It's extremely painful to have to set "trace <src> sink buf1" for all
sources, then to do the same for "level developer" (for example). Let's
have a possibility via a meta-source "all" to apply the change to all
sources at once. This currently supports level and sink, which are not
dependent on the source, this is a good start.
The event emitted by this trace was of type CLOSE instead of NEW, which
would somtimes temporarily pause a started trace.
This can be backported to 3.0, probably 2.6.
The test was was performed but there's no way to set the option! Let's
just add "qconn" to select the quic conn when the source supports it.
This can be backported at least to 3.0, probably 2.6.
The doc clearly says that "start <evt>" should leave the trace in pause
mode until the indicated event appears. However it's not what's happening,
the state is not changed until one command uses "now", so it's typically
needed to configure the events with "start <evt>" then enable the waiting
mode using "pause now". This is counter-intuitive and does not match the
doc, so let's fix it so that "start <evt>" switches from stopped to waiting
as long as at least one event is enabled.
This can be backported to all versions.
When calling TRACE_ENABLED(), which is called by TRACE_PRINTF(), we pass
a NULL plockptr to __trace_enabled(). This argument is used when lockon
is active, and may update the pointer. This is an overlook which also
broke the lockon mechanism because now for calls from __trace(), it
dereferences a pointer pointing to NULL, and never updates it due to the
broken condition, so that trace() never sets up src->lockon_ptr.
The bug was introduced in 2.8 by commit 8f9a9704bb ("MINOR: trace: add a
TRACE_ENABLED() macro to determine if a trace is active"), so the fix must
be backported there.
These ones were not proposed in the list of trackable elements. Note
that this depends on previous commit:
BUG/MINOR: trace/quic: enable conn/session pointer recovery from quic_conn
This should be backported to at least 3.0, maybe even 2.6.
In __trace_enabled(), a quic_conn was detected, but it was not possible
to derive the connection nor the session from it, which was quite limiting
in terms of ability to track a same instance.
This should be backported to at least 3.0, maybe even 2.6.
Reuse newly defined tot_time structure to measure various values related
to a QCS lifetime.
First, a timer is used to comptabilize the total QCS lifetime. Then, two
other timers are used to account the total time during which Tx from
stream layer to MUX is blocked, either on lack of buffer or due to
flow-control.
These three timers are reported in qmux_dump_qcs_info(). Thus, they are
available in traces and for QUIC MUX debug string sample.
Define a new utility type tot_time. Its purpose is to be able to account
elapsed time accross multiple periods. Functions are defined to easily
start and stop measures, and return the current value.
Define a new xprt_ops callback named dump_info. This can be used to
extend MUX debug string with infos from the lower layer.
Implement dump_info for QUIC stack. For now, only minimal info are
reported : bytes in flight and size of the sending window. This should
allow to detect if the congestion controller is fine. These info are
reported via QUIC MUX debug string sample.
Implement MUX_SCTL_DBG_STR for QUIC MUX. This returns info for the
current QCS and QCC instances, reusing qmux_dump_qc{c,s}_info functions
already used for traces, and the connection flags.
This stream operation is useful for debug string sample support.
Extract trace code to dump QCC and QCS instances into dedicated
functions named qmux_dump_qc{c,s}_info(). This will allow to easily
print QCC/QCS infos outside of traces.
These are passed to the underlying mux to retrieve debug information
at the mux level (stream/connection) as a string that's meant to be
added to logs.
The API is quite complex just because we can't pass any info to the
bottom function. So we construct a union and pass the argument as an
int, and expect the callee to fill that with its buffer in return.
Most likely the mux->ctl and ->sctl API should be reworked before
the release to simplify this.
The functions take an optional argument that is a bit mask of the
layers to dump:
muxs=1
muxc=2
xprt=4
conn=8
sock=16
The default (0) logs everything available.
These must be before {bs,fs}.id, not after. Should be backported wherever
068ce2d5d2 ("MINOR: stconn: Add samples to retrieve about stream aborts")
is (normally 3.0).
Add a new BUG_ON() in qc-stream_desc_ack(). It ensures that
acknowledgement are always notify in-order. This is because out-of-order
ACKs cannot be handled by qc_stream_desc layer which does not support
gap in STREAM sent data.
Prior to this fix, out-of-order ACKs are simply ignored without any
error. This currently cannot happen thanks to careful
qc_stream_desc_ack() invokation. If this assumption is broken in the
future by inatteion, this would cause loss of ACK notification which
will prevent qc_stream_desc release.
STREAM frames have dedicated handling on retransmission. A special check
is done to remove data already acked in case of duplicated frames, thus
only unacked data are retransmitted.
This handling is faulty in case of an empty STREAM frame with FIN set.
On retransmission, this frame does not cover any unacked range as it is
empty and is thus discarded. This may cause the transfer to freeze with
the client waiting indefinitely for the FIN notification.
To handle retransmission of empty FIN STREAM frame, qc_stream_desc layer
have been extended. A new flag QC_SD_FL_WAIT_FOR_FIN is set by MUX QUIC
when FIN has been transmitted. If set, it prevents qc_stream_desc to be
freed until FIN is acknowledged. On retransmission side,
qc_stream_frm_is_acked() has been updated. It now reports false if
FIN bit is set on the frame and qc_stream_desc has QC_SD_FL_WAIT_FOR_FIN
set.
This must be backported up to 2.6. However, this modifies heavily
critical section for ACK handling and retransmission. As such, it must
be backported only after a period of observation.
This issue can be reproduced by using the following socat command as
server to add delay between the response and connection closure :
$ socat TCP-LISTEN:<port>,fork,reuseaddr,crlf SYSTEM:'echo "HTTP/1.1 200 OK"; echo ""; sleep 1;'
On the client side, ngtcp2 can be used to simulate packet drop. Without
this patch, connection will be interrupted on QUIC idle timeout or
haproxy client timeout with ERR_DRAINING on ngtcp2 :
$ ngtcp2-client --exit-on-all-streams-close -r 0.3 <host> <port> "http://<host>:<port>/?s=32o"
Alternatively to ngtcp2 random loss, an extra haproxy patch can also be
used to force skipping the emission of the empty STREAM frame :
diff --git a/include/haproxy/quic_tx-t.h b/include/haproxy/quic_tx-t.h
index efbdfe687..1ff899acd 100644
--- a/include/haproxy/quic_tx-t.h
+++ b/include/haproxy/quic_tx-t.h
@@ -26,6 +26,8 @@ extern struct pool_head *pool_head_quic_cc_buf;
/* Flag a sent packet as being probing with old data */
#define QUIC_FL_TX_PACKET_PROBE_WITH_OLD_DATA (1UL << 5)
+#define QUIC_FL_TX_PACKET_SKIP_SENDTO (1UL << 6)
+
/* Structure to store enough information about TX QUIC packets. */
struct quic_tx_packet {
/* List entry point. */
diff --git a/src/quic_tx.c b/src/quic_tx.c
index 2f199ac3c..2702fc9b9 100644
--- a/src/quic_tx.c
+++ b/src/quic_tx.c
@@ -318,7 +318,7 @@ static int qc_send_ppkts(struct buffer *buf, struct ssl_sock_ctx *ctx)
tmpbuf.size = tmpbuf.data = dglen;
TRACE_PROTO("TX dgram", QUIC_EV_CONN_SPPKTS, qc);
- if (!skip_sendto) {
+ if (!skip_sendto && !(first_pkt->flags & QUIC_FL_TX_PACKET_SKIP_SENDTO)) {
int ret = qc_snd_buf(qc, &tmpbuf, tmpbuf.data, 0, gso);
if (ret < 0) {
if (gso && ret == -EIO) {
@@ -354,6 +354,7 @@ static int qc_send_ppkts(struct buffer *buf, struct ssl_sock_ctx *ctx)
qc->cntrs.sent_bytes_gso += ret;
}
}
+ first_pkt->flags &= ~QUIC_FL_TX_PACKET_SKIP_SENDTO;
b_del(buf, dglen + QUIC_DGRAM_HEADLEN);
qc->bytes.tx += tmpbuf.data;
@@ -2066,6 +2067,17 @@ static int qc_do_build_pkt(unsigned char *pos, const unsigned char *end,
continue;
}
+ switch (cf->type) {
+ case QUIC_FT_STREAM_8 ... QUIC_FT_STREAM_F:
+ if (!cf->stream.len && (qc->flags & QUIC_FL_CONN_TX_MUX_CONTEXT)) {
+ TRACE_USER("artificially drop packet with empty STREAM frame", QUIC_EV_CONN_TXPKT, qc);
+ pkt->flags |= QUIC_FL_TX_PACKET_SKIP_SENDTO;
+ }
+ break;
+ default:
+ break;
+ }
+
quic_tx_packet_refinc(pkt);
cf->pkt = pkt;
}
When a STREAM frame is retransmitted, a check is performed to remove
range of data already acked from it. This is useful when STREAM frames
are duplicated and splitted to cover different data ranges. The newly
retransmitted frame contains only unacked data.
This process is performed similarly in qc_dup_pkt_frms() and
qc_build_frms(). Refactor the code into a new function named
qc_stream_frm_is_acked(). It returns true if frame data are already
fully acked and retransmission can be avoided. If only a partial range
of data is acknowledged, frame content is updated to only cover the
unacked data.
This patch does not have any functional change. However, it simplifies
retransmission for STREAM frames. Also, it will be reused to fix
retransmission for empty STREAM frames with FIN set from the following
patch :
BUG/MEDIUM: quic: handle retransmit for standalone FIN STREAM
As such, it must be backported prior to it.
qc_stream_desc had a field <release> used as a boolean. Convert it with
a new <flags> field and QC_SD_FL_RELEASE value as equivalent.
The purpose of this patch is to be able to extend qc_stream_desc by
adding newer flags values. This patch is required for the following
patch
BUG/MEDIUM: quic: handle retransmit for standalone FIN STREAM
As such, it must be backported prior to it.
An issue has been introduced with cd99440 ("BUG/MAJOR: server/addr: fix
a race during server addr:svc_port updates").
Indeed, in the above commit we implemented the atomic_sync task which is
responsible for consuming pending server events to apply the changes
atomically. For now only server's addr updates are concerned.
To prevent the task from causing contention, a budget was assigned to it.
It can be controlled with the global tunable
'tune.events.max-events-at-once': the task may not process more than this
number of events at once.
However, a bug was introduced with this budget logic: each time the task
has to be interrupted because it runs out of budget, we reschedule the
task to finish where it left off, but the current event which was already
removed from the queue wasn't processed yet. This means that this pending
event (each tune.events.max-events-at-once) is effectively lost.
When the atomic_sync task deals with large number of concurrent events,
this bug has 2 known consequences: first a server's addr/port update
will be lost every 'tune.events.max-events-at-once'. This can of course
cause reliability issues because if the event is not republished
periodically, the server could stay in a stale state for indefinite amount
of time. This is the case when the DNS server flaps for instance: some
servers may not come back UP after the incident as described in GH #2666.
Another issue is that the lost event was not cleaned up, resulting in a
small memory leak. So in the end, it means that the bug is likely to
cause more and more degradation over time until haproxy is restarted.
As a workaround, 'tune.events.max-events-at-once' may be set to the
maximum number of events expected per batch. Note however that this value
cannot exceed 10 000, otherwise it could cause the watchdog to trigger due
to the task being busy for too long and preventing other threads from
making any progress. Setting higher values may not be optimal for common
workloads so it should only be used to mitigate the bug while waiting for
this fix.
Since tune.events.max-events-at-once defaults to 100, this bug only
affects configs that involve more than 100 servers whose addr:port
properties are likely to be updated at the same time (batched updates
from cli, lua, dns..)
To fix the bug, we move the budget check after the current event is fully
handled. For that we went from a basic 'while' to 'do..while' loop as we
assume from the config that 'tune.events.max-events-at-once' cannot be 0.
While at it, we reschedule the task once thread isolation ends (it was not
required to perform the reschedule while under isolation) to give the hand
back faster to waiting threads.
This patch should be backported up to 2.9 with cd99440. It should fix
GH #2666.
This issue was reported by Ilya (@Chipitsine) when building haproxy against
aws-lc in GH #2663 where handshakeloss and handshakecorruption interop tests could
lead haproxy to crash after having built too short datagrams:
FATAL: bug condition "first_pkt->type == QUIC_PACKET_TYPE_INITIAL && (first_pkt->flags & (1UL << 0)) && length < 1200" matched at src/quic_tx.c:163
call trace(13):
| 0x55f4ee4dcc02 [ba d9 00 00 00 48 8d 35]: main-0x195bf2
| 0x55f4ee4e3112 [83 3d 2f 16 35 00 00 0f]: qc_send+0x11f3/0x1b5d
| 0x55f4ee4e9ab4 [85 c0 0f 85 00 f6 ff ff]: quic_conn_io_cb+0xab1/0xf1c
| 0x55f4ee6efa82 [48 c7 c0 f8 55 ff ff 64]: run_tasks_from_lists+0x173/0x9c2
| 0x55f4ee6f05d3 [8b 7d a0 29 c7 85 ff 0f]: process_runnable_tasks+0x302/0x6e6
| 0x55f4ee671bb7 [83 3d 86 72 44 00 01 0f]: run_poll_loop+0x6e/0x57b
| 0x55f4ee672367 [48 8b 1d 22 d4 1d 00 48]: main-0x48d
| 0x55f4ee6755e0 [b8 00 00 00 00 e8 08 61]: main+0x2dec/0x335d
This could happen after Handshake packet building failures which follow a successful
Initial packet into the same datagram. In this case, the datagram could be emitted
with a too short length (<1200 bytes).
To fix this, store the datagram only if the first packet is not an Initial packet
or if its length is big enough (>=1200 bytes).
Must be backported as far as 2.6.
By "aws-lc only", one means that this bug was first revealed by aws-lc stack.
This does not mean it will not appeared for new versions of other TLS stacks which
have never revealed this bug.
This bug was reported by Ilya (@chipitsine) in GH #2657 where some QUIC interop
tests (resumption, zerortt) could lead to crash with haproxy compiled against
aws-lc TLS stack. These crashed were triggered by this BUG_ON() which detects
that too short datagrams with at least one ack-eliciting Initial packet inside
could be built.
<0>2024-07-31T15:13:42.562717+02:00 [01|quic|5|quic_tx.c:739] qc_prep_pkts():
next encryption level : qc@0x61d000041080 idle_timer_task@0x60d000006b80 flags=0x6000058
FATAL: bug condition "first_pkt->type == QUIC_PACKET_TYPE_INITIAL && (first_pkt->flags & (1UL << 0)) && length < 1200" matched at src/quic_tx.c:163
call trace(12):
| 0x563ea447bc02 [ba d9 00 00 00 48 8d 35]: main-0x1958ce
| 0x563ea4482703 [e9 73 fe ff ff ba 03 00]: qc_send+0x17e4/0x1b5d
| 0x563ea4488ab4 [85 c0 0f 85 00 f6 ff ff]: quic_conn_io_cb+0xab1/0xf1c
| 0x563ea468e6f9 [48 c7 c0 f8 55 ff ff 64]: run_tasks_from_lists+0x173/0x9c2
| 0x563ea468f24a [8b 7d a0 29 c7 85 ff 0f]: process_runnable_tasks+0x302/0x6e6
| 0x563ea4610893 [83 3d aa 65 44 00 01 0f]: run_poll_loop+0x6e/0x57b
| 0x563ea4611043 [48 8b 1d 46 c7 1d 00 48]: main-0x48d
| 0x7f64d05fb609 [64 48 89 04 25 30 06 00]: libpthread:+0x8609
| 0x7f64d0520353 [48 89 c7 b8 3c 00 00 00]: libc:clone+0x43/0x5e
That said everything was correctly done by qc_prep_ptks() to prevent such a case.
But this relied on the hypothesis that the list of encryption levels it used
was always built in the same order as follows for 0-RTT sessions:
initial, early-data, handshake, application
But this order is determined but the order the TLS stack derives the secrets
for these encryption levels. For aws-lc, this order is not the same but
as follows:
initial, handshake, application, early-data
During 0-RTT sessions, the server may have to build three ack-eliciting packets
(with CRYPTO data inside) to reply to the first client packet: initial, hanshake,
application. qc_prep_pkts() adds a PADDING frame to the last built packet
for the last encryption level in the list. But after application level encryption,
there is early-data encryption level. This prevented qc_prep_pkts() to build
a padded applicaiton level last packet to send a 1200-bytes datagram.
To fix this, always insert early-data encryption level after the initial
encryption level into the encryption levels list when initializing this encryption
level from quic_conn_enc_level_init().
Must be backported as far as 2.9.
When the peer applet is waiting for a synchronisation with the global sync
task, we must notify it won't consume data. Otherwise, if some data are
already waiting in the input buffer, the applet will be woken up in loop and
this wil trigger the watchdog. Once synchronized, the applet is woken up. In
that case, the peer applet must indicate it is going to consume data again.
This patch should fix the issue #2656. It must be backported to 3.0.
When a stream is explicitly woken up by the H2 conneciton, if an error
condition is detected, the corresponding error flag is set on the SE. So
SE_FL_ERROR or SE_FL_ERR_PENDING, depending if the end of stream was
reported or not.
However, there is no attempt to propagate other termination flags. We must
be sure to properly set SE_FL_EOI and SE_FL_EOS when appropriate to be able
to switch a pending error to a fatal error.
Because of this bug, the SE remains with a pending error and no end of
stream, preventing the applicative stream to trully abort it. It means on
some abort scenario, it is possible to block a stream infinitely.
This patch must be backported at least as far as 2.8. No bug was observed on
older versions while the same code is inuse.
For regular H2 messages, the HTX EOM flag is synonymous the end of input. So
SE_FL_EOI flag must also be set on the stream-endpoint descriptor. However,
there is an exception. For tunneled streams, the end of message is reported
on the HTX message just after the headers. But in that case, no end of input
is reported on the SE.
But here, there is a bug. The "early" EOM is also report on the HTX messages
when there is no payload (for instance a content-length set to 0). If there
is no ES flag on the H2 HEADERS frame, it is an unexpected case. Because for
the applicative stream and most probably for the opposite endpoint, the
message is considered as finihsed. It is switched in its DONE state (or the
equivalent on the endpoint). But, if an extra H2 frame with the ES flag is
received, a TRAILERS frame or an emtpy DATA frame, an extra EOT HTX block is
pushed to carry the HTX EOM flag. So an extra HTX block is emitted for a
regular HTX message. It is totally invalid, it must never happen.
Because it is an undefined behavior, it is difficult to predict the result.
But it definitly prevent the applicative stream to properly handle aborts
and errors because data remain blocked in the channel buffer. Indeed, the
end of the message was seen, so no more data are forwarded.
It seems to be an issue for 2.8 and upper. Harder to evaluate for older
versions.
This patch must be backported as far as 2.4.
When we are waiting for the server response, if an error is pending on the
frontend side (a write error on client), it is handled as an abort and all
regular response analyzers are removed, except the one responsible to
release the filters, if any. However, while it is handled as an abort, the
error is not reported, as usual, via http_reply_and_close() function. It is
an issue because in that, the channels buffers are not reset.
Because of this bug, it is possible to block a stream infinitely. The
request side is waiting for the response side and the response side is
blocked because filters must be released and this cannot be done because
data remain blocked in channels buffers.
So, in that case, calling http_reply_and_close() with no message is enough
to unblock the stream.
This patch must be backported as far as 2.8.
This commit prevents forwarding of an HTTP/2 Extended CONNECT when "h2c"
or "h2" token is set as targetted protocol. Contrary to the previous
commit which deals with HTTP/1 mux, this time the request is rejected
and a RESET_STREAM is reported to the client.
This must be backported up to 2.4 after a period of observation.
haproxy supports tunnel establishment through HTTP Upgrade mechanism.
Since the following commit, extended CONNECT is also supported for
HTTP/2 both on frontend and backend side.
commit 9bf957335e
MEDIUM: mux_h2: generate Extended CONNECT from htx upgrade
As specified by HTTP/2 rfc, "h2c" can be used by an HTTP/1.1 client to
request an upgrade to HTTP/2. In haproxy, this is not supported so it
silently ignores this. However, Connection and Upgrade headers are
forwarded as-is on the backend side.
If using HTTP/1 on the backend side and the server supports this upgrade
mechanism, haproxy won't be able to parse the HTTP response. If using
HTTP/2, mux backend tries to incorrectly convert the request to an
Extended CONNECT with h2c protocol, which may also prevent the response
to be transmitted.
To fix this, flag HTTP/1 request with "h2c" or "h2" token in an upgrade
header. On converting the header list to HTX, the upgrade header is
skipped if any of this token is present and the H1_MF_CONN_UPG flag is
removed.
This issue can easily be reproduced using curl --http2 argument to
connect to an HTTP/1 frontend.
This must be backported up to 2.4 after a period of observation.
Control layer callback get_info has recently been implemented for QUIC.
However, fc_lost always returned 0. This is because quic_get_info() does
not use the correct input argument value to identify lost value.
This does not need to be backported.
QUIC has recently implement get_info callback to return RTT/sRTT values.
However, it uses milliseconds, contrary to TCP which uses microseconds.
This cause smp fetch functions to return invalid values. Fix this by
converting QUIC values to microseconds.
This does not need to be backported.
Decode QUIC MUX connection and stream elements via qcc_show_flags() and
qcs_show_flags(). Flags definition have been moved outside of USE_QUIC
to ease compilation of flags binary.
This low level callback may be called by several sample fetches for
frontend connections like "fc_rtt", "fc_rttvar" etc.
Define this callback for QUIC protocol as pointer to quic_get_info().
This latter supports these sample fetches:
"fc_lost", "fc_reordering", "fc_rtt" and "fc_rttvar".
Update the documentation consequently.
Add ->get_info() new control layer callback definition to protocol struct to
retreive statiscal counters information at transport layer (TCPv4/TCPv6) identified by
an integer into a long long int.
Move the TCP specific code from get_tcp_info() to the tcp_get_info() control layer
function (src/proto_tcp.c) and define it as the ->get_info() callback for
TCPv4 and TCPv6.
Note that get_tcp_info() is called for several TCP sample fetches.
This patch is useful to support some of these sample fetches for QUIC and to
keep the code simple and easy to maintain.
Received QUIC packets are stored in quic_conn Rx buffer after header
protection removal in qc_rx_pkt_handle(). These packets are then removed
after quic_conn IO handler via qc_treat_rx_pkts().
If HP cannot be removed, packets are still copied into quic_conn Rx
buffer. This can happen if encryption level TLS keys are not yet
available. The packet remains in the buffer until HP can be removed and
its content processed.
An issue occurs if client emits a 0-RTT packet but haproxy does not have
the shared secret, for example after a haproxy process restart. In this
case, the packet is copied in quic_conn Rx buffer but its HP won't ever
be removed. This prevents the buffer to be purged. After some time, if
the client has emitted enough packets, Rx buffer won't have any space
left and received packets are dropped. This will cause the connection to
freeze.
To fix this, remove any 0-RTT buffered packets on handshake completion.
At this stage, 0-RTT packets are unnecessary anymore. The client is
expected to reemit its content in 1-RTT packet which are properly
deciphered.
This can easily reproduce with HTTP/3 POST requests or retrieving a big
enough object, which will fill the Rx buffer with ACK frames. Here is a
picoquic command to provoke the issue on haproxy startup :
$ picoquicdemo -Q -v 00000001 -a h3 <hostname> 20443 "/?s=1g"
Note that allow-0rtt must be present on the bind line to trigger the
issue. Else haproxy will reject any 0-RTT packets.
This must be backported up to 2.6.
This could be one of the reason for github issue #2549 but it's unsure
for now.
Revert patch fcc8255 "MINOR: ssl_sock: Early data disabled during
SSL_CTX switching (aws-lc)". The patch was done in the wrong callback
which is never built for AWS-LC, and applies options on the SSL_CTX
instead of the SSL, which should never be done elsewhere than in the
configuration parsing.
This was probably triggered by successfully linking haproxy against
AWS-LC without using USE_OPENSSL_AWSLC.
The patch also reintroduced SSL_CTX_set_early_data_enabled() in the
ssl_quic_initial_ctx() and ssl_sock_initial_ctx(). So the initial_ctx
does have the right setting, but it still needs to be applied to the
selected SSL_CTX in the clienthello, because we need it on the selected
SSL_CTX.
Must be backported to 3.0. (ssl_clienthello.c part was in ssl_sock.c)
Then reactivate HAVE_SSL_0RTT and HAVE_SSL_0RTT_QUIC for AWS-LC, which
were wrongly deactivated in f5353f2c ("MINOR: ssl: add HAVE_SSL_0RTT
constant").
Must be backported to 3.0.
The backend depends on the response and the frontend on the request, not
the other way around. In addition, they used to depend on L6 (hence
contents in the channel buffers) while they should only depend on L5
(permanent info known in the mux).
This came in 2.9 with commit 24059615a7 ("MINOR: Add sample fetches to
get the frontend and backend stream ID") so this can be backported there.
(cherry picked from commit 61dd0156c82ea051779e6524cad403871c31fc5a)
Signed-off-by: Willy Tarreau <w@1wt.eu>
A typo was introduced in 760d26a86 ("BUG/MEDIUM: mux-pt/mux-h1: Release the
pipe on connection error on sending path"). The sedesc variable is 'sd', not
'se'.
This patch must be backported with the commit above.
When data are sent using the kernel splicing, if a connection error
occurred, the pipe must be released. Indeed, in that case, no more data can
be sent and there is no reason to not release the pipe. But it is in fact an
issue for the stream because the channel will appear are not empty. This may
prevent the stream to be released. This happens on 2.8 when a filter is also
attached on it. On 2.9 and upper, it seems there is not issue. But it is
hard to be sure and the current patch remains valid is all cases. On 2.6 and
lower, the code is not the same and, AFAIK, there is no issue.
This patch must be backported to 2.8. However, on 2.8, there is no zero-copy
data forwarding. The patch must be adapted. There is no done_ff/resume_ff
callback functions for muxes. The pipe must released in sc_conn_send() when
an error flag is set on the SE, after the call to snd_pipe callback
function.
When a send on a connection is performed, if a SE error (or a pending error)
was already reported earlier, we leave immediately. No send is performed.
However, we must be sure to report the error at the SC level if necessary.
Indeed, the SE error may have been reported during the zero-copy data
forwarding. So during receive on the opposite side. In that case, we may
have missed the opportunity to report it at the SC level.
The patch must be backported as far as 2.8.
The SPOE was refactored. Now backends referenced by a SPOE filter must use
the spop mode to be able to use the spop multiplexer for server connections.
The "spop" mode was added in the list of supported mode for backends.
After checking that a server or backend is full, it remains possible
to call pendconn_add() just after the last pending requests finishes, so
that there's no more connection on the server for very low maxconn (typ 1),
leaving new ones in queue till the timeout.
The approach depends on where the request was queued, though:
- when queued on a server, we can simply detect that we may dequeue
pending requests and wake them up, it will wake our request and
that's fine. This needs to be done in srv_redispatch_connect() when
the server is set.
- when queued on a backend, it means that all servers are done with
their requests. It means that all servers were full before the
check and all were empty after. In practice this will only concern
configs with less servers than threads. It's where the issue was
first spotted, and it's very hard to reproduce with more than one
server. In this case we need to load-balance again in order to find
a spare server (or even to fail). For this, we call the newly added
dedicated function pendconn_must_try_again() that tells whether or
not a blocked pending request was dequeued and needs to be retried.
This should be backported along with pendconn_must_try_again() to all
stable versions, but with extreme care because over time the queue's
locking evolved.
There's a rare TOCTOU case that happens from time to time with maxconn 1
and multiple threads. Between the moment we see the queue full and the
moment we queue a request, it's possible that the last request on the
server or proxy ended and that no other one is left to offer it its place.
Given that all this code path is performance-critical and we cannot afford
to increase the lock duration, better recheck for the condition after
queueing. For this we need to be able to check for the condition and
cleanly dequeue a request. That's what this patch provides via the new
function pendconn_must_try_again(). It will catch more requests than
absolutely needed though it will catch them all. It may find that around
1/1000 of requests are at risk, though testing shows that in practice,
it's around 1 per million that really gets stuck (other ones benefit
from timing and finishing late requests). Maybe in the future some
conditions might be refined but it's harmless.
What happens to such requests is that they're dequeued and their pendconn
freed, so that the caller can decide to try to LB or queue them again. For
now the function is not used, it's just added separately for easier tracking.
Since commit 35470d518 ("MINOR: quic: activate UDP GSO for QUIC if
supported"), Solaris build fails due to netinet/udp.h being included
without netinet/in.h. Adding it is sufficient to fix the problem. No
backport is needed.
When the signature included in a JWT is verified, if an error occurred, one
or more SSL errors are queued and never cleared. These errors may be then
caught by the SSL stack and a fatal SSL error may be erroneously reported
during a SSL received or send.
So we must take care to clear the SSL error queue when the signature
verification failed.
This patch should fix issue #2643. It must be backported as far as 2.6.
