zdb: add -B option to generate backup stream

This is more-or-less like `zfs send`, but specifying the snapshot by its
objset id for situations where it can't be referenced any other way.

Sponsored-By: Klara, Inc.
Reviewed-by: Tino Reichardt <milky-zfs@mcmilk.de>
Reviewed-by: WHR <msl0000023508@gmail.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes #14642

cmd/zdb/zdb.c

@@ -33,6 +33,7 @@
  *     under sponsorship from the FreeBSD Foundation.
  * Copyright (c) 2021 Allan Jude
  * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
+ * Copyright (c) 2023, Klara Inc.
  */
 
 #include <stdio.h>
@@ -789,6 +790,9 @@ usage(void)
 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
 	    "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
+	    "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
+	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
+	    "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
 	    "\t%s [-v] <bookmark>\n"
 	    "\t%s -C [-A] [-U <cache>]\n"
 	    "\t%s -l [-Aqu] <device>\n"
@@ -802,7 +806,7 @@ usage(void)
 	    "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
 	    "<poolname>\n\n",
 	    cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
-	    cmdname, cmdname, cmdname, cmdname);
+	    cmdname, cmdname, cmdname, cmdname, cmdname);
 
 	(void) fprintf(stderr, "    Dataset name must include at least one "
 	    "separator character '/' or '@'\n");
@@ -825,6 +829,8 @@ usage(void)
 	(void) fprintf(stderr, "    Options to control amount of output:\n");
 	(void) fprintf(stderr, "        -b --block-stats             "
 	    "block statistics\n");
+	(void) fprintf(stderr, "        -B --backup                  "
+	    "backup stream\n");
 	(void) fprintf(stderr, "        -c --checksum                "
 	    "checksum all metadata (twice for all data) blocks\n");
 	(void) fprintf(stderr, "        -C --config                  "
@@ -4875,6 +4881,81 @@ dump_path(char *ds, char *path, uint64_t *retobj)
 	return (err);
 }
 
+static int
+dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
+{
+	const char *p = (const char *)buf;
+	ssize_t nwritten;
+
+	(void) os;
+	(void) arg;
+
+	/* Write the data out, handling short writes and signals. */
+	while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
+		if (nwritten < 0) {
+			if (errno == EINTR)
+				continue;
+			return (errno);
+		}
+		p += nwritten;
+		len -= nwritten;
+	}
+
+	return (0);
+}
+
+static void
+dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
+{
+	boolean_t embed = B_FALSE;
+	boolean_t large_block = B_FALSE;
+	boolean_t compress = B_FALSE;
+	boolean_t raw = B_FALSE;
+
+	const char *c;
+	for (c = flagstr; c != NULL && *c != '\0'; c++) {
+		switch (*c) {
+			case 'e':
+				embed = B_TRUE;
+				break;
+			case 'L':
+				large_block = B_TRUE;
+				break;
+			case 'c':
+				compress = B_TRUE;
+				break;
+			case 'w':
+				raw = B_TRUE;
+				break;
+			default:
+				fprintf(stderr, "dump_backup: invalid flag "
+				    "'%c'\n", *c);
+				return;
+		}
+	}
+
+	if (isatty(STDOUT_FILENO)) {
+		fprintf(stderr, "dump_backup: stream cannot be written "
+		    "to a terminal\n");
+		return;
+	}
+
+	offset_t off = 0;
+	dmu_send_outparams_t out = {
+	    .dso_outfunc = dump_backup_bytes,
+	    .dso_dryrun  = B_FALSE,
+	};
+
+	int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
+	    large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
+	    &off, &out);
+	if (err != 0) {
+		fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
+		    strerror(err));
+		return;
+	}
+}
+
 static int
 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
 {
@@ -8695,6 +8776,7 @@ main(int argc, char **argv)
 	struct option long_options[] = {
 		{"ignore-assertions",	no_argument,		NULL, 'A'},
 		{"block-stats",		no_argument,		NULL, 'b'},
+		{"backup",		no_argument,		NULL, 'B'},
 		{"checksum",		no_argument,		NULL, 'c'},
 		{"config",		no_argument,		NULL, 'C'},
 		{"datasets",		no_argument,		NULL, 'd'},
@@ -8736,10 +8818,11 @@ main(int argc, char **argv)
 	};
 
 	while ((c = getopt_long(argc, argv,
-	    "AbcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:uU:vVx:XYyZ",
+	    "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:uU:vVx:XYyZ",
 	    long_options, NULL)) != -1) {
 		switch (c) {
 		case 'b':
+		case 'B':
 		case 'c':
 		case 'C':
 		case 'd':
@@ -8887,7 +8970,7 @@ main(int argc, char **argv)
 		verbose = MAX(verbose, 1);
 
 	for (c = 0; c < 256; c++) {
-		if (dump_all && strchr("AeEFkKlLNOPrRSXy", c) == NULL)
+		if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
 			dump_opt[c] = 1;
 		if (dump_opt[c])
 			dump_opt[c] += verbose;
@@ -9073,7 +9156,8 @@ main(int argc, char **argv)
 				    checkpoint_pool, error);
 			}
 
-		} else if (target_is_spa || dump_opt['R'] || objset_id == 0) {
+		} else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
+		    objset_id == 0) {
 			zdb_set_skip_mmp(target);
 			error = spa_open_rewind(target, &spa, FTAG, policy,
 			    NULL);
@@ -9209,7 +9293,10 @@ retry_lookup:
 					    strerror(errno));
 			}
 		}
-		if (os != NULL) {
+		if (dump_opt['B']) {
+			dump_backup(target, objset_id,
+			    argc > 0 ? argv[0] : NULL);
+		} else if (os != NULL) {
 			dump_objset(os);
 		} else if (zopt_object_args > 0 && !dump_opt['m']) {
 			dump_objset(spa->spa_meta_objset);

man/man8/zdb.8

@@ -14,7 +14,7 @@
 .\" Copyright (c) 2017 Lawrence Livermore National Security, LLC.
 .\" Copyright (c) 2017 Intel Corporation.
 .\"
-.Dd October 7, 2020
+.Dd June 4, 2023
 .Dt ZDB 8
 .Os
 .
@@ -41,6 +41,13 @@
 .Ar poolname Ns Op Ar / Ns Ar dataset Ns | Ns Ar objset-ID
 .Op Ar object Ns | Ns Ar range Ns …
 .Nm
+.Fl B
+.Op Fl e Oo Fl V Oc Oo Fl p Ar path Oc Ns …
+.Op Fl U Ar cache
+.Op Fl K Ar key
+.Ar poolname Ns Ar / Ns Ar objset-ID
+.Op Ar backup-flags
+.Nm
 .Fl C
 .Op Fl A
 .Op Fl U Ar cache
@@ -123,6 +130,22 @@ Display options:
 Display statistics regarding the number, size
 .Pq logical, physical and allocated
 and deduplication of blocks.
+.It Fl B , -backup
+Generate a backup stream, similar to
+.Nm zfs Cm send ,
+but for the numeric objset ID, and without opening the dataset.
+This can be useful in recovery scenarios if dataset metadata has become
+corrupted but the dataset itself is readable.
+The optional
+.Ar flags
+argument is a string of one or more of the letters
+.Sy e ,
+.Sy L ,
+.Sy c ,
+and
+.Sy w ,
+which correspond to the same flags in
+.Xr zfs-send 8 .
 .It Fl c , -checksum
 Verify the checksum of all metadata blocks while printing block statistics
 .Po see

module/zfs/dmu_send.c

@@ -1955,7 +1955,7 @@ setup_featureflags(struct dmu_send_params *dspp, objset_t *os,
 {
 	dsl_dataset_t *to_ds = dspp->to_ds;
 	dsl_pool_t *dp = dspp->dp;
-#ifdef _KERNEL
+
 	if (dmu_objset_type(os) == DMU_OST_ZFS) {
 		uint64_t version;
 		if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0)
@@ -1964,7 +1964,6 @@ setup_featureflags(struct dmu_send_params *dspp, objset_t *os,
 		if (version >= ZPL_VERSION_SA)
 			*featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
 	}
-#endif
 
 	/* raw sends imply large_block_ok */
 	if ((dspp->rawok || dspp->large_block_ok) &&

tests/runfiles/common.run

@@ -128,7 +128,7 @@ tests = ['zdb_002_pos', 'zdb_003_pos', 'zdb_004_pos', 'zdb_005_pos',
     'zdb_block_size_histogram', 'zdb_checksum', 'zdb_decompress',
     'zdb_display_block', 'zdb_encrypted', 'zdb_label_checksum',
     'zdb_object_range_neg', 'zdb_object_range_pos', 'zdb_objset_id',
-    'zdb_decompress_zstd', 'zdb_recover', 'zdb_recover_2']
+    'zdb_decompress_zstd', 'zdb_recover', 'zdb_recover_2', 'zdb_backup']
 pre =
 post =
 tags = ['functional', 'cli_root', 'zdb']

tests/zfs-tests/tests/Makefile.am

@@ -572,6 +572,7 @@ nobase_dist_datadir_zfs_tests_tests_SCRIPTS += \
 	functional/cli_root/zdb/zdb_006_pos.ksh \
 	functional/cli_root/zdb/zdb_args_neg.ksh \
 	functional/cli_root/zdb/zdb_args_pos.ksh \
+	functional/cli_root/zdb/zdb_backup.ksh \
 	functional/cli_root/zdb/zdb_block_size_histogram.ksh \
 	functional/cli_root/zdb/zdb_checksum.ksh \
 	functional/cli_root/zdb/zdb_decompress.ksh \

tests/zfs-tests/tests/functional/cli_root/zdb/zdb_backup.ksh

@@ -0,0 +1,55 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2023, Klara Inc.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+write_count=8
+blksize=131072
+
+tmpfile=$TEST_BASE_DIR/tmpfile
+
+function cleanup
+{
+	datasetexists $TESTPOOL && destroy_pool $TESTPOOL
+	rm $tmpfile.1 $tmpfile.2
+}
+
+log_onexit cleanup
+
+log_assert "Verify that zfs send and zdb -B produce the same stream"
+
+verify_runnable "global"
+verify_disk_count "$DISKS" 2
+
+default_mirror_setup_noexit $DISKS
+file_write -o create -w -f $TESTDIR/file -b $blksize -c $write_count
+
+snap=$TESTPOOL/$TESTFS@snap
+log_must zfs snapshot $snap
+typeset -i objsetid=$(zfs get -Ho value objsetid $snap)
+
+sync_pool $TESTPOOL
+
+log_must eval "zfs send -ecL $snap > $tmpfile.1"
+log_must eval "zdb -B $TESTPOOL/$objsetid ecL > $tmpfile.2"
+
+typeset sum1=$(cat $tmpfile.1 | md5sum)
+typeset sum2=$(cat $tmpfile.2 | md5sum)
+
+log_must test "$sum1" = "$sum2"
+
+log_pass "zfs send and zdb -B produce the same stream"