* [PATCH] fuse: add fusex filesystem
@ 2026-04-29 10:20 Miklos Szeredi
2026-05-07 8:31 ` Horst Birthelmer
` (3 more replies)
0 siblings, 4 replies; 17+ messages in thread
From: Miklos Szeredi @ 2026-04-29 10:20 UTC (permalink / raw)
To: fuse-devel, linux-fsdevel
This stands for "fuse extended/experimental".
The purpose is to provide a clean base for big features like the FUSE_IOMAP
api.
It's also a good way to try new stuff like file handles and compound
requests without the risk of breaking something in the large and complex
fuse codebase.
Whether these features will be migrated back into the main fuse codebase,
or fusex is going to end up as a major version update is still up in the
air.
Major differences from regular fuse:
- local filesystem mode only
- only synchronous FUSE_INIT is supported
- only no-open mode
- new requests:
+ FUSE_LOOKUP_ROOT - return nodeid of root
+ FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
+ FUSE_MKOBJX - merged FUSE_MKNOD, MKDIR, SYMLINK and TMPFILE
+ FUSE_SETSTATX - extended version of FUSE_SETATTR
Missing features:
- file handles / export ops
- compound requests
- xattr caching
- fileattr
- fiemap
- ioctl
- copy_file_range
- lazy dir open
Test server can be found at:
https://github.com/szmi/fuse-utils
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
---
Patch is against
git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse.git#for-next
fs/fuse/Makefile | 2 +-
fs/fuse/args.h | 1 +
fs/fuse/dev.c | 13 +-
fs/fuse/dir.c | 4 +-
fs/fuse/fuse_i.h | 8 +
fs/fuse/fusex.c | 1751 +++++++++++++++++++++++++++++++++++++
fs/fuse/fusex.h | 4 +
fs/fuse/inode.c | 16 +-
include/uapi/linux/fuse.h | 36 +
9 files changed, 1826 insertions(+), 9 deletions(-)
create mode 100644 fs/fuse/fusex.c
create mode 100644 fs/fuse/fusex.h
diff --git a/fs/fuse/Makefile b/fs/fuse/Makefile
index 245e67852b03..d9963e411b62 100644
--- a/fs/fuse/Makefile
+++ b/fs/fuse/Makefile
@@ -12,7 +12,7 @@ obj-$(CONFIG_VIRTIO_FS) += virtiofs.o
fuse-y := trace.o # put trace.o first so we see ftrace errors sooner
fuse-y += dev.o dir.o file.o inode.o control.o xattr.o acl.o readdir.o ioctl.o req_timeout.o req.o
-fuse-y += poll.o notify.o
+fuse-y += poll.o notify.o fusex.o
fuse-y += iomode.o
fuse-$(CONFIG_FUSE_DAX) += dax.o
fuse-$(CONFIG_FUSE_PASSTHROUGH) += passthrough.o backing.o
diff --git a/fs/fuse/args.h b/fs/fuse/args.h
index ecfe51a192af..1c1e0a25ea07 100644
--- a/fs/fuse/args.h
+++ b/fs/fuse/args.h
@@ -35,6 +35,7 @@ struct fuse_args {
bool out_pages:1;
bool user_pages:1;
bool out_argvar:1;
+ bool out_var_alloc:1;
bool page_zeroing:1;
bool page_replace:1;
bool may_block:1;
diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index 6fe0d8c263df..134572b1a9af 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -1847,15 +1847,24 @@ int fuse_copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
reqsize += fuse_len_args(args->out_numargs, args->out_args);
- if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
+ if (reqsize < nbytes)
return -EINVAL;
- else if (reqsize > nbytes) {
+
+ if (args->out_argvar) {
struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
unsigned diffsize = reqsize - nbytes;
if (diffsize > lastarg->size)
return -EINVAL;
lastarg->size -= diffsize;
+
+ if (args->out_var_alloc) {
+ lastarg->value = kvmalloc(lastarg->size, GFP_KERNEL);
+ if (!lastarg->value)
+ return -ENOMEM;
+ }
+ } else if (reqsize > nbytes) {
+ return -EINVAL;
}
return fuse_copy_args(cs, args->out_numargs, args->out_pages,
args->out_args, args->page_zeroing);
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index be41c14ef329..cbe0d4b65d49 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -542,7 +542,7 @@ int fuse_valid_type(int m)
S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
}
-static bool fuse_valid_size(u64 size)
+bool fuse_valid_size(u64 size)
{
return size <= LLONG_MAX;
}
@@ -2485,7 +2485,7 @@ static int fuse_symlink_read_folio(struct file *null, struct folio *folio)
return err;
}
-static const struct address_space_operations fuse_symlink_aops = {
+const struct address_space_operations fuse_symlink_aops = {
.read_folio = fuse_symlink_read_folio,
};
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 3a7ac74a23ed..fe66281b7554 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -69,6 +69,9 @@ extern struct mutex fuse_mutex;
extern unsigned int max_user_bgreq;
extern unsigned int max_user_congthresh;
+extern struct kmem_cache *fuse_inode_cachep;
+extern const struct address_space_operations fuse_symlink_aops;
+
struct fuse_forget_link;
/**
@@ -911,6 +914,8 @@ struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
struct fuse_entry_out *outarg, struct inode **inode);
+void fuse_umount_begin(struct super_block *sb);
+
/*
* Initialize READ or READDIR request
*/
@@ -1102,6 +1107,7 @@ void fuse_ctl_remove_conn(struct fuse_conn *fc);
* Is file type valid?
*/
int fuse_valid_type(int m);
+bool fuse_valid_size(u64 size);
bool fuse_invalid_attr(struct fuse_attr *attr);
@@ -1204,6 +1210,8 @@ struct posix_acl *fuse_get_acl(struct mnt_idmap *idmap,
int fuse_set_acl(struct mnt_idmap *, struct dentry *dentry,
struct posix_acl *acl, int type);
+void fuse_convert_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr);
+
/* readdir.c */
int fuse_readdir(struct file *file, struct dir_context *ctx);
diff --git a/fs/fuse/fusex.c b/fs/fuse/fusex.c
new file mode 100644
index 000000000000..98e239e7e00e
--- /dev/null
+++ b/fs/fuse/fusex.c
@@ -0,0 +1,1751 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include "fusex.h"
+#include "dev.h"
+#include "fuse_i.h"
+
+#include <linux/fs_context.h>
+#include <linux/miscdevice.h>
+#include <linux/xxhash.h>
+#include <linux/pagemap.h>
+#include <linux/exportfs.h>
+#include <linux/iversion.h>
+#include <linux/posix_acl_xattr.h>
+#include <linux/statfs.h>
+#include <linux/falloc.h>
+#include <linux/fs_parser.h>
+
+static void fusex_init_inode(struct inode *inode);
+
+#define ADD_IN_ARG(_args, _size, _value) \
+ (*NEXT_IN_ARG(&(_args)) = (struct fuse_in_arg) { .size = (_size), .value = (_value) })
+
+#define ADD_IN_ARG_S(args, ptr) \
+ ADD_IN_ARG(args, sizeof(*(ptr)), ptr)
+
+#define ADD_IN_ARG_ZERO(args) \
+ ADD_IN_ARG(args, 0, NULL)
+
+#define ADD_OUT_ARG(_args, _size, _value) \
+ (*NEXT_OUT_ARG(&(_args)) = (struct fuse_arg) { .size = (_size), .value = (_value) })
+
+#define ADD_OUT_ARG_S(args, ptr) \
+ ADD_OUT_ARG(args, sizeof(*(ptr)), ptr)
+
+static struct fuse_in_arg *NEXT_IN_ARG(struct fuse_args *args)
+{
+ if (WARN_ON(args->in_numargs >= ARRAY_SIZE(args->in_args)))
+ return NULL;
+
+ return &args->in_args[args->in_numargs++];
+}
+
+static struct fuse_arg *NEXT_OUT_ARG(struct fuse_args *args)
+{
+ if (WARN_ON(args->out_numargs >= ARRAY_SIZE(args->out_args)))
+ return NULL;
+
+ return &args->out_args[args->out_numargs++];
+}
+
+struct fusex_id {
+ u64 nodeid;
+ /* will extend with file handle */
+};
+
+static int fusex_id_from_args(struct fuse_args *args, struct fusex_id *id)
+{
+ struct fuse_entryx_out *outarg = args->out_args[0].value;
+
+ /* will extract file handle */
+ id->nodeid = outarg->nodeid;
+ if (!id->nodeid)
+ return -EIO;
+
+ return 0;
+}
+
+static ssize_t fusex_inode_request(struct inode *inode, struct fuse_args *args)
+{
+ args->nodeid = get_node_id(inode);
+ /* will add file handle */
+ return fuse_simple_request(get_fuse_mount(inode), args);
+}
+
+static ssize_t fusex_id_request(struct inode *inode, const struct fusex_id *id,
+ struct fuse_args *args)
+{
+ args->nodeid = id->nodeid;
+ /* will add file handle */
+ return fuse_simple_request(get_fuse_mount(inode), args);
+}
+
+static ssize_t fusex_inode2_request(struct inode *inode, struct inode *inode2,
+ struct fuse_args *args)
+{
+ /* will add file handle for both inodes */
+ return fusex_inode_request(inode, args);
+}
+
+static struct inode *fusex_alloc_inode(struct super_block *sb)
+{
+ struct fuse_inode *fi;
+ struct fuse_forget_link *forget __free(kfree) = fuse_alloc_forget();
+
+ if (!forget)
+ return NULL;
+
+ fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
+ if (!fi)
+ return NULL;
+
+ /* Initialize private data (i.e. everything except fi->inode) */
+ BUILD_BUG_ON(offsetof(struct fuse_inode, inode) != 0);
+ memset((void *) fi + sizeof(fi->inode), 0, sizeof(*fi) - sizeof(fi->inode));
+
+ fi->forget = no_free_ptr(forget);
+ return &fi->inode;
+}
+
+static void fusex_free_inode(struct inode *inode)
+{
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ kfree(fi->forget);
+ kmem_cache_free(fuse_inode_cachep, fi);
+}
+
+static void fusex_evict_inode(struct inode *inode)
+{
+ struct fuse_conn *fc = get_fuse_conn(inode);
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ truncate_inode_pages_final(&inode->i_data);
+ clear_inode(inode);
+ if (fi->forget) {
+ fuse_chan_queue_forget(fc->chan, fi->forget, fi->nodeid, 1);
+ fi->forget = NULL;
+ }
+}
+
+static int fusex_send_setstatx(struct inode *inode, struct fuse_setstatx_in *inarg)
+{
+ FUSE_ARGS(args);
+ struct fuse_statx_out outarg;
+
+ args.opcode = FUSE_SETSTATX;
+ ADD_IN_ARG_S(args, inarg);
+ ADD_OUT_ARG_S(args, &outarg);
+
+ return fusex_inode_request(inode, &args);
+}
+
+static void fusex_get_atime(const struct inode *inode, struct fuse_statx *sx)
+{
+ sx->mask |= STATX_ATIME;
+ sx->atime.tv_sec = inode->i_atime_sec;
+ sx->atime.tv_nsec = inode->i_atime_nsec;
+}
+
+static void fusex_get_mtime(const struct inode *inode, struct fuse_statx *sx)
+{
+ sx->mask |= STATX_MTIME;
+ sx->mtime.tv_sec = inode->i_mtime_sec;
+ sx->mtime.tv_nsec = inode->i_mtime_nsec;
+}
+
+static void fusex_get_ctime(const struct inode *inode, struct fuse_statx *sx)
+{
+ sx->mask |= STATX_CTIME;
+ sx->ctime.tv_sec = inode->i_ctime_sec;
+ sx->ctime.tv_nsec = inode->i_ctime_nsec;
+}
+
+static int fusex_write_inode(struct inode *inode, struct writeback_control *wbc)
+{
+ struct fuse_setstatx_in inarg;
+
+ memset(&inarg, 0, sizeof(inarg));
+
+ fusex_get_atime(inode, &inarg.stat);
+ fusex_get_mtime(inode, &inarg.stat);
+ fusex_get_ctime(inode, &inarg.stat);
+
+ return fusex_send_setstatx(inode, &inarg);
+}
+
+static int fusex_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+ FUSE_ARGS(args);
+ struct fuse_statfs_out outarg;
+ int err;
+
+ args.opcode = FUSE_STATFS;
+ ADD_OUT_ARG_S(args, &outarg);
+ err = fusex_inode_request(d_inode(dentry), &args);
+ if (!err)
+ fuse_convert_statfs(buf, &outarg.st);
+ return err;
+}
+
+static const struct super_operations fusex_super_operations = {
+ .alloc_inode = fusex_alloc_inode,
+ .free_inode = fusex_free_inode,
+ .evict_inode = fusex_evict_inode,
+ .write_inode = fusex_write_inode,
+ .umount_begin = fuse_umount_begin,
+ .statfs = fusex_statfs,
+};
+
+static void fusex_set_times(struct inode *inode, struct fuse_statx *attr)
+{
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ /* Sanitize nsecs */
+ attr->atime.tv_nsec = min_t(u32, attr->atime.tv_nsec, NSEC_PER_SEC - 1);
+ attr->mtime.tv_nsec = min_t(u32, attr->mtime.tv_nsec, NSEC_PER_SEC - 1);
+ attr->ctime.tv_nsec = min_t(u32, attr->ctime.tv_nsec, NSEC_PER_SEC - 1);
+
+ inode_set_mtime(inode, attr->mtime.tv_sec, attr->mtime.tv_nsec);
+ inode_set_ctime(inode, attr->ctime.tv_sec, attr->ctime.tv_nsec);
+ inode_set_atime(inode, attr->atime.tv_sec, attr->atime.tv_nsec);
+
+ if (attr->mask & STATX_BTIME) {
+ set_bit(FUSE_I_BTIME, &fi->state);
+ fi->i_btime.tv_sec = attr->btime.tv_sec;
+ fi->i_btime.tv_nsec = attr->btime.tv_nsec;
+ }
+}
+
+static void fusex_set_attr(struct inode *inode, struct fuse_statx *attr)
+{
+ struct user_namespace *user_ns = i_user_ns(inode);
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ inode->i_mode = attr->mode;
+ inode->i_size = attr->size;
+ inode->i_blocks = attr->blocks;
+ inode->i_ino = fi->orig_ino = attr->ino;
+ inode->i_uid = make_kuid(user_ns, attr->uid);
+ inode->i_gid = make_kgid(user_ns, attr->gid);
+ fi->cached_i_blkbits = ilog2(attr->blksize);
+ if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
+ inode->i_rdev = MKDEV(attr->rdev_major, attr->rdev_minor);
+
+ if (S_ISDIR(attr->mode) && attr->nlink > 1)
+ attr->nlink = 1;
+ set_nlink(inode, attr->nlink);
+
+ fusex_set_times(inode, attr);
+}
+
+static int fusex_inode_eq(struct inode *inode, void *_id)
+{
+ struct fusex_id *id = _id;
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ return id->nodeid == fi->nodeid;
+}
+
+static int fusex_inode_set(struct inode *inode, void *_id)
+{
+ struct fusex_id *id = _id;
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ fi->nodeid = id->nodeid;
+
+ return 0;
+}
+
+static unsigned long fusex_hash_id(const struct fusex_id *id)
+{
+ return xxhash(&id->nodeid, sizeof(id->nodeid), 0);
+}
+
+static void fusex_fill_statx(struct fuse_args *args, struct fuse_statx_in *inarg,
+ struct fuse_statx_out *outarg)
+{
+ memset(inarg, 0, sizeof(*inarg));
+ inarg->sx_mask = STATX_BASIC_STATS | STATX_BTIME;
+
+ args->opcode = FUSE_STATX;
+ ADD_IN_ARG_S(*args, inarg);
+ ADD_OUT_ARG_S(*args, outarg);
+}
+
+static int fusex_send_statx(struct inode *inode, struct fuse_statx_out *outarg)
+{
+ FUSE_ARGS(args);
+ struct fuse_statx_in inarg;
+ int err;
+
+ fusex_fill_statx(&args, &inarg, outarg);
+ err = fusex_inode_request(inode, &args);
+ if (err)
+ return err;
+
+ if (!fuse_valid_type(outarg->stat.mode) || !fuse_valid_size(outarg->stat.size))
+ return -EIO;
+
+ return 0;
+}
+
+static struct inode *fusex_iget(struct super_block *sb, struct fusex_id *id)
+{
+ return iget5_locked(sb, fusex_hash_id(id), fusex_inode_eq, fusex_inode_set, id);
+}
+
+static struct inode *fusex_get_inode(struct super_block *sb, struct fusex_id *id)
+{
+ struct inode *inode = fusex_iget(sb, id);
+ struct fuse_statx_out statx;
+ int err;
+
+ if (!inode)
+ return ERR_PTR(-ENOMEM);
+
+ if (inode_state_read_once(inode) & I_NEW) {
+ inode_state_set(inode, I_DONTCACHE);
+ err = fusex_send_statx(inode, &statx);
+ if (err) {
+ discard_new_inode(inode);
+ return ERR_PTR(err);
+ }
+ fusex_set_attr(inode, &statx.stat);
+ fusex_init_inode(inode);
+ unlock_new_inode(inode);
+ }
+ return inode;
+}
+
+static void fusex_extend_file(struct inode *inode, loff_t old_size, loff_t new_size)
+{
+ WARN_ON(new_size > inode->i_size);
+
+ if (old_size < new_size)
+ truncate_pagecache_range(inode, old_size, new_size - 1);
+}
+
+static long fusex_file_fallocate(struct file *file, int mode, loff_t offset, loff_t length)
+{
+ struct inode *inode = file_inode(file);
+ FUSE_ARGS(args);
+ struct fuse_fallocate_in inarg = {
+ .offset = offset,
+ .length = length,
+ .mode = mode
+ };
+ off_t end = offset + length;
+ int err;
+
+ if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
+ FALLOC_FL_ZERO_RANGE))
+ return -EOPNOTSUPP;
+
+ guard(rwsem_write)(&inode->i_rwsem);
+
+ if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) {
+ err = filemap_write_and_wait_range(inode->i_mapping, offset, end - 1);
+ if (err)
+ return err;
+ }
+
+ if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->i_size) {
+ err = inode_newsize_ok(inode, end);
+ if (err)
+ return err;
+ }
+
+ err = file_modified(file);
+ if (err)
+ return err;
+
+ args.opcode = FUSE_FALLOCATE;
+ args.in_numargs = 1;
+ ADD_IN_ARG_S(args, &inarg);
+ err = fusex_inode_request(inode, &args);
+ if (err)
+ return err;
+
+ if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->i_size) {
+ loff_t old_size = inode->i_size;
+
+ i_size_write(inode, end);
+ fusex_extend_file(inode, old_size, inode->i_size);
+ }
+
+ if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
+ truncate_pagecache_range(inode, offset, end - 1);
+
+ return 0;
+}
+
+static const struct file_operations fusex_file_operations = {
+ .read_iter = generic_file_read_iter,
+ .write_iter = generic_file_write_iter,
+ .splice_read = filemap_splice_read,
+ .splice_write = iter_file_splice_write,
+ .llseek = generic_file_llseek,
+ .mmap_prepare = generic_file_mmap_prepare,
+ .fsync = simple_fsync_noflush,
+ .fallocate = fusex_file_fallocate,
+};
+
+static int fusex_send_read(struct inode *inode, loff_t pos, struct file *file,
+ struct folio *folio, unsigned int off, unsigned int len)
+{
+ struct fuse_args_pages ap = {};
+ struct fuse_folio_desc desc = { .offset = off, .length = len };
+ struct fuse_read_in inarg;
+ ssize_t res;
+
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.offset = pos;
+ inarg.size = len;
+ inarg.flags = file->f_flags;
+ ap.args.opcode = FUSE_READ;
+ ADD_IN_ARG_S(ap.args, &inarg);
+ ADD_OUT_ARG(ap.args, len, NULL);
+ ap.args.out_argvar = true;
+ ap.args.out_pages = true;
+ ap.num_folios = 1;
+ ap.folios = &folio;
+ ap.descs = &desc;
+
+ res = fusex_inode_request(inode, &ap.args);
+ if (res < 0)
+ return res;
+
+ WARN_ON(res > len);
+ if (res < len)
+ folio_zero_segment(folio, off + res, off + len);
+
+ return 0;
+}
+
+static int fusex_do_read_folio(struct file *file, struct folio *folio)
+{
+ struct inode *inode = folio->mapping->host;
+ loff_t folio_start = folio_pos(folio);
+ loff_t i_size = i_size_read(inode);
+ size_t full_len = folio_size(folio), len = full_len;
+ int err;
+
+ WARN_ON(i_size <= folio_start);
+
+ if (i_size < folio_start + full_len) {
+ len = i_size - folio_start;
+ folio_zero_segment(folio, len, full_len);
+ }
+ err = fusex_send_read(inode, folio_start, file, folio, 0, len);
+ if (err)
+ return err;
+
+ folio_mark_uptodate(folio);
+
+ return 0;
+}
+
+static int fusex_read_folio(struct file *file, struct folio *folio)
+{
+ int err = fusex_do_read_folio(file, folio);
+
+ folio_unlock(folio);
+ return err;
+}
+
+static int fusex_send_write(struct inode *inode, loff_t pos,
+ struct folio *folio, unsigned int off, unsigned int len)
+{
+ struct fuse_args_pages ap = {};
+ struct fuse_folio_desc desc = { .offset = off, .length = len };
+ struct fuse_write_in inarg;
+ struct fuse_write_out outarg;
+ int err;
+
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.offset = pos;
+ inarg.size = len;
+
+ ap.args.opcode = FUSE_WRITE;
+ ADD_IN_ARG_S(ap.args, &inarg);
+ ADD_IN_ARG(ap.args, len, NULL);
+ ap.args.in_pages = true;
+ ap.num_folios = 1;
+ ap.folios = &folio;
+ ap.descs = &desc;
+ ADD_OUT_ARG_S(ap.args, &outarg);
+
+ err = fusex_inode_request(inode, &ap.args);
+ if (err)
+ return err;
+
+ if (outarg.size != len)
+ return -EIO;
+
+ return 0;
+}
+
+static int fusex_writepages(struct address_space *mapping, struct writeback_control *wbc)
+{
+ struct folio *folio = NULL;
+ int err;
+
+ while ((folio = writeback_iter(mapping, wbc, folio, &err))) {
+ struct inode *inode = folio->mapping->host;
+ loff_t folio_start = folio_pos(folio);
+ loff_t i_size = i_size_read(inode);
+ size_t full_len = folio_size(folio), len = full_len;
+
+ if (folio_start < i_size) {
+ if (i_size < folio_start + full_len)
+ len = i_size - folio_start;
+
+ err = fusex_send_write(inode, folio_start, folio, 0, len);
+ }
+ folio_unlock(folio);
+ }
+
+ return err;
+}
+
+static int fusex_write_begin(const struct kiocb *iocb, struct address_space *mapping,
+ loff_t pos, unsigned int len,
+ struct folio **foliop, void **fsdata)
+{
+ struct folio *folio;
+
+ folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
+ mapping_gfp_mask(mapping));
+ if (IS_ERR(folio))
+ return PTR_ERR(folio);
+
+ if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
+ if (folio->mapping->host->i_size <= folio_pos(folio)) {
+ folio_zero_segment(folio, 0, folio_size(folio));
+ folio_mark_uptodate(folio);
+ } else {
+ int err = fusex_do_read_folio(iocb->ki_filp, folio);
+ if (err) {
+ folio_unlock(folio);
+ folio_put(folio);
+ return err;
+ }
+ }
+ }
+ *foliop = folio;
+
+ return 0;
+}
+
+static int fusex_write_end(const struct kiocb *iocb, struct address_space *mapping,
+ loff_t pos, unsigned int len, unsigned int copied,
+ struct folio *folio, void *fsdata)
+{
+ struct inode *inode = folio->mapping->host;
+ loff_t old_size = inode->i_size;
+ loff_t end_pos = pos + copied;
+ int err = 0;
+
+ if (!folio_test_uptodate(folio)) {
+ if (copied < len) {
+ size_t off = offset_in_folio(folio, end_pos);
+
+ err = fusex_send_read(inode, end_pos, iocb->ki_filp, folio,
+ off, len - copied);
+ if (err)
+ goto out;
+ }
+ folio_mark_uptodate(folio);
+ }
+ if (end_pos > old_size)
+ i_size_write(inode, end_pos);
+out:
+ folio_mark_dirty(folio);
+ folio_unlock(folio);
+ folio_put(folio);
+
+ if (!err)
+ fusex_extend_file(inode, old_size, pos);
+
+ return err ? err : copied;
+}
+
+static ssize_t fusex_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
+{
+ struct inode *inode = file_inode(iocb->ki_filp);
+ loff_t old_size = i_size_read(inode);
+ loff_t pos = iocb->ki_pos;
+ int err = 0;
+
+ if (!iov_iter_count(iter) || (iov_iter_rw(iter) == READ && pos >= old_size))
+ return 0;
+
+ for (;;) {
+ struct page **pages = NULL;
+ struct folio *folio;
+ ssize_t len;
+ size_t off;
+
+ len = iov_iter_extract_pages(iter, &pages, PAGE_SIZE, 1, 0, &off);
+ if (len <= 0) {
+ err = len;
+ break;
+ }
+
+ folio = page_folio(pages[0]);
+ off += folio_page_idx(folio, pages[0]) * PAGE_SIZE;
+ kvfree(pages);
+
+ if (iov_iter_rw(iter) == WRITE) {
+ err = fusex_send_write(inode, pos, folio, off, len);
+ } else {
+ if (pos + len > old_size)
+ len = old_size - pos;
+
+ err = fusex_send_read(inode, pos, iocb->ki_filp, folio, off, len);
+ if (!err && user_backed_iter(iter))
+ folio_mark_dirty_lock(folio);
+ }
+ if (iov_iter_extract_will_pin(iter))
+ unpin_folio(folio);
+
+ if (err || !len)
+ break;
+ pos += len;
+ }
+ if (pos > iocb->ki_pos) {
+ if (iov_iter_rw(iter) == WRITE) {
+ if (pos > old_size)
+ i_size_write(inode, pos);
+ fusex_extend_file(inode, old_size, iocb->ki_pos);
+ }
+ }
+
+ return (pos - iocb->ki_pos) ?: err;
+}
+
+static const struct address_space_operations fusex_file_aops = {
+ .read_folio = fusex_read_folio,
+ .writepages = fusex_writepages,
+ .write_begin = fusex_write_begin,
+ .write_end = fusex_write_end,
+ .direct_IO = fusex_direct_IO,
+ .dirty_folio = filemap_dirty_folio,
+ .migrate_folio = filemap_migrate_folio,
+};
+
+static void fusex_dir_modified(struct inode *dir)
+{
+ inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
+ inode_inc_iversion(dir);
+ __mark_inode_dirty(dir, I_DIRTY_SYNC);
+}
+
+static void fusex_update_ctime(struct inode *inode)
+{
+ inode_set_ctime_current(inode);
+ __mark_inode_dirty(inode, I_DIRTY_SYNC);
+}
+
+static struct inode *fusex_do_lookup(struct inode *base, const struct qstr *name)
+{
+ FUSE_ARGS(args);
+ struct fuse_entryx_out outarg;
+ struct fusex_id id;
+ int err;
+
+ args.opcode = FUSE_LOOKUPX;
+ ADD_IN_ARG_ZERO(args);
+ ADD_IN_ARG(args, name->len + 1, name->name);
+ ADD_OUT_ARG_S(args, &outarg);
+
+ err = fusex_inode_request(base, &args);
+ if (err < 0)
+ return ERR_PTR(err);
+
+ if (outarg.flags & FUSE_ENTRYX_NEGATIVE) {
+ if (err > 0)
+ return ERR_PTR(-EIO);
+ return NULL;
+ }
+
+ err = fusex_id_from_args(&args, &id);
+ if (err)
+ return ERR_PTR(err);
+
+ return fusex_get_inode(base->i_sb, &id);
+}
+
+
+static struct dentry *fusex_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
+{
+ struct inode *inode = fusex_do_lookup(dir, &dentry->d_name);
+
+ return d_splice_alias(inode, dentry);
+}
+
+static int fusex_getattr(struct mnt_idmap *idmap, const struct path *path, struct kstat *stat,
+ u32 request_mask, unsigned int flags)
+{
+ struct inode *inode = d_inode(path->dentry);
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ generic_fillattr(idmap, request_mask, inode, stat);
+ stat->ino = fi->orig_ino;
+ stat->blksize = 1 << fi->cached_i_blkbits;
+ if (test_bit(FUSE_I_BTIME, &fi->state)) {
+ stat->btime = fi->i_btime;
+ stat->result_mask |= STATX_BTIME;
+ }
+
+ return 0;
+}
+
+static void *fusex_send_lgxattr(struct inode *inode, const char *name, size_t *sizep)
+{
+ FUSE_ARGS(args);
+ struct fuse_getxattr_in inarg;
+ ssize_t res;
+
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.size = name ? XATTR_SIZE_MAX : XATTR_LIST_MAX;
+
+ args.opcode = name ? FUSE_GETXATTR : FUSE_LISTXATTR;
+ ADD_IN_ARG_S(args, &inarg);
+ if (name)
+ ADD_IN_ARG(args, strlen(name) + 1, name);
+ ADD_OUT_ARG(args, inarg.size, NULL);
+ args.out_argvar = true;
+ args.out_var_alloc = true;
+
+ res = fusex_inode_request(inode, &args);
+ if (res < 0) {
+ kfree(args.out_args[0].value);
+ if (res == -ENOSYS)
+ res = -EOPNOTSUPP;
+ return ERR_PTR(res);
+ }
+
+ *sizep = res;
+
+ return args.out_args[0].value;
+}
+
+static bool fusex_verify_xattr_list(char *list, size_t size)
+{
+ while (size) {
+ size_t thislen = strnlen(list, size);
+
+ if (!thislen || thislen == size)
+ return false;
+
+ size -= thislen + 1;
+ list += thislen + 1;
+ }
+ return true;
+}
+
+static char *fusex_send_listxattr(struct inode *inode, size_t *gotsize)
+{
+ char *list = fusex_send_lgxattr(inode, NULL, gotsize);
+
+ if (IS_ERR(list))
+ return list;
+
+ if (!fusex_verify_xattr_list(list, *gotsize)) {
+ kfree(list);
+ return ERR_PTR(-EIO);
+ }
+
+ return list;
+}
+
+static ssize_t fusex_listxattr(struct dentry *dentry, char *list, size_t size)
+{
+ struct inode *inode = d_inode(dentry);
+ ssize_t res;
+ char *gotlist __free(kfree) = fusex_send_listxattr(inode, &res);
+
+ if (IS_ERR(gotlist))
+ return PTR_ERR(gotlist);
+
+ if (size) {
+ if (size < res)
+ res = -ERANGE;
+ else
+ memcpy(list, gotlist, res);
+ }
+
+ return res;
+}
+
+static struct posix_acl *fusex_get_acl(struct inode *inode, int type, bool rcu)
+{
+ struct user_namespace *user_ns = i_user_ns(inode);
+ const char *name = posix_acl_xattr_name(type);
+ size_t size;
+ void *value __free(kfree) = fusex_send_lgxattr(inode, name, &size);
+ struct posix_acl *acl = NULL;
+
+ WARN_ON(rcu);
+
+ if (IS_ERR(value)) {
+ switch (PTR_ERR(value)) {
+ case -ENODATA:
+ case -EOPNOTSUPP:
+ value = NULL;
+ break;
+ default:
+ return ERR_CAST(value);
+ }
+ }
+ if (value)
+ acl = posix_acl_from_xattr(user_ns, value, size);
+
+ if (!IS_ERR(acl))
+ set_cached_acl(inode, type, acl);
+
+ return acl;
+}
+
+static void fusex_fill_setxattr(struct fuse_args *args, struct fuse_setxattr_in *inarg,
+ const char *name, const void *value, size_t size, int flags)
+{
+ if (value) {
+ memset(inarg, 0, sizeof(*inarg));
+ inarg->size = size;
+ inarg->flags = flags;
+
+ args->opcode = FUSE_SETXATTR;
+ ADD_IN_ARG_S(*args, inarg);
+ ADD_IN_ARG(*args, strlen(name) + 1, name);
+ ADD_IN_ARG(*args, size, value);
+ } else {
+ args->opcode = FUSE_REMOVEXATTR;
+ ADD_IN_ARG_ZERO(*args);
+ ADD_IN_ARG(*args, strlen(name) + 1, name);
+ }
+}
+
+
+static int fusex_send_setxattr(struct inode *inode, const char *name, const void *value,
+ size_t size, int flags)
+{
+ FUSE_ARGS(args);
+ struct fuse_setxattr_in inarg;
+ int err;
+
+ fusex_fill_setxattr(&args, &inarg, name, value, size, flags);
+ err = fusex_inode_request(inode, &args);
+ if (err) {
+ if (err == -ENOSYS)
+ err = -EOPNOTSUPP;
+ return err;
+ }
+
+ fusex_update_ctime(inode);
+
+ return 0;
+}
+
+static int fusex_send_setacl(struct inode *inode, int type, struct posix_acl *acl)
+{
+ struct user_namespace *user_ns = i_user_ns(inode);
+ const char *name = posix_acl_xattr_name(type);
+ size_t size;
+
+ if (!acl)
+ return fusex_send_setxattr(inode, name, NULL, 0, 0);
+
+ void *value __free(kfree) = posix_acl_to_xattr(user_ns, acl, &size, GFP_KERNEL);
+ if (!value)
+ return -ENOMEM;
+
+ return fusex_send_setxattr(inode, name, value, size, 0);
+}
+
+static int fusex_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
+ struct posix_acl *acl, int type)
+{
+ struct inode *inode = d_inode(dentry);
+ umode_t mode = inode->i_mode;
+ bool update_mode;
+ int err;
+
+ if (type == ACL_TYPE_ACCESS && acl) {
+ err = posix_acl_update_mode(idmap, inode, &mode, &acl);
+ if (err)
+ return err;
+ update_mode = true;
+ }
+
+ err = fusex_send_setacl(inode, type, acl);
+ if (err)
+ return err;
+
+ set_cached_acl(inode, type, acl);
+ inode_set_ctime_current(inode);
+ if (update_mode) {
+ struct fuse_setstatx_in inarg;
+
+ inode->i_mode = mode;
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.stat.mask |= STATX_MODE;
+ inarg.stat.mode = inode->i_mode & 07777;
+ fusex_get_ctime(inode, &inarg.stat);
+
+ err = fusex_send_setstatx(inode, &inarg);
+ } else {
+ __mark_inode_dirty(inode, I_DIRTY_SYNC);
+ }
+
+ return err;
+}
+
+static int fusex_send_mkobjx(struct inode *dir, struct inode *inode,
+ const struct qstr *name, const char *link_body, struct fusex_id *id)
+{
+ FUSE_ARGS(args);
+ struct fuse_mkobjx_in inarg;
+ struct fuse_entryx_out outarg;
+ struct user_namespace *u = i_user_ns(inode);
+ int flags = 0;
+ int err;
+
+ if (!name->len) {
+ WARN_ON((inode->i_mode & S_IFMT) != S_IFREG);
+ flags |= FUSE_MKOBJX_TMPFILE;
+ }
+
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.namesize = name->len + 1;
+ inarg.stat.mask = STATX_UID | STATX_GID | STATX_MODE | STATX_TYPE | STATX_BTIME;
+ inarg.stat.uid = from_kuid(u, inode->i_uid);
+ inarg.stat.gid = from_kgid(u, inode->i_gid);
+ inarg.stat.rdev_major = MAJOR(inode->i_rdev);
+ inarg.stat.rdev_minor = MINOR(inode->i_rdev);
+ inarg.stat.mode = inode->i_mode;
+ fusex_get_atime(inode, &inarg.stat);
+ fusex_get_mtime(inode, &inarg.stat);
+ fusex_get_ctime(inode, &inarg.stat);
+ inarg.stat.btime = inarg.stat.ctime;
+ inarg.flags = flags;
+
+ args.opcode = FUSE_MKOBJX;
+ ADD_IN_ARG_S(args, &inarg);
+ ADD_IN_ARG(args, inarg.namesize, name->name);
+ if (S_ISLNK(inode->i_mode))
+ ADD_IN_ARG(args, strlen(link_body) + 1, link_body);
+ ADD_OUT_ARG_S(args, &outarg);
+
+ err = fusex_inode_request(dir, &args);
+ if (err)
+ return err;
+
+ return fusex_id_from_args(&args, id);
+}
+
+static int fusex_set_initial_acl(struct inode *inode, const struct fusex_id *id, int type)
+{
+ struct posix_acl *acl = (type == ACL_TYPE_ACCESS) ? inode->i_acl : inode->i_default_acl;
+ if (!acl)
+ return 0;
+
+ FUSE_ARGS(args);
+ struct fuse_setxattr_in inarg;
+ size_t size;
+ const char *name = posix_acl_xattr_name(type);
+ const void *value __free(kfree) =
+ posix_acl_to_xattr(i_user_ns(inode), acl, &size, GFP_KERNEL);
+ if (!value)
+ return -ENOMEM;
+
+ fusex_fill_setxattr(&args, &inarg, name, value, size, 0);
+
+ return fusex_id_request(inode, id, &args);
+}
+
+static int fusex_do_mkobjx(struct inode *dir, struct inode *inode, const struct qstr *name,
+ const char *link_body, struct fuse_statx_out *outarg_sx)
+{
+ FUSE_ARGS(args_sx);
+ struct fuse_statx_in inarg_sx;
+ struct inode *old;
+ struct fusex_id id;
+ int err;
+
+ err = fusex_send_mkobjx(dir, inode, name, link_body, &id);
+ if (err)
+ return err;
+
+ err = fusex_set_initial_acl(inode, &id, ACL_TYPE_ACCESS);
+ if (err)
+ return err;
+
+ err = fusex_set_initial_acl(inode, &id, ACL_TYPE_DEFAULT);
+ if (err)
+ return err;
+
+ fusex_fill_statx(&args_sx, &inarg_sx, outarg_sx);
+ err = fusex_id_request(inode, &id, &args_sx);
+ if (err)
+ return err;
+
+ old = inode_insert5(inode, fusex_hash_id(&id), fusex_inode_eq, fusex_inode_set, &id);
+ if (old != inode) {
+ iput(old);
+ return -EBUSY;
+ }
+ return 0;
+}
+
+static int fusex_setup_new_inode(struct inode *inode, struct fuse_statx *statx)
+{
+ struct fuse_inode *fi;
+
+ if (inode->i_mode != statx->mode)
+ return -EIO;
+ if (!uid_eq(inode->i_uid, make_kuid(i_user_ns(inode), statx->uid)))
+ return -EIO;
+ if (!gid_eq(inode->i_gid, make_kgid(i_user_ns(inode), statx->gid)))
+ return -EIO;
+ if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
+ if (MKDEV(statx->rdev_major, statx->rdev_minor) != inode->i_rdev)
+ return -EIO;
+ }
+
+ fi = get_fuse_inode(inode);
+ inode->i_size = statx->size;
+ inode->i_blocks = statx->blocks;
+ inode->i_ino = fi->orig_ino = statx->ino;
+ fi->cached_i_blkbits = ilog2(statx->blksize);
+ fusex_set_times(inode, statx);
+
+ return 0;
+}
+
+static struct inode *fusex_new_inode(struct mnt_idmap *idmap, struct inode *dir,
+ const struct qstr *name, umode_t mode, dev_t rdev,
+ const char *link_body)
+{
+ struct inode *inode;
+ struct fuse_statx_out statx;
+ int err;
+
+ inode = new_inode(dir->i_sb);
+ if (!inode)
+ return ERR_PTR(-ENOMEM);
+
+ inode_init_owner(idmap, inode, dir, mode);
+ if (S_ISCHR(mode) || S_ISBLK(mode))
+ inode->i_rdev = rdev;
+
+ simple_inode_init_ts(inode);
+ fusex_init_inode(inode);
+ inode->i_size = 0;
+
+ err = posix_acl_create(dir, &inode->i_mode, &inode->i_default_acl, &inode->i_acl);
+ if (err)
+ goto iput_noforget;
+
+ err = fusex_do_mkobjx(dir, inode, name, link_body, &statx);
+ if (err)
+ goto iput_noforget;
+
+ err = fusex_setup_new_inode(inode, &statx.stat);
+ if (err) {
+ discard_new_inode(inode);
+ return ERR_PTR(err);
+ }
+
+ return inode;
+
+iput_noforget:
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ kfree(fi->forget);
+ fi->forget = NULL;
+ iput(inode);
+ return ERR_PTR(err);
+}
+
+static int fusex_mkobj(struct mnt_idmap *idmap, struct inode *dir,
+ struct dentry *dentry, umode_t mode, dev_t rdev, const char *link_body)
+{
+ struct inode *inode = fusex_new_inode(idmap, dir, &dentry->d_name, mode, rdev, link_body);
+ if (IS_ERR(inode))
+ return PTR_ERR(inode);
+
+ fusex_dir_modified(dir);
+ d_instantiate_new(dentry, inode);
+
+ return 0;
+}
+
+static int fusex_create(struct mnt_idmap *idmap, struct inode *dir,
+ struct dentry *dentry, umode_t mode, bool excl)
+{
+ return fusex_mkobj(idmap, dir, dentry, mode, 0, NULL);
+}
+
+static int fusex_mknod(struct mnt_idmap *idmap, struct inode *dir,
+ struct dentry *dentry, umode_t mode, dev_t rdev)
+{
+ return fusex_mkobj(idmap, dir, dentry, mode, rdev, NULL);
+}
+
+static struct dentry *fusex_mkdir(struct mnt_idmap *idmap, struct inode *dir,
+ struct dentry *dentry, umode_t mode)
+{
+ int err = fusex_mkobj(idmap, dir, dentry, S_IFDIR | mode, 0, NULL);
+
+ return ERR_PTR(err);
+}
+
+static int fusex_symlink(struct mnt_idmap *idmap, struct inode *dir,
+ struct dentry *dentry, const char *link_body)
+{
+ return fusex_mkobj(idmap, dir, dentry, S_IFLNK | 0777, 0, link_body);
+}
+
+static int fusex_tmpfile(struct mnt_idmap *idmap, struct inode *dir, struct file *file,
+ umode_t mode)
+{
+ struct inode *inode = fusex_new_inode(idmap, dir, &empty_name, mode, 0, NULL);
+ if (IS_ERR(inode))
+ return PTR_ERR(inode);
+
+ d_tmpfile(file, inode);
+ unlock_new_inode(inode);
+ return finish_open_simple(file, 0);
+}
+
+static int fusex_remove(struct inode *dir, struct dentry *dentry)
+{
+ struct inode *inode = d_inode(dentry);
+ FUSE_ARGS(args);
+ int err;
+
+ rwsem_assert_held_write(&inode->i_rwsem);
+
+ args.opcode = d_is_dir(dentry) ? FUSE_RMDIR : FUSE_UNLINK;
+ ADD_IN_ARG_ZERO(args);
+ ADD_IN_ARG(args, dentry->d_name.len + 1, dentry->d_name.name);
+
+ err = fusex_inode_request(dir, &args);
+ if (err)
+ return err;
+
+ drop_nlink(inode);
+ fusex_update_ctime(inode);
+ fusex_dir_modified(dir);
+
+ return 0;
+}
+
+static int fusex_setattr(struct mnt_idmap *idmap, struct dentry *dentry, struct iattr *iattr)
+{
+ struct inode *inode = d_inode(dentry);
+ struct user_namespace *user_ns = i_user_ns(inode);
+ struct fuse_setstatx_in inarg;
+ int err;
+
+ err = setattr_prepare(idmap, dentry, iattr);
+ if (err)
+ return err;
+
+ if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size != inode->i_size)
+ iattr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
+
+ memset(&inarg, 0, sizeof(inarg));
+ if (iattr->ia_valid & ATTR_SIZE) {
+ inarg.stat.mask |= STATX_SIZE;
+ inarg.stat.size = iattr->ia_size;
+ }
+ if (iattr->ia_valid & ATTR_UID) {
+ inarg.stat.mask |= STATX_UID;
+ inarg.stat.uid = from_kuid(user_ns, from_vfsuid(idmap, user_ns, iattr->ia_vfsuid));
+ }
+ if (iattr->ia_valid & ATTR_GID) {
+ inarg.stat.mask |= STATX_GID;
+ inarg.stat.gid = from_kgid(user_ns, from_vfsgid(idmap, user_ns, iattr->ia_vfsgid));
+ }
+ if (iattr->ia_valid & ATTR_MODE) {
+ inarg.stat.mask |= STATX_MODE;
+ err = posix_acl_chmod(idmap, dentry, iattr->ia_mode);
+ if (err)
+ return err;
+
+ inarg.stat.mode = iattr->ia_mode & 07777;
+ }
+ if (iattr->ia_valid & ATTR_ATIME) {
+ inarg.stat.mask |= STATX_ATIME;
+ inarg.stat.atime.tv_sec = iattr->ia_atime.tv_sec;
+ inarg.stat.atime.tv_nsec = iattr->ia_atime.tv_nsec;
+ }
+ if (iattr->ia_valid & ATTR_CTIME) {
+ inarg.stat.mask |= STATX_CTIME;
+ inarg.stat.ctime.tv_sec = iattr->ia_ctime.tv_sec;
+ inarg.stat.ctime.tv_nsec = iattr->ia_ctime.tv_nsec;
+ }
+ if (iattr->ia_valid & ATTR_MTIME) {
+ inarg.stat.mask |= STATX_MTIME;
+ inarg.stat.mtime.tv_sec = iattr->ia_mtime.tv_sec;
+ inarg.stat.mtime.tv_nsec = iattr->ia_mtime.tv_nsec;
+ }
+ err = fusex_send_setstatx(inode, &inarg);
+ if (err)
+ return err;
+
+ setattr_copy(idmap, inode, iattr);
+ if (iattr->ia_valid & ATTR_SIZE) {
+ loff_t old_size = inode->i_size;
+
+ i_size_write(inode, iattr->ia_size);
+ fusex_extend_file(inode, old_size, inode->i_size);
+ truncate_pagecache(inode, inode->i_size);
+ }
+
+ return 0;
+}
+
+static int fusex_rename(struct mnt_idmap *idmap, struct inode *olddir,
+ struct dentry *olddentry, struct inode *newdir,
+ struct dentry *newdentry, unsigned int flags)
+{
+ struct inode *oldinode = d_inode(olddentry);
+ struct inode *newinode = d_inode(newdentry);
+ struct fuse_rename2_in inarg;
+ FUSE_ARGS(args);
+ int err;
+
+ if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
+ return -EINVAL;
+
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.newdir = get_node_id(newdir);
+ inarg.flags = flags;
+
+ args.opcode = FUSE_RENAME2;
+ ADD_IN_ARG_S(args, &inarg);
+ ADD_IN_ARG(args, olddentry->d_name.len + 1, olddentry->d_name.name);
+ ADD_IN_ARG(args, newdentry->d_name.len + 1, newdentry->d_name.name);
+
+ err = fusex_inode2_request(olddir, newdir, &args);
+ if (err)
+ return err;
+
+ fusex_update_ctime(oldinode);
+ if (newinode) {
+ if (!(flags & RENAME_EXCHANGE))
+ drop_nlink(newinode);
+ fusex_update_ctime(newinode);
+ }
+
+ fusex_dir_modified(olddir);
+ fusex_dir_modified(newdir);
+
+ return 0;
+}
+
+static int fusex_link(struct dentry *dentry, struct inode *newdir, struct dentry *newdentry)
+{
+ struct fuse_link_in inarg;
+ struct inode *inode = d_inode(dentry);
+ FUSE_ARGS(args);
+ int err;
+
+ memset(&inarg, 0, sizeof(inarg));
+ inarg.oldnodeid = get_node_id(inode);
+
+ args.opcode = FUSE_LINK;
+ ADD_IN_ARG_S(args, &inarg);
+ ADD_IN_ARG(args, newdentry->d_name.len + 1, newdentry->d_name.name);
+
+ err = fusex_inode2_request(newdir, inode, &args);
+ if (err)
+ return err;
+
+ ihold(inode);
+ inc_nlink(inode);
+ fusex_update_ctime(inode);
+ fusex_dir_modified(newdir);
+ d_instantiate(newdentry, inode);
+
+ return 0;
+}
+
+static const struct inode_operations fusex_file_inode_operations = {
+ .getattr = fusex_getattr,
+ .setattr = fusex_setattr,
+ .get_inode_acl = fusex_get_acl,
+ .set_acl = fusex_set_acl,
+ .listxattr = fusex_listxattr,
+};
+
+static const struct inode_operations fusex_dir_inode_operations = {
+ .getattr = fusex_getattr,
+ .setattr = fusex_setattr,
+ .get_inode_acl = fusex_get_acl,
+ .set_acl = fusex_set_acl,
+ .listxattr = fusex_listxattr,
+
+ .lookup = fusex_lookup,
+ .create = fusex_create,
+ .tmpfile = fusex_tmpfile,
+ .mkdir = fusex_mkdir,
+ .mknod = fusex_mknod,
+ .symlink = fusex_symlink,
+ .unlink = fusex_remove,
+ .rmdir = fusex_remove,
+ .rename = fusex_rename,
+ .link = fusex_link,
+};
+
+static const struct inode_operations fusex_symlink_inode_operations = {
+ .getattr = fusex_getattr,
+ .get_link = page_get_link,
+ .listxattr = fusex_listxattr,
+};
+
+static struct fuse_file *fusex_file_alloc(void)
+{
+ struct fuse_file *ff;
+
+ ff = kzalloc(sizeof(*ff) + sizeof(*ff->args), GFP_KERNEL_ACCOUNT);
+ if (ff)
+ ff->args = (void *)(ff + 1);
+
+ return ff;
+}
+
+static int fusex_send_opendir(struct fuse_file *ff, struct inode *inode)
+{
+ struct fuse_open_in inarg;
+ struct fuse_open_out outarg;
+ FUSE_ARGS(args);
+ int err;
+
+ memset(&inarg, 0, sizeof(inarg));
+
+ args.opcode = FUSE_OPENDIR;
+ ADD_IN_ARG_S(args, &inarg);
+ ADD_OUT_ARG_S(args, &outarg);
+
+ err = fusex_inode_request(inode, &args);
+ if (!err) {
+ ff->fh = outarg.fh;
+ ff->open_flags = FOPEN_CACHE_DIR;
+ }
+ return err;
+}
+
+static int fusex_dir_open(struct inode *inode, struct file *file)
+{
+ struct fuse_file *ff __free(kfree) = fusex_file_alloc();
+ struct fuse_release_args *ra;
+ int err;
+
+ if (!ff)
+ return -ENOMEM;
+
+ ra = &ff->args->release_args;
+ ADD_IN_ARG_S(ra->args, &ra->inarg);
+
+ err = fusex_send_opendir(ff, inode);
+ if (err)
+ return err;
+
+ file->private_data = no_free_ptr(ff);
+ return 0;
+}
+
+static void fusex_release_end(struct fuse_args *args, int error)
+{
+ struct fuse_release_args *ra = container_of(args, typeof(*ra), args);
+ struct fuse_file *ff = (struct fuse_file *) ra - 1;
+
+ iput(ra->inode);
+ kfree(ff);
+}
+
+static int fusex_dir_release(struct inode *inode, struct file *file)
+{
+ struct fuse_mount *fm = get_fuse_mount(inode);
+ struct fuse_file *ff = file->private_data;
+ struct fuse_release_args *ra = &ff->args->release_args;
+
+ ra->inarg.fh = ff->fh;
+
+ ra->args.opcode = FUSE_RELEASEDIR;
+ ra->args.force = true;
+ ra->args.nocreds = true;
+ ra->args.end = fusex_release_end;
+ ra->inode = igrab(inode);
+
+ if (fuse_simple_background(fm, &ra->args, GFP_KERNEL | __GFP_NOFAIL))
+ fusex_release_end(&ra->args, -ENOTCONN);
+
+ return 0;
+}
+
+static const struct file_operations fusex_dir_operations = {
+ .llseek = generic_file_llseek,
+ .read = generic_read_dir,
+ .iterate_shared = fuse_readdir,
+ .open = fusex_dir_open,
+ .release = fusex_dir_release,
+ .fsync = simple_fsync_noflush,
+};
+
+static int fusex_symlink_read_folio(struct file *null, struct folio *folio)
+{
+ struct inode *inode = folio->mapping->host;
+ struct fuse_folio_desc desc = { .length = folio_size(folio) - 1 };
+ struct fuse_args_pages ap = {};
+ ssize_t res;
+
+ ap.args.opcode = FUSE_READLINK;
+ ADD_IN_ARG_ZERO(ap.args);
+ ADD_OUT_ARG(ap.args, desc.length, NULL);
+ ap.args.out_pages = true;
+ ap.args.out_argvar = true;
+ ap.args.page_zeroing = true;
+ ap.num_folios = 1;
+ ap.folios = &folio;
+ ap.descs = &desc;
+ res = fusex_inode_request(inode, &ap.args);
+ if (res >= 0) {
+ folio_mark_uptodate(folio);
+ res = 0;
+ }
+ folio_unlock(folio);
+ return res;
+}
+
+static const struct address_space_operations fusex_symlink_aops = {
+ .read_folio = fusex_symlink_read_folio,
+};
+
+static void fusex_init_inode(struct inode *inode)
+{
+ struct fuse_inode *fi = get_fuse_inode(inode);
+
+ switch (inode->i_mode & S_IFMT) {
+ case S_IFREG:
+ inode->i_op = &fusex_file_inode_operations;
+ inode->i_fop = &fusex_file_operations;
+ inode->i_data.a_ops = &fusex_file_aops;
+ mapping_set_writeback_may_deadlock_on_reclaim(&inode->i_data);
+ break;
+
+ case S_IFDIR:
+ spin_lock_init(&fi->rdc.lock);
+ inode->i_op = &fusex_dir_inode_operations;
+ inode->i_fop = &fusex_dir_operations;
+ break;
+
+ case S_IFLNK:
+ inode->i_op = &fusex_symlink_inode_operations;
+ inode->i_data.a_ops = &fusex_symlink_aops;
+ inode_nohighmem(inode);
+ break;
+
+ case S_IFCHR:
+ case S_IFBLK:
+ case S_IFIFO:
+ case S_IFSOCK:
+ inode->i_op = &fusex_file_inode_operations;
+ init_special_inode(inode, inode->i_mode, inode->i_rdev);
+ break;
+
+ default:
+ WARN_ON(1);
+ }
+}
+
+static int fusex_xattr_get(const struct xattr_handler *handler, struct dentry *dentry,
+ struct inode *inode, const char *name, void *buffer, size_t size)
+{
+ size_t attr_size;
+ void *value __free(kfree) =
+ fusex_send_lgxattr(inode, name - strlen(handler->prefix), &attr_size);
+
+ if (IS_ERR(value))
+ return PTR_ERR(value);
+
+ if (!size)
+ return attr_size;
+
+ if (size < attr_size)
+ return -ERANGE;
+
+ memcpy(buffer, value, attr_size);
+ return attr_size;
+
+}
+
+static int fusex_xattr_set(const struct xattr_handler *handler, struct mnt_idmap *idmap,
+ struct dentry *dentry, struct inode *inode,
+ const char *name, const void *value, size_t size, int flags)
+{
+ return fusex_send_setxattr(inode, name - strlen(handler->prefix), value, size, flags);
+}
+
+const struct xattr_handler fusex_xattr_user_handler = {
+ .prefix = XATTR_USER_PREFIX,
+ .get = fusex_xattr_get,
+ .set = fusex_xattr_set,
+};
+
+const struct xattr_handler fusex_xattr_handler = {
+ .prefix = "",
+ .get = fusex_xattr_get,
+ .set = fusex_xattr_set,
+};
+
+const struct xattr_handler *const fusex_xattr_handlers[] = {
+ &fusex_xattr_user_handler,
+ &fusex_xattr_handler,
+ NULL,
+};
+
+static int fusex_send_init(struct fuse_mount *fm, struct fusex_id *id,
+ struct fuse_statx_out *statx)
+{
+ FUSE_ARGS(args_in);
+ FUSE_ARGS(args_lr);
+ FUSE_ARGS(args_sx);
+ struct fuse_init_in inarg_in;
+ struct fuse_init_out outarg_in;
+ struct fuse_entryx_out outarg_lr;
+ struct fuse_statx_in inarg_sx;
+ u64 flags = FUSE_INIT_EXT;
+ int err;
+
+ if (fuse_uring_enabled())
+ flags |= FUSE_OVER_IO_URING;
+
+ memset(&inarg_in, 0, sizeof(inarg_in));
+ inarg_in.major = FUSE_KERNEL_VERSION;
+ inarg_in.minor = FUSE_KERNEL_MINOR_VERSION;
+ inarg_in.flags = flags;
+ inarg_in.flags2 = flags >> 32;
+ args_in.opcode = FUSE_INIT;
+ ADD_IN_ARG_S(args_in, &inarg_in);
+ ADD_OUT_ARG_S(args_in, &outarg_in);
+ err = fuse_simple_request(fm, &args_in);
+ if (err)
+ return err;
+
+ flags = outarg_in.flags;
+ if (!(flags & FUSE_INIT_EXT))
+ return -EINVAL;
+
+ flags |= (u64) outarg_in.flags2 << 32;
+
+ fm->fc->minor = outarg_in.minor;
+ fm->fc->max_write = outarg_in.max_write;
+ fm->fc->max_pages = min_t(unsigned int, fm->fc->max_pages_limit,
+ max_t(unsigned int, outarg_in.max_pages, 1));
+
+ args_lr.opcode = FUSE_LOOKUP_ROOT;
+ ADD_OUT_ARG_S(args_lr, &outarg_lr);
+ err = fuse_simple_request(fm, &args_lr);
+ if (err)
+ return err;
+
+ err = fusex_id_from_args(&args_lr, id);
+ if (err)
+ return err;
+
+ fusex_fill_statx(&args_sx, &inarg_sx, statx);
+ args_sx.nodeid = id->nodeid;
+ err = fuse_simple_request(fm, &args_sx);
+ if (err)
+ return err;
+
+ if (flags & FUSE_OVER_IO_URING && fuse_uring_enabled())
+ fuse_chan_io_uring_enable(fm->fc->chan);
+
+ return 0;
+}
+
+static int fusex_fill_super(struct super_block *sb, struct fs_context *fsc)
+{
+ struct fuse_mount *fm = get_fuse_mount_super(sb);
+ struct fuse_conn *fc = fm->fc;
+ struct inode *inode;
+ struct fusex_id id;
+ struct fuse_statx_out statx;
+ struct fuse_chan_param cp;
+ int err;
+
+ /* Dropped in fuse_mount_destroy() */
+ fuse_conn_get(fc);
+ fm->sb = sb;
+ fc->dev = sb->s_dev;
+
+ scoped_guard(mutex, &fuse_mutex) {
+ list_add_tail(&fc->entry, &fuse_conn_list);
+ err = fuse_ctl_add_conn(fc);
+ if (err)
+ return err;
+ }
+
+ err = super_setup_bdi(sb);
+ if (err)
+ return err;
+
+ sb->s_magic = FUSE_SUPER_MAGIC;
+ sb->s_op = &fusex_super_operations;
+ sb->s_xattr = fusex_xattr_handlers;
+ sb->s_maxbytes = MAX_LFS_FILESIZE;
+ sb->s_time_gran = 1;
+ sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
+ if (sb->s_user_ns != &init_user_ns)
+ sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
+ sb->s_blocksize = PAGE_SIZE;
+ sb->s_blocksize_bits = PAGE_SHIFT;
+ sb->s_flags &= SB_RDONLY;
+ sb->s_flags |= SB_POSIXACL;
+
+ err = fusex_send_init(fm, &id, &statx);
+ if (err)
+ return err;
+
+ inode = fusex_iget(sb, &id);
+ if (inode) {
+ WARN_ON(!(inode_state_read_once(inode) & I_NEW));
+ fusex_set_attr(inode, &statx.stat);
+ fusex_init_inode(inode);
+ unlock_new_inode(inode);
+ }
+ sb->s_root = d_make_root(inode);
+ if (!sb->s_root)
+ return -ENOMEM;
+
+ fc->parallel_dirops = true;
+ fc->destroy = true;
+
+ cp.minor = fc->minor;
+ cp.max_write = fc->max_write;
+ cp.max_pages = fc->max_pages;
+ fuse_chan_set_initialized(fc->chan, &cp);
+
+ return 0;
+}
+
+static int fusex_get_tree(struct fs_context *fsc)
+{
+ struct fuse_dev *fud = fsc->fs_private;
+ struct fuse_conn *fc __free(kfree) = kmalloc_obj(*fc);
+ struct fuse_mount *fm __free(kfree) = kzalloc_obj(*fm);
+ struct fuse_chan *fch __free(fuse_chan_free) = fuse_dev_chan_new();
+
+ if (!fch || !fc || !fm)
+ return -ENOMEM;
+
+ fc->release = fuse_free_conn;
+ fsc->s_fs_info = fm;
+ fuse_conn_init(no_free_ptr(fc), no_free_ptr(fm), fsc->user_ns, fch);
+ fuse_chan_set_initialized(fch, NULL);
+ fuse_dev_install(fud, no_free_ptr(fch));
+
+ return get_tree_nodev(fsc, fusex_fill_super);
+}
+
+enum {
+ FUSEX_OPT_FD,
+};
+
+static const struct fs_parameter_spec fusex_fs_parameters[] = {
+ fsparam_fd("fd", FUSEX_OPT_FD),
+ {}
+};
+
+static int fusex_parse_param(struct fs_context *fsc, struct fs_parameter *param)
+{
+ struct fs_parse_result result;
+ struct fuse_dev *fud;
+ int opt;
+
+ if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE)
+ return invalfc(fsc, "No changes allowed in reconfigure");
+
+ opt = fs_parse(fsc, fusex_fs_parameters, param, &result);
+ if (opt < 0)
+ return opt;
+
+ switch (opt) {
+ case FUSEX_OPT_FD:
+ if (param->type != fs_value_is_file)
+ return invalfc(fsc, "FSCONFIG_SET_FD is required for fd");
+ if (param->file->f_op != &fuse_dev_operations)
+ return invalfc(fsc, "fd is not a fuse device");
+ if (param->file->f_cred->user_ns != fsc->user_ns)
+ return invalfc(fsc, "wrong user namespace for fuse device");
+
+ fud = fuse_dev_grab(param->file);
+ if (!fuse_dev_is_sync_init(fud)) {
+ fuse_dev_put(fud);
+ return invalfc(fsc, "synchronous INIT is mandatory");
+ }
+ fsc->fs_private = fud;
+ break;
+
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static void fusex_free_fsc(struct fs_context *fsc)
+{
+ struct fuse_dev *fud = fsc->fs_private;
+
+ if (fud)
+ fuse_dev_put(fud);
+}
+
+static const struct fs_context_operations fusex_context_ops = {
+ .get_tree = fusex_get_tree,
+ .parse_param = fusex_parse_param,
+ .free = fusex_free_fsc,
+};
+
+static int fusex_init_fs_context(struct fs_context *fsc)
+{
+ fsc->ops = &fusex_context_ops;
+ return 0;
+}
+
+static void fusex_kill_sb_anon(struct super_block *sb)
+{
+ struct fuse_mount *fm = get_fuse_mount_super(sb);
+
+ kill_anon_super(sb);
+ fuse_conn_destroy(fm);
+ fuse_mount_remove(fm);
+ fuse_mount_destroy(fm);
+}
+
+static struct file_system_type fusex_fs_type = {
+ .owner = THIS_MODULE,
+ .name = "fusex",
+ .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
+ .init_fs_context = fusex_init_fs_context,
+ .kill_sb = fusex_kill_sb_anon,
+};
+
+int __init fusex_init(void)
+{
+ return register_filesystem(&fusex_fs_type);
+}
+
+void __exit fusex_cleanup(void)
+{
+ unregister_filesystem(&fusex_fs_type);
+}
diff --git a/fs/fuse/fusex.h b/fs/fuse/fusex.h
new file mode 100644
index 000000000000..c7b7c6f8b442
--- /dev/null
+++ b/fs/fuse/fusex.h
@@ -0,0 +1,4 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+int fusex_init(void);
+void fusex_cleanup(void);
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 0897f8e62b4d..2f129f7c168c 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -8,6 +8,7 @@
#include "dev.h"
#include "fuse_i.h"
+#include "fusex.h"
#include <linux/dax.h>
#include <linux/pagemap.h>
@@ -31,7 +32,7 @@ MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
MODULE_DESCRIPTION("Filesystem in Userspace");
MODULE_LICENSE("GPL");
-static struct kmem_cache *fuse_inode_cachep;
+struct kmem_cache *fuse_inode_cachep;
struct list_head fuse_conn_list;
DEFINE_MUTEX(fuse_mutex);
@@ -605,7 +606,7 @@ void fuse_unlock_inode(struct inode *inode, bool locked)
mutex_unlock(&get_fuse_inode(inode)->mutex);
}
-static void fuse_umount_begin(struct super_block *sb)
+void fuse_umount_begin(struct super_block *sb)
{
struct fuse_conn *fc = get_fuse_conn_super(sb);
@@ -631,7 +632,7 @@ static void fuse_send_destroy(struct fuse_mount *fm)
}
}
-static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
+void fuse_convert_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
{
stbuf->f_type = FUSE_SUPER_MAGIC;
stbuf->f_bsize = attr->bsize;
@@ -667,7 +668,7 @@ static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
args.out_args[0].value = &outarg;
err = fuse_simple_request(fm, &args);
if (!err)
- convert_fuse_statfs(buf, &outarg.st);
+ fuse_convert_statfs(buf, &outarg.st);
return err;
}
@@ -2160,6 +2161,10 @@ static int __init fuse_init(void)
if (res)
goto err_sysfs_cleanup;
+ res = fusex_init();
+ if (res)
+ goto err_ctl_cleanup;
+
fuse_dentry_tree_init();
sanitize_global_limit(&max_user_bgreq);
@@ -2167,6 +2172,8 @@ static int __init fuse_init(void)
return 0;
+err_ctl_cleanup:
+ fuse_ctl_cleanup();
err_sysfs_cleanup:
fuse_sysfs_cleanup();
err_dev_cleanup:
@@ -2182,6 +2189,7 @@ static void __exit fuse_exit(void)
pr_debug("exit\n");
fuse_dentry_tree_cleanup();
+ fusex_cleanup();
fuse_ctl_cleanup();
fuse_sysfs_cleanup();
fuse_fs_cleanup();
diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index c13e1f9a2f12..f5d90be821ae 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -663,6 +663,10 @@ enum fuse_opcode {
FUSE_TMPFILE = 51,
FUSE_STATX = 52,
FUSE_COPY_FILE_RANGE_64 = 53,
+ FUSE_LOOKUP_ROOT = 54,
+ FUSE_LOOKUPX = 55,
+ FUSE_MKOBJX = 56,
+ FUSE_SETSTATX = 58,
/* CUSE specific operations */
CUSE_INIT = 4096,
@@ -700,6 +704,20 @@ struct fuse_entry_out {
struct fuse_attr attr;
};
+/*
+ * entryx flags
+ * FUSE_ENTRYX_NEGATIVE: file does not exist, can cache this result
+ */
+#define FUSE_ENTRYX_NEGATIVE (1 << 0)
+
+struct fuse_entryx_out {
+ uint64_t nodeid;
+ uint64_t entry_valid;
+ uint32_t entry_valid_nsec;
+ uint32_t flags;
+ uint64_t spare;
+};
+
struct fuse_forget_in {
uint64_t nlookup;
};
@@ -754,6 +772,17 @@ struct fuse_mknod_in {
uint32_t padding;
};
+enum fuse_mkobjx_flags {
+ FUSE_MKOBJX_TMPFILE = 1 << 0,
+};
+
+struct fuse_mkobjx_in {
+ struct fuse_statx stat;
+ uint32_t namesize;
+ uint32_t flags;
+ uint64_t spare[7];
+};
+
struct fuse_mkdir_in {
uint32_t mode;
uint32_t umask;
@@ -792,6 +821,13 @@ struct fuse_setattr_in {
uint32_t unused5;
};
+struct fuse_setstatx_in {
+ uint64_t fh;
+ uint32_t flags;
+ uint32_t reserved;
+ struct fuse_statx stat;
+};
+
struct fuse_open_in {
uint32_t flags;
uint32_t open_flags; /* FUSE_OPEN_... */
--
2.53.0
^ permalink raw reply related [flat|nested] 17+ messages in thread* Re: [PATCH] fuse: add fusex filesystem
2026-04-29 10:20 [PATCH] fuse: add fusex filesystem Miklos Szeredi
@ 2026-05-07 8:31 ` Horst Birthelmer
2026-05-08 13:01 ` Amir Goldstein
2026-05-12 8:11 ` Miklos Szeredi
2026-05-08 17:29 ` Horst Birthelmer
` (2 subsequent siblings)
3 siblings, 2 replies; 17+ messages in thread
From: Horst Birthelmer @ 2026-05-07 8:31 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: fuse-devel, linux-fsdevel
On Wed, Apr 29, 2026 at 12:20:57PM +0200, Miklos Szeredi wrote:
> This stands for "fuse extended/experimental".
>
> The purpose is to provide a clean base for big features like the FUSE_IOMAP
> api.
>
> It's also a good way to try new stuff like file handles and compound
> requests without the risk of breaking something in the large and complex
> fuse codebase.
>
> Whether these features will be migrated back into the main fuse codebase,
> or fusex is going to end up as a major version update is still up in the
> air.
>
> Major differences from regular fuse:
>
> - local filesystem mode only
> - only synchronous FUSE_INIT is supported
> - only no-open mode
> - new requests:
> + FUSE_LOOKUP_ROOT - return nodeid of root
> + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
> + FUSE_MKOBJX - merged FUSE_MKNOD, MKDIR, SYMLINK and TMPFILE
> + FUSE_SETSTATX - extended version of FUSE_SETATTR
>
> Missing features:
>
> - file handles / export ops
> - compound requests
> - xattr caching
> - fileattr
> - fiemap
> - ioctl
> - copy_file_range
> - lazy dir open
>
> Test server can be found at:
>
> https://github.com/szmi/fuse-utils
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> ---
>
...
Hi Miklos,
I really like the new shiny fusex. It just has one heavy drawback for me.
It is not set to support remote file systems.
I have a coupe of questions that I would like to ask here on the list.
What was the reasoning behind this? Complexity, lack of interest from the community/users of fuse?
Where do you see the biggest challenge for this. To me, it doesn't look impossible to add that,
but I'm sure I'm missing a lot in this context.
Thanks,
Horst
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [PATCH] fuse: add fusex filesystem
2026-05-07 8:31 ` Horst Birthelmer
@ 2026-05-08 13:01 ` Amir Goldstein
2026-05-12 8:17 ` Miklos Szeredi
2026-05-12 8:11 ` Miklos Szeredi
1 sibling, 1 reply; 17+ messages in thread
From: Amir Goldstein @ 2026-05-08 13:01 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: fuse-devel, linux-fsdevel, Horst Birthelmer
On Thu, May 7, 2026 at 10:49 AM Horst Birthelmer <horst@birthelmer.de> wrote:
>
> On Wed, Apr 29, 2026 at 12:20:57PM +0200, Miklos Szeredi wrote:
> > This stands for "fuse extended/experimental".
> >
> > The purpose is to provide a clean base for big features like the FUSE_IOMAP
> > api.
> >
> > It's also a good way to try new stuff like file handles and compound
> > requests without the risk of breaking something in the large and complex
> > fuse codebase.
> >
> > Whether these features will be migrated back into the main fuse codebase,
> > or fusex is going to end up as a major version update is still up in the
> > air.
> >
> > Major differences from regular fuse:
> >
> > - local filesystem mode only
What about things like?
FUSE_HANDLE_KILLPRIV | FUSE_HANDLE_KILLPRIV_V2
FUSE_ATOMIC_O_TRUNC
Does local-only mean that all those (and default_permissions)
are handled by the kernel?
> > - only synchronous FUSE_INIT is supported
> > - only no-open mode
> > - new requests:
> > + FUSE_LOOKUP_ROOT - return nodeid of root
> > + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
> > + FUSE_MKOBJX - merged FUSE_MKNOD, MKDIR, SYMLINK and TMPFILE
> > + FUSE_SETSTATX - extended version of FUSE_SETATTR
> >
> > Missing features:
> >
> > - file handles / export ops
Need to sort out this API that grew odd organically over the years:
FUSE_EXPORT_SUPPORT | FUSE_NO_EXPORT_SUPPORT
The problem is that libfuse declares FUSE_EXPORT_SUPPORT
on behalf of high level filesystems which do not really provide thse
persistent NFS file handle guarantees.
I see no reason for fusex to not support lookup by ".", which is
anyway needed for "reconnect", so a fresh FUSEX_NFS_EXPORT opt-in
for "re-export to NFS" may make more sense.
Thanks,
Amir.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-08 13:01 ` Amir Goldstein
@ 2026-05-12 8:17 ` Miklos Szeredi
2026-05-12 13:08 ` Amir Goldstein
0 siblings, 1 reply; 17+ messages in thread
From: Miklos Szeredi @ 2026-05-12 8:17 UTC (permalink / raw)
To: Amir Goldstein
Cc: Miklos Szeredi, fuse-devel, linux-fsdevel, Horst Birthelmer
On Fri, 8 May 2026 at 15:01, Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Thu, May 7, 2026 at 10:49 AM Horst Birthelmer <horst@birthelmer.de> wrote:
> >
> > On Wed, Apr 29, 2026 at 12:20:57PM +0200, Miklos Szeredi wrote:
> > > Major differences from regular fuse:
> > >
> > > - local filesystem mode only
>
> What about things like?
> FUSE_HANDLE_KILLPRIV | FUSE_HANDLE_KILLPRIV_V2
> FUSE_ATOMIC_O_TRUNC
>
> Does local-only mean that all those (and default_permissions)
> are handled by the kernel?
Yes.
> I see no reason for fusex to not support lookup by ".", which is
> anyway needed for "reconnect", so a fresh FUSEX_NFS_EXPORT opt-in
> for "re-export to NFS" may make more sense.
Why add reconnect? If server supports persistent file handles, and
all req's get the file handle, then no state (nodeid) needs to be
stored by the server.
Thanks,
Miklos
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-12 8:17 ` Miklos Szeredi
@ 2026-05-12 13:08 ` Amir Goldstein
2026-05-12 13:46 ` Bernd Schubert
0 siblings, 1 reply; 17+ messages in thread
From: Amir Goldstein @ 2026-05-12 13:08 UTC (permalink / raw)
To: Miklos Szeredi
Cc: Miklos Szeredi, fuse-devel, linux-fsdevel, Horst Birthelmer
On Tue, May 12, 2026 at 10:18 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> > I see no reason for fusex to not support lookup by ".", which is
> > anyway needed for "reconnect", so a fresh FUSEX_NFS_EXPORT opt-in
> > for "re-export to NFS" may make more sense.
>
> Why add reconnect? If server supports persistent file handles, and
> all req's get the file handle, then no state (nodeid) needs to be
> stored by the server.
>
Yes, we are saying the same thing.
What I am saying is that legacy fuse has FUSE_EXPORT_SUPPORT
which means server supports lookup of "." which libfuse always enabled
and then we added FUSE_NO_EXPORT_SUPPORT to opt-out of nfs export.
I am saying that in FUSEX, lookup "." support is implied.
And that nfs export should always be opt-in (FUSEX_NFS_EXPORT).
Thanks,
Amir.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-12 13:08 ` Amir Goldstein
@ 2026-05-12 13:46 ` Bernd Schubert
0 siblings, 0 replies; 17+ messages in thread
From: Bernd Schubert @ 2026-05-12 13:46 UTC (permalink / raw)
To: Amir Goldstein, Miklos Szeredi
Cc: Miklos Szeredi, fuse-devel, linux-fsdevel, Horst Birthelmer
On 5/12/26 15:08, Amir Goldstein wrote:
> On Tue, May 12, 2026 at 10:18 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
>>
>>> I see no reason for fusex to not support lookup by ".", which is
>>> anyway needed for "reconnect", so a fresh FUSEX_NFS_EXPORT opt-in
>>> for "re-export to NFS" may make more sense.
>>
>> Why add reconnect? If server supports persistent file handles, and
>> all req's get the file handle, then no state (nodeid) needs to be
>> stored by the server.
>>
>
> Yes, we are saying the same thing.
> What I am saying is that legacy fuse has FUSE_EXPORT_SUPPORT
> which means server supports lookup of "." which libfuse always enabled
> and then we added FUSE_NO_EXPORT_SUPPORT to opt-out of nfs export.
>
> I am saying that in FUSEX, lookup "." support is implied.
> And that nfs export should always be opt-in (FUSEX_NFS_EXPORT).
I'm confused here, why do we need lookup of "." without NFS export? And
which file system actually supports lookup of "."?
Shouldn't all that be scratched in favor of LOOKUP_HANDLE?
Thanks,
Bernd
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-07 8:31 ` Horst Birthelmer
2026-05-08 13:01 ` Amir Goldstein
@ 2026-05-12 8:11 ` Miklos Szeredi
2026-05-12 10:29 ` Horst Birthelmer
1 sibling, 1 reply; 17+ messages in thread
From: Miklos Szeredi @ 2026-05-12 8:11 UTC (permalink / raw)
To: Horst Birthelmer; +Cc: Miklos Szeredi, fuse-devel, linux-fsdevel
On Thu, 7 May 2026 at 10:37, Horst Birthelmer <horst@birthelmer.de> wrote:
> Where do you see the biggest challenge for this. To me, it doesn't look impossible to add that,
> but I'm sure I'm missing a lot in this context.
The biggest challenge I see is API design. I'd really like to have a
simple and consistent API for this.
First task: need to document the current interface, which is long
overdue. Here's a start:
https://docs.google.com/document/d/1SInG6nc5dF-db3WtCqFOnRDCY1jRl5JQ0Toiryn2PRc/edit?usp=sharing
This needs refinement, i.e. try to document by cache type, how is the
cache used, invalidated, etc.
Thanks,
Miklos
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: Re: [PATCH] fuse: add fusex filesystem
2026-05-12 8:11 ` Miklos Szeredi
@ 2026-05-12 10:29 ` Horst Birthelmer
0 siblings, 0 replies; 17+ messages in thread
From: Horst Birthelmer @ 2026-05-12 10:29 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: Miklos Szeredi, fuse-devel, linux-fsdevel
On Tue, May 12, 2026 at 10:11:16AM +0200, Miklos Szeredi wrote:
> On Thu, 7 May 2026 at 10:37, Horst Birthelmer <horst@birthelmer.de> wrote:
>
> > Where do you see the biggest challenge for this. To me, it doesn't look impossible to add that,
> > but I'm sure I'm missing a lot in this context.
>
> The biggest challenge I see is API design. I'd really like to have a
> simple and consistent API for this.
OK, that is a valid goal, I guess.
I have tried to come up with a fusex 'module' that does all operations over
io-uring since it is a lot easier to just support one way of transport.
Then I have written a very primitive passthrough to test the user space
interface and I can get it to work with some small changes to the interface
mostly due to the small header space on the ring.
I have done this to check how fast I can have something workable with the
new interface since I'm supposed to add compound commands.
Do you think it is worth pursuing further?
>
> First task: need to document the current interface, which is long
> overdue. Here's a start:
>
> https://docs.google.com/document/d/1SInG6nc5dF-db3WtCqFOnRDCY1jRl5JQ0Toiryn2PRc/edit?usp=sharing
>
> This needs refinement, i.e. try to document by cache type, how is the
> cache used, invalidated, etc.
>
> Thanks,
> Miklos
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-04-29 10:20 [PATCH] fuse: add fusex filesystem Miklos Szeredi
2026-05-07 8:31 ` Horst Birthelmer
@ 2026-05-08 17:29 ` Horst Birthelmer
2026-05-12 8:20 ` Miklos Szeredi
2026-05-11 8:50 ` Horst Birthelmer
2026-05-12 5:05 ` Joanne Koong
3 siblings, 1 reply; 17+ messages in thread
From: Horst Birthelmer @ 2026-05-08 17:29 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: fuse-devel, linux-fsdevel
On Wed, Apr 29, 2026 at 12:20:57PM +0200, Miklos Szeredi wrote:
> This stands for "fuse extended/experimental".
>
> The purpose is to provide a clean base for big features like the FUSE_IOMAP
> api.
>
> It's also a good way to try new stuff like file handles and compound
> requests without the risk of breaking something in the large and complex
> fuse codebase.
>
> Whether these features will be migrated back into the main fuse codebase,
> or fusex is going to end up as a major version update is still up in the
> air.
>
> Major differences from regular fuse:
>
> - local filesystem mode only
> - only synchronous FUSE_INIT is supported
> - only no-open mode
> - new requests:
> + FUSE_LOOKUP_ROOT - return nodeid of root
> + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
> + FUSE_MKOBJX - merged FUSE_MKNOD, MKDIR, SYMLINK and TMPFILE
> + FUSE_SETSTATX - extended version of FUSE_SETATTR
>
> Missing features:
>
> - file handles / export ops
> - compound requests
> - xattr caching
> - fileattr
> - fiemap
> - ioctl
> - copy_file_range
> - lazy dir open
>
> Test server can be found at:
>
> https://github.com/szmi/fuse-utils
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> ---
>
> Patch is against
>
> git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse.git#for-next
>
> fs/fuse/Makefile | 2 +-
> fs/fuse/args.h | 1 +
> fs/fuse/dev.c | 13 +-
> fs/fuse/dir.c | 4 +-
> fs/fuse/fuse_i.h | 8 +
> fs/fuse/fusex.c | 1751 +++++++++++++++++++++++++++++++++++++
> fs/fuse/fusex.h | 4 +
> fs/fuse/inode.c | 16 +-
> include/uapi/linux/fuse.h | 36 +
> 9 files changed, 1826 insertions(+), 9 deletions(-)
> create mode 100644 fs/fuse/fusex.c
> create mode 100644 fs/fuse/fusex.h
>
> diff --git a/fs/fuse/Makefile b/fs/fuse/Makefile
> index 245e67852b03..d9963e411b62 100644
> --- a/fs/fuse/Makefile
> +++ b/fs/fuse/Makefile
> @@ -12,7 +12,7 @@ obj-$(CONFIG_VIRTIO_FS) += virtiofs.o
>
> fuse-y := trace.o # put trace.o first so we see ftrace errors sooner
> fuse-y += dev.o dir.o file.o inode.o control.o xattr.o acl.o readdir.o ioctl.o req_timeout.o req.o
> -fuse-y += poll.o notify.o
> +fuse-y += poll.o notify.o fusex.o
> fuse-y += iomode.o
> fuse-$(CONFIG_FUSE_DAX) += dax.o
> fuse-$(CONFIG_FUSE_PASSTHROUGH) += passthrough.o backing.o
> diff --git a/fs/fuse/args.h b/fs/fuse/args.h
> index ecfe51a192af..1c1e0a25ea07 100644
> --- a/fs/fuse/args.h
> +++ b/fs/fuse/args.h
> @@ -35,6 +35,7 @@ struct fuse_args {
> bool out_pages:1;
> bool user_pages:1;
> bool out_argvar:1;
> + bool out_var_alloc:1;
> bool page_zeroing:1;
> bool page_replace:1;
> bool may_block:1;
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 6fe0d8c263df..134572b1a9af 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -1847,15 +1847,24 @@ int fuse_copy_out_args(struct fuse_copy_state *cs, struct fuse_args *args,
>
> reqsize += fuse_len_args(args->out_numargs, args->out_args);
>
> - if (reqsize < nbytes || (reqsize > nbytes && !args->out_argvar))
> + if (reqsize < nbytes)
> return -EINVAL;
> - else if (reqsize > nbytes) {
> +
> + if (args->out_argvar) {
> struct fuse_arg *lastarg = &args->out_args[args->out_numargs-1];
> unsigned diffsize = reqsize - nbytes;
>
> if (diffsize > lastarg->size)
> return -EINVAL;
> lastarg->size -= diffsize;
> +
> + if (args->out_var_alloc) {
> + lastarg->value = kvmalloc(lastarg->size, GFP_KERNEL);
> + if (!lastarg->value)
> + return -ENOMEM;
> + }
> + } else if (reqsize > nbytes) {
> + return -EINVAL;
> }
> return fuse_copy_args(cs, args->out_numargs, args->out_pages,
> args->out_args, args->page_zeroing);
> diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
> index be41c14ef329..cbe0d4b65d49 100644
> --- a/fs/fuse/dir.c
> +++ b/fs/fuse/dir.c
> @@ -542,7 +542,7 @@ int fuse_valid_type(int m)
> S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
> }
>
> -static bool fuse_valid_size(u64 size)
> +bool fuse_valid_size(u64 size)
> {
> return size <= LLONG_MAX;
> }
> @@ -2485,7 +2485,7 @@ static int fuse_symlink_read_folio(struct file *null, struct folio *folio)
> return err;
> }
>
> -static const struct address_space_operations fuse_symlink_aops = {
> +const struct address_space_operations fuse_symlink_aops = {
> .read_folio = fuse_symlink_read_folio,
> };
>
> diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
> index 3a7ac74a23ed..fe66281b7554 100644
> --- a/fs/fuse/fuse_i.h
> +++ b/fs/fuse/fuse_i.h
> @@ -69,6 +69,9 @@ extern struct mutex fuse_mutex;
> extern unsigned int max_user_bgreq;
> extern unsigned int max_user_congthresh;
>
> +extern struct kmem_cache *fuse_inode_cachep;
> +extern const struct address_space_operations fuse_symlink_aops;
> +
> struct fuse_forget_link;
>
> /**
> @@ -911,6 +914,8 @@ struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
> int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
> struct fuse_entry_out *outarg, struct inode **inode);
>
> +void fuse_umount_begin(struct super_block *sb);
> +
> /*
> * Initialize READ or READDIR request
> */
> @@ -1102,6 +1107,7 @@ void fuse_ctl_remove_conn(struct fuse_conn *fc);
> * Is file type valid?
> */
> int fuse_valid_type(int m);
> +bool fuse_valid_size(u64 size);
>
> bool fuse_invalid_attr(struct fuse_attr *attr);
>
> @@ -1204,6 +1210,8 @@ struct posix_acl *fuse_get_acl(struct mnt_idmap *idmap,
> int fuse_set_acl(struct mnt_idmap *, struct dentry *dentry,
> struct posix_acl *acl, int type);
>
> +void fuse_convert_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr);
> +
> /* readdir.c */
> int fuse_readdir(struct file *file, struct dir_context *ctx);
>
> diff --git a/fs/fuse/fusex.c b/fs/fuse/fusex.c
> new file mode 100644
> index 000000000000..98e239e7e00e
> --- /dev/null
> +++ b/fs/fuse/fusex.c
> @@ -0,0 +1,1751 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +
> +#include "fusex.h"
> +#include "dev.h"
> +#include "fuse_i.h"
> +
> +#include <linux/fs_context.h>
> +#include <linux/miscdevice.h>
> +#include <linux/xxhash.h>
> +#include <linux/pagemap.h>
> +#include <linux/exportfs.h>
> +#include <linux/iversion.h>
> +#include <linux/posix_acl_xattr.h>
> +#include <linux/statfs.h>
> +#include <linux/falloc.h>
> +#include <linux/fs_parser.h>
> +
I think you're missing
#include <uapi/linux/magic.h>
here for FUSE_SUPER_MAGIC
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [PATCH] fuse: add fusex filesystem
2026-04-29 10:20 [PATCH] fuse: add fusex filesystem Miklos Szeredi
2026-05-07 8:31 ` Horst Birthelmer
2026-05-08 17:29 ` Horst Birthelmer
@ 2026-05-11 8:50 ` Horst Birthelmer
2026-05-12 8:34 ` Miklos Szeredi
2026-05-12 5:05 ` Joanne Koong
3 siblings, 1 reply; 17+ messages in thread
From: Horst Birthelmer @ 2026-05-11 8:50 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: fuse-devel, linux-fsdevel
On Wed, Apr 29, 2026 at 12:20:57PM +0200, Miklos Szeredi wrote:
> This stands for "fuse extended/experimental".
>
> The purpose is to provide a clean base for big features like the FUSE_IOMAP
> api.
>
> It's also a good way to try new stuff like file handles and compound
> requests without the risk of breaking something in the large and complex
> fuse codebase.
>
> Whether these features will be migrated back into the main fuse codebase,
> or fusex is going to end up as a major version update is still up in the
> air.
>
> Major differences from regular fuse:
>
> - local filesystem mode only
> - only synchronous FUSE_INIT is supported
> - only no-open mode
> - new requests:
> + FUSE_LOOKUP_ROOT - return nodeid of root
> + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
> + FUSE_MKOBJX - merged FUSE_MKNOD, MKDIR, SYMLINK and TMPFILE
> + FUSE_SETSTATX - extended version of FUSE_SETATTR
>
> Missing features:
>
> - file handles / export ops
> - compound requests
> - xattr caching
> - fileattr
> - fiemap
> - ioctl
> - copy_file_range
> - lazy dir open
>
> Test server can be found at:
>
> https://github.com/szmi/fuse-utils
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> ---
>
Hi Miklos,
I have patched passthrough_hp to use fusex and have noticed, that I had to do
the following to make it work:
diff --git a/fs/fuse/fusex.c b/fs/fuse/fusex.c
index 98e239e7e00e..87fcfc16645a 100644
--- a/fs/fuse/fusex.c
+++ b/fs/fuse/fusex.c
@@ -14,6 +14,7 @@
#include <linux/statfs.h>
#include <linux/falloc.h>
#include <linux/fs_parser.h>
+#include <uapi/linux/magic.h>
static void fusex_init_inode(struct inode *inode);
@@ -1337,6 +1338,7 @@ static int fusex_send_opendir(struct fuse_file *ff, struct inode *inode)
if (!err) {
ff->fh = outarg.fh;
ff->open_flags = FOPEN_CACHE_DIR;
+ ff->nodeid = get_node_id(inode);
}
return err;
}
---
After that I could actually mount and do simple meta operations.
The nodeid is used for listing.
Thanks,
Horst
^ permalink raw reply related [flat|nested] 17+ messages in thread* Re: [PATCH] fuse: add fusex filesystem
2026-05-11 8:50 ` Horst Birthelmer
@ 2026-05-12 8:34 ` Miklos Szeredi
0 siblings, 0 replies; 17+ messages in thread
From: Miklos Szeredi @ 2026-05-12 8:34 UTC (permalink / raw)
To: Horst Birthelmer; +Cc: Miklos Szeredi, fuse-devel, linux-fsdevel
[-- Attachment #1: Type: text/plain, Size: 1385 bytes --]
On Mon, 11 May 2026 at 11:04, Horst Birthelmer <horst@birthelmer.de> wrote:
> @@ -1337,6 +1338,7 @@ static int fusex_send_opendir(struct fuse_file *ff, struct inode *inode)
> if (!err) {
> ff->fh = outarg.fh;
> ff->open_flags = FOPEN_CACHE_DIR;
> + ff->nodeid = get_node_id(inode);
How about this (untested):
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 3bdab8d03373..1b5a262147ef 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -594,7 +594,7 @@ void fuse_read_args_fill(struct fuse_io_args *ia,
struct file *file, loff_t pos,
ia->read.in.size = count;
ia->read.in.flags = file->f_flags;
args->opcode = opcode;
- args->nodeid = ff->nodeid;
+ args->nodeid = get_node_id(file_inode(file));
args->in_numargs = 1;
args->in_args[0].size = sizeof(ia->read.in);
args->in_args[0].value = &ia->read.in;
diff --git a/fs/fuse/fusex.c b/fs/fuse/fusex.c
index c4bc03e57ed7..43c03d7f73ab 100644
--- a/fs/fuse/fusex.c
+++ b/fs/fuse/fusex.c
@@ -1379,6 +1380,7 @@ static int fusex_dir_release(struct inode
*inode, struct file *file)
ra->inarg.fh = ff->fh;
ra->args.opcode = FUSE_RELEASEDIR;
+ ra->args.nodeid = get_node_id(inode);
ra->args.force = true;
ra->args.nocreds = true;
ra->args.end = fusex_release_end;
Thanks,
Miklos
[-- Attachment #2: fusex-readdir-nodeid.patch --]
[-- Type: text/x-patch, Size: 1005 bytes --]
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 3bdab8d03373..1b5a262147ef 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -594,7 +594,7 @@ void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos,
ia->read.in.size = count;
ia->read.in.flags = file->f_flags;
args->opcode = opcode;
- args->nodeid = ff->nodeid;
+ args->nodeid = get_node_id(file_inode(file));
args->in_numargs = 1;
args->in_args[0].size = sizeof(ia->read.in);
args->in_args[0].value = &ia->read.in;
diff --git a/fs/fuse/fusex.c b/fs/fuse/fusex.c
index c4bc03e57ed7..43c03d7f73ab 100644
--- a/fs/fuse/fusex.c
+++ b/fs/fuse/fusex.c
@@ -1379,6 +1380,7 @@ static int fusex_dir_release(struct inode *inode, struct file *file)
ra->inarg.fh = ff->fh;
ra->args.opcode = FUSE_RELEASEDIR;
+ ra->args.nodeid = get_node_id(inode);
ra->args.force = true;
ra->args.nocreds = true;
ra->args.end = fusex_release_end;
^ permalink raw reply related [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-04-29 10:20 [PATCH] fuse: add fusex filesystem Miklos Szeredi
` (2 preceding siblings ...)
2026-05-11 8:50 ` Horst Birthelmer
@ 2026-05-12 5:05 ` Joanne Koong
2026-05-12 9:18 ` Miklos Szeredi
3 siblings, 1 reply; 17+ messages in thread
From: Joanne Koong @ 2026-05-12 5:05 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: fuse-devel, linux-fsdevel
On Wed, Apr 29, 2026 at 3:21 AM Miklos Szeredi <mszeredi@redhat.com> wrote:
>
> This stands for "fuse extended/experimental".
>
> The purpose is to provide a clean base for big features like the FUSE_IOMAP
> api.
>
> It's also a good way to try new stuff like file handles and compound
> requests without the risk of breaking something in the large and complex
> fuse codebase.
>
> Whether these features will be migrated back into the main fuse codebase,
> or fusex is going to end up as a major version update is still up in the
> air.
Could you explain a bit what the main motivation behind fusex is? Is
it mainly to clean up interfaces that are now clunky/stale? Or is it
more driven by limitations in the current codebase that make newer
features hard to add? From the paragraphs above, it kind of seems like
the latter?
If it's mostly to clean up interfaces that are now clunky/stale, I
wonder if now is the best time given that, as Amir phrased it in [1],
"FUSE is experiencing a renaissance of new features and protocol
enhancements", and I'm not sure if we know yet if these new interfaces
will also have some things we'll have wished in hindsight we had done
differently, especially since some of these new features are still
actively evolving and gaining new capabilities (including io-uring and
passthrough). Would it make sense to let these new features and
protocol enhancements evolve and stabilize first before baking them
into fusex?
There's a good chance I'm missing something here, but if fusex is
mostly to make new features easier to add, I'm not sure I see a lot of
difference between doing it in fuse vs fusex when fusex is expanded to
have feature-parity with legacy fuse. Is there something in particular
that fusex makes easier?
I wonder if libfuse could serve as a bridge for the gap between legacy
fuse and fusex, eg libfuse filling any missing gaps / translating
between legacy fuse protocols and fusex protocols such that any fuse
server written for legacy fuse could run using fusex, which would
remove the need to have to do any "backporting" to legacy fuse for new
features that are added to fusex and could accelerate the deprecation
timeline of legacy fuse.
>
> Major differences from regular fuse:
>
> - local filesystem mode only
> - only synchronous FUSE_INIT is supported
Some other things that might be nice:
* handle_killpriv / handle_killpriv2 - sgid/suid + file capabilities
clearing will always be handled by the server (default implementation
in libfuse)
* in fuse io-uring: headers and payload passed together as one
contiguous buffer during registration for pinning, deprecate passing
separate buffers per SQE
* in passthrough: backing_fd passed at lookup/creation time only (not
at open time), expanded struct fuse_backing_map
> - only no-open mode
> - new requests:
> + FUSE_LOOKUP_ROOT - return nodeid of root
> + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
Could you explain why it's preferred to have lookup disentangled from
attributes, eg why compounding lookup + attribute fetching is
preferred over embedding the statx directly inside the lookup request
without compounding? it was brought up in Luis's lookup handles
patchset [2] but it's still a bit unclear to me why compounding the
two is better. I think every operation/request that returns a struct
fuse_entry_out uses the attributes immediately after through
fuse_iget(..., &outarg->attr, ...) or fuse_change_attributes(...,
&outarg.attr, ...), so the attributes seemed pretty tightly
intercoupled with lookup from what I could see.
Thanks,
Joanne
[1] https://lore.kernel.org/linux-fsdevel/CAOQ4uxjRgQGxWYWbRbjVZy19oA5qKZSd9eANKyD8yCoBNGPNNw@mail.gmail.com/
[2] https://lore.kernel.org/linux-fsdevel/20260225112439.27276-1-luis@igalia.com/T/#t
> + FUSE_MKOBJX - merged FUSE_MKNOD, MKDIR, SYMLINK and TMPFILE
> + FUSE_SETSTATX - extended version of FUSE_SETATTR
>
> Missing features:
>
> - file handles / export ops
> - compound requests
> - xattr caching
> - fileattr
> - fiemap
> - ioctl
> - copy_file_range
> - lazy dir open
>
> Test server can be found at:
>
> https://github.com/szmi/fuse-utils
>
> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
> ---
>
> Patch is against
>
> git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse.git#for-next
>
> fs/fuse/Makefile | 2 +-
> fs/fuse/args.h | 1 +
> fs/fuse/dev.c | 13 +-
> fs/fuse/dir.c | 4 +-
> fs/fuse/fuse_i.h | 8 +
> fs/fuse/fusex.c | 1751 +++++++++++++++++++++++++++++++++++++
> fs/fuse/fusex.h | 4 +
> fs/fuse/inode.c | 16 +-
> include/uapi/linux/fuse.h | 36 +
> 9 files changed, 1826 insertions(+), 9 deletions(-)
> create mode 100644 fs/fuse/fusex.c
> create mode 100644 fs/fuse/fusex.h
>
^ permalink raw reply [flat|nested] 17+ messages in thread* Re: [PATCH] fuse: add fusex filesystem
2026-05-12 5:05 ` Joanne Koong
@ 2026-05-12 9:18 ` Miklos Szeredi
2026-05-12 13:22 ` Amir Goldstein
2026-05-12 17:33 ` Joanne Koong
0 siblings, 2 replies; 17+ messages in thread
From: Miklos Szeredi @ 2026-05-12 9:18 UTC (permalink / raw)
To: Joanne Koong; +Cc: Miklos Szeredi, fuse-devel, linux-fsdevel
On Tue, 12 May 2026 at 07:05, Joanne Koong <joannelkoong@gmail.com> wrote:
> Could you explain a bit what the main motivation behind fusex is? Is
> it mainly to clean up interfaces that are now clunky/stale? Or is it
> more driven by limitations in the current codebase that make newer
> features hard to add? From the paragraphs above, it kind of seems like
> the latter?
I think the answer is: both.
Adding new features also hard because the old interface is convoluted
and inconsistent.
> If it's mostly to clean up interfaces that are now clunky/stale, I
> wonder if now is the best time given that, as Amir phrased it in [1],
> "FUSE is experiencing a renaissance of new features and protocol
> enhancements", and I'm not sure if we know yet if these new interfaces
> will also have some things we'll have wished in hindsight we had done
> differently, especially since some of these new features are still
> actively evolving and gaining new capabilities (including io-uring and
> passthrough). Would it make sense to let these new features and
> protocol enhancements evolve and stabilize first before baking them
> into fusex?
Excellent point.
Currently fusex concerns only the filesystem layer, and is agnostic to
the transport layer (except maybe the SYNC_INIT thing). That doesn't
mean that we shouldn't have the same API cleanup in the transport
layer as well. For example, I'd love to get rid of splice on
/dev/fuse, which I feel is more burden than blessing. And I guess
fuse-uring could also do with an API cleanup.
As to when and how this should be done? I guess that's up to you and
Bernd to decide.
But I feel that we do need to have more high level discussion of APIs,
if we want to avoid repeating past mistakes.
> I wonder if libfuse could serve as a bridge for the gap between legacy
> fuse and fusex, eg libfuse filling any missing gaps / translating
> between legacy fuse protocols and fusex protocols such that any fuse
> server written for legacy fuse could run using fusex, which would
> remove the need to have to do any "backporting" to legacy fuse for new
> features that are added to fusex and could accelerate the deprecation
> timeline of legacy fuse.
Definitely. However libfuse does export a lot of details from the
kernel API (e.g. INIT flags) that would no longer work with fusex and
not really feasible to emulate with libfuse. So some porting would be
inevitable.
> Some other things that might be nice:
> * handle_killpriv / handle_killpriv2 - sgid/suid + file capabilities
> clearing will always be handled by the server (default implementation
> in libfuse)
Which is in total contrast to what fusex now does (all that handled by
the VFS). Need to resolve this sanely somehow.
> * in fuse io-uring: headers and payload passed together as one
> contiguous buffer during registration for pinning, deprecate passing
> separate buffers per SQE
Okay.
> * in passthrough: backing_fd passed at lookup/creation time only (not
> at open time), expanded struct fuse_backing_map
Yes, this needs discussion. Amir may have additional info on this.
>
> > - only no-open mode
> > - new requests:
> > + FUSE_LOOKUP_ROOT - return nodeid of root
> > + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
>
> Could you explain why it's preferred to have lookup disentangled from
> attributes, eg why compounding lookup + attribute fetching is
> preferred over embedding the statx directly inside the lookup request
> without compounding? it was brought up in Luis's lookup handles
> patchset [2] but it's still a bit unclear to me why compounding the
> two is better. I think every operation/request that returns a struct
> fuse_entry_out uses the attributes immediately after through
> fuse_iget(..., &outarg->attr, ...) or fuse_change_attributes(...,
> &outarg.attr, ...),
That's one argument for decoupling: why include the same fields
(attributes, validity) in several ops (LOOKUP_ROOT, LOOKUPX, MKOBJX)
if we already have a separate one to do that? It simplifies the
protocol. If we started with decoupled GETATTR, then we wouldn't be
taking about adding these X versions, because it would be a simple
GETATTR -> STATX conversion.
There's also an interesting side effect of coupling STATX with LOOKUP,
which is inherently racy:
Thread A:
send LOOKUP + STATX for /fuse/dir1/filea
[server performs lookup and statx, sends reply]
Thread B:
send SETATTR for /fuse/dir2/fileb, which is hard linked to /fuse/dir1/filea
evict inode for /fuse/dir2/fileb
Thread A:
looks up inode for /fuse/dir1/filea in cache, doesn't find any
creates inode, fills stale attributes
This issue is currently solved with fc->evict_ctr in regular fuse, but
works fine without any additional checks in fusex because of the
decoupled ops.
Of course, it is probably desirable to compound LOOKUP with STATX
despite this, but it gives the implementation more leeway (e.g. on a
different OS).
To conclude: separating out STATX from LOOKUPX, etc. makes the protocol:
a) simpler
b) more flexible
Thanks,
Miklos
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-12 9:18 ` Miklos Szeredi
@ 2026-05-12 13:22 ` Amir Goldstein
2026-05-12 19:22 ` Joanne Koong
2026-05-12 17:33 ` Joanne Koong
1 sibling, 1 reply; 17+ messages in thread
From: Amir Goldstein @ 2026-05-12 13:22 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: Joanne Koong, Miklos Szeredi, fuse-devel, linux-fsdevel
On Tue, May 12, 2026 at 11:31 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> > * in passthrough: backing_fd passed at lookup/creation time only (not
> > at open time), expanded struct fuse_backing_map
>
> Yes, this needs discussion. Amir may have additional info on this.
>
Having the two different lifecycles (open-to-close, lookup-to-forget)
is by design.
This allows the server to potentially move files from passthrough mode
to caching mode and vice versa, as long as all fds are closed.
Adding inode passthrough ops does not play well with this mode
this is why it requires the lookup-to-forget lifecycle, but I see no reason
kill the open-to-close lifecycle as it seems useful for servers that
mostly care about I/O passthrough.
w.r.t expanded struct fuse_backing_map, at the moment it does not
sound right to me to map a backing id X with more than a single backing
file or device.
I think that the maps of fuse_file to multiple fuse_backing is better
handled by another layer of mapping (sort of a backing_mapper)
which is what famfs and iomap need, but it is just my initial thoughts.
Joanne, if you are a concrete use case for that, let's try to see
how a design to this use case may look like.
Thanks,
Amir.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-12 13:22 ` Amir Goldstein
@ 2026-05-12 19:22 ` Joanne Koong
0 siblings, 0 replies; 17+ messages in thread
From: Joanne Koong @ 2026-05-12 19:22 UTC (permalink / raw)
To: Amir Goldstein; +Cc: Miklos Szeredi, Miklos Szeredi, fuse-devel, linux-fsdevel
On Tue, May 12, 2026 at 6:22 AM Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Tue, May 12, 2026 at 11:31 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
> >
> > > * in passthrough: backing_fd passed at lookup/creation time only (not
> > > at open time), expanded struct fuse_backing_map
> >
> > Yes, this needs discussion. Amir may have additional info on this.
> >
>
> Having the two different lifecycles (open-to-close, lookup-to-forget)
> is by design.
>
> This allows the server to potentially move files from passthrough mode
> to caching mode and vice versa, as long as all fds are closed.
I'm unsure how useful mode switching is between caching and
passthrough because the server has no control over when the client
application closes all its fds (eg it can only react but not trigger
it). For my internal use case of switching on passthrough, I was
envisioning the server handling everything through direct io until
passthrough is possible (eg the remote file has been fully downloaded
over the network) so that passthrough can be switched on even with the
application having open fds.
>
> Adding inode passthrough ops does not play well with this mode
> this is why it requires the lookup-to-forget lifecycle, but I see no reason
> kill the open-to-close lifecycle as it seems useful for servers that
> mostly care about I/O passthrough.
>
> w.r.t expanded struct fuse_backing_map, at the moment it does not
> sound right to me to map a backing id X with more than a single backing
> file or device.
My thinking with the struct fuse_backing_map expansion was that future
ioctls for managing backing files (eg attaching a backing file to a
live inode or swapping one backing file for another) might benefit
from a shared struct instead of needing to add new structs for each
type.
>
> I think that the maps of fuse_file to multiple fuse_backing is better
> handled by another layer of mapping (sort of a backing_mapper)
> which is what famfs and iomap need, but it is just my initial thoughts.
I like this idea a lot, it makes sense to me.
>
> Joanne, if you are a concrete use case for that, let's try to see
> how a design to this use case may look like.
The internal team I'm working with is planning to start working on
their server in H2. They will run some benchmarks to see if current
passthrough and/or io-uring+zero-copy might get them the performance
gains they need, or whether they will still need the set of advanced
passthrough features they requested. I'll keep you posted once there's
performance data and if there is still a need for the more advanced
passthrough features (file mapping to multiple backing files, backing
files dynamically changing, enabling passthrough on a live inode),
I'll draft up a design doc to get your thoughts on. In the meantime,
I'm planning to send out v2 of the inode metadata passthrough patchset
sometime this week or next with your feedback from v1.
Thanks,
Joanne
>
> Thanks,
> Amir.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH] fuse: add fusex filesystem
2026-05-12 9:18 ` Miklos Szeredi
2026-05-12 13:22 ` Amir Goldstein
@ 2026-05-12 17:33 ` Joanne Koong
1 sibling, 0 replies; 17+ messages in thread
From: Joanne Koong @ 2026-05-12 17:33 UTC (permalink / raw)
To: Miklos Szeredi; +Cc: Miklos Szeredi, fuse-devel, linux-fsdevel
On Tue, May 12, 2026 at 2:18 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Tue, 12 May 2026 at 07:05, Joanne Koong <joannelkoong@gmail.com> wrote:
>
> > If it's mostly to clean up interfaces that are now clunky/stale, I
> > wonder if now is the best time given that, as Amir phrased it in [1],
> > "FUSE is experiencing a renaissance of new features and protocol
> > enhancements", and I'm not sure if we know yet if these new interfaces
> > will also have some things we'll have wished in hindsight we had done
> > differently, especially since some of these new features are still
> > actively evolving and gaining new capabilities (including io-uring and
> > passthrough). Would it make sense to let these new features and
> > protocol enhancements evolve and stabilize first before baking them
> > into fusex?
>
> Excellent point.
>
> Currently fusex concerns only the filesystem layer, and is agnostic to
> the transport layer (except maybe the SYNC_INIT thing). That doesn't
As soon as fusex lands, doesn't this mean with the linux backwards
compatibility policy that any user-facing transport layer decisions
are also locked in? Or is the uapi for it still able to change later
because fusex is marked as "experimental"?
> mean that we shouldn't have the same API cleanup in the transport
> layer as well. For example, I'd love to get rid of splice on
> /dev/fuse, which I feel is more burden than blessing. And I guess
> fuse-uring could also do with an API cleanup.
>
> As to when and how this should be done? I guess that's up to you and
> Bernd to decide.
>
> But I feel that we do need to have more high level discussion of APIs,
> if we want to avoid repeating past mistakes.
>
> >
> > > - only no-open mode
> > > - new requests:
> > > + FUSE_LOOKUP_ROOT - return nodeid of root
> > > + FUSE_LOOKUPX - FUSE_LOOKUP without the getattr
> >
> > Could you explain why it's preferred to have lookup disentangled from
> > attributes, eg why compounding lookup + attribute fetching is
> > preferred over embedding the statx directly inside the lookup request
> > without compounding? it was brought up in Luis's lookup handles
> > patchset [2] but it's still a bit unclear to me why compounding the
> > two is better. I think every operation/request that returns a struct
> > fuse_entry_out uses the attributes immediately after through
> > fuse_iget(..., &outarg->attr, ...) or fuse_change_attributes(...,
> > &outarg.attr, ...),
>
> That's one argument for decoupling: why include the same fields
> (attributes, validity) in several ops (LOOKUP_ROOT, LOOKUPX, MKOBJX)
> if we already have a separate one to do that? It simplifies the
> protocol. If we started with decoupled GETATTR, then we wouldn't be
> taking about adding these X versions, because it would be a simple
> GETATTR -> STATX conversion.
>
> There's also an interesting side effect of coupling STATX with LOOKUP,
> which is inherently racy:
>
> Thread A:
> send LOOKUP + STATX for /fuse/dir1/filea
> [server performs lookup and statx, sends reply]
>
> Thread B:
> send SETATTR for /fuse/dir2/fileb, which is hard linked to /fuse/dir1/filea
> evict inode for /fuse/dir2/fileb
>
> Thread A:
> looks up inode for /fuse/dir1/filea in cache, doesn't find any
> creates inode, fills stale attributes
>
> This issue is currently solved with fc->evict_ctr in regular fuse, but
> works fine without any additional checks in fusex because of the
> decoupled ops.
>
> Of course, it is probably desirable to compound LOOKUP with STATX
> despite this, but it gives the implementation more leeway (e.g. on a
> different OS).
>
> To conclude: separating out STATX from LOOKUPX, etc. makes the protocol:
>
> a) simpler
> b) more flexible
This makes sense to me, thanks for the explanation.
Thanks,
Joanne
^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2026-05-12 19:22 UTC | newest]
Thread overview: 17+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-04-29 10:20 [PATCH] fuse: add fusex filesystem Miklos Szeredi
2026-05-07 8:31 ` Horst Birthelmer
2026-05-08 13:01 ` Amir Goldstein
2026-05-12 8:17 ` Miklos Szeredi
2026-05-12 13:08 ` Amir Goldstein
2026-05-12 13:46 ` Bernd Schubert
2026-05-12 8:11 ` Miklos Szeredi
2026-05-12 10:29 ` Horst Birthelmer
2026-05-08 17:29 ` Horst Birthelmer
2026-05-12 8:20 ` Miklos Szeredi
2026-05-11 8:50 ` Horst Birthelmer
2026-05-12 8:34 ` Miklos Szeredi
2026-05-12 5:05 ` Joanne Koong
2026-05-12 9:18 ` Miklos Szeredi
2026-05-12 13:22 ` Amir Goldstein
2026-05-12 19:22 ` Joanne Koong
2026-05-12 17:33 ` Joanne Koong
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox