1 // SPDX-License-Identifier: GPL-2.0-or-later
5 * metadata alloc and free
6 * Inspired by ext3 block groups.
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/highmem.h>
16 #include <cluster/masklog.h>
21 #include "blockcheck.h"
25 #include "localalloc.h"
30 #include "ocfs2_trace.h"
32 #include "buffer_head_io.h"
34 #define NOT_ALLOC_NEW_GROUP 0
35 #define ALLOC_NEW_GROUP 0x1
36 #define ALLOC_GROUPS_FROM_GLOBAL 0x2
38 #define OCFS2_MAX_TO_STEAL 1024
40 struct ocfs2_suballoc_result {
41 u64 sr_bg_blkno; /* The bg we allocated from. Set
42 to 0 when a block group is
44 u64 sr_bg_stable_blkno; /*
45 * Doesn't change, always
50 u64 sr_blkno; /* The first allocated block */
51 unsigned int sr_bit_offset; /* The bit in the bg */
52 unsigned int sr_bits; /* How many bits we claimed */
53 unsigned int sr_max_contig_bits; /* The length for contiguous
54 * free bits, only available
59 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
61 if (res->sr_blkno == 0)
65 return res->sr_bg_blkno;
67 return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
70 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
71 static int ocfs2_block_group_fill(handle_t *handle,
72 struct inode *alloc_inode,
73 struct buffer_head *bg_bh,
75 unsigned int group_clusters,
77 struct ocfs2_chain_list *cl);
78 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
79 struct inode *alloc_inode,
80 struct buffer_head *bh,
82 u64 *last_alloc_group,
85 static int ocfs2_cluster_group_search(struct inode *inode,
86 struct buffer_head *group_bh,
87 u32 bits_wanted, u32 min_bits,
89 struct ocfs2_suballoc_result *res);
90 static int ocfs2_block_group_search(struct inode *inode,
91 struct buffer_head *group_bh,
92 u32 bits_wanted, u32 min_bits,
94 struct ocfs2_suballoc_result *res);
95 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
99 struct ocfs2_suballoc_result *res);
100 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
102 static int ocfs2_relink_block_group(handle_t *handle,
103 struct inode *alloc_inode,
104 struct buffer_head *fe_bh,
105 struct buffer_head *bg_bh,
106 struct buffer_head *prev_bg_bh,
108 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
110 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
113 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
117 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
118 u32 bits_wanted, u64 max_block,
120 struct ocfs2_alloc_context **ac);
122 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
124 struct inode *inode = ac->ac_inode;
127 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
128 ocfs2_inode_unlock(inode, 1);
138 kfree(ac->ac_find_loc_priv);
139 ac->ac_find_loc_priv = NULL;
142 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
144 ocfs2_free_ac_resource(ac);
148 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
150 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
153 #define do_error(fmt, ...) \
156 mlog(ML_ERROR, fmt, ##__VA_ARGS__); \
158 return ocfs2_error(sb, fmt, ##__VA_ARGS__); \
161 static int ocfs2_validate_gd_self(struct super_block *sb,
162 struct buffer_head *bh,
165 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
167 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
168 do_error("Group descriptor #%llu has bad signature %.*s\n",
169 (unsigned long long)bh->b_blocknr, 7,
173 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
174 do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
175 (unsigned long long)bh->b_blocknr,
176 (unsigned long long)le64_to_cpu(gd->bg_blkno));
179 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
180 do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
181 (unsigned long long)bh->b_blocknr,
182 le32_to_cpu(gd->bg_generation));
185 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
186 do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
187 (unsigned long long)bh->b_blocknr,
188 le16_to_cpu(gd->bg_bits),
189 le16_to_cpu(gd->bg_free_bits_count));
192 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
193 do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
194 (unsigned long long)bh->b_blocknr,
195 le16_to_cpu(gd->bg_bits),
196 8 * le16_to_cpu(gd->bg_size));
202 static int ocfs2_validate_gd_parent(struct super_block *sb,
203 struct ocfs2_dinode *di,
204 struct buffer_head *bh,
207 unsigned int max_bits;
208 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
210 if (di->i_blkno != gd->bg_parent_dinode) {
211 do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
212 (unsigned long long)bh->b_blocknr,
213 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
214 (unsigned long long)le64_to_cpu(di->i_blkno));
217 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
218 if (le16_to_cpu(gd->bg_bits) > max_bits) {
219 do_error("Group descriptor #%llu has bit count of %u\n",
220 (unsigned long long)bh->b_blocknr,
221 le16_to_cpu(gd->bg_bits));
224 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
225 if ((le16_to_cpu(gd->bg_chain) >
226 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
227 ((le16_to_cpu(gd->bg_chain) ==
228 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
229 do_error("Group descriptor #%llu has bad chain %u\n",
230 (unsigned long long)bh->b_blocknr,
231 le16_to_cpu(gd->bg_chain));
240 * This version only prints errors. It does not fail the filesystem, and
241 * exists only for resize.
243 int ocfs2_check_group_descriptor(struct super_block *sb,
244 struct ocfs2_dinode *di,
245 struct buffer_head *bh)
248 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
250 BUG_ON(!buffer_uptodate(bh));
253 * If the ecc fails, we return the error but otherwise
254 * leave the filesystem running. We know any error is
255 * local to this block.
257 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
260 "Checksum failed for group descriptor %llu\n",
261 (unsigned long long)bh->b_blocknr);
263 rc = ocfs2_validate_gd_self(sb, bh, 1);
265 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
270 static int ocfs2_validate_group_descriptor(struct super_block *sb,
271 struct buffer_head *bh)
274 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
276 trace_ocfs2_validate_group_descriptor(
277 (unsigned long long)bh->b_blocknr);
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);
291 * Errors after here are fatal.
294 return ocfs2_validate_gd_self(sb, bh, 0);
297 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
298 u64 gd_blkno, struct buffer_head **bh)
301 struct buffer_head *tmp = *bh;
303 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
304 ocfs2_validate_group_descriptor);
308 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
314 /* If ocfs2_read_block() got us a new bh, pass it up. */
322 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
323 struct ocfs2_group_desc *bg,
324 struct ocfs2_chain_list *cl,
325 u64 p_blkno, unsigned int clusters)
327 struct ocfs2_extent_list *el = &bg->bg_list;
328 struct ocfs2_extent_rec *rec;
330 BUG_ON(!ocfs2_supports_discontig_bg(osb));
331 if (!el->l_next_free_rec)
332 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
333 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
334 rec->e_blkno = cpu_to_le64(p_blkno);
335 rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
336 le16_to_cpu(cl->cl_bpc));
337 rec->e_leaf_clusters = cpu_to_le16(clusters);
338 le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
339 le16_add_cpu(&bg->bg_free_bits_count,
340 clusters * le16_to_cpu(cl->cl_bpc));
341 le16_add_cpu(&el->l_next_free_rec, 1);
344 static int ocfs2_block_group_fill(handle_t *handle,
345 struct inode *alloc_inode,
346 struct buffer_head *bg_bh,
348 unsigned int group_clusters,
350 struct ocfs2_chain_list *cl)
353 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
354 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
355 struct super_block * sb = alloc_inode->i_sb;
357 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
358 status = ocfs2_error(alloc_inode->i_sb,
359 "group block (%llu) != b_blocknr (%llu)\n",
360 (unsigned long long)group_blkno,
361 (unsigned long long) bg_bh->b_blocknr);
365 status = ocfs2_journal_access_gd(handle,
366 INODE_CACHE(alloc_inode),
368 OCFS2_JOURNAL_ACCESS_CREATE);
374 memset(bg, 0, sb->s_blocksize);
375 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
376 bg->bg_generation = cpu_to_le32(osb->fs_generation);
377 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
378 osb->s_feature_incompat));
379 bg->bg_chain = cpu_to_le16(my_chain);
380 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
381 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
382 bg->bg_blkno = cpu_to_le64(group_blkno);
383 if (group_clusters == le16_to_cpu(cl->cl_cpg))
384 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
386 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
389 /* set the 1st bit in the bitmap to account for the descriptor block */
390 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
391 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
393 ocfs2_journal_dirty(handle, bg_bh);
395 /* There is no need to zero out or otherwise initialize the
396 * other blocks in a group - All valid FS metadata in a block
397 * group stores the superblock fs_generation value at
398 * allocation time. */
406 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
411 while (curr < le16_to_cpu(cl->cl_count)) {
412 if (le32_to_cpu(cl->cl_recs[best].c_total) >
413 le32_to_cpu(cl->cl_recs[curr].c_total))
420 static struct buffer_head *
421 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
422 struct inode *alloc_inode,
423 struct ocfs2_alloc_context *ac,
424 struct ocfs2_chain_list *cl)
427 u32 bit_off, num_bits;
429 struct buffer_head *bg_bh;
430 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
432 status = ocfs2_claim_clusters(handle, ac,
433 le16_to_cpu(cl->cl_cpg), &bit_off,
436 if (status != -ENOSPC)
441 /* setup the group */
442 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
443 trace_ocfs2_block_group_alloc_contig(
444 (unsigned long long)bg_blkno, alloc_rec);
446 bg_bh = sb_getblk(osb->sb, bg_blkno);
452 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
454 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
455 bg_blkno, num_bits, alloc_rec, cl);
462 return status ? ERR_PTR(status) : bg_bh;
465 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
467 struct ocfs2_alloc_context *ac,
468 unsigned int min_bits,
469 u32 *bit_off, u32 *num_bits)
474 status = ocfs2_claim_clusters(handle, ac, min_bits,
476 if (status != -ENOSPC)
485 static int ocfs2_block_group_grow_discontig(handle_t *handle,
486 struct inode *alloc_inode,
487 struct buffer_head *bg_bh,
488 struct ocfs2_alloc_context *ac,
489 struct ocfs2_chain_list *cl,
490 unsigned int min_bits)
493 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
494 struct ocfs2_group_desc *bg =
495 (struct ocfs2_group_desc *)bg_bh->b_data;
496 unsigned int needed = le16_to_cpu(cl->cl_cpg) -
497 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
498 u32 p_cpos, clusters;
500 struct ocfs2_extent_list *el = &bg->bg_list;
502 status = ocfs2_journal_access_gd(handle,
503 INODE_CACHE(alloc_inode),
505 OCFS2_JOURNAL_ACCESS_CREATE);
511 while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
512 le16_to_cpu(el->l_count))) {
513 if (min_bits > needed)
515 status = ocfs2_block_group_claim_bits(osb, handle, ac,
519 if (status != -ENOSPC)
523 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
524 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
528 needed = le16_to_cpu(cl->cl_cpg) -
529 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
534 * We have used up all the extent rec but can't fill up
535 * the cpg. So bail out.
541 ocfs2_journal_dirty(handle, bg_bh);
547 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
548 struct ocfs2_alloc_context *cluster_ac,
549 struct inode *alloc_inode,
550 struct buffer_head *bg_bh)
553 struct ocfs2_group_desc *bg;
554 struct ocfs2_extent_list *el;
555 struct ocfs2_extent_rec *rec;
560 bg = (struct ocfs2_group_desc *)bg_bh->b_data;
562 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
563 rec = &el->l_recs[i];
564 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
566 le64_to_cpu(rec->e_blkno),
567 le16_to_cpu(rec->e_leaf_clusters));
570 /* Try all the clusters to free */
573 ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
577 static struct buffer_head *
578 ocfs2_block_group_alloc_discontig(handle_t *handle,
579 struct inode *alloc_inode,
580 struct ocfs2_alloc_context *ac,
581 struct ocfs2_chain_list *cl)
584 u32 bit_off, num_bits;
586 unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
587 struct buffer_head *bg_bh = NULL;
588 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
589 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
591 if (!ocfs2_supports_discontig_bg(osb)) {
596 status = ocfs2_extend_trans(handle,
597 ocfs2_calc_bg_discontig_credits(osb->sb));
604 * We're going to be grabbing from multiple cluster groups.
605 * We don't have enough credits to relink them all, and the
606 * cluster groups will be staying in cache for the duration of
609 ac->ac_disable_chain_relink = 1;
611 /* Claim the first region */
612 status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
613 &bit_off, &num_bits);
615 if (status != -ENOSPC)
621 /* setup the group */
622 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
623 trace_ocfs2_block_group_alloc_discontig(
624 (unsigned long long)bg_blkno, alloc_rec);
626 bg_bh = sb_getblk(osb->sb, bg_blkno);
632 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
634 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
635 bg_blkno, num_bits, alloc_rec, cl);
641 status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
642 bg_bh, ac, cl, min_bits);
648 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
649 return status ? ERR_PTR(status) : bg_bh;
653 * We expect the block group allocator to already be locked.
655 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
656 struct inode *alloc_inode,
657 struct buffer_head *bh,
659 u64 *last_alloc_group,
663 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
664 struct ocfs2_chain_list *cl;
665 struct ocfs2_alloc_context *ac = NULL;
666 handle_t *handle = NULL;
668 struct buffer_head *bg_bh = NULL;
669 struct ocfs2_group_desc *bg;
671 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
673 cl = &fe->id2.i_chain;
674 status = ocfs2_reserve_clusters_with_limit(osb,
675 le16_to_cpu(cl->cl_cpg),
676 max_block, flags, &ac);
678 if (status != -ENOSPC)
683 credits = ocfs2_calc_group_alloc_credits(osb->sb,
684 le16_to_cpu(cl->cl_cpg));
685 handle = ocfs2_start_trans(osb, credits);
686 if (IS_ERR(handle)) {
687 status = PTR_ERR(handle);
693 if (last_alloc_group && *last_alloc_group != 0) {
694 trace_ocfs2_block_group_alloc(
695 (unsigned long long)*last_alloc_group);
696 ac->ac_last_group = *last_alloc_group;
699 bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
701 if (PTR_ERR(bg_bh) == -ENOSPC)
702 bg_bh = ocfs2_block_group_alloc_discontig(handle,
706 status = PTR_ERR(bg_bh);
708 if (status != -ENOSPC)
712 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
714 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
715 bh, OCFS2_JOURNAL_ACCESS_WRITE);
721 alloc_rec = le16_to_cpu(bg->bg_chain);
722 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
723 le16_to_cpu(bg->bg_free_bits_count));
724 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
725 le16_to_cpu(bg->bg_bits));
726 cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
727 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
728 le16_add_cpu(&cl->cl_next_free_rec, 1);
730 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
731 le16_to_cpu(bg->bg_free_bits_count));
732 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
733 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
735 ocfs2_journal_dirty(handle, bh);
737 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
738 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
739 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
740 le32_to_cpu(fe->i_clusters)));
741 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
742 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
743 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
744 ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
748 /* save the new last alloc group so that the caller can cache it. */
749 if (last_alloc_group)
750 *last_alloc_group = ac->ac_last_group;
754 ocfs2_commit_trans(osb, handle);
757 ocfs2_free_alloc_context(ac);
766 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
767 struct ocfs2_alloc_context *ac,
770 u64 *last_alloc_group,
774 u32 bits_wanted = ac->ac_bits_wanted;
775 struct inode *alloc_inode;
776 struct buffer_head *bh = NULL;
777 struct ocfs2_dinode *fe;
780 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
786 inode_lock(alloc_inode);
788 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
790 inode_unlock(alloc_inode);
797 ac->ac_inode = alloc_inode;
798 ac->ac_alloc_slot = slot;
800 fe = (struct ocfs2_dinode *) bh->b_data;
802 /* The bh was validated by the inode read inside
803 * ocfs2_inode_lock(). Any corruption is a code bug. */
804 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
806 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
807 status = ocfs2_error(alloc_inode->i_sb,
808 "Invalid chain allocator %llu\n",
809 (unsigned long long)le64_to_cpu(fe->i_blkno));
813 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
814 le32_to_cpu(fe->id1.bitmap1.i_used);
816 if (bits_wanted > free_bits) {
817 /* cluster bitmap never grows */
818 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
819 trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
825 if (!(flags & ALLOC_NEW_GROUP)) {
826 trace_ocfs2_reserve_suballoc_bits_no_new_group(
827 slot, bits_wanted, free_bits);
832 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
834 last_alloc_group, flags);
836 if (status != -ENOSPC)
840 atomic_inc(&osb->alloc_stats.bg_extends);
842 /* You should never ask for this much metadata */
844 (le32_to_cpu(fe->id1.bitmap1.i_total)
845 - le32_to_cpu(fe->id1.bitmap1.i_used)));
858 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
860 spin_lock(&osb->osb_lock);
861 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
862 spin_unlock(&osb->osb_lock);
863 atomic_set(&osb->s_num_inodes_stolen, 0);
866 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
868 spin_lock(&osb->osb_lock);
869 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
870 spin_unlock(&osb->osb_lock);
871 atomic_set(&osb->s_num_meta_stolen, 0);
874 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
876 ocfs2_init_inode_steal_slot(osb);
877 ocfs2_init_meta_steal_slot(osb);
880 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
882 spin_lock(&osb->osb_lock);
883 if (type == INODE_ALLOC_SYSTEM_INODE)
884 osb->s_inode_steal_slot = (u16)slot;
885 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
886 osb->s_meta_steal_slot = (u16)slot;
887 spin_unlock(&osb->osb_lock);
890 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
892 int slot = OCFS2_INVALID_SLOT;
894 spin_lock(&osb->osb_lock);
895 if (type == INODE_ALLOC_SYSTEM_INODE)
896 slot = osb->s_inode_steal_slot;
897 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
898 slot = osb->s_meta_steal_slot;
899 spin_unlock(&osb->osb_lock);
904 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
906 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
909 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
911 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
914 static int ocfs2_steal_resource(struct ocfs2_super *osb,
915 struct ocfs2_alloc_context *ac,
918 int i, status = -ENOSPC;
919 int slot = __ocfs2_get_steal_slot(osb, type);
921 /* Start to steal resource from the first slot after ours. */
922 if (slot == OCFS2_INVALID_SLOT)
923 slot = osb->slot_num + 1;
925 for (i = 0; i < osb->max_slots; i++, slot++) {
926 if (slot == osb->max_slots)
929 if (slot == osb->slot_num)
932 status = ocfs2_reserve_suballoc_bits(osb, ac,
935 NOT_ALLOC_NEW_GROUP);
937 __ocfs2_set_steal_slot(osb, slot, type);
941 ocfs2_free_ac_resource(ac);
947 static int ocfs2_steal_inode(struct ocfs2_super *osb,
948 struct ocfs2_alloc_context *ac)
950 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
953 static int ocfs2_steal_meta(struct ocfs2_super *osb,
954 struct ocfs2_alloc_context *ac)
956 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
959 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
961 struct ocfs2_alloc_context **ac)
964 int slot = ocfs2_get_meta_steal_slot(osb);
966 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
973 (*ac)->ac_bits_wanted = blocks;
974 (*ac)->ac_which = OCFS2_AC_USE_META;
975 (*ac)->ac_group_search = ocfs2_block_group_search;
977 if (slot != OCFS2_INVALID_SLOT &&
978 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
981 atomic_set(&osb->s_num_meta_stolen, 0);
982 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
983 EXTENT_ALLOC_SYSTEM_INODE,
984 (u32)osb->slot_num, NULL,
985 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
990 if (slot != OCFS2_INVALID_SLOT)
991 ocfs2_init_meta_steal_slot(osb);
993 } else if (status < 0 && status != -ENOSPC) {
998 ocfs2_free_ac_resource(*ac);
1001 status = ocfs2_steal_meta(osb, *ac);
1002 atomic_inc(&osb->s_num_meta_stolen);
1004 if (status != -ENOSPC)
1011 if ((status < 0) && *ac) {
1012 ocfs2_free_alloc_context(*ac);
1021 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1022 struct ocfs2_extent_list *root_el,
1023 struct ocfs2_alloc_context **ac)
1025 return ocfs2_reserve_new_metadata_blocks(osb,
1026 ocfs2_extend_meta_needed(root_el),
1030 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1031 struct ocfs2_alloc_context **ac)
1034 int slot = ocfs2_get_inode_steal_slot(osb);
1037 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1044 (*ac)->ac_bits_wanted = 1;
1045 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1047 (*ac)->ac_group_search = ocfs2_block_group_search;
1050 * stat(2) can't handle i_ino > 32bits, so we tell the
1051 * lower levels not to allocate us a block group past that
1052 * limit. The 'inode64' mount option avoids this behavior.
1054 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1055 (*ac)->ac_max_block = (u32)~0U;
1058 * slot is set when we successfully steal inode from other nodes.
1059 * It is reset in 3 places:
1060 * 1. when we flush the truncate log
1061 * 2. when we complete local alloc recovery.
1062 * 3. when we successfully allocate from our own slot.
1063 * After it is set, we will go on stealing inodes until we find the
1064 * need to check our slots to see whether there is some space for us.
1066 if (slot != OCFS2_INVALID_SLOT &&
1067 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1070 atomic_set(&osb->s_num_inodes_stolen, 0);
1071 alloc_group = osb->osb_inode_alloc_group;
1072 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1073 INODE_ALLOC_SYSTEM_INODE,
1077 ALLOC_GROUPS_FROM_GLOBAL);
1081 spin_lock(&osb->osb_lock);
1082 osb->osb_inode_alloc_group = alloc_group;
1083 spin_unlock(&osb->osb_lock);
1084 trace_ocfs2_reserve_new_inode_new_group(
1085 (unsigned long long)alloc_group);
1088 * Some inodes must be freed by us, so try to allocate
1089 * from our own next time.
1091 if (slot != OCFS2_INVALID_SLOT)
1092 ocfs2_init_inode_steal_slot(osb);
1094 } else if (status < 0 && status != -ENOSPC) {
1099 ocfs2_free_ac_resource(*ac);
1102 status = ocfs2_steal_inode(osb, *ac);
1103 atomic_inc(&osb->s_num_inodes_stolen);
1105 if (status != -ENOSPC)
1112 if ((status < 0) && *ac) {
1113 ocfs2_free_alloc_context(*ac);
1122 /* local alloc code has to do the same thing, so rather than do this
1124 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1125 struct ocfs2_alloc_context *ac)
1129 ac->ac_which = OCFS2_AC_USE_MAIN;
1130 ac->ac_group_search = ocfs2_cluster_group_search;
1132 status = ocfs2_reserve_suballoc_bits(osb, ac,
1133 GLOBAL_BITMAP_SYSTEM_INODE,
1134 OCFS2_INVALID_SLOT, NULL,
1136 if (status < 0 && status != -ENOSPC)
1142 /* Callers don't need to care which bitmap (local alloc or main) to
1143 * use so we figure it out for them, but unfortunately this clutters
1145 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1146 u32 bits_wanted, u64 max_block,
1148 struct ocfs2_alloc_context **ac)
1150 int status, ret = 0;
1153 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1160 (*ac)->ac_bits_wanted = bits_wanted;
1161 (*ac)->ac_max_block = max_block;
1164 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1165 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1166 status = ocfs2_reserve_local_alloc_bits(osb,
1169 if ((status < 0) && (status != -ENOSPC)) {
1175 if (status == -ENOSPC) {
1177 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1178 /* Retry if there is sufficient space cached in truncate log */
1179 if (status == -ENOSPC && !retried) {
1181 ocfs2_inode_unlock((*ac)->ac_inode, 1);
1182 inode_unlock((*ac)->ac_inode);
1184 ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1186 iput((*ac)->ac_inode);
1187 (*ac)->ac_inode = NULL;
1194 inode_lock((*ac)->ac_inode);
1195 ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1198 inode_unlock((*ac)->ac_inode);
1199 iput((*ac)->ac_inode);
1200 (*ac)->ac_inode = NULL;
1205 if (status != -ENOSPC)
1213 if ((status < 0) && *ac) {
1214 ocfs2_free_alloc_context(*ac);
1223 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1225 struct ocfs2_alloc_context **ac)
1227 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1228 ALLOC_NEW_GROUP, ac);
1232 * More or less lifted from ext3. I'll leave their description below:
1234 * "For ext3 allocations, we must not reuse any blocks which are
1235 * allocated in the bitmap buffer's "last committed data" copy. This
1236 * prevents deletes from freeing up the page for reuse until we have
1237 * committed the delete transaction.
1239 * If we didn't do this, then deleting something and reallocating it as
1240 * data would allow the old block to be overwritten before the
1241 * transaction committed (because we force data to disk before commit).
1242 * This would lead to corruption if we crashed between overwriting the
1243 * data and committing the delete.
1245 * @@@ We may want to make this allocation behaviour conditional on
1246 * data-writes at some point, and disable it for metadata allocations or
1247 * sync-data inodes."
1249 * Note: OCFS2 already does this differently for metadata vs data
1250 * allocations, as those bitmaps are separate and undo access is never
1251 * called on a metadata group descriptor.
1253 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1256 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1257 struct journal_head *jh;
1260 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1263 jh = jbd2_journal_grab_journal_head(bg_bh);
1267 spin_lock(&jh->b_state_lock);
1268 bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1270 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1273 spin_unlock(&jh->b_state_lock);
1274 jbd2_journal_put_journal_head(jh);
1279 u16 ocfs2_find_max_contig_free_bits(void *bitmap,
1280 u16 total_bits, u16 start)
1282 u16 offset, free_bits;
1283 u16 contig_bits = 0;
1285 while (start < total_bits) {
1286 offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start);
1287 if (offset == total_bits)
1290 start = ocfs2_find_next_bit(bitmap, total_bits, offset);
1291 free_bits = start - offset;
1292 if (contig_bits < free_bits)
1293 contig_bits = free_bits;
1299 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1300 struct buffer_head *bg_bh,
1301 unsigned int bits_wanted,
1302 unsigned int total_bits,
1303 struct ocfs2_suballoc_result *res)
1306 u16 best_offset, best_size;
1307 u16 prev_best_size = 0;
1308 int offset, start, found, status = 0;
1309 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1311 /* Callers got this descriptor from
1312 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1313 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1315 found = start = best_offset = best_size = 0;
1316 bitmap = bg->bg_bitmap;
1318 while ((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) <
1320 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1321 /* We found a zero, but we can't use it as it
1322 * hasn't been put to disk yet! */
1325 } else if (offset == start) {
1326 /* we found a zero */
1328 /* move start to the next bit to test */
1331 /* got a zero after some ones */
1334 prev_best_size = best_size;
1336 if (found > best_size) {
1338 best_offset = start - found;
1340 /* we got everything we needed */
1341 if (found == bits_wanted) {
1342 /* mlog(0, "Found it all!\n"); */
1347 /* best_size will be allocated, we save prev_best_size */
1348 res->sr_max_contig_bits = prev_best_size;
1350 res->sr_bit_offset = best_offset;
1351 res->sr_bits = best_size;
1354 /* No error log here -- see the comment above
1355 * ocfs2_test_bg_bit_allocatable */
1361 int ocfs2_block_group_set_bits(handle_t *handle,
1362 struct inode *alloc_inode,
1363 struct ocfs2_group_desc *bg,
1364 struct buffer_head *group_bh,
1365 unsigned int bit_off,
1366 unsigned int num_bits,
1367 unsigned int max_contig_bits,
1371 void *bitmap = bg->bg_bitmap;
1372 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1373 unsigned int start = bit_off + num_bits;
1375 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
1377 /* All callers get the descriptor via
1378 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1379 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1380 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1382 trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1384 if (ocfs2_is_cluster_bitmap(alloc_inode))
1385 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1387 status = ocfs2_journal_access_gd(handle,
1388 INODE_CACHE(alloc_inode),
1396 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1397 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1398 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1399 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1400 le16_to_cpu(bg->bg_bits),
1401 le16_to_cpu(bg->bg_free_bits_count),
1405 ocfs2_set_bit(bit_off++, bitmap);
1408 * this is optimize path, caller set old contig value
1409 * in max_contig_bits to bypass finding action.
1412 bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
1413 } else if (ocfs2_is_cluster_bitmap(alloc_inode)) {
1415 * Usually, the block group bitmap allocates only 1 bit
1416 * at a time, while the cluster group allocates n bits
1417 * each time. Therefore, we only save the contig bits for
1418 * the cluster group.
1420 contig_bits = ocfs2_find_max_contig_free_bits(bitmap,
1421 le16_to_cpu(bg->bg_bits), start);
1422 if (contig_bits > max_contig_bits)
1423 max_contig_bits = contig_bits;
1424 bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
1425 ocfs2_local_alloc_seen_free_bits(osb, max_contig_bits);
1427 bg->bg_contig_free_bits = 0;
1430 ocfs2_journal_dirty(handle, group_bh);
1436 /* find the one with the most empty bits */
1437 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1441 BUG_ON(!cl->cl_next_free_rec);
1444 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1445 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1446 le32_to_cpu(cl->cl_recs[best].c_free))
1451 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1455 static int ocfs2_relink_block_group(handle_t *handle,
1456 struct inode *alloc_inode,
1457 struct buffer_head *fe_bh,
1458 struct buffer_head *bg_bh,
1459 struct buffer_head *prev_bg_bh,
1463 /* there is a really tiny chance the journal calls could fail,
1464 * but we wouldn't want inconsistent blocks in *any* case. */
1465 u64 bg_ptr, prev_bg_ptr;
1466 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1467 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1468 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1470 /* The caller got these descriptors from
1471 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1472 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1473 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1475 trace_ocfs2_relink_block_group(
1476 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1477 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1478 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1480 bg_ptr = le64_to_cpu(bg->bg_next_group);
1481 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1483 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1485 OCFS2_JOURNAL_ACCESS_WRITE);
1489 prev_bg->bg_next_group = bg->bg_next_group;
1490 ocfs2_journal_dirty(handle, prev_bg_bh);
1492 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1493 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1495 goto out_rollback_prev_bg;
1497 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1498 ocfs2_journal_dirty(handle, bg_bh);
1500 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1501 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1503 goto out_rollback_bg;
1505 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1506 ocfs2_journal_dirty(handle, fe_bh);
1514 bg->bg_next_group = cpu_to_le64(bg_ptr);
1515 out_rollback_prev_bg:
1516 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1520 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1523 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1526 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1527 * value on error. */
1528 static int ocfs2_cluster_group_search(struct inode *inode,
1529 struct buffer_head *group_bh,
1530 u32 bits_wanted, u32 min_bits,
1532 struct ocfs2_suballoc_result *res)
1534 int search = -ENOSPC;
1537 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1538 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1539 unsigned int max_bits, gd_cluster_off;
1541 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1543 if (le16_to_cpu(gd->bg_contig_free_bits) &&
1544 le16_to_cpu(gd->bg_contig_free_bits) < bits_wanted)
1547 /* ->bg_contig_free_bits may un-initialized, so compare again */
1548 if (le16_to_cpu(gd->bg_free_bits_count) >= bits_wanted) {
1549 max_bits = le16_to_cpu(gd->bg_bits);
1551 /* Tail groups in cluster bitmaps which aren't cpg
1552 * aligned are prone to partial extension by a failed
1553 * fs resize. If the file system resize never got to
1554 * update the dinode cluster count, then we don't want
1555 * to trust any clusters past it, regardless of what
1556 * the group descriptor says. */
1557 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1558 le64_to_cpu(gd->bg_blkno));
1559 if ((gd_cluster_off + max_bits) >
1560 OCFS2_I(inode)->ip_clusters) {
1561 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1562 trace_ocfs2_cluster_group_search_wrong_max_bits(
1563 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1564 le16_to_cpu(gd->bg_bits),
1565 OCFS2_I(inode)->ip_clusters, max_bits);
1568 ret = ocfs2_block_group_find_clear_bits(osb,
1569 group_bh, bits_wanted,
1575 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1577 res->sr_bit_offset +
1579 trace_ocfs2_cluster_group_search_max_block(
1580 (unsigned long long)blkoff,
1581 (unsigned long long)max_block);
1582 if (blkoff > max_block)
1586 /* ocfs2_block_group_find_clear_bits() might
1587 * return success, but we still want to return
1588 * -ENOSPC unless it found the minimum number
1590 if (min_bits <= res->sr_bits)
1591 search = 0; /* success */
1597 static int ocfs2_block_group_search(struct inode *inode,
1598 struct buffer_head *group_bh,
1599 u32 bits_wanted, u32 min_bits,
1601 struct ocfs2_suballoc_result *res)
1605 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1607 BUG_ON(min_bits != 1);
1608 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1610 if (le16_to_cpu(bg->bg_free_bits_count) >= bits_wanted) {
1611 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1612 group_bh, bits_wanted,
1613 le16_to_cpu(bg->bg_bits),
1615 if (!ret && max_block) {
1616 blkoff = le64_to_cpu(bg->bg_blkno) +
1617 res->sr_bit_offset + res->sr_bits;
1618 trace_ocfs2_block_group_search_max_block(
1619 (unsigned long long)blkoff,
1620 (unsigned long long)max_block);
1621 if (blkoff > max_block)
1629 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1631 struct buffer_head *di_bh,
1637 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1638 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1640 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1641 OCFS2_JOURNAL_ACCESS_WRITE);
1647 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1648 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1649 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1650 ocfs2_journal_dirty(handle, di_bh);
1656 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1657 struct buffer_head *di_bh,
1662 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1663 struct ocfs2_chain_list *cl;
1665 cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1666 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1667 di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1668 le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1671 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1672 struct ocfs2_extent_rec *rec,
1673 struct ocfs2_chain_list *cl)
1675 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1676 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1677 unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1679 if (res->sr_bit_offset < bitoff)
1681 if (res->sr_bit_offset >= (bitoff + bitcount))
1683 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1684 (res->sr_bit_offset - bitoff);
1685 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1686 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1690 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1691 struct ocfs2_group_desc *bg,
1692 struct ocfs2_suballoc_result *res)
1695 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
1696 struct ocfs2_extent_rec *rec;
1697 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1698 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1700 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1705 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1706 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
1707 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1708 !bg->bg_list.l_next_free_rec)
1711 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1712 rec = &bg->bg_list.l_recs[i];
1713 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1714 res->sr_bg_blkno = bg_blkno; /* Restore */
1720 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1724 struct ocfs2_suballoc_result *res,
1728 struct buffer_head *group_bh = NULL;
1729 struct ocfs2_group_desc *gd;
1730 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1731 struct inode *alloc_inode = ac->ac_inode;
1733 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1734 res->sr_bg_blkno, &group_bh);
1740 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1741 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1742 ac->ac_max_block, res);
1750 ocfs2_bg_discontig_fix_result(ac, gd, res);
1753 * sr_bg_blkno might have been changed by
1754 * ocfs2_bg_discontig_fix_result
1756 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1758 if (ac->ac_find_loc_only)
1761 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1763 le16_to_cpu(gd->bg_chain));
1769 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1770 res->sr_bit_offset, res->sr_bits,
1771 res->sr_max_contig_bits, 0);
1773 ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1775 le16_to_cpu(gd->bg_chain));
1780 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1788 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1792 struct ocfs2_suballoc_result *res,
1798 struct inode *alloc_inode = ac->ac_inode;
1799 struct buffer_head *group_bh = NULL;
1800 struct buffer_head *prev_group_bh = NULL;
1801 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1802 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1803 struct ocfs2_group_desc *bg;
1805 chain = ac->ac_chain;
1806 trace_ocfs2_search_chain_begin(
1807 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1808 bits_wanted, chain);
1810 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1811 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1817 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1820 /* for now, the chain search is a bit simplistic. We just use
1821 * the 1st group with any empty bits. */
1822 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1823 bits_wanted, min_bits,
1826 if (!bg->bg_next_group)
1829 brelse(prev_group_bh);
1830 prev_group_bh = NULL;
1832 next_group = le64_to_cpu(bg->bg_next_group);
1833 prev_group_bh = group_bh;
1835 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1836 next_group, &group_bh);
1841 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1844 if (status != -ENOSPC)
1849 trace_ocfs2_search_chain_succ(
1850 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1852 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1854 BUG_ON(res->sr_bits == 0);
1856 ocfs2_bg_discontig_fix_result(ac, bg, res);
1859 * sr_bg_blkno might have been changed by
1860 * ocfs2_bg_discontig_fix_result
1862 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1865 * Keep track of previous block descriptor read. When
1866 * we find a target, if we have read more than X
1867 * number of descriptors, and the target is reasonably
1868 * empty, relink him to top of his chain.
1870 * We've read 0 extra blocks and only send one more to
1871 * the transaction, yet the next guy to search has a
1874 * Do this *after* figuring out how many bits we're taking out
1875 * of our target group.
1877 if (!ac->ac_disable_chain_relink &&
1879 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1880 status = ocfs2_relink_block_group(handle, alloc_inode,
1881 ac->ac_bh, group_bh,
1882 prev_group_bh, chain);
1889 if (ac->ac_find_loc_only)
1892 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1893 ac->ac_bh, res->sr_bits,
1900 status = ocfs2_block_group_set_bits(handle,
1906 res->sr_max_contig_bits,
1909 ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1910 ac->ac_bh, res->sr_bits, chain);
1915 trace_ocfs2_search_chain_end(
1916 (unsigned long long)le64_to_cpu(fe->i_blkno),
1920 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1923 brelse(prev_group_bh);
1930 /* will give out up to bits_wanted contiguous bits. */
1931 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1935 struct ocfs2_suballoc_result *res)
1940 u64 hint = ac->ac_last_group;
1941 struct ocfs2_chain_list *cl;
1942 struct ocfs2_dinode *fe;
1944 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1945 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1948 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1950 /* The bh was validated by the inode read during
1951 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1952 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1954 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1955 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1956 status = ocfs2_error(ac->ac_inode->i_sb,
1957 "Chain allocator dinode %llu has %u used bits but only %u total\n",
1958 (unsigned long long)le64_to_cpu(fe->i_blkno),
1959 le32_to_cpu(fe->id1.bitmap1.i_used),
1960 le32_to_cpu(fe->id1.bitmap1.i_total));
1964 res->sr_bg_blkno = hint;
1965 if (res->sr_bg_blkno) {
1966 /* Attempt to short-circuit the usual search mechanism
1967 * by jumping straight to the most recently used
1968 * allocation group. This helps us maintain some
1969 * contiguousness across allocations. */
1970 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1971 min_bits, res, &bits_left);
1974 if (status < 0 && status != -ENOSPC) {
1980 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1982 victim = ocfs2_find_victim_chain(cl);
1983 ac->ac_chain = victim;
1985 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1988 if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1989 hint = res->sr_bg_blkno;
1991 hint = ocfs2_group_from_res(res);
1994 if (status < 0 && status != -ENOSPC) {
1999 trace_ocfs2_claim_suballoc_bits(victim);
2001 /* If we didn't pick a good victim, then just default to
2002 * searching each chain in order. Don't allow chain relinking
2003 * because we only calculate enough journal credits for one
2004 * relink per alloc. */
2005 ac->ac_disable_chain_relink = 1;
2006 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
2009 if (le32_to_cpu(cl->cl_recs[i].c_free) < bits_wanted)
2013 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
2016 hint = ocfs2_group_from_res(res);
2019 if (status < 0 && status != -ENOSPC) {
2026 if (status != -ENOSPC) {
2027 /* If the next search of this group is not likely to
2028 * yield a suitable extent, then we reset the last
2029 * group hint so as to not waste a disk read */
2030 if (bits_left < min_bits)
2031 ac->ac_last_group = 0;
2033 ac->ac_last_group = hint;
2042 int ocfs2_claim_metadata(handle_t *handle,
2043 struct ocfs2_alloc_context *ac,
2046 u16 *suballoc_bit_start,
2047 unsigned int *num_bits,
2051 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2054 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2055 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2057 status = ocfs2_claim_suballoc_bits(ac,
2066 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2068 *suballoc_loc = res.sr_bg_blkno;
2069 *suballoc_bit_start = res.sr_bit_offset;
2070 *blkno_start = res.sr_blkno;
2071 ac->ac_bits_given += res.sr_bits;
2072 *num_bits = res.sr_bits;
2080 static void ocfs2_init_inode_ac_group(struct inode *dir,
2081 struct buffer_head *parent_di_bh,
2082 struct ocfs2_alloc_context *ac)
2084 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2086 * Try to allocate inodes from some specific group.
2088 * If the parent dir has recorded the last group used in allocation,
2089 * cool, use it. Otherwise if we try to allocate new inode from the
2090 * same slot the parent dir belongs to, use the same chunk.
2092 * We are very careful here to avoid the mistake of setting
2093 * ac_last_group to a group descriptor from a different (unlocked) slot.
2095 if (OCFS2_I(dir)->ip_last_used_group &&
2096 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2097 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2098 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2099 if (di->i_suballoc_loc)
2100 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2102 ac->ac_last_group = ocfs2_which_suballoc_group(
2103 le64_to_cpu(di->i_blkno),
2104 le16_to_cpu(di->i_suballoc_bit));
2108 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2109 struct ocfs2_alloc_context *ac)
2111 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2112 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2115 int ocfs2_find_new_inode_loc(struct inode *dir,
2116 struct buffer_head *parent_fe_bh,
2117 struct ocfs2_alloc_context *ac,
2121 handle_t *handle = NULL;
2122 struct ocfs2_suballoc_result *res;
2125 BUG_ON(ac->ac_bits_given != 0);
2126 BUG_ON(ac->ac_bits_wanted != 1);
2127 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2129 res = kzalloc(sizeof(*res), GFP_NOFS);
2136 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2139 * The handle started here is for chain relink. Alternatively,
2140 * we could just disable relink for these calls.
2142 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2143 if (IS_ERR(handle)) {
2144 ret = PTR_ERR(handle);
2151 * This will instruct ocfs2_claim_suballoc_bits and
2152 * ocfs2_search_one_group to search but save actual allocation
2155 ac->ac_find_loc_only = 1;
2157 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2163 ac->ac_find_loc_priv = res;
2164 *fe_blkno = res->sr_blkno;
2165 ocfs2_update_inode_fsync_trans(handle, dir, 0);
2168 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2176 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2178 struct ocfs2_alloc_context *ac,
2185 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2186 struct buffer_head *bg_bh = NULL;
2187 struct ocfs2_group_desc *bg;
2188 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2191 * Since di_blkno is being passed back in, we check for any
2192 * inconsistencies which may have happened between
2193 * calls. These are code bugs as di_blkno is not expected to
2194 * change once returned from ocfs2_find_new_inode_loc()
2196 BUG_ON(res->sr_blkno != di_blkno);
2198 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2199 res->sr_bg_stable_blkno, &bg_bh);
2205 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2206 chain = le16_to_cpu(bg->bg_chain);
2208 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2209 ac->ac_bh, res->sr_bits,
2216 ret = ocfs2_block_group_set_bits(handle,
2222 res->sr_max_contig_bits,
2225 ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2226 ac->ac_bh, res->sr_bits, chain);
2231 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2234 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2236 BUG_ON(res->sr_bits != 1);
2238 *suballoc_loc = res->sr_bg_blkno;
2239 *suballoc_bit = res->sr_bit_offset;
2240 ac->ac_bits_given++;
2241 ocfs2_save_inode_ac_group(dir, ac);
2249 int ocfs2_claim_new_inode(handle_t *handle,
2251 struct buffer_head *parent_fe_bh,
2252 struct ocfs2_alloc_context *ac,
2258 struct ocfs2_suballoc_result res;
2261 BUG_ON(ac->ac_bits_given != 0);
2262 BUG_ON(ac->ac_bits_wanted != 1);
2263 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2265 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2267 status = ocfs2_claim_suballoc_bits(ac,
2276 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2278 BUG_ON(res.sr_bits != 1);
2280 *suballoc_loc = res.sr_bg_blkno;
2281 *suballoc_bit = res.sr_bit_offset;
2282 *fe_blkno = res.sr_blkno;
2283 ac->ac_bits_given++;
2284 ocfs2_save_inode_ac_group(dir, ac);
2292 /* translate a group desc. blkno and it's bitmap offset into
2293 * disk cluster offset. */
2294 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2298 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2301 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2303 if (bg_blkno != osb->first_cluster_group_blkno)
2304 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2305 cluster += (u32) bg_bit_off;
2309 /* given a cluster offset, calculate which block group it belongs to
2310 * and return that block offset. */
2311 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2313 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2316 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2318 group_no = cluster / osb->bitmap_cpg;
2320 return osb->first_cluster_group_blkno;
2321 return ocfs2_clusters_to_blocks(inode->i_sb,
2322 group_no * osb->bitmap_cpg);
2325 /* given the block number of a cluster start, calculate which cluster
2326 * group and descriptor bitmap offset that corresponds to. */
2327 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2332 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2333 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2335 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2337 *bg_blkno = ocfs2_which_cluster_group(inode,
2340 if (*bg_blkno == osb->first_cluster_group_blkno)
2341 *bg_bit_off = (u16) data_cluster;
2343 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2344 data_blkno - *bg_blkno);
2348 * min_bits - minimum contiguous chunk from this total allocation we
2349 * can handle. set to what we asked for originally for a full
2350 * contig. allocation, set to '1' to indicate we can deal with extents
2353 int __ocfs2_claim_clusters(handle_t *handle,
2354 struct ocfs2_alloc_context *ac,
2361 unsigned int bits_wanted = max_clusters;
2362 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2363 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2365 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2367 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2368 && ac->ac_which != OCFS2_AC_USE_MAIN);
2370 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2371 WARN_ON(min_clusters > 1);
2373 status = ocfs2_claim_local_alloc_bits(osb,
2380 atomic_inc(&osb->alloc_stats.local_data);
2382 if (min_clusters > (osb->bitmap_cpg - 1)) {
2383 /* The only paths asking for contiguousness
2384 * should know about this already. */
2385 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2386 "group bitmap size %u!\n", min_clusters,
2391 /* clamp the current request down to a realistic size. */
2392 if (bits_wanted > (osb->bitmap_cpg - 1))
2393 bits_wanted = osb->bitmap_cpg - 1;
2395 status = ocfs2_claim_suballoc_bits(ac,
2401 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2403 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2406 atomic_inc(&osb->alloc_stats.bitmap_data);
2407 *num_clusters = res.sr_bits;
2411 if (status != -ENOSPC)
2416 ac->ac_bits_given += *num_clusters;
2424 int ocfs2_claim_clusters(handle_t *handle,
2425 struct ocfs2_alloc_context *ac,
2430 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2432 return __ocfs2_claim_clusters(handle, ac, min_clusters,
2433 bits_wanted, cluster_start, num_clusters);
2436 static int ocfs2_block_group_clear_bits(handle_t *handle,
2437 struct inode *alloc_inode,
2438 struct ocfs2_group_desc *bg,
2439 struct buffer_head *group_bh,
2440 unsigned int bit_off,
2441 unsigned int num_bits,
2442 unsigned int max_contig_bits,
2443 void (*undo_fn)(unsigned int bit,
2444 unsigned long *bmap))
2449 struct ocfs2_group_desc *undo_bg = NULL;
2450 struct journal_head *jh;
2452 /* The caller got this descriptor from
2453 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2454 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2456 trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2458 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2459 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2462 OCFS2_JOURNAL_ACCESS_UNDO :
2463 OCFS2_JOURNAL_ACCESS_WRITE);
2469 jh = bh2jh(group_bh);
2471 spin_lock(&jh->b_state_lock);
2472 undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2478 ocfs2_clear_bit((bit_off + tmp),
2479 (unsigned long *) bg->bg_bitmap);
2481 undo_fn(bit_off + tmp,
2482 (unsigned long *) undo_bg->bg_bitmap);
2484 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2485 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2487 spin_unlock(&jh->b_state_lock);
2488 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2489 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2490 le16_to_cpu(bg->bg_bits),
2491 le16_to_cpu(bg->bg_free_bits_count),
2496 * TODO: even 'num_bits == 1' (the worst case, release 1 cluster),
2497 * we still need to rescan whole bitmap.
2499 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
2500 contig_bits = ocfs2_find_max_contig_free_bits(bg->bg_bitmap,
2501 le16_to_cpu(bg->bg_bits), 0);
2502 if (contig_bits > max_contig_bits)
2503 max_contig_bits = contig_bits;
2504 bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
2506 bg->bg_contig_free_bits = 0;
2510 spin_unlock(&jh->b_state_lock);
2512 ocfs2_journal_dirty(handle, group_bh);
2518 * expects the suballoc inode to already be locked.
2520 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2521 struct inode *alloc_inode,
2522 struct buffer_head *alloc_bh,
2523 unsigned int start_bit,
2526 void (*undo_fn)(unsigned int bit,
2527 unsigned long *bitmap))
2531 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2532 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2533 struct buffer_head *group_bh = NULL;
2534 struct ocfs2_group_desc *group;
2535 __le16 old_bg_contig_free_bits = 0;
2537 /* The alloc_bh comes from ocfs2_free_dinode() or
2538 * ocfs2_free_clusters(). The callers have all locked the
2539 * allocator and gotten alloc_bh from the lock call. This
2540 * validates the dinode buffer. Any corruption that has happened
2542 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2543 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2545 trace_ocfs2_free_suballoc_bits(
2546 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2547 (unsigned long long)bg_blkno,
2550 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2556 group = (struct ocfs2_group_desc *) group_bh->b_data;
2558 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2560 if (ocfs2_is_cluster_bitmap(alloc_inode))
2561 old_bg_contig_free_bits = group->bg_contig_free_bits;
2562 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2564 start_bit, count, 0, undo_fn);
2570 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2571 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2574 ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2576 le16_to_cpu(old_bg_contig_free_bits), 1);
2580 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2582 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2583 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2584 ocfs2_journal_dirty(handle, alloc_bh);
2591 int ocfs2_free_suballoc_bits(handle_t *handle,
2592 struct inode *alloc_inode,
2593 struct buffer_head *alloc_bh,
2594 unsigned int start_bit,
2598 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2599 start_bit, bg_blkno, count, NULL);
2602 int ocfs2_free_dinode(handle_t *handle,
2603 struct inode *inode_alloc_inode,
2604 struct buffer_head *inode_alloc_bh,
2605 struct ocfs2_dinode *di)
2607 u64 blk = le64_to_cpu(di->i_blkno);
2608 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2609 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2611 if (di->i_suballoc_loc)
2612 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2613 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2614 inode_alloc_bh, bit, bg_blkno, 1);
2617 static int _ocfs2_free_clusters(handle_t *handle,
2618 struct inode *bitmap_inode,
2619 struct buffer_head *bitmap_bh,
2621 unsigned int num_clusters,
2622 void (*undo_fn)(unsigned int bit,
2623 unsigned long *bitmap))
2629 /* You can't ever have a contiguous set of clusters
2630 * bigger than a block group bitmap so we never have to worry
2631 * about looping on them.
2632 * This is expensive. We can safely remove once this stuff has
2633 * gotten tested really well. */
2634 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2635 ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2639 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2642 trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2643 (unsigned long long)start_blk,
2644 bg_start_bit, num_clusters);
2646 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2647 bg_start_bit, bg_blkno,
2648 num_clusters, undo_fn);
2654 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2661 int ocfs2_free_clusters(handle_t *handle,
2662 struct inode *bitmap_inode,
2663 struct buffer_head *bitmap_bh,
2665 unsigned int num_clusters)
2667 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2668 start_blk, num_clusters,
2673 * Give never-used clusters back to the global bitmap. We don't need
2674 * to protect these bits in the undo buffer.
2676 int ocfs2_release_clusters(handle_t *handle,
2677 struct inode *bitmap_inode,
2678 struct buffer_head *bitmap_bh,
2680 unsigned int num_clusters)
2682 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2683 start_blk, num_clusters,
2688 * For a given allocation, determine which allocators will need to be
2689 * accessed, and lock them, reserving the appropriate number of bits.
2691 * Sparse file systems call this from ocfs2_write_begin_nolock()
2692 * and ocfs2_allocate_unwritten_extents().
2694 * File systems which don't support holes call this from
2695 * ocfs2_extend_allocation().
2697 int ocfs2_lock_allocators(struct inode *inode,
2698 struct ocfs2_extent_tree *et,
2699 u32 clusters_to_add, u32 extents_to_split,
2700 struct ocfs2_alloc_context **data_ac,
2701 struct ocfs2_alloc_context **meta_ac)
2703 int ret = 0, num_free_extents;
2704 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2705 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2711 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2713 num_free_extents = ocfs2_num_free_extents(et);
2714 if (num_free_extents < 0) {
2715 ret = num_free_extents;
2721 * Sparse allocation file systems need to be more conservative
2722 * with reserving room for expansion - the actual allocation
2723 * happens while we've got a journal handle open so re-taking
2724 * a cluster lock (because we ran out of room for another
2725 * extent) will violate ordering rules.
2727 * Most of the time we'll only be seeing this 1 cluster at a time
2730 * Always lock for any unwritten extents - we might want to
2731 * add blocks during a split.
2733 if (!num_free_extents ||
2734 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2735 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2743 if (clusters_to_add == 0)
2746 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2756 ocfs2_free_alloc_context(*meta_ac);
2761 * We cannot have an error and a non null *data_ac.
2769 * Read the inode specified by blkno to get suballoc_slot and
2772 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2773 u16 *suballoc_slot, u64 *group_blkno,
2777 struct buffer_head *inode_bh = NULL;
2778 struct ocfs2_dinode *inode_fe;
2780 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2782 /* dirty read disk */
2783 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2785 mlog(ML_ERROR, "read block %llu failed %d\n",
2786 (unsigned long long)blkno, status);
2790 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2791 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2792 mlog(ML_ERROR, "invalid inode %llu requested\n",
2793 (unsigned long long)blkno);
2798 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2799 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2800 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2801 (unsigned long long)blkno,
2802 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2808 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2810 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2812 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2823 * test whether bit is SET in allocator bitmap or not. on success, 0
2824 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2825 * is returned and *res is meaningless. Call this after you have
2826 * cluster locked against suballoc, or you may get a result based on
2827 * non-up2date contents
2829 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2830 struct inode *suballoc,
2831 struct buffer_head *alloc_bh,
2832 u64 group_blkno, u64 blkno,
2835 struct ocfs2_dinode *alloc_di;
2836 struct ocfs2_group_desc *group;
2837 struct buffer_head *group_bh = NULL;
2841 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2844 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2845 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2846 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2848 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2853 bg_blkno = group_blkno ? group_blkno :
2854 ocfs2_which_suballoc_group(blkno, bit);
2855 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2858 mlog(ML_ERROR, "read group %llu failed %d\n",
2859 (unsigned long long)bg_blkno, status);
2863 group = (struct ocfs2_group_desc *) group_bh->b_data;
2864 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2875 * Test if the bit representing this inode (blkno) is set in the
2878 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2880 * In the event of failure, a negative value is returned and *res is
2883 * Callers must make sure to hold nfs_sync_lock to prevent
2884 * ocfs2_delete_inode() on another node from accessing the same
2885 * suballocator concurrently.
2887 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2890 u64 group_blkno = 0;
2891 u16 suballoc_bit = 0, suballoc_slot = 0;
2892 struct inode *inode_alloc_inode;
2893 struct buffer_head *alloc_bh = NULL;
2895 trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2897 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2898 &group_blkno, &suballoc_bit);
2900 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2904 if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2905 inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2906 GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2908 inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2909 INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2910 if (!inode_alloc_inode) {
2911 /* the error code could be inaccurate, but we are not able to
2912 * get the correct one. */
2914 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2915 (u32)suballoc_slot);
2919 inode_lock(inode_alloc_inode);
2920 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2922 inode_unlock(inode_alloc_inode);
2923 iput(inode_alloc_inode);
2924 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2925 (u32)suballoc_slot, status);
2929 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2930 group_blkno, blkno, suballoc_bit, res);
2932 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2934 ocfs2_inode_unlock(inode_alloc_inode, 0);
2935 inode_unlock(inode_alloc_inode);
2937 iput(inode_alloc_inode);