */
685 		if (pag->pag_agno == start_ag)
686 			memset(&info->low, 0, sizeof(info->low));
687 
688 		/*
689 		 * If this is the last AG, report any gap at the end of it
690 		 * before we drop the reference to the perag when the loop
691 		 * terminates.
692 		 */
693 		if (pag->pag_agno == end_ag) {
694 			info->last = true;
695 			error = query_fn(tp, info, &bt_cur, priv);
696 			if (error)
697 				break;
698 		}
699 		info->pag = NULL;
700 	}
701 
702 	if (bt_cur)
703 		xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
704 							 XFS_BTREE_NOERROR);
705 	if (info->agf_bp) {
706 		xfs_trans_brelse(tp, info->agf_bp);
707 		info->agf_bp = NULL;
708 	}
709 	if (info->pag) {
710 		xfs_perag_rele(info->pag);
711 		info->pag = NULL;
712 	} else if (pag) {
713 		/* loop termination case */
714 		xfs_perag_rele(pag);
715 	}
716 
717 	return error;
718 }
719 
720 /* Actually query the rmap btree. */
721 STATIC int
722 xfs_getfsmap_datadev_rmapbt_query(
723 	struct xfs_trans		*tp,
724 	struct xfs_getfsmap_info	*info,
725 	struct xfs_btree_cur		**curpp,
726 	void				*priv)
727 {
728 	/* Report any gap at the end of the last AG. */
729 	if (info->last)
730 		return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
731 
732 	/* Allocate cursor for this AG and query_range it. */
733 	*curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
734 			info->pag);
735 	return xfs_rmap_query_range(*curpp, &info->low, &info->high,
736 			xfs_getfsmap_datadev_helper, info);
737 }
738 
739 /* Execute a getfsmap query against the regular data device rmapbt. */
740 STATIC int
741 xfs_getfsmap_datadev_rmapbt(
742 	struct xfs_trans		*tp,
743 	const struct xfs_fsmap		*keys,
744 	struct xfs_getfsmap_info	*info)
745 {
746 	info->missing_owner = XFS_FMR_OWN_FREE;
747 	return __xfs_getfsmap_datadev(tp, keys, info,
748 			xfs_getfsmap_datadev_rmapbt_query, NULL);
749 }
750 
751 /* Actually query the bno btree. */
752 STATIC int
753 xfs_getfsmap_datadev_bnobt_query(
754 	struct xfs_trans		*tp,
755 	struct xfs_getfsmap_info	*info,
756 	struct xfs_btree_cur		**curpp,
757 	void				*priv)
758 {
759 	struct xfs_alloc_rec_incore	*key = priv;
760 
761 	/* Report any gap at the end of the last AG. */
762 	if (info->last)
763 		return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
764 
765 	/* Allocate cursor for this AG and query_range it. */
766 	*curpp = xfs_bnobt_init_cursor(tp->t_mountp, tp, info->agf_bp,
767 			info->pag);
768 	key->ar_startblock = info->low.rm_startblock;
769 	key[1].ar_startblock = info->high.rm_startblock;
770 	return xfs_alloc_query_range(*curpp, key, &key[1],
771 			xfs_getfsmap_datadev_bnobt_helper, info);
772 }
773 
774 /* Execute a getfsmap query against the regular data device's bnobt. */
775 STATIC int
776 xfs_getfsmap_datadev_bnobt(
777 	struct xfs_trans		*tp,
778 	const struct xfs_fsmap		*keys,
779 	struct xfs_getfsmap_info	*info)
780 {
781 	struct xfs_alloc_rec_incore	akeys[2];
782 
783 	memset(akeys, 0, sizeof(akeys));
784 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
785 	return __xfs_getfsmap_datadev(tp, keys, info,
786 			xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
787 }
788 
789 /* Do we recognize the device? */
790 STATIC bool
791 xfs_getfsmap_is_valid_device(
792 	struct xfs_mount	*mp,
793 	struct xfs_fsmap	*fm)
794 {
795 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
796 	    fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
797 		return true;
798 	if (mp->m_logdev_targp &&
799 	    fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
800 		return true;
801 	if (mp->m_rtdev_targp &&
802 	    fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
803 		return true;
804 	return false;
805 }
806 
807 /* Ensure that the low key is less than the high key. */
808 STATIC bool
809 xfs_getfsmap_check_keys(
810 	struct xfs_fsmap		*low_key,
811 	struct xfs_fsmap		*high_key)
812 {
813 	if (low_key->fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
814 		if (low_key->fmr_offset)
815 			return false;
816 	}
817 	if (high_key->fmr_flags != -1U &&
818 	    (high_key->fmr_flags & (FMR_OF_SPECIAL_OWNER |
819 				    FMR_OF_EXTENT_MAP))) {
820 		if (high_key->fmr_offset && high_key->fmr_offset != -1ULL)
821 			return false;
822 	}
823 	if (high_key->fmr_length && high_key->fmr_length != -1ULL)
824 		return false;
825 
826 	if (low_key->fmr_device > high_key->fmr_device)
827 		return false;
828 	if (low_key->fmr_device < high_key->fmr_device)
829 		return true;
830 
831 	if (low_key->fmr_physical > high_key->fmr_physical)
832 		return false;
833 	if (low_key->fmr_physical < high_key->fmr_physical)
834 		return true;
835 
836 	if (low_key->fmr_owner > high_key->fmr_owner)
837 		return false;
838 	if (low_key->fmr_owner < high_key->fmr_owner)
839 		return true;
840 
841 	if (low_key->fmr_offset > high_key->fmr_offset)
842 		return false;
843 	if (low_key->fmr_offset < high_key->fmr_offset)
844 		return true;
845 
846 	return false;
847 }
848 
849 /*
850  * There are only two devices if we didn't configure RT devices at build time.
851  */
852 #ifdef CONFIG_XFS_RT
853 #define XFS_GETFSMAP_DEVS	3
854 #else
855 #define XFS_GETFSMAP_DEVS	2
856 #endif /* CONFIG_XFS_RT */
857 
858 /*
859  * Get filesystem's extents as described in head, and format for output. Fills
860  * in the supplied records array until there are no more reverse mappings to
861  * return or head.fmh_entries == head.fmh_count.  In the second case, this
862  * function returns -ECANCELED to indicate that more records would have been
863  * returned.
864  *
865  * Key to Confusion
866  * ----------------
867  * There are multiple levels of keys and counters at work here:
868  * xfs_fsmap_head.fmh_keys	-- low and high fsmap keys passed in;
869  *				   these reflect fs-wide sector addrs.
870  * dkeys			-- fmh_keys used to query each device;
871  *				   these are fmh_keys but w/ the low key
872  *				   bumped up by fmr_length.
873  * xfs_getfsmap_info.next_daddr	-- next disk addr we expect to see; this
874  *				   is how we detect gaps in the fsmap
875 				   records and report them.
876  * xfs_getfsmap_info.low/high	-- per-AG low/high keys computed from
877  *				   dkeys; used to query the metadata.
878  */
879 int
880 xfs_getfsmap(
881 	struct xfs_mount		*mp,
882 	struct xfs_fsmap_head		*head,
883 	struct fsmap			*fsmap_recs)
884 {
885 	struct xfs_trans		*tp = NULL;
886 	struct xfs_fsmap		dkeys[2];	/* per-dev keys */
887 	struct xfs_getfsmap_dev		handlers[XFS_GETFSMAP_DEVS];
888 	struct xfs_getfsmap_info	info = { NULL };
889 	bool				use_rmap;
890 	int				i;
891 	int				error = 0;
892 
893 	if (head->fmh_iflags & ~FMH_IF_VALID)
894 		return -EINVAL;
895 	if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
896 	    !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
897 		return -EINVAL;
898 	if (!xfs_getfsmap_check_keys(&head->fmh_keys[0], &head->fmh_keys[1]))
899 		return -EINVAL;
900 
901 	use_rmap = xfs_has_rmapbt(mp) &&
902 		   has_capability_noaudit(current, CAP_SYS_ADMIN);
903 	head->fmh_entries = 0;
904 
905 	/* Set up our device handlers. */
906 	memset(handlers, 0, sizeof(handlers));
907 	handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
908 	if (use_rmap)
909 		handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
910 	else
911 		handlers[0].fn = xfs_getfsmap_datadev_bnobt;
912 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
913 		handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
914 		handlers[1].fn = xfs_getfsmap_logdev;
915 	}
916 #ifdef CONFIG_XFS_RT
917 	if (mp->m_rtdev_targp) {
918 		handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
919 		handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
920 	}
921 #endif /* CONFIG_XFS_RT */
922 
923 	xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
924 			xfs_getfsmap_dev_compare);
925 
926 	/*
927 	 * To continue where we left off, we allow userspace to use the
928 	 * last mapping from a previous call as the low key of the next.
929 	 * This is identified by a non-zero length in the low key. We
930 	 * have to increment the low key in this scenario to ensure we
931 	 * don't return the same mapping again, and instead return the
932 	 * very next mapping.
933 	 *
934 	 * If the low key mapping refers to file data, the same physical
935 	 * blocks could be mapped to several other files/offsets.
936 	 * According to rmapbt record ordering, the minimal next
937 	 * possible record for the block range is the next starting
938 	 * offset in the same inode. Therefore, each fsmap backend bumps
939 	 * the file offset to continue the search appropriately.  For
940 	 * all other low key mapping types (attr blocks, metadata), each
941 	 * fsmap backend bumps the physical offset as there can be no
942 	 * other mapping for the same physical block range.
943 	 */
944 	dkeys[0] = head->fmh_keys[0];
945 	memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
946 
947 	info.next_daddr = head->fmh_keys[0].fmr_physical +
948 			  head->fmh_keys[0].fmr_length;
949 	info.fsmap_recs = fsmap_recs;
950 	info.head = head;
951 
952 	/* For each device we support... */
953 	for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
954 		/* Is this device within the range the user asked for? */
955 		if (!handlers[i].fn)
956 			continue;
957 		if (head->fmh_keys[0].fmr_device > handlers[i].dev)
958 			continue;
959 		if (head->fmh_keys[1].fmr_device < handlers[i].dev)
960 			break;
961 
962 		/*
963 		 * If this device number matches the high key, we have
964 		 * to pass the high key to the handler to limit the
965 		 * query results.  If the device number exceeds the
966 		 * low key, zero out the low key so that we get
967 		 * everything from the beginning.
968 		 */
969 		if (handlers[i].dev == head->fmh_keys[1].fmr_device)
970 			dkeys[1] = head->fmh_keys[1];
971 		if (handlers[i].dev > head->fmh_keys[0].fmr_device)
972 			memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
973 
974 		/*
975 		 * Grab an empty transaction so that we can use its recursive
976 		 * buffer locking abilities to detect cycles in the rmapbt
977 		 * without deadlocking.
978 		 */
979 		error = xfs_trans_alloc_empty(mp, &tp);
980 		if (error)
981 			break;
982 
983 		info.dev = handlers[i].dev;
984 		info.last = false;
985 		info.pag = NULL;
986 		info.low_daddr = -1ULL;
987 		info.low.rm_blockcount = 0;
988 		error = handlers[i].fn(tp, dkeys, &info);
989 		if (error)
990 			break;
991 		xfs_trans_cancel(tp);
992 		tp = NULL;
993 		info.next_daddr = 0;
994 	}
995 
996 	if (tp)
997 		xfs_trans_cancel(tp);
998 	head->fmh_oflags = FMH_OF_DEV_T;
999 	return error;
1000 }
1001