Skip to contentSkip to conversation

zero pointer after bdrv_unref_child

PATCH Applied Vladimir Sementsov-Ogievskiy 4 participants 7 messages Mar 16, 2020 – Mar 23, 2020 Block layer

lore
Share & tools
lore
#0 lore raw

cover zero pointer after bdrv_unref_child

Hi all!

I faced use-after-free of bs->backing pointer after bdrv_unref_child in bdrv_set_backing_hd.

Fix it, and do similar thing for s->data_file in qcow2.c.

I'm not sure that this is the full fix. Is it safe to keep bs->backing during bdrv_unref_child itself? Is it safe to keep bs->backing during all-child-unref loop in bdrv_close?

Vladimir Sementsov-Ogievskiy (2):
block: bdrv_set_backing_bs: fix use-after-free
block/qcow2: zero data_file child after free

 block.c       | 2 +-
 block/qcow2.c | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
Signature · 1 line
2.21.0
#4 lore raw

PATCH 2/2 block/qcow2: zero data_file child after free

data_file being NULL doesn't seem to be a correct state, but it's better than dead pointer and simpler to debug.

---
 block/qcow2.c | 2 ++
 1 file changed, 2 insertions(+)
block/qcow2.c +2 -0
index d44b45633d..6cdefe059f 100644
@@ -1758,6 +1758,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
     g_free(s->image_data_file);
     if (has_data_file(bs)) {
         bdrv_unref_child(bs, s->data_file);
+        s->data_file = NULL;
     }
     g_free(s->unknown_header_fields);
     cleanup_unknown_header_ext(bs);
@@ -2621,6 +2622,7 @@ static void qcow2_close(BlockDriverState *bs)
 
     if (has_data_file(bs)) {
         bdrv_unref_child(bs, s->data_file);
+        s->data_file = NULL;
     }
 
     qcow2_refcount_close(bs);
Signature · 1 line
2.21.0
#1 lore raw

PATCH 1/2 block: bdrv_set_backing_bs: fix use-after-free

In reply to Vladimir Sementsov-Ogievskiy

There is a use-after-free possible: bdrv_unref_child() leaves bs->backing freed but not NULL. bdrv_attach_child may produce nested polling loop due to drain, than access of freed pointer is possible.

I've produced the following crash on 30 iotest with modified code. It does not reproduce on master, but still seems possible:

    #0  __strcmp_avx2 () at /lib64/libc.so.6
    #1  bdrv_backing_overridden (bs=0x55c9d3cc2060) at block.c:6350
    #2  bdrv_refresh_filename (bs=0x55c9d3cc2060) at block.c:6404
    #3  bdrv_backing_attach (c=0x55c9d48e5520) at block.c:1063
    #4  bdrv_replace_child_noperm
        (child=child@entry=0x55c9d48e5520,
        new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2290
    #5  bdrv_replace_child
        (child=child@entry=0x55c9d48e5520,
        new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2320
    #6  bdrv_root_attach_child
        (child_bs=child_bs@entry=0x55c9d3cc2060,
        child_name=child_name@entry=0x55c9d241d478 "backing",
        child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
        ctx=<optimized out>, perm=<optimized out>, shared_perm=21,
        opaque=0x55c9d3c5a3d0, errp=0x7ffd117108e0) at block.c:2424
    #7  bdrv_attach_child
        (parent_bs=parent_bs@entry=0x55c9d3c5a3d0,
        child_bs=child_bs@entry=0x55c9d3cc2060,
        child_name=child_name@entry=0x55c9d241d478 "backing",
        child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
        errp=errp@entry=0x7ffd117108e0) at block.c:5876
    #8  in bdrv_set_backing_hd
        (bs=bs@entry=0x55c9d3c5a3d0,
        backing_hd=backing_hd@entry=0x55c9d3cc2060,
        errp=errp@entry=0x7ffd117108e0)
        at block.c:2576
    #9  stream_prepare (job=0x55c9d49d84a0) at block/stream.c:150
    #10 job_prepare (job=0x55c9d49d84a0) at job.c:761
    #11 job_txn_apply (txn=<optimized out>, fn=<optimized out>) at
        job.c:145
    #12 job_do_finalize (job=0x55c9d49d84a0) at job.c:778
    #13 job_completed_txn_success (job=0x55c9d49d84a0) at job.c:832
    #14 job_completed (job=0x55c9d49d84a0) at job.c:845
    #15 job_completed (job=0x55c9d49d84a0) at job.c:836
    #16 job_exit (opaque=0x55c9d49d84a0) at job.c:864
    #17 aio_bh_call (bh=0x55c9d471a160) at util/async.c:117
    #18 aio_bh_poll (ctx=ctx@entry=0x55c9d3c46720) at util/async.c:117
    #19 aio_poll (ctx=ctx@entry=0x55c9d3c46720,
        blocking=blocking@entry=true)
        at util/aio-posix.c:728
    #20 bdrv_parent_drained_begin_single (poll=true, c=0x55c9d3d558f0)
        at block/io.c:121
    #21 bdrv_parent_drained_begin_single (c=c@entry=0x55c9d3d558f0,
        poll=poll@entry=true)
        at block/io.c:114
    #22 bdrv_replace_child_noperm
        (child=child@entry=0x55c9d3d558f0,
        new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2258
    #23 bdrv_replace_child
        (child=child@entry=0x55c9d3d558f0,
        new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2320
    #24 bdrv_root_attach_child
        (child_bs=child_bs@entry=0x55c9d3d27300,
        child_name=child_name@entry=0x55c9d241d478 "backing",
        child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
        ctx=<optimized out>, perm=<optimized out>, shared_perm=21,
        opaque=0x55c9d3cc2060, errp=0x7ffd11710c60) at block.c:2424
    #25 bdrv_attach_child
        (parent_bs=parent_bs@entry=0x55c9d3cc2060,
        child_bs=child_bs@entry=0x55c9d3d27300,
        child_name=child_name@entry=0x55c9d241d478 "backing",
        child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
        errp=errp@entry=0x7ffd11710c60) at block.c:5876
    #26 bdrv_set_backing_hd
        (bs=bs@entry=0x55c9d3cc2060,
        backing_hd=backing_hd@entry=0x55c9d3d27300,
        errp=errp@entry=0x7ffd11710c60)
        at block.c:2576
    #27 stream_prepare (job=0x55c9d495ead0) at block/stream.c:150
    ...
---
 block.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block.c +1 -1
index 957630b1c5..a862ce4df9 100644
@@ -2735,10 +2735,10 @@ void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
 
     if (bs->backing) {
         bdrv_unref_child(bs, bs->backing);
+        bs->backing = NULL;
     }
 
     if (!backing_hd) {
-        bs->backing = NULL;
         goto out;
     }
 
Signature · 1 line
2.21.0
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] block: bdrv_set_backing_bs: fix use-after-free

Quoted 70 lines · Vladimir Sementsov-Ogievskiy

There is a use-after-free possible: bdrv_unref_child() leaves bs->backing freed but not NULL. bdrv_attach_child may produce nested polling loop due to drain, than access of freed pointer is possible.

I've produced the following crash on 30 iotest with modified code. It does not reproduce on master, but still seems possible:

     #0  __strcmp_avx2 () at /lib64/libc.so.6
     #1  bdrv_backing_overridden (bs=0x55c9d3cc2060) at block.c:6350
     #2  bdrv_refresh_filename (bs=0x55c9d3cc2060) at block.c:6404
     #3  bdrv_backing_attach (c=0x55c9d48e5520) at block.c:1063
     #4  bdrv_replace_child_noperm
         (child=child@entry=0x55c9d48e5520,
         new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2290
     #5  bdrv_replace_child
         (child=child@entry=0x55c9d48e5520,
         new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2320
     #6  bdrv_root_attach_child
         (child_bs=child_bs@entry=0x55c9d3cc2060,
         child_name=child_name@entry=0x55c9d241d478 "backing",
         child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
         ctx=<optimized out>, perm=<optimized out>, shared_perm=21,
         opaque=0x55c9d3c5a3d0, errp=0x7ffd117108e0) at block.c:2424
     #7  bdrv_attach_child
         (parent_bs=parent_bs@entry=0x55c9d3c5a3d0,
         child_bs=child_bs@entry=0x55c9d3cc2060,
         child_name=child_name@entry=0x55c9d241d478 "backing",
         child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
         errp=errp@entry=0x7ffd117108e0) at block.c:5876
     #8  in bdrv_set_backing_hd
         (bs=bs@entry=0x55c9d3c5a3d0,
         backing_hd=backing_hd@entry=0x55c9d3cc2060,
         errp=errp@entry=0x7ffd117108e0)
         at block.c:2576
     #9  stream_prepare (job=0x55c9d49d84a0) at block/stream.c:150
     #10 job_prepare (job=0x55c9d49d84a0) at job.c:761
     #11 job_txn_apply (txn=<optimized out>, fn=<optimized out>) at
         job.c:145
     #12 job_do_finalize (job=0x55c9d49d84a0) at job.c:778
     #13 job_completed_txn_success (job=0x55c9d49d84a0) at job.c:832
     #14 job_completed (job=0x55c9d49d84a0) at job.c:845
     #15 job_completed (job=0x55c9d49d84a0) at job.c:836
     #16 job_exit (opaque=0x55c9d49d84a0) at job.c:864
     #17 aio_bh_call (bh=0x55c9d471a160) at util/async.c:117
     #18 aio_bh_poll (ctx=ctx@entry=0x55c9d3c46720) at util/async.c:117
     #19 aio_poll (ctx=ctx@entry=0x55c9d3c46720,
         blocking=blocking@entry=true)
         at util/aio-posix.c:728
     #20 bdrv_parent_drained_begin_single (poll=true, c=0x55c9d3d558f0)
         at block/io.c:121
     #21 bdrv_parent_drained_begin_single (c=c@entry=0x55c9d3d558f0,
         poll=poll@entry=true)
         at block/io.c:114
     #22 bdrv_replace_child_noperm
         (child=child@entry=0x55c9d3d558f0,
         new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2258
     #23 bdrv_replace_child
         (child=child@entry=0x55c9d3d558f0,
         new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2320
     #24 bdrv_root_attach_child
         (child_bs=child_bs@entry=0x55c9d3d27300,
         child_name=child_name@entry=0x55c9d241d478 "backing",
         child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
         ctx=<optimized out>, perm=<optimized out>, shared_perm=21,
         opaque=0x55c9d3cc2060, errp=0x7ffd11710c60) at block.c:2424
     #25 bdrv_attach_child
         (parent_bs=parent_bs@entry=0x55c9d3cc2060,
         child_bs=child_bs@entry=0x55c9d3d27300,
         child_name=child_name@entry=0x55c9d241d478 "backing",
         child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
Quoted 9 lines
         errp=errp@entry=0x7ffd11710c60) at block.c:5876
     #26 bdrv_set_backing_hd
         (bs=bs@entry=0x55c9d3cc2060,
         backing_hd=backing_hd@entry=0x55c9d3d27300,
         errp=errp@entry=0x7ffd11710c60)
         at block.c:2576
     #27 stream_prepare (job=0x55c9d495ead0) at block/stream.c:150
     ...

Apparently:

Fixes 12fa4af61f (block: Add Error parameter to bdrv_set_backing_hd)

Right?

Quoted 1 line
Quoted 21 lines
---
  block.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
block.c +1 -1
index 957630b1c5..a862ce4df9 100644
@@ -2735,10 +2735,10 @@ void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
 
      if (bs->backing) {
          bdrv_unref_child(bs, bs->backing);
+        bs->backing = NULL;
      }
 
      if (!backing_hd) {
-        bs->backing = NULL;
          goto out;
      }
 
Reviewed-by
#3 lore raw

Re: [PATCH 1/2] block: bdrv_set_backing_bs: fix use-after-free

Quoted 81 lines · Philippe Mathieu-Daudé
Quoted 70 lines · Vladimir Sementsov-Ogievskiy

There is a use-after-free possible: bdrv_unref_child() leaves bs->backing freed but not NULL. bdrv_attach_child may produce nested polling loop due to drain, than access of freed pointer is possible.

I've produced the following crash on 30 iotest with modified code. It does not reproduce on master, but still seems possible:

     #0  __strcmp_avx2 () at /lib64/libc.so.6
     #1  bdrv_backing_overridden (bs=0x55c9d3cc2060) at block.c:6350 #2  bdrv_refresh_filename (bs=0x55c9d3cc2060) at block.c:6404 #3  bdrv_backing_attach (c=0x55c9d48e5520) at block.c:1063 #4  bdrv_replace_child_noperm (child=child@entry=0x55c9d48e5520, new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2290 #5  bdrv_replace_child (child=child@entry=0x55c9d48e5520, new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2320 #6  bdrv_root_attach_child (child_bs=child_bs@entry=0x55c9d3cc2060, child_name=child_name@entry=0x55c9d241d478 "backing", child_role=child_role@entry=0x55c9d26ecee0 <child_backing>, ctx=<optimized out>, perm=<optimized out>, shared_perm=21, opaque=0x55c9d3c5a3d0, errp=0x7ffd117108e0) at block.c:2424 #7  bdrv_attach_child (parent_bs=parent_bs@entry=0x55c9d3c5a3d0, child_bs=child_bs@entry=0x55c9d3cc2060, child_name=child_name@entry=0x55c9d241d478 "backing", child_role=child_role@entry=0x55c9d26ecee0 <child_backing>, errp=errp@entry=0x7ffd117108e0) at block.c:5876
     #8  in bdrv_set_backing_hd
         (bs=bs@entry=0x55c9d3c5a3d0, backing_hd=backing_hd@entry=0x55c9d3cc2060, errp=errp@entry=0x7ffd117108e0)
         at block.c:2576
     #9  stream_prepare (job=0x55c9d49d84a0) at block/stream.c:150 #10 job_prepare (job=0x55c9d49d84a0) at job.c:761
     #11 job_txn_apply (txn=<optimized out>, fn=<optimized out>) at job.c:145
     #12 job_do_finalize (job=0x55c9d49d84a0) at job.c:778 #13 job_completed_txn_success (job=0x55c9d49d84a0) at job.c:832 #14 job_completed (job=0x55c9d49d84a0) at job.c:845
     #15 job_completed (job=0x55c9d49d84a0) at job.c:836
     #16 job_exit (opaque=0x55c9d49d84a0) at job.c:864
     #17 aio_bh_call (bh=0x55c9d471a160) at util/async.c:117 #18 aio_bh_poll (ctx=ctx@entry=0x55c9d3c46720) at util/async.c:117 #19 aio_poll (ctx=ctx@entry=0x55c9d3c46720, blocking=blocking@entry=true)
         at util/aio-posix.c:728
     #20 bdrv_parent_drained_begin_single (poll=true, c=0x55c9d3d558f0) at block/io.c:121
     #21 bdrv_parent_drained_begin_single (c=c@entry=0x55c9d3d558f0, poll=poll@entry=true)
         at block/io.c:114
     #22 bdrv_replace_child_noperm (child=child@entry=0x55c9d3d558f0, new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2258 #23 bdrv_replace_child (child=child@entry=0x55c9d3d558f0, new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2320 #24 bdrv_root_attach_child (child_bs=child_bs@entry=0x55c9d3d27300, child_name=child_name@entry=0x55c9d241d478 "backing", child_role=child_role@entry=0x55c9d26ecee0 <child_backing>, ctx=<optimized out>, perm=<optimized out>, shared_perm=21, opaque=0x55c9d3cc2060, errp=0x7ffd11710c60) at block.c:2424 #25 bdrv_attach_child (parent_bs=parent_bs@entry=0x55c9d3cc2060, child_bs=child_bs@entry=0x55c9d3d27300, child_name=child_name@entry=0x55c9d241d478 "backing", child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,

Quoted 9 lines

         errp=errp@entry=0x7ffd11710c60) at block.c:5876 #26 bdrv_set_backing_hd (bs=bs@entry=0x55c9d3cc2060, backing_hd=backing_hd@entry=0x55c9d3d27300, errp=errp@entry=0x7ffd11710c60)
         at block.c:2576
     #27 stream_prepare (job=0x55c9d495ead0) at block/stream.c:150
     ...

Quoted 4 lines

Apparently:

Fixes 12fa4af61f (block: Add Error parameter to bdrv_set_backing_hd)

Right?

It's possible, but the way the bug manifests is very likely to have changed dramatically since 2.10.0.

This is (probably?) not a regression in any stable version supported upstream, or caused by that commit. It likely happened later when attach and recursive flushes were reworked.

Quoted 8 lines
Quoted 1 line
Quoted 4 lines

---
  block.c | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 17 lines
Quoted 14 lines
block.c +1 -1
index 957630b1c5..a862ce4df9 100644
@@ -2735,10 +2735,10 @@ void bdrv_set_backing_hd(BlockDriverState *bs,
 BlockDriverState *backing_hd,
         if (bs->backing) {
           bdrv_unref_child(bs, bs->backing);
+        bs->backing = NULL;
       }
         if (!backing_hd) {
-        bs->backing = NULL;
           goto out;
       }
Quoted 1 line

This seems more obviously correct; but what's the error policy if bdrv_attach_child fails? We'll have dropped the old file but won't pick the new one up.

(That already seems to happen without this patch, so ... not a new concern.)

I feel comfortable to say:

But don't know the answer to more structural questions, like if it's important that the detached child gets flushed or not, in what order that happens, etc etc.

Reviewed-by on v1 2/2
#5 lore raw

Re: [PATCH 2/2] block/qcow2: zero data_file child after free

In reply to Vladimir Sementsov-Ogievskiy
Quoted 3 lines · Vladimir Sementsov-Ogievskiy

data_file being NULL doesn't seem to be a correct state, but it's better than dead pointer and simpler to debug.

How important is it to have correct state in the middle of teardown?

Quoted 10 lines
---
 block/qcow2.c | 2 ++
 1 file changed, 2 insertions(+)
block/qcow2.c +0 -0
index d44b45633d..6cdefe059f 100644

block/qcow2.c static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, · line 1762

Quoted 5 lines
block/qcow2.c +1 -0
@@ -1758,6 +1758,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
     g_free(s->image_data_file);
     if (has_data_file(bs)) {
         bdrv_unref_child(bs, s->data_file);
+        s->data_file = NULL;
     }

Probably OK to set to NULL, since this is at the end of a failed open where it would have been set for the first time anyway.

It's an invalid state, but resulting from a failed call. I think that's OK.

(Are there any callers of bdrv_open or qcow2_open that don't just immediately trash this object if it failed? I don't know of any, but there's a lot of callers to bdrv_open.)

block/qcow2.c static void qcow2_close(BlockDriverState *bs) · line 2627

Quoted 9 lines
block/qcow2.c +1 -0
     g_free(s->unknown_header_fields);
     cleanup_unknown_header_ext(bs);
@@ -2621,6 +2622,7 @@ static void qcow2_close(BlockDriverState *bs)
 
     if (has_data_file(bs)) {
         bdrv_unref_child(bs, s->data_file);
+        s->data_file = NULL;
     }
 

Probably fine here too. I can't imagine it's valid to use this object after close() ... unless we open it again, and that should handle setting this back to a realistic value.

block/qcow2.c static void qcow2_close(BlockDriverState *bs) · line 2628

Quoted 2 lines
block/qcow2.c +0 -0
     qcow2_refcount_close(bs);

So I think this is fine? If I understand right this just makes failures more obvious if we do accidentally use this value after a failed open or close, so that seems fine.

(As always, I'll rely on block maintainers to do more serious structural review for cases I am not aware of)

Applied
#6 lore raw
In reply to Vladimir Sementsov-Ogievskiy
Quoted 7 lines · Vladimir Sementsov-Ogievskiy

Hi all!

I faced use-after-free of bs->backing pointer after bdrv_unref_child in bdrv_set_backing_hd.

Fix it, and do similar thing for s->data_file in qcow2.c.

Quoted 8 lines

I'm not sure that this is the full fix. Is it safe to keep bs->backing during bdrv_unref_child itself? Is it safe to keep bs->backing during all-child-unref loop in bdrv_close?

Vladimir Sementsov-Ogievskiy (2):
block: bdrv_set_backing_bs: fix use-after-free
block/qcow2: zero data_file child after free

Thanks, applied to my block branch:

https://git.xanclic.moe/XanClic/qemu/commits/branch/block

Max