xref: /openbmc/linux/fs/ceph/mdsmap.c (revision 30614cf3)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/bug.h>
4 #include <linux/err.h>
5 #include <linux/random.h>
6 #include <linux/slab.h>
7 #include <linux/types.h>
8 
9 #include <linux/ceph/mdsmap.h>
10 #include <linux/ceph/messenger.h>
11 #include <linux/ceph/decode.h>
12 
13 #include "super.h"
14 
15 
16 /*
17  * choose a random mds that is "up" (i.e. has a state > 0), or -1.
18  */
19 int ceph_mdsmap_get_random_mds(struct ceph_mdsmap *m)
20 {
21 	int n = 0;
22 	int i;
23 
24 	/* special case for one mds */
25 	if (1 == m->m_max_mds && m->m_info[0].state > 0)
26 		return 0;
27 
28 	/* count */
29 	for (i = 0; i < m->m_max_mds; i++)
30 		if (m->m_info[i].state > 0)
31 			n++;
32 	if (n == 0)
33 		return -1;
34 
35 	/* pick */
36 	n = prandom_u32() % n;
37 	i = 0;
38 	for (i = 0; n > 0; i++, n--)
39 		while (m->m_info[i].state <= 0)
40 			i++;
41 
42 	return i;
43 }
44 
45 /*
46  * Decode an MDS map
47  *
48  * Ignore any fields we don't care about (there are quite a few of
49  * them).
50  */
51 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
52 {
53 	struct ceph_mdsmap *m;
54 	const void *start = *p;
55 	int i, j, n;
56 	int err = -EINVAL;
57 	u16 version;
58 
59 	m = kzalloc(sizeof(*m), GFP_NOFS);
60 	if (m == NULL)
61 		return ERR_PTR(-ENOMEM);
62 
63 	ceph_decode_16_safe(p, end, version, bad);
64 	if (version > 3) {
65 		pr_warn("got mdsmap version %d > 3, failing", version);
66 		goto bad;
67 	}
68 
69 	ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
70 	m->m_epoch = ceph_decode_32(p);
71 	m->m_client_epoch = ceph_decode_32(p);
72 	m->m_last_failure = ceph_decode_32(p);
73 	m->m_root = ceph_decode_32(p);
74 	m->m_session_timeout = ceph_decode_32(p);
75 	m->m_session_autoclose = ceph_decode_32(p);
76 	m->m_max_file_size = ceph_decode_64(p);
77 	m->m_max_mds = ceph_decode_32(p);
78 
79 	m->m_info = kcalloc(m->m_max_mds, sizeof(*m->m_info), GFP_NOFS);
80 	if (m->m_info == NULL)
81 		goto badmem;
82 
83 	/* pick out active nodes from mds_info (state > 0) */
84 	n = ceph_decode_32(p);
85 	for (i = 0; i < n; i++) {
86 		u64 global_id;
87 		u32 namelen;
88 		s32 mds, inc, state;
89 		u64 state_seq;
90 		u8 infoversion;
91 		struct ceph_entity_addr addr;
92 		u32 num_export_targets;
93 		void *pexport_targets = NULL;
94 		struct ceph_timespec laggy_since;
95 		struct ceph_mds_info *info;
96 
97 		ceph_decode_need(p, end, sizeof(u64)*2 + 1 + sizeof(u32), bad);
98 		global_id = ceph_decode_64(p);
99 		infoversion = ceph_decode_8(p);
100 		*p += sizeof(u64);
101 		namelen = ceph_decode_32(p);  /* skip mds name */
102 		*p += namelen;
103 
104 		ceph_decode_need(p, end,
105 				 4*sizeof(u32) + sizeof(u64) +
106 				 sizeof(addr) + sizeof(struct ceph_timespec),
107 				 bad);
108 		mds = ceph_decode_32(p);
109 		inc = ceph_decode_32(p);
110 		state = ceph_decode_32(p);
111 		state_seq = ceph_decode_64(p);
112 		ceph_decode_copy(p, &addr, sizeof(addr));
113 		ceph_decode_addr(&addr);
114 		ceph_decode_copy(p, &laggy_since, sizeof(laggy_since));
115 		*p += sizeof(u32);
116 		ceph_decode_32_safe(p, end, namelen, bad);
117 		*p += namelen;
118 		if (infoversion >= 2) {
119 			ceph_decode_32_safe(p, end, num_export_targets, bad);
120 			pexport_targets = *p;
121 			*p += num_export_targets * sizeof(u32);
122 		} else {
123 			num_export_targets = 0;
124 		}
125 
126 		dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n",
127 		     i+1, n, global_id, mds, inc,
128 		     ceph_pr_addr(&addr.in_addr),
129 		     ceph_mds_state_name(state));
130 
131 		if (mds < 0 || mds >= m->m_max_mds || state <= 0)
132 			continue;
133 
134 		info = &m->m_info[mds];
135 		info->global_id = global_id;
136 		info->state = state;
137 		info->addr = addr;
138 		info->laggy = (laggy_since.tv_sec != 0 ||
139 			       laggy_since.tv_nsec != 0);
140 		info->num_export_targets = num_export_targets;
141 		if (num_export_targets) {
142 			info->export_targets = kcalloc(num_export_targets,
143 						       sizeof(u32), GFP_NOFS);
144 			if (info->export_targets == NULL)
145 				goto badmem;
146 			for (j = 0; j < num_export_targets; j++)
147 				info->export_targets[j] =
148 				       ceph_decode_32(&pexport_targets);
149 		} else {
150 			info->export_targets = NULL;
151 		}
152 	}
153 
154 	/* pg_pools */
155 	ceph_decode_32_safe(p, end, n, bad);
156 	m->m_num_data_pg_pools = n;
157 	m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS);
158 	if (!m->m_data_pg_pools)
159 		goto badmem;
160 	ceph_decode_need(p, end, sizeof(u64)*(n+1), bad);
161 	for (i = 0; i < n; i++)
162 		m->m_data_pg_pools[i] = ceph_decode_64(p);
163 	m->m_cas_pg_pool = ceph_decode_64(p);
164 
165 	/* ok, we don't care about the rest. */
166 	dout("mdsmap_decode success epoch %u\n", m->m_epoch);
167 	return m;
168 
169 badmem:
170 	err = -ENOMEM;
171 bad:
172 	pr_err("corrupt mdsmap\n");
173 	print_hex_dump(KERN_DEBUG, "mdsmap: ",
174 		       DUMP_PREFIX_OFFSET, 16, 1,
175 		       start, end - start, true);
176 	ceph_mdsmap_destroy(m);
177 	return ERR_PTR(err);
178 }
179 
180 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
181 {
182 	int i;
183 
184 	for (i = 0; i < m->m_max_mds; i++)
185 		kfree(m->m_info[i].export_targets);
186 	kfree(m->m_info);
187 	kfree(m->m_data_pg_pools);
188 	kfree(m);
189 }
190