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 typedef struct histogram {
46 static int read_mrecords(int fd, char *buf, u_int size,
47 hammer_ioc_mrecord_head_t pickup);
48 static int generate_histogram(int fd, const char *filesystem,
49 histogram_t *histogram_ary,
50 struct hammer_ioc_mirror_rw *mirror_base);
51 static hammer_ioc_mrecord_any_t read_mrecord(int fdin, int *errorp,
52 hammer_ioc_mrecord_head_t pickup);
53 static void write_mrecord(int fdout, u_int32_t type,
54 hammer_ioc_mrecord_any_t mrec, int bytes);
55 static void generate_mrec_header(int fd, int pfs_id,
56 union hammer_ioc_mrecord_any *mrec_tmp);
57 static int validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
58 struct hammer_ioc_mrecord_head *pickup,
59 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp);
60 static void update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id);
61 static ssize_t writebw(int fd, const void *buf, size_t nbytes,
62 u_int64_t *bwcount, struct timeval *tv1);
63 static int getyn(void);
64 static void mirror_usage(int code);
67 * Generate a mirroring data stream from the specific source over the
68 * entire key range, but restricted to the specified transaction range.
70 * The HAMMER VFS does most of the work, we add a few new mrecord
71 * types to negotiate the TID ranges and verify that the entire
72 * stream made it to the destination.
75 hammer_cmd_mirror_read(char **av, int ac, int streaming)
77 struct hammer_ioc_mirror_rw mirror;
78 struct hammer_ioc_pseudofs_rw pfs;
79 union hammer_ioc_mrecord_any mrec_tmp;
80 struct hammer_ioc_mrecord_head pickup;
81 hammer_ioc_mrecord_any_t mrec;
82 hammer_tid_t sync_tid;
83 histogram_t histogram_ary;
84 const char *filesystem;
85 char *buf = malloc(SERIALBUF_SIZE);
95 time_t base_t = time(NULL);
100 if (ac == 0 || ac > 2)
104 pickup.signature = 0;
109 histogram_ary = NULL;
112 bzero(&mirror, sizeof(mirror));
113 hammer_key_beg_init(&mirror.key_beg);
114 hammer_key_end_init(&mirror.key_end);
116 fd = getpfs(&pfs, filesystem);
118 if (streaming && VerboseOpt && VerboseOpt < 2) {
119 fprintf(stderr, "\nRunning");
123 gettimeofday(&bwtv, NULL);
127 * Send initial header for the purpose of determining the
130 generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
131 write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
132 &mrec_tmp, sizeof(mrec_tmp.pfs));
135 * In 2-way mode the target will send us a PFS info packet
136 * first. Use the target's current snapshot TID as our default
141 n = validate_mrec_header(fd, 0, 0, pfs.pfs_id, &pickup,
142 NULL, &mirror.tid_beg);
143 if (n < 0) { /* got TERM record */
151 * Write out the PFS header, tid_beg will be updated if our PFS
152 * has a larger begin sync. tid_end is set to the latest source
153 * TID whos flush cycle has completed.
155 generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
156 if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
157 mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
158 mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
160 mirror.size = SERIALBUF_SIZE;
161 mirror.pfs_id = pfs.pfs_id;
162 mirror.shared_uuid = pfs.ondisk->shared_uuid;
165 * XXX If the histogram is exhausted and the TID delta is large
166 * the stream might have been offline for a while and is
167 * now picking it up again. Do another histogram.
170 if (TwoWayPipeOpt && streaming && histogram && histindex == histend) {
171 if (mirror.tid_end - mirror.tid_beg > BULK_MINIMUM)
177 * Initial bulk startup control, try to do some incremental
178 * mirroring in order to allow the stream to be killed and
179 * restarted without having to start over.
181 if (histogram == 0 && BulkOpt == 0) {
183 fprintf(stderr, "\n");
184 histmax = generate_histogram(fd, filesystem,
185 &histogram_ary, &mirror);
190 if (TwoWayPipeOpt && streaming && histogram) {
192 mirror.tid_end = histogram_ary[histindex].tid;
193 estbytes = histogram_ary[histindex-1].bytes;
194 mrec_tmp.pfs.pfsd.sync_end_tid = mirror.tid_end;
199 write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
200 &mrec_tmp, sizeof(mrec_tmp.pfs));
203 * A cycle file overrides the beginning TID only if we are
204 * not operating in two-way mode.
206 if (TwoWayPipeOpt == 0) {
207 hammer_get_cycle(&mirror.key_beg, &mirror.tid_beg);
211 * An additional argument overrides the beginning TID regardless
212 * of what mode we are in. This is not recommending if operating
216 mirror.tid_beg = strtoull(av[1], NULL, 0);
218 if (streaming == 0 || VerboseOpt >= 2) {
220 "Mirror-read: Mirror %016jx to %016jx",
221 (uintmax_t)mirror.tid_beg, (uintmax_t)mirror.tid_end);
223 fprintf(stderr, " (bulk= %ju)", (uintmax_t)estbytes);
224 fprintf(stderr, "\n");
227 if (mirror.key_beg.obj_id != (int64_t)HAMMER_MIN_OBJID) {
228 fprintf(stderr, "Mirror-read: Resuming at object %016jx\n",
229 (uintmax_t)mirror.key_beg.obj_id);
233 * Nothing to do if begin equals end.
235 if (mirror.tid_beg >= mirror.tid_end) {
236 if (streaming == 0 || VerboseOpt >= 2)
237 fprintf(stderr, "Mirror-read: No work to do\n");
244 * Write out bulk records
247 mirror.size = SERIALBUF_SIZE;
251 mirror.pfs_id = pfs.pfs_id;
252 mirror.shared_uuid = pfs.ondisk->shared_uuid;
253 if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
254 fprintf(stderr, "Mirror-read %s failed: %s\n",
255 filesystem, strerror(errno));
258 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
260 "Mirror-read %s fatal error %d\n",
261 filesystem, mirror.head.error);
266 n = writebw(1, mirror.ubuf, mirror.count,
269 n = write(1, mirror.ubuf, mirror.count);
271 if (n != mirror.count) {
272 fprintf(stderr, "Mirror-read %s failed: "
278 total_bytes += mirror.count;
279 if (streaming && VerboseOpt) {
281 "\robj=%016jx tids=%016jx:%016jx %11jd",
282 (uintmax_t)mirror.key_cur.obj_id,
283 (uintmax_t)mirror.tid_beg,
284 (uintmax_t)mirror.tid_end,
285 (intmax_t)total_bytes);
288 mirror.key_beg = mirror.key_cur;
291 * Deal with time limit option
294 (unsigned)(time(NULL) - base_t) > (unsigned)TimeoutOpt) {
296 "Mirror-read %s interrupted by timer at"
299 (uintmax_t)mirror.key_cur.obj_id);
303 } while (mirror.count != 0);
306 if (streaming && VerboseOpt) {
307 fprintf(stderr, "\n");
312 * Write out the termination sync record - only if not interrupted
314 if (interrupted == 0) {
316 write_mrecord(1, HAMMER_MREC_TYPE_SYNC,
317 &mrec_tmp, sizeof(mrec_tmp.sync));
319 write_mrecord(1, HAMMER_MREC_TYPE_IDLE,
320 &mrec_tmp, sizeof(mrec_tmp.sync));
325 * If the -2 option was given (automatic when doing mirror-copy),
326 * a two-way pipe is assumed and we expect a response mrec from
330 mrec = read_mrecord(0, &error, &pickup);
332 mrec->head.type != HAMMER_MREC_TYPE_UPDATE ||
333 mrec->head.rec_size != sizeof(mrec->update)) {
334 fprintf(stderr, "mirror_read: Did not get final "
335 "acknowledgement packet from target\n");
340 hammer_set_cycle(&mirror.key_cur,
342 fprintf(stderr, "Cyclefile %s updated for "
343 "continuation\n", CyclePath);
346 sync_tid = mrec->update.tid;
348 hammer_key_beg_init(&mirror.key_beg);
349 hammer_set_cycle(&mirror.key_beg, sync_tid);
351 "Cyclefile %s updated to 0x%016jx\n",
352 CyclePath, (uintmax_t)sync_tid);
355 } else if (CyclePath) {
356 /* NOTE! mirror.tid_beg cannot be updated */
357 fprintf(stderr, "Warning: cycle file (-c option) cannot be "
358 "fully updated unless you use mirror-copy\n");
359 hammer_set_cycle(&mirror.key_beg, mirror.tid_beg);
361 if (streaming && interrupted == 0) {
362 time_t t1 = time(NULL);
366 * Two way streaming tries to break down large bulk
367 * transfers into smaller ones so it can sync the
368 * transaction id on the slave. This way if we get
369 * interrupted a restart doesn't have to start from
372 if (TwoWayPipeOpt && streaming && histogram) {
373 if (histindex != histmax) {
374 if (VerboseOpt && VerboseOpt < 2)
375 fprintf(stderr, " (bulk incremental)");
381 fprintf(stderr, " W");
384 pfs.ondisk->sync_end_tid = mirror.tid_end;
385 if (ioctl(fd, HAMMERIOC_WAI_PSEUDOFS, &pfs) < 0) {
386 fprintf(stderr, "Mirror-read %s: cannot stream: %s\n",
387 filesystem, strerror(errno));
389 t2 = time(NULL) - t1;
390 if (t2 >= 0 && t2 < DelayOpt) {
392 fprintf(stderr, "\bD");
395 sleep(DelayOpt - t2);
398 fprintf(stderr, "\b ");
405 write_mrecord(1, HAMMER_MREC_TYPE_TERM,
406 &mrec_tmp, sizeof(mrec_tmp.sync));
408 fprintf(stderr, "Mirror-read %s succeeded\n", filesystem);
412 * What we are trying to do here is figure out how much data is
413 * going to be sent for the TID range and to break the TID range
414 * down into reasonably-sized slices (from the point of view of
415 * data sent) so a lost connection can restart at a reasonable
416 * place and not all the way back at the beginning.
418 * An entry's TID serves as the end_tid for the prior entry
419 * So we have to offset the calculation by 1 so that TID falls into
420 * the previous entry when populating entries.
422 * Because the transaction id space is bursty we need a relatively
423 * large number of buckets (like a million) to do a reasonable job
424 * for things like an initial bulk mirrors on a very large filesystem.
426 #define HIST_COUNT (1024 * 1024)
429 generate_histogram(int fd, const char *filesystem,
430 histogram_t *histogram_ary,
431 struct hammer_ioc_mirror_rw *mirror_base)
433 struct hammer_ioc_mirror_rw mirror;
434 union hammer_ioc_mrecord_any *mrec;
435 hammer_tid_t tid_beg;
436 hammer_tid_t tid_end;
439 u_int64_t *tid_bytes;
447 mirror = *mirror_base;
448 tid_beg = mirror.tid_beg;
449 tid_end = mirror.tid_end;
450 mirror.head.flags |= HAMMER_IOC_MIRROR_NODATA;
452 if (*histogram_ary == NULL) {
453 *histogram_ary = malloc(sizeof(struct histogram) *
456 if (tid_beg >= tid_end)
460 tid_bytes = malloc(sizeof(*tid_bytes) * (HIST_COUNT + 2));
461 bzero(tid_bytes, sizeof(tid_bytes));
463 fprintf(stderr, "Prescan to break up bulk transfer");
465 fprintf(stderr, " (%juMB chunks)",
466 (uintmax_t)(SplitupOpt / (1024 * 1024)));
467 fprintf(stderr, "\n");
470 * Note: (tid_beg,tid_end), range is inclusive of both beg & end.
472 * Note: Estimates can be off when the mirror is way behind due
479 if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
480 fprintf(stderr, "Mirror-read %s failed: %s\n",
481 filesystem, strerror(errno));
484 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
486 "Mirror-read %s fatal error %d\n",
487 filesystem, mirror.head.error);
492 off += HAMMER_HEAD_DOALIGN(mrec->head.rec_size)
494 mrec = (void *)((char *)mirror.ubuf + off);
497 * We only care about general RECs and PASS
498 * records. We ignore SKIPs.
500 switch (mrec->head.type & HAMMER_MRECF_TYPE_LOMASK) {
501 case HAMMER_MREC_TYPE_REC:
502 case HAMMER_MREC_TYPE_PASS:
509 * Calculate for two indices, create_tid and
510 * delete_tid. Record data only applies to
513 * When tid is exactly on the boundary it really
514 * belongs to the previous entry because scans
515 * are inclusive of the ending entry.
517 tid = mrec->rec.leaf.base.delete_tid;
518 if (tid && tid >= tid_beg && tid <= tid_end) {
519 len = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
520 if (mrec->head.type ==
521 HAMMER_MREC_TYPE_REC) {
522 len -= HAMMER_HEAD_DOALIGN(
523 mrec->rec.leaf.data_len);
526 i = (tid - tid_beg) * HIST_COUNT /
528 tidx = tid_beg + i * (tid_end - tid_beg) /
530 if (tid == tidx && i)
532 assert(i >= 0 && i < HIST_COUNT);
538 tid = mrec->rec.leaf.base.create_tid;
539 if (tid && tid >= tid_beg && tid <= tid_end) {
540 len = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
541 if (mrec->head.type ==
542 HAMMER_MREC_TYPE_REC_NODATA) {
543 len += HAMMER_HEAD_DOALIGN(
544 mrec->rec.leaf.data_len);
546 i = (tid - tid_beg) * HIST_COUNT /
548 tidx = tid_beg + i * (tid_end - tid_beg) /
550 if (tid == tidx && i)
552 assert(i >= 0 && i < HIST_COUNT);
558 if (VerboseOpt > 1) {
559 if (accum > SplitupOpt) {
560 fprintf(stderr, ".");
565 if (mirror.count == 0)
567 mirror.key_beg = mirror.key_cur;
571 * Reduce to SplitupOpt (default 100MB) chunks. This code may
572 * use up to two additional elements. Do the array in-place.
574 * Inefficient degenerate cases can occur if we do not accumulate
575 * at least the requested split amount, so error on the side of
579 (*histogram_ary)[res].tid = tid_beg;
580 (*histogram_ary)[res].bytes = tid_bytes[0];
581 for (i = 1; i < HIST_COUNT; ++i) {
582 if ((*histogram_ary)[res].bytes >= SplitupOpt) {
584 (*histogram_ary)[res].tid = tid_beg +
585 i * (tid_end - tid_beg) /
587 (*histogram_ary)[res].bytes = 0;
590 (*histogram_ary)[res].bytes += tid_bytes[i];
593 (*histogram_ary)[res].tid = tid_end;
594 (*histogram_ary)[res].bytes = -1;
597 fprintf(stderr, "\n"); /* newline after ... */
598 assert(res <= HIST_COUNT);
599 fprintf(stderr, "Prescan %d chunks, total %ju MBytes (",
600 res, (uintmax_t)total / (1024 * 1024));
601 for (i = 0; i < res && i < 3; ++i) {
603 fprintf(stderr, ", ");
604 fprintf(stderr, "%ju", (uintmax_t)(*histogram_ary)[i].bytes);
607 fprintf(stderr, ", ...");
608 fprintf(stderr, ")\n");
615 create_pfs(const char *filesystem, uuid_t *s_uuid)
617 if (ForceYesOpt == 1) {
618 fprintf(stderr, "PFS slave %s does not exist. "
619 "Auto create new slave PFS!\n", filesystem);
622 fprintf(stderr, "PFS slave %s does not exist.\n"
623 "Do you want to create a new slave PFS? (yes|no) ",
627 fprintf(stderr, "Aborting operation\n");
633 char *shared_uuid = NULL;
634 uuid_to_string(s_uuid, &shared_uuid, &status);
637 asprintf(&cmd, "/sbin/hammer pfs-slave '%s' shared-uuid=%s 1>&2",
638 filesystem, shared_uuid);
642 fprintf(stderr, "Failed to alloc memory\n");
645 if (system(cmd) != 0) {
646 fprintf(stderr, "Failed to create PFS\n");
652 * Pipe the mirroring data stream on stdin to the HAMMER VFS, adding
653 * some additional packet types to negotiate TID ranges and to verify
654 * completion. The HAMMER VFS does most of the work.
656 * It is important to note that the mirror.key_{beg,end} range must
657 * match the ranged used by the original. For now both sides use
658 * range the entire key space.
660 * It is even more important that the records in the stream conform
661 * to the TID range also supplied in the stream. The HAMMER VFS will
662 * use the REC, PASS, and SKIP record types to track the portions of
663 * the B-Tree being scanned in order to be able to proactively delete
664 * records on the target within those active areas that are not mentioned
667 * The mirror.key_cur field is used by the VFS to do this tracking. It
668 * must be initialized to key_beg but then is persistently updated by
669 * the HAMMER VFS on each successive ioctl() call. If you blow up this
670 * field you will blow up the mirror target, possibly to the point of
671 * deleting everything. As a safety measure the HAMMER VFS simply marks
672 * the records that the source has destroyed as deleted on the target,
673 * and normal pruning operations will deal with their final disposition
674 * at some later time.
677 hammer_cmd_mirror_write(char **av, int ac)
679 struct hammer_ioc_mirror_rw mirror;
680 const char *filesystem;
681 char *buf = malloc(SERIALBUF_SIZE);
682 struct hammer_ioc_pseudofs_rw pfs;
683 struct hammer_ioc_mrecord_head pickup;
684 struct hammer_ioc_synctid synctid;
685 union hammer_ioc_mrecord_any mrec_tmp;
686 hammer_ioc_mrecord_any_t mrec;
696 pickup.signature = 0;
700 bzero(&mirror, sizeof(mirror));
701 hammer_key_beg_init(&mirror.key_beg);
702 hammer_key_end_init(&mirror.key_end);
703 mirror.key_end = mirror.key_beg;
706 * Read initial packet
708 mrec = read_mrecord(0, &error, &pickup);
711 fprintf(stderr, "validate_mrec_header: short read\n");
717 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
720 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
721 fprintf(stderr, "validate_mrec_header: did not get expected "
722 "PFSD record type\n");
725 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
726 fprintf(stderr, "validate_mrec_header: unexpected payload "
732 * Create slave PFS if it doesn't yet exist
734 if (lstat(filesystem, &st) != 0) {
735 create_pfs(filesystem, &mrec->pfs.pfsd.shared_uuid);
740 fd = getpfs(&pfs, filesystem);
743 * In two-way mode the target writes out a PFS packet first.
744 * The source uses our tid_end as its tid_beg by default,
745 * picking up where it left off.
749 generate_mrec_header(fd, pfs.pfs_id, &mrec_tmp);
750 if (mirror.tid_beg < mrec_tmp.pfs.pfsd.sync_beg_tid)
751 mirror.tid_beg = mrec_tmp.pfs.pfsd.sync_beg_tid;
752 mirror.tid_end = mrec_tmp.pfs.pfsd.sync_end_tid;
753 write_mrecord(1, HAMMER_MREC_TYPE_PFSD,
754 &mrec_tmp, sizeof(mrec_tmp.pfs));
758 * Read and process the PFS header. The source informs us of
759 * the TID range the stream represents.
761 n = validate_mrec_header(fd, 0, 1, pfs.pfs_id, &pickup,
762 &mirror.tid_beg, &mirror.tid_end);
763 if (n < 0) { /* got TERM record */
769 mirror.size = SERIALBUF_SIZE;
772 * Read and process bulk records (REC, PASS, and SKIP types).
774 * On your life, do NOT mess with mirror.key_cur or your mirror
775 * target may become history.
779 mirror.pfs_id = pfs.pfs_id;
780 mirror.shared_uuid = pfs.ondisk->shared_uuid;
781 mirror.size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
782 if (mirror.size <= 0)
784 if (ioctl(fd, HAMMERIOC_MIRROR_WRITE, &mirror) < 0) {
785 fprintf(stderr, "Mirror-write %s failed: %s\n",
786 filesystem, strerror(errno));
789 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
791 "Mirror-write %s fatal error %d\n",
792 filesystem, mirror.head.error);
796 if (mirror.head.flags & HAMMER_IOC_HEAD_INTR) {
798 "Mirror-write %s interrupted by timer at"
801 mirror.key_cur.obj_id);
808 * Read and process the termination sync record.
810 mrec = read_mrecord(0, &error, &pickup);
812 if (mrec && mrec->head.type == HAMMER_MREC_TYPE_TERM) {
813 fprintf(stderr, "Mirror-write: received termination request\n");
819 (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
820 mrec->head.type != HAMMER_MREC_TYPE_IDLE) ||
821 mrec->head.rec_size != sizeof(mrec->sync)) {
822 fprintf(stderr, "Mirror-write %s: Did not get termination "
823 "sync record, or rec_size is wrong rt=%d\n",
824 filesystem, mrec->head.type);
829 * Update the PFS info on the target so the user has visibility
830 * into the new snapshot, and sync the target filesystem.
832 if (mrec->head.type == HAMMER_MREC_TYPE_SYNC) {
833 update_pfs_snapshot(fd, mirror.tid_end, pfs.pfs_id);
835 bzero(&synctid, sizeof(synctid));
836 synctid.op = HAMMER_SYNCTID_SYNC2;
837 ioctl(fd, HAMMERIOC_SYNCTID, &synctid);
839 if (VerboseOpt >= 2) {
840 fprintf(stderr, "Mirror-write %s: succeeded\n",
849 * Report back to the originator.
852 mrec_tmp.update.tid = mirror.tid_end;
853 write_mrecord(1, HAMMER_MREC_TYPE_UPDATE,
854 &mrec_tmp, sizeof(mrec_tmp.update));
856 printf("Source can update synctid to 0x%016jx\n",
857 (uintmax_t)mirror.tid_end);
864 hammer_cmd_mirror_dump(void)
866 char *buf = malloc(SERIALBUF_SIZE);
867 struct hammer_ioc_mrecord_head pickup;
868 hammer_ioc_mrecord_any_t mrec;
875 * Read and process the PFS header
877 pickup.signature = 0;
880 mrec = read_mrecord(0, &error, &pickup);
883 * Read and process bulk records
886 size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
890 while (offset < size) {
891 mrec = (void *)((char *)buf + offset);
892 bytes = HAMMER_HEAD_DOALIGN(mrec->head.rec_size);
893 if (offset + bytes > size) {
894 fprintf(stderr, "Misaligned record\n");
898 switch(mrec->head.type & HAMMER_MRECF_TYPE_MASK) {
899 case HAMMER_MREC_TYPE_REC_BADCRC:
900 case HAMMER_MREC_TYPE_REC:
901 printf("Record obj=%016jx key=%016jx "
903 (uintmax_t)mrec->rec.leaf.base.obj_id,
904 (uintmax_t)mrec->rec.leaf.base.key,
905 mrec->rec.leaf.base.rec_type,
906 mrec->rec.leaf.base.obj_type);
907 if (mrec->head.type ==
908 HAMMER_MREC_TYPE_REC_BADCRC) {
909 printf(" (BAD CRC)");
912 printf(" tids %016jx:%016jx data=%d\n",
913 (uintmax_t)mrec->rec.leaf.base.create_tid,
914 (uintmax_t)mrec->rec.leaf.base.delete_tid,
915 mrec->rec.leaf.data_len);
917 case HAMMER_MREC_TYPE_PASS:
918 printf("Pass obj=%016jx key=%016jx "
920 (uintmax_t)mrec->rec.leaf.base.obj_id,
921 (uintmax_t)mrec->rec.leaf.base.key,
922 mrec->rec.leaf.base.rec_type,
923 mrec->rec.leaf.base.obj_type);
924 printf(" tids %016jx:%016jx data=%d\n",
925 (uintmax_t)mrec->rec.leaf.base.create_tid,
926 (uintmax_t)mrec->rec.leaf.base.delete_tid,
927 mrec->rec.leaf.data_len);
929 case HAMMER_MREC_TYPE_SKIP:
930 printf("Skip obj=%016jx key=%016jx rt=%02x to\n"
931 " obj=%016jx key=%016jx rt=%02x\n",
932 (uintmax_t)mrec->skip.skip_beg.obj_id,
933 (uintmax_t)mrec->skip.skip_beg.key,
934 mrec->skip.skip_beg.rec_type,
935 (uintmax_t)mrec->skip.skip_end.obj_id,
936 (uintmax_t)mrec->skip.skip_end.key,
937 mrec->skip.skip_end.rec_type);
946 * Read and process the termination sync record.
948 mrec = read_mrecord(0, &error, &pickup);
950 (mrec->head.type != HAMMER_MREC_TYPE_SYNC &&
951 mrec->head.type != HAMMER_MREC_TYPE_IDLE)
953 fprintf(stderr, "Mirror-dump: Did not get termination "
959 hammer_cmd_mirror_copy(char **av, int ac, int streaming)
973 signal(SIGPIPE, SIG_IGN);
984 if ((pid1 = fork()) == 0) {
985 signal(SIGPIPE, SIG_DFL);
990 if ((ptr = strchr(av[0], ':')) != NULL) {
998 xav[xac++] = SshPort;
1001 xav[xac++] = "hammer";
1003 switch(VerboseOpt) {
1013 xav[xac++] = "-vvv";
1021 snprintf(tbuf, sizeof(tbuf), "%d", TimeoutOpt);
1026 xav[xac++] = "mirror-read-stream";
1028 xav[xac++] = "mirror-read";
1031 execv("/usr/bin/ssh", (void *)xav);
1033 hammer_cmd_mirror_read(av, 1, streaming);
1043 if ((pid2 = fork()) == 0) {
1044 signal(SIGPIPE, SIG_DFL);
1049 if ((ptr = strchr(av[1], ':')) != NULL) {
1057 xav[xac++] = SshPort;
1060 xav[xac++] = "hammer";
1062 switch(VerboseOpt) {
1072 xav[xac++] = "-vvv";
1079 xav[xac++] = "mirror-write";
1082 execv("/usr/bin/ssh", (void *)xav);
1084 hammer_cmd_mirror_write(av + 1, 1);
1093 while (waitpid(pid1, NULL, 0) <= 0)
1095 while (waitpid(pid2, NULL, 0) <= 0)
1099 * If the link is lost restart
1103 fprintf(stderr, "\nLost Link\n");
1106 sleep(15 + DelayOpt);
1113 * Read and return multiple mrecords
1116 read_mrecords(int fd, char *buf, u_int size, hammer_ioc_mrecord_head_t pickup)
1118 hammer_ioc_mrecord_any_t mrec;
1126 while (size - count >= HAMMER_MREC_HEADSIZE) {
1128 * Cached the record header in case we run out of buffer
1132 if (pickup->signature == 0) {
1133 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1134 i = read(fd, (char *)pickup + n,
1135 HAMMER_MREC_HEADSIZE - n);
1141 if (n != HAMMER_MREC_HEADSIZE) {
1142 fprintf(stderr, "read_mrecords: short read on pipe\n");
1145 if (pickup->signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1146 fprintf(stderr, "read_mrecords: malformed record on pipe, "
1151 if (pickup->rec_size < HAMMER_MREC_HEADSIZE ||
1152 pickup->rec_size > sizeof(*mrec) + HAMMER_XBUFSIZE) {
1153 fprintf(stderr, "read_mrecords: malformed record on pipe, "
1154 "illegal rec_size\n");
1159 * Stop if we have insufficient space for the record and data.
1161 bytes = HAMMER_HEAD_DOALIGN(pickup->rec_size);
1162 if (size - count < bytes)
1166 * Stop if the record type is not a REC, SKIP, or PASS,
1167 * which are the only types the ioctl supports. Other types
1168 * are used only by the userland protocol.
1172 type = pickup->type & HAMMER_MRECF_TYPE_LOMASK;
1173 if (type != HAMMER_MREC_TYPE_PFSD &&
1174 type != HAMMER_MREC_TYPE_REC &&
1175 type != HAMMER_MREC_TYPE_SKIP &&
1176 type != HAMMER_MREC_TYPE_PASS) {
1181 * Read the remainder and clear the pickup signature.
1183 for (n = HAMMER_MREC_HEADSIZE; n < bytes; n += i) {
1184 i = read(fd, buf + count + n, bytes - n);
1189 fprintf(stderr, "read_mrecords: short read on pipe\n");
1193 bcopy(pickup, buf + count, HAMMER_MREC_HEADSIZE);
1194 pickup->signature = 0;
1196 mrec = (void *)(buf + count);
1199 * Validate the completed record
1201 if (mrec->head.rec_crc !=
1202 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1203 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1204 fprintf(stderr, "read_mrecords: malformed record "
1205 "on pipe, bad crc\n");
1210 * If its a B-Tree record validate the data crc.
1212 * NOTE: If the VFS passes us an explicitly errorde mrec
1213 * we just pass it through.
1215 type = mrec->head.type & HAMMER_MRECF_TYPE_MASK;
1217 if (type == HAMMER_MREC_TYPE_REC) {
1218 if (mrec->head.rec_size <
1219 sizeof(mrec->rec) + mrec->rec.leaf.data_len) {
1221 "read_mrecords: malformed record on "
1222 "pipe, illegal element data_len\n");
1225 if (mrec->rec.leaf.data_len &&
1226 mrec->rec.leaf.data_offset &&
1227 hammer_crc_test_leaf(&mrec->rec + 1, &mrec->rec.leaf) == 0) {
1229 "read_mrecords: data_crc did not "
1230 "match data! obj=%016jx key=%016jx\n",
1231 (uintmax_t)mrec->rec.leaf.base.obj_id,
1232 (uintmax_t)mrec->rec.leaf.base.key);
1234 "continuing, but there are problems\n");
1243 * Read and return a single mrecord.
1246 hammer_ioc_mrecord_any_t
1247 read_mrecord(int fdin, int *errorp, hammer_ioc_mrecord_head_t pickup)
1249 hammer_ioc_mrecord_any_t mrec;
1250 struct hammer_ioc_mrecord_head mrechd;
1255 if (pickup && pickup->type != 0) {
1257 pickup->signature = 0;
1259 n = HAMMER_MREC_HEADSIZE;
1262 * Read in the PFSD header from the sender.
1264 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1265 i = read(fdin, (char *)&mrechd + n, HAMMER_MREC_HEADSIZE - n);
1270 *errorp = 0; /* EOF */
1273 if (n != HAMMER_MREC_HEADSIZE) {
1274 fprintf(stderr, "short read of mrecord header\n");
1279 if (mrechd.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1280 fprintf(stderr, "read_mrecord: bad signature\n");
1284 bytes = HAMMER_HEAD_DOALIGN(mrechd.rec_size);
1285 assert(bytes >= sizeof(mrechd));
1286 mrec = malloc(bytes);
1287 mrec->head = mrechd;
1290 i = read(fdin, (char *)mrec + n, bytes - n);
1296 fprintf(stderr, "read_mrecord: short read on payload\n");
1300 if (mrec->head.rec_crc !=
1301 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1302 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1303 fprintf(stderr, "read_mrecord: bad CRC\n");
1313 write_mrecord(int fdout, u_int32_t type, hammer_ioc_mrecord_any_t mrec,
1316 char zbuf[HAMMER_HEAD_ALIGN];
1319 pad = HAMMER_HEAD_DOALIGN(bytes) - bytes;
1321 assert(bytes >= (int)sizeof(mrec->head));
1322 bzero(&mrec->head, sizeof(mrec->head));
1323 mrec->head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
1324 mrec->head.type = type;
1325 mrec->head.rec_size = bytes;
1326 mrec->head.rec_crc = crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1327 bytes - HAMMER_MREC_CRCOFF);
1328 if (write(fdout, mrec, bytes) != bytes) {
1329 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1330 errno, strerror(errno));
1335 if (write(fdout, zbuf, pad) != pad) {
1336 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1337 errno, strerror(errno));
1344 * Generate a mirroring header with the pfs information of the
1345 * originating filesytem.
1348 generate_mrec_header(int fd, int pfs_id,
1349 union hammer_ioc_mrecord_any *mrec_tmp)
1351 struct hammer_ioc_pseudofs_rw pfs;
1353 bzero(&pfs, sizeof(pfs));
1354 bzero(mrec_tmp, sizeof(*mrec_tmp));
1355 pfs.pfs_id = pfs_id;
1356 pfs.ondisk = &mrec_tmp->pfs.pfsd;
1357 pfs.bytes = sizeof(mrec_tmp->pfs.pfsd);
1358 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1359 fprintf(stderr, "Mirror-read: not a HAMMER fs/pseudofs!\n");
1362 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1363 fprintf(stderr, "Mirror-read: HAMMER pfs version mismatch!\n");
1366 mrec_tmp->pfs.version = pfs.version;
1370 * Validate the pfs information from the originating filesystem
1371 * against the target filesystem. shared_uuid must match.
1373 * return -1 if we got a TERM record
1376 validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
1377 struct hammer_ioc_mrecord_head *pickup,
1378 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp)
1380 struct hammer_ioc_pseudofs_rw pfs;
1381 struct hammer_pseudofs_data pfsd;
1382 hammer_ioc_mrecord_any_t mrec;
1386 * Get the PFSD info from the target filesystem.
1388 bzero(&pfs, sizeof(pfs));
1389 bzero(&pfsd, sizeof(pfsd));
1390 pfs.pfs_id = pfs_id;
1392 pfs.bytes = sizeof(pfsd);
1393 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1394 fprintf(stderr, "mirror-write: not a HAMMER fs/pseudofs!\n");
1397 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1398 fprintf(stderr, "mirror-write: HAMMER pfs version mismatch!\n");
1402 mrec = read_mrecord(fdin, &error, pickup);
1405 fprintf(stderr, "validate_mrec_header: short read\n");
1408 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
1413 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
1414 fprintf(stderr, "validate_mrec_header: did not get expected "
1415 "PFSD record type\n");
1418 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
1419 fprintf(stderr, "validate_mrec_header: unexpected payload "
1423 if (mrec->pfs.version != pfs.version) {
1424 fprintf(stderr, "validate_mrec_header: Version mismatch\n");
1429 * Whew. Ok, is the read PFS info compatible with the target?
1431 if (bcmp(&mrec->pfs.pfsd.shared_uuid, &pfsd.shared_uuid,
1432 sizeof(pfsd.shared_uuid)) != 0) {
1434 "mirror-write: source and target have "
1435 "different shared-uuid's!\n");
1439 (pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
1440 fprintf(stderr, "mirror-write: target must be in slave mode\n");
1444 *tid_begp = mrec->pfs.pfsd.sync_beg_tid;
1446 *tid_endp = mrec->pfs.pfsd.sync_end_tid;
1452 update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id)
1454 struct hammer_ioc_pseudofs_rw pfs;
1455 struct hammer_pseudofs_data pfsd;
1457 bzero(&pfs, sizeof(pfs));
1458 bzero(&pfsd, sizeof(pfsd));
1459 pfs.pfs_id = pfs_id;
1461 pfs.bytes = sizeof(pfsd);
1462 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1463 perror("update_pfs_snapshot (read)");
1466 if (pfsd.sync_end_tid != snapshot_tid) {
1467 pfsd.sync_end_tid = snapshot_tid;
1468 if (ioctl(fd, HAMMERIOC_SET_PSEUDOFS, &pfs) != 0) {
1469 perror("update_pfs_snapshot (rewrite)");
1472 if (VerboseOpt >= 2) {
1474 "Mirror-write: Completed, updated snapshot "
1476 (uintmax_t)snapshot_tid);
1483 * Bandwidth-limited write in chunks
1487 writebw(int fd, const void *buf, size_t nbytes,
1488 u_int64_t *bwcount, struct timeval *tv1)
1499 if (*bwcount + nbytes > BandwidthOpt)
1500 n = BandwidthOpt - *bwcount;
1504 r = write(fd, buf, n);
1508 buf = (const char *)buf + r;
1513 if (*bwcount >= BandwidthOpt) {
1514 gettimeofday(&tv2, NULL);
1515 usec = (int)(tv2.tv_sec - tv1->tv_sec) * 1000000 +
1516 (int)(tv2.tv_usec - tv1->tv_usec);
1517 if (usec >= 0 && usec < 1000000)
1518 usleep(1000000 - usec);
1519 gettimeofday(tv1, NULL);
1520 *bwcount -= BandwidthOpt;
1527 * Get a yes or no answer from the terminal. The program may be run as
1528 * part of a two-way pipe so we cannot use stdin for this operation.
1537 fp = fopen("/dev/tty", "r");
1539 fprintf(stderr, "No terminal for response\n");
1543 while (fgets(buf, sizeof(buf), fp) != NULL) {
1544 if (buf[0] == 'y' || buf[0] == 'Y') {
1548 if (buf[0] == 'n' || buf[0] == 'N') {
1552 fprintf(stderr, "Response not understood\n");
1560 mirror_usage(int code)
1563 "hammer mirror-read <filesystem> [begin-tid]\n"
1564 "hammer mirror-read-stream <filesystem> [begin-tid]\n"
1565 "hammer mirror-write <filesystem>\n"
1566 "hammer mirror-dump\n"
1567 "hammer mirror-copy [[user@]host:]<filesystem>"
1568 " [[user@]host:]<filesystem>\n"
1569 "hammer mirror-stream [[user@]host:]<filesystem>"
1570 " [[user@]host:]<filesystem>\n"