1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
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.
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>
41 #include "localalloc.h"
46 #include "buffer_head_io.h"
48 #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab))
50 static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb);
52 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
54 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
55 struct ocfs2_dinode *alloc,
58 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
60 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
62 struct ocfs2_dinode *alloc,
63 struct inode *main_bm_inode,
64 struct buffer_head *main_bm_bh);
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);
71 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
73 struct ocfs2_alloc_context *ac);
75 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
76 struct inode *local_alloc_inode);
78 static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb)
80 BUG_ON(osb->s_clustersize_bits > 20);
82 /* Size local alloc windows by the megabyte */
83 return osb->local_alloc_size << (20 - osb->s_clustersize_bits);
87 * Tell us whether a given allocation should use the local alloc
88 * file. Otherwise, it has to go to the main bitmap.
90 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
92 int la_bits = ocfs2_local_alloc_window_bits(osb);
95 if (osb->local_alloc_state != OCFS2_LA_ENABLED)
98 /* la_bits should be at least twice the size (in clusters) of
99 * a new block group. We want to be sure block group
100 * allocations go through the local alloc, so allow an
101 * allocation to take up to half the bitmap. */
102 if (bits > (la_bits / 2))
107 mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
108 osb->local_alloc_state, (unsigned long long)bits, la_bits, ret);
112 int ocfs2_load_local_alloc(struct ocfs2_super *osb)
115 struct ocfs2_dinode *alloc = NULL;
116 struct buffer_head *alloc_bh = NULL;
118 struct inode *inode = NULL;
119 struct ocfs2_local_alloc *la;
123 if (ocfs2_mount_local(osb))
126 if (osb->local_alloc_size == 0)
129 if (ocfs2_local_alloc_window_bits(osb) >= osb->bitmap_cpg) {
130 mlog(ML_NOTICE, "Requested local alloc window %d is larger "
131 "than max possible %u. Using defaults.\n",
132 ocfs2_local_alloc_window_bits(osb), (osb->bitmap_cpg - 1));
133 osb->local_alloc_size = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
136 /* read the alloc off disk */
137 inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
145 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
146 &alloc_bh, 0, inode);
152 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
153 la = OCFS2_LOCAL_ALLOC(alloc);
155 if (!(le32_to_cpu(alloc->i_flags) &
156 (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
157 mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
158 (unsigned long long)OCFS2_I(inode)->ip_blkno);
163 if ((la->la_size == 0) ||
164 (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
165 mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
166 le16_to_cpu(la->la_size));
171 /* do a little verification. */
172 num_used = ocfs2_local_alloc_count_bits(alloc);
174 /* hopefully the local alloc has always been recovered before
177 || alloc->id1.bitmap1.i_used
178 || alloc->id1.bitmap1.i_total
180 mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
181 "found = %u, set = %u, taken = %u, off = %u\n",
182 num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
183 le32_to_cpu(alloc->id1.bitmap1.i_total),
184 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
186 osb->local_alloc_bh = alloc_bh;
187 osb->local_alloc_state = OCFS2_LA_ENABLED;
196 mlog(0, "Local alloc window bits = %d\n",
197 ocfs2_local_alloc_window_bits(osb));
204 * return any unused bits to the bitmap and write out a clean
207 * local_alloc_bh is optional. If not passed, we will simply use the
208 * one off osb. If you do pass it however, be warned that it *will* be
209 * returned brelse'd and NULL'd out.*/
210 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
214 struct inode *local_alloc_inode = NULL;
215 struct buffer_head *bh = NULL;
216 struct buffer_head *main_bm_bh = NULL;
217 struct inode *main_bm_inode = NULL;
218 struct ocfs2_dinode *alloc_copy = NULL;
219 struct ocfs2_dinode *alloc = NULL;
223 if (osb->local_alloc_state == OCFS2_LA_UNUSED)
227 ocfs2_get_system_file_inode(osb,
228 LOCAL_ALLOC_SYSTEM_INODE,
230 if (!local_alloc_inode) {
236 osb->local_alloc_state = OCFS2_LA_DISABLED;
238 main_bm_inode = ocfs2_get_system_file_inode(osb,
239 GLOBAL_BITMAP_SYSTEM_INODE,
241 if (!main_bm_inode) {
247 mutex_lock(&main_bm_inode->i_mutex);
249 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
255 /* WINDOW_MOVE_CREDITS is a bit heavy... */
256 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
257 if (IS_ERR(handle)) {
258 mlog_errno(PTR_ERR(handle));
263 bh = osb->local_alloc_bh;
264 alloc = (struct ocfs2_dinode *) bh->b_data;
266 alloc_copy = kmalloc(bh->b_size, GFP_KERNEL);
271 memcpy(alloc_copy, alloc, bh->b_size);
273 status = ocfs2_journal_access(handle, local_alloc_inode, bh,
274 OCFS2_JOURNAL_ACCESS_WRITE);
280 ocfs2_clear_local_alloc(alloc);
282 status = ocfs2_journal_dirty(handle, bh);
289 osb->local_alloc_bh = NULL;
290 osb->local_alloc_state = OCFS2_LA_UNUSED;
292 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
293 main_bm_inode, main_bm_bh);
298 ocfs2_commit_trans(osb, handle);
304 ocfs2_inode_unlock(main_bm_inode, 1);
307 mutex_unlock(&main_bm_inode->i_mutex);
311 if (local_alloc_inode)
312 iput(local_alloc_inode);
321 * We want to free the bitmap bits outside of any recovery context as
322 * we'll need a cluster lock to do so, but we must clear the local
323 * alloc before giving up the recovered nodes journal. To solve this,
324 * we kmalloc a copy of the local alloc before it's change for the
325 * caller to process with ocfs2_complete_local_alloc_recovery
327 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
329 struct ocfs2_dinode **alloc_copy)
332 struct buffer_head *alloc_bh = NULL;
333 struct inode *inode = NULL;
334 struct ocfs2_dinode *alloc;
336 mlog_entry("(slot_num = %d)\n", slot_num);
340 inode = ocfs2_get_system_file_inode(osb,
341 LOCAL_ALLOC_SYSTEM_INODE,
349 mutex_lock(&inode->i_mutex);
351 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
352 &alloc_bh, 0, inode);
358 *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
359 if (!(*alloc_copy)) {
363 memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
365 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
366 ocfs2_clear_local_alloc(alloc);
368 status = ocfs2_write_block(osb, alloc_bh, inode);
373 if ((status < 0) && (*alloc_copy)) {
382 mutex_unlock(&inode->i_mutex);
391 * Step 2: By now, we've completed the journal recovery, we've stamped
392 * a clean local alloc on disk and dropped the node out of the
393 * recovery map. Dlm locks will no longer stall, so lets clear out the
396 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
397 struct ocfs2_dinode *alloc)
401 struct buffer_head *main_bm_bh = NULL;
402 struct inode *main_bm_inode;
406 main_bm_inode = ocfs2_get_system_file_inode(osb,
407 GLOBAL_BITMAP_SYSTEM_INODE,
409 if (!main_bm_inode) {
415 mutex_lock(&main_bm_inode->i_mutex);
417 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
423 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
424 if (IS_ERR(handle)) {
425 status = PTR_ERR(handle);
431 /* we want the bitmap change to be recorded on disk asap */
434 status = ocfs2_sync_local_to_main(osb, handle, alloc,
435 main_bm_inode, main_bm_bh);
439 ocfs2_commit_trans(osb, handle);
442 ocfs2_inode_unlock(main_bm_inode, 1);
445 mutex_unlock(&main_bm_inode->i_mutex);
458 * make sure we've got at least bitswanted contiguous bits in the
459 * local alloc. You lose them when you drop i_mutex.
461 * We will add ourselves to the transaction passed in, but may start
462 * our own in order to shift windows.
464 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
466 struct ocfs2_alloc_context *ac)
469 struct ocfs2_dinode *alloc;
470 struct inode *local_alloc_inode;
471 unsigned int free_bits;
478 ocfs2_get_system_file_inode(osb,
479 LOCAL_ALLOC_SYSTEM_INODE,
481 if (!local_alloc_inode) {
487 mutex_lock(&local_alloc_inode->i_mutex);
489 if (osb->local_alloc_state != OCFS2_LA_ENABLED) {
494 if (bits_wanted > ocfs2_local_alloc_window_bits(osb)) {
495 mlog(0, "Asking for more than my max window size!\n");
500 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
502 #ifdef OCFS2_DEBUG_FS
503 if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
504 ocfs2_local_alloc_count_bits(alloc)) {
505 ocfs2_error(osb->sb, "local alloc inode %llu says it has "
506 "%u free bits, but a count shows %u",
507 (unsigned long long)le64_to_cpu(alloc->i_blkno),
508 le32_to_cpu(alloc->id1.bitmap1.i_used),
509 ocfs2_local_alloc_count_bits(alloc));
515 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
516 le32_to_cpu(alloc->id1.bitmap1.i_used);
517 if (bits_wanted > free_bits) {
518 /* uhoh, window change time. */
520 ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
522 if (status != -ENOSPC)
528 ac->ac_inode = local_alloc_inode;
529 ac->ac_which = OCFS2_AC_USE_LOCAL;
530 get_bh(osb->local_alloc_bh);
531 ac->ac_bh = osb->local_alloc_bh;
534 if (status < 0 && local_alloc_inode) {
535 mutex_unlock(&local_alloc_inode->i_mutex);
536 iput(local_alloc_inode);
539 mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted, osb->slot_num,
546 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
548 struct ocfs2_alloc_context *ac,
554 struct inode *local_alloc_inode;
556 struct ocfs2_dinode *alloc;
557 struct ocfs2_local_alloc *la;
560 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
562 local_alloc_inode = ac->ac_inode;
563 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
564 la = OCFS2_LOCAL_ALLOC(alloc);
566 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
568 /* TODO: Shouldn't we just BUG here? */
574 bitmap = la->la_bitmap;
575 *bit_off = le32_to_cpu(la->la_bm_off) + start;
576 /* local alloc is always contiguous by nature -- we never
577 * delete bits from it! */
578 *num_bits = bits_wanted;
580 status = ocfs2_journal_access(handle, local_alloc_inode,
582 OCFS2_JOURNAL_ACCESS_WRITE);
589 ocfs2_set_bit(start++, bitmap);
591 alloc->id1.bitmap1.i_used = cpu_to_le32(*num_bits +
592 le32_to_cpu(alloc->id1.bitmap1.i_used));
594 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
606 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
611 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
615 buffer = la->la_bitmap;
616 for (i = 0; i < le16_to_cpu(la->la_size); i++)
617 count += hweight8(buffer[i]);
623 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
624 struct ocfs2_dinode *alloc,
627 int numfound, bitoff, left, startoff, lastzero;
630 mlog_entry("(numbits wanted = %u)\n", numbits);
632 if (!alloc->id1.bitmap1.i_total) {
633 mlog(0, "No bits in my window!\n");
638 bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
640 numfound = bitoff = startoff = 0;
642 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
643 while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
644 if (bitoff == left) {
645 /* mlog(0, "bitoff (%d) == left", bitoff); */
648 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
649 "numfound = %d\n", bitoff, startoff, numfound);*/
651 /* Ok, we found a zero bit... is it contig. or do we
653 if (bitoff == startoff) {
654 /* we found a zero */
658 /* got a zero after some ones */
662 /* we got everything we needed */
663 if (numfound == numbits) {
664 /* mlog(0, "Found it all!\n"); */
669 mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff,
672 if (numfound == numbits)
673 bitoff = startoff - numfound;
682 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
684 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
688 alloc->id1.bitmap1.i_total = 0;
689 alloc->id1.bitmap1.i_used = 0;
691 for(i = 0; i < le16_to_cpu(la->la_size); i++)
692 la->la_bitmap[i] = 0;
698 /* turn this on and uncomment below to aid debugging window shifts. */
699 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
703 unsigned int tmp = count;
705 if (ocfs2_test_bit(start + tmp, bitmap)) {
706 printk("ocfs2_verify_zero_bits: start = %u, count = "
707 "%u\n", start, count);
708 printk("ocfs2_verify_zero_bits: bit %u is set!",
717 * sync the local alloc to main bitmap.
719 * assumes you've already locked the main bitmap -- the bitmap inode
720 * passed is used for caching.
722 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
724 struct ocfs2_dinode *alloc,
725 struct inode *main_bm_inode,
726 struct buffer_head *main_bm_bh)
729 int bit_off, left, count, start;
733 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
735 mlog_entry("total = %u, used = %u\n",
736 le32_to_cpu(alloc->id1.bitmap1.i_total),
737 le32_to_cpu(alloc->id1.bitmap1.i_used));
739 if (!alloc->id1.bitmap1.i_total) {
740 mlog(0, "nothing to sync!\n");
744 if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
745 le32_to_cpu(alloc->id1.bitmap1.i_total)) {
746 mlog(0, "all bits were taken!\n");
750 la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
751 le32_to_cpu(la->la_bm_off));
752 bitmap = la->la_bitmap;
753 start = count = bit_off = 0;
754 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
756 while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
758 if ((bit_off < left) && (bit_off == start)) {
764 blkno = la_start_blk +
765 ocfs2_clusters_to_blocks(osb->sb,
768 mlog(0, "freeing %u bits starting at local alloc bit "
769 "%u (la_start_blk = %llu, blkno = %llu)\n",
770 count, start - count,
771 (unsigned long long)la_start_blk,
772 (unsigned long long)blkno);
774 status = ocfs2_free_clusters(handle, main_bm_inode,
775 main_bm_bh, blkno, count);
792 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
793 struct ocfs2_alloc_context **ac,
794 struct inode **bitmap_inode,
795 struct buffer_head **bitmap_bh)
799 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
806 (*ac)->ac_bits_wanted = ocfs2_local_alloc_window_bits(osb);
808 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
810 if (status != -ENOSPC)
815 *bitmap_inode = (*ac)->ac_inode;
816 igrab(*bitmap_inode);
817 *bitmap_bh = (*ac)->ac_bh;
821 if ((status < 0) && *ac) {
822 ocfs2_free_alloc_context(*ac);
831 * pass it the bitmap lock in lock_bh if you have it.
833 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
835 struct ocfs2_alloc_context *ac)
838 u32 cluster_off, cluster_count;
839 struct ocfs2_dinode *alloc = NULL;
840 struct ocfs2_local_alloc *la;
844 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
845 la = OCFS2_LOCAL_ALLOC(alloc);
847 if (alloc->id1.bitmap1.i_total)
848 mlog(0, "asking me to alloc a new window over a non-empty "
851 mlog(0, "Allocating %u clusters for a new window.\n",
852 ocfs2_local_alloc_window_bits(osb));
854 /* Instruct the allocation code to try the most recently used
855 * cluster group. We'll re-record the group used this pass
857 ac->ac_last_group = osb->la_last_gd;
859 /* we used the generic suballoc reserve function, but we set
860 * everything up nicely, so there's no reason why we can't use
861 * the more specific cluster api to claim bits. */
862 status = ocfs2_claim_clusters(osb, handle, ac,
863 ocfs2_local_alloc_window_bits(osb),
864 &cluster_off, &cluster_count);
866 if (status != -ENOSPC)
871 osb->la_last_gd = ac->ac_last_group;
873 la->la_bm_off = cpu_to_le32(cluster_off);
874 alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
875 /* just in case... In the future when we find space ourselves,
876 * we don't have to get all contiguous -- but we'll have to
877 * set all previously used bits in bitmap and update
878 * la_bits_set before setting the bits in the main bitmap. */
879 alloc->id1.bitmap1.i_used = 0;
880 memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
881 le16_to_cpu(la->la_size));
883 mlog(0, "New window allocated:\n");
884 mlog(0, "window la_bm_off = %u\n",
885 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
886 mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total));
893 /* Note that we do *NOT* lock the local alloc inode here as
894 * it's been locked already for us. */
895 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
896 struct inode *local_alloc_inode)
899 struct buffer_head *main_bm_bh = NULL;
900 struct inode *main_bm_inode = NULL;
901 handle_t *handle = NULL;
902 struct ocfs2_dinode *alloc;
903 struct ocfs2_dinode *alloc_copy = NULL;
904 struct ocfs2_alloc_context *ac = NULL;
908 /* This will lock the main bitmap for us. */
909 status = ocfs2_local_alloc_reserve_for_window(osb,
914 if (status != -ENOSPC)
919 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
920 if (IS_ERR(handle)) {
921 status = PTR_ERR(handle);
927 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
929 /* We want to clear the local alloc before doing anything
930 * else, so that if we error later during this operation,
931 * local alloc shutdown won't try to double free main bitmap
932 * bits. Make a copy so the sync function knows which bits to
934 alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_KERNEL);
940 memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
942 status = ocfs2_journal_access(handle, local_alloc_inode,
944 OCFS2_JOURNAL_ACCESS_WRITE);
950 ocfs2_clear_local_alloc(alloc);
952 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
958 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
959 main_bm_inode, main_bm_bh);
965 status = ocfs2_local_alloc_new_window(osb, handle, ac);
967 if (status != -ENOSPC)
972 atomic_inc(&osb->alloc_stats.moves);
977 ocfs2_commit_trans(osb, handle);
989 ocfs2_free_alloc_context(ac);