Previously, when a stream was closed with other than NGHTTP2_NO_ERROR
by RST_STREAM, underlying TCP connection was dropped. This is
undesirable since there may be other streams multiplexed and they are
very much fine. This change introduce new error code
CURLE_HTTP2_STREAM, which indicates stream error that only affects the
relevant stream, and connection should be kept open. The existing
CURLE_HTTP2 means connection error in general.
Ref: https://github.com/curl/curl/issues/659
Ref: https://github.com/curl/curl/pull/663
Simplify the code by using a single entry that looks for a socket in the
socket hash. As indicated in #712, the code looked for CURL_SOCKET_BAD
at some point and that is ineffective/wrong and this makes it easier to
avoid that.
Such a return value isn't documented but could still happen, and the
curl tool code checks for it. It would happen when the underlying
Curl_poll() function returns an error. Starting now we mask that error
as a user of curl_multi_wait() would have no way to handle it anyway.
Reported-by: Jay Satiro
Closes#707
The internal Curl_done() function uses Curl_expire() at times and that
uses the timeout list. Better clean up the list once we're done using
it. This caused a segfault.
Reported-by: 蔡文凱
Bug: https://curl.haxx.se/mail/lib-2016-02/0097.html
introduced in c6aedf680f. It needs to be CURLM_STATE_LAST big since it
must hande the range 0 .. CURLM_STATE_MSGSENT (18) and CURLM_STATE_LAST
is 19 right now.
Reported-by: Dan Fandrich
Bug: http://curl.haxx.se/mail/lib-2015-10/0069.html
... and assign it from the set.fread_func_set pointer in the
Curl_init_CONNECT function. This A) avoids that we have code that
assigns fields in the 'set' struct (which we always knew was bad) and
more importantly B) it makes it impossibly to accidentally leave the
wrong value for when the handle is re-used etc.
Introducing a state-init functionality in multi.c, so that we can set a
specific function to get called when we enter a state. The
Curl_init_CONNECT is thus called when switching to the CONNECT state.
Bug: https://github.com/bagder/curl/issues/346Closes#346
With many easy handles using the same connection for multiplexing, it is
important we store and keep the transfer-oriented stuff in the
SessionHandle so that callbacks and callback data work fine even when
many easy handles share the same physical connection.
.. also make __func__ replacement in multi.
Prior to this change debug builds would fail to build if the compiler
was building pre-c99 and didn't support __func__.
to allow code to act differently on the situation.
Also added some more info message for the connection re-use function to
make it clearer when connections are not re-used.
All the existing Curl_bundle* functions were only ever used from within
the conncache.c file, so I moved them over and made them static (and
removed the Curl_ prefix).
This avoids unnecessary dynamic allocs and as this also removed the last
users of *hash_alloc() and *hash_destroy(), those two functions are now
removed.
If the handle removed from the multi handle happens to be the one
"owning" the pipeline other transfers will be waiting indefinitely. Now
we move such handles back to connect to have them race (again) for
getting the connection and thus avoid hanging.
Bug: http://curl.haxx.se/bug/view.cgi?id=1465
Reported-by: Jiri Dvorak
... even if they don't have an associated connection anymore. It could
leave the waiting transfers pending with no active one on the
connection.
Bug: http://curl.haxx.se/bug/view.cgi?id=1465
Reported-by: Jiri Dvorak
Since we just started make use of free(NULL) in order to simplify code,
this change takes it a step further and:
- converts lots of Curl_safefree() calls to good old free()
- makes Curl_safefree() not check the pointer before free()
The (new) rule of thumb is: if you really want a function call that
frees a pointer and then assigns it to NULL, then use Curl_safefree().
But we will prefer just using free() from now on.
The function "free" is documented in the way that no action shall occur for
a passed null pointer. It is therefore not needed that a function caller
repeats a corresponding check.
http://stackoverflow.com/questions/18775608/free-a-null-pointer-anyway-or-check-first
This issue was fixed by using the software Coccinelle 1.0.0-rc24.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
The code used some happy eyeballs logic even _after_ CONNECT has been
sent to a proxy, while the happy eyeball phase is already (should be)
over by then.
This is solved by splitting the multi state into two separate states
introducing the new SENDPROTOCONNECT state.
Bug: http://curl.haxx.se/mail/lib-2015-01/0170.html
Reported-by: Peter Laser
Since 1342a96ecf, a timeout detected in the multi state machine didn't
necesarily clear everything up, like formpost data.
Bug: https://github.com/bagder/curl/issues/147
Reported-by: Michel Promonet
Patched-by: Michel Promonet
When the connection code decides to close a socket it informs the multi
system via the Curl_multi_closed function. The multi system may, in
turn, invoke the CURLMOPT_SOCKETFUNCTION function with
CURL_POLL_REMOVE. This happens after the socket has already been
closed. Reorder the code so that CURL_POLL_REMOVE is called before the
socket is closed.
Debug output 'typo' fix.
Don't print an extra "0x" in
* Pipe broke: handle 0x0x2546d88, url = /
Add debug output.
Print the number of connections in the connection cache when
adding one, and not only when one is removed.
Fix typos in comments.
... for the local variable name in functions holding the return
code. Using the same name universally makes code easier to read and
follow.
Also, unify code for checking for CURLcode errors with:
if(result) or if(!result)
instead of
if(result == CURLE_OK), if(CURLE_OK == result) or if(result != CURLE_OK)
Coverify CID 1157776. Removed a superfluous if() that always evaluated
true (and an else clause that never ran), and then re-indented the
function accordingly.
As the current element in the list is free()d by Curl_llist_remove(),
when the associated connection is pending, reworked the loop to avoid
accessing the next element through e->next afterward.
... as the struct is free()d in the end anyway. It was first pointed out
to me that one of the ->msglist assignments were supposed to have been
->pending but was a copy and paste mistake when I realized none of the
clearing of pointers had to be there.
... instead of scanning through all handles, stash only the actual
handles that are in that state in the new ->pending list and scan that
list only. It should be mostly empty or very short. And only used for
pipelining.
This avoids a rather hefty slow-down especially notable if you add many
handles to the same multi handle. Regression introduced in commit
0f147887 (version 7.30.0).
Bug: http://curl.haxx.se/mail/lib-2014-07/0206.html
Reported-by: David Meyer
Introducing Curl_expire_latest(). To be used when we the code flow only
wants to get called at a later time that is "no later than X" so that
something can be checked (and another timeout be added).
The low-speed logic for example could easily be made to set very many
expire timeouts if it would be called faster or sooner than what it had
set its own timer and this goes for a few other timers too that aren't
explictiy checked for timer expiration in the code.
If there's no condition the code that says if(time-passed >= TIME), then
Curl_expire_latest() is preferred to Curl_expire().
If there exists such a condition, it is on the other hand important that
Curl_expire() is used and not the other.
Bug: http://curl.haxx.se/mail/lib-2014-06/0235.html
Reported-by: Florian Weimer
While waiting for a host resolve, check if the host cache may have
gotten the name already (by someone else), for when the same name is
resolved by several simultanoues requests.
The resolver thread occasionally gets stuck in getaddrinfo() when the
DNS or anything else is crappy or slow, so when a host is found in the
DNS cache, leave the thread alone and let itself cleanup the mess.
This reverts commit cb3e6dfa35 and instead fixes the problem
differently.
The reverted commit addressed a test failure in test 1021 by simplifying
and generalizing the code flow in a way that damaged the
performance. Now we modify the flow so that Curl_proxyCONNECT() again
does as much as possible in one go, yet still do test 1021 with and
without valgrind. It failed due to mistakes in the multi state machine.
Bug: http://curl.haxx.se/bug/view.cgi?id=1397
Reported-by: Paul Saab
When an error has been detected, skip the final forced call to the
progress callback by making sure to pass the current return code
variable in the Curl_done() call in the CURLM_STATE_DONE state.
This avoids the "extra" callback that could occur even if you returned
error from the progress callback.
Bug: http://curl.haxx.se/mail/lib-2014-06/0062.html
Reported by: Jonathan Cardoso Machado
Make all code use connclose() and connkeep() when changing the "close
state" for a connection. These two macros take a string argument with an
explanation, and debug builds of curl will include that in the debug
output. Helps tracking connection re-use/close issues.
In commit 0b3750b5c2 (released in 7.36.0) we fixed a timeout issue
but instead broke the timings.
To fix this, I introduce a new timestamp to use for the timeouts and
restored the previous timestamp and timestamp position so that the old
timer functionality is restored.
In addition to that, that change also broke connection timeouts for when
more than one connect was used (as it would then count the total time
from the first connect and not for the most recent one). Now
Curl_timeleft() has been modified so that it checks against different
start times depending on which timeout it checks.
Test 1303 is updated accordingly.
Bug: http://curl.haxx.se/mail/lib-2014-05/0147.html
Reported-by: Ryan Braud
set.infilesize in this case was modified in several places, which could
lead to repeated requests using the same handle to get unintendent/wrong
consequences based on what the previous request did!
ufds might not be allocated in case nfds overflows to zero while
extra_nfds is still non-zero. udfs is then accessed within the
extra_nfds-based for loop.
Setting the TIMER_STARTSINGLE timestamp first in CONNECT has the
drawback that for actions that go back to the CONNECT state, the time
stamp is reset and for the multi_socket API there's no corresponding
Curl_expire() then so the timeout logic gets wrong!
Reported-by: Brad Spencer
Bug: http://curl.haxx.se/mail/lib-2014-02/0036.html
A transfer timeout could result in an error message such as "Operation
timed out after 3000 milliseconds with 19 bytes of -1 received". This
patch removes the non-sensical "of -1" when the size of the transfer
is unknown, mirroring the logic in lib/transfer.c
With the recently added timeout "reminder" functionality, there's no
reason left for us to execute timeout code before the time is
ripe. Simplifies the handling too.
This will make the *TIMEOUT and *CONNECTTIMEOUT options more accurate
again, which probably is most important when the *_MS versions are used.
In multi_socket, make sure to update 'now' after having handled activity
on a socket.
BACKGROUND:
We have learned that on some systems timeout timers are inaccurate and
might occasionally fire off too early. To make the multi_socket API work
with this, we made libcurl execute timeout actions a bit early too if
they are within our MULTI_TIMEOUT_INACCURACY. (added in commit
2c72732ebf, present since 7.21.0)
Switching everything to the multi API made this inaccuracy problem
slightly more notable as now everyone can be affected.
Recently (commit 21091549c0) we tweaked that inaccuracy value to make
timeouts more accurate and made it platform specific. We also figured
out that we have code at places that check for fixed timeout values so
they MUST NOT run too early as then they will not trigger at all (see
commit be28223f35 and a691e04470) - so there are definitately problems
with running timeouts before they're supposed to run. (We've handled
that so far by adding the inaccuracy margin to those specific timeouts.)
The libcurl multi_socket API tells the application with a callback that
a timeout expires in N milliseconds (and it explicitly will not tell it
again for the same timeout), and the application is then supposed to
call libcurl when that timeout expires. When libcurl subsequently gets
called with curl_multi_socket_action(...CURL_SOCKET_TIMEOUT...), it
knows that the application thinks the timeout expired - and alas, if it
is within the inaccuracy level libcurl will run code handling that
handle.
If the application says CURL_SOCKET_TIMEOUT to libcurl and _isn't_
within the inaccuracy level, libcurl will not consider the timeout
expired and it will not tell the application again since the timeout
value is still the same.
NOW:
This change introduces a modified behavior here. If the application says
CURL_SOCKET_TIMEOUT and libcurl finds no timeout code to run, it will
inform the application about the timeout value - *again* even if it is
the same timeout that it already told about before (although libcurl
will of course tell it the updated time so that it'll still get the
correct remaining time). This way, we will not risk that the application
believes it has done its job and libcurl thinks the time hasn't come yet
to run any code and both just sit waiting. This also allows us to
decrease the MULTI_TIMEOUT_INACCURACY margin, but that will be handled
in a separate commit.
A repeated timeout update to the application risk that the timeout will
then fire again immediately and we have what basically is a busy-loop
until the time is fine even for libcurl. If that becomes a problem, we
need to address it.
When the progress callback returned 1 at a very early state, the code
would not make CURLE_ABORTED_BY_CALLBACK get returned but the process
would still be interrupted. In the HTTP case, this would then cause a
CURLE_GOT_NOTHING to erroneously get returned instead.
Reported-by: Petr Novak
Bug: http://curl.haxx.se/bug/view.cgi?id=1318
Following commit 0aafd77fa4, replaced the internal usage of
FORMAT_OFF_T and FORMAT_OFF_TU with the external versions that we
expect API programmers to use.
This negates the need for separate definitions which were subtly
different under different platforms/compilers.
The FILE:// code doesn't support this option - and it doesn't make sense
to support it as long as it works as it does since then it'd only block
even longer.
But: setting CURLOPT_MAX_RECV_SPEED_LARGE would make the transfer first
get done and then libcurl would wait until the average speed would get
low enough. This happened because the transfer happens completely in the
DO state for FILE:// but then it would still unconditionally continue in
to the PERFORM state where the speed check is made.
Starting now, the code will skip from DO_DONE to DONE immediately if no
socket is set to be recv()ed or send()ed to.
Bug: http://curl.haxx.se/bug/view.cgi?id=1312
Reported-by: Mohammad AlSaleh
This is an extension to the fix in 7d80ed64e4. We may
call Curl_disconnect() while cleaning up the multi handle,
which could lead to openssl sending packets, which could get
a SIGPIPE.
Signed-off-by: Jeff King <peff@peff.net>
This patch fixes and issue introduced in commit 7d7df83198, if the
tunnel state was TUNNEL_CONNECT, waitconnect_getsock() would return a
bitmask indicating a readable socket but never stored the socket in the
return array.
This patch adds a 200ms delay between the first and second address
family socket connection attempts.
It also iterates over IP addresses in the order returned by the
system, meaning most dual-stack systems will try IPv6 first.
Additionally, it refactors the connect code, removing most code that
handled synchronous connects. Since all sockets are now non-blocking,
the logic can be made simpler.
The code rejected 0 as a valid timeout while in fact the function could
indeed legitimately return that and it should be respected.
Reported-by: Bjorn Stenberg
This patch invokes two socket connect()s nearly simultaneously, and
the socket that is first connected "wins" and is subsequently used for
the connection. The other is terminated.
There is a very slight IPv4 preference, in that if both sockets connect
simultaneously IPv4 is checked first and thus will win.
When waiting for a 100-continue response from the server, the
Curl_readwrite() will refuse to run if called until the timeout has been
reached.
We timeout code in multi_socket() allows code to run slightly before the
actual timeout time, so for test 154 it could lead to the function being
executed but refused in Curl_readwrite() and then the application would
just sit idling forever.
This was detected with runtests.pl -e on test 154.
Make sure we always return CURLM_CALL_MULTI_PERFORM when we reach
CURLM_STATE_DONE since the state is transient and it can very well
continue executing as there is nothing to wait for.
Bug: http://curl.haxx.se/mail/lib-2013-08/0211.html
Reported-by: Yi Huang
