src/allmydata/mutable/checker.py

   1
   2 from twisted.internet import defer
   3 from twisted.python import failure
   4 from allmydata import hashtree
   5 from allmydata.uri import from_string
   6 from allmydata.util import hashutil, base32, idlib, log
   7 from allmydata.checker_results import CheckAndRepairResults, CheckerResults
   8
   9 from common import MODE_CHECK, CorruptShareError
  10 from servermap import ServerMap, ServermapUpdater
  11 from layout import unpack_share, SIGNED_PREFIX_LENGTH
  12
  13 class MutableChecker:
  14
  15     def __init__(self, node, monitor):
  16         self._node = node
  17         self._monitor = monitor
  18         self.bad_shares = [] # list of (nodeid,shnum,failure)
  19         self._storage_index = self._node.get_storage_index()
  20         self.results = CheckerResults(from_string(node.get_uri()), self._storage_index)
  21         self.need_repair = False
  22         self.responded = set() # set of (binary) nodeids
  23
  24     def check(self, verify=False):
  25         servermap = ServerMap()
  26         u = ServermapUpdater(self._node, self._monitor, servermap, MODE_CHECK)
  27         self._node._client.notify_mapupdate(u.get_status())
  28         d = u.update()
  29         d.addCallback(self._got_mapupdate_results)
  30         if verify:
  31             d.addCallback(self._verify_all_shares)
  32         d.addCallback(lambda res: servermap)
  33         d.addCallback(self._fill_checker_results, self.results)
  34         d.addCallback(lambda res: self.results)
  35         return d
  36
  37     def _got_mapupdate_results(self, servermap):
  38         # the file is healthy if there is exactly one recoverable version, it
  39         # has at least N distinct shares, and there are no unrecoverable
  40         # versions: all existing shares will be for the same version.
  41         self._monitor.raise_if_cancelled()
  42         self.best_version = None
  43         num_recoverable = len(servermap.recoverable_versions())
  44         if num_recoverable:
  45             self.best_version = servermap.best_recoverable_version()
  46
  47         if servermap.unrecoverable_versions():
  48             self.need_repair = True
  49         if num_recoverable != 1:
  50             self.need_repair = True
  51         if self.best_version:
  52             available_shares = servermap.shares_available()
  53             (num_distinct_shares, k, N) = available_shares[self.best_version]
  54             if num_distinct_shares < N:
  55                 self.need_repair = True
  56
  57         return servermap
  58
  59     def _verify_all_shares(self, servermap):
  60         # read every byte of each share
  61         if not self.best_version:
  62             return
  63         versionmap = servermap.make_versionmap()
  64         shares = versionmap[self.best_version]
  65         (seqnum, root_hash, IV, segsize, datalength, k, N, prefix,
  66          offsets_tuple) = self.best_version
  67         offsets = dict(offsets_tuple)
  68         readv = [ (0, offsets["EOF"]) ]
  69         dl = []
  70         for (shnum, peerid, timestamp) in shares:
  71             ss = servermap.connections[peerid]
  72             d = self._do_read(ss, peerid, self._storage_index, [shnum], readv)
  73             d.addCallback(self._got_answer, peerid, servermap)
  74             dl.append(d)
  75         return defer.DeferredList(dl, fireOnOneErrback=True, consumeErrors=True)
  76
  77     def _do_read(self, ss, peerid, storage_index, shnums, readv):
  78         # isolate the callRemote to a separate method, so tests can subclass
  79         # Publish and override it
  80         d = ss.callRemote("slot_readv", storage_index, shnums, readv)
  81         return d
  82
  83     def _got_answer(self, datavs, peerid, servermap):
  84         for shnum,datav in datavs.items():
  85             data = datav[0]
  86             try:
  87                 self._got_results_one_share(shnum, peerid, data)
  88             except CorruptShareError:
  89                 f = failure.Failure()
  90                 self.need_repair = True
  91                 self.bad_shares.append( (peerid, shnum, f) )
  92                 prefix = data[:SIGNED_PREFIX_LENGTH]
  93                 servermap.mark_bad_share(peerid, shnum, prefix)
  94                 ss = servermap.connections[peerid]
  95                 self.notify_server_corruption(ss, shnum, str(f.value))
  96
  97     def check_prefix(self, peerid, shnum, data):
  98         (seqnum, root_hash, IV, segsize, datalength, k, N, prefix,
  99          offsets_tuple) = self.best_version
 100         got_prefix = data[:SIGNED_PREFIX_LENGTH]
 101         if got_prefix != prefix:
 102             raise CorruptShareError(peerid, shnum,
 103                                     "prefix mismatch: share changed while we were reading it")
 104
 105     def _got_results_one_share(self, shnum, peerid, data):
 106         self.check_prefix(peerid, shnum, data)
 107
 108         # the [seqnum:signature] pieces are validated by _compare_prefix,
 109         # which checks their signature against the pubkey known to be
 110         # associated with this file.
 111
 112         (seqnum, root_hash, IV, k, N, segsize, datalen, pubkey, signature,
 113          share_hash_chain, block_hash_tree, share_data,
 114          enc_privkey) = unpack_share(data)
 115
 116         # validate [share_hash_chain,block_hash_tree,share_data]
 117
 118         leaves = [hashutil.block_hash(share_data)]
 119         t = hashtree.HashTree(leaves)
 120         if list(t) != block_hash_tree:
 121             raise CorruptShareError(peerid, shnum, "block hash tree failure")
 122         share_hash_leaf = t[0]
 123         t2 = hashtree.IncompleteHashTree(N)
 124         # root_hash was checked by the signature
 125         t2.set_hashes({0: root_hash})
 126         try:
 127             t2.set_hashes(hashes=share_hash_chain,
 128                           leaves={shnum: share_hash_leaf})
 129         except (hashtree.BadHashError, hashtree.NotEnoughHashesError,
 130                 IndexError), e:
 131             msg = "corrupt hashes: %s" % (e,)
 132             raise CorruptShareError(peerid, shnum, msg)
 133
 134         # validate enc_privkey: only possible if we have a write-cap
 135         if not self._node.is_readonly():
 136             alleged_privkey_s = self._node._decrypt_privkey(enc_privkey)
 137             alleged_writekey = hashutil.ssk_writekey_hash(alleged_privkey_s)
 138             if alleged_writekey != self._node.get_writekey():
 139                 raise CorruptShareError(peerid, shnum, "invalid privkey")
 140
 141     def notify_server_corruption(self, ss, shnum, reason):
 142         ss.callRemoteOnly("advise_corrupt_share",
 143                           "mutable", self._storage_index, shnum, reason)
 144
 145     def _count_shares(self, smap, version):
 146         available_shares = smap.shares_available()
 147         (num_distinct_shares, k, N) = available_shares[version]
 148         counters = {}
 149         counters["count-shares-good"] = num_distinct_shares
 150         counters["count-shares-needed"] = k
 151         counters["count-shares-expected"] = N
 152         good_hosts = smap.all_peers_for_version(version)
 153         counters["count-good-share-hosts"] = len(good_hosts)
 154         vmap = smap.make_versionmap()
 155         counters["count-wrong-shares"] = sum([len(shares)
 156                                           for verinfo,shares in vmap.items()
 157                                           if verinfo != version])
 158
 159         return counters
 160
 161     def _fill_checker_results(self, smap, r):
 162         self._monitor.raise_if_cancelled()
 163         r.set_servermap(smap.copy())
 164         healthy = True
 165         data = {}
 166         report = []
 167         summary = []
 168         vmap = smap.make_versionmap()
 169         recoverable = smap.recoverable_versions()
 170         unrecoverable = smap.unrecoverable_versions()
 171         data["count-recoverable-versions"] = len(recoverable)
 172         data["count-unrecoverable-versions"] = len(unrecoverable)
 173
 174         if recoverable:
 175             report.append("Recoverable Versions: " +
 176                           "/".join(["%d*%s" % (len(vmap[v]),
 177                                                smap.summarize_version(v))
 178                                     for v in recoverable]))
 179         if unrecoverable:
 180             report.append("Unrecoverable Versions: " +
 181                           "/".join(["%d*%s" % (len(vmap[v]),
 182                                                smap.summarize_version(v))
 183                                     for v in unrecoverable]))
 184         if smap.unrecoverable_versions():
 185             healthy = False
 186             summary.append("some versions are unrecoverable")
 187             report.append("Unhealthy: some versions are unrecoverable")
 188         if len(recoverable) == 0:
 189             healthy = False
 190             summary.append("no versions are recoverable")
 191             report.append("Unhealthy: no versions are recoverable")
 192         if len(recoverable) > 1:
 193             healthy = False
 194             summary.append("multiple versions are recoverable")
 195             report.append("Unhealthy: there are multiple recoverable versions")
 196
 197         needs_rebalancing = False
 198         if recoverable:
 199             best_version = smap.best_recoverable_version()
 200             report.append("Best Recoverable Version: " +
 201                           smap.summarize_version(best_version))
 202             counters = self._count_shares(smap, best_version)
 203             data.update(counters)
 204             s = counters["count-shares-good"]
 205             k = counters["count-shares-needed"]
 206             N = counters["count-shares-expected"]
 207             if s < N:
 208                 healthy = False
 209                 report.append("Unhealthy: best version has only %d shares "
 210                               "(encoding is %d-of-%d)" % (s, k, N))
 211                 summary.append("%d shares (enc %d-of-%d)" % (s, k, N))
 212             hosts = smap.all_peers_for_version(best_version)
 213             needs_rebalancing = bool( len(hosts) < N )
 214         elif unrecoverable:
 215             healthy = False
 216             # find a k and N from somewhere
 217             first = list(unrecoverable)[0]
 218             # not exactly the best version, but that doesn't matter too much
 219             data.update(self._count_shares(smap, first))
 220             # leave needs_rebalancing=False: the file being unrecoverable is
 221             # the bigger problem
 222         else:
 223             # couldn't find anything at all
 224             data["count-shares-good"] = 0
 225             data["count-shares-needed"] = 3 # arbitrary defaults
 226             data["count-shares-expected"] = 10
 227             data["count-good-share-hosts"] = 0
 228             data["count-wrong-shares"] = 0
 229
 230         if self.bad_shares:
 231             data["count-corrupt-shares"] = len(self.bad_shares)
 232             data["list-corrupt-shares"] = locators = []
 233             report.append("Corrupt Shares:")
 234             summary.append("Corrupt Shares:")
 235             for (peerid, shnum, f) in sorted(self.bad_shares):
 236                 locators.append( (peerid, self._storage_index, shnum) )
 237                 s = "%s-sh%d" % (idlib.shortnodeid_b2a(peerid), shnum)
 238                 if f.check(CorruptShareError):
 239                     ft = f.value.reason
 240                 else:
 241                     ft = str(f)
 242                 report.append(" %s: %s" % (s, ft))
 243                 summary.append(s)
 244                 p = (peerid, self._storage_index, shnum, f)
 245                 r.problems.append(p)
 246                 msg = ("CorruptShareError during mutable verify, "
 247                        "peerid=%(peerid)s, si=%(si)s, shnum=%(shnum)d, "
 248                        "where=%(where)s")
 249                 log.msg(format=msg, peerid=idlib.nodeid_b2a(peerid),
 250                         si=base32.b2a(self._storage_index),
 251                         shnum=shnum,
 252                         where=ft,
 253                         level=log.WEIRD, umid="EkK8QA")
 254         else:
 255             data["count-corrupt-shares"] = 0
 256             data["list-corrupt-shares"] = []
 257
 258         sharemap = {}
 259         for verinfo in vmap:
 260             for (shnum, peerid, timestamp) in vmap[verinfo]:
 261                 shareid = "%s-sh%d" % (smap.summarize_version(verinfo), shnum)
 262                 if shareid not in sharemap:
 263                     sharemap[shareid] = []
 264                 sharemap[shareid].append(peerid)
 265         data["sharemap"] = sharemap
 266         data["servers-responding"] = list(smap.reachable_peers)
 267
 268         r.set_healthy(healthy)
 269         r.set_recoverable(bool(recoverable))
 270         r.set_needs_rebalancing(needs_rebalancing)
 271         r.set_data(data)
 272         if healthy:
 273             r.set_summary("Healthy")
 274         else:
 275             r.set_summary("Unhealthy: " + " ".join(summary))
 276         r.set_report(report)
 277
 278
 279 class MutableCheckAndRepairer(MutableChecker):
 280     def __init__(self, node, monitor):
 281         MutableChecker.__init__(self, node, monitor)
 282         self.cr_results = CheckAndRepairResults(self._storage_index)
 283         self.cr_results.pre_repair_results = self.results
 284         self.need_repair = False
 285
 286     def check(self, verify=False):
 287         d = MutableChecker.check(self, verify)
 288         d.addCallback(self._maybe_repair)
 289         d.addCallback(lambda res: self.cr_results)
 290         return d
 291
 292     def _maybe_repair(self, res):
 293         self._monitor.raise_if_cancelled()
 294         if not self.need_repair:
 295             self.cr_results.post_repair_results = self.results
 296             return
 297         self.cr_results.repair_attempted = True
 298         d = self._node.repair(self.results)
 299         def _repair_finished(repair_results):
 300             self.cr_results.repair_successful = True
 301             r = CheckerResults(from_string(self._node.get_uri()), self._storage_index)
 302             self.cr_results.post_repair_results = r
 303             self._fill_checker_results(repair_results.servermap, r)
 304             self.cr_results.repair_results = repair_results # TODO?
 305         def _repair_error(f):
 306             # I'm not sure if I want to pass through a failure or not.
 307             self.cr_results.repair_successful = False
 308             self.cr_results.repair_failure = f # TODO?
 309             #self.cr_results.post_repair_results = ??
 310             return f
 311         d.addCallbacks(_repair_finished, _repair_error)
 312         return d