Thanks to alexanderharm for suggesting this fix.

Markdown-ify README

The function was written with only a single return at
the end, which was elegant, but harder to follow.

Add early returns to handle common cases and add a few
comments.

A functional change is that we now catch when we get
OK with fi == nil and return EINVAL and log a message.
This should not happen and is a bug in the filesystem
implementation.

Passes ./all.bash and the gocryptfs test suite.

Consulting the output size is incorrect as it does not take into account the
cases where an xattr with an empty value exists.

Moreover, treat listxattr exactly the same as getxattr as the two both
can be used for two different operations.

Add a test for an xattr with an empty value
Signed-off-by: Shayan Pooya <shayan@arista.com>

This helps settle how much performance improvement splice()ing files
brings. The answer is that end-to-end read performance is improved by
about 10% for reading large files.

This saves a couple of bytes, and avoids a copy on serializing the
response header.

Add and document a hack for _OP_GETXATTR in serializeHeader.

Trivial patch that fixes a blank line in the debug output.

Before:

  2017/10/03 20:25:56 Dispatch 602: READDIRPLUS, NodeId: 1. data: {Fh 2 off 0 sz 4096  L 0 DIRECTORY,NONBLOCK,0x8000}
  2017/10/03 20:25:56 Serialize 602: READDIRPLUS code: OK value:  832 bytes data

  2017/10/03 20:25:56 Dispatch 603: GETXATTR, NodeId: 7. data: {sz 0} names: [system.posix_acl_access] 24 bytes
  2017/10/03 20:25:56 Serialize 603: GETXATTR code: 61=no data available value:

After:

  2017/10/03 21:02:46 Dispatch 22: READDIRPLUS, NodeId: 1. data: {Fh 2 off 0 sz 4096  L 0 RDONLY,0x8000}
  2017/10/03 21:02:46 Serialize 22: READDIRPLUS code: OK value:  648 bytes data
  2017/10/03 21:02:46 Dispatch 23: READDIRPLUS, NodeId: 1. data: {Fh 2 off 4 sz 4096  L 0 RDONLY,0x8000}
  2017/10/03 21:02:46 Serialize 23: READDIRPLUS code: OK value:

With golang 1.9, calling WaitMount is necessary to avoid hitting the
deadlock mentioned in #165 in tests where a single process acts as
both the fuse daemon and the client.
Signed-off-by: Shayan Pooya <shayan@arista.com>

Reduces Sizeof(nodefs.Inode{}) from 112 to 96 bytes.

The "generation" field in "Inode" is not used anywhere.
"portableHandleMap" has it's own generation field, which
is actually used.

The "check" field, embedded through the "handled" struct,
is checked for zero, but never set.

This is part of an effort to reduce the memory usage of go-fuse
and gocryptfs. Details can be seen at
https://github.com/rfjakob/gocryptfs/issues/132 .

When an app in a FUSE mount calls getdents(2), go-fuse receives
READDIR[PLUS] and calls the filesystem's OpenDir function that
returns []DirEntry.

The data returned from getdents(2) contains an inode number for
each directory entry, "d_ino". Until now, struct DirEntry had no
corresponding field and the value passed to the kernel was always
FUSE_UNKNOWN_INO = 0xffffffff

This broke apps that actually look at the d_ino field, like
"find -inum".

This commit adds the "Ino" filed to struct DirEntry. If the field
is not set by the filesystem, it is set to FUSE_UNKNOWN_INO,
as before. Otherwise it is left alone and passed to the kernel.

loopbackFileSystem's OpenDir function is extended to set the inode
number. A test verifies that the returned inode number is sane.

Fixes https://github.com/hanwen/go-fuse/issues/175

When a process forks and execs a new process, the child will not be
interested in the parent's file descriptor to /dev/fuse.
Signed-off-by: Shayan Pooya <shayan@arista.com>

TestUnionFsFdLeak looks at global state of the program, and is flaky:
the splice pool keeps open file descriptors, and its number depends on
how many read threads are scheduled in parallel.

Some versions of FUSE need this; fixes pathfs owner tests.

Inode.Parent() has to take the treeLock before looking
at the parents map.

Fixes https://github.com/hanwen/go-fuse/issues/166

The locking code is not hooked in the fuse/ server code, so this
just exercises the kernel's default locking implementation.

The parent tracking moved into nodefs in
commit 0107672a, so there is no need for
taking that lock anymore.

Go 1.9 uses epoll() for more efficient file I/O. File I/O causes a
call to epoll, and the runtime makes this call take up a GOMAXPROCS
slot.

The FUSE kernel module also supports poll: polling on a file residing
in a FUSE file system causes the kernel to sends a POLL request to the
userspace process.  If the process responds with ENOSYS, the kernel
will stop forwarding poll requests to the FUSE process.

In a test for Go FUSE file systems, it is normal to serve the
filesystem out of the same process that opens files in the file
system. If this happens in Go 1.9, the epoll call can take the only
GOMAXPROCS slot left, leaving the process unable to respond to the
FUSE POLL opcode, deadlocking the process.

This change add support for a magic file "/ .go-fuse-epoll-hack" with
node ID uint64(-1), and on starting up the file system, the library
calls poll() on this file, triggering the POLL opcode before the Go
runtime had a chance to do so.

There are two problem scenarios left:

* File system tests that start I/O before calling WaitMount() still
  risk deadlocking themselves.

* The Linux kernel keeps track of feature support in fuse_conn, which notes

         * The following bitfields are only for optimization purposes
         * and hence races in setting them will not cause malfunction

  if our forced ENOSYS gets lost due to a race condition in the
  kernel, this can still trigger.

Fixes golang/go#21014 and #165

This avoids involving the go 1.9 poller in I/O, which causes hangs, as
the poller does not understand the fcntl(O_NONBLOCK) tweak that we
executed on the pipe's file descriptors.

Fixes #164

The root user can issue the umount syscall and does not need the
fusermount binary. Therefore, it is not necessary to fork and exec a new
process just to run the umount binary.

Golang has a global lock held for forking (See exec_unix.go in the golang
source tree):
"Acquire the fork lock so that no other threads create new fds that are
not yet close-on-exec before we fork."

So it would be best if processes would refrain from forking
unnecessarily.
Signed-off-by: Shayan Pooya <shayan@arista.com>

This fixes 'du'.

Fixes #159.

Fixes #160.

This first race was easy to find and fix in the tests themselves.

It exposes another race in nodefs.FileSystemConnector where a node's
mount (and therefore, treeLock) is is not the same as the grand-parent
node's is.

Updates #138.

Fixes #134

Change-Id: I04eb8c16d45dbeef872036aab4091741b981505c

Fixes #127.

This enhances nodes which embed defaultNode (which is most of them),
allowing those implementations to omit a GetAttr() method. Without this,
when the kernel calls GetAttr() for an open file, FUSE will return a
size of zero, causing a read() of the file to always return 0, even
though fuse returns a nonzero number of bytes.

One can leverage this behavior to create a Linux procfs-like file, which
stat shows to be zero bytes, but can be successfully read.

N.B. This behavior is only useful if the FileSystemConnector Options
specify AttrTimeout of zero. Otherwise, the result of the first call to
GetAttr() (file == nil) is cached, and the kernel allows zero bytes to
be read. With this caching disabled, the next GetAttr() call (file !=
nil) will be used to control the number of readable bytes.