Now instead of closing the existing connection attached to the
stream interface, we first check if the one we pick was attached to
another stream interface, in which case the connections are swapped
if possible (eg: if the current connection is not private). That way
the previous connection remains attached to an existing session and
significantly increases the chances of being reused.
In connect_server(), if we don't have a connection attached to the
stream-int, we first look into the server's idle_conns list and we
pick the first one there, we detach it from its owner if it had one.
If we used to have a connection, we close it.
This mechanism works well but doesn't scale : as servers increase,
the likeliness that the connection attached to the stream interface
doesn't match the server and gets closed increases.
For now it only supports "never", meaning that we never want to reuse a
shared connection, and "always", meaning that we can use any connection
that was not marked private. When "never" is set, this also implies that
no idle connection may become a shared one.
This flag is set on an outgoing connection when this connection gets
some properties that must not be shared with other connections, such
as dynamic transparent source binding, SNI or a proxy protocol header,
or an authentication challenge from the server. This will be needed
later to implement connection reuse.
This function is now dedicated to idle connections only, which means
that it must not be used without any endpoint nor anything not a
connection. The connection remains attached to the stream interface.
For now it's not populated but we have the list entry. It will carry
all idle connections that sessions don't want to share. They may be
used later to reclaim connections upon socket shortage for example.
Since we now always call this function with the reuse parameter cleared,
let's simplify the function's logic as it cannot return the existing
connection anymore. The savings on this inline function are appreciable
(240 bytes) :
$ size haproxy.old haproxy.new
text data bss dec hex filename
1020383 40816 36928 1098127 10c18f haproxy.old
1020143 40816 36928 1097887 10c09f haproxy.new
connect_server() already does most of the check that is done again in
si_alloc_conn(), so let's simply reuse the existing connection instead
of calling the function again. It will also simplify the connection
reuse.
Indeed, for reuse to be set, it also requires srv_conn to be valid. In the
end, the only situation where we have to release the existing connection
and allocate a new one is when reuse == 0.
objt_server() is called multiple times at various places while some
places already make use of srv for this. Let's move the call at the
top of the function and use it all over the place.
Currently it is possible for the current_rule field to be evaluated before
being set, leading to valgrind complaining:
==16783== Conditional jump or move depends on uninitialised value(s)
==16783== at 0x44E662: http_res_get_intercept_rule (proto_http.c:3730)
==16783== by 0x44E662: http_process_res_common (proto_http.c:6528)
==16783== by 0x4797B7: process_stream (stream.c:1851)
==16783== by 0x414634: process_runnable_tasks (task.c:238)
==16783== by 0x40B02F: run_poll_loop (haproxy.c:1528)
==16783== by 0x407F25: main (haproxy.c:1887)
This was introduced by commit 152b81e7b2565862956af883820d4f79177d0651.
Jan A. Bruder reported that some very specific hostnames on server
lines were causing haproxy to crash on startup. Given that hist
backtrace showed some heap corruption, it was obvious there was an
overflow somewhere. The bug in fact is a typo in dns_str_to_dn_label()
which mistakenly copies one extra byte from the host name into the
output value, thus effectively corrupting the structure.
The bug triggers while parsing the next server of similar length
after the corruption, which generally triggers at config time but
could theorically crash at any moment during runtime depending on
what malloc sizes are needed next. This is why it's tagged major.
No backport is needed, this bug was introduced in 1.6-dev2.
This patch allow the existing operators to take a variable as parameter.
This is useful to add the content of two variables. This patch modify
the behavior of operators.
This patch check calculus for overflow and returns capped values.
This permits to protect against integer overflow in certain operations
involving ratios, percentages, limits or anything. That can sometimes
be critically important with some operations (eg: content-length < X).
This patch removes the 32 bits unsigned integer and the 32 bit signed
integer. It replaces these types by a unique type 64 bit signed.
This makes easy the usage of integer and clarify signed and unsigned use.
With the previous version, signed and unsigned are used ones in place of
others, and sometimes the converter loose the sign. For example, divisions
are processed with "unsigned", if one entry is negative, the result is
wrong.
Note that the integer pattern matching and dotted version pattern matching
are already working with signed 64 bits integer values.
There is one user-visible change : the "uint()" and "sint()" sample fetch
functions which used to return a constant integer have been replaced with
a new more natural, unified "int()" function. These functions were only
introduced in the latest 1.6-dev2 so there's no impact on regular
deployments.
This patch adds 3 functions for 64 bit integer conversion.
* lltoa_r : converts signed 64 bit integer to string
* read_uint64 : converts from string to signed 64 bits integer with capping
* read_int64 : converts from string to unsigned 64 bits integer with capping
This patch introduces three new functions which can be used to find a
server in a farm using different server information:
- server unique id (srv->puid)
- server name
- find best match using either name or unique id
When performing best matching, the following applies:
- use the server name first (if provided)
- use the server id if provided
in any case, the function can update the caller about mismatches
encountered.
This flag aims at reporting whether the server unique id (srv->puid) has
been forced by the administrator in HAProxy's configuration.
If not set, it means HAProxy has generated automatically the server's
unique id.
function proxy_find_best_match can update the caller by updating an int
provided in argument.
For now, proxy_find_best_match hardcode bit values 0x01, 0x02 and 0x04,
which is not understandable when reading a code exploiting them.
This patch defines 3 macros with a more explicit wording, so further
reading of a code exploiting the magic bit values will be understandable
more easily.
The man said that gmtime() and localtime() can return a NULL value.
This is not tested. It appears that all the values of a 32 bit integer
are valid, but it is better to check the return of these functions.
However, if the integer move from 32 bits to 64 bits, some 64 values
can be unsupported.
Madison May reported that the timeout applied by the default
configuration is inproperly set up.
This patch fix this:
- hold valid default to 10s
- timeout retry default to 1s
The new "sni" server directive takes a sample fetch expression and
uses its return value as a hostname sent as the TLS SNI extension.
A typical use case consists in forwarding the front connection's SNI
value to the server in a bridged HTTPS forwarder :
sni ssl_fc_sni
ssl_sock_set_servername() is used to set the SNI hostname on an
outgoing connection. This function comes from code originally
provided by Christopher Faulet of Qualys.
When the HTTP forwarder is used, it resets msg->sov so that we know that
the parsing pointer has advanced by exactly (msg->eoh + msg->eol - msg->sov)
bytes which may have to be rewound in case we want to perform an HTTP fetch
after forwarding has started (eg: upon connect).
But when the backend is in TCP mode, there may be no HTTP forwarding
analyser installed, still we may want to perform these HTTP fetches in
case we have already ensured at the TCP layer that we have a properly
parsed HTTP transaction.
In order to solve this, we reset msg->sov before doing a channel_forward()
so that we can still compute http_rewind() on the pending data. That ensures
the buffer is always rewindable even in mixed TCP+HTTP mode.
ARGC_CAP was not added to fmt_directives() which is used to format
error messages when failing to parse log format expressions. The
whole switch/case has been reorganized to match the declaration
order making it easier to spot missing values. The default is not
the "log" directive anymore but "undefined" asking to report the
bug.
Backport to 1.5 is not strictly needed but is desirable at least
for code sanity.
Clients that support ECC cipher suites SHOULD send the specified extension
within the SSL ClientHello message according to RFC4492, section 5.1. We
can use this extension to chain-proxy requests so that, on the same IP
address, a ECC compatible clients gets an EC certificate and a non-ECC
compatible client gets a regular RSA certificate. The main advantage of this
approach compared to the one presented by Dave Zhu on the mailing list
is that we can make it work with OpenSSL versions before 1.0.2.
Example:
frontend ssl-relay
mode tcp
bind 0.0.0.0:443
use_backend ssl-ecc if { req.ssl_ec_ext 1 }
default_backend ssl-rsa
backend ssl-ecc
mode tcp
server ecc unix@/var/run/haproxy_ssl_ecc.sock send-proxy-v2 check
backend ssl-rsa
mode tcp
server rsa unix@/var/run/haproxy_ssl_rsa.sock send-proxy-v2 check
listen all-ssl
bind unix@/var/run/haproxy_ssl_ecc.sock accept-proxy ssl crt /usr/local/haproxy/ecc.foo.com.pem user nobody
bind unix@/var/run/haproxy_ssl_rsa.sock accept-proxy ssl crt /usr/local/haproxy/www.foo.com.pem user nobody
Signed-off-by: Nenad Merdanovic <nmerdan@anine.io>
The current method of retrieving the incoming connection's destination
address to hash it is not compatible with IPv6 nor the proxy protocol
because it directly tries to get an IPv4 address from the socket. Instead
we must ask the connection. This is only used when no SNI is provided.
In src/51d.c, the function _51d_conv(), a final '\0' is added into
smp->data.str.str, which can cause a problem if the SMP_F_CONST flag is
set in smp->flags or if smp->data.str.size is not available.
This patch adds a check on smp->flags and smp->data.str.size, and copies
the smp->data.str.str to another buffer by using smp_dup(). If necessary,
the "const" flag is set after device detection. Also, this patch removes
the unnecessary call to chunk_reset() on temp argument.
This option enables overriding source IP address in a HTTP request. It is
useful when we want to set custom source IP (e.g. front proxy rewrites address,
but provides the correct one in headers) or we wan't to mask source IP address
for privacy or compliance.
It acts on any expression which produces correct IP address.
This modification makes possible to use sample_fetch_string() in more places,
where we might need to fetch sample values which are not plain strings. This
way we don't need to fetch string, and convert it into another type afterwards.
When using aliased types, the caller should explicitly check which exact type
was returned (e.g. SMP_T_IPV4 or SMP_T_IPV6 for SMP_T_ADDR).
All usages of sample_fetch_string() are converted to use new function.
Compression stats were not easy to read and could be confusing because
the saving ratio could be taken for global savings while it was only
relative to compressible input. Let's make that a bit clearer using
the new tooltips with a bit more details and also report the effective
ratio over all output bytes.
Commit cc87a11 ("MEDIUM: tcp: add register keyword system.") broke the
TCP ruleset by merging custom rules and accept. It was fixed a first time
by commit e91ffd0 ("BUG/MAJOR: tcp: only call registered actions when
they're registered") but the accept action still didn't work anymore
and was causing the matching rule to simply be ignored.
Since the code introduced a very fragile behaviour by not even mentionning
that accept and custom were silently merged, let's fix this once for all by
adding an explicit check for the accept action. Nevertheless, as previously
mentionned, the action should be changed so that custom is the only action
and the continue vs break indication directly comes from the callee.
No backport is needed, this bug only affects 1.6-dev.
Until now, the code assumed that it can get the offset to the first TLV
header just by subtracting the length of the TLV part from the length of
the complete buffer. However, if the buffer contains actual data after
the header, this computation is flawed and leads to haproxy trying to
parse TLV headers from the proxied data.
This change fixes this by making sure that the offset to the first TLV
header is calculated based from the start of the buffer -- simply by
adding the size of the proxy protocol v2 header plus the address
family-dependent size of the address information block.
The function buffer_slow_realign() was initially designed for requests
only and did not consider pending outgoing data. This causes a problem
when called on responses where data remain in the buffer, which may
happen with pipelined requests when the client is slow to read data.
The user-visible effect is that if less than <maxrewrite> bytes are
present in the buffer from a previous response and these bytes cross
the <maxrewrite> boundary close to the end of the buffer, then a new
response will cause a realign and will destroy these pending data and
move the pointer to what's believed to contain pending output data.
Thus the client receives the crap that lies in the buffer instead of
the original output bytes.
This new implementation now properly realigns everything including the
outgoing data which are moved to the end of the buffer while the input
data are moved to the beginning.
This implementation still uses a buffer-to-buffer copy which is not
optimal in terms of performance and which should be replaced by a
buffer switch later.
Prior to this patch, the following script would return different hashes
on each round when run from a 100 Mbps-connected machine :
i=0
while usleep 100000; do
echo round $((i++))
set -- $(nc6 0 8001 < 1kreq5k.txt | grep -v '^[0-9A-Z]' | md5sum)
if [ "$1" != "3861afbb6566cd48740ce01edc426020" ]; then echo $1;break;fi
done
The file contains 1000 times this request with "Connection: close" on the
last one :
GET /?s=5k&R=1 HTTP/1.1
The config is very simple :
global
tune.bufsize 16384
tune.maxrewrite 8192
defaults
mode http
timeout client 10s
timeout server 5s
timeout connect 3s
listen px
bind :8001
option http-server-close
server s1 127.0.0.1:8000
And httpterm-1.7.2 is used as the server on port 8000.
After the fix, 1 million requests were sent and all returned the same
contents.
Many thanks to Charlie Smurthwaite of atechmedia.com for his precious
help on this issue, which would not have been diagnosed without his
very detailed traces and numerous tests.
The patch must be backported to 1.5 which is where the bug was introduced.
This is in order to avoid conflicting with NetBSD popcount* functions
since 6.x release, the final l to mentions the argument is a long like
NetBSD does.
This patch could be backported to 1.5 to fix the build issue there as well.
This cache is used by 51d converter. The input User-Agent string, the
converter args and a random seed are used as a hashing key. The cached
entries contains a pointer to the resulting string for specific
User-Agent string detection.
The cache size can be tuned using 51degrees-cache-size parameter.
