[0/3] fix removal of nfsd listeners

Message

Olga Kornievskaia Jan. 15, 2025, 11:24 p.m. UTC

Currently if a root user using nfsdctl command tries to remove a particular
listener from the list of previously added ones, then starting the nfsd
leads to the following problem:

[  158.835354] refcount_t: addition on 0; use-after-free.
[  158.835603] WARNING: CPU: 2 PID: 9145 at lib/refcount.c:25 refcount_warn_saturate+0x160/0x1a0
[  158.836017] Modules linked in: rpcrdma rdma_cm iw_cm ib_cm ib_core nfsd auth_rpcgss nfs_acl lockd grace overlay isofs uinput snd_seq_dummy snd_hrtimer nft_fib_inet nft_fib_ipv4 nft_fib_ipv6 nft_fib nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 rfkill ip_set nf_tables qrtr sunrpc vfat fat uvcvideo videobuf2_vmalloc videobuf2_memops uvc videobuf2_v4l2 videodev videobuf2_common snd_hda_codec_generic mc e1000e snd_hda_intel snd_intel_dspcfg snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore sg loop dm_multipath dm_mod nfnetlink vsock_loopback vmw_vsock_virtio_transport_common vmw_vsock_vmci_transport vmw_vmci vsock xfs libcrc32c crct10dif_ce ghash_ce vmwgfx sha2_ce sha256_arm64 sr_mod sha1_ce cdrom nvme drm_client_lib drm_ttm_helper ttm nvme_core drm_kms_helper nvme_auth drm fuse
[  158.840093] CPU: 2 UID: 0 PID: 9145 Comm: nfsd Kdump: loaded Tainted: G    B   W          6.13.0-rc6+ #7
[  158.840624] Tainted: [B]=BAD_PAGE, [W]=WARN
[  158.840802] Hardware name: VMware, Inc. VMware20,1/VBSA, BIOS VMW201.00V.24006586.BA64.2406042154 06/04/2024
[  158.841220] pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
[  158.841563] pc : refcount_warn_saturate+0x160/0x1a0
[  158.841780] lr : refcount_warn_saturate+0x160/0x1a0
[  158.842000] sp : ffff800089be7d80
[  158.842147] x29: ffff800089be7d80 x28: ffff00008e68c148 x27: ffff00008e68c148
[  158.842492] x26: ffff0002e3b5c000 x25: ffff600011cd1829 x24: ffff00008653c010
[  158.842832] x23: ffff00008653c000 x22: 1fffe00011cd1829 x21: ffff00008653c028
[  158.843175] x20: 0000000000000002 x19: ffff00008653c010 x18: 0000000000000000
[  158.843505] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[  158.843836] x14: 0000000000000000 x13: 0000000000000001 x12: ffff600050a26493
[  158.844143] x11: 1fffe00050a26492 x10: ffff600050a26492 x9 : dfff800000000000
[  158.844475] x8 : 00009fffaf5d9b6e x7 : ffff000285132493 x6 : 0000000000000001
[  158.844823] x5 : ffff000285132490 x4 : ffff600050a26493 x3 : ffff8000805e72bc
[  158.845174] x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff000098588000
[  158.845528] Call trace:
[  158.845658]  refcount_warn_saturate+0x160/0x1a0 (P)
[  158.845894]  svc_recv+0x58c/0x680 [sunrpc]
[  158.846183]  nfsd+0x1fc/0x348 [nfsd]
[  158.846390]  kthread+0x274/0x2f8
[  158.846546]  ret_from_fork+0x10/0x20
[  158.846714] ---[ end trace 0000000000000000 ]---

nfsd_nl_listener_set_doit() would manipulate the list of transports of
server's sv_permsocks and svc_xprt_close() the specified listener but
the other list of transports (server's sp_xprts list) would not be
changed leading to the problem.

the other problem is that sp_xprt is a lwq structure of lockless
list which does not have an ability to remove a single entry from
the list.

this patch series addis a function to remove a single entry, then modifies
nfsd_nl_listener_set_doit() to make sure the to-be-removed listener is
removed from both lists and then it also ensures that the remaining
listeners are added back in the correct state.

Olga Kornievskaia (3):
  llist: add ability to remove a particular entry from the list
  SUNRPC: add ability to remove specific server transport
  nfsd: fix management of listener transports

 fs/nfsd/nfsctl.c           |  4 +++-
 include/linux/llist.h      | 36 ++++++++++++++++++++++++++++++++++++
 include/linux/sunrpc/svc.h |  1 +
 net/sunrpc/svc_xprt.c      | 11 +++++++++++
 4 files changed, 51 insertions(+), 1 deletion(-)