1
0
mirror of https://github.com/opencryptoki/opencryptoki.git synced 2026-10-28 18:37:25 +03:00
Files
opencryptoki/usr/lib/api/apiutil.c
Akash Kulhalli 0c26ed6271 api: Add support for AIX
There are implementation-defined behaviours that need to be handled for various
UNIX platforms. For instance, the file locking featureset from BSD (`flock`)
requires the file to be opened in write mode on AIX, but Linux does not care
about the mode in which the file is opened. To preserve backward compatibility,
however, Linux retains its existing file mode; otherwise previous/leftover
daemon runs will cause future daemon startup failures - the lock can never be
obtained in the new file mode, because it will attempt to open a file with mode
bits `-r--` in write mode, which will never succeed (always returns `EPERM`).
There will be no difference to AIX because the open mode is still decided at
compile time, depending on the platform.

Dynamic token loading has also been changed, such that the symbols from all
token libraries are not visible in the process namespace anymore. This is
important because global symbol visibility affects the functionality that is
loaded - because each token exposes the same set of symbols, which have
differing behaviours based on the type of the token. The order in which the
tokens are specified in the configuration also makes a difference here, since
the token that was loaded first overrides the symbols that any later token
load provides. This basically means that all tokens that are not loaded first
do not have their symbols visible to the loading process, which means their
functions are never executed. Restricting the symbol visibility to the library
itself (`RTLD_LOCAL`) solves this issue.

IBM's xlc requires the `init` and `fini` functions to be specified on the
command-line, in addition to being specified as `pragma`s within the source
code. Furthermore, the format of the symbol export definition file differs
from that of gcc. xlc does not have semver support for shared libraries, and
therefore the export format file necessitates being different. This file is
therefore now generated dynamically during the configuration phase by an
(admittedly) hacky string manipulation within configure.ac. But it works(tm),
so it stays in until it breaks.

Signed-off-by: Akash Kulhalli <akash.kulhalli@ibm.com>
2024-04-24 09:48:07 +02:00

772 lines
20 KiB
C

/*
* COPYRIGHT (c) International Business Machines Corp. 2001-2017
*
* This program is provided under the terms of the Common Public License,
* version 1.0 (CPL-1.0). Any use, reproduction or distribution for this
* software constitutes recipient's acceptance of CPL-1.0 terms which can be
* found in the file LICENSE file or at
* https://opensource.org/licenses/cpl1.0.php
*/
//
//
//AIX Pkcs11 Api Utility functions
//
#include <stdint.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <strings.h>
#include <unistd.h>
#include <alloca.h>
#include <dlfcn.h>
#include <errno.h>
#include <sys/syslog.h>
#include <pthread.h>
#include <sys/ipc.h>
#include <pkcs11types.h>
#include <apiclient.h> // Function prototypes for PKCS11
#include <slotmgr.h>
#include <apictl.h>
#include <apiproto.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/file.h>
static int xplfd = -1;
pthread_rwlock_t xplfd_rwlock = PTHREAD_RWLOCK_INITIALIZER;
#include <libgen.h>
#define LIBLOCATION LIB_PATH
#if defined(_AIX)
#define SO_LDFLAGS (RTLD_LOCAL | RTLD_LAZY | RTLD_MEMBER)
#else
#define SO_LDFLAGS (RTLD_LOCAL | RTLD_LAZY)
#endif
extern API_Proc_Struct_t *Anchor;
#include <stdarg.h>
#include "trace.h"
#include "ock_syslog.h"
CK_RV CreateProcLock(void)
{
if (xplfd == -1) {
/* The slot mgr daemon should have already created lock,
* so just open it so we can get a lock...
*/
xplfd = open(OCK_API_LOCK_FILE, OPEN_MODE);
if (xplfd == -1) {
OCK_SYSLOG(LOG_ERR, "Could not open %s\n", OCK_API_LOCK_FILE);
return CKR_FUNCTION_FAILED;
}
}
return CKR_OK;
}
CK_RV ProcLock(void)
{
if (pthread_rwlock_wrlock(&xplfd_rwlock)) {
TRACE_ERROR("Lock failed.\n");
return CKR_CANT_LOCK;
}
if (xplfd != -1) {
flock(xplfd, LOCK_EX);
} else {
TRACE_DEVEL("No file descriptor to lock with.\n");
return CKR_CANT_LOCK;
}
return CKR_OK;
}
CK_RV ProcUnLock(void)
{
if (xplfd != -1) {
flock(xplfd, LOCK_UN);
} else {
TRACE_DEVEL("No file descriptor to unlock with.\n");
return CKR_CANT_LOCK;
}
if (pthread_rwlock_unlock(&xplfd_rwlock)) {
TRACE_ERROR("Unlock failed.\n");
return CKR_CANT_LOCK;
}
return CKR_OK;
}
CK_RV ProcClose(void)
{
if (xplfd != -1)
close(xplfd);
else
TRACE_DEVEL("ProcClose: No file descriptor open to close.\n");
return CKR_OK;
}
unsigned long AddToSessionList(ST_SESSION_T *pSess)
{
unsigned long handle;
handle = bt_node_add(&(Anchor->sess_btree), pSess);
return handle;
}
void RemoveFromSessionList(CK_SESSION_HANDLE handle)
{
bt_node_free(&(Anchor->sess_btree), handle, TRUE);
}
struct closeme_arg {
CK_SLOT_ID slot_id;
CK_BBOOL in_fork_initializer;
};
/* CloseMe
*
* Callback function used to close an individual session for a slot
*/
void CloseMe(STDLL_TokData_t *tokdata, void *node_value,
unsigned long node_handle, void *arg)
{
CK_RV rv;
struct closeme_arg *closeme_arg = (struct closeme_arg *) arg;
ST_SESSION_T *s = (ST_SESSION_T *) node_value;
API_Slot_t *sltp;
STDLL_FcnList_t *fcn;
UNUSED(tokdata);
if (s->slotID == closeme_arg->slot_id) {
/* the single ugliest part about moving to a binary tree: these are the
* guts of the C_CloseSession function, copied here without tests for
* validity, since if we're here, they must already have been valid */
sltp = &(Anchor->SltList[closeme_arg->slot_id]);
fcn = sltp->FcnList;
BEGIN_HSM_MK_CHANGE_LOCK(sltp, rv)
rv = fcn->ST_CloseSession(sltp->TokData, s,
closeme_arg->in_fork_initializer);
END_HSM_MK_CHANGE_LOCK(sltp, rv)
if (rv == CKR_OK) {
decr_sess_counts(closeme_arg->slot_id, s->rw_session);
bt_node_free(&(Anchor->sess_btree), node_handle, TRUE);
}
}
}
/* CloseAllSessions
*
* Run through all the nodes in the binary tree and call CloseMe on each one.
* CloseMe will look at @slot_id and if it matches, will close the session.
* Once all the nodes are closed, we check to see if the tree is empty and if
* so, destroy it
*/
void CloseAllSessions(CK_SLOT_ID slot_id, CK_BBOOL in_fork_initializer)
{
API_Slot_t *sltp = &(Anchor->SltList[slot_id]);
struct closeme_arg arg;
arg.slot_id = slot_id;
arg.in_fork_initializer = in_fork_initializer;
/* for every node in the API-level session tree, call CloseMe on it */
bt_for_each_node(sltp->TokData, &(Anchor->sess_btree), CloseMe,
(void *)&arg);
}
int Valid_Session(CK_SESSION_HANDLE handle, ST_SESSION_T *rSession)
{
ST_SESSION_T *tmp;
int rc;
tmp = bt_get_node_value(&(Anchor->sess_btree), handle);
if (tmp) {
rSession->slotID = tmp->slotID;
rSession->sessionh = tmp->sessionh;
rSession->rw_session = tmp->rw_session;
}
rc = tmp ? TRUE : FALSE;
bt_put_node_value(&(Anchor->sess_btree), tmp);
tmp = NULL;
return rc;
}
int API_Initialized(void)
{
if (Anchor == NULL)
return FALSE;
return TRUE;
}
int slot_present(CK_SLOT_ID id)
{
Slot_Mgr_Socket_t *shData = &(Anchor->SocketDataP);
#ifdef PKCS64
Slot_Info_t_64 *sinfp;
#else
Slot_Info_t *sinfp;
#endif
sinfp = &(shData->slot_info[id]);
if (sinfp->present == FALSE) {
return FALSE;
}
return TRUE;
}
void get_sess_counts(CK_SLOT_ID slotID, CK_ULONG *ret, CK_ULONG *rw_ret)
{
Slot_Mgr_Shr_t *shm;
shm = Anchor->SharedMemP;
ProcLock();
*ret = shm->slot_global_sessions[slotID];
*rw_ret = shm->slot_global_rw_sessions[slotID];
ProcUnLock();
}
void incr_sess_counts(CK_SLOT_ID slotID, CK_BBOOL rw_session)
{
Slot_Mgr_Shr_t *shm;
#ifdef PKCS64
Slot_Mgr_Proc_t_64 *procp;
#else
Slot_Mgr_Proc_t *procp;
#endif
// Get the slot mutex
shm = Anchor->SharedMemP;
ProcLock();
shm->slot_global_sessions[slotID]++;
if (rw_session)
shm->slot_global_rw_sessions[slotID]++;
procp = &shm->proc_table[Anchor->MgrProcIndex];
procp->slot_session_count[slotID]++;
if (rw_session)
procp->slot_rw_session_count[slotID]++;
ProcUnLock();
}
void decr_sess_counts(CK_SLOT_ID slotID, CK_BBOOL rw_session)
{
Slot_Mgr_Shr_t *shm;
#ifdef PKCS64
Slot_Mgr_Proc_t_64 *procp;
#else
Slot_Mgr_Proc_t *procp;
#endif
// Get the slot mutex
shm = Anchor->SharedMemP;
ProcLock();
if (shm->slot_global_sessions[slotID] > 0) {
shm->slot_global_sessions[slotID]--;
}
if (rw_session && shm->slot_global_rw_sessions[slotID] > 0) {
shm->slot_global_rw_sessions[slotID]--;
}
procp = &shm->proc_table[Anchor->MgrProcIndex];
if (procp->slot_session_count[slotID] > 0) {
procp->slot_session_count[slotID]--;
}
if (rw_session && procp->slot_rw_session_count[slotID] > 0) {
procp->slot_rw_session_count[slotID]--;
}
ProcUnLock();
}
uint32_t get_tokspec_count(STDLL_TokData_t *tokdata)
{
Slot_Mgr_Shr_t *shm;
uint32_t ret;
shm = Anchor->SharedMemP;
if (ProcLock() != CKR_OK)
return 0;
ret = shm->slot_global_tokspec_count[tokdata->slot_id];
ProcUnLock();
return ret;
}
void incr_tokspec_count(STDLL_TokData_t *tokdata)
{
Slot_Mgr_Shr_t *shm;
#ifdef PKCS64
Slot_Mgr_Proc_t_64 *procp;
#else
Slot_Mgr_Proc_t *procp;
#endif
// Get the slot mutex
shm = Anchor->SharedMemP;
ProcLock();
shm->slot_global_tokspec_count[tokdata->slot_id]++;
procp = &shm->proc_table[Anchor->MgrProcIndex];
procp->slot_tokspec_count[tokdata->slot_id]++;
ProcUnLock();
}
void decr_tokspec_count(STDLL_TokData_t *tokdata)
{
Slot_Mgr_Shr_t *shm;
#ifdef PKCS64
Slot_Mgr_Proc_t_64 *procp;
#else
Slot_Mgr_Proc_t *procp;
#endif
// Get the slot mutex
shm = Anchor->SharedMemP;
ProcLock();
if (shm->slot_global_tokspec_count[tokdata->slot_id] > 0)
shm->slot_global_tokspec_count[tokdata->slot_id]--;
procp = &shm->proc_table[Anchor->MgrProcIndex];
if (procp->slot_tokspec_count[tokdata->slot_id] > 0)
procp->slot_tokspec_count[tokdata->slot_id]--;
ProcUnLock();
}
// Check if any sessions from other applicaitons exist on this particular
// token.... This will also validate our own sessions as well.
// There might be an issue with the fact that a session is created but the
// number is not incremented until the session allocation is completed by
// the token. The API may need to lock the shared memory prior to creating
// the session and then unlock when the stdll has completed its work.
// Closing sessions should probably behave the same way.
int sessions_exist(CK_SLOT_ID slotID)
{
Slot_Mgr_Shr_t *shm;
uint32 numSessions;
// Get the slot mutex
shm = Anchor->SharedMemP;
ProcLock();
numSessions = shm->slot_global_sessions[slotID];
ProcUnLock();
return numSessions != 0;
}
// Register the process with PKCSSLOTD in the shared memory.
// This call must be made with the API Global Mutex Locked
// and the Anchor control block initialized with the
// shared memory. No checking for shared memory validity is done
int API_Register(void)
{
long int reuse = -1, free = -1;
Slot_Mgr_Shr_t *shm;
#ifdef PKCS64
Slot_Mgr_Proc_t_64 *procp;
#else
Slot_Mgr_Proc_t *procp;
#endif
uint16 indx;
// Grab the Shared Memory lock to prevent other updates to the
// SHM Process
// The registration is done to allow for future handling of
// the Slot Event List. Which is maintained by the Slotd.
shm = Anchor->SharedMemP;
ProcLock();
procp = shm->proc_table;
for (indx = 0; indx < NUMBER_PROCESSES_ALLOWED; indx++, procp++) {
// Is the entry in use
if (procp->inuse == TRUE) {
// Handle the weird case of the process terminating without
// un-registering, and restarting with exactly the same PID
// before the slot manager garbage collection can performed.
// To eliminate the race condition between garbage collection
// the lock should protect us.
// This should be a VERY rare (if ever) occurrence, given the
// way AIX deals with re-allocation of PID;s, however if this
// ever gets ported over to another platform we want to deal
// with this accordingly since it may re-use pids differently
// (Linux appears to re-use pids more rapidly).
if (procp->proc_id == Anchor->ClientCred.real_pid) {
if (reuse == -1) {
reuse = indx;
}
}
} else {
//Already found the first free
if (free == -1) {
free = indx;
}
}
}
// If we did not find a free entry then we fail the routine
if ((reuse == -1) && (free == -1)) {
ProcUnLock();
return FALSE;
}
// check if we are reusing a control block or taking the first free.
// Since the mutex is held, we don;t have to worry about some other
// process grabbing the slot... Garbage collection from
// the slotd should not affect this since it will grab the mutex
// before doing its thing.
if (reuse != -1) {
procp = &(shm->proc_table[reuse]);
indx = reuse;
} else {
procp = &(shm->proc_table[free]);
indx = free;
}
#ifdef PKCS64
memset((char *) procp, 0, sizeof(Slot_Mgr_Proc_t_64));
#else
memset((char *) procp, 0, sizeof(Slot_Mgr_Proc_t));
#endif
procp->inuse = TRUE;
procp->proc_id = Anchor->ClientCred.real_pid;
procp->reg_time = time(NULL);
Anchor->MgrProcIndex = indx;
TRACE_DEVEL("API_Register MgrProcIndc %ld (real) pid %d \n",
(long int) Anchor->MgrProcIndex, procp->proc_id);
//??? What to do about the Mutex and cond variable
//Does initializing them in the slotd allow for them to not be
//initialized in the application.
ProcUnLock();
return TRUE;
}
// DeRegister the process with PKCSSLOTD in the shared memory.
// This call must be made with the API Global Mutex Locked
// and the Anchor control block initialized with the
// shared memory. No checking for shared memory validity is done
void API_UnRegister(void)
{
Slot_Mgr_Shr_t *shm;
#ifdef PKCS64
Slot_Mgr_Proc_t_64 *procp;
#else
Slot_Mgr_Proc_t *procp;
#endif
// Grab the Shared Memory lock to prevent other updates to the
// SHM Process
// The registration is done to allow for future handling of
// the Slot Event List. Which is maintained by the Slotd.
shm = Anchor->SharedMemP;
ProcLock();
procp = &(shm->proc_table[Anchor->MgrProcIndex]);
#ifdef PKCS64
memset((char *) procp, 0, sizeof(Slot_Mgr_Proc_t_64));
#else
memset((char *) procp, 0, sizeof(Slot_Mgr_Proc_t));
#endif
Anchor->MgrProcIndex = 0;
//??? What to do about the Mutex and cond variable
//Does initializing them in the slotd allow for them to not be
//initialized in the application.
ProcUnLock();
}
void DL_UnLoad(API_Slot_t *sltp, CK_SLOT_ID slotID, CK_BBOOL inchildforkinit)
{
Slot_Mgr_Socket_t *shData = &(Anchor->SocketDataP);
#ifdef PKCS64
Slot_Info_t_64 *sinfp;
#else
Slot_Info_t *sinfp;
#endif
if (sltp->TokData) {
pthread_rwlock_destroy(&sltp->TokData->sess_list_rwlock);
pthread_mutex_destroy(&sltp->TokData->login_mutex);
if (sltp->TokData->hsm_mk_change_supported)
pthread_rwlock_destroy(&sltp->TokData->hsm_mk_change_rwlock);
free(sltp->TokData);
sltp->TokData = NULL;
}
sinfp = &(shData->slot_info[slotID]);
if (sinfp->present == FALSE) {
return;
}
if (!sltp->dlop_p) {
return;
}
if (inchildforkinit)
return;
// Call the routine to properly unload the DLL
DL_Unload(sltp);
return;
}
int DL_Loaded(char *location, DLL_Load_t *dllload)
{
int i;
for (i = 0; i < NUMBER_SLOTS_MANAGED; i++) {
if (dllload[i].dll_name != NULL) {
TRACE_DEBUG("DL_LOADED Looking for index %d name %s\n",
i, dllload[i].dll_name);
if (strcmp(location, dllload[i].dll_name) == 0) {
return i; // Return the index of the dll
}
}
}
return -1; // Indicate failure to find the dll
}
#ifdef PKCS64
int DL_Load(Slot_Info_t_64 *sinfp, API_Slot_t *sltp, DLL_Load_t *dllload)
#else
int DL_Load(Slot_Info_t *sinfp, API_Slot_t *sltp, DLL_Load_t *dllload)
#endif
{
int i;
TRACE_DEBUG("DL_LOAD\n");
for (i = 0; i < NUMBER_SLOTS_MANAGED; i++) {
if (dllload[i].dll_name == NULL) {
TRACE_DEBUG("Empty slot at %d \n", i);
break;
}
}
if (i == NUMBER_SLOTS_MANAGED) {
TRACE_DEBUG("No empty slots.\n");
return 0; // Failed to find it..
}
dllload[i].dll_name = sinfp->dll_location; // Point to the location
dllload[i].dlop_p = dlopen(sinfp->dll_location, SO_LDFLAGS);
if (dllload[i].dlop_p != NULL) {
sltp->dlop_p = dllload[i].dlop_p;
sltp->dll_information = &dllload[i];
dllload[i].dll_load_count++;;
} else {
char *e = dlerror();
OCK_SYSLOG(LOG_WARNING,
"%s: dlopen() failed for [%s]; dlerror = [%s]\n",
__func__, sinfp->dll_location, e);
TRACE_DEVEL("DL_Load of %s failed, dlerror: %s\n",
sinfp->dll_location, e);
sltp->dlop_p = NULL;
return 0;
}
return 1;
}
void DL_Unload(API_Slot_t *sltp)
{
DLL_Load_t *dllload;
// Decrement the count of loads. When 0 then unload this thing;
//
dllload = sltp->dll_information;
dllload->dll_load_count--;
if (dllload->dll_load_count == 0) {
dlclose(dllload->dlop_p);
dllload->dll_name = NULL;
}
// Clear out the slot information
sltp->DLLoaded = FALSE;
sltp->dlop_p = NULL;
sltp->pSTfini = NULL;
sltp->pSTcloseall = NULL;
}
int DL_Load_and_Init(API_Slot_t *sltp, CK_SLOT_ID slotID, policy_t policy,
statistics_t statistics)
{
Slot_Mgr_Socket_t *shData = &(Anchor->SocketDataP);
#ifdef PKCS64
Slot_Info_t_64 *sinfp;
#else
Slot_Info_t *sinfp;
#endif
CK_RV (*pSTinit)(API_Slot_t *, CK_SLOT_ID, SLOT_INFO *,
struct trace_handle_t);
CK_RV rv;
int dl_index;
DLL_Load_t *dllload;
// Get pointer to shared memory from the anchor block
//
sinfp = &(shData->slot_info[slotID]);
dllload = Anchor->DLLs; // list of dll's in the system
if (sinfp->present == FALSE) {
return FALSE;
}
if (sltp->TokData != NULL) {
TRACE_ERROR("Already initialized.\n");
return FALSE;
}
/*
* Create separate memory area for each token specific data
*/
sltp->TokData = (STDLL_TokData_t *) calloc(1, sizeof(STDLL_TokData_t));
if (!sltp->TokData) {
TRACE_ERROR("Allocating host memory failed.\n");
return FALSE;
}
sltp->TokData->slot_id = slotID;
sltp->TokData->real_pid = Anchor->ClientCred.real_pid;
sltp->TokData->real_uid = Anchor->ClientCred.real_uid;
sltp->TokData->real_gid = Anchor->ClientCred.real_gid;
sltp->TokData->tokspec_counter.get_tokspec_count = get_tokspec_count;
sltp->TokData->tokspec_counter.incr_tokspec_count = incr_tokspec_count;
sltp->TokData->tokspec_counter.decr_tokspec_count = decr_tokspec_count;
sltp->TokData->ro_session_count = 0;
sltp->TokData->global_login_state = CKS_RO_PUBLIC_SESSION;
sltp->TokData->spinxplfd = -1;
sltp->TokData->spinxplfd_count = 0;
if (pthread_rwlock_init(&sltp->TokData->sess_list_rwlock, NULL) != 0) {
TRACE_ERROR("Initializing session list lock failed.\n");
free(sltp->TokData);
sltp->TokData = NULL;
return FALSE;
}
if (pthread_mutex_init(&sltp->TokData->login_mutex, NULL) != 0) {
TRACE_ERROR("Initializing login mutex failed.\n");
free(sltp->TokData);
sltp->TokData = NULL;
return FALSE;
}
sltp->TokData->policy = policy;
sltp->TokData->mechtable_funcs = &mechtable_funcs;
sltp->TokData->statistics = statistics;
if (strlen(sinfp->dll_location) > 0) {
// Check if this DLL has been loaded already.. If so, just increment
// the counter in the dllload structure and copy the data to
// the slot pointer.
if ((dl_index = DL_Loaded(sinfp->dll_location, dllload)) != -1) {
dllload[dl_index].dll_load_count++;
sltp->dll_information = &dllload[dl_index];
sltp->dlop_p = dllload[dl_index].dlop_p;
} else {
TRACE_DEBUG("DL_Load_and_Init dll_location %s\n",
sinfp->dll_location);
DL_Load(sinfp, sltp, dllload);
}
} else {
free(sltp->TokData);
sltp->TokData = NULL;
return FALSE;
}
if (!sltp->dlop_p) {
TRACE_DEBUG("DL_Load_and_Init pointer NULL\n");
DL_UnLoad(sltp, slotID, FALSE);
return FALSE;
}
*(void **)(&pSTinit) = dlsym(sltp->dlop_p, "ST_Initialize");
if (!pSTinit) {
// Unload the DLL
DL_UnLoad(sltp, slotID, FALSE);
return FALSE;
}
// Returns true or false
rv = pSTinit(sltp, slotID, sinfp, trace);
TRACE_DEBUG("return from STDDLL Init = %lx\n", rv);
if (rv != CKR_OK) {
// clean up and unload
DL_UnLoad(sltp, slotID, FALSE);
sltp->DLLoaded = FALSE;
return FALSE;
} else {
sltp->DLLoaded = TRUE;
sinfp->pk_slot.flags |= CKF_TOKEN_PRESENT;
// Check if a SC_Finalize function has been exported
*(void **)(&sltp->pSTfini) = dlsym(sltp->dlop_p, "SC_Finalize");
*(void **)(&sltp->pSTcloseall) =
dlsym(sltp->dlop_p, "SC_CloseAllSessions");
return TRUE;
}
return TRUE;
}
// copies internal representation of ck_info structure to local process
// representation
void CK_Info_From_Internal(CK_INFO_PTR dest, CK_INFO_PTR_64 src)
{
dest->cryptokiVersion = src->cryptokiVersion;
memcpy(dest->manufacturerID, src->manufacturerID,
sizeof(dest->manufacturerID));
dest->flags = src->flags;
memcpy(dest->libraryDescription, src->libraryDescription,
sizeof(dest->libraryDescription));
dest->libraryVersion = src->libraryVersion;
}