xref: /linux/fs/xfs/xfs_rtalloc.c (revision 03c305861c70d6db898dd2379b882e7772a5c5d0)
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_trans.h"
18 #include "xfs_trans_space.h"
19 #include "xfs_icache.h"
20 #include "xfs_rtalloc.h"
21 #include "xfs_sb.h"
22 #include "xfs_rtbitmap.h"
23 
24 /*
25  * Read and return the summary information for a given extent size,
26  * bitmap block combination.
27  * Keeps track of a current summary block, so we don't keep reading
28  * it from the buffer cache.
29  */
30 static int
31 xfs_rtget_summary(
32 	struct xfs_rtalloc_args	*args,
33 	int			log,	/* log2 of extent size */
34 	xfs_fileoff_t		bbno,	/* bitmap block number */
35 	xfs_suminfo_t		*sum)	/* out: summary info for this block */
36 {
37 	return xfs_rtmodify_summary_int(args, log, bbno, 0, sum);
38 }
39 
40 /*
41  * Return whether there are any free extents in the size range given
42  * by low and high, for the bitmap block bbno.
43  */
44 STATIC int
45 xfs_rtany_summary(
46 	struct xfs_rtalloc_args	*args,
47 	int			low,	/* low log2 extent size */
48 	int			high,	/* high log2 extent size */
49 	xfs_fileoff_t		bbno,	/* bitmap block number */
50 	int			*maxlog) /* out: max log2 extent size free */
51 {
52 	struct xfs_mount	*mp = args->mp;
53 	int			error;
54 	int			log;	/* loop counter, log2 of ext. size */
55 	xfs_suminfo_t		sum;	/* summary data */
56 
57 	/* There are no extents at levels >= m_rsum_cache[bbno]. */
58 	if (mp->m_rsum_cache) {
59 		high = min(high, mp->m_rsum_cache[bbno] - 1);
60 		if (low > high) {
61 			*maxlog = -1;
62 			return 0;
63 		}
64 	}
65 
66 	/*
67 	 * Loop over logs of extent sizes.
68 	 */
69 	for (log = high; log >= low; log--) {
70 		/*
71 		 * Get one summary datum.
72 		 */
73 		error = xfs_rtget_summary(args, log, bbno, &sum);
74 		if (error) {
75 			return error;
76 		}
77 		/*
78 		 * If there are any, return success.
79 		 */
80 		if (sum) {
81 			*maxlog = log;
82 			goto out;
83 		}
84 	}
85 	/*
86 	 * Found nothing, return failure.
87 	 */
88 	*maxlog = -1;
89 out:
90 	/* There were no extents at levels > log. */
91 	if (mp->m_rsum_cache && log + 1 < mp->m_rsum_cache[bbno])
92 		mp->m_rsum_cache[bbno] = log + 1;
93 	return 0;
94 }
95 
96 
97 /*
98  * Copy and transform the summary file, given the old and new
99  * parameters in the mount structures.
100  */
101 STATIC int
102 xfs_rtcopy_summary(
103 	struct xfs_rtalloc_args	*oargs,
104 	struct xfs_rtalloc_args	*nargs)
105 {
106 	xfs_fileoff_t		bbno;	/* bitmap block number */
107 	int			error;
108 	int			log;	/* summary level number (log length) */
109 	xfs_suminfo_t		sum;	/* summary data */
110 
111 	for (log = oargs->mp->m_rsumlevels - 1; log >= 0; log--) {
112 		for (bbno = oargs->mp->m_sb.sb_rbmblocks - 1;
113 		     (xfs_srtblock_t)bbno >= 0;
114 		     bbno--) {
115 			error = xfs_rtget_summary(oargs, log, bbno, &sum);
116 			if (error)
117 				goto out;
118 			if (sum == 0)
119 				continue;
120 			error = xfs_rtmodify_summary(oargs, log, bbno, -sum);
121 			if (error)
122 				goto out;
123 			error = xfs_rtmodify_summary(nargs, log, bbno, sum);
124 			if (error)
125 				goto out;
126 			ASSERT(sum > 0);
127 		}
128 	}
129 	error = 0;
130 out:
131 	xfs_rtbuf_cache_relse(oargs);
132 	return 0;
133 }
134 /*
135  * Mark an extent specified by start and len allocated.
136  * Updates all the summary information as well as the bitmap.
137  */
138 STATIC int
139 xfs_rtallocate_range(
140 	struct xfs_rtalloc_args	*args,
141 	xfs_rtxnum_t		start,	/* start rtext to allocate */
142 	xfs_rtxlen_t		len)	/* in/out: summary block number */
143 {
144 	struct xfs_mount	*mp = args->mp;
145 	xfs_rtxnum_t		end;	/* end of the allocated rtext */
146 	int			error;
147 	xfs_rtxnum_t		postblock = 0; /* first rtext allocated > end */
148 	xfs_rtxnum_t		preblock = 0; /* first rtext allocated < start */
149 
150 	end = start + len - 1;
151 	/*
152 	 * Assume we're allocating out of the middle of a free extent.
153 	 * We need to find the beginning and end of the extent so we can
154 	 * properly update the summary.
155 	 */
156 	error = xfs_rtfind_back(args, start, 0, &preblock);
157 	if (error) {
158 		return error;
159 	}
160 	/*
161 	 * Find the next allocated block (end of free extent).
162 	 */
163 	error = xfs_rtfind_forw(args, end, mp->m_sb.sb_rextents - 1,
164 			&postblock);
165 	if (error) {
166 		return error;
167 	}
168 	/*
169 	 * Decrement the summary information corresponding to the entire
170 	 * (old) free extent.
171 	 */
172 	error = xfs_rtmodify_summary(args,
173 			XFS_RTBLOCKLOG(postblock + 1 - preblock),
174 			xfs_rtx_to_rbmblock(mp, preblock), -1);
175 	if (error) {
176 		return error;
177 	}
178 	/*
179 	 * If there are blocks not being allocated at the front of the
180 	 * old extent, add summary data for them to be free.
181 	 */
182 	if (preblock < start) {
183 		error = xfs_rtmodify_summary(args,
184 				XFS_RTBLOCKLOG(start - preblock),
185 				xfs_rtx_to_rbmblock(mp, preblock), 1);
186 		if (error) {
187 			return error;
188 		}
189 	}
190 	/*
191 	 * If there are blocks not being allocated at the end of the
192 	 * old extent, add summary data for them to be free.
193 	 */
194 	if (postblock > end) {
195 		error = xfs_rtmodify_summary(args,
196 				XFS_RTBLOCKLOG(postblock - end),
197 				xfs_rtx_to_rbmblock(mp, end + 1), 1);
198 		if (error) {
199 			return error;
200 		}
201 	}
202 	/*
203 	 * Modify the bitmap to mark this extent allocated.
204 	 */
205 	error = xfs_rtmodify_range(args, start, len, 0);
206 	return error;
207 }
208 
209 /*
210  * Make sure we don't run off the end of the rt volume.  Be careful that
211  * adjusting maxlen downwards doesn't cause us to fail the alignment checks.
212  */
213 static inline xfs_rtxlen_t
214 xfs_rtallocate_clamp_len(
215 	struct xfs_mount	*mp,
216 	xfs_rtxnum_t		startrtx,
217 	xfs_rtxlen_t		rtxlen,
218 	xfs_rtxlen_t		prod)
219 {
220 	xfs_rtxlen_t		ret;
221 
222 	ret = min(mp->m_sb.sb_rextents, startrtx + rtxlen) - startrtx;
223 	return rounddown(ret, prod);
224 }
225 
226 /*
227  * Attempt to allocate an extent minlen<=len<=maxlen starting from
228  * bitmap block bbno.  If we don't get maxlen then use prod to trim
229  * the length, if given.  Returns error; returns starting block in *rtx.
230  * The lengths are all in rtextents.
231  */
232 STATIC int
233 xfs_rtallocate_extent_block(
234 	struct xfs_rtalloc_args	*args,
235 	xfs_fileoff_t		bbno,	/* bitmap block number */
236 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
237 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
238 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
239 	xfs_rtxnum_t		*nextp,	/* out: next rtext to try */
240 	xfs_rtxlen_t		prod,	/* extent product factor */
241 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
242 {
243 	struct xfs_mount	*mp = args->mp;
244 	xfs_rtxnum_t		besti;	/* best rtext found so far */
245 	xfs_rtxnum_t		bestlen;/* best length found so far */
246 	xfs_rtxnum_t		end;	/* last rtext in chunk */
247 	int			error;
248 	xfs_rtxnum_t		i;	/* current rtext trying */
249 	xfs_rtxnum_t		next;	/* next rtext to try */
250 	int			stat;	/* status from internal calls */
251 
252 	/*
253 	 * Loop over all the extents starting in this bitmap block,
254 	 * looking for one that's long enough.
255 	 */
256 	for (i = xfs_rbmblock_to_rtx(mp, bbno), besti = -1, bestlen = 0,
257 		end = xfs_rbmblock_to_rtx(mp, bbno + 1) - 1;
258 	     i <= end;
259 	     i++) {
260 		/* Make sure we don't scan off the end of the rt volume. */
261 		maxlen = xfs_rtallocate_clamp_len(mp, i, maxlen, prod);
262 
263 		/*
264 		 * See if there's a free extent of maxlen 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, maxlen, 1, &next, &stat);
268 		if (error) {
269 			return error;
270 		}
271 		if (stat) {
272 			/*
273 			 * i for maxlen is all free, allocate and return that.
274 			 */
275 			error = xfs_rtallocate_range(args, i, maxlen);
276 			if (error) {
277 				return error;
278 			}
279 			*len = maxlen;
280 			*rtx = i;
281 			return 0;
282 		}
283 		/*
284 		 * In the case where we have a variable-sized allocation
285 		 * request, figure out how big this free piece is,
286 		 * and if it's big enough for the minimum, and the best
287 		 * so far, remember it.
288 		 */
289 		if (minlen < maxlen) {
290 			xfs_rtxnum_t	thislen;	/* this extent size */
291 
292 			thislen = next - i;
293 			if (thislen >= minlen && thislen > bestlen) {
294 				besti = i;
295 				bestlen = thislen;
296 			}
297 		}
298 		/*
299 		 * If not done yet, find the start of the next free space.
300 		 */
301 		if (next < end) {
302 			error = xfs_rtfind_forw(args, next, end, &i);
303 			if (error) {
304 				return error;
305 			}
306 		} else
307 			break;
308 	}
309 	/*
310 	 * Searched the whole thing & didn't find a maxlen free extent.
311 	 */
312 	if (minlen < maxlen && besti != -1) {
313 		xfs_rtxlen_t	p;	/* amount to trim length by */
314 
315 		/*
316 		 * If size should be a multiple of prod, make that so.
317 		 */
318 		if (prod > 1) {
319 			div_u64_rem(bestlen, prod, &p);
320 			if (p)
321 				bestlen -= p;
322 		}
323 
324 		/*
325 		 * Allocate besti for bestlen & return that.
326 		 */
327 		error = xfs_rtallocate_range(args, besti, bestlen);
328 		if (error) {
329 			return error;
330 		}
331 		*len = bestlen;
332 		*rtx = besti;
333 		return 0;
334 	}
335 	/*
336 	 * Allocation failed.  Set *nextp to the next block to try.
337 	 */
338 	*nextp = next;
339 	*rtx = NULLRTEXTNO;
340 	return 0;
341 }
342 
343 /*
344  * Allocate an extent of length minlen<=len<=maxlen, starting at block
345  * bno.  If we don't get maxlen then use prod to trim the length, if given.
346  * Returns error; returns starting block in *rtx.
347  * The lengths are all in rtextents.
348  */
349 STATIC int
350 xfs_rtallocate_extent_exact(
351 	struct xfs_rtalloc_args	*args,
352 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
353 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
354 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
355 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
356 	xfs_rtxlen_t		prod,	/* extent product factor */
357 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
358 {
359 	int			error;
360 	xfs_rtxlen_t		i;	/* extent length trimmed due to prod */
361 	int			isfree;	/* extent is free */
362 	xfs_rtxnum_t		next;	/* next rtext to try (dummy) */
363 
364 	ASSERT(minlen % prod == 0);
365 	ASSERT(maxlen % prod == 0);
366 	/*
367 	 * Check if the range in question (for maxlen) is free.
368 	 */
369 	error = xfs_rtcheck_range(args, start, maxlen, 1, &next, &isfree);
370 	if (error) {
371 		return error;
372 	}
373 	if (isfree) {
374 		/*
375 		 * If it is, allocate it and return success.
376 		 */
377 		error = xfs_rtallocate_range(args, start, maxlen);
378 		if (error) {
379 			return error;
380 		}
381 		*len = maxlen;
382 		*rtx = start;
383 		return 0;
384 	}
385 	/*
386 	 * If not, allocate what there is, if it's at least minlen.
387 	 */
388 	maxlen = next - start;
389 	if (maxlen < minlen) {
390 		/*
391 		 * Failed, return failure status.
392 		 */
393 		*rtx = NULLRTEXTNO;
394 		return 0;
395 	}
396 	/*
397 	 * Trim off tail of extent, if prod is specified.
398 	 */
399 	if (prod > 1 && (i = maxlen % prod)) {
400 		maxlen -= i;
401 		if (maxlen < minlen) {
402 			/*
403 			 * Now we can't do it, return failure status.
404 			 */
405 			*rtx = NULLRTEXTNO;
406 			return 0;
407 		}
408 	}
409 	/*
410 	 * Allocate what we can and return it.
411 	 */
412 	error = xfs_rtallocate_range(args, start, maxlen);
413 	if (error) {
414 		return error;
415 	}
416 	*len = maxlen;
417 	*rtx = start;
418 	return 0;
419 }
420 
421 /*
422  * Allocate an extent of length minlen<=len<=maxlen, starting as near
423  * to start as possible.  If we don't get maxlen then use prod to trim
424  * the length, if given.  The lengths are all in rtextents.
425  */
426 STATIC int
427 xfs_rtallocate_extent_near(
428 	struct xfs_rtalloc_args	*args,
429 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
430 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
431 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
432 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
433 	xfs_rtxlen_t		prod,	/* extent product factor */
434 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
435 {
436 	struct xfs_mount	*mp = args->mp;
437 	int			maxlog;	/* max useful extent from summary */
438 	xfs_fileoff_t		bbno;	/* bitmap block number */
439 	int			error;
440 	int			i;	/* bitmap block offset (loop control) */
441 	int			j;	/* secondary loop control */
442 	int			log2len; /* log2 of minlen */
443 	xfs_rtxnum_t		n;	/* next rtext to try */
444 	xfs_rtxnum_t		r;	/* result rtext */
445 
446 	ASSERT(minlen % prod == 0);
447 	ASSERT(maxlen % prod == 0);
448 
449 	/*
450 	 * If the block number given is off the end, silently set it to
451 	 * the last block.
452 	 */
453 	if (start >= mp->m_sb.sb_rextents)
454 		start = mp->m_sb.sb_rextents - 1;
455 
456 	/* Make sure we don't run off the end of the rt volume. */
457 	maxlen = xfs_rtallocate_clamp_len(mp, start, maxlen, prod);
458 	if (maxlen < minlen) {
459 		*rtx = NULLRTEXTNO;
460 		return 0;
461 	}
462 
463 	/*
464 	 * Try the exact allocation first.
465 	 */
466 	error = xfs_rtallocate_extent_exact(args, start, minlen, maxlen, len,
467 			prod, &r);
468 	if (error) {
469 		return error;
470 	}
471 	/*
472 	 * If the exact allocation worked, return that.
473 	 */
474 	if (r != NULLRTEXTNO) {
475 		*rtx = r;
476 		return 0;
477 	}
478 	bbno = xfs_rtx_to_rbmblock(mp, start);
479 	i = 0;
480 	j = -1;
481 	ASSERT(minlen != 0);
482 	log2len = xfs_highbit32(minlen);
483 	/*
484 	 * Loop over all bitmap blocks (bbno + i is current block).
485 	 */
486 	for (;;) {
487 		/*
488 		 * Get summary information of extents of all useful levels
489 		 * starting in this bitmap block.
490 		 */
491 		error = xfs_rtany_summary(args, log2len, mp->m_rsumlevels - 1,
492 				bbno + i, &maxlog);
493 		if (error) {
494 			return error;
495 		}
496 		/*
497 		 * If there are any useful extents starting here, try
498 		 * allocating one.
499 		 */
500 		if (maxlog >= 0) {
501 			xfs_extlen_t maxavail =
502 				min_t(xfs_rtblock_t, maxlen,
503 				      (1ULL << (maxlog + 1)) - 1);
504 			/*
505 			 * On the positive side of the starting location.
506 			 */
507 			if (i >= 0) {
508 				/*
509 				 * Try to allocate an extent starting in
510 				 * this block.
511 				 */
512 				error = xfs_rtallocate_extent_block(args,
513 						bbno + i, minlen, maxavail, len,
514 						&n, prod, &r);
515 				if (error) {
516 					return error;
517 				}
518 				/*
519 				 * If it worked, return it.
520 				 */
521 				if (r != NULLRTEXTNO) {
522 					*rtx = r;
523 					return 0;
524 				}
525 			}
526 			/*
527 			 * On the negative side of the starting location.
528 			 */
529 			else {		/* i < 0 */
530 				int	maxblocks;
531 
532 				/*
533 				 * Loop backwards to find the end of the extent
534 				 * we found in the realtime summary.
535 				 *
536 				 * maxblocks is the maximum possible number of
537 				 * bitmap blocks from the start of the extent
538 				 * to the end of the extent.
539 				 */
540 				if (maxlog == 0)
541 					maxblocks = 0;
542 				else if (maxlog < mp->m_blkbit_log)
543 					maxblocks = 1;
544 				else
545 					maxblocks = 2 << (maxlog - mp->m_blkbit_log);
546 
547 				/*
548 				 * We need to check bbno + i + maxblocks down to
549 				 * bbno + i. We already checked bbno down to
550 				 * bbno + j + 1, so we don't need to check those
551 				 * again.
552 				 */
553 				j = min(i + maxblocks, j);
554 				for (; j >= i; j--) {
555 					error = xfs_rtallocate_extent_block(args,
556 							bbno + j, minlen,
557 							maxavail, len, &n, prod,
558 							&r);
559 					if (error) {
560 						return error;
561 					}
562 					/*
563 					 * If it works, return the extent.
564 					 */
565 					if (r != NULLRTEXTNO) {
566 						*rtx = r;
567 						return 0;
568 					}
569 				}
570 			}
571 		}
572 		/*
573 		 * Loop control.  If we were on the positive side, and there's
574 		 * still more blocks on the negative side, go there.
575 		 */
576 		if (i > 0 && (int)bbno - i >= 0)
577 			i = -i;
578 		/*
579 		 * If positive, and no more negative, but there are more
580 		 * positive, go there.
581 		 */
582 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
583 			i++;
584 		/*
585 		 * If negative or 0 (just started), and there are positive
586 		 * blocks to go, go there.  The 0 case moves to block 1.
587 		 */
588 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
589 			i = 1 - i;
590 		/*
591 		 * If negative or 0 and there are more negative blocks,
592 		 * go there.
593 		 */
594 		else if (i <= 0 && (int)bbno + i > 0)
595 			i--;
596 		/*
597 		 * Must be done.  Return failure.
598 		 */
599 		else
600 			break;
601 	}
602 	*rtx = NULLRTEXTNO;
603 	return 0;
604 }
605 
606 /*
607  * Allocate an extent of length minlen<=len<=maxlen, with no position
608  * specified.  If we don't get maxlen then use prod to trim
609  * the length, if given.  The lengths are all in rtextents.
610  */
611 STATIC int
612 xfs_rtallocate_extent_size(
613 	struct xfs_rtalloc_args	*args,
614 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
615 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
616 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
617 	xfs_rtxlen_t		prod,	/* extent product factor */
618 	xfs_rtxnum_t		*rtx)	/* out: start rtext allocated */
619 {
620 	struct xfs_mount	*mp = args->mp;
621 	int			error;
622 	xfs_fileoff_t		i;	/* bitmap block number */
623 	int			l;	/* level number (loop control) */
624 	xfs_rtxnum_t		n;	/* next rtext to be tried */
625 	xfs_rtxnum_t		r;	/* result rtext number */
626 	xfs_suminfo_t		sum;	/* summary information for extents */
627 
628 	ASSERT(minlen % prod == 0);
629 	ASSERT(maxlen % prod == 0);
630 	ASSERT(maxlen != 0);
631 
632 	/*
633 	 * Loop over all the levels starting with maxlen.
634 	 * At each level, look at all the bitmap blocks, to see if there
635 	 * are extents starting there that are long enough (>= maxlen).
636 	 * Note, only on the initial level can the allocation fail if
637 	 * the summary says there's an extent.
638 	 */
639 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
640 		/*
641 		 * Loop over all the bitmap blocks.
642 		 */
643 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
644 			/*
645 			 * Get the summary for this level/block.
646 			 */
647 			error = xfs_rtget_summary(args, l, i, &sum);
648 			if (error) {
649 				return error;
650 			}
651 			/*
652 			 * Nothing there, on to the next block.
653 			 */
654 			if (!sum)
655 				continue;
656 			/*
657 			 * Try allocating the extent.
658 			 */
659 			error = xfs_rtallocate_extent_block(args, i, maxlen,
660 					maxlen, len, &n, prod, &r);
661 			if (error) {
662 				return error;
663 			}
664 			/*
665 			 * If it worked, return that.
666 			 */
667 			if (r != NULLRTEXTNO) {
668 				*rtx = r;
669 				return 0;
670 			}
671 			/*
672 			 * If the "next block to try" returned from the
673 			 * allocator is beyond the next bitmap block,
674 			 * skip to that bitmap block.
675 			 */
676 			if (xfs_rtx_to_rbmblock(mp, n) > i + 1)
677 				i = xfs_rtx_to_rbmblock(mp, n) - 1;
678 		}
679 	}
680 	/*
681 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
682 	 * we're asking for a fixed size extent.
683 	 */
684 	if (minlen > --maxlen) {
685 		*rtx = NULLRTEXTNO;
686 		return 0;
687 	}
688 	ASSERT(minlen != 0);
689 	ASSERT(maxlen != 0);
690 
691 	/*
692 	 * Loop over sizes, from maxlen down to minlen.
693 	 * This time, when we do the allocations, allow smaller ones
694 	 * to succeed.
695 	 */
696 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
697 		/*
698 		 * Loop over all the bitmap blocks, try an allocation
699 		 * starting in that block.
700 		 */
701 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
702 			/*
703 			 * Get the summary information for this level/block.
704 			 */
705 			error =	xfs_rtget_summary(args, l, i, &sum);
706 			if (error) {
707 				return error;
708 			}
709 			/*
710 			 * If nothing there, go on to next.
711 			 */
712 			if (!sum)
713 				continue;
714 			/*
715 			 * Try the allocation.  Make sure the specified
716 			 * minlen/maxlen are in the possible range for
717 			 * this summary level.
718 			 */
719 			error = xfs_rtallocate_extent_block(args, i,
720 					XFS_RTMAX(minlen, 1 << l),
721 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
722 					len, &n, prod, &r);
723 			if (error) {
724 				return error;
725 			}
726 			/*
727 			 * If it worked, return that extent.
728 			 */
729 			if (r != NULLRTEXTNO) {
730 				*rtx = r;
731 				return 0;
732 			}
733 			/*
734 			 * If the "next block to try" returned from the
735 			 * allocator is beyond the next bitmap block,
736 			 * skip to that bitmap block.
737 			 */
738 			if (xfs_rtx_to_rbmblock(mp, n) > i + 1)
739 				i = xfs_rtx_to_rbmblock(mp, n) - 1;
740 		}
741 	}
742 	/*
743 	 * Got nothing, return failure.
744 	 */
745 	*rtx = NULLRTEXTNO;
746 	return 0;
747 }
748 
749 /*
750  * Allocate space to the bitmap or summary file, and zero it, for growfs.
751  */
752 STATIC int
753 xfs_growfs_rt_alloc(
754 	struct xfs_mount	*mp,		/* file system mount point */
755 	xfs_extlen_t		oblocks,	/* old count of blocks */
756 	xfs_extlen_t		nblocks,	/* new count of blocks */
757 	struct xfs_inode	*ip)		/* inode (bitmap/summary) */
758 {
759 	xfs_fileoff_t		bno;		/* block number in file */
760 	struct xfs_buf		*bp;	/* temporary buffer for zeroing */
761 	xfs_daddr_t		d;		/* disk block address */
762 	int			error;		/* error return value */
763 	xfs_fsblock_t		fsbno;		/* filesystem block for bno */
764 	struct xfs_bmbt_irec	map;		/* block map output */
765 	int			nmap;		/* number of block maps */
766 	int			resblks;	/* space reservation */
767 	enum xfs_blft		buf_type;
768 	struct xfs_trans	*tp;
769 
770 	if (ip == mp->m_rsumip)
771 		buf_type = XFS_BLFT_RTSUMMARY_BUF;
772 	else
773 		buf_type = XFS_BLFT_RTBITMAP_BUF;
774 
775 	/*
776 	 * Allocate space to the file, as necessary.
777 	 */
778 	while (oblocks < nblocks) {
779 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
780 		/*
781 		 * Reserve space & log for one extent added to the file.
782 		 */
783 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtalloc, resblks,
784 				0, 0, &tp);
785 		if (error)
786 			return error;
787 		/*
788 		 * Lock the inode.
789 		 */
790 		xfs_ilock(ip, XFS_ILOCK_EXCL);
791 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
792 
793 		error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
794 				XFS_IEXT_ADD_NOSPLIT_CNT);
795 		if (error == -EFBIG)
796 			error = xfs_iext_count_upgrade(tp, ip,
797 					XFS_IEXT_ADD_NOSPLIT_CNT);
798 		if (error)
799 			goto out_trans_cancel;
800 
801 		/*
802 		 * Allocate blocks to the bitmap file.
803 		 */
804 		nmap = 1;
805 		error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
806 					XFS_BMAPI_METADATA, 0, &map, &nmap);
807 		if (!error && nmap < 1)
808 			error = -ENOSPC;
809 		if (error)
810 			goto out_trans_cancel;
811 		/*
812 		 * Free any blocks freed up in the transaction, then commit.
813 		 */
814 		error = xfs_trans_commit(tp);
815 		if (error)
816 			return error;
817 		/*
818 		 * Now we need to clear the allocated blocks.
819 		 * Do this one block per transaction, to keep it simple.
820 		 */
821 		for (bno = map.br_startoff, fsbno = map.br_startblock;
822 		     bno < map.br_startoff + map.br_blockcount;
823 		     bno++, fsbno++) {
824 			/*
825 			 * Reserve log for one block zeroing.
826 			 */
827 			error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtzero,
828 					0, 0, 0, &tp);
829 			if (error)
830 				return error;
831 			/*
832 			 * Lock the bitmap inode.
833 			 */
834 			xfs_ilock(ip, XFS_ILOCK_EXCL);
835 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
836 			/*
837 			 * Get a buffer for the block.
838 			 */
839 			d = XFS_FSB_TO_DADDR(mp, fsbno);
840 			error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
841 					mp->m_bsize, 0, &bp);
842 			if (error)
843 				goto out_trans_cancel;
844 
845 			xfs_trans_buf_set_type(tp, bp, buf_type);
846 			bp->b_ops = &xfs_rtbuf_ops;
847 			memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
848 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
849 			/*
850 			 * Commit the transaction.
851 			 */
852 			error = xfs_trans_commit(tp);
853 			if (error)
854 				return error;
855 		}
856 		/*
857 		 * Go on to the next extent, if any.
858 		 */
859 		oblocks = map.br_startoff + map.br_blockcount;
860 	}
861 
862 	return 0;
863 
864 out_trans_cancel:
865 	xfs_trans_cancel(tp);
866 	return error;
867 }
868 
869 static void
870 xfs_alloc_rsum_cache(
871 	xfs_mount_t	*mp,		/* file system mount structure */
872 	xfs_extlen_t	rbmblocks)	/* number of rt bitmap blocks */
873 {
874 	/*
875 	 * The rsum cache is initialized to the maximum value, which is
876 	 * trivially an upper bound on the maximum level with any free extents.
877 	 * We can continue without the cache if it couldn't be allocated.
878 	 */
879 	mp->m_rsum_cache = kvmalloc(rbmblocks, GFP_KERNEL);
880 	if (mp->m_rsum_cache)
881 		memset(mp->m_rsum_cache, -1, rbmblocks);
882 	else
883 		xfs_warn(mp, "could not allocate realtime summary cache");
884 }
885 
886 /*
887  * Visible (exported) functions.
888  */
889 
890 /*
891  * Grow the realtime area of the filesystem.
892  */
893 int
894 xfs_growfs_rt(
895 	xfs_mount_t	*mp,		/* mount point for filesystem */
896 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
897 {
898 	xfs_fileoff_t	bmbno;		/* bitmap block number */
899 	struct xfs_buf	*bp;		/* temporary buffer */
900 	int		error;		/* error return value */
901 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
902 	xfs_rfsblock_t	nrblocks;	/* new number of realtime blocks */
903 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
904 	xfs_rtxnum_t	nrextents;	/* new number of realtime extents */
905 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
906 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
907 	uint		nrsumlevels;	/* new rt summary levels */
908 	uint		nrsumsize;	/* new size of rt summary, bytes */
909 	xfs_sb_t	*nsbp;		/* new superblock */
910 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
911 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
912 	xfs_sb_t	*sbp;		/* old superblock */
913 	uint8_t		*rsum_cache;	/* old summary cache */
914 
915 	sbp = &mp->m_sb;
916 
917 	if (!capable(CAP_SYS_ADMIN))
918 		return -EPERM;
919 
920 	/* Needs to have been mounted with an rt device. */
921 	if (!XFS_IS_REALTIME_MOUNT(mp))
922 		return -EINVAL;
923 	/*
924 	 * Mount should fail if the rt bitmap/summary files don't load, but
925 	 * we'll check anyway.
926 	 */
927 	if (!mp->m_rbmip || !mp->m_rsumip)
928 		return -EINVAL;
929 
930 	/* Shrink not supported. */
931 	if (in->newblocks <= sbp->sb_rblocks)
932 		return -EINVAL;
933 
934 	/* Can only change rt extent size when adding rt volume. */
935 	if (sbp->sb_rblocks > 0 && in->extsize != sbp->sb_rextsize)
936 		return -EINVAL;
937 
938 	/* Range check the extent size. */
939 	if (XFS_FSB_TO_B(mp, in->extsize) > XFS_MAX_RTEXTSIZE ||
940 	    XFS_FSB_TO_B(mp, in->extsize) < XFS_MIN_RTEXTSIZE)
941 		return -EINVAL;
942 
943 	/* Unsupported realtime features. */
944 	if (xfs_has_rmapbt(mp) || xfs_has_reflink(mp) || xfs_has_quota(mp))
945 		return -EOPNOTSUPP;
946 
947 	nrblocks = in->newblocks;
948 	error = xfs_sb_validate_fsb_count(sbp, nrblocks);
949 	if (error)
950 		return error;
951 	/*
952 	 * Read in the last block of the device, make sure it exists.
953 	 */
954 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
955 				XFS_FSB_TO_BB(mp, nrblocks - 1),
956 				XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
957 	if (error)
958 		return error;
959 	xfs_buf_relse(bp);
960 
961 	/*
962 	 * Calculate new parameters.  These are the final values to be reached.
963 	 */
964 	nrextents = nrblocks;
965 	do_div(nrextents, in->extsize);
966 	nrbmblocks = xfs_rtbitmap_blockcount(mp, nrextents);
967 	nrextslog = xfs_highbit32(nrextents);
968 	nrsumlevels = nrextslog + 1;
969 	nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels, nrbmblocks);
970 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
971 	/*
972 	 * New summary size can't be more than half the size of
973 	 * the log.  This prevents us from getting a log overflow,
974 	 * since we'll log basically the whole summary file at once.
975 	 */
976 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
977 		return -EINVAL;
978 	/*
979 	 * Get the old block counts for bitmap and summary inodes.
980 	 * These can't change since other growfs callers are locked out.
981 	 */
982 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_disk_size);
983 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_disk_size);
984 	/*
985 	 * Allocate space to the bitmap and summary files, as necessary.
986 	 */
987 	error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
988 	if (error)
989 		return error;
990 	error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
991 	if (error)
992 		return error;
993 
994 	rsum_cache = mp->m_rsum_cache;
995 	if (nrbmblocks != sbp->sb_rbmblocks)
996 		xfs_alloc_rsum_cache(mp, nrbmblocks);
997 
998 	/*
999 	 * Allocate a new (fake) mount/sb.
1000 	 */
1001 	nmp = kmem_alloc(sizeof(*nmp), 0);
1002 	/*
1003 	 * Loop over the bitmap blocks.
1004 	 * We will do everything one bitmap block at a time.
1005 	 * Skip the current block if it is exactly full.
1006 	 * This also deals with the case where there were no rtextents before.
1007 	 */
1008 	for (bmbno = sbp->sb_rbmblocks -
1009 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1010 	     bmbno < nrbmblocks;
1011 	     bmbno++) {
1012 		struct xfs_rtalloc_args	args = {
1013 			.mp		= mp,
1014 		};
1015 		struct xfs_rtalloc_args	nargs = {
1016 			.mp		= nmp,
1017 		};
1018 		struct xfs_trans	*tp;
1019 		xfs_rfsblock_t		nrblocks_step;
1020 
1021 		*nmp = *mp;
1022 		nsbp = &nmp->m_sb;
1023 		/*
1024 		 * Calculate new sb and mount fields for this round.
1025 		 */
1026 		nsbp->sb_rextsize = in->extsize;
1027 		nmp->m_rtxblklog = -1; /* don't use shift or masking */
1028 		nsbp->sb_rbmblocks = bmbno + 1;
1029 		nrblocks_step = (bmbno + 1) * NBBY * nsbp->sb_blocksize *
1030 				nsbp->sb_rextsize;
1031 		nsbp->sb_rblocks = min(nrblocks, nrblocks_step);
1032 		nsbp->sb_rextents = xfs_rtb_to_rtx(nmp, nsbp->sb_rblocks);
1033 		ASSERT(nsbp->sb_rextents != 0);
1034 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1035 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1036 		nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels,
1037 				nsbp->sb_rbmblocks);
1038 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1039 		/*
1040 		 * Start a transaction, get the log reservation.
1041 		 */
1042 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtfree, 0, 0, 0,
1043 				&tp);
1044 		if (error)
1045 			break;
1046 		args.tp = tp;
1047 		nargs.tp = tp;
1048 
1049 		/*
1050 		 * Lock out other callers by grabbing the bitmap inode lock.
1051 		 */
1052 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL | XFS_ILOCK_RTBITMAP);
1053 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
1054 		/*
1055 		 * Update the bitmap inode's size ondisk and incore.  We need
1056 		 * to update the incore size so that inode inactivation won't
1057 		 * punch what it thinks are "posteof" blocks.
1058 		 */
1059 		mp->m_rbmip->i_disk_size =
1060 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1061 		i_size_write(VFS_I(mp->m_rbmip), mp->m_rbmip->i_disk_size);
1062 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1063 		/*
1064 		 * Get the summary inode into the transaction.
1065 		 */
1066 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL | XFS_ILOCK_RTSUM);
1067 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
1068 		/*
1069 		 * Update the summary inode's size.  We need to update the
1070 		 * incore size so that inode inactivation won't punch what it
1071 		 * thinks are "posteof" blocks.
1072 		 */
1073 		mp->m_rsumip->i_disk_size = nmp->m_rsumsize;
1074 		i_size_write(VFS_I(mp->m_rsumip), mp->m_rsumip->i_disk_size);
1075 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
1076 		/*
1077 		 * Copy summary data from old to new sizes.
1078 		 * Do this when the real size (not block-aligned) changes.
1079 		 */
1080 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
1081 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
1082 			error = xfs_rtcopy_summary(&args, &nargs);
1083 			if (error)
1084 				goto error_cancel;
1085 		}
1086 		/*
1087 		 * Update superblock fields.
1088 		 */
1089 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
1090 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
1091 				nsbp->sb_rextsize - sbp->sb_rextsize);
1092 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
1093 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
1094 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
1095 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
1096 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
1097 				nsbp->sb_rblocks - sbp->sb_rblocks);
1098 		if (nsbp->sb_rextents != sbp->sb_rextents)
1099 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
1100 				nsbp->sb_rextents - sbp->sb_rextents);
1101 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
1102 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
1103 				nsbp->sb_rextslog - sbp->sb_rextslog);
1104 		/*
1105 		 * Free new extent.
1106 		 */
1107 		error = xfs_rtfree_range(&nargs, sbp->sb_rextents,
1108 				nsbp->sb_rextents - sbp->sb_rextents);
1109 		xfs_rtbuf_cache_relse(&nargs);
1110 		if (error) {
1111 error_cancel:
1112 			xfs_trans_cancel(tp);
1113 			break;
1114 		}
1115 		/*
1116 		 * Mark more blocks free in the superblock.
1117 		 */
1118 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
1119 			nsbp->sb_rextents - sbp->sb_rextents);
1120 		/*
1121 		 * Update mp values into the real mp structure.
1122 		 */
1123 		mp->m_rsumlevels = nrsumlevels;
1124 		mp->m_rsumsize = nrsumsize;
1125 
1126 		error = xfs_trans_commit(tp);
1127 		if (error)
1128 			break;
1129 
1130 		/* Ensure the mount RT feature flag is now set. */
1131 		mp->m_features |= XFS_FEAT_REALTIME;
1132 	}
1133 	if (error)
1134 		goto out_free;
1135 
1136 	/* Update secondary superblocks now the physical grow has completed */
1137 	error = xfs_update_secondary_sbs(mp);
1138 
1139 out_free:
1140 	/*
1141 	 * Free the fake mp structure.
1142 	 */
1143 	kmem_free(nmp);
1144 
1145 	/*
1146 	 * If we had to allocate a new rsum_cache, we either need to free the
1147 	 * old one (if we succeeded) or free the new one and restore the old one
1148 	 * (if there was an error).
1149 	 */
1150 	if (rsum_cache != mp->m_rsum_cache) {
1151 		if (error) {
1152 			kmem_free(mp->m_rsum_cache);
1153 			mp->m_rsum_cache = rsum_cache;
1154 		} else {
1155 			kmem_free(rsum_cache);
1156 		}
1157 	}
1158 
1159 	return error;
1160 }
1161 
1162 /*
1163  * Allocate an extent in the realtime subvolume, with the usual allocation
1164  * parameters.  The length units are all in realtime extents, as is the
1165  * result block number.
1166  */
1167 int
1168 xfs_rtallocate_extent(
1169 	struct xfs_trans	*tp,
1170 	xfs_rtxnum_t		start,	/* starting rtext number to allocate */
1171 	xfs_rtxlen_t		minlen,	/* minimum length to allocate */
1172 	xfs_rtxlen_t		maxlen,	/* maximum length to allocate */
1173 	xfs_rtxlen_t		*len,	/* out: actual length allocated */
1174 	int			wasdel,	/* was a delayed allocation extent */
1175 	xfs_rtxlen_t		prod,	/* extent product factor */
1176 	xfs_rtxnum_t		*rtblock) /* out: start rtext allocated */
1177 {
1178 	struct xfs_rtalloc_args	args = {
1179 		.mp		= tp->t_mountp,
1180 		.tp		= tp,
1181 	};
1182 	int			error;	/* error value */
1183 	xfs_rtxnum_t		r;	/* result allocated rtext */
1184 
1185 	ASSERT(xfs_isilocked(args.mp->m_rbmip, XFS_ILOCK_EXCL));
1186 	ASSERT(minlen > 0 && minlen <= maxlen);
1187 
1188 	/*
1189 	 * If prod is set then figure out what to do to minlen and maxlen.
1190 	 */
1191 	if (prod > 1) {
1192 		xfs_rtxlen_t	i;
1193 
1194 		if ((i = maxlen % prod))
1195 			maxlen -= i;
1196 		if ((i = minlen % prod))
1197 			minlen += prod - i;
1198 		if (maxlen < minlen) {
1199 			*rtblock = NULLRTEXTNO;
1200 			return 0;
1201 		}
1202 	}
1203 
1204 retry:
1205 	if (start == 0) {
1206 		error = xfs_rtallocate_extent_size(&args, minlen,
1207 				maxlen, len, prod, &r);
1208 	} else {
1209 		error = xfs_rtallocate_extent_near(&args, start, minlen,
1210 				maxlen, len, prod, &r);
1211 	}
1212 
1213 	xfs_rtbuf_cache_relse(&args);
1214 	if (error)
1215 		return error;
1216 
1217 	/*
1218 	 * If it worked, update the superblock.
1219 	 */
1220 	if (r != NULLRTEXTNO) {
1221 		long	slen = (long)*len;
1222 
1223 		ASSERT(*len >= minlen && *len <= maxlen);
1224 		if (wasdel)
1225 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
1226 		else
1227 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
1228 	} else if (prod > 1) {
1229 		prod = 1;
1230 		goto retry;
1231 	}
1232 
1233 	*rtblock = r;
1234 	return 0;
1235 }
1236 
1237 /*
1238  * Initialize realtime fields in the mount structure.
1239  */
1240 int				/* error */
1241 xfs_rtmount_init(
1242 	struct xfs_mount	*mp)	/* file system mount structure */
1243 {
1244 	struct xfs_buf		*bp;	/* buffer for last block of subvolume */
1245 	struct xfs_sb		*sbp;	/* filesystem superblock copy in mount */
1246 	xfs_daddr_t		d;	/* address of last block of subvolume */
1247 	unsigned int		rsumblocks;
1248 	int			error;
1249 
1250 	sbp = &mp->m_sb;
1251 	if (sbp->sb_rblocks == 0)
1252 		return 0;
1253 	if (mp->m_rtdev_targp == NULL) {
1254 		xfs_warn(mp,
1255 	"Filesystem has a realtime volume, use rtdev=device option");
1256 		return -ENODEV;
1257 	}
1258 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
1259 	rsumblocks = xfs_rtsummary_blockcount(mp, mp->m_rsumlevels,
1260 			mp->m_sb.sb_rbmblocks);
1261 	mp->m_rsumsize = XFS_FSB_TO_B(mp, rsumblocks);
1262 	mp->m_rbmip = mp->m_rsumip = NULL;
1263 	/*
1264 	 * Check that the realtime section is an ok size.
1265 	 */
1266 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
1267 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
1268 		xfs_warn(mp, "realtime mount -- %llu != %llu",
1269 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
1270 			(unsigned long long) mp->m_sb.sb_rblocks);
1271 		return -EFBIG;
1272 	}
1273 	error = xfs_buf_read_uncached(mp->m_rtdev_targp,
1274 					d - XFS_FSB_TO_BB(mp, 1),
1275 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
1276 	if (error) {
1277 		xfs_warn(mp, "realtime device size check failed");
1278 		return error;
1279 	}
1280 	xfs_buf_relse(bp);
1281 	return 0;
1282 }
1283 
1284 static int
1285 xfs_rtalloc_count_frextent(
1286 	struct xfs_mount		*mp,
1287 	struct xfs_trans		*tp,
1288 	const struct xfs_rtalloc_rec	*rec,
1289 	void				*priv)
1290 {
1291 	uint64_t			*valp = priv;
1292 
1293 	*valp += rec->ar_extcount;
1294 	return 0;
1295 }
1296 
1297 /*
1298  * Reinitialize the number of free realtime extents from the realtime bitmap.
1299  * Callers must ensure that there is no other activity in the filesystem.
1300  */
1301 int
1302 xfs_rtalloc_reinit_frextents(
1303 	struct xfs_mount	*mp)
1304 {
1305 	uint64_t		val = 0;
1306 	int			error;
1307 
1308 	xfs_ilock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
1309 	error = xfs_rtalloc_query_all(mp, NULL, xfs_rtalloc_count_frextent,
1310 			&val);
1311 	xfs_iunlock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
1312 	if (error)
1313 		return error;
1314 
1315 	spin_lock(&mp->m_sb_lock);
1316 	mp->m_sb.sb_frextents = val;
1317 	spin_unlock(&mp->m_sb_lock);
1318 	percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1319 	return 0;
1320 }
1321 
1322 /*
1323  * Read in the bmbt of an rt metadata inode so that we never have to load them
1324  * at runtime.  This enables the use of shared ILOCKs for rtbitmap scans.  Use
1325  * an empty transaction to avoid deadlocking on loops in the bmbt.
1326  */
1327 static inline int
1328 xfs_rtmount_iread_extents(
1329 	struct xfs_inode	*ip,
1330 	unsigned int		lock_class)
1331 {
1332 	struct xfs_trans	*tp;
1333 	int			error;
1334 
1335 	error = xfs_trans_alloc_empty(ip->i_mount, &tp);
1336 	if (error)
1337 		return error;
1338 
1339 	xfs_ilock(ip, XFS_ILOCK_EXCL | lock_class);
1340 
1341 	error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1342 	if (error)
1343 		goto out_unlock;
1344 
1345 	if (xfs_inode_has_attr_fork(ip)) {
1346 		error = xfs_iread_extents(tp, ip, XFS_ATTR_FORK);
1347 		if (error)
1348 			goto out_unlock;
1349 	}
1350 
1351 out_unlock:
1352 	xfs_iunlock(ip, XFS_ILOCK_EXCL | lock_class);
1353 	xfs_trans_cancel(tp);
1354 	return error;
1355 }
1356 
1357 /*
1358  * Get the bitmap and summary inodes and the summary cache into the mount
1359  * structure at mount time.
1360  */
1361 int					/* error */
1362 xfs_rtmount_inodes(
1363 	xfs_mount_t	*mp)		/* file system mount structure */
1364 {
1365 	int		error;		/* error return value */
1366 	xfs_sb_t	*sbp;
1367 
1368 	sbp = &mp->m_sb;
1369 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
1370 	if (error)
1371 		return error;
1372 	ASSERT(mp->m_rbmip != NULL);
1373 
1374 	error = xfs_rtmount_iread_extents(mp->m_rbmip, XFS_ILOCK_RTBITMAP);
1375 	if (error)
1376 		goto out_rele_bitmap;
1377 
1378 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
1379 	if (error)
1380 		goto out_rele_bitmap;
1381 	ASSERT(mp->m_rsumip != NULL);
1382 
1383 	error = xfs_rtmount_iread_extents(mp->m_rsumip, XFS_ILOCK_RTSUM);
1384 	if (error)
1385 		goto out_rele_summary;
1386 
1387 	xfs_alloc_rsum_cache(mp, sbp->sb_rbmblocks);
1388 	return 0;
1389 
1390 out_rele_summary:
1391 	xfs_irele(mp->m_rsumip);
1392 out_rele_bitmap:
1393 	xfs_irele(mp->m_rbmip);
1394 	return error;
1395 }
1396 
1397 void
1398 xfs_rtunmount_inodes(
1399 	struct xfs_mount	*mp)
1400 {
1401 	kmem_free(mp->m_rsum_cache);
1402 	if (mp->m_rbmip)
1403 		xfs_irele(mp->m_rbmip);
1404 	if (mp->m_rsumip)
1405 		xfs_irele(mp->m_rsumip);
1406 }
1407 
1408 /*
1409  * Pick an extent for allocation at the start of a new realtime file.
1410  * Use the sequence number stored in the atime field of the bitmap inode.
1411  * Translate this to a fraction of the rtextents, and return the product
1412  * of rtextents and the fraction.
1413  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
1414  */
1415 int						/* error */
1416 xfs_rtpick_extent(
1417 	xfs_mount_t		*mp,		/* file system mount point */
1418 	xfs_trans_t		*tp,		/* transaction pointer */
1419 	xfs_rtxlen_t		len,		/* allocation length (rtextents) */
1420 	xfs_rtxnum_t		*pick)		/* result rt extent */
1421 {
1422 	xfs_rtxnum_t		b;		/* result rtext */
1423 	int			log2;		/* log of sequence number */
1424 	uint64_t		resid;		/* residual after log removed */
1425 	uint64_t		seq;		/* sequence number of file creation */
1426 	struct timespec64	ts;		/* timespec in inode */
1427 
1428 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
1429 
1430 	ts = inode_get_atime(VFS_I(mp->m_rbmip));
1431 	if (!(mp->m_rbmip->i_diflags & XFS_DIFLAG_NEWRTBM)) {
1432 		mp->m_rbmip->i_diflags |= XFS_DIFLAG_NEWRTBM;
1433 		seq = 0;
1434 	} else {
1435 		seq = ts.tv_sec;
1436 	}
1437 	if ((log2 = xfs_highbit64(seq)) == -1)
1438 		b = 0;
1439 	else {
1440 		resid = seq - (1ULL << log2);
1441 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
1442 		    (log2 + 1);
1443 		if (b >= mp->m_sb.sb_rextents)
1444 			div64_u64_rem(b, mp->m_sb.sb_rextents, &b);
1445 		if (b + len > mp->m_sb.sb_rextents)
1446 			b = mp->m_sb.sb_rextents - len;
1447 	}
1448 	ts.tv_sec = seq + 1;
1449 	inode_set_atime_to_ts(VFS_I(mp->m_rbmip), ts);
1450 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1451 	*pick = b;
1452 	return 0;
1453 }
1454