xref: /openbsd-src/usr.bin/rsync/receiver.c (revision 24bb5fcea3ed904bc467217bdaadb5dfc618d5bf)
1 /*	$OpenBSD: receiver.c,v 1.28 2021/06/30 13:10:04 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, excl;
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 	/* Client sends zero-length exclusions. */
188 
189 	if (!sess->opts->server && !io_write_int(sess, fdout, 0)) {
190 		ERRX1("io_write_int");
191 		goto out;
192 	}
193 
194 	if (sess->opts->server && sess->opts->del) {
195 		if (!io_read_size(sess, fdin, &excl)) {
196 			ERRX1("io_read_size");
197 			goto out;
198 		} else if (excl != 0) {
199 			ERRX("exclusion list is non-empty");
200 			goto out;
201 		}
202 	}
203 
204 	/*
205 	 * Start by receiving the file list and our mystery number.
206 	 * These we're going to be touching on our local system.
207 	 */
208 
209 	if (!flist_recv(sess, fdin, &fl, &flsz)) {
210 		ERRX1("flist_recv");
211 		goto out;
212 	}
213 
214 	/* The IO error is sent after the file list. */
215 
216 	if (!io_read_int(sess, fdin, &ioerror)) {
217 		ERRX1("io_read_int");
218 		goto out;
219 	} else if (ioerror != 0) {
220 		ERRX1("io_error is non-zero");
221 		goto out;
222 	}
223 
224 	if (flsz == 0 && !sess->opts->server) {
225 		WARNX("receiver has empty file list: exiting");
226 		rc = 1;
227 		goto out;
228 	} else if (!sess->opts->server)
229 		LOG1("Transfer starting: %zu files", flsz);
230 
231 	LOG2("%s: receiver destination", root);
232 
233 	/*
234 	 * Create the path for our destination directory, if we're not
235 	 * in dry-run mode (which would otherwise crash w/the pledge).
236 	 * This uses our current umask: we might set the permissions on
237 	 * this directory in post_dir().
238 	 */
239 
240 	if (!sess->opts->dry_run) {
241 		if ((tofree = strdup(root)) == NULL)
242 			err(ERR_NOMEM, NULL);
243 		if (mkpath(tofree) < 0)
244 			err(ERR_FILE_IO, "%s: mkpath", tofree);
245 		free(tofree);
246 	}
247 
248 	/*
249 	 * Disable umask() so we can set permissions fully.
250 	 * Then open the directory iff we're not in dry_run.
251 	 */
252 
253 	oumask = umask(0);
254 
255 	if (!sess->opts->dry_run) {
256 		dfd = open(root, O_RDONLY | O_DIRECTORY, 0);
257 		if (dfd == -1)
258 			err(ERR_FILE_IO, "%s: open", root);
259 	}
260 
261 	/*
262 	 * Begin by conditionally getting all files we have currently
263 	 * available in our destination.
264 	 */
265 
266 	if (sess->opts->del &&
267 	    sess->opts->recursive &&
268 	    !flist_gen_dels(sess, root, &dfl, &dflsz, fl, flsz)) {
269 		ERRX1("flist_gen_local");
270 		goto out;
271 	}
272 
273 	/*
274 	 * Make our entire view of the file-system be limited to what's
275 	 * in the root directory.
276 	 * This prevents us from accidentally (or "under the influence")
277 	 * writing into other parts of the file-system.
278 	 */
279 
280 	if (unveil(root, "rwc") == -1)
281 		err(ERR_IPC, "%s: unveil", root);
282 	if (unveil(NULL, NULL) == -1)
283 		err(ERR_IPC, "unveil");
284 
285 	/* If we have a local set, go for the deletion. */
286 
287 	if (!flist_del(sess, dfd, dfl, dflsz)) {
288 		ERRX1("flist_del");
289 		goto out;
290 	}
291 
292 	/* Initialise poll events to listen from the sender. */
293 
294 	pfd[PFD_SENDER_IN].fd = fdin;
295 	pfd[PFD_UPLOADER_IN].fd = -1;
296 	pfd[PFD_DOWNLOADER_IN].fd = -1;
297 	pfd[PFD_SENDER_OUT].fd = fdout;
298 
299 	pfd[PFD_SENDER_IN].events = POLLIN;
300 	pfd[PFD_UPLOADER_IN].events = POLLIN;
301 	pfd[PFD_DOWNLOADER_IN].events = POLLIN;
302 	pfd[PFD_SENDER_OUT].events = POLLOUT;
303 
304 	ul = upload_alloc(root, dfd, fdout, CSUM_LENGTH_PHASE1, fl, flsz,
305 	    oumask);
306 
307 	if (ul == NULL) {
308 		ERRX1("upload_alloc");
309 		goto out;
310 	}
311 
312 	dl = download_alloc(sess, fdin, fl, flsz, dfd);
313 	if (dl == NULL) {
314 		ERRX1("download_alloc");
315 		goto out;
316 	}
317 
318 	LOG2("%s: ready for phase 1 data", root);
319 
320 	for (;;) {
321 		if ((c = poll(pfd, PFD__MAX, poll_timeout)) == -1) {
322 			ERR("poll");
323 			goto out;
324 		} else if (c == 0) {
325 			ERRX("poll: timeout");
326 			goto out;
327 		}
328 
329 		for (i = 0; i < PFD__MAX; i++)
330 			if (pfd[i].revents & (POLLERR|POLLNVAL)) {
331 				ERRX("poll: bad fd");
332 				goto out;
333 			} else if (pfd[i].revents & POLLHUP) {
334 				ERRX("poll: hangup");
335 				goto out;
336 			}
337 
338 		/*
339 		 * If we have a read event and we're multiplexing, we
340 		 * might just have error messages in the pipe.
341 		 * It's important to flush these out so that we don't
342 		 * clog the pipe.
343 		 * Unset our polling status if there's nothing that
344 		 * remains in the pipe.
345 		 */
346 
347 		if (sess->mplex_reads &&
348 		    (pfd[PFD_SENDER_IN].revents & POLLIN)) {
349 			if (!io_read_flush(sess, fdin)) {
350 				ERRX1("io_read_flush");
351 				goto out;
352 			} else if (sess->mplex_read_remain == 0)
353 				pfd[PFD_SENDER_IN].revents &= ~POLLIN;
354 		}
355 
356 
357 		/*
358 		 * We run the uploader if we have files left to examine
359 		 * (i < flsz) or if we have a file that we've opened and
360 		 * is read to mmap.
361 		 */
362 
363 		if ((pfd[PFD_UPLOADER_IN].revents & POLLIN) ||
364 		    (pfd[PFD_SENDER_OUT].revents & POLLOUT)) {
365 			c = rsync_uploader(ul,
366 				&pfd[PFD_UPLOADER_IN].fd,
367 				sess, &pfd[PFD_SENDER_OUT].fd);
368 			if (c < 0) {
369 				ERRX1("rsync_uploader");
370 				goto out;
371 			}
372 		}
373 
374 		/*
375 		 * We need to run the downloader when we either have
376 		 * read events from the sender or an asynchronous local
377 		 * open is ready.
378 		 * XXX: we don't disable PFD_SENDER_IN like with the
379 		 * uploader because we might stop getting error
380 		 * messages, which will otherwise clog up the pipes.
381 		 */
382 
383 		if ((pfd[PFD_SENDER_IN].revents & POLLIN) ||
384 		    (pfd[PFD_DOWNLOADER_IN].revents & POLLIN)) {
385 			c = rsync_downloader(dl, sess,
386 				&pfd[PFD_DOWNLOADER_IN].fd);
387 			if (c < 0) {
388 				ERRX1("rsync_downloader");
389 				goto out;
390 			} else if (c == 0) {
391 				assert(phase == 0);
392 				phase++;
393 				LOG2("%s: receiver ready for phase 2 data", root);
394 				break;
395 			}
396 
397 			/*
398 			 * FIXME: if we have any errors during the
399 			 * download, most notably files getting out of
400 			 * sync between the send and the receiver, then
401 			 * here we should bump our checksum length and
402 			 * go into the second phase.
403 			 */
404 		}
405 	}
406 
407 	/* Properly close us out by progressing through the phases. */
408 
409 	if (phase == 1) {
410 		if (!io_write_int(sess, fdout, -1)) {
411 			ERRX1("io_write_int");
412 			goto out;
413 		}
414 		if (!io_read_int(sess, fdin, &ioerror)) {
415 			ERRX1("io_read_int");
416 			goto out;
417 		}
418 		if (ioerror != -1) {
419 			ERRX("expected phase ack");
420 			goto out;
421 		}
422 	}
423 
424 	/*
425 	 * Now all of our transfers are complete, so we can fix up our
426 	 * directory permissions.
427 	 */
428 
429 	if (!rsync_uploader_tail(ul, sess)) {
430 		ERRX1("rsync_uploader_tail");
431 		goto out;
432 	}
433 
434 	/* Process server statistics and say good-bye. */
435 
436 	if (!sess_stats_recv(sess, fdin)) {
437 		ERRX1("sess_stats_recv");
438 		goto out;
439 	}
440 	if (!io_write_int(sess, fdout, -1)) {
441 		ERRX1("io_write_int");
442 		goto out;
443 	}
444 
445 	LOG2("receiver finished updating");
446 	rc = 1;
447 out:
448 	if (dfd != -1)
449 		close(dfd);
450 	upload_free(ul);
451 	download_free(dl);
452 	flist_free(fl, flsz);
453 	flist_free(dfl, dflsz);
454 	return rc;
455 }
456