1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * metadata alloc and free
8 * Inspired by ext3 block groups.
10 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/highmem.h>
18 #include <cluster/masklog.h>
23 #include "blockcheck.h"
27 #include "localalloc.h"
32 #include "ocfs2_trace.h"
34 #include "buffer_head_io.h"
36 #define NOT_ALLOC_NEW_GROUP 0
37 #define ALLOC_NEW_GROUP 0x1
38 #define ALLOC_GROUPS_FROM_GLOBAL 0x2
40 #define OCFS2_MAX_TO_STEAL 1024
42 struct ocfs2_suballoc_result {
43 u64 sr_bg_blkno; /* The bg we allocated from. Set
44 to 0 when a block group is
46 u64 sr_bg_stable_blkno; /*
47 * Doesn't change, always
52 u64 sr_blkno; /* The first allocated block */
53 unsigned int sr_bit_offset; /* The bit in the bg */
54 unsigned int sr_bits; /* How many bits we claimed */
57 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
59 if (res->sr_blkno == 0)
63 return res->sr_bg_blkno;
65 return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
68 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
69 static int ocfs2_block_group_fill(handle_t *handle,
70 struct inode *alloc_inode,
71 struct buffer_head *bg_bh,
73 unsigned int group_clusters,
75 struct ocfs2_chain_list *cl);
76 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
77 struct inode *alloc_inode,
78 struct buffer_head *bh,
80 u64 *last_alloc_group,
83 static int ocfs2_cluster_group_search(struct inode *inode,
84 struct buffer_head *group_bh,
85 u32 bits_wanted, u32 min_bits,
87 struct ocfs2_suballoc_result *res);
88 static int ocfs2_block_group_search(struct inode *inode,
89 struct buffer_head *group_bh,
90 u32 bits_wanted, u32 min_bits,
92 struct ocfs2_suballoc_result *res);
93 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
97 struct ocfs2_suballoc_result *res);
98 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
100 static int ocfs2_relink_block_group(handle_t *handle,
101 struct inode *alloc_inode,
102 struct buffer_head *fe_bh,
103 struct buffer_head *bg_bh,
104 struct buffer_head *prev_bg_bh,
106 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
108 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
111 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
115 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
116 u32 bits_wanted, u64 max_block,
118 struct ocfs2_alloc_context **ac);
120 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
122 struct inode *inode = ac->ac_inode;
125 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
126 ocfs2_inode_unlock(inode, 1);
136 kfree(ac->ac_find_loc_priv);
137 ac->ac_find_loc_priv = NULL;
140 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
142 ocfs2_free_ac_resource(ac);
146 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
148 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
151 #define do_error(fmt, ...) \
154 mlog(ML_ERROR, fmt, ##__VA_ARGS__); \
156 return ocfs2_error(sb, fmt, ##__VA_ARGS__); \
159 static int ocfs2_validate_gd_self(struct super_block *sb,
160 struct buffer_head *bh,
163 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
165 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
166 do_error("Group descriptor #%llu has bad signature %.*s\n",
167 (unsigned long long)bh->b_blocknr, 7,
171 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
172 do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
173 (unsigned long long)bh->b_blocknr,
174 (unsigned long long)le64_to_cpu(gd->bg_blkno));
177 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
178 do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
179 (unsigned long long)bh->b_blocknr,
180 le32_to_cpu(gd->bg_generation));
183 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
184 do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
185 (unsigned long long)bh->b_blocknr,
186 le16_to_cpu(gd->bg_bits),
187 le16_to_cpu(gd->bg_free_bits_count));
190 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
191 do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
192 (unsigned long long)bh->b_blocknr,
193 le16_to_cpu(gd->bg_bits),
194 8 * le16_to_cpu(gd->bg_size));
200 static int ocfs2_validate_gd_parent(struct super_block *sb,
201 struct ocfs2_dinode *di,
202 struct buffer_head *bh,
205 unsigned int max_bits;
206 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
208 if (di->i_blkno != gd->bg_parent_dinode) {
209 do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
210 (unsigned long long)bh->b_blocknr,
211 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
212 (unsigned long long)le64_to_cpu(di->i_blkno));
215 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
216 if (le16_to_cpu(gd->bg_bits) > max_bits) {
217 do_error("Group descriptor #%llu has bit count of %u\n",
218 (unsigned long long)bh->b_blocknr,
219 le16_to_cpu(gd->bg_bits));
222 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
223 if ((le16_to_cpu(gd->bg_chain) >
224 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
225 ((le16_to_cpu(gd->bg_chain) ==
226 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
227 do_error("Group descriptor #%llu has bad chain %u\n",
228 (unsigned long long)bh->b_blocknr,
229 le16_to_cpu(gd->bg_chain));
238 * This version only prints errors. It does not fail the filesystem, and
239 * exists only for resize.
241 int ocfs2_check_group_descriptor(struct super_block *sb,
242 struct ocfs2_dinode *di,
243 struct buffer_head *bh)
246 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
248 BUG_ON(!buffer_uptodate(bh));
251 * If the ecc fails, we return the error but otherwise
252 * leave the filesystem running. We know any error is
253 * local to this block.
255 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
258 "Checksum failed for group descriptor %llu\n",
259 (unsigned long long)bh->b_blocknr);
261 rc = ocfs2_validate_gd_self(sb, bh, 1);
263 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
268 static int ocfs2_validate_group_descriptor(struct super_block *sb,
269 struct buffer_head *bh)
272 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
274 trace_ocfs2_validate_group_descriptor(
275 (unsigned long long)bh->b_blocknr);
277 BUG_ON(!buffer_uptodate(bh));
280 * If the ecc fails, we return the error but otherwise
281 * leave the filesystem running. We know any error is
282 * local to this block.
284 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
289 * Errors after here are fatal.
292 return ocfs2_validate_gd_self(sb, bh, 0);
295 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
296 u64 gd_blkno, struct buffer_head **bh)
299 struct buffer_head *tmp = *bh;
301 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
302 ocfs2_validate_group_descriptor);
306 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
312 /* If ocfs2_read_block() got us a new bh, pass it up. */
320 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
321 struct ocfs2_group_desc *bg,
322 struct ocfs2_chain_list *cl,
323 u64 p_blkno, unsigned int clusters)
325 struct ocfs2_extent_list *el = &bg->bg_list;
326 struct ocfs2_extent_rec *rec;
328 BUG_ON(!ocfs2_supports_discontig_bg(osb));
329 if (!el->l_next_free_rec)
330 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
331 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
332 rec->e_blkno = cpu_to_le64(p_blkno);
333 rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
334 le16_to_cpu(cl->cl_bpc));
335 rec->e_leaf_clusters = cpu_to_le16(clusters);
336 le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
337 le16_add_cpu(&bg->bg_free_bits_count,
338 clusters * le16_to_cpu(cl->cl_bpc));
339 le16_add_cpu(&el->l_next_free_rec, 1);
342 static int ocfs2_block_group_fill(handle_t *handle,
343 struct inode *alloc_inode,
344 struct buffer_head *bg_bh,
346 unsigned int group_clusters,
348 struct ocfs2_chain_list *cl)
351 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
352 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
353 struct super_block * sb = alloc_inode->i_sb;
355 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
356 status = ocfs2_error(alloc_inode->i_sb,
357 "group block (%llu) != b_blocknr (%llu)\n",
358 (unsigned long long)group_blkno,
359 (unsigned long long) bg_bh->b_blocknr);
363 status = ocfs2_journal_access_gd(handle,
364 INODE_CACHE(alloc_inode),
366 OCFS2_JOURNAL_ACCESS_CREATE);
372 memset(bg, 0, sb->s_blocksize);
373 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
374 bg->bg_generation = cpu_to_le32(osb->fs_generation);
375 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
376 osb->s_feature_incompat));
377 bg->bg_chain = cpu_to_le16(my_chain);
378 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
379 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
380 bg->bg_blkno = cpu_to_le64(group_blkno);
381 if (group_clusters == le16_to_cpu(cl->cl_cpg))
382 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
384 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
387 /* set the 1st bit in the bitmap to account for the descriptor block */
388 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
389 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
391 ocfs2_journal_dirty(handle, bg_bh);
393 /* There is no need to zero out or otherwise initialize the
394 * other blocks in a group - All valid FS metadata in a block
395 * group stores the superblock fs_generation value at
396 * allocation time. */
404 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
409 while (curr < le16_to_cpu(cl->cl_count)) {
410 if (le32_to_cpu(cl->cl_recs[best].c_total) >
411 le32_to_cpu(cl->cl_recs[curr].c_total))
418 static struct buffer_head *
419 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
420 struct inode *alloc_inode,
421 struct ocfs2_alloc_context *ac,
422 struct ocfs2_chain_list *cl)
425 u32 bit_off, num_bits;
427 struct buffer_head *bg_bh;
428 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
430 status = ocfs2_claim_clusters(handle, ac,
431 le16_to_cpu(cl->cl_cpg), &bit_off,
434 if (status != -ENOSPC)
439 /* setup the group */
440 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
441 trace_ocfs2_block_group_alloc_contig(
442 (unsigned long long)bg_blkno, alloc_rec);
444 bg_bh = sb_getblk(osb->sb, bg_blkno);
450 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
452 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
453 bg_blkno, num_bits, alloc_rec, cl);
460 return status ? ERR_PTR(status) : bg_bh;
463 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
465 struct ocfs2_alloc_context *ac,
466 unsigned int min_bits,
467 u32 *bit_off, u32 *num_bits)
472 status = ocfs2_claim_clusters(handle, ac, min_bits,
474 if (status != -ENOSPC)
483 static int ocfs2_block_group_grow_discontig(handle_t *handle,
484 struct inode *alloc_inode,
485 struct buffer_head *bg_bh,
486 struct ocfs2_alloc_context *ac,
487 struct ocfs2_chain_list *cl,
488 unsigned int min_bits)
491 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
492 struct ocfs2_group_desc *bg =
493 (struct ocfs2_group_desc *)bg_bh->b_data;
494 unsigned int needed = le16_to_cpu(cl->cl_cpg) -
495 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
496 u32 p_cpos, clusters;
498 struct ocfs2_extent_list *el = &bg->bg_list;
500 status = ocfs2_journal_access_gd(handle,
501 INODE_CACHE(alloc_inode),
503 OCFS2_JOURNAL_ACCESS_CREATE);
509 while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
510 le16_to_cpu(el->l_count))) {
511 if (min_bits > needed)
513 status = ocfs2_block_group_claim_bits(osb, handle, ac,
517 if (status != -ENOSPC)
521 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
522 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
526 needed = le16_to_cpu(cl->cl_cpg) -
527 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
532 * We have used up all the extent rec but can't fill up
533 * the cpg. So bail out.
539 ocfs2_journal_dirty(handle, bg_bh);
545 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
546 struct ocfs2_alloc_context *cluster_ac,
547 struct inode *alloc_inode,
548 struct buffer_head *bg_bh)
551 struct ocfs2_group_desc *bg;
552 struct ocfs2_extent_list *el;
553 struct ocfs2_extent_rec *rec;
558 bg = (struct ocfs2_group_desc *)bg_bh->b_data;
560 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
561 rec = &el->l_recs[i];
562 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
564 le64_to_cpu(rec->e_blkno),
565 le16_to_cpu(rec->e_leaf_clusters));
568 /* Try all the clusters to free */
571 ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
575 static struct buffer_head *
576 ocfs2_block_group_alloc_discontig(handle_t *handle,
577 struct inode *alloc_inode,
578 struct ocfs2_alloc_context *ac,
579 struct ocfs2_chain_list *cl)
582 u32 bit_off, num_bits;
584 unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
585 struct buffer_head *bg_bh = NULL;
586 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
587 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
589 if (!ocfs2_supports_discontig_bg(osb)) {
594 status = ocfs2_extend_trans(handle,
595 ocfs2_calc_bg_discontig_credits(osb->sb));
602 * We're going to be grabbing from multiple cluster groups.
603 * We don't have enough credits to relink them all, and the
604 * cluster groups will be staying in cache for the duration of
607 ac->ac_disable_chain_relink = 1;
609 /* Claim the first region */
610 status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
611 &bit_off, &num_bits);
613 if (status != -ENOSPC)
619 /* setup the group */
620 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
621 trace_ocfs2_block_group_alloc_discontig(
622 (unsigned long long)bg_blkno, alloc_rec);
624 bg_bh = sb_getblk(osb->sb, bg_blkno);
630 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
632 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
633 bg_blkno, num_bits, alloc_rec, cl);
639 status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
640 bg_bh, ac, cl, min_bits);
646 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
647 return status ? ERR_PTR(status) : bg_bh;
651 * We expect the block group allocator to already be locked.
653 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
654 struct inode *alloc_inode,
655 struct buffer_head *bh,
657 u64 *last_alloc_group,
661 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
662 struct ocfs2_chain_list *cl;
663 struct ocfs2_alloc_context *ac = NULL;
664 handle_t *handle = NULL;
666 struct buffer_head *bg_bh = NULL;
667 struct ocfs2_group_desc *bg;
669 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
671 cl = &fe->id2.i_chain;
672 status = ocfs2_reserve_clusters_with_limit(osb,
673 le16_to_cpu(cl->cl_cpg),
674 max_block, flags, &ac);
676 if (status != -ENOSPC)
681 credits = ocfs2_calc_group_alloc_credits(osb->sb,
682 le16_to_cpu(cl->cl_cpg));
683 handle = ocfs2_start_trans(osb, credits);
684 if (IS_ERR(handle)) {
685 status = PTR_ERR(handle);
691 if (last_alloc_group && *last_alloc_group != 0) {
692 trace_ocfs2_block_group_alloc(
693 (unsigned long long)*last_alloc_group);
694 ac->ac_last_group = *last_alloc_group;
697 bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
699 if (PTR_ERR(bg_bh) == -ENOSPC)
700 bg_bh = ocfs2_block_group_alloc_discontig(handle,
704 status = PTR_ERR(bg_bh);
706 if (status != -ENOSPC)
710 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
712 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
713 bh, OCFS2_JOURNAL_ACCESS_WRITE);
719 alloc_rec = le16_to_cpu(bg->bg_chain);
720 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
721 le16_to_cpu(bg->bg_free_bits_count));
722 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
723 le16_to_cpu(bg->bg_bits));
724 cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
725 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
726 le16_add_cpu(&cl->cl_next_free_rec, 1);
728 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
729 le16_to_cpu(bg->bg_free_bits_count));
730 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
731 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
733 ocfs2_journal_dirty(handle, bh);
735 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
736 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
737 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
738 le32_to_cpu(fe->i_clusters)));
739 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
740 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
741 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
742 ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
746 /* save the new last alloc group so that the caller can cache it. */
747 if (last_alloc_group)
748 *last_alloc_group = ac->ac_last_group;
752 ocfs2_commit_trans(osb, handle);
755 ocfs2_free_alloc_context(ac);
764 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
765 struct ocfs2_alloc_context *ac,
768 u64 *last_alloc_group,
772 u32 bits_wanted = ac->ac_bits_wanted;
773 struct inode *alloc_inode;
774 struct buffer_head *bh = NULL;
775 struct ocfs2_dinode *fe;
778 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
784 inode_lock(alloc_inode);
786 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
788 inode_unlock(alloc_inode);
795 ac->ac_inode = alloc_inode;
796 ac->ac_alloc_slot = slot;
798 fe = (struct ocfs2_dinode *) bh->b_data;
800 /* The bh was validated by the inode read inside
801 * ocfs2_inode_lock(). Any corruption is a code bug. */
802 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
804 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
805 status = ocfs2_error(alloc_inode->i_sb,
806 "Invalid chain allocator %llu\n",
807 (unsigned long long)le64_to_cpu(fe->i_blkno));
811 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
812 le32_to_cpu(fe->id1.bitmap1.i_used);
814 if (bits_wanted > free_bits) {
815 /* cluster bitmap never grows */
816 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
817 trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
823 if (!(flags & ALLOC_NEW_GROUP)) {
824 trace_ocfs2_reserve_suballoc_bits_no_new_group(
825 slot, bits_wanted, free_bits);
830 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
832 last_alloc_group, flags);
834 if (status != -ENOSPC)
838 atomic_inc(&osb->alloc_stats.bg_extends);
840 /* You should never ask for this much metadata */
842 (le32_to_cpu(fe->id1.bitmap1.i_total)
843 - le32_to_cpu(fe->id1.bitmap1.i_used)));
856 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
858 spin_lock(&osb->osb_lock);
859 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
860 spin_unlock(&osb->osb_lock);
861 atomic_set(&osb->s_num_inodes_stolen, 0);
864 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
866 spin_lock(&osb->osb_lock);
867 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
868 spin_unlock(&osb->osb_lock);
869 atomic_set(&osb->s_num_meta_stolen, 0);
872 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
874 ocfs2_init_inode_steal_slot(osb);
875 ocfs2_init_meta_steal_slot(osb);
878 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
880 spin_lock(&osb->osb_lock);
881 if (type == INODE_ALLOC_SYSTEM_INODE)
882 osb->s_inode_steal_slot = (u16)slot;
883 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
884 osb->s_meta_steal_slot = (u16)slot;
885 spin_unlock(&osb->osb_lock);
888 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
890 int slot = OCFS2_INVALID_SLOT;
892 spin_lock(&osb->osb_lock);
893 if (type == INODE_ALLOC_SYSTEM_INODE)
894 slot = osb->s_inode_steal_slot;
895 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
896 slot = osb->s_meta_steal_slot;
897 spin_unlock(&osb->osb_lock);
902 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
904 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
907 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
909 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
912 static int ocfs2_steal_resource(struct ocfs2_super *osb,
913 struct ocfs2_alloc_context *ac,
916 int i, status = -ENOSPC;
917 int slot = __ocfs2_get_steal_slot(osb, type);
919 /* Start to steal resource from the first slot after ours. */
920 if (slot == OCFS2_INVALID_SLOT)
921 slot = osb->slot_num + 1;
923 for (i = 0; i < osb->max_slots; i++, slot++) {
924 if (slot == osb->max_slots)
927 if (slot == osb->slot_num)
930 status = ocfs2_reserve_suballoc_bits(osb, ac,
933 NOT_ALLOC_NEW_GROUP);
935 __ocfs2_set_steal_slot(osb, slot, type);
939 ocfs2_free_ac_resource(ac);
945 static int ocfs2_steal_inode(struct ocfs2_super *osb,
946 struct ocfs2_alloc_context *ac)
948 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
951 static int ocfs2_steal_meta(struct ocfs2_super *osb,
952 struct ocfs2_alloc_context *ac)
954 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
957 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
959 struct ocfs2_alloc_context **ac)
962 int slot = ocfs2_get_meta_steal_slot(osb);
964 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
971 (*ac)->ac_bits_wanted = blocks;
972 (*ac)->ac_which = OCFS2_AC_USE_META;
973 (*ac)->ac_group_search = ocfs2_block_group_search;
975 if (slot != OCFS2_INVALID_SLOT &&
976 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
979 atomic_set(&osb->s_num_meta_stolen, 0);
980 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
981 EXTENT_ALLOC_SYSTEM_INODE,
982 (u32)osb->slot_num, NULL,
983 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
988 if (slot != OCFS2_INVALID_SLOT)
989 ocfs2_init_meta_steal_slot(osb);
991 } else if (status < 0 && status != -ENOSPC) {
996 ocfs2_free_ac_resource(*ac);
999 status = ocfs2_steal_meta(osb, *ac);
1000 atomic_inc(&osb->s_num_meta_stolen);
1002 if (status != -ENOSPC)
1009 if ((status < 0) && *ac) {
1010 ocfs2_free_alloc_context(*ac);
1019 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1020 struct ocfs2_extent_list *root_el,
1021 struct ocfs2_alloc_context **ac)
1023 return ocfs2_reserve_new_metadata_blocks(osb,
1024 ocfs2_extend_meta_needed(root_el),
1028 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1029 struct ocfs2_alloc_context **ac)
1032 int slot = ocfs2_get_inode_steal_slot(osb);
1035 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1042 (*ac)->ac_bits_wanted = 1;
1043 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1045 (*ac)->ac_group_search = ocfs2_block_group_search;
1048 * stat(2) can't handle i_ino > 32bits, so we tell the
1049 * lower levels not to allocate us a block group past that
1050 * limit. The 'inode64' mount option avoids this behavior.
1052 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1053 (*ac)->ac_max_block = (u32)~0U;
1056 * slot is set when we successfully steal inode from other nodes.
1057 * It is reset in 3 places:
1058 * 1. when we flush the truncate log
1059 * 2. when we complete local alloc recovery.
1060 * 3. when we successfully allocate from our own slot.
1061 * After it is set, we will go on stealing inodes until we find the
1062 * need to check our slots to see whether there is some space for us.
1064 if (slot != OCFS2_INVALID_SLOT &&
1065 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1068 atomic_set(&osb->s_num_inodes_stolen, 0);
1069 alloc_group = osb->osb_inode_alloc_group;
1070 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1071 INODE_ALLOC_SYSTEM_INODE,
1075 ALLOC_GROUPS_FROM_GLOBAL);
1079 spin_lock(&osb->osb_lock);
1080 osb->osb_inode_alloc_group = alloc_group;
1081 spin_unlock(&osb->osb_lock);
1082 trace_ocfs2_reserve_new_inode_new_group(
1083 (unsigned long long)alloc_group);
1086 * Some inodes must be freed by us, so try to allocate
1087 * from our own next time.
1089 if (slot != OCFS2_INVALID_SLOT)
1090 ocfs2_init_inode_steal_slot(osb);
1092 } else if (status < 0 && status != -ENOSPC) {
1097 ocfs2_free_ac_resource(*ac);
1100 status = ocfs2_steal_inode(osb, *ac);
1101 atomic_inc(&osb->s_num_inodes_stolen);
1103 if (status != -ENOSPC)
1110 if ((status < 0) && *ac) {
1111 ocfs2_free_alloc_context(*ac);
1120 /* local alloc code has to do the same thing, so rather than do this
1122 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1123 struct ocfs2_alloc_context *ac)
1127 ac->ac_which = OCFS2_AC_USE_MAIN;
1128 ac->ac_group_search = ocfs2_cluster_group_search;
1130 status = ocfs2_reserve_suballoc_bits(osb, ac,
1131 GLOBAL_BITMAP_SYSTEM_INODE,
1132 OCFS2_INVALID_SLOT, NULL,
1134 if (status < 0 && status != -ENOSPC)
1140 /* Callers don't need to care which bitmap (local alloc or main) to
1141 * use so we figure it out for them, but unfortunately this clutters
1143 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1144 u32 bits_wanted, u64 max_block,
1146 struct ocfs2_alloc_context **ac)
1148 int status, ret = 0;
1151 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1158 (*ac)->ac_bits_wanted = bits_wanted;
1159 (*ac)->ac_max_block = max_block;
1162 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1163 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1164 status = ocfs2_reserve_local_alloc_bits(osb,
1167 if ((status < 0) && (status != -ENOSPC)) {
1173 if (status == -ENOSPC) {
1175 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1176 /* Retry if there is sufficient space cached in truncate log */
1177 if (status == -ENOSPC && !retried) {
1179 ocfs2_inode_unlock((*ac)->ac_inode, 1);
1180 inode_unlock((*ac)->ac_inode);
1182 ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1184 iput((*ac)->ac_inode);
1185 (*ac)->ac_inode = NULL;
1192 inode_lock((*ac)->ac_inode);
1193 ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1196 inode_unlock((*ac)->ac_inode);
1197 iput((*ac)->ac_inode);
1198 (*ac)->ac_inode = NULL;
1203 if (status != -ENOSPC)
1211 if ((status < 0) && *ac) {
1212 ocfs2_free_alloc_context(*ac);
1221 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1223 struct ocfs2_alloc_context **ac)
1225 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1226 ALLOC_NEW_GROUP, ac);
1230 * More or less lifted from ext3. I'll leave their description below:
1232 * "For ext3 allocations, we must not reuse any blocks which are
1233 * allocated in the bitmap buffer's "last committed data" copy. This
1234 * prevents deletes from freeing up the page for reuse until we have
1235 * committed the delete transaction.
1237 * If we didn't do this, then deleting something and reallocating it as
1238 * data would allow the old block to be overwritten before the
1239 * transaction committed (because we force data to disk before commit).
1240 * This would lead to corruption if we crashed between overwriting the
1241 * data and committing the delete.
1243 * @@@ We may want to make this allocation behaviour conditional on
1244 * data-writes at some point, and disable it for metadata allocations or
1245 * sync-data inodes."
1247 * Note: OCFS2 already does this differently for metadata vs data
1248 * allocations, as those bitmaps are separate and undo access is never
1249 * called on a metadata group descriptor.
1251 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1254 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1255 struct journal_head *jh;
1258 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1261 jh = jbd2_journal_grab_journal_head(bg_bh);
1265 spin_lock(&jh->b_state_lock);
1266 bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1268 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1271 spin_unlock(&jh->b_state_lock);
1272 jbd2_journal_put_journal_head(jh);
1277 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1278 struct buffer_head *bg_bh,
1279 unsigned int bits_wanted,
1280 unsigned int total_bits,
1281 struct ocfs2_suballoc_result *res)
1284 u16 best_offset, best_size;
1285 int offset, start, found, status = 0;
1286 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1288 /* Callers got this descriptor from
1289 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1290 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1292 found = start = best_offset = best_size = 0;
1293 bitmap = bg->bg_bitmap;
1295 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1296 if (offset == total_bits)
1299 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1300 /* We found a zero, but we can't use it as it
1301 * hasn't been put to disk yet! */
1304 } else if (offset == start) {
1305 /* we found a zero */
1307 /* move start to the next bit to test */
1310 /* got a zero after some ones */
1314 if (found > best_size) {
1316 best_offset = start - found;
1318 /* we got everything we needed */
1319 if (found == bits_wanted) {
1320 /* mlog(0, "Found it all!\n"); */
1326 res->sr_bit_offset = best_offset;
1327 res->sr_bits = best_size;
1330 /* No error log here -- see the comment above
1331 * ocfs2_test_bg_bit_allocatable */
1337 int ocfs2_block_group_set_bits(handle_t *handle,
1338 struct inode *alloc_inode,
1339 struct ocfs2_group_desc *bg,
1340 struct buffer_head *group_bh,
1341 unsigned int bit_off,
1342 unsigned int num_bits)
1345 void *bitmap = bg->bg_bitmap;
1346 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1348 /* All callers get the descriptor via
1349 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1350 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1351 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1353 trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1355 if (ocfs2_is_cluster_bitmap(alloc_inode))
1356 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1358 status = ocfs2_journal_access_gd(handle,
1359 INODE_CACHE(alloc_inode),
1367 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1368 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1369 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1370 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1371 le16_to_cpu(bg->bg_bits),
1372 le16_to_cpu(bg->bg_free_bits_count),
1376 ocfs2_set_bit(bit_off++, bitmap);
1378 ocfs2_journal_dirty(handle, group_bh);
1384 /* find the one with the most empty bits */
1385 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1389 BUG_ON(!cl->cl_next_free_rec);
1392 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1393 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1394 le32_to_cpu(cl->cl_recs[best].c_free))
1399 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1403 static int ocfs2_relink_block_group(handle_t *handle,
1404 struct inode *alloc_inode,
1405 struct buffer_head *fe_bh,
1406 struct buffer_head *bg_bh,
1407 struct buffer_head *prev_bg_bh,
1411 /* there is a really tiny chance the journal calls could fail,
1412 * but we wouldn't want inconsistent blocks in *any* case. */
1413 u64 bg_ptr, prev_bg_ptr;
1414 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1415 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1416 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1418 /* The caller got these descriptors from
1419 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1420 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1421 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1423 trace_ocfs2_relink_block_group(
1424 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1425 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1426 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1428 bg_ptr = le64_to_cpu(bg->bg_next_group);
1429 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1431 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1433 OCFS2_JOURNAL_ACCESS_WRITE);
1437 prev_bg->bg_next_group = bg->bg_next_group;
1438 ocfs2_journal_dirty(handle, prev_bg_bh);
1440 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1441 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1443 goto out_rollback_prev_bg;
1445 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1446 ocfs2_journal_dirty(handle, bg_bh);
1448 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1449 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1451 goto out_rollback_bg;
1453 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1454 ocfs2_journal_dirty(handle, fe_bh);
1462 bg->bg_next_group = cpu_to_le64(bg_ptr);
1463 out_rollback_prev_bg:
1464 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1468 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1471 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1474 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1475 * value on error. */
1476 static int ocfs2_cluster_group_search(struct inode *inode,
1477 struct buffer_head *group_bh,
1478 u32 bits_wanted, u32 min_bits,
1480 struct ocfs2_suballoc_result *res)
1482 int search = -ENOSPC;
1485 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1486 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1487 unsigned int max_bits, gd_cluster_off;
1489 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1491 if (gd->bg_free_bits_count) {
1492 max_bits = le16_to_cpu(gd->bg_bits);
1494 /* Tail groups in cluster bitmaps which aren't cpg
1495 * aligned are prone to partial extension by a failed
1496 * fs resize. If the file system resize never got to
1497 * update the dinode cluster count, then we don't want
1498 * to trust any clusters past it, regardless of what
1499 * the group descriptor says. */
1500 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1501 le64_to_cpu(gd->bg_blkno));
1502 if ((gd_cluster_off + max_bits) >
1503 OCFS2_I(inode)->ip_clusters) {
1504 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1505 trace_ocfs2_cluster_group_search_wrong_max_bits(
1506 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1507 le16_to_cpu(gd->bg_bits),
1508 OCFS2_I(inode)->ip_clusters, max_bits);
1511 ret = ocfs2_block_group_find_clear_bits(osb,
1512 group_bh, bits_wanted,
1518 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1520 res->sr_bit_offset +
1522 trace_ocfs2_cluster_group_search_max_block(
1523 (unsigned long long)blkoff,
1524 (unsigned long long)max_block);
1525 if (blkoff > max_block)
1529 /* ocfs2_block_group_find_clear_bits() might
1530 * return success, but we still want to return
1531 * -ENOSPC unless it found the minimum number
1533 if (min_bits <= res->sr_bits)
1534 search = 0; /* success */
1535 else if (res->sr_bits) {
1537 * Don't show bits which we'll be returning
1538 * for allocation to the local alloc bitmap.
1540 ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1547 static int ocfs2_block_group_search(struct inode *inode,
1548 struct buffer_head *group_bh,
1549 u32 bits_wanted, u32 min_bits,
1551 struct ocfs2_suballoc_result *res)
1555 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1557 BUG_ON(min_bits != 1);
1558 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1560 if (bg->bg_free_bits_count) {
1561 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1562 group_bh, bits_wanted,
1563 le16_to_cpu(bg->bg_bits),
1565 if (!ret && max_block) {
1566 blkoff = le64_to_cpu(bg->bg_blkno) +
1567 res->sr_bit_offset + res->sr_bits;
1568 trace_ocfs2_block_group_search_max_block(
1569 (unsigned long long)blkoff,
1570 (unsigned long long)max_block);
1571 if (blkoff > max_block)
1579 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1581 struct buffer_head *di_bh,
1587 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1588 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1590 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1591 OCFS2_JOURNAL_ACCESS_WRITE);
1597 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1598 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1599 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1600 ocfs2_journal_dirty(handle, di_bh);
1606 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1607 struct buffer_head *di_bh,
1612 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1613 struct ocfs2_chain_list *cl;
1615 cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1616 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1617 di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1618 le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1621 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1622 struct ocfs2_extent_rec *rec,
1623 struct ocfs2_chain_list *cl)
1625 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1626 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1627 unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1629 if (res->sr_bit_offset < bitoff)
1631 if (res->sr_bit_offset >= (bitoff + bitcount))
1633 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1634 (res->sr_bit_offset - bitoff);
1635 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1636 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1640 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1641 struct ocfs2_group_desc *bg,
1642 struct ocfs2_suballoc_result *res)
1645 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
1646 struct ocfs2_extent_rec *rec;
1647 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1648 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1650 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1655 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1656 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
1657 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1658 !bg->bg_list.l_next_free_rec)
1661 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1662 rec = &bg->bg_list.l_recs[i];
1663 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1664 res->sr_bg_blkno = bg_blkno; /* Restore */
1670 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1674 struct ocfs2_suballoc_result *res,
1678 struct buffer_head *group_bh = NULL;
1679 struct ocfs2_group_desc *gd;
1680 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1681 struct inode *alloc_inode = ac->ac_inode;
1683 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1684 res->sr_bg_blkno, &group_bh);
1690 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1691 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1692 ac->ac_max_block, res);
1700 ocfs2_bg_discontig_fix_result(ac, gd, res);
1703 * sr_bg_blkno might have been changed by
1704 * ocfs2_bg_discontig_fix_result
1706 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1708 if (ac->ac_find_loc_only)
1711 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1713 le16_to_cpu(gd->bg_chain));
1719 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1720 res->sr_bit_offset, res->sr_bits);
1722 ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1724 le16_to_cpu(gd->bg_chain));
1729 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1737 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1741 struct ocfs2_suballoc_result *res,
1747 struct inode *alloc_inode = ac->ac_inode;
1748 struct buffer_head *group_bh = NULL;
1749 struct buffer_head *prev_group_bh = NULL;
1750 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1751 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1752 struct ocfs2_group_desc *bg;
1754 chain = ac->ac_chain;
1755 trace_ocfs2_search_chain_begin(
1756 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1757 bits_wanted, chain);
1759 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1760 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1766 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1769 /* for now, the chain search is a bit simplistic. We just use
1770 * the 1st group with any empty bits. */
1771 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1772 bits_wanted, min_bits,
1775 if (!bg->bg_next_group)
1778 brelse(prev_group_bh);
1779 prev_group_bh = NULL;
1781 next_group = le64_to_cpu(bg->bg_next_group);
1782 prev_group_bh = group_bh;
1784 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1785 next_group, &group_bh);
1790 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1793 if (status != -ENOSPC)
1798 trace_ocfs2_search_chain_succ(
1799 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1801 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1803 BUG_ON(res->sr_bits == 0);
1805 ocfs2_bg_discontig_fix_result(ac, bg, res);
1808 * sr_bg_blkno might have been changed by
1809 * ocfs2_bg_discontig_fix_result
1811 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1814 * Keep track of previous block descriptor read. When
1815 * we find a target, if we have read more than X
1816 * number of descriptors, and the target is reasonably
1817 * empty, relink him to top of his chain.
1819 * We've read 0 extra blocks and only send one more to
1820 * the transaction, yet the next guy to search has a
1823 * Do this *after* figuring out how many bits we're taking out
1824 * of our target group.
1826 if (!ac->ac_disable_chain_relink &&
1828 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1829 status = ocfs2_relink_block_group(handle, alloc_inode,
1830 ac->ac_bh, group_bh,
1831 prev_group_bh, chain);
1838 if (ac->ac_find_loc_only)
1841 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1842 ac->ac_bh, res->sr_bits,
1849 status = ocfs2_block_group_set_bits(handle,
1856 ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1857 ac->ac_bh, res->sr_bits, chain);
1862 trace_ocfs2_search_chain_end(
1863 (unsigned long long)le64_to_cpu(fe->i_blkno),
1867 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1870 brelse(prev_group_bh);
1877 /* will give out up to bits_wanted contiguous bits. */
1878 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1882 struct ocfs2_suballoc_result *res)
1887 u64 hint = ac->ac_last_group;
1888 struct ocfs2_chain_list *cl;
1889 struct ocfs2_dinode *fe;
1891 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1892 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1895 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1897 /* The bh was validated by the inode read during
1898 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1899 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1901 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1902 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1903 status = ocfs2_error(ac->ac_inode->i_sb,
1904 "Chain allocator dinode %llu has %u used bits but only %u total\n",
1905 (unsigned long long)le64_to_cpu(fe->i_blkno),
1906 le32_to_cpu(fe->id1.bitmap1.i_used),
1907 le32_to_cpu(fe->id1.bitmap1.i_total));
1911 res->sr_bg_blkno = hint;
1912 if (res->sr_bg_blkno) {
1913 /* Attempt to short-circuit the usual search mechanism
1914 * by jumping straight to the most recently used
1915 * allocation group. This helps us maintain some
1916 * contiguousness across allocations. */
1917 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1918 min_bits, res, &bits_left);
1921 if (status < 0 && status != -ENOSPC) {
1927 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1929 victim = ocfs2_find_victim_chain(cl);
1930 ac->ac_chain = victim;
1932 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1935 if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1936 hint = res->sr_bg_blkno;
1938 hint = ocfs2_group_from_res(res);
1941 if (status < 0 && status != -ENOSPC) {
1946 trace_ocfs2_claim_suballoc_bits(victim);
1948 /* If we didn't pick a good victim, then just default to
1949 * searching each chain in order. Don't allow chain relinking
1950 * because we only calculate enough journal credits for one
1951 * relink per alloc. */
1952 ac->ac_disable_chain_relink = 1;
1953 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1956 if (!cl->cl_recs[i].c_free)
1960 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1963 hint = ocfs2_group_from_res(res);
1966 if (status < 0 && status != -ENOSPC) {
1973 if (status != -ENOSPC) {
1974 /* If the next search of this group is not likely to
1975 * yield a suitable extent, then we reset the last
1976 * group hint so as to not waste a disk read */
1977 if (bits_left < min_bits)
1978 ac->ac_last_group = 0;
1980 ac->ac_last_group = hint;
1989 int ocfs2_claim_metadata(handle_t *handle,
1990 struct ocfs2_alloc_context *ac,
1993 u16 *suballoc_bit_start,
1994 unsigned int *num_bits,
1998 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2001 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2002 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2004 status = ocfs2_claim_suballoc_bits(ac,
2013 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2015 *suballoc_loc = res.sr_bg_blkno;
2016 *suballoc_bit_start = res.sr_bit_offset;
2017 *blkno_start = res.sr_blkno;
2018 ac->ac_bits_given += res.sr_bits;
2019 *num_bits = res.sr_bits;
2027 static void ocfs2_init_inode_ac_group(struct inode *dir,
2028 struct buffer_head *parent_di_bh,
2029 struct ocfs2_alloc_context *ac)
2031 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2033 * Try to allocate inodes from some specific group.
2035 * If the parent dir has recorded the last group used in allocation,
2036 * cool, use it. Otherwise if we try to allocate new inode from the
2037 * same slot the parent dir belongs to, use the same chunk.
2039 * We are very careful here to avoid the mistake of setting
2040 * ac_last_group to a group descriptor from a different (unlocked) slot.
2042 if (OCFS2_I(dir)->ip_last_used_group &&
2043 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2044 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2045 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2046 if (di->i_suballoc_loc)
2047 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2049 ac->ac_last_group = ocfs2_which_suballoc_group(
2050 le64_to_cpu(di->i_blkno),
2051 le16_to_cpu(di->i_suballoc_bit));
2055 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2056 struct ocfs2_alloc_context *ac)
2058 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2059 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2062 int ocfs2_find_new_inode_loc(struct inode *dir,
2063 struct buffer_head *parent_fe_bh,
2064 struct ocfs2_alloc_context *ac,
2068 handle_t *handle = NULL;
2069 struct ocfs2_suballoc_result *res;
2072 BUG_ON(ac->ac_bits_given != 0);
2073 BUG_ON(ac->ac_bits_wanted != 1);
2074 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2076 res = kzalloc(sizeof(*res), GFP_NOFS);
2083 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2086 * The handle started here is for chain relink. Alternatively,
2087 * we could just disable relink for these calls.
2089 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2090 if (IS_ERR(handle)) {
2091 ret = PTR_ERR(handle);
2098 * This will instruct ocfs2_claim_suballoc_bits and
2099 * ocfs2_search_one_group to search but save actual allocation
2102 ac->ac_find_loc_only = 1;
2104 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2110 ac->ac_find_loc_priv = res;
2111 *fe_blkno = res->sr_blkno;
2112 ocfs2_update_inode_fsync_trans(handle, dir, 0);
2115 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2123 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2125 struct ocfs2_alloc_context *ac,
2132 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2133 struct buffer_head *bg_bh = NULL;
2134 struct ocfs2_group_desc *bg;
2135 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2138 * Since di_blkno is being passed back in, we check for any
2139 * inconsistencies which may have happened between
2140 * calls. These are code bugs as di_blkno is not expected to
2141 * change once returned from ocfs2_find_new_inode_loc()
2143 BUG_ON(res->sr_blkno != di_blkno);
2145 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2146 res->sr_bg_stable_blkno, &bg_bh);
2152 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2153 chain = le16_to_cpu(bg->bg_chain);
2155 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2156 ac->ac_bh, res->sr_bits,
2163 ret = ocfs2_block_group_set_bits(handle,
2170 ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2171 ac->ac_bh, res->sr_bits, chain);
2176 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2179 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2181 BUG_ON(res->sr_bits != 1);
2183 *suballoc_loc = res->sr_bg_blkno;
2184 *suballoc_bit = res->sr_bit_offset;
2185 ac->ac_bits_given++;
2186 ocfs2_save_inode_ac_group(dir, ac);
2194 int ocfs2_claim_new_inode(handle_t *handle,
2196 struct buffer_head *parent_fe_bh,
2197 struct ocfs2_alloc_context *ac,
2203 struct ocfs2_suballoc_result res;
2206 BUG_ON(ac->ac_bits_given != 0);
2207 BUG_ON(ac->ac_bits_wanted != 1);
2208 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2210 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2212 status = ocfs2_claim_suballoc_bits(ac,
2221 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2223 BUG_ON(res.sr_bits != 1);
2225 *suballoc_loc = res.sr_bg_blkno;
2226 *suballoc_bit = res.sr_bit_offset;
2227 *fe_blkno = res.sr_blkno;
2228 ac->ac_bits_given++;
2229 ocfs2_save_inode_ac_group(dir, ac);
2237 /* translate a group desc. blkno and it's bitmap offset into
2238 * disk cluster offset. */
2239 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2243 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2246 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2248 if (bg_blkno != osb->first_cluster_group_blkno)
2249 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2250 cluster += (u32) bg_bit_off;
2254 /* given a cluster offset, calculate which block group it belongs to
2255 * and return that block offset. */
2256 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2258 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2261 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2263 group_no = cluster / osb->bitmap_cpg;
2265 return osb->first_cluster_group_blkno;
2266 return ocfs2_clusters_to_blocks(inode->i_sb,
2267 group_no * osb->bitmap_cpg);
2270 /* given the block number of a cluster start, calculate which cluster
2271 * group and descriptor bitmap offset that corresponds to. */
2272 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2277 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2278 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2280 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2282 *bg_blkno = ocfs2_which_cluster_group(inode,
2285 if (*bg_blkno == osb->first_cluster_group_blkno)
2286 *bg_bit_off = (u16) data_cluster;
2288 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2289 data_blkno - *bg_blkno);
2293 * min_bits - minimum contiguous chunk from this total allocation we
2294 * can handle. set to what we asked for originally for a full
2295 * contig. allocation, set to '1' to indicate we can deal with extents
2298 int __ocfs2_claim_clusters(handle_t *handle,
2299 struct ocfs2_alloc_context *ac,
2306 unsigned int bits_wanted = max_clusters;
2307 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2308 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2310 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2312 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2313 && ac->ac_which != OCFS2_AC_USE_MAIN);
2315 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2316 WARN_ON(min_clusters > 1);
2318 status = ocfs2_claim_local_alloc_bits(osb,
2325 atomic_inc(&osb->alloc_stats.local_data);
2327 if (min_clusters > (osb->bitmap_cpg - 1)) {
2328 /* The only paths asking for contiguousness
2329 * should know about this already. */
2330 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2331 "group bitmap size %u!\n", min_clusters,
2336 /* clamp the current request down to a realistic size. */
2337 if (bits_wanted > (osb->bitmap_cpg - 1))
2338 bits_wanted = osb->bitmap_cpg - 1;
2340 status = ocfs2_claim_suballoc_bits(ac,
2346 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2348 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2351 atomic_inc(&osb->alloc_stats.bitmap_data);
2352 *num_clusters = res.sr_bits;
2356 if (status != -ENOSPC)
2361 ac->ac_bits_given += *num_clusters;
2369 int ocfs2_claim_clusters(handle_t *handle,
2370 struct ocfs2_alloc_context *ac,
2375 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2377 return __ocfs2_claim_clusters(handle, ac, min_clusters,
2378 bits_wanted, cluster_start, num_clusters);
2381 static int ocfs2_block_group_clear_bits(handle_t *handle,
2382 struct inode *alloc_inode,
2383 struct ocfs2_group_desc *bg,
2384 struct buffer_head *group_bh,
2385 unsigned int bit_off,
2386 unsigned int num_bits,
2387 void (*undo_fn)(unsigned int bit,
2388 unsigned long *bmap))
2392 struct ocfs2_group_desc *undo_bg = NULL;
2393 struct journal_head *jh;
2395 /* The caller got this descriptor from
2396 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2397 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2399 trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2401 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2402 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2405 OCFS2_JOURNAL_ACCESS_UNDO :
2406 OCFS2_JOURNAL_ACCESS_WRITE);
2412 jh = bh2jh(group_bh);
2414 spin_lock(&jh->b_state_lock);
2415 undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2421 ocfs2_clear_bit((bit_off + tmp),
2422 (unsigned long *) bg->bg_bitmap);
2424 undo_fn(bit_off + tmp,
2425 (unsigned long *) undo_bg->bg_bitmap);
2427 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2428 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2430 spin_unlock(&jh->b_state_lock);
2431 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2432 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2433 le16_to_cpu(bg->bg_bits),
2434 le16_to_cpu(bg->bg_free_bits_count),
2439 spin_unlock(&jh->b_state_lock);
2441 ocfs2_journal_dirty(handle, group_bh);
2447 * expects the suballoc inode to already be locked.
2449 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2450 struct inode *alloc_inode,
2451 struct buffer_head *alloc_bh,
2452 unsigned int start_bit,
2455 void (*undo_fn)(unsigned int bit,
2456 unsigned long *bitmap))
2460 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2461 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2462 struct buffer_head *group_bh = NULL;
2463 struct ocfs2_group_desc *group;
2465 /* The alloc_bh comes from ocfs2_free_dinode() or
2466 * ocfs2_free_clusters(). The callers have all locked the
2467 * allocator and gotten alloc_bh from the lock call. This
2468 * validates the dinode buffer. Any corruption that has happened
2470 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2471 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2473 trace_ocfs2_free_suballoc_bits(
2474 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2475 (unsigned long long)bg_blkno,
2478 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2484 group = (struct ocfs2_group_desc *) group_bh->b_data;
2486 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2488 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2490 start_bit, count, undo_fn);
2496 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2497 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2500 ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2505 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2507 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2508 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2509 ocfs2_journal_dirty(handle, alloc_bh);
2516 int ocfs2_free_suballoc_bits(handle_t *handle,
2517 struct inode *alloc_inode,
2518 struct buffer_head *alloc_bh,
2519 unsigned int start_bit,
2523 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2524 start_bit, bg_blkno, count, NULL);
2527 int ocfs2_free_dinode(handle_t *handle,
2528 struct inode *inode_alloc_inode,
2529 struct buffer_head *inode_alloc_bh,
2530 struct ocfs2_dinode *di)
2532 u64 blk = le64_to_cpu(di->i_blkno);
2533 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2534 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2536 if (di->i_suballoc_loc)
2537 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2538 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2539 inode_alloc_bh, bit, bg_blkno, 1);
2542 static int _ocfs2_free_clusters(handle_t *handle,
2543 struct inode *bitmap_inode,
2544 struct buffer_head *bitmap_bh,
2546 unsigned int num_clusters,
2547 void (*undo_fn)(unsigned int bit,
2548 unsigned long *bitmap))
2554 /* You can't ever have a contiguous set of clusters
2555 * bigger than a block group bitmap so we never have to worry
2556 * about looping on them.
2557 * This is expensive. We can safely remove once this stuff has
2558 * gotten tested really well. */
2559 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2560 ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2564 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2567 trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2568 (unsigned long long)start_blk,
2569 bg_start_bit, num_clusters);
2571 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2572 bg_start_bit, bg_blkno,
2573 num_clusters, undo_fn);
2579 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2586 int ocfs2_free_clusters(handle_t *handle,
2587 struct inode *bitmap_inode,
2588 struct buffer_head *bitmap_bh,
2590 unsigned int num_clusters)
2592 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2593 start_blk, num_clusters,
2598 * Give never-used clusters back to the global bitmap. We don't need
2599 * to protect these bits in the undo buffer.
2601 int ocfs2_release_clusters(handle_t *handle,
2602 struct inode *bitmap_inode,
2603 struct buffer_head *bitmap_bh,
2605 unsigned int num_clusters)
2607 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2608 start_blk, num_clusters,
2613 * For a given allocation, determine which allocators will need to be
2614 * accessed, and lock them, reserving the appropriate number of bits.
2616 * Sparse file systems call this from ocfs2_write_begin_nolock()
2617 * and ocfs2_allocate_unwritten_extents().
2619 * File systems which don't support holes call this from
2620 * ocfs2_extend_allocation().
2622 int ocfs2_lock_allocators(struct inode *inode,
2623 struct ocfs2_extent_tree *et,
2624 u32 clusters_to_add, u32 extents_to_split,
2625 struct ocfs2_alloc_context **data_ac,
2626 struct ocfs2_alloc_context **meta_ac)
2628 int ret = 0, num_free_extents;
2629 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2630 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2636 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2638 num_free_extents = ocfs2_num_free_extents(et);
2639 if (num_free_extents < 0) {
2640 ret = num_free_extents;
2646 * Sparse allocation file systems need to be more conservative
2647 * with reserving room for expansion - the actual allocation
2648 * happens while we've got a journal handle open so re-taking
2649 * a cluster lock (because we ran out of room for another
2650 * extent) will violate ordering rules.
2652 * Most of the time we'll only be seeing this 1 cluster at a time
2655 * Always lock for any unwritten extents - we might want to
2656 * add blocks during a split.
2658 if (!num_free_extents ||
2659 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2660 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2668 if (clusters_to_add == 0)
2671 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2681 ocfs2_free_alloc_context(*meta_ac);
2686 * We cannot have an error and a non null *data_ac.
2694 * Read the inode specified by blkno to get suballoc_slot and
2697 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2698 u16 *suballoc_slot, u64 *group_blkno,
2702 struct buffer_head *inode_bh = NULL;
2703 struct ocfs2_dinode *inode_fe;
2705 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2707 /* dirty read disk */
2708 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2710 mlog(ML_ERROR, "read block %llu failed %d\n",
2711 (unsigned long long)blkno, status);
2715 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2716 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2717 mlog(ML_ERROR, "invalid inode %llu requested\n",
2718 (unsigned long long)blkno);
2723 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2724 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2725 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2726 (unsigned long long)blkno,
2727 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2733 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2735 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2737 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2748 * test whether bit is SET in allocator bitmap or not. on success, 0
2749 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2750 * is returned and *res is meaningless. Call this after you have
2751 * cluster locked against suballoc, or you may get a result based on
2752 * non-up2date contents
2754 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2755 struct inode *suballoc,
2756 struct buffer_head *alloc_bh,
2757 u64 group_blkno, u64 blkno,
2760 struct ocfs2_dinode *alloc_di;
2761 struct ocfs2_group_desc *group;
2762 struct buffer_head *group_bh = NULL;
2766 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2769 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2770 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2771 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2773 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2778 bg_blkno = group_blkno ? group_blkno :
2779 ocfs2_which_suballoc_group(blkno, bit);
2780 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2783 mlog(ML_ERROR, "read group %llu failed %d\n",
2784 (unsigned long long)bg_blkno, status);
2788 group = (struct ocfs2_group_desc *) group_bh->b_data;
2789 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2800 * Test if the bit representing this inode (blkno) is set in the
2803 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2805 * In the event of failure, a negative value is returned and *res is
2808 * Callers must make sure to hold nfs_sync_lock to prevent
2809 * ocfs2_delete_inode() on another node from accessing the same
2810 * suballocator concurrently.
2812 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2815 u64 group_blkno = 0;
2816 u16 suballoc_bit = 0, suballoc_slot = 0;
2817 struct inode *inode_alloc_inode;
2818 struct buffer_head *alloc_bh = NULL;
2820 trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2822 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2823 &group_blkno, &suballoc_bit);
2825 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2829 if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2830 inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2831 GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2833 inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2834 INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2835 if (!inode_alloc_inode) {
2836 /* the error code could be inaccurate, but we are not able to
2837 * get the correct one. */
2839 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2840 (u32)suballoc_slot);
2844 inode_lock(inode_alloc_inode);
2845 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2847 inode_unlock(inode_alloc_inode);
2848 iput(inode_alloc_inode);
2849 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2850 (u32)suballoc_slot, status);
2854 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2855 group_blkno, blkno, suballoc_bit, res);
2857 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2859 ocfs2_inode_unlock(inode_alloc_inode, 0);
2860 inode_unlock(inode_alloc_inode);
2862 iput(inode_alloc_inode);