Regarding Charles Fisher's “Secure File Transfer” article in
the January 2016 issue:
I was surprised to see in his article on securing FTP the use of an
ancient and insecure piece of software, upload.c, which is more than 15
years old, and has no protection against trivial stack corruption. strcat,
strcpy, sprintf and the like are dangerous and should be avoided unless
in completely straightforward cases. The standard replacements for these
are strncat, strncpy and snprintf, which are all three POSIX. upload.c
is simply no longer from this time and should in my opinion never be
used without an almost complete rewrite.
—
Mischa Salle
Charles Fisher replies: I agree that Kessels' upload.c is of questionable quality due to its age and (lack of) security. It was, however, the only publicly available C code for complete handling of RFC-1867 as of 2015. I have since written the following prototype to replace it, coded for compactness over clarity, intended as a starting point for discussion—not as a fully featured server. It uses modern libraries (including BSD strlcat/cpy, which are safer for C string handling). Note that Conte's sha256 functions do not appear to work properly on big-endian architectures. This code is (hopefully) some improvement over the predecessor:
#include <stdio.h> /*xmit.c *prototype* RFC-1867 file transfer with:*/
#include <string.h> /* http://libccgi.sourceforge.net - cgi by Losen */
#include <stdlib.h> /* http://bradconte.com/sha256_c - sha by Conte */
#include <unistd.h> /* Copyright 2015 Charles Fisher. Distributed */
#include <sys/types.h>/* under the terms of the GNU Lesser General */
#include <sys/stat.h> /* Public License (LGPL 2.1) */
#include "ccgi.h"
/* Compile with: gcc -static -Wall -I. -O2 -o xmit.cgi xmit.c ccgi.c \
strlcpy.c strlcat.c sha256.c
##BEWARE:
http://www.slideshare.net/phdays/chw00t-breaking-unices-chroot-solutions
*/
#define UPL_PATH "/upload/" /* Trailing slash or filename prefix. */
#define TMP_PATH "/upload/cgi-upload" /* Must point to same filesystem.*/
#define uchar unsigned char /* 8-bit byte */
#define uint unsigned int /* 32-bit word */
typedef struct { uchar data[64]; uint datalen; uint bitlen[2];
uint state[8]; } SHA256_CTX;
void sha256_init(SHA256_CTX *);
void sha256_update(SHA256_CTX *, uchar *, uint len);
void sha256_final(SHA256_CTX *, uchar *hash);
size_t strlcat(char *, const char *, size_t);
size_t strlcpy(char *, const char *, size_t);
int main(int argc, char **argv)
{CGI_varlist *vl; const char *name; int mask_len = strlen(TMP_PATH);
char prefix[BUFSIZ] = UPL_PATH, dst[BUFSIZ], *p = getenv("SCRIPT_NAME");
/* Removing write and execute should constrain uploads to 400. */
umask(umask((mode_t)0)|S_IWUSR|S_IWGRP|S_IWOTH|S_IXUSR|S_IXGRP|S_IXOTH);
printf("Content-type: text/plain\r\n\r\n");
if(p != NULL) /* Use the SCRIPT_NAME as a filename local prefix. */
{char genbuf[BUFSIZ];
if(strlcpy(dst, p, BUFSIZ) >= BUFSIZ) return 1;
if((p = strrchr(dst, '/')) != NULL) p++; else p = dst;
if(strlcpy(genbuf, p, BUFSIZ) >= BUFSIZ) return 1;
if((p = strchr(genbuf, '.')) != NULL) *p = '\0';
if(strlcat(prefix, genbuf, BUFSIZ) >= BUFSIZ ||
strlcat(prefix, "-", BUFSIZ) >= BUFSIZ) return 1;
} else if(strlcat(prefix, "IN-", BUFSIZ) >= BUFSIZ) return 1;
if((vl = CGI_get_all(TMP_PATH "-XXXXXX")) == 0)
{ printf("CGI_get_all() failed\r\n"); return 1; }
sync(); /* Rather: sync && echo 3 > /proc/sys/vm/drop_caches */
for(name = CGI_first_name(vl); name != 0; name = CGI_next_name(vl))
{FILE *fp; CGI_value *val; struct stat junk_buf; int i, j;
if(!(val = CGI_lookup_all(vl, 0))) continue;
for(i = 0; val[i]; i++)
{ /* Does filename match TMP_PATH, and does it exist? */
if(!strncmp(val[i], TMP_PATH, mask_len) && !stat(val[i], &junk_buf))
{ /* Abort if sent an empty|malicious|oversized filename. */
j = i++;
if(!strlen(val[i]) || strchr(val[i], '/') || strchr(val[i], '\\') ||
strlcpy(dst, prefix, BUFSIZ) >= BUFSIZ ||
strlcat(dst, val[i], BUFSIZ) >= BUFSIZ) {printf("error");return 1;}
if(link(val[j], dst))
{ /* On link failure, try our best to keep this data. */
if(strlcat(dst, val[j] + mask_len, BUFSIZ) >= BUFSIZ ||
link(val[j], dst)) /* mkstemp suffix appended to filename. */
{ printf("name_error\t%s\r\n", val[i]); continue; }
}
if(unlink(val[j])) { printf("tmp_error\t%s\r\n", val[i]); }
if((fp = fopen(dst, "r")))
{SHA256_CTX ctx; uchar buf[BUFSIZ];
sha256_init(&ctx);
while((j = fread(buf, 1, BUFSIZ, fp))) sha256_update(&ctx, buf, j);
sha256_final(&ctx, buf); fclose(fp);
for(j = 0; j < 32; j++) printf("%02x", buf[j]);
printf("\t%s\r\n", val[i]);
}}}}
CGI_free_varlist(vl); return 0;
}
I've followed Dave's column for many years, and I must say his
scripting skills are without equal. I thought I would pass along an
idea for a future column. I've been in IT for 47 years, and what
amazes me is parallel tasking. Start a script that starts many
programs and monitors the PID, plus use interrupt to put you back into
a menu to control those tasks. To me, that is bang-for-the-buck in
scripting.
I'd be interested in a generic version where you could just drop those tasks in a list or
keep track of one starting a series of scripts—just an idea.
—
Larry Dalton
Dave Taylor replies: That's a cool idea, Larry, and thanks so much for writing in! I don't have a column due to the boss for a few weeks yet, so let me keep this on the proverbial drawing board and see if I can work it into my next column.
And, you've been in IT for 47 years. Did you start with punch cards or paper tape? I remember visiting my Dad's workplace back in the early '70s and they had paper tape as a storage medium. I was quite impressed. Now I have more storage on my watch.
In Shawn Powers' Editors' Choice article recommending Team Viewer (see “Help Me, Uncle Shawn” in the January 2016 Upfront section), he fails to mention the major problem with this type of solution in that the data path goes through a third-party server and, thus, poses an unknown security risk.
It is much better, in my opinion, to use something like RealVNC that
provides a direct point-to-point encrypted link between the two
computers. It is a simple, one-off job to put port 5900 through the
router.
—
Ian
Shawn Powers replies: Your security and privacy concerns are valid, and perhaps I should have pointed them out. Unfortunately, most folks needing the sort of help I can offer over Team Viewer don't know what a port is, much less how to forward TCP port 5900 through their router to whatever private IP their computer might have. In the case of my daughter at college, she doesn't even have that option. (She uses the university's Wi-Fi.) It would be possible to set up SSH keys and an automatic outgoing SSH tunnel to a server I own with a public IP address, and then reverse-tunnel port 5900 through that. But, for simple help with formatting a college paper or installing a printer, it's not worth the effort to guarantee privacy—at least not for me. You are correct, however, that it would have been good to include that information in my article. Thanks for pointing out the potential privacy issue.
