| Message ID | 1300887304-20932-1-git-send-email-ext-phil.2.carmody@nokia.com |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, 2011-03-23 at 15:35 +0200, Phil Carmody wrote: > I knew I invented IS_ERR_OR_NULL for something, and this was probably > it. NULL has lost all information about what the error was, and the most > appropriate error code is ENODEV. However, that's the only error code > that the debugfs functions can return. So basically, any error = ENODEV. This is not true that any error is -ENODEV. There are many other errors possible, e.g., due to an invocation of simple_pin_fs(). I think the right fix would be to fix debugfs and return an error code in any case.
On 28/03/11 09:51 +0300, ext Artem Bityutskiy wrote: > On Wed, 2011-03-23 at 15:35 +0200, Phil Carmody wrote: > > I knew I invented IS_ERR_OR_NULL for something, and this was probably > > it. NULL has lost all information about what the error was, and the most > > appropriate error code is ENODEV. However, that's the only error code > > that the debugfs functions can return. So basically, any error = ENODEV. > > This is not true that any error is -ENODEV. There are many other errors > possible, e.g., due to an invocation of simple_pin_fs(). I meant that once you've thrown away all information about the actual error by turning it into NULL, then the broadest brush that covers the failure modes is ENODEV. That's why I was asked "what's the most generic 'something's gone wrong' code" on the chat channel the other day. > I think the right fix would be to fix debugfs and return an error code > in any case. Agree. Alas it might require hitting a lot of both active and dead code. Is that something you want me to look at? If you do it yourself, please Cc: me, I'll happily review it. Phil
On Mon, 2011-03-28 at 13:38 +0300, Phil Carmody wrote: > > I think the right fix would be to fix debugfs and return an error code > > in any case. > > Agree. Alas it might require hitting a lot of both active and dead code. > Is that something you want me to look at? If you do it yourself, please > Cc: me, I'll happily review it. No, I do not ask you do change debugfs. Do it only if you fare feeling enthusiastic about this :-) But I do not feel that way, so won't delve into the debugfs story :-) All I mean is that we should not do this: - if (IS_ERR(ptr)) { + if (IS_ERR_OR_NULL(ptr)) - int err = PTR_ERR(ptr); - return err; + return -ENODEV; But rather - if (IS_ERR(ptr)) { + if (IS_ERR_OR_NULL(ptr)) { - int err = PTR_ERR(ptr); + int err = ptr ? PTR_ERR(ptr) : -ENODEV; return err; or something like this.
diff --git a/fs/ubifs/debug.c b/fs/ubifs/debug.c index dbc093a..7b41b30 100644 --- a/fs/ubifs/debug.c +++ b/fs/ubifs/debug.c @@ -2578,11 +2578,10 @@ static struct dentry *dfs_rootdir; int dbg_debugfs_init(void) { dfs_rootdir = debugfs_create_dir("ubifs", NULL); - if (IS_ERR(dfs_rootdir)) { - int err = PTR_ERR(dfs_rootdir); + if (IS_ERR_OR_NULL(dfs_rootdir)) { ubifs_err("cannot create \"ubifs\" debugfs directory, " - "error %d\n", err); - return err; + "error %d\n", PTR_ERR(dfs_rootdir)); + return -ENODEV; } return 0; @@ -2645,47 +2644,44 @@ static const struct file_operations dfs_fops = { */ int dbg_debugfs_init_fs(struct ubifs_info *c) { - int err; const char *fname; struct dentry *dent; struct ubifs_debug_info *d = c->dbg; sprintf(d->dfs_dir_name, "ubi%d_%d", c->vi.ubi_num, c->vi.vol_id); d->dfs_dir = debugfs_create_dir(d->dfs_dir_name, dfs_rootdir); - if (IS_ERR(d->dfs_dir)) { - err = PTR_ERR(d->dfs_dir); + if (IS_ERR_OR_NULL(d->dfs_dir)) { ubifs_err("cannot create \"%s\" debugfs directory, error %d\n", - d->dfs_dir_name, err); + d->dfs_dir_name, PTR_ERR(d->dfs_rootdir)); goto out; } fname = "dump_lprops"; dent = debugfs_create_file(fname, S_IWUGO, d->dfs_dir, c, &dfs_fops); - if (IS_ERR(dent)) + if (IS_ERR_OR_NULL(dent)) goto out_remove; d->dfs_dump_lprops = dent; fname = "dump_budg"; dent = debugfs_create_file(fname, S_IWUGO, d->dfs_dir, c, &dfs_fops); - if (IS_ERR(dent)) + if (IS_ERR_OR_NULL(dent)) goto out_remove; d->dfs_dump_budg = dent; fname = "dump_tnc"; dent = debugfs_create_file(fname, S_IWUGO, d->dfs_dir, c, &dfs_fops); - if (IS_ERR(dent)) + if (IS_ERR_OR_NULL(dent)) goto out_remove; d->dfs_dump_tnc = dent; return 0; out_remove: - err = PTR_ERR(dent); ubifs_err("cannot create \"%s\" debugfs directory, error %d\n", - fname, err); + fname, PTR_ERR(dent)); debugfs_remove_recursive(d->dfs_dir); out: - return err; + return -ENODEV; } /**
I knew I invented IS_ERR_OR_NULL for something, and this was probably it. NULL has lost all information about what the error was, and the most appropriate error code is ENODEV. However, that's the only error code that the debugfs functions can return. So basically, any error = ENODEV. Signed-off-by: Phil Carmody <ext-phil.2.carmody@nokia.com> --- fs/ubifs/debug.c | 24 ++++++++++-------------- 1 files changed, 10 insertions(+), 14 deletions(-)