| Message ID | abr3xq5eankrmzvyhjd5za6itfm5s7wpqwfy7lp3iuwsv33oi3@dx5eg6wmb2so (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: page_pool: check page pool ethtool stats | expand |
On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > If the page_pool variable is null while passing it to > the page_pool_get_stats function we receive a kernel error. > > Check if the page_pool variable is at least valid. IMHO this seems insufficient, the driver still has to check if PP was instantiated when the strings are queried. My weak preference would be to stick to v1 and have the driver check all the conditions. But if nobody else feels this way, it's fine :)
On Mon, Oct 02, 2023 at 12:46:50PM -0700, Jakub Kicinski wrote: > On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > > If the page_pool variable is null while passing it to > > the page_pool_get_stats function we receive a kernel error. > > > > Check if the page_pool variable is at least valid. > > IMHO this seems insufficient, the driver still has to check if PP > was instantiated when the strings are queried. My weak preference > would be to stick to v1 and have the driver check all the conditions. > But if nobody else feels this way, it's fine :) :) let me know which variant is preferred so I can send a new v3.
Hi Jakub, On Mon, Oct 02, 2023 at 12:46:50PM -0700, Jakub Kicinski wrote: > On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > > If the page_pool variable is null while passing it to > > the page_pool_get_stats function we receive a kernel error. > > > > Check if the page_pool variable is at least valid. > > IMHO this seems insufficient, the driver still has to check if PP > was instantiated when the strings are queried. My weak preference > would be to stick to v1 and have the driver check all the conditions. > But if nobody else feels this way, it's fine :) I don't disagree, but OTOH it would be sane for the API not to crash if something invalid is passed. Maybe the best approach would be to keep the driver checks, which are saner, but add the page pool code as well with a printable error indicating a driver bug? Thanks /Ilias
On Wed, Oct 25, 2023 at 08:59:44AM +0300, Ilias Apalodimas wrote: > Hi Jakub, > > On Mon, Oct 02, 2023 at 12:46:50PM -0700, Jakub Kicinski wrote: > > On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > > > If the page_pool variable is null while passing it to > > > the page_pool_get_stats function we receive a kernel error. > > > > > > Check if the page_pool variable is at least valid. > > > > IMHO this seems insufficient, the driver still has to check if PP > > was instantiated when the strings are queried. My weak preference > > would be to stick to v1 and have the driver check all the conditions. > > But if nobody else feels this way, it's fine :) > > I don't disagree, but OTOH it would be sane for the API not to crash if > something invalid is passed. In-kernel API assumes that in-kernel callers know how to use it. > Maybe the best approach would be to keep the > driver checks, which are saner, but add the page pool code as well with a > printable error indicating a driver bug? It is no different from adding extra checks to prevent drivers from random calls with random parameters to well defined API. Thanks > > Thanks > /Ilias >
On Wed, Oct 25, 2023 at 10:53:41AM +0300, Leon Romanovsky wrote: > On Wed, Oct 25, 2023 at 08:59:44AM +0300, Ilias Apalodimas wrote: > > Hi Jakub, > > > > On Mon, Oct 02, 2023 at 12:46:50PM -0700, Jakub Kicinski wrote: > > > On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > > > > If the page_pool variable is null while passing it to > > > > the page_pool_get_stats function we receive a kernel error. > > > > > > > > Check if the page_pool variable is at least valid. > > > > > > IMHO this seems insufficient, the driver still has to check if PP > > > was instantiated when the strings are queried. My weak preference > > > would be to stick to v1 and have the driver check all the conditions. > > > But if nobody else feels this way, it's fine :) > > > > I don't disagree, but OTOH it would be sane for the API not to crash if > > something invalid is passed. > > In-kernel API assumes that in-kernel callers know how to use it. > > > Maybe the best approach would be to keep the > > driver checks, which are saner, but add the page pool code as well with a > > printable error indicating a driver bug? > > It is no different from adding extra checks to prevent drivers from random calls > with random parameters to well defined API. > > Thanks I can see the point for both arguments so I think we should definitely keep the driver check. Is there a consensus on what to do on the page pool side? Do you want me to keep the additional page pool check to prevent a kernel crash? I mean the mvneta change was also implemented with this problem and it leads to serious side effects without an additional check. Especially if the page pool ethtool stats is implemented in more drivers in the future and the implementations are not 100% correct, it will crash the kernel. Best Sven > > > > > Thanks > > /Ilias > >
On Mon, Oct 30, 2023 at 10:39:18AM +0100, Sven Auhagen wrote: > On Wed, Oct 25, 2023 at 10:53:41AM +0300, Leon Romanovsky wrote: > > On Wed, Oct 25, 2023 at 08:59:44AM +0300, Ilias Apalodimas wrote: > > > Hi Jakub, > > > > > > On Mon, Oct 02, 2023 at 12:46:50PM -0700, Jakub Kicinski wrote: > > > > On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > > > > > If the page_pool variable is null while passing it to > > > > > the page_pool_get_stats function we receive a kernel error. > > > > > > > > > > Check if the page_pool variable is at least valid. > > > > > > > > IMHO this seems insufficient, the driver still has to check if PP > > > > was instantiated when the strings are queried. My weak preference > > > > would be to stick to v1 and have the driver check all the conditions. > > > > But if nobody else feels this way, it's fine :) > > > > > > I don't disagree, but OTOH it would be sane for the API not to crash if > > > something invalid is passed. > > > > In-kernel API assumes that in-kernel callers know how to use it. > > > > > Maybe the best approach would be to keep the > > > driver checks, which are saner, but add the page pool code as well with a > > > printable error indicating a driver bug? > > > > It is no different from adding extra checks to prevent drivers from random calls > > with random parameters to well defined API. > > > > Thanks > > I can see the point for both arguments so I think we should definitely > keep the driver check. > > Is there a consensus on what to do on the page pool side? > Do you want me to keep the additional page pool check to prevent > a kernel crash? I don't want to see bloat of checks in kernel API. They hide issues and not prevent them. > I mean the mvneta change was also implemented with this problem > and it leads to serious side effects without an additional check. > Especially if the page pool ethtool stats is implemented in more > drivers in the future and the implementations are not 100% correct, > it will crash the kernel. Like many other driver mistakes. Thanks > > Best > Sven > > > > > > > > > Thanks > > > /Ilias > > >
On Mon, Oct 30, 2023 at 04:23:55PM +0200, Leon Romanovsky wrote: > On Mon, Oct 30, 2023 at 10:39:18AM +0100, Sven Auhagen wrote: > > On Wed, Oct 25, 2023 at 10:53:41AM +0300, Leon Romanovsky wrote: > > > On Wed, Oct 25, 2023 at 08:59:44AM +0300, Ilias Apalodimas wrote: > > > > Hi Jakub, > > > > > > > > On Mon, Oct 02, 2023 at 12:46:50PM -0700, Jakub Kicinski wrote: > > > > > On Sun, 1 Oct 2023 13:41:15 +0200 Sven Auhagen wrote: > > > > > > If the page_pool variable is null while passing it to > > > > > > the page_pool_get_stats function we receive a kernel error. > > > > > > > > > > > > Check if the page_pool variable is at least valid. > > > > > > > > > > IMHO this seems insufficient, the driver still has to check if PP > > > > > was instantiated when the strings are queried. My weak preference > > > > > would be to stick to v1 and have the driver check all the conditions. > > > > > But if nobody else feels this way, it's fine :) > > > > > > > > I don't disagree, but OTOH it would be sane for the API not to crash if > > > > something invalid is passed. > > > > > > In-kernel API assumes that in-kernel callers know how to use it. > > > > > > > Maybe the best approach would be to keep the > > > > driver checks, which are saner, but add the page pool code as well with a > > > > printable error indicating a driver bug? > > > > > > It is no different from adding extra checks to prevent drivers from random calls > > > with random parameters to well defined API. > > > > > > Thanks > > > > I can see the point for both arguments so I think we should definitely > > keep the driver check. > > > > Is there a consensus on what to do on the page pool side? > > Do you want me to keep the additional page pool check to prevent > > a kernel crash? > > I don't want to see bloat of checks in kernel API. They hide issues and > not prevent them. > > > I mean the mvneta change was also implemented with this problem > > and it leads to serious side effects without an additional check. > > Especially if the page pool ethtool stats is implemented in more > > drivers in the future and the implementations are not 100% correct, > > it will crash the kernel. > > Like many other driver mistakes. > > Thanks Understood, thanks. I will resubmit with the initial patch that has only the driver fix in it. Best Sven > > > > > Best > > Sven > > > > > > > > > > > > > Thanks > > > > /Ilias > > > >
diff --git a/net/core/page_pool.c b/net/core/page_pool.c index 2396c99bedea..4c5dca6b4a16 100644 --- a/net/core/page_pool.c +++ b/net/core/page_pool.c @@ -65,6 +65,9 @@ bool page_pool_get_stats(struct page_pool *pool, if (!stats) return false; + if (!pool) + return false; + /* The caller is responsible to initialize stats. */ stats->alloc_stats.fast += pool->alloc_stats.fast; stats->alloc_stats.slow += pool->alloc_stats.slow;
If the page_pool variable is null while passing it to the page_pool_get_stats function we receive a kernel error. Check if the page_pool variable is at least valid. Signed-off-by: Sven Auhagen <sven.auhagen@voleatech.de> Reported-by: Paulo Da Silva <Paulo.DaSilva@kyberna.com>