When I added this code in 1472f6d, I forgot to add tests for it. That's
why I didn't realize that the values were appended to the wrong
FormatBuilder object, so an empty string was returned instead of the
expected "nan"/"inf". This made debugging some FPU issues with the
ScummVM port significantly more difficult.
For some algorithms, such as bloom filters, it's possible to reuse a
hash function (rather than having different hashing functions) if the
seed is different each time the hash function is used.
Modify AK::string_hash() to take a seed parameter, which defaults to 0
(the value the hash value was originally initialized to).
In the long-term, we should probably have a way to signal decoding
failure. For now, it should suffice to at least not crash. This is
particularly relevant because apparently this can be triggered while
parsing a PEM certificate, which happens during every TLS connection.
Found by OSS Fuzz
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=38979
In particular, we implicitly required that the caller initializes the
returned instances themselves (solved by making
UniformBumpAllocator::allocate call the constructor), and BumpAllocator
itself cannot handle classes that are not trivially deconstructible
(solved by deleting the method).
Co-authored-by: Ali Mohammad Pur <ali.mpfard@gmail.com>
This fixes two bugs:
1. `end_offset` was missing the alignment that might have been
introduced while computing `base_ptr`.
2. Ignoring point 1, `end_offset` computed the offset of the first byte
that is outside the current chunk. However, this might be in the
middle of a (hypothetical) object! The loop treats `end_offset` as if
it points to the first byte beyond the last (valid) object. So if the
last few bytes of the chunk are unused, the loop iterates once too
often.
Found by OSS Fuzz, long-standing issue (since 2021-07-31)
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=38733
(This probably also resolves some other issues that go through
RegexMatcher.)
See also: 0f1425c895
Creating a String object from the formatted data is unnecessary, as it
immediately gets turned into a StringView anyways.
With this change, we will no longer allocate on the heap when printing
`VirtualAddress`, `VirtualRange` and `InodeIdentifier` objects, which is
a step towards stronger OOM hardening.
Forcing the formatting to go through `Formatter<FormatString>` is
completely unnecessary, increases code size, performs a String
allocation and prevents us from using the formatting options available
on that type.
This commit also removes explicit formatters from
`BlockBasedFileSystem::BlockIndex` and `Kernel::InodeIndex`, as those
are already covered by the blanket implementation for all
`DistinctNumeric` types.
There is no need to have a user-defined copy constructor that simply
calls the base class's copy constructor. By having the compiler generate
it for us, Span is made trivially copyable, so it can be passed in
registers.
This function converts a single wide character into its multibyte
representation (UTF-8 in our case). It is called from libc++'s
`std::basic_ostream<wchar_t>::flush`, which gets called at program exit
from a global destructor in order to flush `std::wcout`.
The platform independent Processor.h file includes the shared processor
code and includes the specific platform header file.
All references to the Arch/x86/Processor.h file have been replaced with
a reference to Arch/Processor.h.
Both my approach and the previous approach were wrong for different
cases. I've changed the Iterators index from storage-relative to
queue-relative, and it's much simpler and more obviously correct.
fixes#10383
While I was working on LibWeb, I got a page fault at 0xe0e0e0e4.
This indicates a destroyed RefPtr if compiled with SANITIZE_PTRS
defined. However, the page fault handler didn't print out this
indication.
This makes the page fault handler print out a note if the faulting
address looks like a recently destroyed RefPtr, OwnPtr, NonnullRefPtr,
NonnullOwnPtr, ThreadSafeRefPtr or ThreadSafeNonnullRefPtr. It will
only do this if SANITIZE_PTRS is defined, as smart pointers don't get
scrubbed without it being defined.
Some time ago, automatic locking was added to the AK smart pointers to
paper over various race conditions in the kernel. Until we've actually
solved the issues in the kernel, we're stuck with the locking.
However, we don't need to punish single-threaded userspace programs with
the high cost of locking. This patch moves the thread-safe variants of
RefPtr, NonnullRefPtr, WeakPtr and RefCounted into Kernel/Library/.
Example failure:
IDAllocator.h only pulls in AK/Hashtable.h, so any compilation unit that
includes AK/IDAllocator.h without including AK/Traits.h before it used
to be doomed to fail with the cryptic error message "In instantiation of
'AK::HashTable<T, TraitsForT, IsOrdered>::Iterator AK::HashTable<T,
TraitsForT, IsOrdered>::find(const T&) [with T = int; TraitsForT =
AK::Traits: incomplete type 'AK::Traits<int>' used in nested name
specifier".
This change also applys to to_uint
On i686 u64 == long long but on x86_64 u64 == long. Therefor on either
arch, one of the instantiations is missed. This change makes sure that
all integer types have an instantiation.
It's very common to encounter single-character strings in JavaScript on
the web. We can make such strings significantly lighter by having a
1-character inline capacity on the Vectors.
Until we're confident that RequestServer doesn't need this runtime debug
dump helper, it's much nicer if everyone has it built in, so they can
simply send a SIGINFO if they see it acting up.
Otherwise we'd end up trying to delete the wrong connection if a
connection made before us is deleted.
Fixes _some_ RequestServer spins (though not all...).
This commit also adds a small debug mechanism to RequestServer (which
can be enabled by turning REQUEST_SERVER_DEBUG on), that can dump all
the current active connections in the cache, what they're doing, and how
long they've been doing that by sending it a SIGINFO.
The first argument for LexicalPath::join was a String const&, which
resulted in an unnecessary memory copy when invoked with a StringView.
Changing the type of the parameter to StringView avoids the extra cost.
This was noticed while working on #10230.
We already know the length of these substrings, so there's no need to
check their lengths using strlen in the StringView(char*) constructor.
This patch also removes an accidental 1-byte OOB read that was left over
from when this method received null-terminated char pointers instead of
string views, as well removes the unnecessary lambda call (two of the
checks were impossible, and this was only called in one place, so we can
just inline it)
This was required before commit 5f724b6ca1
when we were building LibC before libstdc++ headers were available in
the sysroot. However as noted in that commit, we never actually needed
to be building LibC before libstdc++, so we can go ahead and remove this
ancient hack.
The only use of this define was removed in commit
5f724b6ca1, over a year ago.
The define was there to prevent the LibC build we built before libstdc++
from complaining about missing libgcc symbols. However, as noted in that
commit, we never actually needed to build LibC at all.
"result" is a tad bit too generic to provide a clash-free experience -
we found instances in LibJS where this breaks already. Essentially this
doesn't work:
auto foo = TRY(bar(result));
Because it expands to the following within the TRY() scope:
{
auto result = bar(result);
...
}
And that of course fails:
error: use of ‘result’ before deduction of ‘auto’
The simple solution here is to use a name that is much less likely to
clash with anything used in the expression ("_temporary_result"). :^)
DisjointChunks<T> provides a nice interface over multiple sequential
Vector<T>'s, allowing the user to iterate over/index into/slice from
said buffers as if they were a single contiguous buffer.
To work with views on such objects, DisjointSpans<T> is provided, which
has the same behaviour but does not own the underlying objects.
These are required in the specification and used by the web's URL
built-in, this commit also removes the Badge<AK::URL> from URLParser
to allow other classes that need to call the parser directly like the
web's URL built-in to do so.
This should:
- Accept const& on both sides
- Not involve implicit conversions on either side
otherwise the argument order would be deemed significant, and trip this
warning.
Serenity has explicit_bzero() in LibC with the same implementation,
however we need to be able to use this from Lagom on all platforms
that we support building serenity on. I've implemented it in AK for
this reason.
This removes the awkward String::replace API which was the only String
API which mutated the String and replaces it with a new immutable
version that returns a new String with the replacements applied. This
also fixes a couple of UAFs that were caused by the use of this API.
As an optimization an equivalent StringView::replace API was also added
to remove an unnecessary String allocations in the format of:
`String { view }.replace(...);`
This was needlessly copying StringView arguments, and was also using
strstr internally, which meant it was doing a bunch of unnecessary
strlen calls on it. This also moves the implementation to StringUtils
to allow API consistency between String and StringView.
This variant of dbgputstr does not lock the global log lock, as it is
called before the current or any other processor was initialized,
meaning that:
A) The $gs base was not setup yet, so we cannot enter into critical
sections, and as a result we cannot use SpinLocks
B) No other processors may try to print at the same time anyway
This bit me because I accidentally made the destructor for a class which
was wrapped in an Optional private. This causes none of the Optional
destructors to be able to be deduced, which when combined with concepts
causes an internal compile error in GCC 10.3.0+. This commit adds a note
here to make sure that future encounters of this bug does not surprise
people.
c27abaabc4 moved this out of the global
namespace, but did not qualify its users.
While this seems to be fine (sometimes, somehow), let's qualify it to
avoid random breakage.
This is in preparation for making KBufferBuilder::append() and friends
return a KResult. Long-term we should come up with a solution that works
for both kernel and userspace clients of the JSON API.
This type is useful, as the sizes will be visible in the compiler error
messages, as they will be part of the template parameters. This is not
possible with a normal static_assert of the sizeof a type.
The way we use classes like Kernel::KResultOr<T> and AK::Result<T, E>
makes checking for errors (and short-circuiting returns) quite verbose.
This patch adds a new TRY(expression) macro that either evaluates to
the released result of the expression if successful, or returns the
error if not.
Before:
auto foo_or_error = get_foo();
if (foo_or_error.is_error())
return foo_or_error.release_error();
auto foo = foo_or_error.release_value();
After:
auto foo = TRY(get_foo());
The macro uses a GNU C++ extension which is supported by GCC, Clang,
Intel C++, and possibly others. It's not *ideal*, but since it makes our
codebase considerably nicer, let's try(!) it out. :^)
Co-authored-by: Ali Mohammad Pur <mpfard@serenityos.org>
This commit moves the KResult and KResultOr objects to Kernel/API to
signify that they may now be freely used by userspace code at points
where a syscall-related error result is to be expected. It also exposes
KResult and KResultOr to the global namespace to make it nicer to use
for userspace code.
This function ensures that a key is present in the HashMap.
If it's not present, it is inserted, and the corresponding value
is initialized with whatever the callback returns.
It allows us to express this:
auto it = map.find(key);
if (it == map.end()) {
map.set(it, make_a_value());
it = map.find(key);
}
auto& value = it->value;
Like this:
auto& value = map.ensure(key, [] { return make_a_value(); });
Note that the callback is only invoked if we have to insert a missing
key into the HashMap. This is important in case constructing the default
value is expensive or otherwise undesirable.
This introduces a new define AK_DONT_REPLACE_STD that disables our own
implementation of std::move and std::forward. Some ports include both
STL and AK headers which causes conflicts when trying to resolve those
functions. The port can define AK_DONT_REPLACE_STD before including
Serenity headers in that case.
This avoids a value copy when calling value() or value_or() on a
temporary Optional. This is very common when using the HashMap::get()
API like this:
auto value = hash_map.get(key).value_or(fallback_value);
Our existing implementation did not check the element type of the other
pointer in the constructors and move assignment operators. This meant
that some operations that would require explicit casting on raw pointers
were done implicitly, such as:
- downcasting a base class to a derived class (e.g. `Kernel::Inode` =>
`Kernel::ProcFSDirectoryInode` in Kernel/ProcFS.cpp),
- casting to an unrelated type (e.g. `Promise<bool>` => `Promise<Empty>`
in LibIMAP/Client.cpp)
This, of course, allows gross violations of the type system, and makes
the need to type-check less obvious before downcasting. Luckily, while
adding the `static_ptr_cast`s, only two truly incorrect usages were
found; in the other instances, our casts just needed to be made
explicit.
And also try_create<T> => try_make_ref_counted<T>.
A global "create" was a bit much. The new name matches make<T> better,
which we've used for making single-owner objects since forever.
