Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc
[linux-2.6] / fs / ocfs2 / extent_map.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * extent_map.c
5  *
6  * Block/Cluster mapping functions
7  *
8  * Copyright (C) 2004 Oracle.  All rights reserved.
9  *
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.
13  *
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.
18  *
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.
23  */
24
25 #include <linux/fs.h>
26 #include <linux/init.h>
27 #include <linux/types.h>
28 #include <linux/fiemap.h>
29
30 #define MLOG_MASK_PREFIX ML_EXTENT_MAP
31 #include <cluster/masklog.h>
32
33 #include "ocfs2.h"
34
35 #include "alloc.h"
36 #include "dlmglue.h"
37 #include "extent_map.h"
38 #include "inode.h"
39 #include "super.h"
40
41 #include "buffer_head_io.h"
42
43 /*
44  * The extent caching implementation is intentionally trivial.
45  *
46  * We only cache a small number of extents stored directly on the
47  * inode, so linear order operations are acceptable. If we ever want
48  * to increase the size of the extent map, then these algorithms must
49  * get smarter.
50  */
51
52 void ocfs2_extent_map_init(struct inode *inode)
53 {
54         struct ocfs2_inode_info *oi = OCFS2_I(inode);
55
56         oi->ip_extent_map.em_num_items = 0;
57         INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
58 }
59
60 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
61                                       unsigned int cpos,
62                                       struct ocfs2_extent_map_item **ret_emi)
63 {
64         unsigned int range;
65         struct ocfs2_extent_map_item *emi;
66
67         *ret_emi = NULL;
68
69         list_for_each_entry(emi, &em->em_list, ei_list) {
70                 range = emi->ei_cpos + emi->ei_clusters;
71
72                 if (cpos >= emi->ei_cpos && cpos < range) {
73                         list_move(&emi->ei_list, &em->em_list);
74
75                         *ret_emi = emi;
76                         break;
77                 }
78         }
79 }
80
81 static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
82                                    unsigned int *phys, unsigned int *len,
83                                    unsigned int *flags)
84 {
85         unsigned int coff;
86         struct ocfs2_inode_info *oi = OCFS2_I(inode);
87         struct ocfs2_extent_map_item *emi;
88
89         spin_lock(&oi->ip_lock);
90
91         __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
92         if (emi) {
93                 coff = cpos - emi->ei_cpos;
94                 *phys = emi->ei_phys + coff;
95                 if (len)
96                         *len = emi->ei_clusters - coff;
97                 if (flags)
98                         *flags = emi->ei_flags;
99         }
100
101         spin_unlock(&oi->ip_lock);
102
103         if (emi == NULL)
104                 return -ENOENT;
105
106         return 0;
107 }
108
109 /*
110  * Forget about all clusters equal to or greater than cpos.
111  */
112 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
113 {
114         struct ocfs2_extent_map_item *emi, *n;
115         struct ocfs2_inode_info *oi = OCFS2_I(inode);
116         struct ocfs2_extent_map *em = &oi->ip_extent_map;
117         LIST_HEAD(tmp_list);
118         unsigned int range;
119
120         spin_lock(&oi->ip_lock);
121         list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
122                 if (emi->ei_cpos >= cpos) {
123                         /* Full truncate of this record. */
124                         list_move(&emi->ei_list, &tmp_list);
125                         BUG_ON(em->em_num_items == 0);
126                         em->em_num_items--;
127                         continue;
128                 }
129
130                 range = emi->ei_cpos + emi->ei_clusters;
131                 if (range > cpos) {
132                         /* Partial truncate */
133                         emi->ei_clusters = cpos - emi->ei_cpos;
134                 }
135         }
136         spin_unlock(&oi->ip_lock);
137
138         list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
139                 list_del(&emi->ei_list);
140                 kfree(emi);
141         }
142 }
143
144 /*
145  * Is any part of emi2 contained within emi1
146  */
147 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
148                                  struct ocfs2_extent_map_item *emi2)
149 {
150         unsigned int range1, range2;
151
152         /*
153          * Check if logical start of emi2 is inside emi1
154          */
155         range1 = emi1->ei_cpos + emi1->ei_clusters;
156         if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
157                 return 1;
158
159         /*
160          * Check if logical end of emi2 is inside emi1
161          */
162         range2 = emi2->ei_cpos + emi2->ei_clusters;
163         if (range2 > emi1->ei_cpos && range2 <= range1)
164                 return 1;
165
166         return 0;
167 }
168
169 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
170                                   struct ocfs2_extent_map_item *src)
171 {
172         dest->ei_cpos = src->ei_cpos;
173         dest->ei_phys = src->ei_phys;
174         dest->ei_clusters = src->ei_clusters;
175         dest->ei_flags = src->ei_flags;
176 }
177
178 /*
179  * Try to merge emi with ins. Returns 1 if merge succeeds, zero
180  * otherwise.
181  */
182 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
183                                          struct ocfs2_extent_map_item *ins)
184 {
185         /*
186          * Handle contiguousness
187          */
188         if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
189             ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
190             ins->ei_flags == emi->ei_flags) {
191                 emi->ei_clusters += ins->ei_clusters;
192                 return 1;
193         } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
194                    (ins->ei_cpos + ins->ei_clusters) == emi->ei_phys &&
195                    ins->ei_flags == emi->ei_flags) {
196                 emi->ei_phys = ins->ei_phys;
197                 emi->ei_cpos = ins->ei_cpos;
198                 emi->ei_clusters += ins->ei_clusters;
199                 return 1;
200         }
201
202         /*
203          * Overlapping extents - this shouldn't happen unless we've
204          * split an extent to change it's flags. That is exceedingly
205          * rare, so there's no sense in trying to optimize it yet.
206          */
207         if (ocfs2_ei_is_contained(emi, ins) ||
208             ocfs2_ei_is_contained(ins, emi)) {
209                 ocfs2_copy_emi_fields(emi, ins);
210                 return 1;
211         }
212
213         /* No merge was possible. */
214         return 0;
215 }
216
217 /*
218  * In order to reduce complexity on the caller, this insert function
219  * is intentionally liberal in what it will accept.
220  *
221  * The only rule is that the truncate call *must* be used whenever
222  * records have been deleted. This avoids inserting overlapping
223  * records with different physical mappings.
224  */
225 void ocfs2_extent_map_insert_rec(struct inode *inode,
226                                  struct ocfs2_extent_rec *rec)
227 {
228         struct ocfs2_inode_info *oi = OCFS2_I(inode);
229         struct ocfs2_extent_map *em = &oi->ip_extent_map;
230         struct ocfs2_extent_map_item *emi, *new_emi = NULL;
231         struct ocfs2_extent_map_item ins;
232
233         ins.ei_cpos = le32_to_cpu(rec->e_cpos);
234         ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
235                                                le64_to_cpu(rec->e_blkno));
236         ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
237         ins.ei_flags = rec->e_flags;
238
239 search:
240         spin_lock(&oi->ip_lock);
241
242         list_for_each_entry(emi, &em->em_list, ei_list) {
243                 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
244                         list_move(&emi->ei_list, &em->em_list);
245                         spin_unlock(&oi->ip_lock);
246                         goto out;
247                 }
248         }
249
250         /*
251          * No item could be merged.
252          *
253          * Either allocate and add a new item, or overwrite the last recently
254          * inserted.
255          */
256
257         if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
258                 if (new_emi == NULL) {
259                         spin_unlock(&oi->ip_lock);
260
261                         new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
262                         if (new_emi == NULL)
263                                 goto out;
264
265                         goto search;
266                 }
267
268                 ocfs2_copy_emi_fields(new_emi, &ins);
269                 list_add(&new_emi->ei_list, &em->em_list);
270                 em->em_num_items++;
271                 new_emi = NULL;
272         } else {
273                 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
274                 emi = list_entry(em->em_list.prev,
275                                  struct ocfs2_extent_map_item, ei_list);
276                 list_move(&emi->ei_list, &em->em_list);
277                 ocfs2_copy_emi_fields(emi, &ins);
278         }
279
280         spin_unlock(&oi->ip_lock);
281
282 out:
283         if (new_emi)
284                 kfree(new_emi);
285 }
286
287 static int ocfs2_last_eb_is_empty(struct inode *inode,
288                                   struct ocfs2_dinode *di)
289 {
290         int ret, next_free;
291         u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
292         struct buffer_head *eb_bh = NULL;
293         struct ocfs2_extent_block *eb;
294         struct ocfs2_extent_list *el;
295
296         ret = ocfs2_read_block(inode, last_eb_blk, &eb_bh);
297         if (ret) {
298                 mlog_errno(ret);
299                 goto out;
300         }
301
302         eb = (struct ocfs2_extent_block *) eb_bh->b_data;
303         el = &eb->h_list;
304
305         if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
306                 ret = -EROFS;
307                 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
308                 goto out;
309         }
310
311         if (el->l_tree_depth) {
312                 ocfs2_error(inode->i_sb,
313                             "Inode %lu has non zero tree depth in "
314                             "leaf block %llu\n", inode->i_ino,
315                             (unsigned long long)eb_bh->b_blocknr);
316                 ret = -EROFS;
317                 goto out;
318         }
319
320         next_free = le16_to_cpu(el->l_next_free_rec);
321
322         if (next_free == 0 ||
323             (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
324                 ret = 1;
325
326 out:
327         brelse(eb_bh);
328         return ret;
329 }
330
331 /*
332  * Return the 1st index within el which contains an extent start
333  * larger than v_cluster.
334  */
335 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
336                                        u32 v_cluster)
337 {
338         int i;
339         struct ocfs2_extent_rec *rec;
340
341         for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
342                 rec = &el->l_recs[i];
343
344                 if (v_cluster < le32_to_cpu(rec->e_cpos))
345                         break;
346         }
347
348         return i;
349 }
350
351 /*
352  * Figure out the size of a hole which starts at v_cluster within the given
353  * extent list.
354  *
355  * If there is no more allocation past v_cluster, we return the maximum
356  * cluster size minus v_cluster.
357  *
358  * If we have in-inode extents, then el points to the dinode list and
359  * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
360  * containing el.
361  */
362 static int ocfs2_figure_hole_clusters(struct inode *inode,
363                                       struct ocfs2_extent_list *el,
364                                       struct buffer_head *eb_bh,
365                                       u32 v_cluster,
366                                       u32 *num_clusters)
367 {
368         int ret, i;
369         struct buffer_head *next_eb_bh = NULL;
370         struct ocfs2_extent_block *eb, *next_eb;
371
372         i = ocfs2_search_for_hole_index(el, v_cluster);
373
374         if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
375                 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
376
377                 /*
378                  * Check the next leaf for any extents.
379                  */
380
381                 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
382                         goto no_more_extents;
383
384                 ret = ocfs2_read_block(inode,
385                                        le64_to_cpu(eb->h_next_leaf_blk),
386                                        &next_eb_bh);
387                 if (ret) {
388                         mlog_errno(ret);
389                         goto out;
390                 }
391                 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
392
393                 if (!OCFS2_IS_VALID_EXTENT_BLOCK(next_eb)) {
394                         ret = -EROFS;
395                         OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, next_eb);
396                         goto out;
397                 }
398
399                 el = &next_eb->h_list;
400
401                 i = ocfs2_search_for_hole_index(el, v_cluster);
402         }
403
404 no_more_extents:
405         if (i == le16_to_cpu(el->l_next_free_rec)) {
406                 /*
407                  * We're at the end of our existing allocation. Just
408                  * return the maximum number of clusters we could
409                  * possibly allocate.
410                  */
411                 *num_clusters = UINT_MAX - v_cluster;
412         } else {
413                 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
414         }
415
416         ret = 0;
417 out:
418         brelse(next_eb_bh);
419         return ret;
420 }
421
422 static int ocfs2_get_clusters_nocache(struct inode *inode,
423                                       struct buffer_head *di_bh,
424                                       u32 v_cluster, unsigned int *hole_len,
425                                       struct ocfs2_extent_rec *ret_rec,
426                                       unsigned int *is_last)
427 {
428         int i, ret, tree_height, len;
429         struct ocfs2_dinode *di;
430         struct ocfs2_extent_block *uninitialized_var(eb);
431         struct ocfs2_extent_list *el;
432         struct ocfs2_extent_rec *rec;
433         struct buffer_head *eb_bh = NULL;
434
435         memset(ret_rec, 0, sizeof(*ret_rec));
436         if (is_last)
437                 *is_last = 0;
438
439         di = (struct ocfs2_dinode *) di_bh->b_data;
440         el = &di->id2.i_list;
441         tree_height = le16_to_cpu(el->l_tree_depth);
442
443         if (tree_height > 0) {
444                 ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
445                 if (ret) {
446                         mlog_errno(ret);
447                         goto out;
448                 }
449
450                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
451                 el = &eb->h_list;
452
453                 if (el->l_tree_depth) {
454                         ocfs2_error(inode->i_sb,
455                                     "Inode %lu has non zero tree depth in "
456                                     "leaf block %llu\n", inode->i_ino,
457                                     (unsigned long long)eb_bh->b_blocknr);
458                         ret = -EROFS;
459                         goto out;
460                 }
461         }
462
463         i = ocfs2_search_extent_list(el, v_cluster);
464         if (i == -1) {
465                 /*
466                  * Holes can be larger than the maximum size of an
467                  * extent, so we return their lengths in a seperate
468                  * field.
469                  */
470                 if (hole_len) {
471                         ret = ocfs2_figure_hole_clusters(inode, el, eb_bh,
472                                                          v_cluster, &len);
473                         if (ret) {
474                                 mlog_errno(ret);
475                                 goto out;
476                         }
477
478                         *hole_len = len;
479                 }
480                 goto out_hole;
481         }
482
483         rec = &el->l_recs[i];
484
485         BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
486
487         if (!rec->e_blkno) {
488                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
489                             "record (%u, %u, 0)", inode->i_ino,
490                             le32_to_cpu(rec->e_cpos),
491                             ocfs2_rec_clusters(el, rec));
492                 ret = -EROFS;
493                 goto out;
494         }
495
496         *ret_rec = *rec;
497
498         /*
499          * Checking for last extent is potentially expensive - we
500          * might have to look at the next leaf over to see if it's
501          * empty.
502          *
503          * The first two checks are to see whether the caller even
504          * cares for this information, and if the extent is at least
505          * the last in it's list.
506          *
507          * If those hold true, then the extent is last if any of the
508          * additional conditions hold true:
509          *  - Extent list is in-inode
510          *  - Extent list is right-most
511          *  - Extent list is 2nd to rightmost, with empty right-most
512          */
513         if (is_last) {
514                 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
515                         if (tree_height == 0)
516                                 *is_last = 1;
517                         else if (eb->h_blkno == di->i_last_eb_blk)
518                                 *is_last = 1;
519                         else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
520                                 ret = ocfs2_last_eb_is_empty(inode, di);
521                                 if (ret < 0) {
522                                         mlog_errno(ret);
523                                         goto out;
524                                 }
525                                 if (ret == 1)
526                                         *is_last = 1;
527                         }
528                 }
529         }
530
531 out_hole:
532         ret = 0;
533 out:
534         brelse(eb_bh);
535         return ret;
536 }
537
538 static void ocfs2_relative_extent_offsets(struct super_block *sb,
539                                           u32 v_cluster,
540                                           struct ocfs2_extent_rec *rec,
541                                           u32 *p_cluster, u32 *num_clusters)
542
543 {
544         u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
545
546         *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
547         *p_cluster = *p_cluster + coff;
548
549         if (num_clusters)
550                 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
551 }
552
553 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
554                              u32 *p_cluster, u32 *num_clusters,
555                              struct ocfs2_extent_list *el)
556 {
557         int ret = 0, i;
558         struct buffer_head *eb_bh = NULL;
559         struct ocfs2_extent_block *eb;
560         struct ocfs2_extent_rec *rec;
561         u32 coff;
562
563         if (el->l_tree_depth) {
564                 ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
565                 if (ret) {
566                         mlog_errno(ret);
567                         goto out;
568                 }
569
570                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
571                 el = &eb->h_list;
572
573                 if (el->l_tree_depth) {
574                         ocfs2_error(inode->i_sb,
575                                     "Inode %lu has non zero tree depth in "
576                                     "xattr leaf block %llu\n", inode->i_ino,
577                                     (unsigned long long)eb_bh->b_blocknr);
578                         ret = -EROFS;
579                         goto out;
580                 }
581         }
582
583         i = ocfs2_search_extent_list(el, v_cluster);
584         if (i == -1) {
585                 ret = -EROFS;
586                 mlog_errno(ret);
587                 goto out;
588         } else {
589                 rec = &el->l_recs[i];
590                 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
591
592                 if (!rec->e_blkno) {
593                         ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
594                                     "record (%u, %u, 0) in xattr", inode->i_ino,
595                                     le32_to_cpu(rec->e_cpos),
596                                     ocfs2_rec_clusters(el, rec));
597                         ret = -EROFS;
598                         goto out;
599                 }
600                 coff = v_cluster - le32_to_cpu(rec->e_cpos);
601                 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
602                                                     le64_to_cpu(rec->e_blkno));
603                 *p_cluster = *p_cluster + coff;
604                 if (num_clusters)
605                         *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
606         }
607 out:
608         if (eb_bh)
609                 brelse(eb_bh);
610         return ret;
611 }
612
613 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
614                        u32 *p_cluster, u32 *num_clusters,
615                        unsigned int *extent_flags)
616 {
617         int ret;
618         unsigned int uninitialized_var(hole_len), flags = 0;
619         struct buffer_head *di_bh = NULL;
620         struct ocfs2_extent_rec rec;
621
622         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
623                 ret = -ERANGE;
624                 mlog_errno(ret);
625                 goto out;
626         }
627
628         ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
629                                       num_clusters, extent_flags);
630         if (ret == 0)
631                 goto out;
632
633         ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &di_bh);
634         if (ret) {
635                 mlog_errno(ret);
636                 goto out;
637         }
638
639         ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
640                                          &rec, NULL);
641         if (ret) {
642                 mlog_errno(ret);
643                 goto out;
644         }
645
646         if (rec.e_blkno == 0ULL) {
647                 /*
648                  * A hole was found. Return some canned values that
649                  * callers can key on. If asked for, num_clusters will
650                  * be populated with the size of the hole.
651                  */
652                 *p_cluster = 0;
653                 if (num_clusters) {
654                         *num_clusters = hole_len;
655                 }
656         } else {
657                 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
658                                               p_cluster, num_clusters);
659                 flags = rec.e_flags;
660
661                 ocfs2_extent_map_insert_rec(inode, &rec);
662         }
663
664         if (extent_flags)
665                 *extent_flags = flags;
666
667 out:
668         brelse(di_bh);
669         return ret;
670 }
671
672 /*
673  * This expects alloc_sem to be held. The allocation cannot change at
674  * all while the map is in the process of being updated.
675  */
676 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
677                                 u64 *ret_count, unsigned int *extent_flags)
678 {
679         int ret;
680         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
681         u32 cpos, num_clusters, p_cluster;
682         u64 boff = 0;
683
684         cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
685
686         ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
687                                  extent_flags);
688         if (ret) {
689                 mlog_errno(ret);
690                 goto out;
691         }
692
693         /*
694          * p_cluster == 0 indicates a hole.
695          */
696         if (p_cluster) {
697                 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
698                 boff += (v_blkno & (u64)(bpc - 1));
699         }
700
701         *p_blkno = boff;
702
703         if (ret_count) {
704                 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
705                 *ret_count -= v_blkno & (u64)(bpc - 1);
706         }
707
708 out:
709         return ret;
710 }
711
712 static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
713                                struct fiemap_extent_info *fieinfo,
714                                u64 map_start)
715 {
716         int ret;
717         unsigned int id_count;
718         struct ocfs2_dinode *di;
719         u64 phys;
720         u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
721         struct ocfs2_inode_info *oi = OCFS2_I(inode);
722
723         di = (struct ocfs2_dinode *)di_bh->b_data;
724         id_count = le16_to_cpu(di->id2.i_data.id_count);
725
726         if (map_start < id_count) {
727                 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
728                 phys += offsetof(struct ocfs2_dinode, id2.i_data.id_data);
729
730                 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
731                                               flags);
732                 if (ret < 0)
733                         return ret;
734         }
735
736         return 0;
737 }
738
739 #define OCFS2_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
740
741 int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
742                  u64 map_start, u64 map_len)
743 {
744         int ret, is_last;
745         u32 mapping_end, cpos;
746         unsigned int hole_size;
747         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
748         u64 len_bytes, phys_bytes, virt_bytes;
749         struct buffer_head *di_bh = NULL;
750         struct ocfs2_extent_rec rec;
751
752         ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
753         if (ret)
754                 return ret;
755
756         ret = ocfs2_inode_lock(inode, &di_bh, 0);
757         if (ret) {
758                 mlog_errno(ret);
759                 goto out;
760         }
761
762         down_read(&OCFS2_I(inode)->ip_alloc_sem);
763
764         /*
765          * Handle inline-data separately.
766          */
767         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
768                 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
769                 goto out_unlock;
770         }
771
772         cpos = map_start >> osb->s_clustersize_bits;
773         mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
774                                                map_start + map_len);
775         mapping_end -= cpos;
776         is_last = 0;
777         while (cpos < mapping_end && !is_last) {
778                 u32 fe_flags;
779
780                 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
781                                                  &hole_size, &rec, &is_last);
782                 if (ret) {
783                         mlog_errno(ret);
784                         goto out;
785                 }
786
787                 if (rec.e_blkno == 0ULL) {
788                         cpos += hole_size;
789                         continue;
790                 }
791
792                 fe_flags = 0;
793                 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
794                         fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
795                 if (is_last)
796                         fe_flags |= FIEMAP_EXTENT_LAST;
797                 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
798                 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
799                 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
800
801                 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
802                                               len_bytes, fe_flags);
803                 if (ret)
804                         break;
805
806                 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
807         }
808
809         if (ret > 0)
810                 ret = 0;
811
812 out_unlock:
813         brelse(di_bh);
814
815         up_read(&OCFS2_I(inode)->ip_alloc_sem);
816
817         ocfs2_inode_unlock(inode, 0);
818 out:
819
820         return ret;
821 }