ALP-pool/kernel-livepatch-MICRO-6-0_Update_3
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
ae0d953551
kernel-livepatch-MICRO-6-0_.../kernel-livepatch-MICRO-6-0_Update_3.changes

1112 lines
48 KiB
Plaintext
Raw Blame History

-------------------------------------------------------------------
Mon Sep 30 17:57:42 CEST 2024 - nstange@suse.de
- New branch for MICRO-6-0_Update_3
- commit f95a323
-------------------------------------------------------------------
Mon Sep 2 13:02:29 CEST 2024 - nstange@suse.de
- scripts: make tar-up recognize SLE Micro codestreams
- commit be8c692
-------------------------------------------------------------------
Mon Oct 23 11:19:00 CEST 2023 - nstange@suse.de
- klp_syscalls.h: adapt to kernels >= 6.1 on s390x and ppc64le
There had been a couple of changes to the kernel's architecture specific
syscall related definitions on ppc64le and s390x, which require some
amendments to the klp_syscalls.h abstraction wrappers to enable support:
- 7e92e01b7245 ("powerpc: Provide syscall wrapper"),
- 94746890202c ("powerpc: Don't add __powerpc_ prefix to syscall entry
points"),
- 2213d44e140f ("s390/syscalls: get rid of system call alias functions").
Implement that.
- commit 049524d
-------------------------------------------------------------------
Fri Oct 13 11:26:56 CEST 2023 - nstange@suse.de
- kallsyms_relocs: Drop 'mod' argument from symbol iteration callback
With upstream commit 3703bd54cd37 ("kallsyms: Delete an unused parameter
related to {module_}kallsyms_on_each_symbol()"), the 'mod' argument is
no longer passed to the kallsyms symbol iteration callbacks. Drop it from
the kallsyms_relocs helper implementation accordingly.
- commit e5e774b
-------------------------------------------------------------------
Mon Feb 27 18:56:17 CET 2023 - mpdesouza@suse.com
- create-makefile.sh: Add Kbuild.inc files support
Check if there are Kbuild.inc files and copy it's content into the new
Makefile. An example of Kbuild.inc can be used to enable -Werror for specific
object on all architectures:
CFLAGS_livepatch_main.o = -Werror
- commit a23f264
-------------------------------------------------------------------
Fri Feb 10 19:10:03 CET 2023 - mpdesouza@suse.com
- livepatch_main.c: Remove KLP_NOREG_API checks
All currently supported codestreams define KLP_NOREG_API, making this
checks obsolete.
- commit 06c9fa0
-------------------------------------------------------------------
Mon Dec 12 01:53:36 CET 2022 - mpdesouza@suse.com
- klp_trace.h: Add KLPR_TRACE_EVENT macros
Currently klp-ccp cannot track and redefine macros that use livepatches
or exported symbols, thus generating a large amount of code that needs
to be massaged. These macros define two variants of TRACE_EVENT macros,
for kernels older than 5.10, and another version for > 5.10, which
dropped data_args argument.
- commit 288960b
-------------------------------------------------------------------
Thu Oct 27 15:59:38 CEST 2022 - nstange@suse.de
- klp_syscalls.h: fix KLP_COMPAT_SYSCALL_SYM() macro for s390
Currently, the KLP_COMPAT_SYSCALL_SYM() expands to the __s390_compat_*()
variant, which expects the original types for its arguments, not longs.
For compatibility with the other archs, make it expand to the __se_compat_*()
version expecting longs for its arguments.
- commit 36a47e5
-------------------------------------------------------------------
Tue Oct 18 13:09:46 CEST 2022 - nstange@suse.de
- rpm: enable support for builds against the -RT kernel variant
Currently there is only support for building a kernel-livepatch package against
the default flavor of the canonical (<empty>) kernel variant. Livepatches for
the -RT variant will be provided in the future though. Prepare the packaging
scripts for this.
More specifically, make scripts/tar-up.sh extract an optional "variant"
component from the codestream name (as read from scripts/release-version.sh):
- SLE15-SP4-RT_Update_xy would specify a -RT variant while
- SLE15-SP4_Update_xy retains its meaning and referes to the <empty> variant
as before.
Introduce a %variant macro to the spec file and make tar-up.sh to set its value,
just alongside the other substitutions its already doing.
Make the spec file's package name and dependency specification to depend on
the %variant as appropriate.
- commit c5dc06b
-------------------------------------------------------------------
Thu Jun 16 08:16:19 CEST 2022 - nstange@suse.de
- uname_patch: include livepatch_uname.h from the uname livepatch code
Currently, livepatch_uname.c doesn't include livepatch_uname.h, which prohibits
compile-time protoype checking. Include livepatch_uname.h from
livepatch_uname.c, just as it's always being done for all the other livepatches,
too.
- commit c3ba44c
-------------------------------------------------------------------
Thu Jun 16 08:08:52 CEST 2022 - nstange@suse.de
- klp_syscalls: provide s390x variant of KLP_SYSCALL_DECLx() for kernels >= 5.12
s390x' syscall prototypes have changed with kernel commit 3a790cc1c9ef ("s390:
pass struct pt_regs instead of registers to syscalls") accepted for v5.12.
Add a corresponding KLP_SYSCALL_DECLx() #define to klp_syscalls.h.
- commit 9380841
-------------------------------------------------------------------
Fri Jun 10 10:49:32 CEST 2022 - nstange@suse.de
- kallsyms_relocs: allow for explicit initialization
Currently, the kallsyms_relocs code initializes itself in a lazy fashion upon
the first usage of its klp_resolve_kallsyms_relocs().
However, that initialization code, __kallsyms_relocs_init(), is now also in
charge of populating the klpe_find_module function pointer. As users might
depend on klpe_find_module() before their first call to
klp_resolve_kallsyms_relocs(), give them a means to trigger the initialization
explicitly: rename __kallsyms_relocs_init() to klp_kallsyms_relocs_init() and
make it externally visible.
- commit 5fe7b9f
-------------------------------------------------------------------
Fri Jun 10 10:42:28 CEST 2022 - nstange@suse.de
- kallsyms_relocs: lookup find_module() at initialization time
The livepatch initialization handlers invoking the kallsyms_relocs functionality
for their resp. target modules usually depend on the kernel's find_module().
However, with upstream commit 089049f6c995 ("module: unexport find_module and
module_mutex"), find_module() got unexported and is no longer directly
available.
Make the kallsyms_relocs initialization code look it up via kallsyms and
make the result available to livepatch initialization handlers via the
new, externally visible klpe_find_module function poiner.
- commit 7c53b1f
-------------------------------------------------------------------
Fri Jun 10 10:36:19 CEST 2022 - nstange@suse.de
- kallsyms_relocs: give klp_module_kallsyms_on_each_symbol a "klpe_" prefix
In line with our usual convention of giving symbols populated by means of
kallsyms-lookups a "klpe_" prefix, rename the internal
"klp_module_kallsyms_on_each_symbol" to "klpe_module_kallsyms_on_each_symbol".
- commit 20e54d1
-------------------------------------------------------------------
Fri Jun 10 10:25:55 CEST 2022 - nstange@suse.de
- kallsyms_relocs: factor out lookup code and make __kallsyms_relocs_init() use it
Currently, __kallsyms_relocs_init() contains some open-coded kallsyms lookup for
filling in the klp_module_kallsyms_on_each_symbol function pointer at bootstrap
time. Future commits will make the init code to populate some more symbols,
which more or less resembles the functionality klp_resolve_kallsyms_relocs() is
already providing.
Enable code reuse by factoring out the kallsyms lookup related pieces from
klp_resolve_kallsyms_relocs() into the new __klp_resolve_kallsyms_relocs()
and make both, klp_resolve_kallsyms_relocs() and __kallsyms_relocs_init() to
invoke it.
- commit 1bef553
-------------------------------------------------------------------
Fri Jun 10 09:47:40 CEST 2022 - nstange@suse.de
- kallsyms_relocs: strip underscore prefix from __klp_resolve_kallsyms_relocs()
The double undescore prefix of "__klp_resolve_kallsyms_relocs()" suggests that
the caller is supposed to take care of some locking, of module_mutex in this
case.
However, since upstream commit 013c1667cf78 ("kallsyms: refactor
{,module_}kallsyms_on_each_symbol"), module_kallsyms_on_each_symbol(), and
hence also the __klp_resolve_kallsyms_relocs() using that, doesn't need to
have module_mutex locked by callers anymore.
To reflect this, remove the underscore prefix from
__klp_resolve_kallsyms_relocs() and drop the comment about module_mutex from
the code.
- commit 4476b8b
-------------------------------------------------------------------
Thu Feb 24 10:36:40 CET 2022 - mbenes@suse.cz
- livepatch: Add MODULE_INFO with a git revision
A git HEAD revision was removed from "uname -v" output. Add it as
MODULE_INFO to a live patch kernel module, so it can be acquired by
modinfo tool if needed. The information is also available in a rpm
changelog.
- commit ff67cb6
-------------------------------------------------------------------
Wed Feb 23 19:21:44 CET 2022 - mbenes@suse.cz
- uname_patch: Trim klp tag to fix the overflow
SLE15-SP4 introduces an option to specify a preempt model during boot.
new_utsname->version was updated to take this into account and contains
PREEMPT_DYNAMIC tag now. In the end, there is not much space left to
include our klp_tag and it overflows. Instead of removing the tag
completely, trim it so that the user can at least easily spot that a
live patch is installed on the system.
A git HEAD revision will be stored elsewhere.
While at it, make the tag const.
References: bsc#1196281
- commit 51e46f7
-------------------------------------------------------------------
Wed Feb 23 19:20:37 CET 2022 - mbenes@suse.cz
- uname_patch: Update to v5.14 kernel/sys.c
Backport upstream commit 88a686728b37 ("kbuild: simplify access to the
kernel's version").
- commit 86c9d55
-------------------------------------------------------------------
Thu Aug 5 16:50:11 CEST 2021 - nstange@suse.de
- scripts/register-patches.sh: fix issue with per-klp_object #if-guards
scripts/register-patches.sh is supposed to #if-guard each constructed
klp_object instance by the logical or of the individual functions'
associated conditions as specified in the corresponding
patched_funcs.csv entries. If only one such function entry doesn't have a
condition associated with it, the compound logical || would always evaluate
to true though and thus, register-patches.sh should skip the
per-klp_object #if-guard alltogether in this case.
To this end, the inner loop iterating over the function entries resets the
array o_conds of unique conditions seen for the current object and breaks
out upon encountering an unconditional patch entry, i.e. one w/o an empty
condition field. The problem is that the break from the inner loop has no
effect on the outer loop over the different patched_funcs.csv's and thus,
the emptied o_conds array can get populated again in the course of
processing a later patched_funcs.csv. Later code would then find the
non-empty o_conds and guard the currently constructed klp_object by oring
its individual entries together rather than omitting the #if-guard as a
whole as it should.
Fix this by introducing the boolean variable "any_unconds", flip it to true
upon encountering an unconditional function entry and force the o_conds
array to empty if any_unconds is found to be set once the outer loop has
completed.
- commit dae55a1
-------------------------------------------------------------------
Wed Jan 13 11:41:32 CET 2021 - nstange@suse.de
- klp_syscalls.h: fix syscall prototype mismatch on s390x for kernels >= 4.17
The __SYSCALL_DEFINEx(x, name, ...) macro as defined in
arch/s390/include/asm/syscall_wrapper.h declares two protoypes for
a given syscall: __s390x_sys##name() and __se_sys##name(). The former
symbol is made to be an alias to the latter and the function arguments are
of the "real" type as specified in the macro invocation whereas the
latter's argument types are transformed into longs.
Currently the KLP_SYSCALL_SYM() helper macro from our klp_syscalls.h
evaluates to the __s390x_sys##name() variant, but its expansion result is
intended to be used with KLP_SYSCALL_DECLx(), which does the transformation
of the arguments' types to longs. This results in compilation errors due to
the syscall prototype declaration from KLP_SYSCALL_DECLx() confliciting
with the one from __SYSCALL_DEFINEx(), if visible.
The current behaviour of KLP_SYSCALL_DECLx() should be retained in order
to keep it working for the compatibility stubs, i.e. with
KLP_SYSCALL_COMPAT_STUB_SYM(). So fix the issue by making KLP_SYSCALL_SYM()
to evaluate to the __se_sys##name() variant on 390x for kernel versions >=
4.17.
- commit 862bd77
-------------------------------------------------------------------
Tue Jan 12 13:38:00 CET 2021 - nstange@suse.de
- scripts/register-patches.sh: stringify klp_funcs' ->old_name
In order to enable the use of e.g. KLP_SYSCALL_SYM() for the to be
livepatched function's name in patched_funcs.csv, make register-patches.sh
wrap the emitted klp_funcs' ->old_name initialization values with
__stringify() rather than writing string tokens directly.
- commit f54c4d6
-------------------------------------------------------------------
Tue May 19 15:01:34 CEST 2020 - mbenes@suse.cz
- scripts: Disable use of klp-convert
klp-convert tool was introduced to improve a situation with unexported
symbols while preparing live patches. However, it is still not stable
enough and upstream still needs to decide the purpose of the tool. Given
that it is used only for uname patch and only on SLE15-SP1 it is better
to just disable it for now.
At the same, leave the infrastructure in place, because we might use it
in the future.
- commit 3397b3e
-------------------------------------------------------------------
Wed Apr 8 10:14:20 CEST 2020 - nstange@suse.de
- scripts: enable s390x for SLE12-SP4
The initial live patch shall be built on s390x for future SLE12-SP4 kernel
releases. Make tar-up.sh add s390x to ExclusiveArch from the (not yet
existing) SLE12-SP4_Update_13 onwards.
- commit f49a99e
-------------------------------------------------------------------
Mon Mar 30 11:59:20 CEST 2020 - nstange@suse.de
- scripts: enable s390x for SLE15-SP2
- commit 933574a
-------------------------------------------------------------------
Wed Mar 25 10:45:50 CET 2020 - nstange@suse.de
- scripts: Generate ExclusiveArch in spec file dynamically
s390x support is slowly being introduced for newly created
master-livepatch based branches. In order to avoid problems with existing
branches for e.g. the maintenance team, don't add s390x to the hard-coded
list of ExclusiveArchs, but let tar-up.sh enable it dynamically depending
on the codestream in question.
For now, s390x builds will be enabled on SLE12-SP5, beginning with
SLE12-SP5_Update_3 onwards.
- commit 27b683d
-------------------------------------------------------------------
Mon Dec 2 13:49:24 CET 2019 - mbenes@suse.cz
- Revert "shadow variables: allow for dynamic initialization"
This reverts commit 843c6fa42429afc1682cdb39119e7a011af2abc9.
- commit 23d37c8
-------------------------------------------------------------------
Mon Dec 2 13:40:37 CET 2019 - mbenes@suse.cz
- Revert "shadow variables: introduce upstream patch"
This reverts commit e899c4fd3fe7602ebd70f578d8475f1049de7c78.
- commit c1be24c
-------------------------------------------------------------------
Mon Dec 2 13:38:18 CET 2019 - mbenes@suse.cz
- Revert "shadow variables: drop EXPORT_SYMBOL()s"
This reverts commit ac6cfebd7f831213ebcd4b2690672871572ec49e.
- commit 5771a4b
-------------------------------------------------------------------
Mon Dec 2 13:38:04 CET 2019 - mbenes@suse.cz
- Revert "shadow variables: share shadow data among KGraft modules"
This reverts commit 8e1e705d4d56981949f7ae3854d8e1cc2be7f40f.
- commit 1c87412
-------------------------------------------------------------------
Mon Dec 2 13:37:30 CET 2019 - mbenes@suse.cz
- Revert "shadow variables: add KGR_SHADOW_ID helper"
This reverts commit 237c8f3d13c382321d3e65d138d328eae0b82f6c.
- commit 41936fd
-------------------------------------------------------------------
Sat Sep 7 18:53:16 CEST 2019 - nstange@suse.de
- uname_patch: convert to the syscall stub wrapper macros from klp_syscalls.h
In order to make the live patch to the newuname() syscall work on
kernels >= 4.17 again, convert it to the KLP_SYSCALL_*() wrapper macros
provided by klp_syscalls.h.
References: bsc#1149841
- commit b5af38e
-------------------------------------------------------------------
Sat Sep 7 18:53:15 CEST 2019 - nstange@suse.de
- Provide wrapper macros for syscall naming
Live patching syscall stubs is a common task, for example any live patch
package modifies the newuname syscall.
For the actual definitions of the live patched syscall stubs, the
__SYSCALL_DEFINEx() name can always be (and often has been) used like e.g.
__SYSCALL_DEFINEx(3, _klp_timer_create, const clockid_t, which_clock,
struct sigevent __user *, timer_event_spec,
timer_t __user *, created_timer_id)
{
/* New implementation */
}
Up to kernel 4.16, this used to define a function named
"SyS_klp_timer_create" which could then be used to live patch the
"SyS_timer_create".
However, beginning with kernel version 4.17, resp. upstream commits
- fa697140f9a2 ("syscalls/x86: Use 'struct pt_regs' based syscall calling
convention for 64-bit syscalls")
- e145242ea0df ("syscalls/core, syscalls/x86: Clean up syscall stub
naming convention")
- d5a00528b58c ("syscalls/core, syscalls/x86: Rename struct pt_regs-based
sys_*() to __x64_sys_*()"),
things became more complex:
- The naming of the resulting stubs now varies across architecture.
- Some architectures (x86_64, s390x) instantiate an additional
compat stub for syscalls sharing a common implementation between 32 and
64 bits. (The 32 bit entry code used to convert from the 32 bit ABI to
64 bit and simply call the 64 bit syscall stub afterwards. That's
handled by the new 32 bit stubs now.)
- The stubs' signatures have changed: each argument used to get mapped
to either long or long long, but on x86_64, the stubs are now receiving
a single struct pt_regs only -- it's their responsibility to extract
the arguments as appropriate.
In order to not require each and every live patch touching syscalls to
include an insane amount of ifdeffery, provide a set of #defines hiding it:
1.) KLP_SYSCALL_SYM(name) expands to the syscall stub name for 64 bits
as defined by _SYSCALL_DEFINEx(x, _name, ...).
2.) If the architeture requires 32bit specific stubs for syscalls sharing
a common implementation between 32 and 64bits, the
KLP_ARCH_HAS_SYSCALL_COMPAT_STUBS macro is defined.
3.) If KLP_ARCH_HAS_SYSCALL_COMPAT_STUBS is defined, then
KLP_SYSCALL_COMPAT_STUB_SYM(name) expands to the syscall stub name
for 32 bits as defined by _SYSCALL_DEFINEx(x, _name, ...).
4.) For syscalls not sharing a common implementation between 32 and
64 bits, i.e. those defined by COMPAT_SYSCALL_DEFINEx(),
the macro KLP_COMPAT_SYSCALL_SYM(name) expands to the stub name
defined as defined by COMPAT_SYSCALL_DEFINEx(x, _name, ...).
5.) Finally, for hiding differences between the signatures,
provide the macro KLP_SYSCALL_DECLx(x, sym, ...) which
expands to a declaration of sym, with the x arguments either
mapped to long resp. long long each, or collapsed to a single
struct pt_regs argument as appropriate for the architecture.
Note that these macros are defined as appropriate on kernels before and
after 4.17, so that live patch code can be shared.
References: bsc#1149841
- commit da7b9a5
-------------------------------------------------------------------
Sat Aug 24 19:06:03 CEST 2019 - nstange@suse.de
- scripts/create-makefile.sh: add -I flag for toplevel directory to ccflags-y
Since upstream commit 58156ba4468f ("kbuild: skip 'addtree' and 'flags'
magic for external module build") Kbuild won't add an -I flag for an
external module's toplevel source directory to the compilation flags
anymore.
This results in compilation errors like the following:
uname_patch/livepatch_uname.c:36:10: fatal error: klp_convert.h: No such
file or directory
#include "klp_convert.h"
^~~~~~~~~~~~~~~
Fix this by appending '-I$(obj)' to ccflags-y within the Makefile created
by scripts/create-makefile.sh. Note that "$(obj)" is set to the current
source directory before the Makefile is sourced by Kbuild.
- commit b30a48e
-------------------------------------------------------------------
Thu Mar 7 15:23:42 CET 2019 - mbenes@suse.cz
- livepatch_main.c: Adaptation to a new livepatch API
The atomic replace patch set among others removed the two-stage API.
There is no (un)registration step needed now. SLES backport defines
KLP_NOREG_API macro to easily distinguish whether the kernel provides
the old or the new API. Use it and change the module init and exit
functions accordingly.
- commit 060163b
-------------------------------------------------------------------
Thu Feb 7 14:13:00 CET 2019 - mbenes@suse.cz
- uname_patch: Use klp-convert macros and rely on klp-convert where
possible
- commit 4c9eb70
-------------------------------------------------------------------
Wed Feb 6 14:12:44 CET 2019 - mbenes@suse.cz
- Define macros to switch easily between klp-convert and kallsyms
Kallsyms trick does not have to be used for resolving undefined symbols
when klp-convert is available. It would be great though to share live
patches sources between both modes of operation.
Define macros to help with the task. Their definitions depend on
whether USE_KLP_CONVERT macro is defined. tar-up.sh script is
responsible to decide.
- commit e3a42b7
-------------------------------------------------------------------
Wed Feb 6 10:53:44 CET 2019 - mbenes@suse.cz
- Use klp-convert where provided
klp-convert tool converts undefined symbols in a live patch kernel module
to special relocation records which are resolved by the kernel. It
allows to omit kallsyms tricks.
Wire it to the spec file and let tar-up.sh script decide if it is to be
used depending on a codestream. SLE15-SP1 is supported currently.
- commit 3efd330
-------------------------------------------------------------------
Tue Dec 11 11:27:23 CET 2018 - mbenes@suse.cz
- uname_patch: don't hold uts_sem while accessing userspace memory
Backport upstream patch 42a0cc347858 ("sys: don't hold uts_sem while
accessing userspace memory").
- commit d4e00de
-------------------------------------------------------------------
Tue Oct 2 16:38:19 CEST 2018 - mbenes@suse.cz
- scripts/tar-up.sh: Add ppc64le to ExclusiveArch even for SLE12-SP2
- commit 77a8a8b
-------------------------------------------------------------------
Wed Aug 8 15:08:00 CEST 2018 - nstange@suse.de
- Provide common kallsyms wrapper API
With bsc#1103203, the need for disambiguating between a multiply
defined symbol arose. This is something the kallsyms_lookup_name() based
code snippet we used to copy&paste to every individual CVE fix can't
handle.
Implement a proper wrapper API for doing the kallsyms lookups.
- commit bd113d8
-------------------------------------------------------------------
Wed Aug 8 15:07:59 CEST 2018 - nstange@suse.de
- Provide common kallsyms wrapper API
With bsc#1103203, the need for disambiguating between a multiply
defined symbol arose. This is something the kallsyms_lookup_name() based
code snippet we used to copy&paste to every individual CVE fix can't
handle.
Implement a proper wrapper API for doing the kallsyms lookups.
- commit 4aed7d2
-------------------------------------------------------------------
Wed Jul 11 13:55:14 CEST 2018 - nstange@suse.de
- provide KGR_SHADOW_ID() helper macro
- provide KLP_SHADOW_ID() helper macro
In analogy to the KGR_SHADOW_ID() macro, introduce KLP_SHADOW_ID() for
the construction of unique shadow variable id's.
- commit 7325c49
-------------------------------------------------------------------
Sun Jul 8 13:02:18 CEST 2018 - nstange@suse.de
- scripts/register-patches.sh: implement conditional inclusion
Currently, subpatches provide a patched_funcs.csv file describing what
needs to be patched. register-patches.sh inspects those to assemble one
global klp_patch structure.
The current format for these patched_funcs.csv's is
obj old_func(,sympos) newfun
However, sometimes subpatches depend on some kernel configuration values
like CONFIG_X86_64 and functions shall get patched only if the target
kernel configuration matches.
Extends the patched_funcs.csv format to
obj old_func(,sympos) newfun (cpp condition)
where everything coming after 'newfun' is taken to be a CPP condition to be
used for conditional inclusion. In case there's no condition specified,
assign that entry the same semantics as if a '1' had been given.
Make register-patches.sh guard the corresponding klp_func entries with #if
pragmas.
Furthermore, let it guard the enclosing klp_object instances by or'ing
together all its klp_funcs' conditions.
For the sake of better readability, omit redundant #if pragmas as well as
condition clauses. In particular,
- if a function entry hasn't got any condition explicitly specified,
there won't be any #if pragma, neither at the klp_func nor at the
klp_object level,
- if multiple function entries for an object are protected by the same
condition, it'll be or'ed in at the klp_object level only once,
- if all of an object's functions share the same condition, no #if pragmas
will be emitted at the klp_func level because they would only duplicate
what's already there for the enclosing object and
- multiple subsequent function entries sharing the same condition get
collated.
- commit 56f0729
-------------------------------------------------------------------
Sun Jul 8 13:02:17 CEST 2018 - nstange@suse.de
- scripts/register-patches.sh: allow spaces as patched_funcs.csv separators
Currently there's one single cut(1) usage which requires that (single) tabs
are used as field separators for the patched_funcs.csv.
As the rest of the code can deal with sequences of any whitespace already,
this imposes an unnecessary restriction on the format.
Substitute that cut(1) usage by a sed(1) invocation as appropriate.
- commit 9852661
-------------------------------------------------------------------
Mon Jun 4 15:20:08 CEST 2018 - mbenes@suse.cz
- livepatch_main.c: Set .replace to true
- commit 643f04c
-------------------------------------------------------------------
Mon May 14 08:30:00 CEST 2018 - nstange@suse.de
- scrips/create-makefile.sh: add support for assembly files
- commit cf2464a
-------------------------------------------------------------------
Mon Mar 5 15:44:31 CET 2018 - nstange@suse.de
- shadow variables: allow for dynamic initialization
Currently, the only shadow variable initialization scheme exposed by the
allocation API is to let klp_shadow_alloc() resp. klp_shadow_get_or_alloc()
memcpy some user provided buffer to the freshly allocated shadow variable.
This is too limited for shadow structures containing pointers into
themselves like list_heads or mutexes.
Change the internal __klp_shadow_get_or_alloc() to take a pointer to an
initializer functions and call that in place of the memcpy() operation.
In order to retain former functionality of klp_shadow_alloc() and
klp_shadow_get_or_alloc(), make them pass the new
__klp_shadow_memcpy_init() wrapper to __klp_shadow_get_or_alloc().
Finally, introduce the new klp_shadow_alloc_with_init() and
klp_shadow_get_or_alloc_with_init() which pass a user provided initializer
function pointer onwards to __klp_shadow_get_or_alloc().
- commit 843c6fa
-------------------------------------------------------------------
Wed Dec 6 14:40:14 CET 2017 - mbenes@suse.cz
- Revert "shadow variables: introduce upstream patch"
This reverts commit e899c4fd3fe7602ebd70f578d8475f1049de7c78.
- commit a27c66a
-------------------------------------------------------------------
Wed Dec 6 14:37:09 CET 2017 - mbenes@suse.cz
- Revert "shadow variables: drop EXPORT_SYMBOL()s"
This reverts commit ac6cfebd7f831213ebcd4b2690672871572ec49e.
- commit 40d0ba6
-------------------------------------------------------------------
Wed Dec 6 14:37:06 CET 2017 - mbenes@suse.cz
- Revert "shadow variables: share shadow data among KGraft modules"
This reverts commit 8e1e705d4d56981949f7ae3854d8e1cc2be7f40f.
- commit d184b38
-------------------------------------------------------------------
Wed Dec 6 14:36:56 CET 2017 - mbenes@suse.cz
- Revert "shadow variables: add KGR_SHADOW_ID helper"
This reverts commit 237c8f3d13c382321d3e65d138d328eae0b82f6c.
- commit 22d6153
-------------------------------------------------------------------
Wed Dec 6 12:18:06 CET 2017 - mbenes@suse.cz
- rpm/config.sh: Use SUSE:SLE-15:GA project
- commit ff32fc9
-------------------------------------------------------------------
Wed Dec 6 12:14:17 CET 2017 - mbenes@suse.cz
- Revert "scripts: Generate ExclusiveArch in spec file dynamically"
This reverts commit 95ed856ea8f99b4e48d7d324278b3628d2ac2fa2.
SLE15 will support ppc64le arch from the beginning.
- commit 92e9bdb
-------------------------------------------------------------------
Tue Dec 5 16:42:04 CET 2017 - mbenes@suse.cz
- uname_patch: fix UNAME26 for 4.0
Backport upstream commit 39afb5ee4640 ("kernel/sys.c: fix UNAME26 for
4.0").
- commit 5988feb
-------------------------------------------------------------------
Mon Dec 4 15:25:24 CET 2017 - mbenes@suse.cz
- Revert "Add compat.h to deal with changes of KGR_PATCH macro"
This reverts commit 4186bef35862029a2fd36ba4a73d5fa538992709.
All currently supported kernels (that is, everything since
SLE12_Update_14 and SLE12-SP1_Update_5) have sympos support. We can drop
compat, because we don't need it anymore.
- commit 11e3220
-------------------------------------------------------------------
Thu Nov 30 15:15:20 CET 2017 - mbenes@suse.cz
- scripts: Generate ExclusiveArch in spec file dynamically
ppc64le architecture kernel support is not present in all currently
supported branches. It may cause problem for the maintenance team.
Generate ExclusiveArch dynamically. It should be 'ppc64le x86_64' for
SLE12-SP3 and 'x86_64' for the rest.
- commit 95ed856
-------------------------------------------------------------------
Thu Nov 16 14:27:46 CET 2017 - mbenes@suse.cz
- rpm/kgraft-patch.spec: Add ppc64le as a supported arch
ppc64le is about to be supported in Live Patching product. Add it to
ExclusiveArch tag.
- commit 8437c94
-------------------------------------------------------------------
Thu Nov 16 14:26:35 CET 2017 - mbenes@suse.cz
- rpm/kgraft-patch.spec: Remove s390x from supported archs
s390x is not supported in Live Patching product. Remove it from
ExclusiveArch.
- commit f9614f2
-------------------------------------------------------------------
Tue Oct 31 10:34:53 CET 2017 - nstange@suse.de
- livepatch_main.c: klp_patch_init(): fix error handling
In case either of the invocations of klp_register_patch() or
klp_enable_patch() fails, anything which has been setup by the prior
per-(sub-)patch initialiation code, i.e. the expansion of
@@KLP_PATCHES_INIT_CALLS@@, won't get undone.
Fix this.
Also make klp_patch_init() look more like the common 'goto err' idiom
and adjust scripts/register_patches.sh accordingly.
Fix for commit 7e20201cdcb8 ("kGraft to livepatch migration. API
change.").
- commit 6552b44
-------------------------------------------------------------------
Tue Oct 31 10:34:52 CET 2017 - nstange@suse.de
- scripts/register_patches.sh: generate klp_object array
The KLP API doesn't take a flat list of to be patched functions
like KGraft did, but introduces an intermediate layer: struct
klp_object.
Each klp_patch instance is supposed to reference an array of
klp_object's which in turn provide an array of klp_func's each.
To facilitate merging, we want to generate this list of klp_object's
automatically, exactly like we did for the flat function list with KGraft.
For each klp_patch instance, there must be at most one klp_object entry
referring to the same object.
Hence care must be taken not to add an entry for the same object twice
in case two different (sub-)patches both patch some functions therein.
Require from each (sub-)patch to provide the list of to be patched
symbols in a file named SUBPATCH/patched_funcs.csv with each line
conforming to the
obj old_func(,sympos) new_func
pattern.
Make scripts/register.sh generate an klp_object array initializer based on
this and let it expand the @@KLP_PATCHES_OBJS@@ tag within livepatch_main.c
accordingly.
Do not replace the now obsolete @@KLP_PATCHES_FUNCS@@ anymore.
Add and remove the @@KLP_PATCHES_OBJS@@ and @@KLP_PATCHES_FUNCS@@
markers to and from livepatch_main.c respectively.
[ mb: amend copy&paste error ($newfun at the end of uname klp_func[]) ]
- commit 0fe721b
-------------------------------------------------------------------
Thu Oct 26 13:54:06 CEST 2017 - lpechacek@suse.com
- kGraft to livepatch migration. External rename.
External rename and thus final step of kGraft -> upstream livepatch
migration. kgraft-patch* modules are now livepatch* and live in
/lib/modules/$(uname -r)/livepatch.
References: fate#323682
[ mb: changelog ]
- commit f842fd5
-------------------------------------------------------------------
Thu Oct 5 12:12:29 CEST 2017 - nstange@suse.de
- shadow variables: add KGR_SHADOW_ID helper
As shadow variables are supposed to be shared among different KGraft
modules their id's must be compile time constants.
Introduce the KGR_SHADOW_ID helper macro for generating them in a uniform
manner based on the bsc# number and a local id.
- commit 237c8f3
-------------------------------------------------------------------
Thu Oct 5 12:12:28 CEST 2017 - nstange@suse.de
- shadow variables: share shadow data among KGraft modules
As it stands, each KGraft module maintains its own set of shadow variable
management structures and thus, shadow variables are not sharable between
livepatch modules.
This behaviour is different from the upstream implementation and, as
pointed out by Miroslav Benes, it also opens up an opportunity for a small
window where the system might become vulnerable again during transition as
we stack new livepatches on top.
Let all KGraft patches share the shadow data.
Sharing is implemented by moving the management structures from a KGraft
module's .data to dynamically allocated memory. Each KGraft module will
have specifically named pointers, 'kgr_shadow_hash12' and
'kgr_shadow_lock12', referencing them.
Upon initialization, a KGraft module will discover already existing such
shadow data by kallsyms-searching all loaded modules for these pointer
symbols. If none is found, a new instance is allocated. The newly
introduced kgr_shadow_init() implementing this is idempotent and can thus
be called from the bsc# subpatches' initializers if needed.
Upon KGraft module removal, the new kgr_shadow_cleanup() will conduct
another kallsyms search and deallocate the shadow data in case there are
no more users. kgr_shadow_cleanup() is also idempotent.
Initialization and teardown of the common shadow data is serialized with
the module_mutex which has to be taken for the kallsyms search anyway.
- commit 8e1e705
-------------------------------------------------------------------
Thu Oct 5 12:12:27 CEST 2017 - nstange@suse.de
- shadow variables: drop EXPORT_SYMBOL()s
The shadow variable API will only ever get used by the KGraft module itself
and thus, there's no need for exporting it.
Drop all EXPORT_SYMBOL annotations.
- commit ac6cfeb
-------------------------------------------------------------------
Thu Oct 5 12:12:26 CEST 2017 - nstange@suse.de
- shadow variables: introduce upstream patch
Joe Lawrence posted the sixth version of his shadow variable patch [1]
implementing the association of additional out-of-band data members to
existing structure instances from livepatches.
Jiri Kosina has applied this to his
git://git.kernel.org/pub/scm/linux/kernel/git/jikos/livepatching.git for-4.15/shadow-variables
tree and thus, it's queued up and close to getting merged.
The plan is to eventually backport this shadow variable support to SLE
kernels, but we also want to have it usable from KGraft modules by now.
Port the implementation to the kraft-patches module.
Namely,
- dump shadow.c in it's current upstream state as it is after commits
439e7271dc2b ("livepatch: introduce shadow variable API")
5d9da759f758 ("livepatch: __klp_shadow_get_or_alloc() is local to
shadow.c")
19205da6a0da ("livepatch: Small shadow variable documentation fixes")
- add a shadow.h header and declare the newly introduced functions there
- and incorporate the new files into the KGraft module's build system.
[1] 1504211861-19899-2-git-send-email-joe.lawrence@redhat.com
("[PATCH v6] livepatch: introduce shadow variable API")
- commit e899c4f
-------------------------------------------------------------------
Wed Jul 12 11:14:40 CEST 2017 - lpechacek@suse.com
- kGraft to livepatch migration. API change.
Change from kGraft API to livepatch API.
Note: error handling in _init() function is broken and fixed later.
Automatic generation of klp_objects is not present at all. Added later.
References: fate#323682
[ mb: changelog, patch split, whitespace errors ]
- commit 7e20201
-------------------------------------------------------------------
Wed Jul 12 11:08:57 CEST 2017 - lpechacek@suse.com
- kGraft to livepatch migration. Internal rename.
Internal rename in preparation for kGraft -> upstream livepatch
migration. External module naming stays the same. API is not touched
yet.
References: fate#323682
[ mb: changelog edit ]
- commit 28a04a2
-------------------------------------------------------------------
Tue Jun 13 15:54:27 CEST 2017 - nstange@suse.de
- scripts/register-patches.sh: register subpatch sources in rpm spec
In order to reduce the manual merging work upon addition of new
(sub)patches, commit 4e8dc885be22 ("scripts: create kgr_patch_main.c
dynamically") introduced the register-patches.sh helper. It discovers
those and tweaks the main entry point, kgr_patch_main.c, as needed.
However, a remaining manual merging task is to list a (sub)patch's source
archive in rpm/kgraft-patch.spec and to %setup it.
Make scripts/register-patches.sh do this.
Namely,
- introduce the @@KGR_PATCHES_SOURCES@@ and @@KGR_PATCHES_SETUP_SOURCES@@
placeholders in rpm/kgraft-patch.spec
- and make scripts/register-patches.sh expand those within a spec file
to be given as an additional command line argument.
Finally, adjust scripts/tar-up.sh accordingly.
- commit 9eafc8a
-------------------------------------------------------------------
Tue Jun 13 15:51:42 CEST 2017 - nstange@suse.de
- scripts/register-patches.sh: don't add ','s to @@KGR_PATCHES_FUNCS@@
register-patches.sh expands kgr_patch_main.c's @@KGR_PATCHES_FUNCS@@
placeholder by concatenating all available patches' KGR_PATCH_<XY>_FUNCS
together, separating them by commas.
The KGR_PATCH_<XY>_FUNCS are CPP macros supposed to be provided by each
patch. If one of these happens to be empty, the preprocessed expansion
will contain two consecutive commas which gcc doesn't like in array
initializers.
Do not add any commas to the @@KGR_PATCHES_FUNCS@@ expansion but require
the individual KGR_PATCH_<XY>_FUNCS macros to already contain trailing
ones as needed.
Fixes: 4e8dc885be22 ("scripts: create kgr_patch_main.c dynamically")
- commit ba41416
-------------------------------------------------------------------
Wed Jun 7 12:05:41 CEST 2017 - nstange@suse.de
- scripts: create kgr_patch_main.c dynamically
The kgraft-patches repository has got many branches, each corresponding
to a supported codestream. Each of those carries a potentially different
set of live (sub)patches which are controlled through the entry points in
kgr_patch_main.c. According to Miroslav, merging of a new (sub)patch
based on the pristine master is a pita due to conflicts.
Since all (sub)patches stick to certain conventions already, the required
modifications of the merging-hotspot kgr_patch_main.c are quite mechanic.
Let a script do the work.
Namely,
- insert some special @@-embraced placeholders at the few places depending
on the actual set of (sub)patches,
- let register-patches.sh discover the available (sub)patches by searching
for directories
- and let register-patches.sh replace those placeholders in
kgr_patch_main.c
Finally, add a register-patches.sh invocation to tar-up.sh.
This procedure requires that a SUBPATCH located in directory SUBPATCH/
adheres to the following conventions:
- It must provide a provide a SUBPATCH/kgr_patch_SUBPATCH.h header.
- This header must provide declarations for kgr_patch_SUBPATCH_init()
and kgr_patch_SUBPATCH_cleanup().
- This header must also #define a KGR_PATCH_SUBPATCH_FUNCS macro.
It should expand to a comma separated list of KGR_PATCH*() entries,
each corresponding to a function the subpatch wants to replace.
[mbenes: fixed typos, empty line removed]
- commit 4e8dc88
-------------------------------------------------------------------
Mon Apr 24 16:00:54 CEST 2017 - mbenes@suse.cz
- Replace $(PWD) with $(CURDIR) in Makefile
CURDIR is an internal variable of make and more suitable.
- commit 03bf1d5
-------------------------------------------------------------------
Wed Apr 19 14:02:27 CEST 2017 - mbenes@suse.cz
- Create Makefile automatically
Introduce scripts/create-makefile.sh script to automatically create a
makefile. The scripts is called from tar-up.sh or could be called
manually.
- commit 1af6c29
-------------------------------------------------------------------
Mon Oct 24 13:26:09 CEST 2016 - mbenes@suse.cz
- Better to use SUSE:SLE-12:Update than Devel:kGraft:SLE12 project
- commit bdc7598
-------------------------------------------------------------------
Tue May 10 15:43:59 CEST 2016 - mbenes@suse.cz
- Add compat.h to deal with changes of KGR_PATCH macro
Sympos patch set for kGraft redefined KGR_PATCH macro and added two new
ones. Add new compat.h which contains macro magic so that all kGraft
patches would work on both old and new kernels with the patch set
merged.
- commit 4186bef
-------------------------------------------------------------------
Fri May 6 17:01:17 CEST 2016 - mbenes@suse.cz
- Fix the number of parameters of KGR_PATCH macro
New kernels contain kGraft's sympos patch set which changed number of
paramaters of KGR_PATCH macro and introduced new macros. Fix it in
master so it will be ok for new branches.
- commit 78cf676
-------------------------------------------------------------------
Tue Sep 1 13:00:23 CEST 2015 - mmarek@suse.com
- Include the RPM version number in the module name
- commit 8fa02c6
-------------------------------------------------------------------
Wed Aug 26 11:29:44 CEST 2015 - mbenes@suse.cz
- Remove forgotten debug option in the Makefile
- commit 9c24ab8
-------------------------------------------------------------------
Mon Aug 17 13:42:04 CEST 2015 - mbenes@suse.cz
- Add license and copyright notices
- commit d42d3aa
-------------------------------------------------------------------
Wed Jul 15 15:58:35 CEST 2015 - mbenes@suse.cz
- Remove immediate flag
Fake signal was merged to kGraft and immediate feature removed. Remove
it in kGraft patches from now on too.
- commit c767ad2
-------------------------------------------------------------------
Wed May 20 16:32:17 CEST 2015 - mbenes@suse.cz
- Set immediate flag to false
Using immediate set to true can lead to BUGs and oopses when
downgrading, reverting or applying replace_all patches. There is no way
how to find out if there is a process in the old code which is being
removed. The module would be put, removed and the process will crash.
The consistency model guarantees that there is no one in the old code
when the finalization ends. Thus use it for all case to be safe.
- commit 830e1a3
-------------------------------------------------------------------
Tue May 12 15:48:07 CEST 2015 - mbenes@suse.cz
- Fix description in rpm spec file
Spec file description mentions initial kGraft patch which is only true
for real initial patch. Make it more neutral.
References: bsc#930408
- commit a55e023
-------------------------------------------------------------------
Wed Apr 1 15:36:24 CEST 2015 - mbenes@suse.cz
- Generate archives names automatically in tar-up.sh
- commit 1f34f18
-------------------------------------------------------------------
Wed Apr 1 13:39:26 CEST 2015 - mbenes@suse.cz
- Automatically generate .changes file from git log
Also add comments to tar-up.sh script to distinguish between sections.
- commit 212a7ae
-------------------------------------------------------------------
Thu Mar 26 14:24:21 CET 2015 - mmarek@suse.cz
- Revert "Require exact kernel version in the patch"
This needs to be done differently, so that modprobe --force works as
expected.
References: bnc#920615
This reverts commit c62c11aecd4e3f8822e1b835fea403acc3148c5a.
- commit bc88dd7
-------------------------------------------------------------------
Wed Mar 25 13:10:24 CET 2015 - mmarek@suse.cz
- Require exact kernel version in the patch
References: bnc#920615
- commit c62c11a
-------------------------------------------------------------------
Tue Mar 24 12:15:41 CET 2015 - mmarek@suse.cz
- Add the git commit and branch to the package description
References: bnc#920633
- commit 1ff4e48
-------------------------------------------------------------------
Wed Nov 26 10:09:14 CET 2014 - mbenes@suse.cz
- Set immediate flag for the initial patch
Setting immediate to true will simplify installation of the initial patch and
possibly also of the further updates.
References: bnc#907150
- commit 391b810
-------------------------------------------------------------------
Tue Nov 25 16:26:40 CET 2014 - mbenes@suse.cz
- Add .replace_all set to true
Add .replace_all flag set to true even to the initial patch. Thus we will not
forget to add that later. Also .immediate is there as a comment.
- commit 933e15e
-------------------------------------------------------------------
Mon Nov 24 15:02:33 CET 2014 - mmarek@suse.cz
- Drop the hardcoded kernel release string
The updated kgraft-devel macros set this during build time, so we do not
need to know the kernel release string beforehand. As a name suffix for
the source packages, let's use SLE12_Test in the master branch and
SLE12_Update_<n> in the update branches.
- commit 65f7a25
-------------------------------------------------------------------
Fri Nov 21 15:48:48 CET 2014 - mmarek@suse.cz
- Check that we are building against the set kernel version
- commit 689e44a
-------------------------------------------------------------------
Wed Nov 12 04:11:14 CET 2014 - mmarek@suse.cz
- Mark the module as supported
References: bnc#904970
- commit 6249314
-------------------------------------------------------------------
Tue Nov 11 17:11:28 CET 2014 - mmarek@suse.cz
- Build the test packages against Devel:kGraft:SLE12
- commit c952fbb
-------------------------------------------------------------------
Thu Nov 6 13:55:43 CET 2014 - mbenes@suse.cz
- Add top git commit hash to uname -v
Add top git commit hash to version part of uname. This makes the identification
of current patch level easy (even in crash: p kgr_tag).
References: fate#317769
- commit 54c9595
-------------------------------------------------------------------
Tue Nov 4 16:23:50 CET 2014 - mbenes@suse.cz
- Replace @@RELEASE@@ in kgr_patch->name with @@RPMRELEASE@@
We need to replace @@RELEASE@@ in kgr_patch->name with @@RPMRELEASE@@ due to
sysfs tree. @@RELEASE@@ changes with each new version of package.
- commit 51fd9dd
-------------------------------------------------------------------
Mon Nov 3 17:27:24 CET 2014 - mmarek@suse.cz
- Add a source-timestamp file with the git commit hash and branch
This is required by the bs-upload-kernel script to upload packages to
the BS. It can also be used by the specfile in the future.
- commit feab4f1
-------------------------------------------------------------------
Mon Nov 3 16:56:31 CET 2014 - mbenes@suse.cz
- Initial commit
- commit 600de9d
-------------------------------------------------------------------
Mon Nov 3 14:59:46 CET 2014 - mmarek@suse.cz
- Add config.sh script
This tells the automatic builder which IBS project to use.
- commit aa7f1cb
Reference in New Issue View Git Blame Copy Permalink