generic: test that encrypted filenames are presented without collisions
PATCH 1/1 generic: test that encrypted filenames are presented without collisions
From: Eric Biggers <[email protected]>
[RFC for now since the kernel fixes are currently sitting in fscrypt/master and haven't been merged to Linus's tree yet; I'll resend this once they're merged]
Add a test which creates many similarly-named files in an encrypted directory, then verifies they can be deleted without access to the encryption key. This is a regression test for two related bugs which caused presented names to "collide" and point to the wrong inodes.
--- tests/generic/500 | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++ tests/generic/500.out | 4 +++ tests/generic/group | 1 + 3 files changed, 104 insertions(+) create mode 100755 tests/generic/500 create mode 100644 tests/generic/500.out
tests/generic/500 new file +99 -0
@@ -0,0 +1,99 @@ +#! /bin/bash +# FS QA Test generic/500 +# +# Test that without the encryption key for a directory, long filenames are +# presented in a way which avoids collisions, even though they are abbreviated +# in order to support names up to NAME_MAX bytes. +# +# Regression test for: +# TODO_COMMIT_ID ("f2fs: check entire encrypted bigname when finding a dentry") +# TODO_COMMIT_ID ("fscrypt: avoid collisions when presenting long encrypted filenames") +# +# Even with these two fixes it's still possible to create intentional +# collisions. For now this test covers "accidental" collisions only. +# +#----------------------------------------------------------------------- +# Copyright (c) 2017 Google, Inc. All Rights Reserved. +# +# Author: Eric Biggers <[email protected]> +# +# This program is free software; you can redistribute it and/or +# modify it under the terms of the GNU General Public License as +# published by the Free Software Foundation. +# +# This program is distributed in the hope that it would be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program; if not, write the Free Software Foundation, +# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA +#----------------------------------------------------------------------- +# + +seq=`basename $0` +seqres=$RESULT_DIR/$seq +echo "QA output created by $seq" + +here=`pwd` +tmp=/tmp/$$ +status=1 # failure is the default! +trap "_cleanup; exit \$status" 0 1 2 3 15 + +_cleanup() +{ + cd / + rm -f $tmp.* +} + +# get standard environment, filters and checks +. ./common/rc +. ./common/filter +. ./common/encrypt + +# remove previous $seqres.full before test +rm -f $seqres.full + +# real QA test starts here +_supported_fs generic +_supported_os Linux +_require_scratch_encryption +_require_xfs_io_command "set_encpolicy" +_require_command "$KEYCTL_PROG" keyctl + +# set up an encrypted directory + +_new_session_keyring +_scratch_mkfs_encrypted &>> $seqres.full +_scratch_mount +mkdir $SCRATCH_MNT/edir +keydesc=$(_generate_encryption_key) +# -f 0x2: zero-pad to 16-byte boundary (i.e. encryption block boundary) +$XFS_IO_PROG -c "set_encpolicy -f 0x2 $keydesc" $SCRATCH_MNT/edir + +# Create files with long names (> 32 bytes, long enough to trigger the use of +# "digested" names) in the encrypted directory. +# +# Use 100,000 files so that we have a good chance of detecting buggy filesystems +# that solely use a 32-bit hash to distinguish files, which f2fs was doing. +# +# Furthermore, make the filenames differ only in the last 16-byte encryption +# block. This reproduces the bug where it was not accounted for that ciphertext +# stealing (CTS) causes the last two blocks to appear "flipped". +seq -f "$SCRATCH_MNT/edir/abcdefghijklmnopqrstuvwxyz012345%.0f" 100000 | xargs touch +find $SCRATCH_MNT/edir/ -type f | xargs stat -c %i | sort | uniq | wc -l + +_unlink_encryption_key $keydesc +_scratch_cycle_mount + +# Verify that every file has a unique inode number and can be removed without +# error. With the bug(s), some filenames incorrectly pointed to the same inode, +# and ext4 reported a "Structure needs cleaning" error when removing files. +find $SCRATCH_MNT/edir/ -type f | xargs stat -c %i | sort | uniq | wc -l +rm -rf $SCRATCH_MNT/edir +stat $SCRATCH_MNT/edir |& _filter_scratch + +# success, all done +status=0 +exit
tests/generic/500.out new file +4 -0
@@ -0,0 +1,4 @@ +QA output created by 500 +100000 +100000 +stat: cannot stat 'SCRATCH_MNT/edir': No such file or directory
tests/generic/group +1 -0
@@ -431,3 +431,4 @@ 426 auto quick exportfs 427 auto quick aio rw 428 auto quick +500 auto encrypt
Signature · 1 line
2.13.0.rc1.294.g07d810a77f-goog
Quoted 5 lines · Eric Biggers
From: Eric Biggers <[email protected]>
[RFC for now since the kernel fixes are currently sitting in fscrypt/master and haven't been merged to Linus's tree yet; I'll resend this once they're merged]
Quoted 7 lines
Add a test which creates many similarly-named files in an encrypted directory, then verifies they can be deleted without access to the encryption key. This is a regression test for two related bugs which caused presented names to "collide" and point to the wrong inodes.
Signed-off-by Eric Biggers <[email protected]>
The test looks fine to me, it fails on ext4 with rm reporting "Structure needs cleaning".
One minor issue is that rm reports too many failures (around 98000 lines in my test)
+rm: cannot remove '/mnt/testarea/scratch/edir/_SET,lbqVCybotHoHkKrq4xM7FtbuBCLY': Structure needs cleaning
and that makes the output diff a bit harder to review, and takes too much space in 500.out.bad file unnecessarily (11M). Better to trim or limit the error log somehow, e.g. using "head -n 10"?
Thanks,
Eryu
Quoted 20 lines · Eryu Guan
Quoted 5 lines · Eric Biggers
From: Eric Biggers <[email protected]>
[RFC for now since the kernel fixes are currently sitting in fscrypt/master and haven't been merged to Linus's tree yet; I'll resend this once they're merged]
Quoted 7 lines
Add a test which creates many similarly-named files in an encrypted directory, then verifies they can be deleted without access to the encryption key. This is a regression test for two related bugs which caused presented names to "collide" and point to the wrong inodes.
Signed-off-by Eric Biggers <[email protected]>The test looks fine to me, it fails on ext4 with rm reporting "Structure needs cleaning".
One minor issue is that rm reports too many failures (around 98000 lines in my test)
Quoted 8 lines
+rm: cannot remove '
/mnt/testarea/scratch/edir/_SET,lbqVCybotHoHkKrq4xM7FtbuBCLY': Structure needs cleaningand that makes the output diff a bit harder to review, and takes too much space in 500.out.bad file unnecessarily (11M). Better to trim or limit the error log somehow, e.g. using "head -n 10"?
Thanks,
Eryu
Yes, I'll limit it to 10 error messages. Thanks!
Eric