Skip to contentSkip to conversation

mkfs: custom agcount that renders AG size < XFS_AG_MIN_BYTES gives "Assertion failed. Aborted"

PATCH In review Srikanth C S 3 participants 5 messages Jul 5, 2022 – Jul 6, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mkfs: custom agcount that renders AG size < XFS_AG_MIN_BYTES gives "Assertion failed. Aborted"

For a 2GB FS we have

$ mkfs.xfs -f -d agcount=129 test.img
mkfs.xfs: xfs_mkfs.c:3021: align_ag_geometry: Assertion `!cli_opt_set(&dopts, D_AGCOUNT)' failed.
Aborted

With this change we have

$ mkfs.xfs -f -d agcount=129 test.img
Invalid value 129 for -d agcount option. Value is too large.
---
 mkfs/xfs_mkfs.c | 7 +++++++
 1 file changed, 7 insertions(+)
mkfs/xfs_mkfs.c +7 -0
index 057b3b09..32dcbfff 100644
@@ -2897,6 +2897,13 @@ _("agsize (%s) not a multiple of fs blk size (%d)\n"),
 		cfg->agcount = cli->agcount;
 		cfg->agsize = cfg->dblocks / cfg->agcount +
 				(cfg->dblocks % cfg->agcount != 0);
+		if (cfg->agsize < XFS_AG_MIN_BYTES >> cfg->blocklog)
+		{
+			fprintf(stderr,
+_("Invalid value %lld for -d agcount option. Value is too large.\n"),
+				(long long)cli->agcount);
+			usage();	
+		}
 	} else {
 		calc_default_ag_geometry(cfg->blocklog, cfg->dblocks,
 					 cfg->dsunit, &cfg->agsize,
Signature · 1 line
2.25.1
#1 lore raw
Quoted 4 lines · Srikanth C S

For a 2GB FS we have

$ mkfs.xfs -f -d agcount=129 test.img
mkfs.xfs: xfs_mkfs.c:3021: align_ag_geometry: Assertion `!cli_opt_set(&dopts, D_AGCOUNT)' failed.
Aborted

Ok, that's because the size of the last AG is too small when trying to align the AG size to stripe geometry. It fails an assert that says "we should not get here if the agcount was specified on the CLI".

Quoted 4 lines

With this change we have

$ mkfs.xfs -f -d agcount=129 test.img
Invalid value 129 for -d agcount option. Value is too large.

OK, but....

Quoted 11 lines
---
 mkfs/xfs_mkfs.c | 7 +++++++
 1 file changed, 7 insertions(+)
mkfs/xfs_mkfs.c +0 -0
index 057b3b09..32dcbfff 100644

mkfs/xfs_mkfs.c _("agsize (%s) not a multiple of fs blk size (%d)\n"), · line 2906

Quoted 10 lines
mkfs/xfs_mkfs.c +7 -0
@@ -2897,6 +2897,13 @@ _("agsize (%s) not a multiple of fs blk size (%d)\n"),
 		cfg->agcount = cli->agcount;
 		cfg->agsize = cfg->dblocks / cfg->agcount +
 				(cfg->dblocks % cfg->agcount != 0);
+		if (cfg->agsize < XFS_AG_MIN_BYTES >> cfg->blocklog)
+		{
+			fprintf(stderr,
+_("Invalid value %lld for -d agcount option. Value is too large.\n"),
+				(long long)cli->agcount);
+			usage();	
+		}

.... that's not where we validate the calculated ag size. That happens via align_ag_geometry() -> validate_ag_geometry(). i.e. we can't reject an AG specification until after we've applied all the necessary modifiers to it first (such as stripe alignment requirements).

IOWs, we do actually check for valid AG sizes, it's just that this specific case hit an ASSERT() check before we got to validating the AG size. I'm pretty sure just removing the ASSERT - which assumes that "-d agcount=xxx" is not so large that it produces undersized AGs - will fix the problem and result in the correct error message being returned.

Cheers,

Dave.

Signature · 2 lines
Dave Chinner
[email protected]
#2 lore raw
Quoted 6 lines · Dave Chinner
Quoted 4 lines · Srikanth C S

For a 2GB FS we have

$ mkfs.xfs -f -d agcount=129 test.img
mkfs.xfs: xfs_mkfs.c:3021: align_ag_geometry: Assertion `!cli_opt_set(&dopts, D_AGCOUNT)' failed.
Aborted
Quoted 9 lines

Ok, that's because the size of the last AG is too small when trying to align the AG size to stripe geometry. It fails an assert that says "we should not get here if the agcount was specified on the CLI".

Quoted 4 lines

With this change we have

$ mkfs.xfs -f -d agcount=129 test.img
Invalid value 129 for -d agcount option. Value is too large.

What version of mkfs is this?

$ truncate -s 2g /tmp/a
$ mkfs.xfs -V
mkfs.xfs version 5.18.0
$ mkfs.xfs -f -d agcount=129 /tmp/a
agsize (4065 blocks) too small, need at least 4096 blocks
Quoted 30 lines

OK, but....

Quoted 11 lines
---
 mkfs/xfs_mkfs.c | 7 +++++++
 1 file changed, 7 insertions(+)
mkfs/xfs_mkfs.c +0 -0
index 057b3b09..32dcbfff 100644

mkfs/xfs_mkfs.c _("agsize (%s) not a multiple of fs blk size (%d)\n"), · line 2906

Quoted 10 lines
mkfs/xfs_mkfs.c +7 -0
@@ -2897,6 +2897,13 @@ _("agsize (%s) not a multiple of fs blk size (%d)\n"),
 		cfg->agcount = cli->agcount;
 		cfg->agsize = cfg->dblocks / cfg->agcount +
 				(cfg->dblocks % cfg->agcount != 0);
+		if (cfg->agsize < XFS_AG_MIN_BYTES >> cfg->blocklog)
+		{
+			fprintf(stderr,
+_("Invalid value %lld for -d agcount option. Value is too large.\n"),
+				(long long)cli->agcount);
+			usage();	
+		}

.... that's not where we validate the calculated ag size. That happens via align_ag_geometry() -> validate_ag_geometry(). i.e. we can't reject an AG specification until after we've applied all the necessary modifiers to it first (such as stripe alignment requirements).

Quoted 6 lines

IOWs, we do actually check for valid AG sizes, it's just that this specific case hit an ASSERT() check before we got to validating the AG size. I'm pretty sure just removing the ASSERT - which assumes that "-d agcount=xxx" is not so large that it produces undersized AGs - will fix the problem and result in the correct error message being returned.

(Agreed.)

--D

Quoted 7 lines

Cheers,

Dave.

Signature · 2 lines
Dave Chinner
[email protected]
#3 lore raw
Quoted 19 lines · Dave Chinner
Quoted 4 lines · Srikanth C S
Quoted 3 lines

For a 2GB FS we have

$ mkfs.xfs -f -d agcount=129 test.img
mkfs.xfs: xfs_mkfs.c:3021: align_ag_geometry: Assertion

`!cli_opt_set(&dopts, D_AGCOUNT)' failed.

Quoted 10 lines
Quoted 1 line

Aborted

Ok, that's because the size of the last AG is too small when trying to align the AG size to stripe geometry. It fails an assert that says "we should not get here if the agcount was specified on the CLI".

Quoted 4 lines

With this change we have

$ mkfs.xfs -f -d agcount=129 test.img Invalid value 129 for -d
agcount option. Value is too large.
Quoted 8 lines

What version of mkfs is this?

$ truncate -s 2g /tmp/a
$ mkfs.xfs -V
mkfs.xfs version 5.18.0
$ mkfs.xfs -f -d agcount=129 /tmp/a
agsize (4065 blocks) too small, need at least 4096 blocks

For the same version I get Assertion failed

$ truncate -s 2g /tmp/a
$ mkfs.xfs -V
mkfs.xfs version 5.18.0
$ mkfs.xfs -f -d agcount=129 /tmp/a
mkfs.xfs: xfs_mkfs.c:3033: align_ag_geometry: Assertion `!cli_opt_set(&dopts, D_AGCOUNT)' failed.
Aborted (core dumped)
Quoted 48 lines
Quoted 8 lines

OK, but....

Quoted 6 lines
---
 mkfs/xfs_mkfs.c | 7 +++++++
 1 file changed, 7 insertions(+)
Quoted 5 lines

mkfs/xfs_mkfs.c _("agsize (%s) not a multiple of fs blk size

Quoted 5 lines
mkfs/xfs_mkfs.c index renamed from mkfs/xfs_mkfs.c +0 -0

057b3b09..32dcbfff 100644

mkfs/xfs_mkfs.c +0 -0
@@ -2897,6 +2897,13 @@ _("agsize (%s) not a multiple of fs blk size

(%d)\n"),

Quoted 16 lines

mkfs/xfs_mkfs.c _("agsize (%s) not a multiple of fs blk size (%d)\n"), · line 2906

Quoted 10 lines
mkfs/xfs_mkfs.c +7 -0
 		cfg->agcount = cli->agcount;
 		cfg->agsize = cfg->dblocks / cfg->agcount +
 				(cfg->dblocks % cfg->agcount != 0);
+		if (cfg->agsize < XFS_AG_MIN_BYTES >> cfg->blocklog)
+		{
+			fprintf(stderr,
+_("Invalid value %lld for -d agcount option. Value is too large.\n"),
+				(long long)cli->agcount);
+			usage();
+		}

.... that's not where we validate the calculated ag size. That happens via align_ag_geometry() -> validate_ag_geometry(). i.e. we can't reject an AG specification until after we've applied all the necessary modifiers to it first (such as stripe alignment requirements).

Quoted 6 lines

IOWs, we do actually check for valid AG sizes, it's just that this specific case hit an ASSERT() check before we got to validating the AG size. I'm pretty sure just removing the ASSERT - which assumes that "-d agcount=xxx" is not so large that it produces undersized AGs - will fix the problem and result in the correct error message being returned.

(Agreed.)

--D

Quoted 7 lines

Cheers,

Dave.

--
Dave Chinner
[email protected]

#4 lore raw
Quoted 28 lines · Srikanth C S
Quoted 19 lines · Dave Chinner
Quoted 4 lines · Srikanth C S
Quoted 3 lines

For a 2GB FS we have

$ mkfs.xfs -f -d agcount=129 test.img
mkfs.xfs: xfs_mkfs.c:3021: align_ag_geometry: Assertion

`!cli_opt_set(&dopts, D_AGCOUNT)' failed.

Quoted 10 lines
Quoted 1 line

Aborted

Ok, that's because the size of the last AG is too small when trying to align the AG size to stripe geometry. It fails an assert that says "we should not get here if the agcount was specified on the CLI".

Quoted 4 lines

With this change we have

$ mkfs.xfs -f -d agcount=129 test.img Invalid value 129 for -d
agcount option. Value is too large.
Quoted 8 lines

What version of mkfs is this?

$ truncate -s 2g /tmp/a
$ mkfs.xfs -V
mkfs.xfs version 5.18.0
$ mkfs.xfs -f -d agcount=129 /tmp/a
agsize (4065 blocks) too small, need at least 4096 blocks
Quoted 7 lines

For the same version I get Assertion failed

$ truncate -s 2g /tmp/a
$ mkfs.xfs -V
mkfs.xfs version 5.18.0
$ mkfs.xfs -f -d agcount=129 /tmp/a
mkfs.xfs: xfs_mkfs.c:3033: align_ag_geometry: Assertion `!cli_opt_set(&dopts, D_AGCOUNT)' failed.
Aborted (core dumped)

ahaha, the distro package got built with -DNDEBUG, which turned off ASSERTions. With my upstream tot build I see this problem too...

Quoted 30 lines
Quoted 30 lines
Quoted 8 lines

OK, but....

Quoted 6 lines
---
 mkfs/xfs_mkfs.c | 7 +++++++
 1 file changed, 7 insertions(+)
Quoted 5 lines

mkfs/xfs_mkfs.c _("agsize (%s) not a multiple of fs blk size

Quoted 5 lines
mkfs/xfs_mkfs.c index renamed from mkfs/xfs_mkfs.c +0 -0

057b3b09..32dcbfff 100644

mkfs/xfs_mkfs.c +0 -0
@@ -2897,6 +2897,13 @@ _("agsize (%s) not a multiple of fs blk size

(%d)\n"),

Quoted 16 lines

mkfs/xfs_mkfs.c _("agsize (%s) not a multiple of fs blk size (%d)\n"), · line 2906

Quoted 10 lines
mkfs/xfs_mkfs.c +7 -0
 		cfg->agcount = cli->agcount;
 		cfg->agsize = cfg->dblocks / cfg->agcount +
 				(cfg->dblocks % cfg->agcount != 0);
+		if (cfg->agsize < XFS_AG_MIN_BYTES >> cfg->blocklog)
+		{
+			fprintf(stderr,
+_("Invalid value %lld for -d agcount option. Value is too large.\n"),
+				(long long)cli->agcount);
+			usage();
+		}

.... that's not where we validate the calculated ag size. That happens via align_ag_geometry() -> validate_ag_geometry(). i.e. we can't reject an AG specification until after we've applied all the necessary modifiers to it first (such as stripe alignment requirements).

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

IOWs, we do actually check for valid AG sizes, it's just that this specific case hit an ASSERT() check before we got to validating the AG size. I'm pretty sure just removing the ASSERT - which assumes that "-d agcount=xxx" is not so large that it produces undersized AGs - will fix the problem and result in the correct error message being returned.

...so yeah, what Dave said. :)

--D

Quoted 12 lines
Quoted 12 lines

(Agreed.)

--D

Quoted 7 lines

Cheers,

Dave.

--
Dave Chinner
[email protected]