xref: /openbmc/linux/fs/ceph/mdsmap.c (revision 0883c2c0)
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 	u8 mdsmap_v, mdsmap_cv;
58 
59 	m = kzalloc(sizeof(*m), GFP_NOFS);
60 	if (m == NULL)
61 		return ERR_PTR(-ENOMEM);
62 
63 	ceph_decode_need(p, end, 1 + 1, bad);
64 	mdsmap_v = ceph_decode_8(p);
65 	mdsmap_cv = ceph_decode_8(p);
66 	if (mdsmap_v >= 4) {
67 	       u32 mdsmap_len;
68 	       ceph_decode_32_safe(p, end, mdsmap_len, bad);
69 	       if (end < *p + mdsmap_len)
70 		       goto bad;
71 	       end = *p + mdsmap_len;
72 	}
73 
74 	ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
75 	m->m_epoch = ceph_decode_32(p);
76 	m->m_client_epoch = ceph_decode_32(p);
77 	m->m_last_failure = ceph_decode_32(p);
78 	m->m_root = ceph_decode_32(p);
79 	m->m_session_timeout = ceph_decode_32(p);
80 	m->m_session_autoclose = ceph_decode_32(p);
81 	m->m_max_file_size = ceph_decode_64(p);
82 	m->m_max_mds = ceph_decode_32(p);
83 
84 	m->m_info = kcalloc(m->m_max_mds, sizeof(*m->m_info), GFP_NOFS);
85 	if (m->m_info == NULL)
86 		goto badmem;
87 
88 	/* pick out active nodes from mds_info (state > 0) */
89 	n = ceph_decode_32(p);
90 	for (i = 0; i < n; i++) {
91 		u64 global_id;
92 		u32 namelen;
93 		s32 mds, inc, state;
94 		u64 state_seq;
95 		u8 info_v;
96 		void *info_end = NULL;
97 		struct ceph_entity_addr addr;
98 		u32 num_export_targets;
99 		void *pexport_targets = NULL;
100 		struct ceph_timespec laggy_since;
101 		struct ceph_mds_info *info;
102 
103 		ceph_decode_need(p, end, sizeof(u64) + 1, bad);
104 		global_id = ceph_decode_64(p);
105 		info_v= ceph_decode_8(p);
106 		if (info_v >= 4) {
107 			u32 info_len;
108 			u8 info_cv;
109 			ceph_decode_need(p, end, 1 + sizeof(u32), bad);
110 			info_cv = ceph_decode_8(p);
111 			info_len = ceph_decode_32(p);
112 			info_end = *p + info_len;
113 			if (info_end > end)
114 				goto bad;
115 		}
116 
117 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
118 		*p += sizeof(u64);
119 		namelen = ceph_decode_32(p);  /* skip mds name */
120 		*p += namelen;
121 
122 		ceph_decode_need(p, end,
123 				 4*sizeof(u32) + sizeof(u64) +
124 				 sizeof(addr) + sizeof(struct ceph_timespec),
125 				 bad);
126 		mds = ceph_decode_32(p);
127 		inc = ceph_decode_32(p);
128 		state = ceph_decode_32(p);
129 		state_seq = ceph_decode_64(p);
130 		ceph_decode_copy(p, &addr, sizeof(addr));
131 		ceph_decode_addr(&addr);
132 		ceph_decode_copy(p, &laggy_since, sizeof(laggy_since));
133 		*p += sizeof(u32);
134 		ceph_decode_32_safe(p, end, namelen, bad);
135 		*p += namelen;
136 		if (info_v >= 2) {
137 			ceph_decode_32_safe(p, end, num_export_targets, bad);
138 			pexport_targets = *p;
139 			*p += num_export_targets * sizeof(u32);
140 		} else {
141 			num_export_targets = 0;
142 		}
143 
144 		if (info_end && *p != info_end) {
145 			if (*p > info_end)
146 				goto bad;
147 			*p = info_end;
148 		}
149 
150 		dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n",
151 		     i+1, n, global_id, mds, inc,
152 		     ceph_pr_addr(&addr.in_addr),
153 		     ceph_mds_state_name(state));
154 
155 		if (mds < 0 || mds >= m->m_max_mds || state <= 0)
156 			continue;
157 
158 		info = &m->m_info[mds];
159 		info->global_id = global_id;
160 		info->state = state;
161 		info->addr = addr;
162 		info->laggy = (laggy_since.tv_sec != 0 ||
163 			       laggy_since.tv_nsec != 0);
164 		info->num_export_targets = num_export_targets;
165 		if (num_export_targets) {
166 			info->export_targets = kcalloc(num_export_targets,
167 						       sizeof(u32), GFP_NOFS);
168 			if (info->export_targets == NULL)
169 				goto badmem;
170 			for (j = 0; j < num_export_targets; j++)
171 				info->export_targets[j] =
172 				       ceph_decode_32(&pexport_targets);
173 		} else {
174 			info->export_targets = NULL;
175 		}
176 	}
177 
178 	/* pg_pools */
179 	ceph_decode_32_safe(p, end, n, bad);
180 	m->m_num_data_pg_pools = n;
181 	m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS);
182 	if (!m->m_data_pg_pools)
183 		goto badmem;
184 	ceph_decode_need(p, end, sizeof(u64)*(n+1), bad);
185 	for (i = 0; i < n; i++)
186 		m->m_data_pg_pools[i] = ceph_decode_64(p);
187 	m->m_cas_pg_pool = ceph_decode_64(p);
188 
189 	/* ok, we don't care about the rest. */
190 	*p = end;
191 	dout("mdsmap_decode success epoch %u\n", m->m_epoch);
192 	return m;
193 
194 badmem:
195 	err = -ENOMEM;
196 bad:
197 	pr_err("corrupt mdsmap\n");
198 	print_hex_dump(KERN_DEBUG, "mdsmap: ",
199 		       DUMP_PREFIX_OFFSET, 16, 1,
200 		       start, end - start, true);
201 	ceph_mdsmap_destroy(m);
202 	return ERR_PTR(err);
203 }
204 
205 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
206 {
207 	int i;
208 
209 	for (i = 0; i < m->m_max_mds; i++)
210 		kfree(m->m_info[i].export_targets);
211 	kfree(m->m_info);
212 	kfree(m->m_data_pg_pools);
213 	kfree(m);
214 }
215