On 07/01/2025 13:01, Yeoreum Yun wrote:
Hi Suzuki,
Hi Levi
On 23/12/2024 18:53, Yeoreum Yun wrote:
While enable active config via cscfg_csdev_enable_active_config(), active config could be deactivated via configfs' sysfs interface. This could make UAF issue in below scenario:
CPU0 CPU1 (sysfs enable) load module cscfg_load_config_sets() activate config. // sysfs (sys_active_cnt == 1) ... cscfg_csdev_enable_active_config() lock(csdev->cscfg_csdev_lock) // here load config activate by CPU1 unlock(csdev->cscfg_csdev_lock)
deactivate config // sysfs (sys_activec_cnt == 0) cscfg_unload_config_sets() unload module // access to config_desc which freed // while unloading module. cfs_csdev_enable_configTo address this, introduce sys_enable_cnt in cscfg_mgr to prevent deactivate while there is enabled configuration.
Thanks for the finding the problem and the detailed description + patch. I have some concerns on the fix, please find it below.
So we have 3 atomic counters now ! cscfg_mgr->sys_active_cnt // Global count config->active_cnt // Per config count,
And another one which this one introduces.
cscfg_mgr->sys_enable_cnt // ?
And config->active_cnt is always ever 0 or 1. i.e., it is not really a reference counter at the moment but it indicates whether it is active or not. Could we not use that for tracking the references on the specific config ?
i.e., every time we "enable_active_config" atomic_inc(config->active_cnt)
and disable_active_config() always decrements the config. These could be wrapped in cscfg_get() cscfg_put() which would, inc/dec the refcounts and also drop the "module" reference when the active_cnt == 0
This action is done via _cscfg_activate_config() already but its activation is done via "sysfs". and the checking active_cnt, I think it would increase lots of complex.
I don't understand. We have this today :
cscfg_config_desc which is getting de-activated. This has reference to the module (owner). If someone is using this config_desc in running session (be it perf or sysfg), it must be refcounted. As such I don't see that the cscfg is refcounted anywhere. (The active_cnt indicates if this has been activated, but not used in a session). If we have a refcount on the "cscfg_config_desc" for each active session, we could prevent/delay unloading the module until the last one drops.
Something like this is what I was checking ?
coresight: cscfg: Hold refcount on the config for active sessions
Signed-off-by: Suzuki K Poulose suzuki.poulose@arm.com --- .../hwtracing/coresight/coresight-config.h | 2 +- .../hwtracing/coresight/coresight-syscfg.c | 24 +++++++++++++------ 2 files changed, 18 insertions(+), 8 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-config.h b/drivers/hwtracing/coresight/coresight-config.h index 6ba013975741..84cdde6f0e4d 100644 --- a/drivers/hwtracing/coresight/coresight-config.h +++ b/drivers/hwtracing/coresight/coresight-config.h @@ -228,7 +228,7 @@ struct cscfg_feature_csdev { * @feats_csdev:references to the device features to enable. */ struct cscfg_config_csdev { - const struct cscfg_config_desc *config_desc; + struct cscfg_config_desc *config_desc; struct coresight_device *csdev; bool enabled; struct list_head node; diff --git a/drivers/hwtracing/coresight/coresight-syscfg.c b/drivers/hwtracing/coresight/coresight-syscfg.c index 11138a9762b0..53baeaaf907f 100644 --- a/drivers/hwtracing/coresight/coresight-syscfg.c +++ b/drivers/hwtracing/coresight/coresight-syscfg.c @@ -914,15 +914,20 @@ static int _cscfg_activate_config(unsigned long cfg_hash) return err; }
+static void cscfg_config_desc_put(struct cscfg_config_desc *config_desc) +{ + if (!atomic_dec_return(&config_desc->active_cnt)) + cscfg_owner_put(config_desc->load_owner); +} + static void _cscfg_deactivate_config(unsigned long cfg_hash) { struct cscfg_config_desc *config_desc;
list_for_each_entry(config_desc, &cscfg_mgr->config_desc_list, item) { if ((unsigned long)config_desc->event_ea->var == cfg_hash) { - atomic_dec(&config_desc->active_cnt); atomic_dec(&cscfg_mgr->sys_active_cnt); - cscfg_owner_put(config_desc->load_owner); + cscfg_config_desc_put(config_desc); dev_dbg(cscfg_device(), "Deactivate config %s.\n", config_desc->name); break; } @@ -1047,7 +1052,7 @@ int cscfg_csdev_enable_active_config(struct coresight_device *csdev, unsigned long cfg_hash, int preset) { struct cscfg_config_csdev *config_csdev_active = NULL, *config_csdev_item; - const struct cscfg_config_desc *config_desc; + struct cscfg_config_desc *config_desc; unsigned long flags; int err = 0;
@@ -1062,8 +1067,8 @@ int cscfg_csdev_enable_active_config(struct coresight_device *csdev, spin_lock_irqsave(&csdev->cscfg_csdev_lock, flags); list_for_each_entry(config_csdev_item, &csdev->config_csdev_list, node) { config_desc = config_csdev_item->config_desc; - if ((atomic_read(&config_desc->active_cnt)) && - ((unsigned long)config_desc->event_ea->var == cfg_hash)) { + if ((unsigned long)config_desc->event_ea->var == cfg_hash && + atomic_inc_not_zero(&config_desc->active_cnt)) { config_csdev_active = config_csdev_item; csdev->active_cscfg_ctxt = (void *)config_csdev_active; break; @@ -1091,12 +1096,15 @@ int cscfg_csdev_enable_active_config(struct coresight_device *csdev, * Set enabled if OK, err if not. */ spin_lock_irqsave(&csdev->cscfg_csdev_lock, flags); - if (csdev->active_cscfg_ctxt) + if (csdev->active_cscfg_ctxt == config_csdev_active) config_csdev_active->enabled = true; else err = -EBUSY; spin_unlock_irqrestore(&csdev->cscfg_csdev_lock, flags); } + if (err) + cscfg_config_desc_put(config_desc); + } return err; } @@ -1136,8 +1144,10 @@ void cscfg_csdev_disable_active_config(struct coresight_device *csdev) spin_unlock_irqrestore(&csdev->cscfg_csdev_lock, flags);
/* true if there was an enabled active config */ - if (config_csdev) + if (config_csdev) { cscfg_csdev_disable_config(config_csdev); + cscfg_config_desc_put(config_csdev->config_desc); + } } EXPORT_SYMBOL_GPL(cscfg_csdev_disable_active_config);
-
Suzuki K Poulose