Re: [PATCH] nfsd: don't ignore the return code of svc_proc_register()

[Chuck Lever <chuck.lever@oracle.com>]

On 2/6/25 1:29 PM, Jeff Layton wrote:
> On Thu, 2025-02-06 at 13:17 -0500, Chuck Lever wrote:
>> On 2/6/25 1:12 PM, Jeff Layton wrote:
>>> Currently, nfsd_proc_stat_init() ignores the return value of
>>> svc_proc_register(). If the procfile creation fails, then the kernel
>>> will WARN when it tries to remove the entry later.
>>>
>>> Fix nfsd_proc_stat_init() to return the same type of pointer as
>>> svc_proc_register(), and fix up nfsd_net_init() to check that and fail
>>> the nfsd_net construction if it occurs.
>>>
>>> svc_proc_register() can fail if the dentry can't be allocated, or if an
>>> identical dentry already exists. The second case is pretty unlikely in
>>> the nfsd_net construction codepath, so if this happens, return -ENOMEM.
>>>
>>> Fixes: 93483ac5fec6 ("nfsd: expose /proc/net/sunrpc/nfsd in net namespaces")
>>> Reported-by: syzbot+e34ad04f27991521104c@syzkaller.appspotmail.com
>>> Closes: https://lore.kernel.org/linux-nfs/67a47501.050a0220.19061f.05f9.GAE@google.com/
>>> Signed-off-by: Jeff Layton <jlayton@kernel.org>
>>> ---
>>> I looked at the console log from the report, and syzkaller is doing
>>> fault injection on allocations. You can see the stack where the "nfsd"
>>> directory under /proc failed to be created due to one. This is a pretty
>>> unlikely bug under normal circumstances, but it's simple to fix. The
>>> problem predates the patch in Fixes:, but it's not worth the effort to
>>> backport this to anything earlier.
>>
>> I'd prefer to document this by labeling the actual commit that
>> introduced the problem in the Fixes: tag, then using
>>
>> "Cc: stable # vN.M"
>>
>> to block automatic backporting to LTS kernels where this patch won't
>> apply cleanly. I can derive the values of N and M from the commit you
>> mention above, but do you happen to know the actual culprit commit?
>>
>>
> 
> Unfortunately this bug goes back to the initial 2.6.12 import into git.
> I didn't look earlier. Note that nfsd is not alone here. Ignoring the
> result of proc_create_data() is very common.
> 
> If you want to drop the Fixes: tag, and add the Cc: stable instead,
> then that's fine with me. Whatever works best.

OK. If we don't know the culprit, then a lone "Cc: stable" should be
sufficient.


>>> ---
>>>  fs/nfsd/nfsctl.c | 9 ++++++++-
>>>  fs/nfsd/stats.c  | 4 ++--
>>>  fs/nfsd/stats.h  | 2 +-
>>>  3 files changed, 11 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
>>> index 95ea4393305bd38493b640fbaba2e8f57f5a501d..583eda0df54dca394de4bbe8d148be2892df39cb 100644
>>> --- a/fs/nfsd/nfsctl.c
>>> +++ b/fs/nfsd/nfsctl.c
>>> @@ -2204,8 +2204,14 @@ static __net_init int nfsd_net_init(struct net *net)
>>>  					  NFSD_STATS_COUNTERS_NUM);
>>>  	if (retval)
>>>  		goto out_repcache_error;
>>> +
>>>  	memset(&nn->nfsd_svcstats, 0, sizeof(nn->nfsd_svcstats));
>>>  	nn->nfsd_svcstats.program = &nfsd_programs[0];
>>> +	if (!nfsd_proc_stat_init(net)) {
>>> +		retval = -ENOMEM;
>>> +		goto out_proc_error;
>>> +	}
>>> +
>>>  	for (i = 0; i < sizeof(nn->nfsd_versions); i++)
>>>  		nn->nfsd_versions[i] = nfsd_support_version(i);
>>>  	for (i = 0; i < sizeof(nn->nfsd4_minorversions); i++)
>>> @@ -2215,12 +2221,13 @@ static __net_init int nfsd_net_init(struct net *net)
>>>  	nfsd4_init_leases_net(nn);
>>>  	get_random_bytes(&nn->siphash_key, sizeof(nn->siphash_key));
>>>  	seqlock_init(&nn->writeverf_lock);
>>> -	nfsd_proc_stat_init(net);
>>>  #if IS_ENABLED(CONFIG_NFS_LOCALIO)
>>>  	INIT_LIST_HEAD(&nn->local_clients);
>>>  #endif
>>>  	return 0;
>>>  
>>> +out_proc_error:
>>> +	percpu_counter_destroy_many(nn->counter, NFSD_STATS_COUNTERS_NUM);
>>>  out_repcache_error:
>>>  	nfsd_idmap_shutdown(net);
>>>  out_idmap_error:
>>> diff --git a/fs/nfsd/stats.c b/fs/nfsd/stats.c
>>> index bb22893f1157e4c159e123b6d8e25b8eab52e187..f7eaf95e20fc8758566f469c6c2de79119fea070 100644
>>> --- a/fs/nfsd/stats.c
>>> +++ b/fs/nfsd/stats.c
>>> @@ -73,11 +73,11 @@ static int nfsd_show(struct seq_file *seq, void *v)
>>>  
>>>  DEFINE_PROC_SHOW_ATTRIBUTE(nfsd);
>>>  
>>> -void nfsd_proc_stat_init(struct net *net)
>>> +struct proc_dir_entry *nfsd_proc_stat_init(struct net *net)
>>>  {
>>>  	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
>>>  
>>> -	svc_proc_register(net, &nn->nfsd_svcstats, &nfsd_proc_ops);
>>> +	return svc_proc_register(net, &nn->nfsd_svcstats, &nfsd_proc_ops);
>>>  }
>>>  
>>>  void nfsd_proc_stat_shutdown(struct net *net)
>>> diff --git a/fs/nfsd/stats.h b/fs/nfsd/stats.h
>>> index 04aacb6c36e2576ba231ee481e3a3e9e9f255a61..e4efb0e4e56d467c13eaa5a1dd312c85dadeb4b8 100644
>>> --- a/fs/nfsd/stats.h
>>> +++ b/fs/nfsd/stats.h
>>> @@ -10,7 +10,7 @@
>>>  #include <uapi/linux/nfsd/stats.h>
>>>  #include <linux/percpu_counter.h>
>>>  
>>> -void nfsd_proc_stat_init(struct net *net);
>>> +struct proc_dir_entry *nfsd_proc_stat_init(struct net *net);
>>>  void nfsd_proc_stat_shutdown(struct net *net);
>>>  
>>>  static inline void nfsd_stats_rc_hits_inc(struct nfsd_net *nn)
>>>
>>> ---
>>> base-commit: ebbdc9429c39336a406b191cfe84bca2c12c2f73
>>> change-id: 20250206-nfsd-fixes-8e61bdf66347
>>>
>>> Best regards,
>>
>>
> 


-- 
Chuck Lever