Humm... Looking deeper on my plugin i saw that have a call to the
i_stream_create_header_filter:
full_input[0] = i_stream_create_header_filter(input, HEADER_FILTER_EXCLUDE |
HEADER_FILTER_NO_CR, exclude_headers, 3, filter_callback, &zuser->hash_value);
I need this because i really have 3 lines on the header that i need to hide
from the client...
And i have a call to i_stream_create_concat too:
imail->data.stream = i_stream_create_concat(full_input);
Because i have a header and body separation too just for the fact that the
header could be searched via dovecot cache once it remains un-cryptographed.
So... The size of S flag before dovecot change it is exactly the header size
minus 3 lines of the header plus the un-cryptographed body content.
I notice too two things:
- that dovecot does the rename of the file and can open the message (including
the un-cryptographed body part) but removes the W flag on the filename. Isn't
the W flag used anymore?
- that the size used by dovecot when it renames the file is only the size of
the header part, excluding the body part... I did a test with a larger message
and got this at the log:
Oct 17 17:24:27 test dovecot: imap(reina...@exemplo.com.br): Error: Cached
message size larger than expected (3472 > 1010)
Oct 17 17:24:27 test dovecot: imap(reina...@exemplo.com.br): Error: Maildir
filename has wrong S value, renamed the file from
/storage/test/messages/exemplo.com.br/reinaldo/Maildir/cur/1382041442.M605987P2439V2051I475262.test,S=3472,W=3538:2,
to
/storage/test/messages/exemplo.com.br/reinaldo/Maildir/cur/1382041442.M605987P2439V2051I475262.test,S=1010:2,
Oct 17 17:24:27 test dovecot: imap(reina...@exemplo.com.br): Error: Corrupted
index cache file
/storage/test/messages/exemplo.com.br/reinaldo/Maildir/dovecot.index.cache:
Broken physical size for mail UID 18
Oct 17 17:24:27 test dovecot: imap(reina...@exemplo.com.br): Error: read()
failed: Input/output error (FETCH for mailbox INBOX UID 18)
Oct 17 17:24:27 test dovecot: imap(reina...@exemplo.com.br): Disconnected:
Internal error occurred. Refer to server log for more information. [2013-10-17
17:24:27] in=339 out=2915
So... I'm assuming that on dovecot 2.0.19 the index.cache was indexing all the
header and un-cryptographed body, really using the information on S and W flags.
Does the dovecot 2.1.17 index work in another way, needing to index only the
header part of the message?
> From: ramatuk...@hotmail.com
> To: dovecot@dovecot.org
> Date: Tue, 15 Oct 2013 17:33:44 -0300
> Subject: [Dovecot] Plugin issue with update from 2.0.19 to 2.1.17
>
> Hello. Probably only Timo can help-me with this.
>
> I have a self-made plugin based on the zlib plugin that i use to cryptograph
> the messages at inbox.
>
> As a side-effect of the cryptography, my plugin changes the size of the
> message, but until 2.0.19 this works well with dovecot index and the W/S
> flags.
>
> But now, i'm going to upgrade to 2.1.17 and now i have these messages on log
> at my test ambiance:
>
> Oct 15 20:19:25 test dovecot: imap(reina...@exemplo.com.br): Error: Cached
> message size smaller than expected (367 < 529)
> Oct 15 20:19:25 test dovecot: imap(reina...@exemplo.com.br): Error: Maildir
> filename has wrong S value, renamed the file from
> /storage/test/messages/exemplo.com.br/reinaldo/Maildir/.Sent/cur/1381879158.M634385P5208.test,S=367,W=378:2,S
> to
> /storage/test/messages/exemplo.com.br/reinaldo/Maildir/.Sent/cur/1381879158.M634385P5208.test,S=529:2,S
> Oct 15 20:19:25 test dovecot: imap(reina...@exemplo.com.br): Error: Corrupted
> index cache file
> /storage/test/messages/exemplo.com.br/reinaldo/Maildir/.Sent/dovecot.index.cache:
> Broken physical size for mail UID 6
>
> I understood with these messages that the dovecot is arguing to get the real
> size of the file now, overriding the return of size from my crypto plugin.
> But i don't understood if this is a consequence from the changes to correct
> the index issue (the dovecot.index.cache issue).
>
> So... I want know if the correct way to fix this would be change my plugin to
> return the real size of the file (that will be larger than the effective
> message that the plugin returne after the de-cryptography) or if i need use
> some new function to the plugin replace the expected size based on the real
> size of the message, not of the file.
>
> Thanks
>
> Reinaldo
>
