pnfs: don't check sequence on new stateids in layoutget
When layoutget returns an entirely new layout stateid it should not check the generation counter as the new stateid will start with a new counter entirely unrelated to old one. The current behavior causes constant layoutget failures against a block server which allocates a new stateid after an recall that removed all outstanding layouts. Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
This commit is contained in:
parent
1013df6115
commit
362f74745c
@ -682,17 +682,6 @@ static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
|
|||||||
return (s32)(s1 - s2) > 0;
|
return (s32)(s1 - s2) > 0;
|
||||||
}
|
}
|
||||||
|
|
||||||
static void
|
|
||||||
pnfs_verify_layout_stateid(struct pnfs_layout_hdr *lo,
|
|
||||||
const nfs4_stateid *new,
|
|
||||||
struct list_head *free_me_list)
|
|
||||||
{
|
|
||||||
if (nfs4_stateid_match_other(&lo->plh_stateid, new))
|
|
||||||
return;
|
|
||||||
/* Layout is new! Kill existing layout segments */
|
|
||||||
pnfs_mark_matching_lsegs_invalid(lo, free_me_list, NULL);
|
|
||||||
}
|
|
||||||
|
|
||||||
/* update lo->plh_stateid with new if is more recent */
|
/* update lo->plh_stateid with new if is more recent */
|
||||||
void
|
void
|
||||||
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
|
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
|
||||||
@ -1367,16 +1356,29 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
|
|||||||
goto out_forget_reply;
|
goto out_forget_reply;
|
||||||
}
|
}
|
||||||
|
|
||||||
if (pnfs_layoutgets_blocked(lo, 1) ||
|
if (pnfs_layoutgets_blocked(lo, 1)) {
|
||||||
pnfs_layout_stateid_blocked(lo, &res->stateid)) {
|
|
||||||
dprintk("%s forget reply due to state\n", __func__);
|
dprintk("%s forget reply due to state\n", __func__);
|
||||||
goto out_forget_reply;
|
goto out_forget_reply;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* Check that the new stateid matches the old stateid */
|
if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
|
||||||
pnfs_verify_layout_stateid(lo, &res->stateid, &free_me);
|
/* existing state ID, make sure the sequence number matches. */
|
||||||
/* Done processing layoutget. Set the layout stateid */
|
if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
|
||||||
pnfs_set_layout_stateid(lo, &res->stateid, false);
|
dprintk("%s forget reply due to sequence\n", __func__);
|
||||||
|
goto out_forget_reply;
|
||||||
|
}
|
||||||
|
pnfs_set_layout_stateid(lo, &res->stateid, false);
|
||||||
|
} else {
|
||||||
|
/*
|
||||||
|
* We got an entirely new state ID. Mark all segments for the
|
||||||
|
* inode invalid, and don't bother validating the stateid
|
||||||
|
* sequence number.
|
||||||
|
*/
|
||||||
|
pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
|
||||||
|
|
||||||
|
nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
|
||||||
|
lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
|
||||||
|
}
|
||||||
|
|
||||||
pnfs_get_lseg(lseg);
|
pnfs_get_lseg(lseg);
|
||||||
pnfs_layout_insert_lseg(lo, lseg);
|
pnfs_layout_insert_lseg(lo, lseg);
|
||||||
|
Loading…
Reference in New Issue
Block a user