Another inconsistency in the paths handed to stat in the VFS

[Jeremy Allison]

On Wed, Apr 15, 2015 at 10:58:20AM -0700, Richard Sharpe wrote:
> Hi folks,
> 
> Here is the log from a recent test:
> 
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with /home/shares/share1
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with /
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with /home/shares/share1
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with ./.
>   my_test_vfs_stat called with ./..
>   my_test_vfs_stat called with ./testdir2
>   my_test_vfs_stat called with ./testdir6
>   my_test_vfs_stat called with ./testdir13
>   my_test_vfs_stat called with ./testdir8
>   my_test_vfs_stat called with ./testdir4
>   my_test_vfs_stat called with ./testdir11
>   my_test_vfs_stat called with ./somefunnydir
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6
>   my_test_vfs_stat called with test6/.
>   my_test_vfs_stat called with test6/..
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with .
>   my_test_vfs_stat called with /
> 
> Now, there are some problems here that I am working with Jeremy, but
> the inconsistency I wanted to mention is that in some cases
> SMB_VFS_STAT is called with ./ on the front of the path and in other
> cases it is not.
> 
> The following patch seems to fix this problem:
> 
> diff --git a/source3/smbd/dir.c b/source3/smbd/dir.c
> index 36d95d5..c9c6850 100644
> --- a/source3/smbd/dir.c
> +++ b/source3/smbd/dir.c
> @@ -1135,10 +1135,20 @@ bool smbd_dirptr_get_entry(TALLOC_CTX *ctx,
>                         return false;
>                 }
> 
> -               memcpy(pathreal, dirptr->path, pathlen);
> -               pathreal[pathlen] = '/';
> -               memcpy(pathreal + slashlen + pathlen, dname,
> -                      talloc_get_size(dname));
> +               /*
> +                * We don't want to pass ./xxx to modules below us so don't
> +                * add the path if it is just . by itself. Not sure if we
> +                * will ever see ./xxx as a dirpath ...
> +                */
> +               if (dirptr->path && dirptr->path[0] &&
> +                   dirptr->path[0] == '.' && pathlen == 1) {

We have a standard macro

#ifndef ISDOT
#define ISDOT(path) ( \
                        *((const char *)(path)) == '.' && \
                        *(((const char *)(path)) + 1) == '\0' \
                    )
#endif

for part of the above...

> +                       memcpy(pathreal, dname, talloc_get_size(dname));
> +               } else {
> +                       memcpy(pathreal, dirptr->path, pathlen);
> +                       pathreal[pathlen] = '/';
> +                       memcpy(pathreal + slashlen + pathlen, dname,
> +                              talloc_get_size(dname));
> +               }
> 
>                 /* Create smb_fname with NULL stream_name. */
>                 ZERO_STRUCT(smb_fname);
> 
> 
> 
> -- 
> Regards,
> Richard Sharpe
> (何以解憂？唯有杜康。--曹操)