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/*
3 * extent_map.c
4 *
5 * Block/Cluster mapping functions
6 *
7 * Copyright (C) 2004 Oracle. All rights reserved.
8 */
9
10#include <linux/fs.h>
11#include <linux/init.h>
12#include <linux/slab.h>
13#include <linux/types.h>
14#include <linux/fiemap.h>
15
16#include <cluster/masklog.h>
17
18#include "ocfs2.h"
19
20#include "alloc.h"
21#include "dlmglue.h"
22#include "extent_map.h"
23#include "inode.h"
24#include "super.h"
25#include "symlink.h"
26#include "aops.h"
27#include "ocfs2_trace.h"
28
29#include "buffer_head_io.h"
30
31/*
32 * The extent caching implementation is intentionally trivial.
33 *
34 * We only cache a small number of extents stored directly on the
35 * inode, so linear order operations are acceptable. If we ever want
36 * to increase the size of the extent map, then these algorithms must
37 * get smarter.
38 */
39
40void ocfs2_extent_map_init(struct inode *inode)
41{
42 struct ocfs2_inode_info *oi = OCFS2_I(inode);
43
44 oi->ip_extent_map.em_num_items = 0;
45 INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
46}
47
48static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
49 unsigned int cpos,
50 struct ocfs2_extent_map_item **ret_emi)
51{
52 unsigned int range;
53 struct ocfs2_extent_map_item *emi;
54
55 *ret_emi = NULL;
56
57 list_for_each_entry(emi, &em->em_list, ei_list) {
58 range = emi->ei_cpos + emi->ei_clusters;
59
60 if (cpos >= emi->ei_cpos && cpos < range) {
61 list_move(&emi->ei_list, &em->em_list);
62
63 *ret_emi = emi;
64 break;
65 }
66 }
67}
68
69static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
70 unsigned int *phys, unsigned int *len,
71 unsigned int *flags)
72{
73 unsigned int coff;
74 struct ocfs2_inode_info *oi = OCFS2_I(inode);
75 struct ocfs2_extent_map_item *emi;
76
77 spin_lock(&oi->ip_lock);
78
79 __ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
80 if (emi) {
81 coff = cpos - emi->ei_cpos;
82 *phys = emi->ei_phys + coff;
83 if (len)
84 *len = emi->ei_clusters - coff;
85 if (flags)
86 *flags = emi->ei_flags;
87 }
88
89 spin_unlock(&oi->ip_lock);
90
91 if (emi == NULL)
92 return -ENOENT;
93
94 return 0;
95}
96
97/*
98 * Forget about all clusters equal to or greater than cpos.
99 */
100void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
101{
102 struct ocfs2_extent_map_item *emi, *n;
103 struct ocfs2_inode_info *oi = OCFS2_I(inode);
104 struct ocfs2_extent_map *em = &oi->ip_extent_map;
105 LIST_HEAD(tmp_list);
106 unsigned int range;
107
108 spin_lock(&oi->ip_lock);
109 list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
110 if (emi->ei_cpos >= cpos) {
111 /* Full truncate of this record. */
112 list_move(&emi->ei_list, &tmp_list);
113 BUG_ON(em->em_num_items == 0);
114 em->em_num_items--;
115 continue;
116 }
117
118 range = emi->ei_cpos + emi->ei_clusters;
119 if (range > cpos) {
120 /* Partial truncate */
121 emi->ei_clusters = cpos - emi->ei_cpos;
122 }
123 }
124 spin_unlock(&oi->ip_lock);
125
126 list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
127 list_del(&emi->ei_list);
128 kfree(emi);
129 }
130}
131
132/*
133 * Is any part of emi2 contained within emi1
134 */
135static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
136 struct ocfs2_extent_map_item *emi2)
137{
138 unsigned int range1, range2;
139
140 /*
141 * Check if logical start of emi2 is inside emi1
142 */
143 range1 = emi1->ei_cpos + emi1->ei_clusters;
144 if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
145 return 1;
146
147 /*
148 * Check if logical end of emi2 is inside emi1
149 */
150 range2 = emi2->ei_cpos + emi2->ei_clusters;
151 if (range2 > emi1->ei_cpos && range2 <= range1)
152 return 1;
153
154 return 0;
155}
156
157static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
158 struct ocfs2_extent_map_item *src)
159{
160 dest->ei_cpos = src->ei_cpos;
161 dest->ei_phys = src->ei_phys;
162 dest->ei_clusters = src->ei_clusters;
163 dest->ei_flags = src->ei_flags;
164}
165
166/*
167 * Try to merge emi with ins. Returns 1 if merge succeeds, zero
168 * otherwise.
169 */
170static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
171 struct ocfs2_extent_map_item *ins)
172{
173 /*
174 * Handle contiguousness
175 */
176 if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
177 ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
178 ins->ei_flags == emi->ei_flags) {
179 emi->ei_clusters += ins->ei_clusters;
180 return 1;
181 } else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
182 (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
183 ins->ei_flags == emi->ei_flags) {
184 emi->ei_phys = ins->ei_phys;
185 emi->ei_cpos = ins->ei_cpos;
186 emi->ei_clusters += ins->ei_clusters;
187 return 1;
188 }
189
190 /*
191 * Overlapping extents - this shouldn't happen unless we've
192 * split an extent to change it's flags. That is exceedingly
193 * rare, so there's no sense in trying to optimize it yet.
194 */
195 if (ocfs2_ei_is_contained(emi, ins) ||
196 ocfs2_ei_is_contained(ins, emi)) {
197 ocfs2_copy_emi_fields(emi, ins);
198 return 1;
199 }
200
201 /* No merge was possible. */
202 return 0;
203}
204
205/*
206 * In order to reduce complexity on the caller, this insert function
207 * is intentionally liberal in what it will accept.
208 *
209 * The only rule is that the truncate call *must* be used whenever
210 * records have been deleted. This avoids inserting overlapping
211 * records with different physical mappings.
212 */
213void ocfs2_extent_map_insert_rec(struct inode *inode,
214 struct ocfs2_extent_rec *rec)
215{
216 struct ocfs2_inode_info *oi = OCFS2_I(inode);
217 struct ocfs2_extent_map *em = &oi->ip_extent_map;
218 struct ocfs2_extent_map_item *emi, *new_emi = NULL;
219 struct ocfs2_extent_map_item ins;
220
221 ins.ei_cpos = le32_to_cpu(rec->e_cpos);
222 ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
223 le64_to_cpu(rec->e_blkno));
224 ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
225 ins.ei_flags = rec->e_flags;
226
227search:
228 spin_lock(&oi->ip_lock);
229
230 list_for_each_entry(emi, &em->em_list, ei_list) {
231 if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
232 list_move(&emi->ei_list, &em->em_list);
233 spin_unlock(&oi->ip_lock);
234 goto out;
235 }
236 }
237
238 /*
239 * No item could be merged.
240 *
241 * Either allocate and add a new item, or overwrite the last recently
242 * inserted.
243 */
244
245 if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
246 if (new_emi == NULL) {
247 spin_unlock(&oi->ip_lock);
248
249 new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
250 if (new_emi == NULL)
251 goto out;
252
253 goto search;
254 }
255
256 ocfs2_copy_emi_fields(new_emi, &ins);
257 list_add(&new_emi->ei_list, &em->em_list);
258 em->em_num_items++;
259 new_emi = NULL;
260 } else {
261 BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
262 emi = list_entry(em->em_list.prev,
263 struct ocfs2_extent_map_item, ei_list);
264 list_move(&emi->ei_list, &em->em_list);
265 ocfs2_copy_emi_fields(emi, &ins);
266 }
267
268 spin_unlock(&oi->ip_lock);
269
270out:
271 kfree(new_emi);
272}
273
274static int ocfs2_last_eb_is_empty(struct inode *inode,
275 struct ocfs2_dinode *di)
276{
277 int ret, next_free;
278 u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
279 struct buffer_head *eb_bh = NULL;
280 struct ocfs2_extent_block *eb;
281 struct ocfs2_extent_list *el;
282
283 ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
284 if (ret) {
285 mlog_errno(ret);
286 goto out;
287 }
288
289 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
290 el = &eb->h_list;
291
292 if (el->l_tree_depth) {
293 ocfs2_error(inode->i_sb,
294 "Inode %lu has non zero tree depth in leaf block %llu\n",
295 inode->i_ino,
296 (unsigned long long)eb_bh->b_blocknr);
297 ret = -EROFS;
298 goto out;
299 }
300
301 next_free = le16_to_cpu(el->l_next_free_rec);
302
303 if (next_free == 0 ||
304 (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
305 ret = 1;
306
307out:
308 brelse(eb_bh);
309 return ret;
310}
311
312/*
313 * Return the 1st index within el which contains an extent start
314 * larger than v_cluster.
315 */
316static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
317 u32 v_cluster)
318{
319 int i;
320 struct ocfs2_extent_rec *rec;
321
322 for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
323 rec = &el->l_recs[i];
324
325 if (v_cluster < le32_to_cpu(rec->e_cpos))
326 break;
327 }
328
329 return i;
330}
331
332/*
333 * Figure out the size of a hole which starts at v_cluster within the given
334 * extent list.
335 *
336 * If there is no more allocation past v_cluster, we return the maximum
337 * cluster size minus v_cluster.
338 *
339 * If we have in-inode extents, then el points to the dinode list and
340 * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
341 * containing el.
342 */
343int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
344 struct ocfs2_extent_list *el,
345 struct buffer_head *eb_bh,
346 u32 v_cluster,
347 u32 *num_clusters)
348{
349 int ret, i;
350 struct buffer_head *next_eb_bh = NULL;
351 struct ocfs2_extent_block *eb, *next_eb;
352
353 i = ocfs2_search_for_hole_index(el, v_cluster);
354
355 if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
356 eb = (struct ocfs2_extent_block *)eb_bh->b_data;
357
358 /*
359 * Check the next leaf for any extents.
360 */
361
362 if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
363 goto no_more_extents;
364
365 ret = ocfs2_read_extent_block(ci,
366 le64_to_cpu(eb->h_next_leaf_blk),
367 &next_eb_bh);
368 if (ret) {
369 mlog_errno(ret);
370 goto out;
371 }
372
373 next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
374 el = &next_eb->h_list;
375 i = ocfs2_search_for_hole_index(el, v_cluster);
376 }
377
378no_more_extents:
379 if (i == le16_to_cpu(el->l_next_free_rec)) {
380 /*
381 * We're at the end of our existing allocation. Just
382 * return the maximum number of clusters we could
383 * possibly allocate.
384 */
385 *num_clusters = UINT_MAX - v_cluster;
386 } else {
387 *num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
388 }
389
390 ret = 0;
391out:
392 brelse(next_eb_bh);
393 return ret;
394}
395
396static int ocfs2_get_clusters_nocache(struct inode *inode,
397 struct buffer_head *di_bh,
398 u32 v_cluster, unsigned int *hole_len,
399 struct ocfs2_extent_rec *ret_rec,
400 unsigned int *is_last)
401{
402 int i, ret, tree_height, len;
403 struct ocfs2_dinode *di;
404 struct ocfs2_extent_block *eb;
405 struct ocfs2_extent_list *el;
406 struct ocfs2_extent_rec *rec;
407 struct buffer_head *eb_bh = NULL;
408
409 memset(ret_rec, 0, sizeof(*ret_rec));
410 if (is_last)
411 *is_last = 0;
412
413 di = (struct ocfs2_dinode *) di_bh->b_data;
414 el = &di->id2.i_list;
415 tree_height = le16_to_cpu(el->l_tree_depth);
416
417 if (tree_height > 0) {
418 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
419 &eb_bh);
420 if (ret) {
421 mlog_errno(ret);
422 goto out;
423 }
424
425 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
426 el = &eb->h_list;
427
428 if (el->l_tree_depth) {
429 ocfs2_error(inode->i_sb,
430 "Inode %lu has non zero tree depth in leaf block %llu\n",
431 inode->i_ino,
432 (unsigned long long)eb_bh->b_blocknr);
433 ret = -EROFS;
434 goto out;
435 }
436 }
437
438 i = ocfs2_search_extent_list(el, v_cluster);
439 if (i == -1) {
440 /*
441 * Holes can be larger than the maximum size of an
442 * extent, so we return their lengths in a separate
443 * field.
444 */
445 if (hole_len) {
446 ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
447 el, eb_bh,
448 v_cluster, &len);
449 if (ret) {
450 mlog_errno(ret);
451 goto out;
452 }
453
454 *hole_len = len;
455 }
456 goto out_hole;
457 }
458
459 rec = &el->l_recs[i];
460
461 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
462
463 if (!rec->e_blkno) {
464 ocfs2_error(inode->i_sb,
465 "Inode %lu has bad extent record (%u, %u, 0)\n",
466 inode->i_ino,
467 le32_to_cpu(rec->e_cpos),
468 ocfs2_rec_clusters(el, rec));
469 ret = -EROFS;
470 goto out;
471 }
472
473 *ret_rec = *rec;
474
475 /*
476 * Checking for last extent is potentially expensive - we
477 * might have to look at the next leaf over to see if it's
478 * empty.
479 *
480 * The first two checks are to see whether the caller even
481 * cares for this information, and if the extent is at least
482 * the last in it's list.
483 *
484 * If those hold true, then the extent is last if any of the
485 * additional conditions hold true:
486 * - Extent list is in-inode
487 * - Extent list is right-most
488 * - Extent list is 2nd to rightmost, with empty right-most
489 */
490 if (is_last) {
491 if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
492 if (tree_height == 0)
493 *is_last = 1;
494 else if (eb->h_blkno == di->i_last_eb_blk)
495 *is_last = 1;
496 else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
497 ret = ocfs2_last_eb_is_empty(inode, di);
498 if (ret < 0) {
499 mlog_errno(ret);
500 goto out;
501 }
502 if (ret == 1)
503 *is_last = 1;
504 }
505 }
506 }
507
508out_hole:
509 ret = 0;
510out:
511 brelse(eb_bh);
512 return ret;
513}
514
515static void ocfs2_relative_extent_offsets(struct super_block *sb,
516 u32 v_cluster,
517 struct ocfs2_extent_rec *rec,
518 u32 *p_cluster, u32 *num_clusters)
519
520{
521 u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
522
523 *p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
524 *p_cluster = *p_cluster + coff;
525
526 if (num_clusters)
527 *num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
528}
529
530int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
531 u32 *p_cluster, u32 *num_clusters,
532 struct ocfs2_extent_list *el,
533 unsigned int *extent_flags)
534{
535 int ret = 0, i;
536 struct buffer_head *eb_bh = NULL;
537 struct ocfs2_extent_block *eb;
538 struct ocfs2_extent_rec *rec;
539 u32 coff;
540
541 if (el->l_tree_depth) {
542 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
543 &eb_bh);
544 if (ret) {
545 mlog_errno(ret);
546 goto out;
547 }
548
549 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
550 el = &eb->h_list;
551
552 if (el->l_tree_depth) {
553 ocfs2_error(inode->i_sb,
554 "Inode %lu has non zero tree depth in xattr leaf block %llu\n",
555 inode->i_ino,
556 (unsigned long long)eb_bh->b_blocknr);
557 ret = -EROFS;
558 goto out;
559 }
560 }
561
562 i = ocfs2_search_extent_list(el, v_cluster);
563 if (i == -1) {
564 ret = -EROFS;
565 mlog_errno(ret);
566 goto out;
567 } else {
568 rec = &el->l_recs[i];
569 BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
570
571 if (!rec->e_blkno) {
572 ocfs2_error(inode->i_sb,
573 "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
574 inode->i_ino,
575 le32_to_cpu(rec->e_cpos),
576 ocfs2_rec_clusters(el, rec));
577 ret = -EROFS;
578 goto out;
579 }
580 coff = v_cluster - le32_to_cpu(rec->e_cpos);
581 *p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
582 le64_to_cpu(rec->e_blkno));
583 *p_cluster = *p_cluster + coff;
584 if (num_clusters)
585 *num_clusters = ocfs2_rec_clusters(el, rec) - coff;
586
587 if (extent_flags)
588 *extent_flags = rec->e_flags;
589 }
590out:
591 brelse(eb_bh);
592 return ret;
593}
594
595int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
596 u32 *p_cluster, u32 *num_clusters,
597 unsigned int *extent_flags)
598{
599 int ret;
600 unsigned int hole_len, flags = 0;
601 struct buffer_head *di_bh = NULL;
602 struct ocfs2_extent_rec rec;
603
604 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
605 ret = -ERANGE;
606 mlog_errno(ret);
607 goto out;
608 }
609
610 ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
611 num_clusters, extent_flags);
612 if (ret == 0)
613 goto out;
614
615 ret = ocfs2_read_inode_block(inode, &di_bh);
616 if (ret) {
617 mlog_errno(ret);
618 goto out;
619 }
620
621 ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
622 &rec, NULL);
623 if (ret) {
624 mlog_errno(ret);
625 goto out;
626 }
627
628 if (rec.e_blkno == 0ULL) {
629 /*
630 * A hole was found. Return some canned values that
631 * callers can key on. If asked for, num_clusters will
632 * be populated with the size of the hole.
633 */
634 *p_cluster = 0;
635 if (num_clusters) {
636 *num_clusters = hole_len;
637 }
638 } else {
639 ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
640 p_cluster, num_clusters);
641 flags = rec.e_flags;
642
643 ocfs2_extent_map_insert_rec(inode, &rec);
644 }
645
646 if (extent_flags)
647 *extent_flags = flags;
648
649out:
650 brelse(di_bh);
651 return ret;
652}
653
654/*
655 * This expects alloc_sem to be held. The allocation cannot change at
656 * all while the map is in the process of being updated.
657 */
658int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
659 u64 *ret_count, unsigned int *extent_flags)
660{
661 int ret;
662 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
663 u32 cpos, num_clusters, p_cluster;
664 u64 boff = 0;
665
666 cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
667
668 ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
669 extent_flags);
670 if (ret) {
671 mlog_errno(ret);
672 goto out;
673 }
674
675 /*
676 * p_cluster == 0 indicates a hole.
677 */
678 if (p_cluster) {
679 boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
680 boff += (v_blkno & (u64)(bpc - 1));
681 }
682
683 *p_blkno = boff;
684
685 if (ret_count) {
686 *ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
687 *ret_count -= v_blkno & (u64)(bpc - 1);
688 }
689
690out:
691 return ret;
692}
693
694/*
695 * The ocfs2_fiemap_inline() may be a little bit misleading, since
696 * it not only handles the fiemap for inlined files, but also deals
697 * with the fast symlink, cause they have no difference for extent
698 * mapping per se.
699 */
700static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
701 struct fiemap_extent_info *fieinfo,
702 u64 map_start)
703{
704 int ret;
705 unsigned int id_count;
706 struct ocfs2_dinode *di;
707 u64 phys;
708 u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
709 struct ocfs2_inode_info *oi = OCFS2_I(inode);
710
711 di = (struct ocfs2_dinode *)di_bh->b_data;
712 if (ocfs2_inode_is_fast_symlink(inode))
713 id_count = ocfs2_fast_symlink_chars(inode->i_sb);
714 else
715 id_count = le16_to_cpu(di->id2.i_data.id_count);
716
717 if (map_start < id_count) {
718 phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
719 if (ocfs2_inode_is_fast_symlink(inode))
720 phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
721 else
722 phys += offsetof(struct ocfs2_dinode,
723 id2.i_data.id_data);
724
725 ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
726 flags);
727 if (ret < 0)
728 return ret;
729 }
730
731 return 0;
732}
733
734int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
735 u64 map_start, u64 map_len)
736{
737 int ret, is_last;
738 u32 mapping_end, cpos;
739 unsigned int hole_size;
740 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
741 u64 len_bytes, phys_bytes, virt_bytes;
742 struct buffer_head *di_bh = NULL;
743 struct ocfs2_extent_rec rec;
744
745 ret = fiemap_prep(inode, fieinfo, map_start, &map_len, 0);
746 if (ret)
747 return ret;
748
749 ret = ocfs2_inode_lock(inode, &di_bh, 0);
750 if (ret) {
751 mlog_errno(ret);
752 goto out;
753 }
754
755 down_read(&OCFS2_I(inode)->ip_alloc_sem);
756
757 /*
758 * Handle inline-data and fast symlink separately.
759 */
760 if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
761 ocfs2_inode_is_fast_symlink(inode)) {
762 ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
763 goto out_unlock;
764 }
765
766 cpos = map_start >> osb->s_clustersize_bits;
767 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
768 map_start + map_len);
769 is_last = 0;
770 while (cpos < mapping_end && !is_last) {
771 u32 fe_flags;
772
773 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
774 &hole_size, &rec, &is_last);
775 if (ret) {
776 mlog_errno(ret);
777 goto out_unlock;
778 }
779
780 if (rec.e_blkno == 0ULL) {
781 cpos += hole_size;
782 continue;
783 }
784
785 fe_flags = 0;
786 if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
787 fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
788 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
789 fe_flags |= FIEMAP_EXTENT_SHARED;
790 if (is_last)
791 fe_flags |= FIEMAP_EXTENT_LAST;
792 len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
793 phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
794 virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
795
796 ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
797 len_bytes, fe_flags);
798 if (ret)
799 break;
800
801 cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
802 }
803
804 if (ret > 0)
805 ret = 0;
806
807out_unlock:
808 brelse(di_bh);
809
810 up_read(&OCFS2_I(inode)->ip_alloc_sem);
811
812 ocfs2_inode_unlock(inode, 0);
813out:
814
815 return ret;
816}
817
818/* Is IO overwriting allocated blocks? */
819int ocfs2_overwrite_io(struct inode *inode, struct buffer_head *di_bh,
820 u64 map_start, u64 map_len)
821{
822 int ret = 0, is_last;
823 u32 mapping_end, cpos;
824 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
825 struct ocfs2_extent_rec rec;
826
827 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
828 if (ocfs2_size_fits_inline_data(di_bh, map_start + map_len))
829 return ret;
830 else
831 return -EAGAIN;
832 }
833
834 cpos = map_start >> osb->s_clustersize_bits;
835 mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
836 map_start + map_len);
837 is_last = 0;
838 while (cpos < mapping_end && !is_last) {
839 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
840 NULL, &rec, &is_last);
841 if (ret) {
842 mlog_errno(ret);
843 goto out;
844 }
845
846 if (rec.e_blkno == 0ULL)
847 break;
848
849 if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
850 break;
851
852 cpos = le32_to_cpu(rec.e_cpos) +
853 le16_to_cpu(rec.e_leaf_clusters);
854 }
855
856 if (cpos < mapping_end)
857 ret = -EAGAIN;
858out:
859 return ret;
860}
861
862int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
863{
864 struct inode *inode = file->f_mapping->host;
865 int ret;
866 unsigned int is_last = 0, is_data = 0;
867 u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
868 u32 cpos, cend, clen, hole_size;
869 u64 extoff, extlen;
870 struct buffer_head *di_bh = NULL;
871 struct ocfs2_extent_rec rec;
872
873 BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
874
875 ret = ocfs2_inode_lock(inode, &di_bh, 0);
876 if (ret) {
877 mlog_errno(ret);
878 goto out;
879 }
880
881 down_read(&OCFS2_I(inode)->ip_alloc_sem);
882
883 if (*offset >= i_size_read(inode)) {
884 ret = -ENXIO;
885 goto out_unlock;
886 }
887
888 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
889 if (whence == SEEK_HOLE)
890 *offset = i_size_read(inode);
891 goto out_unlock;
892 }
893
894 clen = 0;
895 cpos = *offset >> cs_bits;
896 cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
897
898 while (cpos < cend && !is_last) {
899 ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
900 &rec, &is_last);
901 if (ret) {
902 mlog_errno(ret);
903 goto out_unlock;
904 }
905
906 extoff = cpos;
907 extoff <<= cs_bits;
908
909 if (rec.e_blkno == 0ULL) {
910 clen = hole_size;
911 is_data = 0;
912 } else {
913 clen = le16_to_cpu(rec.e_leaf_clusters) -
914 (cpos - le32_to_cpu(rec.e_cpos));
915 is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ? 0 : 1;
916 }
917
918 if ((!is_data && whence == SEEK_HOLE) ||
919 (is_data && whence == SEEK_DATA)) {
920 if (extoff > *offset)
921 *offset = extoff;
922 goto out_unlock;
923 }
924
925 if (!is_last)
926 cpos += clen;
927 }
928
929 if (whence == SEEK_HOLE) {
930 extoff = cpos;
931 extoff <<= cs_bits;
932 extlen = clen;
933 extlen <<= cs_bits;
934
935 if ((extoff + extlen) > i_size_read(inode))
936 extlen = i_size_read(inode) - extoff;
937 extoff += extlen;
938 if (extoff > *offset)
939 *offset = extoff;
940 goto out_unlock;
941 }
942
943 ret = -ENXIO;
944
945out_unlock:
946
947 brelse(di_bh);
948
949 up_read(&OCFS2_I(inode)->ip_alloc_sem);
950
951 ocfs2_inode_unlock(inode, 0);
952out:
953 return ret;
954}
955
956int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
957 struct buffer_head *bhs[], int flags,
958 int (*validate)(struct super_block *sb,
959 struct buffer_head *bh))
960{
961 int rc = 0;
962 u64 p_block, p_count;
963 int i, count, done = 0;
964
965 trace_ocfs2_read_virt_blocks(
966 inode, (unsigned long long)v_block, nr, bhs, flags,
967 validate);
968
969 if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
970 i_size_read(inode)) {
971 BUG_ON(!(flags & OCFS2_BH_READAHEAD));
972 goto out;
973 }
974
975 while (done < nr) {
976 if (!down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem)) {
977 rc = -EAGAIN;
978 mlog(ML_ERROR,
979 "Inode #%llu ip_alloc_sem is temporarily unavailable\n",
980 (unsigned long long)OCFS2_I(inode)->ip_blkno);
981 break;
982 }
983 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
984 &p_block, &p_count, NULL);
985 up_read(&OCFS2_I(inode)->ip_alloc_sem);
986 if (rc) {
987 mlog_errno(rc);
988 break;
989 }
990
991 if (!p_block) {
992 rc = -EIO;
993 mlog(ML_ERROR,
994 "Inode #%llu contains a hole at offset %llu\n",
995 (unsigned long long)OCFS2_I(inode)->ip_blkno,
996 (unsigned long long)(v_block + done) <<
997 inode->i_sb->s_blocksize_bits);
998 break;
999 }
1000
1001 count = nr - done;
1002 if (p_count < count)
1003 count = p_count;
1004
1005 /*
1006 * If the caller passed us bhs, they should have come
1007 * from a previous readahead call to this function. Thus,
1008 * they should have the right b_blocknr.
1009 */
1010 for (i = 0; i < count; i++) {
1011 if (!bhs[done + i])
1012 continue;
1013 BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
1014 }
1015
1016 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
1017 bhs + done, flags, validate);
1018 if (rc) {
1019 mlog_errno(rc);
1020 break;
1021 }
1022 done += count;
1023 }
1024
1025out:
1026 return rc;
1027}
1028
1029