8000 Tags · frida/frida-gum · GitHub
[go: up one dir, main page]

Skip to content

Tags

Tags

17.26.1

Toggle 17.26.1's commit message
subprojects: Bump outdated

17.26.0

Toggle 17.26.0's commit message
style: Order prototypes like definitions

17.25.2

Toggle 17.25.2's commit message
gumjs: Strip source in compiled QuickJS bytecode

Since QuickJS BC_VERSION 5, JS_WriteObject() serializes each function's
source slice into b->debug.source unless JS_STRIP_SOURCE is set on the
JSRuntime, duplicating nested function bodies throughout the compiled
bytecode blob. Match gumquickcompile by enabling JS_STRIP_SOURCE in
gum_compile_script_task_run().

17.25.1

Toggle 17.25.1's commit message
module-registry: Tear down with early destructors

On Windows the registry hooks LdrLoadDll() and LdrUnloadDll(), and
reverting those hooks patches code. When an agent delegates code memory
to its host, the memory hooks needed for that are gone by the time final
destructors run, so the patch failed and we aborted.

Tear down in gum_shutdown() instead, so callers can do so while patching
still works.

17.25.0

Toggle 17.25.0's commit message
windows: Bound nearby allocation searches

gum_memory_allocate_near() could return memory outside the requested
range, or memory it had already freed. A downward search reaching
address 0 called VirtualAlloc (NULL, ...), letting the OS pick any
address, and a first step beyond max_distance returned the freed initial
allocation.

Limit the search to the requested range and the application address
bounds, returning NULL when nothing fits. Step by the allocation
granularity, since VirtualAlloc() rounds reservations down to it, and
try the aligned near address first.

17.24.3

Toggle 17.24.3's commit message
subprojects: Bump outdated

17.24.2

Toggle 17.24.2's commit message
tests: Cover darwin-mapper TLV lifecycle

Map a small dylib with thread-locals in-process and check that each
thread gets fresh storage with the initial values, that the storage is
freed when a thread exits and when the module is unloaded, and that the
destructor releases the key.

17.24.1

Toggle 17.24.1's commit message
subprojects: Bump outdated

17.24.0

Toggle 17.24.0's commit message
thumb-writer: Fix double literal commit on flush

The alignment NOP emitted while committing the pool at flush could push
pending literals past the commit threshold. That triggered a nested
commit, after which the outer one patched the same LDRs a second time,
OR-ing a bogus offset into them. Stalker's Thumb callouts hit this
intermittently in large blocks, crashing with SIGILL on QNX.

Clear the earliest literal instruction up front so the pool cannot be
committed from within its own commit.

17.23.3

Toggle 17.23.3's commit message
gumjs: Read REAL columns as doubles on V8

gum_parse_column() fetched SQLITE_FLOAT values with
sqlite3_column_int64(), so on V8 `SELECT 1.5` stepped to [1] while
QuickJS returned [1.5]. Use sqlite3_column_double() like the QuickJS
backend does.
0