mkfs: fix zone capacity check for sequential zones
PATCH 1/1 mkfs: fix zone capacity check for sequential zones
From: Carlos Maiolino <[email protected]>
Sequential zones can have a different, smaller capacity than conventional zones.
Currently mkfs assumes both sequential and conventional zones will have the same capacity and and set the zone_info to the capacity of the first found zone and use that value to validate all the remaining zones's capacity.
Because conventional zones can't have a different capacity than its size, the first zone always have the largest possible capacity, so, mkfs will fail to validate any consecutive sequential zone if its capacity is smaller than the conventional zones.
What we should do instead, is set the zone info capacity accordingly to the settings of first zone found of the respective type and validate the capacity based on that instead of assuming all zones will have the same capacity.
---
FWIS, writing the patch description I'm assuming that the conventional zones always come first and we can't have a zoned device starting with sequential zones followed by conventional zones.
mkfs/xfs_mkfs.c | 50 +++++++++++++++++++++++++++++++------------------ 1 file changed, 32 insertions(+), 18 deletions(-)
mkfs/xfs_mkfs.c +32 -18
@@ -2545,6 +2545,28 @@ struct zone_topology {/* random size that allows efficient processing */ #define ZONES_PER_IOCTL 16384 +static void +zone_validate_capacity ( +struct zone_info *zi, +__u64 capacity, +bool conventional) +{ +if (conventional && (zi->zone_capacity != zi->zone_size)) { +fprintf (stderr,_ ("Zone capacity equal to Zone size required for conventional zones.\n" )); +exit (1 ); + } + +if (zi->zone_capacity > zi->zone_size) { +fprintf (stderr,_ ("Zone capacity larger than zone size!\n" )); +exit (1 ); + } + +if (zi->zone_capacity != capacity) { +fprintf (stderr,_ ("Inconsistent zone capacity!\n" )); +exit (1 ); + } +} + static void report_zones( const char *name,
@@ -2621,6 +2643,11 @@ _("Inconsistent zone size!\n"));switch (zones[i].type) {case BLK_ZONE_TYPE_CONVENTIONAL : +if (!zi->zone_capacity) + zi->zone_capacity = zones[i].capacity; +zone_validate_capacity (zi, zones[i].capacity, + true); +/* * We can only use the conventional space at the * start of the device for metadata, so don't
@@ -2632,6 +2659,11 @@ _("Inconsistent zone size!\n")); zi->nr_conv_zones++;break ;case BLK_ZONE_TYPE_SEQWRITE_REQ : +if (!found_seq) + zi->zone_capacity = zones[i].capacity; +zone_validate_capacity (zi, zones[i].capacity, + false); + found_seq = true;break ;case BLK_ZONE_TYPE_SEQWRITE_PREF :
@@ -2644,19 +2676,6 @@ _("Unknown zone type (0x%x) found.\n"), zones[i].type);exit (1 ); } -if (!n) { - zi->zone_capacity = zones[i].capacity; -if (zi->zone_capacity > zi->zone_size) { -fprintf (stderr, -_ ("Zone capacity larger than zone size!\n" )); -exit (1 ); - } - }else if (zones[i].capacity != zi->zone_capacity) { -fprintf (stderr, -_ ("Inconsistent zone capacity!\n" )); -exit (1 ); - } - n++; } sector = zones[rep->nr_zones - 1].start +
@@ -2683,11 +2702,6 @@ validate_zoned( _("Data devices requires conventional zones.\n"));usage (); } -if (zt->data.zone_capacity != zt->data.zone_size) { -fprintf (stderr, -_ ("Zone capacity equal to Zone size required for conventional zones.\n" )); -usage (); - } cli->sb_feat.zoned = true; cfg->rtstart =
Signature · 1 line
2.51.0
Quoted 3 lines · On Wed, Nov 12, 2025 at 01:33:16PM +0100, [email protected] wrote:
FWIS, writing the patch description I'm assuming that the conventional zones always come first and we can't have a zoned device starting with sequential zones followed by conventional zones.
XFS kinda assumes this by not counting counting conventional zones found after sequential zones, so we'll error out due to the lack of conventional space later unless a separate device is used.
mkfs/xfs_mkfs.c struct zone_topology { · line 2554
Quoted 7 lines
mkfs/xfs_mkfs.c +7
-0+static void +zone_validate_capacity ( +struct zone_info *zi, +__u64 capacity, +bool conventional) +{ +if (conventional && (zi->zone_capacity != zi->zone_size)) {
No need for the inner braces.
Otherwise this looks good:
Note that we should also verify that if a device is used purely as RT device (aka no internal RT device), that there are either no conventional zones, or zone capacity == zone_size. I can look into that as an incremental patch.
Quoted 19 lines · Christoph Hellwig
Quoted 3 lines · On Wed, Nov 12, 2025 at 01:33:16PM +0100, [email protected] wrote:
FWIS, writing the patch description I'm assuming that the conventional zones always come first and we can't have a zoned device starting with sequential zones followed by conventional zones.
XFS kinda assumes this by not counting counting conventional zones found after sequential zones, so we'll error out due to the lack of conventional space later unless a separate device is used.
mkfs/xfs_mkfs.c struct zone_topology { · line 2554
Quoted 7 lines
mkfs/xfs_mkfs.c +7
-0+static void +zone_validate_capacity ( +struct zone_info *zi, +__u64 capacity, +bool conventional) +{ +if (conventional && (zi->zone_capacity != zi->zone_size)) {No need for the inner braces.
Quoted 8 lines
Otherwise this looks good:
Reviewed-by Christoph Hellwig <[email protected]>Note that we should also verify that if a device is used purely as RT device (aka no internal RT device), that there are either no conventional zones, or zone capacity ==
zone_size. I can look into that as an incremental patch.
Your call. I can add it to the V2, giving I'll rewrite the above to remove the unneeded braces.
Couldn't we leverage this check in validate_zoned() for that?
if (cli->xi->rt.name && !cli->xi->rt.isfile) {
[...]I'm not sure though if just detecting a rtdev was passed on the CLI would suffice to cover what you proposed.
Quoted 1 line
Quoted 2 lines · Carlos Maiolino
Your call. I can add it to the V2, giving I'll rewrite the above to remove the unneeded braces.
Nah, let's do one thing at a time.
Quoted 6 lines · Christoph Hellwig
Quoted 2 lines · Carlos Maiolino
Your call. I can add it to the V2, giving I'll rewrite the above to remove the unneeded braces.
Nah, let's do one thing at a time.
okie!