Re: [PATCH] nfsd: fix race between cache_clean and cache_purge

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 6144 bytes --]

On Tue, Mar 24 2020, Yihao Wu wrote:

> cache_purge should hold cache_list_lock as cache_clean does. Otherwise a cache
> can be cache_put twice, which leads to a use-after-free bug.
>
> To reproduce, run ltp. It happens rarely.  /opt/ltp/runltp run -f net.nfs
>
> [14454.137661] ==================================================================
> [14454.138863] BUG: KASAN: use-after-free in cache_purge+0xce/0x160 [sunrpc]
> [14454.139822] Read of size 4 at addr ffff8883d484d560 by task nfsd/31993
> [14454.140746]
> [14454.140995] CPU: 1 PID: 31993 Comm: nfsd Kdump: loaded Not tainted 4.19.91-0.229.git.87bac30.al7.x86_64.debug #1
> [14454.141002] Call Trace:
> [14454.141014]  dump_stack+0xaf/0xfb[14454.141027]  print_address_description+0x6a/0x2a0
> [14454.141037]  kasan_report+0x166/0x2b0[14454.141057]  ? cache_purge+0xce/0x160 [sunrpc]
> [14454.141079]  cache_purge+0xce/0x160 [sunrpc]
> [14454.141099]  nfsd_last_thread+0x267/0x270 [nfsd][14454.141109]  ? nfsd_last_thread+0x5/0x270 [nfsd]
> [14454.141130]  nfsd_destroy+0xcb/0x180 [nfsd]
> [14454.141140]  ? nfsd_destroy+0x5/0x180 [nfsd]
> [14454.141153]  nfsd+0x1e4/0x2b0 [nfsd]
> [14454.141163]  ? nfsd+0x5/0x2b0 [nfsd]
> [14454.141173]  kthread+0x114/0x150
> [14454.141183]  ? nfsd_destroy+0x180/0x180 [nfsd]
> [14454.141187]  ? kthread_park+0xb0/0xb0
> [14454.141197]  ret_from_fork+0x3a/0x50
> [14454.141224]
> [14454.141475] Allocated by task 20918:
> [14454.142011]  kmem_cache_alloc_trace+0x9f/0x2e0
> [14454.142027]  sunrpc_cache_lookup+0xca/0x2f0 [sunrpc]
> [14454.142037]  svc_export_parse+0x1e7/0x930 [nfsd]
> [14454.142051]  cache_do_downcall+0x5a/0x80 [sunrpc]
> [14454.142064]  cache_downcall+0x78/0x180 [sunrpc]
> [14454.142078]  cache_write_procfs+0x57/0x80 [sunrpc]
> [14454.142083]  proc_reg_write+0x90/0xd0
> [14454.142088]  vfs_write+0xc2/0x1c0
> [14454.142092]  ksys_write+0x4d/0xd0
> [14454.142098]  do_syscall_64+0x60/0x250
> [14454.142103]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
> [14454.142106]
> [14454.142344] Freed by task 19165:
> [14454.142804]  kfree+0x114/0x300
> [14454.142819]  cache_clean+0x2a4/0x2e0 [sunrpc]
> [14454.142833]  cache_flush+0x24/0x60 [sunrpc]
> [14454.142845]  write_flush.isra.19+0xbe/0x100 [sunrpc]
> [14454.142849]  proc_reg_write+0x90/0xd0
> [14454.142853]  vfs_write+0xc2/0x1c0
> [14454.142856]  ksys_write+0x4d/0xd0
> [14454.142860]  do_syscall_64+0x60/0x250
> [14454.142865]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
> [14454.142867]
> [14454.143095] The buggy address belongs to the object at ffff8883d484d540 which belongs to the cache kmalloc-256 of size 256
> [14454.144842] The buggy address is located 32 bytes inside of  256-byte region [ffff8883d484d540, ffff8883d484d640)
> [14454.146463] The buggy address belongs to the page:
> [14454.147155] page:ffffea000f521300 count:1 mapcount:0 mapping:ffff888107c02e00 index:0xffff8883d484da40 compound_map count: 0
> [14454.148712] flags: 0x17fffc00010200(slab|head)
> [14454.149356] raw: 0017fffc00010200 ffffea000f4baf00 0000000200000002 ffff888107c02e00
> [14454.150453] raw: ffff8883d484da40 0000000080190001 00000001ffffffff 0000000000000000
> [14454.151557] page dumped because: kasan: bad access detected
> [14454.152364]
> [14454.152606] Memory state around the buggy address:
> [14454.153300]  ffff8883d484d400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
> [14454.154319]  ffff8883d484d480: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
> [14454.155324] >ffff8883d484d500: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
> [14454.156334]                                                        ^
> [14454.157237]  ffff8883d484d580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
> [14454.158262]  ffff8883d484d600: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
> [14454.159282] ==================================================================
> [14454.160224] Disabling lock debugging due to kernel taint
>
> Fixes: 471a930ad7d1(SUNRPC: Drop all entries from cache_detail when cache_purge())
> Cc: stable@vger.kernel.org #v4.11+
> Signed-off-by: Yihao Wu <wuyihao@linux.alibaba.com>
> ---
>  net/sunrpc/cache.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/net/sunrpc/cache.c b/net/sunrpc/cache.c
> index bd843a81afa0..3e523eefc47f 100644
> --- a/net/sunrpc/cache.c
> +++ b/net/sunrpc/cache.c
> @@ -524,9 +524,11 @@ void cache_purge(struct cache_detail *detail)
>  	struct hlist_node *tmp = NULL;
>  	int i = 0;
>  
> +	spin_lock(&cache_list_lock);
>  	spin_lock(&detail->hash_lock);
>  	if (!detail->entries) {
>  		spin_unlock(&detail->hash_lock);
> +		spin_unlock(&cache_list_lock);
>  		return;
>  	}
>  
> @@ -541,6 +543,7 @@ void cache_purge(struct cache_detail *detail)
>  		}
>  	}
>  	spin_unlock(&detail->hash_lock);
> +	spin_unlock(&cache_list_lock);
>  }
>  EXPORT_SYMBOL_GPL(cache_purge);
>  
> -- 
> 2.20.1.2432.ga663e714

I wonder if this is the best solution.
This code:

		hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
			sunrpc_begin_cache_remove_entry(ch, detail);
			spin_unlock(&detail->hash_lock);
			sunrpc_end_cache_remove_entry(ch, detail);
			spin_lock(&detail->hash_lock);
		}

Looks wrong.
Dropping a lock while walking a list if only safe if you hold a
reference to the place-holder - 'tmp' in this case.  but we don't.
As this is trying to remove everything in the list it would be safer to
do something like

 while (!hlist_empty(head)) {
 	ch = hlist_entry(head->first, struct cache_head, h);
	sunrpc_begin_cache_remove_entry(ch, detail);
	spin_unlock(&detail->hash_lock);
	sunrpc_end_cache_remove_entry(ch, detail);
	spin_lock(&detail->hash_lock);
 }

I'm guessing that would fix the problem in a more focused.
But I'm not 100% sure because there is no analysis given of the cause.
What line is
  cache_purge+0xce/0x160
./scripts/faddr2line can tell you.
I suspect it is the hlist_for_each_entry_safe() line.

Can you test the change I suggest and see if it helps?

Thanks,
NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]