Implement refcounter based data protection guard
to avoid unnecessary calls to mprotect when dlopen/dlclose
is called from a constructor.
Bug: 19124318
Big: 7941716
Change-Id: Id221b84ce75443094f99756dc9950b0a1dc87222
Handy when looking at strace output for seeing "where does my code start?",
though you'll need to set LD_DEBUG=1 (or greater) and constructors will
already have run (but you can set LD_DEBUG to a higher value to track that).
Change-Id: I82d53a7123b39febf5b2a956de83c8a30f229651
Must build this linker as a shared object.
A statically linked module does not load during init.
BUG: 18008984
Change-Id: I5f0411129024edcb9381c473ea4e741a075e9184
Make the request structure match the new structure found in debuggerd
code.
On 64 bit systems, make sure that the 32 bit compiled linker is contacted.
Bug: https://code.google.com/p/android/issues/detail?id=97024
Change-Id: I675b8ff222fcf836b05de4e9b5976ff91ca929bf
Position Independent Executables cannot communicate with GDB using DT_MIPS_RLD_TAG.
Hence a new tag DT_MIPS_RLD_MAP2 got introduced to fix the issue.
Change-Id: I77ce3faf1ef2e121003b804ecd291abb25cfbb56
Signed-off-by: Nikola Veljkovic<Nikola.Veljkovic@imgtec.com>
1. Take into consideration that the local_group_root_
is not set for all of not linked libraries.
2. We need to check visited list earlier to avoid double
soinfo_free.
Change-Id: Iabc0a06a97c63f7e6bd4641731f50bb1466fed4f
Count references on the group level to avoid
partially unloading function that might be
referenced by other libraries in the local_group
Bonus: with this change we can correctly unload recursively
linked libraries. is_recursive check is removed.
Also dynamic executables (not .so) with 0 DT_NEEDED libraries
are now correctly linked.
Change-Id: Idfa83baef402840599b93a875f2881d9f020dbcd
Relocate symbol against DF_1_GLOBAL shared libraries
loaded before this shared library. This includes
main executable, ld_preloads and other libraries
that have DF_1_GLOBAL flag set.
Bug: 2643900
Bug: 15432753
Change-Id: Ia788748be59b739d1c7e62b978e7e255f9e65c7b
The local group is a sequence of libraries in default (breadth-first)
order. It allows RTLD_LOCALLY loaded library to correctly relocate
symbols within its group (see test-cases).
Local group lookup is performed after main executable and ld_preloads.
Bug: 2643900
Bug: 15432753
Change-Id: I9bb013b46d17dbb5cbdfb8fef26f552748385541
replace lseek() and use pread() instead
add test for library_fd_offset > file_size case
Bug: 17762003
Change-Id: I4555f0be635124efe849c1f226985bcba72ffcbd
DT_STRSZ Implement strtab boundary checks
DT_FLAGS_1 Warn if flags other than DF_1_NOW|DF_1_GLOBAL are set
Bug: 17552334
Change-Id: Iaad29cd52f5b2d7d2f785fb351697906dc1617d9
This reverts commit 8f61d99183
Despite the fact that static linker does all the work while linking
-Bsymbolic executables, according to the SCO doc following DT_SYMBOLIC
and DF_SYMBOLIC flags is still a requirement for the dynamic linker
as well.
(see http://www.sco.com/developers/gabi/2012-12-31/ch5.dynamic.html)
Change-Id: I23f6a5e9b662e40cdbc462b64feff912431e800c
Get dynamic flags from phdr table's correct entry rather the first
entry, so that the following DT_DEBUG entry can be set.
Also fix the undefined reference to LoadTask::deleter issue under gcc
-O0 option.
Bug: 17524778
Change-Id: I9c679af197b034761fb739d6c980e628ff2ab84c
This is needed when building the linker against a libc that was built
with coverage information.
Bug: 17553780
Change-Id: If131002e067dd7768fc657c30e62c42a45c05ace
The debuggerd case can probably never happen, because you're crashing at this
point anyway. The system property one seems possible though.
Change-Id: Idba6a4f1d68587ec5b320d1e25f0b6a987ea32a0
From the elf-spec: "Symbolically bound shared objects are
identified by the .dynamic entry DT_SYMBOLIC. This tag is
informational only; the runtime linker processes symbol
lookups from these objects in the same manner as any
other object."
Change-Id: I4aae0e502ed9fac096fbf076f4d6b2e055d784f6
Symbols from libraries opened with RTLD_LOCAL (default)
should not be visible via dlsym(RLTD_DEFAULT/RTLD_NEXT, .)
Bug: 17512583
Change-Id: I1758943081a67cf3d49ba5808e061b8251a91964
Expanded test for recursive libs. Fixed bug with unnecessary
soinfo_free of already loaded library.
Change-Id: I2cc19f2650c8b12a35feeac127ef608ebba44d88
There are number of changes in the way IFUNC related relocations are done:
1. IRELATIVE relocations are now supported for x86/x86_64 and arm64.
2. IFUNC relocations are now relying on static linker to generate
them in correct order - this removes necessety of additional
relocation pass for ifuncs.
3. Related to 2: rela?.dyn relocations are preformed before .plt ones.
4. Ifunc are resolved on symbol lookup this approach allowed to avoid
mprotect(PROT_WRITE) call on r-x program segments.
Bug: 17399706
Bug: 17177284
Change-Id: I414dd3e82bd47cc03442c5dfc7c279949aec51ed
Enable the -std=gnu++11 flag for libstdc++ static and
dynamic libs.
ScopeGuard uses DISABLE_ macros instead of '= delete';
Change-Id: I07e21b306f95fffd49345f7fa136cfdac61e0225
This patch fixes the problem with symbol search order
for dlsym(RTLD_DEFAULT/RTLD_NEXT, .) by loading libraries
and ld_preloads in correct order.
Bug: https://code.google.com/p/android/issues/detail?id=74255
Change-Id: If1ba5c807322409faa914e27ecb675e2c4541f0d
Attempt: 2
Previously we duplicated code in build/core/executable.mk and it's hard
to maintain. LOCAL_POST_LINK_CMD is a clean solution for this.
Bug: 17403674
Change-Id: I8c054bf782af68cf1f88b8d787a78ca74a7968b5
Clang is still disabled for x86 and x86_64 long double code,
for x86_64 special assembly instruction, and the linker module.
BUG: 17163651
BUG: 17302991
BUG: 17403674
Change-Id: I43c5f5f0ddc0c2a31918f00b57150dadd8f3f26b
This patch fixes the problem with symbol search order
for dlsym(RTLD_DEFAULT/RTLD_NEXT, .) by loading libraries
and ld_preloads in correct order.
Bug: https://code.google.com/p/android/issues/detail?id=74255
Change-Id: I4cf84c70dbaabe99310230dfda12385ae5401859
Change lookup order during relocation so that
ld_preloads always precede caller (unless caller
is main executable).
Asan needs this change in order to intercept libc->libc
calls.
Bug: 15432753
(cherry picked from commit 05e190c093)
Change-Id: I5bfb58e18015b1ec5b77842dbb37fb122fa1fd1a
When enabling debug malloc, the snprintf calls in the linker fails to
update the buffer.
The problem is that snprintf makes a call to pthread_getspecific that
returns a valid pointer, but the data it points to is zero. This should
never happen and causes the snprintf to stop and do nothing.
Temporarily replace snprintf with a different implementation to work
around this issue.
Bug: 16874447
Bug: 17302493
(cherry pick from commit 172955a4e3)
Change-Id: Idca9d417978403d61debfd0434aaa82fd770f33b
This includes:
1. Placing has_ifunc after fields with version = 0
2. Switch to has_min_version(v) function.
3. Minor soinfo initialization refactoring (placement new + ctor)
Change-Id: I1bf5fde4d930914012ce5f3ad5acb48217da9b2d
Do not run symbol lookup on already visited soinfos
Not taking into account already visited libraries
dramatically slows down dlsym in cases when there
are multiple occurrences of a large library in
dependency tree.
Bug: 16977077
Change-Id: I1379f30ed8b06758dd1cc76b80833ac8589afa50
ifuncs now work in i386 and x86_64 when called in the same library as
well as in a different library.
Bug:6657325
Change-Id: Ic0c48b1b0a76cb90f36c20c79f68294cc3fd44a1
dlsym(3) with handle != RTLD_DEFAULT|RTLD_NEXT performs
breadth first search through the dependency tree.
Bug: 16653281
Change-Id: I017a6975d1a62abb0218a7eb59ae4deba458e324
PR_GET_DUMPABLE is used by an application to indicate whether or
not core dumps / PTRACE_ATTACH should work.
Security sensitive applications often set PR_SET_DUMPABLE to 0 to
disable core dumps, to avoid leaking sensitive memory to persistent
storage. Similarly, they also set PR_SET_DUMPABLE to zero to prevent
PTRACE_ATTACH from working, again to avoid leaking the contents
of sensitive memory.
Honor PR_GET_DUMPABLE when connecting to debuggerd. If an application
has said it doesn't want its memory dumped, then we shouldn't
ask debuggerd to dump memory on its behalf.
FORTIFY_SOURCE tests: Modify the fortify_source tests to set
PR_SET_DUMPABLE=0. This reduces the total runtime of
/data/nativetest/bionic-unit-tests/bionic-unit-tests32 from approx
53 seconds to 25 seconds. There's no need to connect to debuggerd
when running these tests.
Bug: 16513137
Change-Id: Idc7857b089f3545758f4d9b436b783d580fb653f
There's no need: __linker_init only takes one argument.
Also remove the arm __CTOR_LIST__; we use .init_array and .fini_array instead
of .ctor and .dtor anyway, and I don't think we've ever supported the latter.
Change-Id: Ifc91a5a90c6aa39d674bf0509a7af2e1ff0beddd
Also fix a few formatting issues in copyright headers that were confusing
the script (though obviously it would be better if the script were smarter).
Change-Id: I7f561bef4f84fdcbd84f375ee226bd65db0e507b
/system/bin/tf_daemon on Nakasi has been updated so it's
a PIE executable. We can now drop non-PIE support.
This reverts commit d81b3b275d.
Bug: 15086752
Bug: 15732002
Bug: 14566672
Change-Id: I832c4de01a8fa050754ea64da317d1639990265e
On Nakasi builds, /system/bin/tf_daemon is a pre-compiled
vendor supplied binary. When support for non-PIE executables
was removed, that daemon failed to start, which induced
a number of bugs.
* keystore is broken
* CTS runs will hang after a certain period of time
Temporarily reenable non-PIE support for 32-bit platforms
until Nakasi is fixed.
This rolls back the following commits:
Cleanup: updated comments
* 6275f20834
Cleanup: remove AARCH/ARM_COPY relocation support
* b906e13c55
And restricts the following patch to 64 bit only:
Remove support for non-PIE executables
* 2aebf5429b
Bug: 14566672
Bug: 15086752
Bug: 15732002
Change-Id: Ia2501aa14bd30feb4a6ce66bdb7c9f066dba0b5f
While introducing RTLD_NOLOAD in b648a8a5 a change in how soinfo structs are
connected was also introduced. When calling dlopen the library that is loaded
is added as a child to the soinfo from which the caller comes - i.e. building a
dependency graph.
Unfortunately this shows issues upon unloading, such as blowing the stack if
there are loops in the graph.
This change reverts that part of b648a8a5, keeping everything as a child of the
root soinfo.
Change-Id: I604ee9210613b19f693a568158e69707620a95db
Signed-off-by: Bjorn Andersson <bjorn.andersson@sonymobile.com>
* Removed unnecessary NULL check in dlsym
* Fixed dlsym_failure test to account for
correct RTLD_DEFAULT value
* Added temporary check for legacy RTLD_DEFAULT
value for non-yet-recompiled binaries
Bug: 15146875
Change-Id: I089fa673762629f5724b6e4fbca019d9cfc39905
bionic RTLD_NEXT was changed from 0xffffffff to -1L, which breaks
existing binaries compiled. Temporarily look either until we can
get recompiled binaries.
Bug: 15113039
Change-Id: I1568fa0e4a832c5e6df21da74a33b22bde7f16f6
Linker is now able to resolve symlinked libraries correctly.
soinfo is extended to save the graph of dependencies during
load/unload. Dependencies are used only in CallConstructor.
Bug: 9741592
Change-Id: Id9c48a74c46aa89bcdf3d54ec2f8ba3d398130b1
Freeing block mprotects on the page which it turn
may lead to application crash if linker subsequently
tries to modify another block on the page.
Bug: 14895266
Change-Id: I8ff7f5df467d7be184242de652032b3c84e24b76
Don't support the dynamic linker loading ET_EXEC executables. Only
support ET_DYN executables.
Bug: 14566672
Change-Id: Ia1f2bfffe7ea867c1cd62e67d7fd637c60c0728d
If the file has no relro segment, the generated relro file will have
length 0, which caused mmap to fail. If the relro file has nonzero size,
but is too short (e.g. because it's for the wrong version of the
library), the linker would segfault while comparing the data. Fix both
these issues: don't try to map a zero length file, and don't try to
compare data that would be beyond the end of the file.
Improve test to explicitly generate two versions of the library: one
with -z relro, and one with -z norelro, so we can test both cases; also
explicitly test the case where the relro file has length 0.
Bug: 14299541
Change-Id: Id8b95585edda90e8bb5de452a35b70ed2d224934
Because we re-raise various signals, we corrupt the si_code that debuggerd
sees when it ptraces our siginfo. One possible solution (shown here) is to
pass the original si_code value in the message we send to debuggerd.
Change-Id: I76f9aa2c0442e5cab611d132532409e700383907
Add missing second param to soinfo_link_image in a section that's not
compiled on ARM.
Bug: 13005501
Change-Id: Id0ede8e03da4e05b25c0aeb24a840f868031d4e8
This has been annoying me for a while, because it's often quite misleading.
Today, for example, I saw:
Fatal signal 13 (SIGPIPE) at 0x6573 (code=0), thread 25971 (top)
where the apparent address is actually the pid of the signal source (in this
case the kernel on behalf of the thread itself).
This patch isn't as fancy as strace, but it at least means we never say
anything misleading. We could decode the si_code field like strace and
debuggerd, but I'm reluctant to do that without some way to share the code
between at least bionic and debuggerd.
Examples after:
Fatal signal 13 (SIGPIPE), code 0 in tid 9157 (top)
Fatal signal 11 (SIGSEGV), code 1, fault addr 0x0 in tid 9142 (crasher64)
Fatal signal 6 (SIGABRT), code -6 in tid 9132 (crasher64)
(Note that the code still shows as 0 for SIGPIPE in the signal handler itself
but as -6 (SI_TKILL) in debuggerd; this is actually correct --- debuggerd is
showing the re-raised signal sent at the end of the signal handler that
initially showed the correct code 0.)
Change-Id: I71cad4ab61f422a4f6687a60ac770371790278e0
Add flags and a file descriptor to android_dlopen_ext() to allow writing
the RELRO section of the loaded library to a file after relocation
processing, and to allow mapping identical pages from the file over the
top of relocated memory in another process. Explicitly comparing the
pages is required in case a page contains a reference to a symbol
defined in another library loaded at a random base address.
Bug: 13005501
Change-Id: Ibb5b2d384edfaa5acf3e97a5f8b6115c10497a1e
Add flags and parameters to android_dlopen_ext() to allow loading a
library at an already-reserved fixed address. If the library to be
loaded will not fit within the space reserved, then the linker will
either fail, or allocate its own address space as usual, according to
which flag has been specified. This behaviour only applies to the
specific library requested; any other libraries loaded as dependencies
will be loaded in the normal fashion.
There is a new gtest included to cover the functionality added.
Bug: 13005501
Change-Id: I5d1810375b20fc51ba6a9b3191a25f9792c687f1
Add a function "android_dlopen_ext()", defined in <android/dlext.h>.
This is an extended version of dlopen() which takes a struct for passing
additional parameters for Android-specific functionality. This will be
used to support RELRO section sharing between separate processes.
Bug: 13005501
Change-Id: I9c99b2f2a02ee329dedaeba09ef3a1113b17b2d6
Use LOCAL_MODULE_STEM_32 and LOCAL_MODULE_STEM_64 to build
linker and linker64. Set LOCAL_MULTILIB := both for consistency
with other rules of this type, although LOCAL_MULTILIB := both is
only consumed by executable.mk which is skipped by
linker_executable.mk. Duplicate the necessary new parts of
executable.mk.
Change-Id: I888e87cf125cfbcfdee6a55b6377d8bc5ce402ea
Our <machine/asm.h> files were modified from upstream, to the extent
that no architecture was actually using the upstream ENTRY or END macros,
assuming that architecture even had such a macro upstream. This patch moves
everyone to the same macros, with just a few tweaks remaining in the
<machine/asm.h> files, which no one should now use directly.
I've removed most of the unused cruft from the <machine/asm.h> files, though
there's still rather a lot in the mips/mips64 ones.
Bug: 12229603
Change-Id: I2fff287dc571ac1087abe9070362fb9420d85d6d
When bionic's dynamic linker processes the .dynamic section of a MIPS ELF binary and encounters the DT_MIPS_RLD_MAP dynamic array tag, it calculates the address of where to write a pointer to the _r_debug structure. The current implementation simply reads the value given in the d_ptr field and writes the pointer address to that location. However, this value has to be adjusted to reflect the real load address of the binary. Otherwise the linker will write to a faulty location possibly resulting in a crash when linking a MIPS binary that includes DT_MIPS_RLD_MAP. This change corrects that problem.
Change-Id: I1a91874f7ab47289001fe72d9016660c14c70362
Signed-off-by: Benjamin Adolphi <b.adolphi@gmail.com>
Remove the linker's reliance on BSD cruft and use the glibc-style
ElfW macro. (Other code too, but the linker contains the majority
of the code that needs to work for Elf32 and Elf64.)
All platforms need dl_iterate_phdr_static, so it doesn't make sense
to have that part of the per-architecture configuration.
Bug: 12476126
Change-Id: I1d7f918f1303a392794a6cd8b3512ff56bd6e487
Also move some of the stuff that should be in <link.h> out of the
private "linker.h", to make it clearer that these are public API
known to gdb that we can't change.
Bug: 12554197
Change-Id: I830e1260d3d8b833ed99bc1518f1c6b6102be8af
Some system functions like munmap expect soinfo's size to be size_t, but currently
it is unsigned. Change it to size_t to fit 64bit's portability.
Change-Id: I0bf6d522b38b0cd9bf1db05b004b5326217412a2
Signed-off-by: Weiwu Chen <weiwu.chen@intel.com>
Signed-off-by: Qiming Shi <qiming.shi@intel.com>
Do not use the magic number 0xFFFFFFFFU to represent the max value of an address
as it's not correct on 64bit platform. We should use UINTPTR_MAX instead.
Change-Id: I1fc6f5864a651b2eddea2333cb0788f9d9223270
Signed-off-by: Qiming Shi <qiming.shi@intel.com>
Signed-off-by: Weiwu Chen <weiwu.chen@intel.com>
We're not going to have init(1) set LD_LIBRARY_PATH globally on 64-bit.
This patch makes it possible for libnativehelper to set LD_LIBRARY_PATH
in each Java VM (to support System.loadLibrary) without also hard-coding
the default search path there.
Change-Id: If13961fae976e06dd80d5ef522f31e8b7eb01154
We don't need this on architectures other than aarch64, and
we're still investigating why we need it on aarch64, but it
doesn't seem unreasonable to have this flag set when linking
the dynamic linker anyway; it's clearly the intended behavior.
Change-Id: I4fa1b4ae543a818979934bf818eabac03bb9154f
If the linker can't resolve its own internal references to symbols,
we currently exit silently (albeit with EXIT_FAILURE). Not very helpful.
Change-Id: I1614fc970dee4560b38832ede1987b65a8e53a1e
Addition of support for AArch64 in the linker64 target.
Change-Id: I8dfd9711278f6706063e91f626b6007ea7a3dd6e
Signed-off-by: Marcus Oakland <marcus.oakland@arm.com>
Under valgrind ehdr_vdso is null and causing segfault.
Adding debug info for vdso producing plenty of issues on debugging
through gdbserver, removing it. It doesn't seem it should be here.
Unwinding through vdso test still works.
Change-Id: I1a7e233c493f2268d725fa8d7279822d07decc49
Signed-off-by: Pavel Chupin <pavel.v.chupin@intel.com>
Integration of kernel VDSO into internal bionic data structures using
common functions.
Fix for dl_iterate_phdr function: the function provides incorrect
address of object in case of nonzero virtual and base addresses.
Location in address space of a particular program header should be
calculated using the formula: addr = base_addr + virtual_addr.
Signed-off-by: Sergey Melnikov <sergey.melnikov@intel.com>
Change-Id: Ie2ab4257fd456242aab8afed0bd5bd6b29e81d6d
We don't need our own architecture macros; the standard ones will do.
This patch also fixes some __x86_64__ tests to be USE_RELA tests instead,
because they're not actually x86_64-specific.
I've cleaned up architecture-specific code slightly so where possible
all the code corresponding to a particular architecture is together.
This patch also fixes a bug in LP64 DT_PLTGOT handling, which should be
an error rather than falling through into DT_DEBUG! There was another #ifdef
bug where we'd only report unexpected DT_ entries on MIPS.
Change-Id: Id1d04e372611f641c1aa278a18e379f28af9eaf5
Like the new TODO says, though, do we need this at all? Wouldn't we be better
off just using the regular architecture-specific macros? This is a dynamic
linker, so there's never a possibility of cross-linking.
Change-Id: I7be6b8663f5e585c4024a49aae383430c86a3c1b
Also clean up <signal.h> and revert the hacks that were necessary
for 64-bit in linker/debugger.cpp until now.
Change-Id: I3b0554ca8a49ee1c97cda086ce2c1954ebc11892
The x86_64 build was failing because clone.S had a call to __thread_entry which
was being added to a different intermediate .a on the way to making libc.so,
and the linker couldn't guarantee statically that such a relocation would be
possible.
ld: error: out/target/product/generic_x86_64/obj/STATIC_LIBRARIES/libc_common_intermediates/libc_common.a(clone.o): requires dynamic R_X86_64_PC32 reloc against '__thread_entry' which may overflow at runtime; recompile with -fPIC
This patch addresses that by ensuring that the caller and callee end up in the
same intermediate .a. While I'm here, I've tried to clean up some of the mess
that led to this situation too. In particular, this removes libc/private/ from
the default include path (except for the DNS code), and splits out the DNS
code into its own library (since it's a weird special case of upstream NetBSD
code that's diverged so heavily it's unlikely ever to get back in sync).
There's more cleanup of the DNS situation possible, but this is definitely a
step in the right direction, and it's more than enough to get x86_64 building
cleanly.
Change-Id: I00425a7245b7a2573df16cc38798187d0729e7c4
If __get_tls has the right type, a lot of confusing casting can disappear.
It was probably a mistake that __get_tls was exposed as a function for mips
and x86 (but not arm), so let's (a) ensure that the __get_tls function
always matches the macro, (b) that we have the function for arm too, and
(c) that we don't have the function for any 64-bit architecture.
Change-Id: Ie9cb989b66e2006524ad7733eb6e1a65055463be
