[Bug 199072] libmd inefficient read() calls

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=199072

            Bug ID: 199072
           Summary: libmd inefficient read() calls
           Product: Base System
           Version: 11.0-CURRENT
          Hardware: Any
                OS: Any
            Status: New
          Severity: Affects Only Me
          Priority: ---
         Component: bin
          Assignee: freebsd-bugs at FreeBSD.org
          Reporter: kcwu at csie.org

src/lib/libmd/mdXhl.c
MDXFileChunk(const char *filename, char *buf, off_t ofs, off_t len)
{
        unsigned char buffer[BUFSIZ];

where BUFSIZE is 1024 from stdio.h

This will make lots of read(2) syscalls. All utilities depend on libmd,
including md5(1) and sha256(1), are affected.

==============
For example, on my system, (file content is already cached in ram)
$ du -h file.bin
4.9G    file.bin
$ time md5 file.bin
MD5 (file.bin) = 422bb7e6759dee9dd1af940f9d9db5ab

real    0m31.848s
user    0m11.500s
sys     0m19.526s

Notice that 19.526s spend in sys.

You could also verify the read size is 1024 using truss(1)
$ truss md5 file.bin
...
read(3,"^3fEiRA\M^S\^]\M-Rf\M-7\M-E\M-g"...,1024) = 1024 (0x400)
...
==============
If modified buffer[] to larger, say 1MB, md5(1) becomes much faster.
$ time md5 file.bin
MD5 (file.bin) = 422bb7e6759dee9dd1af940f9d9db5ab

real    0m14.369s
user    0m11.220s
sys     0m1.933s

-- 
You are receiving this mail because:
You are the assignee for the bug.