From: Jani Nikula <jani@nikula.org>
To: Austin Clements <amdragon@MIT.EDU>, notmuch@notmuchmail.org
Subject: Re: [PATCH v2 2/2] new: Centralize file type stat-ing logic
In-Reply-To: <87r4uvdryz.fsf@nikula.org>
References: <1336414186-15293-1-git-send-email-amdragon@mit.edu>
	<1336429240-1114-1-git-send-email-amdragon@mit.edu>
	<1336429240-1114-3-git-send-email-amdragon@mit.edu>
	<87r4uvdryz.fsf@nikula.org>
User-Agent: Notmuch/0.11.1+222~ga47a98c (http://notmuchmail.org) Emacs/23.1.1
	(i686-pc-linux-gnu)
Date: Tue, 08 May 2012 08:33:16 +0000
Message-ID: <87obpzdqcz.fsf@nikula.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Vladimir Marek <vlmarek@volny.cz>
Precedence: list

On Tue, 08 May 2012 07:58:28 +0000, Jani Nikula <jani@nikula.org> wrote:
> On Mon,  7 May 2012 18:20:40 -0400, Austin Clements <amdragon@MIT.EDU> wrote:
> > This moves our logic to get a file's type into one function.  This has
> > several benefits: we can support OSes and file systems that do not
> > provide dirent.d_type or always return DT_UNKNOWN, complex
> > symlink-handling logic has been replaced by a simple stat fall-through
> > in one place, and the error message for un-stat-able file is more
> > accurate (previously, the error always mentioned directories, even
> > though a broken symlink is not a directory).
> 
> LGTM.

Okay, it's good, but I think you can make it even better:

add_files_recursive() has check for "! S_ISDIR (st.st_mode)" in the
beginning, returning silently in case it recursed based on a symlink to
regular file. IIUC, this will no longer happen with your patch, as
symlinks are resolved and stat'ed before recursing.

add_files() exists to fail loudly in the same situation, and has
otherwise the same checks in the beginning. I think you could now use
the checks from add_files() to replace the ones in
add_files_recursive(), and get rid of add_files() altogether.

Please double check my thinking. Also this should probably be a separate
patch, no need to change the current one.

Jani.


> 
> > ---
> >  notmuch-new.c |  103 +++++++++++++++++++++++++++++++++++----------------------
> >  test/new      |    2 +-
> >  2 files changed, 64 insertions(+), 41 deletions(-)
> > 
> > diff --git a/notmuch-new.c b/notmuch-new.c
> > index cb720cc..8955677 100644
> > --- a/notmuch-new.c
> > +++ b/notmuch-new.c
> > @@ -154,6 +154,48 @@ dirent_sort_strcmp_name (const struct dirent **a, const struct dirent **b)
> >      return strcmp ((*a)->d_name, (*b)->d_name);
> >  }
> >  
> > +/* Return the type of a directory entry relative to path as a stat(2)
> > + * mode.  Like stat, this follows symlinks.  Returns -1 and sets errno
> > + * if the file's type cannot be determined (which includes dangling
> > + * symlinks).
> > + */
> > +static int
> > +dirent_type (const char *path, const struct dirent *entry)
> > +{
> > +    struct stat statbuf;
> > +    char *abspath;
> > +    int err, saved_errno;
> > +
> > +#ifdef _DIRENT_HAVE_D_TYPE
> > +    /* Mapping from d_type to stat mode_t.  We omit DT_LNK so that
> > +     * we'll fall through to stat and get the real file type. */
> > +    static const mode_t modes[] = {
> > +	[DT_BLK]  = S_IFBLK,
> > +	[DT_CHR]  = S_IFCHR,
> > +	[DT_DIR]  = S_IFDIR,
> > +	[DT_FIFO] = S_IFIFO,
> > +	[DT_REG]  = S_IFREG,
> > +	[DT_SOCK] = S_IFSOCK
> > +    };
> > +    if (entry->d_type < ARRAY_SIZE(modes) && modes[entry->d_type])
> > +	return modes[entry->d_type];
> > +#endif
> > +
> > +    abspath = talloc_asprintf (NULL, "%s/%s", path, entry->d_name);
> > +    if (!abspath) {
> > +	errno = ENOMEM;
> > +	return -1;
> > +    }
> > +    err = stat(abspath, &statbuf);
> > +    saved_errno = errno;
> > +    talloc_free (abspath);
> > +    if (err < 0) {
> > +	errno = saved_errno;
> > +	return -1;
> > +    }
> > +    return statbuf.st_mode & S_IFMT;
> > +}
> > +
> >  /* Test if the directory looks like a Maildir directory.
> >   *
> >   * Search through the array of directory entries to see if we can find all
> > @@ -162,12 +204,12 @@ dirent_sort_strcmp_name (const struct dirent **a, const struct dirent **b)
> >   * Return 1 if the directory looks like a Maildir and 0 otherwise.
> >   */
> >  static int
> > -_entries_resemble_maildir (struct dirent **entries, int count)
> > +_entries_resemble_maildir (const char *path, struct dirent **entries, int count)
> >  {
> >      int i, found = 0;
> >  
> >      for (i = 0; i < count; i++) {
> > -	if (entries[i]->d_type != DT_DIR && entries[i]->d_type != DT_UNKNOWN)
> > +	if (dirent_type (path, entries[i]) != S_IFDIR)
> >  	    continue;
> >  
> >  	if (strcmp(entries[i]->d_name, "new") == 0 ||
> > @@ -250,7 +292,7 @@ add_files_recursive (notmuch_database_t *notmuch,
> >      notmuch_status_t status, ret = NOTMUCH_STATUS_SUCCESS;
> >      notmuch_message_t *message = NULL;
> >      struct dirent **fs_entries = NULL;
> > -    int i, num_fs_entries;
> > +    int i, num_fs_entries, entry_type;
> >      notmuch_directory_t *directory;
> >      notmuch_filenames_t *db_files = NULL;
> >      notmuch_filenames_t *db_subdirs = NULL;
> > @@ -317,7 +359,7 @@ add_files_recursive (notmuch_database_t *notmuch,
> >      }
> >  
> >      /* Pass 1: Recurse into all sub-directories. */
> > -    is_maildir = _entries_resemble_maildir (fs_entries, num_fs_entries);
> > +    is_maildir = _entries_resemble_maildir (path, fs_entries, num_fs_entries);
> >  
> >      for (i = 0; i < num_fs_entries; i++) {
> >  	if (interrupted)
> > @@ -325,17 +367,16 @@ add_files_recursive (notmuch_database_t *notmuch,
> >  
> >  	entry = fs_entries[i];
> >  
> > -	/* We only want to descend into directories.
> > -	 * But symlinks can be to directories too, of course.
> > -	 *
> > -	 * And if the filesystem doesn't tell us the file type in the
> > -	 * scandir results, then it might be a directory (and if not,
> > -	 * then we'll stat and return immediately in the next level of
> > -	 * recursion). */
> > -	if (entry->d_type != DT_DIR &&
> > -	    entry->d_type != DT_LNK &&
> > -	    entry->d_type != DT_UNKNOWN)
> > -	{
> > +	/* We only want to descend into directories (and symlinks to
> > +	 * directories). */
> > +	entry_type = dirent_type (path, entry);
> > +	if (entry_type == -1) {
> > +	    /* Be pessimistic, e.g. so we don't lose lots of mail just
> > +	     * because a user broke a symlink. */
> > +	    fprintf (stderr, "Error reading file %s/%s: %s\n",
> > +		     path, entry->d_name, strerror (errno));
> > +	    return NOTMUCH_STATUS_FILE_ERROR;
> > +	} else if (entry_type != S_IFDIR) {
> >  	    continue;
> >  	}
> >  
> > @@ -425,31 +466,13 @@ add_files_recursive (notmuch_database_t *notmuch,
> >  	    notmuch_filenames_move_to_next (db_subdirs);
> >  	}
> >  
> > -	/* If we're looking at a symlink, we only want to add it if it
> > -	 * links to a regular file, (and not to a directory, say).
> > -	 *
> > -	 * Similarly, if the file is of unknown type (due to filesystem
> > -	 * limitations), then we also need to look closer.
> > -	 *
> > -	 * In either case, a stat does the trick.
> > -	 */
> > -	if (entry->d_type == DT_LNK || entry->d_type == DT_UNKNOWN) {
> > -	    int err;
> > -
> > -	    next = talloc_asprintf (notmuch, "%s/%s", path, entry->d_name);
> > -	    err = stat (next, &st);
> > -	    talloc_free (next);
> > -	    next = NULL;
> > -
> > -	    /* Don't emit an error for a link pointing nowhere, since
> > -	     * the directory-traversal pass will have already done
> > -	     * that. */
> > -	    if (err)
> > -		continue;
> > -
> > -	    if (! S_ISREG (st.st_mode))
> > -		continue;
> > -	} else if (entry->d_type != DT_REG) {
> > +	/* Only add regular files (and symlinks to regular files). */
> > +	entry_type = dirent_type (path, entry);
> > +	if (entry_type == -1) {
> > +	    fprintf (stderr, "Error reading file %s/%s: %s\n",
> > +		     path, entry->d_name, strerror (errno));
> > +	    return NOTMUCH_STATUS_FILE_ERROR;
> > +	} else if (entry_type != S_IFREG) {
> >  	    continue;
> >  	}
> >  
> > diff --git a/test/new b/test/new
> > index 26253db..e3900f5 100755
> > --- a/test/new
> > +++ b/test/new
> > @@ -140,7 +140,7 @@ test_begin_subtest "Broken symlink aborts"
> >  ln -s does-not-exist "${MAIL_DIR}/broken"
> >  output=$(NOTMUCH_NEW 2>&1)
> >  test_expect_equal "$output" \
> > -"Error reading directory /run/shm/nm/tmp.new/mail/broken: No such file or directory
> > +"Error reading file /run/shm/nm/tmp.new/mail/broken: No such file or directory
> >  Note: A fatal error was encountered: Something went wrong trying to read or write a file
> >  No new mail."
> >  rm "${MAIL_DIR}/broken"
> > -- 
> > 1.7.10
> > 
> > _______________________________________________
> > notmuch mailing list
> > notmuch@notmuchmail.org
> > http://notmuchmail.org/mailman/listinfo/notmuch