It's not safe to destroy objects inside a callback from a parent that
still has references to the object.
Formalize what the user code can indicate by its return code from the
callback functions and provide the implementations at the parents.
- LWSSSSRET_OK: no action, OK
- LWSSSSRET_DISCONNECT_ME: disconnect the underlying connection
- LWSSSSRET_DESTROY_ME: destroy the ss object
- LWSSSSRET_TX_DONT_SEND: for tx, give up the tx opportunity since nothing to send
Some streamtypes do not pass or receive payload meaningfully. Allow them
to just leave their related cb NULL. Ditto for state, although I'm not sure
how useful such a streamtype can be.
lws_dll2 has been available for a couple of years and lws_dll1 marked as
deprecated since v3.2. Remove it.
If it affects you, lws_dll2 has an almost identical api but has more
features, such as storing in the lws_dll2_t which owner it is listed by
and keeping a count of objects on the list at the owner object.
Adapt the pt sul owner list to be an array, and define two different lists,
one that acts like before and is the default for existing users, and another
that has the ability to cooperate with systemwide suspend to restrict the
interval spent suspended so that it will wake in time for the earliest
thing on this wake-suspend sul list.
Clean the api a bit and add lws_sul_cancel() that only needs the sul as the
argument.
Add a flag for client creation info to indicate that this client connection
is important enough that, eg, validity checking it to detect silently dead
connections should go on the wake-suspend sul list. That flag is exposed in
secure streams policy so it can be added to a streamtype with
"swake_validity": true
Deprecate out the old vhost timer stuff that predates sul. Add a flag
LWS_WITH_DEPRECATED_THINGS in cmake so users can get it back temporarily
before it will be removed in a v4.2.
Adapt all remaining in-tree users of it to use explicit suls.
The guy calling the callback with LADNS_RET_FAILED is going to
inform his caller that it failed... let him decide to close and
fail the connection attempt.
Allow selection of Unix Domain Sockets on windows since it is supported
for the last couple of years on windows 10
https://devblogs.microsoft.com/commandline/af_unix-comes-to-windows/
... if only they could add a full set of posix pieces to go with it
(and abstract namespace UDS which doesn't work apparently) so that
the parts dealing with uid / gid don't have to be disabled.
Establish a new distributed CMake architecture with CMake code related to
a source directory moving to be in the subdir in its own CMakeLists.txt.
In particular, there's now one in ./lib which calls through to ones
further down the directory tree like ./lib/plat/xxx, ./lib/roles/xxx etc.
This cuts the main CMakelists.txt from 98KB -> 33KB, about a 66% reduction,
and it's much easier to maintain sub-CMakeLists.txt that are in the same
directory as the sources they manage, and conceal all the details that that
level.
Child CMakelists.txt become responsible for:
- include_directories() definition (this is not supported by CMake
directly, it passes it back up via PARENT_SCOPE vars in helper
macros)
- Addition child CMakeLists.txt inclusion, for example toplevel ->
role -> role subdir
- Source file addition to the build
- Dependent library path resolution... this is now a private thing
in the child CMakeLists.txt, it just passes back any adaptations
to include_directories() and the LIB_LIST without filling the
parent namespace with the details
Esp-idf has an improved but still kind of abused cmake-
based build system now.
If we see ESP_PLATFORM coming as a cmake var, we can know we
are being built from inside the esp-idf config system.
Leave the existing esp32 arrangements alone but triggered off
ESP_PLATFORM, adapt to use the cross toolchain file and
various quirks automatically.
In this way you can build lws a part of your project in a
much cleaner way.
Prepare a minimal esp32 test app for use in Sai
Adapt .sai.json to build for esp32
If you encounter an h1 transaction with no content length and you
parsed the headers, it actually means a hangup subsequently is an
indication of a correct transaction completion. So take care to
do the _COMPLETION callback under those circumstances too.
Unlike any other sockaddr variant it turns out when sockaddr_un reports its
sizeof() to connect() or listen(), it is trimmed to the used length of the
sun_path[] member not including any trailing 0x00.
Until now we worked fine, but our actual UDS paths have a large number of
trailing 0x00 (shown as @ in most tools). Clients and servers can still
interoperate if they both have this broken name.
This patch fixes it to trim the sockaddr_un to the path length so the name
is as you would expect.
When most of ss-h2 was combined into ss-h1 during development, the h2 difference
about needing HTTP_FINAL to signal h2 FIN flag was accidentally dropped. In
many cases the peer can infer it, from, eg, content-length reached. But we need
to replace explicitly doing it to cover all cases.
POSIX connect() specifies it will signal POLLOUT available when
the connect result is available. But windows has some non-posix
nonsense.
Improve the plat support to simulate the missing POLLOUT.
Add a member to the vh init struct allowing control of the overall
connection wait introduced in an earlier patch. Set it to 20s
by default.
The timeout_secs member controls the individual DNS result
connect timeout and is reduced to 5s by default.
This patch allows client connections to recover from a nonresponsive
(ie, does not complete connect()) peer and continue to try subsequent
DNS results.
Replace the bash selftest plumbing with CTest.
To use the selftests, build with -DLWS_WITH_MINIMAL_EXAMPLES=1
and `CTEST_OUTPUT_ON_FAILURE=1 make test` or just
`make test`.
To disable tests that require internet access, also give
-DLWS_CTEST_INTERNET_AVAILABLE=0
Remove travis and appveyor scripts on master.
Remove travis and appveyor decals on README.md.
lejp_parse() return type is an int... but in the function, the temp
for it is a char. This leads to badness that is currently worked
around by casting the return through a signed char type.
But that leads to more badness since if there's >127 bytes of buffer
left after the end of the JSON object, we misreport it.
Bite the bullet and fix the temp type, and fix up all the guys
who were working around it at the caller return casting to use the
resulting straight int.
If you are using this api, remove any casting you may have cut-
and-pasted like this
n = (int)(signed char)lejp_parse(...);
... to just be like this...
n = lejp_parse(...);
Until now if the generic transaction completes and the connection is idle,
there's a fixed 5s grace period to keep the connection up in case something
else wants it.
This allows setting it in the client creation info struct .keep_warm_secs.
If left at 0, then it maintains the backward-compatible 5s wait.
You can disconnect the stream by returning -1 from tx(). You can
give up your chance to send anything by returning 1 from tx().
Returning 0 sends `*len` amount of the provided buffer.
Returning <0 from rx() also disconnects the stream.
There is no way to allocate a client wsi mux sid before the headers are
send, because we don't know the order in which new wsi headers will be sent
and so seen by the peer.
The peer inisists that sid indexes only increase... we cannot allocat sids
monotonically at the client and then send them disordered...
Add support for blob type in sqlite3... it's unusual in that it
is created into the table schema as a column of blob type, but is
not serialized or deserialized into or from JSON or sqlite.
Because the size of blobs is open-ended, accessing them in one
hit may not be possible, eg, exceed the size of available heap.
As binary, they would have to be base64-encoded in JSON
representation and that bloating may be excessive, with, eg,
a 500MB blob. So while they can be defined using lws_struct
schema, and coexist inside a column of a table managed by
lws_struct, they must be read and written separately.
X509_VERYFY_PARAM_Set1_host of openSSL allows the third argument, which
is the length of the hostname string, to be 0. Then, it assumes hostname
is a null-terminated C string. BoringSSL enforces the actual length to
be specified, and the hostname string should end with a null char.
Just provide the length, making both OpenSSL and BoringSSL happy.
