In Patrick's trace it was visible that after a stream had been missed,
the next stream would receive a WINDOW_UPDATE with the first one's
credit added to its own. This makes sense because in case of error
h2c->rcvd_s is not reset. Given that this counter is per frame, better
reset it when starting to parse a new frame, it's easier and safer.
This must be backported as far as 1.8.
In h2_process_mux() if we have some room and an attempt to send a window
update for the connection was pending, it's done first. But it's not done
for the stream, which will have for effect of postponing this attempt
till next pass into h2_process_demux(), at the risk of seeing the send
buffer full again. Let's always try to send both pending frames as soon
as possible.
This should be backported as far as 1.8.
Patrick Hemmer reported a rare case where the H2 mux emits spurious
RST_STREAM(STREAM_CLOSED) that are triggered by the send path and do
not even appear to be associated with a previous incoming frame, while
the send path never emits such a thing.
The problem is particularly complex (hence its rarity). What happens is
that when data are uploaded (POST) we must refill the sending stream's
window by sending a WINDOW_UPDATE message (and we must refill the
connection's too). But in a highly bidirectional traffic, it is possible
that the mux's buffer will be full and that there is no more room to build
this WINDOW_UPDATE frame. In this case the demux parser switches to the
H2_CS_FRAME_A state, noting that an "acknowledgement" is needed for the
current frame, and it doesn't change the current stream nor frame type.
But the stream's state was possibly updated (typically OPEN->HREM when
a DATA frame carried the ES flag).
Later the data can leave the buffer, wake up h2_io_cb(), which calls
h2_send() to send pending data, itself calling h2_process_mux() which
detects that there are unacked data in the connection's window so it
emits a WINDOW_UPDATE for the connection and resets the counter. so it
emits a WINDOW_UPDATE for the connection and resets the counter. Then
h2_process() calls h2_process_demux() which continues the processing
based on the current frame type and the current state H2_CS_FRAME_A.
Unfortunately the protocol compliance checks matching the frame type
against the current state are still present. These tests are designed
for new frames only, not for those in progress, but they are not
limited by frame types. Thus the current DATA frame is checked again
against the current stream state that is now HREM, and fails the test
with a STREAM_CLOSED error.
The quick and backportable solution consists in adding the test for
this ACK and bypass all these checks that were already validated prior
to the state transition. A better long-term solution would consist in
having a new state between H and P indicating the frame is new and
needs to be checked ("N" for new?) and apply the protocol tests only
in this state. In addition everywhere we decide to send a window
update, we should send a stream WU first if there are unacked data
for the current stream. Last, rcvd_s should always be reset when
transitioning to FRAME_H (and a BUGON for this in dev would help).
The bug will be way harder to trigger on 2.0 than on 1.8/1.9 because
we have a ring buffer for the connection so the buffer full situations
are extremely rare.
This fix must be backpored to all versions having H2 (as far as 1.8).
Special thanks to Patrick for providing exploitable traces.
If the server decides to close early, we don't want to send a
REFUSED_STREAM error but a CANCEL, so that the client doesn't want
to retry. The test in h2_do_shutw() was wrong for this as it would
handle the HLOC case like the case where nothing had been sent for
this stream, which is wrong. Now h2_do_shutw() does nothing in this
case and lets h2_do_shutr() decide.
Note that this partially undoes f983d00a1 ("BUG/MINOR: mux-h2: make
the do_shut{r,w} functions more robust against retries").
This must be backported to 2.0. The patch above was not backported to
1.9 for being too risky there, but if it eventually gets to it, this
one will be needed as well.
There is a test on the existence of the conn_stream when receiving data,
to be sure to have somewhere to deliver it. Right now it responds with
STREAM_CLOSED, which is not correct since from an H2 point of view the
stream is not closed and a peer could be upset to see this. After some
analysis, it is important to keep this test to be sure not to fill the
rxbuf then stall the connection. Another option could be to modiffy
h2_frt_transfer_data() to silently discard any contents but the CANCEL
error code is designed exactly for this and to save the peer from
continuing to stream data that will be discarded, so better switch to
using this.
This must be backported as far as 1.8.
The test in h2s_send_rst_stream() is excessive, it refrains from sending an
RST_STREAM if *the last frame* was an RST_STREAM, regardless of the stream
ID. In a context where both clients and servers abort a lot, it could happen
that one RST_STREAM is dropped from responses from time to time, causing
delays to the client.
This must be backported to 2.0, 1.9 and 1.8.
The SETTINGS frame parser updates all streams' window for each
INITIAL_WINDOW_SIZE setting received on the connection (like h2spec
does in test 6.5.3), which can start to be expensive if repeated when
there are many streams (up to 100 by default). A quick test shows that
it's possible to parse only 35000 settings per second on a 3 GHz core
for 100 streams, which is rather small.
Given that window sizes are relative and may be negative, there's no
point in pre-initializing them for each stream and update them from
the settings. Instead, let's make them relative to the connection's
initial window size so that any change immediately affects all streams.
The only thing that remains needed is to wake up the streams that were
unblocked by the update, which is now done once at the end of
h2_process_demux() instead of once per setting. This now results in
5.7 million settings being processed per second, which is way better.
In order to keep the change small, the h2s' mws field was renamed to
"sws" for "stream window size", and an h2s_mws() function was added
to add it to the connection's initial window setting and determine the
window size to use when muxing. The h2c_update_all_ws() function was
renamed to h2c_unblock_sfctl() since it's now only used to unblock
previously blocked streams.
This needs to be backported to all versions till 1.8.
Recent optimization in commit 4d7a88482 ("MEDIUM: mux-h2: don't try to
read more than needed") broke the receipt of large DATA frames because
it would unconditionally subscribe if there was some room left, thus
preventing any new rx from being done since subscription may only be
done once the end was reached, as indicated by ret == 0.
However, fixing this uncovered that in HTX mode previous versions might
occasionally be affected as well, when an available frame is the same
size as the maximum data that may fit into an HTX buffer, we may end
up reading that whole frame and still subscribe since it's still allowed
to receive, thus causing issues to read the next frame.
This patch will only work for 2.1-dev but a minor adaptation will be
needed for earlier versions (down to 1.9, where subscribe() was added).
The h2_recv() loop was historically built around a loop to deal with
the callback model but this is not needed anymore, as it the upper
layer wants more data, it will simply try to read again. Right now
50% of the recvfrom() calls made over H2 return EAGAIN. With this
change it doesn't happen anymore. Note that the code simply consists
in breaking the loop, and reporting real data receipt instead of
always returning 1.
A test was made not to subscribe if we actually read data but it
doesn't change anything since we might be subscribed very early
already.
Since the legacy HTTP mode is disabled and no multiplexer relies on it anymore,
there is no reason to have 2 multiplexer protocols for the HTTP. So the protocol
PROTO_MODE_HTX was removed and all HTTP multiplexers use now PROTO_MODE_HTTP.
<head> and <tail> fields are now signed 32-bits integers. For an empty HTX
message, these fields are set to -1. So the field <used> is now useless and can
safely be removed. To know if an HTX message is empty or not, we just compare
<head> against -1 (it also works with <tail>). The function htx_nbblks() has
been added to get the number of used blocks.
Because the infinite forward is HTX aware, it is useless to tinker with the
number of bytes really sent. This was fixed long ago for the H1 and forgotten to
do so for the H2.
When a DATA frame is processed for a message with a content-length, we first
take care to not have a frame size that exceeds the remaining to
read. Otherwise, an error is triggered. But we must remove the padding length
from the frame size because the padding is not included in the announced
content-length.
This patch must be backported to 2.0 and 1.9.
In the function h2_process_demux(), if several frames are parsed, the padding
length must be reset between each frame. Otherwise we may wrongly think a frame
has a padding block because the previous one was padded.
This patch must be backported to 2.0 and 1.9.
When commit 0350b90e3 ("MEDIUM: htx: make htx_add_data() never defragment
the buffer") was introduced, it made htx_add_data() actually be able to
add less data than it was asked for, and the callers must use the returned
value to know how much was added. The H2 code used to rely on the frame
length instead of the return value. A version of the code doing this was
written but is obviously not the one that got merged, resulting in breaking
large uploads or downloads when HTX would have instead defragmented the
buffer because the HTX side sees less contents than what the H2 side sees.
This patch fixes this again. No backport is needed.
Olivier found that commit 99ad1b3e8 ("MINOR: mux-h2: stop relying on
CS_FL_REOS") managed to break abortonclose again with H2. What happens
is that while the CS_FL_REOS flag was set on some transitions to the
HREM state, it's not set on all and is in fact only set when the low
level connection is closed. So making the replacement condition match
the HREM and ERROR states is not correct and causes completely correct
requests to send advertise an early close of the connection layer while
only the stream's input is closed.
In order to avoid this, we now properly split the checks for the CLOSED
state and for the closed connection. This way there is no risk to set
the EOS flag too early on the connection.
No backport is needed.
The function really only operates on tasklets, its arguments are always
tasklets cast as tasks to match the function's type, to be cast back to
a struct tasklet. Let's rename it to tasklet_remove_from_tasklet_list(),
take a struct tasklet, and get rid of the undesired task casts.
It's really confusing to call it a task because it's a tasklet and used
in places where tasks and tasklets are used together. Let's rename it
to tasklet to remove this confusion.
By default, the scheme "https" is always used. But when an explicit scheme was
defined and when this scheme is "http", we use it in the request sent to the
server. This is done by checking flags of the start-line. If the flag
HTX_SL_F_HAS_SCHM is set, it means an explicit scheme was defined on the client
side. And if the flag HTX_SL_F_SCHM_HTTP is set, it means the scheme "http" was
used.
Instead of using the macro MAX_HTTP_HDR to limit the number of headers parsed
before throwing an error, we now use the custom global variable
global.tune.max_http_hdr.
This patch must be backported to 1.9.
There seems to be a tricky case in the H2 mux related to stream flow
control versus buffer a full situation : is a large response cannot
be entirely sent to the client due to the stream window being too
small, the stream is paused with the SFCTL flag. Then the upper
layer stream might get bored and expire this stream. It will then
shut it down first. But the shutdown operation might fail if the
mux buffer is full, resulting in the h2s being subscribed to the
deferred_shut event with the stream *not* added to the send_list
since it's blocked in SFCTL. In the mean time the upper layer completely
closes, calling h2_detach(). There we have a send_wait (the pending
shutw), the stream is marked with SFCTL so we orphan it.
Then if the client finally reads all the data that were clogging the
buffer, the send_list is run again, but our stream is not there. From
this point, the connection's stream list is not empty, the mux buffer
is empty, so the connection's timeout is not set. If the client
disappears without updating the stream's window, nothing will expire
the connection.
This patch makes sure we always keep the connection timeout updated.
There might be finer solutions, such as checking that there are still
living streams in the connection (i.e. streams not blocked in SFCTL
state), though this is not necessarily trivial nor useful, since the
client timeout is the same for the upper level stream and the connection
anyway.
This patch needs to be backported to 1.9 and 1.8 after some observation.
This type of blocks is useless because transition between data and trailers is
obvious. And when there is no trailers, the end-of-message is still there to
know when data end for chunked messages.
HTTP trailers are now parsed in the same way headers are. It means trailers are
converted to K/V blocks followed by an end-of-trailer marker. For now, to make
things simple, the type for trailer blocks are not the same than for header
blocks. But the aim is to make no difference between headers and trailers by
using the same type. Probably for the end-of marker too.
Usually when calling offer_buffer(), we don't expect to offer it to
ourselves. But with h2 we have the same buffer_wait for the two directions
so we can unblock the recv path when completing a send(), or we can unblock
part of the mux buffer after sending the first few buffers that we managed
to collect. Thus it is important to always accept to wake up any requester.
A few parts of this patch could possibly be backported but earlier versions
already have other issues related to low-buffer condition so it's not sure
it's worth taking the risk to make things worse.
The test for the mux alloc failure in h2_send() right after an attempt
at h2_process_mux() used to make sense as it tried to detect that this
latter failed to produce data. But now that we have a list of buffers,
it is a perfectly valid situation where there can still be data in the
buffer(s).
So now when we see this flag we only declare it's the last run on the
loop. In addition we need to make sure we break out of the loop on
snd_buf failure, or we'll loop indefinitely, for example when the buf
is full and we can't send.
No backport is needed.
In h2c_frt_stream_new, if we failed to create the stream for some reason,
don't forget to set h2s->cs to NULL before calling h2s_destroy(), otherwise
h2s_destroy() will call h2s_close(), which will attempt to access
h2s->cs->flags if it's non-NULL.
This should be backported to 1.9.
In lock profiles it's visible that there is a huge contention on the
buffer lock. The reason is that when offer_buffers() is called, it
systematically takes the lock before verifying if there is any
waiter. However doing so doesn't protect against races since a
waiter can happen just after we release the lock as well. Similarly
in h2 we take the lock every time an h2c is going to be released,
even without checking that the h2c belongs to a wait list. These
two have now been addressed by verifying non-emptiness of the list
prior to taking the lock.
In order to later allow htx_add_data() to transmit partial blocks and
avoid defragmenting the buffer, we'll need to return the number of bytes
consumed. This first modification makes the function do this and its
callers take this into account. At the moment the function still works
atomically so it returns either the block size or zero. However all
call places have been adapted to consider any value between zero and
the block size.
1xx informational messages (all except 101) are now part of the HTTP reponse,
semantically speaking. These messages are not followed by an EOM anymore,
because a final reponse is always expected. All these parts can also be
transferred to the channel in same time, if possible. The HTX response analyzer
has been update to forward them in loop, as the legacy one.
In the function htx_xfer_blks(), we take care to transfer all headers in one
time. When the current block is a start-line, we check if there is enough space
to transfer all headers too. If not, and if the destination is empty, a parsing
error is reported on the source.
The H2 multiplexer is the only one to use this function. When a parsing error is
reported during the transfer, the flag CS_FL_EOI is also set on the conn_stream.
Now, the SI calls h2_rcv_buf() with the right count value. So we can rely on
it. Unlike the H1 multiplexer, it is fairly easier for the H2 multiplexer
because the HTX message already exists, we only transfer blocks from the H2S to
the channel. And this part is handled by htx_xfer_blks().
This patch makes the function more accurate. Thanks to the function
htx_get_max_blksz(), the transfer of data has been simplified. Note that now the
total number of bytes copied (metadata + payload) is returned. This slighly
change how the function is used in the H2 multiplexer.
in the H2 multiplexer, when a HEADERS frame is built before sending it, we have
the warranty the start-line is the head of the HTX message. It is safer to rely
on this fact than on the sl_pos value. For now, it's safe to use sl_pos in muxes
because HTTP 1xx messages are considered as full messages in HTX and only one
HTTP message can be stored at a time in HTX. But we are trying to handle 1xx
messages as a part of the reponse message. In this way, an HTTP reponse will be
the sum of all 1xx informational messages followed by the final response. So it
will be possible to have several start-line in the same HTX message. And the
sl_pos will point to the first unprocessed start-line from the analyzers point
of view.
Now when we fail to send because the mux buffer is full, before giving
up and marking MFULL, we try to allocate another buffer in the mux's
ring to try again. Thanks to this (and provided there are enough buffers
allocated to the mux's ring), a single stream picked in the send_list
cannot steal all the mux's room at once. For this, we expand the ring
size to 31 buffers as it seems to be optimal on benchmarks since it
divides the number of context switches by 3. It will inflate each H2
conn's memory by 1 kB.
The bandwidth is now much more stable. Prior to this, it a test on
h2->h1 with very large objects (1 GB), a few tens of connections and
a few tens of streams per connection would show a varying performance
between 34 and 95 Gbps on 2 cores/4 threads, with h2_snd_buf() stopped
on a buffer full condition between 300000 and 600000 times per second.
Now the performance is constantly between 88 and 96 Gbps. Measures show
that buffer full conditions are met around only 159 times per second
in this case, or rougly 2000 to 4000 times less often.
This makes the code more readable and reduces the calls to br_tail().
In addition, all calls to h2_get_buf() are now made via this local
variable, which should significantly help for retries.
Now send() uses a loop to iterate over all buffers to be sent. These
buffers are released and deleted from the vector once completely sent.
If any buffer gets released, offer_buffers() is called to wake up some
waiters.
For now it's only one buffer long so the head and tails are always the
same, thus it doesn't change what used to work. In short, br_tail(h2c->mbuf)
was inserted everywhere we used to have h2c->mbuf.
Transferring large objects over H2 sometimes shows unexplained performance
variations. A long analysis resulted in the following discovery. Often the
mux buffer looks like this :
[ empty_head | data | empty_tail ]
Typical numbers are (very common) :
- empty_head = 31
- empty_tail = 16 (total free=47)
- data = 16337
- size = 16384
- data to copy: 43
The reason for these holes are the blocking factors that are not always
the same in and out (due to keeping 9 bytes for the frame size, or the
56 bytes corresponding to the HTX header). This can easily happen 10000
times a second if the network bandwidth permits it!
In this case, while copying a DATA frame we find that the buffer has its
free space wrapped so we decide to realign it to optimize the copy. It's
possible that this practice stems from the code used to emit headers,
which do not support fragmentation and which had no other option left.
But it comes with two problems :
- we don't check if the data fits, which results in a memcpy for nothing
- we can move huge amounts of data to just copy a small block.
This patch addresses this two ways :
- first, by not forcing a data realignment if what we have to copy does
not fit, as this is totally pointless ;
- second, by refusing to move too large data blocks. The threshold was
set to 1 kB, because it may make sense to move 1 kB of data to copy
a 15 kB one at once, which will leave as a single 16 kB block, but
it doesn't make sense to mvoe 15 kB to copy just 1 kB. In all cases
the data would fit and would just be split into two blocks, which is
not very expensive, hence the low limit to 1 kB
With such changes, realignments are very rare, they show up around once
every 15 seconds at 60 Gbps, and look like this, resulting in a much more
stable bit rate :
buf=0x7fe6ec0c3510,h=16333,d=35,s=16384 room=16349 in=16337
This patch should be safe for backporting to 1.9 if some performance
issues are reported there.
In HTX, when a HEADERS frame is formatted before sending it to the client or the
server, If an EOM is found because there is no body, we must count it in the
number bytes sent.
This patch must be backported to 1.9.
It is not legal to subscribe if we're already subscribed, or to unsubscribe
if we did not subscribe, so instead of trying to handle those cases, just
assert that it's ok using the new BUG_ON() macro.
Just like CS_FL_REOS previously, the CS_FL_EOI flag is abused as a proxy
for H2_SF_ES_RCVD. The problem is that this flag is consumed by the
application layer and is set immediately when an end of stream was met,
which is too early since the application must retrieve the rxbuf's
contents first. The effect is that some transfers are truncated (mostly
the first one of a connection in most tests).
The problem of mixing CS flags and H2S flags in the H2 mux is not new
(and is currently being addressed) but this specific one was emphasized
in commit 63768a63d ("MEDIUM: mux-h2: Don't mix the end of the message
with the end of stream") which was backported to 1.9. Note that other
flags, particularly CS_FL_REOS still need to be asynchronously reported,
though their impact seems more limited for now.
This patch makes sure that all internal uses of CS_FL_EOI are replaced
with a test on H2_SF_ES_RCVD (as there is a 1-to-1 equivalence) and that
CS_FL_EOI is only reported once the rxbuf is empty.
This should ideally be backported to 1.9 unless it causes too much
trouble due to the recent changes in this area, as 1.9 *seems* not
to be directly affected by this bug.
This flag was introduced early in 1.9 development (a3f7efe00) to report
the fact that the rxbuf that was present on the conn_stream was followed
by a shutr. Since then the rxbuf moved from the conn_stream to the h2s
(638b799b0) but the flag remained on the conn_stream. It is problematic
because some state transitions inside the mux depend on it, thus depend
on the CS, and as such have to test for its existence before proceeding.
This patch replaces the test on CS_FL_REOS with a test on the only
states that set this flag (H2_SS_CLOSED, H2_SS_HREM, H2_SS_ERROR).
The few places where the flag was set were removed (the flag is not
used by the data layer).
This flag is currently set when an incoming close was received, which
results in the stream being in either H2_SS_HREM, H2_SS_CLOSED, or
H2_SS_ERROR states, so let's remove the test for the OPEN and HLOC
cases.
