diff options
| -rw-r--r-- | sys/fs/nfs/nfs_var.h | 2 | ||||
| -rw-r--r-- | sys/fs/nfsclient/nfs_clrpcops.c | 1114 | ||||
| -rw-r--r-- | sys/fs/nfsclient/nfs_clstate.c | 55 |
3 files changed, 947 insertions, 224 deletions
diff --git a/sys/fs/nfs/nfs_var.h b/sys/fs/nfs/nfs_var.h index c6cc3c619602..86db40ec1521 100644 --- a/sys/fs/nfs/nfs_var.h +++ b/sys/fs/nfs/nfs_var.h @@ -555,7 +555,7 @@ int nfscl_tryclose(struct nfsclopen *, struct ucred *, struct nfsmount *, NFSPROC_T *); void nfscl_cleanup(NFSPROC_T *); int nfscl_layout(struct nfsmount *, vnode_t, u_int8_t *, int, nfsv4stateid_t *, - int, struct nfsclflayouthead *, struct nfscllayout **, struct ucred *, + int, int, struct nfsclflayouthead *, struct nfscllayout **, struct ucred *, NFSPROC_T *); struct nfscllayout *nfscl_getlayout(struct nfsclclient *, uint8_t *, int, uint64_t, struct nfsclflayout **, int *); diff --git a/sys/fs/nfsclient/nfs_clrpcops.c b/sys/fs/nfsclient/nfs_clrpcops.c index 926d8e617c59..4c9e47740423 100644 --- a/sys/fs/nfsclient/nfs_clrpcops.c +++ b/sys/fs/nfsclient/nfs_clrpcops.c @@ -116,18 +116,28 @@ static int nfscl_doflayoutio(vnode_t, struct uio *, int *, int *, int *, nfsv4stateid_t *, int, struct nfscldevinfo *, struct nfscllayout *, struct nfsclflayout *, uint64_t, uint64_t, int, struct ucred *, NFSPROC_T *); +static int nfscl_dofflayoutio(vnode_t, struct uio *, int *, int *, int *, + nfsv4stateid_t *, int, struct nfscldevinfo *, struct nfscllayout *, + struct nfsclflayout *, uint64_t, uint64_t, int, int, struct mbuf *, + struct ucred *, NFSPROC_T *); +static struct mbuf *nfsm_copym(struct mbuf *, int, int); static int nfsrpc_readds(vnode_t, struct uio *, nfsv4stateid_t *, int *, - struct nfsclds *, uint64_t, int, struct nfsfh *, struct ucred *, - NFSPROC_T *); + struct nfsclds *, uint64_t, int, struct nfsfh *, int, int, int, + struct ucred *, NFSPROC_T *); static int nfsrpc_writeds(vnode_t, struct uio *, int *, int *, nfsv4stateid_t *, struct nfsclds *, uint64_t, int, - struct nfsfh *, int, struct ucred *, NFSPROC_T *); + struct nfsfh *, int, int, int, int, struct ucred *, NFSPROC_T *); +static int nfsrpc_writedsmir(vnode_t, int *, int *, nfsv4stateid_t *, + struct nfsclds *, uint64_t, int, struct nfsfh *, struct mbuf *, int, int, + struct ucred *, NFSPROC_T *); static enum nfsclds_state nfscl_getsameserver(struct nfsmount *, struct nfsclds *, struct nfsclds **); static int nfsrpc_commitds(vnode_t, uint64_t, int, struct nfsclds *, - struct nfsfh *, struct ucred *, NFSPROC_T *); + struct nfsfh *, int, int, struct ucred *, NFSPROC_T *); static void nfsrv_setuplayoutget(struct nfsrv_descript *, int, uint64_t, - uint64_t, uint64_t, nfsv4stateid_t *, int, int); + uint64_t, uint64_t, nfsv4stateid_t *, int, int, int); +static int nfsrv_parseug(struct nfsrv_descript *, int, uid_t *, gid_t *, + NFSPROC_T *); static int nfsrv_parselayoutget(struct nfsrv_descript *, nfsv4stateid_t *, int *, struct nfsclflayouthead *); static int nfsrpc_getopenlayout(struct nfsmount *, vnode_t, u_int8_t *, @@ -139,19 +149,19 @@ static int nfsrpc_getcreatelayout(vnode_t, char *, int, struct vattr *, struct nfsfh **, int *, int *, void *, int *); static int nfsrpc_openlayoutrpc(struct nfsmount *, vnode_t, u_int8_t *, int, uint8_t *, int, uint32_t, struct nfsclopen *, uint8_t *, int, - struct nfscldeleg **, nfsv4stateid_t *, int, int, int *, + struct nfscldeleg **, nfsv4stateid_t *, int, int, int, int *, struct nfsclflayouthead *, int *, struct ucred *, NFSPROC_T *); static int nfsrpc_createlayout(vnode_t, char *, int, struct vattr *, nfsquad_t, int, struct nfsclowner *, struct nfscldeleg **, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *, int *, nfsv4stateid_t *, - int, int, int *, struct nfsclflayouthead *, int *); + int, int, int, int *, struct nfsclflayouthead *, int *); static int nfsrpc_layoutget(struct nfsmount *, uint8_t *, int, int, uint64_t, - uint64_t, uint64_t, int, nfsv4stateid_t *, int *, struct nfsclflayouthead *, - struct ucred *, NFSPROC_T *, void *); + uint64_t, uint64_t, int, int, nfsv4stateid_t *, int *, + struct nfsclflayouthead *, struct ucred *, NFSPROC_T *, void *); static int nfsrpc_layoutgetres(struct nfsmount *, vnode_t, uint8_t *, int, nfsv4stateid_t *, int, uint32_t *, struct nfscllayout **, - struct nfsclflayouthead *, int, int *, struct ucred *, NFSPROC_T *); + struct nfsclflayouthead *, int, int, int *, struct ucred *, NFSPROC_T *); /* * nfs null call from vfs. @@ -4849,9 +4859,10 @@ nfsrpc_destroyclient(struct nfsmount *nmp, struct nfsclclient *clp, */ static int nfsrpc_layoutget(struct nfsmount *nmp, uint8_t *fhp, int fhlen, int iomode, - uint64_t offset, uint64_t len, uint64_t minlen, int layoutlen, - nfsv4stateid_t *stateidp, int *retonclosep, struct nfsclflayouthead *flhp, - struct ucred *cred, NFSPROC_T *p, void *stuff) + uint64_t offset, uint64_t len, uint64_t minlen, int layouttype, + int layoutlen, nfsv4stateid_t *stateidp, int *retonclosep, + struct nfsclflayouthead *flhp, struct ucred *cred, NFSPROC_T *p, + void *stuff) { struct nfsrv_descript nfsd, *nd = &nfsd; int error; @@ -4859,7 +4870,7 @@ nfsrpc_layoutget(struct nfsmount *nmp, uint8_t *fhp, int fhlen, int iomode, nfscl_reqstart(nd, NFSPROC_LAYOUTGET, nmp, fhp, fhlen, NULL, NULL, 0, 0); nfsrv_setuplayoutget(nd, iomode, offset, len, minlen, stateidp, - layoutlen, 0); + layouttype, layoutlen, 0); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); @@ -4882,7 +4893,7 @@ nfsrpc_getdeviceinfo(struct nfsmount *nmp, uint8_t *deviceid, int layouttype, uint32_t *notifybitsp, struct nfscldevinfo **ndip, struct ucred *cred, NFSPROC_T *p) { - uint32_t cnt, *tl; + uint32_t cnt, *tl, vers, minorvers; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; struct sockaddr_in sin, ssin; @@ -4915,51 +4926,68 @@ nfsrpc_getdeviceinfo(struct nfsmount *nmp, uint8_t *deviceid, int layouttype, if (error != 0) return (error); if (nd->nd_repstat == 0) { - NFSM_DISSECT(tl, uint32_t *, 3 * NFSX_UNSIGNED); - if (layouttype != fxdr_unsigned(int, *tl++)) + NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); + if (layouttype != fxdr_unsigned(int, *tl)) printf("EEK! devinfo layout type not same!\n"); - stripecnt = fxdr_unsigned(int, *++tl); - NFSCL_DEBUG(4, "stripecnt=%d\n", stripecnt); - if (stripecnt < 1 || stripecnt > 4096) { - printf("NFS devinfo stripecnt %d: out of range\n", - stripecnt); - error = NFSERR_BADXDR; - goto nfsmout; - } - NFSM_DISSECT(tl, uint32_t *, (stripecnt + 1) * NFSX_UNSIGNED); - addrcnt = fxdr_unsigned(int, *(tl + stripecnt)); - NFSCL_DEBUG(4, "addrcnt=%d\n", addrcnt); - if (addrcnt < 1 || addrcnt > 128) { - printf("NFS devinfo addrcnt %d: out of range\n", - addrcnt); - error = NFSERR_BADXDR; - goto nfsmout; - } - - /* - * Now we know how many stripe indices and addresses, so - * we can allocate the structure the correct size. - */ - i = (stripecnt * sizeof(uint8_t)) / sizeof(struct nfsclds *) - + 1; - NFSCL_DEBUG(4, "stripeindices=%d\n", i); - ndi = malloc(sizeof(*ndi) + (addrcnt + i) * - sizeof(struct nfsclds *), M_NFSDEVINFO, M_WAITOK | M_ZERO); - NFSBCOPY(deviceid, ndi->nfsdi_deviceid, NFSX_V4DEVICEID); - ndi->nfsdi_refcnt = 0; - ndi->nfsdi_stripecnt = stripecnt; - ndi->nfsdi_addrcnt = addrcnt; - /* Fill in the stripe indices. */ - for (i = 0; i < stripecnt; i++) { - stripeindex = fxdr_unsigned(uint8_t, *tl++); - NFSCL_DEBUG(4, "stripeind=%d\n", stripeindex); - if (stripeindex >= addrcnt) { - printf("NFS devinfo stripeindex %d: too big\n", - (int)stripeindex); + if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); + stripecnt = fxdr_unsigned(int, *tl); + NFSCL_DEBUG(4, "stripecnt=%d\n", stripecnt); + if (stripecnt < 1 || stripecnt > 4096) { + printf("pNFS File layout devinfo stripecnt %d:" + " out of range\n", stripecnt); + error = NFSERR_BADXDR; + goto nfsmout; + } + NFSM_DISSECT(tl, uint32_t *, (stripecnt + 1) * + NFSX_UNSIGNED); + addrcnt = fxdr_unsigned(int, *(tl + stripecnt)); + NFSCL_DEBUG(4, "addrcnt=%d\n", addrcnt); + if (addrcnt < 1 || addrcnt > 128) { + printf("NFS devinfo addrcnt %d: out of range\n", + addrcnt); error = NFSERR_BADXDR; goto nfsmout; } - nfsfldi_setstripeindex(ndi, i, stripeindex); + + /* + * Now we know how many stripe indices and addresses, so + * we can allocate the structure the correct size. + */ + i = (stripecnt * sizeof(uint8_t)) / + sizeof(struct nfsclds *) + 1; + NFSCL_DEBUG(4, "stripeindices=%d\n", i); + ndi = malloc(sizeof(*ndi) + (addrcnt + i) * + sizeof(struct nfsclds *), M_NFSDEVINFO, M_WAITOK | + M_ZERO); + NFSBCOPY(deviceid, ndi->nfsdi_deviceid, + NFSX_V4DEVICEID); + ndi->nfsdi_refcnt = 0; + ndi->nfsdi_flags = NFSDI_FILELAYOUT; + ndi->nfsdi_stripecnt = stripecnt; + ndi->nfsdi_addrcnt = addrcnt; + /* Fill in the stripe indices. */ + for (i = 0; i < stripecnt; i++) { + stripeindex = fxdr_unsigned(uint8_t, *tl++); + NFSCL_DEBUG(4, "stripeind=%d\n", stripeindex); + if (stripeindex >= addrcnt) { + printf("pNFS File Layout devinfo" + " stripeindex %d: too big\n", + (int)stripeindex); + error = NFSERR_BADXDR; + goto nfsmout; + } + nfsfldi_setstripeindex(ndi, i, stripeindex); + } + } else if (layouttype == NFSLAYOUT_FLEXFILE) { + /* For Flex File, we only get one address list. */ + ndi = malloc(sizeof(*ndi) + sizeof(struct nfsclds *), + M_NFSDEVINFO, M_WAITOK | M_ZERO); + NFSBCOPY(deviceid, ndi->nfsdi_deviceid, + NFSX_V4DEVICEID); + ndi->nfsdi_refcnt = 0; + ndi->nfsdi_flags = NFSDI_FLEXFILE; + addrcnt = ndi->nfsdi_addrcnt = 1; } /* Now, dissect the server address(es). */ @@ -5005,6 +5033,46 @@ nfsrpc_getdeviceinfo(struct nfsmount *nmp, uint8_t *deviceid, int layouttype, } gotvers = NFS_VER4; /* Always NFSv4 for File Layout. */ + /* For Flex File, we will take one of the versions to use. */ + if (layouttype == NFSLAYOUT_FLEXFILE) { + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); + j = fxdr_unsigned(int, *tl); + if (j < 1 || j > NFSDEV_MAXVERS) { + printf("pNFS: too many versions\n"); + error = NFSERR_BADXDR; + goto nfsmout; + } + gotvers = 0; + for (i = 0; i < j; i++) { + NFSM_DISSECT(tl, uint32_t *, 5 * NFSX_UNSIGNED); + vers = fxdr_unsigned(uint32_t, *tl++); + minorvers = fxdr_unsigned(uint32_t, *tl++); + if ((vers == NFS_VER4 && minorvers == + NFSV41_MINORVERSION) || (vers == NFS_VER3 && + gotvers == 0)) { + gotvers = vers; + /* We'll take this one. */ + ndi->nfsdi_versindex = i; + ndi->nfsdi_vers = vers; + ndi->nfsdi_minorvers = minorvers; + ndi->nfsdi_rsize = fxdr_unsigned( + uint32_t, *tl++); + ndi->nfsdi_wsize = fxdr_unsigned( + uint32_t, *tl++); + if (*tl == newnfs_true) + ndi->nfsdi_flags |= + NFSDI_TIGHTCOUPLED; + else + ndi->nfsdi_flags &= + ~NFSDI_TIGHTCOUPLED; + } + } + if (gotvers == 0) { + printf("pNFS: no NFSv3 or NFSv4.1\n"); + error = NFSERR_BADXDR; + goto nfsmout; + } + } /* And the notify bits. */ NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); @@ -5033,7 +5101,7 @@ nfsrpc_getdeviceinfo(struct nfsmount *nmp, uint8_t *deviceid, int layouttype, *gotdspp = dsp; } } - if (nd->nd_repstat != 0) + if (nd->nd_repstat != 0 && error == 0) error = nd->nd_repstat; nfsmout: if (error != 0 && ndi != NULL) @@ -5126,7 +5194,15 @@ nfsrpc_layoutreturn(struct nfsmount *nmp, uint8_t *fh, int fhlen, int reclaim, *tl++ = stateidp->other[0]; *tl++ = stateidp->other[1]; *tl++ = stateidp->other[2]; - *tl = txdr_unsigned(0); + if (layouttype == NFSLAYOUT_NFSV4_1_FILES) + *tl = txdr_unsigned(0); + else if (layouttype == NFSLAYOUT_FLEXFILE) { + *tl = txdr_unsigned(2 * NFSX_UNSIGNED); + NFSM_BUILD(tl, uint32_t *, 2 * NFSX_UNSIGNED); + /* No ioerrs or stats yet. */ + *tl++ = 0; + *tl = 0; + } } nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, @@ -5162,11 +5238,15 @@ nfsrpc_getlayout(struct nfsmount *nmp, vnode_t vp, struct nfsfh *nfhp, struct nfscllayout *lyp; struct nfsclflayout *flp; struct nfsclflayouthead flh; - int error = 0, islocked, layoutlen, recalled, retonclose; + int error = 0, islocked, layoutlen, layouttype, recalled, retonclose; nfsv4stateid_t stateid; struct nfsclsession *tsep; *lypp = NULL; + if (NFSHASFLEXFILE(nmp)) + layouttype = NFSLAYOUT_FLEXFILE; + else + layouttype = NFSLAYOUT_NFSV4_1_FILES; /* * If lyp is returned non-NULL, there will be a refcnt (shared lock) * on it, iff flp != NULL or a lock (exclusive lock) on it iff @@ -5189,8 +5269,8 @@ nfsrpc_getlayout(struct nfsmount *nmp, vnode_t vp, struct nfsfh *nfhp, stateid.other[2] = stateidp->other[2]; error = nfsrpc_layoutget(nmp, nfhp->nfh_fh, nfhp->nfh_len, iomode, (uint64_t)0, UINT64_MAX, - (uint64_t)0, layoutlen, &stateid, &retonclose, - &flh, cred, p, NULL); + (uint64_t)0, layouttype, layoutlen, &stateid, + &retonclose, &flh, cred, p, NULL); } else { islocked = 1; stateid.seqid = lyp->nfsly_stateid.seqid; @@ -5199,12 +5279,12 @@ nfsrpc_getlayout(struct nfsmount *nmp, vnode_t vp, struct nfsfh *nfhp, stateid.other[2] = lyp->nfsly_stateid.other[2]; error = nfsrpc_layoutget(nmp, nfhp->nfh_fh, nfhp->nfh_len, iomode, off, UINT64_MAX, - (uint64_t)0, layoutlen, &stateid, &retonclose, - &flh, cred, p, NULL); + (uint64_t)0, layouttype, layoutlen, &stateid, + &retonclose, &flh, cred, p, NULL); } error = nfsrpc_layoutgetres(nmp, vp, nfhp->nfh_fh, nfhp->nfh_len, &stateid, retonclose, notifybitsp, &lyp, - &flh, error, NULL, cred, p); + &flh, layouttype, error, NULL, cred, p); if (error == 0) *lypp = lyp; else if (islocked != 0) @@ -5336,38 +5416,53 @@ nfsrpc_fillsa(struct nfsmount *nmp, struct sockaddr_in *sin, error = newnfs_connect(nmp, nrp, NULL, p, 0); NFSCL_DEBUG(3, "DS connect=%d\n", error); + dsp = NULL; /* Now, do the exchangeid and create session. */ if (error == 0) { - error = nfsrpc_exchangeid(nmp, clp, nrp, NFSV4EXCH_USEPNFSDS, - &dsp, nrp->nr_cred, p); - NFSCL_DEBUG(3, "DS exchangeid=%d\n", error); - if (error != 0) - newnfs_disconnect(nrp); + if (vers == NFS_VER4) { + error = nfsrpc_exchangeid(nmp, clp, nrp, + NFSV4EXCH_USEPNFSDS, &dsp, nrp->nr_cred, p); + NFSCL_DEBUG(3, "DS exchangeid=%d\n", error); + if (error != 0) + newnfs_disconnect(nrp); + } else { + dsp = malloc(sizeof(struct nfsclds), M_NFSCLDS, + M_WAITOK | M_ZERO); + dsp->nfsclds_flags |= NFSCLDS_DS; + dsp->nfsclds_expire = INT32_MAX; /* No renews needed. */ + mtx_init(&dsp->nfsclds_mtx, "nfsds", NULL, MTX_DEF); + mtx_init(&dsp->nfsclds_sess.nfsess_mtx, "nfssession", + NULL, MTX_DEF); + } } if (error == 0) { dsp->nfsclds_sockp = nrp; - NFSLOCKMNT(nmp); - retv = nfscl_getsameserver(nmp, dsp, &tdsp); - NFSCL_DEBUG(3, "getsame ret=%d\n", retv); - if (retv == NFSDSP_USETHISSESSION) { + if (vers == NFS_VER4) { + NFSLOCKMNT(nmp); + retv = nfscl_getsameserver(nmp, dsp, &tdsp); + NFSCL_DEBUG(3, "getsame ret=%d\n", retv); + if (retv == NFSDSP_USETHISSESSION) { + NFSUNLOCKMNT(nmp); + /* + * If there is already a session for this + * server, use it. + */ + (void)newnfs_disconnect(nrp); + nfscl_freenfsclds(dsp); + *dspp = tdsp; + return (0); + } + if (retv == NFSDSP_SEQTHISSESSION) + sequenceid = + tdsp->nfsclds_sess.nfsess_sequenceid; + else + sequenceid = + dsp->nfsclds_sess.nfsess_sequenceid; NFSUNLOCKMNT(nmp); - /* - * If there is already a session for this server, - * use it. - */ - (void)newnfs_disconnect(nrp); - nfscl_freenfsclds(dsp); - *dspp = tdsp; - return (0); + error = nfsrpc_createsession(nmp, &dsp->nfsclds_sess, + nrp, sequenceid, 0, nrp->nr_cred, p); + NFSCL_DEBUG(3, "DS createsess=%d\n", error); } - if (retv == NFSDSP_SEQTHISSESSION) - sequenceid = tdsp->nfsclds_sess.nfsess_sequenceid; - else - sequenceid = dsp->nfsclds_sess.nfsess_sequenceid; - NFSUNLOCKMNT(nmp); - error = nfsrpc_createsession(nmp, &dsp->nfsclds_sess, - nrp, sequenceid, 0, nrp->nr_cred, p); - NFSCL_DEBUG(3, "DS createsess=%d\n", error); } else { NFSFREECRED(nrp->nr_cred); NFSFREEMUTEX(&nrp->nr_mtx); @@ -5448,11 +5543,17 @@ nfscl_doiods(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, struct nfscllayout *layp; struct nfscldevinfo *dip; struct nfsclflayout *rflp; + struct mbuf *m; nfsv4stateid_t stateid; struct ucred *newcred; uint64_t lastbyte, len, off, oresid, xfer; - int eof, error, iolaymode, recalled; + int eof, error, firstmirror, i, iolaymode, mirrorcnt, recalled; void *lckp; + uint8_t *dev; + void *iovbase; + size_t iovlen; + off_t offs; + ssize_t resid; if (!NFSHASPNFS(nmp) || nfscl_enablecallb == 0 || nfs_numnfscbd == 0 || (np->n_flag & NNOLAYOUT) != 0) @@ -5520,30 +5621,93 @@ nfscl_doiods(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, oresid = xfer = (uint64_t)uiop->uio_resid; if (xfer > (rflp->nfsfl_end - rflp->nfsfl_off)) xfer = rflp->nfsfl_end - rflp->nfsfl_off; - dip = nfscl_getdevinfo(nmp->nm_clp, rflp->nfsfl_dev, - rflp->nfsfl_devp); - if (dip != NULL) { - error = nfscl_doflayoutio(vp, uiop, iomode, - must_commit, &eof, &stateid, rwaccess, dip, - layp, rflp, off, xfer, docommit, newcred, - p); - nfscl_reldevinfo(dip); - lastbyte = off + xfer - 1; - if (error == 0) { + /* + * For Flex File layout with mirrored DSs, select one + * of them at random for reads. For writes and commits, + * do all mirrors. + */ + m = NULL; + firstmirror = 0; + mirrorcnt = 1; + if ((layp->nfsly_flags & NFSLY_FLEXFILE) != 0 && + (mirrorcnt = rflp->nfsfl_mirrorcnt) > 1) { + if (rwaccess == NFSV4OPEN_ACCESSREAD) { + firstmirror = arc4random() % mirrorcnt; + mirrorcnt = firstmirror + 1; + } else if (docommit == 0) { + /* + * Save values, so uiop can be rolled + * back upon a write error. + */ + offs = uiop->uio_offset; + resid = uiop->uio_resid; + iovbase = uiop->uio_iov->iov_base; + iovlen = uiop->uio_iov->iov_len; + m = nfsm_uiombuflist(uiop, len, NULL, + NULL); + } + } + for (i = firstmirror; i < mirrorcnt && error == 0; i++){ + if ((layp->nfsly_flags & NFSLY_FLEXFILE) != 0) + dev = rflp->nfsfl_ffm[i].dev; + else + dev = rflp->nfsfl_dev; + dip = nfscl_getdevinfo(nmp->nm_clp, dev, + rflp->nfsfl_devp); + if (dip != NULL) { + if ((rflp->nfsfl_flags & NFSFL_FLEXFILE) + != 0) + error = nfscl_dofflayoutio(vp, + uiop, iomode, must_commit, + &eof, &stateid, rwaccess, + dip, layp, rflp, off, xfer, + i, docommit, m, newcred, + p); + else + error = nfscl_doflayoutio(vp, + uiop, iomode, must_commit, + &eof, &stateid, rwaccess, + dip, layp, rflp, off, xfer, + docommit, newcred, p); + nfscl_reldevinfo(dip); + } else + error = EIO; + } + if (m != NULL) + m_freem(m); + if (error == 0) { + if (mirrorcnt > 1 && rwaccess == + NFSV4OPEN_ACCESSWRITE && docommit == 0) { NFSLOCKCLSTATE(); - if (lastbyte > layp->nfsly_lastbyte) - layp->nfsly_lastbyte = lastbyte; + layp->nfsly_flags |= NFSLY_WRITTEN; NFSUNLOCKCLSTATE(); - } else if (error == NFSERR_OPENMODE && - rwaccess == NFSV4OPEN_ACCESSREAD) { - NFSLOCKMNT(nmp); - nmp->nm_state |= NFSSTA_OPENMODE; - NFSUNLOCKMNT(nmp); } + lastbyte = off + xfer - 1; + NFSLOCKCLSTATE(); + if (lastbyte > layp->nfsly_lastbyte) + layp->nfsly_lastbyte = lastbyte; + NFSUNLOCKCLSTATE(); + } else if (error == NFSERR_OPENMODE && + rwaccess == NFSV4OPEN_ACCESSREAD) { + NFSLOCKMNT(nmp); + nmp->nm_state |= NFSSTA_OPENMODE; + NFSUNLOCKMNT(nmp); } else error = EIO; if (error == 0) len -= (oresid - (uint64_t)uiop->uio_resid); + else if (mirrorcnt > 1 && rwaccess == + NFSV4OPEN_ACCESSWRITE && docommit == 0) { + /* + * In case the rpc gets retried, roll the + * uio fields changed by nfsm_uiombuflist() + * back. + */ + uiop->uio_offset = offs; + uiop->uio_resid = resid; + uiop->uio_iov->iov_base = iovbase; + uiop->uio_iov->iov_len = iovlen; + } } } if (lckp != NULL) @@ -5555,6 +5719,38 @@ nfscl_doiods(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, } /* + * Make a copy of the mbuf chain and add an mbuf for null padding, as required. + */ +static struct mbuf * +nfsm_copym(struct mbuf *m, int off, int xfer) +{ + struct mbuf *m2, *m3, *m4; + uint32_t *tl; + int rem; + + m2 = m_copym(m, off, xfer, M_WAITOK); + rem = NFSM_RNDUP(xfer) - xfer; + if (rem > 0) { + /* + * The zero padding to a multiple of 4 bytes is required by + * the XDR. So that the mbufs copied by reference aren't + * modified, add an mbuf with the zero'd bytes to the list. + * rem will be a maximum of 3, so one zero'd uint32_t is + * sufficient. + */ + m3 = m2; + while (m3->m_next != NULL) + m3 = m3->m_next; + NFSMGET(m4); + tl = NFSMTOD(m4, uint32_t *); + *tl = 0; + mbuf_setlen(m4, rem); + mbuf_setnext(m3, m4); + } + return (m2); +} + +/* * Find a file layout that will handle the first bytes of the requested * range and return the information from it needed to to the I/O operation. */ @@ -5659,7 +5855,7 @@ nfscl_doflayoutio(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, if (docommit != 0) { if (error == 0) error = nfsrpc_commitds(vp, io_off, xfer, - *dspp, fhp, cred, p); + *dspp, fhp, 0, 0, cred, p); if (error == 0) { /* * Set both eof and uio_resid = 0 to end any @@ -5674,11 +5870,11 @@ nfscl_doflayoutio(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, } } else if (rwflag == NFSV4OPEN_ACCESSREAD) error = nfsrpc_readds(vp, uiop, stateidp, eofp, *dspp, - io_off, xfer, fhp, cred, p); + io_off, xfer, fhp, 0, 0, 0, cred, p); else { error = nfsrpc_writeds(vp, uiop, iomode, must_commit, stateidp, *dspp, io_off, xfer, fhp, commit_thru_mds, - cred, p); + 0, 0, 0, cred, p); if (error == 0) { NFSLOCKCLSTATE(); lyp->nfsly_flags |= NFSLY_WRITTEN; @@ -5696,42 +5892,174 @@ nfscl_doflayoutio(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, } /* + * Do I/O using an NFSv4.1 flex file layout. + */ +static int +nfscl_dofflayoutio(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, + int *eofp, nfsv4stateid_t *stateidp, int rwflag, struct nfscldevinfo *dp, + struct nfscllayout *lyp, struct nfsclflayout *flp, uint64_t off, + uint64_t len, int mirror, int docommit, struct mbuf *mp, struct ucred *cred, + NFSPROC_T *p) +{ + uint64_t transfer, xfer; + int error, rel_off; + struct nfsnode *np; + struct nfsfh *fhp; + struct nfsclds **dspp; + struct ucred *tcred; + struct mbuf *m; + + np = VTONFS(vp); + error = 0; + rel_off = 0; + NFSCL_DEBUG(4, "nfscl_dofflayoutio: off=%ju len=%ju\n", (uintmax_t)off, + (uintmax_t)len); + /* Loop around, doing I/O for each stripe unit. */ + while (len > 0 && error == 0) { + dspp = nfsfldi_addr(dp, 0); + fhp = flp->nfsfl_ffm[mirror].fh[dp->nfsdi_versindex]; + stateidp = &flp->nfsfl_ffm[mirror].st; + NFSCL_DEBUG(4, "mirror=%d vind=%d fhlen=%d st.seqid=0x%x\n", + mirror, dp->nfsdi_versindex, fhp->nfh_len, stateidp->seqid); + if ((dp->nfsdi_flags & NFSDI_TIGHTCOUPLED) == 0) { + tcred = NFSNEWCRED(cred); + tcred->cr_uid = flp->nfsfl_ffm[mirror].user; + tcred->cr_groups[0] = flp->nfsfl_ffm[mirror].group; + tcred->cr_ngroups = 1; + } else + tcred = cred; + if (rwflag == NFSV4OPEN_ACCESSREAD) + transfer = dp->nfsdi_rsize; + else + transfer = dp->nfsdi_wsize; + mtx_lock(&np->n_mtx); + np->n_flag |= NDSCOMMIT; + mtx_unlock(&np->n_mtx); + if (len > transfer && docommit == 0) + xfer = transfer; + else + xfer = len; + if (docommit != 0) { + if (error == 0) + error = nfsrpc_commitds(vp, off, xfer, *dspp, + fhp, dp->nfsdi_vers, dp->nfsdi_minorvers, + tcred, p); + NFSCL_DEBUG(4, "aft nfsrpc_commitds=%d\n", error); + if (error == 0) { + /* + * Set both eof and uio_resid = 0 to end any + * loops. + */ + *eofp = 1; + uiop->uio_resid = 0; + } else { + mtx_lock(&np->n_mtx); + np->n_flag &= ~NDSCOMMIT; + mtx_unlock(&np->n_mtx); + } + } else if (rwflag == NFSV4OPEN_ACCESSREAD) + error = nfsrpc_readds(vp, uiop, stateidp, eofp, *dspp, + off, xfer, fhp, 1, dp->nfsdi_vers, + dp->nfsdi_minorvers, tcred, p); + else { + if (flp->nfsfl_mirrorcnt == 1) { + error = nfsrpc_writeds(vp, uiop, iomode, + must_commit, stateidp, *dspp, off, xfer, + fhp, 0, 1, dp->nfsdi_vers, + dp->nfsdi_minorvers, tcred, p); + if (error == 0) { + NFSLOCKCLSTATE(); + lyp->nfsly_flags |= NFSLY_WRITTEN; + NFSUNLOCKCLSTATE(); + } + } else { + m = nfsm_copym(mp, rel_off, xfer); + NFSCL_DEBUG(4, "mcopy reloff=%d xfer=%jd\n", + rel_off, (uintmax_t)xfer); + error = nfsrpc_writedsmir(vp, iomode, + must_commit, stateidp, *dspp, off, xfer, + fhp, m, dp->nfsdi_vers, dp->nfsdi_minorvers, + tcred, p); + NFSCL_DEBUG(4, "nfsrpc_writedsmir=%d\n", error); + } + } + NFSCL_DEBUG(4, "aft read/writeds=%d\n", error); + if (error == 0) { + len -= xfer; + off += xfer; + rel_off += xfer; + } + if ((dp->nfsdi_flags & NFSDI_TIGHTCOUPLED) == 0) + NFSFREECRED(tcred); + } + NFSCL_DEBUG(4, "eo nfscl_dofflayoutio=%d\n", error); + return (error); +} + +/* * The actual read RPC done to a DS. */ static int nfsrpc_readds(vnode_t vp, struct uio *uiop, nfsv4stateid_t *stateidp, int *eofp, - struct nfsclds *dsp, uint64_t io_off, int len, struct nfsfh *fhp, - struct ucred *cred, NFSPROC_T *p) + struct nfsclds *dsp, uint64_t io_off, int len, struct nfsfh *fhp, int flex, + int vers, int minorvers, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; - int error, retlen; + int attrflag, error, retlen; struct nfsrv_descript nfsd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsrv_descript *nd = &nfsd; struct nfssockreq *nrp; + struct nfsvattr na; nd->nd_mrep = NULL; - nfscl_reqstart(nd, NFSPROC_READDS, nmp, fhp->nfh_fh, fhp->nfh_len, - NULL, &dsp->nfsclds_sess, 0, 0); - nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSEQIDZERO); + if (vers == 0 || vers == NFS_VER4) { + nfscl_reqstart(nd, NFSPROC_READDS, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + vers = NFS_VER4; + NFSCL_DEBUG(4, "nfsrpc_readds: vers4 minvers=%d\n", minorvers); + if (flex != 0) + nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); + else + nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSEQIDZERO); + } else { + nfscl_reqstart(nd, NFSPROC_READ, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + NFSCL_DEBUG(4, "nfsrpc_readds: vers3\n"); + } NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); txdr_hyper(io_off, tl); *(tl + 2) = txdr_unsigned(len); nrp = dsp->nfsclds_sockp; + NFSCL_DEBUG(4, "nfsrpc_readds: nrp=%p\n", nrp); if (nrp == NULL) /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, - NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); + NFS_PROG, vers, NULL, 1, NULL, &dsp->nfsclds_sess); + NFSCL_DEBUG(4, "nfsrpc_readds: stat=%d err=%d\n", nd->nd_repstat, + error); if (error != 0) return (error); + if (vers == NFS_VER3) { + error = nfscl_postop_attr(nd, &na, &attrflag, NULL); + NFSCL_DEBUG(4, "nfsrpc_readds: postop=%d\n", error); + if (error != 0) + goto nfsmout; + } if (nd->nd_repstat != 0) { error = nd->nd_repstat; goto nfsmout; } - NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); - *eofp = fxdr_unsigned(int, *tl); + if (vers == NFS_VER3) { + NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); + *eofp = fxdr_unsigned(int, *(tl + 1)); + } else { + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); + *eofp = fxdr_unsigned(int, *tl); + } NFSM_STRSIZ(retlen, len); + NFSCL_DEBUG(4, "nfsrpc_readds: retlen=%d eof=%d\n", retlen, *eofp); error = nfsm_mbufuio(nd, uiop, retlen); nfsmout: if (nd->nd_mrep != NULL) @@ -5745,24 +6073,40 @@ nfsmout: static int nfsrpc_writeds(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, nfsv4stateid_t *stateidp, struct nfsclds *dsp, uint64_t io_off, int len, - struct nfsfh *fhp, int commit_thru_mds, struct ucred *cred, NFSPROC_T *p) + struct nfsfh *fhp, int commit_thru_mds, int flex, int vers, int minorvers, + struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); - int error, rlen, commit, committed = NFSWRITE_FILESYNC; + int attrflag, error, rlen, commit, committed = NFSWRITE_FILESYNC; int32_t backup; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; struct nfssockreq *nrp; + struct nfsvattr na; KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1")); nd->nd_mrep = NULL; - nfscl_reqstart(nd, NFSPROC_WRITEDS, nmp, fhp->nfh_fh, fhp->nfh_len, - NULL, &dsp->nfsclds_sess, 0, 0); - nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSEQIDZERO); - NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); + if (vers == 0 || vers == NFS_VER4) { + nfscl_reqstart(nd, NFSPROC_WRITEDS, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + NFSCL_DEBUG(4, "nfsrpc_writeds: vers4 minvers=%d\n", minorvers); + vers = NFS_VER4; + if (flex != 0) + nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); + else + nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSEQIDZERO); + NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); + } else { + nfscl_reqstart(nd, NFSPROC_WRITE, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + NFSCL_DEBUG(4, "nfsrpc_writeds: vers3\n"); + NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + 3 * NFSX_UNSIGNED); + } txdr_hyper(io_off, tl); tl += 2; + if (vers == NFS_VER3) + *tl++ = txdr_unsigned(len); *tl++ = txdr_unsigned(*iomode); *tl = txdr_unsigned(len); nfsm_uiombuf(nd, uiop, len); @@ -5771,7 +6115,9 @@ nfsrpc_writeds(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, - NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); + NFS_PROG, vers, NULL, 1, NULL, &dsp->nfsclds_sess); + NFSCL_DEBUG(4, "nfsrpc_writeds: err=%d stat=%d\n", error, + nd->nd_repstat); if (error != 0) return (error); if (nd->nd_repstat != 0) { @@ -5786,8 +6132,16 @@ nfsrpc_writeds(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, uio_iov_len_add(uiop, len); error = nd->nd_repstat; } else { + if (vers == NFS_VER3) { + error = nfscl_wcc_data(nd, vp, &na, &attrflag, NULL, + NULL); + NFSCL_DEBUG(4, "nfsrpc_writeds: wcc_data=%d\n", error); + if (error != 0) + goto nfsmout; + } NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); rlen = fxdr_unsigned(int, *tl++); + NFSCL_DEBUG(4, "nfsrpc_writeds: len=%d rlen=%d\n", len, rlen); if (rlen == 0) { error = NFSERR_IO; goto nfsmout; @@ -5842,6 +6196,117 @@ nfsmout: } /* + * The actual write RPC done to a DS. + * This variant is called from a separate kernel process for mirrors. + * Any short write is considered an IO error. + */ +static int +nfsrpc_writedsmir(vnode_t vp, int *iomode, int *must_commit, + nfsv4stateid_t *stateidp, struct nfsclds *dsp, uint64_t io_off, int len, + struct nfsfh *fhp, struct mbuf *m, int vers, int minorvers, + struct ucred *cred, NFSPROC_T *p) +{ + uint32_t *tl; + struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); + int attrflag, error, commit, committed = NFSWRITE_FILESYNC, rlen; + struct nfsrv_descript nfsd; + struct nfsrv_descript *nd = &nfsd; + struct nfssockreq *nrp; + struct nfsvattr na; + + nd->nd_mrep = NULL; + if (vers == 0 || vers == NFS_VER4) { + nfscl_reqstart(nd, NFSPROC_WRITEDS, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + vers = NFS_VER4; + NFSCL_DEBUG(4, "nfsrpc_writedsmir: vers4 minvers=%d\n", + minorvers); + nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); + NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); + } else { + nfscl_reqstart(nd, NFSPROC_WRITE, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + NFSCL_DEBUG(4, "nfsrpc_writedsmir: vers3\n"); + NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + 3 * NFSX_UNSIGNED); + } + txdr_hyper(io_off, tl); + tl += 2; + if (vers == NFS_VER3) + *tl++ = txdr_unsigned(len); + *tl++ = txdr_unsigned(*iomode); + *tl = txdr_unsigned(len); + if (len > 0) { + /* Put data in mbuf chain. */ + nd->nd_mb->m_next = m; + /* Set nd_mb and nd_bpos to end of data. */ + while (m->m_next != NULL) + m = m->m_next; + nd->nd_mb = m; + nd->nd_bpos = mtod(m, char *) + m->m_len; + NFSCL_DEBUG(4, "nfsrpc_writedsmir: lastmb len=%d\n", m->m_len); + } + nrp = dsp->nfsclds_sockp; + if (nrp == NULL) + /* If NULL, use the MDS socket. */ + nrp = &nmp->nm_sockreq; + error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, + NFS_PROG, vers, NULL, 1, NULL, &dsp->nfsclds_sess); + NFSCL_DEBUG(4, "nfsrpc_writedsmir: err=%d stat=%d\n", error, + nd->nd_repstat); + if (error != 0) + return (error); + if (nd->nd_repstat != 0) + error = nd->nd_repstat; + else { + if (vers == NFS_VER3) { + error = nfscl_wcc_data(nd, vp, &na, &attrflag, NULL, + NULL); + NFSCL_DEBUG(4, "nfsrpc_writedsmir: wcc_data=%d\n", + error); + if (error != 0) + goto nfsmout; + } + NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); + rlen = fxdr_unsigned(int, *tl++); + NFSCL_DEBUG(4, "nfsrpc_writedsmir: len=%d rlen=%d\n", len, + rlen); + if (rlen != len) { + error = NFSERR_IO; + NFSCL_DEBUG(4, "nfsrpc_writedsmir: len=%d rlen=%d\n", + len, rlen); + goto nfsmout; + } + commit = fxdr_unsigned(int, *tl++); + + /* + * Return the lowest commitment level + * obtained by any of the RPCs. + */ + if (committed == NFSWRITE_FILESYNC) + committed = commit; + else if (committed == NFSWRITE_DATASYNC && + commit == NFSWRITE_UNSTABLE) + committed = commit; + NFSLOCKDS(dsp); + if ((dsp->nfsclds_flags & NFSCLDS_HASWRITEVERF) == 0) { + NFSBCOPY(tl, dsp->nfsclds_verf, NFSX_VERF); + dsp->nfsclds_flags |= NFSCLDS_HASWRITEVERF; + } else if (NFSBCMP(tl, dsp->nfsclds_verf, NFSX_VERF)) { + *must_commit = 1; + NFSBCOPY(tl, dsp->nfsclds_verf, NFSX_VERF); + } + NFSUNLOCKDS(dsp); + } +nfsmout: + if (nd->nd_mrep != NULL) + mbuf_freem(nd->nd_mrep); + *iomode = committed; + if (nd->nd_repstat != 0 && error == 0) + error = nd->nd_repstat; + return (error); +} + +/* * Free up the nfsclds structure. */ void @@ -5913,17 +6378,26 @@ nfscl_getsameserver(struct nfsmount *nmp, struct nfsclds *newdsp, */ static int nfsrpc_commitds(vnode_t vp, uint64_t offset, int cnt, struct nfsclds *dsp, - struct nfsfh *fhp, struct ucred *cred, NFSPROC_T *p) + struct nfsfh *fhp, int vers, int minorvers, struct ucred *cred, + NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfssockreq *nrp; - int error; + struct nfsvattr na; + int attrflag, error; nd->nd_mrep = NULL; - nfscl_reqstart(nd, NFSPROC_COMMITDS, nmp, fhp->nfh_fh, fhp->nfh_len, - NULL, &dsp->nfsclds_sess, 0, 0); + if (vers == 0 || vers == NFS_VER4) { + nfscl_reqstart(nd, NFSPROC_COMMITDS, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + vers = NFS_VER4; + } else + nfscl_reqstart(nd, NFSPROC_COMMIT, nmp, fhp->nfh_fh, + fhp->nfh_len, NULL, &dsp->nfsclds_sess, vers, minorvers); + NFSCL_DEBUG(4, "nfsrpc_commitds: vers=%d minvers=%d\n", vers, + minorvers); NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(offset, tl); tl += 2; @@ -5933,10 +6407,19 @@ nfsrpc_commitds(vnode_t vp, uint64_t offset, int cnt, struct nfsclds *dsp, /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, - NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); + NFS_PROG, vers, NULL, 1, NULL, &dsp->nfsclds_sess); + NFSCL_DEBUG(4, "nfsrpc_commitds: err=%d stat=%d\n", error, + nd->nd_repstat); if (error != 0) return (error); if (nd->nd_repstat == 0) { + if (vers == NFS_VER3) { + error = nfscl_wcc_data(nd, vp, &na, &attrflag, NULL, + NULL); + NFSCL_DEBUG(4, "nfsrpc_commitds: wccdata=%d\n", error); + if (error != 0) + goto nfsmout; + } NFSM_DISSECT(tl, u_int32_t *, NFSX_VERF); NFSLOCKDS(dsp); if (NFSBCMP(tl, dsp->nfsclds_verf, NFSX_VERF)) { @@ -5957,15 +6440,15 @@ nfsmout: */ static void nfsrv_setuplayoutget(struct nfsrv_descript *nd, int iomode, uint64_t offset, - uint64_t len, uint64_t minlen, nfsv4stateid_t *stateidp, int layoutlen, - int usecurstateid) + uint64_t len, uint64_t minlen, nfsv4stateid_t *stateidp, int layouttype, + int layoutlen, int usecurstateid) { uint32_t *tl; NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED + 3 * NFSX_HYPER + NFSX_STATEID); *tl++ = newnfs_false; /* Don't signal availability. */ - *tl++ = txdr_unsigned(NFSLAYOUT_NFSV4_1_FILES); + *tl++ = txdr_unsigned(layouttype); *tl++ = txdr_unsigned(iomode); txdr_hyper(offset, tl); tl += 2; @@ -5998,11 +6481,15 @@ nfsrv_parselayoutget(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, { uint32_t *tl; struct nfsclflayout *flp, *prevflp, *tflp; - int cnt, error, gotiomode, fhcnt, nfhlen, i, j; - uint64_t retlen; + int cnt, error, fhcnt, gotiomode, i, iomode, j, k, l, laytype, nfhlen; + int m, mirrorcnt; + uint64_t retlen, off; struct nfsfh *nfhp; uint8_t *cp; + uid_t user; + gid_t grp; + NFSCL_DEBUG(4, "in nfsrv_parselayoutget\n"); error = 0; flp = NULL; gotiomode = -1; @@ -6025,64 +6512,204 @@ nfsrv_parselayoutget(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, goto nfsmout; } for (i = 0; i < cnt; i++) { - /* Dissect all the way to the file handle cnt. */ - NFSM_DISSECT(tl, uint32_t *, 3 * NFSX_HYPER + - 6 * NFSX_UNSIGNED + NFSX_V4DEVICEID); - fhcnt = fxdr_unsigned(int, *(tl + 11 + - NFSX_V4DEVICEID / NFSX_UNSIGNED)); - NFSCL_DEBUG(4, "fhcnt=%d\n", fhcnt); - if (fhcnt < 0 || fhcnt > 100) { - /* Don't accept more than 100 file handles. */ - error = NFSERR_BADXDR; - goto nfsmout; - } - if (fhcnt > 1) - flp = malloc(sizeof(*flp) + (fhcnt - 1) * - sizeof(struct nfsfh *), M_NFSFLAYOUT, M_WAITOK); - else - flp = malloc(sizeof(*flp), M_NFSFLAYOUT, M_WAITOK); - flp->nfsfl_flags = 0; - flp->nfsfl_fhcnt = 0; - flp->nfsfl_devp = NULL; - flp->nfsfl_off = fxdr_hyper(tl); tl += 2; + /* Dissect to the layout type. */ + NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER + + 3 * NFSX_UNSIGNED); + off = fxdr_hyper(tl); tl += 2; retlen = fxdr_hyper(tl); tl += 2; - if (flp->nfsfl_off + retlen < flp->nfsfl_off) - flp->nfsfl_end = UINT64_MAX - flp->nfsfl_off; - else - flp->nfsfl_end = flp->nfsfl_off + retlen; - flp->nfsfl_iomode = fxdr_unsigned(int, *tl++); - if (gotiomode == -1) - gotiomode = flp->nfsfl_iomode; - if (fxdr_unsigned(int, *tl++) != NFSLAYOUT_NFSV4_1_FILES) { - printf("NFSv4.1: got non-files layout\n"); - error = NFSERR_BADXDR; - goto nfsmout; - } - NFSBCOPY(++tl, flp->nfsfl_dev, NFSX_V4DEVICEID); - tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); - flp->nfsfl_util = fxdr_unsigned(uint32_t, *tl++); - NFSCL_DEBUG(4, "flutil=0x%x\n", flp->nfsfl_util); - flp->nfsfl_stripe1 = fxdr_unsigned(uint32_t, *tl++); - flp->nfsfl_patoff = fxdr_hyper(tl); tl += 2; - if (fxdr_unsigned(int, *tl) != fhcnt) { - printf("EEK! bad fhcnt\n"); - error = NFSERR_BADXDR; - goto nfsmout; - } - for (j = 0; j < fhcnt; j++) { - NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); - nfhlen = fxdr_unsigned(int, *tl); - if (nfhlen <= 0 || nfhlen > NFSX_V4FHMAX) { + iomode = fxdr_unsigned(int, *tl++); + laytype = fxdr_unsigned(int, *tl); + NFSCL_DEBUG(4, "layt=%d off=%ju len=%ju iom=%d\n", laytype, + (uintmax_t)off, (uintmax_t)retlen, iomode); + /* Ignore length of layout body for now. */ + if (laytype == NFSLAYOUT_NFSV4_1_FILES) { + /* Parse the File layout up to fhcnt. */ + NFSM_DISSECT(tl, uint32_t *, 3 * NFSX_UNSIGNED + + NFSX_HYPER + NFSX_V4DEVICEID); + fhcnt = fxdr_unsigned(int, *(tl + 4 + + NFSX_V4DEVICEID / NFSX_UNSIGNED)); + NFSCL_DEBUG(4, "fhcnt=%d\n", fhcnt); + if (fhcnt < 0 || fhcnt > 100) { + /* Don't accept more than 100 file handles. */ + error = NFSERR_BADXDR; + goto nfsmout; + } + if (fhcnt > 0) + flp = malloc(sizeof(*flp) + fhcnt * + sizeof(struct nfsfh *), M_NFSFLAYOUT, + M_WAITOK); + else + flp = malloc(sizeof(*flp), M_NFSFLAYOUT, + M_WAITOK); + flp->nfsfl_flags = NFSFL_FILE; + flp->nfsfl_fhcnt = 0; + flp->nfsfl_devp = NULL; + flp->nfsfl_off = off; + if (flp->nfsfl_off + retlen < flp->nfsfl_off) + flp->nfsfl_end = UINT64_MAX - flp->nfsfl_off; + else + flp->nfsfl_end = flp->nfsfl_off + retlen; + flp->nfsfl_iomode = iomode; + if (gotiomode == -1) + gotiomode = flp->nfsfl_iomode; + /* Ignore layout body length for now. */ + NFSBCOPY(tl, flp->nfsfl_dev, NFSX_V4DEVICEID); + tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); + flp->nfsfl_util = fxdr_unsigned(uint32_t, *tl++); + NFSCL_DEBUG(4, "flutil=0x%x\n", flp->nfsfl_util); + flp->nfsfl_stripe1 = fxdr_unsigned(uint32_t, *tl++); + flp->nfsfl_patoff = fxdr_hyper(tl); tl += 2; + NFSCL_DEBUG(4, "stripe1=%u poff=%ju\n", + flp->nfsfl_stripe1, (uintmax_t)flp->nfsfl_patoff); + for (j = 0; j < fhcnt; j++) { + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); + nfhlen = fxdr_unsigned(int, *tl); + if (nfhlen <= 0 || nfhlen > NFSX_V4FHMAX) { + error = NFSERR_BADXDR; + goto nfsmout; + } + nfhp = malloc(sizeof(*nfhp) + nfhlen - 1, + M_NFSFH, M_WAITOK); + flp->nfsfl_fh[j] = nfhp; + flp->nfsfl_fhcnt++; + nfhp->nfh_len = nfhlen; + NFSM_DISSECT(cp, uint8_t *, NFSM_RNDUP(nfhlen)); + NFSBCOPY(cp, nfhp->nfh_fh, nfhlen); + } + } else if (laytype == NFSLAYOUT_FLEXFILE) { + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + + NFSX_HYPER); + mirrorcnt = fxdr_unsigned(int, *(tl + 2)); + NFSCL_DEBUG(4, "mirrorcnt=%d\n", mirrorcnt); + if (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS) { error = NFSERR_BADXDR; goto nfsmout; } - nfhp = malloc(sizeof(*nfhp) + nfhlen - 1, M_NFSFH, - M_WAITOK); - flp->nfsfl_fh[j] = nfhp; - flp->nfsfl_fhcnt++; - nfhp->nfh_len = nfhlen; - NFSM_DISSECT(cp, uint8_t *, NFSM_RNDUP(nfhlen)); - NFSBCOPY(cp, nfhp->nfh_fh, nfhlen); + flp = malloc(sizeof(*flp) + mirrorcnt * + sizeof(struct nfsffm), M_NFSFLAYOUT, M_WAITOK); + flp->nfsfl_flags = NFSFL_FLEXFILE; + flp->nfsfl_mirrorcnt = mirrorcnt; + flp->nfsfl_devp = NULL; + flp->nfsfl_off = off; + if (flp->nfsfl_off + retlen < flp->nfsfl_off) + flp->nfsfl_end = UINT64_MAX - flp->nfsfl_off; + else + flp->nfsfl_end = flp->nfsfl_off + retlen; + flp->nfsfl_iomode = iomode; + if (gotiomode == -1) + gotiomode = flp->nfsfl_iomode; + flp->nfsfl_stripeunit = fxdr_hyper(tl); + NFSCL_DEBUG(4, "stripeunit=%ju\n", + (uintmax_t)flp->nfsfl_stripeunit); + for (j = 0; j < mirrorcnt; j++) { + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); + k = fxdr_unsigned(int, *tl); + if (k < 1 || k > 128) { + error = NFSERR_BADXDR; + goto nfsmout; + } + NFSCL_DEBUG(4, "servercnt=%d\n", k); + for (l = 0; l < k; l++) { + NFSM_DISSECT(tl, uint32_t *, + NFSX_V4DEVICEID + NFSX_STATEID + + 2 * NFSX_UNSIGNED); + if (l == 0) { + /* Just use the first server. */ + NFSBCOPY(tl, + flp->nfsfl_ffm[j].dev, + NFSX_V4DEVICEID); + tl += (NFSX_V4DEVICEID / + NFSX_UNSIGNED); + tl++; + flp->nfsfl_ffm[j].st.seqid = + *tl++; + flp->nfsfl_ffm[j].st.other[0] = + *tl++; + flp->nfsfl_ffm[j].st.other[1] = + *tl++; + flp->nfsfl_ffm[j].st.other[2] = + *tl++; + NFSCL_DEBUG(4, "st.seqid=%u " + "st.o0=0x%x st.o1=0x%x " + "st.o2=0x%x\n", + flp->nfsfl_ffm[j].st.seqid, + flp->nfsfl_ffm[j].st.other[0], + flp->nfsfl_ffm[j].st.other[1], + flp->nfsfl_ffm[j].st.other[2]); + } else + tl += ((NFSX_V4DEVICEID + + NFSX_STATEID + + NFSX_UNSIGNED) / + NFSX_UNSIGNED); + fhcnt = fxdr_unsigned(int, *tl); + NFSCL_DEBUG(4, "fhcnt=%d\n", fhcnt); + if (fhcnt < 1 || + fhcnt > NFSDEV_MAXVERS) { + error = NFSERR_BADXDR; + goto nfsmout; + } + for (m = 0; m < fhcnt; m++) { + NFSM_DISSECT(tl, uint32_t *, + NFSX_UNSIGNED); + nfhlen = fxdr_unsigned(int, + *tl); + NFSCL_DEBUG(4, "nfhlen=%d\n", + nfhlen); + if (nfhlen <= 0 || nfhlen > + NFSX_V4FHMAX) { + error = NFSERR_BADXDR; + goto nfsmout; + } + NFSM_DISSECT(cp, uint8_t *, + NFSM_RNDUP(nfhlen)); + if (l == 0) { + flp->nfsfl_ffm[j].fhcnt + = fhcnt; + nfhp = malloc( + sizeof(*nfhp) + + nfhlen - 1, M_NFSFH, + M_WAITOK); + flp->nfsfl_ffm[j].fh[m] + = nfhp; + nfhp->nfh_len = nfhlen; + NFSBCOPY(cp, + nfhp->nfh_fh, + nfhlen); + NFSCL_DEBUG(4, + "got fh\n"); + } + } + /* Now, get the ffsd_user/ffds_group. */ + error = nfsrv_parseug(nd, 0, &user, + &grp, curthread); + NFSCL_DEBUG(4, "after parseu=%d\n", + error); + if (error == 0) + error = nfsrv_parseug(nd, 1, + &user, &grp, curthread); + NFSCL_DEBUG(4, "aft parseg=%d\n", + grp); + if (error != 0) + goto nfsmout; + NFSCL_DEBUG(4, "user=%d group=%d\n", + user, grp); + if (l == 0) { + flp->nfsfl_ffm[j].user = user; + flp->nfsfl_ffm[j].group = grp; + NFSCL_DEBUG(4, + "usr=%d grp=%d\n", user, + grp); + } + } + } + NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); + flp->nfsfl_fflags = fxdr_unsigned(uint32_t, *tl++); + flp->nfsfl_statshint = fxdr_unsigned(uint32_t, *tl); + NFSCL_DEBUG(4, "fflags=0x%x statshint=%d\n", + flp->nfsfl_fflags, flp->nfsfl_statshint); + } else { + error = NFSERR_BADXDR; + goto nfsmout; } if (flp->nfsfl_iomode == gotiomode) { /* Keep the list in increasing offset order. */ @@ -6098,6 +6725,7 @@ nfsrv_parselayoutget(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, else LIST_INSERT_AFTER(prevflp, flp, nfsfl_list); + NFSCL_DEBUG(4, "flp inserted\n"); } else { printf("nfscl_layoutget(): got wrong iomode\n"); nfscl_freeflayout(flp); @@ -6105,12 +6733,58 @@ nfsrv_parselayoutget(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, flp = NULL; } nfsmout: + NFSCL_DEBUG(4, "eo nfsrv_parselayoutget=%d\n", error); if (error != 0 && flp != NULL) nfscl_freeflayout(flp); return (error); } /* + * Parse a user/group digit string. + */ +static int +nfsrv_parseug(struct nfsrv_descript *nd, int dogrp, uid_t *uidp, gid_t *gidp, + NFSPROC_T *p) +{ + uint32_t *tl; + char *cp, *str, str0[NFSV4_SMALLSTR + 1]; + uint32_t len = 0; + int error = 0; + + NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); + len = fxdr_unsigned(uint32_t, *tl); + if (len > NFSV4_OPAQUELIMIT) { + error = NFSERR_BADXDR; + goto nfsmout; + } + NFSCL_DEBUG(4, "nfsrv_parseug: len=%d\n", len); + if (len == 0) { + if (dogrp != 0) + *gidp = GID_NOGROUP; + else + *uidp = UID_NOBODY; + return (0); + } + if (len > NFSV4_SMALLSTR) + str = malloc(len + 1, M_TEMP, M_WAITOK); + else + str = str0; + NFSM_DISSECT(cp, char *, NFSM_RNDUP(len)); + NFSBCOPY(cp, str, len); + str[len] = '\0'; + NFSCL_DEBUG(4, "nfsrv_parseug: str=%s\n", str); + if (dogrp != 0) + error = nfsv4_strtogid(nd, str, len, gidp, p); + else + error = nfsv4_strtouid(nd, str, len, uidp, p); +nfsmout: + if (len > NFSV4_SMALLSTR) + free(str, M_TEMP); + NFSCL_DEBUG(4, "eo nfsrv_parseug=%d\n", error); + return (error); +} + +/* * Similar to nfsrpc_getlayout(), except that it uses nfsrpc_openlayget(), * so that it does both an Open and a Layoutget. */ @@ -6124,11 +6798,15 @@ nfsrpc_getopenlayout(struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, struct nfsclflayout *flp; struct nfsclflayouthead flh; int error, islocked, layoutlen, recalled, retonclose, usecurstateid; - int laystat; + int layouttype, laystat; nfsv4stateid_t stateid; struct nfsclsession *tsep; error = 0; + if (NFSHASFLEXFILE(nmp)) + layouttype = NFSLAYOUT_FLEXFILE; + else + layouttype = NFSLAYOUT_NFSV4_1_FILES; /* * If lyp is returned non-NULL, there will be a refcnt (shared lock) * on it, iff flp != NULL or a lock (exclusive lock) on it iff @@ -6159,13 +6837,13 @@ nfsrpc_getopenlayout(struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, } error = nfsrpc_openlayoutrpc(nmp, vp, nfhp, fhlen, newfhp, newfhlen, mode, op, name, namelen, - dpp, &stateid, usecurstateid, layoutlen, + dpp, &stateid, usecurstateid, layouttype, layoutlen, &retonclose, &flh, &laystat, cred, p); NFSCL_DEBUG(4, "aft nfsrpc_openlayoutrpc laystat=%d err=%d\n", laystat, error); laystat = nfsrpc_layoutgetres(nmp, vp, newfhp, newfhlen, - &stateid, retonclose, NULL, &lyp, &flh, laystat, &islocked, - cred, p); + &stateid, retonclose, NULL, &lyp, &flh, layouttype, laystat, + &islocked, cred, p); } else error = nfsrpc_openrpc(nmp, vp, nfhp, fhlen, newfhp, newfhlen, mode, op, name, namelen, dpp, 0, 0, cred, p, 0, 0); @@ -6187,7 +6865,7 @@ static int nfsrpc_openlayoutrpc(struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, int fhlen, uint8_t *newfhp, int newfhlen, uint32_t mode, struct nfsclopen *op, uint8_t *name, int namelen, struct nfscldeleg **dpp, - nfsv4stateid_t *stateidp, int usecurstateid, + nfsv4stateid_t *stateidp, int usecurstateid, int layouttype, int layoutlen, int *retonclosep, struct nfsclflayouthead *flhp, int *laystatp, struct ucred *cred, NFSPROC_T *p) { @@ -6229,7 +6907,7 @@ nfsrpc_openlayoutrpc(struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, else iomode = NFSLAYOUTIOMODE_READ; nfsrv_setuplayoutget(nd, iomode, 0, UINT64_MAX, 0, stateidp, - layoutlen, usecurstateid); + layouttype, layoutlen, usecurstateid); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) @@ -6363,7 +7041,7 @@ nfsrpc_createlayout(vnode_t dvp, char *name, int namelen, struct vattr *vap, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff, int *unlockedp, nfsv4stateid_t *stateidp, - int usecurstateid, int layoutlen, int *retonclosep, + int usecurstateid, int layouttype, int layoutlen, int *retonclosep, struct nfsclflayouthead *flhp, int *laystatp) { uint32_t *tl; @@ -6442,7 +7120,7 @@ nfsrpc_createlayout(vnode_t dvp, char *name, int namelen, struct vattr *vap, *tl++ = txdr_unsigned(NFSV4OP_RESTOREFH); *tl = txdr_unsigned(NFSV4OP_LAYOUTGET); nfsrv_setuplayoutget(nd, NFSLAYOUTIOMODE_RW, 0, UINT64_MAX, 0, stateidp, - layoutlen, usecurstateid); + layouttype, layoutlen, usecurstateid); error = nfscl_request(nd, dvp, p, cred, dstuff); if (error != 0) return (error); @@ -6628,17 +7306,21 @@ nfsrpc_getcreatelayout(vnode_t dvp, char *name, int namelen, struct vattr *vap, struct nfsclsession *tsep; struct nfsmount *nmp; nfsv4stateid_t stateid; - int error, layoutlen, retonclose, laystat; + int error, layoutlen, layouttype, retonclose, laystat; error = 0; nmp = VFSTONFS(dvp->v_mount); + if (NFSHASFLEXFILE(nmp)) + layouttype = NFSLAYOUT_FLEXFILE; + else + layouttype = NFSLAYOUT_NFSV4_1_FILES; LIST_INIT(&flh); tsep = nfsmnt_mdssession(nmp); layoutlen = tsep->nfsess_maxcache - (NFSX_STATEID + 3 * NFSX_UNSIGNED); error = nfsrpc_createlayout(dvp, name, namelen, vap, cverf, fmode, owp, dpp, cred, p, dnap, nnap, nfhpp, attrflagp, dattrflagp, - dstuff, unlockedp, &stateid, 1, layoutlen, &retonclose, &flh, - &laystat); + dstuff, unlockedp, &stateid, 1, layouttype, layoutlen, &retonclose, + &flh, &laystat); NFSCL_DEBUG(4, "aft nfsrpc_createlayoutrpc laystat=%d err=%d\n", laystat, error); lyp = NULL; @@ -6646,10 +7328,11 @@ nfsrpc_getcreatelayout(vnode_t dvp, char *name, int namelen, struct vattr *vap, nfhp = *nfhpp; laystat = nfsrpc_layoutgetres(nmp, dvp, nfhp->nfh_fh, nfhp->nfh_len, &stateid, retonclose, NULL, &lyp, &flh, - laystat, NULL, cred, p); + layouttype, laystat, NULL, cred, p); } else laystat = nfsrpc_layoutgetres(nmp, dvp, NULL, 0, &stateid, - retonclose, NULL, &lyp, &flh, laystat, NULL, cred, p); + retonclose, NULL, &lyp, &flh, layouttype, laystat, NULL, + cred, p); if (laystat == 0) nfscl_rellayout(lyp, 0); return (error); @@ -6661,17 +7344,23 @@ nfsrpc_getcreatelayout(vnode_t dvp, char *name, int namelen, struct vattr *vap, static int nfsrpc_layoutgetres(struct nfsmount *nmp, vnode_t vp, uint8_t *newfhp, int newfhlen, nfsv4stateid_t *stateidp, int retonclose, uint32_t *notifybit, - struct nfscllayout **lypp, struct nfsclflayouthead *flhp, + struct nfscllayout **lypp, struct nfsclflayouthead *flhp, int layouttype, int laystat, int *islockedp, struct ucred *cred, NFSPROC_T *p) { struct nfsclflayout *tflp; struct nfscldevinfo *dip; + uint8_t *dev; if (laystat == NFSERR_UNKNLAYOUTTYPE) { - /* Disable PNFS. */ - NFSCL_DEBUG(1, "disable PNFS\n"); NFSLOCKMNT(nmp); - nmp->nm_state &= ~NFSSTA_PNFS; + if (!NFSHASFLEXFILE(nmp)) { + /* Switch to using Flex File Layout. */ + nmp->nm_state |= NFSSTA_FLEXFILE; + } else if (layouttype == NFSLAYOUT_FLEXFILE) { + /* Disable pNFS. */ + NFSCL_DEBUG(1, "disable PNFS\n"); + nmp->nm_state &= ~(NFSSTA_PNFS | NFSSTA_FLEXFILE); + } NFSUNLOCKMNT(nmp); } if (laystat == 0) { @@ -6680,9 +7369,12 @@ nfsrpc_layoutgetres(struct nfsmount *nmp, vnode_t vp, uint8_t *newfhp, laystat = nfscl_adddevinfo(nmp, NULL, tflp); NFSCL_DEBUG(4, "aft adddev=%d\n", laystat); if (laystat != 0) { - laystat = nfsrpc_getdeviceinfo(nmp, - tflp->nfsfl_dev, NFSLAYOUT_NFSV4_1_FILES, - notifybit, &dip, cred, p); + if (layouttype == NFSLAYOUT_FLEXFILE) + dev = tflp->nfsfl_ffm[0].dev; + else + dev = tflp->nfsfl_dev; + laystat = nfsrpc_getdeviceinfo(nmp, dev, + layouttype, notifybit, &dip, cred, p); NFSCL_DEBUG(4, "aft nfsrpc_gdi=%d\n", laystat); if (laystat != 0) @@ -6701,7 +7393,7 @@ nfsrpc_layoutgetres(struct nfsmount *nmp, vnode_t vp, uint8_t *newfhp, * get the fsid for the file system. */ laystat = nfscl_layout(nmp, vp, newfhp, newfhlen, stateidp, - retonclose, flhp, lypp, cred, p); + layouttype, retonclose, flhp, lypp, cred, p); NFSCL_DEBUG(4, "nfsrpc_layoutgetres: aft nfscl_layout=%d\n", laystat); if (laystat == 0 && islockedp != NULL) diff --git a/sys/fs/nfsclient/nfs_clstate.c b/sys/fs/nfsclient/nfs_clstate.c index e0687eeaf8bf..cedc157b0dac 100644 --- a/sys/fs/nfsclient/nfs_clstate.c +++ b/sys/fs/nfsclient/nfs_clstate.c @@ -4783,7 +4783,7 @@ nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers) */ APPLESTATIC int nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen, - nfsv4stateid_t *stateidp, int retonclose, + nfsv4stateid_t *stateidp, int layouttype, int retonclose, struct nfsclflayouthead *fhlp, struct nfscllayout **lypp, struct ucred *cred, NFSPROC_T *p) { @@ -4835,8 +4835,12 @@ nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen, lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0]; lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1]; lyp->nfsly_clp = clp; - lyp->nfsly_flags = (retonclose != 0) ? - (NFSLY_FILES | NFSLY_RETONCLOSE) : NFSLY_FILES; + if (layouttype == NFSLAYOUT_FLEXFILE) + lyp->nfsly_flags = NFSLY_FLEXFILE; + else + lyp->nfsly_flags = NFSLY_FILES; + if (retonclose != 0) + lyp->nfsly_flags |= NFSLY_RETONCLOSE; lyp->nfsly_fhlen = fhlen; NFSBCOPY(fhp, lyp->nfsly_fh, fhlen); TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list); @@ -5079,6 +5083,7 @@ nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, { struct nfsclclient *clp; struct nfscldevinfo *tdip; + uint8_t *dev; NFSLOCKCLSTATE(); clp = nmp->nm_clp; @@ -5088,7 +5093,11 @@ nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, free(dip, M_NFSDEVINFO); return (ENODEV); } - tdip = nfscl_finddevinfo(clp, flp->nfsfl_dev); + if ((flp->nfsfl_flags & NFSFL_FILE) != 0) + dev = flp->nfsfl_dev; + else + dev = flp->nfsfl_ffm[0].dev; + tdip = nfscl_finddevinfo(clp, dev); if (tdip != NULL) { tdip->nfsdi_layoutrefs++; flp->nfsfl_devp = tdip; @@ -5140,10 +5149,15 @@ nfscl_freelayout(struct nfscllayout *layp) APPLESTATIC void nfscl_freeflayout(struct nfsclflayout *flp) { - int i; - - for (i = 0; i < flp->nfsfl_fhcnt; i++) - free(flp->nfsfl_fh[i], M_NFSFH); + int i, j; + + if ((flp->nfsfl_flags & NFSFL_FILE) != 0) + for (i = 0; i < flp->nfsfl_fhcnt; i++) + free(flp->nfsfl_fh[i], M_NFSFH); + if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0) + for (i = 0; i < flp->nfsfl_mirrorcnt; i++) + for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++) + free(flp->nfsfl_ffm[i].fh[j], M_NFSFH); if (flp->nfsfl_devp != NULL) flp->nfsfl_devp->nfsdi_layoutrefs--; free(flp, M_NFSFLAYOUT); @@ -5235,12 +5249,17 @@ nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp, { struct nfsclrecalllayout *rp; nfsv4stateid_t stateid; + int layouttype; NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER); stateid.seqid = lyp->nfsly_stateid.seqid; + if ((lyp->nfsly_flags & NFSLY_FILES) != 0) + layouttype = NFSLAYOUT_NFSV4_1_FILES; + else + layouttype = NFSLAYOUT_FLEXFILE; LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) { (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh, - lyp->nfsly_fhlen, 0, NFSLAYOUT_NFSV4_1_FILES, + lyp->nfsly_fhlen, 0, layouttype, rp->nfsrecly_iomode, rp->nfsrecly_recalltype, rp->nfsrecly_off, rp->nfsrecly_len, &stateid, cred, p, NULL); @@ -5256,15 +5275,27 @@ nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp, { struct nfsclflayout *flp; uint64_t len; - int error; + int error, layouttype; + if ((lyp->nfsly_flags & NFSLY_FILES) != 0) + layouttype = NFSLAYOUT_NFSV4_1_FILES; + else + layouttype = NFSLAYOUT_FLEXFILE; LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) { - if (flp->nfsfl_off <= lyp->nfsly_lastbyte) { + if (layouttype == NFSLAYOUT_FLEXFILE && + (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) { + NFSCL_DEBUG(4, "Flex file: no layoutcommit\n"); + /* If not supported, don't bother doing it. */ + NFSLOCKMNT(nmp); + nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT; + NFSUNLOCKMNT(nmp); + break; + } else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) { len = flp->nfsfl_end - flp->nfsfl_off; error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh, lyp->nfsly_fhlen, 0, flp->nfsfl_off, len, lyp->nfsly_lastbyte, &lyp->nfsly_stateid, - NFSLAYOUT_NFSV4_1_FILES, cred, p, NULL); + layouttype, cred, p, NULL); NFSCL_DEBUG(4, "layoutcommit err=%d\n", error); if (error == NFSERR_NOTSUPP) { /* If not supported, don't bother doing it. */ |