Re: [PATCH v4 0/4] MR6855: server: Fix renaming file to a hardlink of itself.

[Gabriel Ivăncescu (＠insn)]

On Thu Feb  6 13:12:41 2025 +0000, Jinoh Kang wrote:
> There is probably no fundamental improvements in I/O aspect, although I
> believe the name `is_same_dentry` is much clearer. I like this version better:
> /* Determine if two paths have the same parent directories and
> basenames. */
> static int is_same_dentry( const char *pathname1, const char *pathname2 )
> {
>     const char *lastslash1 = strrchr(pathname1, '/');
>     const char *lastslash2 = strrchr(pathname2, '/');
>     char *dirname1 = NULL, dirname2 = NULL;
>     int res;
>     /* Last slash must be preceded by a non-slash character */
>     assert( lastslash1 && lastslash1 != pathname1 && lastslash1[-1] !=
> '/' );
>     assert( lastslash2 && lastslash2 != pathname2 && lastslash2[-1] !=
> '/' );
>     /* different basenames imply different dentries */
>     if (strcmp( lastslash1 + 1, lastslash2 + 1 ) != 0) return 0;
>     res = -1;
>     if ((dirname1 = memdup( pathname1, lastslash1 + 1 - pathname1 )) &&
>         (dirname2 = memdup( pathname2, lastslash2 + 1 - pathname2 )))
>     {
>         struct stat st1, st2;
>         dirname1[lastslash1 - pathname1] = '\0';
>         dirname2[lastslash2 - pathname2] = '\0';
>         if (stat(dirname1, &st1) || stat(dirname2, &st2))
>         {
>             file_set_error();
>         }
>         else
>         {
>             res = st1.st_dev == st2.st_dev && st1.st_ino == st2.st_ino;
>         }
>     }
>     free(dirname2);
>     free(dirname1);
>     return res;
> }

Are you sure that "name" (I mean the one passed to the function) can't
have trailing slashes or multiple of them? I somehow remember it was a
possible concern at some point long ago when the other case-insens
patchset got reviewed by Rémi, but not sure anymore.