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