xref: /linux/fs/xfs/xfs_iomap.c (revision 858259cf7d1c443c836a2022b78cb281f0a9b95e)
1 /*
2  * Copyright (c) 2000-2005 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_dir.h"
27 #include "xfs_dir2.h"
28 #include "xfs_alloc.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_quota.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_btree.h"
42 #include "xfs_bmap.h"
43 #include "xfs_bit.h"
44 #include "xfs_rtalloc.h"
45 #include "xfs_error.h"
46 #include "xfs_itable.h"
47 #include "xfs_rw.h"
48 #include "xfs_acl.h"
49 #include "xfs_cap.h"
50 #include "xfs_mac.h"
51 #include "xfs_attr.h"
52 #include "xfs_buf_item.h"
53 #include "xfs_trans_space.h"
54 #include "xfs_utils.h"
55 #include "xfs_iomap.h"
56 
57 #if defined(XFS_RW_TRACE)
58 void
59 xfs_iomap_enter_trace(
60 	int		tag,
61 	xfs_iocore_t	*io,
62 	xfs_off_t	offset,
63 	ssize_t		count)
64 {
65 	xfs_inode_t	*ip = XFS_IO_INODE(io);
66 
67 	if (!ip->i_rwtrace)
68 		return;
69 
70 	ktrace_enter(ip->i_rwtrace,
71 		(void *)((unsigned long)tag),
72 		(void *)ip,
73 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
74 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
75 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
76 		(void *)((unsigned long)(offset & 0xffffffff)),
77 		(void *)((unsigned long)count),
78 		(void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
79 		(void *)((unsigned long)(io->io_new_size & 0xffffffff)),
80 		(void *)NULL,
81 		(void *)NULL,
82 		(void *)NULL,
83 		(void *)NULL,
84 		(void *)NULL,
85 		(void *)NULL,
86 		(void *)NULL);
87 }
88 
89 void
90 xfs_iomap_map_trace(
91 	int		tag,
92 	xfs_iocore_t	*io,
93 	xfs_off_t	offset,
94 	ssize_t		count,
95 	xfs_iomap_t	*iomapp,
96 	xfs_bmbt_irec_t	*imapp,
97 	int		flags)
98 {
99 	xfs_inode_t	*ip = XFS_IO_INODE(io);
100 
101 	if (!ip->i_rwtrace)
102 		return;
103 
104 	ktrace_enter(ip->i_rwtrace,
105 		(void *)((unsigned long)tag),
106 		(void *)ip,
107 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
108 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
109 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
110 		(void *)((unsigned long)(offset & 0xffffffff)),
111 		(void *)((unsigned long)count),
112 		(void *)((unsigned long)flags),
113 		(void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
114 		(void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
115 		(void *)((unsigned long)(iomapp->iomap_delta)),
116 		(void *)((unsigned long)(iomapp->iomap_bsize)),
117 		(void *)((unsigned long)(iomapp->iomap_bn)),
118 		(void *)(__psint_t)(imapp->br_startoff),
119 		(void *)((unsigned long)(imapp->br_blockcount)),
120 		(void *)(__psint_t)(imapp->br_startblock));
121 }
122 #else
123 #define xfs_iomap_enter_trace(tag, io, offset, count)
124 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
125 #endif
126 
127 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
128 						<< mp->m_writeio_log)
129 #define XFS_STRAT_WRITE_IMAPS	2
130 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
131 
132 STATIC int
133 xfs_imap_to_bmap(
134 	xfs_iocore_t	*io,
135 	xfs_off_t	offset,
136 	xfs_bmbt_irec_t *imap,
137 	xfs_iomap_t	*iomapp,
138 	int		imaps,			/* Number of imap entries */
139 	int		iomaps,			/* Number of iomap entries */
140 	int		flags)
141 {
142 	xfs_mount_t	*mp;
143 	xfs_fsize_t	nisize;
144 	int		pbm;
145 	xfs_fsblock_t	start_block;
146 
147 	mp = io->io_mount;
148 	nisize = XFS_SIZE(mp, io);
149 	if (io->io_new_size > nisize)
150 		nisize = io->io_new_size;
151 
152 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
153 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
154 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
155 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
156 		iomapp->iomap_flags = flags;
157 
158 		if (io->io_flags & XFS_IOCORE_RT) {
159 			iomapp->iomap_flags |= IOMAP_REALTIME;
160 			iomapp->iomap_target = mp->m_rtdev_targp;
161 		} else {
162 			iomapp->iomap_target = mp->m_ddev_targp;
163 		}
164 		start_block = imap->br_startblock;
165 		if (start_block == HOLESTARTBLOCK) {
166 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
167 			iomapp->iomap_flags |= IOMAP_HOLE;
168 		} else if (start_block == DELAYSTARTBLOCK) {
169 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
170 			iomapp->iomap_flags |= IOMAP_DELAY;
171 		} else {
172 			iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
173 			if (ISUNWRITTEN(imap))
174 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
175 		}
176 
177 		if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
178 			iomapp->iomap_flags |= IOMAP_EOF;
179 		}
180 
181 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
182 	}
183 	return pbm;	/* Return the number filled */
184 }
185 
186 int
187 xfs_iomap(
188 	xfs_iocore_t	*io,
189 	xfs_off_t	offset,
190 	ssize_t		count,
191 	int		flags,
192 	xfs_iomap_t	*iomapp,
193 	int		*niomaps)
194 {
195 	xfs_mount_t	*mp = io->io_mount;
196 	xfs_fileoff_t	offset_fsb, end_fsb;
197 	int		error = 0;
198 	int		lockmode = 0;
199 	xfs_bmbt_irec_t	imap;
200 	int		nimaps = 1;
201 	int		bmapi_flags = 0;
202 	int		iomap_flags = 0;
203 
204 	if (XFS_FORCED_SHUTDOWN(mp))
205 		return XFS_ERROR(EIO);
206 
207 	switch (flags &
208 		(BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
209 		 BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
210 	case BMAPI_READ:
211 		xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
212 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
213 		bmapi_flags = XFS_BMAPI_ENTIRE;
214 		break;
215 	case BMAPI_WRITE:
216 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
217 		lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
218 		if (flags & BMAPI_IGNSTATE)
219 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
220 		XFS_ILOCK(mp, io, lockmode);
221 		break;
222 	case BMAPI_ALLOCATE:
223 		xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
224 		lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
225 		bmapi_flags = XFS_BMAPI_ENTIRE;
226 		/* Attempt non-blocking lock */
227 		if (flags & BMAPI_TRYLOCK) {
228 			if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
229 				return XFS_ERROR(EAGAIN);
230 		} else {
231 			XFS_ILOCK(mp, io, lockmode);
232 		}
233 		break;
234 	case BMAPI_UNWRITTEN:
235 		goto phase2;
236 	case BMAPI_DEVICE:
237 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
238 		iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
239 			mp->m_rtdev_targp : mp->m_ddev_targp;
240 		error = 0;
241 		*niomaps = 1;
242 		goto out;
243 	default:
244 		BUG();
245 	}
246 
247 	ASSERT(offset <= mp->m_maxioffset);
248 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
249 		count = mp->m_maxioffset - offset;
250 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
251 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
252 
253 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
254 			(xfs_filblks_t)(end_fsb - offset_fsb),
255 			bmapi_flags,  NULL, 0, &imap,
256 			&nimaps, NULL);
257 
258 	if (error)
259 		goto out;
260 
261 phase2:
262 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
263 	case BMAPI_WRITE:
264 		/* If we found an extent, return it */
265 		if (nimaps &&
266 		    (imap.br_startblock != HOLESTARTBLOCK) &&
267 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
268 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
269 					offset, count, iomapp, &imap, flags);
270 			break;
271 		}
272 
273 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
274 			error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
275 					count, flags, &imap, &nimaps, nimaps);
276 		} else {
277 			error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
278 					flags, &imap, &nimaps);
279 		}
280 		if (!error) {
281 			xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
282 					offset, count, iomapp, &imap, flags);
283 		}
284 		iomap_flags = IOMAP_NEW;
285 		break;
286 	case BMAPI_ALLOCATE:
287 		/* If we found an extent, return it */
288 		XFS_IUNLOCK(mp, io, lockmode);
289 		lockmode = 0;
290 
291 		if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
292 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
293 					offset, count, iomapp, &imap, flags);
294 			break;
295 		}
296 
297 		error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
298 						 &imap, &nimaps);
299 		break;
300 	case BMAPI_UNWRITTEN:
301 		lockmode = 0;
302 		error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
303 		nimaps = 0;
304 		break;
305 	}
306 
307 	if (nimaps) {
308 		*niomaps = xfs_imap_to_bmap(io, offset, &imap,
309 					    iomapp, nimaps, *niomaps, iomap_flags);
310 	} else if (niomaps) {
311 		*niomaps = 0;
312 	}
313 
314 out:
315 	if (lockmode)
316 		XFS_IUNLOCK(mp, io, lockmode);
317 	return XFS_ERROR(error);
318 }
319 
320 STATIC int
321 xfs_flush_space(
322 	xfs_inode_t	*ip,
323 	int		*fsynced,
324 	int		*ioflags)
325 {
326 	switch (*fsynced) {
327 	case 0:
328 		if (ip->i_delayed_blks) {
329 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
330 			xfs_flush_inode(ip);
331 			xfs_ilock(ip, XFS_ILOCK_EXCL);
332 			*fsynced = 1;
333 		} else {
334 			*ioflags |= BMAPI_SYNC;
335 			*fsynced = 2;
336 		}
337 		return 0;
338 	case 1:
339 		*fsynced = 2;
340 		*ioflags |= BMAPI_SYNC;
341 		return 0;
342 	case 2:
343 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
344 		xfs_flush_device(ip);
345 		xfs_ilock(ip, XFS_ILOCK_EXCL);
346 		*fsynced = 3;
347 		return 0;
348 	}
349 	return 1;
350 }
351 
352 int
353 xfs_iomap_write_direct(
354 	xfs_inode_t	*ip,
355 	xfs_off_t	offset,
356 	size_t		count,
357 	int		flags,
358 	xfs_bmbt_irec_t *ret_imap,
359 	int		*nmaps,
360 	int		found)
361 {
362 	xfs_mount_t	*mp = ip->i_mount;
363 	xfs_iocore_t	*io = &ip->i_iocore;
364 	xfs_fileoff_t	offset_fsb;
365 	xfs_fileoff_t	last_fsb;
366 	xfs_filblks_t	count_fsb;
367 	xfs_fsblock_t	firstfsb;
368 	int		nimaps;
369 	int		error;
370 	int		bmapi_flag;
371 	int		quota_flag;
372 	int		rt;
373 	xfs_trans_t	*tp;
374 	xfs_bmbt_irec_t imap;
375 	xfs_bmap_free_t free_list;
376 	xfs_filblks_t	qblocks, resblks;
377 	int		committed;
378 	int		resrtextents;
379 
380 	/*
381 	 * Make sure that the dquots are there. This doesn't hold
382 	 * the ilock across a disk read.
383 	 */
384 	error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
385 	if (error)
386 		return XFS_ERROR(error);
387 
388 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
389 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
390 	count_fsb = last_fsb - offset_fsb;
391 	if (found && (ret_imap->br_startblock == HOLESTARTBLOCK)) {
392 		xfs_fileoff_t	map_last_fsb;
393 
394 		map_last_fsb = ret_imap->br_blockcount + ret_imap->br_startoff;
395 		if (map_last_fsb < last_fsb) {
396 			last_fsb = map_last_fsb;
397 			count_fsb = last_fsb - offset_fsb;
398 		}
399 		ASSERT(count_fsb > 0);
400 	}
401 
402 	/*
403 	 * Determine if reserving space on the data or realtime partition.
404 	 */
405 	if ((rt = XFS_IS_REALTIME_INODE(ip))) {
406 		xfs_extlen_t	extsz;
407 
408 		if (!(extsz = ip->i_d.di_extsize))
409 			extsz = mp->m_sb.sb_rextsize;
410 		resrtextents = qblocks = (count_fsb + extsz - 1);
411 		do_div(resrtextents, mp->m_sb.sb_rextsize);
412 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
413 		quota_flag = XFS_QMOPT_RES_RTBLKS;
414 	} else {
415 		resrtextents = 0;
416 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, count_fsb);
417 		quota_flag = XFS_QMOPT_RES_REGBLKS;
418 	}
419 
420 	/*
421 	 * Allocate and setup the transaction
422 	 */
423 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
424 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
425 	error = xfs_trans_reserve(tp, resblks,
426 			XFS_WRITE_LOG_RES(mp), resrtextents,
427 			XFS_TRANS_PERM_LOG_RES,
428 			XFS_WRITE_LOG_COUNT);
429 
430 	/*
431 	 * Check for running out of space, note: need lock to return
432 	 */
433 	if (error)
434 		xfs_trans_cancel(tp, 0);
435 	xfs_ilock(ip, XFS_ILOCK_EXCL);
436 	if (error)
437 		goto error_out;
438 
439 	if (XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag)) {
440 		error = (EDQUOT);
441 		goto error1;
442 	}
443 
444 	bmapi_flag = XFS_BMAPI_WRITE;
445 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
446 	xfs_trans_ihold(tp, ip);
447 
448 	if (!(flags & BMAPI_MMAP) && (offset < ip->i_d.di_size || rt))
449 		bmapi_flag |= XFS_BMAPI_PREALLOC;
450 
451 	/*
452 	 * Issue the bmapi() call to allocate the blocks
453 	 */
454 	XFS_BMAP_INIT(&free_list, &firstfsb);
455 	nimaps = 1;
456 	error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
457 		bmapi_flag, &firstfsb, 0, &imap, &nimaps, &free_list);
458 	if (error)
459 		goto error0;
460 
461 	/*
462 	 * Complete the transaction
463 	 */
464 	error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
465 	if (error)
466 		goto error0;
467 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
468 	if (error)
469 		goto error_out;
470 
471 	/*
472 	 * Copy any maps to caller's array and return any error.
473 	 */
474 	if (nimaps == 0) {
475 		error = (ENOSPC);
476 		goto error_out;
477 	}
478 
479 	*ret_imap = imap;
480 	*nmaps = 1;
481 	if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
482                 cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
483                         "start_block : %llx start_off : %llx blkcnt : %llx "
484                         "extent-state : %x \n",
485                         (ip->i_mount)->m_fsname,
486                         (long long)ip->i_ino,
487                         ret_imap->br_startblock, ret_imap->br_startoff,
488                         ret_imap->br_blockcount,ret_imap->br_state);
489         }
490 	return 0;
491 
492 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
493 	xfs_bmap_cancel(&free_list);
494 	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
495 
496 error1:	/* Just cancel transaction */
497 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
498 	*nmaps = 0;	/* nothing set-up here */
499 
500 error_out:
501 	return XFS_ERROR(error);
502 }
503 
504 int
505 xfs_iomap_write_delay(
506 	xfs_inode_t	*ip,
507 	xfs_off_t	offset,
508 	size_t		count,
509 	int		ioflag,
510 	xfs_bmbt_irec_t *ret_imap,
511 	int		*nmaps)
512 {
513 	xfs_mount_t	*mp = ip->i_mount;
514 	xfs_iocore_t	*io = &ip->i_iocore;
515 	xfs_fileoff_t	offset_fsb;
516 	xfs_fileoff_t	last_fsb;
517 	xfs_fsize_t	isize;
518 	xfs_fsblock_t	firstblock;
519 	int		nimaps;
520 	int		error;
521 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
522 	int		aeof;
523 	int		fsynced = 0;
524 
525 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
526 
527 	/*
528 	 * Make sure that the dquots are there. This doesn't hold
529 	 * the ilock across a disk read.
530 	 */
531 
532 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
533 	if (error)
534 		return XFS_ERROR(error);
535 
536 retry:
537 	isize = ip->i_d.di_size;
538 	if (io->io_new_size > isize) {
539 		isize = io->io_new_size;
540 	}
541 
542 	aeof = 0;
543 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
544 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
545 	/*
546 	 * If the caller is doing a write at the end of the file,
547 	 * then extend the allocation (and the buffer used for the write)
548 	 * out to the file system's write iosize.  We clean up any extra
549 	 * space left over when the file is closed in xfs_inactive().
550 	 *
551 	 * For sync writes, we are flushing delayed allocate space to
552 	 * try to make additional space available for allocation near
553 	 * the filesystem full boundary - preallocation hurts in that
554 	 * situation, of course.
555 	 */
556 	if (!(ioflag & BMAPI_SYNC) && ((offset + count) > ip->i_d.di_size)) {
557 		xfs_off_t	aligned_offset;
558 		xfs_filblks_t   count_fsb;
559 		unsigned int	iosize;
560 		xfs_fileoff_t	ioalign;
561 		int		n;
562 		xfs_fileoff_t   start_fsb;
563 
564 		/*
565 		 * If there are any real blocks past eof, then don't
566 		 * do any speculative allocation.
567 		 */
568 		start_fsb = XFS_B_TO_FSBT(mp,
569 					((xfs_ufsize_t)(offset + count - 1)));
570 		count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
571 		while (count_fsb > 0) {
572 			nimaps = XFS_WRITE_IMAPS;
573 			error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb,
574 					0, &firstblock, 0, imap, &nimaps, NULL);
575 			if (error) {
576 				return error;
577 			}
578 			for (n = 0; n < nimaps; n++) {
579 				if ( !(io->io_flags & XFS_IOCORE_RT)  &&
580 					!imap[n].br_startblock) {
581 					cmn_err(CE_PANIC,"Access to block "
582 						"zero:  fs <%s> inode: %lld "
583 						"start_block : %llx start_off "
584 						": %llx blkcnt : %llx "
585 						"extent-state : %x \n",
586 						(ip->i_mount)->m_fsname,
587 						(long long)ip->i_ino,
588 						imap[n].br_startblock,
589 						imap[n].br_startoff,
590 						imap[n].br_blockcount,
591 						imap[n].br_state);
592         			}
593 				if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
594 				    (imap[n].br_startblock != DELAYSTARTBLOCK)) {
595 					goto write_map;
596 				}
597 				start_fsb += imap[n].br_blockcount;
598 				count_fsb -= imap[n].br_blockcount;
599 			}
600 		}
601 		iosize = mp->m_writeio_blocks;
602 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
603 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
604 		last_fsb = ioalign + iosize;
605 		aeof = 1;
606 	}
607 write_map:
608 	nimaps = XFS_WRITE_IMAPS;
609 	firstblock = NULLFSBLOCK;
610 
611 	/*
612 	 * If mounted with the "-o swalloc" option, roundup the allocation
613 	 * request to a stripe width boundary if the file size is >=
614 	 * stripe width and we are allocating past the allocation eof.
615 	 */
616 	if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_swidth
617 	    && (mp->m_flags & XFS_MOUNT_SWALLOC)
618 	    && (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)) && aeof) {
619 		int eof;
620 		xfs_fileoff_t new_last_fsb;
621 
622 		new_last_fsb = roundup_64(last_fsb, mp->m_swidth);
623 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
624 		if (error) {
625 			return error;
626 		}
627 		if (eof) {
628 			last_fsb = new_last_fsb;
629 		}
630 	/*
631 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
632 	 * if the file size is >= stripe unit size, and we are allocating past
633 	 * the allocation eof.
634 	 */
635 	} else if (!(io->io_flags & XFS_IOCORE_RT) && mp->m_dalign &&
636 		   (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)) && aeof) {
637 		int eof;
638 		xfs_fileoff_t new_last_fsb;
639 		new_last_fsb = roundup_64(last_fsb, mp->m_dalign);
640 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
641 		if (error) {
642 			return error;
643 		}
644 		if (eof) {
645 			last_fsb = new_last_fsb;
646 		}
647 	/*
648 	 * Round up the allocation request to a real-time extent boundary
649 	 * if the file is on the real-time subvolume.
650 	 */
651 	} else if (io->io_flags & XFS_IOCORE_RT && aeof) {
652 		int eof;
653 		xfs_fileoff_t new_last_fsb;
654 
655 		new_last_fsb = roundup_64(last_fsb, mp->m_sb.sb_rextsize);
656 		error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
657 		if (error) {
658 			return error;
659 		}
660 		if (eof)
661 			last_fsb = new_last_fsb;
662 	}
663 	error = xfs_bmapi(NULL, ip, offset_fsb,
664 			  (xfs_filblks_t)(last_fsb - offset_fsb),
665 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
666 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
667 			  &nimaps, NULL);
668 	/*
669 	 * This can be EDQUOT, if nimaps == 0
670 	 */
671 	if (error && (error != ENOSPC)) {
672 		return XFS_ERROR(error);
673 	}
674 	/*
675 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
676 	 * then we must have run out of space.
677 	 */
678 	if (nimaps == 0) {
679 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
680 					io, offset, count);
681 		if (xfs_flush_space(ip, &fsynced, &ioflag))
682 			return XFS_ERROR(ENOSPC);
683 
684 		error = 0;
685 		goto retry;
686 	}
687 
688 	*ret_imap = imap[0];
689 	*nmaps = 1;
690 	if ( !(io->io_flags & XFS_IOCORE_RT)  && !ret_imap->br_startblock) {
691 		cmn_err(CE_PANIC,"Access to block zero:  fs <%s> inode: %lld "
692                         "start_block : %llx start_off : %llx blkcnt : %llx "
693                         "extent-state : %x \n",
694                         (ip->i_mount)->m_fsname,
695                         (long long)ip->i_ino,
696                         ret_imap->br_startblock, ret_imap->br_startoff,
697                         ret_imap->br_blockcount,ret_imap->br_state);
698 	}
699 	return 0;
700 }
701 
702 /*
703  * Pass in a delayed allocate extent, convert it to real extents;
704  * return to the caller the extent we create which maps on top of
705  * the originating callers request.
706  *
707  * Called without a lock on the inode.
708  */
709 int
710 xfs_iomap_write_allocate(
711 	xfs_inode_t	*ip,
712 	xfs_off_t	offset,
713 	size_t		count,
714 	xfs_bmbt_irec_t *map,
715 	int		*retmap)
716 {
717 	xfs_mount_t	*mp = ip->i_mount;
718 	xfs_iocore_t    *io = &ip->i_iocore;
719 	xfs_fileoff_t	offset_fsb, last_block;
720 	xfs_fileoff_t	end_fsb, map_start_fsb;
721 	xfs_fsblock_t	first_block;
722 	xfs_bmap_free_t	free_list;
723 	xfs_filblks_t	count_fsb;
724 	xfs_bmbt_irec_t	imap[XFS_STRAT_WRITE_IMAPS];
725 	xfs_trans_t	*tp;
726 	int		i, nimaps, committed;
727 	int		error = 0;
728 	int		nres;
729 
730 	*retmap = 0;
731 
732 	/*
733 	 * Make sure that the dquots are there.
734 	 */
735 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
736 		return XFS_ERROR(error);
737 
738 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
739 	count_fsb = map->br_blockcount;
740 	map_start_fsb = map->br_startoff;
741 
742 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
743 
744 	while (count_fsb != 0) {
745 		/*
746 		 * Set up a transaction with which to allocate the
747 		 * backing store for the file.  Do allocations in a
748 		 * loop until we get some space in the range we are
749 		 * interested in.  The other space that might be allocated
750 		 * is in the delayed allocation extent on which we sit
751 		 * but before our buffer starts.
752 		 */
753 
754 		nimaps = 0;
755 		while (nimaps == 0) {
756 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
757 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
758 			error = xfs_trans_reserve(tp, nres,
759 					XFS_WRITE_LOG_RES(mp),
760 					0, XFS_TRANS_PERM_LOG_RES,
761 					XFS_WRITE_LOG_COUNT);
762 			if (error == ENOSPC) {
763 				error = xfs_trans_reserve(tp, 0,
764 						XFS_WRITE_LOG_RES(mp),
765 						0,
766 						XFS_TRANS_PERM_LOG_RES,
767 						XFS_WRITE_LOG_COUNT);
768 			}
769 			if (error) {
770 				xfs_trans_cancel(tp, 0);
771 				return XFS_ERROR(error);
772 			}
773 			xfs_ilock(ip, XFS_ILOCK_EXCL);
774 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
775 			xfs_trans_ihold(tp, ip);
776 
777 			XFS_BMAP_INIT(&free_list, &first_block);
778 
779 			nimaps = XFS_STRAT_WRITE_IMAPS;
780 			/*
781 			 * Ensure we don't go beyond eof - it is possible
782 			 * the extents changed since we did the read call,
783 			 * we dropped the ilock in the interim.
784 			 */
785 
786 			end_fsb = XFS_B_TO_FSB(mp, ip->i_d.di_size);
787 			xfs_bmap_last_offset(NULL, ip, &last_block,
788 				XFS_DATA_FORK);
789 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
790 			if ((map_start_fsb + count_fsb) > last_block) {
791 				count_fsb = last_block - map_start_fsb;
792 				if (count_fsb == 0) {
793 					error = EAGAIN;
794 					goto trans_cancel;
795 				}
796 			}
797 
798 			/* Go get the actual blocks */
799 			error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
800 					XFS_BMAPI_WRITE, &first_block, 1,
801 					imap, &nimaps, &free_list);
802 			if (error)
803 				goto trans_cancel;
804 
805 			error = xfs_bmap_finish(&tp, &free_list,
806 					first_block, &committed);
807 			if (error)
808 				goto trans_cancel;
809 
810 			error = xfs_trans_commit(tp,
811 					XFS_TRANS_RELEASE_LOG_RES, NULL);
812 			if (error)
813 				goto error0;
814 
815 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
816 		}
817 
818 		/*
819 		 * See if we were able to allocate an extent that
820 		 * covers at least part of the callers request
821 		 */
822 
823 		for (i = 0; i < nimaps; i++) {
824 			if ( !(io->io_flags & XFS_IOCORE_RT)  &&
825 				!imap[i].br_startblock) {
826 				cmn_err(CE_PANIC,"Access to block zero:  "
827 					"fs <%s> inode: %lld "
828 					"start_block : %llx start_off : %llx "
829 					"blkcnt : %llx extent-state : %x \n",
830 					(ip->i_mount)->m_fsname,
831 					(long long)ip->i_ino,
832 					imap[i].br_startblock,
833 					imap[i].br_startoff,
834 				        imap[i].br_blockcount,imap[i].br_state);
835                         }
836 			if ((offset_fsb >= imap[i].br_startoff) &&
837 			    (offset_fsb < (imap[i].br_startoff +
838 					   imap[i].br_blockcount))) {
839 				*map = imap[i];
840 				*retmap = 1;
841 				XFS_STATS_INC(xs_xstrat_quick);
842 				return 0;
843 			}
844 			count_fsb -= imap[i].br_blockcount;
845 		}
846 
847 		/* So far we have not mapped the requested part of the
848 		 * file, just surrounding data, try again.
849 		 */
850 		nimaps--;
851 		map_start_fsb = imap[nimaps].br_startoff +
852 				imap[nimaps].br_blockcount;
853 	}
854 
855 trans_cancel:
856 	xfs_bmap_cancel(&free_list);
857 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
858 error0:
859 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
860 	return XFS_ERROR(error);
861 }
862 
863 int
864 xfs_iomap_write_unwritten(
865 	xfs_inode_t	*ip,
866 	xfs_off_t	offset,
867 	size_t		count)
868 {
869 	xfs_mount_t	*mp = ip->i_mount;
870 	xfs_iocore_t    *io = &ip->i_iocore;
871 	xfs_trans_t	*tp;
872 	xfs_fileoff_t	offset_fsb;
873 	xfs_filblks_t	count_fsb;
874 	xfs_filblks_t	numblks_fsb;
875 	xfs_bmbt_irec_t	imap;
876 	int		committed;
877 	int		error;
878 	int		nres;
879 	int		nimaps;
880 	xfs_fsblock_t	firstfsb;
881 	xfs_bmap_free_t	free_list;
882 
883 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
884 				&ip->i_iocore, offset, count);
885 
886 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
887 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
888 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
889 
890 	do {
891 		nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
892 
893 		/*
894 		 * set up a transaction to convert the range of extents
895 		 * from unwritten to real. Do allocations in a loop until
896 		 * we have covered the range passed in.
897 		 */
898 
899 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
900 		error = xfs_trans_reserve(tp, nres,
901 				XFS_WRITE_LOG_RES(mp), 0,
902 				XFS_TRANS_PERM_LOG_RES,
903 				XFS_WRITE_LOG_COUNT);
904 		if (error) {
905 			xfs_trans_cancel(tp, 0);
906 			goto error0;
907 		}
908 
909 		xfs_ilock(ip, XFS_ILOCK_EXCL);
910 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
911 		xfs_trans_ihold(tp, ip);
912 
913 		/*
914 		 * Modify the unwritten extent state of the buffer.
915 		 */
916 		XFS_BMAP_INIT(&free_list, &firstfsb);
917 		nimaps = 1;
918 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
919 				  XFS_BMAPI_WRITE, &firstfsb,
920 				  1, &imap, &nimaps, &free_list);
921 		if (error)
922 			goto error_on_bmapi_transaction;
923 
924 		error = xfs_bmap_finish(&(tp), &(free_list),
925 				firstfsb, &committed);
926 		if (error)
927 			goto error_on_bmapi_transaction;
928 
929 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
930 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
931 		if (error)
932 			goto error0;
933 
934 		if ( !(io->io_flags & XFS_IOCORE_RT)  && !imap.br_startblock) {
935 			cmn_err(CE_PANIC,"Access to block zero:  fs <%s> "
936 				"inode: %lld start_block : %llx start_off : "
937 				"%llx blkcnt : %llx extent-state : %x \n",
938 				(ip->i_mount)->m_fsname,
939 				(long long)ip->i_ino,
940 				imap.br_startblock,imap.br_startoff,
941 				imap.br_blockcount,imap.br_state);
942         	}
943 
944 		if ((numblks_fsb = imap.br_blockcount) == 0) {
945 			/*
946 			 * The numblks_fsb value should always get
947 			 * smaller, otherwise the loop is stuck.
948 			 */
949 			ASSERT(imap.br_blockcount);
950 			break;
951 		}
952 		offset_fsb += numblks_fsb;
953 		count_fsb -= numblks_fsb;
954 	} while (count_fsb > 0);
955 
956 	return 0;
957 
958 error_on_bmapi_transaction:
959 	xfs_bmap_cancel(&free_list);
960 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
961 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
962 error0:
963 	return XFS_ERROR(error);
964 }
965