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 /* Callers got this descriptor from
846 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
847 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
849 found = start = best_offset = best_size = 0;
850 bitmap = bg->bg_bitmap;
852 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
853 if (offset == total_bits)
856 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
857 /* We found a zero, but we can't use it as it
858 * hasn't been put to disk yet! */
861 } else if (offset == start) {
862 /* we found a zero */
864 /* move start to the next bit to test */
867 /* got a zero after some ones */
871 if (found > best_size) {
873 best_offset = start - found;
875 /* we got everything we needed */
876 if (found == bits_wanted) {
877 /* mlog(0, "Found it all!\n"); */
882 /* XXX: I think the first clause is equivalent to the second
884 if (found == bits_wanted) {
885 *bit_off = start - found;
887 } else if (best_size) {
888 *bit_off = best_offset;
889 *bits_found = best_size;
892 /* No error log here -- see the comment above
893 * ocfs2_test_bg_bit_allocatable */
899 static inline int ocfs2_block_group_set_bits(handle_t *handle,
900 struct inode *alloc_inode,
901 struct ocfs2_group_desc *bg,
902 struct buffer_head *group_bh,
903 unsigned int bit_off,
904 unsigned int num_bits)
907 void *bitmap = bg->bg_bitmap;
908 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
912 /* All callers get the descriptor via
913 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
914 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
915 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
917 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
920 if (ocfs2_is_cluster_bitmap(alloc_inode))
921 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
923 status = ocfs2_journal_access(handle,
932 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
935 ocfs2_set_bit(bit_off++, bitmap);
937 status = ocfs2_journal_dirty(handle,
949 /* find the one with the most empty bits */
950 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
954 BUG_ON(!cl->cl_next_free_rec);
957 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
958 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
959 le32_to_cpu(cl->cl_recs[best].c_free))
964 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
968 static int ocfs2_relink_block_group(handle_t *handle,
969 struct inode *alloc_inode,
970 struct buffer_head *fe_bh,
971 struct buffer_head *bg_bh,
972 struct buffer_head *prev_bg_bh,
976 /* there is a really tiny chance the journal calls could fail,
977 * but we wouldn't want inconsistent blocks in *any* case. */
978 u64 fe_ptr, bg_ptr, prev_bg_ptr;
979 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
980 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
981 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
983 /* The caller got these descriptors from
984 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
985 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
986 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
988 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
989 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
990 (unsigned long long)le64_to_cpu(bg->bg_blkno),
991 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
993 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
994 bg_ptr = le64_to_cpu(bg->bg_next_group);
995 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
997 status = ocfs2_journal_access(handle, alloc_inode, prev_bg_bh,
998 OCFS2_JOURNAL_ACCESS_WRITE);
1004 prev_bg->bg_next_group = bg->bg_next_group;
1006 status = ocfs2_journal_dirty(handle, prev_bg_bh);
1012 status = ocfs2_journal_access(handle, alloc_inode, bg_bh,
1013 OCFS2_JOURNAL_ACCESS_WRITE);
1019 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1021 status = ocfs2_journal_dirty(handle, bg_bh);
1027 status = ocfs2_journal_access(handle, alloc_inode, fe_bh,
1028 OCFS2_JOURNAL_ACCESS_WRITE);
1034 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1036 status = ocfs2_journal_dirty(handle, fe_bh);
1045 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1046 bg->bg_next_group = cpu_to_le64(bg_ptr);
1047 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1054 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1057 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1060 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1061 * value on error. */
1062 static int ocfs2_cluster_group_search(struct inode *inode,
1063 struct buffer_head *group_bh,
1064 u32 bits_wanted, u32 min_bits,
1066 u16 *bit_off, u16 *bits_found)
1068 int search = -ENOSPC;
1071 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1072 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1073 u16 tmp_off, tmp_found;
1074 unsigned int max_bits, gd_cluster_off;
1076 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1078 if (gd->bg_free_bits_count) {
1079 max_bits = le16_to_cpu(gd->bg_bits);
1081 /* Tail groups in cluster bitmaps which aren't cpg
1082 * aligned are prone to partial extention by a failed
1083 * fs resize. If the file system resize never got to
1084 * update the dinode cluster count, then we don't want
1085 * to trust any clusters past it, regardless of what
1086 * the group descriptor says. */
1087 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1088 le64_to_cpu(gd->bg_blkno));
1089 if ((gd_cluster_off + max_bits) >
1090 OCFS2_I(inode)->ip_clusters) {
1091 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1092 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1093 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1094 le16_to_cpu(gd->bg_bits),
1095 OCFS2_I(inode)->ip_clusters, max_bits);
1098 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1099 group_bh, bits_wanted,
1101 &tmp_off, &tmp_found);
1106 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1108 tmp_off + tmp_found);
1109 mlog(0, "Checking %llu against %llu\n",
1110 (unsigned long long)blkoff,
1111 (unsigned long long)max_block);
1112 if (blkoff > max_block)
1116 /* ocfs2_block_group_find_clear_bits() might
1117 * return success, but we still want to return
1118 * -ENOSPC unless it found the minimum number
1120 if (min_bits <= tmp_found) {
1122 *bits_found = tmp_found;
1123 search = 0; /* success */
1124 } else if (tmp_found) {
1126 * Don't show bits which we'll be returning
1127 * for allocation to the local alloc bitmap.
1129 ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1136 static int ocfs2_block_group_search(struct inode *inode,
1137 struct buffer_head *group_bh,
1138 u32 bits_wanted, u32 min_bits,
1140 u16 *bit_off, u16 *bits_found)
1144 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1146 BUG_ON(min_bits != 1);
1147 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1149 if (bg->bg_free_bits_count) {
1150 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1151 group_bh, bits_wanted,
1152 le16_to_cpu(bg->bg_bits),
1153 bit_off, bits_found);
1154 if (!ret && max_block) {
1155 blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1157 mlog(0, "Checking %llu against %llu\n",
1158 (unsigned long long)blkoff,
1159 (unsigned long long)max_block);
1160 if (blkoff > max_block)
1168 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1170 struct buffer_head *di_bh,
1176 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1177 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1179 ret = ocfs2_journal_access(handle, inode, di_bh,
1180 OCFS2_JOURNAL_ACCESS_WRITE);
1186 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1187 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1188 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1190 ret = ocfs2_journal_dirty(handle, di_bh);
1198 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1203 unsigned int *num_bits,
1209 struct buffer_head *group_bh = NULL;
1210 struct ocfs2_group_desc *gd;
1211 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1212 struct inode *alloc_inode = ac->ac_inode;
1214 ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1221 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1222 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1223 ac->ac_max_block, bit_off, &found);
1232 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1234 le16_to_cpu(gd->bg_chain));
1240 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1241 *bit_off, *num_bits);
1245 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1253 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1258 unsigned int *num_bits,
1263 u16 chain, tmp_bits;
1266 struct inode *alloc_inode = ac->ac_inode;
1267 struct buffer_head *group_bh = NULL;
1268 struct buffer_head *prev_group_bh = NULL;
1269 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1270 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1271 struct ocfs2_group_desc *bg;
1273 chain = ac->ac_chain;
1274 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1276 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1278 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1279 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1285 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1288 /* for now, the chain search is a bit simplistic. We just use
1289 * the 1st group with any empty bits. */
1290 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1291 bits_wanted, min_bits,
1292 ac->ac_max_block, bit_off,
1293 &tmp_bits)) == -ENOSPC) {
1294 if (!bg->bg_next_group)
1297 brelse(prev_group_bh);
1298 prev_group_bh = NULL;
1300 next_group = le64_to_cpu(bg->bg_next_group);
1301 prev_group_bh = group_bh;
1303 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1304 next_group, &group_bh);
1309 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1312 if (status != -ENOSPC)
1317 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1318 tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1320 *num_bits = tmp_bits;
1322 BUG_ON(*num_bits == 0);
1325 * Keep track of previous block descriptor read. When
1326 * we find a target, if we have read more than X
1327 * number of descriptors, and the target is reasonably
1328 * empty, relink him to top of his chain.
1330 * We've read 0 extra blocks and only send one more to
1331 * the transaction, yet the next guy to search has a
1334 * Do this *after* figuring out how many bits we're taking out
1335 * of our target group.
1337 if (ac->ac_allow_chain_relink &&
1339 (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1340 status = ocfs2_relink_block_group(handle, alloc_inode,
1341 ac->ac_bh, group_bh,
1342 prev_group_bh, chain);
1349 /* Ok, claim our bits now: set the info on dinode, chainlist
1350 * and then the group */
1351 status = ocfs2_journal_access(handle,
1354 OCFS2_JOURNAL_ACCESS_WRITE);
1360 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1361 fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1362 le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1364 status = ocfs2_journal_dirty(handle,
1371 status = ocfs2_block_group_set_bits(handle,
1382 mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1383 (unsigned long long)le64_to_cpu(fe->i_blkno));
1385 *bg_blkno = le64_to_cpu(bg->bg_blkno);
1386 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1389 brelse(prev_group_bh);
1395 /* will give out up to bits_wanted contiguous bits. */
1396 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1397 struct ocfs2_alloc_context *ac,
1402 unsigned int *num_bits,
1408 u64 hint_blkno = ac->ac_last_group;
1409 struct ocfs2_chain_list *cl;
1410 struct ocfs2_dinode *fe;
1414 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1415 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1418 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1420 /* The bh was validated by the inode read during
1421 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1422 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1424 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1425 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1426 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1427 "bits but only %u total.",
1428 (unsigned long long)le64_to_cpu(fe->i_blkno),
1429 le32_to_cpu(fe->id1.bitmap1.i_used),
1430 le32_to_cpu(fe->id1.bitmap1.i_total));
1436 /* Attempt to short-circuit the usual search mechanism
1437 * by jumping straight to the most recently used
1438 * allocation group. This helps us mantain some
1439 * contiguousness across allocations. */
1440 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1441 min_bits, bit_off, num_bits,
1442 hint_blkno, &bits_left);
1444 /* Be careful to update *bg_blkno here as the
1445 * caller is expecting it to be filled in, and
1446 * ocfs2_search_one_group() won't do that for
1448 *bg_blkno = hint_blkno;
1451 if (status < 0 && status != -ENOSPC) {
1457 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1459 victim = ocfs2_find_victim_chain(cl);
1460 ac->ac_chain = victim;
1461 ac->ac_allow_chain_relink = 1;
1463 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1464 num_bits, bg_blkno, &bits_left);
1467 if (status < 0 && status != -ENOSPC) {
1472 mlog(0, "Search of victim chain %u came up with nothing, "
1473 "trying all chains now.\n", victim);
1475 /* If we didn't pick a good victim, then just default to
1476 * searching each chain in order. Don't allow chain relinking
1477 * because we only calculate enough journal credits for one
1478 * relink per alloc. */
1479 ac->ac_allow_chain_relink = 0;
1480 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1483 if (!cl->cl_recs[i].c_free)
1487 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1488 bit_off, num_bits, bg_blkno,
1492 if (status < 0 && status != -ENOSPC) {
1499 if (status != -ENOSPC) {
1500 /* If the next search of this group is not likely to
1501 * yield a suitable extent, then we reset the last
1502 * group hint so as to not waste a disk read */
1503 if (bits_left < min_bits)
1504 ac->ac_last_group = 0;
1506 ac->ac_last_group = *bg_blkno;
1514 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1516 struct ocfs2_alloc_context *ac,
1518 u16 *suballoc_bit_start,
1519 unsigned int *num_bits,
1526 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1527 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1529 status = ocfs2_claim_suballoc_bits(osb,
1541 atomic_inc(&osb->alloc_stats.bg_allocs);
1543 *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1544 ac->ac_bits_given += (*num_bits);
1551 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1553 struct ocfs2_alloc_context *ac,
1558 unsigned int num_bits;
1564 BUG_ON(ac->ac_bits_given != 0);
1565 BUG_ON(ac->ac_bits_wanted != 1);
1566 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1568 status = ocfs2_claim_suballoc_bits(osb,
1580 atomic_inc(&osb->alloc_stats.bg_allocs);
1582 BUG_ON(num_bits != 1);
1584 *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1585 ac->ac_bits_given++;
1592 /* translate a group desc. blkno and it's bitmap offset into
1593 * disk cluster offset. */
1594 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1598 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1601 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1603 if (bg_blkno != osb->first_cluster_group_blkno)
1604 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1605 cluster += (u32) bg_bit_off;
1609 /* given a cluster offset, calculate which block group it belongs to
1610 * and return that block offset. */
1611 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1613 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1616 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1618 group_no = cluster / osb->bitmap_cpg;
1620 return osb->first_cluster_group_blkno;
1621 return ocfs2_clusters_to_blocks(inode->i_sb,
1622 group_no * osb->bitmap_cpg);
1625 /* given the block number of a cluster start, calculate which cluster
1626 * group and descriptor bitmap offset that corresponds to. */
1627 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1632 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1633 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1635 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1637 *bg_blkno = ocfs2_which_cluster_group(inode,
1640 if (*bg_blkno == osb->first_cluster_group_blkno)
1641 *bg_bit_off = (u16) data_cluster;
1643 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1644 data_blkno - *bg_blkno);
1648 * min_bits - minimum contiguous chunk from this total allocation we
1649 * can handle. set to what we asked for originally for a full
1650 * contig. allocation, set to '1' to indicate we can deal with extents
1653 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1655 struct ocfs2_alloc_context *ac,
1662 unsigned int bits_wanted = max_clusters;
1668 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1670 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1671 && ac->ac_which != OCFS2_AC_USE_MAIN);
1673 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1674 status = ocfs2_claim_local_alloc_bits(osb,
1681 atomic_inc(&osb->alloc_stats.local_data);
1683 if (min_clusters > (osb->bitmap_cpg - 1)) {
1684 /* The only paths asking for contiguousness
1685 * should know about this already. */
1686 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1687 "group bitmap size %u!\n", min_clusters,
1692 /* clamp the current request down to a realistic size. */
1693 if (bits_wanted > (osb->bitmap_cpg - 1))
1694 bits_wanted = osb->bitmap_cpg - 1;
1696 status = ocfs2_claim_suballoc_bits(osb,
1706 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1709 atomic_inc(&osb->alloc_stats.bitmap_data);
1713 if (status != -ENOSPC)
1718 ac->ac_bits_given += *num_clusters;
1725 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1727 struct ocfs2_alloc_context *ac,
1732 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1734 return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1735 bits_wanted, cluster_start, num_clusters);
1738 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1739 struct inode *alloc_inode,
1740 struct ocfs2_group_desc *bg,
1741 struct buffer_head *group_bh,
1742 unsigned int bit_off,
1743 unsigned int num_bits)
1747 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1748 struct ocfs2_group_desc *undo_bg = NULL;
1752 /* The caller got this descriptor from
1753 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1754 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1756 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1758 if (ocfs2_is_cluster_bitmap(alloc_inode))
1759 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1761 status = ocfs2_journal_access(handle, alloc_inode, group_bh,
1768 if (ocfs2_is_cluster_bitmap(alloc_inode))
1769 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1773 ocfs2_clear_bit((bit_off + tmp),
1774 (unsigned long *) bg->bg_bitmap);
1775 if (ocfs2_is_cluster_bitmap(alloc_inode))
1776 ocfs2_set_bit(bit_off + tmp,
1777 (unsigned long *) undo_bg->bg_bitmap);
1779 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1781 status = ocfs2_journal_dirty(handle, group_bh);
1789 * expects the suballoc inode to already be locked.
1791 int ocfs2_free_suballoc_bits(handle_t *handle,
1792 struct inode *alloc_inode,
1793 struct buffer_head *alloc_bh,
1794 unsigned int start_bit,
1800 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1801 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1802 struct buffer_head *group_bh = NULL;
1803 struct ocfs2_group_desc *group;
1807 /* The alloc_bh comes from ocfs2_free_dinode() or
1808 * ocfs2_free_clusters(). The callers have all locked the
1809 * allocator and gotten alloc_bh from the lock call. This
1810 * validates the dinode buffer. Any corruption that has happended
1812 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1813 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1815 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1816 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1817 (unsigned long long)bg_blkno, start_bit);
1819 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
1825 group = (struct ocfs2_group_desc *) group_bh->b_data;
1827 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1829 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1837 status = ocfs2_journal_access(handle, alloc_inode, alloc_bh,
1838 OCFS2_JOURNAL_ACCESS_WRITE);
1844 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1846 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1847 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1849 status = ocfs2_journal_dirty(handle, alloc_bh);
1862 int ocfs2_free_dinode(handle_t *handle,
1863 struct inode *inode_alloc_inode,
1864 struct buffer_head *inode_alloc_bh,
1865 struct ocfs2_dinode *di)
1867 u64 blk = le64_to_cpu(di->i_blkno);
1868 u16 bit = le16_to_cpu(di->i_suballoc_bit);
1869 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1871 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
1872 inode_alloc_bh, bit, bg_blkno, 1);
1875 int ocfs2_free_clusters(handle_t *handle,
1876 struct inode *bitmap_inode,
1877 struct buffer_head *bitmap_bh,
1879 unsigned int num_clusters)
1884 struct ocfs2_dinode *fe;
1886 /* You can't ever have a contiguous set of clusters
1887 * bigger than a block group bitmap so we never have to worry
1888 * about looping on them. */
1892 /* This is expensive. We can safely remove once this stuff has
1893 * gotten tested really well. */
1894 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
1896 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
1898 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
1901 mlog(0, "want to free %u clusters starting at block %llu\n",
1902 num_clusters, (unsigned long long)start_blk);
1903 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
1904 (unsigned long long)bg_blkno, bg_start_bit);
1906 status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
1907 bg_start_bit, bg_blkno,
1914 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
1922 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
1924 printk("Block Group:\n");
1925 printk("bg_signature: %s\n", bg->bg_signature);
1926 printk("bg_size: %u\n", bg->bg_size);
1927 printk("bg_bits: %u\n", bg->bg_bits);
1928 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
1929 printk("bg_chain: %u\n", bg->bg_chain);
1930 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
1931 printk("bg_next_group: %llu\n",
1932 (unsigned long long)bg->bg_next_group);
1933 printk("bg_parent_dinode: %llu\n",
1934 (unsigned long long)bg->bg_parent_dinode);
1935 printk("bg_blkno: %llu\n",
1936 (unsigned long long)bg->bg_blkno);
1939 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
1943 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
1944 printk("i_signature: %s\n", fe->i_signature);
1945 printk("i_size: %llu\n",
1946 (unsigned long long)fe->i_size);
1947 printk("i_clusters: %u\n", fe->i_clusters);
1948 printk("i_generation: %u\n",
1949 le32_to_cpu(fe->i_generation));
1950 printk("id1.bitmap1.i_used: %u\n",
1951 le32_to_cpu(fe->id1.bitmap1.i_used));
1952 printk("id1.bitmap1.i_total: %u\n",
1953 le32_to_cpu(fe->id1.bitmap1.i_total));
1954 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
1955 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
1956 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
1957 printk("id2.i_chain.cl_next_free_rec: %u\n",
1958 fe->id2.i_chain.cl_next_free_rec);
1959 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
1960 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
1961 fe->id2.i_chain.cl_recs[i].c_free);
1962 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
1963 fe->id2.i_chain.cl_recs[i].c_total);
1964 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
1965 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
1970 * For a given allocation, determine which allocators will need to be
1971 * accessed, and lock them, reserving the appropriate number of bits.
1973 * Sparse file systems call this from ocfs2_write_begin_nolock()
1974 * and ocfs2_allocate_unwritten_extents().
1976 * File systems which don't support holes call this from
1977 * ocfs2_extend_allocation().
1979 int ocfs2_lock_allocators(struct inode *inode,
1980 struct ocfs2_extent_tree *et,
1981 u32 clusters_to_add, u32 extents_to_split,
1982 struct ocfs2_alloc_context **data_ac,
1983 struct ocfs2_alloc_context **meta_ac)
1985 int ret = 0, num_free_extents;
1986 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
1987 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1993 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
1995 num_free_extents = ocfs2_num_free_extents(osb, inode, et);
1996 if (num_free_extents < 0) {
1997 ret = num_free_extents;
2003 * Sparse allocation file systems need to be more conservative
2004 * with reserving room for expansion - the actual allocation
2005 * happens while we've got a journal handle open so re-taking
2006 * a cluster lock (because we ran out of room for another
2007 * extent) will violate ordering rules.
2009 * Most of the time we'll only be seeing this 1 cluster at a time
2012 * Always lock for any unwritten extents - we might want to
2013 * add blocks during a split.
2015 if (!num_free_extents ||
2016 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2017 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2025 if (clusters_to_add == 0)
2028 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2038 ocfs2_free_alloc_context(*meta_ac);
2043 * We cannot have an error and a non null *data_ac.