Skip to contentSkip to conversation

ath5k: Fix double free on hw attach error path

PATCH In review Jones Desougi 2 participants 2 messages Oct 27, 2010 – Oct 28, 2010

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ath5k: Fix double free on hw attach error path

If ath5k_hw_attach fails it will free sc->ah (local variable ah) before returning. However, when it reports failure the caller (ath5k_pci_probe) will also free sc->ah. Let the caller handle the deallocation, it does so on further errors as well.

---
The patch is generated from a week-old compat-wireless, but applies cleanly with -p1 to wireless-testing today.

diff -r -u -p compat-wireless-2010-10-19.orig/drivers/net/wireless/ath/ath5k/attach.c compat-wireless-2010-10-19/drivers/net/wireless/ath/ath5k/attach.c
---
compat-wireless-2010-10-19.orig/drivers/net/wireless/ath/ath5k/attach.c

2010-10-19 23:41:54.000000000 +0200
+++ compat-wireless-2010-10-19/drivers/net/wireless/ath/ath5k/attach.c
2010-10-22 14:28:29.000000000 +0200

@@ -139,12 +139,12 @@ int ath5k_hw_attach(struct ath5k_softc *
/* Fill the ath5k_hw struct with the needed functions */ ret = ath5k_hw_init_desc_functions(ah);
if (ret)
- goto err_free;
+ goto err;

/* Bring device out of sleep and reset its units */ ret = ath5k_hw_nic_wakeup(ah, 0, true);
if (ret)
- goto err_free;
+ goto err;

/* Get MAC, PHY and RADIO revisions */
ah->ah_mac_srev = srev;
@@ -234,7 +234,7 @@ int ath5k_hw_attach(struct ath5k_softc *
} else {
ATH5K_ERR(sc, "Couldn't identify radio revision.\n"); ret = -ENODEV;
- goto err_free;
+ goto err;
}
}

@@ -244,7 +244,7 @@ int ath5k_hw_attach(struct ath5k_softc *
(srev < AR5K_SREV_AR2425)) {
ATH5K_ERR(sc, "Device not yet supported.\n"); ret = -ENODEV;
- goto err_free;
+ goto err;
}

/*
@@ -252,7 +252,7 @@ int ath5k_hw_attach(struct ath5k_softc *
*/
ret = ath5k_hw_post(ah);
if (ret)
- goto err_free;
+ goto err;

/* Enable pci core retry fix on Hainan (5213A) and later chips */ if (srev >= AR5K_SREV_AR5213A)
@@ -265,7 +265,7 @@ int ath5k_hw_attach(struct ath5k_softc *
ret = ath5k_eeprom_init(ah);
if (ret) {
ATH5K_ERR(sc, "unable to init EEPROM\n");
- goto err_free;
+ goto err;
}

ee = &ah->ah_capabilities.cap_eeprom;
@@ -307,7 +307,7 @@ int ath5k_hw_attach(struct ath5k_softc *
if (ret) {
ATH5K_ERR(sc, "unable to get device capabilities: 0x%04x\n",
sc->pdev->device);
- goto err_free;
+ goto err;
}

/* Crypto settings */
@@ -341,8 +341,7 @@ int ath5k_hw_attach(struct ath5k_softc *
ath5k_hw_set_ledstate(ah, AR5K_LED_INIT);

return 0;
-err_free:
- kfree(ah);
+err:
return ret;
}

Reviewed-by on v1 1/1
#1 lore raw
Quoted 6 lines · Jones Desougi

If ath5k_hw_attach fails it will free sc->ah (local variable ah) before returning. However, when it reports failure the caller (ath5k_pci_probe) will also free sc->ah. Let the caller handle the deallocation, it does so on further errors as well.

Thanks, yes, looks like that was due to commit 9adca126d, "ath5k: allocate ath5k_hw prior to initializing hw".

This should probably go to stable.

Signature · 1 line
Bob Copeland %% www.bobcopeland.com