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 #include <cluster/masklog.h>
37 #include "blockcheck.h"
41 #include "localalloc.h"
46 #include "ocfs2_trace.h"
48 #include "buffer_head_io.h"
50 #define NOT_ALLOC_NEW_GROUP 0
51 #define ALLOC_NEW_GROUP 0x1
52 #define ALLOC_GROUPS_FROM_GLOBAL 0x2
54 #define OCFS2_MAX_TO_STEAL 1024
56 struct ocfs2_suballoc_result {
57 u64 sr_bg_blkno; /* The bg we allocated from. Set
58 to 0 when a block group is
60 u64 sr_bg_stable_blkno; /*
61 * Doesn't change, always
66 u64 sr_blkno; /* The first allocated block */
67 unsigned int sr_bit_offset; /* The bit in the bg */
68 unsigned int sr_bits; /* How many bits we claimed */
71 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
73 if (res->sr_blkno == 0)
77 return res->sr_bg_blkno;
79 return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
82 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
83 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
84 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
85 static int ocfs2_block_group_fill(handle_t *handle,
86 struct inode *alloc_inode,
87 struct buffer_head *bg_bh,
89 unsigned int group_clusters,
91 struct ocfs2_chain_list *cl);
92 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
93 struct inode *alloc_inode,
94 struct buffer_head *bh,
96 u64 *last_alloc_group,
99 static int ocfs2_cluster_group_search(struct inode *inode,
100 struct buffer_head *group_bh,
101 u32 bits_wanted, u32 min_bits,
103 struct ocfs2_suballoc_result *res);
104 static int ocfs2_block_group_search(struct inode *inode,
105 struct buffer_head *group_bh,
106 u32 bits_wanted, u32 min_bits,
108 struct ocfs2_suballoc_result *res);
109 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
113 struct ocfs2_suballoc_result *res);
114 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
116 static int ocfs2_relink_block_group(handle_t *handle,
117 struct inode *alloc_inode,
118 struct buffer_head *fe_bh,
119 struct buffer_head *bg_bh,
120 struct buffer_head *prev_bg_bh,
122 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
124 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
127 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
131 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
132 u32 bits_wanted, u64 max_block,
134 struct ocfs2_alloc_context **ac);
136 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
138 struct inode *inode = ac->ac_inode;
141 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
142 ocfs2_inode_unlock(inode, 1);
144 mutex_unlock(&inode->i_mutex);
152 if (ac->ac_find_loc_priv) {
153 kfree(ac->ac_find_loc_priv);
154 ac->ac_find_loc_priv = NULL;
158 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
160 ocfs2_free_ac_resource(ac);
164 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
166 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
169 #define do_error(fmt, ...) \
172 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
174 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
177 static int ocfs2_validate_gd_self(struct super_block *sb,
178 struct buffer_head *bh,
181 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
183 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
184 do_error("Group descriptor #%llu has bad signature %.*s",
185 (unsigned long long)bh->b_blocknr, 7,
190 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
191 do_error("Group descriptor #%llu has an invalid bg_blkno "
193 (unsigned long long)bh->b_blocknr,
194 (unsigned long long)le64_to_cpu(gd->bg_blkno));
198 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
199 do_error("Group descriptor #%llu has an invalid "
200 "fs_generation of #%u",
201 (unsigned long long)bh->b_blocknr,
202 le32_to_cpu(gd->bg_generation));
206 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
207 do_error("Group descriptor #%llu has bit count %u but "
208 "claims that %u are free",
209 (unsigned long long)bh->b_blocknr,
210 le16_to_cpu(gd->bg_bits),
211 le16_to_cpu(gd->bg_free_bits_count));
215 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
216 do_error("Group descriptor #%llu has bit count %u but "
217 "max bitmap bits of %u",
218 (unsigned long long)bh->b_blocknr,
219 le16_to_cpu(gd->bg_bits),
220 8 * le16_to_cpu(gd->bg_size));
227 static int ocfs2_validate_gd_parent(struct super_block *sb,
228 struct ocfs2_dinode *di,
229 struct buffer_head *bh,
232 unsigned int max_bits;
233 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
235 if (di->i_blkno != gd->bg_parent_dinode) {
236 do_error("Group descriptor #%llu has bad parent "
237 "pointer (%llu, expected %llu)",
238 (unsigned long long)bh->b_blocknr,
239 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
240 (unsigned long long)le64_to_cpu(di->i_blkno));
244 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
245 if (le16_to_cpu(gd->bg_bits) > max_bits) {
246 do_error("Group descriptor #%llu has bit count of %u",
247 (unsigned long long)bh->b_blocknr,
248 le16_to_cpu(gd->bg_bits));
252 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
253 if ((le16_to_cpu(gd->bg_chain) >
254 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
255 ((le16_to_cpu(gd->bg_chain) ==
256 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
257 do_error("Group descriptor #%llu has bad chain %u",
258 (unsigned long long)bh->b_blocknr,
259 le16_to_cpu(gd->bg_chain));
269 * This version only prints errors. It does not fail the filesystem, and
270 * exists only for resize.
272 int ocfs2_check_group_descriptor(struct super_block *sb,
273 struct ocfs2_dinode *di,
274 struct buffer_head *bh)
277 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
279 BUG_ON(!buffer_uptodate(bh));
282 * If the ecc fails, we return the error but otherwise
283 * leave the filesystem running. We know any error is
284 * local to this block.
286 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
289 "Checksum failed for group descriptor %llu\n",
290 (unsigned long long)bh->b_blocknr);
292 rc = ocfs2_validate_gd_self(sb, bh, 1);
294 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
299 static int ocfs2_validate_group_descriptor(struct super_block *sb,
300 struct buffer_head *bh)
303 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
305 trace_ocfs2_validate_group_descriptor(
306 (unsigned long long)bh->b_blocknr);
308 BUG_ON(!buffer_uptodate(bh));
311 * If the ecc fails, we return the error but otherwise
312 * leave the filesystem running. We know any error is
313 * local to this block.
315 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
320 * Errors after here are fatal.
323 return ocfs2_validate_gd_self(sb, bh, 0);
326 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
327 u64 gd_blkno, struct buffer_head **bh)
330 struct buffer_head *tmp = *bh;
332 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
333 ocfs2_validate_group_descriptor);
337 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
343 /* If ocfs2_read_block() got us a new bh, pass it up. */
351 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
352 struct ocfs2_group_desc *bg,
353 struct ocfs2_chain_list *cl,
354 u64 p_blkno, unsigned int clusters)
356 struct ocfs2_extent_list *el = &bg->bg_list;
357 struct ocfs2_extent_rec *rec;
359 BUG_ON(!ocfs2_supports_discontig_bg(osb));
360 if (!el->l_next_free_rec)
361 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
362 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
363 rec->e_blkno = cpu_to_le64(p_blkno);
364 rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
365 le16_to_cpu(cl->cl_bpc));
366 rec->e_leaf_clusters = cpu_to_le16(clusters);
367 le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
368 le16_add_cpu(&bg->bg_free_bits_count,
369 clusters * le16_to_cpu(cl->cl_bpc));
370 le16_add_cpu(&el->l_next_free_rec, 1);
373 static int ocfs2_block_group_fill(handle_t *handle,
374 struct inode *alloc_inode,
375 struct buffer_head *bg_bh,
377 unsigned int group_clusters,
379 struct ocfs2_chain_list *cl)
382 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
383 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
384 struct super_block * sb = alloc_inode->i_sb;
386 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
387 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
389 (unsigned long long)group_blkno,
390 (unsigned long long) bg_bh->b_blocknr);
395 status = ocfs2_journal_access_gd(handle,
396 INODE_CACHE(alloc_inode),
398 OCFS2_JOURNAL_ACCESS_CREATE);
404 memset(bg, 0, sb->s_blocksize);
405 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
406 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
407 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
408 osb->s_feature_incompat));
409 bg->bg_chain = cpu_to_le16(my_chain);
410 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
411 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
412 bg->bg_blkno = cpu_to_le64(group_blkno);
413 if (group_clusters == le16_to_cpu(cl->cl_cpg))
414 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
416 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
419 /* set the 1st bit in the bitmap to account for the descriptor block */
420 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
421 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
423 ocfs2_journal_dirty(handle, bg_bh);
425 /* There is no need to zero out or otherwise initialize the
426 * other blocks in a group - All valid FS metadata in a block
427 * group stores the superblock fs_generation value at
428 * allocation time. */
436 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
441 while (curr < le16_to_cpu(cl->cl_count)) {
442 if (le32_to_cpu(cl->cl_recs[best].c_total) >
443 le32_to_cpu(cl->cl_recs[curr].c_total))
450 static struct buffer_head *
451 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
452 struct inode *alloc_inode,
453 struct ocfs2_alloc_context *ac,
454 struct ocfs2_chain_list *cl)
457 u32 bit_off, num_bits;
459 struct buffer_head *bg_bh;
460 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
462 status = ocfs2_claim_clusters(handle, ac,
463 le16_to_cpu(cl->cl_cpg), &bit_off,
466 if (status != -ENOSPC)
471 /* setup the group */
472 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
473 trace_ocfs2_block_group_alloc_contig(
474 (unsigned long long)bg_blkno, alloc_rec);
476 bg_bh = sb_getblk(osb->sb, bg_blkno);
482 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
484 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
485 bg_blkno, num_bits, alloc_rec, cl);
492 return status ? ERR_PTR(status) : bg_bh;
495 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
497 struct ocfs2_alloc_context *ac,
498 unsigned int min_bits,
499 u32 *bit_off, u32 *num_bits)
504 status = ocfs2_claim_clusters(handle, ac, min_bits,
506 if (status != -ENOSPC)
515 static int ocfs2_block_group_grow_discontig(handle_t *handle,
516 struct inode *alloc_inode,
517 struct buffer_head *bg_bh,
518 struct ocfs2_alloc_context *ac,
519 struct ocfs2_chain_list *cl,
520 unsigned int min_bits)
523 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
524 struct ocfs2_group_desc *bg =
525 (struct ocfs2_group_desc *)bg_bh->b_data;
526 unsigned int needed = le16_to_cpu(cl->cl_cpg) -
527 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
528 u32 p_cpos, clusters;
530 struct ocfs2_extent_list *el = &bg->bg_list;
532 status = ocfs2_journal_access_gd(handle,
533 INODE_CACHE(alloc_inode),
535 OCFS2_JOURNAL_ACCESS_CREATE);
541 while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
542 le16_to_cpu(el->l_count))) {
543 if (min_bits > needed)
545 status = ocfs2_block_group_claim_bits(osb, handle, ac,
549 if (status != -ENOSPC)
553 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
554 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
558 needed = le16_to_cpu(cl->cl_cpg) -
559 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
564 * We have used up all the extent rec but can't fill up
565 * the cpg. So bail out.
571 ocfs2_journal_dirty(handle, bg_bh);
577 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
578 struct ocfs2_alloc_context *cluster_ac,
579 struct inode *alloc_inode,
580 struct buffer_head *bg_bh)
583 struct ocfs2_group_desc *bg;
584 struct ocfs2_extent_list *el;
585 struct ocfs2_extent_rec *rec;
590 bg = (struct ocfs2_group_desc *)bg_bh->b_data;
592 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
593 rec = &el->l_recs[i];
594 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
596 le64_to_cpu(rec->e_blkno),
597 le16_to_cpu(rec->e_leaf_clusters));
600 /* Try all the clusters to free */
603 ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
607 static struct buffer_head *
608 ocfs2_block_group_alloc_discontig(handle_t *handle,
609 struct inode *alloc_inode,
610 struct ocfs2_alloc_context *ac,
611 struct ocfs2_chain_list *cl)
614 u32 bit_off, num_bits;
616 unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
617 struct buffer_head *bg_bh = NULL;
618 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
619 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
621 if (!ocfs2_supports_discontig_bg(osb)) {
626 status = ocfs2_extend_trans(handle,
627 ocfs2_calc_bg_discontig_credits(osb->sb));
634 * We're going to be grabbing from multiple cluster groups.
635 * We don't have enough credits to relink them all, and the
636 * cluster groups will be staying in cache for the duration of
639 ac->ac_disable_chain_relink = 1;
641 /* Claim the first region */
642 status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
643 &bit_off, &num_bits);
645 if (status != -ENOSPC)
651 /* setup the group */
652 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
653 trace_ocfs2_block_group_alloc_discontig(
654 (unsigned long long)bg_blkno, alloc_rec);
656 bg_bh = sb_getblk(osb->sb, bg_blkno);
662 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
664 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
665 bg_blkno, num_bits, alloc_rec, cl);
671 status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
672 bg_bh, ac, cl, min_bits);
678 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
679 return status ? ERR_PTR(status) : bg_bh;
683 * We expect the block group allocator to already be locked.
685 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
686 struct inode *alloc_inode,
687 struct buffer_head *bh,
689 u64 *last_alloc_group,
693 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
694 struct ocfs2_chain_list *cl;
695 struct ocfs2_alloc_context *ac = NULL;
696 handle_t *handle = NULL;
698 struct buffer_head *bg_bh = NULL;
699 struct ocfs2_group_desc *bg;
701 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
703 cl = &fe->id2.i_chain;
704 status = ocfs2_reserve_clusters_with_limit(osb,
705 le16_to_cpu(cl->cl_cpg),
706 max_block, flags, &ac);
708 if (status != -ENOSPC)
713 credits = ocfs2_calc_group_alloc_credits(osb->sb,
714 le16_to_cpu(cl->cl_cpg));
715 handle = ocfs2_start_trans(osb, credits);
716 if (IS_ERR(handle)) {
717 status = PTR_ERR(handle);
723 if (last_alloc_group && *last_alloc_group != 0) {
724 trace_ocfs2_block_group_alloc(
725 (unsigned long long)*last_alloc_group);
726 ac->ac_last_group = *last_alloc_group;
729 bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
731 if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
732 bg_bh = ocfs2_block_group_alloc_discontig(handle,
736 status = PTR_ERR(bg_bh);
738 if (status != -ENOSPC)
742 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
744 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
745 bh, OCFS2_JOURNAL_ACCESS_WRITE);
751 alloc_rec = le16_to_cpu(bg->bg_chain);
752 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
753 le16_to_cpu(bg->bg_free_bits_count));
754 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
755 le16_to_cpu(bg->bg_bits));
756 cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
757 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
758 le16_add_cpu(&cl->cl_next_free_rec, 1);
760 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
761 le16_to_cpu(bg->bg_free_bits_count));
762 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
763 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
765 ocfs2_journal_dirty(handle, bh);
767 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
768 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
769 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
770 le32_to_cpu(fe->i_clusters)));
771 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
772 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
773 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
777 /* save the new last alloc group so that the caller can cache it. */
778 if (last_alloc_group)
779 *last_alloc_group = ac->ac_last_group;
783 ocfs2_commit_trans(osb, handle);
786 ocfs2_free_alloc_context(ac);
795 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
796 struct ocfs2_alloc_context *ac,
799 u64 *last_alloc_group,
803 u32 bits_wanted = ac->ac_bits_wanted;
804 struct inode *alloc_inode;
805 struct buffer_head *bh = NULL;
806 struct ocfs2_dinode *fe;
809 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
815 mutex_lock(&alloc_inode->i_mutex);
817 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
819 mutex_unlock(&alloc_inode->i_mutex);
826 ac->ac_inode = alloc_inode;
827 ac->ac_alloc_slot = slot;
829 fe = (struct ocfs2_dinode *) bh->b_data;
831 /* The bh was validated by the inode read inside
832 * ocfs2_inode_lock(). Any corruption is a code bug. */
833 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
835 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
836 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
837 (unsigned long long)le64_to_cpu(fe->i_blkno));
842 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
843 le32_to_cpu(fe->id1.bitmap1.i_used);
845 if (bits_wanted > free_bits) {
846 /* cluster bitmap never grows */
847 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
848 trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
854 if (!(flags & ALLOC_NEW_GROUP)) {
855 trace_ocfs2_reserve_suballoc_bits_no_new_group(
856 slot, bits_wanted, free_bits);
861 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
863 last_alloc_group, flags);
865 if (status != -ENOSPC)
869 atomic_inc(&osb->alloc_stats.bg_extends);
871 /* You should never ask for this much metadata */
873 (le32_to_cpu(fe->id1.bitmap1.i_total)
874 - le32_to_cpu(fe->id1.bitmap1.i_used)));
887 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
889 spin_lock(&osb->osb_lock);
890 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
891 spin_unlock(&osb->osb_lock);
892 atomic_set(&osb->s_num_inodes_stolen, 0);
895 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
897 spin_lock(&osb->osb_lock);
898 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
899 spin_unlock(&osb->osb_lock);
900 atomic_set(&osb->s_num_meta_stolen, 0);
903 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
905 ocfs2_init_inode_steal_slot(osb);
906 ocfs2_init_meta_steal_slot(osb);
909 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
911 spin_lock(&osb->osb_lock);
912 if (type == INODE_ALLOC_SYSTEM_INODE)
913 osb->s_inode_steal_slot = slot;
914 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
915 osb->s_meta_steal_slot = slot;
916 spin_unlock(&osb->osb_lock);
919 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
921 int slot = OCFS2_INVALID_SLOT;
923 spin_lock(&osb->osb_lock);
924 if (type == INODE_ALLOC_SYSTEM_INODE)
925 slot = osb->s_inode_steal_slot;
926 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
927 slot = osb->s_meta_steal_slot;
928 spin_unlock(&osb->osb_lock);
933 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
935 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
938 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
940 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
943 static int ocfs2_steal_resource(struct ocfs2_super *osb,
944 struct ocfs2_alloc_context *ac,
947 int i, status = -ENOSPC;
948 int slot = __ocfs2_get_steal_slot(osb, type);
950 /* Start to steal resource from the first slot after ours. */
951 if (slot == OCFS2_INVALID_SLOT)
952 slot = osb->slot_num + 1;
954 for (i = 0; i < osb->max_slots; i++, slot++) {
955 if (slot == osb->max_slots)
958 if (slot == osb->slot_num)
961 status = ocfs2_reserve_suballoc_bits(osb, ac,
964 NOT_ALLOC_NEW_GROUP);
966 __ocfs2_set_steal_slot(osb, slot, type);
970 ocfs2_free_ac_resource(ac);
976 static int ocfs2_steal_inode(struct ocfs2_super *osb,
977 struct ocfs2_alloc_context *ac)
979 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
982 static int ocfs2_steal_meta(struct ocfs2_super *osb,
983 struct ocfs2_alloc_context *ac)
985 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
988 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
990 struct ocfs2_alloc_context **ac)
993 int slot = ocfs2_get_meta_steal_slot(osb);
995 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1002 (*ac)->ac_bits_wanted = blocks;
1003 (*ac)->ac_which = OCFS2_AC_USE_META;
1004 (*ac)->ac_group_search = ocfs2_block_group_search;
1006 if (slot != OCFS2_INVALID_SLOT &&
1007 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1010 atomic_set(&osb->s_num_meta_stolen, 0);
1011 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
1012 EXTENT_ALLOC_SYSTEM_INODE,
1013 (u32)osb->slot_num, NULL,
1014 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1019 if (slot != OCFS2_INVALID_SLOT)
1020 ocfs2_init_meta_steal_slot(osb);
1022 } else if (status < 0 && status != -ENOSPC) {
1027 ocfs2_free_ac_resource(*ac);
1030 status = ocfs2_steal_meta(osb, *ac);
1031 atomic_inc(&osb->s_num_meta_stolen);
1033 if (status != -ENOSPC)
1040 if ((status < 0) && *ac) {
1041 ocfs2_free_alloc_context(*ac);
1050 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1051 struct ocfs2_extent_list *root_el,
1052 struct ocfs2_alloc_context **ac)
1054 return ocfs2_reserve_new_metadata_blocks(osb,
1055 ocfs2_extend_meta_needed(root_el),
1059 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1060 struct ocfs2_alloc_context **ac)
1063 int slot = ocfs2_get_inode_steal_slot(osb);
1066 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1073 (*ac)->ac_bits_wanted = 1;
1074 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1076 (*ac)->ac_group_search = ocfs2_block_group_search;
1079 * stat(2) can't handle i_ino > 32bits, so we tell the
1080 * lower levels not to allocate us a block group past that
1081 * limit. The 'inode64' mount option avoids this behavior.
1083 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1084 (*ac)->ac_max_block = (u32)~0U;
1087 * slot is set when we successfully steal inode from other nodes.
1088 * It is reset in 3 places:
1089 * 1. when we flush the truncate log
1090 * 2. when we complete local alloc recovery.
1091 * 3. when we successfully allocate from our own slot.
1092 * After it is set, we will go on stealing inodes until we find the
1093 * need to check our slots to see whether there is some space for us.
1095 if (slot != OCFS2_INVALID_SLOT &&
1096 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1099 atomic_set(&osb->s_num_inodes_stolen, 0);
1100 alloc_group = osb->osb_inode_alloc_group;
1101 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1102 INODE_ALLOC_SYSTEM_INODE,
1106 ALLOC_GROUPS_FROM_GLOBAL);
1110 spin_lock(&osb->osb_lock);
1111 osb->osb_inode_alloc_group = alloc_group;
1112 spin_unlock(&osb->osb_lock);
1113 trace_ocfs2_reserve_new_inode_new_group(
1114 (unsigned long long)alloc_group);
1117 * Some inodes must be freed by us, so try to allocate
1118 * from our own next time.
1120 if (slot != OCFS2_INVALID_SLOT)
1121 ocfs2_init_inode_steal_slot(osb);
1123 } else if (status < 0 && status != -ENOSPC) {
1128 ocfs2_free_ac_resource(*ac);
1131 status = ocfs2_steal_inode(osb, *ac);
1132 atomic_inc(&osb->s_num_inodes_stolen);
1134 if (status != -ENOSPC)
1141 if ((status < 0) && *ac) {
1142 ocfs2_free_alloc_context(*ac);
1151 /* local alloc code has to do the same thing, so rather than do this
1153 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1154 struct ocfs2_alloc_context *ac)
1158 ac->ac_which = OCFS2_AC_USE_MAIN;
1159 ac->ac_group_search = ocfs2_cluster_group_search;
1161 status = ocfs2_reserve_suballoc_bits(osb, ac,
1162 GLOBAL_BITMAP_SYSTEM_INODE,
1163 OCFS2_INVALID_SLOT, NULL,
1165 if (status < 0 && status != -ENOSPC) {
1174 /* Callers don't need to care which bitmap (local alloc or main) to
1175 * use so we figure it out for them, but unfortunately this clutters
1177 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1178 u32 bits_wanted, u64 max_block,
1180 struct ocfs2_alloc_context **ac)
1184 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1191 (*ac)->ac_bits_wanted = bits_wanted;
1192 (*ac)->ac_max_block = max_block;
1195 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1196 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1197 status = ocfs2_reserve_local_alloc_bits(osb,
1200 if ((status < 0) && (status != -ENOSPC)) {
1206 if (status == -ENOSPC) {
1207 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1209 if (status != -ENOSPC)
1217 if ((status < 0) && *ac) {
1218 ocfs2_free_alloc_context(*ac);
1227 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1229 struct ocfs2_alloc_context **ac)
1231 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1232 ALLOC_NEW_GROUP, ac);
1236 * More or less lifted from ext3. I'll leave their description below:
1238 * "For ext3 allocations, we must not reuse any blocks which are
1239 * allocated in the bitmap buffer's "last committed data" copy. This
1240 * prevents deletes from freeing up the page for reuse until we have
1241 * committed the delete transaction.
1243 * If we didn't do this, then deleting something and reallocating it as
1244 * data would allow the old block to be overwritten before the
1245 * transaction committed (because we force data to disk before commit).
1246 * This would lead to corruption if we crashed between overwriting the
1247 * data and committing the delete.
1249 * @@@ We may want to make this allocation behaviour conditional on
1250 * data-writes at some point, and disable it for metadata allocations or
1251 * sync-data inodes."
1253 * Note: OCFS2 already does this differently for metadata vs data
1254 * allocations, as those bitmaps are separate and undo access is never
1255 * called on a metadata group descriptor.
1257 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1260 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1263 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1266 if (!buffer_jbd(bg_bh))
1269 jbd_lock_bh_state(bg_bh);
1270 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1272 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1275 jbd_unlock_bh_state(bg_bh);
1280 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1281 struct buffer_head *bg_bh,
1282 unsigned int bits_wanted,
1283 unsigned int total_bits,
1284 struct ocfs2_suballoc_result *res)
1287 u16 best_offset, best_size;
1288 int offset, start, found, status = 0;
1289 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1291 /* Callers got this descriptor from
1292 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1293 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1295 found = start = best_offset = best_size = 0;
1296 bitmap = bg->bg_bitmap;
1298 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1299 if (offset == total_bits)
1302 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1303 /* We found a zero, but we can't use it as it
1304 * hasn't been put to disk yet! */
1307 } else if (offset == start) {
1308 /* we found a zero */
1310 /* move start to the next bit to test */
1313 /* got a zero after some ones */
1317 if (found > best_size) {
1319 best_offset = start - found;
1321 /* we got everything we needed */
1322 if (found == bits_wanted) {
1323 /* mlog(0, "Found it all!\n"); */
1329 res->sr_bit_offset = best_offset;
1330 res->sr_bits = best_size;
1333 /* No error log here -- see the comment above
1334 * ocfs2_test_bg_bit_allocatable */
1340 int ocfs2_block_group_set_bits(handle_t *handle,
1341 struct inode *alloc_inode,
1342 struct ocfs2_group_desc *bg,
1343 struct buffer_head *group_bh,
1344 unsigned int bit_off,
1345 unsigned int num_bits)
1348 void *bitmap = bg->bg_bitmap;
1349 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1351 /* All callers get the descriptor via
1352 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1353 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1354 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1356 trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1358 if (ocfs2_is_cluster_bitmap(alloc_inode))
1359 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1361 status = ocfs2_journal_access_gd(handle,
1362 INODE_CACHE(alloc_inode),
1370 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1371 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1372 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1373 " count %u but claims %u are freed. num_bits %d",
1374 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1375 le16_to_cpu(bg->bg_bits),
1376 le16_to_cpu(bg->bg_free_bits_count), num_bits);
1380 ocfs2_set_bit(bit_off++, bitmap);
1382 ocfs2_journal_dirty(handle, group_bh);
1388 /* find the one with the most empty bits */
1389 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1393 BUG_ON(!cl->cl_next_free_rec);
1396 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1397 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1398 le32_to_cpu(cl->cl_recs[best].c_free))
1403 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1407 static int ocfs2_relink_block_group(handle_t *handle,
1408 struct inode *alloc_inode,
1409 struct buffer_head *fe_bh,
1410 struct buffer_head *bg_bh,
1411 struct buffer_head *prev_bg_bh,
1415 /* there is a really tiny chance the journal calls could fail,
1416 * but we wouldn't want inconsistent blocks in *any* case. */
1417 u64 bg_ptr, prev_bg_ptr;
1418 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1419 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1420 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1422 /* The caller got these descriptors from
1423 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1424 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1425 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1427 trace_ocfs2_relink_block_group(
1428 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1429 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1430 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1432 bg_ptr = le64_to_cpu(bg->bg_next_group);
1433 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1435 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1437 OCFS2_JOURNAL_ACCESS_WRITE);
1441 prev_bg->bg_next_group = bg->bg_next_group;
1442 ocfs2_journal_dirty(handle, prev_bg_bh);
1444 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1445 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1447 goto out_rollback_prev_bg;
1449 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1450 ocfs2_journal_dirty(handle, bg_bh);
1452 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1453 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1455 goto out_rollback_bg;
1457 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1458 ocfs2_journal_dirty(handle, fe_bh);
1466 bg->bg_next_group = cpu_to_le64(bg_ptr);
1467 out_rollback_prev_bg:
1468 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1472 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1475 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1478 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1479 * value on error. */
1480 static int ocfs2_cluster_group_search(struct inode *inode,
1481 struct buffer_head *group_bh,
1482 u32 bits_wanted, u32 min_bits,
1484 struct ocfs2_suballoc_result *res)
1486 int search = -ENOSPC;
1489 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1490 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1491 unsigned int max_bits, gd_cluster_off;
1493 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1495 if (gd->bg_free_bits_count) {
1496 max_bits = le16_to_cpu(gd->bg_bits);
1498 /* Tail groups in cluster bitmaps which aren't cpg
1499 * aligned are prone to partial extension by a failed
1500 * fs resize. If the file system resize never got to
1501 * update the dinode cluster count, then we don't want
1502 * to trust any clusters past it, regardless of what
1503 * the group descriptor says. */
1504 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1505 le64_to_cpu(gd->bg_blkno));
1506 if ((gd_cluster_off + max_bits) >
1507 OCFS2_I(inode)->ip_clusters) {
1508 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1509 trace_ocfs2_cluster_group_search_wrong_max_bits(
1510 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1511 le16_to_cpu(gd->bg_bits),
1512 OCFS2_I(inode)->ip_clusters, max_bits);
1515 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1516 group_bh, bits_wanted,
1522 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1524 res->sr_bit_offset +
1526 trace_ocfs2_cluster_group_search_max_block(
1527 (unsigned long long)blkoff,
1528 (unsigned long long)max_block);
1529 if (blkoff > max_block)
1533 /* ocfs2_block_group_find_clear_bits() might
1534 * return success, but we still want to return
1535 * -ENOSPC unless it found the minimum number
1537 if (min_bits <= res->sr_bits)
1538 search = 0; /* success */
1539 else if (res->sr_bits) {
1541 * Don't show bits which we'll be returning
1542 * for allocation to the local alloc bitmap.
1544 ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1551 static int ocfs2_block_group_search(struct inode *inode,
1552 struct buffer_head *group_bh,
1553 u32 bits_wanted, u32 min_bits,
1555 struct ocfs2_suballoc_result *res)
1559 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1561 BUG_ON(min_bits != 1);
1562 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1564 if (bg->bg_free_bits_count) {
1565 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1566 group_bh, bits_wanted,
1567 le16_to_cpu(bg->bg_bits),
1569 if (!ret && max_block) {
1570 blkoff = le64_to_cpu(bg->bg_blkno) +
1571 res->sr_bit_offset + res->sr_bits;
1572 trace_ocfs2_block_group_search_max_block(
1573 (unsigned long long)blkoff,
1574 (unsigned long long)max_block);
1575 if (blkoff > max_block)
1583 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1585 struct buffer_head *di_bh,
1591 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1592 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1594 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1595 OCFS2_JOURNAL_ACCESS_WRITE);
1601 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1602 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1603 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1604 ocfs2_journal_dirty(handle, di_bh);
1610 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1611 struct ocfs2_extent_rec *rec,
1612 struct ocfs2_chain_list *cl)
1614 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1615 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1616 unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1618 if (res->sr_bit_offset < bitoff)
1620 if (res->sr_bit_offset >= (bitoff + bitcount))
1622 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1623 (res->sr_bit_offset - bitoff);
1624 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1625 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1629 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1630 struct ocfs2_group_desc *bg,
1631 struct ocfs2_suballoc_result *res)
1634 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
1635 struct ocfs2_extent_rec *rec;
1636 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1637 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1639 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1644 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1645 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
1646 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1647 !bg->bg_list.l_next_free_rec)
1650 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1651 rec = &bg->bg_list.l_recs[i];
1652 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1653 res->sr_bg_blkno = bg_blkno; /* Restore */
1659 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1663 struct ocfs2_suballoc_result *res,
1667 struct buffer_head *group_bh = NULL;
1668 struct ocfs2_group_desc *gd;
1669 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1670 struct inode *alloc_inode = ac->ac_inode;
1672 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1673 res->sr_bg_blkno, &group_bh);
1679 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1680 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1681 ac->ac_max_block, res);
1689 ocfs2_bg_discontig_fix_result(ac, gd, res);
1692 * sr_bg_blkno might have been changed by
1693 * ocfs2_bg_discontig_fix_result
1695 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1697 if (ac->ac_find_loc_only)
1700 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1702 le16_to_cpu(gd->bg_chain));
1708 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1709 res->sr_bit_offset, res->sr_bits);
1714 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1722 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1726 struct ocfs2_suballoc_result *res,
1732 struct inode *alloc_inode = ac->ac_inode;
1733 struct buffer_head *group_bh = NULL;
1734 struct buffer_head *prev_group_bh = NULL;
1735 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1736 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1737 struct ocfs2_group_desc *bg;
1739 chain = ac->ac_chain;
1740 trace_ocfs2_search_chain_begin(
1741 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1742 bits_wanted, chain);
1744 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1745 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1751 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1754 /* for now, the chain search is a bit simplistic. We just use
1755 * the 1st group with any empty bits. */
1756 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1757 bits_wanted, min_bits,
1760 if (!bg->bg_next_group)
1763 brelse(prev_group_bh);
1764 prev_group_bh = NULL;
1766 next_group = le64_to_cpu(bg->bg_next_group);
1767 prev_group_bh = group_bh;
1769 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1770 next_group, &group_bh);
1775 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1778 if (status != -ENOSPC)
1783 trace_ocfs2_search_chain_succ(
1784 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1786 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1788 BUG_ON(res->sr_bits == 0);
1790 ocfs2_bg_discontig_fix_result(ac, bg, res);
1793 * sr_bg_blkno might have been changed by
1794 * ocfs2_bg_discontig_fix_result
1796 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1799 * Keep track of previous block descriptor read. When
1800 * we find a target, if we have read more than X
1801 * number of descriptors, and the target is reasonably
1802 * empty, relink him to top of his chain.
1804 * We've read 0 extra blocks and only send one more to
1805 * the transaction, yet the next guy to search has a
1808 * Do this *after* figuring out how many bits we're taking out
1809 * of our target group.
1811 if (!ac->ac_disable_chain_relink &&
1813 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1814 status = ocfs2_relink_block_group(handle, alloc_inode,
1815 ac->ac_bh, group_bh,
1816 prev_group_bh, chain);
1823 if (ac->ac_find_loc_only)
1826 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1827 ac->ac_bh, res->sr_bits,
1834 status = ocfs2_block_group_set_bits(handle,
1845 trace_ocfs2_search_chain_end(
1846 (unsigned long long)le64_to_cpu(fe->i_blkno),
1850 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1853 brelse(prev_group_bh);
1860 /* will give out up to bits_wanted contiguous bits. */
1861 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1865 struct ocfs2_suballoc_result *res)
1870 u64 hint = ac->ac_last_group;
1871 struct ocfs2_chain_list *cl;
1872 struct ocfs2_dinode *fe;
1874 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1875 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1878 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1880 /* The bh was validated by the inode read during
1881 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1882 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1884 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1885 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1886 ocfs2_error(ac->ac_inode->i_sb,
1887 "Chain allocator dinode %llu has %u used "
1888 "bits but only %u total.",
1889 (unsigned long long)le64_to_cpu(fe->i_blkno),
1890 le32_to_cpu(fe->id1.bitmap1.i_used),
1891 le32_to_cpu(fe->id1.bitmap1.i_total));
1896 res->sr_bg_blkno = hint;
1897 if (res->sr_bg_blkno) {
1898 /* Attempt to short-circuit the usual search mechanism
1899 * by jumping straight to the most recently used
1900 * allocation group. This helps us maintain some
1901 * contiguousness across allocations. */
1902 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1903 min_bits, res, &bits_left);
1906 if (status < 0 && status != -ENOSPC) {
1912 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1914 victim = ocfs2_find_victim_chain(cl);
1915 ac->ac_chain = victim;
1917 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1920 hint = ocfs2_group_from_res(res);
1923 if (status < 0 && status != -ENOSPC) {
1928 trace_ocfs2_claim_suballoc_bits(victim);
1930 /* If we didn't pick a good victim, then just default to
1931 * searching each chain in order. Don't allow chain relinking
1932 * because we only calculate enough journal credits for one
1933 * relink per alloc. */
1934 ac->ac_disable_chain_relink = 1;
1935 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1938 if (!cl->cl_recs[i].c_free)
1942 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1945 hint = ocfs2_group_from_res(res);
1948 if (status < 0 && status != -ENOSPC) {
1955 if (status != -ENOSPC) {
1956 /* If the next search of this group is not likely to
1957 * yield a suitable extent, then we reset the last
1958 * group hint so as to not waste a disk read */
1959 if (bits_left < min_bits)
1960 ac->ac_last_group = 0;
1962 ac->ac_last_group = hint;
1971 int ocfs2_claim_metadata(handle_t *handle,
1972 struct ocfs2_alloc_context *ac,
1975 u16 *suballoc_bit_start,
1976 unsigned int *num_bits,
1980 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1983 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1984 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1986 status = ocfs2_claim_suballoc_bits(ac,
1995 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
1997 *suballoc_loc = res.sr_bg_blkno;
1998 *suballoc_bit_start = res.sr_bit_offset;
1999 *blkno_start = res.sr_blkno;
2000 ac->ac_bits_given += res.sr_bits;
2001 *num_bits = res.sr_bits;
2009 static void ocfs2_init_inode_ac_group(struct inode *dir,
2010 struct buffer_head *parent_di_bh,
2011 struct ocfs2_alloc_context *ac)
2013 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2015 * Try to allocate inodes from some specific group.
2017 * If the parent dir has recorded the last group used in allocation,
2018 * cool, use it. Otherwise if we try to allocate new inode from the
2019 * same slot the parent dir belongs to, use the same chunk.
2021 * We are very careful here to avoid the mistake of setting
2022 * ac_last_group to a group descriptor from a different (unlocked) slot.
2024 if (OCFS2_I(dir)->ip_last_used_group &&
2025 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2026 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2027 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2028 if (di->i_suballoc_loc)
2029 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2031 ac->ac_last_group = ocfs2_which_suballoc_group(
2032 le64_to_cpu(di->i_blkno),
2033 le16_to_cpu(di->i_suballoc_bit));
2037 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2038 struct ocfs2_alloc_context *ac)
2040 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2041 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2044 int ocfs2_find_new_inode_loc(struct inode *dir,
2045 struct buffer_head *parent_fe_bh,
2046 struct ocfs2_alloc_context *ac,
2050 handle_t *handle = NULL;
2051 struct ocfs2_suballoc_result *res;
2054 BUG_ON(ac->ac_bits_given != 0);
2055 BUG_ON(ac->ac_bits_wanted != 1);
2056 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2058 res = kzalloc(sizeof(*res), GFP_NOFS);
2065 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2068 * The handle started here is for chain relink. Alternatively,
2069 * we could just disable relink for these calls.
2071 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2072 if (IS_ERR(handle)) {
2073 ret = PTR_ERR(handle);
2080 * This will instruct ocfs2_claim_suballoc_bits and
2081 * ocfs2_search_one_group to search but save actual allocation
2084 ac->ac_find_loc_only = 1;
2086 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2092 ac->ac_find_loc_priv = res;
2093 *fe_blkno = res->sr_blkno;
2097 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2105 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2107 struct ocfs2_alloc_context *ac,
2114 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2115 struct buffer_head *bg_bh = NULL;
2116 struct ocfs2_group_desc *bg;
2117 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2120 * Since di_blkno is being passed back in, we check for any
2121 * inconsistencies which may have happened between
2122 * calls. These are code bugs as di_blkno is not expected to
2123 * change once returned from ocfs2_find_new_inode_loc()
2125 BUG_ON(res->sr_blkno != di_blkno);
2127 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2128 res->sr_bg_stable_blkno, &bg_bh);
2134 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2135 chain = le16_to_cpu(bg->bg_chain);
2137 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2138 ac->ac_bh, res->sr_bits,
2145 ret = ocfs2_block_group_set_bits(handle,
2156 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2159 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2161 BUG_ON(res->sr_bits != 1);
2163 *suballoc_loc = res->sr_bg_blkno;
2164 *suballoc_bit = res->sr_bit_offset;
2165 ac->ac_bits_given++;
2166 ocfs2_save_inode_ac_group(dir, ac);
2174 int ocfs2_claim_new_inode(handle_t *handle,
2176 struct buffer_head *parent_fe_bh,
2177 struct ocfs2_alloc_context *ac,
2183 struct ocfs2_suballoc_result res;
2186 BUG_ON(ac->ac_bits_given != 0);
2187 BUG_ON(ac->ac_bits_wanted != 1);
2188 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2190 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2192 status = ocfs2_claim_suballoc_bits(ac,
2201 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2203 BUG_ON(res.sr_bits != 1);
2205 *suballoc_loc = res.sr_bg_blkno;
2206 *suballoc_bit = res.sr_bit_offset;
2207 *fe_blkno = res.sr_blkno;
2208 ac->ac_bits_given++;
2209 ocfs2_save_inode_ac_group(dir, ac);
2217 /* translate a group desc. blkno and it's bitmap offset into
2218 * disk cluster offset. */
2219 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2223 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2226 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2228 if (bg_blkno != osb->first_cluster_group_blkno)
2229 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2230 cluster += (u32) bg_bit_off;
2234 /* given a cluster offset, calculate which block group it belongs to
2235 * and return that block offset. */
2236 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2238 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2241 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2243 group_no = cluster / osb->bitmap_cpg;
2245 return osb->first_cluster_group_blkno;
2246 return ocfs2_clusters_to_blocks(inode->i_sb,
2247 group_no * osb->bitmap_cpg);
2250 /* given the block number of a cluster start, calculate which cluster
2251 * group and descriptor bitmap offset that corresponds to. */
2252 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2257 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2258 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2260 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2262 *bg_blkno = ocfs2_which_cluster_group(inode,
2265 if (*bg_blkno == osb->first_cluster_group_blkno)
2266 *bg_bit_off = (u16) data_cluster;
2268 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2269 data_blkno - *bg_blkno);
2273 * min_bits - minimum contiguous chunk from this total allocation we
2274 * can handle. set to what we asked for originally for a full
2275 * contig. allocation, set to '1' to indicate we can deal with extents
2278 int __ocfs2_claim_clusters(handle_t *handle,
2279 struct ocfs2_alloc_context *ac,
2286 unsigned int bits_wanted = max_clusters;
2287 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2288 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2290 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2292 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2293 && ac->ac_which != OCFS2_AC_USE_MAIN);
2295 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2296 WARN_ON(min_clusters > 1);
2298 status = ocfs2_claim_local_alloc_bits(osb,
2305 atomic_inc(&osb->alloc_stats.local_data);
2307 if (min_clusters > (osb->bitmap_cpg - 1)) {
2308 /* The only paths asking for contiguousness
2309 * should know about this already. */
2310 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2311 "group bitmap size %u!\n", min_clusters,
2316 /* clamp the current request down to a realistic size. */
2317 if (bits_wanted > (osb->bitmap_cpg - 1))
2318 bits_wanted = osb->bitmap_cpg - 1;
2320 status = ocfs2_claim_suballoc_bits(ac,
2326 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2328 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2331 atomic_inc(&osb->alloc_stats.bitmap_data);
2332 *num_clusters = res.sr_bits;
2336 if (status != -ENOSPC)
2341 ac->ac_bits_given += *num_clusters;
2349 int ocfs2_claim_clusters(handle_t *handle,
2350 struct ocfs2_alloc_context *ac,
2355 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2357 return __ocfs2_claim_clusters(handle, ac, min_clusters,
2358 bits_wanted, cluster_start, num_clusters);
2361 static int ocfs2_block_group_clear_bits(handle_t *handle,
2362 struct inode *alloc_inode,
2363 struct ocfs2_group_desc *bg,
2364 struct buffer_head *group_bh,
2365 unsigned int bit_off,
2366 unsigned int num_bits,
2367 void (*undo_fn)(unsigned int bit,
2368 unsigned long *bmap))
2372 struct ocfs2_group_desc *undo_bg = NULL;
2374 /* The caller got this descriptor from
2375 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2376 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2378 trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2380 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2381 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2384 OCFS2_JOURNAL_ACCESS_UNDO :
2385 OCFS2_JOURNAL_ACCESS_WRITE);
2392 jbd_lock_bh_state(group_bh);
2393 undo_bg = (struct ocfs2_group_desc *)
2394 bh2jh(group_bh)->b_committed_data;
2400 ocfs2_clear_bit((bit_off + tmp),
2401 (unsigned long *) bg->bg_bitmap);
2403 undo_fn(bit_off + tmp,
2404 (unsigned long *) undo_bg->bg_bitmap);
2406 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2407 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2408 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2409 " count %u but claims %u are freed. num_bits %d",
2410 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2411 le16_to_cpu(bg->bg_bits),
2412 le16_to_cpu(bg->bg_free_bits_count), num_bits);
2417 jbd_unlock_bh_state(group_bh);
2419 ocfs2_journal_dirty(handle, group_bh);
2425 * expects the suballoc inode to already be locked.
2427 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2428 struct inode *alloc_inode,
2429 struct buffer_head *alloc_bh,
2430 unsigned int start_bit,
2433 void (*undo_fn)(unsigned int bit,
2434 unsigned long *bitmap))
2438 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2439 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2440 struct buffer_head *group_bh = NULL;
2441 struct ocfs2_group_desc *group;
2443 /* The alloc_bh comes from ocfs2_free_dinode() or
2444 * ocfs2_free_clusters(). The callers have all locked the
2445 * allocator and gotten alloc_bh from the lock call. This
2446 * validates the dinode buffer. Any corruption that has happened
2448 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2449 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2451 trace_ocfs2_free_suballoc_bits(
2452 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2453 (unsigned long long)bg_blkno,
2456 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2462 group = (struct ocfs2_group_desc *) group_bh->b_data;
2464 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2466 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2468 start_bit, count, undo_fn);
2474 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2475 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2481 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2483 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2484 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2485 ocfs2_journal_dirty(handle, alloc_bh);
2495 int ocfs2_free_suballoc_bits(handle_t *handle,
2496 struct inode *alloc_inode,
2497 struct buffer_head *alloc_bh,
2498 unsigned int start_bit,
2502 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2503 start_bit, bg_blkno, count, NULL);
2506 int ocfs2_free_dinode(handle_t *handle,
2507 struct inode *inode_alloc_inode,
2508 struct buffer_head *inode_alloc_bh,
2509 struct ocfs2_dinode *di)
2511 u64 blk = le64_to_cpu(di->i_blkno);
2512 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2513 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2515 if (di->i_suballoc_loc)
2516 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2517 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2518 inode_alloc_bh, bit, bg_blkno, 1);
2521 static int _ocfs2_free_clusters(handle_t *handle,
2522 struct inode *bitmap_inode,
2523 struct buffer_head *bitmap_bh,
2525 unsigned int num_clusters,
2526 void (*undo_fn)(unsigned int bit,
2527 unsigned long *bitmap))
2532 struct ocfs2_dinode *fe;
2534 /* You can't ever have a contiguous set of clusters
2535 * bigger than a block group bitmap so we never have to worry
2536 * about looping on them.
2537 * This is expensive. We can safely remove once this stuff has
2538 * gotten tested really well. */
2539 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2541 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2543 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2546 trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2547 (unsigned long long)start_blk,
2548 bg_start_bit, num_clusters);
2550 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2551 bg_start_bit, bg_blkno,
2552 num_clusters, undo_fn);
2558 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2567 int ocfs2_free_clusters(handle_t *handle,
2568 struct inode *bitmap_inode,
2569 struct buffer_head *bitmap_bh,
2571 unsigned int num_clusters)
2573 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2574 start_blk, num_clusters,
2579 * Give never-used clusters back to the global bitmap. We don't need
2580 * to protect these bits in the undo buffer.
2582 int ocfs2_release_clusters(handle_t *handle,
2583 struct inode *bitmap_inode,
2584 struct buffer_head *bitmap_bh,
2586 unsigned int num_clusters)
2588 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2589 start_blk, num_clusters,
2593 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2595 printk("Block Group:\n");
2596 printk("bg_signature: %s\n", bg->bg_signature);
2597 printk("bg_size: %u\n", bg->bg_size);
2598 printk("bg_bits: %u\n", bg->bg_bits);
2599 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2600 printk("bg_chain: %u\n", bg->bg_chain);
2601 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
2602 printk("bg_next_group: %llu\n",
2603 (unsigned long long)bg->bg_next_group);
2604 printk("bg_parent_dinode: %llu\n",
2605 (unsigned long long)bg->bg_parent_dinode);
2606 printk("bg_blkno: %llu\n",
2607 (unsigned long long)bg->bg_blkno);
2610 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2614 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2615 printk("i_signature: %s\n", fe->i_signature);
2616 printk("i_size: %llu\n",
2617 (unsigned long long)fe->i_size);
2618 printk("i_clusters: %u\n", fe->i_clusters);
2619 printk("i_generation: %u\n",
2620 le32_to_cpu(fe->i_generation));
2621 printk("id1.bitmap1.i_used: %u\n",
2622 le32_to_cpu(fe->id1.bitmap1.i_used));
2623 printk("id1.bitmap1.i_total: %u\n",
2624 le32_to_cpu(fe->id1.bitmap1.i_total));
2625 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2626 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2627 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2628 printk("id2.i_chain.cl_next_free_rec: %u\n",
2629 fe->id2.i_chain.cl_next_free_rec);
2630 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2631 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2632 fe->id2.i_chain.cl_recs[i].c_free);
2633 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2634 fe->id2.i_chain.cl_recs[i].c_total);
2635 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2636 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2641 * For a given allocation, determine which allocators will need to be
2642 * accessed, and lock them, reserving the appropriate number of bits.
2644 * Sparse file systems call this from ocfs2_write_begin_nolock()
2645 * and ocfs2_allocate_unwritten_extents().
2647 * File systems which don't support holes call this from
2648 * ocfs2_extend_allocation().
2650 int ocfs2_lock_allocators(struct inode *inode,
2651 struct ocfs2_extent_tree *et,
2652 u32 clusters_to_add, u32 extents_to_split,
2653 struct ocfs2_alloc_context **data_ac,
2654 struct ocfs2_alloc_context **meta_ac)
2656 int ret = 0, num_free_extents;
2657 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2658 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2664 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2666 num_free_extents = ocfs2_num_free_extents(osb, et);
2667 if (num_free_extents < 0) {
2668 ret = num_free_extents;
2674 * Sparse allocation file systems need to be more conservative
2675 * with reserving room for expansion - the actual allocation
2676 * happens while we've got a journal handle open so re-taking
2677 * a cluster lock (because we ran out of room for another
2678 * extent) will violate ordering rules.
2680 * Most of the time we'll only be seeing this 1 cluster at a time
2683 * Always lock for any unwritten extents - we might want to
2684 * add blocks during a split.
2686 if (!num_free_extents ||
2687 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2688 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2696 if (clusters_to_add == 0)
2699 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2709 ocfs2_free_alloc_context(*meta_ac);
2714 * We cannot have an error and a non null *data_ac.
2722 * Read the inode specified by blkno to get suballoc_slot and
2725 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2726 u16 *suballoc_slot, u64 *group_blkno,
2730 struct buffer_head *inode_bh = NULL;
2731 struct ocfs2_dinode *inode_fe;
2733 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2735 /* dirty read disk */
2736 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2738 mlog(ML_ERROR, "read block %llu failed %d\n",
2739 (unsigned long long)blkno, status);
2743 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2744 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2745 mlog(ML_ERROR, "invalid inode %llu requested\n",
2746 (unsigned long long)blkno);
2751 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2752 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2753 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2754 (unsigned long long)blkno,
2755 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2761 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2763 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2765 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2776 * test whether bit is SET in allocator bitmap or not. on success, 0
2777 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2778 * is returned and *res is meaningless. Call this after you have
2779 * cluster locked against suballoc, or you may get a result based on
2780 * non-up2date contents
2782 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2783 struct inode *suballoc,
2784 struct buffer_head *alloc_bh,
2785 u64 group_blkno, u64 blkno,
2788 struct ocfs2_dinode *alloc_di;
2789 struct ocfs2_group_desc *group;
2790 struct buffer_head *group_bh = NULL;
2794 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2797 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2798 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2799 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2801 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2806 bg_blkno = group_blkno ? group_blkno :
2807 ocfs2_which_suballoc_group(blkno, bit);
2808 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2811 mlog(ML_ERROR, "read group %llu failed %d\n",
2812 (unsigned long long)bg_blkno, status);
2816 group = (struct ocfs2_group_desc *) group_bh->b_data;
2817 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2828 * Test if the bit representing this inode (blkno) is set in the
2831 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2833 * In the event of failure, a negative value is returned and *res is
2836 * Callers must make sure to hold nfs_sync_lock to prevent
2837 * ocfs2_delete_inode() on another node from accessing the same
2838 * suballocator concurrently.
2840 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2843 u64 group_blkno = 0;
2844 u16 suballoc_bit = 0, suballoc_slot = 0;
2845 struct inode *inode_alloc_inode;
2846 struct buffer_head *alloc_bh = NULL;
2848 trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2850 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2851 &group_blkno, &suballoc_bit);
2853 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2858 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2860 if (!inode_alloc_inode) {
2861 /* the error code could be inaccurate, but we are not able to
2862 * get the correct one. */
2864 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2865 (u32)suballoc_slot);
2869 mutex_lock(&inode_alloc_inode->i_mutex);
2870 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2872 mutex_unlock(&inode_alloc_inode->i_mutex);
2873 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2874 (u32)suballoc_slot, status);
2878 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2879 group_blkno, blkno, suballoc_bit, res);
2881 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2883 ocfs2_inode_unlock(inode_alloc_inode, 0);
2884 mutex_unlock(&inode_alloc_inode->i_mutex);
2886 iput(inode_alloc_inode);