Re: [RFC] do we want pipefs et.al. in /proc/filesystems?
On Wed, Oct 9, 2013 at 5:47 PM, Al Viro wrote: > > These days there's no reason to register the filesystem types > that can only be used for internal mounts - ia64 pfmfs, anon_inodes, > bdev, pipefs and sockfs, in the current tree. The only user-visible > effect of dropping those register_filesystem() would be shorter > /proc/filesystems - that bunch wouldn't be mentioned there anymore. > Does anything care about that? FWIW, the diff eliminating those > would be as below (net/socket.c init leaks on early failures; > I'd left that as-is - it's a separate story). I like the patch, but I do worry that some user space thing uses this to check whether the particular filesystem module is loaded. Those kinds of things can't use /proc/modules (because it might be built-in), but /proc/filesystems has traditionally contained everything. That said, these particular filesystems I really don't think people would ever possibly check for, so I think it's fine. Linus -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
[RFC] do we want pipefs et.al. in /proc/filesystems?
We used to require register_filesystem() before any mounts
(including the internal ones) could be done. For userland mounts
it's obviously right - how else could fs type name -> file_system_type
lookup be done? For internal ones, it used to be really needed for
a while, but after some API massage (about 10 years ago, IIRC) it really
became pointless - kern_mount() takes a pointer to file_system_type
and doesn't bother with searching for it by name. For a while there
was one remaining reason - the list of struct superblock belonging to
given fs type was a list_head one, with register_filesystem() initializing
the list. A couple of years ago we switched to hlist, which doesn't
require anything of that sort - hlist_head initialized to NULL is an
empty list.
These days there's no reason to register the filesystem types
that can only be used for internal mounts - ia64 pfmfs, anon_inodes,
bdev, pipefs and sockfs, in the current tree. The only user-visible
effect of dropping those register_filesystem() would be shorter
/proc/filesystems - that bunch wouldn't be mentioned there anymore.
Does anything care about that? FWIW, the diff eliminating those
would be as below (net/socket.c init leaks on early failures;
I'd left that as-is - it's a separate story).
diff --git a/arch/ia64/kernel/perfmon.c b/arch/ia64/kernel/perfmon.c
index 5a9ff1c..bb3968e 100644
--- a/arch/ia64/kernel/perfmon.c
+++ b/arch/ia64/kernel/perfmon.c
@@ -1529,16 +1529,10 @@ static struct vfsmount *pfmfs_mnt __read_mostly;
static int __init
init_pfm_fs(void)
{
- int err = register_filesystem(_fs_type);
- if (!err) {
- pfmfs_mnt = kern_mount(_fs_type);
- err = PTR_ERR(pfmfs_mnt);
- if (IS_ERR(pfmfs_mnt))
- unregister_filesystem(_fs_type);
- else
- err = 0;
- }
- return err;
+ pfmfs_mnt = kern_mount(_fs_type);
+ if (IS_ERR(pfmfs_mnt))
+ return PTR_ERR(pfmfs_mnt);
+ return 0;
}
static ssize_t
diff --git a/fs/anon_inodes.c b/fs/anon_inodes.c
index 2408473..ad56c5a 100644
--- a/fs/anon_inodes.c
+++ b/fs/anon_inodes.c
@@ -175,22 +175,11 @@ EXPORT_SYMBOL_GPL(anon_inode_getfd);
static int __init anon_inode_init(void)
{
- int error;
-
- error = register_filesystem(_inode_fs_type);
- if (error)
- goto err_exit;
anon_inode_mnt = kern_mount(_inode_fs_type);
- if (IS_ERR(anon_inode_mnt)) {
- error = PTR_ERR(anon_inode_mnt);
- goto err_unregister_filesystem;
- }
+ if (IS_ERR(anon_inode_mnt))
+ panic("anon_inode_init() failed (%ld)\n",
+ PTR_ERR(anon_inode_mnt));
return 0;
-
-err_unregister_filesystem:
- unregister_filesystem(_inode_fs_type);
-err_exit:
- panic(KERN_ERR "anon_inode_init() failed (%d)\n", error);
}
fs_initcall(anon_inode_init);
diff --git a/fs/block_dev.c b/fs/block_dev.c
index 1e86823..ff3bf6a 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -461,9 +461,6 @@ void __init bdev_cache_init(void)
0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
SLAB_MEM_SPREAD|SLAB_PANIC),
init_once);
- err = register_filesystem(_type);
- if (err)
- panic("Cannot register bdev pseudo-fs");
bd_mnt = kern_mount(_type);
if (IS_ERR(bd_mnt))
panic("Cannot create bdev pseudo-fs");
diff --git a/fs/pipe.c b/fs/pipe.c
index d2c45e14..7a63d0a 100644
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -1311,16 +1311,10 @@ static struct file_system_type pipe_fs_type = {
static int __init init_pipe_fs(void)
{
- int err = register_filesystem(_fs_type);
-
- if (!err) {
- pipe_mnt = kern_mount(_fs_type);
- if (IS_ERR(pipe_mnt)) {
- err = PTR_ERR(pipe_mnt);
- unregister_filesystem(_fs_type);
- }
- }
- return err;
+ pipe_mnt = kern_mount(_fs_type);
+ if (IS_ERR(pipe_mnt))
+ panic("pipe init failed\n");
+ return 0;
}
fs_initcall(init_pipe_fs);
diff --git a/net/socket.c b/net/socket.c
index ebed4b6..ea534a3 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -2617,7 +2617,7 @@ static int __init sock_init(void)
*/
err = net_sysctl_init();
if (err)
- goto out;
+ return err;
/*
* Initialize skbuff SLAB cache
@@ -2630,14 +2630,9 @@ static int __init sock_init(void)
init_inodecache();
- err = register_filesystem(_fs_type);
- if (err)
- goto out_fs;
sock_mnt = kern_mount(_fs_type);
- if (IS_ERR(sock_mnt)) {
- err = PTR_ERR(sock_mnt);
- goto out_mount;
- }
+ if (IS_ERR(sock_mnt))
+ return PTR_ERR(sock_mnt);
/*
[RFC] do we want pipefs et.al. in /proc/filesystems?
We used to require register_filesystem() before any mounts
(including the internal ones) could be done. For userland mounts
it's obviously right - how else could fs type name - file_system_type
lookup be done? For internal ones, it used to be really needed for
a while, but after some API massage (about 10 years ago, IIRC) it really
became pointless - kern_mount() takes a pointer to file_system_type
and doesn't bother with searching for it by name. For a while there
was one remaining reason - the list of struct superblock belonging to
given fs type was a list_head one, with register_filesystem() initializing
the list. A couple of years ago we switched to hlist, which doesn't
require anything of that sort - hlist_head initialized to NULL is an
empty list.
These days there's no reason to register the filesystem types
that can only be used for internal mounts - ia64 pfmfs, anon_inodes,
bdev, pipefs and sockfs, in the current tree. The only user-visible
effect of dropping those register_filesystem() would be shorter
/proc/filesystems - that bunch wouldn't be mentioned there anymore.
Does anything care about that? FWIW, the diff eliminating those
would be as below (net/socket.c init leaks on early failures;
I'd left that as-is - it's a separate story).
diff --git a/arch/ia64/kernel/perfmon.c b/arch/ia64/kernel/perfmon.c
index 5a9ff1c..bb3968e 100644
--- a/arch/ia64/kernel/perfmon.c
+++ b/arch/ia64/kernel/perfmon.c
@@ -1529,16 +1529,10 @@ static struct vfsmount *pfmfs_mnt __read_mostly;
static int __init
init_pfm_fs(void)
{
- int err = register_filesystem(pfm_fs_type);
- if (!err) {
- pfmfs_mnt = kern_mount(pfm_fs_type);
- err = PTR_ERR(pfmfs_mnt);
- if (IS_ERR(pfmfs_mnt))
- unregister_filesystem(pfm_fs_type);
- else
- err = 0;
- }
- return err;
+ pfmfs_mnt = kern_mount(pfm_fs_type);
+ if (IS_ERR(pfmfs_mnt))
+ return PTR_ERR(pfmfs_mnt);
+ return 0;
}
static ssize_t
diff --git a/fs/anon_inodes.c b/fs/anon_inodes.c
index 2408473..ad56c5a 100644
--- a/fs/anon_inodes.c
+++ b/fs/anon_inodes.c
@@ -175,22 +175,11 @@ EXPORT_SYMBOL_GPL(anon_inode_getfd);
static int __init anon_inode_init(void)
{
- int error;
-
- error = register_filesystem(anon_inode_fs_type);
- if (error)
- goto err_exit;
anon_inode_mnt = kern_mount(anon_inode_fs_type);
- if (IS_ERR(anon_inode_mnt)) {
- error = PTR_ERR(anon_inode_mnt);
- goto err_unregister_filesystem;
- }
+ if (IS_ERR(anon_inode_mnt))
+ panic(anon_inode_init() failed (%ld)\n,
+ PTR_ERR(anon_inode_mnt));
return 0;
-
-err_unregister_filesystem:
- unregister_filesystem(anon_inode_fs_type);
-err_exit:
- panic(KERN_ERR anon_inode_init() failed (%d)\n, error);
}
fs_initcall(anon_inode_init);
diff --git a/fs/block_dev.c b/fs/block_dev.c
index 1e86823..ff3bf6a 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -461,9 +461,6 @@ void __init bdev_cache_init(void)
0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
SLAB_MEM_SPREAD|SLAB_PANIC),
init_once);
- err = register_filesystem(bd_type);
- if (err)
- panic(Cannot register bdev pseudo-fs);
bd_mnt = kern_mount(bd_type);
if (IS_ERR(bd_mnt))
panic(Cannot create bdev pseudo-fs);
diff --git a/fs/pipe.c b/fs/pipe.c
index d2c45e14..7a63d0a 100644
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -1311,16 +1311,10 @@ static struct file_system_type pipe_fs_type = {
static int __init init_pipe_fs(void)
{
- int err = register_filesystem(pipe_fs_type);
-
- if (!err) {
- pipe_mnt = kern_mount(pipe_fs_type);
- if (IS_ERR(pipe_mnt)) {
- err = PTR_ERR(pipe_mnt);
- unregister_filesystem(pipe_fs_type);
- }
- }
- return err;
+ pipe_mnt = kern_mount(pipe_fs_type);
+ if (IS_ERR(pipe_mnt))
+ panic(pipe init failed\n);
+ return 0;
}
fs_initcall(init_pipe_fs);
diff --git a/net/socket.c b/net/socket.c
index ebed4b6..ea534a3 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -2617,7 +2617,7 @@ static int __init sock_init(void)
*/
err = net_sysctl_init();
if (err)
- goto out;
+ return err;
/*
* Initialize skbuff SLAB cache
@@ -2630,14 +2630,9 @@ static int __init sock_init(void)
init_inodecache();
- err = register_filesystem(sock_fs_type);
- if (err)
- goto out_fs;
sock_mnt = kern_mount(sock_fs_type);
- if (IS_ERR(sock_mnt)) {
- err = PTR_ERR(sock_mnt);
- goto out_mount;
- }
+ if (IS_ERR(sock_mnt))
+
Re: [RFC] do we want pipefs et.al. in /proc/filesystems?
On Wed, Oct 9, 2013 at 5:47 PM, Al Viro v...@zeniv.linux.org.uk wrote: These days there's no reason to register the filesystem types that can only be used for internal mounts - ia64 pfmfs, anon_inodes, bdev, pipefs and sockfs, in the current tree. The only user-visible effect of dropping those register_filesystem() would be shorter /proc/filesystems - that bunch wouldn't be mentioned there anymore. Does anything care about that? FWIW, the diff eliminating those would be as below (net/socket.c init leaks on early failures; I'd left that as-is - it's a separate story). I like the patch, but I do worry that some user space thing uses this to check whether the particular filesystem module is loaded. Those kinds of things can't use /proc/modules (because it might be built-in), but /proc/filesystems has traditionally contained everything. That said, these particular filesystems I really don't think people would ever possibly check for, so I think it's fine. Linus -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
