Skip to contentSkip to conversation

Zoned XFS code improvements

PATCH Applied Damien Le Moal 4 participants 14 messages Mar 16 – Mar 18 XFS filesystem

lore
Share & tools
lore
#0 lore raw

v2 cover Zoned XFS code improvements

All,

Here are 4 simple patches to cleanup the code for zoned XFS. Overall, there is no functional changes introduced beside the change of information displayed in /proc/self/mountstats (patch 3).

Changes from v1:
- Split former patch 1 into patch 1 and 2.
- Changed patch 3 to add new lines of information instead of changing the existing lines.
- Reworked commit titles and messages.

Damien Le Moal (4):
xfs: avoid unnecessary open zone check in xfs_select_zone_nowait()
xfs: fix a comment typo in xfs_select_zone_nowait()
xfs: display more zone related information in mountstats
xfs: avoid unnecessary calculations in xfs_zoned_need_gc()

 fs/xfs/xfs_zone_alloc.c |  9 +++++----
 fs/xfs/xfs_zone_gc.c    | 24 ++++++++++++++++++------
 fs/xfs/xfs_zone_info.c  |  7 ++++++-
 3 files changed, 29 insertions(+), 11 deletions(-)

base-commit: f8544b654f22b1138ba12bc0971a96963b20311d

Signature · 1 line
2.53.0
#1 lore raw

PATCH v2 1/4 xfs: avoid unnecessary open zone check in xfs_select_zone_nowait()

When xfs_select_zone_nowait() is called with pack_tight equal to true, the function xfs_select_open_zone_mru() is called if no open zone is returned by xfs_select_open_zone_lru(), that is, when oz is NULL. The open zone pointer return of xfs_select_zone_nowait() is then checked, but this check is outside of the "if (pack_tight)" that trigered the call to xfs_select_open_zone_mru(). In other word, this check is unnecessarily done even when pack_tight is false.

Move the check for the return value of the call to xfs_select_open_zone_mru() inside the if that controls the call to this function, so that we do not uselessly test again the value of oz when pack_tight is false.

No functional changes.

---
 fs/xfs/xfs_zone_alloc.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
fs/xfs/xfs_zone_alloc.c +4 -3
index 7d712d5a5ce0..8435ccb018dc 100644
@@ -675,10 +675,11 @@ xfs_select_zone_nowait(
 	if (oz)
 		goto out_unlock;
 
-	if (pack_tight)
+	if (pack_tight) {
 		oz = xfs_select_open_zone_mru(zi, write_hint);
-	if (oz)
-		goto out_unlock;
+		if (oz)
+			goto out_unlock;
+	}
 
 	/*
 	 * See if we can open a new zone and use that so that data for different
Signature · 1 line
2.53.0
Reviewed-by on v2 1/4
#3 lore raw

Re: [PATCH v2 1/4] xfs: avoid unnecessary open zone check in xfs_select_zone_nowait()

In reply to Damien Le Moal
Quoted 29 lines · Damien Le Moal

When xfs_select_zone_nowait() is called with pack_tight equal to true, the function xfs_select_open_zone_mru() is called if no open zone is returned by xfs_select_open_zone_lru(), that is, when oz is NULL. The open zone pointer return of xfs_select_zone_nowait() is then checked, but this check is outside of the "if (pack_tight)" that trigered the call to xfs_select_open_zone_mru(). In other word, this check is unnecessarily done even when pack_tight is false.

Move the check for the return value of the call to xfs_select_open_zone_mru() inside the if that controls the call to this function, so that we do not uselessly test again the value of oz when pack_tight is false.

No functional changes.

---
 fs/xfs/xfs_zone_alloc.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
fs/xfs/xfs_zone_alloc.c +0 -1
index 7d712d5a5ce0..8435ccb018dc 100644
@@ -675,10 +675,11 @@ xfs_select_zone_nowait(
 	if (oz)
 		goto out_unlock;
 
-	if (pack_tight)

fs/xfs/xfs_zone_alloc.c xfs_select_zone_nowait( · line 685

Quoted 10 lines
fs/xfs/xfs_zone_alloc.c +4 -2
+	if (pack_tight) {
 		oz = xfs_select_open_zone_mru(zi, write_hint);
-	if (oz)
-		goto out_unlock;
+		if (oz)
+			goto out_unlock;
+	}
 
 	/*
 	 * See if we can open a new zone and use that so that data for different

Looks good,

#4 lore raw

PATCH v2 2/4 xfs: fix a comment typo in xfs_select_zone_nowait()

In reply to Damien Le Moal

Fix a typo in the comment describing the second call to xfs_select_open_zone_lru() in xfs_select_zone_nowait().

---
 fs/xfs/xfs_zone_alloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/xfs/xfs_zone_alloc.c +1 -1
index 8435ccb018dc..a05418ac6a95 100644
@@ -690,7 +690,7 @@ xfs_select_zone_nowait(
 		goto out_unlock;
 
 	/*
-	 * Try to find an zone that is an ok match to colocate data with.
+	 * Try to find a zone that is an ok match to colocate data with.
 	 */
 	oz = xfs_select_open_zone_lru(zi, write_hint, XFS_ZONE_ALLOC_OK);
 	if (oz)
Signature · 1 line
2.53.0
Reviewed-by on v2 2/4
#6 lore raw

Re: [PATCH v2 2/4] xfs: fix a comment typo in xfs_select_zone_nowait()

In reply to Damien Le Moal
Quoted 13 lines · Damien Le Moal

Fix a typo in the comment describing the second call to xfs_select_open_zone_lru() in xfs_select_zone_nowait().

---
 fs/xfs/xfs_zone_alloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/xfs/xfs_zone_alloc.c +0 -0
index 8435ccb018dc..a05418ac6a95 100644

fs/xfs/xfs_zone_alloc.c xfs_select_zone_nowait( · line 696

Quoted 8 lines
fs/xfs/xfs_zone_alloc.c +1 -1
@@ -690,7 +690,7 @@ xfs_select_zone_nowait(
 		goto out_unlock;
 
 	/*
-	 * Try to find an zone that is an ok match to colocate data with.
+	 * Try to find a zone that is an ok match to colocate data with.
 	 */
 	oz = xfs_select_open_zone_lru(zi, write_hint, XFS_ZONE_ALLOC_OK);
 	if (oz)

Looks good,

#7 lore raw

PATCH v2 3/4 xfs: display more zone related information in mountstats

In reply to Damien Le Moal

Modify xfs_zoned_show_stats() to add to the information displayed with /proc/self/mountstats the total number of zones (RT groups) and the number of open zones together with the maximum number of open zones.

---
 fs/xfs/xfs_zone_info.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
fs/xfs/xfs_zone_info.c +6 -1
index 53eabbc3334c..a2af44011654 100644
@@ -90,9 +90,14 @@ xfs_zoned_show_stats(
 	seq_printf(m, "\tRT GC required: %d\n",
 		xfs_zoned_need_gc(mp));
 
+	seq_printf(m, "\ttotal number of zones: %u\n",
+		mp->m_sb.sb_rgcount);
 	seq_printf(m, "\tfree zones: %d\n", atomic_read(&zi->zi_nr_free_zones));
-	seq_puts(m, "\topen zones:\n");
+
 	spin_lock(&zi->zi_open_zones_lock);
+	seq_printf(m, "\tnumber of open zones: %u / %u\n",
+		zi->zi_nr_open_zones, mp->m_max_open_zones);
+	seq_puts(m, "\topen zones:\n");
 	list_for_each_entry(oz, &zi->zi_open_zones, oz_entry)
 		xfs_show_open_zone(m, oz);
 	if (zi->zi_open_gc_zone) {
Signature · 1 line
2.53.0
Reviewed-by on v2 3/4
#9 lore raw

Re: [PATCH v2 3/4] xfs: display more zone related information in mountstats

In reply to Damien Le Moal
Quoted 19 lines · Damien Le Moal

Modify xfs_zoned_show_stats() to add to the information displayed with /proc/self/mountstats the total number of zones (RT groups) and the number of open zones together with the maximum number of open zones.

---
 fs/xfs/xfs_zone_info.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
fs/xfs/xfs_zone_info.c +2 -0
index 53eabbc3334c..a2af44011654 100644
@@ -90,9 +90,14 @@ xfs_zoned_show_stats(
 	seq_printf(m, "\tRT GC required: %d\n",
 		xfs_zoned_need_gc(mp));
 
+	seq_printf(m, "\ttotal number of zones: %u\n",
+		mp->m_sb.sb_rgcount);

fs/xfs/xfs_zone_info.c xfs_zoned_show_stats( · line 103

Quoted 10 lines
fs/xfs/xfs_zone_info.c +4 -1
 	seq_printf(m, "\tfree zones: %d\n", atomic_read(&zi->zi_nr_free_zones));
-	seq_puts(m, "\topen zones:\n");
+
 	spin_lock(&zi->zi_open_zones_lock);
+	seq_printf(m, "\tnumber of open zones: %u / %u\n",
+		zi->zi_nr_open_zones, mp->m_max_open_zones);
+	seq_puts(m, "\topen zones:\n");
 	list_for_each_entry(oz, &zi->zi_open_zones, oz_entry)
 		xfs_show_open_zone(m, oz);
 	if (zi->zi_open_gc_zone) {

Awesome, i've been wanting to have the current number of open zones as part of these stats for a while :)

#10 lore raw

PATCH v2 4/4 xfs: avoid unnecessary calculations in xfs_zoned_need_gc()

In reply to Damien Le Moal

If zonegc_low_space is set to zero (which is the default), the second condition in xfs_zoned_need_gc() that triggers GC never evaluates to true because the calculated threshold will always be 0. So there is no need to calculate the threshold and to evaluate that condition. Return early when zonegc_low_space is zero.

While at it, add comments to document the intent of each of the 3 tests used to determine the return value to control the execution of garbage collection.

---
 fs/xfs/xfs_zone_gc.c | 24 ++++++++++++++++++------
 1 file changed, 18 insertions(+), 6 deletions(-)
fs/xfs/xfs_zone_gc.c +18 -6
index 309f70098524..0ff710fa0ee7 100644
@@ -170,25 +170,37 @@ xfs_zoned_need_gc(
 	s64			available, free, threshold;
 	s32			remainder;
 
+	/* If we have no reclaimable blocks, running GC is useless. */
 	if (!xfs_zoned_have_reclaimable(mp->m_zone_info))
 		return false;
 
+	/*
+	 * In order to avoid file fragmentation as much as possible, we should
+	 * make sure that we can open enough zones. So trigger GC if the number
+	 * of blocks immediately available for writes is lower than the total
+	 * number of blocks from all possible open zones.
+	 */
 	available = xfs_estimate_freecounter(mp, XC_FREE_RTAVAILABLE);
-
 	if (available <
 	    xfs_rtgs_to_rfsbs(mp, mp->m_max_open_zones - XFS_OPEN_GC_ZONES))
 		return true;
 
-	free = xfs_estimate_freecounter(mp, XC_FREE_RTEXTENTS);
+	/*
+	 * For cases where the user wants to be more aggressive with GC,
+	 * the sysfs attribute zonegc_low_space may be set to a non zero value,
+	 * to indicate that GC should try to maintain at least zonegc_low_space
+	 * percent of the free space to be directly available for writing. Check
+	 * this here.
+	 */
+	if (!mp->m_zonegc_low_space)
+		return false;
 
+	free = xfs_estimate_freecounter(mp, XC_FREE_RTEXTENTS);
 	threshold = div_s64_rem(free, 100, &remainder);
 	threshold = threshold * mp->m_zonegc_low_space +
 		    remainder * div_s64(mp->m_zonegc_low_space, 100);
 
-	if (available < threshold)
-		return true;
-
-	return false;
+	return available < threshold;
 }
 
 static struct xfs_zone_gc_data *
Signature · 1 line
2.53.0
Reviewed-by on v2 4/4
#11 lore raw

Re: [PATCH v2 4/4] xfs: avoid unnecessary calculations in xfs_zoned_need_gc()

Quoted 9 lines · Damien Le Moal

If zonegc_low_space is set to zero (which is the default), the second condition in xfs_zoned_need_gc() that triggers GC never evaluates to true because the calculated threshold will always be 0. So there is no need to calculate the threshold and to evaluate that condition. Return early when zonegc_low_space is zero.

While at it, add comments to document the intent of each of the 3 tests used to determine the return value to control the execution of garbage collection.

Looks good:

Reviewed-by on v2 4/4
#12 lore raw

Re: [PATCH v2 4/4] xfs: avoid unnecessary calculations in xfs_zoned_need_gc()

In reply to Damien Le Moal
Quoted 53 lines · Damien Le Moal

If zonegc_low_space is set to zero (which is the default), the second condition in xfs_zoned_need_gc() that triggers GC never evaluates to true because the calculated threshold will always be 0. So there is no need to calculate the threshold and to evaluate that condition. Return early when zonegc_low_space is zero.

While at it, add comments to document the intent of each of the 3 tests used to determine the return value to control the execution of garbage collection.

---
 fs/xfs/xfs_zone_gc.c | 24 ++++++++++++++++++------
 1 file changed, 18 insertions(+), 6 deletions(-)
fs/xfs/xfs_zone_gc.c +17 -2
index 309f70098524..0ff710fa0ee7 100644
@@ -170,25 +170,37 @@ xfs_zoned_need_gc(
 	s64			available, free, threshold;
 	s32			remainder;
 
+	/* If we have no reclaimable blocks, running GC is useless. */
 	if (!xfs_zoned_have_reclaimable(mp->m_zone_info))
 		return false;
 
+	/*
+	 * In order to avoid file fragmentation as much as possible, we should
+	 * make sure that we can open enough zones. So trigger GC if the number
+	 * of blocks immediately available for writes is lower than the total
+	 * number of blocks from all possible open zones.
+	 */
 	available = xfs_estimate_freecounter(mp, XC_FREE_RTAVAILABLE);
-
 	if (available <
 	    xfs_rtgs_to_rfsbs(mp, mp->m_max_open_zones - XFS_OPEN_GC_ZONES))
 		return true;
 
-	free = xfs_estimate_freecounter(mp, XC_FREE_RTEXTENTS);
+	/*
+	 * For cases where the user wants to be more aggressive with GC,
+	 * the sysfs attribute zonegc_low_space may be set to a non zero value,
+	 * to indicate that GC should try to maintain at least zonegc_low_space
+	 * percent of the free space to be directly available for writing. Check
+	 * this here.
+	 */
+	if (!mp->m_zonegc_low_space)
+		return false;
 
+	free = xfs_estimate_freecounter(mp, XC_FREE_RTEXTENTS);
 	threshold = div_s64_rem(free, 100, &remainder);
 	threshold = threshold * mp->m_zonegc_low_space +

fs/xfs/xfs_zone_gc.c xfs_zoned_need_gc( · line 206

Quoted 10 lines
fs/xfs/xfs_zone_gc.c +1 -4
 		    remainder * div_s64(mp->m_zonegc_low_space, 100);
 
-	if (available < threshold)
-		return true;
-
-	return false;
+	return available < threshold;
 }
 
 static struct xfs_zone_gc_data *

The optimization makes sense & thanks for providing the documentation!

maintainer Applied
#13 lore raw
In reply to Damien Le Moal
Quoted 6 lines · Damien Le Moal

All,

Here are 4 simple patches to cleanup the code for zoned XFS. Overall, there is no functional changes introduced beside the change of information displayed in /proc/self/mountstats (patch 3).

Quoted 7 lines

Changes from v1:
- Split former patch 1 into patch 1 and 2.
- Changed patch 3 to add new lines of information instead of changing the existing lines.
- Reworked commit titles and messages.

[...]

Applied to for-next, thanks!

[1/4] xfs: avoid unnecessary open zone check in xfs_select_zone_nowait()

      commit: 770323d418ed5848cc21af172f77377b2cc0542d
[2/4] xfs: fix a comment typo in xfs_select_zone_nowait()
      commit: 6a82a691b08070ad03b237d7db89aa0bfef389e2
[3/4] xfs: display more zone related information in mountstats
      commit: 68aa101bf2046aa8365333a3768cece07975ca5f
[4/4] xfs: avoid unnecessary calculations in xfs_zoned_need_gc()
      commit: c1f955437440f92632e2efca4b591371bb3caefc

Best regards,

Signature · 1 line
Carlos Maiolino <[email protected]>