xref: /openbsd-src/usr.bin/rsync/receiver.c (revision 4e1ee0786f11cc571bd0be17d38e46f635c719fc)
1 /*	$OpenBSD: receiver.c,v 1.30 2021/10/22 11:10:34 claudio Exp $ */
2 
3 /*
4  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
5  * Copyright (c) 2019 Florian Obser <florian@openbsd.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 #include <sys/mman.h>
20 #include <sys/stat.h>
21 
22 #include <assert.h>
23 #include <err.h>
24 #include <errno.h>
25 #include <fcntl.h>
26 #include <inttypes.h>
27 #include <math.h>
28 #include <poll.h>
29 #include <stdio.h>
30 #include <stdlib.h>
31 #include <string.h>
32 #include <time.h>
33 #include <unistd.h>
34 
35 #include "extern.h"
36 
37 enum	pfdt {
38 	PFD_SENDER_IN = 0, /* input from the sender */
39 	PFD_UPLOADER_IN, /* uploader input from a local file */
40 	PFD_DOWNLOADER_IN, /* downloader input from a local file */
41 	PFD_SENDER_OUT, /* output to the sender */
42 	PFD__MAX
43 };
44 
45 int
46 rsync_set_metadata(struct sess *sess, int newfile,
47 	int fd, const struct flist *f, const char *path)
48 {
49 	uid_t		 uid = (uid_t)-1;
50 	gid_t		 gid = (gid_t)-1;
51 	mode_t		 mode;
52 	struct timespec	 ts[2];
53 
54 	/* Conditionally adjust file modification time. */
55 
56 	if (sess->opts->preserve_times) {
57 		ts[0].tv_nsec = UTIME_NOW;
58 		ts[1].tv_sec = f->st.mtime;
59 		ts[1].tv_nsec = 0;
60 		if (futimens(fd, ts) == -1) {
61 			ERR("%s: futimens", path);
62 			return 0;
63 		}
64 		LOG4("%s: updated date", f->path);
65 	}
66 
67 	/*
68 	 * Conditionally adjust identifiers.
69 	 * If we have an EPERM, report it but continue on: this just
70 	 * means that we're mapping into an unknown (or disallowed)
71 	 * group identifier.
72 	 */
73 	if (getuid() == 0 && sess->opts->preserve_uids)
74 		uid = f->st.uid;
75 	if (sess->opts->preserve_gids)
76 		gid = f->st.gid;
77 
78 	mode = f->st.mode;
79 	if (uid != (uid_t)-1 || gid != (gid_t)-1) {
80 		if (fchown(fd, uid, gid) == -1) {
81 			if (errno != EPERM) {
82 				ERR("%s: fchown", path);
83 				return 0;
84 			}
85 			if (getuid() == 0)
86 				WARNX("%s: identity unknown or not available "
87 				    "to user.group: %u.%u", f->path, uid, gid);
88 		} else
89 			LOG4("%s: updated uid and/or gid", f->path);
90 		mode &= ~(S_ISTXT | S_ISUID | S_ISGID);
91 	}
92 
93 	/* Conditionally adjust file permissions. */
94 
95 	if (newfile || sess->opts->preserve_perms) {
96 		if (fchmod(fd, mode) == -1) {
97 			ERR("%s: fchmod", path);
98 			return 0;
99 		}
100 		LOG4("%s: updated permissions", f->path);
101 	}
102 
103 	return 1;
104 }
105 
106 int
107 rsync_set_metadata_at(struct sess *sess, int newfile, int rootfd,
108 	const struct flist *f, const char *path)
109 {
110 	uid_t		 uid = (uid_t)-1;
111 	gid_t		 gid = (gid_t)-1;
112 	mode_t		 mode;
113 	struct timespec	 ts[2];
114 
115 	/* Conditionally adjust file modification time. */
116 
117 	if (sess->opts->preserve_times) {
118 		ts[0].tv_nsec = UTIME_NOW;
119 		ts[1].tv_sec = f->st.mtime;
120 		ts[1].tv_nsec = 0;
121 		if (utimensat(rootfd, path, ts, AT_SYMLINK_NOFOLLOW) == -1) {
122 			ERR("%s: utimensat", path);
123 			return 0;
124 		}
125 		LOG4("%s: updated date", f->path);
126 	}
127 
128 	/*
129 	 * Conditionally adjust identifiers.
130 	 * If we have an EPERM, report it but continue on: this just
131 	 * means that we're mapping into an unknown (or disallowed)
132 	 * group identifier.
133 	 */
134 	if (getuid() == 0 && sess->opts->preserve_uids)
135 		uid = f->st.uid;
136 	if (sess->opts->preserve_gids)
137 		gid = f->st.gid;
138 
139 	mode = f->st.mode;
140 	if (uid != (uid_t)-1 || gid != (gid_t)-1) {
141 		if (fchownat(rootfd, path, uid, gid, AT_SYMLINK_NOFOLLOW) == -1) {
142 			if (errno != EPERM) {
143 				ERR("%s: fchownat", path);
144 				return 0;
145 			}
146 			if (getuid() == 0)
147 				WARNX("%s: identity unknown or not available "
148 				    "to user.group: %u.%u", f->path, uid, gid);
149 		} else
150 			LOG4("%s: updated uid and/or gid", f->path);
151 		mode &= ~(S_ISTXT | S_ISUID | S_ISGID);
152 	}
153 
154 	/* Conditionally adjust file permissions. */
155 
156 	if (newfile || sess->opts->preserve_perms) {
157 		if (fchmodat(rootfd, path, mode, AT_SYMLINK_NOFOLLOW) == -1) {
158 			ERR("%s: fchmodat", path);
159 			return 0;
160 		}
161 		LOG4("%s: updated permissions", f->path);
162 	}
163 
164 	return 1;
165 }
166 
167 /*
168  * Pledges: unveil, unix, rpath, cpath, wpath, stdio, fattr, chown.
169  * Pledges (dry-run): -unix, -cpath, -wpath, -fattr, -chown.
170  */
171 int
172 rsync_receiver(struct sess *sess, int fdin, int fdout, const char *root)
173 {
174 	struct flist	*fl = NULL, *dfl = NULL;
175 	size_t		 i, flsz = 0, dflsz = 0;
176 	char		*tofree;
177 	int		 rc = 0, dfd = -1, phase = 0, c;
178 	int32_t		 ioerror;
179 	struct pollfd	 pfd[PFD__MAX];
180 	struct download	*dl = NULL;
181 	struct upload	*ul = NULL;
182 	mode_t		 oumask;
183 
184 	if (pledge("stdio unix rpath wpath cpath dpath fattr chown getpw unveil", NULL) == -1)
185 		err(ERR_IPC, "pledge");
186 
187 	/*
188 	 * Create the path for our destination directory, if we're not
189 	 * in dry-run mode (which would otherwise crash w/the pledge).
190 	 * This uses our current umask: we might set the permissions on
191 	 * this directory in post_dir().
192 	 */
193 
194 	if (!sess->opts->dry_run) {
195 		if ((tofree = strdup(root)) == NULL)
196 			err(ERR_NOMEM, NULL);
197 		if (mkpath(tofree) < 0)
198 			err(ERR_FILE_IO, "%s: mkpath", tofree);
199 		free(tofree);
200 	}
201 
202 	/*
203 	 * Make our entire view of the file-system be limited to what's
204 	 * in the root directory.
205 	 * This prevents us from accidentally (or "under the influence")
206 	 * writing into other parts of the file-system.
207 	 */
208 	if (sess->opts->basedir[0]) {
209 		/*
210 		 * XXX just unveil everything for read
211 		 * Could unveil each basedir or maybe a common path
212 		 * also the fact that relative path are relative to the
213 		 * root does not help.
214 		 */
215 		if (unveil("/", "r") == -1)
216 			err(ERR_IPC, "%s: unveil", root);
217 	}
218 
219 	if (unveil(root, "rwc") == -1)
220 		err(ERR_IPC, "%s: unveil", root);
221 
222 	if (unveil(NULL, NULL) == -1)
223 		err(ERR_IPC, "unveil");
224 
225 	/* Client sends exclusions. */
226 	if (!sess->opts->server)
227 		send_rules(sess, fdout);
228 
229 	/* Server receives exclusions if delete is on. */
230 	if (sess->opts->server && sess->opts->del)
231 		recv_rules(sess, fdin);
232 
233 	/*
234 	 * Start by receiving the file list and our mystery number.
235 	 * These we're going to be touching on our local system.
236 	 */
237 
238 	if (!flist_recv(sess, fdin, &fl, &flsz)) {
239 		ERRX1("flist_recv");
240 		goto out;
241 	}
242 
243 	/* The IO error is sent after the file list. */
244 
245 	if (!io_read_int(sess, fdin, &ioerror)) {
246 		ERRX1("io_read_int");
247 		goto out;
248 	} else if (ioerror != 0) {
249 		ERRX1("io_error is non-zero");
250 		goto out;
251 	}
252 
253 	if (flsz == 0 && !sess->opts->server) {
254 		WARNX("receiver has empty file list: exiting");
255 		rc = 1;
256 		goto out;
257 	} else if (!sess->opts->server)
258 		LOG1("Transfer starting: %zu files", flsz);
259 
260 	LOG2("%s: receiver destination", root);
261 
262 	/*
263 	 * Disable umask() so we can set permissions fully.
264 	 * Then open the directory iff we're not in dry_run.
265 	 */
266 
267 	oumask = umask(0);
268 
269 	if (!sess->opts->dry_run) {
270 		dfd = open(root, O_RDONLY | O_DIRECTORY, 0);
271 		if (dfd == -1)
272 			err(ERR_FILE_IO, "%s: open", root);
273 	}
274 
275 	/*
276 	 * Begin by conditionally getting all files we have currently
277 	 * available in our destination.
278 	 */
279 
280 	if (sess->opts->del &&
281 	    sess->opts->recursive &&
282 	    !flist_gen_dels(sess, root, &dfl, &dflsz, fl, flsz)) {
283 		ERRX1("flist_gen_local");
284 		goto out;
285 	}
286 
287 	/* If we have a local set, go for the deletion. */
288 
289 	if (!flist_del(sess, dfd, dfl, dflsz)) {
290 		ERRX1("flist_del");
291 		goto out;
292 	}
293 
294 	/* Initialise poll events to listen from the sender. */
295 
296 	pfd[PFD_SENDER_IN].fd = fdin;
297 	pfd[PFD_UPLOADER_IN].fd = -1;
298 	pfd[PFD_DOWNLOADER_IN].fd = -1;
299 	pfd[PFD_SENDER_OUT].fd = fdout;
300 
301 	pfd[PFD_SENDER_IN].events = POLLIN;
302 	pfd[PFD_UPLOADER_IN].events = POLLIN;
303 	pfd[PFD_DOWNLOADER_IN].events = POLLIN;
304 	pfd[PFD_SENDER_OUT].events = POLLOUT;
305 
306 	ul = upload_alloc(root, dfd, fdout, CSUM_LENGTH_PHASE1, fl, flsz,
307 	    oumask);
308 
309 	if (ul == NULL) {
310 		ERRX1("upload_alloc");
311 		goto out;
312 	}
313 
314 	dl = download_alloc(sess, fdin, fl, flsz, dfd);
315 	if (dl == NULL) {
316 		ERRX1("download_alloc");
317 		goto out;
318 	}
319 
320 	LOG2("%s: ready for phase 1 data", root);
321 
322 	for (;;) {
323 		if ((c = poll(pfd, PFD__MAX, poll_timeout)) == -1) {
324 			ERR("poll");
325 			goto out;
326 		} else if (c == 0) {
327 			ERRX("poll: timeout");
328 			goto out;
329 		}
330 
331 		for (i = 0; i < PFD__MAX; i++)
332 			if (pfd[i].revents & (POLLERR|POLLNVAL)) {
333 				ERRX("poll: bad fd");
334 				goto out;
335 			} else if (pfd[i].revents & POLLHUP) {
336 				ERRX("poll: hangup");
337 				goto out;
338 			}
339 
340 		/*
341 		 * If we have a read event and we're multiplexing, we
342 		 * might just have error messages in the pipe.
343 		 * It's important to flush these out so that we don't
344 		 * clog the pipe.
345 		 * Unset our polling status if there's nothing that
346 		 * remains in the pipe.
347 		 */
348 
349 		if (sess->mplex_reads &&
350 		    (pfd[PFD_SENDER_IN].revents & POLLIN)) {
351 			if (!io_read_flush(sess, fdin)) {
352 				ERRX1("io_read_flush");
353 				goto out;
354 			} else if (sess->mplex_read_remain == 0)
355 				pfd[PFD_SENDER_IN].revents &= ~POLLIN;
356 		}
357 
358 
359 		/*
360 		 * We run the uploader if we have files left to examine
361 		 * (i < flsz) or if we have a file that we've opened and
362 		 * is read to mmap.
363 		 */
364 
365 		if ((pfd[PFD_UPLOADER_IN].revents & POLLIN) ||
366 		    (pfd[PFD_SENDER_OUT].revents & POLLOUT)) {
367 			c = rsync_uploader(ul,
368 				&pfd[PFD_UPLOADER_IN].fd,
369 				sess, &pfd[PFD_SENDER_OUT].fd);
370 			if (c < 0) {
371 				ERRX1("rsync_uploader");
372 				goto out;
373 			}
374 		}
375 
376 		/*
377 		 * We need to run the downloader when we either have
378 		 * read events from the sender or an asynchronous local
379 		 * open is ready.
380 		 * XXX: we don't disable PFD_SENDER_IN like with the
381 		 * uploader because we might stop getting error
382 		 * messages, which will otherwise clog up the pipes.
383 		 */
384 
385 		if ((pfd[PFD_SENDER_IN].revents & POLLIN) ||
386 		    (pfd[PFD_DOWNLOADER_IN].revents & POLLIN)) {
387 			c = rsync_downloader(dl, sess,
388 				&pfd[PFD_DOWNLOADER_IN].fd);
389 			if (c < 0) {
390 				ERRX1("rsync_downloader");
391 				goto out;
392 			} else if (c == 0) {
393 				assert(phase == 0);
394 				phase++;
395 				LOG2("%s: receiver ready for phase 2 data", root);
396 				break;
397 			}
398 
399 			/*
400 			 * FIXME: if we have any errors during the
401 			 * download, most notably files getting out of
402 			 * sync between the send and the receiver, then
403 			 * here we should bump our checksum length and
404 			 * go into the second phase.
405 			 */
406 		}
407 	}
408 
409 	/* Properly close us out by progressing through the phases. */
410 
411 	if (phase == 1) {
412 		if (!io_write_int(sess, fdout, -1)) {
413 			ERRX1("io_write_int");
414 			goto out;
415 		}
416 		if (!io_read_int(sess, fdin, &ioerror)) {
417 			ERRX1("io_read_int");
418 			goto out;
419 		}
420 		if (ioerror != -1) {
421 			ERRX("expected phase ack");
422 			goto out;
423 		}
424 	}
425 
426 	/*
427 	 * Now all of our transfers are complete, so we can fix up our
428 	 * directory permissions.
429 	 */
430 
431 	if (!rsync_uploader_tail(ul, sess)) {
432 		ERRX1("rsync_uploader_tail");
433 		goto out;
434 	}
435 
436 	/* Process server statistics and say good-bye. */
437 
438 	if (!sess_stats_recv(sess, fdin)) {
439 		ERRX1("sess_stats_recv");
440 		goto out;
441 	}
442 	if (!io_write_int(sess, fdout, -1)) {
443 		ERRX1("io_write_int");
444 		goto out;
445 	}
446 
447 	LOG2("receiver finished updating");
448 	rc = 1;
449 out:
450 	if (dfd != -1)
451 		close(dfd);
452 	upload_free(ul);
453 	download_free(dl);
454 	flist_free(fl, flsz);
455 	flist_free(dfl, dflsz);
456 	return rc;
457 }
458