xref: /linux/fs/xfs/xfs_fsmap.c (revision 5cd2340cb6a383d04fd88e48fabc2a21a909d6a1)
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 = src->fmr_owner;
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 	xfs_daddr_t		end_daddr;	/* daddr of high fsmap key */
166 	u64			missing_owner;	/* owner of holes */
167 	u32			dev;		/* device id */
168 	/*
169 	 * Low rmap key for the query.  If low.rm_blockcount is nonzero, this
170 	 * is the second (or later) call to retrieve the recordset in pieces.
171 	 * xfs_getfsmap_rec_before_start will compare all records retrieved
172 	 * by the rmapbt query to filter out any records that start before
173 	 * the last record.
174 	 */
175 	struct xfs_rmap_irec	low;
176 	struct xfs_rmap_irec	high;		/* high rmap key */
177 	bool			last;		/* last extent? */
178 };
179 
180 /* Associate a device with a getfsmap handler. */
181 struct xfs_getfsmap_dev {
182 	u32			dev;
183 	int			(*fn)(struct xfs_trans *tp,
184 				      const struct xfs_fsmap *keys,
185 				      struct xfs_getfsmap_info *info);
186 	sector_t		nr_sectors;
187 };
188 
189 /* Compare two getfsmap device handlers. */
190 static int
191 xfs_getfsmap_dev_compare(
192 	const void			*p1,
193 	const void			*p2)
194 {
195 	const struct xfs_getfsmap_dev	*d1 = p1;
196 	const struct xfs_getfsmap_dev	*d2 = p2;
197 
198 	return d1->dev - d2->dev;
199 }
200 
201 /* Decide if this mapping is shared. */
202 STATIC int
203 xfs_getfsmap_is_shared(
204 	struct xfs_trans		*tp,
205 	struct xfs_getfsmap_info	*info,
206 	const struct xfs_rmap_irec	*rec,
207 	bool				*stat)
208 {
209 	struct xfs_mount		*mp = tp->t_mountp;
210 	struct xfs_btree_cur		*cur;
211 	xfs_agblock_t			fbno;
212 	xfs_extlen_t			flen;
213 	int				error;
214 
215 	*stat = false;
216 	if (!xfs_has_reflink(mp))
217 		return 0;
218 	/* rt files will have no perag structure */
219 	if (!info->pag)
220 		return 0;
221 
222 	/* Are there any shared blocks here? */
223 	flen = 0;
224 	cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp, info->pag);
225 
226 	error = xfs_refcount_find_shared(cur, rec->rm_startblock,
227 			rec->rm_blockcount, &fbno, &flen, false);
228 
229 	xfs_btree_del_cursor(cur, error);
230 	if (error)
231 		return error;
232 
233 	*stat = flen > 0;
234 	return 0;
235 }
236 
237 static inline void
238 xfs_getfsmap_format(
239 	struct xfs_mount		*mp,
240 	struct xfs_fsmap		*xfm,
241 	struct xfs_getfsmap_info	*info)
242 {
243 	struct fsmap			*rec;
244 
245 	trace_xfs_getfsmap_mapping(mp, xfm);
246 
247 	rec = &info->fsmap_recs[info->head->fmh_entries++];
248 	xfs_fsmap_from_internal(rec, xfm);
249 }
250 
251 static inline bool
252 xfs_getfsmap_rec_before_start(
253 	struct xfs_getfsmap_info	*info,
254 	const struct xfs_rmap_irec	*rec,
255 	xfs_daddr_t			rec_daddr)
256 {
257 	if (info->low_daddr != XFS_BUF_DADDR_NULL)
258 		return rec_daddr < info->low_daddr;
259 	if (info->low.rm_blockcount)
260 		return xfs_rmap_compare(rec, &info->low) < 0;
261 	return false;
262 }
263 
264 /*
265  * Format a reverse mapping for getfsmap, having translated rm_startblock
266  * into the appropriate daddr units.  Pass in a nonzero @len_daddr if the
267  * length could be larger than rm_blockcount in struct xfs_rmap_irec.
268  */
269 STATIC int
270 xfs_getfsmap_helper(
271 	struct xfs_trans		*tp,
272 	struct xfs_getfsmap_info	*info,
273 	const struct xfs_rmap_irec	*rec,
274 	xfs_daddr_t			rec_daddr,
275 	xfs_daddr_t			len_daddr)
276 {
277 	struct xfs_fsmap		fmr;
278 	struct xfs_mount		*mp = tp->t_mountp;
279 	bool				shared;
280 	int				error;
281 
282 	if (fatal_signal_pending(current))
283 		return -EINTR;
284 
285 	if (len_daddr == 0)
286 		len_daddr = XFS_FSB_TO_BB(mp, rec->rm_blockcount);
287 
288 	/*
289 	 * Filter out records that start before our startpoint, if the
290 	 * caller requested that.
291 	 */
292 	if (xfs_getfsmap_rec_before_start(info, rec, rec_daddr)) {
293 		rec_daddr += len_daddr;
294 		if (info->next_daddr < rec_daddr)
295 			info->next_daddr = rec_daddr;
296 		return 0;
297 	}
298 
299 	/*
300 	 * For an info->last query, we're looking for a gap between the last
301 	 * mapping emitted and the high key specified by userspace.  If the
302 	 * user's query spans less than 1 fsblock, then info->high and
303 	 * info->low will have the same rm_startblock, which causes rec_daddr
304 	 * and next_daddr to be the same.  Therefore, use the end_daddr that
305 	 * we calculated from userspace's high key to synthesize the record.
306 	 * Note that if the btree query found a mapping, there won't be a gap.
307 	 */
308 	if (info->last && info->end_daddr != XFS_BUF_DADDR_NULL)
309 		rec_daddr = info->end_daddr;
310 
311 	/* Are we just counting mappings? */
312 	if (info->head->fmh_count == 0) {
313 		if (info->head->fmh_entries == UINT_MAX)
314 			return -ECANCELED;
315 
316 		if (rec_daddr > info->next_daddr)
317 			info->head->fmh_entries++;
318 
319 		if (info->last)
320 			return 0;
321 
322 		info->head->fmh_entries++;
323 
324 		rec_daddr += len_daddr;
325 		if (info->next_daddr < rec_daddr)
326 			info->next_daddr = rec_daddr;
327 		return 0;
328 	}
329 
330 	/*
331 	 * If the record starts past the last physical block we saw,
332 	 * then we've found a gap.  Report the gap as being owned by
333 	 * whatever the caller specified is the missing owner.
334 	 */
335 	if (rec_daddr > info->next_daddr) {
336 		if (info->head->fmh_entries >= info->head->fmh_count)
337 			return -ECANCELED;
338 
339 		fmr.fmr_device = info->dev;
340 		fmr.fmr_physical = info->next_daddr;
341 		fmr.fmr_owner = info->missing_owner;
342 		fmr.fmr_offset = 0;
343 		fmr.fmr_length = rec_daddr - info->next_daddr;
344 		fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
345 		xfs_getfsmap_format(mp, &fmr, info);
346 	}
347 
348 	if (info->last)
349 		goto out;
350 
351 	/* Fill out the extent we found */
352 	if (info->head->fmh_entries >= info->head->fmh_count)
353 		return -ECANCELED;
354 
355 	trace_xfs_fsmap_mapping(mp, info->dev,
356 			info->pag ? info->pag->pag_agno : NULLAGNUMBER, rec);
357 
358 	fmr.fmr_device = info->dev;
359 	fmr.fmr_physical = rec_daddr;
360 	error = xfs_fsmap_owner_from_rmap(&fmr, rec);
361 	if (error)
362 		return error;
363 	fmr.fmr_offset = XFS_FSB_TO_BB(mp, rec->rm_offset);
364 	fmr.fmr_length = len_daddr;
365 	if (rec->rm_flags & XFS_RMAP_UNWRITTEN)
366 		fmr.fmr_flags |= FMR_OF_PREALLOC;
367 	if (rec->rm_flags & XFS_RMAP_ATTR_FORK)
368 		fmr.fmr_flags |= FMR_OF_ATTR_FORK;
369 	if (rec->rm_flags & XFS_RMAP_BMBT_BLOCK)
370 		fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
371 	if (fmr.fmr_flags == 0) {
372 		error = xfs_getfsmap_is_shared(tp, info, rec, &shared);
373 		if (error)
374 			return error;
375 		if (shared)
376 			fmr.fmr_flags |= FMR_OF_SHARED;
377 	}
378 
379 	xfs_getfsmap_format(mp, &fmr, info);
380 out:
381 	rec_daddr += len_daddr;
382 	if (info->next_daddr < rec_daddr)
383 		info->next_daddr = rec_daddr;
384 	return 0;
385 }
386 
387 /* Transform a rmapbt irec into a fsmap */
388 STATIC int
389 xfs_getfsmap_datadev_helper(
390 	struct xfs_btree_cur		*cur,
391 	const struct xfs_rmap_irec	*rec,
392 	void				*priv)
393 {
394 	struct xfs_mount		*mp = cur->bc_mp;
395 	struct xfs_getfsmap_info	*info = priv;
396 	xfs_fsblock_t			fsb;
397 	xfs_daddr_t			rec_daddr;
398 
399 	fsb = XFS_AGB_TO_FSB(mp, cur->bc_ag.pag->pag_agno, rec->rm_startblock);
400 	rec_daddr = XFS_FSB_TO_DADDR(mp, fsb);
401 
402 	return xfs_getfsmap_helper(cur->bc_tp, info, rec, rec_daddr, 0);
403 }
404 
405 /* Transform a bnobt irec into a fsmap */
406 STATIC int
407 xfs_getfsmap_datadev_bnobt_helper(
408 	struct xfs_btree_cur		*cur,
409 	const struct xfs_alloc_rec_incore *rec,
410 	void				*priv)
411 {
412 	struct xfs_mount		*mp = cur->bc_mp;
413 	struct xfs_getfsmap_info	*info = priv;
414 	struct xfs_rmap_irec		irec;
415 	xfs_daddr_t			rec_daddr;
416 
417 	rec_daddr = XFS_AGB_TO_DADDR(mp, cur->bc_ag.pag->pag_agno,
418 			rec->ar_startblock);
419 
420 	irec.rm_startblock = rec->ar_startblock;
421 	irec.rm_blockcount = rec->ar_blockcount;
422 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
423 	irec.rm_offset = 0;
424 	irec.rm_flags = 0;
425 
426 	return xfs_getfsmap_helper(cur->bc_tp, info, &irec, rec_daddr, 0);
427 }
428 
429 /* Set rmap flags based on the getfsmap flags */
430 static void
431 xfs_getfsmap_set_irec_flags(
432 	struct xfs_rmap_irec	*irec,
433 	const struct xfs_fsmap	*fmr)
434 {
435 	irec->rm_flags = 0;
436 	if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
437 		irec->rm_flags |= XFS_RMAP_ATTR_FORK;
438 	if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
439 		irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
440 	if (fmr->fmr_flags & FMR_OF_PREALLOC)
441 		irec->rm_flags |= XFS_RMAP_UNWRITTEN;
442 }
443 
444 /* Execute a getfsmap query against the log device. */
445 STATIC int
446 xfs_getfsmap_logdev(
447 	struct xfs_trans		*tp,
448 	const struct xfs_fsmap		*keys,
449 	struct xfs_getfsmap_info	*info)
450 {
451 	struct xfs_mount		*mp = tp->t_mountp;
452 	struct xfs_rmap_irec		rmap;
453 	xfs_daddr_t			rec_daddr, len_daddr;
454 	xfs_fsblock_t			start_fsb, end_fsb;
455 	uint64_t			eofs;
456 
457 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
458 	if (keys[0].fmr_physical >= eofs)
459 		return 0;
460 	start_fsb = XFS_BB_TO_FSBT(mp,
461 				keys[0].fmr_physical + keys[0].fmr_length);
462 	end_fsb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
463 
464 	/* Adjust the low key if we are continuing from where we left off. */
465 	if (keys[0].fmr_length > 0)
466 		info->low_daddr = XFS_FSB_TO_BB(mp, start_fsb);
467 
468 	trace_xfs_fsmap_low_key_linear(mp, info->dev, start_fsb);
469 	trace_xfs_fsmap_high_key_linear(mp, info->dev, end_fsb);
470 
471 	if (start_fsb > 0)
472 		return 0;
473 
474 	/* Fabricate an rmap entry for the external log device. */
475 	rmap.rm_startblock = 0;
476 	rmap.rm_blockcount = mp->m_sb.sb_logblocks;
477 	rmap.rm_owner = XFS_RMAP_OWN_LOG;
478 	rmap.rm_offset = 0;
479 	rmap.rm_flags = 0;
480 
481 	rec_daddr = XFS_FSB_TO_BB(mp, rmap.rm_startblock);
482 	len_daddr = XFS_FSB_TO_BB(mp, rmap.rm_blockcount);
483 	return xfs_getfsmap_helper(tp, info, &rmap, rec_daddr, len_daddr);
484 }
485 
486 #ifdef CONFIG_XFS_RT
487 /* Transform a rtbitmap "record" into a fsmap */
488 STATIC int
489 xfs_getfsmap_rtdev_rtbitmap_helper(
490 	struct xfs_mount		*mp,
491 	struct xfs_trans		*tp,
492 	const struct xfs_rtalloc_rec	*rec,
493 	void				*priv)
494 {
495 	struct xfs_getfsmap_info	*info = priv;
496 	struct xfs_rmap_irec		irec;
497 	xfs_rtblock_t			rtbno;
498 	xfs_daddr_t			rec_daddr, len_daddr;
499 
500 	rtbno = xfs_rtx_to_rtb(mp, rec->ar_startext);
501 	rec_daddr = XFS_FSB_TO_BB(mp, rtbno);
502 	irec.rm_startblock = rtbno;
503 
504 	rtbno = xfs_rtx_to_rtb(mp, rec->ar_extcount);
505 	len_daddr = XFS_FSB_TO_BB(mp, rtbno);
506 	irec.rm_blockcount = rtbno;
507 
508 	irec.rm_owner = XFS_RMAP_OWN_NULL;	/* "free" */
509 	irec.rm_offset = 0;
510 	irec.rm_flags = 0;
511 
512 	return xfs_getfsmap_helper(tp, info, &irec, rec_daddr, len_daddr);
513 }
514 
515 /* Execute a getfsmap query against the realtime device rtbitmap. */
516 STATIC int
517 xfs_getfsmap_rtdev_rtbitmap(
518 	struct xfs_trans		*tp,
519 	const struct xfs_fsmap		*keys,
520 	struct xfs_getfsmap_info	*info)
521 {
522 
523 	struct xfs_rtalloc_rec		alow = { 0 };
524 	struct xfs_rtalloc_rec		ahigh = { 0 };
525 	struct xfs_mount		*mp = tp->t_mountp;
526 	xfs_rtblock_t			start_rtb;
527 	xfs_rtblock_t			end_rtb;
528 	uint64_t			eofs;
529 	int				error;
530 
531 	eofs = XFS_FSB_TO_BB(mp, xfs_rtx_to_rtb(mp, mp->m_sb.sb_rextents));
532 	if (keys[0].fmr_physical >= eofs)
533 		return 0;
534 	start_rtb = XFS_BB_TO_FSBT(mp,
535 				keys[0].fmr_physical + keys[0].fmr_length);
536 	end_rtb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
537 
538 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
539 
540 	/* Adjust the low key if we are continuing from where we left off. */
541 	if (keys[0].fmr_length > 0) {
542 		info->low_daddr = XFS_FSB_TO_BB(mp, start_rtb);
543 		if (info->low_daddr >= eofs)
544 			return 0;
545 	}
546 
547 	trace_xfs_fsmap_low_key_linear(mp, info->dev, start_rtb);
548 	trace_xfs_fsmap_high_key_linear(mp, info->dev, end_rtb);
549 
550 	xfs_rtbitmap_lock_shared(mp, XFS_RBMLOCK_BITMAP);
551 
552 	/*
553 	 * Set up query parameters to return free rtextents covering the range
554 	 * we want.
555 	 */
556 	alow.ar_startext = xfs_rtb_to_rtx(mp, start_rtb);
557 	ahigh.ar_startext = xfs_rtb_to_rtxup(mp, end_rtb);
558 	error = xfs_rtalloc_query_range(mp, tp, &alow, &ahigh,
559 			xfs_getfsmap_rtdev_rtbitmap_helper, info);
560 	if (error)
561 		goto err;
562 
563 	/*
564 	 * Report any gaps at the end of the rtbitmap by simulating a null
565 	 * rmap starting at the block after the end of the query range.
566 	 */
567 	info->last = true;
568 	ahigh.ar_startext = min(mp->m_sb.sb_rextents, ahigh.ar_startext);
569 
570 	error = xfs_getfsmap_rtdev_rtbitmap_helper(mp, tp, &ahigh, info);
571 	if (error)
572 		goto err;
573 err:
574 	xfs_rtbitmap_unlock_shared(mp, XFS_RBMLOCK_BITMAP);
575 	return error;
576 }
577 #endif /* CONFIG_XFS_RT */
578 
579 static inline bool
580 rmap_not_shareable(struct xfs_mount *mp, const struct xfs_rmap_irec *r)
581 {
582 	if (!xfs_has_reflink(mp))
583 		return true;
584 	if (XFS_RMAP_NON_INODE_OWNER(r->rm_owner))
585 		return true;
586 	if (r->rm_flags & (XFS_RMAP_ATTR_FORK | XFS_RMAP_BMBT_BLOCK |
587 			   XFS_RMAP_UNWRITTEN))
588 		return true;
589 	return false;
590 }
591 
592 /* Execute a getfsmap query against the regular data device. */
593 STATIC int
594 __xfs_getfsmap_datadev(
595 	struct xfs_trans		*tp,
596 	const struct xfs_fsmap		*keys,
597 	struct xfs_getfsmap_info	*info,
598 	int				(*query_fn)(struct xfs_trans *,
599 						    struct xfs_getfsmap_info *,
600 						    struct xfs_btree_cur **,
601 						    void *),
602 	void				*priv)
603 {
604 	struct xfs_mount		*mp = tp->t_mountp;
605 	struct xfs_perag		*pag;
606 	struct xfs_btree_cur		*bt_cur = NULL;
607 	xfs_fsblock_t			start_fsb;
608 	xfs_fsblock_t			end_fsb;
609 	xfs_agnumber_t			start_ag;
610 	xfs_agnumber_t			end_ag;
611 	uint64_t			eofs;
612 	int				error = 0;
613 
614 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
615 	if (keys[0].fmr_physical >= eofs)
616 		return 0;
617 	start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
618 	end_fsb = XFS_DADDR_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
619 
620 	/*
621 	 * Convert the fsmap low/high keys to AG based keys.  Initialize
622 	 * low to the fsmap low key and max out the high key to the end
623 	 * of the AG.
624 	 */
625 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
626 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
627 	if (error)
628 		return error;
629 	info->low.rm_blockcount = XFS_BB_TO_FSBT(mp, keys[0].fmr_length);
630 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
631 
632 	/* Adjust the low key if we are continuing from where we left off. */
633 	if (info->low.rm_blockcount == 0) {
634 		/* No previous record from which to continue */
635 	} else if (rmap_not_shareable(mp, &info->low)) {
636 		/* Last record seen was an unshareable extent */
637 		info->low.rm_owner = 0;
638 		info->low.rm_offset = 0;
639 
640 		start_fsb += info->low.rm_blockcount;
641 		if (XFS_FSB_TO_DADDR(mp, start_fsb) >= eofs)
642 			return 0;
643 	} else {
644 		/* Last record seen was a shareable file data extent */
645 		info->low.rm_offset += info->low.rm_blockcount;
646 	}
647 	info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
648 
649 	info->high.rm_startblock = -1U;
650 	info->high.rm_owner = ULLONG_MAX;
651 	info->high.rm_offset = ULLONG_MAX;
652 	info->high.rm_blockcount = 0;
653 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
654 
655 	start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
656 	end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
657 
658 	for_each_perag_range(mp, start_ag, end_ag, pag) {
659 		/*
660 		 * Set the AG high key from the fsmap high key if this
661 		 * is the last AG that we're querying.
662 		 */
663 		info->pag = pag;
664 		if (pag->pag_agno == end_ag) {
665 			info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
666 					end_fsb);
667 			info->high.rm_offset = XFS_BB_TO_FSBT(mp,
668 					keys[1].fmr_offset);
669 			error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
670 			if (error)
671 				break;
672 			xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
673 		}
674 
675 		if (bt_cur) {
676 			xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
677 			bt_cur = NULL;
678 			xfs_trans_brelse(tp, info->agf_bp);
679 			info->agf_bp = NULL;
680 		}
681 
682 		error = xfs_alloc_read_agf(pag, tp, 0, &info->agf_bp);
683 		if (error)
684 			break;
685 
686 		trace_xfs_fsmap_low_key(mp, info->dev, pag->pag_agno,
687 				&info->low);
688 		trace_xfs_fsmap_high_key(mp, info->dev, pag->pag_agno,
689 				&info->high);
690 
691 		error = query_fn(tp, info, &bt_cur, priv);
692 		if (error)
693 			break;
694 
695 		/*
696 		 * Set the AG low key to the start of the AG prior to
697 		 * moving on to the next AG.
698 		 */
699 		if (pag->pag_agno == start_ag)
700 			memset(&info->low, 0, sizeof(info->low));
701 
702 		/*
703 		 * If this is the last AG, report any gap at the end of it
704 		 * before we drop the reference to the perag when the loop
705 		 * terminates.
706 		 */
707 		if (pag->pag_agno == end_ag) {
708 			info->last = true;
709 			error = query_fn(tp, info, &bt_cur, priv);
710 			if (error)
711 				break;
712 		}
713 		info->pag = NULL;
714 	}
715 
716 	if (bt_cur)
717 		xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
718 							 XFS_BTREE_NOERROR);
719 	if (info->agf_bp) {
720 		xfs_trans_brelse(tp, info->agf_bp);
721 		info->agf_bp = NULL;
722 	}
723 	if (info->pag) {
724 		xfs_perag_rele(info->pag);
725 		info->pag = NULL;
726 	} else if (pag) {
727 		/* loop termination case */
728 		xfs_perag_rele(pag);
729 	}
730 
731 	return error;
732 }
733 
734 /* Actually query the rmap btree. */
735 STATIC int
736 xfs_getfsmap_datadev_rmapbt_query(
737 	struct xfs_trans		*tp,
738 	struct xfs_getfsmap_info	*info,
739 	struct xfs_btree_cur		**curpp,
740 	void				*priv)
741 {
742 	/* Report any gap at the end of the last AG. */
743 	if (info->last)
744 		return xfs_getfsmap_datadev_helper(*curpp, &info->high, info);
745 
746 	/* Allocate cursor for this AG and query_range it. */
747 	*curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
748 			info->pag);
749 	return xfs_rmap_query_range(*curpp, &info->low, &info->high,
750 			xfs_getfsmap_datadev_helper, info);
751 }
752 
753 /* Execute a getfsmap query against the regular data device rmapbt. */
754 STATIC int
755 xfs_getfsmap_datadev_rmapbt(
756 	struct xfs_trans		*tp,
757 	const struct xfs_fsmap		*keys,
758 	struct xfs_getfsmap_info	*info)
759 {
760 	info->missing_owner = XFS_FMR_OWN_FREE;
761 	return __xfs_getfsmap_datadev(tp, keys, info,
762 			xfs_getfsmap_datadev_rmapbt_query, NULL);
763 }
764 
765 /* Actually query the bno btree. */
766 STATIC int
767 xfs_getfsmap_datadev_bnobt_query(
768 	struct xfs_trans		*tp,
769 	struct xfs_getfsmap_info	*info,
770 	struct xfs_btree_cur		**curpp,
771 	void				*priv)
772 {
773 	struct xfs_alloc_rec_incore	*key = priv;
774 
775 	/* Report any gap at the end of the last AG. */
776 	if (info->last)
777 		return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
778 
779 	/* Allocate cursor for this AG and query_range it. */
780 	*curpp = xfs_bnobt_init_cursor(tp->t_mountp, tp, info->agf_bp,
781 			info->pag);
782 	key->ar_startblock = info->low.rm_startblock;
783 	key[1].ar_startblock = info->high.rm_startblock;
784 	return xfs_alloc_query_range(*curpp, key, &key[1],
785 			xfs_getfsmap_datadev_bnobt_helper, info);
786 }
787 
788 /* Execute a getfsmap query against the regular data device's bnobt. */
789 STATIC int
790 xfs_getfsmap_datadev_bnobt(
791 	struct xfs_trans		*tp,
792 	const struct xfs_fsmap		*keys,
793 	struct xfs_getfsmap_info	*info)
794 {
795 	struct xfs_alloc_rec_incore	akeys[2];
796 
797 	memset(akeys, 0, sizeof(akeys));
798 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
799 	return __xfs_getfsmap_datadev(tp, keys, info,
800 			xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
801 }
802 
803 /* Do we recognize the device? */
804 STATIC bool
805 xfs_getfsmap_is_valid_device(
806 	struct xfs_mount	*mp,
807 	struct xfs_fsmap	*fm)
808 {
809 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
810 	    fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
811 		return true;
812 	if (mp->m_logdev_targp &&
813 	    fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
814 		return true;
815 	if (mp->m_rtdev_targp &&
816 	    fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
817 		return true;
818 	return false;
819 }
820 
821 /* Ensure that the low key is less than the high key. */
822 STATIC bool
823 xfs_getfsmap_check_keys(
824 	struct xfs_fsmap		*low_key,
825 	struct xfs_fsmap		*high_key)
826 {
827 	if (low_key->fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
828 		if (low_key->fmr_offset)
829 			return false;
830 	}
831 	if (high_key->fmr_flags != -1U &&
832 	    (high_key->fmr_flags & (FMR_OF_SPECIAL_OWNER |
833 				    FMR_OF_EXTENT_MAP))) {
834 		if (high_key->fmr_offset && high_key->fmr_offset != -1ULL)
835 			return false;
836 	}
837 	if (high_key->fmr_length && high_key->fmr_length != -1ULL)
838 		return false;
839 
840 	if (low_key->fmr_device > high_key->fmr_device)
841 		return false;
842 	if (low_key->fmr_device < high_key->fmr_device)
843 		return true;
844 
845 	if (low_key->fmr_physical > high_key->fmr_physical)
846 		return false;
847 	if (low_key->fmr_physical < high_key->fmr_physical)
848 		return true;
849 
850 	if (low_key->fmr_owner > high_key->fmr_owner)
851 		return false;
852 	if (low_key->fmr_owner < high_key->fmr_owner)
853 		return true;
854 
855 	if (low_key->fmr_offset > high_key->fmr_offset)
856 		return false;
857 	if (low_key->fmr_offset < high_key->fmr_offset)
858 		return true;
859 
860 	return false;
861 }
862 
863 /*
864  * There are only two devices if we didn't configure RT devices at build time.
865  */
866 #ifdef CONFIG_XFS_RT
867 #define XFS_GETFSMAP_DEVS	3
868 #else
869 #define XFS_GETFSMAP_DEVS	2
870 #endif /* CONFIG_XFS_RT */
871 
872 /*
873  * Get filesystem's extents as described in head, and format for output. Fills
874  * in the supplied records array until there are no more reverse mappings to
875  * return or head.fmh_entries == head.fmh_count.  In the second case, this
876  * function returns -ECANCELED to indicate that more records would have been
877  * returned.
878  *
879  * Key to Confusion
880  * ----------------
881  * There are multiple levels of keys and counters at work here:
882  * xfs_fsmap_head.fmh_keys	-- low and high fsmap keys passed in;
883  *				   these reflect fs-wide sector addrs.
884  * dkeys			-- fmh_keys used to query each device;
885  *				   these are fmh_keys but w/ the low key
886  *				   bumped up by fmr_length.
887  * xfs_getfsmap_info.next_daddr	-- next disk addr we expect to see; this
888  *				   is how we detect gaps in the fsmap
889 				   records and report them.
890  * xfs_getfsmap_info.low/high	-- per-AG low/high keys computed from
891  *				   dkeys; used to query the metadata.
892  */
893 int
894 xfs_getfsmap(
895 	struct xfs_mount		*mp,
896 	struct xfs_fsmap_head		*head,
897 	struct fsmap			*fsmap_recs)
898 {
899 	struct xfs_trans		*tp = NULL;
900 	struct xfs_fsmap		dkeys[2];	/* per-dev keys */
901 	struct xfs_getfsmap_dev		handlers[XFS_GETFSMAP_DEVS];
902 	struct xfs_getfsmap_info	info = { NULL };
903 	bool				use_rmap;
904 	int				i;
905 	int				error = 0;
906 
907 	if (head->fmh_iflags & ~FMH_IF_VALID)
908 		return -EINVAL;
909 	if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
910 	    !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
911 		return -EINVAL;
912 	if (!xfs_getfsmap_check_keys(&head->fmh_keys[0], &head->fmh_keys[1]))
913 		return -EINVAL;
914 
915 	use_rmap = xfs_has_rmapbt(mp) &&
916 		   has_capability_noaudit(current, CAP_SYS_ADMIN);
917 	head->fmh_entries = 0;
918 
919 	/* Set up our device handlers. */
920 	memset(handlers, 0, sizeof(handlers));
921 	handlers[0].nr_sectors = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
922 	handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
923 	if (use_rmap)
924 		handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
925 	else
926 		handlers[0].fn = xfs_getfsmap_datadev_bnobt;
927 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
928 		handlers[1].nr_sectors = XFS_FSB_TO_BB(mp,
929 						       mp->m_sb.sb_logblocks);
930 		handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
931 		handlers[1].fn = xfs_getfsmap_logdev;
932 	}
933 #ifdef CONFIG_XFS_RT
934 	if (mp->m_rtdev_targp) {
935 		handlers[2].nr_sectors = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
936 		handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
937 		handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
938 	}
939 #endif /* CONFIG_XFS_RT */
940 
941 	xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
942 			xfs_getfsmap_dev_compare);
943 
944 	/*
945 	 * To continue where we left off, we allow userspace to use the
946 	 * last mapping from a previous call as the low key of the next.
947 	 * This is identified by a non-zero length in the low key. We
948 	 * have to increment the low key in this scenario to ensure we
949 	 * don't return the same mapping again, and instead return the
950 	 * very next mapping.
951 	 *
952 	 * If the low key mapping refers to file data, the same physical
953 	 * blocks could be mapped to several other files/offsets.
954 	 * According to rmapbt record ordering, the minimal next
955 	 * possible record for the block range is the next starting
956 	 * offset in the same inode. Therefore, each fsmap backend bumps
957 	 * the file offset to continue the search appropriately.  For
958 	 * all other low key mapping types (attr blocks, metadata), each
959 	 * fsmap backend bumps the physical offset as there can be no
960 	 * other mapping for the same physical block range.
961 	 */
962 	dkeys[0] = head->fmh_keys[0];
963 	memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
964 
965 	info.next_daddr = head->fmh_keys[0].fmr_physical +
966 			  head->fmh_keys[0].fmr_length;
967 	info.end_daddr = XFS_BUF_DADDR_NULL;
968 	info.fsmap_recs = fsmap_recs;
969 	info.head = head;
970 
971 	/* For each device we support... */
972 	for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
973 		/* Is this device within the range the user asked for? */
974 		if (!handlers[i].fn)
975 			continue;
976 		if (head->fmh_keys[0].fmr_device > handlers[i].dev)
977 			continue;
978 		if (head->fmh_keys[1].fmr_device < handlers[i].dev)
979 			break;
980 
981 		/*
982 		 * If this device number matches the high key, we have
983 		 * to pass the high key to the handler to limit the
984 		 * query results.  If the device number exceeds the
985 		 * low key, zero out the low key so that we get
986 		 * everything from the beginning.
987 		 */
988 		if (handlers[i].dev == head->fmh_keys[1].fmr_device) {
989 			dkeys[1] = head->fmh_keys[1];
990 			info.end_daddr = min(handlers[i].nr_sectors - 1,
991 					     dkeys[1].fmr_physical);
992 		}
993 		if (handlers[i].dev > head->fmh_keys[0].fmr_device)
994 			memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
995 
996 		/*
997 		 * Grab an empty transaction so that we can use its recursive
998 		 * buffer locking abilities to detect cycles in the rmapbt
999 		 * without deadlocking.
1000 		 */
1001 		error = xfs_trans_alloc_empty(mp, &tp);
1002 		if (error)
1003 			break;
1004 
1005 		info.dev = handlers[i].dev;
1006 		info.last = false;
1007 		info.pag = NULL;
1008 		info.low_daddr = XFS_BUF_DADDR_NULL;
1009 		info.low.rm_blockcount = 0;
1010 		error = handlers[i].fn(tp, dkeys, &info);
1011 		if (error)
1012 			break;
1013 		xfs_trans_cancel(tp);
1014 		tp = NULL;
1015 		info.next_daddr = 0;
1016 	}
1017 
1018 	if (tp)
1019 		xfs_trans_cancel(tp);
1020 	head->fmh_oflags = FMH_OF_DEV_T;
1021 	return error;
1022 }
1023