Igor Chan reported a very interesting bug which was triggered by the
recent dynamic size change in SSL.
The OpenSSL API refuses to send less data than any failed previous
attempt. So what's happening is that if an SSL_write() in streaming
mode sends 5kB of data and the openssl layer cannot send them all,
it returns SSL_ERROR_WANT_WRITE, which haproxy reacts to by enabling
polling on the file descriptor. In the mean time, haproxy may detect
that the buffer was almost full and will disable streaming mode. Upon
write notification, it will try to send again, but less data this
time (limited to tune.ssl_max_record). OpenSSL disagrees with this
and returns a generic error SSL_ERROR_SSL.
The solution which was found consists in adding a flag to the SSL
context to remind that we must not shrink writes after a failed
attempt. Thus, if EAGAIN is encountered, the next send() will not
be limited in order to retry the same size as before.
The ability to globally override the default client and server cipher
suites has been requested multiple times since the introduction of SSL.
This commit adds two new keywords to the global section for this :
- ssl-default-bind-ciphers
- ssl-default-server-ciphers
It is still possible to preset them at build time by setting the macros
LISTEN_DEFAULT_CIPHERS and CONNECT_DEFAULT_CIPHERS.
By having the stream interface pass the CF_STREAMER flag to the
snd_buf() primitive, we're able to tell the send layer whether
we're sending large chunks or small ones.
We use this information in SSL to adjust the max record dynamically.
This results in small chunks respecting tune.ssl.maxrecord at the
beginning of a transfer or for small transfers, with an automatic
switch to full records if the exchanges last long. This allows the
receiver to parse HTML contents on the fly without having to retrieve
16kB of data, which is even more important with small initcwnd since
the receiver does not need to wait for round trips to start fetching
new objects. However, sending large files still produces large chunks.
For example, with tune.ssl.maxrecord = 2859, we see 5 write(2885)
sent in two segments each and 6 write(16421).
This idea was first proposed on the haproxy mailing list by Ilya Grigorik.
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.
If no CA file specified on a server line, the config parser will show an error.
Adds an cmdline option '-dV' to re-set verify 'none' as global default on
servers side (previous behavior).
Also adds 'ssl-server-verify' global statement to set global default to
'none' or 'required'.
WARNING: this changes the default verify mode from "none" to "required" on
the server side, and it *will* break insecure setups.
In addition to previous outputs, we also emit the cumulated number of
connections, the cumulated number of requests, the maximum allowed
SSL connection concurrency, the current number of SSL connections and
the cumulated number of SSL connections. This will help troubleshoot
systems which experience memory shortage due to SSL.
It's easier and safer to rely on conn_ctrl_ready() everywhere than to
check the flag itself. It will also simplify adding extra checks later
if needed. Some useless controls for !ctrl have been removed, as the
CTRL_READY flag itself guarantees ctrl is set.
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.
Now we can more safely rely on the connection state to decide how to
drain and what to do when data are drained. Callers don't need to
manipulate the file descriptor's state anymore.
Note that it also removes the need for the fix ea90063 ("BUG/MEDIUM:
stream-int: fix the keep-alive idle connection handler") since conn_drain()
correctly sets the polling flags.
Recent commit abf08d9 ("BUG/MAJOR: connection: fix mismatch between rcv_buf's
API and usage") accidentely broke SSL by relying on an uninitialized value to
enter the read loop.
Many thanks to Cyril Bont� and Steve Ruiz for reporting this issue.
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 Bont� 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.
The connection will only remain there as a pre-allocated entity whose
goal is to be placed in ->end when establishing an outgoing connection.
All connection initialization can be made on this connection, but all
information retrieved should be applied to the end point only.
This change is huge because there were many users of si->conn. Now the
only users are those who initialize the new connection. The difficulty
appears in a few places such as backend.c, proto_http.c, peers.c where
si->conn is used to hold the connection's target address before assigning
the connection to the stream interface. This is why we have to keep
si->conn for now. A future improvement might consist in dynamically
allocating the connection when it is needed.
This patch just renames functions, types and enums. No code was changed.
A significant number of files were touched, especially the ACL arrays,
so it is likely that some external patches will not apply anymore.
One important thing is that we had to split ACL_PAT_* into two groups :
- ACL_TEST_{PASS|MISS|FAIL}
- PAT_{MATCH|UNMATCH}
A future patch will enforce enums on all these places to avoid confusion.
This patch just moves code without any change.
The ACL are just the association between sample and pattern. The pattern
contains the match method and the parse method. These two things are
different. This patch cleans the code by splitting it.
This is in preparation for associating a agent check
with a server which runs as well as the server's existing check.
The split has been made by:
* Moving elements of struct server's check element that will
be shared by both checks into a new check_common element
of struct server.
* Moving the remaining elements to a new struct check and
making struct server's check element a struct check.
* Adding a server element to struct check, a back-pointer
to the server element it is a member of.
- At this time the server could be obtained using
container_of, however, this will not be so easy
once a second struct check element is added to struct server
to accommodate an agent health check.
Signed-off-by: Simon Horman <horms@verge.net.au>
RFC6125 does not specify if wildcard matches empty strings but
classical browsers implementations does.
After the fix foo*bar.exemple.om matches foobar.exemple.com.
verifyhost allows you to specify a hostname that the remote server's
SSL certificate must match. Connections that don't match will be
closed with an SSL error.
Apollon Oikonomopoulos reported a build failure on Hurd where PATH_MAX
is not defined. The only place where it is referenced is ssl_sock.c,
all other places use MAXPATHLEN instead, with a fallback to 128 when
the OS does not define it. So let's switch to MAXPATHLEN as well.
We're having a lot of duplicate code just because of minor variants between
fetch functions that could be dealt with if the functions had the pointer to
the original keyword, so let's pass it as the last argument. An earlier
version used to pass a pointer to the sample_fetch element, but this is not
the best solution for two reasons :
- fetch functions will solely rely on the keyword string
- some other smp_fetch_* users do not have the pointer to the original
keyword and were forced to pass NULL.
So finally we're passing a pointer to the keyword as a const char *, which
perfectly fits the original purpose.
Benoit Dolez reported a failure to start haproxy 1.5-dev19. The
process would immediately report an internal error with missing
fetches from some crap instead of ACL names.
The cause is that some versions of gcc seem to trim static structs
containing a variable array when moving them to BSS, and only keep
the fixed size, which is just a list head for all ACL and sample
fetch keywords. This was confirmed at least with gcc 3.4.6. And we
can't move these structs to const because they contain a list element
which is needed to link all of them together during the parsing.
The bug indeed appeared with 1.5-dev19 because it's the first one
to have some empty ACL keyword lists.
One solution is to impose -fno-zero-initialized-in-bss to everyone
but this is not really nice. Another solution consists in ensuring
the struct is never empty so that it does not move there. The easy
solution consists in having a non-null list head since it's not yet
initialized.
A new "ILH" list head type was thus created for this purpose : create
an Initialized List Head so that gcc cannot move the struct to BSS.
This fixes the issue for this version of gcc and does not create any
burden for the declarations.
Since commit cfd97c6f was merged into 1.5-dev14 (BUG/MEDIUM: checks:
prevent TIME_WAITs from appearing also on timeouts), some valid health
checks sometimes used to show some TCP resets. For example, this HTTP
health check sent to a local server :
19:55:15.742818 IP 127.0.0.1.16568 > 127.0.0.1.8000: S 3355859679:3355859679(0) win 32792 <mss 16396,nop,nop,sackOK,nop,wscale 7>
19:55:15.742841 IP 127.0.0.1.8000 > 127.0.0.1.16568: S 1060952566:1060952566(0) ack 3355859680 win 32792 <mss 16396,nop,nop,sackOK,nop,wscale 7>
19:55:15.742863 IP 127.0.0.1.16568 > 127.0.0.1.8000: . ack 1 win 257
19:55:15.745402 IP 127.0.0.1.16568 > 127.0.0.1.8000: P 1:23(22) ack 1 win 257
19:55:15.745488 IP 127.0.0.1.8000 > 127.0.0.1.16568: FP 1:146(145) ack 23 win 257
19:55:15.747109 IP 127.0.0.1.16568 > 127.0.0.1.8000: R 23:23(0) ack 147 win 257
After some discussion with Chris Huang-Leaver, it appeared clear that
what we want is to only send the RST when we have no other choice, which
means when the server has not closed. So we still keep SYN/SYN-ACK/RST
for pure TCP checks, but don't want to see an RST emitted as above when
the server has already sent the FIN.
The solution against this consists in implementing a "drain" function at
the protocol layer, which, when defined, causes as much as possible of
the input socket buffer to be flushed to make recv() return zero so that
we know that the server's FIN was received and ACKed. On Linux, we can make
use of MSG_TRUNC on TCP sockets, which has the benefit of draining everything
at once without even copying data. On other platforms, we read up to one
buffer of data before the close. If recv() manages to get the final zero,
we don't disable lingering. Same for hard errors. Otherwise we do.
In practice, on HTTP health checks we generally find that the close was
pending and is returned upon first recv() call. The network trace becomes
cleaner :
19:55:23.650621 IP 127.0.0.1.16561 > 127.0.0.1.8000: S 3982804816:3982804816(0) win 32792 <mss 16396,nop,nop,sackOK,nop,wscale 7>
19:55:23.650644 IP 127.0.0.1.8000 > 127.0.0.1.16561: S 4082139313:4082139313(0) ack 3982804817 win 32792 <mss 16396,nop,nop,sackOK,nop,wscale 7>
19:55:23.650666 IP 127.0.0.1.16561 > 127.0.0.1.8000: . ack 1 win 257
19:55:23.651615 IP 127.0.0.1.16561 > 127.0.0.1.8000: P 1:23(22) ack 1 win 257
19:55:23.651696 IP 127.0.0.1.8000 > 127.0.0.1.16561: FP 1:146(145) ack 23 win 257
19:55:23.652628 IP 127.0.0.1.16561 > 127.0.0.1.8000: F 23:23(0) ack 147 win 257
19:55:23.652655 IP 127.0.0.1.8000 > 127.0.0.1.16561: . ack 24 win 257
This change should be backported to 1.4 which is where Chris encountered
this issue. The code is different, so probably the tcp_drain() function
will have to be put in the checks only.
I left a mistake in my previous patch bringing the crt-list feature,
it breaks clients with no SNI support.
Also remove the useless wildp = NULL as per a previous discussion.
We were using "tune.ssl.maxrecord 2000" and discovered an interesting
problem: SSL data sent from the server to the client showed occasional
corruption of the payload data.
The root cause was:
When ssl_max_record is smaller than the requested send amount
the ring buffer wrapping wasn't properly adjusting the
number of bytes to send.
I solved this by selecting the initial size based on the number
of output bytes that can be sent without splitting _before_ checking
against ssl_max_record.
This new pattern fetch returns the client certificate's SHA-1 fingerprint
(i.e. SHA-1 hash of DER-encoded certificate) in a binary chunk.
This can be useful to pass it to a server in a header or to stick a client
to a server across multiple SSL connections.
Improve the crt-list file format to allow a rule to negate a certain SNI :
<crtfile> [[!]<snifilter> ...]
This can be useful when a domain supports a wildcard but you don't want to
deliver the wildcard cert for certain specific domains.
The ALPN extension is meant to replace the now deprecated NPN extension.
This patch implements support for it. It requires a version of openssl
with support for this extension. Patches are available here right now :
http://html5labs.interopbridges.com/media/167447/alpn_patches.zip
Now that ACLs solely rely on sample fetch functions, make them use the
same arg mask. All inconsistencies have been fixed separately prior to
this patch, so this patch almost only adds a new pointer indirection
and removes all references to ARG*() in the definitions.
The parsing is still performed by the ACL code though.
ACL fetch functions used to directly reference a fetch function. Now
that all ACL fetches have their sample fetches equivalent, we can make
ACLs reference a sample fetch keyword instead.
In order to simplify the code, a sample keyword name may be NULL if it
is the same as the ACL's, which is the most common case.
A minor change appeared, http_auth always expects one argument though
the ACL allowed it to be missing and reported as such afterwards, so
fix the ACL to match this. This is not really a bug.
Samples fetches were relying on two flags SMP_CAP_REQ/SMP_CAP_RES to describe
whether they were compatible with requests rules or with response rules. This
was never reliable because we need a finer granularity (eg: an HTTP request
method needs to parse an HTTP request, and is available past this point).
Some fetches are also dependant on the context (eg: "hdr" uses request or
response depending where it's involved, causing some abiguity).
In order to solve this, we need to precisely indicate in fetches what they
use, and their users will have to compare with what they have.
So now we have a bunch of bits indicating where the sample is fetched in the
processing chain, with a few variants indicating for some of them if it is
permanent or volatile (eg: an HTTP status is stored into the transaction so
it is permanent, despite being caught in the response contents).
The fetches also have a second mask indicating their validity domain. This one
is computed from a conversion table at registration time, so there is no need
for doing it by hand. This validity domain consists in a bitmask with one bit
set for each usage point in the processing chain. Some provisions were made
for upcoming controls such as connection-based TCP rules which apply on top of
the connection layer but before instantiating the session.
Then everywhere a fetch is used, the bit for the control point is checked in
the fetch's validity domain, and it becomes possible to finely ensure that a
fetch will work or not.
Note that we need these two separate bitfields because some fetches are usable
both in request and response (eg: "hdr", "payload"). So the keyword will have
a "use" field made of a combination of several SMP_USE_* values, which will be
converted into a wider list of SMP_VAL_* flags.
The knowledge of permanent vs dynamic information has disappeared for now, as
it was never used. Later we'll probably reintroduce it differently when
dealing with variables. Its only use at the moment could have been to avoid
caching a dynamic rate measurement, but nothing is cached as of now.
This flag is used on ACL matches that support being looking up patterns
in trees. At the moment, only strings and IPs support tree-based lookups,
but the flag is randomly set also on integers and binary data, and is not
even always set on strings nor IPs.
Better get rid of this mess by only relying on the matching function to
decide whether or not it supports tree-based lookups, this is safer and
easier to maintain.
fe61656b added the ability to load a list of certificates from a file,
but error control was incomplete and misleading, as some errors such
as missing files were not reported, and errors reported with Alert()
instead of memprintf() were inappropriate and mixed with upper errors.
Also, the code really supports a single SNI filter right now, so let's
correct it and the doc for that, leaving room for later change if needed.
It designates a list of PEM file with an optional list of SNI filter
per certificate, with the following format for each line :
<crtfile>[ <snifilter>]*
Wildcards are supported in the SNI filter. The certificates will be
presented to clients who provide a valid TLS Server Name Indication
field matching one of SNI filter. If no SNI filter is specified the
CN and alt subjects are used.
Add new tunable "tune.ssl.maxrecord".
Over SSL/TLS, the client can decipher the data only once it has received
a full record. With large records, it means that clients might have to
download up to 16kB of data before starting to process them. Limiting the
record size can improve page load times on browsers located over high
latency or low bandwidth networks. It is suggested to find optimal values
which fit into 1 or 2 TCP segments (generally 1448 bytes over Ethernet
with TCP timestamps enabled, or 1460 when timestamps are disabled), keeping
in mind that SSL/TLS add some overhead. Typical values of 1419 and 2859
gave good results during tests. Use "strace -e trace=write" to find the
best value.
This trick was first suggested by Mike Belshe :
http://www.belshe.com/2010/12/17/performance-and-the-tls-record-size/
Then requested again by Ilya Grigorik who provides some hints here :
http://ofps.oreilly.com/titles/9781449344764/_transport_layer_security_tls.html#ch04_00000101
Openssl needs to access /dev/urandom to initialize its internal random
number generator. It does so when it needs a random for the first time,
which fails if it is a handshake performed after the chroot(), causing
all SSL incoming connections to fail.
This fix consists in calling RAND_bytes() to produce a random before
the chroot, which will in turn open /dev/urandom before it's too late,
and avoid the issue.
If the random generator fails to work while processing the config,
haproxy now fails with an error instead of causing SSL connections to
fail at runtime.
This new option ensures that there is no possible fallback to a default
certificate if the client does not provide an SNI which is explicitly
handled by a certificate.
