blob: 5ee8004326304d9b34bbd87fc8f7a212b1b6e651 [file] [log] [blame]
/*
* Copyright (c) 1993, 1994, 1995, 1996, 1997
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that: (1) source code distributions
* retain the above copyright notice and this paragraph in its entirety, (2)
* distributions including binary code include the above copyright notice and
* this paragraph in its entirety in the documentation or other materials
* provided with the distribution, and (3) all advertising materials mentioning
* features or use of this software display the following acknowledgement:
* ``This product includes software developed by the University of California,
* Lawrence Berkeley Laboratory and its contributors.'' Neither the name of
* the University nor the names of its contributors may be used to endorse
* or promote products derived from this software without specific prior
* written permission.
* THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED
* WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
*
* pcap-util.c - common code for various files
*/
#ifdef HAVE_CONFIG_H
#include <config.h>
#endif
#include <pcap-types.h>
#include "pcap/can_socketcan.h"
#include "pcap/sll.h"
#include "pcap/usb.h"
#include "pcap/nflog.h"
#include "pcap-int.h"
#include "extract.h"
#include "pcap-usb-linux-common.h"
#include "pcap-util.h"
#include "pflog.h"
/*
* Most versions of the DLT_PFLOG pseudo-header have UID and PID fields
* that are saved in host byte order.
*
* When reading a DLT_PFLOG packet, we need to convert those fields from
* the byte order of the host that wrote the file to this host's byte
* order.
*/
static void
swap_pflog_header(const struct pcap_pkthdr *hdr, u_char *buf)
{
u_int caplen = hdr->caplen;
u_int length = hdr->len;
u_int pfloghdr_length;
struct pfloghdr *pflhdr = (struct pfloghdr *)buf;
if (caplen < (u_int) (offsetof(struct pfloghdr, uid) + sizeof pflhdr->uid) ||
length < (u_int) (offsetof(struct pfloghdr, uid) + sizeof pflhdr->uid)) {
/* Not enough data to have the uid field */
return;
}
pfloghdr_length = pflhdr->length;
if (pfloghdr_length < (u_int) (offsetof(struct pfloghdr, uid) + sizeof pflhdr->uid)) {
/* Header doesn't include uid field */
return;
}
pflhdr->uid = SWAPLONG(pflhdr->uid);
if (caplen < (u_int) (offsetof(struct pfloghdr, pid) + sizeof pflhdr->pid) ||
length < (u_int) (offsetof(struct pfloghdr, pid) + sizeof pflhdr->pid)) {
/* Not enough data to have the pid field */
return;
}
if (pfloghdr_length < (u_int) (offsetof(struct pfloghdr, pid) + sizeof pflhdr->pid)) {
/* Header doesn't include pid field */
return;
}
pflhdr->pid = SWAPLONG(pflhdr->pid);
if (caplen < (u_int) (offsetof(struct pfloghdr, rule_uid) + sizeof pflhdr->rule_uid) ||
length < (u_int) (offsetof(struct pfloghdr, rule_uid) + sizeof pflhdr->rule_uid)) {
/* Not enough data to have the rule_uid field */
return;
}
if (pfloghdr_length < (u_int) (offsetof(struct pfloghdr, rule_uid) + sizeof pflhdr->rule_uid)) {
/* Header doesn't include rule_uid field */
return;
}
pflhdr->rule_uid = SWAPLONG(pflhdr->rule_uid);
if (caplen < (u_int) (offsetof(struct pfloghdr, rule_pid) + sizeof pflhdr->rule_pid) ||
length < (u_int) (offsetof(struct pfloghdr, rule_pid) + sizeof pflhdr->rule_pid)) {
/* Not enough data to have the rule_pid field */
return;
}
if (pfloghdr_length < (u_int) (offsetof(struct pfloghdr, rule_pid) + sizeof pflhdr->rule_pid)) {
/* Header doesn't include rule_pid field */
return;
}
pflhdr->rule_pid = SWAPLONG(pflhdr->rule_pid);
}
/*
* Linux cooked capture packets with a protocol type of LINUX_SLL_P_CAN or
* LINUX_SLL_P_CANFD have SocketCAN CAN classic/CAN FD headers in front
* of the payload,with the CAN ID being in the byte order of the host
* that wrote the packet, and Linux cooked capture packets with a protocol
* type of LINUX_SLL_P_CANXL have SocketCAN CAN XL headers in front of the
* payload with the protocol/VCID field, the payload length, and the
* acceptance field in the byte order of the host that wrote the packet.
*
* When reading a Linux cooked capture packet, we need to check for those
* packets and, if the byte order host that wrote the packet, as
* indicated by the byte order of the pcap file or pcapng section
* containing the packet, is the opposite of our byte order, convert
* the header files to our byte order by byte-swapping them.
*/
static void
swap_socketcan_header(uint16_t protocol, u_int caplen, u_int length,
u_char *buf)
{
pcap_can_socketcan_hdr *hdrp;
pcap_can_socketcan_xl_hdr *xl_hdrp;
switch (protocol) {
case LINUX_SLL_P_CAN:
case LINUX_SLL_P_CANFD:
/*
* CAN classic/CAN FD packet; fix up the packet's header
* by byte-swapping the CAN ID field.
*/
hdrp = (pcap_can_socketcan_hdr *)buf;
if (caplen < (u_int) (offsetof(pcap_can_socketcan_hdr, can_id) + sizeof hdrp->can_id) ||
length < (u_int) (offsetof(pcap_can_socketcan_hdr, can_id) + sizeof hdrp->can_id)) {
/* Not enough data to have the can_id field */
return;
}
hdrp->can_id = SWAPLONG(hdrp->can_id);
break;
case LINUX_SLL_P_CANXL:
/*
* CAN XL packet; fix up the packet's header by
* byte-swapping the priority/VCID field, the
* payload length, and the acceptance field.
*/
xl_hdrp = (pcap_can_socketcan_xl_hdr *)buf;
if (caplen < (u_int) (offsetof(pcap_can_socketcan_xl_hdr, priority_vcid) + sizeof xl_hdrp->priority_vcid) ||
length < (u_int) (offsetof(pcap_can_socketcan_xl_hdr, priority_vcid) + sizeof xl_hdrp->priority_vcid)) {
/* Not enough data to have the priority_vcid field */
return;
}
xl_hdrp->priority_vcid = SWAPLONG(xl_hdrp->priority_vcid);
if (caplen < (u_int) (offsetof(pcap_can_socketcan_xl_hdr, payload_length) + sizeof xl_hdrp->payload_length) ||
length < (u_int) (offsetof(pcap_can_socketcan_xl_hdr, payload_length) + sizeof xl_hdrp->payload_length)) {
/* Not enough data to have the payload_length field */
return;
}
xl_hdrp->payload_length = SWAPSHORT(xl_hdrp->payload_length);
if (caplen < (u_int) (offsetof(pcap_can_socketcan_xl_hdr, acceptance_field) + sizeof xl_hdrp->acceptance_field) ||
length < (u_int) (offsetof(pcap_can_socketcan_xl_hdr, acceptance_field) + sizeof xl_hdrp->acceptance_field)) {
/* Not enough data to have the acceptance_field field */
return;
}
xl_hdrp->acceptance_field = SWAPLONG(xl_hdrp->acceptance_field);
break;
default:
/*
* Not a CAN packet; nothing to do.
*/
break;
}
}
/*
* DLT_LINUX_SLL packets with a protocol type of LINUX_SLL_P_CAN or
* LINUX_SLL_P_CANFD have SocketCAN headers in front of the payload,
* with the CAN ID being in host byte order.
*
* When reading a DLT_LINUX_SLL packet, we need to check for those
* packets and convert the CAN ID from the byte order of the host that
* wrote the file to this host's byte order.
*/
static void
swap_linux_sll_socketcan_header(const struct pcap_pkthdr *hdr, u_char *buf)
{
u_int caplen = hdr->caplen;
u_int length = hdr->len;
struct sll_header *shdr = (struct sll_header *)buf;
if (caplen < (u_int) sizeof(struct sll_header) ||
length < (u_int) sizeof(struct sll_header)) {
/* Not enough data to have the protocol field */
return;
}
/*
* Byte-swap what needs to be byte-swapped.
*/
swap_socketcan_header(EXTRACT_BE_U_2(&shdr->sll_protocol),
caplen - (u_int) sizeof(struct sll_header),
length - (u_int) sizeof(struct sll_header),
buf + sizeof(struct sll_header));
}
/*
* The same applies for DLT_LINUX_SLL2.
*/
static void
swap_linux_sll2_socketcan_header(const struct pcap_pkthdr *hdr, u_char *buf)
{
u_int caplen = hdr->caplen;
u_int length = hdr->len;
struct sll2_header *shdr = (struct sll2_header *)buf;
if (caplen < (u_int) sizeof(struct sll2_header) ||
length < (u_int) sizeof(struct sll2_header)) {
/* Not enough data to have the protocol field */
return;
}
/*
* Byte-swap what needs to be byte-swapped.
*/
swap_socketcan_header(EXTRACT_BE_U_2(&shdr->sll2_protocol),
caplen - (u_int) sizeof(struct sll2_header),
length - (u_int) sizeof(struct sll2_header),
buf + sizeof(struct sll2_header));
}
/*
* The DLT_USB_LINUX and DLT_USB_LINUX_MMAPPED headers are in host
* byte order when capturing (it's supplied directly from a
* memory-mapped buffer shared by the kernel).
*
* When reading a DLT_USB_LINUX or DLT_USB_LINUX_MMAPPED packet, we
* need to convert it from the byte order of the host that wrote the
* file to this host's byte order.
*/
static void
swap_linux_usb_header(const struct pcap_pkthdr *hdr, u_char *buf,
int header_len_64_bytes)
{
pcap_usb_header_mmapped *uhdr = (pcap_usb_header_mmapped *)buf;
bpf_u_int32 offset = 0;
/*
* "offset" is the offset *past* the field we're swapping;
* we skip the field *before* checking to make sure
* the captured data length includes the entire field.
*/
/*
* The URB id is a totally opaque value; do we really need to
* convert it to the reading host's byte order???
*/
offset += 8; /* skip past id */
if (hdr->caplen < offset)
return;
uhdr->id = SWAPLL(uhdr->id);
offset += 4; /* skip past various 1-byte fields */
offset += 2; /* skip past bus_id */
if (hdr->caplen < offset)
return;
uhdr->bus_id = SWAPSHORT(uhdr->bus_id);
offset += 2; /* skip past various 1-byte fields */
offset += 8; /* skip past ts_sec */
if (hdr->caplen < offset)
return;
uhdr->ts_sec = SWAPLL(uhdr->ts_sec);
offset += 4; /* skip past ts_usec */
if (hdr->caplen < offset)
return;
uhdr->ts_usec = SWAPLONG(uhdr->ts_usec);
offset += 4; /* skip past status */
if (hdr->caplen < offset)
return;
uhdr->status = SWAPLONG(uhdr->status);
offset += 4; /* skip past urb_len */
if (hdr->caplen < offset)
return;
uhdr->urb_len = SWAPLONG(uhdr->urb_len);
offset += 4; /* skip past data_len */
if (hdr->caplen < offset)
return;
uhdr->data_len = SWAPLONG(uhdr->data_len);
if (uhdr->transfer_type == URB_ISOCHRONOUS) {
offset += 4; /* skip past s.iso.error_count */
if (hdr->caplen < offset)
return;
uhdr->s.iso.error_count = SWAPLONG(uhdr->s.iso.error_count);
offset += 4; /* skip past s.iso.numdesc */
if (hdr->caplen < offset)
return;
uhdr->s.iso.numdesc = SWAPLONG(uhdr->s.iso.numdesc);
} else
offset += 8; /* skip USB setup header */
/*
* With the old header, there are no isochronous descriptors
* after the header.
*
* With the new header, the actual number of descriptors in
* the header is not s.iso.numdesc, it's ndesc - only the
* first N descriptors, for some value of N, are put into
* the header, and ndesc is set to the actual number copied.
* In addition, if s.iso.numdesc is negative, no descriptors
* are captured, and ndesc is set to 0.
*/
if (header_len_64_bytes) {
/*
* This is either the "version 1" header, with
* 16 bytes of additional fields at the end, or
* a "version 0" header from a memory-mapped
* capture, with 16 bytes of zeroed-out padding
* at the end. Byte swap them as if this were
* a "version 1" header.
*/
offset += 4; /* skip past interval */
if (hdr->caplen < offset)
return;
uhdr->interval = SWAPLONG(uhdr->interval);
offset += 4; /* skip past start_frame */
if (hdr->caplen < offset)
return;
uhdr->start_frame = SWAPLONG(uhdr->start_frame);
offset += 4; /* skip past xfer_flags */
if (hdr->caplen < offset)
return;
uhdr->xfer_flags = SWAPLONG(uhdr->xfer_flags);
offset += 4; /* skip past ndesc */
if (hdr->caplen < offset)
return;
uhdr->ndesc = SWAPLONG(uhdr->ndesc);
if (uhdr->transfer_type == URB_ISOCHRONOUS) {
/* swap the values in struct linux_usb_isodesc */
usb_isodesc *pisodesc;
uint32_t i;
pisodesc = (usb_isodesc *)(void *)(buf+offset);
for (i = 0; i < uhdr->ndesc; i++) {
offset += 4; /* skip past status */
if (hdr->caplen < offset)
return;
pisodesc->status = SWAPLONG(pisodesc->status);
offset += 4; /* skip past offset */
if (hdr->caplen < offset)
return;
pisodesc->offset = SWAPLONG(pisodesc->offset);
offset += 4; /* skip past len */
if (hdr->caplen < offset)
return;
pisodesc->len = SWAPLONG(pisodesc->len);
offset += 4; /* skip past padding */
pisodesc++;
}
}
}
}
/*
* The DLT_NFLOG "packets" have a mixture of big-endian and host-byte-order
* data. They begin with a fixed-length header with big-endian fields,
* followed by a set of TLVs, where the type and length are in host
* byte order but the values are either big-endian or are a raw byte
* sequence that's the same regardless of the host's byte order.
*
* When reading a DLT_NFLOG packet, we need to convert the type and
* length values from the byte order of the host that wrote the file
* to the byte order of this host.
*/
static void
swap_nflog_header(const struct pcap_pkthdr *hdr, u_char *buf)
{
u_char *p = buf;
nflog_hdr_t *nfhdr = (nflog_hdr_t *)buf;
nflog_tlv_t *tlv;
u_int caplen = hdr->caplen;
u_int length = hdr->len;
u_int size;
if (caplen < (u_int) sizeof(nflog_hdr_t) ||
length < (u_int) sizeof(nflog_hdr_t)) {
/* Not enough data to have any TLVs. */
return;
}
if (nfhdr->nflog_version != 0) {
/* Unknown NFLOG version */
return;
}
length -= sizeof(nflog_hdr_t);
caplen -= sizeof(nflog_hdr_t);
p += sizeof(nflog_hdr_t);
while (caplen >= sizeof(nflog_tlv_t)) {
tlv = (nflog_tlv_t *) p;
/* Swap the type and length. */
tlv->tlv_type = SWAPSHORT(tlv->tlv_type);
tlv->tlv_length = SWAPSHORT(tlv->tlv_length);
/* Get the length of the TLV. */
size = tlv->tlv_length;
if (size % 4 != 0)
size += 4 - size % 4;
/* Is the TLV's length less than the minimum? */
if (size < sizeof(nflog_tlv_t)) {
/* Yes. Give up now. */
return;
}
/* Do we have enough data for the full TLV? */
if (caplen < size || length < size) {
/* No. */
return;
}
/* Skip over the TLV. */
length -= size;
caplen -= size;
p += size;
}
}
static void
swap_pseudo_headers(int linktype, struct pcap_pkthdr *hdr, u_char *data)
{
/*
* Convert pseudo-headers from the byte order of
* the host on which the file was saved to our
* byte order, as necessary.
*/
switch (linktype) {
case DLT_PFLOG:
swap_pflog_header(hdr, data);
break;
case DLT_LINUX_SLL:
swap_linux_sll_socketcan_header(hdr, data);
break;
case DLT_LINUX_SLL2:
swap_linux_sll2_socketcan_header(hdr, data);
break;
case DLT_USB_LINUX:
swap_linux_usb_header(hdr, data, 0);
break;
case DLT_USB_LINUX_MMAPPED:
swap_linux_usb_header(hdr, data, 1);
break;
case DLT_NFLOG:
swap_nflog_header(hdr, data);
break;
}
}
static inline int
packet_length_might_be_wrong(struct pcap_pkthdr *hdr,
const pcap_usb_header_mmapped *usb_hdr)
{
uint32_t old_style_packet_length;
/*
* Calculate the packet length the old way.
* We know that the multiplication won't overflow, but
* we don't know that the additions won't. Calculate
* it with no overflow checks, as that's how it
* would have been calculated when it was captured.
*/
old_style_packet_length = iso_pseudo_header_len(usb_hdr) +
usb_hdr->urb_len;
return (hdr->len == old_style_packet_length);
}
void
pcapint_post_process(int linktype, int swapped, struct pcap_pkthdr *hdr,
u_char *data)
{
if (swapped)
swap_pseudo_headers(linktype, hdr, data);
/*
* Is this a memory-mapped Linux USB capture?
*/
if (linktype == DLT_USB_LINUX_MMAPPED) {
/*
* Yes.
*
* In older versions of libpcap, in memory-mapped Linux
* USB captures, the original length of completion events
* for incoming isochronous transfers was miscalculated;
* it needed to be calculated based on the offsets and
* lengths in the descriptors, not on the raw URB length,
* but it wasn't.
*
* If this packet contains transferred data (yes, data_flag
* is 0 if we *do* have data), it's a completion event
* for an incoming isochronous transfer, and the
* transfer length appears to have been calculated
* from the raw URB length, fix it.
*
* We only do this if we have the full USB pseudo-header,
* because we will have to look at that header and at
* all of the isochronous descriptors.
*/
if (hdr->caplen < sizeof (pcap_usb_header_mmapped)) {
/*
* We don't have the full pseudo-header.
*/
return;
}
const pcap_usb_header_mmapped *usb_hdr =
(const pcap_usb_header_mmapped *) data;
/*
* Make sure the number of descriptors is sane.
*
* The Linux binary USB monitor code limits the number of
* isochronous descriptors to 128; if the number in the file
* is larger than that, either 1) the file's been damaged
* or 2) the file was produced after the number was raised
* in the kernel.
*
* In case 1), the number can't be trusted, so don't rely on
* it to attempt to fix the original length field in the pcap
* or pcapng header.
*
* In case 2), the system was probably running a version of
* libpcap that didn't miscalculate the original length, so
* it probably doesn't need to be fixed.
*
* This avoids the possibility of the product of the number of
* descriptors and the size of descriptors won't overflow an
* unsigned 32-bit integer.
*/
if (usb_hdr->ndesc > USB_MAXDESC)
return;
if (!usb_hdr->data_flag &&
is_isochronous_transfer_completion(usb_hdr) &&
packet_length_might_be_wrong(hdr, usb_hdr)) {
u_int len;
/*
* Make sure we have all of the descriptors,
* as we will have to look at all of them.
*
* If not, we don't bother trying to fix
* anything.
*/
if (hdr->caplen < iso_pseudo_header_len(usb_hdr))
return;
/*
* Calculate what the length should have been.
*/
len = incoming_isochronous_transfer_completed_len(hdr,
data);
/*
* len is the smaller of UINT_MAX and the total
* header plus data length. That's guaranteed
* to fit in a UINT_MAX.
*
* Don't reduce the original length to a value
* below the captured length, however, as that
* is bogus.
*/
if (len >= hdr->caplen)
hdr->len = len;
/*
* If the captured length is greater than the
* length, use the captured length.
*
* For completion events for incoming isochronous
* transfers, it's based on data_len, which is
* calculated the same way we calculated
* pre_truncation_data_len above, except that
* it has access to all the isochronous descriptors,
* not just the ones that the kernel were able to
* provide us or, for a capture file, that weren't
* sliced off by a snapshot length.
*
* However, it might have been reduced by the USB
* capture mechanism arbitrarily limiting the amount
* of data it provides to userland, or by the libpcap
* capture code limiting it to being no more than the
* snapshot, so we don't want to just use it all the
* time; we only do so to try to get a better estimate
* of the actual length - and to make sure the
* original length is always >= the captured length.
*/
if (hdr->caplen > hdr->len)
hdr->len = hdr->caplen;
}
}
}