<p><br>
On May 8, 2012 12:08 AM, "Jani Nikula" <<a href="mailto:jani@nikula.org">jani@nikula.org</a>> wrote:<br>
><br>
><br>
> On May 7, 2012 9:10 PM, "Austin Clements" <<a href="mailto:amdragon@mit.edu">amdragon@mit.edu</a>> wrote:<br>
> ><br>
> > This moves our logic to get a file's type into one function. This has<br>
> > several benefits: we can support OSes and file systems that do not<br>
> > provide dirent.d_type or always return DT_UNKNOWN, complex<br>
> > symlink-handling logic has been replaced by a simple stat fall-through<br>
> > in one place, and the error message for un-stat-able file is more<br>
> > accurate (previously, the error always mentioned directories, even<br>
> > though a broken symlink is not a directory).<br>
><br>
> Please find some quick drive-by-review below.</p>
<p>Forgot to add that the general approach seems sensible to me.<br></p>
<p>><br>
> J.<br>
><br>
> > ---<br>
> > notmuch-new.c | 99 ++++++++++++++++++++++++++++++++++-----------------------<br>
> > test/new | 2 +-<br>
> > 2 files changed, 60 insertions(+), 41 deletions(-)<br>
> ><br>
> > diff --git a/notmuch-new.c b/notmuch-new.c<br>
> > index cb720cc..cf2580e 100644<br>
> > --- a/notmuch-new.c<br>
> > +++ b/notmuch-new.c<br>
> > @@ -154,6 +154,44 @@ dirent_sort_strcmp_name (const struct dirent **a, const struct dirent **b)<br>
> > return strcmp ((*a)->d_name, (*b)->d_name);<br>
> > }<br>
> ><br>
> > +/* Return the type of a directory entry relative to path as a stat(2)<br>
> > + * mode. Like stat, this follows symlinks. Returns -1 and sets errno<br>
> > + * if the file's type cannot be determined (which includes dangling<br>
> > + * symlinks).<br>
> > + */<br>
> > +static int<br>
> > +dirent_type (const char *path, const struct dirent *entry)<br>
> > +{<br>
> > + struct stat statbuf;<br>
> > + char *abspath;<br>
> > + int err;<br>
> > +<br>
> > +#ifdef _DIRENT_HAVE_D_TYPE<br>
> > + /* Mapping from d_type to stat mode_t. We omit DT_LNK so that<br>
> > + * we'll fall through to stat and get the real file type. */<br>
> > + static const mode_t modes[] = {<br>
> > + [DT_BLK] = S_IFBLK,<br>
> > + [DT_CHR] = S_IFCHR,<br>
> > + [DT_DIR] = S_IFDIR,<br>
> > + [DT_FIFO] = S_IFIFO,<br>
> > + [DT_REG] = S_IFREG,<br>
> > + [DT_SOCK] = S_IFSOCK<br>
> > + };<br>
> > + if (entry->d_type < sizeof(modes)/sizeof(modes[0]) &&<br>
><br>
> ARRAY_SIZE()<br>
><br>
> > + modes[entry->d_type])<br>
> > + return modes[entry->d_type];<br>
> > +#endif<br>
> > +<br>
> > + abspath = talloc_asprintf (NULL, "%s/%s", path, entry->d_name);<br>
> > + if (!abspath)<br>
> > + return -1;<br>
><br>
> Does talloc set errno in this case? I suspect not.<br>
><br>
> > + err = stat(abspath, &statbuf);<br>
> > + talloc_free (abspath);<br>
><br>
> This likely breaks your promise about errno. You can't trust talloc_free() not calling some function that sets errno.<br>
><br>
> > + if (err < 0)<br>
> > + return -1;<br>
> > + return statbuf.st_mode & S_IFMT;<br>
> > +}<br>
> > +<br>
> > /* Test if the directory looks like a Maildir directory.<br>
> > *<br>
> > * Search through the array of directory entries to see if we can find all<br>
> > @@ -162,12 +200,12 @@ dirent_sort_strcmp_name (const struct dirent **a, const struct dirent **b)<br>
> > * Return 1 if the directory looks like a Maildir and 0 otherwise.<br>
> > */<br>
> > static int<br>
> > -_entries_resemble_maildir (struct dirent **entries, int count)<br>
> > +_entries_resemble_maildir (const char *path, struct dirent **entries, int count)<br>
> > {<br>
> > int i, found = 0;<br>
> ><br>
> > for (i = 0; i < count; i++) {<br>
> > - if (entries[i]->d_type != DT_DIR && entries[i]->d_type != DT_UNKNOWN)<br>
> > + if (dirent_type (path, entries[i]) != S_IFDIR)<br>
> > continue;<br>
> ><br>
> > if (strcmp(entries[i]->d_name, "new") == 0 ||<br>
> > @@ -250,7 +288,7 @@ add_files_recursive (notmuch_database_t *notmuch,<br>
> > notmuch_status_t status, ret = NOTMUCH_STATUS_SUCCESS;<br>
> > notmuch_message_t *message = NULL;<br>
> > struct dirent **fs_entries = NULL;<br>
> > - int i, num_fs_entries;<br>
> > + int i, num_fs_entries, entry_type;<br>
> > notmuch_directory_t *directory;<br>
> > notmuch_filenames_t *db_files = NULL;<br>
> > notmuch_filenames_t *db_subdirs = NULL;<br>
> > @@ -317,7 +355,7 @@ add_files_recursive (notmuch_database_t *notmuch,<br>
> > }<br>
> ><br>
> > /* Pass 1: Recurse into all sub-directories. */<br>
> > - is_maildir = _entries_resemble_maildir (fs_entries, num_fs_entries);<br>
> > + is_maildir = _entries_resemble_maildir (path, fs_entries, num_fs_entries);<br>
> ><br>
> > for (i = 0; i < num_fs_entries; i++) {<br>
> > if (interrupted)<br>
> > @@ -325,17 +363,16 @@ add_files_recursive (notmuch_database_t *notmuch,<br>
> ><br>
> > entry = fs_entries[i];<br>
> ><br>
> > - /* We only want to descend into directories.<br>
> > - * But symlinks can be to directories too, of course.<br>
> > - *<br>
> > - * And if the filesystem doesn't tell us the file type in the<br>
> > - * scandir results, then it might be a directory (and if not,<br>
> > - * then we'll stat and return immediately in the next level of<br>
> > - * recursion). */<br>
> > - if (entry->d_type != DT_DIR &&<br>
> > - entry->d_type != DT_LNK &&<br>
> > - entry->d_type != DT_UNKNOWN)<br>
> > - {<br>
> > + /* We only want to descend into directories (and symlinks to<br>
> > + * directories). */<br>
> > + entry_type = dirent_type (path, entry);<br>
> > + if (entry_type == -1) {<br>
> > + /* Be pessimistic, e.g. so we don't loose lots of mail<br>
><br>
> s/loose/lose/ ?<br>
><br>
> > + * just because a user broke a symlink. */<br>
> > + fprintf (stderr, "Error reading file %s/%s: %s\n",<br>
> > + path, entry->d_name, strerror (errno));<br>
><br>
> You can't trust errno here, as explained above.<br>
><br>
> > + return NOTMUCH_STATUS_FILE_ERROR;<br>
> > + } else if (entry_type != S_IFDIR) {<br>
> > continue;<br>
> > }<br>
> ><br>
> > @@ -425,31 +462,13 @@ add_files_recursive (notmuch_database_t *notmuch,<br>
> > notmuch_filenames_move_to_next (db_subdirs);<br>
> > }<br>
> ><br>
> > - /* If we're looking at a symlink, we only want to add it if it<br>
> > - * links to a regular file, (and not to a directory, say).<br>
> > - *<br>
> > - * Similarly, if the file is of unknown type (due to filesystem<br>
> > - * limitations), then we also need to look closer.<br>
> > - *<br>
> > - * In either case, a stat does the trick.<br>
> > - */<br>
> > - if (entry->d_type == DT_LNK || entry->d_type == DT_UNKNOWN) {<br>
> > - int err;<br>
> > -<br>
> > - next = talloc_asprintf (notmuch, "%s/%s", path, entry->d_name);<br>
> > - err = stat (next, &st);<br>
> > - talloc_free (next);<br>
> > - next = NULL;<br>
> > -<br>
> > - /* Don't emit an error for a link pointing nowhere, since<br>
> > - * the directory-traversal pass will have already done<br>
> > - * that. */<br>
> > - if (err)<br>
> > - continue;<br>
> > -<br>
> > - if (! S_ISREG (st.st_mode))<br>
> > - continue;<br>
> > - } else if (entry->d_type != DT_REG) {<br>
> > + /* Only add regular files (and symlinks to regular files). */<br>
> > + entry_type = dirent_type (path, entry);<br>
> > + if (entry_type == -1) {<br>
> > + fprintf (stderr, "Error reading file %s/%s: %s\n",<br>
> > + path, entry->d_name, strerror (errno));<br>
><br>
> Ditto.<br>
><br>
> > + return NOTMUCH_STATUS_FILE_ERROR;<br>
> > + } else if (entry_type != S_IFREG) {<br>
> > continue;<br>
> > }<br>
> ><br>
> > diff --git a/test/new b/test/new<br>
> > index 26253db..e3900f5 100755<br>
> > --- a/test/new<br>
> > +++ b/test/new<br>
> > @@ -140,7 +140,7 @@ test_begin_subtest "Broken symlink aborts"<br>
> > ln -s does-not-exist "${MAIL_DIR}/broken"<br>
> > output=$(NOTMUCH_NEW 2>&1)<br>
> > test_expect_equal "$output" \<br>
> > -"Error reading directory /run/shm/nm/tmp.new/mail/broken: No such file or directory<br>
> > +"Error reading file /run/shm/nm/tmp.new/mail/broken: No such file or directory<br>
> > Note: A fatal error was encountered: Something went wrong trying to read or write a file<br>
> > No new mail."<br>
> > rm "${MAIL_DIR}/broken"<br>
> > --<br>
> > 1.7.10<br>
> ><br>
> > _______________________________________________<br>
> > notmuch mailing list<br>
> > <a href="mailto:notmuch@notmuchmail.org">notmuch@notmuchmail.org</a><br>
> > <a href="http://notmuchmail.org/mailman/listinfo/notmuch">http://notmuchmail.org/mailman/listinfo/notmuch</a><br>
</p>