CLEANUP: htx: remove the unreachable "append_data" label in htx_reserve_max_data()
htx_reserve_max_data() carries an "append_data:" label which is never the target
of a goto: the function simply falls through it. It looks like a leftover from
htx_add_data(), which does use such a label. Let's drop it, the code is reached
by fallthrough anyway and an unused label is only confusing.
BUG/MINOR: http-ana: fix a one-byte over-read in the client-side cookie parser
In http_manage_client_side_cookies(), each iteration of the cookie loop skips
the blanks in front of the attribute name, then tests it against '$':
while (att_beg < hdr_end && HTTP_IS_SPHT(*att_beg))
att_beg++;
...
if (*att_beg == '$')
continue;
<att_beg> may legitimately have reached <hdr_end>, in which case the test reads
one byte past the end of the Cookie header value. This happens for any value
ending with a delimiter, optionally followed by blanks, e.g. "Cookie: a=b; ":
the last iteration starts on the ';', skips it and the trailing space, and lands
exactly on <hdr_end>.
The extra byte always belongs to the HTX buffer (the payload area is followed by
other blocks, the free space or the block table), so there is no out-of-bounds
access at the allocation level, and the only functional consequence is that a
neighbour byte holding a '$' makes the parser skip the end of the header as if
it were an attribute instead of marking the header as to be preserved. But the
value must not be read past its end, and the check is free since <hdr_end> is
already at hand.
Note that http_manage_server_side_cookies() does not have this issue, it uses
"equal == val_end" to detect the empty trailing element.
This has been there since the HTX rewrite of the analysers by commit f4eb75d17
("MINOR: htx: Add proto_htx.c file") in 1.9-dev7, and the pre-HTX code had the
same construct, so it should be backported to all supported versions.
BUG/MINOR: h3: don't use a block pointer to roll back a partial HTX conversion
h3_resp_headers_to_htx() and h3_trailers_to_htx() save a pointer on the tail
HTX block of the destination message and use it to remove whatever they added
when the conversion fails:
tailblk = htx_get_tail_blk(htx);
...
out:
if (appbuf) {
if ((ssize_t)len < 0)
htx_truncate_blk(htx, tailblk);
An HTX block pointer encodes a block *position* (the table is indexed backwards
from the end of the storage area), so it stops designating the same block as
soon as the table is compacted. htx_add_trailer() may reach
htx_reserve_nxblk(), which calls htx_defrag_blks() when the block table has
grown down to the payload while htx->head > 0, i.e. for a nearly full buffer
whose head was already consumed. All blocks then move down by the old value of
htx->head while the saved pointer does not follow, and htx_truncate_blk()
truncates at the wrong place, leaving partially converted trailers in the
message or removing valid blocks.
Only the trailers are really concerned: h3_resp_headers_to_htx() refuses to
work on a non-empty message, so no defragmentation can happen there, but it is
fixed the same way for consistency.
Let's save the amount of data present before the conversion and use
htx_truncate(), which works on a byte offset and is thus insensitive to any
block move, as htx_append_msg() already does.
This should be backported to all versions where the H3 trailers are supported,
so 2.8 and above.
BUG/MINOR: h2: don't use a block pointer to roll back a partial HTX conversion
h2_make_htx_request(), h2_make_htx_response() and h2_make_htx_trailers() save a
pointer on the tail HTX block on entry and use it on their error path to remove
everything they added:
An HTX block pointer is not a stable reference: it is computed from a block
*position* ("htx->blocks + htx->size - (pos + 1) * sizeof(struct htx_blk)"), so
it only designates the same block as long as the block table is not compacted.
htx_add_header()/htx_add_trailer() end up in htx_reserve_nxblk(), which calls
htx_defrag_blks() when the table has grown down to the payload while
htx->head > 0. After such a compaction all the blocks move down by the old
value of htx->head, but <tailblk> still points at the same address, hence at a
block located <head> positions further in the message. htx_truncate_blk() would
then truncate at the wrong place, either leaving partially converted headers or
trailers in the message, or dropping valid blocks that were there before.
This only concerns the trailers, and possibly the response on the backend side,
because the destination message may already hold payload with a partially
consumed head there, while a request is always converted into an empty buffer
which cannot defragment.
Let's save the amount of data present on entry and use htx_truncate() instead,
which relies on a byte offset and is therefore immune to any block move. This
is the same pattern as the one already used by htx_append_msg().
This has been there since the H2 to HTX conversion was introduced, so it should
be backported to all supported versions.
BUG/MINOR: h1: report the right error position on authority/host mismatch
When an absolute-form request target does not match the Host header value,
h1_headers_to_hdr_list() reports the error at two places depending on whether
the message must be blocked or only captured:
if (h1m->err_pos < -1) {
state = H1_MSG_LAST_LF;
ptr = host.ptr; /* Set ptr on the error */
goto http_msg_invalid;
}
if (h1m->err_pos == -1) /* capture the error pointer */
h1m->err_pos = v.ptr - start + skip; /* >= 0 now */
The strict path correctly points at the Host header value while the tolerant
path uses <v>, which at this point still holds the value of the *last* parsed
header field, whatever it was. So with "option
accept-unsafe-violations-in-http-request" enabled, the offset stored in
h1m->err_pos, later used by h1_capture_bad_message() and reported by "show
errors", designates an unrelated part of the message.
Let's use host.ptr in both paths.
This was introduced by commit 25bcdb1d9 ("BUG/MAJOR: h1: Be stricter on request
target validation during message parsing") in 3.0-dev12, so it should be
backported to 3.0 and above.
BUG/MINOR: http-htx: check the trash allocation in http_scheme_based_normalize()
http_scheme_based_normalize() uses a trash chunk to rebuild the target URI when
the default port must be dropped or when an empty path must be replaced by "/",
but it does not test the allocation:
struct buffer *temp = alloc_trash_chunk();
struct ist meth, vsn;
alloc_trash_chunk() takes its memory from pool_head_trash and returns NULL when
the pool is exhausted, in which case chunk_memcat() dereferences NULL and the
worker crashes. All the other users of alloc_trash_chunk() in the HTTP code
(http_act.c, http_ana.c) do test the result.
Normalization is performed on every absolute-form request URI (from H1, H2 and
H3), so this is reachable under memory pressure. Let's simply report a failure,
which the callers already handle as a rewrite error.
This was introduced by commit 4c0882b1b ("MEDIUM: http: implement scheme-based
normalization") in 2.5-dev2, so it should be backported to all supported
versions.
BUG/MINOR: http: fix an out-of-bounds read in http_get_host_port() on empty host
http_get_host_port() walks backwards from the end of the host looking for the
first non-digit, then checks whether it is a colon:
start = istptr(host);
end = istend(host);
for (ptr = end; ptr > start && isdigit((unsigned char)*--ptr););
/* no port found */
if (likely(*ptr != ':'))
return IST_NULL;
When <host> is empty, the loop condition fails immediately, the pre-decrement
is never evaluated and <ptr> is left equal to <end>, so *ptr reads one byte
past the end of the string. With an IST_NULL argument this is a NULL
dereference.
Both cases are reachable with an empty host: h1_validate_mismatch_authority()
calls it on the Host header value, which may be empty ("Host:\r\n") while an
absolute-form request URI is used, and http_scheme_based_normalize() calls it
on the authority extracted from the URI, which is empty for a request like
"GET http:///x HTTP/1.1". In practice the extra byte always lies inside the
request buffer, so the observable effect is limited to possibly mistaking a
neighbour byte for a colon and returning a bogus port, but it remains an
out-of-bounds read and the helper must be usable with an unset ist.
Let's return IST_NULL right away for an empty host.
This was introduced by commit 658f97162 ("MINOR: http: Add function to get port
part of a host") in 2.7-dev2, so it should be backported to 2.8 and above.
BUG/MINOR: http-fetch: fix a NULL channel dereference in smp_fetch_body()
smp_fetch_body() is also used from the health-check context, where the HTX
message comes from <check->bi> and where there is no channel at all: for
"res.body", SMP_RES_CHN(smp) evaluates to NULL because smp->strm is NULL. The
function is aware of this and guards the channel a few lines below:
if (!finished && (check || (chn && !channel_full(chn, global.tune.maxrewrite) &&
but the trash chunk retrieval isn't:
chk = get_best_trash_chunk(&chn->buf, htx->data);
so as soon as a check response is made of more than one HTX DATA block, a
bogus address close to NULL is dereferenced and the worker crashes. Let's use
the check input buffer when there is no channel, as done elsewhere.
Note that no reproducer could be built: with the h1 mux the successive DATA
blocks of a check response always end up appended to the same HTX block, so
the multi-block case was not reachable in practice. The dereference is
nevertheless plainly invalid.
This is a regression introduced by commit ac37158a6 ("BUG/MEDIUM: chunk:
Review chunks usage to not retrieve a large buffer by error") in 3.5-dev2,
which replaced get_trash_chunk_sz(htx->data), that did not look at the
channel, with get_best_trash_chunk(&chn->buf, htx->data). smp_fetch_body_param()
received the same change but is only reachable through "req.body_param", for
which <check> is always NULL and thus <chn> always set, so it is not affected.
It must be backported wherever the commit above was backported, so 3.4 and
above.
Only the upper bound is verified, and unlike "req.hdr" and friends, which rely
on val_hdr() to enforce the lower bound of the occurrence number, these two
keywords are declared with no argument checker at all. A negative identifier is
therefore accepted at boot, and at runtime req_cap[-1] is read; if the pointer
found there is not NULL it is then passed to strlen() and returned as a string.
This is trivially reproduced with a frontend containing:
capture request header Host len 32
http-request return status 200 hdr X-Cap "%[capture.req.hdr(-1)]"
which segfaults the worker on the very first request.
Since a negative capture identifier is meaningless, the cleanest fix is to
reject it at configuration parsing time, as is done for the header occurrence.
Let's add a val_cap_id() checker and reference it from both keywords.
Note that the "capture-req"/"capture-res" converters in http_conv.c index the
same array with an unchecked value too, but they are saved by the list walk
that precedes the access and which stops on a NULL <hdr>, so they only fail to
capture. They are left untouched.
This bug has been there since the keywords were introduced, so this should be
backported to all supported versions.
BUG/MEDIUM: http-fetch: don't parse a non-HTTP check buffer as an HTX message
In the health-check context, smp_prefetch_htx() unconditionally treats
<check->bi> as an HTX message:
if (!s || !chn) {
if (check) {
htx = htxbuf(&check->bi);
This is only true for an HTTP check, where the h1 mux fills the buffer with
HTX blocks. For any other check ruleset (plain "tcp-check", or the default TCP
connect check), the buffer holds the raw bytes received from the server.
All the response-side HTTP sample fetches (res.body, res.hdr, res.hdrs,
res.ver, status, res.cook...) are declared with SMP_SRC_HRSHV/HRSHP/HRSBO and
the validity table in sample.c makes them usable at the SMP_CKP_BE_CHK_RUL
checkpoint, so referencing one of them from a "tcp-check" rule (typically in an
"on-error" log-format string, or in a "status-code" expression) is accepted at
boot without any warning. When the check then runs, the first bytes of the
server's answer are used as "struct htx" fields: ->size, ->head, ->tail and
->first are entirely provided by the peer. htx_get_first_blk() computes
"htx->blocks + htx->size - (first + 1) * sizeof(struct htx_blk)" and the
resulting block is dereferenced, which is a wild read that crashes the worker,
and which may otherwise return arbitrary process memory as the fetched value.
and a server answering with 24 bytes crafted as a "struct htx" header with a
large ->size (e.g. 0x40000000), ->head = 0, ->tail = 1 and ->first = 0,
followed by any padding: the worker segfaults at the first check.
Let's simply refuse to look at the buffer when the check is not relying on
HTX. HTTP checks are unaffected, the fetches now simply return no sample on
other checks.
Note that this requires a hostile server, which is a trusted component for a
reverse proxy, so this is not a security issue, but the parser must not be fed
a buffer whose format it cannot assume.
This should be backported to all supported versions.
BUG/MINOR: http-htx: fix the length moved when removing a header value
http_remove_header() removes <len> bytes at the offset <off> of the header
value, <off> being "start - v.ptr" once <start> has been adjusted to also eat
the comma surrounding the removed value. The number of bytes that remain to be
moved down is therefore "v.len - off - len", but the function passes
"v.len - len" to memmove(), i.e. <off> bytes too many.
As a result, as soon as the removed value is not the first one of the
comma-delimited list (<off> != 0), memmove() reads <off> bytes past the end of
the header value and, when <off> is larger than <len>, writes (<off> - <len>)
bytes past the new end of the HTX block payload, silently corrupting whatever
follows it in the HTX buffer (typically the payload of the next blocks). For
instance with "x-test: aaaaaaaaaa,b,c", removing the second value gives
off = 10 and len = 2, hence 10 bytes read and 8 bytes written past the 20-byte
block payload. The resulting header value itself is not affected, which is
probably why this went unnoticed.
All the current callers happen to be safe: they either use <full> = 1 when
looking the header up, in which case the whole value matches and the block is
removed as a whole by the early return, or they only ever remove the first
value of the list ("Expect: 100-continue" in http_ana.c and "Age" in cache.c).
So this is only a latent out-of-bounds write for now, but the function is a
generic helper and the next caller removing a non-first value would corrupt
the message.
Let's simply compute the moved length from the end of the value. This can be
verified with the "http_htx" unit test added in the previous commit
(DEBUG_UNIT=1, then "haproxy -U http_htx").
This bug was introduced with the HTX conversion by commit 47596d378 ("MINOR:
http_htx: Add functions to manipulate HTX messages in http_htx.c") in 1.9-dev7
so it should be backported to all supported versions.
CLEANUP: haload: use <arg_thrd> instead of <global.nbthread> where applicable
Replace references to <global.nbthread> with <arg_thrd> across loops,
memory allocations, and thread ID distributions. Having two variables
sharing the same value and meaning is confusing, so it is cleaner to
rely exclusively on <arg_thrd>.
BUG/MINOR: haload: fix CPU topology detection by omitting forced "nbthread"
Always writing "nbthread" in the generated global configuration forced a
static thread count, preventing HAProxy from using its automatic CPU
topology detection and correct core binding. This caused severe performance
degradation on large multi-core machines. Fix this by omitting the "nbthread"
directive when -t is not explicitly specified, while ensuring <arg_thrd>
is properly initialized to global.nbthread for internal calculations such as
connection and request rates.
Also ensure <arg_thrd> <= <arg_usr> when -R is specified for internal
calculations.
MINOR: mux-h1: Use htx version to send default low-level errors
When an error was returned by the H1 multiplexer, the raw version was used
for default low-level errors. It is not really an issue, but it is more
consitent to use the HTX version. This way, by default, header names for
such errors are now sent in lower case. And the case of header names can
still be adjusted if necessary, thanks to the previous fix.
It is not really a bug, there is no reason to backport it, except if someone
ask for it.
MINOR: mux-h1: Lower the case for Sec-Websocket-* headers when manually added
During the message formatting, if websocket handshake key must be added by
the mux itself, the corresponding headers were not inserted in lowercase as
expected. It is not really an issue but it is not consistant with processing
on other headers. So let's fix it.
BUG/MEDIUM: mux-h1: Always adjust case for all outgoing headers as expected
In HTX, all header names are converted to lowercase. However, some legacy H1
applications are still sensitive to the header names case. For this
purpose, it is possible to provide a map to automatically adjust the case of
header names. While it is performed for most responses, it is not true for
the low-level errors triggered during requests parsing. It the same ways,
the case of "Sec-Websocket-Key" a "Sec-Websocket-Accept" headers were
adjusted as expected.
To fix this issue the h1-htx API was slightly changed. Now the map used to
adjust the case of header names, if any, is passed to the function
responsible to format the headers. In the H1 multiplexer, the map is first
retrieved then passed as argument to h1_format_htx_hdr() and
h1_format_htx_msg() functions. A NULL pointer is passed if no rewrite must
be performed.
In the H1 multiplexer, h1_adjust_case_outgoing_hdr() was replaced by
h1_get_hdrs_map().
All other calls to h1_format_htx_hdr() were adapted to use a NULL pointer
(httpclient, http-fetch).
This patch relies on "REORG: h1-htx: Move h1 headers map in h1-htx". Both
commits should be backported to all supported versions.
The map used to adjust the H1 headers case was moved in h1-htx part. For
this purpose h1_htx-t.h file was added and h1_hdrs_map and h1_hdr_entry
structures were moved into this file.
Tim Duesterhus [Wed, 15 Jul 2026 15:14:00 +0000 (17:14 +0200)]
MINOR: halog: Add support filtering on header capture values using -hdr-match
This patch extends the existing support for printing captured header fields
(`-hdr`) by a new filter (`-hdr-match`) that only processes lines where the
given capture has a specific value. It works together with all existing filters
and output formats.
The full syntax is `-hdr-match <block>:<field>=<value>`, where <block> and <field>
work just like `-hdr` and `<value>` is an exact string match:
Example:
capture request header a len 50
capture request header b len 50
capture request header c len 50
capture response header d len 50
capture response header e len 50
capture response header f len 50
- `-hdr-match 1:1=foo` will filter for requests where `a` is equal to "foo".
- `-hdr-match "2:3=foo bar"` will filter for requests where `f` is equal to
"foo bar".
The chosen syntax leaves future scope for allowing `<block>:<field>*<value>`
for substring matches and `<block>:<field>^<value>` for prefix matches without
introducing a breaking change.
Tim Duesterhus [Wed, 15 Jul 2026 15:10:36 +0000 (17:10 +0200)]
CLEANUP: halog: Clean up naming for variables related to `-hdr` processing
This is in preparation of allowing to filter based on the values of a header
capture by having a common "capture" prefix for variables related to extracting
header captures.
Certain rare bugs may cause an fd_want_recv() or any of the other
operations being done on a closed FD. This triggers a BUG_ON() on the
next call trying to insert the FD but it's too late to figure when
this happened. Let's just add some BUG_ON_HOT() to detect an attempt
to modify the state of a closed FD so that the culprit is detected.
It will only be enabled with DEBUG_STRICT=2, since by design this may
never happen so it's not needed to enable it in default builds. It was
verified not to trigger on various tests.
OPTIM: tools: keep a cache of recent localtime() and gmtime()
The log subsystem already keeps a cache of the latest generated time
header to avoid paying the price of snprintf() notably. However, as
reported in GH issue #3444 by @zino7825, localtime() and gmtime() are
affected (at least in glibc) by a tzset_lock held during the call to
__tz_convert(), which ruins performance when logging time such as the
accept date. Even just "option httplog" sees its performance divided
by two on a 64-core machine, from 1.8M to 910k req/s.
The following config even goes down to 388k req/s:
defaults
mode http
frontend fe
bind :8001
log 127.0.0.1:5514 len 8192 local0
log-format '{"t":"%t","tr":"%tr","ci":"%ci"}'
http-request set-var(txn.now) date(),ltime("%Y-%m-%dT%H:%M:%S")
option httplog
default_backend be
backend be
server s1 198.18.0.31:8000
With native_queued_spin_lock_slowpath() taking 80% of the CPU, showing
that it's also involved in futexes.
This patch, suggested by @zino7825, implements a very simple thread-
local cache for the 4 previously seen values for both localtime() and
gmtime(). The cache is visited in reverse order so that most recently
updated values are visited first (the most likely ones to be used). A
test with the config above and 12.8M requests showed 38.4M lookups with
only 12k total misses. A more naive scan from 0 to N caused 59M misses.
With this patch, all variants of the tests above remain at native speed
without native_queued_spin_lock_slowpath() being noticeable at all. If
for any reason a log format was so complex that it needed more stored
local times, it would be easy to change the cache size by redefining
TIME_CACHE_BITS.
The functions are no longer inlined, they were moved to tools.c since
we'd rather avoid loops and complex constructs in inlined functions.
Co-authored-by: Jinho Kong <zino7825@users.noreply.github.com>
OPTIM/MEDIUM: proxy/server: avoid server list reordering on startup
Prior to this patch, each server parsed from the configuration was added
at the front of the proxy list. The list was then reversed once parsing
is finished to reflect the configuration order.
Now that proxy servers list has been converted to a doubly linked list,
there is no more a reason for this. Thus, this patch changes the server
insertion order on configuration parsing : this is now performed
directly at the end of the proxy list. Reversal is unnecessary and has
been removed, so post-config performance may be slightly improved.
Peers parsing is the only module which relies on the order insertion.
Thus it has been adapted to now use the last server in its proxy.
MAJOR: proxy: convert server list to a doubly linked struct list
Servers are stored in a list in their parent proxy. Prior to this patch,
this list was singly linked.
This patch converts the proxy server list to a doubly linked struct
list. Server <next> pointer is replaced by a struct list <el_px> attach
point.
The main benefit from this patch is that it removes the bottleneck
performance for add and delete server operations at runtime. As with
main proxies list conversion, this is labelled as major as it is an API
change.
Most of the changes are straightforward : for/while statements are
replaced by list_for_each_entry() macros. LIST_ISEMPTY() is now used to
detect if a proxy does not contain any server.
Server insertion at the front position during config parsing is kept at
the moment, with reordering on post parsing. With the current patch,
this is not strictly necessary so this will be removed in a next change.
MINOR: proxy: define server list iteration functions
Define wrappers function to iterate over a proxy servers list. These
functions are used when a standard for loop over the whole list is not
desirable.
This patch will ease the conversion of the proxy servers list into a
doubly linked struct list type, as the final conversion patch changes
will be smaller.
MINOR: server: do not return next server on srv_drop()
Previously, srv_drop() returned the next server entry in the parent
proxy. This was used as convenience during server iteration. However,
the code has evolved several times to better deal with the server
deletion risk. Currently, srv_drop() return value is only used in a
single place.
Thus, this patch simplifies srv_drop() by removing its return value. As
a side effect, this will also reduce the modification required to
convert the proxy server list to a doubly linked struct list.
MINOR: server: rename global servers_list to all_servers
Rename global <servers_list> to <all_servers>. This name better reflects
that it contains all the servers and servers a similar purpose to
<all_proxies> list.
BUILD: ssl: Do not use SSL3_MT_KEY_UPDATE, hardcode 24 instead
Not every SSL lib provides SSL3_MT_KEY_UPDATE, so just hardcode 24
instead.
This should be backported up to 2.8, when 91004114fe8816f848025fe71def4ea23e72a5f6 will be backported.
BUG/MEDIUM: ssl: Put CO_ER_SSL_KEYUPDATE at the right place
FOr some reason, CO_ER_SSL_KEYUPDATE has been put in the middle of the
enum containing all the different possible connection errors, but
reg-tests depend on their numeric value, so it broke them. Put it at the
end of the enum instead.
MEDIUM: ssl: Add a way to rate-limit TLSv1.3 KeyUpdate
Processing TLSv1.3 KeyUpdate is expensive in term of CPU, and in normal
usage there is very few reason to get a lot of them. So add a new
keyword, tune.ssl.keyupdate-rate-limit, that gives the maximum number of
KeyUpdate we're okay with receiving per second. The default is 100,
which should be enough. 0 means no rate-limiting at all.
This should mitigate the problem reported in Github issue #3450.
BUG/MEDIUM: ssl: Handle non-application data record while splicing
When using splicing with kTLS, if we receive a record that is not an
application data record, such as a KeyUpdate, then splicing will fail.
If that happens, temporarily disable splicing and go the regular way so that
recvmsg() is used, we get the record, and we can resume splicing.
Please not that KeyUpdate is still not handled with AWS-LC. Only recent
Linux kernels support it, and the code hasn't been written for that yet.
Use HAVE_VANILLA_OPENSSL, which is what we actually define when we're
linked against OpenSSL, and not USE_VANILLA_OPENSSL.
Not using the right macro means we may start splicing with kTLS while
there are still data available in OpenSSL's internal buffers, leading to
data not being properly read.
Rename global list <proxies> to <all_proxies>. This better highlights
the difference with the other list <main_proxies>. The first one contain
every proxy instances, whereas the second list is only a subset of user
visible proxies.
In the future, it should be sufficient to only keep <all_proxies> list.
However, this requires careful code analysis, in particular to ensure
proxies iteration are always executed with the proper capabilities
filter.
CLEANUP: proxy/config: clean up after proxies list conversion
All proxies list (main proxies, log forward and sink) have been
converted to the doubly linked standard list type. Proxy <next> member
is now unneeded, it is thus removed from the structure.
This patch also removes obsolete _get_next_proxy() wrapper used during
check_config_validity().
OPTIM/MEDIUM: proxy: avoid main proxies list reordering on startup
Prior to this patch, proxies were added in front ot the visible proxies
list, resulting in the reverse order of their parsing. The list was then
reversed once parsing was completed to reflect the configuration order.
This was performed for performance reason, as visible proxies list was
simply linked.
This list has recently been converted to a doubly linked struct list.
Thus, it is now possible to append a newly parsed proxy at the end of
the list in constant time.
Thus, with this patch, a newly parsed instance is now directly inserted
at the end of main proxies list. This is implemented by updating
main_proxies_register() wrapper. Post-config performance are slightly
improved as reversal is now unneeded.
Amaury Denoyelle [Fri, 16 Jan 2026 08:26:56 +0000 (09:26 +0100)]
MAJOR: proxy: convert proxies_list to a doubly linked struct list
Frontend, backend and listen proxies are all stored in <proxies_list>
which contain all user visible entries. This list is read in several
places, for example during stats dump.
Dynamic backend have just been introduced in the previous 3.4 release,
with creation and removal of new instances at runtime. The main
bottleneck of these operation is the manipulation of <proxies_list> due
to its singly linked nature.
This patch converts <proxies_list> into a doubly linked standard list
type. Proxy member <el> is reused as attach point. It is already used
for internal proxies in other list, not intersecting with this usage.
The main benefit of this change is to improve performance for add and
delete backend operations. It is labelled as major as it may break
external components which manipulate this list.
Also, old naming <proxies_list> has been replaced by <main_proxies>.
This helps to differentiate with the already existing superset <proxies>
list which contain all proxies instances, even internal ones. In the
future, it should be sufficient to only keep the latter list. However,
this requires careful code analysis, in particular to ensure proxies
iteration are always executed with the proper capabilities filter.
Proxies reversal on post parsing has been updated to use the new list
format. This will be removed in a latter patch as it is now possible to
insert proxies in the order of their parsing in constant time. Most of
the other changes are straightforward : for/while statements are
replaced by list_for_each_entry(). For incomplete iteration, recently
introduced iteration wrappers are updated to use LIST macros.
Define wrappers function for <proxies_list> iteration. These functions
are used when a standard for loop over the whole list is not desirable.
The objective of this patch is to ease the conversion of <proxies_list>
from a singly linked list to a standard doubly linked list type. This
will help to keep the final conversion patch smaller.
Also, iteration over struct list is sligthly more complicated as it
requires more operation, for example for list end detection. Thus, the
wrappers will become handy to avoid tedious code repetion.
MINOR: proxy: centralize proxies_list insert during config parsing
Define a new function wrapper mainpx_register() which is used to insert
a proxy instance in <proxies_list> which contain all user visible
entries. Insertion is performed at the front of the list. Thus it must
only be used during configuration parsing, before proxies list reversal.
The main objective of this patch is to simplify the list conversion to a
doubly linked one by centralizing the required changes in a single
function. A nice benefit is that it is now easier to determine the exact
proxies list composition.
MINOR: sink: convert list to standard doubly linked one
This patch is similar to the previous one, dealing this time with sink
proxies stored in <sink_proxies_list>.
This list is converted to a doubly linked standard list type, in
preparation to the conversion of the visible proxies list. The objective
is similar : code simplification for check_config_validity() when
looping over several proxies list.
MINOR: log: convert list to standard doubly linked one
Log-forward proxies are not stored in <proxies_list> alongside other
visible user proxies. Instead, they are stored in a dedicated list
<cfg_log_forward>.
As with the visible proxies list, log forwarders list was defined as a
simple singly linked one. This patch converts it to the doubly linked
standard list type, frequently used in haproxy source code. Proxy <el>
member is reused as list attach point. It was only used by defaults
instances prior to this patch.
This patch does not bring noticeable change. However it will simplify
visible proxies list similar conversion as both lists are looped over
together during check_config_validity().
MINOR: config: define wrapper for proxies loop during check config
On check_config_validity(), proxy post init code is reexcuted over
several proxies list : the main proxies list, log sections and sinks.
This patch defines an internal wrapper to help performing this
iteration. It will become necessary once each list is converted from a
singly linked format to a doubly linked struct list. Once all lists are
converted, the wrapper should be removed as it will be once again
unnecessary.
Adjusts configuration parsing for log-forward sections : replace
<cfg_log_forward> head pointer list usage in individual keywords parser
by <curproxy> global variable. This parsing patterm is already
implemetned in cfgparse-listen module.
This is only a refactoring change without any visible effect. It is a
preparatoy step before converting <cfg_log_forward> to a full doubly
linked struct list.
MINOR: haload: add rate limiting support using -R option
Implement request/connection rate limiting per second using the new
-R command line option <arg_rate>.
Instead of reinventing the wheel, this implementation aligns with the
h1load approach by relying on existing frequency counters. Dedicated
tasks <hld_rate_task> are spawned on each thread solely to manage and
throttle connection initializations based on <arg_rate>.
The counter <running_usrs> is renamed to <running_tasks> to track both
users and connection rate-limiting tasks.
BUG/MINOR: haload: fix display glitches by flushing stdout in summary
Fix a bug in hld_summary() where the output stream was not explicitly
flushed after printing the periodic statistics line. This caused text
buffering issues and visual overlapping in the terminal output.
By default <arg_thrd> was unset (-1) and fell back to global.nbthread,
causing inconsistencies since <arg_usr> defaults to 1. This patch sets
the default thread count to 1, enforces that <arg_thrd> does not exceed
<arg_usr>, updates usage help, and writes nbthread unconditionally.
BUG/MINOR: haload: fix use-after-free upon updating task expiration
Fix a bug in hld_strm_task() where hs was accessed after it had already been
freed via hldstream_free(&hs). The task expiration update and queueing
now safely rely on <usr> rather than <hs->usr>.
BUG/MEDIUM: proxy: protect "show errors" against backend deletion
Command "show errors" is not safe as it loops over the list of proxies.
A crash may occur if the task yields on a proxy which is deleted just
before the dump is restarted.
As with previous fixes dealing with similar cases, this is fixed via the
watcher mechanism. A new <px_watcher> member is defined in
<show_errors_ctx> and used to loop over the proxies list.
Released version 3.5-dev3 with the following main changes :
- BUILD: haload: Increase a buffer size so that gcc will stop complaining
- BUILD: task: Fix build when no 8B CAS is available at all
- BUG/MINOR: hlua: Apply socket timeout on server side only
- BUG/MEDIUM: applet: Reenable reads in applet context if requesting a connection
- MINOR: stats: factor the proxy vs scope check into its own function
- BUG/MEDIUM: stats: subject "stats admin" accesses to "stats scope" filtering
- BUG/MINOR: shctx: fix shctx_row_data_get() when offset exceeds a block
- MINOR: shctx: clamp shctx_row_data_get() reads against the offset
- DOC: stats: document the per-proxy byte count fields in the CSV list
- BUG/MEDIUM: cache: reattach the row when a secondary entry is incomplete
- MINOR: http: add two header parsing functions
- MINOR: cache: minor changes ahead of support for sending early hints
- MINOR: cache: add config options for early hints support
- MINOR: cache: add helper functions for early hints support
- MINOR: shctx: allow consumers to customize eviction strategy
- MINOR: cache: track full and hints entries in per-pool LRU lists
- MEDIUM: cache: early hints-aware eviction code
- MINOR: cache: indicate whether entries are stripped or not
- MINOR: http: factor 103 emission into start/end helpers
- MEDIUM: cache: emit early hints if configured to do so
- MINOR: cache: add a counter for cache hits serving early hints
- MINOR: cache: allow opting out of early hints at the rule level
- MINOR: cache: allow customizing ratio for early hints
- REGTESTS: cache: validate the emission of 103s
- MINOR: cache: factor cache_extract_link_hints out of cache_extract_hints
- MEDIUM: cache: add support for hints-only HTTP caches
- MEDIUM: ssl: introduce src/fips.c with TLS version check
- MEDIUM: ssl: add FIPS TLS 1.2 cipher check for AWS-LC
- MEDIUM: ssl: set FIPS-approved cipher defaults for AWS-LC FIPS builds
- MEDIUM: ssl: add FIPS TLS 1.3 ciphersuite check for AWS-LC
- MEDIUM: ssl: set FIPS-approved curve defaults for AWS-LC FIPS builds
- MEDIUM: ssl: add FIPS elliptic curve check for AWS-LC
- MEDIUM: ssl: set FIPS-approved sigalgs defaults for AWS-LC FIPS builds
- MEDIUM: ssl: add FIPS signature algorithm check for AWS-LC
- CLEANUP: cache: align the cache_hint_hits increment with its siblings
- BUG/MEDIUM: stats: Ensure that Origin is valid on POSTs
- DOC: stats: Document that stats admin is vulnerable to a CSRF attack
- BUG/MEDIUM: ssl-gencert: Don't forget to free memory when done
- BUG/MEDIUM: protobuf: adjust sample size capacity after pointer shift
- BUG/MEDIUM: protobuf: fix nested path bypass in field lookup
- BUG/MINOR: ssl: fix proxy lookup for show ssl sni
- REGTESTS: protobuf: add regression test for nested vs flat paths
- DEBUG: add BUG_ON_STATIC(): a compile-time BUG_ON()
- CLEANUP: event_hdl: Use BUG_ON_STATIC()
- DOC: internals: update core-principles with initializations
- BUG/MINOR: mux-h1: Don't delay send if message with c-l was fully sent
- BUG/MEDIUM: net-helper: Adjust sample size capacity after pointer shift
- BUG/MEDIUM: sample: Adjust sample size capacity after pointer shift for bytes()
- BUG/MEDIUM: sample: Adjust sample size capacity after pointer shift for ltrim()
- BUG/MINOR: sample: Fix bytes() when length it greater than remaining data
- BUILD: Makefile: error when trying to build with aws-lc with the wrong flags
- DOC: fix typo in "del ssl ech" command
- DOC: remove outdated experimental mention on dynamic backends
- BUG/MEDIUM: proxy: protect "show servers ..." against server deletion
- BUG/MEDIUM: proxy: protect "show servers ..." against backend deletion
- BUG/MEDIUM: proxy: protect show backend against be deletion
- MINOR: proxy: stress CLI commands with backends/servers loop
- BUG/MEDIUM: server: Properly check for streams before deletion
- BUG/MINOR: resolvers: do not index resolvers names in the proxies
BUG/MINOR: resolvers: do not index resolvers names in the proxies
Resolvers are self-sustaining sections that have their own list, but for
the purpose of supporting TCP connections, a backend proxy is created
with them, and it holds the same name as the section. Since we started
to create a "default" resolvers section, a similarly named backend in
TCP mode automatically appeared, resulting in issues such as the one
described in GH #3445:
defaults
mode http
frontend public
bind :8001
default_backend default
It yells: "... tries to use incompatible tcp proxy 'default' (<internal>:0)
in a 'use_backend' rule (see 'mode')" because it in fact finds the hidden
resolvers proxy. Depending on versions, adding such a default backend makes
the issue go away or not.
Worse, the following config:
frontend public
bind :8001
default_backend default
actually routes the traffic to the local resolvers :-(
Resolvers do not need to appear in proxies index tree, as they're looked
up by find_resolvers_by_id() which iterates over all resolvers sections
(i.e. tree not used), so we can safely avoid indexing them.
Some other proxies are now protected against this via the use of PR_CAP_INT
which also prevents indexing, but that flag changed multiple times along
versions and is extremely sensitive in older ones, to the point of not
being suitable for a backport.
Instead, commit 116983ad94 ("MEDIUM: cfgparse: do not store unnamed
defaults in name tree") added the following test to refrain from
indexing an unnamed proxy in setup_new_proxy():
- if (name && !(cap & PR_CAP_INT))
+ /* Internal proxies or with empty name are not stored in the named tree. */
+ if (name && name[0] != '\0' && !(cap & PR_CAP_INT))
proxy_store_name(px);
if (!(cap & PR_CAP_DEF))
This part is totally suitable for backporting and permits to simply
change the resolvers' proxy creation to pass NULL or an empty string
instead of the resolvers section's name so that we don't index that
proxy. That's what this patch does. For versions before 3.4, the
commented patch above must be explicitly backported. This patch must
be backported to 3.0. Thanks to Lukasz Nowak for reporting this bug.
It's worth noting that this area is really falling apart and urgently
needs attention. The internal API is currently too limited to permit
reliable lookups, which is the reason why so many proxies are hidden
that way. In an ideal world, all names would be indexed and caps and
modes properly set to unambiguous values, so that lookup functions
only look up specific caps and modes, and support being passed a mask
to avoid matching undesired caps. At the very least a mask should be
given in lookups, and a few extra caps should permit to distinguish
regular proxies (those declared with the "frontend" and "backend"
keyword, supporting use_backend) from other ones (rings, resolvers,
log-forward etc).
BUG/MEDIUM: server: Properly check for streams before deletion
When a server delete command happens, before we actually delete the
server, we check if there are any streams still attached to the server,
and will refuse the deletion if it is so. Unfortunately, the check was
not exhaustive, we would check if there's any stream in the queue, if
there are any connection to the server established, and if served is
non-zero.
But there are at least two cases where a stream will have a reference to
the server, but none of those will be true : the first one is the case
where we tried a connection to the server, and it failed. At this point
we decremented served, but we will still hold a reference to the server
in target until a new server has been assigned. The second one happens
with cookie persistence, in which case target will be set to the server
address way before any connection is attempted, and so way before served
is incremented.
To fix that, introduce a new per-thread-group counter in the server
structure, nb_strm, that counts the streams whose target points to that
server. It is incremented when a stream's target is set to a server, and
decremented when the target changes to another server, to a non-server,
or the stream ends. To make this reliable, all assignments to a stream's
target now go through the new stream_set_target() helper, which keeps
nb_strm up to date and also clears sv_tgcounters when the target is no
longer a server, so we don't keep a dangling reference to the server's
counters (its failure counters have already been accounted for by that
point). To check if there are still any streams attached to the server,
the code just looks at all the per-thread-group counters, and if one of
them is non-zero, there are still streams on that server. This will
always be accurate because the server deletion command runs with thread
isolation.
MINOR: proxy: stress CLI commands with backends/servers loop
Update handlers for "show backend" and "show servers state/conn" so that
stress mode can be used to force the task yielding each time a new
instance is dumped.
This is useful for debugging commands safety with backend/server
deletion in parallel.
BUG/MEDIUM: proxy: protect show backend against be deletion
Command "show backend" loops over the list of visible proxies to display
information about all the backend instances. This command may yield in
case of a long output, but this is not safe with introduction of dynamic
backend deletion at runtime.
Fixes this by using the watcher mechanism, similarly to what is
implemented for stats dump. To support this, a new context dedicated to
"show backend" has been defined.
BUG/MEDIUM: proxy: protect "show servers ..." against backend deletion
This patch is a direct follow up of the previous one which fixes
commands "show servers conn/state". The current patch ensures that these
commands are safe even with task yielding and backend deletion in
parallel. This is also based on the watcher mechanism.
BUG/MEDIUM: proxy: protect "show servers ..." against server deletion
Command show servers conn/state loops over a server list to display
various information. This command may yield in case of a large output.
This can cause a crash if the server on which the command was paused is
deleted prior to restarting the command.
Fixes this by using the watcher mechanism, similar to what is already
implementing with stats dump.
This must be backported at least up to 3.0. For older releases, watcher
mechanism is not available, server refcount + SRV_F_DELETED flag must be
used instead.
DOC: remove outdated experimental mention on dynamic backends
Dynamic backends commands were first implemented with experimental
status guard. However, this status was removed before the final release
as it is considered stable enough.
This patch cleans up outdated references to experimental status for
add/del backend commands in management documentation.
BUILD: Makefile: error when trying to build with aws-lc with the wrong flags
USE_OPENSSL_AWSLC=1 is required when trying to build with aws-lc. This
was already required before, but kind of worked without it since f76e8e50f460 ("BUILD: ssl: replace USE_OPENSSL_AWSLC by
OPENSSL_IS_AWSLC").
However since 9ac590b5 ("MEDIUM: ssl: introduce src/fips.c with TLS
version check"), this can't work by accident anymore.
This patch emits an error when we find OPENSSL_IS_AWSLC, but we didn't
found USE_OPENSSL_AWSLC=1.
BUG/MINOR: sample: Fix bytes() when length it greater than remaining data
For bytes() converter, when the length parameter is greater than the
remaining data, no truncation on length must be performed. However, in that
case, nothing was performed at all. The changes because of the offset
parameter was just ignored.
Now, when the length value is too large, the sample data are moved
accordingly to the offset value as expected.
The coresponding reg-test was updated to test this case with a non-zero
offset. In addition the documentation was fixed to properly match what the
converter do. It was wrongly updated when the support of variables was
introduced in 2.9.
BUG/MEDIUM: sample: Adjust sample size capacity after pointer shift for ltrim()
For the ltrim() converter, on success, the area pointer of the sample buffer
is moved forward without updating the buffer size accordingly. If this
converter is followed by another one relying on the buffer size to do some
operations on the buffer area, this could lead to a buffer overflow.
Thanks to Charles Vosburgh <theminershive@gmail.com> for reporting this and
providing the fix.
This patch must be backported to all supported versions.
BUG/MEDIUM: sample: Adjust sample size capacity after pointer shift for bytes()
For the bytes() converter, on success, the area pointer of the sample buffer
is moved forward without updating the buffer size accordingly. If this
converter is followed by another one relying on the buffer size to do some
operations on the buffer area, this could lead to a buffer overflow.
Thanks to Charles Vosburgh <theminershive@gmail.com> for reporting this and
providing the fix.
BUG/MEDIUM: net-helper: Adjust sample size capacity after pointer shift
eth.data, eth.src and ip.data converters are concerned. On success, the area
pointer of the sample buffer is moved forward without updating the buffer
size accordingly. If these converters are followed by another one relying on
the buffer size to do some operations on the buffer area, this could lead to
a buffer overflow.
Thanks to Charles Vosburgh <theminershive@gmail.com> for reporting this and
providing the fix.
BUG/MINOR: mux-h1: Don't delay send if message with c-l was fully sent
In the H1 multiplexer, if a message was fully sent considering the specified
content-length, the MSG_MORE flag must be unset, regardless the upper layer
said. This should avoid a extra latency of 200ms when this happens while the
HTX EOM flag is not set on the message. This may happen for H2/H3 messages
with trailers or if the END_OF_STREAM was not reported yet on H2/H3 side. In
both cases, on H1 side, no more data will be sent (trailers will be
dropped). So we must never delay the send.
This patch should fix the issue #3447. It must be backported to all
supported versions.
Maxime Henrion [Thu, 16 Jul 2026 16:18:17 +0000 (12:18 -0400)]
DEBUG: add BUG_ON_STATIC(): a compile-time BUG_ON()
Fails the build when <cond>, a constant expression, is true. Maps to
_Static_assert() on C11+; older dialects declare a negatively-sized
extern array instead, which emits no storage nor symbol. Usable at
file and block scope.
REGTESTS: protobuf: add regression test for nested vs flat paths
This patch adds a new VTC regression test to validate the protobuf()
sample converter and prevent future regressions.
It tests four distinct scenarios to ensure correct behavior:
- validating legitimate nested path extraction (e.g., path "1.2") on a
nested payload.
- ensuring flat path extraction (path "2") is blocked when the field is
actually inside a nested payload.
- validating legitimate flat sibling extraction (path "2") on a flat
payload.
- ensuring nested path extraction (path "1.2") is blocked when run on
a flat sibling payload which was fixed by this commit
BUG/MEDIUM: protobuf: fix nested path bypass in field lookup
Command "show ssl sni" accepts an extra option -f to restrict output to
a single frontend. If the specified proxy is not found, an error message
should be displayed. However, this does not behave as expected as no
error is reported and the first proxy in the list is used as a sort of
fallback.
This patch fixes the command parsing function when -f is used. The loop
is now interrupted as soon as a matching entry is found. After the loop,
if local variable <px> is still NULL, it indicates that no matching
entry was found. The error message is displayed as intended and the
command is not executed.
BUG/MEDIUM: protobuf: fix nested path bypass in field lookup
The previous implementation of protobuf_field_lookup() was susceptible
to a security bypass where flat sibling fields could be incorrectly matched
as nested children. This occurred because the parser did not strictly
enforce hierarchical boundaries and depth matching during sequential
iteration, allowing flat structures to satisfy nested path requirements (e.g.
treating a root-level sibling field as if it were nested under a parent).
This patch refactors the lookup logic into a strict, iterative stream
parser that progresses through the defined configuration path (depth).
The new logic guarantees that:
- intermediate matching parent fields must be of type LENGTH_DELIMITED.
If an intermediate match is found but is a primitive type, the parsing
aborts immediately.
- when descending into an intermediate parent, the scanning boundary
(*len) is strictly narrowed down to the nested sub-message size (vlen)
with an upfront validation against buffer overflow (vlen > *len).
- sibling fields at the current level are skipped cleanly without polluting
higher or lower hierarchy levels.
This implementation achieves absolute stack safety (O(1) memory complexity,
no recursion) and guarantees strict validation of nested Protobuf paths.
Many thanks to Red Hat and AISLE Research for reporting this.
BUG/MEDIUM: protobuf: adjust sample size capacity after pointer shift
This bug impacts "protobuf" and "ungrpc" sample fetches.
When extracting fields from a payload, the data pointer
"smp->data.u.str.area" is forwarded to point directly onto the decoded
sub-field data. However, the sample's maximum capacity size tracking property
"smp->data.u.str.size" was left untouched, remaining set to the total
original buffer size (e.g., tune.bufsize).
This creates an architectural size mismatch. Subsequent processing layers or
converters (such as stick-tables executing padding mechanisms via memset())
may assume that the original total capacity is still fully available starting
from the new forwarded offset. In specific configurations involving writable
cloned buffers (e.g., combining "lower" and "protobuf" converters before
tracking key storage), this logical flaw can easily lead to out-of-bounds
heap memory corruption or crashes.
Under rare conditions, depending on the specific combination of frontend
configurations (e.g., use of wait-for-body rules) and the alignment of
incoming HTTP requests containing specific chunked or large payload patterns,
this out-of-bounds write could lead to an immediate haproxy process crash
or data corruption.
Fix this by introducing "protobuf_adjust_smp_size()". This inline function
safely computes the consumed byte offset after each pointer shift and
decrements the sample's residual capacity size accordingly before updating
the payload reference. All "protobuf_smp_store_*" callbacks are updated.
Many thanks to Red Hat and AISLE Research for reporting this.
BUG/MEDIUM: ssl-gencert: Don't forget to free memory when done
In ssl_sock_do_create_cert(), don't forget to free ctmp and tmp_ssl once
we're done creating the certificate, otherwise we will get a memory leak
each time we have to create a new certificate, which will happen each
time a new SNI is used.
This should be backported as far as 2.6 (though on older releases,
ssl_gencert.c doesn't exist, and that function is found in ssl_sock.c).
This patch was submitted by Red Hat and AISLE Research
DOC: stats: Document that stats admin is vulnerable to a CSRF attack
Document that stats admin is vulnerable to a CSRF attack that can't be
totally mitigated, so if that feature really has to be used, precautions
must be taken.
This should be backported up to 2.6.
Many thanks to Red Hat and AISLE Research for reporting this.
BUG/MEDIUM: stats: Ensure that Origin is valid on POSTs
When receiving a POST on the stats interface, make sure that Origin (or
Referer if no Origin is found) matches the Host, to attempt to mitigate
a CSRF attack.
This should be backported up to 2.6.
Many thanks to Red Hat and AISLE Research for reporting this.
MEDIUM: ssl: add FIPS signature algorithm check for AWS-LC
Add ssl_fips_check_sigalgs() which validates the configured signature
algorithm list against the FIPS-approved set: ECDSA on NIST P-curves
with SHA-256/384/512, RSA-PSS (rsae and pss variants) with SHA-256/
384/512, and RSA-PKCS1 with SHA-256/384/512. SHA-1 based algorithms
and non-FIPS primitives (ed25519, ed448) are rejected.
The check uses the same strchr-based string parsing as
ssl_fips_check_ciphersuites(). A NULL list is silently accepted since
the global defaults were already overwritten with FIPS values at init
time.
The check is called right after SSL_CTX_set1_sigalgs_list() and
SSL_CTX_set1_client_sigalgs_list() in both the bind
(ssl_sock_prepare_ctx) and server (ssl_sock_prepare_srv_ssl_ctx)
configuration paths.
MEDIUM: ssl: set FIPS-approved sigalgs defaults for AWS-LC FIPS builds
When AWS-LC is built in FIPS mode, unconditionally override the
compile-time signature algorithm defaults with the FIPS-approved set
before config parsing. Explicit ssl-default-{bind,server}-sigalgs
keywords in the global section still take precedence over these
defaults.
The approved set is defined as macros in include/haproxy/defaults.h
alongside the existing CONNECT/LISTEN_DEFAULT_FIPS_CIPHERS family:
CONNECT/LISTEN_DEFAULT_FIPS_SIGALGS - ECDSA (P-256/384/521),
RSA-PSS and RSA-PKCS1
with SHA-256/384/512
CONNECT/LISTEN_DEFAULT_FIPS_CLIENT_SIGALGS - same set for client
certificate sigalgs
SHA-1 based algorithms and non-FIPS primitives (ed25519, ed448) are
excluded from the defaults.
MEDIUM: ssl: add FIPS elliptic curve check for AWS-LC
Add ssl_fips_check_curves() which validates the configured curve list
against the FIPS-approved NIST P-curves (P-256, P-384, P-521). Each
colon-separated name is resolved to a NID via OBJ_txt2nid() so all
standard aliases (P-256, prime256v1, secp256r1) are handled uniformly.
A NULL list is silently accepted since the global defaults were already
overwritten with FIPS values at init time.
The check is called right after SSL_CTX_set1_curves_list() in both the
bind (ssl_sock_prepare_ctx) and server (ssl_sock_prepare_srv_ssl_ctx)
configuration paths.
MEDIUM: ssl: set FIPS-approved curve defaults for AWS-LC FIPS builds
When AWS-LC is built in FIPS mode, unconditionally override the
compile-time curve defaults with the FIPS-approved NIST P-curves
before config parsing. Explicit ssl-default-{bind,server}-curves
keywords in the global section still take precedence over these
defaults.
The approved set is defined as macros in include/haproxy/defaults.h
alongside the existing CONNECT/LISTEN_DEFAULT_FIPS_CIPHERS family:
CONNECT/LISTEN_DEFAULT_FIPS_CURVES - P-256, P-384, P-521
This ensures that internal servers (httpclient, Lua SSL sockets) that
inherit global defaults also operate with FIPS-compliant curve lists
without requiring explicit configuration.
MEDIUM: ssl: add FIPS TLS 1.3 ciphersuite check for AWS-LC
AWS-LC does not expose TLS 1.3 ciphersuites set via SSL_CTX_set_ciphersuites()
through SSL_CTX_get_ciphers(), so the existing NID-based cipher check in
ssl_fips_check_ciphers() cannot catch non-FIPS TLS 1.3 suites. This is
further compounded by a defect in the AWS-LC-FIPS 3.x branch where TLS 1.3
ciphers are missing from SSL_get_ciphers() entirely (fixed in
https://github.com/aws/aws-lc/pull/2092), making any SSL_CTX-based
inspection unreliable across versions.
Add ssl_fips_check_ciphersuites() which validates the ciphersuite string
directly against a FIPS-approved allowlist (TLS_AES_128_GCM_SHA256 and
TLS_AES_256_GCM_SHA384). A NULL string is silently accepted since the
global defaults were already overwritten with FIPS values at init time.
The new check is called right after SSL_CTX_set_ciphersuites() in both
the bind (ssl_sock_prepare_ctx) and server (ssl_sock_prepare_srv_ssl_ctx)
configuration paths.
MEDIUM: ssl: set FIPS-approved cipher defaults for AWS-LC FIPS builds
When AWS-LC is built in FIPS mode, unconditionally override the
compile-time cipher defaults with FIPS-approved sets before config
parsing. Explicit ssl-default-{bind,server}-ciphers{suites} keywords
in the global section still take precedence over these defaults.
The approved sets are defined as macros in include/haproxy/defaults.h
alongside the existing CONNECT/LISTEN_DEFAULT_CIPHERS family:
CONNECT/LISTEN_DEFAULT_FIPS_CIPHERS - AES-128-GCM-SHA256 and
AES-256-GCM-SHA384 (TLS 1.2)
CONNECT/LISTEN_DEFAULT_FIPS_CIPHERSUITES - TLS_AES_128_GCM_SHA256 and
TLS_AES_256_GCM_SHA384 (TLS 1.3)
This ensures internal servers (httpclient, Lua SSL sockets) that
inherit global defaults also operate with FIPS-compliant cipher lists
without requiring explicit configuration.
MEDIUM: ssl: add FIPS TLS 1.2 cipher check for AWS-LC
Add ssl_fips_check_ciphers() to check the cipher list already loaded
into an SSL_CTX against the FIPS-approved bulk cipher NID allowlist
(AES-128-GCM, AES-256-GCM). The check is run unconditionally after
SSL_CTX_set_cipher_list() for both bind and server contexts.
The function reuses the fips_obj_info() helper introduced with the
version check to emit precise 'type proxy/name [file:line]' error
messages and returns ERR_ALERT|ERR_ABORT|ERR_FATAL on violation.
MEDIUM: ssl: introduce src/fips.c with TLS version check
Add src/fips.c and include/haproxy/fips.h to centralise FIPS compliance
checks for AWS-LC builds.
ssl_fips_check_version() verifies that ssl-min-ver is not set below
TLS 1.2, which AWS-LC in FIPS mode would refuse to negotiate.
The check accepts an enum obj_type * so it can resolve the owning
bind/server and produce a precise error message including the proxy
and object name in the standard HAProxy 'type proxy/name' form.
Maxime Henrion [Tue, 30 Jun 2026 15:33:37 +0000 (11:33 -0400)]
MEDIUM: cache: add support for hints-only HTTP caches
The "early-hints" option can now be set to "only", in which case the
cache will only store data relevant to the emission of 103 Early Hints
responses.
Hint records are extracted on the fly from the live response HTX
rather than going through a full-storage-then-strip cycle: the entry
is created as CACHE_EF_STRIPPED directly, with hint records written
past the cache_entry header at storage time. Response bodies are
never stored.
Because entries are stripped from the start, eviction in this mode
never has to strip a full entry; the full LRU stays empty and
cache_make_room() falls through to evicting the oldest stripped entry
when the cache fills up.
The max-object-size check (which gates caching based on the upstream-
announced Content-Length) is bypassed in this mode: the actual stored
entry is just the hint records (typically under 1 KB), so applying it
to the body size would needlessly discard hints for arbitrarily large
responses.
As a hints-only entry is never served as a response, the backend
response is forwarded unchanged in this mode: in particular its Age
header is kept rather than stripped as it is for a stored response.
Maxime Henrion [Tue, 30 Jun 2026 15:32:50 +0000 (11:32 -0400)]
MINOR: cache: factor cache_extract_link_hints out of cache_extract_hints
Move the per-Link-value processing into a new cache_extract_link_hints()
helper. cache_extract_hints() continues to walk the cached row's HDR
blocks and now invokes the helper once per Link header it finds.
No behaviour change. This prepares the helper for reuse by an upcoming
hints-only storage path that extracts Link headers directly from the
live response HTX rather than from a cached row.
Maxime Henrion [Tue, 2 Jun 2026 13:08:23 +0000 (09:08 -0400)]
REGTESTS: cache: validate the emission of 103s
Add four regtests to validate the behavior of the HTTP cache when early
hints are enabled.
- early_hints.vtc: covers 103 emission on stripped entries and that the
"no-early-hints" rule keyword suppresses it.
- early_hints_ratio.vtc: covers cap-driven eviction of hints entries when
the ratio is reached.
- early_hints_parsing.vtc: pins down the (deliberately strict) Link
header parsing behaviour of link_is_hint() so future relaxations
cannot silently change hint detection.
- early_hints_multiblock.vtc: checks that hints are still extracted when
the relevant Link headers span more than one cache block.
Maxime Henrion [Mon, 1 Jun 2026 21:12:21 +0000 (17:12 -0400)]
MINOR: cache: allow customizing ratio for early hints
It is now possible to give a "ratio" parameter to the "early-hints"
option to indicate what percentage of blocks to reserve for hints-only
entries. Defaults to 25%.
Maxime Henrion [Fri, 26 Jun 2026 20:45:23 +0000 (16:45 -0400)]
MINOR: cache: allow opting out of early hints at the rule level
A "cache-use" directive can now specify the "no-early-hints" keyword
to suppress 103 Early Hints emission for matching requests, when used
with a cache that has the "early-hints" setting enabled.
Maxime Henrion [Tue, 7 Jul 2026 13:10:59 +0000 (09:10 -0400)]
MINOR: cache: add a counter for cache hits serving early hints
Add a cache_hint_hits counter to the per-proxy HTTP stats, incremented
each time the cache emits a 103 Early Hints response. It is reported
in the CSV/typed outputs, on the stats page next to the cache hits,
and exported to prometheus as http_cache_hint_hits_total. Since the
new field changes the shm stats file mapping, its minor version is
bumped along with the object size assertion.
Maxime Henrion [Fri, 26 Jun 2026 20:44:50 +0000 (16:44 -0400)]
MEDIUM: cache: emit early hints if configured to do so
The HTTP cache now supports sending 103 Early Hints responses if
enabled, without the need for explicit per-rule configuration.
Two paths feed the 103 response depending on the matched entry. A
stripped entry (CACHE_EF_STRIPPED) carries its Link records
pre-extracted alongside the cache entry header and is copied out of
the row as-is. A complete entry only carries the cached HTX; the
Link headers are walked on the fly at request time, deliberately
avoiding the extra storage cost of duplicating them alongside the
body.
The on-the-fly path has a cost: a workload where many hits land on
expired complete entries re-parses the cached response on each hit.
To bound this, the first such hit also strips the entry in place
when conditions allow (the hints pool is below its cap, and the row
is not currently being read by another request); subsequent hits for
the same URL then take the cheap pre-extracted path.
To let the cache-use lookup reach that logic, get_entry() now returns
expired entries to callers passing delete_expired=0; such callers are
responsible for checking entry->expire themselves.
Maxime Henrion [Fri, 26 Jun 2026 20:43:12 +0000 (16:43 -0400)]
MINOR: http: factor 103 emission into start/end helpers
Pull the 103 emission code out of http_action_early_hint() and split
it into http_early_hint_start() (open the response and return the htx)
and http_early_hint_end() (close with EOH and forward). The split
shape lets callers add their own Link headers in between, which is
what the HTTP cache will need when it gets support for sending 103
Early Hints from cached responses.
Maxime Henrion [Tue, 30 Jun 2026 15:06:12 +0000 (11:06 -0400)]
MEDIUM: cache: early hints-aware eviction code
We now use the make_room callback from shctx to customize the eviction
procedure. We attempt to balance eviction between hints and full-blown
entries by capping the number of blocks used by hints entries to 25% of
the size of the cache. Upon eviction of a full entry, we first try to
strip it to a hints-only entry if possible.
A Link header's value (which may carry multiple comma-separated links)
is limited to 1024 bytes for hint extraction. Longer values are skipped
entirely and contribute no hint records.
Maxime Henrion [Tue, 30 Jun 2026 15:05:19 +0000 (11:05 -0400)]
MINOR: cache: track full and hints entries in per-pool LRU lists
Introduce two LRU lists in struct cache (full_lru and hints_lru) and an
"lru" link in struct cache_entry, plus a CACHE_EF_STRIPPED flag to
mark hints-only entries. A new cache_row_reattach() helper places each
entry in the appropriate LRU when a row is returned to the cold pool.
No behaviour change yet: the lists are populated but no code consumes
them. The next commit uses them to drive early hints-aware eviction.
Maxime Henrion [Fri, 26 Jun 2026 20:40:26 +0000 (16:40 -0400)]
MINOR: shctx: allow consumers to customize eviction strategy
This introduces a make_room() callback that consumers can provide in
order to implement more specific eviction strategies. To be used in the
HTTP cache code. This goes in hand with a new shctx_row_truncate()
function that allows consumers to truncate a row, which can be used in
the make_room() callback.
Maxime Henrion [Sat, 23 May 2026 01:18:47 +0000 (21:18 -0400)]
MINOR: http: add two header parsing functions
Add http_next_hdr_value() to iterate over the comma-separated values
that some headers contain. Add http_get_hdr_param() to iterate over the
';'-separated parameters that may follow a value.
Currently unused but will be used for parsing Link headers in the cache
to support 103 Early Hints, and could also clean up Cache-Control
parsing among others.
projects / thirdparty / haproxy.git / log
