Skip to contentSkip to conversation

enable cached zone report v4

PATCH In review Christoph Hellwig 5 participants 16 messages Jan 9 – Jan 27

lore
Share & tools
lore
#0 lore raw

Enable cached zone report to speed up mkfs and repair on a zoned block device (e.g. an SMR disk). Cached zone report support was introduced in the kernel with version 6.19-rc1.

Note: I've taken this series over from Damien as he's busy and it need it as a base for further work.

Changes from v3:
- reorder includes to not need the forward declaration in xfs_zones.h
- use the libfrog/ for #include statements
- fix the include guard in libfrog/zones.h
- fix up i18n string mess
- hide the new ioctl definition in libfrog/zones.c
- don't add userspace includes to libxfs/xfs_zones.h
- reuse the buffer for multiple report zone calls

Changes from v2:
- Complete rework of the series to make the zone reporting code common in libfrog
- Added patch 1 and 2 as small cleanups/improvements.

Changes from v1:
- Fix erroneous handling of ioctl(BLKREPORTZONEV2) error to correctly fallback to the regular ioctl(BLKREPORTZONE) if the kernel does not support BLKREPORTZONEV2.

#1 lore raw

PATCH 1/4 xfs: use blkdev_report_zones_cached()

From: Damien Le Moal <[email protected]>

Source kernel commit: e04ccfc28252f181ea8d469d834b48e7dece65b2

Modify xfs_mount_zones() to replace the call to blkdev_report_zones() with blkdev_report_zones_cached() to speed-up mount operations. Since this causes xfs_zone_validate_seq() to see zones with the BLK_ZONE_COND_ACTIVE condition, this function is also modified to acept this condition as valid.

With this change, mounting a freshly formatted large capacity (30 TB) SMR HDD completes under 2s compared to over 4.7s before.

---
 include/platform_defs.h | 4 ++++
 libxfs/xfs_zones.c      | 1 +
 2 files changed, 5 insertions(+)
include/platform_defs.h +4 -0
index da966490b0f5..cfdaca642645 100644
@@ -307,4 +307,8 @@ struct kvec {
 	size_t iov_len;
 };
 
+#ifndef BLK_ZONE_COND_ACTIVE /* added in Linux 6.19 */
+#define BLK_ZONE_COND_ACTIVE	0xff
+#endif
+
 #endif	/* __XFS_PLATFORM_DEFS_H__ */
libxfs/xfs_zones.c +1 -0
index 7a81d83f5b3e..99ae05ce7473 100644
@@ -97,6 +97,7 @@ xfs_zone_validate_seq(
 	case BLK_ZONE_COND_IMP_OPEN:
 	case BLK_ZONE_COND_EXP_OPEN:
 	case BLK_ZONE_COND_CLOSED:
+	case BLK_ZONE_COND_ACTIVE:
 		return xfs_zone_validate_wp(zone, rtg, write_pointer);
 	case BLK_ZONE_COND_FULL:
 		return xfs_zone_validate_full(zone, rtg, write_pointer);
Signature · 1 line
2.47.3
#2 lore raw

Re: [PATCH 1/4] xfs: use blkdev_report_zones_cached()

Quoted 27 lines · Christoph Hellwig

From: Damien Le Moal <[email protected]>

Source kernel commit: e04ccfc28252f181ea8d469d834b48e7dece65b2

Modify xfs_mount_zones() to replace the call to blkdev_report_zones() with blkdev_report_zones_cached() to speed-up mount operations. Since this causes xfs_zone_validate_seq() to see zones with the BLK_ZONE_COND_ACTIVE condition, this function is also modified to acept this condition as valid.

With this change, mounting a freshly formatted large capacity (30 TB) SMR HDD completes under 2s compared to over 4.7s before.

---
 include/platform_defs.h | 4 ++++
 libxfs/xfs_zones.c      | 1 +
 2 files changed, 5 insertions(+)
include/platform_defs.h +0 -0
index da966490b0f5..cfdaca642645 100644

include/platform_defs.h struct kvec { · line 311

Quoted 5 lines
include/platform_defs.h +2 -0
@@ -307,4 +307,8 @@ struct kvec {
 	size_t iov_len;
 };
 
+#ifndef BLK_ZONE_COND_ACTIVE /* added in Linux 6.19 */
+#define BLK_ZONE_COND_ACTIVE	0xff

hmm I think #ifndef doesn't work for enum member. Compiling against linux 6.19-rc6:

../include/platform_defs.h:311:33: error: expected identifier before numeric constant

  311 | #define BLK_ZONE_COND_ACTIVE    0xff
      |                                 ^~~~
/linux-headers-v6.19-rc6/include/linux/blkzoned.h:84:9: note: in expansion of macro ‘BLK_ZONE_COND_ACTIVE’
   84 |         BLK_ZONE_COND_ACTIVE    = 0xFF,
      |         ^~~~~~~~~~~~~~~~~~~~
Signature · 1 line
- Andrey
#3 lore raw

Re: [PATCH 1/4] xfs: use blkdev_report_zones_cached()

Quoted 37 lines · Andrey Albershteyn
Quoted 27 lines · Christoph Hellwig

From: Damien Le Moal <[email protected]>

Source kernel commit: e04ccfc28252f181ea8d469d834b48e7dece65b2

Modify xfs_mount_zones() to replace the call to blkdev_report_zones() with blkdev_report_zones_cached() to speed-up mount operations. Since this causes xfs_zone_validate_seq() to see zones with the BLK_ZONE_COND_ACTIVE condition, this function is also modified to acept this condition as valid.

With this change, mounting a freshly formatted large capacity (30 TB) SMR HDD completes under 2s compared to over 4.7s before.

---
 include/platform_defs.h | 4 ++++
 libxfs/xfs_zones.c      | 1 +
 2 files changed, 5 insertions(+)
include/platform_defs.h +0 -0
index da966490b0f5..cfdaca642645 100644

include/platform_defs.h struct kvec { · line 311

Quoted 5 lines
include/platform_defs.h +2 -0
@@ -307,4 +307,8 @@ struct kvec {
 	size_t iov_len;
 };
 
+#ifndef BLK_ZONE_COND_ACTIVE /* added in Linux 6.19 */
+#define BLK_ZONE_COND_ACTIVE	0xff

hmm I think #ifndef doesn't work for enum member. Compiling against linux 6.19-rc6:

Quoted 6 lines

../include/platform_defs.h:311:33: error: expected identifier before numeric constant

  311 | #define BLK_ZONE_COND_ACTIVE    0xff
      |                                 ^~~~
/linux-headers-v6.19-rc6/include/linux/blkzoned.h:84:9: note: in expansion of macro ‘BLK_ZONE_COND_ACTIVE’
   84 |         BLK_ZONE_COND_ACTIVE    = 0xFF,
      |         ^~~~~~~~~~~~~~~~~~~~

I hacked around this very crudely:
https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/commit/?h=djwong-wtf&id=d5d9b4bd2da95c7b9429112f1c1098a62f155270

Though I think one of hch's fix patches fixed that. I cannot post a lore link because the #@!%)*&!%!!! anubis b@%#%!!! is completely broken and will not let me in. But we did the review for the kernel-side fixes within the last couple of weeks.

--D

Quoted 5 lines
Signature · 2 lines
-- 
- Andrey
#4 lore raw

Re: [PATCH 1/4] xfs: use blkdev_report_zones_cached()

Quoted 13 lines · Darrick J. Wong
Quoted 6 lines

include/platform_defs.h struct kvec { · line 311

Quoted 2 lines
include/platform_defs.h +2 -0
+#ifndef BLK_ZONE_COND_ACTIVE /* added in Linux 6.19 */
+#define BLK_ZONE_COND_ACTIVE	0xff

hmm I think #ifndef doesn't work for enum member. Compiling against linux 6.19-rc6:

Quoted 6 lines

../include/platform_defs.h:311:33: error: expected identifier before numeric constant

  311 | #define BLK_ZONE_COND_ACTIVE    0xff
      |                                 ^~~~
/linux-headers-v6.19-rc6/include/linux/blkzoned.h:84:9: note: in expansion of macro ‘BLK_ZONE_COND_ACTIVE’
   84 |         BLK_ZONE_COND_ACTIVE    = 0xFF,
      |         ^~~~~~~~~~~~~~~~~~~~
Quoted 2 lines

I hacked around this very crudely:
https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/commit/?h=djwong-wtf&id=d5d9b4bd2da95c7b9429112f1c1098a62f155270

Urgg. That's why we should not use enums in uapi headers..

Damien, I guess we could do a:

#define BLK_ZONE_COND_ACTIVE BLK_ZONE_COND_ACTIVE

as the usual trick to fix that in the blkzoned.h header, and given that 6.19 hasn't been released still get it to Linus in time.

Quoted 4 lines

Though I think one of hch's fix patches fixed that. I cannot post a lore link because the #@!%)*&!%!!! anubis b@%#%!!! is completely broken and will not let me in. But we did the review for the kernel-side fixes within the last couple of weeks.

I don't think I fixed it. Not intentionally, and not unintentionally in a way I'd actually understand at least :)

#5 lore raw

PATCH 1/1 libfrog: adjust header order for BLK_ZONE_COND_ACTIVE #ifndef check

Even after kernel fix via defining BLK_ZONE_COND_ACTIVE the compiler is still failing on this with:

In file included from zones.c:5:
../include/platform_defs.h:311:33: error: expected identifier before numeric constant

  311 | #define BLK_ZONE_COND_ACTIVE    0xff
      |                                 ^~~~
/linux-headers-v6.19-rc7/include/linux/blkzoned.h:84:9: note: in expansion of macro ‘BLK_ZONE_COND_ACTIVE’
   84 |         BLK_ZONE_COND_ACTIVE    = 0xFF, /* added in Linux 6.19 */
      |         ^~~~~~~~~~~~~~~~~~~~
In file included from ../libfrog/zones.h:8,
                 from zones.c:7:
/linux-headers-v6.19-rc7/include/linux/blkzoned.h:85:9: warning: ‘BLK_ZONE_COND_ACTIVE’ redefined
   85 | #define BLK_ZONE_COND_ACTIVE    BLK_ZONE_COND_ACTIVE
      |         ^~~~~~~~~~~~~~~~~~~~
../include/platform_defs.h:311:9: note: this is the location of the previous definition
  311 | #define BLK_ZONE_COND_ACTIVE    0xff
      |         ^~~~~~~~~~~~~~~~~~~~

This is because of the order of #ifndef (in platform_defs.h) and #define (in blkzoned.h). The platform_defs.h defines BLK_ZONE_COND_ACTIVE first and this causes enum to fail and macro being redefined. Fix this by including libfrog/zones.h first, which includes blkzoned.h. Add stdint.h for uint64_t in xfrog_report_zones().

---
 libfrog/zones.c | 2 +-
 libfrog/zones.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)
libfrog/zones.c +1 -1
index f1ef0b24c564..68be7388172b 100644
@@ -2,9 +2,9 @@
 /*
  * Copyright (c) 2025, Western Digital Corporation or its affiliates.
  */
+#include "libfrog/zones.h"
 #include "platform_defs.h"
 #include "atomic.h"
-#include "libfrog/zones.h"
 #include <sys/ioctl.h>
 
 /*
libfrog/zones.h +1 -0
index 33c1da7ef192..4cc4d15c4dab 100644
@@ -6,6 +6,7 @@
 #define __LIBFROG_ZONES_H__
 
 #include <linux/blkzoned.h>
+#include <stdint.h>
 
 struct xfrog_zone_report {
 	struct blk_zone_report	rep;
Signature · 1 line
2.51.2
author Reviewed-by on v1 1/1
#6 lore raw

Re: [PATCH] libfrog: adjust header order for BLK_ZONE_COND_ACTIVE #ifndef check

Quoted 5 lines · Andrey Albershteyn

This is because of the order of #ifndef (in platform_defs.h) and #define (in blkzoned.h). The platform_defs.h defines BLK_ZONE_COND_ACTIVE first and this causes enum to fail and macro being redefined. Fix this by including libfrog/zones.h first, which includes blkzoned.h. Add stdint.h for uint64_t in xfrog_report_zones().

Thanks, this looks good:

please order this before the cached zone reporting series, or fold it in to preserve bisectability. Or should I resend the series with this change included?

#7 lore raw

Re: [PATCH] libfrog: adjust header order for BLK_ZONE_COND_ACTIVE #ifndef check

Quoted 10 lines · Christoph Hellwig
Quoted 5 lines · Andrey Albershteyn

This is because of the order of #ifndef (in platform_defs.h) and #define (in blkzoned.h). The platform_defs.h defines BLK_ZONE_COND_ACTIVE first and this causes enum to fail and macro being redefined. Fix this by including libfrog/zones.h first, which includes blkzoned.h. Add stdint.h for uint64_t in xfrog_report_zones().

Thanks, this looks good:

Thanks!

Quoted 2 lines

please order this before the cached zone reporting series, or fold it in

I will apply it first

Signature · 1 line
- Andrey
#8 lore raw

Re: [PATCH] libfrog: adjust header order for BLK_ZONE_COND_ACTIVE #ifndef check

Quoted 10 lines · Andrey Albershteyn
Quoted 9 lines · Christoph Hellwig
Quoted 5 lines · Andrey Albershteyn

This is because of the order of #ifndef (in platform_defs.h) and #define (in blkzoned.h). The platform_defs.h defines BLK_ZONE_COND_ACTIVE first and this causes enum to fail and macro being redefined. Fix this by including libfrog/zones.h first, which includes blkzoned.h. Add stdint.h for uint64_t in xfrog_report_zones().

Thanks, this looks good:

Quoted 8 lines
Quoted 1 line

Thanks!

Quoted 2 lines

please order this before the cached zone reporting series, or fold it in

I will apply it first

hmm, actually there's no this header to apply fix to, can you resend it with this fix applied?

Signature · 1 line
- Andrey
#9 lore raw

Re: [PATCH] libfrog: adjust header order for BLK_ZONE_COND_ACTIVE #ifndef check

Quoted 4 lines · Andrey Albershteyn
Quoted 1 line

I will apply it first

hmm, actually there's no this header to apply fix to, can you resend it with this fix applied?

Sure.

#10 lore raw

PATCH 2/4 mkfs: remove unnecessary return value affectation

In reply to Christoph Hellwig

From: Damien Le Moal <[email protected]>

The function report_zones() in mkfs/xfs_mkfs.c is a void function. So there is no need to set the variable ret to -EIO before returning if fstat() fails.

Fixes 2e5a737a61d3 ("xfs_mkfs: support creating file system with zoned RT devices")
---
 mkfs/xfs_mkfs.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
mkfs/xfs_mkfs.c +1 -3
index b34407725f76..215cff8db7b1 100644
@@ -2587,10 +2587,8 @@ report_zones(
 		exit(1);
 	}
 
-	if (fstat(fd, &st) < 0) {
-		ret = -EIO;
+	if (fstat(fd, &st) < 0)
 		goto out_close;
-	}
 	if (!S_ISBLK(st.st_mode))
 		goto out_close;
 
Signature · 1 line
2.47.3
#11 lore raw

PATCH 3/4 libfrog: lift common zone reporting code from mkfs and repair

In reply to Christoph Hellwig

From: Damien Le Moal <[email protected]>

Define the new helper function xfrog_report_zones() to report zones of a zoned block device. This function is implemented in the new file libfrog/zones.c and defined in the header file libfrog/zones.h and use it from mkfs and repair instead of the previous open coded versions.

xfrog_report_zones() allocates and returns a struct blk_zone_report structure, which can be be reused by subsequent invocations. It is the responsibility of the caller to free this structure after use.

[hch: refactored to allow buffer reuse]
---
 libfrog/Makefile |  6 ++++--
 libfrog/zones.c  | 39 +++++++++++++++++++++++++++++++++++++++
 libfrog/zones.h  | 18 ++++++++++++++++++
 mkfs/xfs_mkfs.c  | 41 ++++++++++++-----------------------------
 repair/zoned.c   | 35 +++++++++++------------------------
 5 files changed, 84 insertions(+), 55 deletions(-)
 create mode 100644 libfrog/zones.c
 create mode 100644 libfrog/zones.h
libfrog/Makefile +4 -2
index 268fa26638d7..9f405ffe3475 100644
@@ -35,7 +35,8 @@ radix-tree.c \
 randbytes.c \
 scrub.c \
 util.c \
-workqueue.c
+workqueue.c \
+zones.c
 
 HFILES = \
 avl64.h \
@@ -65,7 +66,8 @@ radix-tree.h \
 randbytes.h \
 scrub.h \
 statx.h \
-workqueue.h
+workqueue.h \
+zones.h
 
 GETTEXT_PY = \
 	gettext.py
libfrog/zones.c new file +39 -0
new file mode 100644
index 000000000000..2276c56bec9c
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2025, Western Digital Corporation or its affiliates.
+ */
+#include "platform_defs.h"
+#include "libfrog/zones.h"
+#include <sys/ioctl.h>
+
+/* random size that allows efficient processing */
+#define ZONES_PER_REPORT		16384
+
+struct xfrog_zone_report *
+xfrog_report_zones(
+	int			fd,
+	uint64_t		sector,
+	struct xfrog_zone_report *rep)
+{
+	if (!rep) {
+		rep = calloc(1, struct_size(rep, zones, ZONES_PER_REPORT));
+		if (!rep) {
+			fprintf(stderr, "%s\n",
+_("Failed to allocate memory for reporting zones."));
+			return NULL;
+		}
+	}
+
+	rep->rep.sector = sector;
+	rep->rep.nr_zones = ZONES_PER_REPORT;
+
+	if (ioctl(fd, BLKREPORTZONE, &rep->rep)) {
+		fprintf(stderr, "%s %s\n",
+_("ioctl(BLKREPORTZONE) failed:\n"),
+			strerror(-errno));
+		free(rep);
+		return NULL;
+	}
+
+	return rep;
+}
libfrog/zones.h new file +18 -0
new file mode 100644
index 000000000000..33c1da7ef192
@@ -0,0 +1,18 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2025, Western Digital Corporation or its affiliates.
+ */
+#ifndef __LIBFROG_ZONES_H__
+#define __LIBFROG_ZONES_H__
+
+#include <linux/blkzoned.h>
+
+struct xfrog_zone_report {
+	struct blk_zone_report	rep;
+	struct blk_zone		zones[];
+};
+
+struct xfrog_zone_report *
+xfrog_report_zones(int fd, uint64_t sector, struct xfrog_zone_report *rep);
+
+#endif /* __LIBFROG_ZONES_H__ */
mkfs/xfs_mkfs.c +12 -29

79 lines not on this page · raw message

repair/zoned.c +11 -24
index 206b0158f95f..5102d43e218d 100644
@@ -7,6 +7,7 @@
 #include "libxfs_priv.h"
 #include "libxfs.h"
 #include "xfs_zones.h"
+#include "libfrog/zones.h"
 #include "err_protos.h"
 #include "zoned.h"
 
@@ -51,8 +52,7 @@ check_zones(
 	uint64_t		sector = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rtstart);
 	unsigned int		zone_size, zone_capacity;
 	uint64_t		device_size;
-	size_t			rep_size;
-	struct blk_zone_report	*rep;
+	struct xfrog_zone_report *rep = NULL;
 	unsigned int		i, n = 0;
 
 	if (ioctl(fd, BLKGETSIZE64, &device_size))
@@ -67,31 +67,18 @@ check_zones(
 		return;
 	}
 
-	rep_size = sizeof(struct blk_zone_report) +
-		   sizeof(struct blk_zone) * ZONES_PER_IOCTL;
-	rep = malloc(rep_size);
-	if (!rep) {
-		do_warn(_("malloc failed for zone report\n"));
-		return;
-	}
-
 	while (n < mp->m_sb.sb_rgcount) {
-		struct blk_zone *zones = (struct blk_zone *)(rep + 1);
-		int ret;
+		struct blk_zone *zones;
 
-		memset(rep, 0, rep_size);
-		rep->sector = sector;
-		rep->nr_zones = ZONES_PER_IOCTL;
+		rep = xfrog_report_zones(fd, sector, rep);
+		if (!rep)
+			return;
 
-		ret = ioctl(fd, BLKREPORTZONE, rep);
-		if (ret) {
-			do_error(_("ioctl(BLKREPORTZONE) failed: %d!\n"), ret);
-			goto out_free;
-		}
-		if (!rep->nr_zones)
+		if (!rep->rep.nr_zones)
 			break;
 
-		for (i = 0; i < rep->nr_zones; i++) {
+		zones = rep->zones;
+		for (i = 0; i < rep->rep.nr_zones; i++) {
 			if (n >= mp->m_sb.sb_rgcount)
 				break;
 
@@ -130,8 +117,8 @@ _("Inconsistent zone capacity!\n"));
 			report_zones_cb(mp, &zones[i]);
 			n++;
 		}
-		sector = zones[rep->nr_zones - 1].start +
-			 zones[rep->nr_zones - 1].len;
+		sector = zones[rep->rep.nr_zones - 1].start +
+			 zones[rep->rep.nr_zones - 1].len;
 	}
 
 out_free:
Signature · 1 line
2.47.3
Reviewed-by on v1 3/4
#12 lore raw

Re: [PATCH 3/4] libfrog: lift common zone reporting code from mkfs and repair

Quoted 7 lines · Christoph Hellwig

From: Damien Le Moal <[email protected]>

Define the new helper function xfrog_report_zones() to report zones of a zoned block device. This function is implemented in the new file libfrog/zones.c and defined in the header file libfrog/zones.h and use it from mkfs and repair instead of the previous open coded versions.

Quoted 7 lines

xfrog_report_zones() allocates and returns a struct blk_zone_report structure, which can be be reused by subsequent invocations. It is the responsibility of the caller to free this structure after use.

[hch: refactored to allow buffer reuse]

Looks good now, thanx for avoiding the buffer reallocations

--D

Quoted 267 lines
---
 libfrog/Makefile |  6 ++++--
 libfrog/zones.c  | 39 +++++++++++++++++++++++++++++++++++++++
 libfrog/zones.h  | 18 ++++++++++++++++++
 mkfs/xfs_mkfs.c  | 41 ++++++++++++-----------------------------
 repair/zoned.c   | 35 +++++++++++------------------------
 5 files changed, 84 insertions(+), 55 deletions(-)
 create mode 100644 libfrog/zones.c
 create mode 100644 libfrog/zones.h
libfrog/Makefile +4 -2
index 268fa26638d7..9f405ffe3475 100644
@@ -35,7 +35,8 @@ radix-tree.c \
 randbytes.c \
 scrub.c \
 util.c \
-workqueue.c
+workqueue.c \
+zones.c
 
 HFILES = \
 avl64.h \
@@ -65,7 +66,8 @@ radix-tree.h \
 randbytes.h \
 scrub.h \
 statx.h \
-workqueue.h
+workqueue.h \
+zones.h
 
 GETTEXT_PY = \
 	gettext.py
libfrog/zones.c new file +39 -0
new file mode 100644
index 000000000000..2276c56bec9c
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2025, Western Digital Corporation or its affiliates.
+ */
+#include "platform_defs.h"
+#include "libfrog/zones.h"
+#include <sys/ioctl.h>
+
+/* random size that allows efficient processing */
+#define ZONES_PER_REPORT		16384
+
+struct xfrog_zone_report *
+xfrog_report_zones(
+	int			fd,
+	uint64_t		sector,
+	struct xfrog_zone_report *rep)
+{
+	if (!rep) {
+		rep = calloc(1, struct_size(rep, zones, ZONES_PER_REPORT));
+		if (!rep) {
+			fprintf(stderr, "%s\n",
+_("Failed to allocate memory for reporting zones."));
+			return NULL;
+		}
+	}
+
+	rep->rep.sector = sector;
+	rep->rep.nr_zones = ZONES_PER_REPORT;
+
+	if (ioctl(fd, BLKREPORTZONE, &rep->rep)) {
+		fprintf(stderr, "%s %s\n",
+_("ioctl(BLKREPORTZONE) failed:\n"),
+			strerror(-errno));
+		free(rep);
+		return NULL;
+	}
+
+	return rep;
+}
libfrog/zones.h new file +18 -0
new file mode 100644
index 000000000000..33c1da7ef192
@@ -0,0 +1,18 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2025, Western Digital Corporation or its affiliates.
+ */
+#ifndef __LIBFROG_ZONES_H__
+#define __LIBFROG_ZONES_H__
+
+#include <linux/blkzoned.h>
+
+struct xfrog_zone_report {
+	struct blk_zone_report	rep;
+	struct blk_zone		zones[];
+};
+
+struct xfrog_zone_report *
+xfrog_report_zones(int fd, uint64_t sector, struct xfrog_zone_report *rep);
+
+#endif /* __LIBFROG_ZONES_H__ */
mkfs/xfs_mkfs.c +12 -29
index 215cff8db7b1..9c165f29c298 100644
@@ -15,6 +15,7 @@
 #include "libfrog/crc32cselftest.h"
 #include "libfrog/dahashselftest.h"
 #include "libfrog/fsproperties.h"
+#include "libfrog/zones.h"
 #include "proto.h"
 #include <ini.h>
 
@@ -2542,9 +2543,6 @@ struct zone_topology {
 	struct zone_info	log;
 };
 
-/* random size that allows efficient processing */
-#define ZONES_PER_IOCTL			16384
-
 static void
 zone_validate_capacity(
 	struct zone_info	*zi,
@@ -2572,12 +2570,11 @@ report_zones(
 	const char		*name,
 	struct zone_info	*zi)
 {
-	struct blk_zone_report	*rep;
+	struct xfrog_zone_report *rep = NULL;
 	bool			found_seq = false;
-	int			fd, ret = 0;
+	int			fd;
 	uint64_t		device_size;
 	uint64_t		sector = 0;
-	size_t			rep_size;
 	unsigned int		i, n = 0;
 	struct stat		st;
 
@@ -2604,32 +2601,18 @@ report_zones(
 	zi->nr_zones = device_size / zi->zone_size;
 	zi->nr_conv_zones = 0;
 
-	rep_size = sizeof(struct blk_zone_report) +
-		   sizeof(struct blk_zone) * ZONES_PER_IOCTL;
-	rep = malloc(rep_size);
-	if (!rep) {
-		fprintf(stderr,
-_("Failed to allocate memory for zone reporting.\n"));
-		exit(1);
-	}
-
 	while (n < zi->nr_zones) {
-		struct blk_zone *zones = (struct blk_zone *)(rep + 1);
+		struct blk_zone *zones;
 
-		memset(rep, 0, rep_size);
-		rep->sector = sector;
-		rep->nr_zones = ZONES_PER_IOCTL;
-
-		ret = ioctl(fd, BLKREPORTZONE, rep);
-		if (ret) {
-			fprintf(stderr,
-_("ioctl(BLKREPORTZONE) failed: %d!\n"), -errno);
+		rep = xfrog_report_zones(fd, sector, rep);
+		if (!rep)
 			exit(1);
-		}
-		if (!rep->nr_zones)
+
+		if (!rep->rep.nr_zones)
 			break;
 
-		for (i = 0; i < rep->nr_zones; i++) {
+		zones = rep->zones;
+		for (i = 0; i < rep->rep.nr_zones; i++) {
 			if (n >= zi->nr_zones)
 				break;
 
@@ -2676,8 +2659,8 @@ _("Unknown zone type (0x%x) found.\n"), zones[i].type);
 
 			n++;
 		}
-		sector = zones[rep->nr_zones - 1].start +
-			 zones[rep->nr_zones - 1].len;
+		sector = zones[rep->rep.nr_zones - 1].start +
+			 zones[rep->rep.nr_zones - 1].len;
 	}
 
 	free(rep);
repair/zoned.c +11 -24
index 206b0158f95f..5102d43e218d 100644
@@ -7,6 +7,7 @@
 #include "libxfs_priv.h"
 #include "libxfs.h"
 #include "xfs_zones.h"
+#include "libfrog/zones.h"
 #include "err_protos.h"
 #include "zoned.h"
 
@@ -51,8 +52,7 @@ check_zones(
 	uint64_t		sector = XFS_FSB_TO_BB(mp, mp->m_sb.sb_rtstart);
 	unsigned int		zone_size, zone_capacity;
 	uint64_t		device_size;
-	size_t			rep_size;
-	struct blk_zone_report	*rep;
+	struct xfrog_zone_report *rep = NULL;
 	unsigned int		i, n = 0;
 
 	if (ioctl(fd, BLKGETSIZE64, &device_size))
@@ -67,31 +67,18 @@ check_zones(
 		return;
 	}
 
-	rep_size = sizeof(struct blk_zone_report) +
-		   sizeof(struct blk_zone) * ZONES_PER_IOCTL;
-	rep = malloc(rep_size);
-	if (!rep) {
-		do_warn(_("malloc failed for zone report\n"));
-		return;
-	}
-
 	while (n < mp->m_sb.sb_rgcount) {
-		struct blk_zone *zones = (struct blk_zone *)(rep + 1);
-		int ret;
+		struct blk_zone *zones;
 
-		memset(rep, 0, rep_size);
-		rep->sector = sector;
-		rep->nr_zones = ZONES_PER_IOCTL;
+		rep = xfrog_report_zones(fd, sector, rep);
+		if (!rep)
+			return;
 
-		ret = ioctl(fd, BLKREPORTZONE, rep);
-		if (ret) {
-			do_error(_("ioctl(BLKREPORTZONE) failed: %d!\n"), ret);
-			goto out_free;
-		}
-		if (!rep->nr_zones)
+		if (!rep->rep.nr_zones)
 			break;
 
-		for (i = 0; i < rep->nr_zones; i++) {
+		zones = rep->zones;
+		for (i = 0; i < rep->rep.nr_zones; i++) {
 			if (n >= mp->m_sb.sb_rgcount)
 				break;
 
@@ -130,8 +117,8 @@ _("Inconsistent zone capacity!\n"));
 			report_zones_cb(mp, &zones[i]);
 			n++;
 		}
-		sector = zones[rep->nr_zones - 1].start +
-			 zones[rep->nr_zones - 1].len;
+		sector = zones[rep->rep.nr_zones - 1].start +
+			 zones[rep->rep.nr_zones - 1].len;
 	}
 
 out_free:
Signature · 1 line
2.47.3
#13 lore raw

PATCH 4/4 libfrog: enable cached report zones

In reply to Christoph Hellwig

From: Damien Le Moal <[email protected]>

Modify the function xfrog_report_zones() to default to always trying first a cached report zones using the BLKREPORTZONEV2 ioctl. If the kernel does not support BLKREPORTZONEV2, fall back to the (slower) regular report zones BLKREPORTZONE ioctl.

TO enable this feature even if xfsprogs is compiled on a system where linux/blkzoned.h does not define BLKREPORTZONEV2, this ioctl is defined in libfrog/zones.h, together with the BLK_ZONE_REP_CACHED flag and the BLK_ZONE_COND_ACTIVE zone condition.

Since a cached report zone always return the condition BLK_ZONE_COND_ACTIVE for any zone that is implicitly open, explicitly open or closed, the function xfs_zone_validate_seq() is modified to handle this new condition as being equivalent to the implicit open, explicit open or closed conditions.

[hch: don't try cached reporting again if not supported]
---
 libfrog/zones.c | 31 ++++++++++++++++++++++++++++++-
 1 file changed, 30 insertions(+), 1 deletion(-)
libfrog/zones.c +30 -1
index 2276c56bec9c..f1ef0b24c564 100644
@@ -3,12 +3,24 @@
  * Copyright (c) 2025, Western Digital Corporation or its affiliates.
  */
 #include "platform_defs.h"
+#include "atomic.h"
 #include "libfrog/zones.h"
 #include <sys/ioctl.h>
 
+/*
+ * Cached report ioctl (/usr/include/linux/blkzoned.h).
+ * Add in Linux 6.19.
+ */
+#ifndef BLKREPORTZONEV2
+#define BLKREPORTZONEV2		_IOWR(0x12, 142, struct blk_zone_report)
+#define BLK_ZONE_REP_CACHED	(1U << 31)
+#endif /* BLKREPORTZONEV2 */
+
 /* random size that allows efficient processing */
 #define ZONES_PER_REPORT		16384
 
+static atomic_t cached_reporting_disabled;
+
 struct xfrog_zone_report *
 xfrog_report_zones(
 	int			fd,
@@ -24,10 +36,27 @@ _("Failed to allocate memory for reporting zones."));
 		}
 	}
 
+	/*
+	 * Try cached report zones first. If this fails, fallback to the regular
+	 * (slower) report zones.
+	 */
 	rep->rep.sector = sector;
 	rep->rep.nr_zones = ZONES_PER_REPORT;
 
-	if (ioctl(fd, BLKREPORTZONE, &rep->rep)) {
+	if (atomic_read(&cached_reporting_disabled))
+		goto uncached;
+
+	rep->rep.flags = BLK_ZONE_REP_CACHED;
+	if (ioctl(fd, BLKREPORTZONEV2, &rep->rep)) {
+		atomic_inc(&cached_reporting_disabled);
+		goto uncached;
+	}
+
+	return rep;
+
+uncached:
+	rep->rep.flags = 0;
+	if (ioctl(fd, BLKREPORTZONE, rep)) {
 		fprintf(stderr, "%s %s\n",
 _("ioctl(BLKREPORTZONE) failed:\n"),
 			strerror(-errno));
Signature · 1 line
2.47.3
Reviewed-by on v1 4/4
#14 lore raw

Re: [PATCH 4/4] libfrog: enable cached report zones

Quoted 18 lines · Christoph Hellwig

From: Damien Le Moal <[email protected]>

Modify the function xfrog_report_zones() to default to always trying first a cached report zones using the BLKREPORTZONEV2 ioctl. If the kernel does not support BLKREPORTZONEV2, fall back to the (slower) regular report zones BLKREPORTZONE ioctl.

TO enable this feature even if xfsprogs is compiled on a system where linux/blkzoned.h does not define BLKREPORTZONEV2, this ioctl is defined in libfrog/zones.h, together with the BLK_ZONE_REP_CACHED flag and the BLK_ZONE_COND_ACTIVE zone condition.

Since a cached report zone always return the condition BLK_ZONE_COND_ACTIVE for any zone that is implicitly open, explicitly open or closed, the function xfs_zone_validate_seq() is modified to handle this new condition as being equivalent to the implicit open, explicit open or closed conditions.

Quoted 3 lines
[hch: don't try cached reporting again if not supported]

That's an amusing use of atomic_t :)

--D

Quoted 66 lines
---
 libfrog/zones.c | 31 ++++++++++++++++++++++++++++++-
 1 file changed, 30 insertions(+), 1 deletion(-)
libfrog/zones.c +30 -1
index 2276c56bec9c..f1ef0b24c564 100644
@@ -3,12 +3,24 @@
  * Copyright (c) 2025, Western Digital Corporation or its affiliates.
  */
 #include "platform_defs.h"
+#include "atomic.h"
 #include "libfrog/zones.h"
 #include <sys/ioctl.h>
 
+/*
+ * Cached report ioctl (/usr/include/linux/blkzoned.h).
+ * Add in Linux 6.19.
+ */
+#ifndef BLKREPORTZONEV2
+#define BLKREPORTZONEV2		_IOWR(0x12, 142, struct blk_zone_report)
+#define BLK_ZONE_REP_CACHED	(1U << 31)
+#endif /* BLKREPORTZONEV2 */
+
 /* random size that allows efficient processing */
 #define ZONES_PER_REPORT		16384
 
+static atomic_t cached_reporting_disabled;
+
 struct xfrog_zone_report *
 xfrog_report_zones(
 	int			fd,
@@ -24,10 +36,27 @@ _("Failed to allocate memory for reporting zones."));
 		}
 	}
 
+	/*
+	 * Try cached report zones first. If this fails, fallback to the regular
+	 * (slower) report zones.
+	 */
 	rep->rep.sector = sector;
 	rep->rep.nr_zones = ZONES_PER_REPORT;
 
-	if (ioctl(fd, BLKREPORTZONE, &rep->rep)) {
+	if (atomic_read(&cached_reporting_disabled))
+		goto uncached;
+
+	rep->rep.flags = BLK_ZONE_REP_CACHED;
+	if (ioctl(fd, BLKREPORTZONEV2, &rep->rep)) {
+		atomic_inc(&cached_reporting_disabled);
+		goto uncached;
+	}
+
+	return rep;
+
+uncached:
+	rep->rep.flags = 0;
+	if (ioctl(fd, BLKREPORTZONE, rep)) {
 		fprintf(stderr, "%s %s\n",
 _("ioctl(BLKREPORTZONE) failed:\n"),
 			strerror(-errno));
Signature · 1 line
2.47.3
#15 lore raw
In reply to Christoph Hellwig
Quoted 6 lines · Christoph Hellwig

Enable cached zone report to speed up mkfs and repair on a zoned block device (e.g. an SMR disk). Cached zone report support was introduced in the kernel with version 6.19-rc1.

Note: I've taken this series over from Damien as he's busy and it need it as a base for further work.

Thanks for doing this !

Quoted 21 lines

Changes from v3:
- reorder includes to not need the forward declaration in xfs_zones.h
- use the libfrog/ for #include statements
- fix the include guard in libfrog/zones.h
- fix up i18n string mess
- hide the new ioctl definition in libfrog/zones.c
- don't add userspace includes to libxfs/xfs_zones.h
- reuse the buffer for multiple report zone calls

Changes from v2:
- Complete rework of the series to make the zone reporting code common in libfrog
- Added patch 1 and 2 as small cleanups/improvements.

Changes from v1:
- Fix erroneous handling of ioctl(BLKREPORTZONEV2) error to correctly fallback to the regular ioctl(BLKREPORTZONE) if the kernel does not support BLKREPORTZONEV2.

Signature · 2 lines
Damien Le Moal
Western Digital Research