From e0a6249190402a28f1e2925a81acf573157d55ef Mon Sep 17 00:00:00 2001 From: "Masami Hiramatsu (Google)" Date: Tue, 29 Sep 2026 09:19:12 +0900 Subject: [PATCH 1/2] fprobe: Use guard(rcu_sched_notrace) and check rcu_is_watching() unregister_fprobe() and unregister_fprobe_async() (used by BPF kprobe-multi) rely on standard RCU grace periods (synchronize_rcu() and call_rcu()) to wait until in-flight fprobe handlers complete before freeing the fprobe. However, if an fprobe handler executes while RCU is not watching (such as in the idle loop or nohz_full extended quiescent states), standard RCU does not track preemption-disabled sections. Consequently, synchronize_rcu() does not wait for those executions, which can lead to a use-after-free if the fprobe is freed immediately after unregistration. Ensure handlers exit early when !rcu_is_watching(). Furthermore, fprobe_fgraph_entry() and fprobe_ftrace_entry() previously used guard(rcu)() and rcu_read_lock(), which invoke lockdep on every hit under CONFIG_PROVE_LOCKING. This adds overhead and can cause lockdep recursion if probed functions interact with lockdep. Since rhltable_lookup() and rhl_for_each_entry_rcu() use rcu_dereference_all_check() (which checks rcu_read_lock_any_held()), holding preemption disabled via rcu_read_lock_sched_notrace() is fully valid and sufficient so long as rcu_is_watching() is true. Define and use guard(rcu_sched_notrace)() across fprobe_ftrace_entry(), fprobe_fgraph_entry(), and fprobe_return(). This eliminates fast-path rcu_read_lock() and lockdep overhead while guaranteeing safe grace period synchronization. Link: https://lore.kernel.org/all/179064115227.394389.16910234241400391996.stgit@devnote2/ Reported-by: Sashiko Closes: https://sashiko.dev/#/bug/linux-e46bcd68-4a56-4f19-a255-e3772980e5e3 Fixes: 657b594b2084 ("fprobe: Fix unregister_fprobe() to wait for RCU grace period") Cc: stable@vger.kernel.org Assisted-by: LLM Signed-off-by: Masami Hiramatsu (Google) Reviewed-by: Paul E. McKenney --- kernel/trace/fprobe.c | 26 ++++++++++++++++---------- 1 file changed, 16 insertions(+), 10 deletions(-) diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c index 9f2d98181779..da286619c5d8 100644 --- a/kernel/trace/fprobe.c +++ b/kernel/trace/fprobe.c @@ -47,6 +47,10 @@ static struct rhltable fprobe_ip_table; static DEFINE_MUTEX(fprobe_mutex); static struct fgraph_ops fprobe_graph_ops; +DEFINE_LOCK_GUARD_0(rcu_sched_notrace, + rcu_read_lock_sched_notrace(), + rcu_read_unlock_sched_notrace()) + static u32 fprobe_node_hashfn(const void *data, u32 len, u32 seed) { return hash_ptr(*(unsigned long **)data, 32); @@ -329,16 +333,14 @@ static void fprobe_ftrace_entry(unsigned long ip, unsigned long parent_ip, struct fprobe *fp; int bit; + if (!rcu_is_watching()) + return; + bit = ftrace_test_recursion_trylock(ip, parent_ip); if (bit < 0) return; - /* - * ftrace_test_recursion_trylock() disables preemption, but - * rhltable_lookup() checks whether rcu_read_lcok is held. - * So we take rcu_read_lock() here. - */ - rcu_read_lock(); + guard(rcu_sched_notrace)(); head = rhltable_lookup(&fprobe_ip_table, &ip, fprobe_rht_params); rhl_for_each_entry_rcu(node, pos, head, hlist) { @@ -353,7 +355,6 @@ static void fprobe_ftrace_entry(unsigned long ip, unsigned long parent_ip, else __fprobe_handler(ip, parent_ip, fp, fregs, NULL); } - rcu_read_unlock(); ftrace_test_recursion_unlock(bit); } NOKPROBE_SYMBOL(fprobe_ftrace_entry); @@ -567,10 +568,13 @@ static int fprobe_fgraph_entry(struct ftrace_graph_ent *trace, struct fgraph_ops struct fprobe *fp; int used, ret; + if (!rcu_is_watching()) + return 0; + if (WARN_ON_ONCE(!fregs)) return 0; - guard(rcu)(); + guard(rcu_sched_notrace)(); head = rhltable_lookup(&fprobe_ip_table, &func, fprobe_rht_params); reserved_words = 0; rhl_for_each_entry_rcu(node, pos, head, hlist) { @@ -665,13 +669,16 @@ static void fprobe_return(struct ftrace_graph_ret *trace, int size, curr; int size_words; + if (!rcu_is_watching()) + return; + fgraph_data = (unsigned long *)fgraph_retrieve_data(gops->idx, &size); if (WARN_ON_ONCE(!fgraph_data)) return; size_words = SIZE_IN_LONG(size); ret_ip = ftrace_regs_get_instruction_pointer(fregs); - preempt_disable_notrace(); + guard(rcu_sched_notrace)(); curr = 0; while (size_words > curr) { @@ -687,7 +694,6 @@ static void fprobe_return(struct ftrace_graph_ret *trace, } curr += size; } - preempt_enable_notrace(); } NOKPROBE_SYMBOL(fprobe_return); From 7a39c732a22ec4a369af9e19fbe3d666ec83eec4 Mon Sep 17 00:00:00 2001 From: Leon Hwang Date: Wed, 30 Sep 2026 13:36:18 +0800 Subject: [PATCH 2/2] kprobes: Skip disarmed probes when checking optkprobe overlap On x86, an optkprobe at A replaces five bytes with a jump. If a disabled probe B is at A+2, get_optimized_kprobe() stops at B when arming a new probe C at A+4. It leaves A optimized: A A+1 A+2 A+3 A+4 A's jump | e9 | d0 | d1 | d2 | d3 | after C | e9 | d0 | d1 | d2 | cc | The INT3 for C overwrites the last byte of A's jump displacement, so execution can jump to the wrong address. B can have prepared optinsns while disarmed, but has no jump to unoptimize. Continue past disarmed and unprepared probes to find the active optimized probe before arming a probe in its jump. Link: https://lore.kernel.org/all/20260930053618.104498-1-leon.hwang@linux.dev/ Fixes: afd66255b9a4 ("kprobes: Introduce kprobes jump optimization") Cc: stable@vger.kernel.org Signed-off-by: Leon Hwang Signed-off-by: Masami Hiramatsu (Google) --- kernel/kprobes.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/kernel/kprobes.c b/kernel/kprobes.c index 4edd8ca5c657..e787e4948c8b 100644 --- a/kernel/kprobes.c +++ b/kernel/kprobes.c @@ -496,14 +496,16 @@ static bool kprobe_queued(struct kprobe *p) static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr) { int i; - struct kprobe *p = NULL; + struct kprobe *p; struct optimized_kprobe *op; /* Don't check i == 0, since that is a breakpoint case. */ - for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++) + for (i = 1; i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++) { p = get_kprobe(addr - i); + /* A disabled probe can have prepared, but inactive, optinsns. */ + if (!p || !kprobe_optready(p) || kprobe_disarmed(p)) + continue; - if (p && kprobe_optready(p)) { op = container_of(p, struct optimized_kprobe, kp); if (arch_within_optimized_kprobe(op, addr)) return p;