net: mscc: ocelot: fix error handling bugs in mscc_ocelot_init_ports()
PATCH 1/2 net: mscc: ocelot: fix error handling bugs in mscc_ocelot_init_ports()
There are several error handling bugs in mscc_ocelot_init_ports(). I went through the code, and carefully audited it and made fixes and cleanups.
1) The ocelot_probe_port() function didn't have a mirror release function so it was hard to follow. I created the ocelot_release_port() function.
2) In the ocelot_probe_port() function, if the register_netdev() call failed, then it lead to a double free_netdev(dev) bug. Fix this by moving the "ocelot->ports[port] = ocelot_port;" assignment to the end of the function after everything has succeeded.
3) I was concerned that the "port" which comes from of_property_read_u32() might be out of bounds so I added a check for that.
4) In the original code if ocelot_regmap_init() failed then the driver tried to continue but I think that should be a fatal error.
5) If ocelot_probe_port() failed then the most recent devlink was leaked. Fix this by moving the "registered_ports[port] = true;" assignment earlier.
6) The error handling if the final ocelot_port_devlink_init() failed had two problems. The "while (port-- >= 0)" loop should have been "--port" pre-op instead of a post-op to avoid a buffer underflow. The "if (!registered_ports[port])" condition was reversed leading to resource leaks and double frees.
--- drivers/net/ethernet/mscc/ocelot_vsc7514.c | 39 +++++++++------------- 1 file changed, 16 insertions(+), 23 deletions(-)
drivers/net/ethernet/mscc/ocelot_vsc7514.c +16 -23
@@ -1064,7 +1064,6 @@ static void mscc_ocelot_release_ports(struct ocelot *ocelot)int port;for (port =0 ; port < ocelot->num_phys_ports; port++) { -struct ocelot_port_private *priv;struct ocelot_port *ocelot_port; ocelot_port = ocelot->ports[port];
@@ -1072,12 +1071,7 @@ static void mscc_ocelot_release_ports(struct ocelot *ocelot)continue ;ocelot_deinit_port (ocelot, port); - - priv = container_of(ocelot_port, struct ocelot_port_private, - port); - -unregister_netdev (priv->dev); -free_netdev (priv->dev); +ocelot_release_port (ocelot_port); } }
@@ -1123,14 +1117,22 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,continue ; port = reg; +if (port <0 || port >= ocelot->num_phys_ports) { +dev_err (ocelot->dev, +"invalid port number: %d >= %d\n" , port, + ocelot->num_phys_ports); +continue ; + }snprintf (res_name,sizeof (res_name),"port%d" , port); res =platform_get_resource_byname (pdev,IORESOURCE_MEM , res_name); target =ocelot_regmap_init (ocelot, res); -if (IS_ERR (target)) -continue ; +if (IS_ERR (target)) { + err =PTR_ERR (target); + goto out_teardown; + } phy_node =of_parse_phandle (portnp,"phy-handle" ,0 );if (!phy_node)
@@ -1147,6 +1149,7 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,of_node_put (portnp); goto out_teardown; } + registered_ports[port] = true; err =ocelot_probe_port (ocelot, port, target, phy);if (err) {
@@ -1154,8 +1157,6 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev, goto out_teardown; } - registered_ports[port] = true; - ocelot_port = ocelot->ports[port]; priv = container_of(ocelot_port, struct ocelot_port_private, port);
@@ -1213,15 +1214,9 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev, err =ocelot_port_devlink_init (ocelot, port,DEVLINK_PORT_FLAVOUR_UNUSED ); -if (err) { -while (port-- >=0 ) { -if (!registered_ports[port]) -continue ; -ocelot_port_devlink_teardown (ocelot, port); - } - +if (err) goto out_teardown; - } + registered_ports[port] = true; }kfree (registered_ports);
@@ -1233,10 +1228,8 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,mscc_ocelot_release_ports (ocelot);/* Tear down devlink ports for the registered network interfaces */ for (port =0 ; port < ocelot->num_phys_ports; port++) { -if (!registered_ports[port]) -continue ; - -ocelot_port_devlink_teardown (ocelot, port); +if (registered_ports[port]) +ocelot_port_devlink_teardown (ocelot, port); }kfree (registered_ports);return err;
Signature · 1 line
2.29.2
PATCH 2/2 net: mscc: ocelot: fix error code in mscc_ocelot_probe()
Probe should return an error code if platform_get_irq_byname() fails but it returns success instead.
--- drivers/net/ethernet/mscc/ocelot_vsc7514.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/mscc/ocelot_vsc7514.c +3 -1
@@ -1307,8 +1307,10 @@ static int mscc_ocelot_probe(struct platform_device *pdev) goto out_free_devlink; irq_xtr =platform_get_irq_byname (pdev,"xtr" ); -if (irq_xtr <0 ) +if (irq_xtr <0 ) { + err = irq_xtr; goto out_free_devlink; + } err = devm_request_threaded_irq(&pdev->dev, irq_xtr, NULL, ocelot_xtr_irq_handler, IRQF_ONESHOT,
Signature · 1 line
2.29.2
Oops. Please ignore this patch. Part of it is missing.
regards,
dan carpenter
PATCH v2 1/2 net: mscc: ocelot: fix error handling bugs in mscc_ocelot_init_ports()
There are several error handling bugs in mscc_ocelot_init_ports(). I went through the code, and carefully audited it and made fixes and cleanups.
1) The ocelot_probe_port() function didn't have a mirror release function so it was hard to follow. I created the ocelot_release_port() function.
2) In the ocelot_probe_port() function, if the register_netdev() call failed, then it lead to a double free_netdev(dev) bug. Fix this by moving the "ocelot->ports[port] = ocelot_port;" assignment to the end of the function after everything has succeeded.
3) I was concerned that the "port" which comes from of_property_read_u32() might be out of bounds so I added a check for that.
4) In the original code if ocelot_regmap_init() failed then the driver tried to continue but I think that should be a fatal error.
5) If ocelot_probe_port() failed then the most recent devlink was leaked. Fix this by moving the "registered_ports[port] = true;" assignment earlier.
6) The error handling if the final ocelot_port_devlink_init() failed had two problems. The "while (port-- >= 0)" loop should have been "--port" pre-op instead of a post-op to avoid a buffer underflow. The "if (!registered_ports[port])" condition was reversed leading to resource leaks and double frees.
---
v2: In v1, I accidentally left out parts of the commit so it didn't compile.
drivers/net/ethernet/mscc/ocelot.h | 1 + drivers/net/ethernet/mscc/ocelot_net.c | 15 +++++++-- drivers/net/ethernet/mscc/ocelot_vsc7514.c | 39 +++++++++------------- 3 files changed, 30 insertions(+), 25 deletions(-)
drivers/net/ethernet/mscc/ocelot.h +1 -0
@@ -121,6 +121,7 @@ void ocelot_port_writel(struct ocelot_port *port, u32 val, u32 reg);int ocelot_probe_port (struct ocelot *ocelot,int port,struct regmap *target,struct phy_device *phy); +void ocelot_release_port (struct ocelot_port *ocelot_port);int ocelot_devlink_init (struct ocelot *ocelot);void ocelot_devlink_teardown (struct ocelot *ocelot); int ocelot_port_devlink_init(struct ocelot *ocelot, int port,
drivers/net/ethernet/mscc/ocelot_net.c +13 -2
@@ -1262,7 +1262,6 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, ocelot_port = &priv->port; ocelot_port->ocelot = ocelot; ocelot_port->target = target; - ocelot->ports[port] = ocelot_port; dev->netdev_ops = &ocelot_port_netdev_ops; dev->ethtool_ops = &ocelot_ethtool_ops;
@@ -1282,7 +1281,19 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target,if (err) {dev_err (ocelot->dev,"register_netdev failed\n" );free_netdev (dev); +return err; } -return err; + ocelot->ports[port] = ocelot_port; +return 0 ; +} + +void ocelot_release_port (struct ocelot_port *ocelot_port) +{ +struct ocelot_port_private *priv =container_of (ocelot_port, + struct ocelot_port_private, + port); + +unregister_netdev (priv->dev); +free_netdev (priv->dev); }
drivers/net/ethernet/mscc/ocelot_vsc7514.c +16 -23
@@ -1064,7 +1064,6 @@ static void mscc_ocelot_release_ports(struct ocelot *ocelot)int port;for (port =0 ; port < ocelot->num_phys_ports; port++) { -struct ocelot_port_private *priv;struct ocelot_port *ocelot_port; ocelot_port = ocelot->ports[port];
@@ -1072,12 +1071,7 @@ static void mscc_ocelot_release_ports(struct ocelot *ocelot)continue ;ocelot_deinit_port (ocelot, port); - - priv = container_of(ocelot_port, struct ocelot_port_private, - port); - -unregister_netdev (priv->dev); -free_netdev (priv->dev); +ocelot_release_port (ocelot_port); } }
@@ -1123,14 +1117,22 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,continue ; port = reg; +if (port <0 || port >= ocelot->num_phys_ports) { +dev_err (ocelot->dev, +"invalid port number: %d >= %d\n" , port, + ocelot->num_phys_ports); +continue ; + }snprintf (res_name,sizeof (res_name),"port%d" , port); res =platform_get_resource_byname (pdev,IORESOURCE_MEM , res_name); target =ocelot_regmap_init (ocelot, res); -if (IS_ERR (target)) -continue ; +if (IS_ERR (target)) { + err =PTR_ERR (target); + goto out_teardown; + } phy_node =of_parse_phandle (portnp,"phy-handle" ,0 );if (!phy_node)
@@ -1147,6 +1149,7 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,of_node_put (portnp); goto out_teardown; } + registered_ports[port] = true; err =ocelot_probe_port (ocelot, port, target, phy);if (err) {
@@ -1154,8 +1157,6 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev, goto out_teardown; } - registered_ports[port] = true; - ocelot_port = ocelot->ports[port]; priv = container_of(ocelot_port, struct ocelot_port_private, port);
@@ -1213,15 +1214,9 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev, err =ocelot_port_devlink_init (ocelot, port,DEVLINK_PORT_FLAVOUR_UNUSED ); -if (err) { -while (port-- >=0 ) { -if (!registered_ports[port]) -continue ; -ocelot_port_devlink_teardown (ocelot, port); - } - +if (err) goto out_teardown; - } + registered_ports[port] = true; }kfree (registered_ports);
@@ -1233,10 +1228,8 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,mscc_ocelot_release_ports (ocelot);/* Tear down devlink ports for the registered network interfaces */ for (port =0 ; port < ocelot->num_phys_ports; port++) { -if (!registered_ports[port]) -continue ; - -ocelot_port_devlink_teardown (ocelot, port); +if (registered_ports[port]) +ocelot_port_devlink_teardown (ocelot, port); }kfree (registered_ports);return err;
Signature · 1 line
2.29.2
PATCH v2 2/2 net: mscc: ocelot: fix error code in mscc_ocelot_probe()
In reply to Dan CarpenterProbe should return an error code if platform_get_irq_byname() fails but it returns success instead.
---
v2: no change
drivers/net/ethernet/mscc/ocelot_vsc7514.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/mscc/ocelot_vsc7514.c +3 -1
@@ -1307,8 +1307,10 @@ static int mscc_ocelot_probe(struct platform_device *pdev) goto out_free_devlink; irq_xtr =platform_get_irq_byname (pdev,"xtr" ); -if (irq_xtr <0 ) +if (irq_xtr <0 ) { + err = irq_xtr; goto out_free_devlink; + } err = devm_request_threaded_irq(&pdev->dev, irq_xtr, NULL, ocelot_xtr_irq_handler, IRQF_ONESHOT,
Signature · 1 line
2.29.2
Re: [PATCH 2/2 net-next] net: mscc: ocelot: fix error code in mscc_ocelot_probe()
In reply to Dan CarpenterQuoted 6 lines · Dan Carpenter
Probe should return an error code if
platform_get_irq_byname()fails but it returns success instead.Fixes 6c30384eb1de ("net: mscc: ocelot: register devlink ports")Signed-off-by Dan Carpenter <[email protected]>---
Hi Dan,
drivers/net/ethernet/mscc/ocelot_net.c int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, · line 1265
Quoted 9 lines · Dan Carpenter
drivers/net/ethernet/mscc/ocelot_net.c +0
-1@@ -1262,7 +1262,6 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, ocelot_port = &priv->port; ocelot_port->ocelot = ocelot; ocelot_port->target = target; - ocelot->ports[port] = ocelot_port;
You cannot remove this from here just like that, because ocelot_init_port right below accesses ocelot->ports[port], and it will dereference through a NULL pointer otherwise.
Quoted 9 lines
drivers/net/ethernet/mscc/ocelot_net.c +1
-0dev->netdev_ops = &ocelot_port_netdev_ops; dev->ethtool_ops = &ocelot_ethtool_ops;@@ -1282,7 +1281,19 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target,if (err) {dev_err (ocelot->dev,"register_netdev failed\n" );free_netdev (dev); +return err; }
drivers/net/ethernet/mscc/ocelot_net.c int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, · line 1295
Quoted 10 lines
drivers/net/ethernet/mscc/ocelot_net.c +9
-1-return err; + ocelot->ports[port] = ocelot_port; +return 0 ; +} + +void ocelot_release_port (struct ocelot_port *ocelot_port) +{ +struct ocelot_port_private *priv =container_of (ocelot_port, + struct ocelot_port_private, + port);
Can this assignment please be done separately from the declaration?
struct ocelot_port_private *priv; priv = container_of(ocelot_port, struct ocelot_port_private, port);
drivers/net/ethernet/mscc/ocelot_net.c int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, · line 1299
Quoted 4 lines
drivers/net/ethernet/mscc/ocelot_net.c +3
-0+ +unregister_netdev (priv->dev); +free_netdev (priv->dev); }
Fun, isn't it? :D
Thanks for taking the time to untangle this.
Additionally, you have changed the meaning of "registered_ports" from "this port had its net_device registered" to "this port had its devlink_port registered". This is ok, but I would like the variable renamed now, too. I think devlink_ports_registered would be ok.
In hindsight, I was foolish for using a heap-allocated boolean array for registered_ports, because this switch architecture is guaranteed to not have more than 32 ports, so a u32 bitmask is fine.
If you resend, can you please squash this diff on top of your patch?
-----------------------------[cut here]-----------------------------
drivers/net/ethernet/mscc/ocelot_net.c +2 -1
@@ -1196,6 +1196,7 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, ocelot_port = &priv->port; ocelot_port->ocelot = ocelot; ocelot_port->target = target; + ocelot->ports[port] = ocelot_port; dev->netdev_ops = &ocelot_port_netdev_ops; dev->ethtool_ops = &ocelot_ethtool_ops;
@@ -1215,10 +1216,10 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target,if (err) {dev_err (ocelot->dev,"register_netdev failed\n" );free_netdev (dev); + ocelot->ports[port] =NULL ;return err; } - ocelot->ports[port] = ocelot_port;return 0 ; }
drivers/net/ethernet/mscc/ocelot_vsc7514.c +6 -13
@@ -939,8 +939,8 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev, struct device_node *ports) {struct ocelot *ocelot =platform_get_drvdata (pdev); +u32 devlink_ports_registered =0 ;struct device_node *portnp; -bool *registered_ports;int port, err;u32 reg;
@@ -956,11 +956,6 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,if (!ocelot->devlink_ports)return -ENOMEM ; - registered_ports =kcalloc (ocelot->num_phys_ports,sizeof (bool ), -GFP_KERNEL ); -if (!registered_ports) -return -ENOMEM ; -for_each_available_child_of_node (ports, portnp) {struct ocelot_port_private *priv;struct ocelot_port *ocelot_port;
@@ -1009,7 +1004,7 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,of_node_put (portnp); goto out_teardown; } - registered_ports[port] = true; + devlink_ports_registered |=BIT (port); err =ocelot_probe_port (ocelot, port, target, phy);if (err) {
@@ -1069,17 +1064,16 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,/* Initialize unused devlink ports at the end */ for (port =0 ; port < ocelot->num_phys_ports; port++) { -if (registered_ports[port]) +if (devlink_ports_registered &BIT (port))continue ; err =ocelot_port_devlink_init (ocelot, port,DEVLINK_PORT_FLAVOUR_UNUSED );if (err) goto out_teardown; - registered_ports[port] = true; - } -kfree (registered_ports); + devlink_ports_registered |=BIT (port); + }return 0 ;
@@ -1088,10 +1082,9 @@ static int mscc_ocelot_init_ports(struct platform_device *pdev,mscc_ocelot_release_ports (ocelot);/* Tear down devlink ports for the registered network interfaces */ for (port =0 ; port < ocelot->num_phys_ports; port++) { -if (registered_ports[port]) +if (devlink_ports_registered &BIT (port))ocelot_port_devlink_teardown (ocelot, port); } -kfree (registered_ports);return err; }
-----------------------------[cut here]-----------------------------
Then you can add:
Also, it's strange but I don't see the v2 patches in patchwork. Did you send them in-reply-to v1 or something?
Thanks!
Quoted 17 lines · Vladimir Oltean
Hi Dan,
drivers/net/ethernet/mscc/ocelot_net.c int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, · line 1265
Quoted 9 lines · Dan Carpenter
drivers/net/ethernet/mscc/ocelot_net.c +0
-1@@ -1262,7 +1262,6 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, ocelot_port = &priv->port; ocelot_port->ocelot = ocelot; ocelot_port->target = target; - ocelot->ports[port] = ocelot_port;You cannot remove this from here just like that, because
ocelot_init_portright below accesses ocelot->ports[port], and it will dereference through a NULL pointer otherwise.
Argh... Thanks for spotting that.
Quoted 39 lines
Quoted 9 lines
drivers/net/ethernet/mscc/ocelot_net.c +1
-0dev->netdev_ops = &ocelot_port_netdev_ops; dev->ethtool_ops = &ocelot_ethtool_ops;@@ -1282,7 +1281,19 @@ int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target,if (err) {dev_err (ocelot->dev,"register_netdev failed\n" );free_netdev (dev); +return err; }drivers/net/ethernet/mscc/ocelot_net.c int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, · line 1295
Quoted 10 lines
drivers/net/ethernet/mscc/ocelot_net.c +9
-1-return err; + ocelot->ports[port] = ocelot_port; +return 0 ; +} + +void ocelot_release_port (struct ocelot_port *ocelot_port) +{ +struct ocelot_port_private *priv =container_of (ocelot_port, + struct ocelot_port_private, + port);Can this assignment please be done separately from the declaration?
struct ocelot_port_private *priv; priv = container_of(ocelot_port, struct ocelot_port_private, port);drivers/net/ethernet/mscc/ocelot_net.c int ocelot_probe_port(struct ocelot *ocelot, int port, struct regmap *target, · line 1299
Quoted 4 lines
drivers/net/ethernet/mscc/ocelot_net.c +3
-0+ +unregister_netdev (priv->dev); +free_netdev (priv->dev); }Fun, isn't it? :D
Thanks for taking the time to untangle this.Additionally, you have changed the meaning of "
registered_ports" from "this port had itsnet_deviceregistered" to "this port had itsdevlink_portregistered". This is ok, but I would like the variable renamed now, too. I thinkdevlink_ports_registeredwould be ok.
Quoted 5 lines
In hindsight, I was foolish for using a heap-allocated boolean array for
registered_ports, because this switch architecture is guaranteed to not have more than 32 ports, so a u32 bitmask is fine.If you resend, can you please squash this diff on top of your patch?
Yep. I will resend. Thanks for basically writing v2 for me. Your review comments were very clear but code is always 100% clear so that's really great. I've never seen anyone do that before. I should copy that for my own reviews and hopefully it's a new trend.
Quoted 8 lines
Then you can add:
Reviewed-by Vladimir Oltean <[email protected]>Tested-by Vladimir Oltean <[email protected]>Also, it's strange but I don't see the v2 patches in patchwork. Did you send them in-reply-to v1 or something?
I did send them as a reply to v1. Patchwork doesn't like that?
regards,
dan carpenter
Vladimir Oltean
maintainerQuoted 5 lines · Dan Carpenter
Quoted 1 line
If you resend, can you please squash this diff on top of your patch?
Yep. I will resend. Thanks for basically writing v2 for me. Your review comments were very clear but code is always 100% clear so that's really great.
Until the code is wrong :)
Quoted 2 lines
I've never seen anyone do that before. I should copy that for my own reviews and hopefully it's a new trend.
I took that from Florian, it was helpful to me when I wasn't minding my own business and touching code I couldn't test:
https://patchwork.kernel.org/project/netdevbpf/patch/[email protected]/#23858835
Quoted 4 lines
Quoted 2 lines
Also, it's strange but I don't see the v2 patches in patchwork. Did you send them in-reply-to v1 or something?
I did send them as a reply to v1. Patchwork doesn't like that?
Nope.