xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision 6f52b16c5b29b89d92c0e7236f4655dc8491ad70)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * Copyright (c) 2016 Christoph Hellwig.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include <linux/iomap.h>
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_bmap.h"
32 #include "xfs_bmap_util.h"
33 #include "xfs_error.h"
34 #include "xfs_trans.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_iomap.h"
37 #include "xfs_trace.h"
38 #include "xfs_icache.h"
39 #include "xfs_quota.h"
40 #include "xfs_dquot_item.h"
41 #include "xfs_dquot.h"
42 #include "xfs_reflink.h"
43 
44 
45 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
46 						<< mp->m_writeio_log)
47 
48 void
49 xfs_bmbt_to_iomap(
50 	struct xfs_inode	*ip,
51 	struct iomap		*iomap,
52 	struct xfs_bmbt_irec	*imap)
53 {
54 	struct xfs_mount	*mp = ip->i_mount;
55 
56 	if (imap->br_startblock == HOLESTARTBLOCK) {
57 		iomap->blkno = IOMAP_NULL_BLOCK;
58 		iomap->type = IOMAP_HOLE;
59 	} else if (imap->br_startblock == DELAYSTARTBLOCK) {
60 		iomap->blkno = IOMAP_NULL_BLOCK;
61 		iomap->type = IOMAP_DELALLOC;
62 	} else {
63 		iomap->blkno = xfs_fsb_to_db(ip, imap->br_startblock);
64 		if (imap->br_state == XFS_EXT_UNWRITTEN)
65 			iomap->type = IOMAP_UNWRITTEN;
66 		else
67 			iomap->type = IOMAP_MAPPED;
68 	}
69 	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
70 	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
71 	iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
72 	iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
73 }
74 
75 xfs_extlen_t
76 xfs_eof_alignment(
77 	struct xfs_inode	*ip,
78 	xfs_extlen_t		extsize)
79 {
80 	struct xfs_mount	*mp = ip->i_mount;
81 	xfs_extlen_t		align = 0;
82 
83 	if (!XFS_IS_REALTIME_INODE(ip)) {
84 		/*
85 		 * Round up the allocation request to a stripe unit
86 		 * (m_dalign) boundary if the file size is >= stripe unit
87 		 * size, and we are allocating past the allocation eof.
88 		 *
89 		 * If mounted with the "-o swalloc" option the alignment is
90 		 * increased from the strip unit size to the stripe width.
91 		 */
92 		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
93 			align = mp->m_swidth;
94 		else if (mp->m_dalign)
95 			align = mp->m_dalign;
96 
97 		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
98 			align = 0;
99 	}
100 
101 	/*
102 	 * Always round up the allocation request to an extent boundary
103 	 * (when file on a real-time subvolume or has di_extsize hint).
104 	 */
105 	if (extsize) {
106 		if (align)
107 			align = roundup_64(align, extsize);
108 		else
109 			align = extsize;
110 	}
111 
112 	return align;
113 }
114 
115 STATIC int
116 xfs_iomap_eof_align_last_fsb(
117 	struct xfs_inode	*ip,
118 	xfs_extlen_t		extsize,
119 	xfs_fileoff_t		*last_fsb)
120 {
121 	xfs_extlen_t		align = xfs_eof_alignment(ip, extsize);
122 
123 	if (align) {
124 		xfs_fileoff_t	new_last_fsb = roundup_64(*last_fsb, align);
125 		int		eof, error;
126 
127 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
128 		if (error)
129 			return error;
130 		if (eof)
131 			*last_fsb = new_last_fsb;
132 	}
133 	return 0;
134 }
135 
136 STATIC int
137 xfs_alert_fsblock_zero(
138 	xfs_inode_t	*ip,
139 	xfs_bmbt_irec_t	*imap)
140 {
141 	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
142 			"Access to block zero in inode %llu "
143 			"start_block: %llx start_off: %llx "
144 			"blkcnt: %llx extent-state: %x",
145 		(unsigned long long)ip->i_ino,
146 		(unsigned long long)imap->br_startblock,
147 		(unsigned long long)imap->br_startoff,
148 		(unsigned long long)imap->br_blockcount,
149 		imap->br_state);
150 	return -EFSCORRUPTED;
151 }
152 
153 int
154 xfs_iomap_write_direct(
155 	xfs_inode_t	*ip,
156 	xfs_off_t	offset,
157 	size_t		count,
158 	xfs_bmbt_irec_t *imap,
159 	int		nmaps)
160 {
161 	xfs_mount_t	*mp = ip->i_mount;
162 	xfs_fileoff_t	offset_fsb;
163 	xfs_fileoff_t	last_fsb;
164 	xfs_filblks_t	count_fsb, resaligned;
165 	xfs_fsblock_t	firstfsb;
166 	xfs_extlen_t	extsz;
167 	int		nimaps;
168 	int		quota_flag;
169 	int		rt;
170 	xfs_trans_t	*tp;
171 	struct xfs_defer_ops dfops;
172 	uint		qblocks, resblks, resrtextents;
173 	int		error;
174 	int		lockmode;
175 	int		bmapi_flags = XFS_BMAPI_PREALLOC;
176 	uint		tflags = 0;
177 
178 	rt = XFS_IS_REALTIME_INODE(ip);
179 	extsz = xfs_get_extsz_hint(ip);
180 	lockmode = XFS_ILOCK_SHARED;	/* locked by caller */
181 
182 	ASSERT(xfs_isilocked(ip, lockmode));
183 
184 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
185 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
186 	if ((offset + count) > XFS_ISIZE(ip)) {
187 		/*
188 		 * Assert that the in-core extent list is present since this can
189 		 * call xfs_iread_extents() and we only have the ilock shared.
190 		 * This should be safe because the lock was held around a bmapi
191 		 * call in the caller and we only need it to access the in-core
192 		 * list.
193 		 */
194 		ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
195 								XFS_IFEXTENTS);
196 		error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
197 		if (error)
198 			goto out_unlock;
199 	} else {
200 		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
201 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
202 					imap->br_blockcount +
203 					imap->br_startoff);
204 	}
205 	count_fsb = last_fsb - offset_fsb;
206 	ASSERT(count_fsb > 0);
207 	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
208 
209 	if (unlikely(rt)) {
210 		resrtextents = qblocks = resaligned;
211 		resrtextents /= mp->m_sb.sb_rextsize;
212 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
213 		quota_flag = XFS_QMOPT_RES_RTBLKS;
214 	} else {
215 		resrtextents = 0;
216 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
217 		quota_flag = XFS_QMOPT_RES_REGBLKS;
218 	}
219 
220 	/*
221 	 * Drop the shared lock acquired by the caller, attach the dquot if
222 	 * necessary and move on to transaction setup.
223 	 */
224 	xfs_iunlock(ip, lockmode);
225 	error = xfs_qm_dqattach(ip, 0);
226 	if (error)
227 		return error;
228 
229 	/*
230 	 * For DAX, we do not allocate unwritten extents, but instead we zero
231 	 * the block before we commit the transaction.  Ideally we'd like to do
232 	 * this outside the transaction context, but if we commit and then crash
233 	 * we may not have zeroed the blocks and this will be exposed on
234 	 * recovery of the allocation. Hence we must zero before commit.
235 	 *
236 	 * Further, if we are mapping unwritten extents here, we need to zero
237 	 * and convert them to written so that we don't need an unwritten extent
238 	 * callback for DAX. This also means that we need to be able to dip into
239 	 * the reserve block pool for bmbt block allocation if there is no space
240 	 * left but we need to do unwritten extent conversion.
241 	 */
242 	if (IS_DAX(VFS_I(ip))) {
243 		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
244 		if (imap->br_state == XFS_EXT_UNWRITTEN) {
245 			tflags |= XFS_TRANS_RESERVE;
246 			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
247 		}
248 	}
249 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
250 			tflags, &tp);
251 	if (error)
252 		return error;
253 
254 	lockmode = XFS_ILOCK_EXCL;
255 	xfs_ilock(ip, lockmode);
256 
257 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
258 	if (error)
259 		goto out_trans_cancel;
260 
261 	xfs_trans_ijoin(tp, ip, 0);
262 
263 	/*
264 	 * From this point onwards we overwrite the imap pointer that the
265 	 * caller gave to us.
266 	 */
267 	xfs_defer_init(&dfops, &firstfsb);
268 	nimaps = 1;
269 	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
270 				bmapi_flags, &firstfsb, resblks, imap,
271 				&nimaps, &dfops);
272 	if (error)
273 		goto out_bmap_cancel;
274 
275 	/*
276 	 * Complete the transaction
277 	 */
278 	error = xfs_defer_finish(&tp, &dfops);
279 	if (error)
280 		goto out_bmap_cancel;
281 
282 	error = xfs_trans_commit(tp);
283 	if (error)
284 		goto out_unlock;
285 
286 	/*
287 	 * Copy any maps to caller's array and return any error.
288 	 */
289 	if (nimaps == 0) {
290 		error = -ENOSPC;
291 		goto out_unlock;
292 	}
293 
294 	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
295 		error = xfs_alert_fsblock_zero(ip, imap);
296 
297 out_unlock:
298 	xfs_iunlock(ip, lockmode);
299 	return error;
300 
301 out_bmap_cancel:
302 	xfs_defer_cancel(&dfops);
303 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
304 out_trans_cancel:
305 	xfs_trans_cancel(tp);
306 	goto out_unlock;
307 }
308 
309 STATIC bool
310 xfs_quota_need_throttle(
311 	struct xfs_inode *ip,
312 	int type,
313 	xfs_fsblock_t alloc_blocks)
314 {
315 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
316 
317 	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
318 		return false;
319 
320 	/* no hi watermark, no throttle */
321 	if (!dq->q_prealloc_hi_wmark)
322 		return false;
323 
324 	/* under the lo watermark, no throttle */
325 	if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
326 		return false;
327 
328 	return true;
329 }
330 
331 STATIC void
332 xfs_quota_calc_throttle(
333 	struct xfs_inode *ip,
334 	int type,
335 	xfs_fsblock_t *qblocks,
336 	int *qshift,
337 	int64_t	*qfreesp)
338 {
339 	int64_t freesp;
340 	int shift = 0;
341 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
342 
343 	/* no dq, or over hi wmark, squash the prealloc completely */
344 	if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
345 		*qblocks = 0;
346 		*qfreesp = 0;
347 		return;
348 	}
349 
350 	freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
351 	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
352 		shift = 2;
353 		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
354 			shift += 2;
355 		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
356 			shift += 2;
357 	}
358 
359 	if (freesp < *qfreesp)
360 		*qfreesp = freesp;
361 
362 	/* only overwrite the throttle values if we are more aggressive */
363 	if ((freesp >> shift) < (*qblocks >> *qshift)) {
364 		*qblocks = freesp;
365 		*qshift = shift;
366 	}
367 }
368 
369 /*
370  * If we are doing a write at the end of the file and there are no allocations
371  * past this one, then extend the allocation out to the file system's write
372  * iosize.
373  *
374  * If we don't have a user specified preallocation size, dynamically increase
375  * the preallocation size as the size of the file grows.  Cap the maximum size
376  * at a single extent or less if the filesystem is near full. The closer the
377  * filesystem is to full, the smaller the maximum prealocation.
378  *
379  * As an exception we don't do any preallocation at all if the file is smaller
380  * than the minimum preallocation and we are using the default dynamic
381  * preallocation scheme, as it is likely this is the only write to the file that
382  * is going to be done.
383  *
384  * We clean up any extra space left over when the file is closed in
385  * xfs_inactive().
386  */
387 STATIC xfs_fsblock_t
388 xfs_iomap_prealloc_size(
389 	struct xfs_inode	*ip,
390 	loff_t			offset,
391 	loff_t			count,
392 	xfs_extnum_t		idx)
393 {
394 	struct xfs_mount	*mp = ip->i_mount;
395 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
396 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
397 	struct xfs_bmbt_irec	prev;
398 	int			shift = 0;
399 	int64_t			freesp;
400 	xfs_fsblock_t		qblocks;
401 	int			qshift = 0;
402 	xfs_fsblock_t		alloc_blocks = 0;
403 
404 	if (offset + count <= XFS_ISIZE(ip))
405 		return 0;
406 
407 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
408 	    (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
409 		return 0;
410 
411 	/*
412 	 * If an explicit allocsize is set, the file is small, or we
413 	 * are writing behind a hole, then use the minimum prealloc:
414 	 */
415 	if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
416 	    XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
417 	    !xfs_iext_get_extent(ifp, idx - 1, &prev) ||
418 	    prev.br_startoff + prev.br_blockcount < offset_fsb)
419 		return mp->m_writeio_blocks;
420 
421 	/*
422 	 * Determine the initial size of the preallocation. We are beyond the
423 	 * current EOF here, but we need to take into account whether this is
424 	 * a sparse write or an extending write when determining the
425 	 * preallocation size.  Hence we need to look up the extent that ends
426 	 * at the current write offset and use the result to determine the
427 	 * preallocation size.
428 	 *
429 	 * If the extent is a hole, then preallocation is essentially disabled.
430 	 * Otherwise we take the size of the preceding data extent as the basis
431 	 * for the preallocation size. If the size of the extent is greater than
432 	 * half the maximum extent length, then use the current offset as the
433 	 * basis. This ensures that for large files the preallocation size
434 	 * always extends to MAXEXTLEN rather than falling short due to things
435 	 * like stripe unit/width alignment of real extents.
436 	 */
437 	if (prev.br_blockcount <= (MAXEXTLEN >> 1))
438 		alloc_blocks = prev.br_blockcount << 1;
439 	else
440 		alloc_blocks = XFS_B_TO_FSB(mp, offset);
441 	if (!alloc_blocks)
442 		goto check_writeio;
443 	qblocks = alloc_blocks;
444 
445 	/*
446 	 * MAXEXTLEN is not a power of two value but we round the prealloc down
447 	 * to the nearest power of two value after throttling. To prevent the
448 	 * round down from unconditionally reducing the maximum supported prealloc
449 	 * size, we round up first, apply appropriate throttling, round down and
450 	 * cap the value to MAXEXTLEN.
451 	 */
452 	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
453 				       alloc_blocks);
454 
455 	freesp = percpu_counter_read_positive(&mp->m_fdblocks);
456 	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
457 		shift = 2;
458 		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
459 			shift++;
460 		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
461 			shift++;
462 		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
463 			shift++;
464 		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
465 			shift++;
466 	}
467 
468 	/*
469 	 * Check each quota to cap the prealloc size, provide a shift value to
470 	 * throttle with and adjust amount of available space.
471 	 */
472 	if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
473 		xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
474 					&freesp);
475 	if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
476 		xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
477 					&freesp);
478 	if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
479 		xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
480 					&freesp);
481 
482 	/*
483 	 * The final prealloc size is set to the minimum of free space available
484 	 * in each of the quotas and the overall filesystem.
485 	 *
486 	 * The shift throttle value is set to the maximum value as determined by
487 	 * the global low free space values and per-quota low free space values.
488 	 */
489 	alloc_blocks = MIN(alloc_blocks, qblocks);
490 	shift = MAX(shift, qshift);
491 
492 	if (shift)
493 		alloc_blocks >>= shift;
494 	/*
495 	 * rounddown_pow_of_two() returns an undefined result if we pass in
496 	 * alloc_blocks = 0.
497 	 */
498 	if (alloc_blocks)
499 		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
500 	if (alloc_blocks > MAXEXTLEN)
501 		alloc_blocks = MAXEXTLEN;
502 
503 	/*
504 	 * If we are still trying to allocate more space than is
505 	 * available, squash the prealloc hard. This can happen if we
506 	 * have a large file on a small filesystem and the above
507 	 * lowspace thresholds are smaller than MAXEXTLEN.
508 	 */
509 	while (alloc_blocks && alloc_blocks >= freesp)
510 		alloc_blocks >>= 4;
511 check_writeio:
512 	if (alloc_blocks < mp->m_writeio_blocks)
513 		alloc_blocks = mp->m_writeio_blocks;
514 	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
515 				      mp->m_writeio_blocks);
516 	return alloc_blocks;
517 }
518 
519 static int
520 xfs_file_iomap_begin_delay(
521 	struct inode		*inode,
522 	loff_t			offset,
523 	loff_t			count,
524 	struct iomap		*iomap)
525 {
526 	struct xfs_inode	*ip = XFS_I(inode);
527 	struct xfs_mount	*mp = ip->i_mount;
528 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
529 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
530 	xfs_fileoff_t		maxbytes_fsb =
531 		XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
532 	xfs_fileoff_t		end_fsb;
533 	int			error = 0, eof = 0;
534 	struct xfs_bmbt_irec	got;
535 	xfs_extnum_t		idx;
536 	xfs_fsblock_t		prealloc_blocks = 0;
537 
538 	ASSERT(!XFS_IS_REALTIME_INODE(ip));
539 	ASSERT(!xfs_get_extsz_hint(ip));
540 
541 	xfs_ilock(ip, XFS_ILOCK_EXCL);
542 
543 	if (unlikely(XFS_TEST_ERROR(
544 	    (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
545 	     XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
546 	     mp, XFS_ERRTAG_BMAPIFORMAT))) {
547 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
548 		error = -EFSCORRUPTED;
549 		goto out_unlock;
550 	}
551 
552 	XFS_STATS_INC(mp, xs_blk_mapw);
553 
554 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
555 		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
556 		if (error)
557 			goto out_unlock;
558 	}
559 
560 	eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &idx, &got);
561 	if (!eof && got.br_startoff <= offset_fsb) {
562 		if (xfs_is_reflink_inode(ip)) {
563 			bool		shared;
564 
565 			end_fsb = min(XFS_B_TO_FSB(mp, offset + count),
566 					maxbytes_fsb);
567 			xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
568 			error = xfs_reflink_reserve_cow(ip, &got, &shared);
569 			if (error)
570 				goto out_unlock;
571 		}
572 
573 		trace_xfs_iomap_found(ip, offset, count, 0, &got);
574 		goto done;
575 	}
576 
577 	error = xfs_qm_dqattach_locked(ip, 0);
578 	if (error)
579 		goto out_unlock;
580 
581 	/*
582 	 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
583 	 * to keep the chunks of work done where somewhat symmetric with the
584 	 * work writeback does. This is a completely arbitrary number pulled
585 	 * out of thin air as a best guess for initial testing.
586 	 *
587 	 * Note that the values needs to be less than 32-bits wide until
588 	 * the lower level functions are updated.
589 	 */
590 	count = min_t(loff_t, count, 1024 * PAGE_SIZE);
591 	end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
592 
593 	if (eof) {
594 		prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count, idx);
595 		if (prealloc_blocks) {
596 			xfs_extlen_t	align;
597 			xfs_off_t	end_offset;
598 			xfs_fileoff_t	p_end_fsb;
599 
600 			end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
601 			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
602 					prealloc_blocks;
603 
604 			align = xfs_eof_alignment(ip, 0);
605 			if (align)
606 				p_end_fsb = roundup_64(p_end_fsb, align);
607 
608 			p_end_fsb = min(p_end_fsb, maxbytes_fsb);
609 			ASSERT(p_end_fsb > offset_fsb);
610 			prealloc_blocks = p_end_fsb - end_fsb;
611 		}
612 	}
613 
614 retry:
615 	error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
616 			end_fsb - offset_fsb, prealloc_blocks, &got, &idx, eof);
617 	switch (error) {
618 	case 0:
619 		break;
620 	case -ENOSPC:
621 	case -EDQUOT:
622 		/* retry without any preallocation */
623 		trace_xfs_delalloc_enospc(ip, offset, count);
624 		if (prealloc_blocks) {
625 			prealloc_blocks = 0;
626 			goto retry;
627 		}
628 		/*FALLTHRU*/
629 	default:
630 		goto out_unlock;
631 	}
632 
633 	/*
634 	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
635 	 * them out if the write happens to fail.
636 	 */
637 	iomap->flags = IOMAP_F_NEW;
638 	trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
639 done:
640 	if (isnullstartblock(got.br_startblock))
641 		got.br_startblock = DELAYSTARTBLOCK;
642 
643 	if (!got.br_startblock) {
644 		error = xfs_alert_fsblock_zero(ip, &got);
645 		if (error)
646 			goto out_unlock;
647 	}
648 
649 	xfs_bmbt_to_iomap(ip, iomap, &got);
650 
651 out_unlock:
652 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
653 	return error;
654 }
655 
656 /*
657  * Pass in a delayed allocate extent, convert it to real extents;
658  * return to the caller the extent we create which maps on top of
659  * the originating callers request.
660  *
661  * Called without a lock on the inode.
662  *
663  * We no longer bother to look at the incoming map - all we have to
664  * guarantee is that whatever we allocate fills the required range.
665  */
666 int
667 xfs_iomap_write_allocate(
668 	xfs_inode_t	*ip,
669 	int		whichfork,
670 	xfs_off_t	offset,
671 	xfs_bmbt_irec_t *imap)
672 {
673 	xfs_mount_t	*mp = ip->i_mount;
674 	xfs_fileoff_t	offset_fsb, last_block;
675 	xfs_fileoff_t	end_fsb, map_start_fsb;
676 	xfs_fsblock_t	first_block;
677 	struct xfs_defer_ops	dfops;
678 	xfs_filblks_t	count_fsb;
679 	xfs_trans_t	*tp;
680 	int		nimaps;
681 	int		error = 0;
682 	int		flags = XFS_BMAPI_DELALLOC;
683 	int		nres;
684 
685 	if (whichfork == XFS_COW_FORK)
686 		flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
687 
688 	/*
689 	 * Make sure that the dquots are there.
690 	 */
691 	error = xfs_qm_dqattach(ip, 0);
692 	if (error)
693 		return error;
694 
695 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
696 	count_fsb = imap->br_blockcount;
697 	map_start_fsb = imap->br_startoff;
698 
699 	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
700 
701 	while (count_fsb != 0) {
702 		/*
703 		 * Set up a transaction with which to allocate the
704 		 * backing store for the file.  Do allocations in a
705 		 * loop until we get some space in the range we are
706 		 * interested in.  The other space that might be allocated
707 		 * is in the delayed allocation extent on which we sit
708 		 * but before our buffer starts.
709 		 */
710 		nimaps = 0;
711 		while (nimaps == 0) {
712 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
713 			/*
714 			 * We have already reserved space for the extent and any
715 			 * indirect blocks when creating the delalloc extent,
716 			 * there is no need to reserve space in this transaction
717 			 * again.
718 			 */
719 			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
720 					0, XFS_TRANS_RESERVE, &tp);
721 			if (error)
722 				return error;
723 
724 			xfs_ilock(ip, XFS_ILOCK_EXCL);
725 			xfs_trans_ijoin(tp, ip, 0);
726 
727 			xfs_defer_init(&dfops, &first_block);
728 
729 			/*
730 			 * it is possible that the extents have changed since
731 			 * we did the read call as we dropped the ilock for a
732 			 * while. We have to be careful about truncates or hole
733 			 * punchs here - we are not allowed to allocate
734 			 * non-delalloc blocks here.
735 			 *
736 			 * The only protection against truncation is the pages
737 			 * for the range we are being asked to convert are
738 			 * locked and hence a truncate will block on them
739 			 * first.
740 			 *
741 			 * As a result, if we go beyond the range we really
742 			 * need and hit an delalloc extent boundary followed by
743 			 * a hole while we have excess blocks in the map, we
744 			 * will fill the hole incorrectly and overrun the
745 			 * transaction reservation.
746 			 *
747 			 * Using a single map prevents this as we are forced to
748 			 * check each map we look for overlap with the desired
749 			 * range and abort as soon as we find it. Also, given
750 			 * that we only return a single map, having one beyond
751 			 * what we can return is probably a bit silly.
752 			 *
753 			 * We also need to check that we don't go beyond EOF;
754 			 * this is a truncate optimisation as a truncate sets
755 			 * the new file size before block on the pages we
756 			 * currently have locked under writeback. Because they
757 			 * are about to be tossed, we don't need to write them
758 			 * back....
759 			 */
760 			nimaps = 1;
761 			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
762 			error = xfs_bmap_last_offset(ip, &last_block,
763 							XFS_DATA_FORK);
764 			if (error)
765 				goto trans_cancel;
766 
767 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
768 			if ((map_start_fsb + count_fsb) > last_block) {
769 				count_fsb = last_block - map_start_fsb;
770 				if (count_fsb == 0) {
771 					error = -EAGAIN;
772 					goto trans_cancel;
773 				}
774 			}
775 
776 			/*
777 			 * From this point onwards we overwrite the imap
778 			 * pointer that the caller gave to us.
779 			 */
780 			error = xfs_bmapi_write(tp, ip, map_start_fsb,
781 						count_fsb, flags, &first_block,
782 						nres, imap, &nimaps,
783 						&dfops);
784 			if (error)
785 				goto trans_cancel;
786 
787 			error = xfs_defer_finish(&tp, &dfops);
788 			if (error)
789 				goto trans_cancel;
790 
791 			error = xfs_trans_commit(tp);
792 			if (error)
793 				goto error0;
794 
795 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
796 		}
797 
798 		/*
799 		 * See if we were able to allocate an extent that
800 		 * covers at least part of the callers request
801 		 */
802 		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
803 			return xfs_alert_fsblock_zero(ip, imap);
804 
805 		if ((offset_fsb >= imap->br_startoff) &&
806 		    (offset_fsb < (imap->br_startoff +
807 				   imap->br_blockcount))) {
808 			XFS_STATS_INC(mp, xs_xstrat_quick);
809 			return 0;
810 		}
811 
812 		/*
813 		 * So far we have not mapped the requested part of the
814 		 * file, just surrounding data, try again.
815 		 */
816 		count_fsb -= imap->br_blockcount;
817 		map_start_fsb = imap->br_startoff + imap->br_blockcount;
818 	}
819 
820 trans_cancel:
821 	xfs_defer_cancel(&dfops);
822 	xfs_trans_cancel(tp);
823 error0:
824 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
825 	return error;
826 }
827 
828 int
829 xfs_iomap_write_unwritten(
830 	xfs_inode_t	*ip,
831 	xfs_off_t	offset,
832 	xfs_off_t	count,
833 	bool		update_isize)
834 {
835 	xfs_mount_t	*mp = ip->i_mount;
836 	xfs_fileoff_t	offset_fsb;
837 	xfs_filblks_t	count_fsb;
838 	xfs_filblks_t	numblks_fsb;
839 	xfs_fsblock_t	firstfsb;
840 	int		nimaps;
841 	xfs_trans_t	*tp;
842 	xfs_bmbt_irec_t imap;
843 	struct xfs_defer_ops dfops;
844 	struct inode	*inode = VFS_I(ip);
845 	xfs_fsize_t	i_size;
846 	uint		resblks;
847 	int		error;
848 
849 	trace_xfs_unwritten_convert(ip, offset, count);
850 
851 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
852 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
853 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
854 
855 	/*
856 	 * Reserve enough blocks in this transaction for two complete extent
857 	 * btree splits.  We may be converting the middle part of an unwritten
858 	 * extent and in this case we will insert two new extents in the btree
859 	 * each of which could cause a full split.
860 	 *
861 	 * This reservation amount will be used in the first call to
862 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
863 	 * rest of the operation.
864 	 */
865 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
866 
867 	do {
868 		/*
869 		 * Set up a transaction to convert the range of extents
870 		 * from unwritten to real. Do allocations in a loop until
871 		 * we have covered the range passed in.
872 		 *
873 		 * Note that we can't risk to recursing back into the filesystem
874 		 * here as we might be asked to write out the same inode that we
875 		 * complete here and might deadlock on the iolock.
876 		 */
877 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
878 				XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
879 		if (error)
880 			return error;
881 
882 		xfs_ilock(ip, XFS_ILOCK_EXCL);
883 		xfs_trans_ijoin(tp, ip, 0);
884 
885 		/*
886 		 * Modify the unwritten extent state of the buffer.
887 		 */
888 		xfs_defer_init(&dfops, &firstfsb);
889 		nimaps = 1;
890 		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
891 					XFS_BMAPI_CONVERT, &firstfsb, resblks,
892 					&imap, &nimaps, &dfops);
893 		if (error)
894 			goto error_on_bmapi_transaction;
895 
896 		/*
897 		 * Log the updated inode size as we go.  We have to be careful
898 		 * to only log it up to the actual write offset if it is
899 		 * halfway into a block.
900 		 */
901 		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
902 		if (i_size > offset + count)
903 			i_size = offset + count;
904 		if (update_isize && i_size > i_size_read(inode))
905 			i_size_write(inode, i_size);
906 		i_size = xfs_new_eof(ip, i_size);
907 		if (i_size) {
908 			ip->i_d.di_size = i_size;
909 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
910 		}
911 
912 		error = xfs_defer_finish(&tp, &dfops);
913 		if (error)
914 			goto error_on_bmapi_transaction;
915 
916 		error = xfs_trans_commit(tp);
917 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
918 		if (error)
919 			return error;
920 
921 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
922 			return xfs_alert_fsblock_zero(ip, &imap);
923 
924 		if ((numblks_fsb = imap.br_blockcount) == 0) {
925 			/*
926 			 * The numblks_fsb value should always get
927 			 * smaller, otherwise the loop is stuck.
928 			 */
929 			ASSERT(imap.br_blockcount);
930 			break;
931 		}
932 		offset_fsb += numblks_fsb;
933 		count_fsb -= numblks_fsb;
934 	} while (count_fsb > 0);
935 
936 	return 0;
937 
938 error_on_bmapi_transaction:
939 	xfs_defer_cancel(&dfops);
940 	xfs_trans_cancel(tp);
941 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
942 	return error;
943 }
944 
945 static inline bool imap_needs_alloc(struct inode *inode,
946 		struct xfs_bmbt_irec *imap, int nimaps)
947 {
948 	return !nimaps ||
949 		imap->br_startblock == HOLESTARTBLOCK ||
950 		imap->br_startblock == DELAYSTARTBLOCK ||
951 		(IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
952 }
953 
954 static inline bool need_excl_ilock(struct xfs_inode *ip, unsigned flags)
955 {
956 	/*
957 	 * COW writes will allocate delalloc space, so we need to make sure
958 	 * to take the lock exclusively here.
959 	 */
960 	if (xfs_is_reflink_inode(ip) && (flags & (IOMAP_WRITE | IOMAP_ZERO)))
961 		return true;
962 	if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE))
963 		return true;
964 	return false;
965 }
966 
967 static int
968 xfs_file_iomap_begin(
969 	struct inode		*inode,
970 	loff_t			offset,
971 	loff_t			length,
972 	unsigned		flags,
973 	struct iomap		*iomap)
974 {
975 	struct xfs_inode	*ip = XFS_I(inode);
976 	struct xfs_mount	*mp = ip->i_mount;
977 	struct xfs_bmbt_irec	imap;
978 	xfs_fileoff_t		offset_fsb, end_fsb;
979 	int			nimaps = 1, error = 0;
980 	bool			shared = false, trimmed = false;
981 	unsigned		lockmode;
982 
983 	if (XFS_FORCED_SHUTDOWN(mp))
984 		return -EIO;
985 
986 	if (((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE) &&
987 			!IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
988 		/* Reserve delalloc blocks for regular writeback. */
989 		return xfs_file_iomap_begin_delay(inode, offset, length, iomap);
990 	}
991 
992 	if (need_excl_ilock(ip, flags)) {
993 		lockmode = XFS_ILOCK_EXCL;
994 		xfs_ilock(ip, XFS_ILOCK_EXCL);
995 	} else {
996 		lockmode = xfs_ilock_data_map_shared(ip);
997 	}
998 
999 	if ((flags & IOMAP_NOWAIT) && !(ip->i_df.if_flags & XFS_IFEXTENTS)) {
1000 		error = -EAGAIN;
1001 		goto out_unlock;
1002 	}
1003 
1004 	ASSERT(offset <= mp->m_super->s_maxbytes);
1005 	if ((xfs_fsize_t)offset + length > mp->m_super->s_maxbytes)
1006 		length = mp->m_super->s_maxbytes - offset;
1007 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
1008 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1009 
1010 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1011 			       &nimaps, 0);
1012 	if (error)
1013 		goto out_unlock;
1014 
1015 	if (flags & IOMAP_REPORT) {
1016 		/* Trim the mapping to the nearest shared extent boundary. */
1017 		error = xfs_reflink_trim_around_shared(ip, &imap, &shared,
1018 				&trimmed);
1019 		if (error)
1020 			goto out_unlock;
1021 	}
1022 
1023 	if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && xfs_is_reflink_inode(ip)) {
1024 		if (flags & IOMAP_DIRECT) {
1025 			/*
1026 			 * A reflinked inode will result in CoW alloc.
1027 			 * FIXME: It could still overwrite on unshared extents
1028 			 * and not need allocation.
1029 			 */
1030 			if (flags & IOMAP_NOWAIT) {
1031 				error = -EAGAIN;
1032 				goto out_unlock;
1033 			}
1034 			/* may drop and re-acquire the ilock */
1035 			error = xfs_reflink_allocate_cow(ip, &imap, &shared,
1036 					&lockmode);
1037 			if (error)
1038 				goto out_unlock;
1039 		} else {
1040 			error = xfs_reflink_reserve_cow(ip, &imap, &shared);
1041 			if (error)
1042 				goto out_unlock;
1043 		}
1044 
1045 		end_fsb = imap.br_startoff + imap.br_blockcount;
1046 		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
1047 	}
1048 
1049 	if ((flags & IOMAP_WRITE) && imap_needs_alloc(inode, &imap, nimaps)) {
1050 		/*
1051 		 * If nowait is set bail since we are going to make
1052 		 * allocations.
1053 		 */
1054 		if (flags & IOMAP_NOWAIT) {
1055 			error = -EAGAIN;
1056 			goto out_unlock;
1057 		}
1058 		/*
1059 		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1060 		 * pages to keep the chunks of work done where somewhat symmetric
1061 		 * with the work writeback does. This is a completely arbitrary
1062 		 * number pulled out of thin air as a best guess for initial
1063 		 * testing.
1064 		 *
1065 		 * Note that the values needs to be less than 32-bits wide until
1066 		 * the lower level functions are updated.
1067 		 */
1068 		length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1069 		/*
1070 		 * xfs_iomap_write_direct() expects the shared lock. It
1071 		 * is unlocked on return.
1072 		 */
1073 		if (lockmode == XFS_ILOCK_EXCL)
1074 			xfs_ilock_demote(ip, lockmode);
1075 		error = xfs_iomap_write_direct(ip, offset, length, &imap,
1076 				nimaps);
1077 		if (error)
1078 			return error;
1079 
1080 		iomap->flags = IOMAP_F_NEW;
1081 		trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1082 	} else {
1083 		ASSERT(nimaps);
1084 
1085 		xfs_iunlock(ip, lockmode);
1086 		trace_xfs_iomap_found(ip, offset, length, 0, &imap);
1087 	}
1088 
1089 	xfs_bmbt_to_iomap(ip, iomap, &imap);
1090 
1091 	if (shared)
1092 		iomap->flags |= IOMAP_F_SHARED;
1093 	return 0;
1094 out_unlock:
1095 	xfs_iunlock(ip, lockmode);
1096 	return error;
1097 }
1098 
1099 static int
1100 xfs_file_iomap_end_delalloc(
1101 	struct xfs_inode	*ip,
1102 	loff_t			offset,
1103 	loff_t			length,
1104 	ssize_t			written,
1105 	struct iomap		*iomap)
1106 {
1107 	struct xfs_mount	*mp = ip->i_mount;
1108 	xfs_fileoff_t		start_fsb;
1109 	xfs_fileoff_t		end_fsb;
1110 	int			error = 0;
1111 
1112 	/*
1113 	 * Behave as if the write failed if drop writes is enabled. Set the NEW
1114 	 * flag to force delalloc cleanup.
1115 	 */
1116 	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1117 		iomap->flags |= IOMAP_F_NEW;
1118 		written = 0;
1119 	}
1120 
1121 	/*
1122 	 * start_fsb refers to the first unused block after a short write. If
1123 	 * nothing was written, round offset down to point at the first block in
1124 	 * the range.
1125 	 */
1126 	if (unlikely(!written))
1127 		start_fsb = XFS_B_TO_FSBT(mp, offset);
1128 	else
1129 		start_fsb = XFS_B_TO_FSB(mp, offset + written);
1130 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1131 
1132 	/*
1133 	 * Trim delalloc blocks if they were allocated by this write and we
1134 	 * didn't manage to write the whole range.
1135 	 *
1136 	 * We don't need to care about racing delalloc as we hold i_mutex
1137 	 * across the reserve/allocate/unreserve calls. If there are delalloc
1138 	 * blocks in the range, they are ours.
1139 	 */
1140 	if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1141 		truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1142 					 XFS_FSB_TO_B(mp, end_fsb) - 1);
1143 
1144 		xfs_ilock(ip, XFS_ILOCK_EXCL);
1145 		error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1146 					       end_fsb - start_fsb);
1147 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1148 
1149 		if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1150 			xfs_alert(mp, "%s: unable to clean up ino %lld",
1151 				__func__, ip->i_ino);
1152 			return error;
1153 		}
1154 	}
1155 
1156 	return 0;
1157 }
1158 
1159 static int
1160 xfs_file_iomap_end(
1161 	struct inode		*inode,
1162 	loff_t			offset,
1163 	loff_t			length,
1164 	ssize_t			written,
1165 	unsigned		flags,
1166 	struct iomap		*iomap)
1167 {
1168 	if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1169 		return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1170 				length, written, iomap);
1171 	return 0;
1172 }
1173 
1174 const struct iomap_ops xfs_iomap_ops = {
1175 	.iomap_begin		= xfs_file_iomap_begin,
1176 	.iomap_end		= xfs_file_iomap_end,
1177 };
1178 
1179 static int
1180 xfs_xattr_iomap_begin(
1181 	struct inode		*inode,
1182 	loff_t			offset,
1183 	loff_t			length,
1184 	unsigned		flags,
1185 	struct iomap		*iomap)
1186 {
1187 	struct xfs_inode	*ip = XFS_I(inode);
1188 	struct xfs_mount	*mp = ip->i_mount;
1189 	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1190 	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1191 	struct xfs_bmbt_irec	imap;
1192 	int			nimaps = 1, error = 0;
1193 	unsigned		lockmode;
1194 
1195 	if (XFS_FORCED_SHUTDOWN(mp))
1196 		return -EIO;
1197 
1198 	lockmode = xfs_ilock_attr_map_shared(ip);
1199 
1200 	/* if there are no attribute fork or extents, return ENOENT */
1201 	if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1202 		error = -ENOENT;
1203 		goto out_unlock;
1204 	}
1205 
1206 	ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1207 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1208 			       &nimaps, XFS_BMAPI_ENTIRE | XFS_BMAPI_ATTRFORK);
1209 out_unlock:
1210 	xfs_iunlock(ip, lockmode);
1211 
1212 	if (!error) {
1213 		ASSERT(nimaps);
1214 		xfs_bmbt_to_iomap(ip, iomap, &imap);
1215 	}
1216 
1217 	return error;
1218 }
1219 
1220 const struct iomap_ops xfs_xattr_iomap_ops = {
1221 	.iomap_begin		= xfs_xattr_iomap_begin,
1222 };
1223