sparc64: Fix bootup with mcount in some configs.
[linux-2.6/mini2440.git] / fs / ocfs2 / localalloc.c
blobbac7e6abaf4763f75f873592beec3c0d4cb63472
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * localalloc.c
6 * Node local data allocation
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
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.
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.
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.
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>
32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
35 #include "ocfs2.h"
37 #include "alloc.h"
38 #include "blockcheck.h"
39 #include "dlmglue.h"
40 #include "inode.h"
41 #include "journal.h"
42 #include "localalloc.h"
43 #include "suballoc.h"
44 #include "super.h"
45 #include "sysfile.h"
47 #include "buffer_head_io.h"
49 #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab))
51 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
53 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
54 struct ocfs2_dinode *alloc,
55 u32 numbits);
57 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
59 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
60 handle_t *handle,
61 struct ocfs2_dinode *alloc,
62 struct inode *main_bm_inode,
63 struct buffer_head *main_bm_bh);
65 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
66 struct ocfs2_alloc_context **ac,
67 struct inode **bitmap_inode,
68 struct buffer_head **bitmap_bh);
70 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
71 handle_t *handle,
72 struct ocfs2_alloc_context *ac);
74 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
75 struct inode *local_alloc_inode);
77 static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
79 return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||
80 osb->local_alloc_state == OCFS2_LA_ENABLED);
83 void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
84 unsigned int num_clusters)
86 spin_lock(&osb->osb_lock);
87 if (osb->local_alloc_state == OCFS2_LA_DISABLED ||
88 osb->local_alloc_state == OCFS2_LA_THROTTLED)
89 if (num_clusters >= osb->local_alloc_default_bits) {
90 cancel_delayed_work(&osb->la_enable_wq);
91 osb->local_alloc_state = OCFS2_LA_ENABLED;
93 spin_unlock(&osb->osb_lock);
96 void ocfs2_la_enable_worker(struct work_struct *work)
98 struct ocfs2_super *osb =
99 container_of(work, struct ocfs2_super,
100 la_enable_wq.work);
101 spin_lock(&osb->osb_lock);
102 osb->local_alloc_state = OCFS2_LA_ENABLED;
103 spin_unlock(&osb->osb_lock);
107 * Tell us whether a given allocation should use the local alloc
108 * file. Otherwise, it has to go to the main bitmap.
110 * This function does semi-dirty reads of local alloc size and state!
111 * This is ok however, as the values are re-checked once under mutex.
113 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
115 int ret = 0;
116 int la_bits;
118 spin_lock(&osb->osb_lock);
119 la_bits = osb->local_alloc_bits;
121 if (!ocfs2_la_state_enabled(osb))
122 goto bail;
124 /* la_bits should be at least twice the size (in clusters) of
125 * a new block group. We want to be sure block group
126 * allocations go through the local alloc, so allow an
127 * allocation to take up to half the bitmap. */
128 if (bits > (la_bits / 2))
129 goto bail;
131 ret = 1;
132 bail:
133 mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
134 osb->local_alloc_state, (unsigned long long)bits, la_bits, ret);
135 spin_unlock(&osb->osb_lock);
136 return ret;
139 int ocfs2_load_local_alloc(struct ocfs2_super *osb)
141 int status = 0;
142 struct ocfs2_dinode *alloc = NULL;
143 struct buffer_head *alloc_bh = NULL;
144 u32 num_used;
145 struct inode *inode = NULL;
146 struct ocfs2_local_alloc *la;
148 mlog_entry_void();
150 if (osb->local_alloc_bits == 0)
151 goto bail;
153 if (osb->local_alloc_bits >= osb->bitmap_cpg) {
154 mlog(ML_NOTICE, "Requested local alloc window %d is larger "
155 "than max possible %u. Using defaults.\n",
156 osb->local_alloc_bits, (osb->bitmap_cpg - 1));
157 osb->local_alloc_bits =
158 ocfs2_megabytes_to_clusters(osb->sb,
159 OCFS2_DEFAULT_LOCAL_ALLOC_SIZE);
162 /* read the alloc off disk */
163 inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
164 osb->slot_num);
165 if (!inode) {
166 status = -EINVAL;
167 mlog_errno(status);
168 goto bail;
171 status = ocfs2_read_inode_block_full(inode, &alloc_bh,
172 OCFS2_BH_IGNORE_CACHE);
173 if (status < 0) {
174 mlog_errno(status);
175 goto bail;
178 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
179 la = OCFS2_LOCAL_ALLOC(alloc);
181 if (!(le32_to_cpu(alloc->i_flags) &
182 (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
183 mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
184 (unsigned long long)OCFS2_I(inode)->ip_blkno);
185 status = -EINVAL;
186 goto bail;
189 if ((la->la_size == 0) ||
190 (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
191 mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
192 le16_to_cpu(la->la_size));
193 status = -EINVAL;
194 goto bail;
197 /* do a little verification. */
198 num_used = ocfs2_local_alloc_count_bits(alloc);
200 /* hopefully the local alloc has always been recovered before
201 * we load it. */
202 if (num_used
203 || alloc->id1.bitmap1.i_used
204 || alloc->id1.bitmap1.i_total
205 || la->la_bm_off)
206 mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
207 "found = %u, set = %u, taken = %u, off = %u\n",
208 num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
209 le32_to_cpu(alloc->id1.bitmap1.i_total),
210 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
212 osb->local_alloc_bh = alloc_bh;
213 osb->local_alloc_state = OCFS2_LA_ENABLED;
215 bail:
216 if (status < 0)
217 brelse(alloc_bh);
218 if (inode)
219 iput(inode);
221 mlog(0, "Local alloc window bits = %d\n", osb->local_alloc_bits);
223 mlog_exit(status);
224 return status;
228 * return any unused bits to the bitmap and write out a clean
229 * local_alloc.
231 * local_alloc_bh is optional. If not passed, we will simply use the
232 * one off osb. If you do pass it however, be warned that it *will* be
233 * returned brelse'd and NULL'd out.*/
234 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
236 int status;
237 handle_t *handle;
238 struct inode *local_alloc_inode = NULL;
239 struct buffer_head *bh = NULL;
240 struct buffer_head *main_bm_bh = NULL;
241 struct inode *main_bm_inode = NULL;
242 struct ocfs2_dinode *alloc_copy = NULL;
243 struct ocfs2_dinode *alloc = NULL;
245 mlog_entry_void();
247 cancel_delayed_work(&osb->la_enable_wq);
248 flush_workqueue(ocfs2_wq);
250 if (osb->local_alloc_state == OCFS2_LA_UNUSED)
251 goto out;
253 local_alloc_inode =
254 ocfs2_get_system_file_inode(osb,
255 LOCAL_ALLOC_SYSTEM_INODE,
256 osb->slot_num);
257 if (!local_alloc_inode) {
258 status = -ENOENT;
259 mlog_errno(status);
260 goto out;
263 osb->local_alloc_state = OCFS2_LA_DISABLED;
265 main_bm_inode = ocfs2_get_system_file_inode(osb,
266 GLOBAL_BITMAP_SYSTEM_INODE,
267 OCFS2_INVALID_SLOT);
268 if (!main_bm_inode) {
269 status = -EINVAL;
270 mlog_errno(status);
271 goto out;
274 mutex_lock(&main_bm_inode->i_mutex);
276 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
277 if (status < 0) {
278 mlog_errno(status);
279 goto out_mutex;
282 /* WINDOW_MOVE_CREDITS is a bit heavy... */
283 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
284 if (IS_ERR(handle)) {
285 mlog_errno(PTR_ERR(handle));
286 handle = NULL;
287 goto out_unlock;
290 bh = osb->local_alloc_bh;
291 alloc = (struct ocfs2_dinode *) bh->b_data;
293 alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
294 if (!alloc_copy) {
295 status = -ENOMEM;
296 goto out_commit;
298 memcpy(alloc_copy, alloc, bh->b_size);
300 status = ocfs2_journal_access_di(handle, local_alloc_inode, bh,
301 OCFS2_JOURNAL_ACCESS_WRITE);
302 if (status < 0) {
303 mlog_errno(status);
304 goto out_commit;
307 ocfs2_clear_local_alloc(alloc);
309 status = ocfs2_journal_dirty(handle, bh);
310 if (status < 0) {
311 mlog_errno(status);
312 goto out_commit;
315 brelse(bh);
316 osb->local_alloc_bh = NULL;
317 osb->local_alloc_state = OCFS2_LA_UNUSED;
319 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
320 main_bm_inode, main_bm_bh);
321 if (status < 0)
322 mlog_errno(status);
324 out_commit:
325 ocfs2_commit_trans(osb, handle);
327 out_unlock:
328 brelse(main_bm_bh);
330 ocfs2_inode_unlock(main_bm_inode, 1);
332 out_mutex:
333 mutex_unlock(&main_bm_inode->i_mutex);
334 iput(main_bm_inode);
336 out:
337 if (local_alloc_inode)
338 iput(local_alloc_inode);
340 if (alloc_copy)
341 kfree(alloc_copy);
343 mlog_exit_void();
347 * We want to free the bitmap bits outside of any recovery context as
348 * we'll need a cluster lock to do so, but we must clear the local
349 * alloc before giving up the recovered nodes journal. To solve this,
350 * we kmalloc a copy of the local alloc before it's change for the
351 * caller to process with ocfs2_complete_local_alloc_recovery
353 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
354 int slot_num,
355 struct ocfs2_dinode **alloc_copy)
357 int status = 0;
358 struct buffer_head *alloc_bh = NULL;
359 struct inode *inode = NULL;
360 struct ocfs2_dinode *alloc;
362 mlog_entry("(slot_num = %d)\n", slot_num);
364 *alloc_copy = NULL;
366 inode = ocfs2_get_system_file_inode(osb,
367 LOCAL_ALLOC_SYSTEM_INODE,
368 slot_num);
369 if (!inode) {
370 status = -EINVAL;
371 mlog_errno(status);
372 goto bail;
375 mutex_lock(&inode->i_mutex);
377 status = ocfs2_read_inode_block_full(inode, &alloc_bh,
378 OCFS2_BH_IGNORE_CACHE);
379 if (status < 0) {
380 mlog_errno(status);
381 goto bail;
384 *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
385 if (!(*alloc_copy)) {
386 status = -ENOMEM;
387 goto bail;
389 memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
391 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
392 ocfs2_clear_local_alloc(alloc);
394 ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
395 status = ocfs2_write_block(osb, alloc_bh, inode);
396 if (status < 0)
397 mlog_errno(status);
399 bail:
400 if ((status < 0) && (*alloc_copy)) {
401 kfree(*alloc_copy);
402 *alloc_copy = NULL;
405 brelse(alloc_bh);
407 if (inode) {
408 mutex_unlock(&inode->i_mutex);
409 iput(inode);
412 mlog_exit(status);
413 return status;
417 * Step 2: By now, we've completed the journal recovery, we've stamped
418 * a clean local alloc on disk and dropped the node out of the
419 * recovery map. Dlm locks will no longer stall, so lets clear out the
420 * main bitmap.
422 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
423 struct ocfs2_dinode *alloc)
425 int status;
426 handle_t *handle;
427 struct buffer_head *main_bm_bh = NULL;
428 struct inode *main_bm_inode;
430 mlog_entry_void();
432 main_bm_inode = ocfs2_get_system_file_inode(osb,
433 GLOBAL_BITMAP_SYSTEM_INODE,
434 OCFS2_INVALID_SLOT);
435 if (!main_bm_inode) {
436 status = -EINVAL;
437 mlog_errno(status);
438 goto out;
441 mutex_lock(&main_bm_inode->i_mutex);
443 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
444 if (status < 0) {
445 mlog_errno(status);
446 goto out_mutex;
449 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
450 if (IS_ERR(handle)) {
451 status = PTR_ERR(handle);
452 handle = NULL;
453 mlog_errno(status);
454 goto out_unlock;
457 /* we want the bitmap change to be recorded on disk asap */
458 handle->h_sync = 1;
460 status = ocfs2_sync_local_to_main(osb, handle, alloc,
461 main_bm_inode, main_bm_bh);
462 if (status < 0)
463 mlog_errno(status);
465 ocfs2_commit_trans(osb, handle);
467 out_unlock:
468 ocfs2_inode_unlock(main_bm_inode, 1);
470 out_mutex:
471 mutex_unlock(&main_bm_inode->i_mutex);
473 brelse(main_bm_bh);
475 iput(main_bm_inode);
477 out:
478 if (!status)
479 ocfs2_init_inode_steal_slot(osb);
480 mlog_exit(status);
481 return status;
484 /* Check to see if the local alloc window is within ac->ac_max_block */
485 static int ocfs2_local_alloc_in_range(struct inode *inode,
486 struct ocfs2_alloc_context *ac,
487 u32 bits_wanted)
489 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
490 struct ocfs2_dinode *alloc;
491 struct ocfs2_local_alloc *la;
492 int start;
493 u64 block_off;
495 if (!ac->ac_max_block)
496 return 1;
498 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
499 la = OCFS2_LOCAL_ALLOC(alloc);
501 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
502 if (start == -1) {
503 mlog_errno(-ENOSPC);
504 return 0;
508 * Converting (bm_off + start + bits_wanted) to blocks gives us
509 * the blkno just past our actual allocation. This is perfect
510 * to compare with ac_max_block.
512 block_off = ocfs2_clusters_to_blocks(inode->i_sb,
513 le32_to_cpu(la->la_bm_off) +
514 start + bits_wanted);
515 mlog(0, "Checking %llu against %llu\n",
516 (unsigned long long)block_off,
517 (unsigned long long)ac->ac_max_block);
518 if (block_off > ac->ac_max_block)
519 return 0;
521 return 1;
525 * make sure we've got at least bits_wanted contiguous bits in the
526 * local alloc. You lose them when you drop i_mutex.
528 * We will add ourselves to the transaction passed in, but may start
529 * our own in order to shift windows.
531 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
532 u32 bits_wanted,
533 struct ocfs2_alloc_context *ac)
535 int status;
536 struct ocfs2_dinode *alloc;
537 struct inode *local_alloc_inode;
538 unsigned int free_bits;
540 mlog_entry_void();
542 BUG_ON(!ac);
544 local_alloc_inode =
545 ocfs2_get_system_file_inode(osb,
546 LOCAL_ALLOC_SYSTEM_INODE,
547 osb->slot_num);
548 if (!local_alloc_inode) {
549 status = -ENOENT;
550 mlog_errno(status);
551 goto bail;
554 mutex_lock(&local_alloc_inode->i_mutex);
557 * We must double check state and allocator bits because
558 * another process may have changed them while holding i_mutex.
560 spin_lock(&osb->osb_lock);
561 if (!ocfs2_la_state_enabled(osb) ||
562 (bits_wanted > osb->local_alloc_bits)) {
563 spin_unlock(&osb->osb_lock);
564 status = -ENOSPC;
565 goto bail;
567 spin_unlock(&osb->osb_lock);
569 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
571 #ifdef CONFIG_OCFS2_DEBUG_FS
572 if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
573 ocfs2_local_alloc_count_bits(alloc)) {
574 ocfs2_error(osb->sb, "local alloc inode %llu says it has "
575 "%u free bits, but a count shows %u",
576 (unsigned long long)le64_to_cpu(alloc->i_blkno),
577 le32_to_cpu(alloc->id1.bitmap1.i_used),
578 ocfs2_local_alloc_count_bits(alloc));
579 status = -EIO;
580 goto bail;
582 #endif
584 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
585 le32_to_cpu(alloc->id1.bitmap1.i_used);
586 if (bits_wanted > free_bits) {
587 /* uhoh, window change time. */
588 status =
589 ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
590 if (status < 0) {
591 if (status != -ENOSPC)
592 mlog_errno(status);
593 goto bail;
597 * Under certain conditions, the window slide code
598 * might have reduced the number of bits available or
599 * disabled the the local alloc entirely. Re-check
600 * here and return -ENOSPC if necessary.
602 status = -ENOSPC;
603 if (!ocfs2_la_state_enabled(osb))
604 goto bail;
606 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
607 le32_to_cpu(alloc->id1.bitmap1.i_used);
608 if (bits_wanted > free_bits)
609 goto bail;
612 if (ac->ac_max_block)
613 mlog(0, "Calling in_range for max block %llu\n",
614 (unsigned long long)ac->ac_max_block);
616 if (!ocfs2_local_alloc_in_range(local_alloc_inode, ac,
617 bits_wanted)) {
619 * The window is outside ac->ac_max_block.
620 * This errno tells the caller to keep localalloc enabled
621 * but to get the allocation from the main bitmap.
623 status = -EFBIG;
624 goto bail;
627 ac->ac_inode = local_alloc_inode;
628 /* We should never use localalloc from another slot */
629 ac->ac_alloc_slot = osb->slot_num;
630 ac->ac_which = OCFS2_AC_USE_LOCAL;
631 get_bh(osb->local_alloc_bh);
632 ac->ac_bh = osb->local_alloc_bh;
633 status = 0;
634 bail:
635 if (status < 0 && local_alloc_inode) {
636 mutex_unlock(&local_alloc_inode->i_mutex);
637 iput(local_alloc_inode);
640 mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted, osb->slot_num,
641 status);
643 mlog_exit(status);
644 return status;
647 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
648 handle_t *handle,
649 struct ocfs2_alloc_context *ac,
650 u32 bits_wanted,
651 u32 *bit_off,
652 u32 *num_bits)
654 int status, start;
655 struct inode *local_alloc_inode;
656 void *bitmap;
657 struct ocfs2_dinode *alloc;
658 struct ocfs2_local_alloc *la;
660 mlog_entry_void();
661 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
663 local_alloc_inode = ac->ac_inode;
664 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
665 la = OCFS2_LOCAL_ALLOC(alloc);
667 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
668 if (start == -1) {
669 /* TODO: Shouldn't we just BUG here? */
670 status = -ENOSPC;
671 mlog_errno(status);
672 goto bail;
675 bitmap = la->la_bitmap;
676 *bit_off = le32_to_cpu(la->la_bm_off) + start;
677 /* local alloc is always contiguous by nature -- we never
678 * delete bits from it! */
679 *num_bits = bits_wanted;
681 status = ocfs2_journal_access_di(handle, local_alloc_inode,
682 osb->local_alloc_bh,
683 OCFS2_JOURNAL_ACCESS_WRITE);
684 if (status < 0) {
685 mlog_errno(status);
686 goto bail;
689 while(bits_wanted--)
690 ocfs2_set_bit(start++, bitmap);
692 le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
694 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
695 if (status < 0) {
696 mlog_errno(status);
697 goto bail;
700 status = 0;
701 bail:
702 mlog_exit(status);
703 return status;
706 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
708 int i;
709 u8 *buffer;
710 u32 count = 0;
711 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
713 mlog_entry_void();
715 buffer = la->la_bitmap;
716 for (i = 0; i < le16_to_cpu(la->la_size); i++)
717 count += hweight8(buffer[i]);
719 mlog_exit(count);
720 return count;
723 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
724 struct ocfs2_dinode *alloc,
725 u32 numbits)
727 int numfound, bitoff, left, startoff, lastzero;
728 void *bitmap = NULL;
730 mlog_entry("(numbits wanted = %u)\n", numbits);
732 if (!alloc->id1.bitmap1.i_total) {
733 mlog(0, "No bits in my window!\n");
734 bitoff = -1;
735 goto bail;
738 bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
740 numfound = bitoff = startoff = 0;
741 lastzero = -1;
742 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
743 while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
744 if (bitoff == left) {
745 /* mlog(0, "bitoff (%d) == left", bitoff); */
746 break;
748 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
749 "numfound = %d\n", bitoff, startoff, numfound);*/
751 /* Ok, we found a zero bit... is it contig. or do we
752 * start over?*/
753 if (bitoff == startoff) {
754 /* we found a zero */
755 numfound++;
756 startoff++;
757 } else {
758 /* got a zero after some ones */
759 numfound = 1;
760 startoff = bitoff+1;
762 /* we got everything we needed */
763 if (numfound == numbits) {
764 /* mlog(0, "Found it all!\n"); */
765 break;
769 mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff,
770 numfound);
772 if (numfound == numbits)
773 bitoff = startoff - numfound;
774 else
775 bitoff = -1;
777 bail:
778 mlog_exit(bitoff);
779 return bitoff;
782 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
784 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
785 int i;
786 mlog_entry_void();
788 alloc->id1.bitmap1.i_total = 0;
789 alloc->id1.bitmap1.i_used = 0;
790 la->la_bm_off = 0;
791 for(i = 0; i < le16_to_cpu(la->la_size); i++)
792 la->la_bitmap[i] = 0;
794 mlog_exit_void();
797 #if 0
798 /* turn this on and uncomment below to aid debugging window shifts. */
799 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
800 unsigned int start,
801 unsigned int count)
803 unsigned int tmp = count;
804 while(tmp--) {
805 if (ocfs2_test_bit(start + tmp, bitmap)) {
806 printk("ocfs2_verify_zero_bits: start = %u, count = "
807 "%u\n", start, count);
808 printk("ocfs2_verify_zero_bits: bit %u is set!",
809 start + tmp);
810 BUG();
814 #endif
817 * sync the local alloc to main bitmap.
819 * assumes you've already locked the main bitmap -- the bitmap inode
820 * passed is used for caching.
822 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
823 handle_t *handle,
824 struct ocfs2_dinode *alloc,
825 struct inode *main_bm_inode,
826 struct buffer_head *main_bm_bh)
828 int status = 0;
829 int bit_off, left, count, start;
830 u64 la_start_blk;
831 u64 blkno;
832 void *bitmap;
833 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
835 mlog_entry("total = %u, used = %u\n",
836 le32_to_cpu(alloc->id1.bitmap1.i_total),
837 le32_to_cpu(alloc->id1.bitmap1.i_used));
839 if (!alloc->id1.bitmap1.i_total) {
840 mlog(0, "nothing to sync!\n");
841 goto bail;
844 if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
845 le32_to_cpu(alloc->id1.bitmap1.i_total)) {
846 mlog(0, "all bits were taken!\n");
847 goto bail;
850 la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
851 le32_to_cpu(la->la_bm_off));
852 bitmap = la->la_bitmap;
853 start = count = bit_off = 0;
854 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
856 while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
857 != -1) {
858 if ((bit_off < left) && (bit_off == start)) {
859 count++;
860 start++;
861 continue;
863 if (count) {
864 blkno = la_start_blk +
865 ocfs2_clusters_to_blocks(osb->sb,
866 start - count);
868 mlog(0, "freeing %u bits starting at local alloc bit "
869 "%u (la_start_blk = %llu, blkno = %llu)\n",
870 count, start - count,
871 (unsigned long long)la_start_blk,
872 (unsigned long long)blkno);
874 status = ocfs2_free_clusters(handle, main_bm_inode,
875 main_bm_bh, blkno, count);
876 if (status < 0) {
877 mlog_errno(status);
878 goto bail;
881 if (bit_off >= left)
882 break;
883 count = 1;
884 start = bit_off + 1;
887 bail:
888 mlog_exit(status);
889 return status;
892 enum ocfs2_la_event {
893 OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */
894 OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has
895 * enough bits theoretically
896 * free, but a contiguous
897 * allocation could not be
898 * found. */
899 OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have
900 * enough bits free to satisfy
901 * our request. */
903 #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
905 * Given an event, calculate the size of our next local alloc window.
907 * This should always be called under i_mutex of the local alloc inode
908 * so that local alloc disabling doesn't race with processes trying to
909 * use the allocator.
911 * Returns the state which the local alloc was left in. This value can
912 * be ignored by some paths.
914 static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
915 enum ocfs2_la_event event)
917 unsigned int bits;
918 int state;
920 spin_lock(&osb->osb_lock);
921 if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
922 WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
923 goto out_unlock;
927 * ENOSPC and fragmentation are treated similarly for now.
929 if (event == OCFS2_LA_EVENT_ENOSPC ||
930 event == OCFS2_LA_EVENT_FRAGMENTED) {
932 * We ran out of contiguous space in the primary
933 * bitmap. Drastically reduce the number of bits used
934 * by local alloc until we have to disable it.
936 bits = osb->local_alloc_bits >> 1;
937 if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
939 * By setting state to THROTTLED, we'll keep
940 * the number of local alloc bits used down
941 * until an event occurs which would give us
942 * reason to assume the bitmap situation might
943 * have changed.
945 osb->local_alloc_state = OCFS2_LA_THROTTLED;
946 osb->local_alloc_bits = bits;
947 } else {
948 osb->local_alloc_state = OCFS2_LA_DISABLED;
950 queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
951 OCFS2_LA_ENABLE_INTERVAL);
952 goto out_unlock;
956 * Don't increase the size of the local alloc window until we
957 * know we might be able to fulfill the request. Otherwise, we
958 * risk bouncing around the global bitmap during periods of
959 * low space.
961 if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
962 osb->local_alloc_bits = osb->local_alloc_default_bits;
964 out_unlock:
965 state = osb->local_alloc_state;
966 spin_unlock(&osb->osb_lock);
968 return state;
971 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
972 struct ocfs2_alloc_context **ac,
973 struct inode **bitmap_inode,
974 struct buffer_head **bitmap_bh)
976 int status;
978 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
979 if (!(*ac)) {
980 status = -ENOMEM;
981 mlog_errno(status);
982 goto bail;
985 retry_enospc:
986 (*ac)->ac_bits_wanted = osb->local_alloc_bits;
988 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
989 if (status == -ENOSPC) {
990 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
991 OCFS2_LA_DISABLED)
992 goto bail;
994 ocfs2_free_ac_resource(*ac);
995 memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
996 goto retry_enospc;
998 if (status < 0) {
999 mlog_errno(status);
1000 goto bail;
1003 *bitmap_inode = (*ac)->ac_inode;
1004 igrab(*bitmap_inode);
1005 *bitmap_bh = (*ac)->ac_bh;
1006 get_bh(*bitmap_bh);
1007 status = 0;
1008 bail:
1009 if ((status < 0) && *ac) {
1010 ocfs2_free_alloc_context(*ac);
1011 *ac = NULL;
1014 mlog_exit(status);
1015 return status;
1019 * pass it the bitmap lock in lock_bh if you have it.
1021 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
1022 handle_t *handle,
1023 struct ocfs2_alloc_context *ac)
1025 int status = 0;
1026 u32 cluster_off, cluster_count;
1027 struct ocfs2_dinode *alloc = NULL;
1028 struct ocfs2_local_alloc *la;
1030 mlog_entry_void();
1032 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1033 la = OCFS2_LOCAL_ALLOC(alloc);
1035 if (alloc->id1.bitmap1.i_total)
1036 mlog(0, "asking me to alloc a new window over a non-empty "
1037 "one\n");
1039 mlog(0, "Allocating %u clusters for a new window.\n",
1040 osb->local_alloc_bits);
1042 /* Instruct the allocation code to try the most recently used
1043 * cluster group. We'll re-record the group used this pass
1044 * below. */
1045 ac->ac_last_group = osb->la_last_gd;
1047 /* we used the generic suballoc reserve function, but we set
1048 * everything up nicely, so there's no reason why we can't use
1049 * the more specific cluster api to claim bits. */
1050 status = ocfs2_claim_clusters(osb, handle, ac, osb->local_alloc_bits,
1051 &cluster_off, &cluster_count);
1052 if (status == -ENOSPC) {
1053 retry_enospc:
1055 * Note: We could also try syncing the journal here to
1056 * allow use of any free bits which the current
1057 * transaction can't give us access to. --Mark
1059 if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
1060 OCFS2_LA_DISABLED)
1061 goto bail;
1063 status = ocfs2_claim_clusters(osb, handle, ac,
1064 osb->local_alloc_bits,
1065 &cluster_off,
1066 &cluster_count);
1067 if (status == -ENOSPC)
1068 goto retry_enospc;
1070 * We only shrunk the *minimum* number of in our
1071 * request - it's entirely possible that the allocator
1072 * might give us more than we asked for.
1074 if (status == 0) {
1075 spin_lock(&osb->osb_lock);
1076 osb->local_alloc_bits = cluster_count;
1077 spin_unlock(&osb->osb_lock);
1080 if (status < 0) {
1081 if (status != -ENOSPC)
1082 mlog_errno(status);
1083 goto bail;
1086 osb->la_last_gd = ac->ac_last_group;
1088 la->la_bm_off = cpu_to_le32(cluster_off);
1089 alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
1090 /* just in case... In the future when we find space ourselves,
1091 * we don't have to get all contiguous -- but we'll have to
1092 * set all previously used bits in bitmap and update
1093 * la_bits_set before setting the bits in the main bitmap. */
1094 alloc->id1.bitmap1.i_used = 0;
1095 memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
1096 le16_to_cpu(la->la_size));
1098 mlog(0, "New window allocated:\n");
1099 mlog(0, "window la_bm_off = %u\n",
1100 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
1101 mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total));
1103 bail:
1104 mlog_exit(status);
1105 return status;
1108 /* Note that we do *NOT* lock the local alloc inode here as
1109 * it's been locked already for us. */
1110 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
1111 struct inode *local_alloc_inode)
1113 int status = 0;
1114 struct buffer_head *main_bm_bh = NULL;
1115 struct inode *main_bm_inode = NULL;
1116 handle_t *handle = NULL;
1117 struct ocfs2_dinode *alloc;
1118 struct ocfs2_dinode *alloc_copy = NULL;
1119 struct ocfs2_alloc_context *ac = NULL;
1121 mlog_entry_void();
1123 ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
1125 /* This will lock the main bitmap for us. */
1126 status = ocfs2_local_alloc_reserve_for_window(osb,
1127 &ac,
1128 &main_bm_inode,
1129 &main_bm_bh);
1130 if (status < 0) {
1131 if (status != -ENOSPC)
1132 mlog_errno(status);
1133 goto bail;
1136 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
1137 if (IS_ERR(handle)) {
1138 status = PTR_ERR(handle);
1139 handle = NULL;
1140 mlog_errno(status);
1141 goto bail;
1144 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
1146 /* We want to clear the local alloc before doing anything
1147 * else, so that if we error later during this operation,
1148 * local alloc shutdown won't try to double free main bitmap
1149 * bits. Make a copy so the sync function knows which bits to
1150 * free. */
1151 alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
1152 if (!alloc_copy) {
1153 status = -ENOMEM;
1154 mlog_errno(status);
1155 goto bail;
1157 memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
1159 status = ocfs2_journal_access_di(handle, local_alloc_inode,
1160 osb->local_alloc_bh,
1161 OCFS2_JOURNAL_ACCESS_WRITE);
1162 if (status < 0) {
1163 mlog_errno(status);
1164 goto bail;
1167 ocfs2_clear_local_alloc(alloc);
1169 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
1170 if (status < 0) {
1171 mlog_errno(status);
1172 goto bail;
1175 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
1176 main_bm_inode, main_bm_bh);
1177 if (status < 0) {
1178 mlog_errno(status);
1179 goto bail;
1182 status = ocfs2_local_alloc_new_window(osb, handle, ac);
1183 if (status < 0) {
1184 if (status != -ENOSPC)
1185 mlog_errno(status);
1186 goto bail;
1189 atomic_inc(&osb->alloc_stats.moves);
1191 status = 0;
1192 bail:
1193 if (handle)
1194 ocfs2_commit_trans(osb, handle);
1196 brelse(main_bm_bh);
1198 if (main_bm_inode)
1199 iput(main_bm_inode);
1201 if (alloc_copy)
1202 kfree(alloc_copy);
1204 if (ac)
1205 ocfs2_free_alloc_context(ac);
1207 mlog_exit(status);
1208 return status;