On Fri, 27 Dec 2024, yangerkun wrote:
>
> 在 2024/12/27 7:15, NeilBrown 写道:
> > On Wed, 25 Dec 2024, Yang Erkun wrote:
> >> We can access exp->ex_stats or exp->ex_uuid in rcu context(c_show and
> >> e_show). All these resources should be released using kfree_rcu. Fix this
> >> by using call_rcu, clean them all after a rcu grace period.
> >>
> >> ==================================================================
> >> BUG: KASAN: slab-use-after-free in svc_export_show+0x362/0x430 [nfsd]
> >> Read of size 1 at addr ff11000010fdc120 by task cat/870
> >>
> >> CPU: 1 UID: 0 PID: 870 Comm: cat Not tainted 6.12.0-rc3+ #1
> >> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
> >> 1.16.1-2.fc37 04/01/2014
> >> Call Trace:
> >> <TASK>
> >> dump_stack_lvl+0x53/0x70
> >> print_address_description.constprop.0+0x2c/0x3a0
> >> print_report+0xb9/0x280
> >> kasan_report+0xae/0xe0
> >> svc_export_show+0x362/0x430 [nfsd]
> >> c_show+0x161/0x390 [sunrpc]
> >> seq_read_iter+0x589/0x770
> >> seq_read+0x1e5/0x270
> >> proc_reg_read+0xe1/0x140
> >> vfs_read+0x125/0x530
> >> ksys_read+0xc1/0x160
> >> do_syscall_64+0x5f/0x170
> >> entry_SYSCALL_64_after_hwframe+0x76/0x7e
> >>
> >> Allocated by task 830:
> >> kasan_save_stack+0x20/0x40
> >> kasan_save_track+0x14/0x30
> >> __kasan_kmalloc+0x8f/0xa0
> >> __kmalloc_node_track_caller_noprof+0x1bc/0x400
> >> kmemdup_noprof+0x22/0x50
> >> svc_export_parse+0x8a9/0xb80 [nfsd]
> >> cache_do_downcall+0x71/0xa0 [sunrpc]
> >> cache_write_procfs+0x8e/0xd0 [sunrpc]
> >> proc_reg_write+0xe1/0x140
> >> vfs_write+0x1a5/0x6d0
> >> ksys_write+0xc1/0x160
> >> do_syscall_64+0x5f/0x170
> >> entry_SYSCALL_64_after_hwframe+0x76/0x7e
> >>
> >> Freed by task 868:
> >> kasan_save_stack+0x20/0x40
> >> kasan_save_track+0x14/0x30
> >> kasan_save_free_info+0x3b/0x60
> >> __kasan_slab_free+0x37/0x50
> >> kfree+0xf3/0x3e0
> >> svc_export_put+0x87/0xb0 [nfsd]
> >> cache_purge+0x17f/0x1f0 [sunrpc]
> >> nfsd_destroy_serv+0x226/0x2d0 [nfsd]
> >> nfsd_svc+0x125/0x1e0 [nfsd]
> >> write_threads+0x16a/0x2a0 [nfsd]
> >> nfsctl_transaction_write+0x74/0xa0 [nfsd]
> >> vfs_write+0x1a5/0x6d0
> >> ksys_write+0xc1/0x160
> >> do_syscall_64+0x5f/0x170
> >> entry_SYSCALL_64_after_hwframe+0x76/0x7e
> >>
> >> Fixes: ae74136b4bb6 ("SUNRPC: Allow cache lookups to use RCU protection rather than the r/w spinlock")
> >> Reviewed-by: NeilBrown <neilb@xxxxxxx>
> >> Signed-off-by: Yang Erkun <yangerkun@xxxxxxxxxx>
> >> ---
> >> fs/nfsd/export.c | 19 ++++++++++++++-----
> >> 1 file changed, 14 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/fs/nfsd/export.c b/fs/nfsd/export.c
> >> index c6168bccfb6c..0363720280d4 100644
> >> --- a/fs/nfsd/export.c
> >> +++ b/fs/nfsd/export.c
> >> @@ -355,16 +355,25 @@ static void export_stats_destroy(struct export_stats *stats)
> >> EXP_STATS_COUNTERS_NUM);
> >> }
> >>
> >> -static void svc_export_put(struct kref *ref)
> >> +static void svc_export_release(struct rcu_head *rcu_head)
> >> {
> >> - struct svc_export *exp = container_of(ref, struct svc_export, h.ref);
> >> - path_put(&exp->ex_path);
> >> - auth_domain_put(exp->ex_client);
> >> + struct svc_export *exp = container_of(rcu_head, struct svc_export,
> >> + ex_rcu);
> >> +
> >> nfsd4_fslocs_free(&exp->ex_fslocs);
> >> export_stats_destroy(exp->ex_stats);
> >> kfree(exp->ex_stats);
> >> kfree(exp->ex_uuid);
> >> - kfree_rcu(exp, ex_rcu);
> >> + kfree(exp);
> >> +}
> >> +
> >> +static void svc_export_put(struct kref *ref)
> >> +{
> >> + struct svc_export *exp = container_of(ref, struct svc_export, h.ref);
> >> +
> >> + path_put(&exp->ex_path);
> >> + auth_domain_put(exp->ex_client);
> >> + call_rcu(&exp->ex_rcu, svc_export_release);
> >> }
> >>
> >
> > I think that ip_map_put() needs to be fixed for the same reason that
> > svc_export_put() needed to be fixed.
> >
> > ip_map_put() calls auth_domain_put() in ->im_client immediately, but
> > ip_map_show() accesses ->im_client.
> > So ip_map_put() needs to delay the auth_domain_put() using rcu.
>
> Thanks for your detail review!
>
> auth_domain_put will eventually call auth_domain_release, which all
> .domain_release will call call_rcu to free auth_domain. Maybe it is safe
> now?
Yes, that makes sense. As you say, no change is needed for
ip_map_put().
Thanks,
NeilBrown
