Loading...
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 down_read(&OCFS2_I(inode)->ip_alloc_sem);
977 rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
978 &p_block, &p_count, NULL);
979 up_read(&OCFS2_I(inode)->ip_alloc_sem);
980 if (rc) {
981 mlog_errno(rc);
982 break;
983 }
984
985 if (!p_block) {
986 rc = -EIO;
987 mlog(ML_ERROR,
988 "Inode #%llu contains a hole at offset %llu\n",
989 (unsigned long long)OCFS2_I(inode)->ip_blkno,
990 (unsigned long long)(v_block + done) <<
991 inode->i_sb->s_blocksize_bits);
992 break;
993 }
994
995 count = nr - done;
996 if (p_count < count)
997 count = p_count;
998
999 /*
1000 * If the caller passed us bhs, they should have come
1001 * from a previous readahead call to this function. Thus,
1002 * they should have the right b_blocknr.
1003 */
1004 for (i = 0; i < count; i++) {
1005 if (!bhs[done + i])
1006 continue;
1007 BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
1008 }
1009
1010 rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
1011 bhs + done, flags, validate);
1012 if (rc) {
1013 mlog_errno(rc);
1014 break;
1015 }
1016 done += count;
1017 }
1018
1019out:
1020 return rc;
1021}
1022
1023
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