Re: [PATCH] p9caps: add Plan9 capability devices

[Richard Weinberger]

On Sun, Feb 11, 2018 at 10:50 PM, Enrico Weigelt, metux IT consult
<metux@xxxxxx> wrote:
> From: "Enrico Weigelt, metux IT consult" <info@xxxxxxxxx>
>
> This driver implements the Plan9 capability devices, used for
> switching user id via capability tokens.
>
> https://9p.io/sys/doc/auth.html

Please see some nit-picks below.
I'm still reading the plan9 docs to fully get the big picture.

> ---
>  drivers/staging/Kconfig         |   2 +
>  drivers/staging/Makefile        |   1 +
>  drivers/staging/p9caps/Kconfig  |  11 ++
>  drivers/staging/p9caps/Makefile |   1 +
>  drivers/staging/p9caps/p9caps.c | 369 ++++++++++++++++++++++++++++++++++++++++
>  5 files changed, 384 insertions(+)
>  create mode 100644 drivers/staging/p9caps/Kconfig
>  create mode 100644 drivers/staging/p9caps/Makefile
>  create mode 100644 drivers/staging/p9caps/p9caps.c
>
> diff --git a/drivers/staging/Kconfig b/drivers/staging/Kconfig
> index 554683912cff..23f325339fe8 100644
> --- a/drivers/staging/Kconfig
> +++ b/drivers/staging/Kconfig
> @@ -118,4 +118,6 @@ source "drivers/staging/vboxvideo/Kconfig"
>
>  source "drivers/staging/pi433/Kconfig"
>
> +source "drivers/staging/p9caps/Kconfig"
> +
>  endif # STAGING
> diff --git a/drivers/staging/Makefile b/drivers/staging/Makefile
> index 6e536020029a..eccdf4643453 100644
> --- a/drivers/staging/Makefile
> +++ b/drivers/staging/Makefile
> @@ -3,6 +3,7 @@
>
>  obj-y                          += media/
>  obj-y                          += typec/
> +obj-$(CONFIG_PLAN9CAPS)                += p9caps/
>  obj-$(CONFIG_IRDA)             += irda/net/
>  obj-$(CONFIG_IRDA)             += irda/drivers/
>  obj-$(CONFIG_PRISM2_USB)       += wlan-ng/
> diff --git a/drivers/staging/p9caps/Kconfig b/drivers/staging/p9caps/Kconfig
> new file mode 100644
> index 000000000000..b909daaa79ce
> --- /dev/null
> +++ b/drivers/staging/p9caps/Kconfig
> @@ -0,0 +1,11 @@
> +config PLAN9CAPS
> +       tristate "Plan 9 capability device"
> +       default n
> +       select CRYPTO_HMAC
> +       select CRYPTO_SHA1
> +       help
> +         This module implements the Plan 9 capability devices
> +         /dev/caphash and /dev/capuse
> +
> +         To compile this driver as a module, choose
> +         M here: the module will be called p9caps.
> diff --git a/drivers/staging/p9caps/Makefile b/drivers/staging/p9caps/Makefile
> new file mode 100644
> index 000000000000..67d38099a249
> --- /dev/null
> +++ b/drivers/staging/p9caps/Makefile
> @@ -0,0 +1 @@
> +obj-$(CONFIG_PLAN9CAPS)        += p9caps.o
> diff --git a/drivers/staging/p9caps/p9caps.c b/drivers/staging/p9caps/p9caps.c
> new file mode 100644
> index 000000000000..e46b09821c18
> --- /dev/null
> +++ b/drivers/staging/p9caps/p9caps.c
> @@ -0,0 +1,369 @@
> +
> +#include <linux/kernel.h>
> +#include <linux/module.h>
> +#include <linux/slab.h>
> +#include <linux/fs.h>
> +#include <linux/errno.h>
> +#include <linux/fcntl.h>
> +#include <linux/cdev.h>
> +#include <linux/list.h>
> +#include <linux/mm.h>
> +#include <linux/string.h>
> +#include <linux/scatterlist.h>
> +#include <linux/cred.h>
> +#include <linux/err.h>
> +#include <linux/user_namespace.h>
> +#include <linux/mutex.h>
> +#include <crypto/hash.h>
> +#include <crypto/sha.h>
> +
> +/*
> + * Plan9 /dev/caphash and /dev/capuse device
> + *
> + * 2DO: - caphash should only allow one process (per userns)
> + *      - support textual user names
> + *      - invalidate old caps
> + */
> +
> +#define DEVICE_CAPUSE  "/dev/capuse"
> +#define DEVICE_CAPHASH "/dev/caphash"
> +
> +struct caphash_entry {
> +       struct list_head list;
> +       struct user_namespace *user_ns;
> +       char data[SHA1_DIGEST_SIZE];
> +};
> +
> +struct caphash_writer {
> +       struct list_head list;
> +       struct user_namespace *user_ns;
> +};
> +
> +static dev_t caphash_devid = 0;
> +static dev_t capuse_devid = 0;

Static vars are already 0, no need to initialize them.

> +static LIST_HEAD(caphash_entries);
> +static LIST_HEAD(caphash_writers);
> +
> +static DEFINE_MUTEX(lock);
> +
> +struct crypto_ahash *hmac_tfm = NULL;
> +
> +static int caphash_open(struct inode *inode, struct file *filp)
> +{
> +       struct caphash_writer *tmp = NULL;
> +       struct user_namespace *user_ns = current_user_ns();
> +       int retval = 0;
> +       struct list_head *pos, *q;
> +
> +       /* make sure only one instance per namespace can be opened */
> +       mutex_lock(&lock);
> +
> +       list_for_each_safe(pos, q, &(caphash_writers)) {
> +               tmp = list_entry(pos, struct caphash_writer, list);
> +               if (tmp->user_ns == user_ns) {
> +                       pr_err("already locked in this namespace\n");

So, evil guy opens the device but does not close it, therefore the
whole service is blocked in a namespace?

In general, I think we should open that device in
kernel_init_freeable() and hand over the fd to init/systemd.
As the plan9 docs state "The write-only file /dev/caphash can be
opened only by the host owner, and only once. Factotum opens this file
immediately after booting."

> +                       retval = -EBUSY;
> +                       goto out;
> +               }
> +       }
> +
> +       if (!(tmp = kzalloc(sizeof(struct caphash_writer), GFP_KERNEL))) {
> +               retval = -ENOMEM;
> +               goto out;
> +       }
> +
> +       tmp->user_ns = get_user_ns(user_ns);
> +       list_add(&(tmp->list), &caphash_writers);
> +
> +out:
> +       mutex_unlock(&lock);
> +       return retval;
> +}
> +
> +static int caphash_release(struct inode *inode, struct file *filp)
> +{
> +       int retval = 0;
> +       struct user_namespace *user_ns = current_user_ns();

Why not obtaining the user namespace from the open file?
That way one can close a caphash file hande she never opened.
Think of open, followed by nsenter, ...

> +       struct list_head *pos, *q;
> +       struct caphash_entry *tmp;
> +
> +       mutex_lock(&lock);
> +
> +       list_for_each_safe(pos, q, &(caphash_writers)) {
> +               tmp = list_entry(pos, struct caphash_entry, list);

list_for_each_entry.

> +               if (tmp->user_ns == user_ns) {
> +                       list_del(pos);
> +                       kfree(tmp);
> +                       goto out;
> +               }
> +       }
> +
> +out:
> +       mutex_unlock(&lock);
> +       return retval;
> +}
> +
> +static ssize_t caphash_write(struct file *filp, const char __user *buf,
> +                                  size_t count, loff_t *f_pos)
> +{
> +       struct caphash_entry *ent;
> +
> +       if (count > SHA1_DIGEST_SIZE) {
> +               pr_err("SHA1 digest size too large: %d\n", count);
> +               return -E2BIG;
> +       }
> +
> +       if (!(ent = kzalloc(sizeof(struct caphash_entry), GFP_KERNEL)))
> +               return -ENOMEM;
> +
> +       if (copy_from_user(&(ent->data), buf, count)) {
> +               kfree(ent);
> +               return -EFAULT;
> +       }
> +
> +       ent->user_ns = get_user_ns(current_user_ns());

Again, why not taking the user namespace from the file handle?

> +       mutex_lock(&lock);
> +       list_add(&(ent->list), &caphash_entries);
> +       mutex_unlock(&lock);
> +
> +       return count;
> +}
> +
> +/* called w/ lock held. we can relieve this by allocating tfm locally */
> +static ssize_t hash(const char *src, const char* dst, const char *key, u8 *result)
> +{
> +       struct scatterlist sg;
> +       struct ahash_request *req;
> +       int retval;
> +       char *text = NULL;
> +       size_t text_len;
> +       int digest_len;
> +       u8* digest = NULL;
> +
> +       text_len = strlen(src)+strlen(dst)+1;           /* src@dst */
> +       digest_len = crypto_ahash_reqsize(hmac_tfm);
> +
> +       digest = kzalloc(digest_len, GFP_KERNEL);
> +       text = kzalloc(text_len+1, GFP_KERNEL);
> +
> +       if (!digest || !text) {
> +               retval = -ENOMEM;
> +               goto out;
> +       }
> +
> +       if (!(req = ahash_request_alloc(hmac_tfm, GFP_KERNEL))) {
> +               pr_err("failed to alloc ahash_request\n");
> +               retval = -ENOMEM;
> +               goto out;
> +       }
> +
> +       snprintf(text, text_len+1, "%s@%s", src, dst);
> +       sg_set_buf(&sg, text, text_len);
> +
> +       ahash_request_set_callback(req, 0, NULL, NULL);
> +       ahash_request_set_crypt(req, &sg, digest, text_len);
> +
> +       if ((retval = crypto_ahash_setkey(hmac_tfm, key, strlen(key)))) {
> +               pr_err("crypto_ahash_setkey() failed ret=%d\n", retval);
> +               goto out;
> +       }
> +
> +       if ((retval = crypto_ahash_digest(req))) {
> +               pr_err("digest() failed ret=%d\n", retval);
> +               goto out;
> +       }
> +
> +       memcpy(result, digest, SHA1_DIGEST_SIZE);
> +
> +out:
> +       kfree(text);
> +       kfree(digest);
> +
> +       return 0;
> +}
> +
> +static inline kuid_t convert_uid(const char* uname)
> +{
> +       return make_kuid(current_user_ns(), simple_strtol(uname, NULL, 0));
> +}
> +
> +static ssize_t switch_uid(const char *src_uname, const char *dst_uname)
> +{
> +       struct cred *creds = prepare_creds();
> +
> +       kuid_t src_uid = convert_uid(src_uname);
> +       kuid_t dst_uid = convert_uid(dst_uname);
> +
> +       if (!uid_eq(src_uid, current_uid())) {
> +               pr_info("src uid mismatch\n");
> +               return -EPERM;
> +       }
> +
> +       if (!(creds = prepare_creds()))
> +               return -ENOMEM;
> +
> +       creds->uid = dst_uid;
> +       creds->euid = dst_uid;
> +
> +       pr_info("switching from kuid %d to %d\n", src_uid.val, dst_uid.val);
> +       return commit_creds(creds);
> +}
> +
> +static ssize_t try_switch(const char* src_uname, const char* dst_uname, const u8* hashval)
> +{
> +       struct list_head *pos;
> +       list_for_each(pos, &(caphash_entries)) {
> +               struct caphash_entry *tmp = list_entry(pos, struct caphash_entry, list);
> +               if ((0 == memcmp(hashval, tmp->data, SHA1_DIGEST_SIZE)) &&
> +                   (tmp->user_ns == current_user_ns())) {
> +
> +                       int retval;
> +
> +                       if ((retval = switch_uid(src_uname, dst_uname))) {
> +                               pr_info("uid switch failed\n");
> +                               return retval;
> +                       }
> +
> +                       tmp = list_entry(pos, struct caphash_entry, list);
> +                       list_del(pos);
> +                       put_user_ns(tmp->user_ns);
> +                       kfree(tmp);
> +
> +                       return 0;
> +               }
> +       }
> +
> +       pr_info("cap not found\n");
> +
> +       return -ENOENT;
> +}
> +
> +static ssize_t capuse_write(struct file *filp, const char __user *buf,
> +                                 size_t count, loff_t *f_pos)
> +{
> +       ssize_t retval = count;
> +       char  *rand_str, *src_uname, *dst_uname;
> +       u8 hashval[SHA1_DIGEST_SIZE] = { 0 };
> +       char *cmdbuf;
> +
> +       if (!(cmdbuf = kzalloc(count, GFP_KERNEL)))

count is unbound, please apply a limit check.

> +               return -ENOMEM;
> +
> +       if (copy_from_user(cmdbuf, buf, count)) {
> +               retval = -EFAULT;
> +               goto out_free;
> +       }
> +
> +       {
> +               char *walk = cmdbuf;

cmdbuf is toxic, make sure it is at least nul-terminated.

> +               src_uname = strsep(&walk, "@");
> +               dst_uname = strsep(&walk, "@");
> +               rand_str = walk;
> +               if (!src_uname || !dst_uname || !rand_str) {
> +                       retval = -EINVAL;
> +                       goto out_free;
> +               }
> +       }
> +
> +       mutex_lock(&lock);
> +
> +       if ((retval = hash(src_uname, dst_uname, rand_str, hashval)))
> +               goto out_unlock;
> +
> +       if ((retval = try_switch(src_uname, dst_uname, hashval)))
> +               goto out_unlock;
> +
> +       retval = count;
> +
> +out_unlock:
> +       mutex_unlock(&lock);
> +
> +out_free:
> +       kfree(cmdbuf);
> +       return retval;
> +}
> +
> +static const struct file_operations caphash_fops = {
> +       .owner          = THIS_MODULE,
> +       .write          = caphash_write,
> +       .open           = caphash_open,
> +       .release        = caphash_release,
> +};
> +
> +static const struct file_operations capuse_fops = {
> +       .owner          = THIS_MODULE,
> +       .write          = capuse_write,
> +};
> +
> +static struct cdev caphash_dev;
> +static struct cdev capuse_dev;
> +
> +static int clear(void)
> +{
> +       struct caphash_entry *tmp;
> +       struct list_head *pos, *q;
> +
> +       list_for_each_safe(pos, q, &(caphash_entries)) {
> +               tmp = list_entry(pos, struct caphash_entry, list);

list_for_each_entry_safe?

> +               list_del(pos);
> +               kfree(tmp);
> +       }
> +
> +       return 0;
> +}
> +
> +static void _cleanup_module(void)
> +{
> +       clear();
> +
> +       cdev_del(&caphash_dev);
> +       cdev_del(&capuse_dev);
> +
> +       unregister_chrdev_region(caphash_devid, 1);
> +       unregister_chrdev_region(capuse_devid, 1);
> +
> +       if (hmac_tfm)

IS_ERR()? Otherwise you free a error value, if crypto_alloc_ahash() fails.

> +               crypto_free_ahash(hmac_tfm);
> +}
> +
> +static int _init_module(void)
> +{
> +       int retval;
> +
> +       hmac_tfm = crypto_alloc_ahash("hmac(sha1)", 0, CRYPTO_ALG_ASYNC);
> +       if (IS_ERR(hmac_tfm)) {
> +               retval = -PTR_ERR(hmac_tfm);
> +               pr_err("failed to load transform for hmac(sha1): %d\n", retval);
> +               goto fail;
> +       }
> +
> +       if ((retval = alloc_chrdev_region(&caphash_devid, 0, 1, DEVICE_CAPHASH)))
> +               goto fail;
> +
> +       if ((retval = alloc_chrdev_region(&capuse_devid, 0, 1, DEVICE_CAPUSE)))
> +               goto fail;
> +
> +       cdev_init(&caphash_dev, &caphash_fops);
> +       caphash_dev.owner = THIS_MODULE;
> +       if ((retval = cdev_add(&caphash_dev, caphash_devid, 1)))
> +               pr_err("failed adding " DEVICE_CAPHASH ": %d\n", retval);
> +
> +       cdev_init(&capuse_dev, &capuse_fops);
> +       capuse_dev.owner = THIS_MODULE;
> +       if ((retval = cdev_add(&capuse_dev, capuse_devid, 1)))
> +               pr_err("failed adding " DEVICE_CAPUSE ": %d\n", retval);
> +
> +       return 0;
> +
> +fail:
> +       _cleanup_module();
> +       return retval;
> +}
> +
> +MODULE_AUTHOR("Enrico Weigelt, metux IT consult <info@xxxxxxxxx>");
> +MODULE_LICENSE("GPL");
> +
> +module_init(_init_module);
> +module_exit(_cleanup_module);
> --
> 2.11.0
>

-- 
Thanks,
//richard