1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Block/Cluster mapping functions
8 * Copyright (C) 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License, version 2, as published by the Free Software Foundation.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
19 * You should have received a copy of the GNU General Public
20 * License along with this program; if not, write to the
21 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
22 * Boston, MA 021110-1307, USA.
26 #include <linux/init.h>
27 #include <linux/slab.h>
28 #include <linux/types.h>
29 #include <linux/fiemap.h>
31 #include <cluster/masklog.h>
37 #include "extent_map.h"
42 #include "ocfs2_trace.h"
44 #include "buffer_head_io.h"
47 * The extent caching implementation is intentionally trivial.
49 * We only cache a small number of extents stored directly on the
50 * inode, so linear order operations are acceptable. If we ever want
51 * to increase the size of the extent map, then these algorithms must
55 void ocfs2_extent_map_init(struct inode
*inode
)
57 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
59 oi
->ip_extent_map
.em_num_items
= 0;
60 INIT_LIST_HEAD(&oi
->ip_extent_map
.em_list
);
63 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map
*em
,
65 struct ocfs2_extent_map_item
**ret_emi
)
68 struct ocfs2_extent_map_item
*emi
;
72 list_for_each_entry(emi
, &em
->em_list
, ei_list
) {
73 range
= emi
->ei_cpos
+ emi
->ei_clusters
;
75 if (cpos
>= emi
->ei_cpos
&& cpos
< range
) {
76 list_move(&emi
->ei_list
, &em
->em_list
);
84 static int ocfs2_extent_map_lookup(struct inode
*inode
, unsigned int cpos
,
85 unsigned int *phys
, unsigned int *len
,
89 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
90 struct ocfs2_extent_map_item
*emi
;
92 spin_lock(&oi
->ip_lock
);
94 __ocfs2_extent_map_lookup(&oi
->ip_extent_map
, cpos
, &emi
);
96 coff
= cpos
- emi
->ei_cpos
;
97 *phys
= emi
->ei_phys
+ coff
;
99 *len
= emi
->ei_clusters
- coff
;
101 *flags
= emi
->ei_flags
;
104 spin_unlock(&oi
->ip_lock
);
113 * Forget about all clusters equal to or greater than cpos.
115 void ocfs2_extent_map_trunc(struct inode
*inode
, unsigned int cpos
)
117 struct ocfs2_extent_map_item
*emi
, *n
;
118 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
119 struct ocfs2_extent_map
*em
= &oi
->ip_extent_map
;
123 spin_lock(&oi
->ip_lock
);
124 list_for_each_entry_safe(emi
, n
, &em
->em_list
, ei_list
) {
125 if (emi
->ei_cpos
>= cpos
) {
126 /* Full truncate of this record. */
127 list_move(&emi
->ei_list
, &tmp_list
);
128 BUG_ON(em
->em_num_items
== 0);
133 range
= emi
->ei_cpos
+ emi
->ei_clusters
;
135 /* Partial truncate */
136 emi
->ei_clusters
= cpos
- emi
->ei_cpos
;
139 spin_unlock(&oi
->ip_lock
);
141 list_for_each_entry_safe(emi
, n
, &tmp_list
, ei_list
) {
142 list_del(&emi
->ei_list
);
148 * Is any part of emi2 contained within emi1
150 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item
*emi1
,
151 struct ocfs2_extent_map_item
*emi2
)
153 unsigned int range1
, range2
;
156 * Check if logical start of emi2 is inside emi1
158 range1
= emi1
->ei_cpos
+ emi1
->ei_clusters
;
159 if (emi2
->ei_cpos
>= emi1
->ei_cpos
&& emi2
->ei_cpos
< range1
)
163 * Check if logical end of emi2 is inside emi1
165 range2
= emi2
->ei_cpos
+ emi2
->ei_clusters
;
166 if (range2
> emi1
->ei_cpos
&& range2
<= range1
)
172 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item
*dest
,
173 struct ocfs2_extent_map_item
*src
)
175 dest
->ei_cpos
= src
->ei_cpos
;
176 dest
->ei_phys
= src
->ei_phys
;
177 dest
->ei_clusters
= src
->ei_clusters
;
178 dest
->ei_flags
= src
->ei_flags
;
182 * Try to merge emi with ins. Returns 1 if merge succeeds, zero
185 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item
*emi
,
186 struct ocfs2_extent_map_item
*ins
)
189 * Handle contiguousness
191 if (ins
->ei_phys
== (emi
->ei_phys
+ emi
->ei_clusters
) &&
192 ins
->ei_cpos
== (emi
->ei_cpos
+ emi
->ei_clusters
) &&
193 ins
->ei_flags
== emi
->ei_flags
) {
194 emi
->ei_clusters
+= ins
->ei_clusters
;
196 } else if ((ins
->ei_phys
+ ins
->ei_clusters
) == emi
->ei_phys
&&
197 (ins
->ei_cpos
+ ins
->ei_clusters
) == emi
->ei_cpos
&&
198 ins
->ei_flags
== emi
->ei_flags
) {
199 emi
->ei_phys
= ins
->ei_phys
;
200 emi
->ei_cpos
= ins
->ei_cpos
;
201 emi
->ei_clusters
+= ins
->ei_clusters
;
206 * Overlapping extents - this shouldn't happen unless we've
207 * split an extent to change it's flags. That is exceedingly
208 * rare, so there's no sense in trying to optimize it yet.
210 if (ocfs2_ei_is_contained(emi
, ins
) ||
211 ocfs2_ei_is_contained(ins
, emi
)) {
212 ocfs2_copy_emi_fields(emi
, ins
);
216 /* No merge was possible. */
221 * In order to reduce complexity on the caller, this insert function
222 * is intentionally liberal in what it will accept.
224 * The only rule is that the truncate call *must* be used whenever
225 * records have been deleted. This avoids inserting overlapping
226 * records with different physical mappings.
228 void ocfs2_extent_map_insert_rec(struct inode
*inode
,
229 struct ocfs2_extent_rec
*rec
)
231 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
232 struct ocfs2_extent_map
*em
= &oi
->ip_extent_map
;
233 struct ocfs2_extent_map_item
*emi
, *new_emi
= NULL
;
234 struct ocfs2_extent_map_item ins
;
236 ins
.ei_cpos
= le32_to_cpu(rec
->e_cpos
);
237 ins
.ei_phys
= ocfs2_blocks_to_clusters(inode
->i_sb
,
238 le64_to_cpu(rec
->e_blkno
));
239 ins
.ei_clusters
= le16_to_cpu(rec
->e_leaf_clusters
);
240 ins
.ei_flags
= rec
->e_flags
;
243 spin_lock(&oi
->ip_lock
);
245 list_for_each_entry(emi
, &em
->em_list
, ei_list
) {
246 if (ocfs2_try_to_merge_extent_map(emi
, &ins
)) {
247 list_move(&emi
->ei_list
, &em
->em_list
);
248 spin_unlock(&oi
->ip_lock
);
254 * No item could be merged.
256 * Either allocate and add a new item, or overwrite the last recently
260 if (em
->em_num_items
< OCFS2_MAX_EXTENT_MAP_ITEMS
) {
261 if (new_emi
== NULL
) {
262 spin_unlock(&oi
->ip_lock
);
264 new_emi
= kmalloc(sizeof(*new_emi
), GFP_NOFS
);
271 ocfs2_copy_emi_fields(new_emi
, &ins
);
272 list_add(&new_emi
->ei_list
, &em
->em_list
);
276 BUG_ON(list_empty(&em
->em_list
) || em
->em_num_items
== 0);
277 emi
= list_entry(em
->em_list
.prev
,
278 struct ocfs2_extent_map_item
, ei_list
);
279 list_move(&emi
->ei_list
, &em
->em_list
);
280 ocfs2_copy_emi_fields(emi
, &ins
);
283 spin_unlock(&oi
->ip_lock
);
289 static int ocfs2_last_eb_is_empty(struct inode
*inode
,
290 struct ocfs2_dinode
*di
)
293 u64 last_eb_blk
= le64_to_cpu(di
->i_last_eb_blk
);
294 struct buffer_head
*eb_bh
= NULL
;
295 struct ocfs2_extent_block
*eb
;
296 struct ocfs2_extent_list
*el
;
298 ret
= ocfs2_read_extent_block(INODE_CACHE(inode
), last_eb_blk
, &eb_bh
);
304 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
307 if (el
->l_tree_depth
) {
308 ocfs2_error(inode
->i_sb
,
309 "Inode %lu has non zero tree depth in leaf block %llu\n",
311 (unsigned long long)eb_bh
->b_blocknr
);
316 next_free
= le16_to_cpu(el
->l_next_free_rec
);
318 if (next_free
== 0 ||
319 (next_free
== 1 && ocfs2_is_empty_extent(&el
->l_recs
[0])))
328 * Return the 1st index within el which contains an extent start
329 * larger than v_cluster.
331 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list
*el
,
335 struct ocfs2_extent_rec
*rec
;
337 for(i
= 0; i
< le16_to_cpu(el
->l_next_free_rec
); i
++) {
338 rec
= &el
->l_recs
[i
];
340 if (v_cluster
< le32_to_cpu(rec
->e_cpos
))
348 * Figure out the size of a hole which starts at v_cluster within the given
351 * If there is no more allocation past v_cluster, we return the maximum
352 * cluster size minus v_cluster.
354 * If we have in-inode extents, then el points to the dinode list and
355 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
358 int ocfs2_figure_hole_clusters(struct ocfs2_caching_info
*ci
,
359 struct ocfs2_extent_list
*el
,
360 struct buffer_head
*eb_bh
,
365 struct buffer_head
*next_eb_bh
= NULL
;
366 struct ocfs2_extent_block
*eb
, *next_eb
;
368 i
= ocfs2_search_for_hole_index(el
, v_cluster
);
370 if (i
== le16_to_cpu(el
->l_next_free_rec
) && eb_bh
) {
371 eb
= (struct ocfs2_extent_block
*)eb_bh
->b_data
;
374 * Check the next leaf for any extents.
377 if (le64_to_cpu(eb
->h_next_leaf_blk
) == 0ULL)
378 goto no_more_extents
;
380 ret
= ocfs2_read_extent_block(ci
,
381 le64_to_cpu(eb
->h_next_leaf_blk
),
388 next_eb
= (struct ocfs2_extent_block
*)next_eb_bh
->b_data
;
389 el
= &next_eb
->h_list
;
390 i
= ocfs2_search_for_hole_index(el
, v_cluster
);
394 if (i
== le16_to_cpu(el
->l_next_free_rec
)) {
396 * We're at the end of our existing allocation. Just
397 * return the maximum number of clusters we could
400 *num_clusters
= UINT_MAX
- v_cluster
;
402 *num_clusters
= le32_to_cpu(el
->l_recs
[i
].e_cpos
) - v_cluster
;
411 static int ocfs2_get_clusters_nocache(struct inode
*inode
,
412 struct buffer_head
*di_bh
,
413 u32 v_cluster
, unsigned int *hole_len
,
414 struct ocfs2_extent_rec
*ret_rec
,
415 unsigned int *is_last
)
417 int i
, ret
, tree_height
, len
;
418 struct ocfs2_dinode
*di
;
419 struct ocfs2_extent_block
*uninitialized_var(eb
);
420 struct ocfs2_extent_list
*el
;
421 struct ocfs2_extent_rec
*rec
;
422 struct buffer_head
*eb_bh
= NULL
;
424 memset(ret_rec
, 0, sizeof(*ret_rec
));
428 di
= (struct ocfs2_dinode
*) di_bh
->b_data
;
429 el
= &di
->id2
.i_list
;
430 tree_height
= le16_to_cpu(el
->l_tree_depth
);
432 if (tree_height
> 0) {
433 ret
= ocfs2_find_leaf(INODE_CACHE(inode
), el
, v_cluster
,
440 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
443 if (el
->l_tree_depth
) {
444 ocfs2_error(inode
->i_sb
,
445 "Inode %lu has non zero tree depth in leaf block %llu\n",
447 (unsigned long long)eb_bh
->b_blocknr
);
453 i
= ocfs2_search_extent_list(el
, v_cluster
);
456 * Holes can be larger than the maximum size of an
457 * extent, so we return their lengths in a separate
461 ret
= ocfs2_figure_hole_clusters(INODE_CACHE(inode
),
474 rec
= &el
->l_recs
[i
];
476 BUG_ON(v_cluster
< le32_to_cpu(rec
->e_cpos
));
479 ocfs2_error(inode
->i_sb
,
480 "Inode %lu has bad extent record (%u, %u, 0)\n",
482 le32_to_cpu(rec
->e_cpos
),
483 ocfs2_rec_clusters(el
, rec
));
491 * Checking for last extent is potentially expensive - we
492 * might have to look at the next leaf over to see if it's
495 * The first two checks are to see whether the caller even
496 * cares for this information, and if the extent is at least
497 * the last in it's list.
499 * If those hold true, then the extent is last if any of the
500 * additional conditions hold true:
501 * - Extent list is in-inode
502 * - Extent list is right-most
503 * - Extent list is 2nd to rightmost, with empty right-most
506 if (i
== (le16_to_cpu(el
->l_next_free_rec
) - 1)) {
507 if (tree_height
== 0)
509 else if (eb
->h_blkno
== di
->i_last_eb_blk
)
511 else if (eb
->h_next_leaf_blk
== di
->i_last_eb_blk
) {
512 ret
= ocfs2_last_eb_is_empty(inode
, di
);
530 static void ocfs2_relative_extent_offsets(struct super_block
*sb
,
532 struct ocfs2_extent_rec
*rec
,
533 u32
*p_cluster
, u32
*num_clusters
)
536 u32 coff
= v_cluster
- le32_to_cpu(rec
->e_cpos
);
538 *p_cluster
= ocfs2_blocks_to_clusters(sb
, le64_to_cpu(rec
->e_blkno
));
539 *p_cluster
= *p_cluster
+ coff
;
542 *num_clusters
= le16_to_cpu(rec
->e_leaf_clusters
) - coff
;
545 int ocfs2_xattr_get_clusters(struct inode
*inode
, u32 v_cluster
,
546 u32
*p_cluster
, u32
*num_clusters
,
547 struct ocfs2_extent_list
*el
,
548 unsigned int *extent_flags
)
551 struct buffer_head
*eb_bh
= NULL
;
552 struct ocfs2_extent_block
*eb
;
553 struct ocfs2_extent_rec
*rec
;
556 if (el
->l_tree_depth
) {
557 ret
= ocfs2_find_leaf(INODE_CACHE(inode
), el
, v_cluster
,
564 eb
= (struct ocfs2_extent_block
*) eb_bh
->b_data
;
567 if (el
->l_tree_depth
) {
568 ocfs2_error(inode
->i_sb
,
569 "Inode %lu has non zero tree depth in xattr leaf block %llu\n",
571 (unsigned long long)eb_bh
->b_blocknr
);
577 i
= ocfs2_search_extent_list(el
, v_cluster
);
583 rec
= &el
->l_recs
[i
];
584 BUG_ON(v_cluster
< le32_to_cpu(rec
->e_cpos
));
587 ocfs2_error(inode
->i_sb
,
588 "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
590 le32_to_cpu(rec
->e_cpos
),
591 ocfs2_rec_clusters(el
, rec
));
595 coff
= v_cluster
- le32_to_cpu(rec
->e_cpos
);
596 *p_cluster
= ocfs2_blocks_to_clusters(inode
->i_sb
,
597 le64_to_cpu(rec
->e_blkno
));
598 *p_cluster
= *p_cluster
+ coff
;
600 *num_clusters
= ocfs2_rec_clusters(el
, rec
) - coff
;
603 *extent_flags
= rec
->e_flags
;
611 int ocfs2_get_clusters(struct inode
*inode
, u32 v_cluster
,
612 u32
*p_cluster
, u32
*num_clusters
,
613 unsigned int *extent_flags
)
616 unsigned int uninitialized_var(hole_len
), flags
= 0;
617 struct buffer_head
*di_bh
= NULL
;
618 struct ocfs2_extent_rec rec
;
620 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
626 ret
= ocfs2_extent_map_lookup(inode
, v_cluster
, p_cluster
,
627 num_clusters
, extent_flags
);
631 ret
= ocfs2_read_inode_block(inode
, &di_bh
);
637 ret
= ocfs2_get_clusters_nocache(inode
, di_bh
, v_cluster
, &hole_len
,
644 if (rec
.e_blkno
== 0ULL) {
646 * A hole was found. Return some canned values that
647 * callers can key on. If asked for, num_clusters will
648 * be populated with the size of the hole.
652 *num_clusters
= hole_len
;
655 ocfs2_relative_extent_offsets(inode
->i_sb
, v_cluster
, &rec
,
656 p_cluster
, num_clusters
);
659 ocfs2_extent_map_insert_rec(inode
, &rec
);
663 *extent_flags
= flags
;
671 * This expects alloc_sem to be held. The allocation cannot change at
672 * all while the map is in the process of being updated.
674 int ocfs2_extent_map_get_blocks(struct inode
*inode
, u64 v_blkno
, u64
*p_blkno
,
675 u64
*ret_count
, unsigned int *extent_flags
)
678 int bpc
= ocfs2_clusters_to_blocks(inode
->i_sb
, 1);
679 u32 cpos
, num_clusters
, p_cluster
;
682 cpos
= ocfs2_blocks_to_clusters(inode
->i_sb
, v_blkno
);
684 ret
= ocfs2_get_clusters(inode
, cpos
, &p_cluster
, &num_clusters
,
692 * p_cluster == 0 indicates a hole.
695 boff
= ocfs2_clusters_to_blocks(inode
->i_sb
, p_cluster
);
696 boff
+= (v_blkno
& (u64
)(bpc
- 1));
702 *ret_count
= ocfs2_clusters_to_blocks(inode
->i_sb
, num_clusters
);
703 *ret_count
-= v_blkno
& (u64
)(bpc
- 1);
711 * The ocfs2_fiemap_inline() may be a little bit misleading, since
712 * it not only handles the fiemap for inlined files, but also deals
713 * with the fast symlink, cause they have no difference for extent
716 static int ocfs2_fiemap_inline(struct inode
*inode
, struct buffer_head
*di_bh
,
717 struct fiemap_extent_info
*fieinfo
,
721 unsigned int id_count
;
722 struct ocfs2_dinode
*di
;
724 u32 flags
= FIEMAP_EXTENT_DATA_INLINE
|FIEMAP_EXTENT_LAST
;
725 struct ocfs2_inode_info
*oi
= OCFS2_I(inode
);
727 di
= (struct ocfs2_dinode
*)di_bh
->b_data
;
728 if (ocfs2_inode_is_fast_symlink(inode
))
729 id_count
= ocfs2_fast_symlink_chars(inode
->i_sb
);
731 id_count
= le16_to_cpu(di
->id2
.i_data
.id_count
);
733 if (map_start
< id_count
) {
734 phys
= oi
->ip_blkno
<< inode
->i_sb
->s_blocksize_bits
;
735 if (ocfs2_inode_is_fast_symlink(inode
))
736 phys
+= offsetof(struct ocfs2_dinode
, id2
.i_symlink
);
738 phys
+= offsetof(struct ocfs2_dinode
,
741 ret
= fiemap_fill_next_extent(fieinfo
, 0, phys
, id_count
,
750 #define OCFS2_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
752 int ocfs2_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
753 u64 map_start
, u64 map_len
)
756 u32 mapping_end
, cpos
;
757 unsigned int hole_size
;
758 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
759 u64 len_bytes
, phys_bytes
, virt_bytes
;
760 struct buffer_head
*di_bh
= NULL
;
761 struct ocfs2_extent_rec rec
;
763 ret
= fiemap_check_flags(fieinfo
, OCFS2_FIEMAP_FLAGS
);
767 ret
= ocfs2_inode_lock(inode
, &di_bh
, 0);
773 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
776 * Handle inline-data and fast symlink separately.
778 if ((OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) ||
779 ocfs2_inode_is_fast_symlink(inode
)) {
780 ret
= ocfs2_fiemap_inline(inode
, di_bh
, fieinfo
, map_start
);
784 cpos
= map_start
>> osb
->s_clustersize_bits
;
785 mapping_end
= ocfs2_clusters_for_bytes(inode
->i_sb
,
786 map_start
+ map_len
);
788 while (cpos
< mapping_end
&& !is_last
) {
791 ret
= ocfs2_get_clusters_nocache(inode
, di_bh
, cpos
,
792 &hole_size
, &rec
, &is_last
);
798 if (rec
.e_blkno
== 0ULL) {
804 if (rec
.e_flags
& OCFS2_EXT_UNWRITTEN
)
805 fe_flags
|= FIEMAP_EXTENT_UNWRITTEN
;
806 if (rec
.e_flags
& OCFS2_EXT_REFCOUNTED
)
807 fe_flags
|= FIEMAP_EXTENT_SHARED
;
809 fe_flags
|= FIEMAP_EXTENT_LAST
;
810 len_bytes
= (u64
)le16_to_cpu(rec
.e_leaf_clusters
) << osb
->s_clustersize_bits
;
811 phys_bytes
= le64_to_cpu(rec
.e_blkno
) << osb
->sb
->s_blocksize_bits
;
812 virt_bytes
= (u64
)le32_to_cpu(rec
.e_cpos
) << osb
->s_clustersize_bits
;
814 ret
= fiemap_fill_next_extent(fieinfo
, virt_bytes
, phys_bytes
,
815 len_bytes
, fe_flags
);
819 cpos
= le32_to_cpu(rec
.e_cpos
)+ le16_to_cpu(rec
.e_leaf_clusters
);
828 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
830 ocfs2_inode_unlock(inode
, 0);
836 /* Is IO overwriting allocated blocks? */
837 int ocfs2_overwrite_io(struct inode
*inode
, struct buffer_head
*di_bh
,
838 u64 map_start
, u64 map_len
)
840 int ret
= 0, is_last
;
841 u32 mapping_end
, cpos
;
842 struct ocfs2_super
*osb
= OCFS2_SB(inode
->i_sb
);
843 struct ocfs2_extent_rec rec
;
845 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
846 if (ocfs2_size_fits_inline_data(di_bh
, map_start
+ map_len
))
852 cpos
= map_start
>> osb
->s_clustersize_bits
;
853 mapping_end
= ocfs2_clusters_for_bytes(inode
->i_sb
,
854 map_start
+ map_len
);
856 while (cpos
< mapping_end
&& !is_last
) {
857 ret
= ocfs2_get_clusters_nocache(inode
, di_bh
, cpos
,
858 NULL
, &rec
, &is_last
);
864 if (rec
.e_blkno
== 0ULL)
867 if (rec
.e_flags
& OCFS2_EXT_REFCOUNTED
)
870 cpos
= le32_to_cpu(rec
.e_cpos
) +
871 le16_to_cpu(rec
.e_leaf_clusters
);
874 if (cpos
< mapping_end
)
880 int ocfs2_seek_data_hole_offset(struct file
*file
, loff_t
*offset
, int whence
)
882 struct inode
*inode
= file
->f_mapping
->host
;
884 unsigned int is_last
= 0, is_data
= 0;
885 u16 cs_bits
= OCFS2_SB(inode
->i_sb
)->s_clustersize_bits
;
886 u32 cpos
, cend
, clen
, hole_size
;
888 struct buffer_head
*di_bh
= NULL
;
889 struct ocfs2_extent_rec rec
;
891 BUG_ON(whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
);
893 ret
= ocfs2_inode_lock(inode
, &di_bh
, 0);
899 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
901 if (*offset
>= i_size_read(inode
)) {
906 if (OCFS2_I(inode
)->ip_dyn_features
& OCFS2_INLINE_DATA_FL
) {
907 if (whence
== SEEK_HOLE
)
908 *offset
= i_size_read(inode
);
913 cpos
= *offset
>> cs_bits
;
914 cend
= ocfs2_clusters_for_bytes(inode
->i_sb
, i_size_read(inode
));
916 while (cpos
< cend
&& !is_last
) {
917 ret
= ocfs2_get_clusters_nocache(inode
, di_bh
, cpos
, &hole_size
,
927 if (rec
.e_blkno
== 0ULL) {
931 clen
= le16_to_cpu(rec
.e_leaf_clusters
) -
932 (cpos
- le32_to_cpu(rec
.e_cpos
));
933 is_data
= (rec
.e_flags
& OCFS2_EXT_UNWRITTEN
) ? 0 : 1;
936 if ((!is_data
&& whence
== SEEK_HOLE
) ||
937 (is_data
&& whence
== SEEK_DATA
)) {
938 if (extoff
> *offset
)
947 if (whence
== SEEK_HOLE
) {
953 if ((extoff
+ extlen
) > i_size_read(inode
))
954 extlen
= i_size_read(inode
) - extoff
;
956 if (extoff
> *offset
)
967 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
969 ocfs2_inode_unlock(inode
, 0);
974 int ocfs2_read_virt_blocks(struct inode
*inode
, u64 v_block
, int nr
,
975 struct buffer_head
*bhs
[], int flags
,
976 int (*validate
)(struct super_block
*sb
,
977 struct buffer_head
*bh
))
980 u64 p_block
, p_count
;
981 int i
, count
, done
= 0;
983 trace_ocfs2_read_virt_blocks(
984 inode
, (unsigned long long)v_block
, nr
, bhs
, flags
,
987 if (((v_block
+ nr
- 1) << inode
->i_sb
->s_blocksize_bits
) >=
988 i_size_read(inode
)) {
989 BUG_ON(!(flags
& OCFS2_BH_READAHEAD
));
994 down_read(&OCFS2_I(inode
)->ip_alloc_sem
);
995 rc
= ocfs2_extent_map_get_blocks(inode
, v_block
+ done
,
996 &p_block
, &p_count
, NULL
);
997 up_read(&OCFS2_I(inode
)->ip_alloc_sem
);
1006 "Inode #%llu contains a hole at offset %llu\n",
1007 (unsigned long long)OCFS2_I(inode
)->ip_blkno
,
1008 (unsigned long long)(v_block
+ done
) <<
1009 inode
->i_sb
->s_blocksize_bits
);
1014 if (p_count
< count
)
1018 * If the caller passed us bhs, they should have come
1019 * from a previous readahead call to this function. Thus,
1020 * they should have the right b_blocknr.
1022 for (i
= 0; i
< count
; i
++) {
1025 BUG_ON(bhs
[done
+ i
]->b_blocknr
!= (p_block
+ i
));
1028 rc
= ocfs2_read_blocks(INODE_CACHE(inode
), p_block
, count
,
1029 bhs
+ done
, flags
, validate
);