zdb: add -B option to generate backup stream

[robn]

Motivation and Context

We were working on a data recovery for a client, and the dataset name table was damaged to the extent that referencing a snapshot we needed, or asking for its name, would panic ZFS.

On further investigation, we found that dmu_send_obj didn't care at all about the name, so we just needed a way to arrange for it to be called.

Description

This adds a -B option, which will produce a standard backup stream from the given pool/objset-id. An optional arg can set the flags.

# zdb -B lucy/34760 Lecw | ./zstream dump
BEGIN record
	hdrtype = 1
	features = 3420000
	magic = 2f5bacbac
	creation_time = 63dc5e6c
...

Sponsored-By: Klara, Inc.

How Has This Been Tested?

Only local testing on the specific feature. Test suite should get anything of interest.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Library ABI change (libzfs, libzfs_core, libnvpair, libuutil and libzfsbootenv)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the OpenZFS code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• I have run the ZFS Test Suite with this change applied.
• All commit messages are properly formatted and contain Signed-off-by.

[mcmilk]

@robn - In general - this is a really cool idea - and also the implementation looks straightforward.

Could you rebase the thing? It's needed for the new testing workflow... So we see, it doesn't break sth.
I will approve this next week, when the testings are run.

[robn]

Could you rebase the thing? It's needed for the new testing workflow... So we see, it doesn't break sth. I will approve this next week, when the testings are run.

Oops! That's a thing I normally just do before a new push, not sure why I missed it this time. Done! Thanks for pointing it out!

[Low-power]

	if (write(1, buf, len) < 0)
		return (errno);

Short writes may occur.

[robn]

Short writes may occur.

Yeah, fair. I've pushed something that should be a little more resilient.

[Low-power]

Don't use arithmetic operation on a void * directly, cast it to char * first, like:

static int
dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
{
	const char *p = buf;
	ssize_t nwritten;

	(void) os;
	(void) arg;

	/* Write the data out, handling short writes and signals. */
	while ((nwritten = write(1, p, len)) < len) {
		if (nwritten < 0) {
			if (errno == EINTR)
				continue;
			return (errno);
		}
		p += nwritten;
		len -= nwritten;
	}

	return (0);
}

[Low-power]

Consider using some C99 features to reduce scpoe of some variables, and make initialization looks better:

--- a/cmd/zdb/zdb.c
+++ b/cmd/zdb/zdb.c
@@ -4897,9 +4897,6 @@ static void
 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
 {
 	const char *c;
-	int err = 0;
-	offset_t off = 0;
-	dmu_send_outparams_t out = {0};
 
 	boolean_t embed = B_FALSE;
 	boolean_t large_block = B_FALSE;
@@ -4927,10 +4924,13 @@ dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
 		}
 	}
 
-	out.dso_outfunc = dump_backup_bytes;
-	out.dso_dryrun = B_FALSE;
+	offset_t off = 0;
+	dmu_send_outparams_t out = {
+		.dso_outfunc = dump_backup_bytes,
+		.dso_dryrun = B_FALSE
+	};
 
-	err = dmu_send_obj(pool, objset_id, 0, embed, large_block, compress,
+	int err = dmu_send_obj(pool, objset_id, 0, embed, large_block, compress,
 	    raw, /* saved */ B_FALSE, 1, &off, &out);
 	if (err != 0) {
 		fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",

[robn]

Don't use arithmetic operation on a void * directly, cast it to char * first, like:

Yeah, silly me.

Consider using some C99 features to reduce scpoe of some variables, and make initialization looks better:

Agreed. I spend a lot of time in codebases where old conventions are still the rules, and a lot of ZFS still looks like that, so I don't think much about it. But reviewer wins here :)

Thanks!

[Low-power]

Lasty in this comment:

/* Write the data out, handling short reads and signals. */

I think the code means short writes not reads.

[Low-power]

When trying out this feature, I accidently runs it without redirecting, thus trashed my terminal...

May be it is better to have a isatty(STDOUT_FILENO) check just like in zfs-send(8).

[robn]

I think the code means short writes not reads.

Sigh, I'm really having a day of it today. Thanks, fixed.

May be it is better to have a isatty(STDOUT_FILENO) check just like in zfs-send(8).

Great suggestion. Done!

[behlendorf]

This is a nice addition to zdb. However, it looks like it subtly broke two of the CI tests you'll need to run those failures down before we can integrate this.

Tests with results other than PASS that are unexpected:
    FAIL redacted_send/redacted_props (expected PASS)
    FAIL rsend/send_freeobjects (expected PASS)

Please also go ahead and add a new test case for this functionality under tests/zfs-tests/tests/functional/cli_root/zdb/.

[robn]

@behlendorf thanks for the heads up, I missed those - I thought they were obviously false positives, with all the churn happening in the tests at the time.

I should have time to get back to this in the next week or two. Cheers.

[robn]

Not quite done with this yet, still have to add a test for it and I want to update the docs a little bit to say that it can handle sending whole real datasets, not just snapshots (which it obviously can, but I hadn't come at it from that angle, and then we used it for that and won the day 😲).

[behlendorf]

@robn sounds good.

[robn]

@behlendorf good to go (pending remaining tests). I added a commit to expose a function I need; up to you if you would prefer to squash it or not. Cheers!

[behlendorf]

This looks good. Thanks for wrapping it up, I can squash it when merging.

[behlendorf]

Merged, I opted to keep the commits split.

[robn]

Thanks all!