xref: /openbmc/linux/fs/xfs/xfs_iomap.c (revision 8e9356c6)
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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_bmap.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_iomap.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_quota.h"
39 #include "xfs_dquot_item.h"
40 #include "xfs_dquot.h"
41 #include "xfs_dinode.h"
42 
43 
44 #define XFS_WRITEIO_ALIGN(mp,off)	(((off) >> mp->m_writeio_log) \
45 						<< mp->m_writeio_log)
46 #define XFS_WRITE_IMAPS		XFS_BMAP_MAX_NMAP
47 
48 STATIC int
49 xfs_iomap_eof_align_last_fsb(
50 	xfs_mount_t	*mp,
51 	xfs_inode_t	*ip,
52 	xfs_extlen_t	extsize,
53 	xfs_fileoff_t	*last_fsb)
54 {
55 	xfs_fileoff_t	new_last_fsb = 0;
56 	xfs_extlen_t	align = 0;
57 	int		eof, error;
58 
59 	if (!XFS_IS_REALTIME_INODE(ip)) {
60 		/*
61 		 * Round up the allocation request to a stripe unit
62 		 * (m_dalign) boundary if the file size is >= stripe unit
63 		 * size, and we are allocating past the allocation eof.
64 		 *
65 		 * If mounted with the "-o swalloc" option the alignment is
66 		 * increased from the strip unit size to the stripe width.
67 		 */
68 		if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
69 			align = mp->m_swidth;
70 		else if (mp->m_dalign)
71 			align = mp->m_dalign;
72 
73 		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
74 			new_last_fsb = roundup_64(*last_fsb, align);
75 	}
76 
77 	/*
78 	 * Always round up the allocation request to an extent boundary
79 	 * (when file on a real-time subvolume or has di_extsize hint).
80 	 */
81 	if (extsize) {
82 		if (new_last_fsb)
83 			align = roundup_64(new_last_fsb, extsize);
84 		else
85 			align = extsize;
86 		new_last_fsb = roundup_64(*last_fsb, align);
87 	}
88 
89 	if (new_last_fsb) {
90 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
91 		if (error)
92 			return error;
93 		if (eof)
94 			*last_fsb = new_last_fsb;
95 	}
96 	return 0;
97 }
98 
99 STATIC int
100 xfs_alert_fsblock_zero(
101 	xfs_inode_t	*ip,
102 	xfs_bmbt_irec_t	*imap)
103 {
104 	xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
105 			"Access to block zero in inode %llu "
106 			"start_block: %llx start_off: %llx "
107 			"blkcnt: %llx extent-state: %x",
108 		(unsigned long long)ip->i_ino,
109 		(unsigned long long)imap->br_startblock,
110 		(unsigned long long)imap->br_startoff,
111 		(unsigned long long)imap->br_blockcount,
112 		imap->br_state);
113 	return EFSCORRUPTED;
114 }
115 
116 int
117 xfs_iomap_write_direct(
118 	xfs_inode_t	*ip,
119 	xfs_off_t	offset,
120 	size_t		count,
121 	xfs_bmbt_irec_t *imap,
122 	int		nmaps)
123 {
124 	xfs_mount_t	*mp = ip->i_mount;
125 	xfs_fileoff_t	offset_fsb;
126 	xfs_fileoff_t	last_fsb;
127 	xfs_filblks_t	count_fsb, resaligned;
128 	xfs_fsblock_t	firstfsb;
129 	xfs_extlen_t	extsz, temp;
130 	int		nimaps;
131 	int		bmapi_flag;
132 	int		quota_flag;
133 	int		rt;
134 	xfs_trans_t	*tp;
135 	xfs_bmap_free_t free_list;
136 	uint		qblocks, resblks, resrtextents;
137 	int		committed;
138 	int		error;
139 
140 	error = xfs_qm_dqattach(ip, 0);
141 	if (error)
142 		return XFS_ERROR(error);
143 
144 	rt = XFS_IS_REALTIME_INODE(ip);
145 	extsz = xfs_get_extsz_hint(ip);
146 
147 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
148 	last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
149 	if ((offset + count) > XFS_ISIZE(ip)) {
150 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
151 		if (error)
152 			return XFS_ERROR(error);
153 	} else {
154 		if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
155 			last_fsb = MIN(last_fsb, (xfs_fileoff_t)
156 					imap->br_blockcount +
157 					imap->br_startoff);
158 	}
159 	count_fsb = last_fsb - offset_fsb;
160 	ASSERT(count_fsb > 0);
161 
162 	resaligned = count_fsb;
163 	if (unlikely(extsz)) {
164 		if ((temp = do_mod(offset_fsb, extsz)))
165 			resaligned += temp;
166 		if ((temp = do_mod(resaligned, extsz)))
167 			resaligned += extsz - temp;
168 	}
169 
170 	if (unlikely(rt)) {
171 		resrtextents = qblocks = resaligned;
172 		resrtextents /= mp->m_sb.sb_rextsize;
173 		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
174 		quota_flag = XFS_QMOPT_RES_RTBLKS;
175 	} else {
176 		resrtextents = 0;
177 		resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
178 		quota_flag = XFS_QMOPT_RES_REGBLKS;
179 	}
180 
181 	/*
182 	 * Allocate and setup the transaction
183 	 */
184 	tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
185 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
186 				  resblks, resrtextents);
187 	/*
188 	 * Check for running out of space, note: need lock to return
189 	 */
190 	if (error) {
191 		xfs_trans_cancel(tp, 0);
192 		return XFS_ERROR(error);
193 	}
194 
195 	xfs_ilock(ip, XFS_ILOCK_EXCL);
196 
197 	error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
198 	if (error)
199 		goto out_trans_cancel;
200 
201 	xfs_trans_ijoin(tp, ip, 0);
202 
203 	bmapi_flag = 0;
204 	if (offset < XFS_ISIZE(ip) || extsz)
205 		bmapi_flag |= XFS_BMAPI_PREALLOC;
206 
207 	/*
208 	 * From this point onwards we overwrite the imap pointer that the
209 	 * caller gave to us.
210 	 */
211 	xfs_bmap_init(&free_list, &firstfsb);
212 	nimaps = 1;
213 	error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
214 				&firstfsb, 0, imap, &nimaps, &free_list);
215 	if (error)
216 		goto out_bmap_cancel;
217 
218 	/*
219 	 * Complete the transaction
220 	 */
221 	error = xfs_bmap_finish(&tp, &free_list, &committed);
222 	if (error)
223 		goto out_bmap_cancel;
224 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
225 	if (error)
226 		goto out_unlock;
227 
228 	/*
229 	 * Copy any maps to caller's array and return any error.
230 	 */
231 	if (nimaps == 0) {
232 		error = XFS_ERROR(ENOSPC);
233 		goto out_unlock;
234 	}
235 
236 	if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
237 		error = xfs_alert_fsblock_zero(ip, imap);
238 
239 out_unlock:
240 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
241 	return error;
242 
243 out_bmap_cancel:
244 	xfs_bmap_cancel(&free_list);
245 	xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
246 out_trans_cancel:
247 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
248 	goto out_unlock;
249 }
250 
251 /*
252  * If the caller is doing a write at the end of the file, then extend the
253  * allocation out to the file system's write iosize.  We clean up any extra
254  * space left over when the file is closed in xfs_inactive().
255  *
256  * If we find we already have delalloc preallocation beyond EOF, don't do more
257  * preallocation as it it not needed.
258  */
259 STATIC int
260 xfs_iomap_eof_want_preallocate(
261 	xfs_mount_t	*mp,
262 	xfs_inode_t	*ip,
263 	xfs_off_t	offset,
264 	size_t		count,
265 	xfs_bmbt_irec_t *imap,
266 	int		nimaps,
267 	int		*prealloc)
268 {
269 	xfs_fileoff_t   start_fsb;
270 	xfs_filblks_t   count_fsb;
271 	xfs_fsblock_t	firstblock;
272 	int		n, error, imaps;
273 	int		found_delalloc = 0;
274 
275 	*prealloc = 0;
276 	if (offset + count <= XFS_ISIZE(ip))
277 		return 0;
278 
279 	/*
280 	 * If the file is smaller than the minimum prealloc and we are using
281 	 * dynamic preallocation, don't do any preallocation at all as it is
282 	 * likely this is the only write to the file that is going to be done.
283 	 */
284 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
285 	    XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
286 		return 0;
287 
288 	/*
289 	 * If there are any real blocks past eof, then don't
290 	 * do any speculative allocation.
291 	 */
292 	start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
293 	count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
294 	while (count_fsb > 0) {
295 		imaps = nimaps;
296 		firstblock = NULLFSBLOCK;
297 		error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
298 				       0);
299 		if (error)
300 			return error;
301 		for (n = 0; n < imaps; n++) {
302 			if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
303 			    (imap[n].br_startblock != DELAYSTARTBLOCK))
304 				return 0;
305 			start_fsb += imap[n].br_blockcount;
306 			count_fsb -= imap[n].br_blockcount;
307 
308 			if (imap[n].br_startblock == DELAYSTARTBLOCK)
309 				found_delalloc = 1;
310 		}
311 	}
312 	if (!found_delalloc)
313 		*prealloc = 1;
314 	return 0;
315 }
316 
317 /*
318  * Determine the initial size of the preallocation. We are beyond the current
319  * EOF here, but we need to take into account whether this is a sparse write or
320  * an extending write when determining the preallocation size.  Hence we need to
321  * look up the extent that ends at the current write offset and use the result
322  * to determine the preallocation size.
323  *
324  * If the extent is a hole, then preallocation is essentially disabled.
325  * Otherwise we take the size of the preceeding data extent as the basis for the
326  * preallocation size. If the size of the extent is greater than half the
327  * maximum extent length, then use the current offset as the basis. This ensures
328  * that for large files the preallocation size always extends to MAXEXTLEN
329  * rather than falling short due to things like stripe unit/width alignment of
330  * real extents.
331  */
332 STATIC xfs_fsblock_t
333 xfs_iomap_eof_prealloc_initial_size(
334 	struct xfs_mount	*mp,
335 	struct xfs_inode	*ip,
336 	xfs_off_t		offset,
337 	xfs_bmbt_irec_t		*imap,
338 	int			nimaps)
339 {
340 	xfs_fileoff_t   start_fsb;
341 	int		imaps = 1;
342 	int		error;
343 
344 	ASSERT(nimaps >= imaps);
345 
346 	/* if we are using a specific prealloc size, return now */
347 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
348 		return 0;
349 
350 	/* If the file is small, then use the minimum prealloc */
351 	if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
352 		return 0;
353 
354 	/*
355 	 * As we write multiple pages, the offset will always align to the
356 	 * start of a page and hence point to a hole at EOF. i.e. if the size is
357 	 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
358 	 * will return FSB 1. Hence if there are blocks in the file, we want to
359 	 * point to the block prior to the EOF block and not the hole that maps
360 	 * directly at @offset.
361 	 */
362 	start_fsb = XFS_B_TO_FSB(mp, offset);
363 	if (start_fsb)
364 		start_fsb--;
365 	error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
366 	if (error)
367 		return 0;
368 
369 	ASSERT(imaps == 1);
370 	if (imap[0].br_startblock == HOLESTARTBLOCK)
371 		return 0;
372 	if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
373 		return imap[0].br_blockcount << 1;
374 	return XFS_B_TO_FSB(mp, offset);
375 }
376 
377 STATIC bool
378 xfs_quota_need_throttle(
379 	struct xfs_inode *ip,
380 	int type,
381 	xfs_fsblock_t alloc_blocks)
382 {
383 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
384 
385 	if (!dq || !xfs_this_quota_on(ip->i_mount, type))
386 		return false;
387 
388 	/* no hi watermark, no throttle */
389 	if (!dq->q_prealloc_hi_wmark)
390 		return false;
391 
392 	/* under the lo watermark, no throttle */
393 	if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
394 		return false;
395 
396 	return true;
397 }
398 
399 STATIC void
400 xfs_quota_calc_throttle(
401 	struct xfs_inode *ip,
402 	int type,
403 	xfs_fsblock_t *qblocks,
404 	int *qshift)
405 {
406 	int64_t freesp;
407 	int shift = 0;
408 	struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
409 
410 	/* over hi wmark, squash the prealloc completely */
411 	if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
412 		*qblocks = 0;
413 		return;
414 	}
415 
416 	freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
417 	if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
418 		shift = 2;
419 		if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
420 			shift += 2;
421 		if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
422 			shift += 2;
423 	}
424 
425 	/* only overwrite the throttle values if we are more aggressive */
426 	if ((freesp >> shift) < (*qblocks >> *qshift)) {
427 		*qblocks = freesp;
428 		*qshift = shift;
429 	}
430 }
431 
432 /*
433  * If we don't have a user specified preallocation size, dynamically increase
434  * the preallocation size as the size of the file grows. Cap the maximum size
435  * at a single extent or less if the filesystem is near full. The closer the
436  * filesystem is to full, the smaller the maximum prealocation.
437  */
438 STATIC xfs_fsblock_t
439 xfs_iomap_prealloc_size(
440 	struct xfs_mount	*mp,
441 	struct xfs_inode	*ip,
442 	xfs_off_t		offset,
443 	struct xfs_bmbt_irec	*imap,
444 	int			nimaps)
445 {
446 	xfs_fsblock_t		alloc_blocks = 0;
447 	int			shift = 0;
448 	int64_t			freesp;
449 	xfs_fsblock_t		qblocks;
450 	int			qshift = 0;
451 
452 	alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
453 							   imap, nimaps);
454 	if (!alloc_blocks)
455 		goto check_writeio;
456 	qblocks = alloc_blocks;
457 
458 	/*
459 	 * MAXEXTLEN is not a power of two value but we round the prealloc down
460 	 * to the nearest power of two value after throttling. To prevent the
461 	 * round down from unconditionally reducing the maximum supported prealloc
462 	 * size, we round up first, apply appropriate throttling, round down and
463 	 * cap the value to MAXEXTLEN.
464 	 */
465 	alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
466 				       alloc_blocks);
467 
468 	xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
469 	freesp = mp->m_sb.sb_fdblocks;
470 	if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
471 		shift = 2;
472 		if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
473 			shift++;
474 		if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
475 			shift++;
476 		if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
477 			shift++;
478 		if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
479 			shift++;
480 	}
481 
482 	/*
483 	 * Check each quota to cap the prealloc size and provide a shift
484 	 * value to throttle with.
485 	 */
486 	if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
487 		xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift);
488 	if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
489 		xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift);
490 	if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
491 		xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift);
492 
493 	/*
494 	 * The final prealloc size is set to the minimum of free space available
495 	 * in each of the quotas and the overall filesystem.
496 	 *
497 	 * The shift throttle value is set to the maximum value as determined by
498 	 * the global low free space values and per-quota low free space values.
499 	 */
500 	alloc_blocks = MIN(alloc_blocks, qblocks);
501 	shift = MAX(shift, qshift);
502 
503 	if (shift)
504 		alloc_blocks >>= shift;
505 	/*
506 	 * rounddown_pow_of_two() returns an undefined result if we pass in
507 	 * alloc_blocks = 0.
508 	 */
509 	if (alloc_blocks)
510 		alloc_blocks = rounddown_pow_of_two(alloc_blocks);
511 	if (alloc_blocks > MAXEXTLEN)
512 		alloc_blocks = MAXEXTLEN;
513 
514 	/*
515 	 * If we are still trying to allocate more space than is
516 	 * available, squash the prealloc hard. This can happen if we
517 	 * have a large file on a small filesystem and the above
518 	 * lowspace thresholds are smaller than MAXEXTLEN.
519 	 */
520 	while (alloc_blocks && alloc_blocks >= freesp)
521 		alloc_blocks >>= 4;
522 
523 check_writeio:
524 	if (alloc_blocks < mp->m_writeio_blocks)
525 		alloc_blocks = mp->m_writeio_blocks;
526 
527 	trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
528 				      mp->m_writeio_blocks);
529 
530 	return alloc_blocks;
531 }
532 
533 int
534 xfs_iomap_write_delay(
535 	xfs_inode_t	*ip,
536 	xfs_off_t	offset,
537 	size_t		count,
538 	xfs_bmbt_irec_t *ret_imap)
539 {
540 	xfs_mount_t	*mp = ip->i_mount;
541 	xfs_fileoff_t	offset_fsb;
542 	xfs_fileoff_t	last_fsb;
543 	xfs_off_t	aligned_offset;
544 	xfs_fileoff_t	ioalign;
545 	xfs_extlen_t	extsz;
546 	int		nimaps;
547 	xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
548 	int		prealloc;
549 	int		error;
550 
551 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
552 
553 	/*
554 	 * Make sure that the dquots are there. This doesn't hold
555 	 * the ilock across a disk read.
556 	 */
557 	error = xfs_qm_dqattach_locked(ip, 0);
558 	if (error)
559 		return XFS_ERROR(error);
560 
561 	extsz = xfs_get_extsz_hint(ip);
562 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
563 
564 	error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
565 				imap, XFS_WRITE_IMAPS, &prealloc);
566 	if (error)
567 		return error;
568 
569 retry:
570 	if (prealloc) {
571 		xfs_fsblock_t	alloc_blocks;
572 
573 		alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
574 						       XFS_WRITE_IMAPS);
575 
576 		aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
577 		ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
578 		last_fsb = ioalign + alloc_blocks;
579 	} else {
580 		last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
581 	}
582 
583 	if (prealloc || extsz) {
584 		error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
585 		if (error)
586 			return error;
587 	}
588 
589 	/*
590 	 * Make sure preallocation does not create extents beyond the range we
591 	 * actually support in this filesystem.
592 	 */
593 	if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
594 		last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
595 
596 	ASSERT(last_fsb > offset_fsb);
597 
598 	nimaps = XFS_WRITE_IMAPS;
599 	error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
600 				imap, &nimaps, XFS_BMAPI_ENTIRE);
601 	switch (error) {
602 	case 0:
603 	case ENOSPC:
604 	case EDQUOT:
605 		break;
606 	default:
607 		return XFS_ERROR(error);
608 	}
609 
610 	/*
611 	 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
612 	 * without EOF preallocation.
613 	 */
614 	if (nimaps == 0) {
615 		trace_xfs_delalloc_enospc(ip, offset, count);
616 		if (prealloc) {
617 			prealloc = 0;
618 			error = 0;
619 			goto retry;
620 		}
621 		return XFS_ERROR(error ? error : ENOSPC);
622 	}
623 
624 	if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
625 		return xfs_alert_fsblock_zero(ip, &imap[0]);
626 
627 	/*
628 	 * Tag the inode as speculatively preallocated so we can reclaim this
629 	 * space on demand, if necessary.
630 	 */
631 	if (prealloc)
632 		xfs_inode_set_eofblocks_tag(ip);
633 
634 	*ret_imap = imap[0];
635 	return 0;
636 }
637 
638 /*
639  * Pass in a delayed allocate extent, convert it to real extents;
640  * return to the caller the extent we create which maps on top of
641  * the originating callers request.
642  *
643  * Called without a lock on the inode.
644  *
645  * We no longer bother to look at the incoming map - all we have to
646  * guarantee is that whatever we allocate fills the required range.
647  */
648 int
649 xfs_iomap_write_allocate(
650 	xfs_inode_t	*ip,
651 	xfs_off_t	offset,
652 	xfs_bmbt_irec_t *imap)
653 {
654 	xfs_mount_t	*mp = ip->i_mount;
655 	xfs_fileoff_t	offset_fsb, last_block;
656 	xfs_fileoff_t	end_fsb, map_start_fsb;
657 	xfs_fsblock_t	first_block;
658 	xfs_bmap_free_t	free_list;
659 	xfs_filblks_t	count_fsb;
660 	xfs_trans_t	*tp;
661 	int		nimaps, committed;
662 	int		error = 0;
663 	int		nres;
664 
665 	/*
666 	 * Make sure that the dquots are there.
667 	 */
668 	error = xfs_qm_dqattach(ip, 0);
669 	if (error)
670 		return XFS_ERROR(error);
671 
672 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
673 	count_fsb = imap->br_blockcount;
674 	map_start_fsb = imap->br_startoff;
675 
676 	XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
677 
678 	while (count_fsb != 0) {
679 		/*
680 		 * Set up a transaction with which to allocate the
681 		 * backing store for the file.  Do allocations in a
682 		 * loop until we get some space in the range we are
683 		 * interested in.  The other space that might be allocated
684 		 * is in the delayed allocation extent on which we sit
685 		 * but before our buffer starts.
686 		 */
687 
688 		nimaps = 0;
689 		while (nimaps == 0) {
690 			tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
691 			tp->t_flags |= XFS_TRANS_RESERVE;
692 			nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
693 			error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
694 						  nres, 0);
695 			if (error) {
696 				xfs_trans_cancel(tp, 0);
697 				return XFS_ERROR(error);
698 			}
699 			xfs_ilock(ip, XFS_ILOCK_EXCL);
700 			xfs_trans_ijoin(tp, ip, 0);
701 
702 			xfs_bmap_init(&free_list, &first_block);
703 
704 			/*
705 			 * it is possible that the extents have changed since
706 			 * we did the read call as we dropped the ilock for a
707 			 * while. We have to be careful about truncates or hole
708 			 * punchs here - we are not allowed to allocate
709 			 * non-delalloc blocks here.
710 			 *
711 			 * The only protection against truncation is the pages
712 			 * for the range we are being asked to convert are
713 			 * locked and hence a truncate will block on them
714 			 * first.
715 			 *
716 			 * As a result, if we go beyond the range we really
717 			 * need and hit an delalloc extent boundary followed by
718 			 * a hole while we have excess blocks in the map, we
719 			 * will fill the hole incorrectly and overrun the
720 			 * transaction reservation.
721 			 *
722 			 * Using a single map prevents this as we are forced to
723 			 * check each map we look for overlap with the desired
724 			 * range and abort as soon as we find it. Also, given
725 			 * that we only return a single map, having one beyond
726 			 * what we can return is probably a bit silly.
727 			 *
728 			 * We also need to check that we don't go beyond EOF;
729 			 * this is a truncate optimisation as a truncate sets
730 			 * the new file size before block on the pages we
731 			 * currently have locked under writeback. Because they
732 			 * are about to be tossed, we don't need to write them
733 			 * back....
734 			 */
735 			nimaps = 1;
736 			end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
737 			error = xfs_bmap_last_offset(NULL, ip, &last_block,
738 							XFS_DATA_FORK);
739 			if (error)
740 				goto trans_cancel;
741 
742 			last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
743 			if ((map_start_fsb + count_fsb) > last_block) {
744 				count_fsb = last_block - map_start_fsb;
745 				if (count_fsb == 0) {
746 					error = EAGAIN;
747 					goto trans_cancel;
748 				}
749 			}
750 
751 			/*
752 			 * From this point onwards we overwrite the imap
753 			 * pointer that the caller gave to us.
754 			 */
755 			error = xfs_bmapi_write(tp, ip, map_start_fsb,
756 						count_fsb,
757 						XFS_BMAPI_STACK_SWITCH,
758 						&first_block, 1,
759 						imap, &nimaps, &free_list);
760 			if (error)
761 				goto trans_cancel;
762 
763 			error = xfs_bmap_finish(&tp, &free_list, &committed);
764 			if (error)
765 				goto trans_cancel;
766 
767 			error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
768 			if (error)
769 				goto error0;
770 
771 			xfs_iunlock(ip, XFS_ILOCK_EXCL);
772 		}
773 
774 		/*
775 		 * See if we were able to allocate an extent that
776 		 * covers at least part of the callers request
777 		 */
778 		if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
779 			return xfs_alert_fsblock_zero(ip, imap);
780 
781 		if ((offset_fsb >= imap->br_startoff) &&
782 		    (offset_fsb < (imap->br_startoff +
783 				   imap->br_blockcount))) {
784 			XFS_STATS_INC(xs_xstrat_quick);
785 			return 0;
786 		}
787 
788 		/*
789 		 * So far we have not mapped the requested part of the
790 		 * file, just surrounding data, try again.
791 		 */
792 		count_fsb -= imap->br_blockcount;
793 		map_start_fsb = imap->br_startoff + imap->br_blockcount;
794 	}
795 
796 trans_cancel:
797 	xfs_bmap_cancel(&free_list);
798 	xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
799 error0:
800 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
801 	return XFS_ERROR(error);
802 }
803 
804 int
805 xfs_iomap_write_unwritten(
806 	xfs_inode_t	*ip,
807 	xfs_off_t	offset,
808 	size_t		count)
809 {
810 	xfs_mount_t	*mp = ip->i_mount;
811 	xfs_fileoff_t	offset_fsb;
812 	xfs_filblks_t	count_fsb;
813 	xfs_filblks_t	numblks_fsb;
814 	xfs_fsblock_t	firstfsb;
815 	int		nimaps;
816 	xfs_trans_t	*tp;
817 	xfs_bmbt_irec_t imap;
818 	xfs_bmap_free_t free_list;
819 	xfs_fsize_t	i_size;
820 	uint		resblks;
821 	int		committed;
822 	int		error;
823 
824 	trace_xfs_unwritten_convert(ip, offset, count);
825 
826 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
827 	count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
828 	count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
829 
830 	/*
831 	 * Reserve enough blocks in this transaction for two complete extent
832 	 * btree splits.  We may be converting the middle part of an unwritten
833 	 * extent and in this case we will insert two new extents in the btree
834 	 * each of which could cause a full split.
835 	 *
836 	 * This reservation amount will be used in the first call to
837 	 * xfs_bmbt_split() to select an AG with enough space to satisfy the
838 	 * rest of the operation.
839 	 */
840 	resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
841 
842 	do {
843 		/*
844 		 * set up a transaction to convert the range of extents
845 		 * from unwritten to real. Do allocations in a loop until
846 		 * we have covered the range passed in.
847 		 *
848 		 * Note that we open code the transaction allocation here
849 		 * to pass KM_NOFS--we can't risk to recursing back into
850 		 * the filesystem here as we might be asked to write out
851 		 * the same inode that we complete here and might deadlock
852 		 * on the iolock.
853 		 */
854 		sb_start_intwrite(mp->m_super);
855 		tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
856 		tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
857 		error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
858 					  resblks, 0);
859 		if (error) {
860 			xfs_trans_cancel(tp, 0);
861 			return XFS_ERROR(error);
862 		}
863 
864 		xfs_ilock(ip, XFS_ILOCK_EXCL);
865 		xfs_trans_ijoin(tp, ip, 0);
866 
867 		/*
868 		 * Modify the unwritten extent state of the buffer.
869 		 */
870 		xfs_bmap_init(&free_list, &firstfsb);
871 		nimaps = 1;
872 		error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
873 				  XFS_BMAPI_CONVERT, &firstfsb,
874 				  1, &imap, &nimaps, &free_list);
875 		if (error)
876 			goto error_on_bmapi_transaction;
877 
878 		/*
879 		 * Log the updated inode size as we go.  We have to be careful
880 		 * to only log it up to the actual write offset if it is
881 		 * halfway into a block.
882 		 */
883 		i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
884 		if (i_size > offset + count)
885 			i_size = offset + count;
886 
887 		i_size = xfs_new_eof(ip, i_size);
888 		if (i_size) {
889 			ip->i_d.di_size = i_size;
890 			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
891 		}
892 
893 		error = xfs_bmap_finish(&tp, &free_list, &committed);
894 		if (error)
895 			goto error_on_bmapi_transaction;
896 
897 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
898 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
899 		if (error)
900 			return XFS_ERROR(error);
901 
902 		if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
903 			return xfs_alert_fsblock_zero(ip, &imap);
904 
905 		if ((numblks_fsb = imap.br_blockcount) == 0) {
906 			/*
907 			 * The numblks_fsb value should always get
908 			 * smaller, otherwise the loop is stuck.
909 			 */
910 			ASSERT(imap.br_blockcount);
911 			break;
912 		}
913 		offset_fsb += numblks_fsb;
914 		count_fsb -= numblks_fsb;
915 	} while (count_fsb > 0);
916 
917 	return 0;
918 
919 error_on_bmapi_transaction:
920 	xfs_bmap_cancel(&free_list);
921 	xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
922 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
923 	return XFS_ERROR(error);
924 }
925