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) 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_need(p, end, 205 4*sizeof(u32) + sizeof(u64) + 206 sizeof(addr) + sizeof(struct ceph_timespec), 207 bad); 208 mds = ceph_decode_32(p); 209 inc = ceph_decode_32(p); 210 state = ceph_decode_32(p); 211 *p += sizeof(u64); /* state_seq */ 212 err = ceph_decode_entity_addr(p, end, &addr); 213 if (err) 214 goto corrupt; 215 ceph_decode_copy(p, &laggy_since, sizeof(laggy_since)); 216 laggy = laggy_since.tv_sec != 0 || laggy_since.tv_nsec != 0; 217 *p += sizeof(u32); 218 ceph_decode_32_safe(p, end, namelen, bad); 219 *p += namelen; 220 if (info_v >= 2) { 221 ceph_decode_32_safe(p, end, num_export_targets, bad); 222 pexport_targets = *p; 223 *p += num_export_targets * sizeof(u32); 224 } else { 225 num_export_targets = 0; 226 } 227 228 if (info_end && *p != info_end) { 229 if (*p > info_end) 230 goto bad; 231 *p = info_end; 232 } 233 234 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s%s\n", 235 i+1, n, global_id, mds, inc, 236 ceph_pr_addr(&addr), 237 ceph_mds_state_name(state), 238 laggy ? "(laggy)" : ""); 239 240 if (mds < 0 || mds >= m->possible_max_rank) { 241 pr_warn("mdsmap_decode got incorrect mds(%d)\n", mds); 242 continue; 243 } 244 245 if (state <= 0) { 246 dout("mdsmap_decode got incorrect state(%s)\n", 247 ceph_mds_state_name(state)); 248 continue; 249 } 250 251 info = &m->m_info[mds]; 252 info->global_id = global_id; 253 info->state = state; 254 info->addr = addr; 255 info->laggy = laggy; 256 info->num_export_targets = num_export_targets; 257 if (num_export_targets) { 258 info->export_targets = kcalloc(num_export_targets, 259 sizeof(u32), GFP_NOFS); 260 if (!info->export_targets) 261 goto nomem; 262 for (j = 0; j < num_export_targets; j++) 263 info->export_targets[j] = 264 ceph_decode_32(&pexport_targets); 265 } else { 266 info->export_targets = NULL; 267 } 268 } 269 270 /* pg_pools */ 271 ceph_decode_32_safe(p, end, n, bad); 272 m->m_num_data_pg_pools = n; 273 m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS); 274 if (!m->m_data_pg_pools) 275 goto nomem; 276 ceph_decode_need(p, end, sizeof(u64)*(n+1), bad); 277 for (i = 0; i < n; i++) 278 m->m_data_pg_pools[i] = ceph_decode_64(p); 279 m->m_cas_pg_pool = ceph_decode_64(p); 280 m->m_enabled = m->m_epoch > 1; 281 282 mdsmap_ev = 1; 283 if (mdsmap_v >= 2) { 284 ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext); 285 } 286 if (mdsmap_ev >= 3) { 287 if (__decode_and_drop_compat_set(p, end) < 0) 288 goto bad_ext; 289 } 290 /* metadata_pool */ 291 if (mdsmap_ev < 5) { 292 __decode_and_drop_type(p, end, u32, bad_ext); 293 } else { 294 __decode_and_drop_type(p, end, u64, bad_ext); 295 } 296 297 /* created + modified + tableserver */ 298 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 299 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext); 300 __decode_and_drop_type(p, end, u32, bad_ext); 301 302 /* in */ 303 { 304 int num_laggy = 0; 305 ceph_decode_32_safe(p, end, n, bad_ext); 306 ceph_decode_need(p, end, sizeof(u32) * n, bad_ext); 307 308 for (i = 0; i < n; i++) { 309 s32 mds = ceph_decode_32(p); 310 if (mds >= 0 && mds < m->possible_max_rank) { 311 if (m->m_info[mds].laggy) 312 num_laggy++; 313 } 314 } 315 m->m_num_laggy = num_laggy; 316 317 if (n > m->possible_max_rank) { 318 void *new_m_info = krealloc(m->m_info, 319 n * sizeof(*m->m_info), 320 GFP_NOFS | __GFP_ZERO); 321 if (!new_m_info) 322 goto nomem; 323 m->m_info = new_m_info; 324 } 325 m->possible_max_rank = n; 326 } 327 328 /* inc */ 329 __decode_and_drop_map(p, end, u32, u32, bad_ext); 330 /* up */ 331 __decode_and_drop_map(p, end, u32, u64, bad_ext); 332 /* failed */ 333 __decode_and_drop_set(p, end, u32, bad_ext); 334 /* stopped */ 335 __decode_and_drop_set(p, end, u32, bad_ext); 336 337 if (mdsmap_ev >= 4) { 338 /* last_failure_osd_epoch */ 339 __decode_and_drop_type(p, end, u32, bad_ext); 340 } 341 if (mdsmap_ev >= 6) { 342 /* ever_allowed_snaps */ 343 __decode_and_drop_type(p, end, u8, bad_ext); 344 /* explicitly_allowed_snaps */ 345 __decode_and_drop_type(p, end, u8, bad_ext); 346 } 347 if (mdsmap_ev >= 7) { 348 /* inline_data_enabled */ 349 __decode_and_drop_type(p, end, u8, bad_ext); 350 } 351 if (mdsmap_ev >= 8) { 352 u32 name_len; 353 /* enabled */ 354 ceph_decode_8_safe(p, end, m->m_enabled, bad_ext); 355 ceph_decode_32_safe(p, end, name_len, bad_ext); 356 ceph_decode_need(p, end, name_len, bad_ext); 357 *p += name_len; 358 } 359 /* damaged */ 360 if (mdsmap_ev >= 9) { 361 size_t need; 362 ceph_decode_32_safe(p, end, n, bad_ext); 363 need = sizeof(u32) * n; 364 ceph_decode_need(p, end, need, bad_ext); 365 *p += need; 366 m->m_damaged = n > 0; 367 } else { 368 m->m_damaged = false; 369 } 370 bad_ext: 371 dout("mdsmap_decode m_enabled: %d, m_damaged: %d, m_num_laggy: %d\n", 372 !!m->m_enabled, !!m->m_damaged, m->m_num_laggy); 373 *p = end; 374 dout("mdsmap_decode success epoch %u\n", m->m_epoch); 375 return m; 376 nomem: 377 err = -ENOMEM; 378 goto out_err; 379 corrupt: 380 pr_err("corrupt mdsmap\n"); 381 print_hex_dump(KERN_DEBUG, "mdsmap: ", 382 DUMP_PREFIX_OFFSET, 16, 1, 383 start, end - start, true); 384 out_err: 385 ceph_mdsmap_destroy(m); 386 return ERR_PTR(err); 387 bad: 388 err = -EINVAL; 389 goto corrupt; 390 } 391 392 void ceph_mdsmap_destroy(struct ceph_mdsmap *m) 393 { 394 int i; 395 396 for (i = 0; i < m->possible_max_rank; i++) 397 kfree(m->m_info[i].export_targets); 398 kfree(m->m_info); 399 kfree(m->m_data_pg_pools); 400 kfree(m); 401 } 402 403 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m) 404 { 405 int i, nr_active = 0; 406 if (!m->m_enabled) 407 return false; 408 if (m->m_damaged) 409 return false; 410 if (m->m_num_laggy == m->m_num_active_mds) 411 return false; 412 for (i = 0; i < m->possible_max_rank; i++) { 413 if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE) 414 nr_active++; 415 } 416 return nr_active > 0; 417 } 418