1*0713e232SGeorge Wilson /*
2*0713e232SGeorge Wilson * CDDL HEADER START
3*0713e232SGeorge Wilson *
4*0713e232SGeorge Wilson * The contents of this file are subject to the terms of the
5*0713e232SGeorge Wilson * Common Development and Distribution License (the "License").
6*0713e232SGeorge Wilson * You may not use this file except in compliance with the License.
7*0713e232SGeorge Wilson *
8*0713e232SGeorge Wilson * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9*0713e232SGeorge Wilson * or http://www.opensolaris.org/os/licensing.
10*0713e232SGeorge Wilson * See the License for the specific language governing permissions
11*0713e232SGeorge Wilson * and limitations under the License.
12*0713e232SGeorge Wilson *
13*0713e232SGeorge Wilson * When distributing Covered Code, include this CDDL HEADER in each
14*0713e232SGeorge Wilson * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15*0713e232SGeorge Wilson * If applicable, add the following below this CDDL HEADER, with the
16*0713e232SGeorge Wilson * fields enclosed by brackets "[]" replaced with your own identifying
17*0713e232SGeorge Wilson * information: Portions Copyright [yyyy] [name of copyright owner]
18*0713e232SGeorge Wilson *
19*0713e232SGeorge Wilson * CDDL HEADER END
20*0713e232SGeorge Wilson */
21*0713e232SGeorge Wilson /*
22*0713e232SGeorge Wilson * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23*0713e232SGeorge Wilson * Use is subject to license terms.
24*0713e232SGeorge Wilson */
25*0713e232SGeorge Wilson /*
26*0713e232SGeorge Wilson * Copyright (c) 2013 by Delphix. All rights reserved.
27*0713e232SGeorge Wilson */
28*0713e232SGeorge Wilson
29*0713e232SGeorge Wilson #include <sys/zfs_context.h>
30*0713e232SGeorge Wilson #include <sys/range_tree.h>
31*0713e232SGeorge Wilson #include <sys/space_reftree.h>
32*0713e232SGeorge Wilson
33*0713e232SGeorge Wilson /*
34*0713e232SGeorge Wilson * Space reference trees.
35*0713e232SGeorge Wilson *
36*0713e232SGeorge Wilson * A range tree is a collection of integers. Every integer is either
37*0713e232SGeorge Wilson * in the tree, or it's not. A space reference tree generalizes
38*0713e232SGeorge Wilson * the idea: it allows its members to have arbitrary reference counts,
39*0713e232SGeorge Wilson * as opposed to the implicit reference count of 0 or 1 in a range tree.
40*0713e232SGeorge Wilson * This representation comes in handy when computing the union or
41*0713e232SGeorge Wilson * intersection of multiple space maps. For example, the union of
42*0713e232SGeorge Wilson * N range trees is the subset of the reference tree with refcnt >= 1.
43*0713e232SGeorge Wilson * The intersection of N range trees is the subset with refcnt >= N.
44*0713e232SGeorge Wilson *
45*0713e232SGeorge Wilson * [It's very much like a Fourier transform. Unions and intersections
46*0713e232SGeorge Wilson * are hard to perform in the 'range tree domain', so we convert the trees
47*0713e232SGeorge Wilson * into the 'reference count domain', where it's trivial, then invert.]
48*0713e232SGeorge Wilson *
49*0713e232SGeorge Wilson * vdev_dtl_reassess() uses computations of this form to determine
50*0713e232SGeorge Wilson * DTL_MISSING and DTL_OUTAGE for interior vdevs -- e.g. a RAID-Z vdev
51*0713e232SGeorge Wilson * has an outage wherever refcnt >= vdev_nparity + 1, and a mirror vdev
52*0713e232SGeorge Wilson * has an outage wherever refcnt >= vdev_children.
53*0713e232SGeorge Wilson */
54*0713e232SGeorge Wilson static int
space_reftree_compare(const void * x1,const void * x2)55*0713e232SGeorge Wilson space_reftree_compare(const void *x1, const void *x2)
56*0713e232SGeorge Wilson {
57*0713e232SGeorge Wilson const space_ref_t *sr1 = x1;
58*0713e232SGeorge Wilson const space_ref_t *sr2 = x2;
59*0713e232SGeorge Wilson
60*0713e232SGeorge Wilson if (sr1->sr_offset < sr2->sr_offset)
61*0713e232SGeorge Wilson return (-1);
62*0713e232SGeorge Wilson if (sr1->sr_offset > sr2->sr_offset)
63*0713e232SGeorge Wilson return (1);
64*0713e232SGeorge Wilson
65*0713e232SGeorge Wilson if (sr1 < sr2)
66*0713e232SGeorge Wilson return (-1);
67*0713e232SGeorge Wilson if (sr1 > sr2)
68*0713e232SGeorge Wilson return (1);
69*0713e232SGeorge Wilson
70*0713e232SGeorge Wilson return (0);
71*0713e232SGeorge Wilson }
72*0713e232SGeorge Wilson
73*0713e232SGeorge Wilson void
space_reftree_create(avl_tree_t * t)74*0713e232SGeorge Wilson space_reftree_create(avl_tree_t *t)
75*0713e232SGeorge Wilson {
76*0713e232SGeorge Wilson avl_create(t, space_reftree_compare,
77*0713e232SGeorge Wilson sizeof (space_ref_t), offsetof(space_ref_t, sr_node));
78*0713e232SGeorge Wilson }
79*0713e232SGeorge Wilson
80*0713e232SGeorge Wilson void
space_reftree_destroy(avl_tree_t * t)81*0713e232SGeorge Wilson space_reftree_destroy(avl_tree_t *t)
82*0713e232SGeorge Wilson {
83*0713e232SGeorge Wilson space_ref_t *sr;
84*0713e232SGeorge Wilson void *cookie = NULL;
85*0713e232SGeorge Wilson
86*0713e232SGeorge Wilson while ((sr = avl_destroy_nodes(t, &cookie)) != NULL)
87*0713e232SGeorge Wilson kmem_free(sr, sizeof (*sr));
88*0713e232SGeorge Wilson
89*0713e232SGeorge Wilson avl_destroy(t);
90*0713e232SGeorge Wilson }
91*0713e232SGeorge Wilson
92*0713e232SGeorge Wilson static void
space_reftree_add_node(avl_tree_t * t,uint64_t offset,int64_t refcnt)93*0713e232SGeorge Wilson space_reftree_add_node(avl_tree_t *t, uint64_t offset, int64_t refcnt)
94*0713e232SGeorge Wilson {
95*0713e232SGeorge Wilson space_ref_t *sr;
96*0713e232SGeorge Wilson
97*0713e232SGeorge Wilson sr = kmem_alloc(sizeof (*sr), KM_SLEEP);
98*0713e232SGeorge Wilson sr->sr_offset = offset;
99*0713e232SGeorge Wilson sr->sr_refcnt = refcnt;
100*0713e232SGeorge Wilson
101*0713e232SGeorge Wilson avl_add(t, sr);
102*0713e232SGeorge Wilson }
103*0713e232SGeorge Wilson
104*0713e232SGeorge Wilson void
space_reftree_add_seg(avl_tree_t * t,uint64_t start,uint64_t end,int64_t refcnt)105*0713e232SGeorge Wilson space_reftree_add_seg(avl_tree_t *t, uint64_t start, uint64_t end,
106*0713e232SGeorge Wilson int64_t refcnt)
107*0713e232SGeorge Wilson {
108*0713e232SGeorge Wilson space_reftree_add_node(t, start, refcnt);
109*0713e232SGeorge Wilson space_reftree_add_node(t, end, -refcnt);
110*0713e232SGeorge Wilson }
111*0713e232SGeorge Wilson
112*0713e232SGeorge Wilson /*
113*0713e232SGeorge Wilson * Convert (or add) a range tree into a reference tree.
114*0713e232SGeorge Wilson */
115*0713e232SGeorge Wilson void
space_reftree_add_map(avl_tree_t * t,range_tree_t * rt,int64_t refcnt)116*0713e232SGeorge Wilson space_reftree_add_map(avl_tree_t *t, range_tree_t *rt, int64_t refcnt)
117*0713e232SGeorge Wilson {
118*0713e232SGeorge Wilson range_seg_t *rs;
119*0713e232SGeorge Wilson
120*0713e232SGeorge Wilson ASSERT(MUTEX_HELD(rt->rt_lock));
121*0713e232SGeorge Wilson
122*0713e232SGeorge Wilson for (rs = avl_first(&rt->rt_root); rs; rs = AVL_NEXT(&rt->rt_root, rs))
123*0713e232SGeorge Wilson space_reftree_add_seg(t, rs->rs_start, rs->rs_end, refcnt);
124*0713e232SGeorge Wilson }
125*0713e232SGeorge Wilson
126*0713e232SGeorge Wilson /*
127*0713e232SGeorge Wilson * Convert a reference tree into a range tree. The range tree will contain
128*0713e232SGeorge Wilson * all members of the reference tree for which refcnt >= minref.
129*0713e232SGeorge Wilson */
130*0713e232SGeorge Wilson void
space_reftree_generate_map(avl_tree_t * t,range_tree_t * rt,int64_t minref)131*0713e232SGeorge Wilson space_reftree_generate_map(avl_tree_t *t, range_tree_t *rt, int64_t minref)
132*0713e232SGeorge Wilson {
133*0713e232SGeorge Wilson uint64_t start = -1ULL;
134*0713e232SGeorge Wilson int64_t refcnt = 0;
135*0713e232SGeorge Wilson space_ref_t *sr;
136*0713e232SGeorge Wilson
137*0713e232SGeorge Wilson ASSERT(MUTEX_HELD(rt->rt_lock));
138*0713e232SGeorge Wilson
139*0713e232SGeorge Wilson range_tree_vacate(rt, NULL, NULL);
140*0713e232SGeorge Wilson
141*0713e232SGeorge Wilson for (sr = avl_first(t); sr != NULL; sr = AVL_NEXT(t, sr)) {
142*0713e232SGeorge Wilson refcnt += sr->sr_refcnt;
143*0713e232SGeorge Wilson if (refcnt >= minref) {
144*0713e232SGeorge Wilson if (start == -1ULL) {
145*0713e232SGeorge Wilson start = sr->sr_offset;
146*0713e232SGeorge Wilson }
147*0713e232SGeorge Wilson } else {
148*0713e232SGeorge Wilson if (start != -1ULL) {
149*0713e232SGeorge Wilson uint64_t end = sr->sr_offset;
150*0713e232SGeorge Wilson ASSERT(start <= end);
151*0713e232SGeorge Wilson if (end > start)
152*0713e232SGeorge Wilson range_tree_add(rt, start, end - start);
153*0713e232SGeorge Wilson start = -1ULL;
154*0713e232SGeorge Wilson }
155*0713e232SGeorge Wilson }
156*0713e232SGeorge Wilson }
157*0713e232SGeorge Wilson ASSERT(refcnt == 0);
158*0713e232SGeorge Wilson ASSERT(start == -1ULL);
159*0713e232SGeorge Wilson }
160