hush/networking/tftp.c
Denys Vlasenko 6528abeb6a tftpd: another small tweak
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2009-11-08 18:27:18 +01:00

853 lines
23 KiB
C

/* vi: set sw=4 ts=4: */
/*
* A simple tftp client/server for busybox.
* Tries to follow RFC1350.
* Only "octet" mode supported.
* Optional blocksize negotiation (RFC2347 + RFC2348)
*
* Copyright (C) 2001 Magnus Damm <damm@opensource.se>
*
* Parts of the code based on:
*
* atftp: Copyright (C) 2000 Jean-Pierre Lefebvre <helix@step.polymtl.ca>
* and Remi Lefebvre <remi@debian.org>
*
* utftp: Copyright (C) 1999 Uwe Ohse <uwe@ohse.de>
*
* tftpd added by Denys Vlasenko & Vladimir Dronnikov
*
* Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
*/
#include "libbb.h"
#if ENABLE_FEATURE_TFTP_GET || ENABLE_FEATURE_TFTP_PUT
#define TFTP_BLKSIZE_DEFAULT 512 /* according to RFC 1350, don't change */
#define TFTP_BLKSIZE_DEFAULT_STR "512"
#define TFTP_TIMEOUT_MS 50
#define TFTP_MAXTIMEOUT_MS 2000
#define TFTP_NUM_RETRIES 12 /* number of backed-off retries */
/* opcodes we support */
#define TFTP_RRQ 1
#define TFTP_WRQ 2
#define TFTP_DATA 3
#define TFTP_ACK 4
#define TFTP_ERROR 5
#define TFTP_OACK 6
/* error codes sent over network (we use only 0, 1, 3 and 8) */
/* generic (error message is included in the packet) */
#define ERR_UNSPEC 0
#define ERR_NOFILE 1
#define ERR_ACCESS 2
/* disk full or allocation exceeded */
#define ERR_WRITE 3
#define ERR_OP 4
#define ERR_BAD_ID 5
#define ERR_EXIST 6
#define ERR_BAD_USER 7
#define ERR_BAD_OPT 8
/* masks coming from getopt32 */
enum {
TFTP_OPT_GET = (1 << 0),
TFTP_OPT_PUT = (1 << 1),
/* pseudo option: if set, it's tftpd */
TFTPD_OPT = (1 << 7) * ENABLE_TFTPD,
TFTPD_OPT_r = (1 << 8) * ENABLE_TFTPD,
TFTPD_OPT_c = (1 << 9) * ENABLE_TFTPD,
TFTPD_OPT_u = (1 << 10) * ENABLE_TFTPD,
};
#if ENABLE_FEATURE_TFTP_GET && !ENABLE_FEATURE_TFTP_PUT
#define IF_GETPUT(...)
#define CMD_GET(cmd) 1
#define CMD_PUT(cmd) 0
#elif !ENABLE_FEATURE_TFTP_GET && ENABLE_FEATURE_TFTP_PUT
#define IF_GETPUT(...)
#define CMD_GET(cmd) 0
#define CMD_PUT(cmd) 1
#else
#define IF_GETPUT(...) __VA_ARGS__
#define CMD_GET(cmd) ((cmd) & TFTP_OPT_GET)
#define CMD_PUT(cmd) ((cmd) & TFTP_OPT_PUT)
#endif
/* NB: in the code below
* CMD_GET(cmd) and CMD_PUT(cmd) are mutually exclusive
*/
struct globals {
/* u16 TFTP_ERROR; u16 reason; both network-endian, then error text: */
uint8_t error_pkt[4 + 32];
char *user_opt;
/* used in tftpd_main(), a bit big for stack: */
char block_buf[TFTP_BLKSIZE_DEFAULT];
#if ENABLE_FEATURE_TFTP_PROGRESS_BAR
off_t pos;
off_t size;
const char *file;
bb_progress_t pmt;
#endif
};
#define G (*(struct globals*)&bb_common_bufsiz1)
struct BUG_G_too_big {
char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1];
};
#define block_buf (G.block_buf )
#define user_opt (G.user_opt )
#define error_pkt (G.error_pkt )
#define INIT_G() do { } while (0)
#define error_pkt_reason (error_pkt[3])
#define error_pkt_str (error_pkt + 4)
#if ENABLE_FEATURE_TFTP_PROGRESS_BAR
/* SIGALRM logic nicked from the wget applet */
static void progress_meter(int flag)
{
/* We can be called from signal handler */
int save_errno = errno;
if (flag == -1) { /* first call to progress_meter */
bb_progress_init(&G.pmt);
}
bb_progress_update(&G.pmt, G.file, 0, G.pos, G.size);
if (flag == 0) {
/* last call to progress_meter */
alarm(0);
fputc('\n', stderr);
} else {
if (flag == -1) { /* first call to progress_meter */
signal_SA_RESTART_empty_mask(SIGALRM, progress_meter);
}
alarm(1);
}
errno = save_errno;
}
static void tftp_progress_init(void)
{
progress_meter(-1);
}
static void tftp_progress_done(void)
{
progress_meter(0);
}
#else
# define tftp_progress_init() ((void)0)
# define tftp_progress_done() ((void)0)
#endif
#if ENABLE_FEATURE_TFTP_BLOCKSIZE
static int tftp_blksize_check(const char *blksize_str, int maxsize)
{
/* Check if the blksize is valid:
* RFC2348 says between 8 and 65464,
* but our implementation makes it impossible
* to use blksizes smaller than 22 octets. */
unsigned blksize = bb_strtou(blksize_str, NULL, 10);
if (errno
|| (blksize < 24) || (blksize > maxsize)
) {
bb_error_msg("bad blocksize '%s'", blksize_str);
return -1;
}
# if ENABLE_TFTP_DEBUG
bb_error_msg("using blksize %u", blksize);
# endif
return blksize;
}
static char *tftp_get_option(const char *option, char *buf, int len)
{
int opt_val = 0;
int opt_found = 0;
int k;
/* buf points to:
* "opt_name<NUL>opt_val<NUL>opt_name2<NUL>opt_val2<NUL>..." */
while (len > 0) {
/* Make sure options are terminated correctly */
for (k = 0; k < len; k++) {
if (buf[k] == '\0') {
goto nul_found;
}
}
return NULL;
nul_found:
if (opt_val == 0) { /* it's "name" part */
if (strcasecmp(buf, option) == 0) {
opt_found = 1;
}
} else if (opt_found) {
return buf;
}
k++;
buf += k;
len -= k;
opt_val ^= 1;
}
return NULL;
}
#endif
static int tftp_protocol(
/* NULL if tftp, !NULL if tftpd: */
len_and_sockaddr *our_lsa,
len_and_sockaddr *peer_lsa,
const char *local_file
IF_TFTP(, const char *remote_file)
#if !ENABLE_TFTP
# define remote_file NULL
#endif
/* 1 for tftp; 1/0 for tftpd depending whether client asked about it: */
IF_FEATURE_TFTP_BLOCKSIZE(, int want_transfer_size)
IF_FEATURE_TFTP_BLOCKSIZE(, int blksize))
{
#if !ENABLE_FEATURE_TFTP_BLOCKSIZE
enum { blksize = TFTP_BLKSIZE_DEFAULT };
#endif
struct pollfd pfd[1];
#define socket_fd (pfd[0].fd)
int len;
int send_len;
IF_FEATURE_TFTP_BLOCKSIZE(smallint expect_OACK = 0;)
smallint finished = 0;
uint16_t opcode;
uint16_t block_nr;
uint16_t recv_blk;
int open_mode, local_fd;
int retries, waittime_ms;
int io_bufsize = blksize + 4;
char *cp;
/* Can't use RESERVE_CONFIG_BUFFER here since the allocation
* size varies meaning BUFFERS_GO_ON_STACK would fail.
*
* We must keep the transmit and receive buffers separate
* in case we rcv a garbage pkt - we need to rexmit the last pkt.
*/
char *xbuf = xmalloc(io_bufsize);
char *rbuf = xmalloc(io_bufsize);
socket_fd = xsocket(peer_lsa->u.sa.sa_family, SOCK_DGRAM, 0);
setsockopt_reuseaddr(socket_fd);
if (!ENABLE_TFTP || our_lsa) { /* tftpd */
/* Create a socket which is:
* 1. bound to IP:port peer sent 1st datagram to,
* 2. connected to peer's IP:port
* This way we will answer from the IP:port peer
* expects, will not get any other packets on
* the socket, and also plain read/write will work. */
xbind(socket_fd, &our_lsa->u.sa, our_lsa->len);
xconnect(socket_fd, &peer_lsa->u.sa, peer_lsa->len);
/* Is there an error already? Send pkt and bail out */
if (error_pkt_reason || error_pkt_str[0])
goto send_err_pkt;
if (user_opt) {
struct passwd *pw = xgetpwnam(user_opt);
change_identity(pw); /* initgroups, setgid, setuid */
}
}
/* Prepare open mode */
if (CMD_PUT(option_mask32)) {
open_mode = O_RDONLY;
} else {
open_mode = O_WRONLY | O_TRUNC | O_CREAT;
#if ENABLE_TFTPD
if ((option_mask32 & (TFTPD_OPT+TFTPD_OPT_c)) == TFTPD_OPT) {
/* tftpd without -c */
open_mode = O_WRONLY | O_TRUNC;
}
#endif
}
/* Examples of network traffic.
* Note two cases when ACKs with block# of 0 are sent.
*
* Download without options:
* tftp -> "\0\1FILENAME\0octet\0"
* "\0\3\0\1FILEDATA..." <- tftpd
* tftp -> "\0\4\0\1"
* ...
* Download with option of blksize 16384:
* tftp -> "\0\1FILENAME\0octet\0blksize\00016384\0"
* "\0\6blksize\00016384\0" <- tftpd
* tftp -> "\0\4\0\0"
* "\0\3\0\1FILEDATA..." <- tftpd
* tftp -> "\0\4\0\1"
* ...
* Upload without options:
* tftp -> "\0\2FILENAME\0octet\0"
* "\0\4\0\0" <- tftpd
* tftp -> "\0\3\0\1FILEDATA..."
* "\0\4\0\1" <- tftpd
* ...
* Upload with option of blksize 16384:
* tftp -> "\0\2FILENAME\0octet\0blksize\00016384\0"
* "\0\6blksize\00016384\0" <- tftpd
* tftp -> "\0\3\0\1FILEDATA..."
* "\0\4\0\1" <- tftpd
* ...
*/
block_nr = 1;
cp = xbuf + 2;
if (!ENABLE_TFTP || our_lsa) { /* tftpd */
/* Open file (must be after changing user) */
local_fd = open(local_file, open_mode);
if (local_fd < 0) {
error_pkt_reason = ERR_NOFILE;
strcpy((char*)error_pkt_str, "can't open file");
goto send_err_pkt;
}
/* gcc 4.3.1 would NOT optimize it out as it should! */
#if ENABLE_FEATURE_TFTP_BLOCKSIZE
if (blksize != TFTP_BLKSIZE_DEFAULT || want_transfer_size) {
/* Create and send OACK packet. */
/* For the download case, block_nr is still 1 -
* we expect 1st ACK from peer to be for (block_nr-1),
* that is, for "block 0" which is our OACK pkt */
opcode = TFTP_OACK;
goto add_blksize_opt;
}
#endif
if (CMD_GET(option_mask32)) {
/* It's upload and we don't send OACK.
* We must ACK 1st packet (with filename)
* as if it is "block 0" */
block_nr = 0;
}
} else { /* tftp */
/* Open file (must be after changing user) */
local_fd = CMD_GET(option_mask32) ? STDOUT_FILENO : STDIN_FILENO;
if (NOT_LONE_DASH(local_file))
local_fd = xopen(local_file, open_mode);
/* Removing #if, or using if() statement instead of #if may lead to
* "warning: null argument where non-null required": */
#if ENABLE_TFTP
/* tftp */
/* We can't (and don't really need to) bind the socket:
* we don't know from which local IP datagrams will be sent,
* but kernel will pick the same IP every time (unless routing
* table is changed), thus peer will see dgrams consistently
* coming from the same IP.
* We would like to connect the socket, but since peer's
* UDP code can be less perfect than ours, _peer's_ IP:port
* in replies may differ from IP:port we used to send
* our first packet. We can connect() only when we get
* first reply. */
/* build opcode */
opcode = TFTP_WRQ;
if (CMD_GET(option_mask32)) {
opcode = TFTP_RRQ;
}
/* add filename and mode */
/* fill in packet if the filename fits into xbuf */
len = strlen(remote_file) + 1;
if (2 + len + sizeof("octet") >= io_bufsize) {
bb_error_msg("remote filename is too long");
goto ret;
}
strcpy(cp, remote_file);
cp += len;
/* add "mode" part of the packet */
strcpy(cp, "octet");
cp += sizeof("octet");
# if ENABLE_FEATURE_TFTP_BLOCKSIZE
if (blksize == TFTP_BLKSIZE_DEFAULT && !want_transfer_size)
goto send_pkt;
/* Need to add option to pkt */
if ((&xbuf[io_bufsize - 1] - cp) < sizeof("blksize NNNNN tsize ") + sizeof(off_t)*3) {
bb_error_msg("remote filename is too long");
goto ret;
}
expect_OACK = 1;
# endif
#endif /* ENABLE_TFTP */
#if ENABLE_FEATURE_TFTP_BLOCKSIZE
add_blksize_opt:
if (blksize != TFTP_BLKSIZE_DEFAULT) {
/* add "blksize", <nul>, blksize, <nul> */
strcpy(cp, "blksize");
cp += sizeof("blksize");
cp += snprintf(cp, 6, "%d", blksize) + 1;
}
if (want_transfer_size) {
/* add "tsize", <nul>, size, <nul> (see RFC2349) */
/* if tftp and downloading, we send "0" (since we opened local_fd with O_TRUNC)
* and this makes server to send "tsize" option with the size */
/* if tftp and uploading, we send file size (maybe dont, to not confuse old servers???) */
/* if tftpd and downloading, we are answering to client's request */
/* if tftpd and uploading: !want_transfer_size, this code is not executed */
struct stat st;
strcpy(cp, "tsize");
cp += sizeof("tsize");
st.st_size = 0;
fstat(local_fd, &st);
cp += sprintf(cp, "%"OFF_FMT"u", (off_t)st.st_size) + 1;
# if ENABLE_FEATURE_TFTP_PROGRESS_BAR
/* Save for progress bar. If 0 (tftp downloading),
* we look at server's reply later */
G.size = st.st_size;
if (remote_file && st.st_size)
tftp_progress_init();
# endif
}
#endif
/* First packet is built, so skip packet generation */
goto send_pkt;
}
/* Using mostly goto's - continue/break will be less clear
* in where we actually jump to */
while (1) {
/* Build ACK or DATA */
cp = xbuf + 2;
*((uint16_t*)cp) = htons(block_nr);
cp += 2;
block_nr++;
opcode = TFTP_ACK;
if (CMD_PUT(option_mask32)) {
opcode = TFTP_DATA;
len = full_read(local_fd, cp, blksize);
if (len < 0) {
goto send_read_err_pkt;
}
if (len != blksize) {
finished = 1;
}
cp += len;
}
send_pkt:
/* Send packet */
*((uint16_t*)xbuf) = htons(opcode); /* fill in opcode part */
send_len = cp - xbuf;
/* NB: send_len value is preserved in code below
* for potential resend */
retries = TFTP_NUM_RETRIES; /* re-initialize */
waittime_ms = TFTP_TIMEOUT_MS;
send_again:
#if ENABLE_TFTP_DEBUG
fprintf(stderr, "sending %u bytes\n", send_len);
for (cp = xbuf; cp < &xbuf[send_len]; cp++)
fprintf(stderr, "%02x ", (unsigned char) *cp);
fprintf(stderr, "\n");
#endif
xsendto(socket_fd, xbuf, send_len, &peer_lsa->u.sa, peer_lsa->len);
#if ENABLE_FEATURE_TFTP_PROGRESS_BAR
if (ENABLE_TFTP && remote_file) { /* tftp */
G.pos = (block_nr - 1) * (uoff_t)blksize;
}
#endif
/* Was it final ACK? then exit */
if (finished && (opcode == TFTP_ACK))
goto ret;
recv_again:
/* Receive packet */
/*pfd[0].fd = socket_fd;*/
pfd[0].events = POLLIN;
switch (safe_poll(pfd, 1, waittime_ms)) {
default:
/*bb_perror_msg("poll"); - done in safe_poll */
goto ret;
case 0:
retries--;
if (retries == 0) {
bb_error_msg("timeout");
goto ret; /* no err packet sent */
}
/* exponential backoff with limit */
waittime_ms += waittime_ms/2;
if (waittime_ms > TFTP_MAXTIMEOUT_MS) {
waittime_ms = TFTP_MAXTIMEOUT_MS;
}
goto send_again; /* resend last sent pkt */
case 1:
if (!our_lsa) {
/* tftp (not tftpd!) receiving 1st packet */
our_lsa = ((void*)(ptrdiff_t)-1); /* not NULL */
len = recvfrom(socket_fd, rbuf, io_bufsize, 0,
&peer_lsa->u.sa, &peer_lsa->len);
/* Our first dgram went to port 69
* but reply may come from different one.
* Remember and use this new port (and IP) */
if (len >= 0)
xconnect(socket_fd, &peer_lsa->u.sa, peer_lsa->len);
} else {
/* tftpd, or not the very first packet:
* socket is connect()ed, can just read from it. */
/* Don't full_read()!
* This is not TCP, one read == one pkt! */
len = safe_read(socket_fd, rbuf, io_bufsize);
}
if (len < 0) {
goto send_read_err_pkt;
}
if (len < 4) { /* too small? */
goto recv_again;
}
}
/* Process recv'ed packet */
opcode = ntohs( ((uint16_t*)rbuf)[0] );
recv_blk = ntohs( ((uint16_t*)rbuf)[1] );
#if ENABLE_TFTP_DEBUG
fprintf(stderr, "received %d bytes: %04x %04x\n", len, opcode, recv_blk);
#endif
if (opcode == TFTP_ERROR) {
static const char errcode_str[] ALIGN1 =
"\0"
"file not found\0"
"access violation\0"
"disk full\0"
"bad operation\0"
"unknown transfer id\0"
"file already exists\0"
"no such user\0"
"bad option";
const char *msg = "";
if (len > 4 && rbuf[4] != '\0') {
msg = &rbuf[4];
rbuf[io_bufsize - 1] = '\0'; /* paranoia */
} else if (recv_blk <= 8) {
msg = nth_string(errcode_str, recv_blk);
}
bb_error_msg("server error: (%u) %s", recv_blk, msg);
goto ret;
}
#if ENABLE_FEATURE_TFTP_BLOCKSIZE
if (expect_OACK) {
expect_OACK = 0;
if (opcode == TFTP_OACK) {
/* server seems to support options */
char *res;
res = tftp_get_option("blksize", &rbuf[2], len - 2);
if (res) {
blksize = tftp_blksize_check(res, blksize);
if (blksize < 0) {
error_pkt_reason = ERR_BAD_OPT;
goto send_err_pkt;
}
io_bufsize = blksize + 4;
}
# if ENABLE_FEATURE_TFTP_PROGRESS_BAR
if (remote_file && G.size == 0) { /* if we don't know it yet */
res = tftp_get_option("tsize", &rbuf[2], len - 2);
if (res) {
G.size = bb_strtoull(res, NULL, 10);
if (G.size)
tftp_progress_init();
}
}
# endif
if (CMD_GET(option_mask32)) {
/* We'll send ACK for OACK,
* such ACK has "block no" of 0 */
block_nr = 0;
}
continue;
}
/* rfc2347:
* "An option not acknowledged by the server
* must be ignored by the client and server
* as if it were never requested." */
bb_error_msg("server only supports blocksize of 512");
blksize = TFTP_BLKSIZE_DEFAULT;
io_bufsize = TFTP_BLKSIZE_DEFAULT + 4;
}
#endif
/* block_nr is already advanced to next block# we expect
* to get / block# we are about to send next time */
if (CMD_GET(option_mask32) && (opcode == TFTP_DATA)) {
if (recv_blk == block_nr) {
int sz = full_write(local_fd, &rbuf[4], len - 4);
if (sz != len - 4) {
strcpy((char*)error_pkt_str, bb_msg_write_error);
error_pkt_reason = ERR_WRITE;
goto send_err_pkt;
}
if (sz != blksize) {
finished = 1;
}
continue; /* send ACK */
}
/* Disabled to cope with servers with Sorcerer's Apprentice Syndrome */
#if 0
if (recv_blk == (block_nr - 1)) {
/* Server lost our TFTP_ACK. Resend it */
block_nr = recv_blk;
continue;
}
#endif
}
if (CMD_PUT(option_mask32) && (opcode == TFTP_ACK)) {
/* did peer ACK our last DATA pkt? */
if (recv_blk == (uint16_t) (block_nr - 1)) {
if (finished)
goto ret;
continue; /* send next block */
}
}
/* Awww... recv'd packet is not recognized! */
goto recv_again;
/* why recv_again? - rfc1123 says:
* "The sender (i.e., the side originating the DATA packets)
* must never resend the current DATA packet on receipt
* of a duplicate ACK".
* DATA pkts are resent ONLY on timeout.
* Thus "goto send_again" will ba a bad mistake above.
* See:
* http://en.wikipedia.org/wiki/Sorcerer's_Apprentice_Syndrome
*/
} /* end of "while (1)" */
ret:
if (ENABLE_FEATURE_CLEAN_UP) {
close(local_fd);
close(socket_fd);
free(xbuf);
free(rbuf);
}
return finished == 0; /* returns 1 on failure */
send_read_err_pkt:
strcpy((char*)error_pkt_str, bb_msg_read_error);
send_err_pkt:
if (error_pkt_str[0])
bb_error_msg("%s", (char*)error_pkt_str);
error_pkt[1] = TFTP_ERROR;
xsendto(socket_fd, error_pkt, 4 + 1 + strlen((char*)error_pkt_str),
&peer_lsa->u.sa, peer_lsa->len);
return EXIT_FAILURE;
#undef remote_file
}
#if ENABLE_TFTP
int tftp_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
int tftp_main(int argc UNUSED_PARAM, char **argv)
{
len_and_sockaddr *peer_lsa;
const char *local_file = NULL;
const char *remote_file = NULL;
# if ENABLE_FEATURE_TFTP_BLOCKSIZE
const char *blksize_str = TFTP_BLKSIZE_DEFAULT_STR;
int blksize;
# endif
int result;
int port;
IF_GETPUT(int opt;)
INIT_G();
/* -p or -g is mandatory, and they are mutually exclusive */
opt_complementary = "" IF_FEATURE_TFTP_GET("g:") IF_FEATURE_TFTP_PUT("p:")
IF_GETPUT("g--p:p--g:");
IF_GETPUT(opt =) getopt32(argv,
IF_FEATURE_TFTP_GET("g") IF_FEATURE_TFTP_PUT("p")
"l:r:" IF_FEATURE_TFTP_BLOCKSIZE("b:"),
&local_file, &remote_file
IF_FEATURE_TFTP_BLOCKSIZE(, &blksize_str));
argv += optind;
# if ENABLE_FEATURE_TFTP_BLOCKSIZE
/* Check if the blksize is valid:
* RFC2348 says between 8 and 65464 */
blksize = tftp_blksize_check(blksize_str, 65564);
if (blksize < 0) {
//bb_error_msg("bad block size");
return EXIT_FAILURE;
}
# endif
if (remote_file) {
if (!local_file) {
const char *slash = strrchr(remote_file, '/');
local_file = slash ? slash + 1 : remote_file;
}
} else {
remote_file = local_file;
}
/* Error if filename or host is not known */
if (!remote_file || !argv[0])
bb_show_usage();
port = bb_lookup_port(argv[1], "udp", 69);
peer_lsa = xhost2sockaddr(argv[0], port);
# if ENABLE_TFTP_DEBUG
fprintf(stderr, "using server '%s', remote_file '%s', local_file '%s'\n",
xmalloc_sockaddr2dotted(&peer_lsa->u.sa),
remote_file, local_file);
# endif
# if ENABLE_FEATURE_TFTP_PROGRESS_BAR
G.file = remote_file;
# endif
result = tftp_protocol(
NULL /*our_lsa*/, peer_lsa,
local_file, remote_file
IF_FEATURE_TFTP_BLOCKSIZE(, 1 /* want_transfer_size */)
IF_FEATURE_TFTP_BLOCKSIZE(, blksize)
);
tftp_progress_done();
if (result != EXIT_SUCCESS && NOT_LONE_DASH(local_file) && CMD_GET(opt)) {
unlink(local_file);
}
return result;
}
#endif /* ENABLE_TFTP */
#if ENABLE_TFTPD
int tftpd_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
int tftpd_main(int argc UNUSED_PARAM, char **argv)
{
len_and_sockaddr *our_lsa;
len_and_sockaddr *peer_lsa;
char *local_file, *mode;
const char *error_msg;
int opt, result, opcode;
IF_FEATURE_TFTP_BLOCKSIZE(int blksize = TFTP_BLKSIZE_DEFAULT;)
IF_FEATURE_TFTP_BLOCKSIZE(int want_transfer_size;)
INIT_G();
our_lsa = get_sock_lsa(STDIN_FILENO);
if (!our_lsa) {
/* This is confusing:
*bb_error_msg_and_die("stdin is not a socket");
* Better: */
bb_show_usage();
/* Help text says that tftpd must be used as inetd service,
* which is by far the most usual cause of get_sock_lsa
* failure */
}
peer_lsa = xzalloc(LSA_LEN_SIZE + our_lsa->len);
peer_lsa->len = our_lsa->len;
/* Shifting to not collide with TFTP_OPTs */
opt = option_mask32 = TFTPD_OPT | (getopt32(argv, "rcu:", &user_opt) << 8);
argv += optind;
if (argv[0])
xchdir(argv[0]);
result = recv_from_to(STDIN_FILENO, block_buf, sizeof(block_buf),
0 /* flags */,
&peer_lsa->u.sa, &our_lsa->u.sa, our_lsa->len);
error_msg = "malformed packet";
opcode = ntohs(*(uint16_t*)block_buf);
if (result < 4 || result >= sizeof(block_buf)
|| block_buf[result-1] != '\0'
|| (IF_FEATURE_TFTP_PUT(opcode != TFTP_RRQ) /* not download */
IF_GETPUT(&&)
IF_FEATURE_TFTP_GET(opcode != TFTP_WRQ) /* not upload */
)
) {
goto err;
}
local_file = block_buf + 2;
if (local_file[0] == '.' || strstr(local_file, "/.")) {
error_msg = "dot in file name";
goto err;
}
mode = local_file + strlen(local_file) + 1;
if (mode >= block_buf + result || strcmp(mode, "octet") != 0) {
goto err;
}
# if ENABLE_FEATURE_TFTP_BLOCKSIZE
want_transfer_size = 0;
{
char *res;
char *opt_str = mode + sizeof("octet");
int opt_len = block_buf + result - opt_str;
if (opt_len > 0) {
res = tftp_get_option("blksize", opt_str, opt_len);
if (res) {
blksize = tftp_blksize_check(res, 65564);
if (blksize < 0) {
error_pkt_reason = ERR_BAD_OPT;
/* will just send error pkt */
goto do_proto;
}
}
if (opcode != TFTP_WRQ /* download? */
/* did client ask us about file size? */
&& tftp_get_option("tsize", opt_str, opt_len)
) {
want_transfer_size = 1;
}
}
}
# endif
if (!ENABLE_FEATURE_TFTP_PUT || opcode == TFTP_WRQ) {
if (opt & TFTPD_OPT_r) {
/* This would mean "disk full" - not true */
/*error_pkt_reason = ERR_WRITE;*/
error_msg = bb_msg_write_error;
goto err;
}
IF_GETPUT(option_mask32 |= TFTP_OPT_GET;) /* will receive file's data */
} else {
IF_GETPUT(option_mask32 |= TFTP_OPT_PUT;) /* will send file's data */
}
/* NB: if error_pkt_str or error_pkt_reason is set up,
* tftp_protocol() just sends one error pkt and returns */
do_proto:
close(STDIN_FILENO); /* close old, possibly wildcard socket */
/* tftp_protocol() will create new one, bound to particular local IP */
result = tftp_protocol(
our_lsa, peer_lsa,
local_file IF_TFTP(, NULL /*remote_file*/)
IF_FEATURE_TFTP_BLOCKSIZE(, want_transfer_size)
IF_FEATURE_TFTP_BLOCKSIZE(, blksize)
);
return result;
err:
strcpy((char*)error_pkt_str, error_msg);
goto do_proto;
}
#endif /* ENABLE_TFTPD */
#endif /* ENABLE_FEATURE_TFTP_GET || ENABLE_FEATURE_TFTP_PUT */