[PATCHv3 net 0/3] hsr: fix lock warnings

[Hangbin Liu <liuhangbin@gmail.com>]

hsr_for_each_port is called in many places without holding the RCU read
lock, this may trigger warnings on debug kernels like:

  [   40.457015] [  T201] WARNING: suspicious RCU usage
  [   40.457020] [  T201] 6.17.0-rc2-virtme #1 Not tainted
  [   40.457025] [  T201] -----------------------------
  [   40.457029] [  T201] net/hsr/hsr_main.c:137 RCU-list traversed in non-reader section!!
  [   40.457036] [  T201]
                          other info that might help us debug this:

  [   40.457040] [  T201]
                          rcu_scheduler_active = 2, debug_locks = 1
  [   40.457045] [  T201] 2 locks held by ip/201:
  [   40.457050] [  T201]  #0: ffffffff93040a40 (&ops->srcu){.+.+}-{0:0}, at: rtnl_link_ops_get+0xf2/0x280
  [   40.457080] [  T201]  #1: ffffffff92e7f968 (rtnl_mutex){+.+.}-{4:4}, at: rtnl_newlink+0x5e1/0xb20
  [   40.457102] [  T201]
                          stack backtrace:
  [   40.457108] [  T201] CPU: 2 UID: 0 PID: 201 Comm: ip Not tainted 6.17.0-rc2-virtme #1 PREEMPT(full)
  [   40.457114] [  T201] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
  [   40.457117] [  T201] Call Trace:
  [   40.457120] [  T201]  <TASK>
  [   40.457126] [  T201]  dump_stack_lvl+0x6f/0xb0
  [   40.457136] [  T201]  lockdep_rcu_suspicious.cold+0x4f/0xb1
  [   40.457148] [  T201]  hsr_port_get_hsr+0xfe/0x140
  [   40.457158] [  T201]  hsr_add_port+0x192/0x940
  [   40.457167] [  T201]  ? __pfx_hsr_add_port+0x10/0x10
  [   40.457176] [  T201]  ? lockdep_init_map_type+0x5c/0x270
  [   40.457189] [  T201]  hsr_dev_finalize+0x4bc/0xbf0
  [   40.457204] [  T201]  hsr_newlink+0x3c3/0x8f0
  [   40.457212] [  T201]  ? __pfx_hsr_newlink+0x10/0x10
  [   40.457222] [  T201]  ? rtnl_create_link+0x173/0xe40
  [   40.457233] [  T201]  rtnl_newlink_create+0x2cf/0x750
  [   40.457243] [  T201]  ? __pfx_rtnl_newlink_create+0x10/0x10
  [   40.457247] [  T201]  ? __dev_get_by_name+0x12/0x50
  [   40.457252] [  T201]  ? rtnl_dev_get+0xac/0x140
  [   40.457259] [  T201]  ? __pfx_rtnl_dev_get+0x10/0x10
  [   40.457285] [  T201]  __rtnl_newlink+0x22c/0xa50
  [   40.457305] [  T201]  rtnl_newlink+0x637/0xb20

Adding rcu_read_lock() for all hsr_for_each_port() looks confusing.

Introduce a new helper, hsr_for_each_port_rtnl(), that assumes the
RTNL lock is held. This allows callers in suitable contexts to iterate
ports safely without explicit RCU locking.

Other code paths that rely on RCU protection continue to use
hsr_for_each_port() with rcu_read_lock().

v3:
hold dev lock for hsr_get_port_ndev caller (Paolo Abeni)
deal with hsr_port_get_hsr() separtely (Paolo Abeni)
split the patch for easier review (Paolo Abeni)

v2:
check rtnl lock for net_device_ops (Kuniyuki Iwashima)
hold rcu read lock in timer function (Kuniyuki Iwashima)

Hangbin Liu (3):
  hsr: use rtnl lock when iterating over ports
  hsr: use hsr_for_each_port_rtnl in hsr_port_get_hsr
  hsr: hold rcu and dev lock for hsr_get_port_ndev

 drivers/net/ethernet/ti/icssg/icssg_prueth.c | 20 +++++++++-----
 net/hsr/hsr_device.c                         | 28 +++++++++++++-------
 net/hsr/hsr_main.c                           |  4 +--
 net/hsr/hsr_main.h                           |  3 +++
 4 files changed, 37 insertions(+), 18 deletions(-)

-- 
2.50.1