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);
774 ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
778 /* save the new last alloc group so that the caller can cache it. */
779 if (last_alloc_group)
780 *last_alloc_group = ac->ac_last_group;
784 ocfs2_commit_trans(osb, handle);
787 ocfs2_free_alloc_context(ac);
796 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
797 struct ocfs2_alloc_context *ac,
800 u64 *last_alloc_group,
804 u32 bits_wanted = ac->ac_bits_wanted;
805 struct inode *alloc_inode;
806 struct buffer_head *bh = NULL;
807 struct ocfs2_dinode *fe;
810 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
816 mutex_lock(&alloc_inode->i_mutex);
818 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
820 mutex_unlock(&alloc_inode->i_mutex);
827 ac->ac_inode = alloc_inode;
828 ac->ac_alloc_slot = slot;
830 fe = (struct ocfs2_dinode *) bh->b_data;
832 /* The bh was validated by the inode read inside
833 * ocfs2_inode_lock(). Any corruption is a code bug. */
834 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
836 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
837 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
838 (unsigned long long)le64_to_cpu(fe->i_blkno));
843 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
844 le32_to_cpu(fe->id1.bitmap1.i_used);
846 if (bits_wanted > free_bits) {
847 /* cluster bitmap never grows */
848 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
849 trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
855 if (!(flags & ALLOC_NEW_GROUP)) {
856 trace_ocfs2_reserve_suballoc_bits_no_new_group(
857 slot, bits_wanted, free_bits);
862 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
864 last_alloc_group, flags);
866 if (status != -ENOSPC)
870 atomic_inc(&osb->alloc_stats.bg_extends);
872 /* You should never ask for this much metadata */
874 (le32_to_cpu(fe->id1.bitmap1.i_total)
875 - le32_to_cpu(fe->id1.bitmap1.i_used)));
888 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
890 spin_lock(&osb->osb_lock);
891 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
892 spin_unlock(&osb->osb_lock);
893 atomic_set(&osb->s_num_inodes_stolen, 0);
896 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
898 spin_lock(&osb->osb_lock);
899 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
900 spin_unlock(&osb->osb_lock);
901 atomic_set(&osb->s_num_meta_stolen, 0);
904 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
906 ocfs2_init_inode_steal_slot(osb);
907 ocfs2_init_meta_steal_slot(osb);
910 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
912 spin_lock(&osb->osb_lock);
913 if (type == INODE_ALLOC_SYSTEM_INODE)
914 osb->s_inode_steal_slot = slot;
915 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
916 osb->s_meta_steal_slot = slot;
917 spin_unlock(&osb->osb_lock);
920 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
922 int slot = OCFS2_INVALID_SLOT;
924 spin_lock(&osb->osb_lock);
925 if (type == INODE_ALLOC_SYSTEM_INODE)
926 slot = osb->s_inode_steal_slot;
927 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
928 slot = osb->s_meta_steal_slot;
929 spin_unlock(&osb->osb_lock);
934 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
936 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
939 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
941 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
944 static int ocfs2_steal_resource(struct ocfs2_super *osb,
945 struct ocfs2_alloc_context *ac,
948 int i, status = -ENOSPC;
949 int slot = __ocfs2_get_steal_slot(osb, type);
951 /* Start to steal resource from the first slot after ours. */
952 if (slot == OCFS2_INVALID_SLOT)
953 slot = osb->slot_num + 1;
955 for (i = 0; i < osb->max_slots; i++, slot++) {
956 if (slot == osb->max_slots)
959 if (slot == osb->slot_num)
962 status = ocfs2_reserve_suballoc_bits(osb, ac,
965 NOT_ALLOC_NEW_GROUP);
967 __ocfs2_set_steal_slot(osb, slot, type);
971 ocfs2_free_ac_resource(ac);
977 static int ocfs2_steal_inode(struct ocfs2_super *osb,
978 struct ocfs2_alloc_context *ac)
980 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
983 static int ocfs2_steal_meta(struct ocfs2_super *osb,
984 struct ocfs2_alloc_context *ac)
986 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
989 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
991 struct ocfs2_alloc_context **ac)
994 int slot = ocfs2_get_meta_steal_slot(osb);
996 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1003 (*ac)->ac_bits_wanted = blocks;
1004 (*ac)->ac_which = OCFS2_AC_USE_META;
1005 (*ac)->ac_group_search = ocfs2_block_group_search;
1007 if (slot != OCFS2_INVALID_SLOT &&
1008 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1011 atomic_set(&osb->s_num_meta_stolen, 0);
1012 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
1013 EXTENT_ALLOC_SYSTEM_INODE,
1014 (u32)osb->slot_num, NULL,
1015 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1020 if (slot != OCFS2_INVALID_SLOT)
1021 ocfs2_init_meta_steal_slot(osb);
1023 } else if (status < 0 && status != -ENOSPC) {
1028 ocfs2_free_ac_resource(*ac);
1031 status = ocfs2_steal_meta(osb, *ac);
1032 atomic_inc(&osb->s_num_meta_stolen);
1034 if (status != -ENOSPC)
1041 if ((status < 0) && *ac) {
1042 ocfs2_free_alloc_context(*ac);
1051 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1052 struct ocfs2_extent_list *root_el,
1053 struct ocfs2_alloc_context **ac)
1055 return ocfs2_reserve_new_metadata_blocks(osb,
1056 ocfs2_extend_meta_needed(root_el),
1060 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1061 struct ocfs2_alloc_context **ac)
1064 int slot = ocfs2_get_inode_steal_slot(osb);
1067 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1074 (*ac)->ac_bits_wanted = 1;
1075 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1077 (*ac)->ac_group_search = ocfs2_block_group_search;
1080 * stat(2) can't handle i_ino > 32bits, so we tell the
1081 * lower levels not to allocate us a block group past that
1082 * limit. The 'inode64' mount option avoids this behavior.
1084 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1085 (*ac)->ac_max_block = (u32)~0U;
1088 * slot is set when we successfully steal inode from other nodes.
1089 * It is reset in 3 places:
1090 * 1. when we flush the truncate log
1091 * 2. when we complete local alloc recovery.
1092 * 3. when we successfully allocate from our own slot.
1093 * After it is set, we will go on stealing inodes until we find the
1094 * need to check our slots to see whether there is some space for us.
1096 if (slot != OCFS2_INVALID_SLOT &&
1097 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1100 atomic_set(&osb->s_num_inodes_stolen, 0);
1101 alloc_group = osb->osb_inode_alloc_group;
1102 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1103 INODE_ALLOC_SYSTEM_INODE,
1107 ALLOC_GROUPS_FROM_GLOBAL);
1111 spin_lock(&osb->osb_lock);
1112 osb->osb_inode_alloc_group = alloc_group;
1113 spin_unlock(&osb->osb_lock);
1114 trace_ocfs2_reserve_new_inode_new_group(
1115 (unsigned long long)alloc_group);
1118 * Some inodes must be freed by us, so try to allocate
1119 * from our own next time.
1121 if (slot != OCFS2_INVALID_SLOT)
1122 ocfs2_init_inode_steal_slot(osb);
1124 } else if (status < 0 && status != -ENOSPC) {
1129 ocfs2_free_ac_resource(*ac);
1132 status = ocfs2_steal_inode(osb, *ac);
1133 atomic_inc(&osb->s_num_inodes_stolen);
1135 if (status != -ENOSPC)
1142 if ((status < 0) && *ac) {
1143 ocfs2_free_alloc_context(*ac);
1152 /* local alloc code has to do the same thing, so rather than do this
1154 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1155 struct ocfs2_alloc_context *ac)
1159 ac->ac_which = OCFS2_AC_USE_MAIN;
1160 ac->ac_group_search = ocfs2_cluster_group_search;
1162 status = ocfs2_reserve_suballoc_bits(osb, ac,
1163 GLOBAL_BITMAP_SYSTEM_INODE,
1164 OCFS2_INVALID_SLOT, NULL,
1166 if (status < 0 && status != -ENOSPC) {
1175 /* Callers don't need to care which bitmap (local alloc or main) to
1176 * use so we figure it out for them, but unfortunately this clutters
1178 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1179 u32 bits_wanted, u64 max_block,
1181 struct ocfs2_alloc_context **ac)
1185 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1192 (*ac)->ac_bits_wanted = bits_wanted;
1193 (*ac)->ac_max_block = max_block;
1196 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1197 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1198 status = ocfs2_reserve_local_alloc_bits(osb,
1201 if ((status < 0) && (status != -ENOSPC)) {
1207 if (status == -ENOSPC) {
1208 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1210 if (status != -ENOSPC)
1218 if ((status < 0) && *ac) {
1219 ocfs2_free_alloc_context(*ac);
1228 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1230 struct ocfs2_alloc_context **ac)
1232 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1233 ALLOC_NEW_GROUP, ac);
1237 * More or less lifted from ext3. I'll leave their description below:
1239 * "For ext3 allocations, we must not reuse any blocks which are
1240 * allocated in the bitmap buffer's "last committed data" copy. This
1241 * prevents deletes from freeing up the page for reuse until we have
1242 * committed the delete transaction.
1244 * If we didn't do this, then deleting something and reallocating it as
1245 * data would allow the old block to be overwritten before the
1246 * transaction committed (because we force data to disk before commit).
1247 * This would lead to corruption if we crashed between overwriting the
1248 * data and committing the delete.
1250 * @@@ We may want to make this allocation behaviour conditional on
1251 * data-writes at some point, and disable it for metadata allocations or
1252 * sync-data inodes."
1254 * Note: OCFS2 already does this differently for metadata vs data
1255 * allocations, as those bitmaps are separate and undo access is never
1256 * called on a metadata group descriptor.
1258 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1261 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1264 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1267 if (!buffer_jbd(bg_bh))
1270 jbd_lock_bh_state(bg_bh);
1271 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1273 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1276 jbd_unlock_bh_state(bg_bh);
1281 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1282 struct buffer_head *bg_bh,
1283 unsigned int bits_wanted,
1284 unsigned int total_bits,
1285 struct ocfs2_suballoc_result *res)
1288 u16 best_offset, best_size;
1289 int offset, start, found, status = 0;
1290 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1292 /* Callers got this descriptor from
1293 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1294 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1296 found = start = best_offset = best_size = 0;
1297 bitmap = bg->bg_bitmap;
1299 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1300 if (offset == total_bits)
1303 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1304 /* We found a zero, but we can't use it as it
1305 * hasn't been put to disk yet! */
1308 } else if (offset == start) {
1309 /* we found a zero */
1311 /* move start to the next bit to test */
1314 /* got a zero after some ones */
1318 if (found > best_size) {
1320 best_offset = start - found;
1322 /* we got everything we needed */
1323 if (found == bits_wanted) {
1324 /* mlog(0, "Found it all!\n"); */
1330 res->sr_bit_offset = best_offset;
1331 res->sr_bits = best_size;
1334 /* No error log here -- see the comment above
1335 * ocfs2_test_bg_bit_allocatable */
1341 int ocfs2_block_group_set_bits(handle_t *handle,
1342 struct inode *alloc_inode,
1343 struct ocfs2_group_desc *bg,
1344 struct buffer_head *group_bh,
1345 unsigned int bit_off,
1346 unsigned int num_bits)
1349 void *bitmap = bg->bg_bitmap;
1350 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1352 /* All callers get the descriptor via
1353 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1354 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1355 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1357 trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1359 if (ocfs2_is_cluster_bitmap(alloc_inode))
1360 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1362 status = ocfs2_journal_access_gd(handle,
1363 INODE_CACHE(alloc_inode),
1371 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1372 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1373 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1374 " count %u but claims %u are freed. num_bits %d",
1375 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1376 le16_to_cpu(bg->bg_bits),
1377 le16_to_cpu(bg->bg_free_bits_count), num_bits);
1381 ocfs2_set_bit(bit_off++, bitmap);
1383 ocfs2_journal_dirty(handle, group_bh);
1389 /* find the one with the most empty bits */
1390 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1394 BUG_ON(!cl->cl_next_free_rec);
1397 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1398 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1399 le32_to_cpu(cl->cl_recs[best].c_free))
1404 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1408 static int ocfs2_relink_block_group(handle_t *handle,
1409 struct inode *alloc_inode,
1410 struct buffer_head *fe_bh,
1411 struct buffer_head *bg_bh,
1412 struct buffer_head *prev_bg_bh,
1416 /* there is a really tiny chance the journal calls could fail,
1417 * but we wouldn't want inconsistent blocks in *any* case. */
1418 u64 bg_ptr, prev_bg_ptr;
1419 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1420 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1421 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1423 /* The caller got these descriptors from
1424 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1425 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1426 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1428 trace_ocfs2_relink_block_group(
1429 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1430 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1431 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1433 bg_ptr = le64_to_cpu(bg->bg_next_group);
1434 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1436 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1438 OCFS2_JOURNAL_ACCESS_WRITE);
1442 prev_bg->bg_next_group = bg->bg_next_group;
1443 ocfs2_journal_dirty(handle, prev_bg_bh);
1445 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1446 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1448 goto out_rollback_prev_bg;
1450 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1451 ocfs2_journal_dirty(handle, bg_bh);
1453 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1454 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1456 goto out_rollback_bg;
1458 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1459 ocfs2_journal_dirty(handle, fe_bh);
1467 bg->bg_next_group = cpu_to_le64(bg_ptr);
1468 out_rollback_prev_bg:
1469 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1473 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1476 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1479 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1480 * value on error. */
1481 static int ocfs2_cluster_group_search(struct inode *inode,
1482 struct buffer_head *group_bh,
1483 u32 bits_wanted, u32 min_bits,
1485 struct ocfs2_suballoc_result *res)
1487 int search = -ENOSPC;
1490 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1492 unsigned int max_bits, gd_cluster_off;
1494 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1496 if (gd->bg_free_bits_count) {
1497 max_bits = le16_to_cpu(gd->bg_bits);
1499 /* Tail groups in cluster bitmaps which aren't cpg
1500 * aligned are prone to partial extension by a failed
1501 * fs resize. If the file system resize never got to
1502 * update the dinode cluster count, then we don't want
1503 * to trust any clusters past it, regardless of what
1504 * the group descriptor says. */
1505 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1506 le64_to_cpu(gd->bg_blkno));
1507 if ((gd_cluster_off + max_bits) >
1508 OCFS2_I(inode)->ip_clusters) {
1509 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1510 trace_ocfs2_cluster_group_search_wrong_max_bits(
1511 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1512 le16_to_cpu(gd->bg_bits),
1513 OCFS2_I(inode)->ip_clusters, max_bits);
1516 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1517 group_bh, bits_wanted,
1523 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1525 res->sr_bit_offset +
1527 trace_ocfs2_cluster_group_search_max_block(
1528 (unsigned long long)blkoff,
1529 (unsigned long long)max_block);
1530 if (blkoff > max_block)
1534 /* ocfs2_block_group_find_clear_bits() might
1535 * return success, but we still want to return
1536 * -ENOSPC unless it found the minimum number
1538 if (min_bits <= res->sr_bits)
1539 search = 0; /* success */
1540 else if (res->sr_bits) {
1542 * Don't show bits which we'll be returning
1543 * for allocation to the local alloc bitmap.
1545 ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1552 static int ocfs2_block_group_search(struct inode *inode,
1553 struct buffer_head *group_bh,
1554 u32 bits_wanted, u32 min_bits,
1556 struct ocfs2_suballoc_result *res)
1560 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1562 BUG_ON(min_bits != 1);
1563 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1565 if (bg->bg_free_bits_count) {
1566 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1567 group_bh, bits_wanted,
1568 le16_to_cpu(bg->bg_bits),
1570 if (!ret && max_block) {
1571 blkoff = le64_to_cpu(bg->bg_blkno) +
1572 res->sr_bit_offset + res->sr_bits;
1573 trace_ocfs2_block_group_search_max_block(
1574 (unsigned long long)blkoff,
1575 (unsigned long long)max_block);
1576 if (blkoff > max_block)
1584 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1586 struct buffer_head *di_bh,
1592 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1593 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1595 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1596 OCFS2_JOURNAL_ACCESS_WRITE);
1602 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1603 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1604 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1605 ocfs2_journal_dirty(handle, di_bh);
1611 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1612 struct ocfs2_extent_rec *rec,
1613 struct ocfs2_chain_list *cl)
1615 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1616 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1617 unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1619 if (res->sr_bit_offset < bitoff)
1621 if (res->sr_bit_offset >= (bitoff + bitcount))
1623 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1624 (res->sr_bit_offset - bitoff);
1625 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1626 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1630 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1631 struct ocfs2_group_desc *bg,
1632 struct ocfs2_suballoc_result *res)
1635 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
1636 struct ocfs2_extent_rec *rec;
1637 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1638 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1640 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1645 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1646 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
1647 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1648 !bg->bg_list.l_next_free_rec)
1651 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1652 rec = &bg->bg_list.l_recs[i];
1653 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1654 res->sr_bg_blkno = bg_blkno; /* Restore */
1660 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1664 struct ocfs2_suballoc_result *res,
1668 struct buffer_head *group_bh = NULL;
1669 struct ocfs2_group_desc *gd;
1670 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1671 struct inode *alloc_inode = ac->ac_inode;
1673 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1674 res->sr_bg_blkno, &group_bh);
1680 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1681 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1682 ac->ac_max_block, res);
1690 ocfs2_bg_discontig_fix_result(ac, gd, res);
1693 * sr_bg_blkno might have been changed by
1694 * ocfs2_bg_discontig_fix_result
1696 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1698 if (ac->ac_find_loc_only)
1701 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1703 le16_to_cpu(gd->bg_chain));
1709 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1710 res->sr_bit_offset, res->sr_bits);
1715 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1723 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1727 struct ocfs2_suballoc_result *res,
1733 struct inode *alloc_inode = ac->ac_inode;
1734 struct buffer_head *group_bh = NULL;
1735 struct buffer_head *prev_group_bh = NULL;
1736 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1737 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1738 struct ocfs2_group_desc *bg;
1740 chain = ac->ac_chain;
1741 trace_ocfs2_search_chain_begin(
1742 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1743 bits_wanted, chain);
1745 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1746 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1752 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1755 /* for now, the chain search is a bit simplistic. We just use
1756 * the 1st group with any empty bits. */
1757 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1758 bits_wanted, min_bits,
1761 if (!bg->bg_next_group)
1764 brelse(prev_group_bh);
1765 prev_group_bh = NULL;
1767 next_group = le64_to_cpu(bg->bg_next_group);
1768 prev_group_bh = group_bh;
1770 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1771 next_group, &group_bh);
1776 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1779 if (status != -ENOSPC)
1784 trace_ocfs2_search_chain_succ(
1785 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1787 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1789 BUG_ON(res->sr_bits == 0);
1791 ocfs2_bg_discontig_fix_result(ac, bg, res);
1794 * sr_bg_blkno might have been changed by
1795 * ocfs2_bg_discontig_fix_result
1797 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1800 * Keep track of previous block descriptor read. When
1801 * we find a target, if we have read more than X
1802 * number of descriptors, and the target is reasonably
1803 * empty, relink him to top of his chain.
1805 * We've read 0 extra blocks and only send one more to
1806 * the transaction, yet the next guy to search has a
1809 * Do this *after* figuring out how many bits we're taking out
1810 * of our target group.
1812 if (!ac->ac_disable_chain_relink &&
1814 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1815 status = ocfs2_relink_block_group(handle, alloc_inode,
1816 ac->ac_bh, group_bh,
1817 prev_group_bh, chain);
1824 if (ac->ac_find_loc_only)
1827 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1828 ac->ac_bh, res->sr_bits,
1835 status = ocfs2_block_group_set_bits(handle,
1846 trace_ocfs2_search_chain_end(
1847 (unsigned long long)le64_to_cpu(fe->i_blkno),
1851 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1854 brelse(prev_group_bh);
1861 /* will give out up to bits_wanted contiguous bits. */
1862 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1866 struct ocfs2_suballoc_result *res)
1871 u64 hint = ac->ac_last_group;
1872 struct ocfs2_chain_list *cl;
1873 struct ocfs2_dinode *fe;
1875 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1876 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1879 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1881 /* The bh was validated by the inode read during
1882 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1883 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1885 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1886 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1887 ocfs2_error(ac->ac_inode->i_sb,
1888 "Chain allocator dinode %llu has %u used "
1889 "bits but only %u total.",
1890 (unsigned long long)le64_to_cpu(fe->i_blkno),
1891 le32_to_cpu(fe->id1.bitmap1.i_used),
1892 le32_to_cpu(fe->id1.bitmap1.i_total));
1897 res->sr_bg_blkno = hint;
1898 if (res->sr_bg_blkno) {
1899 /* Attempt to short-circuit the usual search mechanism
1900 * by jumping straight to the most recently used
1901 * allocation group. This helps us maintain some
1902 * contiguousness across allocations. */
1903 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1904 min_bits, res, &bits_left);
1907 if (status < 0 && status != -ENOSPC) {
1913 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1915 victim = ocfs2_find_victim_chain(cl);
1916 ac->ac_chain = victim;
1918 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1921 hint = ocfs2_group_from_res(res);
1924 if (status < 0 && status != -ENOSPC) {
1929 trace_ocfs2_claim_suballoc_bits(victim);
1931 /* If we didn't pick a good victim, then just default to
1932 * searching each chain in order. Don't allow chain relinking
1933 * because we only calculate enough journal credits for one
1934 * relink per alloc. */
1935 ac->ac_disable_chain_relink = 1;
1936 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1939 if (!cl->cl_recs[i].c_free)
1943 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1946 hint = ocfs2_group_from_res(res);
1949 if (status < 0 && status != -ENOSPC) {
1956 if (status != -ENOSPC) {
1957 /* If the next search of this group is not likely to
1958 * yield a suitable extent, then we reset the last
1959 * group hint so as to not waste a disk read */
1960 if (bits_left < min_bits)
1961 ac->ac_last_group = 0;
1963 ac->ac_last_group = hint;
1972 int ocfs2_claim_metadata(handle_t *handle,
1973 struct ocfs2_alloc_context *ac,
1976 u16 *suballoc_bit_start,
1977 unsigned int *num_bits,
1981 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1984 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1985 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1987 status = ocfs2_claim_suballoc_bits(ac,
1996 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
1998 *suballoc_loc = res.sr_bg_blkno;
1999 *suballoc_bit_start = res.sr_bit_offset;
2000 *blkno_start = res.sr_blkno;
2001 ac->ac_bits_given += res.sr_bits;
2002 *num_bits = res.sr_bits;
2010 static void ocfs2_init_inode_ac_group(struct inode *dir,
2011 struct buffer_head *parent_di_bh,
2012 struct ocfs2_alloc_context *ac)
2014 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2016 * Try to allocate inodes from some specific group.
2018 * If the parent dir has recorded the last group used in allocation,
2019 * cool, use it. Otherwise if we try to allocate new inode from the
2020 * same slot the parent dir belongs to, use the same chunk.
2022 * We are very careful here to avoid the mistake of setting
2023 * ac_last_group to a group descriptor from a different (unlocked) slot.
2025 if (OCFS2_I(dir)->ip_last_used_group &&
2026 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2027 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2028 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2029 if (di->i_suballoc_loc)
2030 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2032 ac->ac_last_group = ocfs2_which_suballoc_group(
2033 le64_to_cpu(di->i_blkno),
2034 le16_to_cpu(di->i_suballoc_bit));
2038 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2039 struct ocfs2_alloc_context *ac)
2041 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2042 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2045 int ocfs2_find_new_inode_loc(struct inode *dir,
2046 struct buffer_head *parent_fe_bh,
2047 struct ocfs2_alloc_context *ac,
2051 handle_t *handle = NULL;
2052 struct ocfs2_suballoc_result *res;
2055 BUG_ON(ac->ac_bits_given != 0);
2056 BUG_ON(ac->ac_bits_wanted != 1);
2057 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2059 res = kzalloc(sizeof(*res), GFP_NOFS);
2066 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2069 * The handle started here is for chain relink. Alternatively,
2070 * we could just disable relink for these calls.
2072 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2073 if (IS_ERR(handle)) {
2074 ret = PTR_ERR(handle);
2081 * This will instruct ocfs2_claim_suballoc_bits and
2082 * ocfs2_search_one_group to search but save actual allocation
2085 ac->ac_find_loc_only = 1;
2087 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2093 ac->ac_find_loc_priv = res;
2094 *fe_blkno = res->sr_blkno;
2095 ocfs2_update_inode_fsync_trans(handle, dir, 0);
2098 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2106 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2108 struct ocfs2_alloc_context *ac,
2115 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2116 struct buffer_head *bg_bh = NULL;
2117 struct ocfs2_group_desc *bg;
2118 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2121 * Since di_blkno is being passed back in, we check for any
2122 * inconsistencies which may have happened between
2123 * calls. These are code bugs as di_blkno is not expected to
2124 * change once returned from ocfs2_find_new_inode_loc()
2126 BUG_ON(res->sr_blkno != di_blkno);
2128 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2129 res->sr_bg_stable_blkno, &bg_bh);
2135 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2136 chain = le16_to_cpu(bg->bg_chain);
2138 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2139 ac->ac_bh, res->sr_bits,
2146 ret = ocfs2_block_group_set_bits(handle,
2157 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2160 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2162 BUG_ON(res->sr_bits != 1);
2164 *suballoc_loc = res->sr_bg_blkno;
2165 *suballoc_bit = res->sr_bit_offset;
2166 ac->ac_bits_given++;
2167 ocfs2_save_inode_ac_group(dir, ac);
2175 int ocfs2_claim_new_inode(handle_t *handle,
2177 struct buffer_head *parent_fe_bh,
2178 struct ocfs2_alloc_context *ac,
2184 struct ocfs2_suballoc_result res;
2187 BUG_ON(ac->ac_bits_given != 0);
2188 BUG_ON(ac->ac_bits_wanted != 1);
2189 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2191 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2193 status = ocfs2_claim_suballoc_bits(ac,
2202 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2204 BUG_ON(res.sr_bits != 1);
2206 *suballoc_loc = res.sr_bg_blkno;
2207 *suballoc_bit = res.sr_bit_offset;
2208 *fe_blkno = res.sr_blkno;
2209 ac->ac_bits_given++;
2210 ocfs2_save_inode_ac_group(dir, ac);
2218 /* translate a group desc. blkno and it's bitmap offset into
2219 * disk cluster offset. */
2220 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2224 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2227 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2229 if (bg_blkno != osb->first_cluster_group_blkno)
2230 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2231 cluster += (u32) bg_bit_off;
2235 /* given a cluster offset, calculate which block group it belongs to
2236 * and return that block offset. */
2237 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2239 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2242 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2244 group_no = cluster / osb->bitmap_cpg;
2246 return osb->first_cluster_group_blkno;
2247 return ocfs2_clusters_to_blocks(inode->i_sb,
2248 group_no * osb->bitmap_cpg);
2251 /* given the block number of a cluster start, calculate which cluster
2252 * group and descriptor bitmap offset that corresponds to. */
2253 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2258 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2259 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2261 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2263 *bg_blkno = ocfs2_which_cluster_group(inode,
2266 if (*bg_blkno == osb->first_cluster_group_blkno)
2267 *bg_bit_off = (u16) data_cluster;
2269 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2270 data_blkno - *bg_blkno);
2274 * min_bits - minimum contiguous chunk from this total allocation we
2275 * can handle. set to what we asked for originally for a full
2276 * contig. allocation, set to '1' to indicate we can deal with extents
2279 int __ocfs2_claim_clusters(handle_t *handle,
2280 struct ocfs2_alloc_context *ac,
2287 unsigned int bits_wanted = max_clusters;
2288 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2289 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2291 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2293 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2294 && ac->ac_which != OCFS2_AC_USE_MAIN);
2296 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2297 WARN_ON(min_clusters > 1);
2299 status = ocfs2_claim_local_alloc_bits(osb,
2306 atomic_inc(&osb->alloc_stats.local_data);
2308 if (min_clusters > (osb->bitmap_cpg - 1)) {
2309 /* The only paths asking for contiguousness
2310 * should know about this already. */
2311 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2312 "group bitmap size %u!\n", min_clusters,
2317 /* clamp the current request down to a realistic size. */
2318 if (bits_wanted > (osb->bitmap_cpg - 1))
2319 bits_wanted = osb->bitmap_cpg - 1;
2321 status = ocfs2_claim_suballoc_bits(ac,
2327 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2329 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2332 atomic_inc(&osb->alloc_stats.bitmap_data);
2333 *num_clusters = res.sr_bits;
2337 if (status != -ENOSPC)
2342 ac->ac_bits_given += *num_clusters;
2350 int ocfs2_claim_clusters(handle_t *handle,
2351 struct ocfs2_alloc_context *ac,
2356 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2358 return __ocfs2_claim_clusters(handle, ac, min_clusters,
2359 bits_wanted, cluster_start, num_clusters);
2362 static int ocfs2_block_group_clear_bits(handle_t *handle,
2363 struct inode *alloc_inode,
2364 struct ocfs2_group_desc *bg,
2365 struct buffer_head *group_bh,
2366 unsigned int bit_off,
2367 unsigned int num_bits,
2368 void (*undo_fn)(unsigned int bit,
2369 unsigned long *bmap))
2373 struct ocfs2_group_desc *undo_bg = NULL;
2375 /* The caller got this descriptor from
2376 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2377 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2379 trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2381 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2382 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2385 OCFS2_JOURNAL_ACCESS_UNDO :
2386 OCFS2_JOURNAL_ACCESS_WRITE);
2393 jbd_lock_bh_state(group_bh);
2394 undo_bg = (struct ocfs2_group_desc *)
2395 bh2jh(group_bh)->b_committed_data;
2401 ocfs2_clear_bit((bit_off + tmp),
2402 (unsigned long *) bg->bg_bitmap);
2404 undo_fn(bit_off + tmp,
2405 (unsigned long *) undo_bg->bg_bitmap);
2407 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2408 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2409 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2410 " count %u but claims %u are freed. num_bits %d",
2411 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2412 le16_to_cpu(bg->bg_bits),
2413 le16_to_cpu(bg->bg_free_bits_count), num_bits);
2418 jbd_unlock_bh_state(group_bh);
2420 ocfs2_journal_dirty(handle, group_bh);
2426 * expects the suballoc inode to already be locked.
2428 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2429 struct inode *alloc_inode,
2430 struct buffer_head *alloc_bh,
2431 unsigned int start_bit,
2434 void (*undo_fn)(unsigned int bit,
2435 unsigned long *bitmap))
2439 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2440 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2441 struct buffer_head *group_bh = NULL;
2442 struct ocfs2_group_desc *group;
2444 /* The alloc_bh comes from ocfs2_free_dinode() or
2445 * ocfs2_free_clusters(). The callers have all locked the
2446 * allocator and gotten alloc_bh from the lock call. This
2447 * validates the dinode buffer. Any corruption that has happened
2449 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2450 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2452 trace_ocfs2_free_suballoc_bits(
2453 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2454 (unsigned long long)bg_blkno,
2457 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2463 group = (struct ocfs2_group_desc *) group_bh->b_data;
2465 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2467 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2469 start_bit, count, undo_fn);
2475 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2476 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2482 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2484 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2485 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2486 ocfs2_journal_dirty(handle, alloc_bh);
2496 int ocfs2_free_suballoc_bits(handle_t *handle,
2497 struct inode *alloc_inode,
2498 struct buffer_head *alloc_bh,
2499 unsigned int start_bit,
2503 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2504 start_bit, bg_blkno, count, NULL);
2507 int ocfs2_free_dinode(handle_t *handle,
2508 struct inode *inode_alloc_inode,
2509 struct buffer_head *inode_alloc_bh,
2510 struct ocfs2_dinode *di)
2512 u64 blk = le64_to_cpu(di->i_blkno);
2513 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2514 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2516 if (di->i_suballoc_loc)
2517 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2518 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2519 inode_alloc_bh, bit, bg_blkno, 1);
2522 static int _ocfs2_free_clusters(handle_t *handle,
2523 struct inode *bitmap_inode,
2524 struct buffer_head *bitmap_bh,
2526 unsigned int num_clusters,
2527 void (*undo_fn)(unsigned int bit,
2528 unsigned long *bitmap))
2533 struct ocfs2_dinode *fe;
2535 /* You can't ever have a contiguous set of clusters
2536 * bigger than a block group bitmap so we never have to worry
2537 * about looping on them.
2538 * This is expensive. We can safely remove once this stuff has
2539 * gotten tested really well. */
2540 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2542 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2544 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2547 trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2548 (unsigned long long)start_blk,
2549 bg_start_bit, num_clusters);
2551 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2552 bg_start_bit, bg_blkno,
2553 num_clusters, undo_fn);
2559 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2568 int ocfs2_free_clusters(handle_t *handle,
2569 struct inode *bitmap_inode,
2570 struct buffer_head *bitmap_bh,
2572 unsigned int num_clusters)
2574 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2575 start_blk, num_clusters,
2580 * Give never-used clusters back to the global bitmap. We don't need
2581 * to protect these bits in the undo buffer.
2583 int ocfs2_release_clusters(handle_t *handle,
2584 struct inode *bitmap_inode,
2585 struct buffer_head *bitmap_bh,
2587 unsigned int num_clusters)
2589 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2590 start_blk, num_clusters,
2594 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2596 printk("Block Group:\n");
2597 printk("bg_signature: %s\n", bg->bg_signature);
2598 printk("bg_size: %u\n", bg->bg_size);
2599 printk("bg_bits: %u\n", bg->bg_bits);
2600 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2601 printk("bg_chain: %u\n", bg->bg_chain);
2602 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
2603 printk("bg_next_group: %llu\n",
2604 (unsigned long long)bg->bg_next_group);
2605 printk("bg_parent_dinode: %llu\n",
2606 (unsigned long long)bg->bg_parent_dinode);
2607 printk("bg_blkno: %llu\n",
2608 (unsigned long long)bg->bg_blkno);
2611 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2615 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2616 printk("i_signature: %s\n", fe->i_signature);
2617 printk("i_size: %llu\n",
2618 (unsigned long long)fe->i_size);
2619 printk("i_clusters: %u\n", fe->i_clusters);
2620 printk("i_generation: %u\n",
2621 le32_to_cpu(fe->i_generation));
2622 printk("id1.bitmap1.i_used: %u\n",
2623 le32_to_cpu(fe->id1.bitmap1.i_used));
2624 printk("id1.bitmap1.i_total: %u\n",
2625 le32_to_cpu(fe->id1.bitmap1.i_total));
2626 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2627 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2628 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2629 printk("id2.i_chain.cl_next_free_rec: %u\n",
2630 fe->id2.i_chain.cl_next_free_rec);
2631 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2632 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2633 fe->id2.i_chain.cl_recs[i].c_free);
2634 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2635 fe->id2.i_chain.cl_recs[i].c_total);
2636 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2637 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2642 * For a given allocation, determine which allocators will need to be
2643 * accessed, and lock them, reserving the appropriate number of bits.
2645 * Sparse file systems call this from ocfs2_write_begin_nolock()
2646 * and ocfs2_allocate_unwritten_extents().
2648 * File systems which don't support holes call this from
2649 * ocfs2_extend_allocation().
2651 int ocfs2_lock_allocators(struct inode *inode,
2652 struct ocfs2_extent_tree *et,
2653 u32 clusters_to_add, u32 extents_to_split,
2654 struct ocfs2_alloc_context **data_ac,
2655 struct ocfs2_alloc_context **meta_ac)
2657 int ret = 0, num_free_extents;
2658 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2659 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2665 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2667 num_free_extents = ocfs2_num_free_extents(osb, et);
2668 if (num_free_extents < 0) {
2669 ret = num_free_extents;
2675 * Sparse allocation file systems need to be more conservative
2676 * with reserving room for expansion - the actual allocation
2677 * happens while we've got a journal handle open so re-taking
2678 * a cluster lock (because we ran out of room for another
2679 * extent) will violate ordering rules.
2681 * Most of the time we'll only be seeing this 1 cluster at a time
2684 * Always lock for any unwritten extents - we might want to
2685 * add blocks during a split.
2687 if (!num_free_extents ||
2688 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2689 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2697 if (clusters_to_add == 0)
2700 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2710 ocfs2_free_alloc_context(*meta_ac);
2715 * We cannot have an error and a non null *data_ac.
2723 * Read the inode specified by blkno to get suballoc_slot and
2726 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2727 u16 *suballoc_slot, u64 *group_blkno,
2731 struct buffer_head *inode_bh = NULL;
2732 struct ocfs2_dinode *inode_fe;
2734 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2736 /* dirty read disk */
2737 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2739 mlog(ML_ERROR, "read block %llu failed %d\n",
2740 (unsigned long long)blkno, status);
2744 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2745 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2746 mlog(ML_ERROR, "invalid inode %llu requested\n",
2747 (unsigned long long)blkno);
2752 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2753 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2754 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2755 (unsigned long long)blkno,
2756 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2762 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2764 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2766 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2777 * test whether bit is SET in allocator bitmap or not. on success, 0
2778 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2779 * is returned and *res is meaningless. Call this after you have
2780 * cluster locked against suballoc, or you may get a result based on
2781 * non-up2date contents
2783 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2784 struct inode *suballoc,
2785 struct buffer_head *alloc_bh,
2786 u64 group_blkno, u64 blkno,
2789 struct ocfs2_dinode *alloc_di;
2790 struct ocfs2_group_desc *group;
2791 struct buffer_head *group_bh = NULL;
2795 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2798 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2799 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2800 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2802 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2807 bg_blkno = group_blkno ? group_blkno :
2808 ocfs2_which_suballoc_group(blkno, bit);
2809 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2812 mlog(ML_ERROR, "read group %llu failed %d\n",
2813 (unsigned long long)bg_blkno, status);
2817 group = (struct ocfs2_group_desc *) group_bh->b_data;
2818 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2829 * Test if the bit representing this inode (blkno) is set in the
2832 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2834 * In the event of failure, a negative value is returned and *res is
2837 * Callers must make sure to hold nfs_sync_lock to prevent
2838 * ocfs2_delete_inode() on another node from accessing the same
2839 * suballocator concurrently.
2841 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2844 u64 group_blkno = 0;
2845 u16 suballoc_bit = 0, suballoc_slot = 0;
2846 struct inode *inode_alloc_inode;
2847 struct buffer_head *alloc_bh = NULL;
2849 trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2851 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2852 &group_blkno, &suballoc_bit);
2854 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2859 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2861 if (!inode_alloc_inode) {
2862 /* the error code could be inaccurate, but we are not able to
2863 * get the correct one. */
2865 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2866 (u32)suballoc_slot);
2870 mutex_lock(&inode_alloc_inode->i_mutex);
2871 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2873 mutex_unlock(&inode_alloc_inode->i_mutex);
2874 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2875 (u32)suballoc_slot, status);
2879 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2880 group_blkno, blkno, suballoc_bit, res);
2882 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2884 ocfs2_inode_unlock(inode_alloc_inode, 0);
2885 mutex_unlock(&inode_alloc_inode->i_mutex);
2887 iput(inode_alloc_inode);