xref: /linux/fs/ocfs2/localalloc.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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 	ac->ac_inode = local_alloc_inode;
475 	ac->ac_which = OCFS2_AC_USE_LOCAL;
476 
477 	if (osb->local_alloc_state != OCFS2_LA_ENABLED) {
478 		status = -ENOSPC;
479 		goto bail;
480 	}
481 
482 	if (bits_wanted > ocfs2_local_alloc_window_bits(osb)) {
483 		mlog(0, "Asking for more than my max window size!\n");
484 		status = -ENOSPC;
485 		goto bail;
486 	}
487 
488 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
489 
490 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
491 	    ocfs2_local_alloc_count_bits(alloc)) {
492 		ocfs2_error(osb->sb, "local alloc inode %llu says it has "
493 			    "%u free bits, but a count shows %u",
494 			    (unsigned long long)le64_to_cpu(alloc->i_blkno),
495 			    le32_to_cpu(alloc->id1.bitmap1.i_used),
496 			    ocfs2_local_alloc_count_bits(alloc));
497 		status = -EIO;
498 		goto bail;
499 	}
500 
501 	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
502 		le32_to_cpu(alloc->id1.bitmap1.i_used);
503 	if (bits_wanted > free_bits) {
504 		/* uhoh, window change time. */
505 		status =
506 			ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
507 		if (status < 0) {
508 			if (status != -ENOSPC)
509 				mlog_errno(status);
510 			goto bail;
511 		}
512 	}
513 
514 	get_bh(osb->local_alloc_bh);
515 	ac->ac_bh = osb->local_alloc_bh;
516 	status = 0;
517 bail:
518 
519 	mlog_exit(status);
520 	return status;
521 }
522 
523 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
524 				 handle_t *handle,
525 				 struct ocfs2_alloc_context *ac,
526 				 u32 min_bits,
527 				 u32 *bit_off,
528 				 u32 *num_bits)
529 {
530 	int status, start;
531 	struct inode *local_alloc_inode;
532 	u32 bits_wanted;
533 	void *bitmap;
534 	struct ocfs2_dinode *alloc;
535 	struct ocfs2_local_alloc *la;
536 
537 	mlog_entry_void();
538 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
539 
540 	bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
541 	local_alloc_inode = ac->ac_inode;
542 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
543 	la = OCFS2_LOCAL_ALLOC(alloc);
544 
545 	start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
546 	if (start == -1) {
547 		/* TODO: Shouldn't we just BUG here? */
548 		status = -ENOSPC;
549 		mlog_errno(status);
550 		goto bail;
551 	}
552 
553 	bitmap = la->la_bitmap;
554 	*bit_off = le32_to_cpu(la->la_bm_off) + start;
555 	/* local alloc is always contiguous by nature -- we never
556 	 * delete bits from it! */
557 	*num_bits = bits_wanted;
558 
559 	status = ocfs2_journal_access(handle, local_alloc_inode,
560 				      osb->local_alloc_bh,
561 				      OCFS2_JOURNAL_ACCESS_WRITE);
562 	if (status < 0) {
563 		mlog_errno(status);
564 		goto bail;
565 	}
566 
567 	while(bits_wanted--)
568 		ocfs2_set_bit(start++, bitmap);
569 
570 	alloc->id1.bitmap1.i_used = cpu_to_le32(*num_bits +
571 				le32_to_cpu(alloc->id1.bitmap1.i_used));
572 
573 	status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
574 	if (status < 0) {
575 		mlog_errno(status);
576 		goto bail;
577 	}
578 
579 	status = 0;
580 bail:
581 	mlog_exit(status);
582 	return status;
583 }
584 
585 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
586 {
587 	int i;
588 	u8 *buffer;
589 	u32 count = 0;
590 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
591 
592 	mlog_entry_void();
593 
594 	buffer = la->la_bitmap;
595 	for (i = 0; i < le16_to_cpu(la->la_size); i++)
596 		count += hweight8(buffer[i]);
597 
598 	mlog_exit(count);
599 	return count;
600 }
601 
602 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
603 					     struct ocfs2_dinode *alloc,
604 					     u32 numbits)
605 {
606 	int numfound, bitoff, left, startoff, lastzero;
607 	void *bitmap = NULL;
608 
609 	mlog_entry("(numbits wanted = %u)\n", numbits);
610 
611 	if (!alloc->id1.bitmap1.i_total) {
612 		mlog(0, "No bits in my window!\n");
613 		bitoff = -1;
614 		goto bail;
615 	}
616 
617 	bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
618 
619 	numfound = bitoff = startoff = 0;
620 	lastzero = -1;
621 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
622 	while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
623 		if (bitoff == left) {
624 			/* mlog(0, "bitoff (%d) == left", bitoff); */
625 			break;
626 		}
627 		/* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
628 		   "numfound = %d\n", bitoff, startoff, numfound);*/
629 
630 		/* Ok, we found a zero bit... is it contig. or do we
631 		 * start over?*/
632 		if (bitoff == startoff) {
633 			/* we found a zero */
634 			numfound++;
635 			startoff++;
636 		} else {
637 			/* got a zero after some ones */
638 			numfound = 1;
639 			startoff = bitoff+1;
640 		}
641 		/* we got everything we needed */
642 		if (numfound == numbits) {
643 			/* mlog(0, "Found it all!\n"); */
644 			break;
645 		}
646 	}
647 
648 	mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff,
649 	     numfound);
650 
651 	if (numfound == numbits)
652 		bitoff = startoff - numfound;
653 	else
654 		bitoff = -1;
655 
656 bail:
657 	mlog_exit(bitoff);
658 	return bitoff;
659 }
660 
661 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
662 {
663 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
664 	int i;
665 	mlog_entry_void();
666 
667 	alloc->id1.bitmap1.i_total = 0;
668 	alloc->id1.bitmap1.i_used = 0;
669 	la->la_bm_off = 0;
670 	for(i = 0; i < le16_to_cpu(la->la_size); i++)
671 		la->la_bitmap[i] = 0;
672 
673 	mlog_exit_void();
674 }
675 
676 #if 0
677 /* turn this on and uncomment below to aid debugging window shifts. */
678 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
679 				   unsigned int start,
680 				   unsigned int count)
681 {
682 	unsigned int tmp = count;
683 	while(tmp--) {
684 		if (ocfs2_test_bit(start + tmp, bitmap)) {
685 			printk("ocfs2_verify_zero_bits: start = %u, count = "
686 			       "%u\n", start, count);
687 			printk("ocfs2_verify_zero_bits: bit %u is set!",
688 			       start + tmp);
689 			BUG();
690 		}
691 	}
692 }
693 #endif
694 
695 /*
696  * sync the local alloc to main bitmap.
697  *
698  * assumes you've already locked the main bitmap -- the bitmap inode
699  * passed is used for caching.
700  */
701 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
702 				    handle_t *handle,
703 				    struct ocfs2_dinode *alloc,
704 				    struct inode *main_bm_inode,
705 				    struct buffer_head *main_bm_bh)
706 {
707 	int status = 0;
708 	int bit_off, left, count, start;
709 	u64 la_start_blk;
710 	u64 blkno;
711 	void *bitmap;
712 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
713 
714 	mlog_entry("total = %u, COUNT = %u, used = %u\n",
715 		   le32_to_cpu(alloc->id1.bitmap1.i_total),
716 		   ocfs2_local_alloc_count_bits(alloc),
717 		   le32_to_cpu(alloc->id1.bitmap1.i_used));
718 
719 	if (!alloc->id1.bitmap1.i_total) {
720 		mlog(0, "nothing to sync!\n");
721 		goto bail;
722 	}
723 
724 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
725 	    le32_to_cpu(alloc->id1.bitmap1.i_total)) {
726 		mlog(0, "all bits were taken!\n");
727 		goto bail;
728 	}
729 
730 	la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
731 						le32_to_cpu(la->la_bm_off));
732 	bitmap = la->la_bitmap;
733 	start = count = bit_off = 0;
734 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
735 
736 	while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
737 	       != -1) {
738 		if ((bit_off < left) && (bit_off == start)) {
739 			count++;
740 			start++;
741 			continue;
742 		}
743 		if (count) {
744 			blkno = la_start_blk +
745 				ocfs2_clusters_to_blocks(osb->sb,
746 							 start - count);
747 
748 			mlog(0, "freeing %u bits starting at local alloc bit "
749 			     "%u (la_start_blk = %llu, blkno = %llu)\n",
750 			     count, start - count,
751 			     (unsigned long long)la_start_blk,
752 			     (unsigned long long)blkno);
753 
754 			status = ocfs2_free_clusters(handle, main_bm_inode,
755 						     main_bm_bh, blkno, count);
756 			if (status < 0) {
757 				mlog_errno(status);
758 				goto bail;
759 			}
760 		}
761 		if (bit_off >= left)
762 			break;
763 		count = 1;
764 		start = bit_off + 1;
765 	}
766 
767 bail:
768 	mlog_exit(status);
769 	return status;
770 }
771 
772 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
773 						struct ocfs2_alloc_context **ac,
774 						struct inode **bitmap_inode,
775 						struct buffer_head **bitmap_bh)
776 {
777 	int status;
778 
779 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
780 	if (!(*ac)) {
781 		status = -ENOMEM;
782 		mlog_errno(status);
783 		goto bail;
784 	}
785 
786 	(*ac)->ac_bits_wanted = ocfs2_local_alloc_window_bits(osb);
787 
788 	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
789 	if (status < 0) {
790 		if (status != -ENOSPC)
791 			mlog_errno(status);
792 		goto bail;
793 	}
794 
795 	*bitmap_inode = (*ac)->ac_inode;
796 	igrab(*bitmap_inode);
797 	*bitmap_bh = (*ac)->ac_bh;
798 	get_bh(*bitmap_bh);
799 	status = 0;
800 bail:
801 	if ((status < 0) && *ac) {
802 		ocfs2_free_alloc_context(*ac);
803 		*ac = NULL;
804 	}
805 
806 	mlog_exit(status);
807 	return status;
808 }
809 
810 /*
811  * pass it the bitmap lock in lock_bh if you have it.
812  */
813 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
814 					handle_t *handle,
815 					struct ocfs2_alloc_context *ac)
816 {
817 	int status = 0;
818 	u32 cluster_off, cluster_count;
819 	struct ocfs2_dinode *alloc = NULL;
820 	struct ocfs2_local_alloc *la;
821 
822 	mlog_entry_void();
823 
824 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
825 	la = OCFS2_LOCAL_ALLOC(alloc);
826 
827 	if (alloc->id1.bitmap1.i_total)
828 		mlog(0, "asking me to alloc a new window over a non-empty "
829 		     "one\n");
830 
831 	mlog(0, "Allocating %u clusters for a new window.\n",
832 	     ocfs2_local_alloc_window_bits(osb));
833 
834 	/* Instruct the allocation code to try the most recently used
835 	 * cluster group. We'll re-record the group used this pass
836 	 * below. */
837 	ac->ac_last_group = osb->la_last_gd;
838 
839 	/* we used the generic suballoc reserve function, but we set
840 	 * everything up nicely, so there's no reason why we can't use
841 	 * the more specific cluster api to claim bits. */
842 	status = ocfs2_claim_clusters(osb, handle, ac,
843 				      ocfs2_local_alloc_window_bits(osb),
844 				      &cluster_off, &cluster_count);
845 	if (status < 0) {
846 		if (status != -ENOSPC)
847 			mlog_errno(status);
848 		goto bail;
849 	}
850 
851 	osb->la_last_gd = ac->ac_last_group;
852 
853 	la->la_bm_off = cpu_to_le32(cluster_off);
854 	alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
855 	/* just in case... In the future when we find space ourselves,
856 	 * we don't have to get all contiguous -- but we'll have to
857 	 * set all previously used bits in bitmap and update
858 	 * la_bits_set before setting the bits in the main bitmap. */
859 	alloc->id1.bitmap1.i_used = 0;
860 	memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
861 	       le16_to_cpu(la->la_size));
862 
863 	mlog(0, "New window allocated:\n");
864 	mlog(0, "window la_bm_off = %u\n",
865 	     OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
866 	mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total));
867 
868 bail:
869 	mlog_exit(status);
870 	return status;
871 }
872 
873 /* Note that we do *NOT* lock the local alloc inode here as
874  * it's been locked already for us. */
875 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
876 					  struct inode *local_alloc_inode)
877 {
878 	int status = 0;
879 	struct buffer_head *main_bm_bh = NULL;
880 	struct inode *main_bm_inode = NULL;
881 	handle_t *handle = NULL;
882 	struct ocfs2_dinode *alloc;
883 	struct ocfs2_dinode *alloc_copy = NULL;
884 	struct ocfs2_alloc_context *ac = NULL;
885 
886 	mlog_entry_void();
887 
888 	/* This will lock the main bitmap for us. */
889 	status = ocfs2_local_alloc_reserve_for_window(osb,
890 						      &ac,
891 						      &main_bm_inode,
892 						      &main_bm_bh);
893 	if (status < 0) {
894 		if (status != -ENOSPC)
895 			mlog_errno(status);
896 		goto bail;
897 	}
898 
899 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
900 	if (IS_ERR(handle)) {
901 		status = PTR_ERR(handle);
902 		handle = NULL;
903 		mlog_errno(status);
904 		goto bail;
905 	}
906 
907 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
908 
909 	/* We want to clear the local alloc before doing anything
910 	 * else, so that if we error later during this operation,
911 	 * local alloc shutdown won't try to double free main bitmap
912 	 * bits. Make a copy so the sync function knows which bits to
913 	 * free. */
914 	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_KERNEL);
915 	if (!alloc_copy) {
916 		status = -ENOMEM;
917 		mlog_errno(status);
918 		goto bail;
919 	}
920 	memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
921 
922 	status = ocfs2_journal_access(handle, local_alloc_inode,
923 				      osb->local_alloc_bh,
924 				      OCFS2_JOURNAL_ACCESS_WRITE);
925 	if (status < 0) {
926 		mlog_errno(status);
927 		goto bail;
928 	}
929 
930 	ocfs2_clear_local_alloc(alloc);
931 
932 	status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
933 	if (status < 0) {
934 		mlog_errno(status);
935 		goto bail;
936 	}
937 
938 	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
939 					  main_bm_inode, main_bm_bh);
940 	if (status < 0) {
941 		mlog_errno(status);
942 		goto bail;
943 	}
944 
945 	status = ocfs2_local_alloc_new_window(osb, handle, ac);
946 	if (status < 0) {
947 		if (status != -ENOSPC)
948 			mlog_errno(status);
949 		goto bail;
950 	}
951 
952 	atomic_inc(&osb->alloc_stats.moves);
953 
954 	status = 0;
955 bail:
956 	if (handle)
957 		ocfs2_commit_trans(osb, handle);
958 
959 	if (main_bm_bh)
960 		brelse(main_bm_bh);
961 
962 	if (main_bm_inode)
963 		iput(main_bm_inode);
964 
965 	if (alloc_copy)
966 		kfree(alloc_copy);
967 
968 	if (ac)
969 		ocfs2_free_alloc_context(ac);
970 
971 	mlog_exit(status);
972 	return status;
973 }
974 
975