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
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * implementation of iam format for fixed size records.
35 * Author: Wang Di <wangdi@clusterfs.com>
36 * Author: Nikita Danilov <nikita@clusterfs.com>
39 #include <linux/types.h>
40 #include "osd_internal.h"
47 IAM_LEAF_HEADER_MAGIC = 0x1976 /* This is duplicated in
48 * lustre/utils/create_iam.c */
51 /* This is duplicated in lustre/utils/create_iam.c */
52 struct iam_leaf_head {
57 static inline int iam_lfix_entry_size(const struct iam_leaf *l)
59 return iam_leaf_descr(l)->id_key_size + iam_leaf_descr(l)->id_rec_size;
62 static inline struct iam_lentry *
63 iam_lfix_shift(const struct iam_leaf *l, struct iam_lentry *entry, int shift)
65 return (void *)entry + shift * iam_lfix_entry_size(l);
68 static inline struct iam_key *iam_leaf_key_at(struct iam_lentry *entry)
70 return (struct iam_key *)entry;
73 static inline int lfix_keycmp(const struct iam_container *c,
74 const struct iam_key *k1,
75 const struct iam_key *k2)
77 return memcmp(k1, k2, c->ic_descr->id_key_size);
80 static struct iam_leaf_head *iam_get_head(const struct iam_leaf *l)
82 return (struct iam_leaf_head *)l->il_bh->b_data;
85 static struct iam_lentry *iam_entries(const struct buffer_head *bh)
87 return (void *)bh->b_data + sizeof(struct iam_leaf_head);
90 static struct iam_lentry *iam_get_lentries(const struct iam_leaf *l)
92 return iam_entries(l->il_bh);
95 static int leaf_count_limit(const struct iam_leaf *leaf)
99 free_space = iam_leaf_container(leaf)->ic_object->i_sb->s_blocksize;
100 free_space -= sizeof(struct iam_leaf_head);
101 return free_space / iam_lfix_entry_size(leaf);
104 static int lentry_count_get(const struct iam_leaf *leaf)
106 return le16_to_cpu(iam_get_head(leaf)->ill_count);
109 static void lentry_count_set(struct iam_leaf *leaf, unsigned count)
111 assert_corr(0 <= count && count <= leaf_count_limit(leaf));
112 iam_get_head(leaf)->ill_count = cpu_to_le16(count);
115 static struct iam_lentry *iam_lfix_get_end(const struct iam_leaf *l);
117 #if LDISKFS_CORRECTNESS_ON || LDISKFS_INVARIANT_ON
118 static int iam_leaf_at_rec(const struct iam_leaf *folio)
120 return iam_get_lentries(folio) <= folio->il_at &&
121 folio->il_at < iam_lfix_get_end(folio);
125 static struct iam_ikey *iam_lfix_ikey(const struct iam_leaf *l,
126 struct iam_ikey *key)
130 assert_corr(iam_leaf_at_rec(l));
131 return (struct iam_ikey *)ie;
134 static struct iam_key *iam_lfix_key(const struct iam_leaf *l)
138 assert_corr(iam_leaf_at_rec(l));
139 return (struct iam_key *)ie;
142 static int iam_lfix_key_size(const struct iam_leaf *l)
144 return iam_leaf_descr(l)->id_key_size;
147 static void iam_lfix_start(struct iam_leaf *l)
149 l->il_at = iam_get_lentries(l);
152 static inline ptrdiff_t iam_lfix_diff(const struct iam_leaf *l,
153 const struct iam_lentry *e1,
154 const struct iam_lentry *e2)
159 esize = iam_lfix_entry_size(l);
160 diff = (void *)e1 - (void *)e2;
161 assert_corr(diff / esize * esize == diff);
165 static int iam_lfix_init(struct iam_leaf *l)
168 struct iam_leaf_head *ill;
171 assert_corr(l->il_bh != NULL);
173 ill = iam_get_head(l);
174 count = le16_to_cpu(ill->ill_count);
175 if (le16_to_cpu(ill->ill_magic) == IAM_LEAF_HEADER_MAGIC &&
176 0 <= count && count <= leaf_count_limit(l)) {
177 l->il_at = l->il_entries = iam_get_lentries(l);
183 obj = iam_leaf_container(l)->ic_object;
185 "Bad magic in node %llu #%lu: %#x != %#x or bad cnt: %d %d: rc = %d\n",
186 (unsigned long long)l->il_bh->b_blocknr, obj->i_ino,
187 le16_to_cpu(ill->ill_magic), IAM_LEAF_HEADER_MAGIC,
188 count, leaf_count_limit(l), result);
193 static void iam_lfix_fini(struct iam_leaf *l)
195 l->il_entries = l->il_at = NULL;
198 static struct iam_lentry *iam_lfix_get_end(const struct iam_leaf *l)
200 int count = lentry_count_get(l);
201 struct iam_lentry *ile = iam_lfix_shift(l, l->il_entries, count);
206 static struct iam_rec *iam_lfix_rec(const struct iam_leaf *l)
210 assert_corr(iam_leaf_at_rec(l));
211 return e + iam_leaf_descr(l)->id_key_size;
214 static void iam_lfix_next(struct iam_leaf *l)
216 assert_corr(iam_leaf_at_rec(l));
217 l->il_at = iam_lfix_shift(l, l->il_at, 1);
225 static char hdigit(char ch)
227 static const char d[] = "0123456789abcdef";
232 static char *hex(char ch, char *area)
234 area[0] = hdigit(ch >> 4);
235 area[1] = hdigit(ch);
240 static void l_print(struct iam_leaf *leaf, struct iam_lentry *entry)
246 area = (char *)entry;
247 printk(KERN_EMERG "[");
248 for (i = iam_lfix_key_size(leaf); i > 0; --i, ++area)
249 printk("%s", hex(*area, h));
251 for (i = iam_leaf_descr(leaf)->id_rec_size; i > 0; --i, ++area)
252 printk("%s", hex(*area, h));
256 static void lfix_print(struct iam_leaf *leaf)
258 struct iam_lentry *entry;
262 entry = leaf->il_entries;
263 count = lentry_count_get(leaf);
264 printk(KERN_EMERG "lfix: %p %p %d\n", leaf, leaf->il_at, count);
265 for (i = 0; i < count; ++i, entry = iam_lfix_shift(leaf, entry, 1))
266 l_print(leaf, entry);
269 static int iam_lfix_lookup(struct iam_leaf *l, const struct iam_key *k)
271 struct iam_lentry *p, *q, *m, *t;
272 struct iam_container *c;
276 count = lentry_count_get(l);
278 return IAM_LOOKUP_EMPTY;
280 result = IAM_LOOKUP_OK;
281 c = iam_leaf_container(l);
284 q = iam_lfix_shift(l, p, count - 1);
285 if (lfix_keycmp(c, k, iam_leaf_key_at(p)) < 0) {
287 * @k is less than the least key in the leaf
290 result = IAM_LOOKUP_BEFORE;
291 } else if (lfix_keycmp(c, iam_leaf_key_at(q), k) <= 0) {
297 while (iam_lfix_shift(l, p, 1) != q) {
298 m = iam_lfix_shift(l, p, iam_lfix_diff(l, q, p) / 2);
299 assert_corr(p < m && m < q);
300 if (lfix_keycmp(c, iam_leaf_key_at(m), k) <= 0)
305 assert_corr(lfix_keycmp(c, iam_leaf_key_at(p), k) <= 0 &&
306 lfix_keycmp(c, k, iam_leaf_key_at(q)) < 0);
308 * skip over records with duplicate keys.
310 while (p > l->il_entries) {
311 t = iam_lfix_shift(l, p, -1);
312 if (lfix_keycmp(c, iam_leaf_key_at(t), k) == 0)
319 assert_corr(iam_leaf_at_rec(l));
321 if (lfix_keycmp(c, iam_leaf_key_at(l->il_at), k) == 0)
322 result = IAM_LOOKUP_EXACT;
330 static int iam_lfix_ilookup(struct iam_leaf *l, const struct iam_ikey *ik)
332 return iam_lfix_lookup(l, (const struct iam_key *)ik);
335 static void iam_lfix_key_set(struct iam_leaf *l, const struct iam_key *k)
337 assert_corr(iam_leaf_at_rec(l));
338 memcpy(iam_leaf_key_at(l->il_at), k, iam_leaf_descr(l)->id_key_size);
341 static int iam_lfix_key_cmp(const struct iam_leaf *l, const struct iam_key *k)
343 return lfix_keycmp(iam_leaf_container(l), iam_leaf_key_at(l->il_at), k);
346 static int iam_lfix_key_eq(const struct iam_leaf *l, const struct iam_key *k)
348 return !lfix_keycmp(iam_leaf_container(l),
349 iam_leaf_key_at(l->il_at), k);
352 static void iam_lfix_rec_set(struct iam_leaf *l, const struct iam_rec *r)
354 assert_corr(iam_leaf_at_rec(l));
355 memcpy(iam_lfix_rec(l), r, iam_leaf_descr(l)->id_rec_size);
358 static inline int lfix_reccmp(const struct iam_container *c,
359 const struct iam_rec *r1,
360 const struct iam_rec *r2)
362 return memcmp(r1, r2, c->ic_descr->id_rec_size);
365 static int iam_lfix_rec_eq(const struct iam_leaf *l, const struct iam_rec *r)
367 return !lfix_reccmp(iam_leaf_container(l), iam_lfix_rec(l), r);
370 static void iam_lfix_rec_get(const struct iam_leaf *l, struct iam_rec *r)
372 assert_corr(iam_leaf_at_rec(l));
373 memcpy(r, iam_lfix_rec(l), iam_leaf_descr(l)->id_rec_size);
376 static void iam_lfix_rec_add(struct iam_leaf *leaf,
377 const struct iam_key *k, const struct iam_rec *r)
379 struct iam_lentry *end;
380 struct iam_lentry *cur;
381 struct iam_lentry *start;
385 assert_corr(iam_leaf_can_add(leaf, k, r));
387 count = lentry_count_get(leaf);
389 * This branch handles two exceptional cases:
391 * - leaf positioned beyond last record, and
395 if (!iam_leaf_at_end(leaf)) {
396 end = iam_lfix_get_end(leaf);
398 if (lfix_keycmp(iam_leaf_container(leaf),
399 k, iam_leaf_key_at(cur)) >= 0)
403 * Another exceptional case: insertion with the key
404 * less than least key in the leaf.
406 assert_corr(cur == leaf->il_entries);
409 diff = (void *)end - (void *)start;
410 assert_corr(diff >= 0);
411 memmove(iam_lfix_shift(leaf, start, 1), start, diff);
413 lentry_count_set(leaf, count + 1);
414 iam_lfix_key_set(leaf, k);
415 iam_lfix_rec_set(leaf, r);
416 assert_corr(iam_leaf_at_rec(leaf));
419 static void iam_lfix_rec_del(struct iam_leaf *leaf, int shift)
421 struct iam_lentry *next, *end;
425 assert_corr(iam_leaf_at_rec(leaf));
427 count = lentry_count_get(leaf);
428 end = iam_lfix_get_end(leaf);
429 next = iam_lfix_shift(leaf, leaf->il_at, 1);
430 diff = (void *)end - (void *)next;
431 memmove(leaf->il_at, next, diff);
433 lentry_count_set(leaf, count - 1);
436 static int iam_lfix_can_add(const struct iam_leaf *l,
437 const struct iam_key *k, const struct iam_rec *r)
439 return lentry_count_get(l) < leaf_count_limit(l);
442 static int iam_lfix_at_end(const struct iam_leaf *folio)
444 return folio->il_at == iam_lfix_get_end(folio);
447 static void iam_lfix_init_new(struct iam_container *c, struct buffer_head *bh)
449 struct iam_leaf_head *hdr;
451 hdr = (struct iam_leaf_head *)bh->b_data;
452 hdr->ill_magic = cpu_to_le16(IAM_LEAF_HEADER_MAGIC);
453 hdr->ill_count = cpu_to_le16(0);
456 static void iam_lfix_split(struct iam_leaf *l, struct buffer_head **bh,
459 struct iam_path *path;
460 struct iam_leaf_head *hdr;
461 const struct iam_ikey *pivot;
462 struct buffer_head *new_leaf;
471 path = iam_leaf_path(l);
473 hdr = (void *)new_leaf->b_data;
475 count = lentry_count_get(l);
478 start = iam_lfix_shift(l, iam_get_lentries(l), split);
479 finis = iam_lfix_shift(l, iam_get_lentries(l), count);
481 pivot = (const struct iam_ikey *)iam_leaf_key_at(start);
483 memmove(iam_entries(new_leaf), start, finis - start);
484 hdr->ill_count = cpu_to_le16(count - split);
485 lentry_count_set(l, split);
486 if ((void *)l->il_at >= start) {
488 * insertion point moves into new leaf.
492 shift = iam_lfix_diff(l, l->il_at, start);
495 l->il_curidx = new_blknr;
498 * init cannot fail, as node was just initialized.
500 assert_corr(result == 0);
501 l->il_at = iam_lfix_shift(l, iam_get_lentries(l), shift);
504 * Insert pointer to the new node (together with the least key in
505 * the node) into index node.
507 iam_insert_key_lock(path, path->ip_frame, pivot, new_blknr);
510 static int iam_lfix_leaf_empty(struct iam_leaf *leaf)
512 return lentry_count_get(leaf) == 0;
515 static struct iam_leaf_operations iam_lfix_leaf_ops = {
516 .init = iam_lfix_init,
517 .init_new = iam_lfix_init_new,
518 .fini = iam_lfix_fini,
519 .start = iam_lfix_start,
520 .next = iam_lfix_next,
522 .ikey = iam_lfix_ikey,
524 .key_set = iam_lfix_key_set,
525 .key_cmp = iam_lfix_key_cmp,
526 .key_eq = iam_lfix_key_eq,
527 .key_size = iam_lfix_key_size,
528 .rec_set = iam_lfix_rec_set,
529 .rec_eq = iam_lfix_rec_eq,
530 .rec_get = iam_lfix_rec_get,
531 .lookup = iam_lfix_lookup,
532 .ilookup = iam_lfix_ilookup,
533 .at_end = iam_lfix_at_end,
534 .rec_add = iam_lfix_rec_add,
535 .rec_del = iam_lfix_rec_del,
536 .can_add = iam_lfix_can_add,
537 .split = iam_lfix_split,
538 .leaf_empty = iam_lfix_leaf_empty,
546 /* This is duplicated in lustre/utils/create_iam.c */
548 * Then shalt thou see the dew-BEDABBLED wretch
549 * Turn, and return, indenting with the way;
550 * Each envious brier his weary legs doth scratch,
551 * Each shadow makes him stop, each murmur stay:
552 * For misery is trodden on by many,
553 * And being low never relieved by any.
555 IAM_LFIX_ROOT_MAGIC = 0xbedabb1edULL /* d01efull */
558 /* This is duplicated in lustre/utils/create_iam.c */
559 struct iam_lfix_root {
564 u8 ilr_indirect_levels;
568 static __u32 iam_lfix_root_ptr(struct iam_container *c)
573 static int iam_lfix_node_init(struct iam_container *c, struct buffer_head *bh,
579 static struct iam_entry *iam_lfix_root_inc(struct iam_container *c,
580 struct iam_path *path,
581 struct iam_frame *frame)
583 struct iam_lfix_root *root;
584 struct iam_entry *entries;
586 entries = frame->entries;
588 dx_set_count(entries, 2);
589 assert_corr(dx_get_limit(entries) == dx_root_limit(path));
591 root = (void *)frame->bh->b_data;
592 assert_corr(le64_to_cpu(root->ilr_magic) == IAM_LFIX_ROOT_MAGIC);
593 root->ilr_indirect_levels++;
594 frame->at = entries = iam_entry_shift(path, entries, 1);
595 memset(iam_ikey_at(path, entries), 0,
596 iam_path_descr(path)->id_ikey_size);
600 static int iam_lfix_node_check(struct iam_path *path, struct iam_frame *frame)
604 unsigned int limit_correct;
605 struct iam_entry *entries;
607 entries = dx_node_get_entries(path, frame);
609 if (frame == path->ip_frames) {
610 struct iam_lfix_root *root;
612 root = (void *)frame->bh->b_data;
613 if (le64_to_cpu(root->ilr_magic) != IAM_LFIX_ROOT_MAGIC)
615 limit_correct = dx_root_limit(path);
617 limit_correct = dx_node_limit(path);
619 count = dx_get_count(entries);
620 limit = dx_get_limit(entries);
623 if (limit != limit_correct)
628 static int iam_lfix_node_load(struct iam_path *path, struct iam_frame *frame)
630 struct iam_entry *entries;
633 entries = dx_node_get_entries(path, frame);
634 data = frame->bh->b_data;
636 if (frame == path->ip_frames) {
637 struct iam_lfix_root *root;
640 path->ip_indirect = root->ilr_indirect_levels;
641 if (path->ip_ikey_target == NULL)
642 path->ip_ikey_target =
643 (struct iam_ikey *)path->ip_key_target;
645 frame->entries = frame->at = entries;
649 static int iam_lfix_ikeycmp(const struct iam_container *c,
650 const struct iam_ikey *k1,
651 const struct iam_ikey *k2)
653 return memcmp(k1, k2, c->ic_descr->id_ikey_size);
656 static struct iam_path_descr *iam_lfix_ipd_alloc(const struct iam_container *c,
659 return iam_ipd_alloc(area, c->ic_descr->id_ikey_size);
662 static struct iam_operations iam_lfix_ops = {
663 .id_root_ptr = iam_lfix_root_ptr,
664 .id_node_read = iam_node_read,
665 .id_node_init = iam_lfix_node_init,
666 .id_node_check = iam_lfix_node_check,
667 .id_node_load = iam_lfix_node_load,
668 .id_ikeycmp = iam_lfix_ikeycmp,
669 .id_root_inc = iam_lfix_root_inc,
670 .id_ipd_alloc = iam_lfix_ipd_alloc,
671 .id_ipd_free = iam_ipd_free,
675 int iam_lfix_guess(struct iam_container *c)
678 struct buffer_head *bh;
679 const struct iam_lfix_root *root;
681 assert_corr(c->ic_object != NULL);
683 result = iam_node_read(c, iam_lfix_root_ptr(c), NULL, &bh);
685 root = (void *)bh->b_data;
686 if (le64_to_cpu(root->ilr_magic) == IAM_LFIX_ROOT_MAGIC) {
687 struct iam_descr *descr;
690 descr->id_key_size = le16_to_cpu(root->ilr_keysize);
691 descr->id_ikey_size = le16_to_cpu(root->ilr_keysize);
692 descr->id_rec_size = le16_to_cpu(root->ilr_recsize);
693 descr->id_ptr_size = le16_to_cpu(root->ilr_ptrsize);
694 descr->id_root_gap = sizeof(struct iam_lfix_root);
695 descr->id_node_gap = 0;
696 descr->id_ops = &iam_lfix_ops;
697 descr->id_leaf_ops = &iam_lfix_leaf_ops;
715 #define LFIX_ROOT_RECNO \
716 ((4096 - sizeof(struct iam_lfix_root)) / (KEYSIZE + PTRSIZE))
718 #define LFIX_INDEX_RECNO (4096 / (KEYSIZE + PTRSIZE))
720 #define LFIX_LEAF_RECNO \
721 ((4096 - sizeof(struct iam_leaf_head)) / (KEYSIZE + RECSIZE))
724 struct iam_lfix_root lr_root;
728 } lr_entry[LFIX_ROOT_RECNO];
732 struct dx_countlimit li_cl;
733 char li_padding[KEYSIZE + PTRSIZE - sizeof(struct dx_countlimit)];
737 } li_entry[LFIX_INDEX_RECNO - 1];
741 struct iam_leaf_head ll_head;
745 } ll_entry[LFIX_LEAF_RECNO];
748 #define STORE_UNALIGNED(val, dst) \
750 typeof(val) __val = (val); \
751 BUILD_BUG_ON(sizeof(val) != sizeof(*(dst))); \
752 memcpy(dst, &__val, sizeof(*(dst))); \
755 static void lfix_root(void *buf,
756 int blocksize, int keysize, int ptrsize, int recsize)
758 struct iam_lfix_root *root;
759 struct dx_countlimit *limit;
763 *root = (typeof(*root)) {
764 .ilr_magic = cpu_to_le64(IAM_LFIX_ROOT_MAGIC),
765 .ilr_keysize = cpu_to_le16(keysize),
766 .ilr_recsize = cpu_to_le16(recsize),
767 .ilr_ptrsize = cpu_to_le16(ptrsize),
768 .ilr_indirect_levels = 0
771 limit = (void *)(root + 1);
772 *limit = (typeof(*limit)){
774 * limit itself + one pointer to the leaf.
776 .count = cpu_to_le16(2),
777 .limit = iam_root_limit(sizeof(struct iam_lfix_root),
778 blocksize, keysize + ptrsize)
781 /* To guarantee that the padding "keysize + ptrsize"
782 * covers the "dx_countlimit" and the "idle_blocks". */
783 LASSERT((keysize + ptrsize) >=
784 (sizeof(struct dx_countlimit) + sizeof(__u32)));
786 entry = (void *)(limit + 1);
787 /* Put "idle_blocks" just after the limit. There was padding after
788 * the limit, the "idle_blocks" re-uses part of the padding, so no
789 * compatibility issues with old layout.
796 entry = (void *)(root + 1) + keysize + ptrsize;
799 * Entry format is <key> followed by <ptr>. In the minimal tree
800 * consisting of a root and single node, <key> is a minimal possible
803 * XXX: this key is hard-coded to be a sequence of 0's.
806 memset(entry, 0, keysize);
808 /* now @entry points to <ptr> */
810 STORE_UNALIGNED(cpu_to_le32(1), (u_int32_t *)entry);
812 STORE_UNALIGNED(cpu_to_le64(1), (u_int64_t *)entry);
815 static void lfix_leaf(void *buf,
816 int blocksize, int keysize, int ptrsize, int recsize)
818 struct iam_leaf_head *head;
823 *head = (struct iam_leaf_head) {
824 .ill_magic = cpu_to_le16(IAM_LEAF_HEADER_MAGIC),
826 * Leaf contains an entry with the smallest possible key
827 * (created by zeroing).
829 .ill_count = cpu_to_le16(1),
832 entry = (void *)(head + 1);
833 memset(entry, 0, keysize + recsize);
836 int iam_lfix_create(struct inode *obj,
837 int keysize, int ptrsize, int recsize, handle_t *handle)
839 struct buffer_head *root_node;
840 struct buffer_head *leaf_node;
841 struct super_block *sb;
846 assert_corr(obj->i_size == 0);
849 bsize = sb->s_blocksize;
850 root_node = osd_ldiskfs_append(handle, obj, &blknr);
851 if (IS_ERR(root_node))
852 GOTO(out, result = PTR_ERR(root_node));
854 leaf_node = osd_ldiskfs_append(handle, obj, &blknr);
855 if (IS_ERR(leaf_node))
856 GOTO(out_root, result = PTR_ERR(leaf_node));
858 lfix_root(root_node->b_data, bsize, keysize, ptrsize, recsize);
859 lfix_leaf(leaf_node->b_data, bsize, keysize, ptrsize, recsize);
860 ldiskfs_mark_inode_dirty(handle, obj);
861 result = ldiskfs_handle_dirty_metadata(handle, NULL, root_node);
863 result = ldiskfs_handle_dirty_metadata(handle, NULL, leaf_node);
865 ldiskfs_std_error(sb, result);
869 GOTO(out_root, result);