If a system call returns EMFILE, then admin should increase
server.max-fds and check/increase rlimits for num files (ulimit -Hn)
Alternatively, the admin might decrease server.max-connections to limit
the number of connections served in parallel.
avoids separate memory allocation for list of pointers
adds ability to check if con is already in joblist,
so do not re-add con if already in joblist
since con is checked if in joblist before being added to joblist,
there is no longer need for two lists and jobs can be processed
before poll() for to process new events
srv->lim_conns tracks remaining conns until limit is reached,
replacing (srv->max_conns - srv->conns.used)
srv->srvconf.max_conns is now updated at startup, so
srv->srvconf.max_conns serves as srv->max_conns
keep conns_pool of struct connection separate from conns list
and allocate conns list to srv->srvconf.max_conns size at startup
x-ref:
"Memory fragmentation with HTTP/2 enabled"
https://redmine.lighttpd.net/issues/3084
Most OS platforms have already provided solutions to
Y2038 32-bit signed time_t 5 - 10 years ago (or more!)
Notable exceptions are Linux i686 and FreeBSD i386.
Since 32-bit systems tend to be embedded systems,
and since many distros take years to pick up new software,
this commit aims to provide Y2038 mitigations for lighttpd
running on 32-bit systems with Y2038-unsafe 32-bit signed time_t
* Y2038: lighttpd 1.4.60 and later report Y2038 safety
$ lighttpd -V
+ Y2038 support # Y2038-SAFE
$ lighttpd -V
- Y2038 support (unsafe 32-bit signed time_t) # Y2038-UNSAFE
* Y2038: general platform info
* Y2038-SAFE: lighttpd 64-bit builds on platforms using 64-bit time_t
- all major 64-bit platforms (known to this author) use 64-bit time_t
* Y2038-SAFE: lighttpd 32-bit builds on platforms using 64-bit time_t
- Linux x32 ABI (different from i686)
- FreeBSD all 32-bit and 64-bit architectures *except* 32-bit i386
- NetBSD 6.0 (released Oct 2012) all 32-bit and 64-bit architectures
- OpenBSD 5.5 (released May 2014) all 32-bit and 64-bit architectures
- Microsoft Windows XP and Visual Studio 2005 (? unsure ?)
Another reference suggests Visual Studio 2015 defaults to 64-bit time_t
- MacOS 10.15 Catalina (released 2019) drops support for 32-bit apps
* Y2038-SAFE: lighttpd 32-bit builds on platforms using 32-bit unsigned time_t
- e.g. OpenVMS (unknown if lighttpd builds on this platform)
* Y2038-UNSAFE: lighttpd 32-bit builds on platforms using 32-bit signed time_t
- Linux 32-bit (including i686)
- glibc 32-bit library support not yet available for 64-bit time_t
- https://sourceware.org/glibc/wiki/Y2038ProofnessDesign
- Linux kernel 5.6 on 32-bit platforms does support 64-bit time_t
https://itsubuntu.com/linux-kernel-5-6-to-fix-the-year-2038-issue-unix-y2k/
- https://www.gnu.org/software/libc/manual/html_node/64_002dbit-time-symbol-handling.html
"Note: at this point, 64-bit time support in dual-time
configurations is work-in-progress, so for these
configurations, the public API only makes the 32-bit time
support available. In a later change, the public API will
allow user code to choose the time size for a given
compilation unit."
- compiling with -D_TIME_BITS=64 currently has no effect
- glibc recent (Jul 2021) mailing list discussion
- https://public-inbox.org/bug-gnulib/878s2ozq70.fsf@oldenburg.str.redhat.com/T/
- FreeBSD i386
- DragonFlyBSD 32-bit
* Y2038 mitigations attempted on Y2038-UNSAFE platforms (32-bit signed time_t)
* lighttpd prefers system monotonic clock instead of realtime clock
in places where realtime clock is not required
* lighttpd treats negative time_t values as after 19 Jan 2038 03:14:07 GMT
* (lighttpd presumes that lighttpd will not encounter dates before 1970
during normal operation.)
* lighttpd casts struct stat st.st_mtime (and st.st_*time) through uint64_t
to convert negative timestamps for comparisions with 64-bit timestamps
(treating negative timestamp values as after 19 Jan 2038 03:14:07 GMT)
* lighttpd provides unix_time64_t (int64_t) and
* lighttpd provides struct unix_timespec64 (unix_timespec64_t)
(struct timespec equivalent using unix_time64_t tv_sec member)
* lighttpd provides gmtime64_r() and localtime64_r() wrappers
for platforms 32-bit platforms using 32-bit time_t and
lighttpd temporarily shifts the year in order to use
gmtime_r() and localtime_r() (or gmtime() and localtime())
from standard libraries, before readjusting year and passing
struct tm to formatting functions such as strftime()
* lighttpd provides TIME64_CAST() macro to cast signed 32-bit time_t to
unsigned 32-bit and then to unix_time64_t
* Note: while lighttpd tries handle times past 19 Jan 2038 03:14:07 GMT
on 32-bit platforms using 32-bit signed time_t, underlying libraries and
underlying filesystems might not behave properly after 32-bit signed time_t
overflows (19 Jan 2038 03:14:08 GMT). If a given 32-bit OS does not work
properly using negative time_t values, then lighttpd likely will not work
properly on that system.
* Other references and blogs
- https://en.wikipedia.org/wiki/Year_2038_problem
- https://en.wikipedia.org/wiki/Time_formatting_and_storage_bugs
- http://www.lieberbiber.de/2017/03/14/a-look-at-the-year-20362038-problems-and-time-proofness-in-various-systems/
save parsed listen addrs at startup for reuse at runtime
srv_socket->srv_token is normalized at startup and contains IP and port.
save offset to colon, if present, or else length of string (unix socket)
At runtime, srv_token_colon can be quickly used as length of IP string
(without port) or, if not length of string, offset of stringified port
following the colon.
server.feature-flags += ("server.graceful-shutdown-timeout" => 10)
After receiving SIGINT or SIGUSR1, lighttpd will gracefully shutdown,
waiting for existing connections to complete. In the case of SIGUSR1,
this wait occurs before restarting lighttpd. The default timeout is
none (unlimited).
When "server.graceful-shutdown-timeout" option is set, it defines the
number of seconds that lighttpd will wait for existing connections to
complete before shutting down the connection.
Sites which expect large uploads or downloads, or those with very slow
clients, might want to set a much longer timeout, e.g 60 seconds
For more immediate graceful restarts, while still allowing existing
connections time to complete, sites should additionally consider
whether or not
server.feature-flags += ("server.graceful-restart-bg" => "enable")
is appropriate and compatible with their lighttpd.conf settings
graceful and (nearly) immediate lighttpd restart option
For *some* configurations, it *may* be safe to background the current
lighttpd server (or workers) to continue processing active requests
and, in parallel, to start up a new lighttpd server with a new
configuration. For other configurations, doing so might not be safe!
Therefore, this option must be explicitly configured to enable:
server.feature-flags += ("server.graceful-restart-bg" => "enable")
server.systemd-socket-activation = "enable"
Along with enabling server.feature-flags "server.graceful-restart-bg",
enabling server.systemd-socket-activation allows transfer of open
listening sockets to the new lighttpd server instance, and occurs
without closing the listening sockets and without destroying the
kernel listen backlog queue on the socket.
Safe configurations may include lighttpd.conf which connect to
standalone backend daemons, e.g. proxying to other servers,
including PHP-FPM backends.
Unsafe configurations include lighttpd.conf which use "bin-path" option
in *.server configs, instructing lighttpd to execute the backends.
Using the graceful-and-immediate-restart option is likely *unsafe* if
the backend daemon expects only one instance of itself to run at a time.
Current implementation of graceful and immediate restart option keeps
the backgrounded lighttpd in the same process group, so that subsequent
SIGINT or SIGTERM will shut down both the new and the backgrounded
servers. (An alternative option (commented out in the code) is to
background and detach from the new lighttpd process.) Regardless,
existing subprocesses, such as CGI, remain in original process group.
As a result, the new lighttpd server may receive SIGCHLD for unknown
processes inherited from the old server, which the new lighttpd server
will reap and discard. The original lighttpd server, now a child, will
be unable to detect exit or reap and report status on those pre-existing
subprocesses.
Graceful restart is triggered in lighttpd by sending lighttpd SIGUSR1.
If lighttpd is configured with workers, then SIGINT (not SIGUSR1) is
sent to the process group, including other processes started by
lighttpd, e.g. CGI. To work well with graceful restart, CGI scripts and
other processes should trap SIGINT (and SIGUSR1 for good measure).
Long-running scripts may want to checkpoint and close, e.g. a CGI script
implementing a long-running websocket connection.
Location response header is permitted to use relative-path in
RFC 7231 Section 7.1.2. Location
Prefer relative path in redirection for the benefit of reverse proxies
and CDNs. Doing so also avoids potentially disclosing internal schemes
and server names which client might not be able to directly reach.
To restore prior behavior of sending a fully-qualified absolute URI:
server.feature-flags += ("absolute-dir-redirect" => "enable")
x-ref:
https://bz.apache.org/bugzilla/show_bug.cgi?id=63357
(subsequently incrementally updated using git rebase)
huge props and many thank yous to writers of testing tools used while
developing HTTP/2 support in lighttpd:
h2spec - conformance testing tool for HTTP/2 implementation
https://github.com/summerwind/h2spec
h2load - HTTP/2 benchmarking tool
https://nghttp2.org/documentation/h2load-howto.html
curl - command line tool and library for transferring data with URLs
https://curl.haxx.se/
r->con->reqbody_read() replaces connection_handle_read_post_state()
future: might provide different callbacks for request body with
Content-Length versus request body sent via Transfer-Encoding: chunked
(pedantic; no impact)
upon error, server will exit, so the impact of momentarily leaking fd
has no impact. This commit holds the fd in srv->stdin_fd to address
Coverity warning about leaking fd when using server.bind = "/dev/stdin"
NB: r->tmp_buf == srv->tmp_buf (pointer is copied for quicker access)
NB: request read and write chunkqueues currently point to connection
chunkqueues; per-request and per-connection chunkqueues are
not distinct from one another
con->read_queue == r->read_queue
con->write_queue == r->write_queue
NB: in the future, a separate connection config may be needed for
connection-level module hooks. Similarly, might need to have
per-request chunkqueues separate from per-connection chunkqueues.
Should probably also have a request_reset() which is distinct from
connection_reset().
fdevent.c no longer directly uses struct server *srv
srv->srvconf.max_fds (if set) is used to set rlimits
set max_conns in server.c after fdevent_init(), which sets srv->max_fds
using srv->srvconf.max_fds (if set) as input hint
e.g. different server.errorlog for different virtual hosts
Also, support different server.breakagelog to have separate script
error logs, applicable to mod_cgi and mod_ssi exec.
use global rather than passing around (server *) just for that
li_itostrn() and li_utostrn() return string length
(rather than requiring subsequent strlen() to find length)
convert all log_error_write() to log_error() and pass (log_error_st *)
use con->errh in preference to srv->errh (even though currently same)
avoid passing (server *) when previously used only for logging (errh)
