[PATCH] build: file-systems: Allow for bind mounting regular files.

[PATCH] build: file-systems: Allow for bind mounting regular files.

[David Thompson]

[-- Attachment #1: Type: text/plain, Size: 210 bytes --]

As I was working on my container implementation I noticed that
'mount-file-system' doesn't support bind mounting regular files because
it assumes that all mount points are directories.  This patch fixes
that.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-build-file-systems-Allow-for-bind-mounting-regular-f.patch --]
[-- Type: text/x-patch, Size: 1799 bytes --]

From f94fec6cde3826f20c0d69a45c2aa1928c1d0a78 Mon Sep 17 00:00:00 2001
From: David Thompson <dthompson2@worcester.edu>
Date: Sat, 1 Aug 2015 13:43:33 -0400
Subject: [PATCH] build: file-systems: Allow for bind mounting regular files.

* gnu/build/file-systems.scm (regular-file?): New procedure.
  (mount-file-system): Create a regular file instead of a directory when bind
  mounting a regular file.
---
 gnu/build/file-systems.scm | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm
index c58d23c..f0d6f70 100644
--- a/gnu/build/file-systems.scm
+++ b/gnu/build/file-systems.scm
@@ -305,6 +305,10 @@ the following:
                fsck code device)
        (start-repl)))))
 
+(define (regular-file? file-name)
+  "Return #t if FILE-NAME is a regular file."
+  (eq? (stat:type (stat file-name)) 'regular))
+
 (define (mount-flags->bit-mask flags)
   "Return the number suitable for the 'flags' argument of 'mount' that
 corresponds to the symbols listed in FLAGS."
@@ -339,7 +343,16 @@ run a file system check."
            (flags       (mount-flags->bit-mask flags)))
        (when check?
          (check-file-system source type))
-       (mkdir-p mount-point)
+
+       ;; Create the mount point.  Most of the time this is a directory, but
+       ;; in the case of a bind mount, a regular file may be needed.
+       (if (and (= MS_BIND (logand flags MS_BIND))
+                (regular-file? source))
+           (begin
+             (mkdir-p (dirname mount-point))
+             (call-with-output-file mount-point (const #t)))
+           (mkdir-p mount-point))
+
        (mount source mount-point type flags options)
 
        ;; For read-only bind mounts, an extra remount is needed, as per
-- 
2.4.3


[-- Attachment #3: Type: text/plain, Size: 38 bytes --]


-- 
David Thompson
GPG Key: 0FF1D807

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Alex Kost]

David Thompson (2015-08-01 22:17 +0300) wrote:

> diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm
> index c58d23c..f0d6f70 100644
> --- a/gnu/build/file-systems.scm
> +++ b/gnu/build/file-systems.scm
> @@ -305,6 +305,10 @@ the following:
>                 fsck code device)
>         (start-repl)))))
>  
> +(define (regular-file? file-name)
> +  "Return #t if FILE-NAME is a regular file."
> +  (eq? (stat:type (stat file-name)) 'regular))

There are similar procedures in (guix build utils): 'directory-exists?',
'executable-file?' and 'symbolic-link?'.  So I think it is better to put
'regular-file?' there.  WDYT?

-- 
Alex

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Thompson, David]

[-- Attachment #1: Type: text/plain, Size: 767 bytes --]

On Sun, Aug 2, 2015 at 8:10 AM, Alex Kost <alezost@gmail.com> wrote:
> David Thompson (2015-08-01 22:17 +0300) wrote:
>
>> diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm
>> index c58d23c..f0d6f70 100644
>> --- a/gnu/build/file-systems.scm
>> +++ b/gnu/build/file-systems.scm
>> @@ -305,6 +305,10 @@ the following:
>>                 fsck code device)
>>         (start-repl)))))
>>
>> +(define (regular-file? file-name)
>> +  "Return #t if FILE-NAME is a regular file."
>> +  (eq? (stat:type (stat file-name)) 'regular))
>
> There are similar procedures in (guix build utils): 'directory-exists?',
> 'executable-file?' and 'symbolic-link?'.  So I think it is better to put
> 'regular-file?' there.  WDYT?

Sure, that makes sense.  Done.

- Dave

[-- Attachment #2: 0001-build-file-systems-Allow-for-bind-mounting-regular-f.patch --]
[-- Type: text/x-patch, Size: 2283 bytes --]

From 1b2413bd06b1e769edfbe4d170de41398015a67d Mon Sep 17 00:00:00 2001
From: David Thompson <dthompson2@worcester.edu>
Date: Sat, 1 Aug 2015 13:43:33 -0400
Subject: [PATCH] build: file-systems: Allow for bind mounting regular files.

* guix/build/utils.scm (regular-file?): New procedure.
* gnu/build/file-systems.scm (mount-file-system): Create a regular file
  instead of a directory when bind mounting a regular file.
---
 gnu/build/file-systems.scm | 11 ++++++++++-
 guix/build/utils.scm       |  5 +++++
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm
index c58d23c..f0b4b79 100644
--- a/gnu/build/file-systems.scm
+++ b/gnu/build/file-systems.scm
@@ -339,7 +339,16 @@ run a file system check."
            (flags       (mount-flags->bit-mask flags)))
        (when check?
          (check-file-system source type))
-       (mkdir-p mount-point)
+
+       ;; Create the mount point.  Most of the time this is a directory, but
+       ;; in the case of a bind mount, a regular file may be needed.
+       (if (and (= MS_BIND (logand flags MS_BIND))
+                (regular-file? source))
+           (begin
+             (mkdir-p (dirname mount-point))
+             (call-with-output-file mount-point (const #t)))
+           (mkdir-p mount-point))
+
        (mount source mount-point type flags options)
 
        ;; For read-only bind mounts, an extra remount is needed, as per
diff --git a/guix/build/utils.scm b/guix/build/utils.scm
index 676a012..b9543ed 100644
--- a/guix/build/utils.scm
+++ b/guix/build/utils.scm
@@ -38,6 +38,7 @@
             directory-exists?
             executable-file?
             symbolic-link?
+            regular-file?
             call-with-ascii-input-file
             elf-file?
             ar-file?
@@ -110,6 +111,10 @@
   "Return #t if FILE is a symbolic link (aka. \"symlink\".)"
   (eq? (stat:type (lstat file)) 'symlink))
 
+(define (regular-file? file-name)
+  "Return #t if FILE-NAME is a regular file."
+  (eq? (stat:type (stat file-name)) 'regular))
+
 (define (call-with-ascii-input-file file proc)
   "Open FILE as an ASCII or binary file, and pass the resulting port to
 PROC.  FILE is closed when PROC's dynamic extent is left.  Return the
-- 
2.4.3

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Thompson, David]

On Sun, Aug 2, 2015 at 8:43 AM, Thompson, David
<dthompson2@worcester.edu> wrote:
> On Sun, Aug 2, 2015 at 8:10 AM, Alex Kost <alezost@gmail.com> wrote:
>> David Thompson (2015-08-01 22:17 +0300) wrote:
>>
>>> diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm
>>> index c58d23c..f0d6f70 100644
>>> --- a/gnu/build/file-systems.scm
>>> +++ b/gnu/build/file-systems.scm
>>> @@ -305,6 +305,10 @@ the following:
>>>                 fsck code device)
>>>         (start-repl)))))
>>>
>>> +(define (regular-file? file-name)
>>> +  "Return #t if FILE-NAME is a regular file."
>>> +  (eq? (stat:type (stat file-name)) 'regular))
>>
>> There are similar procedures in (guix build utils): 'directory-exists?',
>> 'executable-file?' and 'symbolic-link?'.  So I think it is better to put
>> 'regular-file?' there.  WDYT?
>
> Sure, that makes sense.  Done.

Ah, of course, I forgot about something:  This patch triggers a
rebuild of *everything* now!  I guess it should be applied to
core-updates.  Or, the first patch I submitted can be applied to
master, and then a patch that moves 'regular-file?' to (guix build
utils) can be applied to core-updates later.

Thoughts?

- Dave

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Andreas Enge]

On Sun, Aug 02, 2015 at 08:51:46AM -0400, Thompson, David wrote:
> Ah, of course, I forgot about something:  This patch triggers a
> rebuild of *everything* now!  I guess it should be applied to
> core-updates.

I think it would be better to wait until hydra is done building the
security updates Mark is adding.

Andreas

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Alex Kost]

Thompson, David (2015-08-02 15:51 +0300) wrote:

> On Sun, Aug 2, 2015 at 8:43 AM, Thompson, David
> <dthompson2@worcester.edu> wrote:
>> On Sun, Aug 2, 2015 at 8:10 AM, Alex Kost <alezost@gmail.com> wrote:
>>> David Thompson (2015-08-01 22:17 +0300) wrote:
>>>
>>>> diff --git a/gnu/build/file-systems.scm b/gnu/build/file-systems.scm
>>>> index c58d23c..f0d6f70 100644
>>>> --- a/gnu/build/file-systems.scm
>>>> +++ b/gnu/build/file-systems.scm
>>>> @@ -305,6 +305,10 @@ the following:
>>>>                 fsck code device)
>>>>         (start-repl)))))
>>>>
>>>> +(define (regular-file? file-name)
>>>> +  "Return #t if FILE-NAME is a regular file."
>>>> +  (eq? (stat:type (stat file-name)) 'regular))
>>>
>>> There are similar procedures in (guix build utils): 'directory-exists?',
>>> 'executable-file?' and 'symbolic-link?'.  So I think it is better to put
>>> 'regular-file?' there.  WDYT?
>>
>> Sure, that makes sense.  Done.
>
> Ah, of course, I forgot about something:  This patch triggers a
> rebuild of *everything* now!  I guess it should be applied to
> core-updates.  Or, the first patch I submitted can be applied to
> master, and then a patch that moves 'regular-file?' to (guix build
> utils) can be applied to core-updates later.
>
> Thoughts?

In my opinion it is redundant to use another commit here, so I would
apply this patch to core-updates (taking into account Andreas' note).
But I don't really know how such situations are handled.  Mark should
know better.

-- 
Alex

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Thompson, David]

On Mon, Aug 3, 2015 at 8:47 AM, Alex Kost <alezost@gmail.com> wrote:
> Thompson, David (2015-08-02 15:51 +0300) wrote:
>
>> Ah, of course, I forgot about something:  This patch triggers a
>> rebuild of *everything* now!  I guess it should be applied to
>> core-updates.  Or, the first patch I submitted can be applied to
>> master, and then a patch that moves 'regular-file?' to (guix build
>> utils) can be applied to core-updates later.
>>
>> Thoughts?
>
> In my opinion it is redundant to use another commit here, so I would
> apply this patch to core-updates (taking into account Andreas' note).
> But I don't really know how such situations are handled.  Mark should
> know better.

The reason I want two commits is strategic.  This first patch can go
into master *now*, without breaking anything or causing a full
rebuild.  Moving that procedure into (guix build utils) is nice, but
not strictly necessary.  I am trying to get basic container support in
shape and into master.  Waiting on a core-updates cycle would just
stall the process.

- Dave

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Alex Kost]

Thompson, David (2015-08-06 15:22 +0300) wrote:

> On Mon, Aug 3, 2015 at 8:47 AM, Alex Kost <alezost@gmail.com> wrote:
>> Thompson, David (2015-08-02 15:51 +0300) wrote:
>>
>>> Ah, of course, I forgot about something:  This patch triggers a
>>> rebuild of *everything* now!  I guess it should be applied to
>>> core-updates.  Or, the first patch I submitted can be applied to
>>> master, and then a patch that moves 'regular-file?' to (guix build
>>> utils) can be applied to core-updates later.
>>>
>>> Thoughts?
>>
>> In my opinion it is redundant to use another commit here, so I would
>> apply this patch to core-updates (taking into account Andreas' note).
>> But I don't really know how such situations are handled.  Mark should
>> know better.
>
> The reason I want two commits is strategic.  This first patch can go
> into master *now*, without breaking anything or causing a full
> rebuild.  Moving that procedure into (guix build utils) is nice, but
> not strictly necessary.  I am trying to get basic container support in
> shape and into master.  Waiting on a core-updates cycle would just
> stall the process.

Ah, OK, I don't mind then.

-- 
Alex

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Thompson, David]

On Sat, Aug 1, 2015 at 3:17 PM, David Thompson <dthompson2@worcester.edu> wrote:
> As I was working on my container implementation I noticed that
> 'mount-file-system' doesn't support bind mounting regular files because
> it assumes that all mount points are directories.  This patch fixes
> that.

I pushed this.

- Dave

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Ludovic Courtès]

"Thompson, David" <dthompson2@worcester.edu> skribis:

> +(define (regular-file? file-name)
> +  "Return #t if FILE-NAME is a regular file."
> +  (eq? (stat:type (stat file-name)) 'regular))

I see you moved this procedure to (gnu build file-systems), which is
reasonable.

I wonder if it should use ‘lstat’ instead of ‘stat’?

Ludo’.

Re: [PATCH] build: file-systems: Allow for bind mounting regular files.

[Thompson, David]

On Tue, Aug 18, 2015 at 11:52 AM, Ludovic Courtès <ludo@gnu.org> wrote:
> "Thompson, David" <dthompson2@worcester.edu> skribis:
>
>> +(define (regular-file? file-name)
>> +  "Return #t if FILE-NAME is a regular file."
>> +  (eq? (stat:type (stat file-name)) 'regular))
>
> I see you moved this procedure to (gnu build file-systems), which is
> reasonable.

I did this to avoid a full rebuild.  A patch for core-updates could move it.

> I wonder if it should use ‘lstat’ instead of ‘stat’?

To follow symlinks or not?  I guess the answer is "no".  I can change
it to lstat.

- Dave