xref: /openbmc/linux/fs/xfs/xfs_rtalloc.c (revision 9c1f8594)
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_ref(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(tp, ip, oblocks, nblocks - oblocks,
124 			XFS_BMAPI_WRITE | 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_ref(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 error:
187 	return error;
188 }
189 
190 /*
191  * Attempt to allocate an extent minlen<=len<=maxlen starting from
192  * bitmap block bbno.  If we don't get maxlen then use prod to trim
193  * the length, if given.  Returns error; returns starting block in *rtblock.
194  * The lengths are all in rtextents.
195  */
196 STATIC int				/* error */
197 xfs_rtallocate_extent_block(
198 	xfs_mount_t	*mp,		/* file system mount point */
199 	xfs_trans_t	*tp,		/* transaction pointer */
200 	xfs_rtblock_t	bbno,		/* bitmap block number */
201 	xfs_extlen_t	minlen,		/* minimum length to allocate */
202 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
203 	xfs_extlen_t	*len,		/* out: actual length allocated */
204 	xfs_rtblock_t	*nextp,		/* out: next block to try */
205 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
206 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
207 	xfs_extlen_t	prod,		/* extent product factor */
208 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
209 {
210 	xfs_rtblock_t	besti;		/* best rtblock found so far */
211 	xfs_rtblock_t	bestlen;	/* best length found so far */
212 	xfs_rtblock_t	end;		/* last rtblock in chunk */
213 	int		error;		/* error value */
214 	xfs_rtblock_t	i;		/* current rtblock trying */
215 	xfs_rtblock_t	next;		/* next rtblock to try */
216 	int		stat;		/* status from internal calls */
217 
218 	/*
219 	 * Loop over all the extents starting in this bitmap block,
220 	 * looking for one that's long enough.
221 	 */
222 	for (i = XFS_BLOCKTOBIT(mp, bbno), besti = -1, bestlen = 0,
223 		end = XFS_BLOCKTOBIT(mp, bbno + 1) - 1;
224 	     i <= end;
225 	     i++) {
226 		/*
227 		 * See if there's a free extent of maxlen starting at i.
228 		 * If it's not so then next will contain the first non-free.
229 		 */
230 		error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
231 		if (error) {
232 			return error;
233 		}
234 		if (stat) {
235 			/*
236 			 * i for maxlen is all free, allocate and return that.
237 			 */
238 			error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
239 				rsb);
240 			if (error) {
241 				return error;
242 			}
243 			*len = maxlen;
244 			*rtblock = i;
245 			return 0;
246 		}
247 		/*
248 		 * In the case where we have a variable-sized allocation
249 		 * request, figure out how big this free piece is,
250 		 * and if it's big enough for the minimum, and the best
251 		 * so far, remember it.
252 		 */
253 		if (minlen < maxlen) {
254 			xfs_rtblock_t	thislen;	/* this extent size */
255 
256 			thislen = next - i;
257 			if (thislen >= minlen && thislen > bestlen) {
258 				besti = i;
259 				bestlen = thislen;
260 			}
261 		}
262 		/*
263 		 * If not done yet, find the start of the next free space.
264 		 */
265 		if (next < end) {
266 			error = xfs_rtfind_forw(mp, tp, next, end, &i);
267 			if (error) {
268 				return error;
269 			}
270 		} else
271 			break;
272 	}
273 	/*
274 	 * Searched the whole thing & didn't find a maxlen free extent.
275 	 */
276 	if (minlen < maxlen && besti != -1) {
277 		xfs_extlen_t	p;	/* amount to trim length by */
278 
279 		/*
280 		 * If size should be a multiple of prod, make that so.
281 		 */
282 		if (prod > 1 && (p = do_mod(bestlen, prod)))
283 			bestlen -= p;
284 		/*
285 		 * Allocate besti for bestlen & return that.
286 		 */
287 		error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
288 		if (error) {
289 			return error;
290 		}
291 		*len = bestlen;
292 		*rtblock = besti;
293 		return 0;
294 	}
295 	/*
296 	 * Allocation failed.  Set *nextp to the next block to try.
297 	 */
298 	*nextp = next;
299 	*rtblock = NULLRTBLOCK;
300 	return 0;
301 }
302 
303 /*
304  * Allocate an extent of length minlen<=len<=maxlen, starting at block
305  * bno.  If we don't get maxlen then use prod to trim the length, if given.
306  * Returns error; returns starting block in *rtblock.
307  * The lengths are all in rtextents.
308  */
309 STATIC int				/* error */
310 xfs_rtallocate_extent_exact(
311 	xfs_mount_t	*mp,		/* file system mount point */
312 	xfs_trans_t	*tp,		/* transaction pointer */
313 	xfs_rtblock_t	bno,		/* starting block number to allocate */
314 	xfs_extlen_t	minlen,		/* minimum length to allocate */
315 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
316 	xfs_extlen_t	*len,		/* out: actual length allocated */
317 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
318 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
319 	xfs_extlen_t	prod,		/* extent product factor */
320 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
321 {
322 	int		error;		/* error value */
323 	xfs_extlen_t	i;		/* extent length trimmed due to prod */
324 	int		isfree;		/* extent is free */
325 	xfs_rtblock_t	next;		/* next block to try (dummy) */
326 
327 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
328 	/*
329 	 * Check if the range in question (for maxlen) is free.
330 	 */
331 	error = xfs_rtcheck_range(mp, tp, bno, maxlen, 1, &next, &isfree);
332 	if (error) {
333 		return error;
334 	}
335 	if (isfree) {
336 		/*
337 		 * If it is, allocate it and return success.
338 		 */
339 		error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
340 		if (error) {
341 			return error;
342 		}
343 		*len = maxlen;
344 		*rtblock = bno;
345 		return 0;
346 	}
347 	/*
348 	 * If not, allocate what there is, if it's at least minlen.
349 	 */
350 	maxlen = next - bno;
351 	if (maxlen < minlen) {
352 		/*
353 		 * Failed, return failure status.
354 		 */
355 		*rtblock = NULLRTBLOCK;
356 		return 0;
357 	}
358 	/*
359 	 * Trim off tail of extent, if prod is specified.
360 	 */
361 	if (prod > 1 && (i = maxlen % prod)) {
362 		maxlen -= i;
363 		if (maxlen < minlen) {
364 			/*
365 			 * Now we can't do it, return failure status.
366 			 */
367 			*rtblock = NULLRTBLOCK;
368 			return 0;
369 		}
370 	}
371 	/*
372 	 * Allocate what we can and return it.
373 	 */
374 	error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
375 	if (error) {
376 		return error;
377 	}
378 	*len = maxlen;
379 	*rtblock = bno;
380 	return 0;
381 }
382 
383 /*
384  * Allocate an extent of length minlen<=len<=maxlen, starting as near
385  * to bno as possible.  If we don't get maxlen then use prod to trim
386  * the length, if given.  The lengths are all in rtextents.
387  */
388 STATIC int				/* error */
389 xfs_rtallocate_extent_near(
390 	xfs_mount_t	*mp,		/* file system mount point */
391 	xfs_trans_t	*tp,		/* transaction pointer */
392 	xfs_rtblock_t	bno,		/* starting block number to allocate */
393 	xfs_extlen_t	minlen,		/* minimum length to allocate */
394 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
395 	xfs_extlen_t	*len,		/* out: actual length allocated */
396 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
397 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
398 	xfs_extlen_t	prod,		/* extent product factor */
399 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
400 {
401 	int		any;		/* any useful extents from summary */
402 	xfs_rtblock_t	bbno;		/* bitmap block number */
403 	int		error;		/* error value */
404 	int		i;		/* bitmap block offset (loop control) */
405 	int		j;		/* secondary loop control */
406 	int		log2len;	/* log2 of minlen */
407 	xfs_rtblock_t	n;		/* next block to try */
408 	xfs_rtblock_t	r;		/* result block */
409 
410 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
411 	/*
412 	 * If the block number given is off the end, silently set it to
413 	 * the last block.
414 	 */
415 	if (bno >= mp->m_sb.sb_rextents)
416 		bno = mp->m_sb.sb_rextents - 1;
417 	/*
418 	 * Try the exact allocation first.
419 	 */
420 	error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen, len,
421 		rbpp, rsb, prod, &r);
422 	if (error) {
423 		return error;
424 	}
425 	/*
426 	 * If the exact allocation worked, return that.
427 	 */
428 	if (r != NULLRTBLOCK) {
429 		*rtblock = r;
430 		return 0;
431 	}
432 	bbno = XFS_BITTOBLOCK(mp, bno);
433 	i = 0;
434 	ASSERT(minlen != 0);
435 	log2len = xfs_highbit32(minlen);
436 	/*
437 	 * Loop over all bitmap blocks (bbno + i is current block).
438 	 */
439 	for (;;) {
440 		/*
441 		 * Get summary information of extents of all useful levels
442 		 * starting in this bitmap block.
443 		 */
444 		error = xfs_rtany_summary(mp, tp, log2len, mp->m_rsumlevels - 1,
445 			bbno + i, rbpp, rsb, &any);
446 		if (error) {
447 			return error;
448 		}
449 		/*
450 		 * If there are any useful extents starting here, try
451 		 * allocating one.
452 		 */
453 		if (any) {
454 			/*
455 			 * On the positive side of the starting location.
456 			 */
457 			if (i >= 0) {
458 				/*
459 				 * Try to allocate an extent starting in
460 				 * this block.
461 				 */
462 				error = xfs_rtallocate_extent_block(mp, tp,
463 					bbno + i, minlen, maxlen, len, &n, rbpp,
464 					rsb, prod, &r);
465 				if (error) {
466 					return error;
467 				}
468 				/*
469 				 * If it worked, return it.
470 				 */
471 				if (r != NULLRTBLOCK) {
472 					*rtblock = r;
473 					return 0;
474 				}
475 			}
476 			/*
477 			 * On the negative side of the starting location.
478 			 */
479 			else {		/* i < 0 */
480 				/*
481 				 * Loop backwards through the bitmap blocks from
482 				 * the starting point-1 up to where we are now.
483 				 * There should be an extent which ends in this
484 				 * bitmap block and is long enough.
485 				 */
486 				for (j = -1; j > i; j--) {
487 					/*
488 					 * Grab the summary information for
489 					 * this bitmap block.
490 					 */
491 					error = xfs_rtany_summary(mp, tp,
492 						log2len, mp->m_rsumlevels - 1,
493 						bbno + j, rbpp, rsb, &any);
494 					if (error) {
495 						return error;
496 					}
497 					/*
498 					 * If there's no extent given in the
499 					 * summary that means the extent we
500 					 * found must carry over from an
501 					 * earlier block.  If there is an
502 					 * extent given, we've already tried
503 					 * that allocation, don't do it again.
504 					 */
505 					if (any)
506 						continue;
507 					error = xfs_rtallocate_extent_block(mp,
508 						tp, bbno + j, minlen, maxlen,
509 						len, &n, rbpp, rsb, prod, &r);
510 					if (error) {
511 						return error;
512 					}
513 					/*
514 					 * If it works, return the extent.
515 					 */
516 					if (r != NULLRTBLOCK) {
517 						*rtblock = r;
518 						return 0;
519 					}
520 				}
521 				/*
522 				 * There weren't intervening bitmap blocks
523 				 * with a long enough extent, or the
524 				 * allocation didn't work for some reason
525 				 * (i.e. it's a little * too short).
526 				 * Try to allocate from the summary block
527 				 * that we found.
528 				 */
529 				error = xfs_rtallocate_extent_block(mp, tp,
530 					bbno + i, minlen, maxlen, len, &n, rbpp,
531 					rsb, prod, &r);
532 				if (error) {
533 					return error;
534 				}
535 				/*
536 				 * If it works, return the extent.
537 				 */
538 				if (r != NULLRTBLOCK) {
539 					*rtblock = r;
540 					return 0;
541 				}
542 			}
543 		}
544 		/*
545 		 * Loop control.  If we were on the positive side, and there's
546 		 * still more blocks on the negative side, go there.
547 		 */
548 		if (i > 0 && (int)bbno - i >= 0)
549 			i = -i;
550 		/*
551 		 * If positive, and no more negative, but there are more
552 		 * positive, go there.
553 		 */
554 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
555 			i++;
556 		/*
557 		 * If negative or 0 (just started), and there are positive
558 		 * blocks to go, go there.  The 0 case moves to block 1.
559 		 */
560 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
561 			i = 1 - i;
562 		/*
563 		 * If negative or 0 and there are more negative blocks,
564 		 * go there.
565 		 */
566 		else if (i <= 0 && (int)bbno + i > 0)
567 			i--;
568 		/*
569 		 * Must be done.  Return failure.
570 		 */
571 		else
572 			break;
573 	}
574 	*rtblock = NULLRTBLOCK;
575 	return 0;
576 }
577 
578 /*
579  * Allocate an extent of length minlen<=len<=maxlen, with no position
580  * specified.  If we don't get maxlen then use prod to trim
581  * the length, if given.  The lengths are all in rtextents.
582  */
583 STATIC int				/* error */
584 xfs_rtallocate_extent_size(
585 	xfs_mount_t	*mp,		/* file system mount point */
586 	xfs_trans_t	*tp,		/* transaction pointer */
587 	xfs_extlen_t	minlen,		/* minimum length to allocate */
588 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
589 	xfs_extlen_t	*len,		/* out: actual length allocated */
590 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
591 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
592 	xfs_extlen_t	prod,		/* extent product factor */
593 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
594 {
595 	int		error;		/* error value */
596 	int		i;		/* bitmap block number */
597 	int		l;		/* level number (loop control) */
598 	xfs_rtblock_t	n;		/* next block to be tried */
599 	xfs_rtblock_t	r;		/* result block number */
600 	xfs_suminfo_t	sum;		/* summary information for extents */
601 
602 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
603 	ASSERT(maxlen != 0);
604 
605 	/*
606 	 * Loop over all the levels starting with maxlen.
607 	 * At each level, look at all the bitmap blocks, to see if there
608 	 * are extents starting there that are long enough (>= maxlen).
609 	 * Note, only on the initial level can the allocation fail if
610 	 * the summary says there's an extent.
611 	 */
612 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
613 		/*
614 		 * Loop over all the bitmap blocks.
615 		 */
616 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
617 			/*
618 			 * Get the summary for this level/block.
619 			 */
620 			error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
621 				&sum);
622 			if (error) {
623 				return error;
624 			}
625 			/*
626 			 * Nothing there, on to the next block.
627 			 */
628 			if (!sum)
629 				continue;
630 			/*
631 			 * Try allocating the extent.
632 			 */
633 			error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
634 				maxlen, len, &n, rbpp, rsb, prod, &r);
635 			if (error) {
636 				return error;
637 			}
638 			/*
639 			 * If it worked, return that.
640 			 */
641 			if (r != NULLRTBLOCK) {
642 				*rtblock = r;
643 				return 0;
644 			}
645 			/*
646 			 * If the "next block to try" returned from the
647 			 * allocator is beyond the next bitmap block,
648 			 * skip to that bitmap block.
649 			 */
650 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
651 				i = XFS_BITTOBLOCK(mp, n) - 1;
652 		}
653 	}
654 	/*
655 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
656 	 * we're asking for a fixed size extent.
657 	 */
658 	if (minlen > --maxlen) {
659 		*rtblock = NULLRTBLOCK;
660 		return 0;
661 	}
662 	ASSERT(minlen != 0);
663 	ASSERT(maxlen != 0);
664 
665 	/*
666 	 * Loop over sizes, from maxlen down to minlen.
667 	 * This time, when we do the allocations, allow smaller ones
668 	 * to succeed.
669 	 */
670 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
671 		/*
672 		 * Loop over all the bitmap blocks, try an allocation
673 		 * starting in that block.
674 		 */
675 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
676 			/*
677 			 * Get the summary information for this level/block.
678 			 */
679 			error =	xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
680 						  &sum);
681 			if (error) {
682 				return error;
683 			}
684 			/*
685 			 * If nothing there, go on to next.
686 			 */
687 			if (!sum)
688 				continue;
689 			/*
690 			 * Try the allocation.  Make sure the specified
691 			 * minlen/maxlen are in the possible range for
692 			 * this summary level.
693 			 */
694 			error = xfs_rtallocate_extent_block(mp, tp, i,
695 					XFS_RTMAX(minlen, 1 << l),
696 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
697 					len, &n, rbpp, rsb, prod, &r);
698 			if (error) {
699 				return error;
700 			}
701 			/*
702 			 * If it worked, return that extent.
703 			 */
704 			if (r != NULLRTBLOCK) {
705 				*rtblock = r;
706 				return 0;
707 			}
708 			/*
709 			 * If the "next block to try" returned from the
710 			 * allocator is beyond the next bitmap block,
711 			 * skip to that bitmap block.
712 			 */
713 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
714 				i = XFS_BITTOBLOCK(mp, n) - 1;
715 		}
716 	}
717 	/*
718 	 * Got nothing, return failure.
719 	 */
720 	*rtblock = NULLRTBLOCK;
721 	return 0;
722 }
723 
724 /*
725  * Mark an extent specified by start and len allocated.
726  * Updates all the summary information as well as the bitmap.
727  */
728 STATIC int				/* error */
729 xfs_rtallocate_range(
730 	xfs_mount_t	*mp,		/* file system mount point */
731 	xfs_trans_t	*tp,		/* transaction pointer */
732 	xfs_rtblock_t	start,		/* start block to allocate */
733 	xfs_extlen_t	len,		/* length to allocate */
734 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
735 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
736 {
737 	xfs_rtblock_t	end;		/* end of the allocated extent */
738 	int		error;		/* error value */
739 	xfs_rtblock_t	postblock;	/* first block allocated > end */
740 	xfs_rtblock_t	preblock;	/* first block allocated < start */
741 
742 	end = start + len - 1;
743 	/*
744 	 * Assume we're allocating out of the middle of a free extent.
745 	 * We need to find the beginning and end of the extent so we can
746 	 * properly update the summary.
747 	 */
748 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
749 	if (error) {
750 		return error;
751 	}
752 	/*
753 	 * Find the next allocated block (end of free extent).
754 	 */
755 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
756 		&postblock);
757 	if (error) {
758 		return error;
759 	}
760 	/*
761 	 * Decrement the summary information corresponding to the entire
762 	 * (old) free extent.
763 	 */
764 	error = xfs_rtmodify_summary(mp, tp,
765 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
766 		XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
767 	if (error) {
768 		return error;
769 	}
770 	/*
771 	 * If there are blocks not being allocated at the front of the
772 	 * old extent, add summary data for them to be free.
773 	 */
774 	if (preblock < start) {
775 		error = xfs_rtmodify_summary(mp, tp,
776 			XFS_RTBLOCKLOG(start - preblock),
777 			XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
778 		if (error) {
779 			return error;
780 		}
781 	}
782 	/*
783 	 * If there are blocks not being allocated at the end of the
784 	 * old extent, add summary data for them to be free.
785 	 */
786 	if (postblock > end) {
787 		error = xfs_rtmodify_summary(mp, tp,
788 			XFS_RTBLOCKLOG(postblock - end),
789 			XFS_BITTOBLOCK(mp, end + 1), 1, rbpp, rsb);
790 		if (error) {
791 			return error;
792 		}
793 	}
794 	/*
795 	 * Modify the bitmap to mark this extent allocated.
796 	 */
797 	error = xfs_rtmodify_range(mp, tp, start, len, 0);
798 	return error;
799 }
800 
801 /*
802  * Return whether there are any free extents in the size range given
803  * by low and high, for the bitmap block bbno.
804  */
805 STATIC int				/* error */
806 xfs_rtany_summary(
807 	xfs_mount_t	*mp,		/* file system mount structure */
808 	xfs_trans_t	*tp,		/* transaction pointer */
809 	int		low,		/* low log2 extent size */
810 	int		high,		/* high log2 extent size */
811 	xfs_rtblock_t	bbno,		/* bitmap block number */
812 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
813 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
814 	int		*stat)		/* out: any good extents here? */
815 {
816 	int		error;		/* error value */
817 	int		log;		/* loop counter, log2 of ext. size */
818 	xfs_suminfo_t	sum;		/* summary data */
819 
820 	/*
821 	 * Loop over logs of extent sizes.  Order is irrelevant.
822 	 */
823 	for (log = low; log <= high; log++) {
824 		/*
825 		 * Get one summary datum.
826 		 */
827 		error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
828 		if (error) {
829 			return error;
830 		}
831 		/*
832 		 * If there are any, return success.
833 		 */
834 		if (sum) {
835 			*stat = 1;
836 			return 0;
837 		}
838 	}
839 	/*
840 	 * Found nothing, return failure.
841 	 */
842 	*stat = 0;
843 	return 0;
844 }
845 
846 /*
847  * Get a buffer for the bitmap or summary file block specified.
848  * The buffer is returned read and locked.
849  */
850 STATIC int				/* error */
851 xfs_rtbuf_get(
852 	xfs_mount_t	*mp,		/* file system mount structure */
853 	xfs_trans_t	*tp,		/* transaction pointer */
854 	xfs_rtblock_t	block,		/* block number in bitmap or summary */
855 	int		issum,		/* is summary not bitmap */
856 	xfs_buf_t	**bpp)		/* output: buffer for the block */
857 {
858 	xfs_buf_t	*bp;		/* block buffer, result */
859 	xfs_daddr_t	d;		/* disk addr of block */
860 	int		error;		/* error value */
861 	xfs_fsblock_t	fsb;		/* fs block number for block */
862 	xfs_inode_t	*ip;		/* bitmap or summary inode */
863 
864 	ip = issum ? mp->m_rsumip : mp->m_rbmip;
865 	/*
866 	 * Map from the file offset (block) and inode number to the
867 	 * file system block.
868 	 */
869 	error = xfs_bmapi_single(tp, ip, XFS_DATA_FORK, &fsb, block);
870 	if (error) {
871 		return error;
872 	}
873 	ASSERT(fsb != NULLFSBLOCK);
874 	/*
875 	 * Convert to disk address for buffer cache.
876 	 */
877 	d = XFS_FSB_TO_DADDR(mp, fsb);
878 	/*
879 	 * Read the buffer.
880 	 */
881 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
882 				   mp->m_bsize, 0, &bp);
883 	if (error) {
884 		return error;
885 	}
886 	ASSERT(!xfs_buf_geterror(bp));
887 	*bpp = bp;
888 	return 0;
889 }
890 
891 #ifdef DEBUG
892 /*
893  * Check that the given extent (block range) is allocated already.
894  */
895 STATIC int				/* error */
896 xfs_rtcheck_alloc_range(
897 	xfs_mount_t	*mp,		/* file system mount point */
898 	xfs_trans_t	*tp,		/* transaction pointer */
899 	xfs_rtblock_t	bno,		/* starting block number of extent */
900 	xfs_extlen_t	len,		/* length of extent */
901 	int		*stat)		/* out: 1 for allocated, 0 for not */
902 {
903 	xfs_rtblock_t	new;		/* dummy for xfs_rtcheck_range */
904 
905 	return xfs_rtcheck_range(mp, tp, bno, len, 0, &new, stat);
906 }
907 #endif
908 
909 /*
910  * Check that the given range is either all allocated (val = 0) or
911  * all free (val = 1).
912  */
913 STATIC int				/* error */
914 xfs_rtcheck_range(
915 	xfs_mount_t	*mp,		/* file system mount point */
916 	xfs_trans_t	*tp,		/* transaction pointer */
917 	xfs_rtblock_t	start,		/* starting block number of extent */
918 	xfs_extlen_t	len,		/* length of extent */
919 	int		val,		/* 1 for free, 0 for allocated */
920 	xfs_rtblock_t	*new,		/* out: first block not matching */
921 	int		*stat)		/* out: 1 for matches, 0 for not */
922 {
923 	xfs_rtword_t	*b;		/* current word in buffer */
924 	int		bit;		/* bit number in the word */
925 	xfs_rtblock_t	block;		/* bitmap block number */
926 	xfs_buf_t	*bp;		/* buf for the block */
927 	xfs_rtword_t	*bufp;		/* starting word in buffer */
928 	int		error;		/* error value */
929 	xfs_rtblock_t	i;		/* current bit number rel. to start */
930 	xfs_rtblock_t	lastbit;	/* last useful bit in word */
931 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
932 	xfs_rtword_t	wdiff;		/* difference from wanted value */
933 	int		word;		/* word number in the buffer */
934 
935 	/*
936 	 * Compute starting bitmap block number
937 	 */
938 	block = XFS_BITTOBLOCK(mp, start);
939 	/*
940 	 * Read the bitmap block.
941 	 */
942 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
943 	if (error) {
944 		return error;
945 	}
946 	bufp = bp->b_addr;
947 	/*
948 	 * Compute the starting word's address, and starting bit.
949 	 */
950 	word = XFS_BITTOWORD(mp, start);
951 	b = &bufp[word];
952 	bit = (int)(start & (XFS_NBWORD - 1));
953 	/*
954 	 * 0 (allocated) => all zero's; 1 (free) => all one's.
955 	 */
956 	val = -val;
957 	/*
958 	 * If not starting on a word boundary, deal with the first
959 	 * (partial) word.
960 	 */
961 	if (bit) {
962 		/*
963 		 * Compute first bit not examined.
964 		 */
965 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
966 		/*
967 		 * Mask of relevant bits.
968 		 */
969 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
970 		/*
971 		 * Compute difference between actual and desired value.
972 		 */
973 		if ((wdiff = (*b ^ val) & mask)) {
974 			/*
975 			 * Different, compute first wrong bit and return.
976 			 */
977 			xfs_trans_brelse(tp, bp);
978 			i = XFS_RTLOBIT(wdiff) - bit;
979 			*new = start + i;
980 			*stat = 0;
981 			return 0;
982 		}
983 		i = lastbit - bit;
984 		/*
985 		 * Go on to next block if that's where the next word is
986 		 * and we need the next word.
987 		 */
988 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
989 			/*
990 			 * If done with this block, get the next one.
991 			 */
992 			xfs_trans_brelse(tp, bp);
993 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
994 			if (error) {
995 				return error;
996 			}
997 			b = bufp = bp->b_addr;
998 			word = 0;
999 		} else {
1000 			/*
1001 			 * Go on to the next word in the buffer.
1002 			 */
1003 			b++;
1004 		}
1005 	} else {
1006 		/*
1007 		 * Starting on a word boundary, no partial word.
1008 		 */
1009 		i = 0;
1010 	}
1011 	/*
1012 	 * Loop over whole words in buffers.  When we use up one buffer
1013 	 * we move on to the next one.
1014 	 */
1015 	while (len - i >= XFS_NBWORD) {
1016 		/*
1017 		 * Compute difference between actual and desired value.
1018 		 */
1019 		if ((wdiff = *b ^ val)) {
1020 			/*
1021 			 * Different, compute first wrong bit and return.
1022 			 */
1023 			xfs_trans_brelse(tp, bp);
1024 			i += XFS_RTLOBIT(wdiff);
1025 			*new = start + i;
1026 			*stat = 0;
1027 			return 0;
1028 		}
1029 		i += XFS_NBWORD;
1030 		/*
1031 		 * Go on to next block if that's where the next word is
1032 		 * and we need the next word.
1033 		 */
1034 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1035 			/*
1036 			 * If done with this block, get the next one.
1037 			 */
1038 			xfs_trans_brelse(tp, bp);
1039 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1040 			if (error) {
1041 				return error;
1042 			}
1043 			b = bufp = bp->b_addr;
1044 			word = 0;
1045 		} else {
1046 			/*
1047 			 * Go on to the next word in the buffer.
1048 			 */
1049 			b++;
1050 		}
1051 	}
1052 	/*
1053 	 * If not ending on a word boundary, deal with the last
1054 	 * (partial) word.
1055 	 */
1056 	if ((lastbit = len - i)) {
1057 		/*
1058 		 * Mask of relevant bits.
1059 		 */
1060 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1061 		/*
1062 		 * Compute difference between actual and desired value.
1063 		 */
1064 		if ((wdiff = (*b ^ val) & mask)) {
1065 			/*
1066 			 * Different, compute first wrong bit and return.
1067 			 */
1068 			xfs_trans_brelse(tp, bp);
1069 			i += XFS_RTLOBIT(wdiff);
1070 			*new = start + i;
1071 			*stat = 0;
1072 			return 0;
1073 		} else
1074 			i = len;
1075 	}
1076 	/*
1077 	 * Successful, return.
1078 	 */
1079 	xfs_trans_brelse(tp, bp);
1080 	*new = start + i;
1081 	*stat = 1;
1082 	return 0;
1083 }
1084 
1085 /*
1086  * Copy and transform the summary file, given the old and new
1087  * parameters in the mount structures.
1088  */
1089 STATIC int				/* error */
1090 xfs_rtcopy_summary(
1091 	xfs_mount_t	*omp,		/* old file system mount point */
1092 	xfs_mount_t	*nmp,		/* new file system mount point */
1093 	xfs_trans_t	*tp)		/* transaction pointer */
1094 {
1095 	xfs_rtblock_t	bbno;		/* bitmap block number */
1096 	xfs_buf_t	*bp;		/* summary buffer */
1097 	int		error;		/* error return value */
1098 	int		log;		/* summary level number (log length) */
1099 	xfs_suminfo_t	sum;		/* summary data */
1100 	xfs_fsblock_t	sumbno;		/* summary block number */
1101 
1102 	bp = NULL;
1103 	for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
1104 		for (bbno = omp->m_sb.sb_rbmblocks - 1;
1105 		     (xfs_srtblock_t)bbno >= 0;
1106 		     bbno--) {
1107 			error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
1108 				&sumbno, &sum);
1109 			if (error)
1110 				return error;
1111 			if (sum == 0)
1112 				continue;
1113 			error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
1114 				&bp, &sumbno);
1115 			if (error)
1116 				return error;
1117 			error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
1118 				&bp, &sumbno);
1119 			if (error)
1120 				return error;
1121 			ASSERT(sum > 0);
1122 		}
1123 	}
1124 	return 0;
1125 }
1126 
1127 /*
1128  * Searching backward from start to limit, find the first block whose
1129  * allocated/free state is different from start's.
1130  */
1131 STATIC int				/* error */
1132 xfs_rtfind_back(
1133 	xfs_mount_t	*mp,		/* file system mount point */
1134 	xfs_trans_t	*tp,		/* transaction pointer */
1135 	xfs_rtblock_t	start,		/* starting block to look at */
1136 	xfs_rtblock_t	limit,		/* last block to look at */
1137 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1138 {
1139 	xfs_rtword_t	*b;		/* current word in buffer */
1140 	int		bit;		/* bit number in the word */
1141 	xfs_rtblock_t	block;		/* bitmap block number */
1142 	xfs_buf_t	*bp;		/* buf for the block */
1143 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1144 	int		error;		/* error value */
1145 	xfs_rtblock_t	firstbit;	/* first useful bit in the word */
1146 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1147 	xfs_rtblock_t	len;		/* length of inspected area */
1148 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1149 	xfs_rtword_t	want;		/* mask for "good" values */
1150 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1151 	int		word;		/* word number in the buffer */
1152 
1153 	/*
1154 	 * Compute and read in starting bitmap block for starting block.
1155 	 */
1156 	block = XFS_BITTOBLOCK(mp, start);
1157 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1158 	if (error) {
1159 		return error;
1160 	}
1161 	bufp = bp->b_addr;
1162 	/*
1163 	 * Get the first word's index & point to it.
1164 	 */
1165 	word = XFS_BITTOWORD(mp, start);
1166 	b = &bufp[word];
1167 	bit = (int)(start & (XFS_NBWORD - 1));
1168 	len = start - limit + 1;
1169 	/*
1170 	 * Compute match value, based on the bit at start: if 1 (free)
1171 	 * then all-ones, else all-zeroes.
1172 	 */
1173 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1174 	/*
1175 	 * If the starting position is not word-aligned, deal with the
1176 	 * partial word.
1177 	 */
1178 	if (bit < XFS_NBWORD - 1) {
1179 		/*
1180 		 * Calculate first (leftmost) bit number to look at,
1181 		 * and mask for all the relevant bits in this word.
1182 		 */
1183 		firstbit = XFS_RTMAX((xfs_srtblock_t)(bit - len + 1), 0);
1184 		mask = (((xfs_rtword_t)1 << (bit - firstbit + 1)) - 1) <<
1185 			firstbit;
1186 		/*
1187 		 * Calculate the difference between the value there
1188 		 * and what we're looking for.
1189 		 */
1190 		if ((wdiff = (*b ^ want) & mask)) {
1191 			/*
1192 			 * Different.  Mark where we are and return.
1193 			 */
1194 			xfs_trans_brelse(tp, bp);
1195 			i = bit - XFS_RTHIBIT(wdiff);
1196 			*rtblock = start - i + 1;
1197 			return 0;
1198 		}
1199 		i = bit - firstbit + 1;
1200 		/*
1201 		 * Go on to previous block if that's where the previous word is
1202 		 * and we need the previous word.
1203 		 */
1204 		if (--word == -1 && i < len) {
1205 			/*
1206 			 * If done with this block, get the previous one.
1207 			 */
1208 			xfs_trans_brelse(tp, bp);
1209 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1210 			if (error) {
1211 				return error;
1212 			}
1213 			bufp = bp->b_addr;
1214 			word = XFS_BLOCKWMASK(mp);
1215 			b = &bufp[word];
1216 		} else {
1217 			/*
1218 			 * Go on to the previous word in the buffer.
1219 			 */
1220 			b--;
1221 		}
1222 	} else {
1223 		/*
1224 		 * Starting on a word boundary, no partial word.
1225 		 */
1226 		i = 0;
1227 	}
1228 	/*
1229 	 * Loop over whole words in buffers.  When we use up one buffer
1230 	 * we move on to the previous one.
1231 	 */
1232 	while (len - i >= XFS_NBWORD) {
1233 		/*
1234 		 * Compute difference between actual and desired value.
1235 		 */
1236 		if ((wdiff = *b ^ want)) {
1237 			/*
1238 			 * Different, mark where we are and return.
1239 			 */
1240 			xfs_trans_brelse(tp, bp);
1241 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1242 			*rtblock = start - i + 1;
1243 			return 0;
1244 		}
1245 		i += XFS_NBWORD;
1246 		/*
1247 		 * Go on to previous block if that's where the previous word is
1248 		 * and we need the previous word.
1249 		 */
1250 		if (--word == -1 && i < len) {
1251 			/*
1252 			 * If done with this block, get the previous one.
1253 			 */
1254 			xfs_trans_brelse(tp, bp);
1255 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1256 			if (error) {
1257 				return error;
1258 			}
1259 			bufp = bp->b_addr;
1260 			word = XFS_BLOCKWMASK(mp);
1261 			b = &bufp[word];
1262 		} else {
1263 			/*
1264 			 * Go on to the previous word in the buffer.
1265 			 */
1266 			b--;
1267 		}
1268 	}
1269 	/*
1270 	 * If not ending on a word boundary, deal with the last
1271 	 * (partial) word.
1272 	 */
1273 	if (len - i) {
1274 		/*
1275 		 * Calculate first (leftmost) bit number to look at,
1276 		 * and mask for all the relevant bits in this word.
1277 		 */
1278 		firstbit = XFS_NBWORD - (len - i);
1279 		mask = (((xfs_rtword_t)1 << (len - i)) - 1) << firstbit;
1280 		/*
1281 		 * Compute difference between actual and desired value.
1282 		 */
1283 		if ((wdiff = (*b ^ want) & mask)) {
1284 			/*
1285 			 * Different, mark where we are and return.
1286 			 */
1287 			xfs_trans_brelse(tp, bp);
1288 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1289 			*rtblock = start - i + 1;
1290 			return 0;
1291 		} else
1292 			i = len;
1293 	}
1294 	/*
1295 	 * No match, return that we scanned the whole area.
1296 	 */
1297 	xfs_trans_brelse(tp, bp);
1298 	*rtblock = start - i + 1;
1299 	return 0;
1300 }
1301 
1302 /*
1303  * Searching forward from start to limit, find the first block whose
1304  * allocated/free state is different from start's.
1305  */
1306 STATIC int				/* error */
1307 xfs_rtfind_forw(
1308 	xfs_mount_t	*mp,		/* file system mount point */
1309 	xfs_trans_t	*tp,		/* transaction pointer */
1310 	xfs_rtblock_t	start,		/* starting block to look at */
1311 	xfs_rtblock_t	limit,		/* last block to look at */
1312 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1313 {
1314 	xfs_rtword_t	*b;		/* current word in buffer */
1315 	int		bit;		/* bit number in the word */
1316 	xfs_rtblock_t	block;		/* bitmap block number */
1317 	xfs_buf_t	*bp;		/* buf for the block */
1318 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1319 	int		error;		/* error value */
1320 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1321 	xfs_rtblock_t	lastbit;	/* last useful bit in the word */
1322 	xfs_rtblock_t	len;		/* length of inspected area */
1323 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1324 	xfs_rtword_t	want;		/* mask for "good" values */
1325 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1326 	int		word;		/* word number in the buffer */
1327 
1328 	/*
1329 	 * Compute and read in starting bitmap block for starting block.
1330 	 */
1331 	block = XFS_BITTOBLOCK(mp, start);
1332 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1333 	if (error) {
1334 		return error;
1335 	}
1336 	bufp = bp->b_addr;
1337 	/*
1338 	 * Get the first word's index & point to it.
1339 	 */
1340 	word = XFS_BITTOWORD(mp, start);
1341 	b = &bufp[word];
1342 	bit = (int)(start & (XFS_NBWORD - 1));
1343 	len = limit - start + 1;
1344 	/*
1345 	 * Compute match value, based on the bit at start: if 1 (free)
1346 	 * then all-ones, else all-zeroes.
1347 	 */
1348 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1349 	/*
1350 	 * If the starting position is not word-aligned, deal with the
1351 	 * partial word.
1352 	 */
1353 	if (bit) {
1354 		/*
1355 		 * Calculate last (rightmost) bit number to look at,
1356 		 * and mask for all the relevant bits in this word.
1357 		 */
1358 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1359 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1360 		/*
1361 		 * Calculate the difference between the value there
1362 		 * and what we're looking for.
1363 		 */
1364 		if ((wdiff = (*b ^ want) & mask)) {
1365 			/*
1366 			 * Different.  Mark where we are and return.
1367 			 */
1368 			xfs_trans_brelse(tp, bp);
1369 			i = XFS_RTLOBIT(wdiff) - bit;
1370 			*rtblock = start + i - 1;
1371 			return 0;
1372 		}
1373 		i = lastbit - bit;
1374 		/*
1375 		 * Go on to next block if that's where the next word is
1376 		 * and we need the next word.
1377 		 */
1378 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1379 			/*
1380 			 * If done with this block, get the previous one.
1381 			 */
1382 			xfs_trans_brelse(tp, bp);
1383 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1384 			if (error) {
1385 				return error;
1386 			}
1387 			b = bufp = bp->b_addr;
1388 			word = 0;
1389 		} else {
1390 			/*
1391 			 * Go on to the previous word in the buffer.
1392 			 */
1393 			b++;
1394 		}
1395 	} else {
1396 		/*
1397 		 * Starting on a word boundary, no partial word.
1398 		 */
1399 		i = 0;
1400 	}
1401 	/*
1402 	 * Loop over whole words in buffers.  When we use up one buffer
1403 	 * we move on to the next one.
1404 	 */
1405 	while (len - i >= XFS_NBWORD) {
1406 		/*
1407 		 * Compute difference between actual and desired value.
1408 		 */
1409 		if ((wdiff = *b ^ want)) {
1410 			/*
1411 			 * Different, mark where we are and return.
1412 			 */
1413 			xfs_trans_brelse(tp, bp);
1414 			i += XFS_RTLOBIT(wdiff);
1415 			*rtblock = start + i - 1;
1416 			return 0;
1417 		}
1418 		i += XFS_NBWORD;
1419 		/*
1420 		 * Go on to next block if that's where the next word is
1421 		 * and we need the next word.
1422 		 */
1423 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1424 			/*
1425 			 * If done with this block, get the next one.
1426 			 */
1427 			xfs_trans_brelse(tp, bp);
1428 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1429 			if (error) {
1430 				return error;
1431 			}
1432 			b = bufp = bp->b_addr;
1433 			word = 0;
1434 		} else {
1435 			/*
1436 			 * Go on to the next word in the buffer.
1437 			 */
1438 			b++;
1439 		}
1440 	}
1441 	/*
1442 	 * If not ending on a word boundary, deal with the last
1443 	 * (partial) word.
1444 	 */
1445 	if ((lastbit = len - i)) {
1446 		/*
1447 		 * Calculate mask for all the relevant bits in this word.
1448 		 */
1449 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1450 		/*
1451 		 * Compute difference between actual and desired value.
1452 		 */
1453 		if ((wdiff = (*b ^ want) & mask)) {
1454 			/*
1455 			 * Different, mark where we are and return.
1456 			 */
1457 			xfs_trans_brelse(tp, bp);
1458 			i += XFS_RTLOBIT(wdiff);
1459 			*rtblock = start + i - 1;
1460 			return 0;
1461 		} else
1462 			i = len;
1463 	}
1464 	/*
1465 	 * No match, return that we scanned the whole area.
1466 	 */
1467 	xfs_trans_brelse(tp, bp);
1468 	*rtblock = start + i - 1;
1469 	return 0;
1470 }
1471 
1472 /*
1473  * Mark an extent specified by start and len freed.
1474  * Updates all the summary information as well as the bitmap.
1475  */
1476 STATIC int				/* error */
1477 xfs_rtfree_range(
1478 	xfs_mount_t	*mp,		/* file system mount point */
1479 	xfs_trans_t	*tp,		/* transaction pointer */
1480 	xfs_rtblock_t	start,		/* starting block to free */
1481 	xfs_extlen_t	len,		/* length to free */
1482 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1483 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1484 {
1485 	xfs_rtblock_t	end;		/* end of the freed extent */
1486 	int		error;		/* error value */
1487 	xfs_rtblock_t	postblock;	/* first block freed > end */
1488 	xfs_rtblock_t	preblock;	/* first block freed < start */
1489 
1490 	end = start + len - 1;
1491 	/*
1492 	 * Modify the bitmap to mark this extent freed.
1493 	 */
1494 	error = xfs_rtmodify_range(mp, tp, start, len, 1);
1495 	if (error) {
1496 		return error;
1497 	}
1498 	/*
1499 	 * Assume we're freeing out of the middle of an allocated extent.
1500 	 * We need to find the beginning and end of the extent so we can
1501 	 * properly update the summary.
1502 	 */
1503 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
1504 	if (error) {
1505 		return error;
1506 	}
1507 	/*
1508 	 * Find the next allocated block (end of allocated extent).
1509 	 */
1510 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
1511 		&postblock);
1512 	if (error)
1513 		return error;
1514 	/*
1515 	 * If there are blocks not being freed at the front of the
1516 	 * old extent, add summary data for them to be allocated.
1517 	 */
1518 	if (preblock < start) {
1519 		error = xfs_rtmodify_summary(mp, tp,
1520 			XFS_RTBLOCKLOG(start - preblock),
1521 			XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
1522 		if (error) {
1523 			return error;
1524 		}
1525 	}
1526 	/*
1527 	 * If there are blocks not being freed at the end of the
1528 	 * old extent, add summary data for them to be allocated.
1529 	 */
1530 	if (postblock > end) {
1531 		error = xfs_rtmodify_summary(mp, tp,
1532 			XFS_RTBLOCKLOG(postblock - end),
1533 			XFS_BITTOBLOCK(mp, end + 1), -1, rbpp, rsb);
1534 		if (error) {
1535 			return error;
1536 		}
1537 	}
1538 	/*
1539 	 * Increment the summary information corresponding to the entire
1540 	 * (new) free extent.
1541 	 */
1542 	error = xfs_rtmodify_summary(mp, tp,
1543 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
1544 		XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
1545 	return error;
1546 }
1547 
1548 /*
1549  * Read and return the summary information for a given extent size,
1550  * bitmap block combination.
1551  * Keeps track of a current summary block, so we don't keep reading
1552  * it from the buffer cache.
1553  */
1554 STATIC int				/* error */
1555 xfs_rtget_summary(
1556 	xfs_mount_t	*mp,		/* file system mount structure */
1557 	xfs_trans_t	*tp,		/* transaction pointer */
1558 	int		log,		/* log2 of extent size */
1559 	xfs_rtblock_t	bbno,		/* bitmap block number */
1560 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1561 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
1562 	xfs_suminfo_t	*sum)		/* out: summary info for this block */
1563 {
1564 	xfs_buf_t	*bp;		/* buffer for summary block */
1565 	int		error;		/* error value */
1566 	xfs_fsblock_t	sb;		/* summary fsblock */
1567 	int		so;		/* index into the summary file */
1568 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1569 
1570 	/*
1571 	 * Compute entry number in the summary file.
1572 	 */
1573 	so = XFS_SUMOFFS(mp, log, bbno);
1574 	/*
1575 	 * Compute the block number in the summary file.
1576 	 */
1577 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1578 	/*
1579 	 * If we have an old buffer, and the block number matches, use that.
1580 	 */
1581 	if (rbpp && *rbpp && *rsb == sb)
1582 		bp = *rbpp;
1583 	/*
1584 	 * Otherwise we have to get the buffer.
1585 	 */
1586 	else {
1587 		/*
1588 		 * If there was an old one, get rid of it first.
1589 		 */
1590 		if (rbpp && *rbpp)
1591 			xfs_trans_brelse(tp, *rbpp);
1592 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1593 		if (error) {
1594 			return error;
1595 		}
1596 		/*
1597 		 * Remember this buffer and block for the next call.
1598 		 */
1599 		if (rbpp) {
1600 			*rbpp = bp;
1601 			*rsb = sb;
1602 		}
1603 	}
1604 	/*
1605 	 * Point to the summary information & copy it out.
1606 	 */
1607 	sp = XFS_SUMPTR(mp, bp, so);
1608 	*sum = *sp;
1609 	/*
1610 	 * Drop the buffer if we're not asked to remember it.
1611 	 */
1612 	if (!rbpp)
1613 		xfs_trans_brelse(tp, bp);
1614 	return 0;
1615 }
1616 
1617 /*
1618  * Set the given range of bitmap bits to the given value.
1619  * Do whatever I/O and logging is required.
1620  */
1621 STATIC int				/* error */
1622 xfs_rtmodify_range(
1623 	xfs_mount_t	*mp,		/* file system mount point */
1624 	xfs_trans_t	*tp,		/* transaction pointer */
1625 	xfs_rtblock_t	start,		/* starting block to modify */
1626 	xfs_extlen_t	len,		/* length of extent to modify */
1627 	int		val)		/* 1 for free, 0 for allocated */
1628 {
1629 	xfs_rtword_t	*b;		/* current word in buffer */
1630 	int		bit;		/* bit number in the word */
1631 	xfs_rtblock_t	block;		/* bitmap block number */
1632 	xfs_buf_t	*bp;		/* buf for the block */
1633 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1634 	int		error;		/* error value */
1635 	xfs_rtword_t	*first;		/* first used word in the buffer */
1636 	int		i;		/* current bit number rel. to start */
1637 	int		lastbit;	/* last useful bit in word */
1638 	xfs_rtword_t	mask;		/* mask o frelevant bits for value */
1639 	int		word;		/* word number in the buffer */
1640 
1641 	/*
1642 	 * Compute starting bitmap block number.
1643 	 */
1644 	block = XFS_BITTOBLOCK(mp, start);
1645 	/*
1646 	 * Read the bitmap block, and point to its data.
1647 	 */
1648 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1649 	if (error) {
1650 		return error;
1651 	}
1652 	bufp = bp->b_addr;
1653 	/*
1654 	 * Compute the starting word's address, and starting bit.
1655 	 */
1656 	word = XFS_BITTOWORD(mp, start);
1657 	first = b = &bufp[word];
1658 	bit = (int)(start & (XFS_NBWORD - 1));
1659 	/*
1660 	 * 0 (allocated) => all zeroes; 1 (free) => all ones.
1661 	 */
1662 	val = -val;
1663 	/*
1664 	 * If not starting on a word boundary, deal with the first
1665 	 * (partial) word.
1666 	 */
1667 	if (bit) {
1668 		/*
1669 		 * Compute first bit not changed and mask of relevant bits.
1670 		 */
1671 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1672 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1673 		/*
1674 		 * Set/clear the active bits.
1675 		 */
1676 		if (val)
1677 			*b |= mask;
1678 		else
1679 			*b &= ~mask;
1680 		i = lastbit - bit;
1681 		/*
1682 		 * Go on to the next block if that's where the next word is
1683 		 * and we need the next word.
1684 		 */
1685 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1686 			/*
1687 			 * Log the changed part of this block.
1688 			 * Get the next one.
1689 			 */
1690 			xfs_trans_log_buf(tp, bp,
1691 				(uint)((char *)first - (char *)bufp),
1692 				(uint)((char *)b - (char *)bufp));
1693 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1694 			if (error) {
1695 				return error;
1696 			}
1697 			first = b = bufp = bp->b_addr;
1698 			word = 0;
1699 		} else {
1700 			/*
1701 			 * Go on to the next word in the buffer
1702 			 */
1703 			b++;
1704 		}
1705 	} else {
1706 		/*
1707 		 * Starting on a word boundary, no partial word.
1708 		 */
1709 		i = 0;
1710 	}
1711 	/*
1712 	 * Loop over whole words in buffers.  When we use up one buffer
1713 	 * we move on to the next one.
1714 	 */
1715 	while (len - i >= XFS_NBWORD) {
1716 		/*
1717 		 * Set the word value correctly.
1718 		 */
1719 		*b = val;
1720 		i += XFS_NBWORD;
1721 		/*
1722 		 * Go on to the next block if that's where the next word is
1723 		 * and we need the next word.
1724 		 */
1725 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1726 			/*
1727 			 * Log the changed part of this block.
1728 			 * Get the next one.
1729 			 */
1730 			xfs_trans_log_buf(tp, bp,
1731 				(uint)((char *)first - (char *)bufp),
1732 				(uint)((char *)b - (char *)bufp));
1733 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1734 			if (error) {
1735 				return error;
1736 			}
1737 			first = b = bufp = bp->b_addr;
1738 			word = 0;
1739 		} else {
1740 			/*
1741 			 * Go on to the next word in the buffer
1742 			 */
1743 			b++;
1744 		}
1745 	}
1746 	/*
1747 	 * If not ending on a word boundary, deal with the last
1748 	 * (partial) word.
1749 	 */
1750 	if ((lastbit = len - i)) {
1751 		/*
1752 		 * Compute a mask of relevant bits.
1753 		 */
1754 		bit = 0;
1755 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1756 		/*
1757 		 * Set/clear the active bits.
1758 		 */
1759 		if (val)
1760 			*b |= mask;
1761 		else
1762 			*b &= ~mask;
1763 		b++;
1764 	}
1765 	/*
1766 	 * Log any remaining changed bytes.
1767 	 */
1768 	if (b > first)
1769 		xfs_trans_log_buf(tp, bp, (uint)((char *)first - (char *)bufp),
1770 			(uint)((char *)b - (char *)bufp - 1));
1771 	return 0;
1772 }
1773 
1774 /*
1775  * Read and modify the summary information for a given extent size,
1776  * bitmap block combination.
1777  * Keeps track of a current summary block, so we don't keep reading
1778  * it from the buffer cache.
1779  */
1780 STATIC int				/* error */
1781 xfs_rtmodify_summary(
1782 	xfs_mount_t	*mp,		/* file system mount point */
1783 	xfs_trans_t	*tp,		/* transaction pointer */
1784 	int		log,		/* log2 of extent size */
1785 	xfs_rtblock_t	bbno,		/* bitmap block number */
1786 	int		delta,		/* change to make to summary info */
1787 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1788 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1789 {
1790 	xfs_buf_t	*bp;		/* buffer for the summary block */
1791 	int		error;		/* error value */
1792 	xfs_fsblock_t	sb;		/* summary fsblock */
1793 	int		so;		/* index into the summary file */
1794 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1795 
1796 	/*
1797 	 * Compute entry number in the summary file.
1798 	 */
1799 	so = XFS_SUMOFFS(mp, log, bbno);
1800 	/*
1801 	 * Compute the block number in the summary file.
1802 	 */
1803 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1804 	/*
1805 	 * If we have an old buffer, and the block number matches, use that.
1806 	 */
1807 	if (rbpp && *rbpp && *rsb == sb)
1808 		bp = *rbpp;
1809 	/*
1810 	 * Otherwise we have to get the buffer.
1811 	 */
1812 	else {
1813 		/*
1814 		 * If there was an old one, get rid of it first.
1815 		 */
1816 		if (rbpp && *rbpp)
1817 			xfs_trans_brelse(tp, *rbpp);
1818 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1819 		if (error) {
1820 			return error;
1821 		}
1822 		/*
1823 		 * Remember this buffer and block for the next call.
1824 		 */
1825 		if (rbpp) {
1826 			*rbpp = bp;
1827 			*rsb = sb;
1828 		}
1829 	}
1830 	/*
1831 	 * Point to the summary information, modify and log it.
1832 	 */
1833 	sp = XFS_SUMPTR(mp, bp, so);
1834 	*sp += delta;
1835 	xfs_trans_log_buf(tp, bp, (uint)((char *)sp - (char *)bp->b_addr),
1836 		(uint)((char *)sp - (char *)bp->b_addr + sizeof(*sp) - 1));
1837 	return 0;
1838 }
1839 
1840 /*
1841  * Visible (exported) functions.
1842  */
1843 
1844 /*
1845  * Grow the realtime area of the filesystem.
1846  */
1847 int
1848 xfs_growfs_rt(
1849 	xfs_mount_t	*mp,		/* mount point for filesystem */
1850 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
1851 {
1852 	xfs_rtblock_t	bmbno;		/* bitmap block number */
1853 	xfs_buf_t	*bp;		/* temporary buffer */
1854 	int		error;		/* error return value */
1855 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
1856 	xfs_drfsbno_t	nrblocks;	/* new number of realtime blocks */
1857 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
1858 	xfs_drtbno_t	nrextents;	/* new number of realtime extents */
1859 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
1860 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
1861 	uint		nrsumlevels;	/* new rt summary levels */
1862 	uint		nrsumsize;	/* new size of rt summary, bytes */
1863 	xfs_sb_t	*nsbp;		/* new superblock */
1864 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
1865 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
1866 	xfs_sb_t	*sbp;		/* old superblock */
1867 	xfs_fsblock_t	sumbno;		/* summary block number */
1868 
1869 	sbp = &mp->m_sb;
1870 	/*
1871 	 * Initial error checking.
1872 	 */
1873 	if (!capable(CAP_SYS_ADMIN))
1874 		return XFS_ERROR(EPERM);
1875 	if (mp->m_rtdev_targp == NULL || mp->m_rbmip == NULL ||
1876 	    (nrblocks = in->newblocks) <= sbp->sb_rblocks ||
1877 	    (sbp->sb_rblocks && (in->extsize != sbp->sb_rextsize)))
1878 		return XFS_ERROR(EINVAL);
1879 	if ((error = xfs_sb_validate_fsb_count(sbp, nrblocks)))
1880 		return error;
1881 	/*
1882 	 * Read in the last block of the device, make sure it exists.
1883 	 */
1884 	bp = xfs_buf_read_uncached(mp, mp->m_rtdev_targp,
1885 				XFS_FSB_TO_BB(mp, nrblocks - 1),
1886 				XFS_FSB_TO_B(mp, 1), 0);
1887 	if (!bp)
1888 		return EIO;
1889 	xfs_buf_relse(bp);
1890 
1891 	/*
1892 	 * Calculate new parameters.  These are the final values to be reached.
1893 	 */
1894 	nrextents = nrblocks;
1895 	do_div(nrextents, in->extsize);
1896 	nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
1897 	nrextslog = xfs_highbit32(nrextents);
1898 	nrsumlevels = nrextslog + 1;
1899 	nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
1900 	nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1901 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1902 	/*
1903 	 * New summary size can't be more than half the size of
1904 	 * the log.  This prevents us from getting a log overflow,
1905 	 * since we'll log basically the whole summary file at once.
1906 	 */
1907 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
1908 		return XFS_ERROR(EINVAL);
1909 	/*
1910 	 * Get the old block counts for bitmap and summary inodes.
1911 	 * These can't change since other growfs callers are locked out.
1912 	 */
1913 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_d.di_size);
1914 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_d.di_size);
1915 	/*
1916 	 * Allocate space to the bitmap and summary files, as necessary.
1917 	 */
1918 	error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
1919 	if (error)
1920 		return error;
1921 	error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
1922 	if (error)
1923 		return error;
1924 	/*
1925 	 * Allocate a new (fake) mount/sb.
1926 	 */
1927 	nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
1928 	/*
1929 	 * Loop over the bitmap blocks.
1930 	 * We will do everything one bitmap block at a time.
1931 	 * Skip the current block if it is exactly full.
1932 	 * This also deals with the case where there were no rtextents before.
1933 	 */
1934 	for (bmbno = sbp->sb_rbmblocks -
1935 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1936 	     bmbno < nrbmblocks;
1937 	     bmbno++) {
1938 		xfs_trans_t	*tp;
1939 		int		cancelflags = 0;
1940 
1941 		*nmp = *mp;
1942 		nsbp = &nmp->m_sb;
1943 		/*
1944 		 * Calculate new sb and mount fields for this round.
1945 		 */
1946 		nsbp->sb_rextsize = in->extsize;
1947 		nsbp->sb_rbmblocks = bmbno + 1;
1948 		nsbp->sb_rblocks =
1949 			XFS_RTMIN(nrblocks,
1950 				  nsbp->sb_rbmblocks * NBBY *
1951 				  nsbp->sb_blocksize * nsbp->sb_rextsize);
1952 		nsbp->sb_rextents = nsbp->sb_rblocks;
1953 		do_div(nsbp->sb_rextents, nsbp->sb_rextsize);
1954 		ASSERT(nsbp->sb_rextents != 0);
1955 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1956 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1957 		nrsumsize =
1958 			(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
1959 			nsbp->sb_rbmblocks;
1960 		nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1961 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1962 		/*
1963 		 * Start a transaction, get the log reservation.
1964 		 */
1965 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_FREE);
1966 		if ((error = xfs_trans_reserve(tp, 0,
1967 				XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
1968 			goto error_cancel;
1969 		/*
1970 		 * Lock out other callers by grabbing the bitmap inode lock.
1971 		 */
1972 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
1973 		xfs_trans_ijoin_ref(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
1974 		/*
1975 		 * Update the bitmap inode's size.
1976 		 */
1977 		mp->m_rbmip->i_d.di_size =
1978 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1979 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1980 		cancelflags |= XFS_TRANS_ABORT;
1981 		/*
1982 		 * Get the summary inode into the transaction.
1983 		 */
1984 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL);
1985 		xfs_trans_ijoin_ref(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
1986 		/*
1987 		 * Update the summary inode's size.
1988 		 */
1989 		mp->m_rsumip->i_d.di_size = nmp->m_rsumsize;
1990 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
1991 		/*
1992 		 * Copy summary data from old to new sizes.
1993 		 * Do this when the real size (not block-aligned) changes.
1994 		 */
1995 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
1996 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
1997 			error = xfs_rtcopy_summary(mp, nmp, tp);
1998 			if (error)
1999 				goto error_cancel;
2000 		}
2001 		/*
2002 		 * Update superblock fields.
2003 		 */
2004 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
2005 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
2006 				nsbp->sb_rextsize - sbp->sb_rextsize);
2007 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
2008 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
2009 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
2010 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
2011 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
2012 				nsbp->sb_rblocks - sbp->sb_rblocks);
2013 		if (nsbp->sb_rextents != sbp->sb_rextents)
2014 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
2015 				nsbp->sb_rextents - sbp->sb_rextents);
2016 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
2017 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
2018 				nsbp->sb_rextslog - sbp->sb_rextslog);
2019 		/*
2020 		 * Free new extent.
2021 		 */
2022 		bp = NULL;
2023 		error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
2024 			nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
2025 		if (error) {
2026 error_cancel:
2027 			xfs_trans_cancel(tp, cancelflags);
2028 			break;
2029 		}
2030 		/*
2031 		 * Mark more blocks free in the superblock.
2032 		 */
2033 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
2034 			nsbp->sb_rextents - sbp->sb_rextents);
2035 		/*
2036 		 * Update mp values into the real mp structure.
2037 		 */
2038 		mp->m_rsumlevels = nrsumlevels;
2039 		mp->m_rsumsize = nrsumsize;
2040 
2041 		error = xfs_trans_commit(tp, 0);
2042 		if (error)
2043 			break;
2044 	}
2045 
2046 	/*
2047 	 * Free the fake mp structure.
2048 	 */
2049 	kmem_free(nmp);
2050 
2051 	return error;
2052 }
2053 
2054 /*
2055  * Allocate an extent in the realtime subvolume, with the usual allocation
2056  * parameters.  The length units are all in realtime extents, as is the
2057  * result block number.
2058  */
2059 int					/* error */
2060 xfs_rtallocate_extent(
2061 	xfs_trans_t	*tp,		/* transaction pointer */
2062 	xfs_rtblock_t	bno,		/* starting block number to allocate */
2063 	xfs_extlen_t	minlen,		/* minimum length to allocate */
2064 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
2065 	xfs_extlen_t	*len,		/* out: actual length allocated */
2066 	xfs_alloctype_t	type,		/* allocation type XFS_ALLOCTYPE... */
2067 	int		wasdel,		/* was a delayed allocation extent */
2068 	xfs_extlen_t	prod,		/* extent product factor */
2069 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
2070 {
2071 	xfs_mount_t	*mp = tp->t_mountp;
2072 	int		error;		/* error value */
2073 	xfs_rtblock_t	r;		/* result allocated block */
2074 	xfs_fsblock_t	sb;		/* summary file block number */
2075 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2076 
2077 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2078 	ASSERT(minlen > 0 && minlen <= maxlen);
2079 
2080 	/*
2081 	 * If prod is set then figure out what to do to minlen and maxlen.
2082 	 */
2083 	if (prod > 1) {
2084 		xfs_extlen_t	i;
2085 
2086 		if ((i = maxlen % prod))
2087 			maxlen -= i;
2088 		if ((i = minlen % prod))
2089 			minlen += prod - i;
2090 		if (maxlen < minlen) {
2091 			*rtblock = NULLRTBLOCK;
2092 			return 0;
2093 		}
2094 	}
2095 
2096 	sumbp = NULL;
2097 	/*
2098 	 * Allocate by size, or near another block, or exactly at some block.
2099 	 */
2100 	switch (type) {
2101 	case XFS_ALLOCTYPE_ANY_AG:
2102 		error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
2103 				&sumbp,	&sb, prod, &r);
2104 		break;
2105 	case XFS_ALLOCTYPE_NEAR_BNO:
2106 		error = xfs_rtallocate_extent_near(mp, tp, bno, minlen, maxlen,
2107 				len, &sumbp, &sb, prod, &r);
2108 		break;
2109 	case XFS_ALLOCTYPE_THIS_BNO:
2110 		error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen,
2111 				len, &sumbp, &sb, prod, &r);
2112 		break;
2113 	default:
2114 		error = EIO;
2115 		ASSERT(0);
2116 	}
2117 	if (error)
2118 		return error;
2119 
2120 	/*
2121 	 * If it worked, update the superblock.
2122 	 */
2123 	if (r != NULLRTBLOCK) {
2124 		long	slen = (long)*len;
2125 
2126 		ASSERT(*len >= minlen && *len <= maxlen);
2127 		if (wasdel)
2128 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
2129 		else
2130 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
2131 	}
2132 	*rtblock = r;
2133 	return 0;
2134 }
2135 
2136 /*
2137  * Free an extent in the realtime subvolume.  Length is expressed in
2138  * realtime extents, as is the block number.
2139  */
2140 int					/* error */
2141 xfs_rtfree_extent(
2142 	xfs_trans_t	*tp,		/* transaction pointer */
2143 	xfs_rtblock_t	bno,		/* starting block number to free */
2144 	xfs_extlen_t	len)		/* length of extent freed */
2145 {
2146 	int		error;		/* error value */
2147 	xfs_mount_t	*mp;		/* file system mount structure */
2148 	xfs_fsblock_t	sb;		/* summary file block number */
2149 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2150 
2151 	mp = tp->t_mountp;
2152 	/*
2153 	 * Synchronize by locking the bitmap inode.
2154 	 */
2155 	xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2156 	xfs_trans_ijoin_ref(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2157 
2158 #if defined(__KERNEL__) && defined(DEBUG)
2159 	/*
2160 	 * Check to see that this whole range is currently allocated.
2161 	 */
2162 	{
2163 		int	stat;		/* result from checking range */
2164 
2165 		error = xfs_rtcheck_alloc_range(mp, tp, bno, len, &stat);
2166 		if (error) {
2167 			return error;
2168 		}
2169 		ASSERT(stat);
2170 	}
2171 #endif
2172 	sumbp = NULL;
2173 	/*
2174 	 * Free the range of realtime blocks.
2175 	 */
2176 	error = xfs_rtfree_range(mp, tp, bno, len, &sumbp, &sb);
2177 	if (error) {
2178 		return error;
2179 	}
2180 	/*
2181 	 * Mark more blocks free in the superblock.
2182 	 */
2183 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, (long)len);
2184 	/*
2185 	 * If we've now freed all the blocks, reset the file sequence
2186 	 * number to 0.
2187 	 */
2188 	if (tp->t_frextents_delta + mp->m_sb.sb_frextents ==
2189 	    mp->m_sb.sb_rextents) {
2190 		if (!(mp->m_rbmip->i_d.di_flags & XFS_DIFLAG_NEWRTBM))
2191 			mp->m_rbmip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2192 		*(__uint64_t *)&mp->m_rbmip->i_d.di_atime = 0;
2193 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
2194 	}
2195 	return 0;
2196 }
2197 
2198 /*
2199  * Initialize realtime fields in the mount structure.
2200  */
2201 int				/* error */
2202 xfs_rtmount_init(
2203 	xfs_mount_t	*mp)	/* file system mount structure */
2204 {
2205 	xfs_buf_t	*bp;	/* buffer for last block of subvolume */
2206 	xfs_daddr_t	d;	/* address of last block of subvolume */
2207 	xfs_sb_t	*sbp;	/* filesystem superblock copy in mount */
2208 
2209 	sbp = &mp->m_sb;
2210 	if (sbp->sb_rblocks == 0)
2211 		return 0;
2212 	if (mp->m_rtdev_targp == NULL) {
2213 		xfs_warn(mp,
2214 	"Filesystem has a realtime volume, use rtdev=device option");
2215 		return XFS_ERROR(ENODEV);
2216 	}
2217 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
2218 	mp->m_rsumsize =
2219 		(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
2220 		sbp->sb_rbmblocks;
2221 	mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
2222 	mp->m_rbmip = mp->m_rsumip = NULL;
2223 	/*
2224 	 * Check that the realtime section is an ok size.
2225 	 */
2226 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
2227 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
2228 		xfs_warn(mp, "realtime mount -- %llu != %llu",
2229 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
2230 			(unsigned long long) mp->m_sb.sb_rblocks);
2231 		return XFS_ERROR(EFBIG);
2232 	}
2233 	bp = xfs_buf_read_uncached(mp, mp->m_rtdev_targp,
2234 					d - XFS_FSB_TO_BB(mp, 1),
2235 					XFS_FSB_TO_B(mp, 1), 0);
2236 	if (!bp) {
2237 		xfs_warn(mp, "realtime device size check failed");
2238 		return EIO;
2239 	}
2240 	xfs_buf_relse(bp);
2241 	return 0;
2242 }
2243 
2244 /*
2245  * Get the bitmap and summary inodes into the mount structure
2246  * at mount time.
2247  */
2248 int					/* error */
2249 xfs_rtmount_inodes(
2250 	xfs_mount_t	*mp)		/* file system mount structure */
2251 {
2252 	int		error;		/* error return value */
2253 	xfs_sb_t	*sbp;
2254 
2255 	sbp = &mp->m_sb;
2256 	if (sbp->sb_rbmino == NULLFSINO)
2257 		return 0;
2258 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
2259 	if (error)
2260 		return error;
2261 	ASSERT(mp->m_rbmip != NULL);
2262 	ASSERT(sbp->sb_rsumino != NULLFSINO);
2263 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
2264 	if (error) {
2265 		IRELE(mp->m_rbmip);
2266 		return error;
2267 	}
2268 	ASSERT(mp->m_rsumip != NULL);
2269 	return 0;
2270 }
2271 
2272 void
2273 xfs_rtunmount_inodes(
2274 	struct xfs_mount	*mp)
2275 {
2276 	if (mp->m_rbmip)
2277 		IRELE(mp->m_rbmip);
2278 	if (mp->m_rsumip)
2279 		IRELE(mp->m_rsumip);
2280 }
2281 
2282 /*
2283  * Pick an extent for allocation at the start of a new realtime file.
2284  * Use the sequence number stored in the atime field of the bitmap inode.
2285  * Translate this to a fraction of the rtextents, and return the product
2286  * of rtextents and the fraction.
2287  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
2288  */
2289 int					/* error */
2290 xfs_rtpick_extent(
2291 	xfs_mount_t	*mp,		/* file system mount point */
2292 	xfs_trans_t	*tp,		/* transaction pointer */
2293 	xfs_extlen_t	len,		/* allocation length (rtextents) */
2294 	xfs_rtblock_t	*pick)		/* result rt extent */
2295 {
2296 	xfs_rtblock_t	b;		/* result block */
2297 	int		log2;		/* log of sequence number */
2298 	__uint64_t	resid;		/* residual after log removed */
2299 	__uint64_t	seq;		/* sequence number of file creation */
2300 	__uint64_t	*seqp;		/* pointer to seqno in inode */
2301 
2302 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2303 
2304 	seqp = (__uint64_t *)&mp->m_rbmip->i_d.di_atime;
2305 	if (!(mp->m_rbmip->i_d.di_flags & XFS_DIFLAG_NEWRTBM)) {
2306 		mp->m_rbmip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2307 		*seqp = 0;
2308 	}
2309 	seq = *seqp;
2310 	if ((log2 = xfs_highbit64(seq)) == -1)
2311 		b = 0;
2312 	else {
2313 		resid = seq - (1ULL << log2);
2314 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
2315 		    (log2 + 1);
2316 		if (b >= mp->m_sb.sb_rextents)
2317 			b = do_mod(b, mp->m_sb.sb_rextents);
2318 		if (b + len > mp->m_sb.sb_rextents)
2319 			b = mp->m_sb.sb_rextents - len;
2320 	}
2321 	*seqp = seq + 1;
2322 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
2323 	*pick = b;
2324 	return 0;
2325 }
2326