From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:30
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd")
The original e_show() repro is now fixed. However, the same
sosreport workload still reproduces a closely related fault on
post-v7.0 mainline (Misbah, ppc64le) and on master.20260424
(internal report, aarch64). In both cases the fault is in
cache_check_rcu() reached through c_show() rather than e_show(),
and the cache_head pointer is plain garbage:
pc : cache_check_rcu+0x40 [sunrpc]
lr : c_show+0x60 [sunrpc]
...faulting on h->flags off h = 0x0000000200000000
c_show() is the generic show callback used by
/proc/net/rpc/<cd>/content for every per-net cache_detail
(auth.unix.ip, auth.unix.gid, nfsd.fh, nfsd.export). Two
bugs combine in that path:
1. cache_unregister_net() / cache_destroy_net() free cd and
cd->hash_table synchronously when the namespace exits. The
/proc/net/rpc/.../content open path takes only a module
reference, so a fd kept open across a netns exit walks a
freed hash_table and returns garbage cache_head pointers.
This is the same hazard that e7fcf179b82d closed for the
/proc/fs/nfs/exports file alone.
2. ip_map_put() drops auth_domain_put() before kfree_rcu(), so
sub-objects can be freed before the RCU grace period -- the
same hazard that 48db892356d6 fixed for svc_export_put() and
expkey_put(). unix_gid_put() does not have this bug
structurally (its put_group_info() runs inside the call_rcu()
callback) but it uses a separate idiom from the other three
caches.
This series replaces the v1 narrow fixes with shared
infrastructure that covers all four cache_detail .put paths
and all three per-cache file types:
Patch 1 hoists nfsd_export_wq up to the sunrpc layer as
sunrpc_cache_wq, exposed through sunrpc_cache_queue_release()
and sunrpc_cache_drain() so all four put callbacks share one
workqueue and one drain primitive.
Patch 2 converts ip_map_put() to the queue_rcu_work() pattern,
moving auth_domain_put() into a deferred ip_map_release() that
runs after the RCU grace period.
Patch 3 unifies unix_gid_put() onto the same pattern for
consistency (not a bug fix on its own).
Patch 4 takes a get_net(cd->net) in content_open(), cache_open(),
and open_flush() and drops it in the matching release helpers,
so cache_destroy_net() cannot run while a sunrpc cache fd is
open.
Series has been compile-tested only.
---
Chuck Lever (6):
SUNRPC: Move cache_initialize() declaration to sunrpc-private header
SUNRPC: Provide a shared workqueue for cache release callbacks
SUNRPC: Defer ip_map sub-object cleanup past RCU grace period
SUNRPC: Use shared release pattern for the unix_gid cache
SUNRPC: Hold cd->net for the lifetime of cache files
NFSD: Convert nfsd_export_shutdown() to sunrpc_cache_destroy_net()
fs/nfsd/export.c | 45 ++--------------------
fs/nfsd/export.h | 2 -
fs/nfsd/nfsctl.c | 8 +---
include/linux/sunrpc/cache.h | 3 +-
net/sunrpc/cache.c | 90 ++++++++++++++++++++++++++++++++++++++++++--
net/sunrpc/sunrpc.h | 2 +
net/sunrpc/sunrpc_syms.c | 23 ++++++-----
net/sunrpc/svcauth_unix.c | 46 ++++++++++++----------
8 files changed, 135 insertions(+), 84 deletions(-)
---
base-commit: f3a313ecd1fdab1f5da119db355363b13af6fcac
change-id: 20260430-cache-uaf-fix-a13000f67c37
Best regards,
--
Chuck Lever
From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:31
From: Chuck Lever <redacted>
cache_initialize() was introduced by commit 8eab945c5616 ("sunrpc:
make the cache cleaner workqueue deferrable", 2010) and placed in
the public include/linux/sunrpc/cache.h from the start, but it
has never been EXPORT_SYMBOL_GPL'd. The only caller, init_sunrpc()
in net/sunrpc/sunrpc_syms.c, is built into the same module that
defines the function, so external modules could not link against
the symbol even if they tried. The public declaration has been a
stale-public hygiene leak ever since.
Relocate the declaration to net/sunrpc/sunrpc.h alongside other
sunrpc-internal helpers and include that header from
net/sunrpc/cache.c so the compiler enforces prototype consistency
at the definition site. The public include/linux/sunrpc/cache.h
now reflects the actual external API surface.
No functional change.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Chuck Lever <redacted>
---
include/linux/sunrpc/cache.h | 1 -
net/sunrpc/cache.c | 1 +
net/sunrpc/sunrpc.h | 1 +
3 files changed, 2 insertions(+), 1 deletion(-)
From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:33
From: Chuck Lever <redacted>
Cache .put callbacks may need to release sub-objects whose
cleanup sleeps (path_put, auth_domain_put, put_group_info), which
precludes running the release from a call_rcu() softirq callback.
Commit 48db892356d6 ("NFSD: Defer sub-object cleanup in export
put callbacks") introduced nfsd_export_wq for that purpose, with
a dedicated workqueue chosen so that flush_workqueue() in the
per-namespace teardown path drains only NFSD export release work
rather than blocking on unrelated work queued to system_unbound_wq.
Subsequent patches in this series convert the sunrpc ip_map and
unix_gid put callbacks to the same queue_rcu_work() pattern, and
those would otherwise need their own per-cache workqueue for the
same reason. Hoist the workqueue up to the sunrpc layer so that
all four cache_detail put callbacks share a single workqueue,
managed entirely within net/sunrpc/cache.c.
Expose the workqueue through three helpers.
sunrpc_cache_queue_release() schedules a deferred release after
the next RCU grace period. sunrpc_cache_destroy_net()
encapsulates the cache_unregister_net() + drain +
cache_destroy_net() sequence that single-cache teardowns
otherwise have to open-code, putting the ordering rule in one
place. sunrpc_cache_drain() exposes the underlying
rcu_barrier() + flush_workqueue() primitive for the rare caller
that drains multiple cache_details together, such as
nfsd_export_shutdown(). Allocate the workqueue in
cache_initialize() and destroy it in a new cache_destroy()
called from cleanup_sunrpc(). Replace the local nfsd_export_wq
with the shared sunrpc helpers and drop the
nfsd_export_wq_init/shutdown helpers and their callers.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Chuck Lever <redacted>
---
fs/nfsd/export.c | 41 +++-----------------------
fs/nfsd/export.h | 2 --
fs/nfsd/nfsctl.c | 8 +----
include/linux/sunrpc/cache.h | 3 ++
net/sunrpc/cache.c | 70 +++++++++++++++++++++++++++++++++++++++++++-
net/sunrpc/sunrpc.h | 3 +-
net/sunrpc/sunrpc_syms.c | 23 +++++++++------
7 files changed, 93 insertions(+), 57 deletions(-)
From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:34
From: Chuck Lever <redacted>
ip_map_put() is already correct in that auth_domain_put() of
im->m_client reaches svcauth_unix_domain_release(), which defers
the actual kfree() of the unix_domain with call_rcu(); the ip_map
itself is freed via kfree_rcu(). Readers in c_show() under
cache_seq_start_rcu() therefore observe both objects until the
next RCU grace period.
This patch is a consistency change rather than a bug fix. The
svc_export and svc_expkey caches were converted to the
queue_rcu_work() pattern in commit 48db892356d6 ("NFSD: Defer
sub-object cleanup in export put callbacks") because path_put()
and auth_domain_put() must run in process context after the RCU
grace period. The next patch routes unix_gid through the same
mechanism. Sending ip_map through sunrpc_cache_queue_release()
unifies all four cache_detail .put callbacks on a single release
path and removes the implicit reliance on every current and
future auth_ops .domain_release implementation deferring its own
kfree() behind call_rcu().
Replace the rcu_head field with an rcu_work, move the kfree() and
auth_domain_put() into a new ip_map_release() taking a
work_struct, and have ip_map_put() invoke INIT_RCU_WORK() and
sunrpc_cache_queue_release() in place of kfree_rcu(). Switch
ip_map_cache_destroy() to sunrpc_cache_destroy_net() so
per-namespace teardown waits for outstanding release work
before freeing the cache_detail.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Chuck Lever <redacted>
---
net/sunrpc/svcauth_unix.c | 25 ++++++++++++++++---------
1 file changed, 16 insertions(+), 9 deletions(-)
From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:35
From: Chuck Lever <redacted>
unix_gid_put() is already correct in that put_group_info() runs
inside its call_rcu() callback, after the RCU grace period.
This patch is a consistency change rather than a bug fix:
the three other cache_detail .put callbacks (svc_export,
svc_expkey, ip_map) now use the queue_rcu_work() pattern via
sunrpc_cache_queue_release(), and routing unix_gid through the same
path keeps a single release mechanism for all four caches.
Replace the rcu_head field with an rcu_work, rename
unix_gid_free() to unix_gid_release() and convert it to take
a work_struct, and have unix_gid_put() invoke INIT_RCU_WORK()
and sunrpc_cache_queue_release() in place of call_rcu().
Switch unix_gid_cache_destroy() to sunrpc_cache_destroy_net()
so per-namespace teardown waits for outstanding release work
before freeing the cache_detail.
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Chuck Lever <redacted>
---
net/sunrpc/svcauth_unix.c | 21 ++++++++++-----------
1 file changed, 10 insertions(+), 11 deletions(-)
From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:37
From: Chuck Lever <redacted>
Each per-net sunrpc cache exposes three files under
/proc/net/rpc/<cachename>/: content, channel, and flush.
Their open helpers (content_open, cache_open, open_flush) take a
reference on cd->owner via try_module_get() but no reference on
cd->net. When the network namespace exits, cache_unregister_net()
followed by cache_destroy_net() free cd->hash_table and cd
synchronously without RCU deferral. Any subsequent operation on
the still-open file dereferences the freed cache_detail.
The fault produced by sosreport on aarch64 and ppc64le shows the
typical signature: cache_check_rcu() faults reading h->flags off
a garbage cache_head pointer that came from __cache_seq_start()
walking a freed cd->hash_table. Commit e7fcf179b82d ("NFSD: Hold
net reference for the lifetime of /proc/fs/nfs/exports fd") closed
this hole only for the /proc/fs/nfs/exports file, which has its own
open path; the sunrpc cache files were left exposed.
Take a get_net(cd->net) in content_open(), cache_open(), and
open_flush() once the open has otherwise succeeded, and a matching
put_net() at the tail of each release helper. Holding the net
reference for the open file lifetime prevents the namespace from
exiting while a cache fd is open, which in turn prevents
cache_destroy_net() from running and freeing cd from under the
reader.
put_net() can drop the last namespace reference, in which case
__put_net() queues net_cleanup_work on netns_wq. That work runs
ops_undo_list() on another CPU, which invokes sunrpc_exit_net()
and frees cd via cache_destroy_net(). The release helper must
not dereference cd after put_net(): cache_release(),
content_release(), and release_flush() therefore capture
cd->owner and cd->net into local variables before calling
put_net(net) and module_put(owner).
Reported-by: Misbah Anjum N <redacted>
Closes: https://lore.kernel.org/linux-nfs/8cf80f450085ac17164e8fa1391e9635@linux.ibm.com/
Fixes: 1b10f0b603c0 ("SUNRPC: no need get cache ref when protected by rcu")
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Chuck Lever <redacted>
---
net/sunrpc/cache.c | 22 +++++++++++++++++++---
1 file changed, 19 insertions(+), 3 deletions(-)
From: Chuck Lever <cel@kernel.org> Date: 2026-05-01 14:51:38
From: Chuck Lever <redacted>
The earlier conversion to the shared sunrpc cache workqueue left
nfsd_export_shutdown() open-coding the unregister + drain + destroy
sequence to amortize a single rcu_barrier()+flush across both
caches. That micro-optimization runs only on namespace teardown
and module unload, neither of which is performance-sensitive.
Switch both caches to sunrpc_cache_destroy_net() so that the
canonical teardown idiom is applied uniformly. A future change to
the per-cache release callback that requires the cache_detail to
remain valid through the drain remains correct without per-call-site
auditing.
With the last external caller of sunrpc_cache_drain() gone, drop
the symbol export and the public declaration, and mark the
function static so the compiler can inline it into
sunrpc_cache_destroy_net().
Assisted-by: Claude:claude-opus-4-7[1m]
Signed-off-by: Chuck Lever <redacted>
---
fs/nfsd/export.c | 8 ++------
include/linux/sunrpc/cache.h | 1 -
net/sunrpc/cache.c | 7 ++-----
3 files changed, 4 insertions(+), 12 deletions(-)
From: Jeff Layton <jlayton@kernel.org> Date: 2026-05-05 05:32:56
On Fri, 2026-05-01 at 10:51 -0400, Chuck Lever wrote:
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd")
The original e_show() repro is now fixed. However, the same
sosreport workload still reproduces a closely related fault on
post-v7.0 mainline (Misbah, ppc64le) and on master.20260424
(internal report, aarch64). In both cases the fault is in
cache_check_rcu() reached through c_show() rather than e_show(),
and the cache_head pointer is plain garbage:
pc : cache_check_rcu+0x40 [sunrpc]
lr : c_show+0x60 [sunrpc]
...faulting on h->flags off h = 0x0000000200000000
c_show() is the generic show callback used by
/proc/net/rpc/<cd>/content for every per-net cache_detail
(auth.unix.ip, auth.unix.gid, nfsd.fh, nfsd.export). Two
bugs combine in that path:
1. cache_unregister_net() / cache_destroy_net() free cd and
cd->hash_table synchronously when the namespace exits. The
/proc/net/rpc/.../content open path takes only a module
reference, so a fd kept open across a netns exit walks a
freed hash_table and returns garbage cache_head pointers.
This is the same hazard that e7fcf179b82d closed for the
/proc/fs/nfs/exports file alone.
2. ip_map_put() drops auth_domain_put() before kfree_rcu(), so
sub-objects can be freed before the RCU grace period -- the
same hazard that 48db892356d6 fixed for svc_export_put() and
expkey_put(). unix_gid_put() does not have this bug
structurally (its put_group_info() runs inside the call_rcu()
callback) but it uses a separate idiom from the other three
caches.
This series replaces the v1 narrow fixes with shared
infrastructure that covers all four cache_detail .put paths
and all three per-cache file types:
Patch 1 hoists nfsd_export_wq up to the sunrpc layer as
sunrpc_cache_wq, exposed through sunrpc_cache_queue_release()
and sunrpc_cache_drain() so all four put callbacks share one
workqueue and one drain primitive.
Patch 2 converts ip_map_put() to the queue_rcu_work() pattern,
moving auth_domain_put() into a deferred ip_map_release() that
runs after the RCU grace period.
Patch 3 unifies unix_gid_put() onto the same pattern for
consistency (not a bug fix on its own).
Patch 4 takes a get_net(cd->net) in content_open(), cache_open(),
and open_flush() and drops it in the matching release helpers,
so cache_destroy_net() cannot run while a sunrpc cache fd is
open.
Series has been compile-tested only.
---
Chuck Lever (6):
SUNRPC: Move cache_initialize() declaration to sunrpc-private header
SUNRPC: Provide a shared workqueue for cache release callbacks
SUNRPC: Defer ip_map sub-object cleanup past RCU grace period
SUNRPC: Use shared release pattern for the unix_gid cache
SUNRPC: Hold cd->net for the lifetime of cache files
NFSD: Convert nfsd_export_shutdown() to sunrpc_cache_destroy_net()
fs/nfsd/export.c | 45 ++--------------------
fs/nfsd/export.h | 2 -
fs/nfsd/nfsctl.c | 8 +---
include/linux/sunrpc/cache.h | 3 +-
net/sunrpc/cache.c | 90 ++++++++++++++++++++++++++++++++++++++++++--
net/sunrpc/sunrpc.h | 2 +
net/sunrpc/sunrpc_syms.c | 23 ++++++-----
net/sunrpc/svcauth_unix.c | 46 ++++++++++++----------
8 files changed, 135 insertions(+), 84 deletions(-)
---
base-commit: f3a313ecd1fdab1f5da119db355363b13af6fcac
change-id: 20260430-cache-uaf-fix-a13000f67c37
Best regards,
--
Chuck Lever
The series looks sane.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd")
The original e_show() repro is now fixed. However, the same
sosreport workload still reproduces a closely related fault on
post-v7.0 mainline (Misbah, ppc64le) and on master.20260424
(internal report, aarch64). In both cases the fault is in
cache_check_rcu() reached through c_show() rather than e_show(),
and the cache_head pointer is plain garbage:
pc : cache_check_rcu+0x40 [sunrpc]
lr : c_show+0x60 [sunrpc]
...faulting on h->flags off h = 0x0000000200000000
c_show() is the generic show callback used by
/proc/net/rpc/<cd>/content for every per-net cache_detail
(auth.unix.ip, auth.unix.gid, nfsd.fh, nfsd.export). Two
bugs combine in that path:
1. cache_unregister_net() / cache_destroy_net() free cd and
cd->hash_table synchronously when the namespace exits. The
/proc/net/rpc/.../content open path takes only a module
reference, so a fd kept open across a netns exit walks a
freed hash_table and returns garbage cache_head pointers.
This is the same hazard that e7fcf179b82d closed for the
/proc/fs/nfs/exports file alone.
2. ip_map_put() drops auth_domain_put() before kfree_rcu(), so
sub-objects can be freed before the RCU grace period -- the
same hazard that 48db892356d6 fixed for svc_export_put() and
expkey_put(). unix_gid_put() does not have this bug
structurally (its put_group_info() runs inside the call_rcu()
callback) but it uses a separate idiom from the other three
caches.
This series replaces the v1 narrow fixes with shared
infrastructure that covers all four cache_detail .put paths
and all three per-cache file types:
Patch 1 hoists nfsd_export_wq up to the sunrpc layer as
sunrpc_cache_wq, exposed through sunrpc_cache_queue_release()
and sunrpc_cache_drain() so all four put callbacks share one
workqueue and one drain primitive.
Patch 2 converts ip_map_put() to the queue_rcu_work() pattern,
moving auth_domain_put() into a deferred ip_map_release() that
runs after the RCU grace period.
Patch 3 unifies unix_gid_put() onto the same pattern for
consistency (not a bug fix on its own).
Patch 4 takes a get_net(cd->net) in content_open(), cache_open(),
and open_flush() and drops it in the matching release helpers,
so cache_destroy_net() cannot run while a sunrpc cache fd is
open.
Series has been compile-tested only.
---
Chuck Lever (6):
SUNRPC: Move cache_initialize() declaration to sunrpc-private header
SUNRPC: Provide a shared workqueue for cache release callbacks
SUNRPC: Defer ip_map sub-object cleanup past RCU grace period
SUNRPC: Use shared release pattern for the unix_gid cache
SUNRPC: Hold cd->net for the lifetime of cache files
NFSD: Convert nfsd_export_shutdown() to sunrpc_cache_destroy_net()
fs/nfsd/export.c | 45 ++--------------------
fs/nfsd/export.h | 2 -
fs/nfsd/nfsctl.c | 8 +---
include/linux/sunrpc/cache.h | 3 +-
net/sunrpc/cache.c | 90 ++++++++++++++++++++++++++++++++++++++++++--
net/sunrpc/sunrpc.h | 2 +
net/sunrpc/sunrpc_syms.c | 23 ++++++-----
net/sunrpc/svcauth_unix.c | 46 ++++++++++++----------
8 files changed, 135 insertions(+), 84 deletions(-)
---
base-commit: f3a313ecd1fdab1f5da119db355363b13af6fcac
change-id: 20260430-cache-uaf-fix-a13000f67c37
Best regards,
--
Chuck Lever
Looks good Chuck, thanks very much.
With these patches, testing shows no crashes, sosreport no longer hangs, no seq_file errors.
Tested-by: Alexandr Alexandrov <redacted>
cheers,
c.
From: Chuck Lever <cel@kernel.org> Date: 2026-05-05 10:53:39
On 5/5/26 12:49 PM, Calum Mackay wrote:
On 01/05/2026 3:51 pm, Chuck Lever wrote:
quoted
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put
callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/
fs/nfs/exports fd")
The original e_show() repro is now fixed. However, the same
sosreport workload still reproduces a closely related fault on
post-v7.0 mainline (Misbah, ppc64le) and on master.20260424
(internal report, aarch64). In both cases the fault is in
cache_check_rcu() reached through c_show() rather than e_show(),
and the cache_head pointer is plain garbage:
pc : cache_check_rcu+0x40 [sunrpc]
lr : c_show+0x60 [sunrpc]
...faulting on h->flags off h = 0x0000000200000000
c_show() is the generic show callback used by
/proc/net/rpc/<cd>/content for every per-net cache_detail
(auth.unix.ip, auth.unix.gid, nfsd.fh, nfsd.export). Two
bugs combine in that path:
1. cache_unregister_net() / cache_destroy_net() free cd and
cd->hash_table synchronously when the namespace exits. The
/proc/net/rpc/.../content open path takes only a module
reference, so a fd kept open across a netns exit walks a
freed hash_table and returns garbage cache_head pointers.
This is the same hazard that e7fcf179b82d closed for the
/proc/fs/nfs/exports file alone.
2. ip_map_put() drops auth_domain_put() before kfree_rcu(), so
sub-objects can be freed before the RCU grace period -- the
same hazard that 48db892356d6 fixed for svc_export_put() and
expkey_put(). unix_gid_put() does not have this bug
structurally (its put_group_info() runs inside the call_rcu()
callback) but it uses a separate idiom from the other three
caches.
This series replaces the v1 narrow fixes with shared
infrastructure that covers all four cache_detail .put paths
and all three per-cache file types:
Patch 1 hoists nfsd_export_wq up to the sunrpc layer as
sunrpc_cache_wq, exposed through sunrpc_cache_queue_release()
and sunrpc_cache_drain() so all four put callbacks share one
workqueue and one drain primitive.
Patch 2 converts ip_map_put() to the queue_rcu_work() pattern,
moving auth_domain_put() into a deferred ip_map_release() that
runs after the RCU grace period.
Patch 3 unifies unix_gid_put() onto the same pattern for
consistency (not a bug fix on its own).
Patch 4 takes a get_net(cd->net) in content_open(), cache_open(),
and open_flush() and drops it in the matching release helpers,
so cache_destroy_net() cannot run while a sunrpc cache fd is
open.
Series has been compile-tested only.
---
Chuck Lever (6):
SUNRPC: Move cache_initialize() declaration to sunrpc-private
header
SUNRPC: Provide a shared workqueue for cache release callbacks
SUNRPC: Defer ip_map sub-object cleanup past RCU grace period
SUNRPC: Use shared release pattern for the unix_gid cache
SUNRPC: Hold cd->net for the lifetime of cache files
NFSD: Convert nfsd_export_shutdown() to sunrpc_cache_destroy_net()
fs/nfsd/export.c | 45 ++--------------------
fs/nfsd/export.h | 2 -
fs/nfsd/nfsctl.c | 8 +---
include/linux/sunrpc/cache.h | 3 +-
net/sunrpc/cache.c | 90 ++++++++++++++++++++++++++++++++++
++++++++--
net/sunrpc/sunrpc.h | 2 +
net/sunrpc/sunrpc_syms.c | 23 ++++++-----
net/sunrpc/svcauth_unix.c | 46 ++++++++++++----------
8 files changed, 135 insertions(+), 84 deletions(-)
---
base-commit: f3a313ecd1fdab1f5da119db355363b13af6fcac
change-id: 20260430-cache-uaf-fix-a13000f67c37
Best regards,
--
Chuck Lever
Looks good Chuck, thanks very much.
With these patches, testing shows no crashes, sosreport no longer hangs,
no seq_file errors.
Tested-by: Alexandr Alexandrov <redacted>
cheers,
c.
Excellent; pushed with Jeff's R-b and Alexandr's T-b.
--
Chuck Lever
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd")
Back to the problem fixed by this patches, I'm a little confused why
this UAF can be trigged.
Before this patches, svc_export_put show as follow:
368 static void svc_export_put(struct kref *ref)
369 {
370 struct svc_export *exp = container_of(ref, struct svc_export, h.ref);
371
372 path_put(&exp->ex_path);
373 auth_domain_put(exp->ex_client);
374 call_rcu(&exp->ex_rcu, svc_export_release);
375 }
The auth_domain_put function releases ->name using call_rcu, and
path_put may release the dentry also via call_rcu. All of this seems to
prevent e_show from causing a UAF. Could you point out which line in
d_path triggers the issue?
Thanks,
Erkun.
The original e_show() repro is now fixed. However, the same
sosreport workload still reproduces a closely related fault on
post-v7.0 mainline (Misbah, ppc64le) and on master.20260424
(internal report, aarch64). In both cases the fault is in
cache_check_rcu() reached through c_show() rather than e_show(),
and the cache_head pointer is plain garbage:
pc : cache_check_rcu+0x40 [sunrpc]
lr : c_show+0x60 [sunrpc]
...faulting on h->flags off h = 0x0000000200000000
c_show() is the generic show callback used by
/proc/net/rpc/<cd>/content for every per-net cache_detail
(auth.unix.ip, auth.unix.gid, nfsd.fh, nfsd.export). Two
bugs combine in that path:
1. cache_unregister_net() / cache_destroy_net() free cd and
cd->hash_table synchronously when the namespace exits. The
/proc/net/rpc/.../content open path takes only a module
reference, so a fd kept open across a netns exit walks a
freed hash_table and returns garbage cache_head pointers.
This is the same hazard that e7fcf179b82d closed for the
/proc/fs/nfs/exports file alone.
2. ip_map_put() drops auth_domain_put() before kfree_rcu(), so
sub-objects can be freed before the RCU grace period -- the
same hazard that 48db892356d6 fixed for svc_export_put() and
expkey_put(). unix_gid_put() does not have this bug
structurally (its put_group_info() runs inside the call_rcu()
callback) but it uses a separate idiom from the other three
caches.
This series replaces the v1 narrow fixes with shared
infrastructure that covers all four cache_detail .put paths
and all three per-cache file types:
Patch 1 hoists nfsd_export_wq up to the sunrpc layer as
sunrpc_cache_wq, exposed through sunrpc_cache_queue_release()
and sunrpc_cache_drain() so all four put callbacks share one
workqueue and one drain primitive.
Patch 2 converts ip_map_put() to the queue_rcu_work() pattern,
moving auth_domain_put() into a deferred ip_map_release() that
runs after the RCU grace period.
Patch 3 unifies unix_gid_put() onto the same pattern for
consistency (not a bug fix on its own).
Patch 4 takes a get_net(cd->net) in content_open(), cache_open(),
and open_flush() and drops it in the matching release helpers,
so cache_destroy_net() cannot run while a sunrpc cache fd is
open.
Series has been compile-tested only.
---
Chuck Lever (6):
SUNRPC: Move cache_initialize() declaration to sunrpc-private header
SUNRPC: Provide a shared workqueue for cache release callbacks
SUNRPC: Defer ip_map sub-object cleanup past RCU grace period
SUNRPC: Use shared release pattern for the unix_gid cache
SUNRPC: Hold cd->net for the lifetime of cache files
NFSD: Convert nfsd_export_shutdown() to sunrpc_cache_destroy_net()
fs/nfsd/export.c | 45 ++--------------------
fs/nfsd/export.h | 2 -
fs/nfsd/nfsctl.c | 8 +---
include/linux/sunrpc/cache.h | 3 +-
net/sunrpc/cache.c | 90 ++++++++++++++++++++++++++++++++++++++++++--
net/sunrpc/sunrpc.h | 2 +
net/sunrpc/sunrpc_syms.c | 23 ++++++-----
net/sunrpc/svcauth_unix.c | 46 ++++++++++++----------
8 files changed, 135 insertions(+), 84 deletions(-)
---
base-commit: f3a313ecd1fdab1f5da119db355363b13af6fcac
change-id: 20260430-cache-uaf-fix-a13000f67c37
Best regards,
--
Chuck Lever
Hello Erkun -
On Thu, May 7, 2026, at 11:09 AM, yangerkun wrote:
Hi,
在 2026/5/1 22:51, Chuck Lever 写道:
quoted
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd")
Back to the problem fixed by this patches, I'm a little confused why
this UAF can be trigged.
Before this patches, svc_export_put show as follow:
368 static void svc_export_put(struct kref *ref)
369 {
370 struct svc_export *exp = container_of(ref, struct
svc_export, h.ref);
371
372 path_put(&exp->ex_path);
373 auth_domain_put(exp->ex_client);
374 call_rcu(&exp->ex_rcu, svc_export_release);
375 }
The auth_domain_put function releases ->name using call_rcu, and
path_put may release the dentry also via call_rcu. All of this seems to
prevent e_show from causing a UAF. Could you point out which line in
d_path triggers the issue?
The dentry, the mount, and the auth_domain ->name buffer all
end up RCU-freed (dentry_free() and delayed_free_vfsmnt in
fs/, svcauth_unix_domain_release_rcu() in svcauth_unix.c).
The eventual kfree isn't the problem.
The problem is the synchronous teardown inside path_put(),
which runs before svc_export_put() ever reaches its own
call_rcu():
path_put(&exp->ex_path)
-> dput(dentry)
-> __dentry_kill() [if last ref]
-> __d_drop() /* unhashes */
-> dentry_unlink_inode() /* d_inode = NULL */
-> d_op->d_release() if set
-> drops parent d_lockref /* may cascade up */
-> dentry_free() /* call_rcu deferred */
-> mntput(mnt) /* deferred via task_work */
The dentry pointer itself is RCU-safe, so prepend_path()'s walk
of d_parent and d_name doesn't read freed memory. But by the
time the reader gets there, __d_clear_type_and_inode() has
already stored NULL into d_inode, __d_drop() has broken the
hash linkage, and the parent's d_lockref has been decremented
-- which can in turn fire __dentry_kill() on the parent, and
on up the tree. An e_show() that's still inside its cache RCU
read section walks into that half-dismantled state through
seq_path(), and that's the NULL deref Misbah reported.
The earlier fix (2530766492ec, "nfsd: fix UAF when access
ex_uuid or ex_stats") moved the kfree of ex_uuid and ex_stats
into svc_export_release() so those are RCU-safe now.
path_put() and auth_domain_put() couldn't go in there because
both may sleep, and call_rcu callbacks run in softirq context.
This series uses queue_rcu_work() instead: it defers past the
grace period AND runs the callback in process context, so the
sleeping puts move into the deferred path and the window
closes.
--
Chuck Lever
Hello Erkun -
On Thu, May 7, 2026, at 11:09 AM, yangerkun wrote:
quoted
Hi,
在 2026/5/1 22:51, Chuck Lever 写道:
quoted
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd")
Back to the problem fixed by this patches, I'm a little confused why
this UAF can be trigged.
Before this patches, svc_export_put show as follow:
368 static void svc_export_put(struct kref *ref)
369 {
370 struct svc_export *exp = container_of(ref, struct
svc_export, h.ref);
371
372 path_put(&exp->ex_path);
373 auth_domain_put(exp->ex_client);
374 call_rcu(&exp->ex_rcu, svc_export_release);
375 }
The auth_domain_put function releases ->name using call_rcu, and
path_put may release the dentry also via call_rcu. All of this seems to
prevent e_show from causing a UAF. Could you point out which line in
d_path triggers the issue?
The dentry, the mount, and the auth_domain ->name buffer all
end up RCU-freed (dentry_free() and delayed_free_vfsmnt in
fs/, svcauth_unix_domain_release_rcu() in svcauth_unix.c).
The eventual kfree isn't the problem.
The problem is the synchronous teardown inside path_put(),
which runs before svc_export_put() ever reaches its own
call_rcu():
path_put(&exp->ex_path)
-> dput(dentry)
-> __dentry_kill() [if last ref]
-> __d_drop() /* unhashes */
-> dentry_unlink_inode() /* d_inode = NULL */
-> d_op->d_release() if set
-> drops parent d_lockref /* may cascade up */
-> dentry_free() /* call_rcu deferred */
-> mntput(mnt) /* deferred via task_work */
The dentry pointer itself is RCU-safe, so prepend_path()'s walk
of d_parent and d_name doesn't read freed memory. But by the
time the reader gets there, __d_clear_type_and_inode() has
already stored NULL into d_inode, __d_drop() has broken the
hash linkage, and the parent's d_lockref has been decremented
-- which can in turn fire __dentry_kill() on the parent, and
on up the tree. An e_show() that's still inside its cache RCU
read section walks into that half-dismantled state through
seq_path(), and that's the NULL deref Misbah reported.
Thank you for your detailed explanation! Yes, e_show might be called when the state is partially dismantled, but after carefully reviewing the code with dput up to __dentry_kill, I still cannot find anything that could cause this issue. Additionally, the comments for prepend_path indicate that they have already taken into account that the dentry can be removed concurrently. I have also run some tests on my arm64 QEMU, but I couldn't reproduce the problem either. Could you please help me identify the specific line or pointer in the dentry that triggers this use-after-free or null pointer issue?
Maybe I am not be very familiar with the code, which caused me to fail to identify the real root cause. I'm so sorry for that.
265 char *d_path(const struct path *path, char *buf, int buflen)
266 {
267 DECLARE_BUFFER(b, buf, buflen);
268 struct path root;
269
270 /*
271 * We have various synthetic filesystems that never get mounted. On
272 * these filesystems dentries are never used for lookup purposes, and
273 * thus don't need to be hashed. They also don't need a name until a
274 * user wants to identify the object in /proc/pid/fd/. The little hack
275 * below allows us to generate a name for these objects on demand:
276 *
277 * Some pseudo inodes are mountable. When they are mounted
278 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
279 * and instead have d_path return the mounted path.
280 */
281 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
282 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
283 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
284
285 rcu_read_lock();
286 get_fs_root_rcu(current->fs, &root);
287 if (unlikely(d_unlinked(path->dentry)))
288 prepend(&b, " (deleted)", 11);
289 else
290 prepend_char(&b, 0);
291 prepend_path(path, &root, &b);
292 rcu_read_unlock();
293
294 return extract_string(&b);
295 }
The earlier fix (2530766492ec, "nfsd: fix UAF when access
ex_uuid or ex_stats") moved the kfree of ex_uuid and ex_stats
into svc_export_release() so those are RCU-safe now.
path_put() and auth_domain_put() couldn't go in there because
both may sleep, and call_rcu callbacks run in softirq context.
This series uses queue_rcu_work() instead: it defers past the
grace period AND runs the callback in process context, so the
sleeping puts move into the deferred path and the window
closes.
Yeah, I can get this! Thanks again for your detail explanation!
Thanks,
Erkun.
Hello Erkun -
On Thu, May 7, 2026, at 11:09 AM, yangerkun wrote:
quoted
Hi,
在 2026/5/1 22:51, Chuck Lever 写道:
quoted
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of / proc/fs/nfs/exports fd")
Back to the problem fixed by this patches, I'm a little confused why
this UAF can be trigged.
Before this patches, svc_export_put show as follow:
368 static void svc_export_put(struct kref *ref)
369 {
370 struct svc_export *exp = container_of(ref, struct
svc_export, h.ref);
371
372 path_put(&exp->ex_path);
373 auth_domain_put(exp->ex_client);
374 call_rcu(&exp->ex_rcu, svc_export_release);
375 }
The auth_domain_put function releases ->name using call_rcu, and
path_put may release the dentry also via call_rcu. All of this seems to
prevent e_show from causing a UAF. Could you point out which line in
d_path triggers the issue?
The dentry, the mount, and the auth_domain ->name buffer all
end up RCU-freed (dentry_free() and delayed_free_vfsmnt in
fs/, svcauth_unix_domain_release_rcu() in svcauth_unix.c).
The eventual kfree isn't the problem.
The problem is the synchronous teardown inside path_put(),
which runs before svc_export_put() ever reaches its own
call_rcu():
path_put(&exp->ex_path)
-> dput(dentry)
-> __dentry_kill() [if last ref]
-> __d_drop() /* unhashes */
-> dentry_unlink_inode() /* d_inode = NULL */
-> d_op->d_release() if set
-> drops parent d_lockref /* may cascade up */
-> dentry_free() /* call_rcu deferred */
-> mntput(mnt) /* deferred via task_work */
The dentry pointer itself is RCU-safe, so prepend_path()'s walk
of d_parent and d_name doesn't read freed memory. But by the
time the reader gets there, __d_clear_type_and_inode() has
already stored NULL into d_inode, __d_drop() has broken the
hash linkage, and the parent's d_lockref has been decremented
-- which can in turn fire __dentry_kill() on the parent, and
on up the tree. An e_show() that's still inside its cache RCU
read section walks into that half-dismantled state through
seq_path(), and that's the NULL deref Misbah reported.
Thank you for your detailed explanation! Yes, e_show might be called when the state is partially dismantled, but after carefully reviewing the code with dput up to __dentry_kill, I still cannot find anything that could cause this issue. Additionally, the comments for prepend_path indicate that they have already taken into account that the dentry can be removed concurrently. I have also run some tests on my arm64 QEMU, but I couldn't reproduce the problem either. Could you please help me identify the specific line or pointer in the dentry that triggers this use-after-free or null pointer issue?
Maybe I am not be very familiar with the code, which caused me to fail to identify the real root cause. I'm so sorry for that.
265 char *d_path(const struct path *path, char *buf, int buflen)
266 {
267 DECLARE_BUFFER(b, buf, buflen);
268 struct path root;
269
270 /*
271 * We have various synthetic filesystems that never get mounted. On
272 * these filesystems dentries are never used for lookup purposes, and
273 * thus don't need to be hashed. They also don't need a name until a
274 * user wants to identify the object in /proc/pid/fd/. The little hack
275 * below allows us to generate a name for these objects on demand:
276 *
277 * Some pseudo inodes are mountable. When they are mounted
278 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
279 * and instead have d_path return the mounted path.
280 */
281 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
282 (!IS_ROOT(path->dentry) || path->dentry != path->mnt- >mnt_root))
283 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
284
285 rcu_read_lock();
286 get_fs_root_rcu(current->fs, &root);
287 if (unlikely(d_unlinked(path->dentry)))
288 prepend(&b, " (deleted)", 11);
289 else
290 prepend_char(&b, 0);
291 prepend_path(path, &root, &b);
292 rcu_read_unlock();
293
294 return extract_string(&b);
295 }
quoted
The earlier fix (2530766492ec, "nfsd: fix UAF when access
ex_uuid or ex_stats") moved the kfree of ex_uuid and ex_stats
into svc_export_release() so those are RCU-safe now.
path_put() and auth_domain_put() couldn't go in there because
both may sleep, and call_rcu callbacks run in softirq context.
This series uses queue_rcu_work() instead: it defers past the
grace period AND runs the callback in process context, so the
sleeping puts move into the deferred path and the window
closes.
Yeah, I can get this! Thanks again for your detail explanation!
Also, could the scenario described in this commit be triggered again?
commit 69d803c40edeaf94089fbc8751c9b746cdc35044
Author: Yang Erkun [off-list ref]
Date: Mon Dec 16 22:21:52 2024 +0800
nfsd: Revert "nfsd: release svc_expkey/svc_export with rcu_work"
This reverts commit f8c989a0c89a75d30f899a7cabdc14d72522bb8d.
Before this commit, svc_export_put or expkey_put will call path_put with
sync mode. After this commit, path_put will be called with async mode.
And this can lead the unexpected results show as follow.
mkfs.xfs -f /dev/sda
echo "/ *(rw,no_root_squash,fsid=0)" > /etc/exports
echo "/mnt *(rw,no_root_squash,fsid=1)" >> /etc/exports
exportfs -ra
service nfs-server start
mount -t nfs -o vers=4.0 127.0.0.1:/mnt /mnt1
mount /dev/sda /mnt/sda
touch /mnt1/sda/file
exportfs -r
umount /mnt/sda # failed unexcepted
The touch will finally call nfsd_cross_mnt, add refcount to mount, and
then add cache_head. Before this commit, exportfs -r will call
cache_flush to cleanup all cache_head, and path_put in
svc_export_put/expkey_put will be finished with sync mode. So, the
latter umount will always success. However, after this commit, path_put
will be called with async mode, the latter umount may failed, and if
we add some delay, umount will success too. Personally I think this bug
and should be fixed. We first revert before bugfix patch, and then fix
the original bug with a different way.
Fixes: f8c989a0c89a ("nfsd: release svc_expkey/svc_export with rcu_work")
Signed-off-by: Yang Erkun [off-list ref]
Signed-off-by: Chuck Lever [off-list ref]
Hello Erkun -
On Thu, May 7, 2026, at 11:09 AM, yangerkun wrote:
quoted
Hi,
在 2026/5/1 22:51, Chuck Lever 写道:
quoted
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of / proc/fs/nfs/exports fd")
Back to the problem fixed by this patches, I'm a little confused why
this UAF can be trigged.
Before this patches, svc_export_put show as follow:
368 static void svc_export_put(struct kref *ref)
369 {
370 struct svc_export *exp = container_of(ref, struct
svc_export, h.ref);
371
372 path_put(&exp->ex_path);
373 auth_domain_put(exp->ex_client);
374 call_rcu(&exp->ex_rcu, svc_export_release);
375 }
The auth_domain_put function releases ->name using call_rcu, and
path_put may release the dentry also via call_rcu. All of this seems to
prevent e_show from causing a UAF. Could you point out which line in
d_path triggers the issue?
The dentry, the mount, and the auth_domain ->name buffer all
end up RCU-freed (dentry_free() and delayed_free_vfsmnt in
fs/, svcauth_unix_domain_release_rcu() in svcauth_unix.c).
The eventual kfree isn't the problem.
The problem is the synchronous teardown inside path_put(),
which runs before svc_export_put() ever reaches its own
call_rcu():
path_put(&exp->ex_path)
-> dput(dentry)
-> __dentry_kill() [if last ref]
-> __d_drop() /* unhashes */
-> dentry_unlink_inode() /* d_inode = NULL */
-> d_op->d_release() if set
-> drops parent d_lockref /* may cascade up */
-> dentry_free() /* call_rcu deferred */
-> mntput(mnt) /* deferred via task_work */
The dentry pointer itself is RCU-safe, so prepend_path()'s walk
of d_parent and d_name doesn't read freed memory. But by the
time the reader gets there, __d_clear_type_and_inode() has
already stored NULL into d_inode, __d_drop() has broken the
hash linkage, and the parent's d_lockref has been decremented
-- which can in turn fire __dentry_kill() on the parent, and
on up the tree. An e_show() that's still inside its cache RCU
read section walks into that half-dismantled state through
seq_path(), and that's the NULL deref Misbah reported.
Thank you for your detailed explanation! Yes, e_show might be called when the state is partially dismantled, but after carefully reviewing the code with dput up to __dentry_kill, I still cannot find anything that could cause this issue. Additionally, the comments for prepend_path indicate that they have already taken into account that the dentry can be removed concurrently. I have also run some tests on my arm64 QEMU, but I couldn't reproduce the problem either. Could you please help me identify the specific line or pointer in the dentry that triggers this use-after-free or null pointer issue?
Maybe I am not be very familiar with the code, which caused me to fail to identify the real root cause. I'm so sorry for that.
265 char *d_path(const struct path *path, char *buf, int buflen)
266 {
267 DECLARE_BUFFER(b, buf, buflen);
268 struct path root;
269
270 /*
271 * We have various synthetic filesystems that never get mounted. On
272 * these filesystems dentries are never used for lookup purposes, and
273 * thus don't need to be hashed. They also don't need a name until a
274 * user wants to identify the object in /proc/pid/fd/. The little hack
275 * below allows us to generate a name for these objects on demand:
276 *
277 * Some pseudo inodes are mountable. When they are mounted
278 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
279 * and instead have d_path return the mounted path.
280 */
281 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
282 (!IS_ROOT(path->dentry) || path->dentry != path->mnt- >mnt_root))
283 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
284
285 rcu_read_lock();
286 get_fs_root_rcu(current->fs, &root);
287 if (unlikely(d_unlinked(path->dentry)))
288 prepend(&b, " (deleted)", 11);
289 else
290 prepend_char(&b, 0);
291 prepend_path(path, &root, &b);
292 rcu_read_unlock();
293
294 return extract_string(&b);
295 }
quoted
The earlier fix (2530766492ec, "nfsd: fix UAF when access
ex_uuid or ex_stats") moved the kfree of ex_uuid and ex_stats
into svc_export_release() so those are RCU-safe now.
path_put() and auth_domain_put() couldn't go in there because
both may sleep, and call_rcu callbacks run in softirq context.
This series uses queue_rcu_work() instead: it defers past the
grace period AND runs the callback in process context, so the
sleeping puts move into the deferred path and the window
closes.
Yeah, I can get this! Thanks again for your detail explanation!
Also, could the scenario described in this commit be triggered again?
commit 69d803c40edeaf94089fbc8751c9b746cdc35044
Author: Yang Erkun [off-list ref]
Date: Mon Dec 16 22:21:52 2024 +0800
nfsd: Revert "nfsd: release svc_expkey/svc_export with rcu_work"
This reverts commit f8c989a0c89a75d30f899a7cabdc14d72522bb8d.
Before this commit, svc_export_put or expkey_put will call path_put with
sync mode. After this commit, path_put will be called with async mode.
And this can lead the unexpected results show as follow.
mkfs.xfs -f /dev/sda
echo "/ *(rw,no_root_squash,fsid=0)" > /etc/exports
echo "/mnt *(rw,no_root_squash,fsid=1)" >> /etc/exports
exportfs -ra
service nfs-server start
mount -t nfs -o vers=4.0 127.0.0.1:/mnt /mnt1
mount /dev/sda /mnt/sda
touch /mnt1/sda/file
exportfs -r
umount /mnt/sda # failed unexcepted
The touch will finally call nfsd_cross_mnt, add refcount to mount, and
then add cache_head. Before this commit, exportfs -r will call
cache_flush to cleanup all cache_head, and path_put in
svc_export_put/expkey_put will be finished with sync mode. So, the
latter umount will always success. However, after this commit, path_put
will be called with async mode, the latter umount may failed, and if
we add some delay, umount will success too. Personally I think this bug
and should be fixed. We first revert before bugfix patch, and then fix
the original bug with a different way.
Fixes: f8c989a0c89a ("nfsd: release svc_expkey/svc_export with rcu_work")
Signed-off-by: Yang Erkun [off-list ref]
Signed-off-by: Chuck Lever [off-list ref]
After reviewing these two commits:
e7fcf179b82d NFSD: Hold net reference for the lifetime of /proc/fs/nfs/exports fd
48db892356d6 NFSD: Defer sub-object cleanup in export put callbacks
I believe that the issue described in commit e7fcf179b82d might be the
root cause of the null pointer dereferences mentioned in [1]. This is
because we do not call get_net when opening /proc/fs/nfs/exports. As a
result, when the network namespace exits, nfsd_net_exit is triggered.
If, at the same time, the contents of /proc/fs/nfs/exports are being
read, a use-after-free (UAF) can occur on the struct cache_detail. I
think all three bugs referenced in [1] stem from this issue. Therefore,
commit e7fcf179b82d has already addressed the problem. To prevent the
issue described in commit 69d803c40ede, should we consider reverting
commit 48db892356d6 first? Please let me know if I have misunderstood
any aspect of this problem.
Hello Erkun -
On Thu, May 7, 2026, at 11:09 AM, yangerkun wrote:
quoted
Hi,
在 2026/5/1 22:51, Chuck Lever 写道:
quoted
Misbah Anjum reported a use-after-free in cache_check_rcu()
reached through e_show() while sosreport was reading
/proc/fs/nfsd/exports on ppc64le. Two fixes for that report
landed in v7.0:
48db892356d6 ("NFSD: Defer sub-object cleanup in export put callbacks")
e7fcf179b82d ("NFSD: Hold net reference for the lifetime of / proc/fs/nfs/exports fd")
Back to the problem fixed by this patches, I'm a little confused why
this UAF can be trigged.
Before this patches, svc_export_put show as follow:
368 static void svc_export_put(struct kref *ref)
369 {
370 struct svc_export *exp = container_of(ref, struct
svc_export, h.ref);
371
372 path_put(&exp->ex_path);
373 auth_domain_put(exp->ex_client);
374 call_rcu(&exp->ex_rcu, svc_export_release);
375 }
The auth_domain_put function releases ->name using call_rcu, and
path_put may release the dentry also via call_rcu. All of this seems to
prevent e_show from causing a UAF. Could you point out which line in
d_path triggers the issue?
The dentry, the mount, and the auth_domain ->name buffer all
end up RCU-freed (dentry_free() and delayed_free_vfsmnt in
fs/, svcauth_unix_domain_release_rcu() in svcauth_unix.c).
The eventual kfree isn't the problem.
The problem is the synchronous teardown inside path_put(),
which runs before svc_export_put() ever reaches its own
call_rcu():
path_put(&exp->ex_path)
-> dput(dentry)
-> __dentry_kill() [if last ref]
-> __d_drop() /* unhashes */
-> dentry_unlink_inode() /* d_inode = NULL */
-> d_op->d_release() if set
-> drops parent d_lockref /* may cascade up */
-> dentry_free() /* call_rcu deferred */
-> mntput(mnt) /* deferred via task_work */
The dentry pointer itself is RCU-safe, so prepend_path()'s walk
of d_parent and d_name doesn't read freed memory. But by the
time the reader gets there, __d_clear_type_and_inode() has
already stored NULL into d_inode, __d_drop() has broken the
hash linkage, and the parent's d_lockref has been decremented
-- which can in turn fire __dentry_kill() on the parent, and
on up the tree. An e_show() that's still inside its cache RCU
read section walks into that half-dismantled state through
seq_path(), and that's the NULL deref Misbah reported.
Thank you for your detailed explanation! Yes, e_show might be called when the state is partially dismantled, but after carefully reviewing the code with dput up to __dentry_kill, I still cannot find anything that could cause this issue. Additionally, the comments for prepend_path indicate that they have already taken into account that the dentry can be removed concurrently. I have also run some tests on my arm64 QEMU, but I couldn't reproduce the problem either. Could you please help me identify the specific line or pointer in the dentry that triggers this use-after-free or null pointer issue?
Maybe I am not be very familiar with the code, which caused me to fail to identify the real root cause. I'm so sorry for that.
265 char *d_path(const struct path *path, char *buf, int buflen)
266 {
267 DECLARE_BUFFER(b, buf, buflen);
268 struct path root;
269
270 /*
271 * We have various synthetic filesystems that never get mounted. On
272 * these filesystems dentries are never used for lookup purposes, and
273 * thus don't need to be hashed. They also don't need a name until a
274 * user wants to identify the object in /proc/pid/fd/. The little hack
275 * below allows us to generate a name for these objects on demand:
276 *
277 * Some pseudo inodes are mountable. When they are mounted
278 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
279 * and instead have d_path return the mounted path.
280 */
281 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
282 (!IS_ROOT(path->dentry) || path->dentry != path->mnt- >mnt_root))
283 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
284
285 rcu_read_lock();
286 get_fs_root_rcu(current->fs, &root);
287 if (unlikely(d_unlinked(path->dentry)))
288 prepend(&b, " (deleted)", 11);
289 else
290 prepend_char(&b, 0);
291 prepend_path(path, &root, &b);
292 rcu_read_unlock();
293
294 return extract_string(&b);
295 }
quoted
The earlier fix (2530766492ec, "nfsd: fix UAF when access
ex_uuid or ex_stats") moved the kfree of ex_uuid and ex_stats
into svc_export_release() so those are RCU-safe now.
path_put() and auth_domain_put() couldn't go in there because
both may sleep, and call_rcu callbacks run in softirq context.
This series uses queue_rcu_work() instead: it defers past the
grace period AND runs the callback in process context, so the
sleeping puts move into the deferred path and the window
closes.
Yeah, I can get this! Thanks again for your detail explanation!
Also, could the scenario described in this commit be triggered again?
commit 69d803c40edeaf94089fbc8751c9b746cdc35044
Author: Yang Erkun [off-list ref]
Date: Mon Dec 16 22:21:52 2024 +0800
nfsd: Revert "nfsd: release svc_expkey/svc_export with rcu_work"
This reverts commit f8c989a0c89a75d30f899a7cabdc14d72522bb8d.
Before this commit, svc_export_put or expkey_put will call path_put with
sync mode. After this commit, path_put will be called with async mode.
And this can lead the unexpected results show as follow.
mkfs.xfs -f /dev/sda
echo "/ *(rw,no_root_squash,fsid=0)" > /etc/exports
echo "/mnt *(rw,no_root_squash,fsid=1)" >> /etc/exports
exportfs -ra
service nfs-server start
mount -t nfs -o vers=4.0 127.0.0.1:/mnt /mnt1
mount /dev/sda /mnt/sda
touch /mnt1/sda/file
exportfs -r
umount /mnt/sda # failed unexcepted
The touch will finally call nfsd_cross_mnt, add refcount to mount, and
then add cache_head. Before this commit, exportfs -r will call
cache_flush to cleanup all cache_head, and path_put in
svc_export_put/expkey_put will be finished with sync mode. So, the
latter umount will always success. However, after this commit, path_put
will be called with async mode, the latter umount may failed, and if
we add some delay, umount will success too. Personally I think this bug
and should be fixed. We first revert before bugfix patch, and then fix
the original bug with a different way.
Fixes: f8c989a0c89a ("nfsd: release svc_expkey/svc_export with rcu_work")
Signed-off-by: Yang Erkun [off-list ref]
Signed-off-by: Chuck Lever [off-list ref]
After reviewing these two commits:
e7fcf179b82d NFSD: Hold net reference for the lifetime of /proc/fs/nfs/ exports fd
48db892356d6 NFSD: Defer sub-object cleanup in export put callbacks
I believe that the issue described in commit e7fcf179b82d might be the
root cause of the null pointer dereferences mentioned in [1]. This is
because we do not call get_net when opening /proc/fs/nfs/exports. As a
result, when the network namespace exits, nfsd_net_exit is triggered.
If, at the same time, the contents of /proc/fs/nfs/exports are being
read, a use-after-free (UAF) can occur on the struct cache_detail. I
think all three bugs referenced in [1] stem from this issue. Therefore,
commit e7fcf179b82d has already addressed the problem. To prevent the
issue described in commit 69d803c40ede, should we consider reverting
commit 48db892356d6 first? Please let me know if I have misunderstood
any aspect of this problem.
Locally, I wrote a stable regression test case. I also reverted to commit 9189d23b835cec646ba5010db35d1557a77c5857 (which is before commits 2862eee078a4 "SUNRPC: make sure cache entry active before cache_show" and be8f982c369c "nfsd: make sure exp active before svc_export_show"). Even then, a panic can still be triggered without any actual export path...
Hi Erkun -
On Fri, May 8, 2026, at 9:00 AM, yangerkun wrote:
在 2026/5/8 16:16, yangerkun 写道:
quoted
在 2026/5/8 11:08, yangerkun 写道:
After reviewing these two commits:
e7fcf179b82d NFSD: Hold net reference for the lifetime of /proc/fs/nfs/
exports fd
48db892356d6 NFSD: Defer sub-object cleanup in export put callbacks
I believe that the issue described in commit e7fcf179b82d might be the
root cause of the null pointer dereferences mentioned in [1].
That's where I landed too. e7fcf179b82d closed the specific
oops Misbah hit on /proc/fs/nfs/exports. The matching patch
in this series is 5/6 ("SUNRPC: Hold cd->net for the lifetime
of cache files"), which extends the same get_net()/put_net()
guard to the sunrpc cache files at
/proc/net/rpc/<cache>/{content,channel,flush} .
Those open helpers had the same hole; sosreport just hit the
nfsd-specific file first because it reads /proc/fs/nfsd/exports.
Patch 5/6's changelog pins down the deref site you asked
about: cache_check_rcu() faults reading h->flags off a
garbage cache_head returned by __cache_seq_start() walking a
cd->hash_table that cache_destroy_net() already freed. Not a
dentry deref. The dentry-teardown path is a separate failure
mode that 48db892356d6 closed for the export and expkey caches.
quoted
To prevent the
issue described in commit 69d803c40ede, should we consider reverting
commit 48db892356d6 first?
Not for this series. Patches 3/6 and 4/6 don't add any new
path_put deferral; their commit messages call them out as
consistency changes, not bug fixes. ip_map holds only an
auth_domain reference and unix_gid holds only a group_info,
so neither cache reaches mntput from the deferred release.
The exportfs-r-then-umount sequence isn't touched by this
series.
The svc_export and svc_expkey path_put deferral lives in
48db892356d6, which is already in v7.0. If the umount window
from 69d803c40ede is still reachable through that commit,
that's a regression in 48db892356d6 and worth a separate
thread.
Locally, I wrote a stable regression test case. I also reverted to
commit 9189d23b835cec646ba5010db35d1557a77c5857 (which is before commits
2862eee078a4 "SUNRPC: make sure cache entry active before cache_show"
and be8f982c369c "nfsd: make sure exp active before svc_export_show").
Even then, a panic can still be triggered without any actual export path...
That fits 5/6's failure mode. Without an export no svc_export
or svc_expkey entry is populated, but rpc.mountd reads
auth.unix.ip/content and auth.unix.gid/content directly,
and on a pre-5/6 tree the open helpers in cache.c hold no
reference on cd->net. cache_destroy_net() at namespace exit
then races a reader still inside cache_seq_start_rcu(), and
the reader walks a freed cd->hash_table.
Could you share the reproducer and the panic stack trace?
If the fault is in cache_check_rcu() through one of the
sunrpc cache files, that confirms 5/6 is the right fix, and
I'll happily carry your Tested-by on it.
--
Chuck Lever
Hi Erkun -
On Fri, May 8, 2026, at 9:00 AM, yangerkun wrote:
quoted
在 2026/5/8 16:16, yangerkun 写道:
quoted
在 2026/5/8 11:08, yangerkun 写道:
After reviewing these two commits:
e7fcf179b82d NFSD: Hold net reference for the lifetime of /proc/fs/nfs/
exports fd
48db892356d6 NFSD: Defer sub-object cleanup in export put callbacks
I believe that the issue described in commit e7fcf179b82d might be the
root cause of the null pointer dereferences mentioned in [1].
That's where I landed too. e7fcf179b82d closed the specific
oops Misbah hit on /proc/fs/nfs/exports. The matching patch
Yeah!
in this series is 5/6 ("SUNRPC: Hold cd->net for the lifetime
of cache files"), which extends the same get_net()/put_net()
guard to the sunrpc cache files at
/proc/net/rpc/<cache>/{content,channel,flush} .
Those open helpers had the same hole; sosreport just hit the
nfsd-specific file first because it reads /proc/fs/nfsd/exports.
Hmm... /proc/net is always a symlink to /proc/self/net. After opening /proc/net/rpc/<cache>/content and attempting to read it, the proc_reg_read function calls use_pde before pde_read. This sequence can prevent a race condition because nfsd_export_shutdown leads to cache_unregister_net, which calls remove_cache_proc_entries, then proc_remove, and eventually proc_entry_rundown. The proc_entry_rundown function waits until unuse_pde is called in proc_reg_read. Therefore, I'm not sure if forgetting to call get_net when opening /proc/net/rpc/<cache>/content is the root cause of the null pointer in c_show. I've tried to find any other possible root causes but have been unsuccessful. Sorry....
Patch 5/6's changelog pins down the deref site you asked
about: cache_check_rcu() faults reading h->flags off a
garbage cache_head returned by __cache_seq_start() walking a
cd->hash_table that cache_destroy_net() already freed. Not a
dentry deref. The dentry-teardown path is a separate failure
mode that 48db892356d6 closed for the export and expkey caches.
quoted
quoted
To prevent the
issue described in commit 69d803c40ede, should we consider reverting
commit 48db892356d6 first?
Not for this series. Patches 3/6 and 4/6 don't add any new
path_put deferral; their commit messages call them out as
consistency changes, not bug fixes. ip_map holds only an
auth_domain reference and unix_gid holds only a group_info,
so neither cache reaches mntput from the deferred release.
The exportfs-r-then-umount sequence isn't touched by this
series.
The svc_export and svc_expkey path_put deferral lives in
48db892356d6, which is already in v7.0. If the umount window
from 69d803c40ede is still reachable through that commit,
that's a regression in 48db892356d6 and worth a separate
thread.
Yeah! Totally agree!
quoted
Locally, I wrote a stable regression test case. I also reverted to
commit 9189d23b835cec646ba5010db35d1557a77c5857 (which is before commits
2862eee078a4 "SUNRPC: make sure cache entry active before cache_show"
and be8f982c369c "nfsd: make sure exp active before svc_export_show").
Even then, a panic can still be triggered without any actual export path...
That fits 5/6's failure mode. Without an export no svc_export
or svc_expkey entry is populated, but rpc.mountd reads
auth.unix.ip/content and auth.unix.gid/content directly,
and on a pre-5/6 tree the open helpers in cache.c hold no
reference on cd->net. cache_destroy_net() at namespace exit
then races a reader still inside cache_seq_start_rcu(), and
the reader walks a freed cd->hash_table.
Could you share the reproducer and the panic stack trace?
If the fault is in cache_check_rcu() through one of the
sunrpc cache files, that confirms 5/6 is the right fix, and
I'll happily carry your Tested-by on it.
The shell(Created will AI assist):
#!/bin/bash
#
# Test for e7fcf179b82d ("NFSD: Hold net reference for ...")
#
# Reproduces the scenario described in the commit:
# 1. Process opens /proc/fs/nfsd/exports in netns A
# 2. Process leaves A (joins B), emptying A
# 3. ip netns del A triggers nfsd_export_shutdown → cache_detail freed
# 4. Process reads from still-open fd → UAF on UNFIXED kernel
#
# On current kernel (with e7fcf179b82d applied):
# get_net in exports_net_open prevents netns A from being destroyed
# → read succeeds safely (test output: "SUCCESS")
#
# On kernel WITHOUT e7fcf179b82d:
# No get_net → A destroyed → read triggers UAF:
# - KASAN: use-after-free, or
# - NULL deref, or
# - slab corruption (ASCII strings like "cap_type", "libz.so.")
#
# Usage: sudo ./test_nfsd_exports_uaf.sh
set -e -u
NS_A="nfsd_test_A_$$"
NS_B="nfsd_test_B_$$"
SYNC="/tmp/nfsd_uaf_sync_$$"
GO="/tmp/nfsd_uaf_go_$$"
REPRO="/tmp/uaf_repro_$$"
cleanup() {
set +e
kill $REPRO_PID 2>/dev/null || true
wait $REPRO_PID 2>/dev/null || true
ip netns del "$NS_B" 2>/dev/null || true
ip netns del "$NS_A" 2>/dev/null || true
rm -f "$REPRO" "$SYNC" "$GO"
}
trap cleanup EXIT
echo "=== Reproduce e7fcf179b82d scenario ==="
# --- Setup ---
echo "[setup] creating netns A and B..."
ip netns add "$NS_A"
ip netns add "$NS_B"
echo "[setup] loading nfsd..."
modprobe nfsd || true
echo "[setup] compiling repro..."
gcc -o "$REPRO" /tmp/uaf_repro.c 2>/dev/null || \
gcc -o "$REPRO" -x c - <<'SRCEOF' 2>/dev/null
#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <sched.h>
#include <signal.h>
#include <string.h>
#include <sys/types.h>
#include <sys/stat.h>
static volatile sig_atomic_t go_flag = 0;
static void handler(int sig) { go_flag = 1; }
int main(int argc, char *argv[]) {
const char *netns_b = argv[1], *sync_f = argv[2], *go_f = argv[3];
int fd, nsfd; ssize_t n; char buf[4096];
fd = open("/proc/fs/nfs/exports", O_RDONLY);
if (fd < 0) { perror("open exports debug"); return 1; }
fprintf(stderr, "[repro] opened exports (fd=%d) in netns A\n", fd);
nsfd = open(netns_b, O_RDONLY);
if (nsfd < 0) { perror("open B"); return 1; }
if (setns(nsfd, CLONE_NEWNET) < 0) { perror("setns"); return 1; }
close(nsfd);
fprintf(stderr, "[repro] moved to B; A has no processes\n");
close(open(sync_f, O_CREAT | O_WRONLY, 0666));
signal(SIGCONT, handler);
while (!go_flag) { struct stat st; if (stat(go_f, &st) == 0) break; pause(); }
fprintf(stderr, "[repro] reading exports fd...\n");
lseek(fd, 0, SEEK_SET);
sleep(1);
n = read(fd, buf, sizeof(buf)-1);
if (n < 0) { perror("read"); close(fd); return 1; }
buf[n] = '\0';
fprintf(stderr, "[repro] SUCCESS: read %zd bytes (no UAF)\n", n);
close(fd);
return 0;
}
SRCEOF
# --- Run repro inside A ---
rm -f "$SYNC" "$GO"
echo "[test] starting repro inside netns A..."
ip netns exec "$NS_A" "$REPRO" /var/run/netns/"$NS_B" "$SYNC" "$GO" &
REPRO_PID=$!
# --- Wait for repro to move to B ---
echo "[test] waiting for repro to signal that A is empty..."
for i in $(seq 1 30); do
if [ -f "$SYNC" ]; then break; fi
if ! kill -0 $REPRO_PID 2>/dev/null; then
echo "[FAIL] repro exited prematurely"
wait $REPRO_PID || true
exit 1
fi
sleep 0.2
done
if [ ! -f "$SYNC" ]; then
echo "[FAIL] timeout waiting for repro"
exit 1
fi
echo "[test] repro moved to B"
# --- Destroy netns A ---
echo "[test] destroying netns A (ip netns del $NS_A)..."
set +e
ip netns del "$NS_A" 2>&1
RC=$?
set -e
if [ $RC -eq 0 ]; then
echo "[test] 'ip netns del $NS_A' returned success"
else
echo "[test] 'ip netns del $NS_A' returned $RC"
fi
# --- Signal repro to read from the exports fd ---
echo "[test] signaling repro to read from exports fd..."
touch "$GO"
kill -CONT $REPRO_PID 2>/dev/null || true
# --- Wait for repro and check result ---
set +e
wait $REPRO_PID
RC=$?
set -e
if [ $RC -eq 0 ]; then
echo ""
echo "=== TEST PASSED: no UAF detected (kernel has e7fcf179b82d fix) ==="
echo " get_net() holds netns A alive while the fd is open."
else
echo ""
echo "=== TEST FAILED with exit code $RC ==="
echo " Possible UAF or other error."
echo " If running on a kernel WITHOUT e7fcf179b82d, this crash is EXPECTED."
fi
exit $RC
Panic show as follow with commit:
commit 9189d23b835cec646ba5010db35d1557a77c5857 (HEAD -> master)
Author: Chuck Lever [off-list ref]
Date: Thu Oct 17 09:36:31 2024 -0400
lockd: Remove unneeded initialization of file_lock::c.flc_flags
localhost login: [ 39.462598][ T579] ==================================================================
[202/363]
[ 39.463541][ T579] BUG: KASAN: slab-use-after-free in cache_seq_next_rcu+0xa4/0x180 [sunrpc]
[ 39.464551][ T579] Read of size 4 at addr ffff00000fbe8408 by task uaf_repro_563/579
[ 39.465291][ T579]
[ 39.465513][ T579] CPU: 1 UID: 0 PID: 579 Comm: uaf_repro_563 Not tainted 6.12.0-rc7+ #17
[ 39.466349][ T579] Hardware name: linux,dummy-virt (DT)
[ 39.466897][ T579] Call trace:
[ 39.467224][ T579] dump_backtrace+0xa4/0x140
[ 39.467742][ T579] show_stack+0x20/0x38
[ 39.468156][ T579] dump_stack_lvl+0x80/0xf8
[ 39.468694][ T579] print_report+0xfc/0x5c8
[ 39.469237][ T579] kasan_report+0x78/0xc8
[ 39.469676][ T579] __asan_load4+0x9c/0xc0
[ 39.470115][ T579] cache_seq_next_rcu+0xa4/0x180 [sunrpc]
[ 39.470842][ T579] seq_read_iter+0x4a0/0x6c0
[ 39.471355][ T579] seq_read+0x194/0x218
[ 39.471770][ T579] proc_reg_read+0x110/0x198
[ 39.472235][ T579] vfs_read+0x150/0x490
[ 39.472656][ T579] ksys_read+0xd4/0x198
[ 39.473070][ T579] __arm64_sys_read+0x4c/0x68
[ 39.473537][ T579] invoke_syscall+0x64/0x188
[ 39.473992][ T579] el0_svc_common.constprop.1+0xd8/0x158
[ 39.474558][ T579] do_el0_svc+0x38/0x50
[ 39.474981][ T579] el0_svc+0x34/0xc0
[ 39.475422][ T579] el0t_64_sync_handler+0xa0/0xc8
[ 39.475933][ T579] el0t_64_sync+0x188/0x190
[ 39.476385][ T579]
[ 39.476618][ T579] Allocated by task 566:
[ 39.477087][ T579] kasan_save_stack+0x2c/0x58
[ 39.477561][ T579] kasan_save_track+0x20/0x40
[ 39.478030][ T579] kasan_save_alloc_info+0x40/0x58
[ 39.478539][ T579] __kasan_kmalloc+0xa0/0xb8
[ 39.478997][ T579] __kmalloc_node_track_caller_noprof+0x194/0x370
[ 39.479646][ T579] kmemdup_noprof+0x34/0x68
[ 39.480094][ T579] cache_create_net+0x30/0x108 [sunrpc]
[ 39.480800][ T579] nfsd_export_init+0x78/0x188 [nfsd]
[ 39.481505][ T579] nfsd_net_init+0x50/0x1e8 [nfsd]
[ 39.482136][ T579] ops_init+0xcc/0x210
[ 39.482615][ T579] register_pernet_operations+0x218/0x348
[ 39.483180][ T579] register_pernet_subsys+0x38/0x60
[ 39.483698][ T579] 0xffffb6c9bf5b90c0
[ 39.484096][ T579] do_one_initcall+0xa8/0x3c8
[ 39.484563][ T579] do_init_module+0x100/0x378
[ 39.485070][ T579] load_module+0x2d78/0x2e80
[ 39.485532][ T579] init_module_from_file+0xec/0x148
[ 39.486044][ T579] __arm64_sys_finit_module+0x394/0x618
[ 39.486604][ T579] invoke_syscall+0x64/0x188
[ 39.487065][ T579] el0_svc_common.constprop.1+0xd8/0x158
[ 39.487629][ T579] do_el0_svc+0x38/0x50
[ 39.488044][ T579] el0_svc+0x34/0xc0
[ 39.488437][ T579] el0t_64_sync_handler+0xa0/0xc8
[ 39.488939][ T579] el0t_64_sync+0x188/0x190
[ 39.489398][ T579]
[ 39.489635][ T579] Freed by task 53:
[ 39.490013][ T579] kasan_save_stack+0x2c/0x58
[ 39.490479][ T579] kasan_save_track+0x20/0x40
[ 39.490948][ T579] kasan_save_free_info+0x4c/0x78
[ 39.491449][ T579] __kasan_slab_free+0x50/0x70
[ 39.491924][ T579] kfree+0x160/0x310
[ 39.492312][ T579] cache_destroy_net+0x34/0x50 [sunrpc]
[ 39.493015][ T579] nfsd_export_shutdown+0xc0/0x150 [nfsd]
[ 39.493711][ T579] nfsd_net_exit+0x68/0x88 [nfsd]
[ 39.494338][ T579] ops_exit_list.isra.13+0x64/0xc0
[ 39.494856][ T579] cleanup_net+0x508/0x788
[ 39.495300][ T579] process_scheduled_works+0x3d8/0x7e8
[ 39.495895][ T579] worker_thread+0x29c/0x630
[ 39.496364][ T579] kthread+0x170/0x188
[ 39.496773][ T579] ret_from_fork+0x10/0x20
[ 39.497217][ T579]
[ 39.497453][ T579] The buggy address belongs to the object at ffff00000fbe8400
I have try to replace
fd = open("/proc/fs/nfs/exports", O_RDONLY);
with
fd = open("/proc/fs/nfs/exports", O_RDONLY);
No c_show UAF trigger...
Hi Erkun -
On Sat, May 9, 2026, at 5:41 AM, yangerkun wrote:
Hmm... /proc/net is always a symlink to /proc/self/net. After opening
/proc/net/rpc/<cache>/content and attempting to read it, the
proc_reg_read function calls use_pde before pde_read. This sequence can
prevent a race condition because nfsd_export_shutdown leads to
cache_unregister_net, which calls remove_cache_proc_entries, then
proc_remove, and eventually proc_entry_rundown. The proc_entry_rundown
function waits until unuse_pde is called in proc_reg_read. Therefore,
I'm not sure if forgetting to call get_net when opening
/proc/net/rpc/<cache>/content is the root cause of the null pointer in
c_show.
Walked the synchronization. You're right.
cache_unregister_net() calls remove_cache_proc_entries(),
which runs proc_remove(); remove_proc_subtree() then invokes
proc_entry_rundown() on each per-cache file. Rundown does
atomic_add_return(BIAS, &de->in_use), where BIAS = -1U << 31.
No active readers means the post-add value equals BIAS and
rundown returns at once. Readers present means the value
exceeds BIAS, and wait_for_completion() blocks until the last
unuse_pde() decrements the counter to exactly BIAS and signals
the completion. atomic_inc_unless_negative() in use_pde() then
fails, so any later read() on a still-open userspace fd
returns -EIO without touching cd. close_pdeo() forces release
on the remaining openers while cd is still valid.
cache_destroy_net() runs only after that whole sequence has
finished, so cd->hash_table is freed once no reader can be
inside cache_seq_*_rcu() and no fd can dereference cd through
a release callback.
The 5/6 changelog overstates the window. Your reproducer
opens /proc/fs/nfs/exports through exports_nfsd_open(), which
bypasses use_pde() and is the path e7fcf179b82d closed. The
sunrpc cache files reach c_show through proc_reg_read(), which
goes through use_pde()/unuse_pde() and is covered by rundown.
5/6 doesn't close the hazard its changelog describes.
Patch 3/6 is what matches Misbah's reproducer. Pre-series
ip_map_put() drops auth_domain_put() synchronously, with only
the ip_map free deferred:
auth_domain_put(&im->m_client->h); /* synchronous */
kfree_rcu(im, m_rcu);
A reader walking auth.unix.ip/content under rcu_read_lock()
can dereference im->m_client after the auth_domain has been
freed. Same shape as 48db892356d6's svc_export fix, applied
to ip_map. 3/6 moves auth_domain_put() into a deferred
ip_map_release() scheduled via queue_rcu_work(), so the
sub-object free waits for the grace period.
For v2: re-test Misbah's reproducer with patches 1-4 and 6
only and see whether 3/6 alone closes the crash. If it does,
drop 5/6; if it doesn't, reframe 5/6 as a consistency change
without the UAF claim (and without the behavioral change that
pins a netns alive while a debug fd is open). Either way, the
cover letter needs a rewrite to match.
Thanks for your analysis and review.
--
Chuck Lever