2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
34 * $DragonFly: src/sbin/hammer/cmd_mirror.c,v 1.16 2008/11/09 05:22:56 dillon Exp $
39 #define SERIALBUF_SIZE (512 * 1024)
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);
61 #define BULK_MINIMUM 20000
64 * Generate a mirroring data stream from the specific source over the
65 * entire key range, but restricted to the specified transaction range.
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.
72 hammer_cmd_mirror_read(char **av, int ac, int streaming)
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);
90 time_t base_t = time(NULL);
94 if (ac == 0 || ac > 2)
101 histogram_ary = NULL;
104 bzero(&mirror, sizeof(mirror));
105 hammer_key_beg_init(&mirror.key_beg);
106 hammer_key_end_init(&mirror.key_end);
108 fd = getpfs(&pfs, filesystem);
110 if (streaming && VerboseOpt) {
111 fprintf(stderr, "\nRunning");
115 gettimeofday(&bwtv, NULL);
119 * Send initial header for the purpose of determining the
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));
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
133 n = validate_mrec_header(fd, 0, 0, pfs.pfs_id, &pickup,
134 NULL, &mirror.tid_beg);
135 if (n < 0) { /* got TERM record */
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.
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;
152 mirror.size = SERIALBUF_SIZE;
153 mirror.pfs_id = pfs.pfs_id;
154 mirror.shared_uuid = pfs.ondisk->shared_uuid;
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.
162 if (TwoWayPipeOpt && streaming && histogram == 0) {
163 if (mirror.tid_end - mirror.tid_beg > BULK_MINIMUM)
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.
173 if (histogram < 0 && BulkOpt == 0) {
175 fprintf(stderr, "\n");
176 histogram = generate_histogram(fd, filesystem,
177 &histogram_ary, &mirror);
180 if (TwoWayPipeOpt && streaming && histogram > 0) {
181 mirror.tid_end = histogram_ary[--histogram];
182 mrec_tmp.pfs.pfsd.sync_end_tid = mirror.tid_end;
185 write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
186 &mrec_tmp, sizeof(mrec_tmp.pfs));
189 * A cycle file overrides the beginning TID only if we are
190 * not operating in two-way mode.
192 if (TwoWayPipeOpt == 0) {
193 hammer_get_cycle(&mirror.key_beg, &mirror.tid_beg);
197 * An additional argument overrides the beginning TID regardless
198 * of what mode we are in. This is not recommending if operating
202 mirror.tid_beg = strtoull(av[1], NULL, 0);
204 if (streaming == 0 || VerboseOpt >= 2) {
206 "Mirror-read: Mirror from %016jx to %016jx\n",
207 (uintmax_t)mirror.tid_beg, (uintmax_t)mirror.tid_end);
209 if (mirror.key_beg.obj_id != (int64_t)HAMMER_MIN_OBJID) {
210 fprintf(stderr, "Mirror-read: Resuming at object %016jx\n",
211 (uintmax_t)mirror.key_beg.obj_id);
215 * Nothing to do if begin equals end.
217 if (mirror.tid_beg >= mirror.tid_end) {
218 if (streaming == 0 || VerboseOpt >= 2)
219 fprintf(stderr, "Mirror-read: No work to do\n");
226 * Write out bulk records
229 mirror.size = SERIALBUF_SIZE;
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));
240 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
242 "Mirror-read %s fatal error %d\n",
243 filesystem, mirror.head.error);
248 n = writebw(1, mirror.ubuf, mirror.count,
251 n = write(1, mirror.ubuf, mirror.count);
253 if (n != mirror.count) {
254 fprintf(stderr, "Mirror-read %s failed: "
260 total_bytes += mirror.count;
261 if (streaming && VerboseOpt) {
263 "\robj=%016jx tids=%016jx:%016jx %11jd",
264 (uintmax_t)mirror.key_cur.obj_id,
265 (uintmax_t)mirror.tid_beg,
266 (uintmax_t)mirror.tid_end,
267 (intmax_t)total_bytes);
270 mirror.key_beg = mirror.key_cur;
273 * Deal with time limit option
276 (unsigned)(time(NULL) - base_t) > (unsigned)TimeoutOpt) {
278 "Mirror-read %s interrupted by timer at"
281 (uintmax_t)mirror.key_cur.obj_id);
285 } while (mirror.count != 0);
289 * Write out the termination sync record - only if not interrupted
291 if (interrupted == 0) {
293 write_mrecord(1, HAMMER_MREC_TYPE_SYNC,
294 &mrec_tmp, sizeof(mrec_tmp.sync));
296 write_mrecord(1, HAMMER_MREC_TYPE_IDLE,
297 &mrec_tmp, sizeof(mrec_tmp.sync));
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
307 mrec = read_mrecord(0, &error, &pickup);
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");
317 hammer_set_cycle(&mirror.key_cur, mirror.tid_beg);
318 fprintf(stderr, "Cyclefile %s updated for "
319 "continuation\n", CyclePath);
322 sync_tid = mrec->update.tid;
324 hammer_key_beg_init(&mirror.key_beg);
325 hammer_set_cycle(&mirror.key_beg, sync_tid);
327 "Cyclefile %s updated to 0x%016jx\n",
328 CyclePath, (uintmax_t)sync_tid);
331 } else if (CyclePath) {
332 /* NOTE! mirror.tid_beg cannot be updated */
333 fprintf(stderr, "Warning: cycle file (-c option) cannot be "
334 "fully updated unless you use mirror-copy\n");
335 hammer_set_cycle(&mirror.key_beg, mirror.tid_beg);
337 if (streaming && interrupted == 0) {
338 time_t t1 = time(NULL);
342 * Two way streaming tries to break down large bulk
343 * transfers into smaller ones so it can sync the
344 * transaction id on the slave. This way if we get
345 * interrupted a restart doesn't have to start from
348 if (TwoWayPipeOpt && streaming && histogram > 0) {
350 fprintf(stderr, " (bulk incremental)");
355 fprintf(stderr, " W");
358 pfs.ondisk->sync_end_tid = mirror.tid_end;
359 if (ioctl(fd, HAMMERIOC_WAI_PSEUDOFS, &pfs) < 0) {
360 fprintf(stderr, "Mirror-read %s: cannot stream: %s\n",
361 filesystem, strerror(errno));
363 t2 = time(NULL) - t1;
364 if (t2 >= 0 && t2 < DelayOpt) {
366 fprintf(stderr, "\bD");
369 sleep(DelayOpt - t2);
372 fprintf(stderr, "\b ");
379 write_mrecord(1, HAMMER_MREC_TYPE_TERM,
380 &mrec_tmp, sizeof(mrec_tmp.sync));
382 fprintf(stderr, "Mirror-read %s succeeded\n", filesystem);
386 * Ok, this isn't really a histogram. What we are trying to do
387 * here is find the first tid_end for the scan that returns
388 * at least some data. The frontend of the TID space will generally
389 * return nothing so we can't just divide out the full mirroring
390 * range. Once we find the point where a real data stream starts
391 * to get generated we can divide out the range from that point.
393 * When starting a new mirroring operation completely from scratch
394 * this code will take some time to run, but once some mirroring
395 * data is synchronized on the target you will be able to interrupt
396 * the stream and restart it and the later invocations of this
397 * code will be such that it should run much faster.
400 generate_histogram(int fd, const char *filesystem,
401 hammer_tid_t **histogram_ary,
402 struct hammer_ioc_mirror_rw *mirror_base)
404 struct hammer_ioc_mirror_rw mirror;
405 hammer_tid_t tid_beg;
406 hammer_tid_t tid_end;
407 hammer_tid_t tid_half;
410 mirror = *mirror_base;
411 tid_beg = mirror.tid_beg;
412 tid_end = mirror.tid_end;
415 free(*histogram_ary);
416 if (tid_beg + BULK_MINIMUM >= tid_end)
420 fprintf(stderr, "Doing Range Test\n");
421 while (tid_end - tid_beg > BULK_MINIMUM) {
422 tid_half = tid_beg + (tid_end - tid_beg) * 2 / 3;
424 mirror.tid_beg = tid_beg;
425 mirror.tid_end = tid_half;
427 if (VerboseOpt > 1) {
428 fprintf(stderr, "RangeTest %016jx/%016jx - %016jx (%jd) ",
432 (intmax_t)(tid_half - tid_beg));
435 if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
436 fprintf(stderr, "Mirror-read %s failed: %s\n",
437 filesystem, strerror(errno));
440 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
442 "Mirror-read %s fatal error %d\n",
443 filesystem, mirror.head.error);
447 fprintf(stderr, "%d\n", mirror.count);
448 if (mirror.count > SERIALBUF_SIZE / 2) {
455 tid_end = mirror_base->tid_end;
456 fprintf(stderr, "histogram range %016llx - %016llx\n",
457 (long long)tid_beg, (long long)tid_end);
460 * The final array generates our incremental ending tids in
461 * reverse order. The caller also picks them off in reverse order.
463 *histogram_ary = malloc(sizeof(hammer_tid_t) * 20);
464 for (i = 0; i < 20; ++i) {
465 (*histogram_ary)[i] = tid_end - (tid_end - tid_beg) / 20 * i;
471 create_pfs(const char *filesystem, uuid_t *s_uuid)
473 if (ForceYesOpt == 1) {
474 fprintf(stderr, "PFS slave %s does not exist. "
475 "Auto create new slave PFS!\n", filesystem);
478 fprintf(stderr, "PFS slave %s does not exist.\n"
479 "Do you want to create a new slave PFS? (yes|no) ",
483 fprintf(stderr, "Aborting operation\n");
489 char *shared_uuid = NULL;
490 uuid_to_string(s_uuid, &shared_uuid, &status);
493 asprintf(&cmd, "/sbin/hammer pfs-slave '%s' shared-uuid=%s 1>&2",
494 filesystem, shared_uuid);
498 fprintf(stderr, "Failed to alloc memory\n");
501 if (system(cmd) != 0) {
502 fprintf(stderr, "Failed to create PFS\n");
508 * Pipe the mirroring data stream on stdin to the HAMMER VFS, adding
509 * some additional packet types to negotiate TID ranges and to verify
510 * completion. The HAMMER VFS does most of the work.
512 * It is important to note that the mirror.key_{beg,end} range must
513 * match the ranged used by the original. For now both sides use
514 * range the entire key space.
516 * It is even more important that the records in the stream conform
517 * to the TID range also supplied in the stream. The HAMMER VFS will
518 * use the REC, PASS, and SKIP record types to track the portions of
519 * the B-Tree being scanned in order to be able to proactively delete
520 * records on the target within those active areas that are not mentioned
523 * The mirror.key_cur field is used by the VFS to do this tracking. It
524 * must be initialized to key_beg but then is persistently updated by
525 * the HAMMER VFS on each successive ioctl() call. If you blow up this
526 * field you will blow up the mirror target, possibly to the point of
527 * deleting everything. As a safety measure the HAMMER VFS simply marks
528 * the records that the source has destroyed as deleted on the target,
529 * and normal pruning operations will deal with their final disposition
530 * at some later time.
533 hammer_cmd_mirror_write(char **av, int ac)
535 struct hammer_ioc_mirror_rw mirror;
536 const char *filesystem;
537 char *buf = malloc(SERIALBUF_SIZE);
538 struct hammer_ioc_pseudofs_rw pfs;
539 struct hammer_ioc_mrecord_head pickup;
540 struct hammer_ioc_synctid synctid;
541 union hammer_ioc_mrecord_any mrec_tmp;
542 hammer_ioc_mrecord_any_t mrec;
552 pickup.signature = 0;
556 bzero(&mirror, sizeof(mirror));
557 hammer_key_beg_init(&mirror.key_beg);
558 hammer_key_end_init(&mirror.key_end);
559 mirror.key_end = mirror.key_beg;
562 * Read initial packet
564 mrec = read_mrecord(0, &error, &pickup);
567 fprintf(stderr, "validate_mrec_header: short read\n");
573 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
576 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
577 fprintf(stderr, "validate_mrec_header: did not get expected "
578 "PFSD record type\n");
581 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
582 fprintf(stderr, "validate_mrec_header: unexpected payload "
588 * Create slave PFS if it doesn't yet exist
590 if (lstat(filesystem, &st) != 0) {
591 create_pfs(filesystem, &mrec->pfs.pfsd.shared_uuid);
596 fd = getpfs(&pfs, filesystem);
599 * In two-way mode the target writes out a PFS packet first.
600 * The source uses our tid_end as its tid_beg by default,
601 * picking up where it left off.
605 generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
606 if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
607 mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
608 mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
609 write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
610 &mrec_tmp, sizeof(mrec_tmp.pfs));
614 * Read and process the PFS header. The source informs us of
615 * the TID range the stream represents.
617 n = validate_mrec_header(fd, 0, 1, pfs.pfs_id, &pickup,
618 &mirror.tid_beg, &mirror.tid_end);
619 if (n < 0) { /* got TERM record */
625 mirror.size = SERIALBUF_SIZE;
628 * Read and process bulk records (REC, PASS, and SKIP types).
630 * On your life, do NOT mess with mirror.key_cur or your mirror
631 * target may become history.
635 mirror.pfs_id = pfs.pfs_id;
636 mirror.shared_uuid = pfs.ondisk->shared_uuid;
637 mirror.size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
638 if (mirror.size <= 0)
640 if (ioctl(fd, HAMMERIOC_MIRROR_WRITE, &mirror) < 0) {
641 fprintf(stderr, "Mirror-write %s failed: %s\n",
642 filesystem, strerror(errno));
645 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
647 "Mirror-write %s fatal error %d\n",
648 filesystem, mirror.head.error);
652 if (mirror.head.flags & HAMMER_IOC_HEAD_INTR) {
654 "Mirror-write %s interrupted by timer at"
657 mirror.key_cur.obj_id);
664 * Read and process the termination sync record.
666 mrec = read_mrecord(0, &error, &pickup);
668 if (mrec && mrec->head.type == HAMMER_MREC_TYPE_TERM) {
669 fprintf(stderr, "Mirror-write: received termination request\n");
675 (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
676 mrec->head.type != HAMMER_MREC_TYPE_IDLE) ||
677 mrec->head.rec_size != sizeof(mrec->sync)) {
678 fprintf(stderr, "Mirror-write %s: Did not get termination "
679 "sync record, or rec_size is wrong rt=%d\n",
680 filesystem, mrec->head.type);
685 * Update the PFS info on the target so the user has visibility
686 * into the new snapshot, and sync the target filesystem.
688 if (mrec->head.type == HAMMER_MREC_TYPE_SYNC) {
689 update_pfs_snapshot(fd, mirror.tid_end, pfs.pfs_id);
691 bzero(&synctid, sizeof(synctid));
692 synctid.op = HAMMER_SYNCTID_SYNC2;
693 ioctl(fd, HAMMERIOC_SYNCTID, &synctid);
695 if (VerboseOpt >= 2) {
696 fprintf(stderr, "Mirror-write %s: succeeded\n",
705 * Report back to the originator.
708 mrec_tmp.update.tid = mirror.tid_end;
709 write_mrecord(1, HAMMER_MREC_TYPE_UPDATE,
710 &mrec_tmp, sizeof(mrec_tmp.update));
712 printf("Source can update synctid to 0x%016jx\n",
713 (uintmax_t)mirror.tid_end);
720 hammer_cmd_mirror_dump(void)
722 char *buf = malloc(SERIALBUF_SIZE);
723 struct hammer_ioc_mrecord_head pickup;
724 hammer_ioc_mrecord_any_t mrec;
731 * Read and process the PFS header
733 pickup.signature = 0;
736 mrec = read_mrecord(0, &error, &pickup);
739 * Read and process bulk records
742 size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
746 while (offset < size) {
747 mrec = (void *)((char *)buf + offset);
748 bytes = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
749 if (offset + bytes > size) {
750 fprintf(stderr, "Misaligned record\n");
754 switch(mrec->head.type & HAMMER_MRECF_TYPE_MASK) {
755 case HAMMER_MREC_TYPE_REC_BADCRC:
756 case HAMMER_MREC_TYPE_REC:
757 printf("Record obj=%016jx key=%016jx "
759 (uintmax_t)mrec->rec.leaf.base.obj_id,
760 (uintmax_t)mrec->rec.leaf.base.key,
761 mrec->rec.leaf.base.rec_type,
762 mrec->rec.leaf.base.obj_type);
763 if (mrec->head.type ==
764 HAMMER_MREC_TYPE_REC_BADCRC) {
765 printf(" (BAD CRC)");
768 printf(" tids %016jx:%016jx data=%d\n",
769 (uintmax_t)mrec->rec.leaf.base.create_tid,
770 (uintmax_t)mrec->rec.leaf.base.delete_tid,
771 mrec->rec.leaf.data_len);
773 case HAMMER_MREC_TYPE_PASS:
774 printf("Pass obj=%016jx key=%016jx "
776 (uintmax_t)mrec->rec.leaf.base.obj_id,
777 (uintmax_t)mrec->rec.leaf.base.key,
778 mrec->rec.leaf.base.rec_type,
779 mrec->rec.leaf.base.obj_type);
780 printf(" tids %016jx:%016jx data=%d\n",
781 (uintmax_t)mrec->rec.leaf.base.create_tid,
782 (uintmax_t)mrec->rec.leaf.base.delete_tid,
783 mrec->rec.leaf.data_len);
785 case HAMMER_MREC_TYPE_SKIP:
786 printf("Skip obj=%016jx key=%016jx rt=%02x to\n"
787 " obj=%016jx key=%016jx rt=%02x\n",
788 (uintmax_t)mrec->skip.skip_beg.obj_id,
789 (uintmax_t)mrec->skip.skip_beg.key,
790 mrec->skip.skip_beg.rec_type,
791 (uintmax_t)mrec->skip.skip_end.obj_id,
792 (uintmax_t)mrec->skip.skip_end.key,
793 mrec->skip.skip_end.rec_type);
802 * Read and process the termination sync record.
804 mrec = read_mrecord(0, &error, &pickup);
806 (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
807 mrec->head.type != HAMMER_MREC_TYPE_IDLE)
809 fprintf(stderr, "Mirror-dump: Did not get termination "
815 hammer_cmd_mirror_copy(char **av, int ac, int streaming)
829 signal(SIGPIPE, SIG_IGN);
840 if ((pid1 = fork()) == 0) {
841 signal(SIGPIPE, SIG_DFL);
846 if ((ptr = strchr(av[0], ':')) != NULL) {
852 xav[xac++] = SshPort;
855 xav[xac++] = "hammer";
875 snprintf(tbuf, sizeof(tbuf), "%d", TimeoutOpt);
880 xav[xac++] = "mirror-read-stream";
882 xav[xac++] = "mirror-read";
885 execv("/usr/bin/ssh", (void *)xav);
887 hammer_cmd_mirror_read(av, 1, streaming);
897 if ((pid2 = fork()) == 0) {
898 signal(SIGPIPE, SIG_DFL);
903 if ((ptr = strchr(av[1], ':')) != NULL) {
909 xav[xac++] = SshPort;
912 xav[xac++] = "hammer";
931 xav[xac++] = "mirror-write";
934 execv("/usr/bin/ssh", (void *)xav);
936 hammer_cmd_mirror_write(av + 1, 1);
945 while (waitpid(pid1, NULL, 0) <= 0)
947 while (waitpid(pid2, NULL, 0) <= 0)
951 * If the link is lost restart
955 fprintf(stderr, "\nLost Link\n");
958 sleep(15 + DelayOpt);
965 * Read and return multiple mrecords
968 read_mrecords(int fd, char *buf, u_int size, hammer_ioc_mrecord_head_t pickup)
970 hammer_ioc_mrecord_any_t mrec;
978 while (size - count >= HAMMER_MREC_HEADSIZE) {
980 * Cached the record header in case we run out of buffer
984 if (pickup->signature == 0) {
985 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
986 i = read(fd, (char *)pickup + n,
987 HAMMER_MREC_HEADSIZE - n);
993 if (n != HAMMER_MREC_HEADSIZE) {
994 fprintf(stderr, "read_mrecords: short read on pipe\n");
997 if (pickup->signature != HAMMER_IOC_MIRROR_SIGNATURE) {
998 fprintf(stderr, "read_mrecords: malformed record on pipe, "
1003 if (pickup->rec_size < HAMMER_MREC_HEADSIZE ||
1004 pickup->rec_size > sizeof(*mrec) + HAMMER_XBUFSIZE) {
1005 fprintf(stderr, "read_mrecords: malformed record on pipe, "
1006 "illegal rec_size\n");
1011 * Stop if we have insufficient space for the record and data.
1013 bytes = HAMMER_HEAD_DOALIGN(pickup->rec_size);
1014 if (size - count < bytes)
1018 * Stop if the record type is not a REC, SKIP, or PASS,
1019 * which are the only types the ioctl supports. Other types
1020 * are used only by the userland protocol.
1024 type = pickup->type & HAMMER_MRECF_TYPE_LOMASK;
1025 if (type != HAMMER_MREC_TYPE_PFSD &&
1026 type != HAMMER_MREC_TYPE_REC &&
1027 type != HAMMER_MREC_TYPE_SKIP &&
1028 type != HAMMER_MREC_TYPE_PASS) {
1033 * Read the remainder and clear the pickup signature.
1035 for (n = HAMMER_MREC_HEADSIZE; n < bytes; n += i) {
1036 i = read(fd, buf + count + n, bytes - n);
1041 fprintf(stderr, "read_mrecords: short read on pipe\n");
1045 bcopy(pickup, buf + count, HAMMER_MREC_HEADSIZE);
1046 pickup->signature = 0;
1048 mrec = (void *)(buf + count);
1051 * Validate the completed record
1053 if (mrec->head.rec_crc !=
1054 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1055 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1056 fprintf(stderr, "read_mrecords: malformed record "
1057 "on pipe, bad crc\n");
1062 * If its a B-Tree record validate the data crc.
1064 * NOTE: If the VFS passes us an explicitly errorde mrec
1065 * we just pass it through.
1067 type = mrec->head.type & HAMMER_MRECF_TYPE_MASK;
1069 if (type == HAMMER_MREC_TYPE_REC) {
1070 if (mrec->head.rec_size <
1071 sizeof(mrec->rec) + mrec->rec.leaf.data_len) {
1073 "read_mrecords: malformed record on "
1074 "pipe, illegal element data_len\n");
1077 if (mrec->rec.leaf.data_len &&
1078 mrec->rec.leaf.data_offset &&
1079 hammer_crc_test_leaf(&mrec->rec + 1, &mrec->rec.leaf) == 0) {
1081 "read_mrecords: data_crc did not "
1082 "match data! obj=%016jx key=%016jx\n",
1083 (uintmax_t)mrec->rec.leaf.base.obj_id,
1084 (uintmax_t)mrec->rec.leaf.base.key);
1086 "continuing, but there are problems\n");
1095 * Read and return a single mrecord.
1098 hammer_ioc_mrecord_any_t
1099 read_mrecord(int fdin, int *errorp, hammer_ioc_mrecord_head_t pickup)
1101 hammer_ioc_mrecord_any_t mrec;
1102 struct hammer_ioc_mrecord_head mrechd;
1107 if (pickup && pickup->type != 0) {
1109 pickup->signature = 0;
1111 n = HAMMER_MREC_HEADSIZE;
1114 * Read in the PFSD header from the sender.
1116 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1117 i = read(fdin, (char *)&mrechd + n, HAMMER_MREC_HEADSIZE - n);
1122 *errorp = 0; /* EOF */
1125 if (n != HAMMER_MREC_HEADSIZE) {
1126 fprintf(stderr, "short read of mrecord header\n");
1131 if (mrechd.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1132 fprintf(stderr, "read_mrecord: bad signature\n");
1136 bytes = HAMMER_HEAD_DOALIGN(mrechd.rec_size);
1137 assert(bytes >= sizeof(mrechd));
1138 mrec = malloc(bytes);
1139 mrec->head = mrechd;
1142 i = read(fdin, (char *)mrec + n, bytes - n);
1148 fprintf(stderr, "read_mrecord: short read on payload\n");
1152 if (mrec->head.rec_crc !=
1153 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1154 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1155 fprintf(stderr, "read_mrecord: bad CRC\n");
1165 write_mrecord(int fdout, u_int32_t type, hammer_ioc_mrecord_any_t mrec,
1168 char zbuf[HAMMER_HEAD_ALIGN];
1171 pad = HAMMER_HEAD_DOALIGN(bytes) - bytes;
1173 assert(bytes >= (int)sizeof(mrec->head));
1174 bzero(&mrec->head, sizeof(mrec->head));
1175 mrec->head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
1176 mrec->head.type = type;
1177 mrec->head.rec_size = bytes;
1178 mrec->head.rec_crc = crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1179 bytes - HAMMER_MREC_CRCOFF);
1180 if (write(fdout, mrec, bytes) != bytes) {
1181 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1182 errno, strerror(errno));
1187 if (write(fdout, zbuf, pad) != pad) {
1188 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1189 errno, strerror(errno));
1196 * Generate a mirroring header with the pfs information of the
1197 * originating filesytem.
1200 generate_mrec_header(int fd, int pfs_id,
1201 union hammer_ioc_mrecord_any *mrec_tmp)
1203 struct hammer_ioc_pseudofs_rw pfs;
1205 bzero(&pfs, sizeof(pfs));
1206 bzero(mrec_tmp, sizeof(*mrec_tmp));
1207 pfs.pfs_id = pfs_id;
1208 pfs.ondisk = &mrec_tmp->pfs.pfsd;
1209 pfs.bytes = sizeof(mrec_tmp->pfs.pfsd);
1210 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1211 fprintf(stderr, "Mirror-read: not a HAMMER fs/pseudofs!\n");
1214 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1215 fprintf(stderr, "Mirror-read: HAMMER pfs version mismatch!\n");
1218 mrec_tmp->pfs.version = pfs.version;
1222 * Validate the pfs information from the originating filesystem
1223 * against the target filesystem. shared_uuid must match.
1225 * return -1 if we got a TERM record
1228 validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
1229 struct hammer_ioc_mrecord_head *pickup,
1230 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp)
1232 struct hammer_ioc_pseudofs_rw pfs;
1233 struct hammer_pseudofs_data pfsd;
1234 hammer_ioc_mrecord_any_t mrec;
1238 * Get the PFSD info from the target filesystem.
1240 bzero(&pfs, sizeof(pfs));
1241 bzero(&pfsd, sizeof(pfsd));
1242 pfs.pfs_id = pfs_id;
1244 pfs.bytes = sizeof(pfsd);
1245 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1246 fprintf(stderr, "mirror-write: not a HAMMER fs/pseudofs!\n");
1249 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1250 fprintf(stderr, "mirror-write: HAMMER pfs version mismatch!\n");
1254 mrec = read_mrecord(fdin, &error, pickup);
1257 fprintf(stderr, "validate_mrec_header: short read\n");
1260 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
1265 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
1266 fprintf(stderr, "validate_mrec_header: did not get expected "
1267 "PFSD record type\n");
1270 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
1271 fprintf(stderr, "validate_mrec_header: unexpected payload "
1275 if (mrec->pfs.version != pfs.version) {
1276 fprintf(stderr, "validate_mrec_header: Version mismatch\n");
1281 * Whew. Ok, is the read PFS info compatible with the target?
1283 if (bcmp(&mrec->pfs.pfsd.shared_uuid, &pfsd.shared_uuid,
1284 sizeof(pfsd.shared_uuid)) != 0) {
1286 "mirror-write: source and target have "
1287 "different shared-uuid's!\n");
1291 (pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
1292 fprintf(stderr, "mirror-write: target must be in slave mode\n");
1296 *tid_begp = mrec->pfs.pfsd.sync_beg_tid;
1298 *tid_endp = mrec->pfs.pfsd.sync_end_tid;
1304 update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id)
1306 struct hammer_ioc_pseudofs_rw pfs;
1307 struct hammer_pseudofs_data pfsd;
1309 bzero(&pfs, sizeof(pfs));
1310 bzero(&pfsd, sizeof(pfsd));
1311 pfs.pfs_id = pfs_id;
1313 pfs.bytes = sizeof(pfsd);
1314 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1315 perror("update_pfs_snapshot (read)");
1318 if (pfsd.sync_end_tid != snapshot_tid) {
1319 pfsd.sync_end_tid = snapshot_tid;
1320 if (ioctl(fd, HAMMERIOC_SET_PSEUDOFS, &pfs) != 0) {
1321 perror("update_pfs_snapshot (rewrite)");
1324 if (VerboseOpt >= 2) {
1326 "Mirror-write: Completed, updated snapshot "
1328 (uintmax_t)snapshot_tid);
1334 * Bandwidth-limited write in chunks
1338 writebw(int fd, const void *buf, size_t nbytes,
1339 u_int64_t *bwcount, struct timeval *tv1)
1350 if (*bwcount + nbytes > BandwidthOpt)
1351 n = BandwidthOpt - *bwcount;
1355 r = write(fd, buf, n);
1359 buf = (const char *)buf + r;
1364 if (*bwcount >= BandwidthOpt) {
1365 gettimeofday(&tv2, NULL);
1366 usec = (int)(tv2.tv_sec - tv1->tv_sec) * 1000000 +
1367 (int)(tv2.tv_usec - tv1->tv_usec);
1368 if (usec >= 0 && usec < 1000000)
1369 usleep(1000000 - usec);
1370 gettimeofday(tv1, NULL);
1371 *bwcount -= BandwidthOpt;
1378 * Get a yes or no answer from the terminal. The program may be run as
1379 * part of a two-way pipe so we cannot use stdin for this operation.
1388 fp = fopen("/dev/tty", "r");
1390 fprintf(stderr, "No terminal for response\n");
1394 while (fgets(buf, sizeof(buf), fp) != NULL) {
1395 if (buf[0] == 'y' || buf[0] == 'Y') {
1399 if (buf[0] == 'n' || buf[0] == 'N') {
1403 fprintf(stderr, "Response not understood\n");
1411 mirror_usage(int code)
1414 "hammer mirror-read <filesystem> [begin-tid]\n"
1415 "hammer mirror-read-stream <filesystem> [begin-tid]\n"
1416 "hammer mirror-write <filesystem>\n"
1417 "hammer mirror-dump\n"
1418 "hammer mirror-copy [[user@]host:]<filesystem>"
1419 " [[user@]host:]<filesystem>\n"
1420 "hammer mirror-stream [[user@]host:]<filesystem>"
1421 " [[user@]host:]<filesystem>\n"