xref: /openbmc/linux/fs/xfs/xfs_rtalloc.c (revision 1fa6ac37)
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_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_fsops.h"
43 #include "xfs_error.h"
44 #include "xfs_rw.h"
45 #include "xfs_inode_item.h"
46 #include "xfs_trans_space.h"
47 #include "xfs_utils.h"
48 #include "xfs_trace.h"
49 
50 
51 /*
52  * Prototypes for internal functions.
53  */
54 
55 
56 STATIC int xfs_rtallocate_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
57 		xfs_extlen_t, xfs_buf_t **, xfs_fsblock_t *);
58 STATIC int xfs_rtany_summary(xfs_mount_t *, xfs_trans_t *, int, int,
59 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, int *);
60 STATIC int xfs_rtcheck_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
61 		xfs_extlen_t, int, xfs_rtblock_t *, int *);
62 STATIC int xfs_rtfind_back(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
63 		xfs_rtblock_t, xfs_rtblock_t *);
64 STATIC int xfs_rtfind_forw(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
65 		xfs_rtblock_t, xfs_rtblock_t *);
66 STATIC int xfs_rtget_summary( xfs_mount_t *, xfs_trans_t *, int,
67 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, xfs_suminfo_t *);
68 STATIC int xfs_rtmodify_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
69 		xfs_extlen_t, int);
70 STATIC int xfs_rtmodify_summary(xfs_mount_t *, xfs_trans_t *, int,
71 		xfs_rtblock_t, int, xfs_buf_t **, xfs_fsblock_t *);
72 
73 /*
74  * Internal functions.
75  */
76 
77 /*
78  * Allocate space to the bitmap or summary file, and zero it, for growfs.
79  */
80 STATIC int				/* error */
81 xfs_growfs_rt_alloc(
82 	xfs_mount_t	*mp,		/* file system mount point */
83 	xfs_extlen_t	oblocks,	/* old count of blocks */
84 	xfs_extlen_t	nblocks,	/* new count of blocks */
85 	xfs_ino_t	ino)		/* inode number (bitmap/summary) */
86 {
87 	xfs_fileoff_t	bno;		/* block number in file */
88 	xfs_buf_t	*bp;		/* temporary buffer for zeroing */
89 	int		committed;	/* transaction committed flag */
90 	xfs_daddr_t	d;		/* disk block address */
91 	int		error;		/* error return value */
92 	xfs_fsblock_t	firstblock;	/* first block allocated in xaction */
93 	xfs_bmap_free_t	flist;		/* list of freed blocks */
94 	xfs_fsblock_t	fsbno;		/* filesystem block for bno */
95 	xfs_inode_t	*ip;		/* pointer to incore inode */
96 	xfs_bmbt_irec_t	map;		/* block map output */
97 	int		nmap;		/* number of block maps */
98 	int		resblks;	/* space reservation */
99 
100 	/*
101 	 * Allocate space to the file, as necessary.
102 	 */
103 	while (oblocks < nblocks) {
104 		int		cancelflags = 0;
105 		xfs_trans_t	*tp;
106 
107 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ALLOC);
108 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
109 		/*
110 		 * Reserve space & log for one extent added to the file.
111 		 */
112 		if ((error = xfs_trans_reserve(tp, resblks,
113 				XFS_GROWRTALLOC_LOG_RES(mp), 0,
114 				XFS_TRANS_PERM_LOG_RES,
115 				XFS_DEFAULT_PERM_LOG_COUNT)))
116 			goto error_cancel;
117 		cancelflags = XFS_TRANS_RELEASE_LOG_RES;
118 		/*
119 		 * Lock the inode.
120 		 */
121 		if ((error = xfs_trans_iget(mp, tp, ino, 0,
122 						XFS_ILOCK_EXCL, &ip)))
123 			goto error_cancel;
124 		xfs_bmap_init(&flist, &firstblock);
125 		/*
126 		 * Allocate blocks to the bitmap file.
127 		 */
128 		nmap = 1;
129 		cancelflags |= XFS_TRANS_ABORT;
130 		error = xfs_bmapi(tp, ip, oblocks, nblocks - oblocks,
131 			XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, &firstblock,
132 			resblks, &map, &nmap, &flist, NULL);
133 		if (!error && nmap < 1)
134 			error = XFS_ERROR(ENOSPC);
135 		if (error)
136 			goto error_cancel;
137 		/*
138 		 * Free any blocks freed up in the transaction, then commit.
139 		 */
140 		error = xfs_bmap_finish(&tp, &flist, &committed);
141 		if (error)
142 			goto error_cancel;
143 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
144 		if (error)
145 			goto error;
146 		/*
147 		 * Now we need to clear the allocated blocks.
148 		 * Do this one block per transaction, to keep it simple.
149 		 */
150 		cancelflags = 0;
151 		for (bno = map.br_startoff, fsbno = map.br_startblock;
152 		     bno < map.br_startoff + map.br_blockcount;
153 		     bno++, fsbno++) {
154 			tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ZERO);
155 			/*
156 			 * Reserve log for one block zeroing.
157 			 */
158 			if ((error = xfs_trans_reserve(tp, 0,
159 					XFS_GROWRTZERO_LOG_RES(mp), 0, 0, 0)))
160 				goto error_cancel;
161 			/*
162 			 * Lock the bitmap inode.
163 			 */
164 			if ((error = xfs_trans_iget(mp, tp, ino, 0,
165 							XFS_ILOCK_EXCL, &ip)))
166 				goto error_cancel;
167 			/*
168 			 * Get a buffer for the block.
169 			 */
170 			d = XFS_FSB_TO_DADDR(mp, fsbno);
171 			bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
172 				mp->m_bsize, 0);
173 			if (bp == NULL) {
174 				error = XFS_ERROR(EIO);
175 error_cancel:
176 				xfs_trans_cancel(tp, cancelflags);
177 				goto error;
178 			}
179 			memset(XFS_BUF_PTR(bp), 0, mp->m_sb.sb_blocksize);
180 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
181 			/*
182 			 * Commit the transaction.
183 			 */
184 			error = xfs_trans_commit(tp, 0);
185 			if (error)
186 				goto error;
187 		}
188 		/*
189 		 * Go on to the next extent, if any.
190 		 */
191 		oblocks = map.br_startoff + map.br_blockcount;
192 	}
193 	return 0;
194 error:
195 	return error;
196 }
197 
198 /*
199  * Attempt to allocate an extent minlen<=len<=maxlen starting from
200  * bitmap block bbno.  If we don't get maxlen then use prod to trim
201  * the length, if given.  Returns error; returns starting block in *rtblock.
202  * The lengths are all in rtextents.
203  */
204 STATIC int				/* error */
205 xfs_rtallocate_extent_block(
206 	xfs_mount_t	*mp,		/* file system mount point */
207 	xfs_trans_t	*tp,		/* transaction pointer */
208 	xfs_rtblock_t	bbno,		/* bitmap block number */
209 	xfs_extlen_t	minlen,		/* minimum length to allocate */
210 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
211 	xfs_extlen_t	*len,		/* out: actual length allocated */
212 	xfs_rtblock_t	*nextp,		/* out: next block to try */
213 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
214 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
215 	xfs_extlen_t	prod,		/* extent product factor */
216 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
217 {
218 	xfs_rtblock_t	besti;		/* best rtblock found so far */
219 	xfs_rtblock_t	bestlen;	/* best length found so far */
220 	xfs_rtblock_t	end;		/* last rtblock in chunk */
221 	int		error;		/* error value */
222 	xfs_rtblock_t	i;		/* current rtblock trying */
223 	xfs_rtblock_t	next;		/* next rtblock to try */
224 	int		stat;		/* status from internal calls */
225 
226 	/*
227 	 * Loop over all the extents starting in this bitmap block,
228 	 * looking for one that's long enough.
229 	 */
230 	for (i = XFS_BLOCKTOBIT(mp, bbno), besti = -1, bestlen = 0,
231 		end = XFS_BLOCKTOBIT(mp, bbno + 1) - 1;
232 	     i <= end;
233 	     i++) {
234 		/*
235 		 * See if there's a free extent of maxlen starting at i.
236 		 * If it's not so then next will contain the first non-free.
237 		 */
238 		error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
239 		if (error) {
240 			return error;
241 		}
242 		if (stat) {
243 			/*
244 			 * i for maxlen is all free, allocate and return that.
245 			 */
246 			error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
247 				rsb);
248 			if (error) {
249 				return error;
250 			}
251 			*len = maxlen;
252 			*rtblock = i;
253 			return 0;
254 		}
255 		/*
256 		 * In the case where we have a variable-sized allocation
257 		 * request, figure out how big this free piece is,
258 		 * and if it's big enough for the minimum, and the best
259 		 * so far, remember it.
260 		 */
261 		if (minlen < maxlen) {
262 			xfs_rtblock_t	thislen;	/* this extent size */
263 
264 			thislen = next - i;
265 			if (thislen >= minlen && thislen > bestlen) {
266 				besti = i;
267 				bestlen = thislen;
268 			}
269 		}
270 		/*
271 		 * If not done yet, find the start of the next free space.
272 		 */
273 		if (next < end) {
274 			error = xfs_rtfind_forw(mp, tp, next, end, &i);
275 			if (error) {
276 				return error;
277 			}
278 		} else
279 			break;
280 	}
281 	/*
282 	 * Searched the whole thing & didn't find a maxlen free extent.
283 	 */
284 	if (minlen < maxlen && besti != -1) {
285 		xfs_extlen_t	p;	/* amount to trim length by */
286 
287 		/*
288 		 * If size should be a multiple of prod, make that so.
289 		 */
290 		if (prod > 1 && (p = do_mod(bestlen, prod)))
291 			bestlen -= p;
292 		/*
293 		 * Allocate besti for bestlen & return that.
294 		 */
295 		error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
296 		if (error) {
297 			return error;
298 		}
299 		*len = bestlen;
300 		*rtblock = besti;
301 		return 0;
302 	}
303 	/*
304 	 * Allocation failed.  Set *nextp to the next block to try.
305 	 */
306 	*nextp = next;
307 	*rtblock = NULLRTBLOCK;
308 	return 0;
309 }
310 
311 /*
312  * Allocate an extent of length minlen<=len<=maxlen, starting at block
313  * bno.  If we don't get maxlen then use prod to trim the length, if given.
314  * Returns error; returns starting block in *rtblock.
315  * The lengths are all in rtextents.
316  */
317 STATIC int				/* error */
318 xfs_rtallocate_extent_exact(
319 	xfs_mount_t	*mp,		/* file system mount point */
320 	xfs_trans_t	*tp,		/* transaction pointer */
321 	xfs_rtblock_t	bno,		/* starting block number to allocate */
322 	xfs_extlen_t	minlen,		/* minimum length to allocate */
323 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
324 	xfs_extlen_t	*len,		/* out: actual length allocated */
325 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
326 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
327 	xfs_extlen_t	prod,		/* extent product factor */
328 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
329 {
330 	int		error;		/* error value */
331 	xfs_extlen_t	i;		/* extent length trimmed due to prod */
332 	int		isfree;		/* extent is free */
333 	xfs_rtblock_t	next;		/* next block to try (dummy) */
334 
335 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
336 	/*
337 	 * Check if the range in question (for maxlen) is free.
338 	 */
339 	error = xfs_rtcheck_range(mp, tp, bno, maxlen, 1, &next, &isfree);
340 	if (error) {
341 		return error;
342 	}
343 	if (isfree) {
344 		/*
345 		 * If it is, allocate it and return success.
346 		 */
347 		error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
348 		if (error) {
349 			return error;
350 		}
351 		*len = maxlen;
352 		*rtblock = bno;
353 		return 0;
354 	}
355 	/*
356 	 * If not, allocate what there is, if it's at least minlen.
357 	 */
358 	maxlen = next - bno;
359 	if (maxlen < minlen) {
360 		/*
361 		 * Failed, return failure status.
362 		 */
363 		*rtblock = NULLRTBLOCK;
364 		return 0;
365 	}
366 	/*
367 	 * Trim off tail of extent, if prod is specified.
368 	 */
369 	if (prod > 1 && (i = maxlen % prod)) {
370 		maxlen -= i;
371 		if (maxlen < minlen) {
372 			/*
373 			 * Now we can't do it, return failure status.
374 			 */
375 			*rtblock = NULLRTBLOCK;
376 			return 0;
377 		}
378 	}
379 	/*
380 	 * Allocate what we can and return it.
381 	 */
382 	error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
383 	if (error) {
384 		return error;
385 	}
386 	*len = maxlen;
387 	*rtblock = bno;
388 	return 0;
389 }
390 
391 /*
392  * Allocate an extent of length minlen<=len<=maxlen, starting as near
393  * to bno as possible.  If we don't get maxlen then use prod to trim
394  * the length, if given.  The lengths are all in rtextents.
395  */
396 STATIC int				/* error */
397 xfs_rtallocate_extent_near(
398 	xfs_mount_t	*mp,		/* file system mount point */
399 	xfs_trans_t	*tp,		/* transaction pointer */
400 	xfs_rtblock_t	bno,		/* starting block number to allocate */
401 	xfs_extlen_t	minlen,		/* minimum length to allocate */
402 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
403 	xfs_extlen_t	*len,		/* out: actual length allocated */
404 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
405 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
406 	xfs_extlen_t	prod,		/* extent product factor */
407 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
408 {
409 	int		any;		/* any useful extents from summary */
410 	xfs_rtblock_t	bbno;		/* bitmap block number */
411 	int		error;		/* error value */
412 	int		i;		/* bitmap block offset (loop control) */
413 	int		j;		/* secondary loop control */
414 	int		log2len;	/* log2 of minlen */
415 	xfs_rtblock_t	n;		/* next block to try */
416 	xfs_rtblock_t	r;		/* result block */
417 
418 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
419 	/*
420 	 * If the block number given is off the end, silently set it to
421 	 * the last block.
422 	 */
423 	if (bno >= mp->m_sb.sb_rextents)
424 		bno = mp->m_sb.sb_rextents - 1;
425 	/*
426 	 * Try the exact allocation first.
427 	 */
428 	error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen, len,
429 		rbpp, rsb, prod, &r);
430 	if (error) {
431 		return error;
432 	}
433 	/*
434 	 * If the exact allocation worked, return that.
435 	 */
436 	if (r != NULLRTBLOCK) {
437 		*rtblock = r;
438 		return 0;
439 	}
440 	bbno = XFS_BITTOBLOCK(mp, bno);
441 	i = 0;
442 	ASSERT(minlen != 0);
443 	log2len = xfs_highbit32(minlen);
444 	/*
445 	 * Loop over all bitmap blocks (bbno + i is current block).
446 	 */
447 	for (;;) {
448 		/*
449 		 * Get summary information of extents of all useful levels
450 		 * starting in this bitmap block.
451 		 */
452 		error = xfs_rtany_summary(mp, tp, log2len, mp->m_rsumlevels - 1,
453 			bbno + i, rbpp, rsb, &any);
454 		if (error) {
455 			return error;
456 		}
457 		/*
458 		 * If there are any useful extents starting here, try
459 		 * allocating one.
460 		 */
461 		if (any) {
462 			/*
463 			 * On the positive side of the starting location.
464 			 */
465 			if (i >= 0) {
466 				/*
467 				 * Try to allocate an extent starting in
468 				 * this block.
469 				 */
470 				error = xfs_rtallocate_extent_block(mp, tp,
471 					bbno + i, minlen, maxlen, len, &n, rbpp,
472 					rsb, prod, &r);
473 				if (error) {
474 					return error;
475 				}
476 				/*
477 				 * If it worked, return it.
478 				 */
479 				if (r != NULLRTBLOCK) {
480 					*rtblock = r;
481 					return 0;
482 				}
483 			}
484 			/*
485 			 * On the negative side of the starting location.
486 			 */
487 			else {		/* i < 0 */
488 				/*
489 				 * Loop backwards through the bitmap blocks from
490 				 * the starting point-1 up to where we are now.
491 				 * There should be an extent which ends in this
492 				 * bitmap block and is long enough.
493 				 */
494 				for (j = -1; j > i; j--) {
495 					/*
496 					 * Grab the summary information for
497 					 * this bitmap block.
498 					 */
499 					error = xfs_rtany_summary(mp, tp,
500 						log2len, mp->m_rsumlevels - 1,
501 						bbno + j, rbpp, rsb, &any);
502 					if (error) {
503 						return error;
504 					}
505 					/*
506 					 * If there's no extent given in the
507 					 * summary that means the extent we
508 					 * found must carry over from an
509 					 * earlier block.  If there is an
510 					 * extent given, we've already tried
511 					 * that allocation, don't do it again.
512 					 */
513 					if (any)
514 						continue;
515 					error = xfs_rtallocate_extent_block(mp,
516 						tp, bbno + j, minlen, maxlen,
517 						len, &n, rbpp, rsb, prod, &r);
518 					if (error) {
519 						return error;
520 					}
521 					/*
522 					 * If it works, return the extent.
523 					 */
524 					if (r != NULLRTBLOCK) {
525 						*rtblock = r;
526 						return 0;
527 					}
528 				}
529 				/*
530 				 * There weren't intervening bitmap blocks
531 				 * with a long enough extent, or the
532 				 * allocation didn't work for some reason
533 				 * (i.e. it's a little * too short).
534 				 * Try to allocate from the summary block
535 				 * that we found.
536 				 */
537 				error = xfs_rtallocate_extent_block(mp, tp,
538 					bbno + i, minlen, maxlen, len, &n, rbpp,
539 					rsb, prod, &r);
540 				if (error) {
541 					return error;
542 				}
543 				/*
544 				 * If it works, return the extent.
545 				 */
546 				if (r != NULLRTBLOCK) {
547 					*rtblock = r;
548 					return 0;
549 				}
550 			}
551 		}
552 		/*
553 		 * Loop control.  If we were on the positive side, and there's
554 		 * still more blocks on the negative side, go there.
555 		 */
556 		if (i > 0 && (int)bbno - i >= 0)
557 			i = -i;
558 		/*
559 		 * If positive, and no more negative, but there are more
560 		 * positive, go there.
561 		 */
562 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
563 			i++;
564 		/*
565 		 * If negative or 0 (just started), and there are positive
566 		 * blocks to go, go there.  The 0 case moves to block 1.
567 		 */
568 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
569 			i = 1 - i;
570 		/*
571 		 * If negative or 0 and there are more negative blocks,
572 		 * go there.
573 		 */
574 		else if (i <= 0 && (int)bbno + i > 0)
575 			i--;
576 		/*
577 		 * Must be done.  Return failure.
578 		 */
579 		else
580 			break;
581 	}
582 	*rtblock = NULLRTBLOCK;
583 	return 0;
584 }
585 
586 /*
587  * Allocate an extent of length minlen<=len<=maxlen, with no position
588  * specified.  If we don't get maxlen then use prod to trim
589  * the length, if given.  The lengths are all in rtextents.
590  */
591 STATIC int				/* error */
592 xfs_rtallocate_extent_size(
593 	xfs_mount_t	*mp,		/* file system mount point */
594 	xfs_trans_t	*tp,		/* transaction pointer */
595 	xfs_extlen_t	minlen,		/* minimum length to allocate */
596 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
597 	xfs_extlen_t	*len,		/* out: actual length allocated */
598 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
599 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
600 	xfs_extlen_t	prod,		/* extent product factor */
601 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
602 {
603 	int		error;		/* error value */
604 	int		i;		/* bitmap block number */
605 	int		l;		/* level number (loop control) */
606 	xfs_rtblock_t	n;		/* next block to be tried */
607 	xfs_rtblock_t	r;		/* result block number */
608 	xfs_suminfo_t	sum;		/* summary information for extents */
609 
610 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
611 	ASSERT(maxlen != 0);
612 
613 	/*
614 	 * Loop over all the levels starting with maxlen.
615 	 * At each level, look at all the bitmap blocks, to see if there
616 	 * are extents starting there that are long enough (>= maxlen).
617 	 * Note, only on the initial level can the allocation fail if
618 	 * the summary says there's an extent.
619 	 */
620 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
621 		/*
622 		 * Loop over all the bitmap blocks.
623 		 */
624 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
625 			/*
626 			 * Get the summary for this level/block.
627 			 */
628 			error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
629 				&sum);
630 			if (error) {
631 				return error;
632 			}
633 			/*
634 			 * Nothing there, on to the next block.
635 			 */
636 			if (!sum)
637 				continue;
638 			/*
639 			 * Try allocating the extent.
640 			 */
641 			error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
642 				maxlen, len, &n, rbpp, rsb, prod, &r);
643 			if (error) {
644 				return error;
645 			}
646 			/*
647 			 * If it worked, return that.
648 			 */
649 			if (r != NULLRTBLOCK) {
650 				*rtblock = r;
651 				return 0;
652 			}
653 			/*
654 			 * If the "next block to try" returned from the
655 			 * allocator is beyond the next bitmap block,
656 			 * skip to that bitmap block.
657 			 */
658 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
659 				i = XFS_BITTOBLOCK(mp, n) - 1;
660 		}
661 	}
662 	/*
663 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
664 	 * we're asking for a fixed size extent.
665 	 */
666 	if (minlen > --maxlen) {
667 		*rtblock = NULLRTBLOCK;
668 		return 0;
669 	}
670 	ASSERT(minlen != 0);
671 	ASSERT(maxlen != 0);
672 
673 	/*
674 	 * Loop over sizes, from maxlen down to minlen.
675 	 * This time, when we do the allocations, allow smaller ones
676 	 * to succeed.
677 	 */
678 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
679 		/*
680 		 * Loop over all the bitmap blocks, try an allocation
681 		 * starting in that block.
682 		 */
683 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
684 			/*
685 			 * Get the summary information for this level/block.
686 			 */
687 			error =	xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
688 						  &sum);
689 			if (error) {
690 				return error;
691 			}
692 			/*
693 			 * If nothing there, go on to next.
694 			 */
695 			if (!sum)
696 				continue;
697 			/*
698 			 * Try the allocation.  Make sure the specified
699 			 * minlen/maxlen are in the possible range for
700 			 * this summary level.
701 			 */
702 			error = xfs_rtallocate_extent_block(mp, tp, i,
703 					XFS_RTMAX(minlen, 1 << l),
704 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
705 					len, &n, rbpp, rsb, prod, &r);
706 			if (error) {
707 				return error;
708 			}
709 			/*
710 			 * If it worked, return that extent.
711 			 */
712 			if (r != NULLRTBLOCK) {
713 				*rtblock = r;
714 				return 0;
715 			}
716 			/*
717 			 * If the "next block to try" returned from the
718 			 * allocator is beyond the next bitmap block,
719 			 * skip to that bitmap block.
720 			 */
721 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
722 				i = XFS_BITTOBLOCK(mp, n) - 1;
723 		}
724 	}
725 	/*
726 	 * Got nothing, return failure.
727 	 */
728 	*rtblock = NULLRTBLOCK;
729 	return 0;
730 }
731 
732 /*
733  * Mark an extent specified by start and len allocated.
734  * Updates all the summary information as well as the bitmap.
735  */
736 STATIC int				/* error */
737 xfs_rtallocate_range(
738 	xfs_mount_t	*mp,		/* file system mount point */
739 	xfs_trans_t	*tp,		/* transaction pointer */
740 	xfs_rtblock_t	start,		/* start block to allocate */
741 	xfs_extlen_t	len,		/* length to allocate */
742 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
743 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
744 {
745 	xfs_rtblock_t	end;		/* end of the allocated extent */
746 	int		error;		/* error value */
747 	xfs_rtblock_t	postblock;	/* first block allocated > end */
748 	xfs_rtblock_t	preblock;	/* first block allocated < start */
749 
750 	end = start + len - 1;
751 	/*
752 	 * Assume we're allocating out of the middle of a free extent.
753 	 * We need to find the beginning and end of the extent so we can
754 	 * properly update the summary.
755 	 */
756 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
757 	if (error) {
758 		return error;
759 	}
760 	/*
761 	 * Find the next allocated block (end of free extent).
762 	 */
763 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
764 		&postblock);
765 	if (error) {
766 		return error;
767 	}
768 	/*
769 	 * Decrement the summary information corresponding to the entire
770 	 * (old) free extent.
771 	 */
772 	error = xfs_rtmodify_summary(mp, tp,
773 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
774 		XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
775 	if (error) {
776 		return error;
777 	}
778 	/*
779 	 * If there are blocks not being allocated at the front of the
780 	 * old extent, add summary data for them to be free.
781 	 */
782 	if (preblock < start) {
783 		error = xfs_rtmodify_summary(mp, tp,
784 			XFS_RTBLOCKLOG(start - preblock),
785 			XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
786 		if (error) {
787 			return error;
788 		}
789 	}
790 	/*
791 	 * If there are blocks not being allocated at the end of the
792 	 * old extent, add summary data for them to be free.
793 	 */
794 	if (postblock > end) {
795 		error = xfs_rtmodify_summary(mp, tp,
796 			XFS_RTBLOCKLOG(postblock - end),
797 			XFS_BITTOBLOCK(mp, end + 1), 1, rbpp, rsb);
798 		if (error) {
799 			return error;
800 		}
801 	}
802 	/*
803 	 * Modify the bitmap to mark this extent allocated.
804 	 */
805 	error = xfs_rtmodify_range(mp, tp, start, len, 0);
806 	return error;
807 }
808 
809 /*
810  * Return whether there are any free extents in the size range given
811  * by low and high, for the bitmap block bbno.
812  */
813 STATIC int				/* error */
814 xfs_rtany_summary(
815 	xfs_mount_t	*mp,		/* file system mount structure */
816 	xfs_trans_t	*tp,		/* transaction pointer */
817 	int		low,		/* low log2 extent size */
818 	int		high,		/* high log2 extent size */
819 	xfs_rtblock_t	bbno,		/* bitmap block number */
820 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
821 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
822 	int		*stat)		/* out: any good extents here? */
823 {
824 	int		error;		/* error value */
825 	int		log;		/* loop counter, log2 of ext. size */
826 	xfs_suminfo_t	sum;		/* summary data */
827 
828 	/*
829 	 * Loop over logs of extent sizes.  Order is irrelevant.
830 	 */
831 	for (log = low; log <= high; log++) {
832 		/*
833 		 * Get one summary datum.
834 		 */
835 		error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
836 		if (error) {
837 			return error;
838 		}
839 		/*
840 		 * If there are any, return success.
841 		 */
842 		if (sum) {
843 			*stat = 1;
844 			return 0;
845 		}
846 	}
847 	/*
848 	 * Found nothing, return failure.
849 	 */
850 	*stat = 0;
851 	return 0;
852 }
853 
854 /*
855  * Get a buffer for the bitmap or summary file block specified.
856  * The buffer is returned read and locked.
857  */
858 STATIC int				/* error */
859 xfs_rtbuf_get(
860 	xfs_mount_t	*mp,		/* file system mount structure */
861 	xfs_trans_t	*tp,		/* transaction pointer */
862 	xfs_rtblock_t	block,		/* block number in bitmap or summary */
863 	int		issum,		/* is summary not bitmap */
864 	xfs_buf_t	**bpp)		/* output: buffer for the block */
865 {
866 	xfs_buf_t	*bp;		/* block buffer, result */
867 	xfs_daddr_t	d;		/* disk addr of block */
868 	int		error;		/* error value */
869 	xfs_fsblock_t	fsb;		/* fs block number for block */
870 	xfs_inode_t	*ip;		/* bitmap or summary inode */
871 
872 	ip = issum ? mp->m_rsumip : mp->m_rbmip;
873 	/*
874 	 * Map from the file offset (block) and inode number to the
875 	 * file system block.
876 	 */
877 	error = xfs_bmapi_single(tp, ip, XFS_DATA_FORK, &fsb, block);
878 	if (error) {
879 		return error;
880 	}
881 	ASSERT(fsb != NULLFSBLOCK);
882 	/*
883 	 * Convert to disk address for buffer cache.
884 	 */
885 	d = XFS_FSB_TO_DADDR(mp, fsb);
886 	/*
887 	 * Read the buffer.
888 	 */
889 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
890 				   mp->m_bsize, 0, &bp);
891 	if (error) {
892 		return error;
893 	}
894 	ASSERT(bp && !XFS_BUF_GETERROR(bp));
895 	*bpp = bp;
896 	return 0;
897 }
898 
899 #ifdef DEBUG
900 /*
901  * Check that the given extent (block range) is allocated already.
902  */
903 STATIC int				/* error */
904 xfs_rtcheck_alloc_range(
905 	xfs_mount_t	*mp,		/* file system mount point */
906 	xfs_trans_t	*tp,		/* transaction pointer */
907 	xfs_rtblock_t	bno,		/* starting block number of extent */
908 	xfs_extlen_t	len,		/* length of extent */
909 	int		*stat)		/* out: 1 for allocated, 0 for not */
910 {
911 	xfs_rtblock_t	new;		/* dummy for xfs_rtcheck_range */
912 
913 	return xfs_rtcheck_range(mp, tp, bno, len, 0, &new, stat);
914 }
915 #endif
916 
917 /*
918  * Check that the given range is either all allocated (val = 0) or
919  * all free (val = 1).
920  */
921 STATIC int				/* error */
922 xfs_rtcheck_range(
923 	xfs_mount_t	*mp,		/* file system mount point */
924 	xfs_trans_t	*tp,		/* transaction pointer */
925 	xfs_rtblock_t	start,		/* starting block number of extent */
926 	xfs_extlen_t	len,		/* length of extent */
927 	int		val,		/* 1 for free, 0 for allocated */
928 	xfs_rtblock_t	*new,		/* out: first block not matching */
929 	int		*stat)		/* out: 1 for matches, 0 for not */
930 {
931 	xfs_rtword_t	*b;		/* current word in buffer */
932 	int		bit;		/* bit number in the word */
933 	xfs_rtblock_t	block;		/* bitmap block number */
934 	xfs_buf_t	*bp;		/* buf for the block */
935 	xfs_rtword_t	*bufp;		/* starting word in buffer */
936 	int		error;		/* error value */
937 	xfs_rtblock_t	i;		/* current bit number rel. to start */
938 	xfs_rtblock_t	lastbit;	/* last useful bit in word */
939 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
940 	xfs_rtword_t	wdiff;		/* difference from wanted value */
941 	int		word;		/* word number in the buffer */
942 
943 	/*
944 	 * Compute starting bitmap block number
945 	 */
946 	block = XFS_BITTOBLOCK(mp, start);
947 	/*
948 	 * Read the bitmap block.
949 	 */
950 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
951 	if (error) {
952 		return error;
953 	}
954 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
955 	/*
956 	 * Compute the starting word's address, and starting bit.
957 	 */
958 	word = XFS_BITTOWORD(mp, start);
959 	b = &bufp[word];
960 	bit = (int)(start & (XFS_NBWORD - 1));
961 	/*
962 	 * 0 (allocated) => all zero's; 1 (free) => all one's.
963 	 */
964 	val = -val;
965 	/*
966 	 * If not starting on a word boundary, deal with the first
967 	 * (partial) word.
968 	 */
969 	if (bit) {
970 		/*
971 		 * Compute first bit not examined.
972 		 */
973 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
974 		/*
975 		 * Mask of relevant bits.
976 		 */
977 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
978 		/*
979 		 * Compute difference between actual and desired value.
980 		 */
981 		if ((wdiff = (*b ^ val) & mask)) {
982 			/*
983 			 * Different, compute first wrong bit and return.
984 			 */
985 			xfs_trans_brelse(tp, bp);
986 			i = XFS_RTLOBIT(wdiff) - bit;
987 			*new = start + i;
988 			*stat = 0;
989 			return 0;
990 		}
991 		i = lastbit - bit;
992 		/*
993 		 * Go on to next block if that's where the next word is
994 		 * and we need the next word.
995 		 */
996 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
997 			/*
998 			 * If done with this block, get the next one.
999 			 */
1000 			xfs_trans_brelse(tp, bp);
1001 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1002 			if (error) {
1003 				return error;
1004 			}
1005 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1006 			word = 0;
1007 		} else {
1008 			/*
1009 			 * Go on to the next word in the buffer.
1010 			 */
1011 			b++;
1012 		}
1013 	} else {
1014 		/*
1015 		 * Starting on a word boundary, no partial word.
1016 		 */
1017 		i = 0;
1018 	}
1019 	/*
1020 	 * Loop over whole words in buffers.  When we use up one buffer
1021 	 * we move on to the next one.
1022 	 */
1023 	while (len - i >= XFS_NBWORD) {
1024 		/*
1025 		 * Compute difference between actual and desired value.
1026 		 */
1027 		if ((wdiff = *b ^ val)) {
1028 			/*
1029 			 * Different, compute first wrong bit and return.
1030 			 */
1031 			xfs_trans_brelse(tp, bp);
1032 			i += XFS_RTLOBIT(wdiff);
1033 			*new = start + i;
1034 			*stat = 0;
1035 			return 0;
1036 		}
1037 		i += XFS_NBWORD;
1038 		/*
1039 		 * Go on to next block if that's where the next word is
1040 		 * and we need the next word.
1041 		 */
1042 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1043 			/*
1044 			 * If done with this block, get the next one.
1045 			 */
1046 			xfs_trans_brelse(tp, bp);
1047 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1048 			if (error) {
1049 				return error;
1050 			}
1051 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1052 			word = 0;
1053 		} else {
1054 			/*
1055 			 * Go on to the next word in the buffer.
1056 			 */
1057 			b++;
1058 		}
1059 	}
1060 	/*
1061 	 * If not ending on a word boundary, deal with the last
1062 	 * (partial) word.
1063 	 */
1064 	if ((lastbit = len - i)) {
1065 		/*
1066 		 * Mask of relevant bits.
1067 		 */
1068 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1069 		/*
1070 		 * Compute difference between actual and desired value.
1071 		 */
1072 		if ((wdiff = (*b ^ val) & mask)) {
1073 			/*
1074 			 * Different, compute first wrong bit and return.
1075 			 */
1076 			xfs_trans_brelse(tp, bp);
1077 			i += XFS_RTLOBIT(wdiff);
1078 			*new = start + i;
1079 			*stat = 0;
1080 			return 0;
1081 		} else
1082 			i = len;
1083 	}
1084 	/*
1085 	 * Successful, return.
1086 	 */
1087 	xfs_trans_brelse(tp, bp);
1088 	*new = start + i;
1089 	*stat = 1;
1090 	return 0;
1091 }
1092 
1093 /*
1094  * Copy and transform the summary file, given the old and new
1095  * parameters in the mount structures.
1096  */
1097 STATIC int				/* error */
1098 xfs_rtcopy_summary(
1099 	xfs_mount_t	*omp,		/* old file system mount point */
1100 	xfs_mount_t	*nmp,		/* new file system mount point */
1101 	xfs_trans_t	*tp)		/* transaction pointer */
1102 {
1103 	xfs_rtblock_t	bbno;		/* bitmap block number */
1104 	xfs_buf_t	*bp;		/* summary buffer */
1105 	int		error;		/* error return value */
1106 	int		log;		/* summary level number (log length) */
1107 	xfs_suminfo_t	sum;		/* summary data */
1108 	xfs_fsblock_t	sumbno;		/* summary block number */
1109 
1110 	bp = NULL;
1111 	for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
1112 		for (bbno = omp->m_sb.sb_rbmblocks - 1;
1113 		     (xfs_srtblock_t)bbno >= 0;
1114 		     bbno--) {
1115 			error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
1116 				&sumbno, &sum);
1117 			if (error)
1118 				return error;
1119 			if (sum == 0)
1120 				continue;
1121 			error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
1122 				&bp, &sumbno);
1123 			if (error)
1124 				return error;
1125 			error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
1126 				&bp, &sumbno);
1127 			if (error)
1128 				return error;
1129 			ASSERT(sum > 0);
1130 		}
1131 	}
1132 	return 0;
1133 }
1134 
1135 /*
1136  * Searching backward from start to limit, find the first block whose
1137  * allocated/free state is different from start's.
1138  */
1139 STATIC int				/* error */
1140 xfs_rtfind_back(
1141 	xfs_mount_t	*mp,		/* file system mount point */
1142 	xfs_trans_t	*tp,		/* transaction pointer */
1143 	xfs_rtblock_t	start,		/* starting block to look at */
1144 	xfs_rtblock_t	limit,		/* last block to look at */
1145 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1146 {
1147 	xfs_rtword_t	*b;		/* current word in buffer */
1148 	int		bit;		/* bit number in the word */
1149 	xfs_rtblock_t	block;		/* bitmap block number */
1150 	xfs_buf_t	*bp;		/* buf for the block */
1151 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1152 	int		error;		/* error value */
1153 	xfs_rtblock_t	firstbit;	/* first useful bit in the word */
1154 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1155 	xfs_rtblock_t	len;		/* length of inspected area */
1156 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1157 	xfs_rtword_t	want;		/* mask for "good" values */
1158 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1159 	int		word;		/* word number in the buffer */
1160 
1161 	/*
1162 	 * Compute and read in starting bitmap block for starting block.
1163 	 */
1164 	block = XFS_BITTOBLOCK(mp, start);
1165 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1166 	if (error) {
1167 		return error;
1168 	}
1169 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1170 	/*
1171 	 * Get the first word's index & point to it.
1172 	 */
1173 	word = XFS_BITTOWORD(mp, start);
1174 	b = &bufp[word];
1175 	bit = (int)(start & (XFS_NBWORD - 1));
1176 	len = start - limit + 1;
1177 	/*
1178 	 * Compute match value, based on the bit at start: if 1 (free)
1179 	 * then all-ones, else all-zeroes.
1180 	 */
1181 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1182 	/*
1183 	 * If the starting position is not word-aligned, deal with the
1184 	 * partial word.
1185 	 */
1186 	if (bit < XFS_NBWORD - 1) {
1187 		/*
1188 		 * Calculate first (leftmost) bit number to look at,
1189 		 * and mask for all the relevant bits in this word.
1190 		 */
1191 		firstbit = XFS_RTMAX((xfs_srtblock_t)(bit - len + 1), 0);
1192 		mask = (((xfs_rtword_t)1 << (bit - firstbit + 1)) - 1) <<
1193 			firstbit;
1194 		/*
1195 		 * Calculate the difference between the value there
1196 		 * and what we're looking for.
1197 		 */
1198 		if ((wdiff = (*b ^ want) & mask)) {
1199 			/*
1200 			 * Different.  Mark where we are and return.
1201 			 */
1202 			xfs_trans_brelse(tp, bp);
1203 			i = bit - XFS_RTHIBIT(wdiff);
1204 			*rtblock = start - i + 1;
1205 			return 0;
1206 		}
1207 		i = bit - firstbit + 1;
1208 		/*
1209 		 * Go on to previous block if that's where the previous word is
1210 		 * and we need the previous word.
1211 		 */
1212 		if (--word == -1 && i < len) {
1213 			/*
1214 			 * If done with this block, get the previous one.
1215 			 */
1216 			xfs_trans_brelse(tp, bp);
1217 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1218 			if (error) {
1219 				return error;
1220 			}
1221 			bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1222 			word = XFS_BLOCKWMASK(mp);
1223 			b = &bufp[word];
1224 		} else {
1225 			/*
1226 			 * Go on to the previous word in the buffer.
1227 			 */
1228 			b--;
1229 		}
1230 	} else {
1231 		/*
1232 		 * Starting on a word boundary, no partial word.
1233 		 */
1234 		i = 0;
1235 	}
1236 	/*
1237 	 * Loop over whole words in buffers.  When we use up one buffer
1238 	 * we move on to the previous one.
1239 	 */
1240 	while (len - i >= XFS_NBWORD) {
1241 		/*
1242 		 * Compute difference between actual and desired value.
1243 		 */
1244 		if ((wdiff = *b ^ want)) {
1245 			/*
1246 			 * Different, mark where we are and return.
1247 			 */
1248 			xfs_trans_brelse(tp, bp);
1249 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1250 			*rtblock = start - i + 1;
1251 			return 0;
1252 		}
1253 		i += XFS_NBWORD;
1254 		/*
1255 		 * Go on to previous block if that's where the previous word is
1256 		 * and we need the previous word.
1257 		 */
1258 		if (--word == -1 && i < len) {
1259 			/*
1260 			 * If done with this block, get the previous one.
1261 			 */
1262 			xfs_trans_brelse(tp, bp);
1263 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1264 			if (error) {
1265 				return error;
1266 			}
1267 			bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1268 			word = XFS_BLOCKWMASK(mp);
1269 			b = &bufp[word];
1270 		} else {
1271 			/*
1272 			 * Go on to the previous word in the buffer.
1273 			 */
1274 			b--;
1275 		}
1276 	}
1277 	/*
1278 	 * If not ending on a word boundary, deal with the last
1279 	 * (partial) word.
1280 	 */
1281 	if (len - i) {
1282 		/*
1283 		 * Calculate first (leftmost) bit number to look at,
1284 		 * and mask for all the relevant bits in this word.
1285 		 */
1286 		firstbit = XFS_NBWORD - (len - i);
1287 		mask = (((xfs_rtword_t)1 << (len - i)) - 1) << firstbit;
1288 		/*
1289 		 * Compute difference between actual and desired value.
1290 		 */
1291 		if ((wdiff = (*b ^ want) & mask)) {
1292 			/*
1293 			 * Different, mark where we are and return.
1294 			 */
1295 			xfs_trans_brelse(tp, bp);
1296 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1297 			*rtblock = start - i + 1;
1298 			return 0;
1299 		} else
1300 			i = len;
1301 	}
1302 	/*
1303 	 * No match, return that we scanned the whole area.
1304 	 */
1305 	xfs_trans_brelse(tp, bp);
1306 	*rtblock = start - i + 1;
1307 	return 0;
1308 }
1309 
1310 /*
1311  * Searching forward from start to limit, find the first block whose
1312  * allocated/free state is different from start's.
1313  */
1314 STATIC int				/* error */
1315 xfs_rtfind_forw(
1316 	xfs_mount_t	*mp,		/* file system mount point */
1317 	xfs_trans_t	*tp,		/* transaction pointer */
1318 	xfs_rtblock_t	start,		/* starting block to look at */
1319 	xfs_rtblock_t	limit,		/* last block to look at */
1320 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1321 {
1322 	xfs_rtword_t	*b;		/* current word in buffer */
1323 	int		bit;		/* bit number in the word */
1324 	xfs_rtblock_t	block;		/* bitmap block number */
1325 	xfs_buf_t	*bp;		/* buf for the block */
1326 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1327 	int		error;		/* error value */
1328 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1329 	xfs_rtblock_t	lastbit;	/* last useful bit in the word */
1330 	xfs_rtblock_t	len;		/* length of inspected area */
1331 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1332 	xfs_rtword_t	want;		/* mask for "good" values */
1333 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1334 	int		word;		/* word number in the buffer */
1335 
1336 	/*
1337 	 * Compute and read in starting bitmap block for starting block.
1338 	 */
1339 	block = XFS_BITTOBLOCK(mp, start);
1340 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1341 	if (error) {
1342 		return error;
1343 	}
1344 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1345 	/*
1346 	 * Get the first word's index & point to it.
1347 	 */
1348 	word = XFS_BITTOWORD(mp, start);
1349 	b = &bufp[word];
1350 	bit = (int)(start & (XFS_NBWORD - 1));
1351 	len = limit - start + 1;
1352 	/*
1353 	 * Compute match value, based on the bit at start: if 1 (free)
1354 	 * then all-ones, else all-zeroes.
1355 	 */
1356 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1357 	/*
1358 	 * If the starting position is not word-aligned, deal with the
1359 	 * partial word.
1360 	 */
1361 	if (bit) {
1362 		/*
1363 		 * Calculate last (rightmost) bit number to look at,
1364 		 * and mask for all the relevant bits in this word.
1365 		 */
1366 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1367 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1368 		/*
1369 		 * Calculate the difference between the value there
1370 		 * and what we're looking for.
1371 		 */
1372 		if ((wdiff = (*b ^ want) & mask)) {
1373 			/*
1374 			 * Different.  Mark where we are and return.
1375 			 */
1376 			xfs_trans_brelse(tp, bp);
1377 			i = XFS_RTLOBIT(wdiff) - bit;
1378 			*rtblock = start + i - 1;
1379 			return 0;
1380 		}
1381 		i = lastbit - bit;
1382 		/*
1383 		 * Go on to next block if that's where the next word is
1384 		 * and we need the next word.
1385 		 */
1386 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1387 			/*
1388 			 * If done with this block, get the previous one.
1389 			 */
1390 			xfs_trans_brelse(tp, bp);
1391 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1392 			if (error) {
1393 				return error;
1394 			}
1395 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1396 			word = 0;
1397 		} else {
1398 			/*
1399 			 * Go on to the previous word in the buffer.
1400 			 */
1401 			b++;
1402 		}
1403 	} else {
1404 		/*
1405 		 * Starting on a word boundary, no partial word.
1406 		 */
1407 		i = 0;
1408 	}
1409 	/*
1410 	 * Loop over whole words in buffers.  When we use up one buffer
1411 	 * we move on to the next one.
1412 	 */
1413 	while (len - i >= XFS_NBWORD) {
1414 		/*
1415 		 * Compute difference between actual and desired value.
1416 		 */
1417 		if ((wdiff = *b ^ want)) {
1418 			/*
1419 			 * Different, mark where we are and return.
1420 			 */
1421 			xfs_trans_brelse(tp, bp);
1422 			i += XFS_RTLOBIT(wdiff);
1423 			*rtblock = start + i - 1;
1424 			return 0;
1425 		}
1426 		i += XFS_NBWORD;
1427 		/*
1428 		 * Go on to next block if that's where the next word is
1429 		 * and we need the next word.
1430 		 */
1431 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1432 			/*
1433 			 * If done with this block, get the next one.
1434 			 */
1435 			xfs_trans_brelse(tp, bp);
1436 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1437 			if (error) {
1438 				return error;
1439 			}
1440 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1441 			word = 0;
1442 		} else {
1443 			/*
1444 			 * Go on to the next word in the buffer.
1445 			 */
1446 			b++;
1447 		}
1448 	}
1449 	/*
1450 	 * If not ending on a word boundary, deal with the last
1451 	 * (partial) word.
1452 	 */
1453 	if ((lastbit = len - i)) {
1454 		/*
1455 		 * Calculate mask for all the relevant bits in this word.
1456 		 */
1457 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1458 		/*
1459 		 * Compute difference between actual and desired value.
1460 		 */
1461 		if ((wdiff = (*b ^ want) & mask)) {
1462 			/*
1463 			 * Different, mark where we are and return.
1464 			 */
1465 			xfs_trans_brelse(tp, bp);
1466 			i += XFS_RTLOBIT(wdiff);
1467 			*rtblock = start + i - 1;
1468 			return 0;
1469 		} else
1470 			i = len;
1471 	}
1472 	/*
1473 	 * No match, return that we scanned the whole area.
1474 	 */
1475 	xfs_trans_brelse(tp, bp);
1476 	*rtblock = start + i - 1;
1477 	return 0;
1478 }
1479 
1480 /*
1481  * Mark an extent specified by start and len freed.
1482  * Updates all the summary information as well as the bitmap.
1483  */
1484 STATIC int				/* error */
1485 xfs_rtfree_range(
1486 	xfs_mount_t	*mp,		/* file system mount point */
1487 	xfs_trans_t	*tp,		/* transaction pointer */
1488 	xfs_rtblock_t	start,		/* starting block to free */
1489 	xfs_extlen_t	len,		/* length to free */
1490 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1491 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1492 {
1493 	xfs_rtblock_t	end;		/* end of the freed extent */
1494 	int		error;		/* error value */
1495 	xfs_rtblock_t	postblock;	/* first block freed > end */
1496 	xfs_rtblock_t	preblock;	/* first block freed < start */
1497 
1498 	end = start + len - 1;
1499 	/*
1500 	 * Modify the bitmap to mark this extent freed.
1501 	 */
1502 	error = xfs_rtmodify_range(mp, tp, start, len, 1);
1503 	if (error) {
1504 		return error;
1505 	}
1506 	/*
1507 	 * Assume we're freeing out of the middle of an allocated extent.
1508 	 * We need to find the beginning and end of the extent so we can
1509 	 * properly update the summary.
1510 	 */
1511 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
1512 	if (error) {
1513 		return error;
1514 	}
1515 	/*
1516 	 * Find the next allocated block (end of allocated extent).
1517 	 */
1518 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
1519 		&postblock);
1520 	if (error)
1521 		return error;
1522 	/*
1523 	 * If there are blocks not being freed at the front of the
1524 	 * old extent, add summary data for them to be allocated.
1525 	 */
1526 	if (preblock < start) {
1527 		error = xfs_rtmodify_summary(mp, tp,
1528 			XFS_RTBLOCKLOG(start - preblock),
1529 			XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
1530 		if (error) {
1531 			return error;
1532 		}
1533 	}
1534 	/*
1535 	 * If there are blocks not being freed at the end of the
1536 	 * old extent, add summary data for them to be allocated.
1537 	 */
1538 	if (postblock > end) {
1539 		error = xfs_rtmodify_summary(mp, tp,
1540 			XFS_RTBLOCKLOG(postblock - end),
1541 			XFS_BITTOBLOCK(mp, end + 1), -1, rbpp, rsb);
1542 		if (error) {
1543 			return error;
1544 		}
1545 	}
1546 	/*
1547 	 * Increment the summary information corresponding to the entire
1548 	 * (new) free extent.
1549 	 */
1550 	error = xfs_rtmodify_summary(mp, tp,
1551 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
1552 		XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
1553 	return error;
1554 }
1555 
1556 /*
1557  * Read and return the summary information for a given extent size,
1558  * bitmap block combination.
1559  * Keeps track of a current summary block, so we don't keep reading
1560  * it from the buffer cache.
1561  */
1562 STATIC int				/* error */
1563 xfs_rtget_summary(
1564 	xfs_mount_t	*mp,		/* file system mount structure */
1565 	xfs_trans_t	*tp,		/* transaction pointer */
1566 	int		log,		/* log2 of extent size */
1567 	xfs_rtblock_t	bbno,		/* bitmap block number */
1568 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1569 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
1570 	xfs_suminfo_t	*sum)		/* out: summary info for this block */
1571 {
1572 	xfs_buf_t	*bp;		/* buffer for summary block */
1573 	int		error;		/* error value */
1574 	xfs_fsblock_t	sb;		/* summary fsblock */
1575 	int		so;		/* index into the summary file */
1576 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1577 
1578 	/*
1579 	 * Compute entry number in the summary file.
1580 	 */
1581 	so = XFS_SUMOFFS(mp, log, bbno);
1582 	/*
1583 	 * Compute the block number in the summary file.
1584 	 */
1585 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1586 	/*
1587 	 * If we have an old buffer, and the block number matches, use that.
1588 	 */
1589 	if (rbpp && *rbpp && *rsb == sb)
1590 		bp = *rbpp;
1591 	/*
1592 	 * Otherwise we have to get the buffer.
1593 	 */
1594 	else {
1595 		/*
1596 		 * If there was an old one, get rid of it first.
1597 		 */
1598 		if (rbpp && *rbpp)
1599 			xfs_trans_brelse(tp, *rbpp);
1600 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1601 		if (error) {
1602 			return error;
1603 		}
1604 		/*
1605 		 * Remember this buffer and block for the next call.
1606 		 */
1607 		if (rbpp) {
1608 			*rbpp = bp;
1609 			*rsb = sb;
1610 		}
1611 	}
1612 	/*
1613 	 * Point to the summary information & copy it out.
1614 	 */
1615 	sp = XFS_SUMPTR(mp, bp, so);
1616 	*sum = *sp;
1617 	/*
1618 	 * Drop the buffer if we're not asked to remember it.
1619 	 */
1620 	if (!rbpp)
1621 		xfs_trans_brelse(tp, bp);
1622 	return 0;
1623 }
1624 
1625 /*
1626  * Set the given range of bitmap bits to the given value.
1627  * Do whatever I/O and logging is required.
1628  */
1629 STATIC int				/* error */
1630 xfs_rtmodify_range(
1631 	xfs_mount_t	*mp,		/* file system mount point */
1632 	xfs_trans_t	*tp,		/* transaction pointer */
1633 	xfs_rtblock_t	start,		/* starting block to modify */
1634 	xfs_extlen_t	len,		/* length of extent to modify */
1635 	int		val)		/* 1 for free, 0 for allocated */
1636 {
1637 	xfs_rtword_t	*b;		/* current word in buffer */
1638 	int		bit;		/* bit number in the word */
1639 	xfs_rtblock_t	block;		/* bitmap block number */
1640 	xfs_buf_t	*bp;		/* buf for the block */
1641 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1642 	int		error;		/* error value */
1643 	xfs_rtword_t	*first;		/* first used word in the buffer */
1644 	int		i;		/* current bit number rel. to start */
1645 	int		lastbit;	/* last useful bit in word */
1646 	xfs_rtword_t	mask;		/* mask o frelevant bits for value */
1647 	int		word;		/* word number in the buffer */
1648 
1649 	/*
1650 	 * Compute starting bitmap block number.
1651 	 */
1652 	block = XFS_BITTOBLOCK(mp, start);
1653 	/*
1654 	 * Read the bitmap block, and point to its data.
1655 	 */
1656 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1657 	if (error) {
1658 		return error;
1659 	}
1660 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1661 	/*
1662 	 * Compute the starting word's address, and starting bit.
1663 	 */
1664 	word = XFS_BITTOWORD(mp, start);
1665 	first = b = &bufp[word];
1666 	bit = (int)(start & (XFS_NBWORD - 1));
1667 	/*
1668 	 * 0 (allocated) => all zeroes; 1 (free) => all ones.
1669 	 */
1670 	val = -val;
1671 	/*
1672 	 * If not starting on a word boundary, deal with the first
1673 	 * (partial) word.
1674 	 */
1675 	if (bit) {
1676 		/*
1677 		 * Compute first bit not changed and mask of relevant bits.
1678 		 */
1679 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1680 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1681 		/*
1682 		 * Set/clear the active bits.
1683 		 */
1684 		if (val)
1685 			*b |= mask;
1686 		else
1687 			*b &= ~mask;
1688 		i = lastbit - bit;
1689 		/*
1690 		 * Go on to the next block if that's where the next word is
1691 		 * and we need the next word.
1692 		 */
1693 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1694 			/*
1695 			 * Log the changed part of this block.
1696 			 * Get the next one.
1697 			 */
1698 			xfs_trans_log_buf(tp, bp,
1699 				(uint)((char *)first - (char *)bufp),
1700 				(uint)((char *)b - (char *)bufp));
1701 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1702 			if (error) {
1703 				return error;
1704 			}
1705 			first = b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1706 			word = 0;
1707 		} else {
1708 			/*
1709 			 * Go on to the next word in the buffer
1710 			 */
1711 			b++;
1712 		}
1713 	} else {
1714 		/*
1715 		 * Starting on a word boundary, no partial word.
1716 		 */
1717 		i = 0;
1718 	}
1719 	/*
1720 	 * Loop over whole words in buffers.  When we use up one buffer
1721 	 * we move on to the next one.
1722 	 */
1723 	while (len - i >= XFS_NBWORD) {
1724 		/*
1725 		 * Set the word value correctly.
1726 		 */
1727 		*b = val;
1728 		i += XFS_NBWORD;
1729 		/*
1730 		 * Go on to the next block if that's where the next word is
1731 		 * and we need the next word.
1732 		 */
1733 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1734 			/*
1735 			 * Log the changed part of this block.
1736 			 * Get the next one.
1737 			 */
1738 			xfs_trans_log_buf(tp, bp,
1739 				(uint)((char *)first - (char *)bufp),
1740 				(uint)((char *)b - (char *)bufp));
1741 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1742 			if (error) {
1743 				return error;
1744 			}
1745 			first = b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1746 			word = 0;
1747 		} else {
1748 			/*
1749 			 * Go on to the next word in the buffer
1750 			 */
1751 			b++;
1752 		}
1753 	}
1754 	/*
1755 	 * If not ending on a word boundary, deal with the last
1756 	 * (partial) word.
1757 	 */
1758 	if ((lastbit = len - i)) {
1759 		/*
1760 		 * Compute a mask of relevant bits.
1761 		 */
1762 		bit = 0;
1763 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1764 		/*
1765 		 * Set/clear the active bits.
1766 		 */
1767 		if (val)
1768 			*b |= mask;
1769 		else
1770 			*b &= ~mask;
1771 		b++;
1772 	}
1773 	/*
1774 	 * Log any remaining changed bytes.
1775 	 */
1776 	if (b > first)
1777 		xfs_trans_log_buf(tp, bp, (uint)((char *)first - (char *)bufp),
1778 			(uint)((char *)b - (char *)bufp - 1));
1779 	return 0;
1780 }
1781 
1782 /*
1783  * Read and modify the summary information for a given extent size,
1784  * bitmap block combination.
1785  * Keeps track of a current summary block, so we don't keep reading
1786  * it from the buffer cache.
1787  */
1788 STATIC int				/* error */
1789 xfs_rtmodify_summary(
1790 	xfs_mount_t	*mp,		/* file system mount point */
1791 	xfs_trans_t	*tp,		/* transaction pointer */
1792 	int		log,		/* log2 of extent size */
1793 	xfs_rtblock_t	bbno,		/* bitmap block number */
1794 	int		delta,		/* change to make to summary info */
1795 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1796 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1797 {
1798 	xfs_buf_t	*bp;		/* buffer for the summary block */
1799 	int		error;		/* error value */
1800 	xfs_fsblock_t	sb;		/* summary fsblock */
1801 	int		so;		/* index into the summary file */
1802 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1803 
1804 	/*
1805 	 * Compute entry number in the summary file.
1806 	 */
1807 	so = XFS_SUMOFFS(mp, log, bbno);
1808 	/*
1809 	 * Compute the block number in the summary file.
1810 	 */
1811 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1812 	/*
1813 	 * If we have an old buffer, and the block number matches, use that.
1814 	 */
1815 	if (rbpp && *rbpp && *rsb == sb)
1816 		bp = *rbpp;
1817 	/*
1818 	 * Otherwise we have to get the buffer.
1819 	 */
1820 	else {
1821 		/*
1822 		 * If there was an old one, get rid of it first.
1823 		 */
1824 		if (rbpp && *rbpp)
1825 			xfs_trans_brelse(tp, *rbpp);
1826 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1827 		if (error) {
1828 			return error;
1829 		}
1830 		/*
1831 		 * Remember this buffer and block for the next call.
1832 		 */
1833 		if (rbpp) {
1834 			*rbpp = bp;
1835 			*rsb = sb;
1836 		}
1837 	}
1838 	/*
1839 	 * Point to the summary information, modify and log it.
1840 	 */
1841 	sp = XFS_SUMPTR(mp, bp, so);
1842 	*sp += delta;
1843 	xfs_trans_log_buf(tp, bp, (uint)((char *)sp - (char *)XFS_BUF_PTR(bp)),
1844 		(uint)((char *)sp - (char *)XFS_BUF_PTR(bp) + sizeof(*sp) - 1));
1845 	return 0;
1846 }
1847 
1848 /*
1849  * Visible (exported) functions.
1850  */
1851 
1852 /*
1853  * Grow the realtime area of the filesystem.
1854  */
1855 int
1856 xfs_growfs_rt(
1857 	xfs_mount_t	*mp,		/* mount point for filesystem */
1858 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
1859 {
1860 	xfs_rtblock_t	bmbno;		/* bitmap block number */
1861 	xfs_buf_t	*bp;		/* temporary buffer */
1862 	int		error;		/* error return value */
1863 	xfs_inode_t	*ip;		/* bitmap inode, used as lock */
1864 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
1865 	xfs_drfsbno_t	nrblocks;	/* new number of realtime blocks */
1866 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
1867 	xfs_drtbno_t	nrextents;	/* new number of realtime extents */
1868 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
1869 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
1870 	uint		nrsumlevels;	/* new rt summary levels */
1871 	uint		nrsumsize;	/* new size of rt summary, bytes */
1872 	xfs_sb_t	*nsbp;		/* new superblock */
1873 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
1874 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
1875 	xfs_sb_t	*sbp;		/* old superblock */
1876 	xfs_fsblock_t	sumbno;		/* summary block number */
1877 
1878 	sbp = &mp->m_sb;
1879 	/*
1880 	 * Initial error checking.
1881 	 */
1882 	if (!capable(CAP_SYS_ADMIN))
1883 		return XFS_ERROR(EPERM);
1884 	if (mp->m_rtdev_targp == NULL || mp->m_rbmip == NULL ||
1885 	    (nrblocks = in->newblocks) <= sbp->sb_rblocks ||
1886 	    (sbp->sb_rblocks && (in->extsize != sbp->sb_rextsize)))
1887 		return XFS_ERROR(EINVAL);
1888 	if ((error = xfs_sb_validate_fsb_count(sbp, nrblocks)))
1889 		return error;
1890 	/*
1891 	 * Read in the last block of the device, make sure it exists.
1892 	 */
1893 	error = xfs_read_buf(mp, mp->m_rtdev_targp,
1894 			XFS_FSB_TO_BB(mp, nrblocks - 1),
1895 			XFS_FSB_TO_BB(mp, 1), 0, &bp);
1896 	if (error)
1897 		return error;
1898 	ASSERT(bp);
1899 	xfs_buf_relse(bp);
1900 	/*
1901 	 * Calculate new parameters.  These are the final values to be reached.
1902 	 */
1903 	nrextents = nrblocks;
1904 	do_div(nrextents, in->extsize);
1905 	nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
1906 	nrextslog = xfs_highbit32(nrextents);
1907 	nrsumlevels = nrextslog + 1;
1908 	nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
1909 	nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1910 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1911 	/*
1912 	 * New summary size can't be more than half the size of
1913 	 * the log.  This prevents us from getting a log overflow,
1914 	 * since we'll log basically the whole summary file at once.
1915 	 */
1916 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
1917 		return XFS_ERROR(EINVAL);
1918 	/*
1919 	 * Get the old block counts for bitmap and summary inodes.
1920 	 * These can't change since other growfs callers are locked out.
1921 	 */
1922 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_d.di_size);
1923 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_d.di_size);
1924 	/*
1925 	 * Allocate space to the bitmap and summary files, as necessary.
1926 	 */
1927 	if ((error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks,
1928 			mp->m_sb.sb_rbmino)))
1929 		return error;
1930 	if ((error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks,
1931 			mp->m_sb.sb_rsumino)))
1932 		return error;
1933 	/*
1934 	 * Allocate a new (fake) mount/sb.
1935 	 */
1936 	nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
1937 	/*
1938 	 * Loop over the bitmap blocks.
1939 	 * We will do everything one bitmap block at a time.
1940 	 * Skip the current block if it is exactly full.
1941 	 * This also deals with the case where there were no rtextents before.
1942 	 */
1943 	for (bmbno = sbp->sb_rbmblocks -
1944 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1945 	     bmbno < nrbmblocks;
1946 	     bmbno++) {
1947 		xfs_trans_t	*tp;
1948 		int		cancelflags = 0;
1949 
1950 		*nmp = *mp;
1951 		nsbp = &nmp->m_sb;
1952 		/*
1953 		 * Calculate new sb and mount fields for this round.
1954 		 */
1955 		nsbp->sb_rextsize = in->extsize;
1956 		nsbp->sb_rbmblocks = bmbno + 1;
1957 		nsbp->sb_rblocks =
1958 			XFS_RTMIN(nrblocks,
1959 				  nsbp->sb_rbmblocks * NBBY *
1960 				  nsbp->sb_blocksize * nsbp->sb_rextsize);
1961 		nsbp->sb_rextents = nsbp->sb_rblocks;
1962 		do_div(nsbp->sb_rextents, nsbp->sb_rextsize);
1963 		ASSERT(nsbp->sb_rextents != 0);
1964 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1965 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1966 		nrsumsize =
1967 			(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
1968 			nsbp->sb_rbmblocks;
1969 		nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1970 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1971 		/*
1972 		 * Start a transaction, get the log reservation.
1973 		 */
1974 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_FREE);
1975 		if ((error = xfs_trans_reserve(tp, 0,
1976 				XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
1977 			goto error_cancel;
1978 		/*
1979 		 * Lock out other callers by grabbing the bitmap inode lock.
1980 		 */
1981 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
1982 						XFS_ILOCK_EXCL, &ip)))
1983 			goto error_cancel;
1984 		ASSERT(ip == mp->m_rbmip);
1985 		/*
1986 		 * Update the bitmap inode's size.
1987 		 */
1988 		mp->m_rbmip->i_d.di_size =
1989 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1990 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1991 		cancelflags |= XFS_TRANS_ABORT;
1992 		/*
1993 		 * Get the summary inode into the transaction.
1994 		 */
1995 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rsumino, 0,
1996 						XFS_ILOCK_EXCL, &ip)))
1997 			goto error_cancel;
1998 		ASSERT(ip == mp->m_rsumip);
1999 		/*
2000 		 * Update the summary inode's size.
2001 		 */
2002 		mp->m_rsumip->i_d.di_size = nmp->m_rsumsize;
2003 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
2004 		/*
2005 		 * Copy summary data from old to new sizes.
2006 		 * Do this when the real size (not block-aligned) changes.
2007 		 */
2008 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
2009 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
2010 			error = xfs_rtcopy_summary(mp, nmp, tp);
2011 			if (error)
2012 				goto error_cancel;
2013 		}
2014 		/*
2015 		 * Update superblock fields.
2016 		 */
2017 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
2018 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
2019 				nsbp->sb_rextsize - sbp->sb_rextsize);
2020 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
2021 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
2022 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
2023 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
2024 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
2025 				nsbp->sb_rblocks - sbp->sb_rblocks);
2026 		if (nsbp->sb_rextents != sbp->sb_rextents)
2027 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
2028 				nsbp->sb_rextents - sbp->sb_rextents);
2029 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
2030 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
2031 				nsbp->sb_rextslog - sbp->sb_rextslog);
2032 		/*
2033 		 * Free new extent.
2034 		 */
2035 		bp = NULL;
2036 		error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
2037 			nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
2038 		if (error) {
2039 error_cancel:
2040 			xfs_trans_cancel(tp, cancelflags);
2041 			break;
2042 		}
2043 		/*
2044 		 * Mark more blocks free in the superblock.
2045 		 */
2046 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
2047 			nsbp->sb_rextents - sbp->sb_rextents);
2048 		/*
2049 		 * Update mp values into the real mp structure.
2050 		 */
2051 		mp->m_rsumlevels = nrsumlevels;
2052 		mp->m_rsumsize = nrsumsize;
2053 
2054 		error = xfs_trans_commit(tp, 0);
2055 		if (error)
2056 			break;
2057 	}
2058 
2059 	/*
2060 	 * Free the fake mp structure.
2061 	 */
2062 	kmem_free(nmp);
2063 
2064 	return error;
2065 }
2066 
2067 /*
2068  * Allocate an extent in the realtime subvolume, with the usual allocation
2069  * parameters.  The length units are all in realtime extents, as is the
2070  * result block number.
2071  */
2072 int					/* error */
2073 xfs_rtallocate_extent(
2074 	xfs_trans_t	*tp,		/* transaction pointer */
2075 	xfs_rtblock_t	bno,		/* starting block number to allocate */
2076 	xfs_extlen_t	minlen,		/* minimum length to allocate */
2077 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
2078 	xfs_extlen_t	*len,		/* out: actual length allocated */
2079 	xfs_alloctype_t	type,		/* allocation type XFS_ALLOCTYPE... */
2080 	int		wasdel,		/* was a delayed allocation extent */
2081 	xfs_extlen_t	prod,		/* extent product factor */
2082 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
2083 {
2084 	int		error;		/* error value */
2085 	xfs_inode_t	*ip;		/* inode for bitmap file */
2086 	xfs_mount_t	*mp;		/* file system mount structure */
2087 	xfs_rtblock_t	r;		/* result allocated block */
2088 	xfs_fsblock_t	sb;		/* summary file block number */
2089 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2090 
2091 	ASSERT(minlen > 0 && minlen <= maxlen);
2092 	mp = tp->t_mountp;
2093 	/*
2094 	 * If prod is set then figure out what to do to minlen and maxlen.
2095 	 */
2096 	if (prod > 1) {
2097 		xfs_extlen_t	i;
2098 
2099 		if ((i = maxlen % prod))
2100 			maxlen -= i;
2101 		if ((i = minlen % prod))
2102 			minlen += prod - i;
2103 		if (maxlen < minlen) {
2104 			*rtblock = NULLRTBLOCK;
2105 			return 0;
2106 		}
2107 	}
2108 	/*
2109 	 * Lock out other callers by grabbing the bitmap inode lock.
2110 	 */
2111 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2112 					XFS_ILOCK_EXCL, &ip)))
2113 		return error;
2114 	sumbp = NULL;
2115 	/*
2116 	 * Allocate by size, or near another block, or exactly at some block.
2117 	 */
2118 	switch (type) {
2119 	case XFS_ALLOCTYPE_ANY_AG:
2120 		error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
2121 				&sumbp,	&sb, prod, &r);
2122 		break;
2123 	case XFS_ALLOCTYPE_NEAR_BNO:
2124 		error = xfs_rtallocate_extent_near(mp, tp, bno, minlen, maxlen,
2125 				len, &sumbp, &sb, prod, &r);
2126 		break;
2127 	case XFS_ALLOCTYPE_THIS_BNO:
2128 		error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen,
2129 				len, &sumbp, &sb, prod, &r);
2130 		break;
2131 	default:
2132 		ASSERT(0);
2133 	}
2134 	if (error) {
2135 		return error;
2136 	}
2137 	/*
2138 	 * If it worked, update the superblock.
2139 	 */
2140 	if (r != NULLRTBLOCK) {
2141 		long	slen = (long)*len;
2142 
2143 		ASSERT(*len >= minlen && *len <= maxlen);
2144 		if (wasdel)
2145 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
2146 		else
2147 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
2148 	}
2149 	*rtblock = r;
2150 	return 0;
2151 }
2152 
2153 /*
2154  * Free an extent in the realtime subvolume.  Length is expressed in
2155  * realtime extents, as is the block number.
2156  */
2157 int					/* error */
2158 xfs_rtfree_extent(
2159 	xfs_trans_t	*tp,		/* transaction pointer */
2160 	xfs_rtblock_t	bno,		/* starting block number to free */
2161 	xfs_extlen_t	len)		/* length of extent freed */
2162 {
2163 	int		error;		/* error value */
2164 	xfs_inode_t	*ip;		/* bitmap file inode */
2165 	xfs_mount_t	*mp;		/* file system mount structure */
2166 	xfs_fsblock_t	sb;		/* summary file block number */
2167 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2168 
2169 	mp = tp->t_mountp;
2170 	/*
2171 	 * Synchronize by locking the bitmap inode.
2172 	 */
2173 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2174 					XFS_ILOCK_EXCL, &ip)))
2175 		return error;
2176 #if defined(__KERNEL__) && defined(DEBUG)
2177 	/*
2178 	 * Check to see that this whole range is currently allocated.
2179 	 */
2180 	{
2181 		int	stat;		/* result from checking range */
2182 
2183 		error = xfs_rtcheck_alloc_range(mp, tp, bno, len, &stat);
2184 		if (error) {
2185 			return error;
2186 		}
2187 		ASSERT(stat);
2188 	}
2189 #endif
2190 	sumbp = NULL;
2191 	/*
2192 	 * Free the range of realtime blocks.
2193 	 */
2194 	error = xfs_rtfree_range(mp, tp, bno, len, &sumbp, &sb);
2195 	if (error) {
2196 		return error;
2197 	}
2198 	/*
2199 	 * Mark more blocks free in the superblock.
2200 	 */
2201 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, (long)len);
2202 	/*
2203 	 * If we've now freed all the blocks, reset the file sequence
2204 	 * number to 0.
2205 	 */
2206 	if (tp->t_frextents_delta + mp->m_sb.sb_frextents ==
2207 	    mp->m_sb.sb_rextents) {
2208 		if (!(ip->i_d.di_flags & XFS_DIFLAG_NEWRTBM))
2209 			ip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2210 		*(__uint64_t *)&ip->i_d.di_atime = 0;
2211 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2212 	}
2213 	return 0;
2214 }
2215 
2216 /*
2217  * Initialize realtime fields in the mount structure.
2218  */
2219 int				/* error */
2220 xfs_rtmount_init(
2221 	xfs_mount_t	*mp)	/* file system mount structure */
2222 {
2223 	xfs_buf_t	*bp;	/* buffer for last block of subvolume */
2224 	xfs_daddr_t	d;	/* address of last block of subvolume */
2225 	int		error;	/* error return value */
2226 	xfs_sb_t	*sbp;	/* filesystem superblock copy in mount */
2227 
2228 	sbp = &mp->m_sb;
2229 	if (sbp->sb_rblocks == 0)
2230 		return 0;
2231 	if (mp->m_rtdev_targp == NULL) {
2232 		cmn_err(CE_WARN,
2233 	"XFS: This filesystem has a realtime volume, use rtdev=device option");
2234 		return XFS_ERROR(ENODEV);
2235 	}
2236 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
2237 	mp->m_rsumsize =
2238 		(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
2239 		sbp->sb_rbmblocks;
2240 	mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
2241 	mp->m_rbmip = mp->m_rsumip = NULL;
2242 	/*
2243 	 * Check that the realtime section is an ok size.
2244 	 */
2245 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
2246 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
2247 		cmn_err(CE_WARN, "XFS: realtime mount -- %llu != %llu",
2248 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
2249 			(unsigned long long) mp->m_sb.sb_rblocks);
2250 		return XFS_ERROR(EFBIG);
2251 	}
2252 	error = xfs_read_buf(mp, mp->m_rtdev_targp,
2253 				d - XFS_FSB_TO_BB(mp, 1),
2254 				XFS_FSB_TO_BB(mp, 1), 0, &bp);
2255 	if (error) {
2256 		cmn_err(CE_WARN,
2257 	"XFS: realtime mount -- xfs_read_buf failed, returned %d", error);
2258 		if (error == ENOSPC)
2259 			return XFS_ERROR(EFBIG);
2260 		return error;
2261 	}
2262 	xfs_buf_relse(bp);
2263 	return 0;
2264 }
2265 
2266 /*
2267  * Get the bitmap and summary inodes into the mount structure
2268  * at mount time.
2269  */
2270 int					/* error */
2271 xfs_rtmount_inodes(
2272 	xfs_mount_t	*mp)		/* file system mount structure */
2273 {
2274 	int		error;		/* error return value */
2275 	xfs_sb_t	*sbp;
2276 
2277 	sbp = &mp->m_sb;
2278 	if (sbp->sb_rbmino == NULLFSINO)
2279 		return 0;
2280 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip, 0);
2281 	if (error)
2282 		return error;
2283 	ASSERT(mp->m_rbmip != NULL);
2284 	ASSERT(sbp->sb_rsumino != NULLFSINO);
2285 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip, 0);
2286 	if (error) {
2287 		IRELE(mp->m_rbmip);
2288 		return error;
2289 	}
2290 	ASSERT(mp->m_rsumip != NULL);
2291 	return 0;
2292 }
2293 
2294 void
2295 xfs_rtunmount_inodes(
2296 	struct xfs_mount	*mp)
2297 {
2298 	if (mp->m_rbmip)
2299 		IRELE(mp->m_rbmip);
2300 	if (mp->m_rsumip)
2301 		IRELE(mp->m_rsumip);
2302 }
2303 
2304 /*
2305  * Pick an extent for allocation at the start of a new realtime file.
2306  * Use the sequence number stored in the atime field of the bitmap inode.
2307  * Translate this to a fraction of the rtextents, and return the product
2308  * of rtextents and the fraction.
2309  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
2310  */
2311 int					/* error */
2312 xfs_rtpick_extent(
2313 	xfs_mount_t	*mp,		/* file system mount point */
2314 	xfs_trans_t	*tp,		/* transaction pointer */
2315 	xfs_extlen_t	len,		/* allocation length (rtextents) */
2316 	xfs_rtblock_t	*pick)		/* result rt extent */
2317 {
2318 	xfs_rtblock_t	b;		/* result block */
2319 	int		error;		/* error return value */
2320 	xfs_inode_t	*ip;		/* bitmap incore inode */
2321 	int		log2;		/* log of sequence number */
2322 	__uint64_t	resid;		/* residual after log removed */
2323 	__uint64_t	seq;		/* sequence number of file creation */
2324 	__uint64_t	*seqp;		/* pointer to seqno in inode */
2325 
2326 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2327 					XFS_ILOCK_EXCL, &ip)))
2328 		return error;
2329 	ASSERT(ip == mp->m_rbmip);
2330 	seqp = (__uint64_t *)&ip->i_d.di_atime;
2331 	if (!(ip->i_d.di_flags & XFS_DIFLAG_NEWRTBM)) {
2332 		ip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2333 		*seqp = 0;
2334 	}
2335 	seq = *seqp;
2336 	if ((log2 = xfs_highbit64(seq)) == -1)
2337 		b = 0;
2338 	else {
2339 		resid = seq - (1ULL << log2);
2340 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
2341 		    (log2 + 1);
2342 		if (b >= mp->m_sb.sb_rextents)
2343 			b = do_mod(b, mp->m_sb.sb_rextents);
2344 		if (b + len > mp->m_sb.sb_rextents)
2345 			b = mp->m_sb.sb_rextents - len;
2346 	}
2347 	*seqp = seq + 1;
2348 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2349 	*pick = b;
2350 	return 0;
2351 }
2352