• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 	if (inode)
367 		iput(inode);
368 
369 	trace_ocfs2_load_local_alloc(osb->local_alloc_bits);
370 
371 	if (status)
372 		mlog_errno(status);
373 	return status;
374 }
375 
376 /*
377  * return any unused bits to the bitmap and write out a clean
378  * local_alloc.
379  *
380  * local_alloc_bh is optional. If not passed, we will simply use the
381  * one off osb. If you do pass it however, be warned that it *will* be
382  * returned brelse'd and NULL'd out.*/
ocfs2_shutdown_local_alloc(struct ocfs2_super * osb)383 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
384 {
385 	int status;
386 	handle_t *handle;
387 	struct inode *local_alloc_inode = NULL;
388 	struct buffer_head *bh = NULL;
389 	struct buffer_head *main_bm_bh = NULL;
390 	struct inode *main_bm_inode = NULL;
391 	struct ocfs2_dinode *alloc_copy = NULL;
392 	struct ocfs2_dinode *alloc = NULL;
393 
394 	cancel_delayed_work(&osb->la_enable_wq);
395 	flush_workqueue(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 	mutex_lock(&main_bm_inode->i_mutex);
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 	mutex_unlock(&main_bm_inode->i_mutex);
478 	iput(main_bm_inode);
479 
480 out:
481 	if (local_alloc_inode)
482 		iput(local_alloc_inode);
483 
484 	kfree(alloc_copy);
485 }
486 
487 /*
488  * We want to free the bitmap bits outside of any recovery context as
489  * we'll need a cluster lock to do so, but we must clear the local
490  * alloc before giving up the recovered nodes journal. To solve this,
491  * we kmalloc a copy of the local alloc before it's change for the
492  * caller to process with ocfs2_complete_local_alloc_recovery
493  */
ocfs2_begin_local_alloc_recovery(struct ocfs2_super * osb,int slot_num,struct ocfs2_dinode ** alloc_copy)494 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
495 				     int slot_num,
496 				     struct ocfs2_dinode **alloc_copy)
497 {
498 	int status = 0;
499 	struct buffer_head *alloc_bh = NULL;
500 	struct inode *inode = NULL;
501 	struct ocfs2_dinode *alloc;
502 
503 	trace_ocfs2_begin_local_alloc_recovery(slot_num);
504 
505 	*alloc_copy = NULL;
506 
507 	inode = ocfs2_get_system_file_inode(osb,
508 					    LOCAL_ALLOC_SYSTEM_INODE,
509 					    slot_num);
510 	if (!inode) {
511 		status = -EINVAL;
512 		mlog_errno(status);
513 		goto bail;
514 	}
515 
516 	mutex_lock(&inode->i_mutex);
517 
518 	status = ocfs2_read_inode_block_full(inode, &alloc_bh,
519 					     OCFS2_BH_IGNORE_CACHE);
520 	if (status < 0) {
521 		mlog_errno(status);
522 		goto bail;
523 	}
524 
525 	*alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
526 	if (!(*alloc_copy)) {
527 		status = -ENOMEM;
528 		goto bail;
529 	}
530 	memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
531 
532 	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
533 	ocfs2_clear_local_alloc(alloc);
534 
535 	ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
536 	status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode));
537 	if (status < 0)
538 		mlog_errno(status);
539 
540 bail:
541 	if (status < 0) {
542 		kfree(*alloc_copy);
543 		*alloc_copy = NULL;
544 	}
545 
546 	brelse(alloc_bh);
547 
548 	if (inode) {
549 		mutex_unlock(&inode->i_mutex);
550 		iput(inode);
551 	}
552 
553 	if (status)
554 		mlog_errno(status);
555 	return status;
556 }
557 
558 /*
559  * Step 2: By now, we've completed the journal recovery, we've stamped
560  * a clean local alloc on disk and dropped the node out of the
561  * recovery map. Dlm locks will no longer stall, so lets clear out the
562  * main bitmap.
563  */
ocfs2_complete_local_alloc_recovery(struct ocfs2_super * osb,struct ocfs2_dinode * alloc)564 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
565 					struct ocfs2_dinode *alloc)
566 {
567 	int status;
568 	handle_t *handle;
569 	struct buffer_head *main_bm_bh = NULL;
570 	struct inode *main_bm_inode;
571 
572 	main_bm_inode = ocfs2_get_system_file_inode(osb,
573 						    GLOBAL_BITMAP_SYSTEM_INODE,
574 						    OCFS2_INVALID_SLOT);
575 	if (!main_bm_inode) {
576 		status = -EINVAL;
577 		mlog_errno(status);
578 		goto out;
579 	}
580 
581 	mutex_lock(&main_bm_inode->i_mutex);
582 
583 	status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
584 	if (status < 0) {
585 		mlog_errno(status);
586 		goto out_mutex;
587 	}
588 
589 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
590 	if (IS_ERR(handle)) {
591 		status = PTR_ERR(handle);
592 		handle = NULL;
593 		mlog_errno(status);
594 		goto out_unlock;
595 	}
596 
597 	/* we want the bitmap change to be recorded on disk asap */
598 	handle->h_sync = 1;
599 
600 	status = ocfs2_sync_local_to_main(osb, handle, alloc,
601 					  main_bm_inode, main_bm_bh);
602 	if (status < 0)
603 		mlog_errno(status);
604 
605 	ocfs2_commit_trans(osb, handle);
606 
607 out_unlock:
608 	ocfs2_inode_unlock(main_bm_inode, 1);
609 
610 out_mutex:
611 	mutex_unlock(&main_bm_inode->i_mutex);
612 
613 	brelse(main_bm_bh);
614 
615 	iput(main_bm_inode);
616 
617 out:
618 	if (!status)
619 		ocfs2_init_steal_slots(osb);
620 	if (status)
621 		mlog_errno(status);
622 	return status;
623 }
624 
625 /*
626  * make sure we've got at least bits_wanted contiguous bits in the
627  * local alloc. You lose them when you drop i_mutex.
628  *
629  * We will add ourselves to the transaction passed in, but may start
630  * our own in order to shift windows.
631  */
ocfs2_reserve_local_alloc_bits(struct ocfs2_super * osb,u32 bits_wanted,struct ocfs2_alloc_context * ac)632 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
633 				   u32 bits_wanted,
634 				   struct ocfs2_alloc_context *ac)
635 {
636 	int status;
637 	struct ocfs2_dinode *alloc;
638 	struct inode *local_alloc_inode;
639 	unsigned int free_bits;
640 
641 	BUG_ON(!ac);
642 
643 	local_alloc_inode =
644 		ocfs2_get_system_file_inode(osb,
645 					    LOCAL_ALLOC_SYSTEM_INODE,
646 					    osb->slot_num);
647 	if (!local_alloc_inode) {
648 		status = -ENOENT;
649 		mlog_errno(status);
650 		goto bail;
651 	}
652 
653 	mutex_lock(&local_alloc_inode->i_mutex);
654 
655 	/*
656 	 * We must double check state and allocator bits because
657 	 * another process may have changed them while holding i_mutex.
658 	 */
659 	spin_lock(&osb->osb_lock);
660 	if (!ocfs2_la_state_enabled(osb) ||
661 	    (bits_wanted > osb->local_alloc_bits)) {
662 		spin_unlock(&osb->osb_lock);
663 		status = -ENOSPC;
664 		goto bail;
665 	}
666 	spin_unlock(&osb->osb_lock);
667 
668 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
669 
670 #ifdef CONFIG_OCFS2_DEBUG_FS
671 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
672 	    ocfs2_local_alloc_count_bits(alloc)) {
673 		ocfs2_error(osb->sb, "local alloc inode %llu says it has %u used bits, but a count shows %u\n",
674 			    (unsigned long long)le64_to_cpu(alloc->i_blkno),
675 			    le32_to_cpu(alloc->id1.bitmap1.i_used),
676 			    ocfs2_local_alloc_count_bits(alloc));
677 		status = -EIO;
678 		goto bail;
679 	}
680 #endif
681 
682 	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
683 		le32_to_cpu(alloc->id1.bitmap1.i_used);
684 	if (bits_wanted > free_bits) {
685 		/* uhoh, window change time. */
686 		status =
687 			ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
688 		if (status < 0) {
689 			if (status != -ENOSPC)
690 				mlog_errno(status);
691 			goto bail;
692 		}
693 
694 		/*
695 		 * Under certain conditions, the window slide code
696 		 * might have reduced the number of bits available or
697 		 * disabled the the local alloc entirely. Re-check
698 		 * here and return -ENOSPC if necessary.
699 		 */
700 		status = -ENOSPC;
701 		if (!ocfs2_la_state_enabled(osb))
702 			goto bail;
703 
704 		free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
705 			le32_to_cpu(alloc->id1.bitmap1.i_used);
706 		if (bits_wanted > free_bits)
707 			goto bail;
708 	}
709 
710 	ac->ac_inode = local_alloc_inode;
711 	/* We should never use localalloc from another slot */
712 	ac->ac_alloc_slot = osb->slot_num;
713 	ac->ac_which = OCFS2_AC_USE_LOCAL;
714 	get_bh(osb->local_alloc_bh);
715 	ac->ac_bh = osb->local_alloc_bh;
716 	status = 0;
717 bail:
718 	if (status < 0 && local_alloc_inode) {
719 		mutex_unlock(&local_alloc_inode->i_mutex);
720 		iput(local_alloc_inode);
721 	}
722 
723 	trace_ocfs2_reserve_local_alloc_bits(
724 		(unsigned long long)ac->ac_max_block,
725 		bits_wanted, osb->slot_num, status);
726 
727 	if (status)
728 		mlog_errno(status);
729 	return status;
730 }
731 
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)732 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
733 				 handle_t *handle,
734 				 struct ocfs2_alloc_context *ac,
735 				 u32 bits_wanted,
736 				 u32 *bit_off,
737 				 u32 *num_bits)
738 {
739 	int status, start;
740 	struct inode *local_alloc_inode;
741 	void *bitmap;
742 	struct ocfs2_dinode *alloc;
743 	struct ocfs2_local_alloc *la;
744 
745 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
746 
747 	local_alloc_inode = ac->ac_inode;
748 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
749 	la = OCFS2_LOCAL_ALLOC(alloc);
750 
751 	start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted,
752 						  ac->ac_resv);
753 	if (start == -1) {
754 		/* TODO: Shouldn't we just BUG here? */
755 		status = -ENOSPC;
756 		mlog_errno(status);
757 		goto bail;
758 	}
759 
760 	bitmap = la->la_bitmap;
761 	*bit_off = le32_to_cpu(la->la_bm_off) + start;
762 	*num_bits = bits_wanted;
763 
764 	status = ocfs2_journal_access_di(handle,
765 					 INODE_CACHE(local_alloc_inode),
766 					 osb->local_alloc_bh,
767 					 OCFS2_JOURNAL_ACCESS_WRITE);
768 	if (status < 0) {
769 		mlog_errno(status);
770 		goto bail;
771 	}
772 
773 	ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start,
774 				  bits_wanted);
775 
776 	while(bits_wanted--)
777 		ocfs2_set_bit(start++, bitmap);
778 
779 	le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
780 	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
781 
782 bail:
783 	if (status)
784 		mlog_errno(status);
785 	return status;
786 }
787 
ocfs2_free_local_alloc_bits(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac,u32 bit_off,u32 num_bits)788 int ocfs2_free_local_alloc_bits(struct ocfs2_super *osb,
789 				handle_t *handle,
790 				struct ocfs2_alloc_context *ac,
791 				u32 bit_off,
792 				u32 num_bits)
793 {
794 	int status, start;
795 	u32 clear_bits;
796 	struct inode *local_alloc_inode;
797 	void *bitmap;
798 	struct ocfs2_dinode *alloc;
799 	struct ocfs2_local_alloc *la;
800 
801 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
802 
803 	local_alloc_inode = ac->ac_inode;
804 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
805 	la = OCFS2_LOCAL_ALLOC(alloc);
806 
807 	bitmap = la->la_bitmap;
808 	start = bit_off - le32_to_cpu(la->la_bm_off);
809 	clear_bits = num_bits;
810 
811 	status = ocfs2_journal_access_di(handle,
812 			INODE_CACHE(local_alloc_inode),
813 			osb->local_alloc_bh,
814 			OCFS2_JOURNAL_ACCESS_WRITE);
815 	if (status < 0) {
816 		mlog_errno(status);
817 		goto bail;
818 	}
819 
820 	while (clear_bits--)
821 		ocfs2_clear_bit(start++, bitmap);
822 
823 	le32_add_cpu(&alloc->id1.bitmap1.i_used, -num_bits);
824 	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
825 
826 bail:
827 	return status;
828 }
829 
ocfs2_local_alloc_count_bits(struct ocfs2_dinode * alloc)830 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
831 {
832 	u32 count;
833 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
834 
835 	count = memweight(la->la_bitmap, le16_to_cpu(la->la_size));
836 
837 	trace_ocfs2_local_alloc_count_bits(count);
838 	return count;
839 }
840 
ocfs2_local_alloc_find_clear_bits(struct ocfs2_super * osb,struct ocfs2_dinode * alloc,u32 * numbits,struct ocfs2_alloc_reservation * resv)841 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
842 				     struct ocfs2_dinode *alloc,
843 				     u32 *numbits,
844 				     struct ocfs2_alloc_reservation *resv)
845 {
846 	int numfound = 0, bitoff, left, startoff, lastzero;
847 	int local_resv = 0;
848 	struct ocfs2_alloc_reservation r;
849 	void *bitmap = NULL;
850 	struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap;
851 
852 	if (!alloc->id1.bitmap1.i_total) {
853 		bitoff = -1;
854 		goto bail;
855 	}
856 
857 	if (!resv) {
858 		local_resv = 1;
859 		ocfs2_resv_init_once(&r);
860 		ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP);
861 		resv = &r;
862 	}
863 
864 	numfound = *numbits;
865 	if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) {
866 		if (numfound < *numbits)
867 			*numbits = numfound;
868 		goto bail;
869 	}
870 
871 	/*
872 	 * Code error. While reservations are enabled, local
873 	 * allocation should _always_ go through them.
874 	 */
875 	BUG_ON(osb->osb_resv_level != 0);
876 
877 	/*
878 	 * Reservations are disabled. Handle this the old way.
879 	 */
880 
881 	bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
882 
883 	numfound = bitoff = startoff = 0;
884 	lastzero = -1;
885 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
886 	while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
887 		if (bitoff == left) {
888 			/* mlog(0, "bitoff (%d) == left", bitoff); */
889 			break;
890 		}
891 		/* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
892 		   "numfound = %d\n", bitoff, startoff, numfound);*/
893 
894 		/* Ok, we found a zero bit... is it contig. or do we
895 		 * start over?*/
896 		if (bitoff == startoff) {
897 			/* we found a zero */
898 			numfound++;
899 			startoff++;
900 		} else {
901 			/* got a zero after some ones */
902 			numfound = 1;
903 			startoff = bitoff+1;
904 		}
905 		/* we got everything we needed */
906 		if (numfound == *numbits) {
907 			/* mlog(0, "Found it all!\n"); */
908 			break;
909 		}
910 	}
911 
912 	trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound);
913 
914 	if (numfound == *numbits)
915 		bitoff = startoff - numfound;
916 	else
917 		bitoff = -1;
918 
919 bail:
920 	if (local_resv)
921 		ocfs2_resv_discard(resmap, resv);
922 
923 	trace_ocfs2_local_alloc_find_clear_bits(*numbits,
924 		le32_to_cpu(alloc->id1.bitmap1.i_total),
925 		bitoff, numfound);
926 
927 	return bitoff;
928 }
929 
ocfs2_clear_local_alloc(struct ocfs2_dinode * alloc)930 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
931 {
932 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
933 	int i;
934 
935 	alloc->id1.bitmap1.i_total = 0;
936 	alloc->id1.bitmap1.i_used = 0;
937 	la->la_bm_off = 0;
938 	for(i = 0; i < le16_to_cpu(la->la_size); i++)
939 		la->la_bitmap[i] = 0;
940 }
941 
942 #if 0
943 /* turn this on and uncomment below to aid debugging window shifts. */
944 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
945 				   unsigned int start,
946 				   unsigned int count)
947 {
948 	unsigned int tmp = count;
949 	while(tmp--) {
950 		if (ocfs2_test_bit(start + tmp, bitmap)) {
951 			printk("ocfs2_verify_zero_bits: start = %u, count = "
952 			       "%u\n", start, count);
953 			printk("ocfs2_verify_zero_bits: bit %u is set!",
954 			       start + tmp);
955 			BUG();
956 		}
957 	}
958 }
959 #endif
960 
961 /*
962  * sync the local alloc to main bitmap.
963  *
964  * assumes you've already locked the main bitmap -- the bitmap inode
965  * passed is used for caching.
966  */
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)967 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
968 				    handle_t *handle,
969 				    struct ocfs2_dinode *alloc,
970 				    struct inode *main_bm_inode,
971 				    struct buffer_head *main_bm_bh)
972 {
973 	int status = 0;
974 	int bit_off, left, count, start;
975 	u64 la_start_blk;
976 	u64 blkno;
977 	void *bitmap;
978 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
979 
980 	trace_ocfs2_sync_local_to_main(
981 	     le32_to_cpu(alloc->id1.bitmap1.i_total),
982 	     le32_to_cpu(alloc->id1.bitmap1.i_used));
983 
984 	if (!alloc->id1.bitmap1.i_total) {
985 		goto bail;
986 	}
987 
988 	if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
989 	    le32_to_cpu(alloc->id1.bitmap1.i_total)) {
990 		goto bail;
991 	}
992 
993 	la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
994 						le32_to_cpu(la->la_bm_off));
995 	bitmap = la->la_bitmap;
996 	start = count = bit_off = 0;
997 	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
998 
999 	while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
1000 	       != -1) {
1001 		if ((bit_off < left) && (bit_off == start)) {
1002 			count++;
1003 			start++;
1004 			continue;
1005 		}
1006 		if (count) {
1007 			blkno = la_start_blk +
1008 				ocfs2_clusters_to_blocks(osb->sb,
1009 							 start - count);
1010 
1011 			trace_ocfs2_sync_local_to_main_free(
1012 			     count, start - count,
1013 			     (unsigned long long)la_start_blk,
1014 			     (unsigned long long)blkno);
1015 
1016 			status = ocfs2_release_clusters(handle,
1017 							main_bm_inode,
1018 							main_bm_bh, blkno,
1019 							count);
1020 			if (status < 0) {
1021 				mlog_errno(status);
1022 				goto bail;
1023 			}
1024 		}
1025 		if (bit_off >= left)
1026 			break;
1027 		count = 1;
1028 		start = bit_off + 1;
1029 	}
1030 
1031 bail:
1032 	if (status)
1033 		mlog_errno(status);
1034 	return status;
1035 }
1036 
1037 enum ocfs2_la_event {
1038 	OCFS2_LA_EVENT_SLIDE,		/* Normal window slide. */
1039 	OCFS2_LA_EVENT_FRAGMENTED,	/* The global bitmap has
1040 					 * enough bits theoretically
1041 					 * free, but a contiguous
1042 					 * allocation could not be
1043 					 * found. */
1044 	OCFS2_LA_EVENT_ENOSPC,		/* Global bitmap doesn't have
1045 					 * enough bits free to satisfy
1046 					 * our request. */
1047 };
1048 #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
1049 /*
1050  * Given an event, calculate the size of our next local alloc window.
1051  *
1052  * This should always be called under i_mutex of the local alloc inode
1053  * so that local alloc disabling doesn't race with processes trying to
1054  * use the allocator.
1055  *
1056  * Returns the state which the local alloc was left in. This value can
1057  * be ignored by some paths.
1058  */
ocfs2_recalc_la_window(struct ocfs2_super * osb,enum ocfs2_la_event event)1059 static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
1060 				  enum ocfs2_la_event event)
1061 {
1062 	unsigned int bits;
1063 	int state;
1064 
1065 	spin_lock(&osb->osb_lock);
1066 	if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
1067 		WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
1068 		goto out_unlock;
1069 	}
1070 
1071 	/*
1072 	 * ENOSPC and fragmentation are treated similarly for now.
1073 	 */
1074 	if (event == OCFS2_LA_EVENT_ENOSPC ||
1075 	    event == OCFS2_LA_EVENT_FRAGMENTED) {
1076 		/*
1077 		 * We ran out of contiguous space in the primary
1078 		 * bitmap. Drastically reduce the number of bits used
1079 		 * by local alloc until we have to disable it.
1080 		 */
1081 		bits = osb->local_alloc_bits >> 1;
1082 		if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
1083 			/*
1084 			 * By setting state to THROTTLED, we'll keep
1085 			 * the number of local alloc bits used down
1086 			 * until an event occurs which would give us
1087 			 * reason to assume the bitmap situation might
1088 			 * have changed.
1089 			 */
1090 			osb->local_alloc_state = OCFS2_LA_THROTTLED;
1091 			osb->local_alloc_bits = bits;
1092 		} else {
1093 			osb->local_alloc_state = OCFS2_LA_DISABLED;
1094 		}
1095 		queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
1096 				   OCFS2_LA_ENABLE_INTERVAL);
1097 		goto out_unlock;
1098 	}
1099 
1100 	/*
1101 	 * Don't increase the size of the local alloc window until we
1102 	 * know we might be able to fulfill the request. Otherwise, we
1103 	 * risk bouncing around the global bitmap during periods of
1104 	 * low space.
1105 	 */
1106 	if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
1107 		osb->local_alloc_bits = osb->local_alloc_default_bits;
1108 
1109 out_unlock:
1110 	state = osb->local_alloc_state;
1111 	spin_unlock(&osb->osb_lock);
1112 
1113 	return state;
1114 }
1115 
ocfs2_local_alloc_reserve_for_window(struct ocfs2_super * osb,struct ocfs2_alloc_context ** ac,struct inode ** bitmap_inode,struct buffer_head ** bitmap_bh)1116 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
1117 						struct ocfs2_alloc_context **ac,
1118 						struct inode **bitmap_inode,
1119 						struct buffer_head **bitmap_bh)
1120 {
1121 	int status;
1122 
1123 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1124 	if (!(*ac)) {
1125 		status = -ENOMEM;
1126 		mlog_errno(status);
1127 		goto bail;
1128 	}
1129 
1130 retry_enospc:
1131 	(*ac)->ac_bits_wanted = osb->local_alloc_bits;
1132 	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1133 	if (status == -ENOSPC) {
1134 		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
1135 		    OCFS2_LA_DISABLED)
1136 			goto bail;
1137 
1138 		ocfs2_free_ac_resource(*ac);
1139 		memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
1140 		goto retry_enospc;
1141 	}
1142 	if (status < 0) {
1143 		mlog_errno(status);
1144 		goto bail;
1145 	}
1146 
1147 	*bitmap_inode = (*ac)->ac_inode;
1148 	igrab(*bitmap_inode);
1149 	*bitmap_bh = (*ac)->ac_bh;
1150 	get_bh(*bitmap_bh);
1151 	status = 0;
1152 bail:
1153 	if ((status < 0) && *ac) {
1154 		ocfs2_free_alloc_context(*ac);
1155 		*ac = NULL;
1156 	}
1157 
1158 	if (status)
1159 		mlog_errno(status);
1160 	return status;
1161 }
1162 
1163 /*
1164  * pass it the bitmap lock in lock_bh if you have it.
1165  */
ocfs2_local_alloc_new_window(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac)1166 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
1167 					handle_t *handle,
1168 					struct ocfs2_alloc_context *ac)
1169 {
1170 	int status = 0;
1171 	u32 cluster_off, cluster_count;
1172 	struct ocfs2_dinode *alloc = NULL;
1173 	struct ocfs2_local_alloc *la;
1174 
1175 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1176 	la = OCFS2_LOCAL_ALLOC(alloc);
1177 
1178 	trace_ocfs2_local_alloc_new_window(
1179 		le32_to_cpu(alloc->id1.bitmap1.i_total),
1180 		osb->local_alloc_bits);
1181 
1182 	/* Instruct the allocation code to try the most recently used
1183 	 * cluster group. We'll re-record the group used this pass
1184 	 * below. */
1185 	ac->ac_last_group = osb->la_last_gd;
1186 
1187 	/* we used the generic suballoc reserve function, but we set
1188 	 * everything up nicely, so there's no reason why we can't use
1189 	 * the more specific cluster api to claim bits. */
1190 	status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits,
1191 				      &cluster_off, &cluster_count);
1192 	if (status == -ENOSPC) {
1193 retry_enospc:
1194 		/*
1195 		 * Note: We could also try syncing the journal here to
1196 		 * allow use of any free bits which the current
1197 		 * transaction can't give us access to. --Mark
1198 		 */
1199 		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
1200 		    OCFS2_LA_DISABLED)
1201 			goto bail;
1202 
1203 		ac->ac_bits_wanted = osb->local_alloc_bits;
1204 		status = ocfs2_claim_clusters(handle, ac,
1205 					      osb->local_alloc_bits,
1206 					      &cluster_off,
1207 					      &cluster_count);
1208 		if (status == -ENOSPC)
1209 			goto retry_enospc;
1210 		/*
1211 		 * We only shrunk the *minimum* number of in our
1212 		 * request - it's entirely possible that the allocator
1213 		 * might give us more than we asked for.
1214 		 */
1215 		if (status == 0) {
1216 			spin_lock(&osb->osb_lock);
1217 			osb->local_alloc_bits = cluster_count;
1218 			spin_unlock(&osb->osb_lock);
1219 		}
1220 	}
1221 	if (status < 0) {
1222 		if (status != -ENOSPC)
1223 			mlog_errno(status);
1224 		goto bail;
1225 	}
1226 
1227 	osb->la_last_gd = ac->ac_last_group;
1228 
1229 	la->la_bm_off = cpu_to_le32(cluster_off);
1230 	alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
1231 	/* just in case... In the future when we find space ourselves,
1232 	 * we don't have to get all contiguous -- but we'll have to
1233 	 * set all previously used bits in bitmap and update
1234 	 * la_bits_set before setting the bits in the main bitmap. */
1235 	alloc->id1.bitmap1.i_used = 0;
1236 	memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
1237 	       le16_to_cpu(la->la_size));
1238 
1239 	ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count,
1240 			     OCFS2_LOCAL_ALLOC(alloc)->la_bitmap);
1241 
1242 	trace_ocfs2_local_alloc_new_window_result(
1243 		OCFS2_LOCAL_ALLOC(alloc)->la_bm_off,
1244 		le32_to_cpu(alloc->id1.bitmap1.i_total));
1245 
1246 bail:
1247 	if (status)
1248 		mlog_errno(status);
1249 	return status;
1250 }
1251 
1252 /* Note that we do *NOT* lock the local alloc inode here as
1253  * it's been locked already for us. */
ocfs2_local_alloc_slide_window(struct ocfs2_super * osb,struct inode * local_alloc_inode)1254 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
1255 					  struct inode *local_alloc_inode)
1256 {
1257 	int status = 0;
1258 	struct buffer_head *main_bm_bh = NULL;
1259 	struct inode *main_bm_inode = NULL;
1260 	handle_t *handle = NULL;
1261 	struct ocfs2_dinode *alloc;
1262 	struct ocfs2_dinode *alloc_copy = NULL;
1263 	struct ocfs2_alloc_context *ac = NULL;
1264 
1265 	ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
1266 
1267 	/* This will lock the main bitmap for us. */
1268 	status = ocfs2_local_alloc_reserve_for_window(osb,
1269 						      &ac,
1270 						      &main_bm_inode,
1271 						      &main_bm_bh);
1272 	if (status < 0) {
1273 		if (status != -ENOSPC)
1274 			mlog_errno(status);
1275 		goto bail;
1276 	}
1277 
1278 	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
1279 	if (IS_ERR(handle)) {
1280 		status = PTR_ERR(handle);
1281 		handle = NULL;
1282 		mlog_errno(status);
1283 		goto bail;
1284 	}
1285 
1286 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1287 
1288 	/* We want to clear the local alloc before doing anything
1289 	 * else, so that if we error later during this operation,
1290 	 * local alloc shutdown won't try to double free main bitmap
1291 	 * bits. Make a copy so the sync function knows which bits to
1292 	 * free. */
1293 	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
1294 	if (!alloc_copy) {
1295 		status = -ENOMEM;
1296 		mlog_errno(status);
1297 		goto bail;
1298 	}
1299 	memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
1300 
1301 	status = ocfs2_journal_access_di(handle,
1302 					 INODE_CACHE(local_alloc_inode),
1303 					 osb->local_alloc_bh,
1304 					 OCFS2_JOURNAL_ACCESS_WRITE);
1305 	if (status < 0) {
1306 		mlog_errno(status);
1307 		goto bail;
1308 	}
1309 
1310 	ocfs2_clear_local_alloc(alloc);
1311 	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
1312 
1313 	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
1314 					  main_bm_inode, main_bm_bh);
1315 	if (status < 0) {
1316 		mlog_errno(status);
1317 		goto bail;
1318 	}
1319 
1320 	status = ocfs2_local_alloc_new_window(osb, handle, ac);
1321 	if (status < 0) {
1322 		if (status != -ENOSPC)
1323 			mlog_errno(status);
1324 		goto bail;
1325 	}
1326 
1327 	atomic_inc(&osb->alloc_stats.moves);
1328 
1329 bail:
1330 	if (handle)
1331 		ocfs2_commit_trans(osb, handle);
1332 
1333 	brelse(main_bm_bh);
1334 
1335 	if (main_bm_inode)
1336 		iput(main_bm_inode);
1337 
1338 	kfree(alloc_copy);
1339 
1340 	if (ac)
1341 		ocfs2_free_alloc_context(ac);
1342 
1343 	if (status)
1344 		mlog_errno(status);
1345 	return status;
1346 }
1347 
1348