xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision b04b4f78)
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_acl.h"
46 #include "xfs_attr.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_trans_space.h"
49 #include "xfs_utils.h"
50 #include "xfs_iomap.h"
51 
52 #if defined(XFS_RW_TRACE)
53 void
54 xfs_iomap_enter_trace(
55 	int		tag,
56 	xfs_inode_t	*ip,
57 	xfs_off_t	offset,
58 	ssize_t		count)
59 {
60 	if (!ip->i_rwtrace)
61 		return;
62 
63 	ktrace_enter(ip->i_rwtrace,
64 		(void *)((unsigned long)tag),
65 		(void *)ip,
66 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
67 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
68 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
69 		(void *)((unsigned long)(offset & 0xffffffff)),
70 		(void *)((unsigned long)count),
71 		(void *)((unsigned long)((ip->i_new_size >> 32) & 0xffffffff)),
72 		(void *)((unsigned long)(ip->i_new_size & 0xffffffff)),
73 		(void *)((unsigned long)current_pid()),
74 		(void *)NULL,
75 		(void *)NULL,
76 		(void *)NULL,
77 		(void *)NULL,
78 		(void *)NULL,
79 		(void *)NULL);
80 }
81 
82 void
83 xfs_iomap_map_trace(
84 	int		tag,
85 	xfs_inode_t	*ip,
86 	xfs_off_t	offset,
87 	ssize_t		count,
88 	xfs_iomap_t	*iomapp,
89 	xfs_bmbt_irec_t	*imapp,
90 	int		flags)
91 {
92 	if (!ip->i_rwtrace)
93 		return;
94 
95 	ktrace_enter(ip->i_rwtrace,
96 		(void *)((unsigned long)tag),
97 		(void *)ip,
98 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
99 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
100 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
101 		(void *)((unsigned long)(offset & 0xffffffff)),
102 		(void *)((unsigned long)count),
103 		(void *)((unsigned long)flags),
104 		(void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
105 		(void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
106 		(void *)((unsigned long)(iomapp->iomap_delta)),
107 		(void *)((unsigned long)(iomapp->iomap_bsize)),
108 		(void *)((unsigned long)(iomapp->iomap_bn)),
109 		(void *)(__psint_t)(imapp->br_startoff),
110 		(void *)((unsigned long)(imapp->br_blockcount)),
111 		(void *)(__psint_t)(imapp->br_startblock));
112 }
113 #else
114 #define xfs_iomap_enter_trace(tag, io, offset, count)
115 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
116 #endif
117 
118 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
119 						<< mp->m_writeio_log)
120 #define XFS_STRAT_WRITE_IMAPS	2
121 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
122 
123 STATIC int
124 xfs_imap_to_bmap(
125 	xfs_inode_t	*ip,
126 	xfs_off_t	offset,
127 	xfs_bmbt_irec_t *imap,
128 	xfs_iomap_t	*iomapp,
129 	int		imaps,			/* Number of imap entries */
130 	int		iomaps,			/* Number of iomap entries */
131 	int		flags)
132 {
133 	xfs_mount_t	*mp = ip->i_mount;
134 	int		pbm;
135 	xfs_fsblock_t	start_block;
136 
137 
138 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
139 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
140 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
141 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
142 		iomapp->iomap_flags = flags;
143 
144 		if (XFS_IS_REALTIME_INODE(ip)) {
145 			iomapp->iomap_flags |= IOMAP_REALTIME;
146 			iomapp->iomap_target = mp->m_rtdev_targp;
147 		} else {
148 			iomapp->iomap_target = mp->m_ddev_targp;
149 		}
150 		start_block = imap->br_startblock;
151 		if (start_block == HOLESTARTBLOCK) {
152 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
153 			iomapp->iomap_flags |= IOMAP_HOLE;
154 		} else if (start_block == DELAYSTARTBLOCK) {
155 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
156 			iomapp->iomap_flags |= IOMAP_DELAY;
157 		} else {
158 			iomapp->iomap_bn = xfs_fsb_to_db(ip, start_block);
159 			if (ISUNWRITTEN(imap))
160 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
161 		}
162 
163 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
164 	}
165 	return pbm;	/* Return the number filled */
166 }
167 
168 int
169 xfs_iomap(
170 	xfs_inode_t	*ip,
171 	xfs_off_t	offset,
172 	ssize_t		count,
173 	int		flags,
174 	xfs_iomap_t	*iomapp,
175 	int		*niomaps)
176 {
177 	xfs_mount_t	*mp = ip->i_mount;
178 	xfs_fileoff_t	offset_fsb, end_fsb;
179 	int		error = 0;
180 	int		lockmode = 0;
181 	xfs_bmbt_irec_t	imap;
182 	int		nimaps = 1;
183 	int		bmapi_flags = 0;
184 	int		iomap_flags = 0;
185 
186 	ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
187 
188 	if (XFS_FORCED_SHUTDOWN(mp))
189 		return XFS_ERROR(EIO);
190 
191 	switch (flags & (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE)) {
192 	case BMAPI_READ:
193 		xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, ip, offset, count);
194 		lockmode = xfs_ilock_map_shared(ip);
195 		bmapi_flags = XFS_BMAPI_ENTIRE;
196 		break;
197 	case BMAPI_WRITE:
198 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, ip, offset, count);
199 		lockmode = XFS_ILOCK_EXCL;
200 		if (flags & BMAPI_IGNSTATE)
201 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
202 		xfs_ilock(ip, lockmode);
203 		break;
204 	case BMAPI_ALLOCATE:
205 		xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, ip, offset, count);
206 		lockmode = XFS_ILOCK_SHARED;
207 		bmapi_flags = XFS_BMAPI_ENTIRE;
208 
209 		/* Attempt non-blocking lock */
210 		if (flags & BMAPI_TRYLOCK) {
211 			if (!xfs_ilock_nowait(ip, lockmode))
212 				return XFS_ERROR(EAGAIN);
213 		} else {
214 			xfs_ilock(ip, lockmode);
215 		}
216 		break;
217 	default:
218 		BUG();
219 	}
220 
221 	ASSERT(offset <= mp->m_maxioffset);
222 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
223 		count = mp->m_maxioffset - offset;
224 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
225 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
226 
227 	error = xfs_bmapi(NULL, ip, offset_fsb,
228 			(xfs_filblks_t)(end_fsb - offset_fsb),
229 			bmapi_flags,  NULL, 0, &imap,
230 			&nimaps, NULL, NULL);
231 
232 	if (error)
233 		goto out;
234 
235 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)) {
236 	case BMAPI_WRITE:
237 		/* If we found an extent, return it */
238 		if (nimaps &&
239 		    (imap.br_startblock != HOLESTARTBLOCK) &&
240 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
241 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip,
242 					offset, count, iomapp, &imap, flags);
243 			break;
244 		}
245 
246 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
247 			error = xfs_iomap_write_direct(ip, offset, count, flags,
248 						       &imap, &nimaps, nimaps);
249 		} else {
250 			error = xfs_iomap_write_delay(ip, offset, count, flags,
251 						      &imap, &nimaps);
252 		}
253 		if (!error) {
254 			xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, ip,
255 					offset, count, iomapp, &imap, flags);
256 		}
257 		iomap_flags = IOMAP_NEW;
258 		break;
259 	case BMAPI_ALLOCATE:
260 		/* If we found an extent, return it */
261 		xfs_iunlock(ip, lockmode);
262 		lockmode = 0;
263 
264 		if (nimaps && !isnullstartblock(imap.br_startblock)) {
265 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip,
266 					offset, count, iomapp, &imap, flags);
267 			break;
268 		}
269 
270 		error = xfs_iomap_write_allocate(ip, offset, count,
271 						 &imap, &nimaps);
272 		break;
273 	}
274 
275 	if (nimaps) {
276 		*niomaps = xfs_imap_to_bmap(ip, offset, &imap,
277 					    iomapp, nimaps, *niomaps, iomap_flags);
278 	} else if (niomaps) {
279 		*niomaps = 0;
280 	}
281 
282 out:
283 	if (lockmode)
284 		xfs_iunlock(ip, lockmode);
285 	return XFS_ERROR(error);
286 }
287 
288 
289 STATIC int
290 xfs_iomap_eof_align_last_fsb(
291 	xfs_mount_t	*mp,
292 	xfs_inode_t	*ip,
293 	xfs_extlen_t	extsize,
294 	xfs_fileoff_t	*last_fsb)
295 {
296 	xfs_fileoff_t	new_last_fsb = 0;
297 	xfs_extlen_t	align;
298 	int		eof, error;
299 
300 	if (XFS_IS_REALTIME_INODE(ip))
301 		;
302 	/*
303 	 * If mounted with the "-o swalloc" option, roundup the allocation
304 	 * request to a stripe width boundary if the file size is >=
305 	 * stripe width and we are allocating past the allocation eof.
306 	 */
307 	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
308 	        (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
309 		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
310 	/*
311 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
312 	 * if the file size is >= stripe unit size, and we are allocating past
313 	 * the allocation eof.
314 	 */
315 	else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
316 		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
317 
318 	/*
319 	 * Always round up the allocation request to an extent boundary
320 	 * (when file on a real-time subvolume or has di_extsize hint).
321 	 */
322 	if (extsize) {
323 		if (new_last_fsb)
324 			align = roundup_64(new_last_fsb, extsize);
325 		else
326 			align = extsize;
327 		new_last_fsb = roundup_64(*last_fsb, align);
328 	}
329 
330 	if (new_last_fsb) {
331 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
332 		if (error)
333 			return error;
334 		if (eof)
335 			*last_fsb = new_last_fsb;
336 	}
337 	return 0;
338 }
339 
340 STATIC int
341 xfs_cmn_err_fsblock_zero(
342 	xfs_inode_t	*ip,
343 	xfs_bmbt_irec_t	*imap)
344 {
345 	xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
346 			"Access to block zero in inode %llu "
347 			"start_block: %llx start_off: %llx "
348 			"blkcnt: %llx extent-state: %x\n",
349 		(unsigned long long)ip->i_ino,
350 		(unsigned long long)imap->br_startblock,
351 		(unsigned long long)imap->br_startoff,
352 		(unsigned long long)imap->br_blockcount,
353 		imap->br_state);
354 	return EFSCORRUPTED;
355 }
356 
357 int
358 xfs_iomap_write_direct(
359 	xfs_inode_t	*ip,
360 	xfs_off_t	offset,
361 	size_t		count,
362 	int		flags,
363 	xfs_bmbt_irec_t *ret_imap,
364 	int		*nmaps,
365 	int		found)
366 {
367 	xfs_mount_t	*mp = ip->i_mount;
368 	xfs_fileoff_t	offset_fsb;
369 	xfs_fileoff_t	last_fsb;
370 	xfs_filblks_t	count_fsb, resaligned;
371 	xfs_fsblock_t	firstfsb;
372 	xfs_extlen_t	extsz, temp;
373 	int		nimaps;
374 	int		bmapi_flag;
375 	int		quota_flag;
376 	int		rt;
377 	xfs_trans_t	*tp;
378 	xfs_bmbt_irec_t imap;
379 	xfs_bmap_free_t free_list;
380 	uint		qblocks, resblks, resrtextents;
381 	int		committed;
382 	int		error;
383 
384 	/*
385 	 * Make sure that the dquots are there. This doesn't hold
386 	 * the ilock across a disk read.
387 	 */
388 	error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
389 	if (error)
390 		return XFS_ERROR(error);
391 
392 	rt = XFS_IS_REALTIME_INODE(ip);
393 	extsz = xfs_get_extsz_hint(ip);
394 
395 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
396 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
397 	if ((offset + count) > ip->i_size) {
398 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
399 		if (error)
400 			goto error_out;
401 	} else {
402 		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
403 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
404 					ret_imap->br_blockcount +
405 					ret_imap->br_startoff);
406 	}
407 	count_fsb = last_fsb - offset_fsb;
408 	ASSERT(count_fsb > 0);
409 
410 	resaligned = count_fsb;
411 	if (unlikely(extsz)) {
412 		if ((temp = do_mod(offset_fsb, extsz)))
413 			resaligned += temp;
414 		if ((temp = do_mod(resaligned, extsz)))
415 			resaligned += extsz - temp;
416 	}
417 
418 	if (unlikely(rt)) {
419 		resrtextents = qblocks = resaligned;
420 		resrtextents /= mp->m_sb.sb_rextsize;
421 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
422 		quota_flag = XFS_QMOPT_RES_RTBLKS;
423 	} else {
424 		resrtextents = 0;
425 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
426 		quota_flag = XFS_QMOPT_RES_REGBLKS;
427 	}
428 
429 	/*
430 	 * Allocate and setup the transaction
431 	 */
432 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
433 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
434 	error = xfs_trans_reserve(tp, resblks,
435 			XFS_WRITE_LOG_RES(mp), resrtextents,
436 			XFS_TRANS_PERM_LOG_RES,
437 			XFS_WRITE_LOG_COUNT);
438 	/*
439 	 * Check for running out of space, note: need lock to return
440 	 */
441 	if (error)
442 		xfs_trans_cancel(tp, 0);
443 	xfs_ilock(ip, XFS_ILOCK_EXCL);
444 	if (error)
445 		goto error_out;
446 
447 	error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
448 					      qblocks, 0, quota_flag);
449 	if (error)
450 		goto error1;
451 
452 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
453 	xfs_trans_ihold(tp, ip);
454 
455 	bmapi_flag = XFS_BMAPI_WRITE;
456 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
457 		bmapi_flag |= XFS_BMAPI_PREALLOC;
458 
459 	/*
460 	 * Issue the xfs_bmapi() call to allocate the blocks
461 	 */
462 	xfs_bmap_init(&free_list, &firstfsb);
463 	nimaps = 1;
464 	error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
465 		&firstfsb, 0, &imap, &nimaps, &free_list, NULL);
466 	if (error)
467 		goto error0;
468 
469 	/*
470 	 * Complete the transaction
471 	 */
472 	error = xfs_bmap_finish(&tp, &free_list, &committed);
473 	if (error)
474 		goto error0;
475 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
476 	if (error)
477 		goto error_out;
478 
479 	/*
480 	 * Copy any maps to caller's array and return any error.
481 	 */
482 	if (nimaps == 0) {
483 		error = ENOSPC;
484 		goto error_out;
485 	}
486 
487 	if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) {
488 		error = xfs_cmn_err_fsblock_zero(ip, &imap);
489 		goto error_out;
490 	}
491 
492 	*ret_imap = imap;
493 	*nmaps = 1;
494 	return 0;
495 
496 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
497 	xfs_bmap_cancel(&free_list);
498 	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
499 
500 error1:	/* Just cancel transaction */
501 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
502 	*nmaps = 0;	/* nothing set-up here */
503 
504 error_out:
505 	return XFS_ERROR(error);
506 }
507 
508 /*
509  * If the caller is doing a write at the end of the file, then extend the
510  * allocation out to the file system's write iosize.  We clean up any extra
511  * space left over when the file is closed in xfs_inactive().
512  */
513 STATIC int
514 xfs_iomap_eof_want_preallocate(
515 	xfs_mount_t	*mp,
516 	xfs_inode_t	*ip,
517 	xfs_off_t	offset,
518 	size_t		count,
519 	int		ioflag,
520 	xfs_bmbt_irec_t *imap,
521 	int		nimaps,
522 	int		*prealloc)
523 {
524 	xfs_fileoff_t   start_fsb;
525 	xfs_filblks_t   count_fsb;
526 	xfs_fsblock_t	firstblock;
527 	int		n, error, imaps;
528 
529 	*prealloc = 0;
530 	if ((offset + count) <= ip->i_size)
531 		return 0;
532 
533 	/*
534 	 * If there are any real blocks past eof, then don't
535 	 * do any speculative allocation.
536 	 */
537 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
538 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
539 	while (count_fsb > 0) {
540 		imaps = nimaps;
541 		firstblock = NULLFSBLOCK;
542 		error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
543 				  &firstblock, 0, imap, &imaps, NULL, NULL);
544 		if (error)
545 			return error;
546 		for (n = 0; n < imaps; n++) {
547 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
548 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
549 				return 0;
550 			start_fsb += imap[n].br_blockcount;
551 			count_fsb -= imap[n].br_blockcount;
552 		}
553 	}
554 	*prealloc = 1;
555 	return 0;
556 }
557 
558 int
559 xfs_iomap_write_delay(
560 	xfs_inode_t	*ip,
561 	xfs_off_t	offset,
562 	size_t		count,
563 	int		ioflag,
564 	xfs_bmbt_irec_t *ret_imap,
565 	int		*nmaps)
566 {
567 	xfs_mount_t	*mp = ip->i_mount;
568 	xfs_fileoff_t	offset_fsb;
569 	xfs_fileoff_t	last_fsb;
570 	xfs_off_t	aligned_offset;
571 	xfs_fileoff_t	ioalign;
572 	xfs_fsblock_t	firstblock;
573 	xfs_extlen_t	extsz;
574 	int		nimaps;
575 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
576 	int		prealloc, flushed = 0;
577 	int		error;
578 
579 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
580 
581 	/*
582 	 * Make sure that the dquots are there. This doesn't hold
583 	 * the ilock across a disk read.
584 	 */
585 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
586 	if (error)
587 		return XFS_ERROR(error);
588 
589 	extsz = xfs_get_extsz_hint(ip);
590 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
591 
592 	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
593 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
594 	if (error)
595 		return error;
596 
597 retry:
598 	if (prealloc) {
599 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
600 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
601 		last_fsb = ioalign + mp->m_writeio_blocks;
602 	} else {
603 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
604 	}
605 
606 	if (prealloc || extsz) {
607 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
608 		if (error)
609 			return error;
610 	}
611 
612 	nimaps = XFS_WRITE_IMAPS;
613 	firstblock = NULLFSBLOCK;
614 	error = xfs_bmapi(NULL, ip, offset_fsb,
615 			  (xfs_filblks_t)(last_fsb - offset_fsb),
616 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
617 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
618 			  &nimaps, NULL, NULL);
619 	if (error && (error != ENOSPC))
620 		return XFS_ERROR(error);
621 
622 	/*
623 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
624 	 * then we must have run out of space - flush all other inodes with
625 	 * delalloc blocks and retry without EOF preallocation.
626 	 */
627 	if (nimaps == 0) {
628 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
629 					ip, offset, count);
630 		if (flushed)
631 			return XFS_ERROR(ENOSPC);
632 
633 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
634 		xfs_flush_inodes(ip);
635 		xfs_ilock(ip, XFS_ILOCK_EXCL);
636 
637 		flushed = 1;
638 		error = 0;
639 		prealloc = 0;
640 		goto retry;
641 	}
642 
643 	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
644 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
645 
646 	*ret_imap = imap[0];
647 	*nmaps = 1;
648 
649 	return 0;
650 }
651 
652 /*
653  * Pass in a delayed allocate extent, convert it to real extents;
654  * return to the caller the extent we create which maps on top of
655  * the originating callers request.
656  *
657  * Called without a lock on the inode.
658  *
659  * We no longer bother to look at the incoming map - all we have to
660  * guarantee is that whatever we allocate fills the required range.
661  */
662 int
663 xfs_iomap_write_allocate(
664 	xfs_inode_t	*ip,
665 	xfs_off_t	offset,
666 	size_t		count,
667 	xfs_bmbt_irec_t *map,
668 	int		*retmap)
669 {
670 	xfs_mount_t	*mp = ip->i_mount;
671 	xfs_fileoff_t	offset_fsb, last_block;
672 	xfs_fileoff_t	end_fsb, map_start_fsb;
673 	xfs_fsblock_t	first_block;
674 	xfs_bmap_free_t	free_list;
675 	xfs_filblks_t	count_fsb;
676 	xfs_bmbt_irec_t	imap;
677 	xfs_trans_t	*tp;
678 	int		nimaps, committed;
679 	int		error = 0;
680 	int		nres;
681 
682 	*retmap = 0;
683 
684 	/*
685 	 * Make sure that the dquots are there.
686 	 */
687 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
688 		return XFS_ERROR(error);
689 
690 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
691 	count_fsb = map->br_blockcount;
692 	map_start_fsb = map->br_startoff;
693 
694 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
695 
696 	while (count_fsb != 0) {
697 		/*
698 		 * Set up a transaction with which to allocate the
699 		 * backing store for the file.  Do allocations in a
700 		 * loop until we get some space in the range we are
701 		 * interested in.  The other space that might be allocated
702 		 * is in the delayed allocation extent on which we sit
703 		 * but before our buffer starts.
704 		 */
705 
706 		nimaps = 0;
707 		while (nimaps == 0) {
708 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
709 			tp->t_flags |= XFS_TRANS_RESERVE;
710 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
711 			error = xfs_trans_reserve(tp, nres,
712 					XFS_WRITE_LOG_RES(mp),
713 					0, XFS_TRANS_PERM_LOG_RES,
714 					XFS_WRITE_LOG_COUNT);
715 			if (error) {
716 				xfs_trans_cancel(tp, 0);
717 				return XFS_ERROR(error);
718 			}
719 			xfs_ilock(ip, XFS_ILOCK_EXCL);
720 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
721 			xfs_trans_ihold(tp, ip);
722 
723 			xfs_bmap_init(&free_list, &first_block);
724 
725 			/*
726 			 * it is possible that the extents have changed since
727 			 * we did the read call as we dropped the ilock for a
728 			 * while. We have to be careful about truncates or hole
729 			 * punchs here - we are not allowed to allocate
730 			 * non-delalloc blocks here.
731 			 *
732 			 * The only protection against truncation is the pages
733 			 * for the range we are being asked to convert are
734 			 * locked and hence a truncate will block on them
735 			 * first.
736 			 *
737 			 * As a result, if we go beyond the range we really
738 			 * need and hit an delalloc extent boundary followed by
739 			 * a hole while we have excess blocks in the map, we
740 			 * will fill the hole incorrectly and overrun the
741 			 * transaction reservation.
742 			 *
743 			 * Using a single map prevents this as we are forced to
744 			 * check each map we look for overlap with the desired
745 			 * range and abort as soon as we find it. Also, given
746 			 * that we only return a single map, having one beyond
747 			 * what we can return is probably a bit silly.
748 			 *
749 			 * We also need to check that we don't go beyond EOF;
750 			 * this is a truncate optimisation as a truncate sets
751 			 * the new file size before block on the pages we
752 			 * currently have locked under writeback. Because they
753 			 * are about to be tossed, we don't need to write them
754 			 * back....
755 			 */
756 			nimaps = 1;
757 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
758 			error = xfs_bmap_last_offset(NULL, ip, &last_block,
759 							XFS_DATA_FORK);
760 			if (error)
761 				goto trans_cancel;
762 
763 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
764 			if ((map_start_fsb + count_fsb) > last_block) {
765 				count_fsb = last_block - map_start_fsb;
766 				if (count_fsb == 0) {
767 					error = EAGAIN;
768 					goto trans_cancel;
769 				}
770 			}
771 
772 			/* Go get the actual blocks */
773 			error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
774 					XFS_BMAPI_WRITE, &first_block, 1,
775 					&imap, &nimaps, &free_list, NULL);
776 			if (error)
777 				goto trans_cancel;
778 
779 			error = xfs_bmap_finish(&tp, &free_list, &committed);
780 			if (error)
781 				goto trans_cancel;
782 
783 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
784 			if (error)
785 				goto error0;
786 
787 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
788 		}
789 
790 		/*
791 		 * See if we were able to allocate an extent that
792 		 * covers at least part of the callers request
793 		 */
794 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
795 			return xfs_cmn_err_fsblock_zero(ip, &imap);
796 
797 		if ((offset_fsb >= imap.br_startoff) &&
798 		    (offset_fsb < (imap.br_startoff +
799 				   imap.br_blockcount))) {
800 			*map = imap;
801 			*retmap = 1;
802 			XFS_STATS_INC(xs_xstrat_quick);
803 			return 0;
804 		}
805 
806 		/*
807 		 * So far we have not mapped the requested part of the
808 		 * file, just surrounding data, try again.
809 		 */
810 		count_fsb -= imap.br_blockcount;
811 		map_start_fsb = imap.br_startoff + imap.br_blockcount;
812 	}
813 
814 trans_cancel:
815 	xfs_bmap_cancel(&free_list);
816 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
817 error0:
818 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
819 	return XFS_ERROR(error);
820 }
821 
822 int
823 xfs_iomap_write_unwritten(
824 	xfs_inode_t	*ip,
825 	xfs_off_t	offset,
826 	size_t		count)
827 {
828 	xfs_mount_t	*mp = ip->i_mount;
829 	xfs_fileoff_t	offset_fsb;
830 	xfs_filblks_t	count_fsb;
831 	xfs_filblks_t	numblks_fsb;
832 	xfs_fsblock_t	firstfsb;
833 	int		nimaps;
834 	xfs_trans_t	*tp;
835 	xfs_bmbt_irec_t imap;
836 	xfs_bmap_free_t free_list;
837 	uint		resblks;
838 	int		committed;
839 	int		error;
840 
841 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN, ip, offset, count);
842 
843 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
844 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
845 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
846 
847 	/*
848 	 * Reserve enough blocks in this transaction for two complete extent
849 	 * btree splits.  We may be converting the middle part of an unwritten
850 	 * extent and in this case we will insert two new extents in the btree
851 	 * each of which could cause a full split.
852 	 *
853 	 * This reservation amount will be used in the first call to
854 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
855 	 * rest of the operation.
856 	 */
857 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
858 
859 	do {
860 		/*
861 		 * set up a transaction to convert the range of extents
862 		 * from unwritten to real. Do allocations in a loop until
863 		 * we have covered the range passed in.
864 		 */
865 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
866 		tp->t_flags |= XFS_TRANS_RESERVE;
867 		error = xfs_trans_reserve(tp, resblks,
868 				XFS_WRITE_LOG_RES(mp), 0,
869 				XFS_TRANS_PERM_LOG_RES,
870 				XFS_WRITE_LOG_COUNT);
871 		if (error) {
872 			xfs_trans_cancel(tp, 0);
873 			return XFS_ERROR(error);
874 		}
875 
876 		xfs_ilock(ip, XFS_ILOCK_EXCL);
877 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
878 		xfs_trans_ihold(tp, ip);
879 
880 		/*
881 		 * Modify the unwritten extent state of the buffer.
882 		 */
883 		xfs_bmap_init(&free_list, &firstfsb);
884 		nimaps = 1;
885 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
886 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
887 				  1, &imap, &nimaps, &free_list, NULL);
888 		if (error)
889 			goto error_on_bmapi_transaction;
890 
891 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
892 		if (error)
893 			goto error_on_bmapi_transaction;
894 
895 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
896 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
897 		if (error)
898 			return XFS_ERROR(error);
899 
900 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
901 			return xfs_cmn_err_fsblock_zero(ip, &imap);
902 
903 		if ((numblks_fsb = imap.br_blockcount) == 0) {
904 			/*
905 			 * The numblks_fsb value should always get
906 			 * smaller, otherwise the loop is stuck.
907 			 */
908 			ASSERT(imap.br_blockcount);
909 			break;
910 		}
911 		offset_fsb += numblks_fsb;
912 		count_fsb -= numblks_fsb;
913 	} while (count_fsb > 0);
914 
915 	return 0;
916 
917 error_on_bmapi_transaction:
918 	xfs_bmap_cancel(&free_list);
919 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
920 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
921 	return XFS_ERROR(error);
922 }
923