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 #include <cluster/masklog.h>
33 
34 #include "ocfs2.h"
35 
36 #include "alloc.h"
37 #include "blockcheck.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 #include "ocfs2_trace.h"
46 
47 #include "buffer_head_io.h"
48 
49 #define OCFS2_LOCAL_ALLOC(dinode)	(&((dinode)->id2.i_lab))
50 
51 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
52 
53 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
54 					     struct ocfs2_dinode *alloc,
55 					     u32 *numbits,
56 					     struct ocfs2_alloc_reservation *resv);
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  * ocfs2_la_default_mb() - determine a default size, in megabytes of
80  * the local alloc.
81  *
82  * Generally, we'd like to pick as large a local alloc as
83  * possible. Performance on large workloads tends to scale
84  * proportionally to la size. In addition to that, the reservations
85  * code functions more efficiently as it can reserve more windows for
86  * write.
87  *
88  * Some things work against us when trying to choose a large local alloc:
89  *
90  * - We need to ensure our sizing is picked to leave enough space in
91  *   group descriptors for other allocations (such as block groups,
92  *   etc). Picking default sizes which are a multiple of 4 could help
93  *   - block groups are allocated in 2mb and 4mb chunks.
94  *
95  * - Likewise, we don't want to starve other nodes of bits on small
96  *   file systems. This can easily be taken care of by limiting our
97  *   default to a reasonable size (256M) on larger cluster sizes.
98  *
99  * - Some file systems can't support very large sizes - 4k and 8k in
100  *   particular are limited to less than 128 and 256 megabytes respectively.
101  *
102  * The following reference table shows group descriptor and local
103  * alloc maximums at various cluster sizes (4k blocksize)
104  *
105  * csize: 4K	group: 126M	la: 121M
106  * csize: 8K	group: 252M	la: 243M
107  * csize: 16K	group: 504M	la: 486M
108  * csize: 32K	group: 1008M	la: 972M
109  * csize: 64K	group: 2016M	la: 1944M
110  * csize: 128K	group: 4032M	la: 3888M
111  * csize: 256K	group: 8064M	la: 7776M
112  * csize: 512K	group: 16128M	la: 15552M
113  * csize: 1024K	group: 32256M	la: 31104M
114  */
115 #define	OCFS2_LA_MAX_DEFAULT_MB	256
116 #define	OCFS2_LA_OLD_DEFAULT	8
ocfs2_la_default_mb(struct ocfs2_super * osb)117 unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb)
118 {
119 	unsigned int la_mb;
120 	unsigned int gd_mb;
121 	unsigned int la_max_mb;
122 	unsigned int megs_per_slot;
123 	struct super_block *sb = osb->sb;
124 
125 	gd_mb = ocfs2_clusters_to_megabytes(osb->sb,
126 		8 * ocfs2_group_bitmap_size(sb, 0, osb->s_feature_incompat));
127 
128 	/*
129 	 * This takes care of files systems with very small group
130 	 * descriptors - 512 byte blocksize at cluster sizes lower
131 	 * than 16K and also 1k blocksize with 4k cluster size.
132 	 */
133 	if ((sb->s_blocksize == 512 && osb->s_clustersize <= 8192)
134 	    || (sb->s_blocksize == 1024 && osb->s_clustersize == 4096))
135 		return OCFS2_LA_OLD_DEFAULT;
136 
137 	/*
138 	 * Leave enough room for some block groups and make the final
139 	 * value we work from a multiple of 4.
140 	 */
141 	gd_mb -= 16;
142 	gd_mb &= 0xFFFFFFFB;
143 
144 	la_mb = gd_mb;
145 
146 	/*
147 	 * Keep window sizes down to a reasonable default
148 	 */
149 	if (la_mb > OCFS2_LA_MAX_DEFAULT_MB) {
150 		/*
151 		 * Some clustersize / blocksize combinations will have
152 		 * given us a larger than OCFS2_LA_MAX_DEFAULT_MB
153 		 * default size, but get poor distribution when
154 		 * limited to exactly 256 megabytes.
155 		 *
156 		 * As an example, 16K clustersize at 4K blocksize
157 		 * gives us a cluster group size of 504M. Paring the
158 		 * local alloc size down to 256 however, would give us
159 		 * only one window and around 200MB left in the
160 		 * cluster group. Instead, find the first size below
161 		 * 256 which would give us an even distribution.
162 		 *
163 		 * Larger cluster group sizes actually work out pretty
164 		 * well when pared to 256, so we don't have to do this
165 		 * for any group that fits more than two
166 		 * OCFS2_LA_MAX_DEFAULT_MB windows.
167 		 */
168 		if (gd_mb > (2 * OCFS2_LA_MAX_DEFAULT_MB))
169 			la_mb = 256;
170 		else {
171 			unsigned int gd_mult = gd_mb;
172 
173 			while (gd_mult > 256)
174 				gd_mult = gd_mult >> 1;
175 
176 			la_mb = gd_mult;
177 		}
178 	}
179 
180 	megs_per_slot = osb->osb_clusters_at_boot / osb->max_slots;
181 	megs_per_slot = ocfs2_clusters_to_megabytes(osb->sb, megs_per_slot);
182 	/* Too many nodes, too few disk clusters. */
183 	if (megs_per_slot < la_mb)
184 		la_mb = megs_per_slot;
185 
186 	/* We can't store more bits than we can in a block. */
187 	la_max_mb = ocfs2_clusters_to_megabytes(osb->sb,
188 						ocfs2_local_alloc_size(sb) * 8);
189 	if (la_mb > la_max_mb)
190 		la_mb = la_max_mb;
191 
192 	return la_mb;
193 }
194 
ocfs2_la_set_sizes(struct ocfs2_super * osb,int requested_mb)195 void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb)
196 {
197 	struct super_block *sb = osb->sb;
198 	unsigned int la_default_mb = ocfs2_la_default_mb(osb);
199 	unsigned int la_max_mb;
200 
201 	la_max_mb = ocfs2_clusters_to_megabytes(sb,
202 						ocfs2_local_alloc_size(sb) * 8);
203 
204 	trace_ocfs2_la_set_sizes(requested_mb, la_max_mb, la_default_mb);
205 
206 	if (requested_mb == -1) {
207 		/* No user request - use defaults */
208 		osb->local_alloc_default_bits =
209 			ocfs2_megabytes_to_clusters(sb, la_default_mb);
210 	} else if (requested_mb > la_max_mb) {
211 		/* Request is too big, we give the maximum available */
212 		osb->local_alloc_default_bits =
213 			ocfs2_megabytes_to_clusters(sb, la_max_mb);
214 	} else {
215 		osb->local_alloc_default_bits =
216 			ocfs2_megabytes_to_clusters(sb, requested_mb);
217 	}
218 
219 	osb->local_alloc_bits = osb->local_alloc_default_bits;
220 }
221 
ocfs2_la_state_enabled(struct ocfs2_super * osb)222 static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
223 {
224 	return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||
225 		osb->local_alloc_state == OCFS2_LA_ENABLED);
226 }
227 
ocfs2_local_alloc_seen_free_bits(struct ocfs2_super * osb,unsigned int num_clusters)228 void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
229 				      unsigned int num_clusters)
230 {
231 	spin_lock(&osb->osb_lock);
232 	if (osb->local_alloc_state == OCFS2_LA_DISABLED ||
233 	    osb->local_alloc_state == OCFS2_LA_THROTTLED)
234 		if (num_clusters >= osb->local_alloc_default_bits) {
235 			cancel_delayed_work(&osb->la_enable_wq);
236 			osb->local_alloc_state = OCFS2_LA_ENABLED;
237 		}
238 	spin_unlock(&osb->osb_lock);
239 }
240 
ocfs2_la_enable_worker(struct work_struct * work)241 void ocfs2_la_enable_worker(struct work_struct *work)
242 {
243 	struct ocfs2_super *osb =
244 		container_of(work, struct ocfs2_super,
245 			     la_enable_wq.work);
246 	spin_lock(&osb->osb_lock);
247 	osb->local_alloc_state = OCFS2_LA_ENABLED;
248 	spin_unlock(&osb->osb_lock);
249 }
250 
251 /*
252  * Tell us whether a given allocation should use the local alloc
253  * file. Otherwise, it has to go to the main bitmap.
254  *
255  * This function does semi-dirty reads of local alloc size and state!
256  * This is ok however, as the values are re-checked once under mutex.
257  */
ocfs2_alloc_should_use_local(struct ocfs2_super * osb,u64 bits)258 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
259 {
260 	int ret = 0;
261 	int la_bits;
262 
263 	spin_lock(&osb->osb_lock);
264 	la_bits = osb->local_alloc_bits;
265 
266 	if (!ocfs2_la_state_enabled(osb))
267 		goto bail;
268 
269 	/* la_bits should be at least twice the size (in clusters) of
270 	 * a new block group. We want to be sure block group
271 	 * allocations go through the local alloc, so allow an
272 	 * allocation to take up to half the bitmap. */
273 	if (bits > (la_bits / 2))
274 		goto bail;
275 
276 	ret = 1;
277 bail:
278 	trace_ocfs2_alloc_should_use_local(
279 	     (unsigned long long)bits, osb->local_alloc_state, la_bits, ret);
280 	spin_unlock(&osb->osb_lock);
281 	return ret;
282 }
283 
ocfs2_load_local_alloc(struct ocfs2_super * osb)284 int ocfs2_load_local_alloc(struct ocfs2_super *osb)
285 {
286 	int status = 0;
287 	struct ocfs2_dinode *alloc = NULL;
288 	struct buffer_head *alloc_bh = NULL;
289 	u32 num_used;
290 	struct inode *inode = NULL;
291 	struct ocfs2_local_alloc *la;
292 
293 	if (osb->local_alloc_bits == 0)
294 		goto bail;
295 
296 	if (osb->local_alloc_bits >= osb->bitmap_cpg) {
297 		mlog(ML_NOTICE, "Requested local alloc window %d is larger "
298 		     "than max possible %u. Using defaults.\n",
299 		     osb->local_alloc_bits, (osb->bitmap_cpg - 1));
300 		osb->local_alloc_bits =
301 			ocfs2_megabytes_to_clusters(osb->sb,
302 						    ocfs2_la_default_mb(osb));
303 	}
304 
305 	/* read the alloc off disk */
306 	inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
307 					    osb->slot_num);
308 	if (!inode) {
309 		status = -EINVAL;
310 		mlog_errno(status);
311 		goto bail;
312 	}
313 
314 	status = ocfs2_read_inode_block_full(inode, &alloc_bh,
315 					     OCFS2_BH_IGNORE_CACHE);
316 	if (status < 0) {
317 		mlog_errno(status);
318 		goto bail;
319 	}
320 
321 	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
322 	la = OCFS2_LOCAL_ALLOC(alloc);
323 
324 	if (!(le32_to_cpu(alloc->i_flags) &
325 	    (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
326 		mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
327 		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
328 		status = -EINVAL;
329 		goto bail;
330 	}
331 
332 	if ((la->la_size == 0) ||
333 	    (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
334 		mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
335 		     le16_to_cpu(la->la_size));
336 		status = -EINVAL;
337 		goto bail;
338 	}
339 
340 	/* do a little verification. */
341 	num_used = ocfs2_local_alloc_count_bits(alloc);
342 
343 	/* hopefully the local alloc has always been recovered before
344 	 * we load it. */
345 	if (num_used
346 	    || alloc->id1.bitmap1.i_used
347 	    || alloc->id1.bitmap1.i_total
348 	    || la->la_bm_off) {
349 		mlog(ML_ERROR, "inconsistent detected, clean journal with"
350 		     " unrecovered local alloc, please run fsck.ocfs2!\n"
351 		     "found = %u, set = %u, taken = %u, off = %u\n",
352 		     num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
353 		     le32_to_cpu(alloc->id1.bitmap1.i_total),
354 		     OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
355 
356 		status = -EINVAL;
357 		goto bail;
358 	}
359 
360 	osb->local_alloc_bh = alloc_bh;
361 	osb->local_alloc_state = OCFS2_LA_ENABLED;
362 
363 bail:
364 	if (status < 0)
365 		brelse(alloc_bh);
366 	iput(inode);
367 
368 	trace_ocfs2_load_local_alloc(osb->local_alloc_bits);
369 
370 	if (status)
371 		mlog_errno(status);
372 	return status;
373 }
374 
375 /*
376  * return any unused bits to the bitmap and write out a clean
377  * local_alloc.
378  *
379  * local_alloc_bh is optional. If not passed, we will simply use the
380  * one off osb. If you do pass it however, be warned that it *will* be
381  * returned brelse'd and NULL'd out.*/
ocfs2_shutdown_local_alloc(struct ocfs2_super * osb)382 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
383 {
384 	int status;
385 	handle_t *handle;
386 	struct inode *local_alloc_inode = NULL;
387 	struct buffer_head *bh = NULL;
388 	struct buffer_head *main_bm_bh = NULL;
389 	struct inode *main_bm_inode = NULL;
390 	struct ocfs2_dinode *alloc_copy = NULL;
391 	struct ocfs2_dinode *alloc = NULL;
392 
393 	cancel_delayed_work(&osb->la_enable_wq);
394 	if (osb->ocfs2_wq)
395 		flush_workqueue(osb->ocfs2_wq);
396 
397 	if (osb->local_alloc_state == OCFS2_LA_UNUSED)
398 		goto out;
399 
400 	local_alloc_inode =
401 		ocfs2_get_system_file_inode(osb,
402 					    LOCAL_ALLOC_SYSTEM_INODE,
403 					    osb->slot_num);
404 	if (!local_alloc_inode) {
405 		status = -ENOENT;
406 		mlog_errno(status);
407 		goto out;
408 	}
409 
410 	osb->local_alloc_state = OCFS2_LA_DISABLED;
411 
412 	ocfs2_resmap_uninit(&osb->osb_la_resmap);
413 
414 	main_bm_inode = ocfs2_get_system_file_inode(osb,
415 						    GLOBAL_BITMAP_SYSTEM_INODE,
416 						    OCFS2_INVALID_SLOT);
417 	if (!main_bm_inode) {
418 		status = -EINVAL;
419 		mlog_errno(status);
420 		goto out;
421 	}
422 
423 	inode_lock(main_bm_inode);
424 
425 	status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
426 	if (status < 0) {
427 		mlog_errno(status);
428 		goto out_mutex;
429 	}
430 
431 	/* WINDOW_MOVE_CREDITS is a bit heavy... */
432 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
433 	if (IS_ERR(handle)) {
434 		mlog_errno(PTR_ERR(handle));
435 		handle = NULL;
436 		goto out_unlock;
437 	}
438 
439 	bh = osb->local_alloc_bh;
440 	alloc = (struct ocfs2_dinode *) bh->b_data;
441 
442 	alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
443 	if (!alloc_copy) {
444 		status = -ENOMEM;
445 		goto out_commit;
446 	}
447 	memcpy(alloc_copy, alloc, bh->b_size);
448 
449 	status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode),
450 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
451 	if (status < 0) {
452 		mlog_errno(status);
453 		goto out_commit;
454 	}
455 
456 	ocfs2_clear_local_alloc(alloc);
457 	ocfs2_journal_dirty(handle, bh);
458 
459 	brelse(bh);
460 	osb->local_alloc_bh = NULL;
461 	osb->local_alloc_state = OCFS2_LA_UNUSED;
462 
463 	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
464 					  main_bm_inode, main_bm_bh);
465 	if (status < 0)
466 		mlog_errno(status);
467 
468 out_commit:
469 	ocfs2_commit_trans(osb, handle);
470 
471 out_unlock:
472 	brelse(main_bm_bh);
473 
474 	ocfs2_inode_unlock(main_bm_inode, 1);
475 
476 out_mutex:
477 	inode_unlock(main_bm_inode);
478 	iput(main_bm_inode);
479 
480 out:
481 	iput(local_alloc_inode);
482 
483 	kfree(alloc_copy);
484 }
485 
486 /*
487  * We want to free the bitmap bits outside of any recovery context as
488  * we'll need a cluster lock to do so, but we must clear the local
489  * alloc before giving up the recovered nodes journal. To solve this,
490  * we kmalloc a copy of the local alloc before it's change for the
491  * caller to process with ocfs2_complete_local_alloc_recovery
492  */
ocfs2_begin_local_alloc_recovery(struct ocfs2_super * osb,int slot_num,struct ocfs2_dinode ** alloc_copy)493 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
494 				     int slot_num,
495 				     struct ocfs2_dinode **alloc_copy)
496 {
497 	int status = 0;
498 	struct buffer_head *alloc_bh = NULL;
499 	struct inode *inode = NULL;
500 	struct ocfs2_dinode *alloc;
501 
502 	trace_ocfs2_begin_local_alloc_recovery(slot_num);
503 
504 	*alloc_copy = NULL;
505 
506 	inode = ocfs2_get_system_file_inode(osb,
507 					    LOCAL_ALLOC_SYSTEM_INODE,
508 					    slot_num);
509 	if (!inode) {
510 		status = -EINVAL;
511 		mlog_errno(status);
512 		goto bail;
513 	}
514 
515 	inode_lock(inode);
516 
517 	status = ocfs2_read_inode_block_full(inode, &alloc_bh,
518 					     OCFS2_BH_IGNORE_CACHE);
519 	if (status < 0) {
520 		mlog_errno(status);
521 		goto bail;
522 	}
523 
524 	*alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
525 	if (!(*alloc_copy)) {
526 		status = -ENOMEM;
527 		goto bail;
528 	}
529 	memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
530 
531 	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
532 	ocfs2_clear_local_alloc(alloc);
533 
534 	ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
535 	status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode));
536 	if (status < 0)
537 		mlog_errno(status);
538 
539 bail:
540 	if (status < 0) {
541 		kfree(*alloc_copy);
542 		*alloc_copy = NULL;
543 	}
544 
545 	brelse(alloc_bh);
546 
547 	if (inode) {
548 		inode_unlock(inode);
549 		iput(inode);
550 	}
551 
552 	if (status)
553 		mlog_errno(status);
554 	return status;
555 }
556 
557 /*
558  * Step 2: By now, we've completed the journal recovery, we've stamped
559  * a clean local alloc on disk and dropped the node out of the
560  * recovery map. Dlm locks will no longer stall, so lets clear out the
561  * main bitmap.
562  */
ocfs2_complete_local_alloc_recovery(struct ocfs2_super * osb,struct ocfs2_dinode * alloc)563 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
564 					struct ocfs2_dinode *alloc)
565 {
566 	int status;
567 	handle_t *handle;
568 	struct buffer_head *main_bm_bh = NULL;
569 	struct inode *main_bm_inode;
570 
571 	main_bm_inode = ocfs2_get_system_file_inode(osb,
572 						    GLOBAL_BITMAP_SYSTEM_INODE,
573 						    OCFS2_INVALID_SLOT);
574 	if (!main_bm_inode) {
575 		status = -EINVAL;
576 		mlog_errno(status);
577 		goto out;
578 	}
579 
580 	inode_lock(main_bm_inode);
581 
582 	status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
583 	if (status < 0) {
584 		mlog_errno(status);
585 		goto out_mutex;
586 	}
587 
588 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
589 	if (IS_ERR(handle)) {
590 		status = PTR_ERR(handle);
591 		handle = NULL;
592 		mlog_errno(status);
593 		goto out_unlock;
594 	}
595 
596 	/* we want the bitmap change to be recorded on disk asap */
597 	handle->h_sync = 1;
598 
599 	status = ocfs2_sync_local_to_main(osb, handle, alloc,
600 					  main_bm_inode, main_bm_bh);
601 	if (status < 0)
602 		mlog_errno(status);
603 
604 	ocfs2_commit_trans(osb, handle);
605 
606 out_unlock:
607 	ocfs2_inode_unlock(main_bm_inode, 1);
608 
609 out_mutex:
610 	inode_unlock(main_bm_inode);
611 
612 	brelse(main_bm_bh);
613 
614 	iput(main_bm_inode);
615 
616 out:
617 	if (!status)
618 		ocfs2_init_steal_slots(osb);
619 	if (status)
620 		mlog_errno(status);
621 	return status;
622 }
623 
624 /*
625  * make sure we've got at least bits_wanted contiguous bits in the
626  * local alloc. You lose them when you drop i_mutex.
627  *
628  * We will add ourselves to the transaction passed in, but may start
629  * our own in order to shift windows.
630  */
ocfs2_reserve_local_alloc_bits(struct ocfs2_super * osb,u32 bits_wanted,struct ocfs2_alloc_context * ac)631 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
632 				   u32 bits_wanted,
633 				   struct ocfs2_alloc_context *ac)
634 {
635 	int status;
636 	struct ocfs2_dinode *alloc;
637 	struct inode *local_alloc_inode;
638 	unsigned int free_bits;
639 
640 	BUG_ON(!ac);
641 
642 	local_alloc_inode =
643 		ocfs2_get_system_file_inode(osb,
644 					    LOCAL_ALLOC_SYSTEM_INODE,
645 					    osb->slot_num);
646 	if (!local_alloc_inode) {
647 		status = -ENOENT;
648 		mlog_errno(status);
649 		goto bail;
650 	}
651 
652 	inode_lock(local_alloc_inode);
653 
654 	/*
655 	 * We must double check state and allocator bits because
656 	 * another process may have changed them while holding i_mutex.
657 	 */
658 	spin_lock(&osb->osb_lock);
659 	if (!ocfs2_la_state_enabled(osb) ||
660 	    (bits_wanted > osb->local_alloc_bits)) {
661 		spin_unlock(&osb->osb_lock);
662 		status = -ENOSPC;
663 		goto bail;
664 	}
665 	spin_unlock(&osb->osb_lock);
666 
667 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
668 
669 #ifdef CONFIG_OCFS2_DEBUG_FS
670 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
671 	    ocfs2_local_alloc_count_bits(alloc)) {
672 		status = ocfs2_error(osb->sb, "local alloc inode %llu says it has %u used bits, but a count shows %u\n",
673 				(unsigned long long)le64_to_cpu(alloc->i_blkno),
674 				le32_to_cpu(alloc->id1.bitmap1.i_used),
675 				ocfs2_local_alloc_count_bits(alloc));
676 		goto bail;
677 	}
678 #endif
679 
680 	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
681 		le32_to_cpu(alloc->id1.bitmap1.i_used);
682 	if (bits_wanted > free_bits) {
683 		/* uhoh, window change time. */
684 		status =
685 			ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
686 		if (status < 0) {
687 			if (status != -ENOSPC)
688 				mlog_errno(status);
689 			goto bail;
690 		}
691 
692 		/*
693 		 * Under certain conditions, the window slide code
694 		 * might have reduced the number of bits available or
695 		 * disabled the the local alloc entirely. Re-check
696 		 * here and return -ENOSPC if necessary.
697 		 */
698 		status = -ENOSPC;
699 		if (!ocfs2_la_state_enabled(osb))
700 			goto bail;
701 
702 		free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
703 			le32_to_cpu(alloc->id1.bitmap1.i_used);
704 		if (bits_wanted > free_bits)
705 			goto bail;
706 	}
707 
708 	ac->ac_inode = local_alloc_inode;
709 	/* We should never use localalloc from another slot */
710 	ac->ac_alloc_slot = osb->slot_num;
711 	ac->ac_which = OCFS2_AC_USE_LOCAL;
712 	get_bh(osb->local_alloc_bh);
713 	ac->ac_bh = osb->local_alloc_bh;
714 	status = 0;
715 bail:
716 	if (status < 0 && local_alloc_inode) {
717 		inode_unlock(local_alloc_inode);
718 		iput(local_alloc_inode);
719 	}
720 
721 	trace_ocfs2_reserve_local_alloc_bits(
722 		(unsigned long long)ac->ac_max_block,
723 		bits_wanted, osb->slot_num, status);
724 
725 	if (status)
726 		mlog_errno(status);
727 	return status;
728 }
729 
ocfs2_claim_local_alloc_bits(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac,u32 bits_wanted,u32 * bit_off,u32 * num_bits)730 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
731 				 handle_t *handle,
732 				 struct ocfs2_alloc_context *ac,
733 				 u32 bits_wanted,
734 				 u32 *bit_off,
735 				 u32 *num_bits)
736 {
737 	int status, start;
738 	struct inode *local_alloc_inode;
739 	void *bitmap;
740 	struct ocfs2_dinode *alloc;
741 	struct ocfs2_local_alloc *la;
742 
743 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
744 
745 	local_alloc_inode = ac->ac_inode;
746 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
747 	la = OCFS2_LOCAL_ALLOC(alloc);
748 
749 	start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted,
750 						  ac->ac_resv);
751 	if (start == -1) {
752 		/* TODO: Shouldn't we just BUG here? */
753 		status = -ENOSPC;
754 		mlog_errno(status);
755 		goto bail;
756 	}
757 
758 	bitmap = la->la_bitmap;
759 	*bit_off = le32_to_cpu(la->la_bm_off) + start;
760 	*num_bits = bits_wanted;
761 
762 	status = ocfs2_journal_access_di(handle,
763 					 INODE_CACHE(local_alloc_inode),
764 					 osb->local_alloc_bh,
765 					 OCFS2_JOURNAL_ACCESS_WRITE);
766 	if (status < 0) {
767 		mlog_errno(status);
768 		goto bail;
769 	}
770 
771 	ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start,
772 				  bits_wanted);
773 
774 	while(bits_wanted--)
775 		ocfs2_set_bit(start++, bitmap);
776 
777 	le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
778 	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
779 
780 bail:
781 	if (status)
782 		mlog_errno(status);
783 	return status;
784 }
785 
ocfs2_free_local_alloc_bits(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac,u32 bit_off,u32 num_bits)786 int ocfs2_free_local_alloc_bits(struct ocfs2_super *osb,
787 				handle_t *handle,
788 				struct ocfs2_alloc_context *ac,
789 				u32 bit_off,
790 				u32 num_bits)
791 {
792 	int status, start;
793 	u32 clear_bits;
794 	struct inode *local_alloc_inode;
795 	void *bitmap;
796 	struct ocfs2_dinode *alloc;
797 	struct ocfs2_local_alloc *la;
798 
799 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
800 
801 	local_alloc_inode = ac->ac_inode;
802 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
803 	la = OCFS2_LOCAL_ALLOC(alloc);
804 
805 	bitmap = la->la_bitmap;
806 	start = bit_off - le32_to_cpu(la->la_bm_off);
807 	clear_bits = num_bits;
808 
809 	status = ocfs2_journal_access_di(handle,
810 			INODE_CACHE(local_alloc_inode),
811 			osb->local_alloc_bh,
812 			OCFS2_JOURNAL_ACCESS_WRITE);
813 	if (status < 0) {
814 		mlog_errno(status);
815 		goto bail;
816 	}
817 
818 	while (clear_bits--)
819 		ocfs2_clear_bit(start++, bitmap);
820 
821 	le32_add_cpu(&alloc->id1.bitmap1.i_used, -num_bits);
822 	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
823 
824 bail:
825 	return status;
826 }
827 
ocfs2_local_alloc_count_bits(struct ocfs2_dinode * alloc)828 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
829 {
830 	u32 count;
831 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
832 
833 	count = memweight(la->la_bitmap, le16_to_cpu(la->la_size));
834 
835 	trace_ocfs2_local_alloc_count_bits(count);
836 	return count;
837 }
838 
ocfs2_local_alloc_find_clear_bits(struct ocfs2_super * osb,struct ocfs2_dinode * alloc,u32 * numbits,struct ocfs2_alloc_reservation * resv)839 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
840 				     struct ocfs2_dinode *alloc,
841 				     u32 *numbits,
842 				     struct ocfs2_alloc_reservation *resv)
843 {
844 	int numfound = 0, bitoff, left, startoff, lastzero;
845 	int local_resv = 0;
846 	struct ocfs2_alloc_reservation r;
847 	void *bitmap = NULL;
848 	struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap;
849 
850 	if (!alloc->id1.bitmap1.i_total) {
851 		bitoff = -1;
852 		goto bail;
853 	}
854 
855 	if (!resv) {
856 		local_resv = 1;
857 		ocfs2_resv_init_once(&r);
858 		ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP);
859 		resv = &r;
860 	}
861 
862 	numfound = *numbits;
863 	if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) {
864 		if (numfound < *numbits)
865 			*numbits = numfound;
866 		goto bail;
867 	}
868 
869 	/*
870 	 * Code error. While reservations are enabled, local
871 	 * allocation should _always_ go through them.
872 	 */
873 	BUG_ON(osb->osb_resv_level != 0);
874 
875 	/*
876 	 * Reservations are disabled. Handle this the old way.
877 	 */
878 
879 	bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
880 
881 	numfound = bitoff = startoff = 0;
882 	lastzero = -1;
883 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
884 	while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
885 		if (bitoff == left) {
886 			/* mlog(0, "bitoff (%d) == left", bitoff); */
887 			break;
888 		}
889 		/* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
890 		   "numfound = %d\n", bitoff, startoff, numfound);*/
891 
892 		/* Ok, we found a zero bit... is it contig. or do we
893 		 * start over?*/
894 		if (bitoff == startoff) {
895 			/* we found a zero */
896 			numfound++;
897 			startoff++;
898 		} else {
899 			/* got a zero after some ones */
900 			numfound = 1;
901 			startoff = bitoff+1;
902 		}
903 		/* we got everything we needed */
904 		if (numfound == *numbits) {
905 			/* mlog(0, "Found it all!\n"); */
906 			break;
907 		}
908 	}
909 
910 	trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound);
911 
912 	if (numfound == *numbits)
913 		bitoff = startoff - numfound;
914 	else
915 		bitoff = -1;
916 
917 bail:
918 	if (local_resv)
919 		ocfs2_resv_discard(resmap, resv);
920 
921 	trace_ocfs2_local_alloc_find_clear_bits(*numbits,
922 		le32_to_cpu(alloc->id1.bitmap1.i_total),
923 		bitoff, numfound);
924 
925 	return bitoff;
926 }
927 
ocfs2_clear_local_alloc(struct ocfs2_dinode * alloc)928 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
929 {
930 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
931 	int i;
932 
933 	alloc->id1.bitmap1.i_total = 0;
934 	alloc->id1.bitmap1.i_used = 0;
935 	la->la_bm_off = 0;
936 	for(i = 0; i < le16_to_cpu(la->la_size); i++)
937 		la->la_bitmap[i] = 0;
938 }
939 
940 #if 0
941 /* turn this on and uncomment below to aid debugging window shifts. */
942 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
943 				   unsigned int start,
944 				   unsigned int count)
945 {
946 	unsigned int tmp = count;
947 	while(tmp--) {
948 		if (ocfs2_test_bit(start + tmp, bitmap)) {
949 			printk("ocfs2_verify_zero_bits: start = %u, count = "
950 			       "%u\n", start, count);
951 			printk("ocfs2_verify_zero_bits: bit %u is set!",
952 			       start + tmp);
953 			BUG();
954 		}
955 	}
956 }
957 #endif
958 
959 /*
960  * sync the local alloc to main bitmap.
961  *
962  * assumes you've already locked the main bitmap -- the bitmap inode
963  * passed is used for caching.
964  */
ocfs2_sync_local_to_main(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_dinode * alloc,struct inode * main_bm_inode,struct buffer_head * main_bm_bh)965 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
966 				    handle_t *handle,
967 				    struct ocfs2_dinode *alloc,
968 				    struct inode *main_bm_inode,
969 				    struct buffer_head *main_bm_bh)
970 {
971 	int status = 0;
972 	int bit_off, left, count, start;
973 	u64 la_start_blk;
974 	u64 blkno;
975 	void *bitmap;
976 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
977 
978 	trace_ocfs2_sync_local_to_main(
979 	     le32_to_cpu(alloc->id1.bitmap1.i_total),
980 	     le32_to_cpu(alloc->id1.bitmap1.i_used));
981 
982 	if (!alloc->id1.bitmap1.i_total) {
983 		goto bail;
984 	}
985 
986 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
987 	    le32_to_cpu(alloc->id1.bitmap1.i_total)) {
988 		goto bail;
989 	}
990 
991 	la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
992 						le32_to_cpu(la->la_bm_off));
993 	bitmap = la->la_bitmap;
994 	start = count = bit_off = 0;
995 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
996 
997 	while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
998 	       != -1) {
999 		if ((bit_off < left) && (bit_off == start)) {
1000 			count++;
1001 			start++;
1002 			continue;
1003 		}
1004 		if (count) {
1005 			blkno = la_start_blk +
1006 				ocfs2_clusters_to_blocks(osb->sb,
1007 							 start - count);
1008 
1009 			trace_ocfs2_sync_local_to_main_free(
1010 			     count, start - count,
1011 			     (unsigned long long)la_start_blk,
1012 			     (unsigned long long)blkno);
1013 
1014 			status = ocfs2_release_clusters(handle,
1015 							main_bm_inode,
1016 							main_bm_bh, blkno,
1017 							count);
1018 			if (status < 0) {
1019 				mlog_errno(status);
1020 				goto bail;
1021 			}
1022 		}
1023 		if (bit_off >= left)
1024 			break;
1025 		count = 1;
1026 		start = bit_off + 1;
1027 	}
1028 
1029 bail:
1030 	if (status)
1031 		mlog_errno(status);
1032 	return status;
1033 }
1034 
1035 enum ocfs2_la_event {
1036 	OCFS2_LA_EVENT_SLIDE,		/* Normal window slide. */
1037 	OCFS2_LA_EVENT_FRAGMENTED,	/* The global bitmap has
1038 					 * enough bits theoretically
1039 					 * free, but a contiguous
1040 					 * allocation could not be
1041 					 * found. */
1042 	OCFS2_LA_EVENT_ENOSPC,		/* Global bitmap doesn't have
1043 					 * enough bits free to satisfy
1044 					 * our request. */
1045 };
1046 #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
1047 /*
1048  * Given an event, calculate the size of our next local alloc window.
1049  *
1050  * This should always be called under i_mutex of the local alloc inode
1051  * so that local alloc disabling doesn't race with processes trying to
1052  * use the allocator.
1053  *
1054  * Returns the state which the local alloc was left in. This value can
1055  * be ignored by some paths.
1056  */
ocfs2_recalc_la_window(struct ocfs2_super * osb,enum ocfs2_la_event event)1057 static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
1058 				  enum ocfs2_la_event event)
1059 {
1060 	unsigned int bits;
1061 	int state;
1062 
1063 	spin_lock(&osb->osb_lock);
1064 	if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
1065 		WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
1066 		goto out_unlock;
1067 	}
1068 
1069 	/*
1070 	 * ENOSPC and fragmentation are treated similarly for now.
1071 	 */
1072 	if (event == OCFS2_LA_EVENT_ENOSPC ||
1073 	    event == OCFS2_LA_EVENT_FRAGMENTED) {
1074 		/*
1075 		 * We ran out of contiguous space in the primary
1076 		 * bitmap. Drastically reduce the number of bits used
1077 		 * by local alloc until we have to disable it.
1078 		 */
1079 		bits = osb->local_alloc_bits >> 1;
1080 		if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
1081 			/*
1082 			 * By setting state to THROTTLED, we'll keep
1083 			 * the number of local alloc bits used down
1084 			 * until an event occurs which would give us
1085 			 * reason to assume the bitmap situation might
1086 			 * have changed.
1087 			 */
1088 			osb->local_alloc_state = OCFS2_LA_THROTTLED;
1089 			osb->local_alloc_bits = bits;
1090 		} else {
1091 			osb->local_alloc_state = OCFS2_LA_DISABLED;
1092 		}
1093 		queue_delayed_work(osb->ocfs2_wq, &osb->la_enable_wq,
1094 				   OCFS2_LA_ENABLE_INTERVAL);
1095 		goto out_unlock;
1096 	}
1097 
1098 	/*
1099 	 * Don't increase the size of the local alloc window until we
1100 	 * know we might be able to fulfill the request. Otherwise, we
1101 	 * risk bouncing around the global bitmap during periods of
1102 	 * low space.
1103 	 */
1104 	if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
1105 		osb->local_alloc_bits = osb->local_alloc_default_bits;
1106 
1107 out_unlock:
1108 	state = osb->local_alloc_state;
1109 	spin_unlock(&osb->osb_lock);
1110 
1111 	return state;
1112 }
1113 
ocfs2_local_alloc_reserve_for_window(struct ocfs2_super * osb,struct ocfs2_alloc_context ** ac,struct inode ** bitmap_inode,struct buffer_head ** bitmap_bh)1114 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
1115 						struct ocfs2_alloc_context **ac,
1116 						struct inode **bitmap_inode,
1117 						struct buffer_head **bitmap_bh)
1118 {
1119 	int status;
1120 
1121 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1122 	if (!(*ac)) {
1123 		status = -ENOMEM;
1124 		mlog_errno(status);
1125 		goto bail;
1126 	}
1127 
1128 retry_enospc:
1129 	(*ac)->ac_bits_wanted = osb->local_alloc_bits;
1130 	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1131 	if (status == -ENOSPC) {
1132 		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
1133 		    OCFS2_LA_DISABLED)
1134 			goto bail;
1135 
1136 		ocfs2_free_ac_resource(*ac);
1137 		memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
1138 		goto retry_enospc;
1139 	}
1140 	if (status < 0) {
1141 		mlog_errno(status);
1142 		goto bail;
1143 	}
1144 
1145 	*bitmap_inode = (*ac)->ac_inode;
1146 	igrab(*bitmap_inode);
1147 	*bitmap_bh = (*ac)->ac_bh;
1148 	get_bh(*bitmap_bh);
1149 	status = 0;
1150 bail:
1151 	if ((status < 0) && *ac) {
1152 		ocfs2_free_alloc_context(*ac);
1153 		*ac = NULL;
1154 	}
1155 
1156 	if (status)
1157 		mlog_errno(status);
1158 	return status;
1159 }
1160 
1161 /*
1162  * pass it the bitmap lock in lock_bh if you have it.
1163  */
ocfs2_local_alloc_new_window(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac)1164 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
1165 					handle_t *handle,
1166 					struct ocfs2_alloc_context *ac)
1167 {
1168 	int status = 0;
1169 	u32 cluster_off, cluster_count;
1170 	struct ocfs2_dinode *alloc = NULL;
1171 	struct ocfs2_local_alloc *la;
1172 
1173 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1174 	la = OCFS2_LOCAL_ALLOC(alloc);
1175 
1176 	trace_ocfs2_local_alloc_new_window(
1177 		le32_to_cpu(alloc->id1.bitmap1.i_total),
1178 		osb->local_alloc_bits);
1179 
1180 	/* Instruct the allocation code to try the most recently used
1181 	 * cluster group. We'll re-record the group used this pass
1182 	 * below. */
1183 	ac->ac_last_group = osb->la_last_gd;
1184 
1185 	/* we used the generic suballoc reserve function, but we set
1186 	 * everything up nicely, so there's no reason why we can't use
1187 	 * the more specific cluster api to claim bits. */
1188 	status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits,
1189 				      &cluster_off, &cluster_count);
1190 	if (status == -ENOSPC) {
1191 retry_enospc:
1192 		/*
1193 		 * Note: We could also try syncing the journal here to
1194 		 * allow use of any free bits which the current
1195 		 * transaction can't give us access to. --Mark
1196 		 */
1197 		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
1198 		    OCFS2_LA_DISABLED)
1199 			goto bail;
1200 
1201 		ac->ac_bits_wanted = osb->local_alloc_bits;
1202 		status = ocfs2_claim_clusters(handle, ac,
1203 					      osb->local_alloc_bits,
1204 					      &cluster_off,
1205 					      &cluster_count);
1206 		if (status == -ENOSPC)
1207 			goto retry_enospc;
1208 		/*
1209 		 * We only shrunk the *minimum* number of in our
1210 		 * request - it's entirely possible that the allocator
1211 		 * might give us more than we asked for.
1212 		 */
1213 		if (status == 0) {
1214 			spin_lock(&osb->osb_lock);
1215 			osb->local_alloc_bits = cluster_count;
1216 			spin_unlock(&osb->osb_lock);
1217 		}
1218 	}
1219 	if (status < 0) {
1220 		if (status != -ENOSPC)
1221 			mlog_errno(status);
1222 		goto bail;
1223 	}
1224 
1225 	osb->la_last_gd = ac->ac_last_group;
1226 
1227 	la->la_bm_off = cpu_to_le32(cluster_off);
1228 	alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
1229 	/* just in case... In the future when we find space ourselves,
1230 	 * we don't have to get all contiguous -- but we'll have to
1231 	 * set all previously used bits in bitmap and update
1232 	 * la_bits_set before setting the bits in the main bitmap. */
1233 	alloc->id1.bitmap1.i_used = 0;
1234 	memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
1235 	       le16_to_cpu(la->la_size));
1236 
1237 	ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count,
1238 			     OCFS2_LOCAL_ALLOC(alloc)->la_bitmap);
1239 
1240 	trace_ocfs2_local_alloc_new_window_result(
1241 		OCFS2_LOCAL_ALLOC(alloc)->la_bm_off,
1242 		le32_to_cpu(alloc->id1.bitmap1.i_total));
1243 
1244 bail:
1245 	if (status)
1246 		mlog_errno(status);
1247 	return status;
1248 }
1249 
1250 /* Note that we do *NOT* lock the local alloc inode here as
1251  * it's been locked already for us. */
ocfs2_local_alloc_slide_window(struct ocfs2_super * osb,struct inode * local_alloc_inode)1252 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
1253 					  struct inode *local_alloc_inode)
1254 {
1255 	int status = 0;
1256 	struct buffer_head *main_bm_bh = NULL;
1257 	struct inode *main_bm_inode = NULL;
1258 	handle_t *handle = NULL;
1259 	struct ocfs2_dinode *alloc;
1260 	struct ocfs2_dinode *alloc_copy = NULL;
1261 	struct ocfs2_alloc_context *ac = NULL;
1262 
1263 	ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
1264 
1265 	/* This will lock the main bitmap for us. */
1266 	status = ocfs2_local_alloc_reserve_for_window(osb,
1267 						      &ac,
1268 						      &main_bm_inode,
1269 						      &main_bm_bh);
1270 	if (status < 0) {
1271 		if (status != -ENOSPC)
1272 			mlog_errno(status);
1273 		goto bail;
1274 	}
1275 
1276 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
1277 	if (IS_ERR(handle)) {
1278 		status = PTR_ERR(handle);
1279 		handle = NULL;
1280 		mlog_errno(status);
1281 		goto bail;
1282 	}
1283 
1284 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1285 
1286 	/* We want to clear the local alloc before doing anything
1287 	 * else, so that if we error later during this operation,
1288 	 * local alloc shutdown won't try to double free main bitmap
1289 	 * bits. Make a copy so the sync function knows which bits to
1290 	 * free. */
1291 	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
1292 	if (!alloc_copy) {
1293 		status = -ENOMEM;
1294 		mlog_errno(status);
1295 		goto bail;
1296 	}
1297 	memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
1298 
1299 	status = ocfs2_journal_access_di(handle,
1300 					 INODE_CACHE(local_alloc_inode),
1301 					 osb->local_alloc_bh,
1302 					 OCFS2_JOURNAL_ACCESS_WRITE);
1303 	if (status < 0) {
1304 		mlog_errno(status);
1305 		goto bail;
1306 	}
1307 
1308 	ocfs2_clear_local_alloc(alloc);
1309 	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
1310 
1311 	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
1312 					  main_bm_inode, main_bm_bh);
1313 	if (status < 0) {
1314 		mlog_errno(status);
1315 		goto bail;
1316 	}
1317 
1318 	status = ocfs2_local_alloc_new_window(osb, handle, ac);
1319 	if (status < 0) {
1320 		if (status != -ENOSPC)
1321 			mlog_errno(status);
1322 		goto bail;
1323 	}
1324 
1325 	atomic_inc(&osb->alloc_stats.moves);
1326 
1327 bail:
1328 	if (handle)
1329 		ocfs2_commit_trans(osb, handle);
1330 
1331 	brelse(main_bm_bh);
1332 
1333 	iput(main_bm_inode);
1334 	kfree(alloc_copy);
1335 
1336 	if (ac)
1337 		ocfs2_free_alloc_context(ac);
1338 
1339 	if (status)
1340 		mlog_errno(status);
1341 	return status;
1342 }
1343 
1344