xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
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_iocore_t	*io,
57 	xfs_off_t	offset,
58 	ssize_t		count)
59 {
60 	xfs_inode_t	*ip = XFS_IO_INODE(io);
61 
62 	if (!ip->i_rwtrace)
63 		return;
64 
65 	ktrace_enter(ip->i_rwtrace,
66 		(void *)((unsigned long)tag),
67 		(void *)ip,
68 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
69 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
70 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
71 		(void *)((unsigned long)(offset & 0xffffffff)),
72 		(void *)((unsigned long)count),
73 		(void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
74 		(void *)((unsigned long)(io->io_new_size & 0xffffffff)),
75 		(void *)((unsigned long)current_pid()),
76 		(void *)NULL,
77 		(void *)NULL,
78 		(void *)NULL,
79 		(void *)NULL,
80 		(void *)NULL,
81 		(void *)NULL);
82 }
83 
84 void
85 xfs_iomap_map_trace(
86 	int		tag,
87 	xfs_iocore_t	*io,
88 	xfs_off_t	offset,
89 	ssize_t		count,
90 	xfs_iomap_t	*iomapp,
91 	xfs_bmbt_irec_t	*imapp,
92 	int		flags)
93 {
94 	xfs_inode_t	*ip = XFS_IO_INODE(io);
95 
96 	if (!ip->i_rwtrace)
97 		return;
98 
99 	ktrace_enter(ip->i_rwtrace,
100 		(void *)((unsigned long)tag),
101 		(void *)ip,
102 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
103 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
104 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
105 		(void *)((unsigned long)(offset & 0xffffffff)),
106 		(void *)((unsigned long)count),
107 		(void *)((unsigned long)flags),
108 		(void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
109 		(void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
110 		(void *)((unsigned long)(iomapp->iomap_delta)),
111 		(void *)((unsigned long)(iomapp->iomap_bsize)),
112 		(void *)((unsigned long)(iomapp->iomap_bn)),
113 		(void *)(__psint_t)(imapp->br_startoff),
114 		(void *)((unsigned long)(imapp->br_blockcount)),
115 		(void *)(__psint_t)(imapp->br_startblock));
116 }
117 #else
118 #define xfs_iomap_enter_trace(tag, io, offset, count)
119 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
120 #endif
121 
122 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
123 						<< mp->m_writeio_log)
124 #define XFS_STRAT_WRITE_IMAPS	2
125 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
126 
127 STATIC int
128 xfs_imap_to_bmap(
129 	xfs_iocore_t	*io,
130 	xfs_off_t	offset,
131 	xfs_bmbt_irec_t *imap,
132 	xfs_iomap_t	*iomapp,
133 	int		imaps,			/* Number of imap entries */
134 	int		iomaps,			/* Number of iomap entries */
135 	int		flags)
136 {
137 	xfs_mount_t	*mp;
138 	int		pbm;
139 	xfs_fsblock_t	start_block;
140 
141 	mp = io->io_mount;
142 
143 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
144 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
145 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
146 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
147 		iomapp->iomap_flags = flags;
148 
149 		if (io->io_flags & XFS_IOCORE_RT) {
150 			iomapp->iomap_flags |= IOMAP_REALTIME;
151 			iomapp->iomap_target = mp->m_rtdev_targp;
152 		} else {
153 			iomapp->iomap_target = mp->m_ddev_targp;
154 		}
155 		start_block = imap->br_startblock;
156 		if (start_block == HOLESTARTBLOCK) {
157 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
158 			iomapp->iomap_flags |= IOMAP_HOLE;
159 		} else if (start_block == DELAYSTARTBLOCK) {
160 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
161 			iomapp->iomap_flags |= IOMAP_DELAY;
162 		} else {
163 			iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
164 			if (ISUNWRITTEN(imap))
165 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
166 		}
167 
168 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
169 	}
170 	return pbm;	/* Return the number filled */
171 }
172 
173 int
174 xfs_iomap(
175 	xfs_iocore_t	*io,
176 	xfs_off_t	offset,
177 	ssize_t		count,
178 	int		flags,
179 	xfs_iomap_t	*iomapp,
180 	int		*niomaps)
181 {
182 	xfs_mount_t	*mp = io->io_mount;
183 	xfs_fileoff_t	offset_fsb, end_fsb;
184 	int		error = 0;
185 	int		lockmode = 0;
186 	xfs_bmbt_irec_t	imap;
187 	int		nimaps = 1;
188 	int		bmapi_flags = 0;
189 	int		iomap_flags = 0;
190 
191 	if (XFS_FORCED_SHUTDOWN(mp))
192 		return XFS_ERROR(EIO);
193 
194 	switch (flags &
195 		(BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
196 		 BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
197 	case BMAPI_READ:
198 		xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
199 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
200 		bmapi_flags = XFS_BMAPI_ENTIRE;
201 		break;
202 	case BMAPI_WRITE:
203 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
204 		lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
205 		if (flags & BMAPI_IGNSTATE)
206 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
207 		XFS_ILOCK(mp, io, lockmode);
208 		break;
209 	case BMAPI_ALLOCATE:
210 		xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
211 		lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
212 		bmapi_flags = XFS_BMAPI_ENTIRE;
213 		/* Attempt non-blocking lock */
214 		if (flags & BMAPI_TRYLOCK) {
215 			if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
216 				return XFS_ERROR(EAGAIN);
217 		} else {
218 			XFS_ILOCK(mp, io, lockmode);
219 		}
220 		break;
221 	case BMAPI_UNWRITTEN:
222 		goto phase2;
223 	case BMAPI_DEVICE:
224 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
225 		iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
226 			mp->m_rtdev_targp : mp->m_ddev_targp;
227 		error = 0;
228 		*niomaps = 1;
229 		goto out;
230 	default:
231 		BUG();
232 	}
233 
234 	ASSERT(offset <= mp->m_maxioffset);
235 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
236 		count = mp->m_maxioffset - offset;
237 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
238 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
239 
240 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
241 			(xfs_filblks_t)(end_fsb - offset_fsb),
242 			bmapi_flags,  NULL, 0, &imap,
243 			&nimaps, NULL, NULL);
244 
245 	if (error)
246 		goto out;
247 
248 phase2:
249 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
250 	case BMAPI_WRITE:
251 		/* If we found an extent, return it */
252 		if (nimaps &&
253 		    (imap.br_startblock != HOLESTARTBLOCK) &&
254 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
255 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
256 					offset, count, iomapp, &imap, flags);
257 			break;
258 		}
259 
260 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
261 			error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
262 					count, flags, &imap, &nimaps, nimaps);
263 		} else {
264 			error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
265 					flags, &imap, &nimaps);
266 		}
267 		if (!error) {
268 			xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
269 					offset, count, iomapp, &imap, flags);
270 		}
271 		iomap_flags = IOMAP_NEW;
272 		break;
273 	case BMAPI_ALLOCATE:
274 		/* If we found an extent, return it */
275 		XFS_IUNLOCK(mp, io, lockmode);
276 		lockmode = 0;
277 
278 		if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
279 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
280 					offset, count, iomapp, &imap, flags);
281 			break;
282 		}
283 
284 		error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
285 						 &imap, &nimaps);
286 		break;
287 	case BMAPI_UNWRITTEN:
288 		lockmode = 0;
289 		error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
290 		nimaps = 0;
291 		break;
292 	}
293 
294 	if (nimaps) {
295 		*niomaps = xfs_imap_to_bmap(io, offset, &imap,
296 					    iomapp, nimaps, *niomaps, iomap_flags);
297 	} else if (niomaps) {
298 		*niomaps = 0;
299 	}
300 
301 out:
302 	if (lockmode)
303 		XFS_IUNLOCK(mp, io, lockmode);
304 	return XFS_ERROR(error);
305 }
306 
307 STATIC int
308 xfs_iomap_eof_align_last_fsb(
309 	xfs_mount_t	*mp,
310 	xfs_iocore_t	*io,
311 	xfs_fsize_t	isize,
312 	xfs_extlen_t	extsize,
313 	xfs_fileoff_t	*last_fsb)
314 {
315 	xfs_fileoff_t	new_last_fsb = 0;
316 	xfs_extlen_t	align;
317 	int		eof, error;
318 
319 	if (io->io_flags & XFS_IOCORE_RT)
320 		;
321 	/*
322 	 * If mounted with the "-o swalloc" option, roundup the allocation
323 	 * request to a stripe width boundary if the file size is >=
324 	 * stripe width and we are allocating past the allocation eof.
325 	 */
326 	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
327 	        (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
328 		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
329 	/*
330 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
331 	 * if the file size is >= stripe unit size, and we are allocating past
332 	 * the allocation eof.
333 	 */
334 	else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
335 		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
336 
337 	/*
338 	 * Always round up the allocation request to an extent boundary
339 	 * (when file on a real-time subvolume or has di_extsize hint).
340 	 */
341 	if (extsize) {
342 		if (new_last_fsb)
343 			align = roundup_64(new_last_fsb, extsize);
344 		else
345 			align = extsize;
346 		new_last_fsb = roundup_64(*last_fsb, align);
347 	}
348 
349 	if (new_last_fsb) {
350 		error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
351 		if (error)
352 			return error;
353 		if (eof)
354 			*last_fsb = new_last_fsb;
355 	}
356 	return 0;
357 }
358 
359 STATIC int
360 xfs_flush_space(
361 	xfs_inode_t	*ip,
362 	int		*fsynced,
363 	int		*ioflags)
364 {
365 	switch (*fsynced) {
366 	case 0:
367 		if (ip->i_delayed_blks) {
368 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
369 			xfs_flush_inode(ip);
370 			xfs_ilock(ip, XFS_ILOCK_EXCL);
371 			*fsynced = 1;
372 		} else {
373 			*ioflags |= BMAPI_SYNC;
374 			*fsynced = 2;
375 		}
376 		return 0;
377 	case 1:
378 		*fsynced = 2;
379 		*ioflags |= BMAPI_SYNC;
380 		return 0;
381 	case 2:
382 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
383 		xfs_flush_device(ip);
384 		xfs_ilock(ip, XFS_ILOCK_EXCL);
385 		*fsynced = 3;
386 		return 0;
387 	}
388 	return 1;
389 }
390 
391 STATIC int
392 xfs_cmn_err_fsblock_zero(
393 	xfs_inode_t	*ip,
394 	xfs_bmbt_irec_t	*imap)
395 {
396 	xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
397 			"Access to block zero in inode %llu "
398 			"start_block: %llx start_off: %llx "
399 			"blkcnt: %llx extent-state: %x\n",
400 		(unsigned long long)ip->i_ino,
401 		(unsigned long long)imap->br_startblock,
402 		(unsigned long long)imap->br_startoff,
403 		(unsigned long long)imap->br_blockcount,
404 		imap->br_state);
405 	return EFSCORRUPTED;
406 }
407 
408 int
409 xfs_iomap_write_direct(
410 	xfs_inode_t	*ip,
411 	xfs_off_t	offset,
412 	size_t		count,
413 	int		flags,
414 	xfs_bmbt_irec_t *ret_imap,
415 	int		*nmaps,
416 	int		found)
417 {
418 	xfs_mount_t	*mp = ip->i_mount;
419 	xfs_iocore_t	*io = &ip->i_iocore;
420 	xfs_fileoff_t	offset_fsb;
421 	xfs_fileoff_t	last_fsb;
422 	xfs_filblks_t	count_fsb, resaligned;
423 	xfs_fsblock_t	firstfsb;
424 	xfs_extlen_t	extsz, temp;
425 	xfs_fsize_t	isize;
426 	int		nimaps;
427 	int		bmapi_flag;
428 	int		quota_flag;
429 	int		rt;
430 	xfs_trans_t	*tp;
431 	xfs_bmbt_irec_t imap;
432 	xfs_bmap_free_t free_list;
433 	uint		qblocks, resblks, resrtextents;
434 	int		committed;
435 	int		error;
436 
437 	/*
438 	 * Make sure that the dquots are there. This doesn't hold
439 	 * the ilock across a disk read.
440 	 */
441 	error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
442 	if (error)
443 		return XFS_ERROR(error);
444 
445 	rt = XFS_IS_REALTIME_INODE(ip);
446 	extsz = xfs_get_extsz_hint(ip);
447 
448 	isize = ip->i_size;
449 	if (io->io_new_size > isize)
450 		isize = io->io_new_size;
451 
452 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
453 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
454 	if ((offset + count) > isize) {
455 		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
456 							&last_fsb);
457 		if (error)
458 			goto error_out;
459 	} else {
460 		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
461 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
462 					ret_imap->br_blockcount +
463 					ret_imap->br_startoff);
464 	}
465 	count_fsb = last_fsb - offset_fsb;
466 	ASSERT(count_fsb > 0);
467 
468 	resaligned = count_fsb;
469 	if (unlikely(extsz)) {
470 		if ((temp = do_mod(offset_fsb, extsz)))
471 			resaligned += temp;
472 		if ((temp = do_mod(resaligned, extsz)))
473 			resaligned += extsz - temp;
474 	}
475 
476 	if (unlikely(rt)) {
477 		resrtextents = qblocks = resaligned;
478 		resrtextents /= mp->m_sb.sb_rextsize;
479 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
480 		quota_flag = XFS_QMOPT_RES_RTBLKS;
481 	} else {
482 		resrtextents = 0;
483 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
484 		quota_flag = XFS_QMOPT_RES_REGBLKS;
485 	}
486 
487 	/*
488 	 * Allocate and setup the transaction
489 	 */
490 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
491 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
492 	error = xfs_trans_reserve(tp, resblks,
493 			XFS_WRITE_LOG_RES(mp), resrtextents,
494 			XFS_TRANS_PERM_LOG_RES,
495 			XFS_WRITE_LOG_COUNT);
496 	/*
497 	 * Check for running out of space, note: need lock to return
498 	 */
499 	if (error)
500 		xfs_trans_cancel(tp, 0);
501 	xfs_ilock(ip, XFS_ILOCK_EXCL);
502 	if (error)
503 		goto error_out;
504 
505 	error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
506 					      qblocks, 0, quota_flag);
507 	if (error)
508 		goto error1;
509 
510 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
511 	xfs_trans_ihold(tp, ip);
512 
513 	bmapi_flag = XFS_BMAPI_WRITE;
514 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
515 		bmapi_flag |= XFS_BMAPI_PREALLOC;
516 
517 	/*
518 	 * Issue the xfs_bmapi() call to allocate the blocks
519 	 */
520 	XFS_BMAP_INIT(&free_list, &firstfsb);
521 	nimaps = 1;
522 	error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb, bmapi_flag,
523 		&firstfsb, 0, &imap, &nimaps, &free_list, NULL);
524 	if (error)
525 		goto error0;
526 
527 	/*
528 	 * Complete the transaction
529 	 */
530 	error = xfs_bmap_finish(&tp, &free_list, &committed);
531 	if (error)
532 		goto error0;
533 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
534 	if (error)
535 		goto error_out;
536 
537 	/*
538 	 * Copy any maps to caller's array and return any error.
539 	 */
540 	if (nimaps == 0) {
541 		error = ENOSPC;
542 		goto error_out;
543 	}
544 
545 	if (unlikely(!imap.br_startblock && !(io->io_flags & XFS_IOCORE_RT))) {
546 		error = xfs_cmn_err_fsblock_zero(ip, &imap);
547 		goto error_out;
548 	}
549 
550 	*ret_imap = imap;
551 	*nmaps = 1;
552 	return 0;
553 
554 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
555 	xfs_bmap_cancel(&free_list);
556 	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
557 
558 error1:	/* Just cancel transaction */
559 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
560 	*nmaps = 0;	/* nothing set-up here */
561 
562 error_out:
563 	return XFS_ERROR(error);
564 }
565 
566 /*
567  * If the caller is doing a write at the end of the file,
568  * then extend the allocation out to the file system's write
569  * iosize.  We clean up any extra space left over when the
570  * file is closed in xfs_inactive().
571  *
572  * For sync writes, we are flushing delayed allocate space to
573  * try to make additional space available for allocation near
574  * the filesystem full boundary - preallocation hurts in that
575  * situation, of course.
576  */
577 STATIC int
578 xfs_iomap_eof_want_preallocate(
579 	xfs_mount_t	*mp,
580 	xfs_iocore_t	*io,
581 	xfs_fsize_t	isize,
582 	xfs_off_t	offset,
583 	size_t		count,
584 	int		ioflag,
585 	xfs_bmbt_irec_t *imap,
586 	int		nimaps,
587 	int		*prealloc)
588 {
589 	xfs_fileoff_t   start_fsb;
590 	xfs_filblks_t   count_fsb;
591 	xfs_fsblock_t	firstblock;
592 	int		n, error, imaps;
593 
594 	*prealloc = 0;
595 	if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
596 		return 0;
597 
598 	/*
599 	 * If there are any real blocks past eof, then don't
600 	 * do any speculative allocation.
601 	 */
602 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
603 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
604 	while (count_fsb > 0) {
605 		imaps = nimaps;
606 		firstblock = NULLFSBLOCK;
607 		error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb, 0,
608 				  &firstblock, 0, imap, &imaps, NULL, NULL);
609 		if (error)
610 			return error;
611 		for (n = 0; n < imaps; n++) {
612 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
613 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
614 				return 0;
615 			start_fsb += imap[n].br_blockcount;
616 			count_fsb -= imap[n].br_blockcount;
617 		}
618 	}
619 	*prealloc = 1;
620 	return 0;
621 }
622 
623 int
624 xfs_iomap_write_delay(
625 	xfs_inode_t	*ip,
626 	xfs_off_t	offset,
627 	size_t		count,
628 	int		ioflag,
629 	xfs_bmbt_irec_t *ret_imap,
630 	int		*nmaps)
631 {
632 	xfs_mount_t	*mp = ip->i_mount;
633 	xfs_iocore_t	*io = &ip->i_iocore;
634 	xfs_fileoff_t	offset_fsb;
635 	xfs_fileoff_t	last_fsb;
636 	xfs_off_t	aligned_offset;
637 	xfs_fileoff_t	ioalign;
638 	xfs_fsblock_t	firstblock;
639 	xfs_extlen_t	extsz;
640 	xfs_fsize_t	isize;
641 	int		nimaps;
642 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
643 	int		prealloc, fsynced = 0;
644 	int		error;
645 
646 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
647 
648 	/*
649 	 * Make sure that the dquots are there. This doesn't hold
650 	 * the ilock across a disk read.
651 	 */
652 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
653 	if (error)
654 		return XFS_ERROR(error);
655 
656 	extsz = xfs_get_extsz_hint(ip);
657 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
658 
659 retry:
660 	isize = ip->i_size;
661 	if (io->io_new_size > isize)
662 		isize = io->io_new_size;
663 
664 	error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
665 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
666 	if (error)
667 		return error;
668 
669 	if (prealloc) {
670 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
671 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
672 		last_fsb = ioalign + mp->m_writeio_blocks;
673 	} else {
674 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
675 	}
676 
677 	if (prealloc || extsz) {
678 		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
679 							&last_fsb);
680 		if (error)
681 			return error;
682 	}
683 
684 	nimaps = XFS_WRITE_IMAPS;
685 	firstblock = NULLFSBLOCK;
686 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
687 			  (xfs_filblks_t)(last_fsb - offset_fsb),
688 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
689 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
690 			  &nimaps, NULL, NULL);
691 	if (error && (error != ENOSPC))
692 		return XFS_ERROR(error);
693 
694 	/*
695 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
696 	 * then we must have run out of space - flush delalloc, and retry..
697 	 */
698 	if (nimaps == 0) {
699 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
700 					io, offset, count);
701 		if (xfs_flush_space(ip, &fsynced, &ioflag))
702 			return XFS_ERROR(ENOSPC);
703 
704 		error = 0;
705 		goto retry;
706 	}
707 
708 	if (unlikely(!imap[0].br_startblock && !(io->io_flags & XFS_IOCORE_RT)))
709 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
710 
711 	*ret_imap = imap[0];
712 	*nmaps = 1;
713 
714 	return 0;
715 }
716 
717 /*
718  * Pass in a delayed allocate extent, convert it to real extents;
719  * return to the caller the extent we create which maps on top of
720  * the originating callers request.
721  *
722  * Called without a lock on the inode.
723  */
724 int
725 xfs_iomap_write_allocate(
726 	xfs_inode_t	*ip,
727 	xfs_off_t	offset,
728 	size_t		count,
729 	xfs_bmbt_irec_t *map,
730 	int		*retmap)
731 {
732 	xfs_mount_t	*mp = ip->i_mount;
733 	xfs_iocore_t    *io = &ip->i_iocore;
734 	xfs_fileoff_t	offset_fsb, last_block;
735 	xfs_fileoff_t	end_fsb, map_start_fsb;
736 	xfs_fsblock_t	first_block;
737 	xfs_bmap_free_t	free_list;
738 	xfs_filblks_t	count_fsb;
739 	xfs_bmbt_irec_t	imap[XFS_STRAT_WRITE_IMAPS];
740 	xfs_trans_t	*tp;
741 	int		i, nimaps, committed;
742 	int		error = 0;
743 	int		nres;
744 
745 	*retmap = 0;
746 
747 	/*
748 	 * Make sure that the dquots are there.
749 	 */
750 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
751 		return XFS_ERROR(error);
752 
753 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
754 	count_fsb = map->br_blockcount;
755 	map_start_fsb = map->br_startoff;
756 
757 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
758 
759 	while (count_fsb != 0) {
760 		/*
761 		 * Set up a transaction with which to allocate the
762 		 * backing store for the file.  Do allocations in a
763 		 * loop until we get some space in the range we are
764 		 * interested in.  The other space that might be allocated
765 		 * is in the delayed allocation extent on which we sit
766 		 * but before our buffer starts.
767 		 */
768 
769 		nimaps = 0;
770 		while (nimaps == 0) {
771 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
772 			tp->t_flags |= XFS_TRANS_RESERVE;
773 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
774 			error = xfs_trans_reserve(tp, nres,
775 					XFS_WRITE_LOG_RES(mp),
776 					0, XFS_TRANS_PERM_LOG_RES,
777 					XFS_WRITE_LOG_COUNT);
778 			if (error) {
779 				xfs_trans_cancel(tp, 0);
780 				return XFS_ERROR(error);
781 			}
782 			xfs_ilock(ip, XFS_ILOCK_EXCL);
783 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
784 			xfs_trans_ihold(tp, ip);
785 
786 			XFS_BMAP_INIT(&free_list, &first_block);
787 
788 			nimaps = XFS_STRAT_WRITE_IMAPS;
789 			/*
790 			 * Ensure we don't go beyond eof - it is possible
791 			 * the extents changed since we did the read call,
792 			 * we dropped the ilock in the interim.
793 			 */
794 
795 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
796 			xfs_bmap_last_offset(NULL, ip, &last_block,
797 				XFS_DATA_FORK);
798 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
799 			if ((map_start_fsb + count_fsb) > last_block) {
800 				count_fsb = last_block - map_start_fsb;
801 				if (count_fsb == 0) {
802 					error = EAGAIN;
803 					goto trans_cancel;
804 				}
805 			}
806 
807 			/* Go get the actual blocks */
808 			error = XFS_BMAPI(mp, tp, io, map_start_fsb, count_fsb,
809 					XFS_BMAPI_WRITE, &first_block, 1,
810 					imap, &nimaps, &free_list, NULL);
811 			if (error)
812 				goto trans_cancel;
813 
814 			error = xfs_bmap_finish(&tp, &free_list, &committed);
815 			if (error)
816 				goto trans_cancel;
817 
818 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
819 			if (error)
820 				goto error0;
821 
822 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
823 		}
824 
825 		/*
826 		 * See if we were able to allocate an extent that
827 		 * covers at least part of the callers request
828 		 */
829 		for (i = 0; i < nimaps; i++) {
830 			if (unlikely(!imap[i].br_startblock &&
831 				     !(io->io_flags & XFS_IOCORE_RT)))
832 				return xfs_cmn_err_fsblock_zero(ip, &imap[i]);
833 			if ((offset_fsb >= imap[i].br_startoff) &&
834 			    (offset_fsb < (imap[i].br_startoff +
835 					   imap[i].br_blockcount))) {
836 				*map = imap[i];
837 				*retmap = 1;
838 				XFS_STATS_INC(xs_xstrat_quick);
839 				return 0;
840 			}
841 			count_fsb -= imap[i].br_blockcount;
842 		}
843 
844 		/* So far we have not mapped the requested part of the
845 		 * file, just surrounding data, try again.
846 		 */
847 		nimaps--;
848 		map_start_fsb = imap[nimaps].br_startoff +
849 				imap[nimaps].br_blockcount;
850 	}
851 
852 trans_cancel:
853 	xfs_bmap_cancel(&free_list);
854 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
855 error0:
856 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
857 	return XFS_ERROR(error);
858 }
859 
860 int
861 xfs_iomap_write_unwritten(
862 	xfs_inode_t	*ip,
863 	xfs_off_t	offset,
864 	size_t		count)
865 {
866 	xfs_mount_t	*mp = ip->i_mount;
867 	xfs_iocore_t    *io = &ip->i_iocore;
868 	xfs_fileoff_t	offset_fsb;
869 	xfs_filblks_t	count_fsb;
870 	xfs_filblks_t	numblks_fsb;
871 	xfs_fsblock_t	firstfsb;
872 	int		nimaps;
873 	xfs_trans_t	*tp;
874 	xfs_bmbt_irec_t imap;
875 	xfs_bmap_free_t free_list;
876 	uint		resblks;
877 	int		committed;
878 	int		error;
879 
880 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
881 				&ip->i_iocore, offset, count);
882 
883 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
884 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
885 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
886 
887 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
888 
889 	do {
890 		/*
891 		 * set up a transaction to convert the range of extents
892 		 * from unwritten to real. Do allocations in a loop until
893 		 * we have covered the range passed in.
894 		 */
895 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
896 		tp->t_flags |= XFS_TRANS_RESERVE;
897 		error = xfs_trans_reserve(tp, resblks,
898 				XFS_WRITE_LOG_RES(mp), 0,
899 				XFS_TRANS_PERM_LOG_RES,
900 				XFS_WRITE_LOG_COUNT);
901 		if (error) {
902 			xfs_trans_cancel(tp, 0);
903 			return XFS_ERROR(error);
904 		}
905 
906 		xfs_ilock(ip, XFS_ILOCK_EXCL);
907 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
908 		xfs_trans_ihold(tp, ip);
909 
910 		/*
911 		 * Modify the unwritten extent state of the buffer.
912 		 */
913 		XFS_BMAP_INIT(&free_list, &firstfsb);
914 		nimaps = 1;
915 		error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb,
916 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
917 				  1, &imap, &nimaps, &free_list, NULL);
918 		if (error)
919 			goto error_on_bmapi_transaction;
920 
921 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
922 		if (error)
923 			goto error_on_bmapi_transaction;
924 
925 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
926 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
927 		if (error)
928 			return XFS_ERROR(error);
929 
930 		if (unlikely(!imap.br_startblock &&
931 			     !(io->io_flags & XFS_IOCORE_RT)))
932 			return xfs_cmn_err_fsblock_zero(ip, &imap);
933 
934 		if ((numblks_fsb = imap.br_blockcount) == 0) {
935 			/*
936 			 * The numblks_fsb value should always get
937 			 * smaller, otherwise the loop is stuck.
938 			 */
939 			ASSERT(imap.br_blockcount);
940 			break;
941 		}
942 		offset_fsb += numblks_fsb;
943 		count_fsb -= numblks_fsb;
944 	} while (count_fsb > 0);
945 
946 	return 0;
947 
948 error_on_bmapi_transaction:
949 	xfs_bmap_cancel(&free_list);
950 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
951 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
952 	return XFS_ERROR(error);
953 }
954