zfs-2.4.2 patchset

.github/workflows/checkstyle.yaml

@@ -56,9 +56,10 @@ jobs:
     - name: Prepare artifacts
       if: failure() && steps.CheckABI.outcome == 'failure'
       run: |
-        find -name *.abi | tar -cf abi_files.tar -T -
-    - uses: actions/upload-artifact@v4
+        find -name *.abi | tar -cjf abi_files.tar.bz2 -T -
+    - uses: actions/upload-artifact@v7
       if: failure() && steps.CheckABI.outcome == 'failure'
       with:
         name: New ABI files (use only if you're sure about interface changes)
-        path: abi_files.tar
+        path: abi_files.tar.bz2
+        archive: false

.github/workflows/codeql.yml

@@ -28,7 +28,7 @@ jobs:
         echo "MAKEFLAGS=-j$(nproc)" >> $GITHUB_ENV
 
     - name: Checkout repository
-      uses: actions/checkout@v4
+      uses: actions/checkout@v6
 
     - name: Initialize CodeQL
       uses: github/codeql-action/init@v3

.github/workflows/scripts/generate-ci-type.py

@@ -3,13 +3,16 @@
 """
 Determine the CI type based on the change list and commit message.
 
-Prints "quick" if (explicity required by user):
+Output format: "<type> <source>" where source is "manual" (from
+ZFS-CI-Type commit tag) or "auto" (from file change heuristics).
+
+Prints "quick manual" if:
 - the *last* commit message contains 'ZFS-CI-Type: quick'
-or if (heuristics):
+or "quick auto" if (heuristics):
 - the files changed are not in the list of specified directories, and
 - all commit messages do not contain 'ZFS-CI-Type: (full|linux|freebsd)'
 
-Otherwise prints "full".
+Otherwise prints "full auto" (or "<type> manual" if explicitly requested).
 """
 
 import sys
@@ -58,9 +61,10 @@
 
     head, base = sys.argv[1:3]
 
-    def output_type(type, reason):
-        print(f'{prog}: will run {type} CI: {reason}', file=sys.stderr)
-        print(type)
+    def output_type(type, source, reason):
+        print(f'{prog}: will run {type} CI ({source}): {reason}',
+              file=sys.stderr)
+        print(f'{type} {source}')
         sys.exit(0)
 
     # check last (HEAD) commit message
@@ -70,7 +74,8 @@ def output_type(type, reason):
 
     for line in last_commit_message_raw.stdout.decode().splitlines():
         if line.strip().lower() == 'zfs-ci-type: quick':
-            output_type('quick', f'requested by HEAD commit {head}')
+            output_type('quick', 'manual',
+                        f'requested by HEAD commit {head}')
 
     # check all commit messages
     all_commit_message_raw = subprocess.run([
@@ -84,11 +89,14 @@ def output_type(type, reason):
         if line.startswith('ZFS-CI-Commit:'):
             commit_ref = line.lstrip('ZFS-CI-Commit:').rstrip()
         if line.strip().lower() == 'zfs-ci-type: freebsd':
-            output_type('freebsd', f'requested by commit {commit_ref}')
+            output_type('freebsd', 'manual',
+                        f'requested by commit {commit_ref}')
         if line.strip().lower() == 'zfs-ci-type: linux':
-            output_type('linux', f'requested by commit {commit_ref}')
+            output_type('linux', 'manual',
+                        f'requested by commit {commit_ref}')
         if line.strip().lower() == 'zfs-ci-type: full':
-            output_type('full', f'requested by commit {commit_ref}')
+            output_type('full', 'manual',
+                        f'requested by commit {commit_ref}')
 
     # check changed files
     changed_files_raw = subprocess.run([
@@ -104,9 +112,10 @@ def output_type(type, reason):
             for r in FULL_RUN_REGEX:
                 if r.match(f):
                     output_type(
-                        'full',
+                        'full', 'auto',
                         f'changed file "{f}" matches pattern "{r.pattern}"'
                         )
 
     # catch-all
-    output_type('quick', 'no changed file matches full CI patterns')
+    output_type('quick', 'auto',
+                'no changed file matches full CI patterns')

.github/workflows/scripts/merge_summary.awk

@@ -17,6 +17,7 @@ BEGIN {
 	pass=0
 	fail=0
 	skip=0
+	killed=0
 	state=""
 	cl=0
 	el=0
@@ -49,6 +50,37 @@ BEGIN {
 /PASS/{ if (state=="pass_count") {pass += $2}}
 /FAIL/{ if (state=="pass_count") {fail += $2}}
 /SKIP/{ if (state=="pass_count") {skip += $2}}
+
+# If the test was killed, you'll get a line like:
+#
+# [2026-04-22T03:34:17.694616] Test (Linux): /usr/share/zfs/zfs-tests/tests/functional/io/setup (run as root) [10:00] [KILLED]
+#
+# Parse out the test name minus the /usr/share/zfs/zfs-tests/tests/functional/'
+# part, and include the optional "(Linux): " line, as you can have the killed
+# tests in two categories, like:
+#
+#   KILLED (Linux): io/setup
+#   KILLED io/setup
+#
+/KILLED/{
+	extra=""
+	for(i=1; i<=NF; i++) {
+		# Look for optional "(Linux):" field
+		if ($i ~ "\\("){
+			extra=$i" "}
+
+		# Look for a field with a '/' in it.  It is the test name.
+		if($i ~ "/") {
+			testname=$i
+			# Remove /usr/share/zfs/zfs-test/test/functional string
+			sub(/\/usr\/share\/zfs\/zfs-tests\/tests\/functional\//,"",testname)
+			testname=extra""testname
+			killed_tests[killed] = testname
+			killed++
+			break
+		}
+	}
+}
 /Running Time/{
 	state="";
 	running[i]=$3;
@@ -106,4 +138,10 @@ END {
 	asort(unexpected_lines, sorted)
 	for (j in sorted)
 		print sorted[j]
+
+	# We don't want to sort killed tests, as the first test that was killed
+	# most likely caused the others to be killed.
+	print "\n\nTests that were killed:"
+	for (j in killed_tests)
+		print "    KILLED "killed_tests[j]
 }

.github/workflows/scripts/qemu-1-setup.sh

@@ -6,6 +6,27 @@
 
 set -eu
 
+# The default runner has a bunch of development tools and other things
+# that we do not need.  Remove them here to free up a total of 35GB.
+#
+# First remove packages - this frees up ~10GB
+echo "Disk space before purge:"
+df -h /
+sudo docker image prune --all --force
+sudo docker builder prune -a
+unneeded="microsoft-edge-stable|azure-cli|google-cloud|google-chrome-stable|"\
+"temurin|llvm|firefox|mysql-server|snapd|android|dotnet|haskell|ghcup|"\
+"powershell|julia|swift|miniconda|chromium"
+sudo apt-get -y remove $(dpkg-query -f '${binary:Package}\n' -W | grep -E "'$unneeded'")
+sudo apt-get -y autoremove
+
+# Next, remove unneeded files in /usr.  This frees up an additional 25GB.
+sudo rm -fr /usr/local/lib/android /usr/share/dotnet /usr/local/.ghcup \
+        /usr/share/swift /usr/local/share/powershell /usr/local/julia* \
+        /usr/share/miniconda /usr/local/share/chromium
+echo "Disk space after:"
+df -h /
+
 # The default 'azure.archive.ubuntu.com' mirrors can be really slow.
 # Prioritize the official Ubuntu mirrors.
 #
@@ -42,7 +63,7 @@ sudo swapoff -a
 # configurations.  On one config you get two 75GB block devices, and on the
 # other you get a single 150GB block device. Here's what both look like:
 #
-# --- Two 75GB block devices ---
+# --- One 150GB block device ---
 # NAME    MAJ:MIN RM  SIZE RO TYPE MOUNTPOINTS
 # sda       8:0    0  150G  0 disk
 # ├─sda1    8:1    0  149G  0 part /
@@ -56,7 +77,7 @@ sudo swapoff -a
 # lrwxrwxrwx 1 root root 11 Jan 29 18:07 azure_root-part15 -> ../../sda15
 # lrwxrwxrwx 1 root root 11 Jan 29 18:07 azure_root-part16 -> ../../sda16
 #
-# --- One 150GB block device ---
+# --- Two 75GB block devices ---
 # NAME    MAJ:MIN RM  SIZE RO TYPE MOUNTPOINTS
 # sda       8:0    0   75G  0 disk
 # ├─sda1    8:1    0   74G  0 part /
@@ -118,18 +139,20 @@ fi
 sudo mkswap $SWAP
 sudo swapon $SWAP
 
+echo "Block devices:"
+lsblk
+
 # adjust zfs module parameter and create pool
-exec 1>/dev/null
 ARC_MIN=$((1024*1024*256))
 ARC_MAX=$((1024*1024*512))
-echo $ARC_MIN | sudo tee /sys/module/zfs/parameters/zfs_arc_min
-echo $ARC_MAX | sudo tee /sys/module/zfs/parameters/zfs_arc_max
-echo 1 | sudo tee /sys/module/zfs/parameters/zvol_use_blk_mq
+echo $ARC_MIN | sudo tee /sys/module/zfs/parameters/zfs_arc_min >/dev/null
+echo $ARC_MAX | sudo tee /sys/module/zfs/parameters/zfs_arc_max >/dev/null
+echo 1 | sudo tee /sys/module/zfs/parameters/zvol_use_blk_mq >/dev/null
 sudo zpool create -f -o ashift=12 zpool $DISKS -O relatime=off \
   -O atime=off -O xattr=sa -O compression=lz4 -O sync=disabled \
   -O redundant_metadata=none -O mountpoint=/mnt/tests
+echo "Status:"
+zpool status
 
-# no need for some scheduler
-for i in /sys/block/s*/queue/scheduler; do
-  echo "none" | sudo tee $i
-done
+echo "Last dmesg:"
+sudo dmesg | tail -n 10

.github/workflows/scripts/qemu-2-start.sh

@@ -78,11 +78,6 @@ case "$OS" in
     OPTS[0]="--boot"
     OPTS[1]="uefi=on"
     ;;
-  fedora41)
-    OSNAME="Fedora 41"
-    OSv="fedora-unknown"
-    URL="https://download.fedoraproject.org/pub/fedora/linux/releases/41/Cloud/x86_64/images/Fedora-Cloud-Base-Generic-41-1.4.x86_64.qcow2"
-    ;;
   fedora42)
     OSNAME="Fedora 42"
     OSv="fedora-unknown"
@@ -101,8 +96,8 @@ case "$OS" in
     KSRC="$FREEBSD_REL/../amd64/$FreeBSD/src.txz"
     NIC="rtl8139"
     ;;
-  freebsd14-3r)
-    FreeBSD="14.3-RELEASE"
+  freebsd14-4r)
+    FreeBSD="14.4-RELEASE"
     OSNAME="FreeBSD $FreeBSD"
     OSv="freebsd14.0"
     URLxz="$FREEBSD_REL/$FreeBSD/amd64/Latest/FreeBSD-$FreeBSD-amd64-BASIC-CI.raw.xz"
@@ -116,8 +111,8 @@ case "$OS" in
     KSRC="$FREEBSD_SNAP/../amd64/$FreeBSD/src.txz"
     NIC="rtl8139"
     ;;
-  freebsd14-3s)
-    FreeBSD="14.3-STABLE"
+  freebsd14-4s)
+    FreeBSD="14.4-STABLE"
     OSNAME="FreeBSD $FreeBSD"
     OSv="freebsd14.0"
     URLxz="$FREEBSD_SNAP/$FreeBSD/amd64/Latest/FreeBSD-$FreeBSD-amd64-BASIC-CI-ufs.raw.xz"
@@ -193,17 +188,49 @@ DISK="/dev/zvol/zpool/openzfs"
 sudo zfs create -ps -b 64k -V 80g zpool/openzfs
 while true; do test -b $DISK && break; sleep 1; done
 
-# we are downloading via axel, curl and wget are mostly slower and
-# require more return value checking
+# We first try to download with 'axel', which is faster than curl, but fallback
+# to curl if that doesn't work.  It is hoped that the curl fallback will get
+# around the occasional "ERROR 502: Bad Gateway" errors.
 IMG="/mnt/tests/cloud-image"
-if [ ! -z "$URLxz" ]; then
-  echo "Loading $URLxz ..."
-  time axel -q -o "$IMG" "$URLxz"
-  echo "Loading $KSRC ..."
-  time axel -q -o ~/src.txz $KSRC
-else
-  echo "Loading $URL ..."
-  time axel -q -o "$IMG" "$URL"
+for cmd in 'axel -q -o' 'curl --fail -LSs -o' ; do
+  if [ ! -z "$URLxz" ]; then
+    echo "Loading $URLxz with $cmd..."
+    time eval "$cmd $IMG $URLxz" || true
+
+    if [ ! -s ~/src.txz ] ; then
+      echo "Loading $KSRC with $cmd..."
+      time eval "$cmd ~/src.txz $KSRC" || true
+    fi
+  else
+    echo "Loading $URL with $cmd..."
+    time eval "$cmd $IMG $URL" || true
+  fi
+
+  if [ -s "$IMG" ] ; then
+    # Successful download
+    break
+  fi
+done
+
+# SPECIAL CASE
+# FreeBSD sometimes has broken links in their "current/" URL.  Go back up a
+# level and look for other images that might work.  For example:
+#
+# https://download.freebsd.org/snapshots/CI-IMAGES/16.0-CURRENT/amd64/:
+#
+# 20251110/
+# 20251209/
+# 20260420/
+# current/
+#
+# In this case let's say the raw.xz link in current/ is bad, so look though the
+# other snapshot links for the newest existing raw.xz file.
+if [ ! -z "$URLxz" ] && [ ! -s "$IMG" ] ; then
+  URLxz=$(wget --accept "*.raw.xz" --spider -np --recursive  --no-verbose \
+    $(dirname $(dirname $URLxz)) 2>&1  | awk '/200 OK/{print $(NF-2)}' | \
+    sort -n | tail -n 1)
+  echo "Couldn't download FreeBSD raw.xz.  Trying fallback snapshot $URLxz"
+  curl --fail -LSs -o $IMG $URLxz
 fi
 
 echo "Importing VM image to zvol..."

.github/workflows/scripts/qemu-3-deps-vm.sh

@@ -3,8 +3,11 @@
 ######################################################################
 # 3) install dependencies for compiling and loading
 #
-# $1: OS name (like 'fedora41')
-# $2: (optional) Experimental Fedora kernel version, like "6.14" to
+# qemu-3-deps-vm.sh [--poweroff] OS_NAME [FEDORA_VERSION]
+#
+# --poweroff: Power off the VM after installing dependencies
+# OS_NAME: OS name (like 'fedora41')
+# FEDORA_VERSION: (optional) Experimental Fedora kernel version, like "6.14" to
 #     install instead of Fedora defaults.
 ######################################################################
 
@@ -153,6 +156,12 @@ function install_fedora_experimental_kernel {
   sudo dnf -y copr disable @kernel-vanilla/mainline
 }
 
+POWEROFF=""
+if [ "$1" == "--poweroff" ] ; then
+        POWEROFF=1
+        shift
+fi
+
 # Install dependencies
 case "$1" in
   almalinux8)
@@ -212,6 +221,11 @@ case "$1" in
     sudo apt-get install -yq linux-tools-common libtirpc-dev \
       linux-modules-extra-$(uname -r)
     sudo apt-get install -yq dh-sequence-dkms
+
+    # Need 'build-essential' explicitly for ARM builder
+    # https://github.com/actions/runner-images/issues/9946
+    sudo apt-get install -yq build-essential
+
     echo "##[endgroup]"
     echo "##[group]Delete Ubuntu OpenZFS modules"
     for i in $(find /lib/modules -name zfs -type d); do sudo rm -rvf $i; done
@@ -306,5 +320,7 @@ esac
 
 # reset cloud-init configuration and poweroff
 sudo cloud-init clean --logs
-sleep 2 && sudo poweroff &
+if [ "$POWEROFF" == "1" ] ; then
+        sleep 2 && sudo poweroff &
+fi
 exit 0

.github/workflows/scripts/qemu-4-build-vm.sh

@@ -350,7 +350,16 @@ fi
 # save some sysinfo
 uname -a > /var/tmp/uname.txt
 
-cd $HOME/zfs
+# Check if we're running this script from within a VM or on the runner itself.
+# Most of the time we will be running in a VM, but the ARM builder actually
+# runs this script on the runner.  If we happen to be running on the ARM
+# runner, we will start in the ZFS source directory.  If we're running on a VM
+# then we'll just start in our home directory, and will need to 'cd' into our
+# source directory.
+if [ ! -e META ] ; then
+  cd $HOME/zfs
+fi
+
 export PATH="$PATH:/sbin:/usr/sbin:/usr/local/sbin"
 
 extra=""