[RFC PATCH] fat: add noflush mount option

[RFC PATCH] fat: add noflush mount option

[David Timber]

On most distros, udisks manages users' removable volume mount requests.
Udisks is configured to always mount FAT volumes with the 'flush' mount
option. This is largely okay for populating directory in small sizes,
but when a large number of files are involved, the flush behaviour acts
as a bottleneck point in fat_file_release(). The user may want to
disable the flush option temporarily before commencing such an intensive
operation.

To cover this use case, introduce the new 'noflush' mount option. When
used in the mount options to mount a volume, it overrides the 'flush'
option previously specified. When used in remount, update it updates the
flag.

This is a breaking change as traditionally, the mount options other than
rw and ro are ignored. The patch breaks this tradition by allowing
reconfiguration of 'flush' and 'noflush' mount options.

Signed-off-by: David Timber <dxdt@dev.snart.me>
---
 fs/fat/inode.c | 26 +++++++++++++++++++++-----
 1 file changed, 21 insertions(+), 5 deletions(-)

diff --git a/fs/fat/inode.c b/fs/fat/inode.c
index 28f78df086ef..850df43ba354 100644
--- a/fs/fat/inode.c
+++ b/fs/fat/inode.c
@@ -813,10 +813,14 @@ int fat_reconfigure(struct fs_context *fc)
 	bool new_rdonly;
 	struct super_block *sb = fc->root->d_sb;
 	struct msdos_sb_info *sbi = MSDOS_SB(sb);
+	struct fat_mount_options *new_opts = fc->fs_private;
 	fc->sb_flags |= SB_NODIRATIME | (sbi->options.isvfat ? 0 : SB_NOATIME);
 
 	sync_filesystem(sb);
 
+	/* allow reconfiguring "flush" or "noflush" */
+	sbi->options.flush = new_opts->flush;
+
 	/* make sure we update state on remount. */
 	new_rdonly = fc->sb_flags & SB_RDONLY;
 	if (new_rdonly != sb_rdonly(sb)) {
@@ -1047,7 +1051,7 @@ enum {
 	Opt_charset, Opt_shortname, Opt_utf8, Opt_utf8_bool,
 	Opt_uni_xl, Opt_uni_xl_bool, Opt_nonumtail, Opt_nonumtail_bool,
 	Opt_obsolete, Opt_flush, Opt_tz, Opt_rodir, Opt_errors, Opt_discard,
-	Opt_nfs, Opt_nfs_enum, Opt_time_offset, Opt_dos1xfloppy,
+	Opt_nfs, Opt_nfs_enum, Opt_time_offset, Opt_dos1xfloppy, Opt_noflush
 };
 
 static const struct constant_table fat_param_check[] = {
@@ -1110,6 +1114,7 @@ const struct fs_parameter_spec fat_param_spec[] = {
 	fsparam_flag	("debug",	Opt_debug),
 	fsparam_flag	("sys_immutable", Opt_immutable),
 	fsparam_flag	("flush",	Opt_flush),
+	fsparam_flag	("noflush",	Opt_noflush),
 	fsparam_enum	("tz",		Opt_tz, fat_param_tz),
 	fsparam_s32	("time_offset",	Opt_time_offset),
 	fsparam_enum	("errors",	Opt_errors, fat_param_errors),
@@ -1167,10 +1172,6 @@ int fat_parse_param(struct fs_context *fc, struct fs_parameter *param,
 	struct fs_parse_result result;
 	int opt;
 
-	/* remount options have traditionally been ignored */
-	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE)
-		return 0;
-
 	opt = fs_parse(fc, fat_param_spec, param, &result);
 	/* If option not found in fat_param_spec, try vfat/msdos options */
 	if (opt == -ENOPARAM) {
@@ -1183,6 +1184,18 @@ int fat_parse_param(struct fs_context *fc, struct fs_parameter *param,
 	if (opt < 0)
 		return opt;
 
+	/* remount options have traditionally been ignored */
+	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
+		switch (opt) {
+		/* but there are exceptions */
+		case Opt_flush:
+		case Opt_noflush:
+			break;
+		default:
+			return 0;
+		}
+	}
+
 	switch (opt) {
 	case Opt_check:
 		opts->name_check = result.uint_32;
@@ -1235,6 +1248,9 @@ int fat_parse_param(struct fs_context *fc, struct fs_parameter *param,
 	case Opt_flush:
 		opts->flush = 1;
 		break;
+	case Opt_noflush:
+		opts->flush = 0;
+		break;
 	case Opt_time_offset:
 		/*
 		 * GMT+-12 zones may have DST corrections so at least
-- 
2.55.0

Re: [RFC PATCH] fat: add noflush mount option

[OGAWA Hirofumi]

David Timber <dxdt@dev.snart.me> writes:

> On most distros, udisks manages users' removable volume mount requests.
> Udisks is configured to always mount FAT volumes with the 'flush' mount
> option. This is largely okay for populating directory in small sizes,
> but when a large number of files are involved, the flush behaviour acts
> as a bottleneck point in fat_file_release(). The user may want to
> disable the flush option temporarily before commencing such an intensive
> operation.
>
> To cover this use case, introduce the new 'noflush' mount option. When
> used in the mount options to mount a volume, it overrides the 'flush'
> option previously specified. When used in remount, update it updates the
> flag.
>
> This is a breaking change as traditionally, the mount options other than
> rw and ro are ignored. The patch breaks this tradition by allowing
> reconfiguration of 'flush' and 'noflush' mount options.
>
> Signed-off-by: David Timber <dxdt@dev.snart.me>
> ---
>  fs/fat/inode.c | 26 +++++++++++++++++++++-----
>  1 file changed, 21 insertions(+), 5 deletions(-)
>
> diff --git a/fs/fat/inode.c b/fs/fat/inode.c
> index 28f78df086ef..850df43ba354 100644
> --- a/fs/fat/inode.c
> +++ b/fs/fat/inode.c
> @@ -813,10 +813,14 @@ int fat_reconfigure(struct fs_context *fc)
>  	bool new_rdonly;
>  	struct super_block *sb = fc->root->d_sb;
>  	struct msdos_sb_info *sbi = MSDOS_SB(sb);
> +	struct fat_mount_options *new_opts = fc->fs_private;
>  	fc->sb_flags |= SB_NODIRATIME | (sbi->options.isvfat ? 0 : SB_NOATIME);
>  
>  	sync_filesystem(sb);
>  
> +	/* allow reconfiguring "flush" or "noflush" */
> +	sbi->options.flush = new_opts->flush;
> +
>  	/* make sure we update state on remount. */
>  	new_rdonly = fc->sb_flags & SB_RDONLY;
>  	if (new_rdonly != sb_rdonly(sb)) {

Maybe, better to set after changed the read-only?

> @@ -1047,7 +1051,7 @@ enum {
>  	Opt_charset, Opt_shortname, Opt_utf8, Opt_utf8_bool,
>  	Opt_uni_xl, Opt_uni_xl_bool, Opt_nonumtail, Opt_nonumtail_bool,
>  	Opt_obsolete, Opt_flush, Opt_tz, Opt_rodir, Opt_errors, Opt_discard,
> -	Opt_nfs, Opt_nfs_enum, Opt_time_offset, Opt_dos1xfloppy,
> +	Opt_nfs, Opt_nfs_enum, Opt_time_offset, Opt_dos1xfloppy, Opt_noflush
>  };
>  
>  static const struct constant_table fat_param_check[] = {
> @@ -1110,6 +1114,7 @@ const struct fs_parameter_spec fat_param_spec[] = {
>  	fsparam_flag	("debug",	Opt_debug),
>  	fsparam_flag	("sys_immutable", Opt_immutable),
>  	fsparam_flag	("flush",	Opt_flush),
> +	fsparam_flag	("noflush",	Opt_noflush),

fsparam_flag() is not including the "noflush" too?

>  	fsparam_enum	("tz",		Opt_tz, fat_param_tz),
>  	fsparam_s32	("time_offset",	Opt_time_offset),
>  	fsparam_enum	("errors",	Opt_errors, fat_param_errors),
> @@ -1167,10 +1172,6 @@ int fat_parse_param(struct fs_context *fc, struct fs_parameter *param,
>  	struct fs_parse_result result;
>  	int opt;
>  
> -	/* remount options have traditionally been ignored */
> -	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE)
> -		return 0;
> -
>  	opt = fs_parse(fc, fat_param_spec, param, &result);
>  	/* If option not found in fat_param_spec, try vfat/msdos options */
>  	if (opt == -ENOPARAM) {
> @@ -1183,6 +1184,18 @@ int fat_parse_param(struct fs_context *fc, struct fs_parameter *param,
>  	if (opt < 0)
>  		return opt;
>  
> +	/* remount options have traditionally been ignored */
> +	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
> +		switch (opt) {
> +		/* but there are exceptions */
> +		case Opt_flush:
> +		case Opt_noflush:
> +			break;
> +		default:
> +			return 0;
> +		}
> +	}
> +
>  	switch (opt) {
>  	case Opt_check:
>  		opts->name_check = result.uint_32;
> @@ -1235,6 +1248,9 @@ int fat_parse_param(struct fs_context *fc, struct fs_parameter *param,
>  	case Opt_flush:
>  		opts->flush = 1;
>  		break;
> +	case Opt_noflush:
> +		opts->flush = 0;
> +		break;
>  	case Opt_time_offset:
>  		/*
>  		 * GMT+-12 zones may have DST corrections so at least

-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Re: [RFC PATCH] fat: add noflush mount option

[David Timber]

On 8/24/26 18:45, OGAWA Hirofumi wrote:
> David Timber <dxdt@dev.snart.me> writes:
>
>> On most distros, udisks manages users' removable volume mount requests.
>> Udisks is configured to always mount FAT volumes with the 'flush' mount
>> option. This is largely okay for populating directory in small sizes,
>> but when a large number of files are involved, the flush behaviour acts
>> as a bottleneck point in fat_file_release(). The user may want to
>> disable the flush option temporarily before commencing such an intensive
>> operation.
>>
>> To cover this use case, introduce the new 'noflush' mount option. When
>> used in the mount options to mount a volume, it overrides the 'flush'
>> option previously specified. When used in remount, update it updates the
>> flag.
>>
>> This is a breaking change as traditionally, the mount options other than
>> rw and ro are ignored. The patch breaks this tradition by allowing
>> reconfiguration of 'flush' and 'noflush' mount options.
>>
>> Signed-off-by: David Timber <dxdt@dev.snart.me>
>> ---
>>  fs/fat/inode.c | 26 +++++++++++++++++++++-----
>>  1 file changed, 21 insertions(+), 5 deletions(-)
>>
>> diff --git a/fs/fat/inode.c b/fs/fat/inode.c
>> index 28f78df086ef..850df43ba354 100644
>> --- a/fs/fat/inode.c
>> +++ b/fs/fat/inode.c
>> @@ -813,10 +813,14 @@ int fat_reconfigure(struct fs_context *fc)
>>  	bool new_rdonly;
>>  	struct super_block *sb = fc->root->d_sb;
>>  	struct msdos_sb_info *sbi = MSDOS_SB(sb);
>> +	struct fat_mount_options *new_opts = fc->fs_private;
>>  	fc->sb_flags |= SB_NODIRATIME | (sbi->options.isvfat ? 0 : SB_NOATIME);
>>  
>>  	sync_filesystem(sb);
>>  
>> +	/* allow reconfiguring "flush" or "noflush" */
>> +	sbi->options.flush = new_opts->flush;
>> +
>>  	/* make sure we update state on remount. */
>>  	new_rdonly = fc->sb_flags & SB_RDONLY;
>>  	if (new_rdonly != sb_rdonly(sb)) {
> Maybe, better to set after changed the read-only?
ACK

>> @@ -1047,7 +1051,7 @@ enum {
>>  	Opt_charset, Opt_shortname, Opt_utf8, Opt_utf8_bool,
>>  	Opt_uni_xl, Opt_uni_xl_bool, Opt_nonumtail, Opt_nonumtail_bool,
>>  	Opt_obsolete, Opt_flush, Opt_tz, Opt_rodir, Opt_errors, Opt_discard,
>> -	Opt_nfs, Opt_nfs_enum, Opt_time_offset, Opt_dos1xfloppy,
>> +	Opt_nfs, Opt_nfs_enum, Opt_time_offset, Opt_dos1xfloppy, Opt_noflush
>>  };
>>  
>>  static const struct constant_table fat_param_check[] = {
>> @@ -1110,6 +1114,7 @@ const struct fs_parameter_spec fat_param_spec[] = {
>>  	fsparam_flag	("debug",	Opt_debug),
>>  	fsparam_flag	("sys_immutable", Opt_immutable),
>>  	fsparam_flag	("flush",	Opt_flush),
>> +	fsparam_flag	("noflush",	Opt_noflush),
> fsparam_flag() is not including the "noflush" too?
I think I should use fsparam_flag_no() here. My bad.

It seems that you don't object to the idea. That's a good start!

If we want to do this, the
documentation(Documentation/filesystems/vfat.rst and mount(8)) should be
updated, too. fyi, `udisksctl mount -o` won't accept "noflush" so
that'll have to be fixed, too. Also have to make sure if "noflush" is
used in fstab, udisks2 overrides the system default settings.

Davo

Re: [RFC PATCH] fat: add noflush mount option

[OGAWA Hirofumi]

David Timber <dxdt@dev.snart.me> writes:


[...]

>>> @@ -1110,6 +1114,7 @@ const struct fs_parameter_spec fat_param_spec[] = {
>>>  	fsparam_flag	("debug",	Opt_debug),
>>>  	fsparam_flag	("sys_immutable", Opt_immutable),
>>>  	fsparam_flag	("flush",	Opt_flush),
>>> +	fsparam_flag	("noflush",	Opt_noflush),
>> fsparam_flag() is not including the "noflush" too?
> I think I should use fsparam_flag_no() here. My bad.
>
> It seems that you don't object to the idea. That's a good start!
>
> If we want to do this, the
> documentation(Documentation/filesystems/vfat.rst and mount(8)) should be
> updated, too. fyi, `udisksctl mount -o` won't accept "noflush" so
> that'll have to be fixed, too. Also have to make sure if "noflush" is
> used in fstab, udisks2 overrides the system default settings.

Well, with a your patch, just remounting without "flush" option doesn't
remove flush?  And this is a user visible change more or less, so if
there is an objection from users, then it will be dropped.

Thanks.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Re: [RFC PATCH] fat: add noflush mount option

[Christoph Hellwig]

On Sun, Aug 23, 2026 at 02:37:32AM +0900, David Timber wrote:
> On most distros, udisks manages users' removable volume mount requests.
> Udisks is configured to always mount FAT volumes with the 'flush' mount
> option. This is largely okay for populating directory in small sizes,

No, it's not.  It is an amaingly stupid idea that leads to horrible
performance and flash wearout.  I'm not objecting to your patch, but
if you care about the usability of FAT volumes in this scenario, the
more important thing is to fix this default.