| Age | Commit message (Collapse) | Author |
|---|
|
Addresses feedback from PR #8188.
|
|
Closes #6630.
|
|
Closes #8020 and #8021.
|
|
|
|
|
|
BinaryViewType gains a HasNoInitialContent function that views can use
to suppress layout restoration when opening a file. The layout will
still be restored when the view is loaded from a saved database.
Fixes https://github.com/Vector35/binaryninja-api/issues/8083.
|
|
|
|
The processed Objective-C metadata is not saved to the .bdnb. It must be
recomputed when the view is loaded.
Fixes https://github.com/Vector35/binaryninja-api/issues/8030.
|
|
The `FileAccessorCache`'s LRU eviction could discard a file's mapped
data after slide info had already been applied to it. Future accesses to
the file produced a fresh mapping, but failed to reapply the slide info.
This could result in pointers not correctly being slid, such as in
https://github.com/Vector35/binaryninja-api/issues/7689.
The LRU cache existed to stay within OS file descriptor limits, since
the old `MappedFile` held its fd open for the lifetime of the mapping.
There's no real reason for it to hold the file descriptor open like
this. Closing it after `mmap` is sufficient to avoid the file descriptor
limits.
`MappedFileRegion` replaces the combination of `FileAccessorCache`,
`WeakFileAccessor`, and `MappedFileAccessor`. It closes the fd
immediately after mmap, so all files can stay mapped without consuming
descriptors, making the cache unnecessary.
`MappedFileRegion` is owned directly by the `CacheEntry` for its full
lifetime. Slide info is applied exactly once to each `MappedFileRegion`.
|
|
|
|
|
|
inlining during analysis
Previously the address of the instruction in the function being inlined
was used as the new instruction's address when copying it during
inlining. Now there is an additional option: use the address of the
call instruction that is being replaced as the new instruction's
address.
This new mode is useful when inlining thunks or stub functions, but care
must be taken if using it beyond that.
The benefit is that it ensures that when a function contains multiple
calls to the same stub function, each inlined copy ends up with distinct
addresses. This ensures that call type adjustments and other overrides
that are stored on the function and keyed by address can be applied
independently to each callsite that was inlined.
The trade-off is that if the function being inlined contains non-trivial
logic, all of the inlined instructions sharing an address will limit
what type of adjustments can be applied to them.
The Objective-C and shared cache workflows are updated to take advantage
of this new mode when they enable inlining of stub functions. This will
make it possible for multiple calls to the same runtime function within
a single function to have separate call type adjustments applied in the
future.
|
|
|
|
|
|
This defers rebuilding of the section map until after all sections for
an image have been added, rather than having the section map be rebuilt
after adding each section.
|
|
Fixes https://github.com/Vector35/binaryninja-api/issues/7666.
Correctly managing the state of bulk symbol modifications via
`BeginBulkModifySymbols` / `EndBulkModifySymbols` is error-prone in the
face of exceptions and early returns. Leaking a bulk symbol modification
can leave the view in a state where no further changes to symbols will
be applied.
All users of the C++ API are encouraged to move from
`BeginBulkModifySymbols` / `EndBulkModifySymbols` to the new
`BulkSymbolModification` class.
|
|
unbacked regions
Fixes https://github.com/Vector35/binaryninja-api/issues/7724
|
|
This ensures they do not overlap with any images that may later be loaded from the shared cache.
|
|
|
|
The down and enter keys now preserve an existing selection in the
associated list or table views, rather than unconditionally selecting or
activating the first item.
|
|
See https://github.com/Vector35/binaryninja-api/issues/7247#issuecomment-3392192797 for details.
|
|
|
|
Extra noise in the CMake configure step, much of the value from these has been removed after the refactor(s), e.g. metadata version is no longer used
|
|
by default
The table views previously visually suggested they were being sorted by
address, but in practice they were not sorted until the user set an
explicit sort order by clicking on a column header.
|
|
These files are present alongside iOS 26 shared caches. Explicitly
ignoring them avoids an error being logged to the console.
|
|
In some iOS 15 caches, the .symbols file's mapping has an address of 0.
This would cause it to be returned by `SharedCache::GetEntryContaining`
and loaded into the view.
The .symbols file contains the local symbol tables for images in the
shared cache. It is not intended to be mapped into the same address
space as the rest of the shared cache.
`SharedCache` now tracks the symbols cache entry separately from other
entries. A dedicated `VirtualMemory` region is used when accessing the
data it contains. This could be a `FileAccessor`, but that would require
additional changes within `SharedCacheMachOHeader`.
`SharedCacheMachOProcessor` now directly accesses the local symbols
cache entry rather than needing to search for it.
|
|
Prior to macOS 13 / iOS 16, the base offset to use for relative direct
selector references within Objective-C message lists was stored within
the `__TEXT,__objc_opt_ro` section of /usr/lib/libobjc.A.dylib.
|
|
These calls set the rpaths used for release builds of the dylibs using a
macro that isn't available in the API repository. They are not necessary
when building the shared cache plug-in outside of the context of the
app.
|
|
This is functionally equivalent to the previous workflow_objc, with the
following changes:
1. It mutates the `core.function.metaAnalysis` workflow rather than
registering a new named workflow. The activities now all check for
the presence of the Objective-C metadata added by `ObjCProcessor` to
determine whether they should do work, rather than relying on
`MachoView` to override the function workflow when Objective-C
metadata is present. This fixes
https://github.com/Vector35/binaryninja-api/issues/6779.
2. The auto-inlining of `objc_msgSend` selector stub functions is
performed in a separate activity from the processing of
`objc_msgSend` call sites. The selector stub inlining activity is
configured so that it does not run in `DSCView` as the shared cache
needs different behavior for stub functions more generally that
`SharedCacheWorkflow` already provides.
3. The way that types like `id` and `SEL` are referenced is fixed so
that they show up as `id` rather than `objc_struct*`.
This also replaces the Objective-C portion of the shared cache's
workflow, and incorporates several bug fixes that had been applied to it
but not the standalone Objective-C workflow.
|
|
Calls to `Workflow::Instance` that were looking up a built-in workflow
name are updated to use `Workflow::Get`. Others use `Workflow::GetOrCreate`.
|
|
C++ plug-ins now consistently use the `plugin_rpath` or
`ui_plugin_rpath` macros to ensure they have `SKIP_BUILD_RPATH` set when
building on Mac (i.e., no `LC_RPATH` is added).
Rust plug-ins have their build.rs updated to only specify `-Wl,-rpath`
when building for Linux. It is not needed on macOS. On macOS we instead
explicitly specify an `@rpath`-relative install name. This doesn't
change any behavior, but avoids leaving an absolute path as the
library's install name and is consistent with CMake's behavior for C++
plug-ins.
|
|
|
|
message
|
|
This ensures that the Python source files are only generated and copied
into the output directory if inputs have changed, rather than being done
unconditionally.
|
|
This ensures that CMake detects when files that match the glob are added
or removed.
|
|
|
|
|
|
|
|
Avoid copying the vector of symbols where possible. When no filter
string is applied we no longer copy the entire vector symbols by having
the symbols used for display be indirected via a pointer. It points
either to the unfiltered symbols or the filtered symbols.
Some unncessary copies have been removed by using `std::move` where
appropriate.
Filtering has been updated to avoid `std::vector::reserve` /
`std::vector::shrink_to_fit` in favor of letting the filtered vector
control its own growth and reuse its buffer. This avoids allocating a
~100MB buffer every time we update the filter only to later reallocate
and move the contents into a smaller buffer. Instead the buffer grows
via `std::vector`'s usual growth algorithm and it is preserved across
filter calls to avoid unnecessarily reallocating it, and only shrink the
buffer if it has shrunk significantly vs previous iterations.
|
|
We're copying `std::string` objects that know their length. There's no
reason to force a call to `strlen`.
|
|
`GetRegionAt` was really providing the semantics of
`GetRegionContaining` due to `m_regions` being an `AddressRangeMap` and
using transparent comparators to allow `find` to work with any address
in the range.
`GetRegionAt` is updated to verify that the start address of the region
that was found matches the requested address. `GetRegionContaining` is
updated to use `AddressRangeMap::find` rather than doing a linear search
over all regions.
|
|
This adds support for the `__objc_arrayobj`, `_objc_dictobj`,
`__objc_intobj`, `__objc_floatobj`, `__objc_doubleobj` and
`__objc_dateobj` sections that contain Objective-C constants. These are
emitted by Apple's versions of Clang for `const` literals, amongst other
things.
|
|
The order that the operands to `+` are evaluated in is unspecified. Clang
happens to evaluate them left to right and gives the expected answer.
MSVC picks the opposite order and so the value it computes is off by 4.
This was resulting in missing method names when arm64 binaries
containing Objective-C are analyzed on Windows.
|
|
This is enabled as is, there might need to be some restrictions later on, does not seem to have an unreasonable impact analysis time
|
|
Apparently the compiler on macOS is less strict about this change?
|
|
|
|
|
|
Also update minimum CMake version.
|
|
cause bugs and also issues with C++20
|
|
backed by a database
|
