(Including the pstore folks and quoting in full)
On Sun, 2013-01-27 at 06:52 +0800, Jeremy Kerr wrote:
> We can cause an oops by unlinking an open efivarfs file, then reading
> (or writing) from the deleted file. The unlink operation causes the
> efivarfs_entry struct to be freed, which is referenced again in the
> read.
>
> This change allow the efivar_entry structures to remain present after
> the variable has been removed, but flagged as not present in firmware by
> way of having an empty ->list member. We keep a refcount to determine
> when the entry can actually be freed.
>
> Signed-off-by: Jeremy Kerr <[email protected]>
>
> ---
> drivers/firmware/efivars.c | 105 +++++++++++++++++++++++++++++++------
> 1 file changed, 89 insertions(+), 16 deletions(-)
>
> diff --git a/drivers/firmware/efivars.c b/drivers/firmware/efivars.c
> index 053f28b..764f548 100644
> --- a/drivers/firmware/efivars.c
> +++ b/drivers/firmware/efivars.c
> @@ -121,6 +121,15 @@ struct efi_variable {
> struct efivar_entry {
> struct efivars *efivars;
> struct efi_variable var;
> +
> + /* references to the kref are held by:
> + * - open efivarfs files
> + * - efivarfs inodes
> + * - the kobj
> + * - presence in the global efivar list
> + */
> + struct kref kref;
> +
> struct list_head list;
> struct kobject kobj;
> };
> @@ -147,7 +156,8 @@ struct efivar_attribute efivar_attr_##_name = { \
> };
>
> #define to_efivar_attr(_attr) container_of(_attr, struct efivar_attribute,
> attr)
> -#define to_efivar_entry(obj) container_of(obj, struct efivar_entry, kobj)
> +#define kobj_to_efivar_entry(obj) container_of(obj, struct efivar_entry,
> kobj)
> +#define kref_to_efivar_entry(obj) container_of(obj, struct efivar_entry,
> kref)
>
> /*
> * Prototype for sysfs creation function
> @@ -580,7 +590,7 @@ efivar_show_raw(struct efivar_entry *entry, char *buf)
> static ssize_t efivar_attr_show(struct kobject *kobj, struct attribute *attr,
> char *buf)
> {
> - struct efivar_entry *var = to_efivar_entry(kobj);
> + struct efivar_entry *var = kobj_to_efivar_entry(kobj);
> struct efivar_attribute *efivar_attr = to_efivar_attr(attr);
> ssize_t ret = -EIO;
>
> @@ -596,7 +606,7 @@ static ssize_t efivar_attr_show(struct kobject *kobj,
> struct attribute *attr,
> static ssize_t efivar_attr_store(struct kobject *kobj, struct attribute
> *attr,
> const char *buf, size_t count)
> {
> - struct efivar_entry *var = to_efivar_entry(kobj);
> + struct efivar_entry *var = kobj_to_efivar_entry(kobj);
> struct efivar_attribute *efivar_attr = to_efivar_attr(attr);
> ssize_t ret = -EIO;
>
> @@ -614,12 +624,28 @@ static const struct sysfs_ops efivar_attr_ops = {
> .store = efivar_attr_store,
> };
>
> -static void efivar_release(struct kobject *kobj)
> +static void efivar_entry_release(struct kref *kref)
> {
> - struct efivar_entry *var = container_of(kobj, struct efivar_entry,
> kobj);
> + struct efivar_entry *var = kref_to_efivar_entry(kref);
> kfree(var);
> }
>
> +static void efivar_entry_put(struct efivar_entry *entry)
> +{
> + kref_put(&entry->kref, efivar_entry_release);
> +}
> +
> +static void efivar_entry_get(struct efivar_entry *entry)
> +{
> + kref_get(&entry->kref);
> +}
> +
> +static void efivar_release(struct kobject *kobj)
> +{
> + struct efivar_entry *var = kobj_to_efivar_entry(kobj);
> + efivar_entry_put(var);
> +}
> +
> static EFIVAR_ATTR(guid, 0400, efivar_guid_read, NULL);
> static EFIVAR_ATTR(attributes, 0400, efivar_attr_read, NULL);
> static EFIVAR_ATTR(size, 0400, efivar_size_read, NULL);
> @@ -647,9 +673,22 @@ efivar_unregister(struct efivar_entry *var)
> kobject_put(&var->kobj);
> }
>
> +static bool efivar_is_present(struct efivar_entry *entry)
> +{
> + bool present;
> +
> + spin_lock(&entry->efivars->lock);
> + present = !list_empty(&entry->list);
> + spin_unlock(&entry->efivars->lock);
> +
> + return present;
> +}
> +
> static int efivarfs_file_open(struct inode *inode, struct file *file)
> {
> - file->private_data = inode->i_private;
> + struct efivar_entry *entry = inode->i_private;
> + file->private_data = entry;
> + efivar_entry_get(entry);
> return 0;
> }
Isn't there a race here? Can't the efivar_entry be deleted and freed
before we call efivar_get_entry()?
The problem is you need to be able to ensure the validity of
inode->i_private, but you can't in the open() function if you haven't
already taken a reference to the variable. A ref count of some
description needs to be incremented in efivarfs_fill_super() before the
efivar_entry pointer is stored in the inode, and while we're holding the
lock.
> @@ -706,6 +745,15 @@ static ssize_t efivarfs_file_write(struct file *file,
> if (attributes & ~(EFI_VARIABLE_MASK))
> return -EINVAL;
>
> + /*
> + * It's possible that the variable has been removed from firmware
> + * since the open(), in which case we need to flag an error
> + * TODO: allow writes that would reinstate the variable, but check
> + * against duplicate variables...
> + */
> + if (!efivar_is_present(var))
> + return -EIO;
> +
> efivars = var->efivars;
>
> /*
> @@ -789,9 +837,10 @@ static ssize_t efivarfs_file_write(struct file *file,
> mutex_unlock(&inode->i_mutex);
>
> } else if (status == EFI_NOT_FOUND) {
> - list_del(&var->list);
> + list_del_init(&var->list);
> spin_unlock(&efivars->lock);
> efivar_unregister(var);
> + efivar_entry_put(var);
> drop_nlink(inode);
> d_delete(file->f_dentry);
> dput(file->f_dentry);
> @@ -819,6 +868,13 @@ static ssize_t efivarfs_file_read(struct file *file,
> char __user *userbuf,
> void *data;
> ssize_t size = 0;
>
> + /*
> + * It's possible that the variable has been removed from firmware
> + * since the open(), in which case we need to flag an error.
> + */
> + if (!efivar_is_present(var))
> + return -EIO;
> +
> spin_lock(&efivars->lock);
> status = efivars->ops->get_variable(var->var.VariableName,
> &var->var.VendorGuid,
> @@ -854,8 +910,15 @@ out_free:
> return size;
> }
>
> +static int efivarfs_file_release(struct inode *inode, struct file *file)
> +{
> + efivar_entry_put(file->private_data);
> + return 0;
> +}
> +
> static void efivarfs_evict_inode(struct inode *inode)
> {
> + efivar_entry_put(inode->i_private);
> clear_inode(inode);
> }
>
> @@ -871,10 +934,11 @@ static struct super_block *efivarfs_sb;
> static const struct inode_operations efivarfs_dir_inode_operations;
>
> static const struct file_operations efivarfs_file_operations = {
> - .open = efivarfs_file_open,
> - .read = efivarfs_file_read,
> - .write = efivarfs_file_write,
> - .llseek = no_llseek,
> + .open = efivarfs_file_open,
> + .read = efivarfs_file_read,
> + .write = efivarfs_file_write,
> + .release = efivarfs_file_release,
> + .llseek = no_llseek,
> };
>
> static struct inode *efivarfs_get_inode(struct super_block *sb,
> @@ -946,6 +1010,8 @@ static int efivarfs_create(struct inode *dir, struct
> dentry *dentry,
> goto out;
> }
>
> + kref_init(&var->kref);
> +
> /* length of the variable name itself: remove GUID and separator */
> namelen = dentry->d_name.len - GUID_LEN - 1;
>
> @@ -969,6 +1035,7 @@ static int efivarfs_create(struct inode *dir, struct
> dentry *dentry,
> kobject_uevent(&var->kobj, KOBJ_ADD);
> spin_lock(&efivars->lock);
> list_add(&var->list, &efivars->list);
> + efivar_entry_get(var);
> spin_unlock(&efivars->lock);
> d_instantiate(dentry, inode);
> dget(dentry);
> @@ -993,9 +1060,10 @@ static int efivarfs_unlink(struct inode *dir, struct
> dentry *dentry)
> 0, 0, NULL);
>
> if (status == EFI_SUCCESS || status == EFI_NOT_FOUND) {
> - list_del(&var->list);
> + list_del_init(&var->list);
> spin_unlock(&efivars->lock);
> efivar_unregister(var);
> + efivar_entry_put(var);
> drop_nlink(dentry->d_inode);
> dput(dentry);
> return 0;
> @@ -1077,6 +1145,7 @@ int efivarfs_fill_super(struct super_block *sb, void
> *data, int silent)
>
> mutex_lock(&inode->i_mutex);
> inode->i_private = entry;
> + efivar_entry_get(entry);
> i_size_write(inode, size+4);
> mutex_unlock(&inode->i_mutex);
> d_add(dentry, inode);
> @@ -1221,8 +1290,10 @@ static int efi_pstore_write(enum pstore_type_id type,
> 0, NULL);
> }
>
> - if (found)
> - list_del(&found->list);
> + if (found) {
> + list_del_init(&found->list);
> + efivar_entry_put(found);
> + }
>
> for (i = 0; i < DUMP_NAME_LEN; i++)
> efi_name[i] = name[i];
> @@ -1414,7 +1485,7 @@ static ssize_t efivar_delete(struct file *filp, struct
> kobject *kobj,
> spin_unlock(&efivars->lock);
> return -EIO;
> }
> - list_del(&search_efivar->list);
> + list_del_init(&search_efivar->list);
> /* We need to release this lock before unregistering. */
> spin_unlock(&efivars->lock);
> efivar_unregister(search_efivar);
> @@ -1533,6 +1604,7 @@ efivar_create_sysfs_entry(struct efivars *efivars,
>
> spin_lock(&efivars->lock);
> list_add(&new_efivar->list, &efivars->list);
> + efivar_entry_get(new_efivar);
> spin_unlock(&efivars->lock);
>
> return 0;
> @@ -1603,8 +1675,9 @@ void unregister_efivars(struct efivars *efivars)
>
> list_for_each_entry_safe(entry, n, &efivars->list, list) {
> spin_lock(&efivars->lock);
> - list_del(&entry->list);
> + list_del_init(&entry->list);
> spin_unlock(&efivars->lock);
> + efivar_entry_put(entry);
> efivar_unregister(entry);
> }
> if (efivars->new_var)
--
Matt Fleming, Intel Open Source Technology Center
--
To unsubscribe from this list: send the line "unsubscribe linux-efi" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
