STRUCT FUSE_CONN(9)

💅 struct fuse_conn(9) — A Fuse connection

okay bestie, so this is a page from the Kernel API Manual, and it's literally all about struct fuse_conn — the little (huge) struct that is a FUSE connection. no cap, this thing is the main character of the whole FUSE mount situation, so let's get into it 🎀

✨ SYNOPSIS ✨

like, here's the whole vibe, every single field, verbatim, don't touch anything:

struct fuse_conn {
    spinlock_t lock;
    refcount_t count;
    atomic_t epoch;
    struct work_struct epoch_work;
    struct rcu_head rcu;
    kuid_t user_id;
    kgid_t group_id;
    struct pid_namespace *pid_ns;
    struct user_namespace *user_ns;
    unsigned max_read;
    unsigned max_write;
    unsigned int max_pages;
    unsigned int max_pages_limit;
    struct fuse_chan *chan;
    atomic64_t khctr;
    struct rb_root polled_files;
    unsigned congestion_threshold;
    unsigned conn_error:1;
    unsigned conn_init:1;
    unsigned async_read:1;
    unsigned abort_err:1;
    unsigned atomic_o_trunc:1;
    unsigned export_support:1;
    unsigned writeback_cache:1;
    unsigned parallel_dirops:1;
    unsigned handle_killpriv:1;
    unsigned cache_symlinks:1;
    unsigned int legacy_opts_show:1;
    unsigned handle_killpriv_v2:1;
    unsigned no_open:1;
    unsigned no_opendir:1;
    unsigned no_fsync:1;
    unsigned no_fsyncdir:1;
    unsigned no_flush:1;
    unsigned no_setxattr:1;
    unsigned setxattr_ext:1;
    unsigned no_getxattr:1;
    unsigned no_listxattr:1;
    unsigned no_removexattr:1;
    unsigned no_lock:1;
    unsigned no_access:1;
    unsigned no_create:1;
    unsigned no_bmap:1;
    unsigned no_poll:1;
    unsigned big_writes:1;
    unsigned dont_mask:1;
    unsigned no_flock:1;
    unsigned no_fallocate:1;
    unsigned no_rename2:1;
    unsigned auto_inval_data:1;
    unsigned explicit_inval_data:1;
    unsigned do_readdirplus:1;
    unsigned readdirplus_auto:1;
    unsigned async_dio:1;
    unsigned no_lseek:1;
    unsigned posix_acl:1;
    unsigned default_permissions:1;
    unsigned allow_other:1;
    unsigned no_copy_file_range:1;
    unsigned no_copy_file_range_64:1;
    unsigned int destroy:1;
    unsigned int delete_stale:1;
    unsigned int no_control:1;
    unsigned int no_force_umount:1;
    unsigned int auto_submounts:1;
    unsigned int sync_fs:1;
    unsigned int init_security:1;
    unsigned int create_supp_group:1;
    unsigned int inode_dax:1;
    unsigned int no_tmpfile:1;
    unsigned int direct_io_allow_mmap:1;
    unsigned int no_statx:1;
    unsigned int passthrough:1;
    unsigned int use_pages_for_kvec_io:1;
    unsigned int no_link:1;
    unsigned int sync_init:1;
    int max_stack_depth;
    unsigned minor;
    struct list_head entry;
    dev_t dev;
    u32 scramble_key[4];
    atomic64_t attr_version;
    atomic64_t evict_ctr;
    u32 name_max;
    void (*release)(struct fuse_conn *);
    struct rw_semaphore killsb;
  #ifdef CONFIG_FUSE_DAX;
    enum fuse_dax_mode dax_mode;
    struct fuse_conn_dax *dax;
  #endif;
    struct list_head mounts;
    struct fuse_sync_bucket *curr_bucket;
  #ifdef CONFIG_FUSE_PASSTHROUGH;
    struct idr backing_files_map;
  #endif;
 };

🦋 Members

🔒 the locking & lifetime girlies

lock — the lock that protects: polled_files, backing_files_map, curr_bucket. she's guarding three things at once, multitasking icon 💅
count — the refcount, bestie
epoch — the current epoch, for up-to-date dentries
epoch_work — used to invalidate dentries from old epochs
rcu — used to delay freeing fuse_conn, making it safe

🫶 the "who even are you" identity block

user_id — the user id for this mount
group_id — the group id for this mount
pid_ns — the pid namespace for this mount
user_ns — the user namespace for this mount

📏 sizes & limits

max_read — maximum read size
max_write — maximum write size
max_pages — maximum number of pages that can be used in a single request
max_pages_limit — constrain ->max_pages to this value during feature negotiation

🚚 transport & handles

chan — transport layer object
khctr — the next unique kernel file handle
polled_files — rbtree of fuse_files waiting for poll events indexed by ph
congestion_threshold — number of background requests at which congestion starts

🩹 connection-status bits

conn_error — connection failed (version mismatch). Cannot race with setting other bitfields since it is only set once in INIT reply, before any other request, and never cleared
conn_init — connection successful. Only set in INIT
async_read — do readahead asynchronously? Only set in INIT
abort_err — return an unique read error after abort. Only set in INIT

🎀 the behaviour flags, in original order (yes, all of them)

atomic_o_trunc — do not send separate SETATTR request before open(O_TRUNC)
export_support — filesystem supports NFS exporting. Only set in INIT
writeback_cache — write-back cache policy (default is write-through)
parallel_dirops — allow parallel lookups and readdir (default is serialized)
handle_killpriv — handle fs handles killing suid/sgid/cap on write/chown/trunc
cache_symlinks — cache READLINK responses in page cache
legacy_opts_show — show legacy mount options
handle_killpriv_v2 — fs kills suid/sgid/cap on write/chown/trunc. suid is killed on write/trunc only if caller did not have CAP_FSETID. sgid is killed on write/truncate only if caller did not have CAP_FSETID as well as file has group execute permission.
no_open — is open/release not implemented by fs?
no_opendir — is opendir/releasedir not implemented by fs?
no_fsync — is fsync not implemented by fs?
no_fsyncdir — is fsyncdir not implemented by fs?
no_flush — is flush not implemented by fs?
no_setxattr — is setxattr not implemented by fs?
setxattr_ext — does file server support extended setxattr
no_getxattr — is getxattr not implemented by fs?
no_listxattr — is listxattr not implemented by fs?
no_removexattr — is removexattr not implemented by fs?
no_lock — are posix file locking primitives not implemented by fs?
no_access — is access not implemented by fs?
no_create — is create not implemented by fs?
no_bmap — is bmap not implemented by fs?
no_poll — is poll not implemented by fs?
big_writes — do multi-page cached writes
dont_mask — don't apply umask to creation modes
no_flock — are BSD file locking primitives not implemented by fs?
no_fallocate — is fallocate not implemented by fs?
no_rename2 — is rename with flags implemented by fs?
auto_inval_data — use enhanced/automatic page cache invalidation.
explicit_inval_data — filesystem is fully responsible for page cache invalidation.
do_readdirplus — does the filesystem support readdirplus?
readdirplus_auto — does the filesystem want adaptive readdirplus?
async_dio — does the filesystem support asynchronous direct-IO submission?
no_lseek — is lseek not implemented by fs?
posix_acl — does the filesystem support posix acls?
default_permissions — check permissions based on the file mode or not?
allow_other — allow other than the mounter user to access the filesystem ?
no_copy_file_range — does the filesystem support copy_file_range?
no_copy_file_range_64 — does the filesystem support copy_file_range_64?
destroy — send DESTROY request
delete_stale — delete dentries that have gone stale
no_control — do not create entry in fusectl fs
no_force_umount — do not allow MNT_FORCE umount
auto_submounts — auto-mount submounts announced by the server
sync_fs — propagate syncfs to server
init_security — initialize security xattrs when creating a new inode
create_supp_group — add supplementary group info when creating a new inode
inode_dax — does the filesystem support per inode DAX?
no_tmpfile — is tmpfile not implemented by fs?
direct_io_allow_mmap — relax restrictions to allow shared mmap in FOPEN_DIRECT_IO mode
no_statx — is statx not implemented by fs?
passthrough — passthrough support for read/write IO
use_pages_for_kvec_io — use pages instead of pointer for kernel I/O
no_link — is link not implemented by fs?
sync_init — is synchronous FUSE_INIT allowed?

🌸 more numbers & bookkeeping

max_stack_depth — maximum stack depth for passthrough backing files
minor — negotiated minor version
entry — entry on the fuse_conn_list
dev — device ID from the root super block
scramble_key — key for lock owner ID scrambling
attr_version — version counter for attribute changes
evict_ctr — version counter for evict inode
name_max — maximum file name length
release — called on final put
killsb — read/write semaphore to hold when accessing the sb of any fuse_mount belonging to this connection

🧚 the CONFIG-gated ones

dax_mode — Dax mode
dax — Dax specific conn data, non-NULL if DAX is enabled
mounts — list of filesystems using this connection
curr_bucket — new writepages go into this bucket
backing_files_map — IDR for backing files ids

👑 Description

so this structure is created when the root filesystem is mounted, and it's destroyed when the client device is closed and the last fuse_mount is destroyed. literally birth and death of a connection, it's giving full life cycle 🦋✨

🌈 SEE ALSO

kernel file ./fs/fuse/fuse_i.h — plus these besties: struct fuse_submount_lookup(9), struct fuse_inode(9), struct fuse_file(9), fuse_fill_super_common(9), fuse_mount_remove(9), fuse_ctl_add_conn(9), fuse_ctl_remove_conn(9)

💀 real talk though: those #ifdef CONFIG_FUSE_DAX and #ifdef CONFIG_FUSE_PASSTHROUGH blocks mean some of these fields only exist in certain kernel configs. not every build gets the whole squad, iykyk 🙈

okay that's the whole page, go off bestie 💖 this one came straight out of ./fs/fuse and it's dated October 2026 — save the date, screenshot it, put it on the fridge 🎀✨🫶

📦 where this page comes from

nixpkgs unstable: linux-manual 7.3.1
💖 girlified by deepseek-ai/DeepSeek-V4.1-Flash-1017 on 2026-11-04: a language model's rewrite of the original, under its terms