/*
* Find the largest index block in the MRC cache. Return NULL if none is
* found.
*/
static struct mrc_data_container *find_current_mrc_cache_local
(struct mrc_data_container *mrc_cache, u32 region_size)
{
u32 region_end;
u32 entry_id = 0;
struct mrc_data_container *mrc_next = mrc_cache;
region_end = (u32) mrc_cache + region_size;
/* Search for the last filled entry in the region */
while (is_mrc_cache(mrc_next)) {
entry_id++;
mrc_cache = mrc_next;
mrc_next = next_mrc_block(mrc_next);
if ((u32)mrc_next >= region_end) {
/* Stay in the MRC data region */
break;
}
}
if (entry_id == 0) {
printk(BIOS_ERR, "%s: No valid fast boot cache found.\n", __func__);
return NULL;
}
/* Verify checksum */
if (mrc_cache->mrc_checksum !=
compute_ip_checksum(mrc_cache->mrc_data,
mrc_cache->mrc_data_size)) {
printk(BIOS_ERR, "%s: fast boot cache checksum mismatch\n", __func__);
return NULL;
}
printk(BIOS_DEBUG, "%s: picked entry %u from cache block\n", __func__,
entry_id - 1);
return mrc_cache;
}
/*---------------------------------------------------------------------
* Method: ip_header_create(struct frame_t *outgoing)
* Scope: Local
*
* This method fills in the IP header on send_datagram based on the
* parameters given.
*--------------------------------------------------------------------*/
void ip_header_create(struct frame_t *outgoing)
{
assert(outgoing);
struct ip *send_header = outgoing->ip_header;
assert(send_header);
send_header->ip_v = 4;
send_header->ip_hl = 5;
send_header->ip_tos = 0;
send_header->ip_id = 0;
send_header->ip_off = 0;
send_header->ip_ttl = INIT_TTL;
send_header->ip_p = IPPROTO_ICMP;
send_header->ip_len = htons(outgoing->ip_len);
send_header->ip_src.s_addr = outgoing->from_ip;
send_header->ip_dst.s_addr = outgoing->to_ip;
send_header->ip_sum = 0;
compute_ip_checksum(outgoing);
return;
}
struct mrc_data_container *mrccache_find_current(struct mrc_region *entry)
{
struct mrc_data_container *cache, *next;
ulong base_addr, end_addr;
uint id;
base_addr = entry->base + entry->offset;
end_addr = base_addr + entry->length;
cache = NULL;
/* Search for the last filled entry in the region */
for (id = 0, next = (struct mrc_data_container *)base_addr;
is_mrc_cache(next);
id++) {
cache = next;
next = next_mrc_block(next);
if ((ulong)next >= end_addr)
break;
}
if (id-- == 0) {
debug("%s: No valid MRC cache found.\n", __func__);
return NULL;
}
/* Verify checksum */
if (cache->checksum != compute_ip_checksum(cache->data,
cache->data_size)) {
printf("%s: MRC cache checksum mismatch\n", __func__);
return NULL;
}
debug("%s: picked entry %u from cache block\n", __func__, id);
return cache;
}
/* When the ramstage is relocatable the elf loading ensures an elf image cannot
* be loaded over the ramstage code. */
void jmp_to_elf_entry(void *entry, unsigned long unused1, unsigned long unused2)
{
elf_boot_notes.hdr.b_checksum =
compute_ip_checksum(&elf_boot_notes, sizeof(elf_boot_notes));
/* Jump to kernel */
__asm__ __volatile__(
" cld \n\t"
/* Now jump to the loaded image */
" call *%0\n\t"
/* The loaded image returned? */
" cli \n\t"
" cld \n\t"
::
"r" (entry),
#if CONFIG_MULTIBOOT
"b"(mbi), "a" (MB_MAGIC2)
#else
"b"(&elf_boot_notes), "a" (0x0E1FB007)
#endif
);
}
ssize_t dst_entry_tcp::fast_send(const struct iovec* p_iov, const ssize_t sz_iov, bool b_blocked /*= true*/, bool is_rexmit /*= false*/, bool dont_inline /*= false*/)
{
tx_packet_template_t* p_pkt;
mem_buf_desc_t *p_mem_buf_desc;
size_t total_packet_len = 0;
// The header is aligned for fast copy but we need to maintain this diff in order to get the real header pointer easily
size_t hdr_alignment_diff = m_header.m_aligned_l2_l3_len - m_header.m_total_hdr_len;
tcp_iovec* p_tcp_iov = NULL;
bool no_copy = true;
if (likely(sz_iov == 1 && !is_rexmit)) {
p_tcp_iov = (tcp_iovec*)p_iov;
if (unlikely(!m_p_ring->is_active_member(p_tcp_iov->p_desc->p_desc_owner, m_id))) {
no_copy = false;
dst_tcp_logdbg("p_desc=%p wrong desc_owner=%p, this ring=%p. did migration occurred?", p_tcp_iov->p_desc, p_tcp_iov->p_desc->p_desc_owner, m_p_ring);
//todo can we handle this in migration (by going over all buffers lwip hold) instead for every send?
}
} else {
no_copy = false;
}
if (unlikely(is_rexmit))
m_p_ring->inc_ring_stats(m_id);
if (likely(no_copy)) {
p_pkt = (tx_packet_template_t*)((uint8_t*)p_tcp_iov[0].iovec.iov_base - m_header.m_aligned_l2_l3_len);
total_packet_len = p_tcp_iov[0].iovec.iov_len + m_header.m_total_hdr_len;
m_header.copy_l2_ip_hdr(p_pkt);
// We've copied to aligned address, and now we must update p_pkt to point to real
// L2 header
//p_pkt = (tx_packet_template_t*)((uint8_t*)p_pkt + hdr_alignment_diff);
p_pkt->hdr.m_ip_hdr.tot_len = (htons)(p_tcp_iov[0].iovec.iov_len + m_header.m_ip_header_len);
m_sge[0].addr = (uintptr_t)((uint8_t*)p_pkt + hdr_alignment_diff);
m_sge[0].length = total_packet_len;
/* for DEBUG */
if ((uint8_t*)m_sge[0].addr < p_tcp_iov[0].p_desc->p_buffer || (uint8_t*)p_pkt < p_tcp_iov[0].p_desc->p_buffer) {
dst_tcp_logerr("p_buffer - addr=%d, m_total_hdr_len=%zd, p_buffer=%p, type=%d, len=%d, tot_len=%d, payload=%p, hdr_alignment_diff=%zd\n",
(int)(p_tcp_iov[0].p_desc->p_buffer - (uint8_t*)m_sge[0].addr), m_header.m_total_hdr_len,
p_tcp_iov[0].p_desc->p_buffer, p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.type,
p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.len, p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.tot_len,
p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.payload, hdr_alignment_diff);
}
if (!dont_inline && (total_packet_len < m_max_inline)) { // inline send
m_p_send_wqe = &m_inline_send_wqe;
} else {
m_p_send_wqe = &m_not_inline_send_wqe;
}
m_p_send_wqe->wr_id = (uintptr_t)p_tcp_iov[0].p_desc;
#ifdef VMA_NO_HW_CSUM
p_pkt->hdr.m_ip_hdr.check = 0; // use 0 at csum calculation time
p_pkt->hdr.m_ip_hdr.check = compute_ip_checksum((unsigned short*)&p_pkt->hdr.m_ip_hdr, p_pkt->hdr.m_ip_hdr.ihl * 2);
struct tcphdr* p_tcphdr = (struct tcphdr*)(((uint8_t*)(&(p_pkt->hdr.m_ip_hdr))+sizeof(p_pkt->hdr.m_ip_hdr)));
p_tcphdr->check = 0;
p_tcphdr->check = compute_tcp_checksum(&p_pkt->hdr.m_ip_hdr, (const uint16_t *)p_tcphdr);
dst_tcp_logfine("using SW checksum calculation: p_pkt->hdr.m_ip_hdr.check=%d, p_tcphdr->check=%d", (int)p_pkt->hdr.m_ip_hdr.check, (int)p_tcphdr->check);
#endif
m_p_ring->send_lwip_buffer(m_id, m_p_send_wqe, b_blocked);
}
else { // We don'nt support inline in this case, since we believe that this a very rare case
p_mem_buf_desc = get_buffer(b_blocked);
if (p_mem_buf_desc == NULL) {
return -1;
}
m_header.copy_l2_ip_hdr((tx_packet_template_t*)p_mem_buf_desc->p_buffer);
// Actually this is not the real packet len we will subtract the alignment diff at the end of the copy
total_packet_len = m_header.m_aligned_l2_l3_len;
for (int i = 0; i < sz_iov; ++i) {
memcpy(p_mem_buf_desc->p_buffer + total_packet_len, p_iov[i].iov_base, p_iov[i].iov_len);
total_packet_len += p_iov[i].iov_len;
}
m_sge[0].addr = (uintptr_t)(p_mem_buf_desc->p_buffer + hdr_alignment_diff);
m_sge[0].length = total_packet_len - hdr_alignment_diff;
// LKey will be updated in ring->send() // m_sge[0].lkey = p_mem_buf_desc->lkey;
/* for DEBUG */
if ((uint8_t*)m_sge[0].addr < p_mem_buf_desc->p_buffer) {
dst_tcp_logerr("p_buffer - addr=%d, m_total_hdr_len=%zd, p_buffer=%p, type=%d, len=%d, tot_len=%d, payload=%p, hdr_alignment_diff=%zd\n",
(int)(p_mem_buf_desc->p_buffer - (uint8_t*)m_sge[0].addr), m_header.m_total_hdr_len,
p_mem_buf_desc->p_buffer, p_mem_buf_desc->lwip_pbuf.pbuf.type,
p_mem_buf_desc->lwip_pbuf.pbuf.len, p_mem_buf_desc->lwip_pbuf.pbuf.tot_len,
p_mem_buf_desc->lwip_pbuf.pbuf.payload, hdr_alignment_diff);
}
p_pkt = (tx_packet_template_t*)((uint8_t*)p_mem_buf_desc->p_buffer);
p_pkt->hdr.m_ip_hdr.tot_len = (htons)(m_sge[0].length - m_header.m_transport_header_len);
#ifdef VMA_NO_HW_CSUM
p_pkt->hdr.m_ip_hdr.check = 0; // use 0 at csum calculation time
p_pkt->hdr.m_ip_hdr.check = compute_ip_checksum((unsigned short*)&p_pkt->hdr.m_ip_hdr, p_pkt->hdr.m_ip_hdr.ihl * 2);
struct tcphdr* p_tcphdr = (struct tcphdr*)(((uint8_t*)(&(p_pkt->hdr.m_ip_hdr))+sizeof(p_pkt->hdr.m_ip_hdr)));
p_tcphdr->check = 0;
p_tcphdr->check = compute_tcp_checksum(&p_pkt->hdr.m_ip_hdr, (const uint16_t *)p_tcphdr);
dst_tcp_logfine("using SW checksum calculation: p_pkt->hdr.m_ip_hdr.check=%d, p_tcphdr->check=%d", (int)p_pkt->hdr.m_ip_hdr.check, (int)p_tcphdr->check);
#endif
m_p_send_wqe = &m_not_inline_send_wqe;
m_p_send_wqe->wr_id = (uintptr_t)p_mem_buf_desc;
m_p_ring->send_ring_buffer(m_id, m_p_send_wqe, b_blocked);
}
#ifndef __COVERITY__
struct tcphdr* p_tcp_h = (struct tcphdr*)(((uint8_t*)(&(p_pkt->hdr.m_ip_hdr))+sizeof(p_pkt->hdr.m_ip_hdr)));
NOT_IN_USE(p_tcp_h); /* to supress warning in case VMA_OPTIMIZE_LOG */
dst_tcp_logfunc("Tx TCP segment info: src_port=%d, dst_port=%d, flags='%s%s%s%s%s%s' seq=%u, ack=%u, win=%u, payload_sz=%u",
ntohs(p_tcp_h->source), ntohs(p_tcp_h->dest),
p_tcp_h->urg?"U":"", p_tcp_h->ack?"A":"", p_tcp_h->psh?"P":"",
p_tcp_h->rst?"R":"", p_tcp_h->syn?"S":"", p_tcp_h->fin?"F":"",
ntohl(p_tcp_h->seq), ntohl(p_tcp_h->ack_seq), ntohs(p_tcp_h->window),
total_packet_len- p_tcp_h->doff*4 -34);
#endif
if (unlikely(m_p_tx_mem_buf_desc_list == NULL)) {
m_p_tx_mem_buf_desc_list = m_p_ring->mem_buf_tx_get(m_id, b_blocked, m_n_sysvar_tx_bufs_batch_tcp);
}
return 0;
}
void jmp_to_elf_entry(void *entry, unsigned long buffer, unsigned long size)
{
extern unsigned char _ram_seg, _eram_seg;
unsigned long lb_start, lb_size;
unsigned long adjust, adjusted_boot_notes;
elf_boot_notes.hdr.b_checksum =
compute_ip_checksum(&elf_boot_notes, sizeof(elf_boot_notes));
lb_start = (unsigned long)&_ram_seg;
lb_size = (unsigned long)(&_eram_seg - &_ram_seg);
adjust = buffer + size - lb_start;
adjusted_boot_notes = (unsigned long)&elf_boot_notes;
adjusted_boot_notes += adjust;
printk(BIOS_SPEW, "entry = 0x%08lx\n", (unsigned long)entry);
printk(BIOS_SPEW, "lb_start = 0x%08lx\n", lb_start);
printk(BIOS_SPEW, "lb_size = 0x%08lx\n", lb_size);
printk(BIOS_SPEW, "adjust = 0x%08lx\n", adjust);
printk(BIOS_SPEW, "buffer = 0x%08lx\n", buffer);
printk(BIOS_SPEW, " elf_boot_notes = 0x%08lx\n", (unsigned long)&elf_boot_notes);
printk(BIOS_SPEW, "adjusted_boot_notes = 0x%08lx\n", adjusted_boot_notes);
/* Jump to kernel */
__asm__ __volatile__(
" cld \n\t"
/* Save the callee save registers... */
" pushl %%esi\n\t"
" pushl %%edi\n\t"
" pushl %%ebx\n\t"
/* Save the parameters I was passed */
" pushl $0\n\t" /* 20 adjust */
" pushl %0\n\t" /* 16 lb_start */
" pushl %1\n\t" /* 12 buffer */
" pushl %2\n\t" /* 8 lb_size */
" pushl %3\n\t" /* 4 entry */
" pushl %4\n\t" /* 0 elf_boot_notes */
/* Compute the adjustment */
" xorl %%eax, %%eax\n\t"
" subl 16(%%esp), %%eax\n\t"
" addl 12(%%esp), %%eax\n\t"
" addl 8(%%esp), %%eax\n\t"
" movl %%eax, 20(%%esp)\n\t"
/* Place a copy of coreboot in its new location */
/* Move ``longs'' the coreboot size is 4 byte aligned */
" movl 12(%%esp), %%edi\n\t"
" addl 8(%%esp), %%edi\n\t"
" movl 16(%%esp), %%esi\n\t"
" movl 8(%%esp), %%ecx\n\n"
" shrl $2, %%ecx\n\t"
" rep movsl\n\t"
/* Adjust the stack pointer to point into the new coreboot image */
" addl 20(%%esp), %%esp\n\t"
/* Adjust the instruction pointer to point into the new coreboot image */
" movl $1f, %%eax\n\t"
" addl 20(%%esp), %%eax\n\t"
" jmp *%%eax\n\t"
"1: \n\t"
/* Copy the coreboot bounce buffer over coreboot */
/* Move ``longs'' the coreboot size is 4 byte aligned */
" movl 16(%%esp), %%edi\n\t"
" movl 12(%%esp), %%esi\n\t"
" movl 8(%%esp), %%ecx\n\t"
" shrl $2, %%ecx\n\t"
" rep movsl\n\t"
/* Now jump to the loaded image */
" movl %5, %%eax\n\t"
" movl 0(%%esp), %%ebx\n\t"
" call *4(%%esp)\n\t"
/* The loaded image returned? */
" cli \n\t"
" cld \n\t"
/* Copy the saved copy of coreboot where coreboot runs */
/* Move ``longs'' the coreboot size is 4 byte aligned */
" movl 16(%%esp), %%edi\n\t"
" movl 12(%%esp), %%esi\n\t"
" addl 8(%%esp), %%esi\n\t"
" movl 8(%%esp), %%ecx\n\t"
" shrl $2, %%ecx\n\t"
" rep movsl\n\t"
/* Adjust the stack pointer to point into the old coreboot image */
" subl 20(%%esp), %%esp\n\t"
/* Adjust the instruction pointer to point into the old coreboot image */
" movl $1f, %%eax\n\t"
" subl 20(%%esp), %%eax\n\t"
" jmp *%%eax\n\t"
"1: \n\t"
/* Drop the parameters I was passed */
" addl $24, %%esp\n\t"
/* Restore the callee save registers */
" popl %%ebx\n\t"
" popl %%edi\n\t"
" popl %%esi\n\t"
::
"ri" (lb_start), "ri" (buffer), "ri" (lb_size),
"ri" (entry),
#if CONFIG_MULTIBOOT
"ri"(mbi), "ri" (MB_MAGIC2)
#else
"ri"(adjusted_boot_notes), "ri" (0x0E1FB007)
#endif
);
}
ssize_t dst_entry_tcp::fast_send(const iovec* p_iov, const ssize_t sz_iov, bool is_dummy, bool b_blocked /*= true*/, bool is_rexmit /*= false*/)
{
int ret = 0;
tx_packet_template_t* p_pkt;
mem_buf_desc_t *p_mem_buf_desc;
size_t total_packet_len = 0;
// The header is aligned for fast copy but we need to maintain this diff in order to get the real header pointer easily
size_t hdr_alignment_diff = m_header.m_aligned_l2_l3_len - m_header.m_total_hdr_len;
tcp_iovec* p_tcp_iov = NULL;
bool no_copy = true;
if (likely(sz_iov == 1 && !is_rexmit)) {
p_tcp_iov = (tcp_iovec*)p_iov;
if (unlikely(!m_p_ring->is_active_member(p_tcp_iov->p_desc->p_desc_owner, m_id))) {
no_copy = false;
dst_tcp_logdbg("p_desc=%p wrong desc_owner=%p, this ring=%p. did migration occurred?", p_tcp_iov->p_desc, p_tcp_iov->p_desc->p_desc_owner, m_p_ring);
//todo can we handle this in migration (by going over all buffers lwip hold) instead for every send?
}
} else {
no_copy = false;
}
if (likely(no_copy)) {
p_pkt = (tx_packet_template_t*)((uint8_t*)p_tcp_iov[0].iovec.iov_base - m_header.m_aligned_l2_l3_len);
total_packet_len = p_tcp_iov[0].iovec.iov_len + m_header.m_total_hdr_len;
m_header.copy_l2_ip_hdr(p_pkt);
// We've copied to aligned address, and now we must update p_pkt to point to real
// L2 header
//p_pkt = (tx_packet_template_t*)((uint8_t*)p_pkt + hdr_alignment_diff);
p_pkt->hdr.m_ip_hdr.tot_len = (htons)(p_tcp_iov[0].iovec.iov_len + m_header.m_ip_header_len);
m_sge[0].addr = (uintptr_t)((uint8_t*)p_pkt + hdr_alignment_diff);
m_sge[0].length = total_packet_len;
if (total_packet_len < m_max_inline) { // inline send
m_p_send_wqe = &m_inline_send_wqe;
} else {
m_p_send_wqe = &m_not_inline_send_wqe;
}
m_p_send_wqe->wr_id = (uintptr_t)p_tcp_iov[0].p_desc;
#ifdef VMA_NO_HW_CSUM
p_pkt->hdr.m_ip_hdr.check = 0; // use 0 at csum calculation time
p_pkt->hdr.m_ip_hdr.check = compute_ip_checksum((unsigned short*)&p_pkt->hdr.m_ip_hdr, p_pkt->hdr.m_ip_hdr.ihl * 2);
struct tcphdr* p_tcphdr = (struct tcphdr*)(((uint8_t*)(&(p_pkt->hdr.m_ip_hdr))+sizeof(p_pkt->hdr.m_ip_hdr)));
p_tcphdr->check = 0;
p_tcphdr->check = compute_tcp_checksum(&p_pkt->hdr.m_ip_hdr, (const uint16_t *)p_tcphdr);
dst_tcp_logfine("using SW checksum calculation: p_pkt->hdr.m_ip_hdr.check=%d, p_tcphdr->check=%d", (int)p_pkt->hdr.m_ip_hdr.check, (int)p_tcphdr->check);
#endif
send_lwip_buffer(m_id, m_p_send_wqe, b_blocked, is_dummy);
/* for DEBUG */
if ((uint8_t*)m_sge[0].addr < p_tcp_iov[0].p_desc->p_buffer || (uint8_t*)p_pkt < p_tcp_iov[0].p_desc->p_buffer) {
dst_tcp_logerr("p_buffer - addr=%d, m_total_hdr_len=%zd, p_buffer=%p, type=%d, len=%d, tot_len=%d, payload=%p, hdr_alignment_diff=%zd\n",
(int)(p_tcp_iov[0].p_desc->p_buffer - (uint8_t*)m_sge[0].addr), m_header.m_total_hdr_len,
p_tcp_iov[0].p_desc->p_buffer, p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.type,
p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.len, p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.tot_len,
p_tcp_iov[0].p_desc->lwip_pbuf.pbuf.payload, hdr_alignment_diff);
}
}
else { // We don'nt support inline in this case, since we believe that this a very rare case
p_mem_buf_desc = get_buffer(b_blocked);
if (p_mem_buf_desc == NULL) {
ret = -1;
goto out;
}
m_header.copy_l2_ip_hdr((tx_packet_template_t*)p_mem_buf_desc->p_buffer);
// Actually this is not the real packet len we will subtract the alignment diff at the end of the copy
total_packet_len = m_header.m_aligned_l2_l3_len;
for (int i = 0; i < sz_iov; ++i) {
memcpy(p_mem_buf_desc->p_buffer + total_packet_len, p_iov[i].iov_base, p_iov[i].iov_len);
total_packet_len += p_iov[i].iov_len;
}
m_sge[0].addr = (uintptr_t)(p_mem_buf_desc->p_buffer + hdr_alignment_diff);
m_sge[0].length = total_packet_len - hdr_alignment_diff;
// LKey will be updated in ring->send() // m_sge[0].lkey = p_mem_buf_desc->lkey;
p_pkt = (tx_packet_template_t*)((uint8_t*)p_mem_buf_desc->p_buffer);
p_pkt->hdr.m_ip_hdr.tot_len = (htons)(m_sge[0].length - m_header.m_transport_header_len);
#ifdef VMA_NO_HW_CSUM
p_pkt->hdr.m_ip_hdr.check = 0; // use 0 at csum calculation time
p_pkt->hdr.m_ip_hdr.check = compute_ip_checksum((unsigned short*)&p_pkt->hdr.m_ip_hdr, p_pkt->hdr.m_ip_hdr.ihl * 2);
struct tcphdr* p_tcphdr = (struct tcphdr*)(((uint8_t*)(&(p_pkt->hdr.m_ip_hdr))+sizeof(p_pkt->hdr.m_ip_hdr)));
p_tcphdr->check = 0;
p_tcphdr->check = compute_tcp_checksum(&p_pkt->hdr.m_ip_hdr, (const uint16_t *)p_tcphdr);
dst_tcp_logfine("using SW checksum calculation: p_pkt->hdr.m_ip_hdr.check=%d, p_tcphdr->check=%d", (int)p_pkt->hdr.m_ip_hdr.check, (int)p_tcphdr->check);
#endif
m_p_send_wqe = &m_not_inline_send_wqe;
m_p_send_wqe->wr_id = (uintptr_t)p_mem_buf_desc;
vma_wr_tx_packet_attr attr = (vma_wr_tx_packet_attr)((VMA_TX_PACKET_BLOCK*b_blocked) |
(VMA_TX_PACKET_DUMMY*is_dummy) |
VMA_TX_PACKET_L3_CSUM |
VMA_TX_PACKET_L4_CSUM);
send_ring_buffer(m_id, m_p_send_wqe, attr);
/* for DEBUG */
if ((uint8_t*)m_sge[0].addr < p_mem_buf_desc->p_buffer) {
dst_tcp_logerr("p_buffer - addr=%d, m_total_hdr_len=%zd, p_buffer=%p, type=%d, len=%d, tot_len=%d, payload=%p, hdr_alignment_diff=%zd\n",
(int)(p_mem_buf_desc->p_buffer - (uint8_t*)m_sge[0].addr), m_header.m_total_hdr_len,
p_mem_buf_desc->p_buffer, p_mem_buf_desc->lwip_pbuf.pbuf.type,
p_mem_buf_desc->lwip_pbuf.pbuf.len, p_mem_buf_desc->lwip_pbuf.pbuf.tot_len,
p_mem_buf_desc->lwip_pbuf.pbuf.payload, hdr_alignment_diff);
}
}
if (unlikely(m_p_tx_mem_buf_desc_list == NULL)) {
m_p_tx_mem_buf_desc_list = m_p_ring->mem_buf_tx_get(m_id, b_blocked, m_n_sysvar_tx_bufs_batch_tcp);
}
out:
if (unlikely(is_rexmit)) {
m_p_ring->inc_tx_retransmissions(m_id);
}
return ret;
}
static int sb_eth_send(struct udevice *dev, void *packet, int length)
{
struct eth_sandbox_priv *priv = dev_get_priv(dev);
struct ethernet_hdr *eth = packet;
debug("eth_sandbox: Send packet %d\n", length);
if (dev->seq >= 0 && dev->seq < ARRAY_SIZE(disabled) &&
disabled[dev->seq])
return 0;
if (ntohs(eth->et_protlen) == PROT_ARP) {
struct arp_hdr *arp = packet + ETHER_HDR_SIZE;
if (ntohs(arp->ar_op) == ARPOP_REQUEST) {
struct ethernet_hdr *eth_recv;
struct arp_hdr *arp_recv;
/* store this as the assumed IP of the fake host */
priv->fake_host_ipaddr = net_read_ip(&arp->ar_tpa);
/* Formulate a fake response */
eth_recv = (void *)priv->recv_packet_buffer;
memcpy(eth_recv->et_dest, eth->et_src, ARP_HLEN);
memcpy(eth_recv->et_src, priv->fake_host_hwaddr,
ARP_HLEN);
eth_recv->et_protlen = htons(PROT_ARP);
arp_recv = (void *)priv->recv_packet_buffer +
ETHER_HDR_SIZE;
arp_recv->ar_hrd = htons(ARP_ETHER);
arp_recv->ar_pro = htons(PROT_IP);
arp_recv->ar_hln = ARP_HLEN;
arp_recv->ar_pln = ARP_PLEN;
arp_recv->ar_op = htons(ARPOP_REPLY);
memcpy(&arp_recv->ar_sha, priv->fake_host_hwaddr,
ARP_HLEN);
net_write_ip(&arp_recv->ar_spa, priv->fake_host_ipaddr);
memcpy(&arp_recv->ar_tha, &arp->ar_sha, ARP_HLEN);
net_copy_ip(&arp_recv->ar_tpa, &arp->ar_spa);
priv->recv_packet_length = ETHER_HDR_SIZE +
ARP_HDR_SIZE;
}
} else if (ntohs(eth->et_protlen) == PROT_IP) {
struct ip_udp_hdr *ip = packet + ETHER_HDR_SIZE;
if (ip->ip_p == IPPROTO_ICMP) {
struct icmp_hdr *icmp = (struct icmp_hdr *)&ip->udp_src;
if (icmp->type == ICMP_ECHO_REQUEST) {
struct ethernet_hdr *eth_recv;
struct ip_udp_hdr *ipr;
struct icmp_hdr *icmpr;
/* reply to the ping */
memcpy(priv->recv_packet_buffer, packet,
length);
eth_recv = (void *)priv->recv_packet_buffer;
ipr = (void *)priv->recv_packet_buffer +
ETHER_HDR_SIZE;
icmpr = (struct icmp_hdr *)&ipr->udp_src;
memcpy(eth_recv->et_dest, eth->et_src,
ARP_HLEN);
memcpy(eth_recv->et_src, priv->fake_host_hwaddr,
ARP_HLEN);
ipr->ip_sum = 0;
ipr->ip_off = 0;
net_copy_ip((void *)&ipr->ip_dst, &ip->ip_src);
net_write_ip((void *)&ipr->ip_src,
priv->fake_host_ipaddr);
ipr->ip_sum = compute_ip_checksum(ipr,
IP_HDR_SIZE);
icmpr->type = ICMP_ECHO_REPLY;
icmpr->checksum = 0;
icmpr->checksum = compute_ip_checksum(icmpr,
ICMP_HDR_SIZE);
priv->recv_packet_length = length;
}
}
}
return 0;
}
void sr_handlepacket(struct sr_instance* sr,
uint8_t* packet/* lent */,
unsigned int len,
char* interface/* lent */)
{
/* REQUIRES */
assert(sr);
assert(packet);
assert(interface);
printf("*** -> Received packet of length %d \n",len);
if( len < ETHER_HDR_LEN ) {
printf("Bogus packet length\n");
return;
}
struct sr_if *iface;
struct frame_t *incoming = create_frame_t(sr, packet, len, interface);
struct frame_t *outgoing = NULL;
/* First, deal with ARP cache timeouts */
arp_cache_flush(&sr_arp_cache);
arp_queue_flush(sr, &sr_arp_queue);
/* Do we only need to cache ARP replies, or src MAC/IP on regular IP packets too, etc? */
/* Also, do we need to worry about fragmentation? */
/* Then actually handle the packet */
/* Start by determining protocol */
if (incoming->ip_header){
/* sanity checks */
if ( incoming->ip_header->ip_v != 4 ){
printf("IP packet not IPv4\n");
return;
}
compute_ip_checksum(incoming);
/* Check the checksum */
if( incoming->ip_header->ip_sum != 0 ) {
fprintf(stderr, "IP checksum incorrect, packet was dropped\n");
return;
}
//set checksum back to what it was
compute_ip_checksum(incoming);
/* Are we the destination? */
if (iface = if_dst_check(sr, incoming->to_ip)){ //we could change this to just take incoming and then get to_ip
/* Is this an ICMP packet? */
if (incoming->icmp_header){
printf("received ICMP datagram\n");
compute_icmp_checksum(incoming);
if(incoming->icmp_header->icmp_type == ECHO_REQUEST && incoming->icmp_header->icmp_sum == 0){
outgoing = generate_icmp_echo(incoming);
printf("received ICMP echo request\n");
}
else
printf("Dropped packet--we don't deal with that code, or invalid checksum\n");
}
else{
outgoing = generate_icmp_error(incoming, DEST_UNREACH, PORT_UNREACH);
printf("A packet for me! Flattering, but wrong.\n");
}
}
else {
/* Has it timed out? */
if (incoming->ip_header->ip_ttl <= 0){
int err = 0;
//make sure it's not an ICMP error packet alrady
if (incoming->ip_header->ip_p == IPPROTO_ICMP){
incoming->icmp_header = ((void *) incoming->ip_header +
incoming->ip_hl);
uint8_t code = incoming->icmp_header->icmp_type;
//don't send ICMP error messages about ICMP error messages
if (code == DEST_UNREACH || code == TIME_EXCEEDED || code == 12 || code == 31)
//12 and 31 indicate bad IP header and datagram conversion error, respectively
err = 1;
}
if (!err){
outgoing = generate_icmp_error(incoming, TIME_EXCEEDED, TIME_INTRANSIT);
printf("Slowpoke. TTL exceeded.\n");
}
}
else {
/* update and forward packet; if necessary, add it to queue */
struct arpc_entry *incache;
uint32_t arpc_ip;
outgoing = update_ip_hdr(sr, incoming, &arpc_ip);
incache = arp_cache_lookup(sr_arp_cache.first, arpc_ip);
if (!incache){
struct arpq_entry *entry = arp_queue_lookup(sr_arp_queue.first, outgoing->to_ip);
struct arpq_entry *temp_entry = NULL;
if ( entry ){ //if we've already sent an ARP request about this IP
if( time(NULL) - 1 > entry->arpq_last_req ) {
if(entry->arpq_num_reqs >= ARP_MAX_REQ)
{
printf("Too many ARP requests\n");
arpq_packets_icmpsend(sr, &entry->arpq_packets);
destroy_arpq_entry(entry);
return;
}
else if (entry->arpq_packets.first)
{
struct frame_t *arp_req;
struct queued_packet *old_packet = entry->arpq_packets.first;
entry->arpq_last_req = time(NULL);
entry->arpq_num_reqs++;
}
}
assert( (entry->arpq_packets).first );
if (!arpq_add_packet(entry, outgoing, len, incoming->from_MAC, incoming->iface))
printf("ARP queue packet add failed\n");
else printf("added packet to queue\n");
}
/* else, there are no outstanding ARP requests for this particular IP */
else {
printf("outgoing ip is %d\n", ntohl(outgoing->to_ip));
temp_entry = arpq_add_entry(&sr_arp_queue, outgoing->iface, outgoing, outgoing->to_ip, outgoing->ip_len, incoming->from_MAC, incoming->iface);
}
free(outgoing->frame);
/* make ARP request */
outgoing = arp_create(sr, outgoing, outgoing->iface, ARP_REQUEST); //send datagram will now point to an ARP packet, not to the IP datagram
if (temp_entry){
temp_entry->arpq_next_hop = outgoing->to_ip;
}
printf("sending ARP request\n");
}
else{
printf("got the MAC, can actually send this packet\n");
memcpy(outgoing->to_MAC, incache->arpc_mac, ETHER_ADDR_LEN);
encapsulate(outgoing);
}
}
}
}
else if ( incoming->arp_header )
{
printf("received ARP packet\n");
struct sr_arphdr *arp_header = incoming->arp_header;
uint8_t in_cache = 0;
struct arpc_entry *arpc_ent = arp_cache_lookup(sr_arp_cache.first, arp_header->ar_sip);
printf("checking the cache\n");
if( arpc_ent )
{
arp_cache_update(arpc_ent, arp_header->ar_sha);
printf("updated cache\n");
in_cache = 1;
}
struct sr_if *target_if = if_dst_check(sr, arp_header->ar_tip);
printf("checking the target\n");
if( target_if )
{
printf("It's for us\n");
if( !in_cache ) {
if( arp_cache_add(&sr_arp_cache, arp_header->ar_sha, arp_header->ar_sip) ) {
printf("added to cache\n");
printf("ip is %d\n", ntohl(arp_header->ar_sip));
struct arpq_entry *new_ent;
if( new_ent = arpq_next_hop_lookup(sr_arp_queue.first, arp_header->ar_sip) )
arpq_entry_clear(sr, &sr_arp_queue, new_ent, arp_header->ar_sha);
}
else perror("ARP request not added to cache");
}
if( ntohs(arp_header->ar_op) == ARP_REQUEST ){
outgoing = arp_create(sr, incoming, incoming->iface, ARP_REPLY);
printf("created ARP reply\n");
assert(outgoing);
}
}
}
else perror("Unknown protocol");
//send datagram, if appropriate
if (outgoing != NULL){
sr_send_packet(sr, (uint8_t *)outgoing->frame, outgoing->len, outgoing->iface->name);
printf("sent packet of length %d on iface %s\n", outgoing->len, outgoing->iface->name);
}
if (outgoing != NULL) destroy_frame_t(outgoing);
}/* end sr_handlepacket */
/****************************************************************************
* lbtable_scan
*
* Scan the chunk of memory specified by 'start' and 'end' for a coreboot
* table. The first 4 bytes of the table are marked by the signature
* { 'L', 'B', 'I', 'O' }. 'start' and 'end' indicate the addresses of the
* first and last bytes of the chunk of memory to be scanned. For instance,
* values of 0x10000000 and 0x1000ffff for 'start' and 'end' specify a 64k
* chunk of memory starting at address 0x10000000. 'start' and 'end' are
* physical addresses.
*
* If a coreboot table is found, return a pointer to it. Otherwise return
* NULL. On return, *bad_header_count and *bad_table_count are set as
* follows:
*
* *bad_header_count:
* Indicates the number of times in which a valid signature was found
* but the header checksum was invalid.
*
* *bad_table_count:
* Indicates the number of times in which a header with a valid
* checksum was found but the table checksum was invalid.
****************************************************************************/
static const struct lb_header *lbtable_scan(unsigned long start,
unsigned long end,
int *bad_header_count,
int *bad_table_count)
{
static const char signature[4] = { 'L', 'B', 'I', 'O' };
const struct lb_header *table;
const struct lb_forward *forward;
unsigned long p;
uint32_t sig;
assert(end >= start);
memcpy(&sig, signature, sizeof(sig));
table = NULL;
*bad_header_count = 0;
*bad_table_count = 0;
/* Look for signature. Table is aligned on 16-byte boundary. Therefore
* only check every fourth 32-bit memory word. As the loop is coded below,
* this function will behave in a reasonable manner for ALL possible values
* for 'start' and 'end': even weird boundary cases like 0x00000000 and
* 0xffffffff on a 32-bit architecture.
*/
map_pages(start, end - start);
for (p = start;
(p <= end) &&
(end - p >= (sizeof(uint32_t) - 1)); p += 4) {
if (*(uint32_t*)phystov(p) != sig)
continue;
/* We found a valid signature. */
table = (const struct lb_header *)phystov(p);
/* validate header checksum */
if (compute_ip_checksum((void *)table, sizeof(*table))) {
(*bad_header_count)++;
continue;
}
map_pages(p, table->table_bytes + sizeof(*table));
/* validate table checksum */
if (table->table_checksum !=
compute_ip_checksum(((char *)table) + sizeof(*table),
table->table_bytes)) {
(*bad_table_count)++;
continue;
}
/* checksums are ok: we found it! */
/* But it may just be a forwarding table, so look if there's a forwarder */
lbtable = table;
forward = (struct lb_forward *)find_lbrec(LB_TAG_FORWARD);
lbtable = NULL;
if (forward) {
uint64_t new_phys = forward->forward;
table = lbtable_scan(new_phys, new_phys + getpagesize(),
bad_header_count, bad_table_count);
}
return table;
}
return NULL;
}
/*
* HELPER function called from arp_thread
*/
void process_arp_queue(struct sr_instance* sr) {
router_state* rs = get_router_state(sr);
node* n = get_router_state(sr)->arp_queue;
node* next = NULL;
time_t now;
double diff;
while (n) {
next = n->next;
arp_queue_entry* aqe = (arp_queue_entry*)n->data;
/* has it been over a second since the last arp request was sent? */
time(&now);
diff = difftime(now, aqe->last_req_time);
if (diff > 1) {
/* have we sent less than 5 arp requests? */
if (aqe->requests < 5) {
/* send another */
time(&(aqe->last_req_time));
++(aqe->requests);
send_arp_request(sr, aqe->next_hop.s_addr, aqe->out_iface_name);
} else {
/* we have exceeded the max arp requests, return packets to sender */
node* cur_packet_node = aqe->head;
node* next_packet_node = NULL;
while (cur_packet_node) {
/* send icmp for the packet, free it, and its encasing entry */
arp_queue_packet_entry* aqpe = (arp_queue_packet_entry*)cur_packet_node->data;
/* only send an icmp error if the packet is not icmp, or if it is, its an echo request or reply
* also ensure we don't send an icmp error back to one of our interfaces
*/
if ((get_ip_hdr(aqpe->packet, aqpe->len)->ip_p != IP_PROTO_ICMP) ||
(get_icmp_hdr(aqpe->packet, aqpe->len)->icmp_type == ICMP_TYPE_ECHO_REPLY) ||
(get_icmp_hdr(aqpe->packet, aqpe->len)->icmp_type == ICMP_TYPE_ECHO_REQUEST)) {
/* also ensure we don't send an icmp error back to one of our interfaces */
if (!iface_match_ip(rs, get_ip_hdr(aqpe->packet, aqpe->len)->ip_src.s_addr)) {
/* Total hack here to increment the TTL since we already decremented it earlier in the pipeline
* and the ICMP error should return the original packet.
* TODO: Don't decrement the TTL until the packet is ready to be put on the wire
* and we have the next hop ARP address, although checking should be done
* where it is currently being decremented to minimize effort on a doomed packet */
ip_hdr *ip = get_ip_hdr(aqpe->packet, aqpe->len);
if (ip->ip_ttl < 255) {
ip->ip_ttl++;
/* recalculate checksum */
bzero(&ip->ip_sum, sizeof(uint16_t));
uint16_t checksum = htons(compute_ip_checksum(ip));
ip->ip_sum = checksum;
}
send_icmp_packet(sr, aqpe->packet, aqpe->len, ICMP_TYPE_DESTINATION_UNREACHABLE, ICMP_CODE_HOST_UNREACHABLE);
}
}
free(aqpe->packet);
next_packet_node = cur_packet_node->next;
//free(cur_packet_node); /* IS THIS CORRECT TO FREE IT ? */
node_remove(&(aqe->head), cur_packet_node);
cur_packet_node = next_packet_node;
}
/* free the arp queue entry for this destination ip, and patch the list */
node_remove(&(get_router_state(sr)->arp_queue), n);
}
}
n = next;
}
}
static int cb_parse_header(void *addr, int len, struct sysinfo_t *info)
{
struct cb_header *header;
unsigned char *ptr = (unsigned char *)addr;
int i;
for (i = 0; i < len; i += 16, ptr += 16) {
header = (struct cb_header *)ptr;
if (!strncmp((const char *)header->signature, "LBIO", 4))
break;
}
/* We walked the entire space and didn't find anything. */
if (i >= len)
return -1;
if (!header->table_bytes)
return 0;
/* Make sure the checksums match. */
if (!ip_checksum_ok(header, sizeof(*header)))
return -1;
if (compute_ip_checksum(ptr + sizeof(*header), header->table_bytes) !=
header->table_checksum)
return -1;
/* Now, walk the tables. */
ptr += header->header_bytes;
/* Inintialize some fields to sentinel values. */
info->vbnv_start = info->vbnv_size = (uint32_t)(-1);
for (i = 0; i < header->table_entries; i++) {
struct cb_record *rec = (struct cb_record *)ptr;
/* We only care about a few tags here (maybe more later). */
switch (rec->tag) {
case CB_TAG_FORWARD:
return cb_parse_header(
(void *)(unsigned long)
((struct cb_forward *)rec)->forward,
len, info);
continue;
case CB_TAG_MEMORY:
cb_parse_memory(ptr, info);
break;
case CB_TAG_SERIAL:
cb_parse_serial(ptr, info);
break;
case CB_TAG_VERSION:
cb_parse_string(ptr, &info->version);
break;
case CB_TAG_EXTRA_VERSION:
cb_parse_string(ptr, &info->extra_version);
break;
case CB_TAG_BUILD:
cb_parse_string(ptr, &info->build);
break;
case CB_TAG_COMPILE_TIME:
cb_parse_string(ptr, &info->compile_time);
break;
case CB_TAG_COMPILE_BY:
cb_parse_string(ptr, &info->compile_by);
break;
case CB_TAG_COMPILE_HOST:
cb_parse_string(ptr, &info->compile_host);
break;
case CB_TAG_COMPILE_DOMAIN:
cb_parse_string(ptr, &info->compile_domain);
break;
case CB_TAG_COMPILER:
cb_parse_string(ptr, &info->compiler);
break;
case CB_TAG_LINKER:
cb_parse_string(ptr, &info->linker);
break;
case CB_TAG_ASSEMBLER:
cb_parse_string(ptr, &info->assembler);
break;
/*
* FIXME we should warn on serial if coreboot set up a
* framebuffer buf the payload does not know about it.
*/
case CB_TAG_FRAMEBUFFER:
cb_parse_framebuffer(ptr, info);
break;
case CB_TAG_GPIO:
cb_parse_gpios(ptr, info);
break;
case CB_TAG_VDAT:
cb_parse_vdat(ptr, info);
break;
case CB_TAG_TIMESTAMPS:
cb_parse_tstamp(ptr, info);
break;
case CB_TAG_CBMEM_CONSOLE:
cb_parse_cbmem_cons(ptr, info);
break;
case CB_TAG_VBNV:
cb_parse_vbnv(ptr, info);
break;
}
ptr += rec->size;
}
return 1;
}
/****************************************************************************
* lbtable_scan
*
* Scan the chunk of memory specified by 'start' and 'end' for a coreboot
* table. The first 4 bytes of the table are marked by the signature
* { 'L', 'B', 'I', 'O' }. 'start' and 'end' indicate the addresses of the
* first and last bytes of the chunk of memory to be scanned. For instance,
* values of 0x10000000 and 0x1000ffff for 'start' and 'end' specify a 64k
* chunk of memory starting at address 0x10000000. 'start' and 'end' are
* virtual addresses in the address space of the current process. They
* represent a chunk of memory obtained by calling mmap() on /dev/mem.
*
* If a coreboot table is found, return a pointer to it. Otherwise return
* NULL. On return, *bad_header_count and *bad_table_count are set as
* follows:
*
* *bad_header_count:
* Indicates the number of times in which a valid signature was found
* but the header checksum was invalid.
*
* *bad_table_count:
* Indicates the number of times in which a header with a valid
* checksum was found but the table checksum was invalid.
****************************************************************************/
static const struct lb_header *lbtable_scan(unsigned long start,
unsigned long end,
int *bad_header_count,
int *bad_table_count)
{
static const char signature[] = { 'L', 'B', 'I', 'O' };
const struct lb_header *table;
const struct lb_forward *forward;
const uint32_t *p;
uint32_t sig;
assert(end >= start);
sig = (*((const uint32_t *)signature));
table = NULL;
*bad_header_count = 0;
*bad_table_count = 0;
/* Look for signature. Table is aligned on 16-byte boundary. Therefore
* only check every fourth 32-bit memory word. As the loop is coded below,
* this function will behave in a reasonable manner for ALL possible values
* for 'start' and 'end': even weird boundary cases like 0x00000000 and
* 0xffffffff on a 32-bit architecture.
*/
for (p = (const uint32_t *)start;
(((unsigned long)p) <= end) &&
((end - (unsigned long)p) >= (sizeof(uint32_t) - 1)); p += 4) {
if (*p != sig)
continue;
/* We found a valid signature. */
table = (const struct lb_header *)p;
/* validate header checksum */
if (compute_ip_checksum((void *)table, sizeof(*table))) {
(*bad_header_count)++;
continue;
}
/* validate table checksum */
if (table->table_checksum !=
compute_ip_checksum(((char *)table) + sizeof(*table),
table->table_bytes)) {
(*bad_table_count)++;
continue;
}
/* checksums are ok: we found it! */
/* But it may just be a forwarding table, so look if there's a forwarder */
lbtable = table;
forward = (struct lb_forward *)find_lbrec(LB_TAG_FORWARD);
lbtable = NULL;
if (forward) {
uint64_t new_phys = forward->forward;
new_phys &= ~(getpagesize() - 1);
munmap((void *)low_phys_mem, BYTES_TO_MAP);
if ((low_phys_mem =
mmap(NULL, BYTES_TO_MAP, PROT_READ, MAP_SHARED, fd,
(off_t) new_phys)) == MAP_FAILED) {
fprintf(stderr,
"%s: Failed to mmap /dev/mem: %s\n",
prog_name, strerror(errno));
exit(1);
}
low_phys_base = new_phys;
table =
lbtable_scan(phystov(low_phys_base),
phystov(low_phys_base + BYTES_TO_MAP),
bad_header_count, bad_table_count);
}
return table;
}
return NULL;
}
