[PATCH v1] ring-buffer: Fix the sub-buffer array base in rb_meta_subbuf_idx()

[Donggeun Yoo <donggeunyoo.kernel@gmail.com>]

On a persistent ring buffer, rb_meta_subbuf_idx() can return an index one
sub-buffer too high.  rb_setup_ids_meta_page() then fills subbuf_ids[]
shifted by one and leaves subbuf_ids[0] NULL, which __rb_map_vma()
dereferences when userspace maps trace_pipe_raw.  rb_update_meta_reader()
writes two such indices into meta->buffers[], which rb_cpu_meta_valid()
would reject on the next boot, discarding the previous boot's trace.

A CPU's sub-buffer array starts after its meta header and the nr_subbufs
integers that follow, aligned up to a sub-buffer.  rb_meta_subbuf_idx()
inverts that calculation but skips only the integers, not the header, so
its base is one sub-buffer low whenever the omitted header size crosses
the alignment boundary.

Invert the calculation with rb_subbufs_from_meta() rather than repeating
it.  rb_range_buffer() maps an index back to an address through the same
helper, so the two directions can no longer disagree.

Cc: <stable@vger.kernel.org>
Fixes: b14d032973d4 ("ring-buffer: Add ring_buffer_meta data")
Signed-off-by: Donggeun Yoo <donggeunyoo.kernel@gmail.com>
---
x86_64_defconfig plus KASAN, FTRACE, TRACING and TRACER_SNAPSHOT, on
5dd1818b15d9, booted with reserve_mem=<size>:0x1000:trace
trace_instance=boot_mapped@trace.  head_buffer is from
instances/boot_mapped/per_cpu/cpuN/buffer_meta, the map of that CPU's
trace_pipe_raw.

nr_cpus=1, 24 reserved sizes 4096 bytes apart:

  nr_subbufs      head_buffer  map      head_buffer  map
                  unpatched             patched
  997 .. 1004     1            ok       1            ok
  1005 .. 1012    2            oops     1            ok
  1013 .. 1019    1            ok       1            ok

nr_cpus=4, nr_subbufs 1017, one boot:

  cpu 0           1            ok       1            ok
  cpu 1, 2, 3     2            oops     1            ok

cpu 0's meta sits after the buffer-wide header and the scratch area and is not
sub-buffer aligned; every other cpu's is, so the affected nr_subbufs differ
between them.

Unpatched oops:

  Oops: general protection fault, probably for non-canonical address 0xdffffc0000000007: 0000 [#1] SMP KASAN NOPTI
  KASAN: null-ptr-deref in range [0x0000000000000038-0x000000000000003f]
  RIP: 0010:__rb_map_vma+0x418/0xa20

With no reserved range at all, both kernels map the global buffer.  On both
arms, tools/testing/selftests/ring-buffer passes 6/6 with no skips and
CONFIG_RING_BUFFER_STARTUP_TEST reports "Ring buffer PASSED!".  ftracetest
gives identical per-test verdicts on the two arms -- 153 pass, 6 fail, 6
unresolved, 10 unsupported, 2 xfail.  The six failures are the same on both
arms and their cause was not established; the unresolved and unsupported ones
need pahole, a hypervisor trace remote, or userspace this initramfs does not
have.

The discarded-trace path is not measured: a fresh QEMU boot gets fresh guest
memory, so this harness cannot carry a persistent buffer across a reboot.

 kernel/trace/ring_buffer.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 04bb94c29f58..fa2e2ce683aa 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -3627,10 +3627,8 @@ static void rb_inc_iter(struct ring_buffer_iter *iter)
 /* Return the index into the sub-buffers for a given sub-buffer */
 static int rb_meta_subbuf_idx(struct ring_buffer_cpu_meta *meta, void *subbuf)
 {
-	void *subbuf_array;
+	void *subbuf_array = rb_subbufs_from_meta(meta);
 
-	subbuf_array = (void *)meta + sizeof(int) * meta->nr_subbufs;
-	subbuf_array = (void *)ALIGN((unsigned long)subbuf_array, meta->subbuf_size);
 	return (subbuf - subbuf_array) / meta->subbuf_size;
 }
 

base-commit: 5dd1818b15d98d4a20806cd00b1b40320b06004f
-- 
2.53.0