The splice() syscall has been supported in glibc since version 2.5 issued
in 2006 and is present on supported systems so there's no need for having
our own arch-specific syscall definitions anymore.
This partially reverts commit 1113116b4a ("MEDIUM: raw-sock: remove
obsolete calls to fd_{cant,cond,done}_{send,recv}") so that we can mark
the FD not ready as required since commit 19bc201c9f ("MEDIUM: connection:
remove the intermediary polling state from the connection"). Indeed, with
the removal of the latter we don't have any other reliable indication that
the FD is blocked, which explains why there are so many EAGAIN in traces.
It's worth noting that a short read or a short write are also reliable
indicators of exhausted buffers and are even documented as such in the
epoll man page in case of edge-triggered mode. That's why we also report
the FD as blocked in such a case.
With this change we completely got rid of EAGAIN in keep-alive tests, but
they were expectedly transferred to epoll_ctl:
$ ./h1load -n 100000 -t 4 -c 1000 -T 20 -F 127.0.0.1:8001/?s=1k/t=20
before:
266331 epoll_ctl 1
200000 sendto 1
200000 recvfrom 1
135757 recvfrom -1
8626 epoll_wait 1
after:
394865 epoll_ctl 1
200000 sendto 1
200000 recvfrom 1
10748 epoll_wait 1
1999 recvfrom -1
This flag is currently supported by raw_sock to perform a single recv()
attempt and avoid subscribing. Typically on the request and response
paths with keep-alive, with short messages we know that it's very likely
that the first message is enough.
Now that we know that the connection layer is transparent for polling
changes, we have no reason for hiding behind conn_xprt_stop_send() and
can safely call fd_stop_send() on the FD once the buffer is empty.
Given that raw_sock's functions solely act on connections and that all its
callers properly use subscribe() when they want to receive/send more, there
is no more reason for calling fd_{cant,cond,done}_{send,recv} anymore as
this call is immediately overridden by the subscribe call. It's also worth
noting that the purpose of fd_cond_recv() whose purpose was to speculatively
enable reading in the FD cache if the FD was active but not yet polled was
made to save on expensive epoll_ctl() calls and was implicitly covered more
cleanly by recent commit 5d7dcc2a8e ("OPTIM: epoll: always poll for recv if
neither active nor ready").
No change on the number of calls to epoll_ctl() was noticed consecutive to
this change.
The test was added before splice() and send() to make sure we never
accidently send after a shutdown, because upper layers do not all
check and it's not their job to do it. In such a case we also set
errno to EPIPE so that the error can be accurately reported, e.g.,
in health checks.
Commit 477902bd2e ("MEDIUM: connections: Get ride of the xprt_done
callback.") broke the master CLI for a very obscure reason. It happens
that short requests immediately terminated by a shutdown are properly
received, CS_FL_EOS is correctly set, but in si_cs_recv(), we refrain
from setting CF_SHUTR on the channel because CO_FL_CONNECTED was not
yet set on the connection since we've not passed again through
conn_fd_handler() and it was not done in conn_complete_session(). While
commit a8a415d31a ("BUG/MEDIUM: connections: Set CO_FL_CONNECTED in
conn_complete_session()") fixed the issue, such accident may happen
again as the root cause is deeper and actually comes down to the fact
that CO_FL_CONNECTED is lazily set at various check points in the code
but not every time we drop one wait bit. It is not the first time we
face this situation.
Originally this flag was used to detect the transition between WAIT_*
and CONNECTED in order to call ->wake() from the FD handler. But since
at least 1.8-dev1 with commit 7bf3fa3c23 ("BUG/MAJOR: connection: update
CO_FL_CONNECTED before calling the data layer"), CO_FL_CONNECTED is
always synchronized against the two others before being checked. Moreover,
with the I/Os moved to tasklets, the decision to call the ->wake() function
is performed after the I/Os in si_cs_process() and equivalent, which don't
care about this transition either.
So in essence, checking for CO_FL_CONNECTED has become a lazy wait to
check for (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN), but that always
relies on someone else having synchronized it.
This patch addresses it once for all by killing this flag and only checking
the two others (for which a composite mask CO_FL_WAIT_L4L6 was added). This
revealed a number of inconsistencies that were purposely not addressed here
for the sake of bisectability:
- while most places do check both L4+L6 and HANDSHAKE at the same time,
some places like assign_server() or back_handle_st_con() and a few
sample fetches looking for proxy protocol do check for L4+L6 but
don't care about HANDSHAKE ; these ones will probably fail on TCP
request session rules if the handshake is not complete.
- some handshake handlers do validate that a connection is established
at L4 but didn't clear CO_FL_WAIT_L4_CONN
- the ->ctl method of mux_fcgi, mux_pt and mux_h1 only checks for L4+L6
before declaring the mux ready while the snd_buf function also checks
for the handshake's completion. Likely the former should validate the
handshake as well and we should get rid of these extra tests in snd_buf.
- raw_sock_from_buf() would directly set CO_FL_CONNECTED and would only
later clear CO_FL_WAIT_L4_CONN.
- xprt_handshake would set CO_FL_CONNECTED itself without actually
clearing CO_FL_WAIT_L4_CONN, which could apparently happen only if
waiting for a pure Rx handshake.
- most places in ssl_sock that were checking CO_FL_CONNECTED don't need
to include the L4 check as an L6 check is enough to decide whether to
wait for more info or not.
It also becomes obvious when reading the test in si_cs_recv() that caused
the failure mentioned above that once converted it doesn't make any sense
anymore: having CS_FL_EOS set while still waiting for L4 and L6 to complete
cannot happen since for CS_FL_EOS to be set, the other ones must have been
validated.
Some of these parts will still deserve further cleanup, and some of the
observations above may induce some backports of potential bug fixes once
totally analyzed in their context. The risk of breaking existing stuff
is too high to blindly backport everything.
The subscriber used to be passed as a "void *param" that was systematically
cast to a struct wait_event*. By now it appears clear that the subscribe()
call at every layer is well defined and always takes a pointer to an event
subscriber of type wait_event, so let's enforce this in the functions'
prototypes, remove the intermediary variables used to cast it and clean up
the comments to clarify what all these functions do in their context.
These ones used to serve as a set of switches between CO_FL_SOCK_* and
CO_FL_XPRT_*, and now that the SOCK layer is gone, they're always a
copy of the last know CO_FL_XPRT_* ones that is resynchronized before
I/O events by calling conn_refresh_polling_flags(), and that are pushed
back to FDs when detecting changes with conn_xprt_polling_changes().
While these functions are not particularly heavy, what they do is
totally redundant by now because the fd_want_*/fd_stop_*() actions
already perform test-and-set operations to decide to create an entry
or not, so they do the exact same thing that is done by
conn_xprt_polling_changes(). As such it is pointless to call that
one, and given that the only reason to keep CO_FL_CURR_* is to detect
changes there, we can now remove them.
Even if this does only save very few cycles, this removes a significant
complexity that has been responsible for many bugs in the past, including
the last one affecting FreeBSD.
All tests look good, and no performance regressions were observed.
CO_FL_WAIT_ROOM is set by the splicing function in raw_sock, and cleared
by the stream-int when splicing is disabled, as well as in
conn_refresh_polling_flags() so that a new call to ->rcv_pipe() could
be attempted by the I/O callbacks called from conn_fd_handler(). This
clearing in conn_refresh_polling_flags() makes no sense anymore and is
in no way related to the polling at all.
Since we don't call them from there anymore it's better to clear it
before attempting to receive, and to set it again later. So let's move
this operation where it should be, in raw_sock_to_pipe() so that it's
now symmetric. It was also placed in raw_sock_to_buf() so that we're
certain that it gets cleared if an attempt to splice is replaced with
a subsequent attempt to recv(). And these were currently already achieved
by the call to conn_refresh_polling_flags(). Now it could theorically be
removed from the stream-int.
Commit 08fa16e397 made sure
we let the fd layer we didn't want to poll anymore if
we failed to send and sendto() returne EAGAIN.
However, just disabling the polling with fd_stop_send()
while not notifying the connection layer means the
connection layer may believe the polling is activated
and nothing needs to be done when it is wrong.
A better fix is to revamp that whole code, for the
time being, just make sure the fd and connection
layer are properly synchronised.
This should fix the problem recently reported on FreeBSD.
Analysing traces revealed a rare but surprizing pattern :
connect() = -1 EAGAIN
send() = success
epoll_ctl(ADD, EPOLLOUT)
epoll_wait()
recvfrom() = success
close()
What happens is that the failed connect() creates an FD update for pollout,
but the successful synchronous send() doesn't disable it because polling was
only disabled in the FD handler. But a successful synchronous connect()
cancellation is a good opportunity to disable polling before it's effectively
enabled in the next loop, so better disable it when reaching the end. The
cost is very low if it was already disabled anyway (one atomic op).
This only affects local connections but with this the typical number of
epoll_ctl() calls per connection dropped from ~4.2 to ~3.8 for plain TCP
and 10k transfers.
In order to address the absurd polling sequence described in issue #253,
let's make sure we disable receiving on a connection until it's established.
Previously with bottom-top I/Os, we were almost certain that a connection
was ready when the first I/O was confirmed. Now we can enter various
functions, including process_stream(), which will attempt to read
something, will fail, and will then subscribe. But we don't want them
to try to receive if we know the connection didn't complete. The first
prerequisite for this is to mark the connection as not ready for receiving
until it's validated. But we don't want to mark it as not ready for sending
because we know that attempting I/Os later is extremely likely to work
without polling.
Once the connection is confirmed we re-enable recv readiness. In order
for this event to be taken into account, the call to tcp_connect_probe()
was moved earlier, between the attempt to send() and the attempt to recv().
This way if tcp_connect_probe() enables reading, we have a chance to
immediately fall back to this and read the possibly pending data.
Now the trace looks like the following. It's far from being perfect
but we've already saved one recvfrom() and one epollctl():
epoll_wait(3, [], 200, 0) = 0
socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 7
fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
setsockopt(7, SOL_TCP, TCP_NODELAY, [1], 4) = 0
connect(7, {sa_family=AF_INET, sin_port=htons(8000), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress)
epoll_ctl(3, EPOLL_CTL_ADD, 7, {EPOLLIN|EPOLLOUT|EPOLLRDHUP, {u32=7, u64=7}}) = 0
epoll_wait(3, [{EPOLLOUT, {u32=7, u64=7}}], 200, 1000) = 1
connect(7, {sa_family=AF_INET, sin_port=htons(8000), sin_addr=inet_addr("127.0.0.1")}, 16) = 0
getsockopt(7, SOL_SOCKET, SO_ERROR, [0], [4]) = 0
sendto(7, "OPTIONS / HTTP/1.0\r\n\r\n", 22, MSG_DONTWAIT|MSG_NOSIGNAL, NULL, 0) = 22
epoll_ctl(3, EPOLL_CTL_MOD, 7, {EPOLLIN|EPOLLRDHUP, {u32=7, u64=7}}) = 0
epoll_wait(3, [{EPOLLIN|EPOLLRDHUP, {u32=7, u64=7}}], 200, 1000) = 1
getsockopt(7, SOL_SOCKET, SO_ERROR, [0], [4]) = 0
getsockopt(7, SOL_SOCKET, SO_ERROR, [0], [4]) = 0
recvfrom(7, "HTTP/1.0 200\r\nContent-length: 0\r\nX-req: size=22, time=0 ms\r\nX-rsp: id=dummy, code=200, cache=1, size=0, time=0 ms (0 real)\r\n\r\n", 16384, 0, NULL, NULL) = 126
close(7) = 0
Now that the various handshakes come with their own XPRT, there's no
need for the CONN_FL_SOCK* flags, and the conn_sock_want|stop functions,
so garbage-collect them.
Add a new method to xprt_ops, remove_xprt. When called, if the provided
xprt_ctx is the same as the xprt's underlying xprt_ctx, it then uses the
new xprt provided, otherwise it calls the remove_xprt method of the next
xprt.
The goal is to be able to add a temporary xprt, that removes itself from
the chain when it did what it had to do. This will be used to implement
a pseudo-xprt for anything that just requires a handshake (such as the
proxy protocol).
Many times we've been missing per-process traffic statistics. While it
didn't make sense in multi-process mode, with threads it does. Thus we
now have a counter of bytes emitted by raw_sock, and a freq counter for
these as well. However, freq_ctr are limited to 32 bits, and given that
loads of 300 Gbps have already been reached over a loopback using
splicing, we need to downscale this a bit. Here we're storing 1/32 of
the byte rate, which gives a theorical limit of 128 GB/s or ~1 Tbps,
which is more than enough. Let's have fun re-reading this sentence in
2029 :-) The values can be read in "show info" output on the CLI.
We've been dealing with a workaround for a bug in splice that used to
affect version 2.6.25 to 2.6.27.12 and which was fixed 10 years ago
in kernel versions which are not supported anymore. Given that people
who would use a kernel in such a range would face much more serious
stability and security issues, it's about time to get rid of this
workaround and of the ASSUME_SPLICE_WORKS build option used to disable
it.
We still have quite a number of build macros which are mapped 1:1 to a
USE_something setting in the makefile but which have a different name.
This patch cleans this up by renaming them to use the USE_something
one, allowing to clean up the makefile and make it more obvious when
reading the code what build option needs to be added.
The following renames were done :
ENABLE_POLL -> USE_POLL
ENABLE_EPOLL -> USE_EPOLL
ENABLE_KQUEUE -> USE_KQUEUE
ENABLE_EVPORTS -> USE_EVPORTS
TPROXY -> USE_TPROXY
NETFILTER -> USE_NETFILTER
NEED_CRYPT_H -> USE_CRYPT_H
CONFIG_HAP_CRYPT -> USE_LIBCRYPT
CONFIG_HAP_NS -> DUSE_NS
CONFIG_HAP_LINUX_SPLICE -> USE_LINUX_SPLICE
CONFIG_HAP_LINUX_TPROXY -> USE_LINUX_TPROXY
CONFIG_HAP_LINUX_VSYSCALL -> USE_LINUX_VSYSCALL
This implements support for the new API which relies on a call to
setsockopt().
On systems that support it (currently, only Linux >= 4.11), this enables
using TCP fast open when connecting to server.
Please note that you should use the retry-on "conn-failure", "empty-response"
and "response-timeout" keywords, or the request won't be able to be retried
on failure.
Co-authored-by: Olivier Houchard <ohouchard@haproxy.com>
For most of the xprt methods, provide a xprt_ctx. This will be useful later
when we'll want to be able to stack xprts.
The init() method now has to create and provide the said xprt_ctx if needed.
If a send succeeded, add the CO_FL_CONNECTED flag, the send may have been
called by the upper layers before we even realized we were connected, and we
may even read the response before we get the information, and si_cs_recv()
has to know if we were connected or not.
This should be backported to 1.9.
For a long time we've been realigning empty buffers in the transport
layers, where the I/Os were performed based on callbacks. Doing so is
optimal for higher data throughput but makes it trickier to optimize
unaligned data, where mux_h1/h2 have to claim some data are present
in the buffer to force unaligned accesses to skip the frame's header
or the chunk header.
We don't need to do this anymore since the I/O calls are now always
performed from top to bottom, so it's only the mux's responsibility
to realign an empty buffer if it wants to.
In practice it doesn't change anything, it's just a convention, and
it will allow the code to be simplified in a next patch.
As we don't know how subscriptions are handled, we can't just assume we can
use LIST_DEL() to unsubscribe, so introduce a new method to mux and connections
to do so.
Add a new "subscribe" method for connection, conn_stream and mux, so that
upper layer can subscribe to them, to be called when the event happens.
Right now, the only event implemented is "SUB_CAN_SEND", where the upper
layer can register to be called back when it is possible to send data.
The connection and conn_stream got a new "send_wait_list" entry, which
required to move a few struct members around to maintain an efficient
cache alignment (and actually this slightly improved performance).
For the same consistency reasons, let's use b_empty() at the few places
where an empty buffer is expected, or c_empty() if it's done on a channel.
Some of these places were there to realign the buffer so
{b,c}_realign_if_empty() was used instead.
The mux and transport rcv_buf() now takes a "flags" argument, just like
the snd_buf() one or like the equivalent syscall lower part. The upper
layers will use this to pass some information such as indicating whether
the buffer is free from outgoing data or if the lower layer may allocate
the buffer itself.
Just like we have a size_t for xprt->snd_buf(), we adjust to use size_t
for rcv_buf()'s count argument and return value. It also removes the
ambiguity related to the possibility to see a negative value there.
This way the senders don't need to modify the buffer's metadata anymore
nor to know about the output's split point. This way the functions can
take a const buffer and it's clearer who's in charge of updating the
buffer after a send. That's why the buffer realignment is now performed
by the caller of the transport's snd_buf() functions.
The return type was updated to return a size_t to comply with the count
argument.
Now when raw_sock_{to,from}_{pipe,buf} are called, if the connection
doesn't feature CO_FL_WILL_UPDATE, they will first retrieve the updated
flags using conn_refresh_polling_flags() before changing any flag, then
call conn_cond_update_sock_polling() before leaving, to commit such
changes. Note that the only real call to one of the __conn_* functions
is in fact in conn_sock_read0() which is called from here.
Till now connections used to rely exclusively on file descriptors. It
was planned in the past that alternative solutions would be implemented,
leading to member "union t" presenting sock.fd only for now.
With QUIC, the connection will need to continue to exist but will not
rely on a file descriptor but a connection ID.
So this patch introduces a "connection handle" which is either a file
descriptor or a connection ID, to replace the existing "union t". We've
now removed the intermediate "struct sock" which was never used. There
is no functional change at all, though the struct connection was inflated
by 32 bits on 64-bit platforms due to alignment.
I just noticed the raw socket constructor was called __ssl_sock_deinit,
which is a bit confusing, and wrong twice, so the attached patch renames it
to __raw_sock_init, which seems more correct.
Curu Wong reported a case where haproxy used to send RST to a server
after receiving its FIN. The problem is caused by the fact that being
a server connection, its fd is marked with linger_risk=1, and that the
poller didn't report POLLRDHUP, making haproxy unaware of a pending
shutdown that came after the data, so it used to resort to nolinger
for closing.
However when pollers support RDHUP we're pretty certain whether or not
a shutdown comes after the data and we don't need to perform that extra
recv() call.
Similarly when we're dealing with an inbound connection we don't care
and don't want to perform this extra recv after a request for a very
unlikely case, as in any case we'll have to deal with the client-facing
TIME_WAIT socket.
So this patch ensures that only when it's known that there's no risk with
lingering data, as well as in cases where it's known that the poller would
have detected a pending RDHUP, we perform the fd_done_recv() otherwise we
continue, trying a subsequent recv() to try to detect a pending shutdown.
This effectively results in an extra recv() call for keep-alive sockets
connected to a server when POLLRDHUP isn't known to be supported, but
it's the only way to know whether they're still alive or closed.
This fix should be backported to 1.7, 1.6 and 1.5. It relies on the
previous patch bringing support for the HAP_POLL_F_RDHUP flag.
There are still a lot of #ifdef USE_OPENSSL in the code (still 43
occurences) because we never know if we can directly access ssl_sock
or not. This patch attacks the problem differently by providing a
way for transport layers to register themselves and for users to
retrieve the pointer. Unregistered transport layers will point to NULL
so it will be easy to check if SSL is registered or not. The mechanism
is very inexpensive as it relies on a two-entries array of pointers,
so the performance will not be affected.
I was testing haproxy-1.5-dev22 on SmartOS (an illumos-based system)
and ran into a problem. There's a small window after non-blocking
connect() is called, but before the TCP connection is established,
where recv() may return ENOTCONN. On Linux, the behaviour here seems
to be always to return EAGAIN. The fix is relatively trivial, and
appears to make haproxy work reliably on current SmartOS (see patch
below). It's possible that other UNIX platforms exhibit this
behaviour as well.
Note: the equivalent was already done for send() in commit 0ea0cf6
("BUG: raw_sock: also consider ENOTCONN in addition to EAGAIN").
Both patches should be backported to 1.4.
This prevents us from passing other useful info and requires the
upper levels to know these flags. Let's use a new flags category
instead : CO_SFL_*. For now, only MSG_MORE has been remapped.
Due to a typo, the MSG_MORE flag used to replace MSG_NOSIGNAL and
MSG_DONTWAIT. Fortunately, sockets are always marked non-blocking,
so the loss of MSG_DONTWAIT is harmless, and the NOSIGNAL is covered
by the interception of the SIGPIPE. So no issue could have been
caused by this bug.
This is the reimplementation of the "done" action : when we experience
a short read, we're almost certain that we've exhausted the system's
buffers and that we'll meet an EAGAIN if we attempt to read again. If
the FD is not yet polled, the stream interface already takes care of
stopping the speculative read. When the FD is already being polled, we
have two options :
- either we're running from a level-triggered poller, in which case
we'd rather report that we've reached the end so that we don't
speculate over the poller and let it report next time data are
available ;
- or we're running from an edge-triggered poller in which case we
have no choice and have to see the EAGAIN to re-enable events.
At the moment we don't have any edge-triggered poller, so it's desirable
to avoid speculative I/O that we know will fail.
Note that this must not be ported to SSL since SSL hides the real
readiness of the file descriptor.
Thanks to this change, we observe no EAGAIN anymore during keep-alive
transfers, and failed recvfrom() are reduced by half in http-server-close
mode (the client-facing side is always being polled and the second recv
can be avoided). Doing so results in about 5% performance increase in
keep-alive mode. Similarly, we used to have up to about 1.6% of EAGAIN
on accept() (1/maxaccept), and these have completely disappeared under
high loads.
The recv/send callbacks must check for readiness themselves instead of
having their callers do it. This will strengthen the test and will also
ensure we never refrain from calling a handshake handler because a
direction is being polled while the other one is ready.
We simply remove these functions and replace their calls with the
appropriate ones :
- if we're in the data phase, we can simply report wait on the FD
- if we're in the socket phase, we may also have to signal the
desire to read/write on the socket because it might not be
active yet.
Steve Ruiz reported some reproducible crashes with HTTP health checks
on a certain page returning a huge length. The traces he provided
clearly showed that the recv() call was performed twice for a total
size exceeding the buffer's length.
Cyril Bonté tracked down the problem to be caused by the full buffer
size being passed to rcv_buf() in event_srv_chk_r() instead of passing
just the remaining amount of space. Indeed, this change happened during
the connection rework in 1.5-dev13 with the following commit :
f150317 MAJOR: checks: completely use the connection transport layer
But one of the problems is also that the comments at the top of the
rcv_buf() functions suggest that the caller only has to ensure the
requested size doesn't overflow the buffer's size.
Also, these functions already have to care about the buffer's size to
handle wrapping free space when there are pending data in the buffer.
So let's change the API instead to more closely match what could be
expected from these functions :
- the caller asks for the maximum amount of bytes it wants to read ;
This means that only the caller is responsible for enforcing the
reserve if it wants to (eg: checks don't).
- the rcv_buf() functions fix their computations to always consider
this size as a max, and always perform validity checks based on
the buffer's free space.
As a result, the code is simplified and reduced, and made more robust
for callers which now just have to care about whether they want the
buffer to be filled or not.
Since the bug was introduced in 1.5-dev13, no backport to stable versions
is needed.
Currently the control and transport layers of a connection are supposed
to be initialized when their respective pointers are not NULL. This will
not work anymore when we plan to reuse connections, because there is an
asymmetry between the accept() side and the connect() side :
- on accept() side, the fd is set first, then the ctrl layer then the
transport layer ; upon error, they must be undone in the reverse order,
then the FD must be closed. The FD must not be deleted if the control
layer was not yet initialized ;
- on the connect() side, the fd is set last and there is no reliable way
to know if it has been initialized or not. In practice it's initialized
to -1 first but this is hackish and supposes that local FDs only will
be used forever. Also, there are even less solutions for keeping trace
of the transport layer's state.
Also it is possible to support delayed close() when something (eg: logs)
tracks some information requiring the transport and/or control layers,
making it even more difficult to clean them.
So the proposed solution is to add two flags to the connection :
- CO_FL_CTRL_READY is set when the control layer is initialized (fd_insert)
and cleared after it's released (fd_delete).
- CO_FL_XPRT_READY is set when the control layer is initialized (xprt->init)
and cleared after it's released (xprt->close).
The functions have been adapted to rely on this and not on the pointers
anymore. conn_xprt_close() was unused and dangerous : it did not close
the control layer (eg: the socket itself) but still marks the transport
layer as closed, preventing any future call to conn_full_close() from
finishing the job.
The problem comes from conn_full_close() in fact. It needs to close the
xprt and ctrl layers independantly. After that we're still having an issue :
we don't know based on ->ctrl alone whether the fd was registered or not.
For this we use the two new flags CO_FL_XPRT_READY and CO_FL_CTRL_READY. We
now rely on this and not on conn->xprt nor conn->ctrl anymore to decide what
remains to be done on the connection.
In order not to miss some flag assignments, we introduce conn_ctrl_init()
to initialize the control layer, register the fd using fd_insert() and set
the flag, and conn_ctrl_close() which unregisters the fd and removes the
flag, but only if the transport layer was closed.
Similarly, at the transport layer, conn_xprt_init() calls ->init and sets
the flag, while conn_xprt_close() checks the flag, calls ->close and clears
the flag, regardless xprt_ctx or xprt_st. This also ensures that the ->init
and the ->close functions are called only once each and in the correct order.
Note that conn_xprt_close() does nothing if the transport layer is still
tracked.
conn_full_close() now simply calls conn_xprt_close() then conn_full_close()
in turn, which do nothing if CO_FL_XPRT_TRACKED is set.
In order to handle the error path, we also provide conn_force_close() which
ignores CO_FL_XPRT_TRACKED and closes the transport and the control layers
in turns. All relevant instances of fd_delete() have been replaced with
conn_force_close(). Now we always know what state the connection is in and
we can expect to split its initialization.
At some places, we report an error by just detecting FD_POLL_ERR.
The problem is that the caller never knows if it must use errno or
call getsockopt(SO_ERROR). And since this last one clears the
pending error from the queue, it cannot be used inconditionally.
An elegant solution consists in clearing errno prior to inspecting
FD_POLL_ERR. The caller then knows that if it gets CO_FL_ERROR and
errno == 0, it must call getsockopt().
