Connection filter had a `get_select_socks()` method, inspired by the
various `getsocks` functions involved during the lifetime of a
transfer. These, depending on transfer state (CONNECT/DO/DONE/ etc.),
return sockets to monitor and flag if this shall be done for POLLIN
and/or POLLOUT.
Due to this design, sockets and flags could only be added, not
removed. This led to problems in filters like HTTP/2 where flow control
prohibits the sending of data until the peer increases the flow
window. The general transfer loop wants to write, adds POLLOUT, the
socket is writeable but no data can be written.
This leads to cpu busy loops. To prevent that, HTTP/2 did set the
`SEND_HOLD` flag of such a blocked transfer, so the transfer loop cedes
further attempts. This works if only one such filter is involved. If a
HTTP/2 transfer goes through a HTTP/2 proxy, two filters are
setting/clearing this flag and may step on each other's toes.
Connection filters `get_select_socks()` is replaced by
`adjust_pollset()`. They get passed a `struct easy_pollset` that keeps
up to `MAX_SOCKSPEREASYHANDLE` sockets and their `POLLIN|POLLOUT`
flags. This struct is initialized in `multi_getsock()` by calling the
various `getsocks()` implementations based on transfer state, as before.
After protocol handlers/transfer loop have set the sockets and flags
they want, the `easy_pollset` is *always* passed to the filters. Filters
"higher" in the chain are called first, starting at the first
not-yet-connection one. Each filter may add sockets and/or change
flags. When all flags are removed, the socket itself is removed from the
pollset.
Example:
* transfer wants to send, adds POLLOUT
* http/2 filter has a flow control block, removes POLLOUT and adds
POLLIN (it is waiting on a WINDOW_UPDATE from the server)
* TLS filter is connected and changes nothing
* h2-proxy filter also has a flow control block on its tunnel stream,
removes POLLOUT and adds POLLIN also.
* socket filter is connected and changes nothing
* The resulting pollset is then mixed together with all other transfers
and their pollsets, just as before.
Use of `SEND_HOLD` is no longer necessary in the filters.
All filters are adapted for the changed method. The handling in
`multi.c` has been adjusted, but its state handling the the protocol
handlers' `getsocks` method are untouched.
The most affected filters are http/2, ngtcp2, quiche and h2-proxy. TLS
filters needed to be adjusted for the connecting handshake read/write
handling.
No noticeable difference in performance was detected in local scorecard
runs.
Closes#11833
Currently the verbose output does not include which algorithms are used
for the signature and key exchange when using OpenSSL. Including the
algorithms used will enable better debugging when working on using new
algorithm implementations. Know what algorithms are used has become more
important with the fast growing research into new quantum-safe
algorithms.
This implementation includes a build time check for the OpenSSL version
to use a new function that will be included in OpenSSL 3.2 that was
introduced in openssl/openssl@6866824
Based-on-patch-by: Martin Schmatz <mrt@zurich.ibm.com>
Closes#12030
Remove the CURL_CA_FALLBACK logic. That build option was added to allow
primarily OpenSSL to use the default paths for loading the CA certs. For
GnuTLS it was instead made to load the "system certs", which is
different and not desirable.
The native CA store loading is now asked for with this option.
Follow-up to 7b55279d1d
Co-authored-by: Jay Satiro
Closes#12137
This define is set in wolfssl's options.h file when this function and
feature is present. Handles both builds with the feature explicitly
disabled and wolfSSL versions before 5.5.2 - which introduced this API
call.
Closes#12108
Syncing this up with CMake.
Source code uses the built-in `OPENSSL_IS_AWSLC` and
`OPENSSL_IS_BORINSSL` macros to detect BoringSSL and AWS-LC. No help is
necessary from the build tools.
The one use of `HAVE_BORINGSSL` in the source turned out to be no longer
necessary for warning-free BoringSSL + Schannel builds. Ref: #1610#2634
autotools detects this anyway for display purposes.
CMake detects this to decide whether to use the BoringSSL-specific
crypto lib with ngtcp2. It detects AWS-LC, but doesn't use the detection
result just yet (planned in #12066).
Ref: #11964
Reviewed-by: Daniel Stenberg
Reviewed-by: Jay Satiro
Closes#12065
While the struct is still public in OpenSSL, there is a (somewhat
inconvenient) accessor. Use it to remain compatible if it becomes opaque
in the future.
Closes#12038
By using unique static function/variable names in source files
implementing these interfaces.
- OpenLDAP combined with any SSH backend.
- MultiSSL with mbedTLS, OpenSSL, wolfSSL, SecureTransport.
Closes#12027
The default wolfSSL_CTX_load_verify_locations() function is quite picky
with the certificates it loads and will for example return error if just
one of the certs has expired.
With the *_ex() function and its WOLFSSL_LOAD_FLAG_IGNORE_ERR flag, it
behaves more similar to what OpenSSL does by default.
Even the set of default certs on my Debian unstable has several expired
ones.
Assisted-by: Juliusz Sosinowicz
Assisted-by: Michael Osipov
Closes#11987
With new option `CURL_DISABLE_SRP=ON` to force-disable it.
To match existing option and detection logic in autotools.
Also:
- fix detecting GnuTLS.
We assume `nettle` as a GnuTLS dependency.
- add CMake GnuTLS CI job.
- bump AppVeyor CMake OpenSSL MSVC job to OpenSSL 1.1.1 (from 1.0.2)
TLS-SRP fails to detect with 1.0.2 due to an OpenSSL header bug.
- fix compiler warning when building with GnuTLS and disabled TLS-SRP.
- fix comment typos, whitespace.
Ref: #11964Closes#11967
- If SSL shutdown is not finished then make an additional call to
SSL_read to gather additional tracing.
- Fix http2 and h2-proxy filters to forward do_close() calls to the next
filter.
For example h2 and SSL shutdown before and after this change:
Before:
Curl_conn_close -> cf_hc_close -> Curl_conn_cf_discard_chain ->
ssl_cf_destroy
After:
Curl_conn_close -> cf_hc_close -> cf_h2_close -> cf_setup_close ->
ssl_cf_close
Note that currently the tracing does not show output on the connection
closure handle. Refer to discussion in #11878.
Ref: https://github.com/curl/curl/discussions/11878
Closes https://github.com/curl/curl/pull/11858
- Use CERT_CONTEXT's pbCertEncoded to determine chain order.
CERT_CONTEXT from SECPKG_ATTR_REMOTE_CERT_CONTEXT contains
end-entity/server certificate in pbCertEncoded. We can use this pointer
to determine the order of certificates when enumerating hCertStore using
CertEnumCertificatesInStore.
This change is to help ensure that the ordering of the certificate chain
requested by the user via CURLINFO_CERTINFO has the same ordering on all
versions of Windows.
Prior to this change Schannel certificate order was reversed in 8986df80
but that was later reverted in f540a39b when it was discovered that
Windows 11 22H2 does the reversal on its own.
Ref: https://github.com/curl/curl/issues/9706
Closes https://github.com/curl/curl/pull/11632
- we delay loading the x509 store to shorten the handshake time.
However an application callback installed via CURLOPT_SSL_CTX_FUNCTION
may need to have the store loaded and try to manipulate it.
- load the x509 store before invoking the app callback
Fixes#11800
Reported-by: guoxinvmware on github
Cloes #11805
- added test cases for various code paths
- fixed handling of blocked write when stream had
been closed inbetween attempts
- re-enabled DEBUGASSERT on send with smaller data size
- in debug builds, environment variables can be set to simulate a slow
network when sending data. cf-socket.c and vquic.c support
* CURL_DBG_SOCK_WBLOCK: percentage of send() calls that should be
answered with a EAGAIN. TCP/UNIX sockets.
This is chosen randomly.
* CURL_DBG_SOCK_WPARTIAL: percentage of data that shall be written
to the network. TCP/UNIX sockets.
Example: 80 means a send with 1000 bytes would only send 800
This is applied to every send.
* CURL_DBG_QUIC_WBLOCK: percentage of send() calls that should be
answered with EAGAIN. QUIC only.
This is chosen randomly.
Closes#11756
We've seen errors left in the OpenSSL error queue (specifically,
"shutdown while in init") by adding some logging it revealed that the
source was this file.
Since we call SSL_read and SSL_shutdown here, but don't check the return
code for an error, we should clear the OpenSSL error queue in case one
was raised.
This didn't affect curl because we call ERR_clear_error before every
write operation (a0dd9df9ab), but when
libcurl is used in a process with other OpenSSL users, they may detect
an OpenSSL error pushed by libcurl's SSL_shutdown as if it was their
own.
Co-authored-by: Satana de Sant'Ana <satana@skylittlesystem.org>
Closes#11736
- bring bearssl handshake times down from +200ms down to other TLS backends
- vtls: improve generic get_select_socks() implementation
- tests: provide Apache with a suitable ssl session cache
Closes#11675
OpenSSL 1.1.1 defines this macro, but no ealier version, or any of the
popular forks (yet). Use the macro itself to detect its presence,
replacing the hard-wired fork-specific conditions.
This way the feature will enable automatically when forks implement it,
while also shorter and possibly requiring less future maintenance.
Follow-up to 94241a9e78#6721
Reviewed-by: Jay Satiro
Closes#11617
Make sure that context initialization during hash setup works to avoid
going forward with the risk of a null pointer dereference.
Reported-by: Philippe Antoine on HackerOne
Assisted-by: Jay Satiro
Assisted-by: Daniel Stenberg
Closes#11614
LibreSSL 2.7.0 (2018-03-21) introduced automatic initialization,
`OPENSSL_init_ssl()` function and deprecated the old, manual init
method, as seen in OpenSSL 1.1.0. Switch to the modern method when
available.
Ref: https://ftp.openbsd.org/pub/OpenBSD/LibreSSL/libressl-2.7.0-relnotes.txt
Reviewed-by: Daniel Stenberg
Closes#11611
We remove support for building curl with gskit.
- This is a niche TLS library, only running on some IBM systems
- no regular curl contributors use this backend
- no CI builds use or verify this backend
- gskit, or the curl adaption for it, lacks many modern TLS features
making it an inferior solution
- build breakages in this code take weeks or more to get detected
- fixing gskit code is mostly done "flying blind"
This removal has been advertized in DEPRECATED in Jan 2, 2023 and it has
been mentioned on the curl-library mailing list.
It could be brought back, this is not a ban. Given proper effort and
will, gskit support is welcome back into the curl TLS backend family.
Closes#11460
- If the user set a legacy algorithm list (CURLOPT_SSL_CIPHER_LIST) then
use the SCHANNEL_CRED legacy structure to pass the list to Schannel.
- If the user set both a legacy algorithm list and a TLS 1.3 cipher list
then abort.
Although MS doesn't document it, Schannel will not negotiate TLS 1.3
when SCHANNEL_CRED is used. That means setting a legacy algorithm list
limits the user to earlier versions of TLS.
Prior to this change, since 8beff435 (precedes 7.85.0), libcurl would
ignore legacy algorithms in Windows 10 1809 and later.
Reported-by: zhihaoy@users.noreply.github.com
Fixes https://github.com/curl/curl/pull/10741
Closes https://github.com/curl/curl/pull/10746
EGD is Entropy Gathering Daemon, a socket-based entropy source supported
by pre-OpenSSL v1.1 versions and now deprecated. curl also deprecated it
a while ago.
Its detection in CMake was broken all along because OpenSSL libs were
not linked at the point of feature check.
Delete detection from both cmake and autotools, along with the related
source snippet, and the `--with-egd-socket=` `./configure` option.
Closes#11556
- Update the OpenSSL connect state machine to handle
SSL_ERROR_WANT_RETRY_VERIFY.
This allows libcurl users that are using custom certificate validation
to suspend processing while waiting for external I/O during certificate
validation.
Closes https://github.com/curl/curl/pull/11499
When SecCertificateCopyCommonName() returns NULL, the common_name
pointer remains set to NULL which apparently when calling CFRelease() on
(sometimes?) crashes.
Reported-by: Guillaume Algis
Fixes#9194Closes#11554
This reverts commit 8986df802d.
Windows does not guarantee a particular certificate ordering, even
though TLS may have its own ordering/relationship guarantees. Recent
versions of Windows 11 reversed the ordering of ceritifcates returned by
CertEnumCertificatesInStore, therefore this commit no longer works as
initially intended. libcurl makes no guarantees about certificate
ordering if the operating system can't.
Ref: https://github.com/curl/curl/issues/9706
Closes https://github.com/curl/curl/pull/11536
Rename `close` and `connect` in `struct Curl_cftype` for
consistency and to avoid clashes with macros of the same name
(the standard AmigaOS networking connect() function is implemented
via a macro).
Closes#11491
- fix HTTP/2 check to not declare a connection dead when
the read attempt results in EAGAIN
- add H2-PROXY alive check as for HTTP/2 that was missing
and is needed
- add attach/detach around Curl_conn_is_alive() and remove
these in filter methods
- add checks for number of connections used in some test_10
proxy tunneling tests
Closes#11368
