On Tue, Jun 8, 2010 at 12:14 AM, Benny Halevy <bhalevy@xxxxxxxxxxx> wrote:
> On 2010-06-08 00:11, Alexandros Batsakis wrote:
>> In the first iteration of the pNFS code, we support only whole file layouts.
>> To facilitate the move to multiple-segments, we keep the segment processing
>> code, but the segment list should always contain at most one segment per I/O type
>>
>> Signed-off-by: Alexandros Batsakis <batsakis@xxxxxxxxxx>
>> ---
>> fs/nfs/callback_proc.c | 7 ++++---
>> fs/nfs/pnfs.c | 25 ++++++++-----------------
>> 2 files changed, 12 insertions(+), 20 deletions(-)
>>
>> diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
>> index 8752239..16b4510 100644
>> --- a/fs/nfs/callback_proc.c
>> +++ b/fs/nfs/callback_proc.c
>> @@ -213,6 +213,10 @@ static int pnfs_recall_layout(void *data)
>> then return layouts, resume after layoutreturns complete
>> */
>>
>> + /* support whole file layouts only */
>> + rl.cbl_seg.offset = 0;
>> + rl.cbl_seg.length = NFS4_MAX_UINT64;
>> +
>> if (rl.cbl_recall_type == RETURN_FILE) {
>> status = pnfs_return_layout(inode, &rl.cbl_seg, &rl.cbl_stateid,
>> RETURN_FILE, true);
>> @@ -221,9 +225,6 @@ static int pnfs_recall_layout(void *data)
>> goto out;
>> }
>>
>> - rl.cbl_seg.offset = 0;
>> - rl.cbl_seg.length = NFS4_MAX_UINT64;
>> -
>> /* FIXME: This loop is inefficient, running in O(|s_inodes|^2) */
>> while ((ino = nfs_layoutrecall_find_inode(clp, &rl)) != NULL) {
>> /* XXX need to check status on pnfs_return_layout */
>> diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
>> index dd7ad26..8985e9f 100644
>> --- a/fs/nfs/pnfs.c
>> +++ b/fs/nfs/pnfs.c
>> @@ -551,12 +551,6 @@ pnfs_layout_from_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
>> * for now, assume that whole file layouts are requested.
>> * arg->offset: 0
>> * arg->length: all ones
>> -*
>> -* for now, assume the LAYOUTGET operation is triggered by an I/O request.
>> -* the count field is the count in the I/O request, and will be used
>> -* as the minlength. for the file operation that piggy-backs
>> -* the LAYOUTGET operation with an OPEN, s
>> -* arg->minlength = count.
>> */
>> static int
>> get_layout(struct inode *ino,
>> @@ -577,11 +571,11 @@ get_layout(struct inode *ino,
>> return -ENOMEM;
>> }
>> lgp->lo = lo;
>> - lgp->args.minlength = PAGE_CACHE_SIZE;
>> + lgp->args.minlength = NFS4_MAX_UINT64;
>> lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
>> lgp->args.lseg.iomode = range->iomode;
>> - lgp->args.lseg.offset = range->offset;
>> - lgp->args.lseg.length = max(range->length, lgp->args.minlength);
>> + lgp->args.lseg.offset = 0;
>> + lgp->args.lseg.length = NFS4_MAX_UINT64;
>
> I'm not sure how well this will work for pnfs-obj writes.
> We might have to revert it in the pnfs post-submit branch.
>
> Benny
>
Agreed... but bear in mind that there are other things that need to
change too to properly (and efficiently) support partial layouts (e.g.
layout operation sequencing). So maybe a fresh patch will be more
appropriate, but it's up to you of course.
-alexandros
>> lgp->args.type = server->pnfs_curr_ld->id;
>> lgp->args.inode = ino;
>> lgp->lsegpp = lsegpp;
>> @@ -756,7 +750,7 @@ _pnfs_return_layout(struct inode *ino, struct nfs4_pnfs_layout_segment *range,
>> else {
>> arg.iomode = IOMODE_ANY;
>> arg.offset = 0;
>> - arg.length = ~0;
>> + arg.length = NFS4_MAX_UINT64;
>> }
>> if (type == RETURN_FILE) {
>> lo = get_lock_current_layout(nfsi);
>> @@ -1051,8 +1045,8 @@ pnfs_update_layout(struct inode *ino,
>> {
>> struct nfs4_pnfs_layout_segment arg = {
>> .iomode = iomode,
>> - .offset = pos,
>> - .length = count
>> + .offset = 0,
>> + .length = ~0
>> };
>> struct nfs_inode *nfsi = NFS_I(ino);
>> struct pnfs_layout_type *lo;
>> @@ -1142,7 +1136,6 @@ out_put:
>> void
>> pnfs_get_layout_done(struct nfs4_pnfs_layoutget *lgp, int rpc_status)
>> {
>> - struct nfs4_pnfs_layoutget_res *res = &lgp->res;
>> struct pnfs_layout_segment *lseg = NULL;
>> struct nfs_inode *nfsi = PNFS_NFS_INODE(lgp->lo);
>> time_t suspend = 0;
>> @@ -1151,11 +1144,10 @@ pnfs_get_layout_done(struct nfs4_pnfs_layoutget *lgp, int rpc_status)
>>
>> lgp->status = rpc_status;
>> if (likely(!rpc_status)) {
>> - if (unlikely(res->layout.len <= 0)) {
>> + if (unlikely(lgp->res.layout.len < 0)) {
>> printk(KERN_ERR
>> - "%s: ERROR! Layout size is ZERO!\n", __func__);
>> + "%s: ERROR Returned layout size is ZERO\n", __func__);
>> lgp->status = -EIO;
>> - goto get_out;
>> }
>> goto out;
>> }
>> @@ -1233,7 +1225,6 @@ pnfs_get_layout_done(struct nfs4_pnfs_layoutget *lgp, int rpc_status)
>> break;
>> }
>>
>> -get_out:
>> /* remember that get layout failed and suspend trying */
>> nfsi->layout.pnfs_layout_suspend = suspend;
>> set_bit(lo_fail_bit(lgp->args.lseg.iomode),
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
