xref: /openbmc/linux/fs/ceph/mdsmap.c (revision 3ce311af)
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  */
20 int ceph_mdsmap_get_random_mds(struct ceph_mdsmap *m)
21 {
22 	int n = 0;
23 	int i, j;
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 	for (j = 0, i = 0; i < m->m_num_mds; i++) {
39 		if (m->m_info[i].state > 0)
40 			j++;
41 		if (j > n)
42 			break;
43 	}
44 
45 	return i;
46 }
47 
48 #define __decode_and_drop_type(p, end, type, bad)		\
49 	do {							\
50 		if (*p + sizeof(type) > end)			\
51 			goto bad;				\
52 		*p += sizeof(type);				\
53 	} while (0)
54 
55 #define __decode_and_drop_set(p, end, type, bad)		\
56 	do {							\
57 		u32 n;						\
58 		size_t need;					\
59 		ceph_decode_32_safe(p, end, n, bad);		\
60 		need = sizeof(type) * n;			\
61 		ceph_decode_need(p, end, need, bad);		\
62 		*p += need;					\
63 	} while (0)
64 
65 #define __decode_and_drop_map(p, end, ktype, vtype, bad)	\
66 	do {							\
67 		u32 n;						\
68 		size_t need;					\
69 		ceph_decode_32_safe(p, end, n, bad);		\
70 		need = (sizeof(ktype) + sizeof(vtype)) * n;	\
71 		ceph_decode_need(p, end, need, bad);		\
72 		*p += need;					\
73 	} while (0)
74 
75 
76 static int __decode_and_drop_compat_set(void **p, void* end)
77 {
78 	int i;
79 	/* compat, ro_compat, incompat*/
80 	for (i = 0; i < 3; i++) {
81 		u32 n;
82 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
83 		/* mask */
84 		*p += sizeof(u64);
85 		/* names (map<u64, string>) */
86 		n = ceph_decode_32(p);
87 		while (n-- > 0) {
88 			u32 len;
89 			ceph_decode_need(p, end, sizeof(u64) + sizeof(u32),
90 					 bad);
91 			*p += sizeof(u64);
92 			len = ceph_decode_32(p);
93 			ceph_decode_need(p, end, len, bad);
94 			*p += len;
95 		}
96 	}
97 	return 0;
98 bad:
99 	return -1;
100 }
101 
102 /*
103  * Decode an MDS map
104  *
105  * Ignore any fields we don't care about (there are quite a few of
106  * them).
107  */
108 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
109 {
110 	struct ceph_mdsmap *m;
111 	const void *start = *p;
112 	int i, j, n;
113 	int err;
114 	u8 mdsmap_v, mdsmap_cv;
115 	u16 mdsmap_ev;
116 
117 	m = kzalloc(sizeof(*m), GFP_NOFS);
118 	if (!m)
119 		return ERR_PTR(-ENOMEM);
120 
121 	ceph_decode_need(p, end, 1 + 1, bad);
122 	mdsmap_v = ceph_decode_8(p);
123 	mdsmap_cv = ceph_decode_8(p);
124 	if (mdsmap_v >= 4) {
125 	       u32 mdsmap_len;
126 	       ceph_decode_32_safe(p, end, mdsmap_len, bad);
127 	       if (end < *p + mdsmap_len)
128 		       goto bad;
129 	       end = *p + mdsmap_len;
130 	}
131 
132 	ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
133 	m->m_epoch = ceph_decode_32(p);
134 	m->m_client_epoch = ceph_decode_32(p);
135 	m->m_last_failure = ceph_decode_32(p);
136 	m->m_root = ceph_decode_32(p);
137 	m->m_session_timeout = ceph_decode_32(p);
138 	m->m_session_autoclose = ceph_decode_32(p);
139 	m->m_max_file_size = ceph_decode_64(p);
140 	m->m_max_mds = ceph_decode_32(p);
141 	m->m_num_mds = m->m_max_mds;
142 
143 	m->m_info = kcalloc(m->m_num_mds, sizeof(*m->m_info), GFP_NOFS);
144 	if (!m->m_info)
145 		goto nomem;
146 
147 	/* pick out active nodes from mds_info (state > 0) */
148 	n = ceph_decode_32(p);
149 	for (i = 0; i < n; i++) {
150 		u64 global_id;
151 		u32 namelen;
152 		s32 mds, inc, state;
153 		u64 state_seq;
154 		u8 info_v;
155 		void *info_end = NULL;
156 		struct ceph_entity_addr addr;
157 		u32 num_export_targets;
158 		void *pexport_targets = NULL;
159 		struct ceph_timespec laggy_since;
160 		struct ceph_mds_info *info;
161 		bool laggy;
162 
163 		ceph_decode_need(p, end, sizeof(u64) + 1, bad);
164 		global_id = ceph_decode_64(p);
165 		info_v= ceph_decode_8(p);
166 		if (info_v >= 4) {
167 			u32 info_len;
168 			u8 info_cv;
169 			ceph_decode_need(p, end, 1 + sizeof(u32), bad);
170 			info_cv = ceph_decode_8(p);
171 			info_len = ceph_decode_32(p);
172 			info_end = *p + info_len;
173 			if (info_end > end)
174 				goto bad;
175 		}
176 
177 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
178 		*p += sizeof(u64);
179 		namelen = ceph_decode_32(p);  /* skip mds name */
180 		*p += namelen;
181 
182 		ceph_decode_need(p, end,
183 				 4*sizeof(u32) + sizeof(u64) +
184 				 sizeof(addr) + sizeof(struct ceph_timespec),
185 				 bad);
186 		mds = ceph_decode_32(p);
187 		inc = ceph_decode_32(p);
188 		state = ceph_decode_32(p);
189 		state_seq = ceph_decode_64(p);
190 		err = ceph_decode_entity_addr(p, end, &addr);
191 		if (err)
192 			goto corrupt;
193 		ceph_decode_copy(p, &laggy_since, sizeof(laggy_since));
194 		laggy = laggy_since.tv_sec != 0 || laggy_since.tv_nsec != 0;
195 		*p += sizeof(u32);
196 		ceph_decode_32_safe(p, end, namelen, bad);
197 		*p += namelen;
198 		if (info_v >= 2) {
199 			ceph_decode_32_safe(p, end, num_export_targets, bad);
200 			pexport_targets = *p;
201 			*p += num_export_targets * sizeof(u32);
202 		} else {
203 			num_export_targets = 0;
204 		}
205 
206 		if (info_end && *p != info_end) {
207 			if (*p > info_end)
208 				goto bad;
209 			*p = info_end;
210 		}
211 
212 		dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s%s\n",
213 		     i+1, n, global_id, mds, inc,
214 		     ceph_pr_addr(&addr),
215 		     ceph_mds_state_name(state),
216 		     laggy ? "(laggy)" : "");
217 
218 		if (mds < 0 || state <= 0)
219 			continue;
220 
221 		if (mds >= m->m_num_mds) {
222 			int new_num = max(mds + 1, m->m_num_mds * 2);
223 			void *new_m_info = krealloc(m->m_info,
224 						new_num * sizeof(*m->m_info),
225 						GFP_NOFS | __GFP_ZERO);
226 			if (!new_m_info)
227 				goto nomem;
228 			m->m_info = new_m_info;
229 			m->m_num_mds = new_num;
230 		}
231 
232 		info = &m->m_info[mds];
233 		info->global_id = global_id;
234 		info->state = state;
235 		info->addr = addr;
236 		info->laggy = laggy;
237 		info->num_export_targets = num_export_targets;
238 		if (num_export_targets) {
239 			info->export_targets = kcalloc(num_export_targets,
240 						       sizeof(u32), GFP_NOFS);
241 			if (!info->export_targets)
242 				goto nomem;
243 			for (j = 0; j < num_export_targets; j++)
244 				info->export_targets[j] =
245 				       ceph_decode_32(&pexport_targets);
246 		} else {
247 			info->export_targets = NULL;
248 		}
249 	}
250 	if (m->m_num_mds > m->m_max_mds) {
251 		/* find max up mds */
252 		for (i = m->m_num_mds; i >= m->m_max_mds; i--) {
253 			if (i == 0 || m->m_info[i-1].state > 0)
254 				break;
255 		}
256 		m->m_num_mds = i;
257 	}
258 
259 	/* pg_pools */
260 	ceph_decode_32_safe(p, end, n, bad);
261 	m->m_num_data_pg_pools = n;
262 	m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS);
263 	if (!m->m_data_pg_pools)
264 		goto nomem;
265 	ceph_decode_need(p, end, sizeof(u64)*(n+1), bad);
266 	for (i = 0; i < n; i++)
267 		m->m_data_pg_pools[i] = ceph_decode_64(p);
268 	m->m_cas_pg_pool = ceph_decode_64(p);
269 	m->m_enabled = m->m_epoch > 1;
270 
271 	mdsmap_ev = 1;
272 	if (mdsmap_v >= 2) {
273 		ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext);
274 	}
275 	if (mdsmap_ev >= 3) {
276 		if (__decode_and_drop_compat_set(p, end) < 0)
277 			goto bad_ext;
278 	}
279 	/* metadata_pool */
280 	if (mdsmap_ev < 5) {
281 		__decode_and_drop_type(p, end, u32, bad_ext);
282 	} else {
283 		__decode_and_drop_type(p, end, u64, bad_ext);
284 	}
285 
286 	/* created + modified + tableserver */
287 	__decode_and_drop_type(p, end, struct ceph_timespec, bad_ext);
288 	__decode_and_drop_type(p, end, struct ceph_timespec, bad_ext);
289 	__decode_and_drop_type(p, end, u32, bad_ext);
290 
291 	/* in */
292 	{
293 		int num_laggy = 0;
294 		ceph_decode_32_safe(p, end, n, bad_ext);
295 		ceph_decode_need(p, end, sizeof(u32) * n, bad_ext);
296 
297 		for (i = 0; i < n; i++) {
298 			s32 mds = ceph_decode_32(p);
299 			if (mds >= 0 && mds < m->m_num_mds) {
300 				if (m->m_info[mds].laggy)
301 					num_laggy++;
302 			}
303 		}
304 		m->m_num_laggy = num_laggy;
305 
306 		if (n > m->m_num_mds) {
307 			void *new_m_info = krealloc(m->m_info,
308 						    n * sizeof(*m->m_info),
309 						    GFP_NOFS | __GFP_ZERO);
310 			if (!new_m_info)
311 				goto nomem;
312 			m->m_info = new_m_info;
313 		}
314 		m->m_num_mds = n;
315 	}
316 
317 	/* inc */
318 	__decode_and_drop_map(p, end, u32, u32, bad_ext);
319 	/* up */
320 	__decode_and_drop_map(p, end, u32, u64, bad_ext);
321 	/* failed */
322 	__decode_and_drop_set(p, end, u32, bad_ext);
323 	/* stopped */
324 	__decode_and_drop_set(p, end, u32, bad_ext);
325 
326 	if (mdsmap_ev >= 4) {
327 		/* last_failure_osd_epoch */
328 		__decode_and_drop_type(p, end, u32, bad_ext);
329 	}
330 	if (mdsmap_ev >= 6) {
331 		/* ever_allowed_snaps */
332 		__decode_and_drop_type(p, end, u8, bad_ext);
333 		/* explicitly_allowed_snaps */
334 		__decode_and_drop_type(p, end, u8, bad_ext);
335 	}
336 	if (mdsmap_ev >= 7) {
337 		/* inline_data_enabled */
338 		__decode_and_drop_type(p, end, u8, bad_ext);
339 	}
340 	if (mdsmap_ev >= 8) {
341 		u32 name_len;
342 		/* enabled */
343 		ceph_decode_8_safe(p, end, m->m_enabled, bad_ext);
344 		ceph_decode_32_safe(p, end, name_len, bad_ext);
345 		ceph_decode_need(p, end, name_len, bad_ext);
346 		*p += name_len;
347 	}
348 	/* damaged */
349 	if (mdsmap_ev >= 9) {
350 		size_t need;
351 		ceph_decode_32_safe(p, end, n, bad_ext);
352 		need = sizeof(u32) * n;
353 		ceph_decode_need(p, end, need, bad_ext);
354 		*p += need;
355 		m->m_damaged = n > 0;
356 	} else {
357 		m->m_damaged = false;
358 	}
359 bad_ext:
360 	dout("mdsmap_decode m_enabled: %d, m_damaged: %d, m_num_laggy: %d\n",
361 	     !!m->m_enabled, !!m->m_damaged, m->m_num_laggy);
362 	*p = end;
363 	dout("mdsmap_decode success epoch %u\n", m->m_epoch);
364 	return m;
365 nomem:
366 	err = -ENOMEM;
367 	goto out_err;
368 corrupt:
369 	pr_err("corrupt mdsmap\n");
370 	print_hex_dump(KERN_DEBUG, "mdsmap: ",
371 		       DUMP_PREFIX_OFFSET, 16, 1,
372 		       start, end - start, true);
373 out_err:
374 	ceph_mdsmap_destroy(m);
375 	return ERR_PTR(err);
376 bad:
377 	err = -EINVAL;
378 	goto corrupt;
379 }
380 
381 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
382 {
383 	int i;
384 
385 	for (i = 0; i < m->m_num_mds; i++)
386 		kfree(m->m_info[i].export_targets);
387 	kfree(m->m_info);
388 	kfree(m->m_data_pg_pools);
389 	kfree(m);
390 }
391 
392 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m)
393 {
394 	int i, nr_active = 0;
395 	if (!m->m_enabled)
396 		return false;
397 	if (m->m_damaged)
398 		return false;
399 	if (m->m_num_laggy > 0)
400 		return false;
401 	for (i = 0; i < m->m_num_mds; i++) {
402 		if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE)
403 			nr_active++;
404 	}
405 	return nr_active > 0;
406 }
407