client: quit on failed remount during dentry invalidate test

Fixes: http://tracker.ceph.com/issues/22269
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

diff --git a/PendingReleaseNotes b/PendingReleaseNotes
index 4e5d38f8e94bfaf80c05ba418fe62b4098d3ca42..64c23ff1b4e0890b62c1b27181beb04a28f57441 100644 (file)
--- a/PendingReleaseNotes
+++ b/PendingReleaseNotes
@@ -23,3 +23,10 @@
 
 >= 12.2.2
 ---------
+
+- *CephFS*:
+
+  * The CephFS client now catches failures to clear dentries during startup
+    and refuses to start as consistency and untrimmable cache issues may
+    develop. The new option client_die_on_failed_dentry_invalidate (default:
+    true) may be turned off to allow the client to proceed (dangerous!).

diff --git a/src/ceph_fuse.cc b/src/ceph_fuse.cc
index d536365c3fb93a00473c909cc9028932f786ef09..76fedf50afeb6edb8709b888acf819a14acf79cd 100644 (file)
--- a/src/ceph_fuse.cc
+++ b/src/ceph_fuse.cc
@@ -161,12 +161,14 @@ int main(int argc, const char **argv, const char *envp[]) {
 #if defined(__linux__)
        int ver = get_linux_version();
        assert(ver != 0);
-        auto client_try_dentry_invalidate = g_conf->get_val<bool>(
+        bool client_try_dentry_invalidate = g_conf->get_val<bool>(
           "client_try_dentry_invalidate");
        bool can_invalidate_dentries =
           client_try_dentry_invalidate && ver < KERNEL_VERSION(3, 18, 0);
        int tr = client->test_dentry_handling(can_invalidate_dentries);
-       if (tr != 0) {
+        bool client_die_on_failed_dentry_invalidate = g_conf->get_val<bool>(
+          "client_die_on_failed_dentry_invalidate");
+       if (tr != 0 && client_die_on_failed_dentry_invalidate) {
          cerr << "ceph-fuse[" << getpid()
               << "]: fuse failed dentry invalidate/remount test with error "
               << cpp_strerror(tr) << ", stopping" << std::endl;

diff --git a/src/client/Client.cc b/src/client/Client.cc
index 67074b78d3eadfe673773720306334e9076644f8..d9c7fc30fdd49ebe49b7646634764d531b21e010 100644 (file)
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -238,7 +238,6 @@ Client::Client(Messenger *m, MonClient *mc, Objecter *objecter_)
     getgroups_cb(NULL),
     umask_cb(NULL),
     can_invalidate_dentries(false),
-    require_remount(false),
     async_ino_invalidator(m->cct),
     async_dentry_invalidator(m->cct),
     interrupt_finisher(m->cct),
@@ -4009,6 +4008,32 @@ void Client::remove_session_caps(MetaSession *s)
   sync_cond.Signal();
 }
 
+int Client::_do_remount(void)
+{
+  errno = 0;
+  int r = remount_cb(callback_handle);
+  if (r != 0) {
+    int e = errno;
+    client_t whoami = get_nodeid();
+    if (r == -1) {
+      lderr(cct) <<
+          "failed to remount (to trim kernel dentries): "
+          "errno = " << e << " (" << strerror(e) << ")" << dendl;
+    } else {
+      lderr(cct) <<
+          "failed to remount (to trim kernel dentries): "
+          "return code = " << r << dendl;
+    }
+    bool should_abort = cct->_conf->get_val<bool>("client_die_on_failed_remount") ||
+        cct->_conf->get_val<bool>("client_die_on_failed_dentry_invalidate");
+    if (should_abort && !unmounting) {
+      lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl;
+      ceph_abort();
+    }
+  }
+  return r;
+}
+
 class C_Client_Remount : public Context  {
 private:
   Client *client;
@@ -4016,17 +4041,7 @@ public:
   explicit C_Client_Remount(Client *c) : client(c) {}
   void finish(int r) override {
     assert(r == 0);
-    errno = 0;
-    r = client->remount_cb(client->callback_handle);
-    if (r != 0) {
-      int e = errno;
-      client_t whoami = client->get_nodeid();
-      lderr(client->cct) << "tried to remount (to trim kernel dentries) and got error "
-                        << r << " (errno = " << e << "; " << strerror(e) << ")" << dendl;
-      if (client->require_remount && !client->unmounting) {
-       assert(0 == "failed to remount for kernel dentry trimming");
-      }
-    }
+    client->_do_remount();
   }
 };
 
@@ -10051,21 +10066,19 @@ int Client::test_dentry_handling(bool can_invalidate)
   if (can_invalidate_dentries) {
     assert(dentry_invalidate_cb);
     ldout(cct, 1) << "using dentry_invalidate_cb" << dendl;
+    r = 0;
   } else if (remount_cb) {
     ldout(cct, 1) << "using remount_cb" << dendl;
-    int s = remount_cb(callback_handle);
-    if (s) {
-      lderr(cct) << "Failed to invoke remount, needed to ensure kernel dcache consistency"
-                << dendl;
-    }
-    if (cct->_conf->client_die_on_failed_remount) {
-      require_remount = true;
-      r = s;
-    }
-  } else {
-    lderr(cct) << "no method to invalidate kernel dentry cache; expect issues!" << dendl;
-    if (cct->_conf->client_die_on_failed_remount)
+    r = _do_remount();
+  }
+  if (r) {
+    bool should_abort = cct->_conf->get_val<bool>("client_die_on_failed_dentry_invalidate");
+    if (should_abort) {
+      lderr(cct) << "no method to invalidate kernel dentry cache; quitting!" << dendl;
       ceph_abort();
+    } else {
+      lderr(cct) << "no method to invalidate kernel dentry cache; expect issues!" << dendl;
+    }
   }
   return r;
 }

diff --git a/src/client/Client.h b/src/client/Client.h
index d8d0847150d4e5629d8fee5439abf6a781b4fea1..0c1669c10da2716e44dc85cff0e7174ad1844c8c 100644 (file)
--- a/src/client/Client.h
+++ b/src/client/Client.h
@@ -275,7 +275,6 @@ class Client : public Dispatcher, public md_config_obs_t {
   client_getgroups_callback_t getgroups_cb;
   client_umask_callback_t umask_cb;
   bool can_invalidate_dentries;
-  bool require_remount;
 
   Finisher async_ino_invalidator;
   Finisher async_dentry_invalidator;
@@ -761,6 +760,8 @@ private:
   int _release_fh(Fh *fh);
   void _put_fh(Fh *fh);
 
+  int _do_remount(void);
+  friend class C_Client_Remount;
 
   struct C_Readahead : public Context {
     Client *client;

diff --git a/src/common/legacy_config_opts.h b/src/common/legacy_config_opts.h
index 376ea30837bb66c40e6378e7e5d4f0e7a8f503d0..41163641feb33bf60ec863fa8ca1dd3c98300026 100644 (file)
--- a/src/common/legacy_config_opts.h
+++ b/src/common/legacy_config_opts.h
@@ -374,7 +374,6 @@ OPTION(client_permissions, OPT_BOOL)
 OPTION(client_dirsize_rbytes, OPT_BOOL)
 
 OPTION(client_try_dentry_invalidate, OPT_BOOL) // the client should try to use dentry invaldation instead of remounting, on kernels it believes that will work for
-OPTION(client_die_on_failed_remount, OPT_BOOL)
 OPTION(client_check_pool_perm, OPT_BOOL)
 OPTION(client_use_faked_inos, OPT_BOOL)
 OPTION(client_mds_namespace, OPT_STR)

diff --git a/src/common/options.cc b/src/common/options.cc
index 56d8ea1d016ec9dec33a5b64d43cbce8e97f6ef7..4c6cb0eed82c6fb120af2ef211ed92fc899ad237 100644 (file)
--- a/src/common/options.cc
+++ b/src/common/options.cc
@@ -6595,10 +6595,15 @@ std::vector<Option> get_mds_client_options() {
     .set_default(false)
     .set_description(""),
 
-    Option("client_die_on_failed_remount", Option::TYPE_BOOL, Option::LEVEL_ADVANCED)
-    .set_default(true)
+    Option("client_die_on_failed_remount", Option::TYPE_BOOL, Option::LEVEL_DEV)
+    .set_default(false)
     .set_description(""),
 
+    Option("client_die_on_failed_dentry_invalidate", Option::TYPE_BOOL, Option::LEVEL_ADVANCED)
+    .set_default(true)
+    .set_description("kill the client when no dentry invalidation options are available")
+    .set_long_description("The CephFS client requires a mechanism to invalidate dentries in the caller (e.g. the kernel for ceph-fuse) when capabilities must be recalled. If the client cannot do this then the MDS cache cannot shrink which can cause the MDS to fail."),
+
     Option("client_check_pool_perm", Option::TYPE_BOOL, Option::LEVEL_ADVANCED)
     .set_default(true)
     .set_description(""),