4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see [sun.com URL with a
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 * implementation of iam format for fixed size records.
39 * Author: Wang Di <wangdi@clusterfs.com>
40 * Author: Nikita Danilov <nikita@clusterfs.com>
43 #include <linux/types.h>
44 #include "osd_internal.h"
51 IAM_LEAF_HEADER_MAGIC = 0x1976 /* This is duplicated in
52 * lustre/utils/create_iam.c */
55 /* This is duplicated in lustre/utils/create_iam.c */
56 struct iam_leaf_head {
61 static inline int iam_lfix_entry_size(const struct iam_leaf *l)
63 return iam_leaf_descr(l)->id_key_size + iam_leaf_descr(l)->id_rec_size;
66 static inline struct iam_lentry *
67 iam_lfix_shift(const struct iam_leaf *l, struct iam_lentry *entry, int shift)
69 return (void *)entry + shift * iam_lfix_entry_size(l);
72 static inline struct iam_key *iam_leaf_key_at(struct iam_lentry *entry)
74 return (struct iam_key *)entry;
77 static inline int lfix_keycmp(const struct iam_container *c,
78 const struct iam_key *k1,
79 const struct iam_key *k2)
81 return memcmp(k1, k2, c->ic_descr->id_key_size);
84 static struct iam_leaf_head *iam_get_head(const struct iam_leaf *l)
86 return (struct iam_leaf_head *)l->il_bh->b_data;
89 static struct iam_lentry *iam_entries(const struct buffer_head *bh)
91 return (void *)bh->b_data + sizeof(struct iam_leaf_head);
94 static struct iam_lentry *iam_get_lentries(const struct iam_leaf *l)
96 return iam_entries(l->il_bh);
99 static int leaf_count_limit(const struct iam_leaf *leaf)
103 free_space = iam_leaf_container(leaf)->ic_object->i_sb->s_blocksize;
104 free_space -= sizeof(struct iam_leaf_head);
105 return free_space / iam_lfix_entry_size(leaf);
108 static int lentry_count_get(const struct iam_leaf *leaf)
110 return le16_to_cpu(iam_get_head(leaf)->ill_count);
113 static void lentry_count_set(struct iam_leaf *leaf, unsigned count)
115 assert_corr(0 <= count && count <= leaf_count_limit(leaf));
116 iam_get_head(leaf)->ill_count = cpu_to_le16(count);
119 static struct iam_lentry *iam_lfix_get_end(const struct iam_leaf *l);
121 #if LDISKFS_CORRECTNESS_ON || LDISKFS_INVARIANT_ON
122 static int iam_leaf_at_rec(const struct iam_leaf *folio)
125 iam_get_lentries(folio) <= folio->il_at &&
126 folio->il_at < iam_lfix_get_end(folio);
130 static struct iam_ikey *iam_lfix_ikey(const struct iam_leaf *l,
131 struct iam_ikey *key)
134 assert_corr(iam_leaf_at_rec(l));
135 return (struct iam_ikey*)ie;
138 static struct iam_key *iam_lfix_key(const struct iam_leaf *l)
141 assert_corr(iam_leaf_at_rec(l));
142 return (struct iam_key*)ie;
145 static int iam_lfix_key_size(const struct iam_leaf *l)
147 return iam_leaf_descr(l)->id_key_size;
150 static void iam_lfix_start(struct iam_leaf *l)
152 l->il_at = iam_get_lentries(l);
155 static inline ptrdiff_t iam_lfix_diff(const struct iam_leaf *l,
156 const struct iam_lentry *e1,
157 const struct iam_lentry *e2)
162 esize = iam_lfix_entry_size(l);
163 diff = (void *)e1 - (void *)e2;
164 assert_corr(diff / esize * esize == diff);
168 static int iam_lfix_init(struct iam_leaf *l)
171 struct iam_leaf_head *ill;
174 assert_corr(l->il_bh != NULL);
176 ill = iam_get_head(l);
177 count = le16_to_cpu(ill->ill_count);
178 if (le16_to_cpu(ill->ill_magic) == IAM_LEAF_HEADER_MAGIC &&
179 0 <= count && count <= leaf_count_limit(l)) {
180 l->il_at = l->il_entries = iam_get_lentries(l);
185 obj = iam_leaf_container(l)->ic_object;
186 CERROR("Wrong magic in node %llu (#%lu): %#x != %#x or "
187 "wrong count: %d (%d)\n",
188 (unsigned long long)l->il_bh->b_blocknr, obj->i_ino,
189 le16_to_cpu(ill->ill_magic), IAM_LEAF_HEADER_MAGIC,
190 count, leaf_count_limit(l));
196 static void iam_lfix_fini(struct iam_leaf *l)
198 l->il_entries = l->il_at = NULL;
201 static struct iam_lentry *iam_lfix_get_end(const struct iam_leaf *l)
203 int count = lentry_count_get(l);
204 struct iam_lentry *ile = iam_lfix_shift(l, l->il_entries, count);
209 static struct iam_rec *iam_lfix_rec(const struct iam_leaf *l)
212 assert_corr(iam_leaf_at_rec(l));
213 return e + iam_leaf_descr(l)->id_key_size;
216 static void iam_lfix_next(struct iam_leaf *l)
218 assert_corr(iam_leaf_at_rec(l));
219 l->il_at = iam_lfix_shift(l, l->il_at, 1);
227 static char hdigit(char ch)
229 static char d[] = "0123456789abcdef";
233 static char *hex(char ch, char *area)
235 area[0] = hdigit(ch >> 4);
236 area[1] = hdigit(ch);
241 static void l_print(struct iam_leaf *leaf, struct iam_lentry *entry)
247 area = (char *)entry;
248 printk(KERN_EMERG "[");
249 for (i = iam_lfix_key_size(leaf); i > 0; --i, ++area)
250 printk("%s", hex(*area, h));
252 for (i = iam_leaf_descr(leaf)->id_rec_size; i > 0; --i, ++area)
253 printk("%s", hex(*area, h));
257 static void lfix_print(struct iam_leaf *leaf)
259 struct iam_lentry *entry;
263 entry = leaf->il_entries;
264 count = lentry_count_get(leaf);
265 printk(KERN_EMERG "lfix: %p %p %d\n", leaf, leaf->il_at, count);
266 for (i = 0; i < count; ++i, entry = iam_lfix_shift(leaf, entry, 1))
267 l_print(leaf, entry);
270 static int iam_lfix_lookup(struct iam_leaf *l, const struct iam_key *k)
272 struct iam_lentry *p, *q, *m, *t;
273 struct iam_container *c;
277 count = lentry_count_get(l);
279 return IAM_LOOKUP_EMPTY;
281 result = IAM_LOOKUP_OK;
282 c = iam_leaf_container(l);
285 q = iam_lfix_shift(l, p, count - 1);
286 if (lfix_keycmp(c, k, iam_leaf_key_at(p)) < 0) {
288 * @k is less than the least key in the leaf
291 result = IAM_LOOKUP_BEFORE;
292 } else if (lfix_keycmp(c, iam_leaf_key_at(q), k) <= 0) {
298 while (iam_lfix_shift(l, p, 1) != q) {
299 m = iam_lfix_shift(l, p, iam_lfix_diff(l, q, p) / 2);
300 assert_corr(p < m && m < q);
301 if (lfix_keycmp(c, iam_leaf_key_at(m), k) <= 0)
306 assert_corr(lfix_keycmp(c, iam_leaf_key_at(p), k) <= 0 &&
307 lfix_keycmp(c, k, iam_leaf_key_at(q)) < 0);
309 * skip over records with duplicate keys.
311 while (p > l->il_entries) {
312 t = iam_lfix_shift(l, p, -1);
313 if (lfix_keycmp(c, iam_leaf_key_at(t), k) == 0)
320 assert_corr(iam_leaf_at_rec(l));
322 if (lfix_keycmp(c, iam_leaf_key_at(l->il_at), k) == 0)
323 result = IAM_LOOKUP_EXACT;
331 static int iam_lfix_ilookup(struct iam_leaf *l, const struct iam_ikey *ik)
333 return iam_lfix_lookup(l, (const struct iam_key *)ik);
336 static void iam_lfix_key_set(struct iam_leaf *l, const struct iam_key *k)
338 assert_corr(iam_leaf_at_rec(l));
339 memcpy(iam_leaf_key_at(l->il_at), k, iam_leaf_descr(l)->id_key_size);
342 static int iam_lfix_key_cmp(const struct iam_leaf *l, const struct iam_key *k)
344 return lfix_keycmp(iam_leaf_container(l), iam_leaf_key_at(l->il_at), k);
347 static int iam_lfix_key_eq(const struct iam_leaf *l, const struct iam_key *k)
349 return !lfix_keycmp(iam_leaf_container(l),
350 iam_leaf_key_at(l->il_at), k);
353 static void iam_lfix_rec_set(struct iam_leaf *l, const struct iam_rec *r)
355 assert_corr(iam_leaf_at_rec(l));
356 memcpy(iam_lfix_rec(l), r, iam_leaf_descr(l)->id_rec_size);
359 static inline int lfix_reccmp(const struct iam_container *c,
360 const struct iam_rec *r1,
361 const struct iam_rec *r2)
363 return memcmp(r1, r2, c->ic_descr->id_rec_size);
366 static int iam_lfix_rec_eq(const struct iam_leaf *l, const struct iam_rec *r)
368 return !lfix_reccmp(iam_leaf_container(l), iam_lfix_rec(l), r);
371 static void iam_lfix_rec_get(const struct iam_leaf *l, struct iam_rec *r)
373 assert_corr(iam_leaf_at_rec(l));
374 memcpy(r, iam_lfix_rec(l), iam_leaf_descr(l)->id_rec_size);
377 static void iam_lfix_rec_add(struct iam_leaf *leaf,
378 const struct iam_key *k, const struct iam_rec *r)
380 struct iam_lentry *end;
381 struct iam_lentry *cur;
382 struct iam_lentry *start;
386 assert_corr(iam_leaf_can_add(leaf, k, r));
388 count = lentry_count_get(leaf);
390 * This branch handles two exceptional cases:
392 * - leaf positioned beyond last record, and
396 if (!iam_leaf_at_end(leaf)) {
397 end = iam_lfix_get_end(leaf);
399 if (lfix_keycmp(iam_leaf_container(leaf),
400 k, iam_leaf_key_at(cur)) >= 0)
404 * Another exceptional case: insertion with the key
405 * less than least key in the leaf.
407 assert_corr(cur == leaf->il_entries);
410 diff = (void *)end - (void *)start;
411 assert_corr(diff >= 0);
412 memmove(iam_lfix_shift(leaf, start, 1), start, diff);
414 lentry_count_set(leaf, count + 1);
415 iam_lfix_key_set(leaf, k);
416 iam_lfix_rec_set(leaf, r);
417 assert_corr(iam_leaf_at_rec(leaf));
420 static void iam_lfix_rec_del(struct iam_leaf *leaf, int shift)
422 struct iam_lentry *next, *end;
426 assert_corr(iam_leaf_at_rec(leaf));
428 count = lentry_count_get(leaf);
429 end = iam_lfix_get_end(leaf);
430 next = iam_lfix_shift(leaf, leaf->il_at, 1);
431 diff = (void *)end - (void *)next;
432 memmove(leaf->il_at, next, diff);
434 lentry_count_set(leaf, count - 1);
437 static int iam_lfix_can_add(const struct iam_leaf *l,
438 const struct iam_key *k, const struct iam_rec *r)
440 return lentry_count_get(l) < leaf_count_limit(l);
443 static int iam_lfix_at_end(const struct iam_leaf *folio)
445 return folio->il_at == iam_lfix_get_end(folio);
448 static void iam_lfix_init_new(struct iam_container *c, struct buffer_head *bh)
450 struct iam_leaf_head *hdr;
452 hdr = (struct iam_leaf_head*)bh->b_data;
453 hdr->ill_magic = cpu_to_le16(IAM_LEAF_HEADER_MAGIC);
454 hdr->ill_count = cpu_to_le16(0);
457 static void iam_lfix_split(struct iam_leaf *l, struct buffer_head **bh,
460 struct iam_path *path;
461 struct iam_leaf_head *hdr;
462 const struct iam_ikey *pivot;
463 struct buffer_head *new_leaf;
472 path = iam_leaf_path(l);
474 hdr = (void *)new_leaf->b_data;
476 count = lentry_count_get(l);
479 start = iam_lfix_shift(l, iam_get_lentries(l), split);
480 finis = iam_lfix_shift(l, iam_get_lentries(l), count);
482 pivot = (const struct iam_ikey *)iam_leaf_key_at(start);
484 memmove(iam_entries(new_leaf), start, finis - start);
485 hdr->ill_count = cpu_to_le16(count - split);
486 lentry_count_set(l, split);
487 if ((void *)l->il_at >= start) {
489 * insertion point moves into new leaf.
494 shift = iam_lfix_diff(l, l->il_at, start);
497 l->il_curidx = new_blknr;
498 result = iam_lfix_init(l);
500 * init cannot fail, as node was just initialized.
502 assert_corr(result == 0);
503 l->il_at = iam_lfix_shift(l, iam_get_lentries(l), shift);
506 * Insert pointer to the new node (together with the least key in
507 * the node) into index node.
509 iam_insert_key_lock(path, path->ip_frame, pivot, new_blknr);
512 static int iam_lfix_leaf_empty(struct iam_leaf *leaf)
514 return lentry_count_get(leaf) == 0;
517 static struct iam_leaf_operations iam_lfix_leaf_ops = {
518 .init = iam_lfix_init,
519 .init_new = iam_lfix_init_new,
520 .fini = iam_lfix_fini,
521 .start = iam_lfix_start,
522 .next = iam_lfix_next,
524 .ikey = iam_lfix_ikey,
526 .key_set = iam_lfix_key_set,
527 .key_cmp = iam_lfix_key_cmp,
528 .key_eq = iam_lfix_key_eq,
529 .key_size = iam_lfix_key_size,
530 .rec_set = iam_lfix_rec_set,
531 .rec_eq = iam_lfix_rec_eq,
532 .rec_get = iam_lfix_rec_get,
533 .lookup = iam_lfix_lookup,
534 .ilookup = iam_lfix_ilookup,
535 .at_end = iam_lfix_at_end,
536 .rec_add = iam_lfix_rec_add,
537 .rec_del = iam_lfix_rec_del,
538 .can_add = iam_lfix_can_add,
539 .split = iam_lfix_split,
540 .leaf_empty = iam_lfix_leaf_empty,
548 /* This is duplicated in lustre/utils/create_iam.c */
550 * Then shalt thou see the dew-BEDABBLED wretch
551 * Turn, and return, indenting with the way;
552 * Each envious brier his weary legs doth scratch,
553 * Each shadow makes him stop, each murmur stay:
554 * For misery is trodden on by many,
555 * And being low never relieved by any.
557 IAM_LFIX_ROOT_MAGIC = 0xbedabb1edULL // d01efull
560 /* This is duplicated in lustre/utils/create_iam.c */
561 struct iam_lfix_root {
566 u8 ilr_indirect_levels;
570 static __u32 iam_lfix_root_ptr(struct iam_container *c)
575 static int iam_lfix_node_init(struct iam_container *c, struct buffer_head *bh,
581 static struct iam_entry *iam_lfix_root_inc(struct iam_container *c,
582 struct iam_path *path,
583 struct iam_frame *frame)
585 struct iam_lfix_root *root;
586 struct iam_entry *entries;
588 entries = frame->entries;
590 dx_set_count(entries, 2);
591 assert_corr(dx_get_limit(entries) == dx_root_limit(path));
593 root = (void *)frame->bh->b_data;
594 assert_corr(le64_to_cpu(root->ilr_magic) == IAM_LFIX_ROOT_MAGIC);
595 root->ilr_indirect_levels ++;
596 frame->at = entries = iam_entry_shift(path, entries, 1);
597 memset(iam_ikey_at(path, entries), 0,
598 iam_path_descr(path)->id_ikey_size);
602 static int iam_lfix_node_check(struct iam_path *path, struct iam_frame *frame)
606 unsigned limit_correct;
607 struct iam_entry *entries;
609 entries = dx_node_get_entries(path, frame);
611 if (frame == path->ip_frames) {
612 struct iam_lfix_root *root;
614 root = (void *)frame->bh->b_data;
615 if (le64_to_cpu(root->ilr_magic) != IAM_LFIX_ROOT_MAGIC) {
618 limit_correct = dx_root_limit(path);
620 limit_correct = dx_node_limit(path);
621 count = dx_get_count(entries);
622 limit = dx_get_limit(entries);
626 if (limit != limit_correct) {
632 static int iam_lfix_node_load(struct iam_path *path, struct iam_frame *frame)
634 struct iam_entry *entries;
636 entries = dx_node_get_entries(path, frame);
638 data = frame->bh->b_data;
640 if (frame == path->ip_frames) {
641 struct iam_lfix_root *root;
644 path->ip_indirect = root->ilr_indirect_levels;
645 if (path->ip_ikey_target == NULL)
646 path->ip_ikey_target =
647 (struct iam_ikey *)path->ip_key_target;
649 frame->entries = frame->at = entries;
653 static int iam_lfix_ikeycmp(const struct iam_container *c,
654 const struct iam_ikey *k1,
655 const struct iam_ikey *k2)
657 return memcmp(k1, k2, c->ic_descr->id_ikey_size);
660 static struct iam_path_descr *iam_lfix_ipd_alloc(const struct iam_container *c,
663 return iam_ipd_alloc(area, c->ic_descr->id_ikey_size);
666 static struct iam_operations iam_lfix_ops = {
667 .id_root_ptr = iam_lfix_root_ptr,
668 .id_node_read = iam_node_read,
669 .id_node_init = iam_lfix_node_init,
670 .id_node_check = iam_lfix_node_check,
671 .id_node_load = iam_lfix_node_load,
672 .id_ikeycmp = iam_lfix_ikeycmp,
673 .id_root_inc = iam_lfix_root_inc,
674 .id_ipd_alloc = iam_lfix_ipd_alloc,
675 .id_ipd_free = iam_ipd_free,
679 static int iam_lfix_guess(struct iam_container *c)
682 struct buffer_head *bh;
683 const struct iam_lfix_root *root;
685 assert_corr(c->ic_object != NULL);
687 result = iam_node_read(c, iam_lfix_root_ptr(c), NULL, &bh);
689 root = (void *)bh->b_data;
690 if (le64_to_cpu(root->ilr_magic) == IAM_LFIX_ROOT_MAGIC) {
691 struct iam_descr *descr;
694 descr->id_key_size = le16_to_cpu(root->ilr_keysize);
695 descr->id_ikey_size = le16_to_cpu(root->ilr_keysize);
696 descr->id_rec_size = le16_to_cpu(root->ilr_recsize);
697 descr->id_ptr_size = le16_to_cpu(root->ilr_ptrsize);
698 descr->id_root_gap = sizeof(struct iam_lfix_root);
699 descr->id_node_gap = 0;
700 descr->id_ops = &iam_lfix_ops;
701 descr->id_leaf_ops = &iam_lfix_leaf_ops;
712 static struct iam_format iam_lfix_format = {
713 .if_guess = iam_lfix_guess
716 void iam_lfix_format_init(void)
718 iam_format_register(&iam_lfix_format);
729 #define LFIX_ROOT_RECNO \
730 ((4096 - sizeof(struct iam_lfix_root)) / (KEYSIZE + PTRSIZE))
732 #define LFIX_INDEX_RECNO (4096 / (KEYSIZE + PTRSIZE))
734 #define LFIX_LEAF_RECNO \
735 ((4096 - sizeof(struct iam_leaf_head)) / (KEYSIZE + RECSIZE))
738 struct iam_lfix_root lr_root;
742 } lr_entry[LFIX_ROOT_RECNO];
746 struct dx_countlimit li_cl;
747 char li_padding[KEYSIZE + PTRSIZE - sizeof(struct dx_countlimit)];
751 } li_entry[LFIX_INDEX_RECNO - 1];
755 struct iam_leaf_head ll_head;
759 } ll_entry[LFIX_LEAF_RECNO];
762 #define STORE_UNALIGNED(val, dst) \
764 typeof(val) __val = (val); \
765 CLASSERT(sizeof(val) == sizeof(*(dst))); \
766 memcpy(dst, &__val, sizeof(*(dst))); \
769 static void lfix_root(void *buf,
770 int blocksize, int keysize, int ptrsize, int recsize)
772 struct iam_lfix_root *root;
773 struct dx_countlimit *limit;
777 *root = (typeof(*root)) {
778 .ilr_magic = cpu_to_le64(IAM_LFIX_ROOT_MAGIC),
779 .ilr_keysize = cpu_to_le16(keysize),
780 .ilr_recsize = cpu_to_le16(recsize),
781 .ilr_ptrsize = cpu_to_le16(ptrsize),
782 .ilr_indirect_levels = 0
785 limit = (void *)(root + 1);
786 *limit = (typeof(*limit)){
788 * limit itself + one pointer to the leaf.
790 .count = cpu_to_le16(2),
791 .limit = iam_root_limit(sizeof(struct iam_lfix_root),
792 blocksize, keysize + ptrsize)
795 /* To guarantee that the padding "keysize + ptrsize"
796 * covers the "dx_countlimit" and the "idle_blocks". */
797 LASSERT((keysize + ptrsize) >=
798 (sizeof(struct dx_countlimit) + sizeof(__u32)));
800 entry = (void *)(limit + 1);
801 /* Put "idle_blocks" just after the limit. There was padding after
802 * the limit, the "idle_blocks" re-uses part of the padding, so no
803 * compatibility issues with old layout.
810 entry = (void *)(root + 1) + keysize + ptrsize;
813 * Entry format is <key> followed by <ptr>. In the minimal tree
814 * consisting of a root and single node, <key> is a minimal possible
817 * XXX: this key is hard-coded to be a sequence of 0's.
820 memset(entry, 0, keysize);
822 /* now @entry points to <ptr> */
824 STORE_UNALIGNED(cpu_to_le32(1), (u_int32_t *)entry);
826 STORE_UNALIGNED(cpu_to_le64(1), (u_int64_t *)entry);
829 static void lfix_leaf(void *buf,
830 int blocksize, int keysize, int ptrsize, int recsize)
832 struct iam_leaf_head *head;
837 *head = (struct iam_leaf_head) {
838 .ill_magic = cpu_to_le16(IAM_LEAF_HEADER_MAGIC),
840 * Leaf contains an entry with the smallest possible key
841 * (created by zeroing).
843 .ill_count = cpu_to_le16(1),
846 entry = (void *)(head + 1);
847 memset(entry, 0, keysize + recsize);
850 int iam_lfix_create(struct inode *obj,
851 int keysize, int ptrsize, int recsize, handle_t *handle)
853 struct buffer_head *root_node;
854 struct buffer_head *leaf_node;
855 struct super_block *sb;
861 assert_corr(obj->i_size == 0);
864 bsize = sb->s_blocksize;
865 root_node = osd_ldiskfs_append(handle, obj, &blknr);
866 if (IS_ERR(root_node))
867 GOTO(out, result = PTR_ERR(root_node));
869 leaf_node = osd_ldiskfs_append(handle, obj, &blknr);
870 if (IS_ERR(leaf_node))
871 GOTO(out_root, result = PTR_ERR(leaf_node));
873 lfix_root(root_node->b_data, bsize, keysize, ptrsize, recsize);
874 lfix_leaf(leaf_node->b_data, bsize, keysize, ptrsize, recsize);
875 ldiskfs_mark_inode_dirty(handle, obj);
876 result = ldiskfs_handle_dirty_metadata(handle, NULL, root_node);
878 result = ldiskfs_handle_dirty_metadata(handle, NULL, leaf_node);
880 ldiskfs_std_error(sb, result);
884 GOTO(out_root, result);