xref: /openbmc/linux/fs/xfs/xfs_rtalloc.c (revision 63dc02bd)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_alloc.h"
33 #include "xfs_bmap.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_fsops.h"
36 #include "xfs_error.h"
37 #include "xfs_rw.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_utils.h"
41 #include "xfs_trace.h"
42 #include "xfs_buf.h"
43 
44 
45 /*
46  * Prototypes for internal functions.
47  */
48 
49 
50 STATIC int xfs_rtallocate_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
51 		xfs_extlen_t, xfs_buf_t **, xfs_fsblock_t *);
52 STATIC int xfs_rtany_summary(xfs_mount_t *, xfs_trans_t *, int, int,
53 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, int *);
54 STATIC int xfs_rtcheck_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
55 		xfs_extlen_t, int, xfs_rtblock_t *, int *);
56 STATIC int xfs_rtfind_back(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
57 		xfs_rtblock_t, xfs_rtblock_t *);
58 STATIC int xfs_rtfind_forw(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
59 		xfs_rtblock_t, xfs_rtblock_t *);
60 STATIC int xfs_rtget_summary( xfs_mount_t *, xfs_trans_t *, int,
61 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, xfs_suminfo_t *);
62 STATIC int xfs_rtmodify_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
63 		xfs_extlen_t, int);
64 STATIC int xfs_rtmodify_summary(xfs_mount_t *, xfs_trans_t *, int,
65 		xfs_rtblock_t, int, xfs_buf_t **, xfs_fsblock_t *);
66 
67 /*
68  * Internal functions.
69  */
70 
71 /*
72  * Allocate space to the bitmap or summary file, and zero it, for growfs.
73  */
74 STATIC int				/* error */
75 xfs_growfs_rt_alloc(
76 	xfs_mount_t	*mp,		/* file system mount point */
77 	xfs_extlen_t	oblocks,	/* old count of blocks */
78 	xfs_extlen_t	nblocks,	/* new count of blocks */
79 	xfs_inode_t	*ip)		/* inode (bitmap/summary) */
80 {
81 	xfs_fileoff_t	bno;		/* block number in file */
82 	xfs_buf_t	*bp;		/* temporary buffer for zeroing */
83 	int		committed;	/* transaction committed flag */
84 	xfs_daddr_t	d;		/* disk block address */
85 	int		error;		/* error return value */
86 	xfs_fsblock_t	firstblock;	/* first block allocated in xaction */
87 	xfs_bmap_free_t	flist;		/* list of freed blocks */
88 	xfs_fsblock_t	fsbno;		/* filesystem block for bno */
89 	xfs_bmbt_irec_t	map;		/* block map output */
90 	int		nmap;		/* number of block maps */
91 	int		resblks;	/* space reservation */
92 
93 	/*
94 	 * Allocate space to the file, as necessary.
95 	 */
96 	while (oblocks < nblocks) {
97 		int		cancelflags = 0;
98 		xfs_trans_t	*tp;
99 
100 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ALLOC);
101 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
102 		/*
103 		 * Reserve space & log for one extent added to the file.
104 		 */
105 		if ((error = xfs_trans_reserve(tp, resblks,
106 				XFS_GROWRTALLOC_LOG_RES(mp), 0,
107 				XFS_TRANS_PERM_LOG_RES,
108 				XFS_DEFAULT_PERM_LOG_COUNT)))
109 			goto error_cancel;
110 		cancelflags = XFS_TRANS_RELEASE_LOG_RES;
111 		/*
112 		 * Lock the inode.
113 		 */
114 		xfs_ilock(ip, XFS_ILOCK_EXCL);
115 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
116 
117 		xfs_bmap_init(&flist, &firstblock);
118 		/*
119 		 * Allocate blocks to the bitmap file.
120 		 */
121 		nmap = 1;
122 		cancelflags |= XFS_TRANS_ABORT;
123 		error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
124 					XFS_BMAPI_METADATA, &firstblock,
125 					resblks, &map, &nmap, &flist);
126 		if (!error && nmap < 1)
127 			error = XFS_ERROR(ENOSPC);
128 		if (error)
129 			goto error_cancel;
130 		/*
131 		 * Free any blocks freed up in the transaction, then commit.
132 		 */
133 		error = xfs_bmap_finish(&tp, &flist, &committed);
134 		if (error)
135 			goto error_cancel;
136 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
137 		if (error)
138 			goto error;
139 		/*
140 		 * Now we need to clear the allocated blocks.
141 		 * Do this one block per transaction, to keep it simple.
142 		 */
143 		cancelflags = 0;
144 		for (bno = map.br_startoff, fsbno = map.br_startblock;
145 		     bno < map.br_startoff + map.br_blockcount;
146 		     bno++, fsbno++) {
147 			tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ZERO);
148 			/*
149 			 * Reserve log for one block zeroing.
150 			 */
151 			if ((error = xfs_trans_reserve(tp, 0,
152 					XFS_GROWRTZERO_LOG_RES(mp), 0, 0, 0)))
153 				goto error_cancel;
154 			/*
155 			 * Lock the bitmap inode.
156 			 */
157 			xfs_ilock(ip, XFS_ILOCK_EXCL);
158 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
159 			/*
160 			 * Get a buffer for the block.
161 			 */
162 			d = XFS_FSB_TO_DADDR(mp, fsbno);
163 			bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
164 				mp->m_bsize, 0);
165 			if (bp == NULL) {
166 				error = XFS_ERROR(EIO);
167 error_cancel:
168 				xfs_trans_cancel(tp, cancelflags);
169 				goto error;
170 			}
171 			memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
172 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
173 			/*
174 			 * Commit the transaction.
175 			 */
176 			error = xfs_trans_commit(tp, 0);
177 			if (error)
178 				goto error;
179 		}
180 		/*
181 		 * Go on to the next extent, if any.
182 		 */
183 		oblocks = map.br_startoff + map.br_blockcount;
184 	}
185 	return 0;
186 
187 error:
188 	return error;
189 }
190 
191 /*
192  * Attempt to allocate an extent minlen<=len<=maxlen starting from
193  * bitmap block bbno.  If we don't get maxlen then use prod to trim
194  * the length, if given.  Returns error; returns starting block in *rtblock.
195  * The lengths are all in rtextents.
196  */
197 STATIC int				/* error */
198 xfs_rtallocate_extent_block(
199 	xfs_mount_t	*mp,		/* file system mount point */
200 	xfs_trans_t	*tp,		/* transaction pointer */
201 	xfs_rtblock_t	bbno,		/* bitmap block number */
202 	xfs_extlen_t	minlen,		/* minimum length to allocate */
203 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
204 	xfs_extlen_t	*len,		/* out: actual length allocated */
205 	xfs_rtblock_t	*nextp,		/* out: next block to try */
206 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
207 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
208 	xfs_extlen_t	prod,		/* extent product factor */
209 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
210 {
211 	xfs_rtblock_t	besti;		/* best rtblock found so far */
212 	xfs_rtblock_t	bestlen;	/* best length found so far */
213 	xfs_rtblock_t	end;		/* last rtblock in chunk */
214 	int		error;		/* error value */
215 	xfs_rtblock_t	i;		/* current rtblock trying */
216 	xfs_rtblock_t	next;		/* next rtblock to try */
217 	int		stat;		/* status from internal calls */
218 
219 	/*
220 	 * Loop over all the extents starting in this bitmap block,
221 	 * looking for one that's long enough.
222 	 */
223 	for (i = XFS_BLOCKTOBIT(mp, bbno), besti = -1, bestlen = 0,
224 		end = XFS_BLOCKTOBIT(mp, bbno + 1) - 1;
225 	     i <= end;
226 	     i++) {
227 		/*
228 		 * See if there's a free extent of maxlen starting at i.
229 		 * If it's not so then next will contain the first non-free.
230 		 */
231 		error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
232 		if (error) {
233 			return error;
234 		}
235 		if (stat) {
236 			/*
237 			 * i for maxlen is all free, allocate and return that.
238 			 */
239 			error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
240 				rsb);
241 			if (error) {
242 				return error;
243 			}
244 			*len = maxlen;
245 			*rtblock = i;
246 			return 0;
247 		}
248 		/*
249 		 * In the case where we have a variable-sized allocation
250 		 * request, figure out how big this free piece is,
251 		 * and if it's big enough for the minimum, and the best
252 		 * so far, remember it.
253 		 */
254 		if (minlen < maxlen) {
255 			xfs_rtblock_t	thislen;	/* this extent size */
256 
257 			thislen = next - i;
258 			if (thislen >= minlen && thislen > bestlen) {
259 				besti = i;
260 				bestlen = thislen;
261 			}
262 		}
263 		/*
264 		 * If not done yet, find the start of the next free space.
265 		 */
266 		if (next < end) {
267 			error = xfs_rtfind_forw(mp, tp, next, end, &i);
268 			if (error) {
269 				return error;
270 			}
271 		} else
272 			break;
273 	}
274 	/*
275 	 * Searched the whole thing & didn't find a maxlen free extent.
276 	 */
277 	if (minlen < maxlen && besti != -1) {
278 		xfs_extlen_t	p;	/* amount to trim length by */
279 
280 		/*
281 		 * If size should be a multiple of prod, make that so.
282 		 */
283 		if (prod > 1 && (p = do_mod(bestlen, prod)))
284 			bestlen -= p;
285 		/*
286 		 * Allocate besti for bestlen & return that.
287 		 */
288 		error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
289 		if (error) {
290 			return error;
291 		}
292 		*len = bestlen;
293 		*rtblock = besti;
294 		return 0;
295 	}
296 	/*
297 	 * Allocation failed.  Set *nextp to the next block to try.
298 	 */
299 	*nextp = next;
300 	*rtblock = NULLRTBLOCK;
301 	return 0;
302 }
303 
304 /*
305  * Allocate an extent of length minlen<=len<=maxlen, starting at block
306  * bno.  If we don't get maxlen then use prod to trim the length, if given.
307  * Returns error; returns starting block in *rtblock.
308  * The lengths are all in rtextents.
309  */
310 STATIC int				/* error */
311 xfs_rtallocate_extent_exact(
312 	xfs_mount_t	*mp,		/* file system mount point */
313 	xfs_trans_t	*tp,		/* transaction pointer */
314 	xfs_rtblock_t	bno,		/* starting block number to allocate */
315 	xfs_extlen_t	minlen,		/* minimum length to allocate */
316 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
317 	xfs_extlen_t	*len,		/* out: actual length allocated */
318 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
319 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
320 	xfs_extlen_t	prod,		/* extent product factor */
321 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
322 {
323 	int		error;		/* error value */
324 	xfs_extlen_t	i;		/* extent length trimmed due to prod */
325 	int		isfree;		/* extent is free */
326 	xfs_rtblock_t	next;		/* next block to try (dummy) */
327 
328 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
329 	/*
330 	 * Check if the range in question (for maxlen) is free.
331 	 */
332 	error = xfs_rtcheck_range(mp, tp, bno, maxlen, 1, &next, &isfree);
333 	if (error) {
334 		return error;
335 	}
336 	if (isfree) {
337 		/*
338 		 * If it is, allocate it and return success.
339 		 */
340 		error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
341 		if (error) {
342 			return error;
343 		}
344 		*len = maxlen;
345 		*rtblock = bno;
346 		return 0;
347 	}
348 	/*
349 	 * If not, allocate what there is, if it's at least minlen.
350 	 */
351 	maxlen = next - bno;
352 	if (maxlen < minlen) {
353 		/*
354 		 * Failed, return failure status.
355 		 */
356 		*rtblock = NULLRTBLOCK;
357 		return 0;
358 	}
359 	/*
360 	 * Trim off tail of extent, if prod is specified.
361 	 */
362 	if (prod > 1 && (i = maxlen % prod)) {
363 		maxlen -= i;
364 		if (maxlen < minlen) {
365 			/*
366 			 * Now we can't do it, return failure status.
367 			 */
368 			*rtblock = NULLRTBLOCK;
369 			return 0;
370 		}
371 	}
372 	/*
373 	 * Allocate what we can and return it.
374 	 */
375 	error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
376 	if (error) {
377 		return error;
378 	}
379 	*len = maxlen;
380 	*rtblock = bno;
381 	return 0;
382 }
383 
384 /*
385  * Allocate an extent of length minlen<=len<=maxlen, starting as near
386  * to bno 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				/* error */
390 xfs_rtallocate_extent_near(
391 	xfs_mount_t	*mp,		/* file system mount point */
392 	xfs_trans_t	*tp,		/* transaction pointer */
393 	xfs_rtblock_t	bno,		/* starting block number to allocate */
394 	xfs_extlen_t	minlen,		/* minimum length to allocate */
395 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
396 	xfs_extlen_t	*len,		/* out: actual length allocated */
397 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
398 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
399 	xfs_extlen_t	prod,		/* extent product factor */
400 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
401 {
402 	int		any;		/* any useful extents from summary */
403 	xfs_rtblock_t	bbno;		/* bitmap block number */
404 	int		error;		/* error value */
405 	int		i;		/* bitmap block offset (loop control) */
406 	int		j;		/* secondary loop control */
407 	int		log2len;	/* log2 of minlen */
408 	xfs_rtblock_t	n;		/* next block to try */
409 	xfs_rtblock_t	r;		/* result block */
410 
411 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
412 	/*
413 	 * If the block number given is off the end, silently set it to
414 	 * the last block.
415 	 */
416 	if (bno >= mp->m_sb.sb_rextents)
417 		bno = mp->m_sb.sb_rextents - 1;
418 	/*
419 	 * Try the exact allocation first.
420 	 */
421 	error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen, len,
422 		rbpp, rsb, prod, &r);
423 	if (error) {
424 		return error;
425 	}
426 	/*
427 	 * If the exact allocation worked, return that.
428 	 */
429 	if (r != NULLRTBLOCK) {
430 		*rtblock = r;
431 		return 0;
432 	}
433 	bbno = XFS_BITTOBLOCK(mp, bno);
434 	i = 0;
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(mp, tp, log2len, mp->m_rsumlevels - 1,
446 			bbno + i, rbpp, rsb, &any);
447 		if (error) {
448 			return error;
449 		}
450 		/*
451 		 * If there are any useful extents starting here, try
452 		 * allocating one.
453 		 */
454 		if (any) {
455 			/*
456 			 * On the positive side of the starting location.
457 			 */
458 			if (i >= 0) {
459 				/*
460 				 * Try to allocate an extent starting in
461 				 * this block.
462 				 */
463 				error = xfs_rtallocate_extent_block(mp, tp,
464 					bbno + i, minlen, maxlen, len, &n, rbpp,
465 					rsb, prod, &r);
466 				if (error) {
467 					return error;
468 				}
469 				/*
470 				 * If it worked, return it.
471 				 */
472 				if (r != NULLRTBLOCK) {
473 					*rtblock = r;
474 					return 0;
475 				}
476 			}
477 			/*
478 			 * On the negative side of the starting location.
479 			 */
480 			else {		/* i < 0 */
481 				/*
482 				 * Loop backwards through the bitmap blocks from
483 				 * the starting point-1 up to where we are now.
484 				 * There should be an extent which ends in this
485 				 * bitmap block and is long enough.
486 				 */
487 				for (j = -1; j > i; j--) {
488 					/*
489 					 * Grab the summary information for
490 					 * this bitmap block.
491 					 */
492 					error = xfs_rtany_summary(mp, tp,
493 						log2len, mp->m_rsumlevels - 1,
494 						bbno + j, rbpp, rsb, &any);
495 					if (error) {
496 						return error;
497 					}
498 					/*
499 					 * If there's no extent given in the
500 					 * summary that means the extent we
501 					 * found must carry over from an
502 					 * earlier block.  If there is an
503 					 * extent given, we've already tried
504 					 * that allocation, don't do it again.
505 					 */
506 					if (any)
507 						continue;
508 					error = xfs_rtallocate_extent_block(mp,
509 						tp, bbno + j, minlen, maxlen,
510 						len, &n, rbpp, rsb, prod, &r);
511 					if (error) {
512 						return error;
513 					}
514 					/*
515 					 * If it works, return the extent.
516 					 */
517 					if (r != NULLRTBLOCK) {
518 						*rtblock = r;
519 						return 0;
520 					}
521 				}
522 				/*
523 				 * There weren't intervening bitmap blocks
524 				 * with a long enough extent, or the
525 				 * allocation didn't work for some reason
526 				 * (i.e. it's a little * too short).
527 				 * Try to allocate from the summary block
528 				 * that we found.
529 				 */
530 				error = xfs_rtallocate_extent_block(mp, tp,
531 					bbno + i, minlen, maxlen, len, &n, rbpp,
532 					rsb, prod, &r);
533 				if (error) {
534 					return error;
535 				}
536 				/*
537 				 * If it works, return the extent.
538 				 */
539 				if (r != NULLRTBLOCK) {
540 					*rtblock = r;
541 					return 0;
542 				}
543 			}
544 		}
545 		/*
546 		 * Loop control.  If we were on the positive side, and there's
547 		 * still more blocks on the negative side, go there.
548 		 */
549 		if (i > 0 && (int)bbno - i >= 0)
550 			i = -i;
551 		/*
552 		 * If positive, and no more negative, but there are more
553 		 * positive, go there.
554 		 */
555 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
556 			i++;
557 		/*
558 		 * If negative or 0 (just started), and there are positive
559 		 * blocks to go, go there.  The 0 case moves to block 1.
560 		 */
561 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
562 			i = 1 - i;
563 		/*
564 		 * If negative or 0 and there are more negative blocks,
565 		 * go there.
566 		 */
567 		else if (i <= 0 && (int)bbno + i > 0)
568 			i--;
569 		/*
570 		 * Must be done.  Return failure.
571 		 */
572 		else
573 			break;
574 	}
575 	*rtblock = NULLRTBLOCK;
576 	return 0;
577 }
578 
579 /*
580  * Allocate an extent of length minlen<=len<=maxlen, with no position
581  * specified.  If we don't get maxlen then use prod to trim
582  * the length, if given.  The lengths are all in rtextents.
583  */
584 STATIC int				/* error */
585 xfs_rtallocate_extent_size(
586 	xfs_mount_t	*mp,		/* file system mount point */
587 	xfs_trans_t	*tp,		/* transaction pointer */
588 	xfs_extlen_t	minlen,		/* minimum length to allocate */
589 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
590 	xfs_extlen_t	*len,		/* out: actual length allocated */
591 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
592 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
593 	xfs_extlen_t	prod,		/* extent product factor */
594 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
595 {
596 	int		error;		/* error value */
597 	int		i;		/* bitmap block number */
598 	int		l;		/* level number (loop control) */
599 	xfs_rtblock_t	n;		/* next block to be tried */
600 	xfs_rtblock_t	r;		/* result block number */
601 	xfs_suminfo_t	sum;		/* summary information for extents */
602 
603 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
604 	ASSERT(maxlen != 0);
605 
606 	/*
607 	 * Loop over all the levels starting with maxlen.
608 	 * At each level, look at all the bitmap blocks, to see if there
609 	 * are extents starting there that are long enough (>= maxlen).
610 	 * Note, only on the initial level can the allocation fail if
611 	 * the summary says there's an extent.
612 	 */
613 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
614 		/*
615 		 * Loop over all the bitmap blocks.
616 		 */
617 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
618 			/*
619 			 * Get the summary for this level/block.
620 			 */
621 			error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
622 				&sum);
623 			if (error) {
624 				return error;
625 			}
626 			/*
627 			 * Nothing there, on to the next block.
628 			 */
629 			if (!sum)
630 				continue;
631 			/*
632 			 * Try allocating the extent.
633 			 */
634 			error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
635 				maxlen, len, &n, rbpp, rsb, prod, &r);
636 			if (error) {
637 				return error;
638 			}
639 			/*
640 			 * If it worked, return that.
641 			 */
642 			if (r != NULLRTBLOCK) {
643 				*rtblock = r;
644 				return 0;
645 			}
646 			/*
647 			 * If the "next block to try" returned from the
648 			 * allocator is beyond the next bitmap block,
649 			 * skip to that bitmap block.
650 			 */
651 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
652 				i = XFS_BITTOBLOCK(mp, n) - 1;
653 		}
654 	}
655 	/*
656 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
657 	 * we're asking for a fixed size extent.
658 	 */
659 	if (minlen > --maxlen) {
660 		*rtblock = NULLRTBLOCK;
661 		return 0;
662 	}
663 	ASSERT(minlen != 0);
664 	ASSERT(maxlen != 0);
665 
666 	/*
667 	 * Loop over sizes, from maxlen down to minlen.
668 	 * This time, when we do the allocations, allow smaller ones
669 	 * to succeed.
670 	 */
671 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
672 		/*
673 		 * Loop over all the bitmap blocks, try an allocation
674 		 * starting in that block.
675 		 */
676 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
677 			/*
678 			 * Get the summary information for this level/block.
679 			 */
680 			error =	xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
681 						  &sum);
682 			if (error) {
683 				return error;
684 			}
685 			/*
686 			 * If nothing there, go on to next.
687 			 */
688 			if (!sum)
689 				continue;
690 			/*
691 			 * Try the allocation.  Make sure the specified
692 			 * minlen/maxlen are in the possible range for
693 			 * this summary level.
694 			 */
695 			error = xfs_rtallocate_extent_block(mp, tp, i,
696 					XFS_RTMAX(minlen, 1 << l),
697 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
698 					len, &n, rbpp, rsb, prod, &r);
699 			if (error) {
700 				return error;
701 			}
702 			/*
703 			 * If it worked, return that extent.
704 			 */
705 			if (r != NULLRTBLOCK) {
706 				*rtblock = r;
707 				return 0;
708 			}
709 			/*
710 			 * If the "next block to try" returned from the
711 			 * allocator is beyond the next bitmap block,
712 			 * skip to that bitmap block.
713 			 */
714 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
715 				i = XFS_BITTOBLOCK(mp, n) - 1;
716 		}
717 	}
718 	/*
719 	 * Got nothing, return failure.
720 	 */
721 	*rtblock = NULLRTBLOCK;
722 	return 0;
723 }
724 
725 /*
726  * Mark an extent specified by start and len allocated.
727  * Updates all the summary information as well as the bitmap.
728  */
729 STATIC int				/* error */
730 xfs_rtallocate_range(
731 	xfs_mount_t	*mp,		/* file system mount point */
732 	xfs_trans_t	*tp,		/* transaction pointer */
733 	xfs_rtblock_t	start,		/* start block to allocate */
734 	xfs_extlen_t	len,		/* length to allocate */
735 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
736 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
737 {
738 	xfs_rtblock_t	end;		/* end of the allocated extent */
739 	int		error;		/* error value */
740 	xfs_rtblock_t	postblock;	/* first block allocated > end */
741 	xfs_rtblock_t	preblock;	/* first block allocated < start */
742 
743 	end = start + len - 1;
744 	/*
745 	 * Assume we're allocating out of the middle of a free extent.
746 	 * We need to find the beginning and end of the extent so we can
747 	 * properly update the summary.
748 	 */
749 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
750 	if (error) {
751 		return error;
752 	}
753 	/*
754 	 * Find the next allocated block (end of free extent).
755 	 */
756 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
757 		&postblock);
758 	if (error) {
759 		return error;
760 	}
761 	/*
762 	 * Decrement the summary information corresponding to the entire
763 	 * (old) free extent.
764 	 */
765 	error = xfs_rtmodify_summary(mp, tp,
766 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
767 		XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
768 	if (error) {
769 		return error;
770 	}
771 	/*
772 	 * If there are blocks not being allocated at the front of the
773 	 * old extent, add summary data for them to be free.
774 	 */
775 	if (preblock < start) {
776 		error = xfs_rtmodify_summary(mp, tp,
777 			XFS_RTBLOCKLOG(start - preblock),
778 			XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
779 		if (error) {
780 			return error;
781 		}
782 	}
783 	/*
784 	 * If there are blocks not being allocated at the end of the
785 	 * old extent, add summary data for them to be free.
786 	 */
787 	if (postblock > end) {
788 		error = xfs_rtmodify_summary(mp, tp,
789 			XFS_RTBLOCKLOG(postblock - end),
790 			XFS_BITTOBLOCK(mp, end + 1), 1, rbpp, rsb);
791 		if (error) {
792 			return error;
793 		}
794 	}
795 	/*
796 	 * Modify the bitmap to mark this extent allocated.
797 	 */
798 	error = xfs_rtmodify_range(mp, tp, start, len, 0);
799 	return error;
800 }
801 
802 /*
803  * Return whether there are any free extents in the size range given
804  * by low and high, for the bitmap block bbno.
805  */
806 STATIC int				/* error */
807 xfs_rtany_summary(
808 	xfs_mount_t	*mp,		/* file system mount structure */
809 	xfs_trans_t	*tp,		/* transaction pointer */
810 	int		low,		/* low log2 extent size */
811 	int		high,		/* high log2 extent size */
812 	xfs_rtblock_t	bbno,		/* bitmap block number */
813 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
814 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
815 	int		*stat)		/* out: any good extents here? */
816 {
817 	int		error;		/* error value */
818 	int		log;		/* loop counter, log2 of ext. size */
819 	xfs_suminfo_t	sum;		/* summary data */
820 
821 	/*
822 	 * Loop over logs of extent sizes.  Order is irrelevant.
823 	 */
824 	for (log = low; log <= high; log++) {
825 		/*
826 		 * Get one summary datum.
827 		 */
828 		error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
829 		if (error) {
830 			return error;
831 		}
832 		/*
833 		 * If there are any, return success.
834 		 */
835 		if (sum) {
836 			*stat = 1;
837 			return 0;
838 		}
839 	}
840 	/*
841 	 * Found nothing, return failure.
842 	 */
843 	*stat = 0;
844 	return 0;
845 }
846 
847 /*
848  * Get a buffer for the bitmap or summary file block specified.
849  * The buffer is returned read and locked.
850  */
851 STATIC int				/* error */
852 xfs_rtbuf_get(
853 	xfs_mount_t	*mp,		/* file system mount structure */
854 	xfs_trans_t	*tp,		/* transaction pointer */
855 	xfs_rtblock_t	block,		/* block number in bitmap or summary */
856 	int		issum,		/* is summary not bitmap */
857 	xfs_buf_t	**bpp)		/* output: buffer for the block */
858 {
859 	xfs_buf_t	*bp;		/* block buffer, result */
860 	xfs_inode_t	*ip;		/* bitmap or summary inode */
861 	xfs_bmbt_irec_t	map;
862 	int		nmap;
863 	int		error;		/* error value */
864 
865 	ip = issum ? mp->m_rsumip : mp->m_rbmip;
866 
867 	error = xfs_bmapi_read(ip, block, 1, &map, &nmap, XFS_DATA_FORK);
868 	if (error)
869 		return error;
870 
871 	ASSERT(map.br_startblock != NULLFSBLOCK);
872 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
873 				   XFS_FSB_TO_DADDR(mp, map.br_startblock),
874 				   mp->m_bsize, 0, &bp);
875 	if (error)
876 		return error;
877 	ASSERT(!xfs_buf_geterror(bp));
878 	*bpp = bp;
879 	return 0;
880 }
881 
882 #ifdef DEBUG
883 /*
884  * Check that the given extent (block range) is allocated already.
885  */
886 STATIC int				/* error */
887 xfs_rtcheck_alloc_range(
888 	xfs_mount_t	*mp,		/* file system mount point */
889 	xfs_trans_t	*tp,		/* transaction pointer */
890 	xfs_rtblock_t	bno,		/* starting block number of extent */
891 	xfs_extlen_t	len,		/* length of extent */
892 	int		*stat)		/* out: 1 for allocated, 0 for not */
893 {
894 	xfs_rtblock_t	new;		/* dummy for xfs_rtcheck_range */
895 
896 	return xfs_rtcheck_range(mp, tp, bno, len, 0, &new, stat);
897 }
898 #endif
899 
900 /*
901  * Check that the given range is either all allocated (val = 0) or
902  * all free (val = 1).
903  */
904 STATIC int				/* error */
905 xfs_rtcheck_range(
906 	xfs_mount_t	*mp,		/* file system mount point */
907 	xfs_trans_t	*tp,		/* transaction pointer */
908 	xfs_rtblock_t	start,		/* starting block number of extent */
909 	xfs_extlen_t	len,		/* length of extent */
910 	int		val,		/* 1 for free, 0 for allocated */
911 	xfs_rtblock_t	*new,		/* out: first block not matching */
912 	int		*stat)		/* out: 1 for matches, 0 for not */
913 {
914 	xfs_rtword_t	*b;		/* current word in buffer */
915 	int		bit;		/* bit number in the word */
916 	xfs_rtblock_t	block;		/* bitmap block number */
917 	xfs_buf_t	*bp;		/* buf for the block */
918 	xfs_rtword_t	*bufp;		/* starting word in buffer */
919 	int		error;		/* error value */
920 	xfs_rtblock_t	i;		/* current bit number rel. to start */
921 	xfs_rtblock_t	lastbit;	/* last useful bit in word */
922 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
923 	xfs_rtword_t	wdiff;		/* difference from wanted value */
924 	int		word;		/* word number in the buffer */
925 
926 	/*
927 	 * Compute starting bitmap block number
928 	 */
929 	block = XFS_BITTOBLOCK(mp, start);
930 	/*
931 	 * Read the bitmap block.
932 	 */
933 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
934 	if (error) {
935 		return error;
936 	}
937 	bufp = bp->b_addr;
938 	/*
939 	 * Compute the starting word's address, and starting bit.
940 	 */
941 	word = XFS_BITTOWORD(mp, start);
942 	b = &bufp[word];
943 	bit = (int)(start & (XFS_NBWORD - 1));
944 	/*
945 	 * 0 (allocated) => all zero's; 1 (free) => all one's.
946 	 */
947 	val = -val;
948 	/*
949 	 * If not starting on a word boundary, deal with the first
950 	 * (partial) word.
951 	 */
952 	if (bit) {
953 		/*
954 		 * Compute first bit not examined.
955 		 */
956 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
957 		/*
958 		 * Mask of relevant bits.
959 		 */
960 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
961 		/*
962 		 * Compute difference between actual and desired value.
963 		 */
964 		if ((wdiff = (*b ^ val) & mask)) {
965 			/*
966 			 * Different, compute first wrong bit and return.
967 			 */
968 			xfs_trans_brelse(tp, bp);
969 			i = XFS_RTLOBIT(wdiff) - bit;
970 			*new = start + i;
971 			*stat = 0;
972 			return 0;
973 		}
974 		i = lastbit - bit;
975 		/*
976 		 * Go on to next block if that's where the next word is
977 		 * and we need the next word.
978 		 */
979 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
980 			/*
981 			 * If done with this block, get the next one.
982 			 */
983 			xfs_trans_brelse(tp, bp);
984 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
985 			if (error) {
986 				return error;
987 			}
988 			b = bufp = bp->b_addr;
989 			word = 0;
990 		} else {
991 			/*
992 			 * Go on to the next word in the buffer.
993 			 */
994 			b++;
995 		}
996 	} else {
997 		/*
998 		 * Starting on a word boundary, no partial word.
999 		 */
1000 		i = 0;
1001 	}
1002 	/*
1003 	 * Loop over whole words in buffers.  When we use up one buffer
1004 	 * we move on to the next one.
1005 	 */
1006 	while (len - i >= XFS_NBWORD) {
1007 		/*
1008 		 * Compute difference between actual and desired value.
1009 		 */
1010 		if ((wdiff = *b ^ val)) {
1011 			/*
1012 			 * Different, compute first wrong bit and return.
1013 			 */
1014 			xfs_trans_brelse(tp, bp);
1015 			i += XFS_RTLOBIT(wdiff);
1016 			*new = start + i;
1017 			*stat = 0;
1018 			return 0;
1019 		}
1020 		i += XFS_NBWORD;
1021 		/*
1022 		 * Go on to next block if that's where the next word is
1023 		 * and we need the next word.
1024 		 */
1025 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1026 			/*
1027 			 * If done with this block, get the next one.
1028 			 */
1029 			xfs_trans_brelse(tp, bp);
1030 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1031 			if (error) {
1032 				return error;
1033 			}
1034 			b = bufp = bp->b_addr;
1035 			word = 0;
1036 		} else {
1037 			/*
1038 			 * Go on to the next word in the buffer.
1039 			 */
1040 			b++;
1041 		}
1042 	}
1043 	/*
1044 	 * If not ending on a word boundary, deal with the last
1045 	 * (partial) word.
1046 	 */
1047 	if ((lastbit = len - i)) {
1048 		/*
1049 		 * Mask of relevant bits.
1050 		 */
1051 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1052 		/*
1053 		 * Compute difference between actual and desired value.
1054 		 */
1055 		if ((wdiff = (*b ^ val) & mask)) {
1056 			/*
1057 			 * Different, compute first wrong bit and return.
1058 			 */
1059 			xfs_trans_brelse(tp, bp);
1060 			i += XFS_RTLOBIT(wdiff);
1061 			*new = start + i;
1062 			*stat = 0;
1063 			return 0;
1064 		} else
1065 			i = len;
1066 	}
1067 	/*
1068 	 * Successful, return.
1069 	 */
1070 	xfs_trans_brelse(tp, bp);
1071 	*new = start + i;
1072 	*stat = 1;
1073 	return 0;
1074 }
1075 
1076 /*
1077  * Copy and transform the summary file, given the old and new
1078  * parameters in the mount structures.
1079  */
1080 STATIC int				/* error */
1081 xfs_rtcopy_summary(
1082 	xfs_mount_t	*omp,		/* old file system mount point */
1083 	xfs_mount_t	*nmp,		/* new file system mount point */
1084 	xfs_trans_t	*tp)		/* transaction pointer */
1085 {
1086 	xfs_rtblock_t	bbno;		/* bitmap block number */
1087 	xfs_buf_t	*bp;		/* summary buffer */
1088 	int		error;		/* error return value */
1089 	int		log;		/* summary level number (log length) */
1090 	xfs_suminfo_t	sum;		/* summary data */
1091 	xfs_fsblock_t	sumbno;		/* summary block number */
1092 
1093 	bp = NULL;
1094 	for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
1095 		for (bbno = omp->m_sb.sb_rbmblocks - 1;
1096 		     (xfs_srtblock_t)bbno >= 0;
1097 		     bbno--) {
1098 			error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
1099 				&sumbno, &sum);
1100 			if (error)
1101 				return error;
1102 			if (sum == 0)
1103 				continue;
1104 			error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
1105 				&bp, &sumbno);
1106 			if (error)
1107 				return error;
1108 			error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
1109 				&bp, &sumbno);
1110 			if (error)
1111 				return error;
1112 			ASSERT(sum > 0);
1113 		}
1114 	}
1115 	return 0;
1116 }
1117 
1118 /*
1119  * Searching backward from start to limit, find the first block whose
1120  * allocated/free state is different from start's.
1121  */
1122 STATIC int				/* error */
1123 xfs_rtfind_back(
1124 	xfs_mount_t	*mp,		/* file system mount point */
1125 	xfs_trans_t	*tp,		/* transaction pointer */
1126 	xfs_rtblock_t	start,		/* starting block to look at */
1127 	xfs_rtblock_t	limit,		/* last block to look at */
1128 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1129 {
1130 	xfs_rtword_t	*b;		/* current word in buffer */
1131 	int		bit;		/* bit number in the word */
1132 	xfs_rtblock_t	block;		/* bitmap block number */
1133 	xfs_buf_t	*bp;		/* buf for the block */
1134 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1135 	int		error;		/* error value */
1136 	xfs_rtblock_t	firstbit;	/* first useful bit in the word */
1137 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1138 	xfs_rtblock_t	len;		/* length of inspected area */
1139 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1140 	xfs_rtword_t	want;		/* mask for "good" values */
1141 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1142 	int		word;		/* word number in the buffer */
1143 
1144 	/*
1145 	 * Compute and read in starting bitmap block for starting block.
1146 	 */
1147 	block = XFS_BITTOBLOCK(mp, start);
1148 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1149 	if (error) {
1150 		return error;
1151 	}
1152 	bufp = bp->b_addr;
1153 	/*
1154 	 * Get the first word's index & point to it.
1155 	 */
1156 	word = XFS_BITTOWORD(mp, start);
1157 	b = &bufp[word];
1158 	bit = (int)(start & (XFS_NBWORD - 1));
1159 	len = start - limit + 1;
1160 	/*
1161 	 * Compute match value, based on the bit at start: if 1 (free)
1162 	 * then all-ones, else all-zeroes.
1163 	 */
1164 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1165 	/*
1166 	 * If the starting position is not word-aligned, deal with the
1167 	 * partial word.
1168 	 */
1169 	if (bit < XFS_NBWORD - 1) {
1170 		/*
1171 		 * Calculate first (leftmost) bit number to look at,
1172 		 * and mask for all the relevant bits in this word.
1173 		 */
1174 		firstbit = XFS_RTMAX((xfs_srtblock_t)(bit - len + 1), 0);
1175 		mask = (((xfs_rtword_t)1 << (bit - firstbit + 1)) - 1) <<
1176 			firstbit;
1177 		/*
1178 		 * Calculate the difference between the value there
1179 		 * and what we're looking for.
1180 		 */
1181 		if ((wdiff = (*b ^ want) & mask)) {
1182 			/*
1183 			 * Different.  Mark where we are and return.
1184 			 */
1185 			xfs_trans_brelse(tp, bp);
1186 			i = bit - XFS_RTHIBIT(wdiff);
1187 			*rtblock = start - i + 1;
1188 			return 0;
1189 		}
1190 		i = bit - firstbit + 1;
1191 		/*
1192 		 * Go on to previous block if that's where the previous word is
1193 		 * and we need the previous word.
1194 		 */
1195 		if (--word == -1 && i < len) {
1196 			/*
1197 			 * If done with this block, get the previous one.
1198 			 */
1199 			xfs_trans_brelse(tp, bp);
1200 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1201 			if (error) {
1202 				return error;
1203 			}
1204 			bufp = bp->b_addr;
1205 			word = XFS_BLOCKWMASK(mp);
1206 			b = &bufp[word];
1207 		} else {
1208 			/*
1209 			 * Go on to the previous word in the buffer.
1210 			 */
1211 			b--;
1212 		}
1213 	} else {
1214 		/*
1215 		 * Starting on a word boundary, no partial word.
1216 		 */
1217 		i = 0;
1218 	}
1219 	/*
1220 	 * Loop over whole words in buffers.  When we use up one buffer
1221 	 * we move on to the previous one.
1222 	 */
1223 	while (len - i >= XFS_NBWORD) {
1224 		/*
1225 		 * Compute difference between actual and desired value.
1226 		 */
1227 		if ((wdiff = *b ^ want)) {
1228 			/*
1229 			 * Different, mark where we are and return.
1230 			 */
1231 			xfs_trans_brelse(tp, bp);
1232 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1233 			*rtblock = start - i + 1;
1234 			return 0;
1235 		}
1236 		i += XFS_NBWORD;
1237 		/*
1238 		 * Go on to previous block if that's where the previous word is
1239 		 * and we need the previous word.
1240 		 */
1241 		if (--word == -1 && i < len) {
1242 			/*
1243 			 * If done with this block, get the previous one.
1244 			 */
1245 			xfs_trans_brelse(tp, bp);
1246 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1247 			if (error) {
1248 				return error;
1249 			}
1250 			bufp = bp->b_addr;
1251 			word = XFS_BLOCKWMASK(mp);
1252 			b = &bufp[word];
1253 		} else {
1254 			/*
1255 			 * Go on to the previous word in the buffer.
1256 			 */
1257 			b--;
1258 		}
1259 	}
1260 	/*
1261 	 * If not ending on a word boundary, deal with the last
1262 	 * (partial) word.
1263 	 */
1264 	if (len - i) {
1265 		/*
1266 		 * Calculate first (leftmost) bit number to look at,
1267 		 * and mask for all the relevant bits in this word.
1268 		 */
1269 		firstbit = XFS_NBWORD - (len - i);
1270 		mask = (((xfs_rtword_t)1 << (len - i)) - 1) << firstbit;
1271 		/*
1272 		 * Compute difference between actual and desired value.
1273 		 */
1274 		if ((wdiff = (*b ^ want) & mask)) {
1275 			/*
1276 			 * Different, mark where we are and return.
1277 			 */
1278 			xfs_trans_brelse(tp, bp);
1279 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1280 			*rtblock = start - i + 1;
1281 			return 0;
1282 		} else
1283 			i = len;
1284 	}
1285 	/*
1286 	 * No match, return that we scanned the whole area.
1287 	 */
1288 	xfs_trans_brelse(tp, bp);
1289 	*rtblock = start - i + 1;
1290 	return 0;
1291 }
1292 
1293 /*
1294  * Searching forward from start to limit, find the first block whose
1295  * allocated/free state is different from start's.
1296  */
1297 STATIC int				/* error */
1298 xfs_rtfind_forw(
1299 	xfs_mount_t	*mp,		/* file system mount point */
1300 	xfs_trans_t	*tp,		/* transaction pointer */
1301 	xfs_rtblock_t	start,		/* starting block to look at */
1302 	xfs_rtblock_t	limit,		/* last block to look at */
1303 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1304 {
1305 	xfs_rtword_t	*b;		/* current word in buffer */
1306 	int		bit;		/* bit number in the word */
1307 	xfs_rtblock_t	block;		/* bitmap block number */
1308 	xfs_buf_t	*bp;		/* buf for the block */
1309 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1310 	int		error;		/* error value */
1311 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1312 	xfs_rtblock_t	lastbit;	/* last useful bit in the word */
1313 	xfs_rtblock_t	len;		/* length of inspected area */
1314 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1315 	xfs_rtword_t	want;		/* mask for "good" values */
1316 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1317 	int		word;		/* word number in the buffer */
1318 
1319 	/*
1320 	 * Compute and read in starting bitmap block for starting block.
1321 	 */
1322 	block = XFS_BITTOBLOCK(mp, start);
1323 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1324 	if (error) {
1325 		return error;
1326 	}
1327 	bufp = bp->b_addr;
1328 	/*
1329 	 * Get the first word's index & point to it.
1330 	 */
1331 	word = XFS_BITTOWORD(mp, start);
1332 	b = &bufp[word];
1333 	bit = (int)(start & (XFS_NBWORD - 1));
1334 	len = limit - start + 1;
1335 	/*
1336 	 * Compute match value, based on the bit at start: if 1 (free)
1337 	 * then all-ones, else all-zeroes.
1338 	 */
1339 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1340 	/*
1341 	 * If the starting position is not word-aligned, deal with the
1342 	 * partial word.
1343 	 */
1344 	if (bit) {
1345 		/*
1346 		 * Calculate last (rightmost) bit number to look at,
1347 		 * and mask for all the relevant bits in this word.
1348 		 */
1349 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1350 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1351 		/*
1352 		 * Calculate the difference between the value there
1353 		 * and what we're looking for.
1354 		 */
1355 		if ((wdiff = (*b ^ want) & mask)) {
1356 			/*
1357 			 * Different.  Mark where we are and return.
1358 			 */
1359 			xfs_trans_brelse(tp, bp);
1360 			i = XFS_RTLOBIT(wdiff) - bit;
1361 			*rtblock = start + i - 1;
1362 			return 0;
1363 		}
1364 		i = lastbit - bit;
1365 		/*
1366 		 * Go on to next block if that's where the next word is
1367 		 * and we need the next word.
1368 		 */
1369 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1370 			/*
1371 			 * If done with this block, get the previous one.
1372 			 */
1373 			xfs_trans_brelse(tp, bp);
1374 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1375 			if (error) {
1376 				return error;
1377 			}
1378 			b = bufp = bp->b_addr;
1379 			word = 0;
1380 		} else {
1381 			/*
1382 			 * Go on to the previous word in the buffer.
1383 			 */
1384 			b++;
1385 		}
1386 	} else {
1387 		/*
1388 		 * Starting on a word boundary, no partial word.
1389 		 */
1390 		i = 0;
1391 	}
1392 	/*
1393 	 * Loop over whole words in buffers.  When we use up one buffer
1394 	 * we move on to the next one.
1395 	 */
1396 	while (len - i >= XFS_NBWORD) {
1397 		/*
1398 		 * Compute difference between actual and desired value.
1399 		 */
1400 		if ((wdiff = *b ^ want)) {
1401 			/*
1402 			 * Different, mark where we are and return.
1403 			 */
1404 			xfs_trans_brelse(tp, bp);
1405 			i += XFS_RTLOBIT(wdiff);
1406 			*rtblock = start + i - 1;
1407 			return 0;
1408 		}
1409 		i += XFS_NBWORD;
1410 		/*
1411 		 * Go on to next block if that's where the next word is
1412 		 * and we need the next word.
1413 		 */
1414 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1415 			/*
1416 			 * If done with this block, get the next one.
1417 			 */
1418 			xfs_trans_brelse(tp, bp);
1419 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1420 			if (error) {
1421 				return error;
1422 			}
1423 			b = bufp = bp->b_addr;
1424 			word = 0;
1425 		} else {
1426 			/*
1427 			 * Go on to the next word in the buffer.
1428 			 */
1429 			b++;
1430 		}
1431 	}
1432 	/*
1433 	 * If not ending on a word boundary, deal with the last
1434 	 * (partial) word.
1435 	 */
1436 	if ((lastbit = len - i)) {
1437 		/*
1438 		 * Calculate mask for all the relevant bits in this word.
1439 		 */
1440 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1441 		/*
1442 		 * Compute difference between actual and desired value.
1443 		 */
1444 		if ((wdiff = (*b ^ want) & mask)) {
1445 			/*
1446 			 * Different, mark where we are and return.
1447 			 */
1448 			xfs_trans_brelse(tp, bp);
1449 			i += XFS_RTLOBIT(wdiff);
1450 			*rtblock = start + i - 1;
1451 			return 0;
1452 		} else
1453 			i = len;
1454 	}
1455 	/*
1456 	 * No match, return that we scanned the whole area.
1457 	 */
1458 	xfs_trans_brelse(tp, bp);
1459 	*rtblock = start + i - 1;
1460 	return 0;
1461 }
1462 
1463 /*
1464  * Mark an extent specified by start and len freed.
1465  * Updates all the summary information as well as the bitmap.
1466  */
1467 STATIC int				/* error */
1468 xfs_rtfree_range(
1469 	xfs_mount_t	*mp,		/* file system mount point */
1470 	xfs_trans_t	*tp,		/* transaction pointer */
1471 	xfs_rtblock_t	start,		/* starting block to free */
1472 	xfs_extlen_t	len,		/* length to free */
1473 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1474 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1475 {
1476 	xfs_rtblock_t	end;		/* end of the freed extent */
1477 	int		error;		/* error value */
1478 	xfs_rtblock_t	postblock;	/* first block freed > end */
1479 	xfs_rtblock_t	preblock;	/* first block freed < start */
1480 
1481 	end = start + len - 1;
1482 	/*
1483 	 * Modify the bitmap to mark this extent freed.
1484 	 */
1485 	error = xfs_rtmodify_range(mp, tp, start, len, 1);
1486 	if (error) {
1487 		return error;
1488 	}
1489 	/*
1490 	 * Assume we're freeing out of the middle of an allocated extent.
1491 	 * We need to find the beginning and end of the extent so we can
1492 	 * properly update the summary.
1493 	 */
1494 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
1495 	if (error) {
1496 		return error;
1497 	}
1498 	/*
1499 	 * Find the next allocated block (end of allocated extent).
1500 	 */
1501 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
1502 		&postblock);
1503 	if (error)
1504 		return error;
1505 	/*
1506 	 * If there are blocks not being freed at the front of the
1507 	 * old extent, add summary data for them to be allocated.
1508 	 */
1509 	if (preblock < start) {
1510 		error = xfs_rtmodify_summary(mp, tp,
1511 			XFS_RTBLOCKLOG(start - preblock),
1512 			XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
1513 		if (error) {
1514 			return error;
1515 		}
1516 	}
1517 	/*
1518 	 * If there are blocks not being freed at the end of the
1519 	 * old extent, add summary data for them to be allocated.
1520 	 */
1521 	if (postblock > end) {
1522 		error = xfs_rtmodify_summary(mp, tp,
1523 			XFS_RTBLOCKLOG(postblock - end),
1524 			XFS_BITTOBLOCK(mp, end + 1), -1, rbpp, rsb);
1525 		if (error) {
1526 			return error;
1527 		}
1528 	}
1529 	/*
1530 	 * Increment the summary information corresponding to the entire
1531 	 * (new) free extent.
1532 	 */
1533 	error = xfs_rtmodify_summary(mp, tp,
1534 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
1535 		XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
1536 	return error;
1537 }
1538 
1539 /*
1540  * Read and return the summary information for a given extent size,
1541  * bitmap block combination.
1542  * Keeps track of a current summary block, so we don't keep reading
1543  * it from the buffer cache.
1544  */
1545 STATIC int				/* error */
1546 xfs_rtget_summary(
1547 	xfs_mount_t	*mp,		/* file system mount structure */
1548 	xfs_trans_t	*tp,		/* transaction pointer */
1549 	int		log,		/* log2 of extent size */
1550 	xfs_rtblock_t	bbno,		/* bitmap block number */
1551 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1552 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
1553 	xfs_suminfo_t	*sum)		/* out: summary info for this block */
1554 {
1555 	xfs_buf_t	*bp;		/* buffer for summary block */
1556 	int		error;		/* error value */
1557 	xfs_fsblock_t	sb;		/* summary fsblock */
1558 	int		so;		/* index into the summary file */
1559 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1560 
1561 	/*
1562 	 * Compute entry number in the summary file.
1563 	 */
1564 	so = XFS_SUMOFFS(mp, log, bbno);
1565 	/*
1566 	 * Compute the block number in the summary file.
1567 	 */
1568 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1569 	/*
1570 	 * If we have an old buffer, and the block number matches, use that.
1571 	 */
1572 	if (rbpp && *rbpp && *rsb == sb)
1573 		bp = *rbpp;
1574 	/*
1575 	 * Otherwise we have to get the buffer.
1576 	 */
1577 	else {
1578 		/*
1579 		 * If there was an old one, get rid of it first.
1580 		 */
1581 		if (rbpp && *rbpp)
1582 			xfs_trans_brelse(tp, *rbpp);
1583 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1584 		if (error) {
1585 			return error;
1586 		}
1587 		/*
1588 		 * Remember this buffer and block for the next call.
1589 		 */
1590 		if (rbpp) {
1591 			*rbpp = bp;
1592 			*rsb = sb;
1593 		}
1594 	}
1595 	/*
1596 	 * Point to the summary information & copy it out.
1597 	 */
1598 	sp = XFS_SUMPTR(mp, bp, so);
1599 	*sum = *sp;
1600 	/*
1601 	 * Drop the buffer if we're not asked to remember it.
1602 	 */
1603 	if (!rbpp)
1604 		xfs_trans_brelse(tp, bp);
1605 	return 0;
1606 }
1607 
1608 /*
1609  * Set the given range of bitmap bits to the given value.
1610  * Do whatever I/O and logging is required.
1611  */
1612 STATIC int				/* error */
1613 xfs_rtmodify_range(
1614 	xfs_mount_t	*mp,		/* file system mount point */
1615 	xfs_trans_t	*tp,		/* transaction pointer */
1616 	xfs_rtblock_t	start,		/* starting block to modify */
1617 	xfs_extlen_t	len,		/* length of extent to modify */
1618 	int		val)		/* 1 for free, 0 for allocated */
1619 {
1620 	xfs_rtword_t	*b;		/* current word in buffer */
1621 	int		bit;		/* bit number in the word */
1622 	xfs_rtblock_t	block;		/* bitmap block number */
1623 	xfs_buf_t	*bp;		/* buf for the block */
1624 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1625 	int		error;		/* error value */
1626 	xfs_rtword_t	*first;		/* first used word in the buffer */
1627 	int		i;		/* current bit number rel. to start */
1628 	int		lastbit;	/* last useful bit in word */
1629 	xfs_rtword_t	mask;		/* mask o frelevant bits for value */
1630 	int		word;		/* word number in the buffer */
1631 
1632 	/*
1633 	 * Compute starting bitmap block number.
1634 	 */
1635 	block = XFS_BITTOBLOCK(mp, start);
1636 	/*
1637 	 * Read the bitmap block, and point to its data.
1638 	 */
1639 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1640 	if (error) {
1641 		return error;
1642 	}
1643 	bufp = bp->b_addr;
1644 	/*
1645 	 * Compute the starting word's address, and starting bit.
1646 	 */
1647 	word = XFS_BITTOWORD(mp, start);
1648 	first = b = &bufp[word];
1649 	bit = (int)(start & (XFS_NBWORD - 1));
1650 	/*
1651 	 * 0 (allocated) => all zeroes; 1 (free) => all ones.
1652 	 */
1653 	val = -val;
1654 	/*
1655 	 * If not starting on a word boundary, deal with the first
1656 	 * (partial) word.
1657 	 */
1658 	if (bit) {
1659 		/*
1660 		 * Compute first bit not changed and mask of relevant bits.
1661 		 */
1662 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1663 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1664 		/*
1665 		 * Set/clear the active bits.
1666 		 */
1667 		if (val)
1668 			*b |= mask;
1669 		else
1670 			*b &= ~mask;
1671 		i = lastbit - bit;
1672 		/*
1673 		 * Go on to the next block if that's where the next word is
1674 		 * and we need the next word.
1675 		 */
1676 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1677 			/*
1678 			 * Log the changed part of this block.
1679 			 * Get the next one.
1680 			 */
1681 			xfs_trans_log_buf(tp, bp,
1682 				(uint)((char *)first - (char *)bufp),
1683 				(uint)((char *)b - (char *)bufp));
1684 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1685 			if (error) {
1686 				return error;
1687 			}
1688 			first = b = bufp = bp->b_addr;
1689 			word = 0;
1690 		} else {
1691 			/*
1692 			 * Go on to the next word in the buffer
1693 			 */
1694 			b++;
1695 		}
1696 	} else {
1697 		/*
1698 		 * Starting on a word boundary, no partial word.
1699 		 */
1700 		i = 0;
1701 	}
1702 	/*
1703 	 * Loop over whole words in buffers.  When we use up one buffer
1704 	 * we move on to the next one.
1705 	 */
1706 	while (len - i >= XFS_NBWORD) {
1707 		/*
1708 		 * Set the word value correctly.
1709 		 */
1710 		*b = val;
1711 		i += XFS_NBWORD;
1712 		/*
1713 		 * Go on to the next block if that's where the next word is
1714 		 * and we need the next word.
1715 		 */
1716 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1717 			/*
1718 			 * Log the changed part of this block.
1719 			 * Get the next one.
1720 			 */
1721 			xfs_trans_log_buf(tp, bp,
1722 				(uint)((char *)first - (char *)bufp),
1723 				(uint)((char *)b - (char *)bufp));
1724 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1725 			if (error) {
1726 				return error;
1727 			}
1728 			first = b = bufp = bp->b_addr;
1729 			word = 0;
1730 		} else {
1731 			/*
1732 			 * Go on to the next word in the buffer
1733 			 */
1734 			b++;
1735 		}
1736 	}
1737 	/*
1738 	 * If not ending on a word boundary, deal with the last
1739 	 * (partial) word.
1740 	 */
1741 	if ((lastbit = len - i)) {
1742 		/*
1743 		 * Compute a mask of relevant bits.
1744 		 */
1745 		bit = 0;
1746 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1747 		/*
1748 		 * Set/clear the active bits.
1749 		 */
1750 		if (val)
1751 			*b |= mask;
1752 		else
1753 			*b &= ~mask;
1754 		b++;
1755 	}
1756 	/*
1757 	 * Log any remaining changed bytes.
1758 	 */
1759 	if (b > first)
1760 		xfs_trans_log_buf(tp, bp, (uint)((char *)first - (char *)bufp),
1761 			(uint)((char *)b - (char *)bufp - 1));
1762 	return 0;
1763 }
1764 
1765 /*
1766  * Read and modify the summary information for a given extent size,
1767  * bitmap block combination.
1768  * Keeps track of a current summary block, so we don't keep reading
1769  * it from the buffer cache.
1770  */
1771 STATIC int				/* error */
1772 xfs_rtmodify_summary(
1773 	xfs_mount_t	*mp,		/* file system mount point */
1774 	xfs_trans_t	*tp,		/* transaction pointer */
1775 	int		log,		/* log2 of extent size */
1776 	xfs_rtblock_t	bbno,		/* bitmap block number */
1777 	int		delta,		/* change to make to summary info */
1778 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1779 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1780 {
1781 	xfs_buf_t	*bp;		/* buffer for the summary block */
1782 	int		error;		/* error value */
1783 	xfs_fsblock_t	sb;		/* summary fsblock */
1784 	int		so;		/* index into the summary file */
1785 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1786 
1787 	/*
1788 	 * Compute entry number in the summary file.
1789 	 */
1790 	so = XFS_SUMOFFS(mp, log, bbno);
1791 	/*
1792 	 * Compute the block number in the summary file.
1793 	 */
1794 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1795 	/*
1796 	 * If we have an old buffer, and the block number matches, use that.
1797 	 */
1798 	if (rbpp && *rbpp && *rsb == sb)
1799 		bp = *rbpp;
1800 	/*
1801 	 * Otherwise we have to get the buffer.
1802 	 */
1803 	else {
1804 		/*
1805 		 * If there was an old one, get rid of it first.
1806 		 */
1807 		if (rbpp && *rbpp)
1808 			xfs_trans_brelse(tp, *rbpp);
1809 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1810 		if (error) {
1811 			return error;
1812 		}
1813 		/*
1814 		 * Remember this buffer and block for the next call.
1815 		 */
1816 		if (rbpp) {
1817 			*rbpp = bp;
1818 			*rsb = sb;
1819 		}
1820 	}
1821 	/*
1822 	 * Point to the summary information, modify and log it.
1823 	 */
1824 	sp = XFS_SUMPTR(mp, bp, so);
1825 	*sp += delta;
1826 	xfs_trans_log_buf(tp, bp, (uint)((char *)sp - (char *)bp->b_addr),
1827 		(uint)((char *)sp - (char *)bp->b_addr + sizeof(*sp) - 1));
1828 	return 0;
1829 }
1830 
1831 /*
1832  * Visible (exported) functions.
1833  */
1834 
1835 /*
1836  * Grow the realtime area of the filesystem.
1837  */
1838 int
1839 xfs_growfs_rt(
1840 	xfs_mount_t	*mp,		/* mount point for filesystem */
1841 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
1842 {
1843 	xfs_rtblock_t	bmbno;		/* bitmap block number */
1844 	xfs_buf_t	*bp;		/* temporary buffer */
1845 	int		error;		/* error return value */
1846 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
1847 	xfs_drfsbno_t	nrblocks;	/* new number of realtime blocks */
1848 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
1849 	xfs_drtbno_t	nrextents;	/* new number of realtime extents */
1850 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
1851 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
1852 	uint		nrsumlevels;	/* new rt summary levels */
1853 	uint		nrsumsize;	/* new size of rt summary, bytes */
1854 	xfs_sb_t	*nsbp;		/* new superblock */
1855 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
1856 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
1857 	xfs_sb_t	*sbp;		/* old superblock */
1858 	xfs_fsblock_t	sumbno;		/* summary block number */
1859 
1860 	sbp = &mp->m_sb;
1861 	/*
1862 	 * Initial error checking.
1863 	 */
1864 	if (!capable(CAP_SYS_ADMIN))
1865 		return XFS_ERROR(EPERM);
1866 	if (mp->m_rtdev_targp == NULL || mp->m_rbmip == NULL ||
1867 	    (nrblocks = in->newblocks) <= sbp->sb_rblocks ||
1868 	    (sbp->sb_rblocks && (in->extsize != sbp->sb_rextsize)))
1869 		return XFS_ERROR(EINVAL);
1870 	if ((error = xfs_sb_validate_fsb_count(sbp, nrblocks)))
1871 		return error;
1872 	/*
1873 	 * Read in the last block of the device, make sure it exists.
1874 	 */
1875 	bp = xfs_buf_read_uncached(mp, mp->m_rtdev_targp,
1876 				XFS_FSB_TO_BB(mp, nrblocks - 1),
1877 				XFS_FSB_TO_B(mp, 1), 0);
1878 	if (!bp)
1879 		return EIO;
1880 	xfs_buf_relse(bp);
1881 
1882 	/*
1883 	 * Calculate new parameters.  These are the final values to be reached.
1884 	 */
1885 	nrextents = nrblocks;
1886 	do_div(nrextents, in->extsize);
1887 	nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
1888 	nrextslog = xfs_highbit32(nrextents);
1889 	nrsumlevels = nrextslog + 1;
1890 	nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
1891 	nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1892 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1893 	/*
1894 	 * New summary size can't be more than half the size of
1895 	 * the log.  This prevents us from getting a log overflow,
1896 	 * since we'll log basically the whole summary file at once.
1897 	 */
1898 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
1899 		return XFS_ERROR(EINVAL);
1900 	/*
1901 	 * Get the old block counts for bitmap and summary inodes.
1902 	 * These can't change since other growfs callers are locked out.
1903 	 */
1904 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_d.di_size);
1905 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_d.di_size);
1906 	/*
1907 	 * Allocate space to the bitmap and summary files, as necessary.
1908 	 */
1909 	error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
1910 	if (error)
1911 		return error;
1912 	error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
1913 	if (error)
1914 		return error;
1915 	/*
1916 	 * Allocate a new (fake) mount/sb.
1917 	 */
1918 	nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
1919 	/*
1920 	 * Loop over the bitmap blocks.
1921 	 * We will do everything one bitmap block at a time.
1922 	 * Skip the current block if it is exactly full.
1923 	 * This also deals with the case where there were no rtextents before.
1924 	 */
1925 	for (bmbno = sbp->sb_rbmblocks -
1926 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1927 	     bmbno < nrbmblocks;
1928 	     bmbno++) {
1929 		xfs_trans_t	*tp;
1930 		int		cancelflags = 0;
1931 
1932 		*nmp = *mp;
1933 		nsbp = &nmp->m_sb;
1934 		/*
1935 		 * Calculate new sb and mount fields for this round.
1936 		 */
1937 		nsbp->sb_rextsize = in->extsize;
1938 		nsbp->sb_rbmblocks = bmbno + 1;
1939 		nsbp->sb_rblocks =
1940 			XFS_RTMIN(nrblocks,
1941 				  nsbp->sb_rbmblocks * NBBY *
1942 				  nsbp->sb_blocksize * nsbp->sb_rextsize);
1943 		nsbp->sb_rextents = nsbp->sb_rblocks;
1944 		do_div(nsbp->sb_rextents, nsbp->sb_rextsize);
1945 		ASSERT(nsbp->sb_rextents != 0);
1946 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1947 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1948 		nrsumsize =
1949 			(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
1950 			nsbp->sb_rbmblocks;
1951 		nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1952 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1953 		/*
1954 		 * Start a transaction, get the log reservation.
1955 		 */
1956 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_FREE);
1957 		if ((error = xfs_trans_reserve(tp, 0,
1958 				XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
1959 			goto error_cancel;
1960 		/*
1961 		 * Lock out other callers by grabbing the bitmap inode lock.
1962 		 */
1963 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
1964 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
1965 		/*
1966 		 * Update the bitmap inode's size.
1967 		 */
1968 		mp->m_rbmip->i_d.di_size =
1969 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1970 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1971 		cancelflags |= XFS_TRANS_ABORT;
1972 		/*
1973 		 * Get the summary inode into the transaction.
1974 		 */
1975 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL);
1976 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
1977 		/*
1978 		 * Update the summary inode's size.
1979 		 */
1980 		mp->m_rsumip->i_d.di_size = nmp->m_rsumsize;
1981 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
1982 		/*
1983 		 * Copy summary data from old to new sizes.
1984 		 * Do this when the real size (not block-aligned) changes.
1985 		 */
1986 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
1987 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
1988 			error = xfs_rtcopy_summary(mp, nmp, tp);
1989 			if (error)
1990 				goto error_cancel;
1991 		}
1992 		/*
1993 		 * Update superblock fields.
1994 		 */
1995 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
1996 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
1997 				nsbp->sb_rextsize - sbp->sb_rextsize);
1998 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
1999 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
2000 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
2001 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
2002 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
2003 				nsbp->sb_rblocks - sbp->sb_rblocks);
2004 		if (nsbp->sb_rextents != sbp->sb_rextents)
2005 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
2006 				nsbp->sb_rextents - sbp->sb_rextents);
2007 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
2008 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
2009 				nsbp->sb_rextslog - sbp->sb_rextslog);
2010 		/*
2011 		 * Free new extent.
2012 		 */
2013 		bp = NULL;
2014 		error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
2015 			nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
2016 		if (error) {
2017 error_cancel:
2018 			xfs_trans_cancel(tp, cancelflags);
2019 			break;
2020 		}
2021 		/*
2022 		 * Mark more blocks free in the superblock.
2023 		 */
2024 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
2025 			nsbp->sb_rextents - sbp->sb_rextents);
2026 		/*
2027 		 * Update mp values into the real mp structure.
2028 		 */
2029 		mp->m_rsumlevels = nrsumlevels;
2030 		mp->m_rsumsize = nrsumsize;
2031 
2032 		error = xfs_trans_commit(tp, 0);
2033 		if (error)
2034 			break;
2035 	}
2036 
2037 	/*
2038 	 * Free the fake mp structure.
2039 	 */
2040 	kmem_free(nmp);
2041 
2042 	return error;
2043 }
2044 
2045 /*
2046  * Allocate an extent in the realtime subvolume, with the usual allocation
2047  * parameters.  The length units are all in realtime extents, as is the
2048  * result block number.
2049  */
2050 int					/* error */
2051 xfs_rtallocate_extent(
2052 	xfs_trans_t	*tp,		/* transaction pointer */
2053 	xfs_rtblock_t	bno,		/* starting block number to allocate */
2054 	xfs_extlen_t	minlen,		/* minimum length to allocate */
2055 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
2056 	xfs_extlen_t	*len,		/* out: actual length allocated */
2057 	xfs_alloctype_t	type,		/* allocation type XFS_ALLOCTYPE... */
2058 	int		wasdel,		/* was a delayed allocation extent */
2059 	xfs_extlen_t	prod,		/* extent product factor */
2060 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
2061 {
2062 	xfs_mount_t	*mp = tp->t_mountp;
2063 	int		error;		/* error value */
2064 	xfs_rtblock_t	r;		/* result allocated block */
2065 	xfs_fsblock_t	sb;		/* summary file block number */
2066 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2067 
2068 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2069 	ASSERT(minlen > 0 && minlen <= maxlen);
2070 
2071 	/*
2072 	 * If prod is set then figure out what to do to minlen and maxlen.
2073 	 */
2074 	if (prod > 1) {
2075 		xfs_extlen_t	i;
2076 
2077 		if ((i = maxlen % prod))
2078 			maxlen -= i;
2079 		if ((i = minlen % prod))
2080 			minlen += prod - i;
2081 		if (maxlen < minlen) {
2082 			*rtblock = NULLRTBLOCK;
2083 			return 0;
2084 		}
2085 	}
2086 
2087 	sumbp = NULL;
2088 	/*
2089 	 * Allocate by size, or near another block, or exactly at some block.
2090 	 */
2091 	switch (type) {
2092 	case XFS_ALLOCTYPE_ANY_AG:
2093 		error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
2094 				&sumbp,	&sb, prod, &r);
2095 		break;
2096 	case XFS_ALLOCTYPE_NEAR_BNO:
2097 		error = xfs_rtallocate_extent_near(mp, tp, bno, minlen, maxlen,
2098 				len, &sumbp, &sb, prod, &r);
2099 		break;
2100 	case XFS_ALLOCTYPE_THIS_BNO:
2101 		error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen,
2102 				len, &sumbp, &sb, prod, &r);
2103 		break;
2104 	default:
2105 		error = EIO;
2106 		ASSERT(0);
2107 	}
2108 	if (error)
2109 		return error;
2110 
2111 	/*
2112 	 * If it worked, update the superblock.
2113 	 */
2114 	if (r != NULLRTBLOCK) {
2115 		long	slen = (long)*len;
2116 
2117 		ASSERT(*len >= minlen && *len <= maxlen);
2118 		if (wasdel)
2119 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
2120 		else
2121 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
2122 	}
2123 	*rtblock = r;
2124 	return 0;
2125 }
2126 
2127 /*
2128  * Free an extent in the realtime subvolume.  Length is expressed in
2129  * realtime extents, as is the block number.
2130  */
2131 int					/* error */
2132 xfs_rtfree_extent(
2133 	xfs_trans_t	*tp,		/* transaction pointer */
2134 	xfs_rtblock_t	bno,		/* starting block number to free */
2135 	xfs_extlen_t	len)		/* length of extent freed */
2136 {
2137 	int		error;		/* error value */
2138 	xfs_mount_t	*mp;		/* file system mount structure */
2139 	xfs_fsblock_t	sb;		/* summary file block number */
2140 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2141 
2142 	mp = tp->t_mountp;
2143 
2144 	ASSERT(mp->m_rbmip->i_itemp != NULL);
2145 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2146 
2147 #if defined(__KERNEL__) && defined(DEBUG)
2148 	/*
2149 	 * Check to see that this whole range is currently allocated.
2150 	 */
2151 	{
2152 		int	stat;		/* result from checking range */
2153 
2154 		error = xfs_rtcheck_alloc_range(mp, tp, bno, len, &stat);
2155 		if (error) {
2156 			return error;
2157 		}
2158 		ASSERT(stat);
2159 	}
2160 #endif
2161 	sumbp = NULL;
2162 	/*
2163 	 * Free the range of realtime blocks.
2164 	 */
2165 	error = xfs_rtfree_range(mp, tp, bno, len, &sumbp, &sb);
2166 	if (error) {
2167 		return error;
2168 	}
2169 	/*
2170 	 * Mark more blocks free in the superblock.
2171 	 */
2172 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, (long)len);
2173 	/*
2174 	 * If we've now freed all the blocks, reset the file sequence
2175 	 * number to 0.
2176 	 */
2177 	if (tp->t_frextents_delta + mp->m_sb.sb_frextents ==
2178 	    mp->m_sb.sb_rextents) {
2179 		if (!(mp->m_rbmip->i_d.di_flags & XFS_DIFLAG_NEWRTBM))
2180 			mp->m_rbmip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2181 		*(__uint64_t *)&mp->m_rbmip->i_d.di_atime = 0;
2182 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
2183 	}
2184 	return 0;
2185 }
2186 
2187 /*
2188  * Initialize realtime fields in the mount structure.
2189  */
2190 int				/* error */
2191 xfs_rtmount_init(
2192 	xfs_mount_t	*mp)	/* file system mount structure */
2193 {
2194 	xfs_buf_t	*bp;	/* buffer for last block of subvolume */
2195 	xfs_daddr_t	d;	/* address of last block of subvolume */
2196 	xfs_sb_t	*sbp;	/* filesystem superblock copy in mount */
2197 
2198 	sbp = &mp->m_sb;
2199 	if (sbp->sb_rblocks == 0)
2200 		return 0;
2201 	if (mp->m_rtdev_targp == NULL) {
2202 		xfs_warn(mp,
2203 	"Filesystem has a realtime volume, use rtdev=device option");
2204 		return XFS_ERROR(ENODEV);
2205 	}
2206 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
2207 	mp->m_rsumsize =
2208 		(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
2209 		sbp->sb_rbmblocks;
2210 	mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
2211 	mp->m_rbmip = mp->m_rsumip = NULL;
2212 	/*
2213 	 * Check that the realtime section is an ok size.
2214 	 */
2215 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
2216 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
2217 		xfs_warn(mp, "realtime mount -- %llu != %llu",
2218 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
2219 			(unsigned long long) mp->m_sb.sb_rblocks);
2220 		return XFS_ERROR(EFBIG);
2221 	}
2222 	bp = xfs_buf_read_uncached(mp, mp->m_rtdev_targp,
2223 					d - XFS_FSB_TO_BB(mp, 1),
2224 					XFS_FSB_TO_B(mp, 1), 0);
2225 	if (!bp) {
2226 		xfs_warn(mp, "realtime device size check failed");
2227 		return EIO;
2228 	}
2229 	xfs_buf_relse(bp);
2230 	return 0;
2231 }
2232 
2233 /*
2234  * Get the bitmap and summary inodes into the mount structure
2235  * at mount time.
2236  */
2237 int					/* error */
2238 xfs_rtmount_inodes(
2239 	xfs_mount_t	*mp)		/* file system mount structure */
2240 {
2241 	int		error;		/* error return value */
2242 	xfs_sb_t	*sbp;
2243 
2244 	sbp = &mp->m_sb;
2245 	if (sbp->sb_rbmino == NULLFSINO)
2246 		return 0;
2247 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
2248 	if (error)
2249 		return error;
2250 	ASSERT(mp->m_rbmip != NULL);
2251 	ASSERT(sbp->sb_rsumino != NULLFSINO);
2252 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
2253 	if (error) {
2254 		IRELE(mp->m_rbmip);
2255 		return error;
2256 	}
2257 	ASSERT(mp->m_rsumip != NULL);
2258 	return 0;
2259 }
2260 
2261 void
2262 xfs_rtunmount_inodes(
2263 	struct xfs_mount	*mp)
2264 {
2265 	if (mp->m_rbmip)
2266 		IRELE(mp->m_rbmip);
2267 	if (mp->m_rsumip)
2268 		IRELE(mp->m_rsumip);
2269 }
2270 
2271 /*
2272  * Pick an extent for allocation at the start of a new realtime file.
2273  * Use the sequence number stored in the atime field of the bitmap inode.
2274  * Translate this to a fraction of the rtextents, and return the product
2275  * of rtextents and the fraction.
2276  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
2277  */
2278 int					/* error */
2279 xfs_rtpick_extent(
2280 	xfs_mount_t	*mp,		/* file system mount point */
2281 	xfs_trans_t	*tp,		/* transaction pointer */
2282 	xfs_extlen_t	len,		/* allocation length (rtextents) */
2283 	xfs_rtblock_t	*pick)		/* result rt extent */
2284 {
2285 	xfs_rtblock_t	b;		/* result block */
2286 	int		log2;		/* log of sequence number */
2287 	__uint64_t	resid;		/* residual after log removed */
2288 	__uint64_t	seq;		/* sequence number of file creation */
2289 	__uint64_t	*seqp;		/* pointer to seqno in inode */
2290 
2291 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2292 
2293 	seqp = (__uint64_t *)&mp->m_rbmip->i_d.di_atime;
2294 	if (!(mp->m_rbmip->i_d.di_flags & XFS_DIFLAG_NEWRTBM)) {
2295 		mp->m_rbmip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2296 		*seqp = 0;
2297 	}
2298 	seq = *seqp;
2299 	if ((log2 = xfs_highbit64(seq)) == -1)
2300 		b = 0;
2301 	else {
2302 		resid = seq - (1ULL << log2);
2303 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
2304 		    (log2 + 1);
2305 		if (b >= mp->m_sb.sb_rextents)
2306 			b = do_mod(b, mp->m_sb.sb_rextents);
2307 		if (b + len > mp->m_sb.sb_rextents)
2308 			b = mp->m_sb.sb_rextents - len;
2309 	}
2310 	*seqp = seq + 1;
2311 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
2312 	*pick = b;
2313 	return 0;
2314 }
2315