xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision d78c317f)
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_bit.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_bmap.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_itable.h"
39 #include "xfs_rw.h"
40 #include "xfs_attr.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_utils.h"
44 #include "xfs_iomap.h"
45 #include "xfs_trace.h"
46 
47 
48 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
49 						<< mp->m_writeio_log)
50 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
51 
52 STATIC int
53 xfs_iomap_eof_align_last_fsb(
54 	xfs_mount_t	*mp,
55 	xfs_inode_t	*ip,
56 	xfs_extlen_t	extsize,
57 	xfs_fileoff_t	*last_fsb)
58 {
59 	xfs_fileoff_t	new_last_fsb = 0;
60 	xfs_extlen_t	align = 0;
61 	int		eof, error;
62 
63 	if (!XFS_IS_REALTIME_INODE(ip)) {
64 		/*
65 		 * Round up the allocation request to a stripe unit
66 		 * (m_dalign) boundary if the file size is >= stripe unit
67 		 * size, and we are allocating past the allocation eof.
68 		 *
69 		 * If mounted with the "-o swalloc" option the alignment is
70 		 * increased from the strip unit size to the stripe width.
71 		 */
72 		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
73 			align = mp->m_swidth;
74 		else if (mp->m_dalign)
75 			align = mp->m_dalign;
76 
77 		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
78 			new_last_fsb = roundup_64(*last_fsb, align);
79 	}
80 
81 	/*
82 	 * Always round up the allocation request to an extent boundary
83 	 * (when file on a real-time subvolume or has di_extsize hint).
84 	 */
85 	if (extsize) {
86 		if (new_last_fsb)
87 			align = roundup_64(new_last_fsb, extsize);
88 		else
89 			align = extsize;
90 		new_last_fsb = roundup_64(*last_fsb, align);
91 	}
92 
93 	if (new_last_fsb) {
94 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
95 		if (error)
96 			return error;
97 		if (eof)
98 			*last_fsb = new_last_fsb;
99 	}
100 	return 0;
101 }
102 
103 STATIC int
104 xfs_alert_fsblock_zero(
105 	xfs_inode_t	*ip,
106 	xfs_bmbt_irec_t	*imap)
107 {
108 	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
109 			"Access to block zero in inode %llu "
110 			"start_block: %llx start_off: %llx "
111 			"blkcnt: %llx extent-state: %x\n",
112 		(unsigned long long)ip->i_ino,
113 		(unsigned long long)imap->br_startblock,
114 		(unsigned long long)imap->br_startoff,
115 		(unsigned long long)imap->br_blockcount,
116 		imap->br_state);
117 	return EFSCORRUPTED;
118 }
119 
120 int
121 xfs_iomap_write_direct(
122 	xfs_inode_t	*ip,
123 	xfs_off_t	offset,
124 	size_t		count,
125 	xfs_bmbt_irec_t *imap,
126 	int		nmaps)
127 {
128 	xfs_mount_t	*mp = ip->i_mount;
129 	xfs_fileoff_t	offset_fsb;
130 	xfs_fileoff_t	last_fsb;
131 	xfs_filblks_t	count_fsb, resaligned;
132 	xfs_fsblock_t	firstfsb;
133 	xfs_extlen_t	extsz, temp;
134 	int		nimaps;
135 	int		bmapi_flag;
136 	int		quota_flag;
137 	int		rt;
138 	xfs_trans_t	*tp;
139 	xfs_bmap_free_t free_list;
140 	uint		qblocks, resblks, resrtextents;
141 	int		committed;
142 	int		error;
143 
144 	/*
145 	 * Make sure that the dquots are there. This doesn't hold
146 	 * the ilock across a disk read.
147 	 */
148 	error = xfs_qm_dqattach_locked(ip, 0);
149 	if (error)
150 		return XFS_ERROR(error);
151 
152 	rt = XFS_IS_REALTIME_INODE(ip);
153 	extsz = xfs_get_extsz_hint(ip);
154 
155 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
156 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
157 	if ((offset + count) > XFS_ISIZE(ip)) {
158 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
159 		if (error)
160 			goto error_out;
161 	} else {
162 		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
163 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
164 					imap->br_blockcount +
165 					imap->br_startoff);
166 	}
167 	count_fsb = last_fsb - offset_fsb;
168 	ASSERT(count_fsb > 0);
169 
170 	resaligned = count_fsb;
171 	if (unlikely(extsz)) {
172 		if ((temp = do_mod(offset_fsb, extsz)))
173 			resaligned += temp;
174 		if ((temp = do_mod(resaligned, extsz)))
175 			resaligned += extsz - temp;
176 	}
177 
178 	if (unlikely(rt)) {
179 		resrtextents = qblocks = resaligned;
180 		resrtextents /= mp->m_sb.sb_rextsize;
181 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
182 		quota_flag = XFS_QMOPT_RES_RTBLKS;
183 	} else {
184 		resrtextents = 0;
185 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
186 		quota_flag = XFS_QMOPT_RES_REGBLKS;
187 	}
188 
189 	/*
190 	 * Allocate and setup the transaction
191 	 */
192 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
193 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
194 	error = xfs_trans_reserve(tp, resblks,
195 			XFS_WRITE_LOG_RES(mp), resrtextents,
196 			XFS_TRANS_PERM_LOG_RES,
197 			XFS_WRITE_LOG_COUNT);
198 	/*
199 	 * Check for running out of space, note: need lock to return
200 	 */
201 	if (error)
202 		xfs_trans_cancel(tp, 0);
203 	xfs_ilock(ip, XFS_ILOCK_EXCL);
204 	if (error)
205 		goto error_out;
206 
207 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
208 	if (error)
209 		goto error1;
210 
211 	xfs_trans_ijoin(tp, ip, 0);
212 
213 	bmapi_flag = 0;
214 	if (offset < XFS_ISIZE(ip) || extsz)
215 		bmapi_flag |= XFS_BMAPI_PREALLOC;
216 
217 	/*
218 	 * From this point onwards we overwrite the imap pointer that the
219 	 * caller gave to us.
220 	 */
221 	xfs_bmap_init(&free_list, &firstfsb);
222 	nimaps = 1;
223 	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
224 				&firstfsb, 0, imap, &nimaps, &free_list);
225 	if (error)
226 		goto error0;
227 
228 	/*
229 	 * Complete the transaction
230 	 */
231 	error = xfs_bmap_finish(&tp, &free_list, &committed);
232 	if (error)
233 		goto error0;
234 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
235 	if (error)
236 		goto error_out;
237 
238 	/*
239 	 * Copy any maps to caller's array and return any error.
240 	 */
241 	if (nimaps == 0) {
242 		error = ENOSPC;
243 		goto error_out;
244 	}
245 
246 	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) {
247 		error = xfs_alert_fsblock_zero(ip, imap);
248 		goto error_out;
249 	}
250 
251 	return 0;
252 
253 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
254 	xfs_bmap_cancel(&free_list);
255 	xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
256 
257 error1:	/* Just cancel transaction */
258 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
259 
260 error_out:
261 	return XFS_ERROR(error);
262 }
263 
264 /*
265  * If the caller is doing a write at the end of the file, then extend the
266  * allocation out to the file system's write iosize.  We clean up any extra
267  * space left over when the file is closed in xfs_inactive().
268  *
269  * If we find we already have delalloc preallocation beyond EOF, don't do more
270  * preallocation as it it not needed.
271  */
272 STATIC int
273 xfs_iomap_eof_want_preallocate(
274 	xfs_mount_t	*mp,
275 	xfs_inode_t	*ip,
276 	xfs_off_t	offset,
277 	size_t		count,
278 	xfs_bmbt_irec_t *imap,
279 	int		nimaps,
280 	int		*prealloc)
281 {
282 	xfs_fileoff_t   start_fsb;
283 	xfs_filblks_t   count_fsb;
284 	xfs_fsblock_t	firstblock;
285 	int		n, error, imaps;
286 	int		found_delalloc = 0;
287 
288 	*prealloc = 0;
289 	if (offset + count <= XFS_ISIZE(ip))
290 		return 0;
291 
292 	/*
293 	 * If there are any real blocks past eof, then don't
294 	 * do any speculative allocation.
295 	 */
296 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
297 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
298 	while (count_fsb > 0) {
299 		imaps = nimaps;
300 		firstblock = NULLFSBLOCK;
301 		error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
302 				       0);
303 		if (error)
304 			return error;
305 		for (n = 0; n < imaps; n++) {
306 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
307 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
308 				return 0;
309 			start_fsb += imap[n].br_blockcount;
310 			count_fsb -= imap[n].br_blockcount;
311 
312 			if (imap[n].br_startblock == DELAYSTARTBLOCK)
313 				found_delalloc = 1;
314 		}
315 	}
316 	if (!found_delalloc)
317 		*prealloc = 1;
318 	return 0;
319 }
320 
321 /*
322  * If we don't have a user specified preallocation size, dynamically increase
323  * the preallocation size as the size of the file grows. Cap the maximum size
324  * at a single extent or less if the filesystem is near full. The closer the
325  * filesystem is to full, the smaller the maximum prealocation.
326  */
327 STATIC xfs_fsblock_t
328 xfs_iomap_prealloc_size(
329 	struct xfs_mount	*mp,
330 	struct xfs_inode	*ip)
331 {
332 	xfs_fsblock_t		alloc_blocks = 0;
333 
334 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
335 		int shift = 0;
336 		int64_t freesp;
337 
338 		/*
339 		 * rounddown_pow_of_two() returns an undefined result
340 		 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
341 		 * ensure we always pass in a non-zero value.
342 		 */
343 		alloc_blocks = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)) + 1;
344 		alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
345 					rounddown_pow_of_two(alloc_blocks));
346 
347 		xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
348 		freesp = mp->m_sb.sb_fdblocks;
349 		if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
350 			shift = 2;
351 			if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
352 				shift++;
353 			if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
354 				shift++;
355 			if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
356 				shift++;
357 			if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
358 				shift++;
359 		}
360 		if (shift)
361 			alloc_blocks >>= shift;
362 	}
363 
364 	if (alloc_blocks < mp->m_writeio_blocks)
365 		alloc_blocks = mp->m_writeio_blocks;
366 
367 	return alloc_blocks;
368 }
369 
370 int
371 xfs_iomap_write_delay(
372 	xfs_inode_t	*ip,
373 	xfs_off_t	offset,
374 	size_t		count,
375 	xfs_bmbt_irec_t *ret_imap)
376 {
377 	xfs_mount_t	*mp = ip->i_mount;
378 	xfs_fileoff_t	offset_fsb;
379 	xfs_fileoff_t	last_fsb;
380 	xfs_off_t	aligned_offset;
381 	xfs_fileoff_t	ioalign;
382 	xfs_extlen_t	extsz;
383 	int		nimaps;
384 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
385 	int		prealloc, flushed = 0;
386 	int		error;
387 
388 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
389 
390 	/*
391 	 * Make sure that the dquots are there. This doesn't hold
392 	 * the ilock across a disk read.
393 	 */
394 	error = xfs_qm_dqattach_locked(ip, 0);
395 	if (error)
396 		return XFS_ERROR(error);
397 
398 	extsz = xfs_get_extsz_hint(ip);
399 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
400 
401 
402 	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
403 				imap, XFS_WRITE_IMAPS, &prealloc);
404 	if (error)
405 		return error;
406 
407 retry:
408 	if (prealloc) {
409 		xfs_fsblock_t	alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
410 
411 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
412 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
413 		last_fsb = ioalign + alloc_blocks;
414 	} else {
415 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
416 	}
417 
418 	if (prealloc || extsz) {
419 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
420 		if (error)
421 			return error;
422 	}
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  */
476 int
477 xfs_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 
625 trans_cancel:
626 	xfs_bmap_cancel(&free_list);
627 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
628 error0:
629 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
630 	return XFS_ERROR(error);
631 }
632 
633 int
634 xfs_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 	uint		resblks;
649 	int		committed;
650 	int		error;
651 
652 	trace_xfs_unwritten_convert(ip, offset, count);
653 
654 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
655 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
656 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
657 
658 	/*
659 	 * Reserve enough blocks in this transaction for two complete extent
660 	 * btree splits.  We may be converting the middle part of an unwritten
661 	 * extent and in this case we will insert two new extents in the btree
662 	 * each of which could cause a full split.
663 	 *
664 	 * This reservation amount will be used in the first call to
665 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
666 	 * rest of the operation.
667 	 */
668 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
669 
670 	do {
671 		/*
672 		 * set up a transaction to convert the range of extents
673 		 * from unwritten to real. Do allocations in a loop until
674 		 * we have covered the range passed in.
675 		 *
676 		 * Note that we open code the transaction allocation here
677 		 * to pass KM_NOFS--we can't risk to recursing back into
678 		 * the filesystem here as we might be asked to write out
679 		 * the same inode that we complete here and might deadlock
680 		 * on the iolock.
681 		 */
682 		xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
683 		tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
684 		tp->t_flags |= XFS_TRANS_RESERVE;
685 		error = xfs_trans_reserve(tp, resblks,
686 				XFS_WRITE_LOG_RES(mp), 0,
687 				XFS_TRANS_PERM_LOG_RES,
688 				XFS_WRITE_LOG_COUNT);
689 		if (error) {
690 			xfs_trans_cancel(tp, 0);
691 			return XFS_ERROR(error);
692 		}
693 
694 		xfs_ilock(ip, XFS_ILOCK_EXCL);
695 		xfs_trans_ijoin(tp, ip, 0);
696 
697 		/*
698 		 * Modify the unwritten extent state of the buffer.
699 		 */
700 		xfs_bmap_init(&free_list, &firstfsb);
701 		nimaps = 1;
702 		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
703 				  XFS_BMAPI_CONVERT, &firstfsb,
704 				  1, &imap, &nimaps, &free_list);
705 		if (error)
706 			goto error_on_bmapi_transaction;
707 
708 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
709 		if (error)
710 			goto error_on_bmapi_transaction;
711 
712 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
713 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
714 		if (error)
715 			return XFS_ERROR(error);
716 
717 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
718 			return xfs_alert_fsblock_zero(ip, &imap);
719 
720 		if ((numblks_fsb = imap.br_blockcount) == 0) {
721 			/*
722 			 * The numblks_fsb value should always get
723 			 * smaller, otherwise the loop is stuck.
724 			 */
725 			ASSERT(imap.br_blockcount);
726 			break;
727 		}
728 		offset_fsb += numblks_fsb;
729 		count_fsb -= numblks_fsb;
730 	} while (count_fsb > 0);
731 
732 	return 0;
733 
734 error_on_bmapi_transaction:
735 	xfs_bmap_cancel(&free_list);
736 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
737 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
738 	return XFS_ERROR(error);
739 }
740