xref: /linux/fs/xfs/xfs_fsmap.c (revision 8cbd01ba9c38eb16f3a572300da486ac544519b7)
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 #include "xfs_rtgroup.h"
29 
30 /* Convert an xfs_fsmap to an fsmap. */
31 static void
xfs_fsmap_from_internal(struct fsmap * dest,struct xfs_fsmap * src)32 xfs_fsmap_from_internal(
33 	struct fsmap		*dest,
34 	struct xfs_fsmap	*src)
35 {
36 	dest->fmr_device = src->fmr_device;
37 	dest->fmr_flags = src->fmr_flags;
38 	dest->fmr_physical = BBTOB(src->fmr_physical);
39 	dest->fmr_owner = src->fmr_owner;
40 	dest->fmr_offset = BBTOB(src->fmr_offset);
41 	dest->fmr_length = BBTOB(src->fmr_length);
42 	dest->fmr_reserved[0] = 0;
43 	dest->fmr_reserved[1] = 0;
44 	dest->fmr_reserved[2] = 0;
45 }
46 
47 /* Convert an fsmap to an xfs_fsmap. */
48 static void
xfs_fsmap_to_internal(struct xfs_fsmap * dest,struct fsmap * src)49 xfs_fsmap_to_internal(
50 	struct xfs_fsmap	*dest,
51 	struct fsmap		*src)
52 {
53 	dest->fmr_device = src->fmr_device;
54 	dest->fmr_flags = src->fmr_flags;
55 	dest->fmr_physical = BTOBBT(src->fmr_physical);
56 	dest->fmr_owner = src->fmr_owner;
57 	dest->fmr_offset = BTOBBT(src->fmr_offset);
58 	dest->fmr_length = BTOBBT(src->fmr_length);
59 }
60 
61 /* Convert an fsmap owner into an rmapbt owner. */
62 static int
xfs_fsmap_owner_to_rmap(struct xfs_rmap_irec * dest,const struct xfs_fsmap * src)63 xfs_fsmap_owner_to_rmap(
64 	struct xfs_rmap_irec	*dest,
65 	const struct xfs_fsmap	*src)
66 {
67 	if (!(src->fmr_flags & FMR_OF_SPECIAL_OWNER)) {
68 		dest->rm_owner = src->fmr_owner;
69 		return 0;
70 	}
71 
72 	switch (src->fmr_owner) {
73 	case 0:			/* "lowest owner id possible" */
74 	case -1ULL:		/* "highest owner id possible" */
75 		dest->rm_owner = src->fmr_owner;
76 		break;
77 	case XFS_FMR_OWN_FREE:
78 		dest->rm_owner = XFS_RMAP_OWN_NULL;
79 		break;
80 	case XFS_FMR_OWN_UNKNOWN:
81 		dest->rm_owner = XFS_RMAP_OWN_UNKNOWN;
82 		break;
83 	case XFS_FMR_OWN_FS:
84 		dest->rm_owner = XFS_RMAP_OWN_FS;
85 		break;
86 	case XFS_FMR_OWN_LOG:
87 		dest->rm_owner = XFS_RMAP_OWN_LOG;
88 		break;
89 	case XFS_FMR_OWN_AG:
90 		dest->rm_owner = XFS_RMAP_OWN_AG;
91 		break;
92 	case XFS_FMR_OWN_INOBT:
93 		dest->rm_owner = XFS_RMAP_OWN_INOBT;
94 		break;
95 	case XFS_FMR_OWN_INODES:
96 		dest->rm_owner = XFS_RMAP_OWN_INODES;
97 		break;
98 	case XFS_FMR_OWN_REFC:
99 		dest->rm_owner = XFS_RMAP_OWN_REFC;
100 		break;
101 	case XFS_FMR_OWN_COW:
102 		dest->rm_owner = XFS_RMAP_OWN_COW;
103 		break;
104 	case XFS_FMR_OWN_DEFECTIVE:	/* not implemented */
105 		/* fall through */
106 	default:
107 		return -EINVAL;
108 	}
109 	return 0;
110 }
111 
112 /* Convert an rmapbt owner into an fsmap owner. */
113 static int
xfs_fsmap_owner_from_frec(struct xfs_fsmap * dest,const struct xfs_fsmap_irec * frec)114 xfs_fsmap_owner_from_frec(
115 	struct xfs_fsmap		*dest,
116 	const struct xfs_fsmap_irec	*frec)
117 {
118 	dest->fmr_flags = 0;
119 	if (!XFS_RMAP_NON_INODE_OWNER(frec->owner)) {
120 		dest->fmr_owner = frec->owner;
121 		return 0;
122 	}
123 	dest->fmr_flags |= FMR_OF_SPECIAL_OWNER;
124 
125 	switch (frec->owner) {
126 	case XFS_RMAP_OWN_FS:
127 		dest->fmr_owner = XFS_FMR_OWN_FS;
128 		break;
129 	case XFS_RMAP_OWN_LOG:
130 		dest->fmr_owner = XFS_FMR_OWN_LOG;
131 		break;
132 	case XFS_RMAP_OWN_AG:
133 		dest->fmr_owner = XFS_FMR_OWN_AG;
134 		break;
135 	case XFS_RMAP_OWN_INOBT:
136 		dest->fmr_owner = XFS_FMR_OWN_INOBT;
137 		break;
138 	case XFS_RMAP_OWN_INODES:
139 		dest->fmr_owner = XFS_FMR_OWN_INODES;
140 		break;
141 	case XFS_RMAP_OWN_REFC:
142 		dest->fmr_owner = XFS_FMR_OWN_REFC;
143 		break;
144 	case XFS_RMAP_OWN_COW:
145 		dest->fmr_owner = XFS_FMR_OWN_COW;
146 		break;
147 	case XFS_RMAP_OWN_NULL:	/* "free" */
148 		dest->fmr_owner = XFS_FMR_OWN_FREE;
149 		break;
150 	default:
151 		ASSERT(0);
152 		return -EFSCORRUPTED;
153 	}
154 	return 0;
155 }
156 
157 /* getfsmap query state */
158 struct xfs_getfsmap_info {
159 	struct xfs_fsmap_head	*head;
160 	struct fsmap		*fsmap_recs;	/* mapping records */
161 	struct xfs_buf		*agf_bp;	/* AGF, for refcount queries */
162 	struct xfs_group	*group;		/* group info, if applicable */
163 	xfs_daddr_t		next_daddr;	/* next daddr we expect */
164 	/* daddr of low fsmap key when we're using the rtbitmap */
165 	xfs_daddr_t		low_daddr;
166 	/* daddr of high fsmap key, or the last daddr on the device */
167 	xfs_daddr_t		end_daddr;
168 	u64			missing_owner;	/* owner of holes */
169 	u32			dev;		/* device id */
170 	/*
171 	 * Low rmap key for the query.  If low.rm_blockcount is nonzero, this
172 	 * is the second (or later) call to retrieve the recordset in pieces.
173 	 * xfs_getfsmap_rec_before_start will compare all records retrieved
174 	 * by the rmapbt query to filter out any records that start before
175 	 * the last record.
176 	 */
177 	struct xfs_rmap_irec	low;
178 	struct xfs_rmap_irec	high;		/* high rmap key */
179 	bool			last;		/* last extent? */
180 };
181 
182 /* Associate a device with a getfsmap handler. */
183 struct xfs_getfsmap_dev {
184 	u32			dev;
185 	int			(*fn)(struct xfs_trans *tp,
186 				      const struct xfs_fsmap *keys,
187 				      struct xfs_getfsmap_info *info);
188 	sector_t		nr_sectors;
189 };
190 
191 /* Compare two getfsmap device handlers. */
192 static int
xfs_getfsmap_dev_compare(const void * p1,const void * p2)193 xfs_getfsmap_dev_compare(
194 	const void			*p1,
195 	const void			*p2)
196 {
197 	const struct xfs_getfsmap_dev	*d1 = p1;
198 	const struct xfs_getfsmap_dev	*d2 = p2;
199 
200 	return d1->dev - d2->dev;
201 }
202 
203 /* Decide if this mapping is shared. */
204 STATIC int
xfs_getfsmap_is_shared(struct xfs_trans * tp,struct xfs_getfsmap_info * info,const struct xfs_fsmap_irec * frec,bool * stat)205 xfs_getfsmap_is_shared(
206 	struct xfs_trans		*tp,
207 	struct xfs_getfsmap_info	*info,
208 	const struct xfs_fsmap_irec	*frec,
209 	bool				*stat)
210 {
211 	struct xfs_mount		*mp = tp->t_mountp;
212 	struct xfs_btree_cur		*cur;
213 	xfs_agblock_t			fbno;
214 	xfs_extlen_t			flen;
215 	int				error;
216 
217 	*stat = false;
218 	if (!xfs_has_reflink(mp))
219 		return 0;
220 	/* rt files will have no perag structure */
221 	if (!info->group)
222 		return 0;
223 
224 	/* Are there any shared blocks here? */
225 	flen = 0;
226 	cur = xfs_refcountbt_init_cursor(mp, tp, info->agf_bp,
227 			to_perag(info->group));
228 
229 	error = xfs_refcount_find_shared(cur, frec->rec_key,
230 			XFS_BB_TO_FSBT(mp, frec->len_daddr), &fbno, &flen,
231 			false);
232 
233 	xfs_btree_del_cursor(cur, error);
234 	if (error)
235 		return error;
236 
237 	*stat = flen > 0;
238 	return 0;
239 }
240 
241 static inline void
xfs_getfsmap_format(struct xfs_mount * mp,struct xfs_fsmap * xfm,struct xfs_getfsmap_info * info)242 xfs_getfsmap_format(
243 	struct xfs_mount		*mp,
244 	struct xfs_fsmap		*xfm,
245 	struct xfs_getfsmap_info	*info)
246 {
247 	struct fsmap			*rec;
248 
249 	trace_xfs_getfsmap_mapping(mp, xfm);
250 
251 	rec = &info->fsmap_recs[info->head->fmh_entries++];
252 	xfs_fsmap_from_internal(rec, xfm);
253 }
254 
255 static inline bool
xfs_getfsmap_frec_before_start(struct xfs_getfsmap_info * info,const struct xfs_fsmap_irec * frec)256 xfs_getfsmap_frec_before_start(
257 	struct xfs_getfsmap_info	*info,
258 	const struct xfs_fsmap_irec	*frec)
259 {
260 	if (info->low_daddr != XFS_BUF_DADDR_NULL)
261 		return frec->start_daddr < info->low_daddr;
262 	if (info->low.rm_blockcount) {
263 		struct xfs_rmap_irec	rec = {
264 			.rm_startblock	= frec->rec_key,
265 			.rm_owner	= frec->owner,
266 			.rm_flags	= frec->rm_flags,
267 		};
268 
269 		return xfs_rmap_compare(&rec, &info->low) < 0;
270 	}
271 
272 	return false;
273 }
274 
275 /*
276  * Format a reverse mapping for getfsmap, having translated rm_startblock
277  * into the appropriate daddr units.  Pass in a nonzero @len_daddr if the
278  * length could be larger than rm_blockcount in struct xfs_rmap_irec.
279  */
280 STATIC int
xfs_getfsmap_helper(struct xfs_trans * tp,struct xfs_getfsmap_info * info,const struct xfs_fsmap_irec * frec)281 xfs_getfsmap_helper(
282 	struct xfs_trans		*tp,
283 	struct xfs_getfsmap_info	*info,
284 	const struct xfs_fsmap_irec	*frec)
285 {
286 	struct xfs_fsmap		fmr;
287 	struct xfs_mount		*mp = tp->t_mountp;
288 	bool				shared;
289 	int				error = 0;
290 
291 	if (fatal_signal_pending(current))
292 		return -EINTR;
293 
294 	/*
295 	 * Filter out records that start before our startpoint, if the
296 	 * caller requested that.
297 	 */
298 	if (xfs_getfsmap_frec_before_start(info, frec))
299 		goto out;
300 
301 	/* Are we just counting mappings? */
302 	if (info->head->fmh_count == 0) {
303 		if (info->head->fmh_entries == UINT_MAX)
304 			return -ECANCELED;
305 
306 		if (frec->start_daddr > info->next_daddr)
307 			info->head->fmh_entries++;
308 
309 		if (info->last)
310 			return 0;
311 
312 		info->head->fmh_entries++;
313 		goto out;
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 (frec->start_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 = frec->start_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->group ? info->group->xg_gno : NULLAGNUMBER,
343 			frec);
344 
345 	fmr.fmr_device = info->dev;
346 	fmr.fmr_physical = frec->start_daddr;
347 	error = xfs_fsmap_owner_from_frec(&fmr, frec);
348 	if (error)
349 		return error;
350 	fmr.fmr_offset = XFS_FSB_TO_BB(mp, frec->offset);
351 	fmr.fmr_length = frec->len_daddr;
352 	if (frec->rm_flags & XFS_RMAP_UNWRITTEN)
353 		fmr.fmr_flags |= FMR_OF_PREALLOC;
354 	if (frec->rm_flags & XFS_RMAP_ATTR_FORK)
355 		fmr.fmr_flags |= FMR_OF_ATTR_FORK;
356 	if (frec->rm_flags & XFS_RMAP_BMBT_BLOCK)
357 		fmr.fmr_flags |= FMR_OF_EXTENT_MAP;
358 	if (fmr.fmr_flags == 0) {
359 		error = xfs_getfsmap_is_shared(tp, info, frec, &shared);
360 		if (error)
361 			return error;
362 		if (shared)
363 			fmr.fmr_flags |= FMR_OF_SHARED;
364 	}
365 
366 	xfs_getfsmap_format(mp, &fmr, info);
367 out:
368 	info->next_daddr = max(info->next_daddr,
369 			       frec->start_daddr + frec->len_daddr);
370 	return 0;
371 }
372 
373 static inline int
xfs_getfsmap_group_helper(struct xfs_getfsmap_info * info,struct xfs_trans * tp,struct xfs_group * xg,xfs_agblock_t startblock,xfs_extlen_t blockcount,struct xfs_fsmap_irec * frec)374 xfs_getfsmap_group_helper(
375 	struct xfs_getfsmap_info	*info,
376 	struct xfs_trans		*tp,
377 	struct xfs_group		*xg,
378 	xfs_agblock_t			startblock,
379 	xfs_extlen_t			blockcount,
380 	struct xfs_fsmap_irec		*frec)
381 {
382 	/*
383 	 * For an info->last query, we're looking for a gap between the last
384 	 * mapping emitted and the high key specified by userspace.  If the
385 	 * user's query spans less than 1 fsblock, then info->high and
386 	 * info->low will have the same rm_startblock, which causes rec_daddr
387 	 * and next_daddr to be the same.  Therefore, use the end_daddr that
388 	 * we calculated from userspace's high key to synthesize the record.
389 	 * Note that if the btree query found a mapping, there won't be a gap.
390 	 */
391 	if (info->last)
392 		frec->start_daddr = info->end_daddr + 1;
393 	else
394 		frec->start_daddr = xfs_gbno_to_daddr(xg, startblock);
395 
396 	frec->len_daddr = XFS_FSB_TO_BB(xg->xg_mount, blockcount);
397 	return xfs_getfsmap_helper(tp, info, frec);
398 }
399 
400 /* Transform a rmapbt irec into a fsmap */
401 STATIC int
xfs_getfsmap_rmapbt_helper(struct xfs_btree_cur * cur,const struct xfs_rmap_irec * rec,void * priv)402 xfs_getfsmap_rmapbt_helper(
403 	struct xfs_btree_cur		*cur,
404 	const struct xfs_rmap_irec	*rec,
405 	void				*priv)
406 {
407 	struct xfs_fsmap_irec		frec = {
408 		.owner			= rec->rm_owner,
409 		.offset			= rec->rm_offset,
410 		.rm_flags		= rec->rm_flags,
411 		.rec_key		= rec->rm_startblock,
412 	};
413 	struct xfs_getfsmap_info	*info = priv;
414 
415 	return xfs_getfsmap_group_helper(info, cur->bc_tp, cur->bc_group,
416 			rec->rm_startblock, rec->rm_blockcount, &frec);
417 }
418 
419 /* Transform a bnobt irec into a fsmap */
420 STATIC int
xfs_getfsmap_datadev_bnobt_helper(struct xfs_btree_cur * cur,const struct xfs_alloc_rec_incore * rec,void * priv)421 xfs_getfsmap_datadev_bnobt_helper(
422 	struct xfs_btree_cur		*cur,
423 	const struct xfs_alloc_rec_incore *rec,
424 	void				*priv)
425 {
426 	struct xfs_fsmap_irec		frec = {
427 		.owner			= XFS_RMAP_OWN_NULL, /* "free" */
428 		.rec_key		= rec->ar_startblock,
429 	};
430 	struct xfs_getfsmap_info	*info = priv;
431 
432 	return xfs_getfsmap_group_helper(info, cur->bc_tp, cur->bc_group,
433 			rec->ar_startblock, rec->ar_blockcount, &frec);
434 }
435 
436 /* Set rmap flags based on the getfsmap flags */
437 static void
xfs_getfsmap_set_irec_flags(struct xfs_rmap_irec * irec,const struct xfs_fsmap * fmr)438 xfs_getfsmap_set_irec_flags(
439 	struct xfs_rmap_irec	*irec,
440 	const struct xfs_fsmap	*fmr)
441 {
442 	irec->rm_flags = 0;
443 	if (fmr->fmr_flags & FMR_OF_ATTR_FORK)
444 		irec->rm_flags |= XFS_RMAP_ATTR_FORK;
445 	if (fmr->fmr_flags & FMR_OF_EXTENT_MAP)
446 		irec->rm_flags |= XFS_RMAP_BMBT_BLOCK;
447 	if (fmr->fmr_flags & FMR_OF_PREALLOC)
448 		irec->rm_flags |= XFS_RMAP_UNWRITTEN;
449 }
450 
451 static inline bool
rmap_not_shareable(struct xfs_mount * mp,const struct xfs_rmap_irec * r)452 rmap_not_shareable(struct xfs_mount *mp, const struct xfs_rmap_irec *r)
453 {
454 	if (!xfs_has_reflink(mp))
455 		return true;
456 	if (XFS_RMAP_NON_INODE_OWNER(r->rm_owner))
457 		return true;
458 	if (r->rm_flags & (XFS_RMAP_ATTR_FORK | XFS_RMAP_BMBT_BLOCK |
459 			   XFS_RMAP_UNWRITTEN))
460 		return true;
461 	return false;
462 }
463 
464 /* Execute a getfsmap query against the regular data device. */
465 STATIC int
__xfs_getfsmap_datadev(struct xfs_trans * tp,const struct xfs_fsmap * keys,struct xfs_getfsmap_info * info,int (* query_fn)(struct xfs_trans *,struct xfs_getfsmap_info *,struct xfs_btree_cur **,void *),void * priv)466 __xfs_getfsmap_datadev(
467 	struct xfs_trans		*tp,
468 	const struct xfs_fsmap		*keys,
469 	struct xfs_getfsmap_info	*info,
470 	int				(*query_fn)(struct xfs_trans *,
471 						    struct xfs_getfsmap_info *,
472 						    struct xfs_btree_cur **,
473 						    void *),
474 	void				*priv)
475 {
476 	struct xfs_mount		*mp = tp->t_mountp;
477 	struct xfs_perag		*pag = NULL;
478 	struct xfs_btree_cur		*bt_cur = NULL;
479 	xfs_fsblock_t			start_fsb;
480 	xfs_fsblock_t			end_fsb;
481 	xfs_agnumber_t			start_ag, end_ag;
482 	uint64_t			eofs;
483 	int				error = 0;
484 
485 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
486 	if (keys[0].fmr_physical >= eofs)
487 		return 0;
488 	start_fsb = XFS_DADDR_TO_FSB(mp, keys[0].fmr_physical);
489 	end_fsb = XFS_DADDR_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
490 
491 	/*
492 	 * Convert the fsmap low/high keys to AG based keys.  Initialize
493 	 * low to the fsmap low key and max out the high key to the end
494 	 * of the AG.
495 	 */
496 	info->low.rm_offset = XFS_BB_TO_FSBT(mp, keys[0].fmr_offset);
497 	error = xfs_fsmap_owner_to_rmap(&info->low, &keys[0]);
498 	if (error)
499 		return error;
500 	info->low.rm_blockcount = XFS_BB_TO_FSBT(mp, keys[0].fmr_length);
501 	xfs_getfsmap_set_irec_flags(&info->low, &keys[0]);
502 
503 	/* Adjust the low key if we are continuing from where we left off. */
504 	if (info->low.rm_blockcount == 0) {
505 		/* No previous record from which to continue */
506 	} else if (rmap_not_shareable(mp, &info->low)) {
507 		/* Last record seen was an unshareable extent */
508 		info->low.rm_owner = 0;
509 		info->low.rm_offset = 0;
510 
511 		start_fsb += info->low.rm_blockcount;
512 		if (XFS_FSB_TO_DADDR(mp, start_fsb) >= eofs)
513 			return 0;
514 	} else {
515 		/* Last record seen was a shareable file data extent */
516 		info->low.rm_offset += info->low.rm_blockcount;
517 	}
518 	info->low.rm_startblock = XFS_FSB_TO_AGBNO(mp, start_fsb);
519 
520 	info->high.rm_startblock = -1U;
521 	info->high.rm_owner = ULLONG_MAX;
522 	info->high.rm_offset = ULLONG_MAX;
523 	info->high.rm_blockcount = 0;
524 	info->high.rm_flags = XFS_RMAP_KEY_FLAGS | XFS_RMAP_REC_FLAGS;
525 
526 	start_ag = XFS_FSB_TO_AGNO(mp, start_fsb);
527 	end_ag = XFS_FSB_TO_AGNO(mp, end_fsb);
528 
529 	while ((pag = xfs_perag_next_range(mp, pag, start_ag, end_ag))) {
530 		/*
531 		 * Set the AG high key from the fsmap high key if this
532 		 * is the last AG that we're querying.
533 		 */
534 		info->group = pag_group(pag);
535 		if (pag_agno(pag) == end_ag) {
536 			info->high.rm_startblock = XFS_FSB_TO_AGBNO(mp,
537 					end_fsb);
538 			info->high.rm_offset = XFS_BB_TO_FSBT(mp,
539 					keys[1].fmr_offset);
540 			error = xfs_fsmap_owner_to_rmap(&info->high, &keys[1]);
541 			if (error)
542 				break;
543 			xfs_getfsmap_set_irec_flags(&info->high, &keys[1]);
544 		}
545 
546 		if (bt_cur) {
547 			xfs_btree_del_cursor(bt_cur, XFS_BTREE_NOERROR);
548 			bt_cur = NULL;
549 			xfs_trans_brelse(tp, info->agf_bp);
550 			info->agf_bp = NULL;
551 		}
552 
553 		error = xfs_alloc_read_agf(pag, tp, 0, &info->agf_bp);
554 		if (error)
555 			break;
556 
557 		trace_xfs_fsmap_low_group_key(mp, info->dev, pag_agno(pag),
558 				&info->low);
559 		trace_xfs_fsmap_high_group_key(mp, info->dev, pag_agno(pag),
560 				&info->high);
561 
562 		error = query_fn(tp, info, &bt_cur, priv);
563 		if (error)
564 			break;
565 
566 		/*
567 		 * Set the AG low key to the start of the AG prior to
568 		 * moving on to the next AG.
569 		 */
570 		if (pag_agno(pag) == start_ag)
571 			memset(&info->low, 0, sizeof(info->low));
572 
573 		/*
574 		 * If this is the last AG, report any gap at the end of it
575 		 * before we drop the reference to the perag when the loop
576 		 * terminates.
577 		 */
578 		if (pag_agno(pag) == end_ag) {
579 			info->last = true;
580 			error = query_fn(tp, info, &bt_cur, priv);
581 			if (error)
582 				break;
583 		}
584 		info->group = NULL;
585 	}
586 
587 	if (bt_cur)
588 		xfs_btree_del_cursor(bt_cur, error < 0 ? XFS_BTREE_ERROR :
589 							 XFS_BTREE_NOERROR);
590 	if (info->agf_bp) {
591 		xfs_trans_brelse(tp, info->agf_bp);
592 		info->agf_bp = NULL;
593 	}
594 	if (info->group) {
595 		xfs_perag_rele(pag);
596 		info->group = NULL;
597 	} else if (pag) {
598 		/* loop termination case */
599 		xfs_perag_rele(pag);
600 	}
601 
602 	return error;
603 }
604 
605 /* Actually query the rmap btree. */
606 STATIC int
xfs_getfsmap_datadev_rmapbt_query(struct xfs_trans * tp,struct xfs_getfsmap_info * info,struct xfs_btree_cur ** curpp,void * priv)607 xfs_getfsmap_datadev_rmapbt_query(
608 	struct xfs_trans		*tp,
609 	struct xfs_getfsmap_info	*info,
610 	struct xfs_btree_cur		**curpp,
611 	void				*priv)
612 {
613 	/* Report any gap at the end of the last AG. */
614 	if (info->last)
615 		return xfs_getfsmap_rmapbt_helper(*curpp, &info->high, info);
616 
617 	/* Allocate cursor for this AG and query_range it. */
618 	*curpp = xfs_rmapbt_init_cursor(tp->t_mountp, tp, info->agf_bp,
619 			to_perag(info->group));
620 	return xfs_rmap_query_range(*curpp, &info->low, &info->high,
621 			xfs_getfsmap_rmapbt_helper, info);
622 }
623 
624 /* Execute a getfsmap query against the regular data device rmapbt. */
625 STATIC int
xfs_getfsmap_datadev_rmapbt(struct xfs_trans * tp,const struct xfs_fsmap * keys,struct xfs_getfsmap_info * info)626 xfs_getfsmap_datadev_rmapbt(
627 	struct xfs_trans		*tp,
628 	const struct xfs_fsmap		*keys,
629 	struct xfs_getfsmap_info	*info)
630 {
631 	info->missing_owner = XFS_FMR_OWN_FREE;
632 	return __xfs_getfsmap_datadev(tp, keys, info,
633 			xfs_getfsmap_datadev_rmapbt_query, NULL);
634 }
635 
636 /* Actually query the bno btree. */
637 STATIC int
xfs_getfsmap_datadev_bnobt_query(struct xfs_trans * tp,struct xfs_getfsmap_info * info,struct xfs_btree_cur ** curpp,void * priv)638 xfs_getfsmap_datadev_bnobt_query(
639 	struct xfs_trans		*tp,
640 	struct xfs_getfsmap_info	*info,
641 	struct xfs_btree_cur		**curpp,
642 	void				*priv)
643 {
644 	struct xfs_alloc_rec_incore	*key = priv;
645 
646 	/* Report any gap at the end of the last AG. */
647 	if (info->last)
648 		return xfs_getfsmap_datadev_bnobt_helper(*curpp, &key[1], info);
649 
650 	/* Allocate cursor for this AG and query_range it. */
651 	*curpp = xfs_bnobt_init_cursor(tp->t_mountp, tp, info->agf_bp,
652 			to_perag(info->group));
653 	key->ar_startblock = info->low.rm_startblock;
654 	key[1].ar_startblock = info->high.rm_startblock;
655 	return xfs_alloc_query_range(*curpp, key, &key[1],
656 			xfs_getfsmap_datadev_bnobt_helper, info);
657 }
658 
659 /* Execute a getfsmap query against the regular data device's bnobt. */
660 STATIC int
xfs_getfsmap_datadev_bnobt(struct xfs_trans * tp,const struct xfs_fsmap * keys,struct xfs_getfsmap_info * info)661 xfs_getfsmap_datadev_bnobt(
662 	struct xfs_trans		*tp,
663 	const struct xfs_fsmap		*keys,
664 	struct xfs_getfsmap_info	*info)
665 {
666 	struct xfs_alloc_rec_incore	akeys[2];
667 
668 	memset(akeys, 0, sizeof(akeys));
669 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
670 	return __xfs_getfsmap_datadev(tp, keys, info,
671 			xfs_getfsmap_datadev_bnobt_query, &akeys[0]);
672 }
673 
674 /* Execute a getfsmap query against the log device. */
675 STATIC int
xfs_getfsmap_logdev(struct xfs_trans * tp,const struct xfs_fsmap * keys,struct xfs_getfsmap_info * info)676 xfs_getfsmap_logdev(
677 	struct xfs_trans		*tp,
678 	const struct xfs_fsmap		*keys,
679 	struct xfs_getfsmap_info	*info)
680 {
681 	struct xfs_fsmap_irec		frec = {
682 		.start_daddr		= 0,
683 		.rec_key		= 0,
684 		.owner			= XFS_RMAP_OWN_LOG,
685 	};
686 	struct xfs_mount		*mp = tp->t_mountp;
687 	xfs_fsblock_t			start_fsb, end_fsb;
688 	uint64_t			eofs;
689 
690 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
691 	if (keys[0].fmr_physical >= eofs)
692 		return 0;
693 	start_fsb = XFS_BB_TO_FSBT(mp,
694 				keys[0].fmr_physical + keys[0].fmr_length);
695 	end_fsb = XFS_BB_TO_FSB(mp, min(eofs - 1, keys[1].fmr_physical));
696 
697 	/* Adjust the low key if we are continuing from where we left off. */
698 	if (keys[0].fmr_length > 0)
699 		info->low_daddr = XFS_FSB_TO_BB(mp, start_fsb);
700 
701 	trace_xfs_fsmap_low_linear_key(mp, info->dev, start_fsb);
702 	trace_xfs_fsmap_high_linear_key(mp, info->dev, end_fsb);
703 
704 	if (start_fsb > 0)
705 		return 0;
706 
707 	/* Fabricate an rmap entry for the external log device. */
708 	frec.len_daddr = XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
709 	return xfs_getfsmap_helper(tp, info, &frec);
710 }
711 
712 #ifdef CONFIG_XFS_RT
713 /* Transform a rtbitmap "record" into a fsmap */
714 STATIC int
xfs_getfsmap_rtdev_rtbitmap_helper(struct xfs_rtgroup * rtg,struct xfs_trans * tp,const struct xfs_rtalloc_rec * rec,void * priv)715 xfs_getfsmap_rtdev_rtbitmap_helper(
716 	struct xfs_rtgroup		*rtg,
717 	struct xfs_trans		*tp,
718 	const struct xfs_rtalloc_rec	*rec,
719 	void				*priv)
720 {
721 	struct xfs_fsmap_irec		frec = {
722 		.owner			= XFS_RMAP_OWN_NULL, /* "free" */
723 	};
724 	struct xfs_mount		*mp = rtg_mount(rtg);
725 	struct xfs_getfsmap_info	*info = priv;
726 	xfs_rtblock_t			start_rtb =
727 				xfs_rtx_to_rtb(rtg, rec->ar_startext);
728 	uint64_t			rtbcount =
729 				xfs_rtbxlen_to_blen(mp, rec->ar_extcount);
730 
731 	/*
732 	 * For an info->last query, we're looking for a gap between the last
733 	 * mapping emitted and the high key specified by userspace.  If the
734 	 * user's query spans less than 1 fsblock, then info->high and
735 	 * info->low will have the same rm_startblock, which causes rec_daddr
736 	 * and next_daddr to be the same.  Therefore, use the end_daddr that
737 	 * we calculated from userspace's high key to synthesize the record.
738 	 * Note that if the btree query found a mapping, there won't be a gap.
739 	 */
740 	if (info->last)
741 		frec.start_daddr = info->end_daddr + 1;
742 	else
743 		frec.start_daddr = xfs_rtb_to_daddr(mp, start_rtb);
744 
745 	frec.len_daddr = XFS_FSB_TO_BB(mp, rtbcount);
746 	return xfs_getfsmap_helper(tp, info, &frec);
747 }
748 
749 /* Execute a getfsmap query against the realtime device rtbitmap. */
750 STATIC int
xfs_getfsmap_rtdev_rtbitmap(struct xfs_trans * tp,const struct xfs_fsmap * keys,struct xfs_getfsmap_info * info)751 xfs_getfsmap_rtdev_rtbitmap(
752 	struct xfs_trans		*tp,
753 	const struct xfs_fsmap		*keys,
754 	struct xfs_getfsmap_info	*info)
755 {
756 	struct xfs_mount		*mp = tp->t_mountp;
757 	xfs_rtblock_t			start_rtbno, end_rtbno;
758 	xfs_rtxnum_t			start_rtx, end_rtx;
759 	xfs_rgnumber_t			start_rgno, end_rgno;
760 	struct xfs_rtgroup		*rtg = NULL;
761 	uint64_t			eofs;
762 	int				error;
763 
764 	eofs = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
765 	if (keys[0].fmr_physical >= eofs)
766 		return 0;
767 
768 	info->missing_owner = XFS_FMR_OWN_UNKNOWN;
769 
770 	/* Adjust the low key if we are continuing from where we left off. */
771 	start_rtbno = xfs_daddr_to_rtb(mp,
772 			keys[0].fmr_physical + keys[0].fmr_length);
773 	if (keys[0].fmr_length > 0) {
774 		info->low_daddr = xfs_rtb_to_daddr(mp, start_rtbno);
775 		if (info->low_daddr >= eofs)
776 			return 0;
777 	}
778 	start_rtx = xfs_rtb_to_rtx(mp, start_rtbno);
779 	start_rgno = xfs_rtb_to_rgno(mp, start_rtbno);
780 
781 	end_rtbno = xfs_daddr_to_rtb(mp, min(eofs - 1, keys[1].fmr_physical));
782 	end_rgno = xfs_rtb_to_rgno(mp, end_rtbno);
783 
784 	trace_xfs_fsmap_low_linear_key(mp, info->dev, start_rtbno);
785 	trace_xfs_fsmap_high_linear_key(mp, info->dev, end_rtbno);
786 
787 	end_rtx = -1ULL;
788 
789 	while ((rtg = xfs_rtgroup_next_range(mp, rtg, start_rgno, end_rgno))) {
790 		if (rtg_rgno(rtg) == end_rgno)
791 			end_rtx = xfs_rtb_to_rtx(mp,
792 					end_rtbno + mp->m_sb.sb_rextsize - 1);
793 
794 		info->group = rtg_group(rtg);
795 		xfs_rtgroup_lock(rtg, XFS_RTGLOCK_BITMAP_SHARED);
796 		error = xfs_rtalloc_query_range(rtg, tp, start_rtx, end_rtx,
797 				xfs_getfsmap_rtdev_rtbitmap_helper, info);
798 		if (error)
799 			break;
800 
801 		/*
802 		 * Report any gaps at the end of the rtbitmap by simulating a
803 		 * zero-length free extent starting at the rtx after the end
804 		 * of the query range.
805 		 */
806 		if (rtg_rgno(rtg) == end_rgno) {
807 			struct xfs_rtalloc_rec	ahigh = {
808 				.ar_startext	= min(end_rtx + 1,
809 						      rtg->rtg_extents),
810 			};
811 
812 			info->last = true;
813 			error = xfs_getfsmap_rtdev_rtbitmap_helper(rtg, tp,
814 					&ahigh, info);
815 			if (error)
816 				break;
817 		}
818 
819 		xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_BITMAP_SHARED);
820 		info->group = NULL;
821 		start_rtx = 0;
822 	}
823 
824 	/* loop termination case */
825 	if (rtg) {
826 		if (info->group) {
827 			xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_BITMAP_SHARED);
828 			info->group = NULL;
829 		}
830 		xfs_rtgroup_rele(rtg);
831 	}
832 
833 	return error;
834 }
835 #endif /* CONFIG_XFS_RT */
836 
837 /* Do we recognize the device? */
838 STATIC bool
xfs_getfsmap_is_valid_device(struct xfs_mount * mp,struct xfs_fsmap * fm)839 xfs_getfsmap_is_valid_device(
840 	struct xfs_mount	*mp,
841 	struct xfs_fsmap	*fm)
842 {
843 	if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
844 	    fm->fmr_device == new_encode_dev(mp->m_ddev_targp->bt_dev))
845 		return true;
846 	if (mp->m_logdev_targp &&
847 	    fm->fmr_device == new_encode_dev(mp->m_logdev_targp->bt_dev))
848 		return true;
849 	if (mp->m_rtdev_targp &&
850 	    fm->fmr_device == new_encode_dev(mp->m_rtdev_targp->bt_dev))
851 		return true;
852 	return false;
853 }
854 
855 /* Ensure that the low key is less than the high key. */
856 STATIC bool
xfs_getfsmap_check_keys(struct xfs_fsmap * low_key,struct xfs_fsmap * high_key)857 xfs_getfsmap_check_keys(
858 	struct xfs_fsmap		*low_key,
859 	struct xfs_fsmap		*high_key)
860 {
861 	if (low_key->fmr_flags & (FMR_OF_SPECIAL_OWNER | FMR_OF_EXTENT_MAP)) {
862 		if (low_key->fmr_offset)
863 			return false;
864 	}
865 	if (high_key->fmr_flags != -1U &&
866 	    (high_key->fmr_flags & (FMR_OF_SPECIAL_OWNER |
867 				    FMR_OF_EXTENT_MAP))) {
868 		if (high_key->fmr_offset && high_key->fmr_offset != -1ULL)
869 			return false;
870 	}
871 	if (high_key->fmr_length && high_key->fmr_length != -1ULL)
872 		return false;
873 
874 	if (low_key->fmr_device > high_key->fmr_device)
875 		return false;
876 	if (low_key->fmr_device < high_key->fmr_device)
877 		return true;
878 
879 	if (low_key->fmr_physical > high_key->fmr_physical)
880 		return false;
881 	if (low_key->fmr_physical < high_key->fmr_physical)
882 		return true;
883 
884 	if (low_key->fmr_owner > high_key->fmr_owner)
885 		return false;
886 	if (low_key->fmr_owner < high_key->fmr_owner)
887 		return true;
888 
889 	if (low_key->fmr_offset > high_key->fmr_offset)
890 		return false;
891 	if (low_key->fmr_offset < high_key->fmr_offset)
892 		return true;
893 
894 	return false;
895 }
896 
897 /*
898  * There are only two devices if we didn't configure RT devices at build time.
899  */
900 #ifdef CONFIG_XFS_RT
901 #define XFS_GETFSMAP_DEVS	3
902 #else
903 #define XFS_GETFSMAP_DEVS	2
904 #endif /* CONFIG_XFS_RT */
905 
906 /*
907  * Get filesystem's extents as described in head, and format for output. Fills
908  * in the supplied records array until there are no more reverse mappings to
909  * return or head.fmh_entries == head.fmh_count.  In the second case, this
910  * function returns -ECANCELED to indicate that more records would have been
911  * returned.
912  *
913  * Key to Confusion
914  * ----------------
915  * There are multiple levels of keys and counters at work here:
916  * xfs_fsmap_head.fmh_keys	-- low and high fsmap keys passed in;
917  *				   these reflect fs-wide sector addrs.
918  * dkeys			-- fmh_keys used to query each device;
919  *				   these are fmh_keys but w/ the low key
920  *				   bumped up by fmr_length.
921  * xfs_getfsmap_info.next_daddr	-- next disk addr we expect to see; this
922  *				   is how we detect gaps in the fsmap
923 				   records and report them.
924  * xfs_getfsmap_info.low/high	-- per-AG low/high keys computed from
925  *				   dkeys; used to query the metadata.
926  */
927 STATIC int
xfs_getfsmap(struct xfs_mount * mp,struct xfs_fsmap_head * head,struct fsmap * fsmap_recs)928 xfs_getfsmap(
929 	struct xfs_mount		*mp,
930 	struct xfs_fsmap_head		*head,
931 	struct fsmap			*fsmap_recs)
932 {
933 	struct xfs_trans		*tp = NULL;
934 	struct xfs_fsmap		dkeys[2];	/* per-dev keys */
935 	struct xfs_getfsmap_dev		handlers[XFS_GETFSMAP_DEVS];
936 	struct xfs_getfsmap_info	info = {
937 		.fsmap_recs		= fsmap_recs,
938 		.head			= head,
939 	};
940 	bool				use_rmap;
941 	int				i;
942 	int				error = 0;
943 
944 	if (head->fmh_iflags & ~FMH_IF_VALID)
945 		return -EINVAL;
946 	if (!xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[0]) ||
947 	    !xfs_getfsmap_is_valid_device(mp, &head->fmh_keys[1]))
948 		return -EINVAL;
949 	if (!xfs_getfsmap_check_keys(&head->fmh_keys[0], &head->fmh_keys[1]))
950 		return -EINVAL;
951 
952 	use_rmap = xfs_has_rmapbt(mp) &&
953 		   has_capability_noaudit(current, CAP_SYS_ADMIN);
954 	head->fmh_entries = 0;
955 
956 	/* Set up our device handlers. */
957 	memset(handlers, 0, sizeof(handlers));
958 	handlers[0].nr_sectors = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
959 	handlers[0].dev = new_encode_dev(mp->m_ddev_targp->bt_dev);
960 	if (use_rmap)
961 		handlers[0].fn = xfs_getfsmap_datadev_rmapbt;
962 	else
963 		handlers[0].fn = xfs_getfsmap_datadev_bnobt;
964 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
965 		handlers[1].nr_sectors = XFS_FSB_TO_BB(mp,
966 						       mp->m_sb.sb_logblocks);
967 		handlers[1].dev = new_encode_dev(mp->m_logdev_targp->bt_dev);
968 		handlers[1].fn = xfs_getfsmap_logdev;
969 	}
970 #ifdef CONFIG_XFS_RT
971 	if (mp->m_rtdev_targp) {
972 		handlers[2].nr_sectors = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
973 		handlers[2].dev = new_encode_dev(mp->m_rtdev_targp->bt_dev);
974 		handlers[2].fn = xfs_getfsmap_rtdev_rtbitmap;
975 	}
976 #endif /* CONFIG_XFS_RT */
977 
978 	xfs_sort(handlers, XFS_GETFSMAP_DEVS, sizeof(struct xfs_getfsmap_dev),
979 			xfs_getfsmap_dev_compare);
980 
981 	/*
982 	 * To continue where we left off, we allow userspace to use the
983 	 * last mapping from a previous call as the low key of the next.
984 	 * This is identified by a non-zero length in the low key. We
985 	 * have to increment the low key in this scenario to ensure we
986 	 * don't return the same mapping again, and instead return the
987 	 * very next mapping.
988 	 *
989 	 * If the low key mapping refers to file data, the same physical
990 	 * blocks could be mapped to several other files/offsets.
991 	 * According to rmapbt record ordering, the minimal next
992 	 * possible record for the block range is the next starting
993 	 * offset in the same inode. Therefore, each fsmap backend bumps
994 	 * the file offset to continue the search appropriately.  For
995 	 * all other low key mapping types (attr blocks, metadata), each
996 	 * fsmap backend bumps the physical offset as there can be no
997 	 * other mapping for the same physical block range.
998 	 */
999 	dkeys[0] = head->fmh_keys[0];
1000 	memset(&dkeys[1], 0xFF, sizeof(struct xfs_fsmap));
1001 
1002 	info.next_daddr = head->fmh_keys[0].fmr_physical +
1003 			  head->fmh_keys[0].fmr_length;
1004 
1005 	/* For each device we support... */
1006 	for (i = 0; i < XFS_GETFSMAP_DEVS; i++) {
1007 		/* Is this device within the range the user asked for? */
1008 		if (!handlers[i].fn)
1009 			continue;
1010 		if (head->fmh_keys[0].fmr_device > handlers[i].dev)
1011 			continue;
1012 		if (head->fmh_keys[1].fmr_device < handlers[i].dev)
1013 			break;
1014 
1015 		/*
1016 		 * If this device number matches the high key, we have to pass
1017 		 * the high key to the handler to limit the query results, and
1018 		 * set the end_daddr so that we can synthesize records at the
1019 		 * end of the query range or device.
1020 		 */
1021 		if (handlers[i].dev == head->fmh_keys[1].fmr_device) {
1022 			dkeys[1] = head->fmh_keys[1];
1023 			info.end_daddr = min(handlers[i].nr_sectors - 1,
1024 					     dkeys[1].fmr_physical);
1025 		} else {
1026 			info.end_daddr = handlers[i].nr_sectors - 1;
1027 		}
1028 
1029 		/*
1030 		 * If the device number exceeds the low key, zero out the low
1031 		 * key so that we get everything from the beginning.
1032 		 */
1033 		if (handlers[i].dev > head->fmh_keys[0].fmr_device)
1034 			memset(&dkeys[0], 0, sizeof(struct xfs_fsmap));
1035 
1036 		/*
1037 		 * Grab an empty transaction so that we can use its recursive
1038 		 * buffer locking abilities to detect cycles in the rmapbt
1039 		 * without deadlocking.
1040 		 */
1041 		error = xfs_trans_alloc_empty(mp, &tp);
1042 		if (error)
1043 			break;
1044 
1045 		info.dev = handlers[i].dev;
1046 		info.last = false;
1047 		info.group = NULL;
1048 		info.low_daddr = XFS_BUF_DADDR_NULL;
1049 		info.low.rm_blockcount = 0;
1050 		error = handlers[i].fn(tp, dkeys, &info);
1051 		if (error)
1052 			break;
1053 		xfs_trans_cancel(tp);
1054 		tp = NULL;
1055 		info.next_daddr = 0;
1056 	}
1057 
1058 	if (tp)
1059 		xfs_trans_cancel(tp);
1060 	head->fmh_oflags = FMH_OF_DEV_T;
1061 	return error;
1062 }
1063 
1064 int
xfs_ioc_getfsmap(struct xfs_inode * ip,struct fsmap_head __user * arg)1065 xfs_ioc_getfsmap(
1066 	struct xfs_inode	*ip,
1067 	struct fsmap_head	__user *arg)
1068 {
1069 	struct xfs_fsmap_head	xhead = {0};
1070 	struct fsmap_head	head;
1071 	struct fsmap		*recs;
1072 	unsigned int		count;
1073 	__u32			last_flags = 0;
1074 	bool			done = false;
1075 	int			error;
1076 
1077 	if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
1078 		return -EFAULT;
1079 	if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
1080 	    memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
1081 		       sizeof(head.fmh_keys[0].fmr_reserved)) ||
1082 	    memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
1083 		       sizeof(head.fmh_keys[1].fmr_reserved)))
1084 		return -EINVAL;
1085 
1086 	/*
1087 	 * Use an internal memory buffer so that we don't have to copy fsmap
1088 	 * data to userspace while holding locks.  Start by trying to allocate
1089 	 * up to 128k for the buffer, but fall back to a single page if needed.
1090 	 */
1091 	count = min_t(unsigned int, head.fmh_count,
1092 			131072 / sizeof(struct fsmap));
1093 	recs = kvcalloc(count, sizeof(struct fsmap), GFP_KERNEL);
1094 	if (!recs) {
1095 		count = min_t(unsigned int, head.fmh_count,
1096 				PAGE_SIZE / sizeof(struct fsmap));
1097 		recs = kvcalloc(count, sizeof(struct fsmap), GFP_KERNEL);
1098 		if (!recs)
1099 			return -ENOMEM;
1100 	}
1101 
1102 	xhead.fmh_iflags = head.fmh_iflags;
1103 	xfs_fsmap_to_internal(&xhead.fmh_keys[0], &head.fmh_keys[0]);
1104 	xfs_fsmap_to_internal(&xhead.fmh_keys[1], &head.fmh_keys[1]);
1105 
1106 	trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1107 	trace_xfs_getfsmap_high_key(ip->i_mount, &xhead.fmh_keys[1]);
1108 
1109 	head.fmh_entries = 0;
1110 	do {
1111 		struct fsmap __user	*user_recs;
1112 		struct fsmap		*last_rec;
1113 
1114 		user_recs = &arg->fmh_recs[head.fmh_entries];
1115 		xhead.fmh_entries = 0;
1116 		xhead.fmh_count = min_t(unsigned int, count,
1117 					head.fmh_count - head.fmh_entries);
1118 
1119 		/* Run query, record how many entries we got. */
1120 		error = xfs_getfsmap(ip->i_mount, &xhead, recs);
1121 		switch (error) {
1122 		case 0:
1123 			/*
1124 			 * There are no more records in the result set.  Copy
1125 			 * whatever we got to userspace and break out.
1126 			 */
1127 			done = true;
1128 			break;
1129 		case -ECANCELED:
1130 			/*
1131 			 * The internal memory buffer is full.  Copy whatever
1132 			 * records we got to userspace and go again if we have
1133 			 * not yet filled the userspace buffer.
1134 			 */
1135 			error = 0;
1136 			break;
1137 		default:
1138 			goto out_free;
1139 		}
1140 		head.fmh_entries += xhead.fmh_entries;
1141 		head.fmh_oflags = xhead.fmh_oflags;
1142 
1143 		/*
1144 		 * If the caller wanted a record count or there aren't any
1145 		 * new records to return, we're done.
1146 		 */
1147 		if (head.fmh_count == 0 || xhead.fmh_entries == 0)
1148 			break;
1149 
1150 		/* Copy all the records we got out to userspace. */
1151 		if (copy_to_user(user_recs, recs,
1152 				 xhead.fmh_entries * sizeof(struct fsmap))) {
1153 			error = -EFAULT;
1154 			goto out_free;
1155 		}
1156 
1157 		/* Remember the last record flags we copied to userspace. */
1158 		last_rec = &recs[xhead.fmh_entries - 1];
1159 		last_flags = last_rec->fmr_flags;
1160 
1161 		/* Set up the low key for the next iteration. */
1162 		xfs_fsmap_to_internal(&xhead.fmh_keys[0], last_rec);
1163 		trace_xfs_getfsmap_low_key(ip->i_mount, &xhead.fmh_keys[0]);
1164 	} while (!done && head.fmh_entries < head.fmh_count);
1165 
1166 	/*
1167 	 * If there are no more records in the query result set and we're not
1168 	 * in counting mode, mark the last record returned with the LAST flag.
1169 	 */
1170 	if (done && head.fmh_count > 0 && head.fmh_entries > 0) {
1171 		struct fsmap __user	*user_rec;
1172 
1173 		last_flags |= FMR_OF_LAST;
1174 		user_rec = &arg->fmh_recs[head.fmh_entries - 1];
1175 
1176 		if (copy_to_user(&user_rec->fmr_flags, &last_flags,
1177 					sizeof(last_flags))) {
1178 			error = -EFAULT;
1179 			goto out_free;
1180 		}
1181 	}
1182 
1183 	/* copy back header */
1184 	if (copy_to_user(arg, &head, sizeof(struct fsmap_head))) {
1185 		error = -EFAULT;
1186 		goto out_free;
1187 	}
1188 
1189 out_free:
1190 	kvfree(recs);
1191 	return error;
1192 }
1193