[HDFS-11851] getGlobalJNIEnv() may deadlock if exception is thrown

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 3.0.0-alpha4
• Fix Version/s: 3.0.0-alpha4
• Component/s: libhdfs
• Target Version/s: 3.0.0-alpha4
• Assignee: Sailesh Mukil
• Reporter: Henry Robinson
• Created: 18/May/17 19:51
• Updated: 18/Oct/17 08:34
• Resolved: 08/Jun/17 21:20

Description

HDFS-11529 introduced a deadlock into getGlobalJNIEnv() if an exception is thrown. getGlobalJNIEnv() holds jvmMutex, but printExceptionAndFree() will eventually try to acquire that lock in setTLSExceptionStrings().

The exception might get caught from loadFileSystems:

jthr = invokeMethod(env, NULL, STATIC, NULL,
                         "org/apache/hadoop/fs/FileSystem",
                         "loadFileSystems", "()V");
        if (jthr) {
            printExceptionAndFree(env, jthr, PRINT_EXC_ALL, "loadFileSystems");
        }
    }

and here's the relevant parts of the stack trace from where I call this API in Impala, which uses libhdfs:

#0  __lll_lock_wait () at ../nptl/sysdeps/unix/sysv/linux/x86_64/lowlevellock.S:135
#1  0x00007ffff4a8d657 in _L_lock_909 () from /lib/x86_64-linux-gnu/libpthread.so.0
#2  0x00007ffff4a8d480 in __GI___pthread_mutex_lock (mutex=0x47ce960 <jvmMutex>) at ../nptl/pthread_mutex_lock.c:79
#3  0x0000000002f06056 in mutexLock (m=<optimized out>) at /data/2/jenkins/workspace/impala-hadoop-dependency/hadoop/hadoop-hdfs-project/hadoop-hdfs/src/main/native/libhdfs/os/posix/mutexes.c:28
#4  0x0000000002efe817 in setTLSExceptionStrings (rootCause=0x0, stackTrace=0x0) at /data/2/jenkins/workspace/impala-hadoop-dependency/hadoop/hadoop-hdfs-project/hadoop-hdfs/src/main/native/libhdfs/jni_helper.c:581
#5  0x0000000002f065d7 in printExceptionAndFreeV (env=0x513c1e8, exc=0x508a8c0, noPrintFlags=<optimized out>, fmt=0x34349cf "loadFileSystems", ap=0x7fffffffb660)
    at /data/2/jenkins/workspace/impala-hadoop-dependency/hadoop/hadoop-hdfs-project/hadoop-hdfs/src/main/native/libhdfs/exception.c:183
#6  0x0000000002f0683d in printExceptionAndFree (env=<optimized out>, exc=<optimized out>, noPrintFlags=<optimized out>, fmt=<optimized out>)
    at /data/2/jenkins/workspace/impala-hadoop-dependency/hadoop/hadoop-hdfs-project/hadoop-hdfs/src/main/native/libhdfs/exception.c:213
#7  0x0000000002eff60f in getGlobalJNIEnv () at /data/2/jenkins/workspace/impala-hadoop-dependency/hadoop/hadoop-hdfs-project/hadoop-hdfs/src/main/native/libhdfs/jni_helper.c:463

Issue Links

• is broken by: HDFS-11529 Add libHDFS API to return last exception
• is related to: HDFS-12494 libhdfs SIGSEGV in setTLSExceptionStrings
• relates to: IMPALA-6073 No response from expr-codegen-test/expr-test