xref: /linux/fs/xfs/xfs_rtalloc.c (revision 40ccd6aa3e2e05be93394e3cd560c718dedfcc77)
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 #include "xfs_log_priv.h"
26 #include "xfs_health.h"
27 
28 /*
29  * Return whether there are any free extents in the size range given
30  * by low and high, for the bitmap block bbno.
31  */
32 STATIC int
33 xfs_rtany_summary(
34 	struct xfs_rtalloc_args	*args,
35 	int			low,	/* low log2 extent size */
36 	int			high,	/* high log2 extent size */
37 	xfs_fileoff_t		bbno,	/* bitmap block number */
38 	int			*maxlog) /* out: max log2 extent size free */
39 {
40 	struct xfs_mount	*mp = args->mp;
41 	int			error;
42 	int			log;	/* loop counter, log2 of ext. size */
43 	xfs_suminfo_t		sum;	/* summary data */
44 
45 	/* There are no extents at levels >= m_rsum_cache[bbno]. */
46 	if (mp->m_rsum_cache) {
47 		high = min(high, mp->m_rsum_cache[bbno] - 1);
48 		if (low > high) {
49 			*maxlog = -1;
50 			return 0;
51 		}
52 	}
53 
54 	/*
55 	 * Loop over logs of extent sizes.
56 	 */
57 	for (log = high; log >= low; log--) {
58 		/*
59 		 * Get one summary datum.
60 		 */
61 		error = xfs_rtget_summary(args, log, bbno, &sum);
62 		if (error) {
63 			return error;
64 		}
65 		/*
66 		 * If there are any, return success.
67 		 */
68 		if (sum) {
69 			*maxlog = log;
70 			goto out;
71 		}
72 	}
73 	/*
74 	 * Found nothing, return failure.
75 	 */
76 	*maxlog = -1;
77 out:
78 	/* There were no extents at levels > log. */
79 	if (mp->m_rsum_cache && log + 1 < mp->m_rsum_cache[bbno])
80 		mp->m_rsum_cache[bbno] = log + 1;
81 	return 0;
82 }
83 
84 
85 /*
86  * Copy and transform the summary file, given the old and new
87  * parameters in the mount structures.
88  */
89 STATIC int
90 xfs_rtcopy_summary(
91 	struct xfs_rtalloc_args	*oargs,
92 	struct xfs_rtalloc_args	*nargs)
93 {
94 	xfs_fileoff_t		bbno;	/* bitmap block number */
95 	int			error;
96 	int			log;	/* summary level number (log length) */
97 	xfs_suminfo_t		sum;	/* summary data */
98 
99 	for (log = oargs->mp->m_rsumlevels - 1; log >= 0; log--) {
100 		for (bbno = oargs->mp->m_sb.sb_rbmblocks - 1;
101 		     (xfs_srtblock_t)bbno >= 0;
102 		     bbno--) {
103 			error = xfs_rtget_summary(oargs, log, bbno, &sum);
104 			if (error)
105 				goto out;
106 			if (sum == 0)
107 				continue;
108 			error = xfs_rtmodify_summary(oargs, log, bbno, -sum);
109 			if (error)
110 				goto out;
111 			error = xfs_rtmodify_summary(nargs, log, bbno, sum);
112 			if (error)
113 				goto out;
114 			ASSERT(sum > 0);
115 		}
116 	}
117 	error = 0;
118 out:
119 	xfs_rtbuf_cache_relse(oargs);
120 	return 0;
121 }
122 /*
123  * Mark an extent specified by start and len allocated.
124  * Updates all the summary information as well as the bitmap.
125  */
126 STATIC int
127 xfs_rtallocate_range(
128 	struct xfs_rtalloc_args	*args,
129 	xfs_rtxnum_t		start,	/* start rtext to allocate */
130 	xfs_rtxlen_t		len)	/* in/out: summary block number */
131 {
132 	struct xfs_mount	*mp = args->mp;
133 	xfs_rtxnum_t		end;	/* end of the allocated rtext */
134 	int			error;
135 	xfs_rtxnum_t		postblock = 0; /* first rtext allocated > end */
136 	xfs_rtxnum_t		preblock = 0; /* first rtext allocated < start */
137 
138 	end = start + len - 1;
139 	/*
140 	 * Assume we're allocating out of the middle of a free extent.
141 	 * We need to find the beginning and end of the extent so we can
142 	 * properly update the summary.
143 	 */
144 	error = xfs_rtfind_back(args, start, 0, &preblock);
145 	if (error)
146 		return error;
147 
148 	/*
149 	 * Find the next allocated block (end of free extent).
150 	 */
151 	error = xfs_rtfind_forw(args, end, mp->m_sb.sb_rextents - 1,
152 			&postblock);
153 	if (error)
154 		return error;
155 
156 	/*
157 	 * Decrement the summary information corresponding to the entire
158 	 * (old) free extent.
159 	 */
160 	error = xfs_rtmodify_summary(args,
161 			xfs_highbit64(postblock + 1 - preblock),
162 			xfs_rtx_to_rbmblock(mp, preblock), -1);
163 	if (error)
164 		return error;
165 
166 	/*
167 	 * If there are blocks not being allocated at the front of the
168 	 * old extent, add summary data for them to be free.
169 	 */
170 	if (preblock < start) {
171 		error = xfs_rtmodify_summary(args,
172 				xfs_highbit64(start - preblock),
173 				xfs_rtx_to_rbmblock(mp, preblock), 1);
174 		if (error)
175 			return error;
176 	}
177 
178 	/*
179 	 * If there are blocks not being allocated at the end of the
180 	 * old extent, add summary data for them to be free.
181 	 */
182 	if (postblock > end) {
183 		error = xfs_rtmodify_summary(args,
184 				xfs_highbit64(postblock - end),
185 				xfs_rtx_to_rbmblock(mp, end + 1), 1);
186 		if (error)
187 			return error;
188 	}
189 
190 	/*
191 	 * Modify the bitmap to mark this extent allocated.
192 	 */
193 	return xfs_rtmodify_range(args, start, len, 0);
194 }
195 
196 /*
197  * Make sure we don't run off the end of the rt volume.  Be careful that
198  * adjusting maxlen downwards doesn't cause us to fail the alignment checks.
199  */
200 static inline xfs_rtxlen_t
201 xfs_rtallocate_clamp_len(
202 	struct xfs_mount	*mp,
203 	xfs_rtxnum_t		startrtx,
204 	xfs_rtxlen_t		rtxlen,
205 	xfs_rtxlen_t		prod)
206 {
207 	xfs_rtxlen_t		ret;
208 
209 	ret = min(mp->m_sb.sb_rextents, startrtx + rtxlen) - startrtx;
210 	return rounddown(ret, prod);
211 }
212 
213 /*
214  * Attempt to allocate an extent minlen<=len<=maxlen starting from
215  * bitmap block bbno.  If we don't get maxlen then use prod to trim
216  * the length, if given.  Returns error; returns starting block in *rtx.
217  * The lengths are all in rtextents.
218  */
219 STATIC int
220 xfs_rtallocate_extent_block(
221 	struct xfs_rtalloc_args	*args,
222 	xfs_fileoff_t		bbno,	/* bitmap block number */
223 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
224 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
225 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
226 	xfs_rtxnum_t		*nextp,	/* out: next rtext to try */
227 	xfs_rtxlen_t		prod,	/* extent product factor */
228 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
229 {
230 	struct xfs_mount	*mp = args->mp;
231 	xfs_rtxnum_t		besti;	/* best rtext found so far */
232 	xfs_rtxnum_t		bestlen;/* best length found so far */
233 	xfs_rtxnum_t		end;	/* last rtext in chunk */
234 	int			error;
235 	xfs_rtxnum_t		i;	/* current rtext trying */
236 	xfs_rtxnum_t		next;	/* next rtext to try */
237 	int			stat;	/* status from internal calls */
238 
239 	/*
240 	 * Loop over all the extents starting in this bitmap block,
241 	 * looking for one that's long enough.
242 	 */
243 	for (i = xfs_rbmblock_to_rtx(mp, bbno), besti = -1, bestlen = 0,
244 		end = xfs_rbmblock_to_rtx(mp, bbno + 1) - 1;
245 	     i <= end;
246 	     i++) {
247 		/* Make sure we don't scan off the end of the rt volume. */
248 		maxlen = xfs_rtallocate_clamp_len(mp, i, maxlen, prod);
249 
250 		/*
251 		 * See if there's a free extent of maxlen starting at i.
252 		 * If it's not so then next will contain the first non-free.
253 		 */
254 		error = xfs_rtcheck_range(args, i, maxlen, 1, &next, &stat);
255 		if (error)
256 			return error;
257 		if (stat) {
258 			/*
259 			 * i for maxlen is all free, allocate and return that.
260 			 */
261 			bestlen = maxlen;
262 			besti = i;
263 			goto allocate;
264 		}
265 
266 		/*
267 		 * In the case where we have a variable-sized allocation
268 		 * request, figure out how big this free piece is,
269 		 * and if it's big enough for the minimum, and the best
270 		 * so far, remember it.
271 		 */
272 		if (minlen < maxlen) {
273 			xfs_rtxnum_t	thislen;	/* this extent size */
274 
275 			thislen = next - i;
276 			if (thislen >= minlen && thislen > bestlen) {
277 				besti = i;
278 				bestlen = thislen;
279 			}
280 		}
281 		/*
282 		 * If not done yet, find the start of the next free space.
283 		 */
284 		if (next >= end)
285 			break;
286 		error = xfs_rtfind_forw(args, next, end, &i);
287 		if (error)
288 			return error;
289 	}
290 
291 	/*
292 	 * Searched the whole thing & didn't find a maxlen free extent.
293 	 */
294 	if (minlen > maxlen || besti == -1) {
295 		/*
296 		 * Allocation failed.  Set *nextp to the next block to try.
297 		 */
298 		*nextp = next;
299 		return -ENOSPC;
300 	}
301 
302 	/*
303 	 * If size should be a multiple of prod, make that so.
304 	 */
305 	if (prod > 1) {
306 		xfs_rtxlen_t	p;	/* amount to trim length by */
307 
308 		div_u64_rem(bestlen, prod, &p);
309 		if (p)
310 			bestlen -= p;
311 	}
312 
313 	/*
314 	 * Allocate besti for bestlen & return that.
315 	 */
316 allocate:
317 	error = xfs_rtallocate_range(args, besti, bestlen);
318 	if (error)
319 		return error;
320 	*len = bestlen;
321 	*rtx = besti;
322 	return 0;
323 }
324 
325 /*
326  * Allocate an extent of length minlen<=len<=maxlen, starting at block
327  * bno.  If we don't get maxlen then use prod to trim the length, if given.
328  * Returns error; returns starting block in *rtx.
329  * The lengths are all in rtextents.
330  */
331 STATIC int
332 xfs_rtallocate_extent_exact(
333 	struct xfs_rtalloc_args	*args,
334 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
335 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
336 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
337 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
338 	xfs_rtxlen_t		prod,	/* extent product factor */
339 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
340 {
341 	int			error;
342 	xfs_rtxlen_t		i;	/* extent length trimmed due to prod */
343 	int			isfree;	/* extent is free */
344 	xfs_rtxnum_t		next;	/* next rtext to try (dummy) */
345 
346 	ASSERT(minlen % prod == 0);
347 	ASSERT(maxlen % prod == 0);
348 	/*
349 	 * Check if the range in question (for maxlen) is free.
350 	 */
351 	error = xfs_rtcheck_range(args, start, maxlen, 1, &next, &isfree);
352 	if (error)
353 		return error;
354 
355 	if (!isfree) {
356 		/*
357 		 * If not, allocate what there is, if it's at least minlen.
358 		 */
359 		maxlen = next - start;
360 		if (maxlen < minlen)
361 			return -ENOSPC;
362 
363 		/*
364 		 * Trim off tail of extent, if prod is specified.
365 		 */
366 		if (prod > 1 && (i = maxlen % prod)) {
367 			maxlen -= i;
368 			if (maxlen < minlen)
369 				return -ENOSPC;
370 		}
371 	}
372 
373 	/*
374 	 * Allocate what we can and return it.
375 	 */
376 	error = xfs_rtallocate_range(args, start, maxlen);
377 	if (error)
378 		return error;
379 	*len = maxlen;
380 	*rtx = start;
381 	return 0;
382 }
383 
384 /*
385  * Allocate an extent of length minlen<=len<=maxlen, starting as near
386  * to start as possible.  If we don't get maxlen then use prod to trim
387  * the length, if given.  The lengths are all in rtextents.
388  */
389 STATIC int
390 xfs_rtallocate_extent_near(
391 	struct xfs_rtalloc_args	*args,
392 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
393 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
394 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
395 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
396 	xfs_rtxlen_t		prod,	/* extent product factor */
397 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
398 {
399 	struct xfs_mount	*mp = args->mp;
400 	int			maxlog;	/* max useful extent from summary */
401 	xfs_fileoff_t		bbno;	/* bitmap block number */
402 	int			error;
403 	int			i;	/* bitmap block offset (loop control) */
404 	int			j;	/* secondary loop control */
405 	int			log2len; /* log2 of minlen */
406 	xfs_rtxnum_t		n;	/* next rtext to try */
407 
408 	ASSERT(minlen % prod == 0);
409 	ASSERT(maxlen % prod == 0);
410 
411 	/*
412 	 * If the block number given is off the end, silently set it to
413 	 * the last block.
414 	 */
415 	if (start >= mp->m_sb.sb_rextents)
416 		start = mp->m_sb.sb_rextents - 1;
417 
418 	/* Make sure we don't run off the end of the rt volume. */
419 	maxlen = xfs_rtallocate_clamp_len(mp, start, maxlen, prod);
420 	if (maxlen < minlen)
421 		return -ENOSPC;
422 
423 	/*
424 	 * Try the exact allocation first.
425 	 */
426 	error = xfs_rtallocate_extent_exact(args, start, minlen, maxlen, len,
427 			prod, rtx);
428 	if (error != -ENOSPC)
429 		return error;
430 
431 
432 	bbno = xfs_rtx_to_rbmblock(mp, start);
433 	i = 0;
434 	j = -1;
435 	ASSERT(minlen != 0);
436 	log2len = xfs_highbit32(minlen);
437 	/*
438 	 * Loop over all bitmap blocks (bbno + i is current block).
439 	 */
440 	for (;;) {
441 		/*
442 		 * Get summary information of extents of all useful levels
443 		 * starting in this bitmap block.
444 		 */
445 		error = xfs_rtany_summary(args, log2len, mp->m_rsumlevels - 1,
446 				bbno + i, &maxlog);
447 		if (error)
448 			return error;
449 
450 		/*
451 		 * If there are any useful extents starting here, try
452 		 * allocating one.
453 		 */
454 		if (maxlog >= 0) {
455 			xfs_extlen_t maxavail =
456 				min_t(xfs_rtblock_t, maxlen,
457 				      (1ULL << (maxlog + 1)) - 1);
458 			/*
459 			 * On the positive side of the starting location.
460 			 */
461 			if (i >= 0) {
462 				/*
463 				 * Try to allocate an extent starting in
464 				 * this block.
465 				 */
466 				error = xfs_rtallocate_extent_block(args,
467 						bbno + i, minlen, maxavail, len,
468 						&n, prod, rtx);
469 				if (error != -ENOSPC)
470 					return error;
471 			}
472 			/*
473 			 * On the negative side of the starting location.
474 			 */
475 			else {		/* i < 0 */
476 				int	maxblocks;
477 
478 				/*
479 				 * Loop backwards to find the end of the extent
480 				 * we found in the realtime summary.
481 				 *
482 				 * maxblocks is the maximum possible number of
483 				 * bitmap blocks from the start of the extent
484 				 * to the end of the extent.
485 				 */
486 				if (maxlog == 0)
487 					maxblocks = 0;
488 				else if (maxlog < mp->m_blkbit_log)
489 					maxblocks = 1;
490 				else
491 					maxblocks = 2 << (maxlog - mp->m_blkbit_log);
492 
493 				/*
494 				 * We need to check bbno + i + maxblocks down to
495 				 * bbno + i. We already checked bbno down to
496 				 * bbno + j + 1, so we don't need to check those
497 				 * again.
498 				 */
499 				j = min(i + maxblocks, j);
500 				for (; j >= i; j--) {
501 					error = xfs_rtallocate_extent_block(args,
502 							bbno + j, minlen,
503 							maxavail, len, &n, prod,
504 							rtx);
505 					if (error != -ENOSPC)
506 						return error;
507 				}
508 			}
509 		}
510 		/*
511 		 * Loop control.  If we were on the positive side, and there's
512 		 * still more blocks on the negative side, go there.
513 		 */
514 		if (i > 0 && (int)bbno - i >= 0)
515 			i = -i;
516 		/*
517 		 * If positive, and no more negative, but there are more
518 		 * positive, go there.
519 		 */
520 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
521 			i++;
522 		/*
523 		 * If negative or 0 (just started), and there are positive
524 		 * blocks to go, go there.  The 0 case moves to block 1.
525 		 */
526 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
527 			i = 1 - i;
528 		/*
529 		 * If negative or 0 and there are more negative blocks,
530 		 * go there.
531 		 */
532 		else if (i <= 0 && (int)bbno + i > 0)
533 			i--;
534 		/*
535 		 * Must be done.  Return failure.
536 		 */
537 		else
538 			break;
539 	}
540 	return -ENOSPC;
541 }
542 
543 static int
544 xfs_rtalloc_sumlevel(
545 	struct xfs_rtalloc_args	*args,
546 	int			l,	/* level number */
547 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
548 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
549 	xfs_rtxlen_t		prod,	/* extent product factor */
550 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
551 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
552 {
553 	xfs_fileoff_t		i;	/* bitmap block number */
554 
555 	for (i = 0; i < args->mp->m_sb.sb_rbmblocks; i++) {
556 		xfs_suminfo_t	sum;	/* summary information for extents */
557 		xfs_rtxnum_t	n;	/* next rtext to be tried */
558 		int		error;
559 
560 		error = xfs_rtget_summary(args, l, i, &sum);
561 		if (error)
562 			return error;
563 
564 		/*
565 		 * Nothing there, on to the next block.
566 		 */
567 		if (!sum)
568 			continue;
569 
570 		/*
571 		 * Try allocating the extent.
572 		 */
573 		error = xfs_rtallocate_extent_block(args, i, minlen, maxlen,
574 				len, &n, prod, rtx);
575 		if (error != -ENOSPC)
576 			return error;
577 
578 		/*
579 		 * If the "next block to try" returned from the allocator is
580 		 * beyond the next bitmap block, skip to that bitmap block.
581 		 */
582 		if (xfs_rtx_to_rbmblock(args->mp, n) > i + 1)
583 			i = xfs_rtx_to_rbmblock(args->mp, n) - 1;
584 	}
585 
586 	return -ENOSPC;
587 }
588 
589 /*
590  * Allocate an extent of length minlen<=len<=maxlen, with no position
591  * specified.  If we don't get maxlen then use prod to trim
592  * the length, if given.  The lengths are all in rtextents.
593  */
594 STATIC int
595 xfs_rtallocate_extent_size(
596 	struct xfs_rtalloc_args	*args,
597 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
598 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
599 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
600 	xfs_rtxlen_t		prod,	/* extent product factor */
601 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
602 {
603 	int			error;
604 	int			l;	/* level number (loop control) */
605 
606 	ASSERT(minlen % prod == 0);
607 	ASSERT(maxlen % prod == 0);
608 	ASSERT(maxlen != 0);
609 
610 	/*
611 	 * Loop over all the levels starting with maxlen.
612 	 *
613 	 * At each level, look at all the bitmap blocks, to see if there are
614 	 * extents starting there that are long enough (>= maxlen).
615 	 *
616 	 * Note, only on the initial level can the allocation fail if the
617 	 * summary says there's an extent.
618 	 */
619 	for (l = xfs_highbit32(maxlen); l < args->mp->m_rsumlevels; l++) {
620 		error = xfs_rtalloc_sumlevel(args, l, minlen, maxlen, prod, len,
621 				rtx);
622 		if (error != -ENOSPC)
623 			return error;
624 	}
625 
626 	/*
627 	 * Didn't find any maxlen blocks.  Try smaller ones, unless we are
628 	 * looking for a fixed size extent.
629 	 */
630 	if (minlen > --maxlen)
631 		return -ENOSPC;
632 	ASSERT(minlen != 0);
633 	ASSERT(maxlen != 0);
634 
635 	/*
636 	 * Loop over sizes, from maxlen down to minlen.
637 	 *
638 	 * This time, when we do the allocations, allow smaller ones to succeed,
639 	 * but make sure the specified minlen/maxlen are in the possible range
640 	 * for this summary level.
641 	 */
642 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
643 		error = xfs_rtalloc_sumlevel(args, l,
644 				max_t(xfs_rtxlen_t, minlen, 1 << l),
645 				min_t(xfs_rtxlen_t, maxlen, (1 << (l + 1)) - 1),
646 				prod, len, rtx);
647 		if (error != -ENOSPC)
648 			return error;
649 	}
650 
651 	return -ENOSPC;
652 }
653 
654 /*
655  * Allocate space to the bitmap or summary file, and zero it, for growfs.
656  */
657 STATIC int
658 xfs_growfs_rt_alloc(
659 	struct xfs_mount	*mp,		/* file system mount point */
660 	xfs_extlen_t		oblocks,	/* old count of blocks */
661 	xfs_extlen_t		nblocks,	/* new count of blocks */
662 	struct xfs_inode	*ip)		/* inode (bitmap/summary) */
663 {
664 	xfs_fileoff_t		bno;		/* block number in file */
665 	struct xfs_buf		*bp;	/* temporary buffer for zeroing */
666 	xfs_daddr_t		d;		/* disk block address */
667 	int			error;		/* error return value */
668 	xfs_fsblock_t		fsbno;		/* filesystem block for bno */
669 	struct xfs_bmbt_irec	map;		/* block map output */
670 	int			nmap;		/* number of block maps */
671 	int			resblks;	/* space reservation */
672 	enum xfs_blft		buf_type;
673 	struct xfs_trans	*tp;
674 
675 	if (ip == mp->m_rsumip)
676 		buf_type = XFS_BLFT_RTSUMMARY_BUF;
677 	else
678 		buf_type = XFS_BLFT_RTBITMAP_BUF;
679 
680 	/*
681 	 * Allocate space to the file, as necessary.
682 	 */
683 	while (oblocks < nblocks) {
684 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
685 		/*
686 		 * Reserve space & log for one extent added to the file.
687 		 */
688 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtalloc, resblks,
689 				0, 0, &tp);
690 		if (error)
691 			return error;
692 		/*
693 		 * Lock the inode.
694 		 */
695 		xfs_ilock(ip, XFS_ILOCK_EXCL);
696 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
697 
698 		error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
699 				XFS_IEXT_ADD_NOSPLIT_CNT);
700 		if (error)
701 			goto out_trans_cancel;
702 
703 		/*
704 		 * Allocate blocks to the bitmap file.
705 		 */
706 		nmap = 1;
707 		error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
708 					XFS_BMAPI_METADATA, 0, &map, &nmap);
709 		if (error)
710 			goto out_trans_cancel;
711 		/*
712 		 * Free any blocks freed up in the transaction, then commit.
713 		 */
714 		error = xfs_trans_commit(tp);
715 		if (error)
716 			return error;
717 		/*
718 		 * Now we need to clear the allocated blocks.
719 		 * Do this one block per transaction, to keep it simple.
720 		 */
721 		for (bno = map.br_startoff, fsbno = map.br_startblock;
722 		     bno < map.br_startoff + map.br_blockcount;
723 		     bno++, fsbno++) {
724 			/*
725 			 * Reserve log for one block zeroing.
726 			 */
727 			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtzero,
728 					0, 0, 0, &tp);
729 			if (error)
730 				return error;
731 			/*
732 			 * Lock the bitmap inode.
733 			 */
734 			xfs_ilock(ip, XFS_ILOCK_EXCL);
735 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
736 			/*
737 			 * Get a buffer for the block.
738 			 */
739 			d = XFS_FSB_TO_DADDR(mp, fsbno);
740 			error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
741 					mp->m_bsize, 0, &bp);
742 			if (error)
743 				goto out_trans_cancel;
744 
745 			xfs_trans_buf_set_type(tp, bp, buf_type);
746 			bp->b_ops = &xfs_rtbuf_ops;
747 			memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
748 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
749 			/*
750 			 * Commit the transaction.
751 			 */
752 			error = xfs_trans_commit(tp);
753 			if (error)
754 				return error;
755 		}
756 		/*
757 		 * Go on to the next extent, if any.
758 		 */
759 		oblocks = map.br_startoff + map.br_blockcount;
760 	}
761 
762 	return 0;
763 
764 out_trans_cancel:
765 	xfs_trans_cancel(tp);
766 	return error;
767 }
768 
769 static void
770 xfs_alloc_rsum_cache(
771 	xfs_mount_t	*mp,		/* file system mount structure */
772 	xfs_extlen_t	rbmblocks)	/* number of rt bitmap blocks */
773 {
774 	/*
775 	 * The rsum cache is initialized to the maximum value, which is
776 	 * trivially an upper bound on the maximum level with any free extents.
777 	 * We can continue without the cache if it couldn't be allocated.
778 	 */
779 	mp->m_rsum_cache = kvmalloc(rbmblocks, GFP_KERNEL);
780 	if (mp->m_rsum_cache)
781 		memset(mp->m_rsum_cache, -1, rbmblocks);
782 	else
783 		xfs_warn(mp, "could not allocate realtime summary cache");
784 }
785 
786 /*
787  * Visible (exported) functions.
788  */
789 
790 /*
791  * Grow the realtime area of the filesystem.
792  */
793 int
794 xfs_growfs_rt(
795 	xfs_mount_t	*mp,		/* mount point for filesystem */
796 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
797 {
798 	xfs_fileoff_t	bmbno;		/* bitmap block number */
799 	struct xfs_buf	*bp;		/* temporary buffer */
800 	int		error;		/* error return value */
801 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
802 	xfs_rfsblock_t	nrblocks;	/* new number of realtime blocks */
803 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
804 	xfs_rtxnum_t	nrextents;	/* new number of realtime extents */
805 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
806 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
807 	uint		nrsumlevels;	/* new rt summary levels */
808 	uint		nrsumsize;	/* new size of rt summary, bytes */
809 	xfs_sb_t	*nsbp;		/* new superblock */
810 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
811 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
812 	xfs_sb_t	*sbp;		/* old superblock */
813 	uint8_t		*rsum_cache;	/* old summary cache */
814 
815 	sbp = &mp->m_sb;
816 
817 	if (!capable(CAP_SYS_ADMIN))
818 		return -EPERM;
819 
820 	/* Needs to have been mounted with an rt device. */
821 	if (!XFS_IS_REALTIME_MOUNT(mp))
822 		return -EINVAL;
823 	/*
824 	 * Mount should fail if the rt bitmap/summary files don't load, but
825 	 * we'll check anyway.
826 	 */
827 	if (!mp->m_rbmip || !mp->m_rsumip)
828 		return -EINVAL;
829 
830 	/* Shrink not supported. */
831 	if (in->newblocks <= sbp->sb_rblocks)
832 		return -EINVAL;
833 
834 	/* Can only change rt extent size when adding rt volume. */
835 	if (sbp->sb_rblocks > 0 && in->extsize != sbp->sb_rextsize)
836 		return -EINVAL;
837 
838 	/* Range check the extent size. */
839 	if (XFS_FSB_TO_B(mp, in->extsize) > XFS_MAX_RTEXTSIZE ||
840 	    XFS_FSB_TO_B(mp, in->extsize) < XFS_MIN_RTEXTSIZE)
841 		return -EINVAL;
842 
843 	/* Unsupported realtime features. */
844 	if (xfs_has_rmapbt(mp) || xfs_has_reflink(mp) || xfs_has_quota(mp))
845 		return -EOPNOTSUPP;
846 
847 	nrblocks = in->newblocks;
848 	error = xfs_sb_validate_fsb_count(sbp, nrblocks);
849 	if (error)
850 		return error;
851 	/*
852 	 * Read in the last block of the device, make sure it exists.
853 	 */
854 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
855 				XFS_FSB_TO_BB(mp, nrblocks - 1),
856 				XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
857 	if (error)
858 		return error;
859 	xfs_buf_relse(bp);
860 
861 	/*
862 	 * Calculate new parameters.  These are the final values to be reached.
863 	 */
864 	nrextents = nrblocks;
865 	do_div(nrextents, in->extsize);
866 	if (!xfs_validate_rtextents(nrextents))
867 		return -EINVAL;
868 	nrbmblocks = xfs_rtbitmap_blockcount(mp, nrextents);
869 	nrextslog = xfs_compute_rextslog(nrextents);
870 	nrsumlevels = nrextslog + 1;
871 	nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels, nrbmblocks);
872 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
873 	/*
874 	 * New summary size can't be more than half the size of
875 	 * the log.  This prevents us from getting a log overflow,
876 	 * since we'll log basically the whole summary file at once.
877 	 */
878 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
879 		return -EINVAL;
880 	/*
881 	 * Get the old block counts for bitmap and summary inodes.
882 	 * These can't change since other growfs callers are locked out.
883 	 */
884 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_disk_size);
885 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_disk_size);
886 	/*
887 	 * Allocate space to the bitmap and summary files, as necessary.
888 	 */
889 	error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
890 	if (error)
891 		return error;
892 	error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
893 	if (error)
894 		return error;
895 
896 	rsum_cache = mp->m_rsum_cache;
897 	if (nrbmblocks != sbp->sb_rbmblocks)
898 		xfs_alloc_rsum_cache(mp, nrbmblocks);
899 
900 	/*
901 	 * Allocate a new (fake) mount/sb.
902 	 */
903 	nmp = kmalloc(sizeof(*nmp), GFP_KERNEL | __GFP_NOFAIL);
904 	/*
905 	 * Loop over the bitmap blocks.
906 	 * We will do everything one bitmap block at a time.
907 	 * Skip the current block if it is exactly full.
908 	 * This also deals with the case where there were no rtextents before.
909 	 */
910 	for (bmbno = sbp->sb_rbmblocks -
911 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
912 	     bmbno < nrbmblocks;
913 	     bmbno++) {
914 		struct xfs_rtalloc_args	args = {
915 			.mp		= mp,
916 		};
917 		struct xfs_rtalloc_args	nargs = {
918 			.mp		= nmp,
919 		};
920 		struct xfs_trans	*tp;
921 		xfs_rfsblock_t		nrblocks_step;
922 
923 		*nmp = *mp;
924 		nsbp = &nmp->m_sb;
925 		/*
926 		 * Calculate new sb and mount fields for this round.
927 		 */
928 		nsbp->sb_rextsize = in->extsize;
929 		nmp->m_rtxblklog = -1; /* don't use shift or masking */
930 		nsbp->sb_rbmblocks = bmbno + 1;
931 		nrblocks_step = (bmbno + 1) * NBBY * nsbp->sb_blocksize *
932 				nsbp->sb_rextsize;
933 		nsbp->sb_rblocks = min(nrblocks, nrblocks_step);
934 		nsbp->sb_rextents = xfs_rtb_to_rtx(nmp, nsbp->sb_rblocks);
935 		ASSERT(nsbp->sb_rextents != 0);
936 		nsbp->sb_rextslog = xfs_compute_rextslog(nsbp->sb_rextents);
937 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
938 		nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels,
939 				nsbp->sb_rbmblocks);
940 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
941 		/* recompute growfsrt reservation from new rsumsize */
942 		xfs_trans_resv_calc(nmp, &nmp->m_resv);
943 
944 		/*
945 		 * Start a transaction, get the log reservation.
946 		 */
947 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtfree, 0, 0, 0,
948 				&tp);
949 		if (error)
950 			break;
951 		args.tp = tp;
952 		nargs.tp = tp;
953 
954 		/*
955 		 * Lock out other callers by grabbing the bitmap and summary
956 		 * inode locks and joining them to the transaction.
957 		 */
958 		xfs_rtbitmap_lock(tp, mp);
959 		/*
960 		 * Update the bitmap inode's size ondisk and incore.  We need
961 		 * to update the incore size so that inode inactivation won't
962 		 * punch what it thinks are "posteof" blocks.
963 		 */
964 		mp->m_rbmip->i_disk_size =
965 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
966 		i_size_write(VFS_I(mp->m_rbmip), mp->m_rbmip->i_disk_size);
967 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
968 		/*
969 		 * Update the summary inode's size.  We need to update the
970 		 * incore size so that inode inactivation won't punch what it
971 		 * thinks are "posteof" blocks.
972 		 */
973 		mp->m_rsumip->i_disk_size = nmp->m_rsumsize;
974 		i_size_write(VFS_I(mp->m_rsumip), mp->m_rsumip->i_disk_size);
975 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
976 		/*
977 		 * Copy summary data from old to new sizes.
978 		 * Do this when the real size (not block-aligned) changes.
979 		 */
980 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
981 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
982 			error = xfs_rtcopy_summary(&args, &nargs);
983 			if (error)
984 				goto error_cancel;
985 		}
986 		/*
987 		 * Update superblock fields.
988 		 */
989 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
990 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
991 				nsbp->sb_rextsize - sbp->sb_rextsize);
992 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
993 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
994 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
995 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
996 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
997 				nsbp->sb_rblocks - sbp->sb_rblocks);
998 		if (nsbp->sb_rextents != sbp->sb_rextents)
999 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
1000 				nsbp->sb_rextents - sbp->sb_rextents);
1001 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
1002 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
1003 				nsbp->sb_rextslog - sbp->sb_rextslog);
1004 		/*
1005 		 * Free new extent.
1006 		 */
1007 		error = xfs_rtfree_range(&nargs, sbp->sb_rextents,
1008 				nsbp->sb_rextents - sbp->sb_rextents);
1009 		xfs_rtbuf_cache_relse(&nargs);
1010 		if (error) {
1011 error_cancel:
1012 			xfs_trans_cancel(tp);
1013 			break;
1014 		}
1015 		/*
1016 		 * Mark more blocks free in the superblock.
1017 		 */
1018 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
1019 			nsbp->sb_rextents - sbp->sb_rextents);
1020 		/*
1021 		 * Update mp values into the real mp structure.
1022 		 */
1023 		mp->m_rsumlevels = nrsumlevels;
1024 		mp->m_rsumsize = nrsumsize;
1025 		/* recompute growfsrt reservation from new rsumsize */
1026 		xfs_trans_resv_calc(mp, &mp->m_resv);
1027 
1028 		error = xfs_trans_commit(tp);
1029 		if (error)
1030 			break;
1031 
1032 		/* Ensure the mount RT feature flag is now set. */
1033 		mp->m_features |= XFS_FEAT_REALTIME;
1034 	}
1035 	if (error)
1036 		goto out_free;
1037 
1038 	/* Update secondary superblocks now the physical grow has completed */
1039 	error = xfs_update_secondary_sbs(mp);
1040 
1041 out_free:
1042 	/*
1043 	 * Free the fake mp structure.
1044 	 */
1045 	kfree(nmp);
1046 
1047 	/*
1048 	 * If we had to allocate a new rsum_cache, we either need to free the
1049 	 * old one (if we succeeded) or free the new one and restore the old one
1050 	 * (if there was an error).
1051 	 */
1052 	if (rsum_cache != mp->m_rsum_cache) {
1053 		if (error) {
1054 			kvfree(mp->m_rsum_cache);
1055 			mp->m_rsum_cache = rsum_cache;
1056 		} else {
1057 			kvfree(rsum_cache);
1058 		}
1059 	}
1060 
1061 	return error;
1062 }
1063 
1064 /*
1065  * Initialize realtime fields in the mount structure.
1066  */
1067 int				/* error */
1068 xfs_rtmount_init(
1069 	struct xfs_mount	*mp)	/* file system mount structure */
1070 {
1071 	struct xfs_buf		*bp;	/* buffer for last block of subvolume */
1072 	struct xfs_sb		*sbp;	/* filesystem superblock copy in mount */
1073 	xfs_daddr_t		d;	/* address of last block of subvolume */
1074 	unsigned int		rsumblocks;
1075 	int			error;
1076 
1077 	sbp = &mp->m_sb;
1078 	if (sbp->sb_rblocks == 0)
1079 		return 0;
1080 	if (mp->m_rtdev_targp == NULL) {
1081 		xfs_warn(mp,
1082 	"Filesystem has a realtime volume, use rtdev=device option");
1083 		return -ENODEV;
1084 	}
1085 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
1086 	rsumblocks = xfs_rtsummary_blockcount(mp, mp->m_rsumlevels,
1087 			mp->m_sb.sb_rbmblocks);
1088 	mp->m_rsumsize = XFS_FSB_TO_B(mp, rsumblocks);
1089 	mp->m_rbmip = mp->m_rsumip = NULL;
1090 	/*
1091 	 * Check that the realtime section is an ok size.
1092 	 */
1093 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
1094 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
1095 		xfs_warn(mp, "realtime mount -- %llu != %llu",
1096 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
1097 			(unsigned long long) mp->m_sb.sb_rblocks);
1098 		return -EFBIG;
1099 	}
1100 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
1101 					d - XFS_FSB_TO_BB(mp, 1),
1102 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
1103 	if (error) {
1104 		xfs_warn(mp, "realtime device size check failed");
1105 		return error;
1106 	}
1107 	xfs_buf_relse(bp);
1108 	return 0;
1109 }
1110 
1111 static int
1112 xfs_rtalloc_count_frextent(
1113 	struct xfs_mount		*mp,
1114 	struct xfs_trans		*tp,
1115 	const struct xfs_rtalloc_rec	*rec,
1116 	void				*priv)
1117 {
1118 	uint64_t			*valp = priv;
1119 
1120 	*valp += rec->ar_extcount;
1121 	return 0;
1122 }
1123 
1124 /*
1125  * Reinitialize the number of free realtime extents from the realtime bitmap.
1126  * Callers must ensure that there is no other activity in the filesystem.
1127  */
1128 int
1129 xfs_rtalloc_reinit_frextents(
1130 	struct xfs_mount	*mp)
1131 {
1132 	uint64_t		val = 0;
1133 	int			error;
1134 
1135 	xfs_rtbitmap_lock_shared(mp, XFS_RBMLOCK_BITMAP);
1136 	error = xfs_rtalloc_query_all(mp, NULL, xfs_rtalloc_count_frextent,
1137 			&val);
1138 	xfs_rtbitmap_unlock_shared(mp, XFS_RBMLOCK_BITMAP);
1139 	if (error)
1140 		return error;
1141 
1142 	spin_lock(&mp->m_sb_lock);
1143 	mp->m_sb.sb_frextents = val;
1144 	spin_unlock(&mp->m_sb_lock);
1145 	percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1146 	return 0;
1147 }
1148 
1149 /*
1150  * Read in the bmbt of an rt metadata inode so that we never have to load them
1151  * at runtime.  This enables the use of shared ILOCKs for rtbitmap scans.  Use
1152  * an empty transaction to avoid deadlocking on loops in the bmbt.
1153  */
1154 static inline int
1155 xfs_rtmount_iread_extents(
1156 	struct xfs_inode	*ip,
1157 	unsigned int		lock_class)
1158 {
1159 	struct xfs_trans	*tp;
1160 	int			error;
1161 
1162 	error = xfs_trans_alloc_empty(ip->i_mount, &tp);
1163 	if (error)
1164 		return error;
1165 
1166 	xfs_ilock(ip, XFS_ILOCK_EXCL | lock_class);
1167 
1168 	error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1169 	if (error)
1170 		goto out_unlock;
1171 
1172 	if (xfs_inode_has_attr_fork(ip)) {
1173 		error = xfs_iread_extents(tp, ip, XFS_ATTR_FORK);
1174 		if (error)
1175 			goto out_unlock;
1176 	}
1177 
1178 out_unlock:
1179 	xfs_iunlock(ip, XFS_ILOCK_EXCL | lock_class);
1180 	xfs_trans_cancel(tp);
1181 	return error;
1182 }
1183 
1184 /*
1185  * Get the bitmap and summary inodes and the summary cache into the mount
1186  * structure at mount time.
1187  */
1188 int					/* error */
1189 xfs_rtmount_inodes(
1190 	xfs_mount_t	*mp)		/* file system mount structure */
1191 {
1192 	int		error;		/* error return value */
1193 	xfs_sb_t	*sbp;
1194 
1195 	sbp = &mp->m_sb;
1196 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
1197 	if (xfs_metadata_is_sick(error))
1198 		xfs_rt_mark_sick(mp, XFS_SICK_RT_BITMAP);
1199 	if (error)
1200 		return error;
1201 	ASSERT(mp->m_rbmip != NULL);
1202 
1203 	error = xfs_rtmount_iread_extents(mp->m_rbmip, XFS_ILOCK_RTBITMAP);
1204 	if (error)
1205 		goto out_rele_bitmap;
1206 
1207 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
1208 	if (xfs_metadata_is_sick(error))
1209 		xfs_rt_mark_sick(mp, XFS_SICK_RT_SUMMARY);
1210 	if (error)
1211 		goto out_rele_bitmap;
1212 	ASSERT(mp->m_rsumip != NULL);
1213 
1214 	error = xfs_rtmount_iread_extents(mp->m_rsumip, XFS_ILOCK_RTSUM);
1215 	if (error)
1216 		goto out_rele_summary;
1217 
1218 	xfs_alloc_rsum_cache(mp, sbp->sb_rbmblocks);
1219 	return 0;
1220 
1221 out_rele_summary:
1222 	xfs_irele(mp->m_rsumip);
1223 out_rele_bitmap:
1224 	xfs_irele(mp->m_rbmip);
1225 	return error;
1226 }
1227 
1228 void
1229 xfs_rtunmount_inodes(
1230 	struct xfs_mount	*mp)
1231 {
1232 	kvfree(mp->m_rsum_cache);
1233 	if (mp->m_rbmip)
1234 		xfs_irele(mp->m_rbmip);
1235 	if (mp->m_rsumip)
1236 		xfs_irele(mp->m_rsumip);
1237 }
1238 
1239 /*
1240  * Pick an extent for allocation at the start of a new realtime file.
1241  * Use the sequence number stored in the atime field of the bitmap inode.
1242  * Translate this to a fraction of the rtextents, and return the product
1243  * of rtextents and the fraction.
1244  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
1245  */
1246 static int
1247 xfs_rtpick_extent(
1248 	xfs_mount_t		*mp,		/* file system mount point */
1249 	xfs_trans_t		*tp,		/* transaction pointer */
1250 	xfs_rtxlen_t		len,		/* allocation length (rtextents) */
1251 	xfs_rtxnum_t		*pick)		/* result rt extent */
1252 {
1253 	xfs_rtxnum_t		b;		/* result rtext */
1254 	int			log2;		/* log of sequence number */
1255 	uint64_t		resid;		/* residual after log removed */
1256 	uint64_t		seq;		/* sequence number of file creation */
1257 	struct timespec64	ts;		/* timespec in inode */
1258 
1259 	xfs_assert_ilocked(mp->m_rbmip, XFS_ILOCK_EXCL);
1260 
1261 	ts = inode_get_atime(VFS_I(mp->m_rbmip));
1262 	if (!(mp->m_rbmip->i_diflags & XFS_DIFLAG_NEWRTBM)) {
1263 		mp->m_rbmip->i_diflags |= XFS_DIFLAG_NEWRTBM;
1264 		seq = 0;
1265 	} else {
1266 		seq = ts.tv_sec;
1267 	}
1268 	if ((log2 = xfs_highbit64(seq)) == -1)
1269 		b = 0;
1270 	else {
1271 		resid = seq - (1ULL << log2);
1272 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
1273 		    (log2 + 1);
1274 		if (b >= mp->m_sb.sb_rextents)
1275 			div64_u64_rem(b, mp->m_sb.sb_rextents, &b);
1276 		if (b + len > mp->m_sb.sb_rextents)
1277 			b = mp->m_sb.sb_rextents - len;
1278 	}
1279 	ts.tv_sec = seq + 1;
1280 	inode_set_atime_to_ts(VFS_I(mp->m_rbmip), ts);
1281 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1282 	*pick = b;
1283 	return 0;
1284 }
1285 
1286 static void
1287 xfs_rtalloc_align_minmax(
1288 	xfs_rtxlen_t		*raminlen,
1289 	xfs_rtxlen_t		*ramaxlen,
1290 	xfs_rtxlen_t		*prod)
1291 {
1292 	xfs_rtxlen_t		newmaxlen = *ramaxlen;
1293 	xfs_rtxlen_t		newminlen = *raminlen;
1294 	xfs_rtxlen_t		slack;
1295 
1296 	slack = newmaxlen % *prod;
1297 	if (slack)
1298 		newmaxlen -= slack;
1299 	slack = newminlen % *prod;
1300 	if (slack)
1301 		newminlen += *prod - slack;
1302 
1303 	/*
1304 	 * If adjusting for extent size hint alignment produces an invalid
1305 	 * min/max len combination, go ahead without it.
1306 	 */
1307 	if (newmaxlen < newminlen) {
1308 		*prod = 1;
1309 		return;
1310 	}
1311 	*ramaxlen = newmaxlen;
1312 	*raminlen = newminlen;
1313 }
1314 
1315 int
1316 xfs_bmap_rtalloc(
1317 	struct xfs_bmalloca	*ap)
1318 {
1319 	struct xfs_mount	*mp = ap->ip->i_mount;
1320 	xfs_fileoff_t		orig_offset = ap->offset;
1321 	xfs_rtxnum_t		start;	   /* allocation hint rtextent no */
1322 	xfs_rtxnum_t		rtx;	   /* actually allocated rtextent no */
1323 	xfs_rtxlen_t		prod = 0;  /* product factor for allocators */
1324 	xfs_extlen_t		mod = 0;   /* product factor for allocators */
1325 	xfs_rtxlen_t		ralen = 0; /* realtime allocation length */
1326 	xfs_extlen_t		align;     /* minimum allocation alignment */
1327 	xfs_extlen_t		orig_length = ap->length;
1328 	xfs_extlen_t		minlen = mp->m_sb.sb_rextsize;
1329 	xfs_rtxlen_t		raminlen;
1330 	bool			rtlocked = false;
1331 	bool			ignore_locality = false;
1332 	struct xfs_rtalloc_args	args = {
1333 		.mp		= mp,
1334 		.tp		= ap->tp,
1335 	};
1336 	int			error;
1337 
1338 	align = xfs_get_extsz_hint(ap->ip);
1339 	if (!align)
1340 		align = 1;
1341 retry:
1342 	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
1343 					align, 1, ap->eof, 0,
1344 					ap->conv, &ap->offset, &ap->length);
1345 	if (error)
1346 		return error;
1347 	ASSERT(ap->length);
1348 	ASSERT(xfs_extlen_to_rtxmod(mp, ap->length) == 0);
1349 
1350 	/*
1351 	 * If we shifted the file offset downward to satisfy an extent size
1352 	 * hint, increase minlen by that amount so that the allocator won't
1353 	 * give us an allocation that's too short to cover at least one of the
1354 	 * blocks that the caller asked for.
1355 	 */
1356 	if (ap->offset != orig_offset)
1357 		minlen += orig_offset - ap->offset;
1358 
1359 	/*
1360 	 * Set ralen to be the actual requested length in rtextents.
1361 	 *
1362 	 * If the old value was close enough to XFS_BMBT_MAX_EXTLEN that
1363 	 * we rounded up to it, cut it back so it's valid again.
1364 	 * Note that if it's a really large request (bigger than
1365 	 * XFS_BMBT_MAX_EXTLEN), we don't hear about that number, and can't
1366 	 * adjust the starting point to match it.
1367 	 */
1368 	ralen = xfs_extlen_to_rtxlen(mp, min(ap->length, XFS_MAX_BMBT_EXTLEN));
1369 	raminlen = max_t(xfs_rtxlen_t, 1, xfs_extlen_to_rtxlen(mp, minlen));
1370 	ASSERT(raminlen > 0);
1371 	ASSERT(raminlen <= ralen);
1372 
1373 	/*
1374 	 * Lock out modifications to both the RT bitmap and summary inodes
1375 	 */
1376 	if (!rtlocked) {
1377 		xfs_rtbitmap_lock(ap->tp, mp);
1378 		rtlocked = true;
1379 	}
1380 
1381 	if (ignore_locality) {
1382 		start = 0;
1383 	} else if (xfs_bmap_adjacent(ap)) {
1384 		start = xfs_rtb_to_rtx(mp, ap->blkno);
1385 	} else if (ap->eof && ap->offset == 0) {
1386 		/*
1387 		 * If it's an allocation to an empty file at offset 0, pick an
1388 		 * extent that will space things out in the rt area.
1389 		 */
1390 		error = xfs_rtpick_extent(mp, ap->tp, ralen, &start);
1391 		if (error)
1392 			return error;
1393 	} else {
1394 		start = 0;
1395 	}
1396 
1397 	/*
1398 	 * Only bother calculating a real prod factor if offset & length are
1399 	 * perfectly aligned, otherwise it will just get us in trouble.
1400 	 */
1401 	div_u64_rem(ap->offset, align, &mod);
1402 	if (mod || ap->length % align) {
1403 		prod = 1;
1404 	} else {
1405 		prod = xfs_extlen_to_rtxlen(mp, align);
1406 		if (prod > 1)
1407 			xfs_rtalloc_align_minmax(&raminlen, &ralen, &prod);
1408 	}
1409 
1410 	if (start) {
1411 		error = xfs_rtallocate_extent_near(&args, start, raminlen,
1412 				ralen, &ralen, prod, &rtx);
1413 	} else {
1414 		error = xfs_rtallocate_extent_size(&args, raminlen,
1415 				ralen, &ralen, prod, &rtx);
1416 	}
1417 	xfs_rtbuf_cache_relse(&args);
1418 
1419 	if (error == -ENOSPC) {
1420 		if (align > mp->m_sb.sb_rextsize) {
1421 			/*
1422 			 * We previously enlarged the request length to try to
1423 			 * satisfy an extent size hint.  The allocator didn't
1424 			 * return anything, so reset the parameters to the
1425 			 * original values and try again without alignment
1426 			 * criteria.
1427 			 */
1428 			ap->offset = orig_offset;
1429 			ap->length = orig_length;
1430 			minlen = align = mp->m_sb.sb_rextsize;
1431 			goto retry;
1432 		}
1433 
1434 		if (!ignore_locality && start != 0) {
1435 			/*
1436 			 * If we can't allocate near a specific rt extent, try
1437 			 * again without locality criteria.
1438 			 */
1439 			ignore_locality = true;
1440 			goto retry;
1441 		}
1442 
1443 		ap->blkno = NULLFSBLOCK;
1444 		ap->length = 0;
1445 		return 0;
1446 	}
1447 	if (error)
1448 		return error;
1449 
1450 	xfs_trans_mod_sb(ap->tp, ap->wasdel ?
1451 			XFS_TRANS_SB_RES_FREXTENTS : XFS_TRANS_SB_FREXTENTS,
1452 			-(long)ralen);
1453 	ap->blkno = xfs_rtx_to_rtb(mp, rtx);
1454 	ap->length = xfs_rtxlen_to_extlen(mp, ralen);
1455 	xfs_bmap_alloc_account(ap);
1456 	return 0;
1457 }
1458