From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- Documentation/livepatch/shadow-vars.rst | 226 ++++++++++++++++++++++++++++++++ 1 file changed, 226 insertions(+) create mode 100644 Documentation/livepatch/shadow-vars.rst (limited to 'Documentation/livepatch/shadow-vars.rst') diff --git a/Documentation/livepatch/shadow-vars.rst b/Documentation/livepatch/shadow-vars.rst new file mode 100644 index 000000000..7a7098bfb --- /dev/null +++ b/Documentation/livepatch/shadow-vars.rst @@ -0,0 +1,226 @@ +================ +Shadow Variables +================ + +Shadow variables are a simple way for livepatch modules to associate +additional "shadow" data with existing data structures. Shadow data is +allocated separately from parent data structures, which are left +unmodified. The shadow variable API described in this document is used +to allocate/add and remove/free shadow variables to/from their parents. + +The implementation introduces a global, in-kernel hashtable that +associates pointers to parent objects and a numeric identifier of the +shadow data. The numeric identifier is a simple enumeration that may be +used to describe shadow variable version, class or type, etc. More +specifically, the parent pointer serves as the hashtable key while the +numeric id subsequently filters hashtable queries. Multiple shadow +variables may attach to the same parent object, but their numeric +identifier distinguishes between them. + + +1. Brief API summary +==================== + +(See the full API usage docbook notes in livepatch/shadow.c.) + +A hashtable references all shadow variables. These references are +stored and retrieved through a pair. + +* The klp_shadow variable data structure encapsulates both tracking + meta-data and shadow-data: + + - meta-data + + - obj - pointer to parent object + - id - data identifier + + - data[] - storage for shadow data + +It is important to note that the klp_shadow_alloc() and +klp_shadow_get_or_alloc() are zeroing the variable by default. +They also allow to call a custom constructor function when a non-zero +value is needed. Callers should provide whatever mutual exclusion +is required. + +Note that the constructor is called under klp_shadow_lock spinlock. It allows +to do actions that can be done only once when a new variable is allocated. + +* klp_shadow_get() - retrieve a shadow variable data pointer + - search hashtable for pair + +* klp_shadow_alloc() - allocate and add a new shadow variable + - search hashtable for pair + + - if exists + + - WARN and return NULL + + - if doesn't already exist + + - allocate a new shadow variable + - initialize the variable using a custom constructor and data when provided + - add to the global hashtable + +* klp_shadow_get_or_alloc() - get existing or alloc a new shadow variable + - search hashtable for pair + + - if exists + + - return existing shadow variable + + - if doesn't already exist + + - allocate a new shadow variable + - initialize the variable using a custom constructor and data when provided + - add pair to the global hashtable + +* klp_shadow_free() - detach and free a shadow variable + - find and remove a reference from global hashtable + + - if found + + - call destructor function if defined + - free shadow variable + +* klp_shadow_free_all() - detach and free all <_, id> shadow variables + - find and remove any <_, id> references from global hashtable + + - if found + + - call destructor function if defined + - free shadow variable + + +2. Use cases +============ + +(See the example shadow variable livepatch modules in samples/livepatch/ +for full working demonstrations.) + +For the following use-case examples, consider commit 1d147bfa6429 +("mac80211: fix AP powersave TX vs. wakeup race"), which added a +spinlock to net/mac80211/sta_info.h :: struct sta_info. Each use-case +example can be considered a stand-alone livepatch implementation of this +fix. + + +Matching parent's lifecycle +--------------------------- + +If parent data structures are frequently created and destroyed, it may +be easiest to align their shadow variables lifetimes to the same +allocation and release functions. In this case, the parent data +structure is typically allocated, initialized, then registered in some +manner. Shadow variable allocation and setup can then be considered +part of the parent's initialization and should be completed before the +parent "goes live" (ie, any shadow variable get-API requests are made +for this pair.) + +For commit 1d147bfa6429, when a parent sta_info structure is allocated, +allocate a shadow copy of the ps_lock pointer, then initialize it:: + + #define PS_LOCK 1 + struct sta_info *sta_info_alloc(struct ieee80211_sub_if_data *sdata, + const u8 *addr, gfp_t gfp) + { + struct sta_info *sta; + spinlock_t *ps_lock; + + /* Parent structure is created */ + sta = kzalloc(sizeof(*sta) + hw->sta_data_size, gfp); + + /* Attach a corresponding shadow variable, then initialize it */ + ps_lock = klp_shadow_alloc(sta, PS_LOCK, sizeof(*ps_lock), gfp, + NULL, NULL); + if (!ps_lock) + goto shadow_fail; + spin_lock_init(ps_lock); + ... + +When requiring a ps_lock, query the shadow variable API to retrieve one +for a specific struct sta_info::: + + void ieee80211_sta_ps_deliver_wakeup(struct sta_info *sta) + { + spinlock_t *ps_lock; + + /* sync with ieee80211_tx_h_unicast_ps_buf */ + ps_lock = klp_shadow_get(sta, PS_LOCK); + if (ps_lock) + spin_lock(ps_lock); + ... + +When the parent sta_info structure is freed, first free the shadow +variable:: + + void sta_info_free(struct ieee80211_local *local, struct sta_info *sta) + { + klp_shadow_free(sta, PS_LOCK, NULL); + kfree(sta); + ... + + +In-flight parent objects +------------------------ + +Sometimes it may not be convenient or possible to allocate shadow +variables alongside their parent objects. Or a livepatch fix may +require shadow variables for only a subset of parent object instances. +In these cases, the klp_shadow_get_or_alloc() call can be used to attach +shadow variables to parents already in-flight. + +For commit 1d147bfa6429, a good spot to allocate a shadow spinlock is +inside ieee80211_sta_ps_deliver_wakeup():: + + int ps_lock_shadow_ctor(void *obj, void *shadow_data, void *ctor_data) + { + spinlock_t *lock = shadow_data; + + spin_lock_init(lock); + return 0; + } + + #define PS_LOCK 1 + void ieee80211_sta_ps_deliver_wakeup(struct sta_info *sta) + { + spinlock_t *ps_lock; + + /* sync with ieee80211_tx_h_unicast_ps_buf */ + ps_lock = klp_shadow_get_or_alloc(sta, PS_LOCK, + sizeof(*ps_lock), GFP_ATOMIC, + ps_lock_shadow_ctor, NULL); + + if (ps_lock) + spin_lock(ps_lock); + ... + +This usage will create a shadow variable, only if needed, otherwise it +will use one that was already created for this pair. + +Like the previous use-case, the shadow spinlock needs to be cleaned up. +A shadow variable can be freed just before its parent object is freed, +or even when the shadow variable itself is no longer required. + + +Other use-cases +--------------- + +Shadow variables can also be used as a flag indicating that a data +structure was allocated by new, livepatched code. In this case, it +doesn't matter what data value the shadow variable holds, its existence +suggests how to handle the parent object. + + +3. References +============= + +* https://github.com/dynup/kpatch + + The livepatch implementation is based on the kpatch version of shadow + variables. + +* http://files.mkgnu.net/files/dynamos/doc/papers/dynamos_eurosys_07.pdf + + Dynamic and Adaptive Updates of Non-Quiescent Subsystems in Commodity + Operating System Kernels (Kritis Makris, Kyung Dong Ryu 2007) presented + a datatype update technique called "shadow data structures". -- cgit v1.3.1