allow debug.log-state-handling to be enabled in config conditions,
even though the connection and request may already have begun
(and therefore not logging state changes which already occurred)
(also, minor additional structure cleanups made in the vicinity)
Existing behavior: strict header parsing, strict host parsing, and
basic url normalization are enabled by default in lighttpd, and
protect against h2 request tunnelling attempts using invalid chars
Keeping strict parsing settings enabled is strongly recommended.
If any are explicitly disabled in lighttpd.conf, then this patch takes
steps to thwart h2 request tunnelling.
"HTTP/2: The Sequel is Always Worse"
Existing behavior: mod_proxy *does not* forward Upgrade header
unless explicitly enabled in lighttpd.conf (default: not enabled)
(proxy.header += ("upgrade" => "enable"))
mod_cgi previously used to forward Upgrade request header, but would
remove Upgrade response header if cgi.upgrade was not explicitly enabled
(cgi.upgrade = "enable")
This patch thwarts h2c smuggling when lighttpd.conf has also been
explicitly configured to pass "Upgrade" request header
"h2c Smuggling: Request Smuggling Via HTTP/2 Cleartext (h2c)"
make ck_memeq_const_time() more generically reusable
remove implementation requirements that strings be '\0' terminated,
or at least have an initialized byte following each string (s[len])
"mod_auth cache password doesn't match"
reschedule HTTP/2 streams w/ pending data if con->write_queue is empty
reschedule to read deferred (and pending) HTTP/2 frames when
MAX_CONCURRENT_STREAMS are active and a stream completes.
"HTTP/2 requests sometimes take very long (missing last chunk)"
ignore PROPFIND "Depth" request header for files (non-collections)
RFC4918 10.2. Depth Header
"If a resource does not have internal members, then the Depth header MUST be ignored."
"Webdav + rclone backup"
add comment for handling of streaming with no response body
add commented-out code to disable streaming to wait for backend protocol
to signal end of response (prevent http_response_write_prepare() from
short-circuiting and finishing responses without response body)
"FastCGI premature socket close with response streaming and 204 status"
Basic auth passwd cache might fail to match when it should have
matched (false negative) when comparing an uninitialized byte.
That bug "fails closed" and does not use the cache when it could.
This patch allows for proper match in the cache when it should match.
"mod_auth cache password doesn't match"
fix HTTP/2 upload > 64k with server.max-request-size > 0
(regression present only in lighttpd 1.4.60)
"File upload is broken after upgrade from 1.4.59 to 1.4.60"
preserve remote IP until request reset
(historical IP restore in request_done hook was obsoleted by
commit fea5bdc8 in which request plugin context was split from
connection plugin context, and by much older commits which
ensured that request_reset hook was always run)
"Remote address behind reverse proxy not logged"
Earlier macOS need _DARWIN_C_SOURCE defined for errno_t from sys/types.h
Alternatively, define __STDC_WANT_LIB_EXT1__ >= 1 before include errno.h
"error: unknown type name 'errno_t' (macOS 10.13.6)"
do not read-ahead past '\0' while url-decoding
lighttpd 1.4.60 could previously have read one byte of potentially
uninitialized data. lighttpd detects the '\0' so there is no exposure
of data. This also can not cause a crash in lighttpd 1.4.60 due to how
lighttpd 1.4.60 allocates memory for buffers in sizes (power-2 + 1),
and typical system malloc alignment of 4- or 8- bytes.
chunkqueue_peek_data() experiment to mmap temporary files
(currently disabled in code due to not making measurable difference
in throughput (during a specific microbenchmark load test),
though it does reduce CPU use by ~10% in the same microbenchmark)
enabling this may cause large spikes in RSS mem usage reported by the
system, due to the read-only memory maps of the temporary files,
but this is nothing to be alarmed about, as the memory maps are
file-backed and read-only, so minimally add to memory pressure
allow up to 32k of data frames per stream per round
(previously limited to single max_frame_size (default 16k))
For 8 streams, 32k*8 is 256k, which is current lighttpd MAX_WRITE_LIMIT,
so each stream still gets a chance to write data (unless write queue
not emptied on previous attempt, reducing add limit this round)
cap size of data framed for HTTP/2 response until more data sent to
make sure to reschedule connection in job queue if max_bytes reached
and then the entire con->write_queue was flushed to network, or else
there is a chance the request may not get rescheduled (and then will
timeout) if the request is completed from the backend and there is
no other traffic or streams to trigger connection processing.
(check con->write_queue > 8k rather than empty from last round,
since small frames such as connection preface may have been added
this round while processing con->read_queue)
ignore SIGINT, SIGUSR1 in fcgi-responder if HAVE_SIGNAL is defined
(must be defined separately since config.h is not included)
Not required for test framework. Added as an example in the code,
e.g. if code is reused with lighttpd and graceful shutdown or restart.
(backend will be sent SIGTERM when server is ready to restart)