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_num_mds && m->m_info[0].state > 0) 26 return 0; 27 28 /* count */ 29 for (i = 0; i < m->m_num_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 #define __decode_and_drop_type(p, end, type, bad) \ 46 do { \ 47 if (*p + sizeof(type) > end) \ 48 goto bad; \ 49 *p += sizeof(type); \ 50 } while (0) 51 52 #define __decode_and_drop_set(p, end, type, bad) \ 53 do { \ 54 u32 n; \ 55 size_t need; \ 56 ceph_decode_32_safe(p, end, n, bad); \ 57 need = sizeof(type) * n; \ 58 ceph_decode_need(p, end, need, bad); \ 59 *p += need; \ 60 } while (0) 61 62 #define __decode_and_drop_map(p, end, ktype, vtype, bad) \ 63 do { \ 64 u32 n; \ 65 size_t need; \ 66 ceph_decode_32_safe(p, end, n, bad); \ 67 need = (sizeof(ktype) + sizeof(vtype)) * n; \ 68 ceph_decode_need(p, end, need, bad); \ 69 *p += need; \ 70 } while (0) 71 72 73 static int __decode_and_drop_compat_set(void **p, void* end) 74 { 75 int i; 76 /* compat, ro_compat, incompat*/ 77 for (i = 0; i < 3; i++) { 78 u32 n; 79 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); 80 /* mask */ 81 *p += sizeof(u64); 82 /* names (map<u64, string>) */ 83 n = ceph_decode_32(p); 84 while (n-- > 0) { 85 u32 len; 86 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), 87 bad); 88 *p += sizeof(u64); 89 len = ceph_decode_32(p); 90 ceph_decode_need(p, end, len, bad); 91 *p += len; 92 } 93 } 94 return 0; 95 bad: 96 return -1; 97 } 98 99 /* 100 * Decode an MDS map 101 * 102 * Ignore any fields we don't care about (there are quite a few of 103 * them). 104 */ 105 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end) 106 { 107 struct ceph_mdsmap *m; 108 const void *start = *p; 109 int i, j, n; 110 int err = -EINVAL; 111 u8 mdsmap_v, mdsmap_cv; 112 u16 mdsmap_ev; 113 114 m = kzalloc(sizeof(*m), GFP_NOFS); 115 if (!m) 116 return ERR_PTR(-ENOMEM); 117 118 ceph_decode_need(p, end, 1 + 1, bad); 119 mdsmap_v = ceph_decode_8(p); 120 mdsmap_cv = ceph_decode_8(p); 121 if (mdsmap_v >= 4) { 122 u32 mdsmap_len; 123 ceph_decode_32_safe(p, end, mdsmap_len, bad); 124 if (end < *p + mdsmap_len) 125 goto bad; 126 end = *p + mdsmap_len; 127 } 128 129 ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad); 130 m->m_epoch = ceph_decode_32(p); 131 m->m_client_epoch = ceph_decode_32(p); 132 m->m_last_failure = ceph_decode_32(p); 133 m->m_root = ceph_decode_32(p); 134 m->m_session_timeout = ceph_decode_32(p); 135 m->m_session_autoclose = ceph_decode_32(p); 136 m->m_max_file_size = ceph_decode_64(p); 137 m->m_max_mds = ceph_decode_32(p); 138 m->m_num_mds = m->m_max_mds; 139 140 m->m_info = kcalloc(m->m_num_mds, sizeof(*m->m_info), GFP_NOFS); 141 if (!m->m_info) 142 goto nomem; 143 144 /* pick out active nodes from mds_info (state > 0) */ 145 n = ceph_decode_32(p); 146 for (i = 0; i < n; i++) { 147 u64 global_id; 148 u32 namelen; 149 s32 mds, inc, state; 150 u64 state_seq; 151 u8 info_v; 152 void *info_end = NULL; 153 struct ceph_entity_addr addr; 154 u32 num_export_targets; 155 void *pexport_targets = NULL; 156 struct ceph_timespec laggy_since; 157 struct ceph_mds_info *info; 158 159 ceph_decode_need(p, end, sizeof(u64) + 1, bad); 160 global_id = ceph_decode_64(p); 161 info_v= ceph_decode_8(p); 162 if (info_v >= 4) { 163 u32 info_len; 164 u8 info_cv; 165 ceph_decode_need(p, end, 1 + sizeof(u32), bad); 166 info_cv = ceph_decode_8(p); 167 info_len = ceph_decode_32(p); 168 info_end = *p + info_len; 169 if (info_end > end) 170 goto bad; 171 } 172 173 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); 174 *p += sizeof(u64); 175 namelen = ceph_decode_32(p); /* skip mds name */ 176 *p += namelen; 177 178 ceph_decode_need(p, end, 179 4*sizeof(u32) + sizeof(u64) + 180 sizeof(addr) + sizeof(struct ceph_timespec), 181 bad); 182 mds = ceph_decode_32(p); 183 inc = ceph_decode_32(p); 184 state = ceph_decode_32(p); 185 state_seq = ceph_decode_64(p); 186 ceph_decode_copy(p, &addr, sizeof(addr)); 187 ceph_decode_addr(&addr); 188 ceph_decode_copy(p, &laggy_since, sizeof(laggy_since)); 189 *p += sizeof(u32); 190 ceph_decode_32_safe(p, end, namelen, bad); 191 *p += namelen; 192 if (info_v >= 2) { 193 ceph_decode_32_safe(p, end, num_export_targets, bad); 194 pexport_targets = *p; 195 *p += num_export_targets * sizeof(u32); 196 } else { 197 num_export_targets = 0; 198 } 199 200 if (info_end && *p != info_end) { 201 if (*p > info_end) 202 goto bad; 203 *p = info_end; 204 } 205 206 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n", 207 i+1, n, global_id, mds, inc, 208 ceph_pr_addr(&addr.in_addr), 209 ceph_mds_state_name(state)); 210 211 if (mds < 0 || state <= 0) 212 continue; 213 214 if (mds >= m->m_num_mds) { 215 int new_num = max(mds + 1, m->m_num_mds * 2); 216 void *new_m_info = krealloc(m->m_info, 217 new_num * sizeof(*m->m_info), 218 GFP_NOFS | __GFP_ZERO); 219 if (!new_m_info) 220 goto nomem; 221 m->m_info = new_m_info; 222 m->m_num_mds = new_num; 223 } 224 225 info = &m->m_info[mds]; 226 info->global_id = global_id; 227 info->state = state; 228 info->addr = addr; 229 info->laggy = (laggy_since.tv_sec != 0 || 230 laggy_since.tv_nsec != 0); 231 info->num_export_targets = num_export_targets; 232 if (num_export_targets) { 233 info->export_targets = kcalloc(num_export_targets, 234 sizeof(u32), GFP_NOFS); 235 if (!info->export_targets) 236 goto nomem; 237 for (j = 0; j < num_export_targets; j++) 238 info->export_targets[j] = 239 ceph_decode_32(&pexport_targets); 240 } else { 241 info->export_targets = NULL; 242 } 243 } 244 if (m->m_num_mds > m->m_max_mds) { 245 /* find max up mds */ 246 for (i = m->m_num_mds; i >= m->m_max_mds; i--) { 247 if (i == 0 || m->m_info[i-1].state > 0) 248 break; 249 } 250 m->m_num_mds = i; 251 } 252 253 /* pg_pools */ 254 ceph_decode_32_safe(p, end, n, bad); 255 m->m_num_data_pg_pools = n; 256 m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS); 257 if (!m->m_data_pg_pools) 258 goto nomem; 259 ceph_decode_need(p, end, sizeof(u64)*(n+1), bad); 260 for (i = 0; i < n; i++) 261 m->m_data_pg_pools[i] = ceph_decode_64(p); 262 m->m_cas_pg_pool = ceph_decode_64(p); 263 m->m_enabled = m->m_epoch > 1; 264 265 mdsmap_ev = 1; 266 if (mdsmap_v >= 2) { 267 ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext); 268 } 269 if (mdsmap_ev >= 3) { 270 if (__decode_and_drop_compat_set(p, end) < 0) 271 goto bad_ext; 272 } 273 /* metadata_pool */ 274 if (mdsmap_ev < 5) { 275 __decode_and_drop_type(p, end, u32, bad_ext); 276 } else { 277 __decode_and_drop_type(p, end, u64, bad_ext); 278 } 279 280 /* created + modified + tableserver */ 281 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 282 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 283 __decode_and_drop_type(p, end, u32, bad_ext); 284 285 /* in */ 286 { 287 int num_laggy = 0; 288 ceph_decode_32_safe(p, end, n, bad_ext); 289 ceph_decode_need(p, end, sizeof(u32) * n, bad_ext); 290 291 for (i = 0; i < n; i++) { 292 s32 mds = ceph_decode_32(p); 293 if (mds >= 0 && mds < m->m_num_mds) { 294 if (m->m_info[mds].laggy) 295 num_laggy++; 296 } 297 } 298 m->m_num_laggy = num_laggy; 299 300 if (n > m->m_num_mds) { 301 void *new_m_info = krealloc(m->m_info, 302 n * sizeof(*m->m_info), 303 GFP_NOFS | __GFP_ZERO); 304 if (!new_m_info) 305 goto nomem; 306 m->m_info = new_m_info; 307 } 308 m->m_num_mds = n; 309 } 310 311 /* inc */ 312 __decode_and_drop_map(p, end, u32, u32, bad_ext); 313 /* up */ 314 __decode_and_drop_map(p, end, u32, u64, bad_ext); 315 /* failed */ 316 __decode_and_drop_set(p, end, u32, bad_ext); 317 /* stopped */ 318 __decode_and_drop_set(p, end, u32, bad_ext); 319 320 if (mdsmap_ev >= 4) { 321 /* last_failure_osd_epoch */ 322 __decode_and_drop_type(p, end, u32, bad_ext); 323 } 324 if (mdsmap_ev >= 6) { 325 /* ever_allowed_snaps */ 326 __decode_and_drop_type(p, end, u8, bad_ext); 327 /* explicitly_allowed_snaps */ 328 __decode_and_drop_type(p, end, u8, bad_ext); 329 } 330 if (mdsmap_ev >= 7) { 331 /* inline_data_enabled */ 332 __decode_and_drop_type(p, end, u8, bad_ext); 333 } 334 if (mdsmap_ev >= 8) { 335 u32 name_len; 336 /* enabled */ 337 ceph_decode_8_safe(p, end, m->m_enabled, bad_ext); 338 ceph_decode_32_safe(p, end, name_len, bad_ext); 339 ceph_decode_need(p, end, name_len, bad_ext); 340 *p += name_len; 341 } 342 /* damaged */ 343 if (mdsmap_ev >= 9) { 344 size_t need; 345 ceph_decode_32_safe(p, end, n, bad_ext); 346 need = sizeof(u32) * n; 347 ceph_decode_need(p, end, need, bad_ext); 348 *p += need; 349 m->m_damaged = n > 0; 350 } else { 351 m->m_damaged = false; 352 } 353 bad_ext: 354 *p = end; 355 dout("mdsmap_decode success epoch %u\n", m->m_epoch); 356 return m; 357 nomem: 358 err = -ENOMEM; 359 goto out_err; 360 bad: 361 pr_err("corrupt mdsmap\n"); 362 print_hex_dump(KERN_DEBUG, "mdsmap: ", 363 DUMP_PREFIX_OFFSET, 16, 1, 364 start, end - start, true); 365 out_err: 366 ceph_mdsmap_destroy(m); 367 return ERR_PTR(err); 368 } 369 370 void ceph_mdsmap_destroy(struct ceph_mdsmap *m) 371 { 372 int i; 373 374 for (i = 0; i < m->m_num_mds; i++) 375 kfree(m->m_info[i].export_targets); 376 kfree(m->m_info); 377 kfree(m->m_data_pg_pools); 378 kfree(m); 379 } 380 381 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m) 382 { 383 int i, nr_active = 0; 384 if (!m->m_enabled) 385 return false; 386 if (m->m_damaged) 387 return false; 388 if (m->m_num_laggy > 0) 389 return false; 390 for (i = 0; i < m->m_num_mds; i++) { 391 if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE) 392 nr_active++; 393 } 394 return nr_active > 0; 395 } 396