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}
v5.9
   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 && (mp->m_flags & XFS_MOUNT_SWALLOC))
 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(ifp->if_flags & XFS_IFEXTENTS);
 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	int			quota_flag;
 198	uint			qblocks, resblks;
 199	unsigned int		resrtextents = 0;
 200	int			error;
 201	int			bmapi_flags = XFS_BMAPI_PREALLOC;
 202	uint			tflags = 0;
 
 
 
 
 
 
 
 
 
 203
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 204	ASSERT(count_fsb > 0);
 205
 206	resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
 207					   xfs_get_extsz_hint(ip));
 208	if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
 
 
 
 
 
 
 209		resrtextents = qblocks = resaligned;
 210		resrtextents /= mp->m_sb.sb_rextsize;
 211		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
 212		quota_flag = XFS_QMOPT_RES_RTBLKS;
 213	} else {
 
 214		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
 215		quota_flag = XFS_QMOPT_RES_REGBLKS;
 216	}
 217
 218	error = xfs_qm_dqattach(ip);
 219	if (error)
 220		return error;
 221
 222	/*
 223	 * For DAX, we do not allocate unwritten extents, but instead we zero
 224	 * the block before we commit the transaction.  Ideally we'd like to do
 225	 * this outside the transaction context, but if we commit and then crash
 226	 * we may not have zeroed the blocks and this will be exposed on
 227	 * recovery of the allocation. Hence we must zero before commit.
 228	 *
 229	 * Further, if we are mapping unwritten extents here, we need to zero
 230	 * and convert them to written so that we don't need an unwritten extent
 231	 * callback for DAX. This also means that we need to be able to dip into
 232	 * the reserve block pool for bmbt block allocation if there is no space
 233	 * left but we need to do unwritten extent conversion.
 234	 */
 235	if (IS_DAX(VFS_I(ip))) {
 236		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
 237		if (imap->br_state == XFS_EXT_UNWRITTEN) {
 238			tflags |= XFS_TRANS_RESERVE;
 239			resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 240		}
 
 
 
 
 
 241	}
 242	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
 243			tflags, &tp);
 244	if (error)
 245		return error;
 246
 247	xfs_ilock(ip, XFS_ILOCK_EXCL);
 248
 249	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
 250	if (error)
 251		goto out_trans_cancel;
 252
 253	xfs_trans_ijoin(tp, ip, 0);
 254
 
 
 
 
 255	/*
 256	 * From this point onwards we overwrite the imap pointer that the
 257	 * caller gave to us.
 258	 */
 
 259	nimaps = 1;
 260	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
 261				imap, &nimaps);
 262	if (error)
 263		goto out_res_cancel;
 264
 265	/*
 266	 * Complete the transaction
 267	 */
 268	error = xfs_trans_commit(tp);
 
 
 
 269	if (error)
 270		goto out_unlock;
 271
 272	/*
 273	 * Copy any maps to caller's array and return any error.
 274	 */
 275	if (nimaps == 0) {
 276		error = -ENOSPC;
 277		goto out_unlock;
 278	}
 279
 280	if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
 281		error = xfs_alert_fsblock_zero(ip, imap);
 282
 283out_unlock:
 284	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 285	return error;
 286
 287out_res_cancel:
 
 288	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
 289out_trans_cancel:
 290	xfs_trans_cancel(tp);
 291	goto out_unlock;
 292}
 293
 294STATIC bool
 295xfs_quota_need_throttle(
 296	struct xfs_inode	*ip,
 297	xfs_dqtype_t		type,
 298	xfs_fsblock_t		alloc_blocks)
 299{
 300	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 301
 302	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
 303		return false;
 
 304
 305	/* no hi watermark, no throttle */
 306	if (!dq->q_prealloc_hi_wmark)
 307		return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 308
 309	/* under the lo watermark, no throttle */
 310	if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
 311		return false;
 312
 313	return true;
 314}
 315
 316STATIC void
 317xfs_quota_calc_throttle(
 318	struct xfs_inode	*ip,
 319	xfs_dqtype_t		type,
 320	xfs_fsblock_t		*qblocks,
 321	int			*qshift,
 322	int64_t			*qfreesp)
 323{
 324	struct xfs_dquot	*dq = xfs_inode_dquot(ip, type);
 325	int64_t			freesp;
 326	int			shift = 0;
 327
 328	/* no dq, or over hi wmark, squash the prealloc completely */
 329	if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
 330		*qblocks = 0;
 331		*qfreesp = 0;
 332		return;
 333	}
 334
 335	freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
 336	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
 337		shift = 2;
 338		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
 339			shift += 2;
 340		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
 341			shift += 2;
 342	}
 343
 344	if (freesp < *qfreesp)
 345		*qfreesp = freesp;
 346
 347	/* only overwrite the throttle values if we are more aggressive */
 348	if ((freesp >> shift) < (*qblocks >> *qshift)) {
 349		*qblocks = freesp;
 350		*qshift = shift;
 351	}
 
 
 
 352}
 353
 354/*
 355 * If we don't have a user specified preallocation size, dynamically increase
 356 * the preallocation size as the size of the file grows.  Cap the maximum size
 357 * at a single extent or less if the filesystem is near full. The closer the
 358 * filesystem is to being full, the smaller the maximum preallocation.
 359 */
 360STATIC xfs_fsblock_t
 361xfs_iomap_prealloc_size(
 362	struct xfs_inode	*ip,
 363	int			whichfork,
 364	loff_t			offset,
 365	loff_t			count,
 366	struct xfs_iext_cursor	*icur)
 367{
 368	struct xfs_iext_cursor	ncur = *icur;
 369	struct xfs_bmbt_irec	prev, got;
 370	struct xfs_mount	*mp = ip->i_mount;
 371	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
 372	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 373	int64_t			freesp;
 374	xfs_fsblock_t		qblocks;
 375	xfs_fsblock_t		alloc_blocks = 0;
 376	xfs_extlen_t		plen;
 377	int			shift = 0;
 378	int			qshift = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 379
 380	/*
 381	 * As an exception we don't do any preallocation at all if the file is
 382	 * smaller than the minimum preallocation and we are using the default
 383	 * dynamic preallocation scheme, as it is likely this is the only write
 384	 * to the file that is going to be done.
 385	 */
 386	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
 387		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 388
 389	/*
 390	 * Use the minimum preallocation size for small files or if we are
 391	 * writing right after a hole.
 392	 */
 393	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
 394	    !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
 395	    prev.br_startoff + prev.br_blockcount < offset_fsb)
 396		return mp->m_allocsize_blocks;
 397
 398	/*
 399	 * Take the size of the preceding data extents as the basis for the
 400	 * preallocation size. Note that we don't care if the previous extents
 401	 * are written or not.
 402	 */
 403	plen = prev.br_blockcount;
 404	while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
 405		if (plen > MAXEXTLEN / 2 ||
 406		    isnullstartblock(got.br_startblock) ||
 407		    got.br_startoff + got.br_blockcount != prev.br_startoff ||
 408		    got.br_startblock + got.br_blockcount != prev.br_startblock)
 409			break;
 410		plen += got.br_blockcount;
 411		prev = got;
 412	}
 413
 414	/*
 415	 * If the size of the extents is greater than half the maximum extent
 416	 * length, then use the current offset as the basis.  This ensures that
 417	 * for large files the preallocation size always extends to MAXEXTLEN
 418	 * rather than falling short due to things like stripe unit/width
 419	 * alignment of real extents.
 420	 */
 421	alloc_blocks = plen * 2;
 422	if (alloc_blocks > MAXEXTLEN)
 423		alloc_blocks = XFS_B_TO_FSB(mp, offset);
 424	qblocks = alloc_blocks;
 425
 426	/*
 427	 * MAXEXTLEN is not a power of two value but we round the prealloc down
 428	 * to the nearest power of two value after throttling. To prevent the
 429	 * round down from unconditionally reducing the maximum supported
 430	 * prealloc size, we round up first, apply appropriate throttling,
 431	 * round down and cap the value to MAXEXTLEN.
 432	 */
 433	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
 434				       alloc_blocks);
 435
 436	freesp = percpu_counter_read_positive(&mp->m_fdblocks);
 437	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
 438		shift = 2;
 439		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
 440			shift++;
 441		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
 442			shift++;
 443		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
 444			shift++;
 445		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
 446			shift++;
 447	}
 448
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 449	/*
 450	 * Check each quota to cap the prealloc size, provide a shift value to
 451	 * throttle with and adjust amount of available space.
 452	 */
 453	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
 454		xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
 455					&freesp);
 456	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
 457		xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
 458					&freesp);
 459	if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
 460		xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
 461					&freesp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 462
 463	/*
 464	 * The final prealloc size is set to the minimum of free space available
 465	 * in each of the quotas and the overall filesystem.
 466	 *
 467	 * The shift throttle value is set to the maximum value as determined by
 468	 * the global low free space values and per-quota low free space values.
 469	 */
 470	alloc_blocks = min(alloc_blocks, qblocks);
 471	shift = max(shift, qshift);
 472
 473	if (shift)
 474		alloc_blocks >>= shift;
 475	/*
 476	 * rounddown_pow_of_two() returns an undefined result if we pass in
 477	 * alloc_blocks = 0.
 478	 */
 479	if (alloc_blocks)
 480		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
 481	if (alloc_blocks > MAXEXTLEN)
 482		alloc_blocks = MAXEXTLEN;
 483
 484	/*
 485	 * If we are still trying to allocate more space than is
 486	 * available, squash the prealloc hard. This can happen if we
 487	 * have a large file on a small filesystem and the above
 488	 * lowspace thresholds are smaller than MAXEXTLEN.
 489	 */
 490	while (alloc_blocks && alloc_blocks >= freesp)
 491		alloc_blocks >>= 4;
 492	if (alloc_blocks < mp->m_allocsize_blocks)
 493		alloc_blocks = mp->m_allocsize_blocks;
 494	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
 495				      mp->m_allocsize_blocks);
 496	return alloc_blocks;
 497}
 498
 499int
 500xfs_iomap_write_unwritten(
 501	xfs_inode_t	*ip,
 502	xfs_off_t	offset,
 503	xfs_off_t	count,
 504	bool		update_isize)
 505{
 506	xfs_mount_t	*mp = ip->i_mount;
 507	xfs_fileoff_t	offset_fsb;
 508	xfs_filblks_t	count_fsb;
 509	xfs_filblks_t	numblks_fsb;
 
 510	int		nimaps;
 511	xfs_trans_t	*tp;
 512	xfs_bmbt_irec_t imap;
 513	struct inode	*inode = VFS_I(ip);
 514	xfs_fsize_t	i_size;
 515	uint		resblks;
 
 516	int		error;
 517
 518	trace_xfs_unwritten_convert(ip, offset, count);
 519
 520	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 521	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
 522	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
 523
 524	/*
 525	 * Reserve enough blocks in this transaction for two complete extent
 526	 * btree splits.  We may be converting the middle part of an unwritten
 527	 * extent and in this case we will insert two new extents in the btree
 528	 * each of which could cause a full split.
 529	 *
 530	 * This reservation amount will be used in the first call to
 531	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
 532	 * rest of the operation.
 533	 */
 534	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
 535
 536	/* Attach dquots so that bmbt splits are accounted correctly. */
 537	error = xfs_qm_dqattach(ip);
 538	if (error)
 539		return error;
 540
 541	do {
 542		/*
 543		 * Set up a transaction to convert the range of extents
 544		 * from unwritten to real. Do allocations in a loop until
 545		 * we have covered the range passed in.
 546		 *
 547		 * Note that we can't risk to recursing back into the filesystem
 548		 * here as we might be asked to write out the same inode that we
 549		 * complete here and might deadlock on the iolock.
 550		 */
 551		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
 552				XFS_TRANS_RESERVE, &tp);
 553		if (error)
 554			return error;
 
 
 
 
 
 
 
 
 
 555
 556		xfs_ilock(ip, XFS_ILOCK_EXCL);
 557		xfs_trans_ijoin(tp, ip, 0);
 558
 559		error = xfs_trans_reserve_quota_nblks(tp, ip, resblks, 0,
 560				XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES);
 561		if (error)
 562			goto error_on_bmapi_transaction;
 563
 564		/*
 565		 * Modify the unwritten extent state of the buffer.
 566		 */
 
 567		nimaps = 1;
 568		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
 569					XFS_BMAPI_CONVERT, resblks, &imap,
 570					&nimaps);
 571		if (error)
 572			goto error_on_bmapi_transaction;
 573
 574		/*
 575		 * Log the updated inode size as we go.  We have to be careful
 576		 * to only log it up to the actual write offset if it is
 577		 * halfway into a block.
 578		 */
 579		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
 580		if (i_size > offset + count)
 581			i_size = offset + count;
 582		if (update_isize && i_size > i_size_read(inode))
 583			i_size_write(inode, i_size);
 584		i_size = xfs_new_eof(ip, i_size);
 585		if (i_size) {
 586			ip->i_d.di_size = i_size;
 587			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 588		}
 589
 590		error = xfs_trans_commit(tp);
 
 
 
 
 591		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 592		if (error)
 593			return error;
 594
 595		if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
 596			return xfs_alert_fsblock_zero(ip, &imap);
 597
 598		if ((numblks_fsb = imap.br_blockcount) == 0) {
 599			/*
 600			 * The numblks_fsb value should always get
 601			 * smaller, otherwise the loop is stuck.
 602			 */
 603			ASSERT(imap.br_blockcount);
 604			break;
 605		}
 606		offset_fsb += numblks_fsb;
 607		count_fsb -= numblks_fsb;
 608	} while (count_fsb > 0);
 609
 610	return 0;
 611
 612error_on_bmapi_transaction:
 613	xfs_trans_cancel(tp);
 614	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 615	return error;
 616}
 617
 618static inline bool
 619imap_needs_alloc(
 620	struct inode		*inode,
 621	unsigned		flags,
 622	struct xfs_bmbt_irec	*imap,
 623	int			nimaps)
 624{
 625	/* don't allocate blocks when just zeroing */
 626	if (flags & IOMAP_ZERO)
 627		return false;
 628	if (!nimaps ||
 629	    imap->br_startblock == HOLESTARTBLOCK ||
 630	    imap->br_startblock == DELAYSTARTBLOCK)
 631		return true;
 632	/* we convert unwritten extents before copying the data for DAX */
 633	if (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN)
 634		return true;
 635	return false;
 636}
 637
 638static inline bool
 639imap_needs_cow(
 640	struct xfs_inode	*ip,
 641	unsigned int		flags,
 642	struct xfs_bmbt_irec	*imap,
 643	int			nimaps)
 644{
 645	if (!xfs_is_cow_inode(ip))
 646		return false;
 647
 648	/* when zeroing we don't have to COW holes or unwritten extents */
 649	if (flags & IOMAP_ZERO) {
 650		if (!nimaps ||
 651		    imap->br_startblock == HOLESTARTBLOCK ||
 652		    imap->br_state == XFS_EXT_UNWRITTEN)
 653			return false;
 654	}
 655
 656	return true;
 657}
 658
 659static int
 660xfs_ilock_for_iomap(
 661	struct xfs_inode	*ip,
 662	unsigned		flags,
 663	unsigned		*lockmode)
 664{
 665	unsigned		mode = XFS_ILOCK_SHARED;
 666	bool			is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
 667
 668	/*
 669	 * COW writes may allocate delalloc space or convert unwritten COW
 670	 * extents, so we need to make sure to take the lock exclusively here.
 671	 */
 672	if (xfs_is_cow_inode(ip) && is_write)
 673		mode = XFS_ILOCK_EXCL;
 674
 675	/*
 676	 * Extents not yet cached requires exclusive access, don't block.  This
 677	 * is an opencoded xfs_ilock_data_map_shared() call but with
 678	 * non-blocking behaviour.
 679	 */
 680	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
 681		if (flags & IOMAP_NOWAIT)
 682			return -EAGAIN;
 683		mode = XFS_ILOCK_EXCL;
 684	}
 685
 686relock:
 687	if (flags & IOMAP_NOWAIT) {
 688		if (!xfs_ilock_nowait(ip, mode))
 689			return -EAGAIN;
 690	} else {
 691		xfs_ilock(ip, mode);
 692	}
 693
 694	/*
 695	 * The reflink iflag could have changed since the earlier unlocked
 696	 * check, so if we got ILOCK_SHARED for a write and but we're now a
 697	 * reflink inode we have to switch to ILOCK_EXCL and relock.
 698	 */
 699	if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
 700		xfs_iunlock(ip, mode);
 701		mode = XFS_ILOCK_EXCL;
 702		goto relock;
 703	}
 704
 705	*lockmode = mode;
 706	return 0;
 707}
 708
 709static int
 710xfs_direct_write_iomap_begin(
 711	struct inode		*inode,
 712	loff_t			offset,
 713	loff_t			length,
 714	unsigned		flags,
 715	struct iomap		*iomap,
 716	struct iomap		*srcmap)
 717{
 718	struct xfs_inode	*ip = XFS_I(inode);
 719	struct xfs_mount	*mp = ip->i_mount;
 720	struct xfs_bmbt_irec	imap, cmap;
 721	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 722	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
 723	int			nimaps = 1, error = 0;
 724	bool			shared = false;
 725	u16			iomap_flags = 0;
 726	unsigned		lockmode;
 727
 728	ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
 729
 730	if (XFS_FORCED_SHUTDOWN(mp))
 731		return -EIO;
 732
 733	/*
 734	 * Writes that span EOF might trigger an IO size update on completion,
 735	 * so consider them to be dirty for the purposes of O_DSYNC even if
 736	 * there is no other metadata changes pending or have been made here.
 737	 */
 738	if (offset + length > i_size_read(inode))
 739		iomap_flags |= IOMAP_F_DIRTY;
 740
 741	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
 742	if (error)
 743		return error;
 744
 745	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
 746			       &nimaps, 0);
 747	if (error)
 748		goto out_unlock;
 749
 750	if (imap_needs_cow(ip, flags, &imap, nimaps)) {
 751		error = -EAGAIN;
 752		if (flags & IOMAP_NOWAIT)
 753			goto out_unlock;
 754
 755		/* may drop and re-acquire the ilock */
 756		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
 757				&lockmode, flags & IOMAP_DIRECT);
 758		if (error)
 759			goto out_unlock;
 760		if (shared)
 761			goto out_found_cow;
 762		end_fsb = imap.br_startoff + imap.br_blockcount;
 763		length = XFS_FSB_TO_B(mp, end_fsb) - offset;
 764	}
 765
 766	if (imap_needs_alloc(inode, flags, &imap, nimaps))
 767		goto allocate_blocks;
 768
 769	xfs_iunlock(ip, lockmode);
 770	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
 771	return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags);
 772
 773allocate_blocks:
 774	error = -EAGAIN;
 775	if (flags & IOMAP_NOWAIT)
 776		goto out_unlock;
 777
 778	/*
 779	 * We cap the maximum length we map to a sane size  to keep the chunks
 780	 * of work done where somewhat symmetric with the work writeback does.
 781	 * This is a completely arbitrary number pulled out of thin air as a
 782	 * best guess for initial testing.
 783	 *
 784	 * Note that the values needs to be less than 32-bits wide until the
 785	 * lower level functions are updated.
 786	 */
 787	length = min_t(loff_t, length, 1024 * PAGE_SIZE);
 788	end_fsb = xfs_iomap_end_fsb(mp, offset, length);
 789
 790	if (offset + length > XFS_ISIZE(ip))
 791		end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
 792	else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
 793		end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
 794	xfs_iunlock(ip, lockmode);
 795
 796	error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
 797			&imap);
 798	if (error)
 799		return error;
 800
 801	trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
 802	return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags | IOMAP_F_NEW);
 803
 804out_found_cow:
 805	xfs_iunlock(ip, lockmode);
 806	length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
 807	trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
 808	if (imap.br_startblock != HOLESTARTBLOCK) {
 809		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
 810		if (error)
 811			return error;
 812	}
 813	return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
 814
 815out_unlock:
 816	xfs_iunlock(ip, lockmode);
 817	return error;
 818}
 819
 820const struct iomap_ops xfs_direct_write_iomap_ops = {
 821	.iomap_begin		= xfs_direct_write_iomap_begin,
 822};
 823
 824static int
 825xfs_buffered_write_iomap_begin(
 826	struct inode		*inode,
 827	loff_t			offset,
 828	loff_t			count,
 829	unsigned		flags,
 830	struct iomap		*iomap,
 831	struct iomap		*srcmap)
 832{
 833	struct xfs_inode	*ip = XFS_I(inode);
 834	struct xfs_mount	*mp = ip->i_mount;
 835	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
 836	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
 837	struct xfs_bmbt_irec	imap, cmap;
 838	struct xfs_iext_cursor	icur, ccur;
 839	xfs_fsblock_t		prealloc_blocks = 0;
 840	bool			eof = false, cow_eof = false, shared = false;
 841	int			allocfork = XFS_DATA_FORK;
 842	int			error = 0;
 843
 844	/* we can't use delayed allocations when using extent size hints */
 845	if (xfs_get_extsz_hint(ip))
 846		return xfs_direct_write_iomap_begin(inode, offset, count,
 847				flags, iomap, srcmap);
 848
 849	ASSERT(!XFS_IS_REALTIME_INODE(ip));
 850
 851	xfs_ilock(ip, XFS_ILOCK_EXCL);
 852
 853	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
 854	    XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
 855		error = -EFSCORRUPTED;
 856		goto out_unlock;
 857	}
 858
 859	XFS_STATS_INC(mp, xs_blk_mapw);
 860
 861	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
 862		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
 863		if (error)
 864			goto out_unlock;
 865	}
 866
 867	/*
 868	 * Search the data fork first to look up our source mapping.  We
 869	 * always need the data fork map, as we have to return it to the
 870	 * iomap code so that the higher level write code can read data in to
 871	 * perform read-modify-write cycles for unaligned writes.
 872	 */
 873	eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
 874	if (eof)
 875		imap.br_startoff = end_fsb; /* fake hole until the end */
 876
 877	/* We never need to allocate blocks for zeroing a hole. */
 878	if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
 879		xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
 880		goto out_unlock;
 881	}
 882
 883	/*
 884	 * Search the COW fork extent list even if we did not find a data fork
 885	 * extent.  This serves two purposes: first this implements the
 886	 * speculative preallocation using cowextsize, so that we also unshare
 887	 * block adjacent to shared blocks instead of just the shared blocks
 888	 * themselves.  Second the lookup in the extent list is generally faster
 889	 * than going out to the shared extent tree.
 890	 */
 891	if (xfs_is_cow_inode(ip)) {
 892		if (!ip->i_cowfp) {
 893			ASSERT(!xfs_is_reflink_inode(ip));
 894			xfs_ifork_init_cow(ip);
 895		}
 896		cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
 897				&ccur, &cmap);
 898		if (!cow_eof && cmap.br_startoff <= offset_fsb) {
 899			trace_xfs_reflink_cow_found(ip, &cmap);
 900			goto found_cow;
 901		}
 902	}
 903
 904	if (imap.br_startoff <= offset_fsb) {
 905		/*
 906		 * For reflink files we may need a delalloc reservation when
 907		 * overwriting shared extents.   This includes zeroing of
 908		 * existing extents that contain data.
 909		 */
 910		if (!xfs_is_cow_inode(ip) ||
 911		    ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
 912			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
 913					&imap);
 914			goto found_imap;
 915		}
 916
 917		xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
 918
 919		/* Trim the mapping to the nearest shared extent boundary. */
 920		error = xfs_bmap_trim_cow(ip, &imap, &shared);
 921		if (error)
 922			goto out_unlock;
 923
 924		/* Not shared?  Just report the (potentially capped) extent. */
 925		if (!shared) {
 926			trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
 927					&imap);
 928			goto found_imap;
 929		}
 930
 931		/*
 932		 * Fork all the shared blocks from our write offset until the
 933		 * end of the extent.
 934		 */
 935		allocfork = XFS_COW_FORK;
 936		end_fsb = imap.br_startoff + imap.br_blockcount;
 937	} else {
 938		/*
 939		 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
 940		 * pages to keep the chunks of work done where somewhat
 941		 * symmetric with the work writeback does.  This is a completely
 942		 * arbitrary number pulled out of thin air.
 943		 *
 944		 * Note that the values needs to be less than 32-bits wide until
 945		 * the lower level functions are updated.
 946		 */
 947		count = min_t(loff_t, count, 1024 * PAGE_SIZE);
 948		end_fsb = xfs_iomap_end_fsb(mp, offset, count);
 949
 950		if (xfs_is_always_cow_inode(ip))
 951			allocfork = XFS_COW_FORK;
 952	}
 953
 954	error = xfs_qm_dqattach_locked(ip, false);
 955	if (error)
 956		goto out_unlock;
 957
 958	if (eof && offset + count > XFS_ISIZE(ip)) {
 959		/*
 960		 * Determine the initial size of the preallocation.
 961		 * We clean up any extra preallocation when the file is closed.
 962		 */
 963		if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
 964			prealloc_blocks = mp->m_allocsize_blocks;
 965		else
 966			prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
 967						offset, count, &icur);
 968		if (prealloc_blocks) {
 969			xfs_extlen_t	align;
 970			xfs_off_t	end_offset;
 971			xfs_fileoff_t	p_end_fsb;
 972
 973			end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
 974			p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
 975					prealloc_blocks;
 976
 977			align = xfs_eof_alignment(ip);
 978			if (align)
 979				p_end_fsb = roundup_64(p_end_fsb, align);
 980
 981			p_end_fsb = min(p_end_fsb,
 982				XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
 983			ASSERT(p_end_fsb > offset_fsb);
 984			prealloc_blocks = p_end_fsb - end_fsb;
 985		}
 986	}
 987
 988retry:
 989	error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
 990			end_fsb - offset_fsb, prealloc_blocks,
 991			allocfork == XFS_DATA_FORK ? &imap : &cmap,
 992			allocfork == XFS_DATA_FORK ? &icur : &ccur,
 993			allocfork == XFS_DATA_FORK ? eof : cow_eof);
 994	switch (error) {
 995	case 0:
 996		break;
 997	case -ENOSPC:
 998	case -EDQUOT:
 999		/* retry without any preallocation */
1000		trace_xfs_delalloc_enospc(ip, offset, count);
1001		if (prealloc_blocks) {
1002			prealloc_blocks = 0;
1003			goto retry;
1004		}
1005		/*FALLTHRU*/
1006	default:
1007		goto out_unlock;
1008	}
1009
1010	if (allocfork == XFS_COW_FORK) {
1011		trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1012		goto found_cow;
1013	}
1014
1015	/*
1016	 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1017	 * them out if the write happens to fail.
1018	 */
1019	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1020	trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1021	return xfs_bmbt_to_iomap(ip, iomap, &imap, IOMAP_F_NEW);
1022
1023found_imap:
1024	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1025	return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1026
1027found_cow:
1028	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1029	if (imap.br_startoff <= offset_fsb) {
1030		error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
1031		if (error)
1032			return error;
1033	} else {
1034		xfs_trim_extent(&cmap, offset_fsb,
1035				imap.br_startoff - offset_fsb);
1036	}
1037	return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1038
1039out_unlock:
1040	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1041	return error;
1042}
1043
1044static int
1045xfs_buffered_write_iomap_end(
1046	struct inode		*inode,
1047	loff_t			offset,
1048	loff_t			length,
1049	ssize_t			written,
1050	unsigned		flags,
1051	struct iomap		*iomap)
1052{
1053	struct xfs_inode	*ip = XFS_I(inode);
1054	struct xfs_mount	*mp = ip->i_mount;
1055	xfs_fileoff_t		start_fsb;
1056	xfs_fileoff_t		end_fsb;
1057	int			error = 0;
1058
1059	if (iomap->type != IOMAP_DELALLOC)
1060		return 0;
1061
1062	/*
1063	 * Behave as if the write failed if drop writes is enabled. Set the NEW
1064	 * flag to force delalloc cleanup.
1065	 */
1066	if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1067		iomap->flags |= IOMAP_F_NEW;
1068		written = 0;
1069	}
1070
1071	/*
1072	 * start_fsb refers to the first unused block after a short write. If
1073	 * nothing was written, round offset down to point at the first block in
1074	 * the range.
1075	 */
1076	if (unlikely(!written))
1077		start_fsb = XFS_B_TO_FSBT(mp, offset);
1078	else
1079		start_fsb = XFS_B_TO_FSB(mp, offset + written);
1080	end_fsb = XFS_B_TO_FSB(mp, offset + length);
1081
1082	/*
1083	 * Trim delalloc blocks if they were allocated by this write and we
1084	 * didn't manage to write the whole range.
1085	 *
1086	 * We don't need to care about racing delalloc as we hold i_mutex
1087	 * across the reserve/allocate/unreserve calls. If there are delalloc
1088	 * blocks in the range, they are ours.
1089	 */
1090	if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1091		truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1092					 XFS_FSB_TO_B(mp, end_fsb) - 1);
1093
1094		error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1095					       end_fsb - start_fsb);
1096		if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1097			xfs_alert(mp, "%s: unable to clean up ino %lld",
1098				__func__, ip->i_ino);
1099			return error;
1100		}
1101	}
1102
1103	return 0;
1104}
1105
1106const struct iomap_ops xfs_buffered_write_iomap_ops = {
1107	.iomap_begin		= xfs_buffered_write_iomap_begin,
1108	.iomap_end		= xfs_buffered_write_iomap_end,
1109};
1110
1111static int
1112xfs_read_iomap_begin(
1113	struct inode		*inode,
1114	loff_t			offset,
1115	loff_t			length,
1116	unsigned		flags,
1117	struct iomap		*iomap,
1118	struct iomap		*srcmap)
1119{
1120	struct xfs_inode	*ip = XFS_I(inode);
1121	struct xfs_mount	*mp = ip->i_mount;
1122	struct xfs_bmbt_irec	imap;
1123	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1124	xfs_fileoff_t		end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1125	int			nimaps = 1, error = 0;
1126	bool			shared = false;
1127	unsigned		lockmode;
1128
1129	ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1130
1131	if (XFS_FORCED_SHUTDOWN(mp))
1132		return -EIO;
1133
1134	error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1135	if (error)
1136		return error;
1137	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1138			       &nimaps, 0);
1139	if (!error && (flags & IOMAP_REPORT))
1140		error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1141	xfs_iunlock(ip, lockmode);
1142
1143	if (error)
1144		return error;
1145	trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1146	return xfs_bmbt_to_iomap(ip, iomap, &imap, shared ? IOMAP_F_SHARED : 0);
1147}
1148
1149const struct iomap_ops xfs_read_iomap_ops = {
1150	.iomap_begin		= xfs_read_iomap_begin,
1151};
1152
1153static int
1154xfs_seek_iomap_begin(
1155	struct inode		*inode,
1156	loff_t			offset,
1157	loff_t			length,
1158	unsigned		flags,
1159	struct iomap		*iomap,
1160	struct iomap		*srcmap)
1161{
1162	struct xfs_inode	*ip = XFS_I(inode);
1163	struct xfs_mount	*mp = ip->i_mount;
1164	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1165	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1166	xfs_fileoff_t		cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1167	struct xfs_iext_cursor	icur;
1168	struct xfs_bmbt_irec	imap, cmap;
1169	int			error = 0;
1170	unsigned		lockmode;
1171
1172	if (XFS_FORCED_SHUTDOWN(mp))
1173		return -EIO;
1174
1175	lockmode = xfs_ilock_data_map_shared(ip);
1176	if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
1177		error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1178		if (error)
1179			goto out_unlock;
1180	}
1181
1182	if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1183		/*
1184		 * If we found a data extent we are done.
1185		 */
1186		if (imap.br_startoff <= offset_fsb)
1187			goto done;
1188		data_fsb = imap.br_startoff;
1189	} else {
1190		/*
1191		 * Fake a hole until the end of the file.
1192		 */
1193		data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1194	}
1195
1196	/*
1197	 * If a COW fork extent covers the hole, report it - capped to the next
1198	 * data fork extent:
1199	 */
1200	if (xfs_inode_has_cow_data(ip) &&
1201	    xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1202		cow_fsb = cmap.br_startoff;
1203	if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1204		if (data_fsb < cow_fsb + cmap.br_blockcount)
1205			end_fsb = min(end_fsb, data_fsb);
1206		xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1207		error = xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1208		/*
1209		 * This is a COW extent, so we must probe the page cache
1210		 * because there could be dirty page cache being backed
1211		 * by this extent.
1212		 */
1213		iomap->type = IOMAP_UNWRITTEN;
1214		goto out_unlock;
1215	}
1216
1217	/*
1218	 * Else report a hole, capped to the next found data or COW extent.
1219	 */
1220	if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1221		imap.br_blockcount = cow_fsb - offset_fsb;
1222	else
1223		imap.br_blockcount = data_fsb - offset_fsb;
1224	imap.br_startoff = offset_fsb;
1225	imap.br_startblock = HOLESTARTBLOCK;
1226	imap.br_state = XFS_EXT_NORM;
1227done:
1228	xfs_trim_extent(&imap, offset_fsb, end_fsb);
1229	error = xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1230out_unlock:
1231	xfs_iunlock(ip, lockmode);
1232	return error;
1233}
1234
1235const struct iomap_ops xfs_seek_iomap_ops = {
1236	.iomap_begin		= xfs_seek_iomap_begin,
1237};
1238
1239static int
1240xfs_xattr_iomap_begin(
1241	struct inode		*inode,
1242	loff_t			offset,
1243	loff_t			length,
1244	unsigned		flags,
1245	struct iomap		*iomap,
1246	struct iomap		*srcmap)
1247{
1248	struct xfs_inode	*ip = XFS_I(inode);
1249	struct xfs_mount	*mp = ip->i_mount;
1250	xfs_fileoff_t		offset_fsb = XFS_B_TO_FSBT(mp, offset);
1251	xfs_fileoff_t		end_fsb = XFS_B_TO_FSB(mp, offset + length);
1252	struct xfs_bmbt_irec	imap;
1253	int			nimaps = 1, error = 0;
1254	unsigned		lockmode;
1255
1256	if (XFS_FORCED_SHUTDOWN(mp))
1257		return -EIO;
1258
1259	lockmode = xfs_ilock_attr_map_shared(ip);
1260
1261	/* if there are no attribute fork or extents, return ENOENT */
1262	if (!XFS_IFORK_Q(ip) || !ip->i_afp->if_nextents) {
1263		error = -ENOENT;
1264		goto out_unlock;
1265	}
1266
1267	ASSERT(ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL);
1268	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1269			       &nimaps, XFS_BMAPI_ATTRFORK);
1270out_unlock:
1271	xfs_iunlock(ip, lockmode);
1272
1273	if (error)
1274		return error;
1275	ASSERT(nimaps);
1276	return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1277}
1278
1279const struct iomap_ops xfs_xattr_iomap_ops = {
1280	.iomap_begin		= xfs_xattr_iomap_begin,
1281};