xref: /linux/fs/xfs/xfs_iomap.c (revision 776cfebb430c7b22c208b1b17add97f354d97cab)
1 /*
2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.	 Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 
33 #include "xfs.h"
34 
35 #include "xfs_fs.h"
36 #include "xfs_inum.h"
37 #include "xfs_log.h"
38 #include "xfs_trans.h"
39 #include "xfs_sb.h"
40 #include "xfs_ag.h"
41 #include "xfs_dir.h"
42 #include "xfs_dir2.h"
43 #include "xfs_alloc.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_quota.h"
46 #include "xfs_mount.h"
47 #include "xfs_alloc_btree.h"
48 #include "xfs_bmap_btree.h"
49 #include "xfs_ialloc_btree.h"
50 #include "xfs_btree.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode.h"
57 #include "xfs_bmap.h"
58 #include "xfs_bit.h"
59 #include "xfs_rtalloc.h"
60 #include "xfs_error.h"
61 #include "xfs_itable.h"
62 #include "xfs_rw.h"
63 #include "xfs_acl.h"
64 #include "xfs_cap.h"
65 #include "xfs_mac.h"
66 #include "xfs_attr.h"
67 #include "xfs_buf_item.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_utils.h"
70 #include "xfs_iomap.h"
71 
72 #if defined(XFS_RW_TRACE)
73 void
74 xfs_iomap_enter_trace(
75 	int		tag,
76 	xfs_iocore_t	*io,
77 	xfs_off_t	offset,
78 	ssize_t		count)
79 {
80 	xfs_inode_t	*ip = XFS_IO_INODE(io);
81 
82 	if (!ip->i_rwtrace)
83 		return;
84 
85 	ktrace_enter(ip->i_rwtrace,
86 		(void *)((unsigned long)tag),
87 		(void *)ip,
88 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
89 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
90 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
91 		(void *)((unsigned long)(offset & 0xffffffff)),
92 		(void *)((unsigned long)count),
93 		(void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
94 		(void *)((unsigned long)(io->io_new_size & 0xffffffff)),
95 		(void *)NULL,
96 		(void *)NULL,
97 		(void *)NULL,
98 		(void *)NULL,
99 		(void *)NULL,
100 		(void *)NULL,
101 		(void *)NULL);
102 }
103 
104 void
105 xfs_iomap_map_trace(
106 	int		tag,
107 	xfs_iocore_t	*io,
108 	xfs_off_t	offset,
109 	ssize_t		count,
110 	xfs_iomap_t	*iomapp,
111 	xfs_bmbt_irec_t	*imapp,
112 	int		flags)
113 {
114 	xfs_inode_t	*ip = XFS_IO_INODE(io);
115 
116 	if (!ip->i_rwtrace)
117 		return;
118 
119 	ktrace_enter(ip->i_rwtrace,
120 		(void *)((unsigned long)tag),
121 		(void *)ip,
122 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
123 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
124 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
125 		(void *)((unsigned long)(offset & 0xffffffff)),
126 		(void *)((unsigned long)count),
127 		(void *)((unsigned long)flags),
128 		(void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
129 		(void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
130 		(void *)((unsigned long)(iomapp->iomap_delta)),
131 		(void *)((unsigned long)(iomapp->iomap_bsize)),
132 		(void *)((unsigned long)(iomapp->iomap_bn)),
133 		(void *)(__psint_t)(imapp->br_startoff),
134 		(void *)((unsigned long)(imapp->br_blockcount)),
135 		(void *)(__psint_t)(imapp->br_startblock));
136 }
137 #else
138 #define xfs_iomap_enter_trace(tag, io, offset, count)
139 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
140 #endif
141 
142 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
143 						<< mp->m_writeio_log)
144 #define XFS_STRAT_WRITE_IMAPS	2
145 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
146 
147 STATIC int
148 xfs_imap_to_bmap(
149 	xfs_iocore_t	*io,
150 	xfs_off_t	offset,
151 	xfs_bmbt_irec_t *imap,
152 	xfs_iomap_t	*iomapp,
153 	int		imaps,			/* Number of imap entries */
154 	int		iomaps,			/* Number of iomap entries */
155 	int		flags)
156 {
157 	xfs_mount_t	*mp;
158 	xfs_fsize_t	nisize;
159 	int		pbm;
160 	xfs_fsblock_t	start_block;
161 
162 	mp = io->io_mount;
163 	nisize = XFS_SIZE(mp, io);
164 	if (io->io_new_size > nisize)
165 		nisize = io->io_new_size;
166 
167 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
168 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
169 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
170 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
171 		iomapp->iomap_flags = flags;
172 
173 		if (io->io_flags & XFS_IOCORE_RT) {
174 			iomapp->iomap_flags |= IOMAP_REALTIME;
175 			iomapp->iomap_target = mp->m_rtdev_targp;
176 		} else {
177 			iomapp->iomap_target = mp->m_ddev_targp;
178 		}
179 		start_block = imap->br_startblock;
180 		if (start_block == HOLESTARTBLOCK) {
181 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
182 			iomapp->iomap_flags |= IOMAP_HOLE;
183 		} else if (start_block == DELAYSTARTBLOCK) {
184 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
185 			iomapp->iomap_flags |= IOMAP_DELAY;
186 		} else {
187 			iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
188 			if (ISUNWRITTEN(imap))
189 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
190 		}
191 
192 		if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
193 			iomapp->iomap_flags |= IOMAP_EOF;
194 		}
195 
196 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
197 	}
198 	return pbm;	/* Return the number filled */
199 }
200 
201 int
202 xfs_iomap(
203 	xfs_iocore_t	*io,
204 	xfs_off_t	offset,
205 	ssize_t		count,
206 	int		flags,
207 	xfs_iomap_t	*iomapp,
208 	int		*niomaps)
209 {
210 	xfs_mount_t	*mp = io->io_mount;
211 	xfs_fileoff_t	offset_fsb, end_fsb;
212 	int		error = 0;
213 	int		lockmode = 0;
214 	xfs_bmbt_irec_t	imap;
215 	int		nimaps = 1;
216 	int		bmapi_flags = 0;
217 	int		iomap_flags = 0;
218 
219 	if (XFS_FORCED_SHUTDOWN(mp))
220 		return XFS_ERROR(EIO);
221 
222 	switch (flags &
223 		(BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
224 		 BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
225 	case BMAPI_READ:
226 		xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
227 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
228 		bmapi_flags = XFS_BMAPI_ENTIRE;
229 		if (flags & BMAPI_IGNSTATE)
230 			bmapi_flags |= XFS_BMAPI_IGSTATE;
231 		break;
232 	case BMAPI_WRITE:
233 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
234 		lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
235 		bmapi_flags = 0;
236 		XFS_ILOCK(mp, io, lockmode);
237 		break;
238 	case BMAPI_ALLOCATE:
239 		xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
240 		lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
241 		bmapi_flags = XFS_BMAPI_ENTIRE;
242 		/* Attempt non-blocking lock */
243 		if (flags & BMAPI_TRYLOCK) {
244 			if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
245 				return XFS_ERROR(EAGAIN);
246 		} else {
247 			XFS_ILOCK(mp, io, lockmode);
248 		}
249 		break;
250 	case BMAPI_UNWRITTEN:
251 		goto phase2;
252 	case BMAPI_DEVICE:
253 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
254 		iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
255 			mp->m_rtdev_targp : mp->m_ddev_targp;
256 		error = 0;
257 		*niomaps = 1;
258 		goto out;
259 	default:
260 		BUG();
261 	}
262 
263 	ASSERT(offset <= mp->m_maxioffset);
264 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
265 		count = mp->m_maxioffset - offset;
266 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
267 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
268 
269 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
270 			(xfs_filblks_t)(end_fsb - offset_fsb),
271 			bmapi_flags,  NULL, 0, &imap,
272 			&nimaps, NULL);
273 
274 	if (error)
275 		goto out;
276 
277 phase2:
278 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
279 	case BMAPI_WRITE:
280 		/* If we found an extent, return it */
281 		if (nimaps && (imap.br_startblock != HOLESTARTBLOCK)) {
282 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
283 					offset, count, iomapp, &imap, flags);
284 			break;
285 		}
286 
287 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
288 			error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
289 					count, flags, &imap, &nimaps, nimaps);
290 		} else {
291 			error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
292 					flags, &imap, &nimaps);
293 		}
294 		if (!error) {
295 			xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
296 					offset, count, iomapp, &imap, flags);
297 		}
298 		iomap_flags = IOMAP_NEW;
299 		break;
300 	case BMAPI_ALLOCATE:
301 		/* If we found an extent, return it */
302 		XFS_IUNLOCK(mp, io, lockmode);
303 		lockmode = 0;
304 
305 		if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
306 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
307 					offset, count, iomapp, &imap, flags);
308 			break;
309 		}
310 
311 		error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
312 						 &imap, &nimaps);
313 		break;
314 	case BMAPI_UNWRITTEN:
315 		lockmode = 0;
316 		error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
317 		nimaps = 0;
318 		break;
319 	}
320 
321 	if (nimaps) {
322 		*niomaps = xfs_imap_to_bmap(io, offset, &imap,
323 					    iomapp, nimaps, *niomaps, iomap_flags);
324 	} else if (niomaps) {
325 		*niomaps = 0;
326 	}
327 
328 out:
329 	if (lockmode)
330 		XFS_IUNLOCK(mp, io, lockmode);
331 	return XFS_ERROR(error);
332 }
333 
334 STATIC int
335 xfs_flush_space(
336 	xfs_inode_t	*ip,
337 	int		*fsynced,
338 	int		*ioflags)
339 {
340 	switch (*fsynced) {
341 	case 0:
342 		if (ip->i_delayed_blks) {
343 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
344 			xfs_flush_inode(ip);
345 			xfs_ilock(ip, XFS_ILOCK_EXCL);
346 			*fsynced = 1;
347 		} else {
348 			*ioflags |= BMAPI_SYNC;
349 			*fsynced = 2;
350 		}
351 		return 0;
352 	case 1:
353 		*fsynced = 2;
354 		*ioflags |= BMAPI_SYNC;
355 		return 0;
356 	case 2:
357 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
358 		xfs_flush_device(ip);
359 		xfs_ilock(ip, XFS_ILOCK_EXCL);
360 		*fsynced = 3;
361 		return 0;
362 	}
363 	return 1;
364 }
365 
366 int
367 xfs_iomap_write_direct(
368 	xfs_inode_t	*ip,
369 	xfs_off_t	offset,
370 	size_t		count,
371 	int		flags,
372 	xfs_bmbt_irec_t *ret_imap,
373 	int		*nmaps,
374 	int		found)
375 {
376 	xfs_mount_t	*mp = ip->i_mount;
377 	xfs_iocore_t	*io = &ip->i_iocore;
378 	xfs_fileoff_t	offset_fsb;
379 	xfs_fileoff_t	last_fsb;
380 	xfs_filblks_t	count_fsb;
381 	xfs_fsize_t	isize;
382 	xfs_fsblock_t	firstfsb;
383 	int		nimaps, maps;
384 	int		error;
385 	int		bmapi_flag;
386 	int		rt;
387 	xfs_trans_t	*tp;
388 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS], *imapp;
389 	xfs_bmap_free_t free_list;
390 	int		aeof;
391 	xfs_filblks_t	datablocks;
392 	int		committed;
393 	int		numrtextents;
394 	uint		resblks;
395 
396 	/*
397 	 * Make sure that the dquots are there. This doesn't hold
398 	 * the ilock across a disk read.
399 	 */
400 	error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
401 	if (error)
402 		return XFS_ERROR(error);
403 
404 	maps = min(XFS_WRITE_IMAPS, *nmaps);
405 	nimaps = maps;
406 
407 	isize = ip->i_d.di_size;
408 	aeof = (offset + count) > isize;
409 
410 	if (io->io_new_size > isize)
411 		isize = io->io_new_size;
412 
413 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
414 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
415 	count_fsb = last_fsb - offset_fsb;
416 	if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
417 		xfs_fileoff_t	map_last_fsb;
418 
419 		map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
420 
421 		if (map_last_fsb < last_fsb) {
422 			last_fsb = map_last_fsb;
423 			count_fsb = last_fsb - offset_fsb;
424 		}
425 		ASSERT(count_fsb > 0);
426 	}
427 
428 	/*
429 	 * determine if reserving space on
430 	 * the data or realtime partition.
431 	 */
432 	if ((rt = XFS_IS_REALTIME_INODE(ip))) {
433 		int	sbrtextsize, iprtextsize;
434 
435 		sbrtextsize = mp->m_sb.sb_rextsize;
436 		iprtextsize =
437 			ip->i_d.di_extsize ? ip->i_d.di_extsize : sbrtextsize;
438 		numrtextents = (count_fsb + iprtextsize - 1);
439 		do_div(numrtextents, sbrtextsize);
440 		datablocks = 0;
441 	} else {
442 		datablocks = count_fsb;
443 		numrtextents = 0;
444 	}
445 
446 	/*
447 	 * allocate and setup the transaction
448 	 */
449 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
450 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
451 
452 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
453 
454 	error = xfs_trans_reserve(tp, resblks,
455 			XFS_WRITE_LOG_RES(mp), numrtextents,
456 			XFS_TRANS_PERM_LOG_RES,
457 			XFS_WRITE_LOG_COUNT);
458 
459 	/*
460 	 * check for running out of space
461 	 */
462 	if (error)
463 		/*
464 		 * Free the transaction structure.
465 		 */
466 		xfs_trans_cancel(tp, 0);
467 
468 	xfs_ilock(ip, XFS_ILOCK_EXCL);
469 
470 	if (error)
471 		goto error_out; /* Don't return in above if .. trans ..,
472 					need lock to return */
473 
474 	if (XFS_TRANS_RESERVE_BLKQUOTA(mp, tp, ip, resblks)) {
475 		error = (EDQUOT);
476 		goto error1;
477 	}
478 	nimaps = 1;
479 
480 	bmapi_flag = XFS_BMAPI_WRITE;
481 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
482 	xfs_trans_ihold(tp, ip);
483 
484 	if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
485 		bmapi_flag |= XFS_BMAPI_PREALLOC;
486 
487 	/*
488 	 * issue the bmapi() call to allocate the blocks
489 	 */
490 	XFS_BMAP_INIT(&free_list, &firstfsb);
491 	imapp = &imap[0];
492 	error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
493 		bmapi_flag, &firstfsb, 0, imapp, &nimaps, &free_list);
494 	if (error) {
495 		goto error0;
496 	}
497 
498 	/*
499 	 * complete the transaction
500 	 */
501 
502 	error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
503 	if (error) {
504 		goto error0;
505 	}
506 
507 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
508 	if (error) {
509 		goto error_out;
510 	}
511 
512 	/* copy any maps to caller's array and return any error. */
513 	if (nimaps == 0) {
514 		error = (ENOSPC);
515 		goto error_out;
516 	}
517 
518 	*ret_imap = imap[0];
519 	*nmaps = 1;
520 	if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
521                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
522                         "start_block : %llx start_off : %llx blkcnt : %llx "
523                         "extent-state : %x \n",
524                         (ip->i_mount)->m_fsname,
525                         (long long)ip->i_ino,
526                         ret_imap->br_startblock, ret_imap->br_startoff,
527                         ret_imap->br_blockcount,ret_imap->br_state);
528         }
529 	return 0;
530 
531  error0:	/* Cancel bmap, unlock inode, and cancel trans */
532 	xfs_bmap_cancel(&free_list);
533 
534  error1:	/* Just cancel transaction */
535 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
536 	*nmaps = 0;	/* nothing set-up here */
537 
538 error_out:
539 	return XFS_ERROR(error);
540 }
541 
542 int
543 xfs_iomap_write_delay(
544 	xfs_inode_t	*ip,
545 	xfs_off_t	offset,
546 	size_t		count,
547 	int		ioflag,
548 	xfs_bmbt_irec_t *ret_imap,
549 	int		*nmaps)
550 {
551 	xfs_mount_t	*mp = ip->i_mount;
552 	xfs_iocore_t	*io = &ip->i_iocore;
553 	xfs_fileoff_t	offset_fsb;
554 	xfs_fileoff_t	last_fsb;
555 	xfs_fsize_t	isize;
556 	xfs_fsblock_t	firstblock;
557 	int		nimaps;
558 	int		error;
559 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
560 	int		aeof;
561 	int		fsynced = 0;
562 
563 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
564 
565 	/*
566 	 * Make sure that the dquots are there. This doesn't hold
567 	 * the ilock across a disk read.
568 	 */
569 
570 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
571 	if (error)
572 		return XFS_ERROR(error);
573 
574 retry:
575 	isize = ip->i_d.di_size;
576 	if (io->io_new_size > isize) {
577 		isize = io->io_new_size;
578 	}
579 
580 	aeof = 0;
581 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
582 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
583 	/*
584 	 * If the caller is doing a write at the end of the file,
585 	 * then extend the allocation (and the buffer used for the write)
586 	 * out to the file system's write iosize.  We clean up any extra
587 	 * space left over when the file is closed in xfs_inactive().
588 	 *
589 	 * For sync writes, we are flushing delayed allocate space to
590 	 * try to make additional space available for allocation near
591 	 * the filesystem full boundary - preallocation hurts in that
592 	 * situation, of course.
593 	 */
594 	if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
595 		xfs_off_t	aligned_offset;
596 		xfs_filblks_t   count_fsb;
597 		unsigned int	iosize;
598 		xfs_fileoff_t	ioalign;
599 		int		n;
600 		xfs_fileoff_t   start_fsb;
601 
602 		/*
603 		 * If there are any real blocks past eof, then don't
604 		 * do any speculative allocation.
605 		 */
606 		start_fsb = XFS_B_TO_FSBT(mp,
607 					((xfs_ufsize_t)(offset + count - 1)));
608 		count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
609 		while (count_fsb > 0) {
610 			nimaps = XFS_WRITE_IMAPS;
611 			error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
612 					0, &firstblock, 0, imap, &nimaps, NULL);
613 			if (error) {
614 				return error;
615 			}
616 			for (n = 0; n < nimaps; n++) {
617 				if ( !(io->io_flags & XFS_IOCORE_RT)  &&
618 					!imap[n].br_startblock) {
619 					cmn_err(CE_PANIC,"Access to block "
620 						"zero:  fs <%s> inode: %lld "
621 						"start_block : %llx start_off "
622 						": %llx blkcnt : %llx "
623 						"extent-state : %x \n",
624 						(ip->i_mount)->m_fsname,
625 						(long long)ip->i_ino,
626 						imap[n].br_startblock,
627 						imap[n].br_startoff,
628 						imap[n].br_blockcount,
629 						imap[n].br_state);
630         			}
631 				if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
632 				    (imap[n].br_startblock != DELAYSTARTBLOCK)) {
633 					goto write_map;
634 				}
635 				start_fsb += imap[n].br_blockcount;
636 				count_fsb -= imap[n].br_blockcount;
637 			}
638 		}
639 		iosize = mp->m_writeio_blocks;
640 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
641 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
642 		last_fsb = ioalign + iosize;
643 		aeof = 1;
644 	}
645 write_map:
646 	nimaps = XFS_WRITE_IMAPS;
647 	firstblock = NULLFSBLOCK;
648 
649 	/*
650 	 * If mounted with the "-o swalloc" option, roundup the allocation
651 	 * request to a stripe width boundary if the file size is >=
652 	 * stripe width and we are allocating past the allocation eof.
653 	 */
654 	if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth
655 	    && (mp->m_flags & XFS_MOUNT_SWALLOC)
656 	    && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
657 		int eof;
658 		xfs_fileoff_t new_last_fsb;
659 
660 		new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
661 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
662 		if (error) {
663 			return error;
664 		}
665 		if (eof) {
666 			last_fsb = new_last_fsb;
667 		}
668 	/*
669 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
670 	 * if the file size is >= stripe unit size, and we are allocating past
671 	 * the allocation eof.
672 	 */
673 	} else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
674 		   (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
675 		int eof;
676 		xfs_fileoff_t new_last_fsb;
677 		new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
678 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
679 		if (error) {
680 			return error;
681 		}
682 		if (eof) {
683 			last_fsb = new_last_fsb;
684 		}
685 	/*
686 	 * Round up the allocation request to a real-time extent boundary
687 	 * if the file is on the real-time subvolume.
688 	 */
689 	} else if (io->io_flags & XFS_IOCORE_RT && aeof) {
690 		int eof;
691 		xfs_fileoff_t new_last_fsb;
692 
693 		new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
694 		error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
695 		if (error) {
696 			return error;
697 		}
698 		if (eof)
699 			last_fsb = new_last_fsb;
700 	}
701 	error = xfs_bmapi(NULL, ip, offset_fsb,
702 			  (xfs_filblks_t)(last_fsb - offset_fsb),
703 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
704 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
705 			  &nimaps, NULL);
706 	/*
707 	 * This can be EDQUOT, if nimaps == 0
708 	 */
709 	if (error && (error != ENOSPC)) {
710 		return XFS_ERROR(error);
711 	}
712 	/*
713 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
714 	 * then we must have run out of space.
715 	 */
716 	if (nimaps == 0) {
717 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
718 					io, offset, count);
719 		if (xfs_flush_space(ip, &fsynced, &ioflag))
720 			return XFS_ERROR(ENOSPC);
721 
722 		error = 0;
723 		goto retry;
724 	}
725 
726 	*ret_imap = imap[0];
727 	*nmaps = 1;
728 	if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
729 		cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
730                         "start_block : %llx start_off : %llx blkcnt : %llx "
731                         "extent-state : %x \n",
732                         (ip->i_mount)->m_fsname,
733                         (long long)ip->i_ino,
734                         ret_imap->br_startblock, ret_imap->br_startoff,
735                         ret_imap->br_blockcount,ret_imap->br_state);
736 	}
737 	return 0;
738 }
739 
740 /*
741  * Pass in a delayed allocate extent, convert it to real extents;
742  * return to the caller the extent we create which maps on top of
743  * the originating callers request.
744  *
745  * Called without a lock on the inode.
746  */
747 int
748 xfs_iomap_write_allocate(
749 	xfs_inode_t	*ip,
750 	xfs_off_t	offset,
751 	size_t		count,
752 	xfs_bmbt_irec_t *map,
753 	int		*retmap)
754 {
755 	xfs_mount_t	*mp = ip->i_mount;
756 	xfs_iocore_t    *io = &ip->i_iocore;
757 	xfs_fileoff_t	offset_fsb, last_block;
758 	xfs_fileoff_t	end_fsb, map_start_fsb;
759 	xfs_fsblock_t	first_block;
760 	xfs_bmap_free_t	free_list;
761 	xfs_filblks_t	count_fsb;
762 	xfs_bmbt_irec_t	imap[XFS_STRAT_WRITE_IMAPS];
763 	xfs_trans_t	*tp;
764 	int		i, nimaps, committed;
765 	int		error = 0;
766 	int		nres;
767 
768 	*retmap = 0;
769 
770 	/*
771 	 * Make sure that the dquots are there.
772 	 */
773 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
774 		return XFS_ERROR(error);
775 
776 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
777 	count_fsb = map->br_blockcount;
778 	map_start_fsb = map->br_startoff;
779 
780 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
781 
782 	while (count_fsb != 0) {
783 		/*
784 		 * Set up a transaction with which to allocate the
785 		 * backing store for the file.  Do allocations in a
786 		 * loop until we get some space in the range we are
787 		 * interested in.  The other space that might be allocated
788 		 * is in the delayed allocation extent on which we sit
789 		 * but before our buffer starts.
790 		 */
791 
792 		nimaps = 0;
793 		while (nimaps == 0) {
794 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
795 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
796 			error = xfs_trans_reserve(tp, nres,
797 					XFS_WRITE_LOG_RES(mp),
798 					0, XFS_TRANS_PERM_LOG_RES,
799 					XFS_WRITE_LOG_COUNT);
800 			if (error == ENOSPC) {
801 				error = xfs_trans_reserve(tp, 0,
802 						XFS_WRITE_LOG_RES(mp),
803 						0,
804 						XFS_TRANS_PERM_LOG_RES,
805 						XFS_WRITE_LOG_COUNT);
806 			}
807 			if (error) {
808 				xfs_trans_cancel(tp, 0);
809 				return XFS_ERROR(error);
810 			}
811 			xfs_ilock(ip, XFS_ILOCK_EXCL);
812 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
813 			xfs_trans_ihold(tp, ip);
814 
815 			XFS_BMAP_INIT(&free_list, &first_block);
816 
817 			nimaps = XFS_STRAT_WRITE_IMAPS;
818 			/*
819 			 * Ensure we don't go beyond eof - it is possible
820 			 * the extents changed since we did the read call,
821 			 * we dropped the ilock in the interim.
822 			 */
823 
824 			end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
825 			xfs_bmap_last_offset(NULL, ip, &last_block,
826 				XFS_DATA_FORK);
827 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
828 			if ((map_start_fsb + count_fsb) > last_block) {
829 				count_fsb = last_block - map_start_fsb;
830 				if (count_fsb == 0) {
831 					error = EAGAIN;
832 					goto trans_cancel;
833 				}
834 			}
835 
836 			/* Go get the actual blocks */
837 			error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
838 					XFS_BMAPI_WRITE, &first_block, 1,
839 					imap, &nimaps, &free_list);
840 			if (error)
841 				goto trans_cancel;
842 
843 			error = xfs_bmap_finish(&tp, &free_list,
844 					first_block, &committed);
845 			if (error)
846 				goto trans_cancel;
847 
848 			error = xfs_trans_commit(tp,
849 					XFS_TRANS_RELEASE_LOG_RES, NULL);
850 			if (error)
851 				goto error0;
852 
853 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
854 		}
855 
856 		/*
857 		 * See if we were able to allocate an extent that
858 		 * covers at least part of the callers request
859 		 */
860 
861 		for (i = 0; i < nimaps; i++) {
862 			if ( !(io->io_flags & XFS_IOCORE_RT)  &&
863 				!imap[i].br_startblock) {
864 				cmn_err(CE_PANIC,"Access to block zero:  "
865 					"fs <%s> inode: %lld "
866 					"start_block : %llx start_off : %llx "
867 					"blkcnt : %llx extent-state : %x \n",
868 					(ip->i_mount)->m_fsname,
869 					(long long)ip->i_ino,
870 					imap[i].br_startblock,
871 					imap[i].br_startoff,
872 				        imap[i].br_blockcount,imap[i].br_state);
873                         }
874 			if ((offset_fsb >= imap[i].br_startoff) &&
875 			    (offset_fsb < (imap[i].br_startoff +
876 					   imap[i].br_blockcount))) {
877 				*map = imap[i];
878 				*retmap = 1;
879 				XFS_STATS_INC(xs_xstrat_quick);
880 				return 0;
881 			}
882 			count_fsb -= imap[i].br_blockcount;
883 		}
884 
885 		/* So far we have not mapped the requested part of the
886 		 * file, just surrounding data, try again.
887 		 */
888 		nimaps--;
889 		map_start_fsb = imap[nimaps].br_startoff +
890 				imap[nimaps].br_blockcount;
891 	}
892 
893 trans_cancel:
894 	xfs_bmap_cancel(&free_list);
895 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
896 error0:
897 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
898 	return XFS_ERROR(error);
899 }
900 
901 int
902 xfs_iomap_write_unwritten(
903 	xfs_inode_t	*ip,
904 	xfs_off_t	offset,
905 	size_t		count)
906 {
907 	xfs_mount_t	*mp = ip->i_mount;
908 	xfs_iocore_t    *io = &ip->i_iocore;
909 	xfs_trans_t	*tp;
910 	xfs_fileoff_t	offset_fsb;
911 	xfs_filblks_t	count_fsb;
912 	xfs_filblks_t	numblks_fsb;
913 	xfs_bmbt_irec_t	imap;
914 	int		committed;
915 	int		error;
916 	int		nres;
917 	int		nimaps;
918 	xfs_fsblock_t	firstfsb;
919 	xfs_bmap_free_t	free_list;
920 
921 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
922 				&ip->i_iocore, offset, count);
923 
924 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
925 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
926 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
927 
928 	do {
929 		nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
930 
931 		/*
932 		 * set up a transaction to convert the range of extents
933 		 * from unwritten to real. Do allocations in a loop until
934 		 * we have covered the range passed in.
935 		 */
936 
937 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
938 		error = xfs_trans_reserve(tp, nres,
939 				XFS_WRITE_LOG_RES(mp), 0,
940 				XFS_TRANS_PERM_LOG_RES,
941 				XFS_WRITE_LOG_COUNT);
942 		if (error) {
943 			xfs_trans_cancel(tp, 0);
944 			goto error0;
945 		}
946 
947 		xfs_ilock(ip, XFS_ILOCK_EXCL);
948 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
949 		xfs_trans_ihold(tp, ip);
950 
951 		/*
952 		 * Modify the unwritten extent state of the buffer.
953 		 */
954 		XFS_BMAP_INIT(&free_list, &firstfsb);
955 		nimaps = 1;
956 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
957 				  XFS_BMAPI_WRITE, &firstfsb,
958 				  1, &imap, &nimaps, &free_list);
959 		if (error)
960 			goto error_on_bmapi_transaction;
961 
962 		error = xfs_bmap_finish(&(tp), &(free_list),
963 				firstfsb, &committed);
964 		if (error)
965 			goto error_on_bmapi_transaction;
966 
967 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
968 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
969 		if (error)
970 			goto error0;
971 
972 		if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
973 			cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
974 				"inode: %lld start_block : %llx start_off : "
975 				"%llx blkcnt : %llx extent-state : %x \n",
976 				(ip->i_mount)->m_fsname,
977 				(long long)ip->i_ino,
978 				imap.br_startblock,imap.br_startoff,
979 				imap.br_blockcount,imap.br_state);
980         	}
981 
982 		if ((numblks_fsb = imap.br_blockcount) == 0) {
983 			/*
984 			 * The numblks_fsb value should always get
985 			 * smaller, otherwise the loop is stuck.
986 			 */
987 			ASSERT(imap.br_blockcount);
988 			break;
989 		}
990 		offset_fsb += numblks_fsb;
991 		count_fsb -= numblks_fsb;
992 	} while (count_fsb > 0);
993 
994 	return 0;
995 
996 error_on_bmapi_transaction:
997 	xfs_bmap_cancel(&free_list);
998 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
999 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1000 error0:
1001 	return XFS_ERROR(error);
1002 }
1003