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 u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
52 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
53 struct ocfs2_dinode *alloc,
56 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
58 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
60 struct ocfs2_dinode *alloc,
61 struct inode *main_bm_inode,
62 struct buffer_head *main_bm_bh);
64 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
65 struct ocfs2_alloc_context **ac,
66 struct inode **bitmap_inode,
67 struct buffer_head **bitmap_bh);
69 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
71 struct ocfs2_alloc_context *ac);
73 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
74 struct inode *local_alloc_inode);
77 * Tell us whether a given allocation should use the local alloc
78 * file. Otherwise, it has to go to the main bitmap.
80 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
82 int la_bits = osb->local_alloc_bits;
85 if (osb->local_alloc_state != OCFS2_LA_ENABLED)
88 /* la_bits should be at least twice the size (in clusters) of
89 * a new block group. We want to be sure block group
90 * allocations go through the local alloc, so allow an
91 * allocation to take up to half the bitmap. */
92 if (bits > (la_bits / 2))
97 mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
98 osb->local_alloc_state, (unsigned long long)bits, la_bits, ret);
102 int ocfs2_load_local_alloc(struct ocfs2_super *osb)
105 struct ocfs2_dinode *alloc = NULL;
106 struct buffer_head *alloc_bh = NULL;
108 struct inode *inode = NULL;
109 struct ocfs2_local_alloc *la;
113 if (osb->local_alloc_bits == 0)
116 if (osb->local_alloc_bits >= osb->bitmap_cpg) {
117 mlog(ML_NOTICE, "Requested local alloc window %d is larger "
118 "than max possible %u. Using defaults.\n",
119 osb->local_alloc_bits, (osb->bitmap_cpg - 1));
120 osb->local_alloc_bits =
121 ocfs2_megabytes_to_clusters(osb->sb,
122 OCFS2_DEFAULT_LOCAL_ALLOC_SIZE);
125 /* read the alloc off disk */
126 inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
134 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
135 &alloc_bh, 0, inode);
141 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
142 la = OCFS2_LOCAL_ALLOC(alloc);
144 if (!(le32_to_cpu(alloc->i_flags) &
145 (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
146 mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
147 (unsigned long long)OCFS2_I(inode)->ip_blkno);
152 if ((la->la_size == 0) ||
153 (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
154 mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
155 le16_to_cpu(la->la_size));
160 /* do a little verification. */
161 num_used = ocfs2_local_alloc_count_bits(alloc);
163 /* hopefully the local alloc has always been recovered before
166 || alloc->id1.bitmap1.i_used
167 || alloc->id1.bitmap1.i_total
169 mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
170 "found = %u, set = %u, taken = %u, off = %u\n",
171 num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
172 le32_to_cpu(alloc->id1.bitmap1.i_total),
173 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
175 osb->local_alloc_bh = alloc_bh;
176 osb->local_alloc_state = OCFS2_LA_ENABLED;
185 mlog(0, "Local alloc window bits = %d\n", osb->local_alloc_bits);
192 * return any unused bits to the bitmap and write out a clean
195 * local_alloc_bh is optional. If not passed, we will simply use the
196 * one off osb. If you do pass it however, be warned that it *will* be
197 * returned brelse'd and NULL'd out.*/
198 void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
202 struct inode *local_alloc_inode = NULL;
203 struct buffer_head *bh = NULL;
204 struct buffer_head *main_bm_bh = NULL;
205 struct inode *main_bm_inode = NULL;
206 struct ocfs2_dinode *alloc_copy = NULL;
207 struct ocfs2_dinode *alloc = NULL;
211 if (osb->local_alloc_state == OCFS2_LA_UNUSED)
215 ocfs2_get_system_file_inode(osb,
216 LOCAL_ALLOC_SYSTEM_INODE,
218 if (!local_alloc_inode) {
224 osb->local_alloc_state = OCFS2_LA_DISABLED;
226 main_bm_inode = ocfs2_get_system_file_inode(osb,
227 GLOBAL_BITMAP_SYSTEM_INODE,
229 if (!main_bm_inode) {
235 mutex_lock(&main_bm_inode->i_mutex);
237 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
243 /* WINDOW_MOVE_CREDITS is a bit heavy... */
244 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
245 if (IS_ERR(handle)) {
246 mlog_errno(PTR_ERR(handle));
251 bh = osb->local_alloc_bh;
252 alloc = (struct ocfs2_dinode *) bh->b_data;
254 alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
259 memcpy(alloc_copy, alloc, bh->b_size);
261 status = ocfs2_journal_access(handle, local_alloc_inode, bh,
262 OCFS2_JOURNAL_ACCESS_WRITE);
268 ocfs2_clear_local_alloc(alloc);
270 status = ocfs2_journal_dirty(handle, bh);
277 osb->local_alloc_bh = NULL;
278 osb->local_alloc_state = OCFS2_LA_UNUSED;
280 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
281 main_bm_inode, main_bm_bh);
286 ocfs2_commit_trans(osb, handle);
292 ocfs2_inode_unlock(main_bm_inode, 1);
295 mutex_unlock(&main_bm_inode->i_mutex);
299 if (local_alloc_inode)
300 iput(local_alloc_inode);
309 * We want to free the bitmap bits outside of any recovery context as
310 * we'll need a cluster lock to do so, but we must clear the local
311 * alloc before giving up the recovered nodes journal. To solve this,
312 * we kmalloc a copy of the local alloc before it's change for the
313 * caller to process with ocfs2_complete_local_alloc_recovery
315 int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
317 struct ocfs2_dinode **alloc_copy)
320 struct buffer_head *alloc_bh = NULL;
321 struct inode *inode = NULL;
322 struct ocfs2_dinode *alloc;
324 mlog_entry("(slot_num = %d)\n", slot_num);
328 inode = ocfs2_get_system_file_inode(osb,
329 LOCAL_ALLOC_SYSTEM_INODE,
337 mutex_lock(&inode->i_mutex);
339 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno,
340 &alloc_bh, 0, inode);
346 *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
347 if (!(*alloc_copy)) {
351 memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
353 alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
354 ocfs2_clear_local_alloc(alloc);
356 status = ocfs2_write_block(osb, alloc_bh, inode);
361 if ((status < 0) && (*alloc_copy)) {
370 mutex_unlock(&inode->i_mutex);
379 * Step 2: By now, we've completed the journal recovery, we've stamped
380 * a clean local alloc on disk and dropped the node out of the
381 * recovery map. Dlm locks will no longer stall, so lets clear out the
384 int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
385 struct ocfs2_dinode *alloc)
389 struct buffer_head *main_bm_bh = NULL;
390 struct inode *main_bm_inode;
394 main_bm_inode = ocfs2_get_system_file_inode(osb,
395 GLOBAL_BITMAP_SYSTEM_INODE,
397 if (!main_bm_inode) {
403 mutex_lock(&main_bm_inode->i_mutex);
405 status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
411 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
412 if (IS_ERR(handle)) {
413 status = PTR_ERR(handle);
419 /* we want the bitmap change to be recorded on disk asap */
422 status = ocfs2_sync_local_to_main(osb, handle, alloc,
423 main_bm_inode, main_bm_bh);
427 ocfs2_commit_trans(osb, handle);
430 ocfs2_inode_unlock(main_bm_inode, 1);
433 mutex_unlock(&main_bm_inode->i_mutex);
442 ocfs2_init_inode_steal_slot(osb);
448 * make sure we've got at least bitswanted contiguous bits in the
449 * local alloc. You lose them when you drop i_mutex.
451 * We will add ourselves to the transaction passed in, but may start
452 * our own in order to shift windows.
454 int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
456 struct ocfs2_alloc_context *ac)
459 struct ocfs2_dinode *alloc;
460 struct inode *local_alloc_inode;
461 unsigned int free_bits;
468 ocfs2_get_system_file_inode(osb,
469 LOCAL_ALLOC_SYSTEM_INODE,
471 if (!local_alloc_inode) {
477 mutex_lock(&local_alloc_inode->i_mutex);
479 if (osb->local_alloc_state != OCFS2_LA_ENABLED) {
484 if (bits_wanted > osb->local_alloc_bits) {
485 mlog(0, "Asking for more than my max window size!\n");
490 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
492 #ifdef CONFIG_OCFS2_DEBUG_FS
493 if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
494 ocfs2_local_alloc_count_bits(alloc)) {
495 ocfs2_error(osb->sb, "local alloc inode %llu says it has "
496 "%u free bits, but a count shows %u",
497 (unsigned long long)le64_to_cpu(alloc->i_blkno),
498 le32_to_cpu(alloc->id1.bitmap1.i_used),
499 ocfs2_local_alloc_count_bits(alloc));
505 free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
506 le32_to_cpu(alloc->id1.bitmap1.i_used);
507 if (bits_wanted > free_bits) {
508 /* uhoh, window change time. */
510 ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
512 if (status != -ENOSPC)
518 ac->ac_inode = local_alloc_inode;
519 /* We should never use localalloc from another slot */
520 ac->ac_alloc_slot = osb->slot_num;
521 ac->ac_which = OCFS2_AC_USE_LOCAL;
522 get_bh(osb->local_alloc_bh);
523 ac->ac_bh = osb->local_alloc_bh;
526 if (status < 0 && local_alloc_inode) {
527 mutex_unlock(&local_alloc_inode->i_mutex);
528 iput(local_alloc_inode);
531 mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted, osb->slot_num,
538 int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
540 struct ocfs2_alloc_context *ac,
546 struct inode *local_alloc_inode;
548 struct ocfs2_dinode *alloc;
549 struct ocfs2_local_alloc *la;
552 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
554 local_alloc_inode = ac->ac_inode;
555 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
556 la = OCFS2_LOCAL_ALLOC(alloc);
558 start = ocfs2_local_alloc_find_clear_bits(osb, alloc, bits_wanted);
560 /* TODO: Shouldn't we just BUG here? */
566 bitmap = la->la_bitmap;
567 *bit_off = le32_to_cpu(la->la_bm_off) + start;
568 /* local alloc is always contiguous by nature -- we never
569 * delete bits from it! */
570 *num_bits = bits_wanted;
572 status = ocfs2_journal_access(handle, local_alloc_inode,
574 OCFS2_JOURNAL_ACCESS_WRITE);
581 ocfs2_set_bit(start++, bitmap);
583 le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
585 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
597 static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
602 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
606 buffer = la->la_bitmap;
607 for (i = 0; i < le16_to_cpu(la->la_size); i++)
608 count += hweight8(buffer[i]);
614 static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
615 struct ocfs2_dinode *alloc,
618 int numfound, bitoff, left, startoff, lastzero;
621 mlog_entry("(numbits wanted = %u)\n", numbits);
623 if (!alloc->id1.bitmap1.i_total) {
624 mlog(0, "No bits in my window!\n");
629 bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
631 numfound = bitoff = startoff = 0;
633 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
634 while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
635 if (bitoff == left) {
636 /* mlog(0, "bitoff (%d) == left", bitoff); */
639 /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
640 "numfound = %d\n", bitoff, startoff, numfound);*/
642 /* Ok, we found a zero bit... is it contig. or do we
644 if (bitoff == startoff) {
645 /* we found a zero */
649 /* got a zero after some ones */
653 /* we got everything we needed */
654 if (numfound == numbits) {
655 /* mlog(0, "Found it all!\n"); */
660 mlog(0, "Exiting loop, bitoff = %d, numfound = %d\n", bitoff,
663 if (numfound == numbits)
664 bitoff = startoff - numfound;
673 static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
675 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
679 alloc->id1.bitmap1.i_total = 0;
680 alloc->id1.bitmap1.i_used = 0;
682 for(i = 0; i < le16_to_cpu(la->la_size); i++)
683 la->la_bitmap[i] = 0;
689 /* turn this on and uncomment below to aid debugging window shifts. */
690 static void ocfs2_verify_zero_bits(unsigned long *bitmap,
694 unsigned int tmp = count;
696 if (ocfs2_test_bit(start + tmp, bitmap)) {
697 printk("ocfs2_verify_zero_bits: start = %u, count = "
698 "%u\n", start, count);
699 printk("ocfs2_verify_zero_bits: bit %u is set!",
708 * sync the local alloc to main bitmap.
710 * assumes you've already locked the main bitmap -- the bitmap inode
711 * passed is used for caching.
713 static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
715 struct ocfs2_dinode *alloc,
716 struct inode *main_bm_inode,
717 struct buffer_head *main_bm_bh)
720 int bit_off, left, count, start;
724 struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
726 mlog_entry("total = %u, used = %u\n",
727 le32_to_cpu(alloc->id1.bitmap1.i_total),
728 le32_to_cpu(alloc->id1.bitmap1.i_used));
730 if (!alloc->id1.bitmap1.i_total) {
731 mlog(0, "nothing to sync!\n");
735 if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
736 le32_to_cpu(alloc->id1.bitmap1.i_total)) {
737 mlog(0, "all bits were taken!\n");
741 la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
742 le32_to_cpu(la->la_bm_off));
743 bitmap = la->la_bitmap;
744 start = count = bit_off = 0;
745 left = le32_to_cpu(alloc->id1.bitmap1.i_total);
747 while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
749 if ((bit_off < left) && (bit_off == start)) {
755 blkno = la_start_blk +
756 ocfs2_clusters_to_blocks(osb->sb,
759 mlog(0, "freeing %u bits starting at local alloc bit "
760 "%u (la_start_blk = %llu, blkno = %llu)\n",
761 count, start - count,
762 (unsigned long long)la_start_blk,
763 (unsigned long long)blkno);
765 status = ocfs2_free_clusters(handle, main_bm_inode,
766 main_bm_bh, blkno, count);
783 static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
784 struct ocfs2_alloc_context **ac,
785 struct inode **bitmap_inode,
786 struct buffer_head **bitmap_bh)
790 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
797 (*ac)->ac_bits_wanted = osb->local_alloc_bits;
799 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
801 if (status != -ENOSPC)
806 *bitmap_inode = (*ac)->ac_inode;
807 igrab(*bitmap_inode);
808 *bitmap_bh = (*ac)->ac_bh;
812 if ((status < 0) && *ac) {
813 ocfs2_free_alloc_context(*ac);
822 * pass it the bitmap lock in lock_bh if you have it.
824 static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
826 struct ocfs2_alloc_context *ac)
829 u32 cluster_off, cluster_count;
830 struct ocfs2_dinode *alloc = NULL;
831 struct ocfs2_local_alloc *la;
835 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
836 la = OCFS2_LOCAL_ALLOC(alloc);
838 if (alloc->id1.bitmap1.i_total)
839 mlog(0, "asking me to alloc a new window over a non-empty "
842 mlog(0, "Allocating %u clusters for a new window.\n",
843 osb->local_alloc_bits);
845 /* Instruct the allocation code to try the most recently used
846 * cluster group. We'll re-record the group used this pass
848 ac->ac_last_group = osb->la_last_gd;
850 /* we used the generic suballoc reserve function, but we set
851 * everything up nicely, so there's no reason why we can't use
852 * the more specific cluster api to claim bits. */
853 status = ocfs2_claim_clusters(osb, handle, ac, osb->local_alloc_bits,
854 &cluster_off, &cluster_count);
856 if (status != -ENOSPC)
861 osb->la_last_gd = ac->ac_last_group;
863 la->la_bm_off = cpu_to_le32(cluster_off);
864 alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
865 /* just in case... In the future when we find space ourselves,
866 * we don't have to get all contiguous -- but we'll have to
867 * set all previously used bits in bitmap and update
868 * la_bits_set before setting the bits in the main bitmap. */
869 alloc->id1.bitmap1.i_used = 0;
870 memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
871 le16_to_cpu(la->la_size));
873 mlog(0, "New window allocated:\n");
874 mlog(0, "window la_bm_off = %u\n",
875 OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
876 mlog(0, "window bits = %u\n", le32_to_cpu(alloc->id1.bitmap1.i_total));
883 /* Note that we do *NOT* lock the local alloc inode here as
884 * it's been locked already for us. */
885 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
886 struct inode *local_alloc_inode)
889 struct buffer_head *main_bm_bh = NULL;
890 struct inode *main_bm_inode = NULL;
891 handle_t *handle = NULL;
892 struct ocfs2_dinode *alloc;
893 struct ocfs2_dinode *alloc_copy = NULL;
894 struct ocfs2_alloc_context *ac = NULL;
898 /* This will lock the main bitmap for us. */
899 status = ocfs2_local_alloc_reserve_for_window(osb,
904 if (status != -ENOSPC)
909 handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
910 if (IS_ERR(handle)) {
911 status = PTR_ERR(handle);
917 alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
919 /* We want to clear the local alloc before doing anything
920 * else, so that if we error later during this operation,
921 * local alloc shutdown won't try to double free main bitmap
922 * bits. Make a copy so the sync function knows which bits to
924 alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
930 memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
932 status = ocfs2_journal_access(handle, local_alloc_inode,
934 OCFS2_JOURNAL_ACCESS_WRITE);
940 ocfs2_clear_local_alloc(alloc);
942 status = ocfs2_journal_dirty(handle, osb->local_alloc_bh);
948 status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
949 main_bm_inode, main_bm_bh);
955 status = ocfs2_local_alloc_new_window(osb, handle, ac);
957 if (status != -ENOSPC)
962 atomic_inc(&osb->alloc_stats.moves);
967 ocfs2_commit_trans(osb, handle);
979 ocfs2_free_alloc_context(ac);