On Fri, Nov 30, 2018 at 10:57:58AM +0530, Chandan Rajendra wrote: > On Friday, November 30, 2018 12:35:13 AM IST Eric Biggers wrote: > > Hi Chandan, > > > > On Thu, Nov 29, 2018 at 04:08:31PM +0530, Chandan Rajendra wrote: > > > On Monday, November 26, 2018 11:04:35 PM IST Theodore Y. Ts'o wrote: > > > > On Sun, Nov 25, 2018 at 11:00:38PM -0500, Theodore Y. Ts'o wrote: > > > > > > > > > > It might be that the simplest way to solve things is to merge the f2fs > > > > > dev branch up to 79c66e75720c. This will have the net effect of > > > > > including the five patches listed above onto the fscrypt git tree. So > > > > > long you don't plan to rebase or otherwise change these five patches, > > > > > it should avoid any merge conflicts. > > > > > > > > I've set up a git branch which has the f2fs dev branch, 4.20-rc4, the > > > > fsverity patches, and part of Chandan's patch series here: > > > > > > > > git://git.kernel.org/pub/scm/linux/kernel/git/tytso/fscrypt.git test-working > > > > > > > > There is a minor conflict when I did a trial merge with f2fs.git's dev > > > > branch, but it's pretty obvious to how to resolve it. > > > > > > > > Jaegeuk, Eric, Chandan, please take a look and let me know what you > > > > think. > > > > > > Ted, > > > > > > I have addressed the review comments provided by Eric. Hence three out of > > > the four patches in the test-working branch have new changes. I also got > > > UBIFS to use CONFIG_FS_ENCRYPTION instead of the per-filesystem config > > > symbol. > > > > > > I am currently executing fstests to verify the changes. > > > > > > > > > Eric, > > > > > > When executing generic/900, I noticed that sometimes xfs_io would get stuck > > > for an indefinite period. /proc/<pid of xfs_io>/stack showed that the task was > > > stuck in tty_read() inside the kernel. The following change fixed it, > > > > > > diff --git a/tests/generic/900 b/tests/generic/900 > > > index 290889ce..0831eed4 100755 > > > --- a/tests/generic/900 > > > +++ b/tests/generic/900 > > > @@ -83,7 +83,7 @@ _fsv_create_enable_file $fsv_file >> $seqres.full > > > echo "* reading" > > > $XFS_IO_PROG -r $fsv_file -c '' > > > echo "* xfs_io writing, should be O_RDWR" > > > -$XFS_IO_PROG $fsv_file |& _filter_scratch > > > +$XFS_IO_PROG -c '' $fsv_file 2>&1 | _filter_scratch > > > echo "* bash >>, should be O_APPEND" > > > bash -c "echo >> $fsv_file" |& _filter_scratch > > > echo "* bash >, should be O_WRONLY|O_CREAT|O_TRUNC" > > > > > > xfs_io gets into interactive mode when invoked without a "-c cmd" string. > > > > > > However, I am not able recreate the scenario once again without the above > > > changes applied. I am not sure about what changed. > > > > > > > The test is opening a verity file for read+write access, which should fail. But > > it's incorrectly succeeding, hence the test is right to not pass. Did you add > > the missing call to ext4_set_inode_flags() in ext4_set_verity() as I > > suggested? > > > > (But I'll make the suggested change to the test too, so it fails cleanly in this > > case rather than hangs reading from stdin.) > > Yes, I did make the suggested changes. But the test would some times hang > indefinitely because of xfs_io waiting on input from stdin. > > With the new changes made to ext4_set_verity(), I see that the fsck fails > consistency check. But the failure is seen even without my patches applied. I > have planned to debug the failure after I post the next version of the > patchset. > You're testing with e2fsprogs "1.44.4-2" or later, right? Note that the original "v1.44.4" had some bugs with verity support, which were fixed shortly after. Specifically, e2fsprogs needs to be commit 3baafde6a8ae7 or later. - Eric
