mount

const (
	DefaultFileModeRW = os.FileMode(0o644)
	DefaultDirModeRW  = os.ModeDir | 0o755
)

Writable mounts (/ipns, /mfs): standard POSIX defaults matching umask 022.

const (
	DefaultFileModeRO = os.FileMode(0o444)
	DefaultDirModeRO  = os.ModeDir | 0o555
)

Read-only mount (/ipfs): no write bits.

const AutomaticIno = 1 << 63

AutomaticIno is where go-fuse starts numbering nodes a filesystem did not number itself. Mounts keep their own numbers below it, so a node that slips through unnumbered can never be mistaken for one that was numbered. NewMount pins fs.Options.FirstAutomaticIno to it rather than relying on go-fuse's default, which is the same value but free to change.

const DefaultBlksize = 1024 * 1024

DefaultBlksize is the preferred I/O size (stat.st_blksize) FUSE mounts advertise when no chunker-derived value applies (readonly /ipfs, or writable /mfs with a rabin/buzhash chunker). Larger hints let tools like cp, dd, and rsync use bigger buffers, amortizing FUSE syscall and DAG-walk overhead. 1 MiB matches the chunk size of Kubo's cross-implementation CID-deterministic import profile (IPIP-499). Hardcoded instead of tracking boxo's chunker default so the stat(2) contract stays stable across Kubo and boxo upgrades.

const FirstIno = 2

FirstIno is the lowest number a mount hands out to an entry, leaving 0 (not a valid inode number) and 1 (the mount point) alone.

const MaxReadAhead = 64 * 1024 * 1024

MaxReadAhead tells the kernel how far ahead to read in a single FUSE request. 64 MiB works well for sequential access (streaming, file copies) because most data is served from the local blockstore after the initial fetch. Network-backed reads are already chunked by the DAG layer, so oversized readahead does not cause extra round-trips.

const NamespaceRootMode = os.ModeDir | 0o111

NamespaceRootMode is for the /ipfs/ and /ipns/ root directories. Execute-only: these are virtual namespaces where users traverse by name (CID or IPNS key) but listing the full namespace is not possible.

const Nlink = 1

Nlink is the link count (stat.st_nlink) reported for every entry on a FUSE mount. Left unset it is 0, which POSIX gives to an inode with no remaining names, so tools read a live file as one that is on its way out.

1 is right for files and symlinks: neither IPFS nor MFS has hard links. Directories conventionally count their own entry, their parent's, and one per subdirectory, but reporting 1 for them as well is deliberate. GNU find skips the stat of a directory's children once it has seen st_nlink-2 of them, and would miss entries if the count were a guess; a count below 2 tells it to walk the directory properly instead.

const RootIno = 1

RootIno is the inode number reported for a mount point. go-fuse leaves the root's inode number at 0 unless it is told otherwise, and 0 is not a valid inode number: tools that check it read the mount point as a file that is not there. 1 is the conventional root number and is already the FUSE node ID the kernel uses for the root.

const StatBlockSize = 512

StatBlockSize is the POSIX stat(2) block unit. The st_blocks field reports allocation in 512-byte units regardless of the filesystem's real block size (see `man 2 stat`). Tools like `du`, `ls -s`, and `find -size` multiply st_blocks by this constant to compute bytes.

const SymlinkMode = os.FileMode(0o777)

SymlinkMode is the POSIX permission bits for symlinks. Symlink permissions are always 0777; access control uses the target's mode.

const WritableMountCapabilities = fuse.CAP_ATOMIC_O_TRUNC

WritableMountCapabilities are FUSE capabilities requested for writable mounts (/ipns, /mfs).

CAP_ATOMIC_O_TRUNC tells the kernel to pass O_TRUNC to Open instead of sending a separate SETATTR(size=0) before Open. Without this, the kernel does SETATTR first, which requires opening a write descriptor inside Setattr. MFS only allows one write descriptor at a time, so that deadlocks. With this capability, O_TRUNC is handled inside Open where we already hold the descriptor.

const XattrCID = "ipfs.cid"

XattrCID is the extended attribute name for the node's CID. Follows the convention used by CephFS (ceph.*), Btrfs (btrfs.*), and GlusterFS (glusterfs.*) of using a project-specific namespace.

const XattrCIDDeprecated = "ipfs_cid"

XattrCIDDeprecated is the old xattr name. Getxattr normalizes it to XattrCID and logs a deprecation error so existing tooling keeps working while users migrate. TODO: remove after 2 releases.