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_warning("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 96 ceph_decode_need(p, end, sizeof(u64)*2 + 1 + sizeof(u32), bad); 97 global_id = ceph_decode_64(p); 98 infoversion = ceph_decode_8(p); 99 *p += sizeof(u64); 100 namelen = ceph_decode_32(p); /* skip mds name */ 101 *p += namelen; 102 103 ceph_decode_need(p, end, 104 4*sizeof(u32) + sizeof(u64) + 105 sizeof(addr) + sizeof(struct ceph_timespec), 106 bad); 107 mds = ceph_decode_32(p); 108 inc = ceph_decode_32(p); 109 state = ceph_decode_32(p); 110 state_seq = ceph_decode_64(p); 111 ceph_decode_copy(p, &addr, sizeof(addr)); 112 ceph_decode_addr(&addr); 113 ceph_decode_copy(p, &laggy_since, sizeof(laggy_since)); 114 *p += sizeof(u32); 115 ceph_decode_32_safe(p, end, namelen, bad); 116 *p += namelen; 117 if (infoversion >= 2) { 118 ceph_decode_32_safe(p, end, num_export_targets, bad); 119 pexport_targets = *p; 120 *p += num_export_targets * sizeof(u32); 121 } else { 122 num_export_targets = 0; 123 } 124 125 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n", 126 i+1, n, global_id, mds, inc, 127 ceph_pr_addr(&addr.in_addr), 128 ceph_mds_state_name(state)); 129 if (mds >= 0 && mds < m->m_max_mds && state > 0) { 130 m->m_info[mds].global_id = global_id; 131 m->m_info[mds].state = state; 132 m->m_info[mds].addr = addr; 133 m->m_info[mds].laggy = 134 (laggy_since.tv_sec != 0 || 135 laggy_since.tv_nsec != 0); 136 m->m_info[mds].num_export_targets = num_export_targets; 137 if (num_export_targets) { 138 m->m_info[mds].export_targets = 139 kcalloc(num_export_targets, sizeof(u32), 140 GFP_NOFS); 141 for (j = 0; j < num_export_targets; j++) 142 m->m_info[mds].export_targets[j] = 143 ceph_decode_32(&pexport_targets); 144 } else { 145 m->m_info[mds].export_targets = NULL; 146 } 147 } 148 } 149 150 /* pg_pools */ 151 ceph_decode_32_safe(p, end, n, bad); 152 m->m_num_data_pg_pools = n; 153 m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS); 154 if (!m->m_data_pg_pools) 155 goto badmem; 156 ceph_decode_need(p, end, sizeof(u64)*(n+1), bad); 157 for (i = 0; i < n; i++) 158 m->m_data_pg_pools[i] = ceph_decode_64(p); 159 m->m_cas_pg_pool = ceph_decode_64(p); 160 161 /* ok, we don't care about the rest. */ 162 dout("mdsmap_decode success epoch %u\n", m->m_epoch); 163 return m; 164 165 badmem: 166 err = -ENOMEM; 167 bad: 168 pr_err("corrupt mdsmap\n"); 169 print_hex_dump(KERN_DEBUG, "mdsmap: ", 170 DUMP_PREFIX_OFFSET, 16, 1, 171 start, end - start, true); 172 ceph_mdsmap_destroy(m); 173 return ERR_PTR(-EINVAL); 174 } 175 176 void ceph_mdsmap_destroy(struct ceph_mdsmap *m) 177 { 178 int i; 179 180 for (i = 0; i < m->m_max_mds; i++) 181 kfree(m->m_info[i].export_targets); 182 kfree(m->m_info); 183 kfree(m->m_data_pg_pools); 184 kfree(m); 185 } 186