The crtlist_entry structure use a pointer to the store as key.
That's a problem with the dynamic update of a certificate over the CLI,
because it allocates a new ckch_store. So updating the pointers is
needed. To achieve that, a linked list of the crtlist_entry was added in
the ckch_store, so it's easy to iterate on this list to update the
pointers. Another solution would have been to rework the system so we
don't allocate a new ckch_store, but it requires a rework of the ckch
code.
When updating a ckch_store we may want to update its pointer in the
crtlist_entry which use it. To do this, we need the list of the entries
using the store.
The instances were wrongly inserted in the crtlist entries, all
instances of a crt-list were inserted in the last crt-list entry.
Which was kind of handy to free all instances upon error.
Now that it's done correctly, the error path was changed, it must
iterate on the entries and find the ckch_insts which were generated for
this bind_conf. To avoid wasting time, it stops the iteration once it
found the first unsuccessful generation.
In order to be able to add new certificate in a crt-list, we need the
list of bind_conf that uses this crt-list so we can create a ckch_inst
for each of them.
The original algorithm always killed half the idle connections. This doesn't
take into account the way the load can change. Instead, we now kill half
of the exceeding connections (exceeding connection being the number of
used + idle connections past the last maximum used connections reached).
That way if we reach a peak, we will kill much less, and it'll slowly go back
down when there's less usage.
Add a counter to know the current number of used connections, as well as the
max, this will be used later to refine the algorithm used to kill idle
connections, based on current usage.
With server-template was introduced the possibility to scale the
number of servers in a backend without needing a configuration change
and associated reload. On the other hand it became impractical to
write use-server rules for these servers as they would only accept
existing server labels as argument. This patch allows the use of
log-format notation to describe targets of a use-server rules, such
as in the example below:
listen test
bind *:1234
use-server %[hdr(srv)] if { hdr(srv) -m found }
use-server s1 if { path / }
server s1 127.0.0.1:18080
server s2 127.0.0.1:18081
If a use-server rule is applied because it was conditionned by an
ACL returning true, but the target of the use-server rule cannot be
resolved, no other use-server rule is evaluated and we fall back to
load balancing.
This feature was requested on the ML, and bumped with issue #563.
Add a sample fetch for the name of a bind. This can be useful to
take decisions when PROXY protocol is used and we can't rely on dst,
such as the sample config below.
defaults
mode http
listen bar
bind 127.0.0.1:1111
server s1 127.0.1.1:1234 send-proxy
listen foo
bind 127.0.1.1:1234 name foo accept-proxy
http-request return status 200 hdr dst %[dst] if { dst 127.0.1.1 }
First: self issued CA, aka root CA, is the enchor for chain validation,
no need to send it, client must have it. HAProxy can skip it in ssl_ctx.
Second: the main motivation to skip root CA in ssl_ctx is to be able to
provide it in the chain without drawback. Use case is to provide issuer
for ocsp without the need for .issuer and be able to share it in
issuers-chain-path. This concerns all certificates without intermediate
certificates. It's useless for BoringSSL, .issuer is ignored because ocsp
bits doesn't need it.
13a9232ebc introduced parsing of
Additionnal DNS response section to pick up IP address when available.
That said, this introduced a side effect for other query types (A and
AAAA) leading to consider those responses invalid when parsing the
Additional section.
This patch avoids this situation by ensuring the Additional section is
parsed only for SRV queries.
In h1_detach(), if our input buffer isn't empty, don't just subscribe(), we
may hold a new request, and there's nothing left to read. Instead, call
h1_process() directly, so that a new stream is created.
Failure to do so means if we received the new request to early, the
connecetion will just hang, as it happens when using svn.
When a "peers" section has not any local peer, it is removed of the list
of "peers" sections by check_config_validity(). But a stick-table which
refers to a "peers" section stores a pointer to this peers section.
These pointer must be reset to NULL value for each stick-table refering to
such a "peers" section to prevent stktable_init() to start the peers frontend
attached to the peers section dereferencing the invalid pointer.
Furthemore this patch stops the peers frontend as this is done for other
configurations invalidated by check_config_validity().
Thank you to Olivier D for having reported this issue with such a
simple configuration file which made haproxy crash when started with
-c option for configuration file validation.
defaults
mode http
peers mypeers
peer toto 127.0.0.1:1024
backend test
stick-table type ip size 10k expire 1h store http_req_rate(1h) peers mypeers
Must be backported to 2.1 and 2.0.
Fix an infinite loop which was added in an attempt to fix#558.
If the peers_fe is NULL, it will loop forever.
Must be backported with a2cfd7e as far as 1.8.
Tim reported that in master-worker mode, if a stick-table is declared
but not used in the configuration, its associated peers listener won't
bind.
This problem is due the fact that the master-worker and the daemon mode,
depend on the bind_proc field of the peers proxy to disable the listener.
Unfortunately the bind_proc is left to 0 if no stick-table were used in
the configuration, stopping the listener on all processes.
This fixes sets the bind_proc to the first process if it wasn't
initialized.
Should fix bug #558. Should be backported as far as 1.8.
SSL_CTX_set1_chain is used for openssl >= 1.0.2 and a loop with
SSL_CTX_add_extra_chain_cert for openssl < 1.0.2.
SSL_CTX_add_extra_chain_cert exist with openssl >= 1.0.2 and can be
used for all openssl version (is new name is SSL_CTX_add0_chain_cert).
This patch use SSL_CTX_add_extra_chain_cert to remove any #ifdef for
compatibilty. In addition sk_X509_num()/sk_X509_value() replace
sk_X509_shift() to extract CA from chain, as it is used in others part
of the code.
This bug was introduced by 85888573 "BUG/MEDIUM: ssl: chain must be
initialized with sk_X509_new_null()". No need to set find_chain with
sk_X509_new_null(), use find_chain conditionally to fix issue #516.
This bug was referenced by issue #559.
[wla: fix some alignment/indentation issue]
In run_thread_poll_loop() we test both for (global_tasks_mask & tid_bit)
and thread_has_tasks(), but the former is useless since this test is
already part of the latter.
Commit 4b3f27b ("BUG/MINOR: haproxy/threads: try to make all threads
leave together") improved the soft-stop synchronization but it left a
small race open because it looks at tasks_run_queue, which can drop
to zero then back to one while another thread picks the task from the
run queue to insert it into the tasklet_list. The risk is very low but
not null. In addition the condition didn't consider the possible presence
of signals in the queue.
This patch moves the stopping detection just after the "wake" calculation
which already takes care of the various queues' sizes and signals. It
avoids needlessly duplicating these tests.
The bug was discovered during a code review but will probably never be
observed. This fix may be backported to 2.1 and 2.0 along with the commit
above.
In the various muxes, add a comment documenting that once
srv_add_to_idle_list() got called, any thread may pick that conenction up,
so it is unsafe to access the mux context/the connection, the only thing we
can do is returning.
In h1_detach(), make sure we subscribe before we call
srv_add_to_idle_list(), not after. As soon as srv_add_to_idle_list() is
called, and it is put in an idle list, another thread can take it, and
we're no longer allowed to subscribe.
This fixes a race condition when another thread grabs a connection as soon
as it is put, the original owner would subscribe, and thus the new thread
would fail to do so, and to activate polling.
Fix a potential NULL dereference in "show ssl cert" when we can't
allocate the <out> trash buffer.
This patch creates a new label so we could jump without trying to do the
ci_putchk in this case.
This bug was introduced by ea987ed ("MINOR: ssl/cli: 'new ssl cert'
command"). 2.2 only.
This bug was referenced by issue #556.
In connect_server(), make sure we properly free a newly created connection
if we somehow fail, and it has not yet been attached to a conn_stream, or
it would lead to a memory leak.
This should appease coverity for backend.c, as reported in inssue #556.
This should be backported to 2.1, 2.0 and 1.9
Fix a memory leak that could happen upon a "show ssl cert" if notBefore:
or notAfter: failed to extract its ASN1 string.
Introduced by d4f946c ("MINOR: ssl/cli: 'show ssl cert' give information
on the certificates"). 2.2 only.
In `crtlist_dup_filters()` add the `1` to the number of elements instead of
the size of a single element.
This bug was introduced in commit 2954c478eb,
which is 2.2+. No backport needed.
In `ckch_inst_sni_ctx_to_sni_filters` use `calloc()` to allocate the filter
array. When the function fails to allocate memory for a single entry the
whole array will be `free()`d using free_sni_filters(). With the previous
`malloc()` the pointers for entries after the failing allocation could
possibly be a garbage value.
This bug was introduced in commit 38df1c8006,
which is 2.2+. No backport needed.
In conn_backend_get(), when we manage to get an idle connection from the
current thread's pool, don't forget to decrement the idle connection
counters, or we may end up not reusing connections when we could, and/or
killing connections when we shouldn't.
In connect_server(), if we notice we have more file descriptors opened than
we should, there's no reason not to close a connection just because we're
reusing one, so do it anyway.
In connect_server(), if we no longer have any idle connections for the
current thread, attempt to use the new "takeover" mux method to steal a
connection from another thread.
This should have no impact right now, given no mux implements it.
Make the "list" element a struct mt_list, and explicitely use
list_from_mt_list to get a struct list * where it is used as such, so that
mt_list_for_each_entry will be usable with it.
Implement a new function, fd_takeover(), that lets you become the thread
responsible for the fd. On architectures that do not have a double-width CAS,
use a global rwlock.
fd_set_running() was also changed to be able to compete with fd_takeover(),
either using a dooble-width CAS on both running_mask and thread_mask, or
by claiming a reader on the global rwlock. This extra operation should not
have any measurable impact on modern architectures where threading is
relevant.
Revamp the server connection lists. We know have 3 lists :
- idle_conns, which contains idling connections
- safe_conns, which contains idling connections that are safe to use even
for the first request
- available_conns, which contains connections that are not idling, but can
still accept new streams (those are HTTP/2 or fastcgi, and are always
considered safe).
Make it so sessions are not responsible for connection anymore, except for
connections that are private, and thus can't be shared, otherwise, as soon
as a request is done, the session will just add the connection to the
orphan connections pool.
This will break http-reuse safe, but it is expected to be fixed later.
The CLI command "new ssl cert" allows one to create a new certificate
store in memory. It can be filed with "set ssl cert" and "commit ssl
cert".
This patch also made a small change in "show ssl cert" to handle an
empty certificate store.
Multi-certificate bundles are not supported since they will probably be
removed soon.
This feature alone is useless since there is no way to associate the
store to a crt-list yet.
Example:
$ echo "new ssl cert foobar.pem" | socat /tmp/sock1 -
New empty certificate store 'foobar.pem'!
$ printf "set ssl cert foobar.pem <<\n$(cat localhost.pem.rsa)\n\n" | socat /tmp/sock1 -
Transaction created for certificate foobar.pem!
$ echo "commit ssl cert foobar.pem" | socat /tmp/sock1 -
Committing foobar.pem
Success!
$ echo "show ssl cert foobar.pem" | socat /tmp/sock1 -
Filename: foobar.pem
[...]
The flush_lock was introduced, mostly to be sure that pool_gc() will never
dereference a pointer that has been free'd. __pool_get_first() was acquiring
the lock to, the fear was that otherwise that pointer could get free'd later,
and then pool_gc() would attempt to dereference it. However, that can not
happen, because the only functions that can free a pointer, when using
lockless pools, are pool_gc() and pool_flush(), and as long as those two
are mutually exclusive, nobody will be able to free the pointer while
pool_gc() attempts to access it.
So change the flush_lock to a spinlock, and don't bother acquire/release
it in __pool_get_first(), that way callers of __pool_get_first() won't have
to wait while the pool is flushed. The worst that can happen is we call
__pool_refill_alloc() while the pool is getting flushed, and memory can
get allocated just to be free'd.
This may help with github issue #552
This may be backported to 2.1, 2.0 and 1.9.
When running __signal_process_queue(), we ignore most signals. We can't,
however, ignore WDTSIG and DEBUGSIG, otherwise that thread may end up
waiting for another one that could hold a glibc lock, while the other thread
wait for this one to enter debug_handler().
So make sure WDTSIG and DEBUGSIG aren't ignored, if they are defined.
This probably explains the watchdog deadlock described in github issue
This should be backported to 2.1, 2.0 and 1.9.
Move the definition of WDTSIG and DEBUGSIG from wdt.c and debug.c into
types/signal.h, so that we can access them in another file.
We need those definition to avoid blocking those signals when running
__signal_process_queue().
This should be backported to 2.1, 2.0 and 1.9.
The behavior of calloc() when being passed `0` as `nelem` is implementation
defined. It may return a NULL pointer.
Avoid this issue by checking before allocating. While doing so adjust the local
integer variables that are used to refer to memory offsets to `size_t`.
This issue was introced in commit f91ac19299. This
patch should be backported together with that commit.
There is a memleak of the entry structure in crtlist_load_cert_dir(), in
the case we can't stat the file, or this is not a regular file. Let's
move the entry allocation so it's done after these tests.
Fix issue #551.
When tasklet were introduced, it has been decided not to provide the tasklet
to the callback, but NULL instead. While it may have been reasonable back
then, maybe to be able to differentiate a task from a tasklet from the
callback, it also means that we can't access the tasklet from the handler if
the context provided can't be trusted.
As no handler is shared between a task and a tasklet, and there are now
other means of distinguishing between task and tasklet, just pass the
tasklet pointer too.
This may be backported to 2.1, 2.0 and 1.9 if needed.
A memory leak happens in an error case when ckchs_load_cert_file()
returns NULL in crtlist_parse_file().
This bug was introduced by commit 2954c47 ("MEDIUM: ssl: allow crt-list caching")
This patch fixes bug #551.
In the struct fdtab, introduce a new mask, running_mask. Each thread should
add its bit before using the fd.
Use the running_mask instead of a lock, in fd_insert/fd_delete, we'll just
spin as long as the mask is non-zero, to be sure we access the data
exclusively.
fd_set_running_excl() spins until the mask is 0, fd_set_running() just
adds the thread bit, and fd_clr_running() removes it.
Implement 2 new commands on the CLI:
show ssl crt-list [<filename>]: Without a specified filename, display
the list of crt-lists used by the configuration. If a filename is
specified, it will displays the content of this crt-list, with a line
identifier at the beginning of each line. This output must not be used
as a crt-list file.
dump ssl crt-list <filename>: Dump the content of a crt-list, the output
can be used as a crt-list file.
Note: It currently displays the default ssl-min-ver and ssl-max-ver
which are potentialy not in the original file.
Don't bother trying to remove the connection from the idle list, as the
only connections the mux_pt handles are now the TCP-mode connections, and
those are never added to the idle list.
The agent's engine_id forgot to dup from trash, all engine_ids point to
the same address "&trash.area", the engine_id changed at run time and will
double free when release agents and trash.
This bug was introduced by the commit ee3bcddef ("MINOR: tools: add a generic
function to generate UUIDs").
No backport is needed, this is 2.2-dev.
The commit 6be66ec ("MINOR: ssl: directories are loaded like crt-list")
broke the directory loading of the certificates. The <crtlist> wasn't
filled by the crtlist_load_cert_dir() function. And the entries were
not correctly initialized. Leading to a segfault during startup.
Generate a directory cache with the crtlist and crtlist_entry structures.
With this new model, directories are a special case of the crt-lists.
A directory is a crt-list which allows only one occurence of each file,
without SSL configuration (ssl_bind_conf) and without filters.
The crtlist structure defines a crt-list in the HAProxy configuration.
It contains crtlist_entry structures which are the lines in a crt-list
file.
crt-list are now loaded in memory using crtlist and crtlist_entry
structures. The file is read only once. The generation algorithm changed
a little bit, new ckch instances are generated from the crtlist
structures, instead of being generated during the file loading.
The loading function was split in two, one that loads and caches the
crt-list and certificates, and one that looks for a crt-list and creates
the ckch instances.
Filters are also stored in crtlist_entry->filters as a char ** so we can
generate the sni_ctx again if needed. I won't be needed anymore to parse
the sni_ctx to do that.
A crtlist_entry stores the list of all ckch_inst that were generated
from this entry.
Pass a pointer to the struct ckch_inst to the ssl_sock_load_ckchs()
function so we can manipulate the ckch_inst from
ssl_sock_load_cert_list_file() and ssl_sock_load_cert().
This bug was introduced with peers.c code re-work (7d0ceeec80):
"struct peer" flags are mistakenly checked instead of
"struct peers" flags to check the resync status of the local peer.
The issue was reported here:
https://github.com/haproxy/haproxy/issues/545
This bug affects all branches >= 2.0 and should be backported.
It's more generic and versatile than the previous shut_your_big_mouth_gcc()
that was used to silence annoying warnings as it's not limited to ignoring
syscalls returns only. This allows us to get rid of the aforementioned
function and the shut_your_big_mouth_gcc_int variable, that started to
look ugly in multi-threaded environments.
This message comes when we run:
haproxy -c -V -f /etc/haproxy/haproxy.cfg
[ALERT] 072/233727 (30865) : parsing [/etc/haproxy/haproxy.cfg:34] : The 'rspirep' directive is not supported anymore sionce HAProxy 2.1. Use 'http-response replace-header' instead.
[ALERT] 072/233727 (30865) : Error(s) found in configuration file : /etc/haproxy/haproxy.cfg
[ALERT] 072/233727 (30865) : Fatal errors found in configuration.
These are mostly comments in the code. A few error messages were fixed
and are of low enough importance not to deserve a backport. Some regtests
were also fixed.
This adds the missing blank lines in `make_proxy_line_v2` and
`conn_recv_proxy`. It also adjusts the type of the temporary variable
used for the return value of `recv` to be `ssize_t` instead of `int`.
This patch adds the `unique-id` option to `proxy-v2-options`. If this
option is set a unique ID will be generated based on the `unique-id-format`
while sending the proxy protocol v2 header and stored as the unique id for
the first stream of the connection.
This feature is meant to be used in `tcp` mode. It works on HTTP mode, but
might result in inconsistent unique IDs for the first request on a keep-alive
connection, because the unique ID for the first stream is generated earlier
than the others.
Now that we can send unique IDs in `tcp` mode the `%ID` log variable is made
available in TCP mode.
There's a small issue with soft stop combined with the incoming
connection load balancing. A thread may dispatch a connection to
another one at the moment stopping=1 is set, and the second one could
stop by seeing (jobs - unstoppable_jobs) == 0 in run_poll_loop(),
without ever picking these connections from the queue. This is
visible in that it may occasionally cause a connection drop on
reload since no remaining thread will ever pick that connection
anymore.
In order to address this, this patch adds a stopping_thread_mask
variable by which threads acknowledge their willingness to stop
when their runqueue is empty. And all threads will only stop at
this moment, so that if finally some late work arrives in the
thread's queue, it still has a chance to process it.
This should be backported to 2.1 and 2.0.
When stopping there is a risk that other threads are already in the
process of stopping, so let's not add new work in their queue and
instead keep the incoming connection local.
This should be backported to 2.1 and 2.0.
Surprizingly the variable was never initialized, though on most
platforms it's zeroed at boot, and it is relatively harmless
anyway since in the worst case the bits are updated around poll().
This was introduced by commit 79321b95a8 and needs to be backported
as far as 1.9.
In pool_gc(), when we're not using lockless pool, always update free_list,
and read from it the next element to free. As we now unlock the pool while
we're freeing the item, another thread could have updated free_list in our
back. Not doing so could lead to segfaults when pool_gc() is called.
This should be backported to 2.1.
Don't assume the connection always has a valid session in "owner".
Instead, attempt to retrieve the session from the stream, and modify
the error snapshot code to not assume we always have a session, or the proxy
for the other end.
x86_64 and ARM64 do support the double-word atomic CAS. However on
ARM it must be done only on aligned data. The random generator makes
use of such double-word atomic CAS when available but didn't enforce
alignment, which causes ARM64 to crash early in the startup since
commit 52bf839 ("BUG/MEDIUM: random: implement a thread-safe and
process-safe PRNG").
This commit just unconditionally aligns the arrays. It must be
backported to all branches where the commit above is backported
(likely till 2.0).
Remove the list of private connections from server, it has been largely
unused, we only inserted connections in it, but we would never actually
use it.
When a maximum memory setting is passed to haproxy and maxconn is not set
and ulimit-n is not set, it is expected that maxconn will be set to the
highest value permitted by this memory setting, possibly affecting the
FD limit.
When maxconn was changed to be deduced from the current process's FD limit,
the automatic setting above was partially lost because it now remains
limited to the current FD limit in addition to being limited to the
memory usage. For unprivileged processes it does not change anything,
but for privileged processes the difference is important. Indeed, the
previous behavior ensured that the new FD limit could be enforced on
the process as long as the user had the privilege to do so. Now this
does not happen anymore, and some people rely on this for automatic
sizing in VM environments.
This patch implements the ability to verify if the setting will be
enforceable on the process or not. First it computes maxconn based on
the memory limits alone, then checks if the process is willing to accept
them, otherwise tries again by respecting the process' hard limit.
Thanks to this we now have the best of the pre-2.0 behavior and the
current one, in that privileged users will be able to get as high a
maxconn as they need just based on the memory limit, while unprivileged
users will still get as high a setting as permitted by the intersection
of the memory limit and the process' FD limit.
Ideally, after some observation period, this patch along with the
previous one "MINOR: init: move the maxsock calculation code to
compute_ideal_maxsock()" should be backported to 2.1 and 2.0.
Thanks to Baptiste for raising the issue.
The maxsock value is currently derived from global.maxconn and a few other
settings, some of which also depend on global.maxconn. This makes it
difficult to check if a limit is already too high or not during the maxconn
automatic sizing.
Let's move this code into a new function, compute_ideal_maxsock() which now
takes a maxconn in argument. It performs the same operations and returns
the maxsock value if global.maxconn were to be set to that value. It now
replaces the previous code to compute maxsock.
When calling MT_LIST_DEL_SAFE(), give him the temporary pointer "tmpelt",
as that's what is expected. We want to be able to set that pointer to NULL,
to let other parts of the code know we deleted an element.
In order to store and cache the directory loading, the directory loading
was separated from ssl_sock_load_cert() and put in a new function
ssl_sock_load_cert_dir() to be more readable.
This patch only splits the function in two.
SI_TKILL is not necessarily defined on older systems and is used only
with the pthread_kill() call a few lines below, so it should also be
subject to the USE_THREAD condition.
Technically speaking the call was implemented in glibc 2.3 so we must
rely on this and not on __USE_GNU which is an internal define of glibc
to track use of GNU_SOURCE.
The splice() syscall has been supported in glibc since version 2.5 issued
in 2006 and is present on supported systems so there's no need for having
our own arch-specific syscall definitions anymore.
This was made to support epoll on patched 2.4 kernels, and on early 2.6
using alternative libcs thanks to the arch-specific syscall definitions.
All the features we support have been around since 2.6.2 and present in
glibc since 2.3.2, neither of which are found in field anymore. Let's
simply drop this and use epoll normally.
The accept4() syscall has been present for a while now, there is no more
reason for maintaining our own arch-specific syscall implementation for
systems lacking it in libc but having it in the kernel.
This was introduced 10 years ago to squeeze a few CPU cycles per syscall
on 32-bit x86 machines and was already quite old by then, requiring to
explicitly enable support for this in the kernel. We don't even know if
it still builds, let alone if it works at all on recent kernels! Let's
completely drop this now.
Since commit 244b070 ("MINOR: ssl/cli: support crt-list filters"),
HAProxy generates a list of filters based on the sni_ctx in memory.
However it's not always relevant, sometimes no filters were configured
and the CN/SAN in the new certificate are not the same.
This patch fixes the issue by using a flag filters in the ckch_inst, so
we are able to know if there were filters or not. In the late case it
uses the CN/SAN of the new certificate to generate the sni_ctx.
note: filters are still only used in the crt-list atm.
We currently have two UUID generation functions, one for the sample
fetch and the other one in the SPOE filter. Both were a bit complicated
since they were made to support random() implementations returning an
arbitrary number of bits, and were throwing away 33 bits every 64. Now
we don't need this anymore, so let's have a generic function consuming
64 bits at once and use it as appropriate.
The rand() sample fetch supports being limited to a certain range, but
it only uses 31 bits and scales them as requested, which means that when
the requested output range is larger than 31 bits, the least significant
one is not random and may even be constant.
Let's make use of the whole 32 bits now that we have access ot them.
In order to honor spread_checks we currently call rand() which is not
thread safe and which must never turn its internal state to zero. This
is not thread safe, let's use ha_random() instead. This is a complement
to commimt 52bf839394 ("BUG/MEDIUM: random: implement a thread-safe and
process-safe PRNG") and may be backported with it.
For the random LB algorithm we need a random 32-bit hashing key that used
to be made of two calls to random(). Now we can simply perform a single
call to ha_random32() and get rid of the useless operations.
This is the replacement of failed attempt to add thread safety and
per-process sequences of random numbers initally tried with commit
1c306aa84d ("BUG/MEDIUM: random: implement per-thread and per-process
random sequences").
This new version takes a completely different approach and doesn't try
to work around the horrible OS-specific and non-portable random API
anymore. Instead it implements "xoroshiro128**", a reputedly high
quality random number generator, which is one of the many variants of
xorshift, which passes all quality tests and which is described here:
http://prng.di.unimi.it/
While not cryptographically secure, it is fast and features a 2^128-1
period. It supports fast jumps allowing to cut the period into smaller
non-overlapping sequences, which we use here to support up to 2^32
processes each having their own, non-overlapping sequence of 2^96
numbers (~7*10^28). This is enough to provide 1 billion randoms per
second and per process for 2200 billion years.
The implementation was made thread-safe either by using a double 64-bit
CAS on platforms supporting it (x86_64, aarch64) or by using a local
lock for the time needed to perform the shift operations. This ensures
that all threads pick numbers from the same pool so that it is not
needed to assign per-thread ranges. For processes we use the fast jump
method to advance the sequence by 2^96 for each process.
Before this patch, the following config:
global
nbproc 8
frontend f
bind :4445
mode http
log stdout format raw daemon
log-format "%[uuid] %pid"
redirect location /
Would produce this output:
a4d0ad64-2645-4b74-b894-48acce0669af 12987
a4d0ad64-2645-4b74-b894-48acce0669af 12992
a4d0ad64-2645-4b74-b894-48acce0669af 12986
a4d0ad64-2645-4b74-b894-48acce0669af 12988
a4d0ad64-2645-4b74-b894-48acce0669af 12991
a4d0ad64-2645-4b74-b894-48acce0669af 12989
a4d0ad64-2645-4b74-b894-48acce0669af 12990
82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12987
82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12992
82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12986
(...)
And now produces:
f94b29b3-da74-4e03-a0c5-a532c635bad9 13011
47470c02-4862-4c33-80e7-a952899570e5 13014
86332123-539a-47bf-853f-8c8ea8b2a2b5 13013
8f9efa99-3143-47b2-83cf-d618c8dea711 13012
3cc0f5c7-d790-496b-8d39-bec77647af5b 13015
3ec64915-8f95-4374-9e66-e777dc8791e0 13009
0f9bf894-dcde-408c-b094-6e0bb3255452 13011
49c7bfde-3ffb-40e9-9a8d-8084d650ed8f 13014
e23f6f2e-35c5-4433-a294-b790ab902653 13012
There are multiple benefits to using this method. First, it doesn't
depend anymore on a non-portable API. Second it's thread safe. Third it
is fast and more proven than any hack we could attempt to try to work
around the deficiencies of the various implementations around.
This commit depends on previous patches "MINOR: tools: add 64-bit rotate
operators" and "BUG/MEDIUM: random: initialize the random pool a bit
better", all of which will need to be backported at least as far as
version 2.0. It doesn't require to backport the build fixes for circular
include files dependecy anymore.
This reverts commit 1c306aa84d.
It breaks the build on all non-glibc platforms. I got confused by the
man page (which possibly is the most confusing man page I've ever read
about a standard libc function) and mistakenly understood that random_r
was portable, especially since it appears in latest freebsd source as
well but not in released versions, and with a slightly different API :-/
We need to find a different solution with a fallback. Among the
possibilities, we may reintroduce this one with a fallback relying on
locking around the standard functions, keeping fingers crossed for no
other library function to call them in parallel, or we may also provide
our own PRNG, which is not necessarily more difficult than working
around the totally broken up design of the portable API.
As mentioned in previous patch, the random number generator was never
made thread-safe, which used not to be a problem for health checks
spreading, until the uuid sample fetch function appeared. Currently
it is possible for two threads or processes to produce exactly the
same UUID. In fact it's extremely likely that this will happen for
processes, as can be seen with this config:
global
nbproc 8
frontend f
bind :4445
mode http
log stdout daemon format raw
log-format "%[uuid] %pid"
redirect location /
It typically produces this log:
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30645
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30641
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30644
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30639
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30646
07764439-c24d-4e6f-a5a6-0138be59e7a8 30645
07764439-c24d-4e6f-a5a6-0138be59e7a8 30639
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30643
07764439-c24d-4e6f-a5a6-0138be59e7a8 30646
b6773fdd-678f-4d04-96f2-4fb11ad15d6b 30646
551ce567-0bfb-4bbd-9b58-cdc7e9365325 30642
07764439-c24d-4e6f-a5a6-0138be59e7a8 30642
What this patch does is to use a distinct per-thread and per-process
seed to make sure the same sequences will not appear, and will then
extend these seeds by "burning" a number of randoms that depends on
the global random seed, the thread ID and the process ID. This adds
roughly 20 extra bits of randomness, resulting in 52 bits total per
thread and per process.
It only takes a few milliseconds to burn these randoms and given
that threads start with a different seed, we know they will not
catch each other. So these random extra bits are essentially added
to ensure randomness between boots and cluster instances.
This replaces all uses of random() with ha_random() which uses the
thread-local state.
This must be backported as far as 2.0 or any version having the
UUID sample-fetch function since it's the main victim here.
It's important to note that this patch, in addition to depending on
the previous one "BUG/MEDIUM: init: initialize the random pool a bit
better", also depends on the preceeding build fixes to address a
circular dependency issue in the include files that prevented it
from building. Part or all of these patches may need to be backported
or adapted as well.
Since the UUID sample fetch was created, some people noticed that in
certain virtualized environments they manage to get exact same UUIDs
on different instances started exactly at the same moment. It turns
out that the randoms were only initialized to spread the health checks
originally, not to provide "clean" randoms.
This patch changes this and collects more randomness from various
sources, including existing randoms, /dev/urandom when available,
RAND_bytes() when OpenSSL is available, as well as the timing for such
operations, then applies a SHA1 on all this to keep a 160 bits random
seed available, 32 of which are passed to srandom().
It's worth mentioning that there's no clean way to pass more than 32
bits to srandom() as even initstate() provides an opaque state that
must absolutely not be tampered with since known implementations
contain state information.
At least this allows to have up to 4 billion different sequences
from the boot, which is not that bad.
Note that the thread safety was still not addressed, which is another
issue for another patch.
This must be backported to all versions containing the UUID sample
fetch function, i.e. as far as 2.0.
In the same way than for the request, when a redirect rule is applied the
transction is aborted. This must be done returning HTTP_RULE_RES_ABRT from
http_res_get_intercept_rule() function.
No backport needed because on previous versions, the action return values are
not handled the same way.
Backend counters must be incremented only if a backend was already assigned to
the stream (when the stream exists). Otherwise, it means we are still on the
frontend side.
This patch may be backported as far as 1.6.
When an action interrupts a transaction, returning a response or not, it must
return the ACT_RET_ABRT value and not ACT_RET_STOP. ACT_RET_STOP is reserved to
stop the processing of the current ruleset.
No backport needed because on previous versions, the action return values are
not handled the same way.
When at least a filter is attached to a stream, FLT_END analyzers must be
preserved on request and response channels.
This patch should be backported as far as 1.7.
Since the HTX mode is the only mode to process HTTP messages, the stream is
created for a uniq transaction. The keep-alive is handled at the mux level. So,
the compression filter can be initialized when the stream is created and
released with the stream. Concretly, .channel_start_analyze and
.channel_end_analyze callback functions are replaced by .attach and .detach
ones.
With this change, it is no longer necessary to call FLT_START_FE/BE and FLT_END
analysers for the compression filter.
Since the HTX mode is the only mode to process HTTP messages, the stream is
created for a uniq transaction. The keep-alive is handled at the mux level. So,
the cache filter can be initialized when the stream is created and released with
the stream. Concretly, .channel_start_analyze and .channel_end_analyze callback
functions are replaced by .attach and .detach ones.
With this change, it is no longer necessary to call FLT_START_FE/BE and FLT_END
analysers for the cache filter.
A typo was introduced by the commit c5bb5a0f2 ("BUG/MINOR: http-rules: Preserve
FLT_END analyzers on reject action").
This patch must be backported with the commit c5bb5a0f2.
When at least a filter is attached to a stream, FLT_END analyzers must be
preserved on request and response channels.
This patch should be backported as far as 1.8.
When an action interrupts a transaction, returning a response or not, it must
return the ACT_RET_ABRT value and not ACT_RET_DONE. ACT_RET_DONE is reserved to
stop the processing on the current channel but some analysers may still be
active. When ACT_RET_ABRT is returned, all analysers are removed, except FLT_END
if it is set.
No backport needed because on previous verions, the action return value was not
handled the same way.
It is stated in the comment the return action returns ACT_RET_ABRT on
success. It it the right code to use to abort a transaction. ACT_RET_DONE must
be used when the message processing must be stopped. This does not means the
transaction is interrupted.
No backport needed.
The wake_time of a lua context is now always set to TICK_ETERNITY when the
context is initialized and when everytime the execution of the lua stack is
started. It is mandatory to not set arbitrary wake_time when an action yields.
No backport needed.
This flag was used in some internal functions to be sure the current stream is
able to handle HTTP content. It was introduced when the legacy HTTP code was
still there. Now, It is possible to rely on stream's flags to be sure we have an
HTX stream.
So the flag HLUA_TXN_HTTP_RDY can be removed. Everywhere it was tested, it is
replaced by a call to the IS_HTX_STRM() macro.
This patch is mandatory to allow the support of the filters written in lua.
In these functions, the lua txn was not used. So it can be removed from the
function argument list.
This patch is mandatory to allow the support of the filters written in lua.
In this function, the lua txn was only used to retrieve the stream. But it can
be retieve from the HTTP message, using its channel pointer. So, the lua txn can
be removed from the function argument list.
This patch is mandatory to allow the support of the filters written in lua.
In this function, the lua txn was only used to test if the HTTP transaction is
defined. But it is always used in a context where it is true. So, the lua txn
can be removed from the function argument list.
This patch is mandatory to allow the support of the filters written in lua.
The Lua function Channel.is_full() should not take care of the reserve because
it is not called from a producer (an applet for instance). From an action, it is
allowed to overwrite the buffer reserve.
This patch should be backported as far as 1.7. But it must be adapted for 1.8
and lower because there is no HTX on these versions.
When a lua action aborts a transaction calling txn:done() function, the action
must return ACT_RET_ABRT instead of ACT_RET_DONE. It is mandatory to
abort the message analysis.
This patch must be backported everywhere the commit 7716cdf45 ("MINOR: lua: Get
the action return code on the stack when an action finishes") was
backported. For now, no backport needed.
When an error occurred on the response side, request analysers must be reset. At
this stage, only AN_REQ_HTTP_XFER_BODY analyser remains, and possibly
AN_REQ_FLT_END, if at least one filter is attached to the stream. So it is safe
to remove the AN_REQ_HTTP_XFER_BODY analyser. An error was already handled and a
response was already returned to the client (or it was at least scheduled to be
sent). So there is no reason to continue to process the request payload. It may
cause some troubles for the filters because when an error occurred, data from
the request buffer are truncated.
This patch must be backported as far as 1.9, for the HTX part only. I don't know
if the legacy HTTP code is affected.
During the payload filtering, the offset is relative to the head of the HTX
message and not its first index. This index is the position of the first block
to (re)start the HTTP analysis. It must be used during HTTP analysis but not
during the payload forwarding.
So, from the compression filter point of view, when we loop on the HTX blocks to
compress the response payload, we must start from the head of the HTX
message. To ease the loop, we use the function htx_find_offset().
This patch must be backported as far as 2.0. It depends on the commit "MINOR:
htx: Add a function to return a block at a specific an offset". So this one must
be backported first.
During the payload filtering, the offset is relative to the head of the HTX
message and not its first index. This index is the position of the first block
to (re)start the HTTP analysis. It must be used during HTTP analysis but not
during the payload forwarding.
So, from the cache point of view, when we loop on the HTX blocks to cache the
response payload, we must start from the head of the HTX message. To ease the
loop, we use the function htx_find_offset().
This patch must be backported as far as 2.0. It depends on the commit "MINOR:
htx: Add a function to return a block at a specific an offset". So this one must
be backported first.
If a filter enable the data filtering, in TCP or in HTTP, but it does not
defined the corresponding callback function (so http_payload() or
tcp_payload()), it will be ignored. If all configured data filter do the same,
we must be sure to forward everything. Otherwise nothing will be forwarded at
all.
This patch must be forwarded as far as 1.9.
When the tcp or http payload is filtered, it is important to use the filter
offset to decude the amount of forwarded data because this offset may change
during the call to the callback function. So we should not rely on a local
variable defined before this call.
For now, existing HAproxy filters don't change this offset, so this bug may only
affect external filters.
This patch must be forwarded as far as 1.9.
The htx_find_offset() function may be used to look for a block at a specific
offset in an HTX message, starting from the message head. A compound result is
returned, an htx_ret structure, with the found block and the position of the
offset in the block. If the offset is ouside of the HTX message, the returned
block is NULL.
This patch fixes PROXYv2 parsing when the payload of the TCP connection is
fused with the PROXYv2 header within a single recv() call.
Previously HAProxy ignored the PROXYv2 header length when attempting to
parse the TLV, possibly interpreting the first byte of the payload as a
TLV type.
This patch adds proper validation. It ensures that:
1. TLV parsing stops when the end of the PROXYv2 header is reached.
2. TLV lengths cannot exceed the PROXYv2 header length.
3. The PROXYv2 header ends together with the last TLV, not allowing for
"stray bytes" to be ignored.
A reg-test was added to ensure proper behavior.
This patch tries to find the sweat spot between a small and easily
backportable one, and a cleaner one that's more easily adaptable to
older versions, hence why it merges the "if" and "while" blocks which
causes a reindent of the whole block. It should be used as-is for
versions 1.9 to 2.1, the block about PP2_TYPE_AUTHORITY should be
dropped for 2.0 and the block about CRC32C should be dropped for 1.8.
This bug was introduced when TLV parsing was added. This happened in commit
b3e54fe387. This commit was first released
with HAProxy 1.6-dev1.
A similar issue was fixed in commit 7209c204bd.
This patch must be backported to HAProxy 1.6+.
James Stroehmann reported something working as documented but that can
be considered as a regression in the way the automatic maxconn is
calculated from the process' limits :
https://www.mail-archive.com/haproxy@formilux.org/msg36523.html
The purpose of the changes in 2.0 was to have maxconn default to the
highest possible value permitted to the user based on the ulimit -n
setting, however the calculation starts from the soft limit, which
can be lower than what users were allowed to with previous versions
where the default value of 2000 would force a higher ulimit -n as
long as it fitted in the hard limit.
Usually this is not noticeable if the user changes the limits, because
quite commonly setting a new value restricts both the soft and hard
values.
Let's instead always use the max between the hard and soft limits, as
we know these values are permitted. This was tried on the following
setup:
$ cat ulimit-n.cfg
global
stats socket /tmp/sock1 level admin
$ ulimit -n
1024
Before the change the limits would show like this:
$ socat - /tmp/sock1 <<< "show info" | grep -im2 ^Max
Maxsock: 1023
Maxconn: 489
After the change the limits are now much better and more in line with
the default settings in earlier versions:
$ socat - /tmp/sock1 <<< "show info" | grep -im2 ^Max
Maxsock: 4095
Maxconn: 2025
The difference becomes even more obvious when running moderately large
configs with hundreds of checked servers and hundreds of listeners:
$ cat ulimit-n.cfg
global
stats socket /tmp/sock1 level admin
listen l
bind :10000-10300
server-template srv- 300 0.0.0.0 check disabled
Before After
Maxsock 1024 4096
Maxconn 189 1725
This issue is tagged as minor since a trivial config change fixes it,
but it would help new users to have it backported as far as 2.0.
pattern_finalize_config() uses an inefficient algorithm which is a
problem with very large configuration files. This affects startup, and
therefore reload time. When haproxy is deployed as a router in a
Kubernetes cluster the generated configuration file may be large and
reloads are frequently occuring, which makes this a significant issue.
The old algorithm is O(n^2)
* allocate missing uids - O(n^2)
* sort linked list - O(n^2)
The new algorithm is O(n log n):
* find the user allocated uids - O(n)
* store them for efficient lookup - O(n log n)
* allocate missing uids - n times O(log n)
* sort all uids - O(n log n)
* convert back to linked list - O(n)
Performance examples, startup time in seconds:
pat_refs old new
1000 0.02 0.01
10000 2.1 0.04
20000 12.3 0.07
30000 27.9 0.10
40000 52.5 0.14
50000 77.5 0.17
Please backport to 1.8, 2.0 and 2.1.
This command is used to produce an arbitrary amount of data on the
output. It can be used to test the CLI's state machine as well as
the internal parts related to applets an I/O. A typical test consists
in asking for all sizes from 0 to 16384:
$ (echo "prompt;expert-mode on";for i in {0..16384}; do
echo "debug dev write $i"; done) | socat - /tmp/sock1 | wc -c
134258738
A better test would consist in first waiting for the response before
sending a new request.
This command is not restricted to the admin since it's harmless.
There's a build error reported here:
c9c6cdbf9c/checks
It's just caused by an inconditional assignment of tmp_filter to
*sni_filter without having been initialized, though it's harmless because
this return pointer is not used when fcount is NULL, which is the only
case where this happens.
No backport is needed as this was brought today by commit 38df1c8006
("MINOR: ssl/cli: support crt-list filters").
It was only possible to go down from the ckch_store to the sni_ctx but
not to go up from the sni_ctx to the ckch_store.
To allow that, 2 pointers were added:
- a ckch_inst pointer in the struct sni_ctx
- a ckckh_store pointer in the struct ckch_inst
Generate a list of the previous filters when updating a certificate
which use filters in crt-list. Then pass this list to the function
generating the sni_ctx during the commit.
This feature allows the update of the crt-list certificates which uses
the filters with "set ssl cert".
This function could be probably replaced by creating a new
ckch_inst_new_load_store() function which take the previous sni_ctx list as
an argument instead of the char **sni_filter, avoiding the
allocation/copy during runtime for each filter. But since are still
handling the multi-cert bundles, it's better this way to avoid code
duplication.
When building with DEBUG_STRICT, there are still some BUG_ON(events&event_type)
in the subscribe() code which are not welcome anymore since we explicitly
permit to wake the caller up on readiness. This causes some regtests to fail
since 2c1f37d353 ("OPTIM: mux-h1: subscribe rather than waking up at a few
other places") when built with this option.
No backport is needed, this is 2.2-dev.
This is another round of conversion from a blind tasklet_wakeup() to
a more careful subscribe(). It has significantly improved the number
of function calls per HTTP request (/?s=1k/t=20) :
before after
tasklet_wakeup: 3 2
conn_subscribe: 3 2
h1_iocb: 3 2
h1_process: 3 2
h1_parse_msg_hdrs: 4 3
h1_rcv_buf: 5 3
h1_send: 5 4
h1_subscribe: 2 1
h1_wake_stream_for_send: 5 4
http_wait_for_request: 2 1
process_stream: 3 2
si_cs_io_cb: 4 2
si_cs_process: 3 1
si_cs_rcv: 5 3
si_sync_send: 2 1
si_update_both: 2 1
stream_int_chk_rcv_conn: 3 2
stream_int_notify: 3 1
stream_release_buffers: 9 4
In order to save a lot on syscalls, we currently don't disable receiving
on a file descriptor anymore if its handler was already woken up. But if
the run queue is huge and the poller collects a lot of events, this causes
excess wakeups which take CPU time which is not used to flush these tasklets.
This patch simply considers the run queue size to decide whether or not to
stop receiving. Tests show that by stopping receiving when the run queue
reaches ~16 times its configured size, we can still hold maximal performance
in extreme situations like maxpollevents=20k for runqueue_depth=2, and still
totally avoid calling epoll_event under moderate load using default settings
on keep-alive connections.
We used to call ->wake() for any I/O event for which there was no
subscriber. But this is a problem because this causes massive wake()
storms since we disabled fd_stop_recv() to save syscalls.
The only reason for the io_available condition is to detect that an
asynchronous connect() just finished and will not be handled by any
registered event handler. Since we now properly handle synchronous
connects, we can detect this situation by the fact that we had a
success on conn_fd_check() and no requested I/O took over.
In the rare case of immediate connect() (unix sockets, socket pairs, and
occasionally TCP over the loopback), it is counter-productive to subscribe
for sending and then getting immediately back to process_stream() after
having passed through si_cs_process() just to update the connection. We
already know it is established and it doesn't have any handshake anymore
so we just have to complete it and return to process_stream() with the
stream_interface in the SI_ST_RDY state. In this case, process_stream will
simply loop back to the beginning to synchronize the state and turn it to
SI_ST_EST/ASS/CLO/TAR etc.
This will save us from having to needlessly subscribe in the connect()
code, something which in addition cannot work with edge-triggered pollers.
With commit 065a025610 ("MEDIUM: connection: don't stop receiving events
in the FD handler") we disabled a number of fd_stop_* in conn_fd_handler(),
in order to wait for their respective handlers to deal with them. But it
is not correct to do that for the send direction, as we may very well
have nothing to send. This is visible when connecting in TCP mode to
a server with no data to send, there's nobody anymore to disable the
polling for the send direction.
And it is logical, on the recv() path we know the system has data to
deliver and that some code will be in charge of it. On the send
direction we simply don't know if it was the result of a successful
connect() or if there is still something to send. In any case we
almost never fill the network buffer on a single send() after being
woken up by the system, so disabling the FD immediately or much later
will not change the number of operations.
No backport is needed, this is 2.2-dev.
The approach was wrong. USE_DL is for the makefile to know if it's required
to append -ldl at link time. Some platforms do not need it (and in fact do
not have it) yet they have a working dladdr(). The real condition is related
to ELF. Given that due to Lua, all platforms that require -ldl already have
USE_DL set, let's replace USE_DL with __ELF__ here and consider that dladdr
is always needed on ELF, which basically is already the case.
resolve_sym_name() doesn't build when USE_DL is set on non-GNU platforms
because "Elf(W)" isn't defined. Since it's only used for dladdr1(), let's
refactor all this so that we can completely ifdef out that part on other
platforms. Now we have a separate function to perform the call depending
on the platform and it only returns the size when available.
Instead of special-casing the use of the symbol resolving to decide
whether to dump a partial or complete trace, let's simply start over
and dump everything when we reach the end after having found nothing.
It will be more robust against dirty traces as well.
It happens that on aarch64 backtrace() only returns one entry (tested
with gcc 4.7.4, 5.5.0 and 7.4.1). Probably that it refrains from unwinding
the stack due to the risk of hitting a bad pointer. Here we can use
may_access() to know when it's safe, so we can actually unwind the stack
without taking risks. It happens that the faulting function (the one
just after the signal handler) is not listed here, very likely because
the signal handler uses a special stack and did not create a new frame.
So this patch creates a new my_backtrace() function in standard.h that
either calls backtrace() or does its own unrolling. The choice depends
on HA_HAVE_WORKING_BACKTRACE which is set in compat.h based on the build
target.
It's useful to get an indication of unresolved stuff or memory
corruption to have the apparent depth of the stack trace in the
output, especially if we dump nothing.
At least FreeBSD has a fully functional CLOCK_THREAD_CPUTIME but it
cannot create a timer on it. This is not a problem since our timer is
only used to measure each thread's usage using now_cpu_time_thread().
So by just replacing this clock with CLOCK_REALTIME we allow such
platforms to periodically call the wdt and check the thread's CPU usage.
The consequence is that even on a totally idle system there will still
be a few extra periodic wakeups, but the watchdog becomes usable there
as well.
On operating systems not supporting to create a timer on
POSIX_THREAD_CPUTIME we emit a warning but we return an error so the
process fails to start, which is absurd. Let's return a success once
the warning is emitted instead.
This may be backported to 2.1 and 2.0.
It's only available for bind line. "ca-verify-file" allows to separate
CA certificates from "ca-file". CA names sent in server hello message is
only compute from "ca-file". Typically, "ca-file" must be defined with
intermediate certificates and "ca-verify-file" with certificates to
ending the chain, like root CA.
Fix issue #404.
Calling backtrace() will access libgcc at runtime. We don't want to do
it after the chroot, so let's perform a first call to have it ready in
memory for later use.
When a panic() occurs due to a stuck thread, we'll try to dump a
backtrace of this thread if the config directive USE_BACKTRACE is
set (which is the case on linux+glibc). For this we use the
backtrace() call provided by glibc and iterate the pointers through
resolve_sym_name(). In order to minimize the output (which is limited
to one buffer), we only do this for stuck threads, and we start the
dump above ha_panic()/ha_thread_dump_all_to_trash(), and stop when
meeting known points such as main/run_tasks_from_list/run_poll_loop.
If enabled without USE_DL, the dump will be complete with no details
except that pointers will all be given relative to main, which is
still better than nothing.
The new USE_BACKTRACE config option is enabled by default on glibc since
it has been present for ages. When it is set, the export-dynamic linker
option is enabled so that all non-static symbols are properly resolved.
Now in "show threads", the task/tasklet handler will be resolved
using this function, which will provide more detailed results and
will still support offsets to main for unresolved symbols.
We use various hacks at a few places to try to identify known function
pointers in debugging outputs (show threads & show fd). Let's centralize
this into a new function dedicated to this. It already knows about the
functions matched by "show threads" and "show fd", and when built with
USE_DL, it can rely on dladdr1() to resolve other functions. There are
some limitations, as static functions are not resolved, linking with
-rdynamic is mandatory, and even then some functions will not necessarily
appear. It's possible to do a better job by rebuilding the whole symbol
table from the ELF headers in memory but it's less portable and the gains
are still limited, so this solution remains a reasonable tradeoff.
This function dumps <n> bytes from <addr> in hex form into buffer <buf>
enclosed in brackets after the address itself, formatted on 14 chars
including the "0x" prefix. This is meant to be used as a prefix for code
areas. For example: "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]: "
It relies on may_access() to know if the bytes are dumpable, otherwise "--"
is emitted. An optional prefix is supported.
Since commit 4c2ae48375 ("MINOR: trace: implement a very basic trace()
function") merged in 2.1, trace() is an inline function. It must not
appear in standard.c anymore and may break build depending on includes.
This can be backported to 2.1.
It happens that just sending the debug signal to the process makes on
thread wait for its turn while nobody wants to dump. We need to at
least verify that a dump was really requested for this thread.
This can be backported to 2.1 and 2.0.
Often in crash dumps we see unknown function pointers. Let's display
them relative to main, that helps quite a lot figure the function
from an executable, for example:
(gdb) x/a main+645360
0x4c56a0 <h1_timeout_task>: 0x2e6666666666feeb
This could be backported to 2.0.
In commit 4eee38a ("BUILD/MINOR: tools: fix build warning in the date
conversion functions") we added some return checks to shut build
warnings but the last test is useless since the tested pointer is not
updated by the last call to utoa_pad() used to convert the milliseconds.
It turns out the original code from 2012 already skipped this part,
probably in order to avoid the risk of seeing a millisecond field not
belonging to the 0-999 range. Better keep the check and put the code
into stricter shape.
No backport is needed. This fixes issue #526.
Commit 80b53ffb1c ("MEDIUM: arg: make make_arg_list() stop after its
own arguments") changed the way we detect the empty list because we
cannot stop by looking up the closing parenthesis anymore, thus for
the first missing arg we have to enter the parsing loop again. And
there, finding an empty arg means we go to the empty_err label, where
it was not initially planned to handle this condition. This results
in %[date()] to fail while %[date] works. Let's simply check if we've
reached the minimally supported args there (it used to be done during
the function entry).
Thanks to Jrme for reporting this issue. No backport is needed,
this is 2.2-dev2+ only.
Since commit "MEDIUM: connection: make the subscribe() call able to wakeup
if ready" we have the guarantee that the tasklet will be woken up if
subscribing to a connection for an even that's ready. Since we have too
many tasklet_wakeup() calls in mux-h1, let's now use this property to
improve the situation a bit.
With this change, no syscall count changed, however the number of useless
calls to some functions significantly went down. Here are the differences
for the test below (100k req), in number of calls per request :
$ ./h1load -n 100000 -t 4 -c 1000 -T 20 -F 127.0.0.1:8001/?s=1k/t=20
before after change note
tasklet_wakeup: 3 1 -66%
h1_io_cb: 4 3 -25%
h1_send: 6.7 5.4 -19%
h1_wake: 0.73 0.44 -39%
h1_process: 4.7 3.4 -27%
h1_wake_stream_for_send: 6.7 5.5 -18%
si_cs_process 3.7 3.4 -7.8%
conn_fd_handler 2.7 2.4 -10%
raw_sock_to_buf: 4 2 -50%
pool_free: 4 2 -50% from failed rx calls
Note that the situation could be further improved by having muxes lazily
subscribe to Rx events in case the FD is already being polled. However
this requires deeper changes to implement a LAZY_RECV subscribe mode,
and to replace the FD's active bit by 3 states representing the desired
action to perform on the FD during the update, among NONE (no need to
change), POLL (can't proceed without), and STOP (buffer full). This
would only impact Rx since on Tx we know what we have to send. The
savings to expect from this might be more visible with splicing and/or
when dealing with many connections having long think times.
The remaining epoll_ctl() calls are exclusively caused by the disagreement
between conn_fd_handler() and the mux receiving the data: the fd handler
wants to stop after having woken up the tasklet, then the mux after
receiving data wants to receive again. Given that they don't happen in
the same poll loop when there are many FDs, this causes a lot of state
changes.
As suggested by Olivier, if the task is already scheduled for running,
we don't need to disable the event because it's in the run queue, poll()
cannot stop, and reporting it again will be harmless. What *might*
happen however is that a sampling-based poller like epoll() would report
many times the same event and has trouble getting others behind. But if
it would happen, it would still indicate the run queue has plenty of
pending operations, so it would in fact only displace the problem from
the poller to the run queue, which doesn't seem to be worse (and in
fact we do support priorities while the poller does not).
By doing this change, the keep-alive test with 1k conns and 100k reqs
completely gets rid of the per-request epoll_ctl changes, while still
not causing extra recvfrom() :
$ ./h1load -n 100000 -t 4 -c 1000 -T 20 -F 127.0.0.1:8001/?s=1k/t=20
200000 sendto 1
200000 recvfrom 1
10762 epoll_wait 1
3664 epoll_ctl 1
1999 recvfrom -1
In close mode, it didn't change anything, we're still in the optimal
case (2 epoll per connection) :
$ ./h1load -n 100000 -r 1 -t 4 -c 1000 -T 20 -F 127.0.0.1:8001/?s=1k/t=20
203764 epoll_ctl 1
200000 sendto 1
200000 recvfrom 1
6091 epoll_wait 1
2994 recvfrom -1
There's currently an internal API limitation at the connection layer
regarding conn_subscribe(). We must not subscribe if we haven't yet
met EAGAIN or such a condition, so we sometimes force ourselves to read
in order to meet this condition and being allowed to call subscribe.
But reading cannot always be done (e.g. at the end of a loop where we
cannot afford to retrieve new data and start again) so we instead
perform a tasklet_wakeup() of the requester's io_cb. This is what is
done in mux_h1 for example. The problem with this is that it forces
a new receive when we're not necessarily certain we need one. And if
the FD is not ready and was already being polled, it's a useless
wakeup.
The current patch improves the connection-level subscribe() so that
it really manipulates the polling if the FD is marked not-ready, but
instead schedules the argument tasklet for a wakeup if the FD is
ready. This guarantees we'll wake this tasklet up in any case once the
FD is ready, either immediately or after polling.
By doing so, a test on pure close mode shows we cut in half the number
of epoll_ctl() calls and almost eliminate failed recvfrom():
$ ./h1load -n 100000 -r 1 -t 4 -c 1000 -T 20 -F 127.0.0.1:8001/?s=1k/t=20
before:
399464 epoll_ctl 1
200007 recvfrom 1
200000 sendto 1
100000 recvfrom -1
7508 epoll_wait 1
after:
205739 epoll_ctl 1
200000 sendto 1
200000 recvfrom 1
6084 epoll_wait 1
2651 recvfrom -1
On keep-alive there is no change however.
This partially reverts commit 1113116b4a ("MEDIUM: raw-sock: remove
obsolete calls to fd_{cant,cond,done}_{send,recv}") so that we can mark
the FD not ready as required since commit 19bc201c9f ("MEDIUM: connection:
remove the intermediary polling state from the connection"). Indeed, with
the removal of the latter we don't have any other reliable indication that
the FD is blocked, which explains why there are so many EAGAIN in traces.
It's worth noting that a short read or a short write are also reliable
indicators of exhausted buffers and are even documented as such in the
epoll man page in case of edge-triggered mode. That's why we also report
the FD as blocked in such a case.
With this change we completely got rid of EAGAIN in keep-alive tests, but
they were expectedly transferred to epoll_ctl:
$ ./h1load -n 100000 -t 4 -c 1000 -T 20 -F 127.0.0.1:8001/?s=1k/t=20
before:
266331 epoll_ctl 1
200000 sendto 1
200000 recvfrom 1
135757 recvfrom -1
8626 epoll_wait 1
after:
394865 epoll_ctl 1
200000 sendto 1
200000 recvfrom 1
10748 epoll_wait 1
1999 recvfrom -1
When a header is added or modified, in http_add_header() or
http_replace_header(), a comparison is performed on its name to know if it is
the Host header and if the authority part of the uri must be updated or
not. This comparision must be case-insensive.
This patch should fix the issue #522. It must be backported to 2.1.
As per issue #435 a hostname with a trailing dot confuses our DNS code,
as for a zero length DNS label we emit a null-byte. This change makes us
ignore the zero length label instead.
Must be backported to 1.8.
Even when there isn't a chain, it must be initialized to a empty X509
structure with sk_X509_new_null().
This patch fixes a segfault which appears with older versions of the SSL
libs (openssl 0.9.8, libressl 2.8.3...) because X509_chain_up_ref() does
not check the pointer.
This bug was introduced by b90d2cb ("MINOR: ssl: resolve issuers chain
later").
Should fix issue #516.
Previously when the `unique-id-format` contained non-deterministic parts,
such as the `uuid` fetch each use of the `unique-id` fetch would generate
a new unique ID, replacing the old one. The following configuration shows
the error:
global
log stdout format short daemon
listen test
log global
log-format "%ID"
unique-id-format %{+X}o\ TEST-%[uuid]
mode http
bind *:8080
http-response set-header A %[unique-id]
http-response set-header B %[unique-id]
server example example.com:80
Without the patch the contents of the `A` and `B` response header would
differ.
This bug was introduced in commit f4011ddcf5,
which was first released with HAProxy 1.7-dev3.
This fix should be backported to HAProxy 1.7+.
valgrind complains that epoll_ctl() uses an epoll_event in which we
have only set the part we use from the data field (i.e. the fd). Tests
show that pre-initializing the struct in the stack doesn't have a
measurable impact so let's do it.
In issue #471 it was reported that valgrind sometimes complains about
timer_create() being called with uninitialized bytes. These are in fact
the bits from sigev_value.sival_ptr that are not part of sival_int that
are tagged as such, as valgrind has no way to know we're using the int
instead of the ptr in the union. It's cheap to initialize the field so
let's do it.
Since commit 92919f7fd5 ("MEDIUM: h2: make the request parser rebuild
a complete URI") we make sure to rebuild a complete URI. Unfortunately
the test for an empty :path pseudo-header that is mandated by #8.1.2.3
appened to be performed on the URI before this patch, which is never
empty anymore after being rebuilt, causing h2spec to complain :
8. HTTP Message Exchanges
8.1. HTTP Request/Response Exchange
8.1.2. HTTP Header Fields
8.1.2.3. Request Pseudo-Header Fields
- 1: Sends a HEADERS frame with empty ":path" pseudo-header field
-> The endpoint MUST respond with a stream error of type PROTOCOL_ERROR.
Expected: GOAWAY Frame (Error Code: PROTOCOL_ERROR)
RST_STREAM Frame (Error Code: PROTOCOL_ERROR)
Connection closed
Actual: DATA Frame (length:0, flags:0x01, stream_id:1)
It's worth noting that this error doesn't trigger when calling h2spec
with a timeout as some scripts do, which explains why it wasn't detected
after the patch above.
This fixes one half of issue #471 and should be backported to 2.1.
The goal is to use the ckch to store data from PEM files or <payload> and
only for that. This patch adresses the ckch->ocsp_issuer case. It finds
issuers chain if no chain is present in the ckch in ssl_sock_put_ckch_into_ctx(),
filling the ocsp_issuer from the chain must be done after.
It changes the way '.issuer' is managed: it tries to load '.issuer' in
ckch->ocsp_issuer first and then look for the issuer in the chain later
(in ssl_sock_load_ocsp() ). "ssl-load-extra-files" without the "issuer"
parameter can negate extra '.issuer' file check.
The goal is to use the ckch to store data from a loaded PEM file or a
<payload> and only for that. This patch addresses the ckch->chain case.
Looking for the issuers chain, if no chain is present in the ckch, can
be done in ssl_sock_put_ckch_into_ctx(). This way it is possible to know
the origin of the certificate chain without an extra state.
This lock was only needed to protect the buffer_wq list, but now we have
the mt_list for this. This patch simply turns the buffer_wq list to an
mt_list and gets rid of the lock.
It's worth noting that the whole buffer_wait thing still looks totally
wrong especially in a threaded context: the wakeup_cb() callback is
called synchronously from any thread and may end up calling some
connection code that was not expected to run on a given thread. The
whole thing should probably be reworked to use tasklets instead and be
a bit more centralized.
Move the cert_issuer_tree outside the global_ssl structure since it's
not a configuration variable. And move the declaration of the
issuer_chain structure in types/ssl_sock.h
Reorder the 'show ssl cert' output so it's easier to see if the whole
chain is correct.
For a chain to be correct, an "Issuer" line must have the same
content as the next "Subject" line.
Example:
Subject: /C=FR/ST=Paris/O=HAProxy Test Certificate/CN=test.haproxy.local
Issuer: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 2/CN=ca2.haproxy.local
Chain Subject: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 2/CN=ca2.haproxy.local
Chain Issuer: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 1/CN=ca1.haproxy.local
Chain Subject: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 1/CN=ca1.haproxy.local
Chain Issuer: /C=FR/ST=Paris/O=HAProxy Test Root CA/CN=root.haproxy.local
For each certificate in the chain, displays the issuer, so it's easy to
know if the chain is right.
Also rename "Chain" to "Chain Subject".
Example:
Chain Subject: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 2/CN=ca2.haproxy.local
Chain Issuer: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 1/CN=ca1.haproxy.local
Chain Subject: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 1/CN=ca1.haproxy.local
Chain Issuer: /C=FR/ST=Paris/O=HAProxy Test Root CA/CN=root.haproxy.local
Display the subject of each certificate contained in the chain in the
output of "show ssl cert <filename>".
Each subjects are on a unique line prefixed by "Chain: "
Example:
Chain: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 2/CN=ca2.haproxy.local
Chain: /C=FR/ST=Paris/O=HAProxy Test Intermediate CA 1/CN=ca1.haproxy.local
This directive has never made any sense and has already caused trouble
by forcing the process to stay in foreground during the boot process.
Let's emit a warning mentioning it's deprecated and will be removed in
2.3.
As reported in issue #511, when sending an outgoing local connection
(e.g. health check) we must set the "local" tag and not a "proxy" tag.
The issue comes from historic support on v1 which required to steal the
address on the outgoing connection for such ones, creating confusion in
the v2 code which believes it sees the incoming connection.
In order not to risk to break existing setups which might rely on seeing
the LB's address in the connection's source field, let's just change the
connection type from proxy to local and keep the addresses. The protocol
spec states that for local, the addresses must be ignored anyway.
This problem has always existed, this can be backported as far as 1.5,
though it's probably not a good idea to change such setups, thus maybe
2.0 would be more reasonable.
There were 8 strict aliasing warnings there due to the dereferences
casting to uint32_t of input and output. We can achieve the same using
two write_u64() and four read_u64() which do not cause this issue and
even let the compiler use 64-bit operations.
Enabling strict aliasing fails on the cache's hash which is a series of
20 bytes cast as u32. And in practice it could even fail on some archs
if the http_txn didn't guarantee the hash was properly aligned. Let's
use read_u32() to read the value and write_u32() to set it, this makes
sure the compiler emits the correct code to access these and knows about
the intentional aliasing.
Enabling strict aliasing fails in fd.c when using the double-word CAS,
let's get rid of the (void**)(void*)&cur_list junk and use a union
instead. This way the compiler knows they do alias.
Sample fetch functions ssl_x_sha1(), ssl_fc_npn(), ssl_fc_alpn(),
ssl_fc_session_id(), as well as the CLI's "show cert details" handler
used to dereference the output buffer's <data> field by casting it to
"unsigned *". But while doing this could work before 1.9, it broke
starting with commit 843b7cbe9d ("MEDIUM: chunks: make the chunk struct's
fields match the buffer struct") which merged chunks and buffers, causing
the <data> field to become a size_t. The impact is only on 64-bit platform
and depends on the endianness: on little endian, there should never be any
non-zero bits in the field as it is supposed to have been zeroed before the
call, so it shouldbe harmless; on big endian, the high part of the value
only is written instead of the lower one, often making the result appear
4 billion times larger, and making such values dropped everywhere due to
being larger than a buffer.
It seems that it would be wise to try to re-enable strict-aliasing to
catch such errors.
This must be backported till 1.9.
About every time there's a pointer cast in the code, there's a hidden
bug, and this one was no exception, as it passes the first octet of the
native representation of an integer as a single-character string, which
obviously only works on little endian machines. On big-endian machines,
something as simple as "str(foo),json" only returns zeroes.
This bug was introduced with the JSON converter in 1.6-dev1 by commit
317e1c4f1e ("MINOR: sample: add "json" converter"), the fix may be
backported to all stable branches.
The isalnum(), isalpha(), isdigit() etc functions from ctype.h are
supposed to take an int in argument which must either reflect an
unsigned char or EOF. In practice on some platforms they're implemented
as macros referencing an array, and when passed a char, they either cause
a warning "array subscript has type 'char'" when lucky, or cause random
segfaults when unlucky. It's quite unconvenient by the way since none of
them may return true for negative values. The recent introduction of
cygwin to the list of regularly tested build platforms revealed a lot
of breakage there due to the same issues again.
So this patch addresses the problem all over the code at once. It adds
unsigned char casts to every valid use case, and also drops the unneeded
double cast to int that was sometimes added on top of it.
It may be backported by dropping irrelevant changes if that helps better
support uncommon platforms. It's unlikely to fix bugs on platforms which
would already not emit any warning though.
A build failure on cygwin was reported on github actions here:
https://github.com/haproxy/haproxy/runs/466507874
It's caused by a signed char being passed to isspace(), and this one
being implemented as a macro instead of a function as the man page
suggests. It's the same issue that regularly pops up on Solaris. This
comes from commit 98263291cc which was merged in 1.8-dev1. A backport
is possible though not incredibly useful.
`curr_idle_thr` is of type `unsigned int`, not `int`. Fix this issue by
taking the size of the dereferenced `curr_idle_thr` array.
This issue was introduced when adding the `curr_idle_thr` struct member
in commit f131481a0a. This commit is first
tagged in 2.0-dev1 and marked for backport to 1.9.
This used to be a minor optimization on ix86 where registers are scarce
and the calling convention not very efficient, but this platform is not
relevant enough anymore to warrant all this dirt in the code for the sake
of saving 1 or 2% of performance. Modern platforms don't use this at all
since their calling convention already defaults to using several registers
so better get rid of this once for all.
Don't try to load a .key in a directory without loading its associated
certificate file.
This patch ignores the .key files when iterating over the files in a
directory.
Introduced by 4c5adbf ("MINOR: ssl: load the key from a dedicated
file").
For a certificate on a bind line, if the private key was not found in
the PEM file, look for a .key and load it.
This default behavior can be changed by using the ssl-load-extra-files
directive in the global section
This feature was mentionned in the issue #221.
Now that we have flags indicating the CPU's capabilities, better use
them instead of missing some updates for new CPU families (ARMv8 was
missing there).
The blocksize and the extra field are not necessarily aligned on a
machine word. This can result in crashing an align-sensitive machine
when initializing the shctx area. Let's round both sizes up to a pointer
size to make this safe everywhere.
This fixes issue #512. This should be backported as far as 1.8.
In sample_fetch_path we can use iststop() instead of a for loop
to find the '?' and return the correct length. This requires commit
"MINOR: ist: add an iststop() function".
In http_action_replace_uri() we call http_get_path() in the case of
a replace-path rule. http_get_path() will return an ist pointing to
the start of the path, but uri.ptr + uri.len points to the end of the
uri. As as result, we are matching against a string containing the
query, which we append to the "path" later, effectively duplicating
the query string.
This patch uses the iststop() function introduced in "MINOR: ist: add
an iststop() function" to find the '?' character and update the ist
length when needed.
This fixes issue #510.
The bug was introduced by commit 262c3f1a ("MINOR: http: add a new
"replace-path" action"), which was backported to 2.1 and 2.0.
When we're in H1_MSG_RQBEFORE or H1_MSG_RPBEFORE, we know that the
first message is highly likely the only one and that it's pointless
to try to perform a second recvfrom() to complete a first partial
read. This is similar to what used to be done in the older I/O methods
with the CF_READ_DONTWAIT flag on the channel. So let's pass
CO_RFL_READ_ONCE to the transport layer during rcv_buf() in this case.
By doing so, in a test involving keep-alive connections with a non-null
client think time, we remove 20% of the recvfrom() calls, all of which
used to systematically fail. More precisely, we observe a drop from 5.0
recvfrom() per request with 60% failure to 4.0 per request with 50%
failure.
This flag is currently supported by raw_sock to perform a single recv()
attempt and avoid subscribing. Typically on the request and response
paths with keep-alive, with short messages we know that it's very likely
that the first message is enough.
This marks the end of the transition from the connection polling states
introduced in 1.5-dev12 and the subscriptions in that arrived in 1.9.
The socket layer can now safely use its FD while all upper layers rely
exclusively on subscriptions. These old functions were removed. Some may
deserve some renaming to improved clarty though. The single call to
conn_xprt_stop_both() was dropped in favor of conn_cond_update_polling()
which already does the same.
The last few calls to conn_xprt_{want,stop}_{recv,send} in the central
connection code were replaced with their strictly exact equivalent fd_*,
adding the call to conn_ctrl_ready() when it was missing.
Just like previous commit, we don't need to pass through the connection
layer anymore to enable polling during a connect(), we know the FD, so
let's simply call fd_want_send().
Now that we know that the connection layer is transparent for polling
changes, we have no reason for hiding behind conn_xprt_stop_send() and
can safely call fd_stop_send() on the FD once the buffer is empty.
Historically we used to require that the connections held the desired
polling states for the data layer and the socket layer. Then with muxes
these were more or less merged into the transport layer, and now it
happens that with all transport layers having their own state, the
"transport layer state" as we have it in the connection (XPRT_RD_ENA,
XPRT_WR_ENA) is only an exact copy of the undelying file descriptor
state, but with a delay. All of this is causing some difficulties at
many places in the code because there are still some locations which
use the conn_want_* API to remain clean and only rely on connection,
and count on a later collection call to conn_cond_update_polling(),
while others need an immediate action and directly use the FD updates.
Since our updates are now much cheaper, most of them being only an
atomic test-and-set operation, and since our I/O callbacks are deferred,
there's no benefit anymore in trying to "cache" the transient state
change in the connection flags hoping to cancel them before they
become an FD event. Better make such calls transparent indirections
to the FD layer instead and get rid of the deferred operations which
needlessly complicate the logic inside.
This removes flags CO_FL_XPRT_{RD,WR}_ENA and CO_FL_WILL_UPDATE.
A number of functions related to polling updates were either greatly
simplified or removed.
Two places were using CO_FL_XPRT_WR_ENA as a hint to know if more data
were expected to be sent after a PROXY protocol or SOCKSv4 header. These
ones were simply replaced with a check on the subscription which is
where we ought to get the autoritative information from.
Now the __conn_xprt_want_* and their conn_xprt_want_* counterparts
are the same. conn_stop_polling() and conn_xprt_stop_both() are the
same as well. conn_cond_update_polling() only causes errors to stop
polling. It also becomes way more obvious that muxes should not at
all employ conn_xprt_{want|stop}_{recv,send}(), and that the call
to __conn_xprt_stop_recv() in case a mux failed to allocate a buffer
is inappropriate, it ought to unsubscribe from reads instead. All of
this definitely requires a serious cleanup.
Historically we used to have a global epoll_event for various
manipulations involving epoll_ctl() and when threads were added,
this was turned to a thread_local, which is needlessly expensive
since it's just a temporary variable. Let's move it to a local
variable wherever it's called instead.
While trying to address issue #253, Commit 5909380c ("BUG/MINOR: checks:
stop polling for write when we have nothing left to send") made sure that
we stop polling for writes when the buffer is empty. This was actually
more a workaround than a bug fix because by doing so we may be stopping
polling for an intermediary transport layer without acting on the check
itself in case there's SSL or send-proxy in the chain for example, thus
the approach is wrong. In practice due to the small size of check
requests, this will not have any impact. At best, we ought to unsubscribe
for sending, but that's already the case when we arrive in this function.
But given that the root cause of the issue was addressed later in commits
cc705a6b, c5940392 and ccf3f6d1, we can now safely revert this change.
It was confirmed on the faulty config that this change doesn't have any
effect anymore on the test.
In H1/H2/FCGI, the *_get_buf() functions try to disable receipt of data
when there's no buffer available. But they do so at the lowest possible
level, which is unrelated to the upper transport layers which may still
be trying to feed data based on subscription. The correct approach here
would theorically be to only disable subscription, though when we get
there, the subscription will already have been dropped, so we can safely
just remove that call.
It's unlikely that this could have had any practical impact, as the upper
xprt layer would call this callback which would fail an not resubscribe.
Having the lowest layer disabled would just be temporary since when
re-enabling reading, a subscribe at the end of data would re-enable it.
Backport should not harm but seems useless at this point.
If an client error is reported on the request channel (CF_READ_ERROR) while a
session is tarpitted, no error is returned to the client. Concretly,
http_reply_and_close() function is not called. This function is reponsible to
forward the error to the client. But not only. It is also responsible to abort
the request. Because this function is not called when a read error is reported
on the request channel, and because the tarpit analyzer is the last one, there
is nothing preventing a connection attempt on a server while it is totally
unexpected.
So, a useless connexion on a backend server may be performed because of this
bug. If an HTTP load-balancing algorithm is used on the backend side, it leads
to a crash of HAProxy because the request was already erased.
If you have tarpit rules and if you use an HTTP load-balancing algorithm on your
backends, you must apply this patch. Otherwise a simple TCP reset on a tarpitted
connexion will most likely crash your HAProxy. A safe workaround is to use a
silent-drop rule or a deny rule instead of a tarpit.
This bug also affect the legacy code. It is in fact an very old hidden bug. But
the refactoring of process_stream() in the 1.9 makes it visible. And,
unfortunately, with the HTX, it is easier to hit it because many processing has
been moved in lower layers, in the muxes.
It must be backported as far as 1.9. For the 2.0 and the 1.9, the legacy HTTP
code must also be patched the same way. For older versions, it may be backported
but the bug seems to not impact them.
Thanks to Olivier D <webmaster@ajeux.com> to have reported the bug and provided
all the infos to analyze it.
gcc complains rightfully:
src/ssl_sock.c: In function ‘ssl_load_global_issuers_from_path’:
src/ssl_sock.c:9860:4: warning: format not a string literal and no format arguments [-Wformat-security]
ha_warning(warn);
^
Introduced in 70df7bf19c.
if the monitor-uri starts by a slash ('/'), the matching is performed against
the request's path instead of the request's uri. It is a workaround to let the
HTTP/2 requests match the monitor-uri. Indeed, in HTTP/2, clients are encouraged
to send absolute URIs only.
This patch is not tagged as a bug, because the previous behavior matched exactly
what the doc describes. But it may surprise that HTTP/2 requests don't match the
monitor-uri.
This patch may be backported to 2.1 because URIs of HTTP/2 are stored using the
absolute-form starting this version. For previous versions, this patch will only
helps explicitely absolute HTTP/1 requests (and only the HTX part because on the
legacy HTTP, all the URI is matched).
It should fix the issue #509.
The monitor-uri should be case-sensitive. In reality, the scheme and the host
part are case-insensitives and only the path is case-sensive. But concretely,
since the start, the matching on the monitor-uri is case-sensitive. And it is
probably the expected behavior of almost all users.
This patch must be backported as far as 1.9. For HAProxy 2.0 and 1.9, it must be
applied on src/proto_htx.c.
Certificates loaded with "crt" and "crt-list" commonly share the same
intermediate certificate in PEM file. "issuers-chain-path" is a global
directive to share intermediate chain certificates in a directory. If
certificates chain is not included in certificate PEM file, haproxy
will complete chain if issuer match the first certificate of the chain
stored via "issuers-chain-path" directive. Such chains will be shared
in memory.
As reported in issue #507, since commiy 07e1e3c93e ("MINOR: sample:
regsub now supports backreferences") we must not proceed in regsub()
if we fali to allocate a trash (which in practice never happens). No
backport needed.
This counter is already incremented when a new request is received (or if an
error occurred waiting it). So it must not be incremented when the stream is
terminated, at the end of process_strem(). This bug was introduced by the commit
cff0f739e ("MINOR: counters: Review conditions to increment counters from
analysers").
No backport needed.
When an Host header is updated, the autority part, if any, is also updated to
keep the both syncrhonized. But, when the update is performed while there is no
change, a failure is reported while, in reality, no update is necessary. This
bug was introduced by the commit d7b7a1ce5 ("MEDIUM: http-htx: Keep the Host
header and the request start-line synchronized").
This commit was pushed in the 2.1. But on this version, the bug is hidden
because rewrite errors are silently ignored. And because it happens when there
is no change, if the rewrite fails, noone notices it. But since the 2.2, rewrite
errors are now fatals by default. So when the bug is hit, a 500 error is
returned to the client. Without this fix, a workaround is to disable the strict
rewriting mode (see the "strict-mode" HTTP rule).
The following HTTP rule is a good way to reproduce the bug if a request with an
authority is received. In HTT2, it is pretty common.
acl host_header_exists req.hdr(host) -m found
http-request set-header host %[req.hdr(host)] if host_header_exists
This patch must be backported to 2.1 and everywhere the commit d7b7a1ce5 is
backported. It should fix the issue #494.
In flt_analyze_http_headers() HTTP analyzer, we must not forward systematically
the headers. We must only count them as filtered data (ie. increment the offset
of the right size). It is the http_payload callback responsibility to decide to
forward headers or not by forwarding at least 1 byte of payload. And there is
always at least 1 byte of payload to forward, the EOM block.
This patch depends on following commits:
* MINOR: filters: Forward data only if the last filter forwards something
* MINOR: http-htx: Add a function to retrieve the headers size of an HTX message
This patch must be backported with commits above as far as 1.9. In HAProxy 2.0
and 1.9, the patch must be adapted because of the legacy HTTP code.
In flt_tcp_payload() and flt_http_payload(), if the last filter does not
forwarding anything, nothing is forwarded, not even the already filtered
data. For now, this patch is useless because the last filter is always sync with
the stream's offset. But it will be mandatory for a bugfix.
http_get_hdrs_size() function may now be used to get the bytes held by headers
in an HTX message. It only works if the headers were not already
forwarded. Metadata are not counted here.
Now that the configuration parser is more flexible with samples,
converters and their arguments, we can leverage this to enable
support for backreferences in regsub.
Recent commit 807aef8a14 ("BUG/MINOR: arg: report an error if an argument
is larger than bufsize") aimed at fixing the argument length check but
relied on the fact that the end of string was not reached, except that
it forgot to consider the delimiters (comma and parenthesis) which are
valid conditions to break out of the loop. This used to break simple
expressions like "hdr(xff,1)". Thanks to Jrme for reporting this.
No backport is needed.
Commit ef21facd99 ("MEDIUM: arg: make make_arg_list() support quotes
in arguments") removed the chunk_strncpy() to fill the trash buffer
as the input is being parsed, and accidently dropped the jump to the
error path in case the argument is too large, which is now fixed.
No backport is needed, this is for 2.2. This addresses issue #502.
For a very long time it used to be impossible to pass a closing square
bracket as a valid character in argument to a sample fetch function or
to a converter because the LF parser used to stop on the first such
character found and to pass what was between the first '[' and the first
']' to sample_parse_expr().
This patch addresses this by passing the whole string to sample_parse_expr()
which is the only one authoritative to indicate the first character that
does not belong to the expression. The LF parser then verifies it matches
a ']' or fails. As a result it is finally possible to write rules such as
the following, which is totally valid an unambigous :
http-request redirect location %[url,regsub([.:/?-],!,g)]
|-----| | |
arg1 | `---> arg3
`-----> arg2
|-----------------|
converter
|---------------------|
sample expression
|------------------------|
log-format tag
When an end pointer is passed, instead of complaining that a comma is
missing after a keyword, sample_parse_expr() will silently return the
pointer to the current location into this return pointer so that the
caller can continue its parsing. This will be used by more complex
expressions which embed sample expressions, and may even permit to
embed sample expressions into arguments of other expressions.
Now it becomes possible to reuse the quotes within arguments, allowing
the parser to distinguish a ',' or ')' that is part of the value from
one which delimits the argument. In addition, ',' and ')' may be escaped
using a backslash. However, it is also important to keep in mind that
just like in shell, quotes are first resolved by the word tokenizer, so
in order to pass quotes that are visible to the argument parser, a second
level is needed, either using backslash escaping, or by using an alternate
type.
For example, it's possible to write this to append a comma:
http-request add-header paren-comma-paren "%[str('(--,--)')]"
or this:
http-request add-header paren-comma-paren '%[str("(--,--)")]'
or this:
http-request add-header paren-comma-paren %[str(\'(--,--)\')]
or this:
http-request add-header paren-comma-paren %[str(\"(--,--)\")]
or this:
http-request add-header paren-comma-paren %[str(\"(\"--\',\'--\")\")]
Note that due to the wide use of '\' in front of parenthesis in regex,
the backslash character will purposely *not* escape parenthesis, so that
'\)' placed in quotes is passed verbatim to a regex engine.
For each and every argument parsed by make_arg_list(), there was an
strndup() call, just so that we have a trailing zero for most functions,
and this temporary buffer is released afterwards except for strings where
it is kept.
Proceeding like this is not convenient because 1) it performs a huge
malloc/free dance, and 2) it forces to decide upfront where the argument
ends, which is what prevents commas and right parenthesis from being used.
This patch makes the function copy the temporary argument into the trash
instead, so that we avoid the malloc/free dance for most all non-string
args (e.g. integers, addresses, time, size etc), and that we can later
produce the contents on the fly while parsing the input. It adds a length
check to make sure that the argument is not longer than the buffer size,
which should obviously never be the case but who knows what people put in
their configuration.
The main problem we're having with argument parsing is that at the
moment the caller looks for the first character looking like an end
of arguments (')') and calls make_arg_list() on the sub-string inside
the parenthesis.
Let's first change the way it works so that make_arg_list() also
consumes the parenthesis and returns the pointer to the first char not
consumed. This will later permit to refine each argument parsing.
For now there is no functional change.
Instead of scanning a string looking for an end of line, ')' or ',',
let's only accept characters which are actually valid identifier
characters. This will let the parser know that in %[src], only "src"
is the sample fetch name, not "src]". This was done both for samples
and ACLs since they are the same here.
Now, only one capture is mandatory in the path-info regex, the one matching the
script-name. The path-info capture is optional. Of couse, it must be defined to
fill the PATH_INFO parameter. But it is not mandatory. This way, it is possible
to get the script-name part from the path, excluding the path-info.
This patch is small enough to be backported to 2.1.
If a regex to match the PATH_INFO parameter is configured, it systematically
fails if a newline or a null character is present in the URL-decoded path. So,
from the moment there is at least a "%0a" or a "%00" in the request path, we
always fail to get the PATH_INFO parameter and all the decoded path is used for
the SCRIPT_NAME parameter.
It is probably not the expected behavior. Because, most of time, these
characters are not expected at all in a path, an error is now triggered when one
of these characters is found in the URL-decoded path before trying to execute
the path_info regex. However, this test is not performed if there is no regex
configured.
Note that in reality, the newline character is only a problem when HAProxy is
complied with pcre or pcre2 library and conversely, the null character is only a
problem for the libc's regex library. But both are always excluded to avoid any
inconsistency depending on compile options.
An alternative, not implemented yet, is to replace these characters by another
one. If someone complains about this behavior, it will be re-evaluated.
This patch must be backported to all versions supporting the FastCGI
applications, so to 2.1 for now.
When calling the mux "destroy" method, the argument should be the mux
context, not the connection. In a few instances in the mux code, the
connection was used (mainly when the session wouldn't handle the idle
connection, and the server pool was fool), and that could lead to random
segfaults.
This should be backported to 2.1, 2.0, and 1.9
we cannot return right after socket opening as we need to move back to
the default namespace first
this should fix github issue #500
this might be backported to all version >= 1.6
Fixes: b3e54fe387 ("MAJOR: namespace: add Linux network namespace
support")
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
when `getsockopt` previously failed, we were trying to set defaultmss
with -2 value.
this is a followup of github issue #499
this should be backported to all versions >= v1.8
Fixes: 153659f1ae ("MINOR: tcp: When binding socket, attempt to
reuse one from the old proc.")
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
we were trying to close file descriptor even when `socket` call was
failing.
this should fix github issue #499
this should be backported to all versions >= v1.8
Fixes: 153659f1ae ("MINOR: tcp: When binding socket, attempt to
reuse one from the old proc.")
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
When intializing a listener, let's make sure the bind_thread mask is
always limited to all_threads_mask when inserting the FD. This will
avoid seeing listening FDs with bits corresponding to threads that are
not active (e.g. when using "bind ... process 1/even"). The side effect
is very limited, all that was identified is that atomic operations are
used in fd_update_events() when not necessary. It's more a matter of
long-term correctness in practice.
This fix might be backported as far as 1.8 (then proto_sockpair must
be dropped).
In bug #495 we found that it is possible to resume a listener on an
inexistent thread. This happens when a bind's thread_mask contains bits
out of the active threads mask, such as when using "1/odd" or "1/even".
The thread_mask was used as-is to pick a thread number to re-enable the
listener, and given that the highest number is used, 1/odd or 1/even can
produce quite high thread numbers and crash the process by queuing some
entries into non-existent lists.
This bug is an incomplete fix of commit 413e926ba ("BUG/MAJOR: listener:
fix thread safety in resume_listener()") though it will only trigger if
some bind lines are explicitly bound to thread numbers higher than the
thread count. The fix must be backported to all branches having the fix
above (as far as 1.8, though the code is different there, see the commit
message in 1.8 for changes).
There are a few other places where bind_thread is used without
enforcing all_thread_mask, namely when doing fd_insert() while creating
listeners. It seems harmless but would probably deserve another fix.
Issue #490 reports that there are a few bogus constructs of the famous
"do { if (cond) continue; } while (0)" in the connection code, that are
used to retry on I/O failures caused by receipt of a signal. Let's turn
them into the more correct "while (1) { if (cond) continue; break }"
instead. This may or may not be backported, it shouldn't have any
visible effect.
We do have some checks for the UNIX socket path length to validate the
full pathname of a unix socket but the pathname extension is only taken
into account when using a bind_prefix. The second check only matches
against MAXPATHLEN. So this means that path names between 98 and 108
might successfully parse but fail to bind. Let's adjust the check in
the address parser and refine the error checking at the bind() step.
This addresses bug #493.
In commit 477902b ("MEDIUM: connections: Get ride of the xprt_done
callback.") we added an inconditional call to h2_wake_some_streams()
in h2_wake(), though we must not do it if the connection is destroyed
or we end up with a use-after-free. In this case it's already done in
h2_process() before destroying the connection anyway.
Let's just add this test for now. A cleaner approach might consist in
doing it in the h2_process() function itself when a connection status
change is detected.
No backport is needed, this is purely 2.2.
A bug was introduced during TCP rules refactoring by the commit ac98d81f4
("MINOR: http-rule/tcp-rules: Make track-sc* custom actions"). There is no
stream when L4/L5 TCP rules are evaluated. For these rulesets, In track-sc*
actions, we must take care to rely on the session instead of the stream.
Because of this bug, any evaluation of L4/L5 TCP rules using a track-sc* action
leads to a crash of HAProxy.
No backport needed, except if the above commit is backported.
The code which is supposed to apply the bind_conf configuration on the
SSL_CTX was not called correctly. Indeed it was called with the previous
SSL_CTX so the new ones were left with default settings. For example the
ciphers were not changed.
This patch fixes#429.
Must be backported in 2.1.
This patch fixes memory leaks and a null pointer dereference found by coverity
on the error path when an HTTP return action is parsed. See issue #491.
No need to backport this patch except the HTT return action is backported too.
In action_http_set_status(), when a rewrite error occurred, the stream error
flag must be set before returning the error.
No need to backport this patch except if commit 333bf8c33 ("MINOR: http-rules:
Set SF_ERR_PRXCOND termination flag when a header rewrite fails") is
backported. This bug was reported in issue #491.
The patch adding this check initially only issued a warning, instead of
being fatal. It was changed before committing. However when making this
change the type of the log message was not changed from `ha_warning` to
`ha-alert`. This patch makes this forgotten adjustment.
see 0cf811a5f9
No backport needed. The initial patch was backported as a warning, thus
the log message type is correct.
Consider a configuration like this:
> acl t always_true
> acl or always_false
>
> http-response set-header Foo Bar if t or t
The 'or' within the condition will be treated as a logical disjunction
and the header will be set, despite the ACL 'or' being falsy.
This patch makes it an error to declare such an ACL that will never
work. This patch may be backported to stable releases, turning the
error into a warning only (the code was written in a way to make this
trivial). It should not break anything and might improve the users'
lifes.
If the longjmp() call is not flagged as "noreturn", for example, because the
operating system doesn't target a gcc-compatible compiler, we may get this
warning when building Lua :
src/hlua.c: In function 'hlua_panic_ljmp':
src/hlua.c:128:1: warning: no return statement in function returning non-void [-Wreturn-type]
static int hlua_panic_ljmp(lua_State *L) { longjmp(safe_ljmp_env, 1); }
^~~~~~
The function's prototype cannot be changed because it must be compatible
with Lua's callbacks. Let's simply enclose the call inside WILL_LJMP()
which we created exactly to signal a call to longjmp(). It lets the compiler
know we won't get back into the function and that the return statement is
not needed.
It is now possible to intercept HTTP messages from a lua action and reply to
clients. To do so, a reply object must be provided to the function
txn:done(). It may contain a status code with a reason, a header list and a
body. By default, if an empty reply object is used, an empty 200 response is
returned. If no reply is passed when txn:done() is called, the previous
behaviour is respected, the transaction is terminated and nothing is returned to
the client. The same is done for TCP streams. When txn:done() is called, the
action is terminated with the code ACT_RET_DONE on success and ACT_RET_ERR on
error, interrupting the message analysis.
The reply object may be created for the lua, by hand. Or txn:reply() may be
called. If so, this object provides some methods to fill it:
* Reply:set_status(<status> [ <reason>]) : Set the status and optionally the
reason. If no reason is provided, the default one corresponding to the status
code is used.
* Reply:add_header(<name>, <value>) : Add a header. For a given name, the
values are stored in an ordered list.
* Reply:del_header(<name>) : Removes all occurrences of a header name.
* Reply:set_body(<body>) : Set the reply body.
Here are some examples, all doing the same:
-- ex. 1
txn:done{
status = 400,
reason = "Bad request",
headers = {
["content-type"] = { "text/html" },
["cache-control"] = { "no-cache", "no-store" },
},
body = "<html><body><h1>invalid request<h1></body></html>"
}
-- ex. 2
local reply = txn:reply{
status = 400,
reason = "Bad request",
headers = {
["content-type"] = { "text/html" },
["cache-control"] = { "no-cache", "no-store" }
},
body = "<html><body><h1>invalid request<h1></body></html>"
}
txn:done(reply)
-- ex. 3
local reply = txn:reply()
reply:set_status(400, "Bad request")
reply:add_header("content-length", "text/html")
reply:add_header("cache-control", "no-cache")
reply:add_header("cache-control", "no-store")
reply:set_body("<html><body><h1>invalid request<h1></body></html>")
txn:done(reply)
This function may be used to defined a timeout when a lua action returns
act:YIELD. It is a way to force to reexecute the script after a short time
(defined in milliseconds).
Unlike core:sleep() or core:yield(), the script is fully reexecuted if it
returns act:YIELD. With core functions to yield, the script is interrupted and
restarts from the yield point. When a script returns act:YIELD, it is finished
but the message analysis is blocked on the action waiting its end.
ACT_RET_* code are now available from lua scripts. The gloabl object "act" is
used to register these codes as constant. Now, lua actions can return any of
following codes :
* act.CONTINUE for ACT_RET_CONT
* act.STOP for ACT_RET_STOP
* act.YIELD for ACT_RET_YIELD
* act.ERROR for ACT_RET_ERR
* act.DONE for ACT_RET_DONE
* act.DENY for ACT_RET_DENY
* act.ABORT for ACT_RET_ABRT
* act.INVALID for ACT_RET_INV
For instance, following script denied all requests :
core.register_action("deny", { "http-req" }, function (txn)
return act.DENY
end)
Thus "http-request lua.deny" do exactly the same than "http-request deny".
When an action successfully finishes, the action return code (ACT_RET_*) is now
retrieve on the stack, ff the first element is an integer. In addition, in
hlua_txn_done(), the value ACT_RET_DONE is pushed on the stack before
exiting. Thus, when a script uses this function, the corresponding action still
finishes with the good code. Thanks to this change, the flag HLUA_STOP is now
useless. So it has been removed.
It is a mandatory step to allow a lua action to return any action return code.
In http_process_res_common() analyzer, when a invalid response is reported, the
failed_resp counters must be incremented.
No need to backport this patch, except if the commit b8a5371a ("MEDIUM:
http-ana: Properly handle internal processing errors") is backported too.
It is not possible anymore to alter the HTTP parser state from lua sample
fetches or lua actions. So there is no reason to still check for the parser
state consistency.
It is now possible to append extra headers to the generated responses by HTTP
return actions, while it is not based on an errorfile. For return actions based
on errorfiles, these extra headers are ignored. To define an extra header, a
"hdr" argument must be used with a name and a value. The value is a log-format
string. For instance:
http-request status 200 hdr "x-src" "%[src]" hdr "x-dst" "%[dst]"
Thanks to this new action, it is now possible to return any responses from
HAProxy, with any status code, based on an errorfile, a file or a string. Unlike
the other internal messages generated by HAProxy, these ones are not interpreted
as errors. And it is not necessary to use a file containing a full HTTP
response, although it is still possible. In addition, using a log-format string
or a log-format file, it is possible to have responses with a dynamic
content. This action can be used on the request path or the response path. The
only constraint is to have a responses smaller than a buffer. And to avoid any
warning the buffer space reserved to the headers rewritting should also be free.
When a response is returned with a file or a string as payload, it only contains
the content-length header and the content-type header, if applicable. Here are
examples:
http-request return content-type image/x-icon file /var/www/favicon.ico \
if { path /favicon.ico }
http-request return status 403 content-type text/plain \
lf-string "Access denied. IP %[src] is blacklisted." \
if { src -f /etc/haproxy/blacklist.lst }
This patch introduces the 'http-after-response' rules. These rules are evaluated
at the end of the response analysis, just before the data forwarding, on ALL
HTTP responses, the server ones but also all responses generated by
HAProxy. Thanks to this ruleset, it is now possible for instance to add some
headers to the responses generated by the stats applet. Following actions are
supported :
* allow
* add-header
* del-header
* replace-header
* replace-value
* set-header
* set-status
* set-var
* strict-mode
* unset-var
Call http_forward_proxy_resp() function when an internal response is
returned. It concerns redirect, auth and error reponses. But also 100-Continue
and 103-Early-Hints responses. For errors, there is a subtlety. if the forward
fails, an HTTP 500 error is generated if it is not already an internal
error. For now http_forward_proxy_resp() cannot fail. But it will be possible
when the new ruleset applied on all responses will be added.
Operations performed when internal responses (redirect/deny/auth/errors) are
returned are always the same. The http_forward_proxy_resp() function is added to
group all of them under a unique function.
The http_server_error() function now relies on http_reply_and_close(). Both do
almost the same actions. In addtion, http_server_error() sets the error flag and
the final state flag on the stream.
The rule direction must be tested to do specific processing on the request
path. intercepted_req counter shoud be updated if the rule is evaluated on the
frontend and remaining request's analyzers must be removed. But only on the
request path. The rule direction must also be tested to set the right final
stream state flag.
This patch depends on the commit "MINOR: http-rules: Add a flag on redirect
rules to know the rule direction". Both must be backported to all stable
versions.
HTTP redirect rules can be evaluated on the request or the response path. So
when a redirect rule is evaluated, it is important to have this information
because some specific processing may be performed depending on the direction. So
the REDIRECT_FLAG_FROM_REQ flag has been added. It is set when applicable on the
redirect rule during the parsing.
This patch is mandatory to fix a bug on redirect rule. It must be backported to
all stable versions.
It is important to not forget to specify the HTX resposne was internally
generated when a server perform a redirect. This information is used by the H1
multiplexer to choose the right connexion mode when the response is sent to the
client.
This patch must be backported to 2.1.
The first index in an HTX message is the HTX block index from which the HTTP
analysis must be performed. When HAProxy sends an HTTP response, on error or
redirect, this index must be reset because all pending incoming data are
considered as forwarded. For now, it is only a bug for 103-Early-Hints
response. For other responses, it is not a problem. But it will be when the new
ruleset applied on all responses will be added. For 103 responses, if the first
index is not reset, if there are rewritting rules on server responses, the
generated 103 responses, if any, are evaluated too.
This patch must be backported and probably adapted, at least for 103 responses,
as far as 1.9.
<.arg.dns.dns_opts> field in the act_rule structure is now dynamically allocated
when a do-resolve rule is parsed. This drastically reduces the structure size.
When an error is returned to a client, the right message is injected into the
response buffer. It is performed by http_server_error() or
http_replay_and_close(). Both ignore any data already present into the channel's
buffer. While it is legitimate to remove all input data, it is important to not
remove any outgoing data.
So now, we try to append the error message to the response buffer, only removing
input data. We rely on the channel_htx_copy_msg() function to do so. So this
patch depends on the following two commits:
* MINOR: htx: Add a function to append an HTX message to another one
* MINOR: htx/channel: Add a function to copy an HTX message in a channel's buffer
This patch must be backported as far as 1.9. However, above patches must be
backported first.
the htx_append_msg() function can now be used to append an HTX message to
another one. All the message is copied or nothing. If an error occurs during the
copy, all changes are rolled back.
This patch is mandatory to fix a bug in http_reply_and_close() function. Be
careful to backport it first.
If an error file is too big and, once converted in HTX, runs over the buffer
space reserved to headers rewritting, a warning is emitted. Because a new set of
rules will be added to allow headers rewritting on all responses, including
HAProxy ones, it is important to always keep this space free for error files.
When a header rewrite fails, an internal errors is triggered. But
SF_ERR_INTERNAL is documented to be the concequence of a bug and must be
reported to the dev teamm. So, when this happens, the SF_ERR_PRXCOND termination
flag is set now.
When the global structure is initialized, instead of setting tune.maxrewrite to
-1, its default value can be immediately set. This way, it is always defined
during the configuration validity check. Otherwise, the only way to have it at
this stage, it is to explicity set it in the global section.
Since the strict rewritting mode was introduced, actions manipulating headers
(set/add/replace) always rely on the request message to test if the
HTTP_MSGF_SOFT_RW flag is set or not. But, of course, we must only rely on the
request for http-request rules. For http-response rules, we must use the
response message.
This patch must be backported if the strict rewritting is backported too.
The function was added in commit 6c39198b57,
but was also used within a single function `free_dcache` which was unused
itself.
see issue #301
see commit 10ce0c2f31 which removed
`free_dcache`
The function was changed to be static in commit
6c39198b57, but even that commit
no longer uses it. The purpose of the change vs. outright removal
is unclear.
see issue #301
As reported in issue #485 the test for !len at the end of the
loop in get_var_int() is useless since it was already done inside
the loop. Actually the code is more readable if we remove the first
one so let's do this instead. The resulting code is exactly the same
since the compiler already optimized the test away.
In ssl_sock_load_dh_params(), if haproxy failed to apply the dhparam
with SSL_CTX_set_tmp_dh(), it will apply the DH with
SSL_CTX_set_dh_auto().
The problem is that we don't clean the OpenSSL errors when leaving this
function so it could fail to load the certificate, even if it's only a
warning.
Fixes bug #483.
Must be backported in 2.1.
We have the ability per bind option to ignore certain errors (CA, crt, ...),
and for this we use a 64-bit field. In issue #479 coverity reports a risk of
too large a left shift. For now as of OpenSSL 1.1.1 the highest error value
that may be reported by X509_STORE_CTX_get_error() seems to be around 50 so
there should be no risk yet, but it's enough of a warning to add a check so
that we don't accidently hide random errors in the future.
This may be backported to relevant stable branches.
This new setting in the global section alters the way HAProxy will look
for unspecified files (.ocsp, .sctl, .issuer, bundles) during the
loading of the SSL certificates.
By default, HAProxy discovers automatically a lot of files not specified
in the configuration, and you may want to disable this behavior if you
want to optimize the startup time.
This patch sets flags in global_ssl.extra_files and then check them
before trying to load an extra file.
When using lockless pools, add a new rwlock, flush_pool. read-lock it when
getting memory from the pool, so that concurrenct access are still
authorized, but write-lock it when we're about to free memory, in
pool_flush() and pool_gc().
The problem is, when removing an item from the pool, we unreference it
to get the next one, however, that pointer may have been free'd in the
meanwhile, and that could provoke a crash if the pointer has been unmapped.
It should be OK to use a rwlock, as normal operations will still be able
to access the pool concurrently, and calls to pool_flush() and pool_gc()
should be pretty rare.
This should be backported to 2.1, 2.0 and 1.9.
In pool_create(), only initialize the pool spinlock if we just created the
pool, in the event we're reusing it, there's no need to initialize it again.
In pool_flush(), we can't just set the free_list to NULL, or we may suffer
the ABA problem. Instead, use a double-width CAS and update the sequence
number.
This should be backported to 2.1, 2.0 and 1.9.
This may, or may not, be related to github issue #476.
We can't clear flags on tasklets because we don't know if they're still
present upon return (they all return NULL, maybe that could change in
the future). As a side effect, once TASK_RUNNING is set, it's never
cleared anymore, which is misleading and resulted in some incorrect
flagging of bulk tasks in the recent scheduler changes. And the only
reason for setting TASK_RUNNING on tasklets was to detect self-wakers,
which is not done using a dedicated flag. So instead of setting this
flags for no opportunity to clear it, let's simply not set it.
Now that we can more accurately watch which connection is really
being woken up from itself, it was desirable to re-adjust the CPU BW
thresholds based on measurements. New tests with 60000 concurrent
connections were run at 100 Gbps with unbounded queues and showed
the following distribution:
scenario TC0 TC1 TC2 observation
-------------------+---+---+----+---------------------------
TCP conn rate : 32, 51, 17
HTTP conn rate : 34, 41, 25
TCP byte rate : 2, 3, 95 (2 MB objets)
splicing byte rate: 11, 6, 83 (2 MB objets)
H2 10k object : 44, 23, 33 client-limited
mixed traffic : 18, 10, 72 2*1m+1*0: 11kcps, 36 Gbps
The H2 experienced a huge change since it uses a persistent connection
that was accidently flagged in the previous test. The splicing test
exhibits a higher need for short tasklets, so does the mixed traffic
test. Given that latency mainly matters for conn rate and H2 here,
the ratios were readjusted as 33% for TC0, 50% for TC1 and 17% for
TC2, keeping in mind that whatever is not consumed by one class is
automatically shared in equal propertions by the next one(s). This
setting immediately provided a nice improvement as with the default
settings (maxpollevents=200, runqueue-depth=200), the same ratios as
above are still reported, while the time to request "show activity"
on the CLI dropped to 30-50ms. The average loop time is around 5.7ms
on the mixed traffic.
In addition, one extra stress test at 90.5 Gbps with 5100 conn/s shows
70-100ms CLI request time, with an average loop time of 17 ms.
sched->current is used to know the current task/tasklet, and is currently
only used by the panic dump code. However it turns out it was not set for
tasklets, which prevents us from using it for more usages, despite the
panic handling code already handling this case very well. Let's make sure
it's now set.
Commit a17664d829 ("MEDIUM: tasks: automatically requeue into the bulk
queue an already running tasklet") tried to inflict a penalty to
self-requeuing tasks/tasklets which correspond to those involved in
large, high-latency data transfers, for the benefit of all other
processing which requires a low latency. However, it turns out that
while it ought to do this on a case-by-case basis, basing itself on
the RUNNING flag isn't accurate because this flag doesn't leave for
tasklets, so we'd rather need a distinct flag to tag such tasklets.
This commit introduces TASK_SELF_WAKING to mark tasklets acting like
this. For now it's still set when TASK_RUNNING is present but this
will have to change. The flag is kept across wakeups.
Commit 140237471e made sure we hold the
toremove_lock for the corresponding thread before removing a connection
from its idle_orphan_conns list, however it failed to unlock it if we
found a connection, leading to a deadlock, so add the missing deadlock.
This should be backported to 2.1 and 2.0.
Measures with unbounded execution ratios under 40000 concurrent
connections at 100 Gbps showed the following CPU bandwidth
distribution between task classes depending on traffic scenarios:
scenario TC0 TC1 TC2 observation
-------------------+---+---+----+---------------------------
TCP conn rate : 29, 48, 23 221 kcps
HTTP conn rate : 29, 47, 24 200 kcps
TCP byte rate : 3, 5, 92 53 Gbps
splicing byte rate: 5, 10, 85 70 Gbps
H2 10k object : 10, 21, 74 client-limited
mixed traffic : 4, 7, 89 2*1m+1*0: 11kcps, 36 Gbps
Thus it seems that we always need a bit of bulk tasks even for short
connections, which seems to imply a suboptimal processing somewhere,
and that there are roughly twice as many tasks (TC1=normal) as regular
tasklets (TC0=urgent). This ratio stands even when data forwarding
increases. So at first glance it looks reasonable to enforce the
following ratio by default:
- 16% for TL_URGENT
- 33% for TL_NORMAL
- 50% for TL_BULK
With this, the TCP conn rate climbs to ~225 kcps, and the mixed traffic
pattern shows a more balanced 17kcps + 35 Gbps with 35ms CLI request
time time instead of 11kcps + 36 Gbps and 400 ms response time. The
byte rate tests (1M objects) are not affected at all. This setting
looks "good enough" to allow immediate merging, and could be refined
later.
It's worth noting that it resists very well to massive increase of
run queue depth and maxpollevents: with the run queue depth changed
from 200 to 10000 and maxpollevents to 10000 as well, the CLI's
request time is back to the previous ~400ms, but the mixed traffic
test reaches 52 Gbps + 7500 CPS, which was never met with the previous
scheduling model, while the CLI used to show ~1 minute response time.
The reason is that in the bulk class it becomes possible to perform
multiple rounds of recv+send and eliminate objects at once, increasing
the L3 cache hit ratio, and keeping the connection count low, without
degrading too much the latency.
Another test with mixed traffic involving 2/3 splicing on huge objects
and 1/3 on empty objects without touching any setting reports 51 Gbps +
5300 cps and 35ms CLI request time.
We used to mix high latency tasks and low latency tasklets in the same
list, and to even refill bulk tasklets there, causing some unfairness
in certain situations (e.g. poll-less transfers between many connections
saturating the machine with similarly-sized in and out network interfaces).
This patch changes the mechanism to split the load into 3 lists depending
on the task/tasklet's desired classes :
- URGENT: this is mainly for tasklets used as deferred callbacks
- NORMAL: this is for regular tasks
- BULK: this is for bulk tasks/tasklets
Arbitrary ratios of max_processed are picked from each of these lists in
turn, with the ability to complete in one list from what was not picked
in the previous one. After some quick tests, the following setup gave
apparently good results both for raw TCP with splicing and for H2-to-H1
request rate:
- 0 to 75% for urgent
- 12 to 50% for normal
- 12 to what remains for bulk
Bulk is not used yet.
New function run_tasks_from_list() will run over a tasklet list and will
run all the tasks and tasklets it finds there within a limit of <max>
that is passed in arggument. This is a preliminary work for scheduler QoS
improvements.
Previous patch 160287b676 ("MEDIUM: pipe/thread: maintain a per-thread
local cache of recently used pipes") didn't replace all pipe counter
updates with atomic ops since some were already under a lock, which is
obviously not a valid reason since these ones can be updated in parallel
to other atomic ops. The result was that the pipes_used could seldom be
seen as negative in the stats (harmless) but also this could result in
slightly more pipes being allocated than permitted, thus stealing a few
file descriptors that were not usable for connections anymore. Let's use
pure atomic ops everywhere these counters are updated.
No backport is needed.
In order to completely remove the pipe locking cost and try to reuse
hot pipes, each thread now maintains a local cache of recently used pipes
that is no larger than its share (maxpipes/nbthreads). All extra pipes
are instead refilled into the global pool. Allocations are made from the
local pool first, and fall back to the global one before allocating one.
This completely removes the observed pipe locking cost at high bit rates,
which was still around 5-6%.
In a quick test involving splicing, we can see that get_pipe() and
put_pipe() together consume up to 12% of the CPU. That's not surprizing
considering how much work is performed under the lock, including the
pipe struct allocation, the pipe creation and its initialization. Same
for releasing, we don't need a lock there to call close() nor to free
to the pool.
Changing this alone was enough to cut the overhead in half. A better
approach should consist in having a per-thread pipe cache, which will
also help keep pages hot in the CPU caches.
src/ssl_sock.c: In function ‘cli_io_handler_show_cert’:
src/ssl_sock.c:10214:6: warning: unused variable ‘n’ [-Wunused-variable]
int n;
^
Fix this problem in the io handler of the "show ssl cert" function.
Given that raw_sock's functions solely act on connections and that all its
callers properly use subscribe() when they want to receive/send more, there
is no more reason for calling fd_{cant,cond,done}_{send,recv} anymore as
this call is immediately overridden by the subscribe call. It's also worth
noting that the purpose of fd_cond_recv() whose purpose was to speculatively
enable reading in the FD cache if the FD was active but not yet polled was
made to save on expensive epoll_ctl() calls and was implicitly covered more
cleanly by recent commit 5d7dcc2a8e ("OPTIM: epoll: always poll for recv if
neither active nor ready").
No change on the number of calls to epoll_ctl() was noticed consecutive to
this change.
this log could be sometimes a bit confusing (depending on the number in
fact) when you read it (e.g is it the number of active connection?) -
only trained eyes knows haproxy output a different log when closing
active connections while stopping.
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
hostname were limited to 62 char, which is not RFC1035 compliant;
- the parsing loop should stop when above max label char
- fix len label test where d[i] was wrongly used
- simplify the whole function to avoid using two extra char* variable
this should fix github issue #387
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
Reviewed-by: Tim Duesterhus <tim@bastelstu.be>
Acked-by: Baptiste <bedis9@gmail.com>
triggered by coverity; src_port is set earlier.
this should fix github issue #467
Fixes: 7fec021537 ("MEDIUM: proxy_protocol: Convert IPs to v6 when
protocols are mixed")
This should be backported to 1.8.
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
Reviewed-by: Tim Duesterhus <tim@bastelstu.be>
Many HTTP actions rely on <.arg.http> in the act_rule structure. Not all actions
use the log-format expression, but it must be initialized anyway. Otherwise,
HAProxy may crash during the deinit when the release function is called.
No backport needed. This patch should fix issue #468.
In issue #465, we see that Coverity detected dead code in checks.c
which is in fact a missing parenthesis to build the connect() flags
consecutive to the API change in commit fdcb007ad8 ("MEDIUM: proto:
Change the prototype of the connect() method.").
The impact should be imperceptible as in the best case it may have
resulted in a missed optimization trying to save a syscall or to merge
outgoing packets.
It may be backported as far as 2.0 though it's not critical.
In ssl_sock_init(), if we fail to allocate the BIO, don't forget to free
the SSL *, or we'd end up with a memory leak.
This should be backported to 2.1 and 2.0.
As the server early data buffer is allocated in the middle of the loop
used to allocate the SSL session without being freed before retrying,
this leads to a memory leak.
To fix this we move the section of code responsible of this early data buffer
alloction after the one reponsible of allocating the SSL session.
Must be backported to 2.1 and 2.0.
When commit 477902bd2e made the conn_stream
allocation unconditional, it unfortunately moved the code doing the allocation
inside #if USE_OPENSSL, which means anybody compiling haproxy without
openssl wouldn't allocate any conn_stream, and would get a segfault later.
Fix that by moving the code that does the allocation outside #if USE_OPENSSL.
In process_stream(), when a client or a server abort is handled, the
corresponding listener's counter is incremented. But, we must be sure to have a
listener attached to the session. This bug was introduced by the commit
cff0f739e5.
Thanks to Fred to reporting me the bug.
No need to backport this patch, except if commit cff0f739e5 is backported.
A stupid cut-paste bug was introduced in the commit cff0f739e5. Backend
counters must of course be incremented on the stream's backend. Not the
frontend.
No need to backport this patch, except if commit cff0f739e5 is backported.
A first patch was made during 2.0-dev to silence a bogus warning emitted
by gcc : dd1c8f1f72 ("MINOR: cfgparse: Add a cast to make gcc happier."),
but it happens it was not sufficient as the warning re-appeared on 32-bit
machines under gcc-8 and gcc-9 :
src/cfgparse.c: In function 'check_config_validity':
src/cfgparse.c:3642:33: warning: argument 1 range [2147483648, 4294967295] exceeds maximum object size 2147483647 [-Walloc-size-larger-than=]
newsrv->idle_orphan_conns = calloc((unsigned int)global.nbthread, sizeof(*newsrv->idle_orphan_conns));
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This warning doesn't trigger in other locations, and it immediately
vanishes if the previous or subsequent loops do not depend on
global.nbthread anymore, or if the field ordering of the struct server
changes! As discussed in the thread at:
https://www.mail-archive.com/haproxy@formilux.org/msg36107.html
playing with -Walloc-size-larger-than has no effect. And a minimal
reproducer could be isolated, indicating it's pointless to circle around
this one. Let's just cast nbthread to ushort so that gcc cannot make
this wrong detection. It's unlikely we'll use more than 65535 threads in
the near future anyway.
This may be backported to older releases if they are also affected, at
least to ease the job of distro maintainers.
Thanks to Ilya for testing.
lua-prepend-path allows the administrator to specify a custom Lua library
path to load custom Lua modules that are useful within the context of HAProxy
without polluting the global Lua library folder.
While the H2 parser properly checks for the absence of anything but
"trailers" in the TE header field, we forget to check this when sending
the request to an H2 server. The problem is that an H2->H2 conversion
may keep "gzip" and fail on the next stage.
This patch makes sure that we only send "TE: trailers" if the TE header
contains the "trailers" token, otherwise it's dropped.
This fixes issue #464 and should be backported till 1.9.
Since commit 1b8e68e89a ("MEDIUM: stick-table: Stop handling stick-tables
as proxies."), a rule referencing the current proxy with no table leads
to the following error :
[ALERT] 023/071924 (16479) : Proxy 'px': unable to find stick-table '(null)'.
[ALERT] 023/071914 (16479) : Fatal errors found in configuration.
for a config like this one:
backend px
stick on src
This patch fixes it and should be backported as far as 2.0.
A few places in health checks and stream-int on the send path were still
checking for this flag. Now we do not and instead we rely on snd_buf()
to report the error if any.
It's worth noting that all 3 real muxes still use CO_FL_SOCK_WR_SH and
CO_FL_ERROR interchangeably at various places to decide to abort and/or
free their data. This should be clarified and fixed so that only
CO_FL_ERROR is used, and this will render the error paths simpler and
more accurate.
The test was added before splice() and send() to make sure we never
accidently send after a shutdown, because upper layers do not all
check and it's not their job to do it. In such a case we also set
errno to EPIPE so that the error can be accurately reported, e.g.,
in health checks.
Just like with CO_FL_SOCK_RD_SH, we don't need to check for this flag too
early because conn_sock_send() already does it. No error was lost so it
was harmless, it was only useless code.
The handshake functions dedicated to proxy proto, netscaler and
socks4 all check for this flag before proceeding. This is wrong,
they must not do and instead perform the call to recv() then
report the close. The reason for this is that the current
construct managed to lose the CO_ER_CIP_EMPTY error code in case
the connection was already shut, thus causing a race condition
with some errors being reported correctly or as unknown depending
on the timing.
There are still leftovers from the pre-xprt_handshake era with lots
of places where I/O callbacks refrain from receiving/sending if they
see that a handshake is present. This needlessly duplicates the
subscribe calls as it will automatically be done by the underlying
xprt_handshake code when attempting the operation.
The only reason for still checking CO_FL_HANDSHAKE is when we decide
to instantiate xprt_handshake. This patch removes all other ones.
As mentioned in commit c192b0ab95 ("MEDIUM: connection: remove
CO_FL_CONNECTED and only rely on CO_FL_WAIT_*"), there is a lack of
consistency on which flags are checked among L4/L6/HANDSHAKE depending
on the code areas. A number of sample fetch functions only check for
L4L6 to report MAY_CHANGE, some places only check for HANDSHAKE and
many check both L4L6 and HANDSHAKE.
This patch starts to make all of this more consistent by introducing a
new mask CO_FL_WAIT_XPRT which is the union of L4/L6/HANDSHAKE and
reports whether the transport layer is ready or not.
All inconsistent call places were updated to rely on this one each time
the goal was to check for the readiness of the transport layer.
Most places continue to check CO_FL_HANDSHAKE while in fact they should
check CO_FL_HANDSHAKE_NOSSL, which contains all handshakes but the one
dedicated to SSL renegotiation. In fact the SSL layer should be the
only one checking CO_FL_SSL_WAIT_HS, so as to avoid processing data
when a renegotiation is in progress, but other ones randomly include it
without knowing. And ideally it should even be an internal flag that's
not exposed in the connection.
This patch takes CO_FL_SSL_WAIT_HS out of CO_FL_HANDSHAKE, uses this flag
consistently all over the code, and gets rid of CO_FL_HANDSHAKE_NOSSL.
In order to limit the confusion that has accumulated over time, the
CO_FL_SSL_WAIT_HS flag which indicates an ongoing SSL handshake,
possibly used by a renegotiation was moved after the other ones.
As mentioned in c192b0ab95 ("MEDIUM: connection: remove CO_FL_CONNECTED
and only rely on CO_FL_WAIT_*"), si_cs_recv() currently does not propagate
CS_FL_EOS to CF_READ_NULL if CO_FL_WAIT_L4L6 is set, while this situation
doesn't exist anymore. Let's get rid of this confusing test.
We only add the Early-data header, or get ssl_fc_has_early to return 1, if
we didn't already did the SSL handshake, as otherwise, we know the early
data were fine, and there's no risk of replay attack. But to do so, we
wrongly checked CO_FL_HANDSHAKE, we have to check CO_FL_SSL_WAIT_HS instead,
as we don't care about the status of any other handshake.
This should be backported to 2.1, 2.0, and 1.9.
When deciding if we should add the Early-Data header, or if the sample fetch
should return
Commit 477902bd2e ("MEDIUM: connections: Get ride of the xprt_done
callback.") broke the master CLI for a very obscure reason. It happens
that short requests immediately terminated by a shutdown are properly
received, CS_FL_EOS is correctly set, but in si_cs_recv(), we refrain
from setting CF_SHUTR on the channel because CO_FL_CONNECTED was not
yet set on the connection since we've not passed again through
conn_fd_handler() and it was not done in conn_complete_session(). While
commit a8a415d31a ("BUG/MEDIUM: connections: Set CO_FL_CONNECTED in
conn_complete_session()") fixed the issue, such accident may happen
again as the root cause is deeper and actually comes down to the fact
that CO_FL_CONNECTED is lazily set at various check points in the code
but not every time we drop one wait bit. It is not the first time we
face this situation.
Originally this flag was used to detect the transition between WAIT_*
and CONNECTED in order to call ->wake() from the FD handler. But since
at least 1.8-dev1 with commit 7bf3fa3c23 ("BUG/MAJOR: connection: update
CO_FL_CONNECTED before calling the data layer"), CO_FL_CONNECTED is
always synchronized against the two others before being checked. Moreover,
with the I/Os moved to tasklets, the decision to call the ->wake() function
is performed after the I/Os in si_cs_process() and equivalent, which don't
care about this transition either.
So in essence, checking for CO_FL_CONNECTED has become a lazy wait to
check for (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN), but that always
relies on someone else having synchronized it.
This patch addresses it once for all by killing this flag and only checking
the two others (for which a composite mask CO_FL_WAIT_L4L6 was added). This
revealed a number of inconsistencies that were purposely not addressed here
for the sake of bisectability:
- while most places do check both L4+L6 and HANDSHAKE at the same time,
some places like assign_server() or back_handle_st_con() and a few
sample fetches looking for proxy protocol do check for L4+L6 but
don't care about HANDSHAKE ; these ones will probably fail on TCP
request session rules if the handshake is not complete.
- some handshake handlers do validate that a connection is established
at L4 but didn't clear CO_FL_WAIT_L4_CONN
- the ->ctl method of mux_fcgi, mux_pt and mux_h1 only checks for L4+L6
before declaring the mux ready while the snd_buf function also checks
for the handshake's completion. Likely the former should validate the
handshake as well and we should get rid of these extra tests in snd_buf.
- raw_sock_from_buf() would directly set CO_FL_CONNECTED and would only
later clear CO_FL_WAIT_L4_CONN.
- xprt_handshake would set CO_FL_CONNECTED itself without actually
clearing CO_FL_WAIT_L4_CONN, which could apparently happen only if
waiting for a pure Rx handshake.
- most places in ssl_sock that were checking CO_FL_CONNECTED don't need
to include the L4 check as an L6 check is enough to decide whether to
wait for more info or not.
It also becomes obvious when reading the test in si_cs_recv() that caused
the failure mentioned above that once converted it doesn't make any sense
anymore: having CS_FL_EOS set while still waiting for L4 and L6 to complete
cannot happen since for CS_FL_EOS to be set, the other ones must have been
validated.
Some of these parts will still deserve further cleanup, and some of the
observations above may induce some backports of potential bug fixes once
totally analyzed in their context. The risk of breaking existing stuff
is too high to blindly backport everything.
ocsp_issuer is primary set from ckch->chain when PEM is loaded from file,
but not set when PEM is loaded via CLI payload. Set ckch->ocsp_issuer in
ssl_sock_load_pem_into_ckch to fix that.
Should be backported in 2.1.
We can't just assume conn_create_mux() will be called, and set CO_FL_CONNECTED,
conn_complete_session() might be call synchronously if we're not using SSL,
so ew haee no choice but to set CO_FL_CONNECTED in there. This should fix
the recent breakage of the mcli reg tests.
When using the OCSP response, if the issuer of the response is in
the certificate chain, its address will be stored in ckch->ocsp_issuer.
However, since the ocsp_issuer could be filled by a separate file, this
pointer is free'd. The refcount of the X509 need to be incremented to
avoid a double free if we free the ocsp_issuer AND the chain.
As reported in bug #447, gcc 9.2 invents impossible code paths and then
complains that we don't check for our pointers to be NULL... This code
path is not critical, better add the test to shut it up than try to
help it being less creative.
This code hasn't changed for a while, so it could help distros to
backport this to older releases.
objt_conn() may return a NULL though here we don't have this situation
anymore since the connection is always there, so let's simply switch
to the unchecked __objt_conn(). This addresses issue #454.
Coverity rightfully reported that it's pointless to test for "conn"
to be null while all code paths leading to it have already
dereferenced it. This addresses issue #461.
When using "set ssl cert" on the CLI, if we load a new PEM, the previous
sctl, issuer and OCSP response are still loaded. This doesn't make any
sense since they won't be usable with a new private key.
This patch free the previous data.
Should be backported in 2.1.
When using multiple filters with "show table", it can be useful to
report which filter entry failed
> show table MY_TABLE data.gpc0 gt 0 data.gpc0a lt 1000
Filter entry #2: Unknown data type
> show table MY_TABLE data.gpc0 gt 0 data.gpc0 lt 1000a
Filter entry #2: Require a valid integer value to compare against
We now also catch garbage data after the filter
> show table MY_TABLE data.gpc0 gt 0 data.gpc0 lt 1000 data.gpc0 gt 1\
data.gpc0 gt 10 a
Detected extra data in filter, 16th word of input, after '10'
Even before multi-filter feature we've also silently accepted garbage
after the input, hiding potential bugs
> show table MY_TABLE data.gpc0 gt 0 data.gpc0
or
> show table MY_TABLE data.gpc0 gt 0 a
In both cases, only first filter entry would be used, silently ignoring
extra filter entry or garbage data.
Last, but not the least, it is now possible to detect multi-filter
feature from cli with something like the following:
> show table MY_TABLE data.blah
Filter entry #1: Unknown data type
The xprt_done_cb callback was used to defer some connection initialization
until we're connected and the handshake are done. As it mostly consists of
creating the mux, instead of using the callback, introduce a conn_create_mux()
function, that will just call conn_complete_session() for frontend, and
create the mux for backend.
In h2_wake(), make sure we call the wake method of the stream_interface,
as we no longer wakeup the stream task.
In connect_server(), when creating a new connection for which we don't yet
know the mux (because it'll be decided by the ALPN), instead of associating
the connection to the stream_interface, always create a conn_stream. This way,
we have less special-casing needed. Store the conn_stream in conn->ctx,
so that we can reach the upper layers if needed.
We don't properly check for missing data values for additional filter
entries, passing out of bounds index to args[], then passing to strlen.
Introduced in commit 1a693fc2: (MEDIUM: cli: Allow multiple filter
entries for "show table")
