Level-7 retries are only possible with a restricted number of HTTP
return codes. While it is usually not safe to retry on 401 and 403, I
came up with an authentication backend which was not synchronizing
authentication of users. While not perfect, being allowed to also retry
on those return codes is really helpful and acts as a hotfix until we
can fix the backend.
Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>
This patch adds -m flag which allows to specify header name
matching method when deleting headers from http request/response.
Currently beg, end, sub, str and reg are supported.
This is related to GitHub issue #909
Baptiste reported a new crash affecting 2.3 which can be triggered
when using H2 on the backend, with http-reuse always and with a tens
of clients doing close only. There are a few combined cases which cause
this to happen, but each time the issue is the same, an already freed
session is dereferenced in session_unown_conn().
Two cases were identified to cause this:
- a connection referencing a session as its owner, which is detached
from the session's list and is destroyed after this session ends.
The test on conn->owner before calling session_unown_conn() is not
sufficent as the pointer is not null but is not valid anymore.
- a connection that never goes idle and that gets killed form the
mux, where session_free() is called first, then conn_free() calls
session_unown_conn() which scans the just freed session for older
connections. This one is only triggered with DEBUG_UAF
The reason for this session to be present here is that it's needed during
the connection setup, to be passed to conn_install_mux_be() to mux->init()
as the owning session, but it's never deleted aftrewards. Furthermore, even
conn_session_free() doesn't delete this pointer after freeing the session
that lies there. Both do definitely result in a use-after-free that's more
easily triggered under DEBUG_UAF.
This patch makes sure that the owner is always deleted after detaching
or killing the session. However it is currently not possible to clear
the owner right after a synchronous init because the proxy protocol
apparently needs it (a reg test checks this), and if we leave it past
the connection setup with the session not attached anywhere, it's hard
to catch the right moment to detach it. This means that the session may
remain in conn->owner as long as the connection has never been added to
nor removed from the session's idle list. Given that this patch needs to
remain simple enough to be backported, instead it adds a workaround in
session_unown_conn() to detect that the element is already not attached
anywhere.
This fix absolutely requires previous patch "CLEANUP: connection: do not
use conn->owner when the session is known" otherwise the situation will
be even worse, as some places used to rely on conn->owner instead of the
session.
The fix could theorically be backported as far as 1.8. However, the code
in this area has significantly changed along versions and there are more
risks of breaking working stuff than fixing real issues there. The issue
was really woken up in two steps during 2.3-dev when slightly reworking
the idle conns with commit 08016ab82 ("MEDIUM: connection: Add private
connections synchronously in session server list") and when adding
support for storing used H2 connections in the session and adding the
necessary call to session_unown_conn() in the muxes. But the same test
managed to crash 2.2 when built in DEBUG_UAF and patched like this,
proving that we used to already leave dangling pointers behind us:
| diff --git a/include/haproxy/connection.h b/include/haproxy/connection.h
| index f8f235c1a..dd30b5f80 100644
| --- a/include/haproxy/connection.h
| +++ b/include/haproxy/connection.h
| @@ -458,6 +458,10 @@ static inline void conn_free(struct connection *conn)
| sess->idle_conns--;
| session_unown_conn(sess, conn);
| }
| + else {
| + struct session *sess = conn->owner;
| + BUG_ON(sess && sess->origin != &conn->obj_type);
| + }
|
| sockaddr_free(&conn->src);
| sockaddr_free(&conn->dst);
It's uncertain whether an existing code path there can lead to dereferencing
conn->owner when it's bad, though certain suspicious memory corruption bugs
make one think it's a likely candidate. The patch should not be hard to
adapt there.
Backports to 2.1 and older are left to the appreciation of the person
doing the backport.
A reproducer consists in this:
global
nbthread 1
listen l
bind :9000
mode http
http-reuse always
server s 127.0.0.1:8999 proto h2
frontend f
bind :8999 proto h2
mode http
http-request return status 200
Then this will make it crash within 2-3 seconds:
$ h1load -e -r 1 -c 10 http://0:9000/
If it does not, it might be that DEBUG_UAF was not used (it's harder then)
and it might be useful to restart.
At a few places we used to rely on conn->owner to retrieve the session
while the session is already known. This is not correct because at some
of these points the reason the connection's owner was still the session
(instead of NULL) is a mistake. At one place a comparison is even made
between the session and conn->owner assuming it's valid without checking
if it's NULL. Let's clean this up to use the session all the time.
Note that this will be needed for a forthcoming fix and will have to be
backported.
HAVE_SSL_CTX_SET_CIPHERSUITES is newly defined macro set in openssl-compat.h,
which helps to identify ssl libs (currently OpenSSL-1.1.1 only) that supports
TLS13 cipersuites manipulation on TLS13 context
in the context of a progressive backend migration, we want to be able to
activate SSL on outgoing connections to the server at runtime without
reloading.
This patch adds a `set server ssl` command; in order to allow that:
- add `srv_use_ssl` to `show servers state` command for compatibility,
also update associated parsing
- when using default-server ssl setting, and `no-ssl` on server line,
init SSL ctx without activating it
- when triggering ssl API, de/activate SSL connections as requested
- clean ongoing connections as it is done for addr/port changes, without
checking prior server state
example config:
backend be_foo
default-server ssl
server srv0 127.0.0.1:6011 weight 1 no-ssl
show servers state:
5 be_foo 1 srv0 127.0.0.1 2 0 1 1 15 1 0 4 0 0 0 0 - 6011 - -1
where srv0 can switch to ssl later during the runtime:
set server be_foo/srv0 ssl on
5 be_foo 1 srv0 127.0.0.1 2 0 1 1 15 1 0 4 0 0 0 0 - 6011 - 1
Also update existing tests and create a new one.
Signed-off-by: William Dauchy <wdauchy@gmail.com>
Functions registered to release memory per-thread have no return value. But the
registering function and the function pointer in per_thread_free_fct structure
specify it should return an integer. This patch fixes it.
This patch may be backported as far as 2.0.
Define a per-thread counters allocated with the greatest size of any
stat module counters. This variable is named trash_counters.
When using a proxy without allocated counters, return the trash counters
from EXTRA_COUNTERS_GET instead of a dangling pointer to prevent
segfault.
This is useful for all the proxies used internally and not
belonging to the global proxy list. As these objects does not appears on
the stat report, it does not matter to use the dummy counters.
For this fix to be functional, the extra counters are explicitly
initialized to NULL on proxy/server/listener init functions.
Most notably, the crash has already been detected with the following
vtc:
- reg-tests/lua/txn_get_priv.vtc
- reg-tests/peers/tls_basic_sync.vtc
- reg-tests/peers/tls_basic_sync_wo_stkt_backend.vtc
There is probably other parts that may be impacted (SPOE for example).
This bug was introduced in the current release and do not need to be
backported. The faulty commits are
"MINOR: ssl: count client hello for stats" and
"MINOR: ssl: add counters for ssl sessions".
Do not cache responses that do not have an explicit expiration time
(s-maxage or max-age Cache-Control directives or Expires header) or a
validator (ETag or Last-Modified headers) anymore, as suggested in
RFC 7234#3.
The TX_FLAG_IGNORE flag is used instead of the TX_FLAG_CACHEABLE so as
not to change the behavior of the checkcache option.
No details are provided when an error occurs during the parsing of an errorfile,
Thus it is a bit hard to diagnose where the problem is. Now, when it happens, an
understandable error message is reported.
This patch is not a bug fix in itself. But it will be required to change an
fatal error into a warning in last stable releases. Thus it must be backported
as far as 2.0.
It does not require heavy deletion from the expr anymore, so we can now
turn this to a single-linked list since most of the time we want to delete
all instances of a given pattern from the head. By doing so we save 32 bytes
of memory per pattern. The pat_unlink_from_head() function was adjusted
accordingly.
This function will be usable to purge at most a specified number of old
entries from a reference. Entries are declared old if their generation
number is in the past compared to the one passed in argument. This will
ease removal of early entries when new ones have been appended.
We also call malloc_trim() when available, at the end of the series,
because this is one place where there is a lot of memory to save. Reloads
of 1M IP addresses used in an ACL made the process grow up to 1.7 GB RSS
after 10 reloads and roughly stabilize there without this call, versus
only 260 MB when the call is present. Sadly there is no direct equivalent
for jemalloc, which stabilizes around 800MB-1GB.
pat_ref_load() basically combines pat_ref_append() and pat_ref_commit().
It's very similar to pat_ref_add() except that it also allows to set the
generation ID and the line number. pat_ref_add() was modified to directly
rely on it to avoid code duplication. Note that a previous declaration
of pat_ref_load() was removed as it was just a leftover of an earlier
incarnation of something possibly similar, so no existing functionality
was changed here.
This function will be used after a successful pat_ref_append() to propagate
the pattern to all use places (including parsing and indexing). On failure,
it will entirely roll back all insertions and free the pattern itself. It
also preserves the generation number so that it is convenient for use in
association with pat_ref_append(). pat_ref_add() was modified to rely on
it instead of open-coding the insertion and roll-back.
Right now it's not possible to perform a safe reload because we don't
know what patterns were recently added or were already present. This
patch adds a generation counter to the reference patterns so that it
is possible to know what generation of the reference they were loaded
with. A reference now has two generations, the current one, used for
all additions, and the next one, allocated to those wishing to update
the contents. The generation wraps at 2^32 so comparisons must be made
relative to the current position.
The idea will be that upon full reload, the caller will first get a new
generation ID, will insert all new patterns using it, will then switch
the current ID to the new one, and will delete all entries older than
the current ID. This has the benefit of supporting chunked updates that
remain consistent and that won't block the whole process for ages like
pat_ref_reload() currently does.
Till now the only way to remove a known reference was via
pat_ref_delete_by_id() which scans the whole list to find a matching pointer.
Let's add pat_ref_delete_by_ptr() which takes a valid pointer. It can be
called by the function above after the pointer is found, and can also be
used to roll back a failed insertion much more efficiently.
These ones are not used anymore, so let's remove them to remove a bit
of the complexity. The ACL keyword's delete() function could be removed
as well, though most keyword declarations are positional and we have a
high risk of introducing a mistake here, so let's not touch the ACL part.
pat_del_tree_gen() was already chained onto pat_del_list_gen() to deal
with remaining cases, so let's complete the merge and have a generic
pattern deletion function acting on the reference and taking care of
reliably removing all elements.
This is the next step in speeding up entry removal. Now we don't scan
the whole lists or trees for elements pointing to the target reference,
instead we start from the reference and delete all linked patterns.
This simplifies some delete functions since we don't need anymore to
delete multiple times from an expression since all nodes appear after
the reference element. We can now have one generic list and one generic
tree deletion function.
This required the replacement of pattern_delete() with an open-coded
version since we now need to lock all expressions first before proceeding.
This means there is a high risk of lock inversion here but given that the
expressions are always scanned in the same order from the same head, this
must not happen.
Now deleting first entries is instantaneous, and it's still slow to
delete the last ones when looking up their ID since it still requires
to look them up by a full scan, but it's already way faster than
previously. Typically removing the last 10 IP from a 20M entries ACL
with a full-scan each took less than 2 seconds.
It would be technically possible to make use of indexed entries to
speed up most lookups for removal by value (e.g. IP addresses) but
that's for later.
There is a data model issue in the current pattern design that makes
pattern deletion extremely expensive: there's no direct way from a
reference to access all indexed occurrences. As such, the only way
to remove all indexed entries corresponding to a reference update
is to scan all expressions's lists and trees to find a link to the
reference. While this was possibly OK when map removal was not
common and most maps were small, this is not conceivable anymore
with GeoIP maps containing 10M+ entries and del-map operations that
are triggered from http-request rulesets.
This patch introduces two list heads from the pattern reference, one
for the objects linked by lists and one for those linked by tree node.
Ideally a single list would be enough but the linked elements are too
much unrelated to be distinguished at the moment, so we'll need two
lists. However for the long term a single-linked list will suffice but
for now it's not possible due to the way elements are removed from
expressions. As such this patch adds 32 bytes of memory usage per
reference plus 16 per indexed entry, but both will be cut in half
later.
The links are not yet used for deletion, this patch only ensures the
list is always consistent.
Now we have a single prune() function to act on an expression, and one
delete function for the lists and one for the trees. The presence of a
pointer in the lists is enough to warrant a free, and we rely on the
PAT_SF_REGFREE flag to decide whether to free using free() or regfree().
Currently we have no way to know how to delete/prune a pattern in a
generic way. A pattern doesn't contain its own type so we don't know
what function to call. Tree nodes are roughly OK but not lists where
regex are possible. Let's add one new bit for sflags at index time to
indicate that regex_free() will be needed upon deletion. It's not used
for now.
It's not possible to uniquely update a single expression without updating
the pattern reference, I don't know why we've put the revision in the
expression back then, given that it in fact provides an update for a
full pattern. Let's move the revision into the reference's head instead.
This is in order to access malloc_trim() which is convenient after
clearing huge maps to reclaim memory. When this is detected, we also
define HA_HAVE_MALLOC_TRIM.
This patch implements a couple of converters to validate and extract data from a
MQTT (Message Queuing Telemetry Transport) message. The validation consists of a
few checks as well as "packet size" validation. The extraction can get any field
from the variable header and the payload.
This is limited to CONNECT and CONNACK packet types only. All other messages are
considered as invalid. It is not a problem for now because only the first packet
on each side can be parsed (CONNECT for the client and CONNACK for the server).
MQTT 3.1.1 and 5.0 are supported.
Reviewed and Fixed by Christopher Faulet <cfaulet@haproxy.com>
This patch implements a couple of converters to validate and extract tag value
from a FIX (Financial Information eXchange) message. The validation consists in
a few checks such as mandatory fields and checksum computation. The extraction
can get any tag value based on a tag string or tag id.
This patch requires the istend() function. Thus it depends on "MINOR: ist: Add
istend() function to return a pointer to the end of the string".
Reviewed and Fixed by Christopher Faulet <cfaulet@haproxy.com>
Released version 2.4-dev0 with the following main changes :
- MINOR: version: it's development again.
- DOC: mention in INSTALL that it's development again
Let us use SSL_CTX_set1_curves_list, defined by OpenSSL, as well as in
openssl-compat when SSL_CTRL_SET_CURVES_LIST is present (BoringSSL),
for feature detection instead of versions.
OpenSSL 1.0.2 and onwards define SSL_CTX_set1_curves_list which is both a
function and a macro. OpenSSL 1.0.2 to 1.1.0 define SSL_CTRL_SET_CURVES_LIST
as a macro, which disappeared from 1.1.1. BoringSSL only has that one and
not the former macro but it does have the function. Let's keep the test on
the macro matching the function name by defining the macro to itself when
needed.
This does the same as fd_want_recv() except that it does check for
fd_updt[] to be allocated, as this may be called during early listener
initialization. Previously we used to check fd_updt[] before calling
fd_want_recv() but this is not correct since it does not update the
FD flags. This method will be safer.
The mem stats are pretty convenient to spot leaks, except that they count
free(NULL) as 1, and the code does actually have quite a number of free(foo)
guards where foo is NULL if the object was already freed. Let's just not
count these ones so that the stats remain consistent. Now it's possible
to compare the strdup()/malloc() and free() and verify they are consistent.
It's more convenient to return the element than to return just 0 or 1,
as the next thing we'll want to do is to act on this element! In addition
it was using variable arguments instead of consts, causing some reuse
constraints which were also addressed. This doesn't change its use as
a boolean, hence why call places were not modified.
The "weight" column on the stats page is somewhat confusing when using
slowstart becaue it reports the effective weight, without being really
explicit about it. In some situations the user-configured weight is more
relevant (especially with long slowstarts where it's important to know
if the configured weight is correct).
This adds a new uweight stat which reports a server's user-configured
weight, and in a backend it receives the sum of all servers' uweights.
In addition it adds the mention of "effective" in a few descriptions
for the "weight" column (help and doc).
As a result, the list of servers in a backend is now always scanned
when dumping the stats. But this is not a problem given that these
servers are already scanned anyway and for way heavier processing.
"no-maint" is a bit similar to "up" except that it will only hide
servers that are in maintenance (or disabled in the configuration), and
not those that are enabled but failed a check. One benefit here is to
significantly reduce the output of the "show stat" command when using
large server-templates containing entries that are not yet provisioned.
Note that the prometheus exporter also has such an option which does
the exact same.
This reverts commit b7ba1d9011. Actually
this test had already been removed in the past by commit fac0f645d
("BUG/MEDIUM: queue: make pendconn_cond_unlink() really thread-safe"),
but the condition to reproduce the bug mentioned there was not clear.
Now after analysis and a certain dose of code cleanup, things start to
appear more obvious. what happens is that if we check the presence of
the node in the tree without taking the lock, we can see the NULL at
the instant the node is being unlinked by another thread in
pendconn_process_next_strm() as part of __pendconn_unlink_prx() or
__pendconn_unlink_srv(). Till now there is no issue except that the
pendconn is not removed from the queue during this operation and that
the task is scheduled to be woken up by pendconn_process_next_strm()
with the stream being added to the list of the server's active
connections by __stream_add_srv_conn(). The first thread finishes
faster and gets back to stream_free() faster than the second one
sets the srv_conn on the stream, so stream_free() skips the s->srv_conn
test and doesn't try to dequeue the freshly queued entry. At the
very least a barrier would be needed there but we can't afford to
free the stream while it's being queued. So there's no other solution
than making sure that either __pendconn_unlink_prx() or
pendconn_cond_unlink() get the entry but never both, which is why the
lock is required around the test. A possible solution would be to set
p->target before unlinking the entry and using it to complete the test.
This would leave no dead period where the pendconn is not seen as
attached.
It is possible, yet extremely difficult, to reproduce this bug, which
was first noticed in bug #880. Running 100 servers with maxconn 1 and
maxqueue 1 on leastconn and a connect timeout of 30ms under 16 threads
with DEBUG_UAF, with a traffic making the backend's queue oscillate
around zero (typically using 250 connections with a local httpterm
server) may rarely manage to trigger a use-after-free.
No backport is needed.
On connection error processing, we can see massive storms of calls to
pendconn_cond_unlink() to release a possible place in the queue. For
example, in issue #908, on average half of the threads are caught in
this function via back_try_conn_req() consecutive to a synchronous
error. However we wait until grabbing the lock to know if the pendconn
is effectively in a queue, which is expensive for many cases. We know
the transition may only happen from in-queue to out-of-queue so it's safe
to first run a preliminary check to see if it's worth going further. This
will allow to avoid the cost of locking for most requests. This should
not change anything for those completing correctly as they're already
run through pendconn_free() which doesn't call pendconn_cond_unlink()
unless deemed necessary.
This is an anticipation of finer grained locking for the queues. For now
all lock places take a write lock so that there is no difference at all
with previous code.
In addition to the previous simplification, most locks don't use the
seek or read lock (e.g. spinlocks etc) so let's split the dump into
distinct operations (write/seek/read) and only report those which
were used. Now the output size is roughly divided by 5 compared
to previous ones.
The lock stats are very verbose and more than half of them are used in
a typical test, making it hard to spot the sought values. Let's simply
report "not used" for those which have not been called at all.
When an internal response is returned to a client, the message payload must be
skipped if it is a reply to a HEAD request. The payload is removed from the HTX
message just before the message forwarding.
This bugs has been around for a long time. It was already there in the pre-HTX
versions. In legacy HTTP mode, internal errors are not parsed. So this bug
cannot be easily fixed. Thus, this patch should only be backported in all HTX
versions, as far as 2.0. However, the code has significantly changed in the
2.2. Thus in the 2.1 and 2.0, the patch must be entirely reworked.
Partial support of conditional HTTP requests. This commit adds the
support of the 'If-None-Match' header (see RFC 7232#3.2).
When a client specifies a list of ETags through one or more
'If-None-Match' headers, they are all compared to the one that might have
been stored in the corresponding http cache entry until one of them
matches.
If a match happens, a specific "304 Not Modified" response is
sent instead of the cached data. This response has all the stored
headers but no other data (see RFC 7232#4.1). Otherwise, the whole cached data
is sent.
Although unlikely in a GET/HEAD request, the "If-None-Match: *" syntax is
valid and also receives a "304 Not Modified" response (RFC 7434#4.3.2).
This resolves a part of GitHub issue #821.
Add a function that compares two etags that might be of different types.
If any of them is weak, the 'W/' prefix is discarded and a strict string
comparison is performed.
Co-authored-by: Tim Duesterhus <tim@bastelstu.be>
In issue #785, users are reporting that it's not convenient to load a
".crt.key" when the configuration contains a ".crt".
This option allows to remove the extension of the certificate before
trying to load any extra SSL file (.key, .ocsp, .sctl, .issuer etc.)
The patch changes a little bit the way ssl_sock_load_files_into_ckch()
looks for the file.
When a connection is released, depending on its state, it may be detached from
the session and it may be removed from the server lists. The first case may
happen for private or unsharable active connections. The second one should only
be performed for idle or available connections. We never try to remove a
connection from the server list if it is attached to a session. But it is also
important to never try to remove a private connecion from the server lists, even
if it is not attached to a session. Otherwise, the curr_used_conn server counter
is decremented once too often.
This bug was introduced by the commit 04a24c5ea ("MINOR: connection: don't check
priv flag on free"). It is related to the issue #881. It only affects the 2.3,
no backport is needed.
This counter is only updated and never used, and in addition it's done
without any atomicity so it's very unlikely to be correct on multi-CPU
systems! Let's just remove it since it's not used.
A few structures were slightly rearranged in order to plug some holes
left around the locks. Sizes ranging from 8 to 32 bytes could be saved
depending on the structures. No performance difference was noticed (none
was expected there), though memory usage might be slightly reduced in
some rare cases.
We don't need to waste the size of a long for the locks: with the plocks,
even an unsigned short would offer enough room for up to 126 threads! Let's
use an unsigned int which will be easier to place in certain structures
and will more conveniently plug some holes, and Atomic ops are at least
as fast on 32-bit as on 64-bit. This will not change anything for 32-bit
platforms.
It's a bit overkill to register an initcall to call a function to set
a lock to zero when not debugging, let's just declare the lock as
pre-initialized to zero.
BoringSSL is a fork of OpenSSL 1.1.0, however in
49e9f67d8b7cbeb3953b5548ad1009d15947a523 it has changed version to 1.1.1.
Should fix issue #895.
This must be backported to 2.2, 2.1, 2.0, 1.8
It was previously a spinlock, and it happens that a number of LB algos
only lock it for lookups, without performing any modification. Let's
first turn it to an rwlock and w-lock it everywhere. This is strictly
identical.
It was carefully checked that every HA_SPIN_LOCK() was turned to
HA_RWLOCK_WRLOCK() and that HA_SPIN_UNLOCK() was turned to
HA_RWLOCK_WRUNLOCK() on this lock. _INIT and _DESTROY were updated too.
Since our locks are based on progressive locks, we support the upgradable
seek lock that is compatible with readers and upgradable to a write lock.
The main purpose is to take it while seeking down a tree for modification
while other threads may seek the same tree for an input (e.g. compute the
next event date).
The newly supported operations are:
HA_RWLOCK_SKLOCK(lbl,l) pl_take_s(l) /* N --> S */
HA_RWLOCK_SKTOWR(lbl,l) pl_stow(l) /* S --> W */
HA_RWLOCK_WRTOSK(lbl,l) pl_wtos(l) /* W --> S */
HA_RWLOCK_SKTORD(lbl,l) pl_stor(l) /* S --> R */
HA_RWLOCK_WRTORD(lbl,l) pl_wtor(l) /* W --> R */
HA_RWLOCK_SKUNLOCK(lbl,l) pl_drop_s(l) /* S --> N */
HA_RWLOCK_TRYSKLOCK(lbl,l) (!pl_try_s(l)) /* N -?> S */
HA_RWLOCK_TRYRDTOSK(lbl,l) (!pl_try_rtos(l)) /* R -?> S */
Existing code paths are left unaffected so this patch doesn't affect
any running code.
We currently use only read and write lock operations with rwlocks, but
ours also support upgradable seek locks for which we do not report any
stats. Let's add them now when DEBUG_THREAD is enabled.
We don't need to specify the handler anymore since it's set in the
receiver. Let's remove this argument from the function and clean up
the remains of code that were still setting it.
Now we define a new sock_accept_iocb() for socket-based stream protocols
and use it as a wrapper for listener_accept() which now takes a listener
and not an FD anymore. This will allow the receiver's I/O cb to be
redefined during registration, and more specifically to get rid of the
hard-coded hacks in protocol_bind_all() made for syslog.
The previous ->accept() callback in the protocol was removed since it
doesn't have anything to do with accept() anymore but is more generic.
A few places where listener_accept() was compared against the FD's IO
callback for debugging purposes on the CLI were updated.
For now we're still using the protocol's default accept() function as
the I/O callback registered by the receiver into the poller. While
this is usable for most TCP connections where a listener is needed,
this is not suitable for UDP where a different handler is needed.
Let's make this configurable in the receiver just like the upper layer
is configurable for listeners. In order to ease stream protocols
handling, the protocols will now provide a default I/O callback
which will be preset into the receivers upon allocation so that
almost none of them has to deal with it.
The socket-specific accept() code in listener_accept() has nothing to
do there. Let's move it to sock.c where it can be significantly cleaned
up. It will now directly return an accepted connection and provide a
status code instead of letting listener_accept() deal with various errno
values. Note that this doesn't support the sockpair specific code.
The function is now responsible for dealing with its own receiver's
polling state and calling fd_cant_recv() when facing EAGAIN.
One tiny change from the previous implementation is that the connection's
sockaddr is now allocated before trying accept(), which saves a memcpy()
of the resulting address for each accept at the expense of a cheap
pool_alloc/pool_free on the final accept returning EAGAIN. This still
apparently slightly improves accept performance in microbencharks.
This per-protocol function will be used to accept an incoming
connection and return it as a struct connection*. As such the protocol
stack's internal representation of a connection will not need to be
handled by the listener code.
This call was introduced by commit 5ced3e887 ("MINOR: sock: add
sock_accept_conn() to test a listening socket") but is actually quite
confusing because it makes one think the socket will accept a connection
(which is what we want to have in a new function) while it only tells
whether it's configured to accept connections. Let's call it
sock_accepting_conn() instead.
The same change was applied to sockpair which had the same issue.
accept_conn() will be used to accept an incoming connection and return it.
It will have to deal with various error codes. The currently identified
ones were created as CO_AC_*.
Till now we would keep a per-thread queue of pending incoming connections
for which we would store:
- the listener
- the accepted FD
- the source address
- the source address' length
And these elements were first used in session_accept_fd() running on the
target thread to allocate a connection and duplicate them again. Doing
this induces various problems. The first one is that session_accept_fd()
may only run on file descriptors and cannot be reused for QUIC. The second
issue is that it induces lots of memory copies and that the listerner
queue thrashes a lot of cache, consuming 64 bytes per entry.
This patch changes this by allocating the connection before queueing it,
and by only placing the connection's pointer into the queue. Indeed, the
first two calls used to initialize the connection already store all the
information above, which can be retrieved from the connection pointer
alone. So we just have to pop one pointer from the target thread, and
pass it to session_accept_fd() which only needs the FD for the final
settings.
This starts to make the accept path a bit more transport-agnostic, and
saves memory and CPU cycles at the same time (1% connection rate increase
was noticed with 4 threads). Thanks to dividing the accept-queue entry
size from 64 to 8 bytes, its size could be increased from 256 to 1024
connections while still dividing the overall size by two. No single
queue full condition was met.
One minor drawback is that connection may be allocated from one thread's
pool to be used into another one. But this already happens a lot with
connection reuse so there is really nothing new here.
Roughly half of the calls to sockadr_alloc() are made to copy an already
known address. Let's optionally pass it in argument so that the function
can handle the copy at the same time, this slightly simplifies its usage.
fd_done_recv() used to be useful with the FD cache because it used to
allow to keep a file descriptor active in the poller without being
marked as ready in the cache, saving it from ringing immediately,
without incurring any system call. It was a way to make it yield
to wait for new events leaving a bit of time for others. The only
user left was the connection accepter (listen_accept()). We used
to suspect that with the FD cache removal it had become totally
useless since changing its readiness or not wouldn't change its
status regarding the poller itself, which would be the only one
deciding to report it again.
Careful tests showed that it indeed has exactly zero effect nowadays,
the syscall numbers are exactly the same with and without, including
when enabling edge-triggered polling.
Given that there's no more API available to manipulate it and that it
was directly called as an optimization from listener_accept(), it's
about time to remove it.
No protocol defines it anymore. The last user used to be the monitor-net
stuff that got partially broken already when the tcp_drain() function
moved to conn_sock_drain() with commit e215bba95 ("MINOR: connection:
make conn_sock_drain() work for all socket families") in 1.9-dev2.
A part of this will surely move back later when non-socket connections
arrive with QUIC but better keep the API clean and implement what's
needed in time instead.
As discussed here during 2.1-dev, "monitor-net" is totally obsolete:
https://www.mail-archive.com/haproxy@formilux.org/msg35204.html
It's fundamentally incompatible with usage of SSL, and imposes the
presence of file descriptors with hard-coded syscalls directly in the
generic accept path.
It's very unlikely that anyone has used it in the last 10 years for
anything beyond testing. In the worst case if anyone would depend
on it, replacing it with "http-request return status 200 if ..." and
"mode http" would certainly do the trick.
The keyword is still detected as special by the config parser to help
users update their configurations appropriately.
As discussed here during 2.1-dev, "mode health" is totally obsolete:
https://www.mail-archive.com/haproxy@formilux.org/msg35204.html
It's fundamentally incompatible with usage of SSL, doesn't support
source filtering, and imposes the presence of file descriptors with
hard-coded syscalls directly in the generic accept path.
It's very unlikely that anyone has used it in the last 10 years for
anything beyond testing. In the worst case if anyone would depend
on it, replacing it with "http-request return status 200" and "mode
http" would certainly do the trick.
The keyword is still detected as special by the config parser to help
users update their configurations appropriately.
Do not check CO_FL_PRIVATE flag to check if the connection is in session
list on conn_free. This is necessary due to the future patches which add
server connections in the session list even if not private, if the mux
protocol is the subject of HOL blocking.
On server connection migration from one thread to another, the wrong
idle thread-specific counter is decremented. This bug was introduced
since commit 3d52f0f1f8 due to the
factorization with srv_use_idle_conn. However, this statement is only
executed from conn_backend_get. Extract the decrement from
srv_use_idle_conn in conn_backend_get and use the correct
thread-specific counter.
Rename the function to srv_use_conn to better reflect its purpose as it
is also used with a newly initialized connection not in the idle list.
As a side change, the connection insertion to available list has also
been extracted to conn_backend_get. This will be useful to be able to
specify an alternative list for protocol subject to HOL risk that should
not be shared between several clients.
This bug is only present in this release and thus do not need a backport.
Now we introdce a new .rx_listening() function to report if a receiver is
actually a listening socket. The reason for this is to help detect shared
sockets that might have been broken by sibling processes.
At several places we need to check if a socket is still valid and still
willing to accept connections. Instead of open-coding this, each time,
let's add a new function for this.
This patch adds "coll" new counter and the heartbeat timer values to "show peers"
command. It also adds the elapsed time since the last handshake to new "last_hdshk"
new peer dump field.
Now we have ->suspend() and ->resume() for listeners at the protocol
level. This means that it now becomes possible for a protocol to redefine
its own way to suspend and resume. The default functions are provided for
TCP, UDP and unix, and they are pass-through to the receiver equivalent
as it used to be till now. Nothing was defined for sockpair since it does
not need to suspend/resume during reloads, hence it will succeed.
The inner part now goes into the protocol and is used to decide how to
unbind a given protocol's listener. The existing code which is able to
also unbind the receiver was provided as a default function that we
currently use everywhere. Some complex listeners like QUIC will use this
to decide how to unbind without impacting existing connections, possibly
by setting up other incoming paths for the traffic.
This is used as a generic way to unbind a receiver at the end of
do_unbind_listener(). This allows to considerably simplify that function
since we can now let the protocol perform the cleanup. The generic code
was moved to sock.c, along with the conditional rx_disable() call. Now
the code also supports that the ->disable() function of the protocol
which acts on the listener performs the close itself and adjusts the
RX_F_BUOND flag accordingly.
This listener flag indicates whether the receiver part of the listener
is specific to the master or to the workers. In practice it's only used
by the master's CLI right now. It's used to know whether or not the FD
must be closed before forking the workers. For this reason it's way more
of a receiver's property than a listener's property, so let's move it
there under the name RX_F_MWORKER. The rest of the code remains
unchanged.
And also remove it from its callers. This subtle distinction was added as
sort of a hack for the seamless reload feature but is not needed anymore
since the do_close turned unused since commit previous commit ("MEDIUM:
listener: let do_unbind_listener() decide whether to close or not").
This also removes the unbind_listener_no_close() function.
This will instantly stop all listeners except those which belong to
a proxy configured with a grace time. This means that UDP listeners,
and peers will also be stopped when called this way.
There are multiple ways a proxy may switch to the disabled state,
but now it's essentially once it loses its last listener. Instead
of keeping duplicate code around and reporting the state change
before actually seeing it, we now report it at the moment it's
performed (from the last listener leaving) which allows to remove
the message from all other places.
For now we cannot easily distinguish a peers frontend from another one,
which will be problematic to avoid reporting them when stopping their
listeners. Let's add PR_MODE_PEERS for this. It's not supposed to cause
any issue since all non-HTTP proxies are handled similarly now.
This function will be used to definitely stop a listener (e.g. during a
soft_stop). This is actually tricky because it may be called for a proxy
or for a protocol, both of which require locks and already hold some. The
function takes booleans indicating which ones are already held, hoping
this will be enough. It's not well defined wether proto->disable() and
proto->rx_disable() are supposed to be called with any lock held, and
they are used from do_unbind_listener() with all these locks. Some back
annotations ought to be added on this point.
The proxy's listeners count is updated, and the proxy is marked as
disabled and woken up after the last one is gone. Note that a
listener in listen state is already not attached anymore since it
was disabled.
We'll need an already locked variant of this function so let's make
__delete_listener() which will be called with the protocol lock held
and the listener's lock held.
These methods will be used to enable/disable accepting new connections
so that listeners do not play with FD directly anymore. Since all the
currently supported protocols work on socket for now, these are identical
to the rx_enable/rx_disable functions. However they were not defined in
sock.c since it's likely that some will quickly start to differ. At the
moment they're not used.
We have to take care of fd_updt before calling fd_{want,stop}_recv()
because it's allocated fairly late in the boot process and some such
functions may be called very early (e.g. to stop a disabled frontend's
listeners).
These methods will be used to enable/disable rx at the receiver level so
that callers don't play with FDs directly anymore. All our protocols use
the generic ones from sock.c at the moment. For now they're not used.
These will be used on receivers, to enable or disable receiving on a
listener, which most of the time just consists in enabling/disabling
the file descriptor.
We have to take care of the existence of fd_updt to know if we may
or not call fd_{want,stop}_recv() since it's not permitted in very
early boot.
This one undoes ->rx_suspend(), it tries to restore an operational socket.
It was only implemented for TCP since it's the only one we support right
now.
The ->pause method is inappropriate since it doesn't exactly "pause" a
listener but rather temporarily disables it so that it's not visible at
all to let another process take its place. The term "suspend" is more
suitable, since the "pause" is actually what we'll need to apply to the
FULL and LIMITED states which really need to make a pause in the accept
process. And it goes well with the use of the "resume" function that
will also need to be made per-protocol.
Let's rename the function and make it act on the receiver since it's
already what it essentially does, hence the prefix "_rx" to make it
more explicit.
The protocol struct was a bit reordered because it was becoming a real
mess between the parts related to the listeners and those for the
receivers.
Since the listeners were split into receiver+listener, this field ought
to have been renamed because it's confusing. It really links receivers
and not listeners, as most of the time it's used via rx.proto_list!
The nb_listeners field was updated accordingly.
These ones have never been called, they were referenced by the protocol's
disable_all for some protocols but there are no traces of their use, so
in addition to not being sure the code works, it has never been tested.
Let's remove a bit of complexity starting from there.
These two functions are used to pause and resume all listeners of
all protocols. They use the standard listener functions for this
so they're supposed to handle the situation gracefully regardless
of the upper proxies' states, and they will report completion on
proxies once the switch is performed.
It might be nice to define a particular "failed" state for listeners
that cannot resume and to count them on proxies in order to mention
that they're definitely stuck. On the other hand, the current
situation is retryable which is quite appreciable as well.
The two functions don't need to be distinguished anymore since they have
all the necessary info to act as needed on their listeners. Let's just
pass via stop_proxy() and make it check for each listener which one to
close or not.
Its sole remaining purpose was to display "proxy foo started", which
has little benefit and pollutes output for those with plenty of proxies.
Let's remove it now.
The VTCs were updated to reflect this, because many of them had explicit
counts of dropped lines to match this message.
This is tagged as MEDIUM because some users may be surprized by the
loss of this quite old message.
The remaining proxy states were only used to distinguish an enabled
proxy from a disabled one. Due to the initialization order, both
PR_STNEW and PR_STREADY were equivalent after startup, and they
would only differ from PR_STSTOPPED when the proxy is disabled or
shutdown (which is effectively another way to disable it).
Now we just have a "disabled" field which allows to distinguish them.
It's becoming obvious that start_proxies() is only used to print a
greeting message now, that we'd rather get rid of. Probably that
zombify_proxy() and stop_proxy() should be merged once their
differences move to the right place.
The enabled/disabled config options were stored into a "state" field
that is an integer but contained only PR_STNEW or PR_STSTOPPED, which
is a bit confusing, and causes a dependency with proxies. This was
renamed to "disabled" and is used as a boolean. The field was also
moved to the end of the struct to stop creating a hole and fill another
one.
This state was used to mention that a proxy was in PAUSED state, as opposed
to the READY state. This was causing some trouble because if a listener
failed to resume (e.g. because its port was temporarily in use during the
resume), it was not possible to retry the operation later. Now by checking
the number of READY or PAUSED listeners instead, we can accurately know if
something went bad and try to fix it again later. The case of the temporary
port conflict during resume now works well:
$ socat readline /tmp/sock1
prompt
> disable frontend testme3
> disable frontend testme3
All sockets are already disabled.
> enable frontend testme3
Failed to resume frontend, check logs for precise cause (port conflict?).
> enable frontend testme3
> enable frontend testme3
All sockets are already enabled.
This state is only set when a pause() fails but isn't even set when a
resume() fails. And we cannot recover from this state. Instead, let's
just count remaining ready listeners to decide to emit an error or not.
It's more accurate and will better support new attempts if needed.
Since v1.4 or so, it's almost not possible anymore to set this state. The
only exception is by using the CLI to change a frontend's maxconn setting
below its current usage. This case makes no sense, and for other cases it
doesn't make sense either because "full" is a vague concept when only
certain listeners are full and not all. Let's just remove this unused
state and make it clear that it's not reported. The "ready" or "open"
states will continue to be reported without being misleading as they
will be opposed to "stop".
The proxy state tries to be synthetic but that doesn't work well with
many listeners, especially for transition phases or after a failed
pause/resume.
In order to address this, we'll instead rely on counters of listeners in
a given state for the 3 major states (ready, paused, listen) and a total
counter. We'll now be able to determine a proxy's state by comparing these
counters only.
This function is used as a wrapper to set a listener's state everywhere.
We'll use it later to maintain some counters in a consistent state when
switching state so it's capital that all state changes go through it.
No functional change was made beyond calling the wrapper.
The zombie state is not used anymore by the listeners, because in the
last two cases where it was tested it couldn't match as it was covered
by the test on the process mask. Instead now the FD is either in the
LISTEN state or the INIT state. This also avoids forcing the listener
to be single-dimensional because actually belonging to another process
isn't totally exclusive with the other states, which explains some of
the difficulties requiring to check the proc_mask and the fd sometimes.
So let's get rid of it now not to be tempted to reuse it.
The doc on the listeners state was updated.
Use the new stats module API to integrate the dns counters in the
standard stats. This is done in order to avoid code duplication, keep
the code related to cli out of dns and use the full possibility of the
stats function, allowing to print dns stats in csv or json format.
Add a boolean 'clearable' on stats module structure. If set, it forces
all the counters to be reset on 'clear counters' cli command. If not,
the counters are reset only when 'clear counters all' is used.
This is executed on startup with the registered statistics module. The
existing statistics have been merged in a list containing all
statistics for each domain. This is useful to print all available
statistics in a generic way.
Allocate extra counters for all proxies/servers/listeners instances.
These counters are allocated with the counters from the stats modules
registered on startup.
Implement a small API to easily add extra counters inside a structure
instance. This will be used to implement dynamic statistics linked on
every type of object as needed.
The counters are stored in a dynamic array inside the relevant objects.
A stat module can be registered to quickly add new statistics on
haproxy. It must be attached to one of the available stats domain. The
register must be done using INITCALL on STG_REGISTER.
The stat module has a name which should be unique for each new module in
a domain. It also contains a statistics list with their name/desc and a
pointer to a function used to fill the stats from the module counters.
The module also provides the initial counters values used on
automatically allocated counters. The offset for these counters
are stored in the module structure.
This flag can be used to determine on what type of proxy object the
statistics should be relevant. It will be useful when adding dynamic
statistics. Currently, this flag is not used.
The domain option will be used to have statistics attached to other
objects than proxies/listeners/servers. At the moment, only the PROXY
domain is available.
Add an argument 'domain' on the 'show stats' cli command to specify the
domain. Only 'domain proxy' is available now. If not specified, proxy
will be considered the default domain.
For HTML output, only proxy statistics will be displayed.
Use an opaque pointer to store proxy instance. Regroup server/listener
as a single opaque pointer. This has the benefit to render the structure
more evolutive to support statistics on other types of objects in the
future.
This patch is needed to extend stat support for components other than
proxies objects.
The prometheus module has been adapted for these changes.
Render the stats size parametric in csv/json dump functions. This is
needed for the future patch which provides dynamic stats. For now the
static value ST_F_TOTAL_FIELDS is provided.
Remove unused parameter px on stats_dump_one_line.
This patch is needed to extend stat support to components other than
proxies objects.
Un-mark stats_dump_one_line and stats_putchk as static and export them
in the header file. These functions will be reusable by other components to
print their statistics.
This patch is needed to extend stat support to components other than
proxies objects.
A crash reported in github issue #880 looks impossible unless
pendconn_cond_unlink() occasionally sees a null leaf_p when attempting
to remove an entry, which seems to be confirmed by the reporter. What
seems to be happening is that depending on compiler optimizations,
this pointer can appear as null while pointers are moved if one of
the node's parents is removed from or inserted into the tree. There's
no explicit null of the pointer during these operations but those
pointers are rewritten in multiple steps and nothing prevents this
situation from happening, and there are no particular barrier nor
atomic ops around this.
This test was used to avoid unnecessary locking, for already deleted
entries, but looking at the code it appears that pendconn_free() already
resets s->pend_pos that's used as <p> there, and that the other call
reasons are after an error where the connection will be dropped as
well. So we don't save anything by doing this test, and make it
unsafe. The older code used to check for list emptiness there and
not inside pendconn_unlink(), which explains why the code has stayed
there. Let's just remove this now.
Thanks to @jaroslawr for reporting this issue in great details and for
testing the proposed fix.
This should be backpored to 1.8, where the test on LIST_ISEMPTY should
be moved to pendconn_unlink() instead (inside the lock, just like 2.0+).
Allow the syntax "${...[*]}" to expand an environment variable
containing several values separated by spaces as individual arguments. A
new flag PARSE_OPT_WORD_EXPAND has been added to toggle this feature on
parse_line invocation. In case of an invalid syntax, a new error
PARSE_ERR_WRONG_EXPAND will be triggered.
This feature has been asked on the github issue #165.
As part of his GREASE experiments on Chromium, Bence Bky reported in
https://lists.w3.org/Archives/Public/ietf-http-wg/2020JulSep/0202.html
and https://bugs.chromium.org/p/chromium/issues/detail?id=1127060 that
a certain combination of frame type and frame flags was causing an error
on app.slack.com. It turns out that it's haproxy that is causing this
issue because the frame type is wrongly assumed to support padding, the
frame flags indicate padding is present, and the frame is too short for
this, resulting in an error.
The reason why only some frame types are affected is due to the frame
type being used in a bit shift to match against a mask, and where the
5 lower bits of the frame type only are used to compute the frame bit.
If the resulting frame bit matches a DATA, HEADERS or PUSH_PROMISE frame
bit, then padding support is assumed and the test is enforced, resulting
in a PROTOCOL_ERROR or FRAME_SIZE_ERROR depending on the payload size.
We must never match any such bit for unsupported frame types so let's
add a check for this. This must be backported as far as 1.8.
Thanks to Cooper Bethea for providing enough context to help narrow the
issue down and to Bence Bky for creating a simple reproducer.
We don't need to cheat with the sock_domain anymore, we now always have
the SOCK_DGRAM sock_type as a complementary selector. This patch restores
the sock_domain to AF_INET* in the udp* protocols and removes all traces
of the now unused AF_CUST_*.
The protocol array used to be only indexed by socket family, which is very
problematic with UDP (requiring an extra family) and with the forthcoming
QUIC (also requiring an extra family), especially since that binds them to
certain families, prevents them from supporting dgram UNIX sockets etc.
In order to address this, we now start to register the protocols with more
info, namely the socket type and the control type (either stream or dgram).
This is sufficient for the protocols we have to deal with, but could also
be extended further if multiple protocol variants were needed. But as is,
it still fits nicely in an array, which is convenient for lookups that are
instant.
This one will be needed to more accurately select a protocol. It may
differ from the socket type for QUIC, which uses dgram at the socket
layer and provides stream at the control layer. The upper level requests
a control layer only so we need this field.
Most callers of str2sa_range() need the protocol only to check that it
provides a ->connect() method. It used to be used to verify that it's a
stream protocol, but it might be a bit early to get rid of it. Let's keep
the test for now but move it to str2sa_range() when the new flag PA_O_CONNECT
is present. This way almost all call places could be cleaned from this.
There's a strange test in the server address parsing code that rechecks
the family from the socket which seems to be a duplicate of the previously
removed tests. It will have to be rechecked.
We'll need this so that it can return pointers to stacked protocol in
the future (for QUIC). In addition this removes a lot of tests for
protocol validity in the callers.
Some of them were checked further apart, or after a call to
str2listener() and they were simplified as well.
There's still a trick, we can fail to return a protocol in case the caller
accepts an fqdn for use later. This is what servers do and in this case it
is valid to return no protocol. A typical example is:
server foo localhost:1111
The function will need to use more than just a family, let's pass it
the selected protocol. The caller will then be able to do all the fancy
stuff required to pick the best protocol.
This is at least temporary, as the migration at once is way too difficuly.
For now it still creates listeners but only allows DGRAM sockets. This
aims at easing the split between listeners and receivers.
If a file descriptor was passed, we can optionally return it. This will
be useful for listening sockets which are both a pre-bound FD and a ready
socket.
These flags indicate whether the call is made to fill a bind or a server
line, or even just send/recv calls (like logs or dns). Some special cases
are made for outgoing FDs (e.g. pipes for logs) or socket FDs (e.g external
listeners), and there's a distinction between stream or dgram usage that's
expected to significantly help str2sa_range() proceed appropriately with
the input information. For now they are not used yet.
These flags indicate what is expected regarding port specifications. Some
callers accept none, some need fixed ports, some have it mandatory, some
support ranges, and some take an offset. Each possibilty is reflected by
an option. For now they are not exploited, but the goal is to instrument
str2sa_range() to properly parse that.
We currently have an argument to require that the address is resolved
but we'll soon add more, so let's turn it into a bit field. The old
"resolve" boolean is now PA_O_RESOLVE.
At some places (log fd@XXX, bind fd@XXX) we support using an explicit
file descriptor number, that is placed into the sockaddr for later use.
The problem is that till now it was done with an AF_UNSPEC family, which
is also used for other situations like missing info or rings (for logs).
Let's create an "official" family AF_CUST_EXISTING_FD for this case so
that we are certain the FD can be found in the address when it is set.
This removes the following fields from struct protocol that are now
retrieved from the protocol family instead: .sock_family, .sock_addrlen,
.l3_addrlen, .addrcmp, .bind, .get_src, .get_dst.
This also removes the UDP-specific udp{,6}_get_{src,dst}() functions
which were referenced but not used yet. Their goal was only to remap
the original AF_INET* addresses to AF_CUST_UDP*.
Note that .sock_domain is still there as it's used as a selector for
the protocol struct to be used.
We now take care of retrieving sock_family, l3_addrlen, bind(),
addrcmp(), get_src() and get_dst() from the protocol family and
not just the protocol itself. There are very few places, this was
only seldom used. Interestingly in sock_inet.c used to rely on
->sock_family instead of ->sock_domain, and sock_unix.c used to
hard-code PF_UNIX instead of using ->sock_domain.
Also it appears obvious we have something wrong it the protocol
selection algorithm because sock_domain is the one set to the custom
protocols while it ought to be sock_family instead, which would avoid
having to hard-code some conversions for UDP namely.
We need to specially handle protocol families which regroup common
functions used for a given address family. These functions include
bind(), addrcmp(), get_src() and get_dst() for now. Some fields are
also added about the address family, socket domain (protocol family
passed to the socket() syscall), and address length.
These protocol families are referenced from the protocols but not yet
used.
Note that for now we don't have a sockpair.c file to host that unusual
family, so the new function was placed directly into proto_sockpair.c.
It's no big deal given that this family is currently not shared with
multiple protocols.
The function does almost nothing but setting up the receiver. This is
normal as the socket the FDs are passed onto are supposed to have been
already created somewhere else, and the only usable identifier for such
a socket pair is the receiving FD itself.
The function was assigned to sockpair's ->bind() and is not used yet.
This function performs all the bind-related stuff for UNIX sockets that
was previously done in uxst_bind_listener(). There is a very tiny
difference however, which is that previously, in the unlikely event
where listen() would fail, it was still possible to roll back the binding
and rename the backup to the original socket. Now we have to rename it
before calling returning, hence it will be done before calling listen().
However, this doesn't cover any particular use case since listen() has no
reason to fail there (and the rollback is not done for inherited sockets),
that was just done that way as a generic error processing path.
The code is not used yet and is referenced in the uxst proto's ->bind().
This function collects all the receiver-specific code from both
tcp_bind_listener() and udp_bind_listener() in order to provide a more
generic AF_INET/AF_INET6 socket binding function. For now the API is
not very elegant because some info are still missing from the receiver
while there's no ideal place to fill them except when calling ->listen()
at the protocol level. It looks like some polishing code is needed in
check_config_validity() or somewhere around this in order to finalize
the receivers' setup. The main issue is that listeners and receivers
are created *before* bind_conf options are parsed and that there's no
finishing step to resolve some of them.
The function currently sets up a receiver and subscribes it to the
poller. In an ideal world we wouldn't subscribe it but let the caller
do it after having finished to configure the L4 stuff. The problem is
that the caller would then need to perform an fd_insert() call and to
possibly set the exported flag on the FD while it's not its job. Maybe
an improvement could be to have a separate sock_start_receiver() call
in sock.c.
For now the function is not used but it will soon be. It's already
referenced as tcp and udp's ->bind().
This will be the function that must be used to bind the receiver. It
solely depends on the address family but for now it's simpler to have
it per protocol.
The new RX_O_FOREIGN, RX_O_V6ONLY and RX_O_V4V6 options are now set into
the rx_settings part during the parsing, so that we don't need to adjust
them in each and every listener anymore. We have to keep both v4v6 and
v6only due to the precedence from v6only over v4v6.
It's the receiver's FD that's inherited from the parent process, not
the listener's so the flag must move to the receiver so that appropriate
actions can be taken.
In order to split the receiver from the listener, we'll need to know that
a socket is already bound and ready to receive. We used to do that via
tha LI_O_ASSIGNED state but that's not sufficient anymore since the
receiver might not belong to a listener anymore. The new RX_F_BOUND flag
is used for this.
A receiver will have to pass a context to be installed into the fdtab
for use by the handler. We need to set this into the receiver struct
as the bind will happen longer after the configuration.
Just like listeners keep a pointer to their bind_conf, receivers now also
have a pointer to their rx_settings. All those belonging to a listener are
automatically initialized with a pointer to the bind_conf's settings.
We'll soon add flags for the receivers, better add them to the final
file, so it's time to move the definition to receiver-t.h. The struct
receiver and rx_settings were placed there.
The receiver is the one which depends on the protocol while the listener
relies on the receiver. Let's move the protocol there. Since there's also
a list element to get back to the listener from the proto list, this list
element (proto_list) was moved as well. For now when scanning protos, we
still see listeners which are linked by their rx.proto_list part.
The listening socket is represented by its file descriptor, which is
generic to all receivers and not just listeners, so it must move to
the rx struct.
It's worth noting that in order to extend receivers and listeners to
other protocols such as QUIC, we'll need other handles than file
descriptors here, and that either a union or a cast to uintptr_t
will have to be used. This was not done yet and the field was
preserved under the name "fd" to avoid adding confusion.
In order to start to split the listeners into the listener part and the
event receiver part, we introduce a new field "rx" into struct listener
that will eventually become a separate struct receiver. This patch only
adds the struct with an options field that the receivers will need.
The netns is common to all listeners/receivers and is used to bind the
listening socket so it must be in the receiver settings and not in the
listener. This removes some yet another set of unnecessary loops.
The interface is common to all listeners/receivers and is used to bind
the listening socket so it must be in the receiver settings and not in
the listener. This removes some unnecessary loops.
There currently is a large inconsistency in how binding parameters are
split between bind_conf and listeners. It happens that for historical
reasons some parameters are available at the listener level but cannot
be configured per-listener but only for a bind_conf, and thus, need to
be replicated. In addition, some of the bind_conf parameters are in fact
for the listening socket itself while others are for the instanciated
sockets.
A previous attempt at splitting listeners into receivers failed because
the boundary between all these settings is not well defined.
This patch introduces a level of listening socket settings in the
bind_conf, that will be detachable later. Such settings that are solely
for the listening socket are:
- unix socket permissions (used only during binding)
- interface (used for binding)
- network namespace (used for binding)
- process mask and thread mask (used during startup)
The rest seems to be used only to initialize the resulting sockets, or
to control the accept rate. For now, only the unix params (bind_conf->ux)
were moved there.
Remove the last utility functions for handling the multi-cert bundles
and remove the multi-variable from the ckch structure.
With this patch, the bundles are completely removed.
Since the removal of the multi-certificates bundle support, this
variable is not useful anymore, we can remove all tests for this
variable and suppose that every ckch contains a single certificate.
Commit 4987a4744 ("CLEANUP: tree-wide: use VAR_ARRAY instead of [0] in
various definitions") broke the build on clang due to the tlv field used
to receive/send the proxy protocol. The problem is that struct tlv is
included at the beginning of struct tlv_ssl, which doesn't make much
sense. In fact the value[] array isn't really a var array but just an
end of struct marker, and must really be an array of size zero.
Surprisingly there were still a number of [0] definitions for variable
sized arrays in certain structures all over the code. We need to use
VAR_ARRAY instead of zero to accommodate various compilers' preferences,
as zero was used only on old ones and tends to report errors on new ones.
tcc supports variadic macros provided that there is always at least one
argument, like older gcc versions. Thus we need to always keep one and
define args as the remaining ones. It's not an issue at all and doesn't
change the way to use them, just the internal definitions.
For whatever reason, glibc decided that the __attribute__ keyword is
the exclusive property of gcc, and redefines it to an empty macro on
other compilers. Some non-gcc compilers also support it (possibly
partially), tinycc is one of them. By doing this, glibc silently
broke all constructors, resulting in code that arrives in main() with
uninitialized variables.
The solution we use here consists in undefining the macro on non-gcc
compilers, and redefining it to itself in order to cause a conflict in
the event the redefinition would happen afterwards. This visibly solved
the problem.
Some checks on __GNUC__ imply that if it's undefined it will match a
low value but that's not always what we want, like for example in the
VAR_ARRAY definition which is not needed on tcc. Let's always be explicit
on these tests.
A SRV record keeps a reference on the corresponding additional record, if
any. But this additional record is also inserted in a separate linked-list into
the dns response. The problems arise when obsolete additional records are
released. The additional records list is purged but the SRV records always
reference these objects, leading to an undefined behavior. Worst, this happens
very quickly because additional records are never renewed. Thus, once received,
an additional record will always expire.
Now, the addtional record are only associated to a SRV record or simply
ignored. And the last version is always used.
This patch helps to fix the issue #841. It must be backported to 2.2.
Ever since the protocols were added in 1.3.13, listeners used to be
started twice:
- once by start_proxies(), which iteratees over all proxies then all
listeners ;
- once by protocol_bind_all() which iterates over all protocols then
all listeners ;
It's a real mess because error reporting is not even consistent, and
more importantly now that some protocols do not appear in regular
proxies (peers, logs), there is no way to retry their binding should
it fail on the last step.
What this patch does is to make sure that listeners are exclusively
started by protocols. The failure to start a listener now causes the
emission of an error indicating the proxy's name (as it used to be
the case per proxy), and retryable failures are silently ignored
during all but last attempts.
The start_proxies() function was kept solely for setting the proxy's
state to READY and emitting the "Proxy started" message and log that
some have likely got used to seeking in their logs.
When calling the http_replace_res_status() function, an optional reason may now
be set. It is ignored if it points to NULL and the original reason is
preserved. Only the response status is replaced. Otherwise both the status and
the reason are replaced.
It simplifies the API and most of time, avoids an extra call to
http_replace_res_reason().
The http_replace_req_path() function now takes a third argument to evaluate the
query-string as part of the path or to preserve it. If <with_qs> is set, the
query-string is replaced with the path. Otherwise, only the path is replaced.
This patch is mandatory to fix issue #829. The next commit depends on it. So be
carefull during backports.
For now we still don't retrieve dgram sockets, but the code must be able
to distinguish them before we switch to receivers. This adds a new flag
to the xfer_sock_list indicating that a socket is of type SOCK_DGRAM. The
way to set the flag for now is by looking at the dummy address family which
equals AF_CUST_UDP{4,6} in this case (given that other dgram sockets are not
yet supported).
We'll want to store more info there and some info that are not represented
in listener options at the moment (such as dgram vs stream) so let's get
rid of these and instead use a new set of options (SOCK_XFER_OPT_*).
The new function was called sock_get_old_sockets() and was left as-is
except a minimum amount of style lifting to make it more readable. It
will never be awesome anyway since it's used very early in the boot
sequence and needs to perform socket I/O without any external help.
This code was highly redundant, existing for TCP clients, TCP servers
and UDP servers. Let's move it to sock_inet where it belongs. The new
functions are sock_inet4_make_foreign() and sock_inet6_make_foreign().
This is essentially a merge from tcp_find_compatible_fd() and
uxst_find_compatible_fd() that relies on a listener's address and
compare function and still checks for other variations. For AF_INET6
it compares a few of the listener's bind options. A minor change for
UNIX sockets is that transparent mode, interface and namespace used
to be ignored when trying to pick a previous socket while now if they
are changed, the socket will not be reused. This could be refined but
it's still better this way as there is no more risk of using a
differently bound socket by accident.
Eventually we should not pass a listener there but a set of binding
parameters (address, interface, namespace etc...) which ultimately will
be grouped into a receiver. For now this still doesn't exist so let's
stick to the listener to break dependencies in the rest of the code.
Let's determine it at boot time instead of doing it on first use. It
also saves us from having to keep it thread local. It's been moved to
the new sock_inet_prepare() function, and the variables were renamed
to sock_inet_tcp_maxseg_default and sock_inet6_tcp_maxseg_default.
The v6only_default variable is not specific to TCP but to AF_INET6, so
let's move it to the right file. It's now immediately filled on startup
during the PREPARE stage so that it doesn't have to be tested each time.
The variable's name was changed to sock_inet6_v6only_default.
The function now makes it clear that it's independent on the socket
type and solely relies on the address family. Note that it supports
both IPv4 and IPv6 as we don't seem to need it per-family.
This one is common to the TCPv4 and UDPv4 code, it retrieves the
destination address of a socket, taking care of the possiblity that for
an incoming connection the traffic was possibly redirected. The TCP and
UDP definitions were updated to rely on it and remove duplicated code.
The new addrcmp() protocol member points to the function to be used to
compare two addresses of the same family.
When picking an FD from a previous process, we can now use the address
specific address comparison functions instead of having to rely on a
local implementation. This will help move that code to a more central
place.
These files will regroup everything specific to AF_INET, AF_INET6 and
AF_UNIX socket definitions and address management. Some code there might
be agnostic to the socket type and could later move to af_xxxx.c but for
now we only support regular sockets so no need to go too far.
The files are quite poor at this step, they only contain the address
comparison function for each address family.
The new file sock.c will contain generic code for standard sockets
relying on file descriptors. We currently have way too much duplication
between proto_uxst, proto_tcp, proto_sockpair and proto_udp.
For now only get_src, get_dst and sock_create_server_socket were moved,
and are used where appropriate.
This is totally ugly, smp_fetch_src() is exported only so that stick_table.c
can (ab)use it in the {sc,src}_* sample fetch functions. It could be argued
that the sample could have been reconstructed there in place, but we don't
even need to duplicate the code. We'd rather simply retrieve the "src"
fetch's function from where it's used at init time and be done with it.
This new flag will be used to mark FDs that must be passed to any future
process across the CLI's "_getsocks" command.
The scheme here is quite complex and full of special cases:
- FDs inherited from parent processes are *not* exported this way, as
they are supposed to instead be passed by the master process itself
across reloads. However such FDs ought never to be paused otherwise
this would disrupt the socket in the parent process as well;
- FDs resulting from a "bind" performed over a socket pair, which are
in fact one side of a socket pair passed inside another control socket
pair must not be passed either. Since all of them are used the same
way, for now it's enough never to put this "exported" flag to FDs
bound by the socketpair code.
- FDs belonging to temporary listeners (e.g. a passive FTP data port)
must not be passed either. Fortunately we don't have such FDs yet.
- the rest of the listeners for now are made of TCP, UNIX stream, ABNS
sockets and are exportable, so they get the flag.
- UDP listeners were wrongly created as listeners and are not suitable
here. Their FDs should be passed but for now they are not since the
client doesn't even distinguish the SO_TYPE of the retrieved sockets.
In addition, it's important to keep in mind that:
- inherited FDs may never be closed in master process but may be closed
in worker processes if the service is shut down (useless since still
bound, but technically possible) ;
- inherited FDs may not be disabled ;
- exported FDs may be disabled because the caller will perform the
subsequent listen() on them. However that might not work for all OSes
- exported FDs may be closed, it just means the service was shut down
from the worker, and will be rebound in the new process. This implies
that we have to disable exported on close().
=> as such, contrary to an apparently obvious equivalence, the "exported"
status doesn't imply anything regarding the ability to close a
listener's FD or not.
This essentially undoes what we did in fd.c in 1.8 to support seamless
reload. Since we don't need to remove an fd anymore we can turn
fd_delete() to the simple function it used to be.
Let's not look at the listener options passed by the original process
and determine from the socket itself whether it is configured for
transparent mode or not. This is cleaner and safer, and doesn't rely
on flag values that could possibly change between versions.
haproxy supports generating SSL certificates based on SNI using a provided
CA signing certificate. Because CA certificates may be signed by multiple
CAs, in some scenarios, it is neccesary for the server to attach the trust chain
in addition to the generated certificate.
The following patch adds the ability to serve the entire trust chain with
the generated certificate. The chain is loaded from the provided
`ca-sign-file` PEM file.
When a regex had been succesfully compiled by the JIT pass, it is better
to use the related match, thanksfully having same signature, for better
performance.
Signed-off-by: David Carlier <devnexen@gmail.com>
The ARGT_PTR argument type may now be used to keep a reference to opaque data in
the argument array used by sample fetches and converters. It is a generic way to
point on data. I guess it could be used for some other arguments, like proxy,
server, map or stick-table.
A dedicated expiration date is now used to apply the inspect-delay of the
tcp-request or tcp-response rulesets. Before, the analyse expiratation date was
used but it may also be updated by the lua (at least). So a lua script may
extend or reduce the inspect-delay by side effect. This is not expected. If it
becomes necessary, a specific function will be added to do this. Because, for
now, it is a bit confusing.
The HTX_FL_EOI flag must now be set on a HTX message when no more data are
expected. Most of time, it must be set before adding the EOM block. Thus, if
there is no space for the EOM, there is still an information to know all data
were received and pushed in the HTX message. There is only an exception for the
HTTP replies (deny, return...). For these messages, the flag is set after all
blocks are pushed in the message, including the EOM block, because, on error,
we remove all inserted data.
__task_free() cannot be called with a task still in the queue. This
test adds a check which confirms there is no concurrency issue on such
a case where a thread could requeue nor wakeup a task being freed.
This aims at catching calls to task_unlink_wq() performed by the wrong
thread based on the shared status for the task, as well as calls to
__task_queue() with the wrong timer queue being used based on the task's
capabilities. This will at least help eliminate some hypothesis during
debugging sessions when suspecting that a wrong thread has attempted to
queue a task at the wrong place.
The BUG_ON() test in task_queue() only tests for the case where
we're queuing a task that doesn't run on the current thread. Let's
refine it a bit further to catch all cases where the task does not
run *exactly* on the current thread alone.
Reported github issue #759 shows there is no name resolving
on server lines for ring and peers sections.
This patch introduce the resolving for those lines.
This patch adds boolean a parameter to parse_server function to specify
if we want the function to perform an initial name resolving using libc.
This boolean is forced to true in case of peers or ring section.
The boolean is kept to false in case of classic servers (from
backend/listen)
This patch should be backported in branches where peers sections
support 'server' lines.
This patch adds a global counter of received syslog messages
and this one is exported on CLI "show info" as "CumRecvLogs".
This patch also updates internal conn counter and freq
of the listener and the proxy for each received log message to
prepare a further export on the "show stats".
Log forwarding:
It is possible to declare one or multiple log forwarding section,
haproxy will forward all received log messages to a log servers list.
log-forward <name>
Creates a new log forwarder proxy identified as <name>.
bind <addr> [param*]
Used to configure a log udp listener to receive messages to forward.
Only udp listeners are allowed, address must be prefixed using
'udp@', 'udp4@' or 'udp6@'. This supports for all "bind" parameters
found in 5.1 paragraph but most of them are irrelevant for udp/syslog case.
log global
log <address> [len <length>] [format <format>] [sample <ranges>:<smp_size>]
<facility> [<level> [<minlevel>]]
Used to configure target log servers. See more details on proxies
documentation.
If no format specified, haproxy tries to keep the incoming log format.
Configured facility is ignored, except if incoming message does not
present a facility but one is mandatory on the outgoing format.
If there is no timestamp available in the input format, but the field
exists in output format, haproxy will use the local date.
Example:
global
log stderr format iso local7
ring myring
description "My local buffer"
format rfc5424
maxlen 1200
size 32764
timeout connect 5s
timeout server 10s
# syslog tcp server
server mysyslogsrv 127.0.0.1:514 log-proto octet-count
log-forward sylog-loadb
bind udp4@127.0.0.1:1514
# all messages on stderr
log global
# all messages on local tcp syslog server
log ring@myring local0
# load balance messages on 4 udp syslog servers
log 127.0.0.1:10001 sample 1:4 local0
log 127.0.0.1:10002 sample 2:4 local0
log 127.0.0.1:10003 sample 3:4 local0
log 127.0.0.1:10004 sample 4:4 local0
This patch introduce a new fd handler used to parse syslog
message on udp.
The parsing function returns level, facility and metadata that
can be immediatly reused to forward message to a log server.
This handler is enabled on udp listeners if proxy is internally set
to mode PR_MODE_SYSLOG
This patch merges build message code between sink and log
and introduce a new API based on struct ist array to
prepare message header with zero copy, targeting the
log forwarding feature.
Log format 'iso' and 'timed' are now avalaible on logs line.
A new log format 'priority' is also added.
This patch introduce proto_udp.c targeting a further support of
log forwarding feature.
This code was originally produced by Frederic Lecaille working on
QUIC support and only minimal requirements for syslog support
have been merged.
In the continuity of the commit 7cf0e4517 ("MINOR: raw_sock: report global
traffic statistics"), we are now able to report the global number of bytes
emitted using the splicing. It can be retrieved in "show info" output on the
CLI.
Note this counter is always declared, regardless the splicing support. This
eases the integration with monitoring tools plugged on the CLI.
The srv_del_conn_from_list() function is now responsible to update the server
counters and the connection flags when a connection is removed from an idle list
(safe, idle or available). It is called when a connection is released or when a
connection is set as private. This function also removes the connection from the
idle list if necessary.
The srv_use_idle_conn() function is now responsible to update the server
counters and the connection flags when an idle connection is reused. The same
function is called when a new connection is created. This simplifies a bit the
connect_server() function.
When a new connection is created, its target is always set just after. So the
connection target may set when it is created instead, during its initialisation
to be precise. It is the purpose of this patch. Now, conn_new() function is
called with the connection target as parameter. The target is then passed to
conn_init(). It means the target must be passed when cs_new() is called. In this
case, the target is only used when the conn-stream is created with no
connection. This only happens for tcpchecks for now.
The session_get_conn() must now be used to look for an available connection
matching a specific target for a given session. This simplifies a bit the
connect_server() function.
When a connection is marked as private, it is now added in the session server
list. We don't wait a stream is detached from the mux to do so. When the
connection is created, this happens after the mux creation. Otherwise, it is
performed when the connection is marked as private.
To allow that, when a connection is created, the session is systematically set
as the connectin owner. Thus, a backend connection has always a owner during its
creation. And a private connection has always a owner until its death.
Note that outside the detach() callback, if the call to session_add_conn()
failed, the error is ignored. In this situation, we retry to add the connection
into the session server list in the detach() callback. If this fails at this
step, the multiplexer is destroyed and the connection is closed.
To set a connection as private, the conn_set_private() function must now be
called. It sets the CO_FL_PRIVATE flags, but it also remove the connection from
the available connection list, if necessary. For now, it never happens because
only HTTP/1 connections may be set as private after their creation. And these
connections are never inserted in the available connection list.
When a connection is added to an idle list, it's already detached and
cannot be seen by two threads at once, so there's no point using
TRY_ADDQ, there will never be any conflict. Let's just use the cheaper
ADDQ.
The TRY_ADDQ there was not needed since the wait list is exclusively
owned by the caller. There's a preliminary test on MT_LIST_ADDED()
that might have been eliminated by keeping MT_LIST_TRY_ADDQ() but
it would have required two more expensive writes before testing so
better keep the test the way it is.
Initially when mt_lists were added, their purpose was to be used with
the scheduler, where anyone may concurrently add the same tasklet, so
it sounded natural to implement a check in MT_LIST_ADD{,Q}. Later their
usage was extended and MT_LIST_ADD{,Q} started to be used on situations
where the element to be added was exclusively owned by the one performing
the operation so a conflict was impossible. This became more obvious with
the idle connections and the new macro was called MT_LIST_ADDQ_NOCHECK.
But this remains confusing and at many places it's not expected that
an MT_LIST_ADD could possibly fail, and worse, at some places we start
by initializing it before adding (and the test is superflous) so let's
rename them to something more conventional to denote the presence of the
check or not:
MT_LIST_ADD{,Q} : inconditional operation, the caller owns the
element, and doesn't care about the element's
current state (exactly like LIST_ADD)
MT_LIST_TRY_ADD{,Q}: only perform the operation if the element is not
already added or in the process of being added.
This means that the previously "safe" MT_LIST_ADD{,Q} are not "safe"
anymore. This also means that in case of backport mistakes in the
future causing this to be overlooked, the slower and safer functions
will still be used by default.
Note that the missing unchecked MT_LIST_ADD macro was added.
The rest of the code will have to be reviewed so that a number of
callers of MT_LIST_TRY_ADDQ are changed to MT_LIST_ADDQ to remove
the unneeded test.
It is now possible to customize TCP keepalive parameters.
These correspond to the socket options TCP_KEEPCNT, TCP_KEEPIDLE, TCP_KEEPINTVL
and are valid for the defaults, listen, frontend and backend sections.
This patch fixes GitHub issue #670.
The torture test run for previous commit 787dc20 ("BUG/MEDIUM: lists: add
missing store barrier on MT_LIST_BEHEAD()") finally broke again after 34M
connections. It appeared that MT_LIST_ADD and MT_LIST_ADDQ were suffering
from the same missing barrier when restoring the original pointers before
giving up, when checking if the element was already added. This is indeed
something which seldom happens with the shared scheduler, in case two
threads simultaneously try to wake up the same tasklet.
With a store barrier there after reverting the pointers, the torture test
survived 750M connections on the NanoPI-Fire3, so it looks good this time.
Probably that MT_LIST_BEHEAD should be added to test-list.c since it seems
to be more sensitive to concurrent accesses with MT_LIST_ADDQ.
It's worth noting that there is no barrier between the last two pointers
update, while there is one in MT_LIST_POP and MT_LIST_BEHEAD, the latter
having shown to be needed, but I cannot demonstrate why we would need
one here. Given that the code seems solid here, let's stick to what is
shown to work.
This fix should be backported to 2.1, just for the sake of safety since
the issue couldn't be triggered there, but it could change with the
compiler or when backporting a fix for example.
When running multi-threaded tests on my NanoPI-Fire3 (8 A53 cores), I
managed to occasionally get either a bus error or a segfault in the
scheduler, but only when running at a high connection rate (injecting
on a tcp-request connection reject rule). The bug is rare and happens
around once per million connections. I could never reproduce it with
less than 4 threads nor on A72 cores.
Haproxy 2.1.0 would also fail there but not 2.1.7.
Every time the crash happened with the TL_URGENT task list corrupted,
though it was not immediately after the LIST_SPLICE() call, indicating
background activity survived the MT_LIST_BEHEAD() operation. This queue
is where the shared runqueue is transferred, and the shared runqueue
gets fast inter-thread tasklet wakeups from idle conn takeover and new
connections.
Comparing the MT_LIST_BEHEAD() and MT_LIST_DEL() implementations, it's
quite obvious that a few barriers are missing from the former, and these
will simply fail on weakly ordered caches.
Two store barriers were added before the break() on failure, to match
what is done on the normal path. Missing them almost always results in
a segfault which is quite rare but consistent (after ~3M connections).
The 3rd one before updating n->prev seems intuitively needed though I
coudln't make the code fail without it. It's present in MT_LIST_DEL so
better not be needlessly creative. The last one is the most important
one, and seems to be the one that helps a concurrent MT_LIST_ADDQ()
detect a late failure and try again. With this, the code survives at
least 30M connections.
Interestingly the exact same issue was addressed in 2.0-dev2 for MT_LIST_DEL
with commit 690d2ad4d ("BUG/MEDIUM: list: add missing store barriers when
updating elements and head").
This fix must be backported to 2.1 as MT_LIST_BEHEAD() is also used there.
It's only tagged as medium because it will only affect entry-level CPUs
like Cortex A53 (x86 are not affected), and requires load levels that are
very hard to achieve on such machines to trigger it. In practice it's
unlikely anyone will ever hit it.
OpenSSL 1.1.1 provides a callback registering function
SSL_CTX_set_keylog_callback, which allows one to receive a string
containing the keys to deciphers TLSv1.3.
Unfortunately it is not possible to store this data in binary form and
we can only get this information using the callback. Which means that we
need to store it until the connection is closed.
This patches add 2 pools, the first one, pool_head_ssl_keylog is used to
store a struct ssl_keylog which will be inserted as a ex_data in a SSL *.
The second one is pool_head_ssl_keylog_str which will be used to store
the hexadecimal strings.
To enable the capture of the keys, you need to set "tune.ssl.keylog on"
in your configuration.
The following fetches were implemented:
ssl_fc_client_early_traffic_secret,
ssl_fc_client_handshake_traffic_secret,
ssl_fc_server_handshake_traffic_secret,
ssl_fc_client_traffic_secret_0,
ssl_fc_server_traffic_secret_0,
ssl_fc_exporter_secret,
ssl_fc_early_exporter_secret
Originally it was made to return a void* because some comparisons in the
code where it was used required a lot of casts. But now we don't need
that anymore. And having it non-const breaks the build on NetBSD 9 as
reported in issue #728.
So let's switch to const and adjust debug.c to accomodate this.
When we takeover a connection, let the xprt layer know. If it has its own
tasklet, and it is already scheduled, then it has to be destroyed, otherwise
it may run the new mux tasklet on the old thread.
Note that we only do this for the ssl xprt for now, because the only other
one that might wake the mux up is the handshake one, which is supposed to
disappear before idle connections exist.
No backport is needed, this is for 2.2.
In the various takeover() methods, make sure we schedule the old tasklet
on the old thread, as we don't want it to run on our own thread! This
was causing a very rare crash when building with DEBUG_STRICT, seeing
that either an FD's thread mask didn't match the thread ID in h1_io_cb(),
or that stream_int_notify() would try to queue a task with the wrong
tid_bit.
In order to reproduce this, it is necessary to maintain many connections
(typically 30k) at a high request rate flowing over H1+SSL between two
proxies, the second of which would randomly reject ~1% of the incoming
connection and randomly killing some idle ones using a very short client
timeout. The request rate must be adjusted so that the CPUs are nearly
saturated, but never reach 100%. It's easier to reproduce this by skipping
local connections and always picking from other threads. The issue
should happen in less than 20s otherwise it's necessary to restart to
reset the idle connections lists.
No backport is needed, takeover() is 2.2 only.
tasklet_wakeup() only checks tl->tid to know whether the task is
programmed to run on the current thread or on a specific thread. We'll
have to ease this selection in a subsequent patch, preferably without
modifying tl->tid, so let's have a new tasklet_wakeup_on() function
to specify the thread number to run on. That the logic has not changed
at all.
Now it's possible to preserve spacing everywhere except in "log-format",
"log-format-sd" and "unique-id-format" directives, where spaces are
delimiters and are merged. That may be useful when the response payload
is specified as a log format string by "lf-file" or "lf-string", or even
for headers or anything else.
In order to merge spaces, a new option LOG_OPT_MERGE_SPACES is applied
exclusively on options passed to function parse_logformat_string().
This patch fixes an issue #701 ("http-request return log-format file
evaluation altering spacing of ASCII output/art").
Now when building with -DDEBUG_MEM_STATS, some malloc/calloc/strdup/realloc
stats are kept per file+line number and may be displayed and even reset on
the CLI using "debug dev memstats". This allows to easily track potential
leakers or abnormal usages.
Enables ('on') or disables ('off') sharing of idle connection pools between
threads for a same server. The default is to share them between threads in
order to minimize the number of persistent connections to a server, and to
optimize the connection reuse rate. But to help with debugging or when
suspecting a bug in HAProxy around connection reuse, it can be convenient to
forcefully disable this idle pool sharing between multiple threads, and force
this option to "off". The default is on.
This could have been nice to have during the idle connections debugging,
but it's not too late to add it!
Add two new macros, MT_LIST_DEL_SAFE_NOINIT makes sure we remove the
element from the list, without reinitializing its next and prev, and
MT_LIST_ADDQ_NOCHECK is similar to MT_LIST_ADDQ(), except it doesn't check
if the element is already in a list.
The goal is to be able to move an element from a list we're currently
parsing to another, keeping it locked in the meanwhile.
task_kill() may be used by any thread to kill any task with less overhead
than a regular wakeup. In order to achieve this, it bypasses the priority
tree and inserts the task directly into the shared tasklets list, cast as
a tasklet. The task_list_size is updated to make sure it is properly
decremented after execution of this task. The task will thus be picked by
process_runnable_tasks() after checking the tree and sent to the TL_URGENT
list, where it will be processed and killed.
If the task is bound to more than one thread, its first thread will be the
one notified.
If the task was already queued or running, nothing is done, only the flag
is added so that it gets killed before or after execution. Of course it's
the caller's responsibility to make sur any resources allocated by this
task were already cleaned up or taken over.
This flag, when set, will be used to indicate that the task must die.
At the moment this may only be placed by the task itself or by the
scheduler when placing it into the TL_NORMAL queue.
The next thread walking algorithm in commit 566df309c ("MEDIUM:
connections: Attempt to get idle connections from other threads.")
proved to be sufficient for most cases, but it still has some rough
edges when threads are unevenly loaded. If one thread wakes up with
10 streams to process in a burst, it will mainly take over connections
from the next one until it doesn't have anymore.
This patch implements a rotating index that is stored into the server
list and that any thread taking over a connection is responsible for
updating. This way it starts mostly random and avoids always picking
from the same place. This results in a smoother distribution overall
and a slightly lower takeover rate.
The problem with the way idle connections currently work is that it's
easy for a thread to steal all of its siblings' connections, then release
them, then it's done by another one, etc. This happens even more easily
due to scheduling latencies, or merged events inside the same pool loop,
which, when dealing with a fast server responding in sub-millisecond
delays, can really result in one thread being fully at work at a time.
In such a case, we perform a huge amount of takeover() which consumes
CPU and requires quite some locking, sometimes resulting in lower
performance than expected.
In order to fight against this problem, this patch introduces a new server
setting "pool-low-conn", whose purpose is to dictate when it is allowed to
steal connections from a sibling. As long as the number of idle connections
remains at least as high as this value, it is permitted to take over another
connection. When the idle connection count becomes lower, a thread may only
use its own connections or create a new one. By proceeding like this even
with a low number (typically 2*nbthreads), we quickly end up in a situation
where all active threads have a few connections. It then becomes possible
to connect to a server without bothering other threads the vast majority
of the time, while still being able to use these connections when the
number of available FDs becomes low.
We also use this threshold instead of global.nbthread in the connection
release logic, allowing to keep more extra connections if needed.
A test performed with 10000 concurrent HTTP/1 connections, 16 threads
and 210 servers with 1 millisecond of server response time showed the
following numbers:
haproxy 2.1.7: 185000 requests per second
haproxy 2.2: 314000 requests per second
haproxy 2.2 lowconn 32: 352000 requests per second
The takeover rate goes down from 300k/s to 13k/s. The difference is
further amplified as the response time shrinks.
There was a logic bug in commit ddfe0743d ("MEDIUM: server: use the two
thresholds for the connection release algorithm"): instead of keeping
only our first idle connection when FDs become scarce, the condition was
inverted resulting in enforcing this constraint unless FDs are scarce.
This results in less idle connections than permitted to be kept under
normal condition.
No backport needed.
When not sharing pools (i.e. when building with -DDEBUG_DONT_SHARE_POOLS)
we have about 47 pools right now, while MAX_BASE_POOLS is only 32, meaning
that only the first 32 ones will benefit from a per-thread cache entry.
This totally kills performance when pools are not shared (roughly -20%).
Let's double the limit to gain some margin, and make it possible to set
it as a build option.
It might be useful to backport this to stable versions as they're likely
to be affected as well.
The algorithm improvement in bdb86bd ("MEDIUM: server: improve estimate
of the need for idle connections") is still not enough because there's
a hard limit between below and above the FD count, so it continues to
end up with many killed connections.
Here we're proceeding differently. Given that there are two configured
limits, a low and a high one, what we do is that we drop connections
when the high limit is reached (what's already done by the killing task
anyway), when we're between the low and the high threshold, we only keep
the connection if our idle entries are empty (with a preference for safe
ones), and below the low threshold, we keep any connection so as to give
them a chance of being reused or taken over by another thread.
Proceeding like this results in much less dropped connections, we
typically see a 99.3% reuse rate (76k conns for 10M requests over 200
servers and 4 threads, with 335k takeovers or 3%), and much less CPU
usage variations because there are no more bursts to try to kill extra
connections.
It should be possible to further improve this by counting the number
of threads exploiting a server and trying to optimize the amount of
per-thread idle connections so that it is approximately balanced among
the threads.
The idle server connection estimates brought in commit bdb86bd ("MEDIUM:
server: improve estimate of the need for idle connections") were committed
without the minimum of 1 idle conn needed for the current thread. The net
effect is that there are bursts of dropped connections when the load varies
because there's no provision for the last connection.
No backport needed, this is 2.2-dev.
This reverts previous commit 347bbf79d20e1cff57075a8a378355dfac2475e2i.
The original code was correct. This patch resulted from a mistaken analysis
and breaks the scheduler:
########################## Starting vtest ##########################
Testing with haproxy version: 2.2-dev11-90b7d9-23
# top TEST reg-tests/lua/close_wait_lf.vtc TIMED OUT (kill -9)
# top TEST reg-tests/lua/close_wait_lf.vtc FAILED (10.008) signal=9
1 tests failed, 0 tests skipped, 88 tests passed
Program terminated with signal SIGABRT, Aborted.
[Current thread is 1 (Thread 0x7fb0dac2c700 (LWP 11292))]
(gdb) bt
#0 0x00007fb0e7c143f8 in raise () from /lib64/libc.so.6
#1 0x00007fb0e7c15ffa in abort () from /lib64/libc.so.6
#2 0x000000000053f5d6 in ha_panic () at src/debug.c:269
#3 0x00000000005a6248 in wdt_handler (sig=14, si=<optimized out>, arg=<optimized out>) at src/wdt.c:119
#4 <signal handler called>
#5 0x00000000004fbccd in tasklet_wakeup (tl=0x1b5abc0) at include/haproxy/task.h:351
#6 listener_accept (fd=<optimized out>) at src/listener.c:999
#7 0x00000000004262df in fd_update_events (evts=<optimized out>, fd=6) at include/haproxy/fd.h:418
#8 _do_poll (p=<optimized out>, exp=<optimized out>, wake=<optimized out>) at src/ev_epoll.c:251
#9 0x0000000000548d0f in run_poll_loop () at src/haproxy.c:2949
#10 0x000000000054908b in run_thread_poll_loop (data=<optimized out>) at src/haproxy.c:3067
#11 0x00007fb0e902b684 in start_thread () from /lib64/libpthread.so.0
#12 0x00007fb0e7ce5eed in clone () from /lib64/libc.so.6
(gdb) up
#5 0x00000000004fbccd in tasklet_wakeup (tl=0x1b5abc0) at include/haproxy/task.h:351
351 if (MT_LIST_ADDQ(&task_per_thread[tl->tid].shared_tasklet_list, (struct mt_list *)&tl->list) == 1) {
If the commit above is ever backported, this one must be as well!
In MT_LIST_ADDQ() and MT_LIST_ADD() we can't just check if the element is
already in a list, because there's a small race condition, it could be added
between the time we checked, and the time we actually set its next and prev.
So we have to lock it first.
This should be backported to 2.1.
The max_used_conns value is used as an estimate of the needed number of
connections on a server to know how many to keep open. But this one is
not reported, making it hard to troubleshoot reuse issues. Let's export
it in the sessions/current column.
Starting with commit 079cb9a ("MEDIUM: connections: Revamp the way idle
connections are killed") we started to improve the way to compute the
need for idle connections. But the condition to keep a connection idle
or drop it when releasing it was not updated. This often results in
storms of close when certain thresholds are met, and long series of
takeover() when there aren't enough connections left for a thread on
a server.
This patch tries to improve the situation this way:
- it keeps an estimate of the number of connections needed for a server.
This estimate is a copy of the max over previous purge period, or is a
max of what is seen over current period; it differs from max_used_conns
in that this one is a counter that's reset on each purge period ;
- when releasing, if the number of current idle+used connections is
lower than this last estimate, then we'll keep the connection;
- when releasing, if the current thread's idle conns head is empty,
and we don't exceed the estimate by the number of threads, then
we'll keep the connection.
- when cleaning up connections, we consider the max of the last two
periods to avoid killing too many idle conns when facing bursty
traffic.
Thanks to this we can better converge towards a situation where, provided
there are enough FDs, each active server keeps at least one idle connection
per thread all the time, with a total number close to what was needed over
the previous measurement period (as defined by pool-purge-delay).
On tests with large numbers of concurrent connections (30k) and many
servers (200), this has quite smoothed the CPU usage pattern, increased
the reuse rate and roughly halved the takeover rate.
The FD takeover operation might have certain impacts explaining
unexpected activities, so it's important to report such a counter
there. We thus count the number of times a thread has stolen an
FD from another thread.
The servers have internal states describing the status of idle connections,
unfortunately these were not exported in the stats. This patch adds the 3
following gauges:
- idle_conn_cur : Current number of unsafe idle connections
- safe_conn_cur : Current number of safe idle connections
- used_conn_cur : Current number of connections in use
The LRU cache head was an array of list, which causes false sharing
between 4 to 8 threads in the same cache line. Let's move it to the
thread_info structure instead. There's no need to do the same for the
pool_cache[] array since it's already quite large (32 pointers each).
By doing this the request rate increased by 1% on a 16-thread machine.
pool-t.h was mistakenly including the full-blown includes for threads,
lists and api instead of the types, and as such, CONFIG_HAP_LOCAL_POOLS
and CONFIG_HAP_LOCKLESS_POOLS were not visible everywhere.
The thread_info struct is convenient to store various per-thread info
without having to resort to a painful thread_local storage which is
slow and painful to initialize.
The problem is, by having this one in thread.h it's very difficult to
add more entries there because everyone already includes thread.h so
conversely thread.h cannot reference certain types.
There's no point in having this there, instead let's create a new pair
of files, tinfo{,-t}.h, which declare the structure. This way it will
become possible to extend them with other includes and have certain
files store their own types there.
We used to have 3 thread-based arrays for toremove_lock, idle_cleanup,
and toremove_connections. The problem is that these items are small,
and that this creates false sharing between threads since it's possible
to pack up to 8-16 of these values into a single cache line. This can
cause real damage where there is contention on the lock.
This patch creates a new array of struct "idle_conns" that is aligned
on a cache line and which contains all three members above. This way
each thread has access to its variables without hindering the other
ones. Just doing this increased the HTTP/1 request rate by 5% on a
16-thread machine.
The definition was moved to connection.{c,h} since it appeared a more
natural evolution of the ongoing changes given that there was already
one of them declared in connection.h previously.
It looked strange to see pool_evict_from_cache() always very present
on "perf top", but there was actually a reason to this: while b_free()
uses pool_free() which properly disposes the buffer into the local cache
and b_alloc_fast() allocates using pool_get_first() which considers the
local cache, b_alloc_margin() does not consider the local cache as it
only uses __pool_get_first() which only allocates from the shared pools.
The impact is that basically everywhere a buffer is allocated (muxes,
streams, applets), it's always picked from the shared pool (hence
involves locking) and is released to the local one and makes it grow
until it's required to trigger a flush using pool_evict_from_cache().
Buffers usage are thus not thread-local at all, and cause eviction of
a lot of possibly useful objects from the local caches.
Just fixing this results in a 10% request rate increase in an HTTP/1 test
on a 16-thread machine.
This bug was caused by recent commit ed891fd ("MEDIUM: memory: make local
pools independent on lockless pools") merged into 2.2-dev9, so not backport
is needed.
BoringSSL does not have X509_get_X509_PUBKEY
let our emulation level define that for BoringSSL as well
Build log:
src/ssl_sample.o: In function `smp_fetch_ssl_x_key_alg':
/home/travis/build/haproxy/haproxy/src/ssl_sample.c:592: undefined reference to `X509_get_X509_PUBKEY'
clang-7: error: linker command failed with exit code 1 (use -v to see invocation)
Makefile:860: recipe for target 'haproxy' failed
make: *** [haproxy] Error 1
travis-ci: https://travis-ci.com/github/haproxy/haproxy/jobs/351670996
With the rework of the config line parser, we've started to emit a dump
of the initial line underlined by a caret character indicating the error
location. But with extremely large lines it starts to take time and can
even cause trouble to slow terminals (e.g. over ssh), and this becomes
useless. In addition, control characters could be dumped as-is which is
bad, especially when the input file is accidently wrong (an executable).
This patch adds a string sanitization function which isolates an area
around the error position in order to report only that area if the string
is too large. The limit was set to 80 characters, which will result in
roughly 40 chars around the error being reported only, prefixed and suffixed
with "..." as needed. In addition, non-printable characters in the line are
now replaced with '?' so as not to corrupt the terminal. This way invalid
variable names, unmatched quotes etc will be easier to spot.
A typical output is now:
[ALERT] 176/092336 (23852) : parsing [bad.cfg:8]: forbidden first char in environment variable name at position 811957:
...c$PATH$PATH$d(xlc`%?$PATH$PATH$dgc?T$%$P?AH?$PATH$PATH$d(?$PATH$PATH$dgc?%...
^
Now that all tasklet queues are scanned at once by run_tasks_from_lists(),
it becomes possible to always check for lower priority classes and jump
back to them when they exist.
This patch adds tune.sched.low-latency global setting to enable this
behavior. What it does is stick to the lowest ranked priority list in
which tasks are still present with an available budget, and leave the
loop to refill the tasklet lists if the trees got new tasks or if new
work arrived into the shared urgent queue.
Doing so allows to cut the latency in half when running with extremely
deep run queues (10k-100k), thus allowing forwarding of small and large
objects to coexist better. It remains off by default since it does have
a small impact on large traffic by default (shorter batches).
Now process_runnable_tasks is responsible for calculating the budgets
for each queue, dequeuing from the tree, and calling run_tasks_from_lists().
This latter one scans the queues, picking tasks there and respecting budgets.
Note that its name was updated with a plural "s" for this reason.
It is neither convenient nor scalable to check each and every tasklet
queue to figure whether it's empty or not while we often need to check
them all at once. This patch introduces a tasklet class mask which gets
a bit 1 set for each queue representing one class of service. A single
test on the mask allows to figure whether there's still some work to be
done. It will later be usable to better factor the runqueue code.
Bits are set when tasklets are queued. They're cleared when queues are
emptied. It is possible that a queue is empty but has a bit if a tasklet
was added then removed, but this is not a problem as this is properly
checked for in run_tasks_from_list().
It will be convenient to have the tasklet queue number soon, better make
current_queue an index rather than a pointer to the queue. When not currently
running (e.g. from I/O), the index is -1.
Add some functions to deinit the whole crtlist and ckch architecture.
It will free all crtlist, crtlist_entry, ckch_store, ckch_inst and their
associated SNI, ssl_conf and SSL_CTX.
The SSL_CTX in the default_ctx and initial_ctx still needs to be free'd
separately.
Since commit 2954c47 ("MEDIUM: ssl: allow crt-list caching"), the
ssl_bind_conf is allocated directly in the crt-list, and the crt-list
can be shared between several bind_conf. The deinit() code wasn't
changed to handle that.
This patch fixes the issue by removing the free of the ssl_conf in
ssl_sock_free_all_ctx().
It should be completed with a patch that free the ssl_conf and the
crt-list.
Fix issue #700.
A test on large objects revealed a big performance loss from 2.1. The
cause was found to be related to cache locality between scheduled
operations that are batched using tasklets. It happens that we now
have several layers of tasklets and that queuing all these operations
leaves time to let memory objects cool down in the CPU cache, effectively
resulting in halving the performance.
A quick test consisting in putting most unknown tasklets into the BULK
queue almost fixed the performance regression, but this is a wrong
approach as it can also slow down some low-latency transfers or access
to applets like the CLI.
What this patch does instead is to queue unknown tasklets into the same
queue as the current one when tasklet_wakeup() is itself called from a
task/tasklet, otherwise it uses urgent for real I/O (when sched->current
is NULL). This results in the called tasklet being woken up much sooner,
often at the end of the current batch of tasklets.
By doing so, a test on 2 cores 4 threads with 256 concurrent H1 conns
transferring 16m objects with 256kB buffers jumped from 55 to 88 Gbps.
It's even possible to go as high as 101 Gbps by evaluating the URGENT
queue after the BULK one, though this was not done as considered
dangerous for latency sensitive operations.
This reinforces the importance of getting back the CPU transfer
mechanisms based on tasklet_wakeup_after() to work at the tasklet level
by supporting an immediate wakeup in certain cases.
No backport is needed, this is strictly 2.2.
In task_per_thread[] we now have current_queue which is a pointer to
the current tasklet_list entry being evaluated. This will be used to
know the class under which the current task/tasklet is currently
running.
When DEBUG_FD is set at build time, we'll keep a counter of per-FD events
in the fdtab. This counter is reported in "show fd" even for closed FDs if
not zero. The purpose is to help spot situations where an apparently closed
FD continues to be reported in loops, or where some events are dismissed.
As reported in issue #419, a "clear map" operation on a very large map
can take a lot of time and freeze the entire process for several seconds.
This patch makes sure that pat_ref_prune() can regularly yield after
clearing some entries so that the rest of the process continues to work.
The first part, the removal of the patterns, can take quite some time
by itself in one run but it's still relatively fast. It may block for
up to 100ms for 16M IP addresses in a tree typically. This change needed
to declare an I/O handler for the clear operation so that we can get
back to it after yielding.
The second part can be much slower because it deconstructs the elements
and its users, but it iterates progressively so we can yield less often
here.
The patch was tested with traffic in parallel sollicitating the map being
released and showed no problem. Some traffic will definitely notice an
incomplete map but the filling is already not atomic anyway thus this is
not different.
It may be backported to stable versions once sufficiently tested for side
effects, at least as far as 2.0 in order to avoid the watchdog triggering
when the process is frozen there. For a better behaviour, all these
prune_* functions should support yielding so that the callers have a
chance to continue also yield in turn.
Some of the recent optimizations around the polling to save a few
epoll_ctl() calls have shown that they could also cause some trouble.
However, over time our code base has become totally asynchronous with
I/Os always attempted from the upper layers and only retried at the
bottom, making it look like we're getting closer to EPOLLET support.
There are showstoppers there such as the listeners which cannot support
this. But given that most of the epoll_ctl() dance comes from the
connections, we can try to enable edge-triggered polling on connections.
What this patch does is to add a new global tunable "tune.fd.edge-triggered",
that makes fd_insert() automatically set an et_possible bit on the fd if
the I/O callback is conn_fd_handler. When the epoll code sees an update
for such an FD, it immediately registers it in both directions the first
time and doesn't update it anymore.
On a few tests it proved quite useful with a 14% request rate increase in
a H2->H1 scenario, reducing the epoll_ctl() calls from 2 per request to
2 per connection.
The option is obviously disabled by default as bugs are still expected,
particularly around the subscribe() code where it is possible that some
layers do not always re-attempt reading data after being woken up.
localpeer <name>
Sets the local instance's peer name. It will be ignored if the "-L"
command line argument is specified or if used after "peers" section
definitions. In such cases, a warning message will be emitted during
the configuration parsing.
This option will also set the HAPROXY_LOCALPEER environment variable.
See also "-L" in the management guide and "peers" section in the
configuration manual.
This one was confusingly called, I thought it was the cumulated number
of streams but it's the number of calls to process_stream(). Let's make
this clearer.
We have poll_drop, poll_dead and poll_skip which are confusingly named
like their poll_io and poll_exp counterparts except that they are not
per poll() call but per-fd. This patch renames them to poll_drop_fd(),
poll_dead_fd() and poll_skip_fd() for this reason.
The "show activity" output mentions a number of indicators to explain
wake up reasons but doesn't have the number of times poll() sees some
I/O. And given that multiple events can happen simultaneously, it's
not always possible to deduce this metric by subtracting.
This patch adds a new "poll_io" counter that allows one to see how
often poll() returns with at least one active FD. This should help
detect stuck events and measure various ratios of poll sub-metrics.
fd_set_running() and fd_takeover() may both use a double-word CAS on the
(running_mask, thread_mask) couple and as such they expect the fields to
be exactly arranged like this. It's critical not to reorder them, so add
a comment to avoid such a potential mistake later.
Since 2.1-dev2, with commit 305d5ab46 ("MAJOR: fd: Get rid of the fd cache.")
we don't have the fd_lock anymore and as such its acitvity counter is always
zero. Let's remove it from the struct and from "show activity" output, as
there are already plenty of indicators to look at.
The cache line comment in the struct activity was updated to reflect
reality as it looks like another one already got removed in the past.
This macro is provided by clang but gcc lacks it. Not having it makes
it painful to test features on both compilers. Better define it to zero
when not available so that __has_feature(foo) never errors.
This function takes on input a string to tokenize, an output storage
(which may be the same) and a number of options indicating how to handle
certain characters (single & double quote support, backslash support,
end of line on '#', environment variables etc). On output it will provide
a list of pointers to individual words after having possibly unescaped
some character sequences, handled quotes and resolved environment
variables, and it will also indicate a status made of:
- a list of failures (overlap between src/dst, wrong quote etc)
- the pointer to the first sequence in error
- the required output length (a-la snprintf()).
This allows a caller to freely unescape/unquote a string by using a
pre-allocated temporary buffer and expand it as necessary. It takes
extreme care at avoiding expensive operations and intentionally does
not use memmove() when removing escapes, hence the reason for the
different input and output buffers. The goal is to use it as the basis
for the config parser.
As reported in issue #689, there is a subtle bug in the ebtree code used
to compared memory blocks. It stems from the platform-dependent memcmp()
implementation. Original implementations used to perform a byte-per-byte
comparison and to stop at the first non-matching byte, as in this old
example:
https://www.retro11.de/ouxr/211bsd/usr/src/lib/libc/compat-sys5/memcmp.c.html
The ebtree code has been relying on this to detect the first non-matching
byte when comparing keys. This is made so that a zero-terminated string
can fail to match against a longer string.
Over time, especially with large busses and SIMD instruction sets,
multi-byte comparisons have appeared, making the processor fetch bytes
past the first different byte, which could possibly be a trailing zero.
This means that it's possible to read past the allocated area for a
string if it was allocated by strdup().
This is not correct and definitely confuses address sanitizers. In real
life the problem doesn't have visible consequences. Indeed, multi-byte
comparisons are implemented so that aligned words are loaded (e.g. 512
bits at once to process a cache line at a time). So there is no way such
a multi-byte access will cross a page boundary and end up reading from
an unallocated zone. This is why it was never noticed before.
This patch addresses this by implementing a one-byte-at-a-time memcmp()
variant for ebtree, called eb_memcmp(). It's optimized for both small and
long strings and guarantees to stop after the first non-matching byte. It
only needs 5 instructions in the loop and was measured to be 3.2 times
faster than the glibc's AVX2-optimized memcmp() on short strings (1 to
257 bytes), since that latter one comes with a significant setup cost.
The break-even seems to be at 512 bytes where both version perform
equally, which is way longer than what's used in general here.
This fix should be backported to stable versions and reintegrated into
the ebtree code.
Commit 0a3b43d9c ("MINOR: haproxy: Make use of deinit_and_exit() for
clean exits") introduced this build warning:
src/haproxy.c: In function 'main':
src/haproxy.c:3775:1: warning: control reaches end of non-void function [-Wreturn-type]
}
^
This is because the new deinit_and_exit() is not marked as "noreturn"
so depending on the optimizations, the noreturn attribute of exit() will
either leak through it and silence the warning or not and confuse the
compiler. Let's just add the attribute to fix this.
No backport is needed, this is purely 2.2.
As reported in issue #686, ARM64 build fails since the include files
reorganization. This is caused by the lack of string.h while a memcpy()
is present in __ha_cas_dw().
