xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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 	if (unlikely(rt)) {
455 		if (!(extsz = ip->i_d.di_extsize))
456 			extsz = mp->m_sb.sb_rextsize;
457 	} else {
458 		extsz = ip->i_d.di_extsize;
459 	}
460 
461 	isize = ip->i_size;
462 	if (io->io_new_size > isize)
463 		isize = io->io_new_size;
464 
465   	offset_fsb = XFS_B_TO_FSBT(mp, offset);
466   	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
467 	if ((offset + count) > isize) {
468 		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
469 							&last_fsb);
470 		if (error)
471 			goto error_out;
472 	} else {
473 		if (found && (ret_imap->br_startblock == HOLESTARTBLOCK))
474 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
475 					ret_imap->br_blockcount +
476 					ret_imap->br_startoff);
477 	}
478 	count_fsb = last_fsb - offset_fsb;
479 	ASSERT(count_fsb > 0);
480 
481 	resaligned = count_fsb;
482 	if (unlikely(extsz)) {
483 		if ((temp = do_mod(offset_fsb, extsz)))
484 			resaligned += temp;
485 		if ((temp = do_mod(resaligned, extsz)))
486 			resaligned += extsz - temp;
487 	}
488 
489 	if (unlikely(rt)) {
490 		resrtextents = qblocks = resaligned;
491 		resrtextents /= mp->m_sb.sb_rextsize;
492   		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
493   		quota_flag = XFS_QMOPT_RES_RTBLKS;
494   	} else {
495   		resrtextents = 0;
496 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
497   		quota_flag = XFS_QMOPT_RES_REGBLKS;
498   	}
499 
500 	/*
501 	 * Allocate and setup the transaction
502 	 */
503 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
504 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
505 	error = xfs_trans_reserve(tp, resblks,
506 			XFS_WRITE_LOG_RES(mp), resrtextents,
507 			XFS_TRANS_PERM_LOG_RES,
508 			XFS_WRITE_LOG_COUNT);
509 	/*
510 	 * Check for running out of space, note: need lock to return
511 	 */
512 	if (error)
513 		xfs_trans_cancel(tp, 0);
514 	xfs_ilock(ip, XFS_ILOCK_EXCL);
515 	if (error)
516 		goto error_out;
517 
518 	error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
519 					      qblocks, 0, quota_flag);
520 	if (error)
521 		goto error1;
522 
523 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
524 	xfs_trans_ihold(tp, ip);
525 
526 	bmapi_flag = XFS_BMAPI_WRITE;
527 	if ((flags & BMAPI_DIRECT) && (offset < ip->i_size || extsz))
528 		bmapi_flag |= XFS_BMAPI_PREALLOC;
529 
530 	/*
531 	 * Issue the xfs_bmapi() call to allocate the blocks
532 	 */
533 	XFS_BMAP_INIT(&free_list, &firstfsb);
534 	nimaps = 1;
535 	error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb, bmapi_flag,
536 		&firstfsb, 0, &imap, &nimaps, &free_list, NULL);
537 	if (error)
538 		goto error0;
539 
540 	/*
541 	 * Complete the transaction
542 	 */
543 	error = xfs_bmap_finish(&tp, &free_list, &committed);
544 	if (error)
545 		goto error0;
546 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
547 	if (error)
548 		goto error_out;
549 
550 	/*
551 	 * Copy any maps to caller's array and return any error.
552 	 */
553 	if (nimaps == 0) {
554 		error = ENOSPC;
555 		goto error_out;
556 	}
557 
558 	if (unlikely(!imap.br_startblock && !(io->io_flags & XFS_IOCORE_RT))) {
559 		error = xfs_cmn_err_fsblock_zero(ip, &imap);
560 		goto error_out;
561 	}
562 
563 	*ret_imap = imap;
564 	*nmaps = 1;
565 	return 0;
566 
567 error0:	/* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
568 	xfs_bmap_cancel(&free_list);
569 	XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
570 
571 error1:	/* Just cancel transaction */
572 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
573 	*nmaps = 0;	/* nothing set-up here */
574 
575 error_out:
576 	return XFS_ERROR(error);
577 }
578 
579 /*
580  * If the caller is doing a write at the end of the file,
581  * then extend the allocation out to the file system's write
582  * iosize.  We clean up any extra space left over when the
583  * file is closed in xfs_inactive().
584  *
585  * For sync writes, we are flushing delayed allocate space to
586  * try to make additional space available for allocation near
587  * the filesystem full boundary - preallocation hurts in that
588  * situation, of course.
589  */
590 STATIC int
591 xfs_iomap_eof_want_preallocate(
592 	xfs_mount_t	*mp,
593 	xfs_iocore_t	*io,
594 	xfs_fsize_t	isize,
595 	xfs_off_t	offset,
596 	size_t		count,
597 	int		ioflag,
598 	xfs_bmbt_irec_t *imap,
599 	int		nimaps,
600 	int		*prealloc)
601 {
602 	xfs_fileoff_t   start_fsb;
603 	xfs_filblks_t   count_fsb;
604 	xfs_fsblock_t	firstblock;
605 	int		n, error, imaps;
606 
607 	*prealloc = 0;
608 	if ((ioflag & BMAPI_SYNC) || (offset + count) <= isize)
609 		return 0;
610 
611 	/*
612 	 * If there are any real blocks past eof, then don't
613 	 * do any speculative allocation.
614 	 */
615 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
616 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
617 	while (count_fsb > 0) {
618 		imaps = nimaps;
619 		firstblock = NULLFSBLOCK;
620 		error = XFS_BMAPI(mp, NULL, io, start_fsb, count_fsb, 0,
621 				  &firstblock, 0, imap, &imaps, NULL, NULL);
622 		if (error)
623 			return error;
624 		for (n = 0; n < imaps; n++) {
625 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
626 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
627 				return 0;
628 			start_fsb += imap[n].br_blockcount;
629 			count_fsb -= imap[n].br_blockcount;
630 		}
631 	}
632 	*prealloc = 1;
633 	return 0;
634 }
635 
636 int
637 xfs_iomap_write_delay(
638 	xfs_inode_t	*ip,
639 	xfs_off_t	offset,
640 	size_t		count,
641 	int		ioflag,
642 	xfs_bmbt_irec_t *ret_imap,
643 	int		*nmaps)
644 {
645 	xfs_mount_t	*mp = ip->i_mount;
646 	xfs_iocore_t	*io = &ip->i_iocore;
647 	xfs_fileoff_t	offset_fsb;
648 	xfs_fileoff_t	last_fsb;
649 	xfs_off_t	aligned_offset;
650 	xfs_fileoff_t	ioalign;
651 	xfs_fsblock_t	firstblock;
652 	xfs_extlen_t	extsz;
653 	xfs_fsize_t	isize;
654 	int		nimaps;
655 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
656 	int		prealloc, fsynced = 0;
657 	int		error;
658 
659 	ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
660 
661 	/*
662 	 * Make sure that the dquots are there. This doesn't hold
663 	 * the ilock across a disk read.
664 	 */
665 	error = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED);
666 	if (error)
667 		return XFS_ERROR(error);
668 
669 	if (XFS_IS_REALTIME_INODE(ip)) {
670 		if (!(extsz = ip->i_d.di_extsize))
671 			extsz = mp->m_sb.sb_rextsize;
672 	} else {
673 		extsz = ip->i_d.di_extsize;
674 	}
675 
676 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
677 
678 retry:
679 	isize = ip->i_size;
680 	if (io->io_new_size > isize)
681 		isize = io->io_new_size;
682 
683 	error = xfs_iomap_eof_want_preallocate(mp, io, isize, offset, count,
684 				ioflag, imap, XFS_WRITE_IMAPS, &prealloc);
685 	if (error)
686 		return error;
687 
688 	if (prealloc) {
689 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
690 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
691 		last_fsb = ioalign + mp->m_writeio_blocks;
692 	} else {
693 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
694 	}
695 
696 	if (prealloc || extsz) {
697 		error = xfs_iomap_eof_align_last_fsb(mp, io, isize, extsz,
698 							&last_fsb);
699 		if (error)
700 			return error;
701 	}
702 
703 	nimaps = XFS_WRITE_IMAPS;
704 	firstblock = NULLFSBLOCK;
705 	error = XFS_BMAPI(mp, NULL, io, offset_fsb,
706 			  (xfs_filblks_t)(last_fsb - offset_fsb),
707 			  XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
708 			  XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
709 			  &nimaps, NULL, NULL);
710 	if (error && (error != ENOSPC))
711 		return XFS_ERROR(error);
712 
713 	/*
714 	 * If bmapi returned us nothing, and if we didn't get back EDQUOT,
715 	 * then we must have run out of space - flush delalloc, and retry..
716 	 */
717 	if (nimaps == 0) {
718 		xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
719 					io, offset, count);
720 		if (xfs_flush_space(ip, &fsynced, &ioflag))
721 			return XFS_ERROR(ENOSPC);
722 
723 		error = 0;
724 		goto retry;
725 	}
726 
727 	if (unlikely(!imap[0].br_startblock && !(io->io_flags & XFS_IOCORE_RT)))
728 		return xfs_cmn_err_fsblock_zero(ip, &imap[0]);
729 
730 	*ret_imap = imap[0];
731 	*nmaps = 1;
732 
733 	return 0;
734 }
735 
736 /*
737  * Pass in a delayed allocate extent, convert it to real extents;
738  * return to the caller the extent we create which maps on top of
739  * the originating callers request.
740  *
741  * Called without a lock on the inode.
742  */
743 int
744 xfs_iomap_write_allocate(
745 	xfs_inode_t	*ip,
746 	xfs_off_t	offset,
747 	size_t		count,
748 	xfs_bmbt_irec_t *map,
749 	int		*retmap)
750 {
751 	xfs_mount_t	*mp = ip->i_mount;
752 	xfs_iocore_t    *io = &ip->i_iocore;
753 	xfs_fileoff_t	offset_fsb, last_block;
754 	xfs_fileoff_t	end_fsb, map_start_fsb;
755 	xfs_fsblock_t	first_block;
756 	xfs_bmap_free_t	free_list;
757 	xfs_filblks_t	count_fsb;
758 	xfs_bmbt_irec_t	imap[XFS_STRAT_WRITE_IMAPS];
759 	xfs_trans_t	*tp;
760 	int		i, nimaps, committed;
761 	int		error = 0;
762 	int		nres;
763 
764 	*retmap = 0;
765 
766 	/*
767 	 * Make sure that the dquots are there.
768 	 */
769 	if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
770 		return XFS_ERROR(error);
771 
772 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
773 	count_fsb = map->br_blockcount;
774 	map_start_fsb = map->br_startoff;
775 
776 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
777 
778 	while (count_fsb != 0) {
779 		/*
780 		 * Set up a transaction with which to allocate the
781 		 * backing store for the file.  Do allocations in a
782 		 * loop until we get some space in the range we are
783 		 * interested in.  The other space that might be allocated
784 		 * is in the delayed allocation extent on which we sit
785 		 * but before our buffer starts.
786 		 */
787 
788 		nimaps = 0;
789 		while (nimaps == 0) {
790 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
791 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
792 			error = xfs_trans_reserve(tp, nres,
793 					XFS_WRITE_LOG_RES(mp),
794 					0, XFS_TRANS_PERM_LOG_RES,
795 					XFS_WRITE_LOG_COUNT);
796 			if (error == ENOSPC) {
797 				error = xfs_trans_reserve(tp, 0,
798 						XFS_WRITE_LOG_RES(mp),
799 						0,
800 						XFS_TRANS_PERM_LOG_RES,
801 						XFS_WRITE_LOG_COUNT);
802 			}
803 			if (error) {
804 				xfs_trans_cancel(tp, 0);
805 				return XFS_ERROR(error);
806 			}
807 			xfs_ilock(ip, XFS_ILOCK_EXCL);
808 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
809 			xfs_trans_ihold(tp, ip);
810 
811 			XFS_BMAP_INIT(&free_list, &first_block);
812 
813 			nimaps = XFS_STRAT_WRITE_IMAPS;
814 			/*
815 			 * Ensure we don't go beyond eof - it is possible
816 			 * the extents changed since we did the read call,
817 			 * we dropped the ilock in the interim.
818 			 */
819 
820 			end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
821 			xfs_bmap_last_offset(NULL, ip, &last_block,
822 				XFS_DATA_FORK);
823 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
824 			if ((map_start_fsb + count_fsb) > last_block) {
825 				count_fsb = last_block - map_start_fsb;
826 				if (count_fsb == 0) {
827 					error = EAGAIN;
828 					goto trans_cancel;
829 				}
830 			}
831 
832 			/* Go get the actual blocks */
833 			error = XFS_BMAPI(mp, tp, io, map_start_fsb, count_fsb,
834 					XFS_BMAPI_WRITE, &first_block, 1,
835 					imap, &nimaps, &free_list, NULL);
836 			if (error)
837 				goto trans_cancel;
838 
839 			error = xfs_bmap_finish(&tp, &free_list, &committed);
840 			if (error)
841 				goto trans_cancel;
842 
843 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
844 			if (error)
845 				goto error0;
846 
847 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
848 		}
849 
850 		/*
851 		 * See if we were able to allocate an extent that
852 		 * covers at least part of the callers request
853 		 */
854 		for (i = 0; i < nimaps; i++) {
855 			if (unlikely(!imap[i].br_startblock &&
856 				     !(io->io_flags & XFS_IOCORE_RT)))
857 				return xfs_cmn_err_fsblock_zero(ip, &imap[i]);
858 			if ((offset_fsb >= imap[i].br_startoff) &&
859 			    (offset_fsb < (imap[i].br_startoff +
860 					   imap[i].br_blockcount))) {
861 				*map = imap[i];
862 				*retmap = 1;
863 				XFS_STATS_INC(xs_xstrat_quick);
864 				return 0;
865 			}
866 			count_fsb -= imap[i].br_blockcount;
867 		}
868 
869 		/* So far we have not mapped the requested part of the
870 		 * file, just surrounding data, try again.
871 		 */
872 		nimaps--;
873 		map_start_fsb = imap[nimaps].br_startoff +
874 				imap[nimaps].br_blockcount;
875 	}
876 
877 trans_cancel:
878 	xfs_bmap_cancel(&free_list);
879 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
880 error0:
881 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
882 	return XFS_ERROR(error);
883 }
884 
885 int
886 xfs_iomap_write_unwritten(
887 	xfs_inode_t	*ip,
888 	xfs_off_t	offset,
889 	size_t		count)
890 {
891 	xfs_mount_t	*mp = ip->i_mount;
892 	xfs_iocore_t    *io = &ip->i_iocore;
893 	xfs_fileoff_t	offset_fsb;
894 	xfs_filblks_t	count_fsb;
895 	xfs_filblks_t	numblks_fsb;
896 	xfs_fsblock_t	firstfsb;
897 	int		nimaps;
898 	xfs_trans_t	*tp;
899 	xfs_bmbt_irec_t imap;
900 	xfs_bmap_free_t free_list;
901 	uint		resblks;
902 	int		committed;
903 	int		error;
904 
905 	xfs_iomap_enter_trace(XFS_IOMAP_UNWRITTEN,
906 				&ip->i_iocore, offset, count);
907 
908 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
909 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
910 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
911 
912 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
913 
914 	do {
915 		/*
916 		 * set up a transaction to convert the range of extents
917 		 * from unwritten to real. Do allocations in a loop until
918 		 * we have covered the range passed in.
919 		 */
920 
921 		tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
922 		error = xfs_trans_reserve(tp, resblks,
923 				XFS_WRITE_LOG_RES(mp), 0,
924 				XFS_TRANS_PERM_LOG_RES,
925 				XFS_WRITE_LOG_COUNT);
926 		if (error) {
927 			xfs_trans_cancel(tp, 0);
928 			return XFS_ERROR(error);
929 		}
930 
931 		xfs_ilock(ip, XFS_ILOCK_EXCL);
932 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
933 		xfs_trans_ihold(tp, ip);
934 
935 		/*
936 		 * Modify the unwritten extent state of the buffer.
937 		 */
938 		XFS_BMAP_INIT(&free_list, &firstfsb);
939 		nimaps = 1;
940 		error = XFS_BMAPI(mp, tp, io, offset_fsb, count_fsb,
941 				  XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
942 				  1, &imap, &nimaps, &free_list, NULL);
943 		if (error)
944 			goto error_on_bmapi_transaction;
945 
946 		error = xfs_bmap_finish(&(tp), &(free_list), &committed);
947 		if (error)
948 			goto error_on_bmapi_transaction;
949 
950 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
951 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
952 		if (error)
953 			return XFS_ERROR(error);
954 
955 		if (unlikely(!imap.br_startblock &&
956 			     !(io->io_flags & XFS_IOCORE_RT)))
957 			return xfs_cmn_err_fsblock_zero(ip, &imap);
958 
959 		if ((numblks_fsb = imap.br_blockcount) == 0) {
960 			/*
961 			 * The numblks_fsb value should always get
962 			 * smaller, otherwise the loop is stuck.
963 			 */
964 			ASSERT(imap.br_blockcount);
965 			break;
966 		}
967 		offset_fsb += numblks_fsb;
968 		count_fsb -= numblks_fsb;
969 	} while (count_fsb > 0);
970 
971 	return 0;
972 
973 error_on_bmapi_transaction:
974 	xfs_bmap_cancel(&free_list);
975 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
976 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
977 	return XFS_ERROR(error);
978 }
979