ethtool: do_sset return correct value on fail
Subsystems (2)
PATCH 1/1 ethtool: do_sset return correct value on fail
From: Roy Novich <[email protected]>
The return value for do_sset was constant and returned 0. This value is misleading when returned on operation failure. Changed return value to the correct function err status.
--- ethtool.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
ethtool.c +1 -1
@@ -3287,7 +3287,7 @@ static int do_sset(struct cmd_context *ctx) } } -return 0 ; +return err; }static int do_gregs (struct cmd_context *ctx)
Signature · 1 line
2.21.0
Quoted 18 lines · Tariq Toukan
From: Roy Novich <[email protected]>
The return value for
do_ssetwas constant and returned 0. This value is misleading when returned on operation failure. Changed return value to the correct function err status.Fixes 32c8037055f5 ("Initial import of ethtool version 3 + a few patches.")Signed-off-by Roy Novich <[email protected]>Signed-off-by Tariq Toukan <[email protected]>--- ethtool.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)ethtool.c +0
-0
ethtool.c static int do_sset(struct cmd_context *ctx) · line 3293
Quoted 8 lines
ethtool.c +1
-1@@ -3287,7 +3287,7 @@ static int do_sset(struct cmd_context *ctx) } } -return 0 ; +return err; }static int do_gregs (struct cmd_context *ctx)
I'm afraid it's not as easy as this. The problem with -s/--set subcommand is that its parameters are in fact implemented by three separate requests (for ioctl, four requests for netlink). Each of them may fail or succeed independently of others. Currently do_sset() always returns 0 which is indeed unfortunate but it's not clear what should we return if some requests fail and some succeed.
With your patch, do_sset() would always return the result of last request it sent to kernel which is inconsistent; consider e.g.
ethtool -s eth0 mdix on wol g
If the ETHTOOL_SLINKSETTINGS request setting MDI-X succeeds and ETHTOOL_SWOL setting WoL fails, the result would be a failure. But if setting MDI-X fails and setting WoL succeeds, do_sset() would report success.
So if we really want to change the behaviour after so many years, it should be consistent: either return non-zero exit code if any request fails or if all fail. (Personally, I would prefer the latter.) And we should probably modify nl_sset() to behave the same way as it currently bails out on first failed request.
Michal