xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision b34e08d5)
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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_bmap.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_iomap.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_quota.h"
39 #include "xfs_dquot_item.h"
40 #include "xfs_dquot.h"
41 #include "xfs_dinode.h"
42 
43 
44 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
45 						<< mp->m_writeio_log)
46 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
47 
48 STATIC int
49 xfs_iomap_eof_align_last_fsb(
50 	xfs_mount_t	*mp,
51 	xfs_inode_t	*ip,
52 	xfs_extlen_t	extsize,
53 	xfs_fileoff_t	*last_fsb)
54 {
55 	xfs_fileoff_t	new_last_fsb = 0;
56 	xfs_extlen_t	align = 0;
57 	int		eof, error;
58 
59 	if (!XFS_IS_REALTIME_INODE(ip)) {
60 		/*
61 		 * Round up the allocation request to a stripe unit
62 		 * (m_dalign) boundary if the file size is >= stripe unit
63 		 * size, and we are allocating past the allocation eof.
64 		 *
65 		 * If mounted with the "-o swalloc" option the alignment is
66 		 * increased from the strip unit size to the stripe width.
67 		 */
68 		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
69 			align = mp->m_swidth;
70 		else if (mp->m_dalign)
71 			align = mp->m_dalign;
72 
73 		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
74 			new_last_fsb = roundup_64(*last_fsb, align);
75 	}
76 
77 	/*
78 	 * Always round up the allocation request to an extent boundary
79 	 * (when file on a real-time subvolume or has di_extsize hint).
80 	 */
81 	if (extsize) {
82 		if (new_last_fsb)
83 			align = roundup_64(new_last_fsb, extsize);
84 		else
85 			align = extsize;
86 		new_last_fsb = roundup_64(*last_fsb, align);
87 	}
88 
89 	if (new_last_fsb) {
90 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
91 		if (error)
92 			return error;
93 		if (eof)
94 			*last_fsb = new_last_fsb;
95 	}
96 	return 0;
97 }
98 
99 STATIC int
100 xfs_alert_fsblock_zero(
101 	xfs_inode_t	*ip,
102 	xfs_bmbt_irec_t	*imap)
103 {
104 	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
105 			"Access to block zero in inode %llu "
106 			"start_block: %llx start_off: %llx "
107 			"blkcnt: %llx extent-state: %x",
108 		(unsigned long long)ip->i_ino,
109 		(unsigned long long)imap->br_startblock,
110 		(unsigned long long)imap->br_startoff,
111 		(unsigned long long)imap->br_blockcount,
112 		imap->br_state);
113 	return EFSCORRUPTED;
114 }
115 
116 int
117 xfs_iomap_write_direct(
118 	xfs_inode_t	*ip,
119 	xfs_off_t	offset,
120 	size_t		count,
121 	xfs_bmbt_irec_t *imap,
122 	int		nmaps)
123 {
124 	xfs_mount_t	*mp = ip->i_mount;
125 	xfs_fileoff_t	offset_fsb;
126 	xfs_fileoff_t	last_fsb;
127 	xfs_filblks_t	count_fsb, resaligned;
128 	xfs_fsblock_t	firstfsb;
129 	xfs_extlen_t	extsz, temp;
130 	int		nimaps;
131 	int		quota_flag;
132 	int		rt;
133 	xfs_trans_t	*tp;
134 	xfs_bmap_free_t free_list;
135 	uint		qblocks, resblks, resrtextents;
136 	int		committed;
137 	int		error;
138 
139 	error = xfs_qm_dqattach(ip, 0);
140 	if (error)
141 		return XFS_ERROR(error);
142 
143 	rt = XFS_IS_REALTIME_INODE(ip);
144 	extsz = xfs_get_extsz_hint(ip);
145 
146 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
147 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
148 	if ((offset + count) > XFS_ISIZE(ip)) {
149 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
150 		if (error)
151 			return XFS_ERROR(error);
152 	} else {
153 		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
154 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
155 					imap->br_blockcount +
156 					imap->br_startoff);
157 	}
158 	count_fsb = last_fsb - offset_fsb;
159 	ASSERT(count_fsb > 0);
160 
161 	resaligned = count_fsb;
162 	if (unlikely(extsz)) {
163 		if ((temp = do_mod(offset_fsb, extsz)))
164 			resaligned += temp;
165 		if ((temp = do_mod(resaligned, extsz)))
166 			resaligned += extsz - temp;
167 	}
168 
169 	if (unlikely(rt)) {
170 		resrtextents = qblocks = resaligned;
171 		resrtextents /= mp->m_sb.sb_rextsize;
172 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
173 		quota_flag = XFS_QMOPT_RES_RTBLKS;
174 	} else {
175 		resrtextents = 0;
176 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
177 		quota_flag = XFS_QMOPT_RES_REGBLKS;
178 	}
179 
180 	/*
181 	 * Allocate and setup the transaction
182 	 */
183 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
184 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
185 				  resblks, resrtextents);
186 	/*
187 	 * Check for running out of space, note: need lock to return
188 	 */
189 	if (error) {
190 		xfs_trans_cancel(tp, 0);
191 		return XFS_ERROR(error);
192 	}
193 
194 	xfs_ilock(ip, XFS_ILOCK_EXCL);
195 
196 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
197 	if (error)
198 		goto out_trans_cancel;
199 
200 	xfs_trans_ijoin(tp, ip, 0);
201 
202 	/*
203 	 * From this point onwards we overwrite the imap pointer that the
204 	 * caller gave to us.
205 	 */
206 	xfs_bmap_init(&free_list, &firstfsb);
207 	nimaps = 1;
208 	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
209 				XFS_BMAPI_PREALLOC, &firstfsb, 0,
210 				imap, &nimaps, &free_list);
211 	if (error)
212 		goto out_bmap_cancel;
213 
214 	/*
215 	 * Complete the transaction
216 	 */
217 	error = xfs_bmap_finish(&tp, &free_list, &committed);
218 	if (error)
219 		goto out_bmap_cancel;
220 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
221 	if (error)
222 		goto out_unlock;
223 
224 	/*
225 	 * Copy any maps to caller's array and return any error.
226 	 */
227 	if (nimaps == 0) {
228 		error = XFS_ERROR(ENOSPC);
229 		goto out_unlock;
230 	}
231 
232 	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
233 		error = xfs_alert_fsblock_zero(ip, imap);
234 
235 out_unlock:
236 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
237 	return error;
238 
239 out_bmap_cancel:
240 	xfs_bmap_cancel(&free_list);
241 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
242 out_trans_cancel:
243 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
244 	goto out_unlock;
245 }
246 
247 /*
248  * If the caller is doing a write at the end of the file, then extend the
249  * allocation out to the file system's write iosize.  We clean up any extra
250  * space left over when the file is closed in xfs_inactive().
251  *
252  * If we find we already have delalloc preallocation beyond EOF, don't do more
253  * preallocation as it it not needed.
254  */
255 STATIC int
256 xfs_iomap_eof_want_preallocate(
257 	xfs_mount_t	*mp,
258 	xfs_inode_t	*ip,
259 	xfs_off_t	offset,
260 	size_t		count,
261 	xfs_bmbt_irec_t *imap,
262 	int		nimaps,
263 	int		*prealloc)
264 {
265 	xfs_fileoff_t   start_fsb;
266 	xfs_filblks_t   count_fsb;
267 	xfs_fsblock_t	firstblock;
268 	int		n, error, imaps;
269 	int		found_delalloc = 0;
270 
271 	*prealloc = 0;
272 	if (offset + count <= XFS_ISIZE(ip))
273 		return 0;
274 
275 	/*
276 	 * If the file is smaller than the minimum prealloc and we are using
277 	 * dynamic preallocation, don't do any preallocation at all as it is
278 	 * likely this is the only write to the file that is going to be done.
279 	 */
280 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
281 	    XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
282 		return 0;
283 
284 	/*
285 	 * If there are any real blocks past eof, then don't
286 	 * do any speculative allocation.
287 	 */
288 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
289 	count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
290 	while (count_fsb > 0) {
291 		imaps = nimaps;
292 		firstblock = NULLFSBLOCK;
293 		error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
294 				       0);
295 		if (error)
296 			return error;
297 		for (n = 0; n < imaps; n++) {
298 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
299 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
300 				return 0;
301 			start_fsb += imap[n].br_blockcount;
302 			count_fsb -= imap[n].br_blockcount;
303 
304 			if (imap[n].br_startblock == DELAYSTARTBLOCK)
305 				found_delalloc = 1;
306 		}
307 	}
308 	if (!found_delalloc)
309 		*prealloc = 1;
310 	return 0;
311 }
312 
313 /*
314  * Determine the initial size of the preallocation. We are beyond the current
315  * EOF here, but we need to take into account whether this is a sparse write or
316  * an extending write when determining the preallocation size.  Hence we need to
317  * look up the extent that ends at the current write offset and use the result
318  * to determine the preallocation size.
319  *
320  * If the extent is a hole, then preallocation is essentially disabled.
321  * Otherwise we take the size of the preceeding data extent as the basis for the
322  * preallocation size. If the size of the extent is greater than half the
323  * maximum extent length, then use the current offset as the basis. This ensures
324  * that for large files the preallocation size always extends to MAXEXTLEN
325  * rather than falling short due to things like stripe unit/width alignment of
326  * real extents.
327  */
328 STATIC xfs_fsblock_t
329 xfs_iomap_eof_prealloc_initial_size(
330 	struct xfs_mount	*mp,
331 	struct xfs_inode	*ip,
332 	xfs_off_t		offset,
333 	xfs_bmbt_irec_t		*imap,
334 	int			nimaps)
335 {
336 	xfs_fileoff_t   start_fsb;
337 	int		imaps = 1;
338 	int		error;
339 
340 	ASSERT(nimaps >= imaps);
341 
342 	/* if we are using a specific prealloc size, return now */
343 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
344 		return 0;
345 
346 	/* If the file is small, then use the minimum prealloc */
347 	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
348 		return 0;
349 
350 	/*
351 	 * As we write multiple pages, the offset will always align to the
352 	 * start of a page and hence point to a hole at EOF. i.e. if the size is
353 	 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
354 	 * will return FSB 1. Hence if there are blocks in the file, we want to
355 	 * point to the block prior to the EOF block and not the hole that maps
356 	 * directly at @offset.
357 	 */
358 	start_fsb = XFS_B_TO_FSB(mp, offset);
359 	if (start_fsb)
360 		start_fsb--;
361 	error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
362 	if (error)
363 		return 0;
364 
365 	ASSERT(imaps == 1);
366 	if (imap[0].br_startblock == HOLESTARTBLOCK)
367 		return 0;
368 	if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
369 		return imap[0].br_blockcount << 1;
370 	return XFS_B_TO_FSB(mp, offset);
371 }
372 
373 STATIC bool
374 xfs_quota_need_throttle(
375 	struct xfs_inode *ip,
376 	int type,
377 	xfs_fsblock_t alloc_blocks)
378 {
379 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
380 
381 	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
382 		return false;
383 
384 	/* no hi watermark, no throttle */
385 	if (!dq->q_prealloc_hi_wmark)
386 		return false;
387 
388 	/* under the lo watermark, no throttle */
389 	if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
390 		return false;
391 
392 	return true;
393 }
394 
395 STATIC void
396 xfs_quota_calc_throttle(
397 	struct xfs_inode *ip,
398 	int type,
399 	xfs_fsblock_t *qblocks,
400 	int *qshift)
401 {
402 	int64_t freesp;
403 	int shift = 0;
404 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
405 
406 	/* over hi wmark, squash the prealloc completely */
407 	if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
408 		*qblocks = 0;
409 		return;
410 	}
411 
412 	freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
413 	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
414 		shift = 2;
415 		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
416 			shift += 2;
417 		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
418 			shift += 2;
419 	}
420 
421 	/* only overwrite the throttle values if we are more aggressive */
422 	if ((freesp >> shift) < (*qblocks >> *qshift)) {
423 		*qblocks = freesp;
424 		*qshift = shift;
425 	}
426 }
427 
428 /*
429  * If we don't have a user specified preallocation size, dynamically increase
430  * the preallocation size as the size of the file grows. Cap the maximum size
431  * at a single extent or less if the filesystem is near full. The closer the
432  * filesystem is to full, the smaller the maximum prealocation.
433  */
434 STATIC xfs_fsblock_t
435 xfs_iomap_prealloc_size(
436 	struct xfs_mount	*mp,
437 	struct xfs_inode	*ip,
438 	xfs_off_t		offset,
439 	struct xfs_bmbt_irec	*imap,
440 	int			nimaps)
441 {
442 	xfs_fsblock_t		alloc_blocks = 0;
443 	int			shift = 0;
444 	int64_t			freesp;
445 	xfs_fsblock_t		qblocks;
446 	int			qshift = 0;
447 
448 	alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
449 							   imap, nimaps);
450 	if (!alloc_blocks)
451 		goto check_writeio;
452 	qblocks = alloc_blocks;
453 
454 	/*
455 	 * MAXEXTLEN is not a power of two value but we round the prealloc down
456 	 * to the nearest power of two value after throttling. To prevent the
457 	 * round down from unconditionally reducing the maximum supported prealloc
458 	 * size, we round up first, apply appropriate throttling, round down and
459 	 * cap the value to MAXEXTLEN.
460 	 */
461 	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
462 				       alloc_blocks);
463 
464 	xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
465 	freesp = mp->m_sb.sb_fdblocks;
466 	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
467 		shift = 2;
468 		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
469 			shift++;
470 		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
471 			shift++;
472 		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
473 			shift++;
474 		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
475 			shift++;
476 	}
477 
478 	/*
479 	 * Check each quota to cap the prealloc size and provide a shift
480 	 * value to throttle with.
481 	 */
482 	if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
483 		xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift);
484 	if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
485 		xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift);
486 	if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
487 		xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift);
488 
489 	/*
490 	 * The final prealloc size is set to the minimum of free space available
491 	 * in each of the quotas and the overall filesystem.
492 	 *
493 	 * The shift throttle value is set to the maximum value as determined by
494 	 * the global low free space values and per-quota low free space values.
495 	 */
496 	alloc_blocks = MIN(alloc_blocks, qblocks);
497 	shift = MAX(shift, qshift);
498 
499 	if (shift)
500 		alloc_blocks >>= shift;
501 	/*
502 	 * rounddown_pow_of_two() returns an undefined result if we pass in
503 	 * alloc_blocks = 0.
504 	 */
505 	if (alloc_blocks)
506 		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
507 	if (alloc_blocks > MAXEXTLEN)
508 		alloc_blocks = MAXEXTLEN;
509 
510 	/*
511 	 * If we are still trying to allocate more space than is
512 	 * available, squash the prealloc hard. This can happen if we
513 	 * have a large file on a small filesystem and the above
514 	 * lowspace thresholds are smaller than MAXEXTLEN.
515 	 */
516 	while (alloc_blocks && alloc_blocks >= freesp)
517 		alloc_blocks >>= 4;
518 
519 check_writeio:
520 	if (alloc_blocks < mp->m_writeio_blocks)
521 		alloc_blocks = mp->m_writeio_blocks;
522 
523 	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
524 				      mp->m_writeio_blocks);
525 
526 	return alloc_blocks;
527 }
528 
529 int
530 xfs_iomap_write_delay(
531 	xfs_inode_t	*ip,
532 	xfs_off_t	offset,
533 	size_t		count,
534 	xfs_bmbt_irec_t *ret_imap)
535 {
536 	xfs_mount_t	*mp = ip->i_mount;
537 	xfs_fileoff_t	offset_fsb;
538 	xfs_fileoff_t	last_fsb;
539 	xfs_off_t	aligned_offset;
540 	xfs_fileoff_t	ioalign;
541 	xfs_extlen_t	extsz;
542 	int		nimaps;
543 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
544 	int		prealloc;
545 	int		error;
546 
547 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
548 
549 	/*
550 	 * Make sure that the dquots are there. This doesn't hold
551 	 * the ilock across a disk read.
552 	 */
553 	error = xfs_qm_dqattach_locked(ip, 0);
554 	if (error)
555 		return XFS_ERROR(error);
556 
557 	extsz = xfs_get_extsz_hint(ip);
558 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
559 
560 	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
561 				imap, XFS_WRITE_IMAPS, &prealloc);
562 	if (error)
563 		return error;
564 
565 retry:
566 	if (prealloc) {
567 		xfs_fsblock_t	alloc_blocks;
568 
569 		alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
570 						       XFS_WRITE_IMAPS);
571 
572 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
573 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
574 		last_fsb = ioalign + alloc_blocks;
575 	} else {
576 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
577 	}
578 
579 	if (prealloc || extsz) {
580 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
581 		if (error)
582 			return error;
583 	}
584 
585 	/*
586 	 * Make sure preallocation does not create extents beyond the range we
587 	 * actually support in this filesystem.
588 	 */
589 	if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
590 		last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
591 
592 	ASSERT(last_fsb > offset_fsb);
593 
594 	nimaps = XFS_WRITE_IMAPS;
595 	error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
596 				imap, &nimaps, XFS_BMAPI_ENTIRE);
597 	switch (error) {
598 	case 0:
599 	case ENOSPC:
600 	case EDQUOT:
601 		break;
602 	default:
603 		return XFS_ERROR(error);
604 	}
605 
606 	/*
607 	 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
608 	 * without EOF preallocation.
609 	 */
610 	if (nimaps == 0) {
611 		trace_xfs_delalloc_enospc(ip, offset, count);
612 		if (prealloc) {
613 			prealloc = 0;
614 			error = 0;
615 			goto retry;
616 		}
617 		return XFS_ERROR(error ? error : ENOSPC);
618 	}
619 
620 	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
621 		return xfs_alert_fsblock_zero(ip, &imap[0]);
622 
623 	/*
624 	 * Tag the inode as speculatively preallocated so we can reclaim this
625 	 * space on demand, if necessary.
626 	 */
627 	if (prealloc)
628 		xfs_inode_set_eofblocks_tag(ip);
629 
630 	*ret_imap = imap[0];
631 	return 0;
632 }
633 
634 /*
635  * Pass in a delayed allocate extent, convert it to real extents;
636  * return to the caller the extent we create which maps on top of
637  * the originating callers request.
638  *
639  * Called without a lock on the inode.
640  *
641  * We no longer bother to look at the incoming map - all we have to
642  * guarantee is that whatever we allocate fills the required range.
643  */
644 int
645 xfs_iomap_write_allocate(
646 	xfs_inode_t	*ip,
647 	xfs_off_t	offset,
648 	xfs_bmbt_irec_t *imap)
649 {
650 	xfs_mount_t	*mp = ip->i_mount;
651 	xfs_fileoff_t	offset_fsb, last_block;
652 	xfs_fileoff_t	end_fsb, map_start_fsb;
653 	xfs_fsblock_t	first_block;
654 	xfs_bmap_free_t	free_list;
655 	xfs_filblks_t	count_fsb;
656 	xfs_trans_t	*tp;
657 	int		nimaps, committed;
658 	int		error = 0;
659 	int		nres;
660 
661 	/*
662 	 * Make sure that the dquots are there.
663 	 */
664 	error = xfs_qm_dqattach(ip, 0);
665 	if (error)
666 		return XFS_ERROR(error);
667 
668 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
669 	count_fsb = imap->br_blockcount;
670 	map_start_fsb = imap->br_startoff;
671 
672 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
673 
674 	while (count_fsb != 0) {
675 		/*
676 		 * Set up a transaction with which to allocate the
677 		 * backing store for the file.  Do allocations in a
678 		 * loop until we get some space in the range we are
679 		 * interested in.  The other space that might be allocated
680 		 * is in the delayed allocation extent on which we sit
681 		 * but before our buffer starts.
682 		 */
683 
684 		nimaps = 0;
685 		while (nimaps == 0) {
686 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
687 			tp->t_flags |= XFS_TRANS_RESERVE;
688 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
689 			error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
690 						  nres, 0);
691 			if (error) {
692 				xfs_trans_cancel(tp, 0);
693 				return XFS_ERROR(error);
694 			}
695 			xfs_ilock(ip, XFS_ILOCK_EXCL);
696 			xfs_trans_ijoin(tp, ip, 0);
697 
698 			xfs_bmap_init(&free_list, &first_block);
699 
700 			/*
701 			 * it is possible that the extents have changed since
702 			 * we did the read call as we dropped the ilock for a
703 			 * while. We have to be careful about truncates or hole
704 			 * punchs here - we are not allowed to allocate
705 			 * non-delalloc blocks here.
706 			 *
707 			 * The only protection against truncation is the pages
708 			 * for the range we are being asked to convert are
709 			 * locked and hence a truncate will block on them
710 			 * first.
711 			 *
712 			 * As a result, if we go beyond the range we really
713 			 * need and hit an delalloc extent boundary followed by
714 			 * a hole while we have excess blocks in the map, we
715 			 * will fill the hole incorrectly and overrun the
716 			 * transaction reservation.
717 			 *
718 			 * Using a single map prevents this as we are forced to
719 			 * check each map we look for overlap with the desired
720 			 * range and abort as soon as we find it. Also, given
721 			 * that we only return a single map, having one beyond
722 			 * what we can return is probably a bit silly.
723 			 *
724 			 * We also need to check that we don't go beyond EOF;
725 			 * this is a truncate optimisation as a truncate sets
726 			 * the new file size before block on the pages we
727 			 * currently have locked under writeback. Because they
728 			 * are about to be tossed, we don't need to write them
729 			 * back....
730 			 */
731 			nimaps = 1;
732 			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
733 			error = xfs_bmap_last_offset(NULL, ip, &last_block,
734 							XFS_DATA_FORK);
735 			if (error)
736 				goto trans_cancel;
737 
738 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
739 			if ((map_start_fsb + count_fsb) > last_block) {
740 				count_fsb = last_block - map_start_fsb;
741 				if (count_fsb == 0) {
742 					error = EAGAIN;
743 					goto trans_cancel;
744 				}
745 			}
746 
747 			/*
748 			 * From this point onwards we overwrite the imap
749 			 * pointer that the caller gave to us.
750 			 */
751 			error = xfs_bmapi_write(tp, ip, map_start_fsb,
752 						count_fsb,
753 						XFS_BMAPI_STACK_SWITCH,
754 						&first_block, 1,
755 						imap, &nimaps, &free_list);
756 			if (error)
757 				goto trans_cancel;
758 
759 			error = xfs_bmap_finish(&tp, &free_list, &committed);
760 			if (error)
761 				goto trans_cancel;
762 
763 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
764 			if (error)
765 				goto error0;
766 
767 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
768 		}
769 
770 		/*
771 		 * See if we were able to allocate an extent that
772 		 * covers at least part of the callers request
773 		 */
774 		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
775 			return xfs_alert_fsblock_zero(ip, imap);
776 
777 		if ((offset_fsb >= imap->br_startoff) &&
778 		    (offset_fsb < (imap->br_startoff +
779 				   imap->br_blockcount))) {
780 			XFS_STATS_INC(xs_xstrat_quick);
781 			return 0;
782 		}
783 
784 		/*
785 		 * So far we have not mapped the requested part of the
786 		 * file, just surrounding data, try again.
787 		 */
788 		count_fsb -= imap->br_blockcount;
789 		map_start_fsb = imap->br_startoff + imap->br_blockcount;
790 	}
791 
792 trans_cancel:
793 	xfs_bmap_cancel(&free_list);
794 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
795 error0:
796 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
797 	return XFS_ERROR(error);
798 }
799 
800 int
801 xfs_iomap_write_unwritten(
802 	xfs_inode_t	*ip,
803 	xfs_off_t	offset,
804 	size_t		count)
805 {
806 	xfs_mount_t	*mp = ip->i_mount;
807 	xfs_fileoff_t	offset_fsb;
808 	xfs_filblks_t	count_fsb;
809 	xfs_filblks_t	numblks_fsb;
810 	xfs_fsblock_t	firstfsb;
811 	int		nimaps;
812 	xfs_trans_t	*tp;
813 	xfs_bmbt_irec_t imap;
814 	xfs_bmap_free_t free_list;
815 	xfs_fsize_t	i_size;
816 	uint		resblks;
817 	int		committed;
818 	int		error;
819 
820 	trace_xfs_unwritten_convert(ip, offset, count);
821 
822 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
823 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
824 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
825 
826 	/*
827 	 * Reserve enough blocks in this transaction for two complete extent
828 	 * btree splits.  We may be converting the middle part of an unwritten
829 	 * extent and in this case we will insert two new extents in the btree
830 	 * each of which could cause a full split.
831 	 *
832 	 * This reservation amount will be used in the first call to
833 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
834 	 * rest of the operation.
835 	 */
836 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
837 
838 	do {
839 		/*
840 		 * set up a transaction to convert the range of extents
841 		 * from unwritten to real. Do allocations in a loop until
842 		 * we have covered the range passed in.
843 		 *
844 		 * Note that we open code the transaction allocation here
845 		 * to pass KM_NOFS--we can't risk to recursing back into
846 		 * the filesystem here as we might be asked to write out
847 		 * the same inode that we complete here and might deadlock
848 		 * on the iolock.
849 		 */
850 		sb_start_intwrite(mp->m_super);
851 		tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
852 		tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
853 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
854 					  resblks, 0);
855 		if (error) {
856 			xfs_trans_cancel(tp, 0);
857 			return XFS_ERROR(error);
858 		}
859 
860 		xfs_ilock(ip, XFS_ILOCK_EXCL);
861 		xfs_trans_ijoin(tp, ip, 0);
862 
863 		/*
864 		 * Modify the unwritten extent state of the buffer.
865 		 */
866 		xfs_bmap_init(&free_list, &firstfsb);
867 		nimaps = 1;
868 		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
869 				  XFS_BMAPI_CONVERT, &firstfsb,
870 				  1, &imap, &nimaps, &free_list);
871 		if (error)
872 			goto error_on_bmapi_transaction;
873 
874 		/*
875 		 * Log the updated inode size as we go.  We have to be careful
876 		 * to only log it up to the actual write offset if it is
877 		 * halfway into a block.
878 		 */
879 		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
880 		if (i_size > offset + count)
881 			i_size = offset + count;
882 
883 		i_size = xfs_new_eof(ip, i_size);
884 		if (i_size) {
885 			ip->i_d.di_size = i_size;
886 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
887 		}
888 
889 		error = xfs_bmap_finish(&tp, &free_list, &committed);
890 		if (error)
891 			goto error_on_bmapi_transaction;
892 
893 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
894 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
895 		if (error)
896 			return XFS_ERROR(error);
897 
898 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
899 			return xfs_alert_fsblock_zero(ip, &imap);
900 
901 		if ((numblks_fsb = imap.br_blockcount) == 0) {
902 			/*
903 			 * The numblks_fsb value should always get
904 			 * smaller, otherwise the loop is stuck.
905 			 */
906 			ASSERT(imap.br_blockcount);
907 			break;
908 		}
909 		offset_fsb += numblks_fsb;
910 		count_fsb -= numblks_fsb;
911 	} while (count_fsb > 0);
912 
913 	return 0;
914 
915 error_on_bmapi_transaction:
916 	xfs_bmap_cancel(&free_list);
917 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
918 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
919 	return XFS_ERROR(error);
920 }
921