Ramping up data preallocation

Ramps up data preallocation based on the number of online
blocks. This results in a simple 2<<n block allocation pattern
until n=11 (2048) - the default value of data_prealloc_blocks.

Signed-off-by: Auke Kok <auke.kok@versity.com>

ReleaseNotes.md

@@ -1,23 +1,6 @@
 Versity ScoutFS Release Notes
 =============================
 
----
-v1.30
-\
-*Apr 21, 2026*
-
-Fix a problem reading the accumulated totals of contributing .totl.
-xattrs when log merging is in progress.  The problem would have readers
-of the totals calculate the sums incorrectly.
-
-Fix a problem updating quota rules.  There was a race where updates
-could be corrupted if they happened while a transaction was being
-written.
-
-Fix a problem deleting files with .indx. xattrs.  The internal indexing
-metadata wouldn't be properly deleted so the files would still claim to
-be present and visible in the index, though the file no longer existed.
-
 ---
 v1.29
 \

kmod/scoutfs-kmod.spec.in

@@ -4,15 +4,9 @@
 %define kmod_git_describe @@GITDESCRIBE@@
 %define pkg_date %(date +%%Y%%m%%d)
 
-# Package type: set --define 'per_kver 1' to tie the package to a specific kernel
-# version (per-kver mode); leave unset for the default per-minor-release behavior.
-
 # take kernel version or default to uname -r
 %{!?kversion: %global kversion %(uname -r)}
 %global kernel_version %{kversion}
-%if 0%{?per_kver}
-%define sanitized_kernel_version %(echo %{kernel_version} | tr - _ |sed -e 's/.x86_64//')
-%endif
 
 %if 0%{?el7}
 %global kernel_source() /usr/src/kernels/%{kernel_version}.$(arch)
@@ -23,34 +17,16 @@
 %{!?_release: %global _release 0.%{pkg_date}git%{kmod_git_hash}}
 
 %if 0%{?el7}
-%if 0%{?per_kver}
-Name:           %{kmod_name}-%{sanitized_kernel_version}
-Provides:       %{kmod_name} = %{kmod_version}
-%else
 Name:           %{kmod_name}
-%endif
-%else
-%if 0%{?per_kver}
-Name:           kmod-%{kmod_name}-%{sanitized_kernel_version}
-Provides:       kmod-%{kmod_name} = %{kmod_version}
 %else
 Name:           kmod-%{kmod_name}
 %endif
-%endif
 Summary:        %{kmod_name} kernel module
 Version:        %{kmod_version}
-%if 0%{?per_kver}
-Release:        %{_release}
-%else
 Release:        %{_release}%{?dist}
-%endif
 License:        GPLv2
 Group:          System/Kernel
 URL:            http://scoutfs.org/
-%if 0%{?per_kver}
-Requires:       kernel-core-uname-r = %{kernel_version}
-Requires:       kernel-modules-uname-r = %{kernel_version}
-%endif
 
 %if 0%{?el7}
 BuildRequires:  %{kernel_module_package_buildreqs}

kmod/src/data.c

@@ -384,6 +384,12 @@ static inline u64 ext_last(struct scoutfs_extent *ext)
  * This can waste a lot of space for small or sparse files but is
  * reasonable when a file population is known to be large and dense but
  * known to be written with non-streaming write patterns.
+ *
+ * In either strategy, preallocation ramps up proportionally with the
+ * file's online block count rather than jumping to the full prealloc
+ * size.  This avoids overallocation for small files in mixed-size
+ * workloads while still allowing large files to benefit from full
+ * preallocation.
  */
 static int alloc_block(struct super_block *sb, struct inode *inode,
 		       struct scoutfs_extent *ext, u64 iblock,
@@ -400,6 +406,7 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
 	struct scoutfs_extent found;
 	struct scoutfs_extent pre = {0,};
 	bool undo_pre = false;
+	bool have_onoff = false;
 	u64 blkno = 0;
 	u64 online;
 	u64 offline;
@@ -445,6 +452,7 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
 		 * blocks.
 		 */
 		scoutfs_inode_get_onoff(inode, &online, &offline);
+		have_onoff = true;
 		if (iblock > 1 && iblock == online) {
 			ret = scoutfs_ext_next(sb, &data_ext_ops, &args,
 					       iblock, 1, &found);
@@ -491,6 +499,16 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
 	/* overall prealloc limit */
 	count = min_t(u64, count, opts.data_prealloc_blocks);
 
+	/*
+	 * Ramp preallocation up proportionally with the file's online
+	 * block count rather than jumping to the full prealloc size.
+	 */
+	if (!ext->len) {
+		if (!have_onoff)
+			scoutfs_inode_get_onoff(inode, &online, &offline);
+		count = max_t(u64, 1, min(count, online));
+	}
+
 	ret = scoutfs_alloc_data(sb, datinf->alloc, datinf->wri,
 				 &datinf->dalloc, count, &blkno, &count);
 	if (ret < 0)

tests/golden/data-prealloc

@@ -8,10 +8,10 @@
 /mnt/test/test/data-prealloc/file-1: extents: 32
 /mnt/test/test/data-prealloc/file-2: extents: 32
 == any writes to region prealloc get full extents
-/mnt/test/test/data-prealloc/file-1: extents: 4
-/mnt/test/test/data-prealloc/file-2: extents: 4
-/mnt/test/test/data-prealloc/file-1: extents: 4
-/mnt/test/test/data-prealloc/file-2: extents: 4
+/mnt/test/test/data-prealloc/file-1: extents: 8
+/mnt/test/test/data-prealloc/file-2: extents: 8
+/mnt/test/test/data-prealloc/file-1: extents: 8
+/mnt/test/test/data-prealloc/file-2: extents: 8
 == streaming offline writes get full extents either way
 /mnt/test/test/data-prealloc/file-1: extents: 4
 /mnt/test/test/data-prealloc/file-2: extents: 4
@@ -20,8 +20,8 @@
 == goofy preallocation amounts work
 /mnt/test/test/data-prealloc/file-1: extents: 6
 /mnt/test/test/data-prealloc/file-2: extents: 6
-/mnt/test/test/data-prealloc/file-1: extents: 6
-/mnt/test/test/data-prealloc/file-2: extents: 6
+/mnt/test/test/data-prealloc/file-1: extents: 10
+/mnt/test/test/data-prealloc/file-2: extents: 10
 /mnt/test/test/data-prealloc/file-1: extents: 3
 /mnt/test/test/data-prealloc/file-2: extents: 3
 == block writes into region allocs hole