On Dienstag, 4. Oktober 2022 12:41:21 CEST Linus Heckemann wrote:
> The previous implementation would iterate over the fid table for
> lookup operations, resulting in an operation with O(n) complexity on
> the number of open files and poor cache locality -- for every open,
> stat, read, write, etc operation.
>
> This change uses a hashtable for this instead, significantly improving
> the performance of the 9p filesystem. The runtime of NixOS's simple
> installer test, which copies ~122k files totalling ~1.8GiB from 9p,
> decreased by a factor of about 10.
>
> Signed-off-by: Linus Heckemann <g...@sphalerite.org>
> Reviewed-by: Philippe Mathieu-Daudé <f4...@amsat.org>
> Reviewed-by: Greg Kurz <gr...@kaod.org>
> Message-Id: <20220908112353.289267-1-...@sphalerite.org>
> [CS: - Retain BUG_ON(f->clunked) in get_fid().
> - Add TODO comment in clunk_fid(). ]
> Signed-off-by: Christian Schoenebeck <qemu_...@crudebyte.com>
> ---
In general: LGTM now, but I will definitely go for some longer test runs
before queuing this patch. Some minor side notes below ...
> Christian Schoenebeck writes:
> > Not a big deal, but I start thinking whether to keep BUG_ON() here as
> > well.
> > That would require using g_hash_table_lookup() here instead of
> > g_hash_table_contains(). Not that I would insist.
>
> Done.
>
> > checkpatch.pl
>
> Fixed.
>
> > OK, I get it that you squashed your previous patch and that you ended up
> > with this comment instead. But if you read the old comment version here,
> > you'll notice that it describes the actual problem more accurately:
> > especially that v9fs_reopen_fid() and put_fid() are the 2 functions that
> > may cause the coroutine to "yield" here. That's an important information
> > to be retained in this comment here as it's not obvious.
>
> Reworded to mention these functions explicitly.
>
> > I would not move this to the 2nd loop. I would still set the
> >
> > FID_NON_RECLAIMABLE flag here, for the same reasons that you are bumping
> > the refcount already in the first loop below.
>
> Good point! Fixed.
>
> > ... that's not the same behaviour as before, is it? Old behaviour was to
> > put>
> > the respective (last) fid only on error. And this would now put all fids.
>
> Indeed it isn't the old behaviour, but I believe it's correct: since
> before we only incremented the reference counter for each one when we
> started reopening it. Now, we increment _all_ their refcounts, so we
> need to put all of them back as well.
Yeah, commented in v9fs_mark_fids_unreclaim() changes below ...
> > Wasn't there something like GLIST_FOREACH()? Then you wouldn't need to
> > add
> >
> > that variable.
>
> glib does provide g_list_foreach, but that requires a function
> pointer. I'm not aware of a macro for this. I could change this to use
> a QLIST (for which we do have a macro) instead of a GList if you
> insist, but I'd default to leaving this as is.
That's better handled by a future cleanup patch. Right now I find it
prioritized to get this performance fix merged ASAP, as it provides a
significant improvement for a lot of people.
I have seen GLIST_FOREACH macros (without function pointer) in other projects,
but I guess that macro was defined locally by those projects.
> Thanks for your repeated reviews and patience!
>
> Linus
>
>
> hw/9pfs/9p.c | 195 +++++++++++++++++++++++++++++----------------------
> hw/9pfs/9p.h | 2 +-
> 2 files changed, 113 insertions(+), 84 deletions(-)
>
> diff --git a/hw/9pfs/9p.c b/hw/9pfs/9p.c
> index aebadeaa03..729d3181e0 100644
> --- a/hw/9pfs/9p.c
> +++ b/hw/9pfs/9p.c
> @@ -256,7 +256,8 @@ static size_t v9fs_string_size(V9fsString *str)
> }
>
> /*
> - * returns 0 if fid got re-opened, 1 if not, < 0 on error */
> + * returns 0 if fid got re-opened, 1 if not, < 0 on error
> + */
> static int coroutine_fn v9fs_reopen_fid(V9fsPDU *pdu, V9fsFidState *f)
> {
> int err = 1;
> @@ -282,33 +283,32 @@ static V9fsFidState *coroutine_fn get_fid(V9fsPDU
> *pdu, int32_t fid) V9fsFidState *f;
> V9fsState *s = pdu->s;
>
> - QSIMPLEQ_FOREACH(f, &s->fid_list, next) {
> + f = g_hash_table_lookup(s->fids, GINT_TO_POINTER(fid));
> + if (f) {
> BUG_ON(f->clunked);
> - if (f->fid == fid) {
> - /*
> - * Update the fid ref upfront so that
> - * we don't get reclaimed when we yield
> - * in open later.
> - */
> - f->ref++;
> - /*
> - * check whether we need to reopen the
> - * file. We might have closed the fd
> - * while trying to free up some file
> - * descriptors.
> - */
> - err = v9fs_reopen_fid(pdu, f);
> - if (err < 0) {
> - f->ref--;
> - return NULL;
> - }
> - /*
> - * Mark the fid as referenced so that the LRU
> - * reclaim won't close the file descriptor
> - */
> - f->flags |= FID_REFERENCED;
> - return f;
> + /*
> + * Update the fid ref upfront so that
> + * we don't get reclaimed when we yield
> + * in open later.
> + */
> + f->ref++;
> + /*
> + * check whether we need to reopen the
> + * file. We might have closed the fd
> + * while trying to free up some file
> + * descriptors.
> + */
> + err = v9fs_reopen_fid(pdu, f);
> + if (err < 0) {
> + f->ref--;
> + return NULL;
> }
> + /*
> + * Mark the fid as referenced so that the LRU
> + * reclaim won't close the file descriptor
> + */
> + f->flags |= FID_REFERENCED;
> + return f;
> }
> return NULL;
> }
> @@ -317,12 +317,11 @@ static V9fsFidState *alloc_fid(V9fsState *s, int32_t
> fid) {
> V9fsFidState *f;
>
> - QSIMPLEQ_FOREACH(f, &s->fid_list, next) {
> + f = g_hash_table_lookup(s->fids, GINT_TO_POINTER(fid));
> + if (f) {
> /* If fid is already there return NULL */
> BUG_ON(f->clunked);
> - if (f->fid == fid) {
> - return NULL;
> - }
> + return NULL;
> }
> f = g_new0(V9fsFidState, 1);
> f->fid = fid;
> @@ -333,7 +332,7 @@ static V9fsFidState *alloc_fid(V9fsState *s, int32_t
> fid) * reclaim won't close the file descriptor
> */
> f->flags |= FID_REFERENCED;
> - QSIMPLEQ_INSERT_TAIL(&s->fid_list, f, next);
> + g_hash_table_insert(s->fids, GINT_TO_POINTER(fid), f);
>
> v9fs_readdir_init(s->proto_version, &f->fs.dir);
> v9fs_readdir_init(s->proto_version, &f->fs_reclaim.dir);
> @@ -424,12 +423,12 @@ static V9fsFidState *clunk_fid(V9fsState *s, int32_t
> fid) {
> V9fsFidState *fidp;
>
> - QSIMPLEQ_FOREACH(fidp, &s->fid_list, next) {
> - if (fidp->fid == fid) {
> - QSIMPLEQ_REMOVE(&s->fid_list, fidp, V9fsFidState, next);
> - fidp->clunked = true;
> - return fidp;
> - }
> + /* TODO: Use g_hash_table_steal_extended() instead? */
> + fidp = g_hash_table_lookup(s->fids, GINT_TO_POINTER(fid));
> + if (fidp) {
> + g_hash_table_remove(s->fids, GINT_TO_POINTER(fid));
> + fidp->clunked = true;
> + return fidp;
> }
> return NULL;
> }
> @@ -439,10 +438,15 @@ void coroutine_fn v9fs_reclaim_fd(V9fsPDU *pdu)
> int reclaim_count = 0;
> V9fsState *s = pdu->s;
> V9fsFidState *f;
> + GHashTableIter iter;
> + gpointer fid;
> +
> + g_hash_table_iter_init(&iter, s->fids);
> +
> QSLIST_HEAD(, V9fsFidState) reclaim_list =
> QSLIST_HEAD_INITIALIZER(reclaim_list);
>
> - QSIMPLEQ_FOREACH(f, &s->fid_list, next) {
> + while (g_hash_table_iter_next(&iter, &fid, (gpointer *) &f)) {
> /*
> * Unlink fids cannot be reclaimed. Check
> * for them and skip them. Also skip fids
> @@ -514,72 +518,86 @@ void coroutine_fn v9fs_reclaim_fd(V9fsPDU *pdu)
> }
> }
>
> +/*
> + * This is used when a path is removed from the directory tree. Any
> + * fids that still reference it must not be closed from then on, since
> + * they cannot be reopened.
> + */
> static int coroutine_fn v9fs_mark_fids_unreclaim(V9fsPDU *pdu, V9fsPath
> *path) {
> - int err;
> + int err = 0;
Not really necessary as `err` was and is never used unitialized, but OK.
> V9fsState *s = pdu->s;
> - V9fsFidState *fidp, *fidp_next;
> + V9fsFidState *fidp;
> + gpointer fid;
> + GHashTableIter iter;
> + /*
> + * The most common case is probably that we have exactly one
> + * fid for the given path, so preallocate exactly one.
> + */
> + g_autoptr(GArray) to_reopen = g_array_sized_new(FALSE, FALSE,
> + sizeof(V9fsFidState *), 1);
> + gint i;
>
> - fidp = QSIMPLEQ_FIRST(&s->fid_list);
> - if (!fidp) {
> - return 0;
> - }
> + g_hash_table_iter_init(&iter, s->fids);
>
> /*
> - * v9fs_reopen_fid() can yield : a reference on the fid must be held
> - * to ensure its pointer remains valid and we can safely pass it to
> - * QSIMPLEQ_NEXT(). The corresponding put_fid() can also yield so
> - * we must keep a reference on the next fid as well. So the logic here
> - * is to get a reference on a fid and only put it back during the next
> - * iteration after we could get a reference on the next fid. Start with
> - * the first one.
> + * We iterate over the fid table looking for the entries we need
> + * to reopen, and store them in to_reopen. This is because
> + * v9fs_reopen_fid() and put_fid() yield. This allows the fid table
> + * to be modified in the meantime, invalidating our iterator.
> */
> - for (fidp->ref++; fidp; fidp = fidp_next) {
> + while (g_hash_table_iter_next(&iter, &fid, (gpointer *) &fidp)) {
> if (fidp->path.size == path->size &&
> !memcmp(fidp->path.data, path->data, path->size)) {
> - /* Mark the fid non reclaimable. */
> - fidp->flags |= FID_NON_RECLAIMABLE;
> -
> - /* reopen the file/dir if already closed */
> - err = v9fs_reopen_fid(pdu, fidp);
> - if (err < 0) {
> - put_fid(pdu, fidp);
> - return err;
> - }
> - }
> -
> - fidp_next = QSIMPLEQ_NEXT(fidp, next);
> -
> - if (fidp_next) {
> /*
> - * Ensure the next fid survives a potential clunk request
> during
> - * put_fid() below and v9fs_reopen_fid() in the next
> iteration.
> + * Ensure the fid survives a potential clunk
> request during
> + * v9fs_reopen_fid or put_fid.
> */
> - fidp_next->ref++;
> + fidp->ref++;
> + fidp->flags |= FID_NON_RECLAIMABLE;
> + g_array_append_val(to_reopen, fidp);
> }
> + }
>
> - /* We're done with this fid */
> - put_fid(pdu, fidp);
> + for (i = 0; i < to_reopen->len; i++) {
> + fidp = g_array_index(to_reopen, V9fsFidState*, i);
> + /* reopen the file/dir if already closed */
> + err = v9fs_reopen_fid(pdu, fidp);
> + if (err < 0) {
> + goto out;
Looks like a simple `break;` here and dropping the `out:` label below is
sufficient. But I can do that on my end.
> + }
> }
>
> - return 0;
> + out:
> + for (i = 0; i < to_reopen->len; i++) {
> + put_fid(pdu, g_array_index(to_reopen, V9fsFidState*, i));
> + }
> + return err;
> }
... so yeah, I think that simple put_fid() loop is fine. Probably I overlooked
the 2nd put_fid() call in the original code. So basically the old == new
behaviour was and is: the sum of refcount increments/decrements after
returning from this function is exactly neutral/zero for each fid, without any
exception.
>
> static void coroutine_fn virtfs_reset(V9fsPDU *pdu)
> {
> V9fsState *s = pdu->s;
> V9fsFidState *fidp;
> + GList *freeing;
> + /*
> + * Get a list of all the values (fid states) in the table, which
> + * we then...
> + */
> + g_autoptr(GList) fids = g_hash_table_get_values(s->fids);
>
> - /* Free all fids */
> - while (!QSIMPLEQ_EMPTY(&s->fid_list)) {
> - /* Get fid */
> - fidp = QSIMPLEQ_FIRST(&s->fid_list);
> - fidp->ref++;
> + /* ... remove from the table, taking over ownership. */
> + g_hash_table_steal_all(s->fids);
>
> - /* Clunk fid */
> - QSIMPLEQ_REMOVE(&s->fid_list, fidp, V9fsFidState, next);
> + /*
> + * This allows us to release our references to them asynchronously
> without
> + * iterating over the hash table and risking iterator
> invalidation
> + * through concurrent modifications.
> + */
> + for (freeing = fids; freeing; freeing = freeing->next) {
> + fidp = freeing->data;
> + fidp->ref++;
> fidp->clunked = true;
> -
> put_fid(pdu, fidp);
> }
> }
> @@ -3205,6 +3223,8 @@ static int coroutine_fn v9fs_complete_rename(V9fsPDU
> *pdu, V9fsFidState *fidp, V9fsFidState *tfidp;
> V9fsState *s = pdu->s;
> V9fsFidState *dirfidp = NULL;
> + GHashTableIter iter;
> + gpointer fid;
>
> v9fs_path_init(&new_path);
> if (newdirfid != -1) {
> @@ -3238,11 +3258,13 @@ static int coroutine_fn v9fs_complete_rename(V9fsPDU
> *pdu, V9fsFidState *fidp, if (err < 0) {
> goto out;
> }
> +
> /*
> * Fixup fid's pointing to the old name to
> * start pointing to the new name
> */
> - QSIMPLEQ_FOREACH(tfidp, &s->fid_list, next) {
> + g_hash_table_iter_init(&iter, s->fids);
> + while (g_hash_table_iter_next(&iter, &fid, (gpointer *) &tfidp)) {
> if (v9fs_path_is_ancestor(&fidp->path, &tfidp->path)) {
> /* replace the name */
> v9fs_fix_path(&tfidp->path, &new_path,
> strlen(fidp->path.data)); @@ -3320,6 +3342,8 @@ static int coroutine_fn
> v9fs_fix_fid_paths(V9fsPDU *pdu, V9fsPath *olddir, V9fsPath oldpath,
> newpath;
> V9fsState *s = pdu->s;
> int err;
> + GHashTableIter iter;
> + gpointer fid;
>
> v9fs_path_init(&oldpath);
> v9fs_path_init(&newpath);
> @@ -3336,7 +3360,8 @@ static int coroutine_fn v9fs_fix_fid_paths(V9fsPDU
> *pdu, V9fsPath *olddir, * Fixup fid's pointing to the old name to
> * start pointing to the new name
> */
> - QSIMPLEQ_FOREACH(tfidp, &s->fid_list, next) {
> + g_hash_table_iter_init(&iter, s->fids);
> + while (g_hash_table_iter_next(&iter, &fid, (gpointer *) &tfidp)) {
> if (v9fs_path_is_ancestor(&oldpath, &tfidp->path)) {
> /* replace the name */
> v9fs_fix_path(&tfidp->path, &newpath, strlen(oldpath.data));
> @@ -4226,7 +4251,7 @@ int v9fs_device_realize_common(V9fsState *s, const
> V9fsTransport *t, s->ctx.fmode = fse->fmode;
> s->ctx.dmode = fse->dmode;
>
> - QSIMPLEQ_INIT(&s->fid_list);
> + s->fids = g_hash_table_new(NULL, NULL);
> qemu_co_rwlock_init(&s->rename_lock);
>
> if (s->ops->init(&s->ctx, errp) < 0) {
> @@ -4286,6 +4311,10 @@ void v9fs_device_unrealize_common(V9fsState *s)
> if (s->ctx.fst) {
> fsdev_throttle_cleanup(s->ctx.fst);
> }
> + if (s->fids) {
> + g_hash_table_destroy(s->fids);
> + s->fids = NULL;
> + }
> g_free(s->tag);
> qp_table_destroy(&s->qpd_table);
> qp_table_destroy(&s->qpp_table);
> diff --git a/hw/9pfs/9p.h b/hw/9pfs/9p.h
> index 994f952600..10fd2076c2 100644
> --- a/hw/9pfs/9p.h
> +++ b/hw/9pfs/9p.h
> @@ -339,7 +339,7 @@ typedef struct {
> struct V9fsState {
> QLIST_HEAD(, V9fsPDU) free_list;
> QLIST_HEAD(, V9fsPDU) active_list;
> - QSIMPLEQ_HEAD(, V9fsFidState) fid_list;
> + GHashTable *fids;
> FileOperations *ops;
> FsContext ctx;
> char *tag;
> --
> 2.36.2
