bsd.dep.mk: fix race condition with beforedepend
[dragonfly.git] / sbin / hammer / cmd_mirror.c
bloba17fced114e8fe5a6ee849242901ced565582a0f
1 /*
2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
34 * $DragonFly: src/sbin/hammer/cmd_mirror.c,v 1.16 2008/11/09 05:22:56 dillon Exp $
37 #include "hammer.h"
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.
71 void
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);
83 int interrupted = 0;
84 int error;
85 int fd;
86 int n;
87 int didwork;
88 int histogram;
89 int64_t total_bytes;
90 time_t base_t = time(NULL);
91 struct timeval bwtv;
92 u_int64_t bwcount;
94 if (ac == 0 || ac > 2)
95 mirror_usage(1);
96 filesystem = av[0];
98 pickup.signature = 0;
99 pickup.type = 0;
100 histogram = -1;
101 histogram_ary = NULL;
103 again:
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");
112 fflush(stderr);
114 total_bytes = 0;
115 gettimeofday(&bwtv, NULL);
116 bwcount = 0;
119 * Send initial header for the purpose of determining the
120 * shared-uuid.
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
129 * begin TID.
131 mirror.tid_beg = 0;
132 if (TwoWayPipeOpt) {
133 n = validate_mrec_header(fd, 0, 0, pfs.pfs_id, &pickup,
134 NULL, &mirror.tid_beg);
135 if (n < 0) { /* got TERM record */
136 relpfs(fd, &pfs);
137 return;
139 ++mirror.tid_beg;
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;
151 mirror.ubuf = buf;
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.
161 #if 0
162 if (TwoWayPipeOpt && streaming && histogram == 0) {
163 if (mirror.tid_end - mirror.tid_beg > BULK_MINIMUM)
164 histogram = -1;
166 #endif
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) {
174 if (VerboseOpt)
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
199 * in two-way mode.
201 if (ac == 2)
202 mirror.tid_beg = strtoull(av[1], NULL, 0);
204 if (streaming == 0 || VerboseOpt >= 2) {
205 fprintf(stderr,
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");
220 didwork = 0;
221 goto done;
223 didwork = 1;
226 * Write out bulk records
228 mirror.ubuf = buf;
229 mirror.size = SERIALBUF_SIZE;
231 do {
232 mirror.count = 0;
233 mirror.pfs_id = pfs.pfs_id;
234 mirror.shared_uuid = pfs.ondisk->shared_uuid;
235 if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
236 fprintf(stderr, "Mirror-read %s failed: %s\n",
237 filesystem, strerror(errno));
238 exit(1);
240 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
241 fprintf(stderr,
242 "Mirror-read %s fatal error %d\n",
243 filesystem, mirror.head.error);
244 exit(1);
246 if (mirror.count) {
247 if (BandwidthOpt) {
248 n = writebw(1, mirror.ubuf, mirror.count,
249 &bwcount, &bwtv);
250 } else {
251 n = write(1, mirror.ubuf, mirror.count);
253 if (n != mirror.count) {
254 fprintf(stderr, "Mirror-read %s failed: "
255 "short write\n",
256 filesystem);
257 exit(1);
260 total_bytes += mirror.count;
261 if (streaming && VerboseOpt) {
262 fprintf(stderr,
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);
268 fflush(stderr);
270 mirror.key_beg = mirror.key_cur;
273 * Deal with time limit option
275 if (TimeoutOpt &&
276 (unsigned)(time(NULL) - base_t) > (unsigned)TimeoutOpt) {
277 fprintf(stderr,
278 "Mirror-read %s interrupted by timer at"
279 " %016jx\n",
280 filesystem,
281 (uintmax_t)mirror.key_cur.obj_id);
282 interrupted = 1;
283 break;
285 } while (mirror.count != 0);
287 done:
289 * Write out the termination sync record - only if not interrupted
291 if (interrupted == 0) {
292 if (didwork) {
293 write_mrecord(1, HAMMER_MREC_TYPE_SYNC,
294 &mrec_tmp, sizeof(mrec_tmp.sync));
295 } else {
296 write_mrecord(1, HAMMER_MREC_TYPE_IDLE,
297 &mrec_tmp, sizeof(mrec_tmp.sync));
302 * If the -2 option was given (automatic when doing mirror-copy),
303 * a two-way pipe is assumed and we expect a response mrec from
304 * the target.
306 if (TwoWayPipeOpt) {
307 mrec = read_mrecord(0, &error, &pickup);
308 if (mrec == NULL ||
309 mrec->head.type != HAMMER_MREC_TYPE_UPDATE ||
310 mrec->head.rec_size != sizeof(mrec->update)) {
311 fprintf(stderr, "mirror_read: Did not get final "
312 "acknowledgement packet from target\n");
313 exit(1);
315 if (interrupted) {
316 if (CyclePath) {
317 hammer_set_cycle(&mirror.key_cur, mirror.tid_beg);
318 fprintf(stderr, "Cyclefile %s updated for "
319 "continuation\n", CyclePath);
321 } else {
322 sync_tid = mrec->update.tid;
323 if (CyclePath) {
324 hammer_key_beg_init(&mirror.key_beg);
325 hammer_set_cycle(&mirror.key_beg, sync_tid);
326 fprintf(stderr,
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);
339 time_t t2;
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
346 * scratch.
348 if (TwoWayPipeOpt && streaming && histogram > 0) {
349 if (VerboseOpt)
350 fprintf(stderr, " (bulk incremental)");
351 goto again;
354 if (VerboseOpt) {
355 fprintf(stderr, " W");
356 fflush(stderr);
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));
362 } else {
363 t2 = time(NULL) - t1;
364 if (t2 >= 0 && t2 < DelayOpt) {
365 if (VerboseOpt) {
366 fprintf(stderr, "\bD");
367 fflush(stderr);
369 sleep(DelayOpt - t2);
371 if (VerboseOpt) {
372 fprintf(stderr, "\b ");
373 fflush(stderr);
375 relpfs(fd, &pfs);
376 goto again;
379 write_mrecord(1, HAMMER_MREC_TYPE_TERM,
380 &mrec_tmp, sizeof(mrec_tmp.sync));
381 relpfs(fd, &pfs);
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.
399 static int
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;
408 int i;
410 mirror = *mirror_base;
411 tid_beg = mirror.tid_beg;
412 tid_end = mirror.tid_end;
414 if (*histogram_ary)
415 free(*histogram_ary);
416 if (tid_beg + BULK_MINIMUM >= tid_end)
417 return(0);
419 if (VerboseOpt)
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;
423 mirror.count = 0;
424 mirror.tid_beg = tid_beg;
425 mirror.tid_end = tid_half;
427 if (VerboseOpt > 1) {
428 fprintf(stderr, "RangeTest %016jx/%016jx - %016jx (%jd) ",
429 (uintmax_t)tid_beg,
430 (uintmax_t)tid_end,
431 (uintmax_t)tid_half,
432 (intmax_t)(tid_half - tid_beg));
434 fflush(stderr);
435 if (ioctl(fd, HAMMERIOC_MIRROR_READ, &mirror) < 0) {
436 fprintf(stderr, "Mirror-read %s failed: %s\n",
437 filesystem, strerror(errno));
438 exit(1);
440 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
441 fprintf(stderr,
442 "Mirror-read %s fatal error %d\n",
443 filesystem, mirror.head.error);
444 exit(1);
446 if (VerboseOpt > 1)
447 fprintf(stderr, "%d\n", mirror.count);
448 if (mirror.count > SERIALBUF_SIZE / 2) {
449 tid_end = tid_half;
450 } else {
451 tid_beg = tid_half;
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;
467 return(20);
470 static void
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);
477 } else {
478 fprintf(stderr, "PFS slave %s does not exist.\n"
479 "Do you want to create a new slave PFS? (yes|no) ",
480 filesystem);
481 fflush(stderr);
482 if (getyn() != 1) {
483 fprintf(stderr, "Aborting operation\n");
484 exit(1);
488 u_int32_t status;
489 char *shared_uuid = NULL;
490 uuid_to_string(s_uuid, &shared_uuid, &status);
492 char *cmd = NULL;
493 asprintf(&cmd, "/sbin/hammer pfs-slave '%s' shared-uuid=%s 1>&2",
494 filesystem, shared_uuid);
495 free(shared_uuid);
497 if (cmd == NULL) {
498 fprintf(stderr, "Failed to alloc memory\n");
499 exit(1);
501 if (system(cmd) != 0) {
502 fprintf(stderr, "Failed to create PFS\n");
504 free(cmd);
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
521 * by the source.
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.
532 void
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;
543 struct stat st;
544 int error;
545 int fd;
546 int n;
548 if (ac != 1)
549 mirror_usage(1);
550 filesystem = av[0];
552 pickup.signature = 0;
553 pickup.type = 0;
555 again:
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);
565 if (mrec == NULL) {
566 if (error == 0)
567 fprintf(stderr, "validate_mrec_header: short read\n");
568 exit(1);
571 * Validate packet
573 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
574 return;
576 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
577 fprintf(stderr, "validate_mrec_header: did not get expected "
578 "PFSD record type\n");
579 exit(1);
581 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
582 fprintf(stderr, "validate_mrec_header: unexpected payload "
583 "size\n");
584 exit(1);
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);
593 free(mrec);
594 mrec = NULL;
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.
603 mirror.tid_beg = 0;
604 if (TwoWayPipeOpt) {
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 */
620 relpfs(fd, &pfs);
621 return;
624 mirror.ubuf = buf;
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.
633 for (;;) {
634 mirror.count = 0;
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)
639 break;
640 if (ioctl(fd, HAMMERIOC_MIRROR_WRITE, &mirror) < 0) {
641 fprintf(stderr, "Mirror-write %s failed: %s\n",
642 filesystem, strerror(errno));
643 exit(1);
645 if (mirror.head.flags & HAMMER_IOC_HEAD_ERROR) {
646 fprintf(stderr,
647 "Mirror-write %s fatal error %d\n",
648 filesystem, mirror.head.error);
649 exit(1);
651 #if 0
652 if (mirror.head.flags & HAMMER_IOC_HEAD_INTR) {
653 fprintf(stderr,
654 "Mirror-write %s interrupted by timer at"
655 " %016llx\n",
656 filesystem,
657 mirror.key_cur.obj_id);
658 exit(0);
660 #endif
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");
670 free(mrec);
671 return;
674 if (mrec == NULL ||
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);
681 exit(1);
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",
697 filesystem);
701 free(mrec);
702 mrec = NULL;
705 * Report back to the originator.
707 if (TwoWayPipeOpt) {
708 mrec_tmp.update.tid = mirror.tid_end;
709 write_mrecord(1, HAMMER_MREC_TYPE_UPDATE,
710 &mrec_tmp, sizeof(mrec_tmp.update));
711 } else {
712 printf("Source can update synctid to 0x%016jx\n",
713 (uintmax_t)mirror.tid_end);
715 relpfs(fd, &pfs);
716 goto again;
719 void
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;
725 int error;
726 int size;
727 int offset;
728 int bytes;
731 * Read and process the PFS header
733 pickup.signature = 0;
734 pickup.type = 0;
736 mrec = read_mrecord(0, &error, &pickup);
739 * Read and process bulk records
741 for (;;) {
742 size = read_mrecords(0, buf, SERIALBUF_SIZE, &pickup);
743 if (size <= 0)
744 break;
745 offset = 0;
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");
751 exit(1);
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 "
758 "rt=%02x ot=%02x",
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)");
767 printf("\n");
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);
772 break;
773 case HAMMER_MREC_TYPE_PASS:
774 printf("Pass obj=%016jx key=%016jx "
775 "rt=%02x ot=%02x\n",
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);
784 break;
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);
794 default:
795 break;
797 offset += bytes;
802 * Read and process the termination sync record.
804 mrec = read_mrecord(0, &error, &pickup);
805 if (mrec == NULL ||
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 "
810 "sync record\n");
814 void
815 hammer_cmd_mirror_copy(char **av, int ac, int streaming)
817 pid_t pid1;
818 pid_t pid2;
819 int fds[2];
820 const char *xav[16];
821 char tbuf[16];
822 char *ptr;
823 int xac;
825 if (ac != 2)
826 mirror_usage(1);
828 TwoWayPipeOpt = 1;
829 signal(SIGPIPE, SIG_IGN);
831 again:
832 if (pipe(fds) < 0) {
833 perror("pipe");
834 exit(1);
838 * Source
840 if ((pid1 = fork()) == 0) {
841 signal(SIGPIPE, SIG_DFL);
842 dup2(fds[0], 0);
843 dup2(fds[0], 1);
844 close(fds[0]);
845 close(fds[1]);
846 if ((ptr = strchr(av[0], ':')) != NULL) {
847 *ptr++ = 0;
848 xac = 0;
849 xav[xac++] = "ssh";
850 xav[xac++] = av[0];
851 xav[xac++] = "hammer";
853 switch(VerboseOpt) {
854 case 0:
855 break;
856 case 1:
857 xav[xac++] = "-v";
858 break;
859 case 2:
860 xav[xac++] = "-vv";
861 break;
862 default:
863 xav[xac++] = "-vvv";
864 break;
866 if (ForceYesOpt) {
867 xav[xac++] = "-y";
869 xav[xac++] = "-2";
870 if (TimeoutOpt) {
871 snprintf(tbuf, sizeof(tbuf), "%d", TimeoutOpt);
872 xav[xac++] = "-t";
873 xav[xac++] = tbuf;
875 if (streaming)
876 xav[xac++] = "mirror-read-stream";
877 else
878 xav[xac++] = "mirror-read";
879 xav[xac++] = ptr;
880 xav[xac++] = NULL;
881 execv("/usr/bin/ssh", (void *)xav);
882 } else {
883 hammer_cmd_mirror_read(av, 1, streaming);
884 fflush(stdout);
885 fflush(stderr);
887 _exit(1);
891 * Target
893 if ((pid2 = fork()) == 0) {
894 signal(SIGPIPE, SIG_DFL);
895 dup2(fds[1], 0);
896 dup2(fds[1], 1);
897 close(fds[0]);
898 close(fds[1]);
899 if ((ptr = strchr(av[1], ':')) != NULL) {
900 *ptr++ = 0;
901 xac = 0;
902 xav[xac++] = "ssh";
903 xav[xac++] = av[1];
904 xav[xac++] = "hammer";
906 switch(VerboseOpt) {
907 case 0:
908 break;
909 case 1:
910 xav[xac++] = "-v";
911 break;
912 case 2:
913 xav[xac++] = "-vv";
914 break;
915 default:
916 xav[xac++] = "-vvv";
917 break;
919 if (ForceYesOpt) {
920 xav[xac++] = "-y";
922 xav[xac++] = "-2";
923 xav[xac++] = "mirror-write";
924 xav[xac++] = ptr;
925 xav[xac++] = NULL;
926 execv("/usr/bin/ssh", (void *)xav);
927 } else {
928 hammer_cmd_mirror_write(av + 1, 1);
929 fflush(stdout);
930 fflush(stderr);
932 _exit(1);
934 close(fds[0]);
935 close(fds[1]);
937 while (waitpid(pid1, NULL, 0) <= 0)
939 while (waitpid(pid2, NULL, 0) <= 0)
943 * If the link is lost restart
945 if (streaming) {
946 if (VerboseOpt) {
947 fprintf(stderr, "\nLost Link\n");
948 fflush(stderr);
950 sleep(15 + DelayOpt);
951 goto again;
957 * Read and return multiple mrecords
959 static int
960 read_mrecords(int fd, char *buf, u_int size, hammer_ioc_mrecord_head_t pickup)
962 hammer_ioc_mrecord_any_t mrec;
963 u_int count;
964 size_t n;
965 size_t i;
966 size_t bytes;
967 int type;
969 count = 0;
970 while (size - count >= HAMMER_MREC_HEADSIZE) {
972 * Cached the record header in case we run out of buffer
973 * space.
975 fflush(stdout);
976 if (pickup->signature == 0) {
977 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
978 i = read(fd, (char *)pickup + n,
979 HAMMER_MREC_HEADSIZE - n);
980 if (i <= 0)
981 break;
983 if (n == 0)
984 break;
985 if (n != HAMMER_MREC_HEADSIZE) {
986 fprintf(stderr, "read_mrecords: short read on pipe\n");
987 exit(1);
989 if (pickup->signature != HAMMER_IOC_MIRROR_SIGNATURE) {
990 fprintf(stderr, "read_mrecords: malformed record on pipe, "
991 "bad signature\n");
992 exit(1);
995 if (pickup->rec_size < HAMMER_MREC_HEADSIZE ||
996 pickup->rec_size > sizeof(*mrec) + HAMMER_XBUFSIZE) {
997 fprintf(stderr, "read_mrecords: malformed record on pipe, "
998 "illegal rec_size\n");
999 exit(1);
1003 * Stop if we have insufficient space for the record and data.
1005 bytes = HAMMER_HEAD_DOALIGN(pickup->rec_size);
1006 if (size - count < bytes)
1007 break;
1010 * Stop if the record type is not a REC, SKIP, or PASS,
1011 * which are the only types the ioctl supports. Other types
1012 * are used only by the userland protocol.
1014 * Ignore all flags.
1016 type = pickup->type & HAMMER_MRECF_TYPE_LOMASK;
1017 if (type != HAMMER_MREC_TYPE_PFSD &&
1018 type != HAMMER_MREC_TYPE_REC &&
1019 type != HAMMER_MREC_TYPE_SKIP &&
1020 type != HAMMER_MREC_TYPE_PASS) {
1021 break;
1025 * Read the remainder and clear the pickup signature.
1027 for (n = HAMMER_MREC_HEADSIZE; n < bytes; n += i) {
1028 i = read(fd, buf + count + n, bytes - n);
1029 if (i <= 0)
1030 break;
1032 if (n != bytes) {
1033 fprintf(stderr, "read_mrecords: short read on pipe\n");
1034 exit(1);
1037 bcopy(pickup, buf + count, HAMMER_MREC_HEADSIZE);
1038 pickup->signature = 0;
1039 pickup->type = 0;
1040 mrec = (void *)(buf + count);
1043 * Validate the completed record
1045 if (mrec->head.rec_crc !=
1046 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1047 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1048 fprintf(stderr, "read_mrecords: malformed record "
1049 "on pipe, bad crc\n");
1050 exit(1);
1054 * If its a B-Tree record validate the data crc.
1056 * NOTE: If the VFS passes us an explicitly errorde mrec
1057 * we just pass it through.
1059 type = mrec->head.type & HAMMER_MRECF_TYPE_MASK;
1061 if (type == HAMMER_MREC_TYPE_REC) {
1062 if (mrec->head.rec_size <
1063 sizeof(mrec->rec) + mrec->rec.leaf.data_len) {
1064 fprintf(stderr,
1065 "read_mrecords: malformed record on "
1066 "pipe, illegal element data_len\n");
1067 exit(1);
1069 if (mrec->rec.leaf.data_len &&
1070 mrec->rec.leaf.data_offset &&
1071 hammer_crc_test_leaf(&mrec->rec + 1, &mrec->rec.leaf) == 0) {
1072 fprintf(stderr,
1073 "read_mrecords: data_crc did not "
1074 "match data! obj=%016jx key=%016jx\n",
1075 (uintmax_t)mrec->rec.leaf.base.obj_id,
1076 (uintmax_t)mrec->rec.leaf.base.key);
1077 fprintf(stderr,
1078 "continuing, but there are problems\n");
1081 count += bytes;
1083 return(count);
1087 * Read and return a single mrecord.
1089 static
1090 hammer_ioc_mrecord_any_t
1091 read_mrecord(int fdin, int *errorp, hammer_ioc_mrecord_head_t pickup)
1093 hammer_ioc_mrecord_any_t mrec;
1094 struct hammer_ioc_mrecord_head mrechd;
1095 size_t bytes;
1096 size_t n;
1097 size_t i;
1099 if (pickup && pickup->type != 0) {
1100 mrechd = *pickup;
1101 pickup->signature = 0;
1102 pickup->type = 0;
1103 n = HAMMER_MREC_HEADSIZE;
1104 } else {
1106 * Read in the PFSD header from the sender.
1108 for (n = 0; n < HAMMER_MREC_HEADSIZE; n += i) {
1109 i = read(fdin, (char *)&mrechd + n, HAMMER_MREC_HEADSIZE - n);
1110 if (i <= 0)
1111 break;
1113 if (n == 0) {
1114 *errorp = 0; /* EOF */
1115 return(NULL);
1117 if (n != HAMMER_MREC_HEADSIZE) {
1118 fprintf(stderr, "short read of mrecord header\n");
1119 *errorp = EPIPE;
1120 return(NULL);
1123 if (mrechd.signature != HAMMER_IOC_MIRROR_SIGNATURE) {
1124 fprintf(stderr, "read_mrecord: bad signature\n");
1125 *errorp = EINVAL;
1126 return(NULL);
1128 bytes = HAMMER_HEAD_DOALIGN(mrechd.rec_size);
1129 assert(bytes >= sizeof(mrechd));
1130 mrec = malloc(bytes);
1131 mrec->head = mrechd;
1133 while (n < bytes) {
1134 i = read(fdin, (char *)mrec + n, bytes - n);
1135 if (i <= 0)
1136 break;
1137 n += i;
1139 if (n != bytes) {
1140 fprintf(stderr, "read_mrecord: short read on payload\n");
1141 *errorp = EPIPE;
1142 return(NULL);
1144 if (mrec->head.rec_crc !=
1145 crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1146 mrec->head.rec_size - HAMMER_MREC_CRCOFF)) {
1147 fprintf(stderr, "read_mrecord: bad CRC\n");
1148 *errorp = EINVAL;
1149 return(NULL);
1151 *errorp = 0;
1152 return(mrec);
1155 static
1156 void
1157 write_mrecord(int fdout, u_int32_t type, hammer_ioc_mrecord_any_t mrec,
1158 int bytes)
1160 char zbuf[HAMMER_HEAD_ALIGN];
1161 int pad;
1163 pad = HAMMER_HEAD_DOALIGN(bytes) - bytes;
1165 assert(bytes >= (int)sizeof(mrec->head));
1166 bzero(&mrec->head, sizeof(mrec->head));
1167 mrec->head.signature = HAMMER_IOC_MIRROR_SIGNATURE;
1168 mrec->head.type = type;
1169 mrec->head.rec_size = bytes;
1170 mrec->head.rec_crc = crc32((char *)mrec + HAMMER_MREC_CRCOFF,
1171 bytes - HAMMER_MREC_CRCOFF);
1172 if (write(fdout, mrec, bytes) != bytes) {
1173 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1174 errno, strerror(errno));
1175 exit(1);
1177 if (pad) {
1178 bzero(zbuf, pad);
1179 if (write(fdout, zbuf, pad) != pad) {
1180 fprintf(stderr, "write_mrecord: error %d (%s)\n",
1181 errno, strerror(errno));
1182 exit(1);
1188 * Generate a mirroring header with the pfs information of the
1189 * originating filesytem.
1191 static void
1192 generate_mrec_header(int fd, int pfs_id,
1193 union hammer_ioc_mrecord_any *mrec_tmp)
1195 struct hammer_ioc_pseudofs_rw pfs;
1197 bzero(&pfs, sizeof(pfs));
1198 bzero(mrec_tmp, sizeof(*mrec_tmp));
1199 pfs.pfs_id = pfs_id;
1200 pfs.ondisk = &mrec_tmp->pfs.pfsd;
1201 pfs.bytes = sizeof(mrec_tmp->pfs.pfsd);
1202 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1203 fprintf(stderr, "Mirror-read: not a HAMMER fs/pseudofs!\n");
1204 exit(1);
1206 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1207 fprintf(stderr, "Mirror-read: HAMMER pfs version mismatch!\n");
1208 exit(1);
1210 mrec_tmp->pfs.version = pfs.version;
1214 * Validate the pfs information from the originating filesystem
1215 * against the target filesystem. shared_uuid must match.
1217 * return -1 if we got a TERM record
1219 static int
1220 validate_mrec_header(int fd, int fdin, int is_target, int pfs_id,
1221 struct hammer_ioc_mrecord_head *pickup,
1222 hammer_tid_t *tid_begp, hammer_tid_t *tid_endp)
1224 struct hammer_ioc_pseudofs_rw pfs;
1225 struct hammer_pseudofs_data pfsd;
1226 hammer_ioc_mrecord_any_t mrec;
1227 int error;
1230 * Get the PFSD info from the target filesystem.
1232 bzero(&pfs, sizeof(pfs));
1233 bzero(&pfsd, sizeof(pfsd));
1234 pfs.pfs_id = pfs_id;
1235 pfs.ondisk = &pfsd;
1236 pfs.bytes = sizeof(pfsd);
1237 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1238 fprintf(stderr, "mirror-write: not a HAMMER fs/pseudofs!\n");
1239 exit(1);
1241 if (pfs.version != HAMMER_IOC_PSEUDOFS_VERSION) {
1242 fprintf(stderr, "mirror-write: HAMMER pfs version mismatch!\n");
1243 exit(1);
1246 mrec = read_mrecord(fdin, &error, pickup);
1247 if (mrec == NULL) {
1248 if (error == 0)
1249 fprintf(stderr, "validate_mrec_header: short read\n");
1250 exit(1);
1252 if (mrec->head.type == HAMMER_MREC_TYPE_TERM) {
1253 free(mrec);
1254 return(-1);
1257 if (mrec->head.type != HAMMER_MREC_TYPE_PFSD) {
1258 fprintf(stderr, "validate_mrec_header: did not get expected "
1259 "PFSD record type\n");
1260 exit(1);
1262 if (mrec->head.rec_size != sizeof(mrec->pfs)) {
1263 fprintf(stderr, "validate_mrec_header: unexpected payload "
1264 "size\n");
1265 exit(1);
1267 if (mrec->pfs.version != pfs.version) {
1268 fprintf(stderr, "validate_mrec_header: Version mismatch\n");
1269 exit(1);
1273 * Whew. Ok, is the read PFS info compatible with the target?
1275 if (bcmp(&mrec->pfs.pfsd.shared_uuid, &pfsd.shared_uuid,
1276 sizeof(pfsd.shared_uuid)) != 0) {
1277 fprintf(stderr,
1278 "mirror-write: source and target have "
1279 "different shared-uuid's!\n");
1280 exit(1);
1282 if (is_target &&
1283 (pfsd.mirror_flags & HAMMER_PFSD_SLAVE) == 0) {
1284 fprintf(stderr, "mirror-write: target must be in slave mode\n");
1285 exit(1);
1287 if (tid_begp)
1288 *tid_begp = mrec->pfs.pfsd.sync_beg_tid;
1289 if (tid_endp)
1290 *tid_endp = mrec->pfs.pfsd.sync_end_tid;
1291 free(mrec);
1292 return(0);
1295 static void
1296 update_pfs_snapshot(int fd, hammer_tid_t snapshot_tid, int pfs_id)
1298 struct hammer_ioc_pseudofs_rw pfs;
1299 struct hammer_pseudofs_data pfsd;
1301 bzero(&pfs, sizeof(pfs));
1302 bzero(&pfsd, sizeof(pfsd));
1303 pfs.pfs_id = pfs_id;
1304 pfs.ondisk = &pfsd;
1305 pfs.bytes = sizeof(pfsd);
1306 if (ioctl(fd, HAMMERIOC_GET_PSEUDOFS, &pfs) != 0) {
1307 perror("update_pfs_snapshot (read)");
1308 exit(1);
1310 if (pfsd.sync_end_tid != snapshot_tid) {
1311 pfsd.sync_end_tid = snapshot_tid;
1312 if (ioctl(fd, HAMMERIOC_SET_PSEUDOFS, &pfs) != 0) {
1313 perror("update_pfs_snapshot (rewrite)");
1314 exit(1);
1316 if (VerboseOpt >= 2) {
1317 fprintf(stderr,
1318 "Mirror-write: Completed, updated snapshot "
1319 "to %016jx\n",
1320 (uintmax_t)snapshot_tid);
1326 * Bandwidth-limited write in chunks
1328 static
1329 ssize_t
1330 writebw(int fd, const void *buf, size_t nbytes,
1331 u_int64_t *bwcount, struct timeval *tv1)
1333 struct timeval tv2;
1334 size_t n;
1335 ssize_t r;
1336 ssize_t a;
1337 int usec;
1339 a = 0;
1340 r = 0;
1341 while (nbytes) {
1342 if (*bwcount + nbytes > BandwidthOpt)
1343 n = BandwidthOpt - *bwcount;
1344 else
1345 n = nbytes;
1346 if (n)
1347 r = write(fd, buf, n);
1348 if (r >= 0) {
1349 a += r;
1350 nbytes -= r;
1351 buf = (const char *)buf + r;
1353 if ((size_t)r != n)
1354 break;
1355 *bwcount += n;
1356 if (*bwcount >= BandwidthOpt) {
1357 gettimeofday(&tv2, NULL);
1358 usec = (int)(tv2.tv_sec - tv1->tv_sec) * 1000000 +
1359 (int)(tv2.tv_usec - tv1->tv_usec);
1360 if (usec >= 0 && usec < 1000000)
1361 usleep(1000000 - usec);
1362 gettimeofday(tv1, NULL);
1363 *bwcount -= BandwidthOpt;
1366 return(a ? a : r);
1370 * Get a yes or no answer from the terminal. The program may be run as
1371 * part of a two-way pipe so we cannot use stdin for this operation.
1373 static int
1374 getyn(void)
1376 char buf[256];
1377 FILE *fp;
1378 int result;
1380 fp = fopen("/dev/tty", "r");
1381 if (fp == NULL) {
1382 fprintf(stderr, "No terminal for response\n");
1383 return(-1);
1385 result = -1;
1386 while (fgets(buf, sizeof(buf), fp) != NULL) {
1387 if (buf[0] == 'y' || buf[0] == 'Y') {
1388 result = 1;
1389 break;
1391 if (buf[0] == 'n' || buf[0] == 'N') {
1392 result = 0;
1393 break;
1395 fprintf(stderr, "Response not understood\n");
1396 break;
1398 fclose(fp);
1399 return(result);
1402 static void
1403 mirror_usage(int code)
1405 fprintf(stderr,
1406 "hammer mirror-read <filesystem> [begin-tid]\n"
1407 "hammer mirror-read-stream <filesystem> [begin-tid]\n"
1408 "hammer mirror-write <filesystem>\n"
1409 "hammer mirror-dump\n"
1410 "hammer mirror-copy [[user@]host:]<filesystem>"
1411 " [[user@]host:]<filesystem>\n"
1412 "hammer mirror-stream [[user@]host:]<filesystem>"
1413 " [[user@]host:]<filesystem>\n"
1415 exit(code);