diff --git a/sys/fs/nfs/nfs_var.h b/sys/fs/nfs/nfs_var.h index 9dd3be3d5c42..ef7d0050e1a3 100644 --- a/sys/fs/nfs/nfs_var.h +++ b/sys/fs/nfs/nfs_var.h @@ -1,788 +1,788 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ /* * XXX needs and because of typedefs */ struct uio; struct ucred; struct nfscred; NFSPROC_T; struct buf; struct sockaddr_in; struct nfs_dlmount; struct file; struct nfsmount; struct socket; struct nfsreq; struct nfssockreq; struct vattr; struct nameidata; struct nfsnode; struct nfsfh; struct sillyrename; struct componentname; struct nfsd_srvargs; struct nfsrv_descript; struct nfs_fattr; union nethostaddr; struct nfsstate; struct nfslock; struct nfsclient; struct nfslayout; struct nfsdsession; struct nfslockconflict; struct nfsd_idargs; struct nfsd_clid; struct nfsusrgrp; struct nfsclowner; struct nfsclopen; struct nfsclopenhead; struct nfsclclient; struct nfsclsession; struct nfscllockowner; struct nfscllock; struct nfscldeleg; struct nfscllayout; struct nfscldevinfo; struct nfsv4lock; struct nfsvattr; struct nfs_vattr; struct NFSSVCARGS; struct nfsdevice; struct pnfsdsfile; struct pnfsdsattr; #ifdef __FreeBSD__ NFS_ACCESS_ARGS; NFS_OPEN_ARGS; NFS_GETATTR_ARGS; NFS_LOOKUP_ARGS; NFS_READDIR_ARGS; #endif /* nfs_nfsdstate.c */ int nfsrv_setclient(struct nfsrv_descript *, struct nfsclient **, nfsquad_t *, nfsquad_t *, NFSPROC_T *); int nfsrv_getclient(nfsquad_t, int, struct nfsclient **, struct nfsdsession *, nfsquad_t, uint32_t, struct nfsrv_descript *, NFSPROC_T *); int nfsrv_destroyclient(nfsquad_t, NFSPROC_T *); int nfsrv_destroysession(struct nfsrv_descript *, uint8_t *); int nfsrv_bindconnsess(struct nfsrv_descript *, uint8_t *, int *); int nfsrv_freestateid(struct nfsrv_descript *, nfsv4stateid_t *, NFSPROC_T *); int nfsrv_teststateid(struct nfsrv_descript *, nfsv4stateid_t *, NFSPROC_T *); int nfsrv_adminrevoke(struct nfsd_clid *, NFSPROC_T *); void nfsrv_dumpclients(struct nfsd_dumpclients *, int); void nfsrv_dumplocks(vnode_t, struct nfsd_dumplocks *, int, NFSPROC_T *); int nfsrv_lockctrl(vnode_t, struct nfsstate **, struct nfslock **, struct nfslockconflict *, nfsquad_t, nfsv4stateid_t *, struct nfsexstuff *, struct nfsrv_descript *, NFSPROC_T *); int nfsrv_openctrl(struct nfsrv_descript *, vnode_t, struct nfsstate **, nfsquad_t, nfsv4stateid_t *, nfsv4stateid_t *, u_int32_t *, struct nfsexstuff *, NFSPROC_T *, u_quad_t); int nfsrv_opencheck(nfsquad_t, nfsv4stateid_t *, struct nfsstate *, vnode_t, struct nfsrv_descript *, NFSPROC_T *, int); int nfsrv_openupdate(vnode_t, struct nfsstate *, nfsquad_t, nfsv4stateid_t *, struct nfsrv_descript *, NFSPROC_T *, int *); int nfsrv_delegupdate(struct nfsrv_descript *, nfsquad_t, nfsv4stateid_t *, vnode_t, int, struct ucred *, NFSPROC_T *, int *); int nfsrv_releaselckown(struct nfsstate *, nfsquad_t, NFSPROC_T *); void nfsrv_zapclient(struct nfsclient *, NFSPROC_T *); int nfssvc_idname(struct nfsd_idargs *); -void nfsrv_servertimer(void); +void nfsrv_servertimer(void * __unused); int nfsrv_getclientipaddr(struct nfsrv_descript *, struct nfsclient *); void nfsrv_setupstable(NFSPROC_T *); void nfsrv_updatestable(NFSPROC_T *); void nfsrv_writestable(u_char *, int, int, NFSPROC_T *); void nfsrv_throwawayopens(NFSPROC_T *); int nfsrv_checkremove(vnode_t, int, struct nfsrv_descript *, nfsquad_t, NFSPROC_T *); void nfsd_recalldelegation(vnode_t, NFSPROC_T *); void nfsd_disabledelegation(vnode_t, NFSPROC_T *); int nfsrv_checksetattr(vnode_t, struct nfsrv_descript *, nfsv4stateid_t *, struct nfsvattr *, nfsattrbit_t *, struct nfsexstuff *, NFSPROC_T *); int nfsrv_checkgetattr(struct nfsrv_descript *, vnode_t, struct nfsvattr *, nfsattrbit_t *, NFSPROC_T *); int nfsrv_nfsuserdport(struct nfsuserd_args *, NFSPROC_T *); void nfsrv_nfsuserddelport(void); void nfsrv_throwawayallstate(NFSPROC_T *); int nfsrv_checksequence(struct nfsrv_descript *, uint32_t, uint32_t *, uint32_t *, int, uint32_t *, NFSPROC_T *); int nfsrv_checkreclaimcomplete(struct nfsrv_descript *, int); void nfsrv_cache_session(struct nfsrv_descript *, struct mbuf **); void nfsrv_freeallbackchannel_xprts(void); int nfsrv_layoutcommit(struct nfsrv_descript *, vnode_t, int, int, uint64_t, uint64_t, uint64_t, int, struct timespec *, int, nfsv4stateid_t *, int, char *, int *, uint64_t *, struct ucred *, NFSPROC_T *); int nfsrv_layoutget(struct nfsrv_descript *, vnode_t, struct nfsexstuff *, int, int *, uint64_t *, uint64_t *, uint64_t, nfsv4stateid_t *, int, int *, int *, char *, struct ucred *, NFSPROC_T *); void nfsrv_flexmirrordel(char *, NFSPROC_T *); void nfsrv_recalloldlayout(NFSPROC_T *); int nfsrv_layoutreturn(struct nfsrv_descript *, vnode_t, int, int, uint64_t, uint64_t, int, int, nfsv4stateid_t *, int, uint32_t *, int *, struct ucred *, NFSPROC_T *); int nfsrv_getdevinfo(char *, int, uint32_t *, uint32_t *, int *, char **); void nfsrv_freeonedevid(struct nfsdevice *); void nfsrv_freealllayoutsanddevids(void); void nfsrv_freefilelayouts(fhandle_t *); int nfsrv_deldsserver(int, char *, NFSPROC_T *); struct nfsdevice *nfsrv_deldsnmp(int, struct nfsmount *, NFSPROC_T *); int nfsrv_delds(char *, NFSPROC_T *); int nfsrv_createdevids(struct nfsd_nfsd_args *, NFSPROC_T *); int nfsrv_checkdsattr(vnode_t, NFSPROC_T *); int nfsrv_copymr(vnode_t, vnode_t, vnode_t, struct nfsdevice *, struct pnfsdsfile *, struct pnfsdsfile *, int, struct ucred *, NFSPROC_T *); int nfsrv_mdscopymr(char *, char *, char *, char *, int *, char *, NFSPROC_T *, struct vnode **, struct vnode **, struct pnfsdsfile **, struct nfsdevice **, struct nfsdevice **); void nfsrv_marknospc(char *, bool); /* nfs_nfsdserv.c */ int nfsrvd_access(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_getattr(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_setattr(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_lookup(struct nfsrv_descript *, int, vnode_t, vnode_t *, fhandle_t *, struct nfsexstuff *); int nfsrvd_readlink(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_read(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_write(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_create(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_mknod(struct nfsrv_descript *, int, vnode_t, vnode_t *, fhandle_t *, struct nfsexstuff *); int nfsrvd_remove(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_rename(struct nfsrv_descript *, int, vnode_t, vnode_t, struct nfsexstuff *, struct nfsexstuff *); int nfsrvd_link(struct nfsrv_descript *, int, vnode_t, vnode_t, struct nfsexstuff *, struct nfsexstuff *); int nfsrvd_symlink(struct nfsrv_descript *, int, vnode_t, vnode_t *, fhandle_t *, struct nfsexstuff *); int nfsrvd_mkdir(struct nfsrv_descript *, int, vnode_t, vnode_t *, fhandle_t *, struct nfsexstuff *); int nfsrvd_readdir(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_readdirplus(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_commit(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_statfs(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_fsinfo(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_close(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_delegpurge(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_delegreturn(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_getfh(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_lock(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_lockt(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_locku(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_openconfirm(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_opendowngrade(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_renew(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_secinfo(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_secinfononame(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_setclientid(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_setclientidcfrm(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_verify(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_open(struct nfsrv_descript *, int, vnode_t, vnode_t *, fhandle_t *, struct nfsexstuff *); int nfsrvd_openattr(struct nfsrv_descript *, int, vnode_t, vnode_t *, fhandle_t *, struct nfsexstuff *); int nfsrvd_releaselckown(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_pathconf(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_exchangeid(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_createsession(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_sequence(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_reclaimcomplete(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_destroyclientid(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_bindconnsess(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_destroysession(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_freestateid(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_layoutget(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_getdevinfo(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_layoutcommit(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_layoutreturn(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_ioadvise(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_layouterror(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_layoutstats(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_teststateid(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_allocate(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_copy_file_range(struct nfsrv_descript *, int, vnode_t, vnode_t, struct nfsexstuff *, struct nfsexstuff *); int nfsrvd_seek(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_getxattr(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_setxattr(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_rmxattr(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_listxattr(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); int nfsrvd_notsupp(struct nfsrv_descript *, int, vnode_t, struct nfsexstuff *); /* nfs_nfsdsocket.c */ void nfsrvd_rephead(struct nfsrv_descript *); void nfsrvd_dorpc(struct nfsrv_descript *, int, u_char *, int, u_int32_t); /* nfs_nfsdcache.c */ void nfsrvd_initcache(void); int nfsrvd_getcache(struct nfsrv_descript *); struct nfsrvcache *nfsrvd_updatecache(struct nfsrv_descript *); void nfsrvd_sentcache(struct nfsrvcache *, int, uint32_t); void nfsrvd_cleancache(void); void nfsrvd_refcache(struct nfsrvcache *); void nfsrvd_derefcache(struct nfsrvcache *); void nfsrvd_delcache(struct nfsrvcache *); void nfsrc_trimcache(uint64_t, uint32_t, int); /* nfs_commonsubs.c */ void nfscl_reqstart(struct nfsrv_descript *, int, struct nfsmount *, u_int8_t *, int, u_int32_t **, struct nfsclsession *, int, int, struct ucred *); void nfsm_stateidtom(struct nfsrv_descript *, nfsv4stateid_t *, int); void nfscl_fillsattr(struct nfsrv_descript *, struct vattr *, vnode_t, int, u_int32_t); void newnfs_init(void); int nfsaddr_match(int, union nethostaddr *, NFSSOCKADDR_T); int nfsaddr2_match(NFSSOCKADDR_T, NFSSOCKADDR_T); int nfsm_strtom(struct nfsrv_descript *, const char *, int); int nfsm_mbufuio(struct nfsrv_descript *, struct uio *, int); int nfsm_fhtom(struct nfsrv_descript *, u_int8_t *, int, int); int nfsm_advance(struct nfsrv_descript *, int, int); void *nfsm_dissct(struct nfsrv_descript *, int, int); void newnfs_copycred(struct nfscred *, struct ucred *); void newnfs_copyincred(struct ucred *, struct nfscred *); int nfsrv_dissectacl(struct nfsrv_descript *, NFSACL_T *, bool, int *, int *, NFSPROC_T *); int nfsrv_getattrbits(struct nfsrv_descript *, nfsattrbit_t *, int *, int *); int nfsv4_loadattr(struct nfsrv_descript *, vnode_t, struct nfsvattr *, struct nfsfh **, fhandle_t *, int, struct nfsv3_pathconf *, struct statfs *, struct nfsstatfs *, struct nfsfsinfo *, NFSACL_T *, int, int *, u_int32_t *, u_int32_t *, NFSPROC_T *, struct ucred *); int nfsv4_lock(struct nfsv4lock *, int, int *, struct mtx *, struct mount *); void nfsv4_unlock(struct nfsv4lock *, int); void nfsv4_relref(struct nfsv4lock *); void nfsv4_getref(struct nfsv4lock *, int *, struct mtx *, struct mount *); int nfsv4_getref_nonblock(struct nfsv4lock *); int nfsv4_testlock(struct nfsv4lock *); int nfsrv_mtostr(struct nfsrv_descript *, char *, int); void nfsrv_cleanusergroup(void); int nfsrv_checkutf8(u_int8_t *, int); int newnfs_sndlock(int *); void newnfs_sndunlock(int *); int nfsv4_getipaddr(struct nfsrv_descript *, struct sockaddr_in *, struct sockaddr_in6 *, sa_family_t *, int *); int nfsv4_seqsession(uint32_t, uint32_t, uint32_t, struct nfsslot *, struct mbuf **, uint16_t); void nfsv4_seqsess_cacherep(uint32_t, struct nfsslot *, int, struct mbuf **); void nfsv4_setsequence(struct nfsmount *, struct nfsrv_descript *, struct nfsclsession *, int, struct ucred *); int nfsv4_sequencelookup(struct nfsmount *, struct nfsclsession *, int *, int *, uint32_t *, uint8_t *, bool); void nfsv4_freeslot(struct nfsclsession *, int, bool); struct ucred *nfsrv_getgrpscred(struct ucred *); struct nfsdevice *nfsv4_findmirror(struct nfsmount *); void nfsm_set(struct nfsrv_descript *, u_int); struct mbuf *nfsm_add_ext_pgs(struct mbuf *, int, int *); int nfsrpc_destroysession(struct nfsmount *, struct nfsclsession *, struct ucred *, NFSPROC_T *); /* nfs_clcomsubs.c */ void nfsm_uiombuf(struct nfsrv_descript *, struct uio *, int); struct mbuf *nfsm_uiombuflist(struct uio *, int, u_int); u_int8_t *nfscl_getmyip(struct nfsmount *, struct in6_addr *, int *); int nfsm_getfh(struct nfsrv_descript *, struct nfsfh **); int nfscl_mtofh(struct nfsrv_descript *, struct nfsfh **, struct nfsvattr *, int *); int nfscl_postop_attr(struct nfsrv_descript *, struct nfsvattr *, int *, void *); int nfscl_wcc_data(struct nfsrv_descript *, vnode_t, struct nfsvattr *, int *, int *, uint64_t *); int nfsm_loadattr(struct nfsrv_descript *, struct nfsvattr *); int nfscl_request(struct nfsrv_descript *, vnode_t, NFSPROC_T *, struct ucred *, void *); /* nfs_nfsdsubs.c */ void nfsd_fhtovp(struct nfsrv_descript *, struct nfsrvfh *, int, vnode_t *, struct nfsexstuff *, mount_t *, int, int); int nfsd_excred(struct nfsrv_descript *, struct nfsexstuff *, struct ucred *, bool); int nfsrv_mtofh(struct nfsrv_descript *, struct nfsrvfh *); int nfsrv_putattrbit(struct nfsrv_descript *, nfsattrbit_t *); void nfsrv_wcc(struct nfsrv_descript *, int, struct nfsvattr *, int, struct nfsvattr *); int nfsv4_fillattr(struct nfsrv_descript *, struct mount *, vnode_t, NFSACL_T *, struct vattr *, fhandle_t *, int, nfsattrbit_t *, struct ucred *, NFSPROC_T *, int, int, int, int, uint64_t, struct statfs *); void nfsrv_fillattr(struct nfsrv_descript *, struct nfsvattr *); struct mbuf *nfsrv_adj(struct mbuf *, int, int); void nfsrv_postopattr(struct nfsrv_descript *, int, struct nfsvattr *); int nfsd_errmap(struct nfsrv_descript *); void nfsv4_uidtostr(uid_t, u_char **, int *); int nfsv4_strtouid(struct nfsrv_descript *, u_char *, int, uid_t *); void nfsv4_gidtostr(gid_t, u_char **, int *); int nfsv4_strtogid(struct nfsrv_descript *, u_char *, int, gid_t *); int nfsrv_checkuidgid(struct nfsrv_descript *, struct nfsvattr *); void nfsrv_fixattr(struct nfsrv_descript *, vnode_t, struct nfsvattr *, NFSACL_T *, NFSPROC_T *, nfsattrbit_t *, struct nfsexstuff *); int nfsrv_errmoved(int); int nfsrv_putreferralattr(struct nfsrv_descript *, nfsattrbit_t *, struct nfsreferral *, int, int *); int nfsrv_parsename(struct nfsrv_descript *, char *, u_long *, NFSPATHLEN_T *); void nfsd_init(void); int nfsd_checkrootexp(struct nfsrv_descript *); void nfsd_getminorvers(struct nfsrv_descript *, u_char *, u_char **, int *, u_int32_t *); /* nfs_clvfsops.c */ void nfscl_retopts(struct nfsmount *, char *, size_t); /* nfs_commonport.c */ int nfsrv_lookupfilename(struct nameidata *, char *, NFSPROC_T *); void nfsrv_object_create(vnode_t, NFSPROC_T *); int nfsrv_mallocmget_limit(void); int nfsvno_v4rootexport(struct nfsrv_descript *); void newnfs_portinit(void); struct ucred *newnfs_getcred(void); void newnfs_setroot(struct ucred *); int nfs_catnap(int, int, const char *); struct nfsreferral *nfsv4root_getreferral(vnode_t, vnode_t, u_int32_t); int nfsvno_pathconf(vnode_t, int, long *, struct ucred *, NFSPROC_T *); int nfsrv_atroot(vnode_t, uint64_t *); int nfs_supportsnfsv4acls(vnode_t); /* nfs_commonacl.c */ int nfsrv_dissectace(struct nfsrv_descript *, struct acl_entry *, bool, int *, int *, NFSPROC_T *); int nfsrv_buildacl(struct nfsrv_descript *, NFSACL_T *, enum vtype, NFSPROC_T *); int nfsrv_compareacl(NFSACL_T *, NFSACL_T *); /* nfs_clrpcops.c */ int nfsrpc_null(vnode_t, struct ucred *, NFSPROC_T *); int nfsrpc_access(vnode_t, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *); int nfsrpc_accessrpc(vnode_t, u_int32_t, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, u_int32_t *, void *); int nfsrpc_open(vnode_t, int, struct ucred *, NFSPROC_T *); int nfsrpc_openrpc(struct nfsmount *, vnode_t, u_int8_t *, int, u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int, struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *, int, int); int nfsrpc_opendowngrade(vnode_t, u_int32_t, struct nfsclopen *, struct ucred *, NFSPROC_T *); int nfsrpc_close(vnode_t, int, NFSPROC_T *); int nfsrpc_closerpc(struct nfsrv_descript *, struct nfsmount *, struct nfsclopen *, struct ucred *, NFSPROC_T *, int); int nfsrpc_openconfirm(vnode_t, u_int8_t *, int, struct nfsclopen *, struct ucred *, NFSPROC_T *); int nfsrpc_setclient(struct nfsmount *, struct nfsclclient *, int, bool *, struct ucred *, NFSPROC_T *); int nfsrpc_getattr(vnode_t, struct ucred *, NFSPROC_T *, struct nfsvattr *, void *); int nfsrpc_getattrnovp(struct nfsmount *, u_int8_t *, int, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, u_int64_t *, uint32_t *); int nfsrpc_setattr(vnode_t, struct vattr *, NFSACL_T *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_lookup(vnode_t, char *, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *); int nfsrpc_readlink(vnode_t, struct uio *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_read(vnode_t, struct uio *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_write(vnode_t, struct uio *, int *, int *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, int, int); int nfsrpc_mknod(vnode_t, char *, int, struct vattr *, u_int32_t, enum vtype, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *); int nfsrpc_create(vnode_t, char *, int, struct vattr *, nfsquad_t, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *); int nfsrpc_remove(vnode_t, char *, int, vnode_t, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_rename(vnode_t, vnode_t, char *, int, vnode_t, vnode_t, char *, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, int *, int *, void *, void *); int nfsrpc_link(vnode_t, vnode_t, char *, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, int *, int *, void *); int nfsrpc_symlink(vnode_t, char *, int, const char *, struct vattr *, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *); int nfsrpc_mkdir(vnode_t, char *, int, struct vattr *, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *); int nfsrpc_rmdir(vnode_t, char *, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_readdir(vnode_t, struct uio *, nfsuint64 *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, int *, void *); int nfsrpc_readdirplus(vnode_t, struct uio *, nfsuint64 *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, int *, void *); int nfsrpc_commit(vnode_t, u_quad_t, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_advlock(vnode_t, off_t, int, struct flock *, int, struct ucred *, NFSPROC_T *, void *, int); int nfsrpc_lockt(struct nfsrv_descript *, vnode_t, struct nfsclclient *, u_int64_t, u_int64_t, struct flock *, struct ucred *, NFSPROC_T *, void *, int); int nfsrpc_lock(struct nfsrv_descript *, struct nfsmount *, vnode_t, u_int8_t *, int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short, struct ucred *, NFSPROC_T *, int); int nfsrpc_statfs(vnode_t, struct nfsstatfs *, struct nfsfsinfo *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_fsinfo(vnode_t, struct nfsfsinfo *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_pathconf(vnode_t, struct nfsv3_pathconf *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); int nfsrpc_renew(struct nfsclclient *, struct nfsclds *, struct ucred *, NFSPROC_T *); int nfsrpc_rellockown(struct nfsmount *, struct nfscllockowner *, uint8_t *, int, struct ucred *, NFSPROC_T *); int nfsrpc_getdirpath(struct nfsmount *, u_char *, struct ucred *, NFSPROC_T *); int nfsrpc_delegreturn(struct nfscldeleg *, struct ucred *, struct nfsmount *, NFSPROC_T *, int); int nfsrpc_getacl(vnode_t, struct ucred *, NFSPROC_T *, NFSACL_T *, void *); int nfsrpc_setacl(vnode_t, struct ucred *, NFSPROC_T *, NFSACL_T *, void *); int nfsrpc_exchangeid(struct nfsmount *, struct nfsclclient *, struct nfssockreq *, int, uint32_t, struct nfsclds **, struct ucred *, NFSPROC_T *); int nfsrpc_createsession(struct nfsmount *, struct nfsclsession *, struct nfssockreq *, struct nfsclds *, uint32_t, int, struct ucred *, NFSPROC_T *); int nfsrpc_destroyclient(struct nfsmount *, struct nfsclclient *, struct ucred *, NFSPROC_T *); int nfsrpc_getdeviceinfo(struct nfsmount *, uint8_t *, int, uint32_t *, struct nfscldevinfo **, struct ucred *, NFSPROC_T *); int nfsrpc_layoutcommit(struct nfsmount *, uint8_t *, int, int, uint64_t, uint64_t, uint64_t, nfsv4stateid_t *, int, struct ucred *, NFSPROC_T *, void *); int nfsrpc_layoutreturn(struct nfsmount *, uint8_t *, int, int, int, uint32_t, int, uint64_t, uint64_t, nfsv4stateid_t *, struct ucred *, NFSPROC_T *, uint32_t, uint32_t, char *); int nfsrpc_reclaimcomplete(struct nfsmount *, struct ucred *, NFSPROC_T *); int nfsrpc_advise(vnode_t, off_t, uint64_t, int, struct ucred *, NFSPROC_T *); int nfscl_doiods(vnode_t, struct uio *, int *, int *, uint32_t, int, struct ucred *, NFSPROC_T *); int nfscl_findlayoutforio(struct nfscllayout *, uint64_t, uint32_t, struct nfsclflayout **); void nfscl_freenfsclds(struct nfsclds *); int nfsrpc_allocate(vnode_t, off_t, off_t, struct nfsvattr *, int *, struct ucred *, NFSPROC_T *, void *); int nfsrpc_copy_file_range(vnode_t, off_t *, vnode_t, off_t *, size_t *, unsigned int, int *, struct nfsvattr *, int *, struct nfsvattr *, struct ucred *, bool, bool *); int nfsrpc_seek(vnode_t, off_t *, bool *, int, struct ucred *, struct nfsvattr *, int *); int nfsrpc_getextattr(vnode_t, const char *, struct uio *, ssize_t *, struct nfsvattr *, int *, struct ucred *, NFSPROC_T *); int nfsrpc_setextattr(vnode_t, const char *, struct uio *, struct nfsvattr *, int *, struct ucred *, NFSPROC_T *); int nfsrpc_listextattr(vnode_t, uint64_t *, struct uio *, size_t *, bool *, struct nfsvattr *, int *, struct ucred *, NFSPROC_T *); int nfsrpc_rmextattr(vnode_t, const char *, struct nfsvattr *, int *, struct ucred *, NFSPROC_T *); void nfsrpc_bindconnsess(CLIENT *, void *, struct ucred *); /* nfs_clstate.c */ int nfscl_open(vnode_t, u_int8_t *, int, u_int32_t, int, struct ucred *, NFSPROC_T *, struct nfsclowner **, struct nfsclopen **, int *, int *, int, bool); int nfscl_getstateid(vnode_t, u_int8_t *, int, u_int32_t, int, struct ucred *, NFSPROC_T *, nfsv4stateid_t *, void **); void nfscl_ownerrelease(struct nfsmount *, struct nfsclowner *, int, int, int); void nfscl_openrelease(struct nfsmount *, struct nfsclopen *, int, int); int nfscl_getcl(struct mount *, struct ucred *, NFSPROC_T *, bool, bool, struct nfsclclient **); struct nfsclclient *nfscl_findcl(struct nfsmount *); void nfscl_clientrelease(struct nfsclclient *); void nfscl_freelock(struct nfscllock *, int); void nfscl_freelockowner(struct nfscllockowner *, int); int nfscl_getbytelock(vnode_t, u_int64_t, u_int64_t, short, struct ucred *, NFSPROC_T *, struct nfsclclient *, int, void *, int, u_int8_t *, u_int8_t *, struct nfscllockowner **, int *, int *); int nfscl_relbytelock(vnode_t, u_int64_t, u_int64_t, struct ucred *, NFSPROC_T *, int, struct nfsclclient *, void *, int, struct nfscllockowner **, int *); int nfscl_checkwritelocked(vnode_t, struct flock *, struct ucred *, NFSPROC_T *, void *, int); void nfscl_lockrelease(struct nfscllockowner *, int, int); void nfscl_fillclid(u_int64_t, char *, u_int8_t *, u_int16_t); void nfscl_filllockowner(void *, u_int8_t *, int); void nfscl_freeopen(struct nfsclopen *, int, bool); void nfscl_umount(struct nfsmount *, NFSPROC_T *, struct nfscldeleghead *); void nfscl_renewthread(struct nfsclclient *, NFSPROC_T *); void nfscl_initiate_recovery(struct nfsclclient *); int nfscl_hasexpired(struct nfsclclient *, u_int32_t, NFSPROC_T *); void nfscl_dumpstate(struct nfsmount *, int, int, int, int); void nfscl_dupopen(vnode_t, int); int nfscl_getclose(vnode_t, struct nfsclclient **); int nfscl_doclose(vnode_t, struct nfsclclient **, NFSPROC_T *); int nfsrpc_doclose(struct nfsmount *, struct nfsclopen *, NFSPROC_T *, bool, bool); int nfscl_deleg(mount_t, struct nfsclclient *, u_int8_t *, int, struct ucred *, NFSPROC_T *, struct nfscldeleg **); void nfscl_lockinit(struct nfsv4lock *); void nfscl_lockexcl(struct nfsv4lock *, void *); void nfscl_lockunlock(struct nfsv4lock *); void nfscl_lockderef(struct nfsv4lock *); void nfscl_delegreturnvp(vnode_t, NFSPROC_T *); void nfscl_docb(struct nfsrv_descript *, NFSPROC_T *); void nfscl_releasealllocks(struct nfsclclient *, vnode_t, NFSPROC_T *, void *, int); int nfscl_lockt(vnode_t, struct nfsclclient *, u_int64_t, u_int64_t, struct flock *, NFSPROC_T *, void *, int); int nfscl_mustflush(vnode_t); int nfscl_nodeleg(vnode_t, int); int nfscl_removedeleg(vnode_t, NFSPROC_T *, nfsv4stateid_t *); int nfscl_getref(struct nfsmount *); void nfscl_relref(struct nfsmount *); int nfscl_renamedeleg(vnode_t, nfsv4stateid_t *, int *, vnode_t, nfsv4stateid_t *, int *, NFSPROC_T *); void nfscl_reclaimnode(vnode_t); void nfscl_newnode(vnode_t); void nfscl_delegmodtime(vnode_t); void nfscl_deleggetmodtime(vnode_t, struct timespec *); int nfscl_tryclose(struct nfsclopen *, struct ucred *, struct nfsmount *, NFSPROC_T *, bool); void nfscl_cleanup(NFSPROC_T *); int nfscl_layout(struct nfsmount *, vnode_t, u_int8_t *, int, nfsv4stateid_t *, int, int, struct nfsclflayouthead *, struct nfscllayout **, struct ucred *, NFSPROC_T *); struct nfscllayout *nfscl_getlayout(struct nfsclclient *, uint8_t *, int, uint64_t, uint32_t, struct nfsclflayout **, int *); void nfscl_dserr(uint32_t, uint32_t, struct nfscldevinfo *, struct nfscllayout *, struct nfsclds *); void nfscl_cancelreqs(struct nfsclds *); void nfscl_rellayout(struct nfscllayout *, int); struct nfscldevinfo *nfscl_getdevinfo(struct nfsclclient *, uint8_t *, struct nfscldevinfo *); void nfscl_reldevinfo(struct nfscldevinfo *); int nfscl_adddevinfo(struct nfsmount *, struct nfscldevinfo *, int, struct nfsclflayout *); void nfscl_freelayout(struct nfscllayout *); void nfscl_freeflayout(struct nfsclflayout *); void nfscl_freedevinfo(struct nfscldevinfo *); int nfscl_layoutcommit(vnode_t, NFSPROC_T *); /* nfs_clport.c */ int nfscl_nget(mount_t, vnode_t, struct nfsfh *, struct componentname *, NFSPROC_T *, struct nfsnode **, void *, int); NFSPROC_T *nfscl_getparent(NFSPROC_T *); void nfscl_start_renewthread(struct nfsclclient *); void nfscl_loadsbinfo(struct nfsmount *, struct nfsstatfs *, void *); void nfscl_loadfsinfo (struct nfsmount *, struct nfsfsinfo *); void nfscl_delegreturn(struct nfscldeleg *, int, struct nfsmount *, struct ucred *, NFSPROC_T *); void nfsrvd_cbinit(int); int nfscl_checksattr(struct vattr *, struct nfsvattr *); int nfscl_ngetreopen(mount_t, u_int8_t *, int, NFSPROC_T *, struct nfsnode **); int nfscl_procdoesntexist(u_int8_t *); int nfscl_maperr(NFSPROC_T *, int, uid_t, gid_t); /* nfs_clsubs.c */ void nfscl_init(void); /* nfs_clbio.c */ int ncl_flush(vnode_t, int, NFSPROC_T *, int, int); /* nfs_clnode.c */ void ncl_invalcaches(vnode_t); /* nfs_nfsdport.c */ int nfsvno_getattr(vnode_t, struct nfsvattr *, struct nfsrv_descript *, NFSPROC_T *, int, nfsattrbit_t *); int nfsvno_setattr(vnode_t, struct nfsvattr *, struct ucred *, NFSPROC_T *, struct nfsexstuff *); int nfsvno_getfh(vnode_t, fhandle_t *, NFSPROC_T *); int nfsvno_accchk(vnode_t, accmode_t, struct ucred *, struct nfsexstuff *, NFSPROC_T *, int, int, u_int32_t *); int nfsvno_namei(struct nfsrv_descript *, struct nameidata *, vnode_t, int, struct nfsexstuff *, NFSPROC_T *, vnode_t *); void nfsvno_setpathbuf(struct nameidata *, char **, u_long **); void nfsvno_relpathbuf(struct nameidata *); int nfsvno_readlink(vnode_t, struct ucred *, int, NFSPROC_T *, struct mbuf **, struct mbuf **, int *); int nfsvno_read(vnode_t, off_t, int, struct ucred *, int, NFSPROC_T *, struct mbuf **, struct mbuf **); int nfsvno_write(vnode_t, off_t, int, int *, struct mbuf *, char *, struct ucred *, NFSPROC_T *); int nfsvno_createsub(struct nfsrv_descript *, struct nameidata *, vnode_t *, struct nfsvattr *, int *, int32_t *, NFSDEV_T, struct nfsexstuff *); int nfsvno_mknod(struct nameidata *, struct nfsvattr *, struct ucred *, NFSPROC_T *); int nfsvno_mkdir(struct nameidata *, struct nfsvattr *, uid_t, struct ucred *, NFSPROC_T *, struct nfsexstuff *); int nfsvno_symlink(struct nameidata *, struct nfsvattr *, char *, int, int, uid_t, struct ucred *, NFSPROC_T *, struct nfsexstuff *); int nfsvno_getsymlink(struct nfsrv_descript *, struct nfsvattr *, NFSPROC_T *, char **, int *); int nfsvno_removesub(struct nameidata *, int, struct ucred *, NFSPROC_T *, struct nfsexstuff *); int nfsvno_rmdirsub(struct nameidata *, int, struct ucred *, NFSPROC_T *, struct nfsexstuff *); int nfsvno_rename(struct nameidata *, struct nameidata *, u_int32_t, u_int32_t, struct ucred *, NFSPROC_T *); int nfsvno_link(struct nameidata *, vnode_t, struct ucred *, NFSPROC_T *, struct nfsexstuff *); int nfsvno_fsync(vnode_t, u_int64_t, int, struct ucred *, NFSPROC_T *); int nfsvno_statfs(vnode_t, struct statfs *); void nfsvno_getfs(struct nfsfsinfo *, int); void nfsvno_open(struct nfsrv_descript *, struct nameidata *, nfsquad_t, nfsv4stateid_t *, struct nfsstate *, int *, struct nfsvattr *, int32_t *, int, NFSACL_T *, nfsattrbit_t *, struct ucred *, struct nfsexstuff *, vnode_t *); int nfsvno_updfilerev(vnode_t, struct nfsvattr *, struct nfsrv_descript *, NFSPROC_T *); int nfsvno_fillattr(struct nfsrv_descript *, struct mount *, vnode_t, struct nfsvattr *, fhandle_t *, int, nfsattrbit_t *, struct ucred *, NFSPROC_T *, int, int, int, int, uint64_t); int nfsrv_sattr(struct nfsrv_descript *, vnode_t, struct nfsvattr *, nfsattrbit_t *, NFSACL_T *, NFSPROC_T *); int nfsv4_sattr(struct nfsrv_descript *, vnode_t, struct nfsvattr *, nfsattrbit_t *, NFSACL_T *, NFSPROC_T *); int nfsvno_checkexp(mount_t, NFSSOCKADDR_T, struct nfsexstuff *, struct ucred **); int nfsvno_fhtovp(mount_t, fhandle_t *, NFSSOCKADDR_T, int, vnode_t *, struct nfsexstuff *, struct ucred **); vnode_t nfsvno_getvp(fhandle_t *); int nfsvno_advlock(vnode_t, int, u_int64_t, u_int64_t, NFSPROC_T *); int nfsrv_v4rootexport(void *, struct ucred *, NFSPROC_T *); int nfsvno_testexp(struct nfsrv_descript *, struct nfsexstuff *); uint32_t nfsrv_hashfh(fhandle_t *); uint32_t nfsrv_hashsessionid(uint8_t *); void nfsrv_backupstable(void); int nfsrv_dsgetdevandfh(struct vnode *, NFSPROC_T *, int *, fhandle_t *, char *); int nfsrv_dsgetsockmnt(struct vnode *, int, char *, int *, int *, NFSPROC_T *, struct vnode **, fhandle_t *, char *, char *, struct vnode **, struct nfsmount **, struct nfsmount *, int *, int *); int nfsrv_dscreate(struct vnode *, struct vattr *, struct vattr *, fhandle_t *, struct pnfsdsfile *, struct pnfsdsattr *, char *, struct ucred *, NFSPROC_T *, struct vnode **); int nfsrv_updatemdsattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); void nfsrv_killrpcs(struct nfsmount *); int nfsrv_setacl(struct vnode *, NFSACL_T *, struct ucred *, NFSPROC_T *); int nfsvno_seek(struct nfsrv_descript *, struct vnode *, u_long, off_t *, int, bool *, struct ucred *, NFSPROC_T *); int nfsvno_allocate(struct vnode *, off_t, off_t, struct ucred *, NFSPROC_T *); int nfsvno_getxattr(struct vnode *, char *, uint32_t, struct ucred *, uint64_t, int, struct thread *, struct mbuf **, struct mbuf **, int *); int nfsvno_setxattr(struct vnode *, char *, int, struct mbuf *, char *, struct ucred *, struct thread *); int nfsvno_rmxattr(struct nfsrv_descript *, struct vnode *, char *, struct ucred *, struct thread *); int nfsvno_listxattr(struct vnode *, uint64_t, struct ucred *, struct thread *, u_char **, uint32_t *, bool *); void nfsm_trimtrailing(struct nfsrv_descript *, struct mbuf *, char *, int, int); bool nfsrv_checkwrongsec(struct nfsrv_descript *, int, enum vtype); void nfsrv_checknospc(void); /* nfs_commonkrpc.c */ int newnfs_nmcancelreqs(struct nfsmount *); void newnfs_set_sigmask(struct thread *, sigset_t *); void newnfs_restore_sigmask(struct thread *, sigset_t *); int newnfs_msleep(struct thread *, void *, struct mtx *, int, char *, int); int newnfs_request(struct nfsrv_descript *, struct nfsmount *, struct nfsclient *, struct nfssockreq *, vnode_t, NFSPROC_T *, struct ucred *, u_int32_t, u_int32_t, u_char *, int, u_int64_t *, struct nfsclsession *); int newnfs_connect(struct nfsmount *, struct nfssockreq *, struct ucred *, NFSPROC_T *, int, bool, struct __rpc_client **); void newnfs_disconnect(struct nfsmount *, struct nfssockreq *); int newnfs_sigintr(struct nfsmount *, NFSPROC_T *); /* nfs_nfsdkrpc.c */ int nfsrvd_addsock(struct file *); int nfsrvd_nfsd(NFSPROC_T *, struct nfsd_nfsd_args *); void nfsrvd_init(int); /* nfs_clkrpc.c */ int nfscbd_addsock(struct file *); int nfscbd_nfsd(NFSPROC_T *, struct nfsd_nfscbd_args *); diff --git a/sys/fs/nfs/nfsport.h b/sys/fs/nfs/nfsport.h index 2955ae43de85..0a0e68787f65 100644 --- a/sys/fs/nfs/nfsport.h +++ b/sys/fs/nfs/nfsport.h @@ -1,1217 +1,1231 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _NFS_NFSPORT_H_ #define _NFS_NFSPORT_H_ /* * In general, I'm not fond of #includes in .h files, but this seems * to be the cleanest way to handle #include files for the ports. */ #ifdef _KERNEL #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_nfs.h" #include "opt_ufs.h" /* * These types must be defined before the nfs includes. */ #define NFSSOCKADDR_T struct sockaddr * #define NFSPROC_T struct thread #define NFSDEV_T dev_t #define NFSSVCARGS nfssvc_args #define NFSACL_T struct acl /* * These should be defined as the types used for the corresponding VOP's * argument type. */ #define NFS_ACCESS_ARGS struct vop_access_args #define NFS_OPEN_ARGS struct vop_open_args #define NFS_GETATTR_ARGS struct vop_getattr_args #define NFS_LOOKUP_ARGS struct vop_lookup_args #define NFS_READDIR_ARGS struct vop_readdir_args /* * Allocate mbufs. Must succeed and never set the mbuf ptr to NULL. */ #define NFSMGET(m) do { \ MGET((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGET((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMGETHDR(m) do { \ MGETHDR((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGETHDR((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMCLGET(m, w) do { \ MGET((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGET((m), M_WAITOK, MT_DATA); \ } \ MCLGET((m), (w)); \ } while (0) #define NFSMCLGETHDR(m, w) do { \ MGETHDR((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGETHDR((m), M_WAITOK, MT_DATA); \ } \ } while (0) /* * Client side constant for size of a lockowner name. */ #define NFSV4CL_LOCKNAMELEN 12 /* * Type for a mutex lock. */ #define NFSMUTEX_T struct mtx +/* Define the NFSD_VNET macros similar to !VIMAGE. */ +#define NFSD_VNET_NAME(n) n +#define NFSD_VNET_DECLARE(t, n) extern t n +#define NFSD_VNET_DEFINE(t, n) t n +#define NFSD_VNET_DEFINE_STATIC(t, n) static t n +#define NFSD_VNET(n) (n) + +#define CTLFLAG_NFSD_VNET 0 + +#define NFSD_CURVNET_SET(n) +#define NFSD_CURVNET_SET_QUIET(n) +#define NFSD_CURVNET_RESTORE() +#define NFSD_TD_TO_VNET(n) NULL + #endif /* _KERNEL */ /* * NFSv4 Operation numbers. */ #define NFSV4OP_ACCESS 3 #define NFSV4OP_CLOSE 4 #define NFSV4OP_COMMIT 5 #define NFSV4OP_CREATE 6 #define NFSV4OP_DELEGPURGE 7 #define NFSV4OP_DELEGRETURN 8 #define NFSV4OP_GETATTR 9 #define NFSV4OP_GETFH 10 #define NFSV4OP_LINK 11 #define NFSV4OP_LOCK 12 #define NFSV4OP_LOCKT 13 #define NFSV4OP_LOCKU 14 #define NFSV4OP_LOOKUP 15 #define NFSV4OP_LOOKUPP 16 #define NFSV4OP_NVERIFY 17 #define NFSV4OP_OPEN 18 #define NFSV4OP_OPENATTR 19 #define NFSV4OP_OPENCONFIRM 20 #define NFSV4OP_OPENDOWNGRADE 21 #define NFSV4OP_PUTFH 22 #define NFSV4OP_PUTPUBFH 23 #define NFSV4OP_PUTROOTFH 24 #define NFSV4OP_READ 25 #define NFSV4OP_READDIR 26 #define NFSV4OP_READLINK 27 #define NFSV4OP_REMOVE 28 #define NFSV4OP_RENAME 29 #define NFSV4OP_RENEW 30 #define NFSV4OP_RESTOREFH 31 #define NFSV4OP_SAVEFH 32 #define NFSV4OP_SECINFO 33 #define NFSV4OP_SETATTR 34 #define NFSV4OP_SETCLIENTID 35 #define NFSV4OP_SETCLIENTIDCFRM 36 #define NFSV4OP_VERIFY 37 #define NFSV4OP_WRITE 38 #define NFSV4OP_RELEASELCKOWN 39 /* * Must be one greater than the last Operation#. */ #define NFSV4OP_NOPS 40 /* * Additional Ops for NFSv4.1. */ #define NFSV4OP_BACKCHANNELCTL 40 #define NFSV4OP_BINDCONNTOSESS 41 #define NFSV4OP_EXCHANGEID 42 #define NFSV4OP_CREATESESSION 43 #define NFSV4OP_DESTROYSESSION 44 #define NFSV4OP_FREESTATEID 45 #define NFSV4OP_GETDIRDELEG 46 #define NFSV4OP_GETDEVINFO 47 #define NFSV4OP_GETDEVLIST 48 #define NFSV4OP_LAYOUTCOMMIT 49 #define NFSV4OP_LAYOUTGET 50 #define NFSV4OP_LAYOUTRETURN 51 #define NFSV4OP_SECINFONONAME 52 #define NFSV4OP_SEQUENCE 53 #define NFSV4OP_SETSSV 54 #define NFSV4OP_TESTSTATEID 55 #define NFSV4OP_WANTDELEG 56 #define NFSV4OP_DESTROYCLIENTID 57 #define NFSV4OP_RECLAIMCOMPL 58 /* * Must be one more than last op#. */ #define NFSV41_NOPS 59 /* * Additional operations for NFSv4.2. */ #define NFSV4OP_ALLOCATE 59 #define NFSV4OP_COPY 60 #define NFSV4OP_COPYNOTIFY 61 #define NFSV4OP_DEALLOCATE 62 #define NFSV4OP_IOADVISE 63 #define NFSV4OP_LAYOUTERROR 64 #define NFSV4OP_LAYOUTSTATS 65 #define NFSV4OP_OFFLOADCANCEL 66 #define NFSV4OP_OFFLOADSTATUS 67 #define NFSV4OP_READPLUS 68 #define NFSV4OP_SEEK 69 #define NFSV4OP_WRITESAME 70 #define NFSV4OP_CLONE 71 /* One greater than the last Operation # defined in RFC-7862. */ #define NFSV42_PURENOPS 72 /* and the optional Extended attribute operations (RFC-8276). */ #define NFSV4OP_GETXATTR 72 #define NFSV4OP_SETXATTR 73 #define NFSV4OP_LISTXATTRS 74 #define NFSV4OP_REMOVEXATTR 75 /* * Must be one more than the last NFSv4.2 op#. */ #define NFSV42_NOPS 76 /* Quirky case if the illegal op code */ #define NFSV4OP_OPILLEGAL 10044 /* * Fake NFSV4OP_xxx used for nfsstat. Start at NFSV42_NOPS. */ #define NFSV4OP_SYMLINK (NFSV42_NOPS) #define NFSV4OP_MKDIR (NFSV42_NOPS + 1) #define NFSV4OP_RMDIR (NFSV42_NOPS + 2) #define NFSV4OP_READDIRPLUS (NFSV42_NOPS + 3) #define NFSV4OP_MKNOD (NFSV42_NOPS + 4) #define NFSV4OP_FSSTAT (NFSV42_NOPS + 5) #define NFSV4OP_FSINFO (NFSV42_NOPS + 6) #define NFSV4OP_PATHCONF (NFSV42_NOPS + 7) #define NFSV4OP_V3CREATE (NFSV42_NOPS + 8) /* * This is the count of the fake operations listed above. */ #define NFSV4OP_FAKENOPS 9 /* * and the Callback OPs */ #define NFSV4OP_CBGETATTR 3 #define NFSV4OP_CBRECALL 4 /* * Must be one greater than the last Callback Operation# for NFSv4.0. */ #define NFSV4OP_CBNOPS 5 /* * Additional Callback Ops for NFSv4.1 only. */ #define NFSV4OP_CBLAYOUTRECALL 5 #define NFSV4OP_CBNOTIFY 6 #define NFSV4OP_CBPUSHDELEG 7 #define NFSV4OP_CBRECALLANY 8 #define NFSV4OP_CBRECALLOBJAVAIL 9 #define NFSV4OP_CBRECALLSLOT 10 #define NFSV4OP_CBSEQUENCE 11 #define NFSV4OP_CBWANTCANCELLED 12 #define NFSV4OP_CBNOTIFYLOCK 13 #define NFSV4OP_CBNOTIFYDEVID 14 #define NFSV41_CBNOPS 15 /* * Additional callback operations for NFSv4.2. */ #define NFSV4OP_CBOFFLOAD 15 #define NFSV42_CBNOPS 16 /* * The lower numbers -> 21 are used by NFSv2 and v3. These define higher * numbers used by NFSv4. * NFS_V3NPROCS is one greater than the last V3 op and NFS_NPROCS is * one greater than the last number. */ #ifndef NFS_V3NPROCS #define NFS_V3NPROCS 22 #define NFSPROC_LOOKUPP 22 #define NFSPROC_SETCLIENTID 23 #define NFSPROC_SETCLIENTIDCFRM 24 #define NFSPROC_LOCK 25 #define NFSPROC_LOCKU 26 #define NFSPROC_OPEN 27 #define NFSPROC_CLOSE 28 #define NFSPROC_OPENCONFIRM 29 #define NFSPROC_LOCKT 30 #define NFSPROC_OPENDOWNGRADE 31 #define NFSPROC_RENEW 32 #define NFSPROC_PUTROOTFH 33 #define NFSPROC_RELEASELCKOWN 34 #define NFSPROC_DELEGRETURN 35 #define NFSPROC_RETDELEGREMOVE 36 #define NFSPROC_RETDELEGRENAME1 37 #define NFSPROC_RETDELEGRENAME2 38 #define NFSPROC_GETACL 39 #define NFSPROC_SETACL 40 /* * Must be defined as one higher than the last Proc# above. */ #define NFSV4_NPROCS 41 /* Additional procedures for NFSv4.1. */ #define NFSPROC_EXCHANGEID 41 #define NFSPROC_CREATESESSION 42 #define NFSPROC_DESTROYSESSION 43 #define NFSPROC_DESTROYCLIENT 44 #define NFSPROC_FREESTATEID 45 #define NFSPROC_LAYOUTGET 46 #define NFSPROC_GETDEVICEINFO 47 #define NFSPROC_LAYOUTCOMMIT 48 #define NFSPROC_LAYOUTRETURN 49 #define NFSPROC_RECLAIMCOMPL 50 #define NFSPROC_WRITEDS 51 #define NFSPROC_READDS 52 #define NFSPROC_COMMITDS 53 #define NFSPROC_OPENLAYGET 54 #define NFSPROC_CREATELAYGET 55 /* * Must be defined as one higher than the last NFSv4.1 Proc# above. */ #define NFSV41_NPROCS 56 /* Additional procedures for NFSv4.2. */ #define NFSPROC_IOADVISE 56 #define NFSPROC_ALLOCATE 57 #define NFSPROC_COPY 58 #define NFSPROC_SEEK 59 #define NFSPROC_SEEKDS 60 /* and the ones for the optional Extended attribute support (RFC-8276). */ #define NFSPROC_GETEXTATTR 61 #define NFSPROC_SETEXTATTR 62 #define NFSPROC_RMEXTATTR 63 #define NFSPROC_LISTEXTATTR 64 /* BindConnectionToSession, done by the krpc for a new connection. */ #define NFSPROC_BINDCONNTOSESS 65 /* Do a Lookup+Open for "oneopenown". */ #define NFSPROC_LOOKUPOPEN 66 /* Do an NFSv4.2 Deallocate. */ #define NFSPROC_DEALLOCATE 67 /* Do an NFSv4.2 LayoutError. */ #define NFSPROC_LAYOUTERROR 68 /* Do an NFSv4 Verify+Write. */ #define NFSPROC_APPENDWRITE 69 /* * Must be defined as one higher than the last NFSv4.2 Proc# above. */ #define NFSV42_NPROCS 70 /* Value of NFSV42_NPROCS for old nfsstats structure. (Always 69) */ #define NFSV42_OLDNPROCS 69 #endif /* NFS_V3NPROCS */ /* * Newest stats structure. * The vers field will be set to NFSSTATS_V1 by the caller. */ #define NFSSTATS_V1 2 struct nfsstatsv1 { int vers; /* Set to version requested by caller. */ uint64_t attrcache_hits; uint64_t attrcache_misses; uint64_t lookupcache_hits; uint64_t lookupcache_misses; uint64_t direofcache_hits; uint64_t direofcache_misses; uint64_t accesscache_hits; uint64_t accesscache_misses; uint64_t biocache_reads; uint64_t read_bios; uint64_t read_physios; uint64_t biocache_writes; uint64_t write_bios; uint64_t write_physios; uint64_t biocache_readlinks; uint64_t readlink_bios; uint64_t biocache_readdirs; uint64_t readdir_bios; uint64_t rpccnt[NFSV42_NPROCS + 10]; uint64_t rpcretries; uint64_t srvrpccnt[NFSV42_NOPS + NFSV4OP_FAKENOPS + 15]; uint64_t srvlayouts; uint64_t cllayouts; uint64_t rpcrequests; uint64_t rpctimeouts; uint64_t rpcunexpected; uint64_t rpcinvalid; uint64_t srvcache_inproghits; uint64_t reserved_2; uint64_t srvcache_nonidemdonehits; uint64_t srvcache_misses; uint64_t srvcache_tcppeak; int srvcache_size; /* Updated by atomic_xx_int(). */ uint64_t srvclients; uint64_t srvopenowners; uint64_t srvopens; uint64_t srvlockowners; uint64_t srvlocks; uint64_t srvdelegates; uint64_t cbrpccnt[NFSV42_CBNOPS + 10]; uint64_t clopenowners; uint64_t clopens; uint64_t cllockowners; uint64_t cllocks; uint64_t cldelegates; uint64_t cllocalopenowners; uint64_t cllocalopens; uint64_t cllocallockowners; uint64_t cllocallocks; uint64_t srvstartcnt; uint64_t srvdonecnt; uint64_t srvbytes[NFSV42_NOPS + NFSV4OP_FAKENOPS + 15]; uint64_t srvops[NFSV42_NOPS + NFSV4OP_FAKENOPS + 15]; struct bintime srvduration[NFSV42_NOPS + NFSV4OP_FAKENOPS + 15]; struct bintime busyfrom; struct bintime busytime; }; /* * Newer stats structure. * The vers field will be set to NFSSTATS_OV1 by the caller. */ #define NFSSTATS_OV1 1 struct nfsstatsov1 { int vers; /* Set to version requested by caller. */ uint64_t attrcache_hits; uint64_t attrcache_misses; uint64_t lookupcache_hits; uint64_t lookupcache_misses; uint64_t direofcache_hits; uint64_t direofcache_misses; uint64_t accesscache_hits; uint64_t accesscache_misses; uint64_t biocache_reads; uint64_t read_bios; uint64_t read_physios; uint64_t biocache_writes; uint64_t write_bios; uint64_t write_physios; uint64_t biocache_readlinks; uint64_t readlink_bios; uint64_t biocache_readdirs; uint64_t readdir_bios; uint64_t rpccnt[NFSV42_OLDNPROCS]; uint64_t rpcretries; uint64_t srvrpccnt[NFSV42_PURENOPS + NFSV4OP_FAKENOPS]; uint64_t reserved_0; uint64_t reserved_1; uint64_t rpcrequests; uint64_t rpctimeouts; uint64_t rpcunexpected; uint64_t rpcinvalid; uint64_t srvcache_inproghits; uint64_t reserved_2; uint64_t srvcache_nonidemdonehits; uint64_t srvcache_misses; uint64_t srvcache_tcppeak; int srvcache_size; /* Updated by atomic_xx_int(). */ uint64_t srvclients; uint64_t srvopenowners; uint64_t srvopens; uint64_t srvlockowners; uint64_t srvlocks; uint64_t srvdelegates; uint64_t cbrpccnt[NFSV42_CBNOPS]; uint64_t clopenowners; uint64_t clopens; uint64_t cllockowners; uint64_t cllocks; uint64_t cldelegates; uint64_t cllocalopenowners; uint64_t cllocalopens; uint64_t cllocallockowners; uint64_t cllocallocks; uint64_t srvstartcnt; uint64_t srvdonecnt; uint64_t srvbytes[NFSV42_PURENOPS + NFSV4OP_FAKENOPS]; uint64_t srvops[NFSV42_PURENOPS + NFSV4OP_FAKENOPS]; struct bintime srvduration[NFSV42_PURENOPS + NFSV4OP_FAKENOPS]; struct bintime busyfrom; struct bintime busytime; }; /* * Old stats structure. */ struct ext_nfsstats { int attrcache_hits; int attrcache_misses; int lookupcache_hits; int lookupcache_misses; int direofcache_hits; int direofcache_misses; int accesscache_hits; int accesscache_misses; int biocache_reads; int read_bios; int read_physios; int biocache_writes; int write_bios; int write_physios; int biocache_readlinks; int readlink_bios; int biocache_readdirs; int readdir_bios; int rpccnt[NFSV4_NPROCS]; int rpcretries; int srvrpccnt[NFSV4OP_NOPS + NFSV4OP_FAKENOPS]; int reserved_0; int reserved_1; int rpcrequests; int rpctimeouts; int rpcunexpected; int rpcinvalid; int srvcache_inproghits; int reserved_2; int srvcache_nonidemdonehits; int srvcache_misses; int srvcache_tcppeak; int srvcache_size; int srvclients; int srvopenowners; int srvopens; int srvlockowners; int srvlocks; int srvdelegates; int cbrpccnt[NFSV4OP_CBNOPS]; int clopenowners; int clopens; int cllockowners; int cllocks; int cldelegates; int cllocalopenowners; int cllocalopens; int cllocallockowners; int cllocallocks; }; #ifdef _KERNEL /* * Define NFS_NPROCS as NFSV4_NPROCS for the experimental kernel code. */ #ifndef NFS_NPROCS #define NFS_NPROCS NFSV4_NPROCS #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Just to keep nfs_var.h happy. */ struct nfs_vattr { int junk; }; struct nfsvattr { struct vattr na_vattr; nfsattrbit_t na_suppattr; u_int64_t na_mntonfileno; u_int64_t na_filesid[2]; }; #define na_type na_vattr.va_type #define na_mode na_vattr.va_mode #define na_nlink na_vattr.va_nlink #define na_uid na_vattr.va_uid #define na_gid na_vattr.va_gid #define na_fsid na_vattr.va_fsid #define na_fileid na_vattr.va_fileid #define na_size na_vattr.va_size #define na_blocksize na_vattr.va_blocksize #define na_atime na_vattr.va_atime #define na_mtime na_vattr.va_mtime #define na_ctime na_vattr.va_ctime #define na_btime na_vattr.va_birthtime #define na_gen na_vattr.va_gen #define na_flags na_vattr.va_flags #define na_rdev na_vattr.va_rdev #define na_bytes na_vattr.va_bytes #define na_filerev na_vattr.va_filerev #define na_vaflags na_vattr.va_vaflags #include /* * This is the header structure used for the lists, etc. (It has the * above record in it. */ struct nfsrv_stablefirst { LIST_HEAD(, nfsrv_stable) nsf_head; /* Head of nfsrv_stable list */ time_t nsf_eograce; /* Time grace period ends */ time_t *nsf_bootvals; /* Previous boottime values */ struct file *nsf_fp; /* File table pointer */ u_char nsf_flags; /* NFSNSF_ flags */ struct nfsf_rec nsf_rec; /* and above first record */ }; #define nsf_lease nsf_rec.lease #define nsf_numboots nsf_rec.numboots /* NFSNSF_xxx flags */ #define NFSNSF_UPDATEDONE 0x01 #define NFSNSF_GRACEOVER 0x02 #define NFSNSF_NEEDLOCK 0x04 #define NFSNSF_EXPIREDCLIENT 0x08 #define NFSNSF_NOOPENS 0x10 #define NFSNSF_OK 0x20 /* * Maximum number of boot times allowed in record. Although there is * really no need for a fixed upper bound, this serves as a sanity check * for a corrupted file. */ #define NFSNSF_MAXNUMBOOTS 10000 /* * This structure defines the other records in the file. The * nst_client array is actually the size of the client string name. */ struct nfst_rec { u_int16_t len; u_char flag; u_char client[1]; }; /* and the values for flag */ #define NFSNST_NEWSTATE 0x1 #define NFSNST_REVOKE 0x2 #define NFSNST_GOTSTATE 0x4 #define NFSNST_RECLAIMED 0x8 /* * This structure is linked onto nfsrv_stablefirst for the duration of * reclaim. */ struct nfsrv_stable { LIST_ENTRY(nfsrv_stable) nst_list; struct nfsclient *nst_clp; struct nfst_rec nst_rec; }; #define nst_timestamp nst_rec.timestamp #define nst_len nst_rec.len #define nst_flag nst_rec.flag #define nst_client nst_rec.client /* * At some point the server will run out of kernel storage for * state structures. For FreeBSD5.2, this results in a panic * kmem_map is full. It happens at well over 1000000 opens plus * locks on a PIII-800 with 256Mbytes, so that is where I've set * the limit. If your server panics due to too many opens/locks, * decrease the size of NFSRV_V4STATELIMIT. If you find the server * returning NFS4ERR_RESOURCE a lot and have lots of memory, try * increasing it. */ #define NFSRV_V4STATELIMIT 500000 /* Max # of Opens + Locks */ /* * The type required differs with BSDen (just the second arg). */ void nfsrvd_rcv(struct socket *, void *, int); /* * Macros for handling socket addresses. (Hopefully this makes the code * more portable, since I've noticed some 'BSD don't have sockaddrs in * mbufs any more.) */ #define NFSSOCKADDR(a, t) ((t)(a)) #define NFSSOCKADDRSIZE(a, s) ((a)->sa_len = (s)) /* * These should be defined as a process or thread structure, as required * for signal handling, etc. */ #define NFSNEWCRED(c) (crdup(c)) #define NFSPROCCRED(p) ((p)->td_ucred) #define NFSFREECRED(c) (crfree(c)) #define NFSUIOPROC(u, p) ((u)->uio_td = NULL) #define NFSPROCP(p) ((p)->td_proc) /* * Define these so that cn_hash and its length is ignored. */ #define NFSCNHASHZERO(c) #define NFSCNHASH(c, v) #define NCHNAMLEN 9999999 /* * Handle SMP stuff: */ #define NFSSTATESPINLOCK extern struct mtx nfs_state_mutex #define NFSLOCKSTATE() mtx_lock(&nfs_state_mutex) #define NFSUNLOCKSTATE() mtx_unlock(&nfs_state_mutex) #define NFSSTATEMUTEXPTR (&nfs_state_mutex) #define NFSREQSPINLOCK extern struct mtx nfs_req_mutex #define NFSLOCKREQ() mtx_lock(&nfs_req_mutex) #define NFSUNLOCKREQ() mtx_unlock(&nfs_req_mutex) #define NFSSOCKMUTEX extern struct mtx nfs_slock_mutex #define NFSSOCKMUTEXPTR (&nfs_slock_mutex) #define NFSLOCKSOCK() mtx_lock(&nfs_slock_mutex) #define NFSUNLOCKSOCK() mtx_unlock(&nfs_slock_mutex) #define NFSNAMEIDMUTEX extern struct mtx nfs_nameid_mutex #define NFSNAMEIDMUTEXPTR (&nfs_nameid_mutex) #define NFSLOCKNAMEID() mtx_lock(&nfs_nameid_mutex) #define NFSUNLOCKNAMEID() mtx_unlock(&nfs_nameid_mutex) #define NFSNAMEIDREQUIRED() mtx_assert(&nfs_nameid_mutex, MA_OWNED) #define NFSCLSTATEMUTEX extern struct mtx nfs_clstate_mutex #define NFSCLSTATEMUTEXPTR (&nfs_clstate_mutex) #define NFSLOCKCLSTATE() mtx_lock(&nfs_clstate_mutex) #define NFSUNLOCKCLSTATE() mtx_unlock(&nfs_clstate_mutex) #define NFSDLOCKMUTEX extern struct mtx newnfsd_mtx #define NFSDLOCKMUTEXPTR (&newnfsd_mtx) #define NFSD_LOCK() mtx_lock(&newnfsd_mtx) #define NFSD_UNLOCK() mtx_unlock(&newnfsd_mtx) #define NFSD_LOCK_ASSERT() mtx_assert(&newnfsd_mtx, MA_OWNED) #define NFSD_UNLOCK_ASSERT() mtx_assert(&newnfsd_mtx, MA_NOTOWNED) #define NFSV4ROOTLOCKMUTEX extern struct mtx nfs_v4root_mutex #define NFSV4ROOTLOCKMUTEXPTR (&nfs_v4root_mutex) #define NFSLOCKV4ROOTMUTEX() mtx_lock(&nfs_v4root_mutex) #define NFSUNLOCKV4ROOTMUTEX() mtx_unlock(&nfs_v4root_mutex) #define NFSLOCKNODE(n) mtx_lock(&((n)->n_mtx)) #define NFSUNLOCKNODE(n) mtx_unlock(&((n)->n_mtx)) #define NFSASSERTNODE(n) mtx_assert(&((n)->n_mtx), MA_OWNED) #define NFSLOCKMNT(m) mtx_lock(&((m)->nm_mtx)) #define NFSUNLOCKMNT(m) mtx_unlock(&((m)->nm_mtx)) #define NFSLOCKIOD() mtx_lock(&ncl_iod_mutex) #define NFSUNLOCKIOD() mtx_unlock(&ncl_iod_mutex) #define NFSASSERTIOD() mtx_assert(&ncl_iod_mutex, MA_OWNED) #define NFSLOCKREQUEST(r) mtx_lock(&((r)->r_mtx)) #define NFSUNLOCKREQUEST(r) mtx_unlock(&((r)->r_mtx)) #define NFSLOCKSOCKREQ(r) mtx_lock(&((r)->nr_mtx)) #define NFSUNLOCKSOCKREQ(r) mtx_unlock(&((r)->nr_mtx)) #define NFSLOCKDS(d) mtx_lock(&((d)->nfsclds_mtx)) #define NFSUNLOCKDS(d) mtx_unlock(&((d)->nfsclds_mtx)) #define NFSSESSIONMUTEXPTR(s) (&((s)->mtx)) #define NFSLOCKSESSION(s) mtx_lock(&((s)->mtx)) #define NFSUNLOCKSESSION(s) mtx_unlock(&((s)->mtx)) #define NFSLAYOUTMUTEXPTR(l) (&((l)->mtx)) #define NFSLOCKLAYOUT(l) mtx_lock(&((l)->mtx)) #define NFSUNLOCKLAYOUT(l) mtx_unlock(&((l)->mtx)) #define NFSDDSMUTEXPTR (&nfsrv_dslock_mtx) #define NFSDDSLOCK() mtx_lock(&nfsrv_dslock_mtx) #define NFSDDSUNLOCK() mtx_unlock(&nfsrv_dslock_mtx) #define NFSDDONTLISTMUTEXPTR (&nfsrv_dontlistlock_mtx) #define NFSDDONTLISTLOCK() mtx_lock(&nfsrv_dontlistlock_mtx) #define NFSDDONTLISTUNLOCK() mtx_unlock(&nfsrv_dontlistlock_mtx) #define NFSDRECALLMUTEXPTR (&nfsrv_recalllock_mtx) #define NFSDRECALLLOCK() mtx_lock(&nfsrv_recalllock_mtx) #define NFSDRECALLUNLOCK() mtx_unlock(&nfsrv_recalllock_mtx) /* * Use these macros to initialize/free a mutex. */ #define NFSINITSOCKMUTEX(m) mtx_init((m), "nfssock", NULL, MTX_DEF) #define NFSFREEMUTEX(m) mtx_destroy((m)) int nfsmsleep(void *, void *, int, const char *, struct timespec *); /* * And weird vm stuff in the nfs server. */ #define PDIRUNLOCK 0x0 #define MAX_COMMIT_COUNT (1024 * 1024) /* * Define these to handle the type of va_rdev. */ #define NFSMAKEDEV(m, n) makedev((m), (n)) #define NFSMAJOR(d) major(d) #define NFSMINOR(d) minor(d) /* * The vnode tag for nfsv4root. */ #define VT_NFSV4ROOT "nfsv4root" /* * Define whatever it takes to do a vn_rdwr(). */ #define NFSD_RDWR(r, v, b, l, o, s, i, c, a, p) \ vn_rdwr((r), (v), (b), (l), (o), (s), (i), (c), NULL, (a), (p)) /* * Macros for handling memory for different BSDen. * NFSBCOPY(src, dst, len) - copies len bytes, non-overlapping * NFSOVBCOPY(src, dst, len) - ditto, but data areas might overlap * NFSBCMP(cp1, cp2, len) - compare len bytes, return 0 if same * NFSBZERO(cp, len) - set len bytes to 0x0 */ #define NFSBCOPY(s, d, l) bcopy((s), (d), (l)) #define NFSOVBCOPY(s, d, l) ovbcopy((s), (d), (l)) #define NFSBCMP(s, d, l) bcmp((s), (d), (l)) #define NFSBZERO(s, l) bzero((s), (l)) /* * Some queue.h files don't have these dfined in them. */ #ifndef LIST_END #define LIST_END(head) NULL #define SLIST_END(head) NULL #define TAILQ_END(head) NULL #endif /* * This must be defined to be a global variable that increments once * per second, but never stops or goes backwards, even when a "date" * command changes the TOD clock. It is used for delta times for * leases, etc. */ #define NFSD_MONOSEC time_uptime /* * Declare the malloc types. */ MALLOC_DECLARE(M_NEWNFSRVCACHE); MALLOC_DECLARE(M_NEWNFSDCLIENT); MALLOC_DECLARE(M_NEWNFSDSTATE); MALLOC_DECLARE(M_NEWNFSDLOCK); MALLOC_DECLARE(M_NEWNFSDLOCKFILE); MALLOC_DECLARE(M_NEWNFSSTRING); MALLOC_DECLARE(M_NEWNFSUSERGROUP); MALLOC_DECLARE(M_NEWNFSDREQ); MALLOC_DECLARE(M_NEWNFSFH); MALLOC_DECLARE(M_NEWNFSCLOWNER); MALLOC_DECLARE(M_NEWNFSCLOPEN); MALLOC_DECLARE(M_NEWNFSCLDELEG); MALLOC_DECLARE(M_NEWNFSCLCLIENT); MALLOC_DECLARE(M_NEWNFSCLLOCKOWNER); MALLOC_DECLARE(M_NEWNFSCLLOCK); MALLOC_DECLARE(M_NEWNFSDIROFF); MALLOC_DECLARE(M_NEWNFSV4NODE); MALLOC_DECLARE(M_NEWNFSDIRECTIO); MALLOC_DECLARE(M_NEWNFSMNT); MALLOC_DECLARE(M_NEWNFSDROLLBACK); MALLOC_DECLARE(M_NEWNFSLAYOUT); MALLOC_DECLARE(M_NEWNFSFLAYOUT); MALLOC_DECLARE(M_NEWNFSDEVINFO); MALLOC_DECLARE(M_NEWNFSSOCKREQ); MALLOC_DECLARE(M_NEWNFSCLDS); MALLOC_DECLARE(M_NEWNFSLAYRECALL); MALLOC_DECLARE(M_NEWNFSDSESSION); #define M_NFSRVCACHE M_NEWNFSRVCACHE #define M_NFSDCLIENT M_NEWNFSDCLIENT #define M_NFSDSTATE M_NEWNFSDSTATE #define M_NFSDLOCK M_NEWNFSDLOCK #define M_NFSDLOCKFILE M_NEWNFSDLOCKFILE #define M_NFSSTRING M_NEWNFSSTRING #define M_NFSUSERGROUP M_NEWNFSUSERGROUP #define M_NFSDREQ M_NEWNFSDREQ #define M_NFSFH M_NEWNFSFH #define M_NFSCLOWNER M_NEWNFSCLOWNER #define M_NFSCLOPEN M_NEWNFSCLOPEN #define M_NFSCLDELEG M_NEWNFSCLDELEG #define M_NFSCLCLIENT M_NEWNFSCLCLIENT #define M_NFSCLLOCKOWNER M_NEWNFSCLLOCKOWNER #define M_NFSCLLOCK M_NEWNFSCLLOCK #define M_NFSDIROFF M_NEWNFSDIROFF #define M_NFSV4NODE M_NEWNFSV4NODE #define M_NFSDIRECTIO M_NEWNFSDIRECTIO #define M_NFSDROLLBACK M_NEWNFSDROLLBACK #define M_NFSLAYOUT M_NEWNFSLAYOUT #define M_NFSFLAYOUT M_NEWNFSFLAYOUT #define M_NFSDEVINFO M_NEWNFSDEVINFO #define M_NFSSOCKREQ M_NEWNFSSOCKREQ #define M_NFSCLDS M_NEWNFSCLDS #define M_NFSLAYRECALL M_NEWNFSLAYRECALL #define M_NFSDSESSION M_NEWNFSDSESSION #define NFSINT_SIGMASK(set) \ (SIGISMEMBER(set, SIGINT) || SIGISMEMBER(set, SIGTERM) || \ SIGISMEMBER(set, SIGHUP) || SIGISMEMBER(set, SIGKILL) || \ SIGISMEMBER(set, SIGQUIT)) /* * Convert a quota block count to byte count. */ #define NFSQUOTABLKTOBYTE(q, b) (q) *= (b) /* * Define this as the largest file size supported. (It should probably * be available via a VFS_xxx Op, but it isn't. */ #define NFSRV_MAXFILESIZE ((u_int64_t)0x800000000000) /* * Set this macro to index() or strchr(), whichever is supported. */ #define STRCHR(s, c) strchr((s), (c)) /* * Set the n_time in the client write rpc, as required. */ #define NFSWRITERPC_SETTIME(w, n, a, v4) \ do { \ if (w) { \ NFSLOCKNODE(n); \ (n)->n_mtime = (a)->na_mtime; \ if (v4) \ (n)->n_change = (a)->na_filerev; \ NFSUNLOCKNODE(n); \ } \ } while (0) /* * Fake value, just to make the client work. */ #define NFS_LATTR_NOSHRINK 1 /* * Prototypes for functions where the arguments vary for different ports. */ int nfscl_loadattrcache(struct vnode **, struct nfsvattr *, void *, void *, int, int); int newnfs_realign(struct mbuf **, int); bool ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep); void ncl_copy_vattr(struct vattr *dst, struct vattr *src); /* * If the port runs on an SMP box that can enforce Atomic ops with low * overheads, define these as atomic increments/decrements. If not, * don't worry about it, since these are used for stats that can be * "out by one" without disastrous consequences. */ #define NFSINCRGLOBAL(a) ((a)++) #define NFSDECRGLOBAL(a) ((a)--) /* * Assorted funky stuff to make things work under Darwin8. */ /* * These macros checks for a field in vattr being set. */ #define NFSATTRISSET(t, v, a) ((v)->a != (t)VNOVAL) #define NFSATTRISSETTIME(v, a) ((v)->a.tv_sec != VNOVAL) /* * Manipulate mount flags. */ #define NFSSTA_HASWRITEVERF 0x00040000 /* Has write verifier */ #define NFSSTA_GOTFSINFO 0x00100000 /* Got the fsinfo */ #define NFSSTA_OPENMODE 0x00200000 /* Must use correct open mode */ #define NFSSTA_FLEXFILE 0x00800000 /* Use Flex File Layout */ #define NFSSTA_NOLAYOUTCOMMIT 0x04000000 /* Don't do LayoutCommit */ #define NFSSTA_SESSPERSIST 0x08000000 /* Has a persistent session */ #define NFSSTA_TIMEO 0x10000000 /* Experiencing a timeout */ #define NFSSTA_LOCKTIMEO 0x20000000 /* Experiencing a lockd timeout */ #define NFSSTA_HASSETFSID 0x40000000 /* Has set the fsid */ #define NFSSTA_PNFS 0x80000000 /* pNFS is enabled */ #define NFSHASNFSV3(n) ((n)->nm_flag & NFSMNT_NFSV3) #define NFSHASNFSV4(n) ((n)->nm_flag & NFSMNT_NFSV4) #define NFSHASNFSV4N(n) ((n)->nm_minorvers > 0) #define NFSHASNFSV3OR4(n) ((n)->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) #define NFSHASGOTFSINFO(n) ((n)->nm_state & NFSSTA_GOTFSINFO) #define NFSHASHASSETFSID(n) ((n)->nm_state & NFSSTA_HASSETFSID) #define NFSHASSTRICT3530(n) ((n)->nm_flag & NFSMNT_STRICT3530) #define NFSHASWRITEVERF(n) ((n)->nm_state & NFSSTA_HASWRITEVERF) #define NFSHASINT(n) ((n)->nm_flag & NFSMNT_INT) #define NFSHASSOFT(n) ((n)->nm_flag & NFSMNT_SOFT) #define NFSHASINTORSOFT(n) ((n)->nm_flag & (NFSMNT_INT | NFSMNT_SOFT)) #define NFSHASDUMBTIMR(n) ((n)->nm_flag & NFSMNT_DUMBTIMR) #define NFSHASNOCONN(n) ((n)->nm_flag & NFSMNT_MNTD) #define NFSHASKERB(n) ((n)->nm_flag & NFSMNT_KERB) #define NFSHASALLGSSNAME(n) ((n)->nm_flag & NFSMNT_ALLGSSNAME) #define NFSHASINTEGRITY(n) ((n)->nm_flag & NFSMNT_INTEGRITY) #define NFSHASPRIVACY(n) ((n)->nm_flag & NFSMNT_PRIVACY) #define NFSSETWRITEVERF(n) ((n)->nm_state |= NFSSTA_HASWRITEVERF) #define NFSSETHASSETFSID(n) ((n)->nm_state |= NFSSTA_HASSETFSID) #define NFSHASPNFSOPT(n) ((n)->nm_flag & NFSMNT_PNFS) #define NFSHASNOLAYOUTCOMMIT(n) ((n)->nm_state & NFSSTA_NOLAYOUTCOMMIT) #define NFSHASSESSPERSIST(n) ((n)->nm_state & NFSSTA_SESSPERSIST) #define NFSHASPNFS(n) ((n)->nm_state & NFSSTA_PNFS) #define NFSHASFLEXFILE(n) ((n)->nm_state & NFSSTA_FLEXFILE) #define NFSHASOPENMODE(n) ((n)->nm_state & NFSSTA_OPENMODE) #define NFSHASONEOPENOWN(n) (((n)->nm_flag & NFSMNT_ONEOPENOWN) != 0 && \ (n)->nm_minorvers > 0) #define NFSHASTLS(n) (((n)->nm_newflag & NFSMNT_TLS) != 0) /* * Set boottime. */ #define NFSSETBOOTTIME(b) (getboottime(&b)) /* * The size of directory blocks in the buffer cache. * MUST BE in the range of PAGE_SIZE <= NFS_DIRBLKSIZ <= MAXBSIZE!! */ #define NFS_DIRBLKSIZ (16 * DIRBLKSIZ) /* Must be a multiple of DIRBLKSIZ */ /* * Define these macros to access mnt_flag fields. */ #define NFSMNT_RDONLY(m) ((m)->mnt_flag & MNT_RDONLY) #endif /* _KERNEL */ /* * Define a structure similar to ufs_args for use in exporting the V4 root. */ struct nfsex_args { char *fspec; struct export_args export; }; struct nfsex_oldargs { char *fspec; struct o2export_args export; }; /* * These export flags should be defined, but there are no bits left. * Maybe a separate mnt_exflag field could be added or the mnt_flag * field increased to 64 bits? */ #ifndef MNT_EXSTRICTACCESS #define MNT_EXSTRICTACCESS 0x0 #endif #ifndef MNT_EXV4ONLY #define MNT_EXV4ONLY 0x0 #endif #ifdef _KERNEL /* * Define this to invalidate the attribute cache for the nfs node. */ #define NFSINVALATTRCACHE(n) ((n)->n_attrstamp = 0) /* Used for FreeBSD only */ void nfsd_mntinit(void); /* * Define these for vnode lock/unlock ops. * * These are good abstractions to macro out, so that they can be added to * later, for debugging or stats, etc. */ #define NFSVOPLOCK(v, f) vn_lock((v), (f)) #define NFSVOPUNLOCK(v) VOP_UNLOCK((v)) #define NFSVOPISLOCKED(v) VOP_ISLOCKED((v)) /* * Define ncl_hash(). */ #define ncl_hash(f, l) (fnv_32_buf((f), (l), FNV1_32_INIT)) int newnfs_iosize(struct nfsmount *); int newnfs_vncmpf(struct vnode *, void *); #ifndef NFS_MINDIRATTRTIMO #define NFS_MINDIRATTRTIMO 3 /* VDIR attrib cache timeout in sec */ #endif #ifndef NFS_MAXDIRATTRTIMO #define NFS_MAXDIRATTRTIMO 60 #endif /* * Nfs outstanding request list element */ struct nfsreq { TAILQ_ENTRY(nfsreq) r_chain; u_int32_t r_flags; /* flags on request, see below */ struct nfsmount *r_nmp; /* Client mnt ptr */ struct mtx r_mtx; /* Mutex lock for this structure */ }; #ifndef NFS_MAXBSIZE #define NFS_MAXBSIZE (maxbcachebuf) #endif /* * This macro checks to see if issuing of delegations is allowed for this * vnode. */ #ifdef VV_DISABLEDELEG #define NFSVNO_DELEGOK(v) \ ((v) == NULL || ((v)->v_vflag & VV_DISABLEDELEG) == 0) #else #define NFSVNO_DELEGOK(v) (1) #endif /* * Name used by getnewvnode() to describe filesystem, "nfs". * For performance reasons it is useful to have the same string * used in both places that call getnewvnode(). */ extern const char nfs_vnode_tag[]; /* * Check for the errors that indicate a DS should be disabled. * ENXIO indicates that the krpc cannot do an RPC on the DS. * EIO is returned by the RPC as an indication of I/O problems on the * server. * Are there other fatal errors? */ #define nfsds_failerr(e) ((e) == ENXIO || (e) == EIO) /* * Get a pointer to the MDS session, which is always the first element * in the list. * This macro can only be safely used when the NFSLOCKMNT() lock is held. * The inline function can be used when the lock isn't held. */ #define NFSMNT_MDSSESSION(m) (&(TAILQ_FIRST(&((m)->nm_sess))->nfsclds_sess)) static __inline struct nfsclsession * nfsmnt_mdssession(struct nfsmount *nmp) { struct nfsclsession *tsep; tsep = NULL; mtx_lock(&nmp->nm_mtx); if (TAILQ_FIRST(&nmp->nm_sess) != NULL) tsep = NFSMNT_MDSSESSION(nmp); mtx_unlock(&nmp->nm_mtx); return (tsep); } #endif /* _KERNEL */ #endif /* _NFS_NFSPORT_H */ diff --git a/sys/fs/nfsserver/nfs_fha_new.c b/sys/fs/nfsserver/nfs_fha_new.c index 1f66842da5b8..59933ef01bd4 100644 --- a/sys/fs/nfsserver/nfs_fha_new.c +++ b/sys/fs/nfsserver/nfs_fha_new.c @@ -1,710 +1,711 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2008 Isilon Inc http://www.isilon.com/ * Copyright (c) 2013 Spectra Logic Corporation * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include static MALLOC_DEFINE(M_NFS_FHA, "NFS FHA", "NFS FHA"); static void fhanew_init(void *foo); static void fhanew_uninit(void *foo); static rpcproc_t fhanew_get_procnum(rpcproc_t procnum); static int fhanew_get_fh(uint64_t *fh, int v3, struct mbuf **md, caddr_t *dpos); static int fhanew_is_read(rpcproc_t procnum); static int fhanew_is_write(rpcproc_t procnum); static int fhanew_get_offset(struct mbuf **md, caddr_t *dpos, int v3, struct fha_info *info); static int fhanew_no_offset(rpcproc_t procnum); static void fhanew_set_locktype(rpcproc_t procnum, struct fha_info *info); static int fhenew_stats_sysctl(SYSCTL_HANDLER_ARGS); static void fha_extract_info(struct svc_req *req, struct fha_info *i); static struct fha_params fhanew_softc; SYSCTL_DECL(_vfs_nfsd); extern int newnfs_nfsv3_procid[]; -extern SVCPOOL *nfsrvd_pool; + +NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool); SYSINIT(nfs_fhanew, SI_SUB_ROOT_CONF, SI_ORDER_ANY, fhanew_init, NULL); SYSUNINIT(nfs_fhanew, SI_SUB_ROOT_CONF, SI_ORDER_ANY, fhanew_uninit, NULL); static void fhanew_init(void *foo) { struct fha_params *softc; int i; softc = &fhanew_softc; bzero(softc, sizeof(*softc)); snprintf(softc->server_name, sizeof(softc->server_name), FHANEW_SERVER_NAME); - softc->pool = &nfsrvd_pool; + softc->pool = &NFSD_VNET(nfsrvd_pool); /* * Initialize the sysctl context list for the fha module. */ sysctl_ctx_init(&softc->sysctl_ctx); softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_vfs_nfsd), OID_AUTO, "fha", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "NFS File Handle Affinity (FHA)"); if (softc->sysctl_tree == NULL) { printf("%s: unable to allocate sysctl tree\n", __func__); return; } for (i = 0; i < FHA_HASH_SIZE; i++) mtx_init(&softc->fha_hash[i].mtx, "fhalock", NULL, MTX_DEF); /* * Set the default tuning parameters. */ softc->ctls.enable = FHA_DEF_ENABLE; softc->ctls.read = FHA_DEF_READ; softc->ctls.write = FHA_DEF_WRITE; softc->ctls.bin_shift = FHA_DEF_BIN_SHIFT; softc->ctls.max_nfsds_per_fh = FHA_DEF_MAX_NFSDS_PER_FH; softc->ctls.max_reqs_per_nfsd = FHA_DEF_MAX_REQS_PER_NFSD; /* * Add sysctls so the user can change the tuning parameters. */ SYSCTL_ADD_UINT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "enable", CTLFLAG_RWTUN, &softc->ctls.enable, 0, "Enable NFS File Handle Affinity (FHA)"); SYSCTL_ADD_UINT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "read", CTLFLAG_RWTUN, &softc->ctls.read, 0, "Enable NFS FHA read locality"); SYSCTL_ADD_UINT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "write", CTLFLAG_RWTUN, &softc->ctls.write, 0, "Enable NFS FHA write locality"); SYSCTL_ADD_UINT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "bin_shift", CTLFLAG_RWTUN, &softc->ctls.bin_shift, 0, "Maximum locality distance 2^(bin_shift) bytes"); SYSCTL_ADD_UINT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "max_nfsds_per_fh", CTLFLAG_RWTUN, &softc->ctls.max_nfsds_per_fh, 0, "Maximum nfsd threads that " "should be working on requests for the same file handle"); SYSCTL_ADD_UINT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "max_reqs_per_nfsd", CTLFLAG_RWTUN, &softc->ctls.max_reqs_per_nfsd, 0, "Maximum requests that " "single nfsd thread should be working on at any time"); SYSCTL_ADD_OID(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "fhe_stats", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, fhenew_stats_sysctl, "A", ""); } static void fhanew_uninit(void *foo) { struct fha_params *softc; int i; softc = &fhanew_softc; sysctl_ctx_free(&softc->sysctl_ctx); for (i = 0; i < FHA_HASH_SIZE; i++) mtx_destroy(&softc->fha_hash[i].mtx); } static rpcproc_t fhanew_get_procnum(rpcproc_t procnum) { if (procnum > NFSV2PROC_STATFS) return (-1); return (newnfs_nfsv3_procid[procnum]); } static int fhanew_get_fh(uint64_t *fh, int v3, struct mbuf **md, caddr_t *dpos) { struct nfsrv_descript lnd, *nd; uint32_t *tl; uint8_t *buf; uint64_t t; int error, len, i; error = 0; len = 0; nd = &lnd; nd->nd_md = *md; nd->nd_dpos = *dpos; if (v3) { NFSM_DISSECT_NONBLOCK(tl, uint32_t *, NFSX_UNSIGNED); if ((len = fxdr_unsigned(int, *tl)) <= 0 || len > NFSX_FHMAX) { error = EBADRPC; goto nfsmout; } } else { len = NFSX_V2FH; } t = 0; if (len != 0) { NFSM_DISSECT_NONBLOCK(buf, uint8_t *, len); for (i = 0; i < len; i++) t ^= ((uint64_t)buf[i] << (i & 7) * 8); } *fh = t; nfsmout: *md = nd->nd_md; *dpos = nd->nd_dpos; return (error); } static int fhanew_is_read(rpcproc_t procnum) { if (procnum == NFSPROC_READ) return (1); else return (0); } static int fhanew_is_write(rpcproc_t procnum) { if (procnum == NFSPROC_WRITE) return (1); else return (0); } static int fhanew_get_offset(struct mbuf **md, caddr_t *dpos, int v3, struct fha_info *info) { struct nfsrv_descript lnd, *nd; uint32_t *tl; int error; error = 0; nd = &lnd; nd->nd_md = *md; nd->nd_dpos = *dpos; if (v3) { NFSM_DISSECT_NONBLOCK(tl, uint32_t *, 2 * NFSX_UNSIGNED); info->offset = fxdr_hyper(tl); } else { NFSM_DISSECT_NONBLOCK(tl, uint32_t *, NFSX_UNSIGNED); info->offset = fxdr_unsigned(uint32_t, *tl); } nfsmout: *md = nd->nd_md; *dpos = nd->nd_dpos; return (error); } static int fhanew_no_offset(rpcproc_t procnum) { if (procnum == NFSPROC_FSSTAT || procnum == NFSPROC_FSINFO || procnum == NFSPROC_PATHCONF || procnum == NFSPROC_NOOP || procnum == NFSPROC_NULL) return (1); else return (0); } static void fhanew_set_locktype(rpcproc_t procnum, struct fha_info *info) { switch (procnum) { case NFSPROC_NULL: case NFSPROC_GETATTR: case NFSPROC_LOOKUP: case NFSPROC_ACCESS: case NFSPROC_READLINK: case NFSPROC_READ: case NFSPROC_READDIR: case NFSPROC_READDIRPLUS: case NFSPROC_WRITE: info->locktype = LK_SHARED; break; case NFSPROC_SETATTR: case NFSPROC_CREATE: case NFSPROC_MKDIR: case NFSPROC_SYMLINK: case NFSPROC_MKNOD: case NFSPROC_REMOVE: case NFSPROC_RMDIR: case NFSPROC_RENAME: case NFSPROC_LINK: case NFSPROC_FSSTAT: case NFSPROC_FSINFO: case NFSPROC_PATHCONF: case NFSPROC_COMMIT: case NFSPROC_NOOP: info->locktype = LK_EXCLUSIVE; break; } } /* * This just specifies that offsets should obey affinity when within * the same 1Mbyte (1<<20) chunk for the file (reads only for now). */ static void fha_extract_info(struct svc_req *req, struct fha_info *i) { struct mbuf *md; caddr_t dpos; static u_int64_t random_fh = 0; int error; int v3 = (req->rq_vers == 3); rpcproc_t procnum; /* * We start off with a random fh. If we get a reasonable * procnum, we set the fh. If there's a concept of offset * that we're interested in, we set that. */ i->fh = ++random_fh; i->offset = 0; i->locktype = LK_EXCLUSIVE; i->read = i->write = 0; /* * Extract the procnum and convert to v3 form if necessary, * taking care to deal with out-of-range procnums. Caller will * ensure that rq_vers is either 2 or 3. */ procnum = req->rq_proc; if (!v3) { rpcproc_t tmp_procnum; tmp_procnum = fhanew_get_procnum(procnum); if (tmp_procnum == -1) goto out; procnum = tmp_procnum; } /* * We do affinity for most. However, we divide a realm of affinity * by file offset so as to allow for concurrent random access. We * only do this for reads today, but this may change when IFS supports * efficient concurrent writes. */ if (fhanew_no_offset(procnum)) goto out; i->read = fhanew_is_read(procnum); i->write = fhanew_is_write(procnum); error = newnfs_realign(&req->rq_args, M_NOWAIT); if (error) goto out; md = req->rq_args; dpos = mtod(md, caddr_t); /* Grab the filehandle. */ error = fhanew_get_fh(&i->fh, v3, &md, &dpos); if (error) goto out; /* Content ourselves with zero offset for all but reads. */ if (i->read || i->write) fhanew_get_offset(&md, &dpos, v3, i); out: fhanew_set_locktype(procnum, i); } static struct fha_hash_entry * fha_hash_entry_new(u_int64_t fh) { struct fha_hash_entry *e; e = malloc(sizeof(*e), M_NFS_FHA, M_WAITOK); e->fh = fh; e->num_rw = 0; e->num_exclusive = 0; e->num_threads = 0; LIST_INIT(&e->threads); return (e); } static void fha_hash_entry_destroy(struct fha_hash_entry *e) { mtx_assert(e->mtx, MA_OWNED); KASSERT(e->num_rw == 0, ("%d reqs on destroyed fhe %p", e->num_rw, e)); KASSERT(e->num_exclusive == 0, ("%d exclusive reqs on destroyed fhe %p", e->num_exclusive, e)); KASSERT(e->num_threads == 0, ("%d threads on destroyed fhe %p", e->num_threads, e)); free(e, M_NFS_FHA); } static void fha_hash_entry_remove(struct fha_hash_entry *e) { mtx_assert(e->mtx, MA_OWNED); LIST_REMOVE(e, link); fha_hash_entry_destroy(e); } static struct fha_hash_entry * fha_hash_entry_lookup(struct fha_params *softc, u_int64_t fh) { struct fha_hash_slot *fhs; struct fha_hash_entry *fhe, *new_fhe; fhs = &softc->fha_hash[fh % FHA_HASH_SIZE]; new_fhe = fha_hash_entry_new(fh); new_fhe->mtx = &fhs->mtx; mtx_lock(&fhs->mtx); LIST_FOREACH(fhe, &fhs->list, link) if (fhe->fh == fh) break; if (!fhe) { fhe = new_fhe; LIST_INSERT_HEAD(&fhs->list, fhe, link); } else fha_hash_entry_destroy(new_fhe); return (fhe); } static void fha_hash_entry_add_thread(struct fha_hash_entry *fhe, SVCTHREAD *thread) { mtx_assert(fhe->mtx, MA_OWNED); thread->st_p2 = 0; LIST_INSERT_HEAD(&fhe->threads, thread, st_alink); fhe->num_threads++; } static void fha_hash_entry_remove_thread(struct fha_hash_entry *fhe, SVCTHREAD *thread) { mtx_assert(fhe->mtx, MA_OWNED); KASSERT(thread->st_p2 == 0, ("%d reqs on removed thread %p", thread->st_p2, thread)); LIST_REMOVE(thread, st_alink); fhe->num_threads--; } /* * Account for an ongoing operation associated with this file. */ static void fha_hash_entry_add_op(struct fha_hash_entry *fhe, int locktype, int count) { mtx_assert(fhe->mtx, MA_OWNED); if (LK_EXCLUSIVE == locktype) fhe->num_exclusive += count; else fhe->num_rw += count; } /* * Get the service thread currently associated with the fhe that is * appropriate to handle this operation. */ static SVCTHREAD * fha_hash_entry_choose_thread(struct fha_params *softc, struct fha_hash_entry *fhe, struct fha_info *i, SVCTHREAD *this_thread) { SVCTHREAD *thread, *min_thread = NULL; int req_count, min_count = 0; off_t offset1, offset2; LIST_FOREACH(thread, &fhe->threads, st_alink) { req_count = thread->st_p2; /* If there are any writes in progress, use the first thread. */ if (fhe->num_exclusive) { #if 0 ITRACE_CURPROC(ITRACE_NFS, ITRACE_INFO, "fha: %p(%d)w", thread, req_count); #endif return (thread); } /* Check whether we should consider locality. */ if ((i->read && !softc->ctls.read) || (i->write && !softc->ctls.write)) goto noloc; /* * Check for locality, making sure that we won't * exceed our per-thread load limit in the process. */ offset1 = i->offset; offset2 = thread->st_p3; if (((offset1 >= offset2) && ((offset1 - offset2) < (1 << softc->ctls.bin_shift))) || ((offset2 > offset1) && ((offset2 - offset1) < (1 << softc->ctls.bin_shift)))) { if ((softc->ctls.max_reqs_per_nfsd == 0) || (req_count < softc->ctls.max_reqs_per_nfsd)) { #if 0 ITRACE_CURPROC(ITRACE_NFS, ITRACE_INFO, "fha: %p(%d)r", thread, req_count); #endif return (thread); } } noloc: /* * We don't have a locality match, so skip this thread, * but keep track of the most attractive thread in case * we need to come back to it later. */ #if 0 ITRACE_CURPROC(ITRACE_NFS, ITRACE_INFO, "fha: %p(%d)s off1 %llu off2 %llu", thread, req_count, offset1, offset2); #endif if ((min_thread == NULL) || (req_count < min_count)) { min_count = req_count; min_thread = thread; } } /* * We didn't find a good match yet. See if we can add * a new thread to this file handle entry's thread list. */ if ((softc->ctls.max_nfsds_per_fh == 0) || (fhe->num_threads < softc->ctls.max_nfsds_per_fh)) { thread = this_thread; #if 0 ITRACE_CURPROC(ITRACE_NFS, ITRACE_INFO, "fha: %p(%d)t", thread, thread->st_p2); #endif fha_hash_entry_add_thread(fhe, thread); } else { /* * We don't want to use any more threads for this file, so * go back to the most attractive nfsd we're already using. */ thread = min_thread; } return (thread); } /* * After getting a request, try to assign it to some thread. Usually we * handle it ourselves. */ SVCTHREAD * fhanew_assign(SVCTHREAD *this_thread, struct svc_req *req) { struct fha_params *softc = &fhanew_softc; SVCTHREAD *thread; struct fha_info i; struct fha_hash_entry *fhe; /* Check to see whether we're enabled. */ if (softc->ctls.enable == 0) goto thist; /* * Only do placement if this is an NFS request. */ if (req->rq_prog != NFS_PROG) goto thist; if (req->rq_vers != 2 && req->rq_vers != 3) goto thist; fha_extract_info(req, &i); /* * We save the offset associated with this request for later * nfsd matching. */ fhe = fha_hash_entry_lookup(softc, i.fh); req->rq_p1 = fhe; req->rq_p2 = i.locktype; req->rq_p3 = i.offset; /* * Choose a thread, taking into consideration locality, thread load, * and the number of threads already working on this file. */ thread = fha_hash_entry_choose_thread(softc, fhe, &i, this_thread); KASSERT(thread, ("fha_assign: NULL thread!")); fha_hash_entry_add_op(fhe, i.locktype, 1); thread->st_p2++; thread->st_p3 = i.offset; /* * Grab the pool lock here to not let chosen thread go away before * the new request inserted to its queue while we drop fhe lock. */ mtx_lock(&thread->st_lock); mtx_unlock(fhe->mtx); return (thread); thist: req->rq_p1 = NULL; mtx_lock(&this_thread->st_lock); return (this_thread); } /* * Called when we're done with an operation. The request has already * been de-queued. */ void fhanew_nd_complete(SVCTHREAD *thread, struct svc_req *req) { struct fha_hash_entry *fhe = req->rq_p1; struct mtx *mtx; /* * This may be called for reqs that didn't go through * fha_assign (e.g. extra NULL ops used for RPCSEC_GSS. */ if (!fhe) return; mtx = fhe->mtx; mtx_lock(mtx); fha_hash_entry_add_op(fhe, req->rq_p2, -1); thread->st_p2--; KASSERT(thread->st_p2 >= 0, ("Negative request count %d on %p", thread->st_p2, thread)); if (thread->st_p2 == 0) { fha_hash_entry_remove_thread(fhe, thread); if (0 == fhe->num_rw + fhe->num_exclusive) fha_hash_entry_remove(fhe); } mtx_unlock(mtx); } static int fhenew_stats_sysctl(SYSCTL_HANDLER_ARGS) { struct fha_params *softc = &fhanew_softc; int error, i; struct sbuf sb; struct fha_hash_entry *fhe; bool_t first, hfirst; SVCTHREAD *thread; sbuf_new(&sb, NULL, 65536, SBUF_FIXEDLEN); if (!*softc->pool) { sbuf_printf(&sb, "NFSD not running\n"); goto out; } for (i = 0; i < FHA_HASH_SIZE; i++) if (!LIST_EMPTY(&softc->fha_hash[i].list)) break; if (i == FHA_HASH_SIZE) { sbuf_printf(&sb, "No file handle entries.\n"); goto out; } hfirst = TRUE; for (; i < FHA_HASH_SIZE; i++) { mtx_lock(&softc->fha_hash[i].mtx); if (LIST_EMPTY(&softc->fha_hash[i].list)) { mtx_unlock(&softc->fha_hash[i].mtx); continue; } sbuf_printf(&sb, "%shash %d: {\n", hfirst ? "" : ", ", i); first = TRUE; LIST_FOREACH(fhe, &softc->fha_hash[i].list, link) { sbuf_printf(&sb, "%sfhe %p: {\n", first ? " " : ", ", fhe); sbuf_printf(&sb, " fh: %ju\n", (uintmax_t) fhe->fh); sbuf_printf(&sb, " num_rw/exclusive: %d/%d\n", fhe->num_rw, fhe->num_exclusive); sbuf_printf(&sb, " num_threads: %d\n", fhe->num_threads); LIST_FOREACH(thread, &fhe->threads, st_alink) { sbuf_printf(&sb, " thread %p offset %ju " "reqs %d\n", thread, thread->st_p3, thread->st_p2); } sbuf_printf(&sb, " }"); first = FALSE; } sbuf_printf(&sb, "\n}"); mtx_unlock(&softc->fha_hash[i].mtx); hfirst = FALSE; } out: sbuf_trim(&sb); sbuf_finish(&sb); error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req); sbuf_delete(&sb); return (error); } diff --git a/sys/fs/nfsserver/nfs_nfsdcache.c b/sys/fs/nfsserver/nfs_nfsdcache.c index 5efcb90411e0..e94d58481286 100644 --- a/sys/fs/nfsserver/nfs_nfsdcache.c +++ b/sys/fs/nfsserver/nfs_nfsdcache.c @@ -1,1036 +1,1053 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); /* * Here is the basic algorithm: * First, some design criteria I used: * - I think a false hit is more serious than a false miss * - A false hit for an RPC that has Op(s) that order via seqid# must be * avoided at all cost * - A valid hit will probably happen a long time after the original reply * and the TCP socket that the original request was received on will no * longer be active * (The long time delay implies to me that LRU is not appropriate.) * - The mechanism will satisfy the requirements of ordering Ops with seqid#s * in them as well as minimizing the risk of redoing retried non-idempotent * Ops. * Because it is biased towards avoiding false hits, multiple entries with * the same xid are to be expected, especially for the case of the entry * in the cache being related to a seqid# sequenced Op. * * The basic algorithm I'm about to code up: * - Null RPCs bypass the cache and are just done * For TCP * - key on (as noted above, there can be several * entries with the same key) * When a request arrives: * For all that match key * - if RPC# != OR request_size != * - not a match with this one * - if NFSv4 and received on same TCP socket OR * received on a TCP connection created before the * entry was cached * - not a match with this one * (V2,3 clients might retry on same TCP socket) * - calculate checksum on first N bytes of NFS XDR * - if checksum != * - not a match for this one * If any of the remaining ones that match has a * seqid_refcnt > 0 * - not a match (go do RPC, using new cache entry) * If one match left * - a hit (reply from cache) * else * - miss (go do RPC, using new cache entry) * * During processing of NFSv4 request: * - set a flag when a non-idempotent Op is processed * - when an Op that uses a seqid# (Open,...) is processed * - if same seqid# as referenced entry in cache * - free new cache entry * - reply from referenced cache entry * else if next seqid# in order * - free referenced cache entry * - increment seqid_refcnt on new cache entry * - set pointer from Openowner/Lockowner to * new cache entry (aka reference it) * else if first seqid# in sequence * - increment seqid_refcnt on new cache entry * - set pointer from Openowner/Lockowner to * new cache entry (aka reference it) * * At end of RPC processing: * - if seqid_refcnt > 0 OR flagged non-idempotent on new * cache entry * - save reply in cache entry * - calculate checksum on first N bytes of NFS XDR * request * - note op and length of XDR request (in bytes) * - timestamp it * else * - free new cache entry * - Send reply (noting info for socket activity check, below) * * For cache entries saved above: * - if saved since seqid_refcnt was > 0 * - free when seqid_refcnt decrements to 0 * (when next one in sequence is processed above, or * when Openowner/Lockowner is discarded) * else { non-idempotent Op(s) } * - free when * - some further activity observed on same * socket * (I'm not yet sure how I'm going to do * this. Maybe look at the TCP connection * to see if the send_tcp_sequence# is well * past sent reply OR K additional RPCs * replied on same socket OR?) * OR * - when very old (hours, days, weeks?) * * For UDP (v2, 3 only), pretty much the old way: * - key on * (at most one entry for each key) * * When a Request arrives: * - if a match with entry via key * - if RPC marked In_progress * - discard request (don't send reply) * else * - reply from cache * - timestamp cache entry * else * - add entry to cache, marked In_progress * - do RPC * - when RPC done * - if RPC# non-idempotent * - mark entry Done (not In_progress) * - save reply * - timestamp cache entry * else * - free cache entry * - send reply * * Later, entries with saved replies are free'd a short time (few minutes) * after reply sent (timestamp). * Reference: Chet Juszczak, "Improving the Performance and Correctness * of an NFS Server", in Proc. Winter 1989 USENIX Conference, * pages 53-63. San Diego, February 1989. * for the UDP case. * nfsrc_floodlevel is set to the allowable upper limit for saved replies * for TCP. For V3, a reply won't be saved when the flood level is * hit. For V4, the non-idempotent Op will return NFSERR_RESOURCE in * that case. This level should be set high enough that this almost * never happens. */ #include -extern struct nfsstatsv1 nfsstatsv1; extern struct mtx nfsrc_udpmtx; -extern struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; -extern struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; -int nfsrc_floodlevel = NFSRVCACHE_FLOODLEVEL, nfsrc_tcpsavedreplies = 0; + +NFSD_VNET_DECLARE(struct nfsrvhashhead *, nfsrvudphashtbl); +NFSD_VNET_DECLARE(struct nfsrchash_bucket *, nfsrchash_table); +NFSD_VNET_DECLARE(struct nfsrchash_bucket *, nfsrcahash_table); +NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); + +NFSD_VNET_DEFINE(int, nfsrc_floodlevel) = NFSRVCACHE_FLOODLEVEL; +NFSD_VNET_DEFINE(int, nfsrc_tcpsavedreplies) = 0; SYSCTL_DECL(_vfs_nfsd); static u_int nfsrc_tcphighwater = 0; static int sysctl_tcphighwater(SYSCTL_HANDLER_ARGS) { int error, newhighwater; newhighwater = nfsrc_tcphighwater; error = sysctl_handle_int(oidp, &newhighwater, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (newhighwater < 0) return (EINVAL); - if (newhighwater >= nfsrc_floodlevel) - nfsrc_floodlevel = newhighwater + newhighwater / 5; + if (newhighwater >= NFSD_VNET(nfsrc_floodlevel)) + NFSD_VNET(nfsrc_floodlevel) = newhighwater + newhighwater / 5; nfsrc_tcphighwater = newhighwater; return (0); } SYSCTL_PROC(_vfs_nfsd, OID_AUTO, tcphighwater, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrc_tcphighwater), sysctl_tcphighwater, "IU", "High water mark for TCP cache entries"); static u_int nfsrc_udphighwater = NFSRVCACHE_UDPHIGHWATER; SYSCTL_UINT(_vfs_nfsd, OID_AUTO, udphighwater, CTLFLAG_RW, &nfsrc_udphighwater, 0, "High water mark for UDP cache entries"); static u_int nfsrc_tcptimeout = NFSRVCACHE_TCPTIMEOUT; SYSCTL_UINT(_vfs_nfsd, OID_AUTO, tcpcachetimeo, CTLFLAG_RW, &nfsrc_tcptimeout, 0, "Timeout for TCP entries in the DRC"); static u_int nfsrc_tcpnonidempotent = 1; SYSCTL_UINT(_vfs_nfsd, OID_AUTO, cachetcp, CTLFLAG_RW, &nfsrc_tcpnonidempotent, 0, "Enable the DRC for NFS over TCP"); -static int nfsrc_udpcachesize = 0; -static TAILQ_HEAD(, nfsrvcache) nfsrvudplru; -static struct nfsrvhashhead nfsrvudphashtbl[NFSRVCACHE_HASHSIZE]; +NFSD_VNET_DEFINE_STATIC(int, nfsrc_udpcachesize) = 0; +NFSD_VNET_DEFINE_STATIC(TAILQ_HEAD(, nfsrvcache), nfsrvudplru); /* * and the reverse mapping from generic to Version 2 procedure numbers */ static int newnfsv2_procid[NFS_V3NPROCS] = { NFSV2PROC_NULL, NFSV2PROC_GETATTR, NFSV2PROC_SETATTR, NFSV2PROC_LOOKUP, NFSV2PROC_NOOP, NFSV2PROC_READLINK, NFSV2PROC_READ, NFSV2PROC_WRITE, NFSV2PROC_CREATE, NFSV2PROC_MKDIR, NFSV2PROC_SYMLINK, NFSV2PROC_CREATE, NFSV2PROC_REMOVE, NFSV2PROC_RMDIR, NFSV2PROC_RENAME, NFSV2PROC_LINK, NFSV2PROC_READDIR, NFSV2PROC_NOOP, NFSV2PROC_STATFS, NFSV2PROC_NOOP, NFSV2PROC_NOOP, NFSV2PROC_NOOP, }; #define nfsrc_hash(xid) (((xid) + ((xid) >> 24)) % NFSRVCACHE_HASHSIZE) #define NFSRCUDPHASH(xid) \ - (&nfsrvudphashtbl[nfsrc_hash(xid)]) + (&NFSD_VNET(nfsrvudphashtbl)[nfsrc_hash(xid)]) #define NFSRCHASH(xid) \ - (&nfsrchash_table[nfsrc_hash(xid)].tbl) -#define NFSRCAHASH(xid) (&nfsrcahash_table[nfsrc_hash(xid)]) + (&NFSD_VNET(nfsrchash_table)[nfsrc_hash(xid)].tbl) +#define NFSRCAHASH(xid) (&NFSD_VNET(nfsrcahash_table)[nfsrc_hash(xid)]) #define TRUE 1 #define FALSE 0 #define NFSRVCACHE_CHECKLEN 100 /* True iff the rpc reply is an nfs status ONLY! */ static int nfsv2_repstat[NFS_V3NPROCS] = { FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, TRUE, TRUE, FALSE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, }; /* * Will NFS want to work over IPv6 someday? */ #define NETFAMILY(rp) \ (((rp)->rc_flag & RC_INETIPV6) ? AF_INET6 : AF_INET) /* local functions */ static int nfsrc_getudp(struct nfsrv_descript *nd, struct nfsrvcache *newrp); static int nfsrc_gettcp(struct nfsrv_descript *nd, struct nfsrvcache *newrp); static void nfsrc_lock(struct nfsrvcache *rp); static void nfsrc_unlock(struct nfsrvcache *rp); static void nfsrc_wanted(struct nfsrvcache *rp); static void nfsrc_freecache(struct nfsrvcache *rp); static int nfsrc_getlenandcksum(struct mbuf *m1, u_int16_t *cksum); static void nfsrc_marksametcpconn(u_int64_t); /* * Return the correct mutex for this cache entry. */ static __inline struct mtx * nfsrc_cachemutex(struct nfsrvcache *rp) { if ((rp->rc_flag & RC_UDP) != 0) return (&nfsrc_udpmtx); - return (&nfsrchash_table[nfsrc_hash(rp->rc_xid)].mtx); + return (&NFSD_VNET(nfsrchash_table)[nfsrc_hash(rp->rc_xid)].mtx); } /* * Initialize the server request cache list */ void nfsrvd_initcache(void) { int i; - static int inited = 0; - if (inited) - return; - inited = 1; + NFSD_VNET(nfsrvudphashtbl) = malloc(sizeof(struct nfsrvhashhead) * + NFSRVCACHE_HASHSIZE, M_NFSRVCACHE, M_WAITOK | M_ZERO); + NFSD_VNET(nfsrchash_table) = malloc(sizeof(struct nfsrchash_bucket) * + NFSRVCACHE_HASHSIZE, M_NFSRVCACHE, M_WAITOK | M_ZERO); + NFSD_VNET(nfsrcahash_table) = malloc(sizeof(struct nfsrchash_bucket) * + NFSRVCACHE_HASHSIZE, M_NFSRVCACHE, M_WAITOK | M_ZERO); + for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { + mtx_init(&NFSD_VNET(nfsrchash_table)[i].mtx, "nfsrtc", NULL, + MTX_DEF); + mtx_init(&NFSD_VNET(nfsrcahash_table)[i].mtx, "nfsrtca", NULL, + MTX_DEF); + } for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { - LIST_INIT(&nfsrvudphashtbl[i]); - LIST_INIT(&nfsrchash_table[i].tbl); - LIST_INIT(&nfsrcahash_table[i].tbl); + LIST_INIT(&NFSD_VNET(nfsrvudphashtbl)[i]); + LIST_INIT(&NFSD_VNET(nfsrchash_table)[i].tbl); + LIST_INIT(&NFSD_VNET(nfsrcahash_table)[i].tbl); } - TAILQ_INIT(&nfsrvudplru); - nfsrc_tcpsavedreplies = 0; - nfsrc_udpcachesize = 0; - nfsstatsv1.srvcache_tcppeak = 0; - nfsstatsv1.srvcache_size = 0; + TAILQ_INIT(&NFSD_VNET(nfsrvudplru)); + NFSD_VNET(nfsrc_tcpsavedreplies) = 0; + NFSD_VNET(nfsrc_udpcachesize) = 0; } /* * Get a cache entry for this request. Basically just malloc a new one * and then call nfsrc_getudp() or nfsrc_gettcp() to do the rest. */ int nfsrvd_getcache(struct nfsrv_descript *nd) { struct nfsrvcache *newrp; int ret; if (nd->nd_procnum == NFSPROC_NULL) panic("nfsd cache null"); newrp = malloc(sizeof (struct nfsrvcache), M_NFSRVCACHE, M_WAITOK); NFSBZERO((caddr_t)newrp, sizeof (struct nfsrvcache)); if (nd->nd_flag & ND_NFSV4) newrp->rc_flag = RC_NFSV4; else if (nd->nd_flag & ND_NFSV3) newrp->rc_flag = RC_NFSV3; else newrp->rc_flag = RC_NFSV2; newrp->rc_xid = nd->nd_retxid; newrp->rc_proc = nd->nd_procnum; newrp->rc_sockref = nd->nd_sockref; newrp->rc_cachetime = nd->nd_tcpconntime; if (nd->nd_flag & ND_SAMETCPCONN) newrp->rc_flag |= RC_SAMETCPCONN; if (nd->nd_nam2 != NULL) { newrp->rc_flag |= RC_UDP; ret = nfsrc_getudp(nd, newrp); } else { ret = nfsrc_gettcp(nd, newrp); } NFSEXITCODE2(0, nd); return (ret); } /* * For UDP (v2, v3): * - key on * (at most one entry for each key) */ static int nfsrc_getudp(struct nfsrv_descript *nd, struct nfsrvcache *newrp) { struct nfsrvcache *rp; struct sockaddr_in *saddr; struct sockaddr_in6 *saddr6; struct nfsrvhashhead *hp; int ret = 0; struct mtx *mutex; mutex = nfsrc_cachemutex(newrp); hp = NFSRCUDPHASH(newrp->rc_xid); loop: mtx_lock(mutex); LIST_FOREACH(rp, hp, rc_hash) { if (newrp->rc_xid == rp->rc_xid && newrp->rc_proc == rp->rc_proc && (newrp->rc_flag & rp->rc_flag & RC_NFSVERS) && nfsaddr_match(NETFAMILY(rp), &rp->rc_haddr, nd->nd_nam)) { if ((rp->rc_flag & RC_LOCKED) != 0) { rp->rc_flag |= RC_WANTED; (void)mtx_sleep(rp, mutex, (PZERO - 1) | PDROP, "nfsrc", 10 * hz); goto loop; } if (rp->rc_flag == 0) panic("nfs udp cache0"); rp->rc_flag |= RC_LOCKED; - TAILQ_REMOVE(&nfsrvudplru, rp, rc_lru); - TAILQ_INSERT_TAIL(&nfsrvudplru, rp, rc_lru); + TAILQ_REMOVE(&NFSD_VNET(nfsrvudplru), rp, rc_lru); + TAILQ_INSERT_TAIL(&NFSD_VNET(nfsrvudplru), rp, rc_lru); if (rp->rc_flag & RC_INPROG) { - nfsstatsv1.srvcache_inproghits++; + nfsstatsv1_p->srvcache_inproghits++; mtx_unlock(mutex); ret = RC_DROPIT; } else if (rp->rc_flag & RC_REPSTATUS) { /* * V2 only. */ - nfsstatsv1.srvcache_nonidemdonehits++; + nfsstatsv1_p->srvcache_nonidemdonehits++; mtx_unlock(mutex); nfsrvd_rephead(nd); *(nd->nd_errp) = rp->rc_status; ret = RC_REPLY; rp->rc_timestamp = NFSD_MONOSEC + NFSRVCACHE_UDPTIMEOUT; } else if (rp->rc_flag & RC_REPMBUF) { - nfsstatsv1.srvcache_nonidemdonehits++; + nfsstatsv1_p->srvcache_nonidemdonehits++; mtx_unlock(mutex); nd->nd_mreq = m_copym(rp->rc_reply, 0, M_COPYALL, M_WAITOK); ret = RC_REPLY; rp->rc_timestamp = NFSD_MONOSEC + NFSRVCACHE_UDPTIMEOUT; } else { panic("nfs udp cache1"); } nfsrc_unlock(rp); free(newrp, M_NFSRVCACHE); goto out; } } - nfsstatsv1.srvcache_misses++; - atomic_add_int(&nfsstatsv1.srvcache_size, 1); - nfsrc_udpcachesize++; + nfsstatsv1_p->srvcache_misses++; + atomic_add_int(&nfsstatsv1_p->srvcache_size, 1); + NFSD_VNET(nfsrc_udpcachesize)++; newrp->rc_flag |= RC_INPROG; saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); if (saddr->sin_family == AF_INET) newrp->rc_inet = saddr->sin_addr.s_addr; else if (saddr->sin_family == AF_INET6) { saddr6 = (struct sockaddr_in6 *)saddr; NFSBCOPY((caddr_t)&saddr6->sin6_addr, (caddr_t)&newrp->rc_inet6, sizeof (struct in6_addr)); newrp->rc_flag |= RC_INETIPV6; } LIST_INSERT_HEAD(hp, newrp, rc_hash); - TAILQ_INSERT_TAIL(&nfsrvudplru, newrp, rc_lru); + TAILQ_INSERT_TAIL(&NFSD_VNET(nfsrvudplru), newrp, rc_lru); mtx_unlock(mutex); nd->nd_rp = newrp; ret = RC_DOIT; out: NFSEXITCODE2(0, nd); return (ret); } /* * Update a request cache entry after the rpc has been done */ struct nfsrvcache * nfsrvd_updatecache(struct nfsrv_descript *nd) { struct nfsrvcache *rp; struct nfsrvcache *retrp = NULL; struct mbuf *m; struct mtx *mutex; rp = nd->nd_rp; if (!rp) panic("nfsrvd_updatecache null rp"); nd->nd_rp = NULL; mutex = nfsrc_cachemutex(rp); mtx_lock(mutex); nfsrc_lock(rp); if (!(rp->rc_flag & RC_INPROG)) panic("nfsrvd_updatecache not inprog"); rp->rc_flag &= ~RC_INPROG; if (rp->rc_flag & RC_UDP) { - TAILQ_REMOVE(&nfsrvudplru, rp, rc_lru); - TAILQ_INSERT_TAIL(&nfsrvudplru, rp, rc_lru); + TAILQ_REMOVE(&NFSD_VNET(nfsrvudplru), rp, rc_lru); + TAILQ_INSERT_TAIL(&NFSD_VNET(nfsrvudplru), rp, rc_lru); } /* * Reply from cache is a special case returned by nfsrv_checkseqid(). */ if (nd->nd_repstat == NFSERR_REPLYFROMCACHE) { - nfsstatsv1.srvcache_nonidemdonehits++; + nfsstatsv1_p->srvcache_nonidemdonehits++; mtx_unlock(mutex); nd->nd_repstat = 0; if (nd->nd_mreq) m_freem(nd->nd_mreq); if (!(rp->rc_flag & RC_REPMBUF)) panic("reply from cache"); nd->nd_mreq = m_copym(rp->rc_reply, 0, M_COPYALL, M_WAITOK); rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; nfsrc_unlock(rp); goto out; } /* * If rc_refcnt > 0, save it * For UDP, save it if ND_SAVEREPLY is set * For TCP, save it if ND_SAVEREPLY and nfsrc_tcpnonidempotent is set */ if (nd->nd_repstat != NFSERR_DONTREPLY && (rp->rc_refcnt > 0 || ((nd->nd_flag & ND_SAVEREPLY) && (rp->rc_flag & RC_UDP)) || ((nd->nd_flag & ND_SAVEREPLY) && !(rp->rc_flag & RC_UDP) && - nfsrc_tcpsavedreplies <= nfsrc_floodlevel && + NFSD_VNET(nfsrc_tcpsavedreplies) <= NFSD_VNET(nfsrc_floodlevel) && nfsrc_tcpnonidempotent))) { if (rp->rc_refcnt > 0) { if (!(rp->rc_flag & RC_NFSV4)) panic("update_cache refcnt"); rp->rc_flag |= RC_REFCNT; } if ((nd->nd_flag & ND_NFSV2) && nfsv2_repstat[newnfsv2_procid[nd->nd_procnum]]) { rp->rc_status = nd->nd_repstat; rp->rc_flag |= RC_REPSTATUS; mtx_unlock(mutex); } else { if (!(rp->rc_flag & RC_UDP)) { - atomic_add_int(&nfsrc_tcpsavedreplies, 1); - if (nfsrc_tcpsavedreplies > - nfsstatsv1.srvcache_tcppeak) - nfsstatsv1.srvcache_tcppeak = - nfsrc_tcpsavedreplies; + atomic_add_int(&NFSD_VNET(nfsrc_tcpsavedreplies), + 1); + if (NFSD_VNET(nfsrc_tcpsavedreplies) > + nfsstatsv1_p->srvcache_tcppeak) + nfsstatsv1_p->srvcache_tcppeak = + NFSD_VNET(nfsrc_tcpsavedreplies); } mtx_unlock(mutex); m = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK); mtx_lock(mutex); rp->rc_reply = m; rp->rc_flag |= RC_REPMBUF; mtx_unlock(mutex); } if (rp->rc_flag & RC_UDP) { rp->rc_timestamp = NFSD_MONOSEC + NFSRVCACHE_UDPTIMEOUT; nfsrc_unlock(rp); } else { rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; if (rp->rc_refcnt > 0) nfsrc_unlock(rp); else retrp = rp; } } else { nfsrc_freecache(rp); mtx_unlock(mutex); } out: NFSEXITCODE2(0, nd); return (retrp); } /* * Invalidate and, if possible, free an in prog cache entry. * Must not sleep. */ void nfsrvd_delcache(struct nfsrvcache *rp) { struct mtx *mutex; mutex = nfsrc_cachemutex(rp); if (!(rp->rc_flag & RC_INPROG)) panic("nfsrvd_delcache not in prog"); mtx_lock(mutex); rp->rc_flag &= ~RC_INPROG; if (rp->rc_refcnt == 0 && !(rp->rc_flag & RC_LOCKED)) nfsrc_freecache(rp); mtx_unlock(mutex); } /* * Called after nfsrvd_updatecache() once the reply is sent, to update * the entry's sequence number and unlock it. The argument is * the pointer returned by nfsrvd_updatecache(). */ void nfsrvd_sentcache(struct nfsrvcache *rp, int have_seq, uint32_t seq) { struct nfsrchash_bucket *hbp; KASSERT(rp->rc_flag & RC_LOCKED, ("nfsrvd_sentcache not locked")); if (have_seq) { hbp = NFSRCAHASH(rp->rc_sockref); mtx_lock(&hbp->mtx); rp->rc_tcpseq = seq; if (rp->rc_acked != RC_NO_ACK) LIST_INSERT_HEAD(&hbp->tbl, rp, rc_ahash); rp->rc_acked = RC_NO_ACK; mtx_unlock(&hbp->mtx); } nfsrc_unlock(rp); } /* * Get a cache entry for TCP * - key on * (allow multiple entries for a given key) */ static int nfsrc_gettcp(struct nfsrv_descript *nd, struct nfsrvcache *newrp) { struct nfsrvcache *rp, *nextrp; int i; struct nfsrvcache *hitrp; struct nfsrvhashhead *hp, nfsrc_templist; int hit, ret = 0; struct mtx *mutex; mutex = nfsrc_cachemutex(newrp); hp = NFSRCHASH(newrp->rc_xid); newrp->rc_reqlen = nfsrc_getlenandcksum(nd->nd_mrep, &newrp->rc_cksum); tryagain: mtx_lock(mutex); hit = 1; LIST_INIT(&nfsrc_templist); /* * Get all the matches and put them on the temp list. */ rp = LIST_FIRST(hp); while (rp != LIST_END(hp)) { nextrp = LIST_NEXT(rp, rc_hash); if (newrp->rc_xid == rp->rc_xid && (!(rp->rc_flag & RC_INPROG) || ((newrp->rc_flag & RC_SAMETCPCONN) && newrp->rc_sockref == rp->rc_sockref)) && (newrp->rc_flag & rp->rc_flag & RC_NFSVERS) && newrp->rc_proc == rp->rc_proc && ((newrp->rc_flag & RC_NFSV4) && newrp->rc_sockref != rp->rc_sockref && newrp->rc_cachetime >= rp->rc_cachetime) && newrp->rc_reqlen == rp->rc_reqlen && newrp->rc_cksum == rp->rc_cksum) { LIST_REMOVE(rp, rc_hash); LIST_INSERT_HEAD(&nfsrc_templist, rp, rc_hash); } rp = nextrp; } /* * Now, use nfsrc_templist to decide if there is a match. */ i = 0; LIST_FOREACH(rp, &nfsrc_templist, rc_hash) { i++; if (rp->rc_refcnt > 0) { hit = 0; break; } } /* * Can be a hit only if one entry left. * Note possible hit entry and put nfsrc_templist back on hash * list. */ if (i != 1) hit = 0; hitrp = rp = LIST_FIRST(&nfsrc_templist); while (rp != LIST_END(&nfsrc_templist)) { nextrp = LIST_NEXT(rp, rc_hash); LIST_REMOVE(rp, rc_hash); LIST_INSERT_HEAD(hp, rp, rc_hash); rp = nextrp; } if (LIST_FIRST(&nfsrc_templist) != LIST_END(&nfsrc_templist)) panic("nfs gettcp cache templist"); if (hit) { rp = hitrp; if ((rp->rc_flag & RC_LOCKED) != 0) { rp->rc_flag |= RC_WANTED; (void)mtx_sleep(rp, mutex, (PZERO - 1) | PDROP, "nfsrc", 10 * hz); goto tryagain; } if (rp->rc_flag == 0) panic("nfs tcp cache0"); rp->rc_flag |= RC_LOCKED; if (rp->rc_flag & RC_INPROG) { - nfsstatsv1.srvcache_inproghits++; + nfsstatsv1_p->srvcache_inproghits++; mtx_unlock(mutex); if (newrp->rc_sockref == rp->rc_sockref) nfsrc_marksametcpconn(rp->rc_sockref); ret = RC_DROPIT; } else if (rp->rc_flag & RC_REPSTATUS) { /* * V2 only. */ - nfsstatsv1.srvcache_nonidemdonehits++; + nfsstatsv1_p->srvcache_nonidemdonehits++; mtx_unlock(mutex); if (newrp->rc_sockref == rp->rc_sockref) nfsrc_marksametcpconn(rp->rc_sockref); ret = RC_REPLY; nfsrvd_rephead(nd); *(nd->nd_errp) = rp->rc_status; rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; } else if (rp->rc_flag & RC_REPMBUF) { - nfsstatsv1.srvcache_nonidemdonehits++; + nfsstatsv1_p->srvcache_nonidemdonehits++; mtx_unlock(mutex); if (newrp->rc_sockref == rp->rc_sockref) nfsrc_marksametcpconn(rp->rc_sockref); ret = RC_REPLY; nd->nd_mreq = m_copym(rp->rc_reply, 0, M_COPYALL, M_WAITOK); rp->rc_timestamp = NFSD_MONOSEC + nfsrc_tcptimeout; } else { panic("nfs tcp cache1"); } nfsrc_unlock(rp); free(newrp, M_NFSRVCACHE); goto out; } - nfsstatsv1.srvcache_misses++; - atomic_add_int(&nfsstatsv1.srvcache_size, 1); + nfsstatsv1_p->srvcache_misses++; + atomic_add_int(&nfsstatsv1_p->srvcache_size, 1); /* * For TCP, multiple entries for a key are allowed, so don't * chain it into the hash table until done. */ newrp->rc_cachetime = NFSD_MONOSEC; newrp->rc_flag |= RC_INPROG; LIST_INSERT_HEAD(hp, newrp, rc_hash); mtx_unlock(mutex); nd->nd_rp = newrp; ret = RC_DOIT; out: NFSEXITCODE2(0, nd); return (ret); } /* * Lock a cache entry. */ static void nfsrc_lock(struct nfsrvcache *rp) { struct mtx *mutex; mutex = nfsrc_cachemutex(rp); mtx_assert(mutex, MA_OWNED); while ((rp->rc_flag & RC_LOCKED) != 0) { rp->rc_flag |= RC_WANTED; (void)mtx_sleep(rp, mutex, PZERO - 1, "nfsrc", 0); } rp->rc_flag |= RC_LOCKED; } /* * Unlock a cache entry. */ static void nfsrc_unlock(struct nfsrvcache *rp) { struct mtx *mutex; mutex = nfsrc_cachemutex(rp); mtx_lock(mutex); rp->rc_flag &= ~RC_LOCKED; nfsrc_wanted(rp); mtx_unlock(mutex); } /* * Wakeup anyone wanting entry. */ static void nfsrc_wanted(struct nfsrvcache *rp) { if (rp->rc_flag & RC_WANTED) { rp->rc_flag &= ~RC_WANTED; wakeup((caddr_t)rp); } } /* * Free up the entry. * Must not sleep. */ static void nfsrc_freecache(struct nfsrvcache *rp) { struct nfsrchash_bucket *hbp; LIST_REMOVE(rp, rc_hash); if (rp->rc_flag & RC_UDP) { - TAILQ_REMOVE(&nfsrvudplru, rp, rc_lru); - nfsrc_udpcachesize--; + TAILQ_REMOVE(&NFSD_VNET(nfsrvudplru), rp, rc_lru); + NFSD_VNET(nfsrc_udpcachesize)--; } else if (rp->rc_acked != RC_NO_SEQ) { hbp = NFSRCAHASH(rp->rc_sockref); mtx_lock(&hbp->mtx); if (rp->rc_acked == RC_NO_ACK) LIST_REMOVE(rp, rc_ahash); mtx_unlock(&hbp->mtx); } nfsrc_wanted(rp); if (rp->rc_flag & RC_REPMBUF) { m_freem(rp->rc_reply); if (!(rp->rc_flag & RC_UDP)) - atomic_add_int(&nfsrc_tcpsavedreplies, -1); + atomic_add_int(&NFSD_VNET(nfsrc_tcpsavedreplies), -1); } free(rp, M_NFSRVCACHE); - atomic_add_int(&nfsstatsv1.srvcache_size, -1); + atomic_add_int(&nfsstatsv1_p->srvcache_size, -1); } /* * Clean out the cache. Called when nfsserver module is unloaded. */ void nfsrvd_cleancache(void) { struct nfsrvcache *rp, *nextrp; int i; for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { - mtx_lock(&nfsrchash_table[i].mtx); - LIST_FOREACH_SAFE(rp, &nfsrchash_table[i].tbl, rc_hash, nextrp) + mtx_lock(&NFSD_VNET(nfsrchash_table)[i].mtx); + LIST_FOREACH_SAFE(rp, &NFSD_VNET(nfsrchash_table)[i].tbl, + rc_hash, nextrp) nfsrc_freecache(rp); - mtx_unlock(&nfsrchash_table[i].mtx); + mtx_unlock(&NFSD_VNET(nfsrchash_table)[i].mtx); } mtx_lock(&nfsrc_udpmtx); for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { - LIST_FOREACH_SAFE(rp, &nfsrvudphashtbl[i], rc_hash, nextrp) { + LIST_FOREACH_SAFE(rp, &NFSD_VNET(nfsrvudphashtbl)[i], rc_hash, + nextrp) { nfsrc_freecache(rp); } } - nfsstatsv1.srvcache_size = 0; + nfsstatsv1_p->srvcache_size = 0; mtx_unlock(&nfsrc_udpmtx); - nfsrc_tcpsavedreplies = 0; + NFSD_VNET(nfsrc_tcpsavedreplies) = 0; } #define HISTSIZE 16 /* * The basic rule is to get rid of entries that are expired. */ void nfsrc_trimcache(u_int64_t sockref, uint32_t snd_una, int final) { struct nfsrchash_bucket *hbp; struct nfsrvcache *rp, *nextrp; int force, lastslot, i, j, k, tto, time_histo[HISTSIZE]; time_t thisstamp; static time_t udp_lasttrim = 0, tcp_lasttrim = 0; static int onethread = 0, oneslot = 0; if (sockref != 0) { hbp = NFSRCAHASH(sockref); mtx_lock(&hbp->mtx); LIST_FOREACH_SAFE(rp, &hbp->tbl, rc_ahash, nextrp) { if (sockref == rp->rc_sockref) { if (SEQ_GEQ(snd_una, rp->rc_tcpseq)) { rp->rc_acked = RC_ACK; LIST_REMOVE(rp, rc_ahash); } else if (final) { rp->rc_acked = RC_NACK; LIST_REMOVE(rp, rc_ahash); } } } mtx_unlock(&hbp->mtx); } if (atomic_cmpset_acq_int(&onethread, 0, 1) == 0) return; if (NFSD_MONOSEC != udp_lasttrim || - nfsrc_udpcachesize >= (nfsrc_udphighwater + + NFSD_VNET(nfsrc_udpcachesize) >= (nfsrc_udphighwater + nfsrc_udphighwater / 2)) { mtx_lock(&nfsrc_udpmtx); udp_lasttrim = NFSD_MONOSEC; - TAILQ_FOREACH_SAFE(rp, &nfsrvudplru, rc_lru, nextrp) { + TAILQ_FOREACH_SAFE(rp, &NFSD_VNET(nfsrvudplru), rc_lru, + nextrp) { if (!(rp->rc_flag & (RC_INPROG|RC_LOCKED|RC_WANTED)) && rp->rc_refcnt == 0 && ((rp->rc_flag & RC_REFCNT) || udp_lasttrim > rp->rc_timestamp || - nfsrc_udpcachesize > nfsrc_udphighwater)) + NFSD_VNET(nfsrc_udpcachesize) > + nfsrc_udphighwater)) nfsrc_freecache(rp); } mtx_unlock(&nfsrc_udpmtx); } if (NFSD_MONOSEC != tcp_lasttrim || - nfsrc_tcpsavedreplies >= nfsrc_tcphighwater) { + NFSD_VNET(nfsrc_tcpsavedreplies) >= nfsrc_tcphighwater) { force = nfsrc_tcphighwater / 4; if (force > 0 && - nfsrc_tcpsavedreplies + force >= nfsrc_tcphighwater) { + NFSD_VNET(nfsrc_tcpsavedreplies) + force >= + nfsrc_tcphighwater) { for (i = 0; i < HISTSIZE; i++) time_histo[i] = 0; i = 0; lastslot = NFSRVCACHE_HASHSIZE - 1; } else { force = 0; if (NFSD_MONOSEC != tcp_lasttrim) { i = 0; lastslot = NFSRVCACHE_HASHSIZE - 1; } else { lastslot = i = oneslot; if (++oneslot >= NFSRVCACHE_HASHSIZE) oneslot = 0; } } tto = nfsrc_tcptimeout; tcp_lasttrim = NFSD_MONOSEC; for (; i <= lastslot; i++) { - mtx_lock(&nfsrchash_table[i].mtx); - LIST_FOREACH_SAFE(rp, &nfsrchash_table[i].tbl, rc_hash, + mtx_lock(&NFSD_VNET(nfsrchash_table)[i].mtx); + LIST_FOREACH_SAFE(rp, + &NFSD_VNET(nfsrchash_table)[i].tbl, rc_hash, nextrp) { if (!(rp->rc_flag & (RC_INPROG|RC_LOCKED|RC_WANTED)) && rp->rc_refcnt == 0) { if ((rp->rc_flag & RC_REFCNT) || tcp_lasttrim > rp->rc_timestamp || rp->rc_acked == RC_ACK) { nfsrc_freecache(rp); continue; } if (force == 0) continue; /* * The timestamps range from roughly the * present (tcp_lasttrim) to the present * + nfsrc_tcptimeout. Generate a simple * histogram of where the timeouts fall. */ j = rp->rc_timestamp - tcp_lasttrim; if (j >= tto) j = HISTSIZE - 1; else if (j < 0) j = 0; else j = j * HISTSIZE / tto; time_histo[j]++; } } - mtx_unlock(&nfsrchash_table[i].mtx); + mtx_unlock(&NFSD_VNET(nfsrchash_table)[i].mtx); } if (force) { /* * Trim some more with a smaller timeout of as little * as 20% of nfsrc_tcptimeout to try and get below * 80% of the nfsrc_tcphighwater. */ k = 0; for (i = 0; i < (HISTSIZE - 2); i++) { k += time_histo[i]; if (k > force) break; } k = tto * (i + 1) / HISTSIZE; if (k < 1) k = 1; thisstamp = tcp_lasttrim + k; for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { - mtx_lock(&nfsrchash_table[i].mtx); - LIST_FOREACH_SAFE(rp, &nfsrchash_table[i].tbl, + mtx_lock(&NFSD_VNET(nfsrchash_table)[i].mtx); + LIST_FOREACH_SAFE(rp, + &NFSD_VNET(nfsrchash_table)[i].tbl, rc_hash, nextrp) { if (!(rp->rc_flag & (RC_INPROG|RC_LOCKED|RC_WANTED)) && rp->rc_refcnt == 0 && ((rp->rc_flag & RC_REFCNT) || thisstamp > rp->rc_timestamp || rp->rc_acked == RC_ACK)) nfsrc_freecache(rp); } - mtx_unlock(&nfsrchash_table[i].mtx); + mtx_unlock(&NFSD_VNET(nfsrchash_table)[i].mtx); } } } atomic_store_rel_int(&onethread, 0); } /* * Add a seqid# reference to the cache entry. */ void nfsrvd_refcache(struct nfsrvcache *rp) { struct mtx *mutex; if (rp == NULL) /* For NFSv4.1, there is no cache entry. */ return; mutex = nfsrc_cachemutex(rp); mtx_lock(mutex); if (rp->rc_refcnt < 0) panic("nfs cache refcnt"); rp->rc_refcnt++; mtx_unlock(mutex); } /* * Dereference a seqid# cache entry. */ void nfsrvd_derefcache(struct nfsrvcache *rp) { struct mtx *mutex; mutex = nfsrc_cachemutex(rp); mtx_lock(mutex); if (rp->rc_refcnt <= 0) panic("nfs cache derefcnt"); rp->rc_refcnt--; if (rp->rc_refcnt == 0 && !(rp->rc_flag & (RC_LOCKED | RC_INPROG))) nfsrc_freecache(rp); mtx_unlock(mutex); } /* * Calculate the length of the mbuf list and a checksum on the first up to * NFSRVCACHE_CHECKLEN bytes. */ static int nfsrc_getlenandcksum(struct mbuf *m1, u_int16_t *cksum) { int len = 0, cklen; struct mbuf *m; m = m1; while (m) { len += m->m_len; m = m->m_next; } cklen = (len > NFSRVCACHE_CHECKLEN) ? NFSRVCACHE_CHECKLEN : len; *cksum = in_cksum(m1, cklen); return (len); } /* * Mark a TCP connection that is seeing retries. Should never happen for * NFSv4. */ static void nfsrc_marksametcpconn(u_int64_t sockref) { } diff --git a/sys/fs/nfsserver/nfs_nfsdkrpc.c b/sys/fs/nfsserver/nfs_nfsdkrpc.c index 44f585ff0beb..7b0f84c397bc 100644 --- a/sys/fs/nfsserver/nfs_nfsdkrpc.c +++ b/sys/fs/nfsserver/nfs_nfsdkrpc.c @@ -1,616 +1,638 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet6.h" #include "opt_kgssapi.h" #include "opt_kern_tls.h" #include #include #include #include #include #include NFSDLOCKMUTEX; NFSV4ROOTLOCKMUTEX; -struct nfsv4lock nfsd_suspend_lock; char *nfsrv_zeropnfsdat = NULL; /* * Mapping of old NFS Version 2 RPC numbers to generic numbers. */ int newnfs_nfsv3_procid[NFS_V3NPROCS] = { NFSPROC_NULL, NFSPROC_GETATTR, NFSPROC_SETATTR, NFSPROC_NOOP, NFSPROC_LOOKUP, NFSPROC_READLINK, NFSPROC_READ, NFSPROC_NOOP, NFSPROC_WRITE, NFSPROC_CREATE, NFSPROC_REMOVE, NFSPROC_RENAME, NFSPROC_LINK, NFSPROC_SYMLINK, NFSPROC_MKDIR, NFSPROC_RMDIR, NFSPROC_READDIR, NFSPROC_FSSTAT, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, NFSPROC_NOOP, }; SYSCTL_DECL(_vfs_nfsd); -SVCPOOL *nfsrvd_pool; - -static int nfs_privport = 0; -SYSCTL_INT(_vfs_nfsd, OID_AUTO, nfs_privport, CTLFLAG_RWTUN, - &nfs_privport, 0, +NFSD_VNET_DEFINE_STATIC(int, nfs_privport) = 0; +SYSCTL_INT(_vfs_nfsd, OID_AUTO, nfs_privport, CTLFLAG_NFSD_VNET | CTLFLAG_RWTUN, + &NFSD_VNET_NAME(nfs_privport), 0, "Only allow clients using a privileged port for NFSv2, 3 and 4"); -static int nfs_minvers = NFS_VER2; -SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_min_nfsvers, CTLFLAG_RWTUN, - &nfs_minvers, 0, "The lowest version of NFS handled by the server"); +NFSD_VNET_DEFINE_STATIC(int, nfs_minvers) = NFS_VER2; +SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_min_nfsvers, + CTLFLAG_NFSD_VNET | CTLFLAG_RWTUN, &NFSD_VNET_NAME(nfs_minvers), 0, + "The lowest version of NFS handled by the server"); -static int nfs_maxvers = NFS_VER4; -SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_max_nfsvers, CTLFLAG_RWTUN, - &nfs_maxvers, 0, "The highest version of NFS handled by the server"); +NFSD_VNET_DEFINE_STATIC(int, nfs_maxvers) = NFS_VER4; +SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_max_nfsvers, + CTLFLAG_NFSD_VNET | CTLFLAG_RWTUN, &NFSD_VNET_NAME(nfs_maxvers), 0, + "The highest version of NFS handled by the server"); static int nfs_proc(struct nfsrv_descript *, u_int32_t, SVCXPRT *xprt, struct nfsrvcache **); extern u_long sb_max_adj; extern int newnfs_numnfsd; -extern struct proc *nfsd_master_proc; extern time_t nfsdev_time; extern int nfsrv_writerpc[NFS_NPROCS]; extern volatile int nfsrv_devidcnt; extern struct nfsv4_opflag nfsv4_opflag[NFSV42_NOPS]; +NFSD_VNET_DECLARE(struct proc *, nfsd_master_proc); + +NFSD_VNET_DEFINE(SVCPOOL *, nfsrvd_pool); +NFSD_VNET_DEFINE(int, nfsrv_numnfsd) = 0; +NFSD_VNET_DEFINE(struct nfsv4lock, nfsd_suspend_lock); + +NFSD_VNET_DEFINE_STATIC(bool, nfsrvd_inited) = false; + /* * NFS server system calls */ static void nfssvc_program(struct svc_req *rqst, SVCXPRT *xprt) { struct nfsrv_descript nd; struct nfsrvcache *rp = NULL; int cacherep, credflavor; #ifdef KERN_TLS u_int maxlen; #endif + NFSD_CURVNET_SET_QUIET(NFSD_TD_TO_VNET(curthread)); memset(&nd, 0, sizeof(nd)); if (rqst->rq_vers == NFS_VER2) { if (rqst->rq_proc > NFSV2PROC_STATFS || newnfs_nfsv3_procid[rqst->rq_proc] == NFSPROC_NOOP) { svcerr_noproc(rqst); svc_freereq(rqst); goto out; } nd.nd_procnum = newnfs_nfsv3_procid[rqst->rq_proc]; nd.nd_flag = ND_NFSV2; } else if (rqst->rq_vers == NFS_VER3) { if (rqst->rq_proc >= NFS_V3NPROCS) { svcerr_noproc(rqst); svc_freereq(rqst); goto out; } nd.nd_procnum = rqst->rq_proc; nd.nd_flag = ND_NFSV3; } else { if (rqst->rq_proc != NFSPROC_NULL && rqst->rq_proc != NFSV4PROC_COMPOUND) { svcerr_noproc(rqst); svc_freereq(rqst); goto out; } nd.nd_procnum = rqst->rq_proc; nd.nd_flag = ND_NFSV4; } /* * Note: we want rq_addr, not svc_getrpccaller for nd_nam2 - * NFS_SRVMAXDATA uses a NULL value for nd_nam2 to detect TCP * mounts. */ nd.nd_mrep = rqst->rq_args; rqst->rq_args = NULL; newnfs_realign(&nd.nd_mrep, M_WAITOK); nd.nd_md = nd.nd_mrep; nd.nd_dpos = mtod(nd.nd_md, caddr_t); nd.nd_nam = svc_getrpccaller(rqst); nd.nd_nam2 = rqst->rq_addr; nd.nd_mreq = NULL; nd.nd_cred = NULL; - if (nfs_privport != 0) { + if (NFSD_VNET(nfs_privport) != 0) { /* Check if source port is privileged */ u_short port; struct sockaddr *nam = nd.nd_nam; struct sockaddr_in *sin; sin = (struct sockaddr_in *)nam; /* * INET/INET6 - same code: * sin_port and sin6_port are at same offset */ port = ntohs(sin->sin_port); if (port >= IPPORT_RESERVED && nd.nd_procnum != NFSPROC_NULL) { #ifdef INET6 char buf[INET6_ADDRSTRLEN]; #else char buf[INET_ADDRSTRLEN]; #endif #ifdef INET6 #if defined(KLD_MODULE) /* Do not use ip6_sprintf: the nfs module should work without INET6. */ #define ip6_sprintf(buf, a) \ (sprintf((buf), "%x:%x:%x:%x:%x:%x:%x:%x", \ (a)->s6_addr16[0], (a)->s6_addr16[1], \ (a)->s6_addr16[2], (a)->s6_addr16[3], \ (a)->s6_addr16[4], (a)->s6_addr16[5], \ (a)->s6_addr16[6], (a)->s6_addr16[7]), \ (buf)) #endif #endif printf("NFS request from unprivileged port (%s:%d)\n", #ifdef INET6 sin->sin_family == AF_INET6 ? ip6_sprintf(buf, &satosin6(sin)->sin6_addr) : #if defined(KLD_MODULE) #undef ip6_sprintf #endif #endif inet_ntoa_r(sin->sin_addr, buf), port); svcerr_weakauth(rqst); svc_freereq(rqst); m_freem(nd.nd_mrep); goto out; } } if (nd.nd_procnum != NFSPROC_NULL) { if (!svc_getcred(rqst, &nd.nd_cred, &credflavor)) { svcerr_weakauth(rqst); svc_freereq(rqst); m_freem(nd.nd_mrep); goto out; } /* Set the flag based on credflavor */ if (credflavor == RPCSEC_GSS_KRB5) { nd.nd_flag |= ND_GSS; } else if (credflavor == RPCSEC_GSS_KRB5I) { nd.nd_flag |= (ND_GSS | ND_GSSINTEGRITY); } else if (credflavor == RPCSEC_GSS_KRB5P) { nd.nd_flag |= (ND_GSS | ND_GSSPRIVACY); } else if (credflavor != AUTH_SYS) { svcerr_weakauth(rqst); svc_freereq(rqst); m_freem(nd.nd_mrep); goto out; } if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0) { nd.nd_flag |= ND_TLS; if ((xprt->xp_tls & RPCTLS_FLAGS_VERIFIED) != 0) nd.nd_flag |= ND_TLSCERT; if ((xprt->xp_tls & RPCTLS_FLAGS_CERTUSER) != 0) nd.nd_flag |= ND_TLSCERTUSER; } nd.nd_maxextsiz = 16384; #ifdef MAC mac_cred_associate_nfsd(nd.nd_cred); #endif /* * Get a refcnt (shared lock) on nfsd_suspend_lock. * NFSSVC_SUSPENDNFSD will take an exclusive lock on * nfsd_suspend_lock to suspend these threads. * The call to nfsv4_lock() that precedes nfsv4_getref() * ensures that the acquisition of the exclusive lock * takes priority over acquisition of the shared lock by * waiting for any exclusive lock request to complete. * This must be done here, before the check of * nfsv4root exports by nfsvno_v4rootexport(). */ NFSLOCKV4ROOTMUTEX(); - nfsv4_lock(&nfsd_suspend_lock, 0, NULL, NFSV4ROOTLOCKMUTEXPTR, - NULL); - nfsv4_getref(&nfsd_suspend_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, - NULL); + nfsv4_lock(&NFSD_VNET(nfsd_suspend_lock), 0, NULL, + NFSV4ROOTLOCKMUTEXPTR, NULL); + nfsv4_getref(&NFSD_VNET(nfsd_suspend_lock), NULL, + NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); if ((nd.nd_flag & ND_NFSV4) != 0) { nd.nd_repstat = nfsvno_v4rootexport(&nd); if (nd.nd_repstat != 0) { NFSLOCKV4ROOTMUTEX(); - nfsv4_relref(&nfsd_suspend_lock); + nfsv4_relref(&NFSD_VNET(nfsd_suspend_lock)); NFSUNLOCKV4ROOTMUTEX(); svcerr_weakauth(rqst); svc_freereq(rqst); m_freem(nd.nd_mrep); goto out; } } #ifdef KERN_TLS if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0 && rpctls_getinfo(&maxlen, false, false)) nd.nd_maxextsiz = maxlen; #endif cacherep = nfs_proc(&nd, rqst->rq_xid, xprt, &rp); NFSLOCKV4ROOTMUTEX(); - nfsv4_relref(&nfsd_suspend_lock); + nfsv4_relref(&NFSD_VNET(nfsd_suspend_lock)); NFSUNLOCKV4ROOTMUTEX(); } else { NFSMGET(nd.nd_mreq); nd.nd_mreq->m_len = 0; cacherep = RC_REPLY; } if (nd.nd_mrep != NULL) m_freem(nd.nd_mrep); if (nd.nd_cred != NULL) crfree(nd.nd_cred); if (cacherep == RC_DROPIT) { if (nd.nd_mreq != NULL) m_freem(nd.nd_mreq); svc_freereq(rqst); goto out; } if (nd.nd_mreq == NULL) { svcerr_decode(rqst); svc_freereq(rqst); goto out; } if (nd.nd_repstat & NFSERR_AUTHERR) { svcerr_auth(rqst, nd.nd_repstat & ~NFSERR_AUTHERR); if (nd.nd_mreq != NULL) m_freem(nd.nd_mreq); } else if (!svc_sendreply_mbuf(rqst, nd.nd_mreq)) { svcerr_systemerr(rqst); } if (rp != NULL) { nfsrvd_sentcache(rp, (rqst->rq_reply_seq != 0 || SVC_ACK(xprt, NULL)), rqst->rq_reply_seq); } svc_freereq(rqst); out: + NFSD_CURVNET_RESTORE(); td_softdep_cleanup(curthread); NFSEXITCODE(0); } /* * Check the cache and, optionally, do the RPC. * Return the appropriate cache response. */ static int nfs_proc(struct nfsrv_descript *nd, u_int32_t xid, SVCXPRT *xprt, struct nfsrvcache **rpp) { int cacherep = RC_DOIT, isdgram, taglen = -1; struct mbuf *m; u_char tag[NFSV4_SMALLSTR + 1], *tagstr = NULL; u_int32_t minorvers = 0; uint32_t ack; *rpp = NULL; if (nd->nd_nam2 == NULL) { nd->nd_flag |= ND_STREAMSOCK; isdgram = 0; } else { isdgram = 1; } /* * Two cases: * 1 - For NFSv2 over UDP, if we are near our malloc/mget * limit, just drop the request. There is no * NFSERR_RESOURCE or NFSERR_DELAY for NFSv2 and the * client will timeout/retry over UDP in a little while. * 2 - nd_repstat == 0 && nd_mreq == NULL, which * means a normal nfs rpc, so check the cache */ if ((nd->nd_flag & ND_NFSV2) && nd->nd_nam2 != NULL && nfsrv_mallocmget_limit()) { cacherep = RC_DROPIT; } else { /* * For NFSv3, play it safe and assume that the client is * doing retries on the same TCP connection. */ if ((nd->nd_flag & (ND_NFSV4 | ND_STREAMSOCK)) == ND_STREAMSOCK) nd->nd_flag |= ND_SAMETCPCONN; nd->nd_retxid = xid; nd->nd_tcpconntime = NFSD_MONOSEC; nd->nd_sockref = xprt->xp_sockref; if ((nd->nd_flag & ND_NFSV4) != 0) nfsd_getminorvers(nd, tag, &tagstr, &taglen, &minorvers); if ((nd->nd_flag & ND_NFSV41) != 0) /* NFSv4.1 caches replies in the session slots. */ cacherep = RC_DOIT; else { cacherep = nfsrvd_getcache(nd); ack = 0; SVC_ACK(xprt, &ack); nfsrc_trimcache(xprt->xp_sockref, ack, 0); } } /* * Handle the request. There are three cases. * RC_DOIT - do the RPC * RC_REPLY - return the reply already created * RC_DROPIT - just throw the request away */ if (cacherep == RC_DOIT) { if ((nd->nd_flag & ND_NFSV41) != 0) nd->nd_xprt = xprt; nfsrvd_dorpc(nd, isdgram, tagstr, taglen, minorvers); if ((nd->nd_flag & ND_NFSV41) != 0) { if (nd->nd_repstat != NFSERR_REPLYFROMCACHE && (nd->nd_flag & ND_SAVEREPLY) != 0) { /* Cache a copy of the reply. */ m = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK); } else m = NULL; if ((nd->nd_flag & ND_HASSEQUENCE) != 0) nfsrv_cache_session(nd, &m); if (nd->nd_repstat == NFSERR_REPLYFROMCACHE) { nd->nd_repstat = 0; if (m != NULL) { m_freem(nd->nd_mreq); nd->nd_mreq = m; } } cacherep = RC_REPLY; } else { if (nd->nd_repstat == NFSERR_DONTREPLY) cacherep = RC_DROPIT; else cacherep = RC_REPLY; *rpp = nfsrvd_updatecache(nd); } } if (tagstr != NULL && taglen > NFSV4_SMALLSTR) free(tagstr, M_TEMP); NFSEXITCODE2(0, nd); return (cacherep); } static void nfssvc_loss(SVCXPRT *xprt) { uint32_t ack; ack = 0; SVC_ACK(xprt, &ack); + NFSD_CURVNET_SET(NFSD_TD_TO_VNET(curthread)); nfsrc_trimcache(xprt->xp_sockref, ack, 1); + NFSD_CURVNET_RESTORE(); } /* * Adds a socket to the list for servicing by nfsds. */ int nfsrvd_addsock(struct file *fp) { int siz; struct socket *so; int error = 0; SVCXPRT *xprt; static u_int64_t sockref = 0; so = fp->f_data; siz = sb_max_adj; error = soreserve(so, siz, siz); if (error) goto out; /* * Steal the socket from userland so that it doesn't close * unexpectedly. */ if (so->so_type == SOCK_DGRAM) - xprt = svc_dg_create(nfsrvd_pool, so, 0, 0); + xprt = svc_dg_create(NFSD_VNET(nfsrvd_pool), so, 0, 0); else - xprt = svc_vc_create(nfsrvd_pool, so, 0, 0); + xprt = svc_vc_create(NFSD_VNET(nfsrvd_pool), so, 0, 0); if (xprt) { fp->f_ops = &badfileops; fp->f_data = NULL; xprt->xp_sockref = ++sockref; - if (nfs_minvers == NFS_VER2) + if (NFSD_VNET(nfs_minvers) == NFS_VER2) svc_reg(xprt, NFS_PROG, NFS_VER2, nfssvc_program, NULL); - if (nfs_minvers <= NFS_VER3 && nfs_maxvers >= NFS_VER3) + if (NFSD_VNET(nfs_minvers) <= NFS_VER3 && + NFSD_VNET(nfs_maxvers) >= NFS_VER3) svc_reg(xprt, NFS_PROG, NFS_VER3, nfssvc_program, NULL); - if (nfs_maxvers >= NFS_VER4) + if (NFSD_VNET(nfs_maxvers) >= NFS_VER4) svc_reg(xprt, NFS_PROG, NFS_VER4, nfssvc_program, NULL); if (so->so_type == SOCK_STREAM) svc_loss_reg(xprt, nfssvc_loss); SVC_RELEASE(xprt); - } + } else + error = EPERM; out: NFSEXITCODE(error); return (error); } /* * Called by nfssvc() for nfsds. Just loops around servicing rpc requests * until it is killed by a signal. */ int nfsrvd_nfsd(struct thread *td, struct nfsd_nfsd_args *args) { char principal[MAXHOSTNAMELEN + 5]; struct proc *p; int error = 0; bool_t ret2, ret3, ret4; error = copyinstr(args->principal, principal, sizeof (principal), NULL); if (error) goto out; /* * Only the first nfsd actually does any work. The RPC code * adds threads to it as needed. Any extra processes offered * by nfsd just exit. If nfsd is new enough, it will call us * once with a structure that specifies how many threads to * use. */ NFSD_LOCK(); - if (newnfs_numnfsd == 0) { + if (NFSD_VNET(nfsrv_numnfsd) == 0) { + nfsrvd_init(0); nfsdev_time = time_second; p = td->td_proc; PROC_LOCK(p); p->p_flag2 |= P2_AST_SU; PROC_UNLOCK(p); - newnfs_numnfsd++; + newnfs_numnfsd++; /* Total num for all vnets. */ + NFSD_VNET(nfsrv_numnfsd)++; /* Num for this vnet. */ NFSD_UNLOCK(); error = nfsrv_createdevids(args, td); if (error == 0) { /* An empty string implies AUTH_SYS only. */ if (principal[0] != '\0') { ret2 = rpc_gss_set_svc_name_call(principal, "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER2); ret3 = rpc_gss_set_svc_name_call(principal, "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER3); ret4 = rpc_gss_set_svc_name_call(principal, "kerberosv5", GSS_C_INDEFINITE, NFS_PROG, NFS_VER4); if (!ret2 || !ret3 || !ret4) - printf( - "nfsd: can't register svc name\n"); + printf("nfsd: can't register svc " + "name %s jid:%d\n", principal, + td->td_ucred->cr_prison->pr_id); } - nfsrvd_pool->sp_minthreads = args->minthreads; - nfsrvd_pool->sp_maxthreads = args->maxthreads; + NFSD_VNET(nfsrvd_pool)->sp_minthreads = + args->minthreads; + NFSD_VNET(nfsrvd_pool)->sp_maxthreads = + args->maxthreads; /* * If this is a pNFS service, make Getattr do a * vn_start_write(), so it can do a vn_set_extattr(). */ if (nfsrv_devidcnt > 0) { nfsrv_writerpc[NFSPROC_GETATTR] = 1; nfsv4_opflag[NFSV4OP_GETATTR].modifyfs = 1; } - svc_run(nfsrvd_pool); + svc_run(NFSD_VNET(nfsrvd_pool)); /* Reset Getattr to not do a vn_start_write(). */ nfsrv_writerpc[NFSPROC_GETATTR] = 0; nfsv4_opflag[NFSV4OP_GETATTR].modifyfs = 0; if (principal[0] != '\0') { rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER2); rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER3); rpc_gss_clear_svc_name_call(NFS_PROG, NFS_VER4); } } NFSD_LOCK(); newnfs_numnfsd--; + NFSD_VNET(nfsrv_numnfsd)--; nfsrvd_init(1); PROC_LOCK(p); p->p_flag2 &= ~P2_AST_SU; PROC_UNLOCK(p); } NFSD_UNLOCK(); out: NFSEXITCODE(error); return (error); } /* * Initialize the data structures for the server. * Handshake with any new nfsds starting up to avoid any chance of * corruption. */ void nfsrvd_init(int terminating) { NFSD_LOCK_ASSERT(); if (terminating) { - nfsd_master_proc = NULL; + NFSD_VNET(nfsd_master_proc) = NULL; NFSD_UNLOCK(); nfsrv_freealllayoutsanddevids(); nfsrv_freeallbackchannel_xprts(); - svcpool_close(nfsrvd_pool); + svcpool_close(NFSD_VNET(nfsrvd_pool)); free(nfsrv_zeropnfsdat, M_TEMP); nfsrv_zeropnfsdat = NULL; NFSD_LOCK(); } else { + /* Initialize per-vnet globals once per vnet. */ + if (NFSD_VNET(nfsrvd_inited)) + return; + NFSD_VNET(nfsrvd_inited) = true; NFSD_UNLOCK(); - nfsrvd_pool = svcpool_create("nfsd", + NFSD_VNET(nfsrvd_pool) = svcpool_create("nfsd", SYSCTL_STATIC_CHILDREN(_vfs_nfsd)); - nfsrvd_pool->sp_rcache = NULL; - nfsrvd_pool->sp_assign = fhanew_assign; - nfsrvd_pool->sp_done = fhanew_nd_complete; + NFSD_VNET(nfsrvd_pool)->sp_rcache = NULL; + NFSD_VNET(nfsrvd_pool)->sp_assign = fhanew_assign; + NFSD_VNET(nfsrvd_pool)->sp_done = fhanew_nd_complete; NFSD_LOCK(); } } diff --git a/sys/fs/nfsserver/nfs_nfsdport.c b/sys/fs/nfsserver/nfs_nfsdport.c index 407987aa767f..a58df7b69420 100644 --- a/sys/fs/nfsserver/nfs_nfsdport.c +++ b/sys/fs/nfsserver/nfs_nfsdport.c @@ -1,6968 +1,7060 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include #include /* * Functions that perform the vfs operations required by the routines in * nfsd_serv.c. It is hoped that this change will make the server more * portable. */ #include #include #include #include #include +#include #include #include #include FEATURE(nfsd, "NFSv4 server"); extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; extern int nfsrv_useacl; extern int newnfs_numnfsd; -extern struct mount nfsv4root_mnt; -extern struct nfsrv_stablefirst nfsrv_stablefirst; -extern SVCPOOL *nfsrvd_pool; -extern struct nfsv4lock nfsd_suspend_lock; -extern struct nfsclienthashhead *nfsclienthash; -extern struct nfslockhashhead *nfslockhash; -extern struct nfssessionhash *nfssessionhash; extern int nfsrv_sessionhashsize; extern struct nfsstatsv1 nfsstatsv1; extern struct nfslayouthash *nfslayouthash; extern int nfsrv_layouthashsize; extern struct mtx nfsrv_dslock_mtx; extern int nfs_pnfsiothreads; -extern struct nfsdontlisthead nfsrv_dontlisthead; -extern volatile int nfsrv_dontlistlen; extern volatile int nfsrv_devidcnt; extern int nfsrv_maxpnfsmirror; extern uint32_t nfs_srvmaxio; extern int nfs_bufpackets; extern u_long sb_max_adj; -struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; + +NFSD_VNET_DECLARE(int, nfsrv_numnfsd); +NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst); +NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool); +NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); +NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash); +NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash); +NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock); + NFSDLOCKMUTEX; NFSSTATESPINLOCK; -struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; -struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; struct mtx nfsrc_udpmtx; struct mtx nfs_v4root_mutex; struct mtx nfsrv_dontlistlock_mtx; struct mtx nfsrv_recalllock_mtx; -struct nfsrvfh nfs_rootfh, nfs_pubfh; -int nfs_pubfhset = 0, nfs_rootfhset = 0; -struct proc *nfsd_master_proc = NULL; +struct nfsrvfh nfs_pubfh; +int nfs_pubfhset = 0; int nfsd_debuglevel = 0; static pid_t nfsd_master_pid = (pid_t)-1; static char nfsd_master_comm[MAXCOMLEN + 1]; static struct timeval nfsd_master_start; static uint32_t nfsv4_sysid = 0; static fhandle_t zerofh; -struct callout nfsd_callout; +static int nfsrv_osd_jail_slot; + +NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL; +NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl); +NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table); +NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table); +NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh); +NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0; +NFSD_VNET_DEFINE(struct callout, nfsd_callout); +NFSD_VNET_DEFINE(struct nfsstatsv1 *, nfsstatsv1_p); +NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt); +NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt); +NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt); +NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false; +NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false; + +static void nfsrv_cleanup(struct prison *); static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, struct ucred *); static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *); int nfsrv_enable_crossmntpt = 1; static int nfs_commit_blks; static int nfs_commit_miss; extern int nfsrv_issuedelegs; extern int nfsrv_dolocallocks; -extern int nfsd_enable_stringtouid; extern struct nfsdevicehead nfsrv_devidhead; static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, struct iovec **); static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, struct mbuf **, struct iovec **); static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, int *); static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, NFSPROC_T *); static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, int *, char *, fhandle_t *); static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, NFSPROC_T *); static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, struct thread *, int, struct mbuf **, char *, struct mbuf **, struct nfsvattr *, struct acl *, off_t *, int, bool *); static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, char *, int *); static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct acl *, int *); static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, struct vnode *, struct nfsmount *, struct nfsvattr *); static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, NFSPROC_T *, struct nfsmount *); static int nfsrv_putfhname(fhandle_t *, char *); static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, struct pnfsdsfile *, struct vnode **, NFSPROC_T *); static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, struct vnode *, NFSPROC_T *); static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, NFSPROC_T *); static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); int nfs_pnfsio(task_fn_t *, void *); SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "NFS server"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, ""); SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, ""); SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 0, "Debug level for NFS server"); -SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, - &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); +NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid); +SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, + CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid), + 0, "Enable nfsd to accept numeric owner_names"); static int nfsrv_pnfsgetdsattr = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); /* * nfsrv_dsdirsize can only be increased and only when the nfsd threads are * not running. * The dsN subdirectories for the increased values must have been created * on all DS servers before this increase is done. */ u_int nfsrv_dsdirsize = 20; static int sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) { int error, newdsdirsize; newdsdirsize = nfsrv_dsdirsize; error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || newnfs_numnfsd != 0) return (EINVAL); nfsrv_dsdirsize = newdsdirsize; return (0); } SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); /* * nfs_srvmaxio can only be increased and only when the nfsd threads are * not running. The setting must be a power of 2, with the current limit of * 1Mbyte. */ static int sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) { int error; u_int newsrvmaxio; uint64_t tval; newsrvmaxio = nfs_srvmaxio; error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (newsrvmaxio == nfs_srvmaxio) return (0); if (newsrvmaxio < nfs_srvmaxio) { printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); return (EINVAL); } if (newsrvmaxio > 1048576) { printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); return (EINVAL); } if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); return (EINVAL); } /* * Check that kern.ipc.maxsockbuf is large enough for * newsrviomax, given the setting of vfs.nfs.bufpackets. */ if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > sb_max_adj) { /* * Suggest vfs.nfs.bufpackets * maximum RPC message for * sb_max_adj. */ tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; /* * Convert suggested sb_max_adj value to a suggested * sb_max value, which is what is set via kern.ipc.maxsockbuf. * Perform the inverse calculation of (from uipc_sockbuf.c): * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / * (MSIZE + MCLBYTES); * XXX If the calculation of sb_max_adj from sb_max changes, * this calculation must be changed as well. */ tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ tval += MCLBYTES - 1; /* Round up divide. */ tval /= MCLBYTES; printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, newsrvmaxio); return (EINVAL); } NFSD_LOCK(); if (newnfs_numnfsd != 0) { NFSD_UNLOCK(); printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " "threads are running\n"); return (EINVAL); } nfs_srvmaxio = newsrvmaxio; NFSD_UNLOCK(); return (0); } SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); #define MAX_REORDERED_RPC 16 #define NUM_HEURISTIC 1031 #define NHUSE_INIT 64 #define NHUSE_INC 16 #define NHUSE_MAX 2048 static struct nfsheur { struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ off_t nh_nextoff; /* next offset for sequential detection */ int nh_use; /* use count for selection */ int nh_seqcount; /* heuristic */ } nfsheur[NUM_HEURISTIC]; /* * Heuristic to detect sequential operation. */ static struct nfsheur * nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) { struct nfsheur *nh; int hi, try; /* Locate best candidate. */ try = 32; hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; nh = &nfsheur[hi]; while (try--) { if (nfsheur[hi].nh_vp == vp) { nh = &nfsheur[hi]; break; } if (nfsheur[hi].nh_use > 0) --nfsheur[hi].nh_use; hi = (hi + 1) % NUM_HEURISTIC; if (nfsheur[hi].nh_use < nh->nh_use) nh = &nfsheur[hi]; } /* Initialize hint if this is a new file. */ if (nh->nh_vp != vp) { nh->nh_vp = vp; nh->nh_nextoff = uio->uio_offset; nh->nh_use = NHUSE_INIT; if (uio->uio_offset == 0) nh->nh_seqcount = 4; else nh->nh_seqcount = 1; } /* Calculate heuristic. */ if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || uio->uio_offset == nh->nh_nextoff) { /* See comments in vfs_vnops.c:sequential_heuristic(). */ nh->nh_seqcount += howmany(uio->uio_resid, 16384); if (nh->nh_seqcount > IO_SEQMAX) nh->nh_seqcount = IO_SEQMAX; } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { /* Probably a reordered RPC, leave seqcount alone. */ } else if (nh->nh_seqcount > 1) { nh->nh_seqcount /= 2; } else { nh->nh_seqcount = 0; } nh->nh_use += NHUSE_INC; if (nh->nh_use > NHUSE_MAX) nh->nh_use = NHUSE_MAX; return (nh); } /* * Get attributes into nfsvattr structure. */ int nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct nfsrv_descript *nd, struct thread *p, int vpislocked, nfsattrbit_t *attrbitp) { int error, gotattr, lockedit = 0; struct nfsvattr na; if (vpislocked == 0) { /* * When vpislocked == 0, the vnode is either exclusively * locked by this thread or not locked by this thread. * As such, shared lock it, if not exclusively locked. */ if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { lockedit = 1; NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); } } /* * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed * attributes, as required. * This needs to be done for regular files if: * - non-NFSv4 RPCs or * - when attrbitp == NULL or * - an NFSv4 RPC with any of the above attributes in attrbitp. * A return of 0 for nfsrv_proxyds() indicates that it has acquired * these attributes. nfsrv_proxyds() will return an error if the * server is not a pNFS one. */ gotattr = 0; if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || (nd->nd_flag & ND_NFSV4) == 0 || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, NULL); if (error == 0) gotattr = 1; } error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); if (lockedit != 0) NFSVOPUNLOCK(vp); /* * If we got the Change, Size and Modify Time from the DS, * replace them. */ if (gotattr != 0) { nvap->na_atime = na.na_atime; nvap->na_mtime = na.na_mtime; nvap->na_filerev = na.na_filerev; nvap->na_size = na.na_size; nvap->na_bytes = na.na_bytes; } NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, error, (uintmax_t)na.na_filerev); NFSEXITCODE(error); return (error); } /* * Get a file handle for a vnode. */ int nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) { int error; NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fhp->fh_fid); NFSEXITCODE(error); return (error); } /* * Perform access checking for vnodes obtained from file handles that would * refer to files already opened by a Unix client. You cannot just use * vn_writechk() and VOP_ACCESSX() for two reasons. * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write * case. * 2 - The owner is to be given access irrespective of mode bits for some * operations, so that processes that chmod after opening a file don't * break. */ int nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, u_int32_t *supportedtypep) { struct vattr vattr; int error = 0, getret = 0; if (vpislocked == 0) { if (NFSVOPLOCK(vp, LK_SHARED) != 0) { error = EPERM; goto out; } } if (accmode & VWRITE) { /* Just vn_writechk() changed to check rdonly */ /* * Disallow write attempts on read-only file systems; * unless the file is a socket or a block or character * device resident on the file system. */ if (NFSVNO_EXRDONLY(exp) || (vp->v_mount->mnt_flag & MNT_RDONLY)) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: error = EROFS; default: break; } } /* * If there's shared text associated with * the inode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp) && error == 0) error = ETXTBSY; } if (error != 0) { if (vpislocked == 0) NFSVOPUNLOCK(vp); goto out; } /* * Should the override still be applied when ACLs are enabled? */ error = VOP_ACCESSX(vp, accmode, cred, p); if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { /* * Try again with VEXPLICIT_DENY, to see if the test for * deletion is supported. */ error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); if (error == 0) { if (vp->v_type == VDIR) { accmode &= ~(VDELETE | VDELETE_CHILD); accmode |= VWRITE; error = VOP_ACCESSX(vp, accmode, cred, p); } else if (supportedtypep != NULL) { *supportedtypep &= ~NFSACCESS_DELETE; } } } /* * Allow certain operations for the owner (reads and writes * on files that are already open). */ if (override != NFSACCCHK_NOOVERRIDE && (error == EPERM || error == EACCES)) { if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) error = 0; else if (override & NFSACCCHK_ALLOWOWNER) { getret = VOP_GETATTR(vp, &vattr, cred); if (getret == 0 && cred->cr_uid == vattr.va_uid) error = 0; } } if (vpislocked == 0) NFSVOPUNLOCK(vp); out: NFSEXITCODE(error); return (error); } /* * Set attribute(s) vnop. */ int nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { u_quad_t savsize = 0; int error, savedit; time_t savbtime; /* * If this is an exported file system and a pNFS service is running, * don't VOP_SETATTR() of size for the MDS file system. */ savedit = 0; error = 0; if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && nvap->na_vattr.va_size > 0) { savsize = nvap->na_vattr.va_size; nvap->na_vattr.va_size = VNOVAL; if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || nvap->na_vattr.va_gid != (gid_t)VNOVAL || nvap->na_vattr.va_mode != (mode_t)VNOVAL || nvap->na_vattr.va_atime.tv_sec != VNOVAL || nvap->na_vattr.va_mtime.tv_sec != VNOVAL) savedit = 1; else savedit = 2; } if (savedit != 2) error = VOP_SETATTR(vp, &nvap->na_vattr, cred); if (savedit != 0) nvap->na_vattr.va_size = savsize; if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || nvap->na_vattr.va_gid != (gid_t)VNOVAL || nvap->na_vattr.va_size != VNOVAL || nvap->na_vattr.va_mode != (mode_t)VNOVAL || nvap->na_vattr.va_atime.tv_sec != VNOVAL || nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { /* Never modify birthtime on a DS file. */ savbtime = nvap->na_vattr.va_birthtime.tv_sec; nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; /* For a pNFS server, set the attributes on the DS file. */ error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); nvap->na_vattr.va_birthtime.tv_sec = savbtime; if (error == ENOENT) error = 0; } NFSEXITCODE(error); return (error); } /* * Set up nameidata for a lookup() call and do it. */ int nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p, struct vnode **retdirp) { struct componentname *cnp = &ndp->ni_cnd; int i; struct iovec aiov; struct uio auio; int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; int error = 0; char *cp; *retdirp = NULL; cnp->cn_nameptr = cnp->cn_pnbuf; ndp->ni_lcf = 0; /* * Extract and set starting directory. */ if (dp->v_type != VDIR) { if (islocked) vput(dp); else vrele(dp); nfsvno_relpathbuf(ndp); error = ENOTDIR; goto out1; } if (islocked) NFSVOPUNLOCK(dp); VREF(dp); *retdirp = dp; if (NFSVNO_EXRDONLY(exp)) cnp->cn_flags |= RDONLY; ndp->ni_segflg = UIO_SYSSPACE; if (nd->nd_flag & ND_PUBLOOKUP) { ndp->ni_loopcnt = 0; if (cnp->cn_pnbuf[0] == '/') { vrele(dp); /* * Check for degenerate pathnames here, since lookup() * panics on them. */ for (i = 1; i < ndp->ni_pathlen; i++) if (cnp->cn_pnbuf[i] != '/') break; if (i == ndp->ni_pathlen) { error = NFSERR_ACCES; goto out; } dp = rootvnode; VREF(dp); } } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || (nd->nd_flag & ND_NFSV4) == 0) { /* * Only cross mount points for NFSv4 when doing a * mount while traversing the file system above * the mount point, unless nfsrv_enable_crossmntpt is set. */ cnp->cn_flags |= NOCROSSMOUNT; } /* * Initialize for scan, set ni_startdir and bump ref on dp again * because lookup() will dereference ni_startdir. */ cnp->cn_thread = p; ndp->ni_startdir = dp; ndp->ni_rootdir = rootvnode; ndp->ni_topdir = NULL; if (!lockleaf) cnp->cn_flags |= LOCKLEAF; for (;;) { cnp->cn_nameptr = cnp->cn_pnbuf; /* * Call lookup() to do the real work. If an error occurs, * ndp->ni_vp and ni_dvp are left uninitialized or NULL and * we do not have to dereference anything before returning. * In either case ni_startdir will be dereferenced and NULLed * out. */ error = lookup(ndp); if (error) break; /* * Check for encountering a symbolic link. Trivial * termination occurs if no symlink encountered. */ if ((cnp->cn_flags & ISSYMLINK) == 0) { if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) nfsvno_relpathbuf(ndp); if (ndp->ni_vp && !lockleaf) NFSVOPUNLOCK(ndp->ni_vp); break; } /* * Validate symlink */ if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) NFSVOPUNLOCK(ndp->ni_dvp); if (!(nd->nd_flag & ND_PUBLOOKUP)) { error = EINVAL; goto badlink2; } if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { error = ELOOP; goto badlink2; } if (ndp->ni_pathlen > 1) cp = uma_zalloc(namei_zone, M_WAITOK); else cp = cnp->cn_pnbuf; aiov.iov_base = cp; aiov.iov_len = MAXPATHLEN; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = NULL; auio.uio_resid = MAXPATHLEN; error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); if (error) { badlink1: if (ndp->ni_pathlen > 1) uma_zfree(namei_zone, cp); badlink2: vrele(ndp->ni_dvp); vput(ndp->ni_vp); break; } linklen = MAXPATHLEN - auio.uio_resid; if (linklen == 0) { error = ENOENT; goto badlink1; } if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { error = ENAMETOOLONG; goto badlink1; } /* * Adjust or replace path */ if (ndp->ni_pathlen > 1) { NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); uma_zfree(namei_zone, cnp->cn_pnbuf); cnp->cn_pnbuf = cp; } else cnp->cn_pnbuf[linklen] = '\0'; ndp->ni_pathlen += linklen; /* * Cleanup refs for next loop and check if root directory * should replace current directory. Normally ni_dvp * becomes the new base directory and is cleaned up when * we loop. Explicitly null pointers after invalidation * to clarify operation. */ vput(ndp->ni_vp); ndp->ni_vp = NULL; if (cnp->cn_pnbuf[0] == '/') { vrele(ndp->ni_dvp); ndp->ni_dvp = ndp->ni_rootdir; VREF(ndp->ni_dvp); } ndp->ni_startdir = ndp->ni_dvp; ndp->ni_dvp = NULL; } if (!lockleaf) cnp->cn_flags &= ~LOCKLEAF; out: if (error) { nfsvno_relpathbuf(ndp); ndp->ni_vp = NULL; ndp->ni_dvp = NULL; ndp->ni_startdir = NULL; } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { ndp->ni_dvp = NULL; } out1: NFSEXITCODE2(error, nd); return (error); } /* * Set up a pathname buffer and return a pointer to it and, optionally * set a hash pointer. */ void nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) { struct componentname *cnp = &ndp->ni_cnd; cnp->cn_flags |= (NOMACCHECK | HASBUF); cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); if (hashpp != NULL) *hashpp = NULL; *bufpp = cnp->cn_pnbuf; } /* * Release the above path buffer, if not released by nfsvno_namei(). */ void nfsvno_relpathbuf(struct nameidata *ndp) { if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) panic("nfsrelpath"); uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); ndp->ni_cnd.cn_flags &= ~HASBUF; } /* * Readlink vnode op into an mbuf list. */ int nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) { struct iovec *iv; struct uio io, *uiop = &io; struct mbuf *mp, *mp3; int len, tlen, error = 0; len = NFS_MAXPATHLEN; if (maxextsiz > 0) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, &mp3, &mp, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); uiop->uio_iov = iv; uiop->uio_offset = 0; uiop->uio_resid = len; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = NULL; error = VOP_READLINK(vp, uiop, cred); free(iv, M_TEMP); if (error) { m_freem(mp3); *lenp = 0; goto out; } if (uiop->uio_resid > 0) { len -= uiop->uio_resid; tlen = NFSM_RNDUP(len); if (tlen == 0) { m_freem(mp3); mp3 = mp = NULL; } else if (tlen != NFS_MAXPATHLEN || tlen != len) mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len); } *lenp = len; *mpp = mp3; *mpendp = mp; out: NFSEXITCODE(error); return (error); } /* * Create an mbuf chain and an associated iovec that can be used to Read * or Getextattr of data. * Upon success, return pointers to the first and last mbufs in the chain * plus the malloc'd iovec and its iovlen. */ static int nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, struct iovec **ivp) { struct mbuf *m, *m2 = NULL, *m3; struct iovec *iv; int i, left, siz; left = len; m3 = NULL; /* * Generate the mbuf list with the uio_iov ref. to it. */ i = 0; while (left > 0) { NFSMGET(m); MCLGET(m, M_WAITOK); m->m_len = 0; siz = min(M_TRAILINGSPACE(m), left); left -= siz; i++; if (m3) m2->m_next = m; else m3 = m; m2 = m; } *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); m = m3; left = len; i = 0; while (left > 0) { if (m == NULL) panic("nfsrv_createiovec iov"); siz = min(M_TRAILINGSPACE(m), left); if (siz > 0) { iv->iov_base = mtod(m, caddr_t) + m->m_len; iv->iov_len = siz; m->m_len += siz; left -= siz; iv++; i++; } m = m->m_next; } *mpp = m3; *mpendp = m2; return (i); } /* * Create an mbuf chain and an associated iovec that can be used to Read * or Getextattr of data. * Upon success, return pointers to the first and last mbufs in the chain * plus the malloc'd iovec and its iovlen. * Same as above, but creates ext_pgs mbuf(s). */ static int nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, struct mbuf **mpendp, struct iovec **ivp) { struct mbuf *m, *m2 = NULL, *m3; struct iovec *iv; int i, left, pgno, siz; left = len; m3 = NULL; /* * Generate the mbuf list with the uio_iov ref. to it. */ i = 0; while (left > 0) { siz = min(left, maxextsiz); m = mb_alloc_ext_plus_pages(siz, M_WAITOK); left -= siz; i += m->m_epg_npgs; if (m3 != NULL) m2->m_next = m; else m3 = m; m2 = m; } *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); m = m3; left = len; i = 0; pgno = 0; while (left > 0) { if (m == NULL) panic("nfsvno_createiovec_extpgs iov"); siz = min(PAGE_SIZE, left); if (siz > 0) { iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); iv->iov_len = siz; m->m_len += siz; if (pgno == m->m_epg_npgs - 1) m->m_epg_last_len = siz; left -= siz; iv++; i++; pgno++; } if (pgno == m->m_epg_npgs && left > 0) { m = m->m_next; if (m == NULL) panic("nfsvno_createiovec_extpgs iov"); pgno = 0; } } *mpp = m3; *mpendp = m2; return (i); } /* * Read vnode op call into mbuf list. */ int nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp) { struct mbuf *m; struct iovec *iv; int error = 0, len, tlen, ioflag = 0; struct mbuf *m3; struct uio io, *uiop = &io; struct nfsheur *nh; /* * Attempt to read from a DS file. A return of ENOENT implies * there is no DS file to read. */ error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, NULL, mpendp, NULL, NULL, NULL, 0, NULL); if (error != ENOENT) return (error); len = NFSM_RNDUP(cnt); if (maxextsiz > 0) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, &m3, &m, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); uiop->uio_iov = iv; uiop->uio_offset = off; uiop->uio_resid = len; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = NULL; nh = nfsrv_sequential_heuristic(uiop, vp); ioflag |= nh->nh_seqcount << IO_SEQSHIFT; /* XXX KDM make this more systematic? */ - nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; + nfsstatsv1_p->srvbytes[NFSV4OP_READ] += uiop->uio_resid; error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); free(iv, M_TEMP); if (error) { m_freem(m3); *mpp = NULL; goto out; } nh->nh_nextoff = uiop->uio_offset; tlen = len - uiop->uio_resid; cnt = cnt < tlen ? cnt : tlen; tlen = NFSM_RNDUP(cnt); if (tlen == 0) { m_freem(m3); m3 = m = NULL; } else if (len != tlen || tlen != cnt) m = nfsrv_adj(m3, len - tlen, tlen - cnt); *mpp = m3; *mpendp = m; out: NFSEXITCODE(error); return (error); } /* * Create the iovec for the mbuf chain passed in as an argument. * The "cp" argument is where the data starts within the first mbuf in * the chain. It returns the iovec and the iovcnt. */ static int nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, int *iovcntp) { struct mbuf *mp; struct iovec *ivp; int cnt, i, len; /* * Loop through the mbuf chain, counting how many mbufs are a * part of this write operation, so the iovec size is known. */ cnt = 0; len = retlen; mp = m; i = mtod(mp, caddr_t) + mp->m_len - cp; while (len > 0) { if (i > 0) { len -= i; cnt++; } mp = mp->m_next; if (!mp) { if (len > 0) return (EBADRPC); } else i = mp->m_len; } /* Now, create the iovec. */ mp = m; *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK); *iovcntp = cnt; i = mtod(mp, caddr_t) + mp->m_len - cp; len = retlen; while (len > 0) { if (mp == NULL) panic("nfsrv_createiovecw"); if (i > 0) { i = min(i, len); ivp->iov_base = cp; ivp->iov_len = i; ivp++; len -= i; } mp = mp->m_next; if (mp) { i = mp->m_len; cp = mtod(mp, caddr_t); } } return (0); } /* * Write vnode op from an mbuf list. */ int nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) { struct iovec *iv; int cnt, ioflags, error; struct uio io, *uiop = &io; struct nfsheur *nh; /* * Attempt to write to a DS file. A return of ENOENT implies * there is no DS file to write. */ error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); if (error != ENOENT) { *stable = NFSWRITE_FILESYNC; return (error); } if (*stable == NFSWRITE_UNSTABLE) ioflags = IO_NODELOCKED; else ioflags = (IO_SYNC | IO_NODELOCKED); error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); if (error != 0) return (error); uiop->uio_iov = iv; uiop->uio_iovcnt = cnt; uiop->uio_resid = retlen; uiop->uio_rw = UIO_WRITE; uiop->uio_segflg = UIO_SYSSPACE; NFSUIOPROC(uiop, p); uiop->uio_offset = off; nh = nfsrv_sequential_heuristic(uiop, vp); ioflags |= nh->nh_seqcount << IO_SEQSHIFT; /* XXX KDM make this more systematic? */ - nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; + nfsstatsv1_p->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; error = VOP_WRITE(vp, uiop, ioflags, cred); if (error == 0) nh->nh_nextoff = uiop->uio_offset; free(iv, M_TEMP); NFSEXITCODE(error); return (error); } /* * Common code for creating a regular file (plus special files for V2). */ int nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) { u_quad_t tempsize; int error; struct thread *p = curthread; error = nd->nd_repstat; if (!error && ndp->ni_vp == NULL) { if (nvap->na_type == VREG || nvap->na_type == VSOCK) { vrele(ndp->ni_startdir); error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); /* For a pNFS server, create the data file on a DS. */ if (error == 0 && nvap->na_type == VREG) { /* * Create a data file on a DS for a pNFS server. * This function just returns if not * running a pNFS DS or the creation fails. */ nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, nd->nd_cred, p); } VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); if (!error) { if (*exclusive_flagp) { *exclusive_flagp = 0; NFSVNO_ATTRINIT(nvap); nvap->na_atime.tv_sec = cverf[0]; nvap->na_atime.tv_nsec = cverf[1]; error = VOP_SETATTR(ndp->ni_vp, &nvap->na_vattr, nd->nd_cred); if (error != 0) { vput(ndp->ni_vp); ndp->ni_vp = NULL; error = NFSERR_NOTSUPP; } } } /* * NFS V2 Only. nfsrvd_mknod() does this for V3. * (This implies, just get out on an error.) */ } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || nvap->na_type == VFIFO) { if (nvap->na_type == VCHR && rdev == 0xffffffff) nvap->na_type = VFIFO; if (nvap->na_type != VFIFO && (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); goto out; } nvap->na_rdev = rdev; error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); vrele(ndp->ni_startdir); if (error) goto out; } else { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = ENXIO; goto out; } *vpp = ndp->ni_vp; } else { /* * Handle cases where error is already set and/or * the file exists. * 1 - clean up the lookup * 2 - iff !error and na_size set, truncate it */ vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); *vpp = ndp->ni_vp; if (ndp->ni_dvp == *vpp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (!error && nvap->na_size != VNOVAL) { error = nfsvno_accchk(*vpp, VWRITE, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (!error) { tempsize = nvap->na_size; NFSVNO_ATTRINIT(nvap); nvap->na_size = tempsize; error = nfsvno_setattr(*vpp, nvap, nd->nd_cred, p, exp); } } if (error) vput(*vpp); } out: NFSEXITCODE(error); return (error); } /* * Do a mknod vnode op. */ int nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, struct thread *p) { int error = 0; enum vtype vtyp; vtyp = nvap->na_type; /* * Iff doesn't exist, create it. */ if (ndp->ni_vp) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = NFSERR_BADTYPE; goto out; } if (vtyp == VSOCK) { vrele(ndp->ni_startdir); error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); } else { if (nvap->na_type != VFIFO && (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); goto out; } error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); vrele(ndp->ni_startdir); /* * Since VOP_MKNOD returns the ni_vp, I can't * see any reason to do the lookup. */ } out: NFSEXITCODE(error); return (error); } /* * Mkdir vnode op. */ int nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { int error = 0; if (ndp->ni_vp != NULL) { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); nfsvno_relpathbuf(ndp); error = EEXIST; goto out; } error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); out: NFSEXITCODE(error); return (error); } /* * symlink vnode op. */ int nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { int error = 0; if (ndp->ni_vp) { vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr, pathcp); /* * Although FreeBSD still had the lookup code in * it for 7/current, there doesn't seem to be any * point, since VOP_SYMLINK() returns the ni_vp. * Just vput it for v2. */ VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); out: NFSEXITCODE(error); return (error); } /* * Parse symbolic link arguments. * This function has an ugly side effect. It will malloc() an area for * the symlink and set iov_base to point to it, only if it succeeds. * So, if it returns with uiop->uio_iov->iov_base != NULL, that must * be FREE'd later. */ int nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, struct thread *p, char **pathcpp, int *lenp) { u_int32_t *tl; char *pathcp = NULL; int error = 0, len; struct nfsv2_sattr *sp; *pathcpp = NULL; *lenp = 0; if ((nd->nd_flag & ND_NFSV3) && (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len > NFS_MAXPATHLEN || len <= 0) { error = EBADRPC; goto nfsmout; } pathcp = malloc(len + 1, M_TEMP, M_WAITOK); error = nfsrv_mtostr(nd, pathcp, len); if (error) goto nfsmout; if (nd->nd_flag & ND_NFSV2) { NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); } *pathcpp = pathcp; *lenp = len; NFSEXITCODE2(0, nd); return (0); nfsmout: if (pathcp) free(pathcp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Remove a non-directory object. */ int nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; int error = 0, mirrorcnt; char fname[PNFS_FILENAME_LEN + 1]; fhandle_t fh; vp = ndp->ni_vp; dsdvp[0] = NULL; if (vp->v_type == VDIR) error = NFSERR_ISDIR; else if (is_v4) error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), p); if (error == 0) nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); if (!error) error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); if (error == 0 && dsdvp[0] != NULL) nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vput(vp); if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Remove a directory. */ int nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *vp; int error = 0; vp = ndp->ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (ndp->ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) error = EBUSY; out: if (!error) error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); vput(vp); if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Rename vnode op. */ int nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) { struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; int error = 0, mirrorcnt; char fname[PNFS_FILENAME_LEN + 1]; fhandle_t fh; dsdvp[0] = NULL; fvp = fromndp->ni_vp; if (ndstat) { vrele(fromndp->ni_dvp); vrele(fvp); error = ndstat; goto out1; } tdvp = tondp->ni_dvp; tvp = tondp->ni_vp; if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; goto out; } if (tvp->v_type == VDIR && tvp->v_mountedhere) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } /* * A rename to '.' or '..' results in a prematurely * unlocked vnode on FreeBSD5, so I'm just going to fail that * here. */ if ((tondp->ni_cnd.cn_namelen == 1 && tondp->ni_cnd.cn_nameptr[0] == '.') || (tondp->ni_cnd.cn_namelen == 2 && tondp->ni_cnd.cn_nameptr[0] == '.' && tondp->ni_cnd.cn_nameptr[1] == '.')) { error = EINVAL; goto out; } } if (fvp->v_type == VDIR && fvp->v_mountedhere) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } if (fvp->v_mount != tdvp->v_mount) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; goto out; } if (fvp == tdvp) { error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; goto out; } if (fvp == tvp) { /* * If source and destination are the same, there is nothing to * do. Set error to -1 to indicate this. */ error = -1; goto out; } if (ndflag & ND_NFSV4) { if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { error = nfsrv_checkremove(fvp, 0, NULL, (nfsquad_t)((u_quad_t)0), p); NFSVOPUNLOCK(fvp); } else error = EPERM; if (tvp && !error) error = nfsrv_checkremove(tvp, 1, NULL, (nfsquad_t)((u_quad_t)0), p); } else { /* * For NFSv2 and NFSv3, try to get rid of the delegation, so * that the NFSv4 client won't be confused by the rename. * Since nfsd_recalldelegation() can only be called on an * unlocked vnode at this point and fvp is the file that will * still exist after the rename, just do fvp. */ nfsd_recalldelegation(fvp, p); } if (error == 0 && tvp != NULL) { nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" " dsdvp=%p\n", dsdvp[0]); } out: if (!error) { error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, &tondp->ni_cnd); } else { if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fromndp->ni_dvp); vrele(fvp); if (error == -1) error = 0; } /* * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and * if the rename succeeded, the DS file for the tvp needs to be * removed. */ if (error == 0 && dsdvp[0] != NULL) { nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); } vrele(tondp->ni_startdir); nfsvno_relpathbuf(tondp); out1: vrele(fromndp->ni_startdir); nfsvno_relpathbuf(fromndp); NFSEXITCODE(error); return (error); } /* * Link vnode op. */ int nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, struct thread *p, struct nfsexstuff *exp) { struct vnode *xp; int error = 0; xp = ndp->ni_vp; if (xp != NULL) { error = EEXIST; } else { xp = ndp->ni_dvp; if (vp->v_mount != xp->v_mount) error = EXDEV; } if (!error) { NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (!VN_IS_DOOMED(vp)) error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); else error = EPERM; if (ndp->ni_dvp == vp) { vrele(ndp->ni_dvp); NFSVOPUNLOCK(vp); } else { vref(vp); VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); } } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (ndp->ni_vp) vrele(ndp->ni_vp); } nfsvno_relpathbuf(ndp); NFSEXITCODE(error); return (error); } /* * Do the fsync() appropriate for the commit. */ int nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, struct thread *td) { int error = 0; /* * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of * file is done. At this time VOP_FSYNC does not accept offset and * byte count parameters so call VOP_FSYNC the whole file for now. * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. * File systems that do not use the buffer cache (as indicated * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). */ if (cnt == 0 || cnt > MAX_COMMIT_COUNT || (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { /* * Give up and do the whole thing */ if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); } else { /* * Locate and synchronously write any buffers that fall * into the requested range. Note: we are assuming that * f_iosize is a power of 2. */ int iosize = vp->v_mount->mnt_stat.f_iosize; int iomask = iosize - 1; struct bufobj *bo; daddr_t lblkno; /* * Align to iosize boundary, super-align to page boundary. */ if (off & iomask) { cnt += off & iomask; off &= ~(u_quad_t)iomask; } if (off & PAGE_MASK) { cnt += off & PAGE_MASK; off &= ~(u_quad_t)PAGE_MASK; } lblkno = off / iosize; if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, off, off + cnt, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } bo = &vp->v_bufobj; BO_LOCK(bo); while (cnt > 0) { struct buf *bp; /* * If we have a buffer and it is marked B_DELWRI we * have to lock and write it. Otherwise the prior * write is assumed to have already been committed. * * gbincore() can return invalid buffers now so we * have to check that bit as well (though B_DELWRI * should not be set if B_INVAL is set there could be * a race here since we haven't locked the buffer). */ if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); continue; /* retry */ } if ((bp->b_flags & (B_DELWRI|B_INVAL)) == B_DELWRI) { bremfree(bp); bp->b_flags &= ~B_ASYNC; bwrite(bp); ++nfs_commit_miss; } else BUF_UNLOCK(bp); BO_LOCK(bo); } ++nfs_commit_blks; if (cnt < iosize) break; cnt -= iosize; ++lblkno; } BO_UNLOCK(bo); } NFSEXITCODE(error); return (error); } /* * Statfs vnode op. */ int nfsvno_statfs(struct vnode *vp, struct statfs *sf) { struct statfs *tsf; int error; tsf = NULL; if (nfsrv_devidcnt > 0) { /* For a pNFS service, get the DS numbers. */ tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); error = nfsrv_pnfsstatfs(tsf, vp->v_mount); if (error != 0) { free(tsf, M_TEMP); tsf = NULL; } } error = VFS_STATFS(vp->v_mount, sf); if (error == 0) { if (tsf != NULL) { sf->f_blocks = tsf->f_blocks; sf->f_bavail = tsf->f_bavail; sf->f_bfree = tsf->f_bfree; sf->f_bsize = tsf->f_bsize; } /* * Since NFS handles these values as unsigned on the * wire, there is no way to represent negative values, * so set them to 0. Without this, they will appear * to be very large positive values for clients like * Solaris10. */ if (sf->f_bavail < 0) sf->f_bavail = 0; if (sf->f_ffree < 0) sf->f_ffree = 0; } free(tsf, M_TEMP); NFSEXITCODE(error); return (error); } /* * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but * must handle nfsrv_opencheck() calls after any other access checks. */ void nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct nfsexstuff *exp, struct vnode **vpp) { struct vnode *vp = NULL; u_quad_t tempsize; struct nfsexstuff nes; struct thread *p = curthread; if (ndp->ni_vp == NULL) nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, NULL, nd, p, nd->nd_repstat); if (!nd->nd_repstat) { if (ndp->ni_vp == NULL) { vrele(ndp->ni_startdir); nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); /* For a pNFS server, create the data file on a DS. */ if (nd->nd_repstat == 0) { /* * Create a data file on a DS for a pNFS server. * This function just returns if not * running a pNFS DS or the creation fails. */ nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, cred, p); } VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? &ndp->ni_vp : NULL, false); nfsvno_relpathbuf(ndp); if (!nd->nd_repstat) { if (*exclusive_flagp) { *exclusive_flagp = 0; NFSVNO_ATTRINIT(nvap); nvap->na_atime.tv_sec = cverf[0]; nvap->na_atime.tv_nsec = cverf[1]; nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, &nvap->na_vattr, cred); if (nd->nd_repstat != 0) { vput(ndp->ni_vp); ndp->ni_vp = NULL; nd->nd_repstat = NFSERR_NOTSUPP; } else NFSSETBIT_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS); } else { nfsrv_fixattr(nd, ndp->ni_vp, nvap, aclp, p, attrbitp, exp); } } vp = ndp->ni_vp; } else { if (ndp->ni_startdir) vrele(ndp->ni_startdir); nfsvno_relpathbuf(ndp); vp = ndp->ni_vp; if (create == NFSV4OPEN_CREATE) { if (ndp->ni_dvp == vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); } if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { if (ndp->ni_cnd.cn_flags & RDONLY) NFSVNO_SETEXRDONLY(&nes); else NFSVNO_EXINIT(&nes); nd->nd_repstat = nfsvno_accchk(vp, VWRITE, cred, &nes, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, vp, nd, p, nd->nd_repstat); if (!nd->nd_repstat) { tempsize = nvap->na_size; NFSVNO_ATTRINIT(nvap); nvap->na_size = tempsize; nd->nd_repstat = nfsvno_setattr(vp, nvap, cred, p, exp); } } else if (vp->v_type == VREG) { nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, vp, nd, p, nd->nd_repstat); } } } else { if (ndp->ni_cnd.cn_flags & HASBUF) nfsvno_relpathbuf(ndp); if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { vrele(ndp->ni_startdir); if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); if (ndp->ni_vp) vput(ndp->ni_vp); } } *vpp = vp; NFSEXITCODE2(0, nd); } /* * Updates the file rev and sets the mtime and ctime * to the current clock time, returning the va_filerev and va_Xtime * values. * Return ESTALE to indicate the vnode is VIRF_DOOMED. */ int nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, struct nfsrv_descript *nd, struct thread *p) { struct vattr va; VATTR_NULL(&va); vfs_timestamp(&va.va_mtime); if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED(vp)) return (ESTALE); } (void) VOP_SETATTR(vp, &va, nd->nd_cred); (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); return (0); } /* * Glue routine to nfsv4_fillattr(). */ int nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p, int isdgram, int reterr, int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) { struct statfs *sf; int error; sf = NULL; if (nfsrv_devidcnt > 0 && (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); error = nfsrv_pnfsstatfs(sf, mp); if (error != 0) { free(sf, M_TEMP); sf = NULL; } } error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, mounted_on_fileno, sf); free(sf, M_TEMP); NFSEXITCODE2(0, nd); return (error); } /* Since the Readdir vnode ops vary, put the entire functions in here. */ /* * nfs readdir service * - mallocs what it thinks is enough to read * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR * - calls VOP_READDIR() * - loops around building the reply * if the output generated exceeds count break out of loop * The NFSM_CLGET macro is used here so that the reply will be packed * tightly in mbuf clusters. * - it trims out records with d_fileno == 0 * this doesn't matter for Unix clients, but they might confuse clients * for other os'. * - it trims out records with d_type == DT_WHT * these cannot be seen through NFS (unless we extend the protocol) * The alternate call nfsrvd_readdirplus() does lookups as well. * PS: The NFS protocol spec. does not clarify what the "count" byte * argument is a count of.. just name strings and file id's or the * entire reply rpc or ... * I tried just file name and id sizes and it confused the Sun client, * so I am using the full rpc size now. The "paranoia.." comment refers * to including the status longwords that are not a part of the dir. * "entry" structures, but are in the rpc. */ int nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, struct vnode *vp, struct nfsexstuff *exp) { struct dirent *dp; u_int32_t *tl; int dirlen; char *cpos, *cend, *rbuf; struct nfsvattr at; int nlen, error = 0, getret = 1; int siz, cnt, fullsiz, eofflag, ncookies; u_int64_t off, toff, verf __unused; u_long *cookies = NULL, *cookiep; struct uio io; struct iovec iv; int is_ufs; struct thread *p = curthread; if (nd->nd_repstat) { nfsrv_postopattr(nd, getret, &at); goto out; } if (nd->nd_flag & ND_NFSV2) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); off = fxdr_unsigned(u_quad_t, *tl++); } else { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); off = fxdr_hyper(tl); tl += 2; verf = fxdr_hyper(tl); tl += 2; } toff = off; cnt = fxdr_unsigned(int, *tl); if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) cnt = NFS_SRVMAXDATA(nd); siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); fullsiz = siz; if (nd->nd_flag & ND_NFSV3) { nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); #if 0 /* * va_filerev is not sufficient as a cookie verifier, * since it is not supposed to change when entries are * removed/added unless that offset cookies returned to * the client are no longer valid. */ if (!nd->nd_repstat && toff && verf != at.na_filerev) nd->nd_repstat = NFSERR_BAD_COOKIE; #endif } if (!nd->nd_repstat && vp->v_type != VDIR) nd->nd_repstat = NFSERR_NOTDIR; if (nd->nd_repstat == 0 && cnt == 0) { if (nd->nd_flag & ND_NFSV2) /* NFSv2 does not have NFSERR_TOOSMALL */ nd->nd_repstat = EPERM; else nd->nd_repstat = NFSERR_TOOSMALL; } if (!nd->nd_repstat) nd->nd_repstat = nfsvno_accchk(vp, VEXEC, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (nd->nd_repstat) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; rbuf = malloc(siz, M_TEMP, M_WAITOK); again: eofflag = 0; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } iv.iov_base = rbuf; iv.iov_len = siz; io.uio_iov = &iv; io.uio_iovcnt = 1; io.uio_offset = (off_t)off; io.uio_resid = siz; io.uio_segflg = UIO_SYSSPACE; io.uio_rw = UIO_READ; io.uio_td = NULL; nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, &cookies); off = (u_int64_t)io.uio_offset; if (io.uio_resid) siz -= io.uio_resid; if (!cookies && !nd->nd_repstat) nd->nd_repstat = NFSERR_PERM; if (nd->nd_flag & ND_NFSV3) { getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); if (!nd->nd_repstat) nd->nd_repstat = getret; } /* * Handles the failed cases. nd->nd_repstat == 0 past here. */ if (nd->nd_repstat) { vput(vp); free(rbuf, M_TEMP); if (cookies) free(cookies, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * If nothing read, return eof * rpc reply */ if (siz == 0) { vput(vp); if (nd->nd_flag & ND_NFSV2) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); } else { nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); tl += 2; } *tl++ = newnfs_false; *tl = newnfs_true; free(rbuf, M_TEMP); free(cookies, M_TEMP); goto out; } /* * Check for degenerate cases of nothing useful read. * If so go try again */ cpos = rbuf; cend = rbuf + siz; dp = (struct dirent *)cpos; cookiep = cookies; /* * For some reason FreeBSD's ufs_readdir() chooses to back the * directory offset up to a block boundary, so it is necessary to * skip over the records that precede the requested offset. This * requires the assumption that file offset cookies monotonically * increase. */ while (cpos < cend && ncookies > 0 && (dp->d_fileno == 0 || dp->d_type == DT_WHT || (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos >= cend || ncookies == 0) { siz = fullsiz; toff = off; goto again; } vput(vp); /* * If cnt > MCLBYTES and the reply will not be saved, use * ext_pgs mbufs for TLS. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. */ if (cnt > MCLBYTES && siz > MCLBYTES && (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) nd->nd_flag |= ND_EXTPG; /* * dirlen is the size of the reply, including all XDR and must * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate * if the XDR should be included in "count", but to be safe, we do. * (Include the two booleans at the end of the reply in dirlen now.) */ if (nd->nd_flag & ND_NFSV3) { nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; } else { dirlen = 2 * NFSX_UNSIGNED; } /* Loop through the records and build reply */ while (cpos < cend && ncookies > 0) { nlen = dp->d_namlen; if (dp->d_fileno != 0 && dp->d_type != DT_WHT && nlen <= NFS_MAXNAMLEN) { if (nd->nd_flag & ND_NFSV3) dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); else dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); if (dirlen > cnt) { eofflag = 0; break; } /* * Build the directory record xdr from * the dirent entry. */ if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; } else { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_true; } *tl = txdr_unsigned(dp->d_fileno); (void) nfsm_strtom(nd, dp->d_name, nlen); if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); txdr_hyper(*cookiep, tl); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(*cookiep); } } cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos < cend) eofflag = 0; NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_false; if (eofflag) *tl = newnfs_true; else *tl = newnfs_false; free(rbuf, M_TEMP); free(cookies, M_TEMP); out: NFSEXITCODE2(0, nd); return (0); nfsmout: vput(vp); NFSEXITCODE2(error, nd); return (error); } /* * Readdirplus for V3 and Readdir for V4. */ int nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, struct vnode *vp, struct nfsexstuff *exp) { struct dirent *dp; u_int32_t *tl; int dirlen; char *cpos, *cend, *rbuf; struct vnode *nvp; fhandle_t nfh; struct nfsvattr nva, at, *nvap = &nva; struct mbuf *mb0, *mb1; struct nfsreferral *refp; int nlen, r, error = 0, getret = 1, usevget = 1; int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; caddr_t bpos0, bpos1; u_int64_t off, toff, verf; u_long *cookies = NULL, *cookiep; nfsattrbit_t attrbits, rderrbits, savbits; struct uio io; struct iovec iv; struct componentname cn; int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; struct mount *mp, *new_mp; uint64_t mounted_on_fileno; struct thread *p = curthread; int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; if (nd->nd_repstat) { nfsrv_postopattr(nd, getret, &at); goto out; } NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); off = fxdr_hyper(tl); toff = off; tl += 2; verf = fxdr_hyper(tl); tl += 2; siz = fxdr_unsigned(int, *tl++); cnt = fxdr_unsigned(int, *tl); /* * Use the server's maximum data transfer size as the upper bound * on reply datalen. */ if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) cnt = NFS_SRVMAXDATA(nd); /* * siz is a "hint" of how much directory information (name, fileid, * cookie) should be in the reply. At least one client "hints" 0, * so I set it to cnt for that case. I also round it up to the * next multiple of DIRBLKSIZ. * Since the size of a Readdirplus directory entry reply will always * be greater than a directory entry returned by VOP_READDIR(), it * does not make sense to read more than NFS_SRVMAXDATA() via * VOP_READDIR(). */ if (siz <= 0) siz = cnt; else if (siz > NFS_SRVMAXDATA(nd)) siz = NFS_SRVMAXDATA(nd); siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); if (nd->nd_flag & ND_NFSV4) { error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error) goto nfsmout; NFSSET_ATTRBIT(&savbits, &attrbits); NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); NFSZERO_ATTRBIT(&rderrbits); NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); } else { NFSZERO_ATTRBIT(&attrbits); } fullsiz = siz; nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); #if 0 if (!nd->nd_repstat) { if (off && verf != at.na_filerev) { /* * va_filerev is not sufficient as a cookie verifier, * since it is not supposed to change when entries are * removed/added unless that offset cookies returned to * the client are no longer valid. */ if (nd->nd_flag & ND_NFSV4) { nd->nd_repstat = NFSERR_NOTSAME; } else { nd->nd_repstat = NFSERR_BAD_COOKIE; } } } #endif if (!nd->nd_repstat && vp->v_type != VDIR) nd->nd_repstat = NFSERR_NOTDIR; if (!nd->nd_repstat && cnt == 0) nd->nd_repstat = NFSERR_TOOSMALL; if (!nd->nd_repstat) nd->nd_repstat = nfsvno_accchk(vp, VEXEC, nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, NFSACCCHK_VPISLOCKED, NULL); if (nd->nd_repstat) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; rbuf = malloc(siz, M_TEMP, M_WAITOK); again: eofflag = 0; if (cookies) { free(cookies, M_TEMP); cookies = NULL; } iv.iov_base = rbuf; iv.iov_len = siz; io.uio_iov = &iv; io.uio_iovcnt = 1; io.uio_offset = (off_t)off; io.uio_resid = siz; io.uio_segflg = UIO_SYSSPACE; io.uio_rw = UIO_READ; io.uio_td = NULL; nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, &cookies); off = (u_int64_t)io.uio_offset; if (io.uio_resid) siz -= io.uio_resid; getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); if (!cookies && !nd->nd_repstat) nd->nd_repstat = NFSERR_PERM; if (!nd->nd_repstat) nd->nd_repstat = getret; if (nd->nd_repstat) { vput(vp); if (cookies) free(cookies, M_TEMP); free(rbuf, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * If nothing read, return eof * rpc reply */ if (siz == 0) { vput(vp); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); txdr_hyper(at.na_filerev, tl); tl += 2; *tl++ = newnfs_false; *tl = newnfs_true; free(cookies, M_TEMP); free(rbuf, M_TEMP); goto out; } /* * Check for degenerate cases of nothing useful read. * If so go try again */ cpos = rbuf; cend = rbuf + siz; dp = (struct dirent *)cpos; cookiep = cookies; /* * For some reason FreeBSD's ufs_readdir() chooses to back the * directory offset up to a block boundary, so it is necessary to * skip over the records that precede the requested offset. This * requires the assumption that file offset cookies monotonically * increase. */ while (cpos < cend && ncookies > 0 && (dp->d_fileno == 0 || dp->d_type == DT_WHT || (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || ((nd->nd_flag & ND_NFSV4) && ((dp->d_namlen == 1 && dp->d_name[0] == '.') || (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } if (cpos >= cend || ncookies == 0) { siz = fullsiz; toff = off; goto again; } /* * Busy the file system so that the mount point won't go away * and, as such, VFS_VGET() can be used safely. */ mp = vp->v_mount; vfs_ref(mp); NFSVOPUNLOCK(vp); nd->nd_repstat = vfs_busy(mp, 0); vfs_rel(mp); if (nd->nd_repstat != 0) { vrele(vp); free(cookies, M_TEMP); free(rbuf, M_TEMP); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); goto out; } /* * Check to see if entries in this directory can be safely acquired * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. * ZFS snapshot directories need VOP_LOOKUP(), so that any * automount of the snapshot directory that is required will * be done. * This needs to be done here for NFSv4, since NFSv4 never does * a VFS_VGET() for "." or "..". */ if (is_zfs == 1) { r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); if (r == EOPNOTSUPP) { usevget = 0; cn.cn_nameiop = LOOKUP; cn.cn_lkflags = LK_SHARED | LK_RETRY; cn.cn_cred = nd->nd_cred; cn.cn_thread = p; } else if (r == 0) vput(nvp); } /* * If the reply is likely to exceed MCLBYTES and the reply will * not be saved, use ext_pgs mbufs for TLS. * It is difficult to predict how large each entry will be and * how many entries have been read, so just assume the directory * entries grow by a factor of 4 when attributes are included. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. */ if (cnt > MCLBYTES && siz > MCLBYTES / 4 && (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) nd->nd_flag |= ND_EXTPG; /* * Save this position, in case there is an error before one entry * is created. */ mb0 = nd->nd_mb; bpos0 = nd->nd_bpos; bextpg0 = nd->nd_bextpg; bextpgsiz0 = nd->nd_bextpgsiz; /* * Fill in the first part of the reply. * dirlen is the reply length in bytes and cannot exceed cnt. * (Include the two booleans at the end of the reply in dirlen now, * so we recognize when we have exceeded cnt.) */ if (nd->nd_flag & ND_NFSV3) { dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; nfsrv_postopattr(nd, getret, &at); } else { dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; } NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); txdr_hyper(at.na_filerev, tl); /* * Save this position, in case there is an empty reply needed. */ mb1 = nd->nd_mb; bpos1 = nd->nd_bpos; bextpg1 = nd->nd_bextpg; bextpgsiz1 = nd->nd_bextpgsiz; /* Loop through the records and build reply */ entrycnt = 0; while (cpos < cend && ncookies > 0 && dirlen < cnt) { nlen = dp->d_namlen; if (dp->d_fileno != 0 && dp->d_type != DT_WHT && nlen <= NFS_MAXNAMLEN && ((nd->nd_flag & ND_NFSV3) || nlen > 2 || (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) || (nlen == 1 && dp->d_name[0] != '.'))) { /* * Save the current position in the reply, in case * this entry exceeds cnt. */ mb1 = nd->nd_mb; bpos1 = nd->nd_bpos; bextpg1 = nd->nd_bextpg; bextpgsiz1 = nd->nd_bextpgsiz; /* * For readdir_and_lookup get the vnode using * the file number. */ nvp = NULL; refp = NULL; r = 0; at_root = 0; needs_unbusy = 0; new_mp = mp; mounted_on_fileno = (uint64_t)dp->d_fileno; if ((nd->nd_flag & ND_NFSV3) || NFSNONZERO_ATTRBIT(&savbits)) { if (nd->nd_flag & ND_NFSV4) refp = nfsv4root_getreferral(NULL, vp, dp->d_fileno); if (refp == NULL) { if (usevget) r = VFS_VGET(mp, dp->d_fileno, LK_SHARED, &nvp); else r = EOPNOTSUPP; if (r == EOPNOTSUPP) { if (usevget) { usevget = 0; cn.cn_nameiop = LOOKUP; cn.cn_lkflags = LK_SHARED | LK_RETRY; cn.cn_cred = nd->nd_cred; cn.cn_thread = p; } cn.cn_nameptr = dp->d_name; cn.cn_namelen = nlen; cn.cn_flags = ISLASTCN | NOFOLLOW | LOCKLEAF; if (nlen == 2 && dp->d_name[0] == '.' && dp->d_name[1] == '.') cn.cn_flags |= ISDOTDOT; if (NFSVOPLOCK(vp, LK_SHARED) != 0) { nd->nd_repstat = EPERM; break; } if ((vp->v_vflag & VV_ROOT) != 0 && (cn.cn_flags & ISDOTDOT) != 0) { vref(vp); nvp = vp; r = 0; } else { r = VOP_LOOKUP(vp, &nvp, &cn); if (vp != nvp) NFSVOPUNLOCK(vp); } } /* * For NFSv4, check to see if nvp is * a mount point and get the mount * point vnode, as required. */ if (r == 0 && nfsrv_enable_crossmntpt != 0 && (nd->nd_flag & ND_NFSV4) != 0 && nvp->v_type == VDIR && nvp->v_mountedhere != NULL) { new_mp = nvp->v_mountedhere; r = vfs_busy(new_mp, 0); vput(nvp); nvp = NULL; if (r == 0) { r = VFS_ROOT(new_mp, LK_SHARED, &nvp); needs_unbusy = 1; if (r == 0) at_root = 1; } } } /* * If we failed to look up the entry, then it * has become invalid, most likely removed. */ if (r != 0) { if (needs_unbusy) vfs_unbusy(new_mp); goto invalid; } KASSERT(refp != NULL || nvp != NULL, ("%s: undetected lookup error", __func__)); if (refp == NULL && ((nd->nd_flag & ND_NFSV3) || NFSNONZERO_ATTRBIT(&attrbits))) { r = nfsvno_getfh(nvp, &nfh, p); if (!r) r = nfsvno_getattr(nvp, nvap, nd, p, 1, &attrbits); if (r == 0 && is_zfs == 1 && nfsrv_enable_crossmntpt != 0 && (nd->nd_flag & ND_NFSV4) != 0 && nvp->v_type == VDIR && vp->v_mount != nvp->v_mount) { /* * For a ZFS snapshot, there is a * pseudo mount that does not set * v_mountedhere, so it needs to * be detected via a different * mount structure. */ at_root = 1; if (new_mp == mp) new_mp = nvp->v_mount; } } /* * If we failed to get attributes of the entry, * then just skip it for NFSv3 (the traditional * behavior in the old NFS server). * For NFSv4 the behavior is controlled by * RDATTRERROR: we either ignore the error or * fail the request. * The exception is EOPNOTSUPP, which can be * returned by nfsvno_getfh() for certain * file systems, such as devfs. This indicates * that the file system cannot be exported, * so just skip over the entry. * Note that RDATTRERROR is never set for NFSv3. */ if (r != 0) { if (!NFSISSET_ATTRBIT(&attrbits, NFSATTRBIT_RDATTRERROR) || r == EOPNOTSUPP) { vput(nvp); if (needs_unbusy != 0) vfs_unbusy(new_mp); if ((nd->nd_flag & ND_NFSV3) || r == EOPNOTSUPP) goto invalid; nd->nd_repstat = r; break; } } } /* * Build the directory record xdr */ if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl++ = 0; *tl = txdr_unsigned(dp->d_fileno); dirlen += nfsm_strtom(nd, dp->d_name, nlen); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); txdr_hyper(*cookiep, tl); nfsrv_postopattr(nd, 0, nvap); dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); if (nvp != NULL) vput(nvp); } else { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; txdr_hyper(*cookiep, tl); dirlen += nfsm_strtom(nd, dp->d_name, nlen); if (nvp != NULL) { supports_nfsv4acls = nfs_supportsnfsv4acls(nvp); NFSVOPUNLOCK(nvp); } else supports_nfsv4acls = 0; if (refp != NULL) { dirlen += nfsrv_putreferralattr(nd, &savbits, refp, 0, &nd->nd_repstat); if (nd->nd_repstat) { if (nvp != NULL) vrele(nvp); if (needs_unbusy != 0) vfs_unbusy(new_mp); break; } } else if (r) { dirlen += nfsvno_fillattr(nd, new_mp, nvp, nvap, &nfh, r, &rderrbits, nd->nd_cred, p, isdgram, 0, supports_nfsv4acls, at_root, mounted_on_fileno); } else { dirlen += nfsvno_fillattr(nd, new_mp, nvp, nvap, &nfh, r, &attrbits, nd->nd_cred, p, isdgram, 0, supports_nfsv4acls, at_root, mounted_on_fileno); } if (nvp != NULL) vrele(nvp); dirlen += (3 * NFSX_UNSIGNED); } if (needs_unbusy != 0) vfs_unbusy(new_mp); if (dirlen <= cnt) entrycnt++; } invalid: cpos += dp->d_reclen; dp = (struct dirent *)cpos; cookiep++; ncookies--; } vrele(vp); vfs_unbusy(mp); /* * If dirlen > cnt, we must strip off the last entry. If that * results in an empty reply, report NFSERR_TOOSMALL. */ if (dirlen > cnt || nd->nd_repstat) { if (!nd->nd_repstat && entrycnt == 0) nd->nd_repstat = NFSERR_TOOSMALL; if (nd->nd_repstat) { nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); if (nd->nd_flag & ND_NFSV3) nfsrv_postopattr(nd, getret, &at); } else nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); eofflag = 0; } else if (cpos < cend) eofflag = 0; if (!nd->nd_repstat) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_false; if (eofflag) *tl = newnfs_true; else *tl = newnfs_false; } free(cookies, M_TEMP); free(rbuf, M_TEMP); out: NFSEXITCODE2(0, nd); return (0); nfsmout: vput(vp); NFSEXITCODE2(error, nd); return (error); } /* * Get the settable attributes out of the mbuf list. * (Return 0 or EBADRPC) */ int nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) { u_int32_t *tl; struct nfsv2_sattr *sp; int error = 0, toclient = 0; switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { case ND_NFSV2: NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); /* * Some old clients didn't fill in the high order 16bits. * --> check the low order 2 bytes for 0xffff */ if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) nvap->na_mode = nfstov_mode(sp->sa_mode); if (sp->sa_uid != newnfs_xdrneg1) nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); if (sp->sa_gid != newnfs_xdrneg1) nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); if (sp->sa_size != newnfs_xdrneg1) nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { #ifdef notyet fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); #else nvap->na_atime.tv_sec = fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); nvap->na_atime.tv_nsec = 0; #endif } if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); break; case ND_NFSV3: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_mode = nfstov_mode(*tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_uid = fxdr_unsigned(uid_t, *tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_gid = fxdr_unsigned(gid_t, *tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); nvap->na_size = fxdr_hyper(tl); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); switch (fxdr_unsigned(int, *tl)) { case NFSV3SATTRTIME_TOCLIENT: NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); fxdr_nfsv3time(tl, &nvap->na_atime); toclient = 1; break; case NFSV3SATTRTIME_TOSERVER: vfs_timestamp(&nvap->na_atime); nvap->na_vaflags |= VA_UTIMES_NULL; break; } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); switch (fxdr_unsigned(int, *tl)) { case NFSV3SATTRTIME_TOCLIENT: NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); fxdr_nfsv3time(tl, &nvap->na_mtime); nvap->na_vaflags &= ~VA_UTIMES_NULL; break; case NFSV3SATTRTIME_TOSERVER: vfs_timestamp(&nvap->na_mtime); if (!toclient) nvap->na_vaflags |= VA_UTIMES_NULL; break; } break; case ND_NFSV4: error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); } nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Handle the setable attributes for V4. * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. */ int nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) { u_int32_t *tl; int attrsum = 0; int i, j; int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; int moderet, toclient = 0; u_char *cp, namestr[NFSV4_SMALLSTR + 1]; uid_t uid; gid_t gid; u_short mode, mask; /* Same type as va_mode. */ struct vattr va; error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); if (error) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsize = fxdr_unsigned(int, *tl); /* * Loop around getting the setable attributes. If an unsupported * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. */ if (retnotsup) { nd->nd_repstat = NFSERR_ATTRNOTSUPP; bitpos = NFSATTRBIT_MAX; } else { bitpos = 0; } moderet = 0; for (; bitpos < NFSATTRBIT_MAX; bitpos++) { if (attrsum > attrsize) { error = NFSERR_BADXDR; goto nfsmout; } if (NFSISSET_ATTRBIT(attrbitp, bitpos)) switch (bitpos) { case NFSATTRBIT_SIZE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (vp != NULL && vp->v_type != VREG) { error = (vp->v_type == VDIR) ? NFSERR_ISDIR : NFSERR_INVAL; goto nfsmout; } nvap->na_size = fxdr_hyper(tl); attrsum += NFSX_HYPER; break; case NFSATTRBIT_ACL: error = nfsrv_dissectacl(nd, aclp, true, &aceerr, &aclsize, p); if (error) goto nfsmout; if (aceerr && !nd->nd_repstat) nd->nd_repstat = aceerr; attrsum += aclsize; break; case NFSATTRBIT_ARCHIVE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_HIDDEN: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_MIMETYPE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); break; case NFSATTRBIT_MODE: moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nvap->na_mode = nfstov_mode(*tl); attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_OWNER: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); j = fxdr_unsigned(int, *tl); if (j < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (j > NFSV4_SMALLSTR) cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); else cp = namestr; error = nfsrv_mtostr(nd, cp, j); if (error) { if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); goto nfsmout; } if (!nd->nd_repstat) { nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid); if (!nd->nd_repstat) nvap->na_uid = uid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); break; case NFSATTRBIT_OWNERGROUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); j = fxdr_unsigned(int, *tl); if (j < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (j > NFSV4_SMALLSTR) cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); else cp = namestr; error = nfsrv_mtostr(nd, cp, j); if (error) { if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); goto nfsmout; } if (!nd->nd_repstat) { nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid); if (!nd->nd_repstat) nvap->na_gid = gid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); break; case NFSATTRBIT_SYSTEM: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_TIMEACCESSSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_atime); toclient = 1; attrsum += NFSX_V4TIME; } else { vfs_timestamp(&nvap->na_atime); nvap->na_vaflags |= VA_UTIMES_NULL; } break; case NFSATTRBIT_TIMEBACKUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); if (!nd->nd_repstat) nd->nd_repstat = NFSERR_ATTRNOTSUPP; attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMECREATE: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_btime); attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFYSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &nvap->na_mtime); nvap->na_vaflags &= ~VA_UTIMES_NULL; attrsum += NFSX_V4TIME; } else { vfs_timestamp(&nvap->na_mtime); if (!toclient) nvap->na_vaflags |= VA_UTIMES_NULL; } break; case NFSATTRBIT_MODESETMASKED: NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); mode = fxdr_unsigned(u_short, *tl++); mask = fxdr_unsigned(u_short, *tl); /* * vp == NULL implies an Open/Create operation. * This attribute can only be used for Setattr and * only for NFSv4.1 or higher. * If moderet != 0, a mode attribute has also been * specified and this attribute cannot be done in the * same Setattr operation. */ if ((nd->nd_flag & ND_NFSV41) == 0) nd->nd_repstat = NFSERR_ATTRNOTSUPP; else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || vp == NULL) nd->nd_repstat = NFSERR_INVAL; else if (moderet == 0) moderet = VOP_GETATTR(vp, &va, nd->nd_cred); if (moderet == 0) nvap->na_mode = (mode & mask) | (va.va_mode & ~mask); else nd->nd_repstat = moderet; attrsum += 2 * NFSX_UNSIGNED; break; default: nd->nd_repstat = NFSERR_ATTRNOTSUPP; /* * set bitpos so we drop out of the loop. */ bitpos = NFSATTRBIT_MAX; break; } } /* * some clients pad the attrlist, so we need to skip over the * padding. */ if (attrsum > attrsize) { error = NFSERR_BADXDR; } else { attrsize = NFSM_RNDUP(attrsize); if (attrsum < attrsize) error = nfsm_advance(nd, attrsize - attrsum, -1); } nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Check/setup export credentials. */ int nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, struct ucred *credanon, bool testsec) { int error; /* * Check/setup credentials. */ if (nd->nd_flag & ND_GSS) exp->nes_exflag &= ~MNT_EXPORTANON; /* * Check to see if the operation is allowed for this security flavor. */ error = 0; if (testsec) { error = nfsvno_testexp(nd, exp); if (error != 0) goto out; } /* * Check to see if the file system is exported V4 only. */ if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { error = NFSERR_PROGNOTV4; goto out; } /* * Now, map the user credentials. * (Note that ND_AUTHNONE will only be set for an NFSv3 * Fsinfo RPC. If set for anything else, this code might need * to change.) */ if (NFSVNO_EXPORTED(exp)) { if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || NFSVNO_EXPORTANON(exp) || (nd->nd_flag & ND_AUTHNONE) != 0) { nd->nd_cred->cr_uid = credanon->cr_uid; nd->nd_cred->cr_gid = credanon->cr_gid; crsetgroups(nd->nd_cred, credanon->cr_ngroups, credanon->cr_groups); } else if ((nd->nd_flag & ND_GSS) == 0) { /* * If using AUTH_SYS, call nfsrv_getgrpscred() to see * if there is a replacement credential with a group * list set up by "nfsuserd -manage-gids". * If there is no replacement, nfsrv_getgrpscred() * simply returns its argument. */ nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); } } out: NFSEXITCODE2(error, nd); return (error); } /* * Check exports. */ int nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, struct ucred **credp) { int error; error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, &exp->nes_numsecflavor, exp->nes_secflavors); if (error) { - if (nfs_rootfhset) { + if (NFSD_VNET(nfs_rootfhset)) { exp->nes_exflag = 0; exp->nes_numsecflavor = 0; error = 0; } } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > MAXSECFLAVORS) { printf("nfsvno_checkexp: numsecflavors out of range\n"); exp->nes_numsecflavor = 0; error = EACCES; } NFSEXITCODE(error); return (error); } /* * Get a vnode for a file handle and export stuff. */ int nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, int lktype, struct vnode **vpp, struct nfsexstuff *exp, struct ucred **credp) { int error; *credp = NULL; exp->nes_numsecflavor = 0; error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); if (error != 0) /* Make sure the server replies ESTALE to the client. */ error = ESTALE; if (nam && !error) { error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, &exp->nes_numsecflavor, exp->nes_secflavors); if (error) { - if (nfs_rootfhset) { + if (NFSD_VNET(nfs_rootfhset)) { exp->nes_exflag = 0; exp->nes_numsecflavor = 0; error = 0; } else { vput(*vpp); } } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > MAXSECFLAVORS) { printf("nfsvno_fhtovp: numsecflavors out of range\n"); exp->nes_numsecflavor = 0; error = EACCES; vput(*vpp); } } NFSEXITCODE(error); return (error); } /* * nfsd_fhtovp() - convert a fh to a vnode ptr * - look up fsid in mount list (if not found ret error) * - get vp and export rights by calling nfsvno_fhtovp() * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon * for AUTH_SYS * - if mpp != NULL, return the mount point so that it can * be used for vn_finished_write() by the caller */ void nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, struct vnode **vpp, struct nfsexstuff *exp, struct mount **mpp, int startwrite, int nextop) { struct mount *mp, *mpw; struct ucred *credanon; fhandle_t *fhp; int error; if (mpp != NULL) *mpp = NULL; *vpp = NULL; fhp = (fhandle_t *)nfp->nfsrvfh_data; mp = vfs_busyfs(&fhp->fh_fsid); if (mp == NULL) { nd->nd_repstat = ESTALE; goto out; } if (startwrite) { mpw = mp; error = vn_start_write(NULL, &mpw, V_WAIT); if (error != 0) { mpw = NULL; vfs_unbusy(mp); nd->nd_repstat = ESTALE; goto out; } if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) lktype = LK_EXCLUSIVE; } else mpw = NULL; nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, &credanon); vfs_unbusy(mp); /* * For NFSv4 without a pseudo root fs, unexported file handles * can be returned, so that Lookup works everywhere. */ if (!nd->nd_repstat && exp->nes_exflag == 0 && !(nd->nd_flag & ND_NFSV4)) { vput(*vpp); *vpp = NULL; nd->nd_repstat = EACCES; } /* * Personally, I've never seen any point in requiring a * reserved port#, since only in the rare case where the * clients are all boxes with secure system privileges, * does it provide any enhanced security, but... some people * believe it to be useful and keep putting this code back in. * (There is also some "security checker" out there that * complains if the nfs server doesn't enforce this.) * However, note the following: * RFC3530 (NFSv4) specifies that a reserved port# not be * required. * RFC2623 recommends that, if a reserved port# is checked for, * that there be a way to turn that off--> ifdef'd. */ #ifdef NFS_REQRSVPORT if (!nd->nd_repstat) { struct sockaddr_in *saddr; struct sockaddr_in6 *saddr6; saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); if (!(nd->nd_flag & ND_NFSV4) && ((saddr->sin_family == AF_INET && ntohs(saddr->sin_port) >= IPPORT_RESERVED) || (saddr6->sin6_family == AF_INET6 && ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { vput(*vpp); nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); } } #endif /* NFS_REQRSVPORT */ /* * Check/setup credentials. */ if (!nd->nd_repstat) { nd->nd_saveduid = nd->nd_cred->cr_uid; nd->nd_repstat = nfsd_excred(nd, exp, credanon, nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); if (nd->nd_repstat) vput(*vpp); } if (credanon != NULL) crfree(credanon); if (nd->nd_repstat) { vn_finished_write(mpw); *vpp = NULL; } else if (mpp != NULL) { *mpp = mpw; } out: NFSEXITCODE2(0, nd); } /* * glue for fp. */ static int fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) { struct filedesc *fdp; struct file *fp; int error = 0; fdp = p->td_proc->p_fd; if (fd < 0 || fd >= fdp->fd_nfiles || (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { error = EBADF; goto out; } *fpp = fp; out: NFSEXITCODE(error); return (error); } /* * Called from nfssvc() to update the exports list. Just call * vfs_export(). This has to be done, since the v4 root fake fs isn't * in the mount list. */ int nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) { struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; int error = 0; struct nameidata nd; fhandle_t fh; - error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); + error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export); if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) - nfs_rootfhset = 0; + NFSD_VNET(nfs_rootfhset) = 0; else if (error == 0) { if (nfsexargp->fspec == NULL) { error = EPERM; goto out; } /* * If fspec != NULL, this is the v4root path. */ NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec, p); if ((error = namei(&nd)) != 0) goto out; error = nfsvno_getfh(nd.ni_vp, &fh, p); vrele(nd.ni_vp); if (!error) { - nfs_rootfh.nfsrvfh_len = NFSX_MYFH; + NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; NFSBCOPY((caddr_t)&fh, - nfs_rootfh.nfsrvfh_data, + NFSD_VNET(nfs_rootfh).nfsrvfh_data, sizeof (fhandle_t)); - nfs_rootfhset = 1; + NFSD_VNET(nfs_rootfhset) = 1; } } out: NFSEXITCODE(error); return (error); } /* * This function needs to test to see if the system is near its limit * for memory allocation via malloc() or mget() and return True iff * either of these resources are near their limit. * XXX (For now, this is just a stub.) */ int nfsrv_testmalloclimit = 0; int nfsrv_mallocmget_limit(void) { static int printmesg = 0; static int testval = 1; if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { if ((printmesg++ % 100) == 0) printf("nfsd: malloc/mget near limit\n"); return (1); } return (0); } /* * BSD specific initialization of a mount point. */ void nfsd_mntinit(void) { - static int inited = 0; - if (inited) + NFSD_LOCK(); + if (NFSD_VNET(nfsrv_mntinited)) { + NFSD_UNLOCK(); return; - inited = 1; - nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); - TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); - TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist); - nfsv4root_mnt.mnt_export = NULL; - TAILQ_INIT(&nfsv4root_opt); - TAILQ_INIT(&nfsv4root_newopt); - nfsv4root_mnt.mnt_opt = &nfsv4root_opt; - nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; - nfsv4root_mnt.mnt_nvnodelistsize = 0; - nfsv4root_mnt.mnt_lazyvnodelistsize = 0; + } + NFSD_VNET(nfsrv_mntinited) = true; + NFSD_UNLOCK(); + + if (curthread->td_ucred->cr_prison == &prison0) + NFSD_VNET(nfsstatsv1_p) = &nfsstatsv1; + else + NFSD_VNET(nfsstatsv1_p) = malloc(sizeof(struct nfsstatsv1), + M_TEMP, M_WAITOK | M_ZERO); + nfsstatsv1_p->srvcache_tcppeak = 0; + nfsstatsv1_p->srvcache_size = 0; + NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP, + M_WAITOK | M_ZERO); + NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED); + mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF); + lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0); + TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist); + TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist); + NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; + TAILQ_INIT(&NFSD_VNET(nfsv4root_opt)); + TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt)); + NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt); + NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt); + NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0; + NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0; + callout_init(&NFSD_VNET(nfsd_callout), 1); + + nfsrvd_initcache(); + nfsd_init(); } static void nfsd_timer(void *arg) { - - nfsrv_servertimer(); - callout_reset_sbt(&nfsd_callout, SBT_1S, SBT_1S, nfsd_timer, NULL, 0); + struct vnet *vnetp; + + vnetp = (struct vnet *)arg; + NFSD_CURVNET_SET_QUIET(vnetp); + nfsrv_servertimer(vnetp); + callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer, + arg, 0); + NFSD_CURVNET_RESTORE(); } /* * Get a vnode for a file handle, without checking exports, etc. */ struct vnode * nfsvno_getvp(fhandle_t *fhp) { struct mount *mp; struct vnode *vp; int error; mp = vfs_busyfs(&fhp->fh_fsid); if (mp == NULL) return (NULL); error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error) return (NULL); return (vp); } /* * Do a local VOP_ADVLOCK(). */ int nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, u_int64_t end, struct thread *td) { int error = 0; struct flock fl; u_int64_t tlen; if (nfsrv_dolocallocks == 0) goto out; ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); fl.l_whence = SEEK_SET; fl.l_type = ftype; fl.l_start = (off_t)first; if (end == NFS64BITSSET) { fl.l_len = 0; } else { tlen = end - first; fl.l_len = (off_t)tlen; } /* * For FreeBSD8, the l_pid and l_sysid must be set to the same * values for all calls, so that all locks will be held by the * nfsd server. (The nfsd server handles conflicts between the * various clients.) * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 * bytes, so it can't be put in l_sysid. */ if (nfsv4_sysid == 0) nfsv4_sysid = nlm_acquire_next_sysid(); fl.l_pid = (pid_t)0; fl.l_sysid = (int)nfsv4_sysid; if (ftype == F_UNLCK) error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, (F_POSIX | F_REMOTE)); else error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, (F_POSIX | F_REMOTE)); out: NFSEXITCODE(error); return (error); } /* * Check the nfsv4 root exports. */ int nfsvno_v4rootexport(struct nfsrv_descript *nd) { struct ucred *credanon; int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; uint64_t exflags; - error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, + error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags, &credanon, &numsecflavor, secflavors); if (error) { error = NFSERR_PROGUNAVAIL; goto out; } if (credanon != NULL) crfree(credanon); for (i = 0; i < numsecflavor; i++) { if (secflavors[i] == AUTH_SYS) nd->nd_flag |= ND_EXAUTHSYS; else if (secflavors[i] == RPCSEC_GSS_KRB5) nd->nd_flag |= ND_EXGSS; else if (secflavors[i] == RPCSEC_GSS_KRB5I) nd->nd_flag |= ND_EXGSSINTEGRITY; else if (secflavors[i] == RPCSEC_GSS_KRB5P) nd->nd_flag |= ND_EXGSSPRIVACY; } /* And set ND_EXxx flags for TLS. */ if ((exflags & MNT_EXTLS) != 0) { nd->nd_flag |= ND_EXTLS; if ((exflags & MNT_EXTLSCERT) != 0) nd->nd_flag |= ND_EXTLSCERT; if ((exflags & MNT_EXTLSCERTUSER) != 0) nd->nd_flag |= ND_EXTLSCERTUSER; } out: NFSEXITCODE(error); return (error); } +/* Osd entry for nfsrv_cleanup. */ +static int +nfsrv_prison_cleanup(void *obj, void *data __unused) +{ + struct prison *pr = obj; + + if ((pr->pr_flags & PR_VNET) == 0) + return (0); + nfsrv_cleanup(pr); + return (0); +} + /* * Nfs server pseudo system call for the nfsd's */ /* * MPSAFE */ static int nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) { struct file *fp; struct nfsd_addsock_args sockarg; struct nfsd_nfsd_args nfsdarg; struct nfsd_nfsd_oargs onfsdarg; struct nfsd_pnfsd_args pnfsdarg; struct vnode *vp, *nvp, *curdvp; struct pnfsdsfile *pf; struct nfsdevice *ds, *fds; cap_rights_t rights; int buflen, error, ret; char *buf, *cp, *cp2, *cp3; char fname[PNFS_FILENAME_LEN + 1]; + NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); if (uap->flag & NFSSVC_NFSDADDSOCK) { error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); if (error) goto out; /* * Since we don't know what rights might be required, * pretend that we need them all. It is better to be too * careful than too reckless. */ error = fget(td, sockarg.sock, cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); if (error != 0) goto out; if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, td); error = EPERM; goto out; } error = nfsrvd_addsock(fp); fdrop(fp, td); } else if (uap->flag & NFSSVC_NFSDNFSD) { if (uap->argp == NULL) { error = EINVAL; goto out; } if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); if (error == 0) { nfsdarg.principal = onfsdarg.principal; nfsdarg.minthreads = onfsdarg.minthreads; nfsdarg.maxthreads = onfsdarg.maxthreads; nfsdarg.version = 1; nfsdarg.addr = NULL; nfsdarg.addrlen = 0; nfsdarg.dnshost = NULL; nfsdarg.dnshostlen = 0; nfsdarg.dspath = NULL; nfsdarg.dspathlen = 0; nfsdarg.mdspath = NULL; nfsdarg.mdspathlen = 0; nfsdarg.mirrorcnt = 1; } } else error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); if (error) goto out; if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && nfsdarg.mirrorcnt >= 1 && nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, nfsdarg.dspathlen, nfsdarg.dnshostlen, nfsdarg.mdspathlen, nfsdarg.mirrorcnt); cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); if (error != 0) { free(cp, M_TEMP); goto out; } cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ nfsdarg.addr = cp; cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ nfsdarg.dnshost = cp; cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ nfsdarg.dspath = cp; cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); if (error != 0) { free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(nfsdarg.dspath, M_TEMP); free(cp, M_TEMP); goto out; } cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ nfsdarg.mdspath = cp; } else { nfsdarg.addr = NULL; nfsdarg.addrlen = 0; nfsdarg.dnshost = NULL; nfsdarg.dnshostlen = 0; nfsdarg.dspath = NULL; nfsdarg.dspathlen = 0; nfsdarg.mdspath = NULL; nfsdarg.mdspathlen = 0; nfsdarg.mirrorcnt = 1; } - nfsd_timer(NULL); + nfsd_timer(NFSD_TD_TO_VNET(td)); error = nfsrvd_nfsd(td, &nfsdarg); + callout_drain(&NFSD_VNET(nfsd_callout)); free(nfsdarg.addr, M_TEMP); free(nfsdarg.dnshost, M_TEMP); free(nfsdarg.dspath, M_TEMP); free(nfsdarg.mdspath, M_TEMP); } else if (uap->flag & NFSSVC_PNFSDS) { error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || pnfsdarg.op == PNFSDOP_FORCEDELDS)) { cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, NULL); if (error == 0) error = nfsrv_deldsserver(pnfsdarg.op, cp, td); free(cp, M_TEMP); } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; buf = malloc(buflen, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); if (error == 0 && pnfsdarg.dspath != NULL) { cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.dspath, cp2, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", error); } else cp2 = NULL; if (error == 0 && pnfsdarg.curdspath != NULL) { cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); error = copyinstr(pnfsdarg.curdspath, cp3, PATH_MAX + 1, NULL); NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", error); } else cp3 = NULL; curdvp = NULL; fds = NULL; if (error == 0) error = nfsrv_mdscopymr(cp, cp2, cp3, buf, &buflen, fname, td, &vp, &nvp, &pf, &ds, &fds); NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); if (error == 0) { if (pf->dsf_dir >= nfsrv_dsdirsize) { printf("copymr: dsdir out of range\n"); pf->dsf_dir = 0; } NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); error = nfsrv_copymr(vp, nvp, ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, (struct pnfsdsfile *)buf, buflen / sizeof(*pf), td->td_ucred, td); vput(vp); vput(nvp); if (fds != NULL && error == 0) { curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; ret = vn_lock(curdvp, LK_EXCLUSIVE); if (ret == 0) { nfsrv_dsremove(curdvp, fname, td->td_ucred, td); NFSVOPUNLOCK(curdvp); } } NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); } free(cp, M_TEMP); free(cp2, M_TEMP); free(cp3, M_TEMP); free(buf, M_TEMP); } } else { error = nfssvc_srvcall(td, uap, td->td_ucred); } out: + NFSD_CURVNET_RESTORE(); NFSEXITCODE(error); return (error); } static int nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) { struct nfsex_args export; struct nfsex_oldargs oexp; struct file *fp = NULL; int stablefd, i, len; struct nfsd_clid adminrevoke; struct nfsd_dumplist dumplist; struct nfsd_dumpclients *dumpclients; struct nfsd_dumplocklist dumplocklist; struct nfsd_dumplocks *dumplocks; struct nameidata nd; vnode_t vp; int error = EINVAL, igotlock; struct proc *procp; gid_t *grps; - static int suspend_nfsd = 0; if (uap->flag & NFSSVC_PUBLICFH) { NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); error = copyin(uap->argp, &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); if (!error) nfs_pubfhset = 1; } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { error = copyin(uap->argp,(caddr_t)&export, sizeof (struct nfsex_args)); if (!error) { grps = NULL; if (export.export.ex_ngroups > NGROUPS_MAX || export.export.ex_ngroups < 0) error = EINVAL; else if (export.export.ex_ngroups > 0) { grps = malloc(export.export.ex_ngroups * sizeof(gid_t), M_TEMP, M_WAITOK); error = copyin(export.export.ex_groups, grps, export.export.ex_ngroups * sizeof(gid_t)); export.export.ex_groups = grps; } else export.export.ex_groups = NULL; if (!error) error = nfsrv_v4rootexport(&export, cred, p); free(grps, M_TEMP); } } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == NFSSVC_V4ROOTEXPORT) { error = copyin(uap->argp,(caddr_t)&oexp, sizeof (struct nfsex_oldargs)); if (!error) { memset(&export.export, 0, sizeof(export.export)); export.export.ex_flags = (uint64_t)oexp.export.ex_flags; export.export.ex_root = oexp.export.ex_root; export.export.ex_uid = oexp.export.ex_anon.cr_uid; export.export.ex_ngroups = oexp.export.ex_anon.cr_ngroups; export.export.ex_groups = NULL; if (export.export.ex_ngroups > XU_NGROUPS || export.export.ex_ngroups < 0) error = EINVAL; else if (export.export.ex_ngroups > 0) { export.export.ex_groups = malloc( export.export.ex_ngroups * sizeof(gid_t), M_TEMP, M_WAITOK); for (i = 0; i < export.export.ex_ngroups; i++) export.export.ex_groups[i] = oexp.export.ex_anon.cr_groups[i]; } export.export.ex_addr = oexp.export.ex_addr; export.export.ex_addrlen = oexp.export.ex_addrlen; export.export.ex_mask = oexp.export.ex_mask; export.export.ex_masklen = oexp.export.ex_masklen; export.export.ex_indexfile = oexp.export.ex_indexfile; export.export.ex_numsecflavors = oexp.export.ex_numsecflavors; if (export.export.ex_numsecflavors >= MAXSECFLAVORS || export.export.ex_numsecflavors < 0) error = EINVAL; else { for (i = 0; i < export.export.ex_numsecflavors; i++) export.export.ex_secflavors[i] = oexp.export.ex_secflavors[i]; } export.fspec = oexp.fspec; if (error == 0) error = nfsrv_v4rootexport(&export, cred, p); free(export.export.ex_groups, M_TEMP); } } else if (uap->flag & NFSSVC_NOPUBLICFH) { nfs_pubfhset = 0; error = 0; } else if (uap->flag & NFSSVC_STABLERESTART) { error = copyin(uap->argp, (caddr_t)&stablefd, sizeof (int)); if (!error) error = fp_getfvp(p, stablefd, &fp, &vp); if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) error = EBADF; - if (!error && newnfs_numnfsd != 0) + if (!error && NFSD_VNET(nfsrv_numnfsd) != 0) error = EPERM; if (!error) { - nfsrv_stablefirst.nsf_fp = fp; + NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp; nfsrv_setupstable(p); } } else if (uap->flag & NFSSVC_ADMINREVOKE) { error = copyin(uap->argp, (caddr_t)&adminrevoke, sizeof (struct nfsd_clid)); if (!error) error = nfsrv_adminrevoke(&adminrevoke, p); } else if (uap->flag & NFSSVC_DUMPCLIENTS) { error = copyin(uap->argp, (caddr_t)&dumplist, sizeof (struct nfsd_dumplist)); if (!error && (dumplist.ndl_size < 1 || dumplist.ndl_size > NFSRV_MAXDUMPLIST)) error = EPERM; if (!error) { len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); nfsrv_dumpclients(dumpclients, dumplist.ndl_size); error = copyout(dumpclients, dumplist.ndl_list, len); free(dumpclients, M_TEMP); } } else if (uap->flag & NFSSVC_DUMPLOCKS) { error = copyin(uap->argp, (caddr_t)&dumplocklist, sizeof (struct nfsd_dumplocklist)); if (!error && (dumplocklist.ndllck_size < 1 || dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) error = EPERM; if (!error) error = nfsrv_lookupfilename(&nd, dumplocklist.ndllck_fname, p); if (!error) { len = sizeof (struct nfsd_dumplocks) * dumplocklist.ndllck_size; dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); nfsrv_dumplocks(nd.ni_vp, dumplocks, dumplocklist.ndllck_size, p); vput(nd.ni_vp); error = copyout(dumplocks, dumplocklist.ndllck_list, len); free(dumplocks, M_TEMP); } } else if (uap->flag & NFSSVC_BACKUPSTABLE) { procp = p->td_proc; PROC_LOCK(procp); nfsd_master_pid = procp->p_pid; bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); nfsd_master_start = procp->p_stats->p_start; - nfsd_master_proc = procp; + NFSD_VNET(nfsd_master_proc) = procp; PROC_UNLOCK(procp); } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { NFSLOCKV4ROOTMUTEX(); - if (suspend_nfsd == 0) { + if (!NFSD_VNET(nfsrv_suspend_nfsd)) { /* Lock out all nfsd threads */ do { - igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, - NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); - } while (igotlock == 0 && suspend_nfsd == 0); - suspend_nfsd = 1; + igotlock = nfsv4_lock( + &NFSD_VNET(nfsd_suspend_lock), 1, NULL, + NFSV4ROOTLOCKMUTEXPTR, NULL); + } while (igotlock == 0 && + !NFSD_VNET(nfsrv_suspend_nfsd)); + NFSD_VNET(nfsrv_suspend_nfsd) = true; } NFSUNLOCKV4ROOTMUTEX(); error = 0; } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { NFSLOCKV4ROOTMUTEX(); - if (suspend_nfsd != 0) { - nfsv4_unlock(&nfsd_suspend_lock, 0); - suspend_nfsd = 0; + if (NFSD_VNET(nfsrv_suspend_nfsd)) { + nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0); + NFSD_VNET(nfsrv_suspend_nfsd) = false; } NFSUNLOCKV4ROOTMUTEX(); error = 0; } NFSEXITCODE(error); return (error); } /* * Check exports. * Returns 0 if ok, 1 otherwise. */ int nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) { int i; if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || (NFSVNO_EXTLSCERT(exp) && (nd->nd_flag & ND_TLSCERT) == 0) || (NFSVNO_EXTLSCERTUSER(exp) && (nd->nd_flag & ND_TLSCERTUSER) == 0)) { if ((nd->nd_flag & ND_NFSV4) != 0) return (NFSERR_WRONGSEC); #ifdef notnow /* There is currently no auth_stat for this. */ else if ((nd->nd_flag & ND_TLS) == 0) return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); else return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); #endif else return (NFSERR_AUTHERR | AUTH_TOOWEAK); } /* * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. */ if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO) return (0); /* * This seems odd, but allow the case where the security flavor * list is empty. This happens when NFSv4 is traversing non-exported * file systems. Exported file systems should always have a non-empty * security flavor list. */ if (exp->nes_numsecflavor == 0) return (0); for (i = 0; i < exp->nes_numsecflavor; i++) { /* * The tests for privacy and integrity must be first, * since ND_GSS is set for everything but AUTH_SYS. */ if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && (nd->nd_flag & ND_GSSPRIVACY)) return (0); if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && (nd->nd_flag & ND_GSSINTEGRITY)) return (0); if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && (nd->nd_flag & ND_GSS)) return (0); if (exp->nes_secflavors[i] == AUTH_SYS && (nd->nd_flag & ND_GSS) == 0) return (0); } if ((nd->nd_flag & ND_NFSV4) != 0) return (NFSERR_WRONGSEC); return (NFSERR_AUTHERR | AUTH_TOOWEAK); } /* * Calculate a hash value for the fid in a file handle. */ uint32_t nfsrv_hashfh(fhandle_t *fhp) { uint32_t hashval; hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); return (hashval); } /* * Calculate a hash value for the sessionid. */ uint32_t nfsrv_hashsessionid(uint8_t *sessionid) { uint32_t hashval; hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); return (hashval); } /* * Signal the userland master nfsd to backup the stable restart file. */ void nfsrv_backupstable(void) { struct proc *procp; - if (nfsd_master_proc != NULL) { + if (NFSD_VNET(nfsd_master_proc) != NULL) { procp = pfind(nfsd_master_pid); /* Try to make sure it is the correct process. */ - if (procp == nfsd_master_proc && + if (procp == NFSD_VNET(nfsd_master_proc) && procp->p_stats->p_start.tv_sec == nfsd_master_start.tv_sec && procp->p_stats->p_start.tv_usec == nfsd_master_start.tv_usec && strcmp(procp->p_comm, nfsd_master_comm) == 0) kern_psignal(procp, SIGUSR2); else - nfsd_master_proc = NULL; + NFSD_VNET(nfsd_master_proc) = NULL; if (procp != NULL) PROC_UNLOCK(procp); } } /* * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. * The arguments are in a structure, so that they can be passed through * taskqueue for a kernel process to execute this function. */ struct nfsrvdscreate { int done; int inprog; struct task tsk; struct ucred *tcred; struct vnode *dvp; NFSPROC_T *p; struct pnfsdsfile *pf; int err; fhandle_t fh; struct vattr va; struct vattr createva; }; int nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) { struct vnode *nvp; struct nameidata named; struct vattr va; char *bufp; u_long *hashp; struct nfsnode *np; struct nfsmount *nmp; int error; NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; named.ni_cnd.cn_thread = p; named.ni_cnd.cn_nameptr = bufp; if (fnamep != NULL) { strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); named.ni_cnd.cn_namelen = strlen(bufp); } else named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); /* Create the date file in the DS mount. */ error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); if (error == 0) { error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); vref(dvp); VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); if (error == 0) { /* Set the ownership of the file. */ error = VOP_SETATTR(nvp, nvap, tcred); NFSD_DEBUG(4, "nfsrv_dscreate:" " setattr-uid=%d\n", error); if (error != 0) vput(nvp); } if (error != 0) printf("pNFS: pnfscreate failed=%d\n", error); } else printf("pNFS: pnfscreate vnlock=%d\n", error); if (error == 0) { np = VTONFS(nvp); nmp = VFSTONFS(nvp->v_mount); if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") != 0 || nmp->nm_nam->sa_len > sizeof( struct sockaddr_in6) || np->n_fhp->nfh_len != NFSX_MYFH) { printf("Bad DS file: fstype=%s salen=%d" " fhlen=%d\n", nvp->v_mount->mnt_vfc->vfc_name, nmp->nm_nam->sa_len, np->n_fhp->nfh_len); error = ENOENT; } /* Set extattrs for the DS on the MDS file. */ if (error == 0) { if (dsa != NULL) { error = VOP_GETATTR(nvp, &va, tcred); if (error == 0) { dsa->dsa_filerev = va.va_filerev; dsa->dsa_size = va.va_size; dsa->dsa_atime = va.va_atime; dsa->dsa_mtime = va.va_mtime; dsa->dsa_bytes = va.va_bytes; } } if (error == 0) { NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, nmp->nm_nam->sa_len); NFSBCOPY(named.ni_cnd.cn_nameptr, pf->dsf_filename, sizeof(pf->dsf_filename)); } } else printf("pNFS: pnfscreate can't get DS" " attr=%d\n", error); if (nvpp != NULL && error == 0) *nvpp = nvp; else vput(nvp); } nfsvno_relpathbuf(&named); return (error); } /* * Start up the thread that will execute nfsrv_dscreate(). */ static void start_dscreate(void *arg, int pending) { struct nfsrvdscreate *dsc; dsc = (struct nfsrvdscreate *)arg; dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); dsc->done = 1; NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); } /* * Create a pNFS data file on the Data Server(s). */ static void nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, NFSPROC_T *p) { struct nfsrvdscreate *dsc, *tdsc = NULL; struct nfsdevice *ds, *tds, *fds; struct mount *mp; struct pnfsdsfile *pf, *tpf; struct pnfsdsattr dsattr; struct vattr va; struct vnode *dvp[NFSDEV_MAXMIRRORS]; struct nfsmount *nmp; fhandle_t fh; uid_t vauid; gid_t vagid; u_short vamode; struct ucred *tcred; int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; int failpos, timo; /* Get a DS server directory in a round-robin order. */ mirrorcnt = 1; mp = vp->v_mount; ds = fds = NULL; NFSDDSLOCK(); /* * Search for the first entry that handles this MDS fs, but use the * first entry for all MDS fs's otherwise. */ TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL) { if (tds->nfsdev_mdsisset == 0 && ds == NULL) ds = tds; else if (tds->nfsdev_mdsisset != 0 && fsidcmp( &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { ds = fds = tds; break; } } } if (ds == NULL) { NFSDDSUNLOCK(); NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); return; } i = dsdir[0] = ds->nfsdev_nextdir; ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; dvp[0] = ds->nfsdev_dsdir[i]; tds = TAILQ_NEXT(ds, nfsdev_list); if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL && ((tds->nfsdev_mdsisset == 0 && fds == NULL) || (tds->nfsdev_mdsisset != 0 && fds != NULL && fsidcmp(&mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0))) { dsdir[mirrorcnt] = i; dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; mirrorcnt++; if (mirrorcnt >= nfsrv_maxpnfsmirror) break; } } } /* Put at end of list to implement round-robin usage. */ TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); NFSDDSUNLOCK(); dsc = NULL; if (mirrorcnt > 1) tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, M_WAITOK | M_ZERO); tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | M_ZERO); error = nfsvno_getfh(vp, &fh, p); if (error == 0) error = VOP_GETATTR(vp, &va, cred); if (error == 0) { /* Set the attributes for "vp" to Setattr the DS vp. */ vauid = va.va_uid; vagid = va.va_gid; vamode = va.va_mode; VATTR_NULL(&va); va.va_uid = vauid; va.va_gid = vagid; va.va_mode = vamode; va.va_size = 0; } else printf("pNFS: pnfscreate getfh+attr=%d\n", error); NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, cred->cr_gid); /* Make data file name based on FH. */ tcred = newnfs_getcred(); /* * Create the file on each DS mirror, using kernel process(es) for the * additional mirrors. */ failpos = -1; for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { tpf->dsf_dir = dsdir[i]; tdsc->tcred = tcred; tdsc->p = p; tdsc->pf = tpf; tdsc->createva = *vap; NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); tdsc->va = va; tdsc->dvp = dvp[i]; tdsc->done = 0; tdsc->inprog = 0; tdsc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_dscreate, tdsc); NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, NULL, tcred, p, NULL); if (ret != 0) { KASSERT(error == 0, ("nfsrv_dscreate err=%d", error)); if (failpos == -1 && nfsds_failerr(ret)) failpos = i; else error = ret; } } } if (error == 0) { tpf->dsf_dir = dsdir[mirrorcnt - 1]; error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, &dsattr, NULL, tcred, p, NULL); if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { failpos = mirrorcnt - 1; error = 0; } } timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; /* Wait for kernel task(s) to complete. */ for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { while (tdsc->inprog != 0 && tdsc->done == 0) tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); if (tdsc->err != 0) { if (failpos == -1 && nfsds_failerr(tdsc->err)) failpos = i; else if (error == 0) error = tdsc->err; } } /* * If failpos has been set, that mirror has failed, so it needs * to be disabled. */ if (failpos >= 0) { nmp = VFSTONFS(dvp[failpos]->v_mount); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); } NFSUNLOCKMNT(nmp); } NFSFREECRED(tcred); if (error == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", mirrorcnt, nfsrv_maxpnfsmirror); /* * For all mirrors that couldn't be created, fill in the * *pf structure, but with an IP address == 0.0.0.0. */ tpf = pf + mirrorcnt; for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { *tpf = *pf; tpf->dsf_sin.sin_family = AF_INET; tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); tpf->dsf_sin.sin_addr.s_addr = 0; tpf->dsf_sin.sin_port = 0; } error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); if (error == 0) error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); if (error != 0) printf("pNFS: pnfscreate setextattr=%d\n", error); } else printf("pNFS: pnfscreate=%d\n", error); free(pf, M_TEMP); free(dsc, M_TEMP); } /* * Get the information needed to remove the pNFS Data Server file from the * Metadata file. Upon success, ddvp is set non-NULL to the locked * DS directory vnode. The caller must unlock *ddvp when done with it. */ static void nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, int *mirrorcntp, char *fname, fhandle_t *fhp) { struct vattr va; struct ucred *tcred; char *buf; int buflen, error; dvpp[0] = NULL; /* If not an exported regular file or not a pNFS server, just return. */ if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || nfsrv_devidcnt == 0) return; /* Check to see if this is the last hard link. */ tcred = newnfs_getcred(); error = VOP_GETATTR(vp, &va, tcred); NFSFREECRED(tcred); if (error != 0) { printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); return; } if (va.va_nlink > 1) return; error = nfsvno_getfh(vp, fhp, p); if (error != 0) { printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); return; } buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); /* Get the directory vnode for the DS mount and the file handle. */ error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); free(buf, M_TEMP); if (error != 0) printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); } /* * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. * The arguments are in a structure, so that they can be passed through * taskqueue for a kernel process to execute this function. */ struct nfsrvdsremove { int done; int inprog; struct task tsk; struct ucred *tcred; struct vnode *dvp; NFSPROC_T *p; int err; char fname[PNFS_FILENAME_LEN + 1]; }; static int nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, NFSPROC_T *p) { struct nameidata named; struct vnode *nvp; char *bufp; u_long *hashp; int error; error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); if (error != 0) return (error); named.ni_cnd.cn_nameiop = DELETE; named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; named.ni_cnd.cn_cred = tcred; named.ni_cnd.cn_thread = p; named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_nameptr = bufp; named.ni_cnd.cn_namelen = strlen(fname); strlcpy(bufp, fname, NAME_MAX); NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); if (error == 0) { error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); vput(nvp); } NFSVOPUNLOCK(dvp); nfsvno_relpathbuf(&named); if (error != 0) printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_dsremove(). */ static void start_dsremove(void *arg, int pending) { struct nfsrvdsremove *dsrm; dsrm = (struct nfsrvdsremove *)arg; dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, dsrm->p); dsrm->done = 1; NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); } /* * Remove a pNFS data file from a Data Server. * nfsrv_pnfsremovesetup() must have been called before the MDS file was * removed to set up the dvp and fill in the FH. */ static void nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, NFSPROC_T *p) { struct ucred *tcred; struct nfsrvdsremove *dsrm, *tdsrm; struct nfsdevice *ds; struct nfsmount *nmp; int failpos, i, ret, timo; tcred = newnfs_getcred(); dsrm = NULL; if (mirrorcnt > 1) dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); /* * Remove the file on each DS mirror, using kernel process(es) for the * additional mirrors. */ failpos = -1; for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { tdsrm->tcred = tcred; tdsrm->p = p; tdsrm->dvp = dvp[i]; strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); tdsrm->inprog = 0; tdsrm->done = 0; tdsrm->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_dsremove, tdsrm); NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_dsremove(dvp[i], fname, tcred, p); if (failpos == -1 && nfsds_failerr(ret)) failpos = i; } } ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) failpos = mirrorcnt - 1; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; /* Wait for kernel task(s) to complete. */ for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { while (tdsrm->inprog != 0 && tdsrm->done == 0) tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); if (failpos == -1 && nfsds_failerr(tdsrm->err)) failpos = i; } /* * If failpos has been set, that mirror has failed, so it needs * to be disabled. */ if (failpos >= 0) { nmp = VFSTONFS(dvp[failpos]->v_mount); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); } NFSUNLOCKMNT(nmp); } /* Get rid all layouts for the file. */ nfsrv_freefilelayouts(fhp); NFSFREECRED(tcred); free(dsrm, M_TEMP); } /* * Generate a file name based on the file handle and put it in *bufp. * Return the number of bytes generated. */ static int nfsrv_putfhname(fhandle_t *fhp, char *bufp) { int i; uint8_t *cp; const uint8_t *hexdigits = "0123456789abcdef"; cp = (uint8_t *)fhp; for (i = 0; i < sizeof(*fhp); i++) { bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; } bufp[2 * i] = '\0'; return (2 * i); } /* * Update the Metadata file's attributes from the DS file when a Read/Write * layout is returned. * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. */ int nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) { struct ucred *tcred; int error; /* Do this as root so that it won't fail with EACCES. */ tcred = newnfs_getcred(); error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); NFSFREECRED(tcred); return (error); } /* * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. */ static int nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, NFSPROC_T *p) { int error; error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); return (error); } static int nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, struct thread *p, int ioproc, struct mbuf **mpp, char *cp, struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, off_t *offp, int content, bool *eofp) { struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; fhandle_t fh[NFSDEV_MAXMIRRORS]; struct vnode *dvp[NFSDEV_MAXMIRRORS]; struct nfsdevice *ds; struct pnfsdsattr dsattr; struct opnfsdsattr odsattr; char *buf; int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; NFSD_DEBUG(4, "in nfsrv_proxyds\n"); /* * If not a regular file, not exported or not a pNFS server, * just return ENOENT. */ if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || nfsrv_devidcnt == 0) return (ENOENT); buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); error = 0; /* * For Getattr, get the Change attribute (va_filerev) and size (va_size) * from the MetaData file's extended attribute. */ if (ioproc == NFSPROC_GETATTR) { error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, p); if (error == 0) { if (buflen == sizeof(odsattr)) { NFSBCOPY(buf, &odsattr, buflen); nap->na_filerev = odsattr.dsa_filerev; nap->na_size = odsattr.dsa_size; nap->na_atime = odsattr.dsa_atime; nap->na_mtime = odsattr.dsa_mtime; /* * Fake na_bytes by rounding up na_size. * Since we don't know the block size, just * use BLKDEV_IOSIZE. */ nap->na_bytes = (odsattr.dsa_size + BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); } else if (buflen == sizeof(dsattr)) { NFSBCOPY(buf, &dsattr, buflen); nap->na_filerev = dsattr.dsa_filerev; nap->na_size = dsattr.dsa_size; nap->na_atime = dsattr.dsa_atime; nap->na_mtime = dsattr.dsa_mtime; nap->na_bytes = dsattr.dsa_bytes; } else error = ENXIO; } if (error == 0) { /* * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() * returns 0, just return now. nfsrv_checkdsattr() * returns 0 if there is no Read/Write layout * plus either an Open/Write_access or Write * delegation issued to a client for the file. */ if (nfsrv_pnfsgetdsattr == 0 || nfsrv_checkdsattr(vp, p) == 0) { free(buf, M_TEMP); return (error); } } /* * Clear ENOATTR so the code below will attempt to do a * nfsrv_getattrdsrpc() to get the attributes and (re)create * the extended attribute. */ if (error == ENOATTR) error = 0; } origmircnt = -1; trycnt = 0; tryagain: if (error == 0) { buflen = 1024; if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == LK_EXCLUSIVE) printf("nfsrv_proxyds: Readds vp exclusively locked\n"); error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, NULL, NULL); if (error == 0) { for (i = 0; i < mirrorcnt; i++) nmp[i] = VFSTONFS(dvp[i]->v_mount); } else printf("pNFS: proxy getextattr sockaddr=%d\n", error); } else printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); if (error == 0) { failpos = -1; if (origmircnt == -1) origmircnt = mirrorcnt; /* * If failpos is set to a mirror#, then that mirror has * failed and will be disabled. For Read, Getattr and Seek, the * function only tries one mirror, so if that mirror has * failed, it will need to be retried. As such, increment * tryitagain for these cases. * For Write, Setattr and Setacl, the function tries all * mirrors and will not return an error for the case where * one mirror has failed. For these cases, the functioning * mirror(s) will have been modified, so a retry isn't * necessary. These functions will set failpos for the * failed mirror#. */ if (ioproc == NFSPROC_READDS) { error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], mpp, mpp2); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * read is required. */ failpos = 0; error = 0; trycnt++; } } else if (ioproc == NFSPROC_WRITEDS) error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, &nmp[0], mirrorcnt, mpp, cp, &failpos); else if (ioproc == NFSPROC_SETATTR) error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], mirrorcnt, nap, &failpos); else if (ioproc == NFSPROC_SETACL) error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], mirrorcnt, aclp, &failpos); else if (ioproc == NFSPROC_SEEKDS) { error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, p, nmp[0]); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * read is required. */ failpos = 0; error = 0; trycnt++; } } else if (ioproc == NFSPROC_ALLOCATE) error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, &nmp[0], mirrorcnt, &failpos); else { error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, vp, nmp[mirrorcnt - 1], nap); if (nfsds_failerr(error) && mirrorcnt > 1) { /* * Setting failpos will cause the mirror * to be disabled and then a retry of this * getattr is required. */ failpos = mirrorcnt - 1; error = 0; trycnt++; } } ds = NULL; if (failpos >= 0) { failnmp = nmp[failpos]; NFSLOCKMNT(failnmp); if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(failnmp); ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, failnmp, p); NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", failpos, ds); if (ds != NULL) nfsrv_killrpcs(failnmp); NFSLOCKMNT(failnmp); failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(failnmp); } NFSUNLOCKMNT(failnmp); } for (i = 0; i < mirrorcnt; i++) NFSVOPUNLOCK(dvp[i]); NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, trycnt); /* Try the Read/Getattr again if a mirror was deleted. */ if (ds != NULL && trycnt > 0 && trycnt < origmircnt) goto tryagain; } else { /* Return ENOENT for any Extended Attribute error. */ error = ENOENT; } free(buf, M_TEMP); NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); return (error); } /* * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended * attribute. * newnmpp - If it points to a non-NULL nmp, that is the destination and needs * to be checked. If it points to a NULL nmp, then it returns * a suitable destination. * curnmp - If non-NULL, it is the source mount for the copy. */ int nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, struct nfsmount *curnmp, int *ippos, int *dsdirp) { struct vnode *dvp, *nvp = NULL, **tdvpp; struct mount *mp; struct nfsmount *nmp, *newnmp; struct sockaddr *sad; struct sockaddr_in *sin; struct nfsdevice *ds, *tds, *fndds; struct pnfsdsfile *pf; uint32_t dsdir; int error, fhiszero, fnd, gotone, i, mirrorcnt; ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); *mirrorcntp = 1; tdvpp = dvpp; if (nvpp != NULL) *nvpp = NULL; if (dvpp != NULL) *dvpp = NULL; if (ippos != NULL) *ippos = -1; if (newnmpp != NULL) newnmp = *newnmpp; else newnmp = NULL; mp = vp->v_mount; error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", buflenp, buf, p); mirrorcnt = *buflenp / sizeof(*pf); if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || *buflenp != sizeof(*pf) * mirrorcnt)) error = ENOATTR; pf = (struct pnfsdsfile *)buf; /* If curnmp != NULL, check for a match in the mirror list. */ if (curnmp != NULL && error == 0) { fnd = 0; for (i = 0; i < mirrorcnt; i++, pf++) { sad = (struct sockaddr *)&pf->dsf_sin; if (nfsaddr2_match(sad, curnmp->nm_nam)) { if (ippos != NULL) *ippos = i; fnd = 1; break; } } if (fnd == 0) error = ENXIO; } gotone = 0; pf = (struct pnfsdsfile *)buf; NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, error); for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { fhiszero = 0; sad = (struct sockaddr *)&pf->dsf_sin; sin = &pf->dsf_sin; dsdir = pf->dsf_dir; if (dsdir >= nfsrv_dsdirsize) { printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); error = ENOATTR; } else if (nvpp != NULL && newnmp != NULL && nfsaddr2_match(sad, newnmp->nm_nam)) error = EEXIST; if (error == 0) { if (ippos != NULL && curnmp == NULL && sad->sa_family == AF_INET && sin->sin_addr.s_addr == 0) *ippos = i; if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) fhiszero = 1; /* Use the socket address to find the mount point. */ fndds = NULL; NFSDDSLOCK(); /* Find a match for the IP address. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL) { dvp = ds->nfsdev_dvp; nmp = VFSTONFS(dvp->v_mount); if (nmp != ds->nfsdev_nmp) printf("different2 nmp %p %p\n", nmp, ds->nfsdev_nmp); if (nfsaddr2_match(sad, nmp->nm_nam)) { fndds = ds; break; } } } if (fndds != NULL && newnmpp != NULL && newnmp == NULL) { /* Search for a place to make a mirror copy. */ TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { if (tds->nfsdev_nmp != NULL && fndds != tds && ((tds->nfsdev_mdsisset == 0 && fndds->nfsdev_mdsisset == 0) || (tds->nfsdev_mdsisset != 0 && fndds->nfsdev_mdsisset != 0 && fsidcmp(&tds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0))) { *newnmpp = tds->nfsdev_nmp; break; } } if (tds != NULL) { /* * Move this entry to the end of the * list, so it won't be selected as * easily the next time. */ TAILQ_REMOVE(&nfsrv_devidhead, tds, nfsdev_list); TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, nfsdev_list); } } NFSDDSUNLOCK(); if (fndds != NULL) { dvp = fndds->nfsdev_dsdir[dsdir]; if (lktype != 0 || fhiszero != 0 || (nvpp != NULL && *nvpp == NULL)) { if (fhiszero != 0) error = vn_lock(dvp, LK_EXCLUSIVE); else if (lktype != 0) error = vn_lock(dvp, lktype); else error = vn_lock(dvp, LK_SHARED); /* * If the file handle is all 0's, try to * do a Lookup against the DS to acquire * it. * If dvpp == NULL or the Lookup fails, * unlock dvp after the call. */ if (error == 0 && (fhiszero != 0 || (nvpp != NULL && *nvpp == NULL))) { error = nfsrv_pnfslookupds(vp, dvp, pf, &nvp, p); if (error == 0) { if (fhiszero != 0) nfsrv_pnfssetfh( vp, pf, devid, fnamep, nvp, p); if (nvpp != NULL && *nvpp == NULL) { *nvpp = nvp; *dsdirp = dsdir; } else vput(nvp); } if (error != 0 || lktype == 0) NFSVOPUNLOCK(dvp); } } if (error == 0) { gotone++; NFSD_DEBUG(4, "gotone=%d\n", gotone); if (devid != NULL) { NFSBCOPY(fndds->nfsdev_deviceid, devid, NFSX_V4DEVICEID); devid += NFSX_V4DEVICEID; } if (dvpp != NULL) *tdvpp++ = dvp; if (fhp != NULL) NFSBCOPY(&pf->dsf_fh, fhp++, NFSX_MYFH); if (fnamep != NULL && gotone == 1) strlcpy(fnamep, pf->dsf_filename, sizeof(pf->dsf_filename)); } else NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " "err=%d\n", error); } } } if (error == 0 && gotone == 0) error = ENOENT; NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, error); if (error == 0) *mirrorcntp = gotone; else { if (gotone > 0 && dvpp != NULL) { /* * If the error didn't occur on the first one and * dvpp != NULL, the one(s) prior to the failure will * have locked dvp's that need to be unlocked. */ for (i = 0; i < gotone; i++) { NFSVOPUNLOCK(*dvpp); *dvpp++ = NULL; } } /* * If it found the vnode to be copied from before a failure, * it needs to be vput()'d. */ if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } } return (error); } /* * Set the extended attribute for the Change attribute. */ static int nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) { struct pnfsdsattr dsattr; int error; ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); dsattr.dsa_filerev = nap->na_filerev; dsattr.dsa_size = nap->na_size; dsattr.dsa_atime = nap->na_atime; dsattr.dsa_mtime = nap->na_mtime; dsattr.dsa_bytes = nap->na_bytes; error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); if (error != 0) printf("pNFS: setextattr=%d\n", error); return (error); } static int nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; struct mbuf *m, *m2; int error = 0, retlen, tlen, trimlen; NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); *mpp = NULL; /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0, cred); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); txdr_hyper(off, tl); *(tl + 2) = txdr_unsigned(len); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); NFSM_STRSIZ(retlen, len); if (retlen > 0) { /* Trim off the pre-data XDR from the mbuf chain. */ m = nd->nd_mrep; while (m != NULL && m != nd->nd_md) { if (m->m_next == nd->nd_md) { m->m_next = NULL; m_freem(nd->nd_mrep); nd->nd_mrep = m = nd->nd_md; } else m = m->m_next; } if (m == NULL) { printf("nfsrv_readdsrpc: busted mbuf list\n"); error = ENOENT; goto nfsmout; } /* * Now, adjust first mbuf so that any XDR before the * read data is skipped over. */ trimlen = nd->nd_dpos - mtod(m, char *); if (trimlen > 0) { m->m_len -= trimlen; NFSM_DATAP(m, trimlen); } /* * Truncate the mbuf chain at retlen bytes of data, * plus XDR padding that brings the length up to a * multiple of 4. */ tlen = NFSM_RNDUP(retlen); do { if (m->m_len >= tlen) { m->m_len = tlen; tlen = 0; m2 = m->m_next; m->m_next = NULL; m_freem(m2); break; } tlen -= m->m_len; m = m->m_next; } while (m != NULL); if (tlen > 0) { printf("nfsrv_readdsrpc: busted mbuf list\n"); error = ENOENT; goto nfsmout; } *mpp = nd->nd_mrep; *mpendp = m; nd->nd_mrep = NULL; } } else error = nd->nd_repstat; nfsmout: /* If nd->nd_mrep is already NULL, this is a no-op. */ m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); return (error); } /* * Do a write RPC on a DS data file, using this structure for the arguments, * so that this function can be executed by a separate kernel process. */ struct nfsrvwritedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; int len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; struct mbuf *m; int err; }; static int nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int commit, error, retlen; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0, cred); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; /* * Do all writes FileSync, since the server doesn't hold onto dirty * buffers. Since clients should be accessing the DS servers directly * using the pNFS layouts, this just needs to work correctly as a * fallback. */ *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); *tl = txdr_unsigned(len); NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); /* 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; nfsm_set(nd, m->m_len); NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); /* Do a Getattr for the attributes that change upon writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); retlen = fxdr_unsigned(int, *tl++); commit = fxdr_unsigned(int, *tl); if (commit != NFSWRITE_FILESYNC) error = NFSERR_IO; NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", retlen, commit, error); } else error = nd->nd_repstat; /* We have no use for the Write Verifier since we use FileSync. */ /* * Get the Change, Size, Access Time and Modify Time attributes and set * on the Metadata file, so its attributes will be what the file's * would be if it had been written. */ if (error == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_writedsdorpc(). */ static void start_writedsdorpc(void *arg, int pending) { struct nfsrvwritedsdorpc *drpc; drpc = (struct nfsrvwritedsdorpc *)arg; drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->m, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); } static int nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct mbuf **mpp, char *cp, int *failposp) { struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; struct mbuf *m; int error, i, offs, ret, timo; NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* Calculate offset in mbuf chain that data starts. */ offs = cp - mtod(*mpp, char *); NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); /* * Do the write RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_writedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, tdrpc->m, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a allocate RPC on a DS data file, using this structure for the arguments, * so that this function can be executed by a separate kernel process. */ struct nfsrvallocatedsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; off_t off; off_t len; struct nfsmount *nmp; struct ucred *cred; NFSPROC_T *p; int err; }; static int nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript *nd; nfsattrbit_t attrbits; nfsv4stateid_t st; int error; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0, cred); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } else error = nd->nd_repstat; NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_allocatedsdorpc(). */ static void start_allocatedsdorpc(void *arg, int pending) { struct nfsrvallocatedsdorpc *drpc; drpc = (struct nfsrvallocatedsdorpc *)arg; drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); } static int nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, int *failposp) { struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the allocate RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->off = off; tdrpc->len = len; tdrpc->nmp = *nmpp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->inprog = 0; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, cred, p); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } static int nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, struct nfsvattr *dsnap) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; nfsattrbit_t attrbits; int error; NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0, cred); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); /* Do a Getattr for the attributes that change due to writing. */ NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", nd->nd_repstat); /* Get rid of weak cache consistency data for now. */ if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); if (error != 0) goto nfsmout; /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); if (*++tl != 0) nd->nd_flag |= ND_NOMOREDATA; } error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error != 0) goto nfsmout; if (nd->nd_repstat != 0) error = nd->nd_repstat; /* * Get the Change, Size, Access Time and Modify Time attributes and set * on the Metadata file, so its attributes will be what the file's * would be if it had been written. */ if (error == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); } NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); return (error); } struct nfsrvsetattrdsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; struct nfsmount *nmp; struct vnode *vp; struct ucred *cred; NFSPROC_T *p; struct nfsvattr na; struct nfsvattr dsna; int err; }; /* * Start up the thread that will execute nfsrv_setattrdsdorpc(). */ static void start_setattrdsdorpc(void *arg, int pending) { struct nfsrvsetattrdsdorpc *drpc; drpc = (struct nfsrvsetattrdsdorpc *)arg; drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); drpc->done = 1; } static int nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct nfsvattr *nap, int *failposp) { struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the setattr RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; tdrpc->inprog = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->nmp = *nmpp; tdrpc->vp = vp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->na = *nap; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; if (error == 0) error = nfsrv_setextattr(vp, &na, p); NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Do a Setattr of an NFSv4 ACL on the DS file. */ static int nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) { struct nfsrv_descript *nd; nfsv4stateid_t st; nfsattrbit_t attrbits; int error; NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), NULL, NULL, 0, 0, cred); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); /* * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), * so passing in the metadata "vp" will be ok, since it is of * the same type (VREG). */ nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, NULL, 0, 0, 0, 0, 0, NULL); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", nd->nd_repstat); error = nd->nd_repstat; m_freem(nd->nd_mrep); free(nd, M_TEMP); return (error); } struct nfsrvsetacldsdorpc { int done; int inprog; struct task tsk; fhandle_t fh; struct nfsmount *nmp; struct vnode *vp; struct ucred *cred; NFSPROC_T *p; struct acl *aclp; int err; }; /* * Start up the thread that will execute nfsrv_setacldsdorpc(). */ static void start_setacldsdorpc(void *arg, int pending) { struct nfsrvsetacldsdorpc *drpc; drpc = (struct nfsrvsetacldsdorpc *)arg; drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, drpc->vp, drpc->nmp, drpc->aclp); drpc->done = 1; } static int nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, int *failposp) { struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the setattr RPC for every DS, using a separate kernel process * for every DS except the last one. */ error = 0; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { tdrpc->done = 0; tdrpc->inprog = 0; NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); tdrpc->nmp = *nmpp; tdrpc->vp = vp; tdrpc->cred = cred; tdrpc->p = p; tdrpc->aclp = aclp; tdrpc->err = 0; ret = EIO; if (nfs_pnfsiothreads != 0) { ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); if (nfsds_failerr(ret) && *failposp == -1) *failposp = i; else if (error == 0 && ret != 0) error = ret; } nmpp++; fhp++; } ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) *failposp = mirrorcnt - 1; else if (error == 0 && ret != 0) error = ret; NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); tdrpc = drpc; timo = hz / 50; /* Wait for 20msec. */ if (timo < 1) timo = 1; for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { /* Wait for RPCs on separate threads to complete. */ while (tdrpc->inprog != 0 && tdrpc->done == 0) tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); if (nfsds_failerr(tdrpc->err) && *failposp == -1) *failposp = i; else if (error == 0 && tdrpc->err != 0) error = tdrpc->err; } free(drpc, M_TEMP); return (error); } /* * Getattr call to the DS for the attributes that change due to writing. */ static int nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) { struct nfsrv_descript *nd; int error; nfsattrbit_t attrbits; NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0, cred); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); (void) nfsrv_putattrbit(nd, &attrbits); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); /* * We can only save the updated values in the extended * attribute if the vp is exclusively locked. * This should happen when any of the following operations * occur on the vnode: * Close, Delegreturn, LayoutCommit, LayoutReturn * As such, the updated extended attribute should get saved * before nfsrv_checkdsattr() returns 0 and allows the cached * attributes to be returned without calling this function. */ if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { error = nfsrv_setextattr(vp, nap, p); NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", error); } } else error = nd->nd_repstat; m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); return (error); } /* * Seek call to a DS. */ static int nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) { uint32_t *tl; struct nfsrv_descript *nd; nfsv4stateid_t st; int error; NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); /* * Use a stateid where other is an alternating 01010 pattern and * seqid is 0xffffffff. This value is not defined as special by * the RFC and is used by the FreeBSD NFS server to indicate an * MDS->DS proxy operation. */ st.other[0] = 0x55555555; st.other[1] = 0x55555555; st.other[2] = 0x55555555; st.seqid = 0xffffffff; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, sizeof(fhandle_t), NULL, NULL, 0, 0, cred); nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(*offp, tl); tl += 2; *tl = txdr_unsigned(content); error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) { free(nd, M_TEMP); return (error); } NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); if (*tl++ == newnfs_true) *eofp = true; else *eofp = false; *offp = fxdr_hyper(tl); } else error = nd->nd_repstat; nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); return (error); } /* * Get the device id and file handle for a DS file. */ int nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, fhandle_t *fhp, char *devid) { int buflen, error; char *buf; buflen = 1024; buf = malloc(buflen, M_TEMP, M_WAITOK); error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); free(buf, M_TEMP); return (error); } /* * Do a Lookup against the DS for the filename. */ static int nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, struct vnode **nvpp, NFSPROC_T *p) { struct nameidata named; struct ucred *tcred; char *bufp; u_long *hashp; struct vnode *nvp; int error; tcred = newnfs_getcred(); named.ni_cnd.cn_nameiop = LOOKUP; named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; named.ni_cnd.cn_cred = tcred; named.ni_cnd.cn_thread = p; named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_nameptr = bufp; named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); strlcpy(bufp, pf->dsf_filename, NAME_MAX); NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); NFSFREECRED(tcred); nfsvno_relpathbuf(&named); if (error == 0) *nvpp = nvp; NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); return (error); } /* * Set the file handle to the correct one. */ static void nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, char *fnamep, struct vnode *nvp, NFSPROC_T *p) { struct nfsnode *np; int ret = 0; np = VTONFS(nvp); NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); /* * We can only do a vn_set_extattr() if the vnode is exclusively * locked and vn_start_write() has been done. If devid != NULL or * fnamep != NULL or the vnode is shared locked, vn_start_write() * may not have been done. * If not done now, it will be done on a future call. */ if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == LK_EXCLUSIVE) ret = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), (char *)pf, p); NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); } /* * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point * when the DS has failed. */ void nfsrv_killrpcs(struct nfsmount *nmp) { /* * Call newnfs_nmcancelreqs() to cause * any RPCs in progress on the mount point to * fail. * This will cause any process waiting for an * RPC to complete while holding a vnode lock * on the mounted-on vnode (such as "df" or * a non-forced "umount") to fail. * This will unlock the mounted-on vnode so * a forced dismount can succeed. * The NFSMNTP_CANCELRPCS flag should be set when this function is * called. */ newnfs_nmcancelreqs(nmp); } /* * Sum up the statfs info for each of the DSs, so that the client will * receive the total for all DSs. */ static int nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) { struct statfs *tsf; struct nfsdevice *ds; struct vnode **dvpp, **tdvpp, *dvp; uint64_t tot; int cnt, error = 0, i; if (nfsrv_devidcnt <= 0) return (ENXIO); dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); /* Get an array of the dvps for the DSs. */ tdvpp = dvpp; i = 0; NFSDDSLOCK(); /* First, search for matches for same file system. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; } } /* * If no matches for same file system, total all servers not assigned * to a file system. */ if (i == 0) { TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset == 0) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; } } } NFSDDSUNLOCK(); cnt = i; /* Do a VFS_STATFS() for each of the DSs and sum them up. */ tdvpp = dvpp; for (i = 0; i < cnt && error == 0; i++) { dvp = *tdvpp++; error = VFS_STATFS(dvp->v_mount, tsf); if (error == 0) { if (sf->f_bsize == 0) { if (tsf->f_bsize > 0) sf->f_bsize = tsf->f_bsize; else sf->f_bsize = 8192; } if (tsf->f_blocks > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_blocks * tsf->f_bsize; sf->f_blocks += (tot / sf->f_bsize); } else sf->f_blocks += tsf->f_blocks; } if (tsf->f_bfree > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_bfree * tsf->f_bsize; sf->f_bfree += (tot / sf->f_bsize); } else sf->f_bfree += tsf->f_bfree; } if (tsf->f_bavail > 0) { if (sf->f_bsize != tsf->f_bsize) { tot = tsf->f_bavail * tsf->f_bsize; sf->f_bavail += (tot / sf->f_bsize); } else sf->f_bavail += tsf->f_bavail; } } } free(tsf, M_TEMP); free(dvpp, M_TEMP); return (error); } /* * Set an NFSv4 acl. */ int nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) { int error; if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { error = NFSERR_ATTRNOTSUPP; goto out; } /* * With NFSv4 ACLs, chmod(2) may need to add additional entries. * Make sure it has enough room for that - splitting every entry * into two and appending "canonical six" entries at the end. * Cribbed out of kern/vfs_acl.c - Rick M. */ if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { error = NFSERR_ATTRNOTSUPP; goto out; } error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); if (error == 0) { error = nfsrv_dssetacl(vp, aclp, cred, p); if (error == ENOENT) error = 0; } out: NFSEXITCODE(error); return (error); } /* * Seek vnode op call (actually it is a VOP_IOCTL()). * This function is called with the vnode locked, but unlocks and vrele()s * the vp before returning. */ int nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr at; int error, ret; ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); /* * Attempt to seek on a DS file. A return of ENOENT implies * there is no DS file to seek on. */ error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, NULL, NULL, NULL, NULL, offp, content, eofp); if (error != ENOENT) { vput(vp); return (error); } /* * Do the VOP_IOCTL() call. For the case where *offp == file_size, * VOP_IOCTL() will return ENXIO. However, the correct reply for * NFSv4.2 is *eofp == true and error == 0 for this case. */ NFSVOPUNLOCK(vp); error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); *eofp = false; if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { /* Handle the cases where we might be at EOF. */ ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); if (ret == 0 && *offp == at.na_size) { *eofp = true; error = 0; } if (ret != 0 && error == 0) error = ret; } vrele(vp); NFSEXITCODE(error); return (error); } /* * Allocate vnode op call. */ int nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, NFSPROC_T *p) { int error; off_t olen; ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); /* * Attempt to allocate on a DS file. A return of ENOENT implies * there is no DS file to allocate on. */ error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, NULL, NULL, NULL, NULL, &len, 0, NULL); if (error != ENOENT) return (error); /* * Do the actual VOP_ALLOCATE(), looping so long as * progress is being made, to achieve completion. */ do { olen = len; error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); if (error == 0 && len > 0 && olen > len) maybe_yield(); } while (error == 0 && len > 0 && olen > len); if (error == 0 && len > 0) error = NFSERR_IO; NFSEXITCODE(error); return (error); } /* * Get Extended Atribute vnode op into an mbuf list. */ int nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) { struct iovec *iv; struct uio io, *uiop = &io; struct mbuf *m, *m2; int alen, error, len, tlen; size_t siz; /* First, find out the size of the extended attribute. */ error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, &siz, cred, p); if (error != 0) return (NFSERR_NOXATTR); if (siz > maxresp - NFS_MAXXDR) return (NFSERR_XATTR2BIG); len = siz; tlen = NFSM_RNDUP(len); if (tlen > 0) { /* * If cnt > MCLBYTES and the reply will not be saved, use * ext_pgs mbufs for TLS. * For NFSv4.0, we do not know for sure if the reply will * be saved, so do not use ext_pgs mbufs for NFSv4.0. * Always use ext_pgs mbufs if ND_EXTPG is set. */ if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, maxextsiz, &m, &m2, &iv); else uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, &iv); uiop->uio_iov = iv; } else { uiop->uio_iovcnt = 0; uiop->uio_iov = iv = NULL; m = m2 = NULL; } uiop->uio_offset = 0; uiop->uio_resid = tlen; uiop->uio_rw = UIO_READ; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = p; #ifdef MAC error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); if (error != 0) goto out; #endif if (tlen > 0) error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, NULL, cred, p); if (error != 0) goto out; if (uiop->uio_resid > 0) { alen = tlen; len = tlen - uiop->uio_resid; tlen = NFSM_RNDUP(len); if (alen != tlen) printf("nfsvno_getxattr: weird size read\n"); if (tlen == 0) { m_freem(m); m = m2 = NULL; } else if (alen != tlen || tlen != len) m2 = nfsrv_adj(m, alen - tlen, tlen - len); } *lenp = len; *mpp = m; *mpendp = m2; out: if (error != 0) { if (m != NULL) m_freem(m); *lenp = 0; } free(iv, M_TEMP); NFSEXITCODE(error); return (error); } /* * Set Extended attribute vnode op from an mbuf list. */ int nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, char *cp, struct ucred *cred, struct thread *p) { struct iovec *iv; struct uio uio, *uiop = &uio; int cnt, error; error = 0; #ifdef MAC error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); #endif if (error != 0) goto out; uiop->uio_rw = UIO_WRITE; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = p; uiop->uio_offset = 0; uiop->uio_resid = len; if (len > 0) { error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); uiop->uio_iov = iv; uiop->uio_iovcnt = cnt; } else { uiop->uio_iov = iv = NULL; uiop->uio_iovcnt = 0; } if (error == 0) { error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, cred, p); if (error == 0) { if (vp->v_type == VREG && nfsrv_devidcnt != 0) nfsvno_updateds(vp, cred, p); error = VOP_FSYNC(vp, MNT_WAIT, p); } free(iv, M_TEMP); } out: NFSEXITCODE(error); return (error); } /* * For a pNFS server, the DS file's ctime and * va_filerev (TimeMetadata and Change) needs to * be updated. This is a hack, but works by * flipping the S_ISGID bit in va_mode and then * flipping it back. * It does result in two MDS->DS RPCs, but creating * a custom RPC just to do this seems overkill, since * Setxattr/Rmxattr will not be done that frequently. * If it fails part way through, that is not too * serious, since the DS file is never executed. */ static void nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr nva; int ret; u_short tmode; ret = VOP_GETATTR(vp, &nva.na_vattr, cred); if (ret == 0) { tmode = nva.na_mode; NFSVNO_ATTRINIT(&nva); tmode ^= S_ISGID; NFSVNO_SETATTRVAL(&nva, mode, tmode); ret = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, NULL, NULL, NULL, &nva, NULL, NULL, 0, NULL); if (ret == 0) { tmode ^= S_ISGID; NFSVNO_SETATTRVAL(&nva, mode, tmode); ret = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, NULL, NULL, NULL, &nva, NULL, NULL, 0, NULL); } } } /* * Remove Extended attribute vnode op. */ int nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, struct ucred *cred, struct thread *p) { int error; /* * Get rid of any delegations. I am not sure why this is required, * but RFC-8276 says so. */ error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); if (error != 0) goto out; #ifdef MAC error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, name); if (error != 0) goto out; #endif error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, cred, p); if (error == 0) { if (vp->v_type == VREG && nfsrv_devidcnt != 0) nfsvno_updateds(vp, cred, p); error = VOP_FSYNC(vp, MNT_WAIT, p); } out: NFSEXITCODE(error); return (error); } /* * List Extended Atribute vnode op into an mbuf list. */ int nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) { struct iovec iv; struct uio io; int error; size_t siz; *bufp = NULL; /* First, find out the size of the extended attribute. */ error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, p); if (error != 0) return (NFSERR_NOXATTR); if (siz <= cookie) { *lenp = 0; *eofp = true; goto out; } if (siz > cookie + *lenp) { siz = cookie + *lenp; *eofp = false; } else *eofp = true; /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ if (siz > 10 * 1024 * 1024) { error = NFSERR_XATTR2BIG; goto out; } *bufp = malloc(siz, M_TEMP, M_WAITOK); iv.iov_base = *bufp; iv.iov_len = siz; io.uio_iovcnt = 1; io.uio_iov = &iv; io.uio_offset = 0; io.uio_resid = siz; io.uio_rw = UIO_READ; io.uio_segflg = UIO_SYSSPACE; io.uio_td = p; #ifdef MAC error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); if (error != 0) goto out; #endif error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, p); if (error != 0) goto out; if (io.uio_resid > 0) siz -= io.uio_resid; *lenp = siz; out: if (error != 0) { free(*bufp, M_TEMP); *bufp = NULL; } NFSEXITCODE(error); return (error); } /* * Trim trailing data off the mbuf list being built. */ void nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, int bextpg, int bextpgsiz) { vm_page_t pg; int fullpgsiz, i; if (mb->m_next != NULL) { m_freem(mb->m_next); mb->m_next = NULL; } if ((mb->m_flags & M_EXTPG) != 0) { KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, ("nfsm_trimtrailing: bextpg out of range")); KASSERT(bpos == (char *)(void *) PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, ("nfsm_trimtrailing: bextpgsiz bad!")); /* First, get rid of any pages after this position. */ for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); vm_page_unwire_noq(pg); vm_page_free(pg); } mb->m_epg_npgs = bextpg + 1; if (bextpg == 0) fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; else fullpgsiz = PAGE_SIZE; mb->m_epg_last_len = fullpgsiz - bextpgsiz; mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); for (i = 1; i < mb->m_epg_npgs; i++) mb->m_len += m_epg_pagelen(mb, i, 0); nd->nd_bextpgsiz = bextpgsiz; nd->nd_bextpg = bextpg; } else mb->m_len = bpos - mtod(mb, char *); nd->nd_mb = mb; nd->nd_bpos = bpos; } /* * Check to see if a put file handle operation should test for * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. * When Open is the next operation, NFSERR_WRONGSEC cannot be * replied for the Open cases that use a component. This can * be identified by the fact that the file handle's type is VDIR. */ bool nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp) { if ((nd->nd_flag & ND_NFSV4) == 0) return (true); if ((nd->nd_flag & ND_LASTOP) != 0) return (false); if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) return (false); if (nextop == NFSV4OP_OPEN && vtyp == VDIR) return (false); return (true); } /* * Check DSs marked no space. */ void nfsrv_checknospc(void) { struct statfs *tsf; struct nfsdevice *ds; struct vnode **dvpp, **tdvpp, *dvp; char *devid, *tdevid; int cnt, error = 0, i; if (nfsrv_devidcnt <= 0) return; dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); /* Get an array of the dvps for the DSs. */ tdvpp = dvpp; tdevid = devid; i = 0; NFSDDSLOCK(); /* First, search for matches for same file system. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { if (++i > nfsrv_devidcnt) break; *tdvpp++ = ds->nfsdev_dvp; NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); tdevid += NFSX_V4DEVICEID; } } NFSDDSUNLOCK(); /* Do a VFS_STATFS() for each of the DSs and clear no space. */ cnt = i; tdvpp = dvpp; tdevid = devid; for (i = 0; i < cnt && error == 0; i++) { dvp = *tdvpp++; error = VFS_STATFS(dvp->v_mount, tsf); if (error == 0 && tsf->f_bavail > 0) { NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); nfsrv_marknospc(tdevid, false); } tdevid += NFSX_V4DEVICEID; } free(tsf, M_TEMP); free(dvpp, M_TEMP); free(devid, M_TEMP); } +/* + * Initialize everything that needs to be initialized for a vnet. + */ +static void +nfsrv_vnetinit(const void *unused __unused) +{ + + nfsd_mntinit(); +} +VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, + nfsrv_vnetinit, NULL); + +/* + * Clean up everything that is in a vnet and needs to be + * done when the jail is destroyed or the module unloaded. + */ +static void +nfsrv_cleanup(struct prison *pr) +{ + int i; + + NFSD_CURVNET_SET(pr->pr_vnet); + NFSD_LOCK(); + if (!NFSD_VNET(nfsrv_mntinited)) { + NFSD_UNLOCK(); + NFSD_CURVNET_RESTORE(); + return; + } + NFSD_VNET(nfsrv_mntinited) = false; + NFSD_UNLOCK(); + + /* Clean out all NFSv4 state. */ + nfsrv_throwawayallstate(curthread); + + /* Clean the NFS server reply cache */ + nfsrvd_cleancache(); + + /* Clean out v4root exports. */ + if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) { + vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export); + free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT); + NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; + } + + /* Free up the krpc server pool. */ + if (NFSD_VNET(nfsrvd_pool) != NULL) + svcpool_destroy(NFSD_VNET(nfsrvd_pool)); + + /* and get rid of the locks */ + for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { + mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx); + mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx); + } + mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx); + for (i = 0; i < nfsrv_sessionhashsize; i++) + mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx); + lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock); + free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE); + free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE); + free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE); + free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT); + free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE); + free(NFSD_VNET(nfssessionhash), M_NFSDSESSION); + free(NFSD_VNET(nfsv4root_mnt), M_TEMP); + NFSD_VNET(nfsv4root_mnt) = NULL; + if (pr != &prison0) + free(NFSD_VNET(nfsstatsv1_p), M_TEMP); + NFSD_CURVNET_RESTORE(); +} + extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); /* * Called once to initialize data structures... */ static int nfsd_modevent(module_t mod, int type, void *data) { int error = 0, i; + osd_method_t methods[PR_MAXMETHOD] = { + [PR_METHOD_REMOVE] = nfsrv_prison_cleanup, + }; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) goto out; newnfs_portinit(); - for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { - mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL, - MTX_DEF); - mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL, - MTX_DEF); - } mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); - mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF); mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); - lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); - callout_init(&nfsd_callout, 1); - nfsrvd_initcache(); - nfsd_init(); - NFSD_LOCK(); - nfsrvd_init(0); - NFSD_UNLOCK(); - nfsd_mntinit(); #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; #endif nfsd_call_nfsd = nfssvc_nfsd; + /* XXX-BZ OSD to VNET? */ + nfsrv_osd_jail_slot = osd_jail_register(NULL, methods); loaded = 1; break; case MOD_UNLOAD: if (newnfs_numnfsd != 0) { error = EBUSY; break; } #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = NULL; vn_deleg_ops.vndeleg_disable = NULL; #endif nfsd_call_nfsd = NULL; - callout_drain(&nfsd_callout); - - /* Clean out all NFSv4 state. */ - nfsrv_throwawayallstate(curthread); - - /* Clean the NFS server reply cache */ - nfsrvd_cleancache(); - - /* Free up the krpc server pool. */ - if (nfsrvd_pool != NULL) - svcpool_destroy(nfsrvd_pool); - - /* and get rid of the locks */ - for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { - mtx_destroy(&nfsrchash_table[i].mtx); - mtx_destroy(&nfsrcahash_table[i].mtx); - } + osd_jail_deregister(nfsrv_osd_jail_slot); + nfsrv_cleanup(&prison0); mtx_destroy(&nfsrc_udpmtx); mtx_destroy(&nfs_v4root_mutex); - mtx_destroy(&nfsv4root_mnt.mnt_mtx); mtx_destroy(&nfsrv_dontlistlock_mtx); mtx_destroy(&nfsrv_recalllock_mtx); - for (i = 0; i < nfsrv_sessionhashsize; i++) - mtx_destroy(&nfssessionhash[i].mtx); if (nfslayouthash != NULL) { for (i = 0; i < nfsrv_layouthashsize; i++) mtx_destroy(&nfslayouthash[i].mtx); free(nfslayouthash, M_NFSDSESSION); } - lockdestroy(&nfsv4root_mnt.mnt_explock); - free(nfsclienthash, M_NFSDCLIENT); - free(nfslockhash, M_NFSDLOCKFILE); - free(nfssessionhash, M_NFSDSESSION); loaded = 0; break; default: error = EOPNOTSUPP; break; } out: NFSEXITCODE(error); return (error); } static moduledata_t nfsd_mod = { "nfsd", nfsd_modevent, NULL, }; DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfsd, 1); MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); MODULE_DEPEND(nfsd, krpc, 1, 1, 1); MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); diff --git a/sys/fs/nfsserver/nfs_nfsdsocket.c b/sys/fs/nfsserver/nfs_nfsdsocket.c index 002236113068..8dd733f2aa73 100644 --- a/sys/fs/nfsserver/nfs_nfsdsocket.c +++ b/sys/fs/nfsserver/nfs_nfsdsocket.c @@ -1,1378 +1,1386 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); /* * Socket operations for use by the nfs server. */ #include -extern struct nfsstatsv1 nfsstatsv1; -extern struct nfsrvfh nfs_pubfh, nfs_rootfh; -extern int nfs_pubfhset, nfs_rootfhset; +extern struct nfsrvfh nfs_pubfh; +extern int nfs_pubfhset; extern struct nfsv4lock nfsv4rootfs_lock; -extern struct nfsrv_stablefirst nfsrv_stablefirst; -extern struct nfsclienthashhead *nfsclienthash; extern int nfsrv_clienthashsize; -extern int nfsrc_floodlevel, nfsrc_tcpsavedreplies; extern int nfsd_debuglevel; extern int nfsrv_layouthighwater; extern volatile int nfsrv_layoutcnt; NFSV4ROOTLOCKMUTEX; NFSSTATESPINLOCK; +NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst); +NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); +NFSD_VNET_DECLARE(int, nfsrc_floodlevel); +NFSD_VNET_DECLARE(int, nfsrc_tcpsavedreplies); +NFSD_VNET_DECLARE(struct nfsrvfh, nfs_rootfh); +NFSD_VNET_DECLARE(int, nfs_rootfhset); +NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); + int (*nfsrv3_procs0[NFS_V3NPROCS])(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *) = { (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_getattr, nfsrvd_setattr, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_access, nfsrvd_readlink, nfsrvd_read, nfsrvd_write, nfsrvd_create, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_remove, nfsrvd_remove, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_readdir, nfsrvd_readdirplus, nfsrvd_statfs, nfsrvd_fsinfo, nfsrvd_pathconf, nfsrvd_commit, }; int (*nfsrv3_procs1[NFS_V3NPROCS])(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *) = { (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, nfsrvd_lookup, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, nfsrvd_mkdir, nfsrvd_symlink, nfsrvd_mknod, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, }; int (*nfsrv3_procs2[NFS_V3NPROCS])(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *) = { (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, nfsrvd_rename, nfsrvd_link, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, }; int (*nfsrv4_ops0[NFSV42_NOPS])(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *) = { (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_access, nfsrvd_close, nfsrvd_commit, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_delegpurge, nfsrvd_delegreturn, nfsrvd_getattr, nfsrvd_getfh, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_lock, nfsrvd_lockt, nfsrvd_locku, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_verify, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_openconfirm, nfsrvd_opendowngrade, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_read, nfsrvd_readdirplus, nfsrvd_readlink, nfsrvd_remove, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_renew, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_secinfo, nfsrvd_setattr, nfsrvd_setclientid, nfsrvd_setclientidcfrm, nfsrvd_verify, nfsrvd_write, nfsrvd_releaselckown, nfsrvd_notsupp, nfsrvd_bindconnsess, nfsrvd_exchangeid, nfsrvd_createsession, nfsrvd_destroysession, nfsrvd_freestateid, nfsrvd_notsupp, nfsrvd_getdevinfo, nfsrvd_notsupp, nfsrvd_layoutcommit, nfsrvd_layoutget, nfsrvd_layoutreturn, nfsrvd_secinfononame, nfsrvd_sequence, nfsrvd_notsupp, nfsrvd_teststateid, nfsrvd_notsupp, nfsrvd_destroyclientid, nfsrvd_reclaimcomplete, nfsrvd_allocate, (int (*)(struct nfsrv_descript *, int, vnode_t , struct nfsexstuff *))0, nfsrvd_notsupp, nfsrvd_notsupp, nfsrvd_ioadvise, nfsrvd_layouterror, nfsrvd_layoutstats, nfsrvd_notsupp, nfsrvd_notsupp, nfsrvd_notsupp, nfsrvd_seek, nfsrvd_notsupp, nfsrvd_notsupp, nfsrvd_getxattr, nfsrvd_setxattr, nfsrvd_listxattr, nfsrvd_rmxattr, }; int (*nfsrv4_ops1[NFSV42_NOPS])(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *) = { (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, nfsrvd_mknod, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, nfsrvd_lookup, nfsrvd_lookup, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, nfsrvd_open, nfsrvd_openattr, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t *, fhandle_t *, struct nfsexstuff *))0, }; int (*nfsrv4_ops2[NFSV42_NOPS])(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *) = { (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, nfsrvd_link, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, nfsrvd_rename, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, nfsrvd_copy_file_range, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, (int (*)(struct nfsrv_descript *, int, vnode_t , vnode_t , struct nfsexstuff *, struct nfsexstuff *))0, }; /* * Static array that defines which nfs rpc's are nonidempotent */ static int nfsrv_nonidempotent[NFS_V3NPROCS] = { FALSE, FALSE, TRUE, FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, }; /* * This static array indicates whether or not the RPC modifies the * file system. */ int nfsrv_writerpc[NFS_NPROCS] = { 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1 }; SYSCTL_DECL(_vfs_nfsd); static int nfs_minminorv4 = NFSV4_MINORVERSION; SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_min_minorversion4, CTLFLAG_RWTUN, &nfs_minminorv4, 0, "The lowest minor version of NFSv4 handled by the server"); static int nfs_maxminorv4 = NFSV42_MINORVERSION; SYSCTL_INT(_vfs_nfsd, OID_AUTO, server_max_minorversion4, CTLFLAG_RWTUN, &nfs_maxminorv4, 0, "The highest minor version of NFSv4 handled by the server"); /* local functions */ static void nfsrvd_compound(struct nfsrv_descript *nd, int isdgram, u_char *tag, int taglen, u_int32_t minorvers); /* * This static array indicates which server procedures require the extra * arguments to return the current file handle for V2, 3. */ static int nfs_retfh[NFS_V3NPROCS] = { 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 2, 2, 0, 0, 0, 0, 0, 0 }; extern struct nfsv4_opflag nfsv4_opflag[NFSV42_NOPS]; static int nfsv3to4op[NFS_V3NPROCS] = { NFSPROC_NULL, NFSV4OP_GETATTR, NFSV4OP_SETATTR, NFSV4OP_LOOKUP, NFSV4OP_ACCESS, NFSV4OP_READLINK, NFSV4OP_READ, NFSV4OP_WRITE, NFSV4OP_V3CREATE, NFSV4OP_MKDIR, NFSV4OP_SYMLINK, NFSV4OP_MKNOD, NFSV4OP_REMOVE, NFSV4OP_RMDIR, NFSV4OP_RENAME, NFSV4OP_LINK, NFSV4OP_READDIR, NFSV4OP_READDIRPLUS, NFSV4OP_FSSTAT, NFSV4OP_FSINFO, NFSV4OP_PATHCONF, NFSV4OP_COMMIT, }; static struct mtx nfsrvd_statmtx; MTX_SYSINIT(nfsst, &nfsrvd_statmtx, "NFSstat", MTX_DEF); static void nfsrvd_statstart(int op, struct bintime *now) { if (op > (NFSV42_NOPS + NFSV4OP_FAKENOPS)) { printf("%s: op %d invalid\n", __func__, op); return; } mtx_lock(&nfsrvd_statmtx); - if (nfsstatsv1.srvstartcnt == nfsstatsv1.srvdonecnt) { + if (nfsstatsv1_p->srvstartcnt == nfsstatsv1_p->srvdonecnt) { if (now != NULL) - nfsstatsv1.busyfrom = *now; + nfsstatsv1_p->busyfrom = *now; else - binuptime(&nfsstatsv1.busyfrom); + binuptime(&nfsstatsv1_p->busyfrom); } - nfsstatsv1.srvrpccnt[op]++; - nfsstatsv1.srvstartcnt++; + nfsstatsv1_p->srvrpccnt[op]++; + nfsstatsv1_p->srvstartcnt++; mtx_unlock(&nfsrvd_statmtx); } static void nfsrvd_statend(int op, uint64_t bytes, struct bintime *now, struct bintime *then) { struct bintime dt, lnow; if (op > (NFSV42_NOPS + NFSV4OP_FAKENOPS)) { printf("%s: op %d invalid\n", __func__, op); return; } if (now == NULL) { now = &lnow; binuptime(now); } mtx_lock(&nfsrvd_statmtx); - nfsstatsv1.srvbytes[op] += bytes; - nfsstatsv1.srvops[op]++; + nfsstatsv1_p->srvbytes[op] += bytes; + nfsstatsv1_p->srvops[op]++; if (then != NULL) { dt = *now; bintime_sub(&dt, then); - bintime_add(&nfsstatsv1.srvduration[op], &dt); + bintime_add(&nfsstatsv1_p->srvduration[op], &dt); } dt = *now; - bintime_sub(&dt, &nfsstatsv1.busyfrom); - bintime_add(&nfsstatsv1.busytime, &dt); - nfsstatsv1.busyfrom = *now; + bintime_sub(&dt, &nfsstatsv1_p->busyfrom); + bintime_add(&nfsstatsv1_p->busytime, &dt); + nfsstatsv1_p->busyfrom = *now; - nfsstatsv1.srvdonecnt++; + nfsstatsv1_p->srvdonecnt++; mtx_unlock(&nfsrvd_statmtx); } /* * Do an RPC. Basically, get the file handles translated to vnode pointers * and then call the appropriate server routine. The server routines are * split into groups, based on whether they use a file handle or file * handle plus name or ... * The NFS V4 Compound RPC is performed separately by nfsrvd_compound(). */ void nfsrvd_dorpc(struct nfsrv_descript *nd, int isdgram, u_char *tag, int taglen, u_int32_t minorvers) { int error = 0, lktype; vnode_t vp; mount_t mp; struct nfsrvfh fh; struct nfsexstuff nes; struct mbuf *md; char *dpos; /* * Save the current position in the request mbuf list so * that a rollback to this location can be done upon an * ERELOOKUP error return from an RPC function. */ md = nd->nd_md; dpos = nd->nd_dpos; tryagain: mp = NULL; /* * Get a locked vnode for the first file handle */ if (!(nd->nd_flag & ND_NFSV4)) { KASSERT(nd->nd_repstat == 0, ("nfsrvd_dorpc")); /* * For NFSv3, if the malloc/mget allocation is near limits, * return NFSERR_DELAY. */ if ((nd->nd_flag & ND_NFSV3) && nfsrv_mallocmget_limit()) { nd->nd_repstat = NFSERR_DELAY; vp = NULL; } else { error = nfsrv_mtofh(nd, &fh); if (error) { if (error != EBADRPC) printf("nfs dorpc err1=%d\n", error); nd->nd_repstat = NFSERR_GARBAGE; goto out; } if (nd->nd_procnum == NFSPROC_READ || nd->nd_procnum == NFSPROC_WRITE || nd->nd_procnum == NFSPROC_READDIR || nd->nd_procnum == NFSPROC_READDIRPLUS || nd->nd_procnum == NFSPROC_READLINK || nd->nd_procnum == NFSPROC_GETATTR || nd->nd_procnum == NFSPROC_ACCESS || nd->nd_procnum == NFSPROC_FSSTAT || nd->nd_procnum == NFSPROC_FSINFO) lktype = LK_SHARED; else lktype = LK_EXCLUSIVE; if (nd->nd_flag & ND_PUBLOOKUP) nfsd_fhtovp(nd, &nfs_pubfh, lktype, &vp, &nes, &mp, nfsrv_writerpc[nd->nd_procnum], -1); else nfsd_fhtovp(nd, &fh, lktype, &vp, &nes, &mp, nfsrv_writerpc[nd->nd_procnum], -1); if (nd->nd_repstat == NFSERR_PROGNOTV4) goto out; } } /* * For V2 and 3, set the ND_SAVEREPLY flag for the recent request * cache, as required. * For V4, nfsrvd_compound() does this. */ if (!(nd->nd_flag & ND_NFSV4) && nfsrv_nonidempotent[nd->nd_procnum]) nd->nd_flag |= ND_SAVEREPLY; nfsrvd_rephead(nd); /* * If nd_repstat is non-zero, just fill in the reply status * to complete the RPC reply for V2. Otherwise, you must do * the RPC. */ if (nd->nd_repstat && (nd->nd_flag & ND_NFSV2)) { *nd->nd_errp = nfsd_errmap(nd); nfsrvd_statstart(nfsv3to4op[nd->nd_procnum], /*now*/ NULL); nfsrvd_statend(nfsv3to4op[nd->nd_procnum], /*bytes*/ 0, /*now*/ NULL, /*then*/ NULL); vn_finished_write(mp); goto out; } /* * Now the procedure can be performed. For V4, nfsrvd_compound() * works through the sub-rpcs, otherwise just call the procedure. * The procedures are in three groups with different arguments. * The group is indicated by the value in nfs_retfh[]. */ if (nd->nd_flag & ND_NFSV4) { nfsrvd_compound(nd, isdgram, tag, taglen, minorvers); } else { struct bintime start_time; binuptime(&start_time); nfsrvd_statstart(nfsv3to4op[nd->nd_procnum], &start_time); if (nfs_retfh[nd->nd_procnum] == 1) { if (vp) NFSVOPUNLOCK(vp); error = (*(nfsrv3_procs1[nd->nd_procnum]))(nd, isdgram, vp, NULL, (fhandle_t *)fh.nfsrvfh_data, &nes); } else if (nfs_retfh[nd->nd_procnum] == 2) { error = (*(nfsrv3_procs2[nd->nd_procnum]))(nd, isdgram, vp, NULL, &nes, NULL); } else { error = (*(nfsrv3_procs0[nd->nd_procnum]))(nd, isdgram, vp, &nes); } vn_finished_write(mp); if (error == 0 && nd->nd_repstat == ERELOOKUP) { /* * Roll back to the beginning of the RPC request * arguments. */ nd->nd_md = md; nd->nd_dpos = dpos; /* Free the junk RPC reply and redo the RPC. */ m_freem(nd->nd_mreq); nd->nd_mreq = nd->nd_mb = NULL; nd->nd_repstat = 0; goto tryagain; } nfsrvd_statend(nfsv3to4op[nd->nd_procnum], /*bytes*/ 0, /*now*/ NULL, /*then*/ &start_time); } if (error) { if (error != EBADRPC) printf("nfs dorpc err2=%d\n", error); nd->nd_repstat = NFSERR_GARBAGE; } *nd->nd_errp = nfsd_errmap(nd); /* * Don't cache certain reply status values. */ if (nd->nd_repstat && (nd->nd_flag & ND_SAVEREPLY) && (nd->nd_repstat == NFSERR_GARBAGE || nd->nd_repstat == NFSERR_BADXDR || nd->nd_repstat == NFSERR_MOVED || nd->nd_repstat == NFSERR_DELAY || nd->nd_repstat == NFSERR_BADSEQID || nd->nd_repstat == NFSERR_RESOURCE || nd->nd_repstat == NFSERR_SERVERFAULT || nd->nd_repstat == NFSERR_STALECLIENTID || nd->nd_repstat == NFSERR_STALESTATEID || nd->nd_repstat == NFSERR_OLDSTATEID || nd->nd_repstat == NFSERR_BADSTATEID || nd->nd_repstat == NFSERR_GRACE || nd->nd_repstat == NFSERR_NOGRACE)) nd->nd_flag &= ~ND_SAVEREPLY; out: NFSEXITCODE2(0, nd); } /* * Breaks down a compound RPC request and calls the server routines for * the subprocedures. * Some suboperations are performed directly here to simplify file handle<--> * vnode pointer handling. */ static void nfsrvd_compound(struct nfsrv_descript *nd, int isdgram, u_char *tag, int taglen, u_int32_t minorvers) { int i, lktype, op, op0 = 0, rstat, statsinprog = 0; u_int32_t *tl; struct nfsclient *clp, *nclp; int error = 0, igotlock, nextop, numops, savefhcnt; u_int32_t retops = 0, *retopsp = NULL, *repp; vnode_t vp, nvp, savevp; struct nfsrvfh fh; mount_t new_mp, temp_mp = NULL; struct ucred *credanon; struct nfsexstuff nes, vpnes, savevpnes; fsid_t cur_fsid, save_fsid; static u_int64_t compref = 0; struct bintime start_time; struct thread *p; struct mbuf *mb, *md; char *bpos, *dpos; int bextpg, bextpgsiz; p = curthread; /* Check for and optionally clear the no space flags for DSs. */ nfsrv_checknospc(); NFSVNO_EXINIT(&vpnes); NFSVNO_EXINIT(&savevpnes); /* * Put the seq# of the current compound RPC in nfsrv_descript. * (This is used by nfsrv_checkgetattr(), to see if the write * delegation was created by the same compound RPC as the one * with that Getattr in it.) * Don't worry about the 64bit number wrapping around. It ain't * gonna happen before this server gets shut down/rebooted. */ nd->nd_compref = compref++; /* * Check for and optionally get a lock on the root. This lock means that * no nfsd will be fiddling with the V4 file system and state stuff. It * is required when the V4 root is being changed, the stable storage * restart file is being updated, or callbacks are being done. * When any of the nfsd are processing an NFSv4 compound RPC, they must * either hold a reference count (nfs_usecnt) or the lock. When * nfsrv_unlock() is called to release the lock, it can optionally * also get a reference count, which saves the need for a call to * nfsrv_getref() after nfsrv_unlock(). */ /* * First, check to see if we need to wait for an update lock. */ igotlock = 0; NFSLOCKV4ROOTMUTEX(); - if (nfsrv_stablefirst.nsf_flags & NFSNSF_NEEDLOCK) + if (NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_NEEDLOCK) igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); else igotlock = nfsv4_lock(&nfsv4rootfs_lock, 0, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); if (igotlock) { /* * If I got the lock, I can update the stable storage file. * Done when the grace period is over or a client has long * since expired. */ - nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NEEDLOCK; - if ((nfsrv_stablefirst.nsf_flags & + NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NEEDLOCK; + if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_GRACEOVER | NFSNSF_UPDATEDONE)) == NFSNSF_GRACEOVER) nfsrv_updatestable(p); /* * If at least one client has long since expired, search * the client list for them, write a REVOKE record on the * stable storage file and then remove them from the client * list. */ - if (nfsrv_stablefirst.nsf_flags & NFSNSF_EXPIREDCLIENT) { - nfsrv_stablefirst.nsf_flags &= ~NFSNSF_EXPIREDCLIENT; + if (NFSD_VNET(nfsrv_stablefirst).nsf_flags & + NFSNSF_EXPIREDCLIENT) { + NFSD_VNET(nfsrv_stablefirst).nsf_flags &= + ~NFSNSF_EXPIREDCLIENT; for (i = 0; i < nfsrv_clienthashsize; i++) { - LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, - nclp) { + LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], + lc_hash, nclp) { if (clp->lc_flags & LCL_EXPIREIT) { if (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg)) nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); nfsrv_zapclient(clp, p); } } } } NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } else { /* * If we didn't get the lock, we need to get a refcnt, * which also checks for and waits for the lock. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); } /* * If flagged, search for open owners that haven't had any opens * for a long time. */ - if (nfsrv_stablefirst.nsf_flags & NFSNSF_NOOPENS) { + if (NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_NOOPENS) { nfsrv_throwawayopens(p); } /* Do a CBLAYOUTRECALL callback if over the high water mark. */ if (nfsrv_layoutcnt > nfsrv_layouthighwater) nfsrv_recalloldlayout(p); savevp = vp = NULL; save_fsid.val[0] = save_fsid.val[1] = 0; cur_fsid.val[0] = cur_fsid.val[1] = 0; nextop = -1; savefhcnt = 0; /* If taglen < 0, there was a parsing error in nfsd_getminorvers(). */ if (taglen < 0) { error = EBADRPC; goto nfsmout; } (void) nfsm_strtom(nd, tag, taglen); NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED); NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if ((minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION && minorvers != NFSV42_MINORVERSION) || minorvers < nfs_minminorv4 || minorvers > nfs_maxminorv4) nd->nd_repstat = NFSERR_MINORVERMISMATCH; if (nd->nd_repstat) numops = 0; else numops = fxdr_unsigned(int, *tl); /* * Loop around doing the sub ops. * vp - is an unlocked vnode pointer for the CFH * savevp - is an unlocked vnode pointer for the SAVEDFH * (at some future date, it might turn out to be more appropriate * to keep the file handles instead of vnode pointers?) * savevpnes and vpnes - are the export flags for the above. */ for (i = 0; i < numops; i++) { NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED); if (savefhcnt > 0) { op = NFSV4OP_SAVEFH; *repp = txdr_unsigned(op); savefhcnt--; } else if (nextop == -1) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); *repp = *tl; op = fxdr_unsigned(int, *tl); } else { op = nextop; *repp = txdr_unsigned(op); nextop = -1; } NFSD_DEBUG(4, "op=%d\n", op); if (op < NFSV4OP_ACCESS || op >= NFSV42_NOPS || (op >= NFSV4OP_NOPS && (nd->nd_flag & ND_NFSV41) == 0) || (op >= NFSV41_NOPS && (nd->nd_flag & ND_NFSV42) == 0)) { nd->nd_repstat = NFSERR_OPILLEGAL; *repp++ = txdr_unsigned(NFSV4OP_OPILLEGAL); *repp = nfsd_errmap(nd); retops++; break; } else { repp++; } binuptime(&start_time); nfsrvd_statstart(op, &start_time); statsinprog = 1; if (i == 0) op0 = op; if (i == numops - 1) nd->nd_flag |= ND_LASTOP; /* * Check for a referral on the current FH and, if so, return * NFSERR_MOVED for all ops that allow it, except Getattr. */ if (vp != NULL && op != NFSV4OP_GETATTR && nfsv4root_getreferral(vp, NULL, 0) != NULL && nfsrv_errmoved(op)) { nd->nd_repstat = NFSERR_MOVED; *repp = nfsd_errmap(nd); retops++; break; } /* * For NFSv4.1, check for a Sequence Operation being first * or one of the other allowed operations by itself. */ if ((nd->nd_flag & ND_NFSV41) != 0) { if (i != 0 && op == NFSV4OP_SEQUENCE) nd->nd_repstat = NFSERR_SEQUENCEPOS; else if (i == 0 && op != NFSV4OP_SEQUENCE && op != NFSV4OP_EXCHANGEID && op != NFSV4OP_CREATESESSION && op != NFSV4OP_BINDCONNTOSESS && op != NFSV4OP_DESTROYCLIENTID && op != NFSV4OP_DESTROYSESSION) nd->nd_repstat = NFSERR_OPNOTINSESS; else if (i != 0 && op0 != NFSV4OP_SEQUENCE) nd->nd_repstat = NFSERR_NOTONLYOP; if (nd->nd_repstat != 0) { *repp = nfsd_errmap(nd); retops++; break; } } nd->nd_procnum = op; /* * If over flood level, reply NFSERR_RESOURCE, if at the first * Op. (Since a client recovery from NFSERR_RESOURCE can get * really nasty for certain Op sequences, I'll play it safe * and only return the error at the beginning.) The cache * will still function over flood level, but uses lots of * mbufs.) * If nfsrv_mallocmget_limit() returns True, the system is near * to its limit for memory that malloc()/mget() can allocate. */ if (i == 0 && (nd->nd_rp == NULL || nd->nd_rp->rc_refcnt == 0) && (nfsrv_mallocmget_limit() || - nfsrc_tcpsavedreplies > nfsrc_floodlevel)) { - if (nfsrc_tcpsavedreplies > nfsrc_floodlevel) + NFSD_VNET(nfsrc_tcpsavedreplies) > + NFSD_VNET(nfsrc_floodlevel))) { + if (NFSD_VNET(nfsrc_tcpsavedreplies) > + NFSD_VNET(nfsrc_floodlevel)) printf("nfsd server cache flooded, try " "increasing vfs.nfsd.tcphighwater\n"); nd->nd_repstat = NFSERR_RESOURCE; *repp = nfsd_errmap(nd); if (op == NFSV4OP_SETATTR) { /* * Setattr replies require a bitmap. * even for errors like these. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = 0; } retops++; break; } if (nfsv4_opflag[op].savereply) nd->nd_flag |= ND_SAVEREPLY; switch (op) { case NFSV4OP_PUTFH: error = nfsrv_mtofh(nd, &fh); if (error) goto nfsmout; if ((nd->nd_flag & ND_LASTOP) == 0) { /* * Pre-parse the next op#. If it is * SaveFH, count it and skip to the * next op#, if not the last op#. * nextop is used to determine if * NFSERR_WRONGSEC can be returned, * per RFC5661 Sec. 2.6. */ do { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); nextop = fxdr_unsigned(int, *tl); if (nextop == NFSV4OP_SAVEFH && i < numops - 1) savefhcnt++; } while (nextop == NFSV4OP_SAVEFH && i < numops - 1); } if (!nd->nd_repstat) nfsd_fhtovp(nd, &fh, LK_SHARED, &nvp, &nes, NULL, 0, nextop); /* For now, allow this for non-export FHs */ if (!nd->nd_repstat) { if (vp) vrele(vp); vp = nvp; cur_fsid = vp->v_mount->mnt_stat.f_fsid; NFSVOPUNLOCK(vp); vpnes = nes; } break; case NFSV4OP_PUTPUBFH: if (nfs_pubfhset) { if ((nd->nd_flag & ND_LASTOP) == 0) { /* * Pre-parse the next op#. If it is * SaveFH, count it and skip to the * next op#, if not the last op#. * nextop is used to determine if * NFSERR_WRONGSEC can be returned, * per RFC5661 Sec. 2.6. */ do { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); nextop = fxdr_unsigned(int, *tl); if (nextop == NFSV4OP_SAVEFH && i < numops - 1) savefhcnt++; } while (nextop == NFSV4OP_SAVEFH && i < numops - 1); } nfsd_fhtovp(nd, &nfs_pubfh, LK_SHARED, &nvp, &nes, NULL, 0, nextop); } else nd->nd_repstat = NFSERR_NOFILEHANDLE; if (!nd->nd_repstat) { if (vp) vrele(vp); vp = nvp; cur_fsid = vp->v_mount->mnt_stat.f_fsid; NFSVOPUNLOCK(vp); vpnes = nes; } break; case NFSV4OP_PUTROOTFH: - if (nfs_rootfhset) { + if (NFSD_VNET(nfs_rootfhset)) { if ((nd->nd_flag & ND_LASTOP) == 0) { /* * Pre-parse the next op#. If it is * SaveFH, count it and skip to the * next op#, if not the last op#. * nextop is used to determine if * NFSERR_WRONGSEC can be returned, * per RFC5661 Sec. 2.6. */ do { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); nextop = fxdr_unsigned(int, *tl); if (nextop == NFSV4OP_SAVEFH && i < numops - 1) savefhcnt++; } while (nextop == NFSV4OP_SAVEFH && i < numops - 1); } - nfsd_fhtovp(nd, &nfs_rootfh, LK_SHARED, &nvp, - &nes, NULL, 0, nextop); + nfsd_fhtovp(nd, &NFSD_VNET(nfs_rootfh), + LK_SHARED, &nvp, &nes, NULL, 0, nextop); if (!nd->nd_repstat) { if (vp) vrele(vp); vp = nvp; cur_fsid = vp->v_mount->mnt_stat.f_fsid; NFSVOPUNLOCK(vp); vpnes = nes; } } else nd->nd_repstat = NFSERR_NOFILEHANDLE; break; case NFSV4OP_SAVEFH: if (vp && NFSVNO_EXPORTED(&vpnes)) { nd->nd_repstat = 0; /* If vp == savevp, a no-op */ if (vp != savevp) { if (savevp) vrele(savevp); VREF(vp); savevp = vp; savevpnes = vpnes; save_fsid = cur_fsid; } if ((nd->nd_flag & ND_CURSTATEID) != 0) { nd->nd_savedcurstateid = nd->nd_curstateid; nd->nd_flag |= ND_SAVEDCURSTATEID; } } else { nd->nd_repstat = NFSERR_NOFILEHANDLE; } break; case NFSV4OP_RESTOREFH: if (savevp) { if ((nd->nd_flag & ND_LASTOP) == 0) { /* * Pre-parse the next op#. If it is * SaveFH, count it and skip to the * next op#, if not the last op#. * nextop is used to determine if * NFSERR_WRONGSEC can be returned, * per RFC5661 Sec. 2.6. */ do { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); nextop = fxdr_unsigned(int, *tl); if (nextop == NFSV4OP_SAVEFH && i < numops - 1) savefhcnt++; } while (nextop == NFSV4OP_SAVEFH && i < numops - 1); } nd->nd_repstat = 0; /* If vp == savevp, a no-op */ if (vp != savevp) { if (nfsrv_checkwrongsec(nd, nextop, savevp->v_type)) nd->nd_repstat = nfsvno_testexp(nd, &savevpnes); if (nd->nd_repstat == 0) { VREF(savevp); vrele(vp); vp = savevp; vpnes = savevpnes; cur_fsid = save_fsid; } } if (nd->nd_repstat == 0 && (nd->nd_flag & ND_SAVEDCURSTATEID) != 0) { nd->nd_curstateid = nd->nd_savedcurstateid; nd->nd_flag |= ND_CURSTATEID; } } else { nd->nd_repstat = NFSERR_RESTOREFH; } break; default: /* * Allow a Lookup, Getattr, GetFH, Secinfo on an * non-exported directory if * nfs_rootfhset. Do I need to allow any other Ops? * (You can only have a non-exported vpnes if * nfs_rootfhset is true. See nfsd_fhtovp()) * Allow AUTH_SYS to be used for file systems * exported GSS only for certain Ops, to allow * clients to do mounts more easily. */ if (nfsv4_opflag[op].needscfh && vp) { if (!NFSVNO_EXPORTED(&vpnes) && op != NFSV4OP_LOOKUP && op != NFSV4OP_GETATTR && op != NFSV4OP_GETFH && op != NFSV4OP_ACCESS && op != NFSV4OP_READLINK && op != NFSV4OP_SECINFO && op != NFSV4OP_SECINFONONAME) nd->nd_repstat = NFSERR_NOFILEHANDLE; if (nd->nd_repstat) { if (op == NFSV4OP_SETATTR) { /* * Setattr reply requires a bitmap * even for errors like these. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = 0; } break; } } /* * Save the current positions in the mbuf lists so * that a rollback to this location can be done upon a * redo due to a ERELOOKUP return for a operation. */ mb = nd->nd_mb; bpos = nd->nd_bpos; bextpg = nd->nd_bextpg; bextpgsiz = nd->nd_bextpgsiz; md = nd->nd_md; dpos = nd->nd_dpos; tryagain: if (nfsv4_opflag[op].retfh == 1) { if (!vp) { nd->nd_repstat = NFSERR_NOFILEHANDLE; break; } if (NFSVNO_EXPORTED(&vpnes) && (op == NFSV4OP_LOOKUP || op == NFSV4OP_LOOKUPP || (op == NFSV4OP_OPEN && vp->v_type == VDIR))) { /* Check for wrong security. */ rstat = nfsvno_testexp(nd, &vpnes); if (rstat != 0) { nd->nd_repstat = rstat; break; } } VREF(vp); if (nfsv4_opflag[op].modifyfs) vn_start_write(vp, &temp_mp, V_WAIT); error = (*(nfsrv4_ops1[op]))(nd, isdgram, vp, &nvp, (fhandle_t *)fh.nfsrvfh_data, &vpnes); if (!error && !nd->nd_repstat) { if (op == NFSV4OP_LOOKUP || op == NFSV4OP_LOOKUPP) { new_mp = nvp->v_mount; if (fsidcmp(&cur_fsid, &new_mp->mnt_stat.f_fsid) != 0) { /* crossed a server mount point */ nd->nd_repstat = nfsvno_checkexp(new_mp, nd->nd_nam, &nes, &credanon); if (!nd->nd_repstat) nd->nd_repstat = nfsd_excred(nd, &nes, credanon, true); if (credanon != NULL) crfree(credanon); if (!nd->nd_repstat) { vpnes = nes; cur_fsid = new_mp->mnt_stat.f_fsid; } } /* Lookup ops return a locked vnode */ NFSVOPUNLOCK(nvp); } if (!nd->nd_repstat) { vrele(vp); vp = nvp; } else vrele(nvp); } if (nfsv4_opflag[op].modifyfs) vn_finished_write(temp_mp); } else if (nfsv4_opflag[op].retfh == 2) { if (vp == NULL || savevp == NULL) { nd->nd_repstat = NFSERR_NOFILEHANDLE; break; } else if (fsidcmp(&cur_fsid, &save_fsid) != 0) { nd->nd_repstat = NFSERR_XDEV; break; } if (nfsv4_opflag[op].modifyfs) vn_start_write(savevp, &temp_mp, V_WAIT); if (NFSVOPLOCK(savevp, LK_EXCLUSIVE) == 0) { VREF(vp); VREF(savevp); error = (*(nfsrv4_ops2[op]))(nd, isdgram, savevp, vp, &savevpnes, &vpnes); } else nd->nd_repstat = NFSERR_PERM; if (nfsv4_opflag[op].modifyfs) vn_finished_write(temp_mp); } else { if (nfsv4_opflag[op].retfh != 0) panic("nfsrvd_compound"); if (nfsv4_opflag[op].needscfh) { if (vp != NULL) { lktype = nfsv4_opflag[op].lktype; if (nfsv4_opflag[op].modifyfs) { vn_start_write(vp, &temp_mp, V_WAIT); if (op == NFSV4OP_WRITE && MNT_SHARED_WRITES(temp_mp)) lktype = LK_SHARED; } if (NFSVOPLOCK(vp, lktype) == 0) VREF(vp); else nd->nd_repstat = NFSERR_PERM; } else { nd->nd_repstat = NFSERR_NOFILEHANDLE; if (op == NFSV4OP_SETATTR) { /* * Setattr reply requires a * bitmap even for errors like * these. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = 0; } break; } if (nd->nd_repstat == 0) { error = (*(nfsrv4_ops0[op]))(nd, isdgram, vp, &vpnes); if ((op == NFSV4OP_SECINFO || op == NFSV4OP_SECINFONONAME) && error == 0 && nd->nd_repstat == 0) { /* * Secinfo and Secinfo_no_name * consume the current FH. */ vrele(vp); vp = NULL; } } if (nfsv4_opflag[op].modifyfs) vn_finished_write(temp_mp); } else { error = (*(nfsrv4_ops0[op]))(nd, isdgram, NULL, &vpnes); } } } if (error) { if (error == EBADRPC || error == NFSERR_BADXDR) { nd->nd_repstat = NFSERR_BADXDR; } else { nd->nd_repstat = error; printf("nfsv4 comperr0=%d\n", error); } error = 0; } if (nd->nd_repstat == ERELOOKUP) { /* * Roll back to the beginning of the operation * arguments. */ nd->nd_md = md; nd->nd_dpos = dpos; /* * Trim off the bogus reply for this operation * and redo the operation. */ nfsm_trimtrailing(nd, mb, bpos, bextpg, bextpgsiz); nd->nd_repstat = 0; nd->nd_flag |= ND_ERELOOKUP; goto tryagain; } nd->nd_flag &= ~ND_ERELOOKUP; if (statsinprog != 0) { nfsrvd_statend(op, /*bytes*/ 0, /*now*/ NULL, /*then*/ &start_time); statsinprog = 0; } retops++; if (nd->nd_repstat) { *repp = nfsd_errmap(nd); break; } else { *repp = 0; /* NFS4_OK */ } } nfsmout: if (statsinprog != 0) { nfsrvd_statend(op, /*bytes*/ 0, /*now*/ NULL, /*then*/ &start_time); statsinprog = 0; } if (error) { if (error == EBADRPC || error == NFSERR_BADXDR) nd->nd_repstat = NFSERR_BADXDR; else printf("nfsv4 comperr1=%d\n", error); } if (taglen == -1) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; *tl = 0; } else { *retopsp = txdr_unsigned(retops); } if (vp) vrele(vp); if (savevp) vrele(savevp); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); NFSEXITCODE2(0, nd); } diff --git a/sys/fs/nfsserver/nfs_nfsdstate.c b/sys/fs/nfsserver/nfs_nfsdstate.c index 26bdeb1c641f..20f8b28d32fd 100644 --- a/sys/fs/nfsserver/nfs_nfsdstate.c +++ b/sys/fs/nfsserver/nfs_nfsdstate.c @@ -1,8869 +1,8885 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009 Rick Macklem, University of Guelph * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include -struct nfsrv_stablefirst nfsrv_stablefirst; int nfsrv_issuedelegs = 0; int nfsrv_dolocallocks = 0; struct nfsv4lock nfsv4rootfs_lock; time_t nfsdev_time = 0; int nfsrv_layouthashsize; volatile int nfsrv_layoutcnt = 0; -extern uint32_t nfs_srvmaxio; -extern int newnfs_numnfsd; -extern struct nfsstatsv1 nfsstatsv1; +NFSD_VNET_DEFINE(struct nfsrv_stablefirst, nfsrv_stablefirst); + +NFSD_VNET_DECLARE(int, nfsrv_numnfsd); +NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); + +extern uint32_t nfs_srvmaxio; extern int nfsrv_lease; extern struct timeval nfsboottime; extern u_int32_t newnfs_true, newnfs_false; extern struct mtx nfsrv_dslock_mtx; extern struct mtx nfsrv_recalllock_mtx; extern struct mtx nfsrv_dontlistlock_mtx; extern int nfsd_debuglevel; extern u_int nfsrv_dsdirsize; extern struct nfsdevicehead nfsrv_devidhead; extern int nfsrv_doflexfile; extern int nfsrv_maxpnfsmirror; NFSV4ROOTLOCKMUTEX; NFSSTATESPINLOCK; extern struct nfsdontlisthead nfsrv_dontlisthead; extern volatile int nfsrv_devidcnt; extern struct nfslayouthead nfsrv_recalllisthead; extern char *nfsrv_zeropnfsdat; SYSCTL_DECL(_vfs_nfsd); int nfsrv_statehashsize = NFSSTATEHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN, &nfsrv_statehashsize, 0, "Size of state hash table set via loader.conf"); int nfsrv_clienthashsize = NFSCLIENTHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN, &nfsrv_clienthashsize, 0, "Size of client hash table set via loader.conf"); int nfsrv_lockhashsize = NFSLOCKHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN, &nfsrv_lockhashsize, 0, "Size of file handle hash table set via loader.conf"); int nfsrv_sessionhashsize = NFSSESSIONHASHSIZE; SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN, &nfsrv_sessionhashsize, 0, "Size of session hash table set via loader.conf"); int nfsrv_layouthighwater = NFSLAYOUTHIGHWATER; SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN, &nfsrv_layouthighwater, 0, "High water mark for number of layouts set via loader.conf"); static int nfsrv_v4statelimit = NFSRV_V4STATELIMIT; SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN, &nfsrv_v4statelimit, 0, "High water limit for NFSv4 opens+locks+delegations"); static int nfsrv_writedelegifpos = 0; SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW, &nfsrv_writedelegifpos, 0, "Issue a write delegation for read opens if possible"); static int nfsrv_allowreadforwriteopen = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW, &nfsrv_allowreadforwriteopen, 0, "Allow Reads to be done with Write Access StateIDs"); int nfsrv_pnfsatime = 0; SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW, &nfsrv_pnfsatime, 0, "For pNFS service, do Getattr ops to keep atime up-to-date"); int nfsrv_flexlinuxhack = 0; SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW, &nfsrv_flexlinuxhack, 0, "For Linux clients, hack around Flex File Layout bug"); /* * Hash lists for nfs V4. */ -struct nfsclienthashhead *nfsclienthash; -struct nfslockhashhead *nfslockhash; -struct nfssessionhash *nfssessionhash; +NFSD_VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash); +NFSD_VNET_DEFINE(struct nfslockhashhead *, nfslockhash); +NFSD_VNET_DEFINE(struct nfssessionhash *, nfssessionhash); + struct nfslayouthash *nfslayouthash; volatile int nfsrv_dontlistlen = 0; static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0; -static time_t nfsrvboottime; static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0; static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER; static int nfsrv_nogsscallback = 0; static volatile int nfsrv_writedelegcnt = 0; static int nfsrv_faildscnt; +NFSD_VNET_DEFINE_STATIC(time_t, nfsrvboottime); + /* local functions */ static void nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp); static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p); static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p); static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p); static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p); static void nfsrv_freenfslock(struct nfslock *lop); static void nfsrv_freenfslockfile(struct nfslockfile *lfp); static void nfsrv_freedeleg(struct nfsstate *); static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, u_int32_t flags, struct nfsstate **stpp); static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, struct nfsstate **stpp); static int nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p); static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit); static void nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp); static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, struct nfslock **other_lopp, struct nfslockfile *lfp); static int nfsrv_getipnumber(u_char *cp); static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, nfsv4stateid_t *stateidp, int specialid); static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, u_int32_t flags); static int nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p); static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, uint32_t callback, int op, const char *optag, struct nfsdsession **sepp, int *slotposp); static u_int32_t nfsrv_nextclientindex(void); static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp); static void nfsrv_markstable(struct nfsclient *clp); static void nfsrv_markreclaim(struct nfsclient *clp); static int nfsrv_checkstable(struct nfsclient *clp); static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct vnode *vp, NFSPROC_T *p); static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p, vnode_t vp); static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, struct nfsclient *clp, int *haslockp, NFSPROC_T *p); static int nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp); static time_t nfsrv_leaseexpiry(void); static void nfsrv_delaydelegtimeout(struct nfsstate *stp); static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, struct nfsstate *stp, struct nfsrvcache *op); static int nfsrv_nootherstate(struct nfsstate *stp); static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p); static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first, uint64_t init_end, NFSPROC_T *p); static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p); static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p); static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end); static void nfsrv_locklf(struct nfslockfile *lfp); static void nfsrv_unlocklf(struct nfslockfile *lfp); static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid); static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid); static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, int dont_replycache, struct nfsdsession **sepp, int *slotposp); static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp); static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p); static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp); static void nfsrv_freelayoutlist(nfsquad_t clientid); static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode); static void nfsrv_freealllayouts(void); static void nfsrv_freedevid(struct nfsdevice *ds); static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, struct nfsdevice **dsp); static void nfsrv_deleteds(struct nfsdevice *fndds); static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost); static void nfsrv_freealldevids(void); static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt, NFSPROC_T *p); static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p); static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, NFSPROC_T *, struct nfslayout **lypp); static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt); static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs); static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs); static int nfsrv_dontlayout(fhandle_t *fhp); static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, vnode_t *tvpp); static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp); /* * Scan the client list for a match and either return the current one, * create a new entry or return an error. * If returning a non-error, the clp structure must either be linked into * the client list or free'd. */ int nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp, nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p) { struct nfsclient *clp = NULL, *new_clp = *new_clpp; int i, error = 0, ret; struct nfsstate *stp, *tstp; #ifdef INET struct sockaddr_in *sin, *rin; #endif #ifdef INET6 struct sockaddr_in6 *sin6, *rin6; #endif struct nfsdsession *sep, *nsep; int zapit = 0, gotit, hasstate = 0, igotlock; static u_int64_t confirm_index = 0; /* * Check for state resource limit exceeded. */ if (nfsrv_openpluslock > nfsrv_v4statelimit) { error = NFSERR_RESOURCE; goto out; } if (nfsrv_issuedelegs == 0 || ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0)) /* * Don't do callbacks when delegations are disabled or * for AUTH_GSS unless enabled via nfsrv_nogsscallback. * If establishing a callback connection is attempted * when a firewall is blocking the callback path, the * server may wait too long for the connect attempt to * succeed during the Open. Some clients, such as Linux, * may timeout and give up on the Open before the server * replies. Also, since AUTH_GSS callbacks are not * yet interoperability tested, they might cause the * server to crap out, if they get past the Init call to * the client. */ new_clp->lc_program = 0; /* Lock out other nfsd threads */ NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!igotlock); NFSUNLOCKV4ROOTMUTEX(); /* * Search for a match in the client list. */ gotit = i = 0; while (i < nfsrv_clienthashsize && !gotit) { - LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { + LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) { if (new_clp->lc_idlen == clp->lc_idlen && !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) { gotit = 1; break; } } if (gotit == 0) i++; } if (!gotit || (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) { if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) { /* * For NFSv4.1, if confirmp->lval[1] is non-zero, the * client is trying to update a confirmed clientid. */ NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); confirmp->lval[1] = 0; error = NFSERR_NOENT; goto out; } /* * Get rid of the old one. */ if (i != nfsrv_clienthashsize) { LIST_REMOVE(clp, lc_hash); nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); zapit = 1; } /* * Add it after assigning a client id to it. */ new_clp->lc_flags |= LCL_NEEDSCONFIRM; if ((nd->nd_flag & ND_NFSV41) != 0) { confirmp->lval[0] = ++confirm_index; new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1; } else confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; clientidp->lval[0] = new_clp->lc_clientid.lval[0] = - (u_int32_t)nfsrvboottime; + NFSD_VNET(nfsrvboottime); clientidp->lval[1] = new_clp->lc_clientid.lval[1] = nfsrv_nextclientindex(); new_clp->lc_stateindex = 0; new_clp->lc_statemaxindex = 0; new_clp->lc_prevsess = 0; new_clp->lc_cbref = 0; new_clp->lc_expiry = nfsrv_leaseexpiry(); LIST_INIT(&new_clp->lc_open); LIST_INIT(&new_clp->lc_deleg); LIST_INIT(&new_clp->lc_olddeleg); LIST_INIT(&new_clp->lc_session); for (i = 0; i < nfsrv_statehashsize; i++) LIST_INIT(&new_clp->lc_stateid[i]); LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, lc_hash); - nfsstatsv1.srvclients++; + nfsstatsv1_p->srvclients++; nfsrv_openpluslock++; nfsrv_clients++; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); if (zapit) nfsrv_zapclient(clp, p); *new_clpp = NULL; goto out; } /* * Now, handle the cases where the id is already issued. */ if (nfsrv_notsamecredname(nd, clp)) { /* * Check to see if there is expired state that should go away. */ if (clp->lc_expiry < NFSD_MONOSEC && (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) { nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); } /* * If there is outstanding state, then reply NFSERR_CLIDINUSE per * RFC3530 Sec. 8.1.2 last para. */ if (!LIST_EMPTY(&clp->lc_deleg)) { hasstate = 1; } else if (LIST_EMPTY(&clp->lc_open)) { hasstate = 0; } else { hasstate = 0; /* Look for an Open on the OpenOwner */ LIST_FOREACH(stp, &clp->lc_open, ls_list) { if (!LIST_EMPTY(&stp->ls_open)) { hasstate = 1; break; } } } if (hasstate) { /* * If the uid doesn't match, return NFSERR_CLIDINUSE after * filling out the correct ipaddr and portnum. */ switch (clp->lc_req.nr_nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam; rin = (struct sockaddr_in *)clp->lc_req.nr_nam; sin->sin_addr.s_addr = rin->sin_addr.s_addr; sin->sin_port = rin->sin_port; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam; rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; sin6->sin6_addr = rin6->sin6_addr; sin6->sin6_port = rin6->sin6_port; break; #endif } NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_CLIDINUSE; goto out; } } if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) { /* * If the verifier has changed, the client has rebooted * and a new client id is issued. The old state info * can be thrown away once the SETCLIENTID_CONFIRM occurs. */ LIST_REMOVE(clp, lc_hash); /* Get rid of all sessions on this clientid. */ LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) { ret = nfsrv_freesession(sep, NULL); if (ret != 0) printf("nfsrv_setclient: verifier changed free" " session failed=%d\n", ret); } new_clp->lc_flags |= LCL_NEEDSCONFIRM; if ((nd->nd_flag & ND_NFSV41) != 0) { confirmp->lval[0] = ++confirm_index; new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1; } else confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; clientidp->lval[0] = new_clp->lc_clientid.lval[0] = - nfsrvboottime; + NFSD_VNET(nfsrvboottime); clientidp->lval[1] = new_clp->lc_clientid.lval[1] = nfsrv_nextclientindex(); new_clp->lc_stateindex = 0; new_clp->lc_statemaxindex = 0; new_clp->lc_prevsess = 0; new_clp->lc_cbref = 0; new_clp->lc_expiry = nfsrv_leaseexpiry(); /* * Save the state until confirmed. */ LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list); LIST_FOREACH(tstp, &new_clp->lc_open, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list) tstp->ls_clp = new_clp; for (i = 0; i < nfsrv_statehashsize; i++) { LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i], ls_hash); LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash) tstp->ls_clp = new_clp; } LIST_INIT(&new_clp->lc_session); LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, lc_hash); - nfsstatsv1.srvclients++; + nfsstatsv1_p->srvclients++; nfsrv_openpluslock++; nfsrv_clients++; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); /* * Must wait until any outstanding callback on the old clp * completes. */ NFSLOCKSTATE(); while (clp->lc_cbref) { clp->lc_flags |= LCL_WAKEUPWANTED; (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsd clp", 10 * hz); } NFSUNLOCKSTATE(); nfsrv_zapclient(clp, p); *new_clpp = NULL; goto out; } /* For NFSv4.1, mark that we found a confirmed clientid. */ if ((nd->nd_flag & ND_NFSV41) != 0) { clientidp->lval[0] = clp->lc_clientid.lval[0]; clientidp->lval[1] = clp->lc_clientid.lval[1]; confirmp->lval[0] = 0; /* Ignored by client */ confirmp->lval[1] = 1; } else { /* * id and verifier match, so update the net address info * and get rid of any existing callback authentication * handle, so a new one will be acquired. */ LIST_REMOVE(clp, lc_hash); new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN); new_clp->lc_expiry = nfsrv_leaseexpiry(); confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; clientidp->lval[0] = new_clp->lc_clientid.lval[0] = clp->lc_clientid.lval[0]; clientidp->lval[1] = new_clp->lc_clientid.lval[1] = clp->lc_clientid.lval[1]; new_clp->lc_delegtime = clp->lc_delegtime; new_clp->lc_stateindex = clp->lc_stateindex; new_clp->lc_statemaxindex = clp->lc_statemaxindex; new_clp->lc_cbref = 0; LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list); LIST_FOREACH(tstp, &new_clp->lc_open, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list) tstp->ls_clp = new_clp; LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list); LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list) tstp->ls_clp = new_clp; for (i = 0; i < nfsrv_statehashsize; i++) { LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i], ls_hash); LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash) tstp->ls_clp = new_clp; } LIST_INIT(&new_clp->lc_session); LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, lc_hash); - nfsstatsv1.srvclients++; + nfsstatsv1_p->srvclients++; nfsrv_openpluslock++; nfsrv_clients++; } NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); if ((nd->nd_flag & ND_NFSV41) == 0) { /* * Must wait until any outstanding callback on the old clp * completes. */ NFSLOCKSTATE(); while (clp->lc_cbref) { clp->lc_flags |= LCL_WAKEUPWANTED; (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsdclp", 10 * hz); } NFSUNLOCKSTATE(); nfsrv_zapclient(clp, p); *new_clpp = NULL; } out: NFSEXITCODE2(error, nd); return (error); } /* * Check to see if the client id exists and optionally confirm it. */ int nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp, struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram, struct nfsrv_descript *nd, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; int i; struct nfsclienthashhead *hp; int error = 0, igotlock, doneok; struct nfssessionhash *shp; struct nfsdsession *sep; uint64_t sessid[2]; bool sess_replay; static uint64_t next_sess = 0; if (clpp) *clpp = NULL; if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 || - opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) { + opflags != CLOPS_RENEW) && NFSD_VNET(nfsrvboottime) != + clientid.lval[0]) { error = NFSERR_STALECLIENTID; goto out; } /* * If called with opflags == CLOPS_RENEW, the State Lock is * already held. Otherwise, we need to get either that or, * for the case of Confirm, lock out the nfsd threads. */ if (opflags & CLOPS_CONFIRM) { NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!igotlock); /* * Create a new sessionid here, since we need to do it where * there is a mutex held to serialize update of next_sess. */ if ((nd->nd_flag & ND_NFSV41) != 0) { sessid[0] = ++next_sess; sessid[1] = clientid.qval; } NFSUNLOCKV4ROOTMUTEX(); } else if (opflags != CLOPS_RENEW) { NFSLOCKSTATE(); } /* For NFSv4.1, the clp is acquired from the associated session. */ if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 && opflags == CLOPS_RENEW) { clp = NULL; if ((nd->nd_flag & ND_HASSEQUENCE) != 0) { shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep != NULL) clp = sep->sess_clp; NFSUNLOCKSESSION(shp); } } else { hp = NFSCLIENTHASH(clientid); LIST_FOREACH(clp, hp, lc_hash) { if (clp->lc_clientid.lval[1] == clientid.lval[1]) break; } } if (clp == NULL) { if (opflags & CLOPS_CONFIRM) error = NFSERR_STALECLIENTID; else error = NFSERR_EXPIRED; } else if (clp->lc_flags & LCL_ADMINREVOKED) { /* * If marked admin revoked, just return the error. */ error = NFSERR_ADMINREVOKED; } if (error) { if (opflags & CLOPS_CONFIRM) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } else if (opflags != CLOPS_RENEW) { NFSUNLOCKSTATE(); } goto out; } /* * Perform any operations specified by the opflags. */ if (opflags & CLOPS_CONFIRM) { sess_replay = false; if ((nd->nd_flag & ND_NFSV41) != 0) { /* * For the case where lc_confirm.lval[0] == confirm.lval[0], * use the new session, but with the previous sessionid. * This is not exactly what the RFC describes, but should * result in the same reply as the previous CreateSession. */ if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) { clp->lc_confirm.lval[0] = confirm.lval[0]; clp->lc_prevsess = sessid[0]; } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) { if (clp->lc_prevsess == 0) error = NFSERR_SEQMISORDERED; else sessid[0] = clp->lc_prevsess; sess_replay = true; } else error = NFSERR_SEQMISORDERED; } else if ((nd->nd_flag & ND_NFSV41) == 0 && clp->lc_confirm.qval != confirm.qval) error = NFSERR_STALECLIENTID; if (error == 0 && nfsrv_notsamecredname(nd, clp)) error = NFSERR_CLIDINUSE; if (!error) { if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) == LCL_NEEDSCONFIRM) { /* * Hang onto the delegations (as old delegations) * for an Open with CLAIM_DELEGATE_PREV unless in * grace, but get rid of the rest of the state. */ nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_olddeleg); if (nfsrv_checkgrace(nd, clp, 0)) { /* In grace, so just delete delegations */ nfsrv_freedeleglist(&clp->lc_deleg); } else { LIST_FOREACH(stp, &clp->lc_deleg, ls_list) stp->ls_flags |= NFSLCK_OLDDELEG; clp->lc_delegtime = NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA; LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg, ls_list); } if ((nd->nd_flag & ND_NFSV41) != 0) clp->lc_program = cbprogram; } clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN); if (clp->lc_program) clp->lc_flags |= LCL_NEEDSCBNULL; /* For NFSv4.1, link the session onto the client. */ if (nsep != NULL) { /* Hold a reference on the xprt for a backchannel. */ if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 && !sess_replay) { if (clp->lc_req.nr_client == NULL) clp->lc_req.nr_client = (struct __rpc_client *) clnt_bck_create(nd->nd_xprt->xp_socket, cbprogram, NFSV4_CBVERS); if (clp->lc_req.nr_client != NULL) { SVC_ACQUIRE(nd->nd_xprt); CLNT_ACQUIRE(clp->lc_req.nr_client); nd->nd_xprt->xp_p2 = clp->lc_req.nr_client; /* Disable idle timeout. */ nd->nd_xprt->xp_idletimeout = 0; nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt; } else nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; } NFSBCOPY(sessid, nsep->sess_sessionid, NFSX_V4SESSIONID); NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid, NFSX_V4SESSIONID); if (!sess_replay) { shp = NFSSESSIONHASH(nsep->sess_sessionid); NFSLOCKSTATE(); NFSLOCKSESSION(shp); LIST_INSERT_HEAD(&shp->list, nsep, sess_hash); LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list); nsep->sess_clp = clp; NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); } } } } else if (clp->lc_flags & LCL_NEEDSCONFIRM) { error = NFSERR_EXPIRED; } /* * If called by the Renew Op, we must check the principal. */ if (!error && (opflags & CLOPS_RENEWOP)) { if (nfsrv_notsamecredname(nd, clp)) { doneok = 0; for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) { LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { if ((stp->ls_flags & NFSLCK_OPEN) && stp->ls_uid == nd->nd_cred->cr_uid) { doneok = 1; break; } } } if (!doneok) error = NFSERR_ACCES; } if (!error && (clp->lc_flags & LCL_CBDOWN)) error = NFSERR_CBPATHDOWN; } if ((!error || error == NFSERR_CBPATHDOWN) && (opflags & CLOPS_RENEW)) { clp->lc_expiry = nfsrv_leaseexpiry(); } if (opflags & CLOPS_CONFIRM) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } else if (opflags != CLOPS_RENEW) { NFSUNLOCKSTATE(); } if (clpp) *clpp = clp; out: NFSEXITCODE2(error, nd); return (error); } /* * Perform the NFSv4.1 destroy clientid. */ int nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p) { struct nfsclient *clp; struct nfsclienthashhead *hp; int error = 0, i, igotlock; - if (nfsrvboottime != clientid.lval[0]) { + if (NFSD_VNET(nfsrvboottime) != clientid.lval[0]) { error = NFSERR_STALECLIENTID; goto out; } /* Lock out other nfsd threads */ NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (igotlock == 0); NFSUNLOCKV4ROOTMUTEX(); hp = NFSCLIENTHASH(clientid); LIST_FOREACH(clp, hp, lc_hash) { if (clp->lc_clientid.lval[1] == clientid.lval[1]) break; } if (clp == NULL) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); /* Just return ok, since it is gone. */ goto out; } /* * Free up all layouts on the clientid. Should the client return the * layouts? */ nfsrv_freelayoutlist(clientid); /* Scan for state on the clientid. */ for (i = 0; i < nfsrv_statehashsize; i++) if (!LIST_EMPTY(&clp->lc_stateid[i])) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_CLIENTIDBUSY; goto out; } if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_CLIENTIDBUSY; goto out; } /* Destroy the clientid and return ok. */ nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); nfsrv_zapclient(clp, p); out: NFSEXITCODE2(error, nd); return (error); } /* * Called from the new nfssvc syscall to admin revoke a clientid. * Returns 0 for success, error otherwise. */ int nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p) { struct nfsclient *clp = NULL; int i, error = 0; int gotit, igotlock; /* * First, lock out the nfsd so that state won't change while the * revocation record is being written to the stable storage restart * file. */ NFSLOCKV4ROOTMUTEX(); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!igotlock); NFSUNLOCKV4ROOTMUTEX(); /* * Search for a match in the client list. */ gotit = i = 0; while (i < nfsrv_clienthashsize && !gotit) { - LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { + LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) { if (revokep->nclid_idlen == clp->lc_idlen && !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) { gotit = 1; break; } } i++; } if (!gotit) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 0); NFSUNLOCKV4ROOTMUTEX(); error = EPERM; goto out; } /* * Now, write out the revocation record */ nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_backupstable(); /* * and clear out the state, marking the clientid revoked. */ clp->lc_flags &= ~LCL_CALLBACKSON; clp->lc_flags |= LCL_ADMINREVOKED; nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 0); NFSUNLOCKV4ROOTMUTEX(); out: NFSEXITCODE(error); return (error); } /* * Dump out stats for all clients. Called from nfssvc(2), that is used * nfsstatsv1. */ void nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt) { struct nfsclient *clp; int i = 0, cnt = 0; /* * First, get a reference on the nfsv4rootfs_lock so that an * exclusive lock cannot be acquired while dumping the clients. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); NFSLOCKSTATE(); /* * Rattle through the client lists until done. */ while (i < nfsrv_clienthashsize && cnt < maxcnt) { - clp = LIST_FIRST(&nfsclienthash[i]); - while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) { + clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]); + while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i]) && cnt < + maxcnt) { nfsrv_dumpaclient(clp, &dumpp[cnt]); cnt++; clp = LIST_NEXT(clp, lc_hash); } i++; } if (cnt < maxcnt) dumpp[cnt].ndcl_clid.nclid_idlen = 0; NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } /* * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd. */ static void nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp) { struct nfsstate *stp, *openstp, *lckownstp; struct nfslock *lop; sa_family_t af; #ifdef INET struct sockaddr_in *rin; #endif #ifdef INET6 struct sockaddr_in6 *rin6; #endif dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0; dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0; dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0; dumpp->ndcl_flags = clp->lc_flags; dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen; NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen); af = clp->lc_req.nr_nam->sa_family; dumpp->ndcl_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)clp->lc_req.nr_nam; dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } /* * Now, scan the state lists and total up the opens and locks. */ LIST_FOREACH(stp, &clp->lc_open, ls_list) { dumpp->ndcl_nopenowners++; LIST_FOREACH(openstp, &stp->ls_open, ls_list) { dumpp->ndcl_nopens++; LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) { dumpp->ndcl_nlockowners++; LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) { dumpp->ndcl_nlocks++; } } } } /* * and the delegation lists. */ LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { dumpp->ndcl_ndelegs++; } LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { dumpp->ndcl_nolddelegs++; } } /* * Dump out lock stats for a file. */ void nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt, NFSPROC_T *p) { struct nfsstate *stp; struct nfslock *lop; int cnt = 0; struct nfslockfile *lfp; sa_family_t af; #ifdef INET struct sockaddr_in *rin; #endif #ifdef INET6 struct sockaddr_in6 *rin6; #endif int ret; fhandle_t nfh; ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p); /* * First, get a reference on the nfsv4rootfs_lock so that an * exclusive lock on it cannot be acquired while dumping the locks. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); NFSLOCKSTATE(); if (!ret) ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0); if (ret) { ldumpp[0].ndlck_clid.nclid_idlen = 0; NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); return; } /* * For each open share on file, dump it out. */ stp = LIST_FIRST(&lfp->lf_open); while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) { ldumpp[cnt].ndlck_flags = stp->ls_flags; ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_openowner->ls_ownerlen; NFSBCOPY(stp->ls_openowner->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id, stp->ls_openowner->ls_ownerlen); ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, stp->ls_clp->lc_idlen); af = stp->ls_clp->lc_req.nr_nam->sa_family; ldumpp[cnt].ndlck_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *) stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } stp = LIST_NEXT(stp, ls_file); cnt++; } /* * and all locks. */ lop = LIST_FIRST(&lfp->lf_lock); while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) { stp = lop->lo_stp; ldumpp[cnt].ndlck_flags = lop->lo_flags; ldumpp[cnt].ndlck_first = lop->lo_first; ldumpp[cnt].ndlck_end = lop->lo_end; ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen; NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id, stp->ls_ownerlen); ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, stp->ls_clp->lc_idlen); af = stp->ls_clp->lc_req.nr_nam->sa_family; ldumpp[cnt].ndlck_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *) stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } lop = LIST_NEXT(lop, lo_lckfile); cnt++; } /* * and the delegations. */ stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) { ldumpp[cnt].ndlck_flags = stp->ls_flags; ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; ldumpp[cnt].ndlck_owner.nclid_idlen = 0; ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, stp->ls_clp->lc_idlen); af = stp->ls_clp->lc_req.nr_nam->sa_family; ldumpp[cnt].ndlck_addrfam = af; switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *) stp->ls_clp->lc_req.nr_nam; ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; break; #endif } stp = LIST_NEXT(stp, ls_file); cnt++; } /* * If list isn't full, mark end of list by setting the client name * to zero length. */ if (cnt < maxcnt) ldumpp[cnt].ndlck_clid.nclid_idlen = 0; NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } /* * Server timer routine. It can scan any linked list, so long * as it holds the spin/mutex lock and there is no exclusive lock on * nfsv4rootfs_lock. * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok * to do this from a callout, since the spin locks work. For * Darwin, I'm not sure what will work correctly yet.) * Should be called once per second. */ void -nfsrv_servertimer(void) +nfsrv_servertimer(void *arg __unused) { struct nfsclient *clp, *nclp; struct nfsstate *stp, *nstp; int got_ref, i; /* * Make sure nfsboottime is set. This is used by V3 as well * as V4. Note that nfsboottime is not nfsrvboottime, which is * only used by the V4 server for leases. */ if (nfsboottime.tv_sec == 0) NFSSETBOOTTIME(nfsboottime); /* * If server hasn't started yet, just return. */ NFSLOCKSTATE(); - if (nfsrv_stablefirst.nsf_eograce == 0) { + if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce == 0) { NFSUNLOCKSTATE(); return; } - if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) { - if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) && - NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce) - nfsrv_stablefirst.nsf_flags |= + if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) { + if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & + NFSNSF_GRACEOVER) && + NFSD_MONOSEC > NFSD_VNET(nfsrv_stablefirst).nsf_eograce) + NFSD_VNET(nfsrv_stablefirst).nsf_flags |= (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); NFSUNLOCKSTATE(); return; } /* * Try and get a reference count on the nfsv4rootfs_lock so that * no nfsd thread can acquire an exclusive lock on it before this * call is done. If it is already exclusively locked, just return. */ NFSLOCKV4ROOTMUTEX(); got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); if (got_ref == 0) { NFSUNLOCKSTATE(); return; } /* * For each client... */ for (i = 0; i < nfsrv_clienthashsize; i++) { - clp = LIST_FIRST(&nfsclienthash[i]); - while (clp != LIST_END(&nfsclienthash[i])) { + clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]); + while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i])) { nclp = LIST_NEXT(clp, lc_hash); if (!(clp->lc_flags & LCL_EXPIREIT)) { if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC && ((LIST_EMPTY(&clp->lc_deleg) && LIST_EMPTY(&clp->lc_open)) || nfsrv_clients > nfsrv_clienthighwater)) || (clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC || (clp->lc_expiry < NFSD_MONOSEC && (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) { /* * Lease has expired several nfsrv_lease times ago: * PLUS * - no state is associated with it * OR * - above high water mark for number of clients * (nfsrv_clienthighwater should be large enough * that this only occurs when clients fail to * use the same nfs_client_id4.id. Maybe somewhat * higher that the maximum number of clients that * will mount this server?) * OR * Lease has expired a very long time ago * OR * Lease has expired PLUS the number of opens + locks * has exceeded 90% of capacity * * --> Mark for expiry. The actual expiry will be done * by an nfsd sometime soon. */ clp->lc_flags |= LCL_EXPIREIT; - nfsrv_stablefirst.nsf_flags |= + NFSD_VNET(nfsrv_stablefirst).nsf_flags |= (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT); } else { /* * If there are no opens, increment no open tick cnt * If time exceeds NFSNOOPEN, mark it to be thrown away * otherwise, if there is an open, reset no open time * Hopefully, this will avoid excessive re-creation * of open owners and subsequent open confirms. */ stp = LIST_FIRST(&clp->lc_open); while (stp != LIST_END(&clp->lc_open)) { nstp = LIST_NEXT(stp, ls_list); if (LIST_EMPTY(&stp->ls_open)) { stp->ls_noopens++; if (stp->ls_noopens > NFSNOOPEN || (nfsrv_openpluslock * 2) > nfsrv_v4statelimit) - nfsrv_stablefirst.nsf_flags |= + NFSD_VNET(nfsrv_stablefirst).nsf_flags |= NFSNSF_NOOPENS; } else { stp->ls_noopens = 0; } stp = nstp; } } } clp = nclp; } } NFSUNLOCKSTATE(); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } /* * The following set of functions free up the various data structures. */ /* * Clear out all open/lock state related to this nfsclient. * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that * there are no other active nfsd threads. */ void nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p) { struct nfsstate *stp, *nstp; struct nfsdsession *sep, *nsep; LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) nfsrv_freeopenowner(stp, 1, p); if ((clp->lc_flags & LCL_ADMINREVOKED) == 0) LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) (void)nfsrv_freesession(sep, NULL); } /* * Free a client that has been cleaned. It should also already have been * removed from the lists. * (Just to be safe w.r.t. newnfs_disconnect(), call this function when * softclock interrupts are enabled.) */ void nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p) { #ifdef notyet if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) == (LCL_GSS | LCL_CALLBACKSON) && (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) && clp->lc_handlelen > 0) { clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE; clp->lc_hand.nfsh_flag |= NFSG_DESTROYED; (void) nfsrv_docallback(clp, NFSV4PROC_CBNULL, NULL, 0, NULL, NULL, NULL, 0, p); } #endif newnfs_disconnect(NULL, &clp->lc_req); free(clp->lc_req.nr_nam, M_SONAME); NFSFREEMUTEX(&clp->lc_req.nr_mtx); free(clp->lc_stateid, M_NFSDCLIENT); free(clp, M_NFSDCLIENT); NFSLOCKSTATE(); - nfsstatsv1.srvclients--; + nfsstatsv1_p->srvclients--; nfsrv_openpluslock--; nfsrv_clients--; NFSUNLOCKSTATE(); } /* * Free a list of delegation state structures. * (This function will also free all nfslockfile structures that no * longer have associated state.) */ void nfsrv_freedeleglist(struct nfsstatehead *sthp) { struct nfsstate *stp, *nstp; LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) { nfsrv_freedeleg(stp); } LIST_INIT(sthp); } /* * Free up a delegation. */ static void nfsrv_freedeleg(struct nfsstate *stp) { struct nfslockfile *lfp; LIST_REMOVE(stp, ls_hash); LIST_REMOVE(stp, ls_list); LIST_REMOVE(stp, ls_file); if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) nfsrv_writedelegcnt--; lfp = stp->ls_lfp; if (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) && lfp->lf_usecount == 0 && nfsv4_testlock(&lfp->lf_locallock_lck) == 0) nfsrv_freenfslockfile(lfp); free(stp, M_NFSDSTATE); - nfsstatsv1.srvdelegates--; + nfsstatsv1_p->srvdelegates--; nfsrv_openpluslock--; nfsrv_delegatecnt--; } /* * This function frees an open owner and all associated opens. */ static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p) { struct nfsstate *nstp, *tstp; LIST_REMOVE(stp, ls_list); /* * Now, free all associated opens. */ nstp = LIST_FIRST(&stp->ls_open); while (nstp != LIST_END(&stp->ls_open)) { tstp = nstp; nstp = LIST_NEXT(nstp, ls_list); (void) nfsrv_freeopen(tstp, NULL, cansleep, p); } if (stp->ls_op) nfsrvd_derefcache(stp->ls_op); free(stp, M_NFSDSTATE); - nfsstatsv1.srvopenowners--; + nfsstatsv1_p->srvopenowners--; nfsrv_openpluslock--; } /* * This function frees an open (nfsstate open structure) with all associated * lock_owners and locks. It also frees the nfslockfile structure iff there * are no other opens on the file. * Returns 1 if it free'd the nfslockfile, 0 otherwise. */ static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) { struct nfsstate *nstp, *tstp; struct nfslockfile *lfp; int ret; LIST_REMOVE(stp, ls_hash); LIST_REMOVE(stp, ls_list); LIST_REMOVE(stp, ls_file); lfp = stp->ls_lfp; /* * Now, free all lockowners associated with this open. */ LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp) nfsrv_freelockowner(tstp, vp, cansleep, p); /* * The nfslockfile is freed here if there are no locks * associated with the open. * If there are locks associated with the open, the * nfslockfile structure can be freed via nfsrv_freelockowner(). * Acquire the state mutex to avoid races with calls to * nfsrv_getlockfile(). */ if (cansleep != 0) NFSLOCKSTATE(); if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) && lfp->lf_usecount == 0 && (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) { nfsrv_freenfslockfile(lfp); ret = 1; } else ret = 0; if (cansleep != 0) NFSUNLOCKSTATE(); free(stp, M_NFSDSTATE); - nfsstatsv1.srvopens--; + nfsstatsv1_p->srvopens--; nfsrv_openpluslock--; return (ret); } /* * Frees a lockowner and all associated locks. */ static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) { LIST_REMOVE(stp, ls_hash); LIST_REMOVE(stp, ls_list); nfsrv_freeallnfslocks(stp, vp, cansleep, p); if (stp->ls_op) nfsrvd_derefcache(stp->ls_op); free(stp, M_NFSDSTATE); - nfsstatsv1.srvlockowners--; + nfsstatsv1_p->srvlockowners--; nfsrv_openpluslock--; } /* * Free all the nfs locks on a lockowner. */ static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) { struct nfslock *lop, *nlop; struct nfsrollback *rlp, *nrlp; struct nfslockfile *lfp = NULL; int gottvp = 0; vnode_t tvp = NULL; uint64_t first, end; if (vp != NULL) ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked"); lop = LIST_FIRST(&stp->ls_lock); while (lop != LIST_END(&stp->ls_lock)) { nlop = LIST_NEXT(lop, lo_lckowner); /* * Since all locks should be for the same file, lfp should * not change. */ if (lfp == NULL) lfp = lop->lo_lfp; else if (lfp != lop->lo_lfp) panic("allnfslocks"); /* * If vp is NULL and cansleep != 0, a vnode must be acquired * from the file handle. This only occurs when called from * nfsrv_cleanclient(). */ if (gottvp == 0) { if (nfsrv_dolocallocks == 0) tvp = NULL; else if (vp == NULL && cansleep != 0) { tvp = nfsvno_getvp(&lfp->lf_fh); if (tvp != NULL) NFSVOPUNLOCK(tvp); } else tvp = vp; gottvp = 1; } if (tvp != NULL) { if (cansleep == 0) panic("allnfs2"); first = lop->lo_first; end = lop->lo_end; nfsrv_freenfslock(lop); nfsrv_localunlock(tvp, lfp, first, end, p); LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) free(rlp, M_NFSDROLLBACK); LIST_INIT(&lfp->lf_rollback); } else nfsrv_freenfslock(lop); lop = nlop; } if (vp == NULL && tvp != NULL) vrele(tvp); } /* * Free an nfslock structure. */ static void nfsrv_freenfslock(struct nfslock *lop) { if (lop->lo_lckfile.le_prev != NULL) { LIST_REMOVE(lop, lo_lckfile); - nfsstatsv1.srvlocks--; + nfsstatsv1_p->srvlocks--; nfsrv_openpluslock--; } LIST_REMOVE(lop, lo_lckowner); free(lop, M_NFSDLOCK); } /* * This function frees an nfslockfile structure. */ static void nfsrv_freenfslockfile(struct nfslockfile *lfp) { LIST_REMOVE(lfp, lf_hash); free(lfp, M_NFSDLOCKFILE); } /* * This function looks up an nfsstate structure via stateid. */ static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags, struct nfsstate **stpp) { struct nfsstate *stp; struct nfsstatehead *hp; int error = 0; *stpp = NULL; hp = NFSSTATEHASH(clp, *stateidp); LIST_FOREACH(stp, hp, ls_hash) { if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, NFSX_STATEIDOTHER)) break; } /* * If no state id in list, return NFSERR_BADSTATEID. */ if (stp == LIST_END(hp)) { error = NFSERR_BADSTATEID; goto out; } *stpp = stp; out: NFSEXITCODE(error); return (error); } /* * This function gets an nfsstate structure via owner string. */ static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, struct nfsstate **stpp) { struct nfsstate *stp; *stpp = NULL; LIST_FOREACH(stp, hp, ls_list) { if (new_stp->ls_ownerlen == stp->ls_ownerlen && !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) { *stpp = stp; return; } } } /* * Lock control function called to update lock status. * Returns 0 upon success, -1 if there is no lock and the flags indicate * that one isn't to be created and an NFSERR_xxx for other errors. * The structures new_stp and new_lop are passed in as pointers that should * be set to NULL if the structure is used and shouldn't be free'd. * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are * never used and can safely be allocated on the stack. For all other * cases, *new_stpp and *new_lopp should be malloc'd before the call, * in case they are used. */ int nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp, struct nfslock **new_lopp, struct nfslockconflict *cfp, nfsquad_t clientid, nfsv4stateid_t *stateidp, __unused struct nfsexstuff *exp, struct nfsrv_descript *nd, NFSPROC_T *p) { struct nfslock *lop; struct nfsstate *new_stp = *new_stpp; struct nfslock *new_lop = *new_lopp; struct nfsstate *tstp, *mystp, *nstp; int specialid = 0; struct nfslockfile *lfp; struct nfslock *other_lop = NULL; struct nfsstate *stp, *lckstp = NULL; struct nfsclient *clp = NULL; u_int32_t bits; int error = 0, haslock = 0, ret, reterr; int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0; fhandle_t nfh; uint64_t first, end; uint32_t lock_flags; if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) { /* * Note the special cases of "all 1s" or "all 0s" stateids and * let reads with all 1s go ahead. */ if (new_stp->ls_stateid.seqid == 0x0 && new_stp->ls_stateid.other[0] == 0x0 && new_stp->ls_stateid.other[1] == 0x0 && new_stp->ls_stateid.other[2] == 0x0) specialid = 1; else if (new_stp->ls_stateid.seqid == 0xffffffff && new_stp->ls_stateid.other[0] == 0xffffffff && new_stp->ls_stateid.other[1] == 0xffffffff && new_stp->ls_stateid.other[2] == 0xffffffff) specialid = 2; } /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, specialid); if (error) goto out; /* * Check for state resource limit exceeded. */ if ((new_stp->ls_flags & NFSLCK_LOCK) && nfsrv_openpluslock > nfsrv_v4statelimit) { error = NFSERR_RESOURCE; goto out; } /* * For the lock case, get another nfslock structure, * just in case we need it. * Malloc now, before we start sifting through the linked lists, * in case we have to wait for memory. */ tryagain: if (new_stp->ls_flags & NFSLCK_LOCK) other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); filestruct_locked = 0; reterr = 0; lfp = NULL; /* * Get the lockfile structure for CFH now, so we can do a sanity * check against the stateid, before incrementing the seqid#, since * we want to return NFSERR_BADSTATEID on failure and the seqid# * shouldn't be incremented for this case. * If nfsrv_getlockfile() returns -1, it means "not found", which * will be handled later. * If we are doing Lock/LockU and local locking is enabled, sleep * lock the nfslockfile structure. */ getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p); NFSLOCKSTATE(); if (getlckret == 0) { if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 && nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) { getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, &lfp, &nfh, 1); if (getlckret == 0) filestruct_locked = 1; } else getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, &lfp, &nfh, 0); } if (getlckret != 0 && getlckret != -1) reterr = getlckret; if (filestruct_locked != 0) { LIST_INIT(&lfp->lf_rollback); if ((new_stp->ls_flags & NFSLCK_LOCK)) { /* * For local locking, do the advisory locking now, so * that any conflict can be detected. A failure later * can be rolled back locally. If an error is returned, * struct nfslockfile has been unlocked and any local * locking rolled back. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } reterr = nfsrv_locallock(vp, lfp, (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)), new_lop->lo_first, new_lop->lo_end, cfp, p); NFSLOCKSTATE(); } } if (specialid == 0) { if (new_stp->ls_flags & NFSLCK_TEST) { /* * RFC 3530 does not list LockT as an op that renews a * lease, but the consensus seems to be that it is ok * for a server to do so. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); /* * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid * error returns for LockT, just go ahead and test for a lock, * since there are no locks for this client, but other locks * can conflict. (ie. same client will always be false) */ if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED) error = 0; lckstp = new_stp; } else { error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error == 0) /* * Look up the stateid */ error = nfsrv_getstate(clp, &new_stp->ls_stateid, new_stp->ls_flags, &stp); /* * do some sanity checks for an unconfirmed open or a * stateid that refers to the wrong file, for an open stateid */ if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) && ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) || (getlckret == 0 && stp->ls_lfp != lfp))){ /* * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID * The only exception is using SETATTR with SIZE. * */ if ((new_stp->ls_flags & (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR) error = NFSERR_BADSTATEID; } if (error == 0 && (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) && getlckret == 0 && stp->ls_lfp != lfp) error = NFSERR_BADSTATEID; /* * If the lockowner stateid doesn't refer to the same file, * I believe that is considered ok, since some clients will * only create a single lockowner and use that for all locks * on all files. * For now, log it as a diagnostic, instead of considering it * a BadStateid. */ if (error == 0 && (stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 && getlckret == 0 && stp->ls_lfp != lfp) { #ifdef DIAGNOSTIC printf("Got a lock statid for different file open\n"); #endif /* error = NFSERR_BADSTATEID; */ } if (error == 0) { if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) { /* * If haslock set, we've already checked the seqid. */ if (!haslock) { if (stp->ls_flags & NFSLCK_OPEN) error = nfsrv_checkseqid(nd, new_stp->ls_seq, stp->ls_openowner, new_stp->ls_op); else error = NFSERR_BADSTATEID; } if (!error) nfsrv_getowner(&stp->ls_open, new_stp, &lckstp); if (lckstp) { /* * For NFSv4.1 and NFSv4.2 allow an * open_to_lock_owner when the lock_owner already * exists. Just clear NFSLCK_OPENTOLOCK so that * a new lock_owner will not be created. * RFC7530 states that the error for NFSv4.0 * is NFS4ERR_BAD_SEQID. */ if ((nd->nd_flag & ND_NFSV41) != 0) new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK; else error = NFSERR_BADSEQID; } else lckstp = new_stp; } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) { /* * If haslock set, ditto above. */ if (!haslock) { if (stp->ls_flags & NFSLCK_OPEN) error = NFSERR_BADSTATEID; else error = nfsrv_checkseqid(nd, new_stp->ls_seq, stp, new_stp->ls_op); } lckstp = stp; } else { lckstp = stp; } } /* * If the seqid part of the stateid isn't the same, return * NFSERR_OLDSTATEID for cases other than I/O Ops. * For I/O Ops, only return NFSERR_OLDSTATEID if * nfsrv_returnoldstateid is set. (The consensus on the email * list was that most clients would prefer to not receive * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that * is what will happen, so I use the nfsrv_returnoldstateid to * allow for either server configuration.) */ if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid && (((nd->nd_flag & ND_NFSV41) == 0 && (!(new_stp->ls_flags & NFSLCK_CHECK) || nfsrv_returnoldstateid)) || ((nd->nd_flag & ND_NFSV41) != 0 && new_stp->ls_stateid.seqid != 0))) error = NFSERR_OLDSTATEID; } } /* * Now we can check for grace. */ if (!error) error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags); if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error && nfsrv_checkstable(clp)) error = NFSERR_NOGRACE; /* * If we successfully Reclaimed state, note that. */ if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error) nfsrv_markstable(clp); /* * At this point, either error == NFSERR_BADSTATEID or the * seqid# has been updated, so we can return any error. * If error == 0, there may be an error in: * nd_repstat - Set by the calling function. * reterr - Set above, if getting the nfslockfile structure * or acquiring the local lock failed. * (If both of these are set, nd_repstat should probably be * returned, since that error was detected before this * function call.) */ if (error != 0 || nd->nd_repstat != 0 || reterr != 0) { if (error == 0) { if (nd->nd_repstat != 0) error = nd->nd_repstat; else error = reterr; } if (filestruct_locked != 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); goto out; } /* * Check the nfsrv_getlockfile return. * Returned -1 if no structure found. */ if (getlckret == -1) { error = NFSERR_EXPIRED; /* * Called from lockt, so no lock is OK. */ if (new_stp->ls_flags & NFSLCK_TEST) { error = 0; } else if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) { /* * Called to check for a lock, OK if the stateid is all * 1s or all 0s, but there should be an nfsstate * otherwise. * (ie. If there is no open, I'll assume no share * deny bits.) */ if (specialid) error = 0; else error = NFSERR_BADSTATEID; } NFSUNLOCKSTATE(); goto out; } /* * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict. * For NFSLCK_CHECK, allow a read if write access is granted, * but check for a deny. For NFSLCK_LOCK, require correct access, * which implies a conflicting deny can't exist. */ if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) { /* * Four kinds of state id: * - specialid (all 0s or all 1s), only for NFSLCK_CHECK * - stateid for an open * - stateid for a delegation * - stateid for a lock owner */ if (!specialid) { if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { delegation = 1; mystp = stp; nfsrv_delaydelegtimeout(stp); } else if (stp->ls_flags & NFSLCK_OPEN) { mystp = stp; } else { mystp = stp->ls_openstp; } /* * If locking or checking, require correct access * bit set. */ if (((new_stp->ls_flags & NFSLCK_LOCK) && !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) & mystp->ls_flags & NFSLCK_ACCESSBITS)) || ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) == (NFSLCK_CHECK | NFSLCK_READACCESS) && !(mystp->ls_flags & NFSLCK_READACCESS) && nfsrv_allowreadforwriteopen == 0) || ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) == (NFSLCK_CHECK | NFSLCK_WRITEACCESS) && !(mystp->ls_flags & NFSLCK_WRITEACCESS))) { if (filestruct_locked != 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl3"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); error = NFSERR_OPENMODE; goto out; } } else mystp = NULL; if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) { /* * Check for a conflicting deny bit. */ LIST_FOREACH(tstp, &lfp->lf_open, ls_file) { if (tstp != mystp) { bits = tstp->ls_flags; bits >>= NFSLCK_SHIFT; if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) { KASSERT(vnode_unlocked == 0, ("nfsrv_lockctrl: vnode unlocked1")); ret = nfsrv_clientconflict(tstp->ls_clp, &haslock, vp, p); if (ret == 1) { /* * nfsrv_clientconflict unlocks state * when it returns non-zero. */ lckstp = NULL; goto tryagain; } if (ret == 0) NFSUNLOCKSTATE(); if (ret == 2) error = NFSERR_PERM; else error = NFSERR_OPENMODE; goto out; } } } /* We're outta here */ NFSUNLOCKSTATE(); goto out; } } /* * For setattr, just get rid of all the Delegations for other clients. */ if (new_stp->ls_flags & NFSLCK_SETATTR) { KASSERT(vnode_unlocked == 0, ("nfsrv_lockctrl: vnode unlocked2")); ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p); if (ret) { /* * nfsrv_cleandeleg() unlocks state when it * returns non-zero. */ if (ret == -1) { lckstp = NULL; goto tryagain; } error = ret; goto out; } if (!(new_stp->ls_flags & NFSLCK_CHECK) || (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg))) { NFSUNLOCKSTATE(); goto out; } } /* * Check for a conflicting delegation. If one is found, call * nfsrv_delegconflict() to handle it. If the v4root lock hasn't * been set yet, it will get the lock. Otherwise, it will recall * the delegation. Then, we try try again... * I currently believe the conflict algorithm to be: * For Lock Ops (Lock/LockT/LockU) * - there is a conflict iff a different client has a write delegation * For Reading (Read Op) * - there is a conflict iff a different client has a write delegation * (the specialids are always a different client) * For Writing (Write/Setattr of size) * - there is a conflict if a different client has any delegation * - there is a conflict if the same client has a read delegation * (I don't understand why this isn't allowed, but that seems to be * the current consensus?) */ tstp = LIST_FIRST(&lfp->lf_deleg); while (tstp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(tstp, ls_file); if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))|| ((new_stp->ls_flags & NFSLCK_CHECK) && (new_lop->lo_flags & NFSLCK_READ))) && clp != tstp->ls_clp && (tstp->ls_flags & NFSLCK_DELEGWRITE)) || ((new_stp->ls_flags & NFSLCK_CHECK) && (new_lop->lo_flags & NFSLCK_WRITE) && (clp != tstp->ls_clp || (tstp->ls_flags & NFSLCK_DELEGREAD)))) { ret = 0; if (filestruct_locked != 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4"); NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); NFSUNLOCKSTATE(); NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); vnode_unlocked = 0; if (VN_IS_DOOMED(vp)) ret = NFSERR_SERVERFAULT; NFSLOCKSTATE(); } if (ret == 0) ret = nfsrv_delegconflict(tstp, &haslock, p, vp); if (ret) { /* * nfsrv_delegconflict unlocks state when it * returns non-zero, which it always does. */ if (other_lop) { free(other_lop, M_NFSDLOCK); other_lop = NULL; } if (ret == -1) { lckstp = NULL; goto tryagain; } error = ret; goto out; } /* Never gets here. */ } tstp = nstp; } /* * Handle the unlock case by calling nfsrv_updatelock(). * (Should I have done some access checking above for unlock? For now, * just let it happen.) */ if (new_stp->ls_flags & NFSLCK_UNLOCK) { first = new_lop->lo_first; end = new_lop->lo_end; nfsrv_updatelock(stp, new_lopp, &other_lop, lfp); stateidp->seqid = ++(stp->ls_stateid.seqid); if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) stateidp->seqid = stp->ls_stateid.seqid = 1; stateidp->other[0] = stp->ls_stateid.other[0]; stateidp->other[1] = stp->ls_stateid.other[1]; stateidp->other[2] = stp->ls_stateid.other[2]; if (filestruct_locked != 0) { NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } /* Update the local locks. */ nfsrv_localunlock(vp, lfp, first, end, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); goto out; } /* * Search for a conflicting lock. A lock conflicts if: * - the lock range overlaps and * - at least one lock is a write lock and * - it is not owned by the same lock owner */ if (!delegation) { LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { if (new_lop->lo_end > lop->lo_first && new_lop->lo_first < lop->lo_end && (new_lop->lo_flags == NFSLCK_WRITE || lop->lo_flags == NFSLCK_WRITE) && lckstp != lop->lo_stp && (clp != lop->lo_stp->ls_clp || lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen || NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner, lckstp->ls_ownerlen))) { if (other_lop) { free(other_lop, M_NFSDLOCK); other_lop = NULL; } if (vnode_unlocked != 0) ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, NULL, p); else ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, vp, p); if (ret == 1) { if (filestruct_locked != 0) { if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6"); NFSVOPUNLOCK(vp); } /* Roll back local locks. */ nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); NFSUNLOCKSTATE(); NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); vnode_unlocked = 0; if (VN_IS_DOOMED(vp)) { error = NFSERR_SERVERFAULT; goto out; } } /* * nfsrv_clientconflict() unlocks state when it * returns non-zero. */ lckstp = NULL; goto tryagain; } /* * Found a conflicting lock, so record the conflict and * return the error. */ if (cfp != NULL && ret == 0) { cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0]; cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1]; cfp->cl_first = lop->lo_first; cfp->cl_end = lop->lo_end; cfp->cl_flags = lop->lo_flags; cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen; NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner, cfp->cl_ownerlen); } if (ret == 2) error = NFSERR_PERM; else if (new_stp->ls_flags & NFSLCK_RECLAIM) error = NFSERR_RECLAIMCONFLICT; else if (new_stp->ls_flags & NFSLCK_CHECK) error = NFSERR_LOCKED; else error = NFSERR_DENIED; if (filestruct_locked != 0 && ret == 0) { /* Roll back local locks. */ NFSUNLOCKSTATE(); if (vnode_unlocked == 0) { ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7"); vnode_unlocked = 1; NFSVOPUNLOCK(vp); } nfsrv_locallock_rollback(vp, lfp, p); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } if (ret == 0) NFSUNLOCKSTATE(); goto out; } } } /* * We only get here if there was no lock that conflicted. */ if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) { NFSUNLOCKSTATE(); goto out; } /* * We only get here when we are creating or modifying a lock. * There are two variants: * - exist_lock_owner where lock_owner exists * - open_to_lock_owner with new lock_owner */ first = new_lop->lo_first; end = new_lop->lo_end; lock_flags = new_lop->lo_flags; if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) { nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp); stateidp->seqid = ++(lckstp->ls_stateid.seqid); if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) stateidp->seqid = lckstp->ls_stateid.seqid = 1; stateidp->other[0] = lckstp->ls_stateid.other[0]; stateidp->other[1] = lckstp->ls_stateid.other[1]; stateidp->other[2] = lckstp->ls_stateid.other[2]; } else { /* * The new open_to_lock_owner case. * Link the new nfsstate into the lists. */ new_stp->ls_seq = new_stp->ls_opentolockseq; nfsrvd_refcache(new_stp->ls_op); stateidp->seqid = new_stp->ls_stateid.seqid = 1; stateidp->other[0] = new_stp->ls_stateid.other[0] = clp->lc_clientid.lval[0]; stateidp->other[1] = new_stp->ls_stateid.other[1] = clp->lc_clientid.lval[1]; stateidp->other[2] = new_stp->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_stp->ls_clp = clp; LIST_INIT(&new_stp->ls_lock); new_stp->ls_openstp = stp; new_stp->ls_lfp = lfp; nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp, lfp); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid), new_stp, ls_hash); LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list); *new_lopp = NULL; *new_stpp = NULL; - nfsstatsv1.srvlockowners++; + nfsstatsv1_p->srvlockowners++; nfsrv_openpluslock++; } if (filestruct_locked != 0) { NFSUNLOCKSTATE(); nfsrv_locallock_commit(lfp, lock_flags, first, end); NFSLOCKSTATE(); nfsrv_unlocklf(lfp); } NFSUNLOCKSTATE(); out: if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } if (vnode_unlocked != 0) { NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (error == 0 && VN_IS_DOOMED(vp)) error = NFSERR_SERVERFAULT; } if (other_lop) free(other_lop, M_NFSDLOCK); NFSEXITCODE2(error, nd); return (error); } /* * Check for state errors for Open. * repstat is passed back out as an error if more critical errors * are not detected. */ int nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd, NFSPROC_T *p, int repstat) { struct nfsstate *stp, *nstp; struct nfsclient *clp; struct nfsstate *ownerstp; struct nfslockfile *lfp, *new_lfp; int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0; if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) readonly = 1; /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) goto out; /* * Check for state resource limit exceeded. * Technically this should be SMP protected, but the worst * case error is "out by one or two" on the count when it * returns NFSERR_RESOURCE and the limit is just a rather * arbitrary high water mark, so no harm is done. */ if (nfsrv_openpluslock > nfsrv_v4statelimit) { error = NFSERR_RESOURCE; goto out; } tryagain: new_lfp = malloc(sizeof (struct nfslockfile), M_NFSDLOCKFILE, M_WAITOK); if (vp) getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, NULL, p); NFSLOCKSTATE(); /* * Get the nfsclient structure. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); /* * Look up the open owner. See if it needs confirmation and * check the seq#, as required. */ if (!error) nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); if (!error && ownerstp) { error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp, new_stp->ls_op); /* * If the OpenOwner hasn't been confirmed, assume the * old one was a replay and this one is ok. * See: RFC3530 Sec. 14.2.18. */ if (error == NFSERR_BADSEQID && (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM)) error = 0; } /* * Check for grace. */ if (!error) error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags); if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error && nfsrv_checkstable(clp)) error = NFSERR_NOGRACE; /* * If none of the above errors occurred, let repstat be * returned. */ if (repstat && !error) error = repstat; if (error) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } free(new_lfp, M_NFSDLOCKFILE); goto out; } /* * If vp == NULL, the file doesn't exist yet, so return ok. * (This always happens on the first pass, so haslock must be 0.) */ if (vp == NULL) { NFSUNLOCKSTATE(); free(new_lfp, M_NFSDLOCKFILE); goto out; } /* * Get the structure for the underlying file. */ if (getfhret) error = getfhret; else error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, NULL, 0); if (new_lfp) free(new_lfp, M_NFSDLOCKFILE); if (error) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } /* * Search for a conflicting open/share. */ if (new_stp->ls_flags & NFSLCK_DELEGCUR) { /* * For Delegate_Cur, search for the matching Delegation, * which indicates no conflict. * An old delegation should have been recovered by the * client doing a Claim_DELEGATE_Prev, so I won't let * it match and return NFSERR_EXPIRED. Should I let it * match? */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (!(stp->ls_flags & NFSLCK_OLDDELEG) && (((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) || stateidp->seqid == stp->ls_stateid.seqid) && !NFSBCMP(stateidp->other, stp->ls_stateid.other, NFSX_STATEIDOTHER)) break; } if (stp == LIST_END(&lfp->lf_deleg) || ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && (stp->ls_flags & NFSLCK_DELEGREAD))) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_EXPIRED; goto out; } } /* * Check for access/deny bit conflicts. I check for the same * owner as well, in case the client didn't bother. */ LIST_FOREACH(stp, &lfp->lf_open, ls_file) { if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) && (((new_stp->ls_flags & NFSLCK_ACCESSBITS) & ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))|| ((stp->ls_flags & NFSLCK_ACCESSBITS) & ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){ ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p); if (ret == 1) { /* * nfsrv_clientconflict() unlocks * state when it returns non-zero. */ goto tryagain; } if (ret == 2) error = NFSERR_PERM; else if (new_stp->ls_flags & NFSLCK_RECLAIM) error = NFSERR_RECLAIMCONFLICT; else error = NFSERR_SHAREDENIED; if (ret == 0) NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } } /* * Check for a conflicting delegation. If one is found, call * nfsrv_delegconflict() to handle it. If the v4root lock hasn't * been set yet, it will get the lock. Otherwise, it will recall * the delegation. Then, we try try again... * (If NFSLCK_DELEGCUR is set, it has a delegation, so there * isn't a conflict.) * I currently believe the conflict algorithm to be: * For Open with Read Access and Deny None * - there is a conflict iff a different client has a write delegation * For Open with other Write Access or any Deny except None * - there is a conflict if a different client has any delegation * - there is a conflict if the same client has a read delegation * (The current consensus is that this last case should be * considered a conflict since the client with a read delegation * could have done an Open with ReadAccess and WriteDeny * locally and then not have checked for the WriteDeny.) * Don't check for a Reclaim, since that will be dealt with * by nfsrv_openctrl(). */ if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) { stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(stp, ls_file); if ((readonly && stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGWRITE)) || (!readonly && (stp->ls_clp != clp || (stp->ls_flags & NFSLCK_DELEGREAD)))) { ret = nfsrv_delegconflict(stp, &haslock, p, vp); if (ret) { /* * nfsrv_delegconflict() unlocks state * when it returns non-zero. */ if (ret == -1) goto tryagain; error = ret; goto out; } } stp = nstp; } } NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } out: NFSEXITCODE2(error, nd); return (error); } /* * Open control function to create/update open state for an open. */ int nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp, struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp, nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp, NFSPROC_T *p, u_quad_t filerev) { struct nfsstate *new_stp = *new_stpp; struct nfsstate *stp, *nstp; struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg; struct nfslockfile *lfp, *new_lfp; struct nfsclient *clp; int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1; int readonly = 0, cbret = 1, getfhret = 0; int gotstate = 0, len = 0; u_char *clidp = NULL; if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) readonly = 1; /* * Check for restart conditions (client and server). * (Paranoia, should have been detected by nfsrv_opencheck().) * If an error does show up, return NFSERR_EXPIRED, since the * the seqid# has already been incremented. */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) { printf("Nfsd: openctrl unexpected restart err=%d\n", error); error = NFSERR_EXPIRED; goto out; } clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); tryagain: new_lfp = malloc(sizeof (struct nfslockfile), M_NFSDLOCKFILE, M_WAITOK); new_open = malloc(sizeof (struct nfsstate), M_NFSDSTATE, M_WAITOK); new_deleg = malloc(sizeof (struct nfsstate), M_NFSDSTATE, M_WAITOK); getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, NULL, p); NFSLOCKSTATE(); /* * Get the client structure. Since the linked lists could be changed * by other nfsd processes if this process does a tsleep(), one of * two things must be done. * 1 - don't tsleep() * or * 2 - get the nfsv4_lock() { indicated by haslock == 1 } * before using the lists, since this lock stops the other * nfsd. This should only be used for rare cases, since it * essentially single threads the nfsd. * At this time, it is only done for cases where the stable * storage file must be written prior to completion of state * expiration. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) && clp->lc_program) { /* * This happens on the first open for a client * that supports callbacks. */ NFSUNLOCKSTATE(); /* * Although nfsrv_docallback() will sleep, clp won't * go away, since they are only removed when the * nfsv4_lock() has blocked the nfsd threads. The * fields in clp can change, but having multiple * threads do this Null callback RPC should be * harmless. */ cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL, NULL, 0, NULL, NULL, NULL, 0, p); NFSLOCKSTATE(); clp->lc_flags &= ~LCL_NEEDSCBNULL; if (!cbret) clp->lc_flags |= LCL_CALLBACKSON; } /* * Look up the open owner. See if it needs confirmation and * check the seq#, as required. */ if (!error) nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); if (error) { NFSUNLOCKSTATE(); printf("Nfsd: openctrl unexpected state err=%d\n", error); free(new_lfp, M_NFSDLOCKFILE); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_EXPIRED; goto out; } if (new_stp->ls_flags & NFSLCK_RECLAIM) nfsrv_markstable(clp); /* * Get the structure for the underlying file. */ if (getfhret) error = getfhret; else error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, NULL, 0); if (new_lfp) free(new_lfp, M_NFSDLOCKFILE); if (error) { NFSUNLOCKSTATE(); printf("Nfsd openctrl unexpected getlockfile err=%d\n", error); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } /* * Search for a conflicting open/share. */ if (new_stp->ls_flags & NFSLCK_DELEGCUR) { /* * For Delegate_Cur, search for the matching Delegation, * which indicates no conflict. * An old delegation should have been recovered by the * client doing a Claim_DELEGATE_Prev, so I won't let * it match and return NFSERR_EXPIRED. Should I let it * match? */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (!(stp->ls_flags & NFSLCK_OLDDELEG) && (((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) || stateidp->seqid == stp->ls_stateid.seqid) && !NFSBCMP(stateidp->other, stp->ls_stateid.other, NFSX_STATEIDOTHER)) break; } if (stp == LIST_END(&lfp->lf_deleg) || ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && (stp->ls_flags & NFSLCK_DELEGREAD))) { NFSUNLOCKSTATE(); printf("Nfsd openctrl unexpected expiry\n"); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_EXPIRED; goto out; } /* * Don't issue a Delegation, since one already exists and * delay delegation timeout, as required. */ delegate = 0; nfsrv_delaydelegtimeout(stp); } /* * Check for access/deny bit conflicts. I also check for the * same owner, since the client might not have bothered to check. * Also, note an open for the same file and owner, if found, * which is all we do here for Delegate_Cur, since conflict * checking is already done. */ LIST_FOREACH(stp, &lfp->lf_open, ls_file) { if (ownerstp && stp->ls_openowner == ownerstp) openstp = stp; if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) { /* * If another client has the file open, the only * delegation that can be issued is a Read delegation * and only if it is a Read open with Deny none. */ if (clp != stp->ls_clp) { if ((stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) writedeleg = 0; else delegate = 0; } if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) & ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))|| ((stp->ls_flags & NFSLCK_ACCESSBITS) & ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){ ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p); if (ret == 1) { /* * nfsrv_clientconflict() unlocks state * when it returns non-zero. */ free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); openstp = NULL; goto tryagain; } if (ret == 2) error = NFSERR_PERM; else if (new_stp->ls_flags & NFSLCK_RECLAIM) error = NFSERR_RECLAIMCONFLICT; else error = NFSERR_SHAREDENIED; if (ret == 0) NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); printf("nfsd openctrl unexpected client cnfl\n"); goto out; } } } /* * Check for a conflicting delegation. If one is found, call * nfsrv_delegconflict() to handle it. If the v4root lock hasn't * been set yet, it will get the lock. Otherwise, it will recall * the delegation. Then, we try try again... * (If NFSLCK_DELEGCUR is set, it has a delegation, so there * isn't a conflict.) * I currently believe the conflict algorithm to be: * For Open with Read Access and Deny None * - there is a conflict iff a different client has a write delegation * For Open with other Write Access or any Deny except None * - there is a conflict if a different client has any delegation * - there is a conflict if the same client has a read delegation * (The current consensus is that this last case should be * considered a conflict since the client with a read delegation * could have done an Open with ReadAccess and WriteDeny * locally and then not have checked for the WriteDeny.) */ if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) { stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(stp, ls_file); if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD)) writedeleg = 0; else delegate = 0; if ((readonly && stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGWRITE)) || (!readonly && (stp->ls_clp != clp || (stp->ls_flags & NFSLCK_DELEGREAD)))) { if (new_stp->ls_flags & NFSLCK_RECLAIM) { delegate = 2; } else { ret = nfsrv_delegconflict(stp, &haslock, p, vp); if (ret) { /* * nfsrv_delegconflict() unlocks state * when it returns non-zero. */ printf("Nfsd openctrl unexpected deleg cnfl\n"); free(new_open, M_NFSDSTATE); free(new_deleg, M_NFSDSTATE); if (ret == -1) { openstp = NULL; goto tryagain; } error = ret; goto out; } } } stp = nstp; } } /* * We only get here if there was no open that conflicted. * If an open for the owner exists, or in the access/deny bits. * Otherwise it is a new open. If the open_owner hasn't been * confirmed, replace the open with the new one needing confirmation, * otherwise add the open. */ if (new_stp->ls_flags & NFSLCK_DELEGPREV) { /* * Handle NFSLCK_DELEGPREV by searching the old delegations for * a match. If found, just move the old delegation to the current * delegation list and issue open. If not found, return * NFSERR_EXPIRED. */ LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { if (stp->ls_lfp == lfp) { /* Found it */ if (stp->ls_clp != clp) panic("olddeleg clp"); LIST_REMOVE(stp, ls_list); LIST_REMOVE(stp, ls_hash); stp->ls_flags &= ~NFSLCK_OLDDELEG; stp->ls_stateid.seqid = delegstateidp->seqid = 1; stp->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; stp->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; stp->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); stp->ls_compref = nd->nd_compref; LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list); LIST_INSERT_HEAD(NFSSTATEHASH(clp, stp->ls_stateid), stp, ls_hash); if (stp->ls_flags & NFSLCK_DELEGWRITE) *rflagsp |= NFSV4OPEN_WRITEDELEGATE; else *rflagsp |= NFSV4OPEN_READDELEGATE; clp->lc_delegtime = NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA; /* * Now, do the associated open. */ new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)| NFSLCK_OPEN; if (stp->ls_flags & NFSLCK_DELEGWRITE) new_open->ls_flags |= (NFSLCK_READACCESS | NFSLCK_WRITEACCESS); else new_open->ls_flags |= NFSLCK_READACCESS; new_open->ls_uid = new_stp->ls_uid; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INIT(&new_open->ls_open); LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); /* * and handle the open owner */ if (ownerstp) { new_open->ls_openowner = ownerstp; LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list); } else { new_open->ls_openowner = new_stp; new_stp->ls_flags = 0; nfsrvd_refcache(new_stp->ls_op); new_stp->ls_noopens = 0; LIST_INIT(&new_stp->ls_open); LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); *new_stpp = NULL; - nfsstatsv1.srvopenowners++; + nfsstatsv1_p->srvopenowners++; nfsrv_openpluslock++; } openstp = new_open; new_open = NULL; - nfsstatsv1.srvopens++; + nfsstatsv1_p->srvopens++; nfsrv_openpluslock++; break; } } if (stp == LIST_END(&clp->lc_olddeleg)) error = NFSERR_EXPIRED; } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { /* * Scan to see that no delegation for this client and file * doesn't already exist. * There also shouldn't yet be an Open for this file and * openowner. */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (stp->ls_clp == clp) break; } if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) { /* * This is the Claim_Previous case with a delegation * type != Delegate_None. */ /* * First, add the delegation. (Although we must issue the * delegation, we can also ask for an immediate return.) */ new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); if (new_stp->ls_flags & NFSLCK_DELEGWRITE) { new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; nfsrv_writedelegcnt++; } else { new_deleg->ls_flags = (NFSLCK_DELEGREAD | NFSLCK_READACCESS); *rflagsp |= NFSV4OPEN_READDELEGATE; } new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; if (delegate == 2 || nfsrv_issuedelegs == 0 || (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != LCL_CALLBACKSON || NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) || !NFSVNO_DELEGOK(vp)) *rflagsp |= NFSV4OPEN_RECALL; - nfsstatsv1.srvdelegates++; + nfsstatsv1_p->srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; /* * Now, do the associated open. */ new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) | NFSLCK_OPEN; if (new_stp->ls_flags & NFSLCK_DELEGWRITE) new_open->ls_flags |= (NFSLCK_READACCESS | NFSLCK_WRITEACCESS); else new_open->ls_flags |= NFSLCK_READACCESS; new_open->ls_uid = new_stp->ls_uid; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INIT(&new_open->ls_open); LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); /* * and handle the open owner */ if (ownerstp) { new_open->ls_openowner = ownerstp; LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); } else { new_open->ls_openowner = new_stp; new_stp->ls_flags = 0; nfsrvd_refcache(new_stp->ls_op); new_stp->ls_noopens = 0; LIST_INIT(&new_stp->ls_open); LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); *new_stpp = NULL; - nfsstatsv1.srvopenowners++; + nfsstatsv1_p->srvopenowners++; nfsrv_openpluslock++; } openstp = new_open; new_open = NULL; - nfsstatsv1.srvopens++; + nfsstatsv1_p->srvopens++; nfsrv_openpluslock++; } else { error = NFSERR_RECLAIMCONFLICT; } } else if (ownerstp) { if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) { /* Replace the open */ if (ownerstp->ls_op) nfsrvd_derefcache(ownerstp->ls_op); ownerstp->ls_op = new_stp->ls_op; nfsrvd_refcache(ownerstp->ls_op); ownerstp->ls_seq = new_stp->ls_seq; *rflagsp |= NFSV4OPEN_RESULTCONFIRM; stp = LIST_FIRST(&ownerstp->ls_open); stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) | NFSLCK_OPEN; stp->ls_stateid.seqid = 1; stp->ls_uid = new_stp->ls_uid; if (lfp != stp->ls_lfp) { LIST_REMOVE(stp, ls_file); LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file); stp->ls_lfp = lfp; } openstp = stp; } else if (openstp) { openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS); openstp->ls_stateid.seqid++; if ((nd->nd_flag & ND_NFSV41) != 0 && openstp->ls_stateid.seqid == 0) openstp->ls_stateid.seqid = 1; /* * This is where we can choose to issue a delegation. */ if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0) *rflagsp |= NFSV4OPEN_WDNOTWANTED; else if (nfsrv_issuedelegs == 0) *rflagsp |= NFSV4OPEN_WDSUPPFTYPE; else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt)) *rflagsp |= NFSV4OPEN_WDRESOURCE; else if (delegate == 0 || writedeleg == 0 || NFSVNO_EXRDONLY(exp) || (readonly != 0 && nfsrv_writedelegifpos == 0) || !NFSVNO_DELEGOK(vp) || (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 || (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != LCL_CALLBACKSON) *rflagsp |= NFSV4OPEN_WDCONTENTION; else { new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; nfsrv_writedelegcnt++; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; - nfsstatsv1.srvdelegates++; + nfsstatsv1_p->srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; } } else { new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)| NFSLCK_OPEN; new_open->ls_uid = new_stp->ls_uid; new_open->ls_openowner = ownerstp; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INIT(&new_open->ls_open); LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); openstp = new_open; new_open = NULL; - nfsstatsv1.srvopens++; + nfsstatsv1_p->srvopens++; nfsrv_openpluslock++; /* * This is where we can choose to issue a delegation. */ if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0) *rflagsp |= NFSV4OPEN_WDNOTWANTED; else if (nfsrv_issuedelegs == 0) *rflagsp |= NFSV4OPEN_WDSUPPFTYPE; else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt)) *rflagsp |= NFSV4OPEN_WDRESOURCE; else if (delegate == 0 || (writedeleg == 0 && readonly == 0) || !NFSVNO_DELEGOK(vp) || (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != LCL_CALLBACKSON) *rflagsp |= NFSV4OPEN_WDCONTENTION; else { new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); if (writedeleg && !NFSVNO_EXRDONLY(exp) && (nfsrv_writedelegifpos || !readonly) && (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) { new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; nfsrv_writedelegcnt++; } else { new_deleg->ls_flags = (NFSLCK_DELEGREAD | NFSLCK_READACCESS); *rflagsp |= NFSV4OPEN_READDELEGATE; } new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; - nfsstatsv1.srvdelegates++; + nfsstatsv1_p->srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; } } } else { /* * New owner case. Start the open_owner sequence with a * Needs confirmation (unless a reclaim) and hang the * new open off it. */ new_open->ls_stateid.seqid = 1; new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) | NFSLCK_OPEN; new_open->ls_uid = new_stp->ls_uid; LIST_INIT(&new_open->ls_open); new_open->ls_openowner = new_stp; new_open->ls_lfp = lfp; new_open->ls_clp = clp; LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); if (new_stp->ls_flags & NFSLCK_RECLAIM) { new_stp->ls_flags = 0; } else if ((nd->nd_flag & ND_NFSV41) != 0) { /* NFSv4.1 never needs confirmation. */ new_stp->ls_flags = 0; /* * This is where we can choose to issue a delegation. */ if (delegate && nfsrv_issuedelegs && (writedeleg || readonly) && (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) == LCL_CALLBACKSON && !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) && NFSVNO_DELEGOK(vp) && ((nd->nd_flag & ND_NFSV41) == 0 || (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) { new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = clp->lc_clientid.lval[0]; new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = clp->lc_clientid.lval[1]; new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = nfsrv_nextstateindex(clp); if (writedeleg && !NFSVNO_EXRDONLY(exp) && (nfsrv_writedelegifpos || !readonly) && ((nd->nd_flag & ND_NFSV41) == 0 || (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0)) { new_deleg->ls_flags = (NFSLCK_DELEGWRITE | NFSLCK_READACCESS | NFSLCK_WRITEACCESS); *rflagsp |= NFSV4OPEN_WRITEDELEGATE; nfsrv_writedelegcnt++; } else { new_deleg->ls_flags = (NFSLCK_DELEGREAD | NFSLCK_READACCESS); *rflagsp |= NFSV4OPEN_READDELEGATE; } new_deleg->ls_uid = new_stp->ls_uid; new_deleg->ls_lfp = lfp; new_deleg->ls_clp = clp; new_deleg->ls_filerev = filerev; new_deleg->ls_compref = nd->nd_compref; new_deleg->ls_lastrecall = 0; LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid), new_deleg, ls_hash); LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); new_deleg = NULL; - nfsstatsv1.srvdelegates++; + nfsstatsv1_p->srvdelegates++; nfsrv_openpluslock++; nfsrv_delegatecnt++; } /* * Since NFSv4.1 never does an OpenConfirm, the first * open state will be acquired here. */ if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) { clp->lc_flags |= LCL_STAMPEDSTABLE; len = clp->lc_idlen; NFSBCOPY(clp->lc_id, clidp, len); gotstate = 1; } } else { *rflagsp |= NFSV4OPEN_RESULTCONFIRM; new_stp->ls_flags = NFSLCK_NEEDSCONFIRM; } nfsrvd_refcache(new_stp->ls_op); new_stp->ls_noopens = 0; LIST_INIT(&new_stp->ls_open); LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), new_open, ls_hash); openstp = new_open; new_open = NULL; *new_stpp = NULL; - nfsstatsv1.srvopens++; + nfsstatsv1_p->srvopens++; nfsrv_openpluslock++; - nfsstatsv1.srvopenowners++; + nfsstatsv1_p->srvopenowners++; nfsrv_openpluslock++; } if (!error) { stateidp->seqid = openstp->ls_stateid.seqid; stateidp->other[0] = openstp->ls_stateid.other[0]; stateidp->other[1] = openstp->ls_stateid.other[1]; stateidp->other[2] = openstp->ls_stateid.other[2]; } NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } if (new_open) free(new_open, M_NFSDSTATE); if (new_deleg) free(new_deleg, M_NFSDSTATE); /* * If the NFSv4.1 client just acquired its first open, write a timestamp * to the stable storage file. */ if (gotstate != 0) { nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); nfsrv_backupstable(); } out: free(clidp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Open update. Does the confirm, downgrade and close. */ int nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p, int *retwriteaccessp) { struct nfsstate *stp; struct nfsclient *clp; struct nfslockfile *lfp; u_int32_t bits; int error = 0, gotstate = 0, len = 0; u_char *clidp = NULL; /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) goto out; clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); NFSLOCKSTATE(); /* * Get the open structure via clientid and stateid. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (!error) error = nfsrv_getstate(clp, &new_stp->ls_stateid, new_stp->ls_flags, &stp); /* * Sanity check the open. */ if (!error && (!(stp->ls_flags & NFSLCK_OPEN) || (!(new_stp->ls_flags & NFSLCK_CONFIRM) && (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) || ((new_stp->ls_flags & NFSLCK_CONFIRM) && (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))))) error = NFSERR_BADSTATEID; if (!error) error = nfsrv_checkseqid(nd, new_stp->ls_seq, stp->ls_openowner, new_stp->ls_op); if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid && (((nd->nd_flag & ND_NFSV41) == 0 && !(new_stp->ls_flags & NFSLCK_CONFIRM)) || ((nd->nd_flag & ND_NFSV41) != 0 && new_stp->ls_stateid.seqid != 0))) error = NFSERR_OLDSTATEID; if (!error && vnode_vtype(vp) != VREG) { if (vnode_vtype(vp) == VDIR) error = NFSERR_ISDIR; else error = NFSERR_INVAL; } if (error) { /* * If a client tries to confirm an Open with a bad * seqid# and there are no byte range locks or other Opens * on the openowner, just throw it away, so the next use of the * openowner will start a fresh seq#. */ if (error == NFSERR_BADSEQID && (new_stp->ls_flags & NFSLCK_CONFIRM) && nfsrv_nootherstate(stp)) nfsrv_freeopenowner(stp->ls_openowner, 0, p); NFSUNLOCKSTATE(); goto out; } /* * Set the return stateid. */ stateidp->seqid = stp->ls_stateid.seqid + 1; if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) stateidp->seqid = 1; stateidp->other[0] = stp->ls_stateid.other[0]; stateidp->other[1] = stp->ls_stateid.other[1]; stateidp->other[2] = stp->ls_stateid.other[2]; /* * Now, handle the three cases. */ if (new_stp->ls_flags & NFSLCK_CONFIRM) { /* * If the open doesn't need confirmation, it seems to me that * there is a client error, but I'll just log it and keep going? */ if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) printf("Nfsv4d: stray open confirm\n"); stp->ls_openowner->ls_flags = 0; stp->ls_stateid.seqid++; if ((nd->nd_flag & ND_NFSV41) != 0 && stp->ls_stateid.seqid == 0) stp->ls_stateid.seqid = 1; if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) { clp->lc_flags |= LCL_STAMPEDSTABLE; len = clp->lc_idlen; NFSBCOPY(clp->lc_id, clidp, len); gotstate = 1; } NFSUNLOCKSTATE(); } else if (new_stp->ls_flags & NFSLCK_CLOSE) { lfp = stp->ls_lfp; if (retwriteaccessp != NULL) { if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0) *retwriteaccessp = 1; else *retwriteaccessp = 0; } if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) { /* Get the lf lock */ nfsrv_locklf(lfp); NFSUNLOCKSTATE(); ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate"); NFSVOPUNLOCK(vp); if (nfsrv_freeopen(stp, vp, 1, p) == 0) { NFSLOCKSTATE(); nfsrv_unlocklf(lfp); NFSUNLOCKSTATE(); } NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); } else { (void) nfsrv_freeopen(stp, NULL, 0, p); NFSUNLOCKSTATE(); } } else { /* * Update the share bits, making sure that the new set are a * subset of the old ones. */ bits = (new_stp->ls_flags & NFSLCK_SHAREBITS); if (~(stp->ls_flags) & bits) { NFSUNLOCKSTATE(); error = NFSERR_INVAL; goto out; } stp->ls_flags = (bits | NFSLCK_OPEN); stp->ls_stateid.seqid++; if ((nd->nd_flag & ND_NFSV41) != 0 && stp->ls_stateid.seqid == 0) stp->ls_stateid.seqid = 1; NFSUNLOCKSTATE(); } /* * If the client just confirmed its first open, write a timestamp * to the stable storage file. */ if (gotstate != 0) { nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); nfsrv_backupstable(); } out: free(clidp, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Delegation update. Does the purge and return. */ int nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid, nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p, int *retwriteaccessp) { struct nfsstate *stp; struct nfsclient *clp; int error = 0; fhandle_t fh; /* * Do a sanity check against the file handle for DelegReturn. */ if (vp) { error = nfsvno_getfh(vp, &fh, p); if (error) goto out; } /* * Check for restart conditions (client and server). */ if (op == NFSV4OP_DELEGRETURN) error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN, stateidp, 0); else error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE, stateidp, 0); NFSLOCKSTATE(); /* * Get the open structure via clientid and stateid. */ if (!error) error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error) { if (error == NFSERR_CBPATHDOWN) error = 0; if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN) error = NFSERR_STALESTATEID; } if (!error && op == NFSV4OP_DELEGRETURN) { error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp); if (!error && stp->ls_stateid.seqid != stateidp->seqid && ((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0)) error = NFSERR_OLDSTATEID; } /* * NFSERR_EXPIRED means that the state has gone away, * so Delegations have been purged. Just return ok. */ if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) { NFSUNLOCKSTATE(); error = 0; goto out; } if (error) { NFSUNLOCKSTATE(); goto out; } if (op == NFSV4OP_DELEGRETURN) { if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh, sizeof (fhandle_t))) { NFSUNLOCKSTATE(); error = NFSERR_BADSTATEID; goto out; } if (retwriteaccessp != NULL) { if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) *retwriteaccessp = 1; else *retwriteaccessp = 0; } nfsrv_freedeleg(stp); } else { nfsrv_freedeleglist(&clp->lc_olddeleg); } NFSUNLOCKSTATE(); error = 0; out: NFSEXITCODE(error); return (error); } /* * Release lock owner. */ int nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid, NFSPROC_T *p) { struct nfsstate *stp, *nstp, *openstp, *ownstp; struct nfsclient *clp; int error = 0; /* * Check for restart conditions (client and server). */ error = nfsrv_checkrestart(clientid, new_stp->ls_flags, &new_stp->ls_stateid, 0); if (error) goto out; NFSLOCKSTATE(); /* * Get the lock owner by name. */ error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, NULL, p); if (error) { NFSUNLOCKSTATE(); goto out; } LIST_FOREACH(ownstp, &clp->lc_open, ls_list) { LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) { stp = LIST_FIRST(&openstp->ls_open); while (stp != LIST_END(&openstp->ls_open)) { nstp = LIST_NEXT(stp, ls_list); /* * If the owner matches, check for locks and * then free or return an error. */ if (stp->ls_ownerlen == new_stp->ls_ownerlen && !NFSBCMP(stp->ls_owner, new_stp->ls_owner, stp->ls_ownerlen)){ if (LIST_EMPTY(&stp->ls_lock)) { nfsrv_freelockowner(stp, NULL, 0, p); } else { NFSUNLOCKSTATE(); error = NFSERR_LOCKSHELD; goto out; } } stp = nstp; } } } NFSUNLOCKSTATE(); out: NFSEXITCODE(error); return (error); } /* * Get the file handle for a lock structure. */ static int nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p) { fhandle_t *fhp = NULL; int error; /* * For lock, use the new nfslock structure, otherwise just * a fhandle_t on the stack. */ if (flags & NFSLCK_OPEN) { KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL")); fhp = &new_lfp->lf_fh; } else if (nfhp) { fhp = nfhp; } else { panic("nfsrv_getlockfh"); } error = nfsvno_getfh(vp, fhp, p); NFSEXITCODE(error); return (error); } /* * Get an nfs lock structure. Allocate one, as required, and return a * pointer to it. * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock. */ static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit) { struct nfslockfile *lfp; fhandle_t *fhp = NULL, *tfhp; struct nfslockhashhead *hp; struct nfslockfile *new_lfp = NULL; /* * For lock, use the new nfslock structure, otherwise just * a fhandle_t on the stack. */ if (flags & NFSLCK_OPEN) { new_lfp = *new_lfpp; fhp = &new_lfp->lf_fh; } else if (nfhp) { fhp = nfhp; } else { panic("nfsrv_getlockfile"); } hp = NFSLOCKHASH(fhp); LIST_FOREACH(lfp, hp, lf_hash) { tfhp = &lfp->lf_fh; if (NFSVNO_CMPFH(fhp, tfhp)) { if (lockit) nfsrv_locklf(lfp); *lfpp = lfp; return (0); } } if (!(flags & NFSLCK_OPEN)) return (-1); /* * No match, so chain the new one into the list. */ LIST_INIT(&new_lfp->lf_open); LIST_INIT(&new_lfp->lf_lock); LIST_INIT(&new_lfp->lf_deleg); LIST_INIT(&new_lfp->lf_locallock); LIST_INIT(&new_lfp->lf_rollback); new_lfp->lf_locallock_lck.nfslock_usecnt = 0; new_lfp->lf_locallock_lck.nfslock_lock = 0; new_lfp->lf_usecount = 0; LIST_INSERT_HEAD(hp, new_lfp, lf_hash); *lfpp = new_lfp; *new_lfpp = NULL; return (0); } /* * This function adds a nfslock lock structure to the list for the associated * nfsstate and nfslockfile structures. It will be inserted after the * entry pointed at by insert_lop. */ static void nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp) { struct nfslock *lop, *nlop; new_lop->lo_stp = stp; new_lop->lo_lfp = lfp; if (stp != NULL) { /* Insert in increasing lo_first order */ lop = LIST_FIRST(&lfp->lf_lock); if (lop == LIST_END(&lfp->lf_lock) || new_lop->lo_first <= lop->lo_first) { LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile); } else { nlop = LIST_NEXT(lop, lo_lckfile); while (nlop != LIST_END(&lfp->lf_lock) && nlop->lo_first < new_lop->lo_first) { lop = nlop; nlop = LIST_NEXT(lop, lo_lckfile); } LIST_INSERT_AFTER(lop, new_lop, lo_lckfile); } } else { new_lop->lo_lckfile.le_prev = NULL; /* list not used */ } /* * Insert after insert_lop, which is overloaded as stp or lfp for * an empty list. */ if (stp == NULL && (struct nfslockfile *)insert_lop == lfp) LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner); else if ((struct nfsstate *)insert_lop == stp) LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner); else LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner); if (stp != NULL) { - nfsstatsv1.srvlocks++; + nfsstatsv1_p->srvlocks++; nfsrv_openpluslock++; } } /* * This function updates the locking for a lock owner and given file. It * maintains a list of lock ranges ordered on increasing file offset that * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style). * It always adds new_lop to the list and sometimes uses the one pointed * at by other_lopp. */ static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, struct nfslock **other_lopp, struct nfslockfile *lfp) { struct nfslock *new_lop = *new_lopp; struct nfslock *lop, *tlop, *ilop; struct nfslock *other_lop = *other_lopp; int unlock = 0, myfile = 0; u_int64_t tmp; /* * Work down the list until the lock is merged. */ if (new_lop->lo_flags & NFSLCK_UNLOCK) unlock = 1; if (stp != NULL) { ilop = (struct nfslock *)stp; lop = LIST_FIRST(&stp->ls_lock); } else { ilop = (struct nfslock *)lfp; lop = LIST_FIRST(&lfp->lf_locallock); } while (lop != NULL) { /* * Only check locks for this file that aren't before the start of * new lock's range. */ if (lop->lo_lfp == lfp) { myfile = 1; if (lop->lo_end >= new_lop->lo_first) { if (new_lop->lo_end < lop->lo_first) { /* * If the new lock ends before the start of the * current lock's range, no merge, just insert * the new lock. */ break; } if (new_lop->lo_flags == lop->lo_flags || (new_lop->lo_first <= lop->lo_first && new_lop->lo_end >= lop->lo_end)) { /* * This lock can be absorbed by the new lock/unlock. * This happens when it covers the entire range * of the old lock or is contiguous * with the old lock and is of the same type or an * unlock. */ if (lop->lo_first < new_lop->lo_first) new_lop->lo_first = lop->lo_first; if (lop->lo_end > new_lop->lo_end) new_lop->lo_end = lop->lo_end; tlop = lop; lop = LIST_NEXT(lop, lo_lckowner); nfsrv_freenfslock(tlop); continue; } /* * All these cases are for contiguous locks that are not the * same type, so they can't be merged. */ if (new_lop->lo_first <= lop->lo_first) { /* * This case is where the new lock overlaps with the * first part of the old lock. Move the start of the * old lock to just past the end of the new lock. The * new lock will be inserted in front of the old, since * ilop hasn't been updated. (We are done now.) */ lop->lo_first = new_lop->lo_end; break; } if (new_lop->lo_end >= lop->lo_end) { /* * This case is where the new lock overlaps with the * end of the old lock's range. Move the old lock's * end to just before the new lock's first and insert * the new lock after the old lock. * Might not be done yet, since the new lock could * overlap further locks with higher ranges. */ lop->lo_end = new_lop->lo_first; ilop = lop; lop = LIST_NEXT(lop, lo_lckowner); continue; } /* * The final case is where the new lock's range is in the * middle of the current lock's and splits the current lock * up. Use *other_lopp to handle the second part of the * split old lock range. (We are done now.) * For unlock, we use new_lop as other_lop and tmp, since * other_lop and new_lop are the same for this case. * We noted the unlock case above, so we don't need * new_lop->lo_flags any longer. */ tmp = new_lop->lo_first; if (other_lop == NULL) { if (!unlock) panic("nfsd srv update unlock"); other_lop = new_lop; *new_lopp = NULL; } other_lop->lo_first = new_lop->lo_end; other_lop->lo_end = lop->lo_end; other_lop->lo_flags = lop->lo_flags; other_lop->lo_stp = stp; other_lop->lo_lfp = lfp; lop->lo_end = tmp; nfsrv_insertlock(other_lop, lop, stp, lfp); *other_lopp = NULL; ilop = lop; break; } } ilop = lop; lop = LIST_NEXT(lop, lo_lckowner); if (myfile && (lop == NULL || lop->lo_lfp != lfp)) break; } /* * Insert the new lock in the list at the appropriate place. */ if (!unlock) { nfsrv_insertlock(new_lop, ilop, stp, lfp); *new_lopp = NULL; } } /* * This function handles sequencing of locks, etc. * It returns an error that indicates what the caller should do. */ static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, struct nfsstate *stp, struct nfsrvcache *op) { int error = 0; if ((nd->nd_flag & ND_NFSV41) != 0) /* NFSv4.1 ignores the open_seqid and lock_seqid. */ goto out; if (op != nd->nd_rp) panic("nfsrvstate checkseqid"); if (!(op->rc_flag & RC_INPROG)) panic("nfsrvstate not inprog"); if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) { printf("refcnt=%d\n", stp->ls_op->rc_refcnt); panic("nfsrvstate op refcnt"); } /* If ND_ERELOOKUP is set, the seqid has already been handled. */ if ((nd->nd_flag & ND_ERELOOKUP) != 0) goto out; if ((stp->ls_seq + 1) == seqid) { if (stp->ls_op) nfsrvd_derefcache(stp->ls_op); stp->ls_op = op; nfsrvd_refcache(op); stp->ls_seq = seqid; goto out; } else if (stp->ls_seq == seqid && stp->ls_op && op->rc_xid == stp->ls_op->rc_xid && op->rc_refcnt == 0 && op->rc_reqlen == stp->ls_op->rc_reqlen && op->rc_cksum == stp->ls_op->rc_cksum) { if (stp->ls_op->rc_flag & RC_INPROG) { error = NFSERR_DONTREPLY; goto out; } nd->nd_rp = stp->ls_op; nd->nd_rp->rc_flag |= RC_INPROG; nfsrvd_delcache(op); error = NFSERR_REPLYFROMCACHE; goto out; } error = NFSERR_BADSEQID; out: NFSEXITCODE2(error, nd); return (error); } /* * Get the client ip address for callbacks. If the strings can't be parsed, * just set lc_program to 0 to indicate no callbacks are possible. * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set * the address to the client's transport address. This won't be used * for callbacks, but can be printed out by nfsstats for info.) * Return error if the xdr can't be parsed, 0 otherwise. */ int nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp) { u_int32_t *tl; u_char *cp, *cp2; int i, j, maxalen = 0, minalen = 0; sa_family_t af; #ifdef INET struct sockaddr_in *rin = NULL, *sin; #endif #ifdef INET6 struct sockaddr_in6 *rin6 = NULL, *sin6; #endif u_char *addr; int error = 0, cantparse = 0; union { in_addr_t ival; u_char cval[4]; } ip; union { in_port_t sval; u_char cval[2]; } port; /* 8 is the maximum length of the port# string. */ addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK); clp->lc_req.nr_client = NULL; clp->lc_req.nr_lock = 0; af = AF_UNSPEC; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); if (i >= 3 && i <= 4) { error = nfsrv_mtostr(nd, addr, i); if (error) goto nfsmout; #ifdef INET if (!strcmp(addr, "tcp")) { clp->lc_flags |= LCL_TCPCALLBACK; clp->lc_req.nr_sotype = SOCK_STREAM; clp->lc_req.nr_soproto = IPPROTO_TCP; af = AF_INET; } else if (!strcmp(addr, "udp")) { clp->lc_req.nr_sotype = SOCK_DGRAM; clp->lc_req.nr_soproto = IPPROTO_UDP; af = AF_INET; } #endif #ifdef INET6 if (af == AF_UNSPEC) { if (!strcmp(addr, "tcp6")) { clp->lc_flags |= LCL_TCPCALLBACK; clp->lc_req.nr_sotype = SOCK_STREAM; clp->lc_req.nr_soproto = IPPROTO_TCP; af = AF_INET6; } else if (!strcmp(addr, "udp6")) { clp->lc_req.nr_sotype = SOCK_DGRAM; clp->lc_req.nr_soproto = IPPROTO_UDP; af = AF_INET6; } } #endif if (af == AF_UNSPEC) { cantparse = 1; } } else { cantparse = 1; if (i > 0) { error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; } } /* * The caller has allocated clp->lc_req.nr_nam to be large enough * for either AF_INET or AF_INET6 and zeroed out the contents. * maxalen is set to the maximum length of the host IP address string * plus 8 for the maximum length of the port#. * minalen is set to the minimum length of the host IP address string * plus 4 for the minimum length of the port#. * These lengths do not include NULL termination, * so INET[6]_ADDRSTRLEN - 1 is used in the calculations. */ switch (af) { #ifdef INET case AF_INET: rin = (struct sockaddr_in *)clp->lc_req.nr_nam; rin->sin_family = AF_INET; rin->sin_len = sizeof(struct sockaddr_in); maxalen = INET_ADDRSTRLEN - 1 + 8; minalen = 7 + 4; break; #endif #ifdef INET6 case AF_INET6: rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; rin6->sin6_family = AF_INET6; rin6->sin6_len = sizeof(struct sockaddr_in6); maxalen = INET6_ADDRSTRLEN - 1 + 8; minalen = 3 + 4; break; #endif } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); if (i < 0) { error = NFSERR_BADXDR; goto nfsmout; } else if (i == 0) { cantparse = 1; } else if (!cantparse && i <= maxalen && i >= minalen) { error = nfsrv_mtostr(nd, addr, i); if (error) goto nfsmout; /* * Parse out the address fields. We expect 6 decimal numbers * separated by '.'s for AF_INET and two decimal numbers * preceeded by '.'s for AF_INET6. */ cp = NULL; switch (af) { #ifdef INET6 /* * For AF_INET6, first parse the host address. */ case AF_INET6: cp = strchr(addr, '.'); if (cp != NULL) { *cp++ = '\0'; if (inet_pton(af, addr, &rin6->sin6_addr) == 1) i = 4; else { cp = NULL; cantparse = 1; } } break; #endif #ifdef INET case AF_INET: cp = addr; i = 0; break; #endif } while (cp != NULL && *cp && i < 6) { cp2 = cp; while (*cp2 && *cp2 != '.') cp2++; if (*cp2) *cp2++ = '\0'; else if (i != 5) { cantparse = 1; break; } j = nfsrv_getipnumber(cp); if (j >= 0) { if (i < 4) ip.cval[3 - i] = j; else port.cval[5 - i] = j; } else { cantparse = 1; break; } cp = cp2; i++; } if (!cantparse) { /* * The host address INADDR_ANY is (mis)used to indicate * "there is no valid callback address". */ switch (af) { #ifdef INET6 case AF_INET6: if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr, &in6addr_any)) rin6->sin6_port = htons(port.sval); else cantparse = 1; break; #endif #ifdef INET case AF_INET: if (ip.ival != INADDR_ANY) { rin->sin_addr.s_addr = htonl(ip.ival); rin->sin_port = htons(port.sval); } else { cantparse = 1; } break; #endif } } } else { cantparse = 1; if (i > 0) { error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; } } if (cantparse) { switch (nd->nd_nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)nd->nd_nam; rin = (struct sockaddr_in *)clp->lc_req.nr_nam; rin->sin_family = AF_INET; rin->sin_len = sizeof(struct sockaddr_in); rin->sin_addr.s_addr = sin->sin_addr.s_addr; rin->sin_port = 0x0; break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)nd->nd_nam; rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; rin6->sin6_family = AF_INET6; rin6->sin6_len = sizeof(struct sockaddr_in6); rin6->sin6_addr = sin6->sin6_addr; rin6->sin6_port = 0x0; break; #endif } clp->lc_program = 0; } nfsmout: free(addr, M_TEMP); NFSEXITCODE2(error, nd); return (error); } /* * Turn a string of up to three decimal digits into a number. Return -1 upon * error. */ static int nfsrv_getipnumber(u_char *cp) { int i = 0, j = 0; while (*cp) { if (j > 2 || *cp < '0' || *cp > '9') return (-1); i *= 10; i += (*cp - '0'); cp++; j++; } if (i < 256) return (i); return (-1); } /* * This function checks for restart conditions. */ static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, nfsv4stateid_t *stateidp, int specialid) { int ret = 0; /* * First check for a server restart. Open, LockT, ReleaseLockOwner * and DelegPurge have a clientid, the rest a stateid. */ if (flags & (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) { - if (clientid.lval[0] != nfsrvboottime) { + if (clientid.lval[0] != NFSD_VNET(nfsrvboottime)) { ret = NFSERR_STALECLIENTID; goto out; } - } else if (stateidp->other[0] != nfsrvboottime && + } else if (stateidp->other[0] != NFSD_VNET(nfsrvboottime) && specialid == 0) { ret = NFSERR_STALESTATEID; goto out; } /* * Read, Write, Setattr and LockT can return NFSERR_GRACE and do * not use a lock/open owner seqid#, so the check can be done now. * (The others will be checked, as required, later.) */ if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST))) goto out; NFSLOCKSTATE(); ret = nfsrv_checkgrace(NULL, NULL, flags); NFSUNLOCKSTATE(); out: NFSEXITCODE(ret); return (ret); } /* * Check for grace. */ static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, u_int32_t flags) { int error = 0, notreclaimed; struct nfsrv_stable *sp; - if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE | + if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE | NFSNSF_GRACEOVER)) == 0) { /* * First, check to see if all of the clients have done a * ReclaimComplete. If so, grace can end now. */ notreclaimed = 0; - LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { + LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, + nst_list) { if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) { notreclaimed = 1; break; } } if (notreclaimed == 0) - nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER | - NFSNSF_NEEDLOCK); + NFSD_VNET(nfsrv_stablefirst).nsf_flags |= + (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); } - if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { + if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) { if (flags & NFSLCK_RECLAIM) { error = NFSERR_NOGRACE; goto out; } } else { if (!(flags & NFSLCK_RECLAIM)) { error = NFSERR_GRACE; goto out; } if (nd != NULL && clp != NULL && (nd->nd_flag & ND_NFSV41) != 0 && (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) { error = NFSERR_NOGRACE; goto out; } /* * If grace is almost over and we are still getting Reclaims, * extend grace a bit. */ if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) > - nfsrv_stablefirst.nsf_eograce) - nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC + - NFSRV_LEASEDELTA; + NFSD_VNET(nfsrv_stablefirst).nsf_eograce) + NFSD_VNET(nfsrv_stablefirst).nsf_eograce = + NFSD_MONOSEC + NFSRV_LEASEDELTA; } out: NFSEXITCODE(error); return (error); } /* * Do a server callback. * The "trunc" argument is slightly overloaded and refers to different * boolean arguments for CBRECALL and CBLAYOUTRECALL. */ static int nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p) { struct mbuf *m; u_int32_t *tl; struct nfsrv_descript *nd; struct ucred *cred; int error = 0, slotpos; u_int32_t callback; struct nfsdsession *sep = NULL; uint64_t tval; bool dotls; nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); cred = newnfs_getcred(); NFSLOCKSTATE(); /* mostly for lc_cbref++ */ if (clp->lc_flags & LCL_NEEDSCONFIRM) { NFSUNLOCKSTATE(); panic("docallb"); } clp->lc_cbref++; /* * Fill the callback program# and version into the request * structure for newnfs_connect() to use. */ clp->lc_req.nr_prog = clp->lc_program; #ifdef notnow if ((clp->lc_flags & LCL_NFSV41) != 0) clp->lc_req.nr_vers = NFSV41_CBVERS; else #endif clp->lc_req.nr_vers = NFSV4_CBVERS; /* * First, fill in some of the fields of nd and cr. */ nd->nd_flag = ND_NFSV4; if (clp->lc_flags & LCL_GSS) nd->nd_flag |= ND_KERBV; if ((clp->lc_flags & LCL_NFSV41) != 0) nd->nd_flag |= ND_NFSV41; if ((clp->lc_flags & LCL_NFSV42) != 0) nd->nd_flag |= ND_NFSV42; nd->nd_repstat = 0; cred->cr_uid = clp->lc_uid; cred->cr_gid = clp->lc_gid; callback = clp->lc_callback; NFSUNLOCKSTATE(); cred->cr_ngroups = 1; /* * Get the first mbuf for the request. */ MGET(m, M_WAITOK, MT_DATA); m->m_len = 0; nd->nd_mreq = nd->nd_mb = m; nd->nd_bpos = mtod(m, caddr_t); /* * and build the callback request. */ if (procnum == NFSV4OP_CBGETATTR) { nd->nd_procnum = NFSV4PROC_CBCOMPOUND; error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR, "CB Getattr", &sep, &slotpos); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0); (void)nfsrv_putattrbit(nd, attrbitp); } else if (procnum == NFSV4OP_CBRECALL) { nd->nd_procnum = NFSV4PROC_CBCOMPOUND; error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL, "CB Recall", &sep, &slotpos); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID); *tl++ = txdr_unsigned(stateidp->seqid); NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl, NFSX_STATEIDOTHER); tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); if (trunc) *tl = newnfs_true; else *tl = newnfs_false; (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0); } else if (procnum == NFSV4OP_CBLAYOUTRECALL) { NFSD_DEBUG(4, "docallback layout recall\n"); nd->nd_procnum = NFSV4PROC_CBCOMPOUND; error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos); NFSD_DEBUG(4, "aft cbcallargs=%d\n", error); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(laytype); *tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY); if (trunc) *tl++ = newnfs_true; else *tl++ = newnfs_false; *tl = txdr_unsigned(NFSV4LAYOUTRET_FILE); nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID); tval = 0; txdr_hyper(tval, tl); tl += 2; tval = UINT64_MAX; txdr_hyper(tval, tl); tl += 2; *tl++ = txdr_unsigned(stateidp->seqid); NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER); tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); NFSD_DEBUG(4, "aft args\n"); } else if (procnum == NFSV4PROC_CBNULL) { nd->nd_procnum = NFSV4PROC_CBNULL; if ((clp->lc_flags & LCL_NFSV41) != 0) { error = nfsv4_getcbsession(clp, &sep); if (error != 0) { m_freem(nd->nd_mreq); goto errout; } } } else { error = NFSERR_SERVERFAULT; m_freem(nd->nd_mreq); goto errout; } /* * Call newnfs_connect(), as required, and then newnfs_request(). */ dotls = false; if ((clp->lc_flags & LCL_TLSCB) != 0) dotls = true; (void) newnfs_sndlock(&clp->lc_req.nr_lock); if (clp->lc_req.nr_client == NULL) { if ((clp->lc_flags & LCL_NFSV41) != 0) { error = ECONNREFUSED; if (procnum != NFSV4PROC_CBNULL) nfsv4_freeslot(&sep->sess_cbsess, slotpos, true); nfsrv_freesession(sep, NULL); } else if (nd->nd_procnum == NFSV4PROC_CBNULL) error = newnfs_connect(NULL, &clp->lc_req, cred, NULL, 1, dotls, &clp->lc_req.nr_client); else error = newnfs_connect(NULL, &clp->lc_req, cred, NULL, 3, dotls, &clp->lc_req.nr_client); } newnfs_sndunlock(&clp->lc_req.nr_lock); NFSD_DEBUG(4, "aft sndunlock=%d\n", error); if (!error) { if ((nd->nd_flag & ND_NFSV41) != 0) { KASSERT(sep != NULL, ("sep NULL")); if (sep->sess_cbsess.nfsess_xprt != NULL) error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL, NULL, cred, clp->lc_program, clp->lc_req.nr_vers, NULL, 1, NULL, &sep->sess_cbsess); else { /* * This should probably never occur, but if a * client somehow does an RPC without a * SequenceID Op that causes a callback just * after the nfsd threads have been terminated * and restared we could conceivably get here * without a backchannel xprt. */ printf("nfsrv_docallback: no xprt\n"); error = ECONNREFUSED; } NFSD_DEBUG(4, "aft newnfs_request=%d\n", error); if (error != 0 && procnum != NFSV4PROC_CBNULL) { /* * It is likely that the callback was never * processed by the client and, as such, * the sequence# for the session slot needs * to be backed up by one to avoid a * NFSERR_SEQMISORDERED error reply. * For the unlikely case where the callback * was processed by the client, this will * make the next callback on the slot * appear to be a retry. * Since callbacks never specify that the * reply be cached, this "apparent retry" * should not be a problem. */ nfsv4_freeslot(&sep->sess_cbsess, slotpos, true); } nfsrv_freesession(sep, NULL); } else error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL, NULL, cred, clp->lc_program, clp->lc_req.nr_vers, NULL, 1, NULL, NULL); } errout: NFSFREECRED(cred); /* * If error is set here, the Callback path isn't working * properly, so twiddle the appropriate LCL_ flags. * (nd_repstat != 0 indicates the Callback path is working, * but the callback failed on the client.) */ if (error) { /* * Mark the callback pathway down, which disabled issuing * of delegations and gets Renew to return NFSERR_CBPATHDOWN. */ NFSLOCKSTATE(); clp->lc_flags |= LCL_CBDOWN; NFSUNLOCKSTATE(); } else { /* * Callback worked. If the callback path was down, disable * callbacks, so no more delegations will be issued. (This * is done on the assumption that the callback pathway is * flakey.) */ NFSLOCKSTATE(); if (clp->lc_flags & LCL_CBDOWN) clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON); NFSUNLOCKSTATE(); if (nd->nd_repstat) { error = nd->nd_repstat; NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n", procnum, error); } else if (error == 0 && procnum == NFSV4OP_CBGETATTR) error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, p, NULL); m_freem(nd->nd_mrep); } NFSLOCKSTATE(); clp->lc_cbref--; if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) { clp->lc_flags &= ~LCL_WAKEUPWANTED; wakeup(clp); } NFSUNLOCKSTATE(); free(nd, M_TEMP); NFSEXITCODE(error); return (error); } /* * Set up the compound RPC for the callback. */ static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, uint32_t callback, int op, const char *optag, struct nfsdsession **sepp, int *slotposp) { uint32_t *tl; int error, len; len = strlen(optag); (void)nfsm_strtom(nd, optag, len); NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED); if ((nd->nd_flag & ND_NFSV41) != 0) { if ((nd->nd_flag & ND_NFSV42) != 0) *tl++ = txdr_unsigned(NFSV42_MINORVERSION); else *tl++ = txdr_unsigned(NFSV41_MINORVERSION); *tl++ = txdr_unsigned(callback); *tl++ = txdr_unsigned(2); *tl = txdr_unsigned(NFSV4OP_CBSEQUENCE); error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp); if (error != 0) return (error); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(op); } else { *tl++ = txdr_unsigned(NFSV4_MINORVERSION); *tl++ = txdr_unsigned(callback); *tl++ = txdr_unsigned(1); *tl = txdr_unsigned(op); } return (0); } /* * Return the next index# for a clientid. Mostly just increment and return * the next one, but... if the 32bit unsigned does actually wrap around, * it should be rebooted. * At an average rate of one new client per second, it will wrap around in * approximately 136 years. (I think the server will have been shut * down or rebooted before then.) */ static u_int32_t nfsrv_nextclientindex(void) { static u_int32_t client_index = 0; client_index++; if (client_index != 0) return (client_index); printf("%s: out of clientids\n", __func__); return (client_index); } /* * Return the next index# for a stateid. Mostly just increment and return * the next one, but... if the 32bit unsigned does actually wrap around * (will a BSD server stay up that long?), find * new start and end values. */ static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp) { struct nfsstate *stp; int i; u_int32_t canuse, min_index, max_index; if (!(clp->lc_flags & LCL_INDEXNOTOK)) { clp->lc_stateindex++; if (clp->lc_stateindex != clp->lc_statemaxindex) return (clp->lc_stateindex); } /* * Yuck, we've hit the end. * Look for a new min and max. */ min_index = 0; max_index = 0xffffffff; for (i = 0; i < nfsrv_statehashsize; i++) { LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { if (stp->ls_stateid.other[2] > 0x80000000) { if (stp->ls_stateid.other[2] < max_index) max_index = stp->ls_stateid.other[2]; } else { if (stp->ls_stateid.other[2] > min_index) min_index = stp->ls_stateid.other[2]; } } } /* * Yikes, highly unlikely, but I'll handle it anyhow. */ if (min_index == 0x80000000 && max_index == 0x80000001) { canuse = 0; /* * Loop around until we find an unused entry. Return that * and set LCL_INDEXNOTOK, so the search will continue next time. * (This is one of those rare cases where a goto is the * cleanest way to code the loop.) */ tryagain: for (i = 0; i < nfsrv_statehashsize; i++) { LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { if (stp->ls_stateid.other[2] == canuse) { canuse++; goto tryagain; } } } clp->lc_flags |= LCL_INDEXNOTOK; return (canuse); } /* * Ok to start again from min + 1. */ clp->lc_stateindex = min_index + 1; clp->lc_statemaxindex = max_index; clp->lc_flags &= ~LCL_INDEXNOTOK; return (clp->lc_stateindex); } /* * The following functions handle the stable storage file that deals with * the edge conditions described in RFC3530 Sec. 8.6.3. * The file is as follows: * - a single record at the beginning that has the lease time of the * previous server instance (before the last reboot) and the nfsrvboottime * values for the previous server boots. * These previous boot times are used to ensure that the current * nfsrvboottime does not, somehow, get set to a previous one. * (This is important so that Stale ClientIDs and StateIDs can * be recognized.) * The number of previous nfsvrboottime values precedes the list. * - followed by some number of appended records with: * - client id string * - flag that indicates it is a record revoking state via lease * expiration or similar * OR has successfully acquired state. * These structures vary in length, with the client string at the end, up * to NFSV4_OPAQUELIMIT in size. * * At the end of the grace period, the file is truncated, the first * record is rewritten with updated information and any acquired state * records for successful reclaims of state are written. * * Subsequent records are appended when the first state is issued to * a client and when state is revoked for a client. * * When reading the file in, state issued records that come later in * the file override older ones, since the append log is in cronological order. * If, for some reason, the file can't be read, the grace period is * immediately terminated and all reclaims get NFSERR_NOGRACE. */ /* * Read in the stable storage file. Called by nfssvc() before the nfsd * processes start servicing requests. */ void nfsrv_setupstable(NFSPROC_T *p) { - struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; + struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst); struct nfsrv_stable *sp, *nsp; struct nfst_rec *tsp; int error, i, tryagain; off_t off = 0; ssize_t aresid, len; /* * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without * a reboot, so state has not been lost. */ if (sf->nsf_flags & NFSNSF_UPDATEDONE) return; /* * Set Grace over just until the file reads successfully. */ - nfsrvboottime = time_second; + NFSD_VNET(nfsrvboottime) = time_second; LIST_INIT(&sf->nsf_head); sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA; if (sf->nsf_fp == NULL) return; error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); if (error || aresid || sf->nsf_numboots == 0 || sf->nsf_numboots > NFSNSF_MAXNUMBOOTS) return; /* * Now, read in the boottimes. */ sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) * - sizeof (time_t), M_TEMP, M_WAITOK); + sizeof(time_t), M_TEMP, M_WAITOK); off = sizeof (struct nfsf_rec); error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); if (error || aresid) { free(sf->nsf_bootvals, M_TEMP); sf->nsf_bootvals = NULL; return; } /* * Make sure this nfsrvboottime is different from all recorded * previous ones. */ do { tryagain = 0; for (i = 0; i < sf->nsf_numboots; i++) { - if (nfsrvboottime == sf->nsf_bootvals[i]) { - nfsrvboottime++; + if (NFSD_VNET(nfsrvboottime) == sf->nsf_bootvals[i]) { + NFSD_VNET(nfsrvboottime)++; tryagain = 1; break; } } } while (tryagain); sf->nsf_flags |= NFSNSF_OK; off += (sf->nsf_numboots * sizeof (time_t)); /* * Read through the file, building a list of records for grace * checking. * Each record is between sizeof (struct nfst_rec) and * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1 * and is actually sizeof (struct nfst_rec) + nst_len - 1. */ tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK); do { error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1, off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid; if (error || (len > 0 && (len < sizeof (struct nfst_rec) || len < (sizeof (struct nfst_rec) + tsp->len - 1)))) { /* * Yuck, the file has been corrupted, so just return * after clearing out any restart state, so the grace period * is over. */ LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { LIST_REMOVE(sp, nst_list); free(sp, M_TEMP); } free(tsp, M_TEMP); sf->nsf_flags &= ~NFSNSF_OK; free(sf->nsf_bootvals, M_TEMP); sf->nsf_bootvals = NULL; return; } if (len > 0) { off += sizeof (struct nfst_rec) + tsp->len - 1; /* * Search the list for a matching client. */ LIST_FOREACH(sp, &sf->nsf_head, nst_list) { if (tsp->len == sp->nst_len && !NFSBCMP(tsp->client, sp->nst_client, tsp->len)) break; } if (sp == LIST_END(&sf->nsf_head)) { sp = (struct nfsrv_stable *)malloc(tsp->len + sizeof (struct nfsrv_stable) - 1, M_TEMP, M_WAITOK); NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec, sizeof (struct nfst_rec) + tsp->len - 1); LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list); } else { if (tsp->flag == NFSNST_REVOKE) sp->nst_flag |= NFSNST_REVOKE; else /* * A subsequent timestamp indicates the client * did a setclientid/confirm and any previous * revoke is no longer relevant. */ sp->nst_flag &= ~NFSNST_REVOKE; } } } while (len > 0); free(tsp, M_TEMP); sf->nsf_flags = NFSNSF_OK; sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease + NFSRV_LEASEDELTA; } /* * Update the stable storage file, now that the grace period is over. */ void nfsrv_updatestable(NFSPROC_T *p) { - struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; + struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst); struct nfsrv_stable *sp, *nsp; int i; struct nfsvattr nva; vnode_t vp; #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000) mount_t mp = NULL; #endif int error; if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE)) return; sf->nsf_flags |= NFSNSF_UPDATEDONE; /* * Ok, we need to rewrite the stable storage file. * - truncate to 0 length * - write the new first structure * - loop through the data structures, writing out any that * have timestamps older than the old boot */ if (sf->nsf_bootvals) { sf->nsf_numboots++; for (i = sf->nsf_numboots - 2; i >= 0; i--) sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i]; } else { sf->nsf_numboots = 1; - sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t), + sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t), M_TEMP, M_WAITOK); } - sf->nsf_bootvals[0] = nfsrvboottime; + sf->nsf_bootvals[0] = NFSD_VNET(nfsrvboottime); sf->nsf_lease = nfsrv_lease; NFSVNO_ATTRINIT(&nva); NFSVNO_SETATTRVAL(&nva, size, 0); vp = NFSFPVNODE(sf->nsf_fp); vn_start_write(vp, &mp, V_WAIT); if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p, NULL); NFSVOPUNLOCK(vp); } else error = EPERM; vn_finished_write(mp); if (!error) error = NFSD_RDWR(UIO_WRITE, vp, (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0, UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p); if (!error) error = NFSD_RDWR(UIO_WRITE, vp, (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), (off_t)(sizeof (struct nfsf_rec)), UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p); free(sf->nsf_bootvals, M_TEMP); sf->nsf_bootvals = NULL; if (error) { sf->nsf_flags &= ~NFSNSF_OK; printf("EEK! Can't write NfsV4 stable storage file\n"); return; } sf->nsf_flags |= NFSNSF_OK; /* * Loop through the list and write out timestamp records for * any clients that successfully reclaimed state. */ LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { if (sp->nst_flag & NFSNST_GOTSTATE) { nfsrv_writestable(sp->nst_client, sp->nst_len, NFSNST_NEWSTATE, p); sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE; } LIST_REMOVE(sp, nst_list); free(sp, M_TEMP); } nfsrv_backupstable(); } /* * Append a record to the stable storage file. */ void nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p) { - struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; + struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst); struct nfst_rec *sp; int error; if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL) return; sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + len - 1, M_TEMP, M_WAITOK); sp->len = len; NFSBCOPY(client, sp->client, len); sp->flag = flag; error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp), (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0, UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p); free(sp, M_TEMP); if (error) { sf->nsf_flags &= ~NFSNSF_OK; printf("EEK! Can't write NfsV4 stable storage file\n"); } } /* * This function is called during the grace period to mark a client * that successfully reclaimed state. */ static void nfsrv_markstable(struct nfsclient *clp) { struct nfsrv_stable *sp; /* * First find the client structure. */ - LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { + LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) { if (sp->nst_len == clp->lc_idlen && !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) break; } - if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) + if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head)) return; /* * Now, just mark it and set the nfsclient back pointer. */ sp->nst_flag |= NFSNST_GOTSTATE; sp->nst_clp = clp; } /* * This function is called when a NFSv4.1 client does a ReclaimComplete. * Very similar to nfsrv_markstable(), except for the flag being set. */ static void nfsrv_markreclaim(struct nfsclient *clp) { struct nfsrv_stable *sp; /* * First find the client structure. */ - LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { + LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) { if (sp->nst_len == clp->lc_idlen && !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) break; } - if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) + if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head)) return; /* * Now, just set the flag. */ sp->nst_flag |= NFSNST_RECLAIMED; } /* * This function is called for a reclaim, to see if it gets grace. * It returns 0 if a reclaim is allowed, 1 otherwise. */ static int nfsrv_checkstable(struct nfsclient *clp) { struct nfsrv_stable *sp; /* * First, find the entry for the client. */ - LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { + LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) { if (sp->nst_len == clp->lc_idlen && !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) break; } /* * If not in the list, state was revoked or no state was issued * since the previous reboot, a reclaim is denied. */ - if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) || + if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head) || (sp->nst_flag & NFSNST_REVOKE) || - !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK)) + !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK)) return (1); return (0); } /* * Test for and try to clear out a conflicting client. This is called by * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients * a found. * The trick here is that it can't revoke a conflicting client with an * expired lease unless it holds the v4root lock, so... * If no v4root lock, get the lock and return 1 to indicate "try again". * Return 0 to indicate the conflict can't be revoked and 1 to indicate * the revocation worked and the conflicting client is "bye, bye", so it * can be tried again. * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK(). * Unlocks State before a non-zero value is returned. */ static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp, NFSPROC_T *p) { int gotlock, lktype = 0; /* * If lease hasn't expired, we can't fix it. */ if (clp->lc_expiry >= NFSD_MONOSEC || - !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) + !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) return (0); if (*haslockp == 0) { NFSUNLOCKSTATE(); if (vp != NULL) { lktype = NFSVOPISLOCKED(vp); NFSVOPUNLOCK(vp); } NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!gotlock); NFSUNLOCKV4ROOTMUTEX(); *haslockp = 1; if (vp != NULL) { NFSVOPLOCK(vp, lktype | LK_RETRY); if (VN_IS_DOOMED(vp)) return (2); } return (1); } NFSUNLOCKSTATE(); /* * Ok, we can expire the conflicting client. */ nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_backupstable(); nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); nfsrv_zapclient(clp, p); return (1); } /* * Resolve a delegation conflict. * Returns 0 to indicate the conflict was resolved without sleeping. * Return -1 to indicate that the caller should check for conflicts again. * Return > 0 for an error that should be returned, normally NFSERR_DELAY. * * Also, manipulate the nfsv4root_lock, as required. It isn't changed * for a return of 0, since there was no sleep and it could be required * later. It is released for a return of NFSERR_DELAY, since the caller * will return that error. It is released when a sleep was done waiting * for the delegation to be returned or expire (so that other nfsds can * handle ops). Then, it must be acquired for the write to stable storage. * (This function is somewhat similar to nfsrv_clientconflict(), but * the semantics differ in a couple of subtle ways. The return of 0 * indicates the conflict was resolved without sleeping here, not * that the conflict can't be resolved and the handling of nfsv4root_lock * differs, as noted above.) * Unlocks State before returning a non-zero value. */ static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p, vnode_t vp) { struct nfsclient *clp = stp->ls_clp; int gotlock, error, lktype = 0, retrycnt, zapped_clp; nfsv4stateid_t tstateid; fhandle_t tfh; /* * If the conflict is with an old delegation... */ if (stp->ls_flags & NFSLCK_OLDDELEG) { /* * You can delete it, if it has expired. */ if (clp->lc_delegtime < NFSD_MONOSEC) { nfsrv_freedeleg(stp); NFSUNLOCKSTATE(); error = -1; goto out; } NFSUNLOCKSTATE(); /* * During this delay, the old delegation could expire or it * could be recovered by the client via an Open with * CLAIM_DELEGATE_PREV. * Release the nfsv4root_lock, if held. */ if (*haslockp) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_DELAY; goto out; } /* * It's a current delegation, so: * - check to see if the delegation has expired * - if so, get the v4root lock and then expire it */ if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall < NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC && stp->ls_delegtime >= NFSD_MONOSEC)) { /* * - do a recall callback, since not yet done * For now, never allow truncate to be set. To use * truncate safely, it must be guaranteed that the * Remove, Rename or Setattr with size of 0 will * succeed and that would require major changes to * the VFS/Vnode OPs. * Set the expiry time large enough so that it won't expire * until after the callback, then set it correctly, once * the callback is done. (The delegation will now time * out whether or not the Recall worked ok. The timeout * will be extended when ops are done on the delegation * stateid, up to the timelimit.) */ if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) { stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) + NFSRV_LEASEDELTA; stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) + NFSRV_LEASEDELTA; stp->ls_flags |= NFSLCK_DELEGRECALL; } stp->ls_lastrecall = time_uptime + 1; /* * Loop NFSRV_CBRETRYCNT times while the CBRecall replies * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done * in order to try and avoid a race that could happen * when a CBRecall request passed the Open reply with * the delegation in it when transitting the network. * Since nfsrv_docallback will sleep, don't use stp after * the call. */ NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid, sizeof (tstateid)); NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh, sizeof (tfh)); NFSUNLOCKSTATE(); if (*haslockp) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } retrycnt = 0; do { error = nfsrv_docallback(clp, NFSV4OP_CBRECALL, &tstateid, 0, &tfh, NULL, NULL, 0, p); retrycnt++; } while ((error == NFSERR_BADSTATEID || error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT); error = NFSERR_DELAY; goto out; } if (clp->lc_expiry >= NFSD_MONOSEC && stp->ls_delegtime >= NFSD_MONOSEC) { NFSUNLOCKSTATE(); /* * A recall has been done, but it has not yet expired. * So, RETURN_DELAY. */ if (*haslockp) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } error = NFSERR_DELAY; goto out; } /* * If we don't yet have the lock, just get it and then return, * since we need that before deleting expired state, such as * this delegation. * When getting the lock, unlock the vnode, so other nfsds that * are in progress, won't get stuck waiting for the vnode lock. */ if (*haslockp == 0) { NFSUNLOCKSTATE(); if (vp != NULL) { lktype = NFSVOPISLOCKED(vp); NFSVOPUNLOCK(vp); } NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (!gotlock); NFSUNLOCKV4ROOTMUTEX(); *haslockp = 1; if (vp != NULL) { NFSVOPLOCK(vp, lktype | LK_RETRY); if (VN_IS_DOOMED(vp)) { *haslockp = 0; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); error = NFSERR_PERM; goto out; } } error = -1; goto out; } NFSUNLOCKSTATE(); /* * Ok, we can delete the expired delegation. * First, write the Revoke record to stable storage and then * clear out the conflict. * Since all other nfsd threads are now blocked, we can safely * sleep without the state changing. */ nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); nfsrv_backupstable(); if (clp->lc_expiry < NFSD_MONOSEC) { nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); LIST_REMOVE(clp, lc_hash); zapped_clp = 1; } else { nfsrv_freedeleg(stp); zapped_clp = 0; } if (zapped_clp) nfsrv_zapclient(clp, p); error = -1; out: NFSEXITCODE(error); return (error); } /* * Check for a remove allowed, if remove is set to 1 and get rid of * delegations. */ int nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; struct nfslockfile *lfp; int error, haslock = 0; fhandle_t nfh; clp = NULL; /* * First, get the lock file structure. * (A return of -1 means no associated state, so remove ok.) */ error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); tryagain: NFSLOCKSTATE(); if (error == 0 && clientid.qval != 0) error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (!error) error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); if (error) { NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } if (error == -1) error = 0; goto out; } /* * Now, we must Recall any delegations. */ error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p); if (error) { /* * nfsrv_cleandeleg() unlocks state for non-zero * return. */ if (error == -1) goto tryagain; if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } goto out; } /* * Now, look for a conflicting open share. */ if (remove) { /* * If the entry in the directory was the last reference to the * corresponding filesystem object, the object can be destroyed * */ if(lfp->lf_usecount>1) LIST_FOREACH(stp, &lfp->lf_open, ls_file) { if (stp->ls_flags & NFSLCK_WRITEDENY) { error = NFSERR_FILEOPEN; break; } } } NFSUNLOCKSTATE(); if (haslock) { NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); } out: NFSEXITCODE(error); return (error); } /* * Clear out all delegations for the file referred to by lfp. * May return NFSERR_DELAY, if there will be a delay waiting for * delegations to expire. * Returns -1 to indicate it slept while recalling a delegation. * This function has the side effect of deleting the nfslockfile structure, * if it no longer has associated state and didn't have to sleep. * Unlocks State before a non-zero value is returned. */ static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, struct nfsclient *clp, int *haslockp, NFSPROC_T *p) { struct nfsstate *stp, *nstp; int ret = 0; stp = LIST_FIRST(&lfp->lf_deleg); while (stp != LIST_END(&lfp->lf_deleg)) { nstp = LIST_NEXT(stp, ls_file); if (stp->ls_clp != clp) { ret = nfsrv_delegconflict(stp, haslockp, p, vp); if (ret) { /* * nfsrv_delegconflict() unlocks state * when it returns non-zero. */ goto out; } } stp = nstp; } out: NFSEXITCODE(ret); return (ret); } /* * There are certain operations that, when being done outside of NFSv4, * require that any NFSv4 delegation for the file be recalled. * This function is to be called for those cases: * VOP_RENAME() - When a delegation is being recalled for any reason, * the client may have to do Opens against the server, using the file's * final component name. If the file has been renamed on the server, * that component name will be incorrect and the Open will fail. * VOP_REMOVE() - Theoretically, a client could Open a file after it has * been removed on the server, if there is a delegation issued to * that client for the file. I say "theoretically" since clients * normally do an Access Op before the Open and that Access Op will * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so * they will detect the file's removal in the same manner. (There is * one case where RFC3530 allows a client to do an Open without first * doing an Access Op, which is passage of a check against the ACE * returned with a Write delegation, but current practice is to ignore * the ACE and always do an Access Op.) * Since the functions can only be called with an unlocked vnode, this * can't be done at this time. * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range * locks locally in the client, which are not visible to the server. To * deal with this, issuing of delegations for a vnode must be disabled * and all delegations for the vnode recalled. This is done via the * second function, using the VV_DISABLEDELEG vflag on the vnode. */ void nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p) { time_t starttime; int error; /* * First, check to see if the server is currently running and it has * been called for a regular file when issuing delegations. */ - if (newnfs_numnfsd == 0 || vp->v_type != VREG || + if (NFSD_VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG || nfsrv_issuedelegs == 0) return; KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp)); /* * First, get a reference on the nfsv4rootfs_lock so that an * exclusive lock cannot be acquired by another thread. */ NFSLOCKV4ROOTMUTEX(); nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); NFSUNLOCKV4ROOTMUTEX(); /* * Now, call nfsrv_checkremove() in a loop while it returns * NFSERR_DELAY. Return upon any other error or when timed out. */ starttime = NFSD_MONOSEC; do { if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { error = nfsrv_checkremove(vp, 0, NULL, (nfsquad_t)((u_quad_t)0), p); NFSVOPUNLOCK(vp); } else error = EPERM; if (error == NFSERR_DELAY) { if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO) break; /* Sleep for a short period of time */ (void) nfs_catnap(PZERO, 0, "nfsremove"); } } while (error == NFSERR_DELAY); NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); NFSUNLOCKV4ROOTMUTEX(); } void nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p) { #ifdef VV_DISABLEDELEG /* * First, flag issuance of delegations disabled. */ atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG); #endif /* * Then call nfsd_recalldelegation() to get rid of all extant * delegations. */ nfsd_recalldelegation(vp, p); } /* * Check for conflicting locks, etc. and then get rid of delegations. * (At one point I thought that I should get rid of delegations for any * Setattr, since it could potentially disallow the I/O op (read or write) * allowed by the delegation. However, Setattr Ops that aren't changing * the size get a stateid of all 0s, so you can't tell if it is a delegation * for the same client or a different one, so I decided to only get rid * of delegations for other clients when the size is being changed.) * In general, a Setattr can disable NFS I/O Ops that are outstanding, such * as Write backs, even if there is no delegation, so it really isn't any * different?) */ int nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct nfsexstuff *exp, NFSPROC_T *p) { struct nfsstate st, *stp = &st; struct nfslock lo, *lop = &lo; int error = 0; nfsquad_t clientid; if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) { stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS); lop->lo_first = nvap->na_size; } else { stp->ls_flags = 0; lop->lo_first = 0; } if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) || NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL)) stp->ls_flags |= NFSLCK_SETATTR; if (stp->ls_flags == 0) goto out; lop->lo_end = NFS64BITSSET; lop->lo_flags = NFSLCK_WRITE; stp->ls_ownerlen = 0; stp->ls_op = NULL; stp->ls_uid = nd->nd_cred->cr_uid; stp->ls_stateid.seqid = stateidp->seqid; clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0]; clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1]; stp->ls_stateid.other[2] = stateidp->other[2]; error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid, stateidp, exp, nd, p); out: NFSEXITCODE2(error, nd); return (error); } /* * Check for a write delegation and do a CBGETATTR if there is one, updating * the attributes, as required. * Should I return an error if I can't get the attributes? (For now, I'll * just return ok. */ int nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p) { struct nfsstate *stp; struct nfslockfile *lfp; struct nfsclient *clp; struct nfsvattr nva; fhandle_t nfh; int error = 0; nfsattrbit_t cbbits; u_quad_t delegfilerev; NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits); if (!NFSNONZERO_ATTRBIT(&cbbits)) goto out; if (nfsrv_writedelegcnt == 0) goto out; /* * Get the lock file structure. * (A return of -1 means no associated state, so return ok.) */ error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); NFSLOCKSTATE(); if (!error) error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); if (error) { NFSUNLOCKSTATE(); if (error == -1) error = 0; goto out; } /* * Now, look for a write delegation. */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if (stp->ls_flags & NFSLCK_DELEGWRITE) break; } if (stp == LIST_END(&lfp->lf_deleg)) { NFSUNLOCKSTATE(); goto out; } clp = stp->ls_clp; /* If the clientid is not confirmed, ignore the delegation. */ if (clp->lc_flags & LCL_NEEDSCONFIRM) { NFSUNLOCKSTATE(); goto out; } delegfilerev = stp->ls_filerev; /* * If the Write delegation was issued as a part of this Compound RPC * or if we have an Implied Clientid (used in a previous Op in this * compound) and it is the client the delegation was issued to, * just return ok. * I also assume that it is from the same client iff the network * host IP address is the same as the callback address. (Not * exactly correct by the RFC, but avoids a lot of Getattr * callbacks.) */ if (nd->nd_compref == stp->ls_compref || ((nd->nd_flag & ND_IMPLIEDCLID) && clp->lc_clientid.qval == nd->nd_clientid.qval) || nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) { NFSUNLOCKSTATE(); goto out; } /* * We are now done with the delegation state structure, * so the statelock can be released and we can now tsleep(). */ /* * Now, we must do the CB Getattr callback, to see if Change or Size * has changed. */ if (clp->lc_expiry >= NFSD_MONOSEC) { NFSUNLOCKSTATE(); NFSVNO_ATTRINIT(&nva); nva.na_filerev = NFS64BITSSET; error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL, 0, &nfh, &nva, &cbbits, 0, p); if (!error) { if ((nva.na_filerev != NFS64BITSSET && nva.na_filerev > delegfilerev) || (NFSVNO_ISSETSIZE(&nva) && nva.na_size != nvap->na_size)) { error = nfsvno_updfilerev(vp, nvap, nd, p); if (NFSVNO_ISSETSIZE(&nva)) nvap->na_size = nva.na_size; } } else error = 0; /* Ignore callback errors for now. */ } else { NFSUNLOCKSTATE(); } out: NFSEXITCODE2(error, nd); return (error); } /* * This function looks for openowners that haven't had any opens for * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS * is set. */ void nfsrv_throwawayopens(NFSPROC_T *p) { struct nfsclient *clp, *nclp; struct nfsstate *stp, *nstp; int i; NFSLOCKSTATE(); - nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS; + NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS; /* * For each client... */ for (i = 0; i < nfsrv_clienthashsize; i++) { - LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { + LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash, + nclp) { LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) { if (LIST_EMPTY(&stp->ls_open) && (stp->ls_noopens > NFSNOOPEN || (nfsrv_openpluslock * 2) > nfsrv_v4statelimit)) nfsrv_freeopenowner(stp, 0, p); } } } NFSUNLOCKSTATE(); } /* * This function checks to see if the credentials are the same. * The check for same credentials is needed for state management operations * for NFSv4.0 where 1 is returned if not same, 0 is returned otherwise. */ static int nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp) { /* For NFSv4.1/4.2, SP4_NONE always allows this. */ if ((nd->nd_flag & ND_NFSV41) != 0) return (0); if (nd->nd_flag & ND_GSS) { if (!(clp->lc_flags & LCL_GSS)) return (1); if (clp->lc_flags & LCL_NAME) { if (nd->nd_princlen != clp->lc_namelen || NFSBCMP(nd->nd_principal, clp->lc_name, clp->lc_namelen)) return (1); else return (0); } if (nd->nd_cred->cr_uid == clp->lc_uid) return (0); else return (1); } else if (clp->lc_flags & LCL_GSS) return (1); /* * For AUTH_SYS, allow the same uid or root. (This is underspecified * in RFC3530, which talks about principals, but doesn't say anything * about uids for AUTH_SYS.) */ if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0) return (0); else return (1); } /* * Calculate the lease expiry time. */ static time_t nfsrv_leaseexpiry(void) { - if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC) + if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC) return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA)); return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA); } /* * Delay the delegation timeout as far as ls_delegtimelimit, as required. */ static void nfsrv_delaydelegtimeout(struct nfsstate *stp) { if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) return; if ((stp->ls_delegtime + 15) > NFSD_MONOSEC && stp->ls_delegtime < stp->ls_delegtimelimit) { stp->ls_delegtime += nfsrv_lease; if (stp->ls_delegtime > stp->ls_delegtimelimit) stp->ls_delegtime = stp->ls_delegtimelimit; } } /* * This function checks to see if there is any other state associated * with the openowner for this Open. * It returns 1 if there is no other state, 0 otherwise. */ static int nfsrv_nootherstate(struct nfsstate *stp) { struct nfsstate *tstp; LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) { if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock)) return (0); } return (1); } /* * Create a list of lock deltas (changes to local byte range locking * that can be rolled back using the list) and apply the changes via * nfsvno_advlock(). Optionally, lock the list. It is expected that either * the rollback or update function will be called after this. * It returns an error (and rolls back, as required), if any nfsvno_advlock() * call fails. If it returns an error, it will unlock the list. */ static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) { struct nfslock *lop, *nlop; int error = 0; /* Loop through the list of locks. */ lop = LIST_FIRST(&lfp->lf_locallock); while (first < end && lop != NULL) { nlop = LIST_NEXT(lop, lo_lckowner); if (first >= lop->lo_end) { /* not there yet */ lop = nlop; } else if (first < lop->lo_first) { /* new one starts before entry in list */ if (end <= lop->lo_first) { /* no overlap between old and new */ error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, end, cfp, p); if (error != 0) break; first = end; } else { /* handle fragment overlapped with new one */ error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, lop->lo_first, cfp, p); if (error != 0) break; first = lop->lo_first; } } else { /* new one overlaps this entry in list */ if (end <= lop->lo_end) { /* overlaps all of new one */ error = nfsrv_dolocal(vp, lfp, flags, lop->lo_flags, first, end, cfp, p); if (error != 0) break; first = end; } else { /* handle fragment overlapped with new one */ error = nfsrv_dolocal(vp, lfp, flags, lop->lo_flags, first, lop->lo_end, cfp, p); if (error != 0) break; first = lop->lo_end; lop = nlop; } } } if (first < end && error == 0) /* handle fragment past end of list */ error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, end, cfp, p); NFSEXITCODE(error); return (error); } /* * Local lock unlock. Unlock all byte ranges that are no longer locked * by NFSv4. To do this, unlock any subranges of first-->end that * do not overlap with the byte ranges of any lock in the lfp->lf_lock * list. This list has all locks for the file held by other * tuples. The list is ordered by increasing * lo_first value, but may have entries that overlap each other, for * the case of read locks. */ static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first, uint64_t init_end, NFSPROC_T *p) { struct nfslock *lop; uint64_t first, end, prevfirst __unused; first = init_first; end = init_end; while (first < init_end) { /* Loop through all nfs locks, adjusting first and end */ prevfirst = 0; LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { KASSERT(prevfirst <= lop->lo_first, ("nfsv4 locks out of order")); KASSERT(lop->lo_first < lop->lo_end, ("nfsv4 bogus lock")); prevfirst = lop->lo_first; if (first >= lop->lo_first && first < lop->lo_end) /* * Overlaps with initial part, so trim * off that initial part by moving first past * it. */ first = lop->lo_end; else if (end > lop->lo_first && lop->lo_first > first) { /* * This lock defines the end of the * segment to unlock, so set end to the * start of it and break out of the loop. */ end = lop->lo_first; break; } if (first >= end) /* * There is no segment left to do, so * break out of this loop and then exit * the outer while() since first will be set * to end, which must equal init_end here. */ break; } if (first < end) { /* Unlock this segment */ (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK, NFSLCK_READ, first, end, NULL, p); nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK, first, end); } /* * Now move past this segment and look for any further * segment in the range, if there is one. */ first = end; end = init_end; } } /* * Do the local lock operation and update the rollback list, as required. * Perform the rollback and return the error if nfsvno_advlock() fails. */ static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) { struct nfsrollback *rlp; int error = 0, ltype, oldltype; if (flags & NFSLCK_WRITE) ltype = F_WRLCK; else if (flags & NFSLCK_READ) ltype = F_RDLCK; else ltype = F_UNLCK; if (oldflags & NFSLCK_WRITE) oldltype = F_WRLCK; else if (oldflags & NFSLCK_READ) oldltype = F_RDLCK; else oldltype = F_UNLCK; if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK)) /* nothing to do */ goto out; error = nfsvno_advlock(vp, ltype, first, end, p); if (error != 0) { if (cfp != NULL) { cfp->cl_clientid.lval[0] = 0; cfp->cl_clientid.lval[1] = 0; cfp->cl_first = 0; cfp->cl_end = NFS64BITSSET; cfp->cl_flags = NFSLCK_WRITE; cfp->cl_ownerlen = 5; NFSBCOPY("LOCAL", cfp->cl_owner, 5); } nfsrv_locallock_rollback(vp, lfp, p); } else if (ltype != F_UNLCK) { rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK, M_WAITOK); rlp->rlck_first = first; rlp->rlck_end = end; rlp->rlck_type = oldltype; LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list); } out: NFSEXITCODE(error); return (error); } /* * Roll back local lock changes and free up the rollback list. */ static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p) { struct nfsrollback *rlp, *nrlp; LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) { (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first, rlp->rlck_end, p); free(rlp, M_NFSDROLLBACK); } LIST_INIT(&lfp->lf_rollback); } /* * Update local lock list and delete rollback list (ie now committed to the * local locks). Most of the work is done by the internal function. */ static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first, uint64_t end) { struct nfsrollback *rlp, *nrlp; struct nfslock *new_lop, *other_lop; new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); if (flags & (NFSLCK_READ | NFSLCK_WRITE)) other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); else other_lop = NULL; new_lop->lo_flags = flags; new_lop->lo_first = first; new_lop->lo_end = end; nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp); if (new_lop != NULL) free(new_lop, M_NFSDLOCK); if (other_lop != NULL) free(other_lop, M_NFSDLOCK); /* and get rid of the rollback list */ LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) free(rlp, M_NFSDROLLBACK); LIST_INIT(&lfp->lf_rollback); } /* * Lock the struct nfslockfile for local lock updating. */ static void nfsrv_locklf(struct nfslockfile *lfp) { int gotlock; /* lf_usecount ensures *lfp won't be free'd */ lfp->lf_usecount++; do { gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL, NFSSTATEMUTEXPTR, NULL); } while (gotlock == 0); lfp->lf_usecount--; } /* * Unlock the struct nfslockfile after local lock updating. */ static void nfsrv_unlocklf(struct nfslockfile *lfp) { nfsv4_unlock(&lfp->lf_locallock_lck, 0); } /* * Clear out all state for the NFSv4 server. * Must be called by a thread that can sleep when no nfsds are running. */ void nfsrv_throwawayallstate(NFSPROC_T *p) { struct nfsclient *clp, *nclp; struct nfslockfile *lfp, *nlfp; int i; /* * For each client, clean out the state and then free the structure. */ for (i = 0; i < nfsrv_clienthashsize; i++) { - LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { + LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash, + nclp) { nfsrv_cleanclient(clp, p); nfsrv_freedeleglist(&clp->lc_deleg); nfsrv_freedeleglist(&clp->lc_olddeleg); free(clp->lc_stateid, M_NFSDCLIENT); free(clp, M_NFSDCLIENT); } } /* * Also, free up any remaining lock file structures. */ for (i = 0; i < nfsrv_lockhashsize; i++) { - LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) { + LIST_FOREACH_SAFE(lfp, &NFSD_VNET(nfslockhash)[i], lf_hash, + nlfp) { printf("nfsd unload: fnd a lock file struct\n"); nfsrv_freenfslockfile(lfp); } } /* And get rid of the deviceid structures and layouts. */ nfsrv_freealllayoutsanddevids(); } /* * Check the sequence# for the session and slot provided as an argument. * Also, renew the lease if the session will return NFS_OK. */ int nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid, uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this, uint32_t *sflagsp, NFSPROC_T *p) { struct nfsdsession *sep; struct nfssessionhash *shp; int error; shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep == NULL) { NFSUNLOCKSESSION(shp); return (NFSERR_BADSESSION); } error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp, sep->sess_slots, NULL, NFSV4_SLOTS - 1); if (error != 0) { NFSUNLOCKSESSION(shp); return (error); } if (cache_this != 0) nd->nd_flag |= ND_SAVEREPLY; /* Renew the lease. */ sep->sess_clp->lc_expiry = nfsrv_leaseexpiry(); nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval; nd->nd_flag |= ND_IMPLIEDCLID; /* Save maximum request and reply sizes. */ nd->nd_maxreq = sep->sess_maxreq; nd->nd_maxresp = sep->sess_maxresp; *sflagsp = 0; if (sep->sess_clp->lc_req.nr_client == NULL || (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0) *sflagsp |= NFSV4SEQ_CBPATHDOWN; NFSUNLOCKSESSION(shp); if (error == NFSERR_EXPIRED) { *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED; error = 0; } else if (error == NFSERR_ADMINREVOKED) { *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED; error = 0; } *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1; return (0); } /* * Check/set reclaim complete for this session/clientid. */ int nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs) { struct nfsdsession *sep; struct nfssessionhash *shp; int error = 0; shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSTATE(); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep == NULL) { NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); return (NFSERR_BADSESSION); } if (onefs != 0) sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS; /* Check to see if reclaim complete has already happened. */ else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) error = NFSERR_COMPLETEALREADY; else { sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE; nfsrv_markreclaim(sep->sess_clp); } NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); return (error); } /* * Cache the reply in a session slot. */ void nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m) { struct nfsdsession *sep; struct nfssessionhash *shp; char *buf, *cp; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; #endif shp = NFSSESSIONHASH(nd->nd_sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(nd->nd_sessionid); if (sep == NULL) { NFSUNLOCKSESSION(shp); - if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { + if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & + NFSNSF_GRACEOVER) != 0) { buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK); switch (nd->nd_nam->sa_family) { #ifdef INET case AF_INET: sin = (struct sockaddr_in *)nd->nd_nam; cp = inet_ntop(sin->sin_family, &sin->sin_addr.s_addr, buf, INET6_ADDRSTRLEN); break; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)nd->nd_nam; cp = inet_ntop(sin6->sin6_family, &sin6->sin6_addr, buf, INET6_ADDRSTRLEN); break; #endif default: cp = NULL; } if (cp != NULL) printf("nfsrv_cache_session: no session " "IPaddr=%s, check NFS clients for unique " "/etc/hostid's\n", cp); else printf("nfsrv_cache_session: no session, " "check NFS clients for unique " "/etc/hostid's\n"); free(buf, M_TEMP); } m_freem(*m); return; } nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat, m); NFSUNLOCKSESSION(shp); } /* * Search for a session that matches the sessionid. */ static struct nfsdsession * nfsrv_findsession(uint8_t *sessionid) { struct nfsdsession *sep; struct nfssessionhash *shp; shp = NFSSESSIONHASH(sessionid); LIST_FOREACH(sep, &shp->list, sess_hash) { if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID)) break; } return (sep); } /* * Destroy a session. */ int nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid) { int error, igotlock, samesess; samesess = 0; if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) && (nd->nd_flag & ND_HASSEQUENCE) != 0) { samesess = 1; if ((nd->nd_flag & ND_LASTOP) == 0) return (NFSERR_BADSESSION); } /* Lock out other nfsd threads */ NFSLOCKV4ROOTMUTEX(); nfsv4_relref(&nfsv4rootfs_lock); do { igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); } while (igotlock == 0); NFSUNLOCKV4ROOTMUTEX(); error = nfsrv_freesession(NULL, sessionid); if (error == 0 && samesess != 0) nd->nd_flag &= ~ND_HASSEQUENCE; NFSLOCKV4ROOTMUTEX(); nfsv4_unlock(&nfsv4rootfs_lock, 1); NFSUNLOCKV4ROOTMUTEX(); return (error); } /* * Bind a connection to a session. * For now, only certain variants are supported, since the current session * structure can only handle a single backchannel entry, which will be * applied to all connections if it is set. */ int nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp) { struct nfssessionhash *shp; struct nfsdsession *sep; struct nfsclient *clp; SVCXPRT *savxprt; int error; error = 0; savxprt = NULL; shp = NFSSESSIONHASH(sessionid); NFSLOCKSTATE(); NFSLOCKSESSION(shp); sep = nfsrv_findsession(sessionid); if (sep != NULL) { clp = sep->sess_clp; if (*foreaftp == NFSCDFC4_BACK || *foreaftp == NFSCDFC4_BACK_OR_BOTH || *foreaftp == NFSCDFC4_FORE_OR_BOTH) { /* Try to set up a backchannel. */ if (clp->lc_req.nr_client == NULL) { NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire " "backchannel\n"); clp->lc_req.nr_client = (struct __rpc_client *) clnt_bck_create(nd->nd_xprt->xp_socket, sep->sess_cbprogram, NFSV4_CBVERS); } if (clp->lc_req.nr_client != NULL) { NFSD_DEBUG(2, "nfsrv_bindconnsess: set up " "backchannel\n"); savxprt = sep->sess_cbsess.nfsess_xprt; SVC_ACQUIRE(nd->nd_xprt); CLNT_ACQUIRE(clp->lc_req.nr_client); nd->nd_xprt->xp_p2 = clp->lc_req.nr_client; /* Disable idle timeout. */ nd->nd_xprt->xp_idletimeout = 0; sep->sess_cbsess.nfsess_xprt = nd->nd_xprt; sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN; clp->lc_flags |= LCL_DONEBINDCONN | LCL_NEEDSCBNULL; clp->lc_flags &= ~LCL_CBDOWN; if (*foreaftp == NFSCDFS4_BACK) *foreaftp = NFSCDFS4_BACK; else *foreaftp = NFSCDFS4_BOTH; } else if (*foreaftp != NFSCDFC4_BACK) { NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set " "up backchannel\n"); sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; clp->lc_flags |= LCL_DONEBINDCONN; *foreaftp = NFSCDFS4_FORE; } else { error = NFSERR_NOTSUPP; printf("nfsrv_bindconnsess: Can't add " "backchannel\n"); } } else { NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n"); clp->lc_flags |= LCL_DONEBINDCONN; *foreaftp = NFSCDFS4_FORE; } } else error = NFSERR_BADSESSION; NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); if (savxprt != NULL) SVC_RELEASE(savxprt); return (error); } /* * Free up a session structure. */ static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid) { struct nfssessionhash *shp; int i; NFSLOCKSTATE(); if (sep == NULL) { shp = NFSSESSIONHASH(sessionid); NFSLOCKSESSION(shp); sep = nfsrv_findsession(sessionid); } else { shp = NFSSESSIONHASH(sep->sess_sessionid); NFSLOCKSESSION(shp); } if (sep != NULL) { sep->sess_refcnt--; if (sep->sess_refcnt > 0) { NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); return (NFSERR_BACKCHANBUSY); } LIST_REMOVE(sep, sess_hash); LIST_REMOVE(sep, sess_list); } NFSUNLOCKSESSION(shp); NFSUNLOCKSTATE(); if (sep == NULL) return (NFSERR_BADSESSION); for (i = 0; i < NFSV4_SLOTS; i++) if (sep->sess_slots[i].nfssl_reply != NULL) m_freem(sep->sess_slots[i].nfssl_reply); if (sep->sess_cbsess.nfsess_xprt != NULL) SVC_RELEASE(sep->sess_cbsess.nfsess_xprt); free(sep, M_NFSDSESSION); return (0); } /* * Free a stateid. * RFC5661 says that it should fail when there are associated opens, locks * or delegations. Since stateids represent opens, I don't see how you can * free an open stateid (it will be free'd when closed), so this function * only works for lock stateids (freeing the lock_owner) or delegations. */ int nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; int error; NFSLOCKSTATE(); /* * Look up the stateid */ error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error == 0) { /* First, check for a delegation. */ LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, NFSX_STATEIDOTHER)) break; } if (stp != NULL) { nfsrv_freedeleg(stp); NFSUNLOCKSTATE(); return (error); } } /* Not a delegation, try for a lock_owner. */ if (error == 0) error = nfsrv_getstate(clp, stateidp, 0, &stp); if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0)) /* Not a lock_owner stateid. */ error = NFSERR_LOCKSHELD; if (error == 0 && !LIST_EMPTY(&stp->ls_lock)) error = NFSERR_LOCKSHELD; if (error == 0) nfsrv_freelockowner(stp, NULL, 0, p); NFSUNLOCKSTATE(); return (error); } /* * Test a stateid. */ int nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, NFSPROC_T *p) { struct nfsclient *clp; struct nfsstate *stp; int error; NFSLOCKSTATE(); /* * Look up the stateid */ error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error == 0) error = nfsrv_getstate(clp, stateidp, 0, &stp); if (error == 0 && stateidp->seqid != 0 && SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid)) error = NFSERR_OLDSTATEID; NFSUNLOCKSTATE(); return (error); } /* * Generate the xdr for an NFSv4.1 CBSequence Operation. */ static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, int dont_replycache, struct nfsdsession **sepp, int *slotposp) { struct nfsdsession *sep; uint32_t *tl, slotseq = 0; int maxslot; uint8_t sessionid[NFSX_V4SESSIONID]; int error; error = nfsv4_getcbsession(clp, sepp); if (error != 0) return (error); sep = *sepp; nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot, &slotseq, sessionid, true); KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot")); /* Build the Sequence arguments. */ NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED); bcopy(sessionid, tl, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; nd->nd_slotseq = tl; nd->nd_slotid = *slotposp; nd->nd_flag |= ND_HASSLOTID; *tl++ = txdr_unsigned(slotseq); *tl++ = txdr_unsigned(*slotposp); *tl++ = txdr_unsigned(maxslot); if (dont_replycache == 0) *tl++ = newnfs_true; else *tl++ = newnfs_false; *tl = 0; /* No referring call list, for now. */ nd->nd_flag |= ND_HASSEQUENCE; return (0); } /* * Get a session for the callback. */ static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp) { struct nfsdsession *sep; NFSLOCKSTATE(); LIST_FOREACH(sep, &clp->lc_session, sess_list) { if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) break; } if (sep == NULL) { NFSUNLOCKSTATE(); return (NFSERR_BADSESSION); } sep->sess_refcnt++; *sepp = sep; NFSUNLOCKSTATE(); return (0); } /* * Free up all backchannel xprts. This needs to be done when the nfsd threads * exit, since those transports will all be going away. * This is only called after all the nfsd threads are done performing RPCs, * so locking shouldn't be an issue. */ void nfsrv_freeallbackchannel_xprts(void) { struct nfsdsession *sep; struct nfsclient *clp; SVCXPRT *xprt; int i; for (i = 0; i < nfsrv_clienthashsize; i++) { - LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { + LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) { LIST_FOREACH(sep, &clp->lc_session, sess_list) { xprt = sep->sess_cbsess.nfsess_xprt; sep->sess_cbsess.nfsess_xprt = NULL; if (xprt != NULL) SVC_RELEASE(xprt); } } } } /* * Do a layout commit. Actually just call nfsrv_updatemdsattr(). * I have no idea if the rest of these arguments will ever be useful? */ int nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype, int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len, int hasnewmtime, struct timespec *newmtimep, int reclaim, nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep, uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr na; int error; error = nfsrv_updatemdsattr(vp, &na, p); if (error == 0) { *hasnewsizep = 1; *newsizep = na.na_size; } return (error); } /* * Try and get a layout. */ int nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp, int layouttype, int *iomode, uint64_t *offset, uint64_t *len, uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose, int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p) { struct nfslayouthash *lhyp; struct nfslayout *lyp; char *devid; fhandle_t fh, *dsfhp; int error, mirrorcnt; if (nfsrv_devidcnt == 0) return (NFSERR_UNKNLAYOUTTYPE); if (*offset != 0) printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset, (uintmax_t)*len); error = nfsvno_getfh(vp, &fh, p); NFSD_DEBUG(4, "layoutget getfh=%d\n", error); if (error != 0) return (error); /* * For now, all layouts are for entire files. * Only issue Read/Write layouts if requested for a non-readonly fs. */ if (NFSVNO_EXRDONLY(exp)) { if (*iomode == NFSLAYOUTIOMODE_RW) return (NFSERR_LAYOUTTRYLATER); *iomode = NFSLAYOUTIOMODE_READ; } if (*iomode != NFSLAYOUTIOMODE_RW) *iomode = NFSLAYOUTIOMODE_READ; /* * Check to see if a write layout can be issued for this file. * This is used during mirror recovery to avoid RW layouts being * issued for a file while it is being copied to the recovered * mirror. */ if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0) return (NFSERR_LAYOUTTRYLATER); *retonclose = 0; *offset = 0; *len = UINT64_MAX; /* First, see if a layout already exists and return if found. */ lhyp = NFSLAYOUTHASH(&fh); NFSLOCKLAYOUT(lhyp); error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp); NFSD_DEBUG(4, "layoutget findlay=%d\n", error); /* * Not sure if the seqid must be the same, so I won't check it. */ if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] || stateidp->other[1] != lyp->lay_stateid.other[1] || stateidp->other[2] != lyp->lay_stateid.other[2])) { if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(1, "ret bad stateid\n"); return (NFSERR_BADSTATEID); } /* * I believe we get here because there is a race between * the client processing the CBLAYOUTRECALL and the layout * being deleted here on the server. * The client has now done a LayoutGet with a non-layout * stateid, as it would when there is no layout. * As such, free this layout and set error == NFSERR_BADSTATEID * so the code below will create a new layout structure as * would happen if no layout was found. * "lyp" will be set before being used below, but set it NULL * as a safety belt. */ nfsrv_freelayout(&lhyp->list, lyp); lyp = NULL; error = NFSERR_BADSTATEID; } if (error == 0) { if (lyp->lay_layoutlen > maxcnt) { NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(1, "ret layout too small\n"); return (NFSERR_TOOSMALL); } if (*iomode == NFSLAYOUTIOMODE_RW) { if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) { NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(1, "ret layout nospace\n"); return (NFSERR_NOSPC); } lyp->lay_flags |= NFSLAY_RW; } else lyp->lay_flags |= NFSLAY_READ; NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); *layoutlenp = lyp->lay_layoutlen; if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; stateidp->seqid = lyp->lay_stateid.seqid; NFSUNLOCKLAYOUT(lhyp); NFSD_DEBUG(4, "ret fnd layout\n"); return (0); } NFSUNLOCKLAYOUT(lhyp); /* Find the device id and file handle. */ dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid); NFSD_DEBUG(4, "layoutget devandfh=%d\n", error); if (error == 0) { if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { if (NFSX_V4FILELAYOUT > maxcnt) error = NFSERR_TOOSMALL; else lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp, devid, vp->v_mount->mnt_stat.f_fsid); } else { if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt) error = NFSERR_TOOSMALL; else lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt, &fh, dsfhp, devid, vp->v_mount->mnt_stat.f_fsid); } } free(dsfhp, M_TEMP); free(devid, M_TEMP); if (error != 0) return (error); /* * Now, add this layout to the list. */ error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p); NFSD_DEBUG(4, "layoutget addl=%d\n", error); /* * The lyp will be set to NULL by nfsrv_addlayout() if it * linked the new structure into the lists. */ free(lyp, M_NFSDSTATE); return (error); } /* * Generate a File Layout. */ static struct nfslayout * nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs) { uint32_t *tl; struct nfslayout *lyp; uint64_t pattern_offset; lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE, M_WAITOK | M_ZERO); lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES; if (iomode == NFSLAYOUTIOMODE_RW) lyp->lay_flags = NFSLAY_RW; else lyp->lay_flags = NFSLAY_READ; NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); lyp->lay_clientid.qval = nd->nd_clientid.qval; lyp->lay_fsid = fs; NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID); /* Fill in the xdr for the files layout. */ tl = (uint32_t *)lyp->lay_xdr; NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); /* Set the stripe size to the maximum I/O size. */ *tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK); *tl++ = 0; /* 1st stripe index. */ pattern_offset = 0; txdr_hyper(pattern_offset, tl); tl += 2; /* Pattern offset. */ *tl++ = txdr_unsigned(1); /* 1 file handle. */ *tl++ = txdr_unsigned(NFSX_V4PNFSFH); NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); lyp->lay_layoutlen = NFSX_V4FILELAYOUT; return (lyp); } #define FLEX_OWNERID "999" #define FLEX_UID0 "0" /* * Generate a Flex File Layout. * The FLEX_OWNERID can be any string of 3 decimal digits. Although this * string goes on the wire, it isn't supposed to be used by the client, * since this server uses tight coupling. * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use * a string of "0". This works around the Linux Flex File Layout driver bug * which uses the synthetic uid/gid strings for the "tightly coupled" case. */ static struct nfslayout * nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs) { uint32_t *tl; struct nfslayout *lyp; uint64_t lenval; int i; lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt), M_NFSDSTATE, M_WAITOK | M_ZERO); lyp->lay_type = NFSLAYOUT_FLEXFILE; if (iomode == NFSLAYOUTIOMODE_RW) lyp->lay_flags = NFSLAY_RW; else lyp->lay_flags = NFSLAY_READ; NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); lyp->lay_clientid.qval = nd->nd_clientid.qval; lyp->lay_fsid = fs; lyp->lay_mirrorcnt = mirrorcnt; NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID); /* Fill in the xdr for the files layout. */ tl = (uint32_t *)lyp->lay_xdr; lenval = 0; txdr_hyper(lenval, tl); tl += 2; /* Stripe unit. */ *tl++ = txdr_unsigned(mirrorcnt); /* # of mirrors. */ for (i = 0; i < mirrorcnt; i++) { *tl++ = txdr_unsigned(1); /* One stripe. */ NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); devid += NFSX_V4DEVICEID; *tl++ = txdr_unsigned(1); /* Efficiency. */ *tl++ = 0; /* Proxy Stateid. */ *tl++ = 0x55555555; *tl++ = 0x55555555; *tl++ = 0x55555555; *tl++ = txdr_unsigned(1); /* 1 file handle. */ *tl++ = txdr_unsigned(NFSX_V4PNFSFH); NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED); dsfhp++; if (nfsrv_flexlinuxhack != 0) { *tl++ = txdr_unsigned(strlen(FLEX_UID0)); *tl = 0; /* 0 pad string. */ NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); *tl++ = txdr_unsigned(strlen(FLEX_UID0)); *tl = 0; /* 0 pad string. */ NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); } else { *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); } } *tl++ = txdr_unsigned(0); /* ff_flags. */ *tl = txdr_unsigned(60); /* Status interval hint. */ lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt); return (lyp); } /* * Parse and process Flex File errors returned via LayoutReturn. */ static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt, NFSPROC_T *p) { uint32_t *tl; int cnt, errcnt, i, j, opnum, stat; char devid[NFSX_V4DEVICEID]; tl = layp; maxcnt -= NFSX_UNSIGNED; if (maxcnt > 0) cnt = fxdr_unsigned(int, *tl++); else cnt = 0; NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt); for (i = 0; i < cnt; i++) { maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER + NFSX_UNSIGNED; if (maxcnt <= 0) break; /* Skip offset, length and stateid for now. */ tl += (4 + NFSX_STATEID / NFSX_UNSIGNED); errcnt = fxdr_unsigned(int, *tl++); NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt); for (j = 0; j < errcnt; j++) { maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED; if (maxcnt < 0) break; NFSBCOPY(tl, devid, NFSX_V4DEVICEID); tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); stat = fxdr_unsigned(int, *tl++); opnum = fxdr_unsigned(int, *tl++); NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum, stat); /* * Except for NFSERR_ACCES, NFSERR_STALE and * NFSERR_NOSPC errors, disable the mirror. */ if (stat != NFSERR_ACCES && stat != NFSERR_STALE && stat != NFSERR_NOSPC) nfsrv_delds(devid, p); /* For NFSERR_NOSPC, mark all devids and layouts. */ if (stat == NFSERR_NOSPC) nfsrv_marknospc(devid, true); } } } /* * This function removes all flex file layouts which has a mirror with * a device id that matches the argument. * Called when the DS represented by the device id has failed. */ void nfsrv_flexmirrordel(char *devid, NFSPROC_T *p) { uint32_t *tl; struct nfslayout *lyp, *nlyp; struct nfslayouthash *lhyp; struct nfslayouthead loclyp; int i, j; NFSD_DEBUG(4, "flexmirrordel\n"); /* Move all layouts found onto a local list. */ TAILQ_INIT(&loclyp); for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (lyp->lay_type == NFSLAYOUT_FLEXFILE && lyp->lay_mirrorcnt > 1) { NFSD_DEBUG(4, "possible match\n"); tl = lyp->lay_xdr; tl += 3; for (j = 0; j < lyp->lay_mirrorcnt; j++) { tl++; if (NFSBCMP(devid, tl, NFSX_V4DEVICEID) == 0) { /* Found one. */ NFSD_DEBUG(4, "fnd one\n"); TAILQ_REMOVE(&lhyp->list, lyp, lay_list); TAILQ_INSERT_HEAD(&loclyp, lyp, lay_list); break; } tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED + NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED + 11 * NFSX_UNSIGNED); } } } NFSUNLOCKLAYOUT(lhyp); } /* Now, try to do a Layout recall for each one found. */ TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) { NFSD_DEBUG(4, "do layout recall\n"); /* * The layout stateid.seqid needs to be incremented * before doing a LAYOUT_RECALL callback. */ if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, &lyp->lay_fh, lyp, 1, lyp->lay_type, p); nfsrv_freelayout(&loclyp, lyp); } } /* * Do a recall callback to the client for this layout. */ static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p) { struct nfsclient *clp; int error; NFSD_DEBUG(4, "nfsrv_recalllayout\n"); error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, NULL, p); NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error); if (error != 0) { printf("nfsrv_recalllayout: getclient err=%d\n", error); return (error); } if ((clp->lc_flags & LCL_NFSV41) != 0) { error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL, stateidp, changed, fhp, NULL, NULL, laytype, p); /* If lyp != NULL, handle an error return here. */ if (error != 0 && lyp != NULL) { NFSDRECALLLOCK(); /* * Mark it returned, since no layout recall * has been done. * All errors seem to be non-recoverable, although * NFSERR_NOMATCHLAYOUT is a normal event. */ if ((lyp->lay_flags & NFSLAY_RECALL) != 0) { lyp->lay_flags |= NFSLAY_RETURNED; wakeup(lyp); } NFSDRECALLUNLOCK(); if (error != NFSERR_NOMATCHLAYOUT) printf("nfsrv_recalllayout: err=%d\n", error); } } else printf("nfsrv_recalllayout: clp not NFSv4.1\n"); return (error); } /* * Find a layout to recall when we exceed our high water mark. */ void nfsrv_recalloldlayout(NFSPROC_T *p) { struct nfslayouthash *lhyp; struct nfslayout *lyp; nfsquad_t clientid; nfsv4stateid_t stateid; fhandle_t fh; int error, laytype = 0, ret; lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) { if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { lyp->lay_flags |= NFSLAY_CALLB; /* * The layout stateid.seqid needs to be incremented * before doing a LAYOUT_RECALL callback. */ if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; clientid = lyp->lay_clientid; stateid = lyp->lay_stateid; NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh)); laytype = lyp->lay_type; break; } } NFSUNLOCKLAYOUT(lhyp); if (lyp != NULL) { error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0, laytype, p); if (error != 0 && error != NFSERR_NOMATCHLAYOUT) NFSD_DEBUG(4, "recallold=%d\n", error); if (error != 0) { NFSLOCKLAYOUT(lhyp); /* * Since the hash list was unlocked, we need to * find it again. */ ret = nfsrv_findlayout(&clientid, &fh, laytype, p, &lyp); if (ret == 0 && (lyp->lay_flags & NFSLAY_CALLB) != 0 && lyp->lay_stateid.other[0] == stateid.other[0] && lyp->lay_stateid.other[1] == stateid.other[1] && lyp->lay_stateid.other[2] == stateid.other[2]) { /* * The client no longer knows this layout, so * it can be free'd now. */ if (error == NFSERR_NOMATCHLAYOUT) nfsrv_freelayout(&lhyp->list, lyp); else { /* * Leave it to be tried later by * clearing NFSLAY_CALLB and moving * it to the head of the list, so it * won't be tried again for a while. */ lyp->lay_flags &= ~NFSLAY_CALLB; TAILQ_REMOVE(&lhyp->list, lyp, lay_list); TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list); } } NFSUNLOCKLAYOUT(lhyp); } } } /* * Try and return layout(s). */ int nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp, int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim, int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp, struct ucred *cred, NFSPROC_T *p) { struct nfsvattr na; struct nfslayouthash *lhyp; struct nfslayout *lyp; fhandle_t fh; int error = 0; *fndp = 0; if (kind == NFSV4LAYOUTRET_FILE) { error = nfsvno_getfh(vp, &fh, p); if (error == 0) { error = nfsrv_updatemdsattr(vp, &na, p); if (error != 0) printf("nfsrv_layoutreturn: updatemdsattr" " failed=%d\n", error); } if (error == 0) { if (reclaim == newnfs_true) { error = nfsrv_checkgrace(NULL, NULL, NFSLCK_RECLAIM); if (error != NFSERR_NOGRACE) error = 0; return (error); } lhyp = NFSLAYOUTHASH(&fh); NFSDRECALLLOCK(); NFSLOCKLAYOUT(lhyp); error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp); NFSD_DEBUG(4, "layoutret findlay=%d\n", error); if (error == 0 && stateidp->other[0] == lyp->lay_stateid.other[0] && stateidp->other[1] == lyp->lay_stateid.other[1] && stateidp->other[2] == lyp->lay_stateid.other[2]) { NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d" " %x %x %x laystateid %d %x %x %x" " off=%ju len=%ju flgs=0x%x\n", stateidp->seqid, stateidp->other[0], stateidp->other[1], stateidp->other[2], lyp->lay_stateid.seqid, lyp->lay_stateid.other[0], lyp->lay_stateid.other[1], lyp->lay_stateid.other[2], (uintmax_t)offset, (uintmax_t)len, lyp->lay_flags); if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; stateidp->seqid = lyp->lay_stateid.seqid; if (offset == 0 && len == UINT64_MAX) { if ((iomode & NFSLAYOUTIOMODE_READ) != 0) lyp->lay_flags &= ~NFSLAY_READ; if ((iomode & NFSLAYOUTIOMODE_RW) != 0) lyp->lay_flags &= ~NFSLAY_RW; if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0) nfsrv_freelayout(&lhyp->list, lyp); else *fndp = 1; } else *fndp = 1; } NFSUNLOCKLAYOUT(lhyp); /* Search the nfsrv_recalllist for a match. */ TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && lyp->lay_clientid.qval == nd->nd_clientid.qval && stateidp->other[0] == lyp->lay_stateid.other[0] && stateidp->other[1] == lyp->lay_stateid.other[1] && stateidp->other[2] == lyp->lay_stateid.other[2]) { lyp->lay_flags |= NFSLAY_RETURNED; wakeup(lyp); error = 0; } } NFSDRECALLUNLOCK(); } if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL) nfsrv_flexlayouterr(nd, layp, maxcnt, p); } else if (kind == NFSV4LAYOUTRET_FSID) nfsrv_freelayouts(&nd->nd_clientid, &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode); else if (kind == NFSV4LAYOUTRET_ALL) nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode); else error = NFSERR_INVAL; if (error == -1) error = 0; return (error); } /* * Look for an existing layout. */ static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, NFSPROC_T *p, struct nfslayout **lypp) { struct nfslayouthash *lhyp; struct nfslayout *lyp; int ret; *lypp = NULL; ret = 0; lhyp = NFSLAYOUTHASH(fhp); TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 && lyp->lay_clientid.qval == clientidp->qval && lyp->lay_type == laytype) break; } if (lyp != NULL) *lypp = lyp; else ret = -1; return (ret); } /* * Add the new layout, as required. */ static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p) { struct nfsclient *clp; struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; fhandle_t *fhp; int error; KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0, ("nfsrv_layoutget: no nd_clientid\n")); lyp = *lypp; fhp = &lyp->lay_fh; NFSLOCKSTATE(); error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); if (error != 0) { NFSUNLOCKSTATE(); return (error); } lyp->lay_stateid.seqid = stateidp->seqid = 1; lyp->lay_stateid.other[0] = stateidp->other[0] = clp->lc_clientid.lval[0]; lyp->lay_stateid.other[1] = stateidp->other[1] = clp->lc_clientid.lval[1]; lyp->lay_stateid.other[2] = stateidp->other[2] = nfsrv_nextstateindex(clp); NFSUNLOCKSTATE(); lhyp = NFSLAYOUTHASH(fhp); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) { if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 && nlyp->lay_clientid.qval == nd->nd_clientid.qval) break; } if (nlyp != NULL) { /* A layout already exists, so use it. */ nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)); NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen); *layoutlenp = nlyp->lay_layoutlen; if (++nlyp->lay_stateid.seqid == 0) nlyp->lay_stateid.seqid = 1; stateidp->seqid = nlyp->lay_stateid.seqid; stateidp->other[0] = nlyp->lay_stateid.other[0]; stateidp->other[1] = nlyp->lay_stateid.other[1]; stateidp->other[2] = nlyp->lay_stateid.other[2]; NFSUNLOCKLAYOUT(lhyp); return (0); } /* Insert the new layout in the lists. */ *lypp = NULL; atomic_add_int(&nfsrv_layoutcnt, 1); - nfsstatsv1.srvlayouts++; + nfsstatsv1_p->srvlayouts++; NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); *layoutlenp = lyp->lay_layoutlen; TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list); NFSUNLOCKLAYOUT(lhyp); return (0); } /* * Get the devinfo for a deviceid. */ int nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt, uint32_t *notify, int *devaddrlen, char **devaddr) { struct nfsdevice *ds; if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype != NFSLAYOUT_FLEXFILE) || (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES)) return (NFSERR_UNKNLAYOUTTYPE); /* * Now, search for the device id. Note that the structures won't go * away, but the order changes in the list. As such, the lock only * needs to be held during the search through the list. */ NFSDDSLOCK(); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 && ds->nfsdev_nmp != NULL) break; } NFSDDSUNLOCK(); if (ds == NULL) return (NFSERR_NOENT); /* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */ *devaddrlen = 0; if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { *devaddrlen = ds->nfsdev_fileaddrlen; *devaddr = ds->nfsdev_fileaddr; } else if (layouttype == NFSLAYOUT_FLEXFILE) { *devaddrlen = ds->nfsdev_flexaddrlen; *devaddr = ds->nfsdev_flexaddr; } if (*devaddrlen == 0) return (NFSERR_UNKNLAYOUTTYPE); /* * The XDR overhead is 3 unsigned values: layout_type, * length_of_address and notify bitmap. * If the notify array is changed to not all zeros, the * count of unsigned values must be increased. */ if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED) { *maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED; return (NFSERR_TOOSMALL); } return (0); } /* * Free a list of layout state structures. */ static void nfsrv_freelayoutlist(nfsquad_t clientid) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; int i; for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (lyp->lay_clientid.qval == clientid.qval) nfsrv_freelayout(&lhyp->list, lyp); } NFSUNLOCKLAYOUT(lhyp); } } /* * Free up a layout. */ static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp) { NFSD_DEBUG(4, "Freelayout=%p\n", lyp); atomic_add_int(&nfsrv_layoutcnt, -1); - nfsstatsv1.srvlayouts--; + nfsstatsv1_p->srvlayouts--; TAILQ_REMOVE(lhp, lyp, lay_list); free(lyp, M_NFSDSTATE); } /* * Free up a device id. */ void nfsrv_freeonedevid(struct nfsdevice *ds) { int i; atomic_add_int(&nfsrv_devidcnt, -1); vrele(ds->nfsdev_dvp); for (i = 0; i < nfsrv_dsdirsize; i++) if (ds->nfsdev_dsdir[i] != NULL) vrele(ds->nfsdev_dsdir[i]); free(ds->nfsdev_fileaddr, M_NFSDSTATE); free(ds->nfsdev_flexaddr, M_NFSDSTATE); free(ds->nfsdev_host, M_NFSDSTATE); free(ds, M_NFSDSTATE); } /* * Free up a device id and its mirrors. */ static void nfsrv_freedevid(struct nfsdevice *ds) { TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); nfsrv_freeonedevid(ds); } /* * Free all layouts and device ids. * Done when the nfsd threads are shut down since there may be a new * modified device id list created when the nfsd is restarted. */ void nfsrv_freealllayoutsanddevids(void) { struct nfsdontlist *mrp, *nmrp; struct nfslayout *lyp, *nlyp; /* Get rid of the deviceid structures. */ nfsrv_freealldevids(); TAILQ_INIT(&nfsrv_devidhead); nfsrv_devidcnt = 0; /* Get rid of all layouts. */ nfsrv_freealllayouts(); /* Get rid of any nfsdontlist entries. */ LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp) free(mrp, M_NFSDSTATE); LIST_INIT(&nfsrv_dontlisthead); nfsrv_dontlistlen = 0; /* Free layouts in the recall list. */ TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp) nfsrv_freelayout(&nfsrv_recalllisthead, lyp); TAILQ_INIT(&nfsrv_recalllisthead); } /* * Free layouts that match the arguments. */ static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; int i; for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (clid->qval != lyp->lay_clientid.qval) continue; if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0) continue; if (laytype != lyp->lay_type) continue; if ((iomode & NFSLAYOUTIOMODE_READ) != 0) lyp->lay_flags &= ~NFSLAY_READ; if ((iomode & NFSLAYOUTIOMODE_RW) != 0) lyp->lay_flags &= ~NFSLAY_RW; if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0) nfsrv_freelayout(&lhyp->list, lyp); } NFSUNLOCKLAYOUT(lhyp); } } /* * Free all layouts for the argument file. */ void nfsrv_freefilelayouts(fhandle_t *fhp) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; lhyp = NFSLAYOUTHASH(fhp); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0) nfsrv_freelayout(&lhyp->list, lyp); } NFSUNLOCKLAYOUT(lhyp); } /* * Free all layouts. */ static void nfsrv_freealllayouts(void) { struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; int i; for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) nfsrv_freelayout(&lhyp->list, lyp); NFSUNLOCKLAYOUT(lhyp); } } /* * Look up the mount path for the DS server. */ static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, struct nfsdevice **dsp) { struct nameidata nd; struct nfsdevice *ds; struct mount *mp; int error, i; char *dsdirpath; size_t dsdirsize; NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp); *dsp = NULL; + if (jailed(p->td_ucred)) { + printf("A pNFS nfsd cannot run in a jail\n"); + return (EPERM); + } NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, dspathp, p); error = namei(&nd); NFSD_DEBUG(4, "lookup=%d\n", error); if (error != 0) return (error); if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); NFSD_DEBUG(4, "dspath not dir\n"); return (ENOTDIR); } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); NFSD_DEBUG(4, "dspath not an NFS mount\n"); return (ENXIO); } /* * Allocate a DS server structure with the NFS mounted directory * vnode reference counted, so that a non-forced dismount will * fail with EBUSY. * This structure is always linked into the list, even if an error * is being returned. The caller will free the entire list upon * an error return. */ *dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t), M_NFSDSTATE, M_WAITOK | M_ZERO); ds->nfsdev_dvp = nd.ni_vp; ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount); NFSVOPUNLOCK(nd.ni_vp); dsdirsize = strlen(dspathp) + 16; dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK); /* Now, create the DS directory structures. */ for (i = 0; i < nfsrv_dsdirsize; i++) { snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, dsdirpath, p); error = namei(&nd); NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error); if (error != 0) break; if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); error = ENOTDIR; NFSD_DEBUG(4, "dsdirpath not a VDIR\n"); break; } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); error = ENXIO; NFSD_DEBUG(4, "dsdirpath not an NFS mount\n"); break; } ds->nfsdev_dsdir[i] = nd.ni_vp; NFSVOPUNLOCK(nd.ni_vp); } free(dsdirpath, M_TEMP); if (strlen(mdspathp) > 0) { /* * This DS stores file for a specific MDS exported file * system. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, mdspathp, p); error = namei(&nd); NFSD_DEBUG(4, "mds lookup=%d\n", error); if (error != 0) goto out; if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); error = ENOTDIR; NFSD_DEBUG(4, "mdspath not dir\n"); goto out; } mp = nd.ni_vp->v_mount; if ((mp->mnt_flag & MNT_EXPORTED) == 0) { vput(nd.ni_vp); error = ENXIO; NFSD_DEBUG(4, "mdspath not an exported fs\n"); goto out; } ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid; ds->nfsdev_mdsisset = 1; vput(nd.ni_vp); } out: TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); atomic_add_int(&nfsrv_devidcnt, 1); return (error); } /* * Look up the mount path for the DS server and delete it. */ int nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p) { struct mount *mp; struct nfsmount *nmp; struct nfsdevice *ds; int error; NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp); /* * Search for the path in the mount list. Avoid looking the path * up, since this mount point may be hung, with associated locked * vnodes, etc. * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked * until this completes. * As noted in the man page, this should be done before any forced * dismount on the mount point, but at least the handshake on * NFSMNTP_CANCELRPCS should make it safe. */ error = 0; ds = NULL; nmp = NULL; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 && strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 && mp->mnt_data != NULL) { nmp = VFSTONFS(mp); NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); } else { NFSUNLOCKMNT(nmp); nmp = NULL; } break; } } mtx_unlock(&mountlist_mtx); if (nmp != NULL) { ds = nfsrv_deldsnmp(op, nmp, p); NFSD_DEBUG(4, "deldsnmp=%p\n", ds); if (ds != NULL) { nfsrv_killrpcs(nmp); NFSD_DEBUG(4, "aft killrpcs\n"); } else error = ENXIO; NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); NFSUNLOCKMNT(nmp); } else error = EINVAL; return (error); } /* * Search for and remove a DS entry which matches the "nmp" argument. * The nfsdevice structure pointer is returned so that the caller can * free it via nfsrv_freeonedevid(). * For the forced case, do not try to do LayoutRecalls, since the server * must be shut down now anyhow. */ struct nfsdevice * nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p) { struct nfsdevice *fndds; NFSD_DEBUG(4, "deldsdvp\n"); NFSDDSLOCK(); if (op == PNFSDOP_FORCEDELDS) fndds = nfsv4_findmirror(nmp); else fndds = nfsrv_findmirroredds(nmp); if (fndds != NULL) nfsrv_deleteds(fndds); NFSDDSUNLOCK(); if (fndds != NULL) { if (op != PNFSDOP_FORCEDELDS) nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); } return (fndds); } /* * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid. * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount * point. * Also, returns an error instead of the nfsdevice found. */ int nfsrv_delds(char *devid, NFSPROC_T *p) { struct nfsdevice *ds, *fndds; struct nfsmount *nmp; int fndmirror; NFSD_DEBUG(4, "delds\n"); /* * Search the DS server list for a match with devid. * Remove the DS entry if found and there is a mirror. */ fndds = NULL; nmp = NULL; fndmirror = 0; NFSDDSLOCK(); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 && ds->nfsdev_nmp != NULL) { NFSD_DEBUG(4, "fnd main ds\n"); fndds = ds; break; } } if (fndds == NULL) { NFSDDSUNLOCK(); return (ENXIO); } if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) fndmirror = 1; else if (fndds->nfsdev_mdsisset != 0) { /* For the fsid is set case, search for a mirror. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds != fndds && ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &fndds->nfsdev_mdsfsid) == 0) { fndmirror = 1; break; } } } if (fndmirror != 0) { nmp = fndds->nfsdev_nmp; NFSLOCKMNT(nmp); if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | NFSMNTP_CANCELRPCS)) == 0) { nmp->nm_privflag |= NFSMNTP_CANCELRPCS; NFSUNLOCKMNT(nmp); nfsrv_deleteds(fndds); } else { NFSUNLOCKMNT(nmp); nmp = NULL; } } NFSDDSUNLOCK(); if (nmp != NULL) { nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); nfsrv_killrpcs(nmp); NFSLOCKMNT(nmp); nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; wakeup(nmp); NFSUNLOCKMNT(nmp); return (0); } return (ENXIO); } /* * Mark a DS as disabled by setting nfsdev_nmp = NULL. */ static void nfsrv_deleteds(struct nfsdevice *fndds) { NFSD_DEBUG(4, "deleteds: deleting a mirror\n"); fndds->nfsdev_nmp = NULL; if (fndds->nfsdev_mdsisset == 0) nfsrv_faildscnt--; } /* * Fill in the addr structures for the File and Flex File layouts. */ static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost) { uint32_t *tl; char *netprot; int addrlen; static uint64_t new_devid = 0; if (strchr(addr, ':') != NULL) netprot = "tcp6"; else netprot = "tcp"; /* Fill in the device id. */ NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time)); new_devid++; NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)], sizeof(new_devid)); /* * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4 * as defined in RFC5661) in XDR. */ addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 6 * NFSX_UNSIGNED; NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot); ds->nfsdev_fileaddrlen = addrlen; tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); ds->nfsdev_fileaddr = (char *)tl; *tl++ = txdr_unsigned(1); /* One stripe with index 0. */ *tl++ = 0; *tl++ = txdr_unsigned(1); /* One multipath list */ *tl++ = txdr_unsigned(1); /* with one entry in it. */ /* The netaddr for this one entry. */ *tl++ = txdr_unsigned(strlen(netprot)); NFSBCOPY(netprot, tl, strlen(netprot)); tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); *tl++ = txdr_unsigned(strlen(addr)); NFSBCOPY(addr, tl, strlen(addr)); /* * Fill in the flex file addr (actually the ff_device_addr4 * as defined for Flexible File Layout) in XDR. */ addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 14 * NFSX_UNSIGNED; ds->nfsdev_flexaddrlen = addrlen; tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); ds->nfsdev_flexaddr = (char *)tl; *tl++ = txdr_unsigned(1); /* One multipath entry. */ /* The netaddr for this one entry. */ *tl++ = txdr_unsigned(strlen(netprot)); NFSBCOPY(netprot, tl, strlen(netprot)); tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); *tl++ = txdr_unsigned(strlen(addr)); NFSBCOPY(addr, tl, strlen(addr)); tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED); *tl++ = txdr_unsigned(2); /* Two NFS Versions. */ *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ *tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max rsize. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max wsize. */ *tl++ = newnfs_true; /* Tightly coupled. */ *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ *tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max rsize. */ *tl++ = txdr_unsigned(nfs_srvmaxio); /* DS max wsize. */ *tl = newnfs_true; /* Tightly coupled. */ ds->nfsdev_hostnamelen = strlen(dnshost); ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE, M_WAITOK); NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1); } /* * Create the device id list. * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument * is misconfigured. */ int nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p) { struct nfsdevice *ds; char *addrp, *dnshostp, *dspathp, *mdspathp; int error, i; addrp = args->addr; dnshostp = args->dnshost; dspathp = args->dspath; mdspathp = args->mdspath; nfsrv_maxpnfsmirror = args->mirrorcnt; if (addrp == NULL || dnshostp == NULL || dspathp == NULL || mdspathp == NULL) return (0); /* * Loop around for each nul-terminated string in args->addr, * args->dnshost, args->dnspath and args->mdspath. */ while (addrp < (args->addr + args->addrlen) && dnshostp < (args->dnshost + args->dnshostlen) && dspathp < (args->dspath + args->dspathlen) && mdspathp < (args->mdspath + args->mdspathlen)) { error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds); if (error != 0) { /* Free all DS servers. */ nfsrv_freealldevids(); nfsrv_devidcnt = 0; return (ENXIO); } nfsrv_allocdevid(ds, addrp, dnshostp); addrp += (strlen(addrp) + 1); dnshostp += (strlen(dnshostp) + 1); dspathp += (strlen(dspathp) + 1); mdspathp += (strlen(mdspathp) + 1); } if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) { /* Free all DS servers. */ nfsrv_freealldevids(); nfsrv_devidcnt = 0; nfsrv_maxpnfsmirror = 1; return (ENXIO); } /* We can fail at most one less DS than the mirror level. */ nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1; /* * Allocate the nfslayout hash table now, since this is a pNFS server. * Make it 1% of the high water mark and at least 100. */ if (nfslayouthash == NULL) { nfsrv_layouthashsize = nfsrv_layouthighwater / 100; if (nfsrv_layouthashsize < 100) nfsrv_layouthashsize = 100; nfslayouthash = mallocarray(nfsrv_layouthashsize, sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_layouthashsize; i++) { mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF); TAILQ_INIT(&nfslayouthash[i].list); } } return (0); } /* * Free all device ids. */ static void nfsrv_freealldevids(void) { struct nfsdevice *ds, *nds; TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds) nfsrv_freedevid(ds); } /* * Check to see if there is a Read/Write Layout plus either: * - A Write Delegation * or * - An Open with Write_access. * Return 1 if this is the case and 0 otherwise. * This function is used by nfsrv_proxyds() to decide if doing a Proxy * Getattr RPC to the Data Server (DS) is necessary. */ #define NFSCLIDVECSIZE 6 int nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p) { fhandle_t fh, *tfhp; struct nfsstate *stp; struct nfslayout *lyp; struct nfslayouthash *lhyp; struct nfslockhashhead *hp; struct nfslockfile *lfp; nfsquad_t clid[NFSCLIDVECSIZE]; int clidcnt, ret; ret = nfsvno_getfh(vp, &fh, p); if (ret != 0) return (0); /* First check for a Read/Write Layout. */ clidcnt = 0; lhyp = NFSLAYOUTHASH(&fh); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && ((lyp->lay_flags & NFSLAY_RW) != 0 || ((lyp->lay_flags & NFSLAY_READ) != 0 && nfsrv_pnfsatime != 0))) { if (clidcnt < NFSCLIDVECSIZE) clid[clidcnt].qval = lyp->lay_clientid.qval; clidcnt++; } } NFSUNLOCKLAYOUT(lhyp); if (clidcnt == 0) { /* None found, so return 0. */ return (0); } /* Get the nfslockfile for this fh. */ NFSLOCKSTATE(); hp = NFSLOCKHASH(&fh); LIST_FOREACH(lfp, hp, lf_hash) { tfhp = &lfp->lf_fh; if (NFSVNO_CMPFH(&fh, tfhp)) break; } if (lfp == NULL) { /* None found, so return 0. */ NFSUNLOCKSTATE(); return (0); } /* Now, look for a Write delegation for this clientid. */ LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 && nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) break; } if (stp != NULL) { /* Found one, so return 1. */ NFSUNLOCKSTATE(); return (1); } /* No Write delegation, so look for an Open with Write_access. */ LIST_FOREACH(stp, &lfp->lf_open, ls_file) { KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0, ("nfsrv_checkdsattr: Non-open in Open list\n")); if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 && nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) break; } NFSUNLOCKSTATE(); if (stp != NULL) return (1); return (0); } /* * Look for a matching clientid in the vector. Return 1 if one might match. */ static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt) { int i; /* If too many for the vector, return 1 since there might be a match. */ if (clidcnt > NFSCLIDVECSIZE) return (1); for (i = 0; i < clidcnt; i++) if (clidvec[i].qval == clid.qval) return (1); return (0); } /* * Check the don't list for "vp" and see if issuing an rw layout is allowed. * Return 1 if issuing an rw layout isn't allowed, 0 otherwise. */ static int nfsrv_dontlayout(fhandle_t *fhp) { struct nfsdontlist *mrp; int ret; if (nfsrv_dontlistlen == 0) return (0); ret = 0; NFSDDONTLISTLOCK(); LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 && (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) { ret = 1; break; } } NFSDDONTLISTUNLOCK(); return (ret); } #define PNFSDS_COPYSIZ 65536 /* * Create a new file on a DS and copy the contents of an extant DS file to it. * This can be used for recovery of a DS file onto a recovered DS. * The steps are: * - When called, the MDS file's vnode is locked, blocking LayoutGet operations. * - Disable issuing of read/write layouts for the file via the nfsdontlist, * so that they will be disabled after the MDS file's vnode is unlocked. * - Set up the nfsrv_recalllist so that recall of read/write layouts can * be done. * - Unlock the MDS file's vnode, so that the client(s) can perform proxied * writes, LayoutCommits and LayoutReturns for the file when completing the * LayoutReturn requested by the LayoutRecall callback. * - Issue a LayoutRecall callback for all read/write layouts and wait for * them to be returned. (If the LayoutRecall callback replies * NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.) * - Exclusively lock the MDS file's vnode. This ensures that no proxied * writes are in progress or can occur during the DS file copy. * It also blocks Setattr operations. * - Create the file on the recovered mirror. * - Copy the file from the operational DS. * - Copy any ACL from the MDS file to the new DS file. * - Set the modify time of the new DS file to that of the MDS file. * - Update the extended attribute for the MDS file. * - Enable issuing of rw layouts by deleting the nfsdontlist entry. * - The caller will unlock the MDS file's vnode allowing operations * to continue normally, since it is now on the mirror again. */ int nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds, struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt, struct ucred *cred, NFSPROC_T *p) { struct nfsdontlist *mrp, *nmrp; struct nfslayouthash *lhyp; struct nfslayout *lyp, *nlyp; struct nfslayouthead thl; struct mount *mp, *tvmp; struct acl *aclp; struct vattr va; struct timespec mtime; fhandle_t fh; vnode_t tvp; off_t rdpos, wrpos; ssize_t aresid; char *dat; int didprintf, ret, retacl, xfer; ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp"); ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp"); /* * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag * so that no more RW layouts will get issued. */ ret = nfsvno_getfh(vp, &fh, p); if (ret != 0) { NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret); return (ret); } nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK); nmrp->nfsmr_flags = NFSMR_DONTLAYOUT; NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh)); NFSDDONTLISTLOCK(); LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0) break; } if (mrp == NULL) { LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list); mrp = nmrp; nmrp = NULL; nfsrv_dontlistlen++; NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n"); } else { NFSDDONTLISTUNLOCK(); free(nmrp, M_NFSDSTATE); NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n"); return (ENXIO); } NFSDDONTLISTUNLOCK(); /* * Search for all RW layouts for this file. Move them to the * recall list, so they can be recalled and their return noted. */ lhyp = NFSLAYOUTHASH(&fh); NFSDRECALLLOCK(); NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && (lyp->lay_flags & NFSLAY_RW) != 0) { TAILQ_REMOVE(&lhyp->list, lyp, lay_list); TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list); lyp->lay_trycnt = 0; } } NFSUNLOCKLAYOUT(lhyp); NFSDRECALLUNLOCK(); ret = 0; mp = tvmp = NULL; didprintf = 0; TAILQ_INIT(&thl); /* Unlock the MDS vp, so that a LayoutReturn can be done on it. */ NFSVOPUNLOCK(vp); /* Now, do a recall for all layouts not yet recalled. */ tryagain: NFSDRECALLLOCK(); TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && (lyp->lay_flags & NFSLAY_RECALL) == 0) { lyp->lay_flags |= NFSLAY_RECALL; /* * The layout stateid.seqid needs to be incremented * before doing a LAYOUT_RECALL callback. */ if (++lyp->lay_stateid.seqid == 0) lyp->lay_stateid.seqid = 1; NFSDRECALLUNLOCK(); nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, &lyp->lay_fh, lyp, 0, lyp->lay_type, p); NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n"); goto tryagain; } } /* Now wait for them to be returned. */ tryagain2: TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) { if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) { TAILQ_REMOVE(&nfsrv_recalllisthead, lyp, lay_list); TAILQ_INSERT_HEAD(&thl, lyp, lay_list); NFSD_DEBUG(4, "nfsrv_copymr: layout returned\n"); } else { lyp->lay_trycnt++; ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR, PVFS | PCATCH, "nfsmrl", hz); NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n", ret); if (ret == EINTR || ret == ERESTART) break; if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) { /* * Give up after 60sec and return * ENXIO, failing the copymr. * This layout will remain on the * recalllist. It can only be cleared * by restarting the nfsd. * This seems the safe way to handle * it, since it cannot be safely copied * with an outstanding RW layout. */ if (lyp->lay_trycnt >= 60) { ret = ENXIO; break; } if (didprintf == 0) { printf("nfsrv_copymr: layout " "not returned\n"); didprintf = 1; } } } goto tryagain2; } } NFSDRECALLUNLOCK(); /* We can now get rid of the layouts that have been returned. */ TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp) nfsrv_freelayout(&thl, lyp); /* * Do the vn_start_write() calls here, before the MDS vnode is * locked and the tvp is created (locked) in the NFS file system * that dvp is in. * For tvmp, this probably isn't necessary, since it will be an * NFS mount and they are not suspendable at this time. */ if (ret == 0) ret = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (ret == 0) { tvmp = dvp->v_mount; ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH); } /* * LK_EXCLUSIVE lock the MDS vnode, so that any * proxied writes through the MDS will be blocked until we have * completed the copy and update of the extended attributes. * This will also ensure that any attributes and ACL will not be * changed until the copy is complete. */ NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (ret == 0 && VN_IS_DOOMED(vp)) { NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n"); ret = ESTALE; } /* Create the data file on the recovered DS. */ if (ret == 0) ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp); /* Copy the DS file, if created successfully. */ if (ret == 0) { /* * Get any NFSv4 ACL on the MDS file, so that it can be set * on the new DS file. */ aclp = acl_alloc(M_WAITOK | M_ZERO); retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); if (retacl != 0 && retacl != ENOATTR) NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl); dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK); /* Malloc a block of 0s used to check for holes. */ if (nfsrv_zeropnfsdat == NULL) nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK | M_ZERO); rdpos = wrpos = 0; ret = VOP_GETATTR(fvp, &va, cred); aresid = 0; while (ret == 0 && aresid == 0) { ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ, rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, &aresid, p); xfer = PNFSDS_COPYSIZ - aresid; if (ret == 0 && xfer > 0) { rdpos += xfer; /* * Skip the write for holes, except for the * last block. */ if (xfer < PNFSDS_COPYSIZ || rdpos == va.va_size || NFSBCMP(dat, nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0) ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer, wrpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, NULL, p); if (ret == 0) wrpos += xfer; } } /* If there is an ACL and the copy succeeded, set the ACL. */ if (ret == 0 && retacl == 0) { ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p); /* * Don't consider these as errors, since VOP_GETACL() * can return an ACL when they are not actually * supported. For example, for UFS, VOP_GETACL() * will return a trivial ACL based on the uid/gid/mode * when there is no ACL on the file. * This case should be recognized as a trivial ACL * by UFS's VOP_SETACL() and succeed, but... */ if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM) ret = 0; } if (ret == 0) ret = VOP_FSYNC(tvp, MNT_WAIT, p); /* Set the DS data file's modify time that of the MDS file. */ if (ret == 0) ret = VOP_GETATTR(vp, &va, cred); if (ret == 0) { mtime = va.va_mtime; VATTR_NULL(&va); va.va_mtime = mtime; ret = VOP_SETATTR(tvp, &va, cred); } vput(tvp); acl_free(aclp); free(dat, M_TEMP); } if (tvmp != NULL) vn_finished_write(tvmp); /* Update the extended attributes for the newly created DS file. */ if (ret == 0) ret = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*wpf) * mirrorcnt, (char *)wpf, p); if (mp != NULL) vn_finished_write(mp); /* Get rid of the dontlist entry, so that Layouts can be issued. */ NFSDDONTLISTLOCK(); LIST_REMOVE(mrp, nfsmr_list); NFSDDONTLISTUNLOCK(); free(mrp, M_NFSDSTATE); return (ret); } /* * Create a data storage file on the recovered DS. */ static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, vnode_t *tvpp) { struct vattr va, nva; int error; /* Make data file name based on FH. */ error = VOP_GETATTR(vp, &va, cred); if (error == 0) { /* Set the attributes for "vp" to Setattr the DS vp. */ VATTR_NULL(&nva); nva.va_uid = va.va_uid; nva.va_gid = va.va_gid; nva.va_mode = va.va_mode; nva.va_size = 0; VATTR_NULL(&va); va.va_type = VREG; va.va_mode = nva.va_mode; NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf); error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL, pf->dsf_filename, cred, p, tvpp); } return (error); } /* * Look up the MDS file shared locked, and then get the extended attribute * to find the extant DS file to be copied to the new mirror. * If successful, *vpp is set to the MDS file's vp and *nvpp is * set to a DS data file for the MDS file, both exclusively locked. * The "buf" argument has the pnfsdsfile structure from the MDS file * in it and buflen is set to its length. */ int nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf, int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp, struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp, struct nfsdevice **fdsp) { struct nameidata nd; struct vnode *vp, *curvp; struct pnfsdsfile *pf; struct nfsmount *nmp, *curnmp; int dsdir, error, mirrorcnt, ippos; vp = NULL; curvp = NULL; curnmp = NULL; *dsp = NULL; *fdsp = NULL; if (dspathp == NULL && curdspathp != NULL) return (EPERM); /* * Look up the MDS file shared locked. The lock will be upgraded * to an exclusive lock after any rw layouts have been returned. */ NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, mdspathp, p); error = namei(&nd); NFSD_DEBUG(4, "lookup=%d\n", error); if (error != 0) return (error); if (nd.ni_vp->v_type != VREG) { vput(nd.ni_vp); NFSD_DEBUG(4, "mdspath not reg\n"); return (EISDIR); } vp = nd.ni_vp; if (curdspathp != NULL) { /* * Look up the current DS path and find the nfsdev structure for * it. */ NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, curdspathp, p); error = namei(&nd); NFSD_DEBUG(4, "ds lookup=%d\n", error); if (error != 0) { vput(vp); return (error); } if (nd.ni_vp->v_type != VDIR) { vput(nd.ni_vp); vput(vp); NFSD_DEBUG(4, "curdspath not dir\n"); return (ENOTDIR); } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); vput(vp); NFSD_DEBUG(4, "curdspath not an NFS mount\n"); return (ENXIO); } curnmp = VFSTONFS(nd.ni_vp->v_mount); /* Search the nfsdev list for a match. */ NFSDDSLOCK(); *fdsp = nfsv4_findmirror(curnmp); NFSDDSUNLOCK(); if (*fdsp == NULL) curnmp = NULL; if (curnmp == NULL) { vput(nd.ni_vp); vput(vp); NFSD_DEBUG(4, "mdscopymr: no current ds\n"); return (ENXIO); } curvp = nd.ni_vp; } if (dspathp != NULL) { /* Look up the nfsdev path and find the nfsdev structure. */ NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, dspathp, p); error = namei(&nd); NFSD_DEBUG(4, "ds lookup=%d\n", error); if (error != 0) { vput(vp); if (curvp != NULL) vput(curvp); return (error); } if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) { vput(nd.ni_vp); vput(vp); if (curvp != NULL) vput(curvp); NFSD_DEBUG(4, "dspath not dir\n"); if (nd.ni_vp == curvp) return (EPERM); return (ENOTDIR); } if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); vput(vp); if (curvp != NULL) vput(curvp); NFSD_DEBUG(4, "dspath not an NFS mount\n"); return (ENXIO); } nmp = VFSTONFS(nd.ni_vp->v_mount); /* * Search the nfsdevice list for a match. If curnmp == NULL, * this is a recovery and there must be a mirror. */ NFSDDSLOCK(); if (curnmp == NULL) *dsp = nfsrv_findmirroredds(nmp); else *dsp = nfsv4_findmirror(nmp); NFSDDSUNLOCK(); if (*dsp == NULL) { vput(nd.ni_vp); vput(vp); if (curvp != NULL) vput(curvp); NFSD_DEBUG(4, "mdscopymr: no ds\n"); return (ENXIO); } } else { nd.ni_vp = NULL; nmp = NULL; } /* * Get a vp for an available DS data file using the extended * attribute on the MDS file. * If there is a valid entry for the new DS in the extended attribute * on the MDS file (as checked via the nmp argument), * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur. */ error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p, NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir); if (curvp != NULL) vput(curvp); if (nd.ni_vp == NULL) { if (error == 0 && nmp != NULL) { /* Search the nfsdev list for a match. */ NFSDDSLOCK(); *dsp = nfsrv_findmirroredds(nmp); NFSDDSUNLOCK(); } if (error == 0 && (nmp == NULL || *dsp == NULL)) { if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } error = ENXIO; } } else vput(nd.ni_vp); /* * When dspathp != NULL and curdspathp == NULL, this is a recovery * and is only allowed if there is a 0.0.0.0 IP address entry. * When curdspathp != NULL, the ippos will be set to that entry. */ if (error == 0 && dspathp != NULL && ippos == -1) { if (nvpp != NULL && *nvpp != NULL) { vput(*nvpp); *nvpp = NULL; } error = ENXIO; } if (error == 0) { *vpp = vp; pf = (struct pnfsdsfile *)buf; if (ippos == -1) { /* If no zeroip pnfsdsfile, add one. */ ippos = *buflenp / sizeof(*pf); *buflenp += sizeof(*pf); pf += ippos; pf->dsf_dir = dsdir; strlcpy(pf->dsf_filename, fname, sizeof(pf->dsf_filename)); } else pf += ippos; *pfp = pf; } else vput(vp); return (error); } /* * Search for a matching pnfsd mirror device structure, base on the nmp arg. * Return one if found, NULL otherwise. */ static struct nfsdevice * nfsrv_findmirroredds(struct nfsmount *nmp) { struct nfsdevice *ds, *fndds; int fndmirror; mtx_assert(NFSDDSMUTEXPTR, MA_OWNED); /* * Search the DS server list for a match with nmp. * Remove the DS entry if found and there is a mirror. */ fndds = NULL; fndmirror = 0; if (nfsrv_devidcnt == 0) return (fndds); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds->nfsdev_nmp == nmp) { NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n"); fndds = ds; break; } } if (fndds == NULL) return (fndds); if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) fndmirror = 1; else if (fndds->nfsdev_mdsisset != 0) { /* For the fsid is set case, search for a mirror. */ TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (ds != fndds && ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && fsidcmp(&ds->nfsdev_mdsfsid, &fndds->nfsdev_mdsfsid) == 0) { fndmirror = 1; break; } } } if (fndmirror == 0) { NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n"); return (NULL); } return (fndds); } /* * Mark the appropriate devid and all associated layout as "out of space". */ void nfsrv_marknospc(char *devid, bool setit) { struct nfsdevice *ds; struct nfslayout *lyp; struct nfslayouthash *lhyp; int i; NFSDDSLOCK(); TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) { NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit); ds->nfsdev_nospc = setit; } } NFSDDSUNLOCK(); for (i = 0; i < nfsrv_layouthashsize; i++) { lhyp = &nfslayouthash[i]; NFSLOCKLAYOUT(lhyp); TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { if (NFSBCMP(lyp->lay_deviceid, devid, NFSX_V4DEVICEID) == 0) { NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n", setit); if (setit) lyp->lay_flags |= NFSLAY_NOSPC; else lyp->lay_flags &= ~NFSLAY_NOSPC; } } NFSUNLOCKLAYOUT(lhyp); } } diff --git a/sys/fs/nfsserver/nfs_nfsdsubs.c b/sys/fs/nfsserver/nfs_nfsdsubs.c index 20f62211f53e..fa5747c9c741 100644 --- a/sys/fs/nfsserver/nfs_nfsdsubs.c +++ b/sys/fs/nfsserver/nfs_nfsdsubs.c @@ -1,2204 +1,2209 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); /* * These functions support the macros and help fiddle mbuf chains for * the nfs op functions. They do things like create the rpc header and * copy data between mbuf chains and uio lists. */ #include extern u_int32_t newnfs_true, newnfs_false; -extern int nfs_rootfhset; extern int nfs_pubfhset; -extern struct nfsclienthashhead *nfsclienthash; extern int nfsrv_clienthashsize; -extern struct nfslockhashhead *nfslockhash; extern int nfsrv_lockhashsize; -extern struct nfssessionhash *nfssessionhash; extern int nfsrv_sessionhashsize; extern int nfsrv_useacl; extern uid_t nfsrv_defaultuid; extern gid_t nfsrv_defaultgid; +NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); +NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash); +NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash); +NFSD_VNET_DECLARE(int, nfs_rootfhset); +NFSD_VNET_DECLARE(uid_t, nfsrv_defaultuid); +NFSD_VNET_DECLARE(gid_t, nfsrv_defaultgid); + +NFSD_VNET_DEFINE(struct nfsdontlisthead, nfsrv_dontlisthead); + + char nfs_v2pubfh[NFSX_V2FH]; struct nfsdontlisthead nfsrv_dontlisthead; struct nfslayouthead nfsrv_recalllisthead; static nfstype newnfsv2_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON }; extern nfstype nfsv34_type[9]; static u_int32_t nfsrv_isannfserr(u_int32_t); SYSCTL_DECL(_vfs_nfsd); static int enable_checkutf8 = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_checkutf8, CTLFLAG_RW, &enable_checkutf8, 0, "Enable the NFSv4 check for the UTF8 compliant name required by rfc3530"); static int enable_nobodycheck = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_nobodycheck, CTLFLAG_RW, &enable_nobodycheck, 0, "Enable the NFSv4 check when setting user nobody as owner"); static int enable_nogroupcheck = 1; SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_nogroupcheck, CTLFLAG_RW, &enable_nogroupcheck, 0, "Enable the NFSv4 check when setting group nogroup as owner"); static char nfsrv_hexdigit(char, int *); /* * Maps errno values to nfs error numbers. * Use NFSERR_IO as the catch all for ones not specifically defined in * RFC 1094. (It now includes the errors added for NFSv3.) */ static u_char nfsrv_v2errmap[NFSERR_REMOTE] = { NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_EXIST, NFSERR_XDEV, NFSERR_NODEV, NFSERR_NOTDIR, NFSERR_ISDIR, NFSERR_INVAL, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, NFSERR_MLINK, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, NFSERR_REMOTE, }; /* * Maps errno values to nfs error numbers. * Although it is not obvious whether or not NFS clients really care if * a returned error value is in the specified list for the procedure, the * safest thing to do is filter them appropriately. For Version 2, the * X/Open XNFS document is the only specification that defines error values * for each RPC (The RFC simply lists all possible error values for all RPCs), * so I have decided to not do this for Version 2. * The first entry is the default error return and the rest are the valid * errors for that RPC in increasing numeric order. */ static short nfsv3err_null[] = { 0, 0, }; static short nfsv3err_getattr[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_setattr[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_PERM, NFSERR_IO, NFSERR_INVAL, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOT_SYNC, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_lookup[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_ACCES, NFSERR_NAMETOL, NFSERR_IO, NFSERR_NOTDIR, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_access[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_readlink[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_INVAL, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_read[] = { NFSERR_IO, NFSERR_IO, NFSERR_NXIO, NFSERR_ACCES, NFSERR_INVAL, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_write[] = { NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_NOSPC, NFSERR_INVAL, NFSERR_FBIG, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_create[] = { NFSERR_IO, NFSERR_EXIST, NFSERR_NAMETOL, NFSERR_ACCES, NFSERR_IO, NFSERR_NOTDIR, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_mkdir[] = { NFSERR_IO, NFSERR_EXIST, NFSERR_ACCES, NFSERR_NAMETOL, NFSERR_IO, NFSERR_NOTDIR, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_symlink[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NAMETOL, NFSERR_NOSPC, NFSERR_IO, NFSERR_NOTDIR, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_mknod[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NAMETOL, NFSERR_NOSPC, NFSERR_IO, NFSERR_NOTDIR, NFSERR_ROFS, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, NFSERR_BADTYPE, 0, }; static short nfsv3err_remove[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_ACCES, NFSERR_NAMETOL, NFSERR_IO, NFSERR_NOTDIR, NFSERR_ROFS, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_rmdir[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_NAMETOL, NFSERR_IO, NFSERR_EXIST, NFSERR_INVAL, NFSERR_ROFS, NFSERR_NOTEMPTY, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_rename[] = { NFSERR_IO, NFSERR_NOENT, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NAMETOL, NFSERR_XDEV, NFSERR_IO, NFSERR_NOTDIR, NFSERR_ISDIR, NFSERR_INVAL, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_MLINK, NFSERR_NOTEMPTY, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_link[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_EXIST, NFSERR_NAMETOL, NFSERR_IO, NFSERR_XDEV, NFSERR_NOTDIR, NFSERR_INVAL, NFSERR_NOSPC, NFSERR_ROFS, NFSERR_MLINK, NFSERR_DQUOT, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_NOTSUPP, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_readdir[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_BAD_COOKIE, NFSERR_TOOSMALL, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_readdirplus[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_NOTDIR, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_BAD_COOKIE, NFSERR_NOTSUPP, NFSERR_TOOSMALL, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_fsstat[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_fsinfo[] = { NFSERR_STALE, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_pathconf[] = { NFSERR_STALE, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short nfsv3err_commit[] = { NFSERR_IO, NFSERR_IO, NFSERR_STALE, NFSERR_BADHANDLE, NFSERR_SERVERFAULT, NFSERR_DELAY, 0, }; static short *nfsrv_v3errmap[] = { nfsv3err_null, nfsv3err_getattr, nfsv3err_setattr, nfsv3err_lookup, nfsv3err_access, nfsv3err_readlink, nfsv3err_read, nfsv3err_write, nfsv3err_create, nfsv3err_mkdir, nfsv3err_symlink, nfsv3err_mknod, nfsv3err_remove, nfsv3err_rmdir, nfsv3err_rename, nfsv3err_link, nfsv3err_readdir, nfsv3err_readdirplus, nfsv3err_fsstat, nfsv3err_fsinfo, nfsv3err_pathconf, nfsv3err_commit, }; /* * And the same for V4. */ static short nfsv4err_null[] = { 0, 0, }; static short nfsv4err_access[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_close[] = { NFSERR_EXPIRED, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADSEQID, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_LOCKSHELD, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_OLDSTATEID, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_commit[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_ISDIR, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_create[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_ATTRNOTSUPP, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADOWNER, NFSERR_BADTYPE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_EXIST, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOFILEHANDLE, NFSERR_NOSPC, NFSERR_NOTDIR, NFSERR_PERM, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_delegpurge[] = { NFSERR_SERVERFAULT, NFSERR_BADXDR, NFSERR_NOTSUPP, NFSERR_LEASEMOVED, NFSERR_MOVED, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALECLIENTID, 0, }; static short nfsv4err_delegreturn[] = { NFSERR_SERVERFAULT, NFSERR_ADMINREVOKED, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_EXPIRED, NFSERR_INVAL, NFSERR_LEASEMOVED, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOTSUPP, NFSERR_OLDSTATEID, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_getattr[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_getfh[] = { NFSERR_BADHANDLE, NFSERR_BADHANDLE, NFSERR_FHEXPIRED, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_link[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_EXIST, NFSERR_FHEXPIRED, NFSERR_FILEOPEN, NFSERR_INVAL, NFSERR_IO, NFSERR_ISDIR, NFSERR_MLINK, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOSPC, NFSERR_NOTDIR, NFSERR_NOTSUPP, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_WRONGSEC, NFSERR_XDEV, 0, }; static short nfsv4err_lock[] = { NFSERR_SERVERFAULT, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADRANGE, NFSERR_BADSEQID, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_DEADLOCK, NFSERR_DELAY, NFSERR_DENIED, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_LOCKNOTSUPP, NFSERR_LOCKRANGE, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOGRACE, NFSERR_OLDSTATEID, NFSERR_OPENMODE, NFSERR_RECLAIMBAD, NFSERR_RECLAIMCONFLICT, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALECLIENTID, NFSERR_STALESTATEID, 0, }; static short nfsv4err_lockt[] = { NFSERR_SERVERFAULT, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_BADRANGE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DENIED, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_LOCKRANGE, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALECLIENTID, 0, }; static short nfsv4err_locku[] = { NFSERR_SERVERFAULT, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADRANGE, NFSERR_BADSEQID, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_LOCKRANGE, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_OLDSTATEID, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_lookup[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADXDR, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOTDIR, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_SYMLINK, NFSERR_WRONGSEC, 0, }; static short nfsv4err_lookupp[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_FHEXPIRED, NFSERR_IO, NFSERR_MOVED, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOTDIR, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_nverify[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_ATTRNOTSUPP, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_SAME, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_open[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_ATTRNOTSUPP, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADOWNER, NFSERR_BADSEQID, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_EXIST, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_IO, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOGRACE, NFSERR_NOSPC, NFSERR_NOTDIR, NFSERR_NOTSUPP, NFSERR_PERM, NFSERR_RECLAIMBAD, NFSERR_RECLAIMCONFLICT, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_SHAREDENIED, NFSERR_STALE, NFSERR_STALECLIENTID, NFSERR_SYMLINK, NFSERR_WRONGSEC, 0, }; static short nfsv4err_openattr[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_FHEXPIRED, NFSERR_IO, NFSERR_MOVED, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOSPC, NFSERR_NOTSUPP, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_openconfirm[] = { NFSERR_SERVERFAULT, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADSEQID, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_OLDSTATEID, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_opendowngrade[] = { NFSERR_SERVERFAULT, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADSEQID, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_OLDSTATEID, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_putfh[] = { NFSERR_SERVERFAULT, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_FHEXPIRED, NFSERR_MOVED, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_WRONGSEC, 0, }; static short nfsv4err_putpubfh[] = { NFSERR_SERVERFAULT, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_WRONGSEC, 0, }; static short nfsv4err_putrootfh[] = { NFSERR_SERVERFAULT, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_WRONGSEC, 0, }; static short nfsv4err_read[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_EXPIRED, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_IO, NFSERR_INVAL, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_LOCKED, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NXIO, NFSERR_OLDSTATEID, NFSERR_OPENMODE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_readdir[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_BAD_COOKIE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOTDIR, NFSERR_NOTSAME, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_TOOSMALL, 0, }; static short nfsv4err_readlink[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADHANDLE, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_IO, NFSERR_ISDIR, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOTSUPP, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_remove[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_FILEOPEN, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOTDIR, NFSERR_NOTEMPTY, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_rename[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_EXIST, NFSERR_FHEXPIRED, NFSERR_FILEOPEN, NFSERR_INVAL, NFSERR_IO, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOSPC, NFSERR_NOTDIR, NFSERR_NOTEMPTY, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_WRONGSEC, NFSERR_XDEV, 0, }; static short nfsv4err_renew[] = { NFSERR_SERVERFAULT, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_BADXDR, NFSERR_CBPATHDOWN, NFSERR_EXPIRED, NFSERR_LEASEMOVED, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALECLIENTID, 0, }; static short nfsv4err_restorefh[] = { NFSERR_SERVERFAULT, NFSERR_BADHANDLE, NFSERR_FHEXPIRED, NFSERR_MOVED, NFSERR_RESOURCE, NFSERR_RESTOREFH, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_WRONGSEC, 0, }; static short nfsv4err_savefh[] = { NFSERR_SERVERFAULT, NFSERR_BADHANDLE, NFSERR_FHEXPIRED, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_secinfo[] = { NFSERR_SERVERFAULT, NFSERR_ACCES, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADNAME, NFSERR_BADXDR, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_MOVED, NFSERR_NAMETOL, NFSERR_NOENT, NFSERR_NOFILEHANDLE, NFSERR_NOTDIR, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_setattr[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_ATTRNOTSUPP, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADOWNER, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_EXPIRED, NFSERR_FBIG, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_INVAL, NFSERR_IO, NFSERR_ISDIR, NFSERR_LOCKED, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOSPC, NFSERR_OLDSTATEID, NFSERR_OPENMODE, NFSERR_PERM, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_setclientid[] = { NFSERR_SERVERFAULT, NFSERR_BADXDR, NFSERR_CLIDINUSE, NFSERR_INVAL, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_WRONGSEC, 0, }; static short nfsv4err_setclientidconfirm[] = { NFSERR_SERVERFAULT, NFSERR_BADXDR, NFSERR_CLIDINUSE, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALECLIENTID, 0, }; static short nfsv4err_verify[] = { NFSERR_SERVERFAULT, NFSERR_ACCES, NFSERR_ATTRNOTSUPP, NFSERR_BADCHAR, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_FHEXPIRED, NFSERR_INVAL, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOTSAME, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALE, 0, }; static short nfsv4err_write[] = { NFSERR_IO, NFSERR_ACCES, NFSERR_ADMINREVOKED, NFSERR_BADHANDLE, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_DELAY, NFSERR_DQUOT, NFSERR_EXPIRED, NFSERR_FBIG, NFSERR_FHEXPIRED, NFSERR_GRACE, NFSERR_INVAL, NFSERR_IO, NFSERR_ISDIR, NFSERR_LEASEMOVED, NFSERR_LOCKED, NFSERR_MOVED, NFSERR_NOFILEHANDLE, NFSERR_NOSPC, NFSERR_NXIO, NFSERR_OLDSTATEID, NFSERR_OPENMODE, NFSERR_RESOURCE, NFSERR_ROFS, NFSERR_SERVERFAULT, NFSERR_STALE, NFSERR_STALESTATEID, 0, }; static short nfsv4err_releaselockowner[] = { NFSERR_SERVERFAULT, NFSERR_ADMINREVOKED, NFSERR_BADXDR, NFSERR_EXPIRED, NFSERR_LEASEMOVED, NFSERR_LOCKSHELD, NFSERR_RESOURCE, NFSERR_SERVERFAULT, NFSERR_STALECLIENTID, 0, }; static short *nfsrv_v4errmap[] = { nfsv4err_null, nfsv4err_null, nfsv4err_null, nfsv4err_access, nfsv4err_close, nfsv4err_commit, nfsv4err_create, nfsv4err_delegpurge, nfsv4err_delegreturn, nfsv4err_getattr, nfsv4err_getfh, nfsv4err_link, nfsv4err_lock, nfsv4err_lockt, nfsv4err_locku, nfsv4err_lookup, nfsv4err_lookupp, nfsv4err_nverify, nfsv4err_open, nfsv4err_openattr, nfsv4err_openconfirm, nfsv4err_opendowngrade, nfsv4err_putfh, nfsv4err_putpubfh, nfsv4err_putrootfh, nfsv4err_read, nfsv4err_readdir, nfsv4err_readlink, nfsv4err_remove, nfsv4err_rename, nfsv4err_renew, nfsv4err_restorefh, nfsv4err_savefh, nfsv4err_secinfo, nfsv4err_setattr, nfsv4err_setclientid, nfsv4err_setclientidconfirm, nfsv4err_verify, nfsv4err_write, nfsv4err_releaselockowner, }; /* * Trim tlen bytes off the end of the mbuf list and then ensure * the end of the last mbuf is nul filled to a long boundary, * as indicated by the value of "nul". * Return the last mbuf in the updated list and free and mbufs * that follow it in the original list. * This is somewhat different than the old nfsrv_adj() with * support for ext_pgs mbufs. It frees the remaining mbufs * instead of setting them 0 length, since lists of ext_pgs * mbufs are all expected to be non-empty. */ struct mbuf * nfsrv_adj(struct mbuf *mp, int len, int nul) { struct mbuf *m, *m2; vm_page_t pg; int i, lastlen, pgno, plen, tlen, trim; uint16_t off; char *cp; /* * Find the last mbuf after adjustment and * how much it needs to be adjusted by. */ tlen = 0; m = mp; for (;;) { tlen += m->m_len; if (m->m_next == NULL) break; m = m->m_next; } /* m is now the last mbuf and tlen the total length. */ if (len >= m->m_len) { /* Need to trim away the last mbuf(s). */ i = tlen - len; m = mp; for (;;) { if (m->m_len >= i) break; i -= m->m_len; m = m->m_next; } lastlen = i; } else lastlen = m->m_len - len; /* * m is now the last mbuf after trimming and its length needs to * be lastlen. * Adjust the last mbuf and set cp to point to where nuls must be * written. */ if ((m->m_flags & M_EXTPG) != 0) { pgno = m->m_epg_npgs - 1; off = (pgno == 0) ? m->m_epg_1st_off : 0; plen = m_epg_pagelen(m, pgno, off); if (m->m_len > lastlen) { /* Trim this mbuf. */ trim = m->m_len - lastlen; while (trim >= plen) { KASSERT(pgno > 0, ("nfsrv_adj: freeing page 0")); /* Free page. */ pg = PHYS_TO_VM_PAGE(m->m_epg_pa[pgno]); vm_page_unwire_noq(pg); vm_page_free(pg); trim -= plen; m->m_epg_npgs--; pgno--; off = (pgno == 0) ? m->m_epg_1st_off : 0; plen = m_epg_pagelen(m, pgno, off); } plen -= trim; m->m_epg_last_len = plen; m->m_len = lastlen; } cp = (char *)(void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); cp += off + plen - nul; } else { m->m_len = lastlen; cp = mtod(m, char *) + m->m_len - nul; } /* Write the nul bytes. */ for (i = 0; i < nul; i++) *cp++ = '\0'; /* Free up any mbufs past "m". */ m2 = m->m_next; m->m_next = NULL; if (m2 != NULL) m_freem(m2); return (m); } /* * Make these functions instead of macros, so that the kernel text size * doesn't get too big... */ void nfsrv_wcc(struct nfsrv_descript *nd, int before_ret, struct nfsvattr *before_nvap, int after_ret, struct nfsvattr *after_nvap) { u_int32_t *tl; if (before_ret) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; } else { NFSM_BUILD(tl, u_int32_t *, 7 * NFSX_UNSIGNED); *tl++ = newnfs_true; txdr_hyper(before_nvap->na_size, tl); tl += 2; txdr_nfsv3time(&(before_nvap->na_mtime), tl); tl += 2; txdr_nfsv3time(&(before_nvap->na_ctime), tl); } nfsrv_postopattr(nd, after_ret, after_nvap); } void nfsrv_postopattr(struct nfsrv_descript *nd, int after_ret, struct nfsvattr *after_nvap) { u_int32_t *tl; NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (after_ret) *tl = newnfs_false; else { *tl = newnfs_true; nfsrv_fillattr(nd, after_nvap); } } /* * Fill in file attributes for V2 and 3. For V4, call a separate * routine that sifts through all the attribute bits. */ void nfsrv_fillattr(struct nfsrv_descript *nd, struct nfsvattr *nvap) { struct nfs_fattr *fp; int fattr_size; /* * Build space for the attribute structure. */ if (nd->nd_flag & ND_NFSV3) fattr_size = NFSX_V3FATTR; else fattr_size = NFSX_V2FATTR; NFSM_BUILD(fp, struct nfs_fattr *, fattr_size); /* * Now just fill it all in. */ fp->fa_nlink = txdr_unsigned(nvap->na_nlink); fp->fa_uid = txdr_unsigned(nvap->na_uid); fp->fa_gid = txdr_unsigned(nvap->na_gid); if (nd->nd_flag & ND_NFSV3) { fp->fa_type = vtonfsv34_type(nvap->na_type); fp->fa_mode = vtonfsv34_mode(nvap->na_mode); txdr_hyper(nvap->na_size, (uint32_t*)&fp->fa3_size); txdr_hyper(nvap->na_bytes, (uint32_t*)&fp->fa3_used); fp->fa3_rdev.specdata1 = txdr_unsigned(NFSMAJOR(nvap->na_rdev)); fp->fa3_rdev.specdata2 = txdr_unsigned(NFSMINOR(nvap->na_rdev)); fp->fa3_fsid.nfsuquad[0] = 0; fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(nvap->na_fsid); txdr_hyper(nvap->na_fileid, (uint32_t*)&fp->fa3_fileid); txdr_nfsv3time(&nvap->na_atime, &fp->fa3_atime); txdr_nfsv3time(&nvap->na_mtime, &fp->fa3_mtime); txdr_nfsv3time(&nvap->na_ctime, &fp->fa3_ctime); } else { fp->fa_type = vtonfsv2_type(nvap->na_type); fp->fa_mode = vtonfsv2_mode(nvap->na_type, nvap->na_mode); fp->fa2_size = txdr_unsigned(nvap->na_size); fp->fa2_blocksize = txdr_unsigned(nvap->na_blocksize); if (nvap->na_type == VFIFO) fp->fa2_rdev = 0xffffffff; else fp->fa2_rdev = txdr_unsigned(nvap->na_rdev); fp->fa2_blocks = txdr_unsigned(nvap->na_bytes / NFS_FABLKSIZE); fp->fa2_fsid = txdr_unsigned(nvap->na_fsid); fp->fa2_fileid = txdr_unsigned(nvap->na_fileid); txdr_nfsv2time(&nvap->na_atime, &fp->fa2_atime); txdr_nfsv2time(&nvap->na_mtime, &fp->fa2_mtime); txdr_nfsv2time(&nvap->na_ctime, &fp->fa2_ctime); } } /* * This function gets a file handle out of an mbuf list. * It returns 0 for success, EBADRPC otherwise. * If sets the third flagp argument to 1 if the file handle is * the public file handle. * For NFSv4, if the length is incorrect, set nd_repstat == NFSERR_BADHANDLE */ int nfsrv_mtofh(struct nfsrv_descript *nd, struct nfsrvfh *fhp) { u_int32_t *tl; int error = 0, len, copylen; if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len == 0 && nfs_pubfhset && (nd->nd_flag & ND_NFSV3) && nd->nd_procnum == NFSPROC_LOOKUP) { nd->nd_flag |= ND_PUBLOOKUP; goto nfsmout; } copylen = len; /* If len == NFSX_V4PNFSFH the RPC is a pNFS DS one. */ if (len == NFSX_V4PNFSFH && (nd->nd_flag & ND_NFSV41) != 0) { copylen = NFSX_MYFH; len = NFSM_RNDUP(len); nd->nd_flag |= ND_DSSERVER; } else if (len < NFSRV_MINFH || len > NFSRV_MAXFH) { if (nd->nd_flag & ND_NFSV4) { if (len > 0 && len <= NFSX_V4FHMAX) { error = nfsm_advance(nd, NFSM_RNDUP(len), -1); if (error) goto nfsmout; nd->nd_repstat = NFSERR_BADHANDLE; goto nfsmout; } else { error = EBADRPC; goto nfsmout; } } else { error = EBADRPC; goto nfsmout; } } } else { /* * For NFSv2, the file handle is always 32 bytes on the * wire, but this server only cares about the first * NFSRV_MAXFH bytes. */ len = NFSX_V2FH; copylen = NFSRV_MAXFH; } NFSM_DISSECT(tl, u_int32_t *, len); if ((nd->nd_flag & ND_NFSV2) && nfs_pubfhset && nd->nd_procnum == NFSPROC_LOOKUP && !NFSBCMP((caddr_t)tl, nfs_v2pubfh, NFSX_V2FH)) { nd->nd_flag |= ND_PUBLOOKUP; goto nfsmout; } NFSBCOPY(tl, (caddr_t)fhp->nfsrvfh_data, copylen); fhp->nfsrvfh_len = copylen; nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Map errnos to NFS error numbers. For Version 3 and 4 also filter out error * numbers not specified for the associated procedure. * NFSPROC_NOOP is a special case, where the high order bits of nd_repstat * should be cleared. NFSPROC_NOOP is used to return errors when a valid * RPC procedure is not involved. * Returns the error number in XDR. */ int nfsd_errmap(struct nfsrv_descript *nd) { short *defaulterrp, *errp; if (!nd->nd_repstat) return (0); if ((nd->nd_repstat & NFSERR_AUTHERR) != 0) return (txdr_unsigned(NFSERR_ACCES)); if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { if (nd->nd_procnum == NFSPROC_NOOP) return (txdr_unsigned(nd->nd_repstat & 0xffff)); if (nd->nd_flag & ND_NFSV3) errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; else if (nd->nd_repstat == EBADRPC) return (txdr_unsigned(NFSERR_BADXDR)); else if (nd->nd_repstat == NFSERR_MINORVERMISMATCH || nd->nd_repstat == NFSERR_OPILLEGAL) return (txdr_unsigned(nd->nd_repstat)); else if (nd->nd_repstat == NFSERR_REPLYFROMCACHE) return (txdr_unsigned(NFSERR_IO)); else if ((nd->nd_flag & ND_NFSV41) != 0) { if (nd->nd_repstat == EOPNOTSUPP) nd->nd_repstat = NFSERR_NOTSUPP; nd->nd_repstat = nfsrv_isannfserr(nd->nd_repstat); return (txdr_unsigned(nd->nd_repstat)); } else errp = defaulterrp = nfsrv_v4errmap[nd->nd_procnum]; while (*++errp) if (*errp == nd->nd_repstat) return (txdr_unsigned(nd->nd_repstat)); return (txdr_unsigned(*defaulterrp)); } if (nd->nd_repstat <= NFSERR_REMOTE) return (txdr_unsigned(nfsrv_v2errmap[nd->nd_repstat - 1])); return (txdr_unsigned(NFSERR_IO)); } /* * Check to see if the error is a valid NFS one. If not, replace it with * NFSERR_IO. */ static u_int32_t nfsrv_isannfserr(u_int32_t errval) { if (errval == NFSERR_OK) return (errval); if (errval >= NFSERR_BADHANDLE && errval <= NFSERR_MAXERRVAL) return (errval); if (errval > 0 && errval <= NFSERR_REMOTE) return (nfsrv_v2errmap[errval - 1]); return (NFSERR_IO); } /* * Check to see if setting a uid/gid is permitted when creating a new * file object. (Called when uid and/or gid is specified in the * settable attributes for V4. */ int nfsrv_checkuidgid(struct nfsrv_descript *nd, struct nfsvattr *nvap) { int error = 0; /* * If not setting either uid nor gid, it's OK. */ if (NFSVNO_NOTSETUID(nvap) && NFSVNO_NOTSETGID(nvap)) goto out; - if ((NFSVNO_ISSETUID(nvap) && nvap->na_uid == nfsrv_defaultuid && - enable_nobodycheck == 1) - || (NFSVNO_ISSETGID(nvap) && nvap->na_gid == nfsrv_defaultgid && - enable_nogroupcheck == 1)) { + if ((NFSVNO_ISSETUID(nvap) && + nvap->na_uid == NFSD_VNET(nfsrv_defaultuid) && + enable_nobodycheck == 1) || + (NFSVNO_ISSETGID(nvap) && + nvap->na_gid == NFSD_VNET(nfsrv_defaultgid) && + enable_nogroupcheck == 1)) { error = NFSERR_BADOWNER; goto out; } if (nd->nd_cred->cr_uid == 0) goto out; if ((NFSVNO_ISSETUID(nvap) && nvap->na_uid != nd->nd_cred->cr_uid) || (NFSVNO_ISSETGID(nvap) && nvap->na_gid != nd->nd_cred->cr_gid && !groupmember(nvap->na_gid, nd->nd_cred))) error = NFSERR_PERM; out: NFSEXITCODE2(error, nd); return (error); } /* * and this routine fixes up the settable attributes for V4 if allowed * by nfsrv_checkuidgid(). */ void nfsrv_fixattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, NFSACL_T *aclp, NFSPROC_T *p, nfsattrbit_t *attrbitp, struct nfsexstuff *exp) { int change = 0; struct nfsvattr nva; uid_t tuid; int error; nfsattrbit_t nattrbits; /* * Maybe this should be done for V2 and 3 but it never has been * and nobody seems to be upset, so I think it's best not to change * the V2 and 3 semantics. */ if ((nd->nd_flag & ND_NFSV4) == 0) goto out; NFSVNO_ATTRINIT(&nva); NFSZERO_ATTRBIT(&nattrbits); tuid = nd->nd_cred->cr_uid; if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) && NFSVNO_ISSETUID(nvap) && nvap->na_uid != nd->nd_cred->cr_uid) { if (nd->nd_cred->cr_uid == 0) { nva.na_uid = nvap->na_uid; change++; NFSSETBIT_ATTRBIT(&nattrbits, NFSATTRBIT_OWNER); } else { NFSCLRBIT_ATTRBIT(attrbitp, NFSATTRBIT_OWNER); } } if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESSSET) && NFSVNO_ISSETATIME(nvap)) { nva.na_atime = nvap->na_atime; change++; NFSSETBIT_ATTRBIT(&nattrbits, NFSATTRBIT_TIMEACCESSSET); } if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFYSET) && NFSVNO_ISSETMTIME(nvap)) { nva.na_mtime = nvap->na_mtime; change++; NFSSETBIT_ATTRBIT(&nattrbits, NFSATTRBIT_TIMEMODIFYSET); } if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) && NFSVNO_ISSETGID(nvap)) { if (nvap->na_gid == nd->nd_cred->cr_gid || groupmember(nvap->na_gid, nd->nd_cred)) { nd->nd_cred->cr_uid = 0; nva.na_gid = nvap->na_gid; change++; NFSSETBIT_ATTRBIT(&nattrbits, NFSATTRBIT_OWNERGROUP); } else { NFSCLRBIT_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP); } } if (change) { error = nfsvno_setattr(vp, &nva, nd->nd_cred, p, exp); if (error) { NFSCLRALL_ATTRBIT(attrbitp, &nattrbits); } } if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) && NFSVNO_ISSETSIZE(nvap) && nvap->na_size != (u_quad_t)0) { NFSCLRBIT_ATTRBIT(attrbitp, NFSATTRBIT_SIZE); } #ifdef NFS4_ACL_EXTATTR_NAME if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL) && nfsrv_useacl != 0 && aclp != NULL) { if (aclp->acl_cnt > 0) { error = nfsrv_setacl(vp, aclp, nd->nd_cred, p); if (error) { NFSCLRBIT_ATTRBIT(attrbitp, NFSATTRBIT_ACL); } } } else #endif NFSCLRBIT_ATTRBIT(attrbitp, NFSATTRBIT_ACL); nd->nd_cred->cr_uid = tuid; out: NFSEXITCODE2(0, nd); } /* * Translate an ASCII hex digit to it's binary value. Return -1 if the * char isn't a hex digit. */ static char nfsrv_hexdigit(char c, int *err) { *err = 0; if (c >= '0' && c <= '9') return (c - '0'); if (c >= 'a' && c <= 'f') return (c - 'a' + ((char)10)); if (c >= 'A' && c <= 'F') return (c - 'A' + ((char)10)); /* Not valid ! */ *err = 1; return (1); /* BOGUS */ } /* * Check to see if NFSERR_MOVED can be returned for this op. Return 1 iff * it can be. */ int nfsrv_errmoved(int op) { short *errp; errp = nfsrv_v4errmap[op]; while (*errp != 0) { if (*errp == NFSERR_MOVED) return (1); errp++; } return (0); } /* * Fill in attributes for a Referral. * (Return the number of bytes of XDR created.) */ int nfsrv_putreferralattr(struct nfsrv_descript *nd, nfsattrbit_t *retbitp, struct nfsreferral *refp, int getattr, int *reterrp) { u_int32_t *tl, *retnump; u_char *cp, *cp2; int prefixnum, retnum = 0, i, len, bitpos, rderrbit = 0, nonrefbit = 0; int fslocationsbit = 0; nfsattrbit_t tmpbits, refbits; NFSREFERRAL_ATTRBIT(&refbits); if (getattr) NFSCLRBIT_ATTRBIT(&refbits, NFSATTRBIT_RDATTRERROR); else if (NFSISSET_ATTRBIT(retbitp, NFSATTRBIT_RDATTRERROR)) rderrbit = 1; if (NFSISSET_ATTRBIT(retbitp, NFSATTRBIT_FSLOCATIONS)) fslocationsbit = 1; /* * Check for the case where unsupported referral attributes are * requested. */ NFSSET_ATTRBIT(&tmpbits, retbitp); NFSCLRALL_ATTRBIT(&tmpbits, &refbits); if (NFSNONZERO_ATTRBIT(&tmpbits)) nonrefbit = 1; if (nonrefbit && !fslocationsbit && (getattr || !rderrbit)) { *reterrp = NFSERR_MOVED; return (0); } /* * Now we can fill in the attributes. */ NFSSET_ATTRBIT(&tmpbits, retbitp); NFSCLRNOT_ATTRBIT(&tmpbits, &refbits); /* * Put out the attribute bitmap for the ones being filled in * and get the field for the number of attributes returned. */ prefixnum = nfsrv_putattrbit(nd, &tmpbits); NFSM_BUILD(retnump, u_int32_t *, NFSX_UNSIGNED); prefixnum += NFSX_UNSIGNED; /* * Now, loop around filling in the attributes for each bit set. */ for (bitpos = 0; bitpos < NFSATTRBIT_MAX; bitpos++) { if (NFSISSET_ATTRBIT(&tmpbits, bitpos)) { switch (bitpos) { case NFSATTRBIT_TYPE: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFDIR); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_FSID: NFSM_BUILD(tl, u_int32_t *, NFSX_V4FSID); *tl++ = 0; *tl++ = txdr_unsigned(NFSV4ROOT_FSID0); *tl++ = 0; *tl = txdr_unsigned(NFSV4ROOT_REFERRAL); retnum += NFSX_V4FSID; break; case NFSATTRBIT_RDATTRERROR: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (nonrefbit) *tl = txdr_unsigned(NFSERR_MOVED); else *tl = 0; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_FSLOCATIONS: retnum += nfsm_strtom(nd, "/", 1); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(refp->nfr_srvcnt); retnum += NFSX_UNSIGNED; cp = refp->nfr_srvlist; for (i = 0; i < refp->nfr_srvcnt; i++) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(1); retnum += NFSX_UNSIGNED; cp2 = STRCHR(cp, ':'); if (cp2 != NULL) len = cp2 - cp; else len = 1; retnum += nfsm_strtom(nd, cp, len); if (cp2 != NULL) cp = cp2 + 1; cp2 = STRCHR(cp, ','); if (cp2 != NULL) len = cp2 - cp; else len = strlen(cp); retnum += nfsm_strtom(nd, cp, len); if (cp2 != NULL) cp = cp2 + 1; } break; case NFSATTRBIT_MOUNTEDONFILEID: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); txdr_hyper(refp->nfr_dfileno, tl); retnum += NFSX_HYPER; break; default: printf("EEK! Bad V4 refattr bitpos=%d\n", bitpos); } } } *retnump = txdr_unsigned(retnum); return (retnum + prefixnum); } /* * Parse a file name out of a request. */ int nfsrv_parsename(struct nfsrv_descript *nd, char *bufp, u_long *hashp, NFSPATHLEN_T *outlenp) { char *fromcp, *tocp, val = '\0'; struct mbuf *md; int i; int rem, len, error = 0, pubtype = 0, outlen = 0, percent = 0; char digit; u_int32_t *tl; u_long hash = 0; if (hashp != NULL) *hashp = 0; tocp = bufp; /* * For V4, check for lookup parent. * Otherwise, get the component name. */ if ((nd->nd_flag & ND_NFSV4) && (nd->nd_procnum == NFSV4OP_LOOKUPP || nd->nd_procnum == NFSV4OP_SECINFONONAME)) { *tocp++ = '.'; hash += ((u_char)'.'); *tocp++ = '.'; hash += ((u_char)'.'); outlen = 2; } else { /* * First, get the name length. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len > NFS_MAXNAMLEN) { nd->nd_repstat = NFSERR_NAMETOL; error = 0; goto nfsmout; } else if (len <= 0) { nd->nd_repstat = NFSERR_INVAL; error = 0; goto nfsmout; } /* * Now, copy the component name into the buffer. */ fromcp = nd->nd_dpos; md = nd->nd_md; rem = mtod(md, caddr_t) + md->m_len - fromcp; for (i = 0; i < len; i++) { while (rem == 0) { md = md->m_next; if (md == NULL) { error = EBADRPC; goto nfsmout; } fromcp = mtod(md, caddr_t); rem = md->m_len; } if (*fromcp == '\0') { nd->nd_repstat = EACCES; error = 0; goto nfsmout; } /* * For lookups on the public filehandle, do some special * processing on the name. (The public file handle is the * root of the public file system for this server.) */ if (nd->nd_flag & ND_PUBLOOKUP) { /* * If the first char is ASCII, it is a canonical * path, otherwise it is a native path. (RFC2054 * doesn't actually state what it is if the first * char isn't ASCII or 0x80, so I assume native.) * pubtype == 1 -> native path * pubtype == 2 -> canonical path */ if (i == 0) { if (*fromcp & 0x80) { /* * Since RFC2054 doesn't indicate * that a native path of just 0x80 * isn't allowed, I'll replace the * 0x80 with '/' instead of just * throwing it away. */ *fromcp = '/'; pubtype = 1; } else { pubtype = 2; } } /* * '/' only allowed in a native path */ if (*fromcp == '/' && pubtype != 1) { nd->nd_repstat = EACCES; error = 0; goto nfsmout; } /* * For the special case of 2 hex digits after a * '%' in an absolute path, calculate the value. * percent == 1 -> indicates "get first hex digit" * percent == 2 -> indicates "get second hex digit" */ if (percent > 0) { digit = nfsrv_hexdigit(*fromcp, &error); if (error) { nd->nd_repstat = EACCES; error = 0; goto nfsmout; } if (percent == 1) { val = (digit << 4); percent = 2; } else { val += digit; percent = 0; *tocp++ = val; hash += ((u_char)val); outlen++; } } else { if (*fromcp == '%' && pubtype == 2) { /* * Must be followed by 2 hex digits */ if ((len - i) < 3) { nd->nd_repstat = EACCES; error = 0; goto nfsmout; } percent = 1; } else { *tocp++ = *fromcp; hash += ((u_char)*fromcp); outlen++; } } } else { /* * Normal, non lookup on public, name. */ if (*fromcp == '/') { if (nd->nd_flag & ND_NFSV4) nd->nd_repstat = NFSERR_BADNAME; else nd->nd_repstat = EACCES; error = 0; goto nfsmout; } hash += ((u_char)*fromcp); *tocp++ = *fromcp; outlen++; } fromcp++; rem--; } nd->nd_md = md; nd->nd_dpos = fromcp; i = NFSM_RNDUP(len) - len; if (i > 0) { if (rem >= i) { nd->nd_dpos += i; } else { error = nfsm_advance(nd, i, rem); if (error) goto nfsmout; } } /* * For v4, don't allow lookups of '.' or '..' and * also check for non-utf8 strings. */ if (nd->nd_flag & ND_NFSV4) { if ((outlen == 1 && bufp[0] == '.') || (outlen == 2 && bufp[0] == '.' && bufp[1] == '.')) { nd->nd_repstat = NFSERR_BADNAME; error = 0; goto nfsmout; } if (enable_checkutf8 == 1 && nfsrv_checkutf8((u_int8_t *)bufp, outlen)) { nd->nd_repstat = NFSERR_INVAL; error = 0; goto nfsmout; } } } *tocp = '\0'; *outlenp = (size_t)outlen + 1; if (hashp != NULL) *hashp = hash; nfsmout: NFSEXITCODE2(error, nd); return (error); } void nfsd_init(void) { int i; - static int inited = 0; - if (inited) - return; - inited = 1; /* * Initialize client queues. Don't free/reinitialize * them when nfsds are restarted. */ - nfsclienthash = malloc(sizeof(struct nfsclienthashhead) * + NFSD_VNET(nfsclienthash) = malloc(sizeof(struct nfsclienthashhead) * nfsrv_clienthashsize, M_NFSDCLIENT, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_clienthashsize; i++) - LIST_INIT(&nfsclienthash[i]); - nfslockhash = malloc(sizeof(struct nfslockhashhead) * + LIST_INIT(&NFSD_VNET(nfsclienthash)[i]); + NFSD_VNET(nfslockhash) = malloc(sizeof(struct nfslockhashhead) * nfsrv_lockhashsize, M_NFSDLOCKFILE, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_lockhashsize; i++) - LIST_INIT(&nfslockhash[i]); - nfssessionhash = malloc(sizeof(struct nfssessionhash) * + LIST_INIT(&NFSD_VNET(nfslockhash)[i]); + NFSD_VNET(nfssessionhash) = malloc(sizeof(struct nfssessionhash) * nfsrv_sessionhashsize, M_NFSDSESSION, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_sessionhashsize; i++) { - mtx_init(&nfssessionhash[i].mtx, "nfssm", NULL, MTX_DEF); - LIST_INIT(&nfssessionhash[i].list); + mtx_init(&NFSD_VNET(nfssessionhash)[i].mtx, "nfssm", NULL, + MTX_DEF); + LIST_INIT(&NFSD_VNET(nfssessionhash)[i].list); } LIST_INIT(&nfsrv_dontlisthead); TAILQ_INIT(&nfsrv_recalllisthead); /* and the v2 pubfh should be all zeros */ NFSBZERO(nfs_v2pubfh, NFSX_V2FH); } /* * Check the v4 root exports. * Return 0 if ok, 1 otherwise. */ int nfsd_checkrootexp(struct nfsrv_descript *nd) { - if (nfs_rootfhset == 0) + if (NFSD_VNET(nfs_rootfhset) == 0) return (NFSERR_AUTHERR | AUTH_FAILED); /* * For NFSv4.1/4.2, if the client specifies SP4_NONE, then these * operations are allowed regardless of the value of the "sec=XXX" * field in the V4: exports line. * As such, these Kerberos checks only apply to NFSv4.0 mounts. */ if ((nd->nd_flag & ND_NFSV41) != 0) goto checktls; if ((nd->nd_flag & (ND_GSS | ND_EXAUTHSYS)) == ND_EXAUTHSYS) goto checktls; if ((nd->nd_flag & (ND_GSSINTEGRITY | ND_EXGSSINTEGRITY)) == (ND_GSSINTEGRITY | ND_EXGSSINTEGRITY)) goto checktls; if ((nd->nd_flag & (ND_GSSPRIVACY | ND_EXGSSPRIVACY)) == (ND_GSSPRIVACY | ND_EXGSSPRIVACY)) goto checktls; if ((nd->nd_flag & (ND_GSS | ND_GSSINTEGRITY | ND_GSSPRIVACY | ND_EXGSS)) == (ND_GSS | ND_EXGSS)) goto checktls; return (NFSERR_AUTHERR | AUTH_TOOWEAK); checktls: if ((nd->nd_flag & ND_EXTLS) == 0) return (0); if ((nd->nd_flag & (ND_TLSCERTUSER | ND_EXTLSCERTUSER)) == (ND_TLSCERTUSER | ND_EXTLSCERTUSER)) return (0); if ((nd->nd_flag & (ND_TLSCERT | ND_EXTLSCERT | ND_EXTLSCERTUSER)) == (ND_TLSCERT | ND_EXTLSCERT)) return (0); if ((nd->nd_flag & (ND_TLS | ND_EXTLSCERTUSER | ND_EXTLSCERT)) == ND_TLS) return (0); #ifdef notnow /* There is currently no auth_stat for this. */ if ((nd->nd_flag & ND_TLS) == 0) return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); #endif return (NFSERR_AUTHERR | AUTH_TOOWEAK); } /* * Parse the first part of an NFSv4 compound to find out what the minor * version# is. */ void nfsd_getminorvers(struct nfsrv_descript *nd, u_char *tag, u_char **tagstrp, int *taglenp, u_int32_t *minversp) { uint32_t *tl; int error = 0, taglen = -1; u_char *tagstr = NULL; NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); taglen = fxdr_unsigned(int, *tl); if (taglen < 0 || taglen > NFSV4_OPAQUELIMIT) { error = EBADRPC; goto nfsmout; } if (taglen <= NFSV4_SMALLSTR) tagstr = tag; else tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK); error = nfsrv_mtostr(nd, tagstr, taglen); if (error != 0) goto nfsmout; NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); *minversp = fxdr_unsigned(u_int32_t, *tl); *tagstrp = tagstr; if (*minversp == NFSV41_MINORVERSION) nd->nd_flag |= ND_NFSV41; else if (*minversp == NFSV42_MINORVERSION) nd->nd_flag |= (ND_NFSV41 | ND_NFSV42); nfsmout: if (error != 0) { if (tagstr != NULL && taglen > NFSV4_SMALLSTR) free(tagstr, M_TEMP); taglen = -1; } *taglenp = taglen; }