OS.FileSys.fullPath gives EINVAL exception on Solaris 10

[binghe]

Hi,

On Solaris 10 (x86), the function OS.FileSys.fullPath strangely gives EINVAL exception on any path string, e.g.:

$ poly
Poly/ML 5.9 Release (Git version v5.9-1-g39d96a2)
> OS.FileSys.fullPath "/";
Exception- SysErr ("Invalid argument", SOME EINVAL) raised

This issue only happens on Solaris 10. (It works fine on Solaris 11.) PolyML 5.8.1 built on Solaris 10 also has the same issue:

$ poly
Poly/ML 5.8.1 Release (Git version v5.8.1)
> OS.FileSys.fullPath "/";
Exception- SysErr ("Invalid argument", SOME EINVAL) raised

Regards,

Chun Tian

[binghe]

Hmmm, I found the root causes. There's a call of realpath(cFileName, NULL) in fullPath (libpolyml/basicio.cpp):

Handle fullPath(TaskData *taskData, Handle filename)
{
    TempString cFileName;

    /* Special case of an empty string. */
    if (PolyStringLength(filename->Word()) == 0) cFileName = strdup(".");
    else cFileName = Poly_string_to_C_alloc(filename->Word());
    if (cFileName == 0) raise_syscall(taskData, "Insufficient memory", NOMEMORY);
    TempCString resBuf(realpath(cFileName, NULL));
    if (resBuf == NULL)
        raise_syscall(taskData, "realpath failed", ERRORNUMBER);
    /* Some versions of Unix don't check the final component
        of a file.  To be consistent try doing a "stat" of
        the resulting string to check it exists. */
    struct stat fbuff;
    if (stat(resBuf, &fbuff) != 0)
        raise_syscall(taskData, "stat failed", ERRORNUMBER);
    return(SAVE(C_string_to_Poly(taskData, resBuf)));
}

On Solaris 10, the man page (3C) of realpath() says the following

Standard C Library Functions                         realpath(3C)

NAME
     realpath - resolve pathname

SYNOPSIS
     #include <stdlib.h>

     char *realpath(const char *restrict  file_name,  char  *res-
     trict resolved_name);

DESCRIPTION
     The realpath() function derives, from the  pathname  pointed
     to  by  file_name,  an absolute pathname that names the same
     file, whose resolution does not involve ".", "..",  or  sym-
     bolic  links.  The  generated  pathname is stored as a null-
     terminated string, up  to  a  maximum  of  {PATH_MAX}  bytes
     (defined  in  limits.h(3HEAD)),  in the buffer pointed to by
     resolved_name.

RETURN VALUES
     On successful completion, realpath() returns  a  pointer  to
     the  resolved  name.   Otherwise,  realpath() returns a null
     pointer and sets errno to indicate the error, and  the  con-
     tents  of the buffer pointed to by resolved_name are left in
     an indeterminate state.

...

ERRORS
     The realpath() function will fail if:

     EACCES          Read or search permission was denied  for  a
                     component of file_name.

     EINVAL          Either the file_name or resolved_name  argu-
                     ment is a null pointer.

Because the 2nd parameter is NULL, realpath returns EINVAL on Solaris 10. I also checked man page of Solaris 11, which only returns EINVAL if the first argument is NULL:

Standard C Library Functions                                      realpath(3C)

NAME
       realpath, frealpath, canonicalize_file_name - resolve pathname

SYNOPSIS
       #include <stdlib.h>

       char *realpath(const char *restrict file_name,
            char *restrict resolved_name);

DESCRIPTION
       The  realpath()  function  derives,  from  the  pathname  pointed to by
       file_name, an absolute pathname that resolves  to  the  same  directory
       entry,  whose resolution does not involve ".", "..", or symbolic links.
       If resolved_name is not null, the generated pathname  is  stored  as  a
       null-terminated  string, up to a maximum of {PATH_MAX} (defined in lim-
       its.h(3HEAD)) bytes in the  buffer  pointed  to  by  resolved_name.  If
       resolved_name  is null, the generated pathname is stored as a null-ter-
       minated string in a buffer that is  allocated  as  if  malloc(3C)  were
       called.

...
ERRORS
       The realpath() function will fail if:

       EACCES          Search permission was denied for  a  component  of  the
                       path prefix of file_name.

       EINVAL          The file_name argument is a null pointer.

Could you please make some changes to support Solaris 10. For example, instead of using the return value of realpath, now we first allocate null-terminated string, up to a maximum of {PATH_MAX}, then use it as the 2nd argument of realpath?

Regards,

Chun Tian

[binghe]

According to Linux man pages [1], glibc versions before 2.3 behaves the same as Solaris 10:

       EINVAL path is NULL.  (In glibc versions before 2.3, this error
              is also returned if resolved_path is NULL.)

Thus my above suggested changes may also make PolyML work better on old Linux systems.

[1] https://man7.org/linux/man-pages/man3/realpath.3.html

[binghe]

The following patch works for me (FILENAME_MAX is a constant defined in C standard [1] but please also see this [2])

diff --git a/libpolyml/basicio.cpp b/libpolyml/basicio.cpp
index 0457524c..77d517ff 100644
--- a/libpolyml/basicio.cpp
+++ b/libpolyml/basicio.cpp
@@ -652,8 +652,10 @@ Handle fullPath(TaskData *taskData, Handle filename)
     if (PolyStringLength(filename->Word()) == 0) cFileName = strdup(".");
     else cFileName = Poly_string_to_C_alloc(filename->Word());
     if (cFileName == 0) raise_syscall(taskData, "Insufficient memory", NOMEMORY);
-    TempCString resBuf(realpath(cFileName, NULL));
-    if (resBuf == NULL)
+
+    /* old UNIX systems require the 2nd argument of realpath() be non-NULL */
+    TempCString resBuf((char *)malloc(FILENAME_MAX * sizeof(char)));
+    if (NULL == realpath(cFileName, resBuf))
         raise_syscall(taskData, "realpath failed", ERRORNUMBER);
     /* Some versions of Unix don't check the final component
         of a file.  To be consistent try doing a "stat" of

[1] https://www.cplusplus.com/reference/cstdio/FILENAME_MAX/
[2] https://stackoverflow.com/questions/65174086/is-there-any-difference-between-filename-max-and-path-max-in-c

[dcjm]

I've added some code to handle this along with a regression test. They are currently in the SolarisRealpath branch. Could you please test this and let me know if it solves it for you. I found an old Solaris 10 installation disc and attempted to install Poly/ML in a virtual machine but ran into so many problems that I have given up.

[binghe]

I've added some code to handle this along with a regression test. They are currently in the SolarisRealpath branch. Could you please test this and let me know if it solves it for you. I found an old Solaris 10 installation disc and attempted to install Poly/ML in a virtual machine but ran into so many problems that I have given up.

Thank you for your time! I just built PolyML in this branch and make a quick test: it works well on Solaris 10:

Poly/ML 5.9.1 Testing (Git version v5.9-182-g0ef5380)
> OS.FileSys.fullPath ".";
val it = "/export/home/binghe": string

P. S. the latest Solaris 10 installation disc can be download at https://www.oracle.com/solaris/solaris10/downloads/solaris10-get-jsp-downloads.html but to build PolyML you need to install GCC 5.5.0 from https://www.opencsw.org and put /opt/csw/bin, etc. into $PATH. I believe GNU make (gmake) is also required.

If you are still interested in having a Solaris x86 VM, please tell me your hypervisor software (VMware, VirtualBox, etc.) and I can prepare an out-of-box Solaris VM for you!

--Chun