Re: 'perf probe' and symbols from .text.<something>

[Masami Hiramatsu]

Hi Evgenii,

On Thu, 18 Feb 2021 20:09:17 +0300
Evgenii Shatokhin <eshatokhin@xxxxxxxxxxxxx> wrote:

> Hi,
> 
> It seems, 'perf probe' can only see functions from .text section in the 
> kernel modules, but not from .text.unlikely or other .text.* sections.
> 
> For example, with kernel 5.11 and nf_conntrack.ko with debug info, 'perf 
> probe' succeeds for nf_conntrack_attach() from .text and fails for 
> nf_ct_resolve_clash() from .text.unlikely:

Thanks for reporting it!

> 
> ------------
> # perf probe -v -m nf_conntrack nf_ct_resolve_clash
> probe-definition(0): nf_ct_resolve_clash
> symbol:nf_ct_resolve_clash file:(null) line:0 offset:0 return:0 lazy:(null)
> 0 arguments
> Failed to get build-id from nf_conntrack.
> Cache open error: -1
> Open Debuginfo file: 
> /lib/modules/5.11.0-test01/kernel/net/netfilter/nf_conntrack.ko
> Try to find probe point from debuginfo.
> Matched function: nf_ct_resolve_clash [33616]
> Probe point found: nf_ct_resolve_clash+0
> Found 1 probe_trace_events.
> Post processing failed or all events are skipped. (-2)
> Probe point 'nf_ct_resolve_clash' not found.
>    Error: Failed to add events. Reason: No such file or directory (Code: -2)

The above log shows, an error occured while post_process_probe_trace_events(),
and the error code is -ENOENT (-2).
  ----
                pr_debug("Found %d probe_trace_events.\n", ntevs);
                ret = post_process_probe_trace_events(pev, *tevs, ntevs,
                                        pev->target, pev->uprobes, dinfo);
                if (ret < 0 || ret == ntevs) {
                        pr_debug("Post processing failed or all events are skipped. (%d)\n", ret);
  ----

In that function, map__find_symbol() failure will return -ENOENT.

  ----
/* Adjust symbol name and address */
static int post_process_probe_trace_point(struct probe_trace_point *tp,
                                           struct map *map, unsigned long offs)
{
        struct symbol *sym;
        u64 addr = tp->address - offs;

        sym = map__find_symbol(map, addr);
        if (!sym)
                return -ENOENT;
  ----

So it seems "map" may not load the symbol out of ".text".
This need to be fixed, since the map is widely used in the perf.

Anyway, since this is on a module, so even if it can not find the symbol
from map (or failed to load a map), it can fail back to the original symbol.
Let me fix that.

> # perf probe -v -m nf_conntrack nf_conntrack_attach
> probe-definition(0): nf_conntrack_attach
> symbol:nf_conntrack_attach file:(null) line:0 offset:0 return:0 lazy:(null)
> 0 arguments
> Failed to get build-id from nf_conntrack.
> Cache open error: -1
> Open Debuginfo file: 
> /lib/modules/5.11.0-test01/kernel/net/netfilter/nf_conntrack.ko
> Try to find probe point from debuginfo.
> Matched function: nf_conntrack_attach [2c8c3]
> Probe point found: nf_conntrack_attach+0
> Found 1 probe_trace_events.
> Opening /sys/kernel/tracing//kprobe_events write=1
> Opening /sys/kernel/tracing//README write=0
> Writing event: p:probe/nf_conntrack_attach 
> nf_conntrack:nf_conntrack_attach+0
> Added new event:
>    probe:nf_conntrack_attach (on nf_conntrack_attach in nf_conntrack)
> ------------
> 
> Is there a way to allow probing of functions in .text.<something> ?

I need to check how machine__kernel_maps() generated maps cut down .text.unlikely.
Arnaldo, I thought the maps in machine__kernel_maps() are generated from
kallsyms (doesn't check .text) right?

> 
> Of course, one could place probes using absolute addresses of the 
> functions but that would be less convenient.
> 
> This also affects many livepatch modules where the kernel code can be 
> compiled with -ffunction-sections and each function may end up in a 
> separate section .text.<function_name>. 'perf probe' cannot be used 
> there, except with the absolute addresses.
> 
> Moreover, if FGKASLR patches are merged 
> (https://lwn.net/Articles/832434/) and the kernel is built with FGKASLR 
> enabled, -ffunction-sections will be used too. 'perf probe' will be 
> unable to see the kernel functions then.

Hmm, if the FGKASLAR really randomizes the symbol address, perf-probe
should give up "_text-relative" probe for that kernel, and must fallback
to the "symbol-based" probe. (Are there any way to check the FGKASLR is on?)
The problem of "symbol-based" probe is that local (static) symbols
may share a same name sometimes. In that case, it can not find correct
symbol. (Maybe I can find a candidate from its size.)
Anyway, sometimes the security and usability are trade-off.

Thank you,

-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>