New connection filter `cf-dns` that manages DNS queries. If hands
out addresses and HTTPS-RR records to anyone interested. Used by
HTTPS and IP happy eyeballing.
Information may become available *before* the libcurl "dns entry"
is complete, e.g. all queries have been answered. The cf-ip-happy
filter uses this information to start connection attempts as soon
as the first address is available.
The multi MSTATE_RESOLVING was removed. A new connection always
goes to MSTATE_CONNECTING. The connectdata bit `dns_resolved`
indicates when DNS information is complete. This is used for
error reporting and starting the progress meter.
Removed dns entries `data->state.dns[i]`, as the `cf-dns` filter
now keeps the reference now.
Many minor tweaks for making this work and pass address information
around safely.
Closes#21027
Use a thread queue and pool for asnyc threaded DNS resolves.
Add pytest test_21_* for verification.
Add `CURLMOPT_RESOLVE_THREADS_MAX` to allow applications to
resize the thread pool used.
Add `CURLMOPT_QUICK_EXIT` to allow applications to skip thread
joins when cleaning up a multi handle. Multi handles in
`curl_easy_perform()` inherit this from `CURLOPT_QUICK_EXIT`.
Add several debug environment variables for testing.
Closes#20936
with `-DCURL_DISABLE_SOCKETPAIR=ON`.
```
lib/asyn-thrdd.c:227:5: error: Value stored to 'do_abort' is never read [clang-analyzer-deadcode.DeadStores,-warnings-as-errors]
227 | do_abort = addr_ctx->do_abort;
| ^ ~~~~~~~~~~~~~~~~~~
```
Closes#21061
Replace the `volatile int dirty` with a reference counter
protected by a mutex when available.
Solve the problem of when to call application's lock function
by adding a volatile flag that indicates a share has been added
to easy handles in its lifetime. That flag ever goes from
FALSE to TRUE, so volatile might work (in the absence of a mutex).
(The problem is that the lock/unlock functions need 2-3
`curl_share_setopt()` invocations to become usable and there
is no way of telling if the third will ever happen. Calling
the lock function before the 3rd setopt may crash the
application.)
When removing a share from an easy handle (or replacing it with
another share), detach the easy connection on a share with a
connection pool.
When cleaning up a share, allow this even if it is still used in
easy handles. It will be destroyed when the reference count
drops to 0.
Closes#20870
- Rename `Curl_resolv_unlink()` to `Curl_dns_entry_unlink()`.
- Change `Curl_dnscache_get()` to return CURLcode result. Returns
now `CURLE_COULDNT_RESOLVE_HOST` for "negative" cache entries.
- Add `Curl_dnscache_add_negative()` to put a "negative" entry
into the cache.
Closes#20864
It's mostly a filler word. I've read through each use of it in the code
base and did minor rephrasings when "simply" carried some meaning. The
overwhelming majority of cases, removing it improved the text
significantly. Inspired by #20793.
Closes#20822
When creating a dns entry, the addrinfo is passed into the entry on
success and needed deallocation by the caller on failure.
Change the signature to have Curl_dnscache_mk_entry() *always* take
ownership of the addrinfo, even on failure. Change parameter to address
of pointer so that call always clears it.
This makes the handling of failures to Curl_dnscache_mk_entry() simpler.
Fixes#20465Closes#20468
Add casts to `bool`, or use `bit` type in local variables, where
neccessary to avoid MSVC compiler warnings C4242.
Note: There may remain places needing the above updates, where not
tested in CI, and missed in manual review.
Also:
- urldata: convert struct field `connect_only` to bitfield to match its
counterpart in another struct.
- rename curl-specific `bit` type to `curl_bit`.
Closes#20142
Declutter the ifdefs in socketpair.h. Introduce Curl_wakeup_*()
function that encapsulate the details about how the socketpair
is implemented.
This moves the EVENTFD specials from the using code into socketpair
implemenatation, avoiding duplications in three places.
Closes#20340
- asyn-thrdd.c: scope an include.
- apply more clang-format suggestions.
- tidy-up PP guard comments.
- delete empty line from the top of headers.
- add empty line after `curl_setup.h` include where missing.
- fix indent.
- CODE_STYLE.md: add `strcpy`.
Follow-up to 8636ad55df#20088
- lib1901.c: drop unnecessary line.
Follow-up to 436e67f65b#20076Closes#20070
- replace `sendf.h` with `curl_trc.h` where it was included just for it.
- drop unused `curl_trc.h` includes.
- easy: delete obsolete comment about `send.h` include reason.
Also:
- move out `curl_trc.h` include from `sendf.h` and include it directly
in users, where not done already. To flatten the include tree and
to less rely on indirect includes.
- stop including `sendf.h` from other headers, replace it with forward
declaration of `Curl_easy`, as done already elsewhere.
Verified with an all non-unity CI run.
Closes#20061
Always use curlx_now() when calling Curl_pgrs_now(data). Tests with the
"manual" updates to now proved differ more then 100ms in parallel testing.
Add `curlx_nowp()` to set current time into a struct curltime.
Add `curlx_ptimediff_ms() and friends, passing pointers.
Update documentation.
Closes#19998
Use `data->progress.now` as the timestamp of proecssing a transfer.
Update it on significant events and refrain from calling `curlx_now()`
in many places.
The problem this addresses is
a) calling curlx_now() has costs, depending on platform. Calling it
every time results in 25% increase `./runtest` duration on macOS.
b) we used to pass a `struct curltime *` around to save on calls, but
when some method directly use `curx_now()` and some use the passed
pointer, the transfer experienes non-linear time. This results in
timeline checks to report events in the wrong order.
By keeping a timestamp in the easy handle and updating it there, no
longer invoking `curlx_now()` in the "lower" methods, the transfer
can observer a steady clock progression.
Add documentation in docs/internals/TIME-KEEPING.md
Reported-by: Viktor Szakats
Fixes#19935Closes#19961
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
- Curl_async_getaddrinfo() always returned NULL so it was pointless.
Return proper curlcode instead to distinguish between errors. Same for
Curl_doh().
- simplify the IP address handling
- make Curl_str2addr() function return CURLcode
Closes#19669
Windows CRTs have a `share.h`. Before this patch when trying to
`#include <share.h>` it, the compiler picked up curl's internal
`lib/share.h` instead. Rename it to avoid this issue.
CRT `share.h` has constants necessary for using safe open CRT functions.
Also rename `lib/share.c` to keep matching the header.
Ref: https://learn.microsoft.com/cpp/c-runtime-library/sharing-constants
Ref: 625f2c1644#16949#16991
Cherry-picked from #19643Closes#19676
Windows CE support was limited to successful builds with ming32ce
(a toolchain that hasn't seen an update since 2009, using an ancient gcc
version and "old mingw"-style SDK headers, that curl deprecated earlier).
Builds with MSVC were broken for a long time. mingw32ce builds were never
actually tested and runtime and unlikely to work due to missing stubs.
Windows CE toolchains also miss to comply with C89. Paired with lack of
demand and support for the platform, curl deprecated it earlier.
This patch removes support from the codebase to ease maintaining Windows
codepaths.
Follow-up to f98c0ba834#17924
Follow-up to 8491e6574c#17379
Follow-up to 2a292c3984#15975Closes#17927
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
When the target host is on a different port than 443, the name
"_[port]._https.[name]" shall be used.
Fixes#19301
Reported-by: Gunni on github
Closes#19324
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
Remove use of pthread_cancel in asnyc threaded resolving. While there
are system where this works, others might leak to resource leakage
(memory, files, etc.). The popular nsswitch is one example where resolve
code can be dragged in that is not prepared.
The overall promise and mechanism of pthread_cancel() is just too
brittle and the historcal design of getaddrinfo() continues to haunt us.
Fixes#18532
Reported-by: Javier Blazquez
Closes#18540
- use wakeup sockets non-locked.
- send wakeup notify only in normal control flow (not cancel). close
wakeup sockets in unlink only.
- remove 5ms thread lifetime wait crutch before pthread_cancel().
Closes#18380
- cancel thread waits until thread start is at least 5ms in the past
to give it some time to get its cancellation setup in place
- cancel thread without holding the mutex. It's supposed to be an
async operation, but better be safe
- set DEFERRED cancel state explicitly, should be default in a pthread,
but better be safe
Closes#18350
Add optional detail to `Curl_resolver_error()` to add to failure message
where available. This makes, for c-ares, the reason for a failed
resource available to the user without extra trace config.
When "dns" tracing enabled, print the c-ares server config at the start
of a resolve.
Closes#18251
Add a flag `thrd_don` to assess if the resolving thread has finished and
only destroy the context when *both* ref_count reaches 0 and thrd_done
is true.
Closes#18345
When the thread started is too fast, the signal will come before the
wait. Add an additional check before the wait to catch the thread
having started or already ended.
Closes#18344
Changed strategy to start up and terminate resolver thread.
When starting up:
Start the thread with mutex acquired, wait for signal from thread that
it started and has incremented the ref counter. Thread set
pthread_cancel() to disabled before that and only enables cancelling
during resolving itself. This assure that the ref counter is correct and
the unlinking of the resolve context always happens.
When shutting down resolving:
If ref counting shows thread has finished, join it, free everything. If
thread has not finished, try pthread_cancel() (non Windows), but keep
the thread handle around.
When destroying resolving:
Shutdown first, then, if the thread is still there and 'quick_exit' is
not set, join it and free everything. This might occur a delay if
getaddrinfo() hangs and cannot be interrupted by pthread_cancel().
Destroying resolving happens when another resolve is started on an
easy handle or when the easy handle is closed.
Add test795 to check that connect timeout triggers correctly
when resolving is delayed. Add debug env var `CURL_DNS_DELAY_MS`
to simulate delays in resolving.
Fix test1557 to set `quick_exit` and use `xxx.invalid` as domain
instead of `nothing` that was leading to hangers in CI.
Closes#18263
- remove a superfluous if()
- edit the #ifdef to only do the timeout checks when socketpair support
is missing, as that code flow could not be reached otherways anyway
CodeSonar pointed out possible issue that led me to see this.
Closes#18306
* Add more tracing information to c-ares errors.
* remove CURL_ASYNC_SUCCESS, rename `ares->last_status` to
`ares->ares_status`. Give trace explanation for "common"
errors
* add ares "csv" information to tracing on failure
* DoH: invoke `Curl_resolver_error()` on failure to populate
error buf
Closes#18247
Hold them for half the normal lifetime. Helps when told to transfer N
URLs in quick succession that all use the same non-resolving hostname.
Done by storing a DNS entry with a NULL pointer for 'addr'.
Previously an attempt was made in #12406 by Björn Stenberg that was
ultimately never merged.
Closes#18157
`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
Replace repeat `#ifdef` code with a macro for the return type of
the thread function.
Also:
- always define `CURL_STDCALL`, allowing to use it without guards.
- lib1307: drop single-use macro `CAINFO`.
Closes#17889
