net: sfp: fix hwmon_name leak on hwmon registration failure
Subsystems (4)
PATCH 1/1 net: sfp: fix hwmon_name leak on hwmon registration failure
When hwmon_device_register_with_info() fails in sfp_hwmon_probe(), the string allocated by hwmon_sanitize_name() is not freed, leaking memory on repeated hotplug/probe attempts.
Also, sfp_hwmon_remove() only freed hwmon_name inside the IS_ERR_OR_NULL(hwmon_dev) block, so a partial-init failure could leave the name allocation behind.
Fix by freeing hwmon_name and clearing both pointers on registration failure, and by freeing hwmon_name unconditionally in remove while still only unregistering hwmon_dev when it is valid.
--- drivers/net/phy/sfp.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/net/phy/sfp.c +5 -1
@@ -1895,9 +1895,13 @@ static void sfp_hwmon_probe(struct work_struct *work) sfp->hwmon_name, sfp, &sfp_hwmon_chip_info, NULL); -if (IS_ERR (sfp->hwmon_dev)) +if (IS_ERR (sfp->hwmon_dev)) {dev_err (sfp->dev,"failed to register hwmon device: %ld\n" ,PTR_ERR (sfp->hwmon_dev)); +kfree (sfp->hwmon_name); + sfp->hwmon_name =NULL ; + sfp->hwmon_dev =NULL ; + } }static int sfp_hwmon_insert (struct sfp *sfp)
Signature · 1 line
2.34.1
Andrew Lunn
maintainerQuoted 3 lines · Krishan Singh
When
hwmon_device_register_with_info()fails insfp_hwmon_probe(), the string allocated byhwmon_sanitize_name()is not freed, leaking memory on repeated hotplug/probe attempts.
Have you actually seen this, or is this theoretical?
I'm not too familiar with this code, but what i find interesting is that sfp_hwmon_probe() is a void function. It has no way to say the probe has failed. So how does the caller know it failed, and it should not call sfp_hwmon_remove()? I suspect sfp_hwmon_remove() is always called. So sfp_hwmon_remove() should be changed to release the hwmon_name if it is not NULL.
Andrew
---
pw-bot: cr
I have looked into the code for sfp_hwmon_remove calls to the best of my abilities and Andrew is right that sfp_hwmon_remove is called on every sfp device.
If sfp_hwmon_probe fails in hwmon_device_register_with_info function call, it leaves sfp->hwmon_dev as an error pointer. When sfp_check_state detects that SFP_F_PRESENT has changed, it sends SFP_E_REMOVE which leads to sfp_hwmon_remove call. In sfp_hwmon_remove sfp->hwmon_name is not freed because !IS_ERR_OR_NULL(sfp->hwmon_dev) is false therefore the whole cleanup block is skipped.
So, as Andrew suggested, freeing sfp->hwmon_name if it is not NULL and it is not an error pointer (hwmon_sanitize_name can return an error pointer) instead of freeing only when sfp->hwmon_dev is valid seems to be the correct approach.