ceph: add request to the tail of unsafe write list
[linux-2.6.git] / fs / ceph / mdsmap.c
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         char r;
24
25         /* count */
26         for (i = 0; i < m->m_max_mds; i++)
27                 if (m->m_info[i].state > 0)
28                         n++;
29         if (n == 0)
30                 return -1;
31
32         /* pick */
33         get_random_bytes(&r, 1);
34         n = r % n;
35         i = 0;
36         for (i = 0; n > 0; i++, n--)
37                 while (m->m_info[i].state <= 0)
38                         i++;
39
40         return i;
41 }
42
43 /*
44  * Decode an MDS map
45  *
46  * Ignore any fields we don't care about (there are quite a few of
47  * them).
48  */
49 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
50 {
51         struct ceph_mdsmap *m;
52         const void *start = *p;
53         int i, j, n;
54         int err = -EINVAL;
55         u16 version;
56
57         m = kzalloc(sizeof(*m), GFP_NOFS);
58         if (m == NULL)
59                 return ERR_PTR(-ENOMEM);
60
61         ceph_decode_16_safe(p, end, version, bad);
62
63         ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
64         m->m_epoch = ceph_decode_32(p);
65         m->m_client_epoch = ceph_decode_32(p);
66         m->m_last_failure = ceph_decode_32(p);
67         m->m_root = ceph_decode_32(p);
68         m->m_session_timeout = ceph_decode_32(p);
69         m->m_session_autoclose = ceph_decode_32(p);
70         m->m_max_file_size = ceph_decode_64(p);
71         m->m_max_mds = ceph_decode_32(p);
72
73         m->m_info = kcalloc(m->m_max_mds, sizeof(*m->m_info), GFP_NOFS);
74         if (m->m_info == NULL)
75                 goto badmem;
76
77         /* pick out active nodes from mds_info (state > 0) */
78         n = ceph_decode_32(p);
79         for (i = 0; i < n; i++) {
80                 u64 global_id;
81                 u32 namelen;
82                 s32 mds, inc, state;
83                 u64 state_seq;
84                 u8 infoversion;
85                 struct ceph_entity_addr addr;
86                 u32 num_export_targets;
87                 void *pexport_targets = NULL;
88                 struct ceph_timespec laggy_since;
89
90                 ceph_decode_need(p, end, sizeof(u64)*2 + 1 + sizeof(u32), bad);
91                 global_id = ceph_decode_64(p);
92                 infoversion = ceph_decode_8(p);
93                 *p += sizeof(u64);
94                 namelen = ceph_decode_32(p);  /* skip mds name */
95                 *p += namelen;
96
97                 ceph_decode_need(p, end,
98                                  4*sizeof(u32) + sizeof(u64) +
99                                  sizeof(addr) + sizeof(struct ceph_timespec),
100                                  bad);
101                 mds = ceph_decode_32(p);
102                 inc = ceph_decode_32(p);
103                 state = ceph_decode_32(p);
104                 state_seq = ceph_decode_64(p);
105                 ceph_decode_copy(p, &addr, sizeof(addr));
106                 ceph_decode_addr(&addr);
107                 ceph_decode_copy(p, &laggy_since, sizeof(laggy_since));
108                 *p += sizeof(u32);
109                 ceph_decode_32_safe(p, end, namelen, bad);
110                 *p += namelen;
111                 if (infoversion >= 2) {
112                         ceph_decode_32_safe(p, end, num_export_targets, bad);
113                         pexport_targets = *p;
114                         *p += num_export_targets * sizeof(u32);
115                 } else {
116                         num_export_targets = 0;
117                 }
118
119                 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n",
120                      i+1, n, global_id, mds, inc,
121                      ceph_pr_addr(&addr.in_addr),
122                      ceph_mds_state_name(state));
123                 if (mds >= 0 && mds < m->m_max_mds && state > 0) {
124                         m->m_info[mds].global_id = global_id;
125                         m->m_info[mds].state = state;
126                         m->m_info[mds].addr = addr;
127                         m->m_info[mds].laggy =
128                                 (laggy_since.tv_sec != 0 ||
129                                  laggy_since.tv_nsec != 0);
130                         m->m_info[mds].num_export_targets = num_export_targets;
131                         if (num_export_targets) {
132                                 m->m_info[mds].export_targets =
133                                         kcalloc(num_export_targets, sizeof(u32),
134                                                 GFP_NOFS);
135                                 for (j = 0; j < num_export_targets; j++)
136                                         m->m_info[mds].export_targets[j] =
137                                                ceph_decode_32(&pexport_targets);
138                         } else {
139                                 m->m_info[mds].export_targets = NULL;
140                         }
141                 }
142         }
143
144         /* pg_pools */
145         ceph_decode_32_safe(p, end, n, bad);
146         m->m_num_data_pg_pools = n;
147         m->m_data_pg_pools = kcalloc(n, sizeof(u32), GFP_NOFS);
148         if (!m->m_data_pg_pools)
149                 goto badmem;
150         ceph_decode_need(p, end, sizeof(u32)*(n+1), bad);
151         for (i = 0; i < n; i++)
152                 m->m_data_pg_pools[i] = ceph_decode_32(p);
153         m->m_cas_pg_pool = ceph_decode_32(p);
154
155         /* ok, we don't care about the rest. */
156         dout("mdsmap_decode success epoch %u\n", m->m_epoch);
157         return m;
158
159 badmem:
160         err = -ENOMEM;
161 bad:
162         pr_err("corrupt mdsmap\n");
163         print_hex_dump(KERN_DEBUG, "mdsmap: ",
164                        DUMP_PREFIX_OFFSET, 16, 1,
165                        start, end - start, true);
166         ceph_mdsmap_destroy(m);
167         return ERR_PTR(-EINVAL);
168 }
169
170 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
171 {
172         int i;
173
174         for (i = 0; i < m->m_max_mds; i++)
175                 kfree(m->m_info[i].export_targets);
176         kfree(m->m_info);
177         kfree(m->m_data_pg_pools);
178         kfree(m);
179 }