xref: /openbmc/linux/fs/ocfs2/localalloc.c (revision 64c70b1c)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * localalloc.c
5  *
6  * Node local data allocation
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/bitops.h>
31 
32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
34 
35 #include "ocfs2.h"
36 
37 #include "alloc.h"
38 #include "dlmglue.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "localalloc.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "sysfile.h"
45 
46 #include "buffer_head_io.h"
47 
48 #define OCFS2_LOCAL_ALLOC(dinode)	(&((dinode)->id2.i_lab))
49 
50 static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb);
51 
52 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
53 
54 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
55 					     struct ocfs2_dinode *alloc,
56 					     u32 numbits);
57 
58 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
59 
60 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
61 				    handle_t *handle,
62 				    struct ocfs2_dinode *alloc,
63 				    struct inode *main_bm_inode,
64 				    struct buffer_head *main_bm_bh);
65 
66 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
67 						struct ocfs2_alloc_context **ac,
68 						struct inode **bitmap_inode,
69 						struct buffer_head **bitmap_bh);
70 
71 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
72 					handle_t *handle,
73 					struct ocfs2_alloc_context *ac);
74 
75 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
76 					  struct inode *local_alloc_inode);
77 
78 /*
79  * Determine how large our local alloc window should be, in bits.
80  *
81  * These values (and the behavior in ocfs2_alloc_should_use_local) have
82  * been chosen so that most allocations, including new block groups go
83  * through local alloc.
84  */
85 static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb)
86 {
87 	BUG_ON(osb->s_clustersize_bits < 12);
88 
89 	return 2048 >> (osb->s_clustersize_bits - 12);
90 }
91 
92 /*
93  * Tell us whether a given allocation should use the local alloc
94  * file. Otherwise, it has to go to the main bitmap.
95  */
96 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
97 {
98 	int la_bits = ocfs2_local_alloc_window_bits(osb);
99 
100 	if (osb->local_alloc_state != OCFS2_LA_ENABLED)
101 		return 0;
102 
103 	/* la_bits should be at least twice the size (in clusters) of
104 	 * a new block group. We want to be sure block group
105 	 * allocations go through the local alloc, so allow an
106 	 * allocation to take up to half the bitmap. */
107 	if (bits > (la_bits / 2))
108 		return 0;
109 
110 	return 1;
111 }
112 
113 int ocfs2_load_local_alloc(struct ocfs2_super *osb)
114 {
115 	int status = 0;
116 	struct ocfs2_dinode *alloc = NULL;
117 	struct buffer_head *alloc_bh = NULL;
118 	u32 num_used;
119 	struct inode *inode = NULL;
120 	struct ocfs2_local_alloc *la;
121 
122 	mlog_entry_void();
123 
124 	/* read the alloc off disk */
125 	inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
126 					    osb->slot_num);
127 	if (!inode) {
128 		status = -EINVAL;
129 		mlog_errno(status);
130 		goto bail;
131 	}
132 
133 	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
134 				  &alloc_bh, 0, inode);
135 	if (status < 0) {
136 		mlog_errno(status);
137 		goto bail;
138 	}
139 
140 	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
141 	la = OCFS2_LOCAL_ALLOC(alloc);
142 
143 	if (!(le32_to_cpu(alloc->i_flags) &
144 	    (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
145 		mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
146 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
147 		status = -EINVAL;
148 		goto bail;
149 	}
150 
151 	if ((la->la_size == 0) ||
152 	    (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
153 		mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
154 		     le16_to_cpu(la->la_size));
155 		status = -EINVAL;
156 		goto bail;
157 	}
158 
159 	/* do a little verification. */
160 	num_used = ocfs2_local_alloc_count_bits(alloc);
161 
162 	/* hopefully the local alloc has always been recovered before
163 	 * we load it. */
164 	if (num_used
165 	    || alloc->id1.bitmap1.i_used
166 	    || alloc->id1.bitmap1.i_total
167 	    || la->la_bm_off)
168 		mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
169 		     "found = %u, set = %u, taken = %u, off = %u\n",
170 		     num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
171 		     le32_to_cpu(alloc->id1.bitmap1.i_total),
172 		     OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
173 
174 	osb->local_alloc_bh = alloc_bh;
175 	osb->local_alloc_state = OCFS2_LA_ENABLED;
176 
177 bail:
178 	if (status < 0)
179 		if (alloc_bh)
180 			brelse(alloc_bh);
181 	if (inode)
182 		iput(inode);
183 
184 	mlog_exit(status);
185 	return status;
186 }
187 
188 /*
189  * return any unused bits to the bitmap and write out a clean
190  * local_alloc.
191  *
192  * local_alloc_bh is optional. If not passed, we will simply use the
193  * one off osb. If you do pass it however, be warned that it *will* be
194  * returned brelse'd and NULL'd out.*/
195 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
196 {
197 	int status;
198 	handle_t *handle;
199 	struct inode *local_alloc_inode = NULL;
200 	struct buffer_head *bh = NULL;
201 	struct buffer_head *main_bm_bh = NULL;
202 	struct inode *main_bm_inode = NULL;
203 	struct ocfs2_dinode *alloc_copy = NULL;
204 	struct ocfs2_dinode *alloc = NULL;
205 
206 	mlog_entry_void();
207 
208 	if (osb->local_alloc_state == OCFS2_LA_UNUSED)
209 		goto out;
210 
211 	local_alloc_inode =
212 		ocfs2_get_system_file_inode(osb,
213 					    LOCAL_ALLOC_SYSTEM_INODE,
214 					    osb->slot_num);
215 	if (!local_alloc_inode) {
216 		status = -ENOENT;
217 		mlog_errno(status);
218 		goto out;
219 	}
220 
221 	osb->local_alloc_state = OCFS2_LA_DISABLED;
222 
223 	main_bm_inode = ocfs2_get_system_file_inode(osb,
224 						    GLOBAL_BITMAP_SYSTEM_INODE,
225 						    OCFS2_INVALID_SLOT);
226 	if (!main_bm_inode) {
227 		status = -EINVAL;
228 		mlog_errno(status);
229 		goto out;
230 	}
231 
232 	mutex_lock(&main_bm_inode->i_mutex);
233 
234 	status = ocfs2_meta_lock(main_bm_inode, &main_bm_bh, 1);
235 	if (status < 0) {
236 		mlog_errno(status);
237 		goto out_mutex;
238 	}
239 
240 	/* WINDOW_MOVE_CREDITS is a bit heavy... */
241 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
242 	if (IS_ERR(handle)) {
243 		mlog_errno(PTR_ERR(handle));
244 		handle = NULL;
245 		goto out_unlock;
246 	}
247 
248 	bh = osb->local_alloc_bh;
249 	alloc = (struct ocfs2_dinode *) bh->b_data;
250 
251 	alloc_copy = kmalloc(bh->b_size, GFP_KERNEL);
252 	if (!alloc_copy) {
253 		status = -ENOMEM;
254 		goto out_commit;
255 	}
256 	memcpy(alloc_copy, alloc, bh->b_size);
257 
258 	status = ocfs2_journal_access(handle, local_alloc_inode, bh,
259 				      OCFS2_JOURNAL_ACCESS_WRITE);
260 	if (status < 0) {
261 		mlog_errno(status);
262 		goto out_commit;
263 	}
264 
265 	ocfs2_clear_local_alloc(alloc);
266 
267 	status = ocfs2_journal_dirty(handle, bh);
268 	if (status < 0) {
269 		mlog_errno(status);
270 		goto out_commit;
271 	}
272 
273 	brelse(bh);
274 	osb->local_alloc_bh = NULL;
275 	osb->local_alloc_state = OCFS2_LA_UNUSED;
276 
277 	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
278 					  main_bm_inode, main_bm_bh);
279 	if (status < 0)
280 		mlog_errno(status);
281 
282 out_commit:
283 	ocfs2_commit_trans(osb, handle);
284 
285 out_unlock:
286 	if (main_bm_bh)
287 		brelse(main_bm_bh);
288 
289 	ocfs2_meta_unlock(main_bm_inode, 1);
290 
291 out_mutex:
292 	mutex_unlock(&main_bm_inode->i_mutex);
293 	iput(main_bm_inode);
294 
295 out:
296 	if (local_alloc_inode)
297 		iput(local_alloc_inode);
298 
299 	if (alloc_copy)
300 		kfree(alloc_copy);
301 
302 	mlog_exit_void();
303 }
304 
305 /*
306  * We want to free the bitmap bits outside of any recovery context as
307  * we'll need a cluster lock to do so, but we must clear the local
308  * alloc before giving up the recovered nodes journal. To solve this,
309  * we kmalloc a copy of the local alloc before it's change for the
310  * caller to process with ocfs2_complete_local_alloc_recovery
311  */
312 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
313 				     int slot_num,
314 				     struct ocfs2_dinode **alloc_copy)
315 {
316 	int status = 0;
317 	struct buffer_head *alloc_bh = NULL;
318 	struct inode *inode = NULL;
319 	struct ocfs2_dinode *alloc;
320 
321 	mlog_entry("(slot_num = %d)\n", slot_num);
322 
323 	*alloc_copy = NULL;
324 
325 	inode = ocfs2_get_system_file_inode(osb,
326 					    LOCAL_ALLOC_SYSTEM_INODE,
327 					    slot_num);
328 	if (!inode) {
329 		status = -EINVAL;
330 		mlog_errno(status);
331 		goto bail;
332 	}
333 
334 	mutex_lock(&inode->i_mutex);
335 
336 	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
337 				  &alloc_bh, 0, inode);
338 	if (status < 0) {
339 		mlog_errno(status);
340 		goto bail;
341 	}
342 
343 	*alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
344 	if (!(*alloc_copy)) {
345 		status = -ENOMEM;
346 		goto bail;
347 	}
348 	memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
349 
350 	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
351 	ocfs2_clear_local_alloc(alloc);
352 
353 	status = ocfs2_write_block(osb, alloc_bh, inode);
354 	if (status < 0)
355 		mlog_errno(status);
356 
357 bail:
358 	if ((status < 0) && (*alloc_copy)) {
359 		kfree(*alloc_copy);
360 		*alloc_copy = NULL;
361 	}
362 
363 	if (alloc_bh)
364 		brelse(alloc_bh);
365 
366 	if (inode) {
367 		mutex_unlock(&inode->i_mutex);
368 		iput(inode);
369 	}
370 
371 	mlog_exit(status);
372 	return status;
373 }
374 
375 /*
376  * Step 2: By now, we've completed the journal recovery, we've stamped
377  * a clean local alloc on disk and dropped the node out of the
378  * recovery map. Dlm locks will no longer stall, so lets clear out the
379  * main bitmap.
380  */
381 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
382 					struct ocfs2_dinode *alloc)
383 {
384 	int status;
385 	handle_t *handle;
386 	struct buffer_head *main_bm_bh = NULL;
387 	struct inode *main_bm_inode;
388 
389 	mlog_entry_void();
390 
391 	main_bm_inode = ocfs2_get_system_file_inode(osb,
392 						    GLOBAL_BITMAP_SYSTEM_INODE,
393 						    OCFS2_INVALID_SLOT);
394 	if (!main_bm_inode) {
395 		status = -EINVAL;
396 		mlog_errno(status);
397 		goto out;
398 	}
399 
400 	mutex_lock(&main_bm_inode->i_mutex);
401 
402 	status = ocfs2_meta_lock(main_bm_inode, &main_bm_bh, 1);
403 	if (status < 0) {
404 		mlog_errno(status);
405 		goto out_mutex;
406 	}
407 
408 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
409 	if (IS_ERR(handle)) {
410 		status = PTR_ERR(handle);
411 		handle = NULL;
412 		mlog_errno(status);
413 		goto out_unlock;
414 	}
415 
416 	/* we want the bitmap change to be recorded on disk asap */
417 	handle->h_sync = 1;
418 
419 	status = ocfs2_sync_local_to_main(osb, handle, alloc,
420 					  main_bm_inode, main_bm_bh);
421 	if (status < 0)
422 		mlog_errno(status);
423 
424 	ocfs2_commit_trans(osb, handle);
425 
426 out_unlock:
427 	ocfs2_meta_unlock(main_bm_inode, 1);
428 
429 out_mutex:
430 	mutex_unlock(&main_bm_inode->i_mutex);
431 
432 	if (main_bm_bh)
433 		brelse(main_bm_bh);
434 
435 	iput(main_bm_inode);
436 
437 out:
438 	mlog_exit(status);
439 	return status;
440 }
441 
442 /*
443  * make sure we've got at least bitswanted contiguous bits in the
444  * local alloc. You lose them when you drop i_mutex.
445  *
446  * We will add ourselves to the transaction passed in, but may start
447  * our own in order to shift windows.
448  */
449 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
450 				   u32 bits_wanted,
451 				   struct ocfs2_alloc_context *ac)
452 {
453 	int status;
454 	struct ocfs2_dinode *alloc;
455 	struct inode *local_alloc_inode;
456 	unsigned int free_bits;
457 
458 	mlog_entry_void();
459 
460 	BUG_ON(!ac);
461 
462 	local_alloc_inode =
463 		ocfs2_get_system_file_inode(osb,
464 					    LOCAL_ALLOC_SYSTEM_INODE,
465 					    osb->slot_num);
466 	if (!local_alloc_inode) {
467 		status = -ENOENT;
468 		mlog_errno(status);
469 		goto bail;
470 	}
471 
472 	mutex_lock(&local_alloc_inode->i_mutex);
473 
474 	if (osb->local_alloc_state != OCFS2_LA_ENABLED) {
475 		status = -ENOSPC;
476 		goto bail;
477 	}
478 
479 	if (bits_wanted > ocfs2_local_alloc_window_bits(osb)) {
480 		mlog(0, "Asking for more than my max window size!\n");
481 		status = -ENOSPC;
482 		goto bail;
483 	}
484 
485 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
486 
487 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
488 	    ocfs2_local_alloc_count_bits(alloc)) {
489 		ocfs2_error(osb->sb, "local alloc inode %llu says it has "
490 			    "%u free bits, but a count shows %u",
491 			    (unsigned long long)le64_to_cpu(alloc->i_blkno),
492 			    le32_to_cpu(alloc->id1.bitmap1.i_used),
493 			    ocfs2_local_alloc_count_bits(alloc));
494 		status = -EIO;
495 		goto bail;
496 	}
497 
498 	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
499 		le32_to_cpu(alloc->id1.bitmap1.i_used);
500 	if (bits_wanted > free_bits) {
501 		/* uhoh, window change time. */
502 		status =
503 			ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
504 		if (status < 0) {
505 			if (status != -ENOSPC)
506 				mlog_errno(status);
507 			goto bail;
508 		}
509 	}
510 
511 	ac->ac_inode = local_alloc_inode;
512 	ac->ac_which = OCFS2_AC_USE_LOCAL;
513 	get_bh(osb->local_alloc_bh);
514 	ac->ac_bh = osb->local_alloc_bh;
515 	status = 0;
516 bail:
517 	if (status < 0 && local_alloc_inode)
518 		iput(local_alloc_inode);
519 
520 	mlog_exit(status);
521 	return status;
522 }
523 
524 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
525 				 handle_t *handle,
526 				 struct ocfs2_alloc_context *ac,
527 				 u32 min_bits,
528 				 u32 *bit_off,
529 				 u32 *num_bits)
530 {
531 	int status, start;
532 	struct inode *local_alloc_inode;
533 	u32 bits_wanted;
534 	void *bitmap;
535 	struct ocfs2_dinode *alloc;
536 	struct ocfs2_local_alloc *la;
537 
538 	mlog_entry_void();
539 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
540 
541 	bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
542 	local_alloc_inode = ac->ac_inode;
543 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
544 	la = OCFS2_LOCAL_ALLOC(alloc);
545 
546 	start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
547 	if (start == -1) {
548 		/* TODO: Shouldn't we just BUG here? */
549 		status = -ENOSPC;
550 		mlog_errno(status);
551 		goto bail;
552 	}
553 
554 	bitmap = la->la_bitmap;
555 	*bit_off = le32_to_cpu(la->la_bm_off) + start;
556 	/* local alloc is always contiguous by nature -- we never
557 	 * delete bits from it! */
558 	*num_bits = bits_wanted;
559 
560 	status = ocfs2_journal_access(handle, local_alloc_inode,
561 				      osb->local_alloc_bh,
562 				      OCFS2_JOURNAL_ACCESS_WRITE);
563 	if (status < 0) {
564 		mlog_errno(status);
565 		goto bail;
566 	}
567 
568 	while(bits_wanted--)
569 		ocfs2_set_bit(start++, bitmap);
570 
571 	alloc->id1.bitmap1.i_used = cpu_to_le32(*num_bits +
572 				le32_to_cpu(alloc->id1.bitmap1.i_used));
573 
574 	status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
575 	if (status < 0) {
576 		mlog_errno(status);
577 		goto bail;
578 	}
579 
580 	status = 0;
581 bail:
582 	mlog_exit(status);
583 	return status;
584 }
585 
586 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
587 {
588 	int i;
589 	u8 *buffer;
590 	u32 count = 0;
591 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
592 
593 	mlog_entry_void();
594 
595 	buffer = la->la_bitmap;
596 	for (i = 0; i < le16_to_cpu(la->la_size); i++)
597 		count += hweight8(buffer[i]);
598 
599 	mlog_exit(count);
600 	return count;
601 }
602 
603 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
604 					     struct ocfs2_dinode *alloc,
605 					     u32 numbits)
606 {
607 	int numfound, bitoff, left, startoff, lastzero;
608 	void *bitmap = NULL;
609 
610 	mlog_entry("(numbits wanted = %u)\n", numbits);
611 
612 	if (!alloc->id1.bitmap1.i_total) {
613 		mlog(0, "No bits in my window!\n");
614 		bitoff = -1;
615 		goto bail;
616 	}
617 
618 	bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
619 
620 	numfound = bitoff = startoff = 0;
621 	lastzero = -1;
622 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
623 	while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
624 		if (bitoff == left) {
625 			/* mlog(0, "bitoff (%d) == left", bitoff); */
626 			break;
627 		}
628 		/* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
629 		   "numfound = %d\n", bitoff, startoff, numfound);*/
630 
631 		/* Ok, we found a zero bit... is it contig. or do we
632 		 * start over?*/
633 		if (bitoff == startoff) {
634 			/* we found a zero */
635 			numfound++;
636 			startoff++;
637 		} else {
638 			/* got a zero after some ones */
639 			numfound = 1;
640 			startoff = bitoff+1;
641 		}
642 		/* we got everything we needed */
643 		if (numfound == numbits) {
644 			/* mlog(0, "Found it all!\n"); */
645 			break;
646 		}
647 	}
648 
649 	mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff,
650 	     numfound);
651 
652 	if (numfound == numbits)
653 		bitoff = startoff - numfound;
654 	else
655 		bitoff = -1;
656 
657 bail:
658 	mlog_exit(bitoff);
659 	return bitoff;
660 }
661 
662 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
663 {
664 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
665 	int i;
666 	mlog_entry_void();
667 
668 	alloc->id1.bitmap1.i_total = 0;
669 	alloc->id1.bitmap1.i_used = 0;
670 	la->la_bm_off = 0;
671 	for(i = 0; i < le16_to_cpu(la->la_size); i++)
672 		la->la_bitmap[i] = 0;
673 
674 	mlog_exit_void();
675 }
676 
677 #if 0
678 /* turn this on and uncomment below to aid debugging window shifts. */
679 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
680 				   unsigned int start,
681 				   unsigned int count)
682 {
683 	unsigned int tmp = count;
684 	while(tmp--) {
685 		if (ocfs2_test_bit(start + tmp, bitmap)) {
686 			printk("ocfs2_verify_zero_bits: start = %u, count = "
687 			       "%u\n", start, count);
688 			printk("ocfs2_verify_zero_bits: bit %u is set!",
689 			       start + tmp);
690 			BUG();
691 		}
692 	}
693 }
694 #endif
695 
696 /*
697  * sync the local alloc to main bitmap.
698  *
699  * assumes you've already locked the main bitmap -- the bitmap inode
700  * passed is used for caching.
701  */
702 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
703 				    handle_t *handle,
704 				    struct ocfs2_dinode *alloc,
705 				    struct inode *main_bm_inode,
706 				    struct buffer_head *main_bm_bh)
707 {
708 	int status = 0;
709 	int bit_off, left, count, start;
710 	u64 la_start_blk;
711 	u64 blkno;
712 	void *bitmap;
713 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
714 
715 	mlog_entry("total = %u, COUNT = %u, used = %u\n",
716 		   le32_to_cpu(alloc->id1.bitmap1.i_total),
717 		   ocfs2_local_alloc_count_bits(alloc),
718 		   le32_to_cpu(alloc->id1.bitmap1.i_used));
719 
720 	if (!alloc->id1.bitmap1.i_total) {
721 		mlog(0, "nothing to sync!\n");
722 		goto bail;
723 	}
724 
725 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
726 	    le32_to_cpu(alloc->id1.bitmap1.i_total)) {
727 		mlog(0, "all bits were taken!\n");
728 		goto bail;
729 	}
730 
731 	la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
732 						le32_to_cpu(la->la_bm_off));
733 	bitmap = la->la_bitmap;
734 	start = count = bit_off = 0;
735 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
736 
737 	while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
738 	       != -1) {
739 		if ((bit_off < left) && (bit_off == start)) {
740 			count++;
741 			start++;
742 			continue;
743 		}
744 		if (count) {
745 			blkno = la_start_blk +
746 				ocfs2_clusters_to_blocks(osb->sb,
747 							 start - count);
748 
749 			mlog(0, "freeing %u bits starting at local alloc bit "
750 			     "%u (la_start_blk = %llu, blkno = %llu)\n",
751 			     count, start - count,
752 			     (unsigned long long)la_start_blk,
753 			     (unsigned long long)blkno);
754 
755 			status = ocfs2_free_clusters(handle, main_bm_inode,
756 						     main_bm_bh, blkno, count);
757 			if (status < 0) {
758 				mlog_errno(status);
759 				goto bail;
760 			}
761 		}
762 		if (bit_off >= left)
763 			break;
764 		count = 1;
765 		start = bit_off + 1;
766 	}
767 
768 bail:
769 	mlog_exit(status);
770 	return status;
771 }
772 
773 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
774 						struct ocfs2_alloc_context **ac,
775 						struct inode **bitmap_inode,
776 						struct buffer_head **bitmap_bh)
777 {
778 	int status;
779 
780 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
781 	if (!(*ac)) {
782 		status = -ENOMEM;
783 		mlog_errno(status);
784 		goto bail;
785 	}
786 
787 	(*ac)->ac_bits_wanted = ocfs2_local_alloc_window_bits(osb);
788 
789 	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
790 	if (status < 0) {
791 		if (status != -ENOSPC)
792 			mlog_errno(status);
793 		goto bail;
794 	}
795 
796 	*bitmap_inode = (*ac)->ac_inode;
797 	igrab(*bitmap_inode);
798 	*bitmap_bh = (*ac)->ac_bh;
799 	get_bh(*bitmap_bh);
800 	status = 0;
801 bail:
802 	if ((status < 0) && *ac) {
803 		ocfs2_free_alloc_context(*ac);
804 		*ac = NULL;
805 	}
806 
807 	mlog_exit(status);
808 	return status;
809 }
810 
811 /*
812  * pass it the bitmap lock in lock_bh if you have it.
813  */
814 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
815 					handle_t *handle,
816 					struct ocfs2_alloc_context *ac)
817 {
818 	int status = 0;
819 	u32 cluster_off, cluster_count;
820 	struct ocfs2_dinode *alloc = NULL;
821 	struct ocfs2_local_alloc *la;
822 
823 	mlog_entry_void();
824 
825 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
826 	la = OCFS2_LOCAL_ALLOC(alloc);
827 
828 	if (alloc->id1.bitmap1.i_total)
829 		mlog(0, "asking me to alloc a new window over a non-empty "
830 		     "one\n");
831 
832 	mlog(0, "Allocating %u clusters for a new window.\n",
833 	     ocfs2_local_alloc_window_bits(osb));
834 
835 	/* Instruct the allocation code to try the most recently used
836 	 * cluster group. We'll re-record the group used this pass
837 	 * below. */
838 	ac->ac_last_group = osb->la_last_gd;
839 
840 	/* we used the generic suballoc reserve function, but we set
841 	 * everything up nicely, so there's no reason why we can't use
842 	 * the more specific cluster api to claim bits. */
843 	status = ocfs2_claim_clusters(osb, handle, ac,
844 				      ocfs2_local_alloc_window_bits(osb),
845 				      &cluster_off, &cluster_count);
846 	if (status < 0) {
847 		if (status != -ENOSPC)
848 			mlog_errno(status);
849 		goto bail;
850 	}
851 
852 	osb->la_last_gd = ac->ac_last_group;
853 
854 	la->la_bm_off = cpu_to_le32(cluster_off);
855 	alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
856 	/* just in case... In the future when we find space ourselves,
857 	 * we don't have to get all contiguous -- but we'll have to
858 	 * set all previously used bits in bitmap and update
859 	 * la_bits_set before setting the bits in the main bitmap. */
860 	alloc->id1.bitmap1.i_used = 0;
861 	memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
862 	       le16_to_cpu(la->la_size));
863 
864 	mlog(0, "New window allocated:\n");
865 	mlog(0, "window la_bm_off = %u\n",
866 	     OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
867 	mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total));
868 
869 bail:
870 	mlog_exit(status);
871 	return status;
872 }
873 
874 /* Note that we do *NOT* lock the local alloc inode here as
875  * it's been locked already for us. */
876 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
877 					  struct inode *local_alloc_inode)
878 {
879 	int status = 0;
880 	struct buffer_head *main_bm_bh = NULL;
881 	struct inode *main_bm_inode = NULL;
882 	handle_t *handle = NULL;
883 	struct ocfs2_dinode *alloc;
884 	struct ocfs2_dinode *alloc_copy = NULL;
885 	struct ocfs2_alloc_context *ac = NULL;
886 
887 	mlog_entry_void();
888 
889 	/* This will lock the main bitmap for us. */
890 	status = ocfs2_local_alloc_reserve_for_window(osb,
891 						      &ac,
892 						      &main_bm_inode,
893 						      &main_bm_bh);
894 	if (status < 0) {
895 		if (status != -ENOSPC)
896 			mlog_errno(status);
897 		goto bail;
898 	}
899 
900 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
901 	if (IS_ERR(handle)) {
902 		status = PTR_ERR(handle);
903 		handle = NULL;
904 		mlog_errno(status);
905 		goto bail;
906 	}
907 
908 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
909 
910 	/* We want to clear the local alloc before doing anything
911 	 * else, so that if we error later during this operation,
912 	 * local alloc shutdown won't try to double free main bitmap
913 	 * bits. Make a copy so the sync function knows which bits to
914 	 * free. */
915 	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_KERNEL);
916 	if (!alloc_copy) {
917 		status = -ENOMEM;
918 		mlog_errno(status);
919 		goto bail;
920 	}
921 	memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
922 
923 	status = ocfs2_journal_access(handle, local_alloc_inode,
924 				      osb->local_alloc_bh,
925 				      OCFS2_JOURNAL_ACCESS_WRITE);
926 	if (status < 0) {
927 		mlog_errno(status);
928 		goto bail;
929 	}
930 
931 	ocfs2_clear_local_alloc(alloc);
932 
933 	status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
934 	if (status < 0) {
935 		mlog_errno(status);
936 		goto bail;
937 	}
938 
939 	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
940 					  main_bm_inode, main_bm_bh);
941 	if (status < 0) {
942 		mlog_errno(status);
943 		goto bail;
944 	}
945 
946 	status = ocfs2_local_alloc_new_window(osb, handle, ac);
947 	if (status < 0) {
948 		if (status != -ENOSPC)
949 			mlog_errno(status);
950 		goto bail;
951 	}
952 
953 	atomic_inc(&osb->alloc_stats.moves);
954 
955 	status = 0;
956 bail:
957 	if (handle)
958 		ocfs2_commit_trans(osb, handle);
959 
960 	if (main_bm_bh)
961 		brelse(main_bm_bh);
962 
963 	if (main_bm_inode)
964 		iput(main_bm_inode);
965 
966 	if (alloc_copy)
967 		kfree(alloc_copy);
968 
969 	if (ac)
970 		ocfs2_free_alloc_context(ac);
971 
972 	mlog_exit(status);
973 	return status;
974 }
975 
976