fs/ceph/mdsmap.c

   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         char r;
  24
  25         /* count */
  26         for (i = 0; i < m->m_max_mds; i++)
  27                 if (m->m_info[i].state > 0)
  28                         n++;
  29         if (n == 0)
  30                 return -1;
  31
  32         /* pick */
  33         get_random_bytes(&r, 1);
  34         n = r % n;
  35         i = 0;
  36         for (i = 0; n > 0; i++, n--)
  37                 while (m->m_info[i].state <= 0)
  38                         i++;
  39
  40         return i;
  41 }
  42
  43 /*
  44  * Decode an MDS map
  45  *
  46  * Ignore any fields we don't care about (there are quite a few of
  47  * them).
  48  */
  49 struct ceph_mdsmap *ceph_mdsmap_decode(void **p, void *end)
  50 {
  51         struct ceph_mdsmap *m;
  52         const void *start = *p;
  53         int i, j, n;
  54         int err = -EINVAL;
  55         u16 version;
  56
  57         m = kzalloc(sizeof(*m), GFP_NOFS);
  58         if (m == NULL)
  59                 return ERR_PTR(-ENOMEM);
  60
  61         ceph_decode_16_safe(p, end, version, bad);
  62         if (version > 3) {
  63                 pr_warning("got mdsmap version %d > 3, failing", version);
  64                 goto bad;
  65         }
  66
  67         ceph_decode_need(p, end, 8*sizeof(u32) + sizeof(u64), bad);
  68         m->m_epoch = ceph_decode_32(p);
  69         m->m_client_epoch = ceph_decode_32(p);
  70         m->m_last_failure = ceph_decode_32(p);
  71         m->m_root = ceph_decode_32(p);
  72         m->m_session_timeout = ceph_decode_32(p);
  73         m->m_session_autoclose = ceph_decode_32(p);
  74         m->m_max_file_size = ceph_decode_64(p);
  75         m->m_max_mds = ceph_decode_32(p);
  76
  77         m->m_info = kcalloc(m->m_max_mds, sizeof(*m->m_info), GFP_NOFS);
  78         if (m->m_info == NULL)
  79                 goto badmem;
  80
  81         /* pick out active nodes from mds_info (state > 0) */
  82         n = ceph_decode_32(p);
  83         for (i = 0; i < n; i++) {
  84                 u64 global_id;
  85                 u32 namelen;
  86                 s32 mds, inc, state;
  87                 u64 state_seq;
  88                 u8 infoversion;
  89                 struct ceph_entity_addr addr;
  90                 u32 num_export_targets;
  91                 void *pexport_targets = NULL;
  92                 struct ceph_timespec laggy_since;
  93
  94                 ceph_decode_need(p, end, sizeof(u64)*2 + 1 + sizeof(u32), bad);
  95                 global_id = ceph_decode_64(p);
  96                 infoversion = ceph_decode_8(p);
  97                 *p += sizeof(u64);
  98                 namelen = ceph_decode_32(p);  /* skip mds name */
  99                 *p += namelen;
 100
 101                 ceph_decode_need(p, end,
 102                                  4*sizeof(u32) + sizeof(u64) +
 103                                  sizeof(addr) + sizeof(struct ceph_timespec),
 104                                  bad);
 105                 mds = ceph_decode_32(p);
 106                 inc = ceph_decode_32(p);
 107                 state = ceph_decode_32(p);
 108                 state_seq = ceph_decode_64(p);
 109                 ceph_decode_copy(p, &addr, sizeof(addr));
 110                 ceph_decode_addr(&addr);
 111                 ceph_decode_copy(p, &laggy_since, sizeof(laggy_since));
 112                 *p += sizeof(u32);
 113                 ceph_decode_32_safe(p, end, namelen, bad);
 114                 *p += namelen;
 115                 if (infoversion >= 2) {
 116                         ceph_decode_32_safe(p, end, num_export_targets, bad);
 117                         pexport_targets = *p;
 118                         *p += num_export_targets * sizeof(u32);
 119                 } else {
 120                         num_export_targets = 0;
 121                 }
 122
 123                 dout("mdsmap_decode %d/%d %lld mds%d.%d %s %s\n",
 124                      i+1, n, global_id, mds, inc,
 125                      ceph_pr_addr(&addr.in_addr),
 126                      ceph_mds_state_name(state));
 127                 if (mds >= 0 && mds < m->m_max_mds && state > 0) {
 128                         m->m_info[mds].global_id = global_id;
 129                         m->m_info[mds].state = state;
 130                         m->m_info[mds].addr = addr;
 131                         m->m_info[mds].laggy =
 132                                 (laggy_since.tv_sec != 0 ||
 133                                  laggy_since.tv_nsec != 0);
 134                         m->m_info[mds].num_export_targets = num_export_targets;
 135                         if (num_export_targets) {
 136                                 m->m_info[mds].export_targets =
 137                                         kcalloc(num_export_targets, sizeof(u32),
 138                                                 GFP_NOFS);
 139                                 for (j = 0; j < num_export_targets; j++)
 140                                         m->m_info[mds].export_targets[j] =
 141                                                ceph_decode_32(&pexport_targets);
 142                         } else {
 143                                 m->m_info[mds].export_targets = NULL;
 144                         }
 145                 }
 146         }
 147
 148         /* pg_pools */
 149         ceph_decode_32_safe(p, end, n, bad);
 150         m->m_num_data_pg_pools = n;
 151         m->m_data_pg_pools = kcalloc(n, sizeof(u64), GFP_NOFS);
 152         if (!m->m_data_pg_pools)
 153                 goto badmem;
 154         ceph_decode_need(p, end, sizeof(u64)*(n+1), bad);
 155         for (i = 0; i < n; i++)
 156                 m->m_data_pg_pools[i] = ceph_decode_64(p);
 157         m->m_cas_pg_pool = ceph_decode_64(p);
 158
 159         /* ok, we don't care about the rest. */
 160         dout("mdsmap_decode success epoch %u\n", m->m_epoch);
 161         return m;
 162
 163 badmem:
 164         err = -ENOMEM;
 165 bad:
 166         pr_err("corrupt mdsmap\n");
 167         print_hex_dump(KERN_DEBUG, "mdsmap: ",
 168                        DUMP_PREFIX_OFFSET, 16, 1,
 169                        start, end - start, true);
 170         ceph_mdsmap_destroy(m);
 171         return ERR_PTR(-EINVAL);
 172 }
 173
 174 void ceph_mdsmap_destroy(struct ceph_mdsmap *m)
 175 {
 176         int i;
 177
 178         for (i = 0; i < m->m_max_mds; i++)
 179                 kfree(m->m_info[i].export_targets);
 180         kfree(m->m_info);
 181         kfree(m->m_data_pg_pools);
 182         kfree(m);
 183 }