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.
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 %016llx to %016llx\n",
207 mirror.tid_beg, mirror.tid_end);
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);
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=%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,
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 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);
326 fprintf(stderr, "Cyclefile %s updated to 0x%016llx\n",
327 CyclePath, sync_tid);
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);
336 if (streaming && interrupted == 0) {
337 time_t t1 = time(NULL);
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
347 if (TwoWayPipeOpt && streaming && histogram > 0) {
349 fprintf(stderr, " (bulk incremental)");
354 fprintf(stderr, " W");
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));
362 t2 = time(NULL) - t1;
363 if (t2 >= 0 && t2 < DelayOpt) {
365 fprintf(stderr, "\bD");
368 sleep(DelayOpt - t2);
371 fprintf(stderr, "\b ");
378 write_mrecord(1, HAMMER_MREC_TYPE_TERM,
379 &mrec_tmp, sizeof(mrec_tmp.sync));
381 fprintf(stderr, "Mirror-read %s succeeded\n", filesystem);
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.
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.
399 generate_histogram(int fd, const char *filesystem,
400 hammer_tid_t **histogram_ary,
401 struct hammer_ioc_mirror_rw *mirror_base)
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;
409 mirror = *mirror_base;
410 tid_beg = mirror.tid_beg;
411 tid_end = mirror.tid_end;
414 free(*histogram_ary);
415 if (tid_beg + BULK_MINIMUM >= tid_end)
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;
423 mirror.tid_beg = tid_beg;
424 mirror.tid_end = tid_half;
426 if (VerboseOpt > 1) {
427 fprintf(stderr, "RangeTest %016llx/%016llx - %016llx (%lld) ",
431 (long long)(tid_half - tid_beg));
434 if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
435 fprintf(stderr, "Mirror-read %s failed: %s\n",
436 filesystem, strerror(errno));
439 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
441 "Mirror-read %s fatal error %d\n",
442 filesystem, mirror.head.error);
446 fprintf(stderr, "%d\n", mirror.count);
447 if (mirror.count > SERIALBUF_SIZE / 2) {
454 tid_end = mirror_base->tid_end;
455 fprintf(stderr, "histogram range %016llx - %016llx\n",
456 (long long)tid_beg, (long long)tid_end);
459 * The final array generates our incremental ending tids in
460 * reverse order. The caller also picks them off in reverse order.
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;
470 create_pfs(const char *filesystem, uuid_t *s_uuid)
472 if (ForceYesOpt == 1) {
473 fprintf(stderr, "PFS slave %s does not exist. "
474 "Auto create new slave PFS!\n", filesystem);
477 fprintf(stderr, "PFS slave %s does not exist.\n"
478 "Do you want to create a new slave PFS? (yes|no) ",
482 fprintf(stderr, "Aborting operation\n");
488 char *shared_uuid = NULL;
489 uuid_to_string(s_uuid, &shared_uuid, &status);
492 asprintf(&cmd, "/sbin/hammer pfs-slave '%s' shared-uuid=%s 1>&2",
493 filesystem, shared_uuid);
497 fprintf(stderr, "Failed to alloc memory\n");
500 if (system(cmd) != 0) {
501 fprintf(stderr, "Failed to create PFS\n");
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.
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.
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
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.
532 hammer_cmd_mirror_write(char **av, int ac)
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;
551 pickup.signature = 0;
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;
561 * Read initial packet
563 mrec = read_mrecord(0, &error, &pickup);
566 fprintf(stderr, "validate_mrec_header: short read\n");
572 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
575 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
576 fprintf(stderr, "validate_mrec_header: did not get expected "
577 "PFSD record type\n");
580 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
581 fprintf(stderr, "validate_mrec_header: unexpected payload "
587 * Create slave PFS if it doesn't yet exist
589 if (lstat(filesystem, &st) != 0) {
590 create_pfs(filesystem, &mrec->pfs.pfsd.shared_uuid);
595 fd = getpfs(&pfs, filesystem);
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.
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));
613 * Read and process the PFS header. The source informs us of
614 * the TID range the stream represents.
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 */
624 mirror.size = SERIALBUF_SIZE;
627 * Read and process bulk records (REC, PASS, and SKIP types).
629 * On your life, do NOT mess with mirror.key_cur or your mirror
630 * target may become history.
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)
639 if (ioctl(fd, HAMMERIOC_MIRROR_WRITE, &mirror) < 0) {
640 fprintf(stderr, "Mirror-write %s failed: %s\n",
641 filesystem, strerror(errno));
644 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
646 "Mirror-write %s fatal error %d\n",
647 filesystem, mirror.head.error);
651 if (mirror.head.flags & HAMMER_IOC_HEAD_INTR) {
653 "Mirror-write %s interrupted by timer at"
656 mirror.key_cur.obj_id);
663 * Read and process the termination sync record.
665 mrec = read_mrecord(0, &error, &pickup);
667 if (mrec && mrec->head.type == HAMMER_MREC_TYPE_TERM) {
668 fprintf(stderr, "Mirror-write: received termination request\n");
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);
684 * Update the PFS info on the target so the user has visibility
685 * into the new snapshot, and sync the target filesystem.
687 if (mrec->head.type == HAMMER_MREC_TYPE_SYNC) {
688 update_pfs_snapshot(fd, mirror.tid_end, pfs.pfs_id);
690 bzero(&synctid, sizeof(synctid));
691 synctid.op = HAMMER_SYNCTID_SYNC2;
692 ioctl(fd, HAMMERIOC_SYNCTID, &synctid);
694 if (VerboseOpt >= 2) {
695 fprintf(stderr, "Mirror-write %s: succeeded\n",
704 * Report back to the originator.
707 mrec_tmp.update.tid = mirror.tid_end;
708 write_mrecord(1, HAMMER_MREC_TYPE_UPDATE,
709 &mrec_tmp, sizeof(mrec_tmp.update));
711 printf("Source can update synctid to 0x%016llx\n",
719 hammer_cmd_mirror_dump(void)
721 char *buf = malloc(SERIALBUF_SIZE);
722 struct hammer_ioc_mrecord_head pickup;
723 hammer_ioc_mrecord_any_t mrec;
730 * Read and process the PFS header
732 pickup.signature = 0;
735 mrec = read_mrecord(0, &error, &pickup);
738 * Read and process bulk records
741 size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
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");
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 "
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)");
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);
772 case HAMMER_MREC_TYPE_PASS:
773 printf("Pass obj=%016llx key=%016llx "
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);
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);
801 * Read and process the termination sync record.
803 mrec = read_mrecord(0, &error, &pickup);
805 (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
806 mrec->head.type != HAMMER_MREC_TYPE_IDLE)
808 fprintf(stderr, "Mirror-dump: Did not get termination "
814 hammer_cmd_mirror_copy(char **av, int ac, int streaming)
838 if ((pid1 = fork()) == 0) {
843 if ((ptr = strchr(av[0], ':')) != NULL) {
848 xav[xac++] = "hammer";
868 snprintf(tbuf, sizeof(tbuf), "%d", TimeoutOpt);
873 xav[xac++] = "mirror-read-stream";
875 xav[xac++] = "mirror-read";
878 execv("/usr/bin/ssh", (void *)xav);
880 hammer_cmd_mirror_read(av, 1, streaming);
890 if ((pid2 = fork()) == 0) {
895 if ((ptr = strchr(av[1], ':')) != NULL) {
900 xav[xac++] = "hammer";
919 xav[xac++] = "mirror-write";
922 execv("/usr/bin/ssh", (void *)xav);
924 hammer_cmd_mirror_write(av + 1, 1);
933 while (waitpid(pid1, NULL, 0) <= 0)
935 while (waitpid(pid2, NULL, 0) <= 0)
939 * If the link is lost restart
943 fprintf(stderr, "\nLost Link\n");
953 * Read and return multiple mrecords
956 read_mrecords(int fd, char *buf, u_int size, hammer_ioc_mrecord_head_t pickup)
958 hammer_ioc_mrecord_any_t mrec;
966 while (size - count >= HAMMER_MREC_HEADSIZE) {
968 * Cached the record header in case we run out of buffer
972 if (pickup->signature == 0) {
973 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
974 i = read(fd, (char *)pickup + n,
975 HAMMER_MREC_HEADSIZE - n);
981 if (n != HAMMER_MREC_HEADSIZE) {
982 fprintf(stderr, "read_mrecords: short read on pipe\n");
985 if (pickup->signature != HAMMER_IOC_MIRROR_SIGNATURE) {
986 fprintf(stderr, "read_mrecords: malformed record on pipe, "
991 if (pickup->rec_size < HAMMER_MREC_HEADSIZE ||
992 pickup->rec_size > sizeof(*mrec) + HAMMER_XBUFSIZE) {
993 fprintf(stderr, "read_mrecords: malformed record on pipe, "
994 "illegal rec_size\n");
999 * Stop if we have insufficient space for the record and data.
1001 bytes = HAMMER_HEAD_DOALIGN(pickup->rec_size);
1002 if (size - count < bytes)
1006 * Stop if the record type is not a REC, SKIP, or PASS,
1007 * which are the only types the ioctl supports. Other types
1008 * are used only by the userland protocol.
1012 type = pickup->type & HAMMER_MRECF_TYPE_LOMASK;
1013 if (type != HAMMER_MREC_TYPE_PFSD &&
1014 type != HAMMER_MREC_TYPE_REC &&
1015 type != HAMMER_MREC_TYPE_SKIP &&
1016 type != HAMMER_MREC_TYPE_PASS) {
1021 * Read the remainder and clear the pickup signature.
1023 for (n = HAMMER_MREC_HEADSIZE; n < bytes; n += i) {
1024 i = read(fd, buf + count + n, bytes - n);
1029 fprintf(stderr, "read_mrecords: short read on pipe\n");
1033 bcopy(pickup, buf + count, HAMMER_MREC_HEADSIZE);
1034 pickup->signature = 0;
1036 mrec = (void *)(buf + count);
1039 * Validate the completed record
1041 if (mrec->head.rec_crc !=
1042 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1043 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1044 fprintf(stderr, "read_mrecords: malformed record "
1045 "on pipe, bad crc\n");
1050 * If its a B-Tree record validate the data crc.
1052 * NOTE: If the VFS passes us an explicitly errorde mrec
1053 * we just pass it through.
1055 type = mrec->head.type & HAMMER_MRECF_TYPE_MASK;
1057 if (type == HAMMER_MREC_TYPE_REC) {
1058 if (mrec->head.rec_size <
1059 sizeof(mrec->rec) + mrec->rec.leaf.data_len) {
1061 "read_mrecords: malformed record on "
1062 "pipe, illegal element data_len\n");
1065 if (mrec->rec.leaf.data_len &&
1066 mrec->rec.leaf.data_offset &&
1067 hammer_crc_test_leaf(&mrec->rec + 1, &mrec->rec.leaf) == 0) {
1069 "read_mrecords: data_crc did not "
1070 "match data! obj=%016llx key=%016llx\n",
1071 mrec->rec.leaf.base.obj_id,
1072 mrec->rec.leaf.base.key);
1074 "continuing, but there are problems\n");
1083 * Read and return a single mrecord.
1086 hammer_ioc_mrecord_any_t
1087 read_mrecord(int fdin, int *errorp, hammer_ioc_mrecord_head_t pickup)
1089 hammer_ioc_mrecord_any_t mrec;
1090 struct hammer_ioc_mrecord_head mrechd;
1095 if (pickup && pickup->type != 0) {
1097 pickup->signature = 0;
1099 n = HAMMER_MREC_HEADSIZE;
1102 * Read in the PFSD header from the sender.
1104 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1105 i = read(fdin, (char *)&mrechd + n, HAMMER_MREC_HEADSIZE - n);
1110 *errorp = 0; /* EOF */
1113 if (n != HAMMER_MREC_HEADSIZE) {
1114 fprintf(stderr, "short read of mrecord header\n");
1119 if (mrechd.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1120 fprintf(stderr, "read_mrecord: bad signature\n");
1124 bytes = HAMMER_HEAD_DOALIGN(mrechd.rec_size);
1125 assert(bytes >= sizeof(mrechd));
1126 mrec = malloc(bytes);
1127 mrec->head = mrechd;
1130 i = read(fdin, (char *)mrec + n, bytes - n);
1136 fprintf(stderr, "read_mrecord: short read on payload\n");
1140 if (mrec->head.rec_crc !=
1141 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1142 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1143 fprintf(stderr, "read_mrecord: bad CRC\n");
1153 write_mrecord(int fdout, u_int32_t type, hammer_ioc_mrecord_any_t mrec,
1156 char zbuf[HAMMER_HEAD_ALIGN];
1159 pad = HAMMER_HEAD_DOALIGN(bytes) - bytes;
1161 assert(bytes >= (int)sizeof(mrec->head));
1162 bzero(&mrec->head, sizeof(mrec->head));
1163 mrec->head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
1164 mrec->head.type = type;
1165 mrec->head.rec_size = bytes;
1166 mrec->head.rec_crc = crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1167 bytes - HAMMER_MREC_CRCOFF);
1168 if (write(fdout, mrec, bytes) != bytes) {
1169 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1170 errno, strerror(errno));
1175 if (write(fdout, zbuf, pad) != pad) {
1176 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1177 errno, strerror(errno));
1184 * Generate a mirroring header with the pfs information of the
1185 * originating filesytem.
1188 generate_mrec_header(int fd, int pfs_id,
1189 union hammer_ioc_mrecord_any *mrec_tmp)
1191 struct hammer_ioc_pseudofs_rw pfs;
1193 bzero(&pfs, sizeof(pfs));
1194 bzero(mrec_tmp, sizeof(*mrec_tmp));
1195 pfs.pfs_id = pfs_id;
1196 pfs.ondisk = &mrec_tmp->pfs.pfsd;
1197 pfs.bytes = sizeof(mrec_tmp->pfs.pfsd);
1198 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1199 fprintf(stderr, "Mirror-read: not a HAMMER fs/pseudofs!\n");
1202 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1203 fprintf(stderr, "Mirror-read: HAMMER pfs version mismatch!\n");
1206 mrec_tmp->pfs.version = pfs.version;
1210 * Validate the pfs information from the originating filesystem
1211 * against the target filesystem. shared_uuid must match.
1213 * return -1 if we got a TERM record
1216 validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
1217 struct hammer_ioc_mrecord_head *pickup,
1218 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp)
1220 struct hammer_ioc_pseudofs_rw pfs;
1221 struct hammer_pseudofs_data pfsd;
1222 hammer_ioc_mrecord_any_t mrec;
1226 * Get the PFSD info from the target filesystem.
1228 bzero(&pfs, sizeof(pfs));
1229 bzero(&pfsd, sizeof(pfsd));
1230 pfs.pfs_id = pfs_id;
1232 pfs.bytes = sizeof(pfsd);
1233 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1234 fprintf(stderr, "mirror-write: not a HAMMER fs/pseudofs!\n");
1237 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1238 fprintf(stderr, "mirror-write: HAMMER pfs version mismatch!\n");
1242 mrec = read_mrecord(fdin, &error, pickup);
1245 fprintf(stderr, "validate_mrec_header: short read\n");
1248 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
1253 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
1254 fprintf(stderr, "validate_mrec_header: did not get expected "
1255 "PFSD record type\n");
1258 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
1259 fprintf(stderr, "validate_mrec_header: unexpected payload "
1263 if (mrec->pfs.version != pfs.version) {
1264 fprintf(stderr, "validate_mrec_header: Version mismatch\n");
1269 * Whew. Ok, is the read PFS info compatible with the target?
1271 if (bcmp(&mrec->pfs.pfsd.shared_uuid, &pfsd.shared_uuid,
1272 sizeof(pfsd.shared_uuid)) != 0) {
1274 "mirror-write: source and target have "
1275 "different shared-uuid's!\n");
1279 (pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
1280 fprintf(stderr, "mirror-write: target must be in slave mode\n");
1284 *tid_begp = mrec->pfs.pfsd.sync_beg_tid;
1286 *tid_endp = mrec->pfs.pfsd.sync_end_tid;
1292 update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id)
1294 struct hammer_ioc_pseudofs_rw pfs;
1295 struct hammer_pseudofs_data pfsd;
1297 bzero(&pfs, sizeof(pfs));
1298 bzero(&pfsd, sizeof(pfsd));
1299 pfs.pfs_id = pfs_id;
1301 pfs.bytes = sizeof(pfsd);
1302 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1303 perror("update_pfs_snapshot (read)");
1306 if (pfsd.sync_end_tid != snapshot_tid) {
1307 pfsd.sync_end_tid = snapshot_tid;
1308 if (ioctl(fd, HAMMERIOC_SET_PSEUDOFS, &pfs) != 0) {
1309 perror("update_pfs_snapshot (rewrite)");
1312 if (VerboseOpt >= 2) {
1314 "Mirror-write: Completed, updated snapshot "
1322 * Bandwidth-limited write in chunks
1326 writebw(int fd, const void *buf, size_t nbytes,
1327 u_int64_t *bwcount, struct timeval *tv1)
1338 if (*bwcount + nbytes > BandwidthOpt)
1339 n = BandwidthOpt - *bwcount;
1343 r = write(fd, buf, n);
1347 buf = (const char *)buf + r;
1352 if (*bwcount >= BandwidthOpt) {
1353 gettimeofday(&tv2, NULL);
1354 usec = (int)(tv2.tv_sec - tv1->tv_sec) * 1000000 +
1355 (int)(tv2.tv_usec - tv1->tv_usec);
1356 if (usec >= 0 && usec < 1000000)
1357 usleep(1000000 - usec);
1358 gettimeofday(tv1, NULL);
1359 *bwcount -= BandwidthOpt;
1366 * Get a yes or no answer from the terminal. The program may be run as
1367 * part of a two-way pipe so we cannot use stdin for this operation.
1376 fp = fopen("/dev/tty", "r");
1378 fprintf(stderr, "No terminal for response\n");
1382 while (fgets(buf, sizeof(buf), fp) != NULL) {
1383 if (buf[0] == 'y' || buf[0] == 'Y') {
1387 if (buf[0] == 'n' || buf[0] == 'N') {
1391 fprintf(stderr, "Response not understood\n");
1399 mirror_usage(int code)
1402 "hammer mirror-read <filesystem> [begin-tid]\n"
1403 "hammer mirror-read-stream <filesystem> [begin-tid]\n"
1404 "hammer mirror-write <filesystem>\n"
1405 "hammer mirror-dump\n"
1406 "hammer mirror-copy [[user@]host:]<filesystem>"
1407 " [[user@]host:]<filesystem>\n"
1408 "hammer mirror-stream [[user@]host:]<filesystem>"
1409 " [[user@]host:]<filesystem>\n"