xref: /openbmc/linux/fs/xfs/xfs_rtalloc.c (revision 545e4006)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_fsops.h"
43 #include "xfs_error.h"
44 #include "xfs_rw.h"
45 #include "xfs_inode_item.h"
46 #include "xfs_trans_space.h"
47 #include "xfs_utils.h"
48 
49 
50 /*
51  * Prototypes for internal functions.
52  */
53 
54 
55 STATIC int xfs_rtallocate_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
56 		xfs_extlen_t, xfs_buf_t **, xfs_fsblock_t *);
57 STATIC int xfs_rtany_summary(xfs_mount_t *, xfs_trans_t *, int, int,
58 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, int *);
59 STATIC int xfs_rtcheck_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
60 		xfs_extlen_t, int, xfs_rtblock_t *, int *);
61 STATIC int xfs_rtfind_back(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
62 		xfs_rtblock_t, xfs_rtblock_t *);
63 STATIC int xfs_rtfind_forw(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
64 		xfs_rtblock_t, xfs_rtblock_t *);
65 STATIC int xfs_rtget_summary( xfs_mount_t *, xfs_trans_t *, int,
66 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, xfs_suminfo_t *);
67 STATIC int xfs_rtmodify_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
68 		xfs_extlen_t, int);
69 STATIC int xfs_rtmodify_summary(xfs_mount_t *, xfs_trans_t *, int,
70 		xfs_rtblock_t, int, xfs_buf_t **, xfs_fsblock_t *);
71 
72 /*
73  * Internal functions.
74  */
75 
76 /*
77  * xfs_lowbit32: get low bit set out of 32-bit argument, -1 if none set.
78  */
79 STATIC int
80 xfs_lowbit32(
81 	__uint32_t	v)
82 {
83 	if (v)
84 		return ffs(v) - 1;
85 	return -1;
86 }
87 
88 /*
89  * Allocate space to the bitmap or summary file, and zero it, for growfs.
90  */
91 STATIC int				/* error */
92 xfs_growfs_rt_alloc(
93 	xfs_mount_t	*mp,		/* file system mount point */
94 	xfs_extlen_t	oblocks,	/* old count of blocks */
95 	xfs_extlen_t	nblocks,	/* new count of blocks */
96 	xfs_ino_t	ino)		/* inode number (bitmap/summary) */
97 {
98 	xfs_fileoff_t	bno;		/* block number in file */
99 	xfs_buf_t	*bp;		/* temporary buffer for zeroing */
100 	int		cancelflags;	/* flags for xfs_trans_cancel */
101 	int		committed;	/* transaction committed flag */
102 	xfs_daddr_t	d;		/* disk block address */
103 	int		error;		/* error return value */
104 	xfs_fsblock_t	firstblock;	/* first block allocated in xaction */
105 	xfs_bmap_free_t	flist;		/* list of freed blocks */
106 	xfs_fsblock_t	fsbno;		/* filesystem block for bno */
107 	xfs_inode_t	*ip;		/* pointer to incore inode */
108 	xfs_bmbt_irec_t	map;		/* block map output */
109 	int		nmap;		/* number of block maps */
110 	int		resblks;	/* space reservation */
111 	xfs_trans_t	*tp;		/* transaction pointer */
112 
113 	/*
114 	 * Allocate space to the file, as necessary.
115 	 */
116 	while (oblocks < nblocks) {
117 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ALLOC);
118 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
119 		cancelflags = 0;
120 		/*
121 		 * Reserve space & log for one extent added to the file.
122 		 */
123 		if ((error = xfs_trans_reserve(tp, resblks,
124 				XFS_GROWRTALLOC_LOG_RES(mp), 0,
125 				XFS_TRANS_PERM_LOG_RES,
126 				XFS_DEFAULT_PERM_LOG_COUNT)))
127 			goto error_cancel;
128 		cancelflags = XFS_TRANS_RELEASE_LOG_RES;
129 		/*
130 		 * Lock the inode.
131 		 */
132 		if ((error = xfs_trans_iget(mp, tp, ino, 0,
133 						XFS_ILOCK_EXCL, &ip)))
134 			goto error_cancel;
135 		XFS_BMAP_INIT(&flist, &firstblock);
136 		/*
137 		 * Allocate blocks to the bitmap file.
138 		 */
139 		nmap = 1;
140 		cancelflags |= XFS_TRANS_ABORT;
141 		error = xfs_bmapi(tp, ip, oblocks, nblocks - oblocks,
142 			XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, &firstblock,
143 			resblks, &map, &nmap, &flist, NULL);
144 		if (!error && nmap < 1)
145 			error = XFS_ERROR(ENOSPC);
146 		if (error)
147 			goto error_cancel;
148 		/*
149 		 * Free any blocks freed up in the transaction, then commit.
150 		 */
151 		error = xfs_bmap_finish(&tp, &flist, &committed);
152 		if (error)
153 			goto error_cancel;
154 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
155 		if (error)
156 			goto error;
157 		/*
158 		 * Now we need to clear the allocated blocks.
159 		 * Do this one block per transaction, to keep it simple.
160 		 */
161 		cancelflags = 0;
162 		for (bno = map.br_startoff, fsbno = map.br_startblock;
163 		     bno < map.br_startoff + map.br_blockcount;
164 		     bno++, fsbno++) {
165 			tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ZERO);
166 			/*
167 			 * Reserve log for one block zeroing.
168 			 */
169 			if ((error = xfs_trans_reserve(tp, 0,
170 					XFS_GROWRTZERO_LOG_RES(mp), 0, 0, 0)))
171 				goto error_cancel;
172 			/*
173 			 * Lock the bitmap inode.
174 			 */
175 			if ((error = xfs_trans_iget(mp, tp, ino, 0,
176 							XFS_ILOCK_EXCL, &ip)))
177 				goto error_cancel;
178 			/*
179 			 * Get a buffer for the block.
180 			 */
181 			d = XFS_FSB_TO_DADDR(mp, fsbno);
182 			bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
183 				mp->m_bsize, 0);
184 			if (bp == NULL) {
185 				error = XFS_ERROR(EIO);
186 				goto error_cancel;
187 			}
188 			memset(XFS_BUF_PTR(bp), 0, mp->m_sb.sb_blocksize);
189 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
190 			/*
191 			 * Commit the transaction.
192 			 */
193 			error = xfs_trans_commit(tp, 0);
194 			if (error)
195 				goto error;
196 		}
197 		/*
198 		 * Go on to the next extent, if any.
199 		 */
200 		oblocks = map.br_startoff + map.br_blockcount;
201 	}
202 	return 0;
203 error_cancel:
204 	xfs_trans_cancel(tp, cancelflags);
205 error:
206 	return error;
207 }
208 
209 /*
210  * Attempt to allocate an extent minlen<=len<=maxlen starting from
211  * bitmap block bbno.  If we don't get maxlen then use prod to trim
212  * the length, if given.  Returns error; returns starting block in *rtblock.
213  * The lengths are all in rtextents.
214  */
215 STATIC int				/* error */
216 xfs_rtallocate_extent_block(
217 	xfs_mount_t	*mp,		/* file system mount point */
218 	xfs_trans_t	*tp,		/* transaction pointer */
219 	xfs_rtblock_t	bbno,		/* bitmap block number */
220 	xfs_extlen_t	minlen,		/* minimum length to allocate */
221 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
222 	xfs_extlen_t	*len,		/* out: actual length allocated */
223 	xfs_rtblock_t	*nextp,		/* out: next block to try */
224 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
225 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
226 	xfs_extlen_t	prod,		/* extent product factor */
227 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
228 {
229 	xfs_rtblock_t	besti;		/* best rtblock found so far */
230 	xfs_rtblock_t	bestlen;	/* best length found so far */
231 	xfs_rtblock_t	end;		/* last rtblock in chunk */
232 	int		error;		/* error value */
233 	xfs_rtblock_t	i;		/* current rtblock trying */
234 	xfs_rtblock_t	next;		/* next rtblock to try */
235 	int		stat;		/* status from internal calls */
236 
237 	/*
238 	 * Loop over all the extents starting in this bitmap block,
239 	 * looking for one that's long enough.
240 	 */
241 	for (i = XFS_BLOCKTOBIT(mp, bbno), besti = -1, bestlen = 0,
242 		end = XFS_BLOCKTOBIT(mp, bbno + 1) - 1;
243 	     i <= end;
244 	     i++) {
245 		/*
246 		 * See if there's a free extent of maxlen starting at i.
247 		 * If it's not so then next will contain the first non-free.
248 		 */
249 		error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
250 		if (error) {
251 			return error;
252 		}
253 		if (stat) {
254 			/*
255 			 * i for maxlen is all free, allocate and return that.
256 			 */
257 			error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
258 				rsb);
259 			if (error) {
260 				return error;
261 			}
262 			*len = maxlen;
263 			*rtblock = i;
264 			return 0;
265 		}
266 		/*
267 		 * In the case where we have a variable-sized allocation
268 		 * request, figure out how big this free piece is,
269 		 * and if it's big enough for the minimum, and the best
270 		 * so far, remember it.
271 		 */
272 		if (minlen < maxlen) {
273 			xfs_rtblock_t	thislen;	/* this extent size */
274 
275 			thislen = next - i;
276 			if (thislen >= minlen && thislen > bestlen) {
277 				besti = i;
278 				bestlen = thislen;
279 			}
280 		}
281 		/*
282 		 * If not done yet, find the start of the next free space.
283 		 */
284 		if (next < end) {
285 			error = xfs_rtfind_forw(mp, tp, next, end, &i);
286 			if (error) {
287 				return error;
288 			}
289 		} else
290 			break;
291 	}
292 	/*
293 	 * Searched the whole thing & didn't find a maxlen free extent.
294 	 */
295 	if (minlen < maxlen && besti != -1) {
296 		xfs_extlen_t	p;	/* amount to trim length by */
297 
298 		/*
299 		 * If size should be a multiple of prod, make that so.
300 		 */
301 		if (prod > 1 && (p = do_mod(bestlen, prod)))
302 			bestlen -= p;
303 		/*
304 		 * Allocate besti for bestlen & return that.
305 		 */
306 		error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
307 		if (error) {
308 			return error;
309 		}
310 		*len = bestlen;
311 		*rtblock = besti;
312 		return 0;
313 	}
314 	/*
315 	 * Allocation failed.  Set *nextp to the next block to try.
316 	 */
317 	*nextp = next;
318 	*rtblock = NULLRTBLOCK;
319 	return 0;
320 }
321 
322 /*
323  * Allocate an extent of length minlen<=len<=maxlen, starting at block
324  * bno.  If we don't get maxlen then use prod to trim the length, if given.
325  * Returns error; returns starting block in *rtblock.
326  * The lengths are all in rtextents.
327  */
328 STATIC int				/* error */
329 xfs_rtallocate_extent_exact(
330 	xfs_mount_t	*mp,		/* file system mount point */
331 	xfs_trans_t	*tp,		/* transaction pointer */
332 	xfs_rtblock_t	bno,		/* starting block number to allocate */
333 	xfs_extlen_t	minlen,		/* minimum length to allocate */
334 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
335 	xfs_extlen_t	*len,		/* out: actual length allocated */
336 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
337 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
338 	xfs_extlen_t	prod,		/* extent product factor */
339 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
340 {
341 	int		error;		/* error value */
342 	xfs_extlen_t	i;		/* extent length trimmed due to prod */
343 	int		isfree;		/* extent is free */
344 	xfs_rtblock_t	next;		/* next block to try (dummy) */
345 
346 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
347 	/*
348 	 * Check if the range in question (for maxlen) is free.
349 	 */
350 	error = xfs_rtcheck_range(mp, tp, bno, maxlen, 1, &next, &isfree);
351 	if (error) {
352 		return error;
353 	}
354 	if (isfree) {
355 		/*
356 		 * If it is, allocate it and return success.
357 		 */
358 		error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
359 		if (error) {
360 			return error;
361 		}
362 		*len = maxlen;
363 		*rtblock = bno;
364 		return 0;
365 	}
366 	/*
367 	 * If not, allocate what there is, if it's at least minlen.
368 	 */
369 	maxlen = next - bno;
370 	if (maxlen < minlen) {
371 		/*
372 		 * Failed, return failure status.
373 		 */
374 		*rtblock = NULLRTBLOCK;
375 		return 0;
376 	}
377 	/*
378 	 * Trim off tail of extent, if prod is specified.
379 	 */
380 	if (prod > 1 && (i = maxlen % prod)) {
381 		maxlen -= i;
382 		if (maxlen < minlen) {
383 			/*
384 			 * Now we can't do it, return failure status.
385 			 */
386 			*rtblock = NULLRTBLOCK;
387 			return 0;
388 		}
389 	}
390 	/*
391 	 * Allocate what we can and return it.
392 	 */
393 	error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
394 	if (error) {
395 		return error;
396 	}
397 	*len = maxlen;
398 	*rtblock = bno;
399 	return 0;
400 }
401 
402 /*
403  * Allocate an extent of length minlen<=len<=maxlen, starting as near
404  * to bno as possible.  If we don't get maxlen then use prod to trim
405  * the length, if given.  The lengths are all in rtextents.
406  */
407 STATIC int				/* error */
408 xfs_rtallocate_extent_near(
409 	xfs_mount_t	*mp,		/* file system mount point */
410 	xfs_trans_t	*tp,		/* transaction pointer */
411 	xfs_rtblock_t	bno,		/* starting block number to allocate */
412 	xfs_extlen_t	minlen,		/* minimum length to allocate */
413 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
414 	xfs_extlen_t	*len,		/* out: actual length allocated */
415 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
416 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
417 	xfs_extlen_t	prod,		/* extent product factor */
418 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
419 {
420 	int		any;		/* any useful extents from summary */
421 	xfs_rtblock_t	bbno;		/* bitmap block number */
422 	int		error;		/* error value */
423 	int		i;		/* bitmap block offset (loop control) */
424 	int		j;		/* secondary loop control */
425 	int		log2len;	/* log2 of minlen */
426 	xfs_rtblock_t	n;		/* next block to try */
427 	xfs_rtblock_t	r;		/* result block */
428 
429 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
430 	/*
431 	 * If the block number given is off the end, silently set it to
432 	 * the last block.
433 	 */
434 	if (bno >= mp->m_sb.sb_rextents)
435 		bno = mp->m_sb.sb_rextents - 1;
436 	/*
437 	 * Try the exact allocation first.
438 	 */
439 	error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen, len,
440 		rbpp, rsb, prod, &r);
441 	if (error) {
442 		return error;
443 	}
444 	/*
445 	 * If the exact allocation worked, return that.
446 	 */
447 	if (r != NULLRTBLOCK) {
448 		*rtblock = r;
449 		return 0;
450 	}
451 	bbno = XFS_BITTOBLOCK(mp, bno);
452 	i = 0;
453 	log2len = xfs_highbit32(minlen);
454 	/*
455 	 * Loop over all bitmap blocks (bbno + i is current block).
456 	 */
457 	for (;;) {
458 		/*
459 		 * Get summary information of extents of all useful levels
460 		 * starting in this bitmap block.
461 		 */
462 		error = xfs_rtany_summary(mp, tp, log2len, mp->m_rsumlevels - 1,
463 			bbno + i, rbpp, rsb, &any);
464 		if (error) {
465 			return error;
466 		}
467 		/*
468 		 * If there are any useful extents starting here, try
469 		 * allocating one.
470 		 */
471 		if (any) {
472 			/*
473 			 * On the positive side of the starting location.
474 			 */
475 			if (i >= 0) {
476 				/*
477 				 * Try to allocate an extent starting in
478 				 * this block.
479 				 */
480 				error = xfs_rtallocate_extent_block(mp, tp,
481 					bbno + i, minlen, maxlen, len, &n, rbpp,
482 					rsb, prod, &r);
483 				if (error) {
484 					return error;
485 				}
486 				/*
487 				 * If it worked, return it.
488 				 */
489 				if (r != NULLRTBLOCK) {
490 					*rtblock = r;
491 					return 0;
492 				}
493 			}
494 			/*
495 			 * On the negative side of the starting location.
496 			 */
497 			else {		/* i < 0 */
498 				/*
499 				 * Loop backwards through the bitmap blocks from
500 				 * the starting point-1 up to where we are now.
501 				 * There should be an extent which ends in this
502 				 * bitmap block and is long enough.
503 				 */
504 				for (j = -1; j > i; j--) {
505 					/*
506 					 * Grab the summary information for
507 					 * this bitmap block.
508 					 */
509 					error = xfs_rtany_summary(mp, tp,
510 						log2len, mp->m_rsumlevels - 1,
511 						bbno + j, rbpp, rsb, &any);
512 					if (error) {
513 						return error;
514 					}
515 					/*
516 					 * If there's no extent given in the
517 					 * summary that means the extent we
518 					 * found must carry over from an
519 					 * earlier block.  If there is an
520 					 * extent given, we've already tried
521 					 * that allocation, don't do it again.
522 					 */
523 					if (any)
524 						continue;
525 					error = xfs_rtallocate_extent_block(mp,
526 						tp, bbno + j, minlen, maxlen,
527 						len, &n, rbpp, rsb, prod, &r);
528 					if (error) {
529 						return error;
530 					}
531 					/*
532 					 * If it works, return the extent.
533 					 */
534 					if (r != NULLRTBLOCK) {
535 						*rtblock = r;
536 						return 0;
537 					}
538 				}
539 				/*
540 				 * There weren't intervening bitmap blocks
541 				 * with a long enough extent, or the
542 				 * allocation didn't work for some reason
543 				 * (i.e. it's a little * too short).
544 				 * Try to allocate from the summary block
545 				 * that we found.
546 				 */
547 				error = xfs_rtallocate_extent_block(mp, tp,
548 					bbno + i, minlen, maxlen, len, &n, rbpp,
549 					rsb, prod, &r);
550 				if (error) {
551 					return error;
552 				}
553 				/*
554 				 * If it works, return the extent.
555 				 */
556 				if (r != NULLRTBLOCK) {
557 					*rtblock = r;
558 					return 0;
559 				}
560 			}
561 		}
562 		/*
563 		 * Loop control.  If we were on the positive side, and there's
564 		 * still more blocks on the negative side, go there.
565 		 */
566 		if (i > 0 && (int)bbno - i >= 0)
567 			i = -i;
568 		/*
569 		 * If positive, and no more negative, but there are more
570 		 * positive, go there.
571 		 */
572 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
573 			i++;
574 		/*
575 		 * If negative or 0 (just started), and there are positive
576 		 * blocks to go, go there.  The 0 case moves to block 1.
577 		 */
578 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
579 			i = 1 - i;
580 		/*
581 		 * If negative or 0 and there are more negative blocks,
582 		 * go there.
583 		 */
584 		else if (i <= 0 && (int)bbno + i > 0)
585 			i--;
586 		/*
587 		 * Must be done.  Return failure.
588 		 */
589 		else
590 			break;
591 	}
592 	*rtblock = NULLRTBLOCK;
593 	return 0;
594 }
595 
596 /*
597  * Allocate an extent of length minlen<=len<=maxlen, with no position
598  * specified.  If we don't get maxlen then use prod to trim
599  * the length, if given.  The lengths are all in rtextents.
600  */
601 STATIC int				/* error */
602 xfs_rtallocate_extent_size(
603 	xfs_mount_t	*mp,		/* file system mount point */
604 	xfs_trans_t	*tp,		/* transaction pointer */
605 	xfs_extlen_t	minlen,		/* minimum length to allocate */
606 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
607 	xfs_extlen_t	*len,		/* out: actual length allocated */
608 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
609 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
610 	xfs_extlen_t	prod,		/* extent product factor */
611 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
612 {
613 	int		error;		/* error value */
614 	int		i;		/* bitmap block number */
615 	int		l;		/* level number (loop control) */
616 	xfs_rtblock_t	n;		/* next block to be tried */
617 	xfs_rtblock_t	r;		/* result block number */
618 	xfs_suminfo_t	sum;		/* summary information for extents */
619 
620 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
621 	/*
622 	 * Loop over all the levels starting with maxlen.
623 	 * At each level, look at all the bitmap blocks, to see if there
624 	 * are extents starting there that are long enough (>= maxlen).
625 	 * Note, only on the initial level can the allocation fail if
626 	 * the summary says there's an extent.
627 	 */
628 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
629 		/*
630 		 * Loop over all the bitmap blocks.
631 		 */
632 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
633 			/*
634 			 * Get the summary for this level/block.
635 			 */
636 			error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
637 				&sum);
638 			if (error) {
639 				return error;
640 			}
641 			/*
642 			 * Nothing there, on to the next block.
643 			 */
644 			if (!sum)
645 				continue;
646 			/*
647 			 * Try allocating the extent.
648 			 */
649 			error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
650 				maxlen, len, &n, rbpp, rsb, prod, &r);
651 			if (error) {
652 				return error;
653 			}
654 			/*
655 			 * If it worked, return that.
656 			 */
657 			if (r != NULLRTBLOCK) {
658 				*rtblock = r;
659 				return 0;
660 			}
661 			/*
662 			 * If the "next block to try" returned from the
663 			 * allocator is beyond the next bitmap block,
664 			 * skip to that bitmap block.
665 			 */
666 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
667 				i = XFS_BITTOBLOCK(mp, n) - 1;
668 		}
669 	}
670 	/*
671 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
672 	 * we're asking for a fixed size extent.
673 	 */
674 	if (minlen > --maxlen) {
675 		*rtblock = NULLRTBLOCK;
676 		return 0;
677 	}
678 	/*
679 	 * Loop over sizes, from maxlen down to minlen.
680 	 * This time, when we do the allocations, allow smaller ones
681 	 * to succeed.
682 	 */
683 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
684 		/*
685 		 * Loop over all the bitmap blocks, try an allocation
686 		 * starting in that block.
687 		 */
688 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
689 			/*
690 			 * Get the summary information for this level/block.
691 			 */
692 			error =	xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
693 						  &sum);
694 			if (error) {
695 				return error;
696 			}
697 			/*
698 			 * If nothing there, go on to next.
699 			 */
700 			if (!sum)
701 				continue;
702 			/*
703 			 * Try the allocation.  Make sure the specified
704 			 * minlen/maxlen are in the possible range for
705 			 * this summary level.
706 			 */
707 			error = xfs_rtallocate_extent_block(mp, tp, i,
708 					XFS_RTMAX(minlen, 1 << l),
709 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
710 					len, &n, rbpp, rsb, prod, &r);
711 			if (error) {
712 				return error;
713 			}
714 			/*
715 			 * If it worked, return that extent.
716 			 */
717 			if (r != NULLRTBLOCK) {
718 				*rtblock = r;
719 				return 0;
720 			}
721 			/*
722 			 * If the "next block to try" returned from the
723 			 * allocator is beyond the next bitmap block,
724 			 * skip to that bitmap block.
725 			 */
726 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
727 				i = XFS_BITTOBLOCK(mp, n) - 1;
728 		}
729 	}
730 	/*
731 	 * Got nothing, return failure.
732 	 */
733 	*rtblock = NULLRTBLOCK;
734 	return 0;
735 }
736 
737 /*
738  * Mark an extent specified by start and len allocated.
739  * Updates all the summary information as well as the bitmap.
740  */
741 STATIC int				/* error */
742 xfs_rtallocate_range(
743 	xfs_mount_t	*mp,		/* file system mount point */
744 	xfs_trans_t	*tp,		/* transaction pointer */
745 	xfs_rtblock_t	start,		/* start block to allocate */
746 	xfs_extlen_t	len,		/* length to allocate */
747 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
748 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
749 {
750 	xfs_rtblock_t	end;		/* end of the allocated extent */
751 	int		error;		/* error value */
752 	xfs_rtblock_t	postblock;	/* first block allocated > end */
753 	xfs_rtblock_t	preblock;	/* first block allocated < start */
754 
755 	end = start + len - 1;
756 	/*
757 	 * Assume we're allocating out of the middle of a free extent.
758 	 * We need to find the beginning and end of the extent so we can
759 	 * properly update the summary.
760 	 */
761 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
762 	if (error) {
763 		return error;
764 	}
765 	/*
766 	 * Find the next allocated block (end of free extent).
767 	 */
768 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
769 		&postblock);
770 	if (error) {
771 		return error;
772 	}
773 	/*
774 	 * Decrement the summary information corresponding to the entire
775 	 * (old) free extent.
776 	 */
777 	error = xfs_rtmodify_summary(mp, tp,
778 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
779 		XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
780 	if (error) {
781 		return error;
782 	}
783 	/*
784 	 * If there are blocks not being allocated at the front of the
785 	 * old extent, add summary data for them to be free.
786 	 */
787 	if (preblock < start) {
788 		error = xfs_rtmodify_summary(mp, tp,
789 			XFS_RTBLOCKLOG(start - preblock),
790 			XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
791 		if (error) {
792 			return error;
793 		}
794 	}
795 	/*
796 	 * If there are blocks not being allocated at the end of the
797 	 * old extent, add summary data for them to be free.
798 	 */
799 	if (postblock > end) {
800 		error = xfs_rtmodify_summary(mp, tp,
801 			XFS_RTBLOCKLOG(postblock - end),
802 			XFS_BITTOBLOCK(mp, end + 1), 1, rbpp, rsb);
803 		if (error) {
804 			return error;
805 		}
806 	}
807 	/*
808 	 * Modify the bitmap to mark this extent allocated.
809 	 */
810 	error = xfs_rtmodify_range(mp, tp, start, len, 0);
811 	return error;
812 }
813 
814 /*
815  * Return whether there are any free extents in the size range given
816  * by low and high, for the bitmap block bbno.
817  */
818 STATIC int				/* error */
819 xfs_rtany_summary(
820 	xfs_mount_t	*mp,		/* file system mount structure */
821 	xfs_trans_t	*tp,		/* transaction pointer */
822 	int		low,		/* low log2 extent size */
823 	int		high,		/* high log2 extent size */
824 	xfs_rtblock_t	bbno,		/* bitmap block number */
825 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
826 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
827 	int		*stat)		/* out: any good extents here? */
828 {
829 	int		error;		/* error value */
830 	int		log;		/* loop counter, log2 of ext. size */
831 	xfs_suminfo_t	sum;		/* summary data */
832 
833 	/*
834 	 * Loop over logs of extent sizes.  Order is irrelevant.
835 	 */
836 	for (log = low; log <= high; log++) {
837 		/*
838 		 * Get one summary datum.
839 		 */
840 		error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
841 		if (error) {
842 			return error;
843 		}
844 		/*
845 		 * If there are any, return success.
846 		 */
847 		if (sum) {
848 			*stat = 1;
849 			return 0;
850 		}
851 	}
852 	/*
853 	 * Found nothing, return failure.
854 	 */
855 	*stat = 0;
856 	return 0;
857 }
858 
859 /*
860  * Get a buffer for the bitmap or summary file block specified.
861  * The buffer is returned read and locked.
862  */
863 STATIC int				/* error */
864 xfs_rtbuf_get(
865 	xfs_mount_t	*mp,		/* file system mount structure */
866 	xfs_trans_t	*tp,		/* transaction pointer */
867 	xfs_rtblock_t	block,		/* block number in bitmap or summary */
868 	int		issum,		/* is summary not bitmap */
869 	xfs_buf_t	**bpp)		/* output: buffer for the block */
870 {
871 	xfs_buf_t	*bp;		/* block buffer, result */
872 	xfs_daddr_t	d;		/* disk addr of block */
873 	int		error;		/* error value */
874 	xfs_fsblock_t	fsb;		/* fs block number for block */
875 	xfs_inode_t	*ip;		/* bitmap or summary inode */
876 
877 	ip = issum ? mp->m_rsumip : mp->m_rbmip;
878 	/*
879 	 * Map from the file offset (block) and inode number to the
880 	 * file system block.
881 	 */
882 	error = xfs_bmapi_single(tp, ip, XFS_DATA_FORK, &fsb, block);
883 	if (error) {
884 		return error;
885 	}
886 	ASSERT(fsb != NULLFSBLOCK);
887 	/*
888 	 * Convert to disk address for buffer cache.
889 	 */
890 	d = XFS_FSB_TO_DADDR(mp, fsb);
891 	/*
892 	 * Read the buffer.
893 	 */
894 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
895 				   mp->m_bsize, 0, &bp);
896 	if (error) {
897 		return error;
898 	}
899 	ASSERT(bp && !XFS_BUF_GETERROR(bp));
900 	*bpp = bp;
901 	return 0;
902 }
903 
904 #ifdef DEBUG
905 /*
906  * Check that the given extent (block range) is allocated already.
907  */
908 STATIC int				/* error */
909 xfs_rtcheck_alloc_range(
910 	xfs_mount_t	*mp,		/* file system mount point */
911 	xfs_trans_t	*tp,		/* transaction pointer */
912 	xfs_rtblock_t	bno,		/* starting block number of extent */
913 	xfs_extlen_t	len,		/* length of extent */
914 	int		*stat)		/* out: 1 for allocated, 0 for not */
915 {
916 	xfs_rtblock_t	new;		/* dummy for xfs_rtcheck_range */
917 
918 	return xfs_rtcheck_range(mp, tp, bno, len, 0, &new, stat);
919 }
920 #endif
921 
922 /*
923  * Check that the given range is either all allocated (val = 0) or
924  * all free (val = 1).
925  */
926 STATIC int				/* error */
927 xfs_rtcheck_range(
928 	xfs_mount_t	*mp,		/* file system mount point */
929 	xfs_trans_t	*tp,		/* transaction pointer */
930 	xfs_rtblock_t	start,		/* starting block number of extent */
931 	xfs_extlen_t	len,		/* length of extent */
932 	int		val,		/* 1 for free, 0 for allocated */
933 	xfs_rtblock_t	*new,		/* out: first block not matching */
934 	int		*stat)		/* out: 1 for matches, 0 for not */
935 {
936 	xfs_rtword_t	*b;		/* current word in buffer */
937 	int		bit;		/* bit number in the word */
938 	xfs_rtblock_t	block;		/* bitmap block number */
939 	xfs_buf_t	*bp;		/* buf for the block */
940 	xfs_rtword_t	*bufp;		/* starting word in buffer */
941 	int		error;		/* error value */
942 	xfs_rtblock_t	i;		/* current bit number rel. to start */
943 	xfs_rtblock_t	lastbit;	/* last useful bit in word */
944 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
945 	xfs_rtword_t	wdiff;		/* difference from wanted value */
946 	int		word;		/* word number in the buffer */
947 
948 	/*
949 	 * Compute starting bitmap block number
950 	 */
951 	block = XFS_BITTOBLOCK(mp, start);
952 	/*
953 	 * Read the bitmap block.
954 	 */
955 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
956 	if (error) {
957 		return error;
958 	}
959 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
960 	/*
961 	 * Compute the starting word's address, and starting bit.
962 	 */
963 	word = XFS_BITTOWORD(mp, start);
964 	b = &bufp[word];
965 	bit = (int)(start & (XFS_NBWORD - 1));
966 	/*
967 	 * 0 (allocated) => all zero's; 1 (free) => all one's.
968 	 */
969 	val = -val;
970 	/*
971 	 * If not starting on a word boundary, deal with the first
972 	 * (partial) word.
973 	 */
974 	if (bit) {
975 		/*
976 		 * Compute first bit not examined.
977 		 */
978 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
979 		/*
980 		 * Mask of relevant bits.
981 		 */
982 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
983 		/*
984 		 * Compute difference between actual and desired value.
985 		 */
986 		if ((wdiff = (*b ^ val) & mask)) {
987 			/*
988 			 * Different, compute first wrong bit and return.
989 			 */
990 			xfs_trans_brelse(tp, bp);
991 			i = XFS_RTLOBIT(wdiff) - bit;
992 			*new = start + i;
993 			*stat = 0;
994 			return 0;
995 		}
996 		i = lastbit - bit;
997 		/*
998 		 * Go on to next block if that's where the next word is
999 		 * and we need the next word.
1000 		 */
1001 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1002 			/*
1003 			 * If done with this block, get the next one.
1004 			 */
1005 			xfs_trans_brelse(tp, bp);
1006 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1007 			if (error) {
1008 				return error;
1009 			}
1010 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1011 			word = 0;
1012 		} else {
1013 			/*
1014 			 * Go on to the next word in the buffer.
1015 			 */
1016 			b++;
1017 		}
1018 	} else {
1019 		/*
1020 		 * Starting on a word boundary, no partial word.
1021 		 */
1022 		i = 0;
1023 	}
1024 	/*
1025 	 * Loop over whole words in buffers.  When we use up one buffer
1026 	 * we move on to the next one.
1027 	 */
1028 	while (len - i >= XFS_NBWORD) {
1029 		/*
1030 		 * Compute difference between actual and desired value.
1031 		 */
1032 		if ((wdiff = *b ^ val)) {
1033 			/*
1034 			 * Different, compute first wrong bit and return.
1035 			 */
1036 			xfs_trans_brelse(tp, bp);
1037 			i += XFS_RTLOBIT(wdiff);
1038 			*new = start + i;
1039 			*stat = 0;
1040 			return 0;
1041 		}
1042 		i += XFS_NBWORD;
1043 		/*
1044 		 * Go on to next block if that's where the next word is
1045 		 * and we need the next word.
1046 		 */
1047 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1048 			/*
1049 			 * If done with this block, get the next one.
1050 			 */
1051 			xfs_trans_brelse(tp, bp);
1052 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1053 			if (error) {
1054 				return error;
1055 			}
1056 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1057 			word = 0;
1058 		} else {
1059 			/*
1060 			 * Go on to the next word in the buffer.
1061 			 */
1062 			b++;
1063 		}
1064 	}
1065 	/*
1066 	 * If not ending on a word boundary, deal with the last
1067 	 * (partial) word.
1068 	 */
1069 	if ((lastbit = len - i)) {
1070 		/*
1071 		 * Mask of relevant bits.
1072 		 */
1073 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1074 		/*
1075 		 * Compute difference between actual and desired value.
1076 		 */
1077 		if ((wdiff = (*b ^ val) & mask)) {
1078 			/*
1079 			 * Different, compute first wrong bit and return.
1080 			 */
1081 			xfs_trans_brelse(tp, bp);
1082 			i += XFS_RTLOBIT(wdiff);
1083 			*new = start + i;
1084 			*stat = 0;
1085 			return 0;
1086 		} else
1087 			i = len;
1088 	}
1089 	/*
1090 	 * Successful, return.
1091 	 */
1092 	xfs_trans_brelse(tp, bp);
1093 	*new = start + i;
1094 	*stat = 1;
1095 	return 0;
1096 }
1097 
1098 /*
1099  * Copy and transform the summary file, given the old and new
1100  * parameters in the mount structures.
1101  */
1102 STATIC int				/* error */
1103 xfs_rtcopy_summary(
1104 	xfs_mount_t	*omp,		/* old file system mount point */
1105 	xfs_mount_t	*nmp,		/* new file system mount point */
1106 	xfs_trans_t	*tp)		/* transaction pointer */
1107 {
1108 	xfs_rtblock_t	bbno;		/* bitmap block number */
1109 	xfs_buf_t	*bp;		/* summary buffer */
1110 	int		error;		/* error return value */
1111 	int		log;		/* summary level number (log length) */
1112 	xfs_suminfo_t	sum;		/* summary data */
1113 	xfs_fsblock_t	sumbno;		/* summary block number */
1114 
1115 	bp = NULL;
1116 	for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
1117 		for (bbno = omp->m_sb.sb_rbmblocks - 1;
1118 		     (xfs_srtblock_t)bbno >= 0;
1119 		     bbno--) {
1120 			error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
1121 				&sumbno, &sum);
1122 			if (error)
1123 				return error;
1124 			if (sum == 0)
1125 				continue;
1126 			error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
1127 				&bp, &sumbno);
1128 			if (error)
1129 				return error;
1130 			error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
1131 				&bp, &sumbno);
1132 			if (error)
1133 				return error;
1134 			ASSERT(sum > 0);
1135 		}
1136 	}
1137 	return 0;
1138 }
1139 
1140 /*
1141  * Searching backward from start to limit, find the first block whose
1142  * allocated/free state is different from start's.
1143  */
1144 STATIC int				/* error */
1145 xfs_rtfind_back(
1146 	xfs_mount_t	*mp,		/* file system mount point */
1147 	xfs_trans_t	*tp,		/* transaction pointer */
1148 	xfs_rtblock_t	start,		/* starting block to look at */
1149 	xfs_rtblock_t	limit,		/* last block to look at */
1150 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1151 {
1152 	xfs_rtword_t	*b;		/* current word in buffer */
1153 	int		bit;		/* bit number in the word */
1154 	xfs_rtblock_t	block;		/* bitmap block number */
1155 	xfs_buf_t	*bp;		/* buf for the block */
1156 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1157 	int		error;		/* error value */
1158 	xfs_rtblock_t	firstbit;	/* first useful bit in the word */
1159 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1160 	xfs_rtblock_t	len;		/* length of inspected area */
1161 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1162 	xfs_rtword_t	want;		/* mask for "good" values */
1163 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1164 	int		word;		/* word number in the buffer */
1165 
1166 	/*
1167 	 * Compute and read in starting bitmap block for starting block.
1168 	 */
1169 	block = XFS_BITTOBLOCK(mp, start);
1170 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1171 	if (error) {
1172 		return error;
1173 	}
1174 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1175 	/*
1176 	 * Get the first word's index & point to it.
1177 	 */
1178 	word = XFS_BITTOWORD(mp, start);
1179 	b = &bufp[word];
1180 	bit = (int)(start & (XFS_NBWORD - 1));
1181 	len = start - limit + 1;
1182 	/*
1183 	 * Compute match value, based on the bit at start: if 1 (free)
1184 	 * then all-ones, else all-zeroes.
1185 	 */
1186 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1187 	/*
1188 	 * If the starting position is not word-aligned, deal with the
1189 	 * partial word.
1190 	 */
1191 	if (bit < XFS_NBWORD - 1) {
1192 		/*
1193 		 * Calculate first (leftmost) bit number to look at,
1194 		 * and mask for all the relevant bits in this word.
1195 		 */
1196 		firstbit = XFS_RTMAX((xfs_srtblock_t)(bit - len + 1), 0);
1197 		mask = (((xfs_rtword_t)1 << (bit - firstbit + 1)) - 1) <<
1198 			firstbit;
1199 		/*
1200 		 * Calculate the difference between the value there
1201 		 * and what we're looking for.
1202 		 */
1203 		if ((wdiff = (*b ^ want) & mask)) {
1204 			/*
1205 			 * Different.  Mark where we are and return.
1206 			 */
1207 			xfs_trans_brelse(tp, bp);
1208 			i = bit - XFS_RTHIBIT(wdiff);
1209 			*rtblock = start - i + 1;
1210 			return 0;
1211 		}
1212 		i = bit - firstbit + 1;
1213 		/*
1214 		 * Go on to previous block if that's where the previous word is
1215 		 * and we need the previous word.
1216 		 */
1217 		if (--word == -1 && i < len) {
1218 			/*
1219 			 * If done with this block, get the previous one.
1220 			 */
1221 			xfs_trans_brelse(tp, bp);
1222 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1223 			if (error) {
1224 				return error;
1225 			}
1226 			bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1227 			word = XFS_BLOCKWMASK(mp);
1228 			b = &bufp[word];
1229 		} else {
1230 			/*
1231 			 * Go on to the previous word in the buffer.
1232 			 */
1233 			b--;
1234 		}
1235 	} else {
1236 		/*
1237 		 * Starting on a word boundary, no partial word.
1238 		 */
1239 		i = 0;
1240 	}
1241 	/*
1242 	 * Loop over whole words in buffers.  When we use up one buffer
1243 	 * we move on to the previous one.
1244 	 */
1245 	while (len - i >= XFS_NBWORD) {
1246 		/*
1247 		 * Compute difference between actual and desired value.
1248 		 */
1249 		if ((wdiff = *b ^ want)) {
1250 			/*
1251 			 * Different, mark where we are and return.
1252 			 */
1253 			xfs_trans_brelse(tp, bp);
1254 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1255 			*rtblock = start - i + 1;
1256 			return 0;
1257 		}
1258 		i += XFS_NBWORD;
1259 		/*
1260 		 * Go on to previous block if that's where the previous word is
1261 		 * and we need the previous word.
1262 		 */
1263 		if (--word == -1 && i < len) {
1264 			/*
1265 			 * If done with this block, get the previous one.
1266 			 */
1267 			xfs_trans_brelse(tp, bp);
1268 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1269 			if (error) {
1270 				return error;
1271 			}
1272 			bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1273 			word = XFS_BLOCKWMASK(mp);
1274 			b = &bufp[word];
1275 		} else {
1276 			/*
1277 			 * Go on to the previous word in the buffer.
1278 			 */
1279 			b--;
1280 		}
1281 	}
1282 	/*
1283 	 * If not ending on a word boundary, deal with the last
1284 	 * (partial) word.
1285 	 */
1286 	if (len - i) {
1287 		/*
1288 		 * Calculate first (leftmost) bit number to look at,
1289 		 * and mask for all the relevant bits in this word.
1290 		 */
1291 		firstbit = XFS_NBWORD - (len - i);
1292 		mask = (((xfs_rtword_t)1 << (len - i)) - 1) << firstbit;
1293 		/*
1294 		 * Compute difference between actual and desired value.
1295 		 */
1296 		if ((wdiff = (*b ^ want) & mask)) {
1297 			/*
1298 			 * Different, mark where we are and return.
1299 			 */
1300 			xfs_trans_brelse(tp, bp);
1301 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1302 			*rtblock = start - i + 1;
1303 			return 0;
1304 		} else
1305 			i = len;
1306 	}
1307 	/*
1308 	 * No match, return that we scanned the whole area.
1309 	 */
1310 	xfs_trans_brelse(tp, bp);
1311 	*rtblock = start - i + 1;
1312 	return 0;
1313 }
1314 
1315 /*
1316  * Searching forward from start to limit, find the first block whose
1317  * allocated/free state is different from start's.
1318  */
1319 STATIC int				/* error */
1320 xfs_rtfind_forw(
1321 	xfs_mount_t	*mp,		/* file system mount point */
1322 	xfs_trans_t	*tp,		/* transaction pointer */
1323 	xfs_rtblock_t	start,		/* starting block to look at */
1324 	xfs_rtblock_t	limit,		/* last block to look at */
1325 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1326 {
1327 	xfs_rtword_t	*b;		/* current word in buffer */
1328 	int		bit;		/* bit number in the word */
1329 	xfs_rtblock_t	block;		/* bitmap block number */
1330 	xfs_buf_t	*bp;		/* buf for the block */
1331 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1332 	int		error;		/* error value */
1333 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1334 	xfs_rtblock_t	lastbit;	/* last useful bit in the word */
1335 	xfs_rtblock_t	len;		/* length of inspected area */
1336 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1337 	xfs_rtword_t	want;		/* mask for "good" values */
1338 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1339 	int		word;		/* word number in the buffer */
1340 
1341 	/*
1342 	 * Compute and read in starting bitmap block for starting block.
1343 	 */
1344 	block = XFS_BITTOBLOCK(mp, start);
1345 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1346 	if (error) {
1347 		return error;
1348 	}
1349 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1350 	/*
1351 	 * Get the first word's index & point to it.
1352 	 */
1353 	word = XFS_BITTOWORD(mp, start);
1354 	b = &bufp[word];
1355 	bit = (int)(start & (XFS_NBWORD - 1));
1356 	len = limit - start + 1;
1357 	/*
1358 	 * Compute match value, based on the bit at start: if 1 (free)
1359 	 * then all-ones, else all-zeroes.
1360 	 */
1361 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1362 	/*
1363 	 * If the starting position is not word-aligned, deal with the
1364 	 * partial word.
1365 	 */
1366 	if (bit) {
1367 		/*
1368 		 * Calculate last (rightmost) bit number to look at,
1369 		 * and mask for all the relevant bits in this word.
1370 		 */
1371 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1372 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1373 		/*
1374 		 * Calculate the difference between the value there
1375 		 * and what we're looking for.
1376 		 */
1377 		if ((wdiff = (*b ^ want) & mask)) {
1378 			/*
1379 			 * Different.  Mark where we are and return.
1380 			 */
1381 			xfs_trans_brelse(tp, bp);
1382 			i = XFS_RTLOBIT(wdiff) - bit;
1383 			*rtblock = start + i - 1;
1384 			return 0;
1385 		}
1386 		i = lastbit - bit;
1387 		/*
1388 		 * Go on to next block if that's where the next word is
1389 		 * and we need the next word.
1390 		 */
1391 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1392 			/*
1393 			 * If done with this block, get the previous one.
1394 			 */
1395 			xfs_trans_brelse(tp, bp);
1396 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1397 			if (error) {
1398 				return error;
1399 			}
1400 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1401 			word = 0;
1402 		} else {
1403 			/*
1404 			 * Go on to the previous word in the buffer.
1405 			 */
1406 			b++;
1407 		}
1408 	} else {
1409 		/*
1410 		 * Starting on a word boundary, no partial word.
1411 		 */
1412 		i = 0;
1413 	}
1414 	/*
1415 	 * Loop over whole words in buffers.  When we use up one buffer
1416 	 * we move on to the next one.
1417 	 */
1418 	while (len - i >= XFS_NBWORD) {
1419 		/*
1420 		 * Compute difference between actual and desired value.
1421 		 */
1422 		if ((wdiff = *b ^ want)) {
1423 			/*
1424 			 * Different, mark where we are and return.
1425 			 */
1426 			xfs_trans_brelse(tp, bp);
1427 			i += XFS_RTLOBIT(wdiff);
1428 			*rtblock = start + i - 1;
1429 			return 0;
1430 		}
1431 		i += XFS_NBWORD;
1432 		/*
1433 		 * Go on to next block if that's where the next word is
1434 		 * and we need the next word.
1435 		 */
1436 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1437 			/*
1438 			 * If done with this block, get the next one.
1439 			 */
1440 			xfs_trans_brelse(tp, bp);
1441 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1442 			if (error) {
1443 				return error;
1444 			}
1445 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1446 			word = 0;
1447 		} else {
1448 			/*
1449 			 * Go on to the next word in the buffer.
1450 			 */
1451 			b++;
1452 		}
1453 	}
1454 	/*
1455 	 * If not ending on a word boundary, deal with the last
1456 	 * (partial) word.
1457 	 */
1458 	if ((lastbit = len - i)) {
1459 		/*
1460 		 * Calculate mask for all the relevant bits in this word.
1461 		 */
1462 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1463 		/*
1464 		 * Compute difference between actual and desired value.
1465 		 */
1466 		if ((wdiff = (*b ^ want) & mask)) {
1467 			/*
1468 			 * Different, mark where we are and return.
1469 			 */
1470 			xfs_trans_brelse(tp, bp);
1471 			i += XFS_RTLOBIT(wdiff);
1472 			*rtblock = start + i - 1;
1473 			return 0;
1474 		} else
1475 			i = len;
1476 	}
1477 	/*
1478 	 * No match, return that we scanned the whole area.
1479 	 */
1480 	xfs_trans_brelse(tp, bp);
1481 	*rtblock = start + i - 1;
1482 	return 0;
1483 }
1484 
1485 /*
1486  * Mark an extent specified by start and len freed.
1487  * Updates all the summary information as well as the bitmap.
1488  */
1489 STATIC int				/* error */
1490 xfs_rtfree_range(
1491 	xfs_mount_t	*mp,		/* file system mount point */
1492 	xfs_trans_t	*tp,		/* transaction pointer */
1493 	xfs_rtblock_t	start,		/* starting block to free */
1494 	xfs_extlen_t	len,		/* length to free */
1495 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1496 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1497 {
1498 	xfs_rtblock_t	end;		/* end of the freed extent */
1499 	int		error;		/* error value */
1500 	xfs_rtblock_t	postblock;	/* first block freed > end */
1501 	xfs_rtblock_t	preblock;	/* first block freed < start */
1502 
1503 	end = start + len - 1;
1504 	/*
1505 	 * Modify the bitmap to mark this extent freed.
1506 	 */
1507 	error = xfs_rtmodify_range(mp, tp, start, len, 1);
1508 	if (error) {
1509 		return error;
1510 	}
1511 	/*
1512 	 * Assume we're freeing out of the middle of an allocated extent.
1513 	 * We need to find the beginning and end of the extent so we can
1514 	 * properly update the summary.
1515 	 */
1516 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
1517 	if (error) {
1518 		return error;
1519 	}
1520 	/*
1521 	 * Find the next allocated block (end of allocated extent).
1522 	 */
1523 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
1524 		&postblock);
1525 	/*
1526 	 * If there are blocks not being freed at the front of the
1527 	 * old extent, add summary data for them to be allocated.
1528 	 */
1529 	if (preblock < start) {
1530 		error = xfs_rtmodify_summary(mp, tp,
1531 			XFS_RTBLOCKLOG(start - preblock),
1532 			XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
1533 		if (error) {
1534 			return error;
1535 		}
1536 	}
1537 	/*
1538 	 * If there are blocks not being freed at the end of the
1539 	 * old extent, add summary data for them to be allocated.
1540 	 */
1541 	if (postblock > end) {
1542 		error = xfs_rtmodify_summary(mp, tp,
1543 			XFS_RTBLOCKLOG(postblock - end),
1544 			XFS_BITTOBLOCK(mp, end + 1), -1, rbpp, rsb);
1545 		if (error) {
1546 			return error;
1547 		}
1548 	}
1549 	/*
1550 	 * Increment the summary information corresponding to the entire
1551 	 * (new) free extent.
1552 	 */
1553 	error = xfs_rtmodify_summary(mp, tp,
1554 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
1555 		XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
1556 	return error;
1557 }
1558 
1559 /*
1560  * Read and return the summary information for a given extent size,
1561  * bitmap block combination.
1562  * Keeps track of a current summary block, so we don't keep reading
1563  * it from the buffer cache.
1564  */
1565 STATIC int				/* error */
1566 xfs_rtget_summary(
1567 	xfs_mount_t	*mp,		/* file system mount structure */
1568 	xfs_trans_t	*tp,		/* transaction pointer */
1569 	int		log,		/* log2 of extent size */
1570 	xfs_rtblock_t	bbno,		/* bitmap block number */
1571 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1572 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
1573 	xfs_suminfo_t	*sum)		/* out: summary info for this block */
1574 {
1575 	xfs_buf_t	*bp;		/* buffer for summary block */
1576 	int		error;		/* error value */
1577 	xfs_fsblock_t	sb;		/* summary fsblock */
1578 	int		so;		/* index into the summary file */
1579 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1580 
1581 	/*
1582 	 * Compute entry number in the summary file.
1583 	 */
1584 	so = XFS_SUMOFFS(mp, log, bbno);
1585 	/*
1586 	 * Compute the block number in the summary file.
1587 	 */
1588 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1589 	/*
1590 	 * If we have an old buffer, and the block number matches, use that.
1591 	 */
1592 	if (rbpp && *rbpp && *rsb == sb)
1593 		bp = *rbpp;
1594 	/*
1595 	 * Otherwise we have to get the buffer.
1596 	 */
1597 	else {
1598 		/*
1599 		 * If there was an old one, get rid of it first.
1600 		 */
1601 		if (rbpp && *rbpp)
1602 			xfs_trans_brelse(tp, *rbpp);
1603 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1604 		if (error) {
1605 			return error;
1606 		}
1607 		/*
1608 		 * Remember this buffer and block for the next call.
1609 		 */
1610 		if (rbpp) {
1611 			*rbpp = bp;
1612 			*rsb = sb;
1613 		}
1614 	}
1615 	/*
1616 	 * Point to the summary information & copy it out.
1617 	 */
1618 	sp = XFS_SUMPTR(mp, bp, so);
1619 	*sum = *sp;
1620 	/*
1621 	 * Drop the buffer if we're not asked to remember it.
1622 	 */
1623 	if (!rbpp)
1624 		xfs_trans_brelse(tp, bp);
1625 	return 0;
1626 }
1627 
1628 /*
1629  * Set the given range of bitmap bits to the given value.
1630  * Do whatever I/O and logging is required.
1631  */
1632 STATIC int				/* error */
1633 xfs_rtmodify_range(
1634 	xfs_mount_t	*mp,		/* file system mount point */
1635 	xfs_trans_t	*tp,		/* transaction pointer */
1636 	xfs_rtblock_t	start,		/* starting block to modify */
1637 	xfs_extlen_t	len,		/* length of extent to modify */
1638 	int		val)		/* 1 for free, 0 for allocated */
1639 {
1640 	xfs_rtword_t	*b;		/* current word in buffer */
1641 	int		bit;		/* bit number in the word */
1642 	xfs_rtblock_t	block;		/* bitmap block number */
1643 	xfs_buf_t	*bp;		/* buf for the block */
1644 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1645 	int		error;		/* error value */
1646 	xfs_rtword_t	*first;		/* first used word in the buffer */
1647 	int		i;		/* current bit number rel. to start */
1648 	int		lastbit;	/* last useful bit in word */
1649 	xfs_rtword_t	mask;		/* mask o frelevant bits for value */
1650 	int		word;		/* word number in the buffer */
1651 
1652 	/*
1653 	 * Compute starting bitmap block number.
1654 	 */
1655 	block = XFS_BITTOBLOCK(mp, start);
1656 	/*
1657 	 * Read the bitmap block, and point to its data.
1658 	 */
1659 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1660 	if (error) {
1661 		return error;
1662 	}
1663 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1664 	/*
1665 	 * Compute the starting word's address, and starting bit.
1666 	 */
1667 	word = XFS_BITTOWORD(mp, start);
1668 	first = b = &bufp[word];
1669 	bit = (int)(start & (XFS_NBWORD - 1));
1670 	/*
1671 	 * 0 (allocated) => all zeroes; 1 (free) => all ones.
1672 	 */
1673 	val = -val;
1674 	/*
1675 	 * If not starting on a word boundary, deal with the first
1676 	 * (partial) word.
1677 	 */
1678 	if (bit) {
1679 		/*
1680 		 * Compute first bit not changed and mask of relevant bits.
1681 		 */
1682 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1683 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1684 		/*
1685 		 * Set/clear the active bits.
1686 		 */
1687 		if (val)
1688 			*b |= mask;
1689 		else
1690 			*b &= ~mask;
1691 		i = lastbit - bit;
1692 		/*
1693 		 * Go on to the next block if that's where the next word is
1694 		 * and we need the next word.
1695 		 */
1696 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1697 			/*
1698 			 * Log the changed part of this block.
1699 			 * Get the next one.
1700 			 */
1701 			xfs_trans_log_buf(tp, bp,
1702 				(uint)((char *)first - (char *)bufp),
1703 				(uint)((char *)b - (char *)bufp));
1704 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1705 			if (error) {
1706 				return error;
1707 			}
1708 			first = b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1709 			word = 0;
1710 		} else {
1711 			/*
1712 			 * Go on to the next word in the buffer
1713 			 */
1714 			b++;
1715 		}
1716 	} else {
1717 		/*
1718 		 * Starting on a word boundary, no partial word.
1719 		 */
1720 		i = 0;
1721 	}
1722 	/*
1723 	 * Loop over whole words in buffers.  When we use up one buffer
1724 	 * we move on to the next one.
1725 	 */
1726 	while (len - i >= XFS_NBWORD) {
1727 		/*
1728 		 * Set the word value correctly.
1729 		 */
1730 		*b = val;
1731 		i += XFS_NBWORD;
1732 		/*
1733 		 * Go on to the next block if that's where the next word is
1734 		 * and we need the next word.
1735 		 */
1736 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1737 			/*
1738 			 * Log the changed part of this block.
1739 			 * Get the next one.
1740 			 */
1741 			xfs_trans_log_buf(tp, bp,
1742 				(uint)((char *)first - (char *)bufp),
1743 				(uint)((char *)b - (char *)bufp));
1744 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1745 			if (error) {
1746 				return error;
1747 			}
1748 			first = b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1749 			word = 0;
1750 		} else {
1751 			/*
1752 			 * Go on to the next word in the buffer
1753 			 */
1754 			b++;
1755 		}
1756 	}
1757 	/*
1758 	 * If not ending on a word boundary, deal with the last
1759 	 * (partial) word.
1760 	 */
1761 	if ((lastbit = len - i)) {
1762 		/*
1763 		 * Compute a mask of relevant bits.
1764 		 */
1765 		bit = 0;
1766 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1767 		/*
1768 		 * Set/clear the active bits.
1769 		 */
1770 		if (val)
1771 			*b |= mask;
1772 		else
1773 			*b &= ~mask;
1774 		b++;
1775 	}
1776 	/*
1777 	 * Log any remaining changed bytes.
1778 	 */
1779 	if (b > first)
1780 		xfs_trans_log_buf(tp, bp, (uint)((char *)first - (char *)bufp),
1781 			(uint)((char *)b - (char *)bufp - 1));
1782 	return 0;
1783 }
1784 
1785 /*
1786  * Read and modify the summary information for a given extent size,
1787  * bitmap block combination.
1788  * Keeps track of a current summary block, so we don't keep reading
1789  * it from the buffer cache.
1790  */
1791 STATIC int				/* error */
1792 xfs_rtmodify_summary(
1793 	xfs_mount_t	*mp,		/* file system mount point */
1794 	xfs_trans_t	*tp,		/* transaction pointer */
1795 	int		log,		/* log2 of extent size */
1796 	xfs_rtblock_t	bbno,		/* bitmap block number */
1797 	int		delta,		/* change to make to summary info */
1798 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1799 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1800 {
1801 	xfs_buf_t	*bp;		/* buffer for the summary block */
1802 	int		error;		/* error value */
1803 	xfs_fsblock_t	sb;		/* summary fsblock */
1804 	int		so;		/* index into the summary file */
1805 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1806 
1807 	/*
1808 	 * Compute entry number in the summary file.
1809 	 */
1810 	so = XFS_SUMOFFS(mp, log, bbno);
1811 	/*
1812 	 * Compute the block number in the summary file.
1813 	 */
1814 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1815 	/*
1816 	 * If we have an old buffer, and the block number matches, use that.
1817 	 */
1818 	if (rbpp && *rbpp && *rsb == sb)
1819 		bp = *rbpp;
1820 	/*
1821 	 * Otherwise we have to get the buffer.
1822 	 */
1823 	else {
1824 		/*
1825 		 * If there was an old one, get rid of it first.
1826 		 */
1827 		if (rbpp && *rbpp)
1828 			xfs_trans_brelse(tp, *rbpp);
1829 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1830 		if (error) {
1831 			return error;
1832 		}
1833 		/*
1834 		 * Remember this buffer and block for the next call.
1835 		 */
1836 		if (rbpp) {
1837 			*rbpp = bp;
1838 			*rsb = sb;
1839 		}
1840 	}
1841 	/*
1842 	 * Point to the summary information, modify and log it.
1843 	 */
1844 	sp = XFS_SUMPTR(mp, bp, so);
1845 	*sp += delta;
1846 	xfs_trans_log_buf(tp, bp, (uint)((char *)sp - (char *)XFS_BUF_PTR(bp)),
1847 		(uint)((char *)sp - (char *)XFS_BUF_PTR(bp) + sizeof(*sp) - 1));
1848 	return 0;
1849 }
1850 
1851 /*
1852  * Visible (exported) functions.
1853  */
1854 
1855 /*
1856  * Grow the realtime area of the filesystem.
1857  */
1858 int
1859 xfs_growfs_rt(
1860 	xfs_mount_t	*mp,		/* mount point for filesystem */
1861 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
1862 {
1863 	xfs_rtblock_t	bmbno;		/* bitmap block number */
1864 	xfs_buf_t	*bp;		/* temporary buffer */
1865 	int		cancelflags;	/* flags for xfs_trans_cancel */
1866 	int		error;		/* error return value */
1867 	xfs_inode_t	*ip;		/* bitmap inode, used as lock */
1868 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
1869 	xfs_drfsbno_t	nrblocks;	/* new number of realtime blocks */
1870 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
1871 	xfs_drtbno_t	nrextents;	/* new number of realtime extents */
1872 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
1873 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
1874 	uint		nrsumlevels;	/* new rt summary levels */
1875 	uint		nrsumsize;	/* new size of rt summary, bytes */
1876 	xfs_sb_t	*nsbp;		/* new superblock */
1877 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
1878 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
1879 	xfs_sb_t	*sbp;		/* old superblock */
1880 	xfs_fsblock_t	sumbno;		/* summary block number */
1881 	xfs_trans_t	*tp;		/* transaction pointer */
1882 
1883 	sbp = &mp->m_sb;
1884 	cancelflags = 0;
1885 	/*
1886 	 * Initial error checking.
1887 	 */
1888 	if (mp->m_rtdev_targp == NULL || mp->m_rbmip == NULL ||
1889 	    (nrblocks = in->newblocks) <= sbp->sb_rblocks ||
1890 	    (sbp->sb_rblocks && (in->extsize != sbp->sb_rextsize)))
1891 		return XFS_ERROR(EINVAL);
1892 	if ((error = xfs_sb_validate_fsb_count(sbp, nrblocks)))
1893 		return error;
1894 	/*
1895 	 * Read in the last block of the device, make sure it exists.
1896 	 */
1897 	error = xfs_read_buf(mp, mp->m_rtdev_targp,
1898 			XFS_FSB_TO_BB(mp, nrblocks - 1),
1899 			XFS_FSB_TO_BB(mp, 1), 0, &bp);
1900 	if (error)
1901 		return error;
1902 	ASSERT(bp);
1903 	xfs_buf_relse(bp);
1904 	/*
1905 	 * Calculate new parameters.  These are the final values to be reached.
1906 	 */
1907 	nrextents = nrblocks;
1908 	do_div(nrextents, in->extsize);
1909 	nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
1910 	nrextslog = xfs_highbit32(nrextents);
1911 	nrsumlevels = nrextslog + 1;
1912 	nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
1913 	nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1914 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1915 	/*
1916 	 * New summary size can't be more than half the size of
1917 	 * the log.  This prevents us from getting a log overflow,
1918 	 * since we'll log basically the whole summary file at once.
1919 	 */
1920 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
1921 		return XFS_ERROR(EINVAL);
1922 	/*
1923 	 * Get the old block counts for bitmap and summary inodes.
1924 	 * These can't change since other growfs callers are locked out.
1925 	 */
1926 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_d.di_size);
1927 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_d.di_size);
1928 	/*
1929 	 * Allocate space to the bitmap and summary files, as necessary.
1930 	 */
1931 	if ((error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks,
1932 			mp->m_sb.sb_rbmino)))
1933 		return error;
1934 	if ((error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks,
1935 			mp->m_sb.sb_rsumino)))
1936 		return error;
1937 	/*
1938 	 * Allocate a new (fake) mount/sb.
1939 	 */
1940 	nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
1941 	/*
1942 	 * Loop over the bitmap blocks.
1943 	 * We will do everything one bitmap block at a time.
1944 	 * Skip the current block if it is exactly full.
1945 	 * This also deals with the case where there were no rtextents before.
1946 	 */
1947 	for (bmbno = sbp->sb_rbmblocks -
1948 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1949 	     bmbno < nrbmblocks;
1950 	     bmbno++) {
1951 		*nmp = *mp;
1952 		nsbp = &nmp->m_sb;
1953 		/*
1954 		 * Calculate new sb and mount fields for this round.
1955 		 */
1956 		nsbp->sb_rextsize = in->extsize;
1957 		nsbp->sb_rbmblocks = bmbno + 1;
1958 		nsbp->sb_rblocks =
1959 			XFS_RTMIN(nrblocks,
1960 				  nsbp->sb_rbmblocks * NBBY *
1961 				  nsbp->sb_blocksize * nsbp->sb_rextsize);
1962 		nsbp->sb_rextents = nsbp->sb_rblocks;
1963 		do_div(nsbp->sb_rextents, nsbp->sb_rextsize);
1964 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1965 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1966 		nrsumsize =
1967 			(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
1968 			nsbp->sb_rbmblocks;
1969 		nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1970 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1971 		/*
1972 		 * Start a transaction, get the log reservation.
1973 		 */
1974 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_FREE);
1975 		cancelflags = 0;
1976 		if ((error = xfs_trans_reserve(tp, 0,
1977 				XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
1978 			break;
1979 		/*
1980 		 * Lock out other callers by grabbing the bitmap inode lock.
1981 		 */
1982 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
1983 						XFS_ILOCK_EXCL, &ip)))
1984 			break;
1985 		ASSERT(ip == mp->m_rbmip);
1986 		/*
1987 		 * Update the bitmap inode's size.
1988 		 */
1989 		mp->m_rbmip->i_d.di_size =
1990 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1991 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1992 		cancelflags |= XFS_TRANS_ABORT;
1993 		/*
1994 		 * Get the summary inode into the transaction.
1995 		 */
1996 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rsumino, 0,
1997 						XFS_ILOCK_EXCL, &ip)))
1998 			break;
1999 		ASSERT(ip == mp->m_rsumip);
2000 		/*
2001 		 * Update the summary inode's size.
2002 		 */
2003 		mp->m_rsumip->i_d.di_size = nmp->m_rsumsize;
2004 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
2005 		/*
2006 		 * Copy summary data from old to new sizes.
2007 		 * Do this when the real size (not block-aligned) changes.
2008 		 */
2009 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
2010 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
2011 			error = xfs_rtcopy_summary(mp, nmp, tp);
2012 			if (error)
2013 				break;
2014 		}
2015 		/*
2016 		 * Update superblock fields.
2017 		 */
2018 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
2019 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
2020 				nsbp->sb_rextsize - sbp->sb_rextsize);
2021 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
2022 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
2023 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
2024 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
2025 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
2026 				nsbp->sb_rblocks - sbp->sb_rblocks);
2027 		if (nsbp->sb_rextents != sbp->sb_rextents)
2028 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
2029 				nsbp->sb_rextents - sbp->sb_rextents);
2030 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
2031 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
2032 				nsbp->sb_rextslog - sbp->sb_rextslog);
2033 		/*
2034 		 * Free new extent.
2035 		 */
2036 		bp = NULL;
2037 		error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
2038 			nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
2039 		if (error)
2040 			break;
2041 		/*
2042 		 * Mark more blocks free in the superblock.
2043 		 */
2044 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
2045 			nsbp->sb_rextents - sbp->sb_rextents);
2046 		/*
2047 		 * Update mp values into the real mp structure.
2048 		 */
2049 		mp->m_rsumlevels = nrsumlevels;
2050 		mp->m_rsumsize = nrsumsize;
2051 
2052 		error = xfs_trans_commit(tp, 0);
2053 		if (error) {
2054 			tp = NULL;
2055 			break;
2056 		}
2057 	}
2058 
2059 	if (error && tp)
2060 		xfs_trans_cancel(tp, cancelflags);
2061 
2062 	/*
2063 	 * Free the fake mp structure.
2064 	 */
2065 	kmem_free(nmp, sizeof(*nmp));
2066 
2067 	return error;
2068 }
2069 
2070 /*
2071  * Allocate an extent in the realtime subvolume, with the usual allocation
2072  * parameters.  The length units are all in realtime extents, as is the
2073  * result block number.
2074  */
2075 int					/* error */
2076 xfs_rtallocate_extent(
2077 	xfs_trans_t	*tp,		/* transaction pointer */
2078 	xfs_rtblock_t	bno,		/* starting block number to allocate */
2079 	xfs_extlen_t	minlen,		/* minimum length to allocate */
2080 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
2081 	xfs_extlen_t	*len,		/* out: actual length allocated */
2082 	xfs_alloctype_t	type,		/* allocation type XFS_ALLOCTYPE... */
2083 	int		wasdel,		/* was a delayed allocation extent */
2084 	xfs_extlen_t	prod,		/* extent product factor */
2085 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
2086 {
2087 	int		error;		/* error value */
2088 	xfs_inode_t	*ip;		/* inode for bitmap file */
2089 	xfs_mount_t	*mp;		/* file system mount structure */
2090 	xfs_rtblock_t	r;		/* result allocated block */
2091 	xfs_fsblock_t	sb;		/* summary file block number */
2092 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2093 
2094 	ASSERT(minlen > 0 && minlen <= maxlen);
2095 	mp = tp->t_mountp;
2096 	/*
2097 	 * If prod is set then figure out what to do to minlen and maxlen.
2098 	 */
2099 	if (prod > 1) {
2100 		xfs_extlen_t	i;
2101 
2102 		if ((i = maxlen % prod))
2103 			maxlen -= i;
2104 		if ((i = minlen % prod))
2105 			minlen += prod - i;
2106 		if (maxlen < minlen) {
2107 			*rtblock = NULLRTBLOCK;
2108 			return 0;
2109 		}
2110 	}
2111 	/*
2112 	 * Lock out other callers by grabbing the bitmap inode lock.
2113 	 */
2114 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2115 					XFS_ILOCK_EXCL, &ip)))
2116 		return error;
2117 	sumbp = NULL;
2118 	/*
2119 	 * Allocate by size, or near another block, or exactly at some block.
2120 	 */
2121 	switch (type) {
2122 	case XFS_ALLOCTYPE_ANY_AG:
2123 		error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
2124 				&sumbp,	&sb, prod, &r);
2125 		break;
2126 	case XFS_ALLOCTYPE_NEAR_BNO:
2127 		error = xfs_rtallocate_extent_near(mp, tp, bno, minlen, maxlen,
2128 				len, &sumbp, &sb, prod, &r);
2129 		break;
2130 	case XFS_ALLOCTYPE_THIS_BNO:
2131 		error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen,
2132 				len, &sumbp, &sb, prod, &r);
2133 		break;
2134 	default:
2135 		ASSERT(0);
2136 	}
2137 	if (error) {
2138 		return error;
2139 	}
2140 	/*
2141 	 * If it worked, update the superblock.
2142 	 */
2143 	if (r != NULLRTBLOCK) {
2144 		long	slen = (long)*len;
2145 
2146 		ASSERT(*len >= minlen && *len <= maxlen);
2147 		if (wasdel)
2148 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
2149 		else
2150 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
2151 	}
2152 	*rtblock = r;
2153 	return 0;
2154 }
2155 
2156 /*
2157  * Free an extent in the realtime subvolume.  Length is expressed in
2158  * realtime extents, as is the block number.
2159  */
2160 int					/* error */
2161 xfs_rtfree_extent(
2162 	xfs_trans_t	*tp,		/* transaction pointer */
2163 	xfs_rtblock_t	bno,		/* starting block number to free */
2164 	xfs_extlen_t	len)		/* length of extent freed */
2165 {
2166 	int		error;		/* error value */
2167 	xfs_inode_t	*ip;		/* bitmap file inode */
2168 	xfs_mount_t	*mp;		/* file system mount structure */
2169 	xfs_fsblock_t	sb;		/* summary file block number */
2170 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2171 
2172 	mp = tp->t_mountp;
2173 	/*
2174 	 * Synchronize by locking the bitmap inode.
2175 	 */
2176 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2177 					XFS_ILOCK_EXCL, &ip)))
2178 		return error;
2179 #if defined(__KERNEL__) && defined(DEBUG)
2180 	/*
2181 	 * Check to see that this whole range is currently allocated.
2182 	 */
2183 	{
2184 		int	stat;		/* result from checking range */
2185 
2186 		error = xfs_rtcheck_alloc_range(mp, tp, bno, len, &stat);
2187 		if (error) {
2188 			return error;
2189 		}
2190 		ASSERT(stat);
2191 	}
2192 #endif
2193 	sumbp = NULL;
2194 	/*
2195 	 * Free the range of realtime blocks.
2196 	 */
2197 	error = xfs_rtfree_range(mp, tp, bno, len, &sumbp, &sb);
2198 	if (error) {
2199 		return error;
2200 	}
2201 	/*
2202 	 * Mark more blocks free in the superblock.
2203 	 */
2204 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, (long)len);
2205 	/*
2206 	 * If we've now freed all the blocks, reset the file sequence
2207 	 * number to 0.
2208 	 */
2209 	if (tp->t_frextents_delta + mp->m_sb.sb_frextents ==
2210 	    mp->m_sb.sb_rextents) {
2211 		if (!(ip->i_d.di_flags & XFS_DIFLAG_NEWRTBM))
2212 			ip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2213 		*(__uint64_t *)&ip->i_d.di_atime = 0;
2214 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2215 	}
2216 	return 0;
2217 }
2218 
2219 /*
2220  * Initialize realtime fields in the mount structure.
2221  */
2222 int				/* error */
2223 xfs_rtmount_init(
2224 	xfs_mount_t	*mp)	/* file system mount structure */
2225 {
2226 	xfs_buf_t	*bp;	/* buffer for last block of subvolume */
2227 	xfs_daddr_t	d;	/* address of last block of subvolume */
2228 	int		error;	/* error return value */
2229 	xfs_sb_t	*sbp;	/* filesystem superblock copy in mount */
2230 
2231 	sbp = &mp->m_sb;
2232 	if (sbp->sb_rblocks == 0)
2233 		return 0;
2234 	if (mp->m_rtdev_targp == NULL) {
2235 		cmn_err(CE_WARN,
2236 	"XFS: This filesystem has a realtime volume, use rtdev=device option");
2237 		return XFS_ERROR(ENODEV);
2238 	}
2239 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
2240 	mp->m_rsumsize =
2241 		(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
2242 		sbp->sb_rbmblocks;
2243 	mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
2244 	mp->m_rbmip = mp->m_rsumip = NULL;
2245 	/*
2246 	 * Check that the realtime section is an ok size.
2247 	 */
2248 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
2249 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
2250 		cmn_err(CE_WARN, "XFS: realtime mount -- %llu != %llu",
2251 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
2252 			(unsigned long long) mp->m_sb.sb_rblocks);
2253 		return XFS_ERROR(E2BIG);
2254 	}
2255 	error = xfs_read_buf(mp, mp->m_rtdev_targp,
2256 				d - XFS_FSB_TO_BB(mp, 1),
2257 				XFS_FSB_TO_BB(mp, 1), 0, &bp);
2258 	if (error) {
2259 		cmn_err(CE_WARN,
2260 	"XFS: realtime mount -- xfs_read_buf failed, returned %d", error);
2261 		if (error == ENOSPC)
2262 			return XFS_ERROR(E2BIG);
2263 		return error;
2264 	}
2265 	xfs_buf_relse(bp);
2266 	return 0;
2267 }
2268 
2269 /*
2270  * Get the bitmap and summary inodes into the mount structure
2271  * at mount time.
2272  */
2273 int					/* error */
2274 xfs_rtmount_inodes(
2275 	xfs_mount_t	*mp)		/* file system mount structure */
2276 {
2277 	int		error;		/* error return value */
2278 	xfs_sb_t	*sbp;
2279 
2280 	sbp = &mp->m_sb;
2281 	if (sbp->sb_rbmino == NULLFSINO)
2282 		return 0;
2283 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip, 0);
2284 	if (error)
2285 		return error;
2286 	ASSERT(mp->m_rbmip != NULL);
2287 	ASSERT(sbp->sb_rsumino != NULLFSINO);
2288 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip, 0);
2289 	if (error) {
2290 		IRELE(mp->m_rbmip);
2291 		return error;
2292 	}
2293 	ASSERT(mp->m_rsumip != NULL);
2294 	return 0;
2295 }
2296 
2297 /*
2298  * Pick an extent for allocation at the start of a new realtime file.
2299  * Use the sequence number stored in the atime field of the bitmap inode.
2300  * Translate this to a fraction of the rtextents, and return the product
2301  * of rtextents and the fraction.
2302  * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
2303  */
2304 int					/* error */
2305 xfs_rtpick_extent(
2306 	xfs_mount_t	*mp,		/* file system mount point */
2307 	xfs_trans_t	*tp,		/* transaction pointer */
2308 	xfs_extlen_t	len,		/* allocation length (rtextents) */
2309 	xfs_rtblock_t	*pick)		/* result rt extent */
2310 {
2311 	xfs_rtblock_t	b;		/* result block */
2312 	int		error;		/* error return value */
2313 	xfs_inode_t	*ip;		/* bitmap incore inode */
2314 	int		log2;		/* log of sequence number */
2315 	__uint64_t	resid;		/* residual after log removed */
2316 	__uint64_t	seq;		/* sequence number of file creation */
2317 	__uint64_t	*seqp;		/* pointer to seqno in inode */
2318 
2319 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2320 					XFS_ILOCK_EXCL, &ip)))
2321 		return error;
2322 	ASSERT(ip == mp->m_rbmip);
2323 	seqp = (__uint64_t *)&ip->i_d.di_atime;
2324 	if (!(ip->i_d.di_flags & XFS_DIFLAG_NEWRTBM)) {
2325 		ip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2326 		*seqp = 0;
2327 	}
2328 	seq = *seqp;
2329 	if ((log2 = xfs_highbit64(seq)) == -1)
2330 		b = 0;
2331 	else {
2332 		resid = seq - (1ULL << log2);
2333 		b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
2334 		    (log2 + 1);
2335 		if (b >= mp->m_sb.sb_rextents)
2336 			b = do_mod(b, mp->m_sb.sb_rextents);
2337 		if (b + len > mp->m_sb.sb_rextents)
2338 			b = mp->m_sb.sb_rextents - len;
2339 	}
2340 	*seqp = seq + 1;
2341 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2342 	*pick = b;
2343 	return 0;
2344 }
2345