Petr Mladek | e553d2a | 2019-10-30 16:43:12 +0100 | [diff] [blame] | 1 | ==================== |
| 2 | System State Changes |
| 3 | ==================== |
| 4 | |
| 5 | Some users are really reluctant to reboot a system. This brings the need |
| 6 | to provide more livepatches and maintain some compatibility between them. |
| 7 | |
| 8 | Maintaining more livepatches is much easier with cumulative livepatches. |
| 9 | Each new livepatch completely replaces any older one. It can keep, |
| 10 | add, and even remove fixes. And it is typically safe to replace any version |
| 11 | of the livepatch with any other one thanks to the atomic replace feature. |
| 12 | |
| 13 | The problems might come with shadow variables and callbacks. They might |
| 14 | change the system behavior or state so that it is no longer safe to |
| 15 | go back and use an older livepatch or the original kernel code. Also |
| 16 | any new livepatch must be able to detect what changes have already been |
| 17 | done by the already installed livepatches. |
| 18 | |
| 19 | This is where the livepatch system state tracking gets useful. It |
| 20 | allows to: |
| 21 | |
| 22 | - store data needed to manipulate and restore the system state |
| 23 | |
| 24 | - define compatibility between livepatches using a change id |
| 25 | and version |
| 26 | |
| 27 | |
| 28 | 1. Livepatch system state API |
| 29 | ============================= |
| 30 | |
| 31 | The state of the system might get modified either by several livepatch callbacks |
| 32 | or by the newly used code. Also it must be possible to find changes done by |
| 33 | already installed livepatches. |
| 34 | |
| 35 | Each modified state is described by struct klp_state, see |
| 36 | include/linux/livepatch.h. |
| 37 | |
| 38 | Each livepatch defines an array of struct klp_states. They mention |
| 39 | all states that the livepatch modifies. |
| 40 | |
| 41 | The livepatch author must define the following two fields for each |
| 42 | struct klp_state: |
| 43 | |
| 44 | - *id* |
| 45 | |
| 46 | - Non-zero number used to identify the affected system state. |
| 47 | |
| 48 | - *version* |
| 49 | |
| 50 | - Number describing the variant of the system state change that |
| 51 | is supported by the given livepatch. |
| 52 | |
| 53 | The state can be manipulated using two functions: |
| 54 | |
| 55 | - *klp_get_state(patch, id)* |
| 56 | |
| 57 | - Get struct klp_state associated with the given livepatch |
| 58 | and state id. |
| 59 | |
| 60 | - *klp_get_prev_state(id)* |
| 61 | |
| 62 | - Get struct klp_state associated with the given feature id and |
| 63 | already installed livepatches. |
| 64 | |
| 65 | 2. Livepatch compatibility |
| 66 | ========================== |
| 67 | |
| 68 | The system state version is used to prevent loading incompatible livepatches. |
| 69 | The check is done when the livepatch is enabled. The rules are: |
| 70 | |
| 71 | - Any completely new system state modification is allowed. |
| 72 | |
| 73 | - System state modifications with the same or higher version are allowed |
| 74 | for already modified system states. |
| 75 | |
| 76 | - Cumulative livepatches must handle all system state modifications from |
| 77 | already installed livepatches. |
| 78 | |
| 79 | - Non-cumulative livepatches are allowed to touch already modified |
| 80 | system states. |
| 81 | |
| 82 | 3. Supported scenarios |
| 83 | ====================== |
| 84 | |
| 85 | Livepatches have their life-cycle and the same is true for the system |
| 86 | state changes. Every compatible livepatch has to support the following |
| 87 | scenarios: |
| 88 | |
| 89 | - Modify the system state when the livepatch gets enabled and the state |
| 90 | has not been already modified by a livepatches that are being |
| 91 | replaced. |
| 92 | |
| 93 | - Take over or update the system state modification when is has already |
| 94 | been done by a livepatch that is being replaced. |
| 95 | |
| 96 | - Restore the original state when the livepatch is disabled. |
| 97 | |
| 98 | - Restore the previous state when the transition is reverted. |
| 99 | It might be the original system state or the state modification |
| 100 | done by livepatches that were being replaced. |
| 101 | |
| 102 | - Remove any already made changes when error occurs and the livepatch |
| 103 | cannot get enabled. |
| 104 | |
| 105 | 4. Expected usage |
| 106 | ================= |
| 107 | |
| 108 | System states are usually modified by livepatch callbacks. The expected |
| 109 | role of each callback is as follows: |
| 110 | |
| 111 | *pre_patch()* |
| 112 | |
| 113 | - Allocate *state->data* when necessary. The allocation might fail |
| 114 | and *pre_patch()* is the only callback that could stop loading |
| 115 | of the livepatch. The allocation is not needed when the data |
| 116 | are already provided by previously installed livepatches. |
| 117 | |
| 118 | - Do any other preparatory action that is needed by |
| 119 | the new code even before the transition gets finished. |
| 120 | For example, initialize *state->data*. |
| 121 | |
| 122 | The system state itself is typically modified in *post_patch()* |
| 123 | when the entire system is able to handle it. |
| 124 | |
| 125 | - Clean up its own mess in case of error. It might be done by a custom |
| 126 | code or by calling *post_unpatch()* explicitly. |
| 127 | |
| 128 | *post_patch()* |
| 129 | |
| 130 | - Copy *state->data* from the previous livepatch when they are |
| 131 | compatible. |
| 132 | |
| 133 | - Do the actual system state modification. Eventually allow |
| 134 | the new code to use it. |
| 135 | |
| 136 | - Make sure that *state->data* has all necessary information. |
| 137 | |
| 138 | - Free *state->data* from replaces livepatches when they are |
| 139 | not longer needed. |
| 140 | |
| 141 | *pre_unpatch()* |
| 142 | |
| 143 | - Prevent the code, added by the livepatch, relying on the system |
| 144 | state change. |
| 145 | |
| 146 | - Revert the system state modification.. |
| 147 | |
| 148 | *post_unpatch()* |
| 149 | |
| 150 | - Distinguish transition reverse and livepatch disabling by |
| 151 | checking *klp_get_prev_state()*. |
| 152 | |
| 153 | - In case of transition reverse, restore the previous system |
| 154 | state. It might mean doing nothing. |
| 155 | |
| 156 | - Remove any not longer needed setting or data. |
| 157 | |
| 158 | .. note:: |
| 159 | |
| 160 | *pre_unpatch()* typically does symmetric operations to *post_patch()*. |
| 161 | Except that it is called only when the livepatch is being disabled. |
| 162 | Therefore it does not need to care about any previously installed |
| 163 | livepatch. |
| 164 | |
| 165 | *post_unpatch()* typically does symmetric operations to *pre_patch()*. |
| 166 | It might be called also during the transition reverse. Therefore it |
| 167 | has to handle the state of the previously installed livepatches. |