Merge pull request #6925 from yandex/aku/msan

Some MemorySanitizer fixes

dbms/src/Common/MemorySanitizer.h

@@ -0,0 +1,9 @@
+#pragma once
+
+#define __msan_unpoison(X, Y)
+#if defined(__has_feature)
+#   if __has_feature(memory_sanitizer)
+#       undef __msan_unpoison
+#       include <sanitizer/msan_interface.h>
+#   endif
+#endif

dbms/src/Functions/GeoUtils.cpp

@@ -254,6 +254,9 @@ void geohashDecode(const char * encoded_string, size_t encoded_len, Float64 * lo
     const UInt8 precision = std::min(encoded_len, static_cast<size_t>(MAX_PRECISION));
     if (precision == 0)
     {
+        // Empty string is converted to (0, 0)
+        *longitude = 0;
+        *latitude = 0;
         return;
     }
 

dbms/src/IO/AIOContextPool.cpp

@@ -2,6 +2,7 @@
 
 #include <Common/Exception.h>
 #include <common/logger_useful.h>
+#include <Common/MemorySanitizer.h>
 #include <Poco/Logger.h>
 #include <boost/range/iterator_range.hpp>
 #include <errno.h>
@@ -69,6 +70,9 @@ int AIOContextPool::getCompletionEvents(io_event events[], const int max_events)
         if (errno != EINTR)
             throwFromErrno("io_getevents: Failed to wait for asynchronous IO completion", ErrorCodes::CANNOT_IO_GETEVENTS, errno);
 
+    /// Unpoison the memory returned from a non-instrumented system call.
+    __msan_unpoison(events, sizeof(*events) * num_events);
+
     return num_events;
 }
 

dbms/src/IO/WriteBufferAIO.cpp

@@ -1,6 +1,7 @@
 #if defined(__linux__) || defined(__FreeBSD__)
 
 #include <IO/WriteBufferAIO.h>
+#include <Common/MemorySanitizer.h>
 #include <Common/ProfileEvents.h>
 
 #include <limits>
@@ -200,6 +201,9 @@ bool WriteBufferAIO::waitForAIOCompletion()
         }
     }
 
+    // Unpoison the memory returned from an uninstrumented system function.
+    __msan_unpoison(&event, sizeof(event));
+
     is_pending_write = false;
 #if defined(__FreeBSD__)
     bytes_written = aio_return(reinterpret_cast<struct aiocb *>(event.udata));

dbms/src/IO/ZlibDeflatingWriteBuffer.cpp

@@ -1,4 +1,5 @@
 #include <IO/ZlibDeflatingWriteBuffer.h>
+#include <Common/MemorySanitizer.h>
 
 
 namespace DB
@@ -70,6 +71,12 @@ void ZlibDeflatingWriteBuffer::nextImpl()
         int rc = deflate(&zstr, Z_NO_FLUSH);
         out.position() = out.buffer().end() - zstr.avail_out;
 
+        // Unpoison the result of deflate explicitly. It uses some custom SSE algo
+        // for computing CRC32, and it looks like msan is unable to comprehend
+        // it fully, so it complains about the resulting value depending on the
+        // uninitialized padding of the input buffer.
+        __msan_unpoison(out.position(), zstr.avail_out);
+
         if (rc != Z_OK)
             throw Exception(std::string("deflate failed: ") + zError(rc), ErrorCodes::ZLIB_DEFLATE_FAILED);
     }
@@ -92,6 +99,12 @@ void ZlibDeflatingWriteBuffer::finish()
         int rc = deflate(&zstr, Z_FINISH);
         out.position() = out.buffer().end() - zstr.avail_out;
 
+        // Unpoison the result of deflate explicitly. It uses some custom SSE algo
+        // for computing CRC32, and it looks like msan is unable to comprehend
+        // it fully, so it complains about the resulting value depending on the
+        // uninitialized padding of the input buffer.
+        __msan_unpoison(out.position(), zstr.avail_out);
+
         if (rc == Z_STREAM_END)
         {
             finished = true;

libs/libcommon/src/phdr_cache.cpp

@@ -20,6 +20,13 @@
     #define USE_PHDR_CACHE 1
 #endif
 
+#define __msan_unpoison(X, Y)
+#if defined(__has_feature)
+#   if __has_feature(memory_sanitizer)
+#       undef __msan_unpoison
+#       include <sanitizer/msan_interface.h>
+#   endif
+#endif
 
 /// Thread Sanitizer uses dl_iterate_phdr function on initialization and fails if we provide our own.
 #ifdef USE_PHDR_CACHE
@@ -99,6 +106,10 @@ void updatePHDRCache()
     PHDRCache * new_phdr_cache = new PHDRCache;
     getOriginalDLIteratePHDR()([] (dl_phdr_info * info, size_t /*size*/, void * data)
     {
+        // `info` is created by dl_iterate_phdr, which is a non-instrumented
+        // libc function, so we have to unpoison it manually.
+        __msan_unpoison(info, sizeof(*info));
+
         reinterpret_cast<PHDRCache *>(data)->push_back(*info);
         return 0;
     }, new_phdr_cache);