xref: /openbmc/linux/fs/ceph/mdsmap.c (revision aac5987a)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/bug.h>
4 #include <linux/err.h>
5 #include <linux/random.h>
6 #include <linux/slab.h>
7 #include <linux/types.h>
8 
9 #include <linux/ceph/mdsmap.h>
10 #include <linux/ceph/messenger.h>
11 #include <linux/ceph/decode.h>
12 
13 #include "super.h"
14 
15 
16 /*
17  * choose a random mds that is "up" (i.e. has a state > 0), or -1.
18  */
19 int ceph_mdsmap_get_random_mds(struct ceph_mdsmap *m)
20 {
21 	int n = 0;
22 	int i;
23 
24 	/* special case for one mds */
25 	if (1 == m->m_max_mds && m->m_info[0].state > 0)
26 		return 0;
27 
28 	/* count */
29 	for (i = 0; i < m->m_max_mds; i++)
30 		if (m->m_info[i].state > 0)
31 			n++;
32 	if (n == 0)
33 		return -1;
34 
35 	/* pick */
36 	n = prandom_u32() % n;
37 	i = 0;
38 	for (i = 0; n > 0; i++, n--)
39 		while (m->m_info[i].state <= 0)
40 			i++;
41 
42 	return i;
43 }
44 
45 #define __decode_and_drop_type(p, end, type, bad)		\
46 	do {							\
47 		if (*p + sizeof(type) > end)			\
48 			goto bad;				\
49 		*p += sizeof(type);				\
50 	} while (0)
51 
52 #define __decode_and_drop_set(p, end, type, bad)		\
53 	do {							\
54 		u32 n;						\
55 		size_t need;					\
56 		ceph_decode_32_safe(p, end, n, bad);		\
57 		need = sizeof(type) * n;			\
58 		ceph_decode_need(p, end, need, bad);		\
59 		*p += need;					\
60 	} while (0)
61 
62 #define __decode_and_drop_map(p, end, ktype, vtype, bad)	\
63 	do {							\
64 		u32 n;						\
65 		size_t need;					\
66 		ceph_decode_32_safe(p, end, n, bad);		\
67 		need = (sizeof(ktype) + sizeof(vtype)) * n;	\
68 		ceph_decode_need(p, end, need, bad);		\
69 		*p += need;					\
70 	} while (0)
71 
72 
73 static int __decode_and_drop_compat_set(void **p, void* end)
74 {
75 	int i;
76 	/* compat, ro_compat, incompat*/
77 	for (i = 0; i < 3; i++) {
78 		u32 n;
79 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
80 		/* mask */
81 		*p += sizeof(u64);
82 		/* names (map<u64, string>) */
83 		n = ceph_decode_32(p);
84 		while (n-- > 0) {
85 			u32 len;
86 			ceph_decode_need(p, end, sizeof(u64) + sizeof(u32),
87 					 bad);
88 			*p += sizeof(u64);
89 			len = ceph_decode_32(p);
90 			ceph_decode_need(p, end, len, bad);
91 			*p += len;
92 		}
93 	}
94 	return 0;
95 bad:
96 	return -1;
97 }
98 
99 /*
100  * Decode an MDS map
101  *
102  * Ignore any fields we don't care about (there are quite a few of
103  * them).
104  */
105 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
106 {
107 	struct ceph_mdsmap *m;
108 	const void *start = *p;
109 	int i, j, n;
110 	int err = -EINVAL;
111 	u8 mdsmap_v, mdsmap_cv;
112 	u16 mdsmap_ev;
113 
114 	m = kzalloc(sizeof(*m), GFP_NOFS);
115 	if (m == NULL)
116 		return ERR_PTR(-ENOMEM);
117 
118 	ceph_decode_need(p, end, 1 + 1, bad);
119 	mdsmap_v = ceph_decode_8(p);
120 	mdsmap_cv = ceph_decode_8(p);
121 	if (mdsmap_v >= 4) {
122 	       u32 mdsmap_len;
123 	       ceph_decode_32_safe(p, end, mdsmap_len, bad);
124 	       if (end < *p + mdsmap_len)
125 		       goto bad;
126 	       end = *p + mdsmap_len;
127 	}
128 
129 	ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
130 	m->m_epoch = ceph_decode_32(p);
131 	m->m_client_epoch = ceph_decode_32(p);
132 	m->m_last_failure = ceph_decode_32(p);
133 	m->m_root = ceph_decode_32(p);
134 	m->m_session_timeout = ceph_decode_32(p);
135 	m->m_session_autoclose = ceph_decode_32(p);
136 	m->m_max_file_size = ceph_decode_64(p);
137 	m->m_max_mds = ceph_decode_32(p);
138 
139 	m->m_info = kcalloc(m->m_max_mds, sizeof(*m->m_info), GFP_NOFS);
140 	if (m->m_info == NULL)
141 		goto nomem;
142 
143 	/* pick out active nodes from mds_info (state > 0) */
144 	n = ceph_decode_32(p);
145 	for (i = 0; i < n; i++) {
146 		u64 global_id;
147 		u32 namelen;
148 		s32 mds, inc, state;
149 		u64 state_seq;
150 		u8 info_v;
151 		void *info_end = NULL;
152 		struct ceph_entity_addr addr;
153 		u32 num_export_targets;
154 		void *pexport_targets = NULL;
155 		struct ceph_timespec laggy_since;
156 		struct ceph_mds_info *info;
157 
158 		ceph_decode_need(p, end, sizeof(u64) + 1, bad);
159 		global_id = ceph_decode_64(p);
160 		info_v= ceph_decode_8(p);
161 		if (info_v >= 4) {
162 			u32 info_len;
163 			u8 info_cv;
164 			ceph_decode_need(p, end, 1 + sizeof(u32), bad);
165 			info_cv = ceph_decode_8(p);
166 			info_len = ceph_decode_32(p);
167 			info_end = *p + info_len;
168 			if (info_end > end)
169 				goto bad;
170 		}
171 
172 		ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
173 		*p += sizeof(u64);
174 		namelen = ceph_decode_32(p);  /* skip mds name */
175 		*p += namelen;
176 
177 		ceph_decode_need(p, end,
178 				 4*sizeof(u32) + sizeof(u64) +
179 				 sizeof(addr) + sizeof(struct ceph_timespec),
180 				 bad);
181 		mds = ceph_decode_32(p);
182 		inc = ceph_decode_32(p);
183 		state = ceph_decode_32(p);
184 		state_seq = ceph_decode_64(p);
185 		ceph_decode_copy(p, &addr, sizeof(addr));
186 		ceph_decode_addr(&addr);
187 		ceph_decode_copy(p, &laggy_since, sizeof(laggy_since));
188 		*p += sizeof(u32);
189 		ceph_decode_32_safe(p, end, namelen, bad);
190 		*p += namelen;
191 		if (info_v >= 2) {
192 			ceph_decode_32_safe(p, end, num_export_targets, bad);
193 			pexport_targets = *p;
194 			*p += num_export_targets * sizeof(u32);
195 		} else {
196 			num_export_targets = 0;
197 		}
198 
199 		if (info_end && *p != info_end) {
200 			if (*p > info_end)
201 				goto bad;
202 			*p = info_end;
203 		}
204 
205 		dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n",
206 		     i+1, n, global_id, mds, inc,
207 		     ceph_pr_addr(&addr.in_addr),
208 		     ceph_mds_state_name(state));
209 
210 		if (mds < 0 || mds >= m->m_max_mds || state <= 0)
211 			continue;
212 
213 		info = &m->m_info[mds];
214 		info->global_id = global_id;
215 		info->state = state;
216 		info->addr = addr;
217 		info->laggy = (laggy_since.tv_sec != 0 ||
218 			       laggy_since.tv_nsec != 0);
219 		info->num_export_targets = num_export_targets;
220 		if (num_export_targets) {
221 			info->export_targets = kcalloc(num_export_targets,
222 						       sizeof(u32), GFP_NOFS);
223 			if (info->export_targets == NULL)
224 				goto nomem;
225 			for (j = 0; j < num_export_targets; j++)
226 				info->export_targets[j] =
227 				       ceph_decode_32(&pexport_targets);
228 		} else {
229 			info->export_targets = NULL;
230 		}
231 	}
232 
233 	/* pg_pools */
234 	ceph_decode_32_safe(p, end, n, bad);
235 	m->m_num_data_pg_pools = n;
236 	m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS);
237 	if (!m->m_data_pg_pools)
238 		goto nomem;
239 	ceph_decode_need(p, end, sizeof(u64)*(n+1), bad);
240 	for (i = 0; i < n; i++)
241 		m->m_data_pg_pools[i] = ceph_decode_64(p);
242 	m->m_cas_pg_pool = ceph_decode_64(p);
243 	m->m_enabled = m->m_epoch > 1;
244 
245 	mdsmap_ev = 1;
246 	if (mdsmap_v >= 2) {
247 		ceph_decode_16_safe(p, end, mdsmap_ev, bad_ext);
248 	}
249 	if (mdsmap_ev >= 3) {
250 		if (__decode_and_drop_compat_set(p, end) < 0)
251 			goto bad_ext;
252 	}
253 	/* metadata_pool */
254 	if (mdsmap_ev < 5) {
255 		__decode_and_drop_type(p, end, u32, bad_ext);
256 	} else {
257 		__decode_and_drop_type(p, end, u64, bad_ext);
258 	}
259 
260 	/* created + modified + tableserver */
261 	__decode_and_drop_type(p, end, struct ceph_timespec, bad_ext);
262 	__decode_and_drop_type(p, end, struct ceph_timespec, bad_ext);
263 	__decode_and_drop_type(p, end, u32, bad_ext);
264 
265 	/* in */
266 	{
267 		int num_laggy = 0;
268 		ceph_decode_32_safe(p, end, n, bad_ext);
269 		ceph_decode_need(p, end, sizeof(u32) * n, bad_ext);
270 
271 		for (i = 0; i < n; i++) {
272 			s32 mds = ceph_decode_32(p);
273 			if (mds >= 0 && mds < m->m_max_mds) {
274 				if (m->m_info[mds].laggy)
275 					num_laggy++;
276 			}
277 		}
278 		m->m_num_laggy = num_laggy;
279 	}
280 
281 	/* inc */
282 	__decode_and_drop_map(p, end, u32, u32, bad_ext);
283 	/* up */
284 	__decode_and_drop_map(p, end, u32, u64, bad_ext);
285 	/* failed */
286 	__decode_and_drop_set(p, end, u32, bad_ext);
287 	/* stopped */
288 	__decode_and_drop_set(p, end, u32, bad_ext);
289 
290 	if (mdsmap_ev >= 4) {
291 		/* last_failure_osd_epoch */
292 		__decode_and_drop_type(p, end, u32, bad_ext);
293 	}
294 	if (mdsmap_ev >= 6) {
295 		/* ever_allowed_snaps */
296 		__decode_and_drop_type(p, end, u8, bad_ext);
297 		/* explicitly_allowed_snaps */
298 		__decode_and_drop_type(p, end, u8, bad_ext);
299 	}
300 	if (mdsmap_ev >= 7) {
301 		/* inline_data_enabled */
302 		__decode_and_drop_type(p, end, u8, bad_ext);
303 	}
304 	if (mdsmap_ev >= 8) {
305 		u32 name_len;
306 		/* enabled */
307 		ceph_decode_8_safe(p, end, m->m_enabled, bad_ext);
308 		ceph_decode_32_safe(p, end, name_len, bad_ext);
309 		ceph_decode_need(p, end, name_len, bad_ext);
310 		*p += name_len;
311 	}
312 	/* damaged */
313 	if (mdsmap_ev >= 9) {
314 		size_t need;
315 		ceph_decode_32_safe(p, end, n, bad_ext);
316 		need = sizeof(u32) * n;
317 		ceph_decode_need(p, end, need, bad_ext);
318 		*p += need;
319 		m->m_damaged = n > 0;
320 	} else {
321 		m->m_damaged = false;
322 	}
323 bad_ext:
324 	*p = end;
325 	dout("mdsmap_decode success epoch %u\n", m->m_epoch);
326 	return m;
327 nomem:
328 	err = -ENOMEM;
329 	goto out_err;
330 bad:
331 	pr_err("corrupt mdsmap\n");
332 	print_hex_dump(KERN_DEBUG, "mdsmap: ",
333 		       DUMP_PREFIX_OFFSET, 16, 1,
334 		       start, end - start, true);
335 out_err:
336 	ceph_mdsmap_destroy(m);
337 	return ERR_PTR(err);
338 }
339 
340 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
341 {
342 	int i;
343 
344 	for (i = 0; i < m->m_max_mds; i++)
345 		kfree(m->m_info[i].export_targets);
346 	kfree(m->m_info);
347 	kfree(m->m_data_pg_pools);
348 	kfree(m);
349 }
350 
351 bool ceph_mdsmap_is_cluster_available(struct ceph_mdsmap *m)
352 {
353 	int i, nr_active = 0;
354 	if (!m->m_enabled)
355 		return false;
356 	if (m->m_damaged)
357 		return false;
358 	if (m->m_num_laggy > 0)
359 		return false;
360 	for (i = 0; i < m->m_max_mds; i++) {
361 		if (m->m_info[i].state == CEPH_MDS_STATE_ACTIVE)
362 			nr_active++;
363 	}
364 	return nr_active > 0;
365 }
366