kubo/fuse · ku/kubo

GGitHubfix(fuse): keep what you write after a rename (#11430)

文件	最后提交记录	最后更新时间
fusetest	fix(fuse): keep what you write after a rename (#11430) * fix(fuse): keep a rename's writes go-fuse hands the kernel's existing node to the new name once Dir.Rename returns, and that node still held the MFS handle the rename had unlinked. MFS treats such a handle as gone: a write through it was accepted and then dropped, so `mv a b` followed by a write to `b` read back the old contents a second later, once the entry cache expired. A directory was worse. Creating a file in one that had just been renamed flushed through the dead handle, which carried the name the rename had moved away from, so the new file was lost and the old directory came back for good. Each node now reaches its MFS handle through an atomic, and a rename points the moved node at the entry that exists afterwards. Entries the kernel had already looked up underneath a renamed directory hang off the handle it was reached through, so the walk follows them down; it covers what the kernel is holding, not the whole tree. Invalidating the entry instead was tried and does not work: the kernel processes FUSE_NOTIFY_INVAL_ENTRY while holding the parent inode lock, so notifying from inside Rename deadlocks, and notifying asynchronously still loses most of the writes it races. Left unfixed: a write through a file descriptor held open across the rename still goes to the descriptor opened from the old handle. * fix(fuse): refuse to replace a non-empty directory A rename may only overwrite a directory that is empty. MFS removes a directory and everything under it without complaint, so `mv -T src dst` took dst's contents with it and reported success. Rmdir already had the check; Rename now makes it too, before it unlinks anything. The check for an absent destination also goes through errors.Is now. It compares against a sentinel that boxo returns bare today, and the cost of that changing is the source file, which by then has been unlinked. * fix(fuse): read /ipfs blocks we cannot decode A UnixFS directory can link to a block of any codec. stat reports one the mount cannot decode as a file the size of the block, but every read of it failed, because the read path went looking for a UnixFS DAG that is not there. A size stat promises has to be a size reads deliver, so serve the block itself. * fix(fuse): list a directory with a missing block One child whose block is not held locally failed the whole listing, and with an errno the caller could make nothing of: ipld.ErrNotFound has no mapping, so it arrived as ENOSYS. The readable entries are worth having, so report the one that is missing with no type and let a stat of it say what is wrong. * fix(fuse): report the CID the path used The ipfs.cid xattr answered with a CID the caller had never seen. A lookup rebuilds the node by decoding the block, which drops the version and codec of the path it came from, so a v1 dag-pb path reported its v0 form. Keep the CID the entry resolved to and report that. The changelog entry also covers the rename check from the commit before it, which landed without one. * test(fuse): make the rename tests catch their bugs TestRenameOntoNamespaceRoot read the file back through the mount, which answers from the entry the kernel still has cached and so succeeds whether or not the rename took the file away. It passed against the bug it was written for. Ask MFS instead. The dirent helper also loops on a record length it never checks, which would spin rather than fail if the kernel ever sent zero. * test(ipns): settle the repo path before mounting TestStatfs assigned Root.RepoPath once the server was already serving, and Statfs reads it from a FUSE handler goroutine, so `go test -race ./fuse/...` reported a data race on every run. The mfs and readonly tests already settle it before their mount; do the same here.	16 天前
ipns	fix(fuse): keep what you write after a rename (#11430) * fix(fuse): keep a rename's writes go-fuse hands the kernel's existing node to the new name once Dir.Rename returns, and that node still held the MFS handle the rename had unlinked. MFS treats such a handle as gone: a write through it was accepted and then dropped, so `mv a b` followed by a write to `b` read back the old contents a second later, once the entry cache expired. A directory was worse. Creating a file in one that had just been renamed flushed through the dead handle, which carried the name the rename had moved away from, so the new file was lost and the old directory came back for good. Each node now reaches its MFS handle through an atomic, and a rename points the moved node at the entry that exists afterwards. Entries the kernel had already looked up underneath a renamed directory hang off the handle it was reached through, so the walk follows them down; it covers what the kernel is holding, not the whole tree. Invalidating the entry instead was tried and does not work: the kernel processes FUSE_NOTIFY_INVAL_ENTRY while holding the parent inode lock, so notifying from inside Rename deadlocks, and notifying asynchronously still loses most of the writes it races. Left unfixed: a write through a file descriptor held open across the rename still goes to the descriptor opened from the old handle. * fix(fuse): refuse to replace a non-empty directory A rename may only overwrite a directory that is empty. MFS removes a directory and everything under it without complaint, so `mv -T src dst` took dst's contents with it and reported success. Rmdir already had the check; Rename now makes it too, before it unlinks anything. The check for an absent destination also goes through errors.Is now. It compares against a sentinel that boxo returns bare today, and the cost of that changing is the source file, which by then has been unlinked. * fix(fuse): read /ipfs blocks we cannot decode A UnixFS directory can link to a block of any codec. stat reports one the mount cannot decode as a file the size of the block, but every read of it failed, because the read path went looking for a UnixFS DAG that is not there. A size stat promises has to be a size reads deliver, so serve the block itself. * fix(fuse): list a directory with a missing block One child whose block is not held locally failed the whole listing, and with an errno the caller could make nothing of: ipld.ErrNotFound has no mapping, so it arrived as ENOSYS. The readable entries are worth having, so report the one that is missing with no type and let a stat of it say what is wrong. * fix(fuse): report the CID the path used The ipfs.cid xattr answered with a CID the caller had never seen. A lookup rebuilds the node by decoding the block, which drops the version and codec of the path it came from, so a v1 dag-pb path reported its v0 form. Keep the CID the entry resolved to and report that. The changelog entry also covers the rename check from the commit before it, which landed without one. * test(fuse): make the rename tests catch their bugs TestRenameOntoNamespaceRoot read the file back through the mount, which answers from the entry the kernel still has cached and so succeeds whether or not the rename took the file away. It passed against the bug it was written for. Ask MFS instead. The dirent helper also loops on a record length it never checks, which would spin rather than fail if the kernel ever sent zero. * test(ipns): settle the repo path before mounting TestStatfs assigned Root.RepoPath once the server was already serving, and Statfs reads it from a FUSE handler goroutine, so `go test -race ./fuse/...` reported a data race on every run. The mfs and readonly tests already settle it before their mount; do the same here.	16 天前
mfs	fix(mfs): stop repo gc from freezing files ops (#11386) * fix(mfs): stop repo gc from freezing files ops Running `ipfs repo gc` alongside `ipfs files` writes could leave MFS permanently hung: GC deleted directory-node blocks a write had written to the blockstore but not yet linked into the persisted MFS root, and the next path lookup blocked forever fetching the missing block while holding the MFS directory lock, so every later files command piled up behind it. MFS mutations now take the pin lock, the same lock `ipfs add` uses, and GC computes the MFS root only after it holds the GC lock. A write's blocks are therefore either fully linked into the root before GC runs, or the write waits for GC to finish, so GC never collects data a live write still needs. - core/commands/files.go: hold the pin lock across write, cp, mkdir, mv, rm, flush, chcid, chmod, and touch - gc/gc.go: take the best-effort MFS root snapshot after acquiring the GC lock, closing the snapshot-before-lock window - core/corerepo/gc.go: pass the root as a callback evaluated under the lock - core/node/core.go: document why MFS keeps its online DAG service so lazy `ipfs files cp /ipfs/<cid>` pointers still resolve - gc/gc_test.go: assert the root snapshot is taken under the GC lock Closes #10842 * fix(mfs): extend gc pin lock to add and fuse The pin lock that stops garbage collection from collecting live MFS blocks now covers every path that mutates MFS, not just `ipfs files`, and every live MFS root is part of the GC live set. - core/commands/add.go: hold the pin lock while `ipfs add --to-files` links the added content into the MFS root - fuse/writable, fuse/mfs: hold the pin lock across FUSE `/mfs` structural writes and file flush, fsync, and release - fuse/ipns: same for the per-key `/ipns` mounts, and register their roots so GC keeps their blocks live - core/core.go: track mounted MFS roots for the GC live set - core/corerepo/gc.go: build the live set from the files root plus every registered root, skipping a root that errors instead of aborting the whole GC Closes #6113 Closes #7008 Closes #9553 * fix(mfs): fail fast on a missing block A directory-node block that is missing locally and unreachable would block an MFS operation forever while it held the directory lock, wedging the whole MFS and a clean shutdown until the process was killed. This is what remained after a repo was damaged by an older Kubo, a manual `ipfs block rm`, or a crash (the lockup in #7844). MFS now bounds those network reads: an unreachable block fails the operation with a timeout and releases the lock, while lazily-referenced content (`ipfs files cp /ipfs/<cid>`) still loads as before. - config: add DefaultMFSFetchTimeout and pass it to the MFS root via mfs.WithFetchTimeout in Import.MFSRootOptions - go.mod: bump boxo to pull in mfs.WithFetchTimeout * feat(mfs): honor --timeout in files read and write Adopt boxo's mfs.File.Open(ctx) so MFS file operations carry a context and a read or write stuck on a missing block can be cancelled instead of blocking forever. - go.mod: bump boxo to pick up mfs.File.Open(ctx) - core/commands/files.go: pass the request context to files read and write, so a client --timeout ends a stuck read or write - fuse/writable: bind long-lived FUSE descriptors (Create, Open) to the mount context via the new Config.MountCtx; the transient Setattr truncate uses the per-operation context - fuse/mfs, fuse/ipns: set MountCtx to the node/mount context - test/cli: cover files read and write honoring --timeout on unreachable content * chore(deps): bump boxo to merged mfs fix Switch from the pre-merge branch pseudo-version to the merged commit of ipfs/boxo#1185, the boxo side of the MFS under-lock work this branch depends on. * refactor(fuse): pass request ctx to pin lock Flush, Release, and Fsync passed context.Background() to the pin lock while their own ctx argument was in scope. Thread the passed ctx through instead, matching the other handlers. boxo's default GCLocker discards the context, so this is a consistency change, not a behavior change. https://github.com/ipfs/kubo/pull/11386#discussion_r3534131824 https://github.com/ipfs/kubo/pull/11386#discussion_r3534138912 https://github.com/ipfs/kubo/pull/11386#discussion_r3534140861 Co-authored-by: Andrew Gillis <11790789+gammazero@users.noreply.github.com> * refactor(gc): const for buffer, simplify roots Name the GC result channel buffer size as a const, drop the temporary in the best-effort root snapshot, and use t.Context() in the snapshot test. The buffer size predates this branch; it only shifted in the diff. https://github.com/ipfs/kubo/pull/11386#discussion_r3534154606 https://github.com/ipfs/kubo/pull/11386#discussion_r3534077507 https://github.com/ipfs/kubo/pull/11386#discussion_r3534085418 https://github.com/ipfs/kubo/pull/11386#discussion_r3534106766 Co-authored-by: Andrew Gillis <11790789+gammazero@users.noreply.github.com> * refactor(add): collapse pin lock defer Fold the pin lock's unlock into the defer at both --to-files call sites, matching the one-line style used elsewhere. https://github.com/ipfs/kubo/pull/11386#discussion_r3534006489 https://github.com/ipfs/kubo/pull/11386#discussion_r3534013057 Co-authored-by: Andrew Gillis <11790789+gammazero@users.noreply.github.com> --------- Co-authored-by: Andrew Gillis <11790789+gammazero@users.noreply.github.com>	1 个月前
mount	fix(fuse): give mounts stable inode numbers (#11429) * fix(fuse): report a link count of 1 st_nlink was left at 0, which POSIX gives an inode with no remaining names, so tools can read a live file as one on its way out. Neither IPFS nor MFS has hard links. Directories report 1 as well, which keeps GNU find from trusting a subdirectory count and skipping entries. * fix(ipns): fill attrs in key directory lookups The /ipns root answers lookups for its key directories and alias symlinks itself, and the reply carried zeroed attributes. Every later lookup refreshed the kernel's cache with the same zeroes, so the Getattr that would have corrected them never ran and a key directory showed up as d--------- with no link count. * fix(fuse): give mounts stable inode numbers go-fuse numbers any node left with a zero StableAttr.Ino itself, and picks a new number every time. The kernel drops a mount's dentries once EntryTimeout expires, so a file nobody touched came back from the next lookup under a different st_ino, and programs that compare file identity over time read that as the file being replaced. vim abandons a save with "E949: File changed while writing", which is what made the FUSE CI job flaky once it moved to slower runners and the save started crossing the one second cache boundary. - /ipfs takes the number from the multihash digest inside the CID, so the same content is one object whichever path reaches it; inline CIDs hash the whole CID instead, their digest being the content itself - /ipns and /mfs allocate per mount from a counter keyed by parent and name, retired on unlink, rmdir and rename so a name that is created again is never handed a removed entry's number - writable nodes carry a generation of their own, so go-fuse builds a fresh node per lookup instead of reusing one bound to an mfs.File that boxo has since replaced - mount points report inode 1 instead of 0, and readdir reports the same numbers as stat fix(fuse): tell two CIDs apart on /ipfs go-fuse matches a lookup against the nodes it already holds by the whole of StableAttr, so two entries that agree on it are served as one object. The inode number alone is 63 bits, and two CIDs can end up sharing one, by chance or by choice. Reading the second one then returned the first one's bytes. A dag-pb CID and the raw CID of the same block hit this every time, because the number ignored the codec. The number and the generation now both come from a hash of the codec and the multihash, so it takes a match on 128 bits to confuse two entries. The mount also sets FirstAutomaticIno instead of relying on go-fuse's default, so the range it keeps for itself stays where the code says it is. * fix(fuse): stat /ipfs entries we cannot read A stat of a child whose block is missing, or of one in a codec this mount does not decode, read UnixFS metadata that was never loaded and panicked. go-fuse does not recover a panic in its serve loop, so this took the whole daemon down. Neither case is exotic: a dag-cbor object linked from a UnixFS directory needs no missing blocks at all. A lookup that cannot read the block now fails instead of building an entry from it, and a block that is not UnixFS is reported as a file of its own size. * fix(fuse): keep a file a rename cannot move `mv /ipns/<key>/f /ipns/f` unlinked the source before finding out that the /ipns root holds no files of its own, then failed with EINVAL and left the file nowhere. The destination is now checked before anything is written. * fix(fuse): store a moved file where it landed A rename across directories only wrote the source directory back, so the file's new name lived in memory until something else flushed it, and a daemon that stopped first lost the file. The destination is written back first, so an interrupted rename leaves the file under both names rather than under neither. * fix(fuse): keep the inode number over a rename Both names gave up their inode numbers on rename, so a moved file came back about a second later as a different file, which is the problem this branch fixes everywhere else. A moved directory was worse: its entries are keyed by its number, so the whole subtree was renumbered and the old keys were left behind until unmount. The number now moves with the entry. Nothing else has to change to make that safe: go-fuse cannot hand back the old node anyway, because every node gets a generation of its own. The comments that credited the renumbering for it were wrong. * test(fuse): check listings and stat agree Each mount fills in the inode number of a directory entry separately from the one it reports to stat, and nothing compared the two. Tools read whichever is cheaper for them. * docs(config): warn about writing to a mounted mfs `ipfs files` writes to the same tree as the /mfs and /ipns mounts without the mount knowing, so the two can lose each other's writes.	16 天前
node	fix(fuse): switch to hanwen/go-fuse (#11272) * test(fuse): consolidate FUSE tests into test/cli/fuse Move FUSE integration tests from sharness shell scripts (t0030, t0031, t0032) and test/cli/fuse_test.go into a dedicated test/cli/fuse/ Go sub-package, ensuring all FUSE test cases run in CI. - git mv test/cli/fuse_test.go to test/cli/fuse/ (package fuse) - convert all sharness FUSE tests to Go subtests under TestFUSE: mount failure, IPNS symlink, IPNS NS map resolution, MFS file/dir creation, xattr (Linux), files write, add --to-files, file removal, nested dirs, publish-while-mounted block, sharded directory reads - add xattr helpers with build tags (linux/other) using unix.Getxattr - split make test_fuse into test_fuse_unit (./fuse/...) and test_fuse_cli (./test/cli/fuse/...) sub-targets - set TEST_FUSE=0 in test_cli so FUSE tests skip in cli-tests CI job - increase fuse-tests CI timeout from 5m to 10m for CLI tests - delete sharness t0030, t0031, t0032 (were always skipped in CI) * docs: document FUSE test split between unit and e2e Add cross-reference comments between the unit tests in fuse/readonly/, fuse/ipns/, fuse/mfs/ and the end-to-end CLI tests in test/cli/fuse/. Also fix AGENTS.md to use a temp dir for fusermount symlink instead of sudo. * ci: prevent stale FUSE mounts from failing fuse-tests On shared self-hosted runners, leftover mount points from previous runs can exhaust the kernel FUSE mount limit. - add job-level concurrency group so only one fuse-tests runs at a time - lazy-unmount stale /tmp/fusetest* mounts before running tests * ci: only symlink fusermount3 when fusermount is missing * fix(fuse): remove goroutine leak in IPNS Flush handler The Flush handler wrapped fi.fi.Flush() in a goroutine so it could return early when the FUSE context was canceled. But the goroutine kept running in the background, and when Release arrived it called Close on the same file descriptor concurrently. The two paths both entered DagModifier.Sync, racing on its internal write buffer and causing a nil pointer panic. The fix is to call Flush directly without a goroutine. The MFS flush cannot be safely canceled mid-operation anyway, so the goroutine only added the illusion of cancellation while leaking work and masking the real error. Also bumps boxo to pick up the matching defense-in-depth fix that serializes FileDescriptor.Flush and Close with a mutex. * fix(fuse): add mutex to IPNS file handle operations bazil/fuse dispatches each FUSE request in its own goroutine. The IPNS File handle had no synchronization, so concurrent Read/Write/Flush/Release calls could overlap on the underlying DagModifier which is not safe for concurrent use. Add sync.Mutex to File, matching the pattern already used by the MFS FileHandler. * refactor(fuse): remove dead File.Forget method bazil/fuse only dispatches Forget to nodes via the NodeForgetter interface. File is a handle, not a node, so this method was never called. The /mfs mount has no equivalent. * fix(fuse): flush IPNS directory after Remove and Rename The /mfs mount flushes the directory after Unlink and Rename so changes propagate to the MFS root immediately. The /ipns mount did not, leaving mutations pending until an unrelated flush. Also add an empty-directory check before removing directories, matching the /mfs mount's safety check. * fix(fuse): inherit CID builder and flush on IPNS Create New files created via the /ipns FUSE mount now inherit the CID builder from their parent directory, preventing CIDv0 nodes from appearing inside a CIDv1 tree. The directory is also flushed after AddChild so the new entry propagates to the MFS root immediately, matching the /mfs mount. * test(fuse): add IPNS Remove and non-empty rmdir tests Cover the file removal path and the empty-directory safety check added in the previous commit. TestRemoveFile verifies a created file can be removed and is gone afterwards. TestRemoveNonEmptyDirectory verifies that rmdir on a directory with children fails, and succeeds once the children are removed first. * feat(fuse): read UnixFS mode/mtime, add StoreMtime/StoreMode config All three FUSE mounts now read mode and mtime from UnixFS metadata when present, falling back to POSIX defaults when absent. Most IPFS data does not include this optional metadata. Writing mode and mtime is opt-in via two new config flags: - Mounts.StoreMtime: persist mtime on file create and open-for-write - Mounts.StoreMode: persist mode on chmod Other changes in this commit: - align default file/dir modes across /ipns and /mfs to 0644/0755 - share mode constants via fuse/mount/mode.go - convert Mounts.FuseAllowOther from bool to Flag for consistency - add Setattr to /ipns FileNode and /mfs File for chmod and touch - move dead File.Setattr from IPNS handle to FileNode (node) - bump boxo for Directory.Mode() and Directory.ModTime() getters * feat(fuse): add ipfs.cid xattr to all mounts All three FUSE mounts now expose the node's CID via the ipfs.cid extended attribute on both files and directories. The /mfs mount also accepts the old ipfs_cid name for backward compatibility. The /ipfs mount previously had a stub that returned nil for all xattrs; it now returns the correct CID. The xattr name follows the convention used by CephFS (ceph.), Btrfs (btrfs.), and GlusterFS (glusterfs.). feat(fuse): switch from bazil.org/fuse to hanwen/go-fuse v2 Replace the unmaintained bazil.org/fuse (last commit 2020) with hanwen/go-fuse v2.9.0, fixing two architectural issues that could not be solved with the old library. ftruncate now works: hanwen/go-fuse passes the open file handle to NodeSetattrer, so Setattr can truncate through the existing write descriptor instead of trying to open a second one (which deadlocks on MFS's single-writer lock). fsync now works: FileFsyncer runs on the handle directly, flushing the write buffer through the open descriptor. Previously a no-op because bazil dispatched Fsync to the inode only. mount package: - NewMount takes (InodeEmbedder, mountpoint, fs.Options) instead of (fs.FS, mountpoint, allowOther) - mount/unmount collapses to a single fs.Mount call - fusermount3 tried before fusermount in ForceUnmount all three mounts: - structs embed fs.Inode (hanwen's InodeEmbedder pattern) - Remove split into Unlink + Rmdir (separate FUSE interfaces) - ReadDirAll replaced with Readdir returning DirStream - fillAttr helper shared between Getattr and Lookup responses - kernel cache invalidation via NotifyContent after Flush - 1s entry/attr timeout for writable mounts (matches go-fuse default, gocryptfs, rclone) - O_APPEND tracked on file handle, writes seek to end - build tags standardized to (linux \|\| darwin \|\| freebsd) && !nofuse tests: - replaced bazil fstestutil.MountedT with shared fusetest.TestMount - fixed TestConcurrentRW: channel drain mismatch and missing sync between write Close and read start - added TestFsync, TestFtruncate, TestReadlink, TestSeekRead, TestLargeFile, TestRmdir, TestCrossDirRename, TestUnknownXattr - added StoreMtime disabled/enabled subtests fix(fuse): close fd on error in Open to prevent leak MFS enforces a single-writer lock, so a leaked write descriptor blocks all subsequent opens of that file until GC. * fix(fuse): detect external unmount via server.Wait Without this, IsActive stays true after `fusermount -u` and Unmount returns nil instead of ErrNotMounted. * fix(fuse): return actual error from Unlink/Rmdir, not ENOENT After confirming the child exists, an Unlink failure could be an IO error. Returning ENOENT would hide the real cause. * fix(fuse): reuse DagReader per open, pass ctx to all reads Readonly Open now returns a file handle holding a DagReader instead of recreating one per Read call. Sequential reads no longer re-traverse the DAG from the root on each kernel request. All three mounts now use CtxReadFull with the kernel's per-request context so killing a process mid-read cancels in-flight block fetches instead of letting them complete uselessly. * chore(fuse): cleanup dead code, add var comments - remove dead `_ = mntDir` in TestXattrCID - comment why immutableAttrCacheTime and mutableCacheTime are var - add TODO for using IPNS record TTL as cache timeout * chore(fuse): replace OSXFUSE 2.x check with macFUSE detection The old check tried to verify OSXFUSE >= 2.7.2 to avoid a kernel panic from 2015. It used sysctl, tried to `go install` a third-party tool at runtime, and referenced paths that no longer exist. Replace with a simple check for the macFUSE mount helper, matching the same paths go-fuse looks for. If neither macFUSE nor OSXFUSE is found, point the user to the install page. Also standardize build tags to (linux \|\| darwin \|\| freebsd) && !nofuse and use strings.ReplaceAll. * fix(fuse): include mountpoint path in mount errors go-fuse's fusermount errors don't include the path, so tools that check error messages for the mountpoint name couldn't tell which mount failed. * chore(ci): remove bazil fusermount workaround go-fuse finds fusermount3 natively, no symlink needed. The stale mount cleanup was for bazil's fstestutil which we no longer use. * docs: update v0.41 changelog for FUSE rewrite * chore(deps): bump boxo for full FileDescriptor serialization boxo@64be0815 extends the mutex from Flush/Close to all FileDescriptor operations (Read, Write, Seek, Truncate, Size), preventing data races on the underlying DagModifier. * chore(deps): bump boxo to merged ipfs/boxo#1133 Picks up full FileDescriptor serialization: the mutex now covers all operations (Read, Write, Seek, Truncate, Size), not just Flush and Close. * feat(fuse): CAP_ATOMIC_O_TRUNC, new integration tests Advertise CAP_ATOMIC_O_TRUNC so the kernel sends O_TRUNC inside Open instead of doing a separate SETATTR(size=0) first. Without this, the kernel's SETATTR needs to open a write descriptor inside Setattr, which deadlocks on MFS's single-writer lock. Move kernel cache invalidation from Flush to Release because mfsFD.Close (in Release) is where the final DAG node is committed. Upgrade go-fuse to latest for ExtraCapabilities support. New tests for both MFS and IPNS: - TestOpenTrunc, TestSeekAndWrite, TestOverwriteExisting - TestTempFileRename, TestVimSavePattern, TestRsyncPattern (skipped pending rename-over-existing and cache fixes) * fix(fuse): rename-over-existing, bump boxo for flushUp race fix IPNS Rename now unlinks the target before AddChild, matching MFS. Without this, renaming onto an existing name returned "directory already has entry". Bump boxo to pick up the flushUp unlinked-entry fix (ipfs/boxo@8ae46d5): when a file descriptor outlives its directory entry (FUSE RELEASE racing with RENAME), flushUp no longer re-adds the stale name. Unskip TestTempFileRename and TestRsyncPattern on both mounts. * fix(fuse): unskip VimSavePattern, bump boxo for setNodeData fix boxo@552d8e7 fixes File.setNodeData dropping content links when updating metadata (mode, mtime). chmod or touch after write no longer makes the file appear empty. Unskip TestVimSavePattern on both mounts. Remove debug logging and temporary test functions added during investigation. * fix(fuse): build tags for cross-compilation go-fuse does not compile on windows/openbsd/netbsd/plan9. Move WritableMountCapabilities (which imports go-fuse) from mode.go (no build tag) to caps.go (platform-gated). Align build tags on fusetest and core/commands/mount stubs so unsupported platforms don't pull in go-fuse transitively. * fix(test): use fusermount3 in CLI FUSE tests The doUnmount helper hardcoded fusermount, but systems with only fuse3 installed have fusermount3. Try fusermount3 first, matching what go-fuse and our ForceUnmount already do. * feat(fuse): symlink support on writable mounts Add NodeSymlinker to MFS and IPNS directories. Symlinks are stored as UnixFS TSymlink nodes in the DAG, the same format used by `ipfs add` for directories containing symlinks. The readonly /ipfs mount already rendered existing symlinks; now /mfs and /ipns can create them too. The target string is cached at Lookup time to avoid re-parsing the DAG node on every Readlink call. Symlink permissions are always 0777 per POSIX convention (access control uses the target's mode). * fix(fuse): checked type assertion in MFS Rename The direct type assertion on newParent could panic if the kernel passed a non-directory inode. Use a checked assertion with EINVAL fallback, matching the type-switch pattern in the IPNS mount. * fix(test): add missing continue in stress test Missing continue after error sends let execution fall through to nil type assertions (read.(files.File)) that would panic on error. Also cancel the context before continuing to avoid leaking it. * fix(fuse): return error from Readdir when DAG.Get fails Abort the directory listing instead of silently omitting the unretrievable entry. Callers get EIO, which is more honest than a partial listing that hides missing blocks. * docs: remove duplicate fsync bullet in changelog * ci: clean up stale FUSE mounts in fuse-tests job On shared self-hosted runners, leftover mounts from crashed runs can exhaust the kernel mount_max limit. Lazy-unmount kubo-test and harness temp mounts before and after tests. * chore(deps): bump boxo to merged ipfs/boxo#1134 Picks up flushUp unlinked-entry guard and setNodeData content link preservation. * docs: add build tag comments, normalize tag style Add a one-line comment above every //go:build directive explaining why the constraint exists. Normalize tag style: positive platform constraints first, then feature flags/negations. Simplify redundant expressions. * fix(fuse): add Setattr to directories for chmod and mtime Tools like tar and rsync call utimensat on directories after extraction. Without Setattr on Dir, this returned ENOTSUP. Add Setattr to Dir (MFS) and Directory (IPNS) that handles mode and mtime the same way as the file-level Setattr. When StoreMtime or StoreMode is disabled the call succeeds silently, matching the file-level behavior. * docs: clarify directory support and spec link for StoreMtime/StoreMode - mention that touch and chmod work on both files and directories - note tar and rsync as practical use cases - link to UnixFS spec for optional metadata storage * fix(fuse): use proper mode conversion, document 9-bit limit Use files.UnixPermsToModePerms and files.ModePermsToUnixPerms for converting between FUSE kernel mode (unix 12-bit layout) and Go's os.FileMode (different bit positions for setuid/setgid/sticky). The UnixFS spec supports all 12 permission bits, but boxo's MFS layer (File.Mode, Directory.Mode) exposes only the lower 9. FUSE mounts are always nosuid so the upper 3 bits would have no effect. Add TestSetuidBitsStripped to both mounts confirming the behavior. * feat(fuse): symlink Setattr with mtime persistence Wire the backing mfs.File into the FUSE Symlink struct so Setattr can call SetModTime when StoreMtime is enabled. boxo's File methods (SetModTime, ModTime) already work on TSymlink nodes since they operate on the FSNode protobuf without checking the type. Without Setattr, rsync -a fails with "failed to set times" on symlinks. Every major FUSE filesystem (gocryptfs, rclone, sshfs, s3fs) implements Setattr on symlinks for this reason. Mode is always 0777 per POSIX convention, so chmod requests are silently accepted but not stored. * fix(fuse): return EIO instead of panicking on unknown node type Replace panic with log.Errorf + syscall.EIO in IPNS Directory.Lookup for unexpected MFS node types. Also remove duplicate comment block on File.Flush. * docs: update FUSE docs for go-fuse migration - fuse.md: replace stale OSXFUSE section with macFUSE, remove obsolete go-fuse-version tool, fix broken FreeBSD sudo echo, update xattr example to ipfs.cid with CIDv1, add mode/mtime section, add unixfs-v1-2025 tip, add debug logging section, add TOC, link to hanwen/go-fuse - changelog: refine bullet wording, link to fuse.md - config.md: fix double space, update fuse.md link text - experimental-features.md: fix double space, soften wording - README.md: add FUSE to features list and docs table * refactor(fuse): extract shared writable types and test suite Extract duplicated code from fuse/mfs and fuse/ipns into a shared fuse/writable package, and consolidate duplicated tests into a reusable suite in fuse/fusetest. - fuse/writable: Dir, FileInode, FileHandle, Symlink types with all FUSE interface methods, shared by both mounts - fuse/fusetest: RunWritableSuite with helpers, exercised by both mfs and ipns via mount-specific factories - fix cache invalidation race: NotifyContent in Flush (synchronous) in addition to Release (async), so stat after close sees new size - drop deprecated ipfs_cid xattr, log error guiding users to ipfs.cid - mfs_unix.go: 632 -> 19 lines (thin wrapper over writable.Dir) - ipns_unix.go: 795 -> 170 lines (Root + key resolution only) - mfs_test.go: 1183 -> 95 lines (factory + persistence test) - ipns_test.go: 1309 -> 162 lines (factory + IPNS-specific tests) - tests that were only in one mount now run on both * feat(fuse): add macOS-specific mount options Set volname, noapplexattr, and noappledouble on macOS via PlatformMountOpts, applied in NewMount so all three mounts benefit automatically. - volname: shows mount name in Finder instead of "macfuse Volume 0" - noapplexattr: suppresses Finder's com.apple.* xattr probes - noappledouble: prevents ._ resource fork sidecar files * fix(fuse): detect symlinks in readdir, fix stale refs Readdir on writable mounts now checks the underlying DAG node type for TFile entries, reporting S_IFLNK for symlinks instead of regular file. This makes ls -l and find -type l work correctly. - writable: Readdir checks SymlinkTarget for TFile entries - writablesuite: add SymlinkReaddir regression test - readonly: add TestReaddirSymlink regression test - test/cli/fuse: fix stale bazil.org/fuse reference in doc comment * fix(fuse): normalize deprecated ipfs_cid xattr to ipfs.cid Getxattr for the old "ipfs_cid" name now returns the CID instead of ENOATTR, keeping existing tooling working during the deprecation period. A log error is emitted on each access to nudge migration. * fix(fuse): serialize concurrent reads on readonly file handles The go-fuse server dispatches each FUSE request in its own goroutine. On files larger than 128 KB the kernel issues concurrent readahead Read requests on the same file handle, racing on the shared DagReader's Seek+CtxReadFull sequence and corrupting its internal state. Add sync.Mutex to roFileHandle (matching the existing pattern in writable.FileHandle) and lock in Read and Release. - fuse/readonly/readonly_unix.go: add mu sync.Mutex to roFileHandle - fuse/readonly/ipfs_test.go: add TestConcurrentLargeFileRead - fuse/fusetest/writablesuite.go: add LargeFileConcurrentRead to shared writable suite (exercised by both /mfs and /ipns tests) * fix(fuse): bypass MFS locking for read-only opens MFS uses an RWMutex (desclock) that holds RLock for the lifetime of a read descriptor and requires exclusive Lock for writes. Tools like rsync --inplace open the same file for reading and writing from separate processes, deadlocking on this mutex. For O_RDONLY opens, create a DagReader directly from the current DAG node instead of going through MFS. The reader gets a point-in-time snapshot and never touches desclock, so writers proceed independently. - fuse/writable/writable.go: add roFileHandle with DagReader for read-only opens, add DAG field to Config - fuse/mfs/mfs_unix.go: pass ipfs.DAG to writable Config - fuse/ipns/ipns_unix.go: pass ipfs.Dag() to writable Config - fuse/fusetest/writablesuite.go: add ConcurrentReadWrite test exercising simultaneous read and write on the same file * fix(fuse): support truncate(path, size) without open fd Open a temporary write descriptor in Setattr when the kernel sends a size change without a file handle (the truncate(2) syscall, as opposed to ftruncate(fd) which passes the handle). Previously this returned ENOTSUP. - fuse/writable: open, truncate, flush, close in Setattr else branch - fuse/fusetest: add TruncatePath to the shared writable suite - test/cli/fuse: add end-to-end truncation test covering ftruncate(fd), syscall.Truncate(path), and open(O_TRUNC) through a real daemon * ci(fuse): get stack traces on test hangs The fuse-tests job was being silently cancelled by GitHub at 10min because Go's per-test timeout (5m) was the same order as the job timeout, and GOTRACEBACK=single hid the hung goroutines anyway. - shrink TEST_FUSE_TIMEOUT to 4m so Go's panic fires first - shrink job timeout-minutes to 6 (normal run is ~3min) - set GOTRACEBACK=all so the panic dumps every goroutine, not just the timer * fix(fuse): fill attrs in FileInode.Setattr response Without this, the kernel could cache zero attrs after a chmod, touch, or ftruncate until AttrTimeout (1s) expired. Dir.Setattr and Symlink.Setattr already fill out.Attr; FileInode.Setattr now matches. * docs(config): clarify Mounts.IPNS writability scope Only directories backed by keys the node holds are writable. All other names resolve via IPNS to read-only symlinks into the /ipfs mount. * fuse: review cleanup for go-fuse migration Final pass on #11272 addressing review feedback. - writable: panic in NewDir if Config.DAG is nil. Both call sites already supply it, but a nil value silently fell back to the MFS path in FileInode.Open, re-introducing the rsync --inplace deadlock the read-only fast path was added to fix. - writable: document Dir.Rename non-atomicity. Source unlink happens before destination add, so any failure between the two loses the source. An atomic fix requires changes in boxo/mfs. - writable: add unit test locking in that Symlink.Setattr accepts a mode-only request without erroring and does not store the requested mode (POSIX symlinks have no meaningful permission bits). - docs/config: correct StoreMode default modes; the previous text listed 0666 for files, which the code never uses. * docs(config): list StoreMtime and StoreMode in Mounts TOC * fix(fuse): fill EntryOut attrs in Dir.Create and Dir.Mkdir Without this, fstat on the file handle returned by Create reports mode 0 and size 0 for up to AttrTimeout (1s), because the kernel caches the empty attrs from the Create response. Path-based stat goes through Lookup which already fills attrs, so the bug only shows up via fstat. Mirrors the same fix already applied to FileInode.Setattr. Dir.Mkdir gets the same fillAttr treatment for consistency, plus a TODO noting that boxo's mfs.Directory.Mkdir accepts no mode arg so the caller's mode is dropped on creation. Adds CreateAttrsImmediate and MkdirAttrsImmediate to the shared writable suite to guard both paths against future regressions. * fix(fuse): map context cancellation to EINTR in read paths When a userspace process is killed mid-read (Ctrl-C, SIGKILL on a stuck cat) the kernel sends FUSE_INTERRUPT and go-fuse cancels the per-request context. fs.ToErrno does not recognise context.Canceled and falls through to "function not implemented", which the kernel cannot act on. Map context.Canceled and DeadlineExceeded to EINTR so the syscall is correctly aborted. - mount/errno.go: new ReadErrno helper used by all context-aware read paths in both readonly and writable mounts - readonly: applied to Node.Open, Node.Readdir, roFileHandle.Read - writable: applied to FileInode.Open, FileHandle.Read, roFileHandle.Read - readonly/ipfs_test.go: TestReadCancellationUnblocks guards the contract via a blocking DagReader fake; without ReadErrno the test reports "function not implemented" instead of EINTR * test(fuse): add OExcl, DirRename, SparseWrite, FsyncCrossHandle Coverage gaps in the shared writable suite: - OExcl: lock files and atomic-create patterns rely on the second open with O_CREATE\|O_EXCL failing with EEXIST - DirRename: previously only file rename and cross-dir file rename were tested; this exercises Rename on a directory inode - SparseWrite: WriteAt past the end of an empty file must report the correct size and return zeros for the gap - FsyncCrossHandle: a reader on a fresh fd must see data flushed by fsync on the writer fd, not just after close * test(fuse): cover external unmount on /ipns and /mfs Previously TestExternalUnmount only exercised /ipfs, leaving the goroutine that watches fuse.Server.Wait() untested for the other two mounts. Refactor into a table-driven test that runs the same fusermount/umount-then-IsActive flow against all three mounts. Switch to coremock.NewMockNode so the node is online: doMount only attaches the /ipns mount when node.IsOnline is true, and the table needs all three populated. * fix(commands): align 'ipfs mount' output columns MountCmd's LongDescription has "MFS mounted at:" with two spaces so the column lines up with the 4-char "IPFS" and "IPNS" rows above, but the runtime encoder and the daemon's startup print used a single space and produced misaligned output. Bring both runtime sites in line with the help text, and update the two existing test fixtures (test/cli/fuse and the sharness test-lib helper that t0040-add-and-cat.sh still uses) to expect the aligned form. * fix(fuse): invalidate kernel cache on Fsync FileHandle.Fsync only flushed the MFS file descriptor and left the kernel's cached attrs and content for the inode untouched. A fresh reader on the same path then saw the size cached from the original Create response (zero), reading zero bytes regardless of how much the writer had synced. Mirror the cache invalidation already done in Flush via inode.NotifyContent(0, 0) so a writer that fsyncs while another process opens the file (vim then a follow-up cat, IDE then a language server) sees consistent state. Sharpen the FsyncCrossHandle assertion to report the size delta on failure; the bug surfaced as got=0/want=500 only after switching from bytes.Equal to require.Equal. * chore(gitignore): ignore test_fuse_unit and test_fuse_cli json output The new test_fuse_unit and test_fuse_cli make targets emit test/fuse/fuse-unit-tests.json and test/fuse/fuse-cli-tests.json respectively, the same gotestsum --jsonfile pattern that test_unit and test_cli already use. Add them to the same .gitignore section so a local test run does not leave the working tree dirty. * test(fuse): end-to-end coverage with real POSIX tools Adds TestFUSERealWorld in test/cli/fuse/realworld_test.go: a single shared-daemon test with 18 subtests that exercise the writable /mfs mount through the actual binaries users invoke (sh, cat, seq, wc, ls, stat, cp, mv, rm, ln, readlink, find, dd, sha256sum, tar, rsync, vim). Each subtest verifies the result both via the FUSE filesystem and via 'ipfs files read\|stat\|ls' so both views agree. Synthetic payloads default to 1 MiB + 1 byte so multi-chunk read/write paths are exercised, not just single-chunk fast paths. External tools are required, not optional: a missing binary fails the test loudly so a CI image change cannot silently turn the suite green. The whole-suite TEST_FUSE gate is the only place a developer is allowed to skip. runCmd forces LC_ALL=C so locale-sensitive tool output (date formats in 'ls -l', decimal separators in 'wc', localized error messages, find/ls collation) is deterministic regardless of the runner's locale settings. One shared daemon across all 18 subtests keeps total runtime under two seconds; isolation comes from per-subtest subdirectories under the mount.	4 个月前
readonly	fix(fuse): keep what you write after a rename (#11430) * fix(fuse): keep a rename's writes go-fuse hands the kernel's existing node to the new name once Dir.Rename returns, and that node still held the MFS handle the rename had unlinked. MFS treats such a handle as gone: a write through it was accepted and then dropped, so `mv a b` followed by a write to `b` read back the old contents a second later, once the entry cache expired. A directory was worse. Creating a file in one that had just been renamed flushed through the dead handle, which carried the name the rename had moved away from, so the new file was lost and the old directory came back for good. Each node now reaches its MFS handle through an atomic, and a rename points the moved node at the entry that exists afterwards. Entries the kernel had already looked up underneath a renamed directory hang off the handle it was reached through, so the walk follows them down; it covers what the kernel is holding, not the whole tree. Invalidating the entry instead was tried and does not work: the kernel processes FUSE_NOTIFY_INVAL_ENTRY while holding the parent inode lock, so notifying from inside Rename deadlocks, and notifying asynchronously still loses most of the writes it races. Left unfixed: a write through a file descriptor held open across the rename still goes to the descriptor opened from the old handle. * fix(fuse): refuse to replace a non-empty directory A rename may only overwrite a directory that is empty. MFS removes a directory and everything under it without complaint, so `mv -T src dst` took dst's contents with it and reported success. Rmdir already had the check; Rename now makes it too, before it unlinks anything. The check for an absent destination also goes through errors.Is now. It compares against a sentinel that boxo returns bare today, and the cost of that changing is the source file, which by then has been unlinked. * fix(fuse): read /ipfs blocks we cannot decode A UnixFS directory can link to a block of any codec. stat reports one the mount cannot decode as a file the size of the block, but every read of it failed, because the read path went looking for a UnixFS DAG that is not there. A size stat promises has to be a size reads deliver, so serve the block itself. * fix(fuse): list a directory with a missing block One child whose block is not held locally failed the whole listing, and with an errno the caller could make nothing of: ipld.ErrNotFound has no mapping, so it arrived as ENOSYS. The readable entries are worth having, so report the one that is missing with no type and let a stat of it say what is wrong. * fix(fuse): report the CID the path used The ipfs.cid xattr answered with a CID the caller had never seen. A lookup rebuilds the node by decoding the block, which drops the version and codec of the path it came from, so a v1 dag-pb path reported its v0 form. Keep the CID the entry resolved to and report that. The changelog entry also covers the rename check from the commit before it, which landed without one. * test(fuse): make the rename tests catch their bugs TestRenameOntoNamespaceRoot read the file back through the mount, which answers from the entry the kernel still has cached and so succeeds whether or not the rename took the file away. It passed against the bug it was written for. Ask MFS instead. The dirent helper also loops on a record length it never checks, which would spin rather than fail if the kernel ever sent zero. * test(ipns): settle the repo path before mounting TestStatfs assigned Root.RepoPath once the server was already serving, and Statfs reads it from a FUSE handler goroutine, so `go test -race ./fuse/...` reported a data race on every run. The mfs and readonly tests already settle it before their mount; do the same here.	16 天前
writable	fix(fuse): keep what you write after a rename (#11430) * fix(fuse): keep a rename's writes go-fuse hands the kernel's existing node to the new name once Dir.Rename returns, and that node still held the MFS handle the rename had unlinked. MFS treats such a handle as gone: a write through it was accepted and then dropped, so `mv a b` followed by a write to `b` read back the old contents a second later, once the entry cache expired. A directory was worse. Creating a file in one that had just been renamed flushed through the dead handle, which carried the name the rename had moved away from, so the new file was lost and the old directory came back for good. Each node now reaches its MFS handle through an atomic, and a rename points the moved node at the entry that exists afterwards. Entries the kernel had already looked up underneath a renamed directory hang off the handle it was reached through, so the walk follows them down; it covers what the kernel is holding, not the whole tree. Invalidating the entry instead was tried and does not work: the kernel processes FUSE_NOTIFY_INVAL_ENTRY while holding the parent inode lock, so notifying from inside Rename deadlocks, and notifying asynchronously still loses most of the writes it races. Left unfixed: a write through a file descriptor held open across the rename still goes to the descriptor opened from the old handle. * fix(fuse): refuse to replace a non-empty directory A rename may only overwrite a directory that is empty. MFS removes a directory and everything under it without complaint, so `mv -T src dst` took dst's contents with it and reported success. Rmdir already had the check; Rename now makes it too, before it unlinks anything. The check for an absent destination also goes through errors.Is now. It compares against a sentinel that boxo returns bare today, and the cost of that changing is the source file, which by then has been unlinked. * fix(fuse): read /ipfs blocks we cannot decode A UnixFS directory can link to a block of any codec. stat reports one the mount cannot decode as a file the size of the block, but every read of it failed, because the read path went looking for a UnixFS DAG that is not there. A size stat promises has to be a size reads deliver, so serve the block itself. * fix(fuse): list a directory with a missing block One child whose block is not held locally failed the whole listing, and with an errno the caller could make nothing of: ipld.ErrNotFound has no mapping, so it arrived as ENOSYS. The readable entries are worth having, so report the one that is missing with no type and let a stat of it say what is wrong. * fix(fuse): report the CID the path used The ipfs.cid xattr answered with a CID the caller had never seen. A lookup rebuilds the node by decoding the block, which drops the version and codec of the path it came from, so a v1 dag-pb path reported its v0 form. Keep the CID the entry resolved to and report that. The changelog entry also covers the rename check from the commit before it, which landed without one. * test(fuse): make the rename tests catch their bugs TestRenameOntoNamespaceRoot read the file back through the mount, which answers from the entry the kernel still has cached and so succeeds whether or not the rename took the file away. It passed against the bug it was written for. Ask MFS instead. The dirent helper also loops on a record length it never checks, which would spin rather than fail if the kernel ever sent zero. * test(ipns): settle the repo path before mounting TestStatfs assigned Root.RepoPath once the server was already serving, and Statfs reads it from a FUSE handler goroutine, so `go test -race ./fuse/...` reported a data race on every run. The mfs and readonly tests already settle it before their mount; do the same here.	16 天前