| Message ID | 20190819192505.483c0bf0@xhacker.debian (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64: KPROBES_ON_FTRACE | expand |
Jisheng Zhang wrote: > For KPROBES_ON_FTRACE case, we need to adjust the kprobe's addr > correspondingly. > > Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com> > --- > kernel/kprobes.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/kernel/kprobes.c b/kernel/kprobes.c > index 9873fc627d61..f8400753a8a9 100644 > --- a/kernel/kprobes.c > +++ b/kernel/kprobes.c > @@ -1560,6 +1560,9 @@ int register_kprobe(struct kprobe *p) > addr = kprobe_addr(p); > if (IS_ERR(addr)) > return PTR_ERR(addr); > +#ifdef CONFIG_KPROBES_ON_FTRACE > + addr = (kprobe_opcode_t *)ftrace_call_adjust((unsigned long)addr); > +#endif > p->addr = addr; I'm not sure what this is achieving, but looks wrong to me. If you intend to have kprobes default to using ftrace entry for probing functions, consider over-riding kprobe_lookup_name() -- see powerpc variant for example. - Naveen
Hi Jisheng, On Mon, 19 Aug 2019 11:36:09 +0000 Jisheng Zhang <Jisheng.Zhang@synaptics.com> wrote: > For KPROBES_ON_FTRACE case, we need to adjust the kprobe's addr > correspondingly. No, I think you have misunderstood what the ftrace_call_adjust() does. Ftrace's rec->ip is already adjusted when initializing it. Kprobes checks the list after initialized (adjusted). So you don't need to adjust it again. BTW, this type of hidden adjustment should be avoided by design. If you find user specifies wrong address, return error instead of adjust it silently. Thank you, > > Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com> > --- > kernel/kprobes.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/kernel/kprobes.c b/kernel/kprobes.c > index 9873fc627d61..f8400753a8a9 100644 > --- a/kernel/kprobes.c > +++ b/kernel/kprobes.c > @@ -1560,6 +1560,9 @@ int register_kprobe(struct kprobe *p) > addr = kprobe_addr(p); > if (IS_ERR(addr)) > return PTR_ERR(addr); > +#ifdef CONFIG_KPROBES_ON_FTRACE > + addr = (kprobe_opcode_t *)ftrace_call_adjust((unsigned long)addr); > +#endif > p->addr = addr; > > ret = check_kprobe_rereg(p); > -- > 2.23.0.rc1 >
On Mon, 19 Aug 2019 22:13:02 +0530 "Naveen N. Rao" wrote: > CAUTION: Email originated externally, do not click links or open attachments unless you recognize the sender and know the content is safe. > > > Jisheng Zhang wrote: > > For KPROBES_ON_FTRACE case, we need to adjust the kprobe's addr > > correspondingly. > > > > Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com> > > --- > > kernel/kprobes.c | 3 +++ > > 1 file changed, 3 insertions(+) > > > > diff --git a/kernel/kprobes.c b/kernel/kprobes.c > > index 9873fc627d61..f8400753a8a9 100644 > > --- a/kernel/kprobes.c > > +++ b/kernel/kprobes.c > > @@ -1560,6 +1560,9 @@ int register_kprobe(struct kprobe *p) > > addr = kprobe_addr(p); > > if (IS_ERR(addr)) > > return PTR_ERR(addr); > > +#ifdef CONFIG_KPROBES_ON_FTRACE > > + addr = (kprobe_opcode_t *)ftrace_call_adjust((unsigned long)addr); > > +#endif > > p->addr = addr; > > I'm not sure what this is achieving, but looks wrong to me. Indeed, I didn't take care of non-ftrace addr when KPROBES_ON_FTRACE, will update in next version. thanks > > If you intend to have kprobes default to using ftrace entry for probing > functions, consider over-riding kprobe_lookup_name() -- see powerpc > variant for example. > > > - Naveen >
On Tue, 20 Aug 2019 09:01:30 +0900 Masami Hiramatsu wrote: > > Hi Jisheng, Hi, > > On Mon, 19 Aug 2019 11:36:09 +0000 > Jisheng Zhang <Jisheng.Zhang@synaptics.com> wrote: > > > For KPROBES_ON_FTRACE case, we need to adjust the kprobe's addr > > correspondingly. > > No, I think you have misunderstood what the ftrace_call_adjust() does. > Ftrace's rec->ip is already adjusted when initializing it. Kprobes > checks the list after initialized (adjusted). So you don't need to > adjust it again. This is not to adjust the ftarce's rec->ip, but to adjust the struct kprobe addr member. Because check_kprobe_address_safe()=>arch_check_ftrace_location will check the kprobe's addr with ftrace's rec->ip. Since ftrace's rec->ip is already adjusted, there will be mismatch if we don't adjust kprobe's addr correspondingly. However, this patch is wrong. I should not update the kprobe's addr for non-ftrace-entry. Will fix this in next version. Thanks > > BTW, this type of hidden adjustment should be avoided by design. > If you find user specifies wrong address, return error instead of > adjust it silently. > > Thank you, > > > > > Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com> > > --- > > kernel/kprobes.c | 3 +++ > > 1 file changed, 3 insertions(+) > > > > diff --git a/kernel/kprobes.c b/kernel/kprobes.c > > index 9873fc627d61..f8400753a8a9 100644 > > --- a/kernel/kprobes.c > > +++ b/kernel/kprobes.c > > @@ -1560,6 +1560,9 @@ int register_kprobe(struct kprobe *p) > > addr = kprobe_addr(p); > > if (IS_ERR(addr)) > > return PTR_ERR(addr); > > +#ifdef CONFIG_KPROBES_ON_FTRACE > > + addr = (kprobe_opcode_t *)ftrace_call_adjust((unsigned long)addr); > > +#endif > > p->addr = addr; > > > > ret = check_kprobe_rereg(p); > > -- > > 2.23.0.rc1 > > > > > -- > Masami Hiramatsu <mhiramat@kernel.org>
diff --git a/kernel/kprobes.c b/kernel/kprobes.c index 9873fc627d61..f8400753a8a9 100644 --- a/kernel/kprobes.c +++ b/kernel/kprobes.c @@ -1560,6 +1560,9 @@ int register_kprobe(struct kprobe *p) addr = kprobe_addr(p); if (IS_ERR(addr)) return PTR_ERR(addr); +#ifdef CONFIG_KPROBES_ON_FTRACE + addr = (kprobe_opcode_t *)ftrace_call_adjust((unsigned long)addr); +#endif p->addr = addr; ret = check_kprobe_rereg(p);
For KPROBES_ON_FTRACE case, we need to adjust the kprobe's addr correspondingly. Signed-off-by: Jisheng Zhang <Jisheng.Zhang@synaptics.com> --- kernel/kprobes.c | 3 +++ 1 file changed, 3 insertions(+)