* src/fdtable.c, src/fdtable.h: Largely extend.
* src/wait-global.c, src/wait-private.c, src/wait-user.c: Remove and
move code to ...
* src/wait.c: here.
(_gpgme_fd_table_init, fd_table_put): Remove. Do not call them.
(_gpgme_add_io_cb, _gpgme_add_io_cb_user): Change to use the fdtable.
(_gpgme_remove_io_cb, _gpgme_remove_io_cb_user): Ditto.
(_gpgme_wait_global_event_cb): Ditto.
(gpgme_wait_ext, _gpgme_wait_on_condition): Ditto.
* src/wait.h (struct io_cb_tag_s): Add fields 'serial' and 'desc'.
Change 'idx' to 'fd'.
(struct fd_table): Remove.
* src/context.h (struct gpgme_context): Remoce 'fdt'. Rename io_cbs
to user_io_cbs for clarity.
* src/engine-gpgsm.c: Unify trace output.
(start): Pass a description along with the IO handlers.
* src/priv-io.h (struct io_select_fd_s): Rename to io_select_s.
(io_select_t): New.
* src/gpgme.c (_gpgme_cancel_with_err): Replace arg 'ctx' by 'serial'.
(gpgme_cancel): Adjust.
--
This is the second part of a larger refactoring of the wait/event
code. Does currently only work on Unix and with the private wait
functions (i.e. the async operations don't yet work).
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/engine-gpg.c (gpg_verify): Supply --verify for fixed gpg
versions.
--
With 2.1.something we print a warning if no explict command was
given. Since 2.1.16 this has been fixed to also allow the combination
of --verify and --output. Thus we now always use "--verify".
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/debug.c (_gpgme_debug_buffer): Switch between two output
formats.
--
The new format is much more practical than the bunch of hex digits
followed by just 16 ascii chars. To get the old behaviour use a debug
level of 10.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/context.h (struct gpgme_context): Add field 'next_ctx'.
* src/gpgme.c (def_lc_lock): Replace by ...
(context_list_lock): new.
(context_list): New variable.
(gpgme_new): Add new context object to the list.
(gpgme_release): Remove context object from the list.
(_gpgme_get_ctx): New function.
--
To allow mapping a context serial number back to a context object and
to check whether a serialno has still a context object, we need to
link them all together. We already take a lock to setup the locale
and thus the only overhead is in freeing the context.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/fdtable.c, src/fdtable.h: New.
* src/posix-io.c (notify_table_item_s): Remove.
(notify_table, notify_table_size, notify_table_lock): Remove.
(_gpgme_io_pipe): Put new fds into the table.
(_gpgme_io_dup): Ditto.
(_gpgme_io_close): Replace notify stuff by a call to the fdtable.
(_gpgme_io_set_close_notify): Remove. Change all callers to to use
_gpgme_fdtable_add_close_notify.
* src/Makefile.am (main_sources): Add new files.
--
This is the first part or a larger change to unify the tracking of
file descriptors. Right now this has only been implemented for
Posix and thus the code will not yet build for Windows.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/posix-io.c (_gpgme_is_fd_valid): New out-commented function.
--
This code is sometimes useful to track down invalid file descriptors.
We keep it commented in the code.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/wait.c (_gpgme_run_io_cb): Fix return code.
--
The function needs to return an gpg_error_t and not ERRNO.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/debug.c (_gpgme_debug): Take better care of NULL userinfo.
(_gpgme_debug_end): Rework.
(_trace_sysres): Print ERRNO and not the supplied RES.
--
The TRACE_SYSRES patch fixes
Regression-due-to: 7a1e7006d0
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/debug.c (_gpgme_debug): Add arg LINE. Chnage all callers.
(_gpgme_debug_begin): Remove.
* src/debug.h (TRACE_SEQ): Use the LINE arg of _gpgme_debug.
--
This includes chnages to always print fds in decimal as weel as
tweaking the TARCE_SEQ function to make use of the new machinery.
The standard 'tag' can now always be NULL and no tag information will
be printed.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/debug.c (debug_lock): Remove. Also remove all users.
(_gpgme_debug): Use gpgrt_bsprintf to prepare the output and finally
print using standard fprintf. Reformat to prefix to be narrower.
--
Note that the locks are now implicitly done using the systems stdio.
The threadid is now printed with 4 digits in hex and thus without the
angle brackets and the 0x. However it is still a hex number even if it
may look like an octal number. The hex letters are uppercase to make
searching in locks easier iff the threadid happens to have a letter in
it.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/cJSON.c: Include stdint.h.
(parse_number): Avoid overflob in SUBSCALE and cap integer values.
--
GnuPG-bug-id: 4330
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/engine-gpg.c (append_args_from_recipients_string): Detect bad
options.
* src/engine-gpgsm.c (set_recipients_from_string): Implement option
parsing.
--
The only option we actually implement is "--" but the code layout is
now very simlar to engine-gpg and can easily be extended if ever
needed.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/engine-gpgsm.c (gpgsm_encrypt): Fix argument check.
--
It is pretty obvious thar the string based new encrypt function has
never been tested for S/MIME. The fix was easy. A followup patch
will extend it to allow for keywords in the future.
GnuPG-bug-id: 4556
Fixes-commit: a1f76b3b54
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/engine-gpgsm.c [!USE_DESCRIPTOR_PASSING] (gpgsm_new): Remove
last call to _gpgme_io_set_close_notify.
--
It is called just after the code in question for all cases.
GnuPG-bug-id: 4456
Fixes-commit: dd21ec997c
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* src/w32-util.c (_gpgme_access): Respect mode parameter.
(_gpgme_create_process_utf8): Convert startupinfo, too.
--
This both did not show up in testing as we only use
mode F_OK and STARTUPINFOA is basically the same as
STARTUPINFOW.
Fixes commit: a82e3a0ae5
GnuPG-Bug-Id: T4453
* src/dirinfo.c (get_gpgconf_item): Use _gpgme_access.
* src/posix-util.c (_gpgme_access): Add forward to normal access.
* src/sys-util.h (_gpgme_access): New for posix and w32.
* src/w32-io.c (_gpgme_io_spawn): Use _gpgme_crate_process_utf8.
* src/w32-util.c (utf8_to_wchar, utf8_to_wchar0): The usual w32 conv.
(find_program_in_dir): Use _gpgme_access.
(find_program_at_standard_place): Use wchar API and convert to UTF-8.
(_gpgme_access): Convert UTF-8 to wchar and use wchar API.
(_gpgme_create_process_utf8): Convert UTF-8 to wchar and use wchar API.
--
While we should not say that we have full support for unicode path
installations of GnuPG, this ensures that GPGME works if GPGME
itself is installed in a unicode path. e.g.: Libreoffice supports
this.
GnuPG-Bug-Id: T4453
Based on a patch provided by Egor Pugin. Thanks.
* src/w32-io.c (_gpgme_io_spawn): Show MessageBox only once.
--
This prevents multiple message boxes from showing when
the GPGME installation is unworkable.
GnuPG-Bug-Id: T4453
* src/debug.c (_gpgme_debug): Use gpgrt_vasprintf instead of
vfprintf to have a more portable format.
--
This fixes crashes on Windows because "%zu" is used which
is not natively supported on Windows but which gpgrt supports.
* src/assuan-support.c (my_spawn): Fix using strtoul for
a long.
--
This was commented on in129def87b262 and is correct.
It is signed here to better handle cases where an
invalid handle value (-1) would be passed.
* src/assuan-support.c (my_spawn): Add hack to
mark the logger fd for w32spawn translation.
--
The w32 spawn code needs to modify argv with
an updated fd that matches the real id
in the spawned process.
It uses spawn_fd_item_s.arg_loc for that.
We hack it here so that the arg_loc is set
for gpgsm's logger-fd without changing
the assuan API.
GnuPG-Bug-Id: T4426
* src/engine-gpgsm.c (struct engine_gpgsm): Add fields diag_cb and
diagnostics.
(close_notify_handler): Close the diag fd on status fd close.
Handle diag close.
(gpgsm_cancel): Handle diag.
(gpgsm_release): Free DIAGNOSTICS.
(gpgsm_new): Support the diag feature.
(start): Set a handler for the diag fd.
(gpgsm_getauditlog): Support GPGME_AUDITLOG_DIAG.
--
Co-authored-by: Andre Heinecke <aheinecke@gnupg.org>
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/cJSON.c (MAX_DEPTH): New. Maximum recursion depth.
(parse_value, parse_array, parse_object): Carry and check
depth argument.
(cJSON_ParseWithOpts): Initialize depth.
--
This fixes a stack overflow if we get weird recursive
json data.
GnuPG-Bug-Id: T4331
* src/w32-util.c (_gpgme_get_gpgconf_path): Add NULL top strconcat.
--
Fortunately this is called early and the stack like cleared out so
that we have not seen wrong behaviour until now. We should really fix
all these annoying HANDLE/int cast warnings and alike so that real
bugs are not drowned by them.
GnuPG-bug-id: 4369
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/gpgme-json.c (subkey_to_json): here
--
Avoid Windows ugliness of allowing different CRTs in the same
process.
Fixes-commit: 7f24233719
Signed-off-by: Werner Koch <wk@gnupg.org>
* configure.ac: Add -Wno-format-truncation and
-Wno-sizeof-pointer-div.
* src/b64dec.c (_gpgme_b64dec_proc): Add fallthrough annotation.
* src/cJSON.c (parse_string): Ditto.
* src/gpgme-json.c (main): Ditto.
--
gcc 8 enables a couple of new warnings. Some of them are useless for
us. In particular:
util.h:42:26: warning: division 'sizeof (char *) / sizeof (char)'
does not compute the number of array elements [-Wsizeof-pointer-div]
#define DIM(v) (sizeof(v)/sizeof((v)[0])) ^
trustlist.c:101:22: note:
in expansion of macro 'DIM' if (strlen (p) == DIM(item->keyid) - 1)
Which is a real standard way to use DIM, here the right hand side is
equivalent to sizeof but nevertheless it is correct. Yes sir, we know
C.
The format string warnings I have seen were assuming that the time
structure returns valued out of scope - but if the system is that
broken, the s_n_printf catches this.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/genkey.c (gpgme_op_genkey_start): Fix check for
parms.
--
This fixes a regression introduced by:
174af15725
So that the error was triggered by usual args.
GnuPG-Bug-Id: T4265
* src/debug.h (TRACE_BEG, TRACE_LOG, TRACE_SUC): Use variadic macros
and remove the TRACE_BEG1 et al. Change all users to always pass a
format string.
(TRACE): Ditto.
* src/debug.c (_gpgme_debugf): New.
* configure.ac <GCC>: Add -Wno-format-zero-length.
--
This makes it easier for use to enable format checks. The zero-length
format is required to allow for an empty format due to the comman
problematic of __VA_ARGS__.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/gpgme.m4 (_AM_PATH_GPGME_CONFIG): Use gpgrt-config with gpgme.pc
when possible.
(AM_PATH_GPGME_GLIB): Likewise with gpgme-glib.pc.
--
Keeping AM_PATH_GPGME_PTHREAD, as is, untouched.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* src/gpgme-glib.pc.in: New.
* src/gpgme.pc.in (avail_lang): Remove.
--
Provide gpgme-glib.pc for gpgme-glib library. The avial_lang
information is no use because *.pc is for C.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* configure.ac: Generate src/gpgme.pc.
* src/Makefile.am (pkgconfigdir, pkgconfig_DATA): New.
(EXTRA_DIST): Add gpgme.pc.in.
* src/gpgme.pc.in: New.
* src/gpgme-config.in: Use variables.
--
Some usages of gpgme-config is not compatible to pkg-config style;
The --glib option and --thread option which affect the output
by --cflags or --libs are not supported by gpgme.pc.
gpgme-config's embedding information for gpg-error and libassuan at
the build time of gpgme is considered inflexible than pkg-config
style. It is now handled by dependency of gpgme.pc (Requires field).
To use gpgme.pc, newer libgpg-error (>= 1.33) and libassuan (>= 2.5.3)
are required, which provide gpg-error.pc and libassuan.pc respectively.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* src/w32-io.c (close_handle, _close_handle): New macro and function;
use in place of all CloseHandle calls.
(wait_for_single_object, _wait_for_single_object): Likewise.
Signed-off-by: Werner Koch <wk@gnupg.org>
