Re: [PATCH 09/11] mount.cc: Implement poor-man's cache

[Corinna Vinschen via Cygwin-patches]

On Feb  3 12:38, Ben wrote:
> 
> 
> On 18-01-2021 12:51, Corinna Vinschen via Cygwin-patches wrote:
> > Ok, so hash_prefix reduces the path to a drive letter or the UNC path
> > prefix and hashes it.  However, what about partitions mounted to a
> > subdir of, say, drive C?  In that case the hashing goes awry, because
> > you're comparing with the hash of drive C while the path is actually
> > pointing to another partition.
> > 
> How can I mount a partition as a subdir of drive C?
> For some reason I can't:
> $ mount /cygdrive/e/Temp/dummy /cygdrive/c/Temp/dummy/dummyone
> mount: /cygdrive/c/Temp/dummy/dummyone: Invalid argument

I wasn't talking about Cygwin mount points, but rather about Windows
mount points.  Since Windows 2000 a partition can be mounted into a
directory of another partition.  Only drive C: (ignoring non-harddisks)
has to be mounted with a drive letter, all others can be mounted just as
on Unix.

But, yeah, Cygwin also supports bind mounts.  Here's an example
from my /etc/fstab.d user mount file:

  //remote/cygwin-src /cygwin nfs binary 0 0
  /cygwin/pub /home/pub nfs bind 0 0


Corinna

Re: [PATCH 09/11] mount.cc: Implement poor-man's cache

[Ben]

On 18-01-2021 12:51, Corinna Vinschen via Cygwin-patches wrote:
> Ok, so hash_prefix reduces the path to a drive letter or the UNC path
> prefix and hashes it.  However, what about partitions mounted to a
> subdir of, say, drive C?  In that case the hashing goes awry, because
> you're comparing with the hash of drive C while the path is actually
> pointing to another partition.
> 
How can I mount a partition as a subdir of drive C?
For some reason I can't:
$ mount /cygdrive/e/Temp/dummy /cygdrive/c/Temp/dummy/dummyone
mount: /cygdrive/c/Temp/dummy/dummyone: Invalid argument

Re: [PATCH 09/11] mount.cc: Implement poor-man's cache

[Corinna Vinschen via Cygwin-patches]

On Jan 15 14:45, Ben Wijen wrote:
> Try to avoid NtQueryVolumeInformationFile.
> ---
>  winsup/cygwin/mount.cc | 78 --
>  winsup/cygwin/mount.h  |  2 +-
>  winsup/cygwin/path.cc  |  2 +-
>  winsup/cygwin/path.h   |  1 +
>  4 files changed, 56 insertions(+), 27 deletions(-)
> 
> diff --git a/winsup/cygwin/mount.cc b/winsup/cygwin/mount.cc
> index e0349815d..1d2b3a61a 100644
> --- a/winsup/cygwin/mount.cc
> +++ b/winsup/cygwin/mount.cc
> @@ -82,6 +82,32 @@ win32_device_name (const char *src_path, char *win32_path, 
> device& dev)
>return true;
>  }
>  
> +static uint32_t
> +hash_prefix (const PUNICODE_STRING upath)
> +{
> +  UNICODE_STRING prefix;
> +  WCHAR *p;
> +
> +  if (upath->Buffer[5] == L':' && upath->Buffer[6] == L'\\')
> +p = upath->Buffer + 6;
> +  else
> +{
> +  /* We're expecting an UNC path.  Move p to the backslash after
> +   "\??\UNC\server\share" or the trailing NUL. */
> +  p = upath->Buffer + 7; /* Skip "\??\UNC" */
> +  int bs_cnt = 0;
> +
> +  while (*++p)
> +if (*p == L'\\')
> +  if (++bs_cnt > 1)
> +break;
> +}
> +  RtlInitCountedUnicodeString (, upath->Buffer,
> +   (p - upath->Buffer) * sizeof(WCHAR));
> +
> +  return hash_path_name ((ino_t) 0, );
> +}
> +

Ok, so hash_prefix reduces the path to a drive letter or the UNC path
prefix and hashes it.  However, what about partitions mounted to a
subdir of, say, drive C?  In that case the hashing goes awry, because
you're comparing with the hash of drive C while the path is actually
pointing to another partition.

> @@ -233,27 +281,7 @@ fs_info::update (PUNICODE_STRING upath, HANDLE in_vol)
>a unique per-drive/share hash. */
>if (ffvi_buf.ffvi.VolumeSerialNumber == 0)
>   {
> -   UNICODE_STRING path_prefix;
> -   WCHAR *p;
> -
> -   if (upath->Buffer[5] == L':' && upath->Buffer[6] == L'\\')
> - p = upath->Buffer + 6;
> -   else
> - {
> -   /* We're expecting an UNC path.  Move p to the backslash after
> -  "\??\UNC\server\share" or the trailing NUL. */
> -   p = upath->Buffer + 7;  /* Skip "\??\UNC" */
> -   int bs_cnt = 0;
> -
> -   while (*++p)
> - if (*p == L'\\')
> - if (++bs_cnt > 1)
> -   break;
> - }
> -   RtlInitCountedUnicodeString (_prefix, upath->Buffer,
> -(p - upath->Buffer) * sizeof (WCHAR));
> -   ffvi_buf.ffvi.VolumeSerialNumber = hash_path_name ((ino_t) 0,
> -  _prefix);
> +   ffvi_buf.ffvi.VolumeSerialNumber = hash_prefix(upath);

Please note that we did this *only* for border case FSes returning a VSN
of 0.  This was sufficient for these cases, but not in a a general sense.


Corinna