Prepare 3.17.1 release

Update ChangeLog.rst and AUTHORS

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

Change version to 3.17.1-rc0

Somehow really hard to set -rcX with meson.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Build fixes for -Og

Cleaning... 18 files.
[35/69] Compiling C object example/passthrough_ll.p/passthrough_ll.c.o
../example/passthrough_ll.c: In function ‘lo_opendir’:
../example/passthrough_ll.c:666:20: warning: ‘fd’ may be used uninitialized [-Wmaybe-uninitialized]
  666 |                 if (fd != -1)
      |                    ^
../example/passthrough_ll.c:637:13: note: ‘fd’ was declared here
  637 |         int fd;
      |             ^~
[38/69] Compiling C object test/test_syscalls.p/test_syscalls.c.o
../test/test_syscalls.c: In function ‘test_seekdir’:
../test/test_syscalls.c:804:16: warning: ‘de’ may be used uninitialized [-Wmaybe-uninitialized]
  804 |         while (de)
      |                ^~
../test/test_syscalls.c:776:24: note: ‘de’ was declared here
  776 |         struct dirent *de;
      |                        ^~

These are actually valid.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

fuse_new version fixes: Change to fuse_new_versioned

Another additon for https://github.com/libfuse/libfuse/issues/1092

Use _fuse_new_versioned() instead of _fuse_new_317 and actually
also remove symbol versioning for it - we don't need it.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

fuse_new: Fix non symboled call to _fuse_new_317

Acidentally the wrong non-existing function was ccalled.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Rename to fuse_session_new_versioned

Similar previous renames to fuse_main_real_versioned,
but here for the low level fuse_session_new.
Also remove symbol versioned as part of "fuse_session_new"
as that function is not part of the official API/ABI
and to allow easier access with dlopen/dlsym.
Also switch back to a macro fuse_session_new, just in
case some code has some expectations on that.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Make fuse_main a macro again and wrap that to fuse_main_fn

As suggested by Bill in Issue #1092 make fuse_main a macro again,
just in case some applications expect it to be a macro.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Rename fuse_main_real_317 to fuse_main_real_versioned

As suggested by Bill in Issue #1092, rename to _versioned
so that applications using dlopen/dlvsym better understand
the meaning of this function.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

build(deps): bump github/codeql-action from 3.28.6 to 3.28.8

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.28.6 to 3.28.8.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/17a820bf2e43b47be2c72b39cc905417bc1ab6d0...dd746615b3b9d728a6a37ca2045b68ca76d4841a)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

fuse_lowlevel.c: drop incorrect comment from convert_to_conn_want_ext()

The comment is probably a leftover from older versions, as the application
of the 32 bits of conn.capable_ext to conn.capable is done before function
convert_to_conn_want_ext() is invoked.

Signed-off-by: Luis Henriques <luis@igalia.com>

fuse_lowlevel.c: fix possible 64 bits value truncation

Because conn.want_ext is a uint64_t, copying it into a uint32_t may result
in truncating it's value.  This patch fixes a bug in do_init() where the
32 bits copy is again converted into a 64 bits value, because it will be
used in convert_to_conn_want_ext().

Signed-off-by: Luis Henriques <luis@igalia.com>

github actions: Add an include-what-you-need check

Probably lots of issues right now, so let's fix it step by step
by only checking modified files - new PRs should fix their
modified files.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

build(deps): bump github/codeql-action from 3.28.1 to 3.28.6

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.28.1 to 3.28.6.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/b6a472f63d85b9c78a3ac5e89422239fc15e9b3c...17a820bf2e43b47be2c72b39cc905417bc1ab6d0)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

lib: Set thread names

Main worker threads: fuse_worker
high level clean up threads: "fuse_prune_nodes"

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

abidiff: Don't warn about added symbols

So far we only want it to warn about changed symbols.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

checkpatch: Ignore ENOSYS_SYSCALL

This ENOSYS is the fuse protocol return code to tell
fuse client/kernel that an operation is not supported,
warning on it is not right.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

codespell: Add 're-used' and 're-using' to the ignore list

These seem to be fine and exist in the code and also seem to
be common English (there are debates which spelling is right).

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

build(deps): bump github/codeql-action from 3.28.0 to 3.28.1

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.28.0 to 3.28.1.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/48ab28a6f5dbc2a99bf1e0131198dd8f1df78169...b6a472f63d85b9c78a3ac5e89422239fc15e9b3c)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

Update doxygen/comments for fuse_reply_open/fuse_reply_create

Comments for fuse_reply_open and fuse_reply_create and with that
doxygen had not been updated for parallel_direct_writes and others.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

Identify the FD hold by parent process

Issue:
There is no directly way to get the FD
hold by parent process which asked do fuse mount.

Use Case:
For auto_unmount case, identify the FD can easy to close
the FD and make automatically unmount manually and explicitly.
The FD[1] can be got via getenv(FUSE_COMMFD2_ENV).
One potential use case is to satisfy FD-Leak checks.

Solution:
Add an extra env _FUSE_COMMFD2 to store the FD.
This will provide a easy way to get the FD via FUSE_COMMFD2_ENV.

Signed-off-by: Zegang Luo <zegang.luo@qq.com>

Make fuse_main_real() not symboled

Addresses https://github.com/libfuse/libfuse/issues/1092

We actually don't need to make fuse_main_real() symboled, as it
is not part of the official API.

The inlined function now always calls into fuse_main_real_317
and the compat ABI function (which should also be available
for dlopen/dlsym) is now always compiled, independent if the
compiler/linker support versioned symbols.

Additionally, fuse_main_real() is also declared as inlined
function and a warning message is created when that function
is called.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

checkpatch: Disable warning for externs in C files

We do actually need these, at least for compat.c.

Also disable git commit id warnings, these are doing
more harm than good (for example trigger long line
warnings when fulfilled).

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Avoid global declarion of internal functions that are new in 3.17

_fuse_new() is not supposed to be called by external users
outside of internal functions or static inlined functions.

This also removes several functions from lib/fuse_versionscript
which where added and exported by
commit 58f85bfa9b7d ("Add in the libfuse version a program...)
as these are libfuse internal only.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Fix fuse_main_real symbols

Commit 58f85bfa9b7d ("Add in the libfuse version a program...")
forgot to add a fuse_main_real function for libfuse compilations
that are not symboled. That is now added in compat.c.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Fix junk readdirplus results when filesystem not filling stat info

Commit dd95d13a ("fix readdirplus when filler is called with zero offset
(#896)) broke readdirplus with passthrough example command:

    passthrough -o auto_cache,modules=subdir,subdir=/src /mnt

The /src directory looks like this:

    ~# ls -l /src
    total 0
    drwx------ 3 root root 60 Jan  2 17:51 testdir

And the fuse directory looks like this:

    ~# ls -l /mnt
    total 0
    d--------- 0 root root 0 Jan  1  1970 testdir

Because readdir_fill_from_list() ignores the fact that filesystem
did not pass the FUSE_FILL_DIR_PLUS flag with valid stat info.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>

example/passthrough: Enable testing of readdirplus without fill offsets

passthrough example supports the --plus command line argument to reply
to readdirplus with fill_dir_plus and unspecified (0) fill offsets.

As explained in this comment:
https://github.com/libfuse/libfuse/pull/896#issuecomment-1978917041
passthrough example needs a few more changes to be able to test commit
dd95d13a ("fix readdirplus when filler is called with zero offset (#896))

With the changes in this commit, readdirplus without fill offsets
can be tested to verify the readdirplus fix above with command line:

passthrough --plus -o auto_cache,modules=subdir,subdir=/src /mnt

Signed-off-by: Amir Goldstein <amir73il@gmail.com>

Fix build of memfs_ll without manual meson reconfigure

After pulling latest code, memfs_ll build would fail because it
builds with C++11.

Changing the default of cpp_std in meson.build is not enough to fix this
problem even if user runs 'meson setup --reconfigure'.

I had to run 'meson setup -Dcpp_std= --reconfigure' to fix the build
as mentioned in this meson issue:
https://github.com/mesonbuild/meson/issues/8062#issuecomment-1568249672

Signed-off-by: Amir Goldstein <amir73il@gmail.com>

Fix libfuse build with FUSE_USE_VERSION 30

Signed-off-by: Amir Goldstein <amir73il@gmail.com>

Remove the DCO check

Actually checkpatch.pl already checks for "Signed-off-by"
and also handles merge commits - no need for duplicate
logic.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Fix the checkpatch.pl workflow

- The MAINTAINERS test is not valid for libfuse.
- Correct the base commit

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

.github/workflows/codespell.yml: checkpatch.pl needs to be skipped

checkpatch.pl has a list of mispelled words and the codespell test
fails on that.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Update AUTHORS for 3.17 and add script for that

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Add a checkpatch.pl github workflow

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Add a github action to check for "Signed-off-by"

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Run github workflow actions on release branches

Running on the 'master' is not enough, actions also
need to run on release branches.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Update ChangeLog.rst for 3.17 (#1085)

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

passthrough_ll: set correct keep_cache flag in opendir (#1083)

In cache=always mode, set keep_cache flag in opendir(), same as done
in open() and same as passthrough_hp does in opendir().

In the default cache=auto mode, use readdir cache, but do not set
keep_cache, same as regular files use page cache for an open file,
but do not keep_cache for a new open.

Note that passthrough_hp by default behaves the same as passthrough_ll
cache=always mode and supports the cache=never mode with --nocache, but
it does not support the equivalent of cache=auto mode.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>

Merge pull request #1081 from bsbernd/so-version-abi-compat

Increase the .so version, but also change back to previous ABI

Increase .so version to 4

We have multiple ABI breakages - increase the .so version.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Convert FUSE_CAP defines to enum fuse_capability

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Add 64-bit conn::{capable,want}_ext fields

The previous fields are left for ABI compatibility, although
it is not beautiful to add that complexity when we have to increase
the so-version as we had ABI breakage anyway.

example/printcap is simplified to use an array, as every line would
have needed to be modified anyway. Missing 'FUSE_CAP_PASSTHROUGH'
was added.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

fuse_lowlevel.c: define FUSE_MAXOP as CUSE_INIT

We don't want FUSE_MAXOP calculated at compilation time,
as the ABI limit is CUSE_INIT - better use that value
directly.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Use uint32_t for libfuse_version

Should be used internally only, but is still in a common file -
we better use arch independent values.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

struct fuse_file_info extension

Add a flags and reserved fields to struct fuse_file_info
and add a static assert on the size.

Also add another static assert for 'struct fuse_conn_info'

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Use single bit for struct fuse_file_info::no_interrupt

Now that we know the exact size of the integer we can
use a single bit and reserve the other bits.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Use (u)int32_t for struct fuse_conn_info

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Add padding3 in struct fuse_file_info

pahole was showing a 4 byte hole - add another padding to
fill that hole.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Use (u)int32_t for struct fuse_file_info

Enforce 32 bit integers for #struct fuse_file_info'.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Use (u)int32_t for struct fuse_config and extend the struct

Enforce 32-bit integers for struct fuse_config and
add future ABI fields to that struct.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Restore compat ABI in 'struct fuse_config'

This fixes commit
dad15aee26835 "Add no_rofd_flush mount option"
and restores the ABI by moving the no_rofd_flush field.

Given that there were already several releases with this commit
we will still need an so version bump.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

Restore compat ABI in 'struct fuse_file_info'

This fixes commit
a5eb7f2 "Enable parallel direct writes on the same file"
and restores the ABI by moving the parallel_direct_writes
bit.

Given that there were already several releases with this commit
we will still need an so version bump.

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>

build(deps): bump github/codeql-action from 3.27.9 to 3.28.0

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.27.9 to 3.28.0.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/df409f7d9260372bd5f19e5b04e83cb3c43714ae...48ab28a6f5dbc2a99bf1e0131198dd8f1df78169)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

Rename _int to _internal

_int can be confused with 'integer'

Allow to have page aligned writes

Read/writes IOs should be page aligned as fuse server
might need to copy data to another buffer otherwise in
order to fulfill network or device storage requirements.

Simple reproducer is example/passthrough*
and opening a file with O_DIRECT - without this change
writing to that file failed with -EINVAL if the underlying
file system was using ext4 (for passthrough_hp the
'passthrough' feature has to be disabled).

The mis-alignment from fuse kernel is not ideal, but we can handle
it by allocation one page more than needed and then using a buffer
that is set up to compensate for kernel misalignment.

This also only set se->buf_reallocable to true when called
by a libfuse internal caller - we do not know what
external callers are doing with the buffer - update to
commit 0e0f43b79b9b

tests: Add debug messages to some tests and umount

On my nfs mount the tests were hanging and it was impossible
to diagnoze what is actually the issue.

Also get rid of 'looseversion' python package dependency,
as that package is not in ubuntu - add a handcode kernel
version parser.

build(deps): bump github/codeql-action from 3.27.5 to 3.27.9

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.27.5 to 3.27.9.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/f09c1c0a94de965c15400f5634aa42fac8fb8f88...df409f7d9260372bd5f19e5b04e83cb3c43714ae)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

examples: Add memfs_ll.cc

This is an initial implementation and quickly hacked together
within a few hours - issues expected.

This also increase to C++17 as memfs_ll makes use of more
recent features.

Background to create this was actually to be able to test large
file names (3 * 1024B), which couldn't be achieved with passthrough
file system as non of the underlying file systems seems to support
that.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

support FUSE_TMPFILE in the low level API

Note that name hashes and using paths as parameters
makes it very hard to support
anonymous files in the high level API.

Known Issues:
- tests have to bail out when O_TMPFILE is not supported.
This will always be the case with high level passthrough implementations.
- test_create_and_link_tmpfile has to be skipped
due to unidentified problems with github runner

build(deps): bump github/codeql-action from 3.27.4 to 3.27.5

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.27.4 to 3.27.5.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/ea9e4e37992a54ee68a9622e985e60c8e8f12d9f...f09c1c0a94de965c15400f5634aa42fac8fb8f88)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

Fix alignment of allocation in fuse_reply_create

When allocating as an array of char only alignment of 1 is guaranteed but
the structure needs an alignment of 8. Specify alignment explicitly

Reallocate fuse_session buffer transparently for extended max writes

A previous PR supported extended max writes (eg write requests larger than 1 MB)
by initializing the fuse session buffer size to use the max_pages_limit set in
/proc/sys/fs/fuse. However, this is a huge problem for machines where multiple
fuse servers may be running but only one server needs large writes. In this case,
a lot of memory will be wasted and will lead to OOM issues.

This PR does a reallocation of the session buffer transparently if the server set
 "se->conn.max_write" to a value larger than 1 MiB. This is only for buffers that
are "owned" by libfuse - if the server wishes to provide its own allocated buffer
for receiving/processing requests, then it should ensure that buffer is allocated
to the proper size from the start.

Local testing showed:
echo 65535 | sudo tee /proc/sys/fs/fuse/max_pages_limit
dd if=/dev/urandom of=hello_file bs=6M count=2

write requests:
write request size is 5242880
write request size is 1048576
write request size is 5242880
write request size is 1048576

build(deps): bump github/codeql-action from 3.27.1 to 3.27.4

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.27.1 to 3.27.4.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/4f3212b61783c3c68e8309a0f18a699764811cda...ea9e4e37992a54ee68a9622e985e60c8e8f12d9f)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

Use fuse_session_{receive/process}_buf() instead of _int version

No functional changes.

Add FUSE_CAP_NO_EXPORT and use it in passthrough_hp

This should stop some more xfstest test failures.

Add fuse_set_feature_flag() / fuse_unset_feature_flag

Simplify setting feature flags a bit by adding a helper
function.

Also move the check for valid flags into a funtion we can re-use
in a later patch.

build(deps): bump github/codeql-action from 3.27.0 to 3.27.1

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.27.0 to 3.27.1.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/662472033e021d55d94146f66f6058822b0b39fd...4f3212b61783c3c68e8309a0f18a699764811cda)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

Correct arg documentation for FUSE_READLINK

It does not have an in-argument.

build(deps): bump github/codeql-action from 3.26.13 to 3.27.0

Bumps [github/codeql-action](https://github.com/github/codeql-action) from 3.26.13 to 3.27.0.
- [Release notes](https://github.com/github/codeql-action/releases)
- [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md)
- [Commits](https://github.com/github/codeql-action/compare/f779452ac5af1c261dce0346a8f964149f49322b...662472033e021d55d94146f66f6058822b0b39fd)

---
updated-dependencies:
- dependency-name: github/codeql-action
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

build(deps): bump actions/checkout from 4.2.1 to 4.2.2

Bumps [actions/checkout](https://github.com/actions/checkout) from 4.2.1 to 4.2.2.
- [Release notes](https://github.com/actions/checkout/releases)
- [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md)
- [Commits](https://github.com/actions/checkout/compare/eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871...11bd71901bbe5b1630ceea73d27597364c9af683)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

codeql workflow: Switch from pip to apt

On some runs:

Run pip install --break-system-packages -r requirements.txt
....
no such option: --break-system-packages
Error: Process completed with exit code 2.

On other runs it refuses to install and asks for that option
as it refuses to override system packages.

Also require ubuntu-latest only, as MacOS is not supported at all
by libfuse.

Update libfuse-operations.txt for FUSE_SYMLINK

Create codeql.yml

fuse_buf_size fix for SIZE_MAX

Add libfuse util strtol wrapper

Add a wrapper around strtol for more rigorous error checking
and convert uses of atoi and strtol to use this instead.

build(deps): bump actions/checkout from 4.2.0 to 4.2.1

Bumps [actions/checkout](https://github.com/actions/checkout) from 4.2.0 to 4.2.1.
- [Release notes](https://github.com/actions/checkout/releases)
- [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md)
- [Commits](https://github.com/actions/checkout/compare/d632683dd7b4114ad314bca15554477dd762a938...eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>

ignore mtab on android

Updating the mtab on Android fails due to differences in toybox's mount
command. Setting IGNORE_MTAB avoids that issue.

Signed-off-by: Daniel Rosenberg <drosen@google.com>

build(deps): bump actions/checkout from 4.1.7 to 4.2.0

Bumps [actions/checkout](https://github.com/actions/checkout) from 4.1.7 to 4.2.0.
- [Release notes](https://github.com/actions/checkout/releases)
- [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md)
- [Commits](https://github.com/actions/checkout/compare/692973e3d937129bcbf40652eb9f2f61becf3332...d632683dd7b4114ad314bca15554477dd762a938)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

fuse_lowlevel FUSE_INIT: Simplify the max_write/bufsize logic

max_write can be limited by se->op.init() and by the buffer size,
we use the minimum of these two.
Required se->bufsize is then set according to the determined
max_write. The current thread will have the old buffer size,
though, as it already had to the allocation to handle the
FUSE_INIT call (unless splice is used and ths variable
and related buffer is not used at all).

The given bufsize is just a hint for minimum size, allocation
could be actually larger (for example to get huge pages).

fuse_lowlevel FUSE_INIT: group bufsize and max_write adjustments

The further these lines are separated from each other the harder
it is to read the code.

There shouldn't be any code change behavior here.

Change FUSE_MAX_MAX_PAGES to FUSE_DEFAULT_MAX_PAGES_LIMIT

A recent upstream patch [1] changed FUSE_MAX_MAX_PAGES to
FUSE_DEFAULT_MAX_PAGES_LIMIT.

Update libfuse to use FUSE_DEFAULT_MAX_PAGES_LIMIT as well
instead of FUSE_MAX_MAX_PAGES.

[1] https://lore.kernel.org/linux-fsdevel/20240923171311.1561917-1-joannelkoong@gmail.com/T/#t

Initialize session buffer size to value set by sysctl

Currently in libfuse, the buffer size for a fuse session is
capped at 1 MiB on a 4k page system. A recent patch
upstream [1] was merged that allows the max number of pages
per fuse request to be dynamically configurable through the
/proc/sys interface (/proc/sys/fs/fuse/max_pages_limit).

This commit adds support for this on the libfuse side to set
the fuse session buffer to take into account the max pages
limit set in /proc/sys/fs/fuse/max_pages_limit. If this
sysctl does not exist (eg older kernels), it will default to
old behavior (using FUSE_MAX_MAX_PAGES (256) as the max pages
limit). This allows for things like bigger write buffers per
request.

[1] https://lore.kernel.org/linux-fsdevel/20240923171311.1561917-1-joannelkoong@gmail.com/T/#t

ABI check: Use base commit / branch of a PR for ABI comparison

The base commit is needed for comparison when a PR has multiple
commits.

Libfuse check ABI differences with 'abidiff'

add version check for the fuse_loop_cfg* operations

fuse_loop_cfg* operations are introduced in fuse version 312 and
can not be used in an early version.
Also fix some typo.

passthrough_hp: Use fuse_loop_cfg_set_max_threads()

fuse_loop_cfg_set_idle_threads() was by accident and
setting it might cause a performance issue.

fusermount: Close file descriptors with close_range() if possible

close_range() is much more efficient.

Also remove the lower limit of 3 and set it to 0, as 0 to 1
might have been closed by the application and might be valid.

fusermount: close inherited fds

When using the auto_unmount option, the fusermount3
watchdog process retains any inherited file descriptors.
This PR ensures they are closed.

Reason is that FDs that are kept open for a long time might
cause issues for applications using libfuse, for example
if these expect a pipe close, but the pipe is kept open
through the inherited file descriptor.
See for example: https://github.com/cvmfs/cvmfs/issues/3645

Signed-off-by: MJ Harvey <mharvey@jumptrading.com>
Signed-off-by: Bernd Schubert <bschubert@ddn.com>

example/passthrough_hp: Set keep_cache and noflush in sfs_create

These flags should be also set for O_CREAT

example/passthrough_hp: No auto FOPEN_DIRECT_IO in passthrough mode

sfs_open and sfs_create set fi->direct_io (FOPEN_DIRECT_IO) when
O_DIRECT is given, in order to benefit from a shared inode lock
in kernel, i.e. to get parallel DIO writes. However, kernel side
disabled passthrough when FOPEN_DIRECT_IO is set. Reads/writes
had been totally failing in this case for O_DIRECT as
sfs_write_buf() and sfs_read() have a sanity check. That sanity
check could be modified, but for performance passthrough is
better than parallel DIO, hence, we only want automatic
FOPEN_DIRECT_IO for O_DIRECT when passthrough is not enabled.

Fixes: https://github.com/libfuse/libfuse/issues/1027
This also fixes automatically switching to FOPEN_DIRECT_IO
for O_DIRECT in sfs_create().

getattr: Make use of FUSE_GETATTR_FH in lowlevel examples

High level examples were already using it, but not
lowlevel. Also update the documentation.

Bump codespell-project/actions-codespell from 2.0 to 2.1

Bumps [codespell-project/actions-codespell](https://github.com/codespell-project/actions-codespell) from 2.0 to 2.1.
- [Release notes](https://github.com/codespell-project/actions-codespell/releases)
- [Commits](https://github.com/codespell-project/actions-codespell/compare/94259cd8be02ad2903ba34a22d9c13de21a74461...406322ec52dd7b488e48c1c4b82e2a8b3a1bf630)

---
updated-dependencies:
- dependency-name: codespell-project/actions-codespell
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

remove unused headers in example/poll_client

fuse_config.h is used in libfuse internally. This may confuse developers
that fuse_config.h is used in the examples.

Rearrange util/fusermount.c umount_fuse_locked() so that
umount2 is called with privs dropped, not raised.  This
works around a clash with NFS permissions: if FUSE mounted
on NFS client directory with root_squash in effect, and
some directory in the path leading to the mount point denies
permissions to others, umount2 will fail because userid 0
cannot search it.  Since drop_privs merely sets the file-
system user- and group-ID without changing the CAP_SYS_ADMIN
capability needed to unmount a file system (which fusermount
has because it is set-user-ID root), umount2 works fine.

Fix FUSE_USE_VERSION in example/notify_store_retrieve.c

This is an addition to commit e75d2c54a347. This example sets
FUSE_USE_VERSION = 34 but uses fuse_loop_cfg_* APIs, which is
not allowed since these APIs are not introduced in version 34.

doc/libfuse-operations: Fix FUSE_STATX in_args[1] description

Fix program-path in util/parse-backtrace and dump_stack function

The option to the path of the binary had been accidentally removed
in the scripts that can parse backtraces.
The dump_stack() function had left over debug messages.

example/passthrough_hp: Remove unused includes

Fix some clang-tidy warnings.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

Fix FUSE_USE_VERSION in example/

This is an addition to commit a8f1ae35af66, which
introduced the 312 API, but didn't set the right
API version in all examples.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

synchronize fuse_kernel.h with linux-6.10

Signed-off-by: Bernd Schubert <bschubert@ddn.com>

example/passthrough_hp: Fix . and .. readdir lookup count

Commit 170edc6a8ef0 added dot and dotdot (. and ..) to readdir
results, but introduced an issue when max number of entries
was reached - lookup count must not be decreased without
doing the lookup.
With ext4 as underlying file system readir seems to return . and ..
at random offsets and randomly failed xfstests for me.

This also fixes indentation, as passthrough_hp.cc does not follow
the linux indentation style (if we decide to fix this, it needs
to be done for the entire file).

Signed-off-by: Bernd Schubert <bschubert@ddn.com>