[PATCH v3 00/12] man2: document "new" mount API

[PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

Back in 2019, the new mount API was merged[1]. David Howells then set
about writing man pages for these new APIs, and sent some patches back
in 2020[2].

Unfortunately, these patches were never merged, which meant that these
APIs were practically undocumented for many years -- arguably this has
been a contributing factor to the relatively slow adoption of these new
(far better) APIs. For instance, I have often discovered that many folks
are unaware of the read(2)-based message retrieval interface provided by
filesystem context file descriptors.

In 2024, Christian Brauner adapted David Howell's original man pages
into the easier-to-edit Markdown format and published them on GitHub[3].
These have been maintained since, including updated information on new
features added since David Howells's 2020 draft pages (such as
MOVE_MOUNT_BENEATH).

While this was a welcome improvement to the previous status quo (that
had lasted over 6 years), speaking personally my experience is that not
having access to these man pages from the terminal has been a fairly
common painpoint.

So, this is a modern version of the man pages for these APIs, in the
hopes that we can finally (6 years later) get proper documentation for
these APIs in the man-pages project.

One important thing to note is that most of these were re-written by me,
with very minimal copying from the versions available from Christian[2].
The reasons for this are two-fold:

 * Both Howells's original version and Christian's maintained versions
   contain crucial mistakes that I have been bitten by in the past (the
   most obvious being that all of these APIs were merged in Linux 5.2,
   but the man pages all claim they were merged in different versions.)

 * As the man pages appear to have been written from Howells's
   perspective while implementing them, some of the wording is a little
   too tied to the implementation (or appears to describe features that
   don't really exist in the merged versions of these APIs).

 * The original versions of the man-pages lacked bigger-picture
   explanations of the reasoning behind the API, which would make it
   easier for readers to understand what operations are doing.

I decided that the best way to resolve these issues is to rewrite them
from the perspective of an actual user of these APIs (me), and check
that we do not repeat the mistakes I found in the originals. I have also
done my best to resolve the issues raised by Michael Kerrisk on the
original patchset sent by Howells[1].

In addition, I have also included a man page for open_tree_attr(2) (as a
subsection of the new open_tree(2) man page), which was merged in Linux
6.15.

[1]: https://lore.kernel.org/all/20190507204921.GL23075@ZenIV.linux.org.uk/
[2]: https://lore.kernel.org/linux-man/159680892602.29015.6551860260436544999.stgit@warthog.procyon.org.uk/
[3]: https://github.com/brauner/man-pages-md

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
Changes in v3:
- `sed -i s|Co-developed-by|Co-authored-by|g`. [Alejandro Colomar]
  - Add Signed-off-by for co-authors. [Christian Brauner]
- `sed -i s|needs-mount|awaiting-mount|g`, to match the kernel parlance.
- Fix VERSIONS/HISTORY mixup in mount_attr(2type) that was copied from
  open_how(2type). [Alejandro Colomar]
- Fix incorrect .BR usage in SYNOPSIS.
- Some more semantic newlines fixes. [Alejandro Colomar]
- Minor fixes suggested by Alejandro. [Alejandro Colomar]
- open_tree_attr(2): heavily reword everything to be better formatted
  and more explicit about its behaviour.
- open_tree(2): write proper explanatory paragraphs for the EXAMPLES.
- mount_setattr(2): fix stray doublequote in SYNOPSIS. [Askar Safin]
- fsopen(2): rework structure of the DESCRIPTION introduction.
- fsopen(2): explicitly say that read(2) errors in the message retrieval
  interface are actual errors, not return 0. [Askar Safin]
- fsopen(2): add BUGS section to describe the unfortunate -ENODATA
  message dropping behaviour that should be fixed by
  <https://lore.kernel.org/r/20250807-fscontext-log-cleanups-v3-0-8d91d6242dc3@cyphar.com/>.
- fsconfig(2): add a NOTES subsection about generic filesystem
  parameters.
- fsconfig(2): add comment about the weirdness surrounding
  FSCONFIG_SET_PATH.
- {fspick,open_tree}(2): Correct AT_NO_AUTOMOUNT description (copied
  from David, who probably copied it from statx(2)) -- AT_NO_AUTOMOUNT
  applies to all path components, not just the final one. [Christian
  Brauner]
- statx(2): fix AT_NO_AUTOMOUNT documentation.
- open_tree(2): swap open(2) reference for openat(2) when saying that
  the result is identical. [Askar Safin]
- fsmount(2): fix DESCRIPTION introduction, and rework attr_flags
  description to better reference mount_setattr(2).
- {fsopen,fspick,fsmount,open_tree}(2): don't use "attach" when talking
  about the file descriptors we return that reference in-kernel objects,
  to avoid confusing readers with mount object attachment status.
- fsconfig(2): remove pidns argument example, as it was kind of unclear
  and referenced kernel features not yet merged.
- fsconfig(2): remove rambling FSCONFIG_SET_PATH_EMPTY text (which
  mostly describes an academic issue that doesn't apply to any existing
  filesystem), and instead add a CAVEATS section which touches on the
  weird type behaviour of fsconfig(2).
- v2: <https://lore.kernel.org/r/20250807-new-mount-api-v2-0-558a27b8068c@cyphar.com>

Changes in v2:
- `make -R lint-man`. [Alejandro Colomar]
- `sed -i s|Glibc|glibc|g`. [Alejandro Colomar]
- `sed -i s|pathname|path|g` [Alejandro Colomar]
- Clean up macro usage, example code, and synopsis. [Alejandro Colomar]
- Try to use semantic newlines. [Alejandro Colomar]
- Make sure the usage of "filesystem context", "filesystem instance",
  and "mount object" are consistent. [Askar Safin]
- Avoid referring to these syscalls without an "at" suffix as "*at()
  syscalls". [Askar Safin]
- Use \% to avoid hyphenation of constants. [Askar Safin, G. Branden Robinson]
- Add a new subsection to mount_setattr(2) to describe the distinction
  between mount attributes and filesystem parameters.
- (Under protest) double-space-after-period formatted commit messages.
- v1: <https://lore.kernel.org/r/20250806-new-mount-api-v1-0-8678f56c6ee0@cyphar.com>

---
Aleksa Sarai (12):
      man/man2/statx.2: correctly document AT_NO_AUTOMOUNT
      man/man2/mount_setattr.2: fix stray quote in SYNOPSIS
      man/man2/mount_setattr.2: move mount_attr struct to mount_attr(2type)
      man/man2/fsopen.2: document "new" mount API
      man/man2/fspick.2: document "new" mount API
      man/man2/fsconfig.2: document "new" mount API
      man/man2/fsmount.2: document "new" mount API
      man/man2/move_mount.2: document "new" mount API
      man/man2/open_tree.2: document "new" mount API
      man/man2/mount_setattr.2: mirror opening sentence from fsopen(2)
      man/man2/open_tree{,_attr}.2: document new open_tree_attr() API
      man/man2/{fsconfig,mount_setattr}.2: add note about attribute-parameter distinction

 man/man2/fsconfig.2           | 681 ++++++++++++++++++++++++++++++++++++++++++
 man/man2/fsmount.2            | 220 ++++++++++++++
 man/man2/fsopen.2             | 384 ++++++++++++++++++++++++
 man/man2/fspick.2             | 309 +++++++++++++++++++
 man/man2/mount_setattr.2      |  62 +++-
 man/man2/move_mount.2         | 640 +++++++++++++++++++++++++++++++++++++++
 man/man2/open_tree.2          | 593 ++++++++++++++++++++++++++++++++++++
 man/man2/open_tree_attr.2     |   1 +
 man/man2/statx.2              |   6 +-
 man/man2type/mount_attr.2type |  61 ++++
 10 files changed, 2940 insertions(+), 17 deletions(-)
---
base-commit: e473affca7b039fd018eedb839d6c80e4fd3df17
change-id: 20250802-new-mount-api-436db984f432


Kind regards,
-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[PATCH v3 01/12] man/man2/statx.2: correctly document AT_NO_AUTOMOUNT

[Aleksa Sarai]

AT_NO_AUTOMOUNT un-sets FOLLOW_AUTOMOUNT, which blocks the automounting
of all automount points encountered during lookup, not just the terminal
component.

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/statx.2 | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/man/man2/statx.2 b/man/man2/statx.2
index 07ac60b3c5d61d919fa790fe2d5c2ba33a771f75..0b4175e994f42c7aab6b0bfd50739971d4d55a4f 100644
--- a/man/man2/statx.2
+++ b/man/man2/statx.2
@@ -184,9 +184,9 @@ .SS Invoking statx():
 the call operates on the current working directory.
 .TP
 .B AT_NO_AUTOMOUNT
-Don't automount the terminal ("basename") component of
-.I path
-if it is a directory that is an automount point.
+Don't automount any automount points encountered
+while resolving
+.IR path .
 This allows the caller to gather attributes of an automount point
 (rather than the location it would mount).
 This flag has no effect if the mount point has already been mounted over.

-- 
2.50.1

[PATCH v3 02/12] man/man2/mount_setattr.2: fix stray quote in SYNOPSIS

[Aleksa Sarai]

Fixes: eb0f8239bc35 ("man/man2/mount_setattr.2: Document glibc >= 2.36 syscall wrappers")
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/mount_setattr.2 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/man/man2/mount_setattr.2 b/man/man2/mount_setattr.2
index c96f0657f0468fc4d2bc0132e08d1328570073b1..e1a975dcc8e2b263f68d18dc0492e8ecc518459e 100644
--- a/man/man2/mount_setattr.2
+++ b/man/man2/mount_setattr.2
@@ -14,7 +14,7 @@ .SH SYNOPSIS
 .B #include <sys/mount.h>
 .P
 .BI "int mount_setattr(int " dirfd ", const char *" path ", unsigned int " flags ","
-.BI "                  struct mount_attr *" attr ", size_t " size );"
+.BI "                  struct mount_attr *" attr ", size_t " size ");"
 .fi
 .SH DESCRIPTION
 The

-- 
2.50.1

[PATCH v3 03/12] man/man2/mount_setattr.2: move mount_attr struct to mount_attr(2type)

[Aleksa Sarai]

As with open_how(2type), it makes sense to move this to a separate man
page.  In addition, future man pages added in this patchset will want to
reference mount_attr(2type).

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/mount_setattr.2      | 17 ++++--------
 man/man2type/mount_attr.2type | 61 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 66 insertions(+), 12 deletions(-)

diff --git a/man/man2/mount_setattr.2 b/man/man2/mount_setattr.2
index e1a975dcc8e2b263f68d18dc0492e8ecc518459e..46fcba927dd8c0959c898b9ba790ae298f514398 100644
--- a/man/man2/mount_setattr.2
+++ b/man/man2/mount_setattr.2
@@ -114,18 +114,11 @@ .SH DESCRIPTION
 .I attr
 argument of
 .BR mount_setattr ()
-is a structure of the following form:
-.P
-.in +4n
-.EX
-struct mount_attr {
-    __u64 attr_set;     /* Mount properties to set */
-    __u64 attr_clr;     /* Mount properties to clear */
-    __u64 propagation;  /* Mount propagation type */
-    __u64 userns_fd;    /* User namespace file descriptor */
-};
-.EE
-.in
+is a pointer to a
+.I mount_attr
+structure,
+described in
+.BR mount_attr (2type).
 .P
 The
 .I attr_set
diff --git a/man/man2type/mount_attr.2type b/man/man2type/mount_attr.2type
new file mode 100644
index 0000000000000000000000000000000000000000..f5c4f48be46ec1e6c0d3a211b6724a1e95311a41
--- /dev/null
+++ b/man/man2type/mount_attr.2type
@@ -0,0 +1,61 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH mount_attr 2type (date) "Linux man-pages (unreleased)"
+.SH NAME
+mount_attr \- what mount properties to set and clear
+.SH LIBRARY
+Linux kernel headers
+.SH SYNOPSIS
+.EX
+.B #include <sys/mount.h>
+.P
+.B struct mount_attr {
+.BR "    u64 attr_set;" "     /* Mount properties to set */"
+.BR "    u64 attr_clr;" "     /* Mount properties to clear */"
+.BR "    u64 propagation;" "  /* Mount propagation type */"
+.BR "    u64 userns_fd;" "    /* User namespace file descriptor */"
+    /* ... */
+.B };
+.EE
+.SH DESCRIPTION
+Specifies which mount properties should be changed with
+.BR mount_setattr (2).
+.P
+The fields are as follows:
+.TP
+.I .attr_set
+This field specifies which
+.BI MOUNT_ATTR_ *
+attribute flags to set.
+.TP
+.I .attr_clr
+This field specifies which
+.BI MOUNT_ATTR_ *
+attribute flags to clear.
+.TP
+.I .propagation
+This field specifies what mount propagation will be applied.
+The valid values of this field are the same propagation types described in
+.BR mount_namespaces (7).
+.TP
+.I .userns_fd
+This field specifies a file descriptor that indicates which user namespace to
+use as a reference for ID-mapped mounts with
+.BR MOUNT_ATTR_IDMAP .
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.12.
+.\" commit 2a1867219c7b27f928e2545782b86daaf9ad50bd
+glibc 2.36.
+.P
+Extra fields may be appended to the structure,
+with a zero value in a new field resulting in
+the kernel behaving as though that extension field was not present.
+Therefore, a user
+.I must
+zero-fill this structure on initialization.
+.SH SEE ALSO
+.BR mount_setattr (2)

-- 
2.50.1

[PATCH v3 04/12] man/man2/fsopen.2: document "new" mount API

[Aleksa Sarai]

This is loosely based on the original documentation written by David
Howells and later maintained by Christian Brauner, but has been
rewritten to be more from a user perspective (as well as fixing a few
critical mistakes).

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/fsopen.2 | 384 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 384 insertions(+)

diff --git a/man/man2/fsopen.2 b/man/man2/fsopen.2
new file mode 100644
index 0000000000000000000000000000000000000000..cce677f316c67de72c359f94a6b415d851a761d6
--- /dev/null
+++ b/man/man2/fsopen.2
@@ -0,0 +1,384 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH fsopen 2 (date) "Linux man-pages (unreleased)"
+.SH NAME
+fsopen \- create a new filesystem context
+.SH LIBRARY
+Standard C library
+.RI ( libc ,\~ \-lc )
+.SH SYNOPSIS
+.nf
+.B #include <sys/mount.h>
+.P
+.BI "int fsopen(const char *" fsname ", unsigned int " flags ");"
+.fi
+.SH DESCRIPTION
+The
+.BR fsopen ()
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.P
+.BR fsopen ()
+creates a blank filesystem configuration context within the kernel
+for the filesystem named by
+.I fsname
+and places it into creation mode.
+A new file descriptor
+associated with the filesystem configuration context
+is then returned.
+The calling process must have the
+.B \%CAP_SYS_ADMIN
+capability in order to create a new filesystem configuration context.
+.P
+A filesystem configuration context is
+an in-kernel representation of a pending transaction,
+containing a set of configuration parameters that are to be applied
+when creating a new instance of a filesystem
+(or modifying the configuration of an existing filesystem instance,
+such as when using
+.BR fspick (2)).
+.P
+After obtaining a filesystem configuration context with
+.BR fsopen (),
+the general workflow for operating on the context looks like the following:
+.IP (1) 5
+Pass the filesystem context file descriptor to
+.BR fsconfig (2)
+to specify any desired filesystem parameters.
+This may be done as many times as necessary.
+.IP (2)
+Pass the same filesystem context file descriptor to
+.BR fsconfig (2)
+with
+.B \%FSCONFIG_CMD_CREATE
+to create an instance of the configured filesystem.
+.IP (3)
+Pass the same filesystem context file descriptor to
+.BR fsmount (2)
+to create a new detached mount object for
+the root of the filesystem instance,
+which is then attached to a new file descriptor.
+(This also places the filesystem context file descriptor into
+reconfiguration mode,
+similar to the mode produced by
+.BR fspick (2).)
+Once a mount object has been created with
+.BR fsmount (2),
+the filesystem context file descriptor can be safely closed.
+.IP (4)
+Now that a mount object has been created,
+you may
+.RS
+.IP (4.1) 7
+use the detached mount object file descriptor as a
+.I dirfd
+argument to "*at()" system calls; and/or
+.IP (4.2) 7
+attach the mount object to a mount point
+by passing the mount object file descriptor to
+.BR move_mount (2).
+This will also prevent the mount object from
+being unmounted and destroyed when
+the mount object file descriptor is closed.
+.RE
+.IP
+The mount object file descriptor will
+remain associated with the mount object
+even after doing the above operations,
+so you may repeatedly use the mount object file descriptor with
+.BR move_mount (2)
+and/or "*at()" system calls
+as many times as necessary.
+.P
+A filesystem context will move between different modes
+throughout its lifecycle
+(such as the creation phase
+when created with
+.BR fsopen (),
+the reconfiguration phase
+when an existing filesystem instance is selected with
+.BR fspick (2),
+and the intermediate "awaiting-mount" phase
+.\" FS_CONTEXT_AWAITING_MOUNT is the term the kernel uses for this.
+between
+.BR \%FSCONFIG_CMD_CREATE
+and
+.BR fsmount (2)),
+which has an impact on
+what operations are permitted on the filesystem context.
+.P
+The file descriptor returned by
+.BR fsopen ()
+also acts as a channel for filesystem drivers to
+provide more comprehensive diagnostic information
+than is normally provided through the standard
+.BR errno (3)
+interface for system calls.
+If an error occurs at any time during the workflow mentioned above,
+calling
+.BR read (2)
+on the filesystem context file descriptor
+will retrieve any ancillary information about the encountered errors.
+(See the "Message retrieval interface" section
+for more details on the message format.)
+.P
+.I flags
+can be used to control aspects of
+the creation of the filesystem configuration context file descriptor.
+A value for
+.I flags
+is constructed by bitwise ORing
+zero or more of the following constants:
+.RS
+.TP
+.B FSOPEN_CLOEXEC
+Set the close-on-exec
+.RB ( FD_CLOEXEC )
+flag on the new file descriptor.
+See the description of the
+.B O_CLOEXEC
+flag in
+.BR open (2)
+for reasons why this may be useful.
+.RE
+.P
+A list of filesystems supported by the running kernel
+(and thus a list of valid values for
+.IR fsname )
+can be obtained from
+.IR /proc/filesystems .
+(See also
+.BR proc_filesystems (5).)
+.SS Message retrieval interface
+When doing operations on a filesystem configuration context,
+the filesystem driver may choose to provide
+ancillary information to userspace
+in the form of message strings.
+.P
+The filesystem context file descriptors returned by
+.BR fsopen ()
+and
+.BR fspick (2)
+may be queried for message strings at any time by calling
+.BR read (2)
+on the file descriptor.
+Each call to
+.BR read (2)
+will return a single message,
+prefixed to indicate its class:
+.RS
+.TP
+\fBe\fP <\fImessage\fP>
+An error message was logged.
+This is usually associated with an error being returned
+from the corresponding system call which triggered this message.
+.TP
+\fBw\fP <\fImessage\fP>
+A warning message was logged.
+.TP
+\fBi\fP <\fImessage\fP>
+An informational message was logged.
+.RE
+.P
+Messages are removed from the queue as they are read.
+Note that the message queue has limited depth,
+so it is possible for messages to get lost.
+If there are no messages in the message queue,
+.B read(2)
+will return \-1 and
+.I errno
+will be set to
+.BR \%ENODATA .
+If the
+.I buf
+argument to
+.BR read (2)
+is not large enough to contain the entire message,
+.BR read (2)
+will return \-1 and
+.I errno
+will be set to
+.BR \%EMSGSIZE .
+(See BUGS.)
+.P
+If there are multiple filesystem contexts
+referencing the same filesystem instance
+(such as if you call
+.BR fspick (2)
+multiple times for the same mount),
+each one gets its own independent message queue.
+This does not apply to multiple file descriptors that are
+tied to the same underlying open file description
+(such as those created with
+.BR dup (2)).
+.P
+Message strings will usually be prefixed by
+the name of the filesystem or kernel subsystem
+that logged the message,
+though this may not always be the case.
+See the Linux kernel source code for details.
+.SH RETURN VALUE
+On success, a new file descriptor is returned.
+On error, \-1 is returned, and
+.I errno
+is set to indicate the error.
+.SH ERRORS
+.TP
+.B EFAULT
+.I fsname
+is NULL
+or a pointer to a location
+outside the calling process's accessible address space.
+.TP
+.B EINVAL
+.I flags
+had an invalid flag set.
+.TP
+.B EMFILE
+The calling process has too many open files to create more.
+.TP
+.B ENFILE
+The system has too many open files to create more.
+.TP
+.B ENODEV
+The filesystem named by
+.I fsname
+is not supported by the kernel.
+.TP
+.B ENOMEM
+The kernel could not allocate sufficient memory to complete the operation.
+.TP
+.B EPERM
+The calling process does not have the required
+.B \%CAP_SYS_ADMIN
+capability.
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.2.
+.\" commit 24dcb3d90a1f67fe08c68a004af37df059d74005
+.\" commit 400913252d09f9cfb8cce33daee43167921fc343
+glibc 2.36.
+.SH BUGS
+.SS Message retrieval interface and \fB\%EMSGSIZE\fP
+As described in the "Message retrieval interface" subsection above,
+calling
+.BR read (2)
+with too small a buffer to contain
+the next pending message in the message queue
+for the filesystem configuration context
+will cause
+.BR read (2)
+to return \-1 and set
+.BR errno (3)
+to
+.BR \%EMSGSIZE .
+.P
+However,
+this failed operation still
+consumes the message from the message queue.
+This effectively discards the message silently,
+as no data is copied into the
+.BR read (2)
+buffer.
+.P
+Programs should take care to ensure that
+their buffers are sufficiently large
+to contain any reasonable message string,
+in order to avoid silently losing valuable diagnostic information.
+.\" Aleksa Sarai
+.\"   This unfortunate behaviour has existed since this feature was merged, but
+.\"   I have sent a patchset which will finally fix it.
+.\"   <https://lore.kernel.org/r/20250807-fscontext-log-cleanups-v3-1-8d91d6242dc3@cyphar.com/>
+.SH EXAMPLES
+To illustrate the workflow for creating a new mount,
+the following is an example of how to mount an
+.BR ext4 (5)
+filesystem stored on
+.I /dev/sdb1
+onto
+.IR /mnt .
+.P
+.in +4n
+.EX
+int fsfd, mntfd;
+\&
+fsfd = fsopen("ext4", FSOPEN_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "ro", NULL, 0);
+fsconfig(fsfd, FSCONFIG_SET_PATH, "source", "/dev/sdb1", AT_FDCWD);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "noatime", NULL, 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "acl", NULL, 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "user_xattr", NULL, 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "iversion", NULL, 0)
+fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0);
+mntfd = fsmount(fsfd, FSMOUNT_CLOEXEC, MOUNT_ATTR_RELATIME);
+move_mount(mntfd, "", AT_FDCWD, "/mnt", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+First,
+an ext4 configuration context is created and attached to the file descriptor
+.IR fsfd .
+Then, a series of parameters
+(such as the source of the filesystem)
+are provided using
+.BR fsconfig (2),
+followed by the filesystem instance being created with
+.BR \%FSCONFIG_CMD_CREATE .
+.BR fsmount (2)
+is then used to create a new mount object attached to the file descriptor
+.IR mntfd ,
+which is then attached to the intended mount point using
+.BR move_mount (2).
+.P
+The above procedure is functionally equivalent to
+the following mount operation using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/dev/sdb1", "/mnt", "ext4", MS_RELATIME,
+      "ro,noatime,acl,user_xattr,iversion");
+.EE
+.in
+.P
+And here's an example of creating a mount object
+of an NFS server share
+and setting a Smack security module label.
+However, instead of attaching it to a mount point,
+the program uses the mount object directly
+to open a file from the NFS share.
+.P
+.in +4n
+.EX
+int fsfd, mntfd, fd;
+\&
+fsfd = fsopen("nfs", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "source", "example.com/pub/linux", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "nfsvers", "3", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "rsize", "65536", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "wsize", "65536", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "smackfsdef", "foolabel", 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "rdma", NULL, 0);
+fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0);
+mntfd = fsmount(fsfd, 0, MOUNT_ATTR_NODEV);
+fd = openat(mntfd, "src/linux-5.2.tar.xz", O_RDONLY);
+.EE
+.in
+.P
+Unlike the previous example,
+this operation has no trivial equivalent with
+.BR mount (2),
+as it was not previously possible to create a mount object
+that is not attached to any mount point.
+.SH SEE ALSO
+.BR fsconfig (2),
+.BR fsmount (2),
+.BR fspick (2),
+.BR mount (2),
+.BR mount_setattr (2),
+.BR move_mount (2),
+.BR open_tree (2),
+.BR mount_namespaces (7)

-- 
2.50.1

[PATCH v3 05/12] man/man2/fspick.2: document "new" mount API

[Aleksa Sarai]

This is loosely based on the original documentation written by David
Howells and later maintained by Christian Brauner, but has been
rewritten to be more from a user perspective (as well as fixing a few
critical mistakes).

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/fspick.2 | 309 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 309 insertions(+)

diff --git a/man/man2/fspick.2 b/man/man2/fspick.2
new file mode 100644
index 0000000000000000000000000000000000000000..a1060bcdb7d57b0656d4065683b5c69407550038
--- /dev/null
+++ b/man/man2/fspick.2
@@ -0,0 +1,309 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH fspick 2 (date) "Linux man-pages (unreleased)"
+.SH NAME
+fspick \- select filesystem for reconfiguration
+.SH LIBRARY
+Standard C library
+.RI ( libc ,\~ \-lc )
+.SH SYNOPSIS
+.nf
+.BR "#include <fcntl.h>" "          /* Definition of " AT_* " constants */"
+.B #include <sys/mount.h>
+.P
+.BI "int fspick(int " dirfd ", const char *" path ", unsigned int " flags ");"
+.fi
+.SH DESCRIPTION
+The
+.BR fspick ()
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.P
+.BR fspick()
+creates a new filesystem configuration context
+for the extant filesystem instance
+associated with the path described by
+.IR dirfd
+and
+.IR path ,
+places it into reconfiguration mode
+(similar to
+.BR mount (8)
+with the
+.I -o remount
+option).
+A new file descriptor
+associated with the filesystem configuration context
+is then returned.
+The calling process must have the
+.BR CAP_SYS_ADMIN
+capability in order to create a new filesystem configuration context.
+.P
+The resultant file descriptor can be used with
+.BR fsconfig (2)
+to specify the desired set of changes to
+filesystem parameters of the filesystem instance.
+Once the desired set of changes have been configured,
+the changes can be effectuated by calling
+.BR fsconfig (2)
+with the
+.B \%FSCONFIG_CMD_RECONFIGURE
+command.
+.P
+As with "*at()" system calls,
+.BR fspick ()
+uses the
+.I dirfd
+argument in conjunction with the
+.I path
+argument to determine the path to operate on, as follows:
+.IP \[bu] 3
+If the pathname given in
+.I path
+is absolute, then
+.I dirfd
+is ignored.
+.IP \[bu]
+If the pathname given in
+.I path
+is relative and
+.I dirfd
+is the special value
+.BR \%AT_FDCWD ,
+then
+.I path
+is interpreted relative to
+the current working directory
+of the calling process (like
+.BR open (2)).
+.IP \[bu]
+If the pathname given in
+.I path
+is relative,
+then it is interpreted relative to
+the directory referred to by the file descriptor
+.I dirfd
+(rather than relative to
+the current working directory
+of the calling process,
+as is done by
+.BR open (2)
+for a relative pathname).
+In this case,
+.I dirfd
+must be a directory
+that was opened for reading
+.RB ( O_RDONLY )
+or using the
+.B O_PATH
+flag.
+.IP \[bu]
+If
+.I path
+is an empty string,
+and
+.I flags
+contains
+.BR \%FSPICK_EMPTY_PATH ,
+then the file descriptor
+.I dirfd
+is operated on directly.
+In this case,
+.I dirfd
+may refer to any type of file,
+not just a directory.
+.P
+.I flags
+can be used to control aspects of how
+.I path
+is resolved and
+properties of the returned file descriptor.
+A value for
+.I flags
+is constructed by bitwise ORing
+zero or more of the following constants:
+.RS
+.TP
+.B FSPICK_CLOEXEC
+Set the close-on-exec
+.RB ( FD_CLOEXEC )
+flag on the new file descriptor.
+See the description of the
+.B O_CLOEXEC
+flag in
+.BR open (2)
+for reasons why this may be useful.
+.TP
+.B FSPICK_EMPTY_PATH
+If
+.I path
+is an empty string,
+operate on the file referred to by
+.I dirfd
+(which may have been obtained from
+.BR open (2),
+.BR fsmount (2),
+or
+.BR open_tree (2)).
+In this case,
+.I dirfd
+may refer to any type of file,
+not just a directory.
+If
+.I dirfd
+is
+.BR \%AT_FDCWD ,
+.BR fspick ()
+will operate on the current working directory
+of the calling process.
+.TP
+.B FSPICK_SYMLINK_NOFOLLOW
+Do not follow symbolic links
+in the terminal component of
+.IR path .
+If
+.I path
+references a symbolic link,
+the returned filesystem context will reference
+the filesystem that the symbolic link itself resides on.
+.TP
+.B FSPICK_NO_AUTOMOUNT
+Do not automount any automount points encountered
+while resolving
+.IR path .
+This allows you to reconfigure an automount point,
+rather than the location that would be mounted.
+This flag has no effect if
+the automount point has already been mounted over.
+.RE
+.P
+As with filesystem contexts created with
+.BR fsopen (2),
+the file descriptor returned by
+.BR fspick ()
+may be queried for message strings at any time by calling
+.BR read (2)
+on the file descriptor.
+(See the "Message retrieval interface" subsection in
+.BR fsopen (2)
+for more details on the message format.)
+.SH RETURN VALUE
+On success, a new file descriptor is returned.
+On error, \-1 is returned, and
+.I errno
+is set to indicate the error.
+.SH ERRORS
+.TP
+.B EACCES
+Search permission is denied
+for one of the directories
+in the path prefix of
+.IR path .
+(See also
+.BR path_resolution (7).)
+.TP
+.B EBADF
+.I path
+is relative but
+.I dirfd
+is neither
+.B \%AT_FDCWD
+nor a valid file descriptor.
+.TP
+.B EFAULT
+.I path
+is NULL
+or a pointer to a location
+outside the calling process's accessible address space.
+.TP
+.B EINVAL
+Invalid flag specified in
+.IR flags .
+.TP
+.B ELOOP
+Too many symbolic links encountered when resolving
+.IR path .
+.TP
+.B EMFILE
+The calling process has too many open files to create more.
+.TP
+.B ENAMETOOLONG
+.I path
+is longer than
+.BR PATH_MAX .
+.TP
+.B ENFILE
+The system has too many open files to create more.
+.TP
+.B ENOENT
+A component of
+.I path
+does not exist,
+or is a dangling symbolic link.
+.TP
+.B ENOENT
+.I path
+is an empty string, but
+.B \%FSPICK_EMPTY_PATH
+is not specified in
+.IR flags .
+.TP
+.B ENOTDIR
+A component of the path prefix of
+.I path
+is not a directory;
+or
+.I path
+is relative and
+.I dirfd
+is a file descriptor referring to a file other than a directory.
+.TP
+.B ENOMEM
+The kernel could not allocate sufficient memory to complete the operation.
+.TP
+.B EPERM
+The calling process does not have the required
+.B \%CAP_SYS_ADMIN
+capability.
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.2.
+.\" commit cf3cba4a429be43e5527a3f78859b1bfd9ebc5fb
+.\" commit 400913252d09f9cfb8cce33daee43167921fc343
+glibc 2.36.
+.SH EXAMPLES
+The following example sets the read-only flag
+on the filesystem instance referenced by
+the mount object attached at
+.IR /tmp .
+.P
+.in +4n
+.EX
+int fsfd = fspick(AT_FDCWD, "/tmp", FSPICK_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "ro", NULL, 0);
+fsconfig(fsfd, FSCONFIG_CMD_RECONFIGURE, NULL, NULL, 0);
+.EE
+.in
+.P
+The above procedure is functionally equivalent to
+the following mount operation using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount(NULL, "/tmp", NULL, MS_REMOUNT | MS_RDONLY, NULL);
+.EE
+.in
+.SH SEE ALSO
+.BR fsconfig (2),
+.BR fsmount (2),
+.BR fsopen (2),
+.BR mount (2),
+.BR mount_setattr (2),
+.BR move_mount (2),
+.BR open_tree (2),
+.BR mount_namespaces (7)
+

-- 
2.50.1

Re: [PATCH v3 05/12] man/man2/fspick.2: document "new" mount API

[Askar Safin]

 ---- On Sat, 09 Aug 2025 00:39:49 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > +The above procedure is functionally equivalent to
 > +the following mount operation using
 > +.BR mount (2):

This is not true.

fspick adds options to superblock. It doesn't remove existing ones.

mount(MS_REMOUNT) replaces options. I. e. mount(2) call provided in
example will unset all other options.

In the end of this message you will find C code, which proves this.

Also, recently I sent patch to mount(2) manpage,
which clarifies MS_REMOUNT | MS_BIND behavior.
This is somewhat related.
The patch comes with another reproducer.
Here is a link: https://lore.kernel.org/linux-man/20250822114315.1571537-1-safinaskar@zohomail.com/

--
Askar Safin
https://types.pl/@safinaskar

// You need to be root (non-initial user namespace will go)

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <stdbool.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sched.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/mount.h>
#include <sys/syscall.h>
#include <sys/vfs.h>
#include <sys/sysmacros.h>
#include <sys/statvfs.h>
#include <linux/openat2.h>

#define MY_ASSERT(cond) do { \
    if (!(cond)) { \
        fprintf (stderr, "%d: %s: assertion failed\n", __LINE__, #cond); \
        exit (1); \
    } \
} while (0)

int
main (void)
{
    // Init
    {
        MY_ASSERT (chdir ("/") == 0);
        MY_ASSERT (unshare (CLONE_NEWNS) == 0);
        MY_ASSERT (mount (NULL, "/", NULL, MS_PRIVATE | MS_REC, NULL) == 0);
        MY_ASSERT (mount (NULL, "/tmp", "tmpfs", 0, NULL) == 0);
    }

    MY_ASSERT (mkdir ("/tmp/a", 0777) == 0);
    MY_ASSERT (mkdir ("/tmp/b", 0777) == 0);

    {
        MY_ASSERT (mount (NULL, "/tmp/a", "tmpfs", MS_SYNCHRONOUS, NULL) == 0);
        {
            struct statfs buf;
            memset (&buf, 0, sizeof buf);
            MY_ASSERT (statfs ("/tmp/a", &buf) == 0);
            MY_ASSERT (buf.f_flags & ST_SYNCHRONOUS);
            MY_ASSERT (!(buf.f_flags & ST_RDONLY));
        }
        {
            int fsfd = fspick (AT_FDCWD, "/tmp/a", FSPICK_CLOEXEC);
            MY_ASSERT (fsfd >= 0);
            MY_ASSERT (fsconfig (fsfd, FSCONFIG_SET_FLAG, "ro", NULL, 0) == 0);
            MY_ASSERT (fsconfig (fsfd, FSCONFIG_CMD_RECONFIGURE, NULL, NULL, 0) == 0);
            MY_ASSERT (close (fsfd) == 0);
        }
        {
            struct statfs buf;
            memset (&buf, 0, sizeof buf);
            MY_ASSERT (statfs ("/tmp/a", &buf) == 0);
            MY_ASSERT (buf.f_flags & ST_SYNCHRONOUS);
            MY_ASSERT (buf.f_flags & ST_RDONLY);
        }
        MY_ASSERT (umount ("/tmp/a") == 0);
    }

    {
        MY_ASSERT (mount (NULL, "/tmp/a", "tmpfs", MS_SYNCHRONOUS, NULL) == 0);
        {
            struct statfs buf;
            memset (&buf, 0, sizeof buf);
            MY_ASSERT (statfs ("/tmp/a", &buf) == 0);
            MY_ASSERT (buf.f_flags & ST_SYNCHRONOUS);
            MY_ASSERT (!(buf.f_flags & ST_RDONLY));
        }
        MY_ASSERT (mount (NULL, "/tmp/a", NULL, MS_REMOUNT | MS_RDONLY, NULL) == 0);
        {
            struct statfs buf;
            memset (&buf, 0, sizeof buf);
            MY_ASSERT (statfs ("/tmp/a", &buf) == 0);
            MY_ASSERT (!(buf.f_flags & ST_SYNCHRONOUS));
            MY_ASSERT (buf.f_flags & ST_RDONLY);
        }
        MY_ASSERT (umount ("/tmp/a") == 0);
    }

    printf ("All tests passed\n");
    exit (0);
}

Re: [PATCH v3 05/12] man/man2/fspick.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 1055 bytes --]

On 2025-08-22, Askar Safin <safinaskar@zohomail.com> wrote:
>  ---- On Sat, 09 Aug 2025 00:39:49 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
>  > +The above procedure is functionally equivalent to
>  > +the following mount operation using
>  > +.BR mount (2):
> 
> This is not true.
> 
> fspick adds options to superblock. It doesn't remove existing ones.

fspick "copies the existing parameters" would be more accurate. I can
reword this, but it's an example and I don't think it makes sense to add
a large amount of clarifying text for each example.

The comparisons to mount(2) are meant to be indicative, but if you I can
also just remove them (David's versions didn't include them).

> mount(MS_REMOUNT) replaces options. I. e. mount(2) call provided in
> example will unset all other options.
> 
> In the end of this message you will find C code, which proves this.

Yes, I am already keenly aware of this behaviour.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 05/12] man/man2/fspick.2: document "new" mount API

[Askar Safin]

 ---- On Fri, 22 Aug 2025 17:40:18 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > On 2025-08-22, Askar Safin <safinaskar@zohomail.com> wrote:
 > >  ---- On Sat, 09 Aug 2025 00:39:49 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > >  > +The above procedure is functionally equivalent to
 > >  > +the following mount operation using
 > >  > +.BR mount (2):
 > > 
 > > This is not true.
 > > 
 > > fspick adds options to superblock. It doesn't remove existing ones.
 > 
 > fspick "copies the existing parameters" would be more accurate. I can
 > reword this, but it's an example and I don't think it makes sense to add
 > a large amount of clarifying text for each example.

I suggest adding "but mount(2) clears existing parameters here, and fspick/fsconfig doesn't".

--
Askar Safin
https://types.pl/@safinaskar

[PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Aleksa Sarai]

This is loosely based on the original documentation written by David
Howells and later maintained by Christian Brauner, but has been
rewritten to be more from a user perspective (as well as fixing a few
critical mistakes).

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/fsconfig.2 | 670 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 670 insertions(+)

diff --git a/man/man2/fsconfig.2 b/man/man2/fsconfig.2
new file mode 100644
index 0000000000000000000000000000000000000000..97c9aff0e0c195e6028e1c7bd70e40905ba9f994
--- /dev/null
+++ b/man/man2/fsconfig.2
@@ -0,0 +1,670 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH fsconfig 2 (date) "Linux man-pages (unreleased)"
+.SH NAME
+fsconfig \- configure new or existing filesystem context
+.SH LIBRARY
+Standard C library
+.RI ( libc ,\~ \-lc )
+.SH SYNOPSIS
+.nf
+.B #include <sys/mount.h>
+.P
+.BI "int fsconfig(int " fd ", unsigned int " cmd ","
+.BI "             const char *" key ", const void *" value ", int " aux ");"
+.fi
+.SH DESCRIPTION
+The
+.BR fsconfig ()
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.P
+.BR fsconfig ()
+is used to supply parameters to
+and issue commands against
+the filesystem configuration context
+associated with the file descriptor
+.IR fd .
+Filesystem configuration contexts can be created with
+.BR fsopen (2)
+or instantiated from an extant filesystem instance with
+.BR fspick (2).
+.P
+The
+.I cmd
+argument indicates the command to be issued.
+Some commands supply parameters to the context
+(equivalent to mount options specified with
+.BR mount (8)),
+while others are meta-operations on the filesystem context.
+The list of valid
+.I cmd
+values are:
+.RS
+.TP
+.B FSCONFIG_SET_FLAG
+Set the flag parameter named by
+.IR key .
+.I value
+must be NULL,
+and
+.I aux
+must be 0.
+.TP
+.B FSCONFIG_SET_STRING
+Set the string parameter named by
+.I key
+to the value specified by
+.IR value .
+.I value
+points to a null-terminated string,
+and
+.I aux
+must be 0.
+.TP
+.B FSCONFIG_SET_BINARY
+Set the blob parameter named by
+.I key
+to the contents of the binary blob
+specified by
+.IR value .
+.I value
+points to
+the start of a buffer
+that is
+.I aux
+bytes in length.
+.TP
+.B FSCONFIG_SET_FD
+Set the file parameter named by
+.I key
+to the open file description
+referenced by the file descriptor
+.IR aux .
+.I value
+must be NULL.
+.IP
+You may also use
+.B \%FSCONFIG_SET_STRING
+for file parameters,
+with
+.I value
+set to a null-terminated string
+containing a base-10 representation
+of the file descriptor number.
+This mechanism is primarily intended for compatibility
+with older
+.BR mount (2)-based
+programs,
+and only works for parameters
+that only accept file descriptor arguments.
+.TP
+.B FSCONFIG_SET_PATH
+Set the path parameter named by
+.I key
+to the object at a provided path,
+resolved in a similar manner to
+.BR openat (2).
+.I value
+points to a null-terminated pathname string,
+and
+.I aux
+is equivalent to the
+.I dirfd
+argument to
+.BR openat (2).
+.IP
+You may also use
+.B \%FSCONFIG_SET_STRING
+for path parameters,
+the behaviour of which is equivalent to
+.B \%FSCONFIG_SET_PATH
+with
+.I aux
+set to
+.BR \%AT_FDCWD .
+.TP
+.B FSCONFIG_SET_PATH_EMPTY
+As with
+.BR \%FSCONFIG_SET_PATH ,
+except that if
+.I value
+is an empty string,
+the file descriptor specified by
+.I aux
+is operated on directly
+and may be any type of file
+(not just a directory).
+This is equivalent to the behaviour of
+.B \%AT_EMPTY_PATH
+with most "*at()" system calls.
+If
+.I aux
+is
+.BR \%AT_FDCWD ,
+the call operates on
+the current working directory of the calling process.
+.TP
+.B FSCONFIG_CMD_CREATE
+This command instructs the filesystem driver
+to instantiate an instance of the filesystem in the kernel
+with the parameters specified in the filesystem configuration context.
+.IR key " and " value
+must be NULL,
+and
+.I aux
+must be 0.
+.IP
+If this operation succeeds,
+the filesystem context
+associated with file descriptor
+.I fd
+now references the created filesystem instance,
+and is placed into a special "awaiting-mount" mode
+that allows you to use
+.BR fsmount (2)
+to create a mount object from the filesystem instance.
+.\" FS_CONTEXT_AWAITING_MOUNT is the term the kernel uses for this.
+.IP
+This is intended for use with filesystem configuration contexts created with
+.BR fsopen (2).
+In order to create a filesystem instance,
+the calling process must have the
+.B \%CAP_SYS_ADMIN
+capability.
+.IP
+Note that the Linux kernel reuses filesystem instances
+for many filesystems,
+so (depending on the filesystem being configured and parameters used)
+it is possible for the filesystem instance "created" by
+.B \%FSCONFIG_CMD_CREATE
+to, in fact, be a reference
+to an existing filesystem instance in the kernel.
+The kernel will attempt to merge the specified parameters
+of this filesystem configuration context
+with those of the filesystem instance being reused,
+but some parameters may be
+.IR "silently ignored" .
+.IP
+Programs that need to ensure
+that they create a new filesystem instance
+with specific parameters
+(notably, security-related parameters
+such as
+.I acl
+to enable POSIX ACLs as described in
+.BR acl (5))
+should use
+.B \%FSCONFIG_CMD_CREATE_EXCL
+instead.
+.TP
+.BR FSCONFIG_CMD_CREATE_EXCL " (since Linux 6.6)"
+.\" commit 22ed7ecdaefe0cac0c6e6295e83048af60435b13
+.\" commit 84ab1277ce5a90a8d1f377707d662ac43cc0918a
+As with
+.BR \%FSCONFIG_CMD_CREATE ,
+except that the kernel is instructed
+to create a new filesystem instance
+("superblock" in kernel-developer parlance)
+rather than reusing an existing one.
+.IP
+If this is not possible
+(such as with disk-backed filesystems
+where multiple filesystem instances
+using the same filesystem driver
+and writing to the same underlying device
+could result in data corruption),
+this operation will incur
+an
+.B EBUSY
+error.
+.IP
+As a result (unlike
+.BR \%FSCONFIG_CMD_CREATE ),
+if this command succeeds
+then the calling process can be sure that
+all of the parameters successfully configured with
+.BR fsconfig ()
+will actually be applied
+to the created filesystem instance.
+.TP
+.B FSCONFIG_CMD_RECONFIGURE
+This command instructs the filesystem driver
+to apply the parameters specified in the filesystem configuration context
+to the extant filesystem instance
+referenced by the filesystem configuration context.
+.IP
+This is primarily intended for use with
+.BR fspick (2),
+but may also be used to modify the parameters of filesystem instance after
+.BR \%FSCONFIG_CMD_CREATE
+was used to create it
+and a mount object was created using
+.BR fsmount (2).
+In order to reconfigure an extant filesystem instance,
+the calling process must have the
+.B CAP_SYS_ADMIN
+capability.
+.IP
+Once this operation succeeds, the filesystem context is reset
+but remains in reconfiguration mode
+and thus can be used for subsequent
+.B \%FSCONFIG_CMD_RECONFIGURE
+commands.
+.RE
+.P
+Parameters specified with
+.BI FSCONFIG_SET_ *
+do not take effect
+until the corresponding
+.B \%FSCONFIG_CMD_CREATE
+or
+.B \%FSCONFIG_CMD_RECONFIGURE
+command is issued.
+.SH RETURN VALUE
+On success,
+.BR fsconfig ()
+returns 0.
+On error, \-1 is returned, and
+.I errno
+is set to indicate the error.
+.SH ERRORS
+If an error occurs, the filesystem driver may provide
+additional information about the error
+through the message retrieval interface for filesystem configuration contexts.
+This additional information can be retrieved at any time by calling
+.BR read (2)
+on the filesystem instance or filesystem configuration context
+referenced by the file descriptor
+.IR fd .
+(See the "Message retrieval interface" subsection in
+.BR fsopen (2)
+for more details on the message format.)
+.P
+Even after an error occurs,
+the filesystem configuration context is
+.I not
+invalidated,
+and thus can still be used with other
+.BR fsconfig ()
+commands.
+This means that users can probe support for filesystem parameters
+on a per-parameter basis,
+and adjust which parameters they wish to set.
+.P
+The error values given below result from
+filesystem type independent errors.
+Each filesystem type may have its own special errors
+and its own special behavior.
+See the Linux kernel source code for details.
+.TP
+.B EACCES
+A component of a path
+provided as a path parameter
+was not searchable.
+(See also
+.BR path_resolution (7).)
+.TP
+.B EACCES
+.B \%FSCONFIG_CMD_CREATE
+was attempted
+for a read-only filesystem
+without specifying the
+.RB ' ro '
+flag parameter.
+.TP
+.B EACCES
+A specified block device parameter
+is located on a filesystem
+mounted with the
+.B \%MS_NODEV
+option.
+.TP
+.B EBADF
+The file descriptor given by
+.I fd
+(or possibly by
+.IR aux ,
+depending on the command)
+is invalid.
+.TP
+.B EBUSY
+The filesystem context associated with
+.I fd
+is in the wrong state
+for the given command.
+.TP
+.B EBUSY
+The filesystem instance cannot be reconfigured as read-only
+with
+.B \%FSCONFIG_CMD_RECONFIGURE
+because some programs
+still hold files open for writing.
+.TP
+.B EBUSY
+A new filesystem instance was requested with
+.B \%FSCONFIG_CMD_CREATE_EXCL
+but a matching superblock already existed.
+.TP
+.B EFAULT
+One of the pointer arguments
+points to a location
+outside the calling process's accessible address space.
+.TP
+.B EINVAL
+.I fd
+does not refer to
+a filesystem configuration context
+or filesystem instance.
+.TP
+.B EINVAL
+One of the values of
+.IR name ,
+.IR value ,
+and/or
+.I aux
+were set to a non-zero value when
+.I cmd
+required that they be zero
+(or NULL).
+.TP
+.B EINVAL
+The parameter named by
+.I name
+cannot be set
+using the type specified with
+.IR cmd .
+.TP
+.B EINVAL
+One of the source parameters
+referred to
+an invalid superblock.
+.TP
+.B ELOOP
+Too many links encountered
+during pathname resolution
+of a path argument.
+.TP
+.B ENAMETOOLONG
+A path argument was longer than
+.BR PATH_MAX .
+.TP
+.B ENOENT
+A path argument had a non-existent component.
+.TP
+.B ENOENT
+A path argument is an empty string,
+but
+.I cmd
+is not
+.BR \%FSCONFIG_SET_PATH_EMPTY .
+.TP
+.B ENOMEM
+The kernel could not allocate sufficient memory to complete the operation.
+.TP
+.B ENOTBLK
+The parameter named by
+.I name
+must be a block device,
+but the provided parameter value was not a block device.
+.TP
+.B ENOTDIR
+A component of the path prefix
+of a path argument
+was not a directory.
+.TP
+.B EOPNOTSUPP
+The command given by
+.I cmd
+is not valid.
+.TP
+.B ENXIO
+The major number
+of a block device parameter
+is out of range.
+.TP
+.B EPERM
+The command given by
+.I cmd
+was
+.BR \%FSCONFIG_CMD_CREATE ,
+.BR \%FSCONFIG_CMD_CREATE_EXCL ,
+or
+.BR \% FSCONFIG_CMD_RECONFIGURE ,
+but the calling process does not have the required
+.B \%CAP_SYS_ADMIN
+capability.
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.2.
+.\" commit ecdab150fddb42fe6a739335257949220033b782
+.\" commit 400913252d09f9cfb8cce33daee43167921fc343
+glibc 2.36.
+.SH NOTES
+.SS Generic filesystem parameters
+Each filesystem driver is responsible for
+parsing most parameters specified with
+.BR fsconfig (),
+meaning that individual filesystems
+may have very different behaviour
+when encountering parameters with the same name.
+In general,
+you should not assume that the behaviour of
+.BR fsconfig ()
+when specifying a parameter to one filesystem type
+will match the behaviour of the same parameter
+with a different filesystem type.
+.P
+However,
+the following generic parameters
+apply to all filesystems and have unified behaviour.
+They are set using the listed
+.BI \%FSCONFIG_SET_ *
+command.
+.TP
+\fIro\fP and \fIrw\fP (\fB\%FSCONFIG_SET_FLAG\fP)
+Configure whether the filesystem instance is read-only.
+.TP
+\fIdirsync\fP (\fB\%FSCONFIG_SET_FLAG\fP)
+Make directory changes on this filesystem instance synchronous.
+.TP
+\fIsync\fP and \fIasync\fP (\fB\%FSCONFIG_SET_FLAG\fP)
+Configure whether writes on this filesystem instance
+will be made synchronous
+(as though the
+.B O_SYNC
+flag to
+.BR open (2)
+was specified for
+all file opens in this filesystem instance).
+.TP
+\fIlazytime\fP and \fInolazytime\fP (\fB\%FSCONFIG_SET_FLAG\fP)
+Configure whether to reduce on-disk updates
+of inode timestamps on this filesystem instance
+(as described in the
+.B \%MS_LAZYTIME
+section of
+.BR mount (2)).
+.TP
+\fImand\fP and \fInomand\fP (\fB\%FSCONFIG_SET_FLAG\fP)
+Configure whether the filesystem instance should permit mandatory locking.
+Since Linux 5.15,
+.\" commit f7e33bdbd6d1bdf9c3df8bba5abcf3399f957ac3
+mandatory locking has been deprecated
+and setting this flag is a no-op.
+.TP
+\fIsource\fP (\fB\%FSCONFIG_SET_STRING\fP)
+This parameter is equivalent to the
+.I source
+parameter passed to
+.BR mount (2)
+for the same filesystem type,
+and is usually the pathname of a block device
+containing the filesystem.
+This parameter may only be set once
+per filesystem configuration context transaction.
+.IP
+Note that individual filesystem drivers
+may choose to additionally permit other
+.BI FSCONFIG_SET_ *
+commands be used with
+.IR source ,
+.\" (i.e., fc->ops->parse_param() parses "source")
+but setting this parameter with
+.B FSCONFIG_SET_STRING
+is always allowed.
+.P
+In addition,
+any filesystem parameters associated with
+Linux Security Modules (LSMs)
+are also generic with respect to the underlying filesystem.
+See the documentation for the LSM you wish to configure for more details.
+.SH CAVEATS
+.SS Filesystem parameter types
+As a result of
+each filesystem driver being responsible for
+parsing most parameters specified with
+.BR fsconfig (),
+some filesystem parameters
+may have unintuitive behaviour
+with regards to which
+.BI \%FSCONFIG_SET_ *
+commands are permitted
+to configure a given parameter.
+.P
+In order for
+filesystem parameters to be backwards compatible with
+.BR mount (2),
+they must be parseable as strings;
+this almost universally means that
+.B \%FSCONFIG_SET_STRING
+can also be used to configure them.
+.\" Aleksa Sarai
+.\"   Theoretically, a filesystem could check fc->oldapi and refuse
+.\"   FSCONFIG_SET_STRING if the operation is coming from the new API, but no
+.\"   filesystems do this (and probably never will).
+However, other
+.BI \%FSCONFIG_SET_ *
+commands need to be opted into
+by each filesystem driver's parameter parser.
+.P
+One of the most user-visible instances of
+this inconsistency is that
+many filesystems do not support
+configuring path parameters with
+.B \%FSCONFIG_SET_PATH
+(despite the name),
+which can lead to somewhat confusing
+.B EINVAL
+errors.
+(For example, the generic
+.I source
+parameter\[em]which is usually a path\[em]can only be configured
+with
+.BR \%FSCONFIG_SET_STRING .)
+.P
+When writing programs that use
+.BR fsconfig ()
+to configure parameters
+with commands other than
+.BR \%FSCONFIG_SET_STRING ,
+users should verify
+that the
+.BI \%FSCONFIG_SET_ *
+commands used to configure each parameter
+are supported by the corresponding filesystem driver.
+.\" Aleksa Sarai
+.\"   While this (quite confusing) inconsistency in behaviour is true today
+.\"   (and has been true since this was merged), this appears to mostly be an
+.\"   unintended consequence of filesystem drivers hand-coding fsparam parsing.
+.\"   Path parameters are the most eggregious causes of confusion. Hopefully we
+.\"   can make this no longer the case in a future kernel.
+.SH EXAMPLES
+To illustrate the different kinds of flags that can be configured with
+.BR fsconfig (),
+here are a few examples of some different filesystems being created:
+.P
+.in +4n
+.EX
+int fsfd, mntfd;
+\&
+fsfd = fsopen("tmpfs", FSOPEN_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "inode64", NULL, 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "uid", "1234", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "huge", "never", 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "casefold", NULL, 0);
+fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0);
+mntfd = fsmount(fsfd, FSMOUNT_CLOEXEC, MOUNT_ATTR_NOEXEC);
+move_mount(mntfd, "", AT_FDCWD, "/tmp", MOVE_MOUNT_F_EMPTY_PATH);
+\&
+fsfd = fsopen("erofs", FSOPEN_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "source", "/dev/loop0", 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "acl", NULL, 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "user_xattr", NULL, 0);
+fsconfig(fsfd, FSCONFIG_CMD_CREATE_EXCL, NULL, NULL, 0);
+mntfd = fsmount(fsfd, FSMOUNT_CLOEXEC, MOUNT_ATTR_NOSUID);
+move_mount(mntfd, "", AT_FDCWD, "/mnt", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+Usually,
+specifying the same parameter named by
+.I key
+multiple times with
+.BR fsconfig ()
+causes the parameter value to be replaced.
+However, some filesystems can have unique behaviour:
+.P
+.in +4n
+.EX
+\&
+int fsfd, mntfd;
+int lowerdirfd = open("/o/ctr/lower1", O_DIRECTORY | O_CLOEXEC);
+\&
+fsfd = fsopen("overlay", FSOPEN_CLOEXEC);
+/* "lowerdir+" appends to the lower dir stack each time. */
+fsconfig(fsfd, FSCONFIG_SET_FD, "lowerdir+", NULL, lowerdirfd);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "lowerdir+", "/o/ctr/lower2", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "lowerdir+", "/o/ctr/lower3", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "lowerdir+", "/o/ctr/lower4", 0);
+.\" fsconfig(fsfd, FSCONFIG_SET_PATH, "lowerdir+", "/o/ctr/lower5", AT_FDCWD);
+.\" fsconfig(fsfd, FSCONFIG_SET_PATH_EMPTY, "lowerdir+", "", lowerdirfd);
+.\" Aleksa Sarai: Hopefully these will also be supported in the future.
+fsconfig(fsfd, FSCONFIG_SET_STRING, "xino", "auto", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "nfs_export", "off", 0);
+fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0);
+mntfd = fsmount(fsfd, FSMOUNT_CLOEXEC, 0);
+move_mount(mntfd, "", AT_FDCWD, "/mnt", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+And here is an example of how
+.BR fspick (2)
+can be used with
+.BR fsconfig ()
+to reconfigure the parameters
+of an extant filesystem instance
+attached to
+.IR /proc :
+.P
+.in +4n
+.EX
+int fsfd = fspick(AT_FDCWD, "/proc", FSPICK_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "hidepid", "ptraceable", 0);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "subset", "pid", 0);
+fsconfig(fsfd, FSCONFIG_CMD_RECONFIGURE, NULL, NULL, 0);
+.EE
+.in
+.SH SEE ALSO
+.BR fsmount (2),
+.BR fsopen (2),
+.BR fspick (2),
+.BR mount (2),
+.BR mount_setattr (2),
+.BR move_mount (2),
+.BR open_tree (2),
+.BR mount_namespaces (7)
+

-- 
2.50.1

Re: [PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 968 bytes --]

On 2025-08-09, Aleksa Sarai <cyphar@cyphar.com> wrote:
> +Note that the Linux kernel reuses filesystem instances
> +for many filesystems,
> +so (depending on the filesystem being configured and parameters used)
> +it is possible for the filesystem instance "created" by
> +.B \%FSCONFIG_CMD_CREATE
> +to, in fact, be a reference
> +to an existing filesystem instance in the kernel.
> +The kernel will attempt to merge the specified parameters
> +of this filesystem configuration context
> +with those of the filesystem instance being reused,
> +but some parameters may be
> +.IR "silently ignored" .

While looking at this again, I realised this explanation is almost
certainly incorrect in a few places (and was based on a misunderstanding
of how sget_fc() works and how it interacts with vfs_get_tree()).

I'll rewrite this in the next version.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Askar Safin]

 ---- On Tue, 12 Aug 2025 22:25:40 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > On 2025-08-09, Aleksa Sarai <cyphar@cyphar.com> wrote:
 > > +Note that the Linux kernel reuses filesystem instances
 > > +for many filesystems,
 > > +so (depending on the filesystem being configured and parameters used)
 > > +it is possible for the filesystem instance "created" by
 > > +.B \%FSCONFIG_CMD_CREATE
 > > +to, in fact, be a reference
 > > +to an existing filesystem instance in the kernel.
 > > +The kernel will attempt to merge the specified parameters
 > > +of this filesystem configuration context
 > > +with those of the filesystem instance being reused,
 > > +but some parameters may be
 > > +.IR "silently ignored" .
 > 
 > While looking at this again, I realised this explanation is almost
 > certainly incorrect in a few places (and was based on a misunderstanding
 > of how sget_fc() works and how it interacts with vfs_get_tree()).
 > 
 > I'll rewrite this in the next version.

This recent patch seems to be relevant:
https://lore.kernel.org/all/20250816-debugfs-mount-opts-v3-1-d271dad57b5b@posteo.net/

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 1750 bytes --]

On 2025-08-21, Askar Safin <safinaskar@zohomail.com> wrote:
>  ---- On Tue, 12 Aug 2025 22:25:40 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
>  > On 2025-08-09, Aleksa Sarai <cyphar@cyphar.com> wrote:
>  > > +Note that the Linux kernel reuses filesystem instances
>  > > +for many filesystems,
>  > > +so (depending on the filesystem being configured and parameters used)
>  > > +it is possible for the filesystem instance "created" by
>  > > +.B \%FSCONFIG_CMD_CREATE
>  > > +to, in fact, be a reference
>  > > +to an existing filesystem instance in the kernel.
>  > > +The kernel will attempt to merge the specified parameters
>  > > +of this filesystem configuration context
>  > > +with those of the filesystem instance being reused,
>  > > +but some parameters may be
>  > > +.IR "silently ignored" .
>  > 
>  > While looking at this again, I realised this explanation is almost
>  > certainly incorrect in a few places (and was based on a misunderstanding
>  > of how sget_fc() works and how it interacts with vfs_get_tree()).
>  > 
>  > I'll rewrite this in the next version.
> 
> This recent patch seems to be relevant:
> https://lore.kernel.org/all/20250816-debugfs-mount-opts-v3-1-d271dad57b5b@posteo.net/

I'm aware of that, I was in one of the previous threads. There are some
deeper consistency issues that I'm writing patches for at the moment.

I'm of two minds whether I should fix the behaviour and then re-send
man-pages with updated text (delaying the next round of man-page reviews
by a month) or just reduce the specificity of this text and then add
more details after it has been fixed.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Askar Safin]

 ---- On Thu, 21 Aug 2025 15:44:42 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > I'm of two minds whether I should fix the behaviour and then re-send
 > man-pages with updated text (delaying the next round of man-page reviews
 > by a month) or just reduce the specificity of this text and then add
 > more details after it has been fixed.

Do what you want.
I'm not in hurry.

CC me if you write any patches, please.
--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Askar Safin]

There is a convention: you can pass invalid fd (such as -1) as dfd to *at-syscalls to enforce that the path is absolute.
This is documented. "man openat" says: "Specifying an invalid file descriptor number in dirfd can be used as a means to ensure that pathname is absolute".
But fsconfig with FSCONFIG_SET_PATH breaks this convention due to this line: https://elixir.bootlin.com/linux/v6.16/source/fs/fsopen.c#L377 .
I think this is a bug, and it should be fixed in kernel. Also, it is possible there are a lot of similarly buggy syscalls. All of them should be fixed,
and moreover a warning should be added to https://docs.kernel.org/process/adding-syscalls.html . And then new fsconfig behavior should be documented.
(Of course, I'm not saying that *you* should do all these. I'm just saying that this bug exists.) (I tested this.)

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 06/12] man/man2/fsconfig.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 1361 bytes --]

On 2025-08-21, Askar Safin <safinaskar@zohomail.com> wrote:
> There is a convention: you can pass invalid fd (such as -1) as dfd to *at-syscalls to enforce that the path is absolute.
> This is documented. "man openat" says: "Specifying an invalid file descriptor number in dirfd can be used as a means to ensure that pathname is absolute".
> But fsconfig with FSCONFIG_SET_PATH breaks this convention due to this line: https://elixir.bootlin.com/linux/v6.16/source/fs/fsopen.c#L377 .
> I think this is a bug, and it should be fixed in kernel. Also, it is possible there are a lot of similarly buggy syscalls. All of them should be fixed,
> and moreover a warning should be added to https://docs.kernel.org/process/adding-syscalls.html . And then new fsconfig behavior should be documented.
> (Of course, I'm not saying that *you* should do all these. I'm just saying that this bug exists.) (I tested this.)

Indeed, good catch! I think we discussed this before --
FSCONFIG_SET_PATH actually doesn't work with any parameters today so
it's not very surprising nobody has noticed this until now. I'll include
it in the set of fixes I have for fscontext.

(FWIW, the convention I see more commonly is -EBADF but that's just a
stylistic I suppose.)

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

[PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Aleksa Sarai]

This is loosely based on the original documentation written by David
Howells and later maintained by Christian Brauner, but has been
rewritten to be more from a user perspective (as well as fixing a few
critical mistakes).

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/fsmount.2 | 220 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 220 insertions(+)

diff --git a/man/man2/fsmount.2 b/man/man2/fsmount.2
new file mode 100644
index 0000000000000000000000000000000000000000..92331cb18272f9ac836e55e7f28faea3a3efbdac
--- /dev/null
+++ b/man/man2/fsmount.2
@@ -0,0 +1,220 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH fsmount 2 (date) "Linux man-pages (unreleased)"
+.SH NAME
+fsmount \- instantiate mount object from filesystem context
+.SH LIBRARY
+Standard C library
+.RI ( libc ,\~ \-lc )
+.SH SYNOPSIS
+.nf
+.B #include <sys/mount.h>
+.P
+.BI "int fsmount(int " fsfd ", unsigned int " flags ", \
+unsigned int " attr_flags ");"
+.fi
+.SH DESCRIPTION
+The
+.BR fsmount ()
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.P
+.BR fsmount ()
+creates a new detached mount object
+for the root of the new filesystem instance
+referenced by the filesystem context file descriptor
+.IR fsfd .
+A new file descriptor
+associated with the detached mount object
+is then returned.
+In order to create a mount object with
+.BR fsmount (),
+the calling process must have the
+.BR \%CAP_SYS_ADMIN
+capability.
+.P
+The filesystem context must have been created with a call to
+.BR fsopen (2)
+and then had a filesystem instance instantiated with a call to
+.BR fsconfig (2)
+with
+.B \%FSCONFIG_CMD_CREATE
+or
+.B \%FSCONFIG_CMD_CREATE_EXCL
+in order to be in the correct state
+for this operation
+(the "awaiting-mount" mode in kernel-developer parlance).
+.\" FS_CONTEXT_AWAITING_MOUNT is the term the kernel uses for this.
+Unlike
+.BR open_tree (2)
+with
+.BR \%OPEN_TREE_CLONE,
+.BR fsmount ()
+can only be called once
+in the lifetime of a filesystem instance
+to produce a mount object.
+.P
+As with file descriptors returned from
+.BR open_tree (2)
+called with
+.BR OPEN_TREE_CLONE ,
+the returned file descriptor
+can then be used with
+.BR move_mount (2),
+.BR mount_setattr (2),
+or other such system calls to do further mount operations.
+This mount object will be unmounted and destroyed
+when the file descriptor is closed
+if it was not otherwise attached to a mount point
+by calling
+.BR move_mount (2).
+The returned file descriptor
+also acts the same as one produced by
+.BR open (2)
+with
+.BR O_PATH ,
+meaning it can also be used as a
+.I dirfd
+argument
+to "*at()" system calls.
+.P
+.I flags
+controls the creation of the returned file descriptor.
+A value for
+.I flags
+is constructed by bitwise ORing
+zero or more of the following constants:
+.RS
+.TP
+.B FSMOUNT_CLOEXEC
+Set the close-on-exec
+.RB ( FD_CLOEXEC )
+flag on the new file descriptor.
+See the description of the
+.B O_CLOEXEC
+flag in
+.BR open (2)
+for reasons why this may be useful.
+.RE
+.P
+.I attr_flags
+specifies mount attributes
+which will be applied to the created mount object,
+in the form of
+.BI \%MOUNT_ATTR_ *
+flags.
+The flags are interpreted as though
+.BR mount_setattr (2)
+was called with
+.I attr.attr_set
+set to the same value as
+.IR attr_flags .
+.BI \% MOUNT_ATTR_ *
+flags which would require
+specifying additional fields in
+.BR mount_attr (2type)
+(such as
+.BR \%MOUNT_ATTR_IDMAP )
+are not valid flag values for
+.IR attr_flags .
+.P
+If the
+.BR fsmount ()
+operation is successful,
+the filesystem context
+associated with the file descriptor
+.I fsfd
+is reset
+and placed into reconfiguration mode,
+as if it were just returned by
+.BR fspick (2).
+You may continue to use
+.BR fsconfig (2)
+with the now-reset filesystem context,
+including issuing the
+.B \%FSCONFIG_CMD_RECONFIGURE
+command
+to reconfigure the filesystem instance.
+.SH RETURN VALUE
+On success, a new file descriptor is returned.
+On error, \-1 is returned, and
+.I errno
+is set to indicate the error.
+.SH ERRORS
+.TP
+.B EBUSY
+The filesystem context associated with
+.I fsfd
+is not in the right state
+to be used by
+.BR fsmount ().
+.TP
+.B EINVAL
+.I flags
+had an invalid flag set.
+.TP
+.B EINVAL
+.I attr_flags
+had an invalid
+.BI MOUNT_ATTR_ *
+flag set.
+.TP
+.B EMFILE
+The calling process has too many open files to create more.
+.TP
+.B ENFILE
+The system has too many open files to create more.
+.TP
+.B ENOSPC
+The "anonymous" mount namespace
+necessary to contain the new mount object
+could not be allocated,
+as doing so would exceed
+the configured per-user limit on
+the number of mount namespaces in the current user namespace.
+(See also
+.BR namespaces (7).)
+.TP
+.B ENOMEM
+The kernel could not allocate sufficient memory to complete the operation.
+.TP
+.B EPERM
+The calling process does not have the required
+.B CAP_SYS_ADMIN
+capability.
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.2.
+.\" commit 93766fbd2696c2c4453dd8e1070977e9cd4e6b6d
+.\" commit 400913252d09f9cfb8cce33daee43167921fc343
+glibc 2.36.
+.SH EXAMPLES
+.in +4n
+.EX
+int fsfd, mntfd, tmpfd;
+\&
+fsfd = fsopen("tmpfs", FSOPEN_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0);
+mntfd = fsmount(fsfd, FSMOUNT_CLOEXEC, MOUNT_ATTR_NODEV | MOUNT_ATTR_NOEXEC);
+\&
+/* Create a new file without attaching the mount object. */
+int tmpfd = openat(mntfd, "tmpfile", O_CREAT | O_EXCL | O_RDWR, 0600);
+unlinkat(mntfd, "tmpfile", 0);
+\&
+/* Attach the mount object to "/tmp". */
+move_mount(mntfd, "", AT_FDCWD, "/tmp", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.SH SEE ALSO
+.BR fsconfig (2),
+.BR fsopen (2),
+.BR fspick (2),
+.BR mount (2),
+.BR mount_setattr (2),
+.BR move_mount (2),
+.BR open_tree (2),
+.BR mount_namespaces (7)
+

-- 
2.50.1

Re: [PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Askar Safin]

fsmount:
> Unlike open_tree(2) with OPEN_TREE_CLONE, fsmount() can only be called once in the lifetime of a filesystem instance to produce a mount object.

I don't understand what you meant here. This phrase in its current form is wrong.
Consider this scenario: we did this:
fsopen(...)
fsconfig(..., FSCONFIG_SET_STRING, "source", ...)
fsconfig(..., FSCONFIG_CMD_CREATE, ...)
fsmount(...)
fsopen(...)
fsconfig(..., FSCONFIG_SET_STRING, "source", ...)
fsconfig(..., FSCONFIG_CMD_CREATE, ...)
fsmount(...)

We used FSCONFIG_CMD_CREATE here as opposed to FSCONFIG_CMD_CREATE_EXCL, thus
it is possible that second fsmount will return mount for the same superblock.
Thus that statement "fsmount() can only be called once in the lifetime of a filesystem instance to produce a mount object"
is not true.

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 2231 bytes --]

On 2025-08-12, Askar Safin <safinaskar@zohomail.com> wrote:
> fsmount:
> > Unlike open_tree(2) with OPEN_TREE_CLONE, fsmount() can only be called once in the lifetime of a filesystem instance to produce a mount object.
>
> I don't understand what you meant here. This phrase in its current form is wrong.
> Consider this scenario: we did this:
> fsopen(...)
> fsconfig(..., FSCONFIG_SET_STRING, "source", ...)
> fsconfig(..., FSCONFIG_CMD_CREATE, ...)
> fsmount(...)
> fsopen(...)
> fsconfig(..., FSCONFIG_SET_STRING, "source", ...)
> fsconfig(..., FSCONFIG_CMD_CREATE, ...)
> fsmount(...)
> 
> We used FSCONFIG_CMD_CREATE here as opposed to FSCONFIG_CMD_CREATE_EXCL, thus
> it is possible that second fsmount will return mount for the same superblock.
> Thus that statement "fsmount() can only be called once in the lifetime of a filesystem instance to produce a mount object"
> is not true.

Yeah, the superblock reuse behaviour makes this description less
coherent than what I was going for. My thinking was that a reused
superblock is (to userspace) conceptually a new filesystem instance
because they create it the same way as any other filesystem instance.
(In fact, the rest of the VFS treats them the same way too -- only
sget_fc() knows about superblock reuse.)

But yeah, "filesystem context" is more accurate here, so probably just:

  Unlike open_tree(2) with OPEN_TREE_CLONE, fsmount() can only be called
  once in the lifetime of a filesystem context.

Though maybe we should mention that it's fsopen(2)-only (even though
it's mentioned earlier in the DESCRIPTION)? If you read the sentence in
isolation you might get the wrong impression. Do you have any
alternative suggestions?

FWIW, superblock reuse is one of those things that is a fairly hairy
implementation detail of the VFS, and as such it has quite odd
semantics. I probably wouldn't have documented it as heavily if it
wasn't for the addition of FSCONFIG_CMD_CREATE_EXCL (maybe an entry in
BUGS or CAVEATS at most -- this behaviour has an even worse impact on
mount(2) but it's completely undocumented there).

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 481 bytes --]

On 2025-08-13, Aleksa Sarai <cyphar@cyphar.com> wrote:
> But yeah, "filesystem context" is more accurate here, so probably just:

Oops, I meant to include

>   Unlike open_tree(2) with OPEN_TREE_CLONE, fsmount() can only be called
>   once in the lifetime of a filesystem context.
                                                 to create a mount object.

at the end.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Askar Safin]

 ---- On Tue, 12 Aug 2025 18:33:04 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 >   Unlike open_tree(2) with OPEN_TREE_CLONE, fsmount() can only be called
 >   once in the lifetime of a filesystem context.

Weird. open_tree doesn't get filesystem context as argument at all.
I suggest just this:

  fsmount() can only be called
  once in the lifetime of a filesystem context.

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 831 bytes --]

On 2025-08-20, Askar Safin <safinaskar@zohomail.com> wrote:
>  ---- On Tue, 12 Aug 2025 18:33:04 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
>  >   Unlike open_tree(2) with OPEN_TREE_CLONE, fsmount() can only be called
>  >   once in the lifetime of a filesystem context.
> 
> Weird. open_tree doesn't get filesystem context as argument at all.
> I suggest just this:
> 
>   fsmount() can only be called
>   once in the lifetime of a filesystem context.

The reason I wanted to include the comparison is that you can create
multiple mount objects from the same underlying object using
open_tree(2) but that's not possible with fsmount(2) (at least, not
without creating a new filesystem context each time).

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 07/12] man/man2/fsmount.2: document "new" mount API

[Askar Safin]

 ---- On Wed, 20 Aug 2025 14:38:48 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > The reason I wanted to include the comparison is that you can create
 > multiple mount objects from the same underlying object using
 > open_tree(2) but that's not possible with fsmount(2) (at least, not
 > without creating a new filesystem context each time).

Okay, you may write that.

--
Askar Safin
https://types.pl/@safinaskar

[PATCH v3 08/12] man/man2/move_mount.2: document "new" mount API

[Aleksa Sarai]

This is loosely based on the original documentation written by David
Howells and later maintained by Christian Brauner, but has been
rewritten to be more from a user perspective (as well as fixing a few
critical mistakes).

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/move_mount.2 | 640 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 640 insertions(+)

diff --git a/man/man2/move_mount.2 b/man/man2/move_mount.2
new file mode 100644
index 0000000000000000000000000000000000000000..ccb5477620ce0ab37a2be11947bf262d50a52cc9
--- /dev/null
+++ b/man/man2/move_mount.2
@@ -0,0 +1,640 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH move_mount 2 (date) "Linux man-pages (unreleased)"
+.SH NAME
+move_mount \- move or attach mount object to filesystem
+.SH LIBRARY
+Standard C library
+.RI ( libc ,\~ \-lc )
+.SH SYNOPSIS
+.nf
+.BR "#include <fcntl.h>" "          /* Definition of " AT_* " constants */"
+.B #include <sys/mount.h>
+.P
+.BI "int move_mount(int " from_dirfd ", const char *" from_path ","
+.BI "               int " to_dirfd ", const char *" to_path ","
+.BI "               unsigned int " flags ");"
+.fi
+.SH DESCRIPTION
+The
+.BR move_mount ()
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.P
+.BR move_mount ()
+moves the mount object indicated by
+.I from_dirfd
+and
+.I from_path
+to the path indicated by
+.I to_dirfd
+and
+.IR to_path .
+The mount object being moved
+can be an existing mount point in the current mount namespace,
+or a detached mount object created by
+.BR fsmount (2)
+or
+.BR open_tree (2)
+with
+.BR \%OPEN_TREE_CLONE .
+.P
+To access the source mount object
+or the destination mount point,
+no permissions are required on the object itself,
+but if either pathname is supplied,
+execute (search) permission is required
+on all of the directories specified in
+.I from_path
+or
+.IR to_path .
+.P
+The calling process must have the
+.BR \%CAP_SYS_ADMIN
+capability in order to move or attach a mount object.
+.P
+As with "*at()" system calls,
+.BR move_mount ()
+uses the
+.I from_dirfd
+and
+.I to_dirfd
+arguments
+in conjunction with the
+.I from_path
+and
+.I to_path
+arguments to determine the source and destination objects to operate on
+(respectively), as follows:
+.IP \[bu] 3
+If the pathname given in
+.I *_path
+is absolute, then
+the corresponding
+.I *_dirfd
+is ignored.
+.IP \[bu]
+If the pathname given in
+.I *_path
+is relative and
+the corresponding
+.I *_dirfd
+is the special value
+.BR \%AT_FDCWD ,
+then
+.I *_path
+is interpreted relative to
+the current working directory
+of the calling process (like
+.BR open (2)).
+.IP \[bu]
+If the pathname given in
+.I *_path
+is relative,
+then it is interpreted relative to
+the directory referred to by
+the corresponding file descriptor
+.I *_dirfd
+(rather than relative to
+the current working directory
+of the calling process,
+as is done by
+.BR open (2)
+for a relative pathname).
+In this case,
+the corresponding
+.I *_dirfd
+must be a directory
+that was opened for reading
+.RB ( O_RDONLY )
+or using the
+.B O_PATH
+flag.
+.IP \[bu]
+If
+.I *_path
+is an empty string,
+and
+.I flags
+contains the appropriate
+.BI \%MOVE_MOUNT_ * _EMPTY_PATH
+flag,
+then the corresponding file descriptor
+.I *_dirfd
+is operated on directly.
+In this case,
+the corresponding
+.I *_dirfd
+may refer to any type of file,
+not just a directory.
+.P
+.I flags
+can be used to control aspects of the path lookup
+for both the source and destination objects,
+as well as other properties of the mount operation.
+A value for
+.I flags
+is constructed by bitwise ORing
+zero or more of the following constants:
+.RS
+.TP
+.B MOVE_MOUNT_F_EMPTY_PATH
+If
+.I from_path
+is an empty string, operate on the file referred to by
+.I from_dirfd
+(which may have been obtained from
+.BR open (2),
+.BR fsmount (2),
+or
+.BR open_tree (2)).
+In this case,
+.I from_dirfd
+may refer to any type of file,
+not just a directory.
+If
+.I from_dirfd
+is
+.BR \%AT_FDCWD ,
+.BR move_mount ()
+will operate on the current working directory
+of the calling process.
+.IP
+This is the most common mechanism
+used to attach detached mount objects
+produced by
+.BR fsmount (2)
+and
+.BR open_tree (2)
+to a mount point.
+.TP
+.B MOVE_MOUNT_T_EMPTY_PATH
+As with
+.BR \%MOVE_MOUNT_F_EMPTY_PATH ,
+except operating on
+.I to_dirfd
+and
+.IR to_path .
+.TP
+.B MOVE_MOUNT_F_SYMLINKS
+If
+.IR from_path
+references a symbolic link,
+then dereference it.
+The default behaviour for
+.BR move_mount ()
+is to
+.I not follow
+symbolic links.
+.TP
+.B MOVE_MOUNT_T_SYMLINKS
+As with
+.BR \%MOVE_MOUNT_F_SYMLINKS ,
+except operating on
+.I to_dirfd
+and
+.IR to_path .
+.TP
+.B MOVE_MOUNT_F_NO_AUTOMOUNT
+Do not automount any automount points encountered
+while resolving
+.IR from_path .
+This allows a mount object
+that has an automount point at its root
+to be moved
+and prevents unintended triggering of an automount point.
+This flag has no effect
+if the automount point has already been mounted over.
+.TP
+.B MOVE_MOUNT_T_NO_AUTOMOUNT
+As with
+.BR \%MOVE_MOUNT_F_NO_AUTOMOUNT ,
+except operating on
+.I to_dirfd
+and
+.IR to_path .
+This allows an automount point to be manually mounted over.
+.TP
+.BR MOVE_MOUNT_SET_GROUP " (since Linux 5.15)"
+Add the attached private-propagation mount object indicated by
+.I to_dirfd
+and
+.I to_path
+into the mount propagation "peer group"
+of the attached non-private-propagation mount object indicated by
+.I from_dirfd
+and
+.IR from_path .
+.IP
+Unlike other
+.BR move_mount ()
+operations,
+this operation does not move or attach any mount objects.
+Instead, it only updates the metadata
+of attached mount objects.
+(Also, take careful note of
+the argument order\[em]the mount object being modified
+by this operation is the one specified by
+.I to_dirfd
+and
+.IR to_path .)
+.IP
+This makes it possible to first create a mount tree
+consisting only of private mounts
+and then configure the desired propagation layout afterwards.
+(See the "SHARED SUBTREES" section of
+.BR mount_namespaces (7)
+for more information about mount propagation and peer groups.)
+.TP
+.BR MOVE_MOUNT_BENEATH " (since Linux 6.5)"
+If the path indicated by
+.I to_dirfd
+and
+.I to_path
+is an existing mount object,
+rather than attaching or moving the mount object
+indicated by
+.I from_dirfd
+and
+.I from_path
+on top of the mount stack,
+attach or move it beneath the current top mount
+on the mount stack.
+.IP
+After using
+.BR \%MOVE_MOUNT_BENEATH ,
+it is possible to
+.BR umount (2)
+the top mount
+in order to reveal the mount object
+which was attached beneath it earlier.
+This allows for the seamless (and atomic) replacement
+of intricate mount trees,
+which can further be used
+to "upgrade" a mount tree with a newer version.
+.IP
+This operation has several restrictions:
+.RS
+.IP \[bu] 3
+Mount objects cannot be attached beneath the filesystem root,
+including cases where
+the filesystem root was configured by
+.BR chroot (2)
+or
+.BR pivot_root (2).
+To mount beneath the filesystem root,
+.BR pivot_root (2)
+must be used.
+.IP \[bu]
+The target path indicated by
+.I to_dirfd
+and
+.I to_path
+must not be a detached mount object,
+such as those produced by
+.BR open_tree (2)
+with
+.B \%OPEN_TREE_CLONE
+or
+.BR fsmount (2).
+.IP \[bu]
+The current top mount
+of the target path's mount stack
+and its parent mount
+must be in the calling process's mount namespace.
+.IP \[bu]
+The caller must have sufficient privileges
+to unmount the top mount
+of the target path's mount stack,
+to prove they have privileges
+to reveal the underlying mount.
+.IP \[bu]
+Mount propagation events triggered by this
+.BR move_mount ()
+operation
+(as described in
+.BR mount_namespaces (7))
+are calculated based on the parent mount
+of the current top mount
+of the target path's mount stack.
+.IP \[bu]
+The target path's mount
+cannot be an ancestor in the mount tree of
+the source mount object.
+.IP \[bu]
+The source mount object
+must not have any overmounts,
+otherwise it would be possible to create "shadow mounts"
+(i.e., two mounts mounted on the same parent mount at the same mount point).
+.IP \[bu]
+It is not possible to move a mount
+beneath a top mount
+if the parent mount
+of the current top mount
+propagates to the top mount itself.
+Otherwise,
+.B \%MOVE_MOUNT_BENEATH
+would cause the mount object
+to be propagated
+to the top mount
+from the parent mount,
+defeating the purpose of using
+.BR \%MOVE_MOUNT_BENEATH .
+.IP \[bu]
+It is not possible to move a mount
+beneath a top mount
+if the parent mount
+of the current top mount
+propagates to the mount object
+being mounted beneath.
+Otherwise, this would cause a similar propagation issue
+to the previous point,
+also defeating the purpose of using
+.BR \%MOVE_MOUNT_BENEATH .
+.RE
+.RE
+.P
+If
+.I from_dirfd
+is a mount object file descriptor and
+.BR move_mount ()
+is operating on it directly,
+.I from_dirfd
+will remain associated with the mount object after
+.BR move_mount ()
+succeeds,
+so you may repeatedly use
+.I from_dirfd
+with
+.BR move_mount (2)
+and/or "*at()" system calls
+as many times as necessary.
+.SH RETURN VALUE
+On success,
+.BR move_mount ()
+returns 0.
+On error, \-1 is returned, and
+.I errno
+is set to indicate the error.
+.SH ERRORS
+.TP
+.B EACCES
+Search permission is denied
+for one of the directories
+in the path prefix of one of
+.I from_path
+or
+.IR to_path .
+(See also
+.BR path_resolution (7).)
+.TP
+.B EBADF
+One of
+.I from_dirfd
+or
+.I to_dirfd
+is not a valid file descriptor.
+.TP
+.B EFAULT
+One of
+.I from_path
+or
+.I to_path
+is NULL
+or a pointer to a location
+outside the calling process's accessible address space.
+.TP
+.B EINVAL
+Invalid flag specified in
+.IR flags .
+.TP
+.B EINVAL
+The path indicated by
+.I from_dirfd
+and
+.I from_path
+is not a mount object.
+.TP
+.B EINVAL
+The mount object type
+of the source mount object and target inode
+are not compatible
+(i.e., the source is a file but the target is a directory, or vice-versa).
+.TP
+.B EINVAL
+The source mount object or target path
+are not in the calling process's mount namespace
+(or an anonymous mount namespace of the calling process).
+.TP
+.B EINVAL
+The source mount object's parent mount
+has shared mount propagation,
+and thus cannot be moved
+(as described in
+.BR mount_namespaces (7)).
+.TP
+.B EINVAL
+The source mount has
+.B MS_UNBINDABLE
+child mounts
+but the target path
+resides on a mount tree with shared mount propagation,
+which would otherwise cause the unbindable mounts to be propagated
+(as described in
+.BR mount_namespaces (7)).
+.TP
+.B EINVAL
+.B \%MOVE_MOUNT_BENEATH
+was attempted,
+but one of the listed restrictions was violated.
+.TP
+.B ELOOP
+Too many symbolic links encountered
+when resolving one of
+.I from_path
+or
+.IR to_path .
+.TP
+.B ENAMETOOLONG
+One of
+.I from_path
+or
+.I to_path
+is longer than
+.BR PATH_MAX .
+.TP
+.B ENOENT
+A component of one of
+.I from_path
+or
+.I to_path
+does not exist.
+.TP
+.B ENOENT
+One of
+.I from_path
+or
+.I to_path
+is an empty string,
+but the corresponding
+.BI MOVE_MOUNT_ * _EMPTY_PATH
+flag is not specified in
+.IR flags .
+.TP
+.B ENOTDIR
+A component of the path prefix of one of
+.I from_path
+or
+.I to_path
+is not a directory,
+or one of
+.I from_path
+or
+.I to_path
+is relative
+and the corresponding
+.I from_dirfd
+or
+.I to_dirfd
+is a file descriptor referring to a file other than a directory.
+.TP
+.B ENOMEM
+The kernel could not allocate sufficient memory to complete the operation.
+.TP
+.B EPERM
+The calling process does not have the required
+.B \%CAP_SYS_ADMIN
+capability.
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.2.
+.\" commit 2db154b3ea8e14b04fee23e3fdfd5e9d17fbc6ae
+.\" commit 400913252d09f9cfb8cce33daee43167921fc343
+glibc 2.36.
+.SH EXAMPLES
+.BR move_mount ()
+can be used to move attached mounts like the following:
+.P
+.in +4n
+.EX
+move_mount(AT_FDCWD, "/a", AT_FDCWD, "/b", 0);
+.EE
+.in
+.P
+This would move the mount object mounted on
+.I /a
+to
+.IR /b .
+The above procedure is functionally equivalent to
+the following mount operation
+using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/a", "/b", NULL, MS_MOVE, NULL);
+.EE
+.in
+.P
+.BR move_mount ()
+can also be used in conjunction with file descriptors returned from
+.BR open_tree (2)
+or
+.BR open (2):
+.P
+.in +4n
+.EX
+int fd = open_tree(AT_FDCWD, "/mnt", 0); /* or open("/mnt", O_PATH); */
+move_mount(fd, "", AT_FDCWD, "/mnt2", MOVE_MOUNT_F_EMPTY_PATH);
+move_mount(fd, "", AT_FDCWD, "/mnt3", MOVE_MOUNT_F_EMPTY_PATH);
+move_mount(fd, "", AT_FDCWD, "/mnt4", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+This would move the mount object mounted at
+.I /mnt
+to
+.IR /mnt2 ,
+then
+.IR /mnt3 ,
+and then
+.IR /mnt4 .
+.P
+If the source mount object
+indicated by
+.I from_dirfd
+and
+.I from_path
+is a detached mount object,
+.BR move_mount ()
+can be used to attach it to a mount point:
+.P
+.in +4n
+.EX
+int fsfd, mntfd;
+\&
+fsfd = fsopen("ext4", FSOPEN_CLOEXEC);
+fsconfig(fsfd, FSCONFIG_SET_STRING, "source", "/dev/sda1", 0);
+fsconfig(fsfd, FSCONFIG_SET_FLAG, "user_xattr", NULL, 0);
+fsconfig(fsfd, FSCONFIG_CMD_CREATE, NULL, NULL, 0);
+mntfd = fsmount(fsfd, FSMOUNT_CLOEXEC, MOUNT_ATTR_NODEV);
+move_mount(mntfd, "", AT_FDCWD, "/home", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+This would create a new filesystem configuration context for ext4,
+configure it,
+create a mount object,
+and then attach it to
+.IR /home .
+The above procedure is functionally equivalent to
+the following mount operation
+using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/dev/sda1", "/home", "ext4", MS_NODEV, "user_xattr");
+.EE
+.in
+.P
+The same operation also works with detached bind-mounts created with
+.BR open_tree (2)
+with
+.BR OPEN_TREE_CLONE :
+.P
+.in +4n
+.EX
+int mntfd = open_tree(AT_FDCWD, "/home/cyphar", OPEN_TREE_CLONE);
+move_mount(mntfd, "", AT_FDCWD, "/root", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+This would create a new bind-mount of
+.I /home/cyphar
+as attached mount object,
+and then attach it to
+.IR /root .
+The above procedure is functionally equivalent to
+the following mount operation
+using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/home/cyphar", "/root", NULL, MS_BIND, NULL);
+.EE
+.in
+.SH SEE ALSO
+.BR fsconfig (2),
+.BR fsmount (2),
+.BR fsopen (2),
+.BR fspick (2),
+.BR mount (2),
+.BR mount_setattr (2),
+.BR open_tree (2),
+.BR mount_namespaces (7)
+

-- 
2.50.1

Re: [PATCH v3 08/12] man/man2/move_mount.2: document "new" mount API

[Askar Safin]

move_mount for v2 contained this:
> Mounts cannot be moved beneath the rootfs

In v3 you changed this to this:
> Mount objects cannot be attached beneath the filesystem root

You made this phrase worse.

"Filesystem root" can be understood as "root of superblock".
So, please, change this to "root directory" or something.

> This would create a new bind-mount of /home/cyphar as attached mount object, and then attach
You meant "as detached mount object"

Also: I see that you addressed all my v2 comments. Thank you!

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 08/12] man/man2/move_mount.2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 1123 bytes --]

On 2025-08-12, Askar Safin <safinaskar@zohomail.com> wrote:
> move_mount for v2 contained this:
> > Mounts cannot be moved beneath the rootfs
> 
> In v3 you changed this to this:
> > Mount objects cannot be attached beneath the filesystem root
> 
> You made this phrase worse.
> 
> "Filesystem root" can be understood as "root of superblock".
> So, please, change this to "root directory" or something.

Maybe I should borrow the "root mount" terminology from pivot_root(2)?
(Though they use "root mount in the mount namespace of the calling
process", which is a little wordy.) I didn't like using "rootfs" as
shorthand in a man-page.

> > This would create a new bind-mount of /home/cyphar as attached mount object, and then attach
> You meant "as detached mount object"

Thanks, I have already fixed this in my branch (and the two other
misuses of "attach" in fsopen(2)). FWIW, open_tree(2) was the first
man-page in this series that I wrote, so I hadn't settled on the wording
the first draft.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 08/12] man/man2/move_mount.2: document "new" mount API

[Askar Safin]

 ---- On Tue, 12 Aug 2025 18:36:53 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > > "Filesystem root" can be understood as "root of superblock".
 > > So, please, change this to "root directory" or something.
 > 
 > Maybe I should borrow the "root mount" terminology from pivot_root(2)?

I don't like this. For me "root mount" is initial root mount, i. e. initramfs.
It is not what you mean here.

 > I didn't like using "rootfs" as
 > shorthand in a man-page.

I agree.

What you mean by "filesystem root" here? "Thing, which is changed by chroot(2)", right?
Then, please, write "root directory" (or "root"), this is standard term for that thing.
Or you can just write "/".

--
Askar Safin
https://types.pl/@safinaskar

[PATCH v3 09/12] man/man2/open_tree.2: document "new" mount API

[Aleksa Sarai]

This is loosely based on the original documentation written by David
Howells and later maintained by Christian Brauner, but has been
rewritten to be more from a user perspective (as well as fixing a few
critical mistakes).

Co-authored-by: David Howells <dhowells@redhat.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Co-authored-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/open_tree.2 | 471 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 471 insertions(+)

diff --git a/man/man2/open_tree.2 b/man/man2/open_tree.2
new file mode 100644
index 0000000000000000000000000000000000000000..07aac7616107d16d05cc71ba7db6aee35f3a9cc6
--- /dev/null
+++ b/man/man2/open_tree.2
@@ -0,0 +1,471 @@
+.\" Copyright, the authors of the Linux man-pages project
+.\"
+.\" SPDX-License-Identifier: Linux-man-pages-copyleft
+.\"
+.TH open_tree 2 (date) "Linux man-pages (unreleased)"
+.SH NAME
+open_tree \- open path or create detached mount object and attach to fd
+.SH LIBRARY
+Standard C library
+.RI ( libc ,\~ \-lc )
+.SH SYNOPSIS
+.nf
+.BR "#include <fcntl.h>" "          /* Definition of " AT_* " constants */"
+.B #include <sys/mount.h>
+.P
+.BI "int open_tree(int " dirfd ", const char *" path ", unsigned int " flags ");"
+.fi
+.SH DESCRIPTION
+The
+.BR open_tree ()
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.IP \[bu] 3
+If
+.I flags
+contains
+.BR \%OPEN_TREE_CLONE ,
+.BR open_tree ()
+creates a detached mount object
+which consists of a bind-mount of
+the path specified by the
+.IR path .
+A new file descriptor
+associated with the detached mount object
+is then returned.
+The mount object is equivalent to a bind-mount
+that would be created by
+.BR mount (2)
+called with
+.BR MS_BIND ,
+except that it is tied to a file descriptor
+and is not mounted onto the filesystem.
+.IP
+As with file descriptors returned from
+.BR fsmount (2),
+the resultant file descriptor can then be used with
+.BR move_mount (2),
+.BR mount_setattr (2),
+or other such system calls to do further mount operations.
+This mount object will be unmounted and destroyed
+when the file descriptor is closed
+if it was not otherwise attached to a mount point
+by calling
+.BR move_mount (2).
+.IP \[bu]
+If
+.I flags
+does not contain
+.BR \%OPEN_TREE_CLONE ,
+.BR open_tree ()
+returns a file descriptor
+that is exactly equivalent to
+one produced by
+.BR openat (2)
+when called with the same
+.I dirfd
+and
+.IR path .
+.P
+In either case, the resultant file descriptor
+acts the same as one produced by
+.BR open (2)
+with
+.BR O_PATH ,
+meaning it can also be used as a
+.I dirfd
+argument to
+"*at()" system calls.
+.P
+As with "*at()" system calls,
+.BR open_tree ()
+uses the
+.I dirfd
+argument in conjunction with the
+.I path
+argument to determine the path to operate on, as follows:
+.IP \[bu] 3
+If the pathname given in
+.I path
+is absolute, then
+.I dirfd
+is ignored.
+.IP \[bu]
+If the pathname given in
+.I path
+is relative and
+.I dirfd
+is the special value
+.BR \%AT_FDCWD ,
+then
+.I path
+is interpreted relative to
+the current working directory
+of the calling process (like
+.BR open (2)).
+.IP \[bu]
+If the pathname given in
+.I path
+is relative,
+then it is interpreted relative to
+the directory referred to by the file descriptor
+.I dirfd
+(rather than relative to
+the current working directory
+of the calling process,
+as is done by
+.BR open (2)
+for a relative pathname).
+In this case,
+.I dirfd
+must be a directory
+that was opened for reading
+.RB ( O_RDONLY )
+or using the
+.B O_PATH
+flag.
+.IP \[bu]
+If
+.I path
+is an empty string,
+and
+.I flags
+contains
+.BR \%AT_EMPTY_PATH ,
+then the file descriptor
+.I dirfd
+is operated on directly.
+In this case,
+.I dirfd
+may refer to any type of file,
+not just a directory.
+.P
+.I flags
+can be used to control aspects of the path lookup
+and properties of the returned file descriptor.
+A value for
+.I flags
+is constructed by bitwise ORing
+zero or more of the following constants:
+.RS
+.TP
+.B AT_EMPTY_PATH
+If
+.I path
+is an empty string, operate on the file referred to by
+.I dirfd
+(which may have been obtained from
+.BR open (2),
+.BR fsmount(2),
+or from another
+.BR open_tree ()
+call).
+In this case,
+.I dirfd
+may refer to any type of file, not just a directory.
+If
+.I dirfd
+is
+.BR \%AT_FDCWD ,
+.BR open_tree ()
+will operate on the current working directory
+of the calling process.
+This flag is Linux-specific; define
+.B \%_GNU_SOURCE
+to obtain its definition.
+.TP
+.B AT_NO_AUTOMOUNT
+Do not automount any automount points encountered
+while resolving
+.IR path .
+This allows you to create a handle to the automount point itself,
+rather than the location it would mount.
+This flag has no effect if the mount point has already been mounted over.
+This flag is Linux-specific; define
+.B \%_GNU_SOURCE
+to obtain its definition.
+.TP
+.B AT_SYMLINK_NOFOLLOW
+If
+.I path
+is a symbolic link, do not dereference it; instead,
+create either a handle to the link itself
+or a bind-mount of it.
+The resultant file descriptor is indistinguishable from one produced by
+.BR openat (2)
+with
+.BR \%O_PATH | O_NOFOLLLOW .
+.TP
+.B OPEN_TREE_CLOEXEC
+Set the close-on-exec
+.RB ( FD_CLOEXEC )
+flag on the new file descriptor.
+See the description of the
+.B O_CLOEXEC
+flag in
+.BR open (2)
+for reasons why this may be useful.
+.TP
+.B OPEN_TREE_CLONE
+Rather than creating an
+.BR openat (2)-style
+.B O_PATH
+file descriptor,
+create a bind-mount of
+.I path
+(akin to
+.IR "mount --bind" )
+as a detached mount object.
+In order to do this operation,
+the calling process must have the
+.BR \%CAP_SYS_ADMIN
+capability.
+.TP
+.B AT_RECURSIVE
+Create a recursive bind-mount of the path
+(akin to
+.IR "mount --rbind" )
+as a detached mount object.
+This flag is only permitted in conjunction with
+.BR \%OPEN_TREE_CLONE .
+.SH RETURN VALUE
+On success, a new file descriptor is returned.
+On error, \-1 is returned, and
+.I errno
+is set to indicate the error.
+.SH ERRORS
+.TP
+.B EACCES
+Search permission is denied for one of the directories
+in the path prefix of
+.IR path .
+(See also
+.BR path_resolution (7).)
+.TP
+.B EBADF
+.I path
+is relative but
+.I dirfd
+is neither
+.B \%AT_FDCWD
+nor a valid file descriptor.
+.TP
+.B EFAULT
+.I path
+is NULL
+or a pointer to a location
+outside the calling process's accessible address space.
+.TP
+.B EINVAL
+Invalid flag specified in
+.IR flags .
+.TP
+.B ELOOP
+Too many symbolic links encountered when resolving
+.IR path .
+.TP
+.B EMFILE
+The calling process has too many open files to create more.
+.TP
+.B ENAMETOOLONG
+.I path
+is longer than
+.BR PATH_MAX .
+.TP
+.B ENFILE
+The system has too many open files to create more.
+.TP
+.B ENOENT
+A component of
+.I path
+does not exist, or is a dangling symbolic link.
+.TP
+.B ENOENT
+.I path
+is an empty string, but
+.B AT_EMPTY_PATH
+is not specified in
+.IR flags .
+.TP
+.B ENOTDIR
+A component of the path prefix of
+.I path
+is not a directory, or
+.I path
+is relative and
+.I dirfd
+is a file descriptor referring to a file other than a directory.
+.TP
+.B ENOSPC
+The "anonymous" mount namespace
+necessary to contain the
+.B \%OPEN_TREE_CLONE
+detached bind-mount mount object
+could not be allocated,
+as doing so would exceed
+the configured per-user limit on
+the number of mount namespaces in the current user namespace.
+(See also
+.BR namespaces (7).)
+.TP
+.B ENOMEM
+The kernel could not allocate sufficient memory to complete the operation.
+.TP
+.B EPERM
+.I flags
+contains
+.B \%OPEN_TREE_CLONE
+but the calling process does not have the required
+.B CAP_SYS_ADMIN
+capability.
+.SH STANDARDS
+Linux.
+.SH HISTORY
+Linux 5.2.
+.\" commit a07b20004793d8926f78d63eb5980559f7813404
+.\" commit 400913252d09f9cfb8cce33daee43167921fc343
+glibc 2.36.
+.SH NOTES
+.SS Anonymous mount namespaces
+The bind-mount mount objects created by
+.BR open_tree ()
+with
+.B \%OPEN_TREE_CLONE
+are not attached to the mount namespace of the calling process.
+Instead, each mount object is attached to
+a newly allocated "anonymous" mount namespace
+associated with the calling process.
+.P
+One of the side-effects of this is that
+(unlike bind-mounts created with
+.BR mount (2)),
+mount propagation
+(as described in
+.BR mount_namespaces (7))
+will not be applied to bind-mounts created by
+.BR open_tree ()
+until the bind-mount is attached with
+.BR move_mount (2),
+at which point the mount
+will be associated with the mount namespace
+where it was mounted
+and mount propagation will resume.
+.SH EXAMPLES
+The following examples show how
+.BR open_tree ()
+can be used in place of more traditional
+.BR mount (2)
+calls with
+.BR MS_BIND .
+.P
+.in +4n
+.EX
+int srcfd = open_tree(AT_FDCWD, "/var", OPEN_TREE_CLONE);
+move_mount(srcfd, "", AT_FDCWD, "/mnt", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+First,
+a detached bind-mount mount object of
+.I /var
+is created and attached to the file descriptor
+.IR srcfd .
+Then, the mount object is attached to
+.I /mnt
+using
+.BR move_mount (2)
+with
+.B \%MOVE_MOUNT_F_EMPTY_PATH
+to request that the detached mount object attached to the file descriptor
+.I srcfd
+be moved (and thus attached) to
+.IR /mnt .
+.P
+The above procedure is functionally equivalent to
+the following mount operation using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/var", "/mnt", NULL, MS_BIND, NULL);
+.EE
+.in
+.P
+.B \%OPEN_TREE_CLONE
+can be combined with
+.B \%AT_RECURSIVE
+to create recursive detached bind-mount mount objects,
+which in turn can be attached to mount points
+to create recursive bind-mounts.
+.P
+.in +4n
+.EX
+int srcfd = open_tree(AT_FDCWD, "/var", OPEN_TREE_CLONE | AT_RECURSIVE);
+move_mount(srcfd, "", AT_FDCWD, "/mnt", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+The above procedure is functionally equivalent to
+the following mount operation using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/var", "/mnt", NULL, MS_BIND | MS_REC, NULL);
+.EE
+.in
+.P
+One of the primary benefits of using
+.BR open_tree ()
+and
+.BR move_mount (2)
+over the traditional
+.BR mount (2)
+is that operating with
+.IR dirfd -style
+file descriptors is far easier and more intuitive.
+.P
+.in +4n
+.EX
+int srcfd = open_tree(100, "", AT_EMPTY_PATH | OPEN_TREE_CLONE);
+move_mount(srcfd, "", 200, "foo", MOVE_MOUNT_F_EMPTY_PATH);
+.EE
+.in
+.P
+The above procedure is roughly equivalent to
+the following mount operation using
+.BR mount (2):
+.P
+.in +4n
+.EX
+mount("/proc/self/fd/100", "/proc/self/fd/200/foo", NULL, MS_BIND, NULL);
+.EE
+.in
+.P
+In addition, you can use the file descriptor returned by
+.BR open_tree ()
+as the
+.I dirfd
+argument to any "*at()" system calls:
+.P
+.in +4n
+.EX
+int dirfd, fd;
+\&
+dirfd = open_tree(AT_FDCWD, "/etc", OPEN_TREE_CLONE);
+fd = openat(dirfd, "passwd", O_RDONLY);
+fchmodat(dirfd, "shadow", 0000, 0);
+close(dirfd);
+close(fd);
+/* The bind-mount is now destroyed. */
+.EE
+.in
+.SH SEE ALSO
+.BR fsconfig (2),
+.BR fsmount (2),
+.BR fsopen (2),
+.BR fspick (2),
+.BR mount (2),
+.BR mount_setattr (2),
+.BR move_mount (2),
+.BR mount_namespaces (7)

-- 
2.50.1

Re: [PATCH v3 09/12] man/man2/open_tree.2: document "new" mount API

[Askar Safin]

man open_tree says:
> mount propagation
> (as described in
> .BR mount_namespaces (7))
> will not be applied to bind-mounts created by
> .BR open_tree ()
> until the bind-mount is attached with
> .BR move_mount (2),

It seems this is wrong, because this commit exists: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=06b1ce966e3f8bfef261c111feb3d4b33ede0cd8 .
I'm not sure about this. (I didn't test this.)



--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 09/12] man/man2/open_tree.2: document "new" mount API

[Christian Brauner]

On Thu, Aug 21, 2025 at 03:27:26PM +0400, Askar Safin wrote:
> man open_tree says:
> > mount propagation
> > (as described in
> > .BR mount_namespaces (7))
> > will not be applied to bind-mounts created by
> > .BR open_tree ()
> > until the bind-mount is attached with
> > .BR move_mount (2),
> 
> It seems this is wrong, because this commit exists: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=06b1ce966e3f8bfef261c111feb3d4b33ede0cd8 .
> I'm not sure about this. (I didn't test this.)

No, it's correct. I reverted this because it broke userspace that relies
on this behavior.

Re: [PATCH v3 09/12] man/man2/open_tree.2: document "new" mount API

[Askar Safin]

 ---- On Sat, 09 Aug 2025 00:39:53 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > +If
 > +.I flags
 > +does not contain
 > +.BR \%OPEN_TREE_CLONE ,
 > +.BR open_tree ()
 > +returns a file descriptor
 > +that is exactly equivalent to
 > +one produced by
 > +.BR openat (2)

This is not true. They differ in handling of automounts.
open_tree follows them in final component (by default),
and openat - not.

See reproducer in the end of this letter.

I suggest merely adding this:
> that is exactly equivalent to one produced by openat(2) (modulo automounts)

--
Askar Safin
https://types.pl/@safinaskar


// Root in initial user namespace

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <stdbool.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sched.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/mount.h>
#include <sys/syscall.h>
#include <linux/openat2.h>

#define MY_ASSERT(cond) do { \
    if (!(cond)) { \
        fprintf (stderr, "%s: assertion failed\n", #cond); \
        exit (1); \
    } \
} while (0)

bool
tracing_mounted (void)
{
    struct statx tracing;
    if (statx (AT_FDCWD, "/tmp/debugfs/tracing", AT_NO_AUTOMOUNT, 0, &tracing) != 0)
        {
            perror ("statx tracing");
            exit (1);
        }
    if (!(tracing.stx_attributes_mask & STATX_ATTR_MOUNT_ROOT))
        {
            fprintf (stderr, "???\n");
            exit (1);
        }
    return tracing.stx_attributes & STATX_ATTR_MOUNT_ROOT;
}

void
mount_debugfs (void)
{
    if (mount (NULL, "/tmp/debugfs", "debugfs", 0, NULL) != 0)
        {
            perror ("mount debugfs");
            exit (1);
        }
    MY_ASSERT (!tracing_mounted ());
}

void
umount_debugfs (void)
{
    umount ("/tmp/debugfs/tracing"); // Ignore errors
    if (umount ("/tmp/debugfs") != 0)
        {
            perror ("umount debugfs");
            exit (1);
        }
}

int
main (void)
{
    // Init
    {
        if (chdir ("/") != 0)
            {
                perror ("chdir /");
                exit (1);
            }
        if (unshare (CLONE_NEWNS) != 0)
            {
                perror ("unshare");
                exit (1);
            }
        if (mount (NULL, "/", NULL, MS_REC | MS_PRIVATE, NULL) != 0)
            {
                perror ("mount(NULL, /, NULL, MS_REC | MS_PRIVATE, NULL)");
                exit (1);
            }
        if (mount (NULL, "/tmp", "tmpfs", 0, NULL) != 0)
            {
                perror ("mount tmpfs");
                exit (1);
            }
    }
    if (mkdir ("/tmp/debugfs", 0777) != 0)
        {
            perror ("mkdir(/tmp/debugfs)");
            exit (1);
        }

    // open(O_PATH) doesn't follow automounts
    {
        mount_debugfs ();
        {
            int fd = open ("/tmp/debugfs/tracing", O_PATH);
            MY_ASSERT (fd >= 0);
            MY_ASSERT (close (fd) == 0);
        }
        MY_ASSERT (!tracing_mounted ());
        umount_debugfs ();
    }

    // open_tree does follow automounts (by default)
    {
        mount_debugfs ();
        {
            int fd = open_tree (AT_FDCWD, "/tmp/debugfs/tracing", 0);
            MY_ASSERT (fd >= 0);
            MY_ASSERT (close (fd) == 0);
        }
        MY_ASSERT (tracing_mounted ());
        umount_debugfs ();
    }

    // open (O_PATH | O_DIRECTORY)
    {
        mount_debugfs ();
        {
            int fd = open ("/tmp/debugfs/tracing", O_PATH | O_DIRECTORY);
            MY_ASSERT (fd >= 0);
            MY_ASSERT (close (fd) == 0);
        }
        MY_ASSERT (tracing_mounted ());
        umount_debugfs ();
    }

    // AT_NO_AUTOMOUNT
    {
        mount_debugfs ();
        {
            int fd = open_tree (AT_FDCWD, "/tmp/debugfs/tracing", AT_NO_AUTOMOUNT);
            MY_ASSERT (fd >= 0);
            MY_ASSERT (close (fd) == 0);
        }
        MY_ASSERT (!tracing_mounted ());
        umount_debugfs ();
    }

    printf ("All tests passed\n");
    exit (0);
}

[PATCH v3 10/12] man/man2/mount_setattr.2: mirror opening sentence from fsopen(2)

[Aleksa Sarai]

All of the other new mount API docs have this lead-in sentence in order
to make this set of APIs feel a little bit more cohesive.  Despite being
a bit of a latecomer, mount_setattr(2) is definitely part of this family
of APIs and so deserves the same treatment.

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/mount_setattr.2 | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/man/man2/mount_setattr.2 b/man/man2/mount_setattr.2
index 46fcba927dd8c0959c898b9ba790ae298f514398..d98e7d70870c082144dfa47e31ddf091c8545e4f 100644
--- a/man/man2/mount_setattr.2
+++ b/man/man2/mount_setattr.2
@@ -19,7 +19,11 @@ .SH SYNOPSIS
 .SH DESCRIPTION
 The
 .BR mount_setattr ()
-system call changes the mount properties of a mount or an entire mount tree.
+system call is part of
+the suite of file descriptor based mount facilities in Linux.
+.P
+.BR mount_setattr ()
+changes the mount properties of a mount or an entire mount tree.
 If
 .I path
 is relative,

-- 
2.50.1

[PATCH v3 11/12] man/man2/open_tree{,_attr}.2: document new open_tree_attr() API

[Aleksa Sarai]

This is a new API added in Linux 6.15, and is effectively just a minor
expansion of open_tree(2) in order to allow for MOUNT_ATTR_IDMAP to be
changed for an existing ID-mapped mount.  glibc does not yet have a
wrapper for this.

Cc: Christian Brauner <brauner@kernel.org>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/open_tree.2      | 122 ++++++++++++++++++++++++++++++++++++++++++++++
 man/man2/open_tree_attr.2 |   1 +
 2 files changed, 123 insertions(+)

diff --git a/man/man2/open_tree.2 b/man/man2/open_tree.2
index 07aac7616107d16d05cc71ba7db6aee35f3a9cc6..f671c401b1f1e2926f7948d6fe5041b7848125ac 100644
--- a/man/man2/open_tree.2
+++ b/man/man2/open_tree.2
@@ -14,7 +14,19 @@ .SH SYNOPSIS
 .B #include <sys/mount.h>
 .P
 .BI "int open_tree(int " dirfd ", const char *" path ", unsigned int " flags ");"
+.P
+.BR "#include <sys/syscall.h>" "    /* Definition of " SYS_* " constants */"
+.P
+.BI "int syscall(SYS_open_tree_attr, int " dirfd ", const char *" path ","
+.BI "            unsigned int " flags ", struct mount_attr *" attr ", \
+size_t " size ");"
 .fi
+.P
+.IR Note :
+glibc provides no wrapper for
+.BR open_tree_attr (),
+necessitating the use of
+.BR syscall (2).
 .SH DESCRIPTION
 The
 .BR open_tree ()
@@ -228,6 +240,111 @@ .SH DESCRIPTION
 as a detached mount object.
 This flag is only permitted in conjunction with
 .BR \%OPEN_TREE_CLONE .
+.SS open_tree_attr()
+The
+.BR open_tree_attr ()
+system call operates in exactly the same way as
+.BR open_tree (),
+except for the differences described here.
+.P
+After performing the same operation as with
+.BR open_tree (),
+(before returning the resulting file descriptor)
+.BR open_tree_attr ()
+will apply the mount attribute changes described in
+.I attr
+to the returned file descriptor.
+(See
+.BR mount_attr (2type)
+for a description of the
+.I mount_attr
+structure.
+As described in
+.BR mount_setattr (2),
+.I size
+must be set to
+.I sizeof(struct mount_attr)
+in order to support future extensions.)
+.P
+The application of
+.I attr
+to the resultant file descriptor
+has identical semantics to
+.BR mount_setattr (2),
+except for the following extensions and general caveats:
+.IP \[bu] 3
+Unlike
+.BR mount_setattr (2)
+called with a regular
+.B OPEN_TREE_CLONE
+detached mount object from
+.BR open_tree (),
+.BR open_tree_attr ()
+can specify a different setting for
+.B \%MOUNT_ATTR_IDMAP
+to the original mount object cloned with
+.BR OPEN_TREE_CLONE .
+.IP
+Adding
+.B \%MOUNT_ATTR_IDMAP
+to
+.I attr.attr_clr
+will disable ID-mapping for the new mount object;
+adding
+.B \%MOUNT_ATTR_IDMAP
+to
+.I attr.attr_set
+will configure the mount object to have the ID-mapping defined by
+the user namespace referenced by the file descriptor
+.IR attr.userns_fd .
+(The semantics of which are identical to when
+.BR mount_setattr (2)
+is used to configure
+.BR \%MOUNT_ATTR_IDMAP .)
+.IP
+Changing or removing the mapping
+of an ID-mapped mount is only permitted
+if a new detached mount object is being created with
+.I flags
+including
+.BR \%OPEN_TREE_CLONE .
+.IP \[bu]
+If
+.I flags
+contains
+.BR \%AT_RECURSIVE ,
+then the attributes described in
+.I attr
+are applied recursively
+(just as when
+.BR mount_setattr (2)
+is called with
+.BR \%AT_RECURSIVE ).
+However, this applies in addition to the
+.BR open_tree ()-specific
+behaviour regarding
+.BR \%AT_RECURSIVE ,
+and thus
+.I flags
+must also contain
+.BR \% OPEN_TREE_CLONE .
+.P
+Note that if
+.I flags
+does not contain
+.BR \% OPEN_TREE_CLONE ,
+.BR open_tree_attr ()
+will attempt to modify the mount attributes of
+the mount object attached at
+the path described by
+.I dirfd
+and
+.IR path .
+As with
+.BR mount_setattr (2),
+if said path is not a mount point,
+.BR open_tree_attr ()
+will return an error.
 .SH RETURN VALUE
 On success, a new file descriptor is returned.
 On error, \-1 is returned, and
@@ -321,10 +438,15 @@ .SH ERRORS
 .SH STANDARDS
 Linux.
 .SH HISTORY
+.SS open_tree()
 Linux 5.2.
 .\" commit a07b20004793d8926f78d63eb5980559f7813404
 .\" commit 400913252d09f9cfb8cce33daee43167921fc343
 glibc 2.36.
+.SS open_tree_attr()
+Linux 6.15.
+.\" commit c4a16820d90199409c9bf01c4f794e1e9e8d8fd8
+.\" commit 7a54947e727b6df840780a66c970395ed9734ebe
 .SH NOTES
 .SS Anonymous mount namespaces
 The bind-mount mount objects created by
diff --git a/man/man2/open_tree_attr.2 b/man/man2/open_tree_attr.2
new file mode 100644
index 0000000000000000000000000000000000000000..e57269bbd269bcce0b0a974425644ba75e379f2f
--- /dev/null
+++ b/man/man2/open_tree_attr.2
@@ -0,0 +1 @@
+.so man2/open_tree.2

-- 
2.50.1

[PATCH v3 12/12] man/man2/{fsconfig,mount_setattr}.2: add note about attribute-parameter distinction

[Aleksa Sarai]

This was not particularly well documented in mount(8) nor mount(2), and
since this is a fairly notable aspect of the new mount API, we should
probably add some words about it.

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 man/man2/fsconfig.2      | 11 +++++++++++
 man/man2/mount_setattr.2 | 37 +++++++++++++++++++++++++++++++++++++
 2 files changed, 48 insertions(+)

diff --git a/man/man2/fsconfig.2 b/man/man2/fsconfig.2
index 97c9aff0e0c195e6028e1c7bd70e40905ba9f994..a7642e1633541bf8f5cd537db22987a4ec70da06 100644
--- a/man/man2/fsconfig.2
+++ b/man/man2/fsconfig.2
@@ -522,6 +522,17 @@ .SS Generic filesystem parameters
 Linux Security Modules (LSMs)
 are also generic with respect to the underlying filesystem.
 See the documentation for the LSM you wish to configure for more details.
+.SS Mount attributes and filesystem parameters
+Some filesystem parameters
+(traditionally associated with
+.BR mount (8)-style
+options)
+are also mount attributes.
+.P
+For a description of the distinction between
+mount attributes and filesystem parameters,
+see the "Mount attributes and filesystem parameters" subsection of
+.BR mount_setattr (2).
 .SH CAVEATS
 .SS Filesystem parameter types
 As a result of
diff --git a/man/man2/mount_setattr.2 b/man/man2/mount_setattr.2
index d98e7d70870c082144dfa47e31ddf091c8545e4f..2927b012eed1569e0d78a2fb91815f364fca124d 100644
--- a/man/man2/mount_setattr.2
+++ b/man/man2/mount_setattr.2
@@ -790,6 +790,43 @@ .SS ID-mapped mounts
 .BR chown (2)
 system call changes the ownership globally and permanently.
 .\"
+.SS Mount attributes and filesystem parameters
+Some mount attributes
+(traditionally associated with
+.BR mount (8)-style
+options)
+are also filesystem parameters.
+For example, the
+.I -o ro
+option to
+.BR mount (8)
+can refer to the
+"read-only" filesystem parameter,
+or the "read-only" mount attribute.
+.P
+The distinction between these two kinds of option is that
+mount object attributes are applied per-mount-object
+(allowing different mount objects
+derived from a given filesystem instance
+to have different attributes),
+while filesystem instance parameters
+("superblock flags" in kernel-developer parlance)
+apply to all mount objects
+derived from the same filesystem instance.
+.P
+When using
+.BR mount (2),
+the line between these two types of mount options was blurred.
+However, with
+.BR mount_setattr ()
+and
+.BR fsconfig (2),
+the distinction is made much clearer.
+Mount attributes are configured with
+.BR mount_setattr (),
+while filesystem parameters can be configured using
+.BR fsconfig (2).
+.\"
 .SS Extensibility
 In order to allow for future extensibility,
 .BR mount_setattr ()

-- 
2.50.1

Re: [PATCH v3 00/12] man2: document "new" mount API

[Askar Safin]

I plan to do a lot of testing of "new" mount API on my computer.
It is quiet possible that I will find some bugs in these manpages during testing.
(I already found some, but I'm not sure.)
I think this will take 3-7 days.
So, Alejandro Colomar, please, don't merge this patchset until then.

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 00/12] man2: document "new" mount API

[Alejandro Colomar]

[-- Attachment #1: Type: text/plain, Size: 461 bytes --]

Hi Askar,

On Sat, Aug 09, 2025 at 07:04:06PM +0400, Askar Safin wrote:
> I plan to do a lot of testing of "new" mount API on my computer.
> It is quiet possible that I will find some bugs in these manpages during testing.
> (I already found some, but I'm not sure.)
> I think this will take 3-7 days.
> So, Alejandro Colomar, please, don't merge this patchset until then.

Sure, thanks!


Cheers,
Alex

-- 
<https://www.alejandro-colomar.es/>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 1392 bytes --]

On 2025-08-09, Askar Safin <safinaskar@zohomail.com> wrote:
> I plan to do a lot of testing of "new" mount API on my computer.
> It is quiet possible that I will find some bugs in these manpages during testing.
> (I already found some, but I'm not sure.)
> I think this will take 3-7 days.
> So, Alejandro Colomar, please, don't merge this patchset until then.

I don't plan to work on this again for the next week at least (I've
already spent over a week on these docs -- writing, rewriting, and then
rewriting once more for good measure; I've started seeing groff in my
nightmares...), so I will go through review comments after you're done.

There are some rough edges on these APIs I found while writing these
docs, so I plan to fix those this cycle if possible (hopefully those
aren't the bugs you said you found in the docs). Two of the fixes have
already been merged in the vfs tree for 6.18 (the -ENODATA handling bug,
as well as a bug in open_tree_attr() that would've let userspace trigger
UAFs). (Once 6.18 is out, I will send a follow-up patchset to document
the fixes.)

FYI, I've already fixed the few ".BR \% FOO" typos. (My terminal font
doesn't have a bold typeface, so when reviewing the rendered man-pages,
mistakes involving .B are hard to spot.)

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Alejandro Colomar]

[-- Attachment #1: Type: text/plain, Size: 2154 bytes --]

Hi Aleksa,

On Sun, Aug 10, 2025 at 03:32:25AM +1000, Aleksa Sarai wrote:
> On 2025-08-09, Askar Safin <safinaskar@zohomail.com> wrote:
> > I plan to do a lot of testing of "new" mount API on my computer.
> > It is quiet possible that I will find some bugs in these manpages during testing.
> > (I already found some, but I'm not sure.)
> > I think this will take 3-7 days.
> > So, Alejandro Colomar, please, don't merge this patchset until then.
> 
> I don't plan to work on this again for the next week at least (I've
> already spent over a week on these docs -- writing, rewriting, and then
> rewriting once more for good measure; I've started seeing groff in my
> nightmares...), so I will go through review comments after you're done.
>
> There are some rough edges on these APIs I found while writing these
> docs, so I plan to fix those this cycle if possible (hopefully those
> aren't the bugs you said you found in the docs). Two of the fixes have
> already been merged in the vfs tree for 6.18 (the -ENODATA handling bug,
> as well as a bug in open_tree_attr() that would've let userspace trigger
> UAFs). (Once 6.18 is out, I will send a follow-up patchset to document
> the fixes.)
> 
> FYI, I've already fixed the few ".BR \% FOO" typos. (My terminal font
> doesn't have a bold typeface, so when reviewing the rendered man-pages,
> mistakes involving .B are hard to spot.)

You can review in PDF if you want.  See the pdfman(1) script under
src/bin/.  It's quite portable:

	$ cat src/bin/pdfman 
	#!/bin/bash
	#
	# Copyright, the authors of the Linux man-pages project
	# SPDX-License-Identifier: GPL-3.0-or-later

	set -Eeuo pipefail;
	shopt -s lastpipe;

	printf '%s\n' "${!#}.XXXXXX" \
	| sed 's,.*/,,' \
	| xargs mktemp -t \
	| read -r tmp;

	man -Tpdf "$@" >"$tmp";
	xdg-open "$tmp";

It works essentially like man(1), so you can pass any man(7) file as its
argument to read it as a PDF.

(You may or may not have it available in your system, if your distro
 packages a recent enough version of the project.)


Have a lovely day!
Alex

-- 
<https://www.alejandro-colomar.es/>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Askar Safin]

I noticed that you changed docs for automounts.
So I dig into automounts implementation.
And I found a bug in openat2.
If RESOLVE_NO_XDEV is specified, then name resolution
doesn't cross automount points (i. e. we get EXDEV),
but automounts still happen!
I think this is a bug.
Bug is reproduced in 6.17-rc1.
In the end of this mail you will find reproducer.
And miniconfig.

If you send patches for this bug, please, CC me.

Are automounts actually used? Is it possible to deprecate or
remove them? It seems for me automounts are rarely tested obscure
feature, which affects core namei code.

This reproducer is based on "tracing" automount, which
actually *IS* already deprecated. But automount mechanism
itself is not deprecated, as well as I know.

Also, I did read namei code, and I think that
options AT_NO_AUTOMOUNT, FSPICK_NO_AUTOMOUNT, etc affect
last component only, not all of them. I didn't test this yet.
I plan to test this within next days.

Also, I still didn't finish my experiments. Hopefully I will
finish them in 7 days. :)

Askar Safin

====

miniconfig:

CONFIG_64BIT=y

CONFIG_EXPERT=y

CONFIG_PRINTK=y
CONFIG_PRINTK_TIME=y

CONFIG_TTY=y
CONFIG_VT=y
CONFIG_VT_CONSOLE=y
CONFIG_FRAMEBUFFER_CONSOLE=y

CONFIG_PROC_FS=y
CONFIG_DEVTMPFS=y
CONFIG_SYSFS=y
CONFIG_TMPFS=y
CONFIG_DEBUG_FS=y
CONFIG_USER_EVENTS=y
CONFIG_FTRACE=y
CONFIG_MULTIUSER=y
CONFIG_NAMESPACES=y
CONFIG_USER_NS=y
CONFIG_PID_NS=y


CONFIG_SERIAL_8250=y
CONFIG_SERIAL_8250_CONSOLE=y

CONFIG_BLK_DEV_INITRD=y
CONFIG_RD_GZIP=y

CONFIG_BINFMT_ELF=y
CONFIG_BINFMT_SCRIPT=y

CONFIG_TRACEFS_AUTOMOUNT_DEPRECATED=y

CONFIG_DEBUG_INFO_DWARF_TOOLCHAIN_DEFAULT=y

====

/*
Author: Askar Safin
Public domain

Make sure your kernel is compiled with CONFIG_TRACEFS_AUTOMOUNT_DEPRECATED=y

If that openat2 bug reproduces, then this program will
print "BUG REPRODUCED". If openat2 is fixed, then
the program will print "BUG NOT REPRODUCED".
Any other output means that something gone wrong,
i. e. results are indeterminate.

This program requires root in initial user namespace
*/

#define _GNU_SOURCE
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sched.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/mount.h>
#include <sys/syscall.h>
#include <linux/openat2.h>

int
main (void)
{
    if (unshare (CLONE_NEWNS) != 0)
        {
            perror ("unshare");
            return 1;
        }
    if (mount (NULL, "/", NULL, MS_REC | MS_PRIVATE, NULL) != 0)
        {
            perror ("mount(NULL, /, NULL, MS_REC | MS_PRIVATE, NULL)");
            return 1;
        }
    if (mount (NULL, "/tmp", "tmpfs", 0, NULL) != 0)
        {
            perror ("mount tmpfs");
            return 1;
        }
    if (mkdir ("/tmp/debugfs", 0777) != 0)
        {
            perror ("mkdir(/tmp/debugfs)");
            return 1;
        }
    if (mount (NULL, "/tmp/debugfs", "debugfs", 0, NULL) != 0)
        {
            perror ("mount debugfs");
            return 1;
        }
    {
        struct statx tracing;
        if (statx (AT_FDCWD, "/tmp/debugfs/tracing", AT_NO_AUTOMOUNT, 0, &tracing) != 0)
            {
                perror ("statx tracing");
                return 1;
            }
        if (!(tracing.stx_attributes_mask & STATX_ATTR_MOUNT_ROOT))
            {
                fprintf (stderr, "???\n");
                return 1;
            }
        // Let's check that nothing is mounted at /tmp/debugfs/tracing yet
        if (tracing.stx_attributes & STATX_ATTR_MOUNT_ROOT)
            {
                fprintf (stderr, "Something already mounted at /tmp/debugfs/tracing\n");
                return 1;
            }
    }
    if (chdir ("/tmp/debugfs") != 0)
        {
            perror ("chdir");
            return 1;
        }
    {
        struct open_how how;
        memset (&how, 0, sizeof how);
        how.flags = O_DIRECTORY;
        how.mode = 0;
        how.resolve = RESOLVE_NO_XDEV | RESOLVE_NO_MAGICLINKS | RESOLVE_NO_SYMLINKS;
        if (syscall (SYS_openat2, AT_FDCWD, "tracing", &how, sizeof how) != -1)
            {
                fprintf (stderr, "openat2 crossed automount point");
                return 1;
            }
        if (errno != EXDEV)
            {
                fprintf (stderr, "wrong errno");
                return 1;
            }
    }
    {
        struct statx tracing;
        if (statx (AT_FDCWD, "/tmp/debugfs/tracing", AT_NO_AUTOMOUNT, 0, &tracing) != 0)
            {
                perror ("statx tracing (2)");
                return 1;
            }
        if (!(tracing.stx_attributes_mask & STATX_ATTR_MOUNT_ROOT))
            {
                fprintf (stderr, "???\n");
                return 1;
            }
        if (tracing.stx_attributes & STATX_ATTR_MOUNT_ROOT)
            {
                fprintf (stderr, "BUG REPRODUCED. Something mounted at /tmp/debugfs/tracing\n");
                return 0;
            }
        else
            {
                fprintf (stderr, "BUG NOT REPRODUCED\n");
                return 0;
            }
    }
}

Re: [PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 2946 bytes --]

On 2025-08-17, Askar Safin <safinaskar@zohomail.com> wrote:
> I noticed that you changed docs for automounts. So I dig into
> automounts implementation. And I found a bug in openat2. If
> RESOLVE_NO_XDEV is specified, then name resolution doesn't cross
> automount points (i. e. we get EXDEV), but automounts still happen! I
> think this is a bug. Bug is reproduced in 6.17-rc1. In the end of this
> mail you will find reproducer. And miniconfig.

Yes, this is a bug -- we check LOOKUP_NO_XDEV after traverse_mounts()
because we want to error out if we actually jumped to a different mount.
We should probably be erroring out in follow_automount() as well, and I
missed this when I wrote openat2().

openat2() also really needs RESOLVE_NO_AUTOMOUNT (and probably
RESOLVE_NO_DOTDOT as well as some other small features). I'll try to
send something soon.

> Are automounts actually used? Is it possible to deprecate or
> remove them? It seems for me automounts are rarely tested obscure
> feature, which affects core namei code.

I use them for auto-mounting NFS shares on my laptop, and I'm sure there
are plenty of other users. They are little bit funky but I highly doubt
they are "unused". Howells probably disagrees in even stronger terms.
Most distributions provide autofs as a supported package (I think it
even comes pre-installed for some distros).

They are not tested by fstests AFAICS, but that's more of a flaw in
fstests (automount requires you to have a running autofs daemon, which
probably makes testing it in fstests or selftests impractical) not the
feature itself.

> This reproducer is based on "tracing" automount, which
> actually *IS* already deprecated. But automount mechanism
> itself is not deprecated, as well as I know.

The automount behaviour of tracefs is different to the general automount
mechanism which is managed by userspace with the autofs daemon. I don't
know the history behind the deprecation, but I expect that it was
deprecated in favour of configuring it with autofs (or just enabling it
by default).

> Also, I did read namei code, and I think that
> options AT_NO_AUTOMOUNT, FSPICK_NO_AUTOMOUNT, etc affect
> last component only, not all of them. I didn't test this yet.
> I plan to test this within next days.

No, LOOKUP_AUTOMOUNT affects all components. I double-checked this with
Christian.

You would think that it's only the last component (like O_DIRECTORY,
O_NOFOLLOW, AT_SYMLINK_{,NO}FOLLOW) but follow_automount() is called for
all components (i.e., as part of step_into()). It hooks into the regular
lookup flow for mountpoints.

Yes, it is quite funky that AT_NO_AUTOMOUNT is the only AT_* flag that
works this way -- hence why I went with a different RESOLVE_* namespace
for openat2() (which _always_ act on _all_ components).

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Askar Safin]

I just sent to fsdevel fix for that RESOLVE_NO_XDEV bug.

Aleksa Sarai <cyphar@cyphar.com>:
> No, LOOKUP_AUTOMOUNT affects all components. I double-checked this with
> Christian.

No. I just tested this. See tests (and miniconfig) in the end of this message.

statx always follows automounts in non-final components no matter what.
I tested this. And it follows automounts in final component depending on
AT_NO_AUTOMOUNT. I tested this too. Also, absolutely all other syscalls always
follow automounts in non-final components no matter what. With sole exception
for openat2 with RESOLVE_NO_XDEV. I didn't test this, but I conclude this
by reading code.

First of all, LOOKUP_PARENT's doc in kernel currently is wrong:
https://elixir.bootlin.com/linux/v6.17-rc1/source/include/linux/namei.h#L31

We see there:
#define LOOKUP_PARENT    BIT(10)    /* Looking up final parent in path */

This is not true. LOOKUP_PARENT means that we are resolving any non-final
component. LOOKUP_PARENT is set when we enter link_path_walk, which
is used for resolving everything except for final component.
And LOOKUP_PARENT is cleared when we leave link_path_walk.

Now let's look here:
https://elixir.bootlin.com/linux/v6.17-rc1/source/fs/namei.c#L1447

	if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
			   LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&

We never return -EISDIR in this "if" if we are in non-final component
thanks to LOOKUP_PARENT here. We fall to finish_automount instead.

Again: if this is non-final component, then LOOKUP_PARENT is set, and thus
LOOKUP_AUTOMOUNT is ignored. If this is final component, then LOOKUP_AUTOMOUNT
may affect things.

Code below tests that:
- statx always follows non-final automounts
- statx follow final automounts depending on options

The code doesn't test other syscalls, they can be added if needed.

The code was tested in Qemu on Linux 6.17-rc1.

I'm not trying to insult you in any way.

Again: thank you a lot for your work! For openat2 and for these mans.

Askar Safin

====

miniconfig:

CONFIG_64BIT=y

CONFIG_EXPERT=y

CONFIG_PRINTK=y
CONFIG_PRINTK_TIME=y

CONFIG_TTY=y
CONFIG_VT=y
CONFIG_VT_CONSOLE=y
CONFIG_FRAMEBUFFER_CONSOLE=y

CONFIG_PROC_FS=y
CONFIG_DEVTMPFS=y
CONFIG_SYSFS=y
CONFIG_TMPFS=y
CONFIG_DEBUG_FS=y
CONFIG_USER_EVENTS=y
CONFIG_FTRACE=y
CONFIG_MULTIUSER=y
CONFIG_NAMESPACES=y
CONFIG_USER_NS=y
CONFIG_PID_NS=y


CONFIG_SERIAL_8250=y
CONFIG_SERIAL_8250_CONSOLE=y

CONFIG_BLK_DEV_INITRD=y
CONFIG_RD_GZIP=y

CONFIG_BINFMT_ELF=y
CONFIG_BINFMT_SCRIPT=y

CONFIG_TRACEFS_AUTOMOUNT_DEPRECATED=y

CONFIG_DEBUG_INFO_DWARF_TOOLCHAIN_DEFAULT=y

====

/*
Author: Askar Safin
Public domain

Make sure your kernel is compiled with CONFIG_TRACEFS_AUTOMOUNT_DEPRECATED=y

If all tests pass, the program
should print "All tests passed".
Any other output means that something gone wrong.

This program requires root in initial user namespace
*/

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <stdbool.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sched.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/mount.h>
#include <sys/syscall.h>
#include <linux/openat2.h>

#define MY_ASSERT(cond) do { \
    if (!(cond)) { \
        fprintf (stderr, "%s: assertion failed\n", #cond); \
        exit (1); \
    } \
} while (0)

bool
tracing_mounted (void)
{
    struct statx tracing;
    if (statx (AT_FDCWD, "/tmp/debugfs/tracing", AT_NO_AUTOMOUNT, 0, &tracing) != 0)
        {
            perror ("statx tracing");
            exit (1);
        }
    if (!(tracing.stx_attributes_mask & STATX_ATTR_MOUNT_ROOT))
        {
            fprintf (stderr, "???\n");
            exit (1);
        }
    return tracing.stx_attributes & STATX_ATTR_MOUNT_ROOT;
}

void
mount_debugfs (void)
{
    if (mount (NULL, "/tmp/debugfs", "debugfs", 0, NULL) != 0)
        {
            perror ("mount debugfs");
            exit (1);
        }
    MY_ASSERT (!tracing_mounted ());
}

void
umount_debugfs (void)
{
    umount ("/tmp/debugfs/tracing"); // Ignore errors
    if (umount ("/tmp/debugfs") != 0)
        {
            perror ("umount debugfs");
            exit (1);
        }
}

int
main (void)
{
    // Init
    {
        if (chdir ("/") != 0)
            {
                perror ("chdir /");
                exit (1);
            }
        if (unshare (CLONE_NEWNS) != 0)
            {
                perror ("unshare");
                exit (1);
            }
        if (mount (NULL, "/", NULL, MS_REC | MS_PRIVATE, NULL) != 0)
            {
                perror ("mount(NULL, /, NULL, MS_REC | MS_PRIVATE, NULL)");
                exit (1);
            }
        if (mount (NULL, "/tmp", "tmpfs", 0, NULL) != 0)
            {
                perror ("mount tmpfs");
                exit (1);
            }
    }
    if (mkdir ("/tmp/debugfs", 0777) != 0)
        {
            perror ("mkdir(/tmp/debugfs)");
            exit (1);
        }

    // statx always follows automounts in non-final components. With AT_NO_AUTOMOUNT and without AT_NO_AUTOMOUNT
    {
        mount_debugfs();
        {
            struct statx readme;
            if (statx (AT_FDCWD, "/tmp/debugfs/tracing/README", 0, 0, &readme) != 0)
                {
                    perror ("statx");
                    exit (1);
                }
        }
        MY_ASSERT (tracing_mounted ());
        umount_debugfs();

        mount_debugfs();
        {
            struct statx readme;
            if (statx (AT_FDCWD, "/tmp/debugfs/tracing/README", AT_NO_AUTOMOUNT, 0, &readme) != 0)
                {
                    perror ("statx");
                    exit (1);
                }
        }
        MY_ASSERT (tracing_mounted ());
        umount_debugfs();
    }

    // statx follows automounts in final components if AT_NO_AUTOMOUNT is not specified
    {
        mount_debugfs();
        {
            struct statx tracing;
            if (statx (AT_FDCWD, "/tmp/debugfs/tracing", 0, 0, &tracing) != 0)
                {
                    perror ("statx");
                    exit (1);
                }
            if (!(tracing.stx_attributes_mask & STATX_ATTR_MOUNT_ROOT))
                {
                    fprintf (stderr, "???\n");
                    exit (1);
                }

            // Checking that this is new mount, not automount point itself
            MY_ASSERT (tracing.stx_attributes & STATX_ATTR_MOUNT_ROOT);
        }
        MY_ASSERT (tracing_mounted ());
        umount_debugfs ();

        mount_debugfs();
        {
            struct statx tracing;
            if (statx (AT_FDCWD, "/tmp/debugfs/tracing", AT_NO_AUTOMOUNT, 0, &tracing) != 0)
                {
                    perror ("statx");
                    exit (1);
                }
            if (!(tracing.stx_attributes_mask & STATX_ATTR_MOUNT_ROOT))
                {
                    fprintf (stderr, "???\n");
                    exit (1);
                }

            MY_ASSERT (!(tracing.stx_attributes & STATX_ATTR_MOUNT_ROOT));
        }
        MY_ASSERT (!tracing_mounted ());
        umount_debugfs ();
    }

    printf ("All tests passed\n");
    exit (0);
}

Re: [PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 638 bytes --]

On 2025-08-17, Askar Safin <safinaskar@zohomail.com> wrote:
> I just sent to fsdevel fix for that RESOLVE_NO_XDEV bug.

Thanks, I've sent some review comments.

> 	if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
> 			   LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
> 
> We never return -EISDIR in this "if" if we are in non-final component
> thanks to LOOKUP_PARENT here. We fall to finish_automount instead.

Grr, I re-read this conditional a few times and I still misunderstood
what it was doing. My bad.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Christian Brauner]

On Mon, Aug 18, 2025 at 02:16:04AM +1000, Aleksa Sarai wrote:
> On 2025-08-17, Askar Safin <safinaskar@zohomail.com> wrote:
> > I noticed that you changed docs for automounts. So I dig into
> > automounts implementation. And I found a bug in openat2. If
> > RESOLVE_NO_XDEV is specified, then name resolution doesn't cross
> > automount points (i. e. we get EXDEV), but automounts still happen! I
> > think this is a bug. Bug is reproduced in 6.17-rc1. In the end of this
> > mail you will find reproducer. And miniconfig.
> 
> Yes, this is a bug -- we check LOOKUP_NO_XDEV after traverse_mounts()
> because we want to error out if we actually jumped to a different mount.
> We should probably be erroring out in follow_automount() as well, and I
> missed this when I wrote openat2().
> 
> openat2() also really needs RESOLVE_NO_AUTOMOUNT (and probably
> RESOLVE_NO_DOTDOT as well as some other small features). I'll try to
> send something soon.
> 
> > Are automounts actually used? Is it possible to deprecate or
> > remove them? It seems for me automounts are rarely tested obscure
> > feature, which affects core namei code.
> 
> I use them for auto-mounting NFS shares on my laptop, and I'm sure there
> are plenty of other users. They are little bit funky but I highly doubt
> they are "unused". Howells probably disagrees in even stronger terms.
> Most distributions provide autofs as a supported package (I think it
> even comes pre-installed for some distros).
> 
> They are not tested by fstests AFAICS, but that's more of a flaw in
> fstests (automount requires you to have a running autofs daemon, which
> probably makes testing it in fstests or selftests impractical) not the
> feature itself.
> 
> > This reproducer is based on "tracing" automount, which
> > actually *IS* already deprecated. But automount mechanism
> > itself is not deprecated, as well as I know.
> 
> The automount behaviour of tracefs is different to the general automount
> mechanism which is managed by userspace with the autofs daemon. I don't
> know the history behind the deprecation, but I expect that it was
> deprecated in favour of configuring it with autofs (or just enabling it
> by default).
> 
> > Also, I did read namei code, and I think that
> > options AT_NO_AUTOMOUNT, FSPICK_NO_AUTOMOUNT, etc affect
> > last component only, not all of them. I didn't test this yet.
> > I plan to test this within next days.
> 
> No, LOOKUP_AUTOMOUNT affects all components. I double-checked this with
> Christian.

Hm? I was asking the question in the chat because I was unsure and not
in front of a computer you then said that it does affect all components. :)

> 
> You would think that it's only the last component (like O_DIRECTORY,
> O_NOFOLLOW, AT_SYMLINK_{,NO}FOLLOW) but follow_automount() is called for
> all components (i.e., as part of step_into()). It hooks into the regular
> lookup flow for mountpoints.
> 
> Yes, it is quite funky that AT_NO_AUTOMOUNT is the only AT_* flag that
> works this way -- hence why I went with a different RESOLVE_* namespace
> for openat2() (which _always_ act on _all_ components).
> 
> -- 
> Aleksa Sarai
> Senior Software Engineer (Containers)
> SUSE Linux GmbH
> https://www.cyphar.com/

Re: [PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 3692 bytes --]

On 2025-08-19, Christian Brauner <brauner@kernel.org> wrote:
> On Mon, Aug 18, 2025 at 02:16:04AM +1000, Aleksa Sarai wrote:
> > On 2025-08-17, Askar Safin <safinaskar@zohomail.com> wrote:
> > > I noticed that you changed docs for automounts. So I dig into
> > > automounts implementation. And I found a bug in openat2. If
> > > RESOLVE_NO_XDEV is specified, then name resolution doesn't cross
> > > automount points (i. e. we get EXDEV), but automounts still happen! I
> > > think this is a bug. Bug is reproduced in 6.17-rc1. In the end of this
> > > mail you will find reproducer. And miniconfig.
> > 
> > Yes, this is a bug -- we check LOOKUP_NO_XDEV after traverse_mounts()
> > because we want to error out if we actually jumped to a different mount.
> > We should probably be erroring out in follow_automount() as well, and I
> > missed this when I wrote openat2().
> > 
> > openat2() also really needs RESOLVE_NO_AUTOMOUNT (and probably
> > RESOLVE_NO_DOTDOT as well as some other small features). I'll try to
> > send something soon.
> > 
> > > Are automounts actually used? Is it possible to deprecate or
> > > remove them? It seems for me automounts are rarely tested obscure
> > > feature, which affects core namei code.
> > 
> > I use them for auto-mounting NFS shares on my laptop, and I'm sure there
> > are plenty of other users. They are little bit funky but I highly doubt
> > they are "unused". Howells probably disagrees in even stronger terms.
> > Most distributions provide autofs as a supported package (I think it
> > even comes pre-installed for some distros).
> > 
> > They are not tested by fstests AFAICS, but that's more of a flaw in
> > fstests (automount requires you to have a running autofs daemon, which
> > probably makes testing it in fstests or selftests impractical) not the
> > feature itself.
> > 
> > > This reproducer is based on "tracing" automount, which
> > > actually *IS* already deprecated. But automount mechanism
> > > itself is not deprecated, as well as I know.
> > 
> > The automount behaviour of tracefs is different to the general automount
> > mechanism which is managed by userspace with the autofs daemon. I don't
> > know the history behind the deprecation, but I expect that it was
> > deprecated in favour of configuring it with autofs (or just enabling it
> > by default).
> > 
> > > Also, I did read namei code, and I think that
> > > options AT_NO_AUTOMOUNT, FSPICK_NO_AUTOMOUNT, etc affect
> > > last component only, not all of them. I didn't test this yet.
> > > I plan to test this within next days.
> > 
> > No, LOOKUP_AUTOMOUNT affects all components. I double-checked this with
> > Christian.
> 
> Hm? I was asking the question in the chat because I was unsure and not
> in front of a computer you then said that it does affect all components. :)

Yeah I misunderstood what you said -- didn't mean to throw you under the
bus, sorry about that!

> > You would think that it's only the last component (like O_DIRECTORY,
> > O_NOFOLLOW, AT_SYMLINK_{,NO}FOLLOW) but follow_automount() is called for
> > all components (i.e., as part of step_into()). It hooks into the regular
> > lookup flow for mountpoints.
> > 
> > Yes, it is quite funky that AT_NO_AUTOMOUNT is the only AT_* flag that
> > works this way -- hence why I went with a different RESOLVE_* namespace
> > for openat2() (which _always_ act on _all_ components).
> > 
> > -- 
> > Aleksa Sarai
> > Senior Software Engineer (Containers)
> > SUSE Linux GmbH
> > https://www.cyphar.com/
> 
> 

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Askar Safin]

 ---- On Sun, 17 Aug 2025 20:16:04 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > They are not tested by fstests AFAICS, but that's more of a flaw in
 > fstests (automount requires you to have a running autofs daemon, which
 > probably makes testing it in fstests or selftests impractical) not the
 > feature itself.

I suggest testing automounts in fstests/selftests using "tracing" automount.
This is what I do in my reproducers.

 > The automount behaviour of tracefs is different to the general automount
 > mechanism which is managed by userspace with the autofs daemon.

Yes. But I still was able to write reproducers using "tracing", so this
automount point is totally okay for tests. (At least for some tests,
such as RESOLVE_NO_XDEV.)

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 1181 bytes --]

On 2025-08-20, Askar Safin <safinaskar@zohomail.com> wrote:
>  ---- On Sun, 17 Aug 2025 20:16:04 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
>  > They are not tested by fstests AFAICS, but that's more of a flaw in
>  > fstests (automount requires you to have a running autofs daemon, which
>  > probably makes testing it in fstests or selftests impractical) not the
>  > feature itself.
> 
> I suggest testing automounts in fstests/selftests using "tracing" automount.
> This is what I do in my reproducers.
> 
>  > The automount behaviour of tracefs is different to the general automount
>  > mechanism which is managed by userspace with the autofs daemon.
> 
> Yes. But I still was able to write reproducers using "tracing", so this
> automount point is totally okay for tests. (At least for some tests,
> such as RESOLVE_NO_XDEV.)

Sure, but I don't think people use allyesconfig when running selftests.
I wonder if the automated test runners even enable deprecated features
like that.

In any case, you can definitely write some tests for it. :D

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Askar Safin]

There is one particular case when open_tree is more powerful than openat with O_PATH. open_tree supports AT_EMPTY_PATH, and openat supports nothing similar.
This means that we can convert normal O_RDONLY file descriptor to O_PATH descriptor using open_tree! I. e.:
  rd = openat(AT_FDCWD, "/tmp/a", O_RDONLY, 0); // Regular file
  open_tree(rd, "", AT_EMPTY_PATH);
You can achieve same effect using /proc:
  rd = openat(AT_FDCWD, "/tmp/a", O_RDONLY, 0); // Regular file
  snprintf(buf, sizeof(buf), "/proc/self/fd/%d", rd);
  openat(AT_FDCWD, buf, O_PATH, 0);
But still I think this has security implications. This means that even if we deny access to /proc for container, it still is able to convert O_RDONLY
descriptors to O_PATH descriptors using open_tree. I. e. this is yet another thing to think about when creating sandboxes.
I know you delivered a talk about similar things a lot of time ago: https://lwn.net/Articles/934460/ . (I tested this.)

--
Askar Safin
https://types.pl/@safinaskar

Re: [PATCH v3 00/12] man2: document "new" mount API

[Aleksa Sarai]

[-- Attachment #1: Type: text/plain, Size: 2380 bytes --]

On 2025-08-21, Askar Safin <safinaskar@zohomail.com> wrote:
> There is one particular case when open_tree is more powerful than openat with O_PATH. open_tree supports AT_EMPTY_PATH, and openat supports nothing similar.
> This means that we can convert normal O_RDONLY file descriptor to O_PATH descriptor using open_tree! I. e.:
>   rd = openat(AT_FDCWD, "/tmp/a", O_RDONLY, 0); // Regular file
>   open_tree(rd, "", AT_EMPTY_PATH);
> You can achieve same effect using /proc:
>   rd = openat(AT_FDCWD, "/tmp/a", O_RDONLY, 0); // Regular file
>   snprintf(buf, sizeof(buf), "/proc/self/fd/%d", rd);
>   openat(AT_FDCWD, buf, O_PATH, 0);
> But still I think this has security implications. This means that even if we deny access to /proc for container, it still is able to convert O_RDONLY
> descriptors to O_PATH descriptors using open_tree. I. e. this is yet another thing to think about when creating sandboxes.
> I know you delivered a talk about similar things a lot of time ago: https://lwn.net/Articles/934460/ . (I tested this.)

O_RDONLY -> O_PATH is less of an issue than the other way around. There
isn't much you can do with O_PATH that you can't do with a properly open
file (by design you actually should have strictly less privileges but
some operations are only really possible with O_PATH, but they're not
security-critical in that way).

I was working on a new patchset for resolving this issue (and adding
O_EMPTYPATH support) late last year but other things fell on my plate
and the design was quite difficult to get to a place where everyone
agreed to it.

The core issue is that we would need to block not just re-opening but
also any operation that is a write (or read) in disguise, which kind of
implies you need to have capabilities attached to file descriptors. This
is already slightly shaky ground if you look at the history of projects
like capsicum -- but also my impression was that just adding it to
"file_permission" was not sufficient, you need to put it in
"path_permission" which means we have to either bloat "struct path" or
come up with some extended structure that you need to plumb through
everywhere.

But yes, this is a thing that is still on my list of things to do, but
not in the immediate future.

-- 
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
https://www.cyphar.com/

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH v3 00/12] man2: document "new" mount API

[Askar Safin]

 ---- On Sat, 09 Aug 2025 00:39:44 +0400  Aleksa Sarai <cyphar@cyphar.com> wrote --- 
 > Back in 2019, the new mount API was merged[1]. David Howells then set

I finished my experiments!

--
Askar Safin
https://types.pl/@safinaskar