Loading...
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/slab.h>
28#include <linux/types.h>
29#include <linux/fiemap.h>
30
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#include "symlink.h"
41#include "ocfs2_trace.h"
42
43#include "buffer_head_io.h"
44
45/*
46 * The extent caching implementation is intentionally trivial.
47 *
48 * We only cache a small number of extents stored directly on the
49 * inode, so linear order operations are acceptable. If we ever want
50 * to increase the size of the extent map, then these algorithms must
51 * get smarter.
52 */
53
54void ocfs2_extent_map_init(struct inode *inode)
55{
56 struct ocfs2_inode_info *oi = OCFS2_I(inode);
57
58 oi->ip_extent_map.em_num_items = 0;
59 INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
60}
61
62static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
63 unsigned int cpos,
64 struct ocfs2_extent_map_item **ret_emi)
65{
66 unsigned int range;
67 struct ocfs2_extent_map_item *emi;
68
69 *ret_emi = NULL;
70
71 list_for_each_entry(emi, &em->em_list, ei_list) {
72 range = emi->ei_cpos + emi->ei_clusters;
73
74 if (cpos >= emi->ei_cpos && cpos < range) {
75 list_move(&emi->ei_list, &em->em_list);
76
77 *ret_emi = emi;
78 break;
79 }
80 }
81}
82
83static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
84 unsigned int *phys, unsigned int *len,
85 unsigned int *flags)
86{
87 unsigned int coff;
88 struct ocfs2_inode_info *oi = OCFS2_I(inode);
89 struct ocfs2_extent_map_item *emi;
90
91 spin_lock(&oi->ip_lock);
92
93 __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
94 if (emi) {
95 coff = cpos - emi->ei_cpos;
96 *phys = emi->ei_phys + coff;
97 if (len)
98 *len = emi->ei_clusters - coff;
99 if (flags)
100 *flags = emi->ei_flags;
101 }
102
103 spin_unlock(&oi->ip_lock);
104
105 if (emi == NULL)
106 return -ENOENT;
107
108 return 0;
109}
110
111/*
112 * Forget about all clusters equal to or greater than cpos.
113 */
114void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
115{
116 struct ocfs2_extent_map_item *emi, *n;
117 struct ocfs2_inode_info *oi = OCFS2_I(inode);
118 struct ocfs2_extent_map *em = &oi->ip_extent_map;
119 LIST_HEAD(tmp_list);
120 unsigned int range;
121
122 spin_lock(&oi->ip_lock);
123 list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
124 if (emi->ei_cpos >= cpos) {
125 /* Full truncate of this record. */
126 list_move(&emi->ei_list, &tmp_list);
127 BUG_ON(em->em_num_items == 0);
128 em->em_num_items--;
129 continue;
130 }
131
132 range = emi->ei_cpos + emi->ei_clusters;
133 if (range > cpos) {
134 /* Partial truncate */
135 emi->ei_clusters = cpos - emi->ei_cpos;
136 }
137 }
138 spin_unlock(&oi->ip_lock);
139
140 list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
141 list_del(&emi->ei_list);
142 kfree(emi);
143 }
144}
145
146/*
147 * Is any part of emi2 contained within emi1
148 */
149static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
150 struct ocfs2_extent_map_item *emi2)
151{
152 unsigned int range1, range2;
153
154 /*
155 * Check if logical start of emi2 is inside emi1
156 */
157 range1 = emi1->ei_cpos + emi1->ei_clusters;
158 if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
159 return 1;
160
161 /*
162 * Check if logical end of emi2 is inside emi1
163 */
164 range2 = emi2->ei_cpos + emi2->ei_clusters;
165 if (range2 > emi1->ei_cpos && range2 <= range1)
166 return 1;
167
168 return 0;
169}
170
171static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
172 struct ocfs2_extent_map_item *src)
173{
174 dest->ei_cpos = src->ei_cpos;
175 dest->ei_phys = src->ei_phys;
176 dest->ei_clusters = src->ei_clusters;
177 dest->ei_flags = src->ei_flags;
178}
179
180/*
181 * Try to merge emi with ins. Returns 1 if merge succeeds, zero
182 * otherwise.
183 */
184static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
185 struct ocfs2_extent_map_item *ins)
186{
187 /*
188 * Handle contiguousness
189 */
190 if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
191 ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
192 ins->ei_flags == emi->ei_flags) {
193 emi->ei_clusters += ins->ei_clusters;
194 return 1;
195 } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
196 (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
197 ins->ei_flags == emi->ei_flags) {
198 emi->ei_phys = ins->ei_phys;
199 emi->ei_cpos = ins->ei_cpos;
200 emi->ei_clusters += ins->ei_clusters;
201 return 1;
202 }
203
204 /*
205 * Overlapping extents - this shouldn't happen unless we've
206 * split an extent to change it's flags. That is exceedingly
207 * rare, so there's no sense in trying to optimize it yet.
208 */
209 if (ocfs2_ei_is_contained(emi, ins) ||
210 ocfs2_ei_is_contained(ins, emi)) {
211 ocfs2_copy_emi_fields(emi, ins);
212 return 1;
213 }
214
215 /* No merge was possible. */
216 return 0;
217}
218
219/*
220 * In order to reduce complexity on the caller, this insert function
221 * is intentionally liberal in what it will accept.
222 *
223 * The only rule is that the truncate call *must* be used whenever
224 * records have been deleted. This avoids inserting overlapping
225 * records with different physical mappings.
226 */
227void ocfs2_extent_map_insert_rec(struct inode *inode,
228 struct ocfs2_extent_rec *rec)
229{
230 struct ocfs2_inode_info *oi = OCFS2_I(inode);
231 struct ocfs2_extent_map *em = &oi->ip_extent_map;
232 struct ocfs2_extent_map_item *emi, *new_emi = NULL;
233 struct ocfs2_extent_map_item ins;
234
235 ins.ei_cpos = le32_to_cpu(rec->e_cpos);
236 ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
237 le64_to_cpu(rec->e_blkno));
238 ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
239 ins.ei_flags = rec->e_flags;
240
241search:
242 spin_lock(&oi->ip_lock);
243
244 list_for_each_entry(emi, &em->em_list, ei_list) {
245 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
246 list_move(&emi->ei_list, &em->em_list);
247 spin_unlock(&oi->ip_lock);
248 goto out;
249 }
250 }
251
252 /*
253 * No item could be merged.
254 *
255 * Either allocate and add a new item, or overwrite the last recently
256 * inserted.
257 */
258
259 if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
260 if (new_emi == NULL) {
261 spin_unlock(&oi->ip_lock);
262
263 new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
264 if (new_emi == NULL)
265 goto out;
266
267 goto search;
268 }
269
270 ocfs2_copy_emi_fields(new_emi, &ins);
271 list_add(&new_emi->ei_list, &em->em_list);
272 em->em_num_items++;
273 new_emi = NULL;
274 } else {
275 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
276 emi = list_entry(em->em_list.prev,
277 struct ocfs2_extent_map_item, ei_list);
278 list_move(&emi->ei_list, &em->em_list);
279 ocfs2_copy_emi_fields(emi, &ins);
280 }
281
282 spin_unlock(&oi->ip_lock);
283
284out:
285 kfree(new_emi);
286}
287
288static int ocfs2_last_eb_is_empty(struct inode *inode,
289 struct ocfs2_dinode *di)
290{
291 int ret, next_free;
292 u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
293 struct buffer_head *eb_bh = NULL;
294 struct ocfs2_extent_block *eb;
295 struct ocfs2_extent_list *el;
296
297 ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
298 if (ret) {
299 mlog_errno(ret);
300 goto out;
301 }
302
303 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
304 el = &eb->h_list;
305
306 if (el->l_tree_depth) {
307 ocfs2_error(inode->i_sb,
308 "Inode %lu has non zero tree depth in leaf block %llu\n",
309 inode->i_ino,
310 (unsigned long long)eb_bh->b_blocknr);
311 ret = -EROFS;
312 goto out;
313 }
314
315 next_free = le16_to_cpu(el->l_next_free_rec);
316
317 if (next_free == 0 ||
318 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
319 ret = 1;
320
321out:
322 brelse(eb_bh);
323 return ret;
324}
325
326/*
327 * Return the 1st index within el which contains an extent start
328 * larger than v_cluster.
329 */
330static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
331 u32 v_cluster)
332{
333 int i;
334 struct ocfs2_extent_rec *rec;
335
336 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
337 rec = &el->l_recs[i];
338
339 if (v_cluster < le32_to_cpu(rec->e_cpos))
340 break;
341 }
342
343 return i;
344}
345
346/*
347 * Figure out the size of a hole which starts at v_cluster within the given
348 * extent list.
349 *
350 * If there is no more allocation past v_cluster, we return the maximum
351 * cluster size minus v_cluster.
352 *
353 * If we have in-inode extents, then el points to the dinode list and
354 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
355 * containing el.
356 */
357int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
358 struct ocfs2_extent_list *el,
359 struct buffer_head *eb_bh,
360 u32 v_cluster,
361 u32 *num_clusters)
362{
363 int ret, i;
364 struct buffer_head *next_eb_bh = NULL;
365 struct ocfs2_extent_block *eb, *next_eb;
366
367 i = ocfs2_search_for_hole_index(el, v_cluster);
368
369 if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
370 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
371
372 /*
373 * Check the next leaf for any extents.
374 */
375
376 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
377 goto no_more_extents;
378
379 ret = ocfs2_read_extent_block(ci,
380 le64_to_cpu(eb->h_next_leaf_blk),
381 &next_eb_bh);
382 if (ret) {
383 mlog_errno(ret);
384 goto out;
385 }
386
387 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
388 el = &next_eb->h_list;
389 i = ocfs2_search_for_hole_index(el, v_cluster);
390 }
391
392no_more_extents:
393 if (i == le16_to_cpu(el->l_next_free_rec)) {
394 /*
395 * We're at the end of our existing allocation. Just
396 * return the maximum number of clusters we could
397 * possibly allocate.
398 */
399 *num_clusters = UINT_MAX - v_cluster;
400 } else {
401 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
402 }
403
404 ret = 0;
405out:
406 brelse(next_eb_bh);
407 return ret;
408}
409
410static int ocfs2_get_clusters_nocache(struct inode *inode,
411 struct buffer_head *di_bh,
412 u32 v_cluster, unsigned int *hole_len,
413 struct ocfs2_extent_rec *ret_rec,
414 unsigned int *is_last)
415{
416 int i, ret, tree_height, len;
417 struct ocfs2_dinode *di;
418 struct ocfs2_extent_block *uninitialized_var(eb);
419 struct ocfs2_extent_list *el;
420 struct ocfs2_extent_rec *rec;
421 struct buffer_head *eb_bh = NULL;
422
423 memset(ret_rec, 0, sizeof(*ret_rec));
424 if (is_last)
425 *is_last = 0;
426
427 di = (struct ocfs2_dinode *) di_bh->b_data;
428 el = &di->id2.i_list;
429 tree_height = le16_to_cpu(el->l_tree_depth);
430
431 if (tree_height > 0) {
432 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
433 &eb_bh);
434 if (ret) {
435 mlog_errno(ret);
436 goto out;
437 }
438
439 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
440 el = &eb->h_list;
441
442 if (el->l_tree_depth) {
443 ocfs2_error(inode->i_sb,
444 "Inode %lu has non zero tree depth in leaf block %llu\n",
445 inode->i_ino,
446 (unsigned long long)eb_bh->b_blocknr);
447 ret = -EROFS;
448 goto out;
449 }
450 }
451
452 i = ocfs2_search_extent_list(el, v_cluster);
453 if (i == -1) {
454 /*
455 * Holes can be larger than the maximum size of an
456 * extent, so we return their lengths in a separate
457 * field.
458 */
459 if (hole_len) {
460 ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
461 el, eb_bh,
462 v_cluster, &len);
463 if (ret) {
464 mlog_errno(ret);
465 goto out;
466 }
467
468 *hole_len = len;
469 }
470 goto out_hole;
471 }
472
473 rec = &el->l_recs[i];
474
475 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
476
477 if (!rec->e_blkno) {
478 ocfs2_error(inode->i_sb,
479 "Inode %lu has bad extent record (%u, %u, 0)\n",
480 inode->i_ino,
481 le32_to_cpu(rec->e_cpos),
482 ocfs2_rec_clusters(el, rec));
483 ret = -EROFS;
484 goto out;
485 }
486
487 *ret_rec = *rec;
488
489 /*
490 * Checking for last extent is potentially expensive - we
491 * might have to look at the next leaf over to see if it's
492 * empty.
493 *
494 * The first two checks are to see whether the caller even
495 * cares for this information, and if the extent is at least
496 * the last in it's list.
497 *
498 * If those hold true, then the extent is last if any of the
499 * additional conditions hold true:
500 * - Extent list is in-inode
501 * - Extent list is right-most
502 * - Extent list is 2nd to rightmost, with empty right-most
503 */
504 if (is_last) {
505 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
506 if (tree_height == 0)
507 *is_last = 1;
508 else if (eb->h_blkno == di->i_last_eb_blk)
509 *is_last = 1;
510 else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
511 ret = ocfs2_last_eb_is_empty(inode, di);
512 if (ret < 0) {
513 mlog_errno(ret);
514 goto out;
515 }
516 if (ret == 1)
517 *is_last = 1;
518 }
519 }
520 }
521
522out_hole:
523 ret = 0;
524out:
525 brelse(eb_bh);
526 return ret;
527}
528
529static void ocfs2_relative_extent_offsets(struct super_block *sb,
530 u32 v_cluster,
531 struct ocfs2_extent_rec *rec,
532 u32 *p_cluster, u32 *num_clusters)
533
534{
535 u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
536
537 *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
538 *p_cluster = *p_cluster + coff;
539
540 if (num_clusters)
541 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
542}
543
544int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
545 u32 *p_cluster, u32 *num_clusters,
546 struct ocfs2_extent_list *el,
547 unsigned int *extent_flags)
548{
549 int ret = 0, i;
550 struct buffer_head *eb_bh = NULL;
551 struct ocfs2_extent_block *eb;
552 struct ocfs2_extent_rec *rec;
553 u32 coff;
554
555 if (el->l_tree_depth) {
556 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
557 &eb_bh);
558 if (ret) {
559 mlog_errno(ret);
560 goto out;
561 }
562
563 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
564 el = &eb->h_list;
565
566 if (el->l_tree_depth) {
567 ocfs2_error(inode->i_sb,
568 "Inode %lu has non zero tree depth in xattr leaf block %llu\n",
569 inode->i_ino,
570 (unsigned long long)eb_bh->b_blocknr);
571 ret = -EROFS;
572 goto out;
573 }
574 }
575
576 i = ocfs2_search_extent_list(el, v_cluster);
577 if (i == -1) {
578 ret = -EROFS;
579 mlog_errno(ret);
580 goto out;
581 } else {
582 rec = &el->l_recs[i];
583 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
584
585 if (!rec->e_blkno) {
586 ocfs2_error(inode->i_sb,
587 "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
588 inode->i_ino,
589 le32_to_cpu(rec->e_cpos),
590 ocfs2_rec_clusters(el, rec));
591 ret = -EROFS;
592 goto out;
593 }
594 coff = v_cluster - le32_to_cpu(rec->e_cpos);
595 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
596 le64_to_cpu(rec->e_blkno));
597 *p_cluster = *p_cluster + coff;
598 if (num_clusters)
599 *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
600
601 if (extent_flags)
602 *extent_flags = rec->e_flags;
603 }
604out:
605 if (eb_bh)
606 brelse(eb_bh);
607 return ret;
608}
609
610int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
611 u32 *p_cluster, u32 *num_clusters,
612 unsigned int *extent_flags)
613{
614 int ret;
615 unsigned int uninitialized_var(hole_len), flags = 0;
616 struct buffer_head *di_bh = NULL;
617 struct ocfs2_extent_rec rec;
618
619 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
620 ret = -ERANGE;
621 mlog_errno(ret);
622 goto out;
623 }
624
625 ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
626 num_clusters, extent_flags);
627 if (ret == 0)
628 goto out;
629
630 ret = ocfs2_read_inode_block(inode, &di_bh);
631 if (ret) {
632 mlog_errno(ret);
633 goto out;
634 }
635
636 ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
637 &rec, NULL);
638 if (ret) {
639 mlog_errno(ret);
640 goto out;
641 }
642
643 if (rec.e_blkno == 0ULL) {
644 /*
645 * A hole was found. Return some canned values that
646 * callers can key on. If asked for, num_clusters will
647 * be populated with the size of the hole.
648 */
649 *p_cluster = 0;
650 if (num_clusters) {
651 *num_clusters = hole_len;
652 }
653 } else {
654 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
655 p_cluster, num_clusters);
656 flags = rec.e_flags;
657
658 ocfs2_extent_map_insert_rec(inode, &rec);
659 }
660
661 if (extent_flags)
662 *extent_flags = flags;
663
664out:
665 brelse(di_bh);
666 return ret;
667}
668
669/*
670 * This expects alloc_sem to be held. The allocation cannot change at
671 * all while the map is in the process of being updated.
672 */
673int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
674 u64 *ret_count, unsigned int *extent_flags)
675{
676 int ret;
677 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
678 u32 cpos, num_clusters, p_cluster;
679 u64 boff = 0;
680
681 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
682
683 ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
684 extent_flags);
685 if (ret) {
686 mlog_errno(ret);
687 goto out;
688 }
689
690 /*
691 * p_cluster == 0 indicates a hole.
692 */
693 if (p_cluster) {
694 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
695 boff += (v_blkno & (u64)(bpc - 1));
696 }
697
698 *p_blkno = boff;
699
700 if (ret_count) {
701 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
702 *ret_count -= v_blkno & (u64)(bpc - 1);
703 }
704
705out:
706 return ret;
707}
708
709/*
710 * The ocfs2_fiemap_inline() may be a little bit misleading, since
711 * it not only handles the fiemap for inlined files, but also deals
712 * with the fast symlink, cause they have no difference for extent
713 * mapping per se.
714 */
715static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
716 struct fiemap_extent_info *fieinfo,
717 u64 map_start)
718{
719 int ret;
720 unsigned int id_count;
721 struct ocfs2_dinode *di;
722 u64 phys;
723 u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
724 struct ocfs2_inode_info *oi = OCFS2_I(inode);
725
726 di = (struct ocfs2_dinode *)di_bh->b_data;
727 if (ocfs2_inode_is_fast_symlink(inode))
728 id_count = ocfs2_fast_symlink_chars(inode->i_sb);
729 else
730 id_count = le16_to_cpu(di->id2.i_data.id_count);
731
732 if (map_start < id_count) {
733 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
734 if (ocfs2_inode_is_fast_symlink(inode))
735 phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
736 else
737 phys += offsetof(struct ocfs2_dinode,
738 id2.i_data.id_data);
739
740 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
741 flags);
742 if (ret < 0)
743 return ret;
744 }
745
746 return 0;
747}
748
749#define OCFS2_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
750
751int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
752 u64 map_start, u64 map_len)
753{
754 int ret, is_last;
755 u32 mapping_end, cpos;
756 unsigned int hole_size;
757 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
758 u64 len_bytes, phys_bytes, virt_bytes;
759 struct buffer_head *di_bh = NULL;
760 struct ocfs2_extent_rec rec;
761
762 ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
763 if (ret)
764 return ret;
765
766 ret = ocfs2_inode_lock(inode, &di_bh, 0);
767 if (ret) {
768 mlog_errno(ret);
769 goto out;
770 }
771
772 down_read(&OCFS2_I(inode)->ip_alloc_sem);
773
774 /*
775 * Handle inline-data and fast symlink separately.
776 */
777 if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
778 ocfs2_inode_is_fast_symlink(inode)) {
779 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
780 goto out_unlock;
781 }
782
783 cpos = map_start >> osb->s_clustersize_bits;
784 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
785 map_start + map_len);
786 is_last = 0;
787 while (cpos < mapping_end && !is_last) {
788 u32 fe_flags;
789
790 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
791 &hole_size, &rec, &is_last);
792 if (ret) {
793 mlog_errno(ret);
794 goto out_unlock;
795 }
796
797 if (rec.e_blkno == 0ULL) {
798 cpos += hole_size;
799 continue;
800 }
801
802 fe_flags = 0;
803 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
804 fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
805 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
806 fe_flags |= FIEMAP_EXTENT_SHARED;
807 if (is_last)
808 fe_flags |= FIEMAP_EXTENT_LAST;
809 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
810 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
811 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
812
813 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
814 len_bytes, fe_flags);
815 if (ret)
816 break;
817
818 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
819 }
820
821 if (ret > 0)
822 ret = 0;
823
824out_unlock:
825 brelse(di_bh);
826
827 up_read(&OCFS2_I(inode)->ip_alloc_sem);
828
829 ocfs2_inode_unlock(inode, 0);
830out:
831
832 return ret;
833}
834
835int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
836{
837 struct inode *inode = file->f_mapping->host;
838 int ret;
839 unsigned int is_last = 0, is_data = 0;
840 u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
841 u32 cpos, cend, clen, hole_size;
842 u64 extoff, extlen;
843 struct buffer_head *di_bh = NULL;
844 struct ocfs2_extent_rec rec;
845
846 BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
847
848 ret = ocfs2_inode_lock(inode, &di_bh, 0);
849 if (ret) {
850 mlog_errno(ret);
851 goto out;
852 }
853
854 down_read(&OCFS2_I(inode)->ip_alloc_sem);
855
856 if (*offset >= i_size_read(inode)) {
857 ret = -ENXIO;
858 goto out_unlock;
859 }
860
861 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
862 if (whence == SEEK_HOLE)
863 *offset = i_size_read(inode);
864 goto out_unlock;
865 }
866
867 clen = 0;
868 cpos = *offset >> cs_bits;
869 cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
870
871 while (cpos < cend && !is_last) {
872 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
873 &rec, &is_last);
874 if (ret) {
875 mlog_errno(ret);
876 goto out_unlock;
877 }
878
879 extoff = cpos;
880 extoff <<= cs_bits;
881
882 if (rec.e_blkno == 0ULL) {
883 clen = hole_size;
884 is_data = 0;
885 } else {
886 clen = le16_to_cpu(rec.e_leaf_clusters) -
887 (cpos - le32_to_cpu(rec.e_cpos));
888 is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ? 0 : 1;
889 }
890
891 if ((!is_data && whence == SEEK_HOLE) ||
892 (is_data && whence == SEEK_DATA)) {
893 if (extoff > *offset)
894 *offset = extoff;
895 goto out_unlock;
896 }
897
898 if (!is_last)
899 cpos += clen;
900 }
901
902 if (whence == SEEK_HOLE) {
903 extoff = cpos;
904 extoff <<= cs_bits;
905 extlen = clen;
906 extlen <<= cs_bits;
907
908 if ((extoff + extlen) > i_size_read(inode))
909 extlen = i_size_read(inode) - extoff;
910 extoff += extlen;
911 if (extoff > *offset)
912 *offset = extoff;
913 goto out_unlock;
914 }
915
916 ret = -ENXIO;
917
918out_unlock:
919
920 brelse(di_bh);
921
922 up_read(&OCFS2_I(inode)->ip_alloc_sem);
923
924 ocfs2_inode_unlock(inode, 0);
925out:
926 return ret;
927}
928
929int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
930 struct buffer_head *bhs[], int flags,
931 int (*validate)(struct super_block *sb,
932 struct buffer_head *bh))
933{
934 int rc = 0;
935 u64 p_block, p_count;
936 int i, count, done = 0;
937
938 trace_ocfs2_read_virt_blocks(
939 inode, (unsigned long long)v_block, nr, bhs, flags,
940 validate);
941
942 if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
943 i_size_read(inode)) {
944 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
945 goto out;
946 }
947
948 while (done < nr) {
949 down_read(&OCFS2_I(inode)->ip_alloc_sem);
950 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
951 &p_block, &p_count, NULL);
952 up_read(&OCFS2_I(inode)->ip_alloc_sem);
953 if (rc) {
954 mlog_errno(rc);
955 break;
956 }
957
958 if (!p_block) {
959 rc = -EIO;
960 mlog(ML_ERROR,
961 "Inode #%llu contains a hole at offset %llu\n",
962 (unsigned long long)OCFS2_I(inode)->ip_blkno,
963 (unsigned long long)(v_block + done) <<
964 inode->i_sb->s_blocksize_bits);
965 break;
966 }
967
968 count = nr - done;
969 if (p_count < count)
970 count = p_count;
971
972 /*
973 * If the caller passed us bhs, they should have come
974 * from a previous readahead call to this function. Thus,
975 * they should have the right b_blocknr.
976 */
977 for (i = 0; i < count; i++) {
978 if (!bhs[done + i])
979 continue;
980 BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
981 }
982
983 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
984 bhs + done, flags, validate);
985 if (rc) {
986 mlog_errno(rc);
987 break;
988 }
989 done += count;
990 }
991
992out:
993 return rc;
994}
995
996
1// SPDX-License-Identifier: GPL-2.0-only
2/* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * extent_map.c
6 *
7 * Block/Cluster mapping functions
8 *
9 * Copyright (C) 2004 Oracle. All rights reserved.
10 */
11
12#include <linux/fs.h>
13#include <linux/init.h>
14#include <linux/slab.h>
15#include <linux/types.h>
16#include <linux/fiemap.h>
17
18#include <cluster/masklog.h>
19
20#include "ocfs2.h"
21
22#include "alloc.h"
23#include "dlmglue.h"
24#include "extent_map.h"
25#include "inode.h"
26#include "super.h"
27#include "symlink.h"
28#include "aops.h"
29#include "ocfs2_trace.h"
30
31#include "buffer_head_io.h"
32
33/*
34 * The extent caching implementation is intentionally trivial.
35 *
36 * We only cache a small number of extents stored directly on the
37 * inode, so linear order operations are acceptable. If we ever want
38 * to increase the size of the extent map, then these algorithms must
39 * get smarter.
40 */
41
42void ocfs2_extent_map_init(struct inode *inode)
43{
44 struct ocfs2_inode_info *oi = OCFS2_I(inode);
45
46 oi->ip_extent_map.em_num_items = 0;
47 INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
48}
49
50static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
51 unsigned int cpos,
52 struct ocfs2_extent_map_item **ret_emi)
53{
54 unsigned int range;
55 struct ocfs2_extent_map_item *emi;
56
57 *ret_emi = NULL;
58
59 list_for_each_entry(emi, &em->em_list, ei_list) {
60 range = emi->ei_cpos + emi->ei_clusters;
61
62 if (cpos >= emi->ei_cpos && cpos < range) {
63 list_move(&emi->ei_list, &em->em_list);
64
65 *ret_emi = emi;
66 break;
67 }
68 }
69}
70
71static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
72 unsigned int *phys, unsigned int *len,
73 unsigned int *flags)
74{
75 unsigned int coff;
76 struct ocfs2_inode_info *oi = OCFS2_I(inode);
77 struct ocfs2_extent_map_item *emi;
78
79 spin_lock(&oi->ip_lock);
80
81 __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
82 if (emi) {
83 coff = cpos - emi->ei_cpos;
84 *phys = emi->ei_phys + coff;
85 if (len)
86 *len = emi->ei_clusters - coff;
87 if (flags)
88 *flags = emi->ei_flags;
89 }
90
91 spin_unlock(&oi->ip_lock);
92
93 if (emi == NULL)
94 return -ENOENT;
95
96 return 0;
97}
98
99/*
100 * Forget about all clusters equal to or greater than cpos.
101 */
102void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
103{
104 struct ocfs2_extent_map_item *emi, *n;
105 struct ocfs2_inode_info *oi = OCFS2_I(inode);
106 struct ocfs2_extent_map *em = &oi->ip_extent_map;
107 LIST_HEAD(tmp_list);
108 unsigned int range;
109
110 spin_lock(&oi->ip_lock);
111 list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
112 if (emi->ei_cpos >= cpos) {
113 /* Full truncate of this record. */
114 list_move(&emi->ei_list, &tmp_list);
115 BUG_ON(em->em_num_items == 0);
116 em->em_num_items--;
117 continue;
118 }
119
120 range = emi->ei_cpos + emi->ei_clusters;
121 if (range > cpos) {
122 /* Partial truncate */
123 emi->ei_clusters = cpos - emi->ei_cpos;
124 }
125 }
126 spin_unlock(&oi->ip_lock);
127
128 list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
129 list_del(&emi->ei_list);
130 kfree(emi);
131 }
132}
133
134/*
135 * Is any part of emi2 contained within emi1
136 */
137static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
138 struct ocfs2_extent_map_item *emi2)
139{
140 unsigned int range1, range2;
141
142 /*
143 * Check if logical start of emi2 is inside emi1
144 */
145 range1 = emi1->ei_cpos + emi1->ei_clusters;
146 if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
147 return 1;
148
149 /*
150 * Check if logical end of emi2 is inside emi1
151 */
152 range2 = emi2->ei_cpos + emi2->ei_clusters;
153 if (range2 > emi1->ei_cpos && range2 <= range1)
154 return 1;
155
156 return 0;
157}
158
159static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
160 struct ocfs2_extent_map_item *src)
161{
162 dest->ei_cpos = src->ei_cpos;
163 dest->ei_phys = src->ei_phys;
164 dest->ei_clusters = src->ei_clusters;
165 dest->ei_flags = src->ei_flags;
166}
167
168/*
169 * Try to merge emi with ins. Returns 1 if merge succeeds, zero
170 * otherwise.
171 */
172static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
173 struct ocfs2_extent_map_item *ins)
174{
175 /*
176 * Handle contiguousness
177 */
178 if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
179 ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
180 ins->ei_flags == emi->ei_flags) {
181 emi->ei_clusters += ins->ei_clusters;
182 return 1;
183 } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
184 (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
185 ins->ei_flags == emi->ei_flags) {
186 emi->ei_phys = ins->ei_phys;
187 emi->ei_cpos = ins->ei_cpos;
188 emi->ei_clusters += ins->ei_clusters;
189 return 1;
190 }
191
192 /*
193 * Overlapping extents - this shouldn't happen unless we've
194 * split an extent to change it's flags. That is exceedingly
195 * rare, so there's no sense in trying to optimize it yet.
196 */
197 if (ocfs2_ei_is_contained(emi, ins) ||
198 ocfs2_ei_is_contained(ins, emi)) {
199 ocfs2_copy_emi_fields(emi, ins);
200 return 1;
201 }
202
203 /* No merge was possible. */
204 return 0;
205}
206
207/*
208 * In order to reduce complexity on the caller, this insert function
209 * is intentionally liberal in what it will accept.
210 *
211 * The only rule is that the truncate call *must* be used whenever
212 * records have been deleted. This avoids inserting overlapping
213 * records with different physical mappings.
214 */
215void ocfs2_extent_map_insert_rec(struct inode *inode,
216 struct ocfs2_extent_rec *rec)
217{
218 struct ocfs2_inode_info *oi = OCFS2_I(inode);
219 struct ocfs2_extent_map *em = &oi->ip_extent_map;
220 struct ocfs2_extent_map_item *emi, *new_emi = NULL;
221 struct ocfs2_extent_map_item ins;
222
223 ins.ei_cpos = le32_to_cpu(rec->e_cpos);
224 ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
225 le64_to_cpu(rec->e_blkno));
226 ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
227 ins.ei_flags = rec->e_flags;
228
229search:
230 spin_lock(&oi->ip_lock);
231
232 list_for_each_entry(emi, &em->em_list, ei_list) {
233 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
234 list_move(&emi->ei_list, &em->em_list);
235 spin_unlock(&oi->ip_lock);
236 goto out;
237 }
238 }
239
240 /*
241 * No item could be merged.
242 *
243 * Either allocate and add a new item, or overwrite the last recently
244 * inserted.
245 */
246
247 if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
248 if (new_emi == NULL) {
249 spin_unlock(&oi->ip_lock);
250
251 new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
252 if (new_emi == NULL)
253 goto out;
254
255 goto search;
256 }
257
258 ocfs2_copy_emi_fields(new_emi, &ins);
259 list_add(&new_emi->ei_list, &em->em_list);
260 em->em_num_items++;
261 new_emi = NULL;
262 } else {
263 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
264 emi = list_entry(em->em_list.prev,
265 struct ocfs2_extent_map_item, ei_list);
266 list_move(&emi->ei_list, &em->em_list);
267 ocfs2_copy_emi_fields(emi, &ins);
268 }
269
270 spin_unlock(&oi->ip_lock);
271
272out:
273 kfree(new_emi);
274}
275
276static int ocfs2_last_eb_is_empty(struct inode *inode,
277 struct ocfs2_dinode *di)
278{
279 int ret, next_free;
280 u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
281 struct buffer_head *eb_bh = NULL;
282 struct ocfs2_extent_block *eb;
283 struct ocfs2_extent_list *el;
284
285 ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
286 if (ret) {
287 mlog_errno(ret);
288 goto out;
289 }
290
291 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
292 el = &eb->h_list;
293
294 if (el->l_tree_depth) {
295 ocfs2_error(inode->i_sb,
296 "Inode %lu has non zero tree depth in leaf block %llu\n",
297 inode->i_ino,
298 (unsigned long long)eb_bh->b_blocknr);
299 ret = -EROFS;
300 goto out;
301 }
302
303 next_free = le16_to_cpu(el->l_next_free_rec);
304
305 if (next_free == 0 ||
306 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
307 ret = 1;
308
309out:
310 brelse(eb_bh);
311 return ret;
312}
313
314/*
315 * Return the 1st index within el which contains an extent start
316 * larger than v_cluster.
317 */
318static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
319 u32 v_cluster)
320{
321 int i;
322 struct ocfs2_extent_rec *rec;
323
324 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
325 rec = &el->l_recs[i];
326
327 if (v_cluster < le32_to_cpu(rec->e_cpos))
328 break;
329 }
330
331 return i;
332}
333
334/*
335 * Figure out the size of a hole which starts at v_cluster within the given
336 * extent list.
337 *
338 * If there is no more allocation past v_cluster, we return the maximum
339 * cluster size minus v_cluster.
340 *
341 * If we have in-inode extents, then el points to the dinode list and
342 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
343 * containing el.
344 */
345int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
346 struct ocfs2_extent_list *el,
347 struct buffer_head *eb_bh,
348 u32 v_cluster,
349 u32 *num_clusters)
350{
351 int ret, i;
352 struct buffer_head *next_eb_bh = NULL;
353 struct ocfs2_extent_block *eb, *next_eb;
354
355 i = ocfs2_search_for_hole_index(el, v_cluster);
356
357 if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
358 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
359
360 /*
361 * Check the next leaf for any extents.
362 */
363
364 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
365 goto no_more_extents;
366
367 ret = ocfs2_read_extent_block(ci,
368 le64_to_cpu(eb->h_next_leaf_blk),
369 &next_eb_bh);
370 if (ret) {
371 mlog_errno(ret);
372 goto out;
373 }
374
375 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
376 el = &next_eb->h_list;
377 i = ocfs2_search_for_hole_index(el, v_cluster);
378 }
379
380no_more_extents:
381 if (i == le16_to_cpu(el->l_next_free_rec)) {
382 /*
383 * We're at the end of our existing allocation. Just
384 * return the maximum number of clusters we could
385 * possibly allocate.
386 */
387 *num_clusters = UINT_MAX - v_cluster;
388 } else {
389 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
390 }
391
392 ret = 0;
393out:
394 brelse(next_eb_bh);
395 return ret;
396}
397
398static int ocfs2_get_clusters_nocache(struct inode *inode,
399 struct buffer_head *di_bh,
400 u32 v_cluster, unsigned int *hole_len,
401 struct ocfs2_extent_rec *ret_rec,
402 unsigned int *is_last)
403{
404 int i, ret, tree_height, len;
405 struct ocfs2_dinode *di;
406 struct ocfs2_extent_block *uninitialized_var(eb);
407 struct ocfs2_extent_list *el;
408 struct ocfs2_extent_rec *rec;
409 struct buffer_head *eb_bh = NULL;
410
411 memset(ret_rec, 0, sizeof(*ret_rec));
412 if (is_last)
413 *is_last = 0;
414
415 di = (struct ocfs2_dinode *) di_bh->b_data;
416 el = &di->id2.i_list;
417 tree_height = le16_to_cpu(el->l_tree_depth);
418
419 if (tree_height > 0) {
420 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
421 &eb_bh);
422 if (ret) {
423 mlog_errno(ret);
424 goto out;
425 }
426
427 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
428 el = &eb->h_list;
429
430 if (el->l_tree_depth) {
431 ocfs2_error(inode->i_sb,
432 "Inode %lu has non zero tree depth in leaf block %llu\n",
433 inode->i_ino,
434 (unsigned long long)eb_bh->b_blocknr);
435 ret = -EROFS;
436 goto out;
437 }
438 }
439
440 i = ocfs2_search_extent_list(el, v_cluster);
441 if (i == -1) {
442 /*
443 * Holes can be larger than the maximum size of an
444 * extent, so we return their lengths in a separate
445 * field.
446 */
447 if (hole_len) {
448 ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
449 el, eb_bh,
450 v_cluster, &len);
451 if (ret) {
452 mlog_errno(ret);
453 goto out;
454 }
455
456 *hole_len = len;
457 }
458 goto out_hole;
459 }
460
461 rec = &el->l_recs[i];
462
463 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
464
465 if (!rec->e_blkno) {
466 ocfs2_error(inode->i_sb,
467 "Inode %lu has bad extent record (%u, %u, 0)\n",
468 inode->i_ino,
469 le32_to_cpu(rec->e_cpos),
470 ocfs2_rec_clusters(el, rec));
471 ret = -EROFS;
472 goto out;
473 }
474
475 *ret_rec = *rec;
476
477 /*
478 * Checking for last extent is potentially expensive - we
479 * might have to look at the next leaf over to see if it's
480 * empty.
481 *
482 * The first two checks are to see whether the caller even
483 * cares for this information, and if the extent is at least
484 * the last in it's list.
485 *
486 * If those hold true, then the extent is last if any of the
487 * additional conditions hold true:
488 * - Extent list is in-inode
489 * - Extent list is right-most
490 * - Extent list is 2nd to rightmost, with empty right-most
491 */
492 if (is_last) {
493 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
494 if (tree_height == 0)
495 *is_last = 1;
496 else if (eb->h_blkno == di->i_last_eb_blk)
497 *is_last = 1;
498 else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
499 ret = ocfs2_last_eb_is_empty(inode, di);
500 if (ret < 0) {
501 mlog_errno(ret);
502 goto out;
503 }
504 if (ret == 1)
505 *is_last = 1;
506 }
507 }
508 }
509
510out_hole:
511 ret = 0;
512out:
513 brelse(eb_bh);
514 return ret;
515}
516
517static void ocfs2_relative_extent_offsets(struct super_block *sb,
518 u32 v_cluster,
519 struct ocfs2_extent_rec *rec,
520 u32 *p_cluster, u32 *num_clusters)
521
522{
523 u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
524
525 *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
526 *p_cluster = *p_cluster + coff;
527
528 if (num_clusters)
529 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
530}
531
532int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
533 u32 *p_cluster, u32 *num_clusters,
534 struct ocfs2_extent_list *el,
535 unsigned int *extent_flags)
536{
537 int ret = 0, i;
538 struct buffer_head *eb_bh = NULL;
539 struct ocfs2_extent_block *eb;
540 struct ocfs2_extent_rec *rec;
541 u32 coff;
542
543 if (el->l_tree_depth) {
544 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
545 &eb_bh);
546 if (ret) {
547 mlog_errno(ret);
548 goto out;
549 }
550
551 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
552 el = &eb->h_list;
553
554 if (el->l_tree_depth) {
555 ocfs2_error(inode->i_sb,
556 "Inode %lu has non zero tree depth in xattr leaf block %llu\n",
557 inode->i_ino,
558 (unsigned long long)eb_bh->b_blocknr);
559 ret = -EROFS;
560 goto out;
561 }
562 }
563
564 i = ocfs2_search_extent_list(el, v_cluster);
565 if (i == -1) {
566 ret = -EROFS;
567 mlog_errno(ret);
568 goto out;
569 } else {
570 rec = &el->l_recs[i];
571 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
572
573 if (!rec->e_blkno) {
574 ocfs2_error(inode->i_sb,
575 "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
576 inode->i_ino,
577 le32_to_cpu(rec->e_cpos),
578 ocfs2_rec_clusters(el, rec));
579 ret = -EROFS;
580 goto out;
581 }
582 coff = v_cluster - le32_to_cpu(rec->e_cpos);
583 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
584 le64_to_cpu(rec->e_blkno));
585 *p_cluster = *p_cluster + coff;
586 if (num_clusters)
587 *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
588
589 if (extent_flags)
590 *extent_flags = rec->e_flags;
591 }
592out:
593 brelse(eb_bh);
594 return ret;
595}
596
597int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
598 u32 *p_cluster, u32 *num_clusters,
599 unsigned int *extent_flags)
600{
601 int ret;
602 unsigned int uninitialized_var(hole_len), flags = 0;
603 struct buffer_head *di_bh = NULL;
604 struct ocfs2_extent_rec rec;
605
606 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
607 ret = -ERANGE;
608 mlog_errno(ret);
609 goto out;
610 }
611
612 ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
613 num_clusters, extent_flags);
614 if (ret == 0)
615 goto out;
616
617 ret = ocfs2_read_inode_block(inode, &di_bh);
618 if (ret) {
619 mlog_errno(ret);
620 goto out;
621 }
622
623 ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
624 &rec, NULL);
625 if (ret) {
626 mlog_errno(ret);
627 goto out;
628 }
629
630 if (rec.e_blkno == 0ULL) {
631 /*
632 * A hole was found. Return some canned values that
633 * callers can key on. If asked for, num_clusters will
634 * be populated with the size of the hole.
635 */
636 *p_cluster = 0;
637 if (num_clusters) {
638 *num_clusters = hole_len;
639 }
640 } else {
641 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
642 p_cluster, num_clusters);
643 flags = rec.e_flags;
644
645 ocfs2_extent_map_insert_rec(inode, &rec);
646 }
647
648 if (extent_flags)
649 *extent_flags = flags;
650
651out:
652 brelse(di_bh);
653 return ret;
654}
655
656/*
657 * This expects alloc_sem to be held. The allocation cannot change at
658 * all while the map is in the process of being updated.
659 */
660int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
661 u64 *ret_count, unsigned int *extent_flags)
662{
663 int ret;
664 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
665 u32 cpos, num_clusters, p_cluster;
666 u64 boff = 0;
667
668 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
669
670 ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
671 extent_flags);
672 if (ret) {
673 mlog_errno(ret);
674 goto out;
675 }
676
677 /*
678 * p_cluster == 0 indicates a hole.
679 */
680 if (p_cluster) {
681 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
682 boff += (v_blkno & (u64)(bpc - 1));
683 }
684
685 *p_blkno = boff;
686
687 if (ret_count) {
688 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
689 *ret_count -= v_blkno & (u64)(bpc - 1);
690 }
691
692out:
693 return ret;
694}
695
696/*
697 * The ocfs2_fiemap_inline() may be a little bit misleading, since
698 * it not only handles the fiemap for inlined files, but also deals
699 * with the fast symlink, cause they have no difference for extent
700 * mapping per se.
701 */
702static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
703 struct fiemap_extent_info *fieinfo,
704 u64 map_start)
705{
706 int ret;
707 unsigned int id_count;
708 struct ocfs2_dinode *di;
709 u64 phys;
710 u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
711 struct ocfs2_inode_info *oi = OCFS2_I(inode);
712
713 di = (struct ocfs2_dinode *)di_bh->b_data;
714 if (ocfs2_inode_is_fast_symlink(inode))
715 id_count = ocfs2_fast_symlink_chars(inode->i_sb);
716 else
717 id_count = le16_to_cpu(di->id2.i_data.id_count);
718
719 if (map_start < id_count) {
720 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
721 if (ocfs2_inode_is_fast_symlink(inode))
722 phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
723 else
724 phys += offsetof(struct ocfs2_dinode,
725 id2.i_data.id_data);
726
727 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
728 flags);
729 if (ret < 0)
730 return ret;
731 }
732
733 return 0;
734}
735
736#define OCFS2_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
737
738int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
739 u64 map_start, u64 map_len)
740{
741 int ret, is_last;
742 u32 mapping_end, cpos;
743 unsigned int hole_size;
744 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
745 u64 len_bytes, phys_bytes, virt_bytes;
746 struct buffer_head *di_bh = NULL;
747 struct ocfs2_extent_rec rec;
748
749 ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
750 if (ret)
751 return ret;
752
753 ret = ocfs2_inode_lock(inode, &di_bh, 0);
754 if (ret) {
755 mlog_errno(ret);
756 goto out;
757 }
758
759 down_read(&OCFS2_I(inode)->ip_alloc_sem);
760
761 /*
762 * Handle inline-data and fast symlink separately.
763 */
764 if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
765 ocfs2_inode_is_fast_symlink(inode)) {
766 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
767 goto out_unlock;
768 }
769
770 cpos = map_start >> osb->s_clustersize_bits;
771 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
772 map_start + map_len);
773 is_last = 0;
774 while (cpos < mapping_end && !is_last) {
775 u32 fe_flags;
776
777 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
778 &hole_size, &rec, &is_last);
779 if (ret) {
780 mlog_errno(ret);
781 goto out_unlock;
782 }
783
784 if (rec.e_blkno == 0ULL) {
785 cpos += hole_size;
786 continue;
787 }
788
789 fe_flags = 0;
790 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
791 fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
792 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
793 fe_flags |= FIEMAP_EXTENT_SHARED;
794 if (is_last)
795 fe_flags |= FIEMAP_EXTENT_LAST;
796 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
797 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
798 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
799
800 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
801 len_bytes, fe_flags);
802 if (ret)
803 break;
804
805 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
806 }
807
808 if (ret > 0)
809 ret = 0;
810
811out_unlock:
812 brelse(di_bh);
813
814 up_read(&OCFS2_I(inode)->ip_alloc_sem);
815
816 ocfs2_inode_unlock(inode, 0);
817out:
818
819 return ret;
820}
821
822/* Is IO overwriting allocated blocks? */
823int ocfs2_overwrite_io(struct inode *inode, struct buffer_head *di_bh,
824 u64 map_start, u64 map_len)
825{
826 int ret = 0, is_last;
827 u32 mapping_end, cpos;
828 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
829 struct ocfs2_extent_rec rec;
830
831 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
832 if (ocfs2_size_fits_inline_data(di_bh, map_start + map_len))
833 return ret;
834 else
835 return -EAGAIN;
836 }
837
838 cpos = map_start >> osb->s_clustersize_bits;
839 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
840 map_start + map_len);
841 is_last = 0;
842 while (cpos < mapping_end && !is_last) {
843 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
844 NULL, &rec, &is_last);
845 if (ret) {
846 mlog_errno(ret);
847 goto out;
848 }
849
850 if (rec.e_blkno == 0ULL)
851 break;
852
853 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
854 break;
855
856 cpos = le32_to_cpu(rec.e_cpos) +
857 le16_to_cpu(rec.e_leaf_clusters);
858 }
859
860 if (cpos < mapping_end)
861 ret = -EAGAIN;
862out:
863 return ret;
864}
865
866int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
867{
868 struct inode *inode = file->f_mapping->host;
869 int ret;
870 unsigned int is_last = 0, is_data = 0;
871 u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
872 u32 cpos, cend, clen, hole_size;
873 u64 extoff, extlen;
874 struct buffer_head *di_bh = NULL;
875 struct ocfs2_extent_rec rec;
876
877 BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
878
879 ret = ocfs2_inode_lock(inode, &di_bh, 0);
880 if (ret) {
881 mlog_errno(ret);
882 goto out;
883 }
884
885 down_read(&OCFS2_I(inode)->ip_alloc_sem);
886
887 if (*offset >= i_size_read(inode)) {
888 ret = -ENXIO;
889 goto out_unlock;
890 }
891
892 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
893 if (whence == SEEK_HOLE)
894 *offset = i_size_read(inode);
895 goto out_unlock;
896 }
897
898 clen = 0;
899 cpos = *offset >> cs_bits;
900 cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
901
902 while (cpos < cend && !is_last) {
903 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
904 &rec, &is_last);
905 if (ret) {
906 mlog_errno(ret);
907 goto out_unlock;
908 }
909
910 extoff = cpos;
911 extoff <<= cs_bits;
912
913 if (rec.e_blkno == 0ULL) {
914 clen = hole_size;
915 is_data = 0;
916 } else {
917 clen = le16_to_cpu(rec.e_leaf_clusters) -
918 (cpos - le32_to_cpu(rec.e_cpos));
919 is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ? 0 : 1;
920 }
921
922 if ((!is_data && whence == SEEK_HOLE) ||
923 (is_data && whence == SEEK_DATA)) {
924 if (extoff > *offset)
925 *offset = extoff;
926 goto out_unlock;
927 }
928
929 if (!is_last)
930 cpos += clen;
931 }
932
933 if (whence == SEEK_HOLE) {
934 extoff = cpos;
935 extoff <<= cs_bits;
936 extlen = clen;
937 extlen <<= cs_bits;
938
939 if ((extoff + extlen) > i_size_read(inode))
940 extlen = i_size_read(inode) - extoff;
941 extoff += extlen;
942 if (extoff > *offset)
943 *offset = extoff;
944 goto out_unlock;
945 }
946
947 ret = -ENXIO;
948
949out_unlock:
950
951 brelse(di_bh);
952
953 up_read(&OCFS2_I(inode)->ip_alloc_sem);
954
955 ocfs2_inode_unlock(inode, 0);
956out:
957 return ret;
958}
959
960int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
961 struct buffer_head *bhs[], int flags,
962 int (*validate)(struct super_block *sb,
963 struct buffer_head *bh))
964{
965 int rc = 0;
966 u64 p_block, p_count;
967 int i, count, done = 0;
968
969 trace_ocfs2_read_virt_blocks(
970 inode, (unsigned long long)v_block, nr, bhs, flags,
971 validate);
972
973 if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
974 i_size_read(inode)) {
975 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
976 goto out;
977 }
978
979 while (done < nr) {
980 down_read(&OCFS2_I(inode)->ip_alloc_sem);
981 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
982 &p_block, &p_count, NULL);
983 up_read(&OCFS2_I(inode)->ip_alloc_sem);
984 if (rc) {
985 mlog_errno(rc);
986 break;
987 }
988
989 if (!p_block) {
990 rc = -EIO;
991 mlog(ML_ERROR,
992 "Inode #%llu contains a hole at offset %llu\n",
993 (unsigned long long)OCFS2_I(inode)->ip_blkno,
994 (unsigned long long)(v_block + done) <<
995 inode->i_sb->s_blocksize_bits);
996 break;
997 }
998
999 count = nr - done;
1000 if (p_count < count)
1001 count = p_count;
1002
1003 /*
1004 * If the caller passed us bhs, they should have come
1005 * from a previous readahead call to this function. Thus,
1006 * they should have the right b_blocknr.
1007 */
1008 for (i = 0; i < count; i++) {
1009 if (!bhs[done + i])
1010 continue;
1011 BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
1012 }
1013
1014 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
1015 bhs + done, flags, validate);
1016 if (rc) {
1017 mlog_errno(rc);
1018 break;
1019 }
1020 done += count;
1021 }
1022
1023out:
1024 return rc;
1025}
1026
1027