xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision 367b8112)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_bit.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_btree.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
44 #include "xfs_rw.h"
45 #include "xfs_acl.h"
46 #include "xfs_attr.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_trans_space.h"
49 #include "xfs_utils.h"
50 #include "xfs_iomap.h"
51 
52 #if defined(XFS_RW_TRACE)
53 void
54 xfs_iomap_enter_trace(
55 	int		tag,
56 	xfs_inode_t	*ip,
57 	xfs_off_t	offset,
58 	ssize_t		count)
59 {
60 	if (!ip->i_rwtrace)
61 		return;
62 
63 	ktrace_enter(ip->i_rwtrace,
64 		(void *)((unsigned long)tag),
65 		(void *)ip,
66 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
67 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
68 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
69 		(void *)((unsigned long)(offset & 0xffffffff)),
70 		(void *)((unsigned long)count),
71 		(void *)((unsigned long)((ip->i_new_size >> 32) & 0xffffffff)),
72 		(void *)((unsigned long)(ip->i_new_size & 0xffffffff)),
73 		(void *)((unsigned long)current_pid()),
74 		(void *)NULL,
75 		(void *)NULL,
76 		(void *)NULL,
77 		(void *)NULL,
78 		(void *)NULL,
79 		(void *)NULL);
80 }
81 
82 void
83 xfs_iomap_map_trace(
84 	int		tag,
85 	xfs_inode_t	*ip,
86 	xfs_off_t	offset,
87 	ssize_t		count,
88 	xfs_iomap_t	*iomapp,
89 	xfs_bmbt_irec_t	*imapp,
90 	int		flags)
91 {
92 	if (!ip->i_rwtrace)
93 		return;
94 
95 	ktrace_enter(ip->i_rwtrace,
96 		(void *)((unsigned long)tag),
97 		(void *)ip,
98 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
99 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
100 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
101 		(void *)((unsigned long)(offset & 0xffffffff)),
102 		(void *)((unsigned long)count),
103 		(void *)((unsigned long)flags),
104 		(void *)((unsigned long)((iomapp->iomap_offset >> 32) & 0xffffffff)),
105 		(void *)((unsigned long)(iomapp->iomap_offset & 0xffffffff)),
106 		(void *)((unsigned long)(iomapp->iomap_delta)),
107 		(void *)((unsigned long)(iomapp->iomap_bsize)),
108 		(void *)((unsigned long)(iomapp->iomap_bn)),
109 		(void *)(__psint_t)(imapp->br_startoff),
110 		(void *)((unsigned long)(imapp->br_blockcount)),
111 		(void *)(__psint_t)(imapp->br_startblock));
112 }
113 #else
114 #define xfs_iomap_enter_trace(tag, io, offset, count)
115 #define xfs_iomap_map_trace(tag, io, offset, count, iomapp, imapp, flags)
116 #endif
117 
118 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
119 						<< mp->m_writeio_log)
120 #define XFS_STRAT_WRITE_IMAPS	2
121 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
122 
123 STATIC int
124 xfs_imap_to_bmap(
125 	xfs_inode_t	*ip,
126 	xfs_off_t	offset,
127 	xfs_bmbt_irec_t *imap,
128 	xfs_iomap_t	*iomapp,
129 	int		imaps,			/* Number of imap entries */
130 	int		iomaps,			/* Number of iomap entries */
131 	int		flags)
132 {
133 	xfs_mount_t	*mp = ip->i_mount;
134 	int		pbm;
135 	xfs_fsblock_t	start_block;
136 
137 
138 	for (pbm = 0; imaps && pbm < iomaps; imaps--, iomapp++, imap++, pbm++) {
139 		iomapp->iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff);
140 		iomapp->iomap_delta = offset - iomapp->iomap_offset;
141 		iomapp->iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount);
142 		iomapp->iomap_flags = flags;
143 
144 		if (XFS_IS_REALTIME_INODE(ip)) {
145 			iomapp->iomap_flags |= IOMAP_REALTIME;
146 			iomapp->iomap_target = mp->m_rtdev_targp;
147 		} else {
148 			iomapp->iomap_target = mp->m_ddev_targp;
149 		}
150 		start_block = imap->br_startblock;
151 		if (start_block == HOLESTARTBLOCK) {
152 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
153 			iomapp->iomap_flags |= IOMAP_HOLE;
154 		} else if (start_block == DELAYSTARTBLOCK) {
155 			iomapp->iomap_bn = IOMAP_DADDR_NULL;
156 			iomapp->iomap_flags |= IOMAP_DELAY;
157 		} else {
158 			iomapp->iomap_bn = XFS_FSB_TO_DB(ip, start_block);
159 			if (ISUNWRITTEN(imap))
160 				iomapp->iomap_flags |= IOMAP_UNWRITTEN;
161 		}
162 
163 		offset += iomapp->iomap_bsize - iomapp->iomap_delta;
164 	}
165 	return pbm;	/* Return the number filled */
166 }
167 
168 int
169 xfs_iomap(
170 	xfs_inode_t	*ip,
171 	xfs_off_t	offset,
172 	ssize_t		count,
173 	int		flags,
174 	xfs_iomap_t	*iomapp,
175 	int		*niomaps)
176 {
177 	xfs_mount_t	*mp = ip->i_mount;
178 	xfs_fileoff_t	offset_fsb, end_fsb;
179 	int		error = 0;
180 	int		lockmode = 0;
181 	xfs_bmbt_irec_t	imap;
182 	int		nimaps = 1;
183 	int		bmapi_flags = 0;
184 	int		iomap_flags = 0;
185 
186 	ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
187 
188 	if (XFS_FORCED_SHUTDOWN(mp))
189 		return XFS_ERROR(EIO);
190 
191 	switch (flags & (BMAPI_READ | BMAPI_WRITE | BMAPI_ALLOCATE)) {
192 	case BMAPI_READ:
193 		xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, ip, offset, count);
194 		lockmode = xfs_ilock_map_shared(ip);
195 		bmapi_flags = XFS_BMAPI_ENTIRE;
196 		break;
197 	case BMAPI_WRITE:
198 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, ip, offset, count);
199 		lockmode = XFS_ILOCK_EXCL;
200 		if (flags & BMAPI_IGNSTATE)
201 			bmapi_flags |= XFS_BMAPI_IGSTATE|XFS_BMAPI_ENTIRE;
202 		xfs_ilock(ip, lockmode);
203 		break;
204 	case BMAPI_ALLOCATE:
205 		xfs_iomap_enter_trace(XFS_IOMAP_ALLOC_ENTER, ip, offset, count);
206 		lockmode = XFS_ILOCK_SHARED;
207 		bmapi_flags = XFS_BMAPI_ENTIRE;
208 
209 		/* Attempt non-blocking lock */
210 		if (flags & BMAPI_TRYLOCK) {
211 			if (!xfs_ilock_nowait(ip, lockmode))
212 				return XFS_ERROR(EAGAIN);
213 		} else {
214 			xfs_ilock(ip, lockmode);
215 		}
216 		break;
217 	default:
218 		BUG();
219 	}
220 
221 	ASSERT(offset <= mp->m_maxioffset);
222 	if ((xfs_fsize_t)offset + count > mp->m_maxioffset)
223 		count = mp->m_maxioffset - offset;
224 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
225 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
226 
227 	error = xfs_bmapi(NULL, ip, offset_fsb,
228 			(xfs_filblks_t)(end_fsb - offset_fsb),
229 			bmapi_flags,  NULL, 0, &imap,
230 			&nimaps, NULL, NULL);
231 
232 	if (error)
233 		goto out;
234 
235 	switch (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)) {
236 	case BMAPI_WRITE:
237 		/* If we found an extent, return it */
238 		if (nimaps &&
239 		    (imap.br_startblock != HOLESTARTBLOCK) &&
240 		    (imap.br_startblock != DELAYSTARTBLOCK)) {
241 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip,
242 					offset, count, iomapp, &imap, flags);
243 			break;
244 		}
245 
246 		if (flags & (BMAPI_DIRECT|BMAPI_MMAP)) {
247 			error = xfs_iomap_write_direct(ip, offset, count, flags,
248 						       &imap, &nimaps, nimaps);
249 		} else {
250 			error = xfs_iomap_write_delay(ip, offset, count, flags,
251 						      &imap, &nimaps);
252 		}
253 		if (!error) {
254 			xfs_iomap_map_trace(XFS_IOMAP_ALLOC_MAP, ip,
255 					offset, count, iomapp, &imap, flags);
256 		}
257 		iomap_flags = IOMAP_NEW;
258 		break;
259 	case BMAPI_ALLOCATE:
260 		/* If we found an extent, return it */
261 		xfs_iunlock(ip, lockmode);
262 		lockmode = 0;
263 
264 		if (nimaps && !ISNULLSTARTBLOCK(imap.br_startblock)) {
265 			xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip,
266 					offset, count, iomapp, &imap, flags);
267 			break;
268 		}
269 
270 		error = xfs_iomap_write_allocate(ip, offset, count,
271 						 &imap, &nimaps);
272 		break;
273 	}
274 
275 	if (nimaps) {
276 		*niomaps = xfs_imap_to_bmap(ip, offset, &imap,
277 					    iomapp, nimaps, *niomaps, iomap_flags);
278 	} else if (niomaps) {
279 		*niomaps = 0;
280 	}
281 
282 out:
283 	if (lockmode)
284 		xfs_iunlock(ip, lockmode);
285 	return XFS_ERROR(error);
286 }
287 
288 
289 STATIC int
290 xfs_iomap_eof_align_last_fsb(
291 	xfs_mount_t	*mp,
292 	xfs_inode_t	*ip,
293 	xfs_fsize_t	isize,
294 	xfs_extlen_t	extsize,
295 	xfs_fileoff_t	*last_fsb)
296 {
297 	xfs_fileoff_t	new_last_fsb = 0;
298 	xfs_extlen_t	align;
299 	int		eof, error;
300 
301 	if (XFS_IS_REALTIME_INODE(ip))
302 		;
303 	/*
304 	 * If mounted with the "-o swalloc" option, roundup the allocation
305 	 * request to a stripe width boundary if the file size is >=
306 	 * stripe width and we are allocating past the allocation eof.
307 	 */
308 	else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
309 	        (isize >= XFS_FSB_TO_B(mp, mp->m_swidth)))
310 		new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
311 	/*
312 	 * Roundup the allocation request to a stripe unit (m_dalign) boundary
313 	 * if the file size is >= stripe unit size, and we are allocating past
314 	 * the allocation eof.
315 	 */
316 	else if (mp->m_dalign && (isize >= XFS_FSB_TO_B(mp, mp->m_dalign)))
317 		new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
318 
319 	/*
320 	 * Always round up the allocation request to an extent boundary
321 	 * (when file on a real-time subvolume or has di_extsize hint).
322 	 */
323 	if (extsize) {
324 		if (new_last_fsb)
325 			align = roundup_64(new_last_fsb, extsize);
326 		else
327 			align = extsize;
328 		new_last_fsb = roundup_64(*last_fsb, align);
329 	}
330 
331 	if (new_last_fsb) {
332 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
333 		if (error)
334 			return error;
335 		if (eof)
336 			*last_fsb = new_last_fsb;
337 	}
338 	return 0;
339 }
340 
341 STATIC int
342 xfs_flush_space(
343 	xfs_inode_t	*ip,
344 	int		*fsynced,
345 	int		*ioflags)
346 {
347 	switch (*fsynced) {
348 	case 0:
349 		if (ip->i_delayed_blks) {
350 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
351 			xfs_flush_inode(ip);
352 			xfs_ilock(ip, XFS_ILOCK_EXCL);
353 			*fsynced = 1;
354 		} else {
355 			*ioflags |= BMAPI_SYNC;
356 			*fsynced = 2;
357 		}
358 		return 0;
359 	case 1:
360 		*fsynced = 2;
361 		*ioflags |= BMAPI_SYNC;
362 		return 0;
363 	case 2:
364 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
365 		xfs_flush_device(ip);
366 		xfs_ilock(ip, XFS_ILOCK_EXCL);
367 		*fsynced = 3;
368 		return 0;
369 	}
370 	return 1;
371 }
372 
373 STATIC int
374 xfs_cmn_err_fsblock_zero(
375 	xfs_inode_t	*ip,
376 	xfs_bmbt_irec_t	*imap)
377 {
378 	xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
379 			"Access to block zero in inode %llu "
380 			"start_block: %llx start_off: %llx "
381 			"blkcnt: %llx extent-state: %x\n",
382 		(unsigned long long)ip->i_ino,
383 		(unsigned long long)imap->br_startblock,
384 		(unsigned long long)imap->br_startoff,
385 		(unsigned long long)imap->br_blockcount,
386 		imap->br_state);
387 	return EFSCORRUPTED;
388 }
389 
390 int
391 xfs_iomap_write_direct(
392 	xfs_inode_t	*ip,
393 	xfs_off_t	offset,
394 	size_t		count,
395 	int		flags,
396 	xfs_bmbt_irec_t *ret_imap,
397 	int		*nmaps,
398 	int		found)
399 {
400 	xfs_mount_t	*mp = ip->i_mount;
401 	xfs_fileoff_t	offset_fsb;
402 	xfs_fileoff_t	last_fsb;
403 	xfs_filblks_t	count_fsb, resaligned;
404 	xfs_fsblock_t	firstfsb;
405 	xfs_extlen_t	extsz, temp;
406 	xfs_fsize_t	isize;
407 	int		nimaps;
408 	int		bmapi_flag;
409 	int		quota_flag;
410 	int		rt;
411 	xfs_trans_t	*tp;
412 	xfs_bmbt_irec_t imap;
413 	xfs_bmap_free_t free_list;
414 	uint		qblocks, resblks, resrtextents;
415 	int		committed;
416 	int		error;
417 
418 	/*
419 	 * Make sure that the dquots are there. This doesn't hold
420 	 * the ilock across a disk read.
421 	 */
422 	error = XFS_QM_DQATTACH(ip->i_mount, ip, XFS_QMOPT_ILOCKED);
423 	if (error)
424 		return XFS_ERROR(error);
425 
426 	rt = XFS_IS_REALTIME_INODE(ip);
427 	extsz = xfs_get_extsz_hint(ip);
428 
429 	isize = ip->i_size;
430 	if (ip->i_new_size > isize)
431 		isize = ip->i_new_size;
432 
433 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
434 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
435 	if ((offset + count) > isize) {
436 		error = xfs_iomap_eof_align_last_fsb(mp, ip, isize, extsz,
437 							&last_fsb);
438 		if (error)
439 			goto error_out;
440 	} else {
441 		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
442 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
443 					ret_imap->br_blockcount +
444 					ret_imap->br_startoff);
445 	}
446 	count_fsb = last_fsb - offset_fsb;
447 	ASSERT(count_fsb > 0);
448 
449 	resaligned = count_fsb;
450 	if (unlikely(extsz)) {
451 		if ((temp = do_mod(offset_fsb, extsz)))
452 			resaligned += temp;
453 		if ((temp = do_mod(resaligned, extsz)))
454 			resaligned += extsz - temp;
455 	}
456 
457 	if (unlikely(rt)) {
458 		resrtextents = qblocks = resaligned;
459 		resrtextents /= mp->m_sb.sb_rextsize;
460 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
461 		quota_flag = XFS_QMOPT_RES_RTBLKS;
462 	} else {
463 		resrtextents = 0;
464 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
465 		quota_flag = XFS_QMOPT_RES_REGBLKS;
466 	}
467 
468 	/*
469 	 * Allocate and setup the transaction
470 	 */
471 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
472 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
473 	error = xfs_trans_reserve(tp, resblks,
474 			XFS_WRITE_LOG_RES(mp), resrtextents,
475 			XFS_TRANS_PERM_LOG_RES,
476 			XFS_WRITE_LOG_COUNT);
477 	/*
478 	 * Check for running out of space, note: need lock to return
479 	 */
480 	if (error)
481 		xfs_trans_cancel(tp, 0);
482 	xfs_ilock(ip, XFS_ILOCK_EXCL);
483 	if (error)
484 		goto error_out;
485 
486 	error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
487 					      qblocks, 0, quota_flag);
488 	if (error)
489 		goto error1;
490 
491 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
492 	xfs_trans_ihold(tp, ip);
493 
494 	bmapi_flag = XFS_BMAPI_WRITE;
495 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
496 		bmapi_flag |= XFS_BMAPI_PREALLOC;
497 
498 	/*
499 	 * Issue the xfs_bmapi() call to allocate the blocks
500 	 */
501 	XFS_BMAP_INIT(&free_list, &firstfsb);
502 	nimaps = 1;
503 	error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
504 		&firstfsb, 0, &imap, &nimaps, &free_list, NULL);
505 	if (error)
506 		goto error0;
507 
508 	/*
509 	 * Complete the transaction
510 	 */
511 	error = xfs_bmap_finish(&tp, &free_list, &committed);
512 	if (error)
513 		goto error0;
514 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
515 	if (error)
516 		goto error_out;
517 
518 	/*
519 	 * Copy any maps to caller's array and return any error.
520 	 */
521 	if (nimaps == 0) {
522 		error = ENOSPC;
523 		goto error_out;
524 	}
525 
526 	if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) {
527 		error = xfs_cmn_err_fsblock_zero(ip, &imap);
528 		goto error_out;
529 	}
530 
531 	*ret_imap = imap;
532 	*nmaps = 1;
533 	return 0;
534 
535 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
536 	xfs_bmap_cancel(&free_list);
537 	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
538 
539 error1:	/* Just cancel transaction */
540 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
541 	*nmaps = 0;	/* nothing set-up here */
542 
543 error_out:
544 	return XFS_ERROR(error);
545 }
546 
547 /*
548  * If the caller is doing a write at the end of the file,
549  * then extend the allocation out to the file system's write
550  * iosize.  We clean up any extra space left over when the
551  * file is closed in xfs_inactive().
552  *
553  * For sync writes, we are flushing delayed allocate space to
554  * try to make additional space available for allocation near
555  * the filesystem full boundary - preallocation hurts in that
556  * situation, of course.
557  */
558 STATIC int
559 xfs_iomap_eof_want_preallocate(
560 	xfs_mount_t	*mp,
561 	xfs_inode_t	*ip,
562 	xfs_fsize_t	isize,
563 	xfs_off_t	offset,
564 	size_t		count,
565 	int		ioflag,
566 	xfs_bmbt_irec_t *imap,
567 	int		nimaps,
568 	int		*prealloc)
569 {
570 	xfs_fileoff_t   start_fsb;
571 	xfs_filblks_t   count_fsb;
572 	xfs_fsblock_t	firstblock;
573 	int		n, error, imaps;
574 
575 	*prealloc = 0;
576 	if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
577 		return 0;
578 
579 	/*
580 	 * If there are any real blocks past eof, then don't
581 	 * do any speculative allocation.
582 	 */
583 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
584 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
585 	while (count_fsb > 0) {
586 		imaps = nimaps;
587 		firstblock = NULLFSBLOCK;
588 		error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
589 				  &firstblock, 0, imap, &imaps, NULL, NULL);
590 		if (error)
591 			return error;
592 		for (n = 0; n < imaps; n++) {
593 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
594 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
595 				return 0;
596 			start_fsb += imap[n].br_blockcount;
597 			count_fsb -= imap[n].br_blockcount;
598 		}
599 	}
600 	*prealloc = 1;
601 	return 0;
602 }
603 
604 int
605 xfs_iomap_write_delay(
606 	xfs_inode_t	*ip,
607 	xfs_off_t	offset,
608 	size_t		count,
609 	int		ioflag,
610 	xfs_bmbt_irec_t *ret_imap,
611 	int		*nmaps)
612 {
613 	xfs_mount_t	*mp = ip->i_mount;
614 	xfs_fileoff_t	offset_fsb;
615 	xfs_fileoff_t	last_fsb;
616 	xfs_off_t	aligned_offset;
617 	xfs_fileoff_t	ioalign;
618 	xfs_fsblock_t	firstblock;
619 	xfs_extlen_t	extsz;
620 	xfs_fsize_t	isize;
621 	int		nimaps;
622 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
623 	int		prealloc, fsynced = 0;
624 	int		error;
625 
626 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
627 
628 	/*
629 	 * Make sure that the dquots are there. This doesn't hold
630 	 * the ilock across a disk read.
631 	 */
632 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
633 	if (error)
634 		return XFS_ERROR(error);
635 
636 	extsz = xfs_get_extsz_hint(ip);
637 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
638 
639 retry:
640 	isize = ip->i_size;
641 	if (ip->i_new_size > isize)
642 		isize = ip->i_new_size;
643 
644 	error = xfs_iomap_eof_want_preallocate(mp, ip, isize, offset, count,
645 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
646 	if (error)
647 		return error;
648 
649 	if (prealloc) {
650 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
651 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
652 		last_fsb = ioalign + mp->m_writeio_blocks;
653 	} else {
654 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
655 	}
656 
657 	if (prealloc || extsz) {
658 		error = xfs_iomap_eof_align_last_fsb(mp, ip, isize, extsz,
659 							&last_fsb);
660 		if (error)
661 			return error;
662 	}
663 
664 	nimaps = XFS_WRITE_IMAPS;
665 	firstblock = NULLFSBLOCK;
666 	error = xfs_bmapi(NULL, ip, offset_fsb,
667 			  (xfs_filblks_t)(last_fsb - offset_fsb),
668 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
669 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
670 			  &nimaps, NULL, NULL);
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 - flush delalloc, and retry..
677 	 */
678 	if (nimaps == 0) {
679 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
680 					ip, 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 	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
689 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
690 
691 	*ret_imap = imap[0];
692 	*nmaps = 1;
693 
694 	return 0;
695 }
696 
697 /*
698  * Pass in a delayed allocate extent, convert it to real extents;
699  * return to the caller the extent we create which maps on top of
700  * the originating callers request.
701  *
702  * Called without a lock on the inode.
703  *
704  * We no longer bother to look at the incoming map - all we have to
705  * guarantee is that whatever we allocate fills the required range.
706  */
707 int
708 xfs_iomap_write_allocate(
709 	xfs_inode_t	*ip,
710 	xfs_off_t	offset,
711 	size_t		count,
712 	xfs_bmbt_irec_t *map,
713 	int		*retmap)
714 {
715 	xfs_mount_t	*mp = ip->i_mount;
716 	xfs_fileoff_t	offset_fsb, last_block;
717 	xfs_fileoff_t	end_fsb, map_start_fsb;
718 	xfs_fsblock_t	first_block;
719 	xfs_bmap_free_t	free_list;
720 	xfs_filblks_t	count_fsb;
721 	xfs_bmbt_irec_t	imap;
722 	xfs_trans_t	*tp;
723 	int		nimaps, committed;
724 	int		error = 0;
725 	int		nres;
726 
727 	*retmap = 0;
728 
729 	/*
730 	 * Make sure that the dquots are there.
731 	 */
732 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
733 		return XFS_ERROR(error);
734 
735 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
736 	count_fsb = map->br_blockcount;
737 	map_start_fsb = map->br_startoff;
738 
739 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
740 
741 	while (count_fsb != 0) {
742 		/*
743 		 * Set up a transaction with which to allocate the
744 		 * backing store for the file.  Do allocations in a
745 		 * loop until we get some space in the range we are
746 		 * interested in.  The other space that might be allocated
747 		 * is in the delayed allocation extent on which we sit
748 		 * but before our buffer starts.
749 		 */
750 
751 		nimaps = 0;
752 		while (nimaps == 0) {
753 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
754 			tp->t_flags |= XFS_TRANS_RESERVE;
755 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
756 			error = xfs_trans_reserve(tp, nres,
757 					XFS_WRITE_LOG_RES(mp),
758 					0, XFS_TRANS_PERM_LOG_RES,
759 					XFS_WRITE_LOG_COUNT);
760 			if (error) {
761 				xfs_trans_cancel(tp, 0);
762 				return XFS_ERROR(error);
763 			}
764 			xfs_ilock(ip, XFS_ILOCK_EXCL);
765 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
766 			xfs_trans_ihold(tp, ip);
767 
768 			XFS_BMAP_INIT(&free_list, &first_block);
769 
770 			/*
771 			 * it is possible that the extents have changed since
772 			 * we did the read call as we dropped the ilock for a
773 			 * while. We have to be careful about truncates or hole
774 			 * punchs here - we are not allowed to allocate
775 			 * non-delalloc blocks here.
776 			 *
777 			 * The only protection against truncation is the pages
778 			 * for the range we are being asked to convert are
779 			 * locked and hence a truncate will block on them
780 			 * first.
781 			 *
782 			 * As a result, if we go beyond the range we really
783 			 * need and hit an delalloc extent boundary followed by
784 			 * a hole while we have excess blocks in the map, we
785 			 * will fill the hole incorrectly and overrun the
786 			 * transaction reservation.
787 			 *
788 			 * Using a single map prevents this as we are forced to
789 			 * check each map we look for overlap with the desired
790 			 * range and abort as soon as we find it. Also, given
791 			 * that we only return a single map, having one beyond
792 			 * what we can return is probably a bit silly.
793 			 *
794 			 * We also need to check that we don't go beyond EOF;
795 			 * this is a truncate optimisation as a truncate sets
796 			 * the new file size before block on the pages we
797 			 * currently have locked under writeback. Because they
798 			 * are about to be tossed, we don't need to write them
799 			 * back....
800 			 */
801 			nimaps = 1;
802 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
803 			error = xfs_bmap_last_offset(NULL, ip, &last_block,
804 							XFS_DATA_FORK);
805 			if (error)
806 				goto trans_cancel;
807 
808 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
809 			if ((map_start_fsb + count_fsb) > last_block) {
810 				count_fsb = last_block - map_start_fsb;
811 				if (count_fsb == 0) {
812 					error = EAGAIN;
813 					goto trans_cancel;
814 				}
815 			}
816 
817 			/* Go get the actual blocks */
818 			error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
819 					XFS_BMAPI_WRITE, &first_block, 1,
820 					&imap, &nimaps, &free_list, NULL);
821 			if (error)
822 				goto trans_cancel;
823 
824 			error = xfs_bmap_finish(&tp, &free_list, &committed);
825 			if (error)
826 				goto trans_cancel;
827 
828 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
829 			if (error)
830 				goto error0;
831 
832 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
833 		}
834 
835 		/*
836 		 * See if we were able to allocate an extent that
837 		 * covers at least part of the callers request
838 		 */
839 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
840 			return xfs_cmn_err_fsblock_zero(ip, &imap);
841 
842 		if ((offset_fsb >= imap.br_startoff) &&
843 		    (offset_fsb < (imap.br_startoff +
844 				   imap.br_blockcount))) {
845 			*map = imap;
846 			*retmap = 1;
847 			XFS_STATS_INC(xs_xstrat_quick);
848 			return 0;
849 		}
850 
851 		/*
852 		 * So far we have not mapped the requested part of the
853 		 * file, just surrounding data, try again.
854 		 */
855 		count_fsb -= imap.br_blockcount;
856 		map_start_fsb = imap.br_startoff + imap.br_blockcount;
857 	}
858 
859 trans_cancel:
860 	xfs_bmap_cancel(&free_list);
861 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
862 error0:
863 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
864 	return XFS_ERROR(error);
865 }
866 
867 int
868 xfs_iomap_write_unwritten(
869 	xfs_inode_t	*ip,
870 	xfs_off_t	offset,
871 	size_t		count)
872 {
873 	xfs_mount_t	*mp = ip->i_mount;
874 	xfs_fileoff_t	offset_fsb;
875 	xfs_filblks_t	count_fsb;
876 	xfs_filblks_t	numblks_fsb;
877 	xfs_fsblock_t	firstfsb;
878 	int		nimaps;
879 	xfs_trans_t	*tp;
880 	xfs_bmbt_irec_t imap;
881 	xfs_bmap_free_t free_list;
882 	uint		resblks;
883 	int		committed;
884 	int		error;
885 
886 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN, ip, offset, count);
887 
888 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
889 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
890 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
891 
892 	/*
893 	 * Reserve enough blocks in this transaction for two complete extent
894 	 * btree splits.  We may be converting the middle part of an unwritten
895 	 * extent and in this case we will insert two new extents in the btree
896 	 * each of which could cause a full split.
897 	 *
898 	 * This reservation amount will be used in the first call to
899 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
900 	 * rest of the operation.
901 	 */
902 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
903 
904 	do {
905 		/*
906 		 * set up a transaction to convert the range of extents
907 		 * from unwritten to real. Do allocations in a loop until
908 		 * we have covered the range passed in.
909 		 */
910 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
911 		tp->t_flags |= XFS_TRANS_RESERVE;
912 		error = xfs_trans_reserve(tp, resblks,
913 				XFS_WRITE_LOG_RES(mp), 0,
914 				XFS_TRANS_PERM_LOG_RES,
915 				XFS_WRITE_LOG_COUNT);
916 		if (error) {
917 			xfs_trans_cancel(tp, 0);
918 			return XFS_ERROR(error);
919 		}
920 
921 		xfs_ilock(ip, XFS_ILOCK_EXCL);
922 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
923 		xfs_trans_ihold(tp, ip);
924 
925 		/*
926 		 * Modify the unwritten extent state of the buffer.
927 		 */
928 		XFS_BMAP_INIT(&free_list, &firstfsb);
929 		nimaps = 1;
930 		error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
931 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
932 				  1, &imap, &nimaps, &free_list, NULL);
933 		if (error)
934 			goto error_on_bmapi_transaction;
935 
936 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
937 		if (error)
938 			goto error_on_bmapi_transaction;
939 
940 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
941 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
942 		if (error)
943 			return XFS_ERROR(error);
944 
945 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
946 			return xfs_cmn_err_fsblock_zero(ip, &imap);
947 
948 		if ((numblks_fsb = imap.br_blockcount) == 0) {
949 			/*
950 			 * The numblks_fsb value should always get
951 			 * smaller, otherwise the loop is stuck.
952 			 */
953 			ASSERT(imap.br_blockcount);
954 			break;
955 		}
956 		offset_fsb += numblks_fsb;
957 		count_fsb -= numblks_fsb;
958 	} while (count_fsb > 0);
959 
960 	return 0;
961 
962 error_on_bmapi_transaction:
963 	xfs_bmap_cancel(&free_list);
964 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
965 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
966 	return XFS_ERROR(error);
967 }
968