wireshark/proto_hier_stats.c

335 lines
8.8 KiB
C
Raw Normal View History

/* proto_hier_stats.c
* Routines for calculating statistics based on protocol.
*
* $Id$
*
* Wireshark - Network traffic analyzer
* By Gerald Combs <gerald@wireshark.org>
* Copyright 1998 Gerald Combs
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include "config.h"
#include <stdio.h>
#include "globals.h"
#include "proto_hier_stats.h"
#include "ui/progress_dlg.h"
#include <epan/epan_dissect.h>
#include <wtap.h>
#include <stdio.h>
#include <string.h>
#include <glib.h>
/* Update the progress bar this many times when scanning the packet list. */
#define N_PROGBAR_UPDATES 100
#define STAT_NODE_STATS(n) ((ph_stats_node_t*)(n)->data)
#define STAT_NODE_HFINFO(n) (STAT_NODE_STATS(n)->hfinfo)
static GNode*
find_stat_node(GNode *parent_stat_node, header_field_info *needle_hfinfo)
{
GNode *needle_stat_node;
header_field_info *hfinfo;
ph_stats_node_t *stats;
needle_stat_node = g_node_first_child(parent_stat_node);
while (needle_stat_node) {
hfinfo = STAT_NODE_HFINFO(needle_stat_node);
if (hfinfo && hfinfo->id == needle_hfinfo->id) {
return needle_stat_node;
}
needle_stat_node = g_node_next_sibling(needle_stat_node);
}
/* None found. Create one. */
stats = g_new(ph_stats_node_t, 1);
/* Intialize counters */
stats->hfinfo = needle_hfinfo;
stats->num_pkts_total = 0;
stats->num_pkts_last = 0;
stats->num_bytes_total = 0;
stats->num_bytes_last = 0;
needle_stat_node = g_node_new(stats);
g_node_append(parent_stat_node, needle_stat_node);
return needle_stat_node;
}
static void
process_node(proto_node *ptree_node, GNode *parent_stat_node, ph_stats_t *ps, guint pkt_len)
{
field_info *finfo;
ph_stats_node_t *stats;
proto_node *proto_sibling_node;
GNode *stat_node;
finfo = PNODE_FINFO(ptree_node);
/* We don't fake protocol nodes we expect them to have a field_info.
* Dissection with faked proto tree? */
g_assert(finfo);
/* If the field info isn't related to a protocol but to a field,
* don't count them, as they don't belong to any protocol.
* (happens e.g. for toplevel tree item of desegmentation "[Reassembled TCP Segments]") */
if (finfo->hfinfo->parent != -1) {
/* Skip this element, use parent status node */
stat_node = parent_stat_node;
stats = STAT_NODE_STATS(stat_node);
} else {
stat_node = find_stat_node(parent_stat_node, finfo->hfinfo);
stats = STAT_NODE_STATS(stat_node);
stats->num_pkts_total++;
stats->num_bytes_total += pkt_len;
}
proto_sibling_node = ptree_node->next;
if (proto_sibling_node) {
/* If the name does not exist for this proto_sibling_node, then it is
* not a normal protocol in the top-level tree. It was instead
* added as a normal tree such as IPv6's Hop-by-hop Option Header and
* should be skipped when creating the protocol hierarchy display. */
if(strlen(PNODE_FINFO(proto_sibling_node)->hfinfo->name) == 0 && ptree_node->next)
proto_sibling_node = proto_sibling_node->next;
process_node(proto_sibling_node, stat_node, ps, pkt_len);
} else {
stats->num_pkts_last++;
stats->num_bytes_last += pkt_len;
}
}
static void
process_tree(proto_tree *protocol_tree, ph_stats_t* ps, guint pkt_len)
{
proto_node *ptree_node;
ptree_node = ((proto_node *)protocol_tree)->first_child;
if (!ptree_node) {
return;
}
process_node(ptree_node, ps->stats_tree, ps, pkt_len);
}
static gboolean
process_frame(frame_data *frame, column_info *cinfo, ph_stats_t* ps)
{
epan_dissect_t edt;
struct wtap_pkthdr phdr;
guint8 pd[WTAP_MAX_PACKET_SIZE];
double cur_time;
/* Load the frame from the capture file */
if (!cf_read_frame_r(&cfile, frame, &phdr, pd))
return FALSE; /* failure */
/* Dissect the frame tree not visible */
epan_dissect_init(&edt, TRUE, FALSE);
/* Don't fake protocols. We need them for the protocol hierarchy */
epan_dissect_fake_protocols(&edt, FALSE);
epan_dissect_run(&edt, &phdr, pd, frame, cinfo);
/* Get stats from this protocol tree */
process_tree(edt.tree, ps, frame->pkt_len);
/* Update times */
cur_time = nstime_to_sec(&frame->abs_ts);
if (cur_time < ps->first_time) {
ps->first_time = cur_time;
}
if (cur_time > ps->last_time){
ps->last_time = cur_time;
}
/* Free our memory. */
epan_dissect_cleanup(&edt);
return TRUE; /* success */
}
ph_stats_t*
ph_stats_new(void)
{
ph_stats_t *ps;
Store the frame_data structures in a tree, rather than a linked list. This lets us get rid of the per-frame_data-structure prev and next pointers, saving memory (at least according to Activity Monitor's report of the virtual address space size on my Snow Leopard machine, it's a noticeable saving), and lets us look up frame_data structures by frame number in O(log2(number of frames)) time rather than O(number of frames) time. It seems to take more CPU time when reading in the file, but seems to go from "finished reading in all the packets" to "displaying the packets" faster and seems to free up the frame_data structures faster when closing the file. It *is* doing more copying, currently, as we now don't allocate the frame_data structure until after the packet has passed the read filter, so that might account for the additional CPU time. (Oh, and, for what it's worth, on an LP64 platform, a frame_data structure is exactly 128 bytes long. However, there's more stuff to remove, so the power-of-2 size is not guaranteed to remain, and it's not a power-of-2 size on an ILP32 platform.) It also means we don't need GLib 2.10 or later for the two-pass mode in TShark. It also means some code in the TCP dissector that was checking pinfo->fd->next to see if it's NULL, in order to see if this is the last packet in the file, no longer works, but that wasn't guaranteed to work anyway: we might be doing a one-pass read through the capture in TShark; we might be dissecting the frame while we're reading in the packets for the first time in Wireshark; we might be doing a live capture in Wireshark; in which case packets might be prematurely considered "the last packet". #if 0 the no-longer-working tests, pending figuring out a better way of doing it. svn path=/trunk/; revision=36849
2011-04-25 19:01:05 +00:00
guint32 framenum;
frame_data *frame;
guint tot_packets, tot_bytes;
progdlg_t *progbar = NULL;
gboolean stop_flag;
int count;
float progbar_val;
GTimeVal start_time;
gchar status_str[100];
int progbar_nextstep;
int progbar_quantum;
/* Initialize the data */
ps = g_new(ph_stats_t, 1);
ps->tot_packets = 0;
ps->tot_bytes = 0;
ps->stats_tree = g_node_new(NULL);
ps->first_time = 0.0;
ps->last_time = 0.0;
/* Update the progress bar when it gets to this value. */
progbar_nextstep = 0;
/* When we reach the value that triggers a progress bar update,
bump that value by this amount. */
progbar_quantum = cfile.count/N_PROGBAR_UPDATES;
/* Count of packets at which we've looked. */
count = 0;
/* Progress so far. */
progbar_val = 0.0f;
stop_flag = FALSE;
g_get_current_time(&start_time);
tot_packets = 0;
tot_bytes = 0;
Store the frame_data structures in a tree, rather than a linked list. This lets us get rid of the per-frame_data-structure prev and next pointers, saving memory (at least according to Activity Monitor's report of the virtual address space size on my Snow Leopard machine, it's a noticeable saving), and lets us look up frame_data structures by frame number in O(log2(number of frames)) time rather than O(number of frames) time. It seems to take more CPU time when reading in the file, but seems to go from "finished reading in all the packets" to "displaying the packets" faster and seems to free up the frame_data structures faster when closing the file. It *is* doing more copying, currently, as we now don't allocate the frame_data structure until after the packet has passed the read filter, so that might account for the additional CPU time. (Oh, and, for what it's worth, on an LP64 platform, a frame_data structure is exactly 128 bytes long. However, there's more stuff to remove, so the power-of-2 size is not guaranteed to remain, and it's not a power-of-2 size on an ILP32 platform.) It also means we don't need GLib 2.10 or later for the two-pass mode in TShark. It also means some code in the TCP dissector that was checking pinfo->fd->next to see if it's NULL, in order to see if this is the last packet in the file, no longer works, but that wasn't guaranteed to work anyway: we might be doing a one-pass read through the capture in TShark; we might be dissecting the frame while we're reading in the packets for the first time in Wireshark; we might be doing a live capture in Wireshark; in which case packets might be prematurely considered "the last packet". #if 0 the no-longer-working tests, pending figuring out a better way of doing it. svn path=/trunk/; revision=36849
2011-04-25 19:01:05 +00:00
for (framenum = 1; framenum <= cfile.count; framenum++) {
frame = frame_data_sequence_find(cfile.frames, framenum);
Store the frame_data structures in a tree, rather than a linked list. This lets us get rid of the per-frame_data-structure prev and next pointers, saving memory (at least according to Activity Monitor's report of the virtual address space size on my Snow Leopard machine, it's a noticeable saving), and lets us look up frame_data structures by frame number in O(log2(number of frames)) time rather than O(number of frames) time. It seems to take more CPU time when reading in the file, but seems to go from "finished reading in all the packets" to "displaying the packets" faster and seems to free up the frame_data structures faster when closing the file. It *is* doing more copying, currently, as we now don't allocate the frame_data structure until after the packet has passed the read filter, so that might account for the additional CPU time. (Oh, and, for what it's worth, on an LP64 platform, a frame_data structure is exactly 128 bytes long. However, there's more stuff to remove, so the power-of-2 size is not guaranteed to remain, and it's not a power-of-2 size on an ILP32 platform.) It also means we don't need GLib 2.10 or later for the two-pass mode in TShark. It also means some code in the TCP dissector that was checking pinfo->fd->next to see if it's NULL, in order to see if this is the last packet in the file, no longer works, but that wasn't guaranteed to work anyway: we might be doing a one-pass read through the capture in TShark; we might be dissecting the frame while we're reading in the packets for the first time in Wireshark; we might be doing a live capture in Wireshark; in which case packets might be prematurely considered "the last packet". #if 0 the no-longer-working tests, pending figuring out a better way of doing it. svn path=/trunk/; revision=36849
2011-04-25 19:01:05 +00:00
/* Create the progress bar if necessary.
We check on every iteration of the loop, so that
it takes no longer than the standard time to create
it (otherwise, for a large file, we might take
considerably longer than that standard time in order
to get to the next progress bar step). */
if (progbar == NULL)
progbar = delayed_create_progress_dlg(
cfile.window, "Computing",
"protocol hierarchy statistics",
TRUE, &stop_flag, &start_time, progbar_val);
/* Update the progress bar, but do it only N_PROGBAR_UPDATES
times; when we update it, we have to run the GTK+ main
loop to get it to repaint what's pending, and doing so
may involve an "ioctl()" to see if there's any pending
input from an X server, and doing that for every packet
can be costly, especially on a big file. */
if (count >= progbar_nextstep) {
/* let's not divide by zero. I should never be started
* with count == 0, so let's assert that
*/
g_assert(cfile.count > 0);
progbar_val = (gfloat) count / cfile.count;
if (progbar != NULL) {
g_snprintf(status_str, sizeof(status_str),
"%4u of %u frames", count, cfile.count);
update_progress_dlg(progbar, progbar_val, status_str);
}
progbar_nextstep += progbar_quantum;
}
if (stop_flag) {
/* Well, the user decided to abort the statistics.
computation process Just stop. */
break;
}
/* Skip frames that are hidden due to the display filter.
XXX - should the progress bar count only packets that
passed the display filter? If so, it should
probably do so for other loops (see "file.c") that
look only at those packets. */
if (frame->flags.passed_dfilter) {
if (tot_packets == 0) {
double cur_time = nstime_to_sec(&frame->abs_ts);
ps->first_time = cur_time;
ps->last_time = cur_time;
}
/* we don't care about colinfo */
if (!process_frame(frame, NULL, ps)) {
/*
* Give up, and set "stop_flag" so we
* just abort rather than popping up
* the statistics window.
*/
stop_flag = TRUE;
break;
}
tot_packets++;
tot_bytes += frame->pkt_len;
}
count++;
}
/* We're done calculating the statistics; destroy the progress bar
if it was created. */
if (progbar != NULL)
destroy_progress_dlg(progbar);
if (stop_flag) {
/*
* We quit in the middle; throw away the statistics
* and return NULL, so our caller doesn't pop up a
* window with the incomplete statistics.
*/
ph_stats_free(ps);
return NULL;
}
ps->tot_packets = tot_packets;
ps->tot_bytes = tot_bytes;
return ps;
}
static gboolean
stat_node_free(GNode *node, gpointer data _U_)
{
ph_stats_node_t *stats = (ph_stats_node_t *)node->data;
if (stats) {
g_free(stats);
}
return FALSE;
}
void
ph_stats_free(ph_stats_t *ps)
{
if (ps->stats_tree) {
g_node_traverse(ps->stats_tree, G_IN_ORDER,
G_TRAVERSE_ALL, -1,
stat_node_free, NULL);
g_node_destroy(ps->stats_tree);
}
g_free(ps);
}