xfs_iomap.c revision 72a048c1
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_btree.h"
16#include "xfs_bmap_btree.h"
17#include "xfs_bmap.h"
18#include "xfs_bmap_util.h"
19#include "xfs_errortag.h"
20#include "xfs_error.h"
21#include "xfs_trans.h"
22#include "xfs_trans_space.h"
23#include "xfs_inode_item.h"
24#include "xfs_iomap.h"
25#include "xfs_trace.h"
26#include "xfs_quota.h"
27#include "xfs_dquot_item.h"
28#include "xfs_dquot.h"
29#include "xfs_reflink.h"
30
31
32#define XFS_ALLOC_ALIGN(mp, off) \
33	(((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
34
35static int
36xfs_alert_fsblock_zero(
37	xfs_inode_t	*ip,
38	xfs_bmbt_irec_t	*imap)
39{
40	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
41			"Access to block zero in inode %llu "
42			"start_block: %llx start_off: %llx "
43			"blkcnt: %llx extent-state: %x",
44		(unsigned long long)ip->i_ino,
45		(unsigned long long)imap->br_startblock,
46		(unsigned long long)imap->br_startoff,
47		(unsigned long long)imap->br_blockcount,
48		imap->br_state);
49	return -EFSCORRUPTED;
50}
51
52int
53xfs_bmbt_to_iomap(
54	struct xfs_inode	*ip,
55	struct iomap		*iomap,
56	struct xfs_bmbt_irec	*imap,
57	u16			flags)
58{
59	struct xfs_mount	*mp = ip->i_mount;
60	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
61
62	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
63		return xfs_alert_fsblock_zero(ip, imap);
64
65	if (imap->br_startblock == HOLESTARTBLOCK) {
66		iomap->addr = IOMAP_NULL_ADDR;
67		iomap->type = IOMAP_HOLE;
68	} else if (imap->br_startblock == DELAYSTARTBLOCK ||
69		   isnullstartblock(imap->br_startblock)) {
70		iomap->addr = IOMAP_NULL_ADDR;
71		iomap->type = IOMAP_DELALLOC;
72	} else {
73		iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
74		if (imap->br_state == XFS_EXT_UNWRITTEN)
75			iomap->type = IOMAP_UNWRITTEN;
76		else
77			iomap->type = IOMAP_MAPPED;
78	}
79	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
80	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
81	iomap->bdev = target->bt_bdev;
82	iomap->dax_dev = target->bt_daxdev;
83	iomap->flags = flags;
84
85	if (xfs_ipincount(ip) &&
86	    (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
87		iomap->flags |= IOMAP_F_DIRTY;
88	return 0;
89}
90
91static void
92xfs_hole_to_iomap(
93	struct xfs_inode	*ip,
94	struct iomap		*iomap,
95	xfs_fileoff_t		offset_fsb,
96	xfs_fileoff_t		end_fsb)
97{
98	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
99
100	iomap->addr = IOMAP_NULL_ADDR;
101	iomap->type = IOMAP_HOLE;
102	iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
103	iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
104	iomap->bdev = target->bt_bdev;
105	iomap->dax_dev = target->bt_daxdev;
106}
107
108static inline xfs_fileoff_t
109xfs_iomap_end_fsb(
110	struct xfs_mount	*mp,
111	loff_t			offset,
112	loff_t			count)
113{
114	ASSERT(offset <= mp->m_super->s_maxbytes);
115	return min(XFS_B_TO_FSB(mp, offset + count),
116		   XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
117}
118
119static xfs_extlen_t
120xfs_eof_alignment(
121	struct xfs_inode	*ip)
122{
123	struct xfs_mount	*mp = ip->i_mount;
124	xfs_extlen_t		align = 0;
125
126	if (!XFS_IS_REALTIME_INODE(ip)) {
127		/*
128		 * Round up the allocation request to a stripe unit
129		 * (m_dalign) boundary if the file size is >= stripe unit
130		 * size, and we are allocating past the allocation eof.
131		 *
132		 * If mounted with the "-o swalloc" option the alignment is
133		 * increased from the strip unit size to the stripe width.
134		 */
135		if (mp->m_swidth && xfs_has_swalloc(mp))
136			align = mp->m_swidth;
137		else if (mp->m_dalign)
138			align = mp->m_dalign;
139
140		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
141			align = 0;
142	}
143
144	return align;
145}
146
147/*
148 * Check if last_fsb is outside the last extent, and if so grow it to the next
149 * stripe unit boundary.
150 */
151xfs_fileoff_t
152xfs_iomap_eof_align_last_fsb(
153	struct xfs_inode	*ip,
154	xfs_fileoff_t		end_fsb)
155{
156	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
157	xfs_extlen_t		extsz = xfs_get_extsz_hint(ip);
158	xfs_extlen_t		align = xfs_eof_alignment(ip);
159	struct xfs_bmbt_irec	irec;
160	struct xfs_iext_cursor	icur;
161
162	ASSERT(!xfs_need_iread_extents(ifp));
163
164	/*
165	 * Always round up the allocation request to the extent hint boundary.
166	 */
167	if (extsz) {
168		if (align)
169			align = roundup_64(align, extsz);
170		else
171			align = extsz;
172	}
173
174	if (align) {
175		xfs_fileoff_t	aligned_end_fsb = roundup_64(end_fsb, align);
176
177		xfs_iext_last(ifp, &icur);
178		if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
179		    aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
180			return aligned_end_fsb;
181	}
182
183	return end_fsb;
184}
185
186int
187xfs_iomap_write_direct(
188	struct xfs_inode	*ip,
189	xfs_fileoff_t		offset_fsb,
190	xfs_fileoff_t		count_fsb,
191	struct xfs_bmbt_irec	*imap)
192{
193	struct xfs_mount	*mp = ip->i_mount;
194	struct xfs_trans	*tp;
195	xfs_filblks_t		resaligned;
196	int			nimaps;
197	unsigned int		dblocks, rblocks;
198	bool			force = false;
199	int			error;
200	int			bmapi_flags = XFS_BMAPI_PREALLOC;
201	int			nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
202
203	ASSERT(count_fsb > 0);
204
205	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
206					   xfs_get_extsz_hint(ip));
207	if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
208		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
209		rblocks = resaligned;
210	} else {
211		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
212		rblocks = 0;
213	}
214
215	error = xfs_qm_dqattach(ip);
216	if (error)
217		return error;
218
219	/*
220	 * For DAX, we do not allocate unwritten extents, but instead we zero
221	 * the block before we commit the transaction.  Ideally we'd like to do
222	 * this outside the transaction context, but if we commit and then crash
223	 * we may not have zeroed the blocks and this will be exposed on
224	 * recovery of the allocation. Hence we must zero before commit.
225	 *
226	 * Further, if we are mapping unwritten extents here, we need to zero
227	 * and convert them to written so that we don't need an unwritten extent
228	 * callback for DAX. This also means that we need to be able to dip into
229	 * the reserve block pool for bmbt block allocation if there is no space
230	 * left but we need to do unwritten extent conversion.
231	 */
232	if (IS_DAX(VFS_I(ip))) {
233		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
234		if (imap->br_state == XFS_EXT_UNWRITTEN) {
235			force = true;
236			nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
237			dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
238		}
239	}
240
241	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
242			rblocks, force, &tp);
243	if (error)
244		return error;
245
246	error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, nr_exts);
247	if (error)
248		goto out_trans_cancel;
249
250	/*
251	 * From this point onwards we overwrite the imap pointer that the
252	 * caller gave to us.
253	 */
254	nimaps = 1;
255	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
256				imap, &nimaps);
257	if (error)
258		goto out_trans_cancel;
259
260	/*
261	 * Complete the transaction
262	 */
263	error = xfs_trans_commit(tp);
264	if (error)
265		goto out_unlock;
266
267	/*
268	 * Copy any maps to caller's array and return any error.
269	 */
270	if (nimaps == 0) {
271		error = -ENOSPC;
272		goto out_unlock;
273	}
274
275	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
276		error = xfs_alert_fsblock_zero(ip, imap);
277
278out_unlock:
279	xfs_iunlock(ip, XFS_ILOCK_EXCL);
280	return error;
281
282out_trans_cancel:
283	xfs_trans_cancel(tp);
284	goto out_unlock;
285}
286
287STATIC bool
288xfs_quota_need_throttle(
289	struct xfs_inode	*ip,
290	xfs_dqtype_t		type,
291	xfs_fsblock_t		alloc_blocks)
292{
293	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
294
295	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
296		return false;
297
298	/* no hi watermark, no throttle */
299	if (!dq->q_prealloc_hi_wmark)
300		return false;
301
302	/* under the lo watermark, no throttle */
303	if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
304		return false;
305
306	return true;
307}
308
309STATIC void
310xfs_quota_calc_throttle(
311	struct xfs_inode	*ip,
312	xfs_dqtype_t		type,
313	xfs_fsblock_t		*qblocks,
314	int			*qshift,
315	int64_t			*qfreesp)
316{
317	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
318	int64_t			freesp;
319	int			shift = 0;
320
321	/* no dq, or over hi wmark, squash the prealloc completely */
322	if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
323		*qblocks = 0;
324		*qfreesp = 0;
325		return;
326	}
327
328	freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
329	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
330		shift = 2;
331		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
332			shift += 2;
333		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
334			shift += 2;
335	}
336
337	if (freesp < *qfreesp)
338		*qfreesp = freesp;
339
340	/* only overwrite the throttle values if we are more aggressive */
341	if ((freesp >> shift) < (*qblocks >> *qshift)) {
342		*qblocks = freesp;
343		*qshift = shift;
344	}
345}
346
347/*
348 * If we don't have a user specified preallocation size, dynamically increase
349 * the preallocation size as the size of the file grows.  Cap the maximum size
350 * at a single extent or less if the filesystem is near full. The closer the
351 * filesystem is to being full, the smaller the maximum preallocation.
352 */
353STATIC xfs_fsblock_t
354xfs_iomap_prealloc_size(
355	struct xfs_inode	*ip,
356	int			whichfork,
357	loff_t			offset,
358	loff_t			count,
359	struct xfs_iext_cursor	*icur)
360{
361	struct xfs_iext_cursor	ncur = *icur;
362	struct xfs_bmbt_irec	prev, got;
363	struct xfs_mount	*mp = ip->i_mount;
364	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
365	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
366	int64_t			freesp;
367	xfs_fsblock_t		qblocks;
368	xfs_fsblock_t		alloc_blocks = 0;
369	xfs_extlen_t		plen;
370	int			shift = 0;
371	int			qshift = 0;
372
373	/*
374	 * As an exception we don't do any preallocation at all if the file is
375	 * smaller than the minimum preallocation and we are using the default
376	 * dynamic preallocation scheme, as it is likely this is the only write
377	 * to the file that is going to be done.
378	 */
379	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
380		return 0;
381
382	/*
383	 * Use the minimum preallocation size for small files or if we are
384	 * writing right after a hole.
385	 */
386	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
387	    !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
388	    prev.br_startoff + prev.br_blockcount < offset_fsb)
389		return mp->m_allocsize_blocks;
390
391	/*
392	 * Take the size of the preceding data extents as the basis for the
393	 * preallocation size. Note that we don't care if the previous extents
394	 * are written or not.
395	 */
396	plen = prev.br_blockcount;
397	while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
398		if (plen > MAXEXTLEN / 2 ||
399		    isnullstartblock(got.br_startblock) ||
400		    got.br_startoff + got.br_blockcount != prev.br_startoff ||
401		    got.br_startblock + got.br_blockcount != prev.br_startblock)
402			break;
403		plen += got.br_blockcount;
404		prev = got;
405	}
406
407	/*
408	 * If the size of the extents is greater than half the maximum extent
409	 * length, then use the current offset as the basis.  This ensures that
410	 * for large files the preallocation size always extends to MAXEXTLEN
411	 * rather than falling short due to things like stripe unit/width
412	 * alignment of real extents.
413	 */
414	alloc_blocks = plen * 2;
415	if (alloc_blocks > MAXEXTLEN)
416		alloc_blocks = XFS_B_TO_FSB(mp, offset);
417	qblocks = alloc_blocks;
418
419	/*
420	 * MAXEXTLEN is not a power of two value but we round the prealloc down
421	 * to the nearest power of two value after throttling. To prevent the
422	 * round down from unconditionally reducing the maximum supported
423	 * prealloc size, we round up first, apply appropriate throttling,
424	 * round down and cap the value to MAXEXTLEN.
425	 */
426	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
427				       alloc_blocks);
428
429	freesp = percpu_counter_read_positive(&mp->m_fdblocks);
430	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
431		shift = 2;
432		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
433			shift++;
434		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
435			shift++;
436		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
437			shift++;
438		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
439			shift++;
440	}
441
442	/*
443	 * Check each quota to cap the prealloc size, provide a shift value to
444	 * throttle with and adjust amount of available space.
445	 */
446	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
447		xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
448					&freesp);
449	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
450		xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
451					&freesp);
452	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
453		xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
454					&freesp);
455
456	/*
457	 * The final prealloc size is set to the minimum of free space available
458	 * in each of the quotas and the overall filesystem.
459	 *
460	 * The shift throttle value is set to the maximum value as determined by
461	 * the global low free space values and per-quota low free space values.
462	 */
463	alloc_blocks = min(alloc_blocks, qblocks);
464	shift = max(shift, qshift);
465
466	if (shift)
467		alloc_blocks >>= shift;
468	/*
469	 * rounddown_pow_of_two() returns an undefined result if we pass in
470	 * alloc_blocks = 0.
471	 */
472	if (alloc_blocks)
473		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
474	if (alloc_blocks > MAXEXTLEN)
475		alloc_blocks = MAXEXTLEN;
476
477	/*
478	 * If we are still trying to allocate more space than is
479	 * available, squash the prealloc hard. This can happen if we
480	 * have a large file on a small filesystem and the above
481	 * lowspace thresholds are smaller than MAXEXTLEN.
482	 */
483	while (alloc_blocks && alloc_blocks >= freesp)
484		alloc_blocks >>= 4;
485	if (alloc_blocks < mp->m_allocsize_blocks)
486		alloc_blocks = mp->m_allocsize_blocks;
487	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
488				      mp->m_allocsize_blocks);
489	return alloc_blocks;
490}
491
492int
493xfs_iomap_write_unwritten(
494	xfs_inode_t	*ip,
495	xfs_off_t	offset,
496	xfs_off_t	count,
497	bool		update_isize)
498{
499	xfs_mount_t	*mp = ip->i_mount;
500	xfs_fileoff_t	offset_fsb;
501	xfs_filblks_t	count_fsb;
502	xfs_filblks_t	numblks_fsb;
503	int		nimaps;
504	xfs_trans_t	*tp;
505	xfs_bmbt_irec_t imap;
506	struct inode	*inode = VFS_I(ip);
507	xfs_fsize_t	i_size;
508	uint		resblks;
509	int		error;
510
511	trace_xfs_unwritten_convert(ip, offset, count);
512
513	offset_fsb = XFS_B_TO_FSBT(mp, offset);
514	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
515	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
516
517	/*
518	 * Reserve enough blocks in this transaction for two complete extent
519	 * btree splits.  We may be converting the middle part of an unwritten
520	 * extent and in this case we will insert two new extents in the btree
521	 * each of which could cause a full split.
522	 *
523	 * This reservation amount will be used in the first call to
524	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
525	 * rest of the operation.
526	 */
527	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
528
529	/* Attach dquots so that bmbt splits are accounted correctly. */
530	error = xfs_qm_dqattach(ip);
531	if (error)
532		return error;
533
534	do {
535		/*
536		 * Set up a transaction to convert the range of extents
537		 * from unwritten to real. Do allocations in a loop until
538		 * we have covered the range passed in.
539		 *
540		 * Note that we can't risk to recursing back into the filesystem
541		 * here as we might be asked to write out the same inode that we
542		 * complete here and might deadlock on the iolock.
543		 */
544		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
545				0, true, &tp);
546		if (error)
547			return error;
548
549		error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
550				XFS_IEXT_WRITE_UNWRITTEN_CNT);
551		if (error)
552			goto error_on_bmapi_transaction;
553
554		/*
555		 * Modify the unwritten extent state of the buffer.
556		 */
557		nimaps = 1;
558		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
559					XFS_BMAPI_CONVERT, resblks, &imap,
560					&nimaps);
561		if (error)
562			goto error_on_bmapi_transaction;
563
564		/*
565		 * Log the updated inode size as we go.  We have to be careful
566		 * to only log it up to the actual write offset if it is
567		 * halfway into a block.
568		 */
569		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
570		if (i_size > offset + count)
571			i_size = offset + count;
572		if (update_isize && i_size > i_size_read(inode))
573			i_size_write(inode, i_size);
574		i_size = xfs_new_eof(ip, i_size);
575		if (i_size) {
576			ip->i_disk_size = i_size;
577			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
578		}
579
580		error = xfs_trans_commit(tp);
581		xfs_iunlock(ip, XFS_ILOCK_EXCL);
582		if (error)
583			return error;
584
585		if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
586			return xfs_alert_fsblock_zero(ip, &imap);
587
588		if ((numblks_fsb = imap.br_blockcount) == 0) {
589			/*
590			 * The numblks_fsb value should always get
591			 * smaller, otherwise the loop is stuck.
592			 */
593			ASSERT(imap.br_blockcount);
594			break;
595		}
596		offset_fsb += numblks_fsb;
597		count_fsb -= numblks_fsb;
598	} while (count_fsb > 0);
599
600	return 0;
601
602error_on_bmapi_transaction:
603	xfs_trans_cancel(tp);
604	xfs_iunlock(ip, XFS_ILOCK_EXCL);
605	return error;
606}
607
608static inline bool
609imap_needs_alloc(
610	struct inode		*inode,
611	unsigned		flags,
612	struct xfs_bmbt_irec	*imap,
613	int			nimaps)
614{
615	/* don't allocate blocks when just zeroing */
616	if (flags & IOMAP_ZERO)
617		return false;
618	if (!nimaps ||
619	    imap->br_startblock == HOLESTARTBLOCK ||
620	    imap->br_startblock == DELAYSTARTBLOCK)
621		return true;
622	/* we convert unwritten extents before copying the data for DAX */
623	if (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN)
624		return true;
625	return false;
626}
627
628static inline bool
629imap_needs_cow(
630	struct xfs_inode	*ip,
631	unsigned int		flags,
632	struct xfs_bmbt_irec	*imap,
633	int			nimaps)
634{
635	if (!xfs_is_cow_inode(ip))
636		return false;
637
638	/* when zeroing we don't have to COW holes or unwritten extents */
639	if (flags & IOMAP_ZERO) {
640		if (!nimaps ||
641		    imap->br_startblock == HOLESTARTBLOCK ||
642		    imap->br_state == XFS_EXT_UNWRITTEN)
643			return false;
644	}
645
646	return true;
647}
648
649static int
650xfs_ilock_for_iomap(
651	struct xfs_inode	*ip,
652	unsigned		flags,
653	unsigned		*lockmode)
654{
655	unsigned		mode = XFS_ILOCK_SHARED;
656	bool			is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
657
658	/*
659	 * COW writes may allocate delalloc space or convert unwritten COW
660	 * extents, so we need to make sure to take the lock exclusively here.
661	 */
662	if (xfs_is_cow_inode(ip) && is_write)
663		mode = XFS_ILOCK_EXCL;
664
665	/*
666	 * Extents not yet cached requires exclusive access, don't block.  This
667	 * is an opencoded xfs_ilock_data_map_shared() call but with
668	 * non-blocking behaviour.
669	 */
670	if (xfs_need_iread_extents(&ip->i_df)) {
671		if (flags & IOMAP_NOWAIT)
672			return -EAGAIN;
673		mode = XFS_ILOCK_EXCL;
674	}
675
676relock:
677	if (flags & IOMAP_NOWAIT) {
678		if (!xfs_ilock_nowait(ip, mode))
679			return -EAGAIN;
680	} else {
681		xfs_ilock(ip, mode);
682	}
683
684	/*
685	 * The reflink iflag could have changed since the earlier unlocked
686	 * check, so if we got ILOCK_SHARED for a write and but we're now a
687	 * reflink inode we have to switch to ILOCK_EXCL and relock.
688	 */
689	if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
690		xfs_iunlock(ip, mode);
691		mode = XFS_ILOCK_EXCL;
692		goto relock;
693	}
694
695	*lockmode = mode;
696	return 0;
697}
698
699/*
700 * Check that the imap we are going to return to the caller spans the entire
701 * range that the caller requested for the IO.
702 */
703static bool
704imap_spans_range(
705	struct xfs_bmbt_irec	*imap,
706	xfs_fileoff_t		offset_fsb,
707	xfs_fileoff_t		end_fsb)
708{
709	if (imap->br_startoff > offset_fsb)
710		return false;
711	if (imap->br_startoff + imap->br_blockcount < end_fsb)
712		return false;
713	return true;
714}
715
716static int
717xfs_direct_write_iomap_begin(
718	struct inode		*inode,
719	loff_t			offset,
720	loff_t			length,
721	unsigned		flags,
722	struct iomap		*iomap,
723	struct iomap		*srcmap)
724{
725	struct xfs_inode	*ip = XFS_I(inode);
726	struct xfs_mount	*mp = ip->i_mount;
727	struct xfs_bmbt_irec	imap, cmap;
728	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
729	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
730	int			nimaps = 1, error = 0;
731	bool			shared = false;
732	u16			iomap_flags = 0;
733	unsigned		lockmode;
734
735	ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
736
737	if (xfs_is_shutdown(mp))
738		return -EIO;
739
740	/*
741	 * Writes that span EOF might trigger an IO size update on completion,
742	 * so consider them to be dirty for the purposes of O_DSYNC even if
743	 * there is no other metadata changes pending or have been made here.
744	 */
745	if (offset + length > i_size_read(inode))
746		iomap_flags |= IOMAP_F_DIRTY;
747
748	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
749	if (error)
750		return error;
751
752	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
753			       &nimaps, 0);
754	if (error)
755		goto out_unlock;
756
757	if (imap_needs_cow(ip, flags, &imap, nimaps)) {
758		error = -EAGAIN;
759		if (flags & IOMAP_NOWAIT)
760			goto out_unlock;
761
762		/* may drop and re-acquire the ilock */
763		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
764				&lockmode, flags & IOMAP_DIRECT);
765		if (error)
766			goto out_unlock;
767		if (shared)
768			goto out_found_cow;
769		end_fsb = imap.br_startoff + imap.br_blockcount;
770		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
771	}
772
773	if (imap_needs_alloc(inode, flags, &imap, nimaps))
774		goto allocate_blocks;
775
776	/*
777	 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
778	 * a single map so that we avoid partial IO failures due to the rest of
779	 * the I/O range not covered by this map triggering an EAGAIN condition
780	 * when it is subsequently mapped and aborting the I/O.
781	 */
782	if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
783		error = -EAGAIN;
784		if (!imap_spans_range(&imap, offset_fsb, end_fsb))
785			goto out_unlock;
786	}
787
788	/*
789	 * For overwrite only I/O, we cannot convert unwritten extents without
790	 * requiring sub-block zeroing.  This can only be done under an
791	 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
792	 * extent to tell the caller to try again.
793	 */
794	if (flags & IOMAP_OVERWRITE_ONLY) {
795		error = -EAGAIN;
796		if (imap.br_state != XFS_EXT_NORM &&
797	            ((offset | length) & mp->m_blockmask))
798			goto out_unlock;
799	}
800
801	xfs_iunlock(ip, lockmode);
802	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
803	return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags);
804
805allocate_blocks:
806	error = -EAGAIN;
807	if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
808		goto out_unlock;
809
810	/*
811	 * We cap the maximum length we map to a sane size  to keep the chunks
812	 * of work done where somewhat symmetric with the work writeback does.
813	 * This is a completely arbitrary number pulled out of thin air as a
814	 * best guess for initial testing.
815	 *
816	 * Note that the values needs to be less than 32-bits wide until the
817	 * lower level functions are updated.
818	 */
819	length = min_t(loff_t, length, 1024 * PAGE_SIZE);
820	end_fsb = xfs_iomap_end_fsb(mp, offset, length);
821
822	if (offset + length > XFS_ISIZE(ip))
823		end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
824	else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
825		end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
826	xfs_iunlock(ip, lockmode);
827
828	error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
829			&imap);
830	if (error)
831		return error;
832
833	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
834	return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags | IOMAP_F_NEW);
835
836out_found_cow:
837	xfs_iunlock(ip, lockmode);
838	length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
839	trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
840	if (imap.br_startblock != HOLESTARTBLOCK) {
841		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
842		if (error)
843			return error;
844	}
845	return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
846
847out_unlock:
848	if (lockmode)
849		xfs_iunlock(ip, lockmode);
850	return error;
851}
852
853const struct iomap_ops xfs_direct_write_iomap_ops = {
854	.iomap_begin		= xfs_direct_write_iomap_begin,
855};
856
857static int
858xfs_buffered_write_iomap_begin(
859	struct inode		*inode,
860	loff_t			offset,
861	loff_t			count,
862	unsigned		flags,
863	struct iomap		*iomap,
864	struct iomap		*srcmap)
865{
866	struct xfs_inode	*ip = XFS_I(inode);
867	struct xfs_mount	*mp = ip->i_mount;
868	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
869	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
870	struct xfs_bmbt_irec	imap, cmap;
871	struct xfs_iext_cursor	icur, ccur;
872	xfs_fsblock_t		prealloc_blocks = 0;
873	bool			eof = false, cow_eof = false, shared = false;
874	int			allocfork = XFS_DATA_FORK;
875	int			error = 0;
876
877	if (xfs_is_shutdown(mp))
878		return -EIO;
879
880	/* we can't use delayed allocations when using extent size hints */
881	if (xfs_get_extsz_hint(ip))
882		return xfs_direct_write_iomap_begin(inode, offset, count,
883				flags, iomap, srcmap);
884
885	ASSERT(!XFS_IS_REALTIME_INODE(ip));
886
887	xfs_ilock(ip, XFS_ILOCK_EXCL);
888
889	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
890	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
891		error = -EFSCORRUPTED;
892		goto out_unlock;
893	}
894
895	XFS_STATS_INC(mp, xs_blk_mapw);
896
897	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
898	if (error)
899		goto out_unlock;
900
901	/*
902	 * Search the data fork first to look up our source mapping.  We
903	 * always need the data fork map, as we have to return it to the
904	 * iomap code so that the higher level write code can read data in to
905	 * perform read-modify-write cycles for unaligned writes.
906	 */
907	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
908	if (eof)
909		imap.br_startoff = end_fsb; /* fake hole until the end */
910
911	/* We never need to allocate blocks for zeroing a hole. */
912	if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
913		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
914		goto out_unlock;
915	}
916
917	/*
918	 * Search the COW fork extent list even if we did not find a data fork
919	 * extent.  This serves two purposes: first this implements the
920	 * speculative preallocation using cowextsize, so that we also unshare
921	 * block adjacent to shared blocks instead of just the shared blocks
922	 * themselves.  Second the lookup in the extent list is generally faster
923	 * than going out to the shared extent tree.
924	 */
925	if (xfs_is_cow_inode(ip)) {
926		if (!ip->i_cowfp) {
927			ASSERT(!xfs_is_reflink_inode(ip));
928			xfs_ifork_init_cow(ip);
929		}
930		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
931				&ccur, &cmap);
932		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
933			trace_xfs_reflink_cow_found(ip, &cmap);
934			goto found_cow;
935		}
936	}
937
938	if (imap.br_startoff <= offset_fsb) {
939		/*
940		 * For reflink files we may need a delalloc reservation when
941		 * overwriting shared extents.   This includes zeroing of
942		 * existing extents that contain data.
943		 */
944		if (!xfs_is_cow_inode(ip) ||
945		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
946			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
947					&imap);
948			goto found_imap;
949		}
950
951		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
952
953		/* Trim the mapping to the nearest shared extent boundary. */
954		error = xfs_bmap_trim_cow(ip, &imap, &shared);
955		if (error)
956			goto out_unlock;
957
958		/* Not shared?  Just report the (potentially capped) extent. */
959		if (!shared) {
960			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
961					&imap);
962			goto found_imap;
963		}
964
965		/*
966		 * Fork all the shared blocks from our write offset until the
967		 * end of the extent.
968		 */
969		allocfork = XFS_COW_FORK;
970		end_fsb = imap.br_startoff + imap.br_blockcount;
971	} else {
972		/*
973		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
974		 * pages to keep the chunks of work done where somewhat
975		 * symmetric with the work writeback does.  This is a completely
976		 * arbitrary number pulled out of thin air.
977		 *
978		 * Note that the values needs to be less than 32-bits wide until
979		 * the lower level functions are updated.
980		 */
981		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
982		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
983
984		if (xfs_is_always_cow_inode(ip))
985			allocfork = XFS_COW_FORK;
986	}
987
988	error = xfs_qm_dqattach_locked(ip, false);
989	if (error)
990		goto out_unlock;
991
992	if (eof && offset + count > XFS_ISIZE(ip)) {
993		/*
994		 * Determine the initial size of the preallocation.
995		 * We clean up any extra preallocation when the file is closed.
996		 */
997		if (xfs_has_allocsize(mp))
998			prealloc_blocks = mp->m_allocsize_blocks;
999		else
1000			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1001						offset, count, &icur);
1002		if (prealloc_blocks) {
1003			xfs_extlen_t	align;
1004			xfs_off_t	end_offset;
1005			xfs_fileoff_t	p_end_fsb;
1006
1007			end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1008			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1009					prealloc_blocks;
1010
1011			align = xfs_eof_alignment(ip);
1012			if (align)
1013				p_end_fsb = roundup_64(p_end_fsb, align);
1014
1015			p_end_fsb = min(p_end_fsb,
1016				XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1017			ASSERT(p_end_fsb > offset_fsb);
1018			prealloc_blocks = p_end_fsb - end_fsb;
1019		}
1020	}
1021
1022retry:
1023	error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1024			end_fsb - offset_fsb, prealloc_blocks,
1025			allocfork == XFS_DATA_FORK ? &imap : &cmap,
1026			allocfork == XFS_DATA_FORK ? &icur : &ccur,
1027			allocfork == XFS_DATA_FORK ? eof : cow_eof);
1028	switch (error) {
1029	case 0:
1030		break;
1031	case -ENOSPC:
1032	case -EDQUOT:
1033		/* retry without any preallocation */
1034		trace_xfs_delalloc_enospc(ip, offset, count);
1035		if (prealloc_blocks) {
1036			prealloc_blocks = 0;
1037			goto retry;
1038		}
1039		fallthrough;
1040	default:
1041		goto out_unlock;
1042	}
1043
1044	if (allocfork == XFS_COW_FORK) {
1045		trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1046		goto found_cow;
1047	}
1048
1049	/*
1050	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1051	 * them out if the write happens to fail.
1052	 */
1053	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1054	trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1055	return xfs_bmbt_to_iomap(ip, iomap, &imap, IOMAP_F_NEW);
1056
1057found_imap:
1058	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1059	return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1060
1061found_cow:
1062	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1063	if (imap.br_startoff <= offset_fsb) {
1064		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
1065		if (error)
1066			return error;
1067		return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1068	}
1069
1070	xfs_trim_extent(&cmap, offset_fsb, imap.br_startoff - offset_fsb);
1071	return xfs_bmbt_to_iomap(ip, iomap, &cmap, 0);
1072
1073out_unlock:
1074	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1075	return error;
1076}
1077
1078static int
1079xfs_buffered_write_iomap_end(
1080	struct inode		*inode,
1081	loff_t			offset,
1082	loff_t			length,
1083	ssize_t			written,
1084	unsigned		flags,
1085	struct iomap		*iomap)
1086{
1087	struct xfs_inode	*ip = XFS_I(inode);
1088	struct xfs_mount	*mp = ip->i_mount;
1089	xfs_fileoff_t		start_fsb;
1090	xfs_fileoff_t		end_fsb;
1091	int			error = 0;
1092
1093	if (iomap->type != IOMAP_DELALLOC)
1094		return 0;
1095
1096	/*
1097	 * Behave as if the write failed if drop writes is enabled. Set the NEW
1098	 * flag to force delalloc cleanup.
1099	 */
1100	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1101		iomap->flags |= IOMAP_F_NEW;
1102		written = 0;
1103	}
1104
1105	/*
1106	 * start_fsb refers to the first unused block after a short write. If
1107	 * nothing was written, round offset down to point at the first block in
1108	 * the range.
1109	 */
1110	if (unlikely(!written))
1111		start_fsb = XFS_B_TO_FSBT(mp, offset);
1112	else
1113		start_fsb = XFS_B_TO_FSB(mp, offset + written);
1114	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1115
1116	/*
1117	 * Trim delalloc blocks if they were allocated by this write and we
1118	 * didn't manage to write the whole range.
1119	 *
1120	 * We don't need to care about racing delalloc as we hold i_mutex
1121	 * across the reserve/allocate/unreserve calls. If there are delalloc
1122	 * blocks in the range, they are ours.
1123	 */
1124	if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1125		truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1126					 XFS_FSB_TO_B(mp, end_fsb) - 1);
1127
1128		error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1129					       end_fsb - start_fsb);
1130		if (error && !xfs_is_shutdown(mp)) {
1131			xfs_alert(mp, "%s: unable to clean up ino %lld",
1132				__func__, ip->i_ino);
1133			return error;
1134		}
1135	}
1136
1137	return 0;
1138}
1139
1140const struct iomap_ops xfs_buffered_write_iomap_ops = {
1141	.iomap_begin		= xfs_buffered_write_iomap_begin,
1142	.iomap_end		= xfs_buffered_write_iomap_end,
1143};
1144
1145static int
1146xfs_read_iomap_begin(
1147	struct inode		*inode,
1148	loff_t			offset,
1149	loff_t			length,
1150	unsigned		flags,
1151	struct iomap		*iomap,
1152	struct iomap		*srcmap)
1153{
1154	struct xfs_inode	*ip = XFS_I(inode);
1155	struct xfs_mount	*mp = ip->i_mount;
1156	struct xfs_bmbt_irec	imap;
1157	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1158	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1159	int			nimaps = 1, error = 0;
1160	bool			shared = false;
1161	unsigned		lockmode;
1162
1163	ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1164
1165	if (xfs_is_shutdown(mp))
1166		return -EIO;
1167
1168	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1169	if (error)
1170		return error;
1171	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1172			       &nimaps, 0);
1173	if (!error && (flags & IOMAP_REPORT))
1174		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1175	xfs_iunlock(ip, lockmode);
1176
1177	if (error)
1178		return error;
1179	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1180	return xfs_bmbt_to_iomap(ip, iomap, &imap, shared ? IOMAP_F_SHARED : 0);
1181}
1182
1183const struct iomap_ops xfs_read_iomap_ops = {
1184	.iomap_begin		= xfs_read_iomap_begin,
1185};
1186
1187static int
1188xfs_seek_iomap_begin(
1189	struct inode		*inode,
1190	loff_t			offset,
1191	loff_t			length,
1192	unsigned		flags,
1193	struct iomap		*iomap,
1194	struct iomap		*srcmap)
1195{
1196	struct xfs_inode	*ip = XFS_I(inode);
1197	struct xfs_mount	*mp = ip->i_mount;
1198	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1199	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1200	xfs_fileoff_t		cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1201	struct xfs_iext_cursor	icur;
1202	struct xfs_bmbt_irec	imap, cmap;
1203	int			error = 0;
1204	unsigned		lockmode;
1205
1206	if (xfs_is_shutdown(mp))
1207		return -EIO;
1208
1209	lockmode = xfs_ilock_data_map_shared(ip);
1210	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1211	if (error)
1212		goto out_unlock;
1213
1214	if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1215		/*
1216		 * If we found a data extent we are done.
1217		 */
1218		if (imap.br_startoff <= offset_fsb)
1219			goto done;
1220		data_fsb = imap.br_startoff;
1221	} else {
1222		/*
1223		 * Fake a hole until the end of the file.
1224		 */
1225		data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1226	}
1227
1228	/*
1229	 * If a COW fork extent covers the hole, report it - capped to the next
1230	 * data fork extent:
1231	 */
1232	if (xfs_inode_has_cow_data(ip) &&
1233	    xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1234		cow_fsb = cmap.br_startoff;
1235	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1236		if (data_fsb < cow_fsb + cmap.br_blockcount)
1237			end_fsb = min(end_fsb, data_fsb);
1238		xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1239		error = xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1240		/*
1241		 * This is a COW extent, so we must probe the page cache
1242		 * because there could be dirty page cache being backed
1243		 * by this extent.
1244		 */
1245		iomap->type = IOMAP_UNWRITTEN;
1246		goto out_unlock;
1247	}
1248
1249	/*
1250	 * Else report a hole, capped to the next found data or COW extent.
1251	 */
1252	if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1253		imap.br_blockcount = cow_fsb - offset_fsb;
1254	else
1255		imap.br_blockcount = data_fsb - offset_fsb;
1256	imap.br_startoff = offset_fsb;
1257	imap.br_startblock = HOLESTARTBLOCK;
1258	imap.br_state = XFS_EXT_NORM;
1259done:
1260	xfs_trim_extent(&imap, offset_fsb, end_fsb);
1261	error = xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1262out_unlock:
1263	xfs_iunlock(ip, lockmode);
1264	return error;
1265}
1266
1267const struct iomap_ops xfs_seek_iomap_ops = {
1268	.iomap_begin		= xfs_seek_iomap_begin,
1269};
1270
1271static int
1272xfs_xattr_iomap_begin(
1273	struct inode		*inode,
1274	loff_t			offset,
1275	loff_t			length,
1276	unsigned		flags,
1277	struct iomap		*iomap,
1278	struct iomap		*srcmap)
1279{
1280	struct xfs_inode	*ip = XFS_I(inode);
1281	struct xfs_mount	*mp = ip->i_mount;
1282	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1283	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1284	struct xfs_bmbt_irec	imap;
1285	int			nimaps = 1, error = 0;
1286	unsigned		lockmode;
1287
1288	if (xfs_is_shutdown(mp))
1289		return -EIO;
1290
1291	lockmode = xfs_ilock_attr_map_shared(ip);
1292
1293	/* if there are no attribute fork or extents, return ENOENT */
1294	if (!XFS_IFORK_Q(ip) || !ip->i_afp->if_nextents) {
1295		error = -ENOENT;
1296		goto out_unlock;
1297	}
1298
1299	ASSERT(ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL);
1300	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1301			       &nimaps, XFS_BMAPI_ATTRFORK);
1302out_unlock:
1303	xfs_iunlock(ip, lockmode);
1304
1305	if (error)
1306		return error;
1307	ASSERT(nimaps);
1308	return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1309}
1310
1311const struct iomap_ops xfs_xattr_iomap_ops = {
1312	.iomap_begin		= xfs_xattr_iomap_begin,
1313};
1314