1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * metadata alloc and free
7 * Inspired by ext3 block groups.
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public
22 * License along with this program; if not, write to the
23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 021110-1307, USA.
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
41 #include "localalloc.h"
47 #include "buffer_head_io.h"
49 #define NOT_ALLOC_NEW_GROUP 0
50 #define ALLOC_NEW_GROUP 1
52 #define OCFS2_MAX_INODES_TO_STEAL 1024
54 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
55 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
56 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
57 static int ocfs2_block_group_fill(handle_t *handle,
58 struct inode *alloc_inode,
59 struct buffer_head *bg_bh,
62 struct ocfs2_chain_list *cl);
63 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
64 struct inode *alloc_inode,
65 struct buffer_head *bh,
68 static int ocfs2_cluster_group_search(struct inode *inode,
69 struct buffer_head *group_bh,
70 u32 bits_wanted, u32 min_bits,
72 u16 *bit_off, u16 *bits_found);
73 static int ocfs2_block_group_search(struct inode *inode,
74 struct buffer_head *group_bh,
75 u32 bits_wanted, u32 min_bits,
77 u16 *bit_off, u16 *bits_found);
78 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
79 struct ocfs2_alloc_context *ac,
84 unsigned int *num_bits,
86 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
88 static inline int ocfs2_block_group_set_bits(handle_t *handle,
89 struct inode *alloc_inode,
90 struct ocfs2_group_desc *bg,
91 struct buffer_head *group_bh,
93 unsigned int num_bits);
94 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
95 struct inode *alloc_inode,
96 struct ocfs2_group_desc *bg,
97 struct buffer_head *group_bh,
99 unsigned int num_bits);
101 static int ocfs2_relink_block_group(handle_t *handle,
102 struct inode *alloc_inode,
103 struct buffer_head *fe_bh,
104 struct buffer_head *bg_bh,
105 struct buffer_head *prev_bg_bh,
107 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
109 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
112 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
116 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
117 u32 bits_wanted, u64 max_block,
118 struct ocfs2_alloc_context **ac);
120 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
122 struct inode *inode = ac->ac_inode;
125 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
126 ocfs2_inode_unlock(inode, 1);
128 mutex_unlock(&inode->i_mutex);
137 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
139 ocfs2_free_ac_resource(ac);
143 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
145 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
148 int ocfs2_validate_group_descriptor(struct super_block *sb,
149 struct ocfs2_dinode *di,
150 struct buffer_head *bh,
153 unsigned int max_bits;
154 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
156 #define do_error(fmt, ...) \
159 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
161 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
164 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
165 do_error("Group descriptor #%llu has bad signature %.*s",
166 (unsigned long long)bh->b_blocknr, 7,
171 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
172 do_error("Group descriptor #%llu has an invalid bg_blkno "
174 (unsigned long long)bh->b_blocknr,
175 (unsigned long long)le64_to_cpu(gd->bg_blkno));
179 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
180 do_error("Group descriptor #%llu has an invalid "
181 "fs_generation of #%u",
182 (unsigned long long)bh->b_blocknr,
183 le32_to_cpu(gd->bg_generation));
187 if (di->i_blkno != gd->bg_parent_dinode) {
188 do_error("Group descriptor #%llu has bad parent "
189 "pointer (%llu, expected %llu)",
190 (unsigned long long)bh->b_blocknr,
191 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
192 (unsigned long long)le64_to_cpu(di->i_blkno));
196 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
197 if (le16_to_cpu(gd->bg_bits) > max_bits) {
198 do_error("Group descriptor #%llu has bit count of %u",
199 (unsigned long long)bh->b_blocknr,
200 le16_to_cpu(gd->bg_bits));
204 if (le16_to_cpu(gd->bg_chain) >=
205 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
206 do_error("Group descriptor #%llu has bad chain %u",
207 (unsigned long long)bh->b_blocknr,
208 le16_to_cpu(gd->bg_chain));
212 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
213 do_error("Group descriptor #%llu has bit count %u but "
214 "claims that %u are free",
215 (unsigned long long)bh->b_blocknr,
216 le16_to_cpu(gd->bg_bits),
217 le16_to_cpu(gd->bg_free_bits_count));
221 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
222 do_error("Group descriptor #%llu has bit count %u but "
223 "max bitmap bits of %u",
224 (unsigned long long)bh->b_blocknr,
225 le16_to_cpu(gd->bg_bits),
226 8 * le16_to_cpu(gd->bg_size));
234 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
235 u64 gd_blkno, struct buffer_head **bh)
238 struct buffer_head *tmp = *bh;
240 rc = ocfs2_read_block(inode, gd_blkno, &tmp);
244 rc = ocfs2_validate_group_descriptor(inode->i_sb, di, tmp, 0);
250 /* If ocfs2_read_block() got us a new bh, pass it up. */
258 static int ocfs2_block_group_fill(handle_t *handle,
259 struct inode *alloc_inode,
260 struct buffer_head *bg_bh,
263 struct ocfs2_chain_list *cl)
266 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
267 struct super_block * sb = alloc_inode->i_sb;
271 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
272 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
274 (unsigned long long)group_blkno,
275 (unsigned long long) bg_bh->b_blocknr);
280 status = ocfs2_journal_access(handle,
283 OCFS2_JOURNAL_ACCESS_CREATE);
289 memset(bg, 0, sb->s_blocksize);
290 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
291 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
292 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
293 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
294 bg->bg_chain = cpu_to_le16(my_chain);
295 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
296 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
297 bg->bg_blkno = cpu_to_le64(group_blkno);
298 /* set the 1st bit in the bitmap to account for the descriptor block */
299 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
300 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
302 status = ocfs2_journal_dirty(handle, bg_bh);
306 /* There is no need to zero out or otherwise initialize the
307 * other blocks in a group - All valid FS metadata in a block
308 * group stores the superblock fs_generation value at
309 * allocation time. */
316 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
321 while (curr < le16_to_cpu(cl->cl_count)) {
322 if (le32_to_cpu(cl->cl_recs[best].c_total) >
323 le32_to_cpu(cl->cl_recs[curr].c_total))
331 * We expect the block group allocator to already be locked.
333 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
334 struct inode *alloc_inode,
335 struct buffer_head *bh,
339 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
340 struct ocfs2_chain_list *cl;
341 struct ocfs2_alloc_context *ac = NULL;
342 handle_t *handle = NULL;
343 u32 bit_off, num_bits;
346 struct buffer_head *bg_bh = NULL;
347 struct ocfs2_group_desc *bg;
349 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
353 cl = &fe->id2.i_chain;
354 status = ocfs2_reserve_clusters_with_limit(osb,
355 le16_to_cpu(cl->cl_cpg),
358 if (status != -ENOSPC)
363 credits = ocfs2_calc_group_alloc_credits(osb->sb,
364 le16_to_cpu(cl->cl_cpg));
365 handle = ocfs2_start_trans(osb, credits);
366 if (IS_ERR(handle)) {
367 status = PTR_ERR(handle);
373 status = ocfs2_claim_clusters(osb,
376 le16_to_cpu(cl->cl_cpg),
380 if (status != -ENOSPC)
385 alloc_rec = ocfs2_find_smallest_chain(cl);
387 /* setup the group */
388 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
389 mlog(0, "new descriptor, record %u, at block %llu\n",
390 alloc_rec, (unsigned long long)bg_blkno);
392 bg_bh = sb_getblk(osb->sb, bg_blkno);
398 ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
400 status = ocfs2_block_group_fill(handle,
411 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
413 status = ocfs2_journal_access(handle, alloc_inode,
414 bh, OCFS2_JOURNAL_ACCESS_WRITE);
420 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
421 le16_to_cpu(bg->bg_free_bits_count));
422 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
423 cl->cl_recs[alloc_rec].c_blkno = cpu_to_le64(bg_blkno);
424 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
425 le16_add_cpu(&cl->cl_next_free_rec, 1);
427 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
428 le16_to_cpu(bg->bg_free_bits_count));
429 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
430 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
432 status = ocfs2_journal_dirty(handle, bh);
438 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
439 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
440 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
441 le32_to_cpu(fe->i_clusters)));
442 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
443 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
444 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
449 ocfs2_commit_trans(osb, handle);
452 ocfs2_free_alloc_context(ac);
460 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
461 struct ocfs2_alloc_context *ac,
467 u32 bits_wanted = ac->ac_bits_wanted;
468 struct inode *alloc_inode;
469 struct buffer_head *bh = NULL;
470 struct ocfs2_dinode *fe;
475 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
481 mutex_lock(&alloc_inode->i_mutex);
483 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
485 mutex_unlock(&alloc_inode->i_mutex);
492 ac->ac_inode = alloc_inode;
493 ac->ac_alloc_slot = slot;
495 fe = (struct ocfs2_dinode *) bh->b_data;
497 /* The bh was validated by the inode read inside
498 * ocfs2_inode_lock(). Any corruption is a code bug. */
499 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
501 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
502 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
503 (unsigned long long)le64_to_cpu(fe->i_blkno));
508 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
509 le32_to_cpu(fe->id1.bitmap1.i_used);
511 if (bits_wanted > free_bits) {
512 /* cluster bitmap never grows */
513 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
514 mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
515 bits_wanted, free_bits);
520 if (alloc_new_group != ALLOC_NEW_GROUP) {
521 mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
522 "and we don't alloc a new group for it.\n",
523 slot, bits_wanted, free_bits);
528 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
531 if (status != -ENOSPC)
535 atomic_inc(&osb->alloc_stats.bg_extends);
537 /* You should never ask for this much metadata */
539 (le32_to_cpu(fe->id1.bitmap1.i_total)
540 - le32_to_cpu(fe->id1.bitmap1.i_used)));
552 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
554 struct ocfs2_alloc_context **ac)
559 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
566 (*ac)->ac_bits_wanted = blocks;
567 (*ac)->ac_which = OCFS2_AC_USE_META;
568 slot = osb->slot_num;
569 (*ac)->ac_group_search = ocfs2_block_group_search;
571 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
572 EXTENT_ALLOC_SYSTEM_INODE,
573 slot, ALLOC_NEW_GROUP);
575 if (status != -ENOSPC)
582 if ((status < 0) && *ac) {
583 ocfs2_free_alloc_context(*ac);
591 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
592 struct ocfs2_extent_list *root_el,
593 struct ocfs2_alloc_context **ac)
595 return ocfs2_reserve_new_metadata_blocks(osb,
596 ocfs2_extend_meta_needed(root_el),
600 static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
601 struct ocfs2_alloc_context *ac)
603 int i, status = -ENOSPC;
604 s16 slot = ocfs2_get_inode_steal_slot(osb);
606 /* Start to steal inodes from the first slot after ours. */
607 if (slot == OCFS2_INVALID_SLOT)
608 slot = osb->slot_num + 1;
610 for (i = 0; i < osb->max_slots; i++, slot++) {
611 if (slot == osb->max_slots)
614 if (slot == osb->slot_num)
617 status = ocfs2_reserve_suballoc_bits(osb, ac,
618 INODE_ALLOC_SYSTEM_INODE,
619 slot, NOT_ALLOC_NEW_GROUP);
621 ocfs2_set_inode_steal_slot(osb, slot);
625 ocfs2_free_ac_resource(ac);
631 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
632 struct ocfs2_alloc_context **ac)
635 s16 slot = ocfs2_get_inode_steal_slot(osb);
637 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
644 (*ac)->ac_bits_wanted = 1;
645 (*ac)->ac_which = OCFS2_AC_USE_INODE;
647 (*ac)->ac_group_search = ocfs2_block_group_search;
650 * stat(2) can't handle i_ino > 32bits, so we tell the
651 * lower levels not to allocate us a block group past that
652 * limit. The 'inode64' mount option avoids this behavior.
654 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
655 (*ac)->ac_max_block = (u32)~0U;
658 * slot is set when we successfully steal inode from other nodes.
659 * It is reset in 3 places:
660 * 1. when we flush the truncate log
661 * 2. when we complete local alloc recovery.
662 * 3. when we successfully allocate from our own slot.
663 * After it is set, we will go on stealing inodes until we find the
664 * need to check our slots to see whether there is some space for us.
666 if (slot != OCFS2_INVALID_SLOT &&
667 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
670 atomic_set(&osb->s_num_inodes_stolen, 0);
671 status = ocfs2_reserve_suballoc_bits(osb, *ac,
672 INODE_ALLOC_SYSTEM_INODE,
673 osb->slot_num, ALLOC_NEW_GROUP);
678 * Some inodes must be freed by us, so try to allocate
679 * from our own next time.
681 if (slot != OCFS2_INVALID_SLOT)
682 ocfs2_init_inode_steal_slot(osb);
684 } else if (status < 0 && status != -ENOSPC) {
689 ocfs2_free_ac_resource(*ac);
692 status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
693 atomic_inc(&osb->s_num_inodes_stolen);
695 if (status != -ENOSPC)
702 if ((status < 0) && *ac) {
703 ocfs2_free_alloc_context(*ac);
711 /* local alloc code has to do the same thing, so rather than do this
713 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
714 struct ocfs2_alloc_context *ac)
718 ac->ac_which = OCFS2_AC_USE_MAIN;
719 ac->ac_group_search = ocfs2_cluster_group_search;
721 status = ocfs2_reserve_suballoc_bits(osb, ac,
722 GLOBAL_BITMAP_SYSTEM_INODE,
725 if (status < 0 && status != -ENOSPC) {
734 /* Callers don't need to care which bitmap (local alloc or main) to
735 * use so we figure it out for them, but unfortunately this clutters
737 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
738 u32 bits_wanted, u64 max_block,
739 struct ocfs2_alloc_context **ac)
745 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
752 (*ac)->ac_bits_wanted = bits_wanted;
753 (*ac)->ac_max_block = max_block;
756 if (ocfs2_alloc_should_use_local(osb, bits_wanted)) {
757 status = ocfs2_reserve_local_alloc_bits(osb,
760 if (status == -EFBIG) {
761 /* The local alloc window is outside ac_max_block.
762 * use the main bitmap. */
764 } else if ((status < 0) && (status != -ENOSPC)) {
770 if (status == -ENOSPC) {
771 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
773 if (status != -ENOSPC)
781 if ((status < 0) && *ac) {
782 ocfs2_free_alloc_context(*ac);
790 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
792 struct ocfs2_alloc_context **ac)
794 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0, ac);
798 * More or less lifted from ext3. I'll leave their description below:
800 * "For ext3 allocations, we must not reuse any blocks which are
801 * allocated in the bitmap buffer's "last committed data" copy. This
802 * prevents deletes from freeing up the page for reuse until we have
803 * committed the delete transaction.
805 * If we didn't do this, then deleting something and reallocating it as
806 * data would allow the old block to be overwritten before the
807 * transaction committed (because we force data to disk before commit).
808 * This would lead to corruption if we crashed between overwriting the
809 * data and committing the delete.
811 * @@@ We may want to make this allocation behaviour conditional on
812 * data-writes at some point, and disable it for metadata allocations or
815 * Note: OCFS2 already does this differently for metadata vs data
816 * allocations, as those bitmaps are separate and undo access is never
817 * called on a metadata group descriptor.
819 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
822 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
824 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
826 if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
829 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
830 return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
833 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
834 struct buffer_head *bg_bh,
835 unsigned int bits_wanted,
836 unsigned int total_bits,
841 u16 best_offset, best_size;
842 int offset, start, found, status = 0;
843 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
845 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
846 OCFS2_RO_ON_INVALID_GROUP_DESC(osb->sb, bg);
850 found = start = best_offset = best_size = 0;
851 bitmap = bg->bg_bitmap;
853 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
854 if (offset == total_bits)
857 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
858 /* We found a zero, but we can't use it as it
859 * hasn't been put to disk yet! */
862 } else if (offset == start) {
863 /* we found a zero */
865 /* move start to the next bit to test */
868 /* got a zero after some ones */
872 if (found > best_size) {
874 best_offset = start - found;
876 /* we got everything we needed */
877 if (found == bits_wanted) {
878 /* mlog(0, "Found it all!\n"); */
883 /* XXX: I think the first clause is equivalent to the second
885 if (found == bits_wanted) {
886 *bit_off = start - found;
888 } else if (best_size) {
889 *bit_off = best_offset;
890 *bits_found = best_size;
893 /* No error log here -- see the comment above
894 * ocfs2_test_bg_bit_allocatable */
900 static inline int ocfs2_block_group_set_bits(handle_t *handle,
901 struct inode *alloc_inode,
902 struct ocfs2_group_desc *bg,
903 struct buffer_head *group_bh,
904 unsigned int bit_off,
905 unsigned int num_bits)
908 void *bitmap = bg->bg_bitmap;
909 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
913 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
914 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
918 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
920 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
923 if (ocfs2_is_cluster_bitmap(alloc_inode))
924 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
926 status = ocfs2_journal_access(handle,
935 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
938 ocfs2_set_bit(bit_off++, bitmap);
940 status = ocfs2_journal_dirty(handle,
952 /* find the one with the most empty bits */
953 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
957 BUG_ON(!cl->cl_next_free_rec);
960 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
961 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
962 le32_to_cpu(cl->cl_recs[best].c_free))
967 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
971 static int ocfs2_relink_block_group(handle_t *handle,
972 struct inode *alloc_inode,
973 struct buffer_head *fe_bh,
974 struct buffer_head *bg_bh,
975 struct buffer_head *prev_bg_bh,
979 /* there is a really tiny chance the journal calls could fail,
980 * but we wouldn't want inconsistent blocks in *any* case. */
981 u64 fe_ptr, bg_ptr, prev_bg_ptr;
982 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
983 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
984 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
986 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
987 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
991 if (!OCFS2_IS_VALID_GROUP_DESC(prev_bg)) {
992 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, prev_bg);
997 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
998 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
999 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1000 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1002 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1003 bg_ptr = le64_to_cpu(bg->bg_next_group);
1004 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1006 status = ocfs2_journal_access(handle, alloc_inode, prev_bg_bh,
1007 OCFS2_JOURNAL_ACCESS_WRITE);
1013 prev_bg->bg_next_group = bg->bg_next_group;
1015 status = ocfs2_journal_dirty(handle, prev_bg_bh);
1021 status = ocfs2_journal_access(handle, alloc_inode, bg_bh,
1022 OCFS2_JOURNAL_ACCESS_WRITE);
1028 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1030 status = ocfs2_journal_dirty(handle, bg_bh);
1036 status = ocfs2_journal_access(handle, alloc_inode, fe_bh,
1037 OCFS2_JOURNAL_ACCESS_WRITE);
1043 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1045 status = ocfs2_journal_dirty(handle, fe_bh);
1054 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1055 bg->bg_next_group = cpu_to_le64(bg_ptr);
1056 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1063 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1066 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1069 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1070 * value on error. */
1071 static int ocfs2_cluster_group_search(struct inode *inode,
1072 struct buffer_head *group_bh,
1073 u32 bits_wanted, u32 min_bits,
1075 u16 *bit_off, u16 *bits_found)
1077 int search = -ENOSPC;
1080 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1081 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1082 u16 tmp_off, tmp_found;
1083 unsigned int max_bits, gd_cluster_off;
1085 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1087 if (gd->bg_free_bits_count) {
1088 max_bits = le16_to_cpu(gd->bg_bits);
1090 /* Tail groups in cluster bitmaps which aren't cpg
1091 * aligned are prone to partial extention by a failed
1092 * fs resize. If the file system resize never got to
1093 * update the dinode cluster count, then we don't want
1094 * to trust any clusters past it, regardless of what
1095 * the group descriptor says. */
1096 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1097 le64_to_cpu(gd->bg_blkno));
1098 if ((gd_cluster_off + max_bits) >
1099 OCFS2_I(inode)->ip_clusters) {
1100 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1101 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1102 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1103 le16_to_cpu(gd->bg_bits),
1104 OCFS2_I(inode)->ip_clusters, max_bits);
1107 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1108 group_bh, bits_wanted,
1110 &tmp_off, &tmp_found);
1115 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1117 tmp_off + tmp_found);
1118 mlog(0, "Checking %llu against %llu\n",
1119 (unsigned long long)blkoff,
1120 (unsigned long long)max_block);
1121 if (blkoff > max_block)
1125 /* ocfs2_block_group_find_clear_bits() might
1126 * return success, but we still want to return
1127 * -ENOSPC unless it found the minimum number
1129 if (min_bits <= tmp_found) {
1131 *bits_found = tmp_found;
1132 search = 0; /* success */
1133 } else if (tmp_found) {
1135 * Don't show bits which we'll be returning
1136 * for allocation to the local alloc bitmap.
1138 ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1145 static int ocfs2_block_group_search(struct inode *inode,
1146 struct buffer_head *group_bh,
1147 u32 bits_wanted, u32 min_bits,
1149 u16 *bit_off, u16 *bits_found)
1153 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1155 BUG_ON(min_bits != 1);
1156 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1158 if (bg->bg_free_bits_count) {
1159 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1160 group_bh, bits_wanted,
1161 le16_to_cpu(bg->bg_bits),
1162 bit_off, bits_found);
1163 if (!ret && max_block) {
1164 blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1166 mlog(0, "Checking %llu against %llu\n",
1167 (unsigned long long)blkoff,
1168 (unsigned long long)max_block);
1169 if (blkoff > max_block)
1177 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1179 struct buffer_head *di_bh,
1185 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1186 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1188 ret = ocfs2_journal_access(handle, inode, di_bh,
1189 OCFS2_JOURNAL_ACCESS_WRITE);
1195 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1196 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1197 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1199 ret = ocfs2_journal_dirty(handle, di_bh);
1207 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1212 unsigned int *num_bits,
1218 struct buffer_head *group_bh = NULL;
1219 struct ocfs2_group_desc *gd;
1220 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1221 struct inode *alloc_inode = ac->ac_inode;
1223 ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1230 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1231 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1232 ac->ac_max_block, bit_off, &found);
1241 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1243 le16_to_cpu(gd->bg_chain));
1249 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1250 *bit_off, *num_bits);
1254 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1262 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1267 unsigned int *num_bits,
1272 u16 chain, tmp_bits;
1275 struct inode *alloc_inode = ac->ac_inode;
1276 struct buffer_head *group_bh = NULL;
1277 struct buffer_head *prev_group_bh = NULL;
1278 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1279 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1280 struct ocfs2_group_desc *bg;
1282 chain = ac->ac_chain;
1283 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1285 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1287 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1288 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1294 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1297 /* for now, the chain search is a bit simplistic. We just use
1298 * the 1st group with any empty bits. */
1299 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1300 bits_wanted, min_bits,
1301 ac->ac_max_block, bit_off,
1302 &tmp_bits)) == -ENOSPC) {
1303 if (!bg->bg_next_group)
1306 brelse(prev_group_bh);
1307 prev_group_bh = NULL;
1309 next_group = le64_to_cpu(bg->bg_next_group);
1310 prev_group_bh = group_bh;
1312 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1313 next_group, &group_bh);
1318 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1321 if (status != -ENOSPC)
1326 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1327 tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1329 *num_bits = tmp_bits;
1331 BUG_ON(*num_bits == 0);
1334 * Keep track of previous block descriptor read. When
1335 * we find a target, if we have read more than X
1336 * number of descriptors, and the target is reasonably
1337 * empty, relink him to top of his chain.
1339 * We've read 0 extra blocks and only send one more to
1340 * the transaction, yet the next guy to search has a
1343 * Do this *after* figuring out how many bits we're taking out
1344 * of our target group.
1346 if (ac->ac_allow_chain_relink &&
1348 (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1349 status = ocfs2_relink_block_group(handle, alloc_inode,
1350 ac->ac_bh, group_bh,
1351 prev_group_bh, chain);
1358 /* Ok, claim our bits now: set the info on dinode, chainlist
1359 * and then the group */
1360 status = ocfs2_journal_access(handle,
1363 OCFS2_JOURNAL_ACCESS_WRITE);
1369 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1370 fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1371 le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1373 status = ocfs2_journal_dirty(handle,
1380 status = ocfs2_block_group_set_bits(handle,
1391 mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1392 (unsigned long long)le64_to_cpu(fe->i_blkno));
1394 *bg_blkno = le64_to_cpu(bg->bg_blkno);
1395 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1398 brelse(prev_group_bh);
1404 /* will give out up to bits_wanted contiguous bits. */
1405 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1406 struct ocfs2_alloc_context *ac,
1411 unsigned int *num_bits,
1417 u64 hint_blkno = ac->ac_last_group;
1418 struct ocfs2_chain_list *cl;
1419 struct ocfs2_dinode *fe;
1423 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1424 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1427 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1429 /* The bh was validated by the inode read during
1430 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1431 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1433 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1434 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1435 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1436 "bits but only %u total.",
1437 (unsigned long long)le64_to_cpu(fe->i_blkno),
1438 le32_to_cpu(fe->id1.bitmap1.i_used),
1439 le32_to_cpu(fe->id1.bitmap1.i_total));
1445 /* Attempt to short-circuit the usual search mechanism
1446 * by jumping straight to the most recently used
1447 * allocation group. This helps us mantain some
1448 * contiguousness across allocations. */
1449 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1450 min_bits, bit_off, num_bits,
1451 hint_blkno, &bits_left);
1453 /* Be careful to update *bg_blkno here as the
1454 * caller is expecting it to be filled in, and
1455 * ocfs2_search_one_group() won't do that for
1457 *bg_blkno = hint_blkno;
1460 if (status < 0 && status != -ENOSPC) {
1466 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1468 victim = ocfs2_find_victim_chain(cl);
1469 ac->ac_chain = victim;
1470 ac->ac_allow_chain_relink = 1;
1472 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1473 num_bits, bg_blkno, &bits_left);
1476 if (status < 0 && status != -ENOSPC) {
1481 mlog(0, "Search of victim chain %u came up with nothing, "
1482 "trying all chains now.\n", victim);
1484 /* If we didn't pick a good victim, then just default to
1485 * searching each chain in order. Don't allow chain relinking
1486 * because we only calculate enough journal credits for one
1487 * relink per alloc. */
1488 ac->ac_allow_chain_relink = 0;
1489 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1492 if (!cl->cl_recs[i].c_free)
1496 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1497 bit_off, num_bits, bg_blkno,
1501 if (status < 0 && status != -ENOSPC) {
1508 if (status != -ENOSPC) {
1509 /* If the next search of this group is not likely to
1510 * yield a suitable extent, then we reset the last
1511 * group hint so as to not waste a disk read */
1512 if (bits_left < min_bits)
1513 ac->ac_last_group = 0;
1515 ac->ac_last_group = *bg_blkno;
1523 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1525 struct ocfs2_alloc_context *ac,
1527 u16 *suballoc_bit_start,
1528 unsigned int *num_bits,
1535 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1536 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1538 status = ocfs2_claim_suballoc_bits(osb,
1550 atomic_inc(&osb->alloc_stats.bg_allocs);
1552 *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1553 ac->ac_bits_given += (*num_bits);
1560 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1562 struct ocfs2_alloc_context *ac,
1567 unsigned int num_bits;
1573 BUG_ON(ac->ac_bits_given != 0);
1574 BUG_ON(ac->ac_bits_wanted != 1);
1575 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1577 status = ocfs2_claim_suballoc_bits(osb,
1589 atomic_inc(&osb->alloc_stats.bg_allocs);
1591 BUG_ON(num_bits != 1);
1593 *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1594 ac->ac_bits_given++;
1601 /* translate a group desc. blkno and it's bitmap offset into
1602 * disk cluster offset. */
1603 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1607 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1610 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1612 if (bg_blkno != osb->first_cluster_group_blkno)
1613 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1614 cluster += (u32) bg_bit_off;
1618 /* given a cluster offset, calculate which block group it belongs to
1619 * and return that block offset. */
1620 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1622 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1625 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1627 group_no = cluster / osb->bitmap_cpg;
1629 return osb->first_cluster_group_blkno;
1630 return ocfs2_clusters_to_blocks(inode->i_sb,
1631 group_no * osb->bitmap_cpg);
1634 /* given the block number of a cluster start, calculate which cluster
1635 * group and descriptor bitmap offset that corresponds to. */
1636 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1641 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1642 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1644 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1646 *bg_blkno = ocfs2_which_cluster_group(inode,
1649 if (*bg_blkno == osb->first_cluster_group_blkno)
1650 *bg_bit_off = (u16) data_cluster;
1652 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1653 data_blkno - *bg_blkno);
1657 * min_bits - minimum contiguous chunk from this total allocation we
1658 * can handle. set to what we asked for originally for a full
1659 * contig. allocation, set to '1' to indicate we can deal with extents
1662 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1664 struct ocfs2_alloc_context *ac,
1671 unsigned int bits_wanted = max_clusters;
1677 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1679 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1680 && ac->ac_which != OCFS2_AC_USE_MAIN);
1682 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1683 status = ocfs2_claim_local_alloc_bits(osb,
1690 atomic_inc(&osb->alloc_stats.local_data);
1692 if (min_clusters > (osb->bitmap_cpg - 1)) {
1693 /* The only paths asking for contiguousness
1694 * should know about this already. */
1695 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1696 "group bitmap size %u!\n", min_clusters,
1701 /* clamp the current request down to a realistic size. */
1702 if (bits_wanted > (osb->bitmap_cpg - 1))
1703 bits_wanted = osb->bitmap_cpg - 1;
1705 status = ocfs2_claim_suballoc_bits(osb,
1715 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1718 atomic_inc(&osb->alloc_stats.bitmap_data);
1722 if (status != -ENOSPC)
1727 ac->ac_bits_given += *num_clusters;
1734 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1736 struct ocfs2_alloc_context *ac,
1741 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1743 return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1744 bits_wanted, cluster_start, num_clusters);
1747 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1748 struct inode *alloc_inode,
1749 struct ocfs2_group_desc *bg,
1750 struct buffer_head *group_bh,
1751 unsigned int bit_off,
1752 unsigned int num_bits)
1756 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1757 struct ocfs2_group_desc *undo_bg = NULL;
1761 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
1762 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
1767 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1769 if (ocfs2_is_cluster_bitmap(alloc_inode))
1770 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1772 status = ocfs2_journal_access(handle, alloc_inode, group_bh,
1779 if (ocfs2_is_cluster_bitmap(alloc_inode))
1780 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1784 ocfs2_clear_bit((bit_off + tmp),
1785 (unsigned long *) bg->bg_bitmap);
1786 if (ocfs2_is_cluster_bitmap(alloc_inode))
1787 ocfs2_set_bit(bit_off + tmp,
1788 (unsigned long *) undo_bg->bg_bitmap);
1790 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1792 status = ocfs2_journal_dirty(handle, group_bh);
1800 * expects the suballoc inode to already be locked.
1802 int ocfs2_free_suballoc_bits(handle_t *handle,
1803 struct inode *alloc_inode,
1804 struct buffer_head *alloc_bh,
1805 unsigned int start_bit,
1811 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1812 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1813 struct buffer_head *group_bh = NULL;
1814 struct ocfs2_group_desc *group;
1818 /* The alloc_bh comes from ocfs2_free_dinode() or
1819 * ocfs2_free_clusters(). The callers have all locked the
1820 * allocator and gotten alloc_bh from the lock call. This
1821 * validates the dinode buffer. Any corruption that has happended
1823 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1824 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1826 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1827 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1828 (unsigned long long)bg_blkno, start_bit);
1830 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
1836 group = (struct ocfs2_group_desc *) group_bh->b_data;
1838 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1840 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1848 status = ocfs2_journal_access(handle, alloc_inode, alloc_bh,
1849 OCFS2_JOURNAL_ACCESS_WRITE);
1855 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1857 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1858 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1860 status = ocfs2_journal_dirty(handle, alloc_bh);
1873 int ocfs2_free_dinode(handle_t *handle,
1874 struct inode *inode_alloc_inode,
1875 struct buffer_head *inode_alloc_bh,
1876 struct ocfs2_dinode *di)
1878 u64 blk = le64_to_cpu(di->i_blkno);
1879 u16 bit = le16_to_cpu(di->i_suballoc_bit);
1880 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1882 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
1883 inode_alloc_bh, bit, bg_blkno, 1);
1886 int ocfs2_free_clusters(handle_t *handle,
1887 struct inode *bitmap_inode,
1888 struct buffer_head *bitmap_bh,
1890 unsigned int num_clusters)
1895 struct ocfs2_dinode *fe;
1897 /* You can't ever have a contiguous set of clusters
1898 * bigger than a block group bitmap so we never have to worry
1899 * about looping on them. */
1903 /* This is expensive. We can safely remove once this stuff has
1904 * gotten tested really well. */
1905 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
1907 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
1909 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
1912 mlog(0, "want to free %u clusters starting at block %llu\n",
1913 num_clusters, (unsigned long long)start_blk);
1914 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
1915 (unsigned long long)bg_blkno, bg_start_bit);
1917 status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
1918 bg_start_bit, bg_blkno,
1925 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
1933 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
1935 printk("Block Group:\n");
1936 printk("bg_signature: %s\n", bg->bg_signature);
1937 printk("bg_size: %u\n", bg->bg_size);
1938 printk("bg_bits: %u\n", bg->bg_bits);
1939 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
1940 printk("bg_chain: %u\n", bg->bg_chain);
1941 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
1942 printk("bg_next_group: %llu\n",
1943 (unsigned long long)bg->bg_next_group);
1944 printk("bg_parent_dinode: %llu\n",
1945 (unsigned long long)bg->bg_parent_dinode);
1946 printk("bg_blkno: %llu\n",
1947 (unsigned long long)bg->bg_blkno);
1950 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
1954 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
1955 printk("i_signature: %s\n", fe->i_signature);
1956 printk("i_size: %llu\n",
1957 (unsigned long long)fe->i_size);
1958 printk("i_clusters: %u\n", fe->i_clusters);
1959 printk("i_generation: %u\n",
1960 le32_to_cpu(fe->i_generation));
1961 printk("id1.bitmap1.i_used: %u\n",
1962 le32_to_cpu(fe->id1.bitmap1.i_used));
1963 printk("id1.bitmap1.i_total: %u\n",
1964 le32_to_cpu(fe->id1.bitmap1.i_total));
1965 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
1966 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
1967 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
1968 printk("id2.i_chain.cl_next_free_rec: %u\n",
1969 fe->id2.i_chain.cl_next_free_rec);
1970 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
1971 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
1972 fe->id2.i_chain.cl_recs[i].c_free);
1973 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
1974 fe->id2.i_chain.cl_recs[i].c_total);
1975 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
1976 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
1981 * For a given allocation, determine which allocators will need to be
1982 * accessed, and lock them, reserving the appropriate number of bits.
1984 * Sparse file systems call this from ocfs2_write_begin_nolock()
1985 * and ocfs2_allocate_unwritten_extents().
1987 * File systems which don't support holes call this from
1988 * ocfs2_extend_allocation().
1990 int ocfs2_lock_allocators(struct inode *inode,
1991 struct ocfs2_extent_tree *et,
1992 u32 clusters_to_add, u32 extents_to_split,
1993 struct ocfs2_alloc_context **data_ac,
1994 struct ocfs2_alloc_context **meta_ac)
1996 int ret = 0, num_free_extents;
1997 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
1998 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2004 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2006 num_free_extents = ocfs2_num_free_extents(osb, inode, et);
2007 if (num_free_extents < 0) {
2008 ret = num_free_extents;
2014 * Sparse allocation file systems need to be more conservative
2015 * with reserving room for expansion - the actual allocation
2016 * happens while we've got a journal handle open so re-taking
2017 * a cluster lock (because we ran out of room for another
2018 * extent) will violate ordering rules.
2020 * Most of the time we'll only be seeing this 1 cluster at a time
2023 * Always lock for any unwritten extents - we might want to
2024 * add blocks during a split.
2026 if (!num_free_extents ||
2027 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2028 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2036 if (clusters_to_add == 0)
2039 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2049 ocfs2_free_alloc_context(*meta_ac);
2054 * We cannot have an error and a non null *data_ac.