Section: libcurl Manual (3)
Updated: December 22, 2020
curl_easy_pause - pause and unpause a connection
CURLcode curl_easy_pause(CURL *handle, int bitmask);
Using this function, you can explicitly mark a running connection to get
paused, and you can unpause a connection that was previously paused.
A connection can be paused by using this function or by letting the read or
the write callbacks return the proper magic return code
(CURL_READFUNC_PAUSE and CURL_WRITEFUNC_PAUSE). A write callback
that returns pause signals to the library that it couldn't take care of any
data at all, and that data will then be delivered again to the callback when
the transfer is unpaused.
While it may feel tempting, take care and notice that you cannot call this
function from another thread. To unpause, you may for example call it from the
progress callback (CURLOPT_PROGRESSFUNCTION(3)), which gets called at
least once per second, even if the connection is paused.
When this function is called to unpause receiving, the chance is high that you
will get your write callback called before this function returns.
The handle argument identifies the transfer you want to pause or
A paused transfer is excluded from low speed cancels via the
CURLOPT_LOW_SPEED_LIMIT(3) option and unpausing a transfer will reset
the time period required for the low speed limit to be met.
The bitmask argument is a set of bits that sets the new state of the
connection. The following bits can be used:
Pause receiving data. There will be no data received on this connection until
this function is called again without this bit set. Thus, the write callback
(CURLOPT_WRITEFUNCTION(3)) won't be called.
Pause sending data. There will be no data sent on this connection until this
function is called again without this bit set. Thus, the read callback
(CURLOPT_READFUNCTION(3)) won't be called.
Convenience define that pauses both directions.
Convenience define that unpauses both directions.
CURLE_OK (zero) means that the option was set properly, and a non-zero return
code means something wrong occurred after the new state was set. See the
man page for the full list with descriptions.
The pausing of transfers does not work with protocols that work without
network connectivity, like FILE://. Trying to pause such a transfer, in any
direction, will cause problems in the worst case or an error in the best case.
When a connection is used multiplexed, like for HTTP/2, and one of the
transfers over the connection is paused and the others continue flowing,
libcurl might end up buffering contents for the paused transfer. It has to do
this because it needs to drain the socket for the other transfers and the
already announced window size for the paused transfer will allow the server to
continue sending data up to that window size amount. By default, libcurl
announces a 32 megabyte window size, which thus can make libcurl end up
buffering 32 megabyte of data for a paused stream.
When such a paused stream is unpaused again, any buffered data will be
Added in libcurl 7.18.0.
When pausing a read by returning the magic return code from a write callback,
the read data is already in libcurl's internal buffers so it'll have to keep
it in an allocated buffer until the receiving is again unpaused using this
If the downloaded data is compressed and is asked to get uncompressed
automatically on download, libcurl will continue to uncompress the entire
downloaded chunk and it will cache the data uncompressed. This has the side-
effect that if you download something that is compressed a lot, it can result
in a very large data amount needing to be allocated to save the data during
the pause. This said, you should probably consider not using paused receiving
if you allow libcurl to uncompress data automatically.