Compare commits

..

1 Commits

Author SHA1 Message Date
Auke Kok 81ea586581 Avoid try_shrink_lock spinlock contention
Every lock_key_range call results in a spinlock on linfo->lock to do
housekeeping and reclaim idle locks. This housekeeping is non-critical
and can be deferred, but more importantly, while a reclaim pass is
ongoing, there's no need to attempt reclaim at the same time.

I ran into this while doing cross-node spam and it caused a near hard
lockup with my VM spinning on the lock for 7+ minutes, without making
any forward progress.

We can just trylock and give up. If the lock is held, another task
is already reclaiming and we're safe to skip. The force flag will
still cause a spin_lock to assure shutdown waits for other reclaims,
which needs to happen before shutdown can progress anyway.

Signed-off-by: Auke Kok <auke.kok@versity.com>
2026-06-02 14:15:56 -07:00
11 changed files with 11 additions and 159 deletions
-14
View File
@@ -1,20 +1,6 @@
Versity ScoutFS Release Notes
=============================
---
v1.32
\
*June 2, 2026*
Fix writing POSIX ACLs over NFS mounts that export the scoutfs
filesystem.
Add support for kernels in the RHEL 9.8 minor release.
Reduce unneeded block allocation when data\_prealloc\_contig\_only was
set to 0. This will help achieve more efficient data space usage when
writing small files.
---
v1.31
\
-10
View File
@@ -479,16 +479,6 @@ ifneq (,$(shell grep '^unsigned int stack_trace_save' include/linux/stacktrace.h
ccflags-y += -DKC_STACK_TRACE_SAVE
endif
#
# v3.14-rc1-7-g4e34e719e457
#
# .set_acl callback added to struct inode_operations. Most kernels
# we target have it, but el7 (3.10 base) does not, so detect.
#
ifneq (,$(shell grep 'int ..set_acl..struct' include/linux/fs.h))
ccflags-y += -DKC_HAS_SET_ACL
endif
#
# v6.1-rc1-2-g138060ba92b3
#
+2 -3
View File
@@ -216,8 +216,7 @@ int scoutfs_set_acl(KC_VFS_NS_DEF
{
struct inode *inode = dentry->d_inode;
#else
int scoutfs_set_acl(KC_VFS_NS_DEF
struct inode *inode, struct posix_acl *acl, int type)
int scoutfs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
{
#endif
struct super_block *sb = inode->i_sb;
@@ -310,7 +309,7 @@ int scoutfs_acl_set_xattr(struct dentry *dentry, const char *name, const void *v
#ifdef KC_SET_ACL_DENTRY
ret = scoutfs_set_acl(KC_VFS_INIT_NS dentry, acl, type);
#else
ret = scoutfs_set_acl(KC_VFS_INIT_NS dentry->d_inode, acl, type);
ret = scoutfs_set_acl(dentry->d_inode, acl, type);
#endif
out:
posix_acl_release(acl);
+1 -2
View File
@@ -5,8 +5,7 @@
int scoutfs_set_acl(KC_VFS_NS_DEF
struct dentry *dentry, struct posix_acl *acl, int type);
#else
int scoutfs_set_acl(KC_VFS_NS_DEF
struct inode *inode, struct posix_acl *acl, int type);
int scoutfs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
#endif
#ifdef KC_GET_INODE_ACL
struct posix_acl *scoutfs_get_acl(struct inode *inode, int type, bool rcu);
+1 -1
View File
@@ -2063,7 +2063,7 @@ const struct inode_operations scoutfs_dir_iops = {
#else
.get_acl = scoutfs_get_acl,
#endif
#ifdef KC_HAS_SET_ACL
#ifdef KC_SET_ACL_DENTRY
.set_acl = scoutfs_set_acl,
#endif
.symlink = scoutfs_symlink,
+2 -2
View File
@@ -154,7 +154,7 @@ static const struct inode_operations scoutfs_file_iops = {
#else
.get_acl = scoutfs_get_acl,
#endif
#ifdef KC_HAS_SET_ACL
#ifdef KC_SET_ACL_DENTRY
.set_acl = scoutfs_set_acl,
#endif
.fiemap = scoutfs_data_fiemap,
@@ -174,7 +174,7 @@ static const struct inode_operations scoutfs_special_iops = {
#else
.get_acl = scoutfs_get_acl,
#endif
#ifdef KC_HAS_SET_ACL
#ifdef KC_SET_ACL_DENTRY
.set_acl = scoutfs_set_acl,
#endif
};
+5 -1
View File
@@ -906,7 +906,11 @@ static bool try_shrink_lock(struct super_block *sb, struct lock_info *linfo, boo
READ_ONCE(linfo->nr_locks) <= opts.lock_idle_count))
return false;
spin_lock(&linfo->lock);
/* Shrinking is best-effort housekeeping unless forced. */
if (force)
spin_lock(&linfo->lock);
else if (!spin_trylock(&linfo->lock))
return false;
lock = list_first_entry_or_null(&linfo->lru_list, struct scoutfs_lock, lru_head);
if (lock && (force || (linfo->nr_locks > opts.lock_idle_count))) {
-7
View File
@@ -171,13 +171,6 @@ t_filter_dmesg()
# orphan log trees reclaim is handled, not an error
re="$re|scoutfs .* reclaiming orphan log trees"
# nfs can emit a whole range of messages we can ignore
re="$re|Installing knfsd .*"
re="$re|nfsd: .*"
re="$re|NFSD: .*"
re="$re|RPC: .*"
re="$re|FS-Cache: .*"
# fencing tests force unmounts and trigger timeouts
re="$re|scoutfs .* forcing unmount"
re="$re|scoutfs .* reconnect timed out"
-32
View File
@@ -1,32 +0,0 @@
== write via NFS, read both sides
== POSIX ACL set via NFS, read both sides
user::rw-
user:22222:rw-
group::r--
mask::rw-
other::r--
user::rw-
user:22222:rw-
group::r--
mask::rw-
other::r--
== POSIX ACL set on scoutfs, read via NFS
user::rw-
user:22222:rw-
group::r--
group:44444:r--
mask::rw-
other::r--
== default ACL inheritance via NFS
user::rw-
user:22222:rwx #effective:rw-
group::r-x #effective:r--
mask::rw-
other::r--
== NFS read demand-stages a released file
1
== cleanup
-1
View File
@@ -3,7 +3,6 @@ basic-block-counts.sh
basic-bad-mounts.sh
basic-posix-acl.sh
basic-acl-consistency.sh
basic-nfs.sh
inode-items-updated.sh
simple-inode-index.sh
simple-staging.sh
-86
View File
@@ -1,86 +0,0 @@
#
# Test basic scoutfs-nfs interactions:
# - read/write
# - stage/release and data wait
# - nfs setacl/getacl mapping
#
t_require_commands scoutfs setfacl getfacl exportfs mount.nfs umount \
stat dd cmp systemctl
systemctl start nfs-server >> "$T_TMPDIR/nfs.log" 2>&1 || \
t_skip "nfs-server not available"
# Keep file creation modes deterministic for the ACL golden output.
umask 022
EXPORT_OPTS="rw,async,no_root_squash,no_subtree_check,fsid=42"
NFS_MNT="$T_TMP.nfs"
NFS_DIR="$NFS_MNT/test/basic-nfs"
filter() { sed "s@$T_TMPDIR@T_TMPDIR@g" | t_filter_fs; }
gf() { getfacl -n --omit-header "$@" 2>/dev/null; }
teardown_nfs()
{
(
umount "$NFS_MNT"
exportfs -u "127.0.0.1:$T_M0"
exportfs -f
systemctl stop nfs-server
rmdir "$NFS_MNT"
) >> "$T_TMPDIR/nfs.log" 2>&1
}
trap teardown_nfs EXIT
exportfs -u "127.0.0.1:$T_M0" >> "$T_TMPDIR/nfs.log" 2>&1 || true
t_quiet mkdir -p "$NFS_MNT"
exportfs -o "$EXPORT_OPTS" "127.0.0.1:$T_M0" >> "$T_TMPDIR/nfs.log" 2>&1
mount.nfs -o vers=3,noac,actimeo=0 "127.0.0.1:$T_M0" "$NFS_MNT" >> "$T_TMPDIR/nfs.log" 2>&1
test -d "$NFS_DIR" || t_fail "test dir $NFS_DIR not visible over NFS"
echo "== write via NFS, read both sides"
dd if=/dev/urandom bs=4096 count=1 of="$T_TMP.data" status=none
cp "$T_TMP.data" "$NFS_DIR/file"
cmp "$T_TMP.data" "$T_D0/file"
cmp "$T_TMP.data" "$NFS_DIR/file"
echo "== POSIX ACL set via NFS, read both sides"
setfacl -m u:22222:rw "$NFS_DIR/file" 2>&1 | filter
gf "$NFS_DIR/file"
gf "$T_D0/file"
echo "== POSIX ACL set on scoutfs, read via NFS"
setfacl -m g:44444:r "$T_D0/file" 2>&1 | filter
gf "$NFS_DIR/file"
echo "== default ACL inheritance via NFS"
mkdir "$NFS_DIR/d"
setfacl -d -m u:22222:rwx "$NFS_DIR/d" 2>&1 | filter
touch "$NFS_DIR/d/child"
gf "$NFS_DIR/d/child"
echo "== NFS read demand-stages a released file"
dd if=/dev/urandom bs=4096 count=1 of="$T_TMP.big" status=none
cp "$T_TMP.big" "$T_D0/big"
sync
vers=$(scoutfs stat -s data_version "$T_D0/big")
t_quiet scoutfs release "$T_D0/big" -V "$vers" -o 0 -l 4K
# NFS read against the offline file blocks in scoutfs_read waiting
# for the data to come back online.
cat "$NFS_DIR/big" > "$T_TMP.read" &
read_pid=$!
sleep 1
scoutfs data-waiting -B 0 -I 0 -p "$T_D0" | wc -l
t_quiet scoutfs stage "$T_TMP.big" "$T_D0/big" -V "$vers" -o 0 -l 4096
wait "$read_pid"
cmp "$T_TMP.big" "$T_TMP.read"
echo "== cleanup"
rm -f "$T_D0/file" "$T_D0/big"
rm -rf "$T_D0/d"
t_pass