[Devel] [PATCH vz10] fs: move FALLOC_FL_PREALLOCATE to a bit upstream has not taken
Alexey Kuznetsov
kuznet at virtuozzo.com
Tue Sep 8 16:43:12 MSK 2026
Ack.
On Tue, Sep 8, 2026 at 9:41 PM Konstantin Khorenko
<khorenko at virtuozzo.com> wrote:
>
> We had FALLOC_FL_PREALLOCATE == 0x80.
>
> Upstream commit 7bd43cc79cab ("fs: introduce FALLOC_FL_WRITE_ZEROES to
> fallocate"), released in v6.17, gives 0x80 to a mode of its own and
> adds it to FALLOC_FL_MODE_MASK.
>
> Move the mode to 0x00100000, well above the range upstream allocates
> modes from and away from the top of the word, where a future user is
> just as likely to start.
>
> The number cannot simply change, though, because userspace carries its
> own copy of it. Export the value as a read-only ext4 module parameter,
> /sys/module/ext4/parameters/falloc_fl_preallocate, and let userspace
> read the mode from there.
>
> The old value stays accepted as FALLOC_FL_PREALLOCATE_OLD, so that
> userspace built before this change keeps working. It has to be dropped
> before the rebase onto v6.17 or later, otherwise we would take
> FALLOC_FL_WRITE_ZEROES requests for our own mode.
>
> Feature: ext4: preallocation flag for small fallocates
>
> https://virtuozzo.atlassian.net/browse/VSTOR-144521
>
> Signed-off-by: Konstantin Khorenko <khorenko at virtuozzo.com>
> ---
> Follow-up to the v3 thread of "fs/ext4: let FALLOC_FL_PREALLOCATE
> through the vfs_fallocate mode switch": 0x80 is FALLOC_FL_WRITE_ZEROES
> upstream since v6.17, so our mode has to move before we rebase onto it,
> and the value has to become discoverable instead of hardcoded on both
> sides.
> ---
> fs/ext4/extents.c | 14 ++++++++++----
> fs/open.c | 1 +
> include/linux/falloc.h | 3 ++-
> include/uapi/linux/falloc.h | 15 ++++++++++++++-
> 4 files changed, 27 insertions(+), 6 deletions(-)
>
> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> index 38f50770f6585..eca3f0312ac53 100644
> --- a/fs/ext4/extents.c
> +++ b/fs/ext4/extents.c
> @@ -39,6 +39,10 @@ static unsigned int falloc_prealloc;
> module_param(falloc_prealloc, uint, 0644);
> MODULE_PARM_DESC(falloc_prealloc, "Preallocate fallocate like it is a write for smaller sizes");
>
> +static unsigned int falloc_fl_preallocate = FALLOC_FL_PREALLOCATE;
> +module_param(falloc_fl_preallocate, uint, 0444);
> +MODULE_PARM_DESC(falloc_fl_preallocate, "fallocate() mode value asking for preallocation");
> +
> /*
> * used by extent splitting.
> */
> @@ -4514,8 +4518,8 @@ static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
> * that it doesn't get unnecessarily split into multiple
> * extents.
> */
> - if (!(mode & FALLOC_FL_PREALLOCATE) && len <= EXT_UNWRITTEN_MAX_LEN &&
> - len > falloc_prealloc)
> + if (!(mode & (FALLOC_FL_PREALLOCATE | FALLOC_FL_PREALLOCATE_OLD)) &&
> + len <= EXT_UNWRITTEN_MAX_LEN && len > falloc_prealloc)
> flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
>
> /*
> @@ -4754,7 +4758,8 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
> /* Return error if mode is not supported */
> if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
> FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
> - FALLOC_FL_INSERT_RANGE | FALLOC_FL_PREALLOCATE))
> + FALLOC_FL_INSERT_RANGE | FALLOC_FL_PREALLOCATE |
> + FALLOC_FL_PREALLOCATE_OLD))
> return -EOPNOTSUPP;
>
> inode_lock(inode);
> @@ -4770,7 +4775,8 @@ long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
> goto out_inode_lock;
>
> if ((mode & FALLOC_FL_MODE_MASK) == FALLOC_FL_ALLOCATE_RANGE ||
> - (mode & FALLOC_FL_MODE_MASK) == FALLOC_FL_PREALLOCATE) {
> + (mode & FALLOC_FL_MODE_MASK) == FALLOC_FL_PREALLOCATE ||
> + (mode & FALLOC_FL_MODE_MASK) == FALLOC_FL_PREALLOCATE_OLD) {
> ret = ext4_do_fallocate(file, offset, len, mode);
> goto out_inode_lock;
> }
> diff --git a/fs/open.c b/fs/open.c
> index fc72ad09e7e81..65c96bf03dc93 100644
> --- a/fs/open.c
> +++ b/fs/open.c
> @@ -268,6 +268,7 @@ int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
> case FALLOC_FL_UNSHARE_RANGE:
> case FALLOC_FL_ZERO_RANGE:
> case FALLOC_FL_PREALLOCATE:
> + case FALLOC_FL_PREALLOCATE_OLD:
> break;
> case FALLOC_FL_PUNCH_HOLE:
> if (!(mode & FALLOC_FL_KEEP_SIZE))
> diff --git a/include/linux/falloc.h b/include/linux/falloc.h
> index f6eee099b3fca..58e991ea6771f 100644
> --- a/include/linux/falloc.h
> +++ b/include/linux/falloc.h
> @@ -37,7 +37,8 @@ struct space_resv {
> FALLOC_FL_ZERO_RANGE | \
> FALLOC_FL_INSERT_RANGE | \
> FALLOC_FL_UNSHARE_RANGE | \
> - FALLOC_FL_PREALLOCATE)
> + FALLOC_FL_PREALLOCATE | \
> + FALLOC_FL_PREALLOCATE_OLD)
>
> /* on ia32 l_start is on a 32-bit boundary */
> #if defined(CONFIG_X86_64)
> diff --git a/include/uapi/linux/falloc.h b/include/uapi/linux/falloc.h
> index 7ac8ffb8e6523..49aa826ad8f8a 100644
> --- a/include/uapi/linux/falloc.h
> +++ b/include/uapi/linux/falloc.h
> @@ -78,6 +78,19 @@
> */
> #define FALLOC_FL_UNSHARE_RANGE 0x40
>
> -#define FALLOC_FL_PREALLOCATE 0x80
> +/*
> + * FALLOC_FL_PREALLOCATE tells the filesystem to allocate the range the way a
> + * write would, with mballoc preallocation, rather than trimming the request
> + * down to a single unwritten extent.
> + *
> + * Userspace may read the mode value from
> + * /sys/module/ext4/parameters/falloc_fl_preallocate.
> + *
> + * FALLOC_FL_PREALLOCATE_OLD is the value used before upstream v6.17 claimed
> + * 0x80 for FALLOC_FL_WRITE_ZEROES. It is accepted as an alias of the above,
> + * and has to be dropped before a rebase onto v6.17 or later.
> + */
> +#define FALLOC_FL_PREALLOCATE_OLD 0x80
> +#define FALLOC_FL_PREALLOCATE 0x00100000
>
> #endif /* _UAPI_FALLOC_H_ */
>
> ---
> base-commit: 3466ec0b2cdf9d49dbb36d7b5ed9b36dfb2fb04d
> change-id: 20260908-pr-khorenko-vstor-144521-falloc_fl_preallocate-cec80ce87000
>
> Best regards,
> --
> Konstantin Khorenko <khorenko at virtuozzo.com>
More information about the Devel
mailing list