Re: [Qemu-devel] [PATCH 4/4] 9pfs: stat_to_qid: implement slow path

[Greg Kurz <groug@kaod.org>]

On Thu, 8 Feb 2018 19:00:19 +0100
<antonios.motakis@huawei.com> wrote:

> From: Antonios Motakis <antonios.motakis@huawei.com>
> 
> stat_to_qid attempts via qid_path_prefixmap to map unique files
> (which are identified by 64bt inode nr and 32 bit device id)
> to a 64 QID path value. However this implementation makes some
> assumptions about inode number generation on the host.
> 
> If qid_path_prefixmap fails, we still have 48bits available in
> the QID path to fall back to a less memory efficient full mapping.
> 
> Signed-off-by: Antonios Motakis <antonios.motakis@huawei.com>
> ---
>  hw/9pfs/9p.c | 66 +++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
>  hw/9pfs/9p.h |  9 +++++++++
>  2 files changed, 70 insertions(+), 5 deletions(-)
> 
> diff --git a/hw/9pfs/9p.c b/hw/9pfs/9p.c
> index f434f05..ae7845d 100644
> --- a/hw/9pfs/9p.c
> +++ b/hw/9pfs/9p.c
> @@ -581,23 +581,72 @@ static uint32_t qpp_hash(QppEntry e)
>      return tb_hash_func7(e.ino_prefix, e.dev, 0, 0, 0);
>  }
>  
> +static uint32_t qpf_hash(QpfEntry e)
> +{
> +    return tb_hash_func7(e.ino, e.dev, 0, 0, 0);

Same remark as with previous patch, obviously :)

> +}
> +
>  static bool qpp_lookup_func(const void *obj, const void *userp)
>  {
>      const QppEntry *e1 = obj, *e2 = userp;
>      return (e1->dev == e2->dev) && (e1->ino_prefix == e2->ino_prefix);

Parenthesitis ? ;)

>  }
>  
> -static void qpp_table_remove(struct qht *ht, void *p, uint32_t h, void *up)
> +static bool qpf_lookup_func(const void *obj, const void *userp)
> +{
> +    const QpfEntry *e1 = obj, *e2 = userp;
> +    return (e1->dev == e2->dev) && (e1->ino == e2->ino);

Parenthesitis ? ;)

> +}
> +
> +static void qp_table_remove(struct qht *ht, void *p, uint32_t h, void *up)
>  {
>      g_free(p);
>  }
>  
> -static void qpp_table_destroy(struct qht *ht)
> +static void qp_table_destroy(struct qht *ht)
>  {
> -    qht_iter(ht, qpp_table_remove, NULL);
> +    qht_iter(ht, qp_table_remove, NULL);
>      qht_destroy(ht);
>  }
>  
> +static int qid_path_fullmap(V9fsPDU *pdu, const struct stat *stbuf,
> +                            uint64_t *path)
> +{
> +    QpfEntry lookup = {
> +        .dev = stbuf->st_dev,
> +        .ino = stbuf->st_ino
> +    }, *val;
> +    uint32_t hash = qpf_hash(lookup);
> +
> +    /* most users won't need the fullmap, so init the table lazily */
> +    if (!pdu->s->qpf_table.map) {
> +        qht_init(&pdu->s->qpf_table, 1 << 16, QHT_MODE_AUTO_RESIZE);
> +    }
> +
> +    val = qht_lookup(&pdu->s->qpf_table, qpf_lookup_func, &lookup, hash);
> +
> +    if (!val) {
> +        if (pdu->s->qp_fullpath_next == 0) {
> +            /* no more files can be mapped :'( */
> +            return -ENFILE;
> +        }
> +
> +        val = g_malloc0(sizeof(QppEntry));
> +        if (!val) {
> +            return -ENOMEM;
> +        }
> +        *val = lookup;
> +
> +        /* new unique inode and device combo */
> +        val->path = pdu->s->qp_fullpath_next++;
> +        pdu->s->qp_fullpath_next &= QPATH_INO_MASK;
> +        qht_insert(&pdu->s->qpf_table, val, hash);
> +    }
> +
> +    *path = val->path;
> +    return 0;
> +}
> +
>  /* stat_to_qid needs to map inode number (64 bits) and device id (32 bits)
>   * to a unique QID path (64 bits). To avoid having to map and keep track
>   * of up to 2^64 objects, we map only the 16 highest bits of the inode plus
> @@ -646,6 +695,10 @@ static int stat_to_qid(V9fsPDU *pdu, const struct stat *stbuf, V9fsQID *qidp)
>  
>      /* map inode+device to qid path (fast path) */
>      err = qid_path_prefixmap(pdu, stbuf, &qidp->path);
> +    if (err == -ENFILE) {
> +        /* fast path didn't work, fal back to full map */
> +        err = qid_path_fullmap(pdu, stbuf, &qidp->path);

Hmm... if we have already generate QIDs with fast path, how
can we be sure we won't collide with the ones from the full
map ?

IIRC, Emilio had suggested to use bit 63 to distinguish between
fast and slow path.

> +    }
>      if (err) {
>          return err;
>      }
> @@ -3693,6 +3746,7 @@ int v9fs_device_realize_common(V9fsState *s, const V9fsTransport *t,
>      /* QID path hash table. 1 entry ought to be enough for anybody ;) */
>      qht_init(&s->qpp_table, 1, QHT_MODE_AUTO_RESIZE);
>      s->qp_prefix_next = 1; /* reserve 0 to detect overflow */
> +    s->qp_fullpath_next = 1;
>  
>      s->ctx.fst = &fse->fst;
>      fsdev_throttle_init(s->ctx.fst);
> @@ -3707,7 +3761,8 @@ out:
>          }
>          g_free(s->tag);
>          g_free(s->ctx.fs_root);
> -        qpp_table_destroy(&s->qpp_table);
> +        qp_table_destroy(&s->qpp_table);
> +        qp_table_destroy(&s->qpf_table);
>          v9fs_path_free(&path);
>      }
>      return rc;
> @@ -3720,7 +3775,8 @@ void v9fs_device_unrealize_common(V9fsState *s, Error **errp)
>      }
>      fsdev_throttle_cleanup(s->ctx.fst);
>      g_free(s->tag);
> -    qpp_table_destroy(&s->qpp_table);
> +    qp_table_destroy(&s->qpp_table);
> +    qp_table_destroy(&s->qpf_table);
>      g_free(s->ctx.fs_root);
>  }
>  
> diff --git a/hw/9pfs/9p.h b/hw/9pfs/9p.h
> index 80428f7..b561f4a 100644
> --- a/hw/9pfs/9p.h
> +++ b/hw/9pfs/9p.h
> @@ -241,6 +241,13 @@ typedef struct {
>      uint16_t qp_prefix;
>  } QppEntry;
>  
> +/* QID path full entry, as above */
> +typedef struct {
> +    dev_t dev;
> +    ino_t ino;
> +    uint64_t path;
> +} QpfEntry;
> +
>  struct V9fsState
>  {
>      QLIST_HEAD(, V9fsPDU) free_list;
> @@ -263,7 +270,9 @@ struct V9fsState
>      V9fsConf fsconf;
>      V9fsQID root_qid;
>      struct qht qpp_table;
> +    struct qht qpf_table;
>      uint16_t qp_prefix_next;
> +    uint64_t qp_fullpath_next;
>  };
>  
>  /* 9p2000.L open flags */