xref: /linux/fs/xfs/xfs_rtalloc.c (revision 3be042cf46feeedf664152d063376b5c17026d1d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_bmap.h"
16 #include "xfs_bmap_btree.h"
17 #include "xfs_bmap_util.h"
18 #include "xfs_trans.h"
19 #include "xfs_trans_space.h"
20 #include "xfs_icache.h"
21 #include "xfs_rtalloc.h"
22 #include "xfs_sb.h"
23 #include "xfs_rtbitmap.h"
24 #include "xfs_quota.h"
25 
26 /*
27  * Return whether there are any free extents in the size range given
28  * by low and high, for the bitmap block bbno.
29  */
30 STATIC int
31 xfs_rtany_summary(
32 	struct xfs_rtalloc_args	*args,
33 	int			low,	/* low log2 extent size */
34 	int			high,	/* high log2 extent size */
35 	xfs_fileoff_t		bbno,	/* bitmap block number */
36 	int			*maxlog) /* out: max log2 extent size free */
37 {
38 	struct xfs_mount	*mp = args->mp;
39 	int			error;
40 	int			log;	/* loop counter, log2 of ext. size */
41 	xfs_suminfo_t		sum;	/* summary data */
42 
43 	/* There are no extents at levels >= m_rsum_cache[bbno]. */
44 	if (mp->m_rsum_cache) {
45 		high = min(high, mp->m_rsum_cache[bbno] - 1);
46 		if (low > high) {
47 			*maxlog = -1;
48 			return 0;
49 		}
50 	}
51 
52 	/*
53 	 * Loop over logs of extent sizes.
54 	 */
55 	for (log = high; log >= low; log--) {
56 		/*
57 		 * Get one summary datum.
58 		 */
59 		error = xfs_rtget_summary(args, log, bbno, &sum);
60 		if (error) {
61 			return error;
62 		}
63 		/*
64 		 * If there are any, return success.
65 		 */
66 		if (sum) {
67 			*maxlog = log;
68 			goto out;
69 		}
70 	}
71 	/*
72 	 * Found nothing, return failure.
73 	 */
74 	*maxlog = -1;
75 out:
76 	/* There were no extents at levels > log. */
77 	if (mp->m_rsum_cache && log + 1 < mp->m_rsum_cache[bbno])
78 		mp->m_rsum_cache[bbno] = log + 1;
79 	return 0;
80 }
81 
82 
83 /*
84  * Copy and transform the summary file, given the old and new
85  * parameters in the mount structures.
86  */
87 STATIC int
88 xfs_rtcopy_summary(
89 	struct xfs_rtalloc_args	*oargs,
90 	struct xfs_rtalloc_args	*nargs)
91 {
92 	xfs_fileoff_t		bbno;	/* bitmap block number */
93 	int			error;
94 	int			log;	/* summary level number (log length) */
95 	xfs_suminfo_t		sum;	/* summary data */
96 
97 	for (log = oargs->mp->m_rsumlevels - 1; log >= 0; log--) {
98 		for (bbno = oargs->mp->m_sb.sb_rbmblocks - 1;
99 		     (xfs_srtblock_t)bbno >= 0;
100 		     bbno--) {
101 			error = xfs_rtget_summary(oargs, log, bbno, &sum);
102 			if (error)
103 				goto out;
104 			if (sum == 0)
105 				continue;
106 			error = xfs_rtmodify_summary(oargs, log, bbno, -sum);
107 			if (error)
108 				goto out;
109 			error = xfs_rtmodify_summary(nargs, log, bbno, sum);
110 			if (error)
111 				goto out;
112 			ASSERT(sum > 0);
113 		}
114 	}
115 	error = 0;
116 out:
117 	xfs_rtbuf_cache_relse(oargs);
118 	return 0;
119 }
120 /*
121  * Mark an extent specified by start and len allocated.
122  * Updates all the summary information as well as the bitmap.
123  */
124 STATIC int
125 xfs_rtallocate_range(
126 	struct xfs_rtalloc_args	*args,
127 	xfs_rtxnum_t		start,	/* start rtext to allocate */
128 	xfs_rtxlen_t		len)	/* in/out: summary block number */
129 {
130 	struct xfs_mount	*mp = args->mp;
131 	xfs_rtxnum_t		end;	/* end of the allocated rtext */
132 	int			error;
133 	xfs_rtxnum_t		postblock = 0; /* first rtext allocated > end */
134 	xfs_rtxnum_t		preblock = 0; /* first rtext allocated < start */
135 
136 	end = start + len - 1;
137 	/*
138 	 * Assume we're allocating out of the middle of a free extent.
139 	 * We need to find the beginning and end of the extent so we can
140 	 * properly update the summary.
141 	 */
142 	error = xfs_rtfind_back(args, start, 0, &preblock);
143 	if (error)
144 		return error;
145 
146 	/*
147 	 * Find the next allocated block (end of free extent).
148 	 */
149 	error = xfs_rtfind_forw(args, end, mp->m_sb.sb_rextents - 1,
150 			&postblock);
151 	if (error)
152 		return error;
153 
154 	/*
155 	 * Decrement the summary information corresponding to the entire
156 	 * (old) free extent.
157 	 */
158 	error = xfs_rtmodify_summary(args,
159 			xfs_highbit64(postblock + 1 - preblock),
160 			xfs_rtx_to_rbmblock(mp, preblock), -1);
161 	if (error)
162 		return error;
163 
164 	/*
165 	 * If there are blocks not being allocated at the front of the
166 	 * old extent, add summary data for them to be free.
167 	 */
168 	if (preblock < start) {
169 		error = xfs_rtmodify_summary(args,
170 				xfs_highbit64(start - preblock),
171 				xfs_rtx_to_rbmblock(mp, preblock), 1);
172 		if (error)
173 			return error;
174 	}
175 
176 	/*
177 	 * If there are blocks not being allocated at the end of the
178 	 * old extent, add summary data for them to be free.
179 	 */
180 	if (postblock > end) {
181 		error = xfs_rtmodify_summary(args,
182 				xfs_highbit64(postblock - end),
183 				xfs_rtx_to_rbmblock(mp, end + 1), 1);
184 		if (error)
185 			return error;
186 	}
187 
188 	/*
189 	 * Modify the bitmap to mark this extent allocated.
190 	 */
191 	return xfs_rtmodify_range(args, start, len, 0);
192 }
193 
194 /*
195  * Make sure we don't run off the end of the rt volume.  Be careful that
196  * adjusting maxlen downwards doesn't cause us to fail the alignment checks.
197  */
198 static inline xfs_rtxlen_t
199 xfs_rtallocate_clamp_len(
200 	struct xfs_mount	*mp,
201 	xfs_rtxnum_t		startrtx,
202 	xfs_rtxlen_t		rtxlen,
203 	xfs_rtxlen_t		prod)
204 {
205 	xfs_rtxlen_t		ret;
206 
207 	ret = min(mp->m_sb.sb_rextents, startrtx + rtxlen) - startrtx;
208 	return rounddown(ret, prod);
209 }
210 
211 /*
212  * Attempt to allocate an extent minlen<=len<=maxlen starting from
213  * bitmap block bbno.  If we don't get maxlen then use prod to trim
214  * the length, if given.  Returns error; returns starting block in *rtx.
215  * The lengths are all in rtextents.
216  */
217 STATIC int
218 xfs_rtallocate_extent_block(
219 	struct xfs_rtalloc_args	*args,
220 	xfs_fileoff_t		bbno,	/* bitmap block number */
221 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
222 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
223 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
224 	xfs_rtxnum_t		*nextp,	/* out: next rtext to try */
225 	xfs_rtxlen_t		prod,	/* extent product factor */
226 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
227 {
228 	struct xfs_mount	*mp = args->mp;
229 	xfs_rtxnum_t		besti;	/* best rtext found so far */
230 	xfs_rtxnum_t		bestlen;/* best length found so far */
231 	xfs_rtxnum_t		end;	/* last rtext in chunk */
232 	int			error;
233 	xfs_rtxnum_t		i;	/* current rtext trying */
234 	xfs_rtxnum_t		next;	/* next rtext to try */
235 	int			stat;	/* status from internal calls */
236 
237 	/*
238 	 * Loop over all the extents starting in this bitmap block,
239 	 * looking for one that's long enough.
240 	 */
241 	for (i = xfs_rbmblock_to_rtx(mp, bbno), besti = -1, bestlen = 0,
242 		end = xfs_rbmblock_to_rtx(mp, bbno + 1) - 1;
243 	     i <= end;
244 	     i++) {
245 		/* Make sure we don't scan off the end of the rt volume. */
246 		maxlen = xfs_rtallocate_clamp_len(mp, i, maxlen, prod);
247 
248 		/*
249 		 * See if there's a free extent of maxlen starting at i.
250 		 * If it's not so then next will contain the first non-free.
251 		 */
252 		error = xfs_rtcheck_range(args, i, maxlen, 1, &next, &stat);
253 		if (error)
254 			return error;
255 		if (stat) {
256 			/*
257 			 * i for maxlen is all free, allocate and return that.
258 			 */
259 			bestlen = maxlen;
260 			besti = i;
261 			goto allocate;
262 		}
263 
264 		/*
265 		 * In the case where we have a variable-sized allocation
266 		 * request, figure out how big this free piece is,
267 		 * and if it's big enough for the minimum, and the best
268 		 * so far, remember it.
269 		 */
270 		if (minlen < maxlen) {
271 			xfs_rtxnum_t	thislen;	/* this extent size */
272 
273 			thislen = next - i;
274 			if (thislen >= minlen && thislen > bestlen) {
275 				besti = i;
276 				bestlen = thislen;
277 			}
278 		}
279 		/*
280 		 * If not done yet, find the start of the next free space.
281 		 */
282 		if (next >= end)
283 			break;
284 		error = xfs_rtfind_forw(args, next, end, &i);
285 		if (error)
286 			return error;
287 	}
288 
289 	/*
290 	 * Searched the whole thing & didn't find a maxlen free extent.
291 	 */
292 	if (minlen > maxlen || besti == -1) {
293 		/*
294 		 * Allocation failed.  Set *nextp to the next block to try.
295 		 */
296 		*nextp = next;
297 		return -ENOSPC;
298 	}
299 
300 	/*
301 	 * If size should be a multiple of prod, make that so.
302 	 */
303 	if (prod > 1) {
304 		xfs_rtxlen_t	p;	/* amount to trim length by */
305 
306 		div_u64_rem(bestlen, prod, &p);
307 		if (p)
308 			bestlen -= p;
309 	}
310 
311 	/*
312 	 * Allocate besti for bestlen & return that.
313 	 */
314 allocate:
315 	error = xfs_rtallocate_range(args, besti, bestlen);
316 	if (error)
317 		return error;
318 	*len = bestlen;
319 	*rtx = besti;
320 	return 0;
321 }
322 
323 /*
324  * Allocate an extent of length minlen<=len<=maxlen, starting at block
325  * bno.  If we don't get maxlen then use prod to trim the length, if given.
326  * Returns error; returns starting block in *rtx.
327  * The lengths are all in rtextents.
328  */
329 STATIC int
330 xfs_rtallocate_extent_exact(
331 	struct xfs_rtalloc_args	*args,
332 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
333 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
334 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
335 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
336 	xfs_rtxlen_t		prod,	/* extent product factor */
337 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
338 {
339 	int			error;
340 	xfs_rtxlen_t		i;	/* extent length trimmed due to prod */
341 	int			isfree;	/* extent is free */
342 	xfs_rtxnum_t		next;	/* next rtext to try (dummy) */
343 
344 	ASSERT(minlen % prod == 0);
345 	ASSERT(maxlen % prod == 0);
346 	/*
347 	 * Check if the range in question (for maxlen) is free.
348 	 */
349 	error = xfs_rtcheck_range(args, start, maxlen, 1, &next, &isfree);
350 	if (error)
351 		return error;
352 
353 	if (!isfree) {
354 		/*
355 		 * If not, allocate what there is, if it's at least minlen.
356 		 */
357 		maxlen = next - start;
358 		if (maxlen < minlen)
359 			return -ENOSPC;
360 
361 		/*
362 		 * Trim off tail of extent, if prod is specified.
363 		 */
364 		if (prod > 1 && (i = maxlen % prod)) {
365 			maxlen -= i;
366 			if (maxlen < minlen)
367 				return -ENOSPC;
368 		}
369 	}
370 
371 	/*
372 	 * Allocate what we can and return it.
373 	 */
374 	error = xfs_rtallocate_range(args, start, maxlen);
375 	if (error)
376 		return error;
377 	*len = maxlen;
378 	*rtx = start;
379 	return 0;
380 }
381 
382 /*
383  * Allocate an extent of length minlen<=len<=maxlen, starting as near
384  * to start as possible.  If we don't get maxlen then use prod to trim
385  * the length, if given.  The lengths are all in rtextents.
386  */
387 STATIC int
388 xfs_rtallocate_extent_near(
389 	struct xfs_rtalloc_args	*args,
390 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
391 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
392 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
393 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
394 	xfs_rtxlen_t		prod,	/* extent product factor */
395 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
396 {
397 	struct xfs_mount	*mp = args->mp;
398 	int			maxlog;	/* max useful extent from summary */
399 	xfs_fileoff_t		bbno;	/* bitmap block number */
400 	int			error;
401 	int			i;	/* bitmap block offset (loop control) */
402 	int			j;	/* secondary loop control */
403 	int			log2len; /* log2 of minlen */
404 	xfs_rtxnum_t		n;	/* next rtext to try */
405 
406 	ASSERT(minlen % prod == 0);
407 	ASSERT(maxlen % prod == 0);
408 
409 	/*
410 	 * If the block number given is off the end, silently set it to
411 	 * the last block.
412 	 */
413 	if (start >= mp->m_sb.sb_rextents)
414 		start = mp->m_sb.sb_rextents - 1;
415 
416 	/* Make sure we don't run off the end of the rt volume. */
417 	maxlen = xfs_rtallocate_clamp_len(mp, start, maxlen, prod);
418 	if (maxlen < minlen)
419 		return -ENOSPC;
420 
421 	/*
422 	 * Try the exact allocation first.
423 	 */
424 	error = xfs_rtallocate_extent_exact(args, start, minlen, maxlen, len,
425 			prod, rtx);
426 	if (error != -ENOSPC)
427 		return error;
428 
429 
430 	bbno = xfs_rtx_to_rbmblock(mp, start);
431 	i = 0;
432 	j = -1;
433 	ASSERT(minlen != 0);
434 	log2len = xfs_highbit32(minlen);
435 	/*
436 	 * Loop over all bitmap blocks (bbno + i is current block).
437 	 */
438 	for (;;) {
439 		/*
440 		 * Get summary information of extents of all useful levels
441 		 * starting in this bitmap block.
442 		 */
443 		error = xfs_rtany_summary(args, log2len, mp->m_rsumlevels - 1,
444 				bbno + i, &maxlog);
445 		if (error)
446 			return error;
447 
448 		/*
449 		 * If there are any useful extents starting here, try
450 		 * allocating one.
451 		 */
452 		if (maxlog >= 0) {
453 			xfs_extlen_t maxavail =
454 				min_t(xfs_rtblock_t, maxlen,
455 				      (1ULL << (maxlog + 1)) - 1);
456 			/*
457 			 * On the positive side of the starting location.
458 			 */
459 			if (i >= 0) {
460 				/*
461 				 * Try to allocate an extent starting in
462 				 * this block.
463 				 */
464 				error = xfs_rtallocate_extent_block(args,
465 						bbno + i, minlen, maxavail, len,
466 						&n, prod, rtx);
467 				if (error != -ENOSPC)
468 					return error;
469 			}
470 			/*
471 			 * On the negative side of the starting location.
472 			 */
473 			else {		/* i < 0 */
474 				int	maxblocks;
475 
476 				/*
477 				 * Loop backwards to find the end of the extent
478 				 * we found in the realtime summary.
479 				 *
480 				 * maxblocks is the maximum possible number of
481 				 * bitmap blocks from the start of the extent
482 				 * to the end of the extent.
483 				 */
484 				if (maxlog == 0)
485 					maxblocks = 0;
486 				else if (maxlog < mp->m_blkbit_log)
487 					maxblocks = 1;
488 				else
489 					maxblocks = 2 << (maxlog - mp->m_blkbit_log);
490 
491 				/*
492 				 * We need to check bbno + i + maxblocks down to
493 				 * bbno + i. We already checked bbno down to
494 				 * bbno + j + 1, so we don't need to check those
495 				 * again.
496 				 */
497 				j = min(i + maxblocks, j);
498 				for (; j >= i; j--) {
499 					error = xfs_rtallocate_extent_block(args,
500 							bbno + j, minlen,
501 							maxavail, len, &n, prod,
502 							rtx);
503 					if (error != -ENOSPC)
504 						return error;
505 				}
506 			}
507 		}
508 		/*
509 		 * Loop control.  If we were on the positive side, and there's
510 		 * still more blocks on the negative side, go there.
511 		 */
512 		if (i > 0 && (int)bbno - i >= 0)
513 			i = -i;
514 		/*
515 		 * If positive, and no more negative, but there are more
516 		 * positive, go there.
517 		 */
518 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
519 			i++;
520 		/*
521 		 * If negative or 0 (just started), and there are positive
522 		 * blocks to go, go there.  The 0 case moves to block 1.
523 		 */
524 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
525 			i = 1 - i;
526 		/*
527 		 * If negative or 0 and there are more negative blocks,
528 		 * go there.
529 		 */
530 		else if (i <= 0 && (int)bbno + i > 0)
531 			i--;
532 		/*
533 		 * Must be done.  Return failure.
534 		 */
535 		else
536 			break;
537 	}
538 	return -ENOSPC;
539 }
540 
541 static int
542 xfs_rtalloc_sumlevel(
543 	struct xfs_rtalloc_args	*args,
544 	int			l,	/* level number */
545 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
546 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
547 	xfs_rtxlen_t		prod,	/* extent product factor */
548 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
549 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
550 {
551 	xfs_fileoff_t		i;	/* bitmap block number */
552 
553 	for (i = 0; i < args->mp->m_sb.sb_rbmblocks; i++) {
554 		xfs_suminfo_t	sum;	/* summary information for extents */
555 		xfs_rtxnum_t	n;	/* next rtext to be tried */
556 		int		error;
557 
558 		error = xfs_rtget_summary(args, l, i, &sum);
559 		if (error)
560 			return error;
561 
562 		/*
563 		 * Nothing there, on to the next block.
564 		 */
565 		if (!sum)
566 			continue;
567 
568 		/*
569 		 * Try allocating the extent.
570 		 */
571 		error = xfs_rtallocate_extent_block(args, i, minlen, maxlen,
572 				len, &n, prod, rtx);
573 		if (error != -ENOSPC)
574 			return error;
575 
576 		/*
577 		 * If the "next block to try" returned from the allocator is
578 		 * beyond the next bitmap block, skip to that bitmap block.
579 		 */
580 		if (xfs_rtx_to_rbmblock(args->mp, n) > i + 1)
581 			i = xfs_rtx_to_rbmblock(args->mp, n) - 1;
582 	}
583 
584 	return -ENOSPC;
585 }
586 
587 /*
588  * Allocate an extent of length minlen<=len<=maxlen, with no position
589  * specified.  If we don't get maxlen then use prod to trim
590  * the length, if given.  The lengths are all in rtextents.
591  */
592 STATIC int
593 xfs_rtallocate_extent_size(
594 	struct xfs_rtalloc_args	*args,
595 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
596 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
597 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
598 	xfs_rtxlen_t		prod,	/* extent product factor */
599 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
600 {
601 	int			error;
602 	int			l;	/* level number (loop control) */
603 
604 	ASSERT(minlen % prod == 0);
605 	ASSERT(maxlen % prod == 0);
606 	ASSERT(maxlen != 0);
607 
608 	/*
609 	 * Loop over all the levels starting with maxlen.
610 	 *
611 	 * At each level, look at all the bitmap blocks, to see if there are
612 	 * extents starting there that are long enough (>= maxlen).
613 	 *
614 	 * Note, only on the initial level can the allocation fail if the
615 	 * summary says there's an extent.
616 	 */
617 	for (l = xfs_highbit32(maxlen); l < args->mp->m_rsumlevels; l++) {
618 		error = xfs_rtalloc_sumlevel(args, l, minlen, maxlen, prod, len,
619 				rtx);
620 		if (error != -ENOSPC)
621 			return error;
622 	}
623 
624 	/*
625 	 * Didn't find any maxlen blocks.  Try smaller ones, unless we are
626 	 * looking for a fixed size extent.
627 	 */
628 	if (minlen > --maxlen)
629 		return -ENOSPC;
630 	ASSERT(minlen != 0);
631 	ASSERT(maxlen != 0);
632 
633 	/*
634 	 * Loop over sizes, from maxlen down to minlen.
635 	 *
636 	 * This time, when we do the allocations, allow smaller ones to succeed,
637 	 * but make sure the specified minlen/maxlen are in the possible range
638 	 * for this summary level.
639 	 */
640 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
641 		error = xfs_rtalloc_sumlevel(args, l,
642 				max_t(xfs_rtxlen_t, minlen, 1 << l),
643 				min_t(xfs_rtxlen_t, maxlen, (1 << (l + 1)) - 1),
644 				prod, len, rtx);
645 		if (error != -ENOSPC)
646 			return error;
647 	}
648 
649 	return -ENOSPC;
650 }
651 
652 /*
653  * Allocate space to the bitmap or summary file, and zero it, for growfs.
654  */
655 STATIC int
656 xfs_growfs_rt_alloc(
657 	struct xfs_mount	*mp,		/* file system mount point */
658 	xfs_extlen_t		oblocks,	/* old count of blocks */
659 	xfs_extlen_t		nblocks,	/* new count of blocks */
660 	struct xfs_inode	*ip)		/* inode (bitmap/summary) */
661 {
662 	xfs_fileoff_t		bno;		/* block number in file */
663 	struct xfs_buf		*bp;	/* temporary buffer for zeroing */
664 	xfs_daddr_t		d;		/* disk block address */
665 	int			error;		/* error return value */
666 	xfs_fsblock_t		fsbno;		/* filesystem block for bno */
667 	struct xfs_bmbt_irec	map;		/* block map output */
668 	int			nmap;		/* number of block maps */
669 	int			resblks;	/* space reservation */
670 	enum xfs_blft		buf_type;
671 	struct xfs_trans	*tp;
672 
673 	if (ip == mp->m_rsumip)
674 		buf_type = XFS_BLFT_RTSUMMARY_BUF;
675 	else
676 		buf_type = XFS_BLFT_RTBITMAP_BUF;
677 
678 	/*
679 	 * Allocate space to the file, as necessary.
680 	 */
681 	while (oblocks < nblocks) {
682 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
683 		/*
684 		 * Reserve space & log for one extent added to the file.
685 		 */
686 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtalloc, resblks,
687 				0, 0, &tp);
688 		if (error)
689 			return error;
690 		/*
691 		 * Lock the inode.
692 		 */
693 		xfs_ilock(ip, XFS_ILOCK_EXCL);
694 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
695 
696 		error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
697 				XFS_IEXT_ADD_NOSPLIT_CNT);
698 		if (error == -EFBIG)
699 			error = xfs_iext_count_upgrade(tp, ip,
700 					XFS_IEXT_ADD_NOSPLIT_CNT);
701 		if (error)
702 			goto out_trans_cancel;
703 
704 		/*
705 		 * Allocate blocks to the bitmap file.
706 		 */
707 		nmap = 1;
708 		error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
709 					XFS_BMAPI_METADATA, 0, &map, &nmap);
710 		if (!error && nmap < 1)
711 			error = -ENOSPC;
712 		if (error)
713 			goto out_trans_cancel;
714 		/*
715 		 * Free any blocks freed up in the transaction, then commit.
716 		 */
717 		error = xfs_trans_commit(tp);
718 		if (error)
719 			return error;
720 		/*
721 		 * Now we need to clear the allocated blocks.
722 		 * Do this one block per transaction, to keep it simple.
723 		 */
724 		for (bno = map.br_startoff, fsbno = map.br_startblock;
725 		     bno < map.br_startoff + map.br_blockcount;
726 		     bno++, fsbno++) {
727 			/*
728 			 * Reserve log for one block zeroing.
729 			 */
730 			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtzero,
731 					0, 0, 0, &tp);
732 			if (error)
733 				return error;
734 			/*
735 			 * Lock the bitmap inode.
736 			 */
737 			xfs_ilock(ip, XFS_ILOCK_EXCL);
738 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
739 			/*
740 			 * Get a buffer for the block.
741 			 */
742 			d = XFS_FSB_TO_DADDR(mp, fsbno);
743 			error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
744 					mp->m_bsize, 0, &bp);
745 			if (error)
746 				goto out_trans_cancel;
747 
748 			xfs_trans_buf_set_type(tp, bp, buf_type);
749 			bp->b_ops = &xfs_rtbuf_ops;
750 			memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
751 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
752 			/*
753 			 * Commit the transaction.
754 			 */
755 			error = xfs_trans_commit(tp);
756 			if (error)
757 				return error;
758 		}
759 		/*
760 		 * Go on to the next extent, if any.
761 		 */
762 		oblocks = map.br_startoff + map.br_blockcount;
763 	}
764 
765 	return 0;
766 
767 out_trans_cancel:
768 	xfs_trans_cancel(tp);
769 	return error;
770 }
771 
772 static void
773 xfs_alloc_rsum_cache(
774 	xfs_mount_t	*mp,		/* file system mount structure */
775 	xfs_extlen_t	rbmblocks)	/* number of rt bitmap blocks */
776 {
777 	/*
778 	 * The rsum cache is initialized to the maximum value, which is
779 	 * trivially an upper bound on the maximum level with any free extents.
780 	 * We can continue without the cache if it couldn't be allocated.
781 	 */
782 	mp->m_rsum_cache = kvmalloc(rbmblocks, GFP_KERNEL);
783 	if (mp->m_rsum_cache)
784 		memset(mp->m_rsum_cache, -1, rbmblocks);
785 	else
786 		xfs_warn(mp, "could not allocate realtime summary cache");
787 }
788 
789 /*
790  * Visible (exported) functions.
791  */
792 
793 /*
794  * Grow the realtime area of the filesystem.
795  */
796 int
797 xfs_growfs_rt(
798 	xfs_mount_t	*mp,		/* mount point for filesystem */
799 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
800 {
801 	xfs_fileoff_t	bmbno;		/* bitmap block number */
802 	struct xfs_buf	*bp;		/* temporary buffer */
803 	int		error;		/* error return value */
804 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
805 	xfs_rfsblock_t	nrblocks;	/* new number of realtime blocks */
806 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
807 	xfs_rtxnum_t	nrextents;	/* new number of realtime extents */
808 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
809 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
810 	uint		nrsumlevels;	/* new rt summary levels */
811 	uint		nrsumsize;	/* new size of rt summary, bytes */
812 	xfs_sb_t	*nsbp;		/* new superblock */
813 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
814 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
815 	xfs_sb_t	*sbp;		/* old superblock */
816 	uint8_t		*rsum_cache;	/* old summary cache */
817 
818 	sbp = &mp->m_sb;
819 
820 	if (!capable(CAP_SYS_ADMIN))
821 		return -EPERM;
822 
823 	/* Needs to have been mounted with an rt device. */
824 	if (!XFS_IS_REALTIME_MOUNT(mp))
825 		return -EINVAL;
826 	/*
827 	 * Mount should fail if the rt bitmap/summary files don't load, but
828 	 * we'll check anyway.
829 	 */
830 	if (!mp->m_rbmip || !mp->m_rsumip)
831 		return -EINVAL;
832 
833 	/* Shrink not supported. */
834 	if (in->newblocks <= sbp->sb_rblocks)
835 		return -EINVAL;
836 
837 	/* Can only change rt extent size when adding rt volume. */
838 	if (sbp->sb_rblocks > 0 && in->extsize != sbp->sb_rextsize)
839 		return -EINVAL;
840 
841 	/* Range check the extent size. */
842 	if (XFS_FSB_TO_B(mp, in->extsize) > XFS_MAX_RTEXTSIZE ||
843 	    XFS_FSB_TO_B(mp, in->extsize) < XFS_MIN_RTEXTSIZE)
844 		return -EINVAL;
845 
846 	/* Unsupported realtime features. */
847 	if (xfs_has_rmapbt(mp) || xfs_has_reflink(mp) || xfs_has_quota(mp))
848 		return -EOPNOTSUPP;
849 
850 	nrblocks = in->newblocks;
851 	error = xfs_sb_validate_fsb_count(sbp, nrblocks);
852 	if (error)
853 		return error;
854 	/*
855 	 * Read in the last block of the device, make sure it exists.
856 	 */
857 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
858 				XFS_FSB_TO_BB(mp, nrblocks - 1),
859 				XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
860 	if (error)
861 		return error;
862 	xfs_buf_relse(bp);
863 
864 	/*
865 	 * Calculate new parameters.  These are the final values to be reached.
866 	 */
867 	nrextents = nrblocks;
868 	do_div(nrextents, in->extsize);
869 	if (!xfs_validate_rtextents(nrextents))
870 		return -EINVAL;
871 	nrbmblocks = xfs_rtbitmap_blockcount(mp, nrextents);
872 	nrextslog = xfs_compute_rextslog(nrextents);
873 	nrsumlevels = nrextslog + 1;
874 	nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels, nrbmblocks);
875 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
876 	/*
877 	 * New summary size can't be more than half the size of
878 	 * the log.  This prevents us from getting a log overflow,
879 	 * since we'll log basically the whole summary file at once.
880 	 */
881 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
882 		return -EINVAL;
883 	/*
884 	 * Get the old block counts for bitmap and summary inodes.
885 	 * These can't change since other growfs callers are locked out.
886 	 */
887 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_disk_size);
888 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_disk_size);
889 	/*
890 	 * Allocate space to the bitmap and summary files, as necessary.
891 	 */
892 	error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
893 	if (error)
894 		return error;
895 	error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
896 	if (error)
897 		return error;
898 
899 	rsum_cache = mp->m_rsum_cache;
900 	if (nrbmblocks != sbp->sb_rbmblocks)
901 		xfs_alloc_rsum_cache(mp, nrbmblocks);
902 
903 	/*
904 	 * Allocate a new (fake) mount/sb.
905 	 */
906 	nmp = kmem_alloc(sizeof(*nmp), 0);
907 	/*
908 	 * Loop over the bitmap blocks.
909 	 * We will do everything one bitmap block at a time.
910 	 * Skip the current block if it is exactly full.
911 	 * This also deals with the case where there were no rtextents before.
912 	 */
913 	for (bmbno = sbp->sb_rbmblocks -
914 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
915 	     bmbno < nrbmblocks;
916 	     bmbno++) {
917 		struct xfs_rtalloc_args	args = {
918 			.mp		= mp,
919 		};
920 		struct xfs_rtalloc_args	nargs = {
921 			.mp		= nmp,
922 		};
923 		struct xfs_trans	*tp;
924 		xfs_rfsblock_t		nrblocks_step;
925 
926 		*nmp = *mp;
927 		nsbp = &nmp->m_sb;
928 		/*
929 		 * Calculate new sb and mount fields for this round.
930 		 */
931 		nsbp->sb_rextsize = in->extsize;
932 		nmp->m_rtxblklog = -1; /* don't use shift or masking */
933 		nsbp->sb_rbmblocks = bmbno + 1;
934 		nrblocks_step = (bmbno + 1) * NBBY * nsbp->sb_blocksize *
935 				nsbp->sb_rextsize;
936 		nsbp->sb_rblocks = min(nrblocks, nrblocks_step);
937 		nsbp->sb_rextents = xfs_rtb_to_rtx(nmp, nsbp->sb_rblocks);
938 		ASSERT(nsbp->sb_rextents != 0);
939 		nsbp->sb_rextslog = xfs_compute_rextslog(nsbp->sb_rextents);
940 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
941 		nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels,
942 				nsbp->sb_rbmblocks);
943 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
944 		/* recompute growfsrt reservation from new rsumsize */
945 		xfs_trans_resv_calc(nmp, &nmp->m_resv);
946 
947 		/*
948 		 * Start a transaction, get the log reservation.
949 		 */
950 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtfree, 0, 0, 0,
951 				&tp);
952 		if (error)
953 			break;
954 		args.tp = tp;
955 		nargs.tp = tp;
956 
957 		/*
958 		 * Lock out other callers by grabbing the bitmap inode lock.
959 		 */
960 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL | XFS_ILOCK_RTBITMAP);
961 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
962 		/*
963 		 * Update the bitmap inode's size ondisk and incore.  We need
964 		 * to update the incore size so that inode inactivation won't
965 		 * punch what it thinks are "posteof" blocks.
966 		 */
967 		mp->m_rbmip->i_disk_size =
968 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
969 		i_size_write(VFS_I(mp->m_rbmip), mp->m_rbmip->i_disk_size);
970 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
971 		/*
972 		 * Get the summary inode into the transaction.
973 		 */
974 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL | XFS_ILOCK_RTSUM);
975 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
976 		/*
977 		 * Update the summary inode's size.  We need to update the
978 		 * incore size so that inode inactivation won't punch what it
979 		 * thinks are "posteof" blocks.
980 		 */
981 		mp->m_rsumip->i_disk_size = nmp->m_rsumsize;
982 		i_size_write(VFS_I(mp->m_rsumip), mp->m_rsumip->i_disk_size);
983 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
984 		/*
985 		 * Copy summary data from old to new sizes.
986 		 * Do this when the real size (not block-aligned) changes.
987 		 */
988 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
989 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
990 			error = xfs_rtcopy_summary(&args, &nargs);
991 			if (error)
992 				goto error_cancel;
993 		}
994 		/*
995 		 * Update superblock fields.
996 		 */
997 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
998 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
999 				nsbp->sb_rextsize - sbp->sb_rextsize);
1000 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
1001 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
1002 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
1003 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
1004 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
1005 				nsbp->sb_rblocks - sbp->sb_rblocks);
1006 		if (nsbp->sb_rextents != sbp->sb_rextents)
1007 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
1008 				nsbp->sb_rextents - sbp->sb_rextents);
1009 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
1010 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
1011 				nsbp->sb_rextslog - sbp->sb_rextslog);
1012 		/*
1013 		 * Free new extent.
1014 		 */
1015 		error = xfs_rtfree_range(&nargs, sbp->sb_rextents,
1016 				nsbp->sb_rextents - sbp->sb_rextents);
1017 		xfs_rtbuf_cache_relse(&nargs);
1018 		if (error) {
1019 error_cancel:
1020 			xfs_trans_cancel(tp);
1021 			break;
1022 		}
1023 		/*
1024 		 * Mark more blocks free in the superblock.
1025 		 */
1026 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
1027 			nsbp->sb_rextents - sbp->sb_rextents);
1028 		/*
1029 		 * Update mp values into the real mp structure.
1030 		 */
1031 		mp->m_rsumlevels = nrsumlevels;
1032 		mp->m_rsumsize = nrsumsize;
1033 		/* recompute growfsrt reservation from new rsumsize */
1034 		xfs_trans_resv_calc(mp, &mp->m_resv);
1035 
1036 		error = xfs_trans_commit(tp);
1037 		if (error)
1038 			break;
1039 
1040 		/* Ensure the mount RT feature flag is now set. */
1041 		mp->m_features |= XFS_FEAT_REALTIME;
1042 	}
1043 	if (error)
1044 		goto out_free;
1045 
1046 	/* Update secondary superblocks now the physical grow has completed */
1047 	error = xfs_update_secondary_sbs(mp);
1048 
1049 out_free:
1050 	/*
1051 	 * Free the fake mp structure.
1052 	 */
1053 	kmem_free(nmp);
1054 
1055 	/*
1056 	 * If we had to allocate a new rsum_cache, we either need to free the
1057 	 * old one (if we succeeded) or free the new one and restore the old one
1058 	 * (if there was an error).
1059 	 */
1060 	if (rsum_cache != mp->m_rsum_cache) {
1061 		if (error) {
1062 			kmem_free(mp->m_rsum_cache);
1063 			mp->m_rsum_cache = rsum_cache;
1064 		} else {
1065 			kmem_free(rsum_cache);
1066 		}
1067 	}
1068 
1069 	return error;
1070 }
1071 
1072 /*
1073  * Initialize realtime fields in the mount structure.
1074  */
1075 int				/* error */
1076 xfs_rtmount_init(
1077 	struct xfs_mount	*mp)	/* file system mount structure */
1078 {
1079 	struct xfs_buf		*bp;	/* buffer for last block of subvolume */
1080 	struct xfs_sb		*sbp;	/* filesystem superblock copy in mount */
1081 	xfs_daddr_t		d;	/* address of last block of subvolume */
1082 	unsigned int		rsumblocks;
1083 	int			error;
1084 
1085 	sbp = &mp->m_sb;
1086 	if (sbp->sb_rblocks == 0)
1087 		return 0;
1088 	if (mp->m_rtdev_targp == NULL) {
1089 		xfs_warn(mp,
1090 	"Filesystem has a realtime volume, use rtdev=device option");
1091 		return -ENODEV;
1092 	}
1093 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
1094 	rsumblocks = xfs_rtsummary_blockcount(mp, mp->m_rsumlevels,
1095 			mp->m_sb.sb_rbmblocks);
1096 	mp->m_rsumsize = XFS_FSB_TO_B(mp, rsumblocks);
1097 	mp->m_rbmip = mp->m_rsumip = NULL;
1098 	/*
1099 	 * Check that the realtime section is an ok size.
1100 	 */
1101 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
1102 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
1103 		xfs_warn(mp, "realtime mount -- %llu != %llu",
1104 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
1105 			(unsigned long long) mp->m_sb.sb_rblocks);
1106 		return -EFBIG;
1107 	}
1108 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
1109 					d - XFS_FSB_TO_BB(mp, 1),
1110 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
1111 	if (error) {
1112 		xfs_warn(mp, "realtime device size check failed");
1113 		return error;
1114 	}
1115 	xfs_buf_relse(bp);
1116 	return 0;
1117 }
1118 
1119 static int
1120 xfs_rtalloc_count_frextent(
1121 	struct xfs_mount		*mp,
1122 	struct xfs_trans		*tp,
1123 	const struct xfs_rtalloc_rec	*rec,
1124 	void				*priv)
1125 {
1126 	uint64_t			*valp = priv;
1127 
1128 	*valp += rec->ar_extcount;
1129 	return 0;
1130 }
1131 
1132 /*
1133  * Reinitialize the number of free realtime extents from the realtime bitmap.
1134  * Callers must ensure that there is no other activity in the filesystem.
1135  */
1136 int
1137 xfs_rtalloc_reinit_frextents(
1138 	struct xfs_mount	*mp)
1139 {
1140 	uint64_t		val = 0;
1141 	int			error;
1142 
1143 	xfs_ilock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
1144 	error = xfs_rtalloc_query_all(mp, NULL, xfs_rtalloc_count_frextent,
1145 			&val);
1146 	xfs_iunlock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
1147 	if (error)
1148 		return error;
1149 
1150 	spin_lock(&mp->m_sb_lock);
1151 	mp->m_sb.sb_frextents = val;
1152 	spin_unlock(&mp->m_sb_lock);
1153 	percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1154 	return 0;
1155 }
1156 
1157 /*
1158  * Read in the bmbt of an rt metadata inode so that we never have to load them
1159  * at runtime.  This enables the use of shared ILOCKs for rtbitmap scans.  Use
1160  * an empty transaction to avoid deadlocking on loops in the bmbt.
1161  */
1162 static inline int
1163 xfs_rtmount_iread_extents(
1164 	struct xfs_inode	*ip,
1165 	unsigned int		lock_class)
1166 {
1167 	struct xfs_trans	*tp;
1168 	int			error;
1169 
1170 	error = xfs_trans_alloc_empty(ip->i_mount, &tp);
1171 	if (error)
1172 		return error;
1173 
1174 	xfs_ilock(ip, XFS_ILOCK_EXCL | lock_class);
1175 
1176 	error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1177 	if (error)
1178 		goto out_unlock;
1179 
1180 	if (xfs_inode_has_attr_fork(ip)) {
1181 		error = xfs_iread_extents(tp, ip, XFS_ATTR_FORK);
1182 		if (error)
1183 			goto out_unlock;
1184 	}
1185 
1186 out_unlock:
1187 	xfs_iunlock(ip, XFS_ILOCK_EXCL | lock_class);
1188 	xfs_trans_cancel(tp);
1189 	return error;
1190 }
1191 
1192 /*
1193  * Get the bitmap and summary inodes and the summary cache into the mount
1194  * structure at mount time.
1195  */
1196 int					/* error */
1197 xfs_rtmount_inodes(
1198 	xfs_mount_t	*mp)		/* file system mount structure */
1199 {
1200 	int		error;		/* error return value */
1201 	xfs_sb_t	*sbp;
1202 
1203 	sbp = &mp->m_sb;
1204 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
1205 	if (error)
1206 		return error;
1207 	ASSERT(mp->m_rbmip != NULL);
1208 
1209 	error = xfs_rtmount_iread_extents(mp->m_rbmip, XFS_ILOCK_RTBITMAP);
1210 	if (error)
1211 		goto out_rele_bitmap;
1212 
1213 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
1214 	if (error)
1215 		goto out_rele_bitmap;
1216 	ASSERT(mp->m_rsumip != NULL);
1217 
1218 	error = xfs_rtmount_iread_extents(mp->m_rsumip, XFS_ILOCK_RTSUM);
1219 	if (error)
1220 		goto out_rele_summary;
1221 
1222 	xfs_alloc_rsum_cache(mp, sbp->sb_rbmblocks);
1223 	return 0;
1224 
1225 out_rele_summary:
1226 	xfs_irele(mp->m_rsumip);
1227 out_rele_bitmap:
1228 	xfs_irele(mp->m_rbmip);
1229 	return error;
1230 }
1231 
1232 void
1233 xfs_rtunmount_inodes(
1234 	struct xfs_mount	*mp)
1235 {
1236 	kmem_free(mp->m_rsum_cache);
1237 	if (mp->m_rbmip)
1238 		xfs_irele(mp->m_rbmip);
1239 	if (mp->m_rsumip)
1240 		xfs_irele(mp->m_rsumip);
1241 }
1242 
1243 /*
1244  * Pick an extent for allocation at the start of a new realtime file.
1245  * Use the sequence number stored in the atime field of the bitmap inode.
1246  * Translate this to a fraction of the rtextents, and return the product
1247  * of rtextents and the fraction.
1248  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
1249  */
1250 static int
1251 xfs_rtpick_extent(
1252 	xfs_mount_t		*mp,		/* file system mount point */
1253 	xfs_trans_t		*tp,		/* transaction pointer */
1254 	xfs_rtxlen_t		len,		/* allocation length (rtextents) */
1255 	xfs_rtxnum_t		*pick)		/* result rt extent */
1256 {
1257 	xfs_rtxnum_t		b;		/* result rtext */
1258 	int			log2;		/* log of sequence number */
1259 	uint64_t		resid;		/* residual after log removed */
1260 	uint64_t		seq;		/* sequence number of file creation */
1261 	struct timespec64	ts;		/* timespec in inode */
1262 
1263 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
1264 
1265 	ts = inode_get_atime(VFS_I(mp->m_rbmip));
1266 	if (!(mp->m_rbmip->i_diflags & XFS_DIFLAG_NEWRTBM)) {
1267 		mp->m_rbmip->i_diflags |= XFS_DIFLAG_NEWRTBM;
1268 		seq = 0;
1269 	} else {
1270 		seq = ts.tv_sec;
1271 	}
1272 	if ((log2 = xfs_highbit64(seq)) == -1)
1273 		b = 0;
1274 	else {
1275 		resid = seq - (1ULL << log2);
1276 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
1277 		    (log2 + 1);
1278 		if (b >= mp->m_sb.sb_rextents)
1279 			div64_u64_rem(b, mp->m_sb.sb_rextents, &b);
1280 		if (b + len > mp->m_sb.sb_rextents)
1281 			b = mp->m_sb.sb_rextents - len;
1282 	}
1283 	ts.tv_sec = seq + 1;
1284 	inode_set_atime_to_ts(VFS_I(mp->m_rbmip), ts);
1285 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1286 	*pick = b;
1287 	return 0;
1288 }
1289 
1290 static void
1291 xfs_rtalloc_align_minmax(
1292 	xfs_rtxlen_t		*raminlen,
1293 	xfs_rtxlen_t		*ramaxlen,
1294 	xfs_rtxlen_t		*prod)
1295 {
1296 	xfs_rtxlen_t		newmaxlen = *ramaxlen;
1297 	xfs_rtxlen_t		newminlen = *raminlen;
1298 	xfs_rtxlen_t		slack;
1299 
1300 	slack = newmaxlen % *prod;
1301 	if (slack)
1302 		newmaxlen -= slack;
1303 	slack = newminlen % *prod;
1304 	if (slack)
1305 		newminlen += *prod - slack;
1306 
1307 	/*
1308 	 * If adjusting for extent size hint alignment produces an invalid
1309 	 * min/max len combination, go ahead without it.
1310 	 */
1311 	if (newmaxlen < newminlen) {
1312 		*prod = 1;
1313 		return;
1314 	}
1315 	*ramaxlen = newmaxlen;
1316 	*raminlen = newminlen;
1317 }
1318 
1319 int
1320 xfs_bmap_rtalloc(
1321 	struct xfs_bmalloca	*ap)
1322 {
1323 	struct xfs_mount	*mp = ap->ip->i_mount;
1324 	xfs_fileoff_t		orig_offset = ap->offset;
1325 	xfs_rtxnum_t		start;	   /* allocation hint rtextent no */
1326 	xfs_rtxnum_t		rtx;	   /* actually allocated rtextent no */
1327 	xfs_rtxlen_t		prod = 0;  /* product factor for allocators */
1328 	xfs_extlen_t		mod = 0;   /* product factor for allocators */
1329 	xfs_rtxlen_t		ralen = 0; /* realtime allocation length */
1330 	xfs_extlen_t		align;     /* minimum allocation alignment */
1331 	xfs_extlen_t		orig_length = ap->length;
1332 	xfs_extlen_t		minlen = mp->m_sb.sb_rextsize;
1333 	xfs_rtxlen_t		raminlen;
1334 	bool			rtlocked = false;
1335 	bool			ignore_locality = false;
1336 	struct xfs_rtalloc_args	args = {
1337 		.mp		= mp,
1338 		.tp		= ap->tp,
1339 	};
1340 	int			error;
1341 
1342 	align = xfs_get_extsz_hint(ap->ip);
1343 retry:
1344 	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
1345 					align, 1, ap->eof, 0,
1346 					ap->conv, &ap->offset, &ap->length);
1347 	if (error)
1348 		return error;
1349 	ASSERT(ap->length);
1350 	ASSERT(xfs_extlen_to_rtxmod(mp, ap->length) == 0);
1351 
1352 	/*
1353 	 * If we shifted the file offset downward to satisfy an extent size
1354 	 * hint, increase minlen by that amount so that the allocator won't
1355 	 * give us an allocation that's too short to cover at least one of the
1356 	 * blocks that the caller asked for.
1357 	 */
1358 	if (ap->offset != orig_offset)
1359 		minlen += orig_offset - ap->offset;
1360 
1361 	/*
1362 	 * Set ralen to be the actual requested length in rtextents.
1363 	 *
1364 	 * If the old value was close enough to XFS_BMBT_MAX_EXTLEN that
1365 	 * we rounded up to it, cut it back so it's valid again.
1366 	 * Note that if it's a really large request (bigger than
1367 	 * XFS_BMBT_MAX_EXTLEN), we don't hear about that number, and can't
1368 	 * adjust the starting point to match it.
1369 	 */
1370 	ralen = xfs_extlen_to_rtxlen(mp, min(ap->length, XFS_MAX_BMBT_EXTLEN));
1371 	raminlen = max_t(xfs_rtxlen_t, 1, xfs_extlen_to_rtxlen(mp, minlen));
1372 	ASSERT(raminlen > 0);
1373 	ASSERT(raminlen <= ralen);
1374 
1375 	/*
1376 	 * Lock out modifications to both the RT bitmap and summary inodes
1377 	 */
1378 	if (!rtlocked) {
1379 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
1380 		xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
1381 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
1382 		xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
1383 		rtlocked = true;
1384 	}
1385 
1386 	if (ignore_locality) {
1387 		start = 0;
1388 	} else if (xfs_bmap_adjacent(ap)) {
1389 		start = xfs_rtb_to_rtx(mp, ap->blkno);
1390 	} else if (ap->eof && ap->offset == 0) {
1391 		/*
1392 		 * If it's an allocation to an empty file at offset 0, pick an
1393 		 * extent that will space things out in the rt area.
1394 		 */
1395 		error = xfs_rtpick_extent(mp, ap->tp, ralen, &start);
1396 		if (error)
1397 			return error;
1398 	} else {
1399 		start = 0;
1400 	}
1401 
1402 	/*
1403 	 * Only bother calculating a real prod factor if offset & length are
1404 	 * perfectly aligned, otherwise it will just get us in trouble.
1405 	 */
1406 	div_u64_rem(ap->offset, align, &mod);
1407 	if (mod || ap->length % align) {
1408 		prod = 1;
1409 	} else {
1410 		prod = xfs_extlen_to_rtxlen(mp, align);
1411 		if (prod > 1)
1412 			xfs_rtalloc_align_minmax(&raminlen, &ralen, &prod);
1413 	}
1414 
1415 	if (start) {
1416 		error = xfs_rtallocate_extent_near(&args, start, raminlen,
1417 				ralen, &ralen, prod, &rtx);
1418 	} else {
1419 		error = xfs_rtallocate_extent_size(&args, raminlen,
1420 				ralen, &ralen, prod, &rtx);
1421 	}
1422 	xfs_rtbuf_cache_relse(&args);
1423 
1424 	if (error == -ENOSPC) {
1425 		if (align > mp->m_sb.sb_rextsize) {
1426 			/*
1427 			 * We previously enlarged the request length to try to
1428 			 * satisfy an extent size hint.  The allocator didn't
1429 			 * return anything, so reset the parameters to the
1430 			 * original values and try again without alignment
1431 			 * criteria.
1432 			 */
1433 			ap->offset = orig_offset;
1434 			ap->length = orig_length;
1435 			minlen = align = mp->m_sb.sb_rextsize;
1436 			goto retry;
1437 		}
1438 
1439 		if (!ignore_locality && start != 0) {
1440 			/*
1441 			 * If we can't allocate near a specific rt extent, try
1442 			 * again without locality criteria.
1443 			 */
1444 			ignore_locality = true;
1445 			goto retry;
1446 		}
1447 
1448 		ap->blkno = NULLFSBLOCK;
1449 		ap->length = 0;
1450 		return 0;
1451 	}
1452 	if (error)
1453 		return error;
1454 
1455 	xfs_trans_mod_sb(ap->tp, ap->wasdel ?
1456 			XFS_TRANS_SB_RES_FREXTENTS : XFS_TRANS_SB_FREXTENTS,
1457 			-(long)ralen);
1458 	ap->blkno = xfs_rtx_to_rtb(mp, rtx);
1459 	ap->length = xfs_rtxlen_to_extlen(mp, ralen);
1460 	xfs_bmap_alloc_account(ap);
1461 	return 0;
1462 }
1463