Re: [PATCH] iwlwifi: mvm: no need to check return value of debugfs_create functions

From: Luca Coelho
Date: Wed Mar 27 2019 - 02:35:52 EST


On Wed, 2019-03-27 at 10:53 +0900, Greg Kroah-Hartman wrote:
> On Tue, Mar 26, 2019 at 06:47:33PM -0700, Laura Abbott wrote:
> > On 3/26/19 6:26 PM, Greg Kroah-Hartman wrote:
> > > On Tue, Mar 26, 2019 at 04:55:54PM -0700, Laura Abbott wrote:
> > > > On 1/22/19 7:21 AM, Greg Kroah-Hartman wrote:
> > > > > diff --git a/drivers/net/wireless/intel/iwlwifi/mvm/debugfs-vif.c b/drivers/net/wireless/intel/iwlwifi/mvm/debugfs-vif.c
> > > > > index 33b0af24a537..c52cdc538678 100644
> > > > > --- a/drivers/net/wireless/intel/iwlwifi/mvm/debugfs-vif.c
> > > > > +++ b/drivers/net/wireless/intel/iwlwifi/mvm/debugfs-vif.c
> > > > > @@ -1446,9 +1446,8 @@ static ssize_t iwl_dbgfs_quota_min_read(struct file *file,
> > > > > #define MVM_DEBUGFS_READ_WRITE_FILE_OPS(name, bufsz) \
> > > > > _MVM_DEBUGFS_READ_WRITE_FILE_OPS(name, bufsz, struct ieee80211_vif)
> > > > > #define MVM_DEBUGFS_ADD_FILE_VIF(name, parent, mode) do { \
> > > > > - if (!debugfs_create_file(#name, mode, parent, vif, \
> > > > > - &iwl_dbgfs_##name##_ops)) \
> > > > > - goto err; \
> > > > > + debugfs_create_file(#name, mode, parent, vif, \
> > > > > + &iwl_dbgfs_##name##_ops); \
> > > > > } while (0)
> > > > > MVM_DEBUGFS_READ_FILE_OPS(mac_params);
> > > > > @@ -1483,12 +1482,6 @@ void iwl_mvm_vif_dbgfs_register(struct iwl_mvm *mvm, struct ieee80211_vif *vif)
> > > > > mvmvif->dbgfs_dir = debugfs_create_dir("iwlmvm", dbgfs_dir);
> > > > > - if (!mvmvif->dbgfs_dir) {
> > > > > - IWL_ERR(mvm, "Failed to create debugfs directory under %pd\n",
> > > > > - dbgfs_dir);
> > > > > - return;
> > > > > - }
> > > > > -
> > > > > if (iwlmvm_mod_params.power_scheme != IWL_POWER_SCHEME_CAM &&
> > > > > ((vif->type == NL80211_IFTYPE_STATION && !vif->p2p) ||
> > > > > (vif->type == NL80211_IFTYPE_STATION && vif->p2p)))
> > > > > @@ -1537,12 +1530,6 @@ void iwl_mvm_vif_dbgfs_register(struct iwl_mvm *mvm, struct ieee80211_vif *vif)
> > > > > mvmvif->dbgfs_slink = debugfs_create_symlink(dbgfs_dir->d_name.name,
> > > > > mvm->debugfs_dir, buf);
> > > > > - if (!mvmvif->dbgfs_slink)
> > > > > - IWL_ERR(mvm, "Can't create debugfs symbolic link under %pd\n",
> > > > > - dbgfs_dir);
> > > > > - return;
> > > > > -err:
> > > > > - IWL_ERR(mvm, "Can't create debugfs entity\n");
> > > > > }
> > > >
> > > > Fedora got a bug report https://bugzilla.redhat.com/show_bug.cgi?id=1691034
> > > > of a crash with 5.0 and the user did a bisect which pointed to ff9fb72bc077
> > > > ("debugfs: return error values, not NULL") because the error checking is
> > > > no longer correct in this driver.
> > > >
> > > > Based on https://patchwork.kernel.org/patch/10865839/, it looks like
> > > > this is supposed to go in for 5.2 but this needs to go in now as
> > > > the error checking is currently broken without it. Can this get queued
> > > > for Linus so we can get it in 5.0 stable?
> > >
> > > That's odd, I can't see how the error checking is wrong here. If the
> > > directory is not created, an error will be returned, which should be
> > > able to be handled by debugfs_create_file().
> > >
> > > So with this patch does the error go away?
> > >
> >
> > The full patch didn't apply cleanly and I didn't try to backport it
> > for the reporter to test. I was going off of the theory that if the
> > patch was there it would fix the problem.
> >
> > What I _think_ is going wrong is dbgfs_dir is actually an errno value:
> >
> >
> > struct dentry *dbgfs_dir = vif->debugfs_dir;
> > struct iwl_mvm_vif *mvmvif = iwl_mvm_vif_from_mac80211(vif);
> > char buf[100];
> >
> > /*
> > * Check if debugfs directory already exist before creating it.
> > * This may happen when, for example, resetting hw or suspend-resume
> > */
> > if (!dbgfs_dir || mvmvif->dbgfs_dir)
> > return;
> >
> >
> > so this blows up in the snprintf
> >
> > snprintf(buf, 100, "../../../%pd3/%pd",
> > dbgfs_dir,
> > mvmvif->dbgfs_dir);
>
> Ah, yeah, that's horrible. They had the name before, why pull it out of
> the dentry again? That will blow up hard, but maybe printk should check
> to see if the pointer really is a pointer first.

I agree this is ugly. But do you mean we could use
("../../../%pd3/%s", dbgfs_dir, "iwlmvm")? Or how did we have the name?

Also, this would solve the sprintf() problem, but still wouldn't solve
the real issue, which is not check for ERR in dbgfs_dir.


> > Unless I misunderstood what the debugfs error change did. I think this
> > also means the if check needs to look for IS_ERR and not just !dbgfs_dir.
>
> Yes, that is correct.

Yeah, we can do that. So this patch doesn't need to be sent for
v5.1-rc* and v5.0, right? At least I don't see how it would fix the
issue. What we need is a new patch with the IS_ERR check.

--
Cheers,
Luca.