Stackprotector support was previously unavailable on s390 because by
default compilers generate code which is not suitable for the kernel:
the canary value is accessed via thread local storage, where the address
of thread local storage is within access registers 0 and 1.
Using those registers also for the kernel would come with a significant
performance impact and more complicated kernel entry/exit code, since
access registers contents would have to be exchanged on every kernel entry
and exit.
With the upcoming gcc 16 release new compiler options will become available
which allow to generate code suitable for the kernel. [1]
Compiler option -mstack-protector-guard=global instructs gcc to generate
stackprotector code that refers to a global stackprotector canary value via
symbol __stack_chk_guard. Access to this value is guaranteed to occur via
larl and lgrl instructions.
Furthermore, compiler option -mstack-protector-guard-record generates a
section containing all code addresses that reference the canary value.
To allow for per task canary values the instructions which load the address
of __stack_chk_guard are patched so they access a lowcore field instead: a
per task canary value is available within the task_struct of each task, and
is written to the per-cpu lowcore location on each context switch.
Also add sanity checks and debugging option to be consistent with other
kernel code patching mechanisms.
Full debugging output can be enabled with the following kernel command line
options:
debug_stackprotector
bootdebug
ignore_loglevel
earlyprintk
dyndbg="file stackprotector.c +p"
Example debug output:
stackprot: 0000021e402d4eda: c010005a9ae3 -> c01f00070240
where "<insn address>: <old insn> -> <new insn>".
[1] gcc commit 0cd1f03939d5 ("s390: Support global stack protector")
Reviewed-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Preinitialize the return value, and break out the for loop in
module_finalize() in case of an error to get rid of an ifdef.
This makes it easier to add additional code, which may also depend
on config options.
Reviewed-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Extend execmem parameters to accommodate more complex overrides of
module_alloc() by architectures.
This includes specification of a fallback range required by arm, arm64
and powerpc, EXECMEM_MODULE_DATA type required by powerpc, support for
allocation of KASAN shadow required by s390 and x86 and support for
late initialization of execmem required by arm64.
The core implementation of execmem_alloc() takes care of suppressing
warnings when the initial allocation fails but there is a fallback range
defined.
Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org>
Acked-by: Will Deacon <will@kernel.org>
Acked-by: Song Liu <song@kernel.org>
Tested-by: Liviu Dudau <liviu@dudau.co.uk>
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
module_alloc() is used everywhere as a mean to allocate memory for code.
Beside being semantically wrong, this unnecessarily ties all subsystems
that need to allocate code, such as ftrace, kprobes and BPF to modules and
puts the burden of code allocation to the modules code.
Several architectures override module_alloc() because of various
constraints where the executable memory can be located and this causes
additional obstacles for improvements of code allocation.
Start splitting code allocation from modules by introducing execmem_alloc()
and execmem_free() APIs.
Initially, execmem_alloc() is a wrapper for module_alloc() and
execmem_free() is a replacement of module_memfree() to allow updating all
call sites to use the new APIs.
Since architectures define different restrictions on placement,
permissions, alignment and other parameters for memory that can be used by
different subsystems that allocate executable memory, execmem_alloc() takes
a type argument, that will be used to identify the calling subsystem and to
allow architectures define parameters for ranges suitable for that
subsystem.
No functional changes.
Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org>
Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Song Liu <song@kernel.org>
Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
During module load, module layout allocation occurs by initially
allowing the architecture to frob the sections. This is performed via
module_frob_arch_sections().
However, the size of each module memory types like text,data,rodata etc
are updated correctly only after layout_sections().
After calculation of required module memory sizes for each types,
move_module() is responsible for allocating the module memory for each
type from modules vaddr range.
Considering the sequence above, module_frob_arch_sections() updates the
module mod_arch_specific got_offset before module memory text type size
is fully updated in layout_sections(). Hence mod_arch_specific
got_offset points to currently zero.
As per s390 ABI,
R_390_GOTENT : (G + O + A - P) >> 1
where
G=me->mem[MOD_TEXT].base+me->arch.got_offset
O=info->got_offset
A=rela->r_addend
P=loc
fix R_390_GOTENT calculation in apply_rela().
Note: currently this doesn't break anything because me->arch.got_offset
is zero. However, reordering of functions in the future could break it.
Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
Pull s390 updates from Vasily Gorbik:
- Add support for stackleak feature. Also allow specifying
architecture-specific stackleak poison function to enable faster
implementation. On s390, the mvc-based implementation helps decrease
typical overhead from a factor of 3 to just 25%
- Convert all assembler files to use SYM* style macros, deprecating the
ENTRY() macro and other annotations. Select ARCH_USE_SYM_ANNOTATIONS
- Improve KASLR to also randomize module and special amode31 code base
load addresses
- Rework decompressor memory tracking to support memory holes and
improve error handling
- Add support for protected virtualization AP binding
- Add support for set_direct_map() calls
- Implement set_memory_rox() and noexec module_alloc()
- Remove obsolete overriding of mem*() functions for KASAN
- Rework kexec/kdump to avoid using nodat_stack to call purgatory
- Convert the rest of the s390 code to use flexible-array member
instead of a zero-length array
- Clean up uaccess inline asm
- Enable ARCH_HAS_MEMBARRIER_SYNC_CORE
- Convert to using CONFIG_FUNCTION_ALIGNMENT and enable
DEBUG_FORCE_FUNCTION_ALIGN_64B
- Resolve last_break in userspace fault reports
- Simplify one-level sysctl registration
- Clean up branch prediction handling
- Rework CPU counter facility to retrieve available counter sets just
once
- Other various small fixes and improvements all over the code
* tag 's390-6.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (118 commits)
s390/stackleak: provide fast __stackleak_poison() implementation
stackleak: allow to specify arch specific stackleak poison function
s390: select ARCH_USE_SYM_ANNOTATIONS
s390/mm: use VM_FLUSH_RESET_PERMS in module_alloc()
s390: wire up memfd_secret system call
s390/mm: enable ARCH_HAS_SET_DIRECT_MAP
s390/mm: use BIT macro to generate SET_MEMORY bit masks
s390/relocate_kernel: adjust indentation
s390/relocate_kernel: use SYM* macros instead of ENTRY(), etc.
s390/entry: use SYM* macros instead of ENTRY(), etc.
s390/purgatory: use SYM* macros instead of ENTRY(), etc.
s390/kprobes: use SYM* macros instead of ENTRY(), etc.
s390/reipl: use SYM* macros instead of ENTRY(), etc.
s390/head64: use SYM* macros instead of ENTRY(), etc.
s390/earlypgm: use SYM* macros instead of ENTRY(), etc.
s390/mcount: use SYM* macros instead of ENTRY(), etc.
s390/crc32le: use SYM* macros instead of ENTRY(), etc.
s390/crc32be: use SYM* macros instead of ENTRY(), etc.
s390/crypto,chacha: use SYM* macros instead of ENTRY(), etc.
s390/amode31: use SYM* macros instead of ENTRY(), etc.
...
Make use of the set_direct_map() calls for module allocations.
In particular:
- All changes to read-only permissions in kernel VA mappings are also
applied to the direct mapping. Note that execute permissions are
intentionally not applied to the direct mapping in order to make
sure that all allocated pages within the direct mapping stay
non-executable
- module_alloc() passes the VM_FLUSH_RESET_PERMS to __vmalloc_node_range()
to make sure that all implicit permission changes made to the direct
mapping are reset when the allocated vm area is freed again
Side effects: the direct mapping will be fragmented depending on how many
vm areas with VM_FLUSH_RESET_PERMS and/or explicit page permission changes
are allocated and freed again.
For example, just after boot of a system the direct mapping statistics look
like:
$cat /proc/meminfo
...
DirectMap4k: 111628 kB
DirectMap1M: 16665600 kB
DirectMap2G: 0 kB
Acked-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
This is the s390 variant of commit 7dfac3c5f4 ("arm64: module: create
module allocations without exec permissions"):
"The core code manages the executable permissions of code regions of
modules explicitly. It is no longer necessary to create the module vmalloc
regions with RWX permissions. So create them with RW- permissions instead,
which is preferred from a security perspective."
Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
The ftrace code assumes at two places that module_alloc() returns
executable memory. While this is currently true, this will be changed
with a subsequent patch to follow other architectures which implement
ARCH_HAS_STRICT_MODULE_RWX.
Acked-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Randomize the load address of modules in the kernel to make KASLR effective
for modules.
This is the s390 variant of commit e2b32e6785 ("x86, kaslr: randomize
module base load address").
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
module_layout manages different types of memory (text, data, rodata, etc.)
in one allocation, which is problematic for some reasons:
1. It is hard to enable CONFIG_STRICT_MODULE_RWX.
2. It is hard to use huge pages in modules (and not break strict rwx).
3. Many archs uses module_layout for arch-specific data, but it is not
obvious how these data are used (are they RO, RX, or RW?)
Improve the scenario by replacing 2 (or 3) module_layout per module with
up to 7 module_memory per module:
MOD_TEXT,
MOD_DATA,
MOD_RODATA,
MOD_RO_AFTER_INIT,
MOD_INIT_TEXT,
MOD_INIT_DATA,
MOD_INIT_RODATA,
and allocating them separately. This adds slightly more entries to
mod_tree (from up to 3 entries per module, to up to 7 entries per
module). However, this at most adds a small constant overhead to
__module_address(), which is expected to be fast.
Various archs use module_layout for different data. These data are put
into different module_memory based on their location in module_layout.
IOW, data that used to go with text is allocated with MOD_MEM_TYPE_TEXT;
data that used to go with data is allocated with MOD_MEM_TYPE_DATA, etc.
module_memory simplifies quite some of the module code. For example,
ARCH_WANTS_MODULES_DATA_IN_VMALLOC is a lot cleaner, as it just uses a
different allocator for the data. kernel/module/strict_rwx.c is also
much cleaner with module_memory.
Signed-off-by: Song Liu <song@kernel.org>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
MIPS is the only remaining architecture that needs to patch jump label
NOP encodings to initialize them at load time. So let's move the module
patching part of that from generic code into arch/mips, and drop it from
the others.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20220615154142.1574619-3-ardb@kernel.org
Merge yet more updates from Andrew Morton:
"This is the material which was staged after willystuff in linux-next.
Subsystems affected by this patch series: mm (debug, selftests,
pagecache, thp, rmap, migration, kasan, hugetlb, pagemap, madvise),
and selftests"
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (113 commits)
selftests: kselftest framework: provide "finished" helper
mm: madvise: MADV_DONTNEED_LOCKED
mm: fix race between MADV_FREE reclaim and blkdev direct IO read
mm: generalize ARCH_HAS_FILTER_PGPROT
mm: unmap_mapping_range_tree() with i_mmap_rwsem shared
mm: warn on deleting redirtied only if accounted
mm/huge_memory: remove stale locking logic from __split_huge_pmd()
mm/huge_memory: remove stale page_trans_huge_mapcount()
mm/swapfile: remove stale reuse_swap_page()
mm/khugepaged: remove reuse_swap_page() usage
mm/huge_memory: streamline COW logic in do_huge_pmd_wp_page()
mm: streamline COW logic in do_swap_page()
mm: slightly clarify KSM logic in do_swap_page()
mm: optimize do_wp_page() for fresh pages in local LRU pagevecs
mm: optimize do_wp_page() for exclusive pages in the swapcache
mm/huge_memory: make is_transparent_hugepage() static
userfaultfd/selftests: enable hugetlb remap and remove event testing
selftests/vm: add hugetlb madvise MADV_DONTNEED MADV_REMOVE test
mm: enable MADV_DONTNEED for hugetlb mappings
kasan: disable LOCKDEP when printing reports
...
Machine generations up to z9 (released in May 2006) have been officially
out of service for several years now (z9 end of service - January 31, 2019).
No distributions build kernels supporting those old machine generations
anymore, except Debian, which seems to pick the oldest supported
generation. The team supporting Debian on s390 has been notified about
the change.
Raising minimum supported machine generation to z10 helps to reduce
maintenance cost and effectively remove code, which is not getting
enough testing coverage due to lack of older hardware and distributions
support. Besides that this unblocks some optimization opportunities and
allows to use wider instruction set in asm files for future features
implementation. Due to this change spectre mitigation and usercopy
implementations could be drastically simplified and many newer instructions
could be converted from ".insn" encoding to instruction names.
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
If the size of the PLT entries generated by apply_rela() exceeds
64KiB, the first ones can no longer reach __jump_r1 with brc. Fix by
using brcl. An alternative solution is to add a __jump_r1 copy after
every 64KiB, however, the space savings are quite small and do not
justify the additional complexity.
Fixes: f19fbd5ed6 ("s390: introduce execute-trampolines for branches")
Cc: stable@vger.kernel.org
Reported-by: Andrea Righi <andrea.righi@canonical.com>
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
s390 allows hotpatching the mask of a conditional jump instruction.
Make use of this feature in order to avoid the expensive stop_machine()
call.
The new trampolines are split in 3 stages:
- A first stage is a 6-byte relative conditional long branch located at
each function's entry point. Its offset always points to the second
stage for the corresponding function, and its mask is either all 0s
(ftrace off) or all 1s (ftrace on). The code for flipping the mask is
borrowed from ftrace_{enable,disable}_ftrace_graph_caller. After
flipping, ftrace_arch_code_modify_post_process() syncs with all the
other CPUs by sending SIGPs.
- Second stages for vmlinux are stored in a separate part of the .text
section reserved by the linker script, and in dynamically allocated
memory for modules. This prevents the icache pollution. The total
size of second stages is about 1.5% of that of the kernel image.
Putting second stages in the .bss section is possible and decreases
the size of the non-compressed vmlinux, but splits the kernel 1:1
mapping, which is a bad tradeoff.
Each second stage contains a call to the third stage, a pointer to
the part of the intercepted function right after the first stage, and
a pointer to an interceptor function (e.g. ftrace_caller).
Second stages are 8-byte aligned for the future direct calls
implementation.
- There are only two copies of the third stage: in the .text section
for vmlinux and in dynamically allocated memory for modules. It can be
an expoline, which is relatively large, so inlining it into each
second stage is prohibitively expensive.
As a result of this organization, phoronix-test-suite with ftrace off
does not show any performance degradation.
Suggested-by: Sven Schnelle <svens@linux.ibm.com>
Suggested-by: Vasily Gorbik <gor@linux.ibm.com>
Co-developed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Link: https://lore.kernel.org/r/20210728212546.128248-3-iii@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Because of late module patching, a livepatch module needs to be able to
apply some of its relocations well after it has been loaded. Instead of
playing games with module_{dis,en}able_ro(), use existing text poking
mechanisms to apply relocations after module loading.
So far only x86, s390 and Power have HAVE_LIVEPATCH but only the first
two also have STRICT_MODULE_RWX.
This will allow removal of the last module_disable_ro() usage in
livepatch. The ultimate goal is to completely disallow making
executable mappings writable.
[ jpoimboe: Split up patches. Use mod state to determine whether
memcpy() can be used. Test and add fixes. ]
Cc: linux-s390@vger.kernel.org
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> # s390
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Reuse str_has_prefix instead of strncmp with hardcoded length to
make the intent of a comparison more obvious.
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Move from modules area entire shadow memory preallocation to dynamic
allocation per module load.
This behaivior has been introduced for x86 with bebf56a1b: "This patch
also forces module_alloc() to return 8*PAGE_SIZE aligned address making
shadow memory handling ( kasan_module_alloc()/kasan_module_free() )
more simple. Such alignment guarantees that each shadow page backing
modules address space correspond to only one module_alloc() allocation"
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The main linker script vmlinux.lds.S for the kernel image merges
the expoline code patch tables into two section ".nospec_call_table"
and ".nospec_return_table". This is *not* done for the modules,
there the sections retain their original names as generated by gcc:
".s390_indirect_call", ".s390_return_mem" and ".s390_return_reg".
The module_finalize code has to check for the compiler generated
section names, otherwise no code patching is done. This slows down
the module code in case of "spectre_v2=off".
Cc: stable@vger.kernel.org # 4.16
Fixes: f19fbd5ed6 ("s390: introduce execute-trampolines for branches")
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Automatically decide between nobp vs. expolines if the spectre_v2=auto
kernel parameter is specified or CONFIG_EXPOLINE_AUTO=y is set.
The decision made at boot time due to CONFIG_EXPOLINE_AUTO=y being set
can be overruled with the nobp, nospec and spectre_v2 kernel parameters.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
