* <laurent.deniel@free.fr>
*
* Almost completely rewritten in order to:
- *
+ *
* - be able to use a unlimited number of ringbuffer files
* - close the current file and open (truncating) the next file at switch
* - set the final file name once open (or reopen)
* - avoid the deletion of files that could not be truncated (can't arise now)
* and do not erase empty files
*
- * The idea behind that is to remove the limitation of the maximum # of
+ * The idea behind that is to remove the limitation of the maximum # of
* ringbuffer files being less than the maximum # of open fd per process
* and to be able to reduce the amount of virtual memory usage (having only
* one file open at most) or the amount of file system usage (by truncating
- * the files at switch and not the capture stop, and by closing them which
+ * the files at switch and not the capture stop, and by closing them which
* makes possible their move or deletion after a switch).
*
*/
#include "pcapio.h"
#include "ringbuffer.h"
-#include "file_util.h"
+#include <wsutil/file_util.h>
/* Ringbuffer file structure */
gchar *fprefix; /* Filename prefix */
gchar *fsuffix; /* Filename suffix */
gboolean unlimited; /* TRUE if unlimited number of files */
- int linktype;
- int snaplen;
int fd; /* Current ringbuffer file descriptor */
FILE *pdh;
/*
- * create the next filename and open a new binary file with that name
+ * create the next filename and open a new binary file with that name
*/
static int ringbuf_open_file(rb_file *rfile, int *err)
{
if (rfile->name != NULL) {
if (rb_data.unlimited == FALSE) {
/* remove old file (if any, so ignore error) */
- eth_unlink(rfile->name);
+ ws_unlink(rfile->name);
}
g_free(rfile->name);
}
#endif
current_time = time(NULL);
- g_snprintf(filenum, sizeof(filenum), "%05u", (rb_data.curr_file_num + 1) % 100000);
+ g_snprintf(filenum, sizeof(filenum), "%05u", (rb_data.curr_file_num + 1) % RINGBUFFER_MAX_NUM_FILES);
strftime(timestr, sizeof(timestr), "%Y%m%d%H%M%S", localtime(¤t_time));
rfile->name = g_strconcat(rb_data.fprefix, "_", filenum, "_", timestr,
rb_data.fsuffix, NULL);
return -1;
}
- rb_data.fd = eth_open(rfile->name, O_RDWR|O_BINARY|O_TRUNC|O_CREAT, 0600);
+ rb_data.fd = ws_open(rfile->name, O_RDWR|O_BINARY|O_TRUNC|O_CREAT, 0600);
if (rb_data.fd == -1 && err != NULL) {
*err = errno;
rb_data.num_files = 1;
}
- rb_data.files = g_malloc(rb_data.num_files * sizeof(rb_file));
+ rb_data.files = (rb_file *)g_malloc(rb_data.num_files * sizeof(rb_file));
if (rb_data.files == NULL) {
return -1;
}
* Calls libpcap_fdopen() for the current ringbuffer file
*/
FILE *
-ringbuf_init_libpcap_fdopen(int linktype, int snaplen,
- long *bytes_written, int *err)
+ringbuf_init_libpcap_fdopen(int *err)
{
-
- rb_data.linktype = linktype;
- rb_data.snaplen = snaplen;
-
- rb_data.pdh = libpcap_fdopen(rb_data.fd, linktype, snaplen, bytes_written,
- err);
+ rb_data.pdh = libpcap_fdopen(rb_data.fd, err);
return rb_data.pdh;
}
* Switches to the next ringbuffer file
*/
gboolean
-ringbuf_switch_file(FILE **pdh, gchar **save_file, int *save_file_fd,
- long *bytes_written, int *err)
+ringbuf_switch_file(FILE **pdh, gchar **save_file, int *save_file_fd, int *err)
{
int next_file_index;
rb_file *next_rfile = NULL;
/* close current file */
if (!libpcap_dump_close(rb_data.pdh, err)) {
- eth_close(rb_data.fd); /* XXX - the above should have closed this already */
+ ws_close(rb_data.fd); /* XXX - the above should have closed this already */
rb_data.pdh = NULL; /* it's still closed, we just got an error while closing */
rb_data.fd = -1;
return FALSE;
return FALSE;
}
- if (ringbuf_init_libpcap_fdopen(rb_data.linktype, rb_data.snaplen,
- bytes_written, err) == NULL) {
+ if (ringbuf_init_libpcap_fdopen(err) == NULL) {
return FALSE;
}
/* close current file, if it's open */
if (rb_data.pdh != NULL) {
if (!libpcap_dump_close(rb_data.pdh, err)) {
- eth_close(rb_data.fd);
+ ws_close(rb_data.fd);
ret_val = FALSE;
}
/* XXX - it shouldn't still be open; "libpcap_dump_close()" should leave the
file closed even if it fails */
if (rb_data.fd != -1) {
- eth_close(rb_data.fd);
+ ws_close(rb_data.fd);
rb_data.fd = -1;
}
if (rb_data.files != NULL) {
for (i=0; i < rb_data.num_files; i++) {
if (rb_data.files[i].name != NULL) {
- eth_unlink(rb_data.files[i].name);
+ ws_unlink(rb_data.files[i].name);
}
}
}