Before this patch curl used the C preprocessor to override standard
memory allocation symbols: malloc, calloc, strdup, realloc, free.
The goal of these is to replace them with curl's debug wrappers in
`CURLDEBUG` builds, another was to replace them with the wrappers
calling user-defined allocators in libcurl. This solution needed a bunch
of workarounds to avoid breaking external headers: it relied on include
order to do the overriding last. For "unity" builds it needed to reset
overrides before external includes. Also in test apps, which are always
built as single source files. It also needed the `(symbol)` trick
to avoid overrides in some places. This would still not fix cases where
the standard symbols were macros. It was also fragile and difficult
to figure out which was the actual function behind an alloc or free call
in a specific piece of code. This in turn caused bugs where the wrong
allocator was accidentally called.
To avoid these problems, this patch replaces this solution with
`curlx_`-prefixed allocator macros, and mapping them _once_ to either
the libcurl wrappers, the debug wrappers or the standard ones, matching
the rest of the code in libtests.
This concludes the long journey to avoid redefining standard functions
in the curl codebase.
Note: I did not update `packages/OS400/*.c` sources. They did not
`#include` `curl_setup.h`, `curl_memory.h` or `memdebug.h`, meaning
the overrides were never applied to them. This may or may not have been
correct. For now I suppressed the direct use of standard allocators
via a local `.checksrc`. Probably they (except for `curlcl.c`) should be
updated to include `curl_setup.h` and use the `curlx_` macros.
This patch changes mappings in two places:
- `lib/curl_threads.c` in libtests: Before this patch it mapped to
libcurl allocators. After, it maps to standard allocators, like
the rest of libtests code.
- `units`: before this patch it mapped to standard allocators. After, it
maps to libcurl allocators.
Also:
- drop all position-dependent `curl_memory.h` and `memdebug.h` includes,
and delete the now unnecessary headers.
- rename `Curl_tcsdup` macro to `curlx_tcsdup` and define like the other
allocators.
- map `curlx_strdup()` to `_strdup()` on Windows (was: `strdup()`).
To fix warnings silenced via `_CRT_NONSTDC_NO_DEPRECATE`.
- multibyte: map `curlx_convert_*()` to `_strdup()` on Windows
(was: `strdup()`).
- src: do not reuse the `strdup` name for the local replacement.
- lib509: call `_strdup()` on Windows (was: `strdup()`).
- test1132: delete test obsoleted by this patch.
- CHECKSRC.md: update text for `SNPRINTF`.
- checksrc: ban standard allocator symbols.
Follow-up to b12da22db1#18866
Follow-up to db98daab05#18844
Follow-up to 4deea9396b#18814
Follow-up to 9678ff5b1b#18776
Follow-up to 10bac43b87#18774
Follow-up to 20142f5d06#18634
Follow-up to bf7375ecc5#18503
Follow-up to 9863599d69#18502
Follow-up to 3bb5e58c10#17827Closes#19626
Change the send parameter from `const void *` to `const uint8_t *` and
adapt calling code. Several had already unsigned chars and were casting.
Closes#19729
Add protocol handler flag `PROTOPT_CONN_REUSE` to indicate that the
protocol allows reusing connections for other tranfers. Add that
to all handlers that support it.
Create connections with `conn->bits.close = FALSE` and remove all
the `connkeep()` calls in protocol handlers setup/connect implementations.
`PROTOPT_CONN_REUSE` assures that the default behaviour applies
at the end of a transfer without need to juggle the close bit.
`conn->bits.close` now serves as an additional indication that a
connection cannot be reused. Only protocol handles that allow
reuse need to set it to override the default behaviour.
Remove all `connclose()` and `connkeep()` calls from connection
filters. Filters should not modify connection flags. They are
supposed to run in eyeballing situations where a filter is just
one of many determining the outcome.
Fix http response header handling to only honour `Connection: close`
for HTTP/1.x versions.
Closes#19333
Description of how this works in `docs/internal/RATELIMITS.ms`.
Notable implementation changes:
- KEEP_SEND_PAUSE/KEEP_SEND_HOLD and KEEP_RECV_PAUSE/KEEP_RECV_HOLD
no longer exist. Pausing is down via blocked the new rlimits.
- KEEP_SEND_TIMED no longer exists. Pausing "100-continue" transfers
is done in the new `Curl_http_perform_pollset()` method.
- HTTP/2 rate limiting implemented via window updates. When
transfer initiaiting connection has a ratelimit, adjust the
initial window size
- HTTP/3 ngtcp2 rate limitin implemnented via ack updates
- HTTP/3 quiche does not seem to support this via its API
- the default progress-meter has been improved for accuracy
in "current speed" results.
pytest speed tests have been improved.
Closes#19384
Add new functions in `curlx/warnless.h` for controlled type
conversions:
* curlx_uitouz, convert unsigned into to size_t (should always work)
* curlx_uztoso, convert size_t to curl_off_t, capping at CURL_OFF_T_MAX
* curlx_sztouz, convert ssize_t to size_t, return TRUE when ok
* curlx_sotouz_range, convert curl_off_t to size_t interval, capping
values to interval bounds
Remove some unnecesary casts, convert some internal recv functions
to the "return result, have size_t* arg" pattern.
Closes#19495
Rename `Curl_timeleft()` to `Curl_timeleft_ms()` to make the units in
the returned `timediff_t` clear. (We used to always have ms there, but
with QUIC started to sometimes calc ns as well).
Rename some assigned vars without `_ms` suffix for clarity as well.
Closes#19486
After this patch, the codebase no longer overrides system printf
functions. Instead it explicitly calls either the curl printf functions
`curl_m*printf()` or the system ones using their original names.
Also:
- drop unused `curl_printf.h` includes.
- checksrc: ban system printf functions, allow where necessary.
Follow-up to db98daab05#18844
Follow-up to 4deea9396b#18814Closes#18866
`getsock()` calls operated on a global limit that could
not be configure beyond 16 sockets. This is no longer adequate
with the new happy eyeballing strategy.
Instead, do the following:
- make `struct easy_pollset` dynamic. Starting with
a minimal room for two sockets, the very common case,
allow it to grow on demand.
- replace all protocol handler getsock() calls with pollsets
and a CURLcode to return failures
- add CURLcode return for all connection filter `adjust_pollset()`
callbacks, since they too can now fail.
- use appropriately in multi.c and multi_ev.c
- fix unit2600 to trigger pollset growth
Closes#18164
Eliminating the socket readability check in the socket connection
filters for the 'data_pending' callback. Improves performance of
handling of transfers, up to ~30%, depending on parallelism and response
size.
Whatever `data_pending()` once was, its semantics are now:
"Is there anything buffered in the connection filters that needs
receive?"
Any checks of the socket's readability are done via `multi_wait()`
and friends.
Fix the one place in HTTP/1 proxy code that checked `data_pending()` and
did an early return if false. Remove that check and actually try to
receive data every time.
Closes#17785
Connection filters had a method `get_host()` which had not really been
documented. Since then, the cf had the `query()` method added. Replace
the separate get_host with query.
Add `CF_QUERY_HOST_PORT` as query to connection filters to retrieve
which remote hostname and port the filter (or its sub-filter) is talking
to. The query is implemented by HTTP and SOCKS filters, all others pass
it through.
Add `Curl_conn_get_current_host()` to retrieve the remote host and port
for a connection. During connect, this will return the host the
connection is talking to right now. Before/After connect, this will
return `conn->host.name`.
This is used by SASL authentication.
Closes#17419
cfilter/conn: change send/recv function signatures. Unify the
calling/return conventions in our send/receive handling.
Curl_conn_recv(), adjust pnread type
Parameter `pnread` was a `ssize_t *`, but `size_t *` is better since the
function returns any error in its `CURLcode` return value.
Closes#17546
Move curlx_ functions into its own subdir.
The idea is to use the curlx_ prefix proper on these functions, and use
these same function names both in tool, lib and test suite source code.
Stop the previous special #define setup for curlx_ names.
The printf defines are now done for the library alone. Tests no longer
use the printf defines. The tool code sets its own defines. The printf
functions are not curlx, they are publicly available.
The strcase defines are not curlx_ functions and should not be used by
tool or server code.
dynbuf, warnless, base64, strparse, timeval, timediff are now proper
curlx functions.
When libcurl is built statically, the functions from the library can be
used as-is. The key is then that the functions must work as-is, without
having to be recompiled for use in tool/tests. This avoids symbol
collisions - when libcurl is built statically, we use those functions
directly when building the tool/tests. When libcurl is shared, we
build/link them separately for the tool/tests.
Assisted-by: Jay Satiro
Closes#17253
- cf-h1-proxy: check return code and return error if the parser fails
- http: make the Retry-After parser check for a date string first then
number to avoid mis-parsing the begining of a date as a number
Closes#16548
Remove `blocking` argument from cfilter's connect method.
Implement blocking behaviour in Curl_conn_connect() instead for all
filter chains.
Update filters implementations. Several of which did never use the
paramter (QUIC for example). Simplifies connect handling in TLS filters
that no longer need to loop
Fixed a blocking connect call in FTP when waiting on a socket accept()
which only worked because the filter did not implement it.
Closes#16397
lib : remove all hyper code
configure: stop detecting hyper
docs: no more mention of hyper
tests: mo more special-handling of hyper builds
CI: no jobs using hyper
Closes#15120
Adds a `bool eos` flag to send methods to indicate that the data
is the last chunk the invovled transfer wants to send to the server.
This will help protocol filters like HTTP/2 and 3 to forward the
stream's EOF flag and also allow to EAGAIN such calls when buffers
are not yet fully flushed.
Closes#14220
Adds a `bool eos` flag to send methods to indicate that the data is the
last chunk the invovled transfer wants to send to the server.
This will help protocol filters like HTTP/2 and 3 to forward the
stream's EOF flag and also allow to EAGAIN such calls when buffers are
not yet fully flushed.
Closes#14220
Based on the standards and guidelines we use for our documentation.
- expand contractions (they're => they are etc)
- host name = > hostname
- file name => filename
- user name = username
- man page => manpage
- run-time => runtime
- set-up => setup
- back-end => backend
- a HTTP => an HTTP
- Two spaces after a period => one space after period
Closes#14073
This adds connection shutdown infrastructure and first use for FTP. FTP
data connections, when not encountering an error, are now shut down in a
blocking way with a 2sec timeout.
- add cfilter `Curl_cft_shutdown` callback
- keep a shutdown start timestamp and timeout at connectdata
- provide shutdown timeout default and member in
`data->set.shutdowntimeout`.
- provide methods for starting, interrogating and clearing
shutdown timers
- provide `Curl_conn_shutdown_blocking()` to shutdown the
`sockindex` filter chain in a blocking way. Use that in FTP.
- add `Curl_conn_cf_poll()` to wait for socket events during
shutdown of a connection filter chain.
This gets the monitoring sockets and events via the filters
"adjust_pollset()" methods. This gives correct behaviour when
shutting down a TLS connection through a HTTP/2 proxy.
- Implement shutdown for all socket filters
- for HTTP/2 and h2 proxying to send GOAWAY
- for TLS backends to the best of their capabilities
- for tcp socket filter to make a final, nonblocking
receive to avoid unwanted RST states
- add shutdown forwarding to happy eyeballers and
https connect ballers when applicable.
Closes#13904
They came up ealier with gcc 12 (Windows), but apparently gcc 14 is
still reporting them, also under Linux.
```
/home/runner/work/curl-for-win/curl-for-win/curl/lib/cf-h1-proxy.c: In function 'cf_h1_proxy_close':
/home/runner/work/curl-for-win/curl-for-win/curl/lib/cf-h1-proxy.c:1060:17: warning: null pointer dereference [-Wnull-dereference]
1060 | cf->connected = FALSE;
/home/runner/work/curl-for-win/curl-for-win/curl/lib/cf-h1-proxy.c:1061:8: warning: null pointer dereference [-Wnull-dereference]
1061 | if(cf->ctx) {
| ~~^~~~~
In function 'tunnel_free',
inlined from 'cf_h1_proxy_destroy' at /home/runner/work/curl-for-win/curl-for-win/curl/lib/cf-h1-proxy.c:1053:3:
/home/runner/work/curl-for-win/curl-for-win/curl/lib/cf-h1-proxy.c:198:27: warning: null pointer dereference [-Wnull-dereference]
198 | struct h1_tunnel_state *ts = cf->ctx;
| ^~
```
Ref: https://github.com/curl/curl-for-win/actions/runs/8985369476/job/24679219528#step:3:6320Fixes#13237Closes#13555
A transfer may do several `SingleRequest`s for its success. This happens
regularly for authentication, follows and retries on failed connections.
The "readwrite()" calls and functions connected to those carried a `bool
*done` parameter to indicate that the current `SingleRequest` is over.
This may happen before `upload_done` or `download_done` bits of
`SingleRequest` are set.
The problem with that is now `write_resp()` protocol handlers are
invoked in places where the `bool *done` cannot be passed up to the
caller. Instead of being a bool in the call chain, it needs to become a
member of `SingleRequest`, reflecting its state.
This removes the `bool *done` parameter and adds the `done` bit to
`SingleRequest` instead. It adds `Curl_req_soft_reset()` for using a
`SingleRequest` in a follow up, clearing `done` and other
flags/counters.
Closes#13096
- Move all the "upload_done" handling to request.c
- add possibility to abort sending of a request
- add `Curl_req_done_sending()` for checks
- transfer.c: readwrite_upload() now clean
- removing data->state.ulbuf and data->req.upload_fromhere
- as well as data->req.upload_present
- set data->req.upload_done on having read all from
the client and completely flushed the send buffer
- tftp, remove setting of data->req.upload_fromhere
- serves no purpose as `upload_present` is not set
and the data itself is directly `sendto()` anyway
- smtp, make upload EOB conversion a client reader
- xfer_ulbuf addition
- add xfer_ulbuf for borrowing, similar to xfer_buf
- use in file upload
- use in c-hyper body sending
- h1-proxy, remove init of data->state.uilbuf that is never used
- smb, add own send_buf instead of using data->state.ulbuf
Closes#13010
- replace `Curl_read()`, `Curl_write()` and `Curl_nwrite()` to
clarify when and at what level they operate
- send/recv of transfer related data is now done via
`Curl_xfer_send()/Curl_xfer_recv()` which no longer has
socket/socketindex as parameter. It decides on the transfer
setup of `conn->sockfd` and `conn->writesockfd` on which
connection filter chain to operate.
- send/recv on a specific connection filter chain is done via
`Curl_conn_send()/Curl_conn_recv()` which get the socket index
as parameter.
- rename `Curl_setup_transfer()` to `Curl_xfer_setup()` for
naming consistency
- clarify that the special CURLE_AGAIN hangling to return
`CURLE_OK` with length 0 only applies to `Curl_xfer_send()`
and CURLE_AGAIN is returned by all other send() variants.
- fix a bug in websocket `curl_ws_recv()` that mixed up data
when it arrived in more than a single chunk (to be made
into a sperate PR, also)
Added as documented [in
CLIENT-READER.md](5b1f31dfba/docs/CLIENT-READERS.md).
- old `Curl_buffer_send()` completely replaced by new `Curl_req_send()`
- old `Curl_fillreadbuffer()` replaced with `Curl_client_read()`
- HTTP chunked uploads are now formatted in a client reader added when
needed.
- FTP line-end conversions are done in a client reader added when
needed.
- when sending requests headers, remaining buffer space is filled with
body data for sending in "one go". This is independent of the request
body size. Resolves#12938 as now small and large requests have the
same code path.
Changes done to test cases:
- test513: now fails before sending request headers as this initial
"client read" triggers the setup fault. Behaves now the same as in
hyper build
- test547, test555, test1620: fix the length check in the lib code to
only fail for reads *smaller* than expected. This was a bug in the
test code that never triggered in the old implementation.
Closes#12969
- replace `Curl_read()`, `Curl_write()` and `Curl_nwrite()` to
clarify when and at what level they operate
- send/recv of transfer related data is now done via
`Curl_xfer_send()/Curl_xfer_recv()` which no longer has
socket/socketindex as parameter. It decides on the transfer
setup of `conn->sockfd` and `conn->writesockfd` on which
connection filter chain to operate.
- send/recv on a specific connection filter chain is done via
`Curl_conn_send()/Curl_conn_recv()` which get the socket index
as parameter.
- rename `Curl_setup_transfer()` to `Curl_xfer_setup()` for
naming consistency
- clarify that the special CURLE_AGAIN hangling to return
`CURLE_OK` with length 0 only applies to `Curl_xfer_send()`
and CURLE_AGAIN is returned by all other send() variants.
- fix a bug in websocket `curl_ws_recv()` that mixed up data
when it arrived in more than a single chunk
The method for sending not just raw bytes, but bytes that are either
"headers" or "body". The send abstraction stack, to to bottom, now is:
* `Curl_req_send()`: has parameter to indicate amount of header bytes,
buffers all data.
* `Curl_xfer_send()`: knows on which socket index to send, returns
amount of bytes sent.
* `Curl_conn_send()`: called with socket index, returns amount of bytes
sent.
In addition there is `Curl_req_flush()` for writing out all buffered
bytes.
`Curl_req_send()` is active for requests without body,
`Curl_buffer_send()` still being used for others. This is because the
special quirks need to be addressed in future parts:
* `expect-100` handling
* `Curl_fillreadbuffer()` needs to add directly to the new
`data->req.sendbuf`
* special body handlings, like `chunked` encodings and line end
conversions will be moved into something like a Client Reader.
In functions of the pattern `CURLcode xxx_send(..., ssize_t *written)`,
replace the `ssize_t` with a `size_t`. It makes no sense to allow for negative
values as the returned `CURLcode` already specifies error conditions. This
allows easier handling of lengths without casting.
Closes#12964
Curl_read/Curl_write clarifications
- replace `Curl_read()`, `Curl_write()` and `Curl_nwrite()` to 1clarify
when and at what level they operate
- send/recv of transfer related data is now done via
`Curl_xfer_send()/Curl_xfer_recv()` which no longer has
socket/socketindex as parameter. It decides on the transfer setup of
`conn->sockfd` and `conn->writesockfd` on which connection filter
chain to operate.
- send/recv on a specific connection filter chain is done via
`Curl_conn_send()/Curl_conn_recv()` which get the socket index as
parameter.
- rename `Curl_setup_transfer()` to `Curl_xfer_setup()` for naming
consistency
- clarify that the special CURLE_AGAIN handling to return `CURLE_OK`
with length 0 only applies to `Curl_xfer_send()` and CURLE_AGAIN is
returned by all other send() variants.
SingleRequest reshuffling
- move functions into request.[ch]
- differentiate between reset and free
- add Curl_req_done() to perform last actions
- add a send `bufq` to SingleRequest for future use in keeping upload data
Closes#12963
This clarifies the handling of server responses by folding the code for
the complicated protocols into their protocol handlers. This concerns
mainly HTTP and its bastard sibling RTSP.
The terms "read" and "write" are often used without clear context if
they refer to the connect or the client/application side of a
transfer. This PR uses "read/write" for operations on the client side
and "send/receive" for the connection, e.g. server side. If this is
considered useful, we can revisit renaming of further methods in another
PR.
Curl's protocol handler `readwrite()` method been changed:
```diff
- CURLcode (*readwrite)(struct Curl_easy *data, struct connectdata *conn,
- const char *buf, size_t blen,
- size_t *pconsumed, bool *readmore);
+ CURLcode (*write_resp)(struct Curl_easy *data, const char *buf, size_t blen,
+ bool is_eos, bool *done);
```
The name was changed to clarify that this writes reponse data to the
client side. The parameter changes are:
* `conn` removed as it always operates on `data->conn`
* `pconsumed` removed as the method needs to handle all data on success
* `readmore` removed as no longer necessary
* `is_eos` as indicator that this is the last call for the transfer
response (end-of-stream).
* `done` TRUE on return iff the transfer response is to be treated as
finished
This change affects many files only because of updated comments in
handlers that provide no implementation. The real change is that the
HTTP protocol handlers now provide an implementation.
The HTTP protocol handlers `write_resp()` implementation will get passed
**all** raw data of a server response for the transfer. The HTTP/1.x
formatted status and headers, as well as the undecoded response
body. `Curl_http_write_resp_hds()` is used internally to parse the
response headers and pass them on. This method is public as the RTSP
protocol handler also uses it.
HTTP/1.1 "chunked" transport encoding is now part of the general
*content encoding* writer stack, just like other encodings. A new flag
`CLIENTWRITE_EOS` was added for the last client write. This allows
writers to verify that they are in a valid end state. The chunked
decoder will check if it indeed has seen the last chunk.
The general response handling in `transfer.c:466` happens in function
`readwrite_data()`. This mainly operates now like:
```
static CURLcode readwrite_data(data, ...)
{
do {
Curl_xfer_recv_resp(data, buf)
...
Curl_xfer_write_resp(data, buf)
...
} while(interested);
...
}
```
All the response data handling is implemented in
`Curl_xfer_write_resp()`. It calls the protocol handler's `write_resp()`
implementation if available, or does the default behaviour.
All raw response data needs to pass through this function. Which also
means that anyone in possession of such data may call
`Curl_xfer_write_resp()`.
Closes#12480
https://best.openssf.org/Compiler-Hardening-Guides/Compiler-Options-Hardening-Guide-for-C-and-C++.html
as of 2023-11-29 [1].
Enable new recommended warnings (except `-Wsign-conversion`):
- enable `-Wformat=2` for clang (in both cmake and autotools).
- add `CURL_PRINTF()` internal attribute and mark functions accepting
printf arguments with it. This is a copy of existing
`CURL_TEMP_PRINTF()` but using `__printf__` to make it compatible
with redefinting the `printf` symbol:
https://gcc.gnu.org/onlinedocs/gcc-3.0.4/gcc_5.html#SEC94
- fix `CURL_PRINTF()` and existing `CURL_TEMP_PRINTF()` for
mingw-w64 and enable it on this platform.
- enable `-Wimplicit-fallthrough`.
- enable `-Wtrampolines`.
- add `-Wsign-conversion` commented with a FIXME.
- cmake: enable `-pedantic-errors` the way we do it with autotools.
Follow-up to d5c0351055#2747
- lib/curl_trc.h: use `CURL_FORMAT()`, this also fixes it to enable format
checks. Previously it was always disabled due to the internal `printf`
macro.
Fix them:
- fix bug where an `set_ipv6_v6only()` call was missed in builds with
`--disable-verbose` / `CURL_DISABLE_VERBOSE_STRINGS=ON`.
- add internal `FALLTHROUGH()` macro.
- replace obsolete fall-through comments with `FALLTHROUGH()`.
- fix fallthrough markups: Delete redundant ones (showing up as
warnings in most cases). Add missing ones. Fix indentation.
- silence `-Wformat-nonliteral` warnings with llvm/clang.
- fix one `-Wformat-nonliteral` warning.
- fix new `-Wformat` and `-Wformat-security` warnings.
- fix `CURL_FORMAT_SOCKET_T` value for mingw-w64. Also move its
definition to `lib/curl_setup.h` allowing use in `tests/server`.
- lib: fix two wrongly passed string arguments in log outputs.
Co-authored-by: Jay Satiro
- fix new `-Wformat` warnings on mingw-w64.
[1] 56c0fde389/docs/Compiler-Hardening-Guides/Compiler-Options-Hardening-Guide-for-C-and-C%2B%2B.mdCloses#12489
- changed header/chunk/handler->readwrite prototypes to accept `buf`,
`blen` and a `pconsumed` pointer. They now get the buffer to work on
and report back how many bytes they consumed
- eliminated `k->str` in SingleRequest
- improved excess data handling to properly calculate with any body data
left in the headerb buffer
- eliminated `k->badheader` enum to only be a bool
Closes#12283
This PR has these changes:
Renaming of unencode_* to cwriter, e.g. client writers
- documentation of sendf.h functions
- move max decode stack checks back to content_encoding.c
- define writer phase which was used as order before
- introduce phases for monitoring inbetween decode phases
- offering default implementations for init/write/close
Add type paramter to client writer's do_write()
- always pass all writes through the writer stack
- writers who only care about BODY data will pass other writes unchanged
add RAW and PROTOCOL client writers
- RAW used for Curl_debug() logging of CURLINFO_DATA_IN
- PROTOCOL used for updates to data->req.bytecount, max_filesize checks and
Curl_pgrsSetDownloadCounter()
- remove all updates of data->req.bytecount and calls to
Curl_pgrsSetDownloadCounter() and Curl_debug() from other code
- adjust test457 expected output to no longer see the excess write
Closes#12184
Connection filter had a `get_select_socks()` method, inspired by the
various `getsocks` functions involved during the lifetime of a
transfer. These, depending on transfer state (CONNECT/DO/DONE/ etc.),
return sockets to monitor and flag if this shall be done for POLLIN
and/or POLLOUT.
Due to this design, sockets and flags could only be added, not
removed. This led to problems in filters like HTTP/2 where flow control
prohibits the sending of data until the peer increases the flow
window. The general transfer loop wants to write, adds POLLOUT, the
socket is writeable but no data can be written.
This leads to cpu busy loops. To prevent that, HTTP/2 did set the
`SEND_HOLD` flag of such a blocked transfer, so the transfer loop cedes
further attempts. This works if only one such filter is involved. If a
HTTP/2 transfer goes through a HTTP/2 proxy, two filters are
setting/clearing this flag and may step on each other's toes.
Connection filters `get_select_socks()` is replaced by
`adjust_pollset()`. They get passed a `struct easy_pollset` that keeps
up to `MAX_SOCKSPEREASYHANDLE` sockets and their `POLLIN|POLLOUT`
flags. This struct is initialized in `multi_getsock()` by calling the
various `getsocks()` implementations based on transfer state, as before.
After protocol handlers/transfer loop have set the sockets and flags
they want, the `easy_pollset` is *always* passed to the filters. Filters
"higher" in the chain are called first, starting at the first
not-yet-connection one. Each filter may add sockets and/or change
flags. When all flags are removed, the socket itself is removed from the
pollset.
Example:
* transfer wants to send, adds POLLOUT
* http/2 filter has a flow control block, removes POLLOUT and adds
POLLIN (it is waiting on a WINDOW_UPDATE from the server)
* TLS filter is connected and changes nothing
* h2-proxy filter also has a flow control block on its tunnel stream,
removes POLLOUT and adds POLLIN also.
* socket filter is connected and changes nothing
* The resulting pollset is then mixed together with all other transfers
and their pollsets, just as before.
Use of `SEND_HOLD` is no longer necessary in the filters.
All filters are adapted for the changed method. The handling in
`multi.c` has been adjusted, but its state handling the the protocol
handlers' `getsocks` method are untouched.
The most affected filters are http/2, ngtcp2, quiche and h2-proxy. TLS
filters needed to be adjusted for the connecting handshake read/write
handling.
No noticeable difference in performance was detected in local scorecard
runs.
Closes#11833
- use shared code for setting up the CONNECT request
when tunneling, used in HTTP/1.x and HTTP/2 proxying
- eliminate use of Curl_buffer_send() and other manipulations
of `data->req` or `data->state.ulbuf`
Closes#11808
- use CLIENTWRITE_BODY *only* when data is actually body data
- add CLIENTWRITE_INFO for meta data that is *not* a HEADER
- debug assertions that BODY/INFO/HEADER is not used mixed
- move `data->set.include_header` check into Curl_client_write
so protocol handlers no longer have to care
- add special in FTP for `data->set.include_header` for historic,
backward compatible reasons
- move unpausing of client writes from easy.c to sendf.c, so that
code is in one place and can forward flags correctly
Closes#11885
Some of these changes come from comparing `Curl_http` and
`start_CONNECT`, which are similar, and adding things to them that are
present in one and missing in another.
The most important changes:
- In `start_CONNECT`, add a missing `hyper_clientconn_free` call on the
happy path.
- In `start_CONNECT`, add a missing `hyper_request_free` on the error
path.
- In `bodysend`, add a missing `hyper_body_free` on an early-exit path.
- In `bodysend`, remove an unnecessary `hyper_body_free` on a different
error path that would cause a double-free.
https://docs.rs/hyper/latest/hyper/ffi/fn.hyper_request_set_body.html
says of `hyper_request_set_body`: "This takes ownership of the
hyper_body *, you must not use it or free it after setting it on the
request." This is true even if `hyper_request_set_body` returns an
error; I confirmed this by looking at the hyper source code.
Other changes are minor but make things slightly nicer.
Closes#11745
