Re: [PATCH 5/6] src/fssum: Allow single file input

[Amir Goldstein <amir73il@gmail.com>]

On Fri, May 22, 2020 at 4:06 AM Arvind Raghavan
<raghavan.arvind@gmail.com> wrote:
>
> On 05/21, Amir Goldstein wrote:
> > On Thu, May 21, 2020 at 3:10 AM Arvind Raghavan
> > <raghavan.arvind@gmail.com> wrote:
> > >
> > > Allow regular links and symlinks to be passed as input to fssum.
> > >
> > > Signed-off-by: Arvind Raghavan <raghavan.arvind@gmail.com>
> > > Signed-off-by: Jayashree Mohan <jaya@cs.utexas.edu>
> > > Signed-off-by: Vijay Chidambaram <vijay@cs.utexas.edu>
> > > ---
> > >  src/fssum.c | 35 ++++++++++++++++++++++++++++++++++-
> > >  1 file changed, 34 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/src/fssum.c b/src/fssum.c
> > > index ece0f556..2d1624ca 100644
> > > --- a/src/fssum.c
> > > +++ b/src/fssum.c
> > > @@ -29,6 +29,7 @@
> > >  #include <inttypes.h>
> > >  #include <assert.h>
> > >  #include <endian.h>
> > > +#include <libgen.h>
> > >
> > >  #define CS_SIZE 16
> > >  #define CHUNKS 128
> > > @@ -884,8 +885,40 @@ main(int argc, char *argv[])
> > >         if (gen_manifest)
> > >                 fprintf(out_fp, "Flags: %s\n", flagstring);
> > >
> > > +       struct stat64 path_st;
> > > +       if (fstat64(fd, &path_st)) {
> > > +               perror("fstat");
> > > +               exit(-1);
> > > +       }
> > > +
> > >         sum_init(&cs);
> > > -       sum(fd, 1, &cs, path, "");
> > > +
> > > +       if (S_ISDIR(path_st.st_mode)) {
> > > +               sum(fd, 1, &cs, path, "");
> > > +       } else if (S_ISREG(path_st.st_mode) || S_ISLNK(path_st.st_mode)) {
> > > +               // Copy because dirname may modify path
> > > +               char* path_copy = alloc(strlen(path));
> > > +               strcpy(path_copy, path);

If you stay with this code please use strdup().

> > > +
> > > +               char* dir_path = dirname(path);
> > > +               char* name = basename(path_copy);
> > > +
> > > +               int dirfd = open(dir_path, O_RDONLY);
> > > +               if (fd == -1) {
> > > +                       fprintf(stderr, "failed to open %s: %s\n", dir_path,
> > > +                               strerror(errno));
> > > +                       exit(-1);
> > > +               }
> > > +
> > > +               sum_one(dirfd, 1, &cs, dir_path, "", name);
> >
> > Instead of all of the above, how about just:
> >                sum_one(fd, 1, &cs, path, "", "");
> >
> > From looking at sum_one() code, it seems to me like that will work,
> > but I may be missing something.
> > It's not that you *want* the name in the checksum, it is not even
> > part of the metadata that is being synced with fsync.
>
> The issue here is that we preserved the code from sum which does
> all its opens using openat with the parent directory fd and a
> filename. Since we're trying to reuse that code I believe we need
> to have this somewhat ugly boilerplate.

Ok. But if you stay with this please add a comment about why
this is done with a hint for the future how to fix this properly.

Or (up to you) you can fix it by calling this helper instead of openat():

int open_one(int dirfd, const char *name)
{
    if (!name || !*name)
        return dup(dirfd);
    return openat(dirfd, name, 0);
}

fstatat() can take empty name with AT_EMPTY_PATH flag.
readlinkat() should be able to take an empty name, but documentation
is not clear whether fd must be O_PATH - need to verify if it works with
non O_PATH fd.

Again, you don't have to do this to get my reviewed-by its just if you
want to and then of course do it in a prep patch, the same one that
gets rid of fchdir and converts to fstatat() and readlinkat().

>
> > Other than that patch set looks excellent.
> > Very pleasant for review :-)
>
> Thanks! :)
>
> > One little thing is missing from the cover letter -
> > Which tests did you run to verify these changes do not regress existing
> > tests?
>
> I just ran the relevant tests and encountered a small issue with
> the refactoring patch. This is my bad, since we changed lstat to
> use fstatat, we are no longer doing a fchdir which a readlink
> call later on relies on. I can fix it by changing the readlink to
> a readlinkat.
>

I see two valid options. please chose the one you like.

1. Revert removal of fchdir. let refactoring be only refactoring.
2. Remove fchdir and convert to fstatat/readlinkat in separate prep patch
    (with or without the empty name support suggested above)

> I'll add that change and add the set of relevant patches to the
> cover letter in a V2.

For patches that did not change from v1 please add my reviewed-by
so I know I do not need to re-review them.

Please include summary of "changes since v1" in cover letter.

Thanks,
Amir.