diff --git a/sys/fs/nfs/nfs_commonport.c b/sys/fs/nfs/nfs_commonport.c index 60131ab66184..99482069c36d 100644 --- a/sys/fs/nfs/nfs_commonport.c +++ b/sys/fs/nfs/nfs_commonport.c @@ -1,972 +1,970 @@ /*- * 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("$FreeBSD$"); /* * Functions that need to be different for different versions of BSD * kernel should be kept here, along with any global storage specific * to this BSD variant. */ #include #include #include #include #include #include #include #include #include #include #include #include #include extern int nfscl_ticks; extern void (*nfsd_call_recall)(struct vnode *, int, struct ucred *, struct thread *); extern int nfsrv_useacl; int newnfs_numnfsd = 0; struct nfsstatsv1 nfsstatsv1; int nfs_numnfscbd = 0; int nfscl_debuglevel = 0; char nfsv4_callbackaddr[INET6_ADDRSTRLEN]; int nfsrv_lughashsize = 100; struct mtx nfsrv_dslock_mtx; struct nfsdevicehead nfsrv_devidhead; volatile int nfsrv_devidcnt = 0; void (*ncl_call_invalcaches)(struct vnode *) = NULL; vop_advlock_t *nfs_advlock_p = NULL; vop_reclaim_t *nfs_reclaim_p = NULL; uint32_t nfs_srvmaxio = NFS_SRVMAXIO; NFSD_VNET_DEFINE(struct nfsstatsv1 *, nfsstatsv1_p); NFSD_VNET_DECLARE(struct nfssockreq, nfsrv_nfsuserdsock); NFSD_VNET_DECLARE(nfsuserd_state, nfsrv_nfsuserd); int nfs_pnfsio(task_fn_t *, void *); static int nfs_realign_test; static int nfs_realign_count; static struct ext_nfsstats oldnfsstats; static struct nfsstatsov1 nfsstatsov1; SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "NFS filesystem"); SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "Number of realign tests done"); SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "Number of mbuf realignments done"); SYSCTL_STRING(_vfs_nfs, OID_AUTO, callback_addr, CTLFLAG_RW, nfsv4_callbackaddr, sizeof(nfsv4_callbackaddr), "NFSv4 callback addr for server to use"); SYSCTL_INT(_vfs_nfs, OID_AUTO, debuglevel, CTLFLAG_RW, &nfscl_debuglevel, 0, "Debug level for NFS client"); SYSCTL_INT(_vfs_nfs, OID_AUTO, userhashsize, CTLFLAG_RDTUN, &nfsrv_lughashsize, 0, "Size of hash tables for uid/name mapping"); int nfs_pnfsiothreads = -1; SYSCTL_INT(_vfs_nfs, OID_AUTO, pnfsiothreads, CTLFLAG_RW, &nfs_pnfsiothreads, 0, "Number of pNFS mirror I/O threads"); /* * Defines for malloc * (Here for FreeBSD, since they allocate storage.) */ MALLOC_DEFINE(M_NEWNFSRVCACHE, "NFSD srvcache", "NFSD Server Request Cache"); MALLOC_DEFINE(M_NEWNFSDCLIENT, "NFSD V4client", "NFSD V4 Client Id"); MALLOC_DEFINE(M_NEWNFSDSTATE, "NFSD V4state", "NFSD V4 State (Openowner, Open, Lockowner, Delegation"); MALLOC_DEFINE(M_NEWNFSDLOCK, "NFSD V4lock", "NFSD V4 byte range lock"); MALLOC_DEFINE(M_NEWNFSDLOCKFILE, "NFSD lckfile", "NFSD Open/Lock file"); MALLOC_DEFINE(M_NEWNFSSTRING, "NFSD string", "NFSD V4 long string"); MALLOC_DEFINE(M_NEWNFSUSERGROUP, "NFSD usrgroup", "NFSD V4 User/group map"); MALLOC_DEFINE(M_NEWNFSDREQ, "NFS req", "NFS request header"); MALLOC_DEFINE(M_NEWNFSFH, "NFS fh", "NFS file handle"); MALLOC_DEFINE(M_NEWNFSCLOWNER, "NFSCL owner", "NFSCL Open Owner"); MALLOC_DEFINE(M_NEWNFSCLOPEN, "NFSCL open", "NFSCL Open"); MALLOC_DEFINE(M_NEWNFSCLDELEG, "NFSCL deleg", "NFSCL Delegation"); MALLOC_DEFINE(M_NEWNFSCLCLIENT, "NFSCL client", "NFSCL Client"); MALLOC_DEFINE(M_NEWNFSCLLOCKOWNER, "NFSCL lckown", "NFSCL Lock Owner"); MALLOC_DEFINE(M_NEWNFSCLLOCK, "NFSCL lck", "NFSCL Lock"); MALLOC_DEFINE(M_NEWNFSV4NODE, "NEWNFSnode", "NFS vnode"); MALLOC_DEFINE(M_NEWNFSDIRECTIO, "NEWdirectio", "NFS Direct IO buffer"); MALLOC_DEFINE(M_NEWNFSDIROFF, "NFSCL diroff", "NFS directory offset data"); MALLOC_DEFINE(M_NEWNFSDROLLBACK, "NFSD rollback", "NFS local lock rollback"); MALLOC_DEFINE(M_NEWNFSLAYOUT, "NFSCL layout", "NFSv4.1 Layout"); MALLOC_DEFINE(M_NEWNFSFLAYOUT, "NFSCL flayout", "NFSv4.1 File Layout"); MALLOC_DEFINE(M_NEWNFSDEVINFO, "NFSCL devinfo", "NFSv4.1 Device Info"); MALLOC_DEFINE(M_NEWNFSSOCKREQ, "NFSCL sockreq", "NFS Sock Req"); MALLOC_DEFINE(M_NEWNFSCLDS, "NFSCL session", "NFSv4.1 Session"); MALLOC_DEFINE(M_NEWNFSLAYRECALL, "NFSCL layrecall", "NFSv4.1 Layout Recall"); MALLOC_DEFINE(M_NEWNFSDSESSION, "NFSD session", "NFSD Session for a client"); /* * Definition of mutex locks. * newnfsd_mtx is used in nfsrvd_nfsd() to protect the nfs socket list * and assorted other nfsd structures. */ struct mtx newnfsd_mtx; struct mtx nfs_sockl_mutex; struct mtx nfs_state_mutex; struct mtx nfs_nameid_mutex; struct mtx nfs_req_mutex; struct mtx nfs_slock_mutex; struct mtx nfs_clstate_mutex; /* local functions */ static int nfssvc_call(struct thread *, struct nfssvc_args *, struct ucred *); -static void nfs_clean(struct prison *); #ifdef __NO_STRICT_ALIGNMENT /* * These architectures don't need re-alignment, so just return. */ int newnfs_realign(struct mbuf **pm, int how) { return (0); } #else /* !__NO_STRICT_ALIGNMENT */ /* * newnfs_realign: * * Check for badly aligned mbuf data and realign by copying the unaligned * portion of the data into a new mbuf chain and freeing the portions * of the old chain that were replaced. * * We cannot simply realign the data within the existing mbuf chain * because the underlying buffers may contain other rpc commands and * we cannot afford to overwrite them. * * We would prefer to avoid this situation entirely. The situation does * not occur with NFS/UDP and is supposed to only occasionally occur * with TCP. Use vfs.nfs.realign_count and realign_test to check this. * */ int newnfs_realign(struct mbuf **pm, int how) { struct mbuf *m, *n; int off, space; ++nfs_realign_test; while ((m = *pm) != NULL) { if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) { /* * NB: we can't depend on m_pkthdr.len to help us * decide what to do here. May not be worth doing * the m_length calculation as m_copyback will * expand the mbuf chain below as needed. */ space = m_length(m, NULL); if (space >= MINCLSIZE) { /* NB: m_copyback handles space > MCLBYTES */ n = m_getcl(how, MT_DATA, 0); } else n = m_get(how, MT_DATA); if (n == NULL) return (ENOMEM); /* * Align the remainder of the mbuf chain. */ n->m_len = 0; off = 0; while (m != NULL) { m_copyback(n, off, m->m_len, mtod(m, caddr_t)); off += m->m_len; m = m->m_next; } m_freem(*pm); *pm = n; ++nfs_realign_count; break; } pm = &m->m_next; } return (0); } #endif /* __NO_STRICT_ALIGNMENT */ #ifdef notdef static void nfsrv_object_create(struct vnode *vp, struct thread *td) { if (vp == NULL || vp->v_type != VREG) return; (void) vfs_object_create(vp, td, td->td_ucred); } #endif /* * Look up a file name. Basically just initialize stuff and call namei(). */ int nfsrv_lookupfilename(struct nameidata *ndp, char *fname, NFSPROC_T *p __unused) { int error; NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fname); error = namei(ndp); if (!error) { NDFREE_PNBUF(ndp); } return (error); } /* * Copy NFS uid, gids to the cred structure. */ void newnfs_copycred(struct nfscred *nfscr, struct ucred *cr) { KASSERT(nfscr->nfsc_ngroups >= 0, ("newnfs_copycred: negative nfsc_ngroups")); cr->cr_uid = nfscr->nfsc_uid; crsetgroups(cr, nfscr->nfsc_ngroups, nfscr->nfsc_groups); } /* * Map args from nfsmsleep() to msleep(). */ int nfsmsleep(void *chan, void *mutex, int prio, const char *wmesg, struct timespec *ts) { u_int64_t nsecval; int error, timeo; if (ts) { timeo = hz * ts->tv_sec; nsecval = (u_int64_t)ts->tv_nsec; nsecval = ((nsecval * ((u_int64_t)hz)) + 500000000) / 1000000000; timeo += (int)nsecval; } else { timeo = 0; } error = msleep(chan, (struct mtx *)mutex, prio, wmesg, timeo); return (error); } /* * Get the file system info for the server. For now, just assume FFS. */ void nfsvno_getfs(struct nfsfsinfo *sip, int isdgram) { int pref; /* * XXX * There should be file system VFS OP(s) to get this information. * For now, assume ufs. */ if (isdgram) pref = NFS_MAXDGRAMDATA; else pref = nfs_srvmaxio; sip->fs_rtmax = nfs_srvmaxio; sip->fs_rtpref = pref; sip->fs_rtmult = NFS_FABLKSIZE; sip->fs_wtmax = nfs_srvmaxio; sip->fs_wtpref = pref; sip->fs_wtmult = NFS_FABLKSIZE; sip->fs_dtpref = pref; sip->fs_maxfilesize = 0xffffffffffffffffull; sip->fs_timedelta.tv_sec = 0; sip->fs_timedelta.tv_nsec = 1; sip->fs_properties = (NFSV3FSINFO_LINK | NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS | NFSV3FSINFO_CANSETTIME); } /* * Do the pathconf vnode op. */ int nfsvno_pathconf(struct vnode *vp, int flag, long *retf, struct ucred *cred, struct thread *p) { int error; error = VOP_PATHCONF(vp, flag, retf); if (error == EOPNOTSUPP || error == EINVAL) { /* * Some file systems return EINVAL for name arguments not * supported and some return EOPNOTSUPP for this case. * So the NFSv3 Pathconf RPC doesn't fail for these cases, * just fake them. */ switch (flag) { case _PC_LINK_MAX: *retf = NFS_LINK_MAX; break; case _PC_NAME_MAX: *retf = NAME_MAX; break; case _PC_CHOWN_RESTRICTED: *retf = 1; break; case _PC_NO_TRUNC: *retf = 1; break; default: /* * Only happens if a _PC_xxx is added to the server, * but this isn't updated. */ *retf = 0; printf("nfsrvd pathconf flag=%d not supp\n", flag); } error = 0; } NFSEXITCODE(error); return (error); } /* Fake nfsrv_atroot. Just return 0 */ int nfsrv_atroot(struct vnode *vp, uint64_t *retp) { return (0); } /* * Set the credentials to refer to root. * If only the various BSDen could agree on whether cr_gid is a separate * field or cr_groups[0]... */ void newnfs_setroot(struct ucred *cred) { cred->cr_uid = 0; cred->cr_groups[0] = 0; cred->cr_ngroups = 1; } /* * Get the client credential. Used for Renew and recovery. */ struct ucred * newnfs_getcred(void) { struct ucred *cred; struct thread *td = curthread; cred = crdup(td->td_ucred); newnfs_setroot(cred); return (cred); } /* * Sleep for a short period of time unless errval == NFSERR_GRACE, where * the sleep should be for 5 seconds. * Since lbolt doesn't exist in FreeBSD-CURRENT, just use a timeout on * an event that never gets a wakeup. Only return EINTR or 0. */ int nfs_catnap(int prio, int errval, const char *wmesg) { static int non_event; int ret; if (errval == NFSERR_GRACE) ret = tsleep(&non_event, prio, wmesg, 5 * hz); else ret = tsleep(&non_event, prio, wmesg, 1); if (ret != EINTR) ret = 0; return (ret); } /* * Get referral. For now, just fail. */ struct nfsreferral * nfsv4root_getreferral(struct vnode *vp, struct vnode *dvp, u_int32_t fileno) { return (NULL); } static int nfssvc_nfscommon(struct thread *td, struct nfssvc_args *uap) { int error; NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); error = nfssvc_call(td, uap, td->td_ucred); NFSD_CURVNET_RESTORE(); NFSEXITCODE(error); return (error); } static int nfssvc_call(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) { int error = EINVAL, i, j; struct nfsd_idargs nid; struct nfsd_oidargs onid; struct { int vers; /* Just the first field of nfsstats. */ } nfsstatver; if (uap->flag & NFSSVC_IDNAME) { if ((uap->flag & NFSSVC_NEWSTRUCT) != 0) error = copyin(uap->argp, &nid, sizeof(nid)); else { error = copyin(uap->argp, &onid, sizeof(onid)); if (error == 0) { nid.nid_flag = onid.nid_flag; nid.nid_uid = onid.nid_uid; nid.nid_gid = onid.nid_gid; nid.nid_usermax = onid.nid_usermax; nid.nid_usertimeout = onid.nid_usertimeout; nid.nid_name = onid.nid_name; nid.nid_namelen = onid.nid_namelen; nid.nid_ngroup = 0; nid.nid_grps = NULL; } } if (error) goto out; error = nfssvc_idname(&nid); goto out; } else if (uap->flag & NFSSVC_GETSTATS) { if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { /* Copy fields to the old ext_nfsstat structure. */ oldnfsstats.attrcache_hits = NFSD_VNET(nfsstatsv1_p)->attrcache_hits; oldnfsstats.attrcache_misses = NFSD_VNET(nfsstatsv1_p)->attrcache_misses; oldnfsstats.lookupcache_hits = NFSD_VNET(nfsstatsv1_p)->lookupcache_hits; oldnfsstats.lookupcache_misses = NFSD_VNET(nfsstatsv1_p)->lookupcache_misses; oldnfsstats.direofcache_hits = NFSD_VNET(nfsstatsv1_p)->direofcache_hits; oldnfsstats.direofcache_misses = NFSD_VNET(nfsstatsv1_p)->direofcache_misses; oldnfsstats.accesscache_hits = NFSD_VNET(nfsstatsv1_p)->accesscache_hits; oldnfsstats.accesscache_misses = NFSD_VNET(nfsstatsv1_p)->accesscache_misses; oldnfsstats.biocache_reads = NFSD_VNET(nfsstatsv1_p)->biocache_reads; oldnfsstats.read_bios = NFSD_VNET(nfsstatsv1_p)->read_bios; oldnfsstats.read_physios = NFSD_VNET(nfsstatsv1_p)->read_physios; oldnfsstats.biocache_writes = NFSD_VNET(nfsstatsv1_p)->biocache_writes; oldnfsstats.write_bios = NFSD_VNET(nfsstatsv1_p)->write_bios; oldnfsstats.write_physios = NFSD_VNET(nfsstatsv1_p)->write_physios; oldnfsstats.biocache_readlinks = NFSD_VNET(nfsstatsv1_p)->biocache_readlinks; oldnfsstats.readlink_bios = NFSD_VNET(nfsstatsv1_p)->readlink_bios; oldnfsstats.biocache_readdirs = NFSD_VNET(nfsstatsv1_p)->biocache_readdirs; oldnfsstats.readdir_bios = NFSD_VNET(nfsstatsv1_p)->readdir_bios; for (i = 0; i < NFSV4_NPROCS; i++) oldnfsstats.rpccnt[i] = NFSD_VNET(nfsstatsv1_p)->rpccnt[i]; oldnfsstats.rpcretries = NFSD_VNET(nfsstatsv1_p)->rpcretries; for (i = 0; i < NFSV4OP_NOPS; i++) oldnfsstats.srvrpccnt[i] = NFSD_VNET(nfsstatsv1_p)->srvrpccnt[i]; for (i = NFSV42_NOPS, j = NFSV4OP_NOPS; i < NFSV42_NOPS + NFSV4OP_FAKENOPS; i++, j++) oldnfsstats.srvrpccnt[j] = NFSD_VNET(nfsstatsv1_p)->srvrpccnt[i]; oldnfsstats.reserved_0 = 0; oldnfsstats.reserved_1 = 0; oldnfsstats.rpcrequests = NFSD_VNET(nfsstatsv1_p)->rpcrequests; oldnfsstats.rpctimeouts = NFSD_VNET(nfsstatsv1_p)->rpctimeouts; oldnfsstats.rpcunexpected = NFSD_VNET(nfsstatsv1_p)->rpcunexpected; oldnfsstats.rpcinvalid = NFSD_VNET(nfsstatsv1_p)->rpcinvalid; oldnfsstats.srvcache_inproghits = NFSD_VNET(nfsstatsv1_p)->srvcache_inproghits; oldnfsstats.reserved_2 = 0; oldnfsstats.srvcache_nonidemdonehits = NFSD_VNET(nfsstatsv1_p)->srvcache_nonidemdonehits; oldnfsstats.srvcache_misses = NFSD_VNET(nfsstatsv1_p)->srvcache_misses; oldnfsstats.srvcache_tcppeak = NFSD_VNET(nfsstatsv1_p)->srvcache_tcppeak; oldnfsstats.srvcache_size = NFSD_VNET(nfsstatsv1_p)->srvcache_size; oldnfsstats.srvclients = NFSD_VNET(nfsstatsv1_p)->srvclients; oldnfsstats.srvopenowners = NFSD_VNET(nfsstatsv1_p)->srvopenowners; oldnfsstats.srvopens = NFSD_VNET(nfsstatsv1_p)->srvopens; oldnfsstats.srvlockowners = NFSD_VNET(nfsstatsv1_p)->srvlockowners; oldnfsstats.srvlocks = NFSD_VNET(nfsstatsv1_p)->srvlocks; oldnfsstats.srvdelegates = NFSD_VNET(nfsstatsv1_p)->srvdelegates; for (i = 0; i < NFSV4OP_CBNOPS; i++) oldnfsstats.cbrpccnt[i] = NFSD_VNET(nfsstatsv1_p)->cbrpccnt[i]; oldnfsstats.clopenowners = NFSD_VNET(nfsstatsv1_p)->clopenowners; oldnfsstats.clopens = NFSD_VNET(nfsstatsv1_p)->clopens; oldnfsstats.cllockowners = NFSD_VNET(nfsstatsv1_p)->cllockowners; oldnfsstats.cllocks = NFSD_VNET(nfsstatsv1_p)->cllocks; oldnfsstats.cldelegates = NFSD_VNET(nfsstatsv1_p)->cldelegates; oldnfsstats.cllocalopenowners = NFSD_VNET(nfsstatsv1_p)->cllocalopenowners; oldnfsstats.cllocalopens = NFSD_VNET(nfsstatsv1_p)->cllocalopens; oldnfsstats.cllocallockowners = NFSD_VNET(nfsstatsv1_p)->cllocallockowners; oldnfsstats.cllocallocks = NFSD_VNET(nfsstatsv1_p)->cllocallocks; error = copyout(&oldnfsstats, uap->argp, sizeof (oldnfsstats)); } else { error = copyin(uap->argp, &nfsstatver, sizeof(nfsstatver)); if (error == 0) { if (nfsstatver.vers == NFSSTATS_OV1) { /* Copy nfsstatsv1 to nfsstatsov1. */ nfsstatsov1.attrcache_hits = NFSD_VNET(nfsstatsv1_p)->attrcache_hits; nfsstatsov1.attrcache_misses = NFSD_VNET(nfsstatsv1_p)->attrcache_misses; nfsstatsov1.lookupcache_hits = NFSD_VNET(nfsstatsv1_p)->lookupcache_hits; nfsstatsov1.lookupcache_misses = NFSD_VNET(nfsstatsv1_p)->lookupcache_misses; nfsstatsov1.direofcache_hits = NFSD_VNET(nfsstatsv1_p)->direofcache_hits; nfsstatsov1.direofcache_misses = NFSD_VNET(nfsstatsv1_p)->direofcache_misses; nfsstatsov1.accesscache_hits = NFSD_VNET(nfsstatsv1_p)->accesscache_hits; nfsstatsov1.accesscache_misses = NFSD_VNET(nfsstatsv1_p)->accesscache_misses; nfsstatsov1.biocache_reads = NFSD_VNET(nfsstatsv1_p)->biocache_reads; nfsstatsov1.read_bios = NFSD_VNET(nfsstatsv1_p)->read_bios; nfsstatsov1.read_physios = NFSD_VNET(nfsstatsv1_p)->read_physios; nfsstatsov1.biocache_writes = NFSD_VNET(nfsstatsv1_p)->biocache_writes; nfsstatsov1.write_bios = NFSD_VNET(nfsstatsv1_p)->write_bios; nfsstatsov1.write_physios = NFSD_VNET(nfsstatsv1_p)->write_physios; nfsstatsov1.biocache_readlinks = NFSD_VNET(nfsstatsv1_p)->biocache_readlinks; nfsstatsov1.readlink_bios = NFSD_VNET(nfsstatsv1_p)->readlink_bios; nfsstatsov1.biocache_readdirs = NFSD_VNET(nfsstatsv1_p)->biocache_readdirs; nfsstatsov1.readdir_bios = NFSD_VNET(nfsstatsv1_p)->readdir_bios; for (i = 0; i < NFSV42_OLDNPROCS; i++) nfsstatsov1.rpccnt[i] = NFSD_VNET(nfsstatsv1_p)->rpccnt[i]; nfsstatsov1.rpcretries = NFSD_VNET(nfsstatsv1_p)->rpcretries; for (i = 0; i < NFSV42_PURENOPS; i++) nfsstatsov1.srvrpccnt[i] = NFSD_VNET(nfsstatsv1_p)->srvrpccnt[i]; for (i = NFSV42_NOPS, j = NFSV42_PURENOPS; i < NFSV42_NOPS + NFSV4OP_FAKENOPS; i++, j++) nfsstatsov1.srvrpccnt[j] = NFSD_VNET(nfsstatsv1_p)->srvrpccnt[i]; nfsstatsov1.reserved_0 = 0; nfsstatsov1.reserved_1 = 0; nfsstatsov1.rpcrequests = NFSD_VNET(nfsstatsv1_p)->rpcrequests; nfsstatsov1.rpctimeouts = NFSD_VNET(nfsstatsv1_p)->rpctimeouts; nfsstatsov1.rpcunexpected = NFSD_VNET(nfsstatsv1_p)->rpcunexpected; nfsstatsov1.rpcinvalid = NFSD_VNET(nfsstatsv1_p)->rpcinvalid; nfsstatsov1.srvcache_inproghits = NFSD_VNET(nfsstatsv1_p)->srvcache_inproghits; nfsstatsov1.reserved_2 = 0; nfsstatsov1.srvcache_nonidemdonehits = NFSD_VNET(nfsstatsv1_p)->srvcache_nonidemdonehits; nfsstatsov1.srvcache_misses = NFSD_VNET(nfsstatsv1_p)->srvcache_misses; nfsstatsov1.srvcache_tcppeak = NFSD_VNET(nfsstatsv1_p)->srvcache_tcppeak; nfsstatsov1.srvcache_size = NFSD_VNET(nfsstatsv1_p)->srvcache_size; nfsstatsov1.srvclients = NFSD_VNET(nfsstatsv1_p)->srvclients; nfsstatsov1.srvopenowners = NFSD_VNET(nfsstatsv1_p)->srvopenowners; nfsstatsov1.srvopens = NFSD_VNET(nfsstatsv1_p)->srvopens; nfsstatsov1.srvlockowners = NFSD_VNET(nfsstatsv1_p)->srvlockowners; nfsstatsov1.srvlocks = NFSD_VNET(nfsstatsv1_p)->srvlocks; nfsstatsov1.srvdelegates = NFSD_VNET(nfsstatsv1_p)->srvdelegates; for (i = 0; i < NFSV42_CBNOPS; i++) nfsstatsov1.cbrpccnt[i] = NFSD_VNET(nfsstatsv1_p)->cbrpccnt[i]; nfsstatsov1.clopenowners = NFSD_VNET(nfsstatsv1_p)->clopenowners; nfsstatsov1.clopens = NFSD_VNET(nfsstatsv1_p)->clopens; nfsstatsov1.cllockowners = NFSD_VNET(nfsstatsv1_p)->cllockowners; nfsstatsov1.cllocks = NFSD_VNET(nfsstatsv1_p)->cllocks; nfsstatsov1.cldelegates = NFSD_VNET(nfsstatsv1_p)->cldelegates; nfsstatsov1.cllocalopenowners = NFSD_VNET(nfsstatsv1_p)->cllocalopenowners; nfsstatsov1.cllocalopens = NFSD_VNET(nfsstatsv1_p)->cllocalopens; nfsstatsov1.cllocallockowners = NFSD_VNET(nfsstatsv1_p)->cllocallockowners; nfsstatsov1.cllocallocks = NFSD_VNET(nfsstatsv1_p)->cllocallocks; nfsstatsov1.srvstartcnt = NFSD_VNET(nfsstatsv1_p)->srvstartcnt; nfsstatsov1.srvdonecnt = NFSD_VNET(nfsstatsv1_p)->srvdonecnt; for (i = NFSV42_NOPS, j = NFSV42_PURENOPS; i < NFSV42_NOPS + NFSV4OP_FAKENOPS; i++, j++) { nfsstatsov1.srvbytes[j] = NFSD_VNET(nfsstatsv1_p)->srvbytes[i]; nfsstatsov1.srvops[j] = NFSD_VNET(nfsstatsv1_p)->srvops[i]; nfsstatsov1.srvduration[j] = NFSD_VNET(nfsstatsv1_p)->srvduration[i]; } nfsstatsov1.busyfrom = NFSD_VNET(nfsstatsv1_p)->busyfrom; nfsstatsov1.busyfrom = NFSD_VNET(nfsstatsv1_p)->busyfrom; error = copyout(&nfsstatsov1, uap->argp, sizeof(nfsstatsov1)); } else if (nfsstatver.vers != NFSSTATS_V1) error = EPERM; else error = copyout(NFSD_VNET(nfsstatsv1_p), uap->argp, sizeof(nfsstatsv1)); } } if (error == 0) { if ((uap->flag & NFSSVC_ZEROCLTSTATS) != 0) { NFSD_VNET(nfsstatsv1_p)->attrcache_hits = 0; NFSD_VNET(nfsstatsv1_p)->attrcache_misses = 0; NFSD_VNET(nfsstatsv1_p)->lookupcache_hits = 0; NFSD_VNET(nfsstatsv1_p)->lookupcache_misses = 0; NFSD_VNET(nfsstatsv1_p)->direofcache_hits = 0; NFSD_VNET(nfsstatsv1_p)->direofcache_misses = 0; NFSD_VNET(nfsstatsv1_p)->accesscache_hits = 0; NFSD_VNET(nfsstatsv1_p)->accesscache_misses = 0; NFSD_VNET(nfsstatsv1_p)->biocache_reads = 0; NFSD_VNET(nfsstatsv1_p)->read_bios = 0; NFSD_VNET(nfsstatsv1_p)->read_physios = 0; NFSD_VNET(nfsstatsv1_p)->biocache_writes = 0; NFSD_VNET(nfsstatsv1_p)->write_bios = 0; NFSD_VNET(nfsstatsv1_p)->write_physios = 0; NFSD_VNET(nfsstatsv1_p)->biocache_readlinks = 0; NFSD_VNET(nfsstatsv1_p)->readlink_bios = 0; NFSD_VNET(nfsstatsv1_p)->biocache_readdirs = 0; NFSD_VNET(nfsstatsv1_p)->readdir_bios = 0; NFSD_VNET(nfsstatsv1_p)->rpcretries = 0; NFSD_VNET(nfsstatsv1_p)->rpcrequests = 0; NFSD_VNET(nfsstatsv1_p)->rpctimeouts = 0; NFSD_VNET(nfsstatsv1_p)->rpcunexpected = 0; NFSD_VNET(nfsstatsv1_p)->rpcinvalid = 0; bzero(NFSD_VNET(nfsstatsv1_p)->rpccnt, sizeof(NFSD_VNET(nfsstatsv1_p)->rpccnt)); } if ((uap->flag & NFSSVC_ZEROSRVSTATS) != 0) { NFSD_VNET(nfsstatsv1_p)->srvcache_inproghits = 0; NFSD_VNET(nfsstatsv1_p)->srvcache_nonidemdonehits = 0; NFSD_VNET(nfsstatsv1_p)->srvcache_misses = 0; NFSD_VNET(nfsstatsv1_p)->srvcache_tcppeak = 0; bzero(NFSD_VNET(nfsstatsv1_p)->srvrpccnt, sizeof(NFSD_VNET(nfsstatsv1_p)->srvrpccnt)); bzero(NFSD_VNET(nfsstatsv1_p)->cbrpccnt, sizeof(NFSD_VNET(nfsstatsv1_p)->cbrpccnt)); } } goto out; } else if (uap->flag & NFSSVC_NFSUSERDPORT) { u_short sockport; struct nfsuserd_args nargs; if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { error = copyin(uap->argp, (caddr_t)&sockport, sizeof (u_short)); if (error == 0) { nargs.nuserd_family = AF_INET; nargs.nuserd_port = sockport; } } else { /* * New nfsuserd_args structure, which indicates * which IP version to use along with the port#. */ error = copyin(uap->argp, &nargs, sizeof(nargs)); } if (!error) error = nfsrv_nfsuserdport(&nargs, p); } else if (uap->flag & NFSSVC_NFSUSERDDELPORT) { nfsrv_nfsuserddelport(); error = 0; } out: NFSEXITCODE(error); return (error); } /* * called by all three modevent routines, so that it gets things * initialized soon enough. */ void newnfs_portinit(void) { static int inited = 0; if (inited) return; inited = 1; /* Initialize SMP locks used by both client and server. */ mtx_init(&newnfsd_mtx, "newnfsd_mtx", NULL, MTX_DEF); mtx_init(&nfs_state_mutex, "nfs_state_mutex", NULL, MTX_DEF); mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL, MTX_DEF); } /* * Determine if the file system supports NFSv4 ACLs. * Return 1 if it does, 0 otherwise. */ int nfs_supportsnfsv4acls(struct vnode *vp) { int error; long retval; ASSERT_VOP_LOCKED(vp, "nfs supports nfsv4acls"); if (nfsrv_useacl == 0) return (0); error = VOP_PATHCONF(vp, _PC_ACL_NFS4, &retval); if (error == 0 && retval != 0) return (1); return (0); } /* * These are the first fields of all the context structures passed into * nfs_pnfsio(). */ struct pnfsio { int done; int inprog; struct task tsk; }; /* * Do a mirror I/O on a pNFS thread. */ int nfs_pnfsio(task_fn_t *func, void *context) { struct pnfsio *pio; int ret; static struct taskqueue *pnfsioq = NULL; pio = (struct pnfsio *)context; if (pnfsioq == NULL) { if (nfs_pnfsiothreads == 0) return (EPERM); if (nfs_pnfsiothreads < 0) nfs_pnfsiothreads = mp_ncpus * 4; pnfsioq = taskqueue_create("pnfsioq", M_WAITOK, taskqueue_thread_enqueue, &pnfsioq); if (pnfsioq == NULL) return (ENOMEM); ret = taskqueue_start_threads(&pnfsioq, nfs_pnfsiothreads, 0, "pnfsiot"); if (ret != 0) { taskqueue_free(pnfsioq); pnfsioq = NULL; return (ret); } } pio->inprog = 1; TASK_INIT(&pio->tsk, 0, func, context); ret = taskqueue_enqueue(pnfsioq, &pio->tsk); if (ret != 0) pio->inprog = 0; return (ret); } -static void -nfs_clean(struct prison *pr) -{ - - NFSD_CURVNET_SET(pr->pr_vnet); - mtx_destroy(&NFSD_VNET(nfsrv_nfsuserdsock).nr_mtx); - if (pr != &prison0) - free(NFSD_VNET(nfsstatsv1_p), M_TEMP); - /* Clean out the name<-->id cache. */ - nfsrv_cleanusergroup(); - NFSD_CURVNET_RESTORE(); -} - /* * Initialize everything that needs to be initialized for a vnet. */ static void nfs_vnetinit(const void *unused __unused) { 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); mtx_init(&NFSD_VNET(nfsrv_nfsuserdsock).nr_mtx, "nfsuserd", NULL, MTX_DEF); } VNET_SYSINIT(nfs_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, nfs_vnetinit, NULL); +static void +nfs_cleanup(void *unused __unused) +{ + + mtx_destroy(&NFSD_VNET(nfsrv_nfsuserdsock).nr_mtx); + if (!IS_DEFAULT_VNET(curvnet)) + free(NFSD_VNET(nfsstatsv1_p), M_TEMP); + /* Clean out the name<-->id cache. */ + nfsrv_cleanusergroup(); +} +VNET_SYSUNINIT(nfs_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, + nfs_cleanup, NULL); + extern int (*nfsd_call_nfscommon)(struct thread *, struct nfssvc_args *); /* * Called once to initialize data structures... */ static int nfscommon_modevent(module_t mod, int type, void *data) { int error = 0; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) goto out; newnfs_portinit(); mtx_init(&nfs_nameid_mutex, "nfs_nameid_mutex", NULL, MTX_DEF); mtx_init(&nfs_sockl_mutex, "nfs_sockl_mutex", NULL, MTX_DEF); mtx_init(&nfs_slock_mutex, "nfs_slock_mutex", NULL, MTX_DEF); mtx_init(&nfs_req_mutex, "nfs_req_mutex", NULL, MTX_DEF); mtx_init(&nfsrv_dslock_mtx, "nfs4ds", NULL, MTX_DEF); TAILQ_INIT(&nfsrv_devidhead); newnfs_init(); nfsd_call_nfscommon = nfssvc_nfscommon; loaded = 1; break; case MOD_UNLOAD: if (newnfs_numnfsd != 0 || NFSD_VNET(nfsrv_nfsuserd) != NOTRUNNING || nfs_numnfscbd != 0) { error = EBUSY; break; } nfsd_call_nfscommon = NULL; - nfs_clean(&prison0); /* and get rid of the mutexes */ mtx_destroy(&nfs_nameid_mutex); mtx_destroy(&newnfsd_mtx); mtx_destroy(&nfs_state_mutex); mtx_destroy(&nfs_clstate_mutex); mtx_destroy(&nfs_sockl_mutex); mtx_destroy(&nfs_slock_mutex); mtx_destroy(&nfs_req_mutex); mtx_destroy(&nfsrv_dslock_mtx); loaded = 0; break; default: error = EOPNOTSUPP; break; } out: NFSEXITCODE(error); return error; } static moduledata_t nfscommon_mod = { "nfscommon", nfscommon_modevent, NULL, }; DECLARE_MODULE(nfscommon, nfscommon_mod, SI_SUB_VFS, SI_ORDER_ANY); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfscommon, 1); MODULE_DEPEND(nfscommon, nfssvc, 1, 1, 1); MODULE_DEPEND(nfscommon, krpc, 1, 1, 1); diff --git a/sys/fs/nfsserver/nfs_nfsdcache.c b/sys/fs/nfsserver/nfs_nfsdcache.c index 2629c795802f..0eb623b6f5ee 100644 --- a/sys/fs/nfsserver/nfs_nfsdcache.c +++ b/sys/fs/nfsserver/nfs_nfsdcache.c @@ -1,1053 +1,1049 @@ /*- * 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 mtx nfsrc_udpmtx; 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 >= 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"); 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) \ (&NFSD_VNET(nfsrvudphashtbl)[nfsrc_hash(xid)]) #define NFSRCHASH(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 (&NFSD_VNET(nfsrchash_table)[nfsrc_hash(rp->rc_xid)].mtx); } /* * Initialize the server request cache list */ void nfsrvd_initcache(void) { int i; 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(&NFSD_VNET(nfsrvudphashtbl)[i]); LIST_INIT(&NFSD_VNET(nfsrchash_table)[i].tbl); LIST_INIT(&NFSD_VNET(nfsrcahash_table)[i].tbl); } 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(&NFSD_VNET(nfsrvudplru), rp, rc_lru); TAILQ_INSERT_TAIL(&NFSD_VNET(nfsrvudplru), rp, rc_lru); if (rp->rc_flag & RC_INPROG) { NFSD_VNET(nfsstatsv1_p)->srvcache_inproghits++; mtx_unlock(mutex); ret = RC_DROPIT; } else if (rp->rc_flag & RC_REPSTATUS) { /* * V2 only. */ NFSD_VNET(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) { NFSD_VNET(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; } } NFSD_VNET(nfsstatsv1_p)->srvcache_misses++; atomic_add_int(&NFSD_VNET(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(&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(&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) { NFSD_VNET(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) && 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(&NFSD_VNET(nfsrc_tcpsavedreplies), 1); if (NFSD_VNET(nfsrc_tcpsavedreplies) > NFSD_VNET(nfsstatsv1_p)->srvcache_tcppeak) NFSD_VNET(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) { NFSD_VNET(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. */ NFSD_VNET(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) { NFSD_VNET(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; } NFSD_VNET(nfsstatsv1_p)->srvcache_misses++; atomic_add_int(&NFSD_VNET(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(&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(&NFSD_VNET(nfsrc_tcpsavedreplies), -1); } free(rp, M_NFSRVCACHE); atomic_add_int(&NFSD_VNET(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(&NFSD_VNET(nfsrchash_table)[i].mtx); LIST_FOREACH_SAFE(rp, &NFSD_VNET(nfsrchash_table)[i].tbl, rc_hash, nextrp) nfsrc_freecache(rp); - mtx_unlock(&NFSD_VNET(nfsrchash_table)[i].mtx); } - mtx_lock(&nfsrc_udpmtx); for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { LIST_FOREACH_SAFE(rp, &NFSD_VNET(nfsrvudphashtbl)[i], rc_hash, nextrp) { nfsrc_freecache(rp); } } NFSD_VNET(nfsstatsv1_p)->srvcache_size = 0; - mtx_unlock(&nfsrc_udpmtx); 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 || NFSD_VNET(nfsrc_udpcachesize) >= (nfsrc_udphighwater + nfsrc_udphighwater / 2)) { mtx_lock(&nfsrc_udpmtx); udp_lasttrim = NFSD_MONOSEC; 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 || NFSD_VNET(nfsrc_udpcachesize) > nfsrc_udphighwater)) nfsrc_freecache(rp); } mtx_unlock(&nfsrc_udpmtx); } if (NFSD_MONOSEC != tcp_lasttrim || NFSD_VNET(nfsrc_tcpsavedreplies) >= nfsrc_tcphighwater) { force = nfsrc_tcphighwater / 4; if (force > 0 && 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(&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(&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(&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(&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_nfsdport.c b/sys/fs/nfsserver/nfs_nfsdport.c index 7708f0325494..cee5f96d7199 100644 --- a/sys/fs/nfsserver/nfs_nfsdport.c +++ b/sys/fs/nfsserver/nfs_nfsdport.c @@ -1,7237 +1,7233 @@ /*- * 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 int nfsrv_sessionhashsize; extern struct nfslayouthash *nfslayouthash; extern int nfsrv_layouthashsize; extern struct mtx nfsrv_dslock_mtx; extern int nfs_pnfsiothreads; extern volatile int nfsrv_devidcnt; extern int nfsrv_maxpnfsmirror; extern uint32_t nfs_srvmaxio; extern int nfs_bufpackets; extern u_long sb_max_adj; 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); NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); NFSDLOCKMUTEX; NFSSTATESPINLOCK; struct mtx nfsrc_udpmtx; struct mtx nfs_v4root_mutex; struct mtx nfsrv_dontlistlock_mtx; struct mtx nfsrv_recalllock_mtx; 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; 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_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 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_deallocatedsrpc(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"); 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 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. */ 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 = vfs_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 (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); 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) { uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); ndp->ni_cnd.cn_pnbuf = NULL; } /* * 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? */ NFSD_VNET(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? */ NFSD_VNET(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) { 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))) { 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); if (error) goto out; } else { 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 */ 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) { nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); vrele(ndp->ni_vp); error = EEXIST; goto out; } if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { nfsvno_relpathbuf(ndp); vput(ndp->ni_dvp); error = NFSERR_BADTYPE; goto out; } if (vtyp == VSOCK) { 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))) { 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); /* * 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) { 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); 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); 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); 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"); } nfsvno_relpathbuf(tondp); out1: 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, bool done_namei, struct nfsexstuff *exp, struct vnode **vpp) { struct vnode *vp = NULL; u_quad_t tempsize; struct nfsexstuff nes; struct thread *p = curthread; uint32_t oldrepstat; if (ndp->ni_vp == NULL) { /* * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be * set true, since cleanup after nfsvno_namei() is needed. */ oldrepstat = nd->nd_repstat; nd->nd_repstat = nfsrv_opencheck(clientid, stateidp, stp, NULL, nd, p, nd->nd_repstat); if (nd->nd_repstat != 0 && oldrepstat == 0) done_namei = true; } if (!nd->nd_repstat) { if (ndp->ni_vp == NULL) { 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 { 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 (done_namei) { KASSERT(create == NFSV4OPEN_CREATE, ("nfsvno_open: not create")); /* * done_namei is set when nfsvno_namei() has completed * successfully, but a subsequent error was set in * nd_repstat. As such, cleanup of the nfsvno_namei() * results is required. */ nfsvno_relpathbuf(ndp); 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; uint64_t *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 __unused; uint64_t *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; } 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_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 (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 (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(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export); if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 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); if ((error = namei(&nd)) != 0) goto out; error = nfsvno_getfh(nd.ni_vp, &fh, p); vrele(nd.ni_vp); if (!error) { NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; NFSBCOPY((caddr_t)&fh, NFSD_VNET(nfs_rootfh).nfsrvfh_data, sizeof (fhandle_t)); 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) { NFSD_LOCK(); if (NFSD_VNET(nfsrv_mntinited)) { NFSD_UNLOCK(); return; } NFSD_VNET(nfsrv_mntinited) = true; nfsrvd_init(0); NFSD_UNLOCK(); 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) { 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(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); } /* * 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(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; 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 && NFSD_VNET(nfsrv_numnfsd) != 0) error = EPERM; if (!error) { 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_VNET(nfsd_master_proc) = procp; PROC_UNLOCK(procp); } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { NFSLOCKV4ROOTMUTEX(); if (!NFSD_VNET(nfsrv_suspend_nfsd)) { /* Lock out all nfsd threads */ do { 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 (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_VNET(nfsd_master_proc) != NULL) { procp = pfind(nfsd_master_pid); /* Try to make sure it is the correct process. */ 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_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 | NOCACHE); nfsvno_setpathbuf(&named, &bufp, &hashp); named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 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_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 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 if (ioproc == NFSPROC_DEALLOCATE) error = nfsrv_deallocatedsrpc(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); } /* * Do a deallocate 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 nfsrvdeallocatedsdorpc { 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_deallocatedsdorpc(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_DEALLOCATE, 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_deallocatedsdorpc: len=%jd\n", (intmax_t)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); *tl = txdr_unsigned(NFSV4OP_GETATTR); 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_deallocatedsdorpc: aft deallocaterpc=%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_deallocatedsdorpc: 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); 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_deallocatedsdorpc: aft loadattr=%d\n", error); nfsmout: m_freem(nd->nd_mrep); free(nd, M_TEMP); NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); return (error); } /* * Start up the thread that will execute nfsrv_deallocatedsdorpc(). */ static void start_deallocatedsdorpc(void *arg, int pending) { struct nfsrvdeallocatedsdorpc *drpc; drpc = (struct nfsrvdeallocatedsdorpc *)arg; drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, drpc->len, NULL, drpc->cred, drpc->p); drpc->done = 1; NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); } static int nfsrv_deallocatedsrpc(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 nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; struct nfsvattr na; int error, i, ret, timo; NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); drpc = NULL; if (mirrorcnt > 1) tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, M_WAITOK); /* * Do the deallocate 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_deallocatedsdorpc, tdrpc); NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", ret); } if (ret != 0) { ret = nfsrv_deallocatedsdorpc(*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_deallocatedsdorpc(*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_deallocatedsrpc: 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_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 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); } /* * Deallocate vnode op call. */ int nfsvno_deallocate(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_deallocate vp"); /* * Attempt to deallocate on a DS file. A return of ENOENT implies * there is no DS file to deallocate on. */ error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, NULL, NULL, NULL, NULL, &len, 0, NULL); if (error != ENOENT) return (error); /* * Do the actual VOP_DEALLOCATE(), looping so long as * progress is being made, to achieve completion. */ do { olen = len; error = VOP_DEALLOCATE(vp, &off, &len, 0, 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) +nfsrv_cleanup(const void *unused __unused) { 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; - NFSD_CURVNET_RESTORE(); } +VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, + nfsrv_cleanup, NULL); 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; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) goto out; newnfs_portinit(); mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); #ifdef VV_DISABLEDELEG vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; #endif nfsd_call_nfsd = nfssvc_nfsd; 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; - nfsrv_cleanup(&prison0); mtx_destroy(&nfsrc_udpmtx); mtx_destroy(&nfs_v4root_mutex); mtx_destroy(&nfsrv_dontlistlock_mtx); mtx_destroy(&nfsrv_recalllock_mtx); if (nfslayouthash != NULL) { for (i = 0; i < nfsrv_layouthashsize; i++) mtx_destroy(&nfslayouthash[i].mtx); free(nfslayouthash, 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/rpc/rpcsec_gss/svc_rpcsec_gss.c b/sys/rpc/rpcsec_gss/svc_rpcsec_gss.c index dc850996a592..03e185bd912b 100644 --- a/sys/rpc/rpcsec_gss/svc_rpcsec_gss.c +++ b/sys/rpc/rpcsec_gss/svc_rpcsec_gss.c @@ -1,1598 +1,1611 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * Copyright (c) 1990 The Regents of the University of California. * * Copyright (c) 2008 Doug Rabson * 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. */ /* svc_rpcsec_gss.c Copyright (c) 2000 The Regents of the University of Michigan. All rights reserved. Copyright (c) 2000 Dug Song . All rights reserved, all wrongs reversed. 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 ``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. $Id: svc_auth_gss.c,v 1.27 2002/01/15 15:43:00 andros Exp $ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "rpcsec_gss_int.h" static bool_t svc_rpc_gss_wrap(SVCAUTH *, struct mbuf **); static bool_t svc_rpc_gss_unwrap(SVCAUTH *, struct mbuf **); static void svc_rpc_gss_release(SVCAUTH *); static enum auth_stat svc_rpc_gss(struct svc_req *, struct rpc_msg *); static int rpc_gss_svc_getcred(struct svc_req *, struct ucred **, int *); static const struct svc_auth_ops svc_auth_gss_ops = { .svc_ah_wrap = svc_rpc_gss_wrap, .svc_ah_unwrap = svc_rpc_gss_unwrap, .svc_ah_release = svc_rpc_gss_release, }; struct sx svc_rpc_gss_lock; struct svc_rpc_gss_callback { SLIST_ENTRY(svc_rpc_gss_callback) cb_link; rpc_gss_callback_t cb_callback; }; SLIST_HEAD(svc_rpc_gss_callback_list, svc_rpc_gss_callback); KGSS_VNET_DEFINE_STATIC(struct svc_rpc_gss_callback_list, svc_rpc_gss_callbacks) = SLIST_HEAD_INITIALIZER(svc_rpc_gss_callbacks); struct svc_rpc_gss_svc_name { SLIST_ENTRY(svc_rpc_gss_svc_name) sn_link; char *sn_principal; gss_OID sn_mech; u_int sn_req_time; gss_cred_id_t sn_cred; u_int sn_program; u_int sn_version; }; SLIST_HEAD(svc_rpc_gss_svc_name_list, svc_rpc_gss_svc_name); KGSS_VNET_DEFINE_STATIC(struct svc_rpc_gss_svc_name_list, svc_rpc_gss_svc_names) = SLIST_HEAD_INITIALIZER(svc_rpc_gss_svc_names); enum svc_rpc_gss_client_state { CLIENT_NEW, /* still authenticating */ CLIENT_ESTABLISHED, /* context established */ CLIENT_STALE /* garbage to collect */ }; #define SVC_RPC_GSS_SEQWINDOW 128 struct svc_rpc_gss_clientid { unsigned long ci_hostid; uint32_t ci_boottime; uint32_t ci_id; }; struct svc_rpc_gss_client { TAILQ_ENTRY(svc_rpc_gss_client) cl_link; TAILQ_ENTRY(svc_rpc_gss_client) cl_alllink; volatile u_int cl_refs; struct sx cl_lock; struct svc_rpc_gss_clientid cl_id; time_t cl_expiration; /* when to gc */ enum svc_rpc_gss_client_state cl_state; /* client state */ bool_t cl_locked; /* fixed service+qop */ gss_ctx_id_t cl_ctx; /* context id */ gss_cred_id_t cl_creds; /* delegated creds */ gss_name_t cl_cname; /* client name */ struct svc_rpc_gss_svc_name *cl_sname; /* server name used */ rpc_gss_rawcred_t cl_rawcred; /* raw credentials */ rpc_gss_ucred_t cl_ucred; /* unix-style credentials */ struct ucred *cl_cred; /* kernel-style credentials */ int cl_rpcflavor; /* RPC pseudo sec flavor */ bool_t cl_done_callback; /* TRUE after call */ void *cl_cookie; /* user cookie from callback */ gid_t cl_gid_storage[NGROUPS]; gss_OID cl_mech; /* mechanism */ gss_qop_t cl_qop; /* quality of protection */ uint32_t cl_seqlast; /* sequence window origin */ uint32_t cl_seqmask[SVC_RPC_GSS_SEQWINDOW/32]; /* bitmask of seqnums */ }; TAILQ_HEAD(svc_rpc_gss_client_list, svc_rpc_gss_client); /* * This structure holds enough information to unwrap arguments or wrap * results for a given request. We use the rq_clntcred area for this * (which is a per-request buffer). */ struct svc_rpc_gss_cookedcred { struct svc_rpc_gss_client *cc_client; rpc_gss_service_t cc_service; uint32_t cc_seq; }; #define CLIENT_HASH_SIZE 256 #define CLIENT_MAX 1024 u_int svc_rpc_gss_client_max = CLIENT_MAX; u_int svc_rpc_gss_client_hash_size = CLIENT_HASH_SIZE; SYSCTL_NODE(_kern, OID_AUTO, rpc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "RPC"); SYSCTL_NODE(_kern_rpc, OID_AUTO, gss, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "GSS"); SYSCTL_UINT(_kern_rpc_gss, OID_AUTO, client_max, CTLFLAG_RW, &svc_rpc_gss_client_max, 0, "Max number of rpc-gss clients"); SYSCTL_UINT(_kern_rpc_gss, OID_AUTO, client_hash, CTLFLAG_RDTUN, &svc_rpc_gss_client_hash_size, 0, "Size of rpc-gss client hash table"); static u_int svc_rpc_gss_lifetime_max = 0; SYSCTL_UINT(_kern_rpc_gss, OID_AUTO, lifetime_max, CTLFLAG_RW, &svc_rpc_gss_lifetime_max, 0, "Maximum lifetime (seconds) of rpc-gss clients"); static u_int svc_rpc_gss_client_count; SYSCTL_UINT(_kern_rpc_gss, OID_AUTO, client_count, CTLFLAG_RD, &svc_rpc_gss_client_count, 0, "Number of rpc-gss clients"); KGSS_VNET_DEFINE(struct svc_rpc_gss_client_list *, svc_rpc_gss_client_hash); KGSS_VNET_DEFINE(struct svc_rpc_gss_client_list, svc_rpc_gss_clients); KGSS_VNET_DEFINE_STATIC(uint32_t, svc_rpc_gss_next_clientid) = 1; static void -svc_rpc_gss_init(void *arg) +svc_rpc_gss_init(void *unused __unused) { int i; KGSS_VNET(svc_rpc_gss_client_hash) = mem_alloc( sizeof(struct svc_rpc_gss_client_list) * svc_rpc_gss_client_hash_size); for (i = 0; i < svc_rpc_gss_client_hash_size; i++) TAILQ_INIT(&KGSS_VNET(svc_rpc_gss_client_hash)[i]); TAILQ_INIT(&KGSS_VNET(svc_rpc_gss_clients)); if (IS_DEFAULT_VNET(curvnet)) { svc_auth_reg(RPCSEC_GSS, svc_rpc_gss, rpc_gss_svc_getcred); sx_init(&svc_rpc_gss_lock, "gsslock"); } } VNET_SYSINIT(svc_rpc_gss_init, SI_SUB_VNET_DONE, SI_ORDER_ANY, svc_rpc_gss_init, NULL); +static void +svc_rpc_gss_cleanup(void *unused __unused) +{ + + mem_free(KGSS_VNET(svc_rpc_gss_client_hash), + sizeof(struct svc_rpc_gss_client_list) * + svc_rpc_gss_client_hash_size); + if (IS_DEFAULT_VNET(curvnet)) + sx_destroy(&svc_rpc_gss_lock); +} +VNET_SYSUNINIT(svc_rpc_gss_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, + svc_rpc_gss_cleanup, NULL); + bool_t rpc_gss_set_callback(rpc_gss_callback_t *cb) { struct svc_rpc_gss_callback *scb; scb = mem_alloc(sizeof(struct svc_rpc_gss_callback)); if (!scb) { _rpc_gss_set_error(RPC_GSS_ER_SYSTEMERROR, ENOMEM); return (FALSE); } scb->cb_callback = *cb; sx_xlock(&svc_rpc_gss_lock); SLIST_INSERT_HEAD(&KGSS_VNET(svc_rpc_gss_callbacks), scb, cb_link); sx_xunlock(&svc_rpc_gss_lock); return (TRUE); } void rpc_gss_clear_callback(rpc_gss_callback_t *cb) { struct svc_rpc_gss_callback *scb; sx_xlock(&svc_rpc_gss_lock); SLIST_FOREACH(scb, &KGSS_VNET(svc_rpc_gss_callbacks), cb_link) { if (scb->cb_callback.program == cb->program && scb->cb_callback.version == cb->version && scb->cb_callback.callback == cb->callback) { SLIST_REMOVE(&KGSS_VNET(svc_rpc_gss_callbacks), scb, svc_rpc_gss_callback, cb_link); sx_xunlock(&svc_rpc_gss_lock); mem_free(scb, sizeof(*scb)); return; } } sx_xunlock(&svc_rpc_gss_lock); } static bool_t rpc_gss_acquire_svc_cred(struct svc_rpc_gss_svc_name *sname) { OM_uint32 maj_stat, min_stat; gss_buffer_desc namebuf; gss_name_t name; gss_OID_set_desc oid_set; oid_set.count = 1; oid_set.elements = sname->sn_mech; namebuf.value = (void *) sname->sn_principal; namebuf.length = strlen(sname->sn_principal); maj_stat = gss_import_name(&min_stat, &namebuf, GSS_C_NT_HOSTBASED_SERVICE, &name); if (maj_stat != GSS_S_COMPLETE) return (FALSE); if (sname->sn_cred != GSS_C_NO_CREDENTIAL) gss_release_cred(&min_stat, &sname->sn_cred); maj_stat = gss_acquire_cred(&min_stat, name, sname->sn_req_time, &oid_set, GSS_C_ACCEPT, &sname->sn_cred, NULL, NULL); if (maj_stat != GSS_S_COMPLETE) { gss_release_name(&min_stat, &name); return (FALSE); } gss_release_name(&min_stat, &name); return (TRUE); } bool_t rpc_gss_set_svc_name(const char *principal, const char *mechanism, u_int req_time, u_int program, u_int version) { struct svc_rpc_gss_svc_name *sname; gss_OID mech_oid; if (!rpc_gss_mech_to_oid(mechanism, &mech_oid)) return (FALSE); sname = mem_alloc(sizeof(*sname)); if (!sname) return (FALSE); sname->sn_principal = strdup(principal, M_RPC); sname->sn_mech = mech_oid; sname->sn_req_time = req_time; sname->sn_cred = GSS_C_NO_CREDENTIAL; sname->sn_program = program; sname->sn_version = version; if (!rpc_gss_acquire_svc_cred(sname)) { free(sname->sn_principal, M_RPC); mem_free(sname, sizeof(*sname)); return (FALSE); } sx_xlock(&svc_rpc_gss_lock); SLIST_INSERT_HEAD(&KGSS_VNET(svc_rpc_gss_svc_names), sname, sn_link); sx_xunlock(&svc_rpc_gss_lock); return (TRUE); } void rpc_gss_clear_svc_name(u_int program, u_int version) { OM_uint32 min_stat; struct svc_rpc_gss_svc_name *sname; sx_xlock(&svc_rpc_gss_lock); SLIST_FOREACH(sname, &KGSS_VNET(svc_rpc_gss_svc_names), sn_link) { if (sname->sn_program == program && sname->sn_version == version) { SLIST_REMOVE(&KGSS_VNET(svc_rpc_gss_svc_names), sname, svc_rpc_gss_svc_name, sn_link); sx_xunlock(&svc_rpc_gss_lock); gss_release_cred(&min_stat, &sname->sn_cred); free(sname->sn_principal, M_RPC); mem_free(sname, sizeof(*sname)); return; } } sx_xunlock(&svc_rpc_gss_lock); } bool_t rpc_gss_get_principal_name(rpc_gss_principal_t *principal, const char *mech, const char *name, const char *node, const char *domain) { OM_uint32 maj_stat, min_stat; gss_OID mech_oid; size_t namelen; gss_buffer_desc buf; gss_name_t gss_name, gss_mech_name; rpc_gss_principal_t result; if (!rpc_gss_mech_to_oid(mech, &mech_oid)) return (FALSE); /* * Construct a gss_buffer containing the full name formatted * as "name/node@domain" where node and domain are optional. */ namelen = strlen(name) + 1; if (node) { namelen += strlen(node) + 1; } if (domain) { namelen += strlen(domain) + 1; } buf.value = mem_alloc(namelen); buf.length = namelen; strcpy((char *) buf.value, name); if (node) { strcat((char *) buf.value, "/"); strcat((char *) buf.value, node); } if (domain) { strcat((char *) buf.value, "@"); strcat((char *) buf.value, domain); } /* * Convert that to a gss_name_t and then convert that to a * mechanism name in the selected mechanism. */ maj_stat = gss_import_name(&min_stat, &buf, GSS_C_NT_USER_NAME, &gss_name); mem_free(buf.value, buf.length); if (maj_stat != GSS_S_COMPLETE) { rpc_gss_log_status("gss_import_name", mech_oid, maj_stat, min_stat); return (FALSE); } maj_stat = gss_canonicalize_name(&min_stat, gss_name, mech_oid, &gss_mech_name); if (maj_stat != GSS_S_COMPLETE) { rpc_gss_log_status("gss_canonicalize_name", mech_oid, maj_stat, min_stat); gss_release_name(&min_stat, &gss_name); return (FALSE); } gss_release_name(&min_stat, &gss_name); /* * Export the mechanism name and use that to construct the * rpc_gss_principal_t result. */ maj_stat = gss_export_name(&min_stat, gss_mech_name, &buf); if (maj_stat != GSS_S_COMPLETE) { rpc_gss_log_status("gss_export_name", mech_oid, maj_stat, min_stat); gss_release_name(&min_stat, &gss_mech_name); return (FALSE); } gss_release_name(&min_stat, &gss_mech_name); result = mem_alloc(sizeof(int) + buf.length); if (!result) { gss_release_buffer(&min_stat, &buf); return (FALSE); } result->len = buf.length; memcpy(result->name, buf.value, buf.length); gss_release_buffer(&min_stat, &buf); *principal = result; return (TRUE); } bool_t rpc_gss_getcred(struct svc_req *req, rpc_gss_rawcred_t **rcred, rpc_gss_ucred_t **ucred, void **cookie) { struct svc_rpc_gss_cookedcred *cc; struct svc_rpc_gss_client *client; if (req->rq_cred.oa_flavor != RPCSEC_GSS) return (FALSE); cc = req->rq_clntcred; client = cc->cc_client; if (rcred) *rcred = &client->cl_rawcred; if (ucred) *ucred = &client->cl_ucred; if (cookie) *cookie = client->cl_cookie; return (TRUE); } /* * This simpler interface is used by svc_getcred to copy the cred data * into a kernel cred structure. */ static int rpc_gss_svc_getcred(struct svc_req *req, struct ucred **crp, int *flavorp) { struct ucred *cr; struct svc_rpc_gss_cookedcred *cc; struct svc_rpc_gss_client *client; rpc_gss_ucred_t *uc; if (req->rq_cred.oa_flavor != RPCSEC_GSS) return (FALSE); cc = req->rq_clntcred; client = cc->cc_client; if (flavorp) *flavorp = client->cl_rpcflavor; if (client->cl_cred) { *crp = crhold(client->cl_cred); return (TRUE); } uc = &client->cl_ucred; cr = client->cl_cred = crget(); cr->cr_uid = cr->cr_ruid = cr->cr_svuid = uc->uid; cr->cr_rgid = cr->cr_svgid = uc->gid; crsetgroups(cr, uc->gidlen, uc->gidlist); cr->cr_prison = curthread->td_ucred->cr_prison; prison_hold(cr->cr_prison); *crp = crhold(cr); return (TRUE); } int rpc_gss_svc_max_data_length(struct svc_req *req, int max_tp_unit_len) { struct svc_rpc_gss_cookedcred *cc = req->rq_clntcred; struct svc_rpc_gss_client *client = cc->cc_client; int want_conf; OM_uint32 max; OM_uint32 maj_stat, min_stat; int result; switch (client->cl_rawcred.service) { case rpc_gss_svc_none: return (max_tp_unit_len); break; case rpc_gss_svc_default: case rpc_gss_svc_integrity: want_conf = FALSE; break; case rpc_gss_svc_privacy: want_conf = TRUE; break; default: return (0); } maj_stat = gss_wrap_size_limit(&min_stat, client->cl_ctx, want_conf, client->cl_qop, max_tp_unit_len, &max); if (maj_stat == GSS_S_COMPLETE) { result = (int) max; if (result < 0) result = 0; return (result); } else { rpc_gss_log_status("gss_wrap_size_limit", client->cl_mech, maj_stat, min_stat); return (0); } } static struct svc_rpc_gss_client * svc_rpc_gss_find_client(struct svc_rpc_gss_clientid *id) { struct svc_rpc_gss_client *client; struct svc_rpc_gss_client_list *list; struct timeval boottime; unsigned long hostid; rpc_gss_log_debug("in svc_rpc_gss_find_client(%d)", id->ci_id); getcredhostid(curthread->td_ucred, &hostid); getboottime(&boottime); if (id->ci_hostid != hostid || id->ci_boottime != boottime.tv_sec) return (NULL); list = &KGSS_VNET(svc_rpc_gss_client_hash) [id->ci_id % svc_rpc_gss_client_hash_size]; sx_xlock(&svc_rpc_gss_lock); TAILQ_FOREACH(client, list, cl_link) { if (client->cl_id.ci_id == id->ci_id) { /* * Move this client to the front of the LRU * list. */ TAILQ_REMOVE(&KGSS_VNET(svc_rpc_gss_clients), client, cl_alllink); TAILQ_INSERT_HEAD(&KGSS_VNET(svc_rpc_gss_clients), client, cl_alllink); refcount_acquire(&client->cl_refs); break; } } sx_xunlock(&svc_rpc_gss_lock); return (client); } static struct svc_rpc_gss_client * svc_rpc_gss_create_client(void) { struct svc_rpc_gss_client *client; struct svc_rpc_gss_client_list *list; struct timeval boottime; unsigned long hostid; rpc_gss_log_debug("in svc_rpc_gss_create_client()"); client = mem_alloc(sizeof(struct svc_rpc_gss_client)); memset(client, 0, sizeof(struct svc_rpc_gss_client)); /* * Set the initial value of cl_refs to two. One for the caller * and the other to hold onto the client structure until it expires. */ refcount_init(&client->cl_refs, 2); sx_init(&client->cl_lock, "GSS-client"); getcredhostid(curthread->td_ucred, &hostid); client->cl_id.ci_hostid = hostid; getboottime(&boottime); client->cl_id.ci_boottime = boottime.tv_sec; client->cl_id.ci_id = KGSS_VNET(svc_rpc_gss_next_clientid)++; /* * Start the client off with a short expiration time. We will * try to get a saner value from the client creds later. */ client->cl_state = CLIENT_NEW; client->cl_locked = FALSE; client->cl_expiration = time_uptime + 5*60; list = &KGSS_VNET(svc_rpc_gss_client_hash) [client->cl_id.ci_id % svc_rpc_gss_client_hash_size]; sx_xlock(&svc_rpc_gss_lock); TAILQ_INSERT_HEAD(list, client, cl_link); TAILQ_INSERT_HEAD(&KGSS_VNET(svc_rpc_gss_clients), client, cl_alllink); svc_rpc_gss_client_count++; sx_xunlock(&svc_rpc_gss_lock); return (client); } static void svc_rpc_gss_destroy_client(struct svc_rpc_gss_client *client) { OM_uint32 min_stat; rpc_gss_log_debug("in svc_rpc_gss_destroy_client()"); if (client->cl_ctx) gss_delete_sec_context(&min_stat, &client->cl_ctx, GSS_C_NO_BUFFER); if (client->cl_cname) gss_release_name(&min_stat, &client->cl_cname); if (client->cl_rawcred.client_principal) mem_free(client->cl_rawcred.client_principal, sizeof(*client->cl_rawcred.client_principal) + client->cl_rawcred.client_principal->len); if (client->cl_cred) crfree(client->cl_cred); sx_destroy(&client->cl_lock); mem_free(client, sizeof(*client)); } /* * Drop a reference to a client and free it if that was the last reference. */ static void svc_rpc_gss_release_client(struct svc_rpc_gss_client *client) { if (!refcount_release(&client->cl_refs)) return; svc_rpc_gss_destroy_client(client); } /* * Remove a client from our global lists. * Must be called with svc_rpc_gss_lock held. */ static void svc_rpc_gss_forget_client_locked(struct svc_rpc_gss_client *client) { struct svc_rpc_gss_client_list *list; sx_assert(&svc_rpc_gss_lock, SX_XLOCKED); list = &KGSS_VNET(svc_rpc_gss_client_hash) [client->cl_id.ci_id % svc_rpc_gss_client_hash_size]; TAILQ_REMOVE(list, client, cl_link); TAILQ_REMOVE(&KGSS_VNET(svc_rpc_gss_clients), client, cl_alllink); svc_rpc_gss_client_count--; } /* * Remove a client from our global lists and free it if we can. */ static void svc_rpc_gss_forget_client(struct svc_rpc_gss_client *client) { struct svc_rpc_gss_client_list *list; struct svc_rpc_gss_client *tclient; list = &KGSS_VNET(svc_rpc_gss_client_hash) [client->cl_id.ci_id % svc_rpc_gss_client_hash_size]; sx_xlock(&svc_rpc_gss_lock); TAILQ_FOREACH(tclient, list, cl_link) { /* * Make sure this client has not already been removed * from the lists by svc_rpc_gss_forget_client() or * svc_rpc_gss_forget_client_locked(). */ if (client == tclient) { svc_rpc_gss_forget_client_locked(client); sx_xunlock(&svc_rpc_gss_lock); svc_rpc_gss_release_client(client); return; } } sx_xunlock(&svc_rpc_gss_lock); } static void svc_rpc_gss_timeout_clients(void) { struct svc_rpc_gss_client *client; time_t now = time_uptime; rpc_gss_log_debug("in svc_rpc_gss_timeout_clients()"); /* * First enforce the max client limit. We keep * svc_rpc_gss_clients in LRU order. */ sx_xlock(&svc_rpc_gss_lock); client = TAILQ_LAST(&KGSS_VNET(svc_rpc_gss_clients), svc_rpc_gss_client_list); while (svc_rpc_gss_client_count > svc_rpc_gss_client_max && client != NULL) { svc_rpc_gss_forget_client_locked(client); sx_xunlock(&svc_rpc_gss_lock); svc_rpc_gss_release_client(client); sx_xlock(&svc_rpc_gss_lock); client = TAILQ_LAST(&KGSS_VNET(svc_rpc_gss_clients), svc_rpc_gss_client_list); } again: TAILQ_FOREACH(client, &KGSS_VNET(svc_rpc_gss_clients), cl_alllink) { if (client->cl_state == CLIENT_STALE || now > client->cl_expiration) { svc_rpc_gss_forget_client_locked(client); sx_xunlock(&svc_rpc_gss_lock); rpc_gss_log_debug("expiring client %p", client); svc_rpc_gss_release_client(client); sx_xlock(&svc_rpc_gss_lock); goto again; } } sx_xunlock(&svc_rpc_gss_lock); } #ifdef DEBUG /* * OID<->string routines. These are uuuuugly. */ static OM_uint32 gss_oid_to_str(OM_uint32 *minor_status, gss_OID oid, gss_buffer_t oid_str) { char numstr[128]; unsigned long number; int numshift; size_t string_length; size_t i; unsigned char *cp; char *bp; /* Decoded according to krb5/gssapi_krb5.c */ /* First determine the size of the string */ string_length = 0; number = 0; numshift = 0; cp = (unsigned char *) oid->elements; number = (unsigned long) cp[0]; sprintf(numstr, "%ld ", number/40); string_length += strlen(numstr); sprintf(numstr, "%ld ", number%40); string_length += strlen(numstr); for (i=1; ilength; i++) { if ( (size_t) (numshift+7) < (sizeof(unsigned long)*8)) { number = (number << 7) | (cp[i] & 0x7f); numshift += 7; } else { *minor_status = 0; return(GSS_S_FAILURE); } if ((cp[i] & 0x80) == 0) { sprintf(numstr, "%ld ", number); string_length += strlen(numstr); number = 0; numshift = 0; } } /* * If we get here, we've calculated the length of "n n n ... n ". Add 4 * here for "{ " and "}\0". */ string_length += 4; if ((bp = malloc(string_length, M_GSSAPI, M_WAITOK | M_ZERO))) { strcpy(bp, "{ "); number = (unsigned long) cp[0]; sprintf(numstr, "%ld ", number/40); strcat(bp, numstr); sprintf(numstr, "%ld ", number%40); strcat(bp, numstr); number = 0; cp = (unsigned char *) oid->elements; for (i=1; ilength; i++) { number = (number << 7) | (cp[i] & 0x7f); if ((cp[i] & 0x80) == 0) { sprintf(numstr, "%ld ", number); strcat(bp, numstr); number = 0; } } strcat(bp, "}"); oid_str->length = strlen(bp)+1; oid_str->value = (void *) bp; *minor_status = 0; return(GSS_S_COMPLETE); } *minor_status = 0; return(GSS_S_FAILURE); } #endif static void svc_rpc_gss_build_ucred(struct svc_rpc_gss_client *client, const gss_name_t name) { OM_uint32 maj_stat, min_stat; rpc_gss_ucred_t *uc = &client->cl_ucred; int numgroups; uc->uid = 65534; uc->gid = 65534; uc->gidlist = client->cl_gid_storage; numgroups = NGROUPS; maj_stat = gss_pname_to_unix_cred(&min_stat, name, client->cl_mech, &uc->uid, &uc->gid, &numgroups, &uc->gidlist[0]); if (GSS_ERROR(maj_stat)) uc->gidlen = 0; else uc->gidlen = numgroups; } static void svc_rpc_gss_set_flavor(struct svc_rpc_gss_client *client) { static gss_OID_desc krb5_mech_oid = {9, (void *) "\x2a\x86\x48\x86\xf7\x12\x01\x02\x02" }; /* * Attempt to translate mech type and service into a * 'pseudo flavor'. Hardwire in krb5 support for now. */ if (kgss_oid_equal(client->cl_mech, &krb5_mech_oid)) { switch (client->cl_rawcred.service) { case rpc_gss_svc_default: case rpc_gss_svc_none: client->cl_rpcflavor = RPCSEC_GSS_KRB5; break; case rpc_gss_svc_integrity: client->cl_rpcflavor = RPCSEC_GSS_KRB5I; break; case rpc_gss_svc_privacy: client->cl_rpcflavor = RPCSEC_GSS_KRB5P; break; } } else { client->cl_rpcflavor = RPCSEC_GSS; } } static bool_t svc_rpc_gss_accept_sec_context(struct svc_rpc_gss_client *client, struct svc_req *rqst, struct rpc_gss_init_res *gr, struct rpc_gss_cred *gc) { gss_buffer_desc recv_tok; gss_OID mech; OM_uint32 maj_stat = 0, min_stat = 0, ret_flags; OM_uint32 cred_lifetime; struct svc_rpc_gss_svc_name *sname; rpc_gss_log_debug("in svc_rpc_gss_accept_context()"); /* Deserialize arguments. */ memset(&recv_tok, 0, sizeof(recv_tok)); if (!svc_getargs(rqst, (xdrproc_t) xdr_gss_buffer_desc, (caddr_t) &recv_tok)) { client->cl_state = CLIENT_STALE; return (FALSE); } /* * First time round, try all the server names we have until * one matches. Afterwards, stick with that one. */ sx_xlock(&svc_rpc_gss_lock); if (!client->cl_sname) { SLIST_FOREACH(sname, &KGSS_VNET(svc_rpc_gss_svc_names), sn_link) { if (sname->sn_program == rqst->rq_prog && sname->sn_version == rqst->rq_vers) { retry: gr->gr_major = gss_accept_sec_context( &gr->gr_minor, &client->cl_ctx, sname->sn_cred, &recv_tok, GSS_C_NO_CHANNEL_BINDINGS, &client->cl_cname, &mech, &gr->gr_token, &ret_flags, &cred_lifetime, &client->cl_creds); if (gr->gr_major == GSS_S_CREDENTIALS_EXPIRED) { /* * Either our creds really did * expire or gssd was * restarted. */ if (rpc_gss_acquire_svc_cred(sname)) goto retry; } client->cl_sname = sname; break; } } if (!sname) { xdr_free((xdrproc_t) xdr_gss_buffer_desc, (char *) &recv_tok); sx_xunlock(&svc_rpc_gss_lock); return (FALSE); } } else { gr->gr_major = gss_accept_sec_context( &gr->gr_minor, &client->cl_ctx, client->cl_sname->sn_cred, &recv_tok, GSS_C_NO_CHANNEL_BINDINGS, &client->cl_cname, &mech, &gr->gr_token, &ret_flags, &cred_lifetime, NULL); } sx_xunlock(&svc_rpc_gss_lock); xdr_free((xdrproc_t) xdr_gss_buffer_desc, (char *) &recv_tok); /* * If we get an error from gss_accept_sec_context, send the * reply anyway so that the client gets a chance to see what * is wrong. */ if (gr->gr_major != GSS_S_COMPLETE && gr->gr_major != GSS_S_CONTINUE_NEEDED) { rpc_gss_log_status("accept_sec_context", client->cl_mech, gr->gr_major, gr->gr_minor); client->cl_state = CLIENT_STALE; return (TRUE); } gr->gr_handle.value = &client->cl_id; gr->gr_handle.length = sizeof(client->cl_id); gr->gr_win = SVC_RPC_GSS_SEQWINDOW; /* Save client info. */ client->cl_mech = mech; client->cl_qop = GSS_C_QOP_DEFAULT; client->cl_done_callback = FALSE; if (gr->gr_major == GSS_S_COMPLETE) { gss_buffer_desc export_name; /* * Change client expiration time to be near when the * client creds expire (or 24 hours if we can't figure * that out). */ if (cred_lifetime == GSS_C_INDEFINITE) cred_lifetime = 24*60*60; /* * Cap cred_lifetime if sysctl kern.rpc.gss.lifetime_max is set. */ if (svc_rpc_gss_lifetime_max > 0 && cred_lifetime > svc_rpc_gss_lifetime_max) cred_lifetime = svc_rpc_gss_lifetime_max; client->cl_expiration = time_uptime + cred_lifetime; /* * Fill in cred details in the rawcred structure. */ client->cl_rawcred.version = RPCSEC_GSS_VERSION; rpc_gss_oid_to_mech(mech, &client->cl_rawcred.mechanism); maj_stat = gss_export_name(&min_stat, client->cl_cname, &export_name); if (maj_stat != GSS_S_COMPLETE) { rpc_gss_log_status("gss_export_name", client->cl_mech, maj_stat, min_stat); return (FALSE); } client->cl_rawcred.client_principal = mem_alloc(sizeof(*client->cl_rawcred.client_principal) + export_name.length); client->cl_rawcred.client_principal->len = export_name.length; memcpy(client->cl_rawcred.client_principal->name, export_name.value, export_name.length); gss_release_buffer(&min_stat, &export_name); client->cl_rawcred.svc_principal = client->cl_sname->sn_principal; client->cl_rawcred.service = gc->gc_svc; /* * Use gss_pname_to_uid to map to unix creds. For * kerberos5, this uses krb5_aname_to_localname. */ svc_rpc_gss_build_ucred(client, client->cl_cname); svc_rpc_gss_set_flavor(client); gss_release_name(&min_stat, &client->cl_cname); #ifdef DEBUG { gss_buffer_desc mechname; gss_oid_to_str(&min_stat, mech, &mechname); rpc_gss_log_debug("accepted context for %s with " "", client->cl_rawcred.client_principal->name, mechname.length, (char *)mechname.value, client->cl_qop, client->cl_rawcred.service); gss_release_buffer(&min_stat, &mechname); } #endif /* DEBUG */ } return (TRUE); } static bool_t svc_rpc_gss_validate(struct svc_rpc_gss_client *client, struct rpc_msg *msg, gss_qop_t *qop, rpc_gss_proc_t gcproc) { struct opaque_auth *oa; gss_buffer_desc rpcbuf, checksum; OM_uint32 maj_stat, min_stat; gss_qop_t qop_state; int32_t rpchdr[128 / sizeof(int32_t)]; int32_t *buf; rpc_gss_log_debug("in svc_rpc_gss_validate()"); memset(rpchdr, 0, sizeof(rpchdr)); /* Reconstruct RPC header for signing (from xdr_callmsg). */ buf = rpchdr; IXDR_PUT_LONG(buf, msg->rm_xid); IXDR_PUT_ENUM(buf, msg->rm_direction); IXDR_PUT_LONG(buf, msg->rm_call.cb_rpcvers); IXDR_PUT_LONG(buf, msg->rm_call.cb_prog); IXDR_PUT_LONG(buf, msg->rm_call.cb_vers); IXDR_PUT_LONG(buf, msg->rm_call.cb_proc); oa = &msg->rm_call.cb_cred; IXDR_PUT_ENUM(buf, oa->oa_flavor); IXDR_PUT_LONG(buf, oa->oa_length); if (oa->oa_length) { memcpy((caddr_t)buf, oa->oa_base, oa->oa_length); buf += RNDUP(oa->oa_length) / sizeof(int32_t); } rpcbuf.value = rpchdr; rpcbuf.length = (u_char *)buf - (u_char *)rpchdr; checksum.value = msg->rm_call.cb_verf.oa_base; checksum.length = msg->rm_call.cb_verf.oa_length; maj_stat = gss_verify_mic(&min_stat, client->cl_ctx, &rpcbuf, &checksum, &qop_state); if (maj_stat != GSS_S_COMPLETE) { rpc_gss_log_status("gss_verify_mic", client->cl_mech, maj_stat, min_stat); /* * A bug in some versions of the Linux client generates a * Destroy operation with a bogus encrypted checksum. Deleting * the credential handle for that case causes the mount to fail. * Since the checksum is bogus (gss_verify_mic() failed), it * doesn't make sense to destroy the handle and not doing so * fixes the Linux mount. */ if (gcproc != RPCSEC_GSS_DESTROY) client->cl_state = CLIENT_STALE; return (FALSE); } *qop = qop_state; return (TRUE); } static bool_t svc_rpc_gss_nextverf(struct svc_rpc_gss_client *client, struct svc_req *rqst, u_int seq) { gss_buffer_desc signbuf; gss_buffer_desc mic; OM_uint32 maj_stat, min_stat; uint32_t nseq; rpc_gss_log_debug("in svc_rpc_gss_nextverf()"); nseq = htonl(seq); signbuf.value = &nseq; signbuf.length = sizeof(nseq); maj_stat = gss_get_mic(&min_stat, client->cl_ctx, client->cl_qop, &signbuf, &mic); if (maj_stat != GSS_S_COMPLETE) { rpc_gss_log_status("gss_get_mic", client->cl_mech, maj_stat, min_stat); client->cl_state = CLIENT_STALE; return (FALSE); } KASSERT(mic.length <= MAX_AUTH_BYTES, ("MIC too large for RPCSEC_GSS")); rqst->rq_verf.oa_flavor = RPCSEC_GSS; rqst->rq_verf.oa_length = mic.length; bcopy(mic.value, rqst->rq_verf.oa_base, mic.length); gss_release_buffer(&min_stat, &mic); return (TRUE); } static bool_t svc_rpc_gss_callback(struct svc_rpc_gss_client *client, struct svc_req *rqst) { struct svc_rpc_gss_callback *scb; rpc_gss_lock_t lock; void *cookie; bool_t cb_res; bool_t result; /* * See if we have a callback for this guy. */ result = TRUE; SLIST_FOREACH(scb, &KGSS_VNET(svc_rpc_gss_callbacks), cb_link) { if (scb->cb_callback.program == rqst->rq_prog && scb->cb_callback.version == rqst->rq_vers) { /* * This one matches. Call the callback and see * if it wants to veto or something. */ lock.locked = FALSE; lock.raw_cred = &client->cl_rawcred; cb_res = scb->cb_callback.callback(rqst, client->cl_creds, client->cl_ctx, &lock, &cookie); if (!cb_res) { client->cl_state = CLIENT_STALE; result = FALSE; break; } /* * The callback accepted the connection - it * is responsible for freeing client->cl_creds * now. */ client->cl_creds = GSS_C_NO_CREDENTIAL; client->cl_locked = lock.locked; client->cl_cookie = cookie; return (TRUE); } } /* * Either no callback exists for this program/version or one * of the callbacks rejected the connection. We just need to * clean up the delegated client creds, if any. */ if (client->cl_creds) { OM_uint32 min_ver; gss_release_cred(&min_ver, &client->cl_creds); } return (result); } static bool_t svc_rpc_gss_check_replay(struct svc_rpc_gss_client *client, uint32_t seq) { uint32_t offset; int word, bit; bool_t result; sx_xlock(&client->cl_lock); if (seq <= client->cl_seqlast) { /* * The request sequence number is less than * the largest we have seen so far. If it is * outside the window or if we have seen a * request with this sequence before, silently * discard it. */ offset = client->cl_seqlast - seq; if (offset >= SVC_RPC_GSS_SEQWINDOW) { result = FALSE; goto out; } word = offset / 32; bit = offset % 32; if (client->cl_seqmask[word] & (1 << bit)) { result = FALSE; goto out; } } result = TRUE; out: sx_xunlock(&client->cl_lock); return (result); } static void svc_rpc_gss_update_seq(struct svc_rpc_gss_client *client, uint32_t seq) { int offset, i, word, bit; uint32_t carry, newcarry; sx_xlock(&client->cl_lock); if (seq > client->cl_seqlast) { /* * This request has a sequence number greater * than any we have seen so far. Advance the * seq window and set bit zero of the window * (which corresponds to the new sequence * number) */ offset = seq - client->cl_seqlast; while (offset > 32) { for (i = (SVC_RPC_GSS_SEQWINDOW / 32) - 1; i > 0; i--) { client->cl_seqmask[i] = client->cl_seqmask[i-1]; } client->cl_seqmask[0] = 0; offset -= 32; } carry = 0; for (i = 0; i < SVC_RPC_GSS_SEQWINDOW / 32; i++) { newcarry = client->cl_seqmask[i] >> (32 - offset); client->cl_seqmask[i] = (client->cl_seqmask[i] << offset) | carry; carry = newcarry; } client->cl_seqmask[0] |= 1; client->cl_seqlast = seq; } else { offset = client->cl_seqlast - seq; word = offset / 32; bit = offset % 32; client->cl_seqmask[word] |= (1 << bit); } sx_xunlock(&client->cl_lock); } enum auth_stat svc_rpc_gss(struct svc_req *rqst, struct rpc_msg *msg) { OM_uint32 min_stat; XDR xdrs; struct svc_rpc_gss_cookedcred *cc; struct svc_rpc_gss_client *client; struct rpc_gss_cred gc; struct rpc_gss_init_res gr; gss_qop_t qop; int call_stat; enum auth_stat result; KGSS_CURVNET_SET_QUIET(KGSS_TD_TO_VNET(curthread)); rpc_gss_log_debug("in svc_rpc_gss()"); /* Garbage collect old clients. */ svc_rpc_gss_timeout_clients(); /* Initialize reply. */ rqst->rq_verf = _null_auth; /* Deserialize client credentials. */ if (rqst->rq_cred.oa_length <= 0) { KGSS_CURVNET_RESTORE(); return (AUTH_BADCRED); } memset(&gc, 0, sizeof(gc)); xdrmem_create(&xdrs, rqst->rq_cred.oa_base, rqst->rq_cred.oa_length, XDR_DECODE); if (!xdr_rpc_gss_cred(&xdrs, &gc)) { XDR_DESTROY(&xdrs); KGSS_CURVNET_RESTORE(); return (AUTH_BADCRED); } XDR_DESTROY(&xdrs); client = NULL; /* Check version. */ if (gc.gc_version != RPCSEC_GSS_VERSION) { result = AUTH_BADCRED; goto out; } /* Check the proc and find the client (or create it) */ if (gc.gc_proc == RPCSEC_GSS_INIT) { if (gc.gc_handle.length != 0) { result = AUTH_BADCRED; goto out; } client = svc_rpc_gss_create_client(); } else { struct svc_rpc_gss_clientid *p; if (gc.gc_handle.length != sizeof(*p)) { result = AUTH_BADCRED; goto out; } p = gc.gc_handle.value; client = svc_rpc_gss_find_client(p); if (!client) { /* * Can't find the client - we may have * destroyed it - tell the other side to * re-authenticate. */ result = RPCSEC_GSS_CREDPROBLEM; goto out; } } cc = rqst->rq_clntcred; cc->cc_client = client; cc->cc_service = gc.gc_svc; cc->cc_seq = gc.gc_seq; /* * The service and sequence number must be ignored for * RPCSEC_GSS_INIT and RPCSEC_GSS_CONTINUE_INIT. */ if (gc.gc_proc != RPCSEC_GSS_INIT && gc.gc_proc != RPCSEC_GSS_CONTINUE_INIT) { /* * Check for sequence number overflow. */ if (gc.gc_seq >= MAXSEQ) { result = RPCSEC_GSS_CTXPROBLEM; goto out; } /* * Check for valid service. */ if (gc.gc_svc != rpc_gss_svc_none && gc.gc_svc != rpc_gss_svc_integrity && gc.gc_svc != rpc_gss_svc_privacy) { result = AUTH_BADCRED; goto out; } } /* Handle RPCSEC_GSS control procedure. */ switch (gc.gc_proc) { case RPCSEC_GSS_INIT: case RPCSEC_GSS_CONTINUE_INIT: if (rqst->rq_proc != NULLPROC) { result = AUTH_REJECTEDCRED; break; } memset(&gr, 0, sizeof(gr)); if (!svc_rpc_gss_accept_sec_context(client, rqst, &gr, &gc)) { result = AUTH_REJECTEDCRED; break; } if (gr.gr_major == GSS_S_COMPLETE) { /* * We borrow the space for the call verf to * pack our reply verf. */ rqst->rq_verf = msg->rm_call.cb_verf; if (!svc_rpc_gss_nextverf(client, rqst, gr.gr_win)) { result = AUTH_REJECTEDCRED; break; } } else { rqst->rq_verf = _null_auth; } call_stat = svc_sendreply(rqst, (xdrproc_t) xdr_rpc_gss_init_res, (caddr_t) &gr); gss_release_buffer(&min_stat, &gr.gr_token); if (!call_stat) { result = AUTH_FAILED; break; } if (gr.gr_major == GSS_S_COMPLETE) client->cl_state = CLIENT_ESTABLISHED; result = RPCSEC_GSS_NODISPATCH; break; case RPCSEC_GSS_DATA: case RPCSEC_GSS_DESTROY: if (!svc_rpc_gss_check_replay(client, gc.gc_seq)) { result = RPCSEC_GSS_NODISPATCH; break; } if (!svc_rpc_gss_validate(client, msg, &qop, gc.gc_proc)) { result = RPCSEC_GSS_CREDPROBLEM; break; } /* * We borrow the space for the call verf to pack our * reply verf. */ rqst->rq_verf = msg->rm_call.cb_verf; if (!svc_rpc_gss_nextverf(client, rqst, gc.gc_seq)) { result = RPCSEC_GSS_CTXPROBLEM; break; } svc_rpc_gss_update_seq(client, gc.gc_seq); /* * Change the SVCAUTH ops on the request to point at * our own code so that we can unwrap the arguments * and wrap the result. The caller will re-set this on * every request to point to a set of null wrap/unwrap * methods. Acquire an extra reference to the client * which will be released by svc_rpc_gss_release() * after the request has finished processing. */ refcount_acquire(&client->cl_refs); rqst->rq_auth.svc_ah_ops = &svc_auth_gss_ops; rqst->rq_auth.svc_ah_private = cc; if (gc.gc_proc == RPCSEC_GSS_DATA) { /* * We might be ready to do a callback to the server to * see if it wants to accept/reject the connection. */ sx_xlock(&client->cl_lock); if (!client->cl_done_callback) { client->cl_done_callback = TRUE; client->cl_qop = qop; client->cl_rawcred.qop = _rpc_gss_num_to_qop( client->cl_rawcred.mechanism, qop); if (!svc_rpc_gss_callback(client, rqst)) { result = AUTH_REJECTEDCRED; sx_xunlock(&client->cl_lock); break; } } sx_xunlock(&client->cl_lock); /* * If the server has locked this client to a * particular service+qop pair, enforce that * restriction now. */ if (client->cl_locked) { if (client->cl_rawcred.service != gc.gc_svc) { result = AUTH_FAILED; break; } else if (client->cl_qop != qop) { result = AUTH_BADVERF; break; } } /* * If the qop changed, look up the new qop * name for rawcred. */ if (client->cl_qop != qop) { client->cl_qop = qop; client->cl_rawcred.qop = _rpc_gss_num_to_qop( client->cl_rawcred.mechanism, qop); } /* * Make sure we use the right service value * for unwrap/wrap. */ if (client->cl_rawcred.service != gc.gc_svc) { client->cl_rawcred.service = gc.gc_svc; svc_rpc_gss_set_flavor(client); } result = AUTH_OK; } else { if (rqst->rq_proc != NULLPROC) { result = AUTH_REJECTEDCRED; break; } call_stat = svc_sendreply(rqst, (xdrproc_t) xdr_void, (caddr_t) NULL); if (!call_stat) { result = AUTH_FAILED; break; } svc_rpc_gss_forget_client(client); result = RPCSEC_GSS_NODISPATCH; break; } break; default: result = AUTH_BADCRED; break; } out: if (client) svc_rpc_gss_release_client(client); xdr_free((xdrproc_t) xdr_rpc_gss_cred, (char *) &gc); KGSS_CURVNET_RESTORE(); return (result); } static bool_t svc_rpc_gss_wrap(SVCAUTH *auth, struct mbuf **mp) { struct svc_rpc_gss_cookedcred *cc; struct svc_rpc_gss_client *client; rpc_gss_log_debug("in svc_rpc_gss_wrap()"); cc = (struct svc_rpc_gss_cookedcred *) auth->svc_ah_private; client = cc->cc_client; if (client->cl_state != CLIENT_ESTABLISHED || cc->cc_service == rpc_gss_svc_none || *mp == NULL) { return (TRUE); } return (xdr_rpc_gss_wrap_data(mp, client->cl_ctx, client->cl_qop, cc->cc_service, cc->cc_seq)); } static bool_t svc_rpc_gss_unwrap(SVCAUTH *auth, struct mbuf **mp) { struct svc_rpc_gss_cookedcred *cc; struct svc_rpc_gss_client *client; rpc_gss_log_debug("in svc_rpc_gss_unwrap()"); cc = (struct svc_rpc_gss_cookedcred *) auth->svc_ah_private; client = cc->cc_client; if (client->cl_state != CLIENT_ESTABLISHED || cc->cc_service == rpc_gss_svc_none) { return (TRUE); } return (xdr_rpc_gss_unwrap_data(mp, client->cl_ctx, client->cl_qop, cc->cc_service, cc->cc_seq)); } static void svc_rpc_gss_release(SVCAUTH *auth) { struct svc_rpc_gss_cookedcred *cc; struct svc_rpc_gss_client *client; rpc_gss_log_debug("in svc_rpc_gss_release()"); cc = (struct svc_rpc_gss_cookedcred *) auth->svc_ah_private; client = cc->cc_client; svc_rpc_gss_release_client(client); } diff --git a/sys/rpc/rpcsec_tls.h b/sys/rpc/rpcsec_tls.h index ba9a754bd276..a1f77a659ee2 100644 --- a/sys/rpc/rpcsec_tls.h +++ b/sys/rpc/rpcsec_tls.h @@ -1,105 +1,102 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2020 Rick Macklem * * 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. * * $FreeBSD$ */ #ifndef _RPC_RPCSEC_TLS_H_ #define _RPC_RPCSEC_TLS_H_ /* Operation values for rpctls syscall. */ #define RPCTLS_SYSC_CLSETPATH 1 #define RPCTLS_SYSC_CLSOCKET 2 #define RPCTLS_SYSC_CLSHUTDOWN 3 #define RPCTLS_SYSC_SRVSETPATH 4 #define RPCTLS_SYSC_SRVSOCKET 5 #define RPCTLS_SYSC_SRVSHUTDOWN 6 #define RPCTLS_SYSC_SRVSTARTUP 7 /* Max nprocs for SRV startup */ #define RPCTLS_SRV_MAXNPROCS 16 /* System call used by the rpctlscd, rpctlssd daemons. */ int rpctls_syscall(int, const char *); /* Flag bits to indicate certificate results. */ #define RPCTLS_FLAGS_HANDSHAKE 0x01 #define RPCTLS_FLAGS_GOTCERT 0x02 #define RPCTLS_FLAGS_SELFSIGNED 0x04 #define RPCTLS_FLAGS_VERIFIED 0x08 #define RPCTLS_FLAGS_DISABLED 0x10 #define RPCTLS_FLAGS_CERTUSER 0x20 #define RPCTLS_FLAGS_HANDSHFAIL 0x40 /* Error return values for upcall rpcs. */ #define RPCTLSERR_OK 0 #define RPCTLSERR_NOCLOSE 1 #define RPCTLSERR_NOSSL 2 #define RPCTLSERR_NOSOCKET 3 #ifdef _KERNEL /* Functions that perform upcalls to the rpctlsd daemon. */ enum clnt_stat rpctls_connect(CLIENT *newclient, char *certname, struct socket *so, uint64_t *sslp, uint32_t *reterr); enum clnt_stat rpctls_cl_handlerecord(uint64_t sec, uint64_t usec, uint64_t ssl, uint32_t *reterr); enum clnt_stat rpctls_srv_handlerecord(uint64_t sec, uint64_t usec, uint64_t ssl, int procpos, uint32_t *reterr); enum clnt_stat rpctls_cl_disconnect(uint64_t sec, uint64_t usec, uint64_t ssl, uint32_t *reterr); enum clnt_stat rpctls_srv_disconnect(uint64_t sec, uint64_t usec, uint64_t ssl, int procpos, uint32_t *reterr); /* Initialization function for rpcsec_tls. */ int rpctls_init(void); -/* Cleanup function for rpcsec_tls. */ -void rpctls_cleanup(void); - /* Get TLS information function. */ bool rpctls_getinfo(u_int *maxlen, bool rpctlscd_run, bool rpctlssd_run); /* String for AUTH_TLS reply verifier. */ #define RPCTLS_START_STRING "STARTTLS" /* ssl refno value to indicate TLS handshake being done. */ #define RPCTLS_REFNO_HANDSHAKE 0xFFFFFFFFFFFFFFFFULL /* Macros for VIMAGE. */ /* Just define the KRPC_VNETxxx() macros as VNETxxx() macros. */ #define KRPC_VNET_DEFINE(t, n) VNET_DEFINE(t, n) #define KRPC_VNET_DEFINE_STATIC(t, n) VNET_DEFINE_STATIC(t, n) #define KRPC_VNET(n) VNET(n) #define KRPC_CURVNET_SET(n) CURVNET_SET(n) #define KRPC_CURVNET_SET_QUIET(n) CURVNET_SET_QUIET(n) #define KRPC_CURVNET_RESTORE() CURVNET_RESTORE() #define KRPC_TD_TO_VNET(n) TD_TO_VNET(n) #endif /* _KERNEL */ #endif /* _RPC_RPCSEC_TLS_H_ */ diff --git a/sys/rpc/rpcsec_tls/rpctls_impl.c b/sys/rpc/rpcsec_tls/rpctls_impl.c index 92b8b9481666..4d81b8612d36 100644 --- a/sys/rpc/rpcsec_tls/rpctls_impl.c +++ b/sys/rpc/rpcsec_tls/rpctls_impl.c @@ -1,853 +1,855 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2008 Isilon Inc http://www.isilon.com/ * Authors: Doug Rabson * Developed with Red Inc: Alfred Perlstein * * 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. */ /* Modified from the kernel GSSAPI code for RPC-over-TLS. */ #include __FBSDID("$FreeBSD$"); #include "opt_kern_tls.h" #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 "rpctlscd.h" #include "rpctlssd.h" /* * Syscall hooks */ static struct syscall_helper_data rpctls_syscalls[] = { SYSCALL_INIT_HELPER(rpctls_syscall), SYSCALL_INIT_LAST }; static CLIENT *rpctls_connect_handle; static struct mtx rpctls_connect_lock; static struct socket *rpctls_connect_so = NULL; static CLIENT *rpctls_connect_cl = NULL; static struct mtx rpctls_server_lock; static struct opaque_auth rpctls_null_verf; KRPC_VNET_DEFINE_STATIC(CLIENT **, rpctls_server_handle); KRPC_VNET_DEFINE_STATIC(struct socket *, rpctls_server_so) = NULL; KRPC_VNET_DEFINE_STATIC(SVCXPRT *, rpctls_server_xprt) = NULL; KRPC_VNET_DEFINE_STATIC(bool, rpctls_srv_newdaemon) = false; KRPC_VNET_DEFINE_STATIC(int, rpctls_srv_prevproc) = 0; KRPC_VNET_DEFINE_STATIC(bool *, rpctls_server_busy); static CLIENT *rpctls_connect_client(void); static CLIENT *rpctls_server_client(int procpos); static enum clnt_stat rpctls_server(SVCXPRT *xprt, struct socket *so, uint32_t *flags, uint64_t *sslp, uid_t *uid, int *ngrps, gid_t **gids, int *procposp); static void rpctls_vnetinit(const void *unused __unused) { int i; KRPC_VNET(rpctls_server_handle) = malloc(sizeof(CLIENT *) * RPCTLS_SRV_MAXNPROCS, M_RPC, M_WAITOK | M_ZERO); KRPC_VNET(rpctls_server_busy) = malloc(sizeof(bool) * RPCTLS_SRV_MAXNPROCS, M_RPC, M_WAITOK | M_ZERO); for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) KRPC_VNET(rpctls_server_busy)[i] = false; } VNET_SYSINIT(rpctls_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, rpctls_vnetinit, NULL); +static void +rpctls_cleanup(void *unused __unused) +{ + + free(KRPC_VNET(rpctls_server_handle), M_RPC); + free(KRPC_VNET(rpctls_server_busy), M_RPC); +} +VNET_SYSUNINIT(rpctls_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, + rpctls_cleanup, NULL); + int rpctls_init(void) { int error; error = syscall_helper_register(rpctls_syscalls, SY_THR_STATIC_KLD); if (error != 0) { printf("rpctls_init: cannot register syscall\n"); return (error); } mtx_init(&rpctls_connect_lock, "rpctls_connect_lock", NULL, MTX_DEF); mtx_init(&rpctls_server_lock, "rpctls_server_lock", NULL, MTX_DEF); rpctls_null_verf.oa_flavor = AUTH_NULL; rpctls_null_verf.oa_base = RPCTLS_START_STRING; rpctls_null_verf.oa_length = strlen(RPCTLS_START_STRING); return (0); } int sys_rpctls_syscall(struct thread *td, struct rpctls_syscall_args *uap) { struct sockaddr_un sun; struct netconfig *nconf; struct file *fp; struct socket *so; SVCXPRT *xprt; char path[MAXPATHLEN]; int fd = -1, error, i, try_count; CLIENT *cl, *oldcl[RPCTLS_SRV_MAXNPROCS], *concl; uint64_t ssl[3]; struct timeval timeo; #ifdef KERN_TLS u_int maxlen; #endif error = priv_check(td, PRIV_NFS_DAEMON); if (error != 0) return (error); KRPC_CURVNET_SET(KRPC_TD_TO_VNET(td)); switch (uap->op) { case RPCTLS_SYSC_SRVSTARTUP: if (jailed(curthread->td_ucred) && !prison_check_nfsd(curthread->td_ucred)) error = EPERM; if (error == 0) { /* Get rid of all old CLIENTs. */ mtx_lock(&rpctls_server_lock); for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { oldcl[i] = KRPC_VNET(rpctls_server_handle)[i]; KRPC_VNET(rpctls_server_handle)[i] = NULL; KRPC_VNET(rpctls_server_busy)[i] = false; } KRPC_VNET(rpctls_srv_newdaemon) = true; KRPC_VNET(rpctls_srv_prevproc) = 0; mtx_unlock(&rpctls_server_lock); for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { if (oldcl[i] != NULL) { CLNT_CLOSE(oldcl[i]); CLNT_RELEASE(oldcl[i]); } } } break; case RPCTLS_SYSC_CLSETPATH: if (jailed(curthread->td_ucred)) error = EPERM; if (error == 0) error = copyinstr(uap->path, path, sizeof(path), NULL); if (error == 0) { error = ENXIO; #ifdef KERN_TLS if (rpctls_getinfo(&maxlen, false, false)) error = 0; #endif } if (error == 0 && (strlen(path) + 1 > sizeof(sun.sun_path) || strlen(path) == 0)) error = EINVAL; cl = NULL; if (error == 0) { sun.sun_family = AF_LOCAL; strlcpy(sun.sun_path, path, sizeof(sun.sun_path)); sun.sun_len = SUN_LEN(&sun); nconf = getnetconfigent("local"); cl = clnt_reconnect_create(nconf, (struct sockaddr *)&sun, RPCTLSCD, RPCTLSCDVERS, RPC_MAXDATASIZE, RPC_MAXDATASIZE); /* * The number of retries defaults to INT_MAX, which * effectively means an infinite, uninterruptable loop. * Set the try_count to 1 so that no retries of the * RPC occur. Since it is an upcall to a local daemon, * requests should not be lost and doing one of these * RPCs multiple times is not correct. * If the server is not working correctly, the * daemon can get stuck in SSL_connect() trying * to read data from the socket during the upcall. * Set a timeout (currently 15sec) and assume the * daemon is hung when the timeout occurs. */ if (cl != NULL) { try_count = 1; CLNT_CONTROL(cl, CLSET_RETRIES, &try_count); timeo.tv_sec = 15; timeo.tv_usec = 0; CLNT_CONTROL(cl, CLSET_TIMEOUT, &timeo); } else error = EINVAL; } mtx_lock(&rpctls_connect_lock); oldcl[0] = rpctls_connect_handle; rpctls_connect_handle = cl; mtx_unlock(&rpctls_connect_lock); if (oldcl[0] != NULL) { CLNT_CLOSE(oldcl[0]); CLNT_RELEASE(oldcl[0]); } break; case RPCTLS_SYSC_SRVSETPATH: if (jailed(curthread->td_ucred) && !prison_check_nfsd(curthread->td_ucred)) error = EPERM; if (error == 0) error = copyinstr(uap->path, path, sizeof(path), NULL); if (error == 0) { error = ENXIO; #ifdef KERN_TLS if (rpctls_getinfo(&maxlen, false, false)) error = 0; #endif } if (error == 0 && (strlen(path) + 1 > sizeof(sun.sun_path) || strlen(path) == 0)) error = EINVAL; cl = NULL; if (error == 0) { sun.sun_family = AF_LOCAL; strlcpy(sun.sun_path, path, sizeof(sun.sun_path)); sun.sun_len = SUN_LEN(&sun); nconf = getnetconfigent("local"); cl = clnt_reconnect_create(nconf, (struct sockaddr *)&sun, RPCTLSSD, RPCTLSSDVERS, RPC_MAXDATASIZE, RPC_MAXDATASIZE); /* * The number of retries defaults to INT_MAX, which * effectively means an infinite, uninterruptable loop. * Set the try_count to 1 so that no retries of the * RPC occur. Since it is an upcall to a local daemon, * requests should not be lost and doing one of these * RPCs multiple times is not correct. * Set a timeout (currently 15sec) and assume that * the daemon is hung if a timeout occurs. */ if (cl != NULL) { try_count = 1; CLNT_CONTROL(cl, CLSET_RETRIES, &try_count); timeo.tv_sec = 15; timeo.tv_usec = 0; CLNT_CONTROL(cl, CLSET_TIMEOUT, &timeo); } else error = EINVAL; } for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) oldcl[i] = NULL; mtx_lock(&rpctls_server_lock); if (KRPC_VNET(rpctls_srv_newdaemon)) { /* * For a new daemon, the rpctls_srv_handles have * already been cleaned up by RPCTLS_SYSC_SRVSTARTUP. * Scan for an available array entry to use. */ for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { if (KRPC_VNET(rpctls_server_handle)[i] == NULL) break; } if (i == RPCTLS_SRV_MAXNPROCS && error == 0) error = ENXIO; } else { /* For an old daemon, clear out old CLIENTs. */ for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { oldcl[i] = KRPC_VNET(rpctls_server_handle)[i]; KRPC_VNET(rpctls_server_handle)[i] = NULL; KRPC_VNET(rpctls_server_busy)[i] = false; } i = 0; /* Set to use rpctls_server_handle[0]. */ } if (error == 0) KRPC_VNET(rpctls_server_handle)[i] = cl; mtx_unlock(&rpctls_server_lock); for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { if (oldcl[i] != NULL) { CLNT_CLOSE(oldcl[i]); CLNT_RELEASE(oldcl[i]); } } break; case RPCTLS_SYSC_CLSHUTDOWN: mtx_lock(&rpctls_connect_lock); oldcl[0] = rpctls_connect_handle; rpctls_connect_handle = NULL; mtx_unlock(&rpctls_connect_lock); if (oldcl[0] != NULL) { CLNT_CLOSE(oldcl[0]); CLNT_RELEASE(oldcl[0]); } break; case RPCTLS_SYSC_SRVSHUTDOWN: mtx_lock(&rpctls_server_lock); for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { oldcl[i] = KRPC_VNET(rpctls_server_handle)[i]; KRPC_VNET(rpctls_server_handle)[i] = NULL; } KRPC_VNET(rpctls_srv_newdaemon) = false; mtx_unlock(&rpctls_server_lock); for (i = 0; i < RPCTLS_SRV_MAXNPROCS; i++) { if (oldcl[i] != NULL) { CLNT_CLOSE(oldcl[i]); CLNT_RELEASE(oldcl[i]); } } break; case RPCTLS_SYSC_CLSOCKET: mtx_lock(&rpctls_connect_lock); so = rpctls_connect_so; rpctls_connect_so = NULL; concl = rpctls_connect_cl; rpctls_connect_cl = NULL; mtx_unlock(&rpctls_connect_lock); if (so != NULL) { error = falloc(td, &fp, &fd, 0); if (error == 0) { /* * Set ssl refno so that clnt_vc_destroy() will * not close the socket and will leave that for * the daemon to do. */ soref(so); ssl[0] = ssl[1] = 0; ssl[2] = RPCTLS_REFNO_HANDSHAKE; CLNT_CONTROL(concl, CLSET_TLS, ssl); finit(fp, FREAD | FWRITE, DTYPE_SOCKET, so, &socketops); fdrop(fp, td); /* Drop fp reference. */ td->td_retval[0] = fd; } } else error = EPERM; break; case RPCTLS_SYSC_SRVSOCKET: mtx_lock(&rpctls_server_lock); so = KRPC_VNET(rpctls_server_so); KRPC_VNET(rpctls_server_so) = NULL; xprt = KRPC_VNET(rpctls_server_xprt); KRPC_VNET(rpctls_server_xprt) = NULL; mtx_unlock(&rpctls_server_lock); if (so != NULL) { error = falloc(td, &fp, &fd, 0); if (error == 0) { /* * Once this file descriptor is associated * with the socket, it cannot be closed by * the server side krpc code (svc_vc.c). */ soref(so); sx_xlock(&xprt->xp_lock); xprt->xp_tls = RPCTLS_FLAGS_HANDSHFAIL; sx_xunlock(&xprt->xp_lock); finit(fp, FREAD | FWRITE, DTYPE_SOCKET, so, &socketops); fdrop(fp, td); /* Drop fp reference. */ td->td_retval[0] = fd; } } else error = EPERM; break; default: error = EINVAL; } KRPC_CURVNET_RESTORE(); return (error); } /* * Acquire the rpctls_connect_handle and return it with a reference count, * if it is available. */ static CLIENT * rpctls_connect_client(void) { CLIENT *cl; mtx_lock(&rpctls_connect_lock); cl = rpctls_connect_handle; if (cl != NULL) CLNT_ACQUIRE(cl); mtx_unlock(&rpctls_connect_lock); return (cl); } /* * Acquire the rpctls_server_handle and return it with a reference count, * if it is available. */ static CLIENT * rpctls_server_client(int procpos) { CLIENT *cl; KRPC_CURVNET_SET_QUIET(KRPC_TD_TO_VNET(curthread)); mtx_lock(&rpctls_server_lock); cl = KRPC_VNET(rpctls_server_handle)[procpos]; if (cl != NULL) CLNT_ACQUIRE(cl); mtx_unlock(&rpctls_server_lock); KRPC_CURVNET_RESTORE(); return (cl); } /* Do an upcall for a new socket connect using TLS. */ enum clnt_stat rpctls_connect(CLIENT *newclient, char *certname, struct socket *so, uint64_t *sslp, uint32_t *reterr) { struct rpctlscd_connect_arg arg; struct rpctlscd_connect_res res; struct rpc_callextra ext; struct timeval utimeout; enum clnt_stat stat; CLIENT *cl; int val; static bool rpctls_connect_busy = false; cl = rpctls_connect_client(); if (cl == NULL) return (RPC_AUTHERROR); /* First, do the AUTH_TLS NULL RPC. */ memset(&ext, 0, sizeof(ext)); utimeout.tv_sec = 30; utimeout.tv_usec = 0; ext.rc_auth = authtls_create(); stat = clnt_call_private(newclient, &ext, NULLPROC, (xdrproc_t)xdr_void, NULL, (xdrproc_t)xdr_void, NULL, utimeout); AUTH_DESTROY(ext.rc_auth); if (stat == RPC_AUTHERROR) return (stat); if (stat != RPC_SUCCESS) return (RPC_SYSTEMERROR); /* Serialize the connect upcalls. */ mtx_lock(&rpctls_connect_lock); while (rpctls_connect_busy) msleep(&rpctls_connect_busy, &rpctls_connect_lock, PVFS, "rtlscn", 0); rpctls_connect_busy = true; rpctls_connect_so = so; rpctls_connect_cl = newclient; mtx_unlock(&rpctls_connect_lock); /* Temporarily block reception during the handshake upcall. */ val = 1; CLNT_CONTROL(newclient, CLSET_BLOCKRCV, &val); /* Do the connect handshake upcall. */ if (certname != NULL) { arg.certname.certname_len = strlen(certname); arg.certname.certname_val = certname; } else arg.certname.certname_len = 0; stat = rpctlscd_connect_1(&arg, &res, cl); if (stat == RPC_SUCCESS) { *reterr = res.reterr; if (res.reterr == 0) { *sslp++ = res.sec; *sslp++ = res.usec; *sslp = res.ssl; } } else if (stat == RPC_TIMEDOUT) { /* * Do a shutdown on the socket, since the daemon is probably * stuck in SSL_connect() trying to read the socket. * Do not soclose() the socket, since the daemon will close() * the socket after SSL_connect() returns an error. */ soshutdown(so, SHUT_RD); } CLNT_RELEASE(cl); /* Unblock reception. */ val = 0; CLNT_CONTROL(newclient, CLSET_BLOCKRCV, &val); /* Once the upcall is done, the daemon is done with the fp and so. */ mtx_lock(&rpctls_connect_lock); rpctls_connect_so = NULL; rpctls_connect_cl = NULL; rpctls_connect_busy = false; wakeup(&rpctls_connect_busy); mtx_unlock(&rpctls_connect_lock); return (stat); } /* Do an upcall to handle an non-application data record using TLS. */ enum clnt_stat rpctls_cl_handlerecord(uint64_t sec, uint64_t usec, uint64_t ssl, uint32_t *reterr) { struct rpctlscd_handlerecord_arg arg; struct rpctlscd_handlerecord_res res; enum clnt_stat stat; CLIENT *cl; cl = rpctls_connect_client(); if (cl == NULL) { *reterr = RPCTLSERR_NOSSL; return (RPC_SUCCESS); } /* Do the handlerecord upcall. */ arg.sec = sec; arg.usec = usec; arg.ssl = ssl; stat = rpctlscd_handlerecord_1(&arg, &res, cl); CLNT_RELEASE(cl); if (stat == RPC_SUCCESS) *reterr = res.reterr; return (stat); } enum clnt_stat rpctls_srv_handlerecord(uint64_t sec, uint64_t usec, uint64_t ssl, int procpos, uint32_t *reterr) { struct rpctlssd_handlerecord_arg arg; struct rpctlssd_handlerecord_res res; enum clnt_stat stat; CLIENT *cl; cl = rpctls_server_client(procpos); if (cl == NULL) { *reterr = RPCTLSERR_NOSSL; return (RPC_SUCCESS); } /* Do the handlerecord upcall. */ arg.sec = sec; arg.usec = usec; arg.ssl = ssl; stat = rpctlssd_handlerecord_1(&arg, &res, cl); CLNT_RELEASE(cl); if (stat == RPC_SUCCESS) *reterr = res.reterr; return (stat); } /* Do an upcall to shut down a socket using TLS. */ enum clnt_stat rpctls_cl_disconnect(uint64_t sec, uint64_t usec, uint64_t ssl, uint32_t *reterr) { struct rpctlscd_disconnect_arg arg; struct rpctlscd_disconnect_res res; enum clnt_stat stat; CLIENT *cl; cl = rpctls_connect_client(); if (cl == NULL) { *reterr = RPCTLSERR_NOSSL; return (RPC_SUCCESS); } /* Do the disconnect upcall. */ arg.sec = sec; arg.usec = usec; arg.ssl = ssl; stat = rpctlscd_disconnect_1(&arg, &res, cl); CLNT_RELEASE(cl); if (stat == RPC_SUCCESS) *reterr = res.reterr; return (stat); } enum clnt_stat rpctls_srv_disconnect(uint64_t sec, uint64_t usec, uint64_t ssl, int procpos, uint32_t *reterr) { struct rpctlssd_disconnect_arg arg; struct rpctlssd_disconnect_res res; enum clnt_stat stat; CLIENT *cl; cl = rpctls_server_client(procpos); if (cl == NULL) { *reterr = RPCTLSERR_NOSSL; return (RPC_SUCCESS); } /* Do the disconnect upcall. */ arg.sec = sec; arg.usec = usec; arg.ssl = ssl; stat = rpctlssd_disconnect_1(&arg, &res, cl); CLNT_RELEASE(cl); if (stat == RPC_SUCCESS) *reterr = res.reterr; return (stat); } /* Do an upcall for a new server socket using TLS. */ static enum clnt_stat rpctls_server(SVCXPRT *xprt, struct socket *so, uint32_t *flags, uint64_t *sslp, uid_t *uid, int *ngrps, gid_t **gids, int *procposp) { enum clnt_stat stat; CLIENT *cl; struct rpctlssd_connect_res res; gid_t *gidp; uint32_t *gidv; int i, procpos; KRPC_CURVNET_SET_QUIET(KRPC_TD_TO_VNET(curthread)); cl = NULL; procpos = -1; mtx_lock(&rpctls_server_lock); for (i = (KRPC_VNET(rpctls_srv_prevproc) + 1) % RPCTLS_SRV_MAXNPROCS; i != KRPC_VNET(rpctls_srv_prevproc); i = (i + 1) % RPCTLS_SRV_MAXNPROCS) { if (KRPC_VNET(rpctls_server_handle)[i] != NULL) break; } if (i == KRPC_VNET(rpctls_srv_prevproc)) { if (KRPC_VNET(rpctls_server_handle)[i] != NULL) procpos = i; } else KRPC_VNET(rpctls_srv_prevproc) = procpos = i; mtx_unlock(&rpctls_server_lock); if (procpos >= 0) cl = rpctls_server_client(procpos); if (cl == NULL) { KRPC_CURVNET_RESTORE(); return (RPC_SYSTEMERROR); } /* Serialize the server upcalls. */ mtx_lock(&rpctls_server_lock); while (KRPC_VNET(rpctls_server_busy)[procpos]) msleep(&KRPC_VNET(rpctls_server_busy)[procpos], &rpctls_server_lock, PVFS, "rtlssn", 0); KRPC_VNET(rpctls_server_busy)[procpos] = true; KRPC_VNET(rpctls_server_so) = so; KRPC_VNET(rpctls_server_xprt) = xprt; mtx_unlock(&rpctls_server_lock); /* Do the server upcall. */ res.gid.gid_val = NULL; stat = rpctlssd_connect_1(NULL, &res, cl); if (stat == RPC_SUCCESS) { *flags = res.flags; *sslp++ = res.sec; *sslp++ = res.usec; *sslp = res.ssl; *procposp = procpos; if ((*flags & (RPCTLS_FLAGS_CERTUSER | RPCTLS_FLAGS_DISABLED)) == RPCTLS_FLAGS_CERTUSER) { *ngrps = res.gid.gid_len; *uid = res.uid; *gids = gidp = mem_alloc(*ngrps * sizeof(gid_t)); gidv = res.gid.gid_val; for (i = 0; i < *ngrps; i++) *gidp++ = *gidv++; } } else if (stat == RPC_TIMEDOUT) { /* * Do a shutdown on the socket, since the daemon is probably * stuck in SSL_accept() trying to read the socket. * Do not soclose() the socket, since the daemon will close() * the socket after SSL_accept() returns an error. */ soshutdown(so, SHUT_RD); } CLNT_RELEASE(cl); mem_free(res.gid.gid_val, 0); /* Once the upcall is done, the daemon is done with the fp and so. */ mtx_lock(&rpctls_server_lock); KRPC_VNET(rpctls_server_so) = NULL; KRPC_VNET(rpctls_server_xprt) = NULL; KRPC_VNET(rpctls_server_busy)[procpos] = false; wakeup(&KRPC_VNET(rpctls_server_busy)[procpos]); mtx_unlock(&rpctls_server_lock); KRPC_CURVNET_RESTORE(); return (stat); } /* * Handle the NULL RPC with authentication flavor of AUTH_TLS. * This is a STARTTLS command, so do the upcall to the rpctlssd daemon, * which will do the TLS handshake. */ enum auth_stat _svcauth_rpcsec_tls(struct svc_req *rqst, struct rpc_msg *msg) { bool_t call_stat; enum clnt_stat stat; SVCXPRT *xprt; uint32_t flags; uint64_t ssl[3]; int ngrps, procpos; uid_t uid; gid_t *gidp; #ifdef KERN_TLS u_int maxlen; #endif /* Initialize reply. */ rqst->rq_verf = rpctls_null_verf; /* Check client credentials. */ if (rqst->rq_cred.oa_length != 0 || msg->rm_call.cb_verf.oa_length != 0 || msg->rm_call.cb_verf.oa_flavor != AUTH_NULL) return (AUTH_BADCRED); if (rqst->rq_proc != NULLPROC) return (AUTH_REJECTEDCRED); call_stat = FALSE; #ifdef KERN_TLS if (rpctls_getinfo(&maxlen, false, true)) call_stat = TRUE; #endif if (!call_stat) return (AUTH_REJECTEDCRED); /* * Disable reception for the krpc so that the TLS handshake can * be done on the socket in the rpctlssd daemon. */ xprt = rqst->rq_xprt; sx_xlock(&xprt->xp_lock); xprt->xp_dontrcv = TRUE; sx_xunlock(&xprt->xp_lock); /* * Send the reply to the NULL RPC with AUTH_TLS, which is the * STARTTLS command for Sun RPC. */ call_stat = svc_sendreply(rqst, (xdrproc_t)xdr_void, NULL); if (!call_stat) { sx_xlock(&xprt->xp_lock); xprt->xp_dontrcv = FALSE; sx_xunlock(&xprt->xp_lock); xprt_active(xprt); /* Harmless if already active. */ return (AUTH_REJECTEDCRED); } /* Do an upcall to do the TLS handshake. */ stat = rpctls_server(xprt, xprt->xp_socket, &flags, ssl, &uid, &ngrps, &gidp, &procpos); /* Re-enable reception on the socket within the krpc. */ sx_xlock(&xprt->xp_lock); xprt->xp_dontrcv = FALSE; if (stat == RPC_SUCCESS) { xprt->xp_tls = flags; xprt->xp_sslsec = ssl[0]; xprt->xp_sslusec = ssl[1]; xprt->xp_sslrefno = ssl[2]; xprt->xp_sslproc = procpos; if ((flags & (RPCTLS_FLAGS_CERTUSER | RPCTLS_FLAGS_DISABLED)) == RPCTLS_FLAGS_CERTUSER) { xprt->xp_ngrps = ngrps; xprt->xp_uid = uid; xprt->xp_gidp = gidp; } } sx_xunlock(&xprt->xp_lock); xprt_active(xprt); /* Harmless if already active. */ return (RPCSEC_GSS_NODISPATCH); } /* * Get kern.ipc.tls.enable and kern.ipc.tls.maxlen. */ bool rpctls_getinfo(u_int *maxlenp, bool rpctlscd_run, bool rpctlssd_run) { u_int maxlen; bool enable; int error; size_t siz; if (!mb_use_ext_pgs) return (false); siz = sizeof(enable); error = kernel_sysctlbyname(curthread, "kern.ipc.tls.enable", &enable, &siz, NULL, 0, NULL, 0); if (error != 0) return (false); siz = sizeof(maxlen); error = kernel_sysctlbyname(curthread, "kern.ipc.tls.maxlen", &maxlen, &siz, NULL, 0, NULL, 0); if (error != 0) return (false); if (rpctlscd_run && rpctls_connect_handle == NULL) return (false); KRPC_CURVNET_SET_QUIET(KRPC_TD_TO_VNET(curthread)); if (rpctlssd_run && KRPC_VNET(rpctls_server_handle)[0] == NULL) { KRPC_CURVNET_RESTORE(); return (false); } KRPC_CURVNET_RESTORE(); *maxlenp = maxlen; return (enable); } -void -rpctls_cleanup(void) -{ - - free(KRPC_VNET(rpctls_server_handle), M_RPC); - free(KRPC_VNET(rpctls_server_busy), M_RPC); -} -