xfs: add selinux labels to whiteout inodes
PATCH 1/1 xfs: add selinux labels to whiteout inodes
We got a report that "renameat2() with flags=RENAME_WHITEOUT doesn't apply an SELinux label on xfs" as it does on other filesystems (for example, ext4 and tmpfs.) While I'm not quite sure how labels may interact w/ whiteout files, leaving them as unlabeled seems inconsistent at best.
--- fs/xfs/xfs_inode.c | 14 +++++++++++++- fs/xfs/xfs_iops.c | 2 +- fs/xfs/xfs_iops.h | 3 +++ 3 files changed, 17 insertions(+), 2 deletions(-)
fs/xfs/xfs_inode.c +13 -1
@@ -3046,10 +3046,12 @@ struct xfs_iunlink {static int xfs_rename_alloc_whiteout (struct user_namespace *mnt_userns, +struct xfs_name *src_name,struct xfs_inode *dp,struct xfs_inode **wip) {struct xfs_inode *tmpfile; +struct qstr name;int error; error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
@@ -3057,6 +3059,15 @@ struct xfs_iunlink {if (error)return error; + name.name = src_name->name; + name.len = src_name->len; + error =xfs_init_security (VFS_I (tmpfile),VFS_I (dp), &name); +if (error) { +xfs_finish_inode_setup (tmpfile); +xfs_irele (tmpfile); +return error; + } +/* * Prepare the tmpfile inode as if it were created through the VFS. * Complete the inode setup and flag it as linkable. nlink is already
@@ -3107,7 +3118,8 @@ struct xfs_iunlink {* appropriately. */ if (flags &RENAME_WHITEOUT ) { - error =xfs_rename_alloc_whiteout (mnt_userns, target_dp, &wip); + error =xfs_rename_alloc_whiteout (mnt_userns, src_name, + target_dp, &wip);if (error)return error;
fs/xfs/xfs_iops.c +1 -1
@@ -76,7 +76,7 @@* inode, of course, such that log replay can't cause these to be lost). */ -STATIC int +int xfs_init_security( struct inode *inode, struct inode *dir,
fs/xfs/xfs_iops.h +3 -0
@@ -17,4 +17,7 @@int xfs_vn_setattr_size (struct user_namespace *mnt_userns,struct dentry *dentry,struct iattr *vap); +int xfs_init_security (struct inode *inode,struct inode *dir, +const struct qstr *qstr); +#endif /* __XFS_IOPS_H__ */
Signature · 1 line
1.8.3.1
I had mentioned this problem in passing to dchinner, and came away with a vague sense that I might be Doing It Wrong. So I should have labeled the patch RFC, I suppose.
An easy way to demonstrate the selinux label result is with this test program, as provided by the bug reporter. ls -lZ will then show you the resulting labels on each file.
#define _GNU_SOURCE
#include <sys/mount.h>
#include <sys/stat.h>
#include <errno.h>
#include <fcntl.h>
#include <sched.h>
#include <stdio.h>
#include <unistd.h>
int main(int argc, char **argv) {
int rc, fd, dirfd;
rc = mkdir("upper", 0700);
if ((rc != 0) && (errno != EEXIST)) {
perror("mkdir");
return rc;
}
rc = unlink("upper/0");
if ((rc != 0) && (errno != ENOENT)) {
perror("unlink");
return rc;
}
rc = unlink("upper/empty");
if ((rc != 0) && (errno != ENOENT)) {
perror("unlink");
return rc;
}
dirfd = open("upper", O_PATH);
if (dirfd == -1) {
perror("open");
return dirfd;
}
fd = creat("upper/empty", 0600);
if (fd == -1) {
perror("creat");
return fd;
}
close(fd);
rc = renameat2(dirfd, "empty", dirfd, "0", RENAME_WHITEOUT);
if (rc == -1) {
perror("renameat2");
return rc;
}
close(dirfd);
return 0;
}Quoted 33 lines · Eric Sandeen
We got a report that "
renameat2()with flags=RENAME_WHITEOUT doesn't apply an SELinux label on xfs" as it does on other filesystems (for example, ext4 and tmpfs.) While I'm not quite sure how labels may interact w/ whiteout files, leaving them as unlabeled seems inconsistent at best.Signed-off-by Eric Sandeen <[email protected]>--- fs/xfs/xfs_inode.c | 14 +++++++++++++- fs/xfs/xfs_iops.c | 2 +- fs/xfs/xfs_iops.h | 3 +++ 3 files changed, 17 insertions(+), 2 deletions(-)fs/xfs/xfs_inode.c +2
-0@@ -3046,10 +3046,12 @@ struct xfs_iunlink {static int xfs_rename_alloc_whiteout (struct user_namespace *mnt_userns, +struct xfs_name *src_name,struct xfs_inode *dp,struct xfs_inode **wip) {struct xfs_inode *tmpfile; +struct qstr name;int error; error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,@@ -3057,6 +3059,15 @@ struct xfs_iunlink {if (error)return error;
fs/xfs/xfs_inode.c struct xfs_iunlink { · line 3070
Quoted 10 lines
fs/xfs/xfs_inode.c +9
-0+ name.name = src_name->name; + name.len = src_name->len; + error =xfs_init_security (VFS_I (tmpfile),VFS_I (dp), &name); +if (error) { +xfs_finish_inode_setup (tmpfile); +xfs_irele (tmpfile); +return error; + } +
I was worried that this would be inside an existing transaction, but the tmpfile create is outside the rename transaction so this will be fine.
Quoted 18 lines
fs/xfs/xfs_inode.c +2
-1/* * Prepare the tmpfile inode as if it were created through the VFS. * Complete the inode setup and flag it as linkable. nlink is already @@ -3107,7 +3118,8 @@ struct xfs_iunlink {* appropriately. */ if (flags &RENAME_WHITEOUT ) { - error =xfs_rename_alloc_whiteout (mnt_userns, target_dp, &wip); + error =xfs_rename_alloc_whiteout (mnt_userns, src_name, + target_dp, &wip);if (error)return error;fs/xfs/xfs_iops.c +0
-0
fs/xfs/xfs_iops.c line 80
Quoted 6 lines
fs/xfs/xfs_iops.c +1
-1@@ -76,7 +76,7 @@* inode, of course, such that log replay can't cause these to be lost). */ -STATIC int +int xfs_init_security(
This function needs renaming, though. As a static function it can get away with not having a namespace, but as a globally visible function it needs to have an "xfs_inode_" prefix....
Otherwise OK.
Signature · 2 lines
Dave Chinner [email protected]
PATCH v2 1/1 xfs: add selinux labels to whiteout inodes
In reply to Eric SandeenWe got a report that "renameat2() with flags=RENAME_WHITEOUT doesn't apply an SELinux label on xfs" as it does on other filesystems (for example, ext4 and tmpfs.) While I'm not quite sure how labels may interact w/ whiteout files, leaving them as unlabeled seems inconsistent at best. Now that xfs_init_security is not static, rename it to xfs_inode_init_security per dchinner's suggestion.
--- fs/xfs/xfs_inode.c | 14 +++++++++++++- fs/xfs/xfs_iops.c | 11 +++++------ fs/xfs/xfs_iops.h | 3 +++ 3 files changed, 21 insertions(+), 7 deletions(-)
fs/xfs/xfs_inode.c +13 -1
@@ -3046,10 +3046,12 @@ struct xfs_iunlink {static int xfs_rename_alloc_whiteout (struct user_namespace *mnt_userns, +struct xfs_name *src_name,struct xfs_inode *dp,struct xfs_inode **wip) {struct xfs_inode *tmpfile; +struct qstr name;int error; error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
@@ -3057,6 +3059,15 @@ struct xfs_iunlink {if (error)return error; + name.name = src_name->name; + name.len = src_name->len; + error =xfs_inode_init_security (VFS_I (tmpfile),VFS_I (dp), &name); +if (error) { +xfs_finish_inode_setup (tmpfile); +xfs_irele (tmpfile); +return error; + } +/* * Prepare the tmpfile inode as if it were created through the VFS. * Complete the inode setup and flag it as linkable. nlink is already
@@ -3107,7 +3118,8 @@ struct xfs_iunlink {* appropriately. */ if (flags &RENAME_WHITEOUT ) { - error =xfs_rename_alloc_whiteout (mnt_userns, target_dp, &wip); + error =xfs_rename_alloc_whiteout (mnt_userns, src_name, + target_dp, &wip);if (error)return error;
fs/xfs/xfs_iops.c +5 -6
@@ -75,9 +75,8 @@* these attrs can be journalled at inode creation time (along with the * inode, of course, such that log replay can't cause these to be lost). */ - -STATIC int -xfs_init_security ( +int +xfs_inode_init_security (struct inode *inode,struct inode *dir,const struct qstr *qstr)
@@ -122,7 +121,7 @@/* Oh, the horror. * If we can't add the ACL or we fail in -* xfs_init_security we must back out. +* xfs_inode_init_security we must back out. * ENOSPC can hit here, among other things. */ xfs_dentry_to_name (&teardown, dentry);
@@ -208,7 +207,7 @@ inode =VFS_I (ip); - error =xfs_init_security (inode, dir, &dentry->d_name); + error =xfs_inode_init_security (inode, dir, &dentry->d_name);if (unlikely (error)) goto out_cleanup_inode;
@@ -424,7 +423,7 @@ inode =VFS_I (cip); - error =xfs_init_security (inode, dir, &dentry->d_name); + error =xfs_inode_init_security (inode, dir, &dentry->d_name);if (unlikely (error)) goto out_cleanup_inode;
fs/xfs/xfs_iops.h +3 -0
@@ -17,4 +17,7 @@int xfs_vn_setattr_size (struct user_namespace *mnt_userns,struct dentry *dentry,struct iattr *vap); +int xfs_inode_init_security (struct inode *inode,struct inode *dir, +const struct qstr *qstr); +#endif /* __XFS_IOPS_H__ */
Signature · 1 line
1.8.3.1
Quoted 9 lines · Eric Sandeen
We got a report that "
renameat2()with flags=RENAME_WHITEOUT doesn't apply an SELinux label on xfs" as it does on other filesystems (for example, ext4 and tmpfs.) While I'm not quite sure how labels may interact w/ whiteout files, leaving them as unlabeled seems inconsistent at best. Now thatxfs_init_securityis not static, rename it toxfs_inode_init_securityper dchinner's suggestion.Signed-off-by Eric Sandeen <[email protected]>Reviewed-by Dave Chinner <[email protected]>
Looks fine to me. I wondered slightly if the label creation needs to be atomic with the file creation, but quickly realized that /never/ happens. Assuming this isn't high priority 5.19 stuff, I'll just roll this into 5.20 if that's ok?
--D
Quoted 107 lines
--- fs/xfs/xfs_inode.c | 14 +++++++++++++- fs/xfs/xfs_iops.c | 11 +++++------ fs/xfs/xfs_iops.h | 3 +++ 3 files changed, 21 insertions(+), 7 deletions(-)fs/xfs/xfs_inode.c +13
-1@@ -3046,10 +3046,12 @@ struct xfs_iunlink {static int xfs_rename_alloc_whiteout (struct user_namespace *mnt_userns, +struct xfs_name *src_name,struct xfs_inode *dp,struct xfs_inode **wip) {struct xfs_inode *tmpfile; +struct qstr name;int error; error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,@@ -3057,6 +3059,15 @@ struct xfs_iunlink {if (error)return error; + name.name = src_name->name; + name.len = src_name->len; + error =xfs_inode_init_security (VFS_I (tmpfile),VFS_I (dp), &name); +if (error) { +xfs_finish_inode_setup (tmpfile); +xfs_irele (tmpfile); +return error; + } +/* * Prepare the tmpfile inode as if it were created through the VFS. * Complete the inode setup and flag it as linkable. nlink is already @@ -3107,7 +3118,8 @@ struct xfs_iunlink {* appropriately. */ if (flags &RENAME_WHITEOUT ) { - error =xfs_rename_alloc_whiteout (mnt_userns, target_dp, &wip); + error =xfs_rename_alloc_whiteout (mnt_userns, src_name, + target_dp, &wip);if (error)return error;fs/xfs/xfs_iops.c +5
-6@@ -75,9 +75,8 @@* these attrs can be journalled at inode creation time (along with the * inode, of course, such that log replay can't cause these to be lost). */ - -STATIC int -xfs_init_security ( +int +xfs_inode_init_security (struct inode *inode,struct inode *dir,const struct qstr *qstr)@@ -122,7 +121,7 @@/* Oh, the horror. * If we can't add the ACL or we fail in -* xfs_init_security we must back out. +* xfs_inode_init_security we must back out. * ENOSPC can hit here, among other things. */ xfs_dentry_to_name (&teardown, dentry);@@ -208,7 +207,7 @@ inode =VFS_I (ip); - error =xfs_init_security (inode, dir, &dentry->d_name); + error =xfs_inode_init_security (inode, dir, &dentry->d_name);if (unlikely (error)) goto out_cleanup_inode;@@ -424,7 +423,7 @@ inode =VFS_I (cip); - error =xfs_init_security (inode, dir, &dentry->d_name); + error =xfs_inode_init_security (inode, dir, &dentry->d_name);if (unlikely (error)) goto out_cleanup_inode;fs/xfs/xfs_iops.h +3
-0@@ -17,4 +17,7 @@int xfs_vn_setattr_size (struct user_namespace *mnt_userns,struct dentry *dentry,struct iattr *vap); +int xfs_inode_init_security (struct inode *inode,struct inode *dir, +const struct qstr *qstr); +#endif /* __XFS_IOPS_H__ */ Signature · 1 line
1.8.3.1
Quoted 11 lines · Darrick J. Wong
Quoted 9 lines · Eric Sandeen
We got a report that "
renameat2()with flags=RENAME_WHITEOUT doesn't apply an SELinux label on xfs" as it does on other filesystems (for example, ext4 and tmpfs.) While I'm not quite sure how labels may interact w/ whiteout files, leaving them as unlabeled seems inconsistent at best. Now thatxfs_init_securityis not static, rename it toxfs_inode_init_securityper dchinner's suggestion.Signed-off-by Eric Sandeen <[email protected]>Reviewed-by Dave Chinner <[email protected]>
Quoted 6 lines
Looks fine to me. I wondered slightly if the label creation needs to be atomic with the file creation, but quickly realized that /never/ happens. Assuming this isn't high priority 5.19 stuff, I'll just roll this into 5.20 if that's ok?
Reviewed-by Darrick J. Wong <[email protected]>
Thanks Darrick. I don't think it's high priority, I got a bug report about the behavior, but there was no indication that it was actively causing visible problems.
-Eric