Since curl's own memory debugging system redefines free() calls to track
and fiddle with memory, it cannot be used on memory allocated by 3rd
party libraries.
Third party libraries SHOULD NOT require free() to release allocated
resources for this reason - and libs can use separate healp allocators
on some systems (like Windows) so free() doesn't necessarily work
anyway.
Filed as an issue with libssh: https://bugs.libssh.org/T268Closes#6481
... in most cases instead of 'struct connectdata *' but in some cases in
addition to.
- We mostly operate on transfers and not connections.
- We need the transfer handle to log, store data and more. Everything in
libcurl is driven by a transfer (the CURL * in the public API).
- This work clarifies and separates the transfers from the connections
better.
- We should avoid "conn->data". Since individual connections can be used
by many transfers when multiplexing, making sure that conn->data
points to the current and correct transfer at all times is difficult
and has been notoriously error-prone over the years. The goal is to
ultimately remove the conn->data pointer for this reason.
Closes#6425
... so that a function can first use MIMEPOST and then set it to NULL to
reset it back to a blank POST.
Added test 584 to verify the fix.
Reported-by: Christoph M. Becker
Fixes#6455Closes#6456
... instead of at end of the DO state. This makes the timer more
accurate for the protocols that use the DOING state (such as FTP), and
simplifies how the function (now called init_perform) is called.
The timer will then include the entire procedure up to PERFORM -
including all instructions for getting the transfer started.
Closes#6454
- During the end-of-headers response phase do not mark the tunnel
complete unless the response body was completely parsed/ignored.
Prior to this change if the entirety of a CONNECT response with chunked
encoding was not received by the time the final header was parsed then
the connection would be marked done prematurely, before all the chunked
data could be read in and ignored (since this is what we do with any
CONNECT response body) and the connection could not be used.
Bug: https://curl.se/mail/lib-2021-01/0033.html
Reported-by: Fabian Keil
Closes https://github.com/curl/curl/pull/6432
When doing a request with a request body expecting a 401/407 back, that
initial request is sent with a zero content-length. Test 177 and more.
Closes#6424
... so that Retry-After and other meta-content can still be used.
Added 1634 to verify. Adjusted test 194 and 281 since --fail now also
includes the header-terminating CRLF in the output before it exits.
Fixes#6408Closes#6409
... to make build tools/valgrind warn if no curl_global_cleanup is
called.
This is conditionally only done for debug builds with the env variable
CURL_GLOBAL_INIT set.
Closes#6410
... and not in the connection setup, as for multiplexed transfers the
connection setup might be skipped and then the transfer would end up
without the set user-agent!
Reported-by: Flameborn on github
Assisted-by: Andrey Gursky
Assisted-by: Jay Satiro
Assisted-by: Mike Gelfand
Fixes#6312Closes#6417
The wolfSSL TLS library defines NO_OLD_TLS in some of their build
configurations and that causes the library to be built without TLS 1.1.
For example if MD5 is explicitly disabled when building wolfSSL then
that defines NO_OLD_TLS and the library is built without TLS 1.1 [1].
Prior to this change attempting to build curl with a wolfSSL that was
built with NO_OLD_TLS would cause a build link error undefined reference
to wolfTLSv1_client_method.
[1]: https://github.com/wolfSSL/wolfssl/blob/v4.5.0-stable/configure.ac#L2366
Bug: https://curl.se/mail/lib-2020-12/0121.html
Reported-by: Julian Montes
Closes https://github.com/curl/curl/pull/6388
When doing HTTP authentication and a port number set with CURLOPT_PORT,
the code would previously have the URL's port number override as if it
had been a redirect to an absolute URL.
Added test 1568 to verify.
Reported-by: UrsusArctos on github
Fixes#6397Closes#6400
We currently use both spellings the british "behaviour" and the american
"behavior". However "behavior" is more used in the project so I think
it's worth dropping the british name.
Closes#6395
Extend the syntax of CURLOPT_RESOLVE strings: allow using a '+' prefix
(similar to the existing '-' prefix for removing entries) to add
DNS cache entries that will time out just like entries that are added
by libcurl itself.
Append " (non-permanent)" to info log message in case a non-permanent
entry is added.
Adjust relevant comments to reflect the new behavior.
Adjust documentation.
Extend unit1607 to test the new functionality.
Closes#6294
Paused transfers should not be stopped due to slow speed even when
CURLOPT_LOW_SPEED_LIMIT is set. Additionally, the slow speed timer is
now reset when the transfer is unpaused - as otherwise it would easily
just trigger immediately after unpausing.
Reported-by: Harry Sintonen
Fixes#6358Closes#6359
... as the socket might be readable all the time when paused and thus
causing a busy-loop.
Reported-by: Harry Sintonen
Reviewed-by: Jay Satiro
Fixes#6356Closes#6357
It is a security process for HTTP.
It doesn't seems to be standard, but it is used by some cloud providers.
Aws:
https://docs.aws.amazon.com/general/latest/gr/signature-version-4.html
Outscale:
https://wiki.outscale.net/display/EN/Creating+a+Canonical+Request
GCP (I didn't test that this code work with GCP though):
https://cloud.google.com/storage/docs/access-control/signing-urls-manually
most of the code is in lib/http_v4_signature.c
Information require by the algorithm:
- The URL
- Current time
- some prefix that are append to some of the signature parameters.
The data extracted from the URL are: the URI, the region,
the host and the API type
example:
https://api.eu-west-2.outscale.com/api/latest/ReadNets
~~~ ~~~~~~~~ ~~~~~~~~~~~~~~~~~~~
^ ^ ^
/ \ URI
API type region
Small description of the algorithm:
- make canonical header using content type, the host, and the date
- hash the post data
- make canonical_request using custom request, the URI,
the get data, the canonical header, the signed header
and post data hash
- hash canonical_request
- make str_to_sign using one of the prefix pass in parameter,
the date, the credential scope and the canonical_request hash
- compute hmac from date, using secret key as key.
- compute hmac from region, using above hmac as key
- compute hmac from api_type, using above hmac as key
- compute hmac from request_type, using above hmac as key
- compute hmac from str_to_sign using above hmac as key
- create Authorization header using above hmac, prefix pass in parameter,
the date, and above hash
Signed-off-by: Matthias Gatto <matthias.gatto@outscale.com>
Closes#5703
It seems current hmac implementation use md5 for the hash,
V4 signature require sha256, so I've added the needed struct in
this commit.
I've added the functions that do the hmac in v4 signature file
as a static function ,in the next patch of the serie,
because it's used only by this file.
Signed-off-by: Matthias Gatto <matthias.gatto@outscale.com>
The linux kernel does not report all ICMP errors back to userspace due
to historical reasons.
IP*_RECVERR sockopt must be turned on to have the correct behaviour
which is to pass all ICMP errors to userspace.
See https://bugzilla.kernel.org/show_bug.cgi?id=202355Closes#6341
When the initial request isn't possible to send in its entirety, the
remainder of request would be delivered to the debug callback as data
and would wrongly be counted internally as body-bytes sent.
Extended test 1295 to verify.
Closes#6328
When failing in TOOFAST, the multi_done() wasn't called so the same
cleanup and handling wasn't done like when it fails in PERFORM, which in
the case of FTP could mean that the control connection wouldn't be
marked as "dead" for the CURLE_ABORTED_BY_CALLBACK case. Which caused
ftp_disconnect() to use it to send "QUIT", which could end up waiting
for a response a long time before giving up!
Reported-by: Tomas Berger
Fixes#6333Closes#6337
This commit introduces a "gophers" handler inside the gopher protocol if
USE_SSL is defined. This protocol is no different than the usual gopher
prococol, with the added TLS encapsulation upon connecting. The protocol
has been adopted in the gopher community, and many people have enabled
TLS in their gopher daemons like geomyidae(8), and clients, like clic(1)
and hurl(1).
I have not implemented test units for this protocol because my knowledge
of Perl is sub-par. However, for someone more knowledgeable it might be
fairly trivial, because the same test that tests the plain gopher
protocol can be used for "gophers" just by adding a TLS listener.
Signed-off-by: parazyd <parazyd@dyne.org>
Closes#6208
The error is shown with infof rather than failf so that the user will
see the extended error message information only in verbose mode, and
will still see the standard CURLE_AUTH_ERROR message. For example:
---
* schannel: InitializeSecurityContext failed: SEC_E_QOP_NOT_SUPPORTED
(0x8009030A) - The per-message Quality of Protection is not supported by
the security package
* multi_done
* Connection #1 to host 127.0.0.1 left intact
curl: (94) An authentication function returned an error
---
Ref: https://github.com/curl/curl/issues/6302
Closes https://github.com/curl/curl/pull/6315
If supported, defer port selection until connect() time
if --interface is given and source port is 0.
Reproducer:
* start fast webserver on port 80
* starve system of ephemeral ports
$ sysctl net.ipv4.ip_local_port_range="60990 60999"
* start a curl/libcurl "crawler"
$curl --keepalive --parallel --parallel-immediate --head --interface
127.0.0.2 "http://127.0.0.[1-254]/file[001-002].txt"
current result:
(possible some successful data)
curl: (45) bind failed with errno 98: Address already in use
result after patch:
(complete success or few connections failing, higlhy depending on load)
Fail only when all the possible 4-tuple combinations are exhausted,
which is impossible to do when port is selected at bind() time becuse
the kernel does not know if socket will be listen()'ed on or connect'ed
yet.
Closes#6295
Valgrind will complain that ssrem buffer usage if not explicit
initialized, hence initialize it to zero.
This completes the change intially started in commit 2c0d721215 ('ftp:
retry getpeername for FTP with TCP_FASTOPEN') where the ssloc buffer has
a similar memset to zero.
Signed-off-by: Hans-Christian Noren Egtvedt <hegtvedt@cisco.com>
Closes#6289
The command line tool also independently sets --ftp-skip-pasv-ip by
default.
Ten test cases updated to adapt the modified --libcurl output.
Bug: https://curl.se/docs/CVE-2020-8284.html
CVE-2020-8284
Reported-by: Varnavas Papaioannou
... as it makes the URL parser accept "very-long-hostname://" as a valid
host name and we don't want that. The parser now only accepts a blank
(no digits) after the colon if the URL starts with a scheme.
Reported-by: d4d on hackerone
Closes#6283
This reverts commit d2a7d7c185.
This commit also reverts the subsequent follow-ups to that commit, which
were all done within windows #ifdefs that are removed in this
change. Marc helped me verify this.
Fixes#6146Closes#6281
In the case of TFO, the remote host name is not resolved at the
connetion time.
For FTP that has lead to missing hostname for the secondary connection.
Therefore the name resolution is done at the time, when FTP requires it.
Fixes#6252Closes#6265Closes#6282
This field needs to be wide enough to hold sockaddr_in6 when
connecting via IPv6. Otherwise, ngtcp2_conn_read_pkt will drop the
packets because of the address mismatch:
I00000022 [...] con ignore packet from unknown path
We can safely assume that struct sockaddr_storage is available, as it
is used in the public interface of ngtcp2.
Closes#6250
The resolve call is done with the right port number, but the subsequent
check used the wrong one, which then could find a previous resolve which
would return and leave the fresh resolve "incomplete" and leaking
memory.
Fixes#6247Closes#6253
