xref: /openbmc/linux/fs/xfs/xfs_rtalloc.c (revision 95e9fd10)
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_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_alloc.h"
32 #include "xfs_bmap.h"
33 #include "xfs_rtalloc.h"
34 #include "xfs_fsops.h"
35 #include "xfs_error.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_trans_space.h"
38 #include "xfs_utils.h"
39 #include "xfs_trace.h"
40 #include "xfs_buf.h"
41 
42 
43 /*
44  * Prototypes for internal functions.
45  */
46 
47 
48 STATIC int xfs_rtallocate_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
49 		xfs_extlen_t, xfs_buf_t **, xfs_fsblock_t *);
50 STATIC int xfs_rtany_summary(xfs_mount_t *, xfs_trans_t *, int, int,
51 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, int *);
52 STATIC int xfs_rtcheck_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
53 		xfs_extlen_t, int, xfs_rtblock_t *, int *);
54 STATIC int xfs_rtfind_back(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
55 		xfs_rtblock_t, xfs_rtblock_t *);
56 STATIC int xfs_rtfind_forw(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
57 		xfs_rtblock_t, xfs_rtblock_t *);
58 STATIC int xfs_rtget_summary( xfs_mount_t *, xfs_trans_t *, int,
59 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, xfs_suminfo_t *);
60 STATIC int xfs_rtmodify_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
61 		xfs_extlen_t, int);
62 STATIC int xfs_rtmodify_summary(xfs_mount_t *, xfs_trans_t *, int,
63 		xfs_rtblock_t, int, xfs_buf_t **, xfs_fsblock_t *);
64 
65 /*
66  * Internal functions.
67  */
68 
69 /*
70  * Allocate space to the bitmap or summary file, and zero it, for growfs.
71  */
72 STATIC int				/* error */
73 xfs_growfs_rt_alloc(
74 	xfs_mount_t	*mp,		/* file system mount point */
75 	xfs_extlen_t	oblocks,	/* old count of blocks */
76 	xfs_extlen_t	nblocks,	/* new count of blocks */
77 	xfs_inode_t	*ip)		/* inode (bitmap/summary) */
78 {
79 	xfs_fileoff_t	bno;		/* block number in file */
80 	xfs_buf_t	*bp;		/* temporary buffer for zeroing */
81 	int		committed;	/* transaction committed flag */
82 	xfs_daddr_t	d;		/* disk block address */
83 	int		error;		/* error return value */
84 	xfs_fsblock_t	firstblock;	/* first block allocated in xaction */
85 	xfs_bmap_free_t	flist;		/* list of freed blocks */
86 	xfs_fsblock_t	fsbno;		/* filesystem block for bno */
87 	xfs_bmbt_irec_t	map;		/* block map output */
88 	int		nmap;		/* number of block maps */
89 	int		resblks;	/* space reservation */
90 
91 	/*
92 	 * Allocate space to the file, as necessary.
93 	 */
94 	while (oblocks < nblocks) {
95 		int		cancelflags = 0;
96 		xfs_trans_t	*tp;
97 
98 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ALLOC);
99 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
100 		/*
101 		 * Reserve space & log for one extent added to the file.
102 		 */
103 		if ((error = xfs_trans_reserve(tp, resblks,
104 				XFS_GROWRTALLOC_LOG_RES(mp), 0,
105 				XFS_TRANS_PERM_LOG_RES,
106 				XFS_DEFAULT_PERM_LOG_COUNT)))
107 			goto error_cancel;
108 		cancelflags = XFS_TRANS_RELEASE_LOG_RES;
109 		/*
110 		 * Lock the inode.
111 		 */
112 		xfs_ilock(ip, XFS_ILOCK_EXCL);
113 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
114 
115 		xfs_bmap_init(&flist, &firstblock);
116 		/*
117 		 * Allocate blocks to the bitmap file.
118 		 */
119 		nmap = 1;
120 		cancelflags |= XFS_TRANS_ABORT;
121 		error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
122 					XFS_BMAPI_METADATA, &firstblock,
123 					resblks, &map, &nmap, &flist);
124 		if (!error && nmap < 1)
125 			error = XFS_ERROR(ENOSPC);
126 		if (error)
127 			goto error_cancel;
128 		/*
129 		 * Free any blocks freed up in the transaction, then commit.
130 		 */
131 		error = xfs_bmap_finish(&tp, &flist, &committed);
132 		if (error)
133 			goto error_cancel;
134 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
135 		if (error)
136 			goto error;
137 		/*
138 		 * Now we need to clear the allocated blocks.
139 		 * Do this one block per transaction, to keep it simple.
140 		 */
141 		cancelflags = 0;
142 		for (bno = map.br_startoff, fsbno = map.br_startblock;
143 		     bno < map.br_startoff + map.br_blockcount;
144 		     bno++, fsbno++) {
145 			tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ZERO);
146 			/*
147 			 * Reserve log for one block zeroing.
148 			 */
149 			if ((error = xfs_trans_reserve(tp, 0,
150 					XFS_GROWRTZERO_LOG_RES(mp), 0, 0, 0)))
151 				goto error_cancel;
152 			/*
153 			 * Lock the bitmap inode.
154 			 */
155 			xfs_ilock(ip, XFS_ILOCK_EXCL);
156 			xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
157 			/*
158 			 * Get a buffer for the block.
159 			 */
160 			d = XFS_FSB_TO_DADDR(mp, fsbno);
161 			bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
162 				mp->m_bsize, 0);
163 			if (bp == NULL) {
164 				error = XFS_ERROR(EIO);
165 error_cancel:
166 				xfs_trans_cancel(tp, cancelflags);
167 				goto error;
168 			}
169 			memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
170 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
171 			/*
172 			 * Commit the transaction.
173 			 */
174 			error = xfs_trans_commit(tp, 0);
175 			if (error)
176 				goto error;
177 		}
178 		/*
179 		 * Go on to the next extent, if any.
180 		 */
181 		oblocks = map.br_startoff + map.br_blockcount;
182 	}
183 	return 0;
184 
185 error:
186 	return error;
187 }
188 
189 /*
190  * Attempt to allocate an extent minlen<=len<=maxlen starting from
191  * bitmap block bbno.  If we don't get maxlen then use prod to trim
192  * the length, if given.  Returns error; returns starting block in *rtblock.
193  * The lengths are all in rtextents.
194  */
195 STATIC int				/* error */
196 xfs_rtallocate_extent_block(
197 	xfs_mount_t	*mp,		/* file system mount point */
198 	xfs_trans_t	*tp,		/* transaction pointer */
199 	xfs_rtblock_t	bbno,		/* bitmap block number */
200 	xfs_extlen_t	minlen,		/* minimum length to allocate */
201 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
202 	xfs_extlen_t	*len,		/* out: actual length allocated */
203 	xfs_rtblock_t	*nextp,		/* out: next block to try */
204 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
205 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
206 	xfs_extlen_t	prod,		/* extent product factor */
207 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
208 {
209 	xfs_rtblock_t	besti;		/* best rtblock found so far */
210 	xfs_rtblock_t	bestlen;	/* best length found so far */
211 	xfs_rtblock_t	end;		/* last rtblock in chunk */
212 	int		error;		/* error value */
213 	xfs_rtblock_t	i;		/* current rtblock trying */
214 	xfs_rtblock_t	next;		/* next rtblock to try */
215 	int		stat;		/* status from internal calls */
216 
217 	/*
218 	 * Loop over all the extents starting in this bitmap block,
219 	 * looking for one that's long enough.
220 	 */
221 	for (i = XFS_BLOCKTOBIT(mp, bbno), besti = -1, bestlen = 0,
222 		end = XFS_BLOCKTOBIT(mp, bbno + 1) - 1;
223 	     i <= end;
224 	     i++) {
225 		/*
226 		 * See if there's a free extent of maxlen starting at i.
227 		 * If it's not so then next will contain the first non-free.
228 		 */
229 		error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
230 		if (error) {
231 			return error;
232 		}
233 		if (stat) {
234 			/*
235 			 * i for maxlen is all free, allocate and return that.
236 			 */
237 			error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
238 				rsb);
239 			if (error) {
240 				return error;
241 			}
242 			*len = maxlen;
243 			*rtblock = i;
244 			return 0;
245 		}
246 		/*
247 		 * In the case where we have a variable-sized allocation
248 		 * request, figure out how big this free piece is,
249 		 * and if it's big enough for the minimum, and the best
250 		 * so far, remember it.
251 		 */
252 		if (minlen < maxlen) {
253 			xfs_rtblock_t	thislen;	/* this extent size */
254 
255 			thislen = next - i;
256 			if (thislen >= minlen && thislen > bestlen) {
257 				besti = i;
258 				bestlen = thislen;
259 			}
260 		}
261 		/*
262 		 * If not done yet, find the start of the next free space.
263 		 */
264 		if (next < end) {
265 			error = xfs_rtfind_forw(mp, tp, next, end, &i);
266 			if (error) {
267 				return error;
268 			}
269 		} else
270 			break;
271 	}
272 	/*
273 	 * Searched the whole thing & didn't find a maxlen free extent.
274 	 */
275 	if (minlen < maxlen && besti != -1) {
276 		xfs_extlen_t	p;	/* amount to trim length by */
277 
278 		/*
279 		 * If size should be a multiple of prod, make that so.
280 		 */
281 		if (prod > 1 && (p = do_mod(bestlen, prod)))
282 			bestlen -= p;
283 		/*
284 		 * Allocate besti for bestlen & return that.
285 		 */
286 		error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
287 		if (error) {
288 			return error;
289 		}
290 		*len = bestlen;
291 		*rtblock = besti;
292 		return 0;
293 	}
294 	/*
295 	 * Allocation failed.  Set *nextp to the next block to try.
296 	 */
297 	*nextp = next;
298 	*rtblock = NULLRTBLOCK;
299 	return 0;
300 }
301 
302 /*
303  * Allocate an extent of length minlen<=len<=maxlen, starting at block
304  * bno.  If we don't get maxlen then use prod to trim the length, if given.
305  * Returns error; returns starting block in *rtblock.
306  * The lengths are all in rtextents.
307  */
308 STATIC int				/* error */
309 xfs_rtallocate_extent_exact(
310 	xfs_mount_t	*mp,		/* file system mount point */
311 	xfs_trans_t	*tp,		/* transaction pointer */
312 	xfs_rtblock_t	bno,		/* starting block number to allocate */
313 	xfs_extlen_t	minlen,		/* minimum length to allocate */
314 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
315 	xfs_extlen_t	*len,		/* out: actual length allocated */
316 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
317 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
318 	xfs_extlen_t	prod,		/* extent product factor */
319 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
320 {
321 	int		error;		/* error value */
322 	xfs_extlen_t	i;		/* extent length trimmed due to prod */
323 	int		isfree;		/* extent is free */
324 	xfs_rtblock_t	next;		/* next block to try (dummy) */
325 
326 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
327 	/*
328 	 * Check if the range in question (for maxlen) is free.
329 	 */
330 	error = xfs_rtcheck_range(mp, tp, bno, maxlen, 1, &next, &isfree);
331 	if (error) {
332 		return error;
333 	}
334 	if (isfree) {
335 		/*
336 		 * If it is, allocate it and return success.
337 		 */
338 		error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
339 		if (error) {
340 			return error;
341 		}
342 		*len = maxlen;
343 		*rtblock = bno;
344 		return 0;
345 	}
346 	/*
347 	 * If not, allocate what there is, if it's at least minlen.
348 	 */
349 	maxlen = next - bno;
350 	if (maxlen < minlen) {
351 		/*
352 		 * Failed, return failure status.
353 		 */
354 		*rtblock = NULLRTBLOCK;
355 		return 0;
356 	}
357 	/*
358 	 * Trim off tail of extent, if prod is specified.
359 	 */
360 	if (prod > 1 && (i = maxlen % prod)) {
361 		maxlen -= i;
362 		if (maxlen < minlen) {
363 			/*
364 			 * Now we can't do it, return failure status.
365 			 */
366 			*rtblock = NULLRTBLOCK;
367 			return 0;
368 		}
369 	}
370 	/*
371 	 * Allocate what we can and return it.
372 	 */
373 	error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
374 	if (error) {
375 		return error;
376 	}
377 	*len = maxlen;
378 	*rtblock = bno;
379 	return 0;
380 }
381 
382 /*
383  * Allocate an extent of length minlen<=len<=maxlen, starting as near
384  * to bno as possible.  If we don't get maxlen then use prod to trim
385  * the length, if given.  The lengths are all in rtextents.
386  */
387 STATIC int				/* error */
388 xfs_rtallocate_extent_near(
389 	xfs_mount_t	*mp,		/* file system mount point */
390 	xfs_trans_t	*tp,		/* transaction pointer */
391 	xfs_rtblock_t	bno,		/* starting block number to allocate */
392 	xfs_extlen_t	minlen,		/* minimum length to allocate */
393 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
394 	xfs_extlen_t	*len,		/* out: actual length allocated */
395 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
396 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
397 	xfs_extlen_t	prod,		/* extent product factor */
398 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
399 {
400 	int		any;		/* any useful extents from summary */
401 	xfs_rtblock_t	bbno;		/* bitmap block number */
402 	int		error;		/* error value */
403 	int		i;		/* bitmap block offset (loop control) */
404 	int		j;		/* secondary loop control */
405 	int		log2len;	/* log2 of minlen */
406 	xfs_rtblock_t	n;		/* next block to try */
407 	xfs_rtblock_t	r;		/* result block */
408 
409 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
410 	/*
411 	 * If the block number given is off the end, silently set it to
412 	 * the last block.
413 	 */
414 	if (bno >= mp->m_sb.sb_rextents)
415 		bno = mp->m_sb.sb_rextents - 1;
416 	/*
417 	 * Try the exact allocation first.
418 	 */
419 	error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen, len,
420 		rbpp, rsb, prod, &r);
421 	if (error) {
422 		return error;
423 	}
424 	/*
425 	 * If the exact allocation worked, return that.
426 	 */
427 	if (r != NULLRTBLOCK) {
428 		*rtblock = r;
429 		return 0;
430 	}
431 	bbno = XFS_BITTOBLOCK(mp, bno);
432 	i = 0;
433 	ASSERT(minlen != 0);
434 	log2len = xfs_highbit32(minlen);
435 	/*
436 	 * Loop over all bitmap blocks (bbno + i is current block).
437 	 */
438 	for (;;) {
439 		/*
440 		 * Get summary information of extents of all useful levels
441 		 * starting in this bitmap block.
442 		 */
443 		error = xfs_rtany_summary(mp, tp, log2len, mp->m_rsumlevels - 1,
444 			bbno + i, rbpp, rsb, &any);
445 		if (error) {
446 			return error;
447 		}
448 		/*
449 		 * If there are any useful extents starting here, try
450 		 * allocating one.
451 		 */
452 		if (any) {
453 			/*
454 			 * On the positive side of the starting location.
455 			 */
456 			if (i >= 0) {
457 				/*
458 				 * Try to allocate an extent starting in
459 				 * this block.
460 				 */
461 				error = xfs_rtallocate_extent_block(mp, tp,
462 					bbno + i, minlen, maxlen, len, &n, rbpp,
463 					rsb, prod, &r);
464 				if (error) {
465 					return error;
466 				}
467 				/*
468 				 * If it worked, return it.
469 				 */
470 				if (r != NULLRTBLOCK) {
471 					*rtblock = r;
472 					return 0;
473 				}
474 			}
475 			/*
476 			 * On the negative side of the starting location.
477 			 */
478 			else {		/* i < 0 */
479 				/*
480 				 * Loop backwards through the bitmap blocks from
481 				 * the starting point-1 up to where we are now.
482 				 * There should be an extent which ends in this
483 				 * bitmap block and is long enough.
484 				 */
485 				for (j = -1; j > i; j--) {
486 					/*
487 					 * Grab the summary information for
488 					 * this bitmap block.
489 					 */
490 					error = xfs_rtany_summary(mp, tp,
491 						log2len, mp->m_rsumlevels - 1,
492 						bbno + j, rbpp, rsb, &any);
493 					if (error) {
494 						return error;
495 					}
496 					/*
497 					 * If there's no extent given in the
498 					 * summary that means the extent we
499 					 * found must carry over from an
500 					 * earlier block.  If there is an
501 					 * extent given, we've already tried
502 					 * that allocation, don't do it again.
503 					 */
504 					if (any)
505 						continue;
506 					error = xfs_rtallocate_extent_block(mp,
507 						tp, bbno + j, minlen, maxlen,
508 						len, &n, rbpp, rsb, prod, &r);
509 					if (error) {
510 						return error;
511 					}
512 					/*
513 					 * If it works, return the extent.
514 					 */
515 					if (r != NULLRTBLOCK) {
516 						*rtblock = r;
517 						return 0;
518 					}
519 				}
520 				/*
521 				 * There weren't intervening bitmap blocks
522 				 * with a long enough extent, or the
523 				 * allocation didn't work for some reason
524 				 * (i.e. it's a little * too short).
525 				 * Try to allocate from the summary block
526 				 * that we found.
527 				 */
528 				error = xfs_rtallocate_extent_block(mp, tp,
529 					bbno + i, minlen, maxlen, len, &n, rbpp,
530 					rsb, prod, &r);
531 				if (error) {
532 					return error;
533 				}
534 				/*
535 				 * If it works, return the extent.
536 				 */
537 				if (r != NULLRTBLOCK) {
538 					*rtblock = r;
539 					return 0;
540 				}
541 			}
542 		}
543 		/*
544 		 * Loop control.  If we were on the positive side, and there's
545 		 * still more blocks on the negative side, go there.
546 		 */
547 		if (i > 0 && (int)bbno - i >= 0)
548 			i = -i;
549 		/*
550 		 * If positive, and no more negative, but there are more
551 		 * positive, go there.
552 		 */
553 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
554 			i++;
555 		/*
556 		 * If negative or 0 (just started), and there are positive
557 		 * blocks to go, go there.  The 0 case moves to block 1.
558 		 */
559 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
560 			i = 1 - i;
561 		/*
562 		 * If negative or 0 and there are more negative blocks,
563 		 * go there.
564 		 */
565 		else if (i <= 0 && (int)bbno + i > 0)
566 			i--;
567 		/*
568 		 * Must be done.  Return failure.
569 		 */
570 		else
571 			break;
572 	}
573 	*rtblock = NULLRTBLOCK;
574 	return 0;
575 }
576 
577 /*
578  * Allocate an extent of length minlen<=len<=maxlen, with no position
579  * specified.  If we don't get maxlen then use prod to trim
580  * the length, if given.  The lengths are all in rtextents.
581  */
582 STATIC int				/* error */
583 xfs_rtallocate_extent_size(
584 	xfs_mount_t	*mp,		/* file system mount point */
585 	xfs_trans_t	*tp,		/* transaction pointer */
586 	xfs_extlen_t	minlen,		/* minimum length to allocate */
587 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
588 	xfs_extlen_t	*len,		/* out: actual length allocated */
589 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
590 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
591 	xfs_extlen_t	prod,		/* extent product factor */
592 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
593 {
594 	int		error;		/* error value */
595 	int		i;		/* bitmap block number */
596 	int		l;		/* level number (loop control) */
597 	xfs_rtblock_t	n;		/* next block to be tried */
598 	xfs_rtblock_t	r;		/* result block number */
599 	xfs_suminfo_t	sum;		/* summary information for extents */
600 
601 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
602 	ASSERT(maxlen != 0);
603 
604 	/*
605 	 * Loop over all the levels starting with maxlen.
606 	 * At each level, look at all the bitmap blocks, to see if there
607 	 * are extents starting there that are long enough (>= maxlen).
608 	 * Note, only on the initial level can the allocation fail if
609 	 * the summary says there's an extent.
610 	 */
611 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
612 		/*
613 		 * Loop over all the bitmap blocks.
614 		 */
615 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
616 			/*
617 			 * Get the summary for this level/block.
618 			 */
619 			error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
620 				&sum);
621 			if (error) {
622 				return error;
623 			}
624 			/*
625 			 * Nothing there, on to the next block.
626 			 */
627 			if (!sum)
628 				continue;
629 			/*
630 			 * Try allocating the extent.
631 			 */
632 			error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
633 				maxlen, len, &n, rbpp, rsb, prod, &r);
634 			if (error) {
635 				return error;
636 			}
637 			/*
638 			 * If it worked, return that.
639 			 */
640 			if (r != NULLRTBLOCK) {
641 				*rtblock = r;
642 				return 0;
643 			}
644 			/*
645 			 * If the "next block to try" returned from the
646 			 * allocator is beyond the next bitmap block,
647 			 * skip to that bitmap block.
648 			 */
649 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
650 				i = XFS_BITTOBLOCK(mp, n) - 1;
651 		}
652 	}
653 	/*
654 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
655 	 * we're asking for a fixed size extent.
656 	 */
657 	if (minlen > --maxlen) {
658 		*rtblock = NULLRTBLOCK;
659 		return 0;
660 	}
661 	ASSERT(minlen != 0);
662 	ASSERT(maxlen != 0);
663 
664 	/*
665 	 * Loop over sizes, from maxlen down to minlen.
666 	 * This time, when we do the allocations, allow smaller ones
667 	 * to succeed.
668 	 */
669 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
670 		/*
671 		 * Loop over all the bitmap blocks, try an allocation
672 		 * starting in that block.
673 		 */
674 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
675 			/*
676 			 * Get the summary information for this level/block.
677 			 */
678 			error =	xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
679 						  &sum);
680 			if (error) {
681 				return error;
682 			}
683 			/*
684 			 * If nothing there, go on to next.
685 			 */
686 			if (!sum)
687 				continue;
688 			/*
689 			 * Try the allocation.  Make sure the specified
690 			 * minlen/maxlen are in the possible range for
691 			 * this summary level.
692 			 */
693 			error = xfs_rtallocate_extent_block(mp, tp, i,
694 					XFS_RTMAX(minlen, 1 << l),
695 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
696 					len, &n, rbpp, rsb, prod, &r);
697 			if (error) {
698 				return error;
699 			}
700 			/*
701 			 * If it worked, return that extent.
702 			 */
703 			if (r != NULLRTBLOCK) {
704 				*rtblock = r;
705 				return 0;
706 			}
707 			/*
708 			 * If the "next block to try" returned from the
709 			 * allocator is beyond the next bitmap block,
710 			 * skip to that bitmap block.
711 			 */
712 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
713 				i = XFS_BITTOBLOCK(mp, n) - 1;
714 		}
715 	}
716 	/*
717 	 * Got nothing, return failure.
718 	 */
719 	*rtblock = NULLRTBLOCK;
720 	return 0;
721 }
722 
723 /*
724  * Mark an extent specified by start and len allocated.
725  * Updates all the summary information as well as the bitmap.
726  */
727 STATIC int				/* error */
728 xfs_rtallocate_range(
729 	xfs_mount_t	*mp,		/* file system mount point */
730 	xfs_trans_t	*tp,		/* transaction pointer */
731 	xfs_rtblock_t	start,		/* start block to allocate */
732 	xfs_extlen_t	len,		/* length to allocate */
733 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
734 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
735 {
736 	xfs_rtblock_t	end;		/* end of the allocated extent */
737 	int		error;		/* error value */
738 	xfs_rtblock_t	postblock;	/* first block allocated > end */
739 	xfs_rtblock_t	preblock;	/* first block allocated < start */
740 
741 	end = start + len - 1;
742 	/*
743 	 * Assume we're allocating out of the middle of a free extent.
744 	 * We need to find the beginning and end of the extent so we can
745 	 * properly update the summary.
746 	 */
747 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
748 	if (error) {
749 		return error;
750 	}
751 	/*
752 	 * Find the next allocated block (end of free extent).
753 	 */
754 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
755 		&postblock);
756 	if (error) {
757 		return error;
758 	}
759 	/*
760 	 * Decrement the summary information corresponding to the entire
761 	 * (old) free extent.
762 	 */
763 	error = xfs_rtmodify_summary(mp, tp,
764 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
765 		XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
766 	if (error) {
767 		return error;
768 	}
769 	/*
770 	 * If there are blocks not being allocated at the front of the
771 	 * old extent, add summary data for them to be free.
772 	 */
773 	if (preblock < start) {
774 		error = xfs_rtmodify_summary(mp, tp,
775 			XFS_RTBLOCKLOG(start - preblock),
776 			XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
777 		if (error) {
778 			return error;
779 		}
780 	}
781 	/*
782 	 * If there are blocks not being allocated at the end of the
783 	 * old extent, add summary data for them to be free.
784 	 */
785 	if (postblock > end) {
786 		error = xfs_rtmodify_summary(mp, tp,
787 			XFS_RTBLOCKLOG(postblock - end),
788 			XFS_BITTOBLOCK(mp, end + 1), 1, rbpp, rsb);
789 		if (error) {
790 			return error;
791 		}
792 	}
793 	/*
794 	 * Modify the bitmap to mark this extent allocated.
795 	 */
796 	error = xfs_rtmodify_range(mp, tp, start, len, 0);
797 	return error;
798 }
799 
800 /*
801  * Return whether there are any free extents in the size range given
802  * by low and high, for the bitmap block bbno.
803  */
804 STATIC int				/* error */
805 xfs_rtany_summary(
806 	xfs_mount_t	*mp,		/* file system mount structure */
807 	xfs_trans_t	*tp,		/* transaction pointer */
808 	int		low,		/* low log2 extent size */
809 	int		high,		/* high log2 extent size */
810 	xfs_rtblock_t	bbno,		/* bitmap block number */
811 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
812 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
813 	int		*stat)		/* out: any good extents here? */
814 {
815 	int		error;		/* error value */
816 	int		log;		/* loop counter, log2 of ext. size */
817 	xfs_suminfo_t	sum;		/* summary data */
818 
819 	/*
820 	 * Loop over logs of extent sizes.  Order is irrelevant.
821 	 */
822 	for (log = low; log <= high; log++) {
823 		/*
824 		 * Get one summary datum.
825 		 */
826 		error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
827 		if (error) {
828 			return error;
829 		}
830 		/*
831 		 * If there are any, return success.
832 		 */
833 		if (sum) {
834 			*stat = 1;
835 			return 0;
836 		}
837 	}
838 	/*
839 	 * Found nothing, return failure.
840 	 */
841 	*stat = 0;
842 	return 0;
843 }
844 
845 /*
846  * Get a buffer for the bitmap or summary file block specified.
847  * The buffer is returned read and locked.
848  */
849 STATIC int				/* error */
850 xfs_rtbuf_get(
851 	xfs_mount_t	*mp,		/* file system mount structure */
852 	xfs_trans_t	*tp,		/* transaction pointer */
853 	xfs_rtblock_t	block,		/* block number in bitmap or summary */
854 	int		issum,		/* is summary not bitmap */
855 	xfs_buf_t	**bpp)		/* output: buffer for the block */
856 {
857 	xfs_buf_t	*bp;		/* block buffer, result */
858 	xfs_inode_t	*ip;		/* bitmap or summary inode */
859 	xfs_bmbt_irec_t	map;
860 	int		nmap = 1;
861 	int		error;		/* error value */
862 
863 	ip = issum ? mp->m_rsumip : mp->m_rbmip;
864 
865 	error = xfs_bmapi_read(ip, block, 1, &map, &nmap, XFS_DATA_FORK);
866 	if (error)
867 		return error;
868 
869 	ASSERT(map.br_startblock != NULLFSBLOCK);
870 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
871 				   XFS_FSB_TO_DADDR(mp, map.br_startblock),
872 				   mp->m_bsize, 0, &bp);
873 	if (error)
874 		return error;
875 	ASSERT(!xfs_buf_geterror(bp));
876 	*bpp = bp;
877 	return 0;
878 }
879 
880 #ifdef DEBUG
881 /*
882  * Check that the given extent (block range) is allocated already.
883  */
884 STATIC int				/* error */
885 xfs_rtcheck_alloc_range(
886 	xfs_mount_t	*mp,		/* file system mount point */
887 	xfs_trans_t	*tp,		/* transaction pointer */
888 	xfs_rtblock_t	bno,		/* starting block number of extent */
889 	xfs_extlen_t	len,		/* length of extent */
890 	int		*stat)		/* out: 1 for allocated, 0 for not */
891 {
892 	xfs_rtblock_t	new;		/* dummy for xfs_rtcheck_range */
893 
894 	return xfs_rtcheck_range(mp, tp, bno, len, 0, &new, stat);
895 }
896 #endif
897 
898 /*
899  * Check that the given range is either all allocated (val = 0) or
900  * all free (val = 1).
901  */
902 STATIC int				/* error */
903 xfs_rtcheck_range(
904 	xfs_mount_t	*mp,		/* file system mount point */
905 	xfs_trans_t	*tp,		/* transaction pointer */
906 	xfs_rtblock_t	start,		/* starting block number of extent */
907 	xfs_extlen_t	len,		/* length of extent */
908 	int		val,		/* 1 for free, 0 for allocated */
909 	xfs_rtblock_t	*new,		/* out: first block not matching */
910 	int		*stat)		/* out: 1 for matches, 0 for not */
911 {
912 	xfs_rtword_t	*b;		/* current word in buffer */
913 	int		bit;		/* bit number in the word */
914 	xfs_rtblock_t	block;		/* bitmap block number */
915 	xfs_buf_t	*bp;		/* buf for the block */
916 	xfs_rtword_t	*bufp;		/* starting word in buffer */
917 	int		error;		/* error value */
918 	xfs_rtblock_t	i;		/* current bit number rel. to start */
919 	xfs_rtblock_t	lastbit;	/* last useful bit in word */
920 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
921 	xfs_rtword_t	wdiff;		/* difference from wanted value */
922 	int		word;		/* word number in the buffer */
923 
924 	/*
925 	 * Compute starting bitmap block number
926 	 */
927 	block = XFS_BITTOBLOCK(mp, start);
928 	/*
929 	 * Read the bitmap block.
930 	 */
931 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
932 	if (error) {
933 		return error;
934 	}
935 	bufp = bp->b_addr;
936 	/*
937 	 * Compute the starting word's address, and starting bit.
938 	 */
939 	word = XFS_BITTOWORD(mp, start);
940 	b = &bufp[word];
941 	bit = (int)(start & (XFS_NBWORD - 1));
942 	/*
943 	 * 0 (allocated) => all zero's; 1 (free) => all one's.
944 	 */
945 	val = -val;
946 	/*
947 	 * If not starting on a word boundary, deal with the first
948 	 * (partial) word.
949 	 */
950 	if (bit) {
951 		/*
952 		 * Compute first bit not examined.
953 		 */
954 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
955 		/*
956 		 * Mask of relevant bits.
957 		 */
958 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
959 		/*
960 		 * Compute difference between actual and desired value.
961 		 */
962 		if ((wdiff = (*b ^ val) & mask)) {
963 			/*
964 			 * Different, compute first wrong bit and return.
965 			 */
966 			xfs_trans_brelse(tp, bp);
967 			i = XFS_RTLOBIT(wdiff) - bit;
968 			*new = start + i;
969 			*stat = 0;
970 			return 0;
971 		}
972 		i = lastbit - bit;
973 		/*
974 		 * Go on to next block if that's where the next word is
975 		 * and we need the next word.
976 		 */
977 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
978 			/*
979 			 * If done with this block, get the next one.
980 			 */
981 			xfs_trans_brelse(tp, bp);
982 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
983 			if (error) {
984 				return error;
985 			}
986 			b = bufp = bp->b_addr;
987 			word = 0;
988 		} else {
989 			/*
990 			 * Go on to the next word in the buffer.
991 			 */
992 			b++;
993 		}
994 	} else {
995 		/*
996 		 * Starting on a word boundary, no partial word.
997 		 */
998 		i = 0;
999 	}
1000 	/*
1001 	 * Loop over whole words in buffers.  When we use up one buffer
1002 	 * we move on to the next one.
1003 	 */
1004 	while (len - i >= XFS_NBWORD) {
1005 		/*
1006 		 * Compute difference between actual and desired value.
1007 		 */
1008 		if ((wdiff = *b ^ val)) {
1009 			/*
1010 			 * Different, compute first wrong bit and return.
1011 			 */
1012 			xfs_trans_brelse(tp, bp);
1013 			i += XFS_RTLOBIT(wdiff);
1014 			*new = start + i;
1015 			*stat = 0;
1016 			return 0;
1017 		}
1018 		i += XFS_NBWORD;
1019 		/*
1020 		 * Go on to next block if that's where the next word is
1021 		 * and we need the next word.
1022 		 */
1023 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1024 			/*
1025 			 * If done with this block, get the next one.
1026 			 */
1027 			xfs_trans_brelse(tp, bp);
1028 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1029 			if (error) {
1030 				return error;
1031 			}
1032 			b = bufp = bp->b_addr;
1033 			word = 0;
1034 		} else {
1035 			/*
1036 			 * Go on to the next word in the buffer.
1037 			 */
1038 			b++;
1039 		}
1040 	}
1041 	/*
1042 	 * If not ending on a word boundary, deal with the last
1043 	 * (partial) word.
1044 	 */
1045 	if ((lastbit = len - i)) {
1046 		/*
1047 		 * Mask of relevant bits.
1048 		 */
1049 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1050 		/*
1051 		 * Compute difference between actual and desired value.
1052 		 */
1053 		if ((wdiff = (*b ^ val) & mask)) {
1054 			/*
1055 			 * Different, compute first wrong bit and return.
1056 			 */
1057 			xfs_trans_brelse(tp, bp);
1058 			i += XFS_RTLOBIT(wdiff);
1059 			*new = start + i;
1060 			*stat = 0;
1061 			return 0;
1062 		} else
1063 			i = len;
1064 	}
1065 	/*
1066 	 * Successful, return.
1067 	 */
1068 	xfs_trans_brelse(tp, bp);
1069 	*new = start + i;
1070 	*stat = 1;
1071 	return 0;
1072 }
1073 
1074 /*
1075  * Copy and transform the summary file, given the old and new
1076  * parameters in the mount structures.
1077  */
1078 STATIC int				/* error */
1079 xfs_rtcopy_summary(
1080 	xfs_mount_t	*omp,		/* old file system mount point */
1081 	xfs_mount_t	*nmp,		/* new file system mount point */
1082 	xfs_trans_t	*tp)		/* transaction pointer */
1083 {
1084 	xfs_rtblock_t	bbno;		/* bitmap block number */
1085 	xfs_buf_t	*bp;		/* summary buffer */
1086 	int		error;		/* error return value */
1087 	int		log;		/* summary level number (log length) */
1088 	xfs_suminfo_t	sum;		/* summary data */
1089 	xfs_fsblock_t	sumbno;		/* summary block number */
1090 
1091 	bp = NULL;
1092 	for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
1093 		for (bbno = omp->m_sb.sb_rbmblocks - 1;
1094 		     (xfs_srtblock_t)bbno >= 0;
1095 		     bbno--) {
1096 			error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
1097 				&sumbno, &sum);
1098 			if (error)
1099 				return error;
1100 			if (sum == 0)
1101 				continue;
1102 			error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
1103 				&bp, &sumbno);
1104 			if (error)
1105 				return error;
1106 			error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
1107 				&bp, &sumbno);
1108 			if (error)
1109 				return error;
1110 			ASSERT(sum > 0);
1111 		}
1112 	}
1113 	return 0;
1114 }
1115 
1116 /*
1117  * Searching backward from start to limit, find the first block whose
1118  * allocated/free state is different from start's.
1119  */
1120 STATIC int				/* error */
1121 xfs_rtfind_back(
1122 	xfs_mount_t	*mp,		/* file system mount point */
1123 	xfs_trans_t	*tp,		/* transaction pointer */
1124 	xfs_rtblock_t	start,		/* starting block to look at */
1125 	xfs_rtblock_t	limit,		/* last block to look at */
1126 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1127 {
1128 	xfs_rtword_t	*b;		/* current word in buffer */
1129 	int		bit;		/* bit number in the word */
1130 	xfs_rtblock_t	block;		/* bitmap block number */
1131 	xfs_buf_t	*bp;		/* buf for the block */
1132 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1133 	int		error;		/* error value */
1134 	xfs_rtblock_t	firstbit;	/* first useful bit in the word */
1135 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1136 	xfs_rtblock_t	len;		/* length of inspected area */
1137 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1138 	xfs_rtword_t	want;		/* mask for "good" values */
1139 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1140 	int		word;		/* word number in the buffer */
1141 
1142 	/*
1143 	 * Compute and read in starting bitmap block for starting block.
1144 	 */
1145 	block = XFS_BITTOBLOCK(mp, start);
1146 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1147 	if (error) {
1148 		return error;
1149 	}
1150 	bufp = bp->b_addr;
1151 	/*
1152 	 * Get the first word's index & point to it.
1153 	 */
1154 	word = XFS_BITTOWORD(mp, start);
1155 	b = &bufp[word];
1156 	bit = (int)(start & (XFS_NBWORD - 1));
1157 	len = start - limit + 1;
1158 	/*
1159 	 * Compute match value, based on the bit at start: if 1 (free)
1160 	 * then all-ones, else all-zeroes.
1161 	 */
1162 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1163 	/*
1164 	 * If the starting position is not word-aligned, deal with the
1165 	 * partial word.
1166 	 */
1167 	if (bit < XFS_NBWORD - 1) {
1168 		/*
1169 		 * Calculate first (leftmost) bit number to look at,
1170 		 * and mask for all the relevant bits in this word.
1171 		 */
1172 		firstbit = XFS_RTMAX((xfs_srtblock_t)(bit - len + 1), 0);
1173 		mask = (((xfs_rtword_t)1 << (bit - firstbit + 1)) - 1) <<
1174 			firstbit;
1175 		/*
1176 		 * Calculate the difference between the value there
1177 		 * and what we're looking for.
1178 		 */
1179 		if ((wdiff = (*b ^ want) & mask)) {
1180 			/*
1181 			 * Different.  Mark where we are and return.
1182 			 */
1183 			xfs_trans_brelse(tp, bp);
1184 			i = bit - XFS_RTHIBIT(wdiff);
1185 			*rtblock = start - i + 1;
1186 			return 0;
1187 		}
1188 		i = bit - firstbit + 1;
1189 		/*
1190 		 * Go on to previous block if that's where the previous word is
1191 		 * and we need the previous word.
1192 		 */
1193 		if (--word == -1 && i < len) {
1194 			/*
1195 			 * If done with this block, get the previous one.
1196 			 */
1197 			xfs_trans_brelse(tp, bp);
1198 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1199 			if (error) {
1200 				return error;
1201 			}
1202 			bufp = bp->b_addr;
1203 			word = XFS_BLOCKWMASK(mp);
1204 			b = &bufp[word];
1205 		} else {
1206 			/*
1207 			 * Go on to the previous word in the buffer.
1208 			 */
1209 			b--;
1210 		}
1211 	} else {
1212 		/*
1213 		 * Starting on a word boundary, no partial word.
1214 		 */
1215 		i = 0;
1216 	}
1217 	/*
1218 	 * Loop over whole words in buffers.  When we use up one buffer
1219 	 * we move on to the previous one.
1220 	 */
1221 	while (len - i >= XFS_NBWORD) {
1222 		/*
1223 		 * Compute difference between actual and desired value.
1224 		 */
1225 		if ((wdiff = *b ^ want)) {
1226 			/*
1227 			 * Different, mark where we are and return.
1228 			 */
1229 			xfs_trans_brelse(tp, bp);
1230 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1231 			*rtblock = start - i + 1;
1232 			return 0;
1233 		}
1234 		i += XFS_NBWORD;
1235 		/*
1236 		 * Go on to previous block if that's where the previous word is
1237 		 * and we need the previous word.
1238 		 */
1239 		if (--word == -1 && i < len) {
1240 			/*
1241 			 * If done with this block, get the previous one.
1242 			 */
1243 			xfs_trans_brelse(tp, bp);
1244 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1245 			if (error) {
1246 				return error;
1247 			}
1248 			bufp = bp->b_addr;
1249 			word = XFS_BLOCKWMASK(mp);
1250 			b = &bufp[word];
1251 		} else {
1252 			/*
1253 			 * Go on to the previous word in the buffer.
1254 			 */
1255 			b--;
1256 		}
1257 	}
1258 	/*
1259 	 * If not ending on a word boundary, deal with the last
1260 	 * (partial) word.
1261 	 */
1262 	if (len - i) {
1263 		/*
1264 		 * Calculate first (leftmost) bit number to look at,
1265 		 * and mask for all the relevant bits in this word.
1266 		 */
1267 		firstbit = XFS_NBWORD - (len - i);
1268 		mask = (((xfs_rtword_t)1 << (len - i)) - 1) << firstbit;
1269 		/*
1270 		 * Compute difference between actual and desired value.
1271 		 */
1272 		if ((wdiff = (*b ^ want) & mask)) {
1273 			/*
1274 			 * Different, mark where we are and return.
1275 			 */
1276 			xfs_trans_brelse(tp, bp);
1277 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1278 			*rtblock = start - i + 1;
1279 			return 0;
1280 		} else
1281 			i = len;
1282 	}
1283 	/*
1284 	 * No match, return that we scanned the whole area.
1285 	 */
1286 	xfs_trans_brelse(tp, bp);
1287 	*rtblock = start - i + 1;
1288 	return 0;
1289 }
1290 
1291 /*
1292  * Searching forward from start to limit, find the first block whose
1293  * allocated/free state is different from start's.
1294  */
1295 STATIC int				/* error */
1296 xfs_rtfind_forw(
1297 	xfs_mount_t	*mp,		/* file system mount point */
1298 	xfs_trans_t	*tp,		/* transaction pointer */
1299 	xfs_rtblock_t	start,		/* starting block to look at */
1300 	xfs_rtblock_t	limit,		/* last block to look at */
1301 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1302 {
1303 	xfs_rtword_t	*b;		/* current word in buffer */
1304 	int		bit;		/* bit number in the word */
1305 	xfs_rtblock_t	block;		/* bitmap block number */
1306 	xfs_buf_t	*bp;		/* buf for the block */
1307 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1308 	int		error;		/* error value */
1309 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1310 	xfs_rtblock_t	lastbit;	/* last useful bit in the word */
1311 	xfs_rtblock_t	len;		/* length of inspected area */
1312 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1313 	xfs_rtword_t	want;		/* mask for "good" values */
1314 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1315 	int		word;		/* word number in the buffer */
1316 
1317 	/*
1318 	 * Compute and read in starting bitmap block for starting block.
1319 	 */
1320 	block = XFS_BITTOBLOCK(mp, start);
1321 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1322 	if (error) {
1323 		return error;
1324 	}
1325 	bufp = bp->b_addr;
1326 	/*
1327 	 * Get the first word's index & point to it.
1328 	 */
1329 	word = XFS_BITTOWORD(mp, start);
1330 	b = &bufp[word];
1331 	bit = (int)(start & (XFS_NBWORD - 1));
1332 	len = limit - start + 1;
1333 	/*
1334 	 * Compute match value, based on the bit at start: if 1 (free)
1335 	 * then all-ones, else all-zeroes.
1336 	 */
1337 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1338 	/*
1339 	 * If the starting position is not word-aligned, deal with the
1340 	 * partial word.
1341 	 */
1342 	if (bit) {
1343 		/*
1344 		 * Calculate last (rightmost) bit number to look at,
1345 		 * and mask for all the relevant bits in this word.
1346 		 */
1347 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1348 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1349 		/*
1350 		 * Calculate the difference between the value there
1351 		 * and what we're looking for.
1352 		 */
1353 		if ((wdiff = (*b ^ want) & mask)) {
1354 			/*
1355 			 * Different.  Mark where we are and return.
1356 			 */
1357 			xfs_trans_brelse(tp, bp);
1358 			i = XFS_RTLOBIT(wdiff) - bit;
1359 			*rtblock = start + i - 1;
1360 			return 0;
1361 		}
1362 		i = lastbit - bit;
1363 		/*
1364 		 * Go on to next block if that's where the next word is
1365 		 * and we need the next word.
1366 		 */
1367 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1368 			/*
1369 			 * If done with this block, get the previous one.
1370 			 */
1371 			xfs_trans_brelse(tp, bp);
1372 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1373 			if (error) {
1374 				return error;
1375 			}
1376 			b = bufp = bp->b_addr;
1377 			word = 0;
1378 		} else {
1379 			/*
1380 			 * Go on to the previous word in the buffer.
1381 			 */
1382 			b++;
1383 		}
1384 	} else {
1385 		/*
1386 		 * Starting on a word boundary, no partial word.
1387 		 */
1388 		i = 0;
1389 	}
1390 	/*
1391 	 * Loop over whole words in buffers.  When we use up one buffer
1392 	 * we move on to the next one.
1393 	 */
1394 	while (len - i >= XFS_NBWORD) {
1395 		/*
1396 		 * Compute difference between actual and desired value.
1397 		 */
1398 		if ((wdiff = *b ^ want)) {
1399 			/*
1400 			 * Different, mark where we are and return.
1401 			 */
1402 			xfs_trans_brelse(tp, bp);
1403 			i += XFS_RTLOBIT(wdiff);
1404 			*rtblock = start + i - 1;
1405 			return 0;
1406 		}
1407 		i += XFS_NBWORD;
1408 		/*
1409 		 * Go on to next block if that's where the next word is
1410 		 * and we need the next word.
1411 		 */
1412 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1413 			/*
1414 			 * If done with this block, get the next one.
1415 			 */
1416 			xfs_trans_brelse(tp, bp);
1417 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1418 			if (error) {
1419 				return error;
1420 			}
1421 			b = bufp = bp->b_addr;
1422 			word = 0;
1423 		} else {
1424 			/*
1425 			 * Go on to the next word in the buffer.
1426 			 */
1427 			b++;
1428 		}
1429 	}
1430 	/*
1431 	 * If not ending on a word boundary, deal with the last
1432 	 * (partial) word.
1433 	 */
1434 	if ((lastbit = len - i)) {
1435 		/*
1436 		 * Calculate mask for all the relevant bits in this word.
1437 		 */
1438 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1439 		/*
1440 		 * Compute difference between actual and desired value.
1441 		 */
1442 		if ((wdiff = (*b ^ want) & mask)) {
1443 			/*
1444 			 * Different, mark where we are and return.
1445 			 */
1446 			xfs_trans_brelse(tp, bp);
1447 			i += XFS_RTLOBIT(wdiff);
1448 			*rtblock = start + i - 1;
1449 			return 0;
1450 		} else
1451 			i = len;
1452 	}
1453 	/*
1454 	 * No match, return that we scanned the whole area.
1455 	 */
1456 	xfs_trans_brelse(tp, bp);
1457 	*rtblock = start + i - 1;
1458 	return 0;
1459 }
1460 
1461 /*
1462  * Mark an extent specified by start and len freed.
1463  * Updates all the summary information as well as the bitmap.
1464  */
1465 STATIC int				/* error */
1466 xfs_rtfree_range(
1467 	xfs_mount_t	*mp,		/* file system mount point */
1468 	xfs_trans_t	*tp,		/* transaction pointer */
1469 	xfs_rtblock_t	start,		/* starting block to free */
1470 	xfs_extlen_t	len,		/* length to free */
1471 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1472 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1473 {
1474 	xfs_rtblock_t	end;		/* end of the freed extent */
1475 	int		error;		/* error value */
1476 	xfs_rtblock_t	postblock;	/* first block freed > end */
1477 	xfs_rtblock_t	preblock;	/* first block freed < start */
1478 
1479 	end = start + len - 1;
1480 	/*
1481 	 * Modify the bitmap to mark this extent freed.
1482 	 */
1483 	error = xfs_rtmodify_range(mp, tp, start, len, 1);
1484 	if (error) {
1485 		return error;
1486 	}
1487 	/*
1488 	 * Assume we're freeing out of the middle of an allocated extent.
1489 	 * We need to find the beginning and end of the extent so we can
1490 	 * properly update the summary.
1491 	 */
1492 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
1493 	if (error) {
1494 		return error;
1495 	}
1496 	/*
1497 	 * Find the next allocated block (end of allocated extent).
1498 	 */
1499 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
1500 		&postblock);
1501 	if (error)
1502 		return error;
1503 	/*
1504 	 * If there are blocks not being freed at the front of the
1505 	 * old extent, add summary data for them to be allocated.
1506 	 */
1507 	if (preblock < start) {
1508 		error = xfs_rtmodify_summary(mp, tp,
1509 			XFS_RTBLOCKLOG(start - preblock),
1510 			XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
1511 		if (error) {
1512 			return error;
1513 		}
1514 	}
1515 	/*
1516 	 * If there are blocks not being freed at the end of the
1517 	 * old extent, add summary data for them to be allocated.
1518 	 */
1519 	if (postblock > end) {
1520 		error = xfs_rtmodify_summary(mp, tp,
1521 			XFS_RTBLOCKLOG(postblock - end),
1522 			XFS_BITTOBLOCK(mp, end + 1), -1, rbpp, rsb);
1523 		if (error) {
1524 			return error;
1525 		}
1526 	}
1527 	/*
1528 	 * Increment the summary information corresponding to the entire
1529 	 * (new) free extent.
1530 	 */
1531 	error = xfs_rtmodify_summary(mp, tp,
1532 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
1533 		XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
1534 	return error;
1535 }
1536 
1537 /*
1538  * Read and return the summary information for a given extent size,
1539  * bitmap block combination.
1540  * Keeps track of a current summary block, so we don't keep reading
1541  * it from the buffer cache.
1542  */
1543 STATIC int				/* error */
1544 xfs_rtget_summary(
1545 	xfs_mount_t	*mp,		/* file system mount structure */
1546 	xfs_trans_t	*tp,		/* transaction pointer */
1547 	int		log,		/* log2 of extent size */
1548 	xfs_rtblock_t	bbno,		/* bitmap block number */
1549 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1550 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
1551 	xfs_suminfo_t	*sum)		/* out: summary info for this block */
1552 {
1553 	xfs_buf_t	*bp;		/* buffer for summary block */
1554 	int		error;		/* error value */
1555 	xfs_fsblock_t	sb;		/* summary fsblock */
1556 	int		so;		/* index into the summary file */
1557 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1558 
1559 	/*
1560 	 * Compute entry number in the summary file.
1561 	 */
1562 	so = XFS_SUMOFFS(mp, log, bbno);
1563 	/*
1564 	 * Compute the block number in the summary file.
1565 	 */
1566 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1567 	/*
1568 	 * If we have an old buffer, and the block number matches, use that.
1569 	 */
1570 	if (rbpp && *rbpp && *rsb == sb)
1571 		bp = *rbpp;
1572 	/*
1573 	 * Otherwise we have to get the buffer.
1574 	 */
1575 	else {
1576 		/*
1577 		 * If there was an old one, get rid of it first.
1578 		 */
1579 		if (rbpp && *rbpp)
1580 			xfs_trans_brelse(tp, *rbpp);
1581 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1582 		if (error) {
1583 			return error;
1584 		}
1585 		/*
1586 		 * Remember this buffer and block for the next call.
1587 		 */
1588 		if (rbpp) {
1589 			*rbpp = bp;
1590 			*rsb = sb;
1591 		}
1592 	}
1593 	/*
1594 	 * Point to the summary information & copy it out.
1595 	 */
1596 	sp = XFS_SUMPTR(mp, bp, so);
1597 	*sum = *sp;
1598 	/*
1599 	 * Drop the buffer if we're not asked to remember it.
1600 	 */
1601 	if (!rbpp)
1602 		xfs_trans_brelse(tp, bp);
1603 	return 0;
1604 }
1605 
1606 /*
1607  * Set the given range of bitmap bits to the given value.
1608  * Do whatever I/O and logging is required.
1609  */
1610 STATIC int				/* error */
1611 xfs_rtmodify_range(
1612 	xfs_mount_t	*mp,		/* file system mount point */
1613 	xfs_trans_t	*tp,		/* transaction pointer */
1614 	xfs_rtblock_t	start,		/* starting block to modify */
1615 	xfs_extlen_t	len,		/* length of extent to modify */
1616 	int		val)		/* 1 for free, 0 for allocated */
1617 {
1618 	xfs_rtword_t	*b;		/* current word in buffer */
1619 	int		bit;		/* bit number in the word */
1620 	xfs_rtblock_t	block;		/* bitmap block number */
1621 	xfs_buf_t	*bp;		/* buf for the block */
1622 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1623 	int		error;		/* error value */
1624 	xfs_rtword_t	*first;		/* first used word in the buffer */
1625 	int		i;		/* current bit number rel. to start */
1626 	int		lastbit;	/* last useful bit in word */
1627 	xfs_rtword_t	mask;		/* mask o frelevant bits for value */
1628 	int		word;		/* word number in the buffer */
1629 
1630 	/*
1631 	 * Compute starting bitmap block number.
1632 	 */
1633 	block = XFS_BITTOBLOCK(mp, start);
1634 	/*
1635 	 * Read the bitmap block, and point to its data.
1636 	 */
1637 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1638 	if (error) {
1639 		return error;
1640 	}
1641 	bufp = bp->b_addr;
1642 	/*
1643 	 * Compute the starting word's address, and starting bit.
1644 	 */
1645 	word = XFS_BITTOWORD(mp, start);
1646 	first = b = &bufp[word];
1647 	bit = (int)(start & (XFS_NBWORD - 1));
1648 	/*
1649 	 * 0 (allocated) => all zeroes; 1 (free) => all ones.
1650 	 */
1651 	val = -val;
1652 	/*
1653 	 * If not starting on a word boundary, deal with the first
1654 	 * (partial) word.
1655 	 */
1656 	if (bit) {
1657 		/*
1658 		 * Compute first bit not changed and mask of relevant bits.
1659 		 */
1660 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1661 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1662 		/*
1663 		 * Set/clear the active bits.
1664 		 */
1665 		if (val)
1666 			*b |= mask;
1667 		else
1668 			*b &= ~mask;
1669 		i = lastbit - bit;
1670 		/*
1671 		 * Go on to the next block if that's where the next word is
1672 		 * and we need the next word.
1673 		 */
1674 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1675 			/*
1676 			 * Log the changed part of this block.
1677 			 * Get the next one.
1678 			 */
1679 			xfs_trans_log_buf(tp, bp,
1680 				(uint)((char *)first - (char *)bufp),
1681 				(uint)((char *)b - (char *)bufp));
1682 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1683 			if (error) {
1684 				return error;
1685 			}
1686 			first = b = bufp = bp->b_addr;
1687 			word = 0;
1688 		} else {
1689 			/*
1690 			 * Go on to the next word in the buffer
1691 			 */
1692 			b++;
1693 		}
1694 	} else {
1695 		/*
1696 		 * Starting on a word boundary, no partial word.
1697 		 */
1698 		i = 0;
1699 	}
1700 	/*
1701 	 * Loop over whole words in buffers.  When we use up one buffer
1702 	 * we move on to the next one.
1703 	 */
1704 	while (len - i >= XFS_NBWORD) {
1705 		/*
1706 		 * Set the word value correctly.
1707 		 */
1708 		*b = val;
1709 		i += XFS_NBWORD;
1710 		/*
1711 		 * Go on to the next block if that's where the next word is
1712 		 * and we need the next word.
1713 		 */
1714 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1715 			/*
1716 			 * Log the changed part of this block.
1717 			 * Get the next one.
1718 			 */
1719 			xfs_trans_log_buf(tp, bp,
1720 				(uint)((char *)first - (char *)bufp),
1721 				(uint)((char *)b - (char *)bufp));
1722 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1723 			if (error) {
1724 				return error;
1725 			}
1726 			first = b = bufp = bp->b_addr;
1727 			word = 0;
1728 		} else {
1729 			/*
1730 			 * Go on to the next word in the buffer
1731 			 */
1732 			b++;
1733 		}
1734 	}
1735 	/*
1736 	 * If not ending on a word boundary, deal with the last
1737 	 * (partial) word.
1738 	 */
1739 	if ((lastbit = len - i)) {
1740 		/*
1741 		 * Compute a mask of relevant bits.
1742 		 */
1743 		bit = 0;
1744 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1745 		/*
1746 		 * Set/clear the active bits.
1747 		 */
1748 		if (val)
1749 			*b |= mask;
1750 		else
1751 			*b &= ~mask;
1752 		b++;
1753 	}
1754 	/*
1755 	 * Log any remaining changed bytes.
1756 	 */
1757 	if (b > first)
1758 		xfs_trans_log_buf(tp, bp, (uint)((char *)first - (char *)bufp),
1759 			(uint)((char *)b - (char *)bufp - 1));
1760 	return 0;
1761 }
1762 
1763 /*
1764  * Read and modify the summary information for a given extent size,
1765  * bitmap block combination.
1766  * Keeps track of a current summary block, so we don't keep reading
1767  * it from the buffer cache.
1768  */
1769 STATIC int				/* error */
1770 xfs_rtmodify_summary(
1771 	xfs_mount_t	*mp,		/* file system mount point */
1772 	xfs_trans_t	*tp,		/* transaction pointer */
1773 	int		log,		/* log2 of extent size */
1774 	xfs_rtblock_t	bbno,		/* bitmap block number */
1775 	int		delta,		/* change to make to summary info */
1776 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1777 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1778 {
1779 	xfs_buf_t	*bp;		/* buffer for the summary block */
1780 	int		error;		/* error value */
1781 	xfs_fsblock_t	sb;		/* summary fsblock */
1782 	int		so;		/* index into the summary file */
1783 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1784 
1785 	/*
1786 	 * Compute entry number in the summary file.
1787 	 */
1788 	so = XFS_SUMOFFS(mp, log, bbno);
1789 	/*
1790 	 * Compute the block number in the summary file.
1791 	 */
1792 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1793 	/*
1794 	 * If we have an old buffer, and the block number matches, use that.
1795 	 */
1796 	if (rbpp && *rbpp && *rsb == sb)
1797 		bp = *rbpp;
1798 	/*
1799 	 * Otherwise we have to get the buffer.
1800 	 */
1801 	else {
1802 		/*
1803 		 * If there was an old one, get rid of it first.
1804 		 */
1805 		if (rbpp && *rbpp)
1806 			xfs_trans_brelse(tp, *rbpp);
1807 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1808 		if (error) {
1809 			return error;
1810 		}
1811 		/*
1812 		 * Remember this buffer and block for the next call.
1813 		 */
1814 		if (rbpp) {
1815 			*rbpp = bp;
1816 			*rsb = sb;
1817 		}
1818 	}
1819 	/*
1820 	 * Point to the summary information, modify and log it.
1821 	 */
1822 	sp = XFS_SUMPTR(mp, bp, so);
1823 	*sp += delta;
1824 	xfs_trans_log_buf(tp, bp, (uint)((char *)sp - (char *)bp->b_addr),
1825 		(uint)((char *)sp - (char *)bp->b_addr + sizeof(*sp) - 1));
1826 	return 0;
1827 }
1828 
1829 /*
1830  * Visible (exported) functions.
1831  */
1832 
1833 /*
1834  * Grow the realtime area of the filesystem.
1835  */
1836 int
1837 xfs_growfs_rt(
1838 	xfs_mount_t	*mp,		/* mount point for filesystem */
1839 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
1840 {
1841 	xfs_rtblock_t	bmbno;		/* bitmap block number */
1842 	xfs_buf_t	*bp;		/* temporary buffer */
1843 	int		error;		/* error return value */
1844 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
1845 	xfs_drfsbno_t	nrblocks;	/* new number of realtime blocks */
1846 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
1847 	xfs_drtbno_t	nrextents;	/* new number of realtime extents */
1848 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
1849 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
1850 	uint		nrsumlevels;	/* new rt summary levels */
1851 	uint		nrsumsize;	/* new size of rt summary, bytes */
1852 	xfs_sb_t	*nsbp;		/* new superblock */
1853 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
1854 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
1855 	xfs_sb_t	*sbp;		/* old superblock */
1856 	xfs_fsblock_t	sumbno;		/* summary block number */
1857 
1858 	sbp = &mp->m_sb;
1859 	/*
1860 	 * Initial error checking.
1861 	 */
1862 	if (!capable(CAP_SYS_ADMIN))
1863 		return XFS_ERROR(EPERM);
1864 	if (mp->m_rtdev_targp == NULL || mp->m_rbmip == NULL ||
1865 	    (nrblocks = in->newblocks) <= sbp->sb_rblocks ||
1866 	    (sbp->sb_rblocks && (in->extsize != sbp->sb_rextsize)))
1867 		return XFS_ERROR(EINVAL);
1868 	if ((error = xfs_sb_validate_fsb_count(sbp, nrblocks)))
1869 		return error;
1870 	/*
1871 	 * Read in the last block of the device, make sure it exists.
1872 	 */
1873 	bp = xfs_buf_read_uncached(mp->m_rtdev_targp,
1874 				XFS_FSB_TO_BB(mp, nrblocks - 1),
1875 				XFS_FSB_TO_BB(mp, 1), 0);
1876 	if (!bp)
1877 		return EIO;
1878 	xfs_buf_relse(bp);
1879 
1880 	/*
1881 	 * Calculate new parameters.  These are the final values to be reached.
1882 	 */
1883 	nrextents = nrblocks;
1884 	do_div(nrextents, in->extsize);
1885 	nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
1886 	nrextslog = xfs_highbit32(nrextents);
1887 	nrsumlevels = nrextslog + 1;
1888 	nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
1889 	nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1890 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1891 	/*
1892 	 * New summary size can't be more than half the size of
1893 	 * the log.  This prevents us from getting a log overflow,
1894 	 * since we'll log basically the whole summary file at once.
1895 	 */
1896 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
1897 		return XFS_ERROR(EINVAL);
1898 	/*
1899 	 * Get the old block counts for bitmap and summary inodes.
1900 	 * These can't change since other growfs callers are locked out.
1901 	 */
1902 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_d.di_size);
1903 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_d.di_size);
1904 	/*
1905 	 * Allocate space to the bitmap and summary files, as necessary.
1906 	 */
1907 	error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
1908 	if (error)
1909 		return error;
1910 	error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
1911 	if (error)
1912 		return error;
1913 	/*
1914 	 * Allocate a new (fake) mount/sb.
1915 	 */
1916 	nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
1917 	/*
1918 	 * Loop over the bitmap blocks.
1919 	 * We will do everything one bitmap block at a time.
1920 	 * Skip the current block if it is exactly full.
1921 	 * This also deals with the case where there were no rtextents before.
1922 	 */
1923 	for (bmbno = sbp->sb_rbmblocks -
1924 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1925 	     bmbno < nrbmblocks;
1926 	     bmbno++) {
1927 		xfs_trans_t	*tp;
1928 		int		cancelflags = 0;
1929 
1930 		*nmp = *mp;
1931 		nsbp = &nmp->m_sb;
1932 		/*
1933 		 * Calculate new sb and mount fields for this round.
1934 		 */
1935 		nsbp->sb_rextsize = in->extsize;
1936 		nsbp->sb_rbmblocks = bmbno + 1;
1937 		nsbp->sb_rblocks =
1938 			XFS_RTMIN(nrblocks,
1939 				  nsbp->sb_rbmblocks * NBBY *
1940 				  nsbp->sb_blocksize * nsbp->sb_rextsize);
1941 		nsbp->sb_rextents = nsbp->sb_rblocks;
1942 		do_div(nsbp->sb_rextents, nsbp->sb_rextsize);
1943 		ASSERT(nsbp->sb_rextents != 0);
1944 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1945 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1946 		nrsumsize =
1947 			(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
1948 			nsbp->sb_rbmblocks;
1949 		nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1950 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1951 		/*
1952 		 * Start a transaction, get the log reservation.
1953 		 */
1954 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_FREE);
1955 		if ((error = xfs_trans_reserve(tp, 0,
1956 				XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
1957 			goto error_cancel;
1958 		/*
1959 		 * Lock out other callers by grabbing the bitmap inode lock.
1960 		 */
1961 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
1962 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
1963 		/*
1964 		 * Update the bitmap inode's size.
1965 		 */
1966 		mp->m_rbmip->i_d.di_size =
1967 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1968 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1969 		cancelflags |= XFS_TRANS_ABORT;
1970 		/*
1971 		 * Get the summary inode into the transaction.
1972 		 */
1973 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL);
1974 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
1975 		/*
1976 		 * Update the summary inode's size.
1977 		 */
1978 		mp->m_rsumip->i_d.di_size = nmp->m_rsumsize;
1979 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
1980 		/*
1981 		 * Copy summary data from old to new sizes.
1982 		 * Do this when the real size (not block-aligned) changes.
1983 		 */
1984 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
1985 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
1986 			error = xfs_rtcopy_summary(mp, nmp, tp);
1987 			if (error)
1988 				goto error_cancel;
1989 		}
1990 		/*
1991 		 * Update superblock fields.
1992 		 */
1993 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
1994 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
1995 				nsbp->sb_rextsize - sbp->sb_rextsize);
1996 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
1997 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
1998 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
1999 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
2000 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
2001 				nsbp->sb_rblocks - sbp->sb_rblocks);
2002 		if (nsbp->sb_rextents != sbp->sb_rextents)
2003 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
2004 				nsbp->sb_rextents - sbp->sb_rextents);
2005 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
2006 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
2007 				nsbp->sb_rextslog - sbp->sb_rextslog);
2008 		/*
2009 		 * Free new extent.
2010 		 */
2011 		bp = NULL;
2012 		error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
2013 			nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
2014 		if (error) {
2015 error_cancel:
2016 			xfs_trans_cancel(tp, cancelflags);
2017 			break;
2018 		}
2019 		/*
2020 		 * Mark more blocks free in the superblock.
2021 		 */
2022 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
2023 			nsbp->sb_rextents - sbp->sb_rextents);
2024 		/*
2025 		 * Update mp values into the real mp structure.
2026 		 */
2027 		mp->m_rsumlevels = nrsumlevels;
2028 		mp->m_rsumsize = nrsumsize;
2029 
2030 		error = xfs_trans_commit(tp, 0);
2031 		if (error)
2032 			break;
2033 	}
2034 
2035 	/*
2036 	 * Free the fake mp structure.
2037 	 */
2038 	kmem_free(nmp);
2039 
2040 	return error;
2041 }
2042 
2043 /*
2044  * Allocate an extent in the realtime subvolume, with the usual allocation
2045  * parameters.  The length units are all in realtime extents, as is the
2046  * result block number.
2047  */
2048 int					/* error */
2049 xfs_rtallocate_extent(
2050 	xfs_trans_t	*tp,		/* transaction pointer */
2051 	xfs_rtblock_t	bno,		/* starting block number to allocate */
2052 	xfs_extlen_t	minlen,		/* minimum length to allocate */
2053 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
2054 	xfs_extlen_t	*len,		/* out: actual length allocated */
2055 	xfs_alloctype_t	type,		/* allocation type XFS_ALLOCTYPE... */
2056 	int		wasdel,		/* was a delayed allocation extent */
2057 	xfs_extlen_t	prod,		/* extent product factor */
2058 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
2059 {
2060 	xfs_mount_t	*mp = tp->t_mountp;
2061 	int		error;		/* error value */
2062 	xfs_rtblock_t	r;		/* result allocated block */
2063 	xfs_fsblock_t	sb;		/* summary file block number */
2064 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2065 
2066 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2067 	ASSERT(minlen > 0 && minlen <= maxlen);
2068 
2069 	/*
2070 	 * If prod is set then figure out what to do to minlen and maxlen.
2071 	 */
2072 	if (prod > 1) {
2073 		xfs_extlen_t	i;
2074 
2075 		if ((i = maxlen % prod))
2076 			maxlen -= i;
2077 		if ((i = minlen % prod))
2078 			minlen += prod - i;
2079 		if (maxlen < minlen) {
2080 			*rtblock = NULLRTBLOCK;
2081 			return 0;
2082 		}
2083 	}
2084 
2085 	sumbp = NULL;
2086 	/*
2087 	 * Allocate by size, or near another block, or exactly at some block.
2088 	 */
2089 	switch (type) {
2090 	case XFS_ALLOCTYPE_ANY_AG:
2091 		error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
2092 				&sumbp,	&sb, prod, &r);
2093 		break;
2094 	case XFS_ALLOCTYPE_NEAR_BNO:
2095 		error = xfs_rtallocate_extent_near(mp, tp, bno, minlen, maxlen,
2096 				len, &sumbp, &sb, prod, &r);
2097 		break;
2098 	case XFS_ALLOCTYPE_THIS_BNO:
2099 		error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen,
2100 				len, &sumbp, &sb, prod, &r);
2101 		break;
2102 	default:
2103 		error = EIO;
2104 		ASSERT(0);
2105 	}
2106 	if (error)
2107 		return error;
2108 
2109 	/*
2110 	 * If it worked, update the superblock.
2111 	 */
2112 	if (r != NULLRTBLOCK) {
2113 		long	slen = (long)*len;
2114 
2115 		ASSERT(*len >= minlen && *len <= maxlen);
2116 		if (wasdel)
2117 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
2118 		else
2119 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
2120 	}
2121 	*rtblock = r;
2122 	return 0;
2123 }
2124 
2125 /*
2126  * Free an extent in the realtime subvolume.  Length is expressed in
2127  * realtime extents, as is the block number.
2128  */
2129 int					/* error */
2130 xfs_rtfree_extent(
2131 	xfs_trans_t	*tp,		/* transaction pointer */
2132 	xfs_rtblock_t	bno,		/* starting block number to free */
2133 	xfs_extlen_t	len)		/* length of extent freed */
2134 {
2135 	int		error;		/* error value */
2136 	xfs_mount_t	*mp;		/* file system mount structure */
2137 	xfs_fsblock_t	sb;		/* summary file block number */
2138 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2139 
2140 	mp = tp->t_mountp;
2141 
2142 	ASSERT(mp->m_rbmip->i_itemp != NULL);
2143 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2144 
2145 #if defined(__KERNEL__) && defined(DEBUG)
2146 	/*
2147 	 * Check to see that this whole range is currently allocated.
2148 	 */
2149 	{
2150 		int	stat;		/* result from checking range */
2151 
2152 		error = xfs_rtcheck_alloc_range(mp, tp, bno, len, &stat);
2153 		if (error) {
2154 			return error;
2155 		}
2156 		ASSERT(stat);
2157 	}
2158 #endif
2159 	sumbp = NULL;
2160 	/*
2161 	 * Free the range of realtime blocks.
2162 	 */
2163 	error = xfs_rtfree_range(mp, tp, bno, len, &sumbp, &sb);
2164 	if (error) {
2165 		return error;
2166 	}
2167 	/*
2168 	 * Mark more blocks free in the superblock.
2169 	 */
2170 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, (long)len);
2171 	/*
2172 	 * If we've now freed all the blocks, reset the file sequence
2173 	 * number to 0.
2174 	 */
2175 	if (tp->t_frextents_delta + mp->m_sb.sb_frextents ==
2176 	    mp->m_sb.sb_rextents) {
2177 		if (!(mp->m_rbmip->i_d.di_flags & XFS_DIFLAG_NEWRTBM))
2178 			mp->m_rbmip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2179 		*(__uint64_t *)&mp->m_rbmip->i_d.di_atime = 0;
2180 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
2181 	}
2182 	return 0;
2183 }
2184 
2185 /*
2186  * Initialize realtime fields in the mount structure.
2187  */
2188 int				/* error */
2189 xfs_rtmount_init(
2190 	xfs_mount_t	*mp)	/* file system mount structure */
2191 {
2192 	xfs_buf_t	*bp;	/* buffer for last block of subvolume */
2193 	xfs_daddr_t	d;	/* address of last block of subvolume */
2194 	xfs_sb_t	*sbp;	/* filesystem superblock copy in mount */
2195 
2196 	sbp = &mp->m_sb;
2197 	if (sbp->sb_rblocks == 0)
2198 		return 0;
2199 	if (mp->m_rtdev_targp == NULL) {
2200 		xfs_warn(mp,
2201 	"Filesystem has a realtime volume, use rtdev=device option");
2202 		return XFS_ERROR(ENODEV);
2203 	}
2204 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
2205 	mp->m_rsumsize =
2206 		(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
2207 		sbp->sb_rbmblocks;
2208 	mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
2209 	mp->m_rbmip = mp->m_rsumip = NULL;
2210 	/*
2211 	 * Check that the realtime section is an ok size.
2212 	 */
2213 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
2214 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
2215 		xfs_warn(mp, "realtime mount -- %llu != %llu",
2216 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
2217 			(unsigned long long) mp->m_sb.sb_rblocks);
2218 		return XFS_ERROR(EFBIG);
2219 	}
2220 	bp = xfs_buf_read_uncached(mp->m_rtdev_targp,
2221 					d - XFS_FSB_TO_BB(mp, 1),
2222 					XFS_FSB_TO_BB(mp, 1), 0);
2223 	if (!bp) {
2224 		xfs_warn(mp, "realtime device size check failed");
2225 		return EIO;
2226 	}
2227 	xfs_buf_relse(bp);
2228 	return 0;
2229 }
2230 
2231 /*
2232  * Get the bitmap and summary inodes into the mount structure
2233  * at mount time.
2234  */
2235 int					/* error */
2236 xfs_rtmount_inodes(
2237 	xfs_mount_t	*mp)		/* file system mount structure */
2238 {
2239 	int		error;		/* error return value */
2240 	xfs_sb_t	*sbp;
2241 
2242 	sbp = &mp->m_sb;
2243 	if (sbp->sb_rbmino == NULLFSINO)
2244 		return 0;
2245 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
2246 	if (error)
2247 		return error;
2248 	ASSERT(mp->m_rbmip != NULL);
2249 	ASSERT(sbp->sb_rsumino != NULLFSINO);
2250 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
2251 	if (error) {
2252 		IRELE(mp->m_rbmip);
2253 		return error;
2254 	}
2255 	ASSERT(mp->m_rsumip != NULL);
2256 	return 0;
2257 }
2258 
2259 void
2260 xfs_rtunmount_inodes(
2261 	struct xfs_mount	*mp)
2262 {
2263 	if (mp->m_rbmip)
2264 		IRELE(mp->m_rbmip);
2265 	if (mp->m_rsumip)
2266 		IRELE(mp->m_rsumip);
2267 }
2268 
2269 /*
2270  * Pick an extent for allocation at the start of a new realtime file.
2271  * Use the sequence number stored in the atime field of the bitmap inode.
2272  * Translate this to a fraction of the rtextents, and return the product
2273  * of rtextents and the fraction.
2274  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
2275  */
2276 int					/* error */
2277 xfs_rtpick_extent(
2278 	xfs_mount_t	*mp,		/* file system mount point */
2279 	xfs_trans_t	*tp,		/* transaction pointer */
2280 	xfs_extlen_t	len,		/* allocation length (rtextents) */
2281 	xfs_rtblock_t	*pick)		/* result rt extent */
2282 {
2283 	xfs_rtblock_t	b;		/* result block */
2284 	int		log2;		/* log of sequence number */
2285 	__uint64_t	resid;		/* residual after log removed */
2286 	__uint64_t	seq;		/* sequence number of file creation */
2287 	__uint64_t	*seqp;		/* pointer to seqno in inode */
2288 
2289 	ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
2290 
2291 	seqp = (__uint64_t *)&mp->m_rbmip->i_d.di_atime;
2292 	if (!(mp->m_rbmip->i_d.di_flags & XFS_DIFLAG_NEWRTBM)) {
2293 		mp->m_rbmip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2294 		*seqp = 0;
2295 	}
2296 	seq = *seqp;
2297 	if ((log2 = xfs_highbit64(seq)) == -1)
2298 		b = 0;
2299 	else {
2300 		resid = seq - (1ULL << log2);
2301 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
2302 		    (log2 + 1);
2303 		if (b >= mp->m_sb.sb_rextents)
2304 			b = do_mod(b, mp->m_sb.sb_rextents);
2305 		if (b + len > mp->m_sb.sb_rextents)
2306 			b = mp->m_sb.sb_rextents - len;
2307 	}
2308 	*seqp = seq + 1;
2309 	xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
2310 	*pick = b;
2311 	return 0;
2312 }
2313