[Top] [All Lists]

Re: [PATCH 12/16] xfs: implement batched inode lookups for AG walking

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 12/16] xfs: implement batched inode lookups for AG walking
From: Alex Elder <aelder@xxxxxxx>
Date: Mon, 27 Sep 2010 12:43:29 -0500
Cc: xfs@xxxxxxxxxxx
In-reply-to: <20100924091522.GT2614@dastard>
References: <1285137869-10310-1-git-send-email-david@xxxxxxxxxxxxx> <1285137869-10310-13-git-send-email-david@xxxxxxxxxxxxx> <1285262225.1973.60.camel@doink> <20100924091522.GT2614@dastard>
Reply-to: aelder@xxxxxxx
On Fri, 2010-09-24 at 19:15 +1000, Dave Chinner wrote:
> On Thu, Sep 23, 2010 at 12:17:05PM -0500, Alex Elder wrote:
> > On Wed, 2010-09-22 at 16:44 +1000, Dave Chinner wrote:

. . .

> > It sounds like you're going to re-work this, but
> > I'll mention this for you to consider anyway.  I
> > don't know that the "done" flag here should be
> > needed.
> This check was added because if we don't detect the special case of
> the last valid inode _number_ in the AG, first_index will loop back
> to 0 and we'll start searching the AG again.  IOWs, we're not
> looking for the last inode in the cache, we're looking for the last
> valid inode number.
> Hence the done flag is ensuring that:
>       a) we terminate the walk at the last valid inode
>       b) if there are inodes at indexes above the last valid inode
>       number, we do not grab them or continue walking them.
> Yes, b) should never happen, but I've had bugs in development code
> that have put inodes in stange places before...
> > The gang lookup should never return
> > anything beyond the end of the AG.  It seems
> > like you ought to be able to detect when you've
> > covered all the whole AG elsewhere,
> AFAICT, there are only two ways - the gang lookup returns nothing,
> or we see the last valid inode number in the AG. If you can come up
> with something that doesn't invlove a tree or inode number lookup,
> I'm all ears....
> > *not*
> > on every entry found in this inner loop and
> > also *not* while holding the lock.
> It has to be done while holding the lock because if we cannot grab
> the inode then the only way we can safely derefence the inode is
> by still holding the inode cache lock. Once we drop the lock, the
> inodes we failed to grab can be removed from the cache and we cannot
> safely dereference them to get the inode number from them.

OK, I have a proposal below--it's not a diff, it's just a
modified version of xfs_inode_ag_walk() for you to consider.
It's not hugely better but it reduces the amount
of computation done inside the inner loop and while the
lock is held.  I haven't done any testing with it.

How this differs from what you have, probably in order
of importance:
- Update first_index only on the last inode returned by
  a gang lookup (not on every inode returned)
- Don't compare new value of first_index against the
  old one when it's updated.
- Use first_index == 0 (rather than done != 0) as an
  indication that we've exhausted the inodes in the AG.
- Tracks only grabbed inodes (rather than filling the
  array with null pointers in slots for those not grabbed)
- Don't bother looking at "error" again if it's zero
  (the normal case)

Anyway, do what you like with this (or do nothing at all).
I was just following up on my suggestion.


        struct xfs_mount        *mp,
        struct xfs_perag        *pag,
        int                     (*execute)(struct xfs_inode *ip,
                                           struct xfs_perag *pag, int
        int                     flags)
        uint32_t                first_index;
        int                     last_error = 0;
        int                     skipped;
        int                     nr_found;
        skipped = 0;
        first_index = 0;
        do {
                int             error = 0;
                int             nr_grabbed = 0;
                int             i;
                struct xfs_inode *batch[XFS_LOOKUP_BATCH];
                struct xfs_inode *ip; /* = NULL if compiler complains */
                nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
                                        (void **) batch, first_index,
                if (!nr_found) {
                /* Grab the inodes while we hold the lock. */
                for (i = 0; i < nr_found; i++) {
                        ip = batch[i];
                        if (!xfs_inode_ag_walk_grab(ip)) {
                                if (i > nr_grabbed)
                                            batch[nr_grabbed] = ip;

                 * Update the index so the next lookup starts after
                 * the last inode found (whether or not we were able
                 * to grab it).  If that inode was the highest one
                 * in the AG, this will evaluate to 0, which will
                 * cause the loop to terminate (below).
                first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);

                /* Done looking at (ungrabbed) inodes; drop the lock */
                for (i = 0; i < nr_grabbed; i++) {
                        ip = batch[i];
                        error = execute(ip, pag, flags);
                        if (error) {
                                if (error == EAGAIN) {
                                else if (last_error != EFSCORRUPTED)
                                        last_error = error;

                /* bail out if the filesystem is corrupted. */
                if (error == EFSCORRUPTED)
        } while (nr_found && first_index);

        if (skipped) {
                goto restart;

        return last_error;

<Prev in Thread] Current Thread [Next in Thread>