xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision c21b37f6)
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 	xfs_fsize_t	nisize;
139 	int		pbm;
140 	xfs_fsblock_t	start_block;
141 
142 	mp = io->io_mount;
143 	nisize = XFS_SIZE(mp, io);
144 	if (io->io_new_size > nisize)
145 		nisize = io->io_new_size;
146 
147 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
148 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
149 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
150 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
151 		iomapp->iomap_flags = flags;
152 
153 		if (io->io_flags & XFS_IOCORE_RT) {
154 			iomapp->iomap_flags |= IOMAP_REALTIME;
155 			iomapp->iomap_target = mp->m_rtdev_targp;
156 		} else {
157 			iomapp->iomap_target = mp->m_ddev_targp;
158 		}
159 		start_block = imap->br_startblock;
160 		if (start_block == HOLESTARTBLOCK) {
161 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
162 			iomapp->iomap_flags |= IOMAP_HOLE;
163 		} else if (start_block == DELAYSTARTBLOCK) {
164 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
165 			iomapp->iomap_flags |= IOMAP_DELAY;
166 		} else {
167 			iomapp->iomap_bn = XFS_FSB_TO_DB_IO(io, start_block);
168 			if (ISUNWRITTEN(imap))
169 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
170 		}
171 
172 		if ((iomapp->iomap_offset + iomapp->iomap_bsize) >= nisize) {
173 			iomapp->iomap_flags |= IOMAP_EOF;
174 		}
175 
176 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
177 	}
178 	return pbm;	/* Return the number filled */
179 }
180 
181 int
182 xfs_iomap(
183 	xfs_iocore_t	*io,
184 	xfs_off_t	offset,
185 	ssize_t		count,
186 	int		flags,
187 	xfs_iomap_t	*iomapp,
188 	int		*niomaps)
189 {
190 	xfs_mount_t	*mp = io->io_mount;
191 	xfs_fileoff_t	offset_fsb, end_fsb;
192 	int		error = 0;
193 	int		lockmode = 0;
194 	xfs_bmbt_irec_t	imap;
195 	int		nimaps = 1;
196 	int		bmapi_flags = 0;
197 	int		iomap_flags = 0;
198 
199 	if (XFS_FORCED_SHUTDOWN(mp))
200 		return XFS_ERROR(EIO);
201 
202 	switch (flags &
203 		(BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE |
204 		 BMAPI_UNWRITTEN | BMAPI_DEVICE)) {
205 	case BMAPI_READ:
206 		xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, io, offset, count);
207 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
208 		bmapi_flags = XFS_BMAPI_ENTIRE;
209 		break;
210 	case BMAPI_WRITE:
211 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, io, offset, count);
212 		lockmode = XFS_ILOCK_EXCL|XFS_EXTSIZE_WR;
213 		if (flags & BMAPI_IGNSTATE)
214 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
215 		XFS_ILOCK(mp, io, lockmode);
216 		break;
217 	case BMAPI_ALLOCATE:
218 		xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, io, offset, count);
219 		lockmode = XFS_ILOCK_SHARED|XFS_EXTSIZE_RD;
220 		bmapi_flags = XFS_BMAPI_ENTIRE;
221 		/* Attempt non-blocking lock */
222 		if (flags & BMAPI_TRYLOCK) {
223 			if (!XFS_ILOCK_NOWAIT(mp, io, lockmode))
224 				return XFS_ERROR(EAGAIN);
225 		} else {
226 			XFS_ILOCK(mp, io, lockmode);
227 		}
228 		break;
229 	case BMAPI_UNWRITTEN:
230 		goto phase2;
231 	case BMAPI_DEVICE:
232 		lockmode = XFS_LCK_MAP_SHARED(mp, io);
233 		iomapp->iomap_target = io->io_flags & XFS_IOCORE_RT ?
234 			mp->m_rtdev_targp : mp->m_ddev_targp;
235 		error = 0;
236 		*niomaps = 1;
237 		goto out;
238 	default:
239 		BUG();
240 	}
241 
242 	ASSERT(offset <= mp->m_maxioffset);
243 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
244 		count = mp->m_maxioffset - offset;
245 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
246 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
247 
248 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
249 			(xfs_filblks_t)(end_fsb - offset_fsb),
250 			bmapi_flags,  NULL, 0, &imap,
251 			&nimaps, NULL, NULL);
252 
253 	if (error)
254 		goto out;
255 
256 phase2:
257 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE|BMAPI_UNWRITTEN)) {
258 	case BMAPI_WRITE:
259 		/* If we found an extent, return it */
260 		if (nimaps &&
261 		    (imap.br_startblock != HOLESTARTBLOCK) &&
262 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
263 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
264 					offset, count, iomapp, &imap, flags);
265 			break;
266 		}
267 
268 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
269 			error = XFS_IOMAP_WRITE_DIRECT(mp, io, offset,
270 					count, flags, &imap, &nimaps, nimaps);
271 		} else {
272 			error = XFS_IOMAP_WRITE_DELAY(mp, io, offset, count,
273 					flags, &imap, &nimaps);
274 		}
275 		if (!error) {
276 			xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, io,
277 					offset, count, iomapp, &imap, flags);
278 		}
279 		iomap_flags = IOMAP_NEW;
280 		break;
281 	case BMAPI_ALLOCATE:
282 		/* If we found an extent, return it */
283 		XFS_IUNLOCK(mp, io, lockmode);
284 		lockmode = 0;
285 
286 		if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
287 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, io,
288 					offset, count, iomapp, &imap, flags);
289 			break;
290 		}
291 
292 		error = XFS_IOMAP_WRITE_ALLOCATE(mp, io, offset, count,
293 						 &imap, &nimaps);
294 		break;
295 	case BMAPI_UNWRITTEN:
296 		lockmode = 0;
297 		error = XFS_IOMAP_WRITE_UNWRITTEN(mp, io, offset, count);
298 		nimaps = 0;
299 		break;
300 	}
301 
302 	if (nimaps) {
303 		*niomaps = xfs_imap_to_bmap(io, offset, &imap,
304 					    iomapp, nimaps, *niomaps, iomap_flags);
305 	} else if (niomaps) {
306 		*niomaps = 0;
307 	}
308 
309 out:
310 	if (lockmode)
311 		XFS_IUNLOCK(mp, io, lockmode);
312 	return XFS_ERROR(error);
313 }
314 
315 STATIC int
316 xfs_iomap_eof_align_last_fsb(
317 	xfs_mount_t	*mp,
318 	xfs_iocore_t	*io,
319 	xfs_fsize_t	isize,
320 	xfs_extlen_t	extsize,
321 	xfs_fileoff_t	*last_fsb)
322 {
323 	xfs_fileoff_t	new_last_fsb = 0;
324 	xfs_extlen_t	align;
325 	int		eof, error;
326 
327 	if (io->io_flags & XFS_IOCORE_RT)
328 		;
329 	/*
330 	 * If mounted with the "-o swalloc" option, roundup the allocation
331 	 * request to a stripe width boundary if the file size is >=
332 	 * stripe width and we are allocating past the allocation eof.
333 	 */
334 	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
335 	        (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
336 		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
337 	/*
338 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
339 	 * if the file size is >= stripe unit size, and we are allocating past
340 	 * the allocation eof.
341 	 */
342 	else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
343 		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
344 
345 	/*
346 	 * Always round up the allocation request to an extent boundary
347 	 * (when file on a real-time subvolume or has di_extsize hint).
348 	 */
349 	if (extsize) {
350 		if (new_last_fsb)
351 			align = roundup_64(new_last_fsb, extsize);
352 		else
353 			align = extsize;
354 		new_last_fsb = roundup_64(*last_fsb, align);
355 	}
356 
357 	if (new_last_fsb) {
358 		error = XFS_BMAP_EOF(mp, io, new_last_fsb, XFS_DATA_FORK, &eof);
359 		if (error)
360 			return error;
361 		if (eof)
362 			*last_fsb = new_last_fsb;
363 	}
364 	return 0;
365 }
366 
367 STATIC int
368 xfs_flush_space(
369 	xfs_inode_t	*ip,
370 	int		*fsynced,
371 	int		*ioflags)
372 {
373 	switch (*fsynced) {
374 	case 0:
375 		if (ip->i_delayed_blks) {
376 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
377 			xfs_flush_inode(ip);
378 			xfs_ilock(ip, XFS_ILOCK_EXCL);
379 			*fsynced = 1;
380 		} else {
381 			*ioflags |= BMAPI_SYNC;
382 			*fsynced = 2;
383 		}
384 		return 0;
385 	case 1:
386 		*fsynced = 2;
387 		*ioflags |= BMAPI_SYNC;
388 		return 0;
389 	case 2:
390 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
391 		xfs_flush_device(ip);
392 		xfs_ilock(ip, XFS_ILOCK_EXCL);
393 		*fsynced = 3;
394 		return 0;
395 	}
396 	return 1;
397 }
398 
399 STATIC int
400 xfs_cmn_err_fsblock_zero(
401 	xfs_inode_t	*ip,
402 	xfs_bmbt_irec_t	*imap)
403 {
404 	xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
405 			"Access to block zero in inode %llu "
406 			"start_block: %llx start_off: %llx "
407 			"blkcnt: %llx extent-state: %x\n",
408 		(unsigned long long)ip->i_ino,
409 		(unsigned long long)imap->br_startblock,
410 		(unsigned long long)imap->br_startoff,
411 		(unsigned long long)imap->br_blockcount,
412 		imap->br_state);
413 	return EFSCORRUPTED;
414 }
415 
416 int
417 xfs_iomap_write_direct(
418 	xfs_inode_t	*ip,
419 	xfs_off_t	offset,
420 	size_t		count,
421 	int		flags,
422 	xfs_bmbt_irec_t *ret_imap,
423 	int		*nmaps,
424 	int		found)
425 {
426 	xfs_mount_t	*mp = ip->i_mount;
427 	xfs_iocore_t	*io = &ip->i_iocore;
428 	xfs_fileoff_t	offset_fsb;
429 	xfs_fileoff_t	last_fsb;
430 	xfs_filblks_t	count_fsb, resaligned;
431 	xfs_fsblock_t	firstfsb;
432 	xfs_extlen_t	extsz, temp;
433 	xfs_fsize_t	isize;
434 	int		nimaps;
435 	int		bmapi_flag;
436 	int		quota_flag;
437 	int		rt;
438 	xfs_trans_t	*tp;
439 	xfs_bmbt_irec_t imap;
440 	xfs_bmap_free_t free_list;
441 	uint		qblocks, resblks, resrtextents;
442 	int		committed;
443 	int		error;
444 
445 	/*
446 	 * Make sure that the dquots are there. This doesn't hold
447 	 * the ilock across a disk read.
448 	 */
449 	error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
450 	if (error)
451 		return XFS_ERROR(error);
452 
453 	rt = XFS_IS_REALTIME_INODE(ip);
454 	extsz = xfs_get_extsz_hint(ip);
455 
456 	isize = ip->i_size;
457 	if (io->io_new_size > isize)
458 		isize = io->io_new_size;
459 
460 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
461 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
462 	if ((offset + count) > isize) {
463 		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
464 							&last_fsb);
465 		if (error)
466 			goto error_out;
467 	} else {
468 		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
469 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
470 					ret_imap->br_blockcount +
471 					ret_imap->br_startoff);
472 	}
473 	count_fsb = last_fsb - offset_fsb;
474 	ASSERT(count_fsb > 0);
475 
476 	resaligned = count_fsb;
477 	if (unlikely(extsz)) {
478 		if ((temp = do_mod(offset_fsb, extsz)))
479 			resaligned += temp;
480 		if ((temp = do_mod(resaligned, extsz)))
481 			resaligned += extsz - temp;
482 	}
483 
484 	if (unlikely(rt)) {
485 		resrtextents = qblocks = resaligned;
486 		resrtextents /= mp->m_sb.sb_rextsize;
487 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
488 		quota_flag = XFS_QMOPT_RES_RTBLKS;
489 	} else {
490 		resrtextents = 0;
491 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
492 		quota_flag = XFS_QMOPT_RES_REGBLKS;
493 	}
494 
495 	/*
496 	 * Allocate and setup the transaction
497 	 */
498 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
499 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
500 	error = xfs_trans_reserve(tp, resblks,
501 			XFS_WRITE_LOG_RES(mp), resrtextents,
502 			XFS_TRANS_PERM_LOG_RES,
503 			XFS_WRITE_LOG_COUNT);
504 	/*
505 	 * Check for running out of space, note: need lock to return
506 	 */
507 	if (error)
508 		xfs_trans_cancel(tp, 0);
509 	xfs_ilock(ip, XFS_ILOCK_EXCL);
510 	if (error)
511 		goto error_out;
512 
513 	error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
514 					      qblocks, 0, quota_flag);
515 	if (error)
516 		goto error1;
517 
518 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
519 	xfs_trans_ihold(tp, ip);
520 
521 	bmapi_flag = XFS_BMAPI_WRITE;
522 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
523 		bmapi_flag |= XFS_BMAPI_PREALLOC;
524 
525 	/*
526 	 * Issue the xfs_bmapi() call to allocate the blocks
527 	 */
528 	XFS_BMAP_INIT(&free_list, &firstfsb);
529 	nimaps = 1;
530 	error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb, bmapi_flag,
531 		&firstfsb, 0, &imap, &nimaps, &free_list, NULL);
532 	if (error)
533 		goto error0;
534 
535 	/*
536 	 * Complete the transaction
537 	 */
538 	error = xfs_bmap_finish(&tp, &free_list, &committed);
539 	if (error)
540 		goto error0;
541 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
542 	if (error)
543 		goto error_out;
544 
545 	/*
546 	 * Copy any maps to caller's array and return any error.
547 	 */
548 	if (nimaps == 0) {
549 		error = ENOSPC;
550 		goto error_out;
551 	}
552 
553 	if (unlikely(!imap.br_startblock && !(io->io_flags & XFS_IOCORE_RT))) {
554 		error = xfs_cmn_err_fsblock_zero(ip, &imap);
555 		goto error_out;
556 	}
557 
558 	*ret_imap = imap;
559 	*nmaps = 1;
560 	return 0;
561 
562 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
563 	xfs_bmap_cancel(&free_list);
564 	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
565 
566 error1:	/* Just cancel transaction */
567 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
568 	*nmaps = 0;	/* nothing set-up here */
569 
570 error_out:
571 	return XFS_ERROR(error);
572 }
573 
574 /*
575  * If the caller is doing a write at the end of the file,
576  * then extend the allocation out to the file system's write
577  * iosize.  We clean up any extra space left over when the
578  * file is closed in xfs_inactive().
579  *
580  * For sync writes, we are flushing delayed allocate space to
581  * try to make additional space available for allocation near
582  * the filesystem full boundary - preallocation hurts in that
583  * situation, of course.
584  */
585 STATIC int
586 xfs_iomap_eof_want_preallocate(
587 	xfs_mount_t	*mp,
588 	xfs_iocore_t	*io,
589 	xfs_fsize_t	isize,
590 	xfs_off_t	offset,
591 	size_t		count,
592 	int		ioflag,
593 	xfs_bmbt_irec_t *imap,
594 	int		nimaps,
595 	int		*prealloc)
596 {
597 	xfs_fileoff_t   start_fsb;
598 	xfs_filblks_t   count_fsb;
599 	xfs_fsblock_t	firstblock;
600 	int		n, error, imaps;
601 
602 	*prealloc = 0;
603 	if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
604 		return 0;
605 
606 	/*
607 	 * If there are any real blocks past eof, then don't
608 	 * do any speculative allocation.
609 	 */
610 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
611 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
612 	while (count_fsb > 0) {
613 		imaps = nimaps;
614 		firstblock = NULLFSBLOCK;
615 		error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb, 0,
616 				  &firstblock, 0, imap, &imaps, NULL, NULL);
617 		if (error)
618 			return error;
619 		for (n = 0; n < imaps; n++) {
620 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
621 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
622 				return 0;
623 			start_fsb += imap[n].br_blockcount;
624 			count_fsb -= imap[n].br_blockcount;
625 		}
626 	}
627 	*prealloc = 1;
628 	return 0;
629 }
630 
631 int
632 xfs_iomap_write_delay(
633 	xfs_inode_t	*ip,
634 	xfs_off_t	offset,
635 	size_t		count,
636 	int		ioflag,
637 	xfs_bmbt_irec_t *ret_imap,
638 	int		*nmaps)
639 {
640 	xfs_mount_t	*mp = ip->i_mount;
641 	xfs_iocore_t	*io = &ip->i_iocore;
642 	xfs_fileoff_t	offset_fsb;
643 	xfs_fileoff_t	last_fsb;
644 	xfs_off_t	aligned_offset;
645 	xfs_fileoff_t	ioalign;
646 	xfs_fsblock_t	firstblock;
647 	xfs_extlen_t	extsz;
648 	xfs_fsize_t	isize;
649 	int		nimaps;
650 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
651 	int		prealloc, fsynced = 0;
652 	int		error;
653 
654 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
655 
656 	/*
657 	 * Make sure that the dquots are there. This doesn't hold
658 	 * the ilock across a disk read.
659 	 */
660 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
661 	if (error)
662 		return XFS_ERROR(error);
663 
664 	extsz = xfs_get_extsz_hint(ip);
665 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
666 
667 retry:
668 	isize = ip->i_size;
669 	if (io->io_new_size > isize)
670 		isize = io->io_new_size;
671 
672 	error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
673 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
674 	if (error)
675 		return error;
676 
677 	if (prealloc) {
678 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
679 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
680 		last_fsb = ioalign + mp->m_writeio_blocks;
681 	} else {
682 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
683 	}
684 
685 	if (prealloc || extsz) {
686 		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
687 							&last_fsb);
688 		if (error)
689 			return error;
690 	}
691 
692 	nimaps = XFS_WRITE_IMAPS;
693 	firstblock = NULLFSBLOCK;
694 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
695 			  (xfs_filblks_t)(last_fsb - offset_fsb),
696 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
697 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
698 			  &nimaps, NULL, NULL);
699 	if (error && (error != ENOSPC))
700 		return XFS_ERROR(error);
701 
702 	/*
703 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
704 	 * then we must have run out of space - flush delalloc, and retry..
705 	 */
706 	if (nimaps == 0) {
707 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
708 					io, offset, count);
709 		if (xfs_flush_space(ip, &fsynced, &ioflag))
710 			return XFS_ERROR(ENOSPC);
711 
712 		error = 0;
713 		goto retry;
714 	}
715 
716 	if (unlikely(!imap[0].br_startblock && !(io->io_flags & XFS_IOCORE_RT)))
717 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
718 
719 	*ret_imap = imap[0];
720 	*nmaps = 1;
721 
722 	return 0;
723 }
724 
725 /*
726  * Pass in a delayed allocate extent, convert it to real extents;
727  * return to the caller the extent we create which maps on top of
728  * the originating callers request.
729  *
730  * Called without a lock on the inode.
731  */
732 int
733 xfs_iomap_write_allocate(
734 	xfs_inode_t	*ip,
735 	xfs_off_t	offset,
736 	size_t		count,
737 	xfs_bmbt_irec_t *map,
738 	int		*retmap)
739 {
740 	xfs_mount_t	*mp = ip->i_mount;
741 	xfs_iocore_t    *io = &ip->i_iocore;
742 	xfs_fileoff_t	offset_fsb, last_block;
743 	xfs_fileoff_t	end_fsb, map_start_fsb;
744 	xfs_fsblock_t	first_block;
745 	xfs_bmap_free_t	free_list;
746 	xfs_filblks_t	count_fsb;
747 	xfs_bmbt_irec_t	imap[XFS_STRAT_WRITE_IMAPS];
748 	xfs_trans_t	*tp;
749 	int		i, nimaps, committed;
750 	int		error = 0;
751 	int		nres;
752 
753 	*retmap = 0;
754 
755 	/*
756 	 * Make sure that the dquots are there.
757 	 */
758 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
759 		return XFS_ERROR(error);
760 
761 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
762 	count_fsb = map->br_blockcount;
763 	map_start_fsb = map->br_startoff;
764 
765 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
766 
767 	while (count_fsb != 0) {
768 		/*
769 		 * Set up a transaction with which to allocate the
770 		 * backing store for the file.  Do allocations in a
771 		 * loop until we get some space in the range we are
772 		 * interested in.  The other space that might be allocated
773 		 * is in the delayed allocation extent on which we sit
774 		 * but before our buffer starts.
775 		 */
776 
777 		nimaps = 0;
778 		while (nimaps == 0) {
779 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
780 			tp->t_flags |= XFS_TRANS_RESERVE;
781 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
782 			error = xfs_trans_reserve(tp, nres,
783 					XFS_WRITE_LOG_RES(mp),
784 					0, XFS_TRANS_PERM_LOG_RES,
785 					XFS_WRITE_LOG_COUNT);
786 			if (error) {
787 				xfs_trans_cancel(tp, 0);
788 				return XFS_ERROR(error);
789 			}
790 			xfs_ilock(ip, XFS_ILOCK_EXCL);
791 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
792 			xfs_trans_ihold(tp, ip);
793 
794 			XFS_BMAP_INIT(&free_list, &first_block);
795 
796 			nimaps = XFS_STRAT_WRITE_IMAPS;
797 			/*
798 			 * Ensure we don't go beyond eof - it is possible
799 			 * the extents changed since we did the read call,
800 			 * we dropped the ilock in the interim.
801 			 */
802 
803 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
804 			xfs_bmap_last_offset(NULL, ip, &last_block,
805 				XFS_DATA_FORK);
806 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
807 			if ((map_start_fsb + count_fsb) > last_block) {
808 				count_fsb = last_block - map_start_fsb;
809 				if (count_fsb == 0) {
810 					error = EAGAIN;
811 					goto trans_cancel;
812 				}
813 			}
814 
815 			/* Go get the actual blocks */
816 			error = XFS_BMAPI(mp, tp, io, map_start_fsb, count_fsb,
817 					XFS_BMAPI_WRITE, &first_block, 1,
818 					imap, &nimaps, &free_list, NULL);
819 			if (error)
820 				goto trans_cancel;
821 
822 			error = xfs_bmap_finish(&tp, &free_list, &committed);
823 			if (error)
824 				goto trans_cancel;
825 
826 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
827 			if (error)
828 				goto error0;
829 
830 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
831 		}
832 
833 		/*
834 		 * See if we were able to allocate an extent that
835 		 * covers at least part of the callers request
836 		 */
837 		for (i = 0; i < nimaps; i++) {
838 			if (unlikely(!imap[i].br_startblock &&
839 				     !(io->io_flags & XFS_IOCORE_RT)))
840 				return xfs_cmn_err_fsblock_zero(ip, &imap[i]);
841 			if ((offset_fsb >= imap[i].br_startoff) &&
842 			    (offset_fsb < (imap[i].br_startoff +
843 					   imap[i].br_blockcount))) {
844 				*map = imap[i];
845 				*retmap = 1;
846 				XFS_STATS_INC(xs_xstrat_quick);
847 				return 0;
848 			}
849 			count_fsb -= imap[i].br_blockcount;
850 		}
851 
852 		/* So far we have not mapped the requested part of the
853 		 * file, just surrounding data, try again.
854 		 */
855 		nimaps--;
856 		map_start_fsb = imap[nimaps].br_startoff +
857 				imap[nimaps].br_blockcount;
858 	}
859 
860 trans_cancel:
861 	xfs_bmap_cancel(&free_list);
862 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
863 error0:
864 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
865 	return XFS_ERROR(error);
866 }
867 
868 int
869 xfs_iomap_write_unwritten(
870 	xfs_inode_t	*ip,
871 	xfs_off_t	offset,
872 	size_t		count)
873 {
874 	xfs_mount_t	*mp = ip->i_mount;
875 	xfs_iocore_t    *io = &ip->i_iocore;
876 	xfs_fileoff_t	offset_fsb;
877 	xfs_filblks_t	count_fsb;
878 	xfs_filblks_t	numblks_fsb;
879 	xfs_fsblock_t	firstfsb;
880 	int		nimaps;
881 	xfs_trans_t	*tp;
882 	xfs_bmbt_irec_t imap;
883 	xfs_bmap_free_t free_list;
884 	uint		resblks;
885 	int		committed;
886 	int		error;
887 
888 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
889 				&ip->i_iocore, offset, count);
890 
891 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
892 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
893 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
894 
895 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
896 
897 	do {
898 		/*
899 		 * set up a transaction to convert the range of extents
900 		 * from unwritten to real. Do allocations in a loop until
901 		 * we have covered the range passed in.
902 		 */
903 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
904 		tp->t_flags |= XFS_TRANS_RESERVE;
905 		error = xfs_trans_reserve(tp, resblks,
906 				XFS_WRITE_LOG_RES(mp), 0,
907 				XFS_TRANS_PERM_LOG_RES,
908 				XFS_WRITE_LOG_COUNT);
909 		if (error) {
910 			xfs_trans_cancel(tp, 0);
911 			return XFS_ERROR(error);
912 		}
913 
914 		xfs_ilock(ip, XFS_ILOCK_EXCL);
915 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
916 		xfs_trans_ihold(tp, ip);
917 
918 		/*
919 		 * Modify the unwritten extent state of the buffer.
920 		 */
921 		XFS_BMAP_INIT(&free_list, &firstfsb);
922 		nimaps = 1;
923 		error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb,
924 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
925 				  1, &imap, &nimaps, &free_list, NULL);
926 		if (error)
927 			goto error_on_bmapi_transaction;
928 
929 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
930 		if (error)
931 			goto error_on_bmapi_transaction;
932 
933 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
934 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
935 		if (error)
936 			return XFS_ERROR(error);
937 
938 		if (unlikely(!imap.br_startblock &&
939 			     !(io->io_flags & XFS_IOCORE_RT)))
940 			return xfs_cmn_err_fsblock_zero(ip, &imap);
941 
942 		if ((numblks_fsb = imap.br_blockcount) == 0) {
943 			/*
944 			 * The numblks_fsb value should always get
945 			 * smaller, otherwise the loop is stuck.
946 			 */
947 			ASSERT(imap.br_blockcount);
948 			break;
949 		}
950 		offset_fsb += numblks_fsb;
951 		count_fsb -= numblks_fsb;
952 	} while (count_fsb > 0);
953 
954 	return 0;
955 
956 error_on_bmapi_transaction:
957 	xfs_bmap_cancel(&free_list);
958 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
959 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
960 	return XFS_ERROR(error);
961 }
962