xref: /linux/fs/xfs/xfs_rtalloc.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_alloc.h"
33 #include "xfs_bmap.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_fsops.h"
36 #include "xfs_error.h"
37 #include "xfs_rw.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_utils.h"
41 #include "xfs_trace.h"
42 
43 
44 /*
45  * Prototypes for internal functions.
46  */
47 
48 
49 STATIC int xfs_rtallocate_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
50 		xfs_extlen_t, xfs_buf_t **, xfs_fsblock_t *);
51 STATIC int xfs_rtany_summary(xfs_mount_t *, xfs_trans_t *, int, int,
52 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, int *);
53 STATIC int xfs_rtcheck_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
54 		xfs_extlen_t, int, xfs_rtblock_t *, int *);
55 STATIC int xfs_rtfind_back(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
56 		xfs_rtblock_t, xfs_rtblock_t *);
57 STATIC int xfs_rtfind_forw(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
58 		xfs_rtblock_t, xfs_rtblock_t *);
59 STATIC int xfs_rtget_summary( xfs_mount_t *, xfs_trans_t *, int,
60 		xfs_rtblock_t, xfs_buf_t **, xfs_fsblock_t *, xfs_suminfo_t *);
61 STATIC int xfs_rtmodify_range(xfs_mount_t *, xfs_trans_t *, xfs_rtblock_t,
62 		xfs_extlen_t, int);
63 STATIC int xfs_rtmodify_summary(xfs_mount_t *, xfs_trans_t *, int,
64 		xfs_rtblock_t, int, xfs_buf_t **, xfs_fsblock_t *);
65 
66 /*
67  * Internal functions.
68  */
69 
70 /*
71  * Allocate space to the bitmap or summary file, and zero it, for growfs.
72  */
73 STATIC int				/* error */
74 xfs_growfs_rt_alloc(
75 	xfs_mount_t	*mp,		/* file system mount point */
76 	xfs_extlen_t	oblocks,	/* old count of blocks */
77 	xfs_extlen_t	nblocks,	/* new count of blocks */
78 	xfs_ino_t	ino)		/* inode number (bitmap/summary) */
79 {
80 	xfs_fileoff_t	bno;		/* block number in file */
81 	xfs_buf_t	*bp;		/* temporary buffer for zeroing */
82 	int		committed;	/* transaction committed flag */
83 	xfs_daddr_t	d;		/* disk block address */
84 	int		error;		/* error return value */
85 	xfs_fsblock_t	firstblock;	/* first block allocated in xaction */
86 	xfs_bmap_free_t	flist;		/* list of freed blocks */
87 	xfs_fsblock_t	fsbno;		/* filesystem block for bno */
88 	xfs_inode_t	*ip;		/* pointer to incore inode */
89 	xfs_bmbt_irec_t	map;		/* block map output */
90 	int		nmap;		/* number of block maps */
91 	int		resblks;	/* space reservation */
92 
93 	/*
94 	 * Allocate space to the file, as necessary.
95 	 */
96 	while (oblocks < nblocks) {
97 		int		cancelflags = 0;
98 		xfs_trans_t	*tp;
99 
100 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ALLOC);
101 		resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
102 		/*
103 		 * Reserve space & log for one extent added to the file.
104 		 */
105 		if ((error = xfs_trans_reserve(tp, resblks,
106 				XFS_GROWRTALLOC_LOG_RES(mp), 0,
107 				XFS_TRANS_PERM_LOG_RES,
108 				XFS_DEFAULT_PERM_LOG_COUNT)))
109 			goto error_cancel;
110 		cancelflags = XFS_TRANS_RELEASE_LOG_RES;
111 		/*
112 		 * Lock the inode.
113 		 */
114 		if ((error = xfs_trans_iget(mp, tp, ino, 0,
115 						XFS_ILOCK_EXCL, &ip)))
116 			goto error_cancel;
117 		xfs_bmap_init(&flist, &firstblock);
118 		/*
119 		 * Allocate blocks to the bitmap file.
120 		 */
121 		nmap = 1;
122 		cancelflags |= XFS_TRANS_ABORT;
123 		error = xfs_bmapi(tp, ip, oblocks, nblocks - oblocks,
124 			XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, &firstblock,
125 			resblks, &map, &nmap, &flist);
126 		if (!error && nmap < 1)
127 			error = XFS_ERROR(ENOSPC);
128 		if (error)
129 			goto error_cancel;
130 		/*
131 		 * Free any blocks freed up in the transaction, then commit.
132 		 */
133 		error = xfs_bmap_finish(&tp, &flist, &committed);
134 		if (error)
135 			goto error_cancel;
136 		error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
137 		if (error)
138 			goto error;
139 		/*
140 		 * Now we need to clear the allocated blocks.
141 		 * Do this one block per transaction, to keep it simple.
142 		 */
143 		cancelflags = 0;
144 		for (bno = map.br_startoff, fsbno = map.br_startblock;
145 		     bno < map.br_startoff + map.br_blockcount;
146 		     bno++, fsbno++) {
147 			tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_ZERO);
148 			/*
149 			 * Reserve log for one block zeroing.
150 			 */
151 			if ((error = xfs_trans_reserve(tp, 0,
152 					XFS_GROWRTZERO_LOG_RES(mp), 0, 0, 0)))
153 				goto error_cancel;
154 			/*
155 			 * Lock the bitmap inode.
156 			 */
157 			if ((error = xfs_trans_iget(mp, tp, ino, 0,
158 							XFS_ILOCK_EXCL, &ip)))
159 				goto error_cancel;
160 			/*
161 			 * Get a buffer for the block.
162 			 */
163 			d = XFS_FSB_TO_DADDR(mp, fsbno);
164 			bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
165 				mp->m_bsize, 0);
166 			if (bp == NULL) {
167 				error = XFS_ERROR(EIO);
168 error_cancel:
169 				xfs_trans_cancel(tp, cancelflags);
170 				goto error;
171 			}
172 			memset(XFS_BUF_PTR(bp), 0, mp->m_sb.sb_blocksize);
173 			xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
174 			/*
175 			 * Commit the transaction.
176 			 */
177 			error = xfs_trans_commit(tp, 0);
178 			if (error)
179 				goto error;
180 		}
181 		/*
182 		 * Go on to the next extent, if any.
183 		 */
184 		oblocks = map.br_startoff + map.br_blockcount;
185 	}
186 	return 0;
187 error:
188 	return error;
189 }
190 
191 /*
192  * Attempt to allocate an extent minlen<=len<=maxlen starting from
193  * bitmap block bbno.  If we don't get maxlen then use prod to trim
194  * the length, if given.  Returns error; returns starting block in *rtblock.
195  * The lengths are all in rtextents.
196  */
197 STATIC int				/* error */
198 xfs_rtallocate_extent_block(
199 	xfs_mount_t	*mp,		/* file system mount point */
200 	xfs_trans_t	*tp,		/* transaction pointer */
201 	xfs_rtblock_t	bbno,		/* bitmap block number */
202 	xfs_extlen_t	minlen,		/* minimum length to allocate */
203 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
204 	xfs_extlen_t	*len,		/* out: actual length allocated */
205 	xfs_rtblock_t	*nextp,		/* out: next block to try */
206 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
207 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
208 	xfs_extlen_t	prod,		/* extent product factor */
209 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
210 {
211 	xfs_rtblock_t	besti;		/* best rtblock found so far */
212 	xfs_rtblock_t	bestlen;	/* best length found so far */
213 	xfs_rtblock_t	end;		/* last rtblock in chunk */
214 	int		error;		/* error value */
215 	xfs_rtblock_t	i;		/* current rtblock trying */
216 	xfs_rtblock_t	next;		/* next rtblock to try */
217 	int		stat;		/* status from internal calls */
218 
219 	/*
220 	 * Loop over all the extents starting in this bitmap block,
221 	 * looking for one that's long enough.
222 	 */
223 	for (i = XFS_BLOCKTOBIT(mp, bbno), besti = -1, bestlen = 0,
224 		end = XFS_BLOCKTOBIT(mp, bbno + 1) - 1;
225 	     i <= end;
226 	     i++) {
227 		/*
228 		 * See if there's a free extent of maxlen starting at i.
229 		 * If it's not so then next will contain the first non-free.
230 		 */
231 		error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
232 		if (error) {
233 			return error;
234 		}
235 		if (stat) {
236 			/*
237 			 * i for maxlen is all free, allocate and return that.
238 			 */
239 			error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
240 				rsb);
241 			if (error) {
242 				return error;
243 			}
244 			*len = maxlen;
245 			*rtblock = i;
246 			return 0;
247 		}
248 		/*
249 		 * In the case where we have a variable-sized allocation
250 		 * request, figure out how big this free piece is,
251 		 * and if it's big enough for the minimum, and the best
252 		 * so far, remember it.
253 		 */
254 		if (minlen < maxlen) {
255 			xfs_rtblock_t	thislen;	/* this extent size */
256 
257 			thislen = next - i;
258 			if (thislen >= minlen && thislen > bestlen) {
259 				besti = i;
260 				bestlen = thislen;
261 			}
262 		}
263 		/*
264 		 * If not done yet, find the start of the next free space.
265 		 */
266 		if (next < end) {
267 			error = xfs_rtfind_forw(mp, tp, next, end, &i);
268 			if (error) {
269 				return error;
270 			}
271 		} else
272 			break;
273 	}
274 	/*
275 	 * Searched the whole thing & didn't find a maxlen free extent.
276 	 */
277 	if (minlen < maxlen && besti != -1) {
278 		xfs_extlen_t	p;	/* amount to trim length by */
279 
280 		/*
281 		 * If size should be a multiple of prod, make that so.
282 		 */
283 		if (prod > 1 && (p = do_mod(bestlen, prod)))
284 			bestlen -= p;
285 		/*
286 		 * Allocate besti for bestlen & return that.
287 		 */
288 		error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
289 		if (error) {
290 			return error;
291 		}
292 		*len = bestlen;
293 		*rtblock = besti;
294 		return 0;
295 	}
296 	/*
297 	 * Allocation failed.  Set *nextp to the next block to try.
298 	 */
299 	*nextp = next;
300 	*rtblock = NULLRTBLOCK;
301 	return 0;
302 }
303 
304 /*
305  * Allocate an extent of length minlen<=len<=maxlen, starting at block
306  * bno.  If we don't get maxlen then use prod to trim the length, if given.
307  * Returns error; returns starting block in *rtblock.
308  * The lengths are all in rtextents.
309  */
310 STATIC int				/* error */
311 xfs_rtallocate_extent_exact(
312 	xfs_mount_t	*mp,		/* file system mount point */
313 	xfs_trans_t	*tp,		/* transaction pointer */
314 	xfs_rtblock_t	bno,		/* starting block number to allocate */
315 	xfs_extlen_t	minlen,		/* minimum length to allocate */
316 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
317 	xfs_extlen_t	*len,		/* out: actual length allocated */
318 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
319 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
320 	xfs_extlen_t	prod,		/* extent product factor */
321 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
322 {
323 	int		error;		/* error value */
324 	xfs_extlen_t	i;		/* extent length trimmed due to prod */
325 	int		isfree;		/* extent is free */
326 	xfs_rtblock_t	next;		/* next block to try (dummy) */
327 
328 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
329 	/*
330 	 * Check if the range in question (for maxlen) is free.
331 	 */
332 	error = xfs_rtcheck_range(mp, tp, bno, maxlen, 1, &next, &isfree);
333 	if (error) {
334 		return error;
335 	}
336 	if (isfree) {
337 		/*
338 		 * If it is, allocate it and return success.
339 		 */
340 		error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
341 		if (error) {
342 			return error;
343 		}
344 		*len = maxlen;
345 		*rtblock = bno;
346 		return 0;
347 	}
348 	/*
349 	 * If not, allocate what there is, if it's at least minlen.
350 	 */
351 	maxlen = next - bno;
352 	if (maxlen < minlen) {
353 		/*
354 		 * Failed, return failure status.
355 		 */
356 		*rtblock = NULLRTBLOCK;
357 		return 0;
358 	}
359 	/*
360 	 * Trim off tail of extent, if prod is specified.
361 	 */
362 	if (prod > 1 && (i = maxlen % prod)) {
363 		maxlen -= i;
364 		if (maxlen < minlen) {
365 			/*
366 			 * Now we can't do it, return failure status.
367 			 */
368 			*rtblock = NULLRTBLOCK;
369 			return 0;
370 		}
371 	}
372 	/*
373 	 * Allocate what we can and return it.
374 	 */
375 	error = xfs_rtallocate_range(mp, tp, bno, maxlen, rbpp, rsb);
376 	if (error) {
377 		return error;
378 	}
379 	*len = maxlen;
380 	*rtblock = bno;
381 	return 0;
382 }
383 
384 /*
385  * Allocate an extent of length minlen<=len<=maxlen, starting as near
386  * to bno as possible.  If we don't get maxlen then use prod to trim
387  * the length, if given.  The lengths are all in rtextents.
388  */
389 STATIC int				/* error */
390 xfs_rtallocate_extent_near(
391 	xfs_mount_t	*mp,		/* file system mount point */
392 	xfs_trans_t	*tp,		/* transaction pointer */
393 	xfs_rtblock_t	bno,		/* starting block number to allocate */
394 	xfs_extlen_t	minlen,		/* minimum length to allocate */
395 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
396 	xfs_extlen_t	*len,		/* out: actual length allocated */
397 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
398 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
399 	xfs_extlen_t	prod,		/* extent product factor */
400 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
401 {
402 	int		any;		/* any useful extents from summary */
403 	xfs_rtblock_t	bbno;		/* bitmap block number */
404 	int		error;		/* error value */
405 	int		i;		/* bitmap block offset (loop control) */
406 	int		j;		/* secondary loop control */
407 	int		log2len;	/* log2 of minlen */
408 	xfs_rtblock_t	n;		/* next block to try */
409 	xfs_rtblock_t	r;		/* result block */
410 
411 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
412 	/*
413 	 * If the block number given is off the end, silently set it to
414 	 * the last block.
415 	 */
416 	if (bno >= mp->m_sb.sb_rextents)
417 		bno = mp->m_sb.sb_rextents - 1;
418 	/*
419 	 * Try the exact allocation first.
420 	 */
421 	error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen, len,
422 		rbpp, rsb, prod, &r);
423 	if (error) {
424 		return error;
425 	}
426 	/*
427 	 * If the exact allocation worked, return that.
428 	 */
429 	if (r != NULLRTBLOCK) {
430 		*rtblock = r;
431 		return 0;
432 	}
433 	bbno = XFS_BITTOBLOCK(mp, bno);
434 	i = 0;
435 	ASSERT(minlen != 0);
436 	log2len = xfs_highbit32(minlen);
437 	/*
438 	 * Loop over all bitmap blocks (bbno + i is current block).
439 	 */
440 	for (;;) {
441 		/*
442 		 * Get summary information of extents of all useful levels
443 		 * starting in this bitmap block.
444 		 */
445 		error = xfs_rtany_summary(mp, tp, log2len, mp->m_rsumlevels - 1,
446 			bbno + i, rbpp, rsb, &any);
447 		if (error) {
448 			return error;
449 		}
450 		/*
451 		 * If there are any useful extents starting here, try
452 		 * allocating one.
453 		 */
454 		if (any) {
455 			/*
456 			 * On the positive side of the starting location.
457 			 */
458 			if (i >= 0) {
459 				/*
460 				 * Try to allocate an extent starting in
461 				 * this block.
462 				 */
463 				error = xfs_rtallocate_extent_block(mp, tp,
464 					bbno + i, minlen, maxlen, len, &n, rbpp,
465 					rsb, prod, &r);
466 				if (error) {
467 					return error;
468 				}
469 				/*
470 				 * If it worked, return it.
471 				 */
472 				if (r != NULLRTBLOCK) {
473 					*rtblock = r;
474 					return 0;
475 				}
476 			}
477 			/*
478 			 * On the negative side of the starting location.
479 			 */
480 			else {		/* i < 0 */
481 				/*
482 				 * Loop backwards through the bitmap blocks from
483 				 * the starting point-1 up to where we are now.
484 				 * There should be an extent which ends in this
485 				 * bitmap block and is long enough.
486 				 */
487 				for (j = -1; j > i; j--) {
488 					/*
489 					 * Grab the summary information for
490 					 * this bitmap block.
491 					 */
492 					error = xfs_rtany_summary(mp, tp,
493 						log2len, mp->m_rsumlevels - 1,
494 						bbno + j, rbpp, rsb, &any);
495 					if (error) {
496 						return error;
497 					}
498 					/*
499 					 * If there's no extent given in the
500 					 * summary that means the extent we
501 					 * found must carry over from an
502 					 * earlier block.  If there is an
503 					 * extent given, we've already tried
504 					 * that allocation, don't do it again.
505 					 */
506 					if (any)
507 						continue;
508 					error = xfs_rtallocate_extent_block(mp,
509 						tp, bbno + j, minlen, maxlen,
510 						len, &n, rbpp, rsb, prod, &r);
511 					if (error) {
512 						return error;
513 					}
514 					/*
515 					 * If it works, return the extent.
516 					 */
517 					if (r != NULLRTBLOCK) {
518 						*rtblock = r;
519 						return 0;
520 					}
521 				}
522 				/*
523 				 * There weren't intervening bitmap blocks
524 				 * with a long enough extent, or the
525 				 * allocation didn't work for some reason
526 				 * (i.e. it's a little * too short).
527 				 * Try to allocate from the summary block
528 				 * that we found.
529 				 */
530 				error = xfs_rtallocate_extent_block(mp, tp,
531 					bbno + i, minlen, maxlen, len, &n, rbpp,
532 					rsb, prod, &r);
533 				if (error) {
534 					return error;
535 				}
536 				/*
537 				 * If it works, return the extent.
538 				 */
539 				if (r != NULLRTBLOCK) {
540 					*rtblock = r;
541 					return 0;
542 				}
543 			}
544 		}
545 		/*
546 		 * Loop control.  If we were on the positive side, and there's
547 		 * still more blocks on the negative side, go there.
548 		 */
549 		if (i > 0 && (int)bbno - i >= 0)
550 			i = -i;
551 		/*
552 		 * If positive, and no more negative, but there are more
553 		 * positive, go there.
554 		 */
555 		else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
556 			i++;
557 		/*
558 		 * If negative or 0 (just started), and there are positive
559 		 * blocks to go, go there.  The 0 case moves to block 1.
560 		 */
561 		else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
562 			i = 1 - i;
563 		/*
564 		 * If negative or 0 and there are more negative blocks,
565 		 * go there.
566 		 */
567 		else if (i <= 0 && (int)bbno + i > 0)
568 			i--;
569 		/*
570 		 * Must be done.  Return failure.
571 		 */
572 		else
573 			break;
574 	}
575 	*rtblock = NULLRTBLOCK;
576 	return 0;
577 }
578 
579 /*
580  * Allocate an extent of length minlen<=len<=maxlen, with no position
581  * specified.  If we don't get maxlen then use prod to trim
582  * the length, if given.  The lengths are all in rtextents.
583  */
584 STATIC int				/* error */
585 xfs_rtallocate_extent_size(
586 	xfs_mount_t	*mp,		/* file system mount point */
587 	xfs_trans_t	*tp,		/* transaction pointer */
588 	xfs_extlen_t	minlen,		/* minimum length to allocate */
589 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
590 	xfs_extlen_t	*len,		/* out: actual length allocated */
591 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
592 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
593 	xfs_extlen_t	prod,		/* extent product factor */
594 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
595 {
596 	int		error;		/* error value */
597 	int		i;		/* bitmap block number */
598 	int		l;		/* level number (loop control) */
599 	xfs_rtblock_t	n;		/* next block to be tried */
600 	xfs_rtblock_t	r;		/* result block number */
601 	xfs_suminfo_t	sum;		/* summary information for extents */
602 
603 	ASSERT(minlen % prod == 0 && maxlen % prod == 0);
604 	ASSERT(maxlen != 0);
605 
606 	/*
607 	 * Loop over all the levels starting with maxlen.
608 	 * At each level, look at all the bitmap blocks, to see if there
609 	 * are extents starting there that are long enough (>= maxlen).
610 	 * Note, only on the initial level can the allocation fail if
611 	 * the summary says there's an extent.
612 	 */
613 	for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
614 		/*
615 		 * Loop over all the bitmap blocks.
616 		 */
617 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
618 			/*
619 			 * Get the summary for this level/block.
620 			 */
621 			error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
622 				&sum);
623 			if (error) {
624 				return error;
625 			}
626 			/*
627 			 * Nothing there, on to the next block.
628 			 */
629 			if (!sum)
630 				continue;
631 			/*
632 			 * Try allocating the extent.
633 			 */
634 			error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
635 				maxlen, len, &n, rbpp, rsb, prod, &r);
636 			if (error) {
637 				return error;
638 			}
639 			/*
640 			 * If it worked, return that.
641 			 */
642 			if (r != NULLRTBLOCK) {
643 				*rtblock = r;
644 				return 0;
645 			}
646 			/*
647 			 * If the "next block to try" returned from the
648 			 * allocator is beyond the next bitmap block,
649 			 * skip to that bitmap block.
650 			 */
651 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
652 				i = XFS_BITTOBLOCK(mp, n) - 1;
653 		}
654 	}
655 	/*
656 	 * Didn't find any maxlen blocks.  Try smaller ones, unless
657 	 * we're asking for a fixed size extent.
658 	 */
659 	if (minlen > --maxlen) {
660 		*rtblock = NULLRTBLOCK;
661 		return 0;
662 	}
663 	ASSERT(minlen != 0);
664 	ASSERT(maxlen != 0);
665 
666 	/*
667 	 * Loop over sizes, from maxlen down to minlen.
668 	 * This time, when we do the allocations, allow smaller ones
669 	 * to succeed.
670 	 */
671 	for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
672 		/*
673 		 * Loop over all the bitmap blocks, try an allocation
674 		 * starting in that block.
675 		 */
676 		for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
677 			/*
678 			 * Get the summary information for this level/block.
679 			 */
680 			error =	xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
681 						  &sum);
682 			if (error) {
683 				return error;
684 			}
685 			/*
686 			 * If nothing there, go on to next.
687 			 */
688 			if (!sum)
689 				continue;
690 			/*
691 			 * Try the allocation.  Make sure the specified
692 			 * minlen/maxlen are in the possible range for
693 			 * this summary level.
694 			 */
695 			error = xfs_rtallocate_extent_block(mp, tp, i,
696 					XFS_RTMAX(minlen, 1 << l),
697 					XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
698 					len, &n, rbpp, rsb, prod, &r);
699 			if (error) {
700 				return error;
701 			}
702 			/*
703 			 * If it worked, return that extent.
704 			 */
705 			if (r != NULLRTBLOCK) {
706 				*rtblock = r;
707 				return 0;
708 			}
709 			/*
710 			 * If the "next block to try" returned from the
711 			 * allocator is beyond the next bitmap block,
712 			 * skip to that bitmap block.
713 			 */
714 			if (XFS_BITTOBLOCK(mp, n) > i + 1)
715 				i = XFS_BITTOBLOCK(mp, n) - 1;
716 		}
717 	}
718 	/*
719 	 * Got nothing, return failure.
720 	 */
721 	*rtblock = NULLRTBLOCK;
722 	return 0;
723 }
724 
725 /*
726  * Mark an extent specified by start and len allocated.
727  * Updates all the summary information as well as the bitmap.
728  */
729 STATIC int				/* error */
730 xfs_rtallocate_range(
731 	xfs_mount_t	*mp,		/* file system mount point */
732 	xfs_trans_t	*tp,		/* transaction pointer */
733 	xfs_rtblock_t	start,		/* start block to allocate */
734 	xfs_extlen_t	len,		/* length to allocate */
735 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
736 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
737 {
738 	xfs_rtblock_t	end;		/* end of the allocated extent */
739 	int		error;		/* error value */
740 	xfs_rtblock_t	postblock;	/* first block allocated > end */
741 	xfs_rtblock_t	preblock;	/* first block allocated < start */
742 
743 	end = start + len - 1;
744 	/*
745 	 * Assume we're allocating out of the middle of a free extent.
746 	 * We need to find the beginning and end of the extent so we can
747 	 * properly update the summary.
748 	 */
749 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
750 	if (error) {
751 		return error;
752 	}
753 	/*
754 	 * Find the next allocated block (end of free extent).
755 	 */
756 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
757 		&postblock);
758 	if (error) {
759 		return error;
760 	}
761 	/*
762 	 * Decrement the summary information corresponding to the entire
763 	 * (old) free extent.
764 	 */
765 	error = xfs_rtmodify_summary(mp, tp,
766 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
767 		XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
768 	if (error) {
769 		return error;
770 	}
771 	/*
772 	 * If there are blocks not being allocated at the front of the
773 	 * old extent, add summary data for them to be free.
774 	 */
775 	if (preblock < start) {
776 		error = xfs_rtmodify_summary(mp, tp,
777 			XFS_RTBLOCKLOG(start - preblock),
778 			XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
779 		if (error) {
780 			return error;
781 		}
782 	}
783 	/*
784 	 * If there are blocks not being allocated at the end of the
785 	 * old extent, add summary data for them to be free.
786 	 */
787 	if (postblock > end) {
788 		error = xfs_rtmodify_summary(mp, tp,
789 			XFS_RTBLOCKLOG(postblock - end),
790 			XFS_BITTOBLOCK(mp, end + 1), 1, rbpp, rsb);
791 		if (error) {
792 			return error;
793 		}
794 	}
795 	/*
796 	 * Modify the bitmap to mark this extent allocated.
797 	 */
798 	error = xfs_rtmodify_range(mp, tp, start, len, 0);
799 	return error;
800 }
801 
802 /*
803  * Return whether there are any free extents in the size range given
804  * by low and high, for the bitmap block bbno.
805  */
806 STATIC int				/* error */
807 xfs_rtany_summary(
808 	xfs_mount_t	*mp,		/* file system mount structure */
809 	xfs_trans_t	*tp,		/* transaction pointer */
810 	int		low,		/* low log2 extent size */
811 	int		high,		/* high log2 extent size */
812 	xfs_rtblock_t	bbno,		/* bitmap block number */
813 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
814 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
815 	int		*stat)		/* out: any good extents here? */
816 {
817 	int		error;		/* error value */
818 	int		log;		/* loop counter, log2 of ext. size */
819 	xfs_suminfo_t	sum;		/* summary data */
820 
821 	/*
822 	 * Loop over logs of extent sizes.  Order is irrelevant.
823 	 */
824 	for (log = low; log <= high; log++) {
825 		/*
826 		 * Get one summary datum.
827 		 */
828 		error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
829 		if (error) {
830 			return error;
831 		}
832 		/*
833 		 * If there are any, return success.
834 		 */
835 		if (sum) {
836 			*stat = 1;
837 			return 0;
838 		}
839 	}
840 	/*
841 	 * Found nothing, return failure.
842 	 */
843 	*stat = 0;
844 	return 0;
845 }
846 
847 /*
848  * Get a buffer for the bitmap or summary file block specified.
849  * The buffer is returned read and locked.
850  */
851 STATIC int				/* error */
852 xfs_rtbuf_get(
853 	xfs_mount_t	*mp,		/* file system mount structure */
854 	xfs_trans_t	*tp,		/* transaction pointer */
855 	xfs_rtblock_t	block,		/* block number in bitmap or summary */
856 	int		issum,		/* is summary not bitmap */
857 	xfs_buf_t	**bpp)		/* output: buffer for the block */
858 {
859 	xfs_buf_t	*bp;		/* block buffer, result */
860 	xfs_daddr_t	d;		/* disk addr of block */
861 	int		error;		/* error value */
862 	xfs_fsblock_t	fsb;		/* fs block number for block */
863 	xfs_inode_t	*ip;		/* bitmap or summary inode */
864 
865 	ip = issum ? mp->m_rsumip : mp->m_rbmip;
866 	/*
867 	 * Map from the file offset (block) and inode number to the
868 	 * file system block.
869 	 */
870 	error = xfs_bmapi_single(tp, ip, XFS_DATA_FORK, &fsb, block);
871 	if (error) {
872 		return error;
873 	}
874 	ASSERT(fsb != NULLFSBLOCK);
875 	/*
876 	 * Convert to disk address for buffer cache.
877 	 */
878 	d = XFS_FSB_TO_DADDR(mp, fsb);
879 	/*
880 	 * Read the buffer.
881 	 */
882 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
883 				   mp->m_bsize, 0, &bp);
884 	if (error) {
885 		return error;
886 	}
887 	ASSERT(bp && !XFS_BUF_GETERROR(bp));
888 	*bpp = bp;
889 	return 0;
890 }
891 
892 #ifdef DEBUG
893 /*
894  * Check that the given extent (block range) is allocated already.
895  */
896 STATIC int				/* error */
897 xfs_rtcheck_alloc_range(
898 	xfs_mount_t	*mp,		/* file system mount point */
899 	xfs_trans_t	*tp,		/* transaction pointer */
900 	xfs_rtblock_t	bno,		/* starting block number of extent */
901 	xfs_extlen_t	len,		/* length of extent */
902 	int		*stat)		/* out: 1 for allocated, 0 for not */
903 {
904 	xfs_rtblock_t	new;		/* dummy for xfs_rtcheck_range */
905 
906 	return xfs_rtcheck_range(mp, tp, bno, len, 0, &new, stat);
907 }
908 #endif
909 
910 /*
911  * Check that the given range is either all allocated (val = 0) or
912  * all free (val = 1).
913  */
914 STATIC int				/* error */
915 xfs_rtcheck_range(
916 	xfs_mount_t	*mp,		/* file system mount point */
917 	xfs_trans_t	*tp,		/* transaction pointer */
918 	xfs_rtblock_t	start,		/* starting block number of extent */
919 	xfs_extlen_t	len,		/* length of extent */
920 	int		val,		/* 1 for free, 0 for allocated */
921 	xfs_rtblock_t	*new,		/* out: first block not matching */
922 	int		*stat)		/* out: 1 for matches, 0 for not */
923 {
924 	xfs_rtword_t	*b;		/* current word in buffer */
925 	int		bit;		/* bit number in the word */
926 	xfs_rtblock_t	block;		/* bitmap block number */
927 	xfs_buf_t	*bp;		/* buf for the block */
928 	xfs_rtword_t	*bufp;		/* starting word in buffer */
929 	int		error;		/* error value */
930 	xfs_rtblock_t	i;		/* current bit number rel. to start */
931 	xfs_rtblock_t	lastbit;	/* last useful bit in word */
932 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
933 	xfs_rtword_t	wdiff;		/* difference from wanted value */
934 	int		word;		/* word number in the buffer */
935 
936 	/*
937 	 * Compute starting bitmap block number
938 	 */
939 	block = XFS_BITTOBLOCK(mp, start);
940 	/*
941 	 * Read the bitmap block.
942 	 */
943 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
944 	if (error) {
945 		return error;
946 	}
947 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
948 	/*
949 	 * Compute the starting word's address, and starting bit.
950 	 */
951 	word = XFS_BITTOWORD(mp, start);
952 	b = &bufp[word];
953 	bit = (int)(start & (XFS_NBWORD - 1));
954 	/*
955 	 * 0 (allocated) => all zero's; 1 (free) => all one's.
956 	 */
957 	val = -val;
958 	/*
959 	 * If not starting on a word boundary, deal with the first
960 	 * (partial) word.
961 	 */
962 	if (bit) {
963 		/*
964 		 * Compute first bit not examined.
965 		 */
966 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
967 		/*
968 		 * Mask of relevant bits.
969 		 */
970 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
971 		/*
972 		 * Compute difference between actual and desired value.
973 		 */
974 		if ((wdiff = (*b ^ val) & mask)) {
975 			/*
976 			 * Different, compute first wrong bit and return.
977 			 */
978 			xfs_trans_brelse(tp, bp);
979 			i = XFS_RTLOBIT(wdiff) - bit;
980 			*new = start + i;
981 			*stat = 0;
982 			return 0;
983 		}
984 		i = lastbit - bit;
985 		/*
986 		 * Go on to next block if that's where the next word is
987 		 * and we need the next word.
988 		 */
989 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
990 			/*
991 			 * If done with this block, get the next one.
992 			 */
993 			xfs_trans_brelse(tp, bp);
994 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
995 			if (error) {
996 				return error;
997 			}
998 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
999 			word = 0;
1000 		} else {
1001 			/*
1002 			 * Go on to the next word in the buffer.
1003 			 */
1004 			b++;
1005 		}
1006 	} else {
1007 		/*
1008 		 * Starting on a word boundary, no partial word.
1009 		 */
1010 		i = 0;
1011 	}
1012 	/*
1013 	 * Loop over whole words in buffers.  When we use up one buffer
1014 	 * we move on to the next one.
1015 	 */
1016 	while (len - i >= XFS_NBWORD) {
1017 		/*
1018 		 * Compute difference between actual and desired value.
1019 		 */
1020 		if ((wdiff = *b ^ val)) {
1021 			/*
1022 			 * Different, compute first wrong bit and return.
1023 			 */
1024 			xfs_trans_brelse(tp, bp);
1025 			i += XFS_RTLOBIT(wdiff);
1026 			*new = start + i;
1027 			*stat = 0;
1028 			return 0;
1029 		}
1030 		i += XFS_NBWORD;
1031 		/*
1032 		 * Go on to next block if that's where the next word is
1033 		 * and we need the next word.
1034 		 */
1035 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1036 			/*
1037 			 * If done with this block, get the next one.
1038 			 */
1039 			xfs_trans_brelse(tp, bp);
1040 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1041 			if (error) {
1042 				return error;
1043 			}
1044 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1045 			word = 0;
1046 		} else {
1047 			/*
1048 			 * Go on to the next word in the buffer.
1049 			 */
1050 			b++;
1051 		}
1052 	}
1053 	/*
1054 	 * If not ending on a word boundary, deal with the last
1055 	 * (partial) word.
1056 	 */
1057 	if ((lastbit = len - i)) {
1058 		/*
1059 		 * Mask of relevant bits.
1060 		 */
1061 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1062 		/*
1063 		 * Compute difference between actual and desired value.
1064 		 */
1065 		if ((wdiff = (*b ^ val) & mask)) {
1066 			/*
1067 			 * Different, compute first wrong bit and return.
1068 			 */
1069 			xfs_trans_brelse(tp, bp);
1070 			i += XFS_RTLOBIT(wdiff);
1071 			*new = start + i;
1072 			*stat = 0;
1073 			return 0;
1074 		} else
1075 			i = len;
1076 	}
1077 	/*
1078 	 * Successful, return.
1079 	 */
1080 	xfs_trans_brelse(tp, bp);
1081 	*new = start + i;
1082 	*stat = 1;
1083 	return 0;
1084 }
1085 
1086 /*
1087  * Copy and transform the summary file, given the old and new
1088  * parameters in the mount structures.
1089  */
1090 STATIC int				/* error */
1091 xfs_rtcopy_summary(
1092 	xfs_mount_t	*omp,		/* old file system mount point */
1093 	xfs_mount_t	*nmp,		/* new file system mount point */
1094 	xfs_trans_t	*tp)		/* transaction pointer */
1095 {
1096 	xfs_rtblock_t	bbno;		/* bitmap block number */
1097 	xfs_buf_t	*bp;		/* summary buffer */
1098 	int		error;		/* error return value */
1099 	int		log;		/* summary level number (log length) */
1100 	xfs_suminfo_t	sum;		/* summary data */
1101 	xfs_fsblock_t	sumbno;		/* summary block number */
1102 
1103 	bp = NULL;
1104 	for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
1105 		for (bbno = omp->m_sb.sb_rbmblocks - 1;
1106 		     (xfs_srtblock_t)bbno >= 0;
1107 		     bbno--) {
1108 			error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
1109 				&sumbno, &sum);
1110 			if (error)
1111 				return error;
1112 			if (sum == 0)
1113 				continue;
1114 			error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
1115 				&bp, &sumbno);
1116 			if (error)
1117 				return error;
1118 			error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
1119 				&bp, &sumbno);
1120 			if (error)
1121 				return error;
1122 			ASSERT(sum > 0);
1123 		}
1124 	}
1125 	return 0;
1126 }
1127 
1128 /*
1129  * Searching backward from start to limit, find the first block whose
1130  * allocated/free state is different from start's.
1131  */
1132 STATIC int				/* error */
1133 xfs_rtfind_back(
1134 	xfs_mount_t	*mp,		/* file system mount point */
1135 	xfs_trans_t	*tp,		/* transaction pointer */
1136 	xfs_rtblock_t	start,		/* starting block to look at */
1137 	xfs_rtblock_t	limit,		/* last block to look at */
1138 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1139 {
1140 	xfs_rtword_t	*b;		/* current word in buffer */
1141 	int		bit;		/* bit number in the word */
1142 	xfs_rtblock_t	block;		/* bitmap block number */
1143 	xfs_buf_t	*bp;		/* buf for the block */
1144 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1145 	int		error;		/* error value */
1146 	xfs_rtblock_t	firstbit;	/* first useful bit in the word */
1147 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1148 	xfs_rtblock_t	len;		/* length of inspected area */
1149 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1150 	xfs_rtword_t	want;		/* mask for "good" values */
1151 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1152 	int		word;		/* word number in the buffer */
1153 
1154 	/*
1155 	 * Compute and read in starting bitmap block for starting block.
1156 	 */
1157 	block = XFS_BITTOBLOCK(mp, start);
1158 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1159 	if (error) {
1160 		return error;
1161 	}
1162 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1163 	/*
1164 	 * Get the first word's index & point to it.
1165 	 */
1166 	word = XFS_BITTOWORD(mp, start);
1167 	b = &bufp[word];
1168 	bit = (int)(start & (XFS_NBWORD - 1));
1169 	len = start - limit + 1;
1170 	/*
1171 	 * Compute match value, based on the bit at start: if 1 (free)
1172 	 * then all-ones, else all-zeroes.
1173 	 */
1174 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1175 	/*
1176 	 * If the starting position is not word-aligned, deal with the
1177 	 * partial word.
1178 	 */
1179 	if (bit < XFS_NBWORD - 1) {
1180 		/*
1181 		 * Calculate first (leftmost) bit number to look at,
1182 		 * and mask for all the relevant bits in this word.
1183 		 */
1184 		firstbit = XFS_RTMAX((xfs_srtblock_t)(bit - len + 1), 0);
1185 		mask = (((xfs_rtword_t)1 << (bit - firstbit + 1)) - 1) <<
1186 			firstbit;
1187 		/*
1188 		 * Calculate the difference between the value there
1189 		 * and what we're looking for.
1190 		 */
1191 		if ((wdiff = (*b ^ want) & mask)) {
1192 			/*
1193 			 * Different.  Mark where we are and return.
1194 			 */
1195 			xfs_trans_brelse(tp, bp);
1196 			i = bit - XFS_RTHIBIT(wdiff);
1197 			*rtblock = start - i + 1;
1198 			return 0;
1199 		}
1200 		i = bit - firstbit + 1;
1201 		/*
1202 		 * Go on to previous block if that's where the previous word is
1203 		 * and we need the previous word.
1204 		 */
1205 		if (--word == -1 && i < len) {
1206 			/*
1207 			 * If done with this block, get the previous one.
1208 			 */
1209 			xfs_trans_brelse(tp, bp);
1210 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1211 			if (error) {
1212 				return error;
1213 			}
1214 			bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1215 			word = XFS_BLOCKWMASK(mp);
1216 			b = &bufp[word];
1217 		} else {
1218 			/*
1219 			 * Go on to the previous word in the buffer.
1220 			 */
1221 			b--;
1222 		}
1223 	} else {
1224 		/*
1225 		 * Starting on a word boundary, no partial word.
1226 		 */
1227 		i = 0;
1228 	}
1229 	/*
1230 	 * Loop over whole words in buffers.  When we use up one buffer
1231 	 * we move on to the previous one.
1232 	 */
1233 	while (len - i >= XFS_NBWORD) {
1234 		/*
1235 		 * Compute difference between actual and desired value.
1236 		 */
1237 		if ((wdiff = *b ^ want)) {
1238 			/*
1239 			 * Different, mark where we are and return.
1240 			 */
1241 			xfs_trans_brelse(tp, bp);
1242 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1243 			*rtblock = start - i + 1;
1244 			return 0;
1245 		}
1246 		i += XFS_NBWORD;
1247 		/*
1248 		 * Go on to previous block if that's where the previous word is
1249 		 * and we need the previous word.
1250 		 */
1251 		if (--word == -1 && i < len) {
1252 			/*
1253 			 * If done with this block, get the previous one.
1254 			 */
1255 			xfs_trans_brelse(tp, bp);
1256 			error = xfs_rtbuf_get(mp, tp, --block, 0, &bp);
1257 			if (error) {
1258 				return error;
1259 			}
1260 			bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1261 			word = XFS_BLOCKWMASK(mp);
1262 			b = &bufp[word];
1263 		} else {
1264 			/*
1265 			 * Go on to the previous word in the buffer.
1266 			 */
1267 			b--;
1268 		}
1269 	}
1270 	/*
1271 	 * If not ending on a word boundary, deal with the last
1272 	 * (partial) word.
1273 	 */
1274 	if (len - i) {
1275 		/*
1276 		 * Calculate first (leftmost) bit number to look at,
1277 		 * and mask for all the relevant bits in this word.
1278 		 */
1279 		firstbit = XFS_NBWORD - (len - i);
1280 		mask = (((xfs_rtword_t)1 << (len - i)) - 1) << firstbit;
1281 		/*
1282 		 * Compute difference between actual and desired value.
1283 		 */
1284 		if ((wdiff = (*b ^ want) & mask)) {
1285 			/*
1286 			 * Different, mark where we are and return.
1287 			 */
1288 			xfs_trans_brelse(tp, bp);
1289 			i += XFS_NBWORD - 1 - XFS_RTHIBIT(wdiff);
1290 			*rtblock = start - i + 1;
1291 			return 0;
1292 		} else
1293 			i = len;
1294 	}
1295 	/*
1296 	 * No match, return that we scanned the whole area.
1297 	 */
1298 	xfs_trans_brelse(tp, bp);
1299 	*rtblock = start - i + 1;
1300 	return 0;
1301 }
1302 
1303 /*
1304  * Searching forward from start to limit, find the first block whose
1305  * allocated/free state is different from start's.
1306  */
1307 STATIC int				/* error */
1308 xfs_rtfind_forw(
1309 	xfs_mount_t	*mp,		/* file system mount point */
1310 	xfs_trans_t	*tp,		/* transaction pointer */
1311 	xfs_rtblock_t	start,		/* starting block to look at */
1312 	xfs_rtblock_t	limit,		/* last block to look at */
1313 	xfs_rtblock_t	*rtblock)	/* out: start block found */
1314 {
1315 	xfs_rtword_t	*b;		/* current word in buffer */
1316 	int		bit;		/* bit number in the word */
1317 	xfs_rtblock_t	block;		/* bitmap block number */
1318 	xfs_buf_t	*bp;		/* buf for the block */
1319 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1320 	int		error;		/* error value */
1321 	xfs_rtblock_t	i;		/* current bit number rel. to start */
1322 	xfs_rtblock_t	lastbit;	/* last useful bit in the word */
1323 	xfs_rtblock_t	len;		/* length of inspected area */
1324 	xfs_rtword_t	mask;		/* mask of relevant bits for value */
1325 	xfs_rtword_t	want;		/* mask for "good" values */
1326 	xfs_rtword_t	wdiff;		/* difference from wanted value */
1327 	int		word;		/* word number in the buffer */
1328 
1329 	/*
1330 	 * Compute and read in starting bitmap block for starting block.
1331 	 */
1332 	block = XFS_BITTOBLOCK(mp, start);
1333 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1334 	if (error) {
1335 		return error;
1336 	}
1337 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1338 	/*
1339 	 * Get the first word's index & point to it.
1340 	 */
1341 	word = XFS_BITTOWORD(mp, start);
1342 	b = &bufp[word];
1343 	bit = (int)(start & (XFS_NBWORD - 1));
1344 	len = limit - start + 1;
1345 	/*
1346 	 * Compute match value, based on the bit at start: if 1 (free)
1347 	 * then all-ones, else all-zeroes.
1348 	 */
1349 	want = (*b & ((xfs_rtword_t)1 << bit)) ? -1 : 0;
1350 	/*
1351 	 * If the starting position is not word-aligned, deal with the
1352 	 * partial word.
1353 	 */
1354 	if (bit) {
1355 		/*
1356 		 * Calculate last (rightmost) bit number to look at,
1357 		 * and mask for all the relevant bits in this word.
1358 		 */
1359 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1360 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1361 		/*
1362 		 * Calculate the difference between the value there
1363 		 * and what we're looking for.
1364 		 */
1365 		if ((wdiff = (*b ^ want) & mask)) {
1366 			/*
1367 			 * Different.  Mark where we are and return.
1368 			 */
1369 			xfs_trans_brelse(tp, bp);
1370 			i = XFS_RTLOBIT(wdiff) - bit;
1371 			*rtblock = start + i - 1;
1372 			return 0;
1373 		}
1374 		i = lastbit - bit;
1375 		/*
1376 		 * Go on to next block if that's where the next word is
1377 		 * and we need the next word.
1378 		 */
1379 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1380 			/*
1381 			 * If done with this block, get the previous one.
1382 			 */
1383 			xfs_trans_brelse(tp, bp);
1384 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1385 			if (error) {
1386 				return error;
1387 			}
1388 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1389 			word = 0;
1390 		} else {
1391 			/*
1392 			 * Go on to the previous word in the buffer.
1393 			 */
1394 			b++;
1395 		}
1396 	} else {
1397 		/*
1398 		 * Starting on a word boundary, no partial word.
1399 		 */
1400 		i = 0;
1401 	}
1402 	/*
1403 	 * Loop over whole words in buffers.  When we use up one buffer
1404 	 * we move on to the next one.
1405 	 */
1406 	while (len - i >= XFS_NBWORD) {
1407 		/*
1408 		 * Compute difference between actual and desired value.
1409 		 */
1410 		if ((wdiff = *b ^ want)) {
1411 			/*
1412 			 * Different, mark where we are and return.
1413 			 */
1414 			xfs_trans_brelse(tp, bp);
1415 			i += XFS_RTLOBIT(wdiff);
1416 			*rtblock = start + i - 1;
1417 			return 0;
1418 		}
1419 		i += XFS_NBWORD;
1420 		/*
1421 		 * Go on to next block if that's where the next word is
1422 		 * and we need the next word.
1423 		 */
1424 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1425 			/*
1426 			 * If done with this block, get the next one.
1427 			 */
1428 			xfs_trans_brelse(tp, bp);
1429 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1430 			if (error) {
1431 				return error;
1432 			}
1433 			b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1434 			word = 0;
1435 		} else {
1436 			/*
1437 			 * Go on to the next word in the buffer.
1438 			 */
1439 			b++;
1440 		}
1441 	}
1442 	/*
1443 	 * If not ending on a word boundary, deal with the last
1444 	 * (partial) word.
1445 	 */
1446 	if ((lastbit = len - i)) {
1447 		/*
1448 		 * Calculate mask for all the relevant bits in this word.
1449 		 */
1450 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1451 		/*
1452 		 * Compute difference between actual and desired value.
1453 		 */
1454 		if ((wdiff = (*b ^ want) & mask)) {
1455 			/*
1456 			 * Different, mark where we are and return.
1457 			 */
1458 			xfs_trans_brelse(tp, bp);
1459 			i += XFS_RTLOBIT(wdiff);
1460 			*rtblock = start + i - 1;
1461 			return 0;
1462 		} else
1463 			i = len;
1464 	}
1465 	/*
1466 	 * No match, return that we scanned the whole area.
1467 	 */
1468 	xfs_trans_brelse(tp, bp);
1469 	*rtblock = start + i - 1;
1470 	return 0;
1471 }
1472 
1473 /*
1474  * Mark an extent specified by start and len freed.
1475  * Updates all the summary information as well as the bitmap.
1476  */
1477 STATIC int				/* error */
1478 xfs_rtfree_range(
1479 	xfs_mount_t	*mp,		/* file system mount point */
1480 	xfs_trans_t	*tp,		/* transaction pointer */
1481 	xfs_rtblock_t	start,		/* starting block to free */
1482 	xfs_extlen_t	len,		/* length to free */
1483 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1484 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1485 {
1486 	xfs_rtblock_t	end;		/* end of the freed extent */
1487 	int		error;		/* error value */
1488 	xfs_rtblock_t	postblock;	/* first block freed > end */
1489 	xfs_rtblock_t	preblock;	/* first block freed < start */
1490 
1491 	end = start + len - 1;
1492 	/*
1493 	 * Modify the bitmap to mark this extent freed.
1494 	 */
1495 	error = xfs_rtmodify_range(mp, tp, start, len, 1);
1496 	if (error) {
1497 		return error;
1498 	}
1499 	/*
1500 	 * Assume we're freeing out of the middle of an allocated extent.
1501 	 * We need to find the beginning and end of the extent so we can
1502 	 * properly update the summary.
1503 	 */
1504 	error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
1505 	if (error) {
1506 		return error;
1507 	}
1508 	/*
1509 	 * Find the next allocated block (end of allocated extent).
1510 	 */
1511 	error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
1512 		&postblock);
1513 	if (error)
1514 		return error;
1515 	/*
1516 	 * If there are blocks not being freed at the front of the
1517 	 * old extent, add summary data for them to be allocated.
1518 	 */
1519 	if (preblock < start) {
1520 		error = xfs_rtmodify_summary(mp, tp,
1521 			XFS_RTBLOCKLOG(start - preblock),
1522 			XFS_BITTOBLOCK(mp, preblock), -1, rbpp, rsb);
1523 		if (error) {
1524 			return error;
1525 		}
1526 	}
1527 	/*
1528 	 * If there are blocks not being freed at the end of the
1529 	 * old extent, add summary data for them to be allocated.
1530 	 */
1531 	if (postblock > end) {
1532 		error = xfs_rtmodify_summary(mp, tp,
1533 			XFS_RTBLOCKLOG(postblock - end),
1534 			XFS_BITTOBLOCK(mp, end + 1), -1, rbpp, rsb);
1535 		if (error) {
1536 			return error;
1537 		}
1538 	}
1539 	/*
1540 	 * Increment the summary information corresponding to the entire
1541 	 * (new) free extent.
1542 	 */
1543 	error = xfs_rtmodify_summary(mp, tp,
1544 		XFS_RTBLOCKLOG(postblock + 1 - preblock),
1545 		XFS_BITTOBLOCK(mp, preblock), 1, rbpp, rsb);
1546 	return error;
1547 }
1548 
1549 /*
1550  * Read and return the summary information for a given extent size,
1551  * bitmap block combination.
1552  * Keeps track of a current summary block, so we don't keep reading
1553  * it from the buffer cache.
1554  */
1555 STATIC int				/* error */
1556 xfs_rtget_summary(
1557 	xfs_mount_t	*mp,		/* file system mount structure */
1558 	xfs_trans_t	*tp,		/* transaction pointer */
1559 	int		log,		/* log2 of extent size */
1560 	xfs_rtblock_t	bbno,		/* bitmap block number */
1561 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1562 	xfs_fsblock_t	*rsb,		/* in/out: summary block number */
1563 	xfs_suminfo_t	*sum)		/* out: summary info for this block */
1564 {
1565 	xfs_buf_t	*bp;		/* buffer for summary block */
1566 	int		error;		/* error value */
1567 	xfs_fsblock_t	sb;		/* summary fsblock */
1568 	int		so;		/* index into the summary file */
1569 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1570 
1571 	/*
1572 	 * Compute entry number in the summary file.
1573 	 */
1574 	so = XFS_SUMOFFS(mp, log, bbno);
1575 	/*
1576 	 * Compute the block number in the summary file.
1577 	 */
1578 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1579 	/*
1580 	 * If we have an old buffer, and the block number matches, use that.
1581 	 */
1582 	if (rbpp && *rbpp && *rsb == sb)
1583 		bp = *rbpp;
1584 	/*
1585 	 * Otherwise we have to get the buffer.
1586 	 */
1587 	else {
1588 		/*
1589 		 * If there was an old one, get rid of it first.
1590 		 */
1591 		if (rbpp && *rbpp)
1592 			xfs_trans_brelse(tp, *rbpp);
1593 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1594 		if (error) {
1595 			return error;
1596 		}
1597 		/*
1598 		 * Remember this buffer and block for the next call.
1599 		 */
1600 		if (rbpp) {
1601 			*rbpp = bp;
1602 			*rsb = sb;
1603 		}
1604 	}
1605 	/*
1606 	 * Point to the summary information & copy it out.
1607 	 */
1608 	sp = XFS_SUMPTR(mp, bp, so);
1609 	*sum = *sp;
1610 	/*
1611 	 * Drop the buffer if we're not asked to remember it.
1612 	 */
1613 	if (!rbpp)
1614 		xfs_trans_brelse(tp, bp);
1615 	return 0;
1616 }
1617 
1618 /*
1619  * Set the given range of bitmap bits to the given value.
1620  * Do whatever I/O and logging is required.
1621  */
1622 STATIC int				/* error */
1623 xfs_rtmodify_range(
1624 	xfs_mount_t	*mp,		/* file system mount point */
1625 	xfs_trans_t	*tp,		/* transaction pointer */
1626 	xfs_rtblock_t	start,		/* starting block to modify */
1627 	xfs_extlen_t	len,		/* length of extent to modify */
1628 	int		val)		/* 1 for free, 0 for allocated */
1629 {
1630 	xfs_rtword_t	*b;		/* current word in buffer */
1631 	int		bit;		/* bit number in the word */
1632 	xfs_rtblock_t	block;		/* bitmap block number */
1633 	xfs_buf_t	*bp;		/* buf for the block */
1634 	xfs_rtword_t	*bufp;		/* starting word in buffer */
1635 	int		error;		/* error value */
1636 	xfs_rtword_t	*first;		/* first used word in the buffer */
1637 	int		i;		/* current bit number rel. to start */
1638 	int		lastbit;	/* last useful bit in word */
1639 	xfs_rtword_t	mask;		/* mask o frelevant bits for value */
1640 	int		word;		/* word number in the buffer */
1641 
1642 	/*
1643 	 * Compute starting bitmap block number.
1644 	 */
1645 	block = XFS_BITTOBLOCK(mp, start);
1646 	/*
1647 	 * Read the bitmap block, and point to its data.
1648 	 */
1649 	error = xfs_rtbuf_get(mp, tp, block, 0, &bp);
1650 	if (error) {
1651 		return error;
1652 	}
1653 	bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1654 	/*
1655 	 * Compute the starting word's address, and starting bit.
1656 	 */
1657 	word = XFS_BITTOWORD(mp, start);
1658 	first = b = &bufp[word];
1659 	bit = (int)(start & (XFS_NBWORD - 1));
1660 	/*
1661 	 * 0 (allocated) => all zeroes; 1 (free) => all ones.
1662 	 */
1663 	val = -val;
1664 	/*
1665 	 * If not starting on a word boundary, deal with the first
1666 	 * (partial) word.
1667 	 */
1668 	if (bit) {
1669 		/*
1670 		 * Compute first bit not changed and mask of relevant bits.
1671 		 */
1672 		lastbit = XFS_RTMIN(bit + len, XFS_NBWORD);
1673 		mask = (((xfs_rtword_t)1 << (lastbit - bit)) - 1) << bit;
1674 		/*
1675 		 * Set/clear the active bits.
1676 		 */
1677 		if (val)
1678 			*b |= mask;
1679 		else
1680 			*b &= ~mask;
1681 		i = lastbit - bit;
1682 		/*
1683 		 * Go on to the next block if that's where the next word is
1684 		 * and we need the next word.
1685 		 */
1686 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1687 			/*
1688 			 * Log the changed part of this block.
1689 			 * Get the next one.
1690 			 */
1691 			xfs_trans_log_buf(tp, bp,
1692 				(uint)((char *)first - (char *)bufp),
1693 				(uint)((char *)b - (char *)bufp));
1694 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1695 			if (error) {
1696 				return error;
1697 			}
1698 			first = b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1699 			word = 0;
1700 		} else {
1701 			/*
1702 			 * Go on to the next word in the buffer
1703 			 */
1704 			b++;
1705 		}
1706 	} else {
1707 		/*
1708 		 * Starting on a word boundary, no partial word.
1709 		 */
1710 		i = 0;
1711 	}
1712 	/*
1713 	 * Loop over whole words in buffers.  When we use up one buffer
1714 	 * we move on to the next one.
1715 	 */
1716 	while (len - i >= XFS_NBWORD) {
1717 		/*
1718 		 * Set the word value correctly.
1719 		 */
1720 		*b = val;
1721 		i += XFS_NBWORD;
1722 		/*
1723 		 * Go on to the next block if that's where the next word is
1724 		 * and we need the next word.
1725 		 */
1726 		if (++word == XFS_BLOCKWSIZE(mp) && i < len) {
1727 			/*
1728 			 * Log the changed part of this block.
1729 			 * Get the next one.
1730 			 */
1731 			xfs_trans_log_buf(tp, bp,
1732 				(uint)((char *)first - (char *)bufp),
1733 				(uint)((char *)b - (char *)bufp));
1734 			error = xfs_rtbuf_get(mp, tp, ++block, 0, &bp);
1735 			if (error) {
1736 				return error;
1737 			}
1738 			first = b = bufp = (xfs_rtword_t *)XFS_BUF_PTR(bp);
1739 			word = 0;
1740 		} else {
1741 			/*
1742 			 * Go on to the next word in the buffer
1743 			 */
1744 			b++;
1745 		}
1746 	}
1747 	/*
1748 	 * If not ending on a word boundary, deal with the last
1749 	 * (partial) word.
1750 	 */
1751 	if ((lastbit = len - i)) {
1752 		/*
1753 		 * Compute a mask of relevant bits.
1754 		 */
1755 		bit = 0;
1756 		mask = ((xfs_rtword_t)1 << lastbit) - 1;
1757 		/*
1758 		 * Set/clear the active bits.
1759 		 */
1760 		if (val)
1761 			*b |= mask;
1762 		else
1763 			*b &= ~mask;
1764 		b++;
1765 	}
1766 	/*
1767 	 * Log any remaining changed bytes.
1768 	 */
1769 	if (b > first)
1770 		xfs_trans_log_buf(tp, bp, (uint)((char *)first - (char *)bufp),
1771 			(uint)((char *)b - (char *)bufp - 1));
1772 	return 0;
1773 }
1774 
1775 /*
1776  * Read and modify the summary information for a given extent size,
1777  * bitmap block combination.
1778  * Keeps track of a current summary block, so we don't keep reading
1779  * it from the buffer cache.
1780  */
1781 STATIC int				/* error */
1782 xfs_rtmodify_summary(
1783 	xfs_mount_t	*mp,		/* file system mount point */
1784 	xfs_trans_t	*tp,		/* transaction pointer */
1785 	int		log,		/* log2 of extent size */
1786 	xfs_rtblock_t	bbno,		/* bitmap block number */
1787 	int		delta,		/* change to make to summary info */
1788 	xfs_buf_t	**rbpp,		/* in/out: summary block buffer */
1789 	xfs_fsblock_t	*rsb)		/* in/out: summary block number */
1790 {
1791 	xfs_buf_t	*bp;		/* buffer for the summary block */
1792 	int		error;		/* error value */
1793 	xfs_fsblock_t	sb;		/* summary fsblock */
1794 	int		so;		/* index into the summary file */
1795 	xfs_suminfo_t	*sp;		/* pointer to returned data */
1796 
1797 	/*
1798 	 * Compute entry number in the summary file.
1799 	 */
1800 	so = XFS_SUMOFFS(mp, log, bbno);
1801 	/*
1802 	 * Compute the block number in the summary file.
1803 	 */
1804 	sb = XFS_SUMOFFSTOBLOCK(mp, so);
1805 	/*
1806 	 * If we have an old buffer, and the block number matches, use that.
1807 	 */
1808 	if (rbpp && *rbpp && *rsb == sb)
1809 		bp = *rbpp;
1810 	/*
1811 	 * Otherwise we have to get the buffer.
1812 	 */
1813 	else {
1814 		/*
1815 		 * If there was an old one, get rid of it first.
1816 		 */
1817 		if (rbpp && *rbpp)
1818 			xfs_trans_brelse(tp, *rbpp);
1819 		error = xfs_rtbuf_get(mp, tp, sb, 1, &bp);
1820 		if (error) {
1821 			return error;
1822 		}
1823 		/*
1824 		 * Remember this buffer and block for the next call.
1825 		 */
1826 		if (rbpp) {
1827 			*rbpp = bp;
1828 			*rsb = sb;
1829 		}
1830 	}
1831 	/*
1832 	 * Point to the summary information, modify and log it.
1833 	 */
1834 	sp = XFS_SUMPTR(mp, bp, so);
1835 	*sp += delta;
1836 	xfs_trans_log_buf(tp, bp, (uint)((char *)sp - (char *)XFS_BUF_PTR(bp)),
1837 		(uint)((char *)sp - (char *)XFS_BUF_PTR(bp) + sizeof(*sp) - 1));
1838 	return 0;
1839 }
1840 
1841 /*
1842  * Visible (exported) functions.
1843  */
1844 
1845 /*
1846  * Grow the realtime area of the filesystem.
1847  */
1848 int
1849 xfs_growfs_rt(
1850 	xfs_mount_t	*mp,		/* mount point for filesystem */
1851 	xfs_growfs_rt_t	*in)		/* growfs rt input struct */
1852 {
1853 	xfs_rtblock_t	bmbno;		/* bitmap block number */
1854 	xfs_buf_t	*bp;		/* temporary buffer */
1855 	int		error;		/* error return value */
1856 	xfs_inode_t	*ip;		/* bitmap inode, used as lock */
1857 	xfs_mount_t	*nmp;		/* new (fake) mount structure */
1858 	xfs_drfsbno_t	nrblocks;	/* new number of realtime blocks */
1859 	xfs_extlen_t	nrbmblocks;	/* new number of rt bitmap blocks */
1860 	xfs_drtbno_t	nrextents;	/* new number of realtime extents */
1861 	uint8_t		nrextslog;	/* new log2 of sb_rextents */
1862 	xfs_extlen_t	nrsumblocks;	/* new number of summary blocks */
1863 	uint		nrsumlevels;	/* new rt summary levels */
1864 	uint		nrsumsize;	/* new size of rt summary, bytes */
1865 	xfs_sb_t	*nsbp;		/* new superblock */
1866 	xfs_extlen_t	rbmblocks;	/* current number of rt bitmap blocks */
1867 	xfs_extlen_t	rsumblocks;	/* current number of rt summary blks */
1868 	xfs_sb_t	*sbp;		/* old superblock */
1869 	xfs_fsblock_t	sumbno;		/* summary block number */
1870 
1871 	sbp = &mp->m_sb;
1872 	/*
1873 	 * Initial error checking.
1874 	 */
1875 	if (!capable(CAP_SYS_ADMIN))
1876 		return XFS_ERROR(EPERM);
1877 	if (mp->m_rtdev_targp == NULL || mp->m_rbmip == NULL ||
1878 	    (nrblocks = in->newblocks) <= sbp->sb_rblocks ||
1879 	    (sbp->sb_rblocks && (in->extsize != sbp->sb_rextsize)))
1880 		return XFS_ERROR(EINVAL);
1881 	if ((error = xfs_sb_validate_fsb_count(sbp, nrblocks)))
1882 		return error;
1883 	/*
1884 	 * Read in the last block of the device, make sure it exists.
1885 	 */
1886 	error = xfs_read_buf(mp, mp->m_rtdev_targp,
1887 			XFS_FSB_TO_BB(mp, nrblocks - 1),
1888 			XFS_FSB_TO_BB(mp, 1), 0, &bp);
1889 	if (error)
1890 		return error;
1891 	ASSERT(bp);
1892 	xfs_buf_relse(bp);
1893 	/*
1894 	 * Calculate new parameters.  These are the final values to be reached.
1895 	 */
1896 	nrextents = nrblocks;
1897 	do_div(nrextents, in->extsize);
1898 	nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
1899 	nrextslog = xfs_highbit32(nrextents);
1900 	nrsumlevels = nrextslog + 1;
1901 	nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
1902 	nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1903 	nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1904 	/*
1905 	 * New summary size can't be more than half the size of
1906 	 * the log.  This prevents us from getting a log overflow,
1907 	 * since we'll log basically the whole summary file at once.
1908 	 */
1909 	if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
1910 		return XFS_ERROR(EINVAL);
1911 	/*
1912 	 * Get the old block counts for bitmap and summary inodes.
1913 	 * These can't change since other growfs callers are locked out.
1914 	 */
1915 	rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_d.di_size);
1916 	rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_d.di_size);
1917 	/*
1918 	 * Allocate space to the bitmap and summary files, as necessary.
1919 	 */
1920 	if ((error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks,
1921 			mp->m_sb.sb_rbmino)))
1922 		return error;
1923 	if ((error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks,
1924 			mp->m_sb.sb_rsumino)))
1925 		return error;
1926 	/*
1927 	 * Allocate a new (fake) mount/sb.
1928 	 */
1929 	nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
1930 	/*
1931 	 * Loop over the bitmap blocks.
1932 	 * We will do everything one bitmap block at a time.
1933 	 * Skip the current block if it is exactly full.
1934 	 * This also deals with the case where there were no rtextents before.
1935 	 */
1936 	for (bmbno = sbp->sb_rbmblocks -
1937 		     ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1938 	     bmbno < nrbmblocks;
1939 	     bmbno++) {
1940 		xfs_trans_t	*tp;
1941 		int		cancelflags = 0;
1942 
1943 		*nmp = *mp;
1944 		nsbp = &nmp->m_sb;
1945 		/*
1946 		 * Calculate new sb and mount fields for this round.
1947 		 */
1948 		nsbp->sb_rextsize = in->extsize;
1949 		nsbp->sb_rbmblocks = bmbno + 1;
1950 		nsbp->sb_rblocks =
1951 			XFS_RTMIN(nrblocks,
1952 				  nsbp->sb_rbmblocks * NBBY *
1953 				  nsbp->sb_blocksize * nsbp->sb_rextsize);
1954 		nsbp->sb_rextents = nsbp->sb_rblocks;
1955 		do_div(nsbp->sb_rextents, nsbp->sb_rextsize);
1956 		ASSERT(nsbp->sb_rextents != 0);
1957 		nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1958 		nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1959 		nrsumsize =
1960 			(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
1961 			nsbp->sb_rbmblocks;
1962 		nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
1963 		nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1964 		/*
1965 		 * Start a transaction, get the log reservation.
1966 		 */
1967 		tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFSRT_FREE);
1968 		if ((error = xfs_trans_reserve(tp, 0,
1969 				XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
1970 			goto error_cancel;
1971 		/*
1972 		 * Lock out other callers by grabbing the bitmap inode lock.
1973 		 */
1974 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
1975 						XFS_ILOCK_EXCL, &ip)))
1976 			goto error_cancel;
1977 		ASSERT(ip == mp->m_rbmip);
1978 		/*
1979 		 * Update the bitmap inode's size.
1980 		 */
1981 		mp->m_rbmip->i_d.di_size =
1982 			nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1983 		xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1984 		cancelflags |= XFS_TRANS_ABORT;
1985 		/*
1986 		 * Get the summary inode into the transaction.
1987 		 */
1988 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rsumino, 0,
1989 						XFS_ILOCK_EXCL, &ip)))
1990 			goto error_cancel;
1991 		ASSERT(ip == mp->m_rsumip);
1992 		/*
1993 		 * Update the summary inode's size.
1994 		 */
1995 		mp->m_rsumip->i_d.di_size = nmp->m_rsumsize;
1996 		xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
1997 		/*
1998 		 * Copy summary data from old to new sizes.
1999 		 * Do this when the real size (not block-aligned) changes.
2000 		 */
2001 		if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
2002 		    mp->m_rsumlevels != nmp->m_rsumlevels) {
2003 			error = xfs_rtcopy_summary(mp, nmp, tp);
2004 			if (error)
2005 				goto error_cancel;
2006 		}
2007 		/*
2008 		 * Update superblock fields.
2009 		 */
2010 		if (nsbp->sb_rextsize != sbp->sb_rextsize)
2011 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
2012 				nsbp->sb_rextsize - sbp->sb_rextsize);
2013 		if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
2014 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
2015 				nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
2016 		if (nsbp->sb_rblocks != sbp->sb_rblocks)
2017 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
2018 				nsbp->sb_rblocks - sbp->sb_rblocks);
2019 		if (nsbp->sb_rextents != sbp->sb_rextents)
2020 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
2021 				nsbp->sb_rextents - sbp->sb_rextents);
2022 		if (nsbp->sb_rextslog != sbp->sb_rextslog)
2023 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
2024 				nsbp->sb_rextslog - sbp->sb_rextslog);
2025 		/*
2026 		 * Free new extent.
2027 		 */
2028 		bp = NULL;
2029 		error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
2030 			nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
2031 		if (error) {
2032 error_cancel:
2033 			xfs_trans_cancel(tp, cancelflags);
2034 			break;
2035 		}
2036 		/*
2037 		 * Mark more blocks free in the superblock.
2038 		 */
2039 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
2040 			nsbp->sb_rextents - sbp->sb_rextents);
2041 		/*
2042 		 * Update mp values into the real mp structure.
2043 		 */
2044 		mp->m_rsumlevels = nrsumlevels;
2045 		mp->m_rsumsize = nrsumsize;
2046 
2047 		error = xfs_trans_commit(tp, 0);
2048 		if (error)
2049 			break;
2050 	}
2051 
2052 	/*
2053 	 * Free the fake mp structure.
2054 	 */
2055 	kmem_free(nmp);
2056 
2057 	return error;
2058 }
2059 
2060 /*
2061  * Allocate an extent in the realtime subvolume, with the usual allocation
2062  * parameters.  The length units are all in realtime extents, as is the
2063  * result block number.
2064  */
2065 int					/* error */
2066 xfs_rtallocate_extent(
2067 	xfs_trans_t	*tp,		/* transaction pointer */
2068 	xfs_rtblock_t	bno,		/* starting block number to allocate */
2069 	xfs_extlen_t	minlen,		/* minimum length to allocate */
2070 	xfs_extlen_t	maxlen,		/* maximum length to allocate */
2071 	xfs_extlen_t	*len,		/* out: actual length allocated */
2072 	xfs_alloctype_t	type,		/* allocation type XFS_ALLOCTYPE... */
2073 	int		wasdel,		/* was a delayed allocation extent */
2074 	xfs_extlen_t	prod,		/* extent product factor */
2075 	xfs_rtblock_t	*rtblock)	/* out: start block allocated */
2076 {
2077 	int		error;		/* error value */
2078 	xfs_inode_t	*ip;		/* inode for bitmap file */
2079 	xfs_mount_t	*mp;		/* file system mount structure */
2080 	xfs_rtblock_t	r;		/* result allocated block */
2081 	xfs_fsblock_t	sb;		/* summary file block number */
2082 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2083 
2084 	ASSERT(minlen > 0 && minlen <= maxlen);
2085 	mp = tp->t_mountp;
2086 	/*
2087 	 * If prod is set then figure out what to do to minlen and maxlen.
2088 	 */
2089 	if (prod > 1) {
2090 		xfs_extlen_t	i;
2091 
2092 		if ((i = maxlen % prod))
2093 			maxlen -= i;
2094 		if ((i = minlen % prod))
2095 			minlen += prod - i;
2096 		if (maxlen < minlen) {
2097 			*rtblock = NULLRTBLOCK;
2098 			return 0;
2099 		}
2100 	}
2101 	/*
2102 	 * Lock out other callers by grabbing the bitmap inode lock.
2103 	 */
2104 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2105 					XFS_ILOCK_EXCL, &ip)))
2106 		return error;
2107 	sumbp = NULL;
2108 	/*
2109 	 * Allocate by size, or near another block, or exactly at some block.
2110 	 */
2111 	switch (type) {
2112 	case XFS_ALLOCTYPE_ANY_AG:
2113 		error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
2114 				&sumbp,	&sb, prod, &r);
2115 		break;
2116 	case XFS_ALLOCTYPE_NEAR_BNO:
2117 		error = xfs_rtallocate_extent_near(mp, tp, bno, minlen, maxlen,
2118 				len, &sumbp, &sb, prod, &r);
2119 		break;
2120 	case XFS_ALLOCTYPE_THIS_BNO:
2121 		error = xfs_rtallocate_extent_exact(mp, tp, bno, minlen, maxlen,
2122 				len, &sumbp, &sb, prod, &r);
2123 		break;
2124 	default:
2125 		ASSERT(0);
2126 	}
2127 	if (error) {
2128 		return error;
2129 	}
2130 	/*
2131 	 * If it worked, update the superblock.
2132 	 */
2133 	if (r != NULLRTBLOCK) {
2134 		long	slen = (long)*len;
2135 
2136 		ASSERT(*len >= minlen && *len <= maxlen);
2137 		if (wasdel)
2138 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
2139 		else
2140 			xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
2141 	}
2142 	*rtblock = r;
2143 	return 0;
2144 }
2145 
2146 /*
2147  * Free an extent in the realtime subvolume.  Length is expressed in
2148  * realtime extents, as is the block number.
2149  */
2150 int					/* error */
2151 xfs_rtfree_extent(
2152 	xfs_trans_t	*tp,		/* transaction pointer */
2153 	xfs_rtblock_t	bno,		/* starting block number to free */
2154 	xfs_extlen_t	len)		/* length of extent freed */
2155 {
2156 	int		error;		/* error value */
2157 	xfs_inode_t	*ip;		/* bitmap file inode */
2158 	xfs_mount_t	*mp;		/* file system mount structure */
2159 	xfs_fsblock_t	sb;		/* summary file block number */
2160 	xfs_buf_t	*sumbp;		/* summary file block buffer */
2161 
2162 	mp = tp->t_mountp;
2163 	/*
2164 	 * Synchronize by locking the bitmap inode.
2165 	 */
2166 	if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
2167 					XFS_ILOCK_EXCL, &ip)))
2168 		return error;
2169 #if defined(__KERNEL__) && defined(DEBUG)
2170 	/*
2171 	 * Check to see that this whole range is currently allocated.
2172 	 */
2173 	{
2174 		int	stat;		/* result from checking range */
2175 
2176 		error = xfs_rtcheck_alloc_range(mp, tp, bno, len, &stat);
2177 		if (error) {
2178 			return error;
2179 		}
2180 		ASSERT(stat);
2181 	}
2182 #endif
2183 	sumbp = NULL;
2184 	/*
2185 	 * Free the range of realtime blocks.
2186 	 */
2187 	error = xfs_rtfree_range(mp, tp, bno, len, &sumbp, &sb);
2188 	if (error) {
2189 		return error;
2190 	}
2191 	/*
2192 	 * Mark more blocks free in the superblock.
2193 	 */
2194 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, (long)len);
2195 	/*
2196 	 * If we've now freed all the blocks, reset the file sequence
2197 	 * number to 0.
2198 	 */
2199 	if (tp->t_frextents_delta + mp->m_sb.sb_frextents ==
2200 	    mp->m_sb.sb_rextents) {
2201 		if (!(ip->i_d.di_flags & XFS_DIFLAG_NEWRTBM))
2202 			ip->i_d.di_flags |= XFS_DIFLAG_NEWRTBM;
2203 		*(__uint64_t *)&ip->i_d.di_atime = 0;
2204 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2205 	}
2206 	return 0;
2207 }
2208 
2209 /*
2210  * Initialize realtime fields in the mount structure.
2211  */
2212 int				/* error */
2213 xfs_rtmount_init(
2214 	xfs_mount_t	*mp)	/* file system mount structure */
2215 {
2216 	xfs_buf_t	*bp;	/* buffer for last block of subvolume */
2217 	xfs_daddr_t	d;	/* address of last block of subvolume */
2218 	int		error;	/* error return value */
2219 	xfs_sb_t	*sbp;	/* filesystem superblock copy in mount */
2220 
2221 	sbp = &mp->m_sb;
2222 	if (sbp->sb_rblocks == 0)
2223 		return 0;
2224 	if (mp->m_rtdev_targp == NULL) {
2225 		cmn_err(CE_WARN,
2226 	"XFS: This filesystem has a realtime volume, use rtdev=device option");
2227 		return XFS_ERROR(ENODEV);
2228 	}
2229 	mp->m_rsumlevels = sbp->sb_rextslog + 1;
2230 	mp->m_rsumsize =
2231 		(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
2232 		sbp->sb_rbmblocks;
2233 	mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
2234 	mp->m_rbmip = mp->m_rsumip = NULL;
2235 	/*
2236 	 * Check that the realtime section is an ok size.
2237 	 */
2238 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
2239 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
2240 		cmn_err(CE_WARN, "XFS: realtime mount -- %llu != %llu",
2241 			(unsigned long long) XFS_BB_TO_FSB(mp, d),
2242 			(unsigned long long) mp->m_sb.sb_rblocks);
2243 		return XFS_ERROR(EFBIG);
2244 	}
2245 	error = xfs_read_buf(mp, mp->m_rtdev_targp,
2246 				d - XFS_FSB_TO_BB(mp, 1),
2247 				XFS_FSB_TO_BB(mp, 1), 0, &bp);
2248 	if (error) {
2249 		cmn_err(CE_WARN,
2250 	"XFS: realtime mount -- xfs_read_buf failed, returned %d", error);
2251 		if (error == ENOSPC)
2252 			return XFS_ERROR(EFBIG);
2253 		return error;
2254 	}
2255 	xfs_buf_relse(bp);
2256 	return 0;
2257 }
2258 
2259 /*
2260  * Get the bitmap and summary inodes into the mount structure
2261  * at mount time.
2262  */
2263 int					/* error */
2264 xfs_rtmount_inodes(
2265 	xfs_mount_t	*mp)		/* file system mount structure */
2266 {
2267 	int		error;		/* error return value */
2268 	xfs_sb_t	*sbp;
2269 
2270 	sbp = &mp->m_sb;
2271 	if (sbp->sb_rbmino == NULLFSINO)
2272 		return 0;
2273 	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
2274 	if (error)
2275 		return error;
2276 	ASSERT(mp->m_rbmip != NULL);
2277 	ASSERT(sbp->sb_rsumino != NULLFSINO);
2278 	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
2279 	if (error) {
2280 		IRELE(mp->m_rbmip);
2281 		return error;
2282 	}
2283 	ASSERT(mp->m_rsumip != NULL);
2284 	return 0;
2285 }
2286 
2287 void
2288 xfs_rtunmount_inodes(
2289 	struct xfs_mount	*mp)
2290 {
2291 	if (mp->m_rbmip)
2292 		IRELE(mp->m_rbmip);
2293 	if (mp->m_rsumip)
2294 		IRELE(mp->m_rsumip);
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