new file mode 100644
@@ -0,0 +1,598 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/*
+ * Copyright (C) 2022 Enjellic Systems Development, LLC
+ * Author: Dr. Greg Wettstein <greg@enjellic.com>
+ *
+ * Implements the an kernel modeling agent.
+ */
+
+#include <linux/list_sort.h>
+#include <crypto/hash.h>
+
+#include "tsem.h"
+
+const u8 pseudonym_digest[WP256_DIGEST_SIZE] = {
+ 0xe3, 0xb0, 0xc4, 0x42, 0x98, 0xfc, 0x1c, 0x14,
+ 0x9a, 0xfb, 0xf4, 0xc8, 0x99, 0x6f, 0xb9, 0x24,
+ 0x27, 0xae, 0x41, 0xe4, 0x64, 0x9b, 0x93, 0x4c,
+ 0xa4, 0x95, 0x99, 0x1b, 0x78, 0x52, 0xb8, 0x55
+};
+
+struct state_point {
+ struct list_head list;
+ struct tsem_event_point *point;
+};
+
+struct pseudonym {
+ struct list_head list;
+ u8 mapping[WP256_DIGEST_SIZE];
+};
+
+static int generate_pseudonym(struct crypto_shash *tfm, struct tsem_file *ep,
+ u8 *pseudonym)
+{
+ int retn = 0;
+ SHASH_DESC_ON_STACK(shash, tfm);
+
+ shash->tfm = tfm;
+
+ retn = crypto_shash_init(shash);
+ if (retn)
+ goto done;
+ retn = crypto_shash_update(shash, (u8 *) &ep->name_length,
+ sizeof(ep->name_length));
+ if (retn)
+ goto done;
+ retn = crypto_shash_finup(shash, ep->name, WP256_DIGEST_SIZE,
+ pseudonym);
+ done:
+ return retn;
+}
+
+static int have_point(u8 *point)
+{
+ int retn = 0;
+ struct tsem_event_point *entry;
+ struct tsem_model *model = tsem_model(current);
+
+ mutex_lock(&model->point_mutex);
+ list_for_each_entry(entry, &model->point_list, list) {
+ if (memcmp(entry->point, point, WP256_DIGEST_SIZE) == 0) {
+ if (entry->valid)
+ retn = 1;
+ else
+ retn = -EPERM;
+ goto done;
+ }
+ }
+
+ done:
+ mutex_unlock(&model->point_mutex);
+ return retn;
+}
+
+static int add_event_point(u8 *point, bool valid)
+{
+ int retn = 1;
+ struct tsem_event_point *entry;
+ struct state_point *state;
+ struct tsem_model *model = tsem_model(current);
+
+ entry = kzalloc(sizeof(*entry), GFP_KERNEL);
+ if (!entry)
+ goto done;
+
+ state = kzalloc(sizeof(*state), GFP_KERNEL);
+ if (!state)
+ goto done;
+ state->point = entry;
+
+ mutex_lock(&model->point_mutex);
+ memcpy(entry->point, point, WP256_DIGEST_SIZE);
+ entry->valid = valid;
+ list_add_tail(&entry->list, &model->point_list);
+ list_add_tail(&state->list, &model->state_list);
+ ++model->point_count;
+ mutex_unlock(&model->point_mutex);
+ retn = 0;
+
+ done:
+ return retn;
+}
+
+static int add_trajectory_point(struct tsem_event *ep)
+{
+ struct tsem_trajectory *entry;
+ struct tsem_model *model = tsem_model(current);
+
+ entry = kzalloc(sizeof(struct tsem_trajectory), GFP_KERNEL);
+ if (!entry)
+ return -ENOMEM;
+ entry->ep = ep;
+ tsem_event_get(ep);
+
+ mutex_lock(&model->trajectory_mutex);
+ list_add_tail(&entry->list, &model->trajectory_list);
+ ++model->trajectory_count;
+ mutex_unlock(&model->trajectory_mutex);
+
+ return 0;
+}
+
+static int add_forensic_point(struct tsem_event *ep)
+{
+ struct tsem_trajectory *entry;
+ struct tsem_model *model = tsem_model(current);
+
+ if (model->forensics_count == model->max_forensics_count)
+ return -E2BIG;
+
+ entry = kzalloc(sizeof(struct tsem_trajectory), GFP_KERNEL);
+ if (!entry)
+ return -ENOMEM;
+ entry->ep = ep;
+ tsem_event_get(ep);
+
+ mutex_lock(&model->forensics_mutex);
+ list_add_tail(&entry->list, &model->forensics_list);
+ ++model->forensics_count;
+ mutex_unlock(&model->forensics_mutex);
+
+ return 0;
+}
+
+static int get_host_measurement(struct crypto_shash *tfm, u8 *id,
+ size_t idlength, u8 *digest)
+{
+ int retn;
+ SHASH_DESC_ON_STACK(shash, tfm);
+ struct tsem_model *model = tsem_model(current);
+
+ shash->tfm = tfm;
+ retn = crypto_shash_init(shash);
+ if (retn)
+ goto done;
+ retn = crypto_shash_update(shash, model->base, WP256_DIGEST_SIZE);
+ if (retn)
+ goto done;
+ retn = crypto_shash_finup(shash, id, idlength, digest);
+
+ done:
+ return retn;
+}
+
+static int update_events_measurement(struct crypto_shash *tfm, u8 *id)
+{
+ int retn;
+ u8 digest[WP256_DIGEST_SIZE];
+ struct tsem_model *model = tsem_model(current);
+ SHASH_DESC_ON_STACK(shash, tfm);
+
+ retn = get_host_measurement(tfm, id, WP256_DIGEST_SIZE, digest);
+ if (retn)
+ goto done;
+
+ shash->tfm = tfm;
+ retn = crypto_shash_init(shash);
+ if (retn)
+ goto done;
+
+ retn = crypto_shash_update(shash, model->measurement,
+ WP256_DIGEST_SIZE);
+ if (retn)
+ goto done;
+
+ retn = crypto_shash_finup(shash, digest, sizeof(digest),
+ model->measurement);
+ if (retn)
+ goto done;
+
+ if (!tsem_context(current)->id)
+ retn = tsem_trust_add_event(digest);
+
+ done:
+ return retn;
+}
+
+static int state_sort(void *priv, const struct list_head *a,
+ const struct list_head *b)
+{
+ unsigned int lp, retn;
+ struct state_point *ap = container_of(a, struct state_point, list);
+ struct state_point *bp = container_of(b, struct state_point, list);
+
+ for (lp = 0; lp < WP256_DIGEST_SIZE - 1; ++lp) {
+ if (ap->point->point[lp] == bp->point->point[lp])
+ continue;
+ retn = ap->point->point[lp] > bp->point->point[lp];
+ goto done;
+ }
+ retn = ap->point->point[lp] > bp->point->point[lp];
+
+ done:
+ return retn;
+}
+
+/**
+ * tesm_model_compute_state() - Calculate a security model state value.
+ *
+ * This value is used to trigger the computation of the security
+ * state description value for a modeling domain.
+ */
+void tsem_model_compute_state(void)
+{
+ u8 state[WP256_DIGEST_SIZE];
+ struct state_point *entry;
+ struct tsem_model *model = tsem_model(current);
+ struct crypto_shash *sha256 = NULL;
+ SHASH_DESC_ON_STACK(shash, tfm);
+
+ sha256 = crypto_alloc_shash("sha256", 0, 0);
+ if (IS_ERR(sha256))
+ return;
+
+ shash->tfm = sha256;
+ if (crypto_shash_init(shash))
+ goto done;
+
+ memset(state, '\0', sizeof(state));
+ if (crypto_shash_update(shash, state, WP256_DIGEST_SIZE))
+ goto done;
+
+ if (get_host_measurement(sha256, tsem_trust_aggregate(),
+ WP256_DIGEST_SIZE, state))
+ goto done;
+
+ if (crypto_shash_finup(shash, state, sizeof(state), state))
+ goto done;
+
+ mutex_lock(&model->point_mutex);
+ list_sort(NULL, &model->state_list, state_sort);
+
+ memcpy(model->state, state, sizeof(model->state));
+ list_for_each_entry(entry, &model->state_list, list) {
+ if (get_host_measurement(sha256, entry->point->point,
+ WP256_DIGEST_SIZE, state))
+ goto unlock_done;
+
+ if (crypto_shash_init(shash))
+ goto unlock_done;
+ if (crypto_shash_update(shash, model->state,
+ WP256_DIGEST_SIZE))
+ goto unlock_done;
+ if (crypto_shash_finup(shash, state, WP256_DIGEST_SIZE,
+ model->state))
+ goto unlock_done;
+ }
+
+ unlock_done:
+ mutex_unlock(&model->point_mutex);
+ done:
+ if (sha256)
+ crypto_free_shash(sha256);
+}
+
+/**
+ * tsem_model_has_pseudonym() - Test for a model pseudonym.
+ * @tsip: A pointer to the TSEM inode security structure.
+ * @ep: A pointer to the TSEM event description structure.
+ * @mapping: A byte array into which the pseudonym state is
+ * to be copied.
+ *
+ * This function is used to test whether a pseudonym has been
+ * declared for a modeling domain. The pseudonym state point is
+ * made available via the array pointed to by the mapping variable.
+ *
+ * Return: If an error occurs during the pseudonym probe a negative
+ * return value is returned. A zero return value indicates that
+ * a pseudonym was not present. A value of one indicates that a
+ * psuedonym had been defined and the mapping variable contains
+ * the pseudonym state point.
+ */
+int tsem_model_has_pseudonym(struct tsem_inode *tsip, struct tsem_file *ep,
+ u8 *mapping)
+{
+ int retn = 0;
+ u8 pseudo_mapping[WP256_DIGEST_SIZE];
+ struct tsem_model *model = tsem_model(current);
+ struct crypto_shash *tfm;
+ struct pseudonym *entry;
+
+ tfm = crypto_alloc_shash("sha256", 0, 0);
+ if (IS_ERR(tfm)) {
+ retn = PTR_ERR(tfm);
+ goto done;
+ }
+ retn = generate_pseudonym(tfm, ep, pseudo_mapping);
+ if (retn)
+ goto done;
+
+ mutex_lock(&model->pseudonym_mutex);
+ list_for_each_entry(entry, &model->pseudonym_list, list) {
+ if (!memcmp(entry->mapping, pseudo_mapping,
+ sizeof(entry->mapping))) {
+ retn = 1;
+ goto done;
+ }
+ }
+ retn = 0;
+
+ done:
+ if (retn)
+ memcpy(mapping, pseudonym_digest, WP256_DIGEST_SIZE);
+
+ mutex_unlock(&model->pseudonym_mutex);
+ crypto_free_shash(tfm);
+ return retn;
+}
+
+/**
+ * tesm_model_event() - Inject a security event into a modeling domain.
+ * @ep: A pointer to the event description structure.
+ *
+ * This function is the entry point for the in kernel Trusted Modeling
+ * Agent (TMA). It takes a description of an event encoded in a
+ * tsem_event structure and generates and updates the security model
+ * description.
+ *
+ * Return: If an error occurs during the injection of an event into a
+ * model a negative error value is returned. A value of zero
+ * is returned if the event was successfully modeled. The
+ * security status of the event is returned by encoding the value
+ * in the bad_COE member of the tsem_task structure.
+ */
+int tsem_model_event(struct tsem_event *ep)
+{
+ int retn;
+ struct crypto_shash *sha256 = NULL;
+ struct tsem_task *task = tsem_task(current);
+ struct tsem_TMA_context *ctx = task->context;
+
+ sha256 = crypto_alloc_shash("sha256", 0, 0);
+ if (IS_ERR(sha256))
+ return PTR_ERR(sha256);
+
+ retn = have_point(ep->mapping);
+ if (retn) {
+ if (retn != 1)
+ task->trust_status = TSEM_TASK_UNTRUSTED;
+ return 0;
+ }
+
+ retn = update_events_measurement(sha256, ep->mapping);
+ if (retn)
+ goto done;
+
+ if (ctx->sealed) {
+ retn = add_event_point(ep->mapping, false);
+ if (!retn)
+ retn = add_forensic_point(ep);
+ task->trust_status = TSEM_TASK_UNTRUSTED;
+ } else {
+ retn = add_event_point(ep->mapping, true);
+ if (!retn)
+ retn = add_trajectory_point(ep);
+ }
+ if (retn)
+ retn = -EPERM;
+
+ done:
+ crypto_free_shash(sha256);
+ return retn;
+}
+
+/**
+ * tesm_model_load_point() - Load a security state event into a model.
+ * @point: A pointer to the array containing the security state
+ * point to be added to the model.
+ *
+ * This function takes the binary representation of a security state
+ * point and loads it into the current model domain.
+ *
+ * Return: If an error occurs during the processing of the security state
+ * point a negative return value is returned. A return value of
+ * zero indicates the point was successfully loaded into the domain.
+ */
+int tsem_model_load_point(u8 *point)
+{
+ ssize_t retn = 0;
+ struct crypto_shash *tfmsha256 = NULL;
+ struct tsem_TMA_context *ctx = tsem_context(current);
+
+ if (have_point(point))
+ goto done;
+ if (add_event_point(point, true)) {
+ retn = -ENOMEM;
+ goto done;
+ }
+
+ tfmsha256 = crypto_alloc_shash("sha256", 0, 0);
+ if (IS_ERR(tfmsha256)) {
+ retn = PTR_ERR(tfmsha256);
+ goto done;
+ }
+
+ if (!ctx->model->have_aggregate) {
+ ctx->model->have_aggregate = true;
+ retn = update_events_measurement(tfmsha256,
+ tsem_trust_aggregate());
+ if (retn)
+ goto done;
+ }
+
+ retn = update_events_measurement(tfmsha256, point);
+
+done:
+ if (tfmsha256)
+ crypto_free_shash(tfmsha256);
+ return retn;
+
+}
+
+/**
+ * tesm_model_load_pseudonym() - Load a pseudonym state point to a model.
+ * @mapping: A pointer to the array containing the pseudonym state
+ * point that is to be added to the model.
+ *
+ * This function takes the binary representation of a file pseudonym
+ * and declares the presence of the pseudonym in the modeling domain.
+ *
+ * Return: If an error occurs during the processing of the pseudonym
+ * state point a negative return value is returned. A return
+ * value of zero indicates the point was successfully loaded
+ * into the model.
+ */
+int tsem_model_load_pseudonym(u8 *mapping)
+{
+ struct pseudonym *psp = NULL;
+ struct tsem_model *model = tsem_model(current);
+
+ psp = kzalloc(sizeof(struct pseudonym), GFP_KERNEL);
+ if (!psp)
+ return -ENOMEM;
+ memcpy(psp->mapping, mapping, sizeof(psp->mapping));
+
+ mutex_lock(&model->pseudonym_mutex);
+ list_add_tail(&psp->list, &model->pseudonym_list);
+ mutex_unlock(&model->pseudonym_mutex);
+ return 0;
+}
+
+/**
+ * tesm_model_load_base() - Load a model base point.
+ * @mapping: A pointer to the array containing the base point to be
+ * set for the model.
+ *
+ * This function takes the binary representation of a base point and
+ * sets this point as the base point for the model.
+ */
+void tsem_model_load_base(u8 *mapping)
+{
+ struct tsem_model *model = tsem_model(current);
+
+ memcpy(model->base, mapping, sizeof(model->base));
+}
+
+/**
+ * tesm_model_add_aggregate() - Add the hardware aggregate to a model.
+ *
+ * This function adds the hardware aggregate value to an internally
+ * modeled security domain.
+ *
+ * Return: If an error occurs during the injection of the aggregate
+ * value into the model a negative error value is returned.
+ * A return value of zero indicates the aggregate was
+ * successfully added.
+ */
+int tsem_model_add_aggregate(void)
+{
+ int retn;
+ struct crypto_shash *tfm = NULL;
+
+ tfm = crypto_alloc_shash("sha256", 0, 0);
+ if (IS_ERR(tfm))
+ return PTR_ERR(tfm);
+
+ retn = update_events_measurement(tfm, tsem_trust_aggregate());
+ crypto_free_shash(tfm);
+ return retn;
+}
+
+/**
+ * tsem_model_allocate() - Allocates a kernel TMA modeling structure.
+ *
+ * This function allocates and initializes a tsem_model structure
+ * that is used to hold modeling information for an in kernel
+ * modeling domain.
+ *
+ * Return: On success a pointer to the model description structure is
+ * returned. If an error occurs an error return value is
+ * encoded in the returned pointer.
+ */
+struct tsem_model *tsem_model_allocate(void)
+{
+ struct tsem_model *model = NULL;
+
+ model = kzalloc(sizeof(struct tsem_model), GFP_KERNEL);
+ if (!model)
+ return NULL;
+
+ mutex_init(&model->point_mutex);
+ INIT_LIST_HEAD(&model->point_list);
+ INIT_LIST_HEAD(&model->state_list);
+
+ mutex_init(&model->trajectory_mutex);
+ INIT_LIST_HEAD(&model->trajectory_list);
+
+ model->max_forensics_count = 100;
+ mutex_init(&model->forensics_mutex);
+ INIT_LIST_HEAD(&model->forensics_list);
+
+ mutex_init(&model->pseudonym_mutex);
+ INIT_LIST_HEAD(&model->pseudonym_list);
+
+ return model;
+}
+
+/**
+ * tsem_model_free() - Frees an a kernel TMA description structure.
+ * @ctx: A pointer to the TMA modeling description structure whose
+ * model definition is to be deleted.
+ *
+ * This function is called when the last reference to a kernel
+ * based TMA model description structure is released.
+ */
+void tsem_model_free(struct tsem_TMA_context *ctx)
+{
+ unsigned int cnt;
+ struct tsem_event_point *centry, *tmp_centry;
+ struct state_point *state, *tmp_state;
+ struct tsem_trajectory *tentry, *tmp_tentry;
+ struct pseudonym *sentry, *tmp_sentry;
+ struct tsem_model *model = ctx->model;
+
+ cnt = 0;
+ list_for_each_entry_safe(centry, tmp_centry, &model->point_list,
+ list) {
+ list_del(¢ry->list);
+ kfree(centry);
+ ++cnt;
+ }
+
+ cnt = 0;
+ list_for_each_entry_safe(state, tmp_state, &model->state_list,
+ list) {
+ list_del(&state->list);
+ kfree(state);
+ ++cnt;
+ }
+
+ cnt = 0;
+ list_for_each_entry_safe(tentry, tmp_tentry, &model->trajectory_list,
+ list) {
+ list_del(&tentry->list);
+ tsem_event_put(tentry->ep);
+ ++cnt;
+ }
+
+ cnt = 0;
+ list_for_each_entry_safe(sentry, tmp_sentry, &model->pseudonym_list,
+ list) {
+ list_del(&sentry->list);
+ kfree(sentry);
+ ++cnt;
+ }
+
+ if (ctx->sealed) {
+ cnt = 0;
+ list_for_each_entry_safe(tentry, tmp_tentry,
+ &model->forensics_list, list) {
+ list_del(&tentry->list);
+ tsem_event_put(tentry->ep);
+ ++cnt;
+ }
+ }
+
+ kfree(model);
+}
A Trusted Modeling Agent (TMA) is an implementation of a modeling algorithm that converts a description of a security event (LSM hook) into a security state point. The sum of these state points is considered to be the functional value of the security model implemented for the modeling domain. The current model implemented is the simple deterministic model used by the Quixote trust orchestrators. These orchestrators represent an initial implementation of the support infrastructure needed to use the TSEM modeling LSM. It is anticipated that other in-kernel modeling implementations will be developed. The TMA takes the mapping of a security event state point, as generated by the tsem_map_event() function, and determines whether or not this point is a valid coefficient in the implemented model. If it is a valid point the process executing the security event is designated as being trusted, otherwise the state of the process is set to be untrusted. By default the TMA runs in free modeling mode where all security events are considered valid. In this mode, the security state point is registered as a valid point and the description of the event is added to the security execution trajectory for the model. The model implementation can be 'sealed' through the TSEM control plane, a condition that causes any state points not registered in the model to be considered a 'forensics' event. The description of such an event is added to the forensics execution trajectory for the model. A forensics event does not result in permission to the event to be denied unless the model is placed in 'enforcing' mode. While the current in-kernel TMA is largely deterministic, one approximation method is provided by this model, which is the notion of a file digest 'pseudonym'. A pseudonym can be declared for an inode by registering the value of the following function with the model: Pseudonym = SHA256(PATH_LENGTH || PATHNAME) If a file pseudonym is detected the file digest value used for the CELL definition is set to a default value that is the SHA256 value of a zero length file. The model.c file implementing the in kernel modeling agent defines this value. The pseudonym value is model specific. A separate modeling domain, with a pseudonym definition, will use the actual computed digest of the file. The TMA implementation also supports the definition of 'base' point that is a 32 byte nonce that can be used to extend each security event state point before it is added to the model. This allows a verifying partner to specify a random value that will allow a verifying partner to confirm the 'freshness' of an attestation of the function state of the model. The modeling implementation supports two different functional values for the model being implemented. The classic linear extension sum of all the security state points and a value referred to as the 'state' of the security model. The 'state' value is designed to make the measurement value invariant to scheduling variations that cause the classic trusted system measurement to be non-deterministic between runs of a workload. The state value is computed by sorting the security event state points in the model in big-endian (natural hash byte order) format and then computing the extension sum over this sorted vector of points. Signed-off-by: Greg Wettstein <greg@enjellic.com> --- security/tsem/model.c | 598 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 598 insertions(+) create mode 100644 security/tsem/model.c