time_t representation is undefined and varies on bitsize and signedness,
and as of C11 could be even non integer.
instead of casting to unsigned long (which would truncate in systems
with a 32bit long after 2106) use difftime to get the elapsed time as a
double and print that (without decimals) instead.
alternatively a cast to curl_off_t and its corresponding print
formatting could have been used (at least in POSIX) but portability and
curl agnostic code was prioritized.
Daniel Stenberg [Fri, 23 Jul 2021 21:52:12 +0000 (23:52 +0200)]
http_proxy: clear 'sending' when the outgoing request is sent
... so that Curl_connect_getsock() will know how to wait for the socket
to become readable and not writable after the entire CONNECT request has
been issued.
Regression added in 7.77.0
Reported-by: zloi-user on github Assisted-by: Jay Satiro
Fixes #7155
Closes #7484
Jay Satiro [Fri, 23 Jul 2021 06:56:52 +0000 (02:56 -0400)]
docs: change max-filesize caveat again
- Add protocols field to max-filesize.d.
- Revert wording on unknown file size caveat and do not discuss specific
protocols in that section.
Partial revert of ecf0225. All max-filesize options now have the list of
protocols and it's clearer just to have that list without discussing
specific protocols in the caveat.
commit 29c7cf79e8b44cf (shipped in 7.78.0) introduced a problem by
assuming that LIB_H2 does not have any leading whitespace. At least
OpenBSD's native pkg-config can produce such whitespace, though:
$ pkg-config --libs-only-l libnghttp2
-lnghttp2
As a result, the configure check for libnghttp2 will erroneously fail.
Jay Satiro [Mon, 19 Jul 2021 23:06:30 +0000 (19:06 -0400)]
connect: fix wrong format specifier in connect error string
0842175 (not in any release) used the wrong format specifier (long int)
for timediff_t. On an OS such as Windows libcurl's timediff_t (usually
64-bit) is bigger than long int (32-bit). In 32-bit Windows builds the
upper 32-bits of the timediff_t were erroneously then used by the next
format specifier. Usually since the timeout isn't larger than 32-bits
this would result in null as a pointer to the string with the reason for
the connection failing. On other OSes or maybe other compilers it could
probably result in garbage values (ie crash on deref).
Before:
Failed to connect to localhost port 12345 after 1201 ms: (nil)
After:
Failed to connect to localhost port 12345 after 1203 ms: Connection refused
docs: spell out directories instead of dirs in create-dirs
Write out directories rather than using the dirs abbrevation. Also
use plural form consistently, even if the code in the end might just
create a single directory.
Closes #7406 Reviewed-by: Daniel Stenberg <daniel@haxx.se> Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Max Zettlmeißl [Sun, 18 Jul 2021 08:42:54 +0000 (10:42 +0200)]
docs: make the documentation for --etag-save match the program behaviour
When using curl with the option `--etag-save` I expected it to save the
ETag without its surrounding quotes, as stated by the documentation in
the repository and by the generated man pages.
My first endeavour was to fix the program, but while investigating the
history of the relevant parts, I discovered that curl once saved the
ETag without the quotes. This was undone by Daniel Stenberg in commit
`98c94596f5928840177b6bd3c7b0f0dd03a431af`, therefore I decided that in
this case the documentation should be adjusted to match the behaviour of
curl.
The changed save behaviour also made parts of the `--etag-compare`
documentation wrong or superfluous, so I adjusted those accordingly.
easy: during upkeep, attach Curl_easy to connections in the cache
During the protocol-specific parts of connection upkeep, some code
assumes that the data->conn pointer already is set correctly. However,
there's currently no guarantee of that in the code.
This fix temporarily attaches each connection to the Curl_easy object
before performing the protocol-specific connection check on it, in a
similar manner to the connection checking in extract_if_dead().
- curl_setup.h: all references to mbedtls_md4* functions and structures
are in the md4.c. This file already includes the <mbedtls/md4.h> file
along with the file existence control (defined (MBEDTLS_MD4_C))
- curl_ntlm_core.c: unnecessary include - repeated below
Wyatt OʼDay reported in #7385 that mbedTLS isn't backwards compatible
and curl no longer builds with it. Document the need to fix our support
until so has been done.
docs: fix inconsistencies in EGDSOCKET documentation
Only the OpenSSL backend actually use the EGDSOCKET, and also use
TLS consistently rather than mixing SSL and TLS. While there, also
fix a minor spelling nit.
Closes: #7391 Reviewed-by: Jay Satiro <raysatiro@yahoo.com>
Revert "ftp: Expression 'ftpc->wait_data_conn' is always false"
The reverted commit introduced a logic error in code that was
correct.
The client using libcurl would notice the error since FTP file
uploads in active transfer mode would somtimes complete with
success despite no transfer having been performed and the
"uploaded" file thus not being on the remote server afterwards.
The FTP server would notice the error because it receives a
RST on the data connection it has established with the client
before any data was transferred at all.
The logic error happens if the STOR response from the server have
arrived by the time ftp_multi_statemach() in the affected code path
is called, but the incoming data connection have not arrived yet.
In that case, the processing of the STOR response will cause
'ftpc->wait_data_conn' to be set to TRUE, contradicting the comment
in the code. Since 'complete' will also be set, later logic would
believe the transfer was done.
In most cases, the STOR response will not have arrived yet when
the affected code path is executed, or the incoming connection will
also have arrived, and thus the error would not express itself.
But if the speed difference of the device using libcurl and the
FTP server is exactly right, the error may happen as often as in
one out of hundred file transfers.
Daniel Stenberg [Mon, 5 Jul 2021 11:28:26 +0000 (13:28 +0200)]
socks4: scan for the IPv4 address in resolve results
Follow-up to 84d2839740 which changed the resolving to always resolve
both address families, but since SOCKS4 only supports IPv4 it should
scan for and use the first available IPv4 address.
Reported-by: shithappens2016 on github
Fixes #7345
Closes #7346
It should not assume that it can copy full lines into the buffer as it
will encourage sloppy coding practices. Instead use byte-wise logic and
check/acknowledge the buffer size appropriately.
Reported-by: Harry Sintonen
Fixes #7330
Closes #7331
Daniel Stenberg [Wed, 30 Jun 2021 09:17:57 +0000 (11:17 +0200)]
url: (void)-prefix a curl_url_get() call
Coverity (CID 1486645) pointed out a use of curl_url_get() in the
parse_proxy function where the return code wasn't checked. A
(void)-prefix makes the intention obvious.
Jay Satiro [Tue, 29 Jun 2021 15:43:35 +0000 (11:43 -0400)]
http: fix crash in rate-limited upload
- Don't set the size of the piece of data to send to the rate limit if
that limit is larger than the buffer size that will hold the piece.
Prior to this change if CURLOPT_MAX_SEND_SPEED_LARGE
(curl tool: --limit-rate) was set then it was possible that a temporary
buffer used for uploading could be written to out of bounds. A likely
scenario for this would be a non-trivial amount of post data combined
with a rate limit larger than CURLOPT_UPLOAD_BUFFERSIZE (default 64k).
The bug was introduced in 24e469f which is in releases since 7.76.0.
projects / thirdparty / curl.git / log
