os/filestore: ceph_abort() on fsync(2) or fdatasync(2) failure

If we get an error from f[data]sync that is always a fatal error.

Fixes: http://tracker.ceph.com/issues/38258
Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 3892b81303bcc35056ca06371651e13b7a4c4f2a)

diff --git a/src/os/filestore/FileStore.cc b/src/os/filestore/FileStore.cc
index 5688131fdfbba65d5c184b5c399c05f2aa72ac98..d3a4259c84dd113f5938f49d55a252e919731b08 100644 (file)
--- a/src/os/filestore/FileStore.cc
+++ b/src/os/filestore/FileStore.cc
@@ -2463,7 +2463,11 @@ void FileStore::_set_global_replay_guard(const coll_t& cid,
   }
 
   // and make sure our xattr is durable.
-  ::fsync(fd);
+  r = ::fsync(fd);
+  if (r < 0) {
+    derr << __func__ << " fsync failed: " << cpp_strerror(errno) << dendl;
+    ceph_abort();
+  }
 
   _inject_failure();
 
@@ -2532,7 +2536,11 @@ void FileStore::_set_replay_guard(int fd,
   _inject_failure();
 
   // first make sure the previous operation commits
-  ::fsync(fd);
+  int r = ::fsync(fd);
+  if (r < 0) {
+    derr << __func__ << " fsync failed: " << cpp_strerror(errno) << dendl;
+    ceph_abort();
+  }
 
   if (!in_progress) {
     // sync object_map too.  even if this object has a header or keys,
@@ -2547,7 +2555,7 @@ void FileStore::_set_replay_guard(int fd,
   bufferlist v(40);
   encode(spos, v);
   encode(in_progress, v);
-  int r = chain_fsetxattr<true, true>(
+  r = chain_fsetxattr<true, true>(
     fd, REPLAY_GUARD_XATTR, v.c_str(), v.length());
   if (r < 0) {
     derr << "fsetxattr " << REPLAY_GUARD_XATTR << " got " << cpp_strerror(r) << dendl;
@@ -2555,7 +2563,11 @@ void FileStore::_set_replay_guard(int fd,
   }
 
   // and make sure our xattr is durable.
-  ::fsync(fd);
+  r = ::fsync(fd);
+  if (r < 0) {
+    derr << __func__ << " fsync failed: " << cpp_strerror(errno) << dendl;
+    ceph_abort();
+  }
 
   _inject_failure();
 
@@ -2605,7 +2617,11 @@ void FileStore::_close_replay_guard(int fd, const SequencerPosition& spos,
   }
 
   // and make sure our xattr is durable.
-  ::fsync(fd);
+  r = ::fsync(fd);
+  if (r < 0) {
+    derr << __func__ << " fsync failed: " << cpp_strerror(errno) << dendl;
+    ceph_abort();
+  }
 
   _inject_failure();
 

diff --git a/src/os/filestore/LFNIndex.cc b/src/os/filestore/LFNIndex.cc
index a35577877d0c48dad1d20d22b8f9276cebbb13f5..2451ae8c705209f4a3e35ca754b296a695aeb50b 100644 (file)
--- a/src/os/filestore/LFNIndex.cc
+++ b/src/os/filestore/LFNIndex.cc
@@ -29,6 +29,7 @@
 #include "common/debug.h"
 #include "include/buffer.h"
 #include "common/ceph_crypto.h"
+#include "common/errno.h"
 #include "include/compat.h"
 #include "chain_xattr.h"
 
@@ -176,10 +177,11 @@ int LFNIndex::fsync_dir(const vector<string> &path)
   maybe_inject_failure();
   int r = ::fsync(fd);
   maybe_inject_failure();
-  if (r < 0)
-    return -errno;
-  else
-    return 0;
+  if (r < 0) {
+    derr << __func__ << " fsync failed: " << cpp_strerror(errno) << dendl;
+    ceph_abort();
+  }
+  return 0;
 }
 
 int LFNIndex::link_object(const vector<string> &from,

diff --git a/src/os/filestore/WBThrottle.cc b/src/os/filestore/WBThrottle.cc
index c87434a4b32a9a21eeb5f8f09dd9953d3744b8a5..988e4378391a52ff676ba4a6cd73be987fcc9633 100644 (file)
--- a/src/os/filestore/WBThrottle.cc
+++ b/src/os/filestore/WBThrottle.cc
@@ -5,6 +5,7 @@
 
 #include "os/filestore/WBThrottle.h"
 #include "common/perf_counters.h"
+#include "common/errno.h"
 
 WBThrottle::WBThrottle(CephContext *cct) :
   cur_ios(0), cur_size(0),
@@ -166,10 +167,14 @@ void *WBThrottle::entry()
     logger->inc(l_wbthrottle_inodes_wb);
     lock.Unlock();
 #if defined(HAVE_FDATASYNC)
-    ::fdatasync(**wb.get<1>());
+    int r = ::fdatasync(**wb.get<1>());
 #else
-    ::fsync(**wb.get<1>());
+    int r = ::fsync(**wb.get<1>());
 #endif
+    if (r < 0) {
+      lderr(cct) << "WBThrottle fsync failed: " << cpp_strerror(errno) << dendl;
+      ceph_abort();
+    }
 #ifdef HAVE_POSIX_FADVISE
     if (cct->_conf->filestore_fadvise && wb.get<2>().nocache) {
       int fa_r = posix_fadvise(**wb.get<1>(), 0, 0, POSIX_FADV_DONTNEED);