Linux Audio

Check our new training course

Loading...
v3.5.6
 
  1/*
  2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
 
  3 * All Rights Reserved.
  4 *
  5 * This program is free software; you can redistribute it and/or
  6 * modify it under the terms of the GNU General Public License as
  7 * published by the Free Software Foundation.
  8 *
  9 * This program is distributed in the hope that it would be useful,
 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12 * GNU General Public License for more details.
 13 *
 14 * You should have received a copy of the GNU General Public License
 15 * along with this program; if not, write the Free Software Foundation,
 16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17 */
 18#include "xfs.h"
 19#include "xfs_fs.h"
 20#include "xfs_log.h"
 21#include "xfs_trans.h"
 22#include "xfs_sb.h"
 23#include "xfs_ag.h"
 24#include "xfs_alloc.h"
 25#include "xfs_quota.h"
 26#include "xfs_mount.h"
 27#include "xfs_bmap_btree.h"
 28#include "xfs_alloc_btree.h"
 29#include "xfs_ialloc_btree.h"
 30#include "xfs_dinode.h"
 31#include "xfs_inode.h"
 32#include "xfs_inode_item.h"
 33#include "xfs_btree.h"
 
 34#include "xfs_bmap.h"
 35#include "xfs_rtalloc.h"
 
 36#include "xfs_error.h"
 37#include "xfs_itable.h"
 38#include "xfs_attr.h"
 39#include "xfs_buf_item.h"
 40#include "xfs_trans_space.h"
 41#include "xfs_utils.h"
 42#include "xfs_iomap.h"
 43#include "xfs_trace.h"
 
 
 
 
 
 
 
 44
 
 
 45
 46#define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
 47						<< mp->m_writeio_log)
 48#define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
 49
 50STATIC int
 51xfs_iomap_eof_align_last_fsb(
 52	xfs_mount_t	*mp,
 53	xfs_inode_t	*ip,
 54	xfs_extlen_t	extsize,
 55	xfs_fileoff_t	*last_fsb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 56{
 57	xfs_fileoff_t	new_last_fsb = 0;
 58	xfs_extlen_t	align = 0;
 59	int		eof, error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 60
 61	if (!XFS_IS_REALTIME_INODE(ip)) {
 62		/*
 63		 * Round up the allocation request to a stripe unit
 64		 * (m_dalign) boundary if the file size is >= stripe unit
 65		 * size, and we are allocating past the allocation eof.
 66		 *
 67		 * If mounted with the "-o swalloc" option the alignment is
 68		 * increased from the strip unit size to the stripe width.
 69		 */
 70		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
 71			align = mp->m_swidth;
 72		else if (mp->m_dalign)
 73			align = mp->m_dalign;
 74
 75		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
 76			new_last_fsb = roundup_64(*last_fsb, align);
 77	}
 78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 79	/*
 80	 * Always round up the allocation request to an extent boundary
 81	 * (when file on a real-time subvolume or has di_extsize hint).
 82	 */
 83	if (extsize) {
 84		if (new_last_fsb)
 85			align = roundup_64(new_last_fsb, extsize);
 86		else
 87			align = extsize;
 88		new_last_fsb = roundup_64(*last_fsb, align);
 89	}
 90
 91	if (new_last_fsb) {
 92		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 93		if (error)
 94			return error;
 95		if (eof)
 96			*last_fsb = new_last_fsb;
 
 97	}
 98	return 0;
 99}
100
101STATIC int
102xfs_alert_fsblock_zero(
103	xfs_inode_t	*ip,
104	xfs_bmbt_irec_t	*imap)
105{
106	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
107			"Access to block zero in inode %llu "
108			"start_block: %llx start_off: %llx "
109			"blkcnt: %llx extent-state: %x\n",
110		(unsigned long long)ip->i_ino,
111		(unsigned long long)imap->br_startblock,
112		(unsigned long long)imap->br_startoff,
113		(unsigned long long)imap->br_blockcount,
114		imap->br_state);
115	return EFSCORRUPTED;
116}
117
118int
119xfs_iomap_write_direct(
120	xfs_inode_t	*ip,
121	xfs_off_t	offset,
122	size_t		count,
123	xfs_bmbt_irec_t *imap,
124	int		nmaps)
 
125{
126	xfs_mount_t	*mp = ip->i_mount;
127	xfs_fileoff_t	offset_fsb;
128	xfs_fileoff_t	last_fsb;
129	xfs_filblks_t	count_fsb, resaligned;
130	xfs_fsblock_t	firstfsb;
131	xfs_extlen_t	extsz, temp;
132	int		nimaps;
133	int		bmapi_flag;
134	int		quota_flag;
135	int		rt;
136	xfs_trans_t	*tp;
137	xfs_bmap_free_t free_list;
138	uint		qblocks, resblks, resrtextents;
139	int		committed;
140	int		error;
141
142	error = xfs_qm_dqattach(ip, 0);
143	if (error)
144		return XFS_ERROR(error);
145
146	rt = XFS_IS_REALTIME_INODE(ip);
147	extsz = xfs_get_extsz_hint(ip);
148
149	offset_fsb = XFS_B_TO_FSBT(mp, offset);
150	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
151	if ((offset + count) > XFS_ISIZE(ip)) {
152		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
153		if (error)
154			return XFS_ERROR(error);
155	} else {
156		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
157			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
158					imap->br_blockcount +
159					imap->br_startoff);
160	}
161	count_fsb = last_fsb - offset_fsb;
162	ASSERT(count_fsb > 0);
163
164	resaligned = count_fsb;
165	if (unlikely(extsz)) {
166		if ((temp = do_mod(offset_fsb, extsz)))
167			resaligned += temp;
168		if ((temp = do_mod(resaligned, extsz)))
169			resaligned += extsz - temp;
170	}
171
172	if (unlikely(rt)) {
173		resrtextents = qblocks = resaligned;
174		resrtextents /= mp->m_sb.sb_rextsize;
175		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
176		quota_flag = XFS_QMOPT_RES_RTBLKS;
177	} else {
178		resrtextents = 0;
179		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
180		quota_flag = XFS_QMOPT_RES_REGBLKS;
181	}
182
 
 
 
 
183	/*
184	 * Allocate and setup the transaction
185	 */
186	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
187	error = xfs_trans_reserve(tp, resblks,
188			XFS_WRITE_LOG_RES(mp), resrtextents,
189			XFS_TRANS_PERM_LOG_RES,
190			XFS_WRITE_LOG_COUNT);
191	/*
192	 * Check for running out of space, note: need lock to return
 
 
193	 */
194	if (error) {
195		xfs_trans_cancel(tp, 0);
196		return XFS_ERROR(error);
 
 
 
 
197	}
198
199	xfs_ilock(ip, XFS_ILOCK_EXCL);
 
 
 
200
201	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
202	if (error)
203		goto out_trans_cancel;
204
205	xfs_trans_ijoin(tp, ip, 0);
206
207	bmapi_flag = 0;
208	if (offset < XFS_ISIZE(ip) || extsz)
209		bmapi_flag |= XFS_BMAPI_PREALLOC;
210
211	/*
212	 * From this point onwards we overwrite the imap pointer that the
213	 * caller gave to us.
214	 */
215	xfs_bmap_init(&free_list, &firstfsb);
216	nimaps = 1;
217	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
218				&firstfsb, 0, imap, &nimaps, &free_list);
219	if (error)
220		goto out_bmap_cancel;
221
222	/*
223	 * Complete the transaction
224	 */
225	error = xfs_bmap_finish(&tp, &free_list, &committed);
226	if (error)
227		goto out_bmap_cancel;
228	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
229	if (error)
230		goto out_unlock;
231
232	/*
233	 * Copy any maps to caller's array and return any error.
234	 */
235	if (nimaps == 0) {
236		error = XFS_ERROR(ENOSPC);
237		goto out_unlock;
238	}
239
240	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
 
241		error = xfs_alert_fsblock_zero(ip, imap);
 
242
243out_unlock:
 
244	xfs_iunlock(ip, XFS_ILOCK_EXCL);
245	return error;
246
247out_bmap_cancel:
248	xfs_bmap_cancel(&free_list);
249	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
250out_trans_cancel:
251	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
252	goto out_unlock;
253}
254
255/*
256 * If the caller is doing a write at the end of the file, then extend the
257 * allocation out to the file system's write iosize.  We clean up any extra
258 * space left over when the file is closed in xfs_inactive().
259 *
260 * If we find we already have delalloc preallocation beyond EOF, don't do more
261 * preallocation as it it not needed.
262 */
263STATIC int
264xfs_iomap_eof_want_preallocate(
265	xfs_mount_t	*mp,
266	xfs_inode_t	*ip,
267	xfs_off_t	offset,
268	size_t		count,
269	xfs_bmbt_irec_t *imap,
270	int		nimaps,
271	int		*prealloc)
272{
273	xfs_fileoff_t   start_fsb;
274	xfs_filblks_t   count_fsb;
275	xfs_fsblock_t	firstblock;
276	int		n, error, imaps;
277	int		found_delalloc = 0;
278
279	*prealloc = 0;
280	if (offset + count <= XFS_ISIZE(ip))
281		return 0;
282
283	/*
284	 * If there are any real blocks past eof, then don't
285	 * do any speculative allocation.
286	 */
287	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
288	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
289	while (count_fsb > 0) {
290		imaps = nimaps;
291		firstblock = NULLFSBLOCK;
292		error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
293				       0);
294		if (error)
295			return error;
296		for (n = 0; n < imaps; n++) {
297			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
298			    (imap[n].br_startblock != DELAYSTARTBLOCK))
299				return 0;
300			start_fsb += imap[n].br_blockcount;
301			count_fsb -= imap[n].br_blockcount;
302
303			if (imap[n].br_startblock == DELAYSTARTBLOCK)
304				found_delalloc = 1;
305		}
306	}
307	if (!found_delalloc)
308		*prealloc = 1;
309	return 0;
310}
311
312/*
313 * If we don't have a user specified preallocation size, dynamically increase
314 * the preallocation size as the size of the file grows. Cap the maximum size
315 * at a single extent or less if the filesystem is near full. The closer the
316 * filesystem is to full, the smaller the maximum prealocation.
317 */
318STATIC xfs_fsblock_t
319xfs_iomap_prealloc_size(
320	struct xfs_mount	*mp,
321	struct xfs_inode	*ip)
322{
323	xfs_fsblock_t		alloc_blocks = 0;
324
325	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
326		int shift = 0;
327		int64_t freesp;
328
329		/*
330		 * rounddown_pow_of_two() returns an undefined result
331		 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
332		 * ensure we always pass in a non-zero value.
333		 */
334		alloc_blocks = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)) + 1;
335		alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
336					rounddown_pow_of_two(alloc_blocks));
337
338		xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
339		freesp = mp->m_sb.sb_fdblocks;
340		if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
341			shift = 2;
342			if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
343				shift++;
344			if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
345				shift++;
346			if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
347				shift++;
348			if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
349				shift++;
350		}
351		if (shift)
352			alloc_blocks >>= shift;
353	}
354
355	if (alloc_blocks < mp->m_writeio_blocks)
356		alloc_blocks = mp->m_writeio_blocks;
 
 
 
 
 
357
358	return alloc_blocks;
359}
360
361int
362xfs_iomap_write_delay(
363	xfs_inode_t	*ip,
364	xfs_off_t	offset,
365	size_t		count,
366	xfs_bmbt_irec_t *ret_imap)
 
367{
368	xfs_mount_t	*mp = ip->i_mount;
369	xfs_fileoff_t	offset_fsb;
370	xfs_fileoff_t	last_fsb;
371	xfs_off_t	aligned_offset;
372	xfs_fileoff_t	ioalign;
373	xfs_extlen_t	extsz;
374	int		nimaps;
375	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
376	int		prealloc, flushed = 0;
377	int		error;
378
379	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
380
381	/*
382	 * Make sure that the dquots are there. This doesn't hold
383	 * the ilock across a disk read.
384	 */
385	error = xfs_qm_dqattach_locked(ip, 0);
386	if (error)
387		return XFS_ERROR(error);
388
389	extsz = xfs_get_extsz_hint(ip);
390	offset_fsb = XFS_B_TO_FSBT(mp, offset);
391
392
393	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
394				imap, XFS_WRITE_IMAPS, &prealloc);
395	if (error)
396		return error;
397
398retry:
399	if (prealloc) {
400		xfs_fsblock_t	alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
401
402		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
403		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
404		last_fsb = ioalign + alloc_blocks;
405	} else {
406		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
 
407	}
408
409	if (prealloc || extsz) {
410		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
411		if (error)
412			return error;
 
413	}
414
415	/*
416	 * Make sure preallocation does not create extents beyond the range we
417	 * actually support in this filesystem.
418	 */
419	if (last_fsb > XFS_B_TO_FSB(mp, mp->m_maxioffset))
420		last_fsb = XFS_B_TO_FSB(mp, mp->m_maxioffset);
421
422	ASSERT(last_fsb > offset_fsb);
423
424	nimaps = XFS_WRITE_IMAPS;
425	error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
426				imap, &nimaps, XFS_BMAPI_ENTIRE);
427	switch (error) {
428	case 0:
429	case ENOSPC:
430	case EDQUOT:
431		break;
432	default:
433		return XFS_ERROR(error);
434	}
435
436	/*
437	 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT.  For
438	 * ENOSPC, * flush all other inodes with delalloc blocks to free up
439	 * some of the excess reserved metadata space. For both cases, retry
440	 * without EOF preallocation.
441	 */
442	if (nimaps == 0) {
443		trace_xfs_delalloc_enospc(ip, offset, count);
444		if (flushed)
445			return XFS_ERROR(error ? error : ENOSPC);
446
447		if (error == ENOSPC) {
448			xfs_iunlock(ip, XFS_ILOCK_EXCL);
449			xfs_flush_inodes(ip);
450			xfs_ilock(ip, XFS_ILOCK_EXCL);
451		}
452
453		flushed = 1;
454		error = 0;
455		prealloc = 0;
456		goto retry;
457	}
 
458
459	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
460		return xfs_alert_fsblock_zero(ip, &imap[0]);
 
 
 
 
 
461
462	*ret_imap = imap[0];
463	return 0;
 
 
 
 
 
 
 
 
 
 
 
464}
465
466/*
467 * Pass in a delayed allocate extent, convert it to real extents;
468 * return to the caller the extent we create which maps on top of
469 * the originating callers request.
470 *
471 * Called without a lock on the inode.
472 *
473 * We no longer bother to look at the incoming map - all we have to
474 * guarantee is that whatever we allocate fills the required range.
475 */
476int
477xfs_iomap_write_allocate(
478	xfs_inode_t	*ip,
479	xfs_off_t	offset,
480	size_t		count,
481	xfs_bmbt_irec_t *imap)
 
482{
483	xfs_mount_t	*mp = ip->i_mount;
484	xfs_fileoff_t	offset_fsb, last_block;
485	xfs_fileoff_t	end_fsb, map_start_fsb;
486	xfs_fsblock_t	first_block;
487	xfs_bmap_free_t	free_list;
488	xfs_filblks_t	count_fsb;
489	xfs_trans_t	*tp;
490	int		nimaps, committed;
491	int		error = 0;
492	int		nres;
 
493
494	/*
495	 * Make sure that the dquots are there.
 
 
 
496	 */
497	error = xfs_qm_dqattach(ip, 0);
498	if (error)
499		return XFS_ERROR(error);
500
501	offset_fsb = XFS_B_TO_FSBT(mp, offset);
502	count_fsb = imap->br_blockcount;
503	map_start_fsb = imap->br_startoff;
504
505	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
506
507	while (count_fsb != 0) {
508		/*
509		 * Set up a transaction with which to allocate the
510		 * backing store for the file.  Do allocations in a
511		 * loop until we get some space in the range we are
512		 * interested in.  The other space that might be allocated
513		 * is in the delayed allocation extent on which we sit
514		 * but before our buffer starts.
515		 */
516
517		nimaps = 0;
518		while (nimaps == 0) {
519			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
520			tp->t_flags |= XFS_TRANS_RESERVE;
521			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
522			error = xfs_trans_reserve(tp, nres,
523					XFS_WRITE_LOG_RES(mp),
524					0, XFS_TRANS_PERM_LOG_RES,
525					XFS_WRITE_LOG_COUNT);
526			if (error) {
527				xfs_trans_cancel(tp, 0);
528				return XFS_ERROR(error);
529			}
530			xfs_ilock(ip, XFS_ILOCK_EXCL);
531			xfs_trans_ijoin(tp, ip, 0);
532
533			xfs_bmap_init(&free_list, &first_block);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
534
535			/*
536			 * it is possible that the extents have changed since
537			 * we did the read call as we dropped the ilock for a
538			 * while. We have to be careful about truncates or hole
539			 * punchs here - we are not allowed to allocate
540			 * non-delalloc blocks here.
541			 *
542			 * The only protection against truncation is the pages
543			 * for the range we are being asked to convert are
544			 * locked and hence a truncate will block on them
545			 * first.
546			 *
547			 * As a result, if we go beyond the range we really
548			 * need and hit an delalloc extent boundary followed by
549			 * a hole while we have excess blocks in the map, we
550			 * will fill the hole incorrectly and overrun the
551			 * transaction reservation.
552			 *
553			 * Using a single map prevents this as we are forced to
554			 * check each map we look for overlap with the desired
555			 * range and abort as soon as we find it. Also, given
556			 * that we only return a single map, having one beyond
557			 * what we can return is probably a bit silly.
558			 *
559			 * We also need to check that we don't go beyond EOF;
560			 * this is a truncate optimisation as a truncate sets
561			 * the new file size before block on the pages we
562			 * currently have locked under writeback. Because they
563			 * are about to be tossed, we don't need to write them
564			 * back....
565			 */
566			nimaps = 1;
567			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
568			error = xfs_bmap_last_offset(NULL, ip, &last_block,
569							XFS_DATA_FORK);
570			if (error)
571				goto trans_cancel;
572
573			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
574			if ((map_start_fsb + count_fsb) > last_block) {
575				count_fsb = last_block - map_start_fsb;
576				if (count_fsb == 0) {
577					error = EAGAIN;
578					goto trans_cancel;
579				}
580			}
581
582			/*
583			 * From this point onwards we overwrite the imap
584			 * pointer that the caller gave to us.
585			 */
586			error = xfs_bmapi_write(tp, ip, map_start_fsb,
587						count_fsb, 0, &first_block, 1,
588						imap, &nimaps, &free_list);
589			if (error)
590				goto trans_cancel;
591
592			error = xfs_bmap_finish(&tp, &free_list, &committed);
593			if (error)
594				goto trans_cancel;
595
596			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
597			if (error)
598				goto error0;
599
600			xfs_iunlock(ip, XFS_ILOCK_EXCL);
601		}
 
 
 
 
 
602
603		/*
604		 * See if we were able to allocate an extent that
605		 * covers at least part of the callers request
606		 */
607		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
608			return xfs_alert_fsblock_zero(ip, imap);
 
 
 
 
 
 
 
609
610		if ((offset_fsb >= imap->br_startoff) &&
611		    (offset_fsb < (imap->br_startoff +
612				   imap->br_blockcount))) {
613			XFS_STATS_INC(xs_xstrat_quick);
614			return 0;
615		}
 
 
 
616
617		/*
618		 * So far we have not mapped the requested part of the
619		 * file, just surrounding data, try again.
620		 */
621		count_fsb -= imap->br_blockcount;
622		map_start_fsb = imap->br_startoff + imap->br_blockcount;
623	}
 
 
 
624
625trans_cancel:
626	xfs_bmap_cancel(&free_list);
627	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
628error0:
629	xfs_iunlock(ip, XFS_ILOCK_EXCL);
630	return XFS_ERROR(error);
 
 
 
 
 
 
 
631}
632
633int
634xfs_iomap_write_unwritten(
635	xfs_inode_t	*ip,
636	xfs_off_t	offset,
637	size_t		count)
 
638{
639	xfs_mount_t	*mp = ip->i_mount;
640	xfs_fileoff_t	offset_fsb;
641	xfs_filblks_t	count_fsb;
642	xfs_filblks_t	numblks_fsb;
643	xfs_fsblock_t	firstfsb;
644	int		nimaps;
645	xfs_trans_t	*tp;
646	xfs_bmbt_irec_t imap;
647	xfs_bmap_free_t free_list;
648	xfs_fsize_t	i_size;
649	uint		resblks;
650	int		committed;
651	int		error;
652
653	trace_xfs_unwritten_convert(ip, offset, count);
654
655	offset_fsb = XFS_B_TO_FSBT(mp, offset);
656	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
657	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
658
659	/*
660	 * Reserve enough blocks in this transaction for two complete extent
661	 * btree splits.  We may be converting the middle part of an unwritten
662	 * extent and in this case we will insert two new extents in the btree
663	 * each of which could cause a full split.
664	 *
665	 * This reservation amount will be used in the first call to
666	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
667	 * rest of the operation.
668	 */
669	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
670
 
 
 
 
 
671	do {
672		/*
673		 * set up a transaction to convert the range of extents
674		 * from unwritten to real. Do allocations in a loop until
675		 * we have covered the range passed in.
676		 *
677		 * Note that we open code the transaction allocation here
678		 * to pass KM_NOFS--we can't risk to recursing back into
679		 * the filesystem here as we might be asked to write out
680		 * the same inode that we complete here and might deadlock
681		 * on the iolock.
682		 */
683		xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
684		tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
685		tp->t_flags |= XFS_TRANS_RESERVE;
686		error = xfs_trans_reserve(tp, resblks,
687				XFS_WRITE_LOG_RES(mp), 0,
688				XFS_TRANS_PERM_LOG_RES,
689				XFS_WRITE_LOG_COUNT);
690		if (error) {
691			xfs_trans_cancel(tp, 0);
692			return XFS_ERROR(error);
693		}
694
695		xfs_ilock(ip, XFS_ILOCK_EXCL);
696		xfs_trans_ijoin(tp, ip, 0);
 
 
697
698		/*
699		 * Modify the unwritten extent state of the buffer.
700		 */
701		xfs_bmap_init(&free_list, &firstfsb);
702		nimaps = 1;
703		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
704				  XFS_BMAPI_CONVERT, &firstfsb,
705				  1, &imap, &nimaps, &free_list);
706		if (error)
707			goto error_on_bmapi_transaction;
708
709		/*
710		 * Log the updated inode size as we go.  We have to be careful
711		 * to only log it up to the actual write offset if it is
712		 * halfway into a block.
713		 */
714		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
715		if (i_size > offset + count)
716			i_size = offset + count;
717
 
718		i_size = xfs_new_eof(ip, i_size);
719		if (i_size) {
720			ip->i_d.di_size = i_size;
721			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
722		}
723
724		error = xfs_bmap_finish(&tp, &free_list, &committed);
725		if (error)
726			goto error_on_bmapi_transaction;
727
728		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
729		xfs_iunlock(ip, XFS_ILOCK_EXCL);
730		if (error)
731			return XFS_ERROR(error);
732
733		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
 
734			return xfs_alert_fsblock_zero(ip, &imap);
 
735
736		if ((numblks_fsb = imap.br_blockcount) == 0) {
737			/*
738			 * The numblks_fsb value should always get
739			 * smaller, otherwise the loop is stuck.
740			 */
741			ASSERT(imap.br_blockcount);
742			break;
743		}
744		offset_fsb += numblks_fsb;
745		count_fsb -= numblks_fsb;
746	} while (count_fsb > 0);
747
748	return 0;
749
750error_on_bmapi_transaction:
751	xfs_bmap_cancel(&free_list);
752	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
753	xfs_iunlock(ip, XFS_ILOCK_EXCL);
754	return XFS_ERROR(error);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
755}
v6.13.7
   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_rtgroup.h"
  28#include "xfs_dquot_item.h"
  29#include "xfs_dquot.h"
  30#include "xfs_reflink.h"
  31#include "xfs_health.h"
  32#include "xfs_rtbitmap.h"
  33
  34#define XFS_ALLOC_ALIGN(mp, off) \
  35	(((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
  36
  37static int
  38xfs_alert_fsblock_zero(
 
 
 
 
 
  39	xfs_inode_t	*ip,
  40	xfs_bmbt_irec_t	*imap)
  41{
  42	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
  43			"Access to block zero in inode %llu "
  44			"start_block: %llx start_off: %llx "
  45			"blkcnt: %llx extent-state: %x",
  46		(unsigned long long)ip->i_ino,
  47		(unsigned long long)imap->br_startblock,
  48		(unsigned long long)imap->br_startoff,
  49		(unsigned long long)imap->br_blockcount,
  50		imap->br_state);
  51	xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
  52	return -EFSCORRUPTED;
  53}
  54
  55u64
  56xfs_iomap_inode_sequence(
  57	struct xfs_inode	*ip,
  58	u16			iomap_flags)
  59{
  60	u64			cookie = 0;
  61
  62	if (iomap_flags & IOMAP_F_XATTR)
  63		return READ_ONCE(ip->i_af.if_seq);
  64	if ((iomap_flags & IOMAP_F_SHARED) && ip->i_cowfp)
  65		cookie = (u64)READ_ONCE(ip->i_cowfp->if_seq) << 32;
  66	return cookie | READ_ONCE(ip->i_df.if_seq);
  67}
  68
  69/*
  70 * Check that the iomap passed to us is still valid for the given offset and
  71 * length.
  72 */
  73static bool
  74xfs_iomap_valid(
  75	struct inode		*inode,
  76	const struct iomap	*iomap)
  77{
  78	struct xfs_inode	*ip = XFS_I(inode);
  79
  80	if (iomap->validity_cookie !=
  81			xfs_iomap_inode_sequence(ip, iomap->flags)) {
  82		trace_xfs_iomap_invalid(ip, iomap);
  83		return false;
  84	}
  85
  86	XFS_ERRORTAG_DELAY(ip->i_mount, XFS_ERRTAG_WRITE_DELAY_MS);
  87	return true;
  88}
  89
  90static const struct iomap_folio_ops xfs_iomap_folio_ops = {
  91	.iomap_valid		= xfs_iomap_valid,
  92};
  93
  94int
  95xfs_bmbt_to_iomap(
  96	struct xfs_inode	*ip,
  97	struct iomap		*iomap,
  98	struct xfs_bmbt_irec	*imap,
  99	unsigned int		mapping_flags,
 100	u16			iomap_flags,
 101	u64			sequence_cookie)
 102{
 103	struct xfs_mount	*mp = ip->i_mount;
 104	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
 105
 106	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
 107		xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
 108		return xfs_alert_fsblock_zero(ip, imap);
 109	}
 110
 111	if (imap->br_startblock == HOLESTARTBLOCK) {
 112		iomap->addr = IOMAP_NULL_ADDR;
 113		iomap->type = IOMAP_HOLE;
 114	} else if (imap->br_startblock == DELAYSTARTBLOCK ||
 115		   isnullstartblock(imap->br_startblock)) {
 116		iomap->addr = IOMAP_NULL_ADDR;
 117		iomap->type = IOMAP_DELALLOC;
 118	} else {
 119		xfs_daddr_t	daddr = xfs_fsb_to_db(ip, imap->br_startblock);
 120
 121		iomap->addr = BBTOB(daddr);
 122		if (mapping_flags & IOMAP_DAX)
 123			iomap->addr += target->bt_dax_part_off;
 124
 125		if (imap->br_state == XFS_EXT_UNWRITTEN)
 126			iomap->type = IOMAP_UNWRITTEN;
 127		else
 128			iomap->type = IOMAP_MAPPED;
 129
 130		/*
 131		 * Mark iomaps starting at the first sector of a RTG as merge
 132		 * boundary so that each I/O completions is contained to a
 133		 * single RTG.
 134		 */
 135		if (XFS_IS_REALTIME_INODE(ip) && xfs_has_rtgroups(mp) &&
 136		    xfs_rtbno_is_group_start(mp, imap->br_startblock))
 137			iomap->flags |= IOMAP_F_BOUNDARY;
 138	}
 139	iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
 140	iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
 141	if (mapping_flags & IOMAP_DAX)
 142		iomap->dax_dev = target->bt_daxdev;
 143	else
 144		iomap->bdev = target->bt_bdev;
 145	iomap->flags = iomap_flags;
 146
 147	if (xfs_ipincount(ip) &&
 148	    (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
 149		iomap->flags |= IOMAP_F_DIRTY;
 150
 151	iomap->validity_cookie = sequence_cookie;
 152	iomap->folio_ops = &xfs_iomap_folio_ops;
 153	return 0;
 154}
 155
 156static void
 157xfs_hole_to_iomap(
 158	struct xfs_inode	*ip,
 159	struct iomap		*iomap,
 160	xfs_fileoff_t		offset_fsb,
 161	xfs_fileoff_t		end_fsb)
 162{
 163	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
 164
 165	iomap->addr = IOMAP_NULL_ADDR;
 166	iomap->type = IOMAP_HOLE;
 167	iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
 168	iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
 169	iomap->bdev = target->bt_bdev;
 170	iomap->dax_dev = target->bt_daxdev;
 171}
 172
 173static inline xfs_fileoff_t
 174xfs_iomap_end_fsb(
 175	struct xfs_mount	*mp,
 176	loff_t			offset,
 177	loff_t			count)
 178{
 179	ASSERT(offset <= mp->m_super->s_maxbytes);
 180	return min(XFS_B_TO_FSB(mp, offset + count),
 181		   XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
 182}
 183
 184static xfs_extlen_t
 185xfs_eof_alignment(
 186	struct xfs_inode	*ip)
 187{
 188	struct xfs_mount	*mp = ip->i_mount;
 189	xfs_extlen_t		align = 0;
 190
 191	if (!XFS_IS_REALTIME_INODE(ip)) {
 192		/*
 193		 * Round up the allocation request to a stripe unit
 194		 * (m_dalign) boundary if the file size is >= stripe unit
 195		 * size, and we are allocating past the allocation eof.
 196		 *
 197		 * If mounted with the "-o swalloc" option the alignment is
 198		 * increased from the strip unit size to the stripe width.
 199		 */
 200		if (mp->m_swidth && xfs_has_swalloc(mp))
 201			align = mp->m_swidth;
 202		else if (mp->m_dalign)
 203			align = mp->m_dalign;
 204
 205		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
 206			align = 0;
 207	}
 208
 209	return align;
 210}
 211
 212/*
 213 * Check if last_fsb is outside the last extent, and if so grow it to the next
 214 * stripe unit boundary.
 215 */
 216xfs_fileoff_t
 217xfs_iomap_eof_align_last_fsb(
 218	struct xfs_inode	*ip,
 219	xfs_fileoff_t		end_fsb)
 220{
 221	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
 222	xfs_extlen_t		extsz = xfs_get_extsz_hint(ip);
 223	xfs_extlen_t		align = xfs_eof_alignment(ip);
 224	struct xfs_bmbt_irec	irec;
 225	struct xfs_iext_cursor	icur;
 226
 227	ASSERT(!xfs_need_iread_extents(ifp));
 228
 229	/*
 230	 * Always round up the allocation request to the extent hint boundary.
 
 231	 */
 232	if (extsz) {
 233		if (align)
 234			align = roundup_64(align, extsz);
 235		else
 236			align = extsz;
 
 237	}
 238
 239	if (align) {
 240		xfs_fileoff_t	aligned_end_fsb = roundup_64(end_fsb, align);
 241
 242		xfs_iext_last(ifp, &icur);
 243		if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
 244		    aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
 245			return aligned_end_fsb;
 246	}
 
 
 247
 248	return end_fsb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 249}
 250
 251int
 252xfs_iomap_write_direct(
 253	struct xfs_inode	*ip,
 254	xfs_fileoff_t		offset_fsb,
 255	xfs_fileoff_t		count_fsb,
 256	unsigned int		flags,
 257	struct xfs_bmbt_irec	*imap,
 258	u64			*seq)
 259{
 260	struct xfs_mount	*mp = ip->i_mount;
 261	struct xfs_trans	*tp;
 262	xfs_filblks_t		resaligned;
 263	int			nimaps;
 264	unsigned int		dblocks, rblocks;
 265	bool			force = false;
 266	int			error;
 267	int			bmapi_flags = XFS_BMAPI_PREALLOC;
 268	int			nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
 
 
 
 
 
 
 269
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270	ASSERT(count_fsb > 0);
 271
 272	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
 273					   xfs_get_extsz_hint(ip));
 274	if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
 275		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 276		rblocks = resaligned;
 
 
 
 
 
 
 
 
 277	} else {
 278		dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 279		rblocks = 0;
 
 280	}
 281
 282	error = xfs_qm_dqattach(ip);
 283	if (error)
 284		return error;
 285
 286	/*
 287	 * For DAX, we do not allocate unwritten extents, but instead we zero
 288	 * the block before we commit the transaction.  Ideally we'd like to do
 289	 * this outside the transaction context, but if we commit and then crash
 290	 * we may not have zeroed the blocks and this will be exposed on
 291	 * recovery of the allocation. Hence we must zero before commit.
 292	 *
 293	 * Further, if we are mapping unwritten extents here, we need to zero
 294	 * and convert them to written so that we don't need an unwritten extent
 295	 * callback for DAX. This also means that we need to be able to dip into
 296	 * the reserve block pool for bmbt block allocation if there is no space
 297	 * left but we need to do unwritten extent conversion.
 298	 */
 299	if (flags & IOMAP_DAX) {
 300		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
 301		if (imap->br_state == XFS_EXT_UNWRITTEN) {
 302			force = true;
 303			nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
 304			dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 305		}
 306	}
 307
 308	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
 309			rblocks, force, &tp);
 310	if (error)
 311		return error;
 312
 313	error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK, nr_exts);
 314	if (error)
 315		goto out_trans_cancel;
 316
 
 
 
 
 
 
 317	/*
 318	 * From this point onwards we overwrite the imap pointer that the
 319	 * caller gave to us.
 320	 */
 
 321	nimaps = 1;
 322	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
 323				imap, &nimaps);
 324	if (error)
 325		goto out_trans_cancel;
 326
 327	/*
 328	 * Complete the transaction
 329	 */
 330	error = xfs_trans_commit(tp);
 331	if (error)
 
 
 
 
 
 
 
 
 
 
 332		goto out_unlock;
 
 333
 334	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
 335		xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
 336		error = xfs_alert_fsblock_zero(ip, imap);
 337	}
 338
 339out_unlock:
 340	*seq = xfs_iomap_inode_sequence(ip, 0);
 341	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 342	return error;
 343
 
 
 
 344out_trans_cancel:
 345	xfs_trans_cancel(tp);
 346	goto out_unlock;
 347}
 348
 349STATIC bool
 350xfs_quota_need_throttle(
 351	struct xfs_inode	*ip,
 352	xfs_dqtype_t		type,
 353	xfs_fsblock_t		alloc_blocks)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 354{
 355	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
 356	struct xfs_dquot_res	*res;
 357	struct xfs_dquot_pre	*pre;
 358
 359	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
 360		return false;
 361
 362	if (XFS_IS_REALTIME_INODE(ip)) {
 363		res = &dq->q_rtb;
 364		pre = &dq->q_rtb_prealloc;
 365	} else {
 366		res = &dq->q_blk;
 367		pre = &dq->q_blk_prealloc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 368	}
 369
 370	/* no hi watermark, no throttle */
 371	if (!pre->q_prealloc_hi_wmark)
 372		return false;
 373
 374	/* under the lo watermark, no throttle */
 375	if (res->reserved + alloc_blocks < pre->q_prealloc_lo_wmark)
 376		return false;
 377
 378	return true;
 379}
 380
 381STATIC void
 382xfs_quota_calc_throttle(
 383	struct xfs_inode	*ip,
 384	xfs_dqtype_t		type,
 385	xfs_fsblock_t		*qblocks,
 386	int			*qshift,
 387	int64_t			*qfreesp)
 388{
 389	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
 390	struct xfs_dquot_res	*res;
 391	struct xfs_dquot_pre	*pre;
 392	int64_t			freesp;
 393	int			shift = 0;
 394
 395	if (!dq) {
 396		res = NULL;
 397		pre = NULL;
 398	} else if (XFS_IS_REALTIME_INODE(ip)) {
 399		res = &dq->q_rtb;
 400		pre = &dq->q_rtb_prealloc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 401	} else {
 402		res = &dq->q_blk;
 403		pre = &dq->q_blk_prealloc;
 404	}
 405
 406	/* no dq, or over hi wmark, squash the prealloc completely */
 407	if (!res || res->reserved >= pre->q_prealloc_hi_wmark) {
 408		*qblocks = 0;
 409		*qfreesp = 0;
 410		return;
 411	}
 412
 413	freesp = pre->q_prealloc_hi_wmark - res->reserved;
 414	if (freesp < pre->q_low_space[XFS_QLOWSP_5_PCNT]) {
 415		shift = 2;
 416		if (freesp < pre->q_low_space[XFS_QLOWSP_3_PCNT])
 417			shift += 2;
 418		if (freesp < pre->q_low_space[XFS_QLOWSP_1_PCNT])
 419			shift += 2;
 
 
 
 
 
 
 
 
 
 
 
 
 420	}
 421
 422	if (freesp < *qfreesp)
 423		*qfreesp = freesp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 424
 425	/* only overwrite the throttle values if we are more aggressive */
 426	if ((freesp >> shift) < (*qblocks >> *qshift)) {
 427		*qblocks = freesp;
 428		*qshift = shift;
 429	}
 430}
 431
 432static int64_t
 433xfs_iomap_freesp(
 434	struct percpu_counter	*counter,
 435	uint64_t		low_space[XFS_LOWSP_MAX],
 436	int			*shift)
 437{
 438	int64_t			freesp;
 439
 440	freesp = percpu_counter_read_positive(counter);
 441	if (freesp < low_space[XFS_LOWSP_5_PCNT]) {
 442		*shift = 2;
 443		if (freesp < low_space[XFS_LOWSP_4_PCNT])
 444			(*shift)++;
 445		if (freesp < low_space[XFS_LOWSP_3_PCNT])
 446			(*shift)++;
 447		if (freesp < low_space[XFS_LOWSP_2_PCNT])
 448			(*shift)++;
 449		if (freesp < low_space[XFS_LOWSP_1_PCNT])
 450			(*shift)++;
 451	}
 452	return freesp;
 453}
 454
 455/*
 456 * If we don't have a user specified preallocation size, dynamically increase
 457 * the preallocation size as the size of the file grows.  Cap the maximum size
 458 * at a single extent or less if the filesystem is near full. The closer the
 459 * filesystem is to being full, the smaller the maximum preallocation.
 
 
 
 
 460 */
 461STATIC xfs_fsblock_t
 462xfs_iomap_prealloc_size(
 463	struct xfs_inode	*ip,
 464	int			whichfork,
 465	loff_t			offset,
 466	loff_t			count,
 467	struct xfs_iext_cursor	*icur)
 468{
 469	struct xfs_iext_cursor	ncur = *icur;
 470	struct xfs_bmbt_irec	prev, got;
 471	struct xfs_mount	*mp = ip->i_mount;
 472	struct xfs_ifork	*ifp = xfs_ifork_ptr(ip, whichfork);
 473	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 474	int64_t			freesp;
 475	xfs_fsblock_t		qblocks;
 476	xfs_fsblock_t		alloc_blocks = 0;
 477	xfs_extlen_t		plen;
 478	int			shift = 0;
 479	int			qshift = 0;
 480
 481	/*
 482	 * As an exception we don't do any preallocation at all if the file is
 483	 * smaller than the minimum preallocation and we are using the default
 484	 * dynamic preallocation scheme, as it is likely this is the only write
 485	 * to the file that is going to be done.
 486	 */
 487	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
 488		return 0;
 
 
 
 
 
 
 
 489
 490	/*
 491	 * Use the minimum preallocation size for small files or if we are
 492	 * writing right after a hole.
 493	 */
 494	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
 495	    !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
 496	    prev.br_startoff + prev.br_blockcount < offset_fsb)
 497		return mp->m_allocsize_blocks;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 498
 499	/*
 500	 * Take the size of the preceding data extents as the basis for the
 501	 * preallocation size. Note that we don't care if the previous extents
 502	 * are written or not.
 503	 */
 504	plen = prev.br_blockcount;
 505	while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
 506		if (plen > XFS_MAX_BMBT_EXTLEN / 2 ||
 507		    isnullstartblock(got.br_startblock) ||
 508		    got.br_startoff + got.br_blockcount != prev.br_startoff ||
 509		    got.br_startblock + got.br_blockcount != prev.br_startblock)
 510			break;
 511		plen += got.br_blockcount;
 512		prev = got;
 513	}
 514
 515	/*
 516	 * If the size of the extents is greater than half the maximum extent
 517	 * length, then use the current offset as the basis.  This ensures that
 518	 * for large files the preallocation size always extends to
 519	 * XFS_BMBT_MAX_EXTLEN rather than falling short due to things like stripe
 520	 * unit/width alignment of real extents.
 521	 */
 522	alloc_blocks = plen * 2;
 523	if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
 524		alloc_blocks = XFS_B_TO_FSB(mp, offset);
 525	qblocks = alloc_blocks;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 526
 527	/*
 528	 * XFS_BMBT_MAX_EXTLEN is not a power of two value but we round the prealloc
 529	 * down to the nearest power of two value after throttling. To prevent
 530	 * the round down from unconditionally reducing the maximum supported
 531	 * prealloc size, we round up first, apply appropriate throttling, round
 532	 * down and cap the value to XFS_BMBT_MAX_EXTLEN.
 533	 */
 534	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(XFS_MAX_BMBT_EXTLEN),
 535				       alloc_blocks);
 
 
 
 
 
 
 
 
 536
 537	if (unlikely(XFS_IS_REALTIME_INODE(ip)))
 538		freesp = xfs_rtbxlen_to_blen(mp,
 539				xfs_iomap_freesp(&mp->m_frextents,
 540					mp->m_low_rtexts, &shift));
 541	else
 542		freesp = xfs_iomap_freesp(&mp->m_fdblocks, mp->m_low_space,
 543				&shift);
 544
 545	/*
 546	 * Check each quota to cap the prealloc size, provide a shift value to
 547	 * throttle with and adjust amount of available space.
 548	 */
 549	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
 550		xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
 551					&freesp);
 552	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
 553		xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
 554					&freesp);
 555	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
 556		xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
 557					&freesp);
 558
 559	/*
 560	 * The final prealloc size is set to the minimum of free space available
 561	 * in each of the quotas and the overall filesystem.
 562	 *
 563	 * The shift throttle value is set to the maximum value as determined by
 564	 * the global low free space values and per-quota low free space values.
 565	 */
 566	alloc_blocks = min(alloc_blocks, qblocks);
 567	shift = max(shift, qshift);
 568
 569	if (shift)
 570		alloc_blocks >>= shift;
 571	/*
 572	 * rounddown_pow_of_two() returns an undefined result if we pass in
 573	 * alloc_blocks = 0.
 574	 */
 575	if (alloc_blocks)
 576		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
 577	if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
 578		alloc_blocks = XFS_MAX_BMBT_EXTLEN;
 579
 580	/*
 581	 * If we are still trying to allocate more space than is
 582	 * available, squash the prealloc hard. This can happen if we
 583	 * have a large file on a small filesystem and the above
 584	 * lowspace thresholds are smaller than XFS_BMBT_MAX_EXTLEN.
 585	 */
 586	while (alloc_blocks && alloc_blocks >= freesp)
 587		alloc_blocks >>= 4;
 588	if (alloc_blocks < mp->m_allocsize_blocks)
 589		alloc_blocks = mp->m_allocsize_blocks;
 590	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
 591				      mp->m_allocsize_blocks);
 592	return alloc_blocks;
 593}
 594
 595int
 596xfs_iomap_write_unwritten(
 597	xfs_inode_t	*ip,
 598	xfs_off_t	offset,
 599	xfs_off_t	count,
 600	bool		update_isize)
 601{
 602	xfs_mount_t	*mp = ip->i_mount;
 603	xfs_fileoff_t	offset_fsb;
 604	xfs_filblks_t	count_fsb;
 605	xfs_filblks_t	numblks_fsb;
 
 606	int		nimaps;
 607	xfs_trans_t	*tp;
 608	xfs_bmbt_irec_t imap;
 609	struct inode	*inode = VFS_I(ip);
 610	xfs_fsize_t	i_size;
 611	uint		resblks;
 
 612	int		error;
 613
 614	trace_xfs_unwritten_convert(ip, offset, count);
 615
 616	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 617	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 618	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 619
 620	/*
 621	 * Reserve enough blocks in this transaction for two complete extent
 622	 * btree splits.  We may be converting the middle part of an unwritten
 623	 * extent and in this case we will insert two new extents in the btree
 624	 * each of which could cause a full split.
 625	 *
 626	 * This reservation amount will be used in the first call to
 627	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
 628	 * rest of the operation.
 629	 */
 630	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 631
 632	/* Attach dquots so that bmbt splits are accounted correctly. */
 633	error = xfs_qm_dqattach(ip);
 634	if (error)
 635		return error;
 636
 637	do {
 638		/*
 639		 * Set up a transaction to convert the range of extents
 640		 * from unwritten to real. Do allocations in a loop until
 641		 * we have covered the range passed in.
 642		 *
 643		 * Note that we can't risk to recursing back into the filesystem
 644		 * here as we might be asked to write out the same inode that we
 645		 * complete here and might deadlock on the iolock.
 646		 */
 647		error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
 648				0, true, &tp);
 649		if (error)
 650			return error;
 
 
 
 
 
 
 
 
 
 651
 652		error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
 653				XFS_IEXT_WRITE_UNWRITTEN_CNT);
 654		if (error)
 655			goto error_on_bmapi_transaction;
 656
 657		/*
 658		 * Modify the unwritten extent state of the buffer.
 659		 */
 
 660		nimaps = 1;
 661		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 662					XFS_BMAPI_CONVERT, resblks, &imap,
 663					&nimaps);
 664		if (error)
 665			goto error_on_bmapi_transaction;
 666
 667		/*
 668		 * Log the updated inode size as we go.  We have to be careful
 669		 * to only log it up to the actual write offset if it is
 670		 * halfway into a block.
 671		 */
 672		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
 673		if (i_size > offset + count)
 674			i_size = offset + count;
 675		if (update_isize && i_size > i_size_read(inode))
 676			i_size_write(inode, i_size);
 677		i_size = xfs_new_eof(ip, i_size);
 678		if (i_size) {
 679			ip->i_disk_size = i_size;
 680			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 681		}
 682
 683		error = xfs_trans_commit(tp);
 
 
 
 
 684		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 685		if (error)
 686			return error;
 687
 688		if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock))) {
 689			xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
 690			return xfs_alert_fsblock_zero(ip, &imap);
 691		}
 692
 693		if ((numblks_fsb = imap.br_blockcount) == 0) {
 694			/*
 695			 * The numblks_fsb value should always get
 696			 * smaller, otherwise the loop is stuck.
 697			 */
 698			ASSERT(imap.br_blockcount);
 699			break;
 700		}
 701		offset_fsb += numblks_fsb;
 702		count_fsb -= numblks_fsb;
 703	} while (count_fsb > 0);
 704
 705	return 0;
 706
 707error_on_bmapi_transaction:
 708	xfs_trans_cancel(tp);
 
 709	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 710	return error;
 711}
 712
 713static inline bool
 714imap_needs_alloc(
 715	struct inode		*inode,
 716	unsigned		flags,
 717	struct xfs_bmbt_irec	*imap,
 718	int			nimaps)
 719{
 720	/* don't allocate blocks when just zeroing */
 721	if (flags & IOMAP_ZERO)
 722		return false;
 723	if (!nimaps ||
 724	    imap->br_startblock == HOLESTARTBLOCK ||
 725	    imap->br_startblock == DELAYSTARTBLOCK)
 726		return true;
 727	/* we convert unwritten extents before copying the data for DAX */
 728	if ((flags & IOMAP_DAX) && imap->br_state == XFS_EXT_UNWRITTEN)
 729		return true;
 730	return false;
 731}
 732
 733static inline bool
 734imap_needs_cow(
 735	struct xfs_inode	*ip,
 736	unsigned int		flags,
 737	struct xfs_bmbt_irec	*imap,
 738	int			nimaps)
 739{
 740	if (!xfs_is_cow_inode(ip))
 741		return false;
 742
 743	/* when zeroing we don't have to COW holes or unwritten extents */
 744	if (flags & (IOMAP_UNSHARE | IOMAP_ZERO)) {
 745		if (!nimaps ||
 746		    imap->br_startblock == HOLESTARTBLOCK ||
 747		    imap->br_state == XFS_EXT_UNWRITTEN)
 748			return false;
 749	}
 750
 751	return true;
 752}
 753
 754/*
 755 * Extents not yet cached requires exclusive access, don't block for
 756 * IOMAP_NOWAIT.
 757 *
 758 * This is basically an opencoded xfs_ilock_data_map_shared() call, but with
 759 * support for IOMAP_NOWAIT.
 760 */
 761static int
 762xfs_ilock_for_iomap(
 763	struct xfs_inode	*ip,
 764	unsigned		flags,
 765	unsigned		*lockmode)
 766{
 767	if (flags & IOMAP_NOWAIT) {
 768		if (xfs_need_iread_extents(&ip->i_df))
 769			return -EAGAIN;
 770		if (!xfs_ilock_nowait(ip, *lockmode))
 771			return -EAGAIN;
 772	} else {
 773		if (xfs_need_iread_extents(&ip->i_df))
 774			*lockmode = XFS_ILOCK_EXCL;
 775		xfs_ilock(ip, *lockmode);
 776	}
 777
 778	return 0;
 779}
 780
 781/*
 782 * Check that the imap we are going to return to the caller spans the entire
 783 * range that the caller requested for the IO.
 784 */
 785static bool
 786imap_spans_range(
 787	struct xfs_bmbt_irec	*imap,
 788	xfs_fileoff_t		offset_fsb,
 789	xfs_fileoff_t		end_fsb)
 790{
 791	if (imap->br_startoff > offset_fsb)
 792		return false;
 793	if (imap->br_startoff + imap->br_blockcount < end_fsb)
 794		return false;
 795	return true;
 796}
 797
 798static int
 799xfs_direct_write_iomap_begin(
 800	struct inode		*inode,
 801	loff_t			offset,
 802	loff_t			length,
 803	unsigned		flags,
 804	struct iomap		*iomap,
 805	struct iomap		*srcmap)
 806{
 807	struct xfs_inode	*ip = XFS_I(inode);
 808	struct xfs_mount	*mp = ip->i_mount;
 809	struct xfs_bmbt_irec	imap, cmap;
 810	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 811	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
 812	int			nimaps = 1, error = 0;
 813	bool			shared = false;
 814	u16			iomap_flags = 0;
 815	unsigned int		lockmode;
 816	u64			seq;
 817
 818	ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
 819
 820	if (xfs_is_shutdown(mp))
 821		return -EIO;
 822
 823	/*
 824	 * Writes that span EOF might trigger an IO size update on completion,
 825	 * so consider them to be dirty for the purposes of O_DSYNC even if
 826	 * there is no other metadata changes pending or have been made here.
 827	 */
 828	if (offset + length > i_size_read(inode))
 829		iomap_flags |= IOMAP_F_DIRTY;
 830
 831	/*
 832	 * COW writes may allocate delalloc space or convert unwritten COW
 833	 * extents, so we need to make sure to take the lock exclusively here.
 834	 */
 835	if (xfs_is_cow_inode(ip))
 836		lockmode = XFS_ILOCK_EXCL;
 837	else
 838		lockmode = XFS_ILOCK_SHARED;
 839
 840relock:
 841	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
 842	if (error)
 843		return error;
 844
 845	/*
 846	 * The reflink iflag could have changed since the earlier unlocked
 847	 * check, check if it again and relock if needed.
 848	 */
 849	if (xfs_is_cow_inode(ip) && lockmode == XFS_ILOCK_SHARED) {
 850		xfs_iunlock(ip, lockmode);
 851		lockmode = XFS_ILOCK_EXCL;
 852		goto relock;
 853	}
 854
 855	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
 856			       &nimaps, 0);
 857	if (error)
 858		goto out_unlock;
 859
 860	if (imap_needs_cow(ip, flags, &imap, nimaps)) {
 861		error = -EAGAIN;
 862		if (flags & IOMAP_NOWAIT)
 863			goto out_unlock;
 864
 865		/* may drop and re-acquire the ilock */
 866		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
 867				&lockmode,
 868				(flags & IOMAP_DIRECT) || IS_DAX(inode));
 869		if (error)
 870			goto out_unlock;
 871		if (shared)
 872			goto out_found_cow;
 873		end_fsb = imap.br_startoff + imap.br_blockcount;
 874		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
 875	}
 876
 877	if (imap_needs_alloc(inode, flags, &imap, nimaps))
 878		goto allocate_blocks;
 879
 880	/*
 881	 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
 882	 * a single map so that we avoid partial IO failures due to the rest of
 883	 * the I/O range not covered by this map triggering an EAGAIN condition
 884	 * when it is subsequently mapped and aborting the I/O.
 885	 */
 886	if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
 887		error = -EAGAIN;
 888		if (!imap_spans_range(&imap, offset_fsb, end_fsb))
 889			goto out_unlock;
 890	}
 891
 892	/*
 893	 * For overwrite only I/O, we cannot convert unwritten extents without
 894	 * requiring sub-block zeroing.  This can only be done under an
 895	 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
 896	 * extent to tell the caller to try again.
 897	 */
 898	if (flags & IOMAP_OVERWRITE_ONLY) {
 899		error = -EAGAIN;
 900		if (imap.br_state != XFS_EXT_NORM &&
 901	            ((offset | length) & mp->m_blockmask))
 902			goto out_unlock;
 903	}
 904
 905	seq = xfs_iomap_inode_sequence(ip, iomap_flags);
 906	xfs_iunlock(ip, lockmode);
 907	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
 908	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
 909
 910allocate_blocks:
 911	error = -EAGAIN;
 912	if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
 913		goto out_unlock;
 914
 915	/*
 916	 * We cap the maximum length we map to a sane size  to keep the chunks
 917	 * of work done where somewhat symmetric with the work writeback does.
 918	 * This is a completely arbitrary number pulled out of thin air as a
 919	 * best guess for initial testing.
 920	 *
 921	 * Note that the values needs to be less than 32-bits wide until the
 922	 * lower level functions are updated.
 923	 */
 924	length = min_t(loff_t, length, 1024 * PAGE_SIZE);
 925	end_fsb = xfs_iomap_end_fsb(mp, offset, length);
 926
 927	if (offset + length > XFS_ISIZE(ip))
 928		end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
 929	else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
 930		end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
 931	xfs_iunlock(ip, lockmode);
 932
 933	error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
 934			flags, &imap, &seq);
 935	if (error)
 936		return error;
 937
 938	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
 939	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
 940				 iomap_flags | IOMAP_F_NEW, seq);
 941
 942out_found_cow:
 943	length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
 944	trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
 945	if (imap.br_startblock != HOLESTARTBLOCK) {
 946		seq = xfs_iomap_inode_sequence(ip, 0);
 947		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
 948		if (error)
 949			goto out_unlock;
 950	}
 951	seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
 952	xfs_iunlock(ip, lockmode);
 953	return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, IOMAP_F_SHARED, seq);
 954
 955out_unlock:
 956	if (lockmode)
 957		xfs_iunlock(ip, lockmode);
 958	return error;
 959}
 960
 961const struct iomap_ops xfs_direct_write_iomap_ops = {
 962	.iomap_begin		= xfs_direct_write_iomap_begin,
 963};
 964
 965static int
 966xfs_dax_write_iomap_end(
 967	struct inode		*inode,
 968	loff_t			pos,
 969	loff_t			length,
 970	ssize_t			written,
 971	unsigned		flags,
 972	struct iomap		*iomap)
 973{
 974	struct xfs_inode	*ip = XFS_I(inode);
 975
 976	if (!xfs_is_cow_inode(ip))
 977		return 0;
 978
 979	if (!written)
 980		return xfs_reflink_cancel_cow_range(ip, pos, length, true);
 981
 982	return xfs_reflink_end_cow(ip, pos, written);
 983}
 984
 985const struct iomap_ops xfs_dax_write_iomap_ops = {
 986	.iomap_begin	= xfs_direct_write_iomap_begin,
 987	.iomap_end	= xfs_dax_write_iomap_end,
 988};
 989
 990static int
 991xfs_buffered_write_iomap_begin(
 992	struct inode		*inode,
 993	loff_t			offset,
 994	loff_t			count,
 995	unsigned		flags,
 996	struct iomap		*iomap,
 997	struct iomap		*srcmap)
 998{
 999	struct xfs_inode	*ip = XFS_I(inode);
1000	struct xfs_mount	*mp = ip->i_mount;
1001	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1002	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1003	struct xfs_bmbt_irec	imap, cmap;
1004	struct xfs_iext_cursor	icur, ccur;
1005	xfs_fsblock_t		prealloc_blocks = 0;
1006	bool			eof = false, cow_eof = false, shared = false;
1007	int			allocfork = XFS_DATA_FORK;
1008	int			error = 0;
1009	unsigned int		lockmode = XFS_ILOCK_EXCL;
1010	unsigned int		iomap_flags = 0;
1011	u64			seq;
1012
1013	if (xfs_is_shutdown(mp))
1014		return -EIO;
1015
1016	/* we can't use delayed allocations when using extent size hints */
1017	if (xfs_get_extsz_hint(ip))
1018		return xfs_direct_write_iomap_begin(inode, offset, count,
1019				flags, iomap, srcmap);
1020
1021	error = xfs_qm_dqattach(ip);
1022	if (error)
1023		return error;
1024
1025	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1026	if (error)
1027		return error;
1028
1029	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
1030	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
1031		xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
1032		error = -EFSCORRUPTED;
1033		goto out_unlock;
1034	}
1035
1036	XFS_STATS_INC(mp, xs_blk_mapw);
1037
1038	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1039	if (error)
1040		goto out_unlock;
1041
1042	/*
1043	 * Search the data fork first to look up our source mapping.  We
1044	 * always need the data fork map, as we have to return it to the
1045	 * iomap code so that the higher level write code can read data in to
1046	 * perform read-modify-write cycles for unaligned writes.
1047	 */
1048	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
1049	if (eof)
1050		imap.br_startoff = end_fsb; /* fake hole until the end */
1051
1052	/* We never need to allocate blocks for zeroing or unsharing a hole. */
1053	if ((flags & (IOMAP_UNSHARE | IOMAP_ZERO)) &&
1054	    imap.br_startoff > offset_fsb) {
1055		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
1056		goto out_unlock;
1057	}
1058
1059	/*
1060	 * For zeroing, trim a delalloc extent that extends beyond the EOF
1061	 * block.  If it starts beyond the EOF block, convert it to an
1062	 * unwritten extent.
1063	 */
1064	if ((flags & IOMAP_ZERO) && imap.br_startoff <= offset_fsb &&
1065	    isnullstartblock(imap.br_startblock)) {
1066		xfs_fileoff_t eof_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
1067
1068		if (offset_fsb >= eof_fsb)
1069			goto convert_delay;
1070		if (end_fsb > eof_fsb) {
1071			end_fsb = eof_fsb;
1072			xfs_trim_extent(&imap, offset_fsb,
1073					end_fsb - offset_fsb);
1074		}
1075	}
1076
1077	/*
1078	 * Search the COW fork extent list even if we did not find a data fork
1079	 * extent.  This serves two purposes: first this implements the
1080	 * speculative preallocation using cowextsize, so that we also unshare
1081	 * block adjacent to shared blocks instead of just the shared blocks
1082	 * themselves.  Second the lookup in the extent list is generally faster
1083	 * than going out to the shared extent tree.
1084	 */
1085	if (xfs_is_cow_inode(ip)) {
1086		if (!ip->i_cowfp) {
1087			ASSERT(!xfs_is_reflink_inode(ip));
1088			xfs_ifork_init_cow(ip);
1089		}
1090		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
1091				&ccur, &cmap);
1092		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
1093			trace_xfs_reflink_cow_found(ip, &cmap);
1094			goto found_cow;
1095		}
1096	}
1097
1098	if (imap.br_startoff <= offset_fsb) {
1099		/*
1100		 * For reflink files we may need a delalloc reservation when
1101		 * overwriting shared extents.   This includes zeroing of
1102		 * existing extents that contain data.
1103		 */
1104		if (!xfs_is_cow_inode(ip) ||
1105		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
1106			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1107					&imap);
1108			goto found_imap;
1109		}
1110
1111		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1112
1113		/* Trim the mapping to the nearest shared extent boundary. */
1114		error = xfs_bmap_trim_cow(ip, &imap, &shared);
1115		if (error)
1116			goto out_unlock;
1117
1118		/* Not shared?  Just report the (potentially capped) extent. */
1119		if (!shared) {
1120			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1121					&imap);
1122			goto found_imap;
1123		}
1124
1125		/*
1126		 * Fork all the shared blocks from our write offset until the
1127		 * end of the extent.
1128		 */
1129		allocfork = XFS_COW_FORK;
1130		end_fsb = imap.br_startoff + imap.br_blockcount;
1131	} else {
1132		/*
1133		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1134		 * pages to keep the chunks of work done where somewhat
1135		 * symmetric with the work writeback does.  This is a completely
1136		 * arbitrary number pulled out of thin air.
1137		 *
1138		 * Note that the values needs to be less than 32-bits wide until
1139		 * the lower level functions are updated.
1140		 */
1141		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
1142		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1143
1144		if (xfs_is_always_cow_inode(ip))
1145			allocfork = XFS_COW_FORK;
1146	}
1147
1148	if (eof && offset + count > XFS_ISIZE(ip)) {
1149		/*
1150		 * Determine the initial size of the preallocation.
1151		 * We clean up any extra preallocation when the file is closed.
1152		 */
1153		if (xfs_has_allocsize(mp))
1154			prealloc_blocks = mp->m_allocsize_blocks;
1155		else if (allocfork == XFS_DATA_FORK)
1156			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1157						offset, count, &icur);
1158		else
1159			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1160						offset, count, &ccur);
1161		if (prealloc_blocks) {
1162			xfs_extlen_t	align;
1163			xfs_off_t	end_offset;
1164			xfs_fileoff_t	p_end_fsb;
1165
1166			end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1167			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1168					prealloc_blocks;
1169
1170			align = xfs_eof_alignment(ip);
1171			if (align)
1172				p_end_fsb = roundup_64(p_end_fsb, align);
1173
1174			p_end_fsb = min(p_end_fsb,
1175				XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1176			ASSERT(p_end_fsb > offset_fsb);
1177			prealloc_blocks = p_end_fsb - end_fsb;
1178		}
1179	}
1180
1181	/*
1182	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1183	 * them out if the write happens to fail.
1184	 */
1185	iomap_flags |= IOMAP_F_NEW;
1186	if (allocfork == XFS_COW_FORK) {
1187		error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1188				end_fsb - offset_fsb, prealloc_blocks, &cmap,
1189				&ccur, cow_eof);
1190		if (error)
1191			goto out_unlock;
1192
1193		trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1194		goto found_cow;
1195	}
1196
1197	error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1198			end_fsb - offset_fsb, prealloc_blocks, &imap, &icur,
1199			eof);
1200	if (error)
1201		goto out_unlock;
1202
1203	trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1204found_imap:
1205	seq = xfs_iomap_inode_sequence(ip, iomap_flags);
1206	xfs_iunlock(ip, lockmode);
1207	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
1208
1209convert_delay:
1210	xfs_iunlock(ip, lockmode);
1211	truncate_pagecache(inode, offset);
1212	error = xfs_bmapi_convert_delalloc(ip, XFS_DATA_FORK, offset,
1213					   iomap, NULL);
1214	if (error)
1215		return error;
1216
1217	trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &imap);
1218	return 0;
1219
1220found_cow:
1221	if (imap.br_startoff <= offset_fsb) {
1222		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0,
1223				xfs_iomap_inode_sequence(ip, 0));
1224		if (error)
1225			goto out_unlock;
1226	} else {
1227		xfs_trim_extent(&cmap, offset_fsb,
1228				imap.br_startoff - offset_fsb);
1229	}
1230
1231	iomap_flags |= IOMAP_F_SHARED;
1232	seq = xfs_iomap_inode_sequence(ip, iomap_flags);
1233	xfs_iunlock(ip, lockmode);
1234	return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, iomap_flags, seq);
1235
1236out_unlock:
1237	xfs_iunlock(ip, lockmode);
1238	return error;
1239}
1240
1241static void
1242xfs_buffered_write_delalloc_punch(
1243	struct inode		*inode,
1244	loff_t			offset,
1245	loff_t			length,
1246	struct iomap		*iomap)
1247{
1248	xfs_bmap_punch_delalloc_range(XFS_I(inode),
1249			(iomap->flags & IOMAP_F_SHARED) ?
1250				XFS_COW_FORK : XFS_DATA_FORK,
1251			offset, offset + length);
1252}
1253
1254static int
1255xfs_buffered_write_iomap_end(
1256	struct inode		*inode,
1257	loff_t			offset,
1258	loff_t			length,
1259	ssize_t			written,
1260	unsigned		flags,
1261	struct iomap		*iomap)
1262{
1263	loff_t			start_byte, end_byte;
1264
1265	/* If we didn't reserve the blocks, we're not allowed to punch them. */
1266	if (iomap->type != IOMAP_DELALLOC || !(iomap->flags & IOMAP_F_NEW))
1267		return 0;
1268
1269	/*
1270	 * iomap_page_mkwrite() will never fail in a way that requires delalloc
1271	 * extents that it allocated to be revoked.  Hence never try to release
1272	 * them here.
1273	 */
1274	if (flags & IOMAP_FAULT)
1275		return 0;
1276
1277	/* Nothing to do if we've written the entire delalloc extent */
1278	start_byte = iomap_last_written_block(inode, offset, written);
1279	end_byte = round_up(offset + length, i_blocksize(inode));
1280	if (start_byte >= end_byte)
1281		return 0;
1282
1283	/* For zeroing operations the callers already hold invalidate_lock. */
1284	if (flags & (IOMAP_UNSHARE | IOMAP_ZERO)) {
1285		rwsem_assert_held_write(&inode->i_mapping->invalidate_lock);
1286		iomap_write_delalloc_release(inode, start_byte, end_byte, flags,
1287				iomap, xfs_buffered_write_delalloc_punch);
1288	} else {
1289		filemap_invalidate_lock(inode->i_mapping);
1290		iomap_write_delalloc_release(inode, start_byte, end_byte, flags,
1291				iomap, xfs_buffered_write_delalloc_punch);
1292		filemap_invalidate_unlock(inode->i_mapping);
1293	}
1294
1295	return 0;
1296}
1297
1298const struct iomap_ops xfs_buffered_write_iomap_ops = {
1299	.iomap_begin		= xfs_buffered_write_iomap_begin,
1300	.iomap_end		= xfs_buffered_write_iomap_end,
1301};
1302
1303static int
1304xfs_read_iomap_begin(
1305	struct inode		*inode,
1306	loff_t			offset,
1307	loff_t			length,
1308	unsigned		flags,
1309	struct iomap		*iomap,
1310	struct iomap		*srcmap)
1311{
1312	struct xfs_inode	*ip = XFS_I(inode);
1313	struct xfs_mount	*mp = ip->i_mount;
1314	struct xfs_bmbt_irec	imap;
1315	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1316	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1317	int			nimaps = 1, error = 0;
1318	bool			shared = false;
1319	unsigned int		lockmode = XFS_ILOCK_SHARED;
1320	u64			seq;
1321
1322	ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1323
1324	if (xfs_is_shutdown(mp))
1325		return -EIO;
1326
1327	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1328	if (error)
1329		return error;
1330	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1331			       &nimaps, 0);
1332	if (!error && ((flags & IOMAP_REPORT) || IS_DAX(inode)))
1333		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1334	seq = xfs_iomap_inode_sequence(ip, shared ? IOMAP_F_SHARED : 0);
1335	xfs_iunlock(ip, lockmode);
1336
1337	if (error)
1338		return error;
1339	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1340	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
1341				 shared ? IOMAP_F_SHARED : 0, seq);
1342}
1343
1344const struct iomap_ops xfs_read_iomap_ops = {
1345	.iomap_begin		= xfs_read_iomap_begin,
1346};
1347
1348static int
1349xfs_seek_iomap_begin(
1350	struct inode		*inode,
1351	loff_t			offset,
1352	loff_t			length,
1353	unsigned		flags,
1354	struct iomap		*iomap,
1355	struct iomap		*srcmap)
1356{
1357	struct xfs_inode	*ip = XFS_I(inode);
1358	struct xfs_mount	*mp = ip->i_mount;
1359	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1360	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1361	xfs_fileoff_t		cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1362	struct xfs_iext_cursor	icur;
1363	struct xfs_bmbt_irec	imap, cmap;
1364	int			error = 0;
1365	unsigned		lockmode;
1366	u64			seq;
1367
1368	if (xfs_is_shutdown(mp))
1369		return -EIO;
1370
1371	lockmode = xfs_ilock_data_map_shared(ip);
1372	error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1373	if (error)
1374		goto out_unlock;
1375
1376	if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1377		/*
1378		 * If we found a data extent we are done.
1379		 */
1380		if (imap.br_startoff <= offset_fsb)
1381			goto done;
1382		data_fsb = imap.br_startoff;
1383	} else {
1384		/*
1385		 * Fake a hole until the end of the file.
1386		 */
1387		data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1388	}
1389
1390	/*
1391	 * If a COW fork extent covers the hole, report it - capped to the next
1392	 * data fork extent:
1393	 */
1394	if (xfs_inode_has_cow_data(ip) &&
1395	    xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1396		cow_fsb = cmap.br_startoff;
1397	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1398		if (data_fsb < cow_fsb + cmap.br_blockcount)
1399			end_fsb = min(end_fsb, data_fsb);
1400		xfs_trim_extent(&cmap, offset_fsb, end_fsb - offset_fsb);
1401		seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1402		error = xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1403				IOMAP_F_SHARED, seq);
1404		/*
1405		 * This is a COW extent, so we must probe the page cache
1406		 * because there could be dirty page cache being backed
1407		 * by this extent.
1408		 */
1409		iomap->type = IOMAP_UNWRITTEN;
1410		goto out_unlock;
1411	}
1412
1413	/*
1414	 * Else report a hole, capped to the next found data or COW extent.
1415	 */
1416	if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1417		imap.br_blockcount = cow_fsb - offset_fsb;
1418	else
1419		imap.br_blockcount = data_fsb - offset_fsb;
1420	imap.br_startoff = offset_fsb;
1421	imap.br_startblock = HOLESTARTBLOCK;
1422	imap.br_state = XFS_EXT_NORM;
1423done:
1424	seq = xfs_iomap_inode_sequence(ip, 0);
1425	xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1426	error = xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1427out_unlock:
1428	xfs_iunlock(ip, lockmode);
1429	return error;
1430}
1431
1432const struct iomap_ops xfs_seek_iomap_ops = {
1433	.iomap_begin		= xfs_seek_iomap_begin,
1434};
1435
1436static int
1437xfs_xattr_iomap_begin(
1438	struct inode		*inode,
1439	loff_t			offset,
1440	loff_t			length,
1441	unsigned		flags,
1442	struct iomap		*iomap,
1443	struct iomap		*srcmap)
1444{
1445	struct xfs_inode	*ip = XFS_I(inode);
1446	struct xfs_mount	*mp = ip->i_mount;
1447	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1448	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1449	struct xfs_bmbt_irec	imap;
1450	int			nimaps = 1, error = 0;
1451	unsigned		lockmode;
1452	int			seq;
1453
1454	if (xfs_is_shutdown(mp))
1455		return -EIO;
1456
1457	lockmode = xfs_ilock_attr_map_shared(ip);
1458
1459	/* if there are no attribute fork or extents, return ENOENT */
1460	if (!xfs_inode_has_attr_fork(ip) || !ip->i_af.if_nextents) {
1461		error = -ENOENT;
1462		goto out_unlock;
1463	}
1464
1465	ASSERT(ip->i_af.if_format != XFS_DINODE_FMT_LOCAL);
1466	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1467			       &nimaps, XFS_BMAPI_ATTRFORK);
1468out_unlock:
1469
1470	seq = xfs_iomap_inode_sequence(ip, IOMAP_F_XATTR);
1471	xfs_iunlock(ip, lockmode);
1472
1473	if (error)
1474		return error;
1475	ASSERT(nimaps);
1476	return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_XATTR, seq);
1477}
1478
1479const struct iomap_ops xfs_xattr_iomap_ops = {
1480	.iomap_begin		= xfs_xattr_iomap_begin,
1481};
1482
1483int
1484xfs_zero_range(
1485	struct xfs_inode	*ip,
1486	loff_t			pos,
1487	loff_t			len,
1488	bool			*did_zero)
1489{
1490	struct inode		*inode = VFS_I(ip);
1491
1492	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
1493
1494	if (IS_DAX(inode))
1495		return dax_zero_range(inode, pos, len, did_zero,
1496				      &xfs_dax_write_iomap_ops);
1497	return iomap_zero_range(inode, pos, len, did_zero,
1498				&xfs_buffered_write_iomap_ops);
1499}
1500
1501int
1502xfs_truncate_page(
1503	struct xfs_inode	*ip,
1504	loff_t			pos,
1505	bool			*did_zero)
1506{
1507	struct inode		*inode = VFS_I(ip);
1508
1509	if (IS_DAX(inode))
1510		return dax_truncate_page(inode, pos, did_zero,
1511					&xfs_dax_write_iomap_ops);
1512	return iomap_truncate_page(inode, pos, did_zero,
1513				   &xfs_buffered_write_iomap_ops);
1514}