]> git.apps.os.sepia.ceph.com Git - ceph.git/commitdiff
crimson/monc: handle_auth_request() doesn't depend on active_con. 41578/head
authorRadoslaw Zarzynski <rzarzyns@redhat.com>
Thu, 27 May 2021 14:55:40 +0000 (14:55 +0000)
committerRadoslaw Zarzynski <rzarzyns@redhat.com>
Thu, 27 May 2021 15:47:45 +0000 (15:47 +0000)
Following crash occured at Sepia [1]:

```
INFO  2021-05-26 20:16:32,872 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] ProtocolV2::start_accept(): targ
et_addr=172.21.15.119:55220/0
DEBUG 2021-05-26 20:16:32,872 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] TRIGGER ACCEPTING, was NONE
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] SEND(26) banner: len_payload=16,
 supported=1, required=0, banner="ceph v2
"
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] RECV(10) banner: "ceph v2
"
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] GOT banner: payload_len=16
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] RECV(16) banner features: supported=1 required=0
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] WRITE HelloFrame: my_type=osd, peer_addr=172.21.15.119:55220/0
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> unknown.? -@55220] GOT HelloFrame: my_type=client peer_addr=v2:172.21.15.119:6803/31733
INFO  2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> client.? -@55220] UPDATE: peer_type=client, policy(lossy=true server=true standby=false resetcheck=false)
DEBUG 2021-05-26 20:16:32,873 [shard 0] ms - [osd.0(client) v2:172.21.15.119:6803/31733 >> client.? -@55220] GOT AuthRequestFrame: method=2, preferred_modes={1, 2}, payload_len=174
/home/jenkins-build/build/workspace/ceph-dev-new-build/ARCH/x86_64/AVAILABLE_ARCH/x86_64/AVAILABLE_DIST/centos8/DIST/centos8/MACHINE_SIZE/gigantic/release/17.0.0-4622-gaa1dc559/rpm/el8/BUILD/ceph-17.0.0-4622-gaa1dc559/src/crimson/mon/MonClient.cc:399:10: runtime error: member access within null pointer of type 'struct Connection'
Segmentation fault on shard 0.
Backtrace:
 0# 0x000055E84CF44C1F in ceph-osd
 1# FatalSignal::signaled(int, siginfo_t const*) in ceph-osd
 2# FatalSignal::install_oneshot_signal_handler<11>()::{lambda(int, siginfo_t*, void*)#1}::_FUN(int, siginfo_t*, void*) in ceph-osd
 3# 0x00007F2BC88C0B20 in /lib64/libpthread.so.0
 4# crimson::mon::Connection::get_conn() in ceph-osd
 5# crimson::mon::Client::handle_auth_request(seastar::shared_ptr<crimson::net::Connection>, seastar::lw_shared_ptr<AuthConnectionMeta>, bool, unsigned int, ceph::buffer::v15_2_0::list const&, ceph::buffer::v15_2_0::list*) in ceph-osd
 6# crimson::net::ProtocolV2::_handle_auth_request(ceph::buffer::v15_2_0::list&, bool) in ceph-osd
 7# 0x000055E84DF67669 in ceph-osd
 8# 0x000055E84DF68775 in ceph-osd
 9# 0x000055E846F47F60 in ceph-osd
10# 0x000055E85296770F in ceph-osd
11# 0x000055E85296CC50 in ceph-osd
12# 0x000055E852B1ECBB in ceph-osd
13# 0x000055E85267C73A in ceph-osd
14# main in ceph-osd
15# __libc_start_main in /lib64/libc.so.6
16# _start in ceph-osd
Fault at location: 0x98
```

[1]: http://pulpito.front.sepia.ceph.com/rzarzynski-2021-05-26_12:20:26-rados-master-distro-basic-smithi/6136907

When the `handle_auth_request()` happens, there is no guarantee
`active_con` is being available. This is reflected in the classical
implementation:

```cpp
int MonClient::handle_auth_request(
  Connection *con,
  // ...
  ceph::buffer::list *reply)
{
  // ...
  bool isvalid = ah->verify_authorizer(
    cct,
    *rotating_secrets,
    payload,
    auth_meta->get_connection_secret_length(),
    reply,
    &con->peer_name,
    &con->peer_global_id,
    &con->peer_caps_info,
    &auth_meta->session_key,
    &auth_meta->connection_secret,
    ac);
```

The patch transplate the same logic to crimson.

Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
src/crimson/mon/MonClient.cc

index 5975c9b4460e1fadd75aeee508c298784008a1c4..20b9d7e602462680f8b07a07fc9ee40edfb34f5d 100644 (file)
@@ -597,7 +597,7 @@ int Client::handle_auth_request(crimson::net::ConnectionRef con,
     auth_meta->get_connection_secret_length(),
     reply,
     &name,
-    &active_con->get_conn()->peer_global_id,
+    &con->peer_global_id,
     &caps_info,
     &auth_meta->session_key,
     &auth_meta->connection_secret,