xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
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_STRAT_WRITE_IMAPS	2
51 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
52 
53 STATIC int xfs_iomap_write_direct(struct xfs_inode *, xfs_off_t, size_t,
54 				  int, struct xfs_bmbt_irec *, int *);
55 STATIC int xfs_iomap_write_delay(struct xfs_inode *, xfs_off_t, size_t, int,
56 				 struct xfs_bmbt_irec *, int *);
57 STATIC int xfs_iomap_write_allocate(struct xfs_inode *, xfs_off_t, size_t,
58 				struct xfs_bmbt_irec *, int *);
59 
60 int
61 xfs_iomap(
62 	struct xfs_inode	*ip,
63 	xfs_off_t		offset,
64 	ssize_t			count,
65 	int			flags,
66 	struct xfs_bmbt_irec	*imap,
67 	int			*nimaps,
68 	int			*new)
69 {
70 	struct xfs_mount	*mp = ip->i_mount;
71 	xfs_fileoff_t		offset_fsb, end_fsb;
72 	int			error = 0;
73 	int			lockmode = 0;
74 	int			bmapi_flags = 0;
75 
76 	ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
77 
78 	*new = 0;
79 
80 	if (XFS_FORCED_SHUTDOWN(mp))
81 		return XFS_ERROR(EIO);
82 
83 	trace_xfs_iomap_enter(ip, offset, count, flags, NULL);
84 
85 	switch (flags & (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE)) {
86 	case BMAPI_READ:
87 		lockmode = xfs_ilock_map_shared(ip);
88 		bmapi_flags = XFS_BMAPI_ENTIRE;
89 		break;
90 	case BMAPI_WRITE:
91 		lockmode = XFS_ILOCK_EXCL;
92 		if (flags & BMAPI_IGNSTATE)
93 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
94 		xfs_ilock(ip, lockmode);
95 		break;
96 	case BMAPI_ALLOCATE:
97 		lockmode = XFS_ILOCK_SHARED;
98 		bmapi_flags = XFS_BMAPI_ENTIRE;
99 
100 		/* Attempt non-blocking lock */
101 		if (flags & BMAPI_TRYLOCK) {
102 			if (!xfs_ilock_nowait(ip, lockmode))
103 				return XFS_ERROR(EAGAIN);
104 		} else {
105 			xfs_ilock(ip, lockmode);
106 		}
107 		break;
108 	default:
109 		BUG();
110 	}
111 
112 	ASSERT(offset <= mp->m_maxioffset);
113 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
114 		count = mp->m_maxioffset - offset;
115 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
116 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
117 
118 	error = xfs_bmapi(NULL, ip, offset_fsb,
119 			(xfs_filblks_t)(end_fsb - offset_fsb),
120 			bmapi_flags,  NULL, 0, imap,
121 			nimaps, NULL);
122 
123 	if (error)
124 		goto out;
125 
126 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)) {
127 	case BMAPI_WRITE:
128 		/* If we found an extent, return it */
129 		if (*nimaps &&
130 		    (imap->br_startblock != HOLESTARTBLOCK) &&
131 		    (imap->br_startblock != DELAYSTARTBLOCK)) {
132 			trace_xfs_iomap_found(ip, offset, count, flags, imap);
133 			break;
134 		}
135 
136 		if (flags & BMAPI_DIRECT) {
137 			error = xfs_iomap_write_direct(ip, offset, count, flags,
138 						       imap, nimaps);
139 		} else {
140 			error = xfs_iomap_write_delay(ip, offset, count, flags,
141 						      imap, nimaps);
142 		}
143 		if (!error) {
144 			trace_xfs_iomap_alloc(ip, offset, count, flags, imap);
145 		}
146 		*new = 1;
147 		break;
148 	case BMAPI_ALLOCATE:
149 		/* If we found an extent, return it */
150 		xfs_iunlock(ip, lockmode);
151 		lockmode = 0;
152 
153 		if (*nimaps && !isnullstartblock(imap->br_startblock)) {
154 			trace_xfs_iomap_found(ip, offset, count, flags, imap);
155 			break;
156 		}
157 
158 		error = xfs_iomap_write_allocate(ip, offset, count,
159 						 imap, nimaps);
160 		break;
161 	}
162 
163 	ASSERT(*nimaps <= 1);
164 
165 out:
166 	if (lockmode)
167 		xfs_iunlock(ip, lockmode);
168 	return XFS_ERROR(error);
169 }
170 
171 STATIC int
172 xfs_iomap_eof_align_last_fsb(
173 	xfs_mount_t	*mp,
174 	xfs_inode_t	*ip,
175 	xfs_extlen_t	extsize,
176 	xfs_fileoff_t	*last_fsb)
177 {
178 	xfs_fileoff_t	new_last_fsb = 0;
179 	xfs_extlen_t	align;
180 	int		eof, error;
181 
182 	if (XFS_IS_REALTIME_INODE(ip))
183 		;
184 	/*
185 	 * If mounted with the "-o swalloc" option, roundup the allocation
186 	 * request to a stripe width boundary if the file size is >=
187 	 * stripe width and we are allocating past the allocation eof.
188 	 */
189 	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
190 	        (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
191 		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
192 	/*
193 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
194 	 * if the file size is >= stripe unit size, and we are allocating past
195 	 * the allocation eof.
196 	 */
197 	else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
198 		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
199 
200 	/*
201 	 * Always round up the allocation request to an extent boundary
202 	 * (when file on a real-time subvolume or has di_extsize hint).
203 	 */
204 	if (extsize) {
205 		if (new_last_fsb)
206 			align = roundup_64(new_last_fsb, extsize);
207 		else
208 			align = extsize;
209 		new_last_fsb = roundup_64(*last_fsb, align);
210 	}
211 
212 	if (new_last_fsb) {
213 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
214 		if (error)
215 			return error;
216 		if (eof)
217 			*last_fsb = new_last_fsb;
218 	}
219 	return 0;
220 }
221 
222 STATIC int
223 xfs_cmn_err_fsblock_zero(
224 	xfs_inode_t	*ip,
225 	xfs_bmbt_irec_t	*imap)
226 {
227 	xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
228 			"Access to block zero in inode %llu "
229 			"start_block: %llx start_off: %llx "
230 			"blkcnt: %llx extent-state: %x\n",
231 		(unsigned long long)ip->i_ino,
232 		(unsigned long long)imap->br_startblock,
233 		(unsigned long long)imap->br_startoff,
234 		(unsigned long long)imap->br_blockcount,
235 		imap->br_state);
236 	return EFSCORRUPTED;
237 }
238 
239 STATIC int
240 xfs_iomap_write_direct(
241 	xfs_inode_t	*ip,
242 	xfs_off_t	offset,
243 	size_t		count,
244 	int		flags,
245 	xfs_bmbt_irec_t *imap,
246 	int		*nmaps)
247 {
248 	xfs_mount_t	*mp = ip->i_mount;
249 	xfs_fileoff_t	offset_fsb;
250 	xfs_fileoff_t	last_fsb;
251 	xfs_filblks_t	count_fsb, resaligned;
252 	xfs_fsblock_t	firstfsb;
253 	xfs_extlen_t	extsz, temp;
254 	int		nimaps;
255 	int		bmapi_flag;
256 	int		quota_flag;
257 	int		rt;
258 	xfs_trans_t	*tp;
259 	xfs_bmap_free_t free_list;
260 	uint		qblocks, resblks, resrtextents;
261 	int		committed;
262 	int		error;
263 
264 	/*
265 	 * Make sure that the dquots are there. This doesn't hold
266 	 * the ilock across a disk read.
267 	 */
268 	error = xfs_qm_dqattach_locked(ip, 0);
269 	if (error)
270 		return XFS_ERROR(error);
271 
272 	rt = XFS_IS_REALTIME_INODE(ip);
273 	extsz = xfs_get_extsz_hint(ip);
274 
275 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
276 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
277 	if ((offset + count) > ip->i_size) {
278 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
279 		if (error)
280 			goto error_out;
281 	} else {
282 		if (*nmaps && (imap->br_startblock == HOLESTARTBLOCK))
283 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
284 					imap->br_blockcount +
285 					imap->br_startoff);
286 	}
287 	count_fsb = last_fsb - offset_fsb;
288 	ASSERT(count_fsb > 0);
289 
290 	resaligned = count_fsb;
291 	if (unlikely(extsz)) {
292 		if ((temp = do_mod(offset_fsb, extsz)))
293 			resaligned += temp;
294 		if ((temp = do_mod(resaligned, extsz)))
295 			resaligned += extsz - temp;
296 	}
297 
298 	if (unlikely(rt)) {
299 		resrtextents = qblocks = resaligned;
300 		resrtextents /= mp->m_sb.sb_rextsize;
301 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
302 		quota_flag = XFS_QMOPT_RES_RTBLKS;
303 	} else {
304 		resrtextents = 0;
305 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
306 		quota_flag = XFS_QMOPT_RES_REGBLKS;
307 	}
308 
309 	/*
310 	 * Allocate and setup the transaction
311 	 */
312 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
313 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
314 	error = xfs_trans_reserve(tp, resblks,
315 			XFS_WRITE_LOG_RES(mp), resrtextents,
316 			XFS_TRANS_PERM_LOG_RES,
317 			XFS_WRITE_LOG_COUNT);
318 	/*
319 	 * Check for running out of space, note: need lock to return
320 	 */
321 	if (error)
322 		xfs_trans_cancel(tp, 0);
323 	xfs_ilock(ip, XFS_ILOCK_EXCL);
324 	if (error)
325 		goto error_out;
326 
327 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
328 	if (error)
329 		goto error1;
330 
331 	xfs_trans_ijoin(tp, ip);
332 
333 	bmapi_flag = XFS_BMAPI_WRITE;
334 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
335 		bmapi_flag |= XFS_BMAPI_PREALLOC;
336 
337 	/*
338 	 * Issue the xfs_bmapi() call to allocate the blocks.
339 	 *
340 	 * From this point onwards we overwrite the imap pointer that the
341 	 * caller gave to us.
342 	 */
343 	xfs_bmap_init(&free_list, &firstfsb);
344 	nimaps = 1;
345 	error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
346 		&firstfsb, 0, imap, &nimaps, &free_list);
347 	if (error)
348 		goto error0;
349 
350 	/*
351 	 * Complete the transaction
352 	 */
353 	error = xfs_bmap_finish(&tp, &free_list, &committed);
354 	if (error)
355 		goto error0;
356 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
357 	if (error)
358 		goto error_out;
359 
360 	/*
361 	 * Copy any maps to caller's array and return any error.
362 	 */
363 	if (nimaps == 0) {
364 		error = ENOSPC;
365 		goto error_out;
366 	}
367 
368 	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) {
369 		error = xfs_cmn_err_fsblock_zero(ip, imap);
370 		goto error_out;
371 	}
372 
373 	*nmaps = 1;
374 	return 0;
375 
376 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
377 	xfs_bmap_cancel(&free_list);
378 	xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
379 
380 error1:	/* Just cancel transaction */
381 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
382 	*nmaps = 0;	/* nothing set-up here */
383 
384 error_out:
385 	return XFS_ERROR(error);
386 }
387 
388 /*
389  * If the caller is doing a write at the end of the file, then extend the
390  * allocation out to the file system's write iosize.  We clean up any extra
391  * space left over when the file is closed in xfs_inactive().
392  */
393 STATIC int
394 xfs_iomap_eof_want_preallocate(
395 	xfs_mount_t	*mp,
396 	xfs_inode_t	*ip,
397 	xfs_off_t	offset,
398 	size_t		count,
399 	int		ioflag,
400 	xfs_bmbt_irec_t *imap,
401 	int		nimaps,
402 	int		*prealloc)
403 {
404 	xfs_fileoff_t   start_fsb;
405 	xfs_filblks_t   count_fsb;
406 	xfs_fsblock_t	firstblock;
407 	int		n, error, imaps;
408 
409 	*prealloc = 0;
410 	if ((offset + count) <= ip->i_size)
411 		return 0;
412 
413 	/*
414 	 * If there are any real blocks past eof, then don't
415 	 * do any speculative allocation.
416 	 */
417 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
418 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
419 	while (count_fsb > 0) {
420 		imaps = nimaps;
421 		firstblock = NULLFSBLOCK;
422 		error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
423 				  &firstblock, 0, imap, &imaps, NULL);
424 		if (error)
425 			return error;
426 		for (n = 0; n < imaps; n++) {
427 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
428 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
429 				return 0;
430 			start_fsb += imap[n].br_blockcount;
431 			count_fsb -= imap[n].br_blockcount;
432 		}
433 	}
434 	*prealloc = 1;
435 	return 0;
436 }
437 
438 STATIC int
439 xfs_iomap_write_delay(
440 	xfs_inode_t	*ip,
441 	xfs_off_t	offset,
442 	size_t		count,
443 	int		ioflag,
444 	xfs_bmbt_irec_t *ret_imap,
445 	int		*nmaps)
446 {
447 	xfs_mount_t	*mp = ip->i_mount;
448 	xfs_fileoff_t	offset_fsb;
449 	xfs_fileoff_t	last_fsb;
450 	xfs_off_t	aligned_offset;
451 	xfs_fileoff_t	ioalign;
452 	xfs_fsblock_t	firstblock;
453 	xfs_extlen_t	extsz;
454 	int		nimaps;
455 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
456 	int		prealloc, flushed = 0;
457 	int		error;
458 
459 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
460 
461 	/*
462 	 * Make sure that the dquots are there. This doesn't hold
463 	 * the ilock across a disk read.
464 	 */
465 	error = xfs_qm_dqattach_locked(ip, 0);
466 	if (error)
467 		return XFS_ERROR(error);
468 
469 	extsz = xfs_get_extsz_hint(ip);
470 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
471 
472 	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
473 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
474 	if (error)
475 		return error;
476 
477 retry:
478 	if (prealloc) {
479 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
480 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
481 		last_fsb = ioalign + mp->m_writeio_blocks;
482 	} else {
483 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
484 	}
485 
486 	if (prealloc || extsz) {
487 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
488 		if (error)
489 			return error;
490 	}
491 
492 	nimaps = XFS_WRITE_IMAPS;
493 	firstblock = NULLFSBLOCK;
494 	error = xfs_bmapi(NULL, ip, offset_fsb,
495 			  (xfs_filblks_t)(last_fsb - offset_fsb),
496 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
497 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
498 			  &nimaps, NULL);
499 	if (error && (error != ENOSPC))
500 		return XFS_ERROR(error);
501 
502 	/*
503 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
504 	 * then we must have run out of space - flush all other inodes with
505 	 * delalloc blocks and retry without EOF preallocation.
506 	 */
507 	if (nimaps == 0) {
508 		trace_xfs_delalloc_enospc(ip, offset, count);
509 		if (flushed)
510 			return XFS_ERROR(ENOSPC);
511 
512 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
513 		xfs_flush_inodes(ip);
514 		xfs_ilock(ip, XFS_ILOCK_EXCL);
515 
516 		flushed = 1;
517 		error = 0;
518 		prealloc = 0;
519 		goto retry;
520 	}
521 
522 	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
523 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
524 
525 	*ret_imap = imap[0];
526 	*nmaps = 1;
527 
528 	return 0;
529 }
530 
531 /*
532  * Pass in a delayed allocate extent, convert it to real extents;
533  * return to the caller the extent we create which maps on top of
534  * the originating callers request.
535  *
536  * Called without a lock on the inode.
537  *
538  * We no longer bother to look at the incoming map - all we have to
539  * guarantee is that whatever we allocate fills the required range.
540  */
541 STATIC int
542 xfs_iomap_write_allocate(
543 	xfs_inode_t	*ip,
544 	xfs_off_t	offset,
545 	size_t		count,
546 	xfs_bmbt_irec_t *imap,
547 	int		*retmap)
548 {
549 	xfs_mount_t	*mp = ip->i_mount;
550 	xfs_fileoff_t	offset_fsb, last_block;
551 	xfs_fileoff_t	end_fsb, map_start_fsb;
552 	xfs_fsblock_t	first_block;
553 	xfs_bmap_free_t	free_list;
554 	xfs_filblks_t	count_fsb;
555 	xfs_trans_t	*tp;
556 	int		nimaps, committed;
557 	int		error = 0;
558 	int		nres;
559 
560 	*retmap = 0;
561 
562 	/*
563 	 * Make sure that the dquots are there.
564 	 */
565 	error = xfs_qm_dqattach(ip, 0);
566 	if (error)
567 		return XFS_ERROR(error);
568 
569 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
570 	count_fsb = imap->br_blockcount;
571 	map_start_fsb = imap->br_startoff;
572 
573 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
574 
575 	while (count_fsb != 0) {
576 		/*
577 		 * Set up a transaction with which to allocate the
578 		 * backing store for the file.  Do allocations in a
579 		 * loop until we get some space in the range we are
580 		 * interested in.  The other space that might be allocated
581 		 * is in the delayed allocation extent on which we sit
582 		 * but before our buffer starts.
583 		 */
584 
585 		nimaps = 0;
586 		while (nimaps == 0) {
587 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
588 			tp->t_flags |= XFS_TRANS_RESERVE;
589 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
590 			error = xfs_trans_reserve(tp, nres,
591 					XFS_WRITE_LOG_RES(mp),
592 					0, XFS_TRANS_PERM_LOG_RES,
593 					XFS_WRITE_LOG_COUNT);
594 			if (error) {
595 				xfs_trans_cancel(tp, 0);
596 				return XFS_ERROR(error);
597 			}
598 			xfs_ilock(ip, XFS_ILOCK_EXCL);
599 			xfs_trans_ijoin(tp, ip);
600 
601 			xfs_bmap_init(&free_list, &first_block);
602 
603 			/*
604 			 * it is possible that the extents have changed since
605 			 * we did the read call as we dropped the ilock for a
606 			 * while. We have to be careful about truncates or hole
607 			 * punchs here - we are not allowed to allocate
608 			 * non-delalloc blocks here.
609 			 *
610 			 * The only protection against truncation is the pages
611 			 * for the range we are being asked to convert are
612 			 * locked and hence a truncate will block on them
613 			 * first.
614 			 *
615 			 * As a result, if we go beyond the range we really
616 			 * need and hit an delalloc extent boundary followed by
617 			 * a hole while we have excess blocks in the map, we
618 			 * will fill the hole incorrectly and overrun the
619 			 * transaction reservation.
620 			 *
621 			 * Using a single map prevents this as we are forced to
622 			 * check each map we look for overlap with the desired
623 			 * range and abort as soon as we find it. Also, given
624 			 * that we only return a single map, having one beyond
625 			 * what we can return is probably a bit silly.
626 			 *
627 			 * We also need to check that we don't go beyond EOF;
628 			 * this is a truncate optimisation as a truncate sets
629 			 * the new file size before block on the pages we
630 			 * currently have locked under writeback. Because they
631 			 * are about to be tossed, we don't need to write them
632 			 * back....
633 			 */
634 			nimaps = 1;
635 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
636 			error = xfs_bmap_last_offset(NULL, ip, &last_block,
637 							XFS_DATA_FORK);
638 			if (error)
639 				goto trans_cancel;
640 
641 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
642 			if ((map_start_fsb + count_fsb) > last_block) {
643 				count_fsb = last_block - map_start_fsb;
644 				if (count_fsb == 0) {
645 					error = EAGAIN;
646 					goto trans_cancel;
647 				}
648 			}
649 
650 			/*
651 			 * Go get the actual blocks.
652 	 	 	 *
653 			 * From this point onwards we overwrite the imap
654 			 * pointer that the caller gave to us.
655 			 */
656 			error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
657 					XFS_BMAPI_WRITE, &first_block, 1,
658 					imap, &nimaps, &free_list);
659 			if (error)
660 				goto trans_cancel;
661 
662 			error = xfs_bmap_finish(&tp, &free_list, &committed);
663 			if (error)
664 				goto trans_cancel;
665 
666 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
667 			if (error)
668 				goto error0;
669 
670 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
671 		}
672 
673 		/*
674 		 * See if we were able to allocate an extent that
675 		 * covers at least part of the callers request
676 		 */
677 		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
678 			return xfs_cmn_err_fsblock_zero(ip, imap);
679 
680 		if ((offset_fsb >= imap->br_startoff) &&
681 		    (offset_fsb < (imap->br_startoff +
682 				   imap->br_blockcount))) {
683 			*retmap = 1;
684 			XFS_STATS_INC(xs_xstrat_quick);
685 			return 0;
686 		}
687 
688 		/*
689 		 * So far we have not mapped the requested part of the
690 		 * file, just surrounding data, try again.
691 		 */
692 		count_fsb -= imap->br_blockcount;
693 		map_start_fsb = imap->br_startoff + imap->br_blockcount;
694 	}
695 
696 trans_cancel:
697 	xfs_bmap_cancel(&free_list);
698 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
699 error0:
700 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
701 	return XFS_ERROR(error);
702 }
703 
704 int
705 xfs_iomap_write_unwritten(
706 	xfs_inode_t	*ip,
707 	xfs_off_t	offset,
708 	size_t		count)
709 {
710 	xfs_mount_t	*mp = ip->i_mount;
711 	xfs_fileoff_t	offset_fsb;
712 	xfs_filblks_t	count_fsb;
713 	xfs_filblks_t	numblks_fsb;
714 	xfs_fsblock_t	firstfsb;
715 	int		nimaps;
716 	xfs_trans_t	*tp;
717 	xfs_bmbt_irec_t imap;
718 	xfs_bmap_free_t free_list;
719 	uint		resblks;
720 	int		committed;
721 	int		error;
722 
723 	trace_xfs_unwritten_convert(ip, offset, count);
724 
725 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
726 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
727 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
728 
729 	/*
730 	 * Reserve enough blocks in this transaction for two complete extent
731 	 * btree splits.  We may be converting the middle part of an unwritten
732 	 * extent and in this case we will insert two new extents in the btree
733 	 * each of which could cause a full split.
734 	 *
735 	 * This reservation amount will be used in the first call to
736 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
737 	 * rest of the operation.
738 	 */
739 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
740 
741 	do {
742 		/*
743 		 * set up a transaction to convert the range of extents
744 		 * from unwritten to real. Do allocations in a loop until
745 		 * we have covered the range passed in.
746 		 *
747 		 * Note that we open code the transaction allocation here
748 		 * to pass KM_NOFS--we can't risk to recursing back into
749 		 * the filesystem here as we might be asked to write out
750 		 * the same inode that we complete here and might deadlock
751 		 * on the iolock.
752 		 */
753 		xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
754 		tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
755 		tp->t_flags |= XFS_TRANS_RESERVE;
756 		error = xfs_trans_reserve(tp, resblks,
757 				XFS_WRITE_LOG_RES(mp), 0,
758 				XFS_TRANS_PERM_LOG_RES,
759 				XFS_WRITE_LOG_COUNT);
760 		if (error) {
761 			xfs_trans_cancel(tp, 0);
762 			return XFS_ERROR(error);
763 		}
764 
765 		xfs_ilock(ip, XFS_ILOCK_EXCL);
766 		xfs_trans_ijoin(tp, ip);
767 
768 		/*
769 		 * Modify the unwritten extent state of the buffer.
770 		 */
771 		xfs_bmap_init(&free_list, &firstfsb);
772 		nimaps = 1;
773 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
774 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
775 				  1, &imap, &nimaps, &free_list);
776 		if (error)
777 			goto error_on_bmapi_transaction;
778 
779 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
780 		if (error)
781 			goto error_on_bmapi_transaction;
782 
783 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
784 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
785 		if (error)
786 			return XFS_ERROR(error);
787 
788 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
789 			return xfs_cmn_err_fsblock_zero(ip, &imap);
790 
791 		if ((numblks_fsb = imap.br_blockcount) == 0) {
792 			/*
793 			 * The numblks_fsb value should always get
794 			 * smaller, otherwise the loop is stuck.
795 			 */
796 			ASSERT(imap.br_blockcount);
797 			break;
798 		}
799 		offset_fsb += numblks_fsb;
800 		count_fsb -= numblks_fsb;
801 	} while (count_fsb > 0);
802 
803 	return 0;
804 
805 error_on_bmapi_transaction:
806 	xfs_bmap_cancel(&free_list);
807 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
808 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
809 	return XFS_ERROR(error);
810 }
811