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 u32 target; 126 127 m = kzalloc(sizeof(*m), GFP_NOFS); 128 if (!m) 129 return ERR_PTR(-ENOMEM); 130 131 ceph_decode_need(p, end, 1 + 1, bad); 132 mdsmap_v = ceph_decode_8(p); 133 *p += sizeof(u8); /* mdsmap_cv */ 134 if (mdsmap_v >= 4) { 135 u32 mdsmap_len; 136 ceph_decode_32_safe(p, end, mdsmap_len, bad); 137 if (end < *p + mdsmap_len) 138 goto bad; 139 end = *p + mdsmap_len; 140 } 141 142 ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad); 143 m->m_epoch = ceph_decode_32(p); 144 m->m_client_epoch = ceph_decode_32(p); 145 m->m_last_failure = ceph_decode_32(p); 146 m->m_root = ceph_decode_32(p); 147 m->m_session_timeout = ceph_decode_32(p); 148 m->m_session_autoclose = ceph_decode_32(p); 149 m->m_max_file_size = ceph_decode_64(p); 150 m->m_max_mds = ceph_decode_32(p); 151 152 /* 153 * pick out the active nodes as the m_num_active_mds, the 154 * m_num_active_mds maybe larger than m_max_mds when decreasing 155 * the max_mds in cluster side, in other case it should less 156 * than or equal to m_max_mds. 157 */ 158 m->m_num_active_mds = n = ceph_decode_32(p); 159 160 /* 161 * the possible max rank, it maybe larger than the m_num_active_mds, 162 * for example if the mds_max == 2 in the cluster, when the MDS(0) 163 * was laggy and being replaced by a new MDS, we will temporarily 164 * receive a new mds map with n_num_mds == 1 and the active MDS(1), 165 * and the mds rank >= m_num_active_mds. 166 */ 167 m->possible_max_rank = max(m->m_num_active_mds, m->m_max_mds); 168 169 m->m_info = kcalloc(m->possible_max_rank, sizeof(*m->m_info), GFP_NOFS); 170 if (!m->m_info) 171 goto nomem; 172 173 /* pick out active nodes from mds_info (state > 0) */ 174 for (i = 0; i < n; i++) { 175 u64 global_id; 176 u32 namelen; 177 s32 mds, inc, state; 178 u8 info_v; 179 void *info_end = NULL; 180 struct ceph_entity_addr addr; 181 u32 num_export_targets; 182 void *pexport_targets = NULL; 183 struct ceph_timespec laggy_since; 184 struct ceph_mds_info *info; 185 bool laggy; 186 187 ceph_decode_need(p, end, sizeof(u64) + 1, bad); 188 global_id = ceph_decode_64(p); 189 info_v= ceph_decode_8(p); 190 if (info_v >= 4) { 191 u32 info_len; 192 ceph_decode_need(p, end, 1 + sizeof(u32), bad); 193 *p += sizeof(u8); /* info_cv */ 194 info_len = ceph_decode_32(p); 195 info_end = *p + info_len; 196 if (info_end > end) 197 goto bad; 198 } 199 200 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); 201 *p += sizeof(u64); 202 namelen = ceph_decode_32(p); /* skip mds name */ 203 *p += namelen; 204 205 ceph_decode_32_safe(p, end, mds, bad); 206 ceph_decode_32_safe(p, end, inc, bad); 207 ceph_decode_32_safe(p, end, state, bad); 208 *p += sizeof(u64); /* state_seq */ 209 if (info_v >= 8) 210 err = ceph_decode_entity_addrvec(p, end, msgr2, &addr); 211 else 212 err = ceph_decode_entity_addr(p, end, &addr); 213 if (err) 214 goto corrupt; 215 216 ceph_decode_copy_safe(p, end, &laggy_since, sizeof(laggy_since), 217 bad); 218 laggy = laggy_since.tv_sec != 0 || laggy_since.tv_nsec != 0; 219 *p += sizeof(u32); 220 ceph_decode_32_safe(p, end, namelen, bad); 221 *p += namelen; 222 if (info_v >= 2) { 223 ceph_decode_32_safe(p, end, num_export_targets, bad); 224 pexport_targets = *p; 225 *p += num_export_targets * sizeof(u32); 226 } else { 227 num_export_targets = 0; 228 } 229 230 if (info_end && *p != info_end) { 231 if (*p > info_end) 232 goto bad; 233 *p = info_end; 234 } 235 236 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s%s\n", 237 i+1, n, global_id, mds, inc, 238 ceph_pr_addr(&addr), 239 ceph_mds_state_name(state), 240 laggy ? "(laggy)" : ""); 241 242 if (mds < 0 || mds >= m->possible_max_rank) { 243 pr_warn("mdsmap_decode got incorrect mds(%d)\n", mds); 244 continue; 245 } 246 247 if (state <= 0) { 248 dout("mdsmap_decode got incorrect state(%s)\n", 249 ceph_mds_state_name(state)); 250 continue; 251 } 252 253 info = &m->m_info[mds]; 254 info->global_id = global_id; 255 info->state = state; 256 info->addr = addr; 257 info->laggy = laggy; 258 info->num_export_targets = num_export_targets; 259 if (num_export_targets) { 260 info->export_targets = kcalloc(num_export_targets, 261 sizeof(u32), GFP_NOFS); 262 if (!info->export_targets) 263 goto nomem; 264 for (j = 0; j < num_export_targets; j++) { 265 target = ceph_decode_32(&pexport_targets); 266 if (target >= m->possible_max_rank) { 267 err = -EIO; 268 goto corrupt; 269 } 270 info->export_targets[j] = target; 271 } 272 } else { 273 info->export_targets = NULL; 274 } 275 } 276 277 /* pg_pools */ 278 ceph_decode_32_safe(p, end, n, bad); 279 m->m_num_data_pg_pools = n; 280 m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS); 281 if (!m->m_data_pg_pools) 282 goto nomem; 283 ceph_decode_need(p, end, sizeof(u64)*(n+1), bad); 284 for (i = 0; i < n; i++) 285 m->m_data_pg_pools[i] = ceph_decode_64(p); 286 m->m_cas_pg_pool = ceph_decode_64(p); 287 m->m_enabled = m->m_epoch > 1; 288 289 mdsmap_ev = 1; 290 if (mdsmap_v >= 2) { 291 ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext); 292 } 293 if (mdsmap_ev >= 3) { 294 if (__decode_and_drop_compat_set(p, end) < 0) 295 goto bad_ext; 296 } 297 /* metadata_pool */ 298 if (mdsmap_ev < 5) { 299 __decode_and_drop_type(p, end, u32, bad_ext); 300 } else { 301 __decode_and_drop_type(p, end, u64, bad_ext); 302 } 303 304 /* created + modified + tableserver */ 305 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 306 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 307 __decode_and_drop_type(p, end, u32, bad_ext); 308 309 /* in */ 310 { 311 int num_laggy = 0; 312 ceph_decode_32_safe(p, end, n, bad_ext); 313 ceph_decode_need(p, end, sizeof(u32) * n, bad_ext); 314 315 for (i = 0; i < n; i++) { 316 s32 mds = ceph_decode_32(p); 317 if (mds >= 0 && mds < m->possible_max_rank) { 318 if (m->m_info[mds].laggy) 319 num_laggy++; 320 } 321 } 322 m->m_num_laggy = num_laggy; 323 324 if (n > m->possible_max_rank) { 325 void *new_m_info = krealloc(m->m_info, 326 n * sizeof(*m->m_info), 327 GFP_NOFS | __GFP_ZERO); 328 if (!new_m_info) 329 goto nomem; 330 m->m_info = new_m_info; 331 } 332 m->possible_max_rank = n; 333 } 334 335 /* inc */ 336 __decode_and_drop_map(p, end, u32, u32, bad_ext); 337 /* up */ 338 __decode_and_drop_map(p, end, u32, u64, bad_ext); 339 /* failed */ 340 __decode_and_drop_set(p, end, u32, bad_ext); 341 /* stopped */ 342 __decode_and_drop_set(p, end, u32, bad_ext); 343 344 if (mdsmap_ev >= 4) { 345 /* last_failure_osd_epoch */ 346 __decode_and_drop_type(p, end, u32, bad_ext); 347 } 348 if (mdsmap_ev >= 6) { 349 /* ever_allowed_snaps */ 350 __decode_and_drop_type(p, end, u8, bad_ext); 351 /* explicitly_allowed_snaps */ 352 __decode_and_drop_type(p, end, u8, bad_ext); 353 } 354 if (mdsmap_ev >= 7) { 355 /* inline_data_enabled */ 356 __decode_and_drop_type(p, end, u8, bad_ext); 357 } 358 if (mdsmap_ev >= 8) { 359 u32 name_len; 360 /* enabled */ 361 ceph_decode_8_safe(p, end, m->m_enabled, bad_ext); 362 ceph_decode_32_safe(p, end, name_len, bad_ext); 363 ceph_decode_need(p, end, name_len, bad_ext); 364 *p += name_len; 365 } 366 /* damaged */ 367 if (mdsmap_ev >= 9) { 368 size_t need; 369 ceph_decode_32_safe(p, end, n, bad_ext); 370 need = sizeof(u32) * n; 371 ceph_decode_need(p, end, need, bad_ext); 372 *p += need; 373 m->m_damaged = n > 0; 374 } else { 375 m->m_damaged = false; 376 } 377 bad_ext: 378 dout("mdsmap_decode m_enabled: %d, m_damaged: %d, m_num_laggy: %d\n", 379 !!m->m_enabled, !!m->m_damaged, m->m_num_laggy); 380 *p = end; 381 dout("mdsmap_decode success epoch %u\n", m->m_epoch); 382 return m; 383 nomem: 384 err = -ENOMEM; 385 goto out_err; 386 corrupt: 387 pr_err("corrupt mdsmap\n"); 388 print_hex_dump(KERN_DEBUG, "mdsmap: ", 389 DUMP_PREFIX_OFFSET, 16, 1, 390 start, end - start, true); 391 out_err: 392 ceph_mdsmap_destroy(m); 393 return ERR_PTR(err); 394 bad: 395 err = -EINVAL; 396 goto corrupt; 397 } 398 399 void ceph_mdsmap_destroy(struct ceph_mdsmap *m) 400 { 401 int i; 402 403 if (m->m_info) { 404 for (i = 0; i < m->possible_max_rank; i++) 405 kfree(m->m_info[i].export_targets); 406 kfree(m->m_info); 407 } 408 kfree(m->m_data_pg_pools); 409 kfree(m); 410 } 411 412 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m) 413 { 414 int i, nr_active = 0; 415 if (!m->m_enabled) 416 return false; 417 if (m->m_damaged) 418 return false; 419 if (m->m_num_laggy == m->m_num_active_mds) 420 return false; 421 for (i = 0; i < m->possible_max_rank; i++) { 422 if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE) 423 nr_active++; 424 } 425 return nr_active > 0; 426 } 427