When the core module loops through an object's list of dynrelas, it
determines whether or not the target location of the dynrela is in a
read-only region of the patch module. If it is, the readonly flag is set to
1 and it calls set_memory_{rw,ro} before and after the probe_kernel_write()
operation. This flag gets set once, and never gets reset for subsequent
iterations. Therefore, if a target happens to be in a RW section of the
patch module, and readonly = 1 had been set before, we may unintentionally
set a normally RW page to RO. Fix this by setting the readonly flag with
each iteration of the loop.
Fixes#681.
Restructure kpatch's sysfs interface and mirror the sysfs tree after
livepatch's sysfs layout. With the current sysfs layout, we cannot
distinguish which object a function belongs to, and we cannot tell which
modules/objects are patched. Therefore, restructure the kpatch sysfs tree
such that module/object information is available. With the new layout, each
patched object has its own directory, with each function being a
subdirectory of its object.
Implement this by embedding a kobject struct within the kpatch_module,
kpatch_func, and kpatch_object structs and supplying their ktypes and
kobject release methods.
Before:
/sys/kernel/kpatch
└── patches
└── <patch_module>
├── checksum
├── enabled
└── functions
├── <function> # from <object1>
│ ├── new_addr
│ └── old_addr
├── <function> # from <object2>
│ ├── new_addr
│ └── old_addr
└─── <function> # from <object3>
├── new_addr
└── old_addr
After:
/sys/kernel/kpatch
└── <patch_module>
├── <object1>
│ └── <function,sympos>
│ ├── new_addr
│ └── old_addr
├── <object2>
│ └── <function,sympos>
│ ├── new_addr
│ └── old_addr
├── checksum
├── enabled
└── <object3>
└── <function,sympos>
├── new_addr
└── old_addr
Upstream 2992ef29ae01 "livepatch/module: make TAINT_LIVEPATCH module-specific"
added a TAINT_LIVEPATCH flag to the module-specific taint flags. This
commit is v4.9+ and the modules taint field is an unsigned int.
Upstream 7fd8329ba502 "taint/module: Clean up global and module taint
flags handling" modified the modules taint field to be an unsigned long.
This commit is v4.10+.
Adjust the module tainting code in kpatch_register() to consider v4.9
kernels as well as v4.10 (and any distro-specific behavior).
Fixes: #666.
The dump_trace interface was deprecated in v4.9: instead of adding yet
another kernel-specific code block to kpatch's stack safety checks, use
save_stack_trace_tsk. It's relatively simple (no callbacks like
dump_trace), arch-independent, and its interface is stable across kernel
releases.
Fixes: #623.
Previous commit "kmod: let kernel apply TAINT_LIVEPATCH" modified the
kpatch patch module to set the "livepatch" module info. This breaks
module loading for kernel config CONFIG_LIVEPATCH=n
kpatch_kmalloc: module is marked as livepatch module, but livepatch support is disabled
kpatch modules can still use TAINT_LIVEPATCH as a per-module taint flag,
but only if it is set after the module loads.
Fixes: 660.
Introduce a second phase in the kpatch-build process that creates kpatch
modules or livepatch modules that use the new klp rela sections depending on
the kernel version being worked on. This change uses the two new programs to
either create a patch module that uses dynrelas (create-kpatch-module) or a
patch module that uses klp rela and arch sections + klp symbols marked with the
correct Elf flags (create-klp-module).
For klp patch modules, the --unique flag for ld is needed to prevent
.parainstructions and .altinstructions sections from different objects
from being merged, as arch_klp_init_object_loaded() applies these sections
per-object.
Upstream commit 2992ef29ae01 ("livepatch/module: make TAINT_LIVEPATCH
module-specific") v4.9+ modified the kernel to add the TAINT_LIVEPATCH
flag on module load. To support this feature, add the "livepatch"
module info in the {k,live}patch modules and drop the add_taint() in the
core module.
Fixes smatch warning:
kmod/core/core.c:64:1: warning: symbol 'kpmod_list' was not declared. Should it be static?
Fixes sparse warnings:
kmod/core/core.c:680 kpatch_write_relocations() warn: inconsistent indenting
kmod/core/core.c:750 kpatch_write_relocations() warn: inconsistent indenting
The error paths in kpatch_shadow_alloc do not free an allocated
kpatch_shadow structure (and it's not added to the kpatch_shadow_hash).
Handle the kfree in the various error return paths.
Fixes the smatch warning:
kmod/core/shadow.c:97 kpatch_shadow_alloc() warn: possible memory leak of 'shadow'
Some features were backported into the 4.4 kernel which change the fields
of the livepatch structures. Ensure we can work with either v4.5 or greater,
or Ubuntu 4.4.0-7 or greater.
If an activeness safety check fails for kernels newer than 4.6, the
error is silently ignored because the newer version of
kpatch_backtrace_address_verify() doesn't set args.ret on error.
It would be an easy fix to just set args->ret on error, but I think a
better approach is just to combine the two versions of the function into
a single function with the use of a little macro trickery.
Backport the symbol lookup and checking code from upstream livepatch
code that relies on a symbol position enumeration rather than a fixed
memory address.
Fixes#617.
ftrace only allows a single user of this flag to register for a given
function. This prevents kpatch conflicts with kprobes handlers which
also might want to change regs->ip for a function.
We should have done this a few years ago. Better late than never...
When building against a different kernel from the one that is currently
running, install the kernel module to a directory based on the version
of the target kernel rather than the currently running kernel.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Upstream commit 568b329a "perf: generalize perf_callchain" modified the
return type (void -> int) of the address member of struct stacktrace_ops.
Use the void function if the kernel version is < 4.6 or return an int
otherwise.
When a patch module is loaded, the kernel facilities like alternatives
and paravirt may alter some of its instructions. This happens before
Kpatch core module is notified and tries to apply dynrelas to it. If an
instruction to apply a dynrela to has already been changed by these
facilities, an incorrect instruction might be written as a result.
The core module now detects such conditions and does not apply dynrela
to the changed instructions.
Suggested by Josh Poimboeuf in the discussion of
https://github.com/dynup/kpatch/issues/580.
Changes in v.2:
* Used pr_notice to give more emphasis to the messages.
* Added an explanation message.
Signed-off-by: Evgenii Shatokhin <eshatokhin@virtuozzo.com>
Commit 7523e4dc5057 upstream ("module: use a structure to encapsulate
layout") uses a new field to access module memory. Account for this change
and ensure backwards compatibility with kernel versions < 4.5
Support patching objects that have duplicated function names. This feature was
introduced upstream in Linux v4.5.
This patch appends the symbol position to the symbol structure when
lookup_local_symbol is called. This pos variable is then used when creating the
funcs and dynrelas sections. Finally, incorporate sympos into the livepatch
patch hook only if the kernel version is greater than v4.5. In other cases the
older format is used.
Fixes: #493
Signed-off-by: Chris J Arges <chris.j.arges@canonical.com>
In some cases when packaging it may not be useful to build kmod/core at
package build time (for example if using DKMS). Add a parameter 'BUILDMOD'
that when set to 'yes' will build kmod/core.
Signed-off-by: Chris J Arges <chris.j.arges@canonical.com>
Don't assume we are building for the current kernel. In addition print out
a proper package necessary for building the module.
Signed-off-by: Chris J Arges <chris.j.arges@canonical.com>
Fixes issue #494. A null pointer dereference can result with patch
modules for multiple objects since the "vmlinux" patch object's "name"
field is null. strcmp therefore crashes trying to compare object->name
if the current object is vmlinux and the supplied "name" argument is
not. Check that object->name is not null before invoking strcmp.
The shadow_get function does't consider the case that
'shadow is inpace', and after the shadow->data be set to the data,
it will not be the pointer. This patch fix it.
Signed-off-by: Li Bin <huawei.libin@huawei.com>
Adds a new patch module scaffold for use when building against a kernel
with CONFIG_LIVE_PATCHING=y.
Signed-off-by: Seth Jennings <sjenning@redhat.com>
Unload of kpatch module (and kpatch_shadow_hash table) before
all shadow variables free requests are processed can lead to
kernel crash.
Add rcu_barrier() to kpatch_exit() to wait for all outstanding
RCU callbacks to complete.
Signed-off-by: Jan Stancek <jstancek@redhat.com>
When patching a kernel module, if we can't find a needed dynrela symbol,
we currently assume it's exported. However, it's also possible that
it's provided by another .o in the patch module. Add support for that.
Fixes#445.
Fix the object unlink error handling so that each function cleans up
after itself properly.
Also use find_symbol() instead of __symbol_get() to make cleanup easier.
When patching a module we don't need a reference to each symbol, since
we already have done a try_module_get() on the module.
Fixes#392.
Make the kpatch-patch-hook.c function and variable names consistent by
prefixing them all with 'patch_'. This makes it easier to distinguish
the patch hook sections from the patched sections when looking at the
ELF section data.
In order to safely re-enable patch modules, add a special
.kpatch.checksum section containing an md5sum of a patch module's
contents. The contents of this section are exported to sysfs via
patch_init and double checked when kpatch load finds that a module of
the same name is already loaded.
This adds support for shadow variables, which allow you to add new
"shadow" fields to existing data structures.
To allow patches to call the shadow functions in the core module, I had
to add a funky hack to use --warn-unresolved-symbols when linking, which
allows the patched vmlinux to link with the missing symbols. I also
added greps to the log file to ensure that only unresolved symbols to
kpatch_shadow_* are allowed. We can remove this hack once the core
module gets moved into the kernel tree.
Fixes#314.
To reduce redundancy, remove/change the old_offset fields in the
kpatch_func and kpatch_patch_func structs to just old_addr. Since
old_offset is being used as a placeholder for old_addr, might as well
consolidate it to just one variable.
Fix incorrect old_offsets for loadable modules during sysfs
initialization in patch_init.
sysfs will be initialized on patch module init regardless of whether
or not the module is loaded. func_old_addr_show() will read from func->old_addr,
which is initially set to 0; it'll be eventually filled in by the core module.
Use this instead of calling printk to avoid unwanted compiler
optimizations which cause kpatch-build errors.
The printk function is annotated with the __cold attribute, which tells
gcc that the function is unlikely to be called. A side effect of this
is that code paths containing calls to printk might also be marked cold,
leading to other functions called in those code paths getting moved into
.text.unlikely or being uninlined.
This macro places printk in its own code path so as not to make the
surrounding code path cold.
I have a related integration test to add, but right now it's broken
because we don't yet properly support the __verbose special section.
That'll be another PR.
Fixes#296.
