upstream: support for "cross"-loading files/directories, i.e.

downloading from one SFTP server while simultaneously uploading to another.

feedback & ok markus@

OpenBSD-Commit-ID: 3982878e29d8df0fa4ddc502f5ff6126ac714235
This commit is contained in:
djm@openbsd.org 2021-08-07 00:02:41 +00:00 committed by Damien Miller
parent a50bd0367f
commit de7115b373
2 changed files with 442 additions and 2 deletions

View File

@ -1,4 +1,4 @@
/* $OpenBSD: sftp-client.c,v 1.144 2021/08/07 00:01:29 djm Exp $ */
/* $OpenBSD: sftp-client.c,v 1.145 2021/08/07 00:02:41 djm Exp $ */
/*
* Copyright (c) 2001-2004 Damien Miller <djm@openbsd.org>
*
@ -2048,6 +2048,429 @@ upload_dir(struct sftp_conn *conn, const char *src, const char *dst,
return ret;
}
static void
handle_dest_replies(struct sftp_conn *to, const char *to_path, int synchronous,
u_int *nreqsp, u_int *write_errorp)
{
struct sshbuf *msg;
u_char type;
u_int id, status;
int r;
struct pollfd pfd;
if ((msg = sshbuf_new()) == NULL)
fatal_f("sshbuf_new failed");
/* Try to eat replies from the upload side */
while (*nreqsp > 0) {
debug3_f("%u outstanding replies", *nreqsp);
if (!synchronous) {
/* Bail out if no data is ready to be read */
pfd.fd = to->fd_in;
pfd.events = POLLIN;
if ((r = poll(&pfd, 1, 0)) == -1) {
if (errno == EINTR)
break;
fatal_f("poll: %s", strerror(errno));
} else if (r == 0)
break; /* fd not ready */
}
sshbuf_reset(msg);
get_msg(to, msg);
if ((r = sshbuf_get_u8(msg, &type)) != 0 ||
(r = sshbuf_get_u32(msg, &id)) != 0)
fatal_fr(r, "dest parse");
debug3("Received dest reply T:%u I:%u R:%u", type, id, *nreqsp);
if (type != SSH2_FXP_STATUS) {
fatal_f("Expected SSH2_FXP_STATUS(%d) packet, got %d",
SSH2_FXP_STATUS, type);
}
if ((r = sshbuf_get_u32(msg, &status)) != 0)
fatal_fr(r, "parse dest status");
debug3("dest SSH2_FXP_STATUS %u", status);
if (status != SSH2_FX_OK) {
/* record first error */
if (*write_errorp == 0)
*write_errorp = status;
}
/*
* XXX this doesn't do full reply matching like do_upload and
* so cannot gracefully truncate terminated uploads at a
* high-water mark. ATM the only caller of this function (scp)
* doesn't support transfer resumption, so this doesn't matter
* a whole lot.
*
* To be safe, do_crossload truncates the destination file to
* zero length on upload failure, since we can't trust the
* server not to have reordered replies that could have
* inserted holes where none existed in the source file.
*
* XXX we could get a more accutate progress bar if we updated
* the counter based on the reply from the destination...
*/
(*nreqsp)--;
}
debug3_f("done: %u outstanding replies", *nreqsp);
}
int
do_crossload(struct sftp_conn *from, struct sftp_conn *to,
const char *from_path, const char *to_path,
Attrib *a, int preserve_flag)
{
struct sshbuf *msg;
int write_error, read_error, lmodified = 0, r;
u_int64_t offset = 0, size;
u_int id, buflen, num_req, max_req, status = SSH2_FX_OK;
u_int num_upload_req;
off_t progress_counter;
u_char *from_handle, *to_handle;
size_t from_handle_len, to_handle_len;
struct request {
u_int id;
size_t len;
u_int64_t offset;
TAILQ_ENTRY(request) tq;
};
TAILQ_HEAD(reqhead, request) requests;
struct request *req;
u_char type;
TAILQ_INIT(&requests);
if (a == NULL && (a = do_stat(from, from_path, 0)) == NULL)
return -1;
if ((a->flags & SSH2_FILEXFER_ATTR_PERMISSIONS) &&
(!S_ISREG(a->perm))) {
error("Cannot download non-regular file: %s", from_path);
return(-1);
}
if (a->flags & SSH2_FILEXFER_ATTR_SIZE)
size = a->size;
else
size = 0;
buflen = from->download_buflen;
if (buflen > to->upload_buflen)
buflen = to->upload_buflen;
/* Send open request to read side */
if (send_open(from, from_path, "origin", SSH2_FXF_READ, NULL,
&from_handle, &from_handle_len) != 0)
return -1;
/* Send open request to write side */
a->flags &= ~SSH2_FILEXFER_ATTR_SIZE;
a->flags &= ~SSH2_FILEXFER_ATTR_UIDGID;
a->perm &= 0777;
if (!preserve_flag)
a->flags &= ~SSH2_FILEXFER_ATTR_ACMODTIME;
if (send_open(to, to_path, "dest",
SSH2_FXF_WRITE|SSH2_FXF_CREAT|SSH2_FXF_TRUNC, a,
&to_handle, &to_handle_len) != 0) {
do_close(from, from_handle, from_handle_len);
return -1;
}
/* Read from remote "from" and write to remote "to" */
offset = 0;
write_error = read_error = num_req = num_upload_req = 0;
max_req = 1;
progress_counter = 0;
if (showprogress && size != 0)
start_progress_meter(from_path, size, &progress_counter);
if ((msg = sshbuf_new()) == NULL)
fatal_f("sshbuf_new failed");
while (num_req > 0 || max_req > 0) {
u_char *data;
size_t len;
/*
* Simulate EOF on interrupt: stop sending new requests and
* allow outstanding requests to drain gracefully
*/
if (interrupted) {
if (num_req == 0) /* If we haven't started yet... */
break;
max_req = 0;
}
/* Send some more requests */
while (num_req < max_req) {
debug3("Request range %llu -> %llu (%d/%d)",
(unsigned long long)offset,
(unsigned long long)offset + buflen - 1,
num_req, max_req);
req = xcalloc(1, sizeof(*req));
req->id = from->msg_id++;
req->len = buflen;
req->offset = offset;
offset += buflen;
num_req++;
TAILQ_INSERT_TAIL(&requests, req, tq);
send_read_request(from, req->id, req->offset,
req->len, from_handle, from_handle_len);
}
/* Try to eat replies from the upload side (nonblocking) */
handle_dest_replies(to, to_path, 0,
&num_upload_req, &write_error);
sshbuf_reset(msg);
get_msg(from, msg);
if ((r = sshbuf_get_u8(msg, &type)) != 0 ||
(r = sshbuf_get_u32(msg, &id)) != 0)
fatal_fr(r, "parse");
debug3("Received origin reply T:%u I:%u R:%d",
type, id, max_req);
/* Find the request in our queue */
for (req = TAILQ_FIRST(&requests);
req != NULL && req->id != id;
req = TAILQ_NEXT(req, tq))
;
if (req == NULL)
fatal("Unexpected reply %u", id);
switch (type) {
case SSH2_FXP_STATUS:
if ((r = sshbuf_get_u32(msg, &status)) != 0)
fatal_fr(r, "parse status");
if (status != SSH2_FX_EOF)
read_error = 1;
max_req = 0;
TAILQ_REMOVE(&requests, req, tq);
free(req);
num_req--;
break;
case SSH2_FXP_DATA:
if ((r = sshbuf_get_string(msg, &data, &len)) != 0)
fatal_fr(r, "parse data");
debug3("Received data %llu -> %llu",
(unsigned long long)req->offset,
(unsigned long long)req->offset + len - 1);
if (len > req->len)
fatal("Received more data than asked for "
"%zu > %zu", len, req->len);
lmodified = 1;
/* Write this chunk out to the destination */
sshbuf_reset(msg);
if ((r = sshbuf_put_u8(msg, SSH2_FXP_WRITE)) != 0 ||
(r = sshbuf_put_u32(msg, to->msg_id++)) != 0 ||
(r = sshbuf_put_string(msg, to_handle,
to_handle_len)) != 0 ||
(r = sshbuf_put_u64(msg, req->offset)) != 0 ||
(r = sshbuf_put_string(msg, data, len)) != 0)
fatal_fr(r, "compose write");
send_msg(to, msg);
debug3("Sent message SSH2_FXP_WRITE I:%u O:%llu S:%zu",
id, (unsigned long long)offset, len);
num_upload_req++;
progress_counter += len;
free(data);
if (len == req->len) {
TAILQ_REMOVE(&requests, req, tq);
free(req);
num_req--;
} else {
/* Resend the request for the missing data */
debug3("Short data block, re-requesting "
"%llu -> %llu (%2d)",
(unsigned long long)req->offset + len,
(unsigned long long)req->offset +
req->len - 1, num_req);
req->id = from->msg_id++;
req->len -= len;
req->offset += len;
send_read_request(from, req->id,
req->offset, req->len,
from_handle, from_handle_len);
/* Reduce the request size */
if (len < buflen)
buflen = MAXIMUM(MIN_READ_SIZE, len);
}
if (max_req > 0) { /* max_req = 0 iff EOF received */
if (size > 0 && offset > size) {
/* Only one request at a time
* after the expected EOF */
debug3("Finish at %llu (%2d)",
(unsigned long long)offset,
num_req);
max_req = 1;
} else if (max_req < from->num_requests) {
++max_req;
}
}
break;
default:
fatal("Expected SSH2_FXP_DATA(%u) packet, got %u",
SSH2_FXP_DATA, type);
}
}
if (showprogress && size)
stop_progress_meter();
/* Drain replies from the server (blocking) */
debug3_f("waiting for %u replies from destination", num_upload_req);
handle_dest_replies(to, to_path, 1, &num_upload_req, &write_error);
/* Sanity check */
if (TAILQ_FIRST(&requests) != NULL)
fatal("Transfer complete, but requests still in queue");
/* Truncate at 0 length on interrupt or error to avoid holes at dest */
if (read_error || write_error || interrupted) {
debug("truncating \"%s\" at 0", to_path);
do_close(to, to_handle, to_handle_len);
free(to_handle);
if (send_open(to, to_path, "dest",
SSH2_FXF_WRITE|SSH2_FXF_CREAT|SSH2_FXF_TRUNC, a,
&to_handle, &to_handle_len) != 0) {
error("truncation failed for \"%s\"", to_path);
to_handle = NULL;
}
}
if (read_error) {
error("Couldn't read from origin file \"%s\" : %s",
from_path, fx2txt(status));
status = -1;
do_close(from, from_handle, from_handle_len);
if (to_handle != NULL)
do_close(to, to_handle, to_handle_len);
} else if (write_error) {
error("Couldn't write to \"%s\": %s",
to_path, fx2txt(write_error));
status = SSH2_FX_FAILURE;
do_close(from, from_handle, from_handle_len);
if (to_handle != NULL)
do_close(to, to_handle, to_handle_len);
} else {
if (do_close(from, from_handle, from_handle_len) != 0 ||
interrupted)
status = -1;
else
status = SSH2_FX_OK;
if (to_handle != NULL) {
/* Need to resend utimes after write */
if (preserve_flag)
do_fsetstat(to, to_handle, to_handle_len, a);
do_close(to, to_handle, to_handle_len);
}
}
sshbuf_free(msg);
free(from_handle);
free(to_handle);
return status == SSH2_FX_OK ? 0 : -1;
}
static int
crossload_dir_internal(struct sftp_conn *from, struct sftp_conn *to,
const char *from_path, const char *to_path,
int depth, Attrib *dirattrib, int preserve_flag, int print_flag)
{
int i, ret = 0;
SFTP_DIRENT **dir_entries;
char *filename, *new_from_path = NULL, *new_to_path = NULL;
mode_t mode = 0777;
if (depth >= MAX_DIR_DEPTH) {
error("Maximum directory depth exceeded: %d levels", depth);
return -1;
}
if (dirattrib == NULL &&
(dirattrib = do_stat(from, from_path, 1)) == NULL) {
error("Unable to stat remote directory \"%s\"", from_path);
return -1;
}
if (!S_ISDIR(dirattrib->perm)) {
error("\"%s\" is not a directory", from_path);
return -1;
}
if (print_flag)
mprintf("Retrieving %s\n", from_path);
dirattrib->flags &= ~SSH2_FILEXFER_ATTR_SIZE;
dirattrib->flags &= ~SSH2_FILEXFER_ATTR_UIDGID;
if (dirattrib->flags & SSH2_FILEXFER_ATTR_PERMISSIONS) {
mode = dirattrib->perm & 01777;
dirattrib->perm = mode | (S_IWUSR|S_IXUSR); /* temp */
} else {
debug("Server did not send permissions for "
"directory \"%s\"", to_path);
}
if (do_mkdir(to, to_path, dirattrib, print_flag) != 0)
return -1;
if (do_readdir(from, from_path, &dir_entries) == -1) {
error("%s: Failed to get directory contents", from_path);
return -1;
}
for (i = 0; dir_entries[i] != NULL && !interrupted; i++) {
free(new_from_path);
free(new_to_path);
filename = dir_entries[i]->filename;
new_from_path = path_append(from_path, filename);
new_to_path = path_append(to_path, filename);
if (S_ISDIR(dir_entries[i]->a.perm)) {
if (strcmp(filename, ".") == 0 ||
strcmp(filename, "..") == 0)
continue;
if (crossload_dir_internal(from, to,
new_from_path, new_to_path,
depth + 1, &(dir_entries[i]->a), preserve_flag,
print_flag) == -1)
ret = -1;
} else if (S_ISREG(dir_entries[i]->a.perm) ) {
if (do_crossload(from, to, new_from_path, new_to_path,
&(dir_entries[i]->a), preserve_flag) == -1) {
error("Transfer of file %s to %s failed",
new_from_path, new_to_path);
ret = -1;
}
} else
logit("%s: not a regular file\n", new_from_path);
}
free(new_to_path);
free(new_from_path);
dirattrib->perm = mode; /* original mode */
do_setstat(to, to_path, dirattrib);
free_sftp_dirents(dir_entries);
return ret;
}
int
crossload_dir(struct sftp_conn *from, struct sftp_conn *to,
const char *from_path, const char *to_path,
Attrib *dirattrib, int preserve_flag, int print_flag)
{
char *from_path_canon;
int ret;
if ((from_path_canon = do_realpath(from, from_path)) == NULL) {
error("Unable to canonicalize path \"%s\"", from_path);
return -1;
}
ret = crossload_dir_internal(from, to, from_path_canon, to_path, 0,
dirattrib, preserve_flag, print_flag);
free(from_path_canon);
return ret;
}
char *
path_append(const char *p1, const char *p2)
{

View File

@ -1,4 +1,4 @@
/* $OpenBSD: sftp-client.h,v 1.30 2021/03/31 22:16:34 djm Exp $ */
/* $OpenBSD: sftp-client.h,v 1.31 2021/08/07 00:02:41 djm Exp $ */
/*
* Copyright (c) 2001-2004 Damien Miller <djm@openbsd.org>
@ -150,6 +150,23 @@ int do_upload(struct sftp_conn *, const char *, const char *, int, int, int);
int upload_dir(struct sftp_conn *, const char *, const char *, int, int, int,
int);
/*
* Download a 'from_path' from the 'from' connection and upload it to
* to 'to' connection at 'to_path'.
*/
int
do_crossload(struct sftp_conn *from, struct sftp_conn *to,
const char *from_path, const char *to_path,
Attrib *a, int preserve_flag);
/*
* Recursively download a directory from 'from_path' from the 'from'
* connection and upload it to 'to' connection at 'to_path'.
*/
int crossload_dir(struct sftp_conn *from, struct sftp_conn *to,
const char *from_path, const char *to_path,
Attrib *dirattrib, int preserve_flag, int print_flag);
/* Concatenate paths, taking care of slashes. Caller must free result. */
char *path_append(const char *, const char *);