Early ELF binaries defined dlopen and friends in crt0.o by
using function pointers in the object handle passed from rtld.
This contract wastes space when many shared objects are allocated
and ties dynamic linker and rest of the system tightly together.
Fake the entry points in a static object and restrict them to the
platforms that had working ELF support at the time. Keep the magic and
version field used by modern crt0.o for all architectures. The checks
will be removed from crt0.o in a follow-up step.
- Move to a shared _rtld_call_ifunc for rel and rela architectures
- Architectures using rel format must patch IRELATIVE non-PLT
relocations like RELATIVE in additition to the later ifunc handling
- Consistently record the delta to the end of the relocation group for
non-PLT IRELATIVE relocations
Hidden ifunc is now supported on all ifunc platforms, even when using
-fno-plt. The combination of -fno-plt and relro is broken due to
incorrect GNU ld output though.
into a header for reuse in crt0.o for static ifunc support. Change the
existing logic for sparc64 to use the Bicc variant of ba,a as it allows
+-8MB displacement compared to the BPcc variant's +-1MB. Teach the sparc
variant the same trick for using ba,a and not sethi+jmp when possible.
thread allocation:
(1) Set the DTV vector up whenever an offset into the static allocation
is assigned, even if the block itself is not initialized. This has been
seen in libstdc++.
(2) Do not free a DTV block if it is part of the static thread
allocation.
functions are used for destructors of thread_local objects.
If a pending destructor exists, prevent unloading of shared objects.
Introduce __dl_cxa_refcount interface for this purpose. When the last
reference is gone and the object has been dlclose'd before, the
unloading is finalized.
Ideally, __cxa_thread_atexit_impl wouldn't exist, but libstdc++ insists
on providing __cxa_thread_atexit as direct wrapper without further
patching.
been changed to cope with more and at least Go actively creates them.
Adjust the mapping size computation to use the maximum and not depend on
PT_LOAD segments to be in order.
The COMBREL logic predates thread-safety of the dynamic linker and
breaks the use of shared locks for the common symbol lookup case. It is
unlikely to provide any benefit for lazy binding or PLT lookups, so
provide equivalent functionality in the non-PLT relocation handling loop
by checking if the symbol used by the current relocation is the same as
the one used during the last lookup. No inter-object cachine is done as
it is also unlikely to be benefical.
Testing with Firefox startup on AMD64 shows a small performance gain by
the new method.
relocations for _SDA_BASE_ and _SDA2_BASE_ that happened to work by
accident. They no longer happen to work when old binaries are run on
newer systems. Explicitly ignore these bogus relocations.
"really unmap the gap between the text and data rather than just removing
all access with mprotect(). the latter results in the kernel having to
keep track of that range separately since the permissions are different.
avoid calling mmap() with a size of zero."
As per toolchain/52054: src/libexec/ld.elf_so update breaks everything,
this commit is very broken for some people (but not others). chs mentioned
he has a fix, but best not to leave -current broken in the meantime.
all access with mprotect(). the latter results in the kernel having to
keep track of that range separately since the permissions are different.
avoid calling mmap() with a size of zero.
This fixes the most of lockups i observed with Open vSwitch
on NetBSD/amd64. ("most of" because it still occasionally
locks up because of other problems. see PR/49816)
function symbol with an implementation choosen at run time.
Refactor calls to functions by address in ld.elf_so to create temporary
function descriptors on the stack, if the address is not leaked outside.
Limitations:
- no support for initialising static storage with function pointers
- no support for unnamed resolver functions
Inspired by FreeBSD's r228435 by kib@freebsd.org.
martin