449 lines · plain
1=========2Livepatch3=========4 5This document outlines basic information about kernel livepatching.6 7.. Table of Contents:8 9.. contents:: :local:10 11 121. Motivation13=============14 15There are many situations where users are reluctant to reboot a system. It may16be because their system is performing complex scientific computations or under17heavy load during peak usage. In addition to keeping systems up and running,18users want to also have a stable and secure system. Livepatching gives users19both by allowing for function calls to be redirected; thus, fixing critical20functions without a system reboot.21 22 232. Kprobes, Ftrace, Livepatching24================================25 26There are multiple mechanisms in the Linux kernel that are directly related27to redirection of code execution; namely: kernel probes, function tracing,28and livepatching:29 30 - The kernel probes are the most generic. The code can be redirected by31 putting a breakpoint instruction instead of any instruction.32 33 - The function tracer calls the code from a predefined location that is34 close to the function entry point. This location is generated by the35 compiler using the '-pg' gcc option.36 37 - Livepatching typically needs to redirect the code at the very beginning38 of the function entry before the function parameters or the stack39 are in any way modified.40 41All three approaches need to modify the existing code at runtime. Therefore42they need to be aware of each other and not step over each other's toes.43Most of these problems are solved by using the dynamic ftrace framework as44a base. A Kprobe is registered as a ftrace handler when the function entry45is probed, see CONFIG_KPROBES_ON_FTRACE. Also an alternative function from46a live patch is called with the help of a custom ftrace handler. But there are47some limitations, see below.48 49 503. Consistency model51====================52 53Functions are there for a reason. They take some input parameters, acquire or54release locks, read, process, and even write some data in a defined way,55have return values. In other words, each function has a defined semantic.56 57Many fixes do not change the semantic of the modified functions. For58example, they add a NULL pointer or a boundary check, fix a race by adding59a missing memory barrier, or add some locking around a critical section.60Most of these changes are self contained and the function presents itself61the same way to the rest of the system. In this case, the functions might62be updated independently one by one.63 64But there are more complex fixes. For example, a patch might change65ordering of locking in multiple functions at the same time. Or a patch66might exchange meaning of some temporary structures and update67all the relevant functions. In this case, the affected unit68(thread, whole kernel) need to start using all new versions of69the functions at the same time. Also the switch must happen only70when it is safe to do so, e.g. when the affected locks are released71or no data are stored in the modified structures at the moment.72 73The theory about how to apply functions a safe way is rather complex.74The aim is to define a so-called consistency model. It attempts to define75conditions when the new implementation could be used so that the system76stays consistent.77 78Livepatch has a consistency model which is a hybrid of kGraft and79kpatch: it uses kGraft's per-task consistency and syscall barrier80switching combined with kpatch's stack trace switching. There are also81a number of fallback options which make it quite flexible.82 83Patches are applied on a per-task basis, when the task is deemed safe to84switch over. When a patch is enabled, livepatch enters into a85transition state where tasks are converging to the patched state.86Usually this transition state can complete in a few seconds. The same87sequence occurs when a patch is disabled, except the tasks converge from88the patched state to the unpatched state.89 90An interrupt handler inherits the patched state of the task it91interrupts. The same is true for forked tasks: the child inherits the92patched state of the parent.93 94Livepatch uses several complementary approaches to determine when it's95safe to patch tasks:96 971. The first and most effective approach is stack checking of sleeping98 tasks. If no affected functions are on the stack of a given task,99 the task is patched. In most cases this will patch most or all of100 the tasks on the first try. Otherwise it'll keep trying101 periodically. This option is only available if the architecture has102 reliable stacks (HAVE_RELIABLE_STACKTRACE).103 1042. The second approach, if needed, is kernel exit switching. A105 task is switched when it returns to user space from a system call, a106 user space IRQ, or a signal. It's useful in the following cases:107 108 a) Patching I/O-bound user tasks which are sleeping on an affected109 function. In this case you have to send SIGSTOP and SIGCONT to110 force it to exit the kernel and be patched.111 b) Patching CPU-bound user tasks. If the task is highly CPU-bound112 then it will get patched the next time it gets interrupted by an113 IRQ.114 1153. For idle "swapper" tasks, since they don't ever exit the kernel, they116 instead have a klp_update_patch_state() call in the idle loop which117 allows them to be patched before the CPU enters the idle state.118 119 (Note there's not yet such an approach for kthreads.)120 121Architectures which don't have HAVE_RELIABLE_STACKTRACE solely rely on122the second approach. It's highly likely that some tasks may still be123running with an old version of the function, until that function124returns. In this case you would have to signal the tasks. This125especially applies to kthreads. They may not be woken up and would need126to be forced. See below for more information.127 128Unless we can come up with another way to patch kthreads, architectures129without HAVE_RELIABLE_STACKTRACE are not considered fully supported by130the kernel livepatching.131 132The /sys/kernel/livepatch/<patch>/transition file shows whether a patch133is in transition. Only a single patch can be in transition at a given134time. A patch can remain in transition indefinitely, if any of the tasks135are stuck in the initial patch state.136 137A transition can be reversed and effectively canceled by writing the138opposite value to the /sys/kernel/livepatch/<patch>/enabled file while139the transition is in progress. Then all the tasks will attempt to140converge back to the original patch state.141 142There's also a /proc/<pid>/patch_state file which can be used to143determine which tasks are blocking completion of a patching operation.144If a patch is in transition, this file shows 0 to indicate the task is145unpatched and 1 to indicate it's patched. Otherwise, if no patch is in146transition, it shows -1. Any tasks which are blocking the transition147can be signaled with SIGSTOP and SIGCONT to force them to change their148patched state. This may be harmful to the system though. Sending a fake signal149to all remaining blocking tasks is a better alternative. No proper signal is150actually delivered (there is no data in signal pending structures). Tasks are151interrupted or woken up, and forced to change their patched state. The fake152signal is automatically sent every 15 seconds.153 154Administrator can also affect a transition through155/sys/kernel/livepatch/<patch>/force attribute. Writing 1 there clears156TIF_PATCH_PENDING flag of all tasks and thus forces the tasks to the patched157state. Important note! The force attribute is intended for cases when the158transition gets stuck for a long time because of a blocking task. Administrator159is expected to collect all necessary data (namely stack traces of such blocking160tasks) and request a clearance from a patch distributor to force the transition.161Unauthorized usage may cause harm to the system. It depends on the nature of the162patch, which functions are (un)patched, and which functions the blocking tasks163are sleeping in (/proc/<pid>/stack may help here). Removal (rmmod) of patch164modules is permanently disabled when the force feature is used. It cannot be165guaranteed there is no task sleeping in such module. It implies unbounded166reference count if a patch module is disabled and enabled in a loop.167 168Moreover, the usage of force may also affect future applications of live169patches and cause even more harm to the system. Administrator should first170consider to simply cancel a transition (see above). If force is used, reboot171should be planned and no more live patches applied.172 1733.1 Adding consistency model support to new architectures174---------------------------------------------------------175 176For adding consistency model support to new architectures, there are a177few options:178 1791) Add CONFIG_HAVE_RELIABLE_STACKTRACE. This means porting objtool, and180 for non-DWARF unwinders, also making sure there's a way for the stack181 tracing code to detect interrupts on the stack.182 1832) Alternatively, ensure that every kthread has a call to184 klp_update_patch_state() in a safe location. Kthreads are typically185 in an infinite loop which does some action repeatedly. The safe186 location to switch the kthread's patch state would be at a designated187 point in the loop where there are no locks taken and all data188 structures are in a well-defined state.189 190 The location is clear when using workqueues or the kthread worker191 API. These kthreads process independent actions in a generic loop.192 193 It's much more complicated with kthreads which have a custom loop.194 There the safe location must be carefully selected on a case-by-case195 basis.196 197 In that case, arches without HAVE_RELIABLE_STACKTRACE would still be198 able to use the non-stack-checking parts of the consistency model:199 200 a) patching user tasks when they cross the kernel/user space201 boundary; and202 203 b) patching kthreads and idle tasks at their designated patch points.204 205 This option isn't as good as option 1 because it requires signaling206 user tasks and waking kthreads to patch them. But it could still be207 a good backup option for those architectures which don't have208 reliable stack traces yet.209 210 2114. Livepatch module212===================213 214Livepatches are distributed using kernel modules, see215samples/livepatch/livepatch-sample.c.216 217The module includes a new implementation of functions that we want218to replace. In addition, it defines some structures describing the219relation between the original and the new implementation. Then there220is code that makes the kernel start using the new code when the livepatch221module is loaded. Also there is code that cleans up before the222livepatch module is removed. All this is explained in more details in223the next sections.224 225 2264.1. New functions227------------------228 229New versions of functions are typically just copied from the original230sources. A good practice is to add a prefix to the names so that they231can be distinguished from the original ones, e.g. in a backtrace. Also232they can be declared as static because they are not called directly233and do not need the global visibility.234 235The patch contains only functions that are really modified. But they236might want to access functions or data from the original source file237that may only be locally accessible. This can be solved by a special238relocation section in the generated livepatch module, see239Documentation/livepatch/module-elf-format.rst for more details.240 241 2424.2. Metadata243-------------244 245The patch is described by several structures that split the information246into three levels:247 248 - struct klp_func is defined for each patched function. It describes249 the relation between the original and the new implementation of a250 particular function.251 252 The structure includes the name, as a string, of the original function.253 The function address is found via kallsyms at runtime.254 255 Then it includes the address of the new function. It is defined256 directly by assigning the function pointer. Note that the new257 function is typically defined in the same source file.258 259 As an optional parameter, the symbol position in the kallsyms database can260 be used to disambiguate functions of the same name. This is not the261 absolute position in the database, but rather the order it has been found262 only for a particular object ( vmlinux or a kernel module ). Note that263 kallsyms allows for searching symbols according to the object name.264 265 - struct klp_object defines an array of patched functions (struct266 klp_func) in the same object. Where the object is either vmlinux267 (NULL) or a module name.268 269 The structure helps to group and handle functions for each object270 together. Note that patched modules might be loaded later than271 the patch itself and the relevant functions might be patched272 only when they are available.273 274 275 - struct klp_patch defines an array of patched objects (struct276 klp_object).277 278 This structure handles all patched functions consistently and eventually,279 synchronously. The whole patch is applied only when all patched280 symbols are found. The only exception are symbols from objects281 (kernel modules) that have not been loaded yet.282 283 For more details on how the patch is applied on a per-task basis,284 see the "Consistency model" section.285 286 2875. Livepatch life-cycle288=======================289 290Livepatching can be described by five basic operations:291loading, enabling, replacing, disabling, removing.292 293Where the replacing and the disabling operations are mutually294exclusive. They have the same result for the given patch but295not for the system.296 297 2985.1. Loading299------------300 301The only reasonable way is to enable the patch when the livepatch kernel302module is being loaded. For this, klp_enable_patch() has to be called303in the module_init() callback. There are two main reasons:304 305First, only the module has an easy access to the related struct klp_patch.306 307Second, the error code might be used to refuse loading the module when308the patch cannot get enabled.309 310 3115.2. Enabling312-------------313 314The livepatch gets enabled by calling klp_enable_patch() from315the module_init() callback. The system will start using the new316implementation of the patched functions at this stage.317 318First, the addresses of the patched functions are found according to their319names. The special relocations, mentioned in the section "New functions",320are applied. The relevant entries are created under321/sys/kernel/livepatch/<name>. The patch is rejected when any above322operation fails.323 324Second, livepatch enters into a transition state where tasks are converging325to the patched state. If an original function is patched for the first326time, a function specific struct klp_ops is created and an universal327ftrace handler is registered\ [#]_. This stage is indicated by a value of '1'328in /sys/kernel/livepatch/<name>/transition. For more information about329this process, see the "Consistency model" section.330 331Finally, once all tasks have been patched, the 'transition' value changes332to '0'.333 334.. [#]335 336 Note that functions might be patched multiple times. The ftrace handler337 is registered only once for a given function. Further patches just add338 an entry to the list (see field `func_stack`) of the struct klp_ops.339 The right implementation is selected by the ftrace handler, see340 the "Consistency model" section.341 342 That said, it is highly recommended to use cumulative livepatches343 because they help keeping the consistency of all changes. In this case,344 functions might be patched two times only during the transition period.345 346 3475.3. Replacing348--------------349 350All enabled patches might get replaced by a cumulative patch that351has the .replace flag set.352 353Once the new patch is enabled and the 'transition' finishes then354all the functions (struct klp_func) associated with the replaced355patches are removed from the corresponding struct klp_ops. Also356the ftrace handler is unregistered and the struct klp_ops is357freed when the related function is not modified by the new patch358and func_stack list becomes empty.359 360See Documentation/livepatch/cumulative-patches.rst for more details.361 362 3635.4. Disabling364--------------365 366Enabled patches might get disabled by writing '0' to367/sys/kernel/livepatch/<name>/enabled.368 369First, livepatch enters into a transition state where tasks are converging370to the unpatched state. The system starts using either the code from371the previously enabled patch or even the original one. This stage is372indicated by a value of '1' in /sys/kernel/livepatch/<name>/transition.373For more information about this process, see the "Consistency model"374section.375 376Second, once all tasks have been unpatched, the 'transition' value changes377to '0'. All the functions (struct klp_func) associated with the to-be-disabled378patch are removed from the corresponding struct klp_ops. The ftrace handler379is unregistered and the struct klp_ops is freed when the func_stack list380becomes empty.381 382Third, the sysfs interface is destroyed.383 384 3855.5. Removing386-------------387 388Module removal is only safe when there are no users of functions provided389by the module. This is the reason why the force feature permanently390disables the removal. Only when the system is successfully transitioned391to a new patch state (patched/unpatched) without being forced it is392guaranteed that no task sleeps or runs in the old code.393 394 3956. Sysfs396========397 398Information about the registered patches can be found under399/sys/kernel/livepatch. The patches could be enabled and disabled400by writing there.401 402/sys/kernel/livepatch/<patch>/force attributes allow administrator to affect a403patching operation.404 405See Documentation/ABI/testing/sysfs-kernel-livepatch for more details.406 407 4087. Limitations409==============410 411The current Livepatch implementation has several limitations:412 413 - Only functions that can be traced could be patched.414 415 Livepatch is based on the dynamic ftrace. In particular, functions416 implementing ftrace or the livepatch ftrace handler could not be417 patched. Otherwise, the code would end up in an infinite loop. A418 potential mistake is prevented by marking the problematic functions419 by "notrace".420 421 422 423 - Livepatch works reliably only when the dynamic ftrace is located at424 the very beginning of the function.425 426 The function need to be redirected before the stack or the function427 parameters are modified in any way. For example, livepatch requires428 using -fentry gcc compiler option on x86_64.429 430 One exception is the PPC port. It uses relative addressing and TOC.431 Each function has to handle TOC and save LR before it could call432 the ftrace handler. This operation has to be reverted on return.433 Fortunately, the generic ftrace code has the same problem and all434 this is handled on the ftrace level.435 436 437 - Kretprobes using the ftrace framework conflict with the patched438 functions.439 440 Both kretprobes and livepatches use a ftrace handler that modifies441 the return address. The first user wins. Either the probe or the patch442 is rejected when the handler is already in use by the other.443 444 445 - Kprobes in the original function are ignored when the code is446 redirected to the new implementation.447 448 There is a work in progress to add warnings about this situation.449