xref: /linux/fs/xfs/xfs_rtalloc.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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_alloc.h"
16 #include "xfs_bmap.h"
17 #include "xfs_bmap_btree.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_trans.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_icache.h"
22 #include "xfs_rtalloc.h"
23 #include "xfs_sb.h"
24 #include "xfs_rtbitmap.h"
25 #include "xfs_quota.h"
26 #include "xfs_log_priv.h"
27 #include "xfs_health.h"
28 
29 /*
30  * Return whether there are any free extents in the size range given
31  * by low and high, for the bitmap block bbno.
32  */
33 STATIC int
34 xfs_rtany_summary(
35 	struct xfs_rtalloc_args	*args,
36 	int			low,	/* low log2 extent size */
37 	int			high,	/* high log2 extent size */
38 	xfs_fileoff_t		bbno,	/* bitmap block number */
39 	int			*maxlog) /* out: max log2 extent size free */
40 {
41 	struct xfs_mount	*mp = args->mp;
42 	int			error;
43 	int			log;	/* loop counter, log2 of ext. size */
44 	xfs_suminfo_t		sum;	/* summary data */
45 
46 	/* There are no extents at levels >= m_rsum_cache[bbno]. */
47 	if (mp->m_rsum_cache) {
48 		high = min(high, mp->m_rsum_cache[bbno] - 1);
49 		if (low > high) {
50 			*maxlog = -1;
51 			return 0;
52 		}
53 	}
54 
55 	/*
56 	 * Loop over logs of extent sizes.
57 	 */
58 	for (log = high; log >= low; log--) {
59 		/*
60 		 * Get one summary datum.
61 		 */
62 		error = xfs_rtget_summary(args, log, bbno, &sum);
63 		if (error) {
64 			return error;
65 		}
66 		/*
67 		 * If there are any, return success.
68 		 */
69 		if (sum) {
70 			*maxlog = log;
71 			goto out;
72 		}
73 	}
74 	/*
75 	 * Found nothing, return failure.
76 	 */
77 	*maxlog = -1;
78 out:
79 	/* There were no extents at levels > log. */
80 	if (mp->m_rsum_cache && log + 1 < mp->m_rsum_cache[bbno])
81 		mp->m_rsum_cache[bbno] = log + 1;
82 	return 0;
83 }
84 
85 
86 /*
87  * Copy and transform the summary file, given the old and new
88  * parameters in the mount structures.
89  */
90 STATIC int
91 xfs_rtcopy_summary(
92 	struct xfs_rtalloc_args	*oargs,
93 	struct xfs_rtalloc_args	*nargs)
94 {
95 	xfs_fileoff_t		bbno;	/* bitmap block number */
96 	int			error;
97 	int			log;	/* summary level number (log length) */
98 	xfs_suminfo_t		sum;	/* summary data */
99 
100 	for (log = oargs->mp->m_rsumlevels - 1; log >= 0; log--) {
101 		for (bbno = oargs->mp->m_sb.sb_rbmblocks - 1;
102 		     (xfs_srtblock_t)bbno >= 0;
103 		     bbno--) {
104 			error = xfs_rtget_summary(oargs, log, bbno, &sum);
105 			if (error)
106 				goto out;
107 			if (sum == 0)
108 				continue;
109 			error = xfs_rtmodify_summary(oargs, log, bbno, -sum);
110 			if (error)
111 				goto out;
112 			error = xfs_rtmodify_summary(nargs, log, bbno, sum);
113 			if (error)
114 				goto out;
115 			ASSERT(sum > 0);
116 		}
117 	}
118 	error = 0;
119 out:
120 	xfs_rtbuf_cache_relse(oargs);
121 	return 0;
122 }
123 /*
124  * Mark an extent specified by start and len allocated.
125  * Updates all the summary information as well as the bitmap.
126  */
127 STATIC int
128 xfs_rtallocate_range(
129 	struct xfs_rtalloc_args	*args,
130 	xfs_rtxnum_t		start,	/* start rtext to allocate */
131 	xfs_rtxlen_t		len)	/* in/out: summary block number */
132 {
133 	struct xfs_mount	*mp = args->mp;
134 	xfs_rtxnum_t		end;	/* end of the allocated rtext */
135 	int			error;
136 	xfs_rtxnum_t		postblock = 0; /* first rtext allocated > end */
137 	xfs_rtxnum_t		preblock = 0; /* first rtext allocated < start */
138 
139 	end = start + len - 1;
140 	/*
141 	 * Assume we're allocating out of the middle of a free extent.
142 	 * We need to find the beginning and end of the extent so we can
143 	 * properly update the summary.
144 	 */
145 	error = xfs_rtfind_back(args, start, &preblock);
146 	if (error)
147 		return error;
148 
149 	/*
150 	 * Find the next allocated block (end of free extent).
151 	 */
152 	error = xfs_rtfind_forw(args, end, mp->m_sb.sb_rextents - 1,
153 			&postblock);
154 	if (error)
155 		return error;
156 
157 	/*
158 	 * Decrement the summary information corresponding to the entire
159 	 * (old) free extent.
160 	 */
161 	error = xfs_rtmodify_summary(args,
162 			xfs_highbit64(postblock + 1 - preblock),
163 			xfs_rtx_to_rbmblock(mp, preblock), -1);
164 	if (error)
165 		return error;
166 
167 	/*
168 	 * If there are blocks not being allocated at the front of the
169 	 * old extent, add summary data for them to be free.
170 	 */
171 	if (preblock < start) {
172 		error = xfs_rtmodify_summary(args,
173 				xfs_highbit64(start - preblock),
174 				xfs_rtx_to_rbmblock(mp, preblock), 1);
175 		if (error)
176 			return error;
177 	}
178 
179 	/*
180 	 * If there are blocks not being allocated at the end of the
181 	 * old extent, add summary data for them to be free.
182 	 */
183 	if (postblock > end) {
184 		error = xfs_rtmodify_summary(args,
185 				xfs_highbit64(postblock - end),
186 				xfs_rtx_to_rbmblock(mp, end + 1), 1);
187 		if (error)
188 			return error;
189 	}
190 
191 	/*
192 	 * Modify the bitmap to mark this extent allocated.
193 	 */
194 	return xfs_rtmodify_range(args, start, len, 0);
195 }
196 
197 /* Reduce @rtxlen until it is a multiple of @prod. */
198 static inline xfs_rtxlen_t
199 xfs_rtalloc_align_len(
200 	xfs_rtxlen_t	rtxlen,
201 	xfs_rtxlen_t	prod)
202 {
203 	if (unlikely(prod > 1))
204 		return rounddown(rtxlen, prod);
205 	return rtxlen;
206 }
207 
208 /*
209  * Make sure we don't run off the end of the rt volume.  Be careful that
210  * adjusting maxlen downwards doesn't cause us to fail the alignment checks.
211  */
212 static inline xfs_rtxlen_t
213 xfs_rtallocate_clamp_len(
214 	struct xfs_mount	*mp,
215 	xfs_rtxnum_t		startrtx,
216 	xfs_rtxlen_t		rtxlen,
217 	xfs_rtxlen_t		prod)
218 {
219 	xfs_rtxlen_t		ret;
220 
221 	ret = min(mp->m_sb.sb_rextents, startrtx + rtxlen) - startrtx;
222 	return xfs_rtalloc_align_len(ret, prod);
223 }
224 
225 /*
226  * Attempt to allocate an extent minlen<=len<=maxlen starting from
227  * bitmap block bbno.  If we don't get maxlen then use prod to trim
228  * the length, if given.  Returns error; returns starting block in *rtx.
229  * The lengths are all in rtextents.
230  */
231 STATIC int
232 xfs_rtallocate_extent_block(
233 	struct xfs_rtalloc_args	*args,
234 	xfs_fileoff_t		bbno,	/* bitmap block number */
235 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
236 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
237 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
238 	xfs_rtxnum_t		*nextp,	/* out: next rtext to try */
239 	xfs_rtxlen_t		prod,	/* extent product factor */
240 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
241 {
242 	struct xfs_mount	*mp = args->mp;
243 	xfs_rtxnum_t		besti = -1; /* best rtext found so far */
244 	xfs_rtxnum_t		end;	/* last rtext in chunk */
245 	xfs_rtxnum_t		i;	/* current rtext trying */
246 	xfs_rtxnum_t		next;	/* next rtext to try */
247 	xfs_rtxlen_t		scanlen; /* number of free rtx to look for */
248 	xfs_rtxlen_t		bestlen = 0; /* best length found so far */
249 	int			stat;	/* status from internal calls */
250 	int			error;
251 
252 	/*
253 	 * Loop over all the extents starting in this bitmap block up to the
254 	 * end of the rt volume, looking for one that's long enough.
255 	 */
256 	end = min(mp->m_sb.sb_rextents, xfs_rbmblock_to_rtx(mp, bbno + 1)) - 1;
257 	for (i = xfs_rbmblock_to_rtx(mp, bbno); i <= end; i++) {
258 		/* Make sure we don't scan off the end of the rt volume. */
259 		scanlen = xfs_rtallocate_clamp_len(mp, i, maxlen, prod);
260 		if (scanlen < minlen)
261 			break;
262 
263 		/*
264 		 * See if there's a free extent of scanlen starting at i.
265 		 * If it's not so then next will contain the first non-free.
266 		 */
267 		error = xfs_rtcheck_range(args, i, scanlen, 1, &next, &stat);
268 		if (error)
269 			return error;
270 		if (stat) {
271 			/*
272 			 * i to scanlen is all free, allocate and return that.
273 			 */
274 			*len = scanlen;
275 			*rtx = i;
276 			return 0;
277 		}
278 
279 		/*
280 		 * In the case where we have a variable-sized allocation
281 		 * request, figure out how big this free piece is,
282 		 * and if it's big enough for the minimum, and the best
283 		 * so far, remember it.
284 		 */
285 		if (minlen < maxlen) {
286 			xfs_rtxnum_t	thislen;	/* this extent size */
287 
288 			thislen = next - i;
289 			if (thislen >= minlen && thislen > bestlen) {
290 				besti = i;
291 				bestlen = thislen;
292 			}
293 		}
294 		/*
295 		 * If not done yet, find the start of the next free space.
296 		 */
297 		if (next >= end)
298 			break;
299 		error = xfs_rtfind_forw(args, next, end, &i);
300 		if (error)
301 			return error;
302 	}
303 
304 	/* Searched the whole thing & didn't find a maxlen free extent. */
305 	if (besti == -1)
306 		goto nospace;
307 
308 	/*
309 	 * Ensure bestlen is a multiple of prod, but don't return a too-short
310 	 * extent.
311 	 */
312 	bestlen = xfs_rtalloc_align_len(bestlen, prod);
313 	if (bestlen < minlen)
314 		goto nospace;
315 
316 	/*
317 	 * Pick besti for bestlen & return that.
318 	 */
319 	*len = bestlen;
320 	*rtx = besti;
321 	return 0;
322 nospace:
323 	/* Allocation failed.  Set *nextp to the next block to try. */
324 	*nextp = next;
325 	return -ENOSPC;
326 }
327 
328 /*
329  * Allocate an extent of length minlen<=len<=maxlen, starting at block
330  * bno.  If we don't get maxlen then use prod to trim the length, if given.
331  * Returns error; returns starting block in *rtx.
332  * The lengths are all in rtextents.
333  */
334 STATIC int
335 xfs_rtallocate_extent_exact(
336 	struct xfs_rtalloc_args	*args,
337 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
338 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
339 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
340 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
341 	xfs_rtxlen_t		prod,	/* extent product factor */
342 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
343 {
344 	struct xfs_mount	*mp = args->mp;
345 	xfs_rtxnum_t		next;	/* next rtext to try (dummy) */
346 	xfs_rtxlen_t		alloclen; /* candidate length */
347 	xfs_rtxlen_t		scanlen; /* number of free rtx to look for */
348 	int			isfree;	/* extent is free */
349 	int			error;
350 
351 	ASSERT(minlen % prod == 0);
352 	ASSERT(maxlen % prod == 0);
353 
354 	/* Make sure we don't run off the end of the rt volume. */
355 	scanlen = xfs_rtallocate_clamp_len(mp, start, maxlen, prod);
356 	if (scanlen < minlen)
357 		return -ENOSPC;
358 
359 	/* Check if the range in question (for scanlen) is free. */
360 	error = xfs_rtcheck_range(args, start, scanlen, 1, &next, &isfree);
361 	if (error)
362 		return error;
363 
364 	if (isfree) {
365 		/* start to scanlen is all free; allocate it. */
366 		*len = scanlen;
367 		*rtx = start;
368 		return 0;
369 	}
370 
371 	/*
372 	 * If not, allocate what there is, if it's at least minlen.
373 	 */
374 	alloclen = next - start;
375 	if (alloclen < minlen)
376 		return -ENOSPC;
377 
378 	/* Ensure alloclen is a multiple of prod. */
379 	alloclen = xfs_rtalloc_align_len(alloclen, prod);
380 	if (alloclen < minlen)
381 		return -ENOSPC;
382 
383 	*len = alloclen;
384 	*rtx = start;
385 	return 0;
386 }
387 
388 /*
389  * Allocate an extent of length minlen<=len<=maxlen, starting as near
390  * to start as possible.  If we don't get maxlen then use prod to trim
391  * the length, if given.  The lengths are all in rtextents.
392  */
393 STATIC int
394 xfs_rtallocate_extent_near(
395 	struct xfs_rtalloc_args	*args,
396 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
397 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
398 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
399 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
400 	xfs_rtxlen_t		prod,	/* extent product factor */
401 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
402 {
403 	struct xfs_mount	*mp = args->mp;
404 	int			maxlog;	/* max useful extent from summary */
405 	xfs_fileoff_t		bbno;	/* bitmap block number */
406 	int			error;
407 	int			i;	/* bitmap block offset (loop control) */
408 	int			j;	/* secondary loop control */
409 	int			log2len; /* log2 of minlen */
410 	xfs_rtxnum_t		n;	/* next rtext to try */
411 
412 	ASSERT(minlen % prod == 0);
413 	ASSERT(maxlen % prod == 0);
414 
415 	/*
416 	 * If the block number given is off the end, silently set it to
417 	 * the last block.
418 	 */
419 	if (start >= mp->m_sb.sb_rextents)
420 		start = mp->m_sb.sb_rextents - 1;
421 
422 	/*
423 	 * Try the exact allocation first.
424 	 */
425 	error = xfs_rtallocate_extent_exact(args, start, minlen, maxlen, len,
426 			prod, rtx);
427 	if (error != -ENOSPC)
428 		return error;
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 	int			error;
553 
554 	for (i = 0; i < args->mp->m_sb.sb_rbmblocks; i++) {
555 		xfs_suminfo_t	sum;	/* summary information for extents */
556 		xfs_rtxnum_t	n;	/* next rtext to be tried */
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 static int
653 xfs_alloc_rsum_cache(
654 	struct xfs_mount	*mp,
655 	xfs_extlen_t		rbmblocks)
656 {
657 	/*
658 	 * The rsum cache is initialized to the maximum value, which is
659 	 * trivially an upper bound on the maximum level with any free extents.
660 	 */
661 	mp->m_rsum_cache = kvmalloc(rbmblocks, GFP_KERNEL);
662 	if (!mp->m_rsum_cache)
663 		return -ENOMEM;
664 	memset(mp->m_rsum_cache, -1, rbmblocks);
665 	return 0;
666 }
667 
668 /*
669  * If we changed the rt extent size (meaning there was no rt volume previously)
670  * and the root directory had EXTSZINHERIT and RTINHERIT set, it's possible
671  * that the extent size hint on the root directory is no longer congruent with
672  * the new rt extent size.  Log the rootdir inode to fix this.
673  */
674 static int
675 xfs_growfs_rt_fixup_extsize(
676 	struct xfs_mount	*mp)
677 {
678 	struct xfs_inode	*ip = mp->m_rootip;
679 	struct xfs_trans	*tp;
680 	int			error = 0;
681 
682 	xfs_ilock(ip, XFS_IOLOCK_EXCL);
683 	if (!(ip->i_diflags & XFS_DIFLAG_RTINHERIT) ||
684 	    !(ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT))
685 		goto out_iolock;
686 
687 	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_ichange, 0, 0, false,
688 			&tp);
689 	if (error)
690 		goto out_iolock;
691 
692 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
693 	error = xfs_trans_commit(tp);
694 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
695 
696 out_iolock:
697 	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
698 	return error;
699 }
700 
701 static int
702 xfs_growfs_rt_bmblock(
703 	struct xfs_mount	*mp,
704 	xfs_rfsblock_t		nrblocks,
705 	xfs_agblock_t		rextsize,
706 	xfs_fileoff_t		bmbno)
707 {
708 	struct xfs_inode	*rbmip = mp->m_rbmip;
709 	struct xfs_inode	*rsumip = mp->m_rsumip;
710 	struct xfs_rtalloc_args	args = {
711 		.mp		= mp,
712 	};
713 	struct xfs_rtalloc_args	nargs = {
714 	};
715 	struct xfs_mount	*nmp;
716 	xfs_rfsblock_t		nrblocks_step;
717 	xfs_rtbxlen_t		freed_rtx;
718 	int			error;
719 
720 
721 	nrblocks_step = (bmbno + 1) * NBBY * mp->m_sb.sb_blocksize * rextsize;
722 
723 	nmp = nargs.mp = kmemdup(mp, sizeof(*mp), GFP_KERNEL);
724 	if (!nmp)
725 		return -ENOMEM;
726 
727 	/*
728 	 * Calculate new sb and mount fields for this round.
729 	 */
730 	nmp->m_sb.sb_rextsize = rextsize;
731 	xfs_mount_sb_set_rextsize(nmp, &nmp->m_sb);
732 	nmp->m_sb.sb_rbmblocks = bmbno + 1;
733 	nmp->m_sb.sb_rblocks = min(nrblocks, nrblocks_step);
734 	nmp->m_sb.sb_rextents = xfs_rtb_to_rtx(nmp, nmp->m_sb.sb_rblocks);
735 	nmp->m_sb.sb_rextslog = xfs_compute_rextslog(nmp->m_sb.sb_rextents);
736 	nmp->m_rsumlevels = nmp->m_sb.sb_rextslog + 1;
737 	nmp->m_rsumblocks = xfs_rtsummary_blockcount(mp, nmp->m_rsumlevels,
738 			nmp->m_sb.sb_rbmblocks);
739 
740 	/*
741 	 * Recompute the growfsrt reservation from the new rsumsize, so that the
742 	 * transaction below use the new, potentially larger value.
743 	 * */
744 	xfs_trans_resv_calc(nmp, &nmp->m_resv);
745 	error = xfs_trans_alloc(mp, &M_RES(nmp)->tr_growrtfree, 0, 0, 0,
746 			&args.tp);
747 	if (error)
748 		goto out_free;
749 	nargs.tp = args.tp;
750 
751 	xfs_rtbitmap_lock(mp);
752 	xfs_rtbitmap_trans_join(args.tp);
753 
754 	/*
755 	 * Update the bitmap inode's size ondisk and incore.  We need to update
756 	 * the incore size so that inode inactivation won't punch what it thinks
757 	 * are "posteof" blocks.
758 	 */
759 	rbmip->i_disk_size = nmp->m_sb.sb_rbmblocks * nmp->m_sb.sb_blocksize;
760 	i_size_write(VFS_I(rbmip), rbmip->i_disk_size);
761 	xfs_trans_log_inode(args.tp, rbmip, XFS_ILOG_CORE);
762 
763 	/*
764 	 * Update the summary inode's size.  We need to update the incore size
765 	 * so that inode inactivation won't punch what it thinks are "posteof"
766 	 * blocks.
767 	 */
768 	rsumip->i_disk_size = nmp->m_rsumblocks * nmp->m_sb.sb_blocksize;
769 	i_size_write(VFS_I(rsumip), rsumip->i_disk_size);
770 	xfs_trans_log_inode(args.tp, rsumip, XFS_ILOG_CORE);
771 
772 	/*
773 	 * Copy summary data from old to new sizes when the real size (not
774 	 * block-aligned) changes.
775 	 */
776 	if (mp->m_sb.sb_rbmblocks != nmp->m_sb.sb_rbmblocks ||
777 	    mp->m_rsumlevels != nmp->m_rsumlevels) {
778 		error = xfs_rtcopy_summary(&args, &nargs);
779 		if (error)
780 			goto out_cancel;
781 	}
782 
783 	/*
784 	 * Update superblock fields.
785 	 */
786 	if (nmp->m_sb.sb_rextsize != mp->m_sb.sb_rextsize)
787 		xfs_trans_mod_sb(args.tp, XFS_TRANS_SB_REXTSIZE,
788 			nmp->m_sb.sb_rextsize - mp->m_sb.sb_rextsize);
789 	if (nmp->m_sb.sb_rbmblocks != mp->m_sb.sb_rbmblocks)
790 		xfs_trans_mod_sb(args.tp, XFS_TRANS_SB_RBMBLOCKS,
791 			nmp->m_sb.sb_rbmblocks - mp->m_sb.sb_rbmblocks);
792 	if (nmp->m_sb.sb_rblocks != mp->m_sb.sb_rblocks)
793 		xfs_trans_mod_sb(args.tp, XFS_TRANS_SB_RBLOCKS,
794 			nmp->m_sb.sb_rblocks - mp->m_sb.sb_rblocks);
795 	if (nmp->m_sb.sb_rextents != mp->m_sb.sb_rextents)
796 		xfs_trans_mod_sb(args.tp, XFS_TRANS_SB_REXTENTS,
797 			nmp->m_sb.sb_rextents - mp->m_sb.sb_rextents);
798 	if (nmp->m_sb.sb_rextslog != mp->m_sb.sb_rextslog)
799 		xfs_trans_mod_sb(args.tp, XFS_TRANS_SB_REXTSLOG,
800 			nmp->m_sb.sb_rextslog - mp->m_sb.sb_rextslog);
801 
802 	/*
803 	 * Free the new extent.
804 	 */
805 	freed_rtx = nmp->m_sb.sb_rextents - mp->m_sb.sb_rextents;
806 	error = xfs_rtfree_range(&nargs, mp->m_sb.sb_rextents, freed_rtx);
807 	xfs_rtbuf_cache_relse(&nargs);
808 	if (error)
809 		goto out_cancel;
810 
811 	/*
812 	 * Mark more blocks free in the superblock.
813 	 */
814 	xfs_trans_mod_sb(args.tp, XFS_TRANS_SB_FREXTENTS, freed_rtx);
815 
816 	/*
817 	 * Update the calculated values in the real mount structure.
818 	 */
819 	mp->m_rsumlevels = nmp->m_rsumlevels;
820 	mp->m_rsumblocks = nmp->m_rsumblocks;
821 	xfs_mount_sb_set_rextsize(mp, &mp->m_sb);
822 
823 	/*
824 	 * Recompute the growfsrt reservation from the new rsumsize.
825 	 */
826 	xfs_trans_resv_calc(mp, &mp->m_resv);
827 
828 	error = xfs_trans_commit(args.tp);
829 	if (error)
830 		goto out_free;
831 
832 	/*
833 	 * Ensure the mount RT feature flag is now set.
834 	 */
835 	mp->m_features |= XFS_FEAT_REALTIME;
836 
837 	kfree(nmp);
838 	return 0;
839 
840 out_cancel:
841 	xfs_trans_cancel(args.tp);
842 out_free:
843 	kfree(nmp);
844 	return error;
845 }
846 
847 /*
848  * Calculate the last rbmblock currently used.
849  *
850  * This also deals with the case where there were no rtextents before.
851  */
852 static xfs_fileoff_t
853 xfs_last_rt_bmblock(
854 	struct xfs_mount	*mp)
855 {
856 	xfs_fileoff_t		bmbno = mp->m_sb.sb_rbmblocks;
857 
858 	/* Skip the current block if it is exactly full. */
859 	if (xfs_rtx_to_rbmword(mp, mp->m_sb.sb_rextents) != 0)
860 		bmbno--;
861 	return bmbno;
862 }
863 
864 /*
865  * Grow the realtime area of the filesystem.
866  */
867 int
868 xfs_growfs_rt(
869 	xfs_mount_t	*mp,		/* mount point for filesystem */
870 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
871 {
872 	xfs_fileoff_t	bmbno;		/* bitmap block number */
873 	struct xfs_buf	*bp;		/* temporary buffer */
874 	int		error;		/* error return value */
875 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
876 	xfs_rtxnum_t	nrextents;	/* new number of realtime extents */
877 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
878 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
879 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
880 	uint8_t		*rsum_cache;	/* old summary cache */
881 	xfs_agblock_t	old_rextsize = mp->m_sb.sb_rextsize;
882 
883 	if (!capable(CAP_SYS_ADMIN))
884 		return -EPERM;
885 
886 	/* Needs to have been mounted with an rt device. */
887 	if (!XFS_IS_REALTIME_MOUNT(mp))
888 		return -EINVAL;
889 
890 	if (!mutex_trylock(&mp->m_growlock))
891 		return -EWOULDBLOCK;
892 	/*
893 	 * Mount should fail if the rt bitmap/summary files don't load, but
894 	 * we'll check anyway.
895 	 */
896 	error = -EINVAL;
897 	if (!mp->m_rbmip || !mp->m_rsumip)
898 		goto out_unlock;
899 
900 	/* Shrink not supported. */
901 	if (in->newblocks <= mp->m_sb.sb_rblocks)
902 		goto out_unlock;
903 	/* Can only change rt extent size when adding rt volume. */
904 	if (mp->m_sb.sb_rblocks > 0 && in->extsize != mp->m_sb.sb_rextsize)
905 		goto out_unlock;
906 
907 	/* Range check the extent size. */
908 	if (XFS_FSB_TO_B(mp, in->extsize) > XFS_MAX_RTEXTSIZE ||
909 	    XFS_FSB_TO_B(mp, in->extsize) < XFS_MIN_RTEXTSIZE)
910 		goto out_unlock;
911 
912 	/* Unsupported realtime features. */
913 	error = -EOPNOTSUPP;
914 	if (xfs_has_rmapbt(mp) || xfs_has_reflink(mp) || xfs_has_quota(mp))
915 		goto out_unlock;
916 
917 	error = xfs_sb_validate_fsb_count(&mp->m_sb, in->newblocks);
918 	if (error)
919 		goto out_unlock;
920 	/*
921 	 * Read in the last block of the device, make sure it exists.
922 	 */
923 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
924 				XFS_FSB_TO_BB(mp, in->newblocks - 1),
925 				XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
926 	if (error)
927 		goto out_unlock;
928 	xfs_buf_relse(bp);
929 
930 	/*
931 	 * Calculate new parameters.  These are the final values to be reached.
932 	 */
933 	nrextents = div_u64(in->newblocks, in->extsize);
934 	if (nrextents == 0) {
935 		error = -EINVAL;
936 		goto out_unlock;
937 	}
938 	nrbmblocks = xfs_rtbitmap_blockcount(mp, nrextents);
939 	nrsumblocks = xfs_rtsummary_blockcount(mp,
940 			xfs_compute_rextslog(nrextents) + 1, nrbmblocks);
941 
942 	/*
943 	 * New summary size can't be more than half the size of
944 	 * the log.  This prevents us from getting a log overflow,
945 	 * since we'll log basically the whole summary file at once.
946 	 */
947 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1)) {
948 		error = -EINVAL;
949 		goto out_unlock;
950 	}
951 
952 	/*
953 	 * Get the old block counts for bitmap and summary inodes.
954 	 * These can't change since other growfs callers are locked out.
955 	 */
956 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_disk_size);
957 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_disk_size);
958 	/*
959 	 * Allocate space to the bitmap and summary files, as necessary.
960 	 */
961 	error = xfs_rtfile_initialize_blocks(mp->m_rbmip, rbmblocks,
962 			nrbmblocks, NULL);
963 	if (error)
964 		goto out_unlock;
965 	error = xfs_rtfile_initialize_blocks(mp->m_rsumip, rsumblocks,
966 			nrsumblocks, NULL);
967 	if (error)
968 		goto out_unlock;
969 
970 	rsum_cache = mp->m_rsum_cache;
971 	if (nrbmblocks != mp->m_sb.sb_rbmblocks) {
972 		error = xfs_alloc_rsum_cache(mp, nrbmblocks);
973 		if (error)
974 			goto out_unlock;
975 	}
976 
977 	/* Initialize the free space bitmap one bitmap block at a time. */
978 	for (bmbno = xfs_last_rt_bmblock(mp); bmbno < nrbmblocks; bmbno++) {
979 		error = xfs_growfs_rt_bmblock(mp, in->newblocks, in->extsize,
980 				bmbno);
981 		if (error)
982 			goto out_free;
983 	}
984 
985 	if (old_rextsize != in->extsize) {
986 		error = xfs_growfs_rt_fixup_extsize(mp);
987 		if (error)
988 			goto out_free;
989 	}
990 
991 	/* Update secondary superblocks now the physical grow has completed */
992 	error = xfs_update_secondary_sbs(mp);
993 
994 out_free:
995 	/*
996 	 * If we had to allocate a new rsum_cache, we either need to free the
997 	 * old one (if we succeeded) or free the new one and restore the old one
998 	 * (if there was an error).
999 	 */
1000 	if (rsum_cache != mp->m_rsum_cache) {
1001 		if (error) {
1002 			kvfree(mp->m_rsum_cache);
1003 			mp->m_rsum_cache = rsum_cache;
1004 		} else {
1005 			kvfree(rsum_cache);
1006 		}
1007 	}
1008 
1009 out_unlock:
1010 	mutex_unlock(&mp->m_growlock);
1011 	return error;
1012 }
1013 
1014 /*
1015  * Initialize realtime fields in the mount structure.
1016  */
1017 int				/* error */
1018 xfs_rtmount_init(
1019 	struct xfs_mount	*mp)	/* file system mount structure */
1020 {
1021 	struct xfs_buf		*bp;	/* buffer for last block of subvolume */
1022 	struct xfs_sb		*sbp;	/* filesystem superblock copy in mount */
1023 	xfs_daddr_t		d;	/* address of last block of subvolume */
1024 	int			error;
1025 
1026 	sbp = &mp->m_sb;
1027 	if (sbp->sb_rblocks == 0)
1028 		return 0;
1029 	if (mp->m_rtdev_targp == NULL) {
1030 		xfs_warn(mp,
1031 	"Filesystem has a realtime volume, use rtdev=device option");
1032 		return -ENODEV;
1033 	}
1034 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
1035 	mp->m_rsumblocks = xfs_rtsummary_blockcount(mp, mp->m_rsumlevels,
1036 			mp->m_sb.sb_rbmblocks);
1037 	mp->m_rbmip = mp->m_rsumip = NULL;
1038 	/*
1039 	 * Check that the realtime section is an ok size.
1040 	 */
1041 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
1042 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
1043 		xfs_warn(mp, "realtime mount -- %llu != %llu",
1044 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
1045 			(unsigned long long) mp->m_sb.sb_rblocks);
1046 		return -EFBIG;
1047 	}
1048 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
1049 					d - XFS_FSB_TO_BB(mp, 1),
1050 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
1051 	if (error) {
1052 		xfs_warn(mp, "realtime device size check failed");
1053 		return error;
1054 	}
1055 	xfs_buf_relse(bp);
1056 	return 0;
1057 }
1058 
1059 static int
1060 xfs_rtalloc_count_frextent(
1061 	struct xfs_mount		*mp,
1062 	struct xfs_trans		*tp,
1063 	const struct xfs_rtalloc_rec	*rec,
1064 	void				*priv)
1065 {
1066 	uint64_t			*valp = priv;
1067 
1068 	*valp += rec->ar_extcount;
1069 	return 0;
1070 }
1071 
1072 /*
1073  * Reinitialize the number of free realtime extents from the realtime bitmap.
1074  * Callers must ensure that there is no other activity in the filesystem.
1075  */
1076 int
1077 xfs_rtalloc_reinit_frextents(
1078 	struct xfs_mount	*mp)
1079 {
1080 	uint64_t		val = 0;
1081 	int			error;
1082 
1083 	xfs_rtbitmap_lock_shared(mp, XFS_RBMLOCK_BITMAP);
1084 	error = xfs_rtalloc_query_all(mp, NULL, xfs_rtalloc_count_frextent,
1085 			&val);
1086 	xfs_rtbitmap_unlock_shared(mp, XFS_RBMLOCK_BITMAP);
1087 	if (error)
1088 		return error;
1089 
1090 	spin_lock(&mp->m_sb_lock);
1091 	mp->m_sb.sb_frextents = val;
1092 	spin_unlock(&mp->m_sb_lock);
1093 	percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1094 	return 0;
1095 }
1096 
1097 /*
1098  * Read in the bmbt of an rt metadata inode so that we never have to load them
1099  * at runtime.  This enables the use of shared ILOCKs for rtbitmap scans.  Use
1100  * an empty transaction to avoid deadlocking on loops in the bmbt.
1101  */
1102 static inline int
1103 xfs_rtmount_iread_extents(
1104 	struct xfs_inode	*ip,
1105 	unsigned int		lock_class)
1106 {
1107 	struct xfs_trans	*tp;
1108 	int			error;
1109 
1110 	error = xfs_trans_alloc_empty(ip->i_mount, &tp);
1111 	if (error)
1112 		return error;
1113 
1114 	xfs_ilock(ip, XFS_ILOCK_EXCL | lock_class);
1115 
1116 	error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1117 	if (error)
1118 		goto out_unlock;
1119 
1120 	if (xfs_inode_has_attr_fork(ip)) {
1121 		error = xfs_iread_extents(tp, ip, XFS_ATTR_FORK);
1122 		if (error)
1123 			goto out_unlock;
1124 	}
1125 
1126 out_unlock:
1127 	xfs_iunlock(ip, XFS_ILOCK_EXCL | lock_class);
1128 	xfs_trans_cancel(tp);
1129 	return error;
1130 }
1131 
1132 /*
1133  * Get the bitmap and summary inodes and the summary cache into the mount
1134  * structure at mount time.
1135  */
1136 int					/* error */
1137 xfs_rtmount_inodes(
1138 	xfs_mount_t	*mp)		/* file system mount structure */
1139 {
1140 	int		error;		/* error return value */
1141 	xfs_sb_t	*sbp;
1142 
1143 	sbp = &mp->m_sb;
1144 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
1145 	if (xfs_metadata_is_sick(error))
1146 		xfs_rt_mark_sick(mp, XFS_SICK_RT_BITMAP);
1147 	if (error)
1148 		return error;
1149 	ASSERT(mp->m_rbmip != NULL);
1150 
1151 	error = xfs_rtmount_iread_extents(mp->m_rbmip, XFS_ILOCK_RTBITMAP);
1152 	if (error)
1153 		goto out_rele_bitmap;
1154 
1155 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
1156 	if (xfs_metadata_is_sick(error))
1157 		xfs_rt_mark_sick(mp, XFS_SICK_RT_SUMMARY);
1158 	if (error)
1159 		goto out_rele_bitmap;
1160 	ASSERT(mp->m_rsumip != NULL);
1161 
1162 	error = xfs_rtmount_iread_extents(mp->m_rsumip, XFS_ILOCK_RTSUM);
1163 	if (error)
1164 		goto out_rele_summary;
1165 
1166 	error = xfs_alloc_rsum_cache(mp, sbp->sb_rbmblocks);
1167 	if (error)
1168 		goto out_rele_summary;
1169 	return 0;
1170 
1171 out_rele_summary:
1172 	xfs_irele(mp->m_rsumip);
1173 out_rele_bitmap:
1174 	xfs_irele(mp->m_rbmip);
1175 	return error;
1176 }
1177 
1178 void
1179 xfs_rtunmount_inodes(
1180 	struct xfs_mount	*mp)
1181 {
1182 	kvfree(mp->m_rsum_cache);
1183 	if (mp->m_rbmip)
1184 		xfs_irele(mp->m_rbmip);
1185 	if (mp->m_rsumip)
1186 		xfs_irele(mp->m_rsumip);
1187 }
1188 
1189 /*
1190  * Pick an extent for allocation at the start of a new realtime file.
1191  * Use the sequence number stored in the atime field of the bitmap inode.
1192  * Translate this to a fraction of the rtextents, and return the product
1193  * of rtextents and the fraction.
1194  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
1195  */
1196 static xfs_rtxnum_t
1197 xfs_rtpick_extent(
1198 	xfs_mount_t		*mp,		/* file system mount point */
1199 	xfs_trans_t		*tp,		/* transaction pointer */
1200 	xfs_rtxlen_t		len)		/* allocation length (rtextents) */
1201 {
1202 	xfs_rtxnum_t		b;		/* result rtext */
1203 	int			log2;		/* log of sequence number */
1204 	uint64_t		resid;		/* residual after log removed */
1205 	uint64_t		seq;		/* sequence number of file creation */
1206 	struct timespec64	ts;		/* timespec in inode */
1207 
1208 	xfs_assert_ilocked(mp->m_rbmip, XFS_ILOCK_EXCL);
1209 
1210 	ts = inode_get_atime(VFS_I(mp->m_rbmip));
1211 	if (!(mp->m_rbmip->i_diflags & XFS_DIFLAG_NEWRTBM)) {
1212 		mp->m_rbmip->i_diflags |= XFS_DIFLAG_NEWRTBM;
1213 		seq = 0;
1214 	} else {
1215 		seq = ts.tv_sec;
1216 	}
1217 	if ((log2 = xfs_highbit64(seq)) == -1)
1218 		b = 0;
1219 	else {
1220 		resid = seq - (1ULL << log2);
1221 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
1222 		    (log2 + 1);
1223 		if (b >= mp->m_sb.sb_rextents)
1224 			div64_u64_rem(b, mp->m_sb.sb_rextents, &b);
1225 		if (b + len > mp->m_sb.sb_rextents)
1226 			b = mp->m_sb.sb_rextents - len;
1227 	}
1228 	ts.tv_sec = seq + 1;
1229 	inode_set_atime_to_ts(VFS_I(mp->m_rbmip), ts);
1230 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1231 	return b;
1232 }
1233 
1234 static void
1235 xfs_rtalloc_align_minmax(
1236 	xfs_rtxlen_t		*raminlen,
1237 	xfs_rtxlen_t		*ramaxlen,
1238 	xfs_rtxlen_t		*prod)
1239 {
1240 	xfs_rtxlen_t		newmaxlen = *ramaxlen;
1241 	xfs_rtxlen_t		newminlen = *raminlen;
1242 	xfs_rtxlen_t		slack;
1243 
1244 	slack = newmaxlen % *prod;
1245 	if (slack)
1246 		newmaxlen -= slack;
1247 	slack = newminlen % *prod;
1248 	if (slack)
1249 		newminlen += *prod - slack;
1250 
1251 	/*
1252 	 * If adjusting for extent size hint alignment produces an invalid
1253 	 * min/max len combination, go ahead without it.
1254 	 */
1255 	if (newmaxlen < newminlen) {
1256 		*prod = 1;
1257 		return;
1258 	}
1259 	*ramaxlen = newmaxlen;
1260 	*raminlen = newminlen;
1261 }
1262 
1263 static int
1264 xfs_rtallocate(
1265 	struct xfs_trans	*tp,
1266 	xfs_rtblock_t		bno_hint,
1267 	xfs_rtxlen_t		minlen,
1268 	xfs_rtxlen_t		maxlen,
1269 	xfs_rtxlen_t		prod,
1270 	bool			wasdel,
1271 	bool			initial_user_data,
1272 	bool			*rtlocked,
1273 	xfs_rtblock_t		*bno,
1274 	xfs_extlen_t		*blen)
1275 {
1276 	struct xfs_rtalloc_args	args = {
1277 		.mp		= tp->t_mountp,
1278 		.tp		= tp,
1279 	};
1280 	xfs_rtxnum_t		start = 0;
1281 	xfs_rtxnum_t		rtx;
1282 	xfs_rtxlen_t		len = 0;
1283 	int			error = 0;
1284 
1285 	/*
1286 	 * Lock out modifications to both the RT bitmap and summary inodes.
1287 	 */
1288 	if (!*rtlocked) {
1289 		xfs_rtbitmap_lock(args.mp);
1290 		xfs_rtbitmap_trans_join(tp);
1291 		*rtlocked = true;
1292 	}
1293 
1294 	/*
1295 	 * For an allocation to an empty file at offset 0, pick an extent that
1296 	 * will space things out in the rt area.
1297 	 */
1298 	if (bno_hint)
1299 		start = xfs_rtb_to_rtx(args.mp, bno_hint);
1300 	else if (initial_user_data)
1301 		start = xfs_rtpick_extent(args.mp, tp, maxlen);
1302 
1303 	if (start) {
1304 		error = xfs_rtallocate_extent_near(&args, start, minlen, maxlen,
1305 				&len, prod, &rtx);
1306 		/*
1307 		 * If we can't allocate near a specific rt extent, try again
1308 		 * without locality criteria.
1309 		 */
1310 		if (error == -ENOSPC) {
1311 			xfs_rtbuf_cache_relse(&args);
1312 			error = 0;
1313 		}
1314 	}
1315 
1316 	if (!error) {
1317 		error = xfs_rtallocate_extent_size(&args, minlen, maxlen, &len,
1318 				prod, &rtx);
1319 	}
1320 
1321 	if (error)
1322 		goto out_release;
1323 
1324 	error = xfs_rtallocate_range(&args, rtx, len);
1325 	if (error)
1326 		goto out_release;
1327 
1328 	xfs_trans_mod_sb(tp, wasdel ?
1329 			XFS_TRANS_SB_RES_FREXTENTS : XFS_TRANS_SB_FREXTENTS,
1330 			-(long)len);
1331 	*bno = xfs_rtx_to_rtb(args.mp, rtx);
1332 	*blen = xfs_rtxlen_to_extlen(args.mp, len);
1333 
1334 out_release:
1335 	xfs_rtbuf_cache_relse(&args);
1336 	return error;
1337 }
1338 
1339 static int
1340 xfs_rtallocate_align(
1341 	struct xfs_bmalloca	*ap,
1342 	xfs_rtxlen_t		*ralen,
1343 	xfs_rtxlen_t		*raminlen,
1344 	xfs_rtxlen_t		*prod,
1345 	bool			*noalign)
1346 {
1347 	struct xfs_mount	*mp = ap->ip->i_mount;
1348 	xfs_fileoff_t		orig_offset = ap->offset;
1349 	xfs_extlen_t		minlen = mp->m_sb.sb_rextsize;
1350 	xfs_extlen_t            align;	/* minimum allocation alignment */
1351 	xfs_extlen_t		mod;	/* product factor for allocators */
1352 	int			error;
1353 
1354 	if (*noalign) {
1355 		align = mp->m_sb.sb_rextsize;
1356 	} else {
1357 		align = xfs_get_extsz_hint(ap->ip);
1358 		if (!align)
1359 			align = 1;
1360 		if (align == mp->m_sb.sb_rextsize)
1361 			*noalign = true;
1362 	}
1363 
1364 	error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 1,
1365 			ap->eof, 0, ap->conv, &ap->offset, &ap->length);
1366 	if (error)
1367 		return error;
1368 	ASSERT(ap->length);
1369 	ASSERT(xfs_extlen_to_rtxmod(mp, ap->length) == 0);
1370 
1371 	/*
1372 	 * If we shifted the file offset downward to satisfy an extent size
1373 	 * hint, increase minlen by that amount so that the allocator won't
1374 	 * give us an allocation that's too short to cover at least one of the
1375 	 * blocks that the caller asked for.
1376 	 */
1377 	if (ap->offset != orig_offset)
1378 		minlen += orig_offset - ap->offset;
1379 
1380 	/*
1381 	 * Set ralen to be the actual requested length in rtextents.
1382 	 *
1383 	 * If the old value was close enough to XFS_BMBT_MAX_EXTLEN that
1384 	 * we rounded up to it, cut it back so it's valid again.
1385 	 * Note that if it's a really large request (bigger than
1386 	 * XFS_BMBT_MAX_EXTLEN), we don't hear about that number, and can't
1387 	 * adjust the starting point to match it.
1388 	 */
1389 	*ralen = xfs_extlen_to_rtxlen(mp, min(ap->length, XFS_MAX_BMBT_EXTLEN));
1390 	*raminlen = max_t(xfs_rtxlen_t, 1, xfs_extlen_to_rtxlen(mp, minlen));
1391 	ASSERT(*raminlen > 0);
1392 	ASSERT(*raminlen <= *ralen);
1393 
1394 	/*
1395 	 * Only bother calculating a real prod factor if offset & length are
1396 	 * perfectly aligned, otherwise it will just get us in trouble.
1397 	 */
1398 	div_u64_rem(ap->offset, align, &mod);
1399 	if (mod || ap->length % align)
1400 		*prod = 1;
1401 	else
1402 		*prod = xfs_extlen_to_rtxlen(mp, align);
1403 
1404 	if (*prod > 1)
1405 		xfs_rtalloc_align_minmax(raminlen, ralen, prod);
1406 	return 0;
1407 }
1408 
1409 int
1410 xfs_bmap_rtalloc(
1411 	struct xfs_bmalloca	*ap)
1412 {
1413 	xfs_fileoff_t		orig_offset = ap->offset;
1414 	xfs_rtxlen_t		prod = 0;  /* product factor for allocators */
1415 	xfs_rtxlen_t		ralen = 0; /* realtime allocation length */
1416 	xfs_rtblock_t		bno_hint = NULLRTBLOCK;
1417 	xfs_extlen_t		orig_length = ap->length;
1418 	xfs_rtxlen_t		raminlen;
1419 	bool			rtlocked = false;
1420 	bool			noalign = false;
1421 	bool			initial_user_data =
1422 		ap->datatype & XFS_ALLOC_INITIAL_USER_DATA;
1423 	int			error;
1424 
1425 retry:
1426 	error = xfs_rtallocate_align(ap, &ralen, &raminlen, &prod, &noalign);
1427 	if (error)
1428 		return error;
1429 
1430 	if (xfs_bmap_adjacent(ap))
1431 		bno_hint = ap->blkno;
1432 
1433 	error = xfs_rtallocate(ap->tp, bno_hint, raminlen, ralen, prod,
1434 			ap->wasdel, initial_user_data, &rtlocked,
1435 			&ap->blkno, &ap->length);
1436 	if (error == -ENOSPC) {
1437 		if (!noalign) {
1438 			/*
1439 			 * We previously enlarged the request length to try to
1440 			 * satisfy an extent size hint.  The allocator didn't
1441 			 * return anything, so reset the parameters to the
1442 			 * original values and try again without alignment
1443 			 * criteria.
1444 			 */
1445 			ap->offset = orig_offset;
1446 			ap->length = orig_length;
1447 			noalign = true;
1448 			goto retry;
1449 		}
1450 
1451 		ap->blkno = NULLFSBLOCK;
1452 		ap->length = 0;
1453 		return 0;
1454 	}
1455 	if (error)
1456 		return error;
1457 
1458 	xfs_bmap_alloc_account(ap);
1459 	return 0;
1460 }
1461