blob: 0c49fbdd4f287f0f338f84a0bebefecad89cf5ac [file] [log] [blame]
/*
* Copyright (c) 2016-2017, The Linux Foundation. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 and
* only version 2 as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*/
/*
* Per-File-Tagger (PFT).
*
* This driver tags enterprise file for encryption/decryption,
* as part of the Per-File-Encryption (PFE) feature.
*
* Enterprise registered applications are identified by their UID.
*
* The PFT exposes character-device interface to the user-space application,
* to handle the following commands:
* 1. Update registered applications list
* 2. Encryption (in-place) of a file that was created before.
* 3. Set State - update the state.
*
* The PFT exposes kernel API hooks that are intercepting file operations
* like create/open/read/write for tagging files and also for access control.
* It utilizes the existing security framework hooks
* that calls the selinux hooks.
*
* The PFT exposes kernel API to the dm-req-crypt driver to provide the info
* if a file is tagged or not. The dm-req-crypt driver is doing the
* actual encryption/decryptiom.
*
* Tagging the file:
* 1. Non-volatile tagging on storage using file extra-attribute (xattr).
* 2. Volatile tagging on the file's inode, for fast access.
*
*/
/* Uncomment the line below to enable debug messages */
/* #define DEBUG 1 */
#define pr_fmt(fmt) "pft [%s]: " fmt, __func__
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/cdev.h>
#include <linux/sched.h>
#include <linux/uaccess.h>
#include <linux/cred.h>
#include <linux/list.h>
#include <linux/errno.h>
#include <linux/printk.h>
#include <linux/blkdev.h>
#include <linux/elevator.h>
#include <linux/bio.h>
#include <linux/bitops.h>
#include <linux/fdtable.h>
#include <linux/selinux.h>
#include <linux/security.h>
#include <linux/lsm_hooks.h>
#include <linux/pft.h>
#include <uapi/linux/msm_pft.h>
#include "objsec.h"
/* File tagging as encrypted/non-encrypted is valid */
#define PFT_TAG_MAGIC ((u32)(0xABC00000))
/* File tagged as encrypted */
#define PFT_TAG_ENCRYPTED BIT(16)
#define PFT_TAG_MAGIC_MASK 0xFFF00000
#define PFT_TAG_FLAGS_MASK 0x000F0000
#define PFT_TAG_KEY_MASK 0x0000FFFF
/* The default encryption key index */
#define PFT_DEFAULT_KEY_INDEX 1
/* The default key index for non-encrypted files */
#define PFT_NO_KEY 0
/* PFT extended attribute name */
#define XATTR_NAME_PFE "security.pfe"
/* PFT driver requested major number */
#define PFT_REQUESTED_MAJOR 213
/* PFT driver name */
#define DEVICE_NAME "pft"
/* Maximum registered applications */
#define PFT_MAX_APPS 1000
/* Maximum command size */
#define PFT_MAX_COMMAND_SIZE (PAGE_SIZE)
/* Current Process ID */
#define current_pid() ((u32)(current->pid))
static const char *pft_state_name[PFT_STATE_MAX_INDEX] = {
"deactivated",
"deactivating",
"key_removed",
"removing_key",
"key_loaded",
};
/**
* struct pft_file_info - pft file node info.
* @file: pointer to file stucture.
* @pid: process ID.
* @list: next list item.
*
* A node in the list of the current open encrypted files.
*/
struct pft_file_info {
struct file *file;
pid_t pid;
struct list_head list;
};
/**
* struct pft_device - device state structure.
*
* @open_count: device open count.
* @major: device major number.
* @state: Per-File-Encryption state.
* @response: command response.
* @pfm_pid: PFM process id.
* @inplace_file: file for in-place encryption.
* @uid_table: registered application array (UID).
* @uid_count: number of registered applications.
* @open_file_list: open encrypted file list.
* @lock: lock protect list access.
*
* The open_count purpose is to ensure that only one user space
* application uses this driver.
* The open_file_list is used to close open encrypted files
* after the key is removed from the encryption hardware.
*/
struct pft_device {
struct cdev cdev;
dev_t device_no;
struct class *driver_class;
int open_count;
int major;
enum pft_state state;
struct pft_command_response response;
u32 pfm_pid;
struct file *inplace_file;
kuid_t *uid_table;
u32 uid_count;
struct list_head open_file_list;
struct mutex lock;
bool is_chosen_lsm;
};
/* Device Driver State */
static struct pft_device *pft_dev;
static struct inode *pft_bio_get_inode(const struct bio *bio);
static int pft_inode_alloc_security(struct inode *inode)
{
struct inode_security_struct *i_sec = NULL;
i_sec = kzalloc(sizeof(*i_sec), GFP_KERNEL);
if (i_sec == NULL)
return -ENOMEM;
inode->i_security = i_sec;
return 0;
}
static void pft_inode_free_security(struct inode *inode)
{
kzfree(inode->i_security);
}
static struct security_hook_list pft_hooks[] = {
LSM_HOOK_INIT(inode_create, pft_inode_create),
LSM_HOOK_INIT(inode_post_create, pft_inode_post_create),
LSM_HOOK_INIT(inode_unlink, pft_inode_unlink),
LSM_HOOK_INIT(inode_mknod, pft_inode_mknod),
LSM_HOOK_INIT(inode_rename, pft_inode_rename),
LSM_HOOK_INIT(inode_setxattr, pft_inode_set_xattr),
LSM_HOOK_INIT(inode_alloc_security, pft_inode_alloc_security),
LSM_HOOK_INIT(inode_free_security, pft_inode_free_security),
LSM_HOOK_INIT(file_open, pft_file_open),
LSM_HOOK_INIT(file_permission, pft_file_permission),
LSM_HOOK_INIT(file_close, pft_file_close),
};
static int __init pft_lsm_init(struct pft_device *dev)
{
/* Check if PFT is the chosen lsm via security_module_enable() */
if (security_module_enable("pft")) {
security_add_hooks(pft_hooks, ARRAY_SIZE(pft_hooks));
dev->is_chosen_lsm = true;
pr_debug("pft is the chosen lsm, registered successfully !\n");
} else {
pr_err("pft is not the chosen lsm\n");
return -ENODEV;
}
return 0;
}
/**
* pft_is_ready() - driver is initialized and ready.
*
* Return: true if the driver is ready.
*/
static bool pft_is_ready(void)
{
return pft_dev != NULL;
}
/**
* file_to_filename() - get the filename from file pointer.
* @filp: file pointer
*
* it is used for debug prints.
*
* Return: filename string or "unknown".
*/
static char *file_to_filename(struct file *filp)
{
struct dentry *dentry = NULL;
char *filename = NULL;
if (!filp || !filp->f_path.dentry)
return "unknown";
dentry = filp->f_path.dentry;
filename = dentry->d_iname;
return filename;
}
/**
* inode_to_filename() - get the filename from inode pointer.
* @inode: inode pointer
*
* it is used for debug prints.
*
* Return: filename string or "unknown".
*/
static char *inode_to_filename(struct inode *inode)
{
struct dentry *dentry = NULL;
char *filename = NULL;
if (hlist_empty(&inode->i_dentry))
return "unknown";
dentry = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
filename = dentry->d_iname;
return filename;
}
/**
* pft_set_response() - set response error code.
*
* @error_code: The error code to return on response.
*/
static inline void pft_set_response(u32 error_code)
{
pft_dev->response.error_code = error_code;
}
/**
* pft_add_file()- Add the file to the list of opened encrypted
* files.
* @filp: file to add.
*
* Return: 0 of successful operation, negative value otherwise.
*/
static int pft_add_file(struct file *filp)
{
struct pft_file_info *node = NULL;
node = kzalloc(sizeof(*node), GFP_KERNEL);
if (!node)
return -ENOMEM;
node->file = filp;
INIT_LIST_HEAD(&node->list);
mutex_lock(&pft_dev->lock);
list_add(&node->list, &pft_dev->open_file_list);
pr_debug("adding file %s to open list.\n", file_to_filename(filp));
mutex_unlock(&pft_dev->lock);
return 0;
}
/**
* pft_remove_file()- Remove the given file from the list of
* open encrypted files.
* @filp: file to remove.
*
* Return: 0 on success, negative value on failure.
*/
static int pft_remove_file(struct file *filp)
{
int ret = -ENOENT;
struct pft_file_info *tmp = NULL;
struct list_head *pos = NULL;
struct list_head *next = NULL;
bool found = false;
mutex_lock(&pft_dev->lock);
list_for_each_safe(pos, next, &pft_dev->open_file_list) {
tmp = list_entry(pos, struct pft_file_info, list);
if (filp == tmp->file) {
found = true;
break;
}
}
if (found) {
pr_debug("remove file %s. from open list.\n ",
file_to_filename(filp));
list_del(&tmp->list);
kfree(tmp);
ret = 0;
}
mutex_unlock(&pft_dev->lock);
return ret;
}
/**
* pft_is_current_process_registered()- Check if current process
* is registered.
*
* Return: true if current process is registered.
*/
static bool pft_is_current_process_registered(void)
{
int is_registered = false;
int i;
kuid_t uid = current_uid();
mutex_lock(&pft_dev->lock);
for (i = 0; i < pft_dev->uid_count; i++) {
if (uid_eq(pft_dev->uid_table[i], uid)) {
pr_debug("current UID [%u] is registered.\n",
__kuid_val(uid));
is_registered = true;
break;
}
}
mutex_unlock(&pft_dev->lock);
return is_registered;
}
/**
* pft_is_xattr_supported() - Check if the filesystem supports
* extended attributes.
* @indoe: pointer to the file inode
*
* Return: true if supported, false if not.
*/
static bool pft_is_xattr_supported(struct inode *inode)
{
if (inode == NULL) {
pr_err("invalid argument inode passed as NULL");
return false;
}
if (inode->i_security == NULL) {
pr_debug("i_security is NULL, not ready yet\n");
return false;
}
if (inode->i_op == NULL) {
pr_debug("i_op is NULL\n");
return false;
}
if (inode->i_op->getxattr == NULL) {
pr_debug_once("getxattr() not supported , filename=%s\n",
inode_to_filename(inode));
return false;
}
if (inode->i_op->setxattr == NULL) {
pr_debug("setxattr() not supported\n");
return false;
}
return true;
}
/**
* pft_get_inode_tag() - get the file tag.
* @indoe: pointer to the file inode
*
* Return: tag
*/
static u32 pft_get_inode_tag(struct inode *inode)
{
struct inode_security_struct *isec = inode->i_security;
if (isec == NULL)
return 0;
return isec->tag;
}
/**
* pft_get_inode_key_index() - get the file key.
* @indoe: pointer to the file inode
*
* Return: key index
*/
static inline u32 pft_get_inode_key_index(struct inode *inode)
{
return pft_get_inode_tag(inode) & PFT_TAG_KEY_MASK;
}
/**
* pft_is_tag_valid() - is the tag valid
* @indoe: pointer to the file inode
*
* The tagging is set to valid when an enterprise file is created
* or when an file is opened first time after power up and the
* xattr was checked to see if the file is encrypted or not.
*
* Return: true if the tag is valid.
*/
static inline bool pft_is_tag_valid(struct inode *inode)
{
struct inode_security_struct *isec = inode->i_security;
if (isec == NULL)
return false;
return ((isec->tag & PFT_TAG_MAGIC_MASK) == PFT_TAG_MAGIC) ?
true : false;
}
/**
* pft_is_file_encrypted() - is inode tagged as encrypted.
*
* @tag: holds the key index and tagging flags.
*
* Return: true if the file is encrypted.
*/
static inline bool pft_is_file_encrypted(u32 tag)
{
return (tag & PFT_TAG_ENCRYPTED) ? true : false;
}
/**
* pft_tag_inode_non_encrypted() - Tag the inode as
* non-encrypted.
* @indoe: pointer to the file inode
*
* Tag file as non-encrypted, only the valid bit is set,
* the encrypted bit is not set.
*/
static inline void pft_tag_inode_non_encrypted(struct inode *inode)
{
struct inode_security_struct *isec = inode->i_security;
isec->tag = (u32)(PFT_TAG_MAGIC);
}
/**
* pft_tag_inode_encrypted() - Tag the inode as encrypted.
* @indoe: pointer to the file inode
*
* Set the valid bit, the encrypted bit, and the key index.
*/
static void pft_tag_inode_encrypted(struct inode *inode, u32 key_index)
{
struct inode_security_struct *isec = inode->i_security;
isec->tag = key_index | PFT_TAG_ENCRYPTED | PFT_TAG_MAGIC;
}
/**
* pft_get_file_tag()- get the file tag.
* @dentry: pointer to file dentry.
* @tag_ptr: pointer to tag.
*
* This is the major function for detecting tag files.
* Get the tag from the inode if tag is valid,
* or from the xattr if this is the 1st time after power up.
*
* Return: 0 on successe, negative value on failure.
*/
static int pft_get_file_tag(struct dentry *dentry, u32 *tag_ptr)
{
ssize_t size = 0;
struct inode *inode;
const char *xattr_name = XATTR_NAME_PFE;
u32 key;
if (!dentry || !dentry->d_inode || !tag_ptr) {
pr_err("invalid param");
return -EINVAL;
}
inode = dentry->d_inode;
if (pft_is_tag_valid(inode)) {
*tag_ptr = pft_get_inode_tag(inode);
return 0;
}
/*
* For the first time reading the tag, the tag is not valid, hence
* get xattr.
*/
size = inode->i_op->getxattr(dentry, xattr_name, &key, sizeof(key));
if (size == -ENODATA || size == -EOPNOTSUPP) {
pft_tag_inode_non_encrypted(inode);
*tag_ptr = pft_get_inode_tag(inode);
} else if (size > 0) {
pr_debug("First time file %s opened, found xattr = %u.\n",
inode_to_filename(inode), key);
pft_tag_inode_encrypted(inode, key);
*tag_ptr = pft_get_inode_tag(inode);
} else {
pr_err("getxattr() failure, ret=%zu.\n", size);
return -EINVAL;
}
return 0;
}
/**
* pft_tag_file() - Tag the file saving the key_index.
* @dentry: file dentry.
* @key_index: encryption key index.
*
* This is the major function for tagging a file.
* Tag the file on both the xattr and the inode.
*
* Return: 0 on successe, negative value on failure.
*/
static int pft_tag_file(struct dentry *dentry, u32 key_index)
{
int size = 0;
const char *xattr_name = XATTR_NAME_PFE;
if (!dentry || !dentry->d_inode) {
pr_err("invalid NULL param");
return -EINVAL;
}
if (!pft_is_xattr_supported(dentry->d_inode)) {
pr_err("set xattr for file %s is not support.\n",
dentry->d_iname);
return -EINVAL;
}
size = dentry->d_inode->i_op->setxattr(dentry, xattr_name, &key_index,
sizeof(key_index), 0);
if (size < 0) {
pr_err("failed to set xattr for file %s, ret =%d.\n",
dentry->d_iname, size);
return -EFAULT;
}
pft_tag_inode_encrypted(dentry->d_inode, key_index);
pr_debug("file %s tagged encrypted\n", dentry->d_iname);
return 0;
}
/**
* pft_get_app_key_index() - get the application key index.
* @uid: registered application UID
*
* Get key index based on the given registered application UID.
* Currently only one key is supported.
*
* Return: encryption key index.
*/
static inline u32 pft_get_app_key_index(kuid_t uid)
{
return PFT_DEFAULT_KEY_INDEX;
}
/**
* pft_is_encrypted_file() - is the file encrypted.
* @dentry: file pointer.
*
* Return: true if the file is encrypted, false otherwise.
*/
static bool pft_is_encrypted_file(struct dentry *dentry)
{
int rc;
u32 tag;
if (!pft_is_ready())
return false;
if (!pft_is_xattr_supported(dentry->d_inode))
return false;
rc = pft_get_file_tag(dentry, &tag);
if (rc < 0)
return false;
return pft_is_file_encrypted(tag);
}
/**
* pft_is_inplace_inode() - is this the inode of file for
* in-place encryption.
* @inode: inode of file to check.
*
* Return: true if this file is being encrypted, false
* otherwise.
*/
static bool pft_is_inplace_inode(struct inode *inode)
{
if (!pft_dev->inplace_file || !pft_dev->inplace_file->f_path.dentry)
return false;
return pft_dev->inplace_file->f_path.dentry->d_inode == inode;
}
/**
* pft_is_inplace_file() - is this the file for in-place
* encryption.
* @filp: file to check.
*
* A file struct might be allocated per process, inode should be
* only one.
*
* Return: true if this file is being encrypted, false
* otherwise.
*/
static inline bool pft_is_inplace_file(struct file *filp)
{
if (!filp || !filp->f_path.dentry || !filp->f_path.dentry->d_inode)
return false;
return pft_is_inplace_inode(filp->f_path.dentry->d_inode);
}
/**
* pft_get_key_index() - get the key index and other indications
* @inode: Pointer to inode struct
* @key_index: Pointer to the return value of key index
* @is_encrypted: Pointer to the return value.
* @is_inplace: Pointer to the return value.
*
* Provides the given inode's encryption key index, and well as
* indications whether the file is encrypted or is it currently
* being in-placed encrypted.
* This API is called by the dm-req-crypt to decide if to
* encrypt/decrypt the file.
* File tagging depends on the hooks to be called from selinux,
* so if selinux is disabled then tagging is also not
* valid.
*
* Return: 0 on successe, negative value on failure.
*/
int pft_get_key_index(struct bio *bio, u32 *key_index,
bool *is_encrypted, bool *is_inplace)
{
u32 tag = 0;
struct inode *inode = NULL;
if (!pft_is_ready())
return -ENODEV;
if (!selinux_is_enabled() && !pft_dev->is_chosen_lsm)
return -ENODEV;
if (!bio)
return -EPERM;
if (!is_encrypted) {
pr_err("is_encrypted is NULL\n");
return -EPERM;
}
if (!is_inplace) {
pr_err("is_inplace is NULL\n");
return -EPERM;
}
if (!key_index) {
pr_err("key_index is NULL\n");
return -EPERM;
}
inode = pft_bio_get_inode(bio);
if (!inode)
return -EINVAL;
if (!pft_is_tag_valid(inode)) {
pr_debug("file %s, Tag not valid\n", inode_to_filename(inode));
return -EINVAL;
}
if (!pft_is_xattr_supported(inode)) {
*is_encrypted = false;
*is_inplace = false;
*key_index = 0;
return 0;
}
tag = pft_get_inode_tag(inode);
*is_encrypted = pft_is_file_encrypted(tag);
*key_index = pft_get_inode_key_index(inode);
*is_inplace = pft_is_inplace_inode(inode);
if (*is_encrypted)
pr_debug("file %s is encrypted\n", inode_to_filename(inode));
return 0;
}
EXPORT_SYMBOL(pft_get_key_index);
/**
* pft_allow_merge_bio()- Check if 2 BIOs can be merged.
* @bio1: Pointer to first BIO structure.
* @bio2: Pointer to second BIO structure.
*
* Prevent merging of BIOs from encrypted and non-encrypted
* files, or files encrypted with different key.
* This API is called by the file system block layer.
*
* Return: true if the BIOs allowed to be merged, false
* otherwise.
*/
bool pft_allow_merge_bio(const struct bio *bio1, const struct bio *bio2)
{
u32 key_index1 = 0, key_index2 = 0;
bool is_encrypted1 = false, is_encrypted2 = false;
bool allow = false;
bool is_inplace = false; /* N.A. */
int ret;
if (!pft_is_ready())
return true;
if (!bio1 || !bio2)
return -EPERM;
/*
* Encrypted BIOs are created only when file encryption is enabled,
* which happens only when key is loaded.
*/
if (pft_dev->state != PFT_STATE_KEY_LOADED)
return true;
ret = pft_get_key_index(bio1, &key_index1,
&is_encrypted1, &is_inplace);
if (ret)
is_encrypted1 = false;
ret = pft_get_key_index(bio2, &key_index2,
&is_encrypted2, &is_inplace);
if (ret)
is_encrypted2 = false;
allow = ((is_encrypted1 == is_encrypted2) &&
(key_index1 == key_index2));
return allow;
}
EXPORT_SYMBOL(pft_allow_merge_bio);
/**
* pft_bio_get_inode() - get the inode from a bio.
* @bio: Pointer to BIO structure.
*
* Walk the bio struct links to get the inode.
* Please note, that in general bio may consist of several pages from
* several files, but in our case we always assume that all pages come
* from the same file, since our logic ensures it. That is why we only
* walk through the first page to look for inode.
*
* Return: pointer to the inode struct if successful, or NULL otherwise.
*
*/
static struct inode *pft_bio_get_inode(const struct bio *bio)
{
if (!bio)
return NULL;
/* check bio vec count > 0 before using the bio->bi_io_vec[] array */
if (!bio->bi_vcnt)
return NULL;
if (!bio->bi_io_vec)
return NULL;
if (!bio->bi_io_vec->bv_page)
return NULL;
if (PageAnon(bio->bi_io_vec->bv_page)) {
struct inode *inode;
/* Using direct-io (O_DIRECT) without page cache */
inode = dio_bio_get_inode((struct bio *)bio);
pr_debug("inode on direct-io, inode = 0x%p.\n", inode);
return inode;
}
if (!bio->bi_io_vec->bv_page->mapping)
return NULL;
if (!bio->bi_io_vec->bv_page->mapping->host)
return NULL;
return bio->bi_io_vec->bv_page->mapping->host;
}
/**
* pft_inode_create() - file creation callback.
* @dir: directory inode pointer
* @dentry: file dentry pointer
* @mode: flags
*
* This hook is called when file is created by VFS.
* This hook is called from the selinux driver.
* This hooks check file creation permission for enterprise
* applications.
* Call path:
* vfs_create()->security_inode_create()->selinux_inode_create()
*
* Return: 0 on successe, negative value on failure.
*/
int pft_inode_create(struct inode *dir, struct dentry *dentry, umode_t mode)
{
if (!dir || !dentry)
return 0;
if (!pft_is_ready())
return 0;
switch (pft_dev->state) {
case PFT_STATE_DEACTIVATED:
case PFT_STATE_KEY_LOADED:
break;
case PFT_STATE_KEY_REMOVED:
case PFT_STATE_DEACTIVATING:
case PFT_STATE_REMOVING_KEY:
/* At this state no new encrypted files can be created */
if (pft_is_current_process_registered()) {
pr_debug("key removed, registered uid %u is denied from creating new file %s\n",
__kuid_val(current_uid()), dentry->d_iname);
return -EACCES;
}
break;
default:
BUG(); /* State is set by "set state" command */
break;
}
return 0;
}
EXPORT_SYMBOL(pft_inode_create);
/**
* pft_inode_post_create() - file creation callback.
* @dir: directory inode pointer
* @dentry: file dentry pointer
* @mode: flags
*
* This hook is called when file is created by VFS.
* This hook is called from the selinux driver.
* This hooks tags new files as encrypted when created by
* enterprise applications.
* Call path:
* vfs_create()->security_inode_post_create()->selinux_inode_post_create()
*
* Return: 0 on successe, negative value on failure.
*/
int pft_inode_post_create(struct inode *dir, struct dentry *dentry,
umode_t mode)
{
int ret;
if (!dir || !dentry)
return 0;
if (!pft_is_ready())
return 0;
switch (pft_dev->state) {
case PFT_STATE_DEACTIVATED:
case PFT_STATE_KEY_REMOVED:
case PFT_STATE_DEACTIVATING:
case PFT_STATE_REMOVING_KEY:
break;
case PFT_STATE_KEY_LOADED:
/* Check whether the new file should be encrypted */
if (pft_is_current_process_registered()) {
u32 key_index = pft_get_app_key_index(current_uid());
ret = pft_tag_file(dentry, key_index);
if (ret == 0)
pr_debug("pid[%u] uid[%d] creating file %s\n",
current_pid(),
__kuid_val(current_uid()),
dentry->d_iname);
else {
pr_err("Failed to tag file %s by pid %d\n",
dentry->d_iname, current_pid());
return -EFAULT;
}
}
break;
default:
BUG(); /* State is set by "set state" command */
break;
}
return 0;
}
EXPORT_SYMBOL(pft_inode_post_create);
/**
* pft_inode_mknod() - mknode file hook (callback)
* @dir: directory inode pointer
* @dentry: file dentry pointer
* @mode: flags
* @dev:
*
* This hook checks encrypted file access permission by
* enterprise application.
* Call path:
* vfs_mknod()->security_inode_mknod()->selinux_inode_mknod()->pft_inode_mknod()
*
* Return: 0 on successful operation, negative value otherwise.
*/
int pft_inode_mknod(struct inode *dir, struct dentry *dentry,
umode_t mode, dev_t dev)
{
int rc;
/* Check if allowed to create new encrypted files */
rc = pft_inode_create(dir, dentry, mode);
return rc;
}
EXPORT_SYMBOL(pft_inode_mknod);
/**
* pft_inode_rename() - file rename hook.
* @inode: directory inode
* @dentry: file dentry
* @new_inode
* @new_dentry
*
* Block attempt to rename enterprise file.
*
* Return: 0 on allowed operation, negative value otherwise.
*/
int pft_inode_rename(struct inode *inode, struct dentry *dentry,
struct inode *new_inode, struct dentry *new_dentry)
{
if (!inode || !dentry || !new_inode || !new_dentry || !dentry->d_inode)
return 0;
if (!pft_is_ready())
return 0;
/* do nothing for non-encrypted files */
if (!pft_is_encrypted_file(dentry))
return 0;
pr_debug("attempt to rename encrypted file [%s]\n", dentry->d_iname);
if (pft_is_inplace_inode(dentry->d_inode)) {
pr_err("access to file %s by uid [%d] pid [%d] is blocked.\n",
inode_to_filename(inode), __kuid_val(current_uid()),
current_pid());
return -EACCES;
}
if (!pft_is_current_process_registered()) {
pr_err("app (uid %u pid %u) can't access file %s\n",
__kuid_val(current_uid()), current_pid(),
dentry->d_iname);
return -EACCES;
}
pr_debug("rename file %s\n", dentry->d_iname);
return 0;
}
EXPORT_SYMBOL(pft_inode_rename);
/**
* pft_file_open() - file open hook (callback).
* @filp: file pointer
* @cred: credentials pointer
*
* This hook is called when file is opened by VFS.
* It is called from the selinux driver.
* It checks enterprise file xattr when first opened.
* It adds encrypted file to the list of open files.
* Call path:
* do_filp_open()->security_dentry_open()->selinux_dentry_open()
*
* Return: 0 on successe, negative value on failure.
*/
int pft_file_open(struct file *filp, const struct cred *cred)
{
int ret;
if (!filp || !filp->f_path.dentry)
return 0;
if (!pft_is_ready())
return 0;
if (filp->f_flags & O_DIRECT)
pr_debug("file %s using O_DIRECT.\n", file_to_filename(filp));
/* do nothing for non-encrypted files */
if (!pft_is_encrypted_file(filp->f_path.dentry))
return 0;
/*
* Only PFM allowed to access in-place-encryption-file
* during in-place-encryption process
*/
if (pft_is_inplace_file(filp) && current_pid() != pft_dev->pfm_pid) {
pr_err("Access to file %s by uid %d pid %d is blocked.\n",
file_to_filename(filp),
__kuid_val(current_uid()),
current_pid());
return -EACCES;
}
switch (pft_dev->state) {
case PFT_STATE_DEACTIVATED:
case PFT_STATE_KEY_REMOVED:
case PFT_STATE_DEACTIVATING:
case PFT_STATE_REMOVING_KEY:
/* Block any access for encrypted files when key not loaded */
pr_debug("key not loaded. uid (%u) can not access file %s\n",
__kuid_val(current_uid()), file_to_filename(filp));
return -EACCES;
case PFT_STATE_KEY_LOADED:
/* Only registered apps may access encrypted files. */
if (!pft_is_current_process_registered()) {
pr_err("unregistered app (uid %u pid %u) is trying to access encrypted file %s\n",
__kuid_val(current_uid()), current_pid(),
file_to_filename(filp));
return -EACCES;
}
ret = pft_add_file(filp);
if (ret) {
pr_err("failed to add file %s to the list.\n",
file_to_filename(filp));
return -EFAULT;
}
break;
default:
BUG(); /* State is set by "set state" command */
break;
}
return 0;
}
EXPORT_SYMBOL(pft_file_open);
/**
* pft_file_permission() - check file access permission.
* @filp: file pointer
* @mask: flags
*
* This hook is called when file is read/write by VFS.
* This hook is called from the selinux driver.
* This hook checks encrypted file access permission by
* enterprise application.
* Call path:
* vfs_read()->security_file_permission()->selinux_file_permission()
*
* Return: 0 on success, negative value on failure.
*/
int pft_file_permission(struct file *filp, int mask)
{
if (!filp)
return 0;
if (!pft_is_ready())
return 0;
/* do nothing for non-encrypted files */
if (!pft_is_encrypted_file(filp->f_path.dentry))
return 0;
/*
* Only PFM allowed to access in-place-encryption-file
* during in-place encryption process
*/
if (pft_is_inplace_file(filp)) {
if (current_pid() == pft_dev->pfm_pid) {
/* mask MAY_WRITE=2 / MAY_READ=4 */
pr_debug("r/w [mask 0x%x] file %s (UID %d, PID %d).\n",
mask, file_to_filename(filp),
__kuid_val(current_uid()), current_pid());
return 0;
}
pr_err("Access to file %s by (UID %d, PID %d) is blocked.\n",
file_to_filename(filp),
__kuid_val(current_uid()), current_pid());
return -EACCES;
}
switch (pft_dev->state) {
case PFT_STATE_DEACTIVATED:
case PFT_STATE_KEY_REMOVED:
case PFT_STATE_DEACTIVATING:
case PFT_STATE_REMOVING_KEY:
/* Block any access for encrypted files when key not loaded */
pr_debug("key not loaded. uid (%u) can not access file %s\n",
__kuid_val(current_uid()), file_to_filename(filp));
return -EACCES;
case PFT_STATE_KEY_LOADED:
/* Only registered apps can access encrypted files. */
if (!pft_is_current_process_registered()) {
pr_err("unregistered app (uid %u pid %u) is trying to access encrypted file %s\n",
__kuid_val(current_uid()), current_pid(),
file_to_filename(filp));
return -EACCES;
}
break;
default:
BUG(); /* State is set by "set state" command */
break;
}
return 0;
}
EXPORT_SYMBOL(pft_file_permission);
/**
* pft_sync_file() - sync the file.
* @filp: file pointer
*
* Complete writing any pending write request of encrypted data
* before key is removed, to avoid writing garbage to
* enterprise files.
*/
static void pft_sync_file(struct file *filp)
{
int ret;
ret = vfs_fsync(filp, false);
if (ret)
pr_debug("failed to sync file %s, ret = %d.\n",
file_to_filename(filp), ret);
else
pr_debug("Sync file %s ok.\n", file_to_filename(filp));
}
/**
* pft_file_close()- handle file close event
* @filp: file pointer
*
* This hook is called when file is closed by VFS.
* This hook is called from the selinux driver.
*
* Return: 0 on successful operation, negative value otherwise.
*/
int pft_file_close(struct file *filp)
{
if (!filp)
return 0;
if (!pft_is_ready())
return 0;
/* do nothing for non-encrypted files */
if (!pft_is_encrypted_file(filp->f_path.dentry))
return 0;
if (pft_is_inplace_file(filp)) {
pr_debug("pid [%u] uid [%u] is closing in-place-encryption file %s\n",
current_pid(), __kuid_val(current_uid()),
file_to_filename(filp));
pft_dev->inplace_file = NULL;
}
switch (pft_dev->state) {
case PFT_STATE_DEACTIVATING:
case PFT_STATE_REMOVING_KEY:
/*
* Do not allow apps to close file when
* pft_close_opened_enc_files() is closing files.
* Normally, all enterprise apps are closed by PFM
* before getting to this state, so the apps files are
* norammly closed by now.
* pft_close_opened_enc_files() is running in PFM context.
*/
if (current_pid() != pft_dev->pfm_pid)
return -EACCES;
case PFT_STATE_DEACTIVATED:
case PFT_STATE_KEY_LOADED:
case PFT_STATE_KEY_REMOVED:
break;
default:
BUG(); /* State is set by "set state" command */
break;
}
pft_sync_file(filp);
pft_remove_file(filp);
return 0;
}
EXPORT_SYMBOL(pft_file_close);
/**
* pft_inode_unlink() - Delete file hook.
* @dir: directory inode pointer
* @dentry: file dentry pointer
*
* call path: vfs_unlink()->security_inode_unlink().
*
* Return: 0 on successful operation, negative value otherwise.
*/
int pft_inode_unlink(struct inode *dir, struct dentry *dentry)
{
struct inode *inode = NULL;
if (!dir || !dentry || !dentry->d_inode)
return 0;
if (!pft_is_ready())
return 0;
inode = dentry->d_inode;
/* do nothing for non-encrypted files */
if (!pft_is_encrypted_file(dentry))
return 0;
if (pft_is_inplace_inode(inode)) {
pr_err("can't delete file %s by uid [%d] pid [%d]\n",
inode_to_filename(inode),
__kuid_val(current_uid()),
current_pid());
return -EBUSY;
}
if (!pft_is_current_process_registered()) {
pr_err("app (uid %u pid %u) is trying to access file %s\n",
__kuid_val(current_uid()),
current_pid(),
inode_to_filename(inode));
return -EACCES;
}
pr_debug("delete file %s\n", inode_to_filename(inode));
return 0;
}
EXPORT_SYMBOL(pft_inode_unlink);
/**
* pft_inode_set_xattr() - set/remove xattr callback.
* @dentry: file dentry pointer
* @name: xattr name.
*
* This hook checks attempt to set/remove PFE xattr.
* Only this kernel driver allows to set the PFE xattr, so block
* any attempt to do it from user space. Allow access for other
* xattr.
*
* Return: 0 on successful operation, negative value otherwise.
*/
int pft_inode_set_xattr(struct dentry *dentry, const char *name,
const void *value, size_t size, int flags)
{
struct inode *inode = NULL;
if (!dentry || !dentry->d_inode)
return 0;
inode = dentry->d_inode;
if (strcmp(name, XATTR_NAME_PFE) != 0) {
pr_debug("xattr name=%s file %s\n", name,
inode_to_filename(inode));
return 0; /* Not PFE xattr so it is ok */
}
pr_err("Attemp to set/remove PFE xattr for file %s\n",
inode_to_filename(inode));
/* Only PFT kernel driver allows to set the PFE xattr */
return -EACCES;
}
EXPORT_SYMBOL(pft_inode_set_xattr);
/**
* pft_close_opened_enc_files() - Close all the currently open
* encrypted files
*
* Close all open encrypted file when removing key or
* deactivating.
*/
static void pft_close_opened_enc_files(void)
{
struct pft_file_info *tmp = NULL;
struct list_head *pos = NULL;
struct list_head *next = NULL;
list_for_each_safe(pos, next, &pft_dev->open_file_list) {
struct file *filp;
tmp = list_entry(pos, struct pft_file_info, list);
filp = tmp->file;
pr_debug("closing file %s.\n", file_to_filename(filp));
/* filp_close() eventually calls pft_file_close() */
filp_close(filp, NULL);
}
}
/**
* pft_set_state() - Handle "Set State" command
* @command: command buffer.
* @size: size of command buffer.
*
* The command execution status is reported by the response.
*
* Return: 0 on successful operation, negative value otherwise.
*/
static int pft_set_state(struct pft_command *command, int size)
{
u32 state = command->set_state.state;
int expected_size = sizeof(command->opcode) +
sizeof(command->set_state);
if (size != expected_size) {
pr_err("Invalid buffer size\n");
pft_set_response(PFT_CMD_RESP_INVALID_CMD_PARAMS);
return -EINVAL;
}
if (state >= PFT_STATE_MAX_INDEX) {
pr_err("Invalid state %d\n", command->set_state.state);
pft_set_response(PFT_CMD_RESP_INVALID_STATE);
return 0;
}
pr_debug("Set State %d [%s].\n", state, pft_state_name[state]);
switch (command->set_state.state) {
case PFT_STATE_DEACTIVATING:
case PFT_STATE_REMOVING_KEY:
pft_close_opened_enc_files();
/* Fall through */
case PFT_STATE_DEACTIVATED:
case PFT_STATE_KEY_LOADED:
case PFT_STATE_KEY_REMOVED:
pft_dev->state = command->set_state.state;
pft_set_response(PFT_CMD_RESP_SUCCESS);
break;
default:
pr_err("Invalid state %d\n", command->set_state.state);
pft_set_response(PFT_CMD_RESP_INVALID_STATE);
break;
}
return 0;
}
/**
* pft_get_process_open_file() - get file pointer using file
* descriptor index.
* @index: file descriptor index.
*
* Return: file pointer on success, NULL on failure.
*/
static struct file *pft_get_process_open_file(int index)
{
struct fdtable *files_table;
files_table = files_fdtable(current->files);
if (files_table == NULL)
return NULL;
if (index >= files_table->max_fds)
return NULL;
else
return files_table->fd[index];
}
/**
* pft_set_inplace_file() - handle "inplace file encryption"
* command.
* @command: command buffer.
* @size: size of command buffer.
*
* The command execution status is reported by the response.
*
* Return: 0 if command is valid, negative value otherwise.
*/
static int pft_set_inplace_file(struct pft_command *command, int size)
{
int expected_size;
u32 fd;
int rc;
struct file *filp = NULL;
struct inode *inode = NULL;
int writecount;
expected_size = sizeof(command->opcode) +
sizeof(command->preform_in_place_file_enc.file_descriptor);
if (size != expected_size) {
pr_err("invalid command size %d expected %d.\n",
size, expected_size);
pft_set_response(PFT_CMD_RESP_INVALID_CMD_PARAMS);
return -EINVAL;
}
if (pft_dev->state != (u32) PFT_STATE_KEY_LOADED) {
pr_err("Key not loaded, state [%d], In-place-encryption is not allowed.\n",
pft_dev->state);
pft_set_response(PFT_CMD_RESP_GENERAL_ERROR);
return 0;
}
/* allow only one in-place file encryption at a time */
if (pft_dev->inplace_file != NULL) {
pr_err("file %s in-place-encryption in progress.\n",
file_to_filename(pft_dev->inplace_file));
/* @todo - use new error code */
pft_set_response(PFT_CMD_RESP_INPLACE_FILE_IS_OPEN);
return 0;
}
fd = command->preform_in_place_file_enc.file_descriptor;
filp = pft_get_process_open_file(fd);
if (filp == NULL) {
pr_err("failed to find file by fd %d.\n", fd);
pft_set_response(PFT_CMD_RESP_GENERAL_ERROR);
return 0;
}
/* Verify the file is not already open by other than PFM */
if (!filp->f_path.dentry || !filp->f_path.dentry->d_inode) {
pr_err("failed to get inode of inplace-file.\n");
pft_set_response(PFT_CMD_RESP_GENERAL_ERROR);
return 0;
}
inode = filp->f_path.dentry->d_inode;
writecount = atomic_read(&inode->i_writecount);
if (writecount > 1) {
pr_err("file %s is opened %d times for write.\n",
file_to_filename(filp), writecount);
pft_set_response(PFT_CMD_RESP_INPLACE_FILE_IS_OPEN);
return 0;
}
/*
* Check if the file was already encryprted.
* In practice, it is unlikely to happen,
* because PFM is not an enterprise application
* it won't be able to open encrypted file.
*/
if (pft_is_encrypted_file(filp->f_path.dentry)) {
pr_err("file %s is already encrypted.\n",
file_to_filename(filp));
pft_set_response(PFT_CMD_RESP_GENERAL_ERROR);
return 0;
}
/* Update the current in-place-encryption file */
pft_dev->inplace_file = filp;
/*
* Now, any new access to this file is allowed only to PFM.
* Lets make sure that all pending writes are completed
* before encrypting the file.
*/
pft_sync_file(filp);
rc = pft_tag_file(pft_dev->inplace_file->f_path.dentry,
pft_get_app_key_index(current_uid()));
if (!rc) {
pr_debug("tagged file %s to be encrypted.\n",
file_to_filename(pft_dev->inplace_file));
pft_set_response(PFT_CMD_RESP_SUCCESS);
} else {
pr_err("failed to tag file %s for encryption.\n",
file_to_filename(pft_dev->inplace_file));
pft_set_response(PFT_CMD_RESP_GENERAL_ERROR);
}
return 0;
}
/**
* pft_update_reg_apps() - Update the registered application
* list.
* @command: command buffer.
* @size: size of command buffer.
*
* The command execution status is reported by the response.
*
* Return: 0 on successful operation, negative value otherwise.
*/
static int pft_update_reg_apps(struct pft_command *command, int size)
{
int i;
int expected_size;
void *buf;
int buf_size;
u32 items_count = command->update_app_list.items_count;
if (items_count > PFT_MAX_APPS) {
pr_err("Number of apps [%d] > max apps [%d]\n",
items_count , PFT_MAX_APPS);
pft_set_response(PFT_CMD_RESP_INVALID_CMD_PARAMS);
return -EINVAL;
}
expected_size =
sizeof(command->opcode) +
sizeof(command->update_app_list.items_count) +
(command->update_app_list.items_count * sizeof(u32));
if (size != expected_size) {
pr_err("invalid command size %d expected %d.\n",
size, expected_size);
pft_set_response(PFT_CMD_RESP_INVALID_CMD_PARAMS);
return -EINVAL;
}
mutex_lock(&pft_dev->lock);
/* Free old table */
kfree(pft_dev->uid_table);
pft_dev->uid_table = NULL;
pft_dev->uid_count = 0;
if (items_count == 0) {
pr_info("empty app list - clear list.\n");
mutex_unlock(&pft_dev->lock);
return 0;
}
buf_size = command->update_app_list.items_count * sizeof(u32);
buf = kzalloc(buf_size, GFP_KERNEL);
if (!buf) {
pft_set_response(PFT_CMD_RESP_GENERAL_ERROR);
mutex_unlock(&pft_dev->lock);
return 0;
}
pft_dev->uid_table = buf;
pft_dev->uid_count = command->update_app_list.items_count;
pr_debug("uid_count = %d\n", pft_dev->uid_count);
for (i = 0; i < pft_dev->uid_count; i++)
pft_dev->uid_table[i] =
KUIDT_INIT(command->update_app_list.table[i]);
pft_set_response(PFT_CMD_RESP_SUCCESS);
mutex_unlock(&pft_dev->lock);
return 0;
}
/**
* pft_handle_command() - Handle user space app commands.
* @buf: command buffer.
* @buf_size: command buffer size.
*
* Return: 0 on successful operation, negative value otherwise.
*/
static int pft_handle_command(void *buf, int buf_size)
{
int ret = 0;
struct pft_command *command = NULL;
/* opcode field is the minimum length of command */
if (buf_size < sizeof(command->opcode)) {
pr_err("Invalid argument used buffer size\n");
return -EINVAL;
}
command = (struct pft_command *)buf;
pft_dev->response.command_id = command->opcode;
switch (command->opcode) {
case PFT_CMD_OPCODE_SET_STATE:
ret = pft_set_state(command, buf_size);
break;
case PFT_CMD_OPCODE_UPDATE_REG_APP_UID:
ret = pft_update_reg_apps(command, buf_size);
break;
case PFT_CMD_OPCODE_PERFORM_IN_PLACE_FILE_ENC:
ret = pft_set_inplace_file(command, buf_size);
break;
default:
pr_err("Invalid command_op_code %u\n", command->opcode);
pft_set_response(PFT_CMD_RESP_INVALID_COMMAND);
return 0;
}
return ret;
}
static int pft_device_open(struct inode *inode, struct file *file)
{
int ret;
mutex_lock(&pft_dev->lock);
if (pft_dev->open_count > 0) {
pr_err("PFT device is already opened (%d)\n",
pft_dev->open_count);
ret = -EBUSY;
} else {
pft_dev->open_count++;
pft_dev->pfm_pid = current_pid();
pr_debug("PFT device opened by %d (%d)\n",
pft_dev->pfm_pid, pft_dev->open_count);
ret = 0;
}
mutex_unlock(&pft_dev->lock);
pr_debug("device opened, count %d\n", pft_dev->open_count);
return ret;
}
static int pft_device_release(struct inode *inode, struct file *file)
{
mutex_lock(&pft_dev->lock);
if (pft_dev->open_count > 0)
pft_dev->open_count--;
pft_dev->pfm_pid = UINT_MAX;
mutex_unlock(&pft_dev->lock);
pr_debug("device released, count %d\n", pft_dev->open_count);
return 0;
}
/**
* pft_device_write() - Get commands from user sapce.
*
* Return: number of bytes to write on success to get the
* command buffer, negative value on failure.
* The error code for handling the command should be retrieve by
* reading the response.
* Note: any reurn value of 0..size-1 will cause retry by the
* OS, so avoid it.
*/
static ssize_t pft_device_write(struct file *filp, const char __user *user_buff,
size_t size, loff_t *f_pos)
{
int ret;
char *cmd_buf;
if (size > PFT_MAX_COMMAND_SIZE || !user_buff || !f_pos) {
pr_err("inavlid parameters.\n");
return -EINVAL;
}
cmd_buf = kzalloc(size, GFP_KERNEL);
if (cmd_buf == NULL)
return -ENOMEM;
ret = copy_from_user(cmd_buf, user_buff, size);
if (ret) {
pr_err("Unable to copy from user (err %d)\n", ret);
kfree(cmd_buf);
return -EFAULT;
}
ret = pft_handle_command(cmd_buf, size);
if (ret) {
kfree(cmd_buf);
return -EFAULT;
}
kfree(cmd_buf);
return size;
}
/**
* pft_device_read() - return response of last command.
*
* Return: number of bytes to read on success, negative value on
* failure.
*/
static ssize_t pft_device_read(struct file *filp, char __user *buffer,
size_t length, loff_t *f_pos)
{
int ret = 0;
if (!buffer || !f_pos || length < sizeof(pft_dev->response)) {
pr_err("inavlid parameters.\n");
return -EFAULT;
}
ret = copy_to_user(buffer, &(pft_dev->response),
sizeof(pft_dev->response));
if (ret) {
pr_err("Unable to copy to user, err = %d.\n", ret);
return -EINVAL;
}
return sizeof(pft_dev->response);
}
static const struct file_operations fops = {
.owner = THIS_MODULE,
.read = pft_device_read,
.write = pft_device_write,
.open = pft_device_open,
.release = pft_device_release,
};
static int __init pft_register_chardev(void)
{
int rc;
unsigned baseminor = 0;
unsigned count = 1;
struct device *class_dev;
rc = alloc_chrdev_region(&pft_dev->device_no, baseminor, count,
DEVICE_NAME);
if (rc < 0) {
pr_err("alloc_chrdev_region failed %d\n", rc);
return rc;
}
pft_dev->driver_class = class_create(THIS_MODULE, DEVICE_NAME);
if (IS_ERR(pft_dev->driver_class)) {
rc = -ENOMEM;
pr_err("class_create failed %d\n", rc);
goto exit_unreg_chrdev_region;
}
class_dev = device_create(pft_dev->driver_class, NULL,
pft_dev->device_no, NULL,
DEVICE_NAME);
if (!class_dev) {
pr_err("class_device_create failed %d\n", rc);
rc = -ENOMEM;
goto exit_destroy_class;
}
cdev_init(&pft_dev->cdev, &fops);
pft_dev->cdev.owner = THIS_MODULE;
rc = cdev_add(&pft_dev->cdev, MKDEV(MAJOR(pft_dev->device_no), 0), 1);
if (rc < 0) {
pr_err("cdev_add failed %d\n", rc);
goto exit_destroy_device;
}
return 0;
exit_destroy_device:
device_destroy(pft_dev->driver_class, pft_dev->device_no);
exit_destroy_class:
class_destroy(pft_dev->driver_class);
exit_unreg_chrdev_region:
unregister_chrdev_region(pft_dev->device_no, 1);
return rc;
}
static void __exit pft_unregister_chrdev(void)
{
cdev_del(&pft_dev->cdev);
device_destroy(pft_dev->driver_class, pft_dev->device_no);
class_destroy(pft_dev->driver_class);
unregister_chrdev_region(pft_dev->device_no, 1);
}
static void __exit pft_free_open_files_list(void)
{
struct pft_file_info *tmp = NULL;
struct list_head *pos = NULL;
struct list_head *next = NULL;
mutex_lock(&pft_dev->lock);
list_for_each_safe(pos, next, &pft_dev->open_file_list) {
tmp = list_entry(pos, struct pft_file_info, list);
list_del(&tmp->list);
kfree(tmp);
}
mutex_unlock(&pft_dev->lock);
}
static void __exit pft_exit(void)
{
if (pft_dev == NULL)
return;
pft_unregister_chrdev();
pft_free_open_files_list();
kfree(pft_dev->uid_table);
kfree(pft_dev);
pft_dev = NULL;
}
static int __init pft_init(void)
{
int ret;
struct pft_device *dev = NULL;
dev = kzalloc(sizeof(struct pft_device), GFP_KERNEL);
if (dev == NULL)
return -ENOMEM;
pft_dev = dev;
dev->state = PFT_STATE_DEACTIVATED;
dev->pfm_pid = UINT_MAX;
INIT_LIST_HEAD(&dev->open_file_list);
mutex_init(&dev->lock);
ret = pft_register_chardev();
if (ret) {
pr_err("create character device failed.\n");
goto fail;
}
pft_lsm_init(dev);
return 0;
fail:
pr_err("Failed to init driver.\n");
kfree(dev);
pft_dev = NULL;
return -ENODEV;
}
module_init(pft_init);
module_exit(pft_exit);
MODULE_LICENSE("GPL v2");
MODULE_DESCRIPTION("Per-File-Tagger driver");