From: Tianyi Chen <hi@tychen.cc>
To: util-linux@vger.kernel.org
Cc: Karel Zak <kzak@redhat.com>
Subject: [PATCH] libmount: support explicit loop direct I/O requests
Date: Mon, 21 Sep 2026 16:36:13 +0900 [thread overview]
Message-ID: <20260921073613.99690-1-hi@tychen.cc> (raw)
Allow X-loopdev.direct-io=on|off to select the direct I/O mode when
mounting a file through a loop device. Pass the flag to the existing
loop setup path and require a matching mode when reusing a device.
Omitting the option preserves the current behavior.
Keep reused devices attached if the mount fails; only release the
reference acquired by this mount. Add tests for option parsing, new
devices, reuse, mode conflicts and failure cleanup.
Fixes: https://github.com/util-linux/util-linux/issues/3464
Assisted-by: LLM
Signed-off-by: Tianyi Chen <hi@tychen.cc>
---
Following the RFC discussion, this uses loopcxt_setup_device() without
adding effective-mode readback for new devices:
https://lore.kernel.org/r/zxpnm3dbgy6vv64gockhcl5egfpe2qpld7jlkgg3enr5jbfha2@q27vjqxnocmb
Built with Meson. The new loop-direct-io test and existing regfile and
fstab-loop tests pass in an x86-64 VM with ext4 backing storage.
libmount/src/hook_loopdev.c | 27 ++++++++-
sys-utils/mount.8.adoc | 10 +++-
tests/expected/mount/loop-direct-io | 6 ++
tests/ts/mount/loop-direct-io | 91 +++++++++++++++++++++++++++++
4 files changed, 132 insertions(+), 2 deletions(-)
create mode 100644 tests/expected/mount/loop-direct-io
create mode 100755 tests/ts/mount/loop-direct-io
diff --git a/libmount/src/hook_loopdev.c b/libmount/src/hook_loopdev.c
index b2291340..cfd07fb1 100644
--- a/libmount/src/hook_loopdev.c
+++ b/libmount/src/hook_loopdev.c
@@ -21,6 +21,7 @@
struct hook_data {
int loopdev_fd;
+ bool reused;
};
static int delete_loopdev(struct libmnt_context *cxt, struct hook_data *hd);
@@ -138,7 +139,7 @@ static int setup_loopdev(struct libmnt_context *cxt,
{
const char *backing_file, *loopdev = NULL;
struct loopdev_cxt lc;
- int rc = 0, lo_flags = 0;
+ int rc = 0, lo_flags = 0, direct_io = -1;
uint64_t offset = 0, sizelimit = 0;
bool reuse = FALSE;
struct libmnt_opt *opt, *loopopt = NULL;
@@ -154,6 +155,19 @@ static int setup_loopdev(struct libmnt_context *cxt,
lo_flags |= LO_FLAGS_READ_ONLY;
}
+ opt = mnt_optlist_get_named(ol, "X-loopdev.direct-io", cxt->map_userspace);
+ if (opt) {
+ const char *value = mnt_opt_get_value(opt);
+
+ if (value && strcmp(value, "on") == 0) {
+ direct_io = 1;
+ lo_flags |= LO_FLAGS_DIRECT_IO;
+ } else if (value && strcmp(value, "off") == 0)
+ direct_io = 0;
+ else
+ return -MNT_ERR_MOUNTOPT;
+ }
+
/*
* loop=
*/
@@ -257,6 +271,13 @@ static int setup_loopdev(struct libmnt_context *cxt,
goto done;
}
+ if (direct_io != -1 && !!loopcxt_is_dio(&lc) != direct_io) {
+ DBG_OBJ(LOOP, cxt, ul_debug("%s has incompatible direct I/O mode",
+ loopcxt_get_device(&lc)));
+ rc = -MNT_ERR_LOOPOVERLAP;
+ goto done;
+ }
+
/* This is no more supported, but check to be safe. */
if (loopcxt_get_encrypt_type(&lc, &lc_encrypt_type) == 0
&& lc_encrypt_type != LO_CRYPT_NONE) {
@@ -350,6 +371,7 @@ static int setup_loopdev(struct libmnt_context *cxt,
} while (1);
success:
+ hd->reused = reuse;
if (!rc)
rc = mnt_fs_set_source(cxt->fs, loopcxt_get_device(&lc));
@@ -408,6 +430,9 @@ static int delete_loopdev(struct libmnt_context *cxt, struct hook_data *hd)
hd->loopdev_fd = -1;
}
+ if (hd && hd->reused)
+ return 0;
+
rc = loopdev_detach(src); /* see lib/loopdev.c */
DBG_OBJ(LOOP, cxt, ul_debug("deleted [rc=%d]", rc));
diff --git a/sys-utils/mount.8.adoc b/sys-utils/mount.8.adoc
index 922ecda7..208a1a16 100644
--- a/sys-utils/mount.8.adoc
+++ b/sys-utils/mount.8.adoc
@@ -796,6 +796,14 @@ ____
mount -t auto -o'X-mount.auto-fstypes="noext2,ext3"' /dev/sdc1 /mnt/test
____
+*X-loopdev.direct-io*=*on*|*off*::
+Enable or disable direct I/O when setting up a loop device. If a loop device
+already exists for the same file, offset and size limit, its direct I/O mode
+must match; otherwise the mount fails without changing the existing device.
+Without this option, new loop devices use buffered I/O and existing devices
+can be reused in either mode. This option does not change an existing mount
+or apply when no loop device is needed. See also *losetup*(8) *--direct-io*.
+
*X-mount.mkdir*[**=**_mode_]::
Allow to make a target directory (mountpoint) if it does not exist yet. The optional argument _mode_ specifies the filesystem access mode used for *mkdir*(2) in octal notation. The default mode is 0755. This functionality is supported only for root users or when *mount* is executed without suid permissions. The option is also supported as *x-mount.mkdir*, but this notation is deprecated since v2.30. See also *--mkdir* command line option.
@@ -1708,7 +1716,7 @@ ____
*mount -t ext4 /tmp/disk.img /mnt*
____
-This type of mount knows about three options, namely *loop*, *offset* and *sizelimit*, that are really options to *losetup*(8). (These options can be used in addition to those specific to the filesystem type.)
+Loop device setup accepts *loop*, *offset*, *sizelimit* and *X-loopdev.direct-io*, corresponding to options of *losetup*(8). (These options can be used in addition to those specific to the filesystem type.)
Since Linux 2.6.25 auto-destruction of loop devices is supported, meaning that any loop device allocated by *mount* will be freed by *umount* independently of _/etc/mtab_.
diff --git a/tests/expected/mount/loop-direct-io b/tests/expected/mount/loop-direct-io
new file mode 100644
index 00000000..990eda84
--- /dev/null
+++ b/tests/expected/mount/loop-direct-io
@@ -0,0 +1,6 @@
+invalid options: OK
+new device (on): OK
+new device (off): OK
+new device (default): OK
+reuse and failed mount (on): OK
+reuse and failed mount (off): OK
diff --git a/tests/ts/mount/loop-direct-io b/tests/ts/mount/loop-direct-io
new file mode 100755
index 00000000..4f30b2bc
--- /dev/null
+++ b/tests/ts/mount/loop-direct-io
@@ -0,0 +1,91 @@
+#!/usr/bin/env bash
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+TS_TOPDIR="${0%/*}/../.."
+TS_DESC="loop direct I/O"
+
+. "$TS_TOPDIR"/functions.sh
+ts_init "$*"
+
+ts_check_test_command "$TS_CMD_MOUNT"
+ts_check_test_command "$TS_CMD_UMOUNT"
+ts_skip_nonroot
+ts_check_losetup
+ts_check_prog "mkfs.ext2"
+
+IMAGE=$(ts_image_init)
+DEVICE=
+
+cleanup()
+{
+ local device
+
+ $TS_CMD_UMOUNT "$TS_MOUNTPOINT" &> /dev/null
+ for device in $($TS_CMD_LOSETUP -j "$IMAGE" | cut -d: -f1); do
+ $TS_CMD_LOSETUP -d "$device" &> /dev/null
+ done
+}
+trap cleanup EXIT
+
+mkfs.ext2 -F "$IMAGE" &> /dev/null || ts_die "Cannot make filesystem"
+mkdir -p "$TS_MOUNTPOINT"
+
+for option in X-loopdev.direct-io X-loopdev.direct-io= X-loopdev.direct-io=invalid; do
+ $TS_CMD_MOUNT -o "$option" "$IMAGE" "$TS_MOUNTPOINT" &> /dev/null &&
+ ts_die "Accepted invalid option: $option"
+done
+[ -z "$($TS_CMD_LOSETUP -j "$IMAGE")" ] || ts_die "Invalid option left a loop device"
+ts_log "invalid options: OK"
+
+DEVICE=$($TS_CMD_LOSETUP --direct-io=on --find --show "$IMAGE" 2>> "$TS_ERRLOG") ||
+ ts_skip "backing filesystem does not support direct I/O"
+[ "$(cat "/sys/class/block/${DEVICE##*/}/loop/dio")" = 1 ] ||
+ ts_skip "loop direct I/O not supported"
+$TS_CMD_LOSETUP -d "$DEVICE" || ts_die "Cannot detach probe device"
+DEVICE=
+
+for mode in on off default; do
+ options=loop
+ expected=0
+ [ "$mode" = default ] || options+=",X-loopdev.direct-io=$mode"
+ [ "$mode" != on ] || expected=1
+ $TS_CMD_MOUNT -o "$options" "$IMAGE" "$TS_MOUNTPOINT" >> "$TS_ERRLOG" 2>&1 ||
+ ts_die "Cannot mount with mode $mode"
+ DEVICE=$($TS_CMD_LOSETUP -j "$IMAGE" | cut -d: -f1)
+ [ "$(cat "/sys/class/block/${DEVICE##*/}/loop/dio")" = "$expected" ] ||
+ ts_die "Wrong direct I/O mode: $mode"
+ $TS_CMD_UMOUNT "$TS_MOUNTPOINT" || ts_die "Cannot unmount"
+ DEVICE=
+ $TS_CMD_MOUNT -t ext2 -o "$options,invalid-fs-option" "$IMAGE" "$TS_MOUNTPOINT" &> /dev/null &&
+ ts_die "Accepted invalid filesystem option on new device"
+ [ -z "$($TS_CMD_LOSETUP -j "$IMAGE")" ] || ts_die "Failed mount left a new loop device"
+ ts_log "new device ($mode): OK"
+done
+
+for mode in on off; do
+ DEVICE=$($TS_CMD_LOSETUP --direct-io="$mode" --find --show "$IMAGE") ||
+ ts_die "Cannot set up device"
+ for options in "loop,X-loopdev.direct-io=$mode" loop; do
+ $TS_CMD_MOUNT -o "$options" "$IMAGE" "$TS_MOUNTPOINT" >> "$TS_ERRLOG" 2>&1 ||
+ ts_die "Cannot reuse device: $options"
+ $TS_CMD_UMOUNT "$TS_MOUNTPOINT" || ts_die "Cannot unmount reused device"
+ done
+ other=on
+ [ "$mode" != on ] || other=off
+ $TS_CMD_MOUNT -o "X-loopdev.direct-io=$other" "$IMAGE" "$TS_MOUNTPOINT" &> /dev/null &&
+ ts_die "Reused device with incompatible direct I/O mode"
+ $TS_CMD_MOUNT -t ext2 -o "X-loopdev.direct-io=$mode,invalid-fs-option" \
+ "$IMAGE" "$TS_MOUNTPOINT" &> /dev/null &&
+ ts_die "Accepted invalid filesystem option"
+ [ "$($TS_CMD_LOSETUP -j "$IMAGE" | cut -d: -f1)" = "$DEVICE" ] ||
+ ts_die "Failed mount detached or duplicated reused device"
+ expected=0
+ [ "$mode" != on ] || expected=1
+ [ "$(cat "/sys/class/block/${DEVICE##*/}/loop/dio")" = "$expected" ] ||
+ ts_die "Reuse changed direct I/O mode"
+ $TS_CMD_LOSETUP -d "$DEVICE" || ts_die "Cannot detach reused device"
+ DEVICE=
+ ts_log "reuse and failed mount ($mode): OK"
+done
+
+ts_finalize
base-commit: 55e3c4dba6101a47ecaae71c91edbe10c59b8d56
--
2.55.0
next reply other threads:[~2026-09-21 7:42 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-09-21 7:36 Tianyi Chen [this message]
2026-09-21 13:25 ` [PATCH] libmount: support explicit loop direct I/O requests Karel Zak
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260921073613.99690-1-hi@tychen.cc \
--to=hi@tychen.cc \
--cc=kzak@redhat.com \
--cc=util-linux@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox