Merge branch 'kprobes: rethook: x86: Replace kretprobe trampoline with rethook'

Masami Hiramatsu says:

====================
Here are the 3rd version for generic kretprobe and kretprobe on x86 for
replacing the kretprobe trampoline with rethook. The previous version
is here[1]

[1] https://lore.kernel.org/all/164821817332.2373735.12048266953420821089.stgit@devnote2/T/#u

This version fixed typo and build issues for bpf-next and CONFIG_RETHOOK=y
error. I also add temporary mitigation lines for ANNOTATE_NOENDBR macro
issue for bpf-next tree [2/4].

This will be removed after merging kernel IBT series.

Background:

This rethook came from Jiri's request of multiple kprobe for bpf[2].
He tried to solve an issue that starting bpf with multiple kprobe will
take a long time because bpf-kprobe will wait for RCU grace period for
sync rcu events.

Jiri wanted to attach a single bpf handler to multiple kprobes and
he tried to introduce multiple-probe interface to kprobe. So I asked
him to use ftrace and kretprobe-like hook if it is only for the
function entry and exit, instead of adding ad-hoc interface
to kprobes.
For this purpose, I introduced the fprobe (kprobe like interface for
ftrace) with the rethook (this is a generic return hook feature for
fprobe exit handler)[3].

[2] https://lore.kernel.org/all/20220104080943.113249-1-jolsa@kernel.org/T/#u
[3] https://lore.kernel.org/all/164191321766.806991.7930388561276940676.stgit@devnote2/T/#u

The rethook is basically same as the kretprobe trampoline. I just made
it decoupled from kprobes. Eventually, the all arch dependent kretprobe
trampolines will be replaced with the rethook trampoline instead of
cloning and set HAVE_RETHOOK=y.
When I port the rethook for all arch which supports kretprobe, the
legacy kretprobe specific code (which is for CONFIG_KRETPROBE_ON_RETHOOK=n)
will be removed eventually.
====================

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Author: Alexei Starovoitov

arch/Kconfig

@@ -164,7 +164,13 @@ config ARCH_USE_BUILTIN_BSWAP
 
 config KRETPROBES
 	def_bool y
-	depends on KPROBES && HAVE_KRETPROBES
+	depends on KPROBES && (HAVE_KRETPROBES || HAVE_RETHOOK)
+
+config KRETPROBE_ON_RETHOOK
+	def_bool y
+	depends on HAVE_RETHOOK
+	depends on KRETPROBES
+	select RETHOOK
 
 config USER_RETURN_NOTIFIER
 	bool

arch/x86/Kconfig

@@ -224,6 +224,7 @@ config X86
 	select HAVE_KPROBES_ON_FTRACE
 	select HAVE_FUNCTION_ERROR_INJECTION
 	select HAVE_KRETPROBES
+	select HAVE_RETHOOK
 	select HAVE_KVM
 	select HAVE_LIVEPATCH			if X86_64
 	select HAVE_MIXED_BREAKPOINTS_REGS

arch/x86/include/asm/unwind.h

@@ -4,7 +4,7 @@
 
 #include <linux/sched.h>
 #include <linux/ftrace.h>
-#include <linux/kprobes.h>
+#include <linux/rethook.h>
 #include <asm/ptrace.h>
 #include <asm/stacktrace.h>
 
@@ -16,7 +16,7 @@ struct unwind_state {
 	unsigned long stack_mask;
 	struct task_struct *task;
 	int graph_idx;
-#ifdef CONFIG_KRETPROBES
+#if defined(CONFIG_RETHOOK)
 	struct llist_node *kr_cur;
 #endif
 	bool error;
@@ -104,19 +104,18 @@ void unwind_module_init(struct module *mod, void *orc_ip, size_t orc_ip_size,
 #endif
 
 static inline
-unsigned long unwind_recover_kretprobe(struct unwind_state *state,
-				       unsigned long addr, unsigned long *addr_p)
+unsigned long unwind_recover_rethook(struct unwind_state *state,
+				     unsigned long addr, unsigned long *addr_p)
 {
-#ifdef CONFIG_KRETPROBES
-	return is_kretprobe_trampoline(addr) ?
-		kretprobe_find_ret_addr(state->task, addr_p, &state->kr_cur) :
-		addr;
-#else
-	return addr;
+#ifdef CONFIG_RETHOOK
+	if (is_rethook_trampoline(addr))
+		return rethook_find_ret_addr(state->task, (unsigned long)addr_p,
+					     &state->kr_cur);
 #endif
+	return addr;
 }
 
-/* Recover the return address modified by kretprobe and ftrace_graph. */
+/* Recover the return address modified by rethook and ftrace_graph. */
 static inline
 unsigned long unwind_recover_ret_addr(struct unwind_state *state,
 				     unsigned long addr, unsigned long *addr_p)
@@ -125,7 +124,7 @@ unsigned long unwind_recover_ret_addr(struct unwind_state *state,
 
 	ret = ftrace_graph_ret_addr(state->task, &state->graph_idx,
 				    addr, addr_p);
-	return unwind_recover_kretprobe(state, ret, addr_p);
+	return unwind_recover_rethook(state, ret, addr_p);
 }
 
 /*

arch/x86/kernel/Makefile

@@ -103,6 +103,7 @@ obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o
 obj-$(CONFIG_FTRACE_SYSCALLS)	+= ftrace.o
 obj-$(CONFIG_X86_TSC)		+= trace_clock.o
 obj-$(CONFIG_TRACING)		+= trace.o
+obj-$(CONFIG_RETHOOK)		+= rethook.o
 obj-$(CONFIG_CRASH_CORE)	+= crash_core_$(BITS).o
 obj-$(CONFIG_KEXEC_CORE)	+= machine_kexec_$(BITS).o
 obj-$(CONFIG_KEXEC_CORE)	+= relocate_kernel_$(BITS).o crash.o

arch/x86/kernel/kprobes/common.h

@@ -6,6 +6,7 @@
 
 #include <asm/asm.h>
 #include <asm/frame.h>
+#include <asm/insn.h>
 
 #ifdef CONFIG_X86_64
 

arch/x86/kernel/kprobes/core.c

@@ -811,18 +811,6 @@ set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
 		= (regs->flags & X86_EFLAGS_IF);
 }
 
-void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
-{
-	unsigned long *sara = stack_addr(regs);
-
-	ri->ret_addr = (kprobe_opcode_t *) *sara;
-	ri->fp = sara;
-
-	/* Replace the return addr with trampoline addr */
-	*sara = (unsigned long) &__kretprobe_trampoline;
-}
-NOKPROBE_SYMBOL(arch_prepare_kretprobe);
-
 static void kprobe_post_process(struct kprobe *cur, struct pt_regs *regs,
 			       struct kprobe_ctlblk *kcb)
 {
@@ -1023,101 +1011,6 @@ int kprobe_int3_handler(struct pt_regs *regs)
 }
 NOKPROBE_SYMBOL(kprobe_int3_handler);
 
-/*
- * When a retprobed function returns, this code saves registers and
- * calls trampoline_handler() runs, which calls the kretprobe's handler.
- */
-asm(
-	".text\n"
-	".global __kretprobe_trampoline\n"
-	".type __kretprobe_trampoline, @function\n"
-	"__kretprobe_trampoline:\n"
-#ifdef CONFIG_X86_64
-	ANNOTATE_NOENDBR
-	/* Push a fake return address to tell the unwinder it's a kretprobe. */
-	"	pushq $__kretprobe_trampoline\n"
-	UNWIND_HINT_FUNC
-	/* Save the 'sp - 8', this will be fixed later. */
-	"	pushq %rsp\n"
-	"	pushfq\n"
-	SAVE_REGS_STRING
-	"	movq %rsp, %rdi\n"
-	"	call trampoline_handler\n"
-	RESTORE_REGS_STRING
-	/* In trampoline_handler(), 'regs->flags' is copied to 'regs->sp'. */
-	"	addq $8, %rsp\n"
-	"	popfq\n"
-#else
-	/* Push a fake return address to tell the unwinder it's a kretprobe. */
-	"	pushl $__kretprobe_trampoline\n"
-	UNWIND_HINT_FUNC
-	/* Save the 'sp - 4', this will be fixed later. */
-	"	pushl %esp\n"
-	"	pushfl\n"
-	SAVE_REGS_STRING
-	"	movl %esp, %eax\n"
-	"	call trampoline_handler\n"
-	RESTORE_REGS_STRING
-	/* In trampoline_handler(), 'regs->flags' is copied to 'regs->sp'. */
-	"	addl $4, %esp\n"
-	"	popfl\n"
-#endif
-	ASM_RET
-	".size __kretprobe_trampoline, .-__kretprobe_trampoline\n"
-);
-NOKPROBE_SYMBOL(__kretprobe_trampoline);
-/*
- * __kretprobe_trampoline() skips updating frame pointer. The frame pointer
- * saved in trampoline_handler() points to the real caller function's
- * frame pointer. Thus the __kretprobe_trampoline() doesn't have a
- * standard stack frame with CONFIG_FRAME_POINTER=y.
- * Let's mark it non-standard function. Anyway, FP unwinder can correctly
- * unwind without the hint.
- */
-STACK_FRAME_NON_STANDARD_FP(__kretprobe_trampoline);
-
-/* This is called from kretprobe_trampoline_handler(). */
-void arch_kretprobe_fixup_return(struct pt_regs *regs,
-				 kprobe_opcode_t *correct_ret_addr)
-{
-	unsigned long *frame_pointer = &regs->sp + 1;
-
-	/* Replace fake return address with real one. */
-	*frame_pointer = (unsigned long)correct_ret_addr;
-}
-
-/*
- * Called from __kretprobe_trampoline
- */
-__used __visible void trampoline_handler(struct pt_regs *regs)
-{
-	unsigned long *frame_pointer;
-
-	/* fixup registers */
-	regs->cs = __KERNEL_CS;
-#ifdef CONFIG_X86_32
-	regs->gs = 0;
-#endif
-	regs->ip = (unsigned long)&__kretprobe_trampoline;
-	regs->orig_ax = ~0UL;
-	regs->sp += sizeof(long);
-	frame_pointer = &regs->sp + 1;
-
-	/*
-	 * The return address at 'frame_pointer' is recovered by the
-	 * arch_kretprobe_fixup_return() which called from the
-	 * kretprobe_trampoline_handler().
-	 */
-	kretprobe_trampoline_handler(regs, frame_pointer);
-
-	/*
-	 * Copy FLAGS to 'pt_regs::sp' so that __kretprobe_trapmoline()
-	 * can do RET right after POPF.
-	 */
-	regs->sp = regs->flags;
-}
-NOKPROBE_SYMBOL(trampoline_handler);
-
 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 {
 	struct kprobe *cur = kprobe_running();

arch/x86/kernel/kprobes/opt.c

@@ -106,7 +106,8 @@ asm (
 			".global optprobe_template_entry\n"
 			"optprobe_template_entry:\n"
 #ifdef CONFIG_X86_64
-			/* We don't bother saving the ss register */
+			"       pushq $" __stringify(__KERNEL_DS) "\n"
+			/* Save the 'sp - 8', this will be fixed later. */
 			"	pushq %rsp\n"
 			"	pushfq\n"
 			".global optprobe_template_clac\n"
@@ -121,14 +122,17 @@ asm (
 			".global optprobe_template_call\n"
 			"optprobe_template_call:\n"
 			ASM_NOP5
-			/* Move flags to rsp */
+			/* Copy 'regs->flags' into 'regs->ss'. */
 			"	movq 18*8(%rsp), %rdx\n"
-			"	movq %rdx, 19*8(%rsp)\n"
+			"	movq %rdx, 20*8(%rsp)\n"
 			RESTORE_REGS_STRING
-			/* Skip flags entry */
-			"	addq $8, %rsp\n"
+			/* Skip 'regs->flags' and 'regs->sp'. */
+			"	addq $16, %rsp\n"
+			/* And pop flags register from 'regs->ss'. */
 			"	popfq\n"
 #else /* CONFIG_X86_32 */
+			"	pushl %ss\n"
+			/* Save the 'sp - 4', this will be fixed later. */
 			"	pushl %esp\n"
 			"	pushfl\n"
 			".global optprobe_template_clac\n"
@@ -142,12 +146,13 @@ asm (
 			".global optprobe_template_call\n"
 			"optprobe_template_call:\n"
 			ASM_NOP5
-			/* Move flags into esp */
+			/* Copy 'regs->flags' into 'regs->ss'. */
 			"	movl 14*4(%esp), %edx\n"
-			"	movl %edx, 15*4(%esp)\n"
+			"	movl %edx, 16*4(%esp)\n"
 			RESTORE_REGS_STRING
-			/* Skip flags entry */
-			"	addl $4, %esp\n"
+			/* Skip 'regs->flags' and 'regs->sp'. */
+			"	addl $8, %esp\n"
+			/* And pop flags register from 'regs->ss'. */
 			"	popfl\n"
 #endif
 			".global optprobe_template_end\n"
@@ -179,6 +184,8 @@ optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
 		kprobes_inc_nmissed_count(&op->kp);
 	} else {
 		struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+		/* Adjust stack pointer */
+		regs->sp += sizeof(long);
 		/* Save skipped registers */
 		regs->cs = __KERNEL_CS;
 #ifdef CONFIG_X86_32