xref: /dflybsd-src/sbin/hammer/cmd_mirror.c (revision 88cdee70b9d467aa0dffa5f24852b83db8c385ef)
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 
35 #include "hammer.h"
36 
37 #define LINE1	0,20
38 #define LINE2	20,78
39 #define LINE3	90,70
40 
41 #define SERIALBUF_SIZE	(512 * 1024)
42 
43 typedef struct histogram {
44 	hammer_tid_t	tid;
45 	uint64_t	bytes;
46 } *histogram_t;
47 
48 const char *ScoreBoardFile;
49 const char *RestrictTarget;
50 
51 static int read_mrecords(int fd, char *buf, u_int size,
52 			 hammer_ioc_mrecord_head_t pickup);
53 static int generate_histogram(int fd, const char *filesystem,
54 			 histogram_t *histogram_ary,
55 			 struct hammer_ioc_mirror_rw *mirror_base,
56 			 int *repeatp);
57 static hammer_ioc_mrecord_any_t read_mrecord(int fdin, int *errorp,
58 			 hammer_ioc_mrecord_head_t pickup);
59 static void write_mrecord(int fdout, uint32_t type,
60 			 hammer_ioc_mrecord_any_t mrec, int bytes);
61 static void generate_mrec_header(int fd, int pfs_id,
62 			 union hammer_ioc_mrecord_any *mrec_tmp);
63 static int validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
64 			 struct hammer_ioc_mrecord_head *pickup,
65 			 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp);
66 static void update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id);
67 static ssize_t writebw(int fd, const void *buf, size_t nbytes,
68 			uint64_t *bwcount, struct timeval *tv1);
69 static int getyntty(void);
70 static void score_printf(size_t i, size_t w, const char *ctl, ...);
71 static void hammer_check_restrict(const char *filesystem);
72 static void mirror_usage(int code);
73 
74 /*
75  * Generate a mirroring data stream from the specific source over the
76  * entire key range, but restricted to the specified transaction range.
77  *
78  * The HAMMER VFS does most of the work, we add a few new mrecord
79  * types to negotiate the TID ranges and verify that the entire
80  * stream made it to the destination.
81  *
82  * streaming will be 0 for mirror-read, 1 for mirror-stream.  The code will
83  * set up a fake value of -1 when running the histogram for mirror-read.
84  */
85 void
86 hammer_cmd_mirror_read(char **av, int ac, int streaming)
87 {
88 	struct hammer_ioc_mirror_rw mirror;
89 	struct hammer_ioc_pseudofs_rw pfs;
90 	union hammer_ioc_mrecord_any mrec_tmp;
91 	struct hammer_ioc_mrecord_head pickup;
92 	hammer_ioc_mrecord_any_t mrec;
93 	hammer_tid_t sync_tid;
94 	histogram_t histogram_ary;
95 	const char *filesystem;
96 	char *buf = malloc(SERIALBUF_SIZE);
97 	int interrupted = 0;
98 	int error;
99 	int fd;
100 	int n;
101 	int didwork;
102 	int histogram;
103 	int histindex;
104 	int histmax;
105 	int repeat = 0;
106 	int sameline;
107 	int64_t total_bytes;
108 	time_t base_t = time(NULL);
109 	struct timeval bwtv;
110 	uint64_t bwcount;
111 	uint64_t estbytes;
112 
113 	if (ac == 0 || ac > 2) {
114 		mirror_usage(1);
115 		/* not reached */
116 	}
117 	filesystem = av[0];
118 	hammer_check_restrict(filesystem);
119 
120 	pickup.signature = 0;
121 	pickup.type = 0;
122 	histogram = 0;
123 	histindex = 0;
124 	histmax = 0;
125 	histogram_ary = NULL;
126 	sameline = 0;
127 
128 again:
129 	bzero(&mirror, sizeof(mirror));
130 	hammer_key_beg_init(&mirror.key_beg);
131 	hammer_key_end_init(&mirror.key_end);
132 
133 	fd = getpfs(&pfs, filesystem);
134 
135 	if (streaming >= 0)
136 		score_printf(LINE1, "Running");
137 
138 	if (streaming >= 0 && VerboseOpt && VerboseOpt < 2) {
139 		fprintf(stderr, "%cRunning  \b\b", (sameline ? '\r' : '\n'));
140 		fflush(stderr);
141 		sameline = 1;
142 	}
143 	sameline = 1;
144 	total_bytes = 0;
145 	gettimeofday(&bwtv, NULL);
146 	bwcount = 0;
147 
148 	/*
149 	 * Send initial header for the purpose of determining the
150 	 * shared-uuid.
151 	 */
152 	generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
153 	write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
154 		      &mrec_tmp, sizeof(mrec_tmp.pfs));
155 
156 	/*
157 	 * In 2-way mode the target will send us a PFS info packet
158 	 * first.  Use the target's current snapshot TID as our default
159 	 * begin TID.
160 	 */
161 	if (TwoWayPipeOpt) {
162 		mirror.tid_beg = 0;
163 		n = validate_mrec_header(fd, 0, 0, pfs.pfs_id, &pickup,
164 					 NULL, &mirror.tid_beg);
165 		if (n < 0) {	/* got TERM record */
166 			relpfs(fd, &pfs);
167 			free(buf);
168 			free(histogram_ary);
169 			return;
170 		}
171 		++mirror.tid_beg;
172 	} else if (streaming && histogram) {
173 		mirror.tid_beg = histogram_ary[histindex].tid + 1;
174 	} else {
175 		mirror.tid_beg = 0;
176 	}
177 
178 	/*
179 	 * Write out the PFS header, tid_beg will be updated if our PFS
180 	 * has a larger begin sync.  tid_end is set to the latest source
181 	 * TID whos flush cycle has completed.
182 	 */
183 	generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
184 	if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
185 		mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
186 	mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
187 	mirror.ubuf = buf;
188 	mirror.size = SERIALBUF_SIZE;
189 	mirror.pfs_id = pfs.pfs_id;
190 	mirror.shared_uuid = pfs.ondisk->shared_uuid;
191 
192 	/*
193 	 * XXX If the histogram is exhausted and the TID delta is large
194 	 *     the stream might have been offline for a while and is
195 	 *     now picking it up again.  Do another histogram.
196 	 */
197 #if 0
198 	if (streaming && histogram && histindex == histend) {
199 		if (mirror.tid_end - mirror.tid_beg > BULK_MINIMUM)
200 			histogram = 0;
201 	}
202 #endif
203 
204 	/*
205 	 * Initial bulk startup control, try to do some incremental
206 	 * mirroring in order to allow the stream to be killed and
207 	 * restarted without having to start over.
208 	 */
209 	if (histogram == 0 && BulkOpt == 0) {
210 		if (VerboseOpt && repeat == 0) {
211 			fprintf(stderr, "\n");
212 			sameline = 0;
213 		}
214 		histmax = generate_histogram(fd, filesystem,
215 					     &histogram_ary, &mirror,
216 					     &repeat);
217 		histindex = 0;
218 		histogram = 1;
219 
220 		/*
221 		 * Just stream the histogram, then stop
222 		 */
223 		if (streaming == 0)
224 			streaming = -1;
225 	}
226 
227 	if (streaming && histogram) {
228 		++histindex;
229 		mirror.tid_end = histogram_ary[histindex].tid;
230 		estbytes = histogram_ary[histindex-1].bytes;
231 		mrec_tmp.pfs.pfsd.sync_end_tid = mirror.tid_end;
232 	} else {
233 		estbytes = 0;
234 	}
235 
236 	write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
237 		      &mrec_tmp, sizeof(mrec_tmp.pfs));
238 
239 	/*
240 	 * A cycle file overrides the beginning TID only if we are
241 	 * not operating in two-way or histogram mode.
242 	 */
243 	if (TwoWayPipeOpt == 0 && histogram == 0)
244 		hammer_get_cycle(&mirror.key_beg, &mirror.tid_beg);
245 
246 	/*
247 	 * An additional argument overrides the beginning TID regardless
248 	 * of what mode we are in.  This is not recommending if operating
249 	 * in two-way mode.
250 	 */
251 	if (ac == 2)
252 		mirror.tid_beg = strtoull(av[1], NULL, 0);
253 
254 	if (streaming == 0 || VerboseOpt >= 2) {
255 		fprintf(stderr,
256 			"Mirror-read: Mirror %016jx to %016jx",
257 			(uintmax_t)mirror.tid_beg, (uintmax_t)mirror.tid_end);
258 		if (histogram)
259 			fprintf(stderr, " (bulk= %ju)", (uintmax_t)estbytes);
260 		fprintf(stderr, "\n");
261 		fflush(stderr);
262 	}
263 	if (mirror.key_beg.obj_id != (int64_t)HAMMER_MIN_OBJID) {
264 		fprintf(stderr, "Mirror-read: Resuming at object %016jx\n",
265 			(uintmax_t)mirror.key_beg.obj_id);
266 	}
267 
268 	/*
269 	 * Nothing to do if begin equals end.
270 	 */
271 	if (mirror.tid_beg >= mirror.tid_end) {
272 		if (streaming == 0 || VerboseOpt >= 2)
273 			fprintf(stderr, "Mirror-read: No work to do\n");
274 		sleep(DelayOpt);
275 		didwork = 0;
276 		histogram = 0;
277 		goto done;
278 	}
279 	didwork = 1;
280 
281 	/*
282 	 * Write out bulk records
283 	 */
284 	mirror.ubuf = buf;
285 	mirror.size = SERIALBUF_SIZE;
286 
287 	do {
288 		mirror.count = 0;
289 		mirror.pfs_id = pfs.pfs_id;
290 		mirror.shared_uuid = pfs.ondisk->shared_uuid;
291 		if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
292 			score_printf(LINE3, "Mirror-read %s failed: %s",
293 				     filesystem, strerror(errno));
294 			err(1, "Mirror-read %s failed", filesystem);
295 			/* not reached */
296 		}
297 		if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
298 			score_printf(LINE3, "Mirror-read %s fatal error %d",
299 				     filesystem, mirror.head.error);
300 			errx(1, "Mirror-read %s fatal error %d",
301 				filesystem, mirror.head.error);
302 			/* not reached */
303 		}
304 		if (mirror.count) {
305 			if (BandwidthOpt) {
306 				n = writebw(1, mirror.ubuf, mirror.count,
307 					    &bwcount, &bwtv);
308 			} else {
309 				n = write(1, mirror.ubuf, mirror.count);
310 			}
311 			if (n != mirror.count) {
312 				score_printf(LINE3,
313 					     "Mirror-read %s failed: "
314 					     "short write",
315 					     filesystem);
316 				errx(1, "Mirror-read %s failed: short write",
317 					filesystem);
318 				/* not reached */
319 			}
320 		}
321 		total_bytes += mirror.count;
322 		if (streaming && VerboseOpt) {
323 			fprintf(stderr,
324 				"\rscan obj=%016jx tids=%016jx:%016jx %11jd",
325 				(uintmax_t)mirror.key_cur.obj_id,
326 				(uintmax_t)mirror.tid_beg,
327 				(uintmax_t)mirror.tid_end,
328 				(intmax_t)total_bytes);
329 			fflush(stderr);
330 			sameline = 0;
331 		} else if (streaming) {
332 			score_printf(LINE2,
333 				"obj=%016jx tids=%016jx:%016jx %11jd",
334 				(uintmax_t)mirror.key_cur.obj_id,
335 				(uintmax_t)mirror.tid_beg,
336 				(uintmax_t)mirror.tid_end,
337 				(intmax_t)total_bytes);
338 		}
339 		mirror.key_beg = mirror.key_cur;
340 
341 		/*
342 		 * Deal with time limit option
343 		 */
344 		if (TimeoutOpt &&
345 		    (unsigned)(time(NULL) - base_t) > (unsigned)TimeoutOpt) {
346 			score_printf(LINE3,
347 				"Mirror-read %s interrupted by timer at"
348 				" %016jx",
349 				filesystem,
350 				(uintmax_t)mirror.key_cur.obj_id);
351 			fprintf(stderr,
352 				"Mirror-read %s interrupted by timer at"
353 				" %016jx\n",
354 				filesystem,
355 				(uintmax_t)mirror.key_cur.obj_id);
356 			interrupted = 1;
357 			break;
358 		}
359 	} while (mirror.count != 0);
360 
361 done:
362 	if (streaming && VerboseOpt && sameline == 0) {
363 		fprintf(stderr, "\n");
364 		fflush(stderr);
365 		sameline = 1;
366 	}
367 
368 	/*
369 	 * Write out the termination sync record - only if not interrupted
370 	 */
371 	if (interrupted == 0) {
372 		if (didwork) {
373 			write_mrecord(1, HAMMER_MREC_TYPE_SYNC,
374 				      &mrec_tmp, sizeof(mrec_tmp.sync));
375 		} else {
376 			write_mrecord(1, HAMMER_MREC_TYPE_IDLE,
377 				      &mrec_tmp, sizeof(mrec_tmp.sync));
378 		}
379 	}
380 
381 	/*
382 	 * If the -2 option was given (automatic when doing mirror-copy),
383 	 * a two-way pipe is assumed and we expect a response mrec from
384 	 * the target.
385 	 */
386 	if (TwoWayPipeOpt) {
387 		mrec = read_mrecord(0, &error, &pickup);
388 		if (mrec == NULL ||
389 		    mrec->head.type != HAMMER_MREC_TYPE_UPDATE ||
390 		    mrec->head.rec_size != sizeof(mrec->update)) {
391 			errx(1, "mirror_read: Did not get final "
392 				"acknowledgement packet from target");
393 			/* not reached */
394 		}
395 		if (interrupted) {
396 			if (CyclePath) {
397 				hammer_set_cycle(&mirror.key_cur,
398 						 mirror.tid_beg);
399 				fprintf(stderr, "Cyclefile %s updated for "
400 					"continuation\n", CyclePath);
401 			}
402 		} else {
403 			sync_tid = mrec->update.tid;
404 			if (CyclePath) {
405 				hammer_key_beg_init(&mirror.key_beg);
406 				hammer_set_cycle(&mirror.key_beg, sync_tid);
407 				fprintf(stderr,
408 					"Cyclefile %s updated to 0x%016jx\n",
409 					CyclePath, (uintmax_t)sync_tid);
410 			}
411 		}
412 		free(mrec);
413 	} else if (CyclePath) {
414 		/* NOTE! mirror.tid_beg cannot be updated */
415 		fprintf(stderr, "Warning: cycle file (-c option) cannot be "
416 				"fully updated unless you use mirror-copy\n");
417 		hammer_set_cycle(&mirror.key_beg, mirror.tid_beg);
418 	}
419 	if (streaming && interrupted == 0) {
420 		time_t t1 = time(NULL);
421 		time_t t2;
422 
423 		/*
424 		 * Try to break down large bulk transfers into smaller ones
425 		 * so it can sync the transaction id on the slave.  This
426 		 * way if we get interrupted a restart doesn't have to
427 		 * start from scratch.
428 		 */
429 		if (streaming && histogram) {
430 			if (histindex != histmax) {
431 				if (VerboseOpt && VerboseOpt < 2 &&
432 				    streaming >= 0) {
433 					fprintf(stderr, " (bulk incremental)");
434 				}
435 				relpfs(fd, &pfs);
436 				goto again;
437 			}
438 		}
439 
440 		if (VerboseOpt && streaming >= 0) {
441 			fprintf(stderr, " W");
442 			fflush(stderr);
443 		} else if (streaming >= 0) {
444 			score_printf(LINE1, "Waiting");
445 		}
446 		pfs.ondisk->sync_end_tid = mirror.tid_end;
447 		if (streaming < 0) {
448 			/*
449 			 * Fake streaming mode when using a histogram to
450 			 * break up a mirror-read, do not wait on source.
451 			 */
452 			streaming = 0;
453 		} else if (ioctl(fd, HAMMERIOC_WAI_PSEUDOFS, &pfs) < 0) {
454 			score_printf(LINE3,
455 				     "Mirror-read %s: cannot stream: %s\n",
456 				     filesystem, strerror(errno));
457 			fprintf(stderr,
458 				"Mirror-read %s: cannot stream: %s\n",
459 				filesystem, strerror(errno));
460 		} else {
461 			t2 = time(NULL) - t1;
462 			if (t2 >= 0 && t2 < DelayOpt) {
463 				if (VerboseOpt) {
464 					fprintf(stderr, "\bD");
465 					fflush(stderr);
466 				}
467 				sleep(DelayOpt - t2);
468 			}
469 			if (VerboseOpt) {
470 				fprintf(stderr, "\b ");
471 				fflush(stderr);
472 			}
473 			relpfs(fd, &pfs);
474 			goto again;
475 		}
476 	}
477 	write_mrecord(1, HAMMER_MREC_TYPE_TERM,
478 		      &mrec_tmp, sizeof(mrec_tmp.sync));
479 	relpfs(fd, &pfs);
480 	free(buf);
481 	free(histogram_ary);
482 	fprintf(stderr, "Mirror-read %s succeeded\n", filesystem);
483 }
484 
485 /*
486  * What we are trying to do here is figure out how much data is
487  * going to be sent for the TID range and to break the TID range
488  * down into reasonably-sized slices (from the point of view of
489  * data sent) so a lost connection can restart at a reasonable
490  * place and not all the way back at the beginning.
491  *
492  * An entry's TID serves as the end_tid for the prior entry
493  * So we have to offset the calculation by 1 so that TID falls into
494  * the previous entry when populating entries.
495  *
496  * Because the transaction id space is bursty we need a relatively
497  * large number of buckets (like a million) to do a reasonable job
498  * for things like an initial bulk mirrors on a very large filesystem.
499  */
500 #define HIST_COUNT	(1024 * 1024)
501 
502 static int
503 generate_histogram(int fd, const char *filesystem,
504 		   histogram_t *histogram_ary,
505 		   struct hammer_ioc_mirror_rw *mirror_base,
506 		   int *repeatp)
507 {
508 	struct hammer_ioc_mirror_rw mirror;
509 	union hammer_ioc_mrecord_any *mrec;
510 	hammer_tid_t tid_beg;
511 	hammer_tid_t tid_end;
512 	hammer_tid_t tid;
513 	hammer_tid_t tidx;
514 	uint64_t *tid_bytes;
515 	uint64_t total;
516 	uint64_t accum;
517 	int chunkno;
518 	int i;
519 	int res;
520 	int off;
521 	int len;
522 
523 	mirror = *mirror_base;
524 	tid_beg = mirror.tid_beg;
525 	tid_end = mirror.tid_end;
526 	mirror.head.flags |= HAMMER_IOC_MIRROR_NODATA;
527 
528 	if (*histogram_ary == NULL) {
529 		*histogram_ary = malloc(sizeof(struct histogram) *
530 					(HIST_COUNT + 2));
531 	}
532 	if (tid_beg >= tid_end)
533 		return(0);
534 
535 	/* needs 2 extra */
536 	tid_bytes = malloc(sizeof(*tid_bytes) * (HIST_COUNT + 2));
537 	bzero(tid_bytes, sizeof(*tid_bytes) * (HIST_COUNT + 2));
538 
539 	if (*repeatp == 0) {
540 		fprintf(stderr, "Prescan to break up bulk transfer");
541 		if (VerboseOpt > 1)
542 			fprintf(stderr, " (%juMB chunks)",
543 				(uintmax_t)(SplitupOpt / (1024 * 1024)));
544 		fprintf(stderr, "\n");
545 	}
546 
547 	/*
548 	 * Note: (tid_beg,tid_end), range is inclusive of both beg & end.
549 	 *
550 	 * Note: Estimates can be off when the mirror is way behind due
551 	 *	 to skips.
552 	 */
553 	total = 0;
554 	accum = 0;
555 	chunkno = 0;
556 	for (;;) {
557 		mirror.count = 0;
558 		if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
559 			err(1, "Mirror-read %s failed", filesystem);
560 			/* not reached */
561 		}
562 		if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
563 			errx(1, "Mirror-read %s fatal error %d",
564 				filesystem, mirror.head.error);
565 			/* not reached */
566 		}
567 		for (off = 0;
568 		     off < mirror.count;
569 		     off += HAMMER_HEAD_DOALIGN(mrec->head.rec_size)) {
570 			mrec = (void *)((char *)mirror.ubuf + off);
571 
572 			/*
573 			 * We only care about general RECs and PASS
574 			 * records.  We ignore SKIPs.
575 			 */
576 			switch (mrec->head.type & HAMMER_MRECF_TYPE_LOMASK) {
577 			case HAMMER_MREC_TYPE_REC:
578 			case HAMMER_MREC_TYPE_PASS:
579 				break;
580 			default:
581 				continue;
582 			}
583 
584 			/*
585 			 * Calculate for two indices, create_tid and
586 			 * delete_tid.  Record data only applies to
587 			 * the create_tid.
588 			 *
589 			 * When tid is exactly on the boundary it really
590 			 * belongs to the previous entry because scans
591 			 * are inclusive of the ending entry.
592 			 */
593 			tid = mrec->rec.leaf.base.delete_tid;
594 			if (tid && tid >= tid_beg && tid <= tid_end) {
595 				len = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
596 				if (mrec->head.type ==
597 				    HAMMER_MREC_TYPE_REC) {
598 					len -= HAMMER_HEAD_DOALIGN(
599 						    mrec->rec.leaf.data_len);
600 					assert(len > 0);
601 				}
602 				i = (tid - tid_beg) * HIST_COUNT /
603 				    (tid_end - tid_beg);
604 				tidx = tid_beg + i * (tid_end - tid_beg) /
605 						 HIST_COUNT;
606 				if (tid == tidx && i)
607 					--i;
608 				assert(i >= 0 && i < HIST_COUNT);
609 				tid_bytes[i] += len;
610 				total += len;
611 				accum += len;
612 			}
613 
614 			tid = mrec->rec.leaf.base.create_tid;
615 			if (tid && tid >= tid_beg && tid <= tid_end) {
616 				len = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
617 				if (mrec->head.type ==
618 				    HAMMER_MREC_TYPE_REC_NODATA) {
619 					len += HAMMER_HEAD_DOALIGN(
620 						    mrec->rec.leaf.data_len);
621 				}
622 				i = (tid - tid_beg) * HIST_COUNT /
623 				    (tid_end - tid_beg);
624 				tidx = tid_beg + i * (tid_end - tid_beg) /
625 						 HIST_COUNT;
626 				if (tid == tidx && i)
627 					--i;
628 				assert(i >= 0 && i < HIST_COUNT);
629 				tid_bytes[i] += len;
630 				total += len;
631 				accum += len;
632 			}
633 		}
634 		if (*repeatp == 0 && accum > SplitupOpt) {
635 			if (VerboseOpt > 1) {
636 				fprintf(stderr, ".");
637 				fflush(stderr);
638 			}
639 			++chunkno;
640 			score_printf(LINE2, "Prescan chunk %d", chunkno);
641 			accum = 0;
642 		}
643 		if (mirror.count == 0)
644 			break;
645 		mirror.key_beg = mirror.key_cur;
646 	}
647 
648 	/*
649 	 * Reduce to SplitupOpt (default 4GB) chunks.  This code may
650 	 * use up to two additional elements.  Do the array in-place.
651 	 *
652 	 * Inefficient degenerate cases can occur if we do not accumulate
653 	 * at least the requested split amount, so error on the side of
654 	 * going over a bit.
655 	 */
656 	res = 0;
657 	(*histogram_ary)[res].tid = tid_beg;
658 	(*histogram_ary)[res].bytes = tid_bytes[0];
659 	for (i = 1; i < HIST_COUNT; ++i) {
660 		if ((*histogram_ary)[res].bytes >= SplitupOpt) {
661 			++res;
662 			(*histogram_ary)[res].tid = tid_beg +
663 					i * (tid_end - tid_beg) /
664 					HIST_COUNT;
665 			(*histogram_ary)[res].bytes = 0;
666 
667 		}
668 		(*histogram_ary)[res].bytes += tid_bytes[i];
669 	}
670 	++res;
671 	(*histogram_ary)[res].tid = tid_end;
672 	(*histogram_ary)[res].bytes = -1;
673 
674 	if (*repeatp == 0) {
675 		if (VerboseOpt > 1)
676 			fprintf(stderr, "\n");	/* newline after ... */
677 		score_printf(LINE3, "Prescan %d chunks, total %ju MBytes",
678 			res, (uintmax_t)total / (1024 * 1024));
679 		fprintf(stderr, "Prescan %d chunks, total %ju MBytes (",
680 			res, (uintmax_t)total / (1024 * 1024));
681 		for (i = 0; i < res && i < 3; ++i) {
682 			if (i)
683 				fprintf(stderr, ", ");
684 			fprintf(stderr, "%ju",
685 				(uintmax_t)(*histogram_ary)[i].bytes);
686 		}
687 		if (i < res)
688 			fprintf(stderr, ", ...");
689 		fprintf(stderr, ")\n");
690 	}
691 	assert(res <= HIST_COUNT);
692 	*repeatp = 1;
693 
694 	free(tid_bytes);
695 	return(res);
696 }
697 
698 static void
699 create_pfs(const char *filesystem, uuid_t *s_uuid)
700 {
701 	if (ForceYesOpt == 1) {
702 		fprintf(stderr, "PFS slave %s does not exist. "
703 			"Auto create new slave PFS!\n", filesystem);
704 
705 	} else {
706 		fprintf(stderr, "PFS slave %s does not exist.\n"
707 			"Do you want to create a new slave PFS? [y/n] ",
708 			filesystem);
709 		fflush(stderr);
710 		if (getyntty() != 1) {
711 			errx(1, "Aborting operation");
712 			/* not reached */
713 		}
714 	}
715 
716 	uint32_t status;
717 	char *shared_uuid = NULL;
718 	uuid_to_string(s_uuid, &shared_uuid, &status);
719 
720 	char *cmd = NULL;
721 	asprintf(&cmd, "/sbin/hammer pfs-slave '%s' shared-uuid=%s 1>&2",
722 		 filesystem, shared_uuid);
723 	free(shared_uuid);
724 
725 	if (cmd == NULL) {
726 		errx(1, "Failed to alloc memory");
727 		/* not reached */
728 	}
729 	if (system(cmd) != 0)
730 		fprintf(stderr, "Failed to create PFS\n");
731 	free(cmd);
732 }
733 
734 /*
735  * Pipe the mirroring data stream on stdin to the HAMMER VFS, adding
736  * some additional packet types to negotiate TID ranges and to verify
737  * completion.  The HAMMER VFS does most of the work.
738  *
739  * It is important to note that the mirror.key_{beg,end} range must
740  * match the ranged used by the original.  For now both sides use
741  * range the entire key space.
742  *
743  * It is even more important that the records in the stream conform
744  * to the TID range also supplied in the stream.  The HAMMER VFS will
745  * use the REC, PASS, and SKIP record types to track the portions of
746  * the B-Tree being scanned in order to be able to proactively delete
747  * records on the target within those active areas that are not mentioned
748  * by the source.
749  *
750  * The mirror.key_cur field is used by the VFS to do this tracking.  It
751  * must be initialized to key_beg but then is persistently updated by
752  * the HAMMER VFS on each successive ioctl() call.  If you blow up this
753  * field you will blow up the mirror target, possibly to the point of
754  * deleting everything.  As a safety measure the HAMMER VFS simply marks
755  * the records that the source has destroyed as deleted on the target,
756  * and normal pruning operations will deal with their final disposition
757  * at some later time.
758  */
759 void
760 hammer_cmd_mirror_write(char **av, int ac)
761 {
762 	struct hammer_ioc_mirror_rw mirror;
763 	const char *filesystem;
764 	char *buf = malloc(SERIALBUF_SIZE);
765 	struct hammer_ioc_pseudofs_rw pfs;
766 	struct hammer_ioc_mrecord_head pickup;
767 	struct hammer_ioc_synctid synctid;
768 	union hammer_ioc_mrecord_any mrec_tmp;
769 	hammer_ioc_mrecord_any_t mrec;
770 	struct stat st;
771 	int error;
772 	int fd;
773 	int n;
774 
775 	if (ac != 1) {
776 		mirror_usage(1);
777 		/* not reached */
778 	}
779 	filesystem = av[0];
780 	hammer_check_restrict(filesystem);
781 
782 	pickup.signature = 0;
783 	pickup.type = 0;
784 
785 again:
786 	bzero(&mirror, sizeof(mirror));
787 	hammer_key_beg_init(&mirror.key_beg);
788 	hammer_key_end_init(&mirror.key_end);
789 	mirror.key_end = mirror.key_beg;
790 
791 	/*
792 	 * Read initial packet
793 	 */
794 	mrec = read_mrecord(0, &error, &pickup);
795 	if (mrec == NULL) {
796 		if (error == 0) {
797 			errx(1, "validate_mrec_header: short read");
798 			/* not reached */
799 		}
800 		exit(1);
801 	}
802 	/*
803 	 * Validate packet
804 	 */
805 	if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
806 		free(buf);
807 		return;
808 	}
809 	if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
810 		errx(1, "validate_mrec_header: did not get expected "
811 			"PFSD record type");
812 		/* not reached */
813 	}
814 	if (mrec->head.rec_size != sizeof(mrec->pfs)) {
815 		errx(1, "validate_mrec_header: unexpected payload size");
816 		/* not reached */
817 	}
818 
819 	/*
820 	 * Create slave PFS if it doesn't yet exist
821 	 */
822 	if (lstat(filesystem, &st) != 0)
823 		create_pfs(filesystem, &mrec->pfs.pfsd.shared_uuid);
824 	free(mrec);
825 	mrec = NULL;
826 
827 	fd = getpfs(&pfs, filesystem);
828 
829 	/*
830 	 * In two-way mode the target writes out a PFS packet first.
831 	 * The source uses our tid_end as its tid_beg by default,
832 	 * picking up where it left off.
833 	 */
834 	mirror.tid_beg = 0;
835 	if (TwoWayPipeOpt) {
836 		generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
837 		if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
838 			mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
839 		mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
840 		write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
841 			      &mrec_tmp, sizeof(mrec_tmp.pfs));
842 	}
843 
844 	/*
845 	 * Read and process the PFS header.  The source informs us of
846 	 * the TID range the stream represents.
847 	 */
848 	n = validate_mrec_header(fd, 0, 1, pfs.pfs_id, &pickup,
849 				 &mirror.tid_beg, &mirror.tid_end);
850 	if (n < 0) {	/* got TERM record */
851 		relpfs(fd, &pfs);
852 		free(buf);
853 		return;
854 	}
855 
856 	mirror.ubuf = buf;
857 	mirror.size = SERIALBUF_SIZE;
858 
859 	/*
860 	 * Read and process bulk records (REC, PASS, and SKIP types).
861 	 *
862 	 * On your life, do NOT mess with mirror.key_cur or your mirror
863 	 * target may become history.
864 	 */
865 	for (;;) {
866 		mirror.count = 0;
867 		mirror.pfs_id = pfs.pfs_id;
868 		mirror.shared_uuid = pfs.ondisk->shared_uuid;
869 		mirror.size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
870 		if (mirror.size <= 0)
871 			break;
872 		if (ioctl(fd, HAMMERIOC_MIRROR_WRITE, &mirror) < 0) {
873 			err(1, "Mirror-write %s failed", filesystem);
874 			/* not reached */
875 		}
876 		if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
877 			errx(1, "Mirror-write %s fatal error %d",
878 				filesystem, mirror.head.error);
879 			/* not reached */
880 		}
881 #if 0
882 		if (mirror.head.flags & HAMMER_IOC_HEAD_INTR) {
883 			errx(1, "Mirror-write %s interrupted by timer at"
884 				" %016llx",
885 				filesystem,
886 				mirror.key_cur.obj_id);
887 			/* not reached */
888 		}
889 #endif
890 	}
891 
892 	/*
893 	 * Read and process the termination sync record.
894 	 */
895 	mrec = read_mrecord(0, &error, &pickup);
896 
897 	if (mrec && mrec->head.type == HAMMER_MREC_TYPE_TERM) {
898 		fprintf(stderr, "Mirror-write: received termination request\n");
899 		relpfs(fd, &pfs);
900 		free(mrec);
901 		free(buf);
902 		return;
903 	}
904 
905 	if (mrec == NULL ||
906 	    (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
907 	     mrec->head.type != HAMMER_MREC_TYPE_IDLE) ||
908 	    mrec->head.rec_size != sizeof(mrec->sync)) {
909 		errx(1, "Mirror-write %s: Did not get termination "
910 			"sync record, or rec_size is wrong rt=%d",
911 			filesystem, (mrec ? (int)mrec->head.type : -1));
912 		/* not reached */
913 	}
914 
915 	/*
916 	 * Update the PFS info on the target so the user has visibility
917 	 * into the new snapshot, and sync the target filesystem.
918 	 */
919 	if (mrec->head.type == HAMMER_MREC_TYPE_SYNC) {
920 		update_pfs_snapshot(fd, mirror.tid_end, pfs.pfs_id);
921 
922 		bzero(&synctid, sizeof(synctid));
923 		synctid.op = HAMMER_SYNCTID_SYNC2;
924 		ioctl(fd, HAMMERIOC_SYNCTID, &synctid);
925 
926 		if (VerboseOpt >= 2) {
927 			fprintf(stderr, "Mirror-write %s: succeeded\n",
928 				filesystem);
929 		}
930 	}
931 
932 	free(mrec);
933 	mrec = NULL;
934 
935 	/*
936 	 * Report back to the originator.
937 	 */
938 	if (TwoWayPipeOpt) {
939 		mrec_tmp.update.tid = mirror.tid_end;
940 		write_mrecord(1, HAMMER_MREC_TYPE_UPDATE,
941 			      &mrec_tmp, sizeof(mrec_tmp.update));
942 	} else {
943 		printf("Source can update synctid to 0x%016jx\n",
944 		       (uintmax_t)mirror.tid_end);
945 	}
946 	relpfs(fd, &pfs);
947 	goto again;
948 }
949 
950 void
951 hammer_cmd_mirror_dump(char **av, int ac)
952 {
953 	char *buf = malloc(SERIALBUF_SIZE);
954 	struct hammer_ioc_mrecord_head pickup;
955 	hammer_ioc_mrecord_any_t mrec;
956 	int error;
957 	int size;
958 	int offset;
959 	int bytes;
960 	int header_only = 0;
961 
962 	if (ac == 1 && strcmp(*av, "header") == 0) {
963 		header_only = 1;
964 	} else if (ac != 0) {
965 		mirror_usage(1);
966 		/* not reached */
967 	}
968 
969 	/*
970 	 * Read and process the PFS header
971 	 */
972 	pickup.signature = 0;
973 	pickup.type = 0;
974 
975 	mrec = read_mrecord(0, &error, &pickup);
976 
977 	/*
978 	 * Dump the PFS header. mirror-dump takes its input from the output
979 	 * of a mirror-read so getpfs() can't be used to get a fd to be passed
980 	 * to dump_pfsd().
981 	 */
982 	if (header_only && mrec != NULL) {
983 		dump_pfsd(&mrec->pfs.pfsd, -1);
984 		free(mrec);
985 		free(buf);
986 		return;
987 	}
988 	free(mrec);
989 
990 again:
991 	/*
992 	 * Read and process bulk records
993 	 */
994 	for (;;) {
995 		size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
996 		if (size <= 0)
997 			break;
998 		offset = 0;
999 		while (offset < size) {
1000 			mrec = (void *)((char *)buf + offset);
1001 			bytes = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
1002 			if (offset + bytes > size) {
1003 				errx(1, "Misaligned record");
1004 				/* not reached */
1005 			}
1006 
1007 			switch(mrec->head.type & HAMMER_MRECF_TYPE_MASK) {
1008 			case HAMMER_MREC_TYPE_REC_BADCRC:
1009 			case HAMMER_MREC_TYPE_REC:
1010 				printf("Record lo=%08x obj=%016jx key=%016jx "
1011 				       "rt=%02x ot=%02x",
1012 				        mrec->rec.leaf.base.localization,
1013 					(uintmax_t)mrec->rec.leaf.base.obj_id,
1014 					(uintmax_t)mrec->rec.leaf.base.key,
1015 					mrec->rec.leaf.base.rec_type,
1016 					mrec->rec.leaf.base.obj_type);
1017 				if (mrec->head.type ==
1018 				    HAMMER_MREC_TYPE_REC_BADCRC) {
1019 					printf(" (BAD CRC)");
1020 				}
1021 				printf("\n");
1022 				printf("       tids %016jx:%016jx data=%d\n",
1023 				    (uintmax_t)mrec->rec.leaf.base.create_tid,
1024 				    (uintmax_t)mrec->rec.leaf.base.delete_tid,
1025 				    mrec->rec.leaf.data_len);
1026 				break;
1027 			case HAMMER_MREC_TYPE_PASS:
1028 				printf("Pass   lo=%08x obj=%016jx key=%016jx "
1029 				       "rt=%02x ot=%02x\n",
1030 				        mrec->rec.leaf.base.localization,
1031 					(uintmax_t)mrec->rec.leaf.base.obj_id,
1032 					(uintmax_t)mrec->rec.leaf.base.key,
1033 					mrec->rec.leaf.base.rec_type,
1034 					mrec->rec.leaf.base.obj_type);
1035 				printf("       tids %016jx:%016jx data=%d\n",
1036 				    (uintmax_t)mrec->rec.leaf.base.create_tid,
1037 				    (uintmax_t)mrec->rec.leaf.base.delete_tid,
1038 					mrec->rec.leaf.data_len);
1039 				break;
1040 			case HAMMER_MREC_TYPE_SKIP:
1041 				printf("Skip   lo=%08x obj=%016jx key=%016jx rt=%02x to\n"
1042 				       "       lo=%08x obj=%016jx key=%016jx rt=%02x\n",
1043 				       mrec->skip.skip_beg.localization,
1044 				       (uintmax_t)mrec->skip.skip_beg.obj_id,
1045 				       (uintmax_t)mrec->skip.skip_beg.key,
1046 				       mrec->skip.skip_beg.rec_type,
1047 				       mrec->skip.skip_end.localization,
1048 				       (uintmax_t)mrec->skip.skip_end.obj_id,
1049 				       (uintmax_t)mrec->skip.skip_end.key,
1050 				       mrec->skip.skip_end.rec_type);
1051 			default:
1052 				break;
1053 			}
1054 			offset += bytes;
1055 		}
1056 	}
1057 
1058 	/*
1059 	 * Read and process the termination sync record.
1060 	 */
1061 	mrec = read_mrecord(0, &error, &pickup);
1062 	if (mrec == NULL ||
1063 	    (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
1064 	     mrec->head.type != HAMMER_MREC_TYPE_IDLE)) {
1065 		fprintf(stderr, "Mirror-dump: Did not get termination "
1066 				"sync record\n");
1067 	}
1068 	free(mrec);
1069 
1070 	/*
1071 	 * Continue with more batches until EOF.
1072 	 */
1073 	mrec = read_mrecord(0, &error, &pickup);
1074 	if (mrec) {
1075 		free(mrec);
1076 		goto again;
1077 	}
1078 	free(buf);
1079 }
1080 
1081 void
1082 hammer_cmd_mirror_copy(char **av, int ac, int streaming)
1083 {
1084 	pid_t pid1;
1085 	pid_t pid2;
1086 	int fds[2];
1087 	const char *xav[32];
1088 	char tbuf[16];
1089 	char *sh, *user, *host, *rfs;
1090 	int xac;
1091 
1092 	if (ac != 2) {
1093 		mirror_usage(1);
1094 		/* not reached */
1095 	}
1096 
1097 	TwoWayPipeOpt = 1;
1098 	signal(SIGPIPE, SIG_IGN);
1099 
1100 again:
1101 	if (pipe(fds) < 0) {
1102 		err(1, "pipe");
1103 		/* not reached */
1104 	}
1105 
1106 	/*
1107 	 * Source
1108 	 */
1109 	if ((pid1 = fork()) == 0) {
1110 		signal(SIGPIPE, SIG_DFL);
1111 		dup2(fds[0], 0);
1112 		dup2(fds[0], 1);
1113 		close(fds[0]);
1114 		close(fds[1]);
1115 		if ((rfs = strchr(av[0], ':')) != NULL) {
1116 			xac = 0;
1117 
1118 			if((sh = getenv("HAMMER_RSH")) == NULL)
1119 				xav[xac++] = "ssh";
1120 			else
1121 				xav[xac++] = sh;
1122 
1123 			if (CompressOpt)
1124 				xav[xac++] = "-C";
1125 
1126 			if ((host = strchr(av[0], '@')) != NULL) {
1127 				user = strndup( av[0], (host++ - av[0]));
1128 				host = strndup( host, (rfs++ - host));
1129 				xav[xac++] = "-l";
1130 				xav[xac++] = user;
1131 				xav[xac++] = host;
1132 			} else {
1133 				host = strndup( av[0], (rfs++ - av[0]));
1134 				user = NULL;
1135 				xav[xac++] = host;
1136 			}
1137 
1138 
1139 			if (SshPort) {
1140 				xav[xac++] = "-p";
1141 				xav[xac++] = SshPort;
1142 			}
1143 
1144 			xav[xac++] = "hammer";
1145 
1146 			switch(VerboseOpt) {
1147 			case 0:
1148 				break;
1149 			case 1:
1150 				xav[xac++] = "-v";
1151 				break;
1152 			case 2:
1153 				xav[xac++] = "-vv";
1154 				break;
1155 			default:
1156 				xav[xac++] = "-vvv";
1157 				break;
1158 			}
1159 			if (ForceYesOpt)
1160 				xav[xac++] = "-y";
1161 			xav[xac++] = "-2";
1162 			if (TimeoutOpt) {
1163 				snprintf(tbuf, sizeof(tbuf), "%d", TimeoutOpt);
1164 				xav[xac++] = "-t";
1165 				xav[xac++] = tbuf;
1166 			}
1167 			if (SplitupOptStr) {
1168 				xav[xac++] = "-S";
1169 				xav[xac++] = SplitupOptStr;
1170 			}
1171 			if (streaming)
1172 				xav[xac++] = "mirror-read-stream";
1173 			else
1174 				xav[xac++] = "mirror-read";
1175 			xav[xac++] = rfs;
1176 			xav[xac++] = NULL;
1177 			execvp(*xav, (void *)xav);
1178 		} else {
1179 			hammer_cmd_mirror_read(av, 1, streaming);
1180 			fflush(stdout);
1181 			fflush(stderr);
1182 		}
1183 		_exit(1);
1184 	}
1185 
1186 	/*
1187 	 * Target
1188 	 */
1189 	if ((pid2 = fork()) == 0) {
1190 		signal(SIGPIPE, SIG_DFL);
1191 		dup2(fds[1], 0);
1192 		dup2(fds[1], 1);
1193 		close(fds[0]);
1194 		close(fds[1]);
1195 		if ((rfs = strchr(av[1], ':')) != NULL) {
1196 			xac = 0;
1197 
1198 			if((sh = getenv("HAMMER_RSH")) == NULL)
1199 				xav[xac++] = "ssh";
1200 			else
1201 				xav[xac++] = sh;
1202 
1203 			if (CompressOpt)
1204 				xav[xac++] = "-C";
1205 
1206 			if ((host = strchr(av[1], '@')) != NULL) {
1207 				user = strndup( av[1], (host++ - av[1]));
1208 				host = strndup( host, (rfs++ - host));
1209 				xav[xac++] = "-l";
1210 				xav[xac++] = user;
1211 				xav[xac++] = host;
1212 			} else {
1213 				host = strndup( av[1], (rfs++ - av[1]));
1214 				user = NULL;
1215 				xav[xac++] = host;
1216 			}
1217 
1218 			if (SshPort) {
1219 				xav[xac++] = "-p";
1220 				xav[xac++] = SshPort;
1221 			}
1222 
1223 			xav[xac++] = "hammer";
1224 
1225 			switch(VerboseOpt) {
1226 			case 0:
1227 				break;
1228 			case 1:
1229 				xav[xac++] = "-v";
1230 				break;
1231 			case 2:
1232 				xav[xac++] = "-vv";
1233 				break;
1234 			default:
1235 				xav[xac++] = "-vvv";
1236 				break;
1237 			}
1238 			if (ForceYesOpt)
1239 				xav[xac++] = "-y";
1240 			xav[xac++] = "-2";
1241 			xav[xac++] = "mirror-write";
1242 			xav[xac++] = rfs;
1243 			xav[xac++] = NULL;
1244 			execvp(*xav, (void *)xav);
1245 		} else {
1246 			hammer_cmd_mirror_write(av + 1, 1);
1247 			fflush(stdout);
1248 			fflush(stderr);
1249 		}
1250 		_exit(1);
1251 	}
1252 	close(fds[0]);
1253 	close(fds[1]);
1254 
1255 	while (waitpid(pid1, NULL, 0) <= 0)
1256 		;
1257 	while (waitpid(pid2, NULL, 0) <= 0)
1258 		;
1259 
1260 	/*
1261 	 * If the link is lost restart
1262 	 */
1263 	if (streaming) {
1264 		if (VerboseOpt) {
1265 			fprintf(stderr, "\nLost Link\n");
1266 			fflush(stderr);
1267 		}
1268 		sleep(15 + DelayOpt);
1269 		goto again;
1270 	}
1271 
1272 }
1273 
1274 /*
1275  * Read and return multiple mrecords
1276  */
1277 static int
1278 read_mrecords(int fd, char *buf, u_int size, hammer_ioc_mrecord_head_t pickup)
1279 {
1280 	hammer_ioc_mrecord_any_t mrec;
1281 	u_int count;
1282 	size_t n;
1283 	size_t i;
1284 	size_t bytes;
1285 	int type;
1286 
1287 	count = 0;
1288 	while (size - count >= HAMMER_MREC_HEADSIZE) {
1289 		/*
1290 		 * Cached the record header in case we run out of buffer
1291 		 * space.
1292 		 */
1293 		fflush(stdout);
1294 		if (pickup->signature == 0) {
1295 			for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1296 				i = read(fd, (char *)pickup + n,
1297 					 HAMMER_MREC_HEADSIZE - n);
1298 				if (i <= 0)
1299 					break;
1300 			}
1301 			if (n == 0)
1302 				break;
1303 			if (n != HAMMER_MREC_HEADSIZE) {
1304 				errx(1, "read_mrecords: short read on pipe");
1305 				/* not reached */
1306 			}
1307 			if (pickup->signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1308 				errx(1, "read_mrecords: malformed record on pipe, "
1309 					"bad signature");
1310 				/* not reached */
1311 			}
1312 		}
1313 		if (pickup->rec_size < HAMMER_MREC_HEADSIZE ||
1314 		    pickup->rec_size > sizeof(*mrec) + HAMMER_XBUFSIZE) {
1315 			errx(1, "read_mrecords: malformed record on pipe, "
1316 				"illegal rec_size");
1317 			/* not reached */
1318 		}
1319 
1320 		/*
1321 		 * Stop if we have insufficient space for the record and data.
1322 		 */
1323 		bytes = HAMMER_HEAD_DOALIGN(pickup->rec_size);
1324 		if (size - count < bytes)
1325 			break;
1326 
1327 		/*
1328 		 * Stop if the record type is not a REC, SKIP, or PASS,
1329 		 * which are the only types the ioctl supports.  Other types
1330 		 * are used only by the userland protocol.
1331 		 *
1332 		 * Ignore all flags.
1333 		 */
1334 		type = pickup->type & HAMMER_MRECF_TYPE_LOMASK;
1335 		if (type != HAMMER_MREC_TYPE_PFSD &&
1336 		    type != HAMMER_MREC_TYPE_REC &&
1337 		    type != HAMMER_MREC_TYPE_SKIP &&
1338 		    type != HAMMER_MREC_TYPE_PASS) {
1339 			break;
1340 		}
1341 
1342 		/*
1343 		 * Read the remainder and clear the pickup signature.
1344 		 */
1345 		for (n = HAMMER_MREC_HEADSIZE; n < bytes; n += i) {
1346 			i = read(fd, buf + count + n, bytes - n);
1347 			if (i <= 0)
1348 				break;
1349 		}
1350 		if (n != bytes) {
1351 			errx(1, "read_mrecords: short read on pipe");
1352 			/* not reached */
1353 		}
1354 
1355 		bcopy(pickup, buf + count, HAMMER_MREC_HEADSIZE);
1356 		pickup->signature = 0;
1357 		pickup->type = 0;
1358 		mrec = (void *)(buf + count);
1359 
1360 		/*
1361 		 * Validate the completed record
1362 		 */
1363 		if (!hammer_crc_test_mrec_head(&mrec->head, mrec->head.rec_size)) {
1364 			errx(1, "read_mrecords: malformed record on pipe, bad crc");
1365 			/* not reached */
1366 		}
1367 
1368 		/*
1369 		 * If its a B-Tree record validate the data crc.
1370 		 *
1371 		 * NOTE: If the VFS passes us an explicitly errorde mrec
1372 		 *	 we just pass it through.
1373 		 */
1374 		type = mrec->head.type & HAMMER_MRECF_TYPE_MASK;
1375 
1376 		if (type == HAMMER_MREC_TYPE_REC) {
1377 			if (mrec->head.rec_size <
1378 			    sizeof(mrec->rec) + mrec->rec.leaf.data_len) {
1379 				errx(1, "read_mrecords: malformed record on "
1380 					"pipe, illegal element data_len");
1381 				/* not reached */
1382 			}
1383 			if (mrec->rec.leaf.data_len &&
1384 			    mrec->rec.leaf.data_offset &&
1385 			    hammer_crc_test_leaf(HammerVersion, &mrec->rec + 1, &mrec->rec.leaf) == 0) {
1386 				fprintf(stderr,
1387 					"read_mrecords: data_crc did not "
1388 					"match data! obj=%016jx key=%016jx\n",
1389 					(uintmax_t)mrec->rec.leaf.base.obj_id,
1390 					(uintmax_t)mrec->rec.leaf.base.key);
1391 				fprintf(stderr,
1392 					"continuing, but there are problems\n");
1393 			}
1394 		}
1395 		count += bytes;
1396 	}
1397 	return(count);
1398 }
1399 
1400 /*
1401  * Read and return a single mrecord.
1402  */
1403 static
1404 hammer_ioc_mrecord_any_t
1405 read_mrecord(int fdin, int *errorp, hammer_ioc_mrecord_head_t pickup)
1406 {
1407 	hammer_ioc_mrecord_any_t mrec;
1408 	struct hammer_ioc_mrecord_head mrechd;
1409 	size_t bytes;
1410 	size_t n;
1411 	size_t i;
1412 
1413 	if (pickup && pickup->type != 0) {
1414 		mrechd = *pickup;
1415 		pickup->signature = 0;
1416 		pickup->type = 0;
1417 		n = HAMMER_MREC_HEADSIZE;
1418 	} else {
1419 		/*
1420 		 * Read in the PFSD header from the sender.
1421 		 */
1422 		for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1423 			i = read(fdin, (char *)&mrechd + n, HAMMER_MREC_HEADSIZE - n);
1424 			if (i <= 0)
1425 				break;
1426 		}
1427 		if (n == 0) {
1428 			*errorp = 0;	/* EOF */
1429 			return(NULL);
1430 		}
1431 		if (n != HAMMER_MREC_HEADSIZE) {
1432 			fprintf(stderr, "short read of mrecord header\n");
1433 			*errorp = EPIPE;
1434 			return(NULL);
1435 		}
1436 	}
1437 	if (mrechd.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1438 		fprintf(stderr, "read_mrecord: bad signature\n");
1439 		*errorp = EINVAL;
1440 		return(NULL);
1441 	}
1442 	bytes = HAMMER_HEAD_DOALIGN(mrechd.rec_size);
1443 	assert(bytes >= sizeof(mrechd));
1444 	mrec = malloc(bytes);
1445 	mrec->head = mrechd;
1446 
1447 	while (n < bytes) {
1448 		i = read(fdin, (char *)mrec + n, bytes - n);
1449 		if (i <= 0)
1450 			break;
1451 		n += i;
1452 	}
1453 	if (n != bytes) {
1454 		fprintf(stderr, "read_mrecord: short read on payload\n");
1455 		*errorp = EPIPE;
1456 		return(NULL);
1457 	}
1458 	if (!hammer_crc_test_mrec_head(&mrec->head, mrec->head.rec_size)) {
1459 		fprintf(stderr, "read_mrecord: bad CRC\n");
1460 		*errorp = EINVAL;
1461 		return(NULL);
1462 	}
1463 	*errorp = 0;
1464 	return(mrec);
1465 }
1466 
1467 static
1468 void
1469 write_mrecord(int fdout, uint32_t type, hammer_ioc_mrecord_any_t mrec,
1470 	      int bytes)
1471 {
1472 	char zbuf[HAMMER_HEAD_ALIGN];
1473 	int pad;
1474 
1475 	pad = HAMMER_HEAD_DOALIGN(bytes) - bytes;
1476 
1477 	assert(bytes >= (int)sizeof(mrec->head));
1478 	bzero(&mrec->head, sizeof(mrec->head));
1479 	mrec->head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
1480 	mrec->head.type = type;
1481 	mrec->head.rec_size = bytes;
1482 	hammer_crc_set_mrec_head(&mrec->head, bytes);
1483 	if (write(fdout, mrec, bytes) != bytes) {
1484 		err(1, "write_mrecord");
1485 		/* not reached */
1486 	}
1487 	if (pad) {
1488 		bzero(zbuf, pad);
1489 		if (write(fdout, zbuf, pad) != pad) {
1490 			err(1, "write_mrecord");
1491 			/* not reached */
1492 		}
1493 	}
1494 }
1495 
1496 /*
1497  * Generate a mirroring header with the pfs information of the
1498  * originating filesytem.
1499  */
1500 static void
1501 generate_mrec_header(int fd, int pfs_id,
1502 		     union hammer_ioc_mrecord_any *mrec_tmp)
1503 {
1504 	struct hammer_ioc_pseudofs_rw pfs;
1505 
1506 	bzero(mrec_tmp, sizeof(*mrec_tmp));
1507 	clrpfs(&pfs, &mrec_tmp->pfs.pfsd, pfs_id);
1508 
1509 	if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1510 		err(1, "Mirror-read: not a HAMMER fs/pseudofs!");
1511 		/* not reached */
1512 	}
1513 	if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1514 		errx(1, "Mirror-read: HAMMER PFS version mismatch!");
1515 		/* not reached */
1516 	}
1517 	mrec_tmp->pfs.version = pfs.version;
1518 }
1519 
1520 /*
1521  * Validate the pfs information from the originating filesystem
1522  * against the target filesystem.  shared_uuid must match.
1523  *
1524  * return -1 if we got a TERM record
1525  */
1526 static int
1527 validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
1528 		     struct hammer_ioc_mrecord_head *pickup,
1529 		     hammer_tid_t *tid_begp, hammer_tid_t *tid_endp)
1530 {
1531 	struct hammer_ioc_pseudofs_rw pfs;
1532 	struct hammer_pseudofs_data pfsd;
1533 	hammer_ioc_mrecord_any_t mrec;
1534 	int error;
1535 
1536 	/*
1537 	 * Get the PFSD info from the target filesystem.
1538 	 */
1539 	clrpfs(&pfs, &pfsd, pfs_id);
1540 	if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1541 		err(1, "mirror-write: not a HAMMER fs/pseudofs!");
1542 		/* not reached */
1543 	}
1544 	if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1545 		errx(1, "mirror-write: HAMMER PFS version mismatch!");
1546 		/* not reached */
1547 	}
1548 
1549 	mrec = read_mrecord(fdin, &error, pickup);
1550 	if (mrec == NULL) {
1551 		if (error == 0) {
1552 			errx(1, "validate_mrec_header: short read");
1553 			/* not reached */
1554 		}
1555 		exit(1);
1556 	}
1557 	if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
1558 		free(mrec);
1559 		return(-1);
1560 	}
1561 
1562 	if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
1563 		errx(1, "validate_mrec_header: did not get expected "
1564 			"PFSD record type");
1565 		/* not reached */
1566 	}
1567 	if (mrec->head.rec_size != sizeof(mrec->pfs)) {
1568 		errx(1, "validate_mrec_header: unexpected payload size");
1569 		/* not reached */
1570 	}
1571 	if (mrec->pfs.version != pfs.version) {
1572 		errx(1, "validate_mrec_header: Version mismatch");
1573 		/* not reached */
1574 	}
1575 
1576 	/*
1577 	 * Whew.  Ok, is the read PFS info compatible with the target?
1578 	 */
1579 	if (bcmp(&mrec->pfs.pfsd.shared_uuid, &pfsd.shared_uuid,
1580 		sizeof(pfsd.shared_uuid)) != 0) {
1581 		errx(1, "mirror-write: source and target have "
1582 			"different shared-uuid's!");
1583 		/* not reached */
1584 	}
1585 	if (is_target && hammer_is_pfs_master(&pfsd)) {
1586 		errx(1, "mirror-write: target must be in slave mode");
1587 		/* not reached */
1588 	}
1589 	if (tid_begp)
1590 		*tid_begp = mrec->pfs.pfsd.sync_beg_tid;
1591 	if (tid_endp)
1592 		*tid_endp = mrec->pfs.pfsd.sync_end_tid;
1593 	free(mrec);
1594 	return(0);
1595 }
1596 
1597 static void
1598 update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id)
1599 {
1600 	struct hammer_ioc_pseudofs_rw pfs;
1601 	struct hammer_pseudofs_data pfsd;
1602 
1603 	clrpfs(&pfs, &pfsd, pfs_id);
1604 	if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1605 		err(1, "update_pfs_snapshot (read)");
1606 		/* not reached */
1607 	}
1608 
1609 	if (pfsd.sync_end_tid != snapshot_tid) {
1610 		pfsd.sync_end_tid = snapshot_tid;
1611 		if (ioctl(fd, HAMMERIOC_SET_PSEUDOFS, &pfs) != 0) {
1612 			err(1, "update_pfs_snapshot (rewrite)");
1613 			/* not reached */
1614 		}
1615 		if (VerboseOpt >= 2) {
1616 			fprintf(stderr,
1617 				"Mirror-write: Completed, updated snapshot "
1618 				"to %016jx\n",
1619 				(uintmax_t)snapshot_tid);
1620 			fflush(stderr);
1621 		}
1622 	}
1623 }
1624 
1625 /*
1626  * Bandwidth-limited write in chunks
1627  */
1628 static
1629 ssize_t
1630 writebw(int fd, const void *buf, size_t nbytes,
1631 	uint64_t *bwcount, struct timeval *tv1)
1632 {
1633 	struct timeval tv2;
1634 	size_t n;
1635 	ssize_t r;
1636 	ssize_t a;
1637 	int usec;
1638 
1639 	a = 0;
1640 	r = 0;
1641 	while (nbytes) {
1642 		if (*bwcount + nbytes > BandwidthOpt)
1643 			n = BandwidthOpt - *bwcount;
1644 		else
1645 			n = nbytes;
1646 		if (n)
1647 			r = write(fd, buf, n);
1648 		if (r >= 0) {
1649 			a += r;
1650 			nbytes -= r;
1651 			buf = (const char *)buf + r;
1652 		}
1653 		if ((size_t)r != n)
1654 			break;
1655 		*bwcount += n;
1656 		if (*bwcount >= BandwidthOpt) {
1657 			gettimeofday(&tv2, NULL);
1658 			usec = (int)(tv2.tv_sec - tv1->tv_sec) * 1000000 +
1659 				(int)(tv2.tv_usec - tv1->tv_usec);
1660 			if (usec >= 0 && usec < 1000000)
1661 				usleep(1000000 - usec);
1662 			gettimeofday(tv1, NULL);
1663 			*bwcount -= BandwidthOpt;
1664 		}
1665 	}
1666 	return(a ? a : r);
1667 }
1668 
1669 /*
1670  * Get a yes or no answer from the terminal.  The program may be run as
1671  * part of a two-way pipe so we cannot use stdin for this operation.
1672  */
1673 static int
1674 getyntty(void)
1675 {
1676 	char buf[256];
1677 	FILE *fp;
1678 	int result;
1679 
1680 	fp = fopen("/dev/tty", "r");
1681 	if (fp == NULL) {
1682 		fprintf(stderr, "No terminal for response\n");
1683 		return(-1);
1684 	}
1685 	result = -1;
1686 	while (fgets(buf, sizeof(buf), fp) != NULL) {
1687 		if (buf[0] == 'y' || buf[0] == 'Y') {
1688 			result = 1;
1689 			break;
1690 		}
1691 		if (buf[0] == 'n' || buf[0] == 'N') {
1692 			result = 0;
1693 			break;
1694 		}
1695 		fprintf(stderr, "Response not understood\n");
1696 		break;
1697 	}
1698 	fclose(fp);
1699 	return(result);
1700 }
1701 
1702 static void
1703 score_printf(size_t i, size_t w, const char *ctl, ...)
1704 {
1705 	va_list va;
1706 	size_t n;
1707 	static size_t SSize;
1708 	static int SFd = -1;
1709 	static char ScoreBuf[1024];
1710 
1711 	if (ScoreBoardFile == NULL)
1712 		return;
1713 	assert(i + w < sizeof(ScoreBuf));
1714 	if (SFd < 0) {
1715 		SFd = open(ScoreBoardFile, O_RDWR|O_CREAT|O_TRUNC, 0644);
1716 		if (SFd < 0)
1717 			return;
1718 		SSize = 0;
1719 	}
1720 	for (n = 0; n < i; ++n) {
1721 		if (ScoreBuf[n] == 0)
1722 			ScoreBuf[n] = ' ';
1723 	}
1724 	va_start(va, ctl);
1725 	vsnprintf(ScoreBuf + i, w - 1, ctl, va);
1726 	va_end(va);
1727 	n = strlen(ScoreBuf + i);
1728 	while (n < w - 1) {
1729 		ScoreBuf[i + n] = ' ';
1730 		++n;
1731 	}
1732 	ScoreBuf[i + n] = '\n';
1733 	if (SSize < i + w)
1734 		SSize = i + w;
1735 	pwrite(SFd, ScoreBuf, SSize, 0);
1736 }
1737 
1738 static void
1739 hammer_check_restrict(const char *filesystem)
1740 {
1741 	size_t rlen;
1742 	int atslash;
1743 
1744 	if (RestrictTarget == NULL)
1745 		return;
1746 	rlen = strlen(RestrictTarget);
1747 	if (strncmp(filesystem, RestrictTarget, rlen) != 0) {
1748 		errx(1, "hammer-remote: restricted target");
1749 		/* not reached */
1750 	}
1751 
1752 	atslash = 1;
1753 	while (filesystem[rlen]) {
1754 		if (atslash &&
1755 		    filesystem[rlen] == '.' &&
1756 		    filesystem[rlen+1] == '.') {
1757 			errx(1, "hammer-remote: '..' not allowed");
1758 			/* not reached */
1759 		}
1760 		if (filesystem[rlen] == '/')
1761 			atslash = 1;
1762 		else
1763 			atslash = 0;
1764 		++rlen;
1765 	}
1766 }
1767 
1768 static void
1769 mirror_usage(int code)
1770 {
1771 	fprintf(stderr,
1772 		"hammer mirror-read <filesystem> [begin-tid]\n"
1773 		"hammer mirror-read-stream <filesystem> [begin-tid]\n"
1774 		"hammer mirror-write <filesystem>\n"
1775 		"hammer mirror-dump [header]\n"
1776 		"hammer mirror-copy [[user@]host:]<filesystem>"
1777 				  " [[user@]host:]<filesystem>\n"
1778 		"hammer mirror-stream [[user@]host:]<filesystem>"
1779 				    " [[user@]host:]<filesystem>\n"
1780 	);
1781 	exit(code);
1782 }
1783 
1784