ocfs2: fiemap support
[linux-2.6.git] / fs / ocfs2 / extent_map.c
blobaed268e80b49173a5a3af264bcd5ca5c7e55ca8a
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * extent_map.c
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.
25 #include <linux/fs.h>
26 #include <linux/init.h>
27 #include <linux/types.h>
28 #include <linux/fiemap.h>
30 #define MLOG_MASK_PREFIX ML_EXTENT_MAP
31 #include <cluster/masklog.h>
33 #include "ocfs2.h"
35 #include "alloc.h"
36 #include "dlmglue.h"
37 #include "extent_map.h"
38 #include "inode.h"
39 #include "super.h"
41 #include "buffer_head_io.h"
44 * The extent caching implementation is intentionally trivial.
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.
52 void ocfs2_extent_map_init(struct inode *inode)
54 struct ocfs2_inode_info *oi = OCFS2_I(inode);
56 oi->ip_extent_map.em_num_items = 0;
57 INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
60 static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
61 unsigned int cpos,
62 struct ocfs2_extent_map_item **ret_emi)
64 unsigned int range;
65 struct ocfs2_extent_map_item *emi;
67 *ret_emi = NULL;
69 list_for_each_entry(emi, &em->em_list, ei_list) {
70 range = emi->ei_cpos + emi->ei_clusters;
72 if (cpos >= emi->ei_cpos && cpos < range) {
73 list_move(&emi->ei_list, &em->em_list);
75 *ret_emi = emi;
76 break;
81 static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
82 unsigned int *phys, unsigned int *len,
83 unsigned int *flags)
85 unsigned int coff;
86 struct ocfs2_inode_info *oi = OCFS2_I(inode);
87 struct ocfs2_extent_map_item *emi;
89 spin_lock(&oi->ip_lock);
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;
101 spin_unlock(&oi->ip_lock);
103 if (emi == NULL)
104 return -ENOENT;
106 return 0;
110 * Forget about all clusters equal to or greater than cpos.
112 void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
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;
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;
130 range = emi->ei_cpos + emi->ei_clusters;
131 if (range > cpos) {
132 /* Partial truncate */
133 emi->ei_clusters = cpos - emi->ei_cpos;
136 spin_unlock(&oi->ip_lock);
138 list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
139 list_del(&emi->ei_list);
140 kfree(emi);
145 * Is any part of emi2 contained within emi1
147 static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
148 struct ocfs2_extent_map_item *emi2)
150 unsigned int range1, range2;
153 * Check if logical start of emi2 is inside emi1
155 range1 = emi1->ei_cpos + emi1->ei_clusters;
156 if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
157 return 1;
160 * Check if logical end of emi2 is inside emi1
162 range2 = emi2->ei_cpos + emi2->ei_clusters;
163 if (range2 > emi1->ei_cpos && range2 <= range1)
164 return 1;
166 return 0;
169 static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
170 struct ocfs2_extent_map_item *src)
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;
179 * Try to merge emi with ins. Returns 1 if merge succeeds, zero
180 * otherwise.
182 static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
183 struct ocfs2_extent_map_item *ins)
186 * Handle contiguousness
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;
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.
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;
213 /* No merge was possible. */
214 return 0;
218 * In order to reduce complexity on the caller, this insert function
219 * is intentionally liberal in what it will accept.
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.
225 void ocfs2_extent_map_insert_rec(struct inode *inode,
226 struct ocfs2_extent_rec *rec)
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;
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;
239 search:
240 spin_lock(&oi->ip_lock);
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;
251 * No item could be merged.
253 * Either allocate and add a new item, or overwrite the last recently
254 * inserted.
257 if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
258 if (new_emi == NULL) {
259 spin_unlock(&oi->ip_lock);
261 new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
262 if (new_emi == NULL)
263 goto out;
265 goto search;
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);
280 spin_unlock(&oi->ip_lock);
282 out:
283 if (new_emi)
284 kfree(new_emi);
287 static int ocfs2_last_eb_is_empty(struct inode *inode,
288 struct ocfs2_dinode *di)
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;
296 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), last_eb_blk,
297 &eb_bh, OCFS2_BH_CACHED, inode);
298 if (ret) {
299 mlog_errno(ret);
300 goto out;
303 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
304 el = &eb->h_list;
306 if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
307 ret = -EROFS;
308 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
309 goto out;
312 if (el->l_tree_depth) {
313 ocfs2_error(inode->i_sb,
314 "Inode %lu has non zero tree depth in "
315 "leaf block %llu\n", inode->i_ino,
316 (unsigned long long)eb_bh->b_blocknr);
317 ret = -EROFS;
318 goto out;
321 next_free = le16_to_cpu(el->l_next_free_rec);
323 if (next_free == 0 ||
324 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
325 ret = 1;
327 out:
328 brelse(eb_bh);
329 return ret;
333 * Return the 1st index within el which contains an extent start
334 * larger than v_cluster.
336 static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
337 u32 v_cluster)
339 int i;
340 struct ocfs2_extent_rec *rec;
342 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
343 rec = &el->l_recs[i];
345 if (v_cluster < le32_to_cpu(rec->e_cpos))
346 break;
349 return i;
353 * Figure out the size of a hole which starts at v_cluster within the given
354 * extent list.
356 * If there is no more allocation past v_cluster, we return the maximum
357 * cluster size minus v_cluster.
359 * If we have in-inode extents, then el points to the dinode list and
360 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
361 * containing el.
363 static int ocfs2_figure_hole_clusters(struct inode *inode,
364 struct ocfs2_extent_list *el,
365 struct buffer_head *eb_bh,
366 u32 v_cluster,
367 u32 *num_clusters)
369 int ret, i;
370 struct buffer_head *next_eb_bh = NULL;
371 struct ocfs2_extent_block *eb, *next_eb;
373 i = ocfs2_search_for_hole_index(el, v_cluster);
375 if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
376 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
379 * Check the next leaf for any extents.
382 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
383 goto no_more_extents;
385 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
386 le64_to_cpu(eb->h_next_leaf_blk),
387 &next_eb_bh, OCFS2_BH_CACHED, inode);
388 if (ret) {
389 mlog_errno(ret);
390 goto out;
392 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
394 if (!OCFS2_IS_VALID_EXTENT_BLOCK(next_eb)) {
395 ret = -EROFS;
396 OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, next_eb);
397 goto out;
400 el = &next_eb->h_list;
402 i = ocfs2_search_for_hole_index(el, v_cluster);
405 no_more_extents:
406 if (i == le16_to_cpu(el->l_next_free_rec)) {
408 * We're at the end of our existing allocation. Just
409 * return the maximum number of clusters we could
410 * possibly allocate.
412 *num_clusters = UINT_MAX - v_cluster;
413 } else {
414 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
417 ret = 0;
418 out:
419 brelse(next_eb_bh);
420 return ret;
423 static int ocfs2_get_clusters_nocache(struct inode *inode,
424 struct buffer_head *di_bh,
425 u32 v_cluster, unsigned int *hole_len,
426 struct ocfs2_extent_rec *ret_rec,
427 unsigned int *is_last)
429 int i, ret, tree_height, len;
430 struct ocfs2_dinode *di;
431 struct ocfs2_extent_block *uninitialized_var(eb);
432 struct ocfs2_extent_list *el;
433 struct ocfs2_extent_rec *rec;
434 struct buffer_head *eb_bh = NULL;
436 memset(ret_rec, 0, sizeof(*ret_rec));
437 if (is_last)
438 *is_last = 0;
440 di = (struct ocfs2_dinode *) di_bh->b_data;
441 el = &di->id2.i_list;
442 tree_height = le16_to_cpu(el->l_tree_depth);
444 if (tree_height > 0) {
445 ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
446 if (ret) {
447 mlog_errno(ret);
448 goto out;
451 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
452 el = &eb->h_list;
454 if (el->l_tree_depth) {
455 ocfs2_error(inode->i_sb,
456 "Inode %lu has non zero tree depth in "
457 "leaf block %llu\n", inode->i_ino,
458 (unsigned long long)eb_bh->b_blocknr);
459 ret = -EROFS;
460 goto out;
464 i = ocfs2_search_extent_list(el, v_cluster);
465 if (i == -1) {
467 * Holes can be larger than the maximum size of an
468 * extent, so we return their lengths in a seperate
469 * field.
471 if (hole_len) {
472 ret = ocfs2_figure_hole_clusters(inode, el, eb_bh,
473 v_cluster, &len);
474 if (ret) {
475 mlog_errno(ret);
476 goto out;
479 *hole_len = len;
481 goto out_hole;
484 rec = &el->l_recs[i];
486 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
488 if (!rec->e_blkno) {
489 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
490 "record (%u, %u, 0)", inode->i_ino,
491 le32_to_cpu(rec->e_cpos),
492 ocfs2_rec_clusters(el, rec));
493 ret = -EROFS;
494 goto out;
497 *ret_rec = *rec;
500 * Checking for last extent is potentially expensive - we
501 * might have to look at the next leaf over to see if it's
502 * empty.
504 * The first two checks are to see whether the caller even
505 * cares for this information, and if the extent is at least
506 * the last in it's list.
508 * If those hold true, then the extent is last if any of the
509 * additional conditions hold true:
510 * - Extent list is in-inode
511 * - Extent list is right-most
512 * - Extent list is 2nd to rightmost, with empty right-most
514 if (is_last) {
515 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
516 if (tree_height == 0)
517 *is_last = 1;
518 else if (eb->h_blkno == di->i_last_eb_blk)
519 *is_last = 1;
520 else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
521 ret = ocfs2_last_eb_is_empty(inode, di);
522 if (ret < 0) {
523 mlog_errno(ret);
524 goto out;
526 if (ret == 1)
527 *is_last = 1;
532 out_hole:
533 ret = 0;
534 out:
535 brelse(eb_bh);
536 return ret;
539 static void ocfs2_relative_extent_offsets(struct super_block *sb,
540 u32 v_cluster,
541 struct ocfs2_extent_rec *rec,
542 u32 *p_cluster, u32 *num_clusters)
545 u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
547 *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
548 *p_cluster = *p_cluster + coff;
550 if (num_clusters)
551 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
554 int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
555 u32 *p_cluster, u32 *num_clusters,
556 unsigned int *extent_flags)
558 int ret;
559 unsigned int uninitialized_var(hole_len), flags = 0;
560 struct buffer_head *di_bh = NULL;
561 struct ocfs2_extent_rec rec;
563 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
564 ret = -ERANGE;
565 mlog_errno(ret);
566 goto out;
569 ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
570 num_clusters, extent_flags);
571 if (ret == 0)
572 goto out;
574 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), OCFS2_I(inode)->ip_blkno,
575 &di_bh, OCFS2_BH_CACHED, inode);
576 if (ret) {
577 mlog_errno(ret);
578 goto out;
581 ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
582 &rec, NULL);
583 if (ret) {
584 mlog_errno(ret);
585 goto out;
588 if (rec.e_blkno == 0ULL) {
590 * A hole was found. Return some canned values that
591 * callers can key on. If asked for, num_clusters will
592 * be populated with the size of the hole.
594 *p_cluster = 0;
595 if (num_clusters) {
596 *num_clusters = hole_len;
598 } else {
599 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
600 p_cluster, num_clusters);
601 flags = rec.e_flags;
603 ocfs2_extent_map_insert_rec(inode, &rec);
606 if (extent_flags)
607 *extent_flags = flags;
609 out:
610 brelse(di_bh);
611 return ret;
615 * This expects alloc_sem to be held. The allocation cannot change at
616 * all while the map is in the process of being updated.
618 int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
619 u64 *ret_count, unsigned int *extent_flags)
621 int ret;
622 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
623 u32 cpos, num_clusters, p_cluster;
624 u64 boff = 0;
626 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
628 ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
629 extent_flags);
630 if (ret) {
631 mlog_errno(ret);
632 goto out;
636 * p_cluster == 0 indicates a hole.
638 if (p_cluster) {
639 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
640 boff += (v_blkno & (u64)(bpc - 1));
643 *p_blkno = boff;
645 if (ret_count) {
646 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
647 *ret_count -= v_blkno & (u64)(bpc - 1);
650 out:
651 return ret;
654 static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
655 struct fiemap_extent_info *fieinfo,
656 u64 map_start)
658 int ret;
659 unsigned int id_count;
660 struct ocfs2_dinode *di;
661 u64 phys;
662 u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
663 struct ocfs2_inode_info *oi = OCFS2_I(inode);
665 di = (struct ocfs2_dinode *)di_bh->b_data;
666 id_count = le16_to_cpu(di->id2.i_data.id_count);
668 if (map_start < id_count) {
669 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
670 phys += offsetof(struct ocfs2_dinode, id2.i_data.id_data);
672 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
673 flags);
674 if (ret < 0)
675 return ret;
678 return 0;
681 #define OCFS2_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
683 int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
684 u64 map_start, u64 map_len)
686 int ret, is_last;
687 u32 mapping_end, cpos;
688 unsigned int hole_size;
689 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
690 u64 len_bytes, phys_bytes, virt_bytes;
691 struct buffer_head *di_bh = NULL;
692 struct ocfs2_extent_rec rec;
694 ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
695 if (ret)
696 return ret;
698 ret = ocfs2_inode_lock(inode, &di_bh, 0);
699 if (ret) {
700 mlog_errno(ret);
701 goto out;
704 down_read(&OCFS2_I(inode)->ip_alloc_sem);
707 * Handle inline-data separately.
709 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
710 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
711 goto out_unlock;
714 cpos = map_start >> osb->s_clustersize_bits;
715 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
716 map_start + map_len);
717 mapping_end -= cpos;
718 is_last = 0;
719 while (cpos < mapping_end && !is_last) {
720 u32 fe_flags;
722 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
723 &hole_size, &rec, &is_last);
724 if (ret) {
725 mlog_errno(ret);
726 goto out;
729 if (rec.e_blkno == 0ULL) {
730 cpos += hole_size;
731 continue;
734 fe_flags = 0;
735 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
736 fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
737 if (is_last)
738 fe_flags |= FIEMAP_EXTENT_LAST;
739 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
740 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
741 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
743 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
744 len_bytes, fe_flags);
745 if (ret)
746 break;
748 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
751 if (ret > 0)
752 ret = 0;
754 out_unlock:
755 brelse(di_bh);
757 up_read(&OCFS2_I(inode)->ip_alloc_sem);
759 ocfs2_inode_unlock(inode, 0);
760 out:
762 return ret;