5 This document outlines basic information about kernel livepatching.
10 2. Kprobes, Ftrace, Livepatching
15 4.3. Livepatch module handling
16 5. Livepatch life-cycle
28 There are many situations where users are reluctant to reboot a system. It may
29 be because their system is performing complex scientific computations or under
30 heavy load during peak usage. In addition to keeping systems up and running,
31 users want to also have a stable and secure system. Livepatching gives users
32 both by allowing for function calls to be redirected; thus, fixing critical
33 functions without a system reboot.
36 2. Kprobes, Ftrace, Livepatching
37 ================================
39 There are multiple mechanisms in the Linux kernel that are directly related
40 to redirection of code execution; namely: kernel probes, function tracing,
43 + The kernel probes are the most generic. The code can be redirected by
44 putting a breakpoint instruction instead of any instruction.
46 + The function tracer calls the code from a predefined location that is
47 close to the function entry point. This location is generated by the
48 compiler using the '-pg' gcc option.
50 + Livepatching typically needs to redirect the code at the very beginning
51 of the function entry before the function parameters or the stack
52 are in any way modified.
54 All three approaches need to modify the existing code at runtime. Therefore
55 they need to be aware of each other and not step over each other's toes.
56 Most of these problems are solved by using the dynamic ftrace framework as
57 a base. A Kprobe is registered as a ftrace handler when the function entry
58 is probed, see CONFIG_KPROBES_ON_FTRACE. Also an alternative function from
59 a live patch is called with the help of a custom ftrace handler. But there are
60 some limitations, see below.
66 Functions are there for a reason. They take some input parameters, get or
67 release locks, read, process, and even write some data in a defined way,
68 have return values. In other words, each function has a defined semantic.
70 Many fixes do not change the semantic of the modified functions. For
71 example, they add a NULL pointer or a boundary check, fix a race by adding
72 a missing memory barrier, or add some locking around a critical section.
73 Most of these changes are self contained and the function presents itself
74 the same way to the rest of the system. In this case, the functions might
75 be updated independently one by one. (This can be done by setting the
76 'immediate' flag in the klp_patch struct.)
78 But there are more complex fixes. For example, a patch might change
79 ordering of locking in multiple functions at the same time. Or a patch
80 might exchange meaning of some temporary structures and update
81 all the relevant functions. In this case, the affected unit
82 (thread, whole kernel) need to start using all new versions of
83 the functions at the same time. Also the switch must happen only
84 when it is safe to do so, e.g. when the affected locks are released
85 or no data are stored in the modified structures at the moment.
87 The theory about how to apply functions a safe way is rather complex.
88 The aim is to define a so-called consistency model. It attempts to define
89 conditions when the new implementation could be used so that the system
92 Livepatch has a consistency model which is a hybrid of kGraft and
93 kpatch: it uses kGraft's per-task consistency and syscall barrier
94 switching combined with kpatch's stack trace switching. There are also
95 a number of fallback options which make it quite flexible.
97 Patches are applied on a per-task basis, when the task is deemed safe to
98 switch over. When a patch is enabled, livepatch enters into a
99 transition state where tasks are converging to the patched state.
100 Usually this transition state can complete in a few seconds. The same
101 sequence occurs when a patch is disabled, except the tasks converge from
102 the patched state to the unpatched state.
104 An interrupt handler inherits the patched state of the task it
105 interrupts. The same is true for forked tasks: the child inherits the
106 patched state of the parent.
108 Livepatch uses several complementary approaches to determine when it's
111 1. The first and most effective approach is stack checking of sleeping
112 tasks. If no affected functions are on the stack of a given task,
113 the task is patched. In most cases this will patch most or all of
114 the tasks on the first try. Otherwise it'll keep trying
115 periodically. This option is only available if the architecture has
116 reliable stacks (HAVE_RELIABLE_STACKTRACE).
118 2. The second approach, if needed, is kernel exit switching. A
119 task is switched when it returns to user space from a system call, a
120 user space IRQ, or a signal. It's useful in the following cases:
122 a) Patching I/O-bound user tasks which are sleeping on an affected
123 function. In this case you have to send SIGSTOP and SIGCONT to
124 force it to exit the kernel and be patched.
125 b) Patching CPU-bound user tasks. If the task is highly CPU-bound
126 then it will get patched the next time it gets interrupted by an
128 c) In the future it could be useful for applying patches for
129 architectures which don't yet have HAVE_RELIABLE_STACKTRACE. In
130 this case you would have to signal most of the tasks on the
131 system. However this isn't supported yet because there's
132 currently no way to patch kthreads without
133 HAVE_RELIABLE_STACKTRACE.
135 3. For idle "swapper" tasks, since they don't ever exit the kernel, they
136 instead have a klp_update_patch_state() call in the idle loop which
137 allows them to be patched before the CPU enters the idle state.
139 (Note there's not yet such an approach for kthreads.)
141 All the above approaches may be skipped by setting the 'immediate' flag
142 in the 'klp_patch' struct, which will disable per-task consistency and
143 patch all tasks immediately. This can be useful if the patch doesn't
144 change any function or data semantics. Note that, even with this flag
145 set, it's possible that some tasks may still be running with an old
146 version of the function, until that function returns.
148 There's also an 'immediate' flag in the 'klp_func' struct which allows
149 you to specify that certain functions in the patch can be applied
150 without per-task consistency. This might be useful if you want to patch
151 a common function like schedule(), and the function change doesn't need
152 consistency but the rest of the patch does.
154 For architectures which don't have HAVE_RELIABLE_STACKTRACE, the user
155 must set patch->immediate which causes all tasks to be patched
156 immediately. This option should be used with care, only when the patch
157 doesn't change any function or data semantics.
159 In the future, architectures which don't have HAVE_RELIABLE_STACKTRACE
160 may be allowed to use per-task consistency if we can come up with
161 another way to patch kthreads.
163 The /sys/kernel/livepatch/<patch>/transition file shows whether a patch
164 is in transition. Only a single patch (the topmost patch on the stack)
165 can be in transition at a given time. A patch can remain in transition
166 indefinitely, if any of the tasks are stuck in the initial patch state.
168 A transition can be reversed and effectively canceled by writing the
169 opposite value to the /sys/kernel/livepatch/<patch>/enabled file while
170 the transition is in progress. Then all the tasks will attempt to
171 converge back to the original patch state.
173 There's also a /proc/<pid>/patch_state file which can be used to
174 determine which tasks are blocking completion of a patching operation.
175 If a patch is in transition, this file shows 0 to indicate the task is
176 unpatched and 1 to indicate it's patched. Otherwise, if no patch is in
177 transition, it shows -1. Any tasks which are blocking the transition
178 can be signaled with SIGSTOP and SIGCONT to force them to change their
182 3.1 Adding consistency model support to new architectures
183 ---------------------------------------------------------
185 For adding consistency model support to new architectures, there are a
188 1) Add CONFIG_HAVE_RELIABLE_STACKTRACE. This means porting objtool, and
189 for non-DWARF unwinders, also making sure there's a way for the stack
190 tracing code to detect interrupts on the stack.
192 2) Alternatively, ensure that every kthread has a call to
193 klp_update_patch_state() in a safe location. Kthreads are typically
194 in an infinite loop which does some action repeatedly. The safe
195 location to switch the kthread's patch state would be at a designated
196 point in the loop where there are no locks taken and all data
197 structures are in a well-defined state.
199 The location is clear when using workqueues or the kthread worker
200 API. These kthreads process independent actions in a generic loop.
202 It's much more complicated with kthreads which have a custom loop.
203 There the safe location must be carefully selected on a case-by-case
206 In that case, arches without HAVE_RELIABLE_STACKTRACE would still be
207 able to use the non-stack-checking parts of the consistency model:
209 a) patching user tasks when they cross the kernel/user space
212 b) patching kthreads and idle tasks at their designated patch points.
214 This option isn't as good as option 1 because it requires signaling
215 user tasks and waking kthreads to patch them. But it could still be
216 a good backup option for those architectures which don't have
217 reliable stack traces yet.
219 In the meantime, patches for such architectures can bypass the
220 consistency model by setting klp_patch.immediate to true. This option
221 is perfectly fine for patches which don't change the semantics of the
222 patched functions. In practice, this is usable for ~90% of security
223 fixes. Use of this option also means the patch can't be unloaded after
224 it has been disabled.
230 Livepatches are distributed using kernel modules, see
231 samples/livepatch/livepatch-sample.c.
233 The module includes a new implementation of functions that we want
234 to replace. In addition, it defines some structures describing the
235 relation between the original and the new implementation. Then there
236 is code that makes the kernel start using the new code when the livepatch
237 module is loaded. Also there is code that cleans up before the
238 livepatch module is removed. All this is explained in more details in
245 New versions of functions are typically just copied from the original
246 sources. A good practice is to add a prefix to the names so that they
247 can be distinguished from the original ones, e.g. in a backtrace. Also
248 they can be declared as static because they are not called directly
249 and do not need the global visibility.
251 The patch contains only functions that are really modified. But they
252 might want to access functions or data from the original source file
253 that may only be locally accessible. This can be solved by a special
254 relocation section in the generated livepatch module, see
255 Documentation/livepatch/module-elf-format.txt for more details.
261 The patch is described by several structures that split the information
264 + struct klp_func is defined for each patched function. It describes
265 the relation between the original and the new implementation of a
268 The structure includes the name, as a string, of the original function.
269 The function address is found via kallsyms at runtime.
271 Then it includes the address of the new function. It is defined
272 directly by assigning the function pointer. Note that the new
273 function is typically defined in the same source file.
275 As an optional parameter, the symbol position in the kallsyms database can
276 be used to disambiguate functions of the same name. This is not the
277 absolute position in the database, but rather the order it has been found
278 only for a particular object ( vmlinux or a kernel module ). Note that
279 kallsyms allows for searching symbols according to the object name.
281 There's also an 'immediate' flag which, when set, patches the
282 function immediately, bypassing the consistency model safety checks.
284 + struct klp_object defines an array of patched functions (struct
285 klp_func) in the same object. Where the object is either vmlinux
286 (NULL) or a module name.
288 The structure helps to group and handle functions for each object
289 together. Note that patched modules might be loaded later than
290 the patch itself and the relevant functions might be patched
291 only when they are available.
294 + struct klp_patch defines an array of patched objects (struct
297 This structure handles all patched functions consistently and eventually,
298 synchronously. The whole patch is applied only when all patched
299 symbols are found. The only exception are symbols from objects
300 (kernel modules) that have not been loaded yet.
302 Setting the 'immediate' flag applies the patch to all tasks
303 immediately, bypassing the consistency model safety checks.
305 For more details on how the patch is applied on a per-task basis,
306 see the "Consistency model" section.
309 4.3. Livepatch module handling
310 ------------------------------
312 The usual behavior is that the new functions will get used when
313 the livepatch module is loaded. For this, the module init() function
314 has to register the patch (struct klp_patch) and enable it. See the
315 section "Livepatch life-cycle" below for more details about these
318 Module removal is only safe when there are no users of the underlying
319 functions. The immediate consistency model is not able to detect this. The
320 code just redirects the functions at the very beginning and it does not
321 check if the functions are in use. In other words, it knows when the
322 functions get called but it does not know when the functions return.
323 Therefore it cannot be decided when the livepatch module can be safely
324 removed. This is solved by a hybrid consistency model. When the system is
325 transitioned to a new patch state (patched/unpatched) it is guaranteed that
326 no task sleeps or runs in the old code.
329 5. Livepatch life-cycle
330 =======================
332 Livepatching defines four basic operations that define the life cycle of each
333 live patch: registration, enabling, disabling and unregistration. There are
334 several reasons why it is done this way.
336 First, the patch is applied only when all patched symbols for already
337 loaded objects are found. The error handling is much easier if this
338 check is done before particular functions get redirected.
340 Second, the immediate consistency model does not guarantee that anyone is not
341 sleeping in the new code after the patch is reverted. This means that the new
342 code needs to stay around "forever". If the code is there, one could apply it
343 again. Therefore it makes sense to separate the operations that might be done
344 once and those that need to be repeated when the patch is enabled (applied)
347 Third, it might take some time until the entire system is migrated
348 when a more complex consistency model is used. The patch revert might
349 block the livepatch module removal for too long. Therefore it is useful
350 to revert the patch using a separate operation that might be called
351 explicitly. But it does not make sense to remove all information
352 until the livepatch module is really removed.
358 Each patch first has to be registered using klp_register_patch(). This makes
359 the patch known to the livepatch framework. Also it does some preliminary
360 computing and checks.
362 In particular, the patch is added into the list of known patches. The
363 addresses of the patched functions are found according to their names.
364 The special relocations, mentioned in the section "New functions", are
365 applied. The relevant entries are created under
366 /sys/kernel/livepatch/<name>. The patch is rejected when any operation
373 Registered patches might be enabled either by calling klp_enable_patch() or
374 by writing '1' to /sys/kernel/livepatch/<name>/enabled. The system will
375 start using the new implementation of the patched functions at this stage.
377 When a patch is enabled, livepatch enters into a transition state where
378 tasks are converging to the patched state. This is indicated by a value
379 of '1' in /sys/kernel/livepatch/<name>/transition. Once all tasks have
380 been patched, the 'transition' value changes to '0'. For more
381 information about this process, see the "Consistency model" section.
383 If an original function is patched for the first time, a function
384 specific struct klp_ops is created and an universal ftrace handler is
387 Functions might be patched multiple times. The ftrace handler is registered
388 only once for the given function. Further patches just add an entry to the
389 list (see field `func_stack`) of the struct klp_ops. The last added
390 entry is chosen by the ftrace handler and becomes the active function
393 Note that the patches might be enabled in a different order than they were
400 Enabled patches might get disabled either by calling klp_disable_patch() or
401 by writing '0' to /sys/kernel/livepatch/<name>/enabled. At this stage
402 either the code from the previously enabled patch or even the original
405 When a patch is disabled, livepatch enters into a transition state where
406 tasks are converging to the unpatched state. This is indicated by a
407 value of '1' in /sys/kernel/livepatch/<name>/transition. Once all tasks
408 have been unpatched, the 'transition' value changes to '0'. For more
409 information about this process, see the "Consistency model" section.
411 Here all the functions (struct klp_func) associated with the to-be-disabled
412 patch are removed from the corresponding struct klp_ops. The ftrace handler
413 is unregistered and the struct klp_ops is freed when the func_stack list
416 Patches must be disabled in exactly the reverse order in which they were
417 enabled. It makes the problem and the implementation much easier.
423 Disabled patches might be unregistered by calling klp_unregister_patch().
424 This can be done only when the patch is disabled and the code is no longer
425 used. It must be called before the livepatch module gets unloaded.
427 At this stage, all the relevant sys-fs entries are removed and the patch
428 is removed from the list of known patches.
434 Information about the registered patches can be found under
435 /sys/kernel/livepatch. The patches could be enabled and disabled
438 See Documentation/ABI/testing/sysfs-kernel-livepatch for more details.
444 The current Livepatch implementation has several limitations:
447 + The patch must not change the semantic of the patched functions.
449 The current implementation guarantees only that either the old
450 or the new function is called. The functions are patched one
451 by one. It means that the patch must _not_ change the semantic
455 + Data structures can not be patched.
457 There is no support to version data structures or anyhow migrate
458 one structure into another. Also the simple consistency model does
459 not allow to switch more functions atomically.
461 Once there is more complex consistency mode, it will be possible to
462 use some workarounds. For example, it will be possible to use a hole
463 for a new member because the data structure is aligned. Or it will
464 be possible to use an existing member for something else.
466 There are no plans to add more generic support for modified structures
470 + Only functions that can be traced could be patched.
472 Livepatch is based on the dynamic ftrace. In particular, functions
473 implementing ftrace or the livepatch ftrace handler could not be
474 patched. Otherwise, the code would end up in an infinite loop. A
475 potential mistake is prevented by marking the problematic functions
480 + Livepatch works reliably only when the dynamic ftrace is located at
481 the very beginning of the function.
483 The function need to be redirected before the stack or the function
484 parameters are modified in any way. For example, livepatch requires
485 using -fentry gcc compiler option on x86_64.
487 One exception is the PPC port. It uses relative addressing and TOC.
488 Each function has to handle TOC and save LR before it could call
489 the ftrace handler. This operation has to be reverted on return.
490 Fortunately, the generic ftrace code has the same problem and all
491 this is handled on the ftrace level.
494 + Kretprobes using the ftrace framework conflict with the patched
497 Both kretprobes and livepatches use a ftrace handler that modifies
498 the return address. The first user wins. Either the probe or the patch
499 is rejected when the handler is already in use by the other.
502 + Kprobes in the original function are ignored when the code is
503 redirected to the new implementation.
505 There is a work in progress to add warnings about this situation.
git.samba.org / sfrench / cifs-2.6.git / blob