DMARC-Filter: OpenDMARC Filter v1.4.2 delorie.com 4APBfr3O2150136
DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org A5E373857B98
MIME-Version: 1.0
References: <CALXu0Uc=wOUbEaTzpfrYdn3Q8rXcaR-if84ZWXPgFJ1qxQbizQ AT mail DOT gmail DOT com>
 <08d6a039-215c-c8e8-0280-15bc0ce28a43 AT jdrake DOT com>
 <CALXu0Ud1ne0F-K5fkHh_Ujp0BvZc6GUMmH3JTSAY_V+sCq8DsQ AT mail DOT gmail DOT com>
In-Reply-To: <CALXu0Ud1ne0F-K5fkHh_Ujp0BvZc6GUMmH3JTSAY_V+sCq8DsQ@mail.gmail.com>
Date: Mon, 25 Nov 2024 12:40:45 +0100
Message-ID: <CAKAoaQnJzpoL0xvi5R9qbmAJV-Q_QtwYaWHCDhTCFzPRoKp+mA@mail.gmail.com>
Subject: Re: [Ms-nfs41-client-devel] Corrupted file name in Cygwin - does
 Cygwin do a silly rename if a file is open?
To: cygwin AT cygwin DOT com, ms-nfs41-client-devel AT lists DOT sourceforge DOT net
From: Roland Mainz via Cygwin <cygwin AT cygwin DOT com>
Reply-To: Roland Mainz <roland DOT mainz AT nrubsig DOT org>
Content-Type: text/plain; charset="utf-8"
Sender: "Cygwin" <cygwin-bounces~archive-cygwin=delorie DOT com AT cygwin DOT com>
Content-Transfer-Encoding: 8bit

On Sun, Nov 24, 2024 at 8:32 AM Cedric Blancher
<cedric DOT blancher AT gmail DOT com> wrote:
>
> On Sat, 23 Nov 2024 at 17:47, Jeremy Drake <cygwin AT jdrake DOT com> wrote:
> >
> > On Sat, 23 Nov 2024, Cedric Blancher via Cygwin wrote:
> >
> > > Good afternoon!
> > >
> > > Does Cygwin do a silly rename if a Cygwin file is open but gets
> > > /bin/rm at the same time?
> >
> > Yes!  See function try_to_bin in winsup/cygwin/syscalls.cc:
> >       /* Create unique filename.  Start with a dot, followed by "cyg"
> >          transposed into the Unicode low surrogate area (U+dc00) on file
> >          systems supporting Unicode (except Samba), followed by the inode
> >          number in hex, followed by a path hash in hex.  The combination
> >          allows to remove multiple hardlinks to the same file. */
>
> That code is wrong.
>
> bash -c 'printf ".\udc63\udc79\udc67#\n"' | iconv -f UTF-8
> .iconv: illegal input sequence at position 1
>
> 334 RtlAppendUnicodeToString (&recycler,
> 335 (pc.fs_flags () & FILE_UNICODE_ON_DISK
> 336 && !pc.fs_is_samba ())
> 337 ? L".\xdc63\xdc79\xdc67" : L".cyg");
>
> SAMBA is right to reject L".\xdc63\xdc79\xdc67", because it is not a
> valid UTF-16 sequence. ReFS with validation, OpenZFS and so on will
> all REJECT such file names, and neither can NFSv4 because file names
> must be valid Unicode (even if nfsd would not validate then filesystem
> being shared via nfsd will reject that).
> So this can only work on ntfs, and only if it is not validating the
> input UTF.16 sequence.
>
> AFAIK FILE_UNICODE_ON_DISK means that the wchar_t sequences must be
> valid UTF-16, and not just be a random sequence of 16bit values.
>
> @Corinna Vinschen Could this sequence please be changed to a VALID
> UTF-8 sequence, such as \u[fffc]\u[fffc]\u[fffc]? That might work with
> SAMBA, ReFS, OpenZFS NFSv4, ...

That does not help with existing Cygwin installations and Cygwin
32bit, which is stuck at Cygwin 3.3.x ... ;-(

I agree that the L".\xdc63\xdc79\xdc67" prefix will backfire on
something like ReFS, OpenZFS etc (SAMBA uses the prefix for
filesystems which do NOT have |FILE_UNICODE_ON_DISK| set), but for
ms-nfs41-client I just stomp over the issues with this patch (wording
still needs to be improved):
---- snip ----
diff --git a/daemon/setattr.c b/daemon/setattr.c
index 9eaafb5..6e9729e 100644
--- a/daemon/setattr.c
+++ b/daemon/setattr.c
@@ -284,6 +284,46 @@ static int handle_nfs41_rename(void
*daemon_context, setattr_upcall_args *args)

    EASSERT((rename->FileNameLength%sizeof(WCHAR)) == 0);

+#define CYGWIN_STOMP_SILLY_RENAME_INVALID_UTF16_SEQUENCE 1
+
+#ifdef CYGWIN_STOMP_SILLY_RENAME_INVALID_UTF16_SEQUENCE
+    /*
+     * Stomp Cygwin "silly rename" invalid Unicode sequence
+     *
+     * Cygwin has it's own variation of "silly rename" (i.e. if
+     * someone deletes a file while someone else still has
+     * a valid fd to that file it first renames that file with a
+     * special prefix, see
+     * newlib-cygwin/winsup/cygwin/syscalls.cc, function
+     * |try_to_bin()|).
+     *
+     * Unfortunately on filesystems supporting Unicode
+     * (i.e. |FILE_UNICODE_ON_DISK|) Cygwin adds the prefix
+     * L".\xdc63\xdc79\xdc67", which is NOT a valid UTF-16 sequence,
+     * and will be rejected by a filesystem validating the
+     * UTF-16 sequence (e.g. SAMBA, ReFS, OpenZFS, ...).
+     * In our case the NFSv4.1 protocol requires valid UTF-8
+     * sequences, and the NFS server will reject filenames if either
+     * the server or the exported filesystem will validate the UTF-8
+     * sequence.
+     *
+     * Since Cygwin only does a |rename()| and never a lookup by
+     * that filename we just stomp the prefix with the prefix used
+     * for non-|FILE_UNICODE_ON_DISK| filesystems.
+     * We ignore the side-effects here, e.g. that Win32 will still
+     * "remember" the original filename in the file name cache.
+     */
+    if ((rename->FileNameLength > (4*sizeof(wchar_t))) &&
+        (!memcmp(rename->FileName,
+            L".\xdc63\xdc79\xdc67", (4*sizeof(wchar_t))))) {
+        DPRINTF(1, ("handle_nfs41_rename(args->path='%s'): "
+            "Cygwin sillyrename prefix \".\\xdc63\\xdc79\\xdc67\" "
+            "detected, squishing prefix to \".cyg\"\n",
+            args->path));
+        (void)memcpy(rename->FileName, L".cyg", 4*sizeof(wchar_t));
+    }
+#endif /* CYGWIN_STOMP_SILLY_RENAME_INVALID_UTF16_SEQUENCE */
+
    dst_path.len = (unsigned short)WideCharToMultiByte(CP_UTF8,
        WC_ERR_INVALID_CHARS|WC_NO_BEST_FIT_CHARS,
        rename->FileName, rename->FileNameLength/sizeof(WCHAR),
---- snip ----

----

Bye,
Roland
-- 
  __ .  . __
 (o.\ \/ /.o) roland DOT mainz AT nrubsig DOT org
  \__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
  /O /==\ O\  TEL +49 641 3992797
 (;O/ \/ \O;)

-- 
Problem reports:      https://cygwin.com/problems.html
FAQ:                  https://cygwin.com/faq/
Documentation:        https://cygwin.com/docs.html
Unsubscribe info:     https://cygwin.com/ml/#unsubscribe-simple