1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/bug.h> 5 #include <linux/err.h> 6 #include <linux/random.h> 7 #include <linux/slab.h> 8 #include <linux/types.h> 9 10 #include <linux/ceph/mdsmap.h> 11 #include <linux/ceph/messenger.h> 12 #include <linux/ceph/decode.h> 13 14 #include "super.h" 15 16 #define CEPH_MDS_IS_READY(i, ignore_laggy) \ 17 (m->m_info[i].state > 0 && ignore_laggy ? true : !m->m_info[i].laggy) 18 19 static int __mdsmap_get_random_mds(struct ceph_mdsmap *m, bool ignore_laggy) 20 { 21 int n = 0; 22 int i, j; 23 24 /* count */ 25 for (i = 0; i < m->possible_max_rank; i++) 26 if (CEPH_MDS_IS_READY(i, ignore_laggy)) 27 n++; 28 if (n == 0) 29 return -1; 30 31 /* pick */ 32 n = prandom_u32() % n; 33 for (j = 0, i = 0; i < m->possible_max_rank; i++) { 34 if (CEPH_MDS_IS_READY(i, ignore_laggy)) 35 j++; 36 if (j > n) 37 break; 38 } 39 40 return i; 41 } 42 43 /* 44 * choose a random mds that is "up" (i.e. has a state > 0), or -1. 45 */ 46 int ceph_mdsmap_get_random_mds(struct ceph_mdsmap *m) 47 { 48 int mds; 49 50 mds = __mdsmap_get_random_mds(m, false); 51 if (mds == m->possible_max_rank || mds == -1) 52 mds = __mdsmap_get_random_mds(m, true); 53 54 return mds == m->possible_max_rank ? -1 : mds; 55 } 56 57 #define __decode_and_drop_type(p, end, type, bad) \ 58 do { \ 59 if (*p + sizeof(type) > end) \ 60 goto bad; \ 61 *p += sizeof(type); \ 62 } while (0) 63 64 #define __decode_and_drop_set(p, end, type, bad) \ 65 do { \ 66 u32 n; \ 67 size_t need; \ 68 ceph_decode_32_safe(p, end, n, bad); \ 69 need = sizeof(type) * n; \ 70 ceph_decode_need(p, end, need, bad); \ 71 *p += need; \ 72 } while (0) 73 74 #define __decode_and_drop_map(p, end, ktype, vtype, bad) \ 75 do { \ 76 u32 n; \ 77 size_t need; \ 78 ceph_decode_32_safe(p, end, n, bad); \ 79 need = (sizeof(ktype) + sizeof(vtype)) * n; \ 80 ceph_decode_need(p, end, need, bad); \ 81 *p += need; \ 82 } while (0) 83 84 85 static int __decode_and_drop_compat_set(void **p, void* end) 86 { 87 int i; 88 /* compat, ro_compat, incompat*/ 89 for (i = 0; i < 3; i++) { 90 u32 n; 91 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); 92 /* mask */ 93 *p += sizeof(u64); 94 /* names (map<u64, string>) */ 95 n = ceph_decode_32(p); 96 while (n-- > 0) { 97 u32 len; 98 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), 99 bad); 100 *p += sizeof(u64); 101 len = ceph_decode_32(p); 102 ceph_decode_need(p, end, len, bad); 103 *p += len; 104 } 105 } 106 return 0; 107 bad: 108 return -1; 109 } 110 111 /* 112 * Decode an MDS map 113 * 114 * Ignore any fields we don't care about (there are quite a few of 115 * them). 116 */ 117 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end, bool msgr2) 118 { 119 struct ceph_mdsmap *m; 120 const void *start = *p; 121 int i, j, n; 122 int err; 123 u8 mdsmap_v; 124 u16 mdsmap_ev; 125 126 m = kzalloc(sizeof(*m), GFP_NOFS); 127 if (!m) 128 return ERR_PTR(-ENOMEM); 129 130 ceph_decode_need(p, end, 1 + 1, bad); 131 mdsmap_v = ceph_decode_8(p); 132 *p += sizeof(u8); /* mdsmap_cv */ 133 if (mdsmap_v >= 4) { 134 u32 mdsmap_len; 135 ceph_decode_32_safe(p, end, mdsmap_len, bad); 136 if (end < *p + mdsmap_len) 137 goto bad; 138 end = *p + mdsmap_len; 139 } 140 141 ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad); 142 m->m_epoch = ceph_decode_32(p); 143 m->m_client_epoch = ceph_decode_32(p); 144 m->m_last_failure = ceph_decode_32(p); 145 m->m_root = ceph_decode_32(p); 146 m->m_session_timeout = ceph_decode_32(p); 147 m->m_session_autoclose = ceph_decode_32(p); 148 m->m_max_file_size = ceph_decode_64(p); 149 m->m_max_mds = ceph_decode_32(p); 150 151 /* 152 * pick out the active nodes as the m_num_active_mds, the 153 * m_num_active_mds maybe larger than m_max_mds when decreasing 154 * the max_mds in cluster side, in other case it should less 155 * than or equal to m_max_mds. 156 */ 157 m->m_num_active_mds = n = ceph_decode_32(p); 158 159 /* 160 * the possible max rank, it maybe larger than the m_num_active_mds, 161 * for example if the mds_max == 2 in the cluster, when the MDS(0) 162 * was laggy and being replaced by a new MDS, we will temporarily 163 * receive a new mds map with n_num_mds == 1 and the active MDS(1), 164 * and the mds rank >= m_num_active_mds. 165 */ 166 m->possible_max_rank = max(m->m_num_active_mds, m->m_max_mds); 167 168 m->m_info = kcalloc(m->possible_max_rank, sizeof(*m->m_info), GFP_NOFS); 169 if (!m->m_info) 170 goto nomem; 171 172 /* pick out active nodes from mds_info (state > 0) */ 173 for (i = 0; i < n; i++) { 174 u64 global_id; 175 u32 namelen; 176 s32 mds, inc, state; 177 u8 info_v; 178 void *info_end = NULL; 179 struct ceph_entity_addr addr; 180 u32 num_export_targets; 181 void *pexport_targets = NULL; 182 struct ceph_timespec laggy_since; 183 struct ceph_mds_info *info; 184 bool laggy; 185 186 ceph_decode_need(p, end, sizeof(u64) + 1, bad); 187 global_id = ceph_decode_64(p); 188 info_v= ceph_decode_8(p); 189 if (info_v >= 4) { 190 u32 info_len; 191 ceph_decode_need(p, end, 1 + sizeof(u32), bad); 192 *p += sizeof(u8); /* info_cv */ 193 info_len = ceph_decode_32(p); 194 info_end = *p + info_len; 195 if (info_end > end) 196 goto bad; 197 } 198 199 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); 200 *p += sizeof(u64); 201 namelen = ceph_decode_32(p); /* skip mds name */ 202 *p += namelen; 203 204 ceph_decode_32_safe(p, end, mds, bad); 205 ceph_decode_32_safe(p, end, inc, bad); 206 ceph_decode_32_safe(p, end, state, bad); 207 *p += sizeof(u64); /* state_seq */ 208 if (info_v >= 8) 209 err = ceph_decode_entity_addrvec(p, end, msgr2, &addr); 210 else 211 err = ceph_decode_entity_addr(p, end, &addr); 212 if (err) 213 goto corrupt; 214 215 ceph_decode_copy_safe(p, end, &laggy_since, sizeof(laggy_since), 216 bad); 217 laggy = laggy_since.tv_sec != 0 || laggy_since.tv_nsec != 0; 218 *p += sizeof(u32); 219 ceph_decode_32_safe(p, end, namelen, bad); 220 *p += namelen; 221 if (info_v >= 2) { 222 ceph_decode_32_safe(p, end, num_export_targets, bad); 223 pexport_targets = *p; 224 *p += num_export_targets * sizeof(u32); 225 } else { 226 num_export_targets = 0; 227 } 228 229 if (info_end && *p != info_end) { 230 if (*p > info_end) 231 goto bad; 232 *p = info_end; 233 } 234 235 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s%s\n", 236 i+1, n, global_id, mds, inc, 237 ceph_pr_addr(&addr), 238 ceph_mds_state_name(state), 239 laggy ? "(laggy)" : ""); 240 241 if (mds < 0 || mds >= m->possible_max_rank) { 242 pr_warn("mdsmap_decode got incorrect mds(%d)\n", mds); 243 continue; 244 } 245 246 if (state <= 0) { 247 dout("mdsmap_decode got incorrect state(%s)\n", 248 ceph_mds_state_name(state)); 249 continue; 250 } 251 252 info = &m->m_info[mds]; 253 info->global_id = global_id; 254 info->state = state; 255 info->addr = addr; 256 info->laggy = laggy; 257 info->num_export_targets = num_export_targets; 258 if (num_export_targets) { 259 info->export_targets = kcalloc(num_export_targets, 260 sizeof(u32), GFP_NOFS); 261 if (!info->export_targets) 262 goto nomem; 263 for (j = 0; j < num_export_targets; j++) 264 info->export_targets[j] = 265 ceph_decode_32(&pexport_targets); 266 } else { 267 info->export_targets = NULL; 268 } 269 } 270 271 /* pg_pools */ 272 ceph_decode_32_safe(p, end, n, bad); 273 m->m_num_data_pg_pools = n; 274 m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS); 275 if (!m->m_data_pg_pools) 276 goto nomem; 277 ceph_decode_need(p, end, sizeof(u64)*(n+1), bad); 278 for (i = 0; i < n; i++) 279 m->m_data_pg_pools[i] = ceph_decode_64(p); 280 m->m_cas_pg_pool = ceph_decode_64(p); 281 m->m_enabled = m->m_epoch > 1; 282 283 mdsmap_ev = 1; 284 if (mdsmap_v >= 2) { 285 ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext); 286 } 287 if (mdsmap_ev >= 3) { 288 if (__decode_and_drop_compat_set(p, end) < 0) 289 goto bad_ext; 290 } 291 /* metadata_pool */ 292 if (mdsmap_ev < 5) { 293 __decode_and_drop_type(p, end, u32, bad_ext); 294 } else { 295 __decode_and_drop_type(p, end, u64, bad_ext); 296 } 297 298 /* created + modified + tableserver */ 299 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 300 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 301 __decode_and_drop_type(p, end, u32, bad_ext); 302 303 /* in */ 304 { 305 int num_laggy = 0; 306 ceph_decode_32_safe(p, end, n, bad_ext); 307 ceph_decode_need(p, end, sizeof(u32) * n, bad_ext); 308 309 for (i = 0; i < n; i++) { 310 s32 mds = ceph_decode_32(p); 311 if (mds >= 0 && mds < m->possible_max_rank) { 312 if (m->m_info[mds].laggy) 313 num_laggy++; 314 } 315 } 316 m->m_num_laggy = num_laggy; 317 318 if (n > m->possible_max_rank) { 319 void *new_m_info = krealloc(m->m_info, 320 n * sizeof(*m->m_info), 321 GFP_NOFS | __GFP_ZERO); 322 if (!new_m_info) 323 goto nomem; 324 m->m_info = new_m_info; 325 } 326 m->possible_max_rank = n; 327 } 328 329 /* inc */ 330 __decode_and_drop_map(p, end, u32, u32, bad_ext); 331 /* up */ 332 __decode_and_drop_map(p, end, u32, u64, bad_ext); 333 /* failed */ 334 __decode_and_drop_set(p, end, u32, bad_ext); 335 /* stopped */ 336 __decode_and_drop_set(p, end, u32, bad_ext); 337 338 if (mdsmap_ev >= 4) { 339 /* last_failure_osd_epoch */ 340 __decode_and_drop_type(p, end, u32, bad_ext); 341 } 342 if (mdsmap_ev >= 6) { 343 /* ever_allowed_snaps */ 344 __decode_and_drop_type(p, end, u8, bad_ext); 345 /* explicitly_allowed_snaps */ 346 __decode_and_drop_type(p, end, u8, bad_ext); 347 } 348 if (mdsmap_ev >= 7) { 349 /* inline_data_enabled */ 350 __decode_and_drop_type(p, end, u8, bad_ext); 351 } 352 if (mdsmap_ev >= 8) { 353 u32 name_len; 354 /* enabled */ 355 ceph_decode_8_safe(p, end, m->m_enabled, bad_ext); 356 ceph_decode_32_safe(p, end, name_len, bad_ext); 357 ceph_decode_need(p, end, name_len, bad_ext); 358 *p += name_len; 359 } 360 /* damaged */ 361 if (mdsmap_ev >= 9) { 362 size_t need; 363 ceph_decode_32_safe(p, end, n, bad_ext); 364 need = sizeof(u32) * n; 365 ceph_decode_need(p, end, need, bad_ext); 366 *p += need; 367 m->m_damaged = n > 0; 368 } else { 369 m->m_damaged = false; 370 } 371 bad_ext: 372 dout("mdsmap_decode m_enabled: %d, m_damaged: %d, m_num_laggy: %d\n", 373 !!m->m_enabled, !!m->m_damaged, m->m_num_laggy); 374 *p = end; 375 dout("mdsmap_decode success epoch %u\n", m->m_epoch); 376 return m; 377 nomem: 378 err = -ENOMEM; 379 goto out_err; 380 corrupt: 381 pr_err("corrupt mdsmap\n"); 382 print_hex_dump(KERN_DEBUG, "mdsmap: ", 383 DUMP_PREFIX_OFFSET, 16, 1, 384 start, end - start, true); 385 out_err: 386 ceph_mdsmap_destroy(m); 387 return ERR_PTR(err); 388 bad: 389 err = -EINVAL; 390 goto corrupt; 391 } 392 393 void ceph_mdsmap_destroy(struct ceph_mdsmap *m) 394 { 395 int i; 396 397 if (m->m_info) { 398 for (i = 0; i < m->possible_max_rank; i++) 399 kfree(m->m_info[i].export_targets); 400 kfree(m->m_info); 401 } 402 kfree(m->m_data_pg_pools); 403 kfree(m); 404 } 405 406 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m) 407 { 408 int i, nr_active = 0; 409 if (!m->m_enabled) 410 return false; 411 if (m->m_damaged) 412 return false; 413 if (m->m_num_laggy == m->m_num_active_mds) 414 return false; 415 for (i = 0; i < m->possible_max_rank; i++) { 416 if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE) 417 nr_active++; 418 } 419 return nr_active > 0; 420 } 421