- 15 Apr, 2015 4 commits
-
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
Close the connection if authentication fails. Syslog authentication failures. Use /dev/[u]random directly to create secret files. Fail creation if none of them are available. Always use "varnishd" as syslog identity, but append "/" + up to seven leading alphanumeric chars from the -i(dent) argument if available.
-
Poul-Henning Kamp authored
-
Nils Goroll authored
-
- 14 Apr, 2015 10 commits
-
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
to run the varnishd worker process as a different user than the VCC and CC subprocesses. It is mandatory that the workuser has the same login group as the user subparamter. Recommended values for packaging: -junix,user=varnish "varnish" user has login group "varnish" -junix,workuser=vrun "vrun" user has login group "varnish"
-
Nils Goroll authored
-
Nils Goroll authored
-
Nils Goroll authored
-
Poul-Henning Kamp authored
Don't assert that we can remove VCL directories, compilers may have left crumbs there, for instance gcov profiling info.
-
Poul-Henning Kamp authored
Also open argv-supplied secret files with elevated privs.
-
Poul-Henning Kamp authored
owned by the initial uid/gid of the varnishd process.
-
Poul-Henning Kamp authored
-
- 13 Apr, 2015 6 commits
-
-
Poul-Henning Kamp authored
timer parameter.
-
Poul-Henning Kamp authored
-
Lasse Karstensen authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
Fixes #1699
-
- 10 Apr, 2015 4 commits
-
-
Federico G. Schwindt authored
-
Federico G. Schwindt authored
-
Martin Blix Grydeland authored
execising that
-
Martin Blix Grydeland authored
EPOLLPRI causes the waiter to report activity on a socket when there is OOB data available. Since HTTP does not make use of OOB the data is never read and the socket thus is always ready for read according to the waiter. This causes the waiter to continously assign the session to a worker only for the worker to find out there is no data sending it back to the waiter. This continues until the request timeout_idle has elapsed. This problem does no constitute a DOS attack vector as the linger timeout in the session SES_RxReq is still invoked, and this one does not trigger on PRI. So the worker sleeps on that poll for timeout_linger before sending the session back to the waiter, and eventually is dropped on timeout_idle.
-
- 09 Apr, 2015 15 commits
-
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
Fixes: #1674
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
This option controls the timeout for the initial VSM open operation
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
This makes query expression errors be reported before any VSM connection related errors.
-
Martin Blix Grydeland authored
This sets the input cursor to be used by the query.
-
Martin Blix Grydeland authored
This is to avoid having to delete the VSLQ context during VSM reopen. Also allows parsing of the query without having to open the VSM first, to facilitate early error messages.
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
After VSLQ_Flush() there should not be any ready vtx' to get rid of.
-
Martin Blix Grydeland authored
It isn't internal
-
Martin Blix Grydeland authored
-
- 08 Apr, 2015 1 commit
-
-
Poul-Henning Kamp authored
proxie'd addresses.
-