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
__mdsmap_get_random_mds(struct ceph_mdsmap * m,bool ignore_laggy)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 = get_random_u32_below(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 */
ceph_mdsmap_get_random_mds(struct ceph_mdsmap * m)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
__decode_and_drop_compat_set(void ** p,void * end)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 */
ceph_mdsmap_decode(struct ceph_mds_client * mdsc,void ** p,void * end,bool msgr2)117 struct ceph_mdsmap *ceph_mdsmap_decode(struct ceph_mds_client *mdsc, void **p,
118 void *end, bool msgr2)
119 {
120 struct ceph_mdsmap *m;
121 const void *start = *p;
122 int i, j, n;
123 int err;
124 u8 mdsmap_v;
125 u16 mdsmap_ev;
126 u32 target;
127
128 m = kzalloc(sizeof(*m), GFP_NOFS);
129 if (!m)
130 return ERR_PTR(-ENOMEM);
131
132 ceph_decode_need(p, end, 1 + 1, bad);
133 mdsmap_v = ceph_decode_8(p);
134 *p += sizeof(u8); /* mdsmap_cv */
135 if (mdsmap_v >= 4) {
136 u32 mdsmap_len;
137 ceph_decode_32_safe(p, end, mdsmap_len, bad);
138 if (end < *p + mdsmap_len)
139 goto bad;
140 end = *p + mdsmap_len;
141 }
142
143 ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
144 m->m_epoch = ceph_decode_32(p);
145 m->m_client_epoch = ceph_decode_32(p);
146 m->m_last_failure = ceph_decode_32(p);
147 m->m_root = ceph_decode_32(p);
148 m->m_session_timeout = ceph_decode_32(p);
149 m->m_session_autoclose = ceph_decode_32(p);
150 m->m_max_file_size = ceph_decode_64(p);
151 m->m_max_mds = ceph_decode_32(p);
152
153 /*
154 * pick out the active nodes as the m_num_active_mds, the
155 * m_num_active_mds maybe larger than m_max_mds when decreasing
156 * the max_mds in cluster side, in other case it should less
157 * than or equal to m_max_mds.
158 */
159 m->m_num_active_mds = n = ceph_decode_32(p);
160
161 /*
162 * the possible max rank, it maybe larger than the m_num_active_mds,
163 * for example if the mds_max == 2 in the cluster, when the MDS(0)
164 * was laggy and being replaced by a new MDS, we will temporarily
165 * receive a new mds map with n_num_mds == 1 and the active MDS(1),
166 * and the mds rank >= m_num_active_mds.
167 */
168 m->possible_max_rank = max(m->m_num_active_mds, m->m_max_mds);
169
170 m->m_info = kcalloc(m->possible_max_rank, sizeof(*m->m_info), GFP_NOFS);
171 if (!m->m_info)
172 goto nomem;
173
174 /* pick out active nodes from mds_info (state > 0) */
175 for (i = 0; i < n; i++) {
176 u64 global_id;
177 u32 namelen;
178 s32 mds, inc, state;
179 u8 info_v;
180 void *info_end = NULL;
181 struct ceph_entity_addr addr;
182 u32 num_export_targets;
183 void *pexport_targets = NULL;
184 struct ceph_timespec laggy_since;
185 struct ceph_mds_info *info;
186 bool laggy;
187
188 ceph_decode_need(p, end, sizeof(u64) + 1, bad);
189 global_id = ceph_decode_64(p);
190 info_v= ceph_decode_8(p);
191 if (info_v >= 4) {
192 u32 info_len;
193 ceph_decode_need(p, end, 1 + sizeof(u32), bad);
194 *p += sizeof(u8); /* info_cv */
195 info_len = ceph_decode_32(p);
196 info_end = *p + info_len;
197 if (info_end > end)
198 goto bad;
199 }
200
201 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
202 *p += sizeof(u64);
203 namelen = ceph_decode_32(p); /* skip mds name */
204 *p += namelen;
205
206 ceph_decode_32_safe(p, end, mds, bad);
207 ceph_decode_32_safe(p, end, inc, bad);
208 ceph_decode_32_safe(p, end, state, bad);
209 *p += sizeof(u64); /* state_seq */
210 if (info_v >= 8)
211 err = ceph_decode_entity_addrvec(p, end, msgr2, &addr);
212 else
213 err = ceph_decode_entity_addr(p, end, &addr);
214 if (err)
215 goto corrupt;
216
217 ceph_decode_copy_safe(p, end, &laggy_since, sizeof(laggy_since),
218 bad);
219 laggy = laggy_since.tv_sec != 0 || laggy_since.tv_nsec != 0;
220 *p += sizeof(u32);
221 ceph_decode_32_safe(p, end, namelen, bad);
222 *p += namelen;
223 if (info_v >= 2) {
224 ceph_decode_32_safe(p, end, num_export_targets, bad);
225 pexport_targets = *p;
226 *p += num_export_targets * sizeof(u32);
227 } else {
228 num_export_targets = 0;
229 }
230
231 if (info_end && *p != info_end) {
232 if (*p > info_end)
233 goto bad;
234 *p = info_end;
235 }
236
237 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s%s\n",
238 i+1, n, global_id, mds, inc,
239 ceph_pr_addr(&addr),
240 ceph_mds_state_name(state),
241 laggy ? "(laggy)" : "");
242
243 if (mds < 0 || mds >= m->possible_max_rank) {
244 pr_warn("mdsmap_decode got incorrect mds(%d)\n", mds);
245 continue;
246 }
247
248 if (state <= 0) {
249 dout("mdsmap_decode got incorrect state(%s)\n",
250 ceph_mds_state_name(state));
251 continue;
252 }
253
254 info = &m->m_info[mds];
255 info->global_id = global_id;
256 info->state = state;
257 info->addr = addr;
258 info->laggy = laggy;
259 info->num_export_targets = num_export_targets;
260 if (num_export_targets) {
261 info->export_targets = kcalloc(num_export_targets,
262 sizeof(u32), GFP_NOFS);
263 if (!info->export_targets)
264 goto nomem;
265 for (j = 0; j < num_export_targets; j++) {
266 target = ceph_decode_32(&pexport_targets);
267 info->export_targets[j] = target;
268 }
269 } else {
270 info->export_targets = NULL;
271 }
272 }
273
274 /* pg_pools */
275 ceph_decode_32_safe(p, end, n, bad);
276 m->m_num_data_pg_pools = n;
277 m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS);
278 if (!m->m_data_pg_pools)
279 goto nomem;
280 ceph_decode_need(p, end, sizeof(u64)*(n+1), bad);
281 for (i = 0; i < n; i++)
282 m->m_data_pg_pools[i] = ceph_decode_64(p);
283 m->m_cas_pg_pool = ceph_decode_64(p);
284 m->m_enabled = m->m_epoch > 1;
285
286 mdsmap_ev = 1;
287 if (mdsmap_v >= 2) {
288 ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext);
289 }
290 if (mdsmap_ev >= 3) {
291 if (__decode_and_drop_compat_set(p, end) < 0)
292 goto bad_ext;
293 }
294 /* metadata_pool */
295 if (mdsmap_ev < 5) {
296 __decode_and_drop_type(p, end, u32, bad_ext);
297 } else {
298 __decode_and_drop_type(p, end, u64, bad_ext);
299 }
300
301 /* created + modified + tableserver */
302 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext);
303 __decode_and_drop_type(p, end, struct ceph_timespec, bad_ext);
304 __decode_and_drop_type(p, end, u32, bad_ext);
305
306 /* in */
307 {
308 int num_laggy = 0;
309 ceph_decode_32_safe(p, end, n, bad_ext);
310 ceph_decode_need(p, end, sizeof(u32) * n, bad_ext);
311
312 for (i = 0; i < n; i++) {
313 s32 mds = ceph_decode_32(p);
314 if (mds >= 0 && mds < m->possible_max_rank) {
315 if (m->m_info[mds].laggy)
316 num_laggy++;
317 }
318 }
319 m->m_num_laggy = num_laggy;
320
321 if (n > m->possible_max_rank) {
322 void *new_m_info = krealloc(m->m_info,
323 n * sizeof(*m->m_info),
324 GFP_NOFS | __GFP_ZERO);
325 if (!new_m_info)
326 goto nomem;
327 m->m_info = new_m_info;
328 }
329 m->possible_max_rank = n;
330 }
331
332 /* inc */
333 __decode_and_drop_map(p, end, u32, u32, bad_ext);
334 /* up */
335 __decode_and_drop_map(p, end, u32, u64, bad_ext);
336 /* failed */
337 __decode_and_drop_set(p, end, u32, bad_ext);
338 /* stopped */
339 __decode_and_drop_set(p, end, u32, bad_ext);
340
341 if (mdsmap_ev >= 4) {
342 /* last_failure_osd_epoch */
343 __decode_and_drop_type(p, end, u32, bad_ext);
344 }
345 if (mdsmap_ev >= 6) {
346 /* ever_allowed_snaps */
347 __decode_and_drop_type(p, end, u8, bad_ext);
348 /* explicitly_allowed_snaps */
349 __decode_and_drop_type(p, end, u8, bad_ext);
350 }
351 if (mdsmap_ev >= 7) {
352 /* inline_data_enabled */
353 __decode_and_drop_type(p, end, u8, bad_ext);
354 }
355 if (mdsmap_ev >= 8) {
356 /* enabled */
357 ceph_decode_8_safe(p, end, m->m_enabled, bad_ext);
358 /* fs_name */
359 ceph_decode_skip_string(p, end, bad_ext);
360 }
361 /* damaged */
362 if (mdsmap_ev >= 9) {
363 size_t need;
364 ceph_decode_32_safe(p, end, n, bad_ext);
365 need = sizeof(u32) * n;
366 ceph_decode_need(p, end, need, bad_ext);
367 *p += need;
368 m->m_damaged = n > 0;
369 } else {
370 m->m_damaged = false;
371 }
372 if (mdsmap_ev >= 17) {
373 /* balancer */
374 ceph_decode_skip_string(p, end, bad_ext);
375 /* standby_count_wanted */
376 ceph_decode_skip_32(p, end, bad_ext);
377 /* old_max_mds */
378 ceph_decode_skip_32(p, end, bad_ext);
379 /* min_compat_client */
380 ceph_decode_skip_8(p, end, bad_ext);
381 /* required_client_features */
382 ceph_decode_skip_set(p, end, 64, bad_ext);
383 /* bal_rank_mask */
384 ceph_decode_skip_string(p, end, bad_ext);
385 }
386 if (mdsmap_ev >= 18) {
387 ceph_decode_64_safe(p, end, m->m_max_xattr_size, bad_ext);
388 }
389 bad_ext:
390 dout("mdsmap_decode m_enabled: %d, m_damaged: %d, m_num_laggy: %d\n",
391 !!m->m_enabled, !!m->m_damaged, m->m_num_laggy);
392 *p = end;
393 dout("mdsmap_decode success epoch %u\n", m->m_epoch);
394 return m;
395 nomem:
396 err = -ENOMEM;
397 goto out_err;
398 corrupt:
399 pr_err("corrupt mdsmap\n");
400 print_hex_dump(KERN_DEBUG, "mdsmap: ",
401 DUMP_PREFIX_OFFSET, 16, 1,
402 start, end - start, true);
403 out_err:
404 ceph_mdsmap_destroy(m);
405 return ERR_PTR(err);
406 bad:
407 err = -EINVAL;
408 goto corrupt;
409 }
410
ceph_mdsmap_destroy(struct ceph_mdsmap * m)411 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
412 {
413 int i;
414
415 if (m->m_info) {
416 for (i = 0; i < m->possible_max_rank; i++)
417 kfree(m->m_info[i].export_targets);
418 kfree(m->m_info);
419 }
420 kfree(m->m_data_pg_pools);
421 kfree(m);
422 }
423
ceph_mdsmap_is_cluster_available(struct ceph_mdsmap * m)424 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m)
425 {
426 int i, nr_active = 0;
427 if (!m->m_enabled)
428 return false;
429 if (m->m_damaged)
430 return false;
431 if (m->m_num_laggy == m->m_num_active_mds)
432 return false;
433 for (i = 0; i < m->possible_max_rank; i++) {
434 if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE)
435 nr_active++;
436 }
437 return nr_active > 0;
438 }
439