We use "conn" everywhere to be a pointer to the connection.
Introduces two functions that "attaches" and "detaches" the connection
to and from the transfer.
Going forward, we should favour using "data->conn" (since a transfer
always only has a single connection or none at all) to "conn->data"
(since a connection can have none, one or many transfers associated with
it and updating conn->data to be correct is error prone and a frequent
reason for internal issues).
Closes#3442
Fixes#3436Closes#3448
Problem 1
After LOTS of scratching my head, I eventually realized that even when doing
10 uploads in parallel, sometimes the socket callback to the application that
tells it what to wait for on the socket, looked like it would reflect the
status of just the single transfer that just changed state.
Digging into the code revealed that this was indeed the truth. When multiple
transfers are using the same connection, the application did not correctly get
the *combined* flags for all transfers which then could make it switch to READ
(only) when in fact most transfers wanted to get told when the socket was
WRITEABLE.
Problem 1b
A separate but related regression had also been introduced by me when I
cleared connection/transfer association better a while ago, as now the logic
couldn't find the connection and see if that was marked as used by more
transfers and then it would also prematurely remove the socket from the socket
hash table even in times other transfers were still using it!
Fix 1
Make sure that each socket stored in the socket hash has a "combined" action
field of what to ask the application to wait for, that is potentially the ORed
action of multiple parallel transfers. And remove that socket hash entry only
if there are no transfers left using it.
Problem 2
The socket hash entry stored an association to a single transfer using that
socket - and when curl_multi_socket_action() was called to tell libcurl about
activities on that specific socket only that transfer was "handled".
This was WRONG, as a single socket/connection can be used by numerous parallel
transfers and not necessarily a single one.
Fix 2
We now store a list of handles in the socket hashtable entry and when libcurl
is told there's traffic for a particular socket, it now iterates over all
known transfers using that single socket.
Added Curl_resolver_kill() for all three resolver modes, which only
blocks when necessary, along with test 1592 to confirm
curl_multi_remove_handle() doesn't block unless it must.
Closes#3428Fixes#3371
Do not assume/store assocation between a given easy handle and the
connection if it can be avoided.
Long-term, the 'conn->data' pointer should probably be removed as it is a
little too error-prone. Still used very widely though.
Reported-by: masbug on github
Fixes#3391Closes#3400
The time_t type is unsigned on some systems and these variables are used
to hold return values from functions that return timediff_t
already. timediff_t is always a signed type.
Closes#3363
This is a companion patch to cbea2fd2c (NTLM: force the connection to
HTTP/1.1, 2018-12-06): with NTLM, we can switch to HTTP/1.1
preemptively. However, with other (Negotiate) authentication it is not
clear to this developer whether there is a way to make it work with
HTTP/2, so let's try HTTP/2 first and fall back in case we encounter the
error HTTP_1_1_REQUIRED.
Note: we will still keep the NTLM workaround, as it avoids an extra
round trip.
Daniel Stenberg helped a lot with this patch, in particular by
suggesting to introduce the Curl_h2_http_1_1_error() function.
Closes#3349
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
curl_multi_wait() was erroneously used from within
curl_easy_perform(). It could lead to it believing there was no socket
to wait for and then instead sleep for a while instead of monitoring the
socket and then miss acting on that activity as swiftly as it should
(causing an up to 1000 ms delay).
Reported-by: Antoni Villalonga
Fixes#3305Closes#3306Closes#3308
The function does not return the same value as snprintf() normally does,
so readers may be mislead into thinking the code works differently than
it actually does. A different function name makes this easier to detect.
Reported-by: Tomas Hoger
Assisted-by: Daniel Gustafsson
Fixes#3296Closes#3297
When using c-ares for asyn dns, the dns socket fd was silently closed
by c-ares without curl being aware. curl would then 'realize' the fd
has been removed at next call of Curl_resolver_getsock, and only then
notify the CURLMOPT_SOCKETFUNCTION to remove fd from its poll set with
CURL_POLL_REMOVE. At this point the fd is already closed.
By using ares socket state callback (ARES_OPT_SOCK_STATE_CB), this
patch allows curl to be notified that the fd is not longer needed
for neither for write nor read. At this point by calling
Curl_multi_closed we are able to notify multi with CURL_POLL_REMOVE
before the fd is actually closed by ares.
In asyn-ares.c Curl_resolver_duphandle we can't use ares_dup anymore
since it does not allow passing a different sock_state_cb_data
Closes#3238
Curl_follow() no longer frees the string. Make sure it happens in the
caller function, like we normally handle allocations.
This bug was introduced with the use of the URL API internally, it has
never been in a release version
Reported-by: Dario Weißer
Closes#3149
Transparently. The related curl_multi_setopt() options all still returns
OK when pipelining is selected.
To re-enable the support, the single line change in lib/multi.c needs to
be reverted.
See docs/DEPRECATE.md
Closes#2705
It was previously erroneously skipped in some situations.
libtest/libntlmconnect.c wrongly depended on wrong behavior (that it
would get a zero timeout) when no handles are "running" in a multi
handle. That behavior is no longer present with this fix. Now libcurl
will always return a -1 timeout when all handles are completed.
Closes#2733
When the application just started the transfer and then stops it while
the name resolve in the background thread hasn't completed, we need to
wait for the resolve to complete and then cleanup data accordingly.
Enabled test 1553 again and added test 1590 to also check when the host
name resolves successfully.
Detected by OSS-fuzz.
Closes#1968
- Get rid of variable that was generating false positive warning
(unitialized)
- Fix issues in tests
- Reduce scope of several variables all over
etc
Closes#2631
... it might call infof() with a NULL first argument that isn't harmful
but makes it not do anything. The infof() line is not very useful
anymore, it has served it purpose. Good riddance!
Fixes#2627
The latest psl is cached in the multi or share handle. It is refreshed
before use after 72 hours.
New share lock CURL_LOCK_DATA_PSL controls the psl cache sharing.
If the latest psl is not available, the builtin psl is used.
Reported-by: Yaakov Selkowitz
Fixes#2553Closes#2601
This extends the INDENTATION case to also handle 'else' statements
and require proper indentation on the following line. Also fixes the
offending cases found in the codebase.
Closes#2532
When receiving REFUSED_STREAM, mark the connection for close and retry
streams accordingly on another/fresh connection.
Reported-by: Terry Wu
Fixes#2416Fixes#1618Closes#2510
When a transfer is requested to get done and it is put in the pending
queue when limited by number of connections, total or per-host, libcurl
would previously very aggressively retry *ALL* pending transfers to get
them transferring. That was very time consuming.
By reducing the aggressiveness in how pending are being retried, we
waste MUCH less time on putting transfers back into pending again.
Some test cases got a factor 30(!) speed improvement with this change.
Reported-by: Cyril B
Fixes#2369Closes#2383
Especially unpausing a transfer might have to move the socket back to the
"currently used sockets" hash to get monitored. Otherwise it would never get
any more data and get stuck. Easily triggered with pausing using the
multi_socket API.
Reported-by: Philip Prindeville
Bug: https://curl.haxx.se/mail/lib-2018-03/0048.htmlFixes#2393Closes#2391
Due to very frequent updates of the rate limit "window", it could
attempt to rate limit within the same milliseconds and that then made
the calculations wrong, leading to it not behaving correctly on very
fast transfers.
This new logic updates the rate limit "window" to be no shorter than the
last three seconds and only updating the timestamps for this when
switching between the states TOOFAST/PERFORM.
Reported-by: 刘佩东
Fixes#2386Closes#2388
Prune the DNS cache immediately after the dns entry is unlocked in
multi_done. Timed out entries will then get discarded in a more orderly
fashion.
Test506 is updated
Reported-by: Oleg Pudeyev
Fixes#2169Closes#2170
If the lock is released before the dealings with the bundle is over, it may
have changed by another thread in the mean time.
Fixes#2132Fixes#2151Closes#2139
returning 'time_t' is problematic when that type is unsigned and we
return values less than zero to signal "already expired", used in
several places in the code.
Closes#2021
