When using some log-format directives in header insertion without HTTP mode,
the config parser used to report a cryptic message about option httplog being
downgraded to tcplog and with "(null):0" as the file name and line number.
This is because the lfs_file and lfs_line were not properly set for some valid
use cases of log-format directives. Now we cover http-request and http-response
as well.
One year ago, commit 5d5b5d8 ("MEDIUM: proto_tcp: add support for tracking
L7 information") brought support for tracking L7 information in tcp-request
content rules. Two years earlier, commit 0a4838c ("[MEDIUM] session-counters:
correctly unbind the counters tracked by the backend") used to flush the
backend counters after processing a request.
While that earliest patch was correct at the time, it became wrong after
the second patch was merged. The code does what it says, but the concept
is flawed. "TCP request content" rules are evaluated for each HTTP request
over a single connection. So if such a rule in the frontend decides to
track any L7 information or to track L4 information when an L7 condition
matches, then it is applied to all requests over the same connection even
if they don't match. This means that a rule such as :
tcp-request content track-sc0 src if { path /index.html }
will count one request for index.html, and another one for each of the
objects present on this page that are fetched over the same connection
which sent the initial matching request.
Worse, it is possible to make the code do stupid things by using multiple
counters:
tcp-request content track-sc0 src if { path /foo }
tcp-request content track-sc1 src if { path /bar }
Just sending two requests first, one with /foo, one with /bar, shows
twice the number of requests for all subsequent requests. Just because
both of them persist after the end of the request.
So the decision to flush backend-tracked counters was not the correct
one. In practice, what is important is to flush countent-based rules
since they are the ones evaluated for each request.
Doing so requires new flags in the session however, to keep track of
which stick-counter was tracked by what ruleset. A later change might
make this easier to maintain over time.
This bug is 1.5-specific, no backport to stable is needed.
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.
Recent commit d7ad9f5 ("MAJOR: channel: add a new flag CF_WAKE_WRITE to
notify the task of writes") introduced this new CF_WAKE_WRITE flag that
an analyser which requires some free space to write must set if it wants
to be notified.
Unfortunately, some places were missing. More specifically, the
compression engine can rarely be stuck by a lack of output space,
especially when dealing with non-compressible data. It then has to
stop until some pending data are flushed and for this it must set
the CF_WAKE_WRITE flag. But these cases were missed by the commit
above.
Fortunately, this change was introduced very recently and never
released, so the impact was limited.
Huge thanks to Sander Klein who first reported this issue and who kindly
and patiently provided lots of traces and test data that made it possible
to reproduce, analyze, then fix this issue.
Patrick Hemmer reported that using unique_id_format and logs did not
report the same unique ID counter since commit 9f09521 ("BUG/MEDIUM:
unique_id: HTTP request counter must be unique!"). This is because
the increment was done while producing the log message, so it was
performed twice.
A better solution consists in fetching a new value once per request
and saving it in the request or session context for all of this
request's life.
It happens that sessions already have a unique ID field which is used
for debugging and reporting errors, and which differs from the one
sent in logs and unique_id header.
So let's change this to reuse this field to have coherent IDs everywhere.
As of now, a session gets a new unique ID once it is instanciated. This
means that TCP sessions will also benefit from a unique ID that can be
logged. And this ID is renewed for each extra HTTP request received on
an existing session. Thus, all TCP sessions and HTTP requests will have
distinct IDs that will be stable along all their life, and coherent
between all places where they're used (logs, unique_id header,
"show sess", "show errors").
This feature is 1.5-specific, no backport to 1.4 is needed.
It's a bit hasardous to wipe out all channel flags, this flag should
be left intact as it protects against recursive calls. Fortunately,
we have no possibility to meet this situation with current applets,
but better fix it before it becomes an issue.
This bug has been there for a long time, but it doesn't seem worth
backporting the fix.
Since commit 6b66f3e ([MAJOR] implement autonomous inter-socket forwarding)
introduced in 1.3.16-rc1, we've been relying on a stupid mechanism to wake
up the task after a write, which was an exact copy-paste of the reader side.
The principle was that if we empty a buffer and there's no forwarding
scheduled or if the *producer* is not in a connected state, then we wake
the task up.
That does not make any sense. It happens to wake up too late sometimes (eg,
when the request analyser waits for some room in the buffer to start to
work), and leads to unneeded wakeups in client-side keep-alive, because
the task is woken up when the response is sent, while the analysers are
simply waiting for a new request.
In order to fix this, we introduce a new channel flag : CF_WAKE_WRITE. It
is designed so that an analyser can explicitly request being notified when
some data were written. It is used only when the HTTP request or response
analysers need to wait for more room in the buffers. It is automatically
cleared upon wake up.
The flag is also automatically set by the functions which try to write into
a buffer from an applet when they fail (bi_putblk() etc...).
That allows us to remove the stupid condition above and avoid some wakeups.
In http-server-close and in http-keep-alive modes, this reduces from 4 to 3
the average number of wakeups per request, and increases the overall
performance by about 1.5%.
This reverts commit f3221f99acdd792352d4ee648d987270d74ca38e.
Igor reported some very strange breakage of his stats page which is
clearly caused by the chunking, though I don't see at first glance
what could be wrong. Better revert it for now.
In theory the principle is simple as we just need to send HTTP chunks
if the client is 1.1 compatible. In practice it's harder because we
have to append a CR LF after each block of data and we're never sure
to have the room for this. In order not to have to deal with this, we
instead send the CR LF prior to each chunk size. The only issue is for
the first chunk and for this reason we avoid to send the empty header
line when using chunked encoding.
We used to unconditionally disable client-side polling after the client
has posted its request. The goal was to avoid subscribing the file
descriptor to the poller for nothing.
This is perfect for the HTTP close mode where we know we won't have to
read on the client side anymore. However, when keep-alive is maintained
with the client, this makes the situation worse. Indeed, after the first
response, we'll have to wait for the client to send a next request and
since this is never immediate, we'll certainly poll. So what happens is
that polling is enabled after a response and disabled after a request,
so the polling is constantly alternating, which is very expensive with
epoll_ctl().
The solution implemented in this patch consists in only disabling the
polling if the client-side is not in keep-alive mode. That way we have
the best of both worlds. In close, we really close, and in keep-alive,
we poll only once.
The performance gained by this change is important, with haproxy jumping
from 158kreq/s to 184kreq/s (+16%) in HTTP keep-alive mode on a machine
which at best does 222k/s in raw TCP mode.
With this patch and the previous one, a keep-alive run with a fast
enough server (or enough concurrent connections to cover the connect
time) does no epoll_ctl() anymore during a run of ab -k. The net
measured gain is 19%.
Compression is normally disabled on HTTP/1.0 since it does not
support chunked encoded responses. But the test was incomplete, and
Bertrand Jacquin reported a case where if the server responded using
1.1 to an 1.0 request, then haproxy still used to compress (and of
course the client could not understand the response).
No backport is needed, this is 1.5-specific.
In HTTP keep-alive mode, if we receive a 401, we still have a chance
of being able to send the visitor again to the same server over the
same connection. This is required by some broken protocols such as
NTLM, and anyway whenever there is an opportunity for sending the
challenge to the proper place, it's better to do it (at least it
helps with debugging).
Idle connections are not monitored right now. So if a server closes after
a response without advertising it, it won't be detected until a next
request wants to use the connection. This is a bit problematic because
it unnecessarily maintains file descriptors and sockets in an idle
state.
This patch implements a very simple idle connection manager for the stream
interface. It presents itself as an I/O callback. The HTTP engine enables
it when it recycles a connection. If a close or an error is detected on the
underlying socket, it tries to drain as much data as possible from the socket,
detect the close and responds with a close as well, then detaches from the
stream interface.
Since comit b805f71 (MEDIUM: sample: let the cast functions set their
output type), the output type of a fetch function is automatically
considered and passed to the next converter. A bug introduced in
1.5-dev9 with commit f853c46 (MEDIUM: pattern/acl: get rid of
temp_pattern in ACLs) was revealed by this last one : the output type
remained string instead of UINT, causing the cast function to try to
cast the contents and to crash on a NULL deref.
Note: this fix was made after a careful review of all fetch functions.
A few non-trivial ones had their comments amended to clearly indicate
the output type.
There are very few users of http_proxy, and all of them complain about
the same thing : the request is passed unmodified to the server (in its
proxy form), and it is not possible to fix it using reqrep rules because
http_proxy happens after.
So let's have http_proxy fix the URL it has analysed to get rid of the
scheme and the host part. This will do what users of this feature expect.
In HTTP keep-alive, if we face a connection error to the server while sending
the request, the error should not be reported, and the client-side connection
should simply be closed, so that client knows it can retry. This can happen if
the server has too short a keep-alive timeout and quits at the same moment the
new request comes in.
When a connection to the server is complete, if the transaction
requests keep-alive mode, we don't shut the connection and we just
reinitialize the stream interface in order to be able to reuse the
connection afterwards.
Note that the server connection count is decremented, just like the
backend's, and that we still try to wake up waiters. But that makes
sense considering that we'll eventually be able to immediately pass
idle connections to waiters.
When allocating a new connection, only the caller knows whether it's
acceptable to reuse the previous one or not. Let's pass this information
to si_alloc_conn() which will do the cleanup if the connection is not
acceptable.
It's common to observe a an recv() call on the client side just after
the connect() to has been issued to the server side when running in
server close mode. The reason is that the whole request has been sent
and the shutw() has been queued in the channel, so the request message
switches to the MSG_CLOSED state, which didn't disable reading. Let's
do it now. That way the reading will only be re-enabled after the
response is transferred to the client. However if abortonclose is set,
we still leave it enabled.
strace shows a lot of EAGAIN on small response messages. This
is caused by the fact that the READ_DONTWAIT flag is not set
on response message, it's only there when we want to flush
pending data.
For small responses, it's a waste of CPU cycles to call recv()
for nothing since most of the time, everything we'll need will
be in the first response. Also, this will offer more opportunities
for using splice() to transfer data.
It's becoming increasingly difficult to ignore unwanted function returns in
debug code with gcc. Now even when you try to work around it, it suggests a
way to write your code differently. For example :
src/frontend.c:187:65: warning: if statement has empty body [-Wempty-body]
if (write(1, trash.str, trash.len) < 0) /* shut gcc warning */;
^
src/frontend.c:187:65: note: put the semicolon on a separate line to silence this warning
1 warning generated.
This is totally unacceptable, this code already had to be written this way
to shut it up in earlier versions. And now it comments the form ? What's the
purpose of the C language if you can't write anymore the code that does what
you want ?
Emeric proposed to just keep a global variable to drain such useless results
so that gcc stops complaining all the time it believes people who write code
are monkeys. The solution is acceptable because the useless assignment is done
only in debug code so it will not impact performance. This patch implements
this, until gcc becomes even "smarter" to detect that we tried to cheat.
With this patch, patterns can be compiled for two modes :
- match
- lookup
The match mode is used for example in ACLs or maps. The lookup mode
is used to lookup a key for pattern maintenance. For example, looking
up a network is different from looking up one address belonging to
this network.
A special case is made for regex. In lookup mode they return the input
regex string and do not compile the regex.
Now, the pat_parse_*() functions parses the incoming data. The input
"pattern" struct can be preallocated. If the parser needs to add some
buffers, it allocates memory.
The function pattern_register() runs the call to the parser, process
the key indexation and associate the "sample_storage" used by maps.
This is used later for increasing the compability with incoming
sample types. When multiple compatible types are supported, one
is arbitrarily used (eg: UINT).
SSL and keep-alive will need to be able to fail on allocation errors,
and the stream interface did not allow to report such a cause. The flag
will then be "RC" as already documented.
This reduces its size which is not reused by anything else. However it
will significantly improve the debugger's output since we'll now get
real state values.
The default case had to be enabled in the parsers because gcc tries
to optimize the switch/case and noticed some values were missing from
the enums and emitted a warning.
Here again we had some oversized and misaligned entries. The method
and the status don't need 4 bytes each, and there was a hole after
the status that does not exist anymore. That's 8 additional bytes
saved from http_txn and as much for the session.
Also some fields were slightly moved to present better memory access
patterns resulting in a steady 0.5% performance increase.
The task returned by stream_int_register_handler() is never used, however we
always need to access the appctx afterwards. So make it return the appctx
instead. We already plan for it to fail, which is the reason for the addition
of a few tests and the possibility for the HTTP analyser to return a status
code 500.
We're about to remove si->appctx, so first let's replace all occurrences
of its usage with a dynamic extract from si->end. A lot of code was changed
by search-n-replace, but the behaviour was intentionally not altered.
The code surrounding calls to stream_int_register_handler() was slightly
changed since we can only use si->end *after* the registration.
The outgoing connection is now allocated dynamically upon the first attempt
to touch the connection's source or destination address. If this allocation
fails, we fail on SN_ERR_RESOURCE.
As we didn't use si->conn anymore, it was removed. The endpoints are released
upon session_free(), on the error path, and upon a new transaction. That way
we are able to carry the existing server's address across retries.
The stream interfaces are not initialized anymore before session_complete(),
so we could even think about allocating them dynamically as well, though
that would not provide much savings.
The session initialization now makes use of conn_new()/conn_free(). This
slightly simplifies the code and makes it more logical. The connection
initialization code is now shorter by about 120 bytes because it's done
at once, allowing the compiler to remove all redundant initializations.
The si_attach_applet() function now takes care of first detaching the
existing endpoint, and it is called from stream_int_register_handler(),
so we can safely remove the calls to si_release_endpoint() in the
application code around this call.
A call to si_detach() was made upon stream_int_unregister_handler() to
ensure we always free the allocated connection if one was allocated in
parallel to setting an applet (eg: detect HTTP proxy while proceeding
with stats maybe).
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.
Since this is the applet context, call it ->appctx to avoid the confusion
with the pointer to the applet. Many places were changed but it's only a
renaming.
At the moment, stats require some preliminary storage just to store
some flags and codes that are parsed very early and used later. In
fact that doesn't make much sense and makes it very hard to allocate
the applet dynamically.
This patch changes this. Now stats_check_uri() only checks for the
validity of the request and the fact that it matches the stats uri.
It's handle_stats() which parses it. It makes more sense because
handle_stats() used to already perform some preliminary processing
such as verifying that POST contents are not missing, etc...
There is only one minor hiccup in doing so : the reqrep rules might
be processed in between. This has been addressed by moving
http_handle_stats() just after stats_check_uri() and setting s->target
at the same time. Now that s->target is totally operational, it's used
to mark the current request as being targetted at the stats, and this
information is used after the request processing to remove the HTTP
analysers and only let the applet handle the request.
Thus we guarantee that the storage for the applet is filled with the
relevant information and not overwritten when we switch to the applet.
There is a big trouble with the way POST is handled for the admin
stats page. The POST parameters are extracted from some http-request
rules, and if not round they return zero hoping for being called again
when more data passes. This results in the HTTP analyser being called
several times and all the rules prior to the stats being executed
multiple times as well. That includes rewrite rules.
So instead of doing this, we now move all the processing of the stats
into the stats applet.
That way we just set the stats applet in the HTTP analyser when a stats
request is detected, and the applet takes the time it needs to read the
arguments and respond. We could even imagine improving the applet to
support requests larger than a single buffer.
The code was almost only moved and minimally changed. Several new HTTP
states were added to the stats applet to emit headers, redirects and
to read POST. It was necessary to do this because the headers sent
depend on the parsing of the POST request. In the end it's beneficial
because we removed two stream_int_retnclose() calls.
In preparation for moving the POST processing to the applet, we first
add new states to the HTTP I/O handler. Till now st0 was only 0/1 for
start/end. We now replace it with an enum.
These two fetch methods predate the samples and used to store the
destination address into the server-facing connection's address field
because we had no other place at this time.
This will become problematic with the current connection changes, so
let's fix this.
This field was used by dumpstats to retrieve a pointer to the current
session, which may already be found from ->owner. With this change,
the stats code doesn't need the connection at all anymore.
We're trying to move the applets out of the struct connection. So
let's remove the dependence on xprt_st and introduce si->applet.st2
to store the missing contextual data instead.
In commit 8c3d0be (MEDIUM: Add DRAIN state and report it on the stats page),
the drain state was updated on every weight change except those that can be
sent via the web interface. This caused inconsistent state combinations to
be reported in the stats depending on the sequence (web then cli vs cli
then web).
It would seem that a call to set_server_drain_state() from within
server_recalc_eweight() would simplify things but that's not completely
certain yet.
We need to initialize the rdr_fmt list inconditionally. Using only
a redirect rule without an http-redirect may cause a crash during
deinit because of the list iterating from null.
We handle "http-request redirect" with a log-format string now, but we
leave "redirect" unaffected.
Note that the control of the special "/" case is move from the runtime
execution to the configuration parsing. If the format rule list is
empty, the build_logline() function does nothing.
We now have the following enums and all related functions return them and
consume them :
enum pat_match_res {
PAT_NOMATCH = 0, /* sample didn't match any pattern */
PAT_MATCH = 3, /* sample matched at least one pattern */
};
enum acl_test_res {
ACL_TEST_FAIL = 0, /* test failed */
ACL_TEST_MISS = 1, /* test may pass with more info */
ACL_TEST_PASS = 3, /* test passed */
};
enum acl_cond_pol {
ACL_COND_NONE, /* no polarity set yet */
ACL_COND_IF, /* positive condition (after 'if') */
ACL_COND_UNLESS, /* negative condition (after 'unless') */
};
It's just in order to avoid doubts when reading some code.
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 will be used later with maps. Each map will associate an entry with
a sample_storage value.
This patch changes the "parse" prototype and all the parsing methods.
The goal is to associate "struct sample_storage" to each entry of
"struct acl_pattern". Only the "parse" function can add the sample value
into the "struct acl_pattern".
