xref: /dflybsd-src/sbin/hammer/cmd_mirror.c (revision 0bd7a37c6472ea619a1815995fa681750bb378cb)
1 /*
2  * Copyright (c) 2008 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sbin/hammer/cmd_mirror.c,v 1.16 2008/11/09 05:22:56 dillon Exp $
35  */
36 
37 #include "hammer.h"
38 
39 #define SERIALBUF_SIZE	(512 * 1024)
40 
41 static int read_mrecords(int fd, char *buf, u_int size,
42 			 hammer_ioc_mrecord_head_t pickup);
43 static int generate_histogram(int fd, const char *filesystem,
44 			 hammer_tid_t **histogram_ary,
45 			 struct hammer_ioc_mirror_rw *mirror_base);
46 static hammer_ioc_mrecord_any_t read_mrecord(int fdin, int *errorp,
47 			 hammer_ioc_mrecord_head_t pickup);
48 static void write_mrecord(int fdout, u_int32_t type,
49 			 hammer_ioc_mrecord_any_t mrec, int bytes);
50 static void generate_mrec_header(int fd, int pfs_id,
51 			 union hammer_ioc_mrecord_any *mrec_tmp);
52 static int validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
53 			 struct hammer_ioc_mrecord_head *pickup,
54 			 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp);
55 static void update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id);
56 static ssize_t writebw(int fd, const void *buf, size_t nbytes,
57 			u_int64_t *bwcount, struct timeval *tv1);
58 static int getyn(void);
59 static void mirror_usage(int code);
60 
61 #define BULK_MINIMUM	20000
62 
63 /*
64  * Generate a mirroring data stream from the specific source over the
65  * entire key range, but restricted to the specified transaction range.
66  *
67  * The HAMMER VFS does most of the work, we add a few new mrecord
68  * types to negotiate the TID ranges and verify that the entire
69  * stream made it to the destination.
70  */
71 void
72 hammer_cmd_mirror_read(char **av, int ac, int streaming)
73 {
74 	struct hammer_ioc_mirror_rw mirror;
75 	struct hammer_ioc_pseudofs_rw pfs;
76 	union hammer_ioc_mrecord_any mrec_tmp;
77 	struct hammer_ioc_mrecord_head pickup;
78 	hammer_ioc_mrecord_any_t mrec;
79 	hammer_tid_t sync_tid;
80 	hammer_tid_t *histogram_ary;
81 	const char *filesystem;
82 	char *buf = malloc(SERIALBUF_SIZE);
83 	int interrupted = 0;
84 	int error;
85 	int fd;
86 	int n;
87 	int didwork;
88 	int histogram;
89 	int64_t total_bytes;
90 	time_t base_t = time(NULL);
91 	struct timeval bwtv;
92 	u_int64_t bwcount;
93 
94 	if (ac == 0 || ac > 2)
95 		mirror_usage(1);
96 	filesystem = av[0];
97 
98 	pickup.signature = 0;
99 	pickup.type = 0;
100 	histogram = -1;
101 	histogram_ary = NULL;
102 
103 again:
104 	bzero(&mirror, sizeof(mirror));
105 	hammer_key_beg_init(&mirror.key_beg);
106 	hammer_key_end_init(&mirror.key_end);
107 
108 	fd = getpfs(&pfs, filesystem);
109 
110 	if (streaming && VerboseOpt) {
111 		fprintf(stderr, "\nRunning");
112 		fflush(stderr);
113 	}
114 	total_bytes = 0;
115 	gettimeofday(&bwtv, NULL);
116 	bwcount = 0;
117 
118 	/*
119 	 * Send initial header for the purpose of determining the
120 	 * shared-uuid.
121 	 */
122 	generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
123 	write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
124 		      &mrec_tmp, sizeof(mrec_tmp.pfs));
125 
126 	/*
127 	 * In 2-way mode the target will send us a PFS info packet
128 	 * first.  Use the target's current snapshot TID as our default
129 	 * begin TID.
130 	 */
131 	mirror.tid_beg = 0;
132 	if (TwoWayPipeOpt) {
133 		n = validate_mrec_header(fd, 0, 0, pfs.pfs_id, &pickup,
134 					 NULL, &mirror.tid_beg);
135 		if (n < 0) {	/* got TERM record */
136 			relpfs(fd, &pfs);
137 			return;
138 		}
139 		++mirror.tid_beg;
140 	}
141 
142 	/*
143 	 * Write out the PFS header, tid_beg will be updated if our PFS
144 	 * has a larger begin sync.  tid_end is set to the latest source
145 	 * TID whos flush cycle has completed.
146 	 */
147 	generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
148 	if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
149 		mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
150 	mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
151 	mirror.ubuf = buf;
152 	mirror.size = SERIALBUF_SIZE;
153 	mirror.pfs_id = pfs.pfs_id;
154 	mirror.shared_uuid = pfs.ondisk->shared_uuid;
155 
156 	/*
157 	 * XXX If the histogram is exhausted and the TID delta is large
158 	 *     the stream might have been offline for a while and is
159 	 *     now picking it up again.  Do another histogram.
160 	 */
161 #if 0
162 	if (TwoWayPipeOpt && streaming && histogram == 0) {
163 		if (mirror.tid_end - mirror.tid_beg > BULK_MINIMUM)
164 			histogram = -1;
165 	}
166 #endif
167 
168 	/*
169 	 * Initial bulk startup control, try to do some incremental
170 	 * mirroring in order to allow the stream to be killed and
171 	 * restarted without having to start over.
172 	 */
173 	if (histogram < 0 && BulkOpt == 0) {
174 		if (VerboseOpt)
175 			fprintf(stderr, "\n");
176 		histogram = generate_histogram(fd, filesystem,
177 					       &histogram_ary, &mirror);
178 	}
179 
180 	if (TwoWayPipeOpt && streaming && histogram > 0) {
181 		mirror.tid_end = histogram_ary[--histogram];
182 		mrec_tmp.pfs.pfsd.sync_end_tid = mirror.tid_end;
183 	}
184 
185 	write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
186 		      &mrec_tmp, sizeof(mrec_tmp.pfs));
187 
188 	/*
189 	 * A cycle file overrides the beginning TID only if we are
190 	 * not operating in two-way mode.
191 	 */
192 	if (TwoWayPipeOpt == 0) {
193 		hammer_get_cycle(&mirror.key_beg, &mirror.tid_beg);
194 	}
195 
196 	/*
197 	 * An additional argument overrides the beginning TID regardless
198 	 * of what mode we are in.  This is not recommending if operating
199 	 * in two-way mode.
200 	 */
201 	if (ac == 2)
202 		mirror.tid_beg = strtoull(av[1], NULL, 0);
203 
204 	if (streaming == 0 || VerboseOpt >= 2) {
205 		fprintf(stderr,
206 			"Mirror-read: Mirror from %016llx to %016llx\n",
207 			mirror.tid_beg, mirror.tid_end);
208 	}
209 	if (mirror.key_beg.obj_id != (int64_t)HAMMER_MIN_OBJID) {
210 		fprintf(stderr, "Mirror-read: Resuming at object %016llx\n",
211 			mirror.key_beg.obj_id);
212 	}
213 
214 	/*
215 	 * Nothing to do if begin equals end.
216 	 */
217 	if (mirror.tid_beg >= mirror.tid_end) {
218 		if (streaming == 0 || VerboseOpt >= 2)
219 			fprintf(stderr, "Mirror-read: No work to do\n");
220 		didwork = 0;
221 		goto done;
222 	}
223 	didwork = 1;
224 
225 	/*
226 	 * Write out bulk records
227 	 */
228 	mirror.ubuf = buf;
229 	mirror.size = SERIALBUF_SIZE;
230 
231 	do {
232 		mirror.count = 0;
233 		mirror.pfs_id = pfs.pfs_id;
234 		mirror.shared_uuid = pfs.ondisk->shared_uuid;
235 		if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
236 			fprintf(stderr, "Mirror-read %s failed: %s\n",
237 				filesystem, strerror(errno));
238 			exit(1);
239 		}
240 		if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
241 			fprintf(stderr,
242 				"Mirror-read %s fatal error %d\n",
243 				filesystem, mirror.head.error);
244 			exit(1);
245 		}
246 		if (mirror.count) {
247 			if (BandwidthOpt) {
248 				n = writebw(1, mirror.ubuf, mirror.count,
249 					    &bwcount, &bwtv);
250 			} else {
251 				n = write(1, mirror.ubuf, mirror.count);
252 			}
253 			if (n != mirror.count) {
254 				fprintf(stderr, "Mirror-read %s failed: "
255 						"short write\n",
256 				filesystem);
257 				exit(1);
258 			}
259 		}
260 		total_bytes += mirror.count;
261 		if (streaming && VerboseOpt) {
262 			fprintf(stderr,
263 				"\robj=%016llx tids=%016llx:%016llx %11lld",
264 				(long long)mirror.key_cur.obj_id,
265 				(long long)mirror.tid_beg,
266 				(long long)mirror.tid_end,
267 				total_bytes);
268 			fflush(stderr);
269 		}
270 		mirror.key_beg = mirror.key_cur;
271 
272 		/*
273 		 * Deal with time limit option
274 		 */
275 		if (TimeoutOpt &&
276 		    (unsigned)(time(NULL) - base_t) > (unsigned)TimeoutOpt) {
277 			fprintf(stderr,
278 				"Mirror-read %s interrupted by timer at"
279 				" %016llx\n",
280 				filesystem,
281 				mirror.key_cur.obj_id);
282 			interrupted = 1;
283 			break;
284 		}
285 	} while (mirror.count != 0);
286 
287 done:
288 	/*
289 	 * Write out the termination sync record - only if not interrupted
290 	 */
291 	if (interrupted == 0) {
292 		if (didwork) {
293 			write_mrecord(1, HAMMER_MREC_TYPE_SYNC,
294 				      &mrec_tmp, sizeof(mrec_tmp.sync));
295 		} else {
296 			write_mrecord(1, HAMMER_MREC_TYPE_IDLE,
297 				      &mrec_tmp, sizeof(mrec_tmp.sync));
298 		}
299 	}
300 
301 	/*
302 	 * If the -2 option was given (automatic when doing mirror-copy),
303 	 * a two-way pipe is assumed and we expect a response mrec from
304 	 * the target.
305 	 */
306 	if (TwoWayPipeOpt) {
307 		mrec = read_mrecord(0, &error, &pickup);
308 		if (mrec == NULL ||
309 		    mrec->head.type != HAMMER_MREC_TYPE_UPDATE ||
310 		    mrec->head.rec_size != sizeof(mrec->update)) {
311 			fprintf(stderr, "mirror_read: Did not get final "
312 					"acknowledgement packet from target\n");
313 			exit(1);
314 		}
315 		if (interrupted) {
316 			if (CyclePath) {
317 				hammer_set_cycle(&mirror.key_cur, mirror.tid_beg);
318 				fprintf(stderr, "Cyclefile %s updated for "
319 					"continuation\n", CyclePath);
320 			}
321 		} else {
322 			sync_tid = mrec->update.tid;
323 			if (CyclePath) {
324 				hammer_key_beg_init(&mirror.key_beg);
325 				hammer_set_cycle(&mirror.key_beg, sync_tid);
326 				fprintf(stderr, "Cyclefile %s updated to 0x%016llx\n",
327 					CyclePath, sync_tid);
328 			}
329 		}
330 	} else if (CyclePath) {
331 		/* NOTE! mirror.tid_beg cannot be updated */
332 		fprintf(stderr, "Warning: cycle file (-c option) cannot be "
333 				"fully updated unless you use mirror-copy\n");
334 		hammer_set_cycle(&mirror.key_beg, mirror.tid_beg);
335 	}
336 	if (streaming && interrupted == 0) {
337 		time_t t1 = time(NULL);
338 		time_t t2;
339 
340 		/*
341 		 * Two way streaming tries to break down large bulk
342 		 * transfers into smaller ones so it can sync the
343 		 * transaction id on the slave.  This way if we get
344 		 * interrupted a restart doesn't have to start from
345 		 * scratch.
346 		 */
347 		if (TwoWayPipeOpt && streaming && histogram > 0) {
348 			if (VerboseOpt)
349 				fprintf(stderr, " (bulk incremental)");
350 			goto again;
351 		}
352 
353 		if (VerboseOpt) {
354 			fprintf(stderr, " W");
355 			fflush(stderr);
356 		}
357 		pfs.ondisk->sync_end_tid = mirror.tid_end;
358 		if (ioctl(fd, HAMMERIOC_WAI_PSEUDOFS, &pfs) < 0) {
359 			fprintf(stderr, "Mirror-read %s: cannot stream: %s\n",
360 				filesystem, strerror(errno));
361 		} else {
362 			t2 = time(NULL) - t1;
363 			if (t2 >= 0 && t2 < DelayOpt) {
364 				if (VerboseOpt) {
365 					fprintf(stderr, "\bD");
366 					fflush(stderr);
367 				}
368 				sleep(DelayOpt - t2);
369 			}
370 			if (VerboseOpt) {
371 				fprintf(stderr, "\b ");
372 				fflush(stderr);
373 			}
374 			relpfs(fd, &pfs);
375 			goto again;
376 		}
377 	}
378 	write_mrecord(1, HAMMER_MREC_TYPE_TERM,
379 		      &mrec_tmp, sizeof(mrec_tmp.sync));
380 	relpfs(fd, &pfs);
381 	fprintf(stderr, "Mirror-read %s succeeded\n", filesystem);
382 }
383 
384 /*
385  * Ok, this isn't really a histogram.  What we are trying to do
386  * here is find the first tid_end for the scan that returns
387  * at least some data.  The frontend of the TID space will generally
388  * return nothing so we can't just divide out the full mirroring
389  * range.  Once we find the point where a real data stream starts
390  * to get generated we can divide out the range from that point.
391  *
392  * When starting a new mirroring operation completely from scratch
393  * this code will take some time to run, but once some mirroring
394  * data is synchronized on the target you will be able to interrupt
395  * the stream and restart it and the later invocations of this
396  * code will be such that it should run much faster.
397  */
398 static int
399 generate_histogram(int fd, const char *filesystem,
400 		   hammer_tid_t **histogram_ary,
401 		   struct hammer_ioc_mirror_rw *mirror_base)
402 {
403 	struct hammer_ioc_mirror_rw mirror;
404 	hammer_tid_t tid_beg;
405 	hammer_tid_t tid_end;
406 	hammer_tid_t tid_half;
407 	int i;
408 
409 	mirror = *mirror_base;
410 	tid_beg = mirror.tid_beg;
411 	tid_end = mirror.tid_end;
412 
413 	if (*histogram_ary)
414 		free(*histogram_ary);
415 	if (tid_beg + BULK_MINIMUM >= tid_end)
416 		return(0);
417 
418 	if (VerboseOpt)
419 		fprintf(stderr, "Doing Range Test\n");
420 	while (tid_end - tid_beg > BULK_MINIMUM) {
421 		tid_half = tid_beg + (tid_end - tid_beg) * 2 / 3;
422 		mirror.count = 0;
423 		mirror.tid_beg = tid_beg;
424 		mirror.tid_end = tid_half;
425 
426 		if (VerboseOpt > 1) {
427 			fprintf(stderr, "RangeTest %016llx/%016llx - %016llx (%lld) ",
428 				(long long)tid_beg,
429 				(long long)tid_end,
430 				(long long)tid_half,
431 				(long long)(tid_half - tid_beg));
432 		}
433 		fflush(stderr);
434 		if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
435 			fprintf(stderr, "Mirror-read %s failed: %s\n",
436 				filesystem, strerror(errno));
437 			exit(1);
438 		}
439 		if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
440 			fprintf(stderr,
441 				"Mirror-read %s fatal error %d\n",
442 				filesystem, mirror.head.error);
443 			exit(1);
444 		}
445 		if (VerboseOpt > 1)
446 			fprintf(stderr, "%d\n", mirror.count);
447 		if (mirror.count > SERIALBUF_SIZE / 2) {
448 			tid_end = tid_half;
449 		} else {
450 			tid_beg = tid_half;
451 		}
452 	}
453 
454 	tid_end = mirror_base->tid_end;
455 	fprintf(stderr, "histogram range %016llx - %016llx\n",
456 		(long long)tid_beg, (long long)tid_end);
457 
458 	/*
459 	 * The final array generates our incremental ending tids in
460 	 * reverse order.  The caller also picks them off in reverse order.
461 	 */
462 	*histogram_ary = malloc(sizeof(hammer_tid_t) * 20);
463 	for (i = 0; i < 20; ++i) {
464 		(*histogram_ary)[i] = tid_end - (tid_end - tid_beg) / 20 * i;
465 	}
466 	return(20);
467 }
468 
469 static void
470 create_pfs(const char *filesystem, uuid_t *s_uuid)
471 {
472 	if (ForceYesOpt == 1) {
473 		fprintf(stderr, "PFS slave %s does not exist. "
474 			"Auto create new slave PFS!\n", filesystem);
475 
476 	} else {
477 		fprintf(stderr, "PFS slave %s does not exist.\n"
478 			"Do you want to create a new slave PFS? (yes|no) ",
479 			filesystem);
480 		fflush(stderr);
481 		if (getyn() != 1) {
482 			fprintf(stderr, "Aborting operation\n");
483 			exit(1);
484 		}
485 	}
486 
487 	u_int32_t status;
488 	char *shared_uuid = NULL;
489 	uuid_to_string(s_uuid, &shared_uuid, &status);
490 
491 	char *cmd = NULL;
492 	asprintf(&cmd, "/sbin/hammer pfs-slave '%s' shared-uuid=%s 1>&2",
493 		 filesystem, shared_uuid);
494 	free(shared_uuid);
495 
496 	if (cmd == NULL) {
497 		fprintf(stderr, "Failed to alloc memory\n");
498 		exit(1);
499 	}
500 	if (system(cmd) != 0) {
501 		fprintf(stderr, "Failed to create PFS\n");
502 	}
503 	free(cmd);
504 }
505 
506 /*
507  * Pipe the mirroring data stream on stdin to the HAMMER VFS, adding
508  * some additional packet types to negotiate TID ranges and to verify
509  * completion.  The HAMMER VFS does most of the work.
510  *
511  * It is important to note that the mirror.key_{beg,end} range must
512  * match the ranged used by the original.  For now both sides use
513  * range the entire key space.
514  *
515  * It is even more important that the records in the stream conform
516  * to the TID range also supplied in the stream.  The HAMMER VFS will
517  * use the REC, PASS, and SKIP record types to track the portions of
518  * the B-Tree being scanned in order to be able to proactively delete
519  * records on the target within those active areas that are not mentioned
520  * by the source.
521  *
522  * The mirror.key_cur field is used by the VFS to do this tracking.  It
523  * must be initialized to key_beg but then is persistently updated by
524  * the HAMMER VFS on each successive ioctl() call.  If you blow up this
525  * field you will blow up the mirror target, possibly to the point of
526  * deleting everything.  As a safety measure the HAMMER VFS simply marks
527  * the records that the source has destroyed as deleted on the target,
528  * and normal pruning operations will deal with their final disposition
529  * at some later time.
530  */
531 void
532 hammer_cmd_mirror_write(char **av, int ac)
533 {
534 	struct hammer_ioc_mirror_rw mirror;
535 	const char *filesystem;
536 	char *buf = malloc(SERIALBUF_SIZE);
537 	struct hammer_ioc_pseudofs_rw pfs;
538 	struct hammer_ioc_mrecord_head pickup;
539 	struct hammer_ioc_synctid synctid;
540 	union hammer_ioc_mrecord_any mrec_tmp;
541 	hammer_ioc_mrecord_any_t mrec;
542 	struct stat st;
543 	int error;
544 	int fd;
545 	int n;
546 
547 	if (ac != 1)
548 		mirror_usage(1);
549 	filesystem = av[0];
550 
551 	pickup.signature = 0;
552 	pickup.type = 0;
553 
554 again:
555 	bzero(&mirror, sizeof(mirror));
556 	hammer_key_beg_init(&mirror.key_beg);
557 	hammer_key_end_init(&mirror.key_end);
558 	mirror.key_end = mirror.key_beg;
559 
560 	/*
561 	 * Read initial packet
562 	 */
563 	mrec = read_mrecord(0, &error, &pickup);
564 	if (mrec == NULL) {
565 		if (error == 0)
566 			fprintf(stderr, "validate_mrec_header: short read\n");
567 		exit(1);
568 	}
569 	/*
570 	 * Validate packet
571 	 */
572 	if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
573 		return;
574 	}
575 	if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
576 		fprintf(stderr, "validate_mrec_header: did not get expected "
577 				"PFSD record type\n");
578 		exit(1);
579 	}
580 	if (mrec->head.rec_size != sizeof(mrec->pfs)) {
581 		fprintf(stderr, "validate_mrec_header: unexpected payload "
582 				"size\n");
583 		exit(1);
584 	}
585 
586 	/*
587 	 * Create slave PFS if it doesn't yet exist
588 	 */
589 	if (lstat(filesystem, &st) != 0) {
590 		create_pfs(filesystem, &mrec->pfs.pfsd.shared_uuid);
591 	}
592 	free(mrec);
593 	mrec = NULL;
594 
595 	fd = getpfs(&pfs, filesystem);
596 
597 	/*
598 	 * In two-way mode the target writes out a PFS packet first.
599 	 * The source uses our tid_end as its tid_beg by default,
600 	 * picking up where it left off.
601 	 */
602 	mirror.tid_beg = 0;
603 	if (TwoWayPipeOpt) {
604 		generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
605 		if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
606 			mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
607 		mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
608 		write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
609 			      &mrec_tmp, sizeof(mrec_tmp.pfs));
610 	}
611 
612 	/*
613 	 * Read and process the PFS header.  The source informs us of
614 	 * the TID range the stream represents.
615 	 */
616 	n = validate_mrec_header(fd, 0, 1, pfs.pfs_id, &pickup,
617 				 &mirror.tid_beg, &mirror.tid_end);
618 	if (n < 0) {	/* got TERM record */
619 		relpfs(fd, &pfs);
620 		return;
621 	}
622 
623 	mirror.ubuf = buf;
624 	mirror.size = SERIALBUF_SIZE;
625 
626 	/*
627 	 * Read and process bulk records (REC, PASS, and SKIP types).
628 	 *
629 	 * On your life, do NOT mess with mirror.key_cur or your mirror
630 	 * target may become history.
631 	 */
632 	for (;;) {
633 		mirror.count = 0;
634 		mirror.pfs_id = pfs.pfs_id;
635 		mirror.shared_uuid = pfs.ondisk->shared_uuid;
636 		mirror.size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
637 		if (mirror.size <= 0)
638 			break;
639 		if (ioctl(fd, HAMMERIOC_MIRROR_WRITE, &mirror) < 0) {
640 			fprintf(stderr, "Mirror-write %s failed: %s\n",
641 				filesystem, strerror(errno));
642 			exit(1);
643 		}
644 		if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
645 			fprintf(stderr,
646 				"Mirror-write %s fatal error %d\n",
647 				filesystem, mirror.head.error);
648 			exit(1);
649 		}
650 #if 0
651 		if (mirror.head.flags & HAMMER_IOC_HEAD_INTR) {
652 			fprintf(stderr,
653 				"Mirror-write %s interrupted by timer at"
654 				" %016llx\n",
655 				filesystem,
656 				mirror.key_cur.obj_id);
657 			exit(0);
658 		}
659 #endif
660 	}
661 
662 	/*
663 	 * Read and process the termination sync record.
664 	 */
665 	mrec = read_mrecord(0, &error, &pickup);
666 
667 	if (mrec && mrec->head.type == HAMMER_MREC_TYPE_TERM) {
668 		fprintf(stderr, "Mirror-write: received termination request\n");
669 		free(mrec);
670 		return;
671 	}
672 
673 	if (mrec == NULL ||
674 	    (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
675 	     mrec->head.type != HAMMER_MREC_TYPE_IDLE) ||
676 	    mrec->head.rec_size != sizeof(mrec->sync)) {
677 		fprintf(stderr, "Mirror-write %s: Did not get termination "
678 				"sync record, or rec_size is wrong rt=%d\n",
679 				filesystem, mrec->head.type);
680 		exit(1);
681 	}
682 
683 	/*
684 	 * Update the PFS info on the target so the user has visibility
685 	 * into the new snapshot, and sync the target filesystem.
686 	 */
687 	if (mrec->head.type == HAMMER_MREC_TYPE_SYNC) {
688 		update_pfs_snapshot(fd, mirror.tid_end, pfs.pfs_id);
689 
690 		bzero(&synctid, sizeof(synctid));
691 		synctid.op = HAMMER_SYNCTID_SYNC2;
692 		ioctl(fd, HAMMERIOC_SYNCTID, &synctid);
693 
694 		if (VerboseOpt >= 2) {
695 			fprintf(stderr, "Mirror-write %s: succeeded\n",
696 				filesystem);
697 		}
698 	}
699 
700 	free(mrec);
701 	mrec = NULL;
702 
703 	/*
704 	 * Report back to the originator.
705 	 */
706 	if (TwoWayPipeOpt) {
707 		mrec_tmp.update.tid = mirror.tid_end;
708 		write_mrecord(1, HAMMER_MREC_TYPE_UPDATE,
709 			      &mrec_tmp, sizeof(mrec_tmp.update));
710 	} else {
711 		printf("Source can update synctid to 0x%016llx\n",
712 		       mirror.tid_end);
713 	}
714 	relpfs(fd, &pfs);
715 	goto again;
716 }
717 
718 void
719 hammer_cmd_mirror_dump(void)
720 {
721 	char *buf = malloc(SERIALBUF_SIZE);
722 	struct hammer_ioc_mrecord_head pickup;
723 	hammer_ioc_mrecord_any_t mrec;
724 	int error;
725 	int size;
726 	int offset;
727 	int bytes;
728 
729 	/*
730 	 * Read and process the PFS header
731 	 */
732 	pickup.signature = 0;
733 	pickup.type = 0;
734 
735 	mrec = read_mrecord(0, &error, &pickup);
736 
737 	/*
738 	 * Read and process bulk records
739 	 */
740 	for (;;) {
741 		size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
742 		if (size <= 0)
743 			break;
744 		offset = 0;
745 		while (offset < size) {
746 			mrec = (void *)((char *)buf + offset);
747 			bytes = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
748 			if (offset + bytes > size) {
749 				fprintf(stderr, "Misaligned record\n");
750 				exit(1);
751 			}
752 
753 			switch(mrec->head.type & HAMMER_MRECF_TYPE_MASK) {
754 			case HAMMER_MREC_TYPE_REC_BADCRC:
755 			case HAMMER_MREC_TYPE_REC:
756 				printf("Record obj=%016llx key=%016llx "
757 				       "rt=%02x ot=%02x",
758 					mrec->rec.leaf.base.obj_id,
759 					mrec->rec.leaf.base.key,
760 					mrec->rec.leaf.base.rec_type,
761 					mrec->rec.leaf.base.obj_type);
762 				if (mrec->head.type ==
763 				    HAMMER_MREC_TYPE_REC_BADCRC) {
764 					printf(" (BAD CRC)");
765 				}
766 				printf("\n");
767 				printf("       tids %016llx:%016llx data=%d\n",
768 					mrec->rec.leaf.base.create_tid,
769 					mrec->rec.leaf.base.delete_tid,
770 					mrec->rec.leaf.data_len);
771 				break;
772 			case HAMMER_MREC_TYPE_PASS:
773 				printf("Pass   obj=%016llx key=%016llx "
774 				       "rt=%02x ot=%02x\n",
775 					mrec->rec.leaf.base.obj_id,
776 					mrec->rec.leaf.base.key,
777 					mrec->rec.leaf.base.rec_type,
778 					mrec->rec.leaf.base.obj_type);
779 				printf("       tids %016llx:%016llx data=%d\n",
780 					mrec->rec.leaf.base.create_tid,
781 					mrec->rec.leaf.base.delete_tid,
782 					mrec->rec.leaf.data_len);
783 				break;
784 			case HAMMER_MREC_TYPE_SKIP:
785 				printf("Skip   obj=%016llx key=%016llx rt=%02x to\n"
786 				       "       obj=%016llx key=%016llx rt=%02x\n",
787 				       mrec->skip.skip_beg.obj_id,
788 				       mrec->skip.skip_beg.key,
789 				       mrec->skip.skip_beg.rec_type,
790 				       mrec->skip.skip_end.obj_id,
791 				       mrec->skip.skip_end.key,
792 				       mrec->skip.skip_end.rec_type);
793 			default:
794 				break;
795 			}
796 			offset += bytes;
797 		}
798 	}
799 
800 	/*
801 	 * Read and process the termination sync record.
802 	 */
803 	mrec = read_mrecord(0, &error, &pickup);
804 	if (mrec == NULL ||
805 	    (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
806 	     mrec->head.type != HAMMER_MREC_TYPE_IDLE)
807 	 ) {
808 		fprintf(stderr, "Mirror-dump: Did not get termination "
809 				"sync record\n");
810 	}
811 }
812 
813 void
814 hammer_cmd_mirror_copy(char **av, int ac, int streaming)
815 {
816 	pid_t pid1;
817 	pid_t pid2;
818 	int fds[2];
819 	const char *xav[16];
820 	char tbuf[16];
821 	char *ptr;
822 	int xac;
823 
824 	if (ac != 2)
825 		mirror_usage(1);
826 
827 	TwoWayPipeOpt = 1;
828 	signal(SIGPIPE, SIG_IGN);
829 
830 again:
831 	if (pipe(fds) < 0) {
832 		perror("pipe");
833 		exit(1);
834 	}
835 
836 	/*
837 	 * Source
838 	 */
839 	if ((pid1 = fork()) == 0) {
840 		signal(SIGPIPE, SIG_DFL);
841 		dup2(fds[0], 0);
842 		dup2(fds[0], 1);
843 		close(fds[0]);
844 		close(fds[1]);
845 		if ((ptr = strchr(av[0], ':')) != NULL) {
846 			*ptr++ = 0;
847 			xac = 0;
848 			xav[xac++] = "ssh";
849 			xav[xac++] = av[0];
850 			xav[xac++] = "hammer";
851 
852 			switch(VerboseOpt) {
853 			case 0:
854 				break;
855 			case 1:
856 				xav[xac++] = "-v";
857 				break;
858 			case 2:
859 				xav[xac++] = "-vv";
860 				break;
861 			default:
862 				xav[xac++] = "-vvv";
863 				break;
864 			}
865 			if (ForceYesOpt) {
866 				xav[xac++] = "-y";
867 			}
868 			xav[xac++] = "-2";
869 			if (TimeoutOpt) {
870 				snprintf(tbuf, sizeof(tbuf), "%d", TimeoutOpt);
871 				xav[xac++] = "-t";
872 				xav[xac++] = tbuf;
873 			}
874 			if (streaming)
875 				xav[xac++] = "mirror-read-stream";
876 			else
877 				xav[xac++] = "mirror-read";
878 			xav[xac++] = ptr;
879 			xav[xac++] = NULL;
880 			execv("/usr/bin/ssh", (void *)xav);
881 		} else {
882 			hammer_cmd_mirror_read(av, 1, streaming);
883 			fflush(stdout);
884 			fflush(stderr);
885 		}
886 		_exit(1);
887 	}
888 
889 	/*
890 	 * Target
891 	 */
892 	if ((pid2 = fork()) == 0) {
893 		signal(SIGPIPE, SIG_DFL);
894 		dup2(fds[1], 0);
895 		dup2(fds[1], 1);
896 		close(fds[0]);
897 		close(fds[1]);
898 		if ((ptr = strchr(av[1], ':')) != NULL) {
899 			*ptr++ = 0;
900 			xac = 0;
901 			xav[xac++] = "ssh";
902 			xav[xac++] = av[1];
903 			xav[xac++] = "hammer";
904 
905 			switch(VerboseOpt) {
906 			case 0:
907 				break;
908 			case 1:
909 				xav[xac++] = "-v";
910 				break;
911 			case 2:
912 				xav[xac++] = "-vv";
913 				break;
914 			default:
915 				xav[xac++] = "-vvv";
916 				break;
917 			}
918 			if (ForceYesOpt) {
919 				xav[xac++] = "-y";
920 			}
921 			xav[xac++] = "-2";
922 			xav[xac++] = "mirror-write";
923 			xav[xac++] = ptr;
924 			xav[xac++] = NULL;
925 			execv("/usr/bin/ssh", (void *)xav);
926 		} else {
927 			hammer_cmd_mirror_write(av + 1, 1);
928 			fflush(stdout);
929 			fflush(stderr);
930 		}
931 		_exit(1);
932 	}
933 	close(fds[0]);
934 	close(fds[1]);
935 
936 	while (waitpid(pid1, NULL, 0) <= 0)
937 		;
938 	while (waitpid(pid2, NULL, 0) <= 0)
939 		;
940 
941 	/*
942 	 * If the link is lost restart
943 	 */
944 	if (streaming) {
945 		if (VerboseOpt) {
946 			fprintf(stderr, "\nLost Link\n");
947 			fflush(stderr);
948 		}
949 		sleep(15 + DelayOpt);
950 		goto again;
951 	}
952 
953 }
954 
955 /*
956  * Read and return multiple mrecords
957  */
958 static int
959 read_mrecords(int fd, char *buf, u_int size, hammer_ioc_mrecord_head_t pickup)
960 {
961 	hammer_ioc_mrecord_any_t mrec;
962 	u_int count;
963 	size_t n;
964 	size_t i;
965 	size_t bytes;
966 	int type;
967 
968 	count = 0;
969 	while (size - count >= HAMMER_MREC_HEADSIZE) {
970 		/*
971 		 * Cached the record header in case we run out of buffer
972 		 * space.
973 		 */
974 		fflush(stdout);
975 		if (pickup->signature == 0) {
976 			for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
977 				i = read(fd, (char *)pickup + n,
978 					 HAMMER_MREC_HEADSIZE - n);
979 				if (i <= 0)
980 					break;
981 			}
982 			if (n == 0)
983 				break;
984 			if (n != HAMMER_MREC_HEADSIZE) {
985 				fprintf(stderr, "read_mrecords: short read on pipe\n");
986 				exit(1);
987 			}
988 			if (pickup->signature != HAMMER_IOC_MIRROR_SIGNATURE) {
989 				fprintf(stderr, "read_mrecords: malformed record on pipe, "
990 					"bad signature\n");
991 				exit(1);
992 			}
993 		}
994 		if (pickup->rec_size < HAMMER_MREC_HEADSIZE ||
995 		    pickup->rec_size > sizeof(*mrec) + HAMMER_XBUFSIZE) {
996 			fprintf(stderr, "read_mrecords: malformed record on pipe, "
997 				"illegal rec_size\n");
998 			exit(1);
999 		}
1000 
1001 		/*
1002 		 * Stop if we have insufficient space for the record and data.
1003 		 */
1004 		bytes = HAMMER_HEAD_DOALIGN(pickup->rec_size);
1005 		if (size - count < bytes)
1006 			break;
1007 
1008 		/*
1009 		 * Stop if the record type is not a REC, SKIP, or PASS,
1010 		 * which are the only types the ioctl supports.  Other types
1011 		 * are used only by the userland protocol.
1012 		 *
1013 		 * Ignore all flags.
1014 		 */
1015 		type = pickup->type & HAMMER_MRECF_TYPE_LOMASK;
1016 		if (type != HAMMER_MREC_TYPE_PFSD &&
1017 		    type != HAMMER_MREC_TYPE_REC &&
1018 		    type != HAMMER_MREC_TYPE_SKIP &&
1019 		    type != HAMMER_MREC_TYPE_PASS) {
1020 			break;
1021 		}
1022 
1023 		/*
1024 		 * Read the remainder and clear the pickup signature.
1025 		 */
1026 		for (n = HAMMER_MREC_HEADSIZE; n < bytes; n += i) {
1027 			i = read(fd, buf + count + n, bytes - n);
1028 			if (i <= 0)
1029 				break;
1030 		}
1031 		if (n != bytes) {
1032 			fprintf(stderr, "read_mrecords: short read on pipe\n");
1033 			exit(1);
1034 		}
1035 
1036 		bcopy(pickup, buf + count, HAMMER_MREC_HEADSIZE);
1037 		pickup->signature = 0;
1038 		pickup->type = 0;
1039 		mrec = (void *)(buf + count);
1040 
1041 		/*
1042 		 * Validate the completed record
1043 		 */
1044 		if (mrec->head.rec_crc !=
1045 		    crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1046 			  mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1047 			fprintf(stderr, "read_mrecords: malformed record "
1048 					"on pipe, bad crc\n");
1049 			exit(1);
1050 		}
1051 
1052 		/*
1053 		 * If its a B-Tree record validate the data crc.
1054 		 *
1055 		 * NOTE: If the VFS passes us an explicitly errorde mrec
1056 		 *	 we just pass it through.
1057 		 */
1058 		type = mrec->head.type & HAMMER_MRECF_TYPE_MASK;
1059 
1060 		if (type == HAMMER_MREC_TYPE_REC) {
1061 			if (mrec->head.rec_size <
1062 			    sizeof(mrec->rec) + mrec->rec.leaf.data_len) {
1063 				fprintf(stderr,
1064 					"read_mrecords: malformed record on "
1065 					"pipe, illegal element data_len\n");
1066 				exit(1);
1067 			}
1068 			if (mrec->rec.leaf.data_len &&
1069 			    mrec->rec.leaf.data_offset &&
1070 			    hammer_crc_test_leaf(&mrec->rec + 1, &mrec->rec.leaf) == 0) {
1071 				fprintf(stderr,
1072 					"read_mrecords: data_crc did not "
1073 					"match data! obj=%016llx key=%016llx\n",
1074 					mrec->rec.leaf.base.obj_id,
1075 					mrec->rec.leaf.base.key);
1076 				fprintf(stderr,
1077 					"continuing, but there are problems\n");
1078 			}
1079 		}
1080 		count += bytes;
1081 	}
1082 	return(count);
1083 }
1084 
1085 /*
1086  * Read and return a single mrecord.
1087  */
1088 static
1089 hammer_ioc_mrecord_any_t
1090 read_mrecord(int fdin, int *errorp, hammer_ioc_mrecord_head_t pickup)
1091 {
1092 	hammer_ioc_mrecord_any_t mrec;
1093 	struct hammer_ioc_mrecord_head mrechd;
1094 	size_t bytes;
1095 	size_t n;
1096 	size_t i;
1097 
1098 	if (pickup && pickup->type != 0) {
1099 		mrechd = *pickup;
1100 		pickup->signature = 0;
1101 		pickup->type = 0;
1102 		n = HAMMER_MREC_HEADSIZE;
1103 	} else {
1104 		/*
1105 		 * Read in the PFSD header from the sender.
1106 		 */
1107 		for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1108 			i = read(fdin, (char *)&mrechd + n, HAMMER_MREC_HEADSIZE - n);
1109 			if (i <= 0)
1110 				break;
1111 		}
1112 		if (n == 0) {
1113 			*errorp = 0;	/* EOF */
1114 			return(NULL);
1115 		}
1116 		if (n != HAMMER_MREC_HEADSIZE) {
1117 			fprintf(stderr, "short read of mrecord header\n");
1118 			*errorp = EPIPE;
1119 			return(NULL);
1120 		}
1121 	}
1122 	if (mrechd.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1123 		fprintf(stderr, "read_mrecord: bad signature\n");
1124 		*errorp = EINVAL;
1125 		return(NULL);
1126 	}
1127 	bytes = HAMMER_HEAD_DOALIGN(mrechd.rec_size);
1128 	assert(bytes >= sizeof(mrechd));
1129 	mrec = malloc(bytes);
1130 	mrec->head = mrechd;
1131 
1132 	while (n < bytes) {
1133 		i = read(fdin, (char *)mrec + n, bytes - n);
1134 		if (i <= 0)
1135 			break;
1136 		n += i;
1137 	}
1138 	if (n != bytes) {
1139 		fprintf(stderr, "read_mrecord: short read on payload\n");
1140 		*errorp = EPIPE;
1141 		return(NULL);
1142 	}
1143 	if (mrec->head.rec_crc !=
1144 	    crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1145 		  mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1146 		fprintf(stderr, "read_mrecord: bad CRC\n");
1147 		*errorp = EINVAL;
1148 		return(NULL);
1149 	}
1150 	*errorp = 0;
1151 	return(mrec);
1152 }
1153 
1154 static
1155 void
1156 write_mrecord(int fdout, u_int32_t type, hammer_ioc_mrecord_any_t mrec,
1157 	      int bytes)
1158 {
1159 	char zbuf[HAMMER_HEAD_ALIGN];
1160 	int pad;
1161 
1162 	pad = HAMMER_HEAD_DOALIGN(bytes) - bytes;
1163 
1164 	assert(bytes >= (int)sizeof(mrec->head));
1165 	bzero(&mrec->head, sizeof(mrec->head));
1166 	mrec->head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
1167 	mrec->head.type = type;
1168 	mrec->head.rec_size = bytes;
1169 	mrec->head.rec_crc = crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1170 				   bytes - HAMMER_MREC_CRCOFF);
1171 	if (write(fdout, mrec, bytes) != bytes) {
1172 		fprintf(stderr, "write_mrecord: error %d (%s)\n",
1173 			errno, strerror(errno));
1174 		exit(1);
1175 	}
1176 	if (pad) {
1177 		bzero(zbuf, pad);
1178 		if (write(fdout, zbuf, pad) != pad) {
1179 			fprintf(stderr, "write_mrecord: error %d (%s)\n",
1180 				errno, strerror(errno));
1181 			exit(1);
1182 		}
1183 	}
1184 }
1185 
1186 /*
1187  * Generate a mirroring header with the pfs information of the
1188  * originating filesytem.
1189  */
1190 static void
1191 generate_mrec_header(int fd, int pfs_id,
1192 		     union hammer_ioc_mrecord_any *mrec_tmp)
1193 {
1194 	struct hammer_ioc_pseudofs_rw pfs;
1195 
1196 	bzero(&pfs, sizeof(pfs));
1197 	bzero(mrec_tmp, sizeof(*mrec_tmp));
1198 	pfs.pfs_id = pfs_id;
1199 	pfs.ondisk = &mrec_tmp->pfs.pfsd;
1200 	pfs.bytes = sizeof(mrec_tmp->pfs.pfsd);
1201 	if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1202 		fprintf(stderr, "Mirror-read: not a HAMMER fs/pseudofs!\n");
1203 		exit(1);
1204 	}
1205 	if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1206 		fprintf(stderr, "Mirror-read: HAMMER pfs version mismatch!\n");
1207 		exit(1);
1208 	}
1209 	mrec_tmp->pfs.version = pfs.version;
1210 }
1211 
1212 /*
1213  * Validate the pfs information from the originating filesystem
1214  * against the target filesystem.  shared_uuid must match.
1215  *
1216  * return -1 if we got a TERM record
1217  */
1218 static int
1219 validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
1220 		     struct hammer_ioc_mrecord_head *pickup,
1221 		     hammer_tid_t *tid_begp, hammer_tid_t *tid_endp)
1222 {
1223 	struct hammer_ioc_pseudofs_rw pfs;
1224 	struct hammer_pseudofs_data pfsd;
1225 	hammer_ioc_mrecord_any_t mrec;
1226 	int error;
1227 
1228 	/*
1229 	 * Get the PFSD info from the target filesystem.
1230 	 */
1231 	bzero(&pfs, sizeof(pfs));
1232 	bzero(&pfsd, sizeof(pfsd));
1233 	pfs.pfs_id = pfs_id;
1234 	pfs.ondisk = &pfsd;
1235 	pfs.bytes = sizeof(pfsd);
1236 	if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1237 		fprintf(stderr, "mirror-write: not a HAMMER fs/pseudofs!\n");
1238 		exit(1);
1239 	}
1240 	if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1241 		fprintf(stderr, "mirror-write: HAMMER pfs version mismatch!\n");
1242 		exit(1);
1243 	}
1244 
1245 	mrec = read_mrecord(fdin, &error, pickup);
1246 	if (mrec == NULL) {
1247 		if (error == 0)
1248 			fprintf(stderr, "validate_mrec_header: short read\n");
1249 		exit(1);
1250 	}
1251 	if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
1252 		free(mrec);
1253 		return(-1);
1254 	}
1255 
1256 	if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
1257 		fprintf(stderr, "validate_mrec_header: did not get expected "
1258 				"PFSD record type\n");
1259 		exit(1);
1260 	}
1261 	if (mrec->head.rec_size != sizeof(mrec->pfs)) {
1262 		fprintf(stderr, "validate_mrec_header: unexpected payload "
1263 				"size\n");
1264 		exit(1);
1265 	}
1266 	if (mrec->pfs.version != pfs.version) {
1267 		fprintf(stderr, "validate_mrec_header: Version mismatch\n");
1268 		exit(1);
1269 	}
1270 
1271 	/*
1272 	 * Whew.  Ok, is the read PFS info compatible with the target?
1273 	 */
1274 	if (bcmp(&mrec->pfs.pfsd.shared_uuid, &pfsd.shared_uuid,
1275 		 sizeof(pfsd.shared_uuid)) != 0) {
1276 		fprintf(stderr,
1277 			"mirror-write: source and target have "
1278 			"different shared-uuid's!\n");
1279 		exit(1);
1280 	}
1281 	if (is_target &&
1282 	    (pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
1283 		fprintf(stderr, "mirror-write: target must be in slave mode\n");
1284 		exit(1);
1285 	}
1286 	if (tid_begp)
1287 		*tid_begp = mrec->pfs.pfsd.sync_beg_tid;
1288 	if (tid_endp)
1289 		*tid_endp = mrec->pfs.pfsd.sync_end_tid;
1290 	free(mrec);
1291 	return(0);
1292 }
1293 
1294 static void
1295 update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id)
1296 {
1297 	struct hammer_ioc_pseudofs_rw pfs;
1298 	struct hammer_pseudofs_data pfsd;
1299 
1300 	bzero(&pfs, sizeof(pfs));
1301 	bzero(&pfsd, sizeof(pfsd));
1302 	pfs.pfs_id = pfs_id;
1303 	pfs.ondisk = &pfsd;
1304 	pfs.bytes = sizeof(pfsd);
1305 	if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1306 		perror("update_pfs_snapshot (read)");
1307 		exit(1);
1308 	}
1309 	if (pfsd.sync_end_tid != snapshot_tid) {
1310 		pfsd.sync_end_tid = snapshot_tid;
1311 		if (ioctl(fd, HAMMERIOC_SET_PSEUDOFS, &pfs) != 0) {
1312 			perror("update_pfs_snapshot (rewrite)");
1313 			exit(1);
1314 		}
1315 		if (VerboseOpt >= 2) {
1316 			fprintf(stderr,
1317 				"Mirror-write: Completed, updated snapshot "
1318 				"to %016llx\n",
1319 				snapshot_tid);
1320 		}
1321 	}
1322 }
1323 
1324 /*
1325  * Bandwidth-limited write in chunks
1326  */
1327 static
1328 ssize_t
1329 writebw(int fd, const void *buf, size_t nbytes,
1330 	u_int64_t *bwcount, struct timeval *tv1)
1331 {
1332 	struct timeval tv2;
1333 	size_t n;
1334 	ssize_t r;
1335 	ssize_t a;
1336 	int usec;
1337 
1338 	a = 0;
1339 	r = 0;
1340 	while (nbytes) {
1341 		if (*bwcount + nbytes > BandwidthOpt)
1342 			n = BandwidthOpt - *bwcount;
1343 		else
1344 			n = nbytes;
1345 		if (n)
1346 			r = write(fd, buf, n);
1347 		if (r >= 0) {
1348 			a += r;
1349 			nbytes -= r;
1350 			buf = (const char *)buf + r;
1351 		}
1352 		if ((size_t)r != n)
1353 			break;
1354 		*bwcount += n;
1355 		if (*bwcount >= BandwidthOpt) {
1356 			gettimeofday(&tv2, NULL);
1357 			usec = (int)(tv2.tv_sec - tv1->tv_sec) * 1000000 +
1358 				(int)(tv2.tv_usec - tv1->tv_usec);
1359 			if (usec >= 0 && usec < 1000000)
1360 				usleep(1000000 - usec);
1361 			gettimeofday(tv1, NULL);
1362 			*bwcount -= BandwidthOpt;
1363 		}
1364 	}
1365 	return(a ? a : r);
1366 }
1367 
1368 /*
1369  * Get a yes or no answer from the terminal.  The program may be run as
1370  * part of a two-way pipe so we cannot use stdin for this operation.
1371  */
1372 static int
1373 getyn(void)
1374 {
1375 	char buf[256];
1376 	FILE *fp;
1377 	int result;
1378 
1379 	fp = fopen("/dev/tty", "r");
1380 	if (fp == NULL) {
1381 		fprintf(stderr, "No terminal for response\n");
1382 		return(-1);
1383 	}
1384 	result = -1;
1385 	while (fgets(buf, sizeof(buf), fp) != NULL) {
1386 		if (buf[0] == 'y' || buf[0] == 'Y') {
1387 			result = 1;
1388 			break;
1389 		}
1390 		if (buf[0] == 'n' || buf[0] == 'N') {
1391 			result = 0;
1392 			break;
1393 		}
1394 		fprintf(stderr, "Response not understood\n");
1395 		break;
1396 	}
1397 	fclose(fp);
1398 	return(result);
1399 }
1400 
1401 static void
1402 mirror_usage(int code)
1403 {
1404 	fprintf(stderr,
1405 		"hammer mirror-read <filesystem> [begin-tid]\n"
1406 		"hammer mirror-read-stream <filesystem> [begin-tid]\n"
1407 		"hammer mirror-write <filesystem>\n"
1408 		"hammer mirror-dump\n"
1409 		"hammer mirror-copy [[user@]host:]<filesystem>"
1410 				  " [[user@]host:]<filesystem>\n"
1411 		"hammer mirror-stream [[user@]host:]<filesystem>"
1412 				    " [[user@]host:]<filesystem>\n"
1413 	);
1414 	exit(code);
1415 }
1416 
1417