This adds three categories:
- qt.core.library
- qt.core.plugin.loader
- qt.core.plugin.factoryloader
plus the "qt.core.plugin.elfparser" category (only available in
developer builds).
All three use the new Q_LOGGING_CATEGORY_WITH_ENV_OVERRIDE technique
which enables their debugging if the QT_DEBUG_PLUGINS=1 variable is set.
As a consequence, some warnings were downgraded to debug messages. I've
only left as warnings situations where a real problem occurred, since
they now get printed by default: failures to open or mmap a file,
corrupt plugin metadata (but not scan of non-plugins), use of
QPluginLoader with a static build of Qt.
Drive-by update of some messages to make them prettier and/or use
qUtf16Printable().
Change-Id: I3eb1bd30e0124f89a052fffd16a752acfe89c19e
Reviewed-by: Kai Koehne <kai.koehne@qt.io>
Since QJsonValue and QCborValue use the same backend, we may as well use
the CBOR frontend classes, which means we avoid an unnecessary
conversion until later.
Change-Id: I2de1b4dfacd443148279fffd16a3e2f56cd74c0b
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
For MSVC it's clear that the plugin and Qt must match, since they would
be linking to different runtime assemblies otherwise. For all other
systems, including MinGW on Windows, there's no such thing.
But we insist on MinGW debug-and-release builds matching.
Change-Id: I3eb1bd30e0124f89a052fffd16a6aa52c7f8b9c0
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
They return a pointer to the actual header, skipping the magic string.
This is done in preparation for the header located in an ELF note, which
won't have the magic.
Change-Id: I3eb1bd30e0124f89a052fffd16a8229bec2ad588
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This rewrite uses the actual structures supplied by the system's C
library, so it should be easier to read. It removes hardcoded constants
with little evident meaning in favor of sizeof() and the macros from
that header. It also removes advancing the data pointer in favor of
having absolute offsets.
The resulting implementation is stricter than the original, checking
more fields in the header. Because the QPluginLoader and QFactoryLoader
users may make decisions based on availability of plugins before
attempting to load them, it's better to be stricter here than to fail
later when trying to dlopen() them.
Debugging and testing are much improved. Instead of stored artifacts, I
added a routine to modify a valid plugin to make it invalid, given the
conditions we've found so far.
If you turn debugging on for this category, you'll see things like:
not-elf.fcqdMq.so : Not an ELF file (invalid signature)
wrong-word-size.QrnSAx.so : ELF 32-bit LSB (GNU/Linux), version 1, shared library or PIC executable, x86-64
invalid-word-size.bOkXvp.so : Invalid ELF file (class 0), LSB (GNU/Linux)
unknown-word-size.ogYKeF.so : Invalid ELF file (class 66), LSB (GNU/Linux)
wrong-endian.owiElX.so : ELF 64-bit MSB (GNU/Linux), version 1, shared library or PIC executable, x86-64
invalid-endian.FRxClR.so : ELF 64-bit invalid endianness (0) (GNU/Linux)
unknown-endian.FfvRrP.so : ELF 64-bit invalid endianness (65) (GNU/Linux)
elf-version-0.gPTdpQ.so : ELF 64-bit LSB (GNU/Linux), file version 0
elf-version-2.jlIUUg.so : ELF 64-bit LSB (GNU/Linux), file version 2
executable.LlXiFp.so : ELF 64-bit LSB (GNU/Linux), version 1, executable, x86-64
relocatable.UsOYuy.so : ELF 64-bit LSB (GNU/Linux), version 1, relocatable, x86-64
core-file.hqvNRz.so : ELF 64-bit LSB (GNU/Linux), version 1, core dump, x86-64
invalid-type.CIJgfS.so : ELF 64-bit LSB (GNU/Linux), version 1, unknown type 259, x86-64
wrong-arch.UcNmgz.so : ELF 64-bit LSB (GNU/Linux), version 1, shared library or PIC executable, AArch64
file-version-0.lZYuda.so : ELF 64-bit LSB (GNU/Linux), version 0, shared library or PIC executable, x86-64
file-version-2.ucfdwL.so : ELF 64-bit LSB (GNU/Linux), version 2, shared library or PIC executable, x86-64
no-sections.rSjsHh.so : ELF 64-bit LSB (GNU/Linux), version 1, shared library or PIC executable, x86-64
no-sections.rSjsHh.so : contains 0 sections of 64 bytes at offset 0 ; section header string table (shstrtab) is entry 0
no-sections.rSjsHh.so : no section table present, not able to find Qt metadata
qtmetadata-executable.vrxcIf.so : ELF 64-bit LSB (GNU/Linux), version 1, shared library or PIC executable, x86-64
qtmetadata-executable.vrxcIf.so : contains 42 sections of 64 bytes at offset 997256 ; section header string table (shstrtab) is entry 41
qtmetadata-executable.vrxcIf.so : shstrtab section is located at offset 996831 size 423
qtmetadata-executable.vrxcIf.so : section 0 name "" type NULL flags X offset 0x0 size 0x0
qtmetadata-executable.vrxcIf.so : section 1 name ".note.gnu.property" type NOTE flags AX offset 0x2a8 size 0x30
qtmetadata-executable.vrxcIf.so : section 2 name ".note.gnu.build-id" type NOTE flags AX offset 0x2d8 size 0x24
qtmetadata-executable.vrxcIf.so : section 3 name ".hash" type HASH flags AX offset 0x300 size 0x44c
qtmetadata-executable.vrxcIf.so : section 4 name ".gnu.hash" type 0x6ffffff6 flags AX offset 0x750 size 0x3b8
qtmetadata-executable.vrxcIf.so : section 5 name ".dynsym" type DYNSYM flags AX offset 0xb08 size 0xd50
qtmetadata-executable.vrxcIf.so : section 6 name ".dynstr" type STRTAB flags AX offset 0x1858 size 0x15d8
qtmetadata-executable.vrxcIf.so : section 7 name ".gnu.version" type 0x6fffffff flags AX offset 0x2e30 size 0x11c
qtmetadata-executable.vrxcIf.so : section 8 name ".gnu.version_r" type 0x6ffffffe flags AX offset 0x2f50 size 0xb0
qtmetadata-executable.vrxcIf.so : section 9 name ".rela.dyn" type RELA flags AX offset 0x3000 size 0x480
qtmetadata-executable.vrxcIf.so : section 10 name ".rela.plt" type RELA flags AX offset 0x3480 size 0x7e0
qtmetadata-executable.vrxcIf.so : section 11 name ".init" type PROGBITS flags AX offset 0x4000 size 0x1b
qtmetadata-executable.vrxcIf.so : section 12 name ".plt" type PROGBITS flags AX offset 0x4020 size 0x550
qtmetadata-executable.vrxcIf.so : section 13 name ".plt.got" type PROGBITS flags AX offset 0x4570 size 0x8
qtmetadata-executable.vrxcIf.so : section 14 name ".text" type PROGBITS flags AX offset 0x4580 size 0x110e
qtmetadata-executable.vrxcIf.so : section 15 name ".fini" type PROGBITS flags AX offset 0x5690 size 0xd
qtmetadata-executable.vrxcIf.so : section 16 name ".rodata" type PROGBITS flags AX offset 0x6000 size 0x473
qtmetadata-executable.vrxcIf.so : section 17 name ".qtversion" type PROGBITS flags AX offset 0x6478 size 0x10
qtmetadata-executable.vrxcIf.so : section 18 name ".qtmetadata" type PROGBITS flags AX offset 0x64a0 size 0x19b
qtmetadata-executable.vrxcIf.so : found .qtmetadata section
qtmetadata-writable.stzwrk.so : ELF 64-bit LSB (GNU/Linux), version 1, shared library or PIC executable, x86-64
Change-Id: I42eb903a916645db9900fffd16a4437af9728eea
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This is done in preparation for storing the metadata without the magic
string in static plugins and in ELF notes.
Change-Id: I3eb1bd30e0124f89a052fffd16a820454dd56d3e
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This is the already-loaded counterpart to the file scanning code
Change-Id: I3eb1bd30e0124f89a052fffd16a81f518fa95f0d
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Those two scanners always return exact results, if the metadata is
present, so we don't need to re-scan (haven't needed since Qt
5.0). Especially since we scan from the end, we were spending cycles
doing unnecessary work.
Change-Id: I42eb903a916645db9900fffd16a4ccfdc7342278
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Both functions took a QString for the input file name, but while the ELF
parser had an optional QLibrary pointer (which was never null) where to
store the error string, the Mach-O parser received a pointer to a
QString. So make both of them take a single in/out QString pointer,
which has the file name on input and is cheap for us because of COW.
Drive-by fix the name of the static function in qmachparser.cpp from
"ns" (which stood for "not suitable") to "notfound".
Change-Id: I3eb1bd30e0124f89a052fffd16a8182f4f8541c3
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The reason it scanned backwards was valid, but unfortunately that
doesn't work if the metadata magic appears in the debug info that comes
after the metadata.
Task-number: QTBUG-67461
Change-Id: I2bbf422288924c198645fffd16a9276767ca2589
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The compiler ought to figure them out, but we can just help it.
Change-Id: I42eb903a916645db9900fffd16a4981ac966f982
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
The multi-state return code was a legacy of how Arvid wrote the ELF
parser code back in the day, the fact that it scanned for two different
types of plugins in Qt 4 and that the metadata could exist in different
places. None of that matters nowadays: who cares if the file is a
corrupt binary, not a valid binary, does not have the right
architecture, or has no suitable section? It's not a plugin, period.
The Qt 4 plugin mechanism was removed for Qt 5.0 in commit
7443895857 ("Remove support for Qt 4 style
plugins").
Change-Id: I42eb903a916645db9900fffd16a442d800399b98
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
If we can't mmap(), then libdl won't be able to either.
Pick-to: 6.2
Change-Id: I42eb903a916645db9900fffd16a492a1ac25903f
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
A comma operator results in a discarded-value expression, meaning the
compiler is going to complain the moment we mark QMutexLocker as
nodiscard. Turn the comma into a functionally equivalent IILE.
Change-Id: I33826902c8471016490aac25160b70c609dafd90
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
For ELF platforms with a GCC-compatible compiler, the QElfParser has
been ignoring the .rodata section ever since Qt 5.0 commit
7443895857 ("Remove support for Qt 4 style
plugins"). That change removed handling of return value
QElfParser::NoQtSection from the ELF parser, which meant that the
plugins without a .qtmetadata section were never considered plugins. In
other words, for those systems, the __attribute__ macro is mandatory.
For systems with a GCC-incompatible compiler, there were only two in Qt
5.x's lifetime: Oracle/Sun Solaris with SunCC and IBM AIX with IBM xlC
compiler. Neither compiler supports C++17 according to [1], so they
can't be in use in Qt 6. IBM xlC now comes with a Clang-based front-end
for users who need Qt 6 and the OpenIndiana distribution of Open Source
Solaris has been using GCC since 2012.
So make the macros mandatory from now on.
[1] https://en.cppreference.com/w/cpp/compiler_support#C.2B.2B17_features
Change-Id: I42eb903a916645db9900fffd16a443745446cc64
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
An enum shows the proper values in the debugger...
Pick-to: 6.2
Change-Id: I2de1b4dfacd443148279fffd16a3a5848196983c
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Rewrite the whole function for readability _and_ performance:
- Extract Method isValidSuffix() as a lambda
- Extract Method isNumeric() as a lambda
- Use a C array of QLatin1String instead of a QStringList to hold the
statically-sized list of candidate suffixes. This has the nice
side-effect that the compiler will now throw an error in case the
#ifdef'ery yields zero candidates (C arrays cannot have no
elements), e.g. when porting to a new platform.
- Last, not least, replace the parsing with a loop that makes clear
what's going on and which is forward-iteration-only-compatible, so
we can use QStringTokenizer directly, without toContainer() to get a
random-access sequence.
Need to use the C++20 version of all_of(), since QStringTokenizer uses
sentinels instead of end-iterators.
Even though we use higher-level constructs now, the code is still more
efficient than the index-twisting we had before.
Change-Id: I9f3faf3e30f58c9eb8a1487a7ca190681e87767b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
The variable `i` is initially `suffixPos + 1` and is then incremented
further. It therefore can never be equal to `suffixPos` (ints don't
overflow, that would be UB, and suffixPos doesn't change its value),
so don't check for that.
Change-Id: I3870ddf6ee550cad6c24fececf2a0b662a33d750
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Drag the cheap check for 'fileName.endsWith(".dylib")' to before the
expensive operations (QFileInfo creation, extraction and splitting of
completeSuffix(), building of suffix candidate list, ...), at the cost
of a duplicated Q_OS_DARWIN check.
Change-Id: I8ed764f18f0beb8ad24c30ab806ecc5452159601
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Either be explicit, or just use QFlags directly.
Change-Id: I18cbea4eaa1a0a4bce7665b735e7d785f7a196b2
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This makes the size nicely aligned and conveniently intializes
everything to the right values.
Change-Id: Ibad2defbbd323fd5cdd4bed8374f40558d80acc0
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Mostly related to qstrlen().
Change-Id: I69e2052c83766e4fc466ed398d0d0eac011a77ec
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
We already manage to compile without this code, and none of it are full
classes or separate functions suitable for qt5compat.
Change-Id: I47facac7ec621cfc4b0b26214b7de37897443519
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Omitting state machine and docs for now.
Task-number: QTBUG-84469
Change-Id: Ibfa5e7035515773461f6cdbff35299315ef65737
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
Conflicts:
examples/opengl/doc/src/cube.qdoc
src/corelib/global/qlibraryinfo.cpp
src/corelib/text/qbytearray_p.h
src/corelib/text/qlocale_data_p.h
src/corelib/time/qhijricalendar_data_p.h
src/corelib/time/qjalalicalendar_data_p.h
src/corelib/time/qromancalendar_data_p.h
src/network/ssl/qsslcertificate.h
src/widgets/doc/src/graphicsview.qdoc
src/widgets/widgets/qcombobox.cpp
src/widgets/widgets/qcombobox.h
tests/auto/corelib/tools/qscopeguard/tst_qscopeguard.cpp
tests/auto/widgets/widgets/qcombobox/tst_qcombobox.cpp
tests/benchmarks/corelib/io/qdiriterator/qdiriterator.pro
tests/manual/diaglib/debugproxystyle.cpp
tests/manual/diaglib/qwidgetdump.cpp
tests/manual/diaglib/qwindowdump.cpp
tests/manual/diaglib/textdump.cpp
util/locale_database/cldr2qlocalexml.py
util/locale_database/qlocalexml.py
util/locale_database/qlocalexml2cpp.py
Resolution of util/locale_database/ are based on:
https://codereview.qt-project.org/c/qt/qtbase/+/294250
and src/corelib/{text,time}/*_data_p.h were then regenerated by
running those scripts.
Updated CMakeLists.txt in each of
tests/auto/corelib/serialization/qcborstreamreader/
tests/auto/corelib/serialization/qcborvalue/
tests/auto/gui/kernel/
and generated new ones in each of
tests/auto/gui/kernel/qaddpostroutine/
tests/auto/gui/kernel/qhighdpiscaling/
tests/libfuzzer/corelib/text/qregularexpression/optimize/
tests/libfuzzer/gui/painting/qcolorspace/fromiccprofile/
tests/libfuzzer/gui/text/qtextdocument/sethtml/
tests/libfuzzer/gui/text/qtextdocument/setmarkdown/
tests/libfuzzer/gui/text/qtextlayout/beginlayout/
by running util/cmake/pro2cmake.py on their changed .pro files.
Changed target name in
tests/auto/gui/kernel/qaction/qaction.pro
tests/auto/gui/kernel/qaction/qactiongroup.pro
tests/auto/gui/kernel/qshortcut/qshortcut.pro
to ensure unique target names for CMake
Changed tst_QComboBox::currentIndex to not test the
currentIndexChanged(QString), as that one does not exist in Qt 6
anymore.
Change-Id: I9a85705484855ae1dc874a81f49d27a50b0dcff7
Commit ae6f73e856 inserted a mutex around
the entire load_sys(). We had reasoed that deadlocks would only occur if
the object creation in instance() recursed into its own instance(),
which was already a bug. But we had forgotten that dlopen()/
LoadLibrary() executes initialization code from the module being loaded,
which could cause a recursion back into the same QPluginLoader or
QLibrary object. This recursion is benign because the module *is* loaded
and dlopen()/LoadLibrary() returns the same handle.
[ChangeLog][QtCore][QLibrary and QPluginLoader] Fixed a deadlock that
would happen if the plugin or library being loaded has load-time
initialization code (C++ global variables) that recursed back into the
same QLibrary or QPluginLoader object.
PS: QLibraryPrivate::loadPlugin() updates pluginState outside a mutex
lock, so pluginState should be made an atomic variable. Once that is
done, we'll only need locking the mutex to update errorString (no
locking before loading).
Fixes: QTBUG-83207
Task-number: QTBUG-39642
Change-Id: Ibdc95e9af7bd456a94ecfffd160209304e5ab2eb
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: David Faure <david.faure@kdab.com>
And make pHnd atomic.
The majority of the variables is updated in QLibraryPrivate::load_sys
and updatePluginState(), which get the mutex protection.
QLibraryPrivate::unload_sys() doesn't need a mutex protection because we
have the refcounting.
[ChangeLog][QtCore][QLibrary & QPluginLoader] Fixed a number of race
conditions caused by having two QLibrary objects pointing to the same
library being operated in different threads.
Fixes: QTBUG-39642
Change-Id: I46bf1f65e8db46afbde5fffd15e1a5b3f5e74ea4
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
There was a race condition in accessing the cached instance factory
member, so rework loadPlugin() to return the cached or newly discovered
instance, with proper, atomic caching. Because I had to change that, I
took the opportunity to fix the QFactoryLoader code that calls
loadPlugin().
Note that QLibraryPrivate::loadPlugin() returns non-nullptr now if the
instance is known, which means the last return in QPluginLoader::load()
will convert to true, not false, if the instance got cached between the
earlier check and the call to loadPlugin(). That's probably what was
intended.
Task-number: QTBUG-39642
Change-Id: I46bf1f65e8db46afbde5fffd15e1a42d2b6cbf2c
Reviewed-by: David Faure <david.faure@kdab.com>
Move away from using 0 as pointer literal.
Done using clang-tidy. This is not complete as
run-clang-tidy can't handle all of qtbase in one go.
Change-Id: I1076a21f32aac0dab078af6f175f7508145eece0
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
In the case when the Qt build is debug_and_release the debug plugin
check should be enabled, otherwise the mixture of Qt debug and Qt
release libraries will cause crashes (e.g. QTBUG-77431)
Task-number: QTBUG-78445
Change-Id: Ice0b03e63ddad893334a0e1a4ede1f6ace83007b
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
This patch caused an error when loading applications built in
debug mode with MinGW and thus has to be reverted.
This reverts commit bba44746f9.
Fixes: QTBUG-77431
Change-Id: I3134878a742b304d10176cc8b0ed5ce06d4de53f
Reviewed-by: Jani Heikkinen <jani.heikkinen@qt.io>
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
Semi-automated, just needed ~20 manual fixes:
$ find \( -iname \*.cpp -or -iname \*.h \) -exec perl -pe 's/(\.|->)load\(\)/$1loadRelaxed\(\)/g' -i \{\} +
$ find \( -iname \*.cpp -or -iname \*.h \) -exec perl -pe 's/(\.|->)store\(/$1storeRelaxed\(/g' -i \{\} +
It can be easily improved (e.g. for store check that there are no commas
after the opening parens). The most common offender is QLibrary::load,
and some code using std::atomic directly.
Change-Id: I07c38a3c8ed32c924ef4999e85c7e45cf48f0f6c
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
... instead of qPrintable(), %s, and explicit qt_error_string().
Saves 2KiB in text size on optimized Linux AMD64 GCC 9.1 builds.
Change-Id: I98b6717da1ed1b678f01167d704a96f10da47966
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
MinGW doesn't have a debug and a release version of the CRT like
Visual C++ does. Disabling the check would allow distribution
only of a Release build of Qt.
Change-Id: Iecfa753217af96ca74091cd1d47400632629abdb
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
Thiago Macieira
Giuseppe D'Angelo
Tor Arne Vestbø
Marc Mutz
Zhang Sheng
Allan Sandfeld Jensen
Ulf Hermann
Friedemann Kleint
Jarek Kobus
Lars Knoll
Qt Forward Merge Bot
Simon Hausmann
Cristian Adam
Oliver Wolff
Alexandru Croitor