xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
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_dir2.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_btree.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
44 #include "xfs_rw.h"
45 #include "xfs_attr.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_utils.h"
49 #include "xfs_iomap.h"
50 #include "xfs_trace.h"
51 
52 
53 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
54 						<< mp->m_writeio_log)
55 #define XFS_STRAT_WRITE_IMAPS	2
56 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
57 
58 STATIC int
59 xfs_imap_to_bmap(
60 	xfs_inode_t	*ip,
61 	xfs_off_t	offset,
62 	xfs_bmbt_irec_t *imap,
63 	xfs_iomap_t	*iomapp,
64 	int		imaps,			/* Number of imap entries */
65 	int		iomaps,			/* Number of iomap entries */
66 	int		flags)
67 {
68 	xfs_mount_t	*mp = ip->i_mount;
69 	int		pbm;
70 	xfs_fsblock_t	start_block;
71 
72 
73 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
74 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
75 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
76 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
77 		iomapp->iomap_flags = flags;
78 
79 		if (XFS_IS_REALTIME_INODE(ip)) {
80 			iomapp->iomap_flags |= IOMAP_REALTIME;
81 			iomapp->iomap_target = mp->m_rtdev_targp;
82 		} else {
83 			iomapp->iomap_target = mp->m_ddev_targp;
84 		}
85 		start_block = imap->br_startblock;
86 		if (start_block == HOLESTARTBLOCK) {
87 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
88 			iomapp->iomap_flags |= IOMAP_HOLE;
89 		} else if (start_block == DELAYSTARTBLOCK) {
90 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
91 			iomapp->iomap_flags |= IOMAP_DELAY;
92 		} else {
93 			iomapp->iomap_bn = xfs_fsb_to_db(ip, start_block);
94 			if (ISUNWRITTEN(imap))
95 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
96 		}
97 
98 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
99 	}
100 	return pbm;	/* Return the number filled */
101 }
102 
103 int
104 xfs_iomap(
105 	xfs_inode_t	*ip,
106 	xfs_off_t	offset,
107 	ssize_t		count,
108 	int		flags,
109 	xfs_iomap_t	*iomapp,
110 	int		*niomaps)
111 {
112 	xfs_mount_t	*mp = ip->i_mount;
113 	xfs_fileoff_t	offset_fsb, end_fsb;
114 	int		error = 0;
115 	int		lockmode = 0;
116 	xfs_bmbt_irec_t	imap;
117 	int		nimaps = 1;
118 	int		bmapi_flags = 0;
119 	int		iomap_flags = 0;
120 
121 	ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
122 
123 	if (XFS_FORCED_SHUTDOWN(mp))
124 		return XFS_ERROR(EIO);
125 
126 	trace_xfs_iomap_enter(ip, offset, count, flags, NULL);
127 
128 	switch (flags & (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE)) {
129 	case BMAPI_READ:
130 		lockmode = xfs_ilock_map_shared(ip);
131 		bmapi_flags = XFS_BMAPI_ENTIRE;
132 		break;
133 	case BMAPI_WRITE:
134 		lockmode = XFS_ILOCK_EXCL;
135 		if (flags & BMAPI_IGNSTATE)
136 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
137 		xfs_ilock(ip, lockmode);
138 		break;
139 	case BMAPI_ALLOCATE:
140 		lockmode = XFS_ILOCK_SHARED;
141 		bmapi_flags = XFS_BMAPI_ENTIRE;
142 
143 		/* Attempt non-blocking lock */
144 		if (flags & BMAPI_TRYLOCK) {
145 			if (!xfs_ilock_nowait(ip, lockmode))
146 				return XFS_ERROR(EAGAIN);
147 		} else {
148 			xfs_ilock(ip, lockmode);
149 		}
150 		break;
151 	default:
152 		BUG();
153 	}
154 
155 	ASSERT(offset <= mp->m_maxioffset);
156 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
157 		count = mp->m_maxioffset - offset;
158 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
159 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
160 
161 	error = xfs_bmapi(NULL, ip, offset_fsb,
162 			(xfs_filblks_t)(end_fsb - offset_fsb),
163 			bmapi_flags,  NULL, 0, &imap,
164 			&nimaps, NULL, NULL);
165 
166 	if (error)
167 		goto out;
168 
169 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)) {
170 	case BMAPI_WRITE:
171 		/* If we found an extent, return it */
172 		if (nimaps &&
173 		    (imap.br_startblock != HOLESTARTBLOCK) &&
174 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
175 			trace_xfs_iomap_found(ip, offset, count, flags, &imap);
176 			break;
177 		}
178 
179 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
180 			error = xfs_iomap_write_direct(ip, offset, count, flags,
181 						       &imap, &nimaps, nimaps);
182 		} else {
183 			error = xfs_iomap_write_delay(ip, offset, count, flags,
184 						      &imap, &nimaps);
185 		}
186 		if (!error) {
187 			trace_xfs_iomap_alloc(ip, offset, count, flags, &imap);
188 		}
189 		iomap_flags = IOMAP_NEW;
190 		break;
191 	case BMAPI_ALLOCATE:
192 		/* If we found an extent, return it */
193 		xfs_iunlock(ip, lockmode);
194 		lockmode = 0;
195 
196 		if (nimaps && !isnullstartblock(imap.br_startblock)) {
197 			trace_xfs_iomap_found(ip, offset, count, flags, &imap);
198 			break;
199 		}
200 
201 		error = xfs_iomap_write_allocate(ip, offset, count,
202 						 &imap, &nimaps);
203 		break;
204 	}
205 
206 	if (nimaps) {
207 		*niomaps = xfs_imap_to_bmap(ip, offset, &imap,
208 					    iomapp, nimaps, *niomaps, iomap_flags);
209 	} else if (niomaps) {
210 		*niomaps = 0;
211 	}
212 
213 out:
214 	if (lockmode)
215 		xfs_iunlock(ip, lockmode);
216 	return XFS_ERROR(error);
217 }
218 
219 
220 STATIC int
221 xfs_iomap_eof_align_last_fsb(
222 	xfs_mount_t	*mp,
223 	xfs_inode_t	*ip,
224 	xfs_extlen_t	extsize,
225 	xfs_fileoff_t	*last_fsb)
226 {
227 	xfs_fileoff_t	new_last_fsb = 0;
228 	xfs_extlen_t	align;
229 	int		eof, error;
230 
231 	if (XFS_IS_REALTIME_INODE(ip))
232 		;
233 	/*
234 	 * If mounted with the "-o swalloc" option, roundup the allocation
235 	 * request to a stripe width boundary if the file size is >=
236 	 * stripe width and we are allocating past the allocation eof.
237 	 */
238 	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
239 	        (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
240 		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
241 	/*
242 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
243 	 * if the file size is >= stripe unit size, and we are allocating past
244 	 * the allocation eof.
245 	 */
246 	else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
247 		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
248 
249 	/*
250 	 * Always round up the allocation request to an extent boundary
251 	 * (when file on a real-time subvolume or has di_extsize hint).
252 	 */
253 	if (extsize) {
254 		if (new_last_fsb)
255 			align = roundup_64(new_last_fsb, extsize);
256 		else
257 			align = extsize;
258 		new_last_fsb = roundup_64(*last_fsb, align);
259 	}
260 
261 	if (new_last_fsb) {
262 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
263 		if (error)
264 			return error;
265 		if (eof)
266 			*last_fsb = new_last_fsb;
267 	}
268 	return 0;
269 }
270 
271 STATIC int
272 xfs_cmn_err_fsblock_zero(
273 	xfs_inode_t	*ip,
274 	xfs_bmbt_irec_t	*imap)
275 {
276 	xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
277 			"Access to block zero in inode %llu "
278 			"start_block: %llx start_off: %llx "
279 			"blkcnt: %llx extent-state: %x\n",
280 		(unsigned long long)ip->i_ino,
281 		(unsigned long long)imap->br_startblock,
282 		(unsigned long long)imap->br_startoff,
283 		(unsigned long long)imap->br_blockcount,
284 		imap->br_state);
285 	return EFSCORRUPTED;
286 }
287 
288 int
289 xfs_iomap_write_direct(
290 	xfs_inode_t	*ip,
291 	xfs_off_t	offset,
292 	size_t		count,
293 	int		flags,
294 	xfs_bmbt_irec_t *ret_imap,
295 	int		*nmaps,
296 	int		found)
297 {
298 	xfs_mount_t	*mp = ip->i_mount;
299 	xfs_fileoff_t	offset_fsb;
300 	xfs_fileoff_t	last_fsb;
301 	xfs_filblks_t	count_fsb, resaligned;
302 	xfs_fsblock_t	firstfsb;
303 	xfs_extlen_t	extsz, temp;
304 	int		nimaps;
305 	int		bmapi_flag;
306 	int		quota_flag;
307 	int		rt;
308 	xfs_trans_t	*tp;
309 	xfs_bmbt_irec_t imap;
310 	xfs_bmap_free_t free_list;
311 	uint		qblocks, resblks, resrtextents;
312 	int		committed;
313 	int		error;
314 
315 	/*
316 	 * Make sure that the dquots are there. This doesn't hold
317 	 * the ilock across a disk read.
318 	 */
319 	error = xfs_qm_dqattach_locked(ip, 0);
320 	if (error)
321 		return XFS_ERROR(error);
322 
323 	rt = XFS_IS_REALTIME_INODE(ip);
324 	extsz = xfs_get_extsz_hint(ip);
325 
326 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
327 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
328 	if ((offset + count) > ip->i_size) {
329 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
330 		if (error)
331 			goto error_out;
332 	} else {
333 		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
334 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
335 					ret_imap->br_blockcount +
336 					ret_imap->br_startoff);
337 	}
338 	count_fsb = last_fsb - offset_fsb;
339 	ASSERT(count_fsb > 0);
340 
341 	resaligned = count_fsb;
342 	if (unlikely(extsz)) {
343 		if ((temp = do_mod(offset_fsb, extsz)))
344 			resaligned += temp;
345 		if ((temp = do_mod(resaligned, extsz)))
346 			resaligned += extsz - temp;
347 	}
348 
349 	if (unlikely(rt)) {
350 		resrtextents = qblocks = resaligned;
351 		resrtextents /= mp->m_sb.sb_rextsize;
352 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
353 		quota_flag = XFS_QMOPT_RES_RTBLKS;
354 	} else {
355 		resrtextents = 0;
356 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
357 		quota_flag = XFS_QMOPT_RES_REGBLKS;
358 	}
359 
360 	/*
361 	 * Allocate and setup the transaction
362 	 */
363 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
364 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
365 	error = xfs_trans_reserve(tp, resblks,
366 			XFS_WRITE_LOG_RES(mp), resrtextents,
367 			XFS_TRANS_PERM_LOG_RES,
368 			XFS_WRITE_LOG_COUNT);
369 	/*
370 	 * Check for running out of space, note: need lock to return
371 	 */
372 	if (error)
373 		xfs_trans_cancel(tp, 0);
374 	xfs_ilock(ip, XFS_ILOCK_EXCL);
375 	if (error)
376 		goto error_out;
377 
378 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
379 	if (error)
380 		goto error1;
381 
382 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
383 	xfs_trans_ihold(tp, ip);
384 
385 	bmapi_flag = XFS_BMAPI_WRITE;
386 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
387 		bmapi_flag |= XFS_BMAPI_PREALLOC;
388 
389 	/*
390 	 * Issue the xfs_bmapi() call to allocate the blocks
391 	 */
392 	xfs_bmap_init(&free_list, &firstfsb);
393 	nimaps = 1;
394 	error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
395 		&firstfsb, 0, &imap, &nimaps, &free_list, NULL);
396 	if (error)
397 		goto error0;
398 
399 	/*
400 	 * Complete the transaction
401 	 */
402 	error = xfs_bmap_finish(&tp, &free_list, &committed);
403 	if (error)
404 		goto error0;
405 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
406 	if (error)
407 		goto error_out;
408 
409 	/*
410 	 * Copy any maps to caller's array and return any error.
411 	 */
412 	if (nimaps == 0) {
413 		error = ENOSPC;
414 		goto error_out;
415 	}
416 
417 	if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) {
418 		error = xfs_cmn_err_fsblock_zero(ip, &imap);
419 		goto error_out;
420 	}
421 
422 	*ret_imap = imap;
423 	*nmaps = 1;
424 	return 0;
425 
426 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
427 	xfs_bmap_cancel(&free_list);
428 	xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
429 
430 error1:	/* Just cancel transaction */
431 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
432 	*nmaps = 0;	/* nothing set-up here */
433 
434 error_out:
435 	return XFS_ERROR(error);
436 }
437 
438 /*
439  * If the caller is doing a write at the end of the file, then extend the
440  * allocation out to the file system's write iosize.  We clean up any extra
441  * space left over when the file is closed in xfs_inactive().
442  */
443 STATIC int
444 xfs_iomap_eof_want_preallocate(
445 	xfs_mount_t	*mp,
446 	xfs_inode_t	*ip,
447 	xfs_off_t	offset,
448 	size_t		count,
449 	int		ioflag,
450 	xfs_bmbt_irec_t *imap,
451 	int		nimaps,
452 	int		*prealloc)
453 {
454 	xfs_fileoff_t   start_fsb;
455 	xfs_filblks_t   count_fsb;
456 	xfs_fsblock_t	firstblock;
457 	int		n, error, imaps;
458 
459 	*prealloc = 0;
460 	if ((offset + count) <= ip->i_size)
461 		return 0;
462 
463 	/*
464 	 * If there are any real blocks past eof, then don't
465 	 * do any speculative allocation.
466 	 */
467 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
468 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
469 	while (count_fsb > 0) {
470 		imaps = nimaps;
471 		firstblock = NULLFSBLOCK;
472 		error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
473 				  &firstblock, 0, imap, &imaps, NULL, NULL);
474 		if (error)
475 			return error;
476 		for (n = 0; n < imaps; n++) {
477 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
478 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
479 				return 0;
480 			start_fsb += imap[n].br_blockcount;
481 			count_fsb -= imap[n].br_blockcount;
482 		}
483 	}
484 	*prealloc = 1;
485 	return 0;
486 }
487 
488 int
489 xfs_iomap_write_delay(
490 	xfs_inode_t	*ip,
491 	xfs_off_t	offset,
492 	size_t		count,
493 	int		ioflag,
494 	xfs_bmbt_irec_t *ret_imap,
495 	int		*nmaps)
496 {
497 	xfs_mount_t	*mp = ip->i_mount;
498 	xfs_fileoff_t	offset_fsb;
499 	xfs_fileoff_t	last_fsb;
500 	xfs_off_t	aligned_offset;
501 	xfs_fileoff_t	ioalign;
502 	xfs_fsblock_t	firstblock;
503 	xfs_extlen_t	extsz;
504 	int		nimaps;
505 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
506 	int		prealloc, flushed = 0;
507 	int		error;
508 
509 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
510 
511 	/*
512 	 * Make sure that the dquots are there. This doesn't hold
513 	 * the ilock across a disk read.
514 	 */
515 	error = xfs_qm_dqattach_locked(ip, 0);
516 	if (error)
517 		return XFS_ERROR(error);
518 
519 	extsz = xfs_get_extsz_hint(ip);
520 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
521 
522 	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
523 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
524 	if (error)
525 		return error;
526 
527 retry:
528 	if (prealloc) {
529 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
530 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
531 		last_fsb = ioalign + mp->m_writeio_blocks;
532 	} else {
533 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
534 	}
535 
536 	if (prealloc || extsz) {
537 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
538 		if (error)
539 			return error;
540 	}
541 
542 	nimaps = XFS_WRITE_IMAPS;
543 	firstblock = NULLFSBLOCK;
544 	error = xfs_bmapi(NULL, ip, offset_fsb,
545 			  (xfs_filblks_t)(last_fsb - offset_fsb),
546 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
547 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
548 			  &nimaps, NULL, NULL);
549 	if (error && (error != ENOSPC))
550 		return XFS_ERROR(error);
551 
552 	/*
553 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
554 	 * then we must have run out of space - flush all other inodes with
555 	 * delalloc blocks and retry without EOF preallocation.
556 	 */
557 	if (nimaps == 0) {
558 		trace_xfs_delalloc_enospc(ip, offset, count);
559 		if (flushed)
560 			return XFS_ERROR(ENOSPC);
561 
562 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
563 		xfs_flush_inodes(ip);
564 		xfs_ilock(ip, XFS_ILOCK_EXCL);
565 
566 		flushed = 1;
567 		error = 0;
568 		prealloc = 0;
569 		goto retry;
570 	}
571 
572 	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
573 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
574 
575 	*ret_imap = imap[0];
576 	*nmaps = 1;
577 
578 	return 0;
579 }
580 
581 /*
582  * Pass in a delayed allocate extent, convert it to real extents;
583  * return to the caller the extent we create which maps on top of
584  * the originating callers request.
585  *
586  * Called without a lock on the inode.
587  *
588  * We no longer bother to look at the incoming map - all we have to
589  * guarantee is that whatever we allocate fills the required range.
590  */
591 int
592 xfs_iomap_write_allocate(
593 	xfs_inode_t	*ip,
594 	xfs_off_t	offset,
595 	size_t		count,
596 	xfs_bmbt_irec_t *map,
597 	int		*retmap)
598 {
599 	xfs_mount_t	*mp = ip->i_mount;
600 	xfs_fileoff_t	offset_fsb, last_block;
601 	xfs_fileoff_t	end_fsb, map_start_fsb;
602 	xfs_fsblock_t	first_block;
603 	xfs_bmap_free_t	free_list;
604 	xfs_filblks_t	count_fsb;
605 	xfs_bmbt_irec_t	imap;
606 	xfs_trans_t	*tp;
607 	int		nimaps, committed;
608 	int		error = 0;
609 	int		nres;
610 
611 	*retmap = 0;
612 
613 	/*
614 	 * Make sure that the dquots are there.
615 	 */
616 	error = xfs_qm_dqattach(ip, 0);
617 	if (error)
618 		return XFS_ERROR(error);
619 
620 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
621 	count_fsb = map->br_blockcount;
622 	map_start_fsb = map->br_startoff;
623 
624 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
625 
626 	while (count_fsb != 0) {
627 		/*
628 		 * Set up a transaction with which to allocate the
629 		 * backing store for the file.  Do allocations in a
630 		 * loop until we get some space in the range we are
631 		 * interested in.  The other space that might be allocated
632 		 * is in the delayed allocation extent on which we sit
633 		 * but before our buffer starts.
634 		 */
635 
636 		nimaps = 0;
637 		while (nimaps == 0) {
638 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
639 			tp->t_flags |= XFS_TRANS_RESERVE;
640 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
641 			error = xfs_trans_reserve(tp, nres,
642 					XFS_WRITE_LOG_RES(mp),
643 					0, XFS_TRANS_PERM_LOG_RES,
644 					XFS_WRITE_LOG_COUNT);
645 			if (error) {
646 				xfs_trans_cancel(tp, 0);
647 				return XFS_ERROR(error);
648 			}
649 			xfs_ilock(ip, XFS_ILOCK_EXCL);
650 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
651 			xfs_trans_ihold(tp, ip);
652 
653 			xfs_bmap_init(&free_list, &first_block);
654 
655 			/*
656 			 * it is possible that the extents have changed since
657 			 * we did the read call as we dropped the ilock for a
658 			 * while. We have to be careful about truncates or hole
659 			 * punchs here - we are not allowed to allocate
660 			 * non-delalloc blocks here.
661 			 *
662 			 * The only protection against truncation is the pages
663 			 * for the range we are being asked to convert are
664 			 * locked and hence a truncate will block on them
665 			 * first.
666 			 *
667 			 * As a result, if we go beyond the range we really
668 			 * need and hit an delalloc extent boundary followed by
669 			 * a hole while we have excess blocks in the map, we
670 			 * will fill the hole incorrectly and overrun the
671 			 * transaction reservation.
672 			 *
673 			 * Using a single map prevents this as we are forced to
674 			 * check each map we look for overlap with the desired
675 			 * range and abort as soon as we find it. Also, given
676 			 * that we only return a single map, having one beyond
677 			 * what we can return is probably a bit silly.
678 			 *
679 			 * We also need to check that we don't go beyond EOF;
680 			 * this is a truncate optimisation as a truncate sets
681 			 * the new file size before block on the pages we
682 			 * currently have locked under writeback. Because they
683 			 * are about to be tossed, we don't need to write them
684 			 * back....
685 			 */
686 			nimaps = 1;
687 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
688 			error = xfs_bmap_last_offset(NULL, ip, &last_block,
689 							XFS_DATA_FORK);
690 			if (error)
691 				goto trans_cancel;
692 
693 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
694 			if ((map_start_fsb + count_fsb) > last_block) {
695 				count_fsb = last_block - map_start_fsb;
696 				if (count_fsb == 0) {
697 					error = EAGAIN;
698 					goto trans_cancel;
699 				}
700 			}
701 
702 			/* Go get the actual blocks */
703 			error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
704 					XFS_BMAPI_WRITE, &first_block, 1,
705 					&imap, &nimaps, &free_list, NULL);
706 			if (error)
707 				goto trans_cancel;
708 
709 			error = xfs_bmap_finish(&tp, &free_list, &committed);
710 			if (error)
711 				goto trans_cancel;
712 
713 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
714 			if (error)
715 				goto error0;
716 
717 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
718 		}
719 
720 		/*
721 		 * See if we were able to allocate an extent that
722 		 * covers at least part of the callers request
723 		 */
724 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
725 			return xfs_cmn_err_fsblock_zero(ip, &imap);
726 
727 		if ((offset_fsb >= imap.br_startoff) &&
728 		    (offset_fsb < (imap.br_startoff +
729 				   imap.br_blockcount))) {
730 			*map = imap;
731 			*retmap = 1;
732 			XFS_STATS_INC(xs_xstrat_quick);
733 			return 0;
734 		}
735 
736 		/*
737 		 * So far we have not mapped the requested part of the
738 		 * file, just surrounding data, try again.
739 		 */
740 		count_fsb -= imap.br_blockcount;
741 		map_start_fsb = imap.br_startoff + imap.br_blockcount;
742 	}
743 
744 trans_cancel:
745 	xfs_bmap_cancel(&free_list);
746 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
747 error0:
748 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
749 	return XFS_ERROR(error);
750 }
751 
752 int
753 xfs_iomap_write_unwritten(
754 	xfs_inode_t	*ip,
755 	xfs_off_t	offset,
756 	size_t		count)
757 {
758 	xfs_mount_t	*mp = ip->i_mount;
759 	xfs_fileoff_t	offset_fsb;
760 	xfs_filblks_t	count_fsb;
761 	xfs_filblks_t	numblks_fsb;
762 	xfs_fsblock_t	firstfsb;
763 	int		nimaps;
764 	xfs_trans_t	*tp;
765 	xfs_bmbt_irec_t imap;
766 	xfs_bmap_free_t free_list;
767 	uint		resblks;
768 	int		committed;
769 	int		error;
770 
771 	trace_xfs_unwritten_convert(ip, offset, count);
772 
773 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
774 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
775 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
776 
777 	/*
778 	 * Reserve enough blocks in this transaction for two complete extent
779 	 * btree splits.  We may be converting the middle part of an unwritten
780 	 * extent and in this case we will insert two new extents in the btree
781 	 * each of which could cause a full split.
782 	 *
783 	 * This reservation amount will be used in the first call to
784 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
785 	 * rest of the operation.
786 	 */
787 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
788 
789 	do {
790 		/*
791 		 * set up a transaction to convert the range of extents
792 		 * from unwritten to real. Do allocations in a loop until
793 		 * we have covered the range passed in.
794 		 *
795 		 * Note that we open code the transaction allocation here
796 		 * to pass KM_NOFS--we can't risk to recursing back into
797 		 * the filesystem here as we might be asked to write out
798 		 * the same inode that we complete here and might deadlock
799 		 * on the iolock.
800 		 */
801 		xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
802 		tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
803 		tp->t_flags |= XFS_TRANS_RESERVE;
804 		error = xfs_trans_reserve(tp, resblks,
805 				XFS_WRITE_LOG_RES(mp), 0,
806 				XFS_TRANS_PERM_LOG_RES,
807 				XFS_WRITE_LOG_COUNT);
808 		if (error) {
809 			xfs_trans_cancel(tp, 0);
810 			return XFS_ERROR(error);
811 		}
812 
813 		xfs_ilock(ip, XFS_ILOCK_EXCL);
814 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
815 		xfs_trans_ihold(tp, ip);
816 
817 		/*
818 		 * Modify the unwritten extent state of the buffer.
819 		 */
820 		xfs_bmap_init(&free_list, &firstfsb);
821 		nimaps = 1;
822 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
823 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
824 				  1, &imap, &nimaps, &free_list, NULL);
825 		if (error)
826 			goto error_on_bmapi_transaction;
827 
828 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
829 		if (error)
830 			goto error_on_bmapi_transaction;
831 
832 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
833 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
834 		if (error)
835 			return XFS_ERROR(error);
836 
837 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
838 			return xfs_cmn_err_fsblock_zero(ip, &imap);
839 
840 		if ((numblks_fsb = imap.br_blockcount) == 0) {
841 			/*
842 			 * The numblks_fsb value should always get
843 			 * smaller, otherwise the loop is stuck.
844 			 */
845 			ASSERT(imap.br_blockcount);
846 			break;
847 		}
848 		offset_fsb += numblks_fsb;
849 		count_fsb -= numblks_fsb;
850 	} while (count_fsb > 0);
851 
852 	return 0;
853 
854 error_on_bmapi_transaction:
855 	xfs_bmap_cancel(&free_list);
856 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
857 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
858 	return XFS_ERROR(error);
859 }
860