d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix the NFSv4.1/pNFS client return layout on close.

Browse Source 
The "return layout on close" case in the pNFS client was badly broken.
Fortunately, extant pNFS servers that I have tested against do not
do this. This patch fixes it. It also changes the way the layout stateid.seqid
is set for LayoutReturn. I think this change is correct w.r.t. the RFC,
but I am not 100% sure.
This was found during recent testing of the pNFS server under development.

MFC after:	2 weeks
This commit is contained in:
Rick Macklem 2017-04-22 22:37:44 +00:00
parent 29924a1c12
commit 4e47dd1885
1 changed files with 33 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
sys/fs/nfsclient/nfs_clstate.c
View File

@ -88,6 +88,8 @@ extern struct nfsstatsv1 nfsstatsv1;
extern struct nfsreqhead nfsd_reqq;
extern u_int32_t newnfs_false, newnfs_true;
extern int nfscl_debuglevel;
extern int nfscl_enablecallb;
extern int nfs_numnfscbd;
NFSREQSPINLOCK;
NFSCLSTATEMUTEX;
int nfscl_inited = 0;
@ -118,7 +120,8 @@ static struct nfsclclient *nfscl_getclnt(u_int32_t);
static struct nfsclclient *nfscl_getclntsess(uint8_t *);
static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
int);
static void nfscl_retoncloselayout(struct nfsclclient *, uint8_t *, int);
static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
int, struct nfsclrecalllayout **);
static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
int);
@ -3121,6 +3124,7 @@ nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
struct nfsclopen *op;
struct nfscldeleg *dp;
struct nfsfh *nfhp;
struct nfsclrecalllayout *recallp;
int error;
error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
@ -3129,6 +3133,7 @@ nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
*clpp = clp;
nfhp = VTONFS(vp)->n_fhp;
recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
NFSLOCKCLSTATE();
/*
* First get rid of the local Open structures, which should be no
@ -3148,7 +3153,7 @@ nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
}
/* Return any layouts marked return on close. */
nfscl_retoncloselayout(clp, nfhp->nfh_fh, nfhp->nfh_len);
nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
/* Now process the opens against the server. */
lookformore:
@ -3171,6 +3176,11 @@ lookformore:
}
}
NFSUNLOCKCLSTATE();
/*
* recallp has been set NULL by nfscl_retoncloselayout() if it was
* used by the function, but calling free() with a NULL pointer is ok.
*/
free(recallp, M_NFSLAYRECALL);
return (0);
}
@ -4890,28 +4900,32 @@ nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
}
/*
* Search for a layout by MDS file handle. If one is found that is marked
* "return on close", delete it, since it should now be forgotten.
* Search for a layout by MDS file handle. If one is found, mark in to be
* recalled, if it already marked "return on close".
*/
static void
nfscl_retoncloselayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen)
nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
int fhlen, struct nfsclrecalllayout **recallpp)
{
struct nfscllayout *lyp;
uint32_t iomode;
tryagain:
if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) ||
nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
(VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
return;
lyp = nfscl_findlayout(clp, fhp, fhlen);
if (lyp != NULL && (lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
/*
* Wait for outstanding I/O ops to be done.
*/
if (lyp->nfsly_lock.nfslock_usecnt != 0 ||
lyp->nfsly_lock.nfslock_lock != 0) {
lyp->nfsly_lock.nfslock_lock |= NFSV4LOCK_WANTED;
(void)mtx_sleep(&lyp->nfsly_lock,
NFSCLSTATEMUTEXPTR, PZERO, "nfslyc", 0);
goto tryagain;
}
nfscl_freelayout(lyp);
if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE |
NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
iomode = 0;
if (!LIST_EMPTY(&lyp->nfsly_flayread))
iomode |= NFSLAYOUTIOMODE_READ;
if (!LIST_EMPTY(&lyp->nfsly_flayrw))
iomode |= NFSLAYOUTIOMODE_RW;
(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
0, UINT64_MAX, lyp->nfsly_stateid.seqid, *recallpp);
NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
*recallpp = NULL;
}
}
@ -5195,8 +5209,8 @@ nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
nfsv4stateid_t stateid;
NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
stateid.seqid = lyp->nfsly_stateid.seqid;
LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
stateid.seqid = rp->nfsrecly_stateseqid;
(void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
lyp->nfsly_fhlen, 0, NFSLAYOUT_NFSV4_1_FILES,
rp->nfsrecly_iomode, rp->nfsrecly_recalltype,