nihilist/tor
1
0
Fork 0
mirror of https://gitlab.torproject.org/tpo/core/tor.git synced 2024-11-24 04:13:28 +01:00
Code Issues Packages Projects Releases Wiki Activity

Remove needless flush-on-write code.

Browse Source 
Long ago, before we had cell queues, it was necessary to maybe call
connection_handle_write() from connectino_write_to_buf_impl() on OR
connections, so that we wouldn't get into a loop of reading infinite
amounts of data and queueing it all on an outbuf before bothering to
write any data.

If that doesn't sounds like what our code does now, you're right:
right now, we won't stick more than OR_CONN_HIGHWATER bytes of cells
on an outbuf, and we won't suck more than CELL_QUEUE_HIGHWATER_SIZE
cells off any edge connection. So, there's no more call for that
code.

Removing this code will simplify our data flow, and that should be
something we can all get behind.
This commit is contained in:
Nick Mathewson 2012-07-25 18:34:14 -04:00
parent a9d56289ee
commit d9746bd468
2 changed files with 12 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
changes/bug6468 Normal file
View File

@ -0,0 +1,8 @@
o Minor bugfixes:
- Remove some now-needless code that tried to aggressively flush
OR connections as data was added to them. Since 0.2.0.1-alpha,
our cell queue logic has saved us from the failure mode that
this code was supposed to prevent. Removing this code will limit
the number of baroque control flow paths through Tor's network
logic. Reported pseudonymously on IRC. Fix for bug 6468;
bigfix on 0.2.0.1-alpha.

28
src/or/connection.c
View File

@ -3365,13 +3365,6 @@ connection_flush(connection_t *conn)
return connection_handle_write(conn, 1);
}
/** OpenSSL TLS record size is 16383; this is close. The goal here is to
* push data out as soon as we know there's enough for a TLS record, so
* during periods of high load we won't read entire megabytes from
* input before pushing any data out. It also has the feature of not
* growing huge outbufs unless something is slow. */
#define MIN_TLS_FLUSHLEN 15872
/** Append <b>len</b> bytes of <b>string</b> onto <b>conn</b>'s
* outbuf, and ask it to start writing.
*
@ -3380,10 +3373,9 @@ connection_flush(connection_t *conn)
* negative, this is the last data to be compressed, and the connection's zlib
* state should be flushed.
*
* If it's an OR conn and an entire TLS record is ready, then try to
* flush the record now. Similarly, if it's a local control connection
* and a 64k chunk is ready, try to flush it all, so we don't end up with
* many megabytes of controller info queued at once.
* If it's a local control connection and a 64k chunk is ready, try to flush
* it all, so we don't end up with many megabytes of controller info queued at
* once.
*/
void
_connection_write_to_buf_impl(const char *string, size_t len,
@ -3451,7 +3443,6 @@ _connection_write_to_buf_impl(const char *string, size_t len,
if (zlib) {
conn->outbuf_flushlen += buf_datalen(conn->outbuf) - old_datalen;
} else {
ssize_t extra = 0;
conn->outbuf_flushlen += len;
/* Should we try flushing the outbuf now? */
@ -3461,14 +3452,7 @@ _connection_write_to_buf_impl(const char *string, size_t len,
return;
}
if (conn->type == CONN_TYPE_OR &&
conn->outbuf_flushlen-len < MIN_TLS_FLUSHLEN &&
conn->outbuf_flushlen >= MIN_TLS_FLUSHLEN) {
/* We just pushed outbuf_flushlen to MIN_TLS_FLUSHLEN or above;
* we can send out a full TLS frame now if we like. */
extra = conn->outbuf_flushlen - MIN_TLS_FLUSHLEN;
conn->outbuf_flushlen = MIN_TLS_FLUSHLEN;
} else if (conn->type == CONN_TYPE_CONTROL &&
if (conn->type == CONN_TYPE_CONTROL &&
!connection_is_rate_limited(conn) &&
conn->outbuf_flushlen-len < 1<<16 &&
conn->outbuf_flushlen >= 1<<16) {
@ -3488,10 +3472,6 @@ _connection_write_to_buf_impl(const char *string, size_t len,
}
return;
}
if (extra) {
conn->outbuf_flushlen += extra;
connection_start_writing(conn);
}
}
}