| Message ID | 20220414155236.249539-1-andrea.righi@canonical.com |
|---|---|
| State | New |
| Headers | show |
| Series | [SRU,J/I,1/1] UBUNTU: SAUCE: overlayfs: fix incorrect mnt_id of files opened from map_files | expand |
NAK-ing because it's missing cover letter email. Will send v2. -Andrea On Thu, Apr 14, 2022 at 05:52:36PM +0200, Andrea Righi wrote: > From: Alexander Mikhalitsyn <alexander@mihalicyn.com> > > BugLink: https://bugs.launchpad.net/bugs/1967924 > BugLink: https://bugs.launchpad.net/bugs/1857257 > > The hack was introduced in ("UBUNTU: SAUCE: overlayfs: allow with > shiftfs as underlay") and it broke checkpoint/restore of docker > contains: > https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1857257 > > The following script can be used to trigger the issue: > #!/bin/bash > > cat > test.py << EOF > import sys > > f = open("/proc/self/maps") > > for l in f.readlines(): > if "python" not in l: > continue > print(l) > s = l.split() > start, end = s[0].split("-") > fname = s[-1] > print(start, end, fname) > break > else: > sys.exit(1) > > test_file1 = open(fname) > test_file2 = open("/proc/self/map_files/%s-%s" % (start, end)) > > fdinfo1 = open("/proc/self/fdinfo/%d" % test_file1.fileno()).read() > fdinfo2 = open("/proc/self/fdinfo/%d" % test_file2.fileno()).read() > > if fdinfo1 != fdinfo2: > print("FAIL") > print(test_file1) > print(fdinfo1) > print(test_file2) > print(fdinfo2) > sys.exit(1) > print("PASS") > EOF > sudo docker run -it --privileged --rm -v `pwd`:/mnt python python /mnt/test.py > > Thanks to Andrei Vagin for the reproducer and investigation of this problem. > > Cc: Andrei Vagin <avagin@gmail.com> > Cc: Adrian Reber <areber@redhat.com> > Cc: Christian Brauner <christian.brauner@ubuntu.com> > Cc: Stefan Bader <stefan.bader@canonical.com> > Cc: Kleber Sacilotto de Souza <kleber.souza@canonical.com> > Cc: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com> > > Fixes: d24b8a5 ("UBUNTU: SAUCE: overlayfs: allow with shiftfs as underlay") > Signed-off-by: Alexander Mikhalitsyn <alexander@mihalicyn.com> > Acked-by: Seth Forshee <seth.forshee@canonical.com> > Acked-by: Stefan Bader <stefan.bader@canonical.com> > Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com> > Signed-off-by: Kelsey Skunberg <kelsey.skunberg@canonical.com> > [small refactoring to add the dependency of AUFS for vma->vm_prfile] > Signed-off-by: Andrea Righi <andrea.righi@canonical.com> > --- > fs/overlayfs/file.c | 39 +++++++++++++++++++++++++++++++++++++++ > 1 file changed, 39 insertions(+) > > diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c > index d0c96ca0202e..362dd17b8a00 100644 > --- a/fs/overlayfs/file.c > +++ b/fs/overlayfs/file.c > @@ -490,6 +490,41 @@ static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync) > return ret; > } > > +#if IS_ENABLED(CONFIG_AUFS_FS) > +/* > + * In map_files_get_link() (fs/proc/base.c) > + * we need to determine correct path from overlayfs. > + * But real_mount(realfile->f_path.mnt) may be not > + * equal to real_mount(file->f_path.mnt). In such case > + * fdinfo of the same file which was opened from > + * /proc/<pid>/map_files/... and "usual" path > + * will show different mnt_id. > + * > + * We solve issue like in aufs by using additional > + * field on struct vm_area_struct called "vm_prfile" > + * which is used only for fdinfo/"printing" needs. > + * > + * See also mm/prfile.c > + */ > +static void ovl_vm_prfile_set(struct vm_area_struct *vma, > + struct file *file) > +{ > + get_file(file); > + vma->vm_prfile = file; > +#ifndef CONFIG_MMU > + get_file(file); > + vma->vm_region->vm_prfile = file; > +#endif > + /* Drop reference count from previous vm_file value */ > + fput(file); > +} > +#else /* !CONFIG_AUFS_FS */ > +static void ovl_vm_prfile_set(struct vm_area_struct *vma, > + struct file *file) > +{ > +} > +#endif/* CONFIG_AUFS_FS */ > + > static int ovl_mmap(struct file *file, struct vm_area_struct *vma) > { > struct file *realfile = file->private_data; > @@ -507,6 +542,10 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma) > old_cred = ovl_override_creds(file_inode(file)->i_sb); > ret = call_mmap(vma->vm_file, vma); > revert_creds(old_cred); > + > + if (!ret) > + ovl_vm_prfile_set(vma, file); > + > ovl_file_accessed(file); > > return ret; > -- > 2.34.1
diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c index d0c96ca0202e..362dd17b8a00 100644 --- a/fs/overlayfs/file.c +++ b/fs/overlayfs/file.c @@ -490,6 +490,41 @@ static int ovl_fsync(struct file *file, loff_t start, loff_t end, int datasync) return ret; } +#if IS_ENABLED(CONFIG_AUFS_FS) +/* + * In map_files_get_link() (fs/proc/base.c) + * we need to determine correct path from overlayfs. + * But real_mount(realfile->f_path.mnt) may be not + * equal to real_mount(file->f_path.mnt). In such case + * fdinfo of the same file which was opened from + * /proc/<pid>/map_files/... and "usual" path + * will show different mnt_id. + * + * We solve issue like in aufs by using additional + * field on struct vm_area_struct called "vm_prfile" + * which is used only for fdinfo/"printing" needs. + * + * See also mm/prfile.c + */ +static void ovl_vm_prfile_set(struct vm_area_struct *vma, + struct file *file) +{ + get_file(file); + vma->vm_prfile = file; +#ifndef CONFIG_MMU + get_file(file); + vma->vm_region->vm_prfile = file; +#endif + /* Drop reference count from previous vm_file value */ + fput(file); +} +#else /* !CONFIG_AUFS_FS */ +static void ovl_vm_prfile_set(struct vm_area_struct *vma, + struct file *file) +{ +} +#endif/* CONFIG_AUFS_FS */ + static int ovl_mmap(struct file *file, struct vm_area_struct *vma) { struct file *realfile = file->private_data; @@ -507,6 +542,10 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma) old_cred = ovl_override_creds(file_inode(file)->i_sb); ret = call_mmap(vma->vm_file, vma); revert_creds(old_cred); + + if (!ret) + ovl_vm_prfile_set(vma, file); + ovl_file_accessed(file); return ret;