int
pfq_sock_tx_unbind(struct pfq_sock *so)
{
size_t n;
/* unbind sync Tx queue */
if (so->opt.txq.def_ifindex != -1) {
dev_put_by_index(sock_net(&so->sk), so->opt.txq.def_ifindex);
}
so->opt.txq.def_ifindex = -1;
so->opt.txq.def_queue = -1;
so->opt.txq.def_dev = NULL;
/* unbind async Tx queue */
if (pfq_unbind_tx_thread(so) < 0)
return -EPERM;
for(n = 0; n < Q_MAX_TX_QUEUES; ++n)
{
if (so->opt.txq_async[n].def_ifindex != -1)
dev_put_by_index(sock_net(&so->sk), so->opt.txq_async[n].def_ifindex);
so->opt.txq_async[n].def_ifindex = -1;
so->opt.txq_async[n].def_queue = -1;
so->opt.txq_async[n].def_dev = NULL;
}
return 0;
}
int pfq_setsockopt(struct socket *sock,
int level, int optname,
char __user * optval,
#if(LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,31))
unsigned
#endif
int optlen)
{
struct pfq_sock *so = pfq_sk(sock->sk);
bool found = true;
if (so == NULL)
return -EINVAL;
switch(optname)
{
case Q_SO_ENABLE:
{
unsigned long addr;
int err = 0;
if (optlen != sizeof(addr))
return -EINVAL;
if (copy_from_user(&addr, optval, optlen))
return -EFAULT;
err = pfq_shared_queue_enable(so, addr);
if (err < 0) {
printk(KERN_INFO "[PFQ|%d] enable error!\n", so->id);
return err;
}
return 0;
} break;
case Q_SO_DISABLE:
{
int err = 0;
pfq_sock_tx_unbind(so);
msleep(Q_GRACE_PERIOD);
err = pfq_shared_queue_disable(so);
if (err < 0) {
printk(KERN_INFO "[PFQ|%d] disable error!\n", so->id);
return err;
}
} break;
case Q_SO_GROUP_BIND:
{
struct pfq_binding bind;
pfq_gid_t gid;
if (optlen != sizeof(struct pfq_binding))
return -EINVAL;
if (copy_from_user(&bind, optval, optlen))
return -EFAULT;
gid = (__force pfq_gid_t)bind.gid;
if (!pfq_has_joined_group(gid, so->id)) {
printk(KERN_INFO "[PFQ|%d] add bind: gid=%d not joined!\n", so->id, bind.gid);
return -EACCES;
}
if (!dev_get_by_index(sock_net(&so->sk), bind.ifindex)) {
printk(KERN_INFO "[PFQ|%d] bind: invalid ifindex=%d!\n", so->id, bind.ifindex);
return -EACCES;
}
pfq_devmap_update(map_set, bind.ifindex, bind.qindex, gid);
pr_devel("[PFQ|%d] group id=%d bind: device ifindex=%d qindex=%d\n",
so->id, bind.gid, bind.ifindex, bind.qindex);
} break;
case Q_SO_GROUP_UNBIND:
{
struct pfq_binding bind;
pfq_gid_t gid;
if (optlen != sizeof(struct pfq_binding))
return -EINVAL;
if (copy_from_user(&bind, optval, optlen))
return -EFAULT;
gid = (__force pfq_gid_t)bind.gid;
if (!pfq_has_joined_group(gid, so->id)) {
printk(KERN_INFO "[PFQ|%d] group id=%d unbind: gid=%d not joined!\n", so->id, gid, bind.gid);
return -EACCES;
}
if (dev_put_by_index(sock_net(&so->sk), bind.ifindex) < 0) {
printk(KERN_INFO "[PFQ|%d] group id=%d unbind: invalid ifindex=%d!\n", so->id, gid, bind.ifindex);
return -EPERM;
}
pfq_devmap_update(map_reset, bind.ifindex, bind.qindex, gid);
pr_devel("[PFQ|%d] group id=%d unbind: device ifindex=%d qindex=%d\n",
so->id, gid, bind.ifindex, bind.qindex);
} break;
case Q_SO_EGRESS_BIND:
{
struct pfq_binding bind;
if (optlen != sizeof(bind))
return -EINVAL;
if (copy_from_user(&bind, optval, optlen))
return -EFAULT;
if (!dev_get_by_index(sock_net(&so->sk), bind.ifindex)) {
printk(KERN_INFO "[PFQ|%d] egress bind: invalid ifindex=%d\n", so->id, bind.ifindex);
return -EPERM;
}
if (bind.qindex < -1) {
printk(KERN_INFO "[PFQ|%d] egress bind: invalid qindex=%d\n", so->id, bind.qindex);
return -EPERM;
}
so->egress_type = pfq_endpoint_device;
so->egress_index = bind.ifindex;
so->egress_queue = bind.qindex;
pr_devel("[PFQ|%d] egress bind: device ifindex=%d qindex=%d\n",
so->id, so->egress_index, so->egress_queue);
} break;
case Q_SO_EGRESS_UNBIND:
{
if (so->egress_index &&
dev_put_by_index(sock_net(&so->sk), so->egress_index) < 0) {
printk(KERN_INFO "[PFQ|%d] egress bind: invalid if_index=%d\n", so->id, so->egress_index);
return -EPERM;
}
so->egress_type = pfq_endpoint_socket;
so->egress_index = 0;
so->egress_queue = 0;
pr_devel("[PFQ|%d] egress unbind.\n", so->id);
} break;
case Q_SO_SET_RX_TSTAMP:
{
int tstamp;
if (optlen != sizeof(so->opt.tstamp))
return -EINVAL;
if (copy_from_user(&tstamp, optval, optlen))
return -EFAULT;
tstamp = tstamp ? 1 : 0;
so->opt.tstamp = tstamp;
pr_devel("[PFQ|%d] timestamp enabled.\n", so->id);
} break;
case Q_SO_SET_RX_CAPLEN:
{
typeof(so->opt.caplen) caplen;
if (optlen != sizeof(caplen))
return -EINVAL;
if (copy_from_user(&caplen, optval, optlen))
return -EFAULT;
if (caplen > (size_t)capt_slot_size) {
printk(KERN_INFO "[PFQ|%d] invalid caplen=%zu (max %d)\n", so->id, caplen, capt_slot_size);
return -EPERM;
}
so->opt.caplen = caplen;
so->opt.rx_slot_size = Q_QUEUE_SLOT_SIZE(so->opt.caplen);
pr_devel("[PFQ|%d] caplen=%zu, slot_size=%zu\n",
so->id, so->opt.caplen, so->opt.rx_slot_size);
} break;
case Q_SO_SET_RX_SLOTS:
{
typeof(so->opt.rx_queue_len) slots;
if (optlen != sizeof(slots))
return -EINVAL;
if (copy_from_user(&slots, optval, optlen))
return -EFAULT;
if (slots > Q_MAX_SOCKQUEUE_LEN) {
printk(KERN_INFO "[PFQ|%d] invalid Rx slots=%zu (max %d)\n",
so->id, slots, Q_MAX_SOCKQUEUE_LEN);
return -EPERM;
}
so->opt.rx_queue_len = slots;
pr_devel("[PFQ|%d] rx_queue slots=%zu\n", so->id, so->opt.rx_queue_len);
} break;
case Q_SO_SET_TX_SLOTS:
{
typeof (so->opt.tx_queue_len) slots;
if (optlen != sizeof(slots))
return -EINVAL;
if (copy_from_user(&slots, optval, optlen))
return -EFAULT;
if (slots > Q_MAX_SOCKQUEUE_LEN) {
printk(KERN_INFO "[PFQ|%d] invalid Tx slots=%zu (max %d)\n",
so->id, slots, Q_MAX_SOCKQUEUE_LEN);
return -EPERM;
}
so->opt.tx_queue_len = slots;
pr_devel("[PFQ|%d] tx_queue slots=%zu\n", so->id, so->opt.tx_queue_len);
} break;
case Q_SO_SET_WEIGHT:
{
int weight;
if (optlen != sizeof(so->weight))
return -EINVAL;
if (copy_from_user(&weight, optval, optlen))
return -EFAULT;
if (weight < 1 || weight > (Q_MAX_SOCK_MASK/Q_MAX_ID)) {
printk(KERN_INFO "[PFQ|%d] weight=%d: invalid range (min 1, max %d)\n", so->id, weight,
Q_MAX_SOCK_MASK/Q_MAX_ID);
return -EPERM;
}
so->weight = weight;
/* invalidate per-cpu sock mask cache */
pfq_invalidate_percpu_eligible_mask(so->id);
pr_devel("[PFQ|%d] new weight set to %d.\n", so->id, weight);
} break;
case Q_SO_GROUP_LEAVE:
{
pfq_gid_t gid;
if (optlen != sizeof(gid))
return -EINVAL;
if (copy_from_user(&gid, optval, optlen))
return -EFAULT;
if (pfq_leave_group(gid, so->id) < 0)
return -EFAULT;
pr_devel("[PFQ|%d] group id=%d left.\n", so->id, gid);
} break;
case Q_SO_GROUP_FPROG:
{
struct pfq_fprog fprog;
pfq_gid_t gid;
if (optlen != sizeof(fprog))
return -EINVAL;
if (copy_from_user(&fprog, optval, optlen))
return -EFAULT;
gid = (__force pfq_gid_t)fprog.gid;
if (!pfq_has_joined_group(gid, so->id)) {
/* don't set the first and return */
return 0;
}
if (fprog.fcode.len > 0) { /* set the filter */
struct sk_filter *filter;
if (fprog.fcode.len == 1) {
struct sock_filter tmp;
/* get the first filter */
if (copy_from_user(&tmp, fprog.fcode.filter, sizeof(tmp)))
return -EFAULT;
/* check whether the first filter is a dummy BPF_RET */
if (BPF_CLASS(tmp.code) == BPF_RET) {
pr_devel("[PFQ|%d] fprog: BPF_RET optimized out!\n", so->id);
return 0;
}
}
filter = pfq_alloc_sk_filter(&fprog.fcode);
if (filter == NULL) {
printk(KERN_INFO "[PFQ|%d] fprog error: alloc_sk_filter for gid=%d\n",
so->id, fprog.gid);
return -EINVAL;
}
pfq_set_group_filter(gid, filter);
pr_devel("[PFQ|%d] fprog: gid=%d (fprog len %d bytes)\n",
so->id, fprog.gid, fprog.fcode.len);
}
else {
/* reset the filter */
pfq_set_group_filter(gid, NULL);
pr_devel("[PFQ|%d] fprog: gid=%d (resetting filter)\n", so->id, fprog.gid);
}
} break;
case Q_SO_GROUP_VLAN_FILT_TOGGLE:
{
struct pfq_vlan_toggle vlan;
pfq_gid_t gid;
if (optlen != sizeof(vlan))
return -EINVAL;
if (copy_from_user(&vlan, optval, optlen))
return -EFAULT;
gid = (__force pfq_gid_t)vlan.gid;
if (!pfq_has_joined_group(gid, so->id)) {
printk(KERN_INFO "[PFQ|%d] vlan filter toggle: gid=%d not joined!\n", so->id, vlan.gid);
return -EACCES;
}
pfq_toggle_group_vlan_filters(gid, vlan.toggle);
pr_devel("[PFQ|%d] vlan filters %s for gid=%d\n",
so->id, (vlan.toggle ? "enabled" : "disabled"), vlan.gid);
} break;
case Q_SO_GROUP_VLAN_FILT:
{
struct pfq_vlan_toggle filt;
pfq_gid_t gid;
if (optlen != sizeof(filt))
return -EINVAL;
if (copy_from_user(&filt, optval, optlen))
return -EFAULT;
gid = (__force pfq_gid_t)filt.gid;
if (!pfq_has_joined_group(gid, so->id)) {
printk(KERN_INFO "[PFQ|%d] vlan filter: gid=%d not joined!\n", so->id, filt.gid);
return -EACCES;
}
if (filt.vid < -1 || filt.vid > 4094) {
printk(KERN_INFO "[PFQ|%d] vlan error: invalid vid=%d for gid=%d!\n",
so->id, filt.vid, filt.gid);
return -EINVAL;
}
if (!pfq_vlan_filters_enabled(gid)) {
printk(KERN_INFO "[PFQ|%d] vlan error: vlan filters disabled for gid=%d!\n",
so->id, filt.gid);
return -EPERM;
}
if (filt.vid == -1) { /* any */
int i;
for(i = 1; i < 4095; i++)
{
pfq_set_group_vlan_filter(gid, filt.toggle, i);
}
}
else {
pfq_set_group_vlan_filter(gid, filt.toggle, filt.vid);
}
pr_devel("[PFQ|%d] vlan filter vid %d set for gid=%d\n", so->id, filt.vid, filt.gid);
} break;
case Q_SO_TX_BIND:
{
struct pfq_binding bind;
struct net_device *dev = NULL;
if (optlen != sizeof(bind))
return -EINVAL;
if (copy_from_user(&bind, optval, optlen))
return -EFAULT;
if (bind.tid < -1) {
printk(KERN_INFO "[PFQ|%d] Tx thread: invalid thread index (%d)!\n", so->id, bind.tid);
return -EPERM;
}
if (bind.tid >= 0 &&
so->opt.tx_num_async_queues >= Q_MAX_TX_QUEUES) {
printk(KERN_INFO "[PFQ|%d] Tx thread: max number of sock queues exceeded!\n", so->id);
return -EPERM;
}
if (bind.qindex < -1) {
printk(KERN_INFO "[PFQ|%d] Tx thread: invalid hw queue (%d)\n", so->id, bind.qindex);
return -EPERM;
}
/* get device */
if (bind.ifindex != -1 &&
!(dev = dev_get_by_index(sock_net(&so->sk), bind.ifindex))) {
printk(KERN_INFO "[PFQ|%d] Tx thread: invalid ifindex=%d\n", so->id, bind.ifindex);
return -EPERM;
}
/* update the socket queue information */
if (bind.tid >= 0) /* async queues */
{
int err = pfq_sock_tx_bind(so, bind.tid, bind.ifindex, bind.qindex, dev);
if (err < 0) {
if (bind.ifindex != -1)
dev_put_by_index(sock_net(&so->sk), bind.ifindex);
return err;
}
pr_devel("[PFQ|%d] Tx[%d] bind: if_index=%d qindex=%d\n", so->id, bind.tid, bind.ifindex, bind.qindex);
}
else /* sync queue */
{
so->opt.txq.def_ifindex = bind.ifindex;
so->opt.txq.def_queue = bind.qindex;
so->opt.txq.def_dev = dev;
pr_devel("[PFQ|%d] Tx bind: if_index=%d qindex=%d\n", so->id,
so->opt.txq.def_ifindex,
so->opt.txq.def_queue);
}
} break;
case Q_SO_TX_UNBIND:
{
pfq_sock_tx_unbind(so);
} break;
case Q_SO_TX_QUEUE:
{
int queue;
if (optlen != sizeof(queue))
return -EINVAL;
if (copy_from_user(&queue, optval, optlen))
return -EFAULT;
if (pfq_get_tx_queue(&so->opt, -1) == NULL) {
printk(KERN_INFO "[PFQ|%d] Tx queue: socket not enabled!\n", so->id);
return -EPERM;
}
if (queue == 0) { /* transmit Tx queue */
atomic_t stop = {0};
pfq_sk_queue_xmit(so, -1, Q_NO_KTHREAD, NUMA_NO_NODE, &stop);
return 0;
}
printk(KERN_INFO "[PFQ|%d] Tx queue: bad queue %d!\n", so->id, queue);
return -EPERM;
} break;
case Q_SO_GROUP_FUNCTION:
{
struct pfq_lang_computation_descr *descr = NULL;
struct pfq_lang_computation_tree *comp = NULL;
struct pfq_group_computation tmp;
size_t psize, ucsize;
void *context = NULL;
pfq_gid_t gid;
int err = 0;
if (optlen != sizeof(tmp))
return -EINVAL;
if (copy_from_user(&tmp, optval, optlen))
return -EFAULT;
gid = (__force pfq_gid_t)tmp.gid;
if (!pfq_has_joined_group(gid, so->id)) {
printk(KERN_INFO "[PFQ|%d] group computation: gid=%d not joined!\n", so->id, tmp.gid);
return -EACCES;
}
if (copy_from_user(&psize, tmp.prog, sizeof(size_t)))
return -EFAULT;
pr_devel("[PFQ|%d] computation size: %zu\n", so->id, psize);
ucsize = sizeof(size_t) * 2 + psize * sizeof(struct pfq_lang_functional_descr);
descr = kmalloc(ucsize, GFP_KERNEL);
if (descr == NULL) {
printk(KERN_INFO "[PFQ|%d] computation: out of memory!\n", so->id);
return -ENOMEM;
}
if (copy_from_user(descr, tmp.prog, ucsize)) {
printk(KERN_INFO "[PFQ|%d] computation: copy_from_user error!\n", so->id);
err = -EFAULT;
goto error;
}
/* print user computation */
pr_devel_computation_descr(descr);
/* check the correctness of computation */
if (pfq_lang_check_computation_descr(descr) < 0) {
printk(KERN_INFO "[PFQ|%d] invalid expression!\n", so->id);
err = -EFAULT;
goto error;
}
/* allocate context */
context = pfq_lang_context_alloc(descr);
if (context == NULL) {
printk(KERN_INFO "[PFQ|%d] context: alloc error!\n", so->id);
err = -EFAULT;
goto error;
}
/* allocate a pfq_lang_computation_tree */
comp = pfq_lang_computation_alloc(descr);
if (comp == NULL) {
printk(KERN_INFO "[PFQ|%d] computation: alloc error!\n", so->id);
err = -EFAULT;
goto error;
}
/* link functions of computation */
if (pfq_lang_computation_rtlink(descr, comp, context) < 0) {
printk(KERN_INFO "[PFQ|%d] computation aborted!", so->id);
err = -EPERM;
goto error;
}
/* print executable tree data structure */
pr_devel_computation_tree(comp);
/* run init functions */
if (pfq_lang_computation_init(comp) < 0) {
printk(KERN_INFO "[PFQ|%d] initialization of computation aborted!", so->id);
pfq_lang_computation_destruct(comp);
err = -EPERM;
goto error;
}
/* enable functional program */
if (pfq_set_group_prog(gid, comp, context) < 0) {
printk(KERN_INFO "[PFQ|%d] set group program error!\n", so->id);
err = -EPERM;
goto error;
}
kfree(descr);
return 0;
error: kfree(comp);
kfree(context);
kfree(descr);
return err;
} break;
default:
{
found = false;
} break;
}
return found ? 0 : sock_setsockopt(sock, level, optname, optval, optlen);
}
