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
= slot
;
883 else if (type
== EXTENT_ALLOC_SYSTEM_INODE
)
884 osb
->s_meta_steal_slot
= 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 if (!buffer_jbd(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
);
1276 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super
*osb
,
1277 struct buffer_head
*bg_bh
,
1278 unsigned int bits_wanted
,
1279 unsigned int total_bits
,
1280 struct ocfs2_suballoc_result
*res
)
1283 u16 best_offset
, best_size
;
1284 int offset
, start
, found
, status
= 0;
1285 struct ocfs2_group_desc
*bg
= (struct ocfs2_group_desc
*) bg_bh
->b_data
;
1287 /* Callers got this descriptor from
1288 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1289 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg
));
1291 found
= start
= best_offset
= best_size
= 0;
1292 bitmap
= bg
->bg_bitmap
;
1294 while((offset
= ocfs2_find_next_zero_bit(bitmap
, total_bits
, start
)) != -1) {
1295 if (offset
== total_bits
)
1298 if (!ocfs2_test_bg_bit_allocatable(bg_bh
, offset
)) {
1299 /* We found a zero, but we can't use it as it
1300 * hasn't been put to disk yet! */
1303 } else if (offset
== start
) {
1304 /* we found a zero */
1306 /* move start to the next bit to test */
1309 /* got a zero after some ones */
1313 if (found
> best_size
) {
1315 best_offset
= start
- found
;
1317 /* we got everything we needed */
1318 if (found
== bits_wanted
) {
1319 /* mlog(0, "Found it all!\n"); */
1325 res
->sr_bit_offset
= best_offset
;
1326 res
->sr_bits
= best_size
;
1329 /* No error log here -- see the comment above
1330 * ocfs2_test_bg_bit_allocatable */
1336 int ocfs2_block_group_set_bits(handle_t
*handle
,
1337 struct inode
*alloc_inode
,
1338 struct ocfs2_group_desc
*bg
,
1339 struct buffer_head
*group_bh
,
1340 unsigned int bit_off
,
1341 unsigned int num_bits
)
1344 void *bitmap
= bg
->bg_bitmap
;
1345 int journal_type
= OCFS2_JOURNAL_ACCESS_WRITE
;
1347 /* All callers get the descriptor via
1348 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1349 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg
));
1350 BUG_ON(le16_to_cpu(bg
->bg_free_bits_count
) < num_bits
);
1352 trace_ocfs2_block_group_set_bits(bit_off
, num_bits
);
1354 if (ocfs2_is_cluster_bitmap(alloc_inode
))
1355 journal_type
= OCFS2_JOURNAL_ACCESS_UNDO
;
1357 status
= ocfs2_journal_access_gd(handle
,
1358 INODE_CACHE(alloc_inode
),
1366 le16_add_cpu(&bg
->bg_free_bits_count
, -num_bits
);
1367 if (le16_to_cpu(bg
->bg_free_bits_count
) > le16_to_cpu(bg
->bg_bits
)) {
1368 return ocfs2_error(alloc_inode
->i_sb
, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1369 (unsigned long long)le64_to_cpu(bg
->bg_blkno
),
1370 le16_to_cpu(bg
->bg_bits
),
1371 le16_to_cpu(bg
->bg_free_bits_count
),
1375 ocfs2_set_bit(bit_off
++, bitmap
);
1377 ocfs2_journal_dirty(handle
, group_bh
);
1383 /* find the one with the most empty bits */
1384 static inline u16
ocfs2_find_victim_chain(struct ocfs2_chain_list
*cl
)
1388 BUG_ON(!cl
->cl_next_free_rec
);
1391 while (curr
< le16_to_cpu(cl
->cl_next_free_rec
)) {
1392 if (le32_to_cpu(cl
->cl_recs
[curr
].c_free
) >
1393 le32_to_cpu(cl
->cl_recs
[best
].c_free
))
1398 BUG_ON(best
>= le16_to_cpu(cl
->cl_next_free_rec
));
1402 static int ocfs2_relink_block_group(handle_t
*handle
,
1403 struct inode
*alloc_inode
,
1404 struct buffer_head
*fe_bh
,
1405 struct buffer_head
*bg_bh
,
1406 struct buffer_head
*prev_bg_bh
,
1410 /* there is a really tiny chance the journal calls could fail,
1411 * but we wouldn't want inconsistent blocks in *any* case. */
1412 u64 bg_ptr
, prev_bg_ptr
;
1413 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) fe_bh
->b_data
;
1414 struct ocfs2_group_desc
*bg
= (struct ocfs2_group_desc
*) bg_bh
->b_data
;
1415 struct ocfs2_group_desc
*prev_bg
= (struct ocfs2_group_desc
*) prev_bg_bh
->b_data
;
1417 /* The caller got these descriptors from
1418 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1419 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg
));
1420 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg
));
1422 trace_ocfs2_relink_block_group(
1423 (unsigned long long)le64_to_cpu(fe
->i_blkno
), chain
,
1424 (unsigned long long)le64_to_cpu(bg
->bg_blkno
),
1425 (unsigned long long)le64_to_cpu(prev_bg
->bg_blkno
));
1427 bg_ptr
= le64_to_cpu(bg
->bg_next_group
);
1428 prev_bg_ptr
= le64_to_cpu(prev_bg
->bg_next_group
);
1430 status
= ocfs2_journal_access_gd(handle
, INODE_CACHE(alloc_inode
),
1432 OCFS2_JOURNAL_ACCESS_WRITE
);
1436 prev_bg
->bg_next_group
= bg
->bg_next_group
;
1437 ocfs2_journal_dirty(handle
, prev_bg_bh
);
1439 status
= ocfs2_journal_access_gd(handle
, INODE_CACHE(alloc_inode
),
1440 bg_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
1442 goto out_rollback_prev_bg
;
1444 bg
->bg_next_group
= fe
->id2
.i_chain
.cl_recs
[chain
].c_blkno
;
1445 ocfs2_journal_dirty(handle
, bg_bh
);
1447 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(alloc_inode
),
1448 fe_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
1450 goto out_rollback_bg
;
1452 fe
->id2
.i_chain
.cl_recs
[chain
].c_blkno
= bg
->bg_blkno
;
1453 ocfs2_journal_dirty(handle
, fe_bh
);
1461 bg
->bg_next_group
= cpu_to_le64(bg_ptr
);
1462 out_rollback_prev_bg
:
1463 prev_bg
->bg_next_group
= cpu_to_le64(prev_bg_ptr
);
1467 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc
*bg
,
1470 return le16_to_cpu(bg
->bg_free_bits_count
) > wanted
;
1473 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1474 * value on error. */
1475 static int ocfs2_cluster_group_search(struct inode
*inode
,
1476 struct buffer_head
*group_bh
,
1477 u32 bits_wanted
, u32 min_bits
,
1479 struct ocfs2_suballoc_result
*res
)
1481 int search
= -ENOSPC
;
1484 struct ocfs2_group_desc
*gd
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
1485 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
1486 unsigned int max_bits
, gd_cluster_off
;
1488 BUG_ON(!ocfs2_is_cluster_bitmap(inode
));
1490 if (gd
->bg_free_bits_count
) {
1491 max_bits
= le16_to_cpu(gd
->bg_bits
);
1493 /* Tail groups in cluster bitmaps which aren't cpg
1494 * aligned are prone to partial extension by a failed
1495 * fs resize. If the file system resize never got to
1496 * update the dinode cluster count, then we don't want
1497 * to trust any clusters past it, regardless of what
1498 * the group descriptor says. */
1499 gd_cluster_off
= ocfs2_blocks_to_clusters(inode
->i_sb
,
1500 le64_to_cpu(gd
->bg_blkno
));
1501 if ((gd_cluster_off
+ max_bits
) >
1502 OCFS2_I(inode
)->ip_clusters
) {
1503 max_bits
= OCFS2_I(inode
)->ip_clusters
- gd_cluster_off
;
1504 trace_ocfs2_cluster_group_search_wrong_max_bits(
1505 (unsigned long long)le64_to_cpu(gd
->bg_blkno
),
1506 le16_to_cpu(gd
->bg_bits
),
1507 OCFS2_I(inode
)->ip_clusters
, max_bits
);
1510 ret
= ocfs2_block_group_find_clear_bits(osb
,
1511 group_bh
, bits_wanted
,
1517 blkoff
= ocfs2_clusters_to_blocks(inode
->i_sb
,
1519 res
->sr_bit_offset
+
1521 trace_ocfs2_cluster_group_search_max_block(
1522 (unsigned long long)blkoff
,
1523 (unsigned long long)max_block
);
1524 if (blkoff
> max_block
)
1528 /* ocfs2_block_group_find_clear_bits() might
1529 * return success, but we still want to return
1530 * -ENOSPC unless it found the minimum number
1532 if (min_bits
<= res
->sr_bits
)
1533 search
= 0; /* success */
1534 else if (res
->sr_bits
) {
1536 * Don't show bits which we'll be returning
1537 * for allocation to the local alloc bitmap.
1539 ocfs2_local_alloc_seen_free_bits(osb
, res
->sr_bits
);
1546 static int ocfs2_block_group_search(struct inode
*inode
,
1547 struct buffer_head
*group_bh
,
1548 u32 bits_wanted
, u32 min_bits
,
1550 struct ocfs2_suballoc_result
*res
)
1554 struct ocfs2_group_desc
*bg
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
1556 BUG_ON(min_bits
!= 1);
1557 BUG_ON(ocfs2_is_cluster_bitmap(inode
));
1559 if (bg
->bg_free_bits_count
) {
1560 ret
= ocfs2_block_group_find_clear_bits(OCFS2_SB(inode
->i_sb
),
1561 group_bh
, bits_wanted
,
1562 le16_to_cpu(bg
->bg_bits
),
1564 if (!ret
&& max_block
) {
1565 blkoff
= le64_to_cpu(bg
->bg_blkno
) +
1566 res
->sr_bit_offset
+ res
->sr_bits
;
1567 trace_ocfs2_block_group_search_max_block(
1568 (unsigned long long)blkoff
,
1569 (unsigned long long)max_block
);
1570 if (blkoff
> max_block
)
1578 int ocfs2_alloc_dinode_update_counts(struct inode
*inode
,
1580 struct buffer_head
*di_bh
,
1586 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
1587 struct ocfs2_chain_list
*cl
= (struct ocfs2_chain_list
*) &di
->id2
.i_chain
;
1589 ret
= ocfs2_journal_access_di(handle
, INODE_CACHE(inode
), di_bh
,
1590 OCFS2_JOURNAL_ACCESS_WRITE
);
1596 tmp_used
= le32_to_cpu(di
->id1
.bitmap1
.i_used
);
1597 di
->id1
.bitmap1
.i_used
= cpu_to_le32(num_bits
+ tmp_used
);
1598 le32_add_cpu(&cl
->cl_recs
[chain
].c_free
, -num_bits
);
1599 ocfs2_journal_dirty(handle
, di_bh
);
1605 void ocfs2_rollback_alloc_dinode_counts(struct inode
*inode
,
1606 struct buffer_head
*di_bh
,
1611 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
1612 struct ocfs2_chain_list
*cl
;
1614 cl
= (struct ocfs2_chain_list
*)&di
->id2
.i_chain
;
1615 tmp_used
= le32_to_cpu(di
->id1
.bitmap1
.i_used
);
1616 di
->id1
.bitmap1
.i_used
= cpu_to_le32(tmp_used
- num_bits
);
1617 le32_add_cpu(&cl
->cl_recs
[chain
].c_free
, num_bits
);
1620 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result
*res
,
1621 struct ocfs2_extent_rec
*rec
,
1622 struct ocfs2_chain_list
*cl
)
1624 unsigned int bpc
= le16_to_cpu(cl
->cl_bpc
);
1625 unsigned int bitoff
= le32_to_cpu(rec
->e_cpos
) * bpc
;
1626 unsigned int bitcount
= le16_to_cpu(rec
->e_leaf_clusters
) * bpc
;
1628 if (res
->sr_bit_offset
< bitoff
)
1630 if (res
->sr_bit_offset
>= (bitoff
+ bitcount
))
1632 res
->sr_blkno
= le64_to_cpu(rec
->e_blkno
) +
1633 (res
->sr_bit_offset
- bitoff
);
1634 if ((res
->sr_bit_offset
+ res
->sr_bits
) > (bitoff
+ bitcount
))
1635 res
->sr_bits
= (bitoff
+ bitcount
) - res
->sr_bit_offset
;
1639 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context
*ac
,
1640 struct ocfs2_group_desc
*bg
,
1641 struct ocfs2_suballoc_result
*res
)
1644 u64 bg_blkno
= res
->sr_bg_blkno
; /* Save off */
1645 struct ocfs2_extent_rec
*rec
;
1646 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)ac
->ac_bh
->b_data
;
1647 struct ocfs2_chain_list
*cl
= &di
->id2
.i_chain
;
1649 if (ocfs2_is_cluster_bitmap(ac
->ac_inode
)) {
1654 res
->sr_blkno
= res
->sr_bg_blkno
+ res
->sr_bit_offset
;
1655 res
->sr_bg_blkno
= 0; /* Clear it for contig block groups */
1656 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac
->ac_inode
->i_sb
)) ||
1657 !bg
->bg_list
.l_next_free_rec
)
1660 for (i
= 0; i
< le16_to_cpu(bg
->bg_list
.l_next_free_rec
); i
++) {
1661 rec
= &bg
->bg_list
.l_recs
[i
];
1662 if (ocfs2_bg_discontig_fix_by_rec(res
, rec
, cl
)) {
1663 res
->sr_bg_blkno
= bg_blkno
; /* Restore */
1669 static int ocfs2_search_one_group(struct ocfs2_alloc_context
*ac
,
1673 struct ocfs2_suballoc_result
*res
,
1677 struct buffer_head
*group_bh
= NULL
;
1678 struct ocfs2_group_desc
*gd
;
1679 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)ac
->ac_bh
->b_data
;
1680 struct inode
*alloc_inode
= ac
->ac_inode
;
1682 ret
= ocfs2_read_group_descriptor(alloc_inode
, di
,
1683 res
->sr_bg_blkno
, &group_bh
);
1689 gd
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
1690 ret
= ac
->ac_group_search(alloc_inode
, group_bh
, bits_wanted
, min_bits
,
1691 ac
->ac_max_block
, res
);
1699 ocfs2_bg_discontig_fix_result(ac
, gd
, res
);
1702 * sr_bg_blkno might have been changed by
1703 * ocfs2_bg_discontig_fix_result
1705 res
->sr_bg_stable_blkno
= group_bh
->b_blocknr
;
1707 if (ac
->ac_find_loc_only
)
1710 ret
= ocfs2_alloc_dinode_update_counts(alloc_inode
, handle
, ac
->ac_bh
,
1712 le16_to_cpu(gd
->bg_chain
));
1718 ret
= ocfs2_block_group_set_bits(handle
, alloc_inode
, gd
, group_bh
,
1719 res
->sr_bit_offset
, res
->sr_bits
);
1721 ocfs2_rollback_alloc_dinode_counts(alloc_inode
, ac
->ac_bh
,
1723 le16_to_cpu(gd
->bg_chain
));
1728 *bits_left
= le16_to_cpu(gd
->bg_free_bits_count
);
1736 static int ocfs2_search_chain(struct ocfs2_alloc_context
*ac
,
1740 struct ocfs2_suballoc_result
*res
,
1746 struct inode
*alloc_inode
= ac
->ac_inode
;
1747 struct buffer_head
*group_bh
= NULL
;
1748 struct buffer_head
*prev_group_bh
= NULL
;
1749 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) ac
->ac_bh
->b_data
;
1750 struct ocfs2_chain_list
*cl
= (struct ocfs2_chain_list
*) &fe
->id2
.i_chain
;
1751 struct ocfs2_group_desc
*bg
;
1753 chain
= ac
->ac_chain
;
1754 trace_ocfs2_search_chain_begin(
1755 (unsigned long long)OCFS2_I(alloc_inode
)->ip_blkno
,
1756 bits_wanted
, chain
);
1758 status
= ocfs2_read_group_descriptor(alloc_inode
, fe
,
1759 le64_to_cpu(cl
->cl_recs
[chain
].c_blkno
),
1765 bg
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
1768 /* for now, the chain search is a bit simplistic. We just use
1769 * the 1st group with any empty bits. */
1770 while ((status
= ac
->ac_group_search(alloc_inode
, group_bh
,
1771 bits_wanted
, min_bits
,
1774 if (!bg
->bg_next_group
)
1777 brelse(prev_group_bh
);
1778 prev_group_bh
= NULL
;
1780 next_group
= le64_to_cpu(bg
->bg_next_group
);
1781 prev_group_bh
= group_bh
;
1783 status
= ocfs2_read_group_descriptor(alloc_inode
, fe
,
1784 next_group
, &group_bh
);
1789 bg
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
1792 if (status
!= -ENOSPC
)
1797 trace_ocfs2_search_chain_succ(
1798 (unsigned long long)le64_to_cpu(bg
->bg_blkno
), res
->sr_bits
);
1800 res
->sr_bg_blkno
= le64_to_cpu(bg
->bg_blkno
);
1802 BUG_ON(res
->sr_bits
== 0);
1804 ocfs2_bg_discontig_fix_result(ac
, bg
, res
);
1807 * sr_bg_blkno might have been changed by
1808 * ocfs2_bg_discontig_fix_result
1810 res
->sr_bg_stable_blkno
= group_bh
->b_blocknr
;
1813 * Keep track of previous block descriptor read. When
1814 * we find a target, if we have read more than X
1815 * number of descriptors, and the target is reasonably
1816 * empty, relink him to top of his chain.
1818 * We've read 0 extra blocks and only send one more to
1819 * the transaction, yet the next guy to search has a
1822 * Do this *after* figuring out how many bits we're taking out
1823 * of our target group.
1825 if (!ac
->ac_disable_chain_relink
&&
1827 (ocfs2_block_group_reasonably_empty(bg
, res
->sr_bits
))) {
1828 status
= ocfs2_relink_block_group(handle
, alloc_inode
,
1829 ac
->ac_bh
, group_bh
,
1830 prev_group_bh
, chain
);
1837 if (ac
->ac_find_loc_only
)
1840 status
= ocfs2_alloc_dinode_update_counts(alloc_inode
, handle
,
1841 ac
->ac_bh
, res
->sr_bits
,
1848 status
= ocfs2_block_group_set_bits(handle
,
1855 ocfs2_rollback_alloc_dinode_counts(alloc_inode
,
1856 ac
->ac_bh
, res
->sr_bits
, chain
);
1861 trace_ocfs2_search_chain_end(
1862 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
1866 *bits_left
= le16_to_cpu(bg
->bg_free_bits_count
);
1869 brelse(prev_group_bh
);
1876 /* will give out up to bits_wanted contiguous bits. */
1877 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context
*ac
,
1881 struct ocfs2_suballoc_result
*res
)
1886 u64 hint
= ac
->ac_last_group
;
1887 struct ocfs2_chain_list
*cl
;
1888 struct ocfs2_dinode
*fe
;
1890 BUG_ON(ac
->ac_bits_given
>= ac
->ac_bits_wanted
);
1891 BUG_ON(bits_wanted
> (ac
->ac_bits_wanted
- ac
->ac_bits_given
));
1894 fe
= (struct ocfs2_dinode
*) ac
->ac_bh
->b_data
;
1896 /* The bh was validated by the inode read during
1897 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1898 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
));
1900 if (le32_to_cpu(fe
->id1
.bitmap1
.i_used
) >=
1901 le32_to_cpu(fe
->id1
.bitmap1
.i_total
)) {
1902 status
= ocfs2_error(ac
->ac_inode
->i_sb
,
1903 "Chain allocator dinode %llu has %u used bits but only %u total\n",
1904 (unsigned long long)le64_to_cpu(fe
->i_blkno
),
1905 le32_to_cpu(fe
->id1
.bitmap1
.i_used
),
1906 le32_to_cpu(fe
->id1
.bitmap1
.i_total
));
1910 res
->sr_bg_blkno
= hint
;
1911 if (res
->sr_bg_blkno
) {
1912 /* Attempt to short-circuit the usual search mechanism
1913 * by jumping straight to the most recently used
1914 * allocation group. This helps us maintain some
1915 * contiguousness across allocations. */
1916 status
= ocfs2_search_one_group(ac
, handle
, bits_wanted
,
1917 min_bits
, res
, &bits_left
);
1920 if (status
< 0 && status
!= -ENOSPC
) {
1926 cl
= (struct ocfs2_chain_list
*) &fe
->id2
.i_chain
;
1928 victim
= ocfs2_find_victim_chain(cl
);
1929 ac
->ac_chain
= victim
;
1931 status
= ocfs2_search_chain(ac
, handle
, bits_wanted
, min_bits
,
1934 if (ocfs2_is_cluster_bitmap(ac
->ac_inode
))
1935 hint
= res
->sr_bg_blkno
;
1937 hint
= ocfs2_group_from_res(res
);
1940 if (status
< 0 && status
!= -ENOSPC
) {
1945 trace_ocfs2_claim_suballoc_bits(victim
);
1947 /* If we didn't pick a good victim, then just default to
1948 * searching each chain in order. Don't allow chain relinking
1949 * because we only calculate enough journal credits for one
1950 * relink per alloc. */
1951 ac
->ac_disable_chain_relink
= 1;
1952 for (i
= 0; i
< le16_to_cpu(cl
->cl_next_free_rec
); i
++) {
1955 if (!cl
->cl_recs
[i
].c_free
)
1959 status
= ocfs2_search_chain(ac
, handle
, bits_wanted
, min_bits
,
1962 hint
= ocfs2_group_from_res(res
);
1965 if (status
< 0 && status
!= -ENOSPC
) {
1972 if (status
!= -ENOSPC
) {
1973 /* If the next search of this group is not likely to
1974 * yield a suitable extent, then we reset the last
1975 * group hint so as to not waste a disk read */
1976 if (bits_left
< min_bits
)
1977 ac
->ac_last_group
= 0;
1979 ac
->ac_last_group
= hint
;
1988 int ocfs2_claim_metadata(handle_t
*handle
,
1989 struct ocfs2_alloc_context
*ac
,
1992 u16
*suballoc_bit_start
,
1993 unsigned int *num_bits
,
1997 struct ocfs2_suballoc_result res
= { .sr_blkno
= 0, };
2000 BUG_ON(ac
->ac_bits_wanted
< (ac
->ac_bits_given
+ bits_wanted
));
2001 BUG_ON(ac
->ac_which
!= OCFS2_AC_USE_META
);
2003 status
= ocfs2_claim_suballoc_bits(ac
,
2012 atomic_inc(&OCFS2_SB(ac
->ac_inode
->i_sb
)->alloc_stats
.bg_allocs
);
2014 *suballoc_loc
= res
.sr_bg_blkno
;
2015 *suballoc_bit_start
= res
.sr_bit_offset
;
2016 *blkno_start
= res
.sr_blkno
;
2017 ac
->ac_bits_given
+= res
.sr_bits
;
2018 *num_bits
= res
.sr_bits
;
2026 static void ocfs2_init_inode_ac_group(struct inode
*dir
,
2027 struct buffer_head
*parent_di_bh
,
2028 struct ocfs2_alloc_context
*ac
)
2030 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*)parent_di_bh
->b_data
;
2032 * Try to allocate inodes from some specific group.
2034 * If the parent dir has recorded the last group used in allocation,
2035 * cool, use it. Otherwise if we try to allocate new inode from the
2036 * same slot the parent dir belongs to, use the same chunk.
2038 * We are very careful here to avoid the mistake of setting
2039 * ac_last_group to a group descriptor from a different (unlocked) slot.
2041 if (OCFS2_I(dir
)->ip_last_used_group
&&
2042 OCFS2_I(dir
)->ip_last_used_slot
== ac
->ac_alloc_slot
)
2043 ac
->ac_last_group
= OCFS2_I(dir
)->ip_last_used_group
;
2044 else if (le16_to_cpu(di
->i_suballoc_slot
) == ac
->ac_alloc_slot
) {
2045 if (di
->i_suballoc_loc
)
2046 ac
->ac_last_group
= le64_to_cpu(di
->i_suballoc_loc
);
2048 ac
->ac_last_group
= ocfs2_which_suballoc_group(
2049 le64_to_cpu(di
->i_blkno
),
2050 le16_to_cpu(di
->i_suballoc_bit
));
2054 static inline void ocfs2_save_inode_ac_group(struct inode
*dir
,
2055 struct ocfs2_alloc_context
*ac
)
2057 OCFS2_I(dir
)->ip_last_used_group
= ac
->ac_last_group
;
2058 OCFS2_I(dir
)->ip_last_used_slot
= ac
->ac_alloc_slot
;
2061 int ocfs2_find_new_inode_loc(struct inode
*dir
,
2062 struct buffer_head
*parent_fe_bh
,
2063 struct ocfs2_alloc_context
*ac
,
2067 handle_t
*handle
= NULL
;
2068 struct ocfs2_suballoc_result
*res
;
2071 BUG_ON(ac
->ac_bits_given
!= 0);
2072 BUG_ON(ac
->ac_bits_wanted
!= 1);
2073 BUG_ON(ac
->ac_which
!= OCFS2_AC_USE_INODE
);
2075 res
= kzalloc(sizeof(*res
), GFP_NOFS
);
2082 ocfs2_init_inode_ac_group(dir
, parent_fe_bh
, ac
);
2085 * The handle started here is for chain relink. Alternatively,
2086 * we could just disable relink for these calls.
2088 handle
= ocfs2_start_trans(OCFS2_SB(dir
->i_sb
), OCFS2_SUBALLOC_ALLOC
);
2089 if (IS_ERR(handle
)) {
2090 ret
= PTR_ERR(handle
);
2097 * This will instruct ocfs2_claim_suballoc_bits and
2098 * ocfs2_search_one_group to search but save actual allocation
2101 ac
->ac_find_loc_only
= 1;
2103 ret
= ocfs2_claim_suballoc_bits(ac
, handle
, 1, 1, res
);
2109 ac
->ac_find_loc_priv
= res
;
2110 *fe_blkno
= res
->sr_blkno
;
2111 ocfs2_update_inode_fsync_trans(handle
, dir
, 0);
2114 ocfs2_commit_trans(OCFS2_SB(dir
->i_sb
), handle
);
2122 int ocfs2_claim_new_inode_at_loc(handle_t
*handle
,
2124 struct ocfs2_alloc_context
*ac
,
2131 struct ocfs2_suballoc_result
*res
= ac
->ac_find_loc_priv
;
2132 struct buffer_head
*bg_bh
= NULL
;
2133 struct ocfs2_group_desc
*bg
;
2134 struct ocfs2_dinode
*di
= (struct ocfs2_dinode
*) ac
->ac_bh
->b_data
;
2137 * Since di_blkno is being passed back in, we check for any
2138 * inconsistencies which may have happened between
2139 * calls. These are code bugs as di_blkno is not expected to
2140 * change once returned from ocfs2_find_new_inode_loc()
2142 BUG_ON(res
->sr_blkno
!= di_blkno
);
2144 ret
= ocfs2_read_group_descriptor(ac
->ac_inode
, di
,
2145 res
->sr_bg_stable_blkno
, &bg_bh
);
2151 bg
= (struct ocfs2_group_desc
*) bg_bh
->b_data
;
2152 chain
= le16_to_cpu(bg
->bg_chain
);
2154 ret
= ocfs2_alloc_dinode_update_counts(ac
->ac_inode
, handle
,
2155 ac
->ac_bh
, res
->sr_bits
,
2162 ret
= ocfs2_block_group_set_bits(handle
,
2169 ocfs2_rollback_alloc_dinode_counts(ac
->ac_inode
,
2170 ac
->ac_bh
, res
->sr_bits
, chain
);
2175 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno
,
2178 atomic_inc(&OCFS2_SB(ac
->ac_inode
->i_sb
)->alloc_stats
.bg_allocs
);
2180 BUG_ON(res
->sr_bits
!= 1);
2182 *suballoc_loc
= res
->sr_bg_blkno
;
2183 *suballoc_bit
= res
->sr_bit_offset
;
2184 ac
->ac_bits_given
++;
2185 ocfs2_save_inode_ac_group(dir
, ac
);
2193 int ocfs2_claim_new_inode(handle_t
*handle
,
2195 struct buffer_head
*parent_fe_bh
,
2196 struct ocfs2_alloc_context
*ac
,
2202 struct ocfs2_suballoc_result res
;
2205 BUG_ON(ac
->ac_bits_given
!= 0);
2206 BUG_ON(ac
->ac_bits_wanted
!= 1);
2207 BUG_ON(ac
->ac_which
!= OCFS2_AC_USE_INODE
);
2209 ocfs2_init_inode_ac_group(dir
, parent_fe_bh
, ac
);
2211 status
= ocfs2_claim_suballoc_bits(ac
,
2220 atomic_inc(&OCFS2_SB(ac
->ac_inode
->i_sb
)->alloc_stats
.bg_allocs
);
2222 BUG_ON(res
.sr_bits
!= 1);
2224 *suballoc_loc
= res
.sr_bg_blkno
;
2225 *suballoc_bit
= res
.sr_bit_offset
;
2226 *fe_blkno
= res
.sr_blkno
;
2227 ac
->ac_bits_given
++;
2228 ocfs2_save_inode_ac_group(dir
, ac
);
2236 /* translate a group desc. blkno and it's bitmap offset into
2237 * disk cluster offset. */
2238 static inline u32
ocfs2_desc_bitmap_to_cluster_off(struct inode
*inode
,
2242 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2245 BUG_ON(!ocfs2_is_cluster_bitmap(inode
));
2247 if (bg_blkno
!= osb
->first_cluster_group_blkno
)
2248 cluster
= ocfs2_blocks_to_clusters(inode
->i_sb
, bg_blkno
);
2249 cluster
+= (u32
) bg_bit_off
;
2253 /* given a cluster offset, calculate which block group it belongs to
2254 * and return that block offset. */
2255 u64
ocfs2_which_cluster_group(struct inode
*inode
, u32 cluster
)
2257 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2260 BUG_ON(!ocfs2_is_cluster_bitmap(inode
));
2262 group_no
= cluster
/ osb
->bitmap_cpg
;
2264 return osb
->first_cluster_group_blkno
;
2265 return ocfs2_clusters_to_blocks(inode
->i_sb
,
2266 group_no
* osb
->bitmap_cpg
);
2269 /* given the block number of a cluster start, calculate which cluster
2270 * group and descriptor bitmap offset that corresponds to. */
2271 static inline void ocfs2_block_to_cluster_group(struct inode
*inode
,
2276 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2277 u32 data_cluster
= ocfs2_blocks_to_clusters(osb
->sb
, data_blkno
);
2279 BUG_ON(!ocfs2_is_cluster_bitmap(inode
));
2281 *bg_blkno
= ocfs2_which_cluster_group(inode
,
2284 if (*bg_blkno
== osb
->first_cluster_group_blkno
)
2285 *bg_bit_off
= (u16
) data_cluster
;
2287 *bg_bit_off
= (u16
) ocfs2_blocks_to_clusters(osb
->sb
,
2288 data_blkno
- *bg_blkno
);
2292 * min_bits - minimum contiguous chunk from this total allocation we
2293 * can handle. set to what we asked for originally for a full
2294 * contig. allocation, set to '1' to indicate we can deal with extents
2297 int __ocfs2_claim_clusters(handle_t
*handle
,
2298 struct ocfs2_alloc_context
*ac
,
2305 unsigned int bits_wanted
= max_clusters
;
2306 struct ocfs2_suballoc_result res
= { .sr_blkno
= 0, };
2307 struct ocfs2_super
*osb
= OCFS2_SB(ac
->ac_inode
->i_sb
);
2309 BUG_ON(ac
->ac_bits_given
>= ac
->ac_bits_wanted
);
2311 BUG_ON(ac
->ac_which
!= OCFS2_AC_USE_LOCAL
2312 && ac
->ac_which
!= OCFS2_AC_USE_MAIN
);
2314 if (ac
->ac_which
== OCFS2_AC_USE_LOCAL
) {
2315 WARN_ON(min_clusters
> 1);
2317 status
= ocfs2_claim_local_alloc_bits(osb
,
2324 atomic_inc(&osb
->alloc_stats
.local_data
);
2326 if (min_clusters
> (osb
->bitmap_cpg
- 1)) {
2327 /* The only paths asking for contiguousness
2328 * should know about this already. */
2329 mlog(ML_ERROR
, "minimum allocation requested %u exceeds "
2330 "group bitmap size %u!\n", min_clusters
,
2335 /* clamp the current request down to a realistic size. */
2336 if (bits_wanted
> (osb
->bitmap_cpg
- 1))
2337 bits_wanted
= osb
->bitmap_cpg
- 1;
2339 status
= ocfs2_claim_suballoc_bits(ac
,
2345 BUG_ON(res
.sr_blkno
); /* cluster alloc can't set */
2347 ocfs2_desc_bitmap_to_cluster_off(ac
->ac_inode
,
2350 atomic_inc(&osb
->alloc_stats
.bitmap_data
);
2351 *num_clusters
= res
.sr_bits
;
2355 if (status
!= -ENOSPC
)
2360 ac
->ac_bits_given
+= *num_clusters
;
2368 int ocfs2_claim_clusters(handle_t
*handle
,
2369 struct ocfs2_alloc_context
*ac
,
2374 unsigned int bits_wanted
= ac
->ac_bits_wanted
- ac
->ac_bits_given
;
2376 return __ocfs2_claim_clusters(handle
, ac
, min_clusters
,
2377 bits_wanted
, cluster_start
, num_clusters
);
2380 static int ocfs2_block_group_clear_bits(handle_t
*handle
,
2381 struct inode
*alloc_inode
,
2382 struct ocfs2_group_desc
*bg
,
2383 struct buffer_head
*group_bh
,
2384 unsigned int bit_off
,
2385 unsigned int num_bits
,
2386 void (*undo_fn
)(unsigned int bit
,
2387 unsigned long *bmap
))
2391 struct ocfs2_group_desc
*undo_bg
= NULL
;
2392 struct journal_head
*jh
;
2394 /* The caller got this descriptor from
2395 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2396 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg
));
2398 trace_ocfs2_block_group_clear_bits(bit_off
, num_bits
);
2400 BUG_ON(undo_fn
&& !ocfs2_is_cluster_bitmap(alloc_inode
));
2401 status
= ocfs2_journal_access_gd(handle
, INODE_CACHE(alloc_inode
),
2404 OCFS2_JOURNAL_ACCESS_UNDO
:
2405 OCFS2_JOURNAL_ACCESS_WRITE
);
2411 jh
= bh2jh(group_bh
);
2413 spin_lock(&jh
->b_state_lock
);
2414 undo_bg
= (struct ocfs2_group_desc
*) jh
->b_committed_data
;
2420 ocfs2_clear_bit((bit_off
+ tmp
),
2421 (unsigned long *) bg
->bg_bitmap
);
2423 undo_fn(bit_off
+ tmp
,
2424 (unsigned long *) undo_bg
->bg_bitmap
);
2426 le16_add_cpu(&bg
->bg_free_bits_count
, num_bits
);
2427 if (le16_to_cpu(bg
->bg_free_bits_count
) > le16_to_cpu(bg
->bg_bits
)) {
2429 spin_unlock(&jh
->b_state_lock
);
2430 return ocfs2_error(alloc_inode
->i_sb
, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2431 (unsigned long long)le64_to_cpu(bg
->bg_blkno
),
2432 le16_to_cpu(bg
->bg_bits
),
2433 le16_to_cpu(bg
->bg_free_bits_count
),
2438 spin_unlock(&jh
->b_state_lock
);
2440 ocfs2_journal_dirty(handle
, group_bh
);
2446 * expects the suballoc inode to already be locked.
2448 static int _ocfs2_free_suballoc_bits(handle_t
*handle
,
2449 struct inode
*alloc_inode
,
2450 struct buffer_head
*alloc_bh
,
2451 unsigned int start_bit
,
2454 void (*undo_fn
)(unsigned int bit
,
2455 unsigned long *bitmap
))
2459 struct ocfs2_dinode
*fe
= (struct ocfs2_dinode
*) alloc_bh
->b_data
;
2460 struct ocfs2_chain_list
*cl
= &fe
->id2
.i_chain
;
2461 struct buffer_head
*group_bh
= NULL
;
2462 struct ocfs2_group_desc
*group
;
2464 /* The alloc_bh comes from ocfs2_free_dinode() or
2465 * ocfs2_free_clusters(). The callers have all locked the
2466 * allocator and gotten alloc_bh from the lock call. This
2467 * validates the dinode buffer. Any corruption that has happened
2469 BUG_ON(!OCFS2_IS_VALID_DINODE(fe
));
2470 BUG_ON((count
+ start_bit
) > ocfs2_bits_per_group(cl
));
2472 trace_ocfs2_free_suballoc_bits(
2473 (unsigned long long)OCFS2_I(alloc_inode
)->ip_blkno
,
2474 (unsigned long long)bg_blkno
,
2477 status
= ocfs2_read_group_descriptor(alloc_inode
, fe
, bg_blkno
,
2483 group
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
2485 BUG_ON((count
+ start_bit
) > le16_to_cpu(group
->bg_bits
));
2487 status
= ocfs2_block_group_clear_bits(handle
, alloc_inode
,
2489 start_bit
, count
, undo_fn
);
2495 status
= ocfs2_journal_access_di(handle
, INODE_CACHE(alloc_inode
),
2496 alloc_bh
, OCFS2_JOURNAL_ACCESS_WRITE
);
2499 ocfs2_block_group_set_bits(handle
, alloc_inode
, group
, group_bh
,
2504 le32_add_cpu(&cl
->cl_recs
[le16_to_cpu(group
->bg_chain
)].c_free
,
2506 tmp_used
= le32_to_cpu(fe
->id1
.bitmap1
.i_used
);
2507 fe
->id1
.bitmap1
.i_used
= cpu_to_le32(tmp_used
- count
);
2508 ocfs2_journal_dirty(handle
, alloc_bh
);
2515 int ocfs2_free_suballoc_bits(handle_t
*handle
,
2516 struct inode
*alloc_inode
,
2517 struct buffer_head
*alloc_bh
,
2518 unsigned int start_bit
,
2522 return _ocfs2_free_suballoc_bits(handle
, alloc_inode
, alloc_bh
,
2523 start_bit
, bg_blkno
, count
, NULL
);
2526 int ocfs2_free_dinode(handle_t
*handle
,
2527 struct inode
*inode_alloc_inode
,
2528 struct buffer_head
*inode_alloc_bh
,
2529 struct ocfs2_dinode
*di
)
2531 u64 blk
= le64_to_cpu(di
->i_blkno
);
2532 u16 bit
= le16_to_cpu(di
->i_suballoc_bit
);
2533 u64 bg_blkno
= ocfs2_which_suballoc_group(blk
, bit
);
2535 if (di
->i_suballoc_loc
)
2536 bg_blkno
= le64_to_cpu(di
->i_suballoc_loc
);
2537 return ocfs2_free_suballoc_bits(handle
, inode_alloc_inode
,
2538 inode_alloc_bh
, bit
, bg_blkno
, 1);
2541 static int _ocfs2_free_clusters(handle_t
*handle
,
2542 struct inode
*bitmap_inode
,
2543 struct buffer_head
*bitmap_bh
,
2545 unsigned int num_clusters
,
2546 void (*undo_fn
)(unsigned int bit
,
2547 unsigned long *bitmap
))
2553 /* You can't ever have a contiguous set of clusters
2554 * bigger than a block group bitmap so we never have to worry
2555 * about looping on them.
2556 * This is expensive. We can safely remove once this stuff has
2557 * gotten tested really well. */
2558 BUG_ON(start_blk
!= ocfs2_clusters_to_blocks(bitmap_inode
->i_sb
,
2559 ocfs2_blocks_to_clusters(bitmap_inode
->i_sb
,
2563 ocfs2_block_to_cluster_group(bitmap_inode
, start_blk
, &bg_blkno
,
2566 trace_ocfs2_free_clusters((unsigned long long)bg_blkno
,
2567 (unsigned long long)start_blk
,
2568 bg_start_bit
, num_clusters
);
2570 status
= _ocfs2_free_suballoc_bits(handle
, bitmap_inode
, bitmap_bh
,
2571 bg_start_bit
, bg_blkno
,
2572 num_clusters
, undo_fn
);
2578 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode
->i_sb
),
2585 int ocfs2_free_clusters(handle_t
*handle
,
2586 struct inode
*bitmap_inode
,
2587 struct buffer_head
*bitmap_bh
,
2589 unsigned int num_clusters
)
2591 return _ocfs2_free_clusters(handle
, bitmap_inode
, bitmap_bh
,
2592 start_blk
, num_clusters
,
2597 * Give never-used clusters back to the global bitmap. We don't need
2598 * to protect these bits in the undo buffer.
2600 int ocfs2_release_clusters(handle_t
*handle
,
2601 struct inode
*bitmap_inode
,
2602 struct buffer_head
*bitmap_bh
,
2604 unsigned int num_clusters
)
2606 return _ocfs2_free_clusters(handle
, bitmap_inode
, bitmap_bh
,
2607 start_blk
, num_clusters
,
2612 * For a given allocation, determine which allocators will need to be
2613 * accessed, and lock them, reserving the appropriate number of bits.
2615 * Sparse file systems call this from ocfs2_write_begin_nolock()
2616 * and ocfs2_allocate_unwritten_extents().
2618 * File systems which don't support holes call this from
2619 * ocfs2_extend_allocation().
2621 int ocfs2_lock_allocators(struct inode
*inode
,
2622 struct ocfs2_extent_tree
*et
,
2623 u32 clusters_to_add
, u32 extents_to_split
,
2624 struct ocfs2_alloc_context
**data_ac
,
2625 struct ocfs2_alloc_context
**meta_ac
)
2627 int ret
= 0, num_free_extents
;
2628 unsigned int max_recs_needed
= clusters_to_add
+ 2 * extents_to_split
;
2629 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
2635 BUG_ON(clusters_to_add
!= 0 && data_ac
== NULL
);
2637 num_free_extents
= ocfs2_num_free_extents(et
);
2638 if (num_free_extents
< 0) {
2639 ret
= num_free_extents
;
2645 * Sparse allocation file systems need to be more conservative
2646 * with reserving room for expansion - the actual allocation
2647 * happens while we've got a journal handle open so re-taking
2648 * a cluster lock (because we ran out of room for another
2649 * extent) will violate ordering rules.
2651 * Most of the time we'll only be seeing this 1 cluster at a time
2654 * Always lock for any unwritten extents - we might want to
2655 * add blocks during a split.
2657 if (!num_free_extents
||
2658 (ocfs2_sparse_alloc(osb
) && num_free_extents
< max_recs_needed
)) {
2659 ret
= ocfs2_reserve_new_metadata(osb
, et
->et_root_el
, meta_ac
);
2667 if (clusters_to_add
== 0)
2670 ret
= ocfs2_reserve_clusters(osb
, clusters_to_add
, data_ac
);
2680 ocfs2_free_alloc_context(*meta_ac
);
2685 * We cannot have an error and a non null *data_ac.
2693 * Read the inode specified by blkno to get suballoc_slot and
2696 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super
*osb
, u64 blkno
,
2697 u16
*suballoc_slot
, u64
*group_blkno
,
2701 struct buffer_head
*inode_bh
= NULL
;
2702 struct ocfs2_dinode
*inode_fe
;
2704 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno
);
2706 /* dirty read disk */
2707 status
= ocfs2_read_blocks_sync(osb
, blkno
, 1, &inode_bh
);
2709 mlog(ML_ERROR
, "read block %llu failed %d\n",
2710 (unsigned long long)blkno
, status
);
2714 inode_fe
= (struct ocfs2_dinode
*) inode_bh
->b_data
;
2715 if (!OCFS2_IS_VALID_DINODE(inode_fe
)) {
2716 mlog(ML_ERROR
, "invalid inode %llu requested\n",
2717 (unsigned long long)blkno
);
2722 if (le16_to_cpu(inode_fe
->i_suballoc_slot
) != (u16
)OCFS2_INVALID_SLOT
&&
2723 (u32
)le16_to_cpu(inode_fe
->i_suballoc_slot
) > osb
->max_slots
- 1) {
2724 mlog(ML_ERROR
, "inode %llu has invalid suballoc slot %u\n",
2725 (unsigned long long)blkno
,
2726 (u32
)le16_to_cpu(inode_fe
->i_suballoc_slot
));
2732 *suballoc_slot
= le16_to_cpu(inode_fe
->i_suballoc_slot
);
2734 *suballoc_bit
= le16_to_cpu(inode_fe
->i_suballoc_bit
);
2736 *group_blkno
= le64_to_cpu(inode_fe
->i_suballoc_loc
);
2747 * test whether bit is SET in allocator bitmap or not. on success, 0
2748 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2749 * is returned and *res is meaningless. Call this after you have
2750 * cluster locked against suballoc, or you may get a result based on
2751 * non-up2date contents
2753 static int ocfs2_test_suballoc_bit(struct ocfs2_super
*osb
,
2754 struct inode
*suballoc
,
2755 struct buffer_head
*alloc_bh
,
2756 u64 group_blkno
, u64 blkno
,
2759 struct ocfs2_dinode
*alloc_di
;
2760 struct ocfs2_group_desc
*group
;
2761 struct buffer_head
*group_bh
= NULL
;
2765 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno
,
2768 alloc_di
= (struct ocfs2_dinode
*)alloc_bh
->b_data
;
2769 if ((bit
+ 1) > ocfs2_bits_per_group(&alloc_di
->id2
.i_chain
)) {
2770 mlog(ML_ERROR
, "suballoc bit %u out of range of %u\n",
2772 ocfs2_bits_per_group(&alloc_di
->id2
.i_chain
));
2777 bg_blkno
= group_blkno
? group_blkno
:
2778 ocfs2_which_suballoc_group(blkno
, bit
);
2779 status
= ocfs2_read_group_descriptor(suballoc
, alloc_di
, bg_blkno
,
2782 mlog(ML_ERROR
, "read group %llu failed %d\n",
2783 (unsigned long long)bg_blkno
, status
);
2787 group
= (struct ocfs2_group_desc
*) group_bh
->b_data
;
2788 *res
= ocfs2_test_bit(bit
, (unsigned long *)group
->bg_bitmap
);
2799 * Test if the bit representing this inode (blkno) is set in the
2802 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2804 * In the event of failure, a negative value is returned and *res is
2807 * Callers must make sure to hold nfs_sync_lock to prevent
2808 * ocfs2_delete_inode() on another node from accessing the same
2809 * suballocator concurrently.
2811 int ocfs2_test_inode_bit(struct ocfs2_super
*osb
, u64 blkno
, int *res
)
2814 u64 group_blkno
= 0;
2815 u16 suballoc_bit
= 0, suballoc_slot
= 0;
2816 struct inode
*inode_alloc_inode
;
2817 struct buffer_head
*alloc_bh
= NULL
;
2819 trace_ocfs2_test_inode_bit((unsigned long long)blkno
);
2821 status
= ocfs2_get_suballoc_slot_bit(osb
, blkno
, &suballoc_slot
,
2822 &group_blkno
, &suballoc_bit
);
2824 mlog(ML_ERROR
, "get alloc slot and bit failed %d\n", status
);
2829 ocfs2_get_system_file_inode(osb
, INODE_ALLOC_SYSTEM_INODE
,
2831 if (!inode_alloc_inode
) {
2832 /* the error code could be inaccurate, but we are not able to
2833 * get the correct one. */
2835 mlog(ML_ERROR
, "unable to get alloc inode in slot %u\n",
2836 (u32
)suballoc_slot
);
2840 inode_lock(inode_alloc_inode
);
2841 status
= ocfs2_inode_lock(inode_alloc_inode
, &alloc_bh
, 0);
2843 inode_unlock(inode_alloc_inode
);
2844 iput(inode_alloc_inode
);
2845 mlog(ML_ERROR
, "lock on alloc inode on slot %u failed %d\n",
2846 (u32
)suballoc_slot
, status
);
2850 status
= ocfs2_test_suballoc_bit(osb
, inode_alloc_inode
, alloc_bh
,
2851 group_blkno
, blkno
, suballoc_bit
, res
);
2853 mlog(ML_ERROR
, "test suballoc bit failed %d\n", status
);
2855 ocfs2_inode_unlock(inode_alloc_inode
, 0);
2856 inode_unlock(inode_alloc_inode
);
2858 iput(inode_alloc_inode
);