The changes in commit 20743e8 removed the s_max_virtual_consoles
constant and hardcoded the number of consoles to 4. But in
PS2KeyboardDevice the keyboard shortcuts for switching to consoles were
hardcoded to 6.
I reintroduced the constant and added it in both places.
For non-x86 targets, it's not very nice to define inline functions in
AK/Memory.h with asm volatile implementations. Guard this inline
assembly with ARCH(I386) and provide portable alternatives. Since we
always compile with optimizations, the hand-vectorized memset and
memcpy seem to be of dubious value, but we'll keep them here until
proven one way or another.
This should fix the Lagom build on native M1 macOS that was reported
on Discord the other day.
JPGLoader used to store component information in a HashTable, indexed
by the ID assigned by the JPEG file. This was fine for most purposes,
however after f89e8fb7 this was revealed to be a flawed implementation
which causes non-deterministic iteration over components.
This issue was previously masked by a perfect storm of int_hash being
stable for the integer values 0, 1 and 2; and AK::HashTable having just
the right amount of buckets for the components to be ordered correctly
after being hashed with int_hash. However, after f89e8fb7,
malloc_good_size was used for determining the amount of space for
allocation; this caused the ordering of the components to change, and
images started showing up with the red and blue channels reversed. The
issue was finally determined to be inconsistent ordering after randomly
changing the order of the components caused Huffman decoding to fail.
This was the result of about 10 hours of hair-pulling and repeatedly
doing full rebuilds due to bisecting between commits that touched AK.
Gunnar, I like you, but please don't make me go through this again. :^)
Credits to Andrew Kaster, bgianf, CxByte and Gunnar for the debugging
help.
The Context and Software Rasterizer now gets the array of texture units
instead of a single texture object. _Technically_, we now support some
primitive form of multi-texturing, though I'm not entirely sure how well
it will work in its current state.
When using BIND_NOW (e.g. via -Wl,-z,now) we would fail to load ELF
images while doing relocations when we encounter a weak symbol. Instead
we should just patch the PLT entry with a null pointer.
This can be reproduced with:
$ cat test.cpp
int main()
{
std::cout << "Hello World!" << std::endl;
}
$ g++ -o test -Wl,-z,now test.cpp
$ ./test
did not find symbol while doing relocations for library test: _ITM_RU1
Since I introduced this functionality there has been a steady stream of
people building with `ALL_THE_DEBUG_MACROS` and trying to boot the
system, and immediately hitting this assert. I have no idea why people
try to build with all the debugging enabled, but I'm tired of seeing the
bug reports about asserts we know are going to happen at this point.
So I'm hiding this value under the new ENABLE_ALL_DEBUG_FACILITIES flag
instead. This is only set by CI, and hopefully no-one will try to build
with this thing (It's documented as not recommended).
Fixes: #7527
There are a bunch of places like drivers which for all intense and
purposes can't really fail allocation during boot, and if they do
fail we should crash immediately.
This change adds `KString::must_create_uninitialized(..)` as well as
`KString::must_create(..)` for use during early boot initialization of
the Kernel. They enforce that they are only used during early boot.
Previously the StringBuilder class would use memcpy() to write
directly into the ByteBuffer's buffer. Instead we should use the
append() method which ensures we don't overrun the buffer.
This allows us to mark the slow part (i.e. where we copy the buffer) as
NEVER_INLINE because this should almost never get called and therefore
should also not get inlined into callers.
Previously ByteBuffer::grow() behaved like Vector<T>::resize().
However the function name was somewhat ambiguous - and so this patch
updates ByteBuffer to behave more like Vector<T> by replacing grow()
with resize() and adding an ensure_capacity() method.
This also lets the user change the buffer's capacity without affecting
the size which was not previously possible.
Additionally this patch makes the capacity() method public (again).
There are lots of people who have issues building serenity because
they don't read the build directions closely enough and have an
unsupported GCC version as their host compiler. Instead of repeatedly
having to answer these kinds of questions, lets just error out upfront.
Previously, we would go crazy and shift things way out of bounds.
Add tests to verify that the decoding algorithm is safe around the
limits of the result type.
Avoid promotion of static strings to AK::String, instead use
AK::StringView and operator ""sv, to force string view's instead
which avoids allocation of String. This code path isn't hot enough
that it makes a huge difference, but every bit counts.
The current CommandLine API unfortunately allocates Strings just to
query the presence of arguments on the command line. Switch the API
to use StringView instead to reduce the number of String allocations.
Since this program is setuid-root, it should be as simple as possible.
To that end, remove `/etc/plsusers` and use filesystem permissions to
achieve the same thing. `/bin/pls` is now only executable by `root` or
members of the `wheel` group.
Also remove all the logic that went to great lengths to `unveil()` a
minimal set of filesystem paths that may be used for the command.
The complexity-to-benefit ratio did not seem justified, and I think
we're better off keeping this simple.
Finally, remove pledge promises the moment they are no longer needed.
This only affects malformed RSA keys. Instead of accepting and
continuing with potentially broken pointers (and in ASAN, crashing), we
now consider bitmaps malformed, and stop parsing.
Found by OSS Fuzz: #31698, long-standing-bug:
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=31698
Fun fact: The "if" only exists because of OSS Fuzz.
8cc279ed74
printf didn't check whether the additional integer variable belongs to
the field width specifier or to the precision specifier, and always
applied it to the field width instead.
Implement the case distinction that we already use in literal width
and precision specifiers for the variable version as well so that
they are correctly attributed.
POSIX (`errno(3p)`) states that errno should not be set to zero.
This helps with applications that don't expect errno to get updated
unless an intermediate syscall also fails.
There were a few cases where we could end up logging profiling events
before or after the associated process or thread exists in the profile:
After enabling profiling we might end up with CPU samples before we
had a chance to synthesize process/thread creation events.
After a thread exits we would still log associated kmalloc/kfree
events. Instead we now just ignore those events.
Previously there was no way to output an empty value into the shadow
file entries when the spwd members were disabled. This would cause new
user entries to the shadow file to be cluttered with disabled values.
This commit checks if the spwd member value is diabled (-1) and will
output as appropriate.
These dbgln's caused excessive load in the WebServer process,
accounting for ~67% of the processing time when serving a webpage
with a bunch of resources like serenityos.org/happy/2nd/.
